Introduction To TH 00 Cars Rich

INTRODUCTION TO THE
THEORY OF* FOURIER'S

SERIES AND INTEGRALS
;
BY
H.
S.
CARSLAW
Sc.D. (CAMBRIDGE), D.Sc. (GLASGOW), F.R.S.E.

PROFESSOR OF MATHEMATICS IN THE UNIVERSITY OF SYDNEY
FORMERLY FELLOW OF EMMANUEL COLLEGE, CAMBRIDGE, AND LECTURER IN MATHEMATICS
IN THE UNIVERSITY OF GLASGOW
SECOND EDITION, COMPLETELY REVISED
MACMILLAN AND
ST.
CO.,
LIMITED
MARTIN'S STREET, LONDON

.1921
COPYRIGHT.
First Published, 1906.
GLASGOW: PRINTED AT THE UNIVERSITV PRESS

BY ROBERT MACLBHQSE AND CO, LTP.
PREFACE
THIS book forms the first volume of the new edition of my book
on Fourier's Series and Integrals and the Mathematical Theory of the
Conduction of Heat, published in 1906. and now for some time out
of print.
Since 1906 so much advance has been made in the Theory
of Fourier's Series and Integrals, as well as in the mathematical
discussion of
Heat Conduction, that it has seemed advisable to

new work, and to issue the same in two volumes.
write a completely
The
first
volume, which
of Infinite Series
Series
and
and
Integrals.
now
appears,
is
concerned with the Theory
Integrals, with special reference to Fourier's

The second volume will be devoted to the
Mathematical Theory of the Conduction of Heat.

No one can properly understand Fourier's Series and Integrals
without a knowledge of what is involved in the convergence of
and integrals. With these questions is bound up

the development of the idea of a limit and a function, and both
The first
are founded upon the modern theory of real numbers.
infinite series
three chapters deal with these matters. In Chapter IV. the Definite
Integral is treated from Riemann's point of view, and special
attention is given to the question of the convergence of infinite
The theory of series whose terms are functions of a

single variable, and the theory of integrals which contain an arbitrary parameter are discussed in Chapters V. and VI. It will be
seen that the two theories are closely related, and can be developed
integrals.
on similar
lines.
The treatment
of Fourier's Series in Chapter VII. depends
Dirichlet's Integrals.
the Second Theorem
on
There, and elsewhere throughout the book,

of Mean Value will be found an essential part
of the argument. In the same chapter the work of Poisson is

adapted to modern standards, and a prominent place is given
to Fejer's work, both in the proof of the fundamental theorem
h ,
PREFACE
VI
and
in the discussion of the nature of the convergence of Fourier's
Chapter IX.
devoted to Gibbs's Phenomenon, and the

last chapter to Fourier's Integrals.
In this chapter the work of
Pringsheim, who has greatly extended the class of functions to
which Fourier's Integral Theorem applies, has been used.
Series.
is
Two appendices are added. The first deals with Practical Harmonic Analysis and Periodogram Analysis. In the second a bibliography of the subject is given.
The functions treated in this book are " ordinary " functions.
An
for which f(x) is defined can be broken up into a

of open partial intervals, in each of which the function
monotonic. If infinities occur in the range, they are isolated
interval
is
(a, b)
number
finite
and finite in number. Such functions will satisfy most of the

demands of the Applied Mathematician.
The modern theory of integration, associated chiefly with the
name
of Lebesgue, has introduced into the Theory of Fourier's

and Integrals functions of a far more complicated nature.
Various writers, notably W. H. Young, are engaged in building up
Series
a theory of these and allied series much more advanced than anything treated in this book. These developments are in the meantime
chiefly interesting to the Pure Mathematician specialising in the
Theory of Functions of a Real Variable. My purpose has been to

remove some of the difficulties of the Applied Mathematician.
The preparation of this book has occupied some time, and much
of
to
it
has been given as a
my
students.
final
To them
it
course in the Infinitesimal Calculus
owes much.
For assistance
in the
revision of the proofs and for many valuable suggestions, I

much indebted to Mr. E. M. Wellish, Mr. K. J. Lyons and Mr.
H. Thorne
of
of the
Department
Sydney.
EMMANUEL COLLEGE,
CAMBRIDGE, Jan. 1921.
of
am
H.
Mathematics in the University

H.
S.
CARSLAW.
CONTENTS
THE THEORY OF FOURIER'S SERIES AND INTEGRALS
PAGE
HISTORICAL INTRODUCTION
CHAPTER
EATIONAL AND IRRATIONAL NUMBERS

SECTION
Numbers
1-2.
Rational
3-5.
Irrational
6-7.
Relations of Magnitude for Real
Numbers
16
17
Numbers
8.
Dedekind's Theorem
9.
The Linear Continuum.
Dedekind's Axiom
The Development
System
10.
23
of the
of Real
CHAPTER
INFINITE SEQUENCES
11.
Infinite
12.
The Upper and Lower Bounds
13.
Limiting Points of an Aggregate
14.
Weierstrass's
15.
Convergent Sequences
Divergent and Oscillatory Sequences
Monotonic Sequences
16.
17.
18.
Aggregates
21
24
Numbers
25
II
AND
SERIES
29
of au
30
Aggregate
31
Theorem
Theorem on an
32
33
....
Infinite Set of Intervals
19-23. Infinite Series
39
40
vii
37
41
CONTENTS
viii
CHAPTER
III
FUNCTIONS OF A SINGLE VARIABLE. LIMITS AND

CONTINUITY
--------
SECTION
24.
The Idea
Lt
25.
>a
of a
Function
PAGE
49
50
/(.r)
26.
Some Theorems on Limits
27.
Lt /(#)
...
52
55
28-29. Necessary
Limit
and
Sufficient Conditions for the Existence of a
56
30-32. Continuous Functions
59
33.
Discontinuous Functions
64
34.
Monotonic Functions
66
35.
Inverse Functions
68
36.
The
37.
Functions of Several Variables
Possibility of Expressing a Function as the Difference of

two Positive and Monotonic Increasing Functions -
69
71
CHAPTER IV
THE DEFINITE INTEGRAL
-
38.
Introductory
39.
The Sums S and
76
77
40.
Darboux's Theorem
41.
The
42.
Necessary and Sufficient Conditions for Integrability
43. Integrable
44.
79
Definite Integral of a
Functions
48.
Some Properties
The First Theorem
49.
The
45-47.
81
82
84
in (a, &) has an Infinite Number of Points

any Partial Interval of (a. b)
of the Definite Integral
of
Mean Value
92
Upper
The Second Theorem
93
of
Mean Value
94
Bounded Integrand. Infinite Interval

Integrals.
The Mean Value Theorems for Infinite Integrals
51-56. Infinite
59-61. Infinite Integrals.
86
87
Definite Integral considered as a Function of its
Limit
57-58.
Function integrable
of Continuity in
50.
Bounded Function
Integrand Infinite
Examples on Chapter IV
98
109
111
120
CONTENTS
ix
CHAPTER V
THE THEORY OF INFINITE SERIES WHOSE TERMS ARE
FUNCTIONS OF A SINGLE VARIABLE
PAGE
SECTION
62.
Introductory
63.
The Sum
122
--.
Series of Continuous Functions
of a
continuous
Uniform Convergence
124
127
129
71.
Term by Term
Term by Term
72.
The Power
73.
Extensions of Abel's Theorem on the Power Series
70.
be Dis-
....
64. Repeated Limits
65-69.
may
140
Integration
Differentiation
143
144
Series
74-76. Integration of Series.
Examples on Chapter
14!)
154
Infinite Integrals
164
CHAPTER
VI
DEFINITE INTEGRALS -CONTAINING AN ARBITRARY
PARAMETER
raf
77.
Continuity of the Integral
78.
Differentiation of the Integral
169
F(x^y)dx
Ja
F(x^y)dx
170
/"'*'
79.
Integration of the Integral
80-83. Infinite Integrals
f(xt y)ax
/""
I
M-l
Uniform Convergence
F(x, y)dx.
84.
Continuity of the Integral ( F(x,y)dx
85.
Integration of the Integral
173
179
I*
/
180
F(x^y)dx
,00
86.
Differentiation of the Integral/
F(x,y)dx
182
ra'
87. Properties of the Infinite Integral
F(x,y)dx
183
88.
Applications of the Preceding Theorems
89.
The Repeated Integral ( dx

f f(x, y) dy
190
Examples on Chapter VI
193
184
CONTENTS
CHAPTER
VII
FOURIER'S SERIES
SECTION
90.
PAGE
196
Introductory
91-92. Dirichlet's Integrals.
200
Form)
(First
Conditions
93. Dirichlet's
94. Dirichlet's Integrals.
(Second Form)
95.
Proof of the Convergence of Fourier's Series
96.
The Cosine
97.
The Sine
98.
Other Forms of Fourier's Series
...
Series
Series
99-100. Poisson's
Treatment
101.
FejeYs Theorem
102.
Two Theorems
103.
FejeYs Theorem and
206
207
210
215
220
228
of Fourier's Series
230
234
on the Arithmetic Means

Fourier's Series
238
240
Examples on Chapter VII
243
CHAPTER
VIII
THE NATURE OF THE CONVERGENCE OF FOURIER'S

SERIES
104-106.
The Order
of the
Terms
248
109.
The Uniform Convergence of Fourier's Series

Differentiation and Integration of Fourier's Series
110.
253
107-108.
more General Theory
of Fourier's Series
261
262
CHAPTER IX
THE APPROXIMATION CURVES AND
GIBBS'S
PHENOMENON
IN FOURIER'S SERIES
111-112.
264
The Approximation Curves

Phenomenon -
268
113-114. Gibbs's
115.
The Trigonometrical Sum 2J
sm ( 2 *'- 1 )'r
271
Sm
116. Gibbs's
Phenomenon
for the Series 2
117. Gibbs's
Phenomenon
for Fourier's Series
~~
'
280
i,.
in general
CONTENTS
xi
CHAPTER X
FOUEIEE'S INTEGRALS
SECTION
PAGE
283
118. Introductory
119. Fourier's
Integral
in its Simplest
120-121.
Theorem
Form
More General Conditions
Sommerfeld's Discussion
Examples on Chapter
APPENDIX
I.
APPENDIX
II.
Practical
for /(#)
and Fourier's Sine Integral
122. Fourier's Cosine Integral

123.
for the Arbitrary Function /(#)
of Fourier's Integrals
Harmonic Analysis and Periodogram Analysis
Bibliography
284
287
292
293
294
295
302
IN the middle of the eighteenth
century there was a prolonged
as
to
the
of
the expansion of an arbitrary
controversy
possibility
function of a real variable in a series of sines and cosines of
multiples of the variable. The question arose in connection with
the problem of the Vibrations of
The theory of these
Strings.
vibrations reduces to the solution of the Differential
Equation
and the
made by D'AlemD'Alembert and Euler

obtained the solution in the functional form
bert,*
earliest
Euler, t
attempts at
y=
The
its
and Bernoulli. +
(x
solution were
Both
+ at) +
\fs
(x
at).
between them lay in the fact that

D'Alembert supposed the initial form of the string to be
given
by a single analytical expression, while Euler regarded it as
principal difference
lying along any arbitrary continuous curve, different parts of

which might be given by different analytical
expressions.
Bernoulli, on the other hand, gave the solution, when the string
starts from rest, in the form of a
trigonometrical series
y = A x sm x cos at -f A 2 sin 2x cos 2at +

and he asserted that
this solution, being perfectly

general,
must
by Euler and D'Alembert. The importance
discovery was immediately recognised,. and Euler pointed
contain those given

of his
out that
if this statement of the solution were

correct, an
function
of a single variable must be
arbitrary
developable in
an infinite series of sines of
multiples of the variable. This he
*
Mem. de I'Acadtmie de
t/oc.
c.i
cit., 4, p.
Berlin, 3, p. 214, 1747.
+ loc.
69, 1748.
cit., 9,
p. 173, 1753.
TTTSTOTtTCAL
INTRODUCTION
held to be obviously impossible, since a series of sines is both

periodic and odd, and he argued that if the arbitrary function
had not both of these properties it could not be expanded in
such a
series.
this stage a memoir appeared in 1759*

a
then
by Lagrange,
young and unknown mathematician, in
which the problem was examined from a totally different point of
While he accepted Euler's solution as the most general, he
view.
While the debate was at
objected to the mode of demonstration, and he proposed to obtain

a satisfactory solution by first considering the case of a finite
number of particles stretched on a weightless string. From the
solution of this problem he deduced that of a continuous string by
making the number of particles infinite.! In this way he showed
that
when the
initial
given by f(x),
at time
t is
displacement of the string of unit length is

initial velocity by F(x), the displacement
and the
given by
ri
Jo
1
( si
7j ( sin mrx' sin mrx sin mrat)F(x')dx.

a-TrJo^n^
This result, and the discussion of the problem which Lagrange
gave in this and other memoirs, have prompted some mathematicians to deny the importance of Fourier's discoveries, and to
attribute to Lagrange the priority in the proof of the development

It is true
of an arbitrary function in trigonometrical series.
that in the formula quoted above it is only necessary to change

the order of summation and integration, and to put t = 0, in order
that
we may
obtain the development of the function f(x) in a

and that the coefficients shall take the definite
series of sines,
Still Lagrange
integral forms with which we are now familiar.
did not take this step, and, as Burkhardt remarks, I the fact that
is a very instructive example of the ease with
which an author omits to draw an almost obvious conclusion
from his results, when his investigation has been undertaken
with another end in view. Lagrange's purpose was to demon-
he did not do so
37.
t/oc. cit.,
*Cf. Lagrange, (Euvres, T. L, p. 37.
"
Entwicklungen nach oscillirenden Functionen," Jahrtsher.
J Burkhardt,
Math.
Ver., Leipzig, 10, Hft. II., p. 32, 1901.
I).
strate the truth of Euler's solution, and to defend- its general

When he had obtained
conclusions against D'Alembert 's attacks.
his solution he therefore proceeded to transform it into the functional
form given by Euler.
Having succeeded
in this, he held
his demonstration to be complete.

The further development of the theory of these series was due
to the astronomical problem of the expansion of the reciprocal
of the distance between two planets in a series of cosines of
As early as 1749
multiples of the angle between the radii.
and 1754 D'Alembert and Euler had published discussions of this
question in which the idea of the definite integral expressions
for the coefficients in Fourier's Series may be traced, and Clairaut,
in 1757,* gave his results in a form which practically contained
these coefficients.
Again, Euler. t in a paper written in 1777 and
in
1793, actually employed the method of multiplying
published
both sides of the equation
f(x)
by
cos
limits
nx and
and
TT.
-f
204 cos x
+ 2a
cos 2x
2a n cos nx -f
integrating the series term

In this way he found that
\x} cos
It is curious that
by term between the
nx dx.
'
ttltJStt
paper's" sêrn
to
have had no
effect
upon the discussion of the problem of the Vibrations of Strings

in which, as we have seen, D'Alembert, Euler, Bernoulli, and
Lagrange were about the same time engaged. The explanation
probably to be found in the fact that these results were not
accepted with confidence, and that they were only used in deter-
is
mining the coefficients of expansions whose existence could be

demonstrated by other means. It was left to Fourier to place
pur knowledge of the theory of trigonometrical series on a firmer
foundation, and, owing to the material advance made by him in
this subject the most important of these expansions are now
generally associated with his name am} called Fourier's Series.
His treatment was suggested by the problems he met in the
Mathematical Theory of the Conduction of Heat. It is to be
*
Paris, Hist. Acad.
tlêtrop.
X. Ada.,
sci. t
1754 [59], Art.
11, 1793 [98], p. 94
iv.
(Jury 1757).
(May
1777).
found in various memoirs, the most important having been

presented to the Paris Academy in 1811, although it was not
printed till 1824-6. These memoirs are practically contained in
In this
his book, Theorie mathematique de la chaleur (1822).
problem of the expansion of a
treatise several discussions of the
function in trigonometrical series are to be found. Some of

them fail in rigour. One is the same as that given by Euler.
However, it is a mistake to suppose that Fourier did not estab-
a rigorous and conclusive manner that a quite arbitrary

function (meaning by this any function capable of being represented by an arc of a continuous curve or by successive
lish in
portions of different continuous curves), could be represented

by the series we now associate with his name, and it is equally
wrong to attribute the first rigorous demonstration of this
theorem to Dirichlet, whose earliest memoir was published in

1829.* A closer examination of Fourier's work will show that
the importance of his investigations merits the fullest recognition, and Darboux, in the latest complete edition of Fourier's
mathematical works,! points out that the method he employed in

the final discussion of the general case is perfectly sound and
practically identical with that used later by Dirichlet in his
memoir.
classical
of
In this discussion Fourier followed the line

is now customary in dealing with infinite
argument which
He
series.
proved that when the values
If*
2-TrJ
1 f
an =
_.
f(x' ) cos nx' dx',
7TJ -TT
bn
= -1 f
TrJ
77
,'
f(x) sin nx' dx,
are inserted in the terms of the series
a
the
sum
+ (a
cos x
+b
of the terms
up
^
t
He
\ \ju
sin x)
+ (a
to cos
cos
nx and
1
sin |
(a;
2oj
+&
sin
nx
sin 2x)
is
x)
then discussed the limiting value of this sum as

*
Dirichlet, J. Math., Berlin, 4, p. 157, 1829.
t (Eurres de Fourier, T. I., p. 512, 1888.
n becomes
infinite,
and thus obtained the sum
of the series
now
called
Fourier's Series.
Fourier
made no claim
coefficients
to the discovery of the values of the

1
I Cn
a,i=-
f(x
cos
nx dx,
~"\
bn
f(x') sin
nx
711
dx'
TTj-n
We
have already seen that they were employed both by Clairaut

and Euler before this time. Still there is an important difference between Fourier's interpretation of these integrals and
that which was current among the mathematicians of the
eighteenth century. The earlier writers by whom they were
employed (with the possible exception of Clairaut) applied them

to the determination of the coefficients of series whose existence
had been demonstrated by other means. Fourier was the first
to apply them to the representation of an entirely arbitrary
function, in the sense in which we have explained this term.
In this he made a distinct advance upon his predecessors.
Indeed Riemann* asserts that when Fourier, in his first paper to
Academy in 1807, stated that a completely arbitrary

function could be expressed in such a series, his statement so
surprised Lagrange that he denied the possibility in the most
the Paris
It should also be noted that he was the first to

allow that the arbitrary function might be given by different
analytical expressions in different parts of the interval also that
definite terms.
he asserted that the sine series could be used for other functions
than odd ones, and the cosine series for other functions than
Further, he was the first to see that when a function
defined for a given range of the variable, its value outside
even ones.
is
that range is in no way determined, and it follows that no one

before him can have properly understood the representation of
an arbitrary function by a trigonometrical series.

The treatment which his work received from the Paris Academy
is evidence of the doubt with which his contemporaries viewed
*
Cf.
Riemann,
" Uher die Darstellbarkeit einer Function durch eine

trigono-
metrische Reihe," Gottingen, Ahh. Ges. Wi*s., 13,
2, 1867.
arguments and results. His first paper upon the Theory of

Heat was presented in 1807. The Academy, wishing to encourage the author to extend and improve his theory, made the
his
question of the propagation of heat the subject of the grand

de matkematiques for 1812.
Fourier submitted his
prix
Memoire sur
la propagation de la Chaleur at the end of 1811

a candidate for the prize. The memoir was referred to
Laplace, Lagrange, Legendre, and the other adjudicators but,
as
while awarding him the prize, they qualified their praise with
criticisms of the rigour of his analysis and methods,* and the
paper was not published
at the time in the

Memoires de
Academic des Sciences. Fourier always resented the treatment
he had received. When publishing his treatise in 1822, he
I'
incorporated in
it,
practically without change, the first part of
memoir; and two years later, having become Secretary of

the Academy on the death of Delambre, he caused his original
paper, in the form in which it had been communicated in 1811,
to be published in these MemoiresA
Probably this step was
this
taken to secure to himself the priority in his important discoveries,

in consequence of the attention the subject was receiving at
the hands of other mathematicians. It is also possible that he
wished to show the injustice of the criticisms which had been
passed upon his work. After the publication of his treatise,
when the results of his different memoirs had become known, it
was recognised that real advance had been made by him in the
treatment of the subject and the substantial accuracy of his
reasoning was admitted.^
*
Their report is quoted by Darboux in his Introduction (p. vii) to (Euvres de

" Cette
Fourier, T. I.
piece renferme les veri tables equations differentielles de la
transmission de la chaleur, soit a Tinterieur des corps, soit a leur surface ; et la
:
nouveaute du sujet, jointe a son importance, a determine la Classe a couronner cet

Ouvrage, en observant cependant que la maniere dont 1'Auteur parvient a ses
equations n'est pas exempte de difficultes, et que son analyse, pour les integrer,
laisse encore
du cote de
quelque chose a desirer, soit relativement a la gene>alite, soit
mcme
la rigueur."
t Memoires de VAcad. des Sc., 4, p. 185, and 5, P- 153.

Jit is interesting to note the following references to his work in the writings
of modern mathematicians
Kelvin, Coll. Works, Vol. III., p. 192 (Article on "Heat," Enc. Brit., 1878).
"
of all to
Returning to the question of the Conduction of Heat, we have first
say that the theory of it was discovered by Fourier, and given to the world
through the French Academy in his Theorie analytique de la Chahnr, with
:
The next writer upon the Theory of Heat was Poisson. He

employed an altogether different method in his discussion of the
a
question of the representation of an arbitrary function by
which
1820
in
from
his
onwards,
papers
are"
practically contained
in
e
(T.I. (2 ed.) 1833),and Theorie
He began with
his
books, Traitt de Me'canique
mathdmatique de la Ckaleur (1835).
the equation
/
CT ,
<
2hcos(xx
x),
solutions of problems naturally arising from it, of which it is difficult to say

whether their uniquely original quality, or their transcendently intense mathematical interest, or their perennially important instruct! veness for physical
science, is
most
to be praised."
Darboux, Introduction, (Euvres de Fourier, T. I., p. v, 1888.

" Par
1'importance de ses decouvertes, par I'influence decisive qu'il a exercee sur
le developpement de la Physique matheniatique, Fourier meritait 1'hommage qui
Son nom figurera digneest rendu aujourd'hui a ses travaux et a sa memoire.
ment a cote des noms, illustres entre tous, dont la liste, destinee a s'accroitre
avec les anne"es, constitue des a present un veritable titre d'honneur pour notre
La
.
,
que Ton peut placer sans
plus parfaits de tous les temps, se
recommande par une exposition interessante et originale des principes fondamentaux ; il eclaire de la lumiere la plus vive et la plus penetrante toutes les
idccs essentielles que nous devons & Fourier et sur lesquelles doit reposer
pays.
Theorie analytique de la Chaleur
injustice a cOte des ecrits scientifiques
les
mais il contient, nous devons le reconnaitre,

la Philosophie naturelle
beaucoup de negligences, des erreurs de calcul et de detail que Fourier a su eViter
dans d'autres ecrits."
d^sormais
1, 1891.
Poincare, Theorie analytiqne de la propagation de la Chaleur, p. 1,
" La theorie de la chaleur de Fourier est un des
premiers exemples de 1'appli-
en partant d'hypotheses simples qui ne sont

cation de 1'analyse a la physique
autre chose que des faits experimentaux generalises, Fourier en a deduit une
surie de consequences dont 1'ensemble constitue une theorie complete et cohe"rente.
;
qu'il a obtenus sont certes inte"ressants par eux-memes, mais ce qui

Test plus encore est la methode qu'il a employee pour y parvenir et qui servira
toujours de moclele a tous ceux qui voudront cultiver une branche quelconque de
Les resultats
la
physique matheniatique.
capitale dans 1'histoire des
J'ajouterai que le livre de Fourier a une importance

mathematiques et que 1'analyse pure lui doit peut-etre
plus encore que 1'analyse applique"e."

Boussinesq, Theorie analytique de la Chaleur, T. I., p. 4, 1901.
" Les admirables
applications qu'il fit de cette m^thode (i.e. his
method
of inte-
grating the equations of Conduction of Heat) sont, a la fois, assez simples et assez
generates, pour avoir servi de modele aux geometres de la premiere moitie de ce
*iecle
et elles leur ont ^t^ d'autant plus utiles, qu'elles ont pu, avec de legeres
tout au plus, etre transportees dans d'autres branches de la
modifications
Physique matheniatique, notamment dans I'Hydrodynamique et dans

de 1'elasticite."
la
Throne
h being numerically
less
than
unity,
and he obtained, by
integration,
1-tf
f(x')cosu(x'-x)dx.
-f
While
it
is
true that
by proceeding
to
the
limit
we may
deduce that
f(x) or i
is
equal to
f
TTJ _ w
we
are not entitled to assert that this holds for the value h=1,
we have already proved that the series converges for this
unless
value.
This
is
the real difficulty of Fourier's Series, and this
Poisson's discussion has been lost sight of in some

There are, however, other
presentations of Fourier's Series.
directions in which Poisson's method has led to most notable
limitation
011
The importance of his work cannot be exaggerated.*

After Poisson, Cauchy attacked the subject in different memoirs
published from ]826 onwards,! using his method of residues, but
results.
his treatment did not attract so
much
attention as that given
about the same time by Dirichlet, to which we
now
turn.
Dirichlet's investigation is contained in two memoirs which

The method which he employed
appeared in 1829 J and 1837.
we have already referred to in speaking of Fourier's work. He

based his proof upon a careful discussion of the limiting values
of the integrals
f
tt
-,
sin
'
Jo
sii
*For a
paper,
N.
full treatment of Poisson's method, reference may be made to Bocher's

" Introduction to the
Theory of Fourier's Series," Ann. Math., Princeton,
J. (Ser. 2), 7, 1906.
fSee Bibliography,
7)o?<e',s
p. 303.
J J.
Repertorium der Physik, Bd.
Math., Berlin,
I., p.
lf>2,
1837.
4, 1829.
as
/UL
sum
increases indefinitely.
this
By
means he showed that the
of the series
a o + ( a i cos x + &i s i n x ) + (2 COS 2# 4- 6 2 sin

where the
coefficients
etc.,
are those given
by
2se) -^
Fourier,
7r<x<^7r, f(x) has only a
provided that, while
finite
is
number
of ordinary discontinuities and turning points, and that it does

not become infinite in this range. In a later paper,* in which he
discussed the expansion in Spherical Harmonics, he showed that

the restriction that f(x) must remain finite is not necessary,
provided that
f(x) dx converges absolutely.
J -IT
The work
of Dirichlet led in a few years to one of the most

advances
not only in the treatment of trigonometrical
important
in
but
also
the
series,
Theory of Functions of a Real Variable
;
be said to have laid the foundations of that theory.

Tliis advance is to be found in the memoir -by Rieinann already
indeed
it
referred
in
1854,
may
which formed his Habilitationsschrift at Gottingen

but was not published till 1867, after his death.
to,
Riemann's aim was to determine the necessary conditions which

f(x) must satisfy, if it can be replaced by its Fourier's Series.
Dirichlet had shown only that certain conditions were sufficient.
The" question which Riemann set himself to answer, he did not
it still remains unsolved.
But in the
he perceived that it was necessary to widen
the concept of the definite integral as then understood.
Cauchy, in 1823", had defined the integral of a continuous
function as the limit of a sum, much in the way in which it is
still treated in our elementary text-books.
The interval of in-
indeed
completely solve
consideration of
it
tegration (a, b)
is first
divided into parts by the points
= &Q,
X^
8= (x, - BO)/^) +
(aj a
Cl
X<
Xn _
X n = 0.
The sum
- xjf(xt) +
Jba f(x)
*./.
fit It.,
I'xrlln, 17, 1S.T7.
dx
is
+ (xn - x n _ 0/OJ
defined as the limit of
10
this
sum when
the
number
of parts
increased indefinitely and
is
length diminished indefinitely. On this understanding

every continuous function has an integral.
For discontinuous functions, he proceeded as follows
their
If a function f(x) is continuous in an interval (a, b) except at

the point c, in the neighbourhood of which f(x) may be bounded
or not, the integral of f(x) in (a, b)
two
limits
Lt
when these
Riemann
in
defined as the
sum
of the
fb
Lt
f(x)dx,
'
/i_>oJ
and
is
fc-7*
/t
f(x]dx,
_> oJc+/t
limits exist.
dismisses altogether the requirement of continuity,

forming the sum S multiplies each interval (x r x r _^) by
the value of /(#), not necessarily at the .beginning (or end) of the
interval, but at a point r arbitrarily chosen between these, or by
a number intermediate between the lower and upper bounds of
is
{ba f(x)dx
such exists,
when
defined as the limit
number
the partial
intervals is increased indefinitely and their length tends to zero.
Riemann's treatment, given in the text in a slightly modified
form, is now generally adopted in a scientific treatment of the
of
this
sum,
if
the
of
Calculus. It is true that a more general theory of integration

has been developed in recent years, chiefly due to the writings
of Lebesgue,* de la Vallee Poussin and Young that theory is
;
Pure MatheBut no mathematician can neglect the concept of the

definite integral which Riemann introduced.
One of the immediate advances it brought was to bring within
the integrable functions a class of discontinuous functions whose
discontinuities were infinitely numerous in any finite interval.
An example, now classical, given by Riemann, was the function
mainly
for the specialist in certain branches of
matics.
defined
by the convergent
series
[x]
[nx]
[2x]
"pTT-sr-
'
tf
where [nx] denotes the positive or negative difference between

nx and the nearest integer, unless nx falls midway between two
consecutive integers,
*
See footnote,
when
p. 77.
the value of [nx]
is
to be taken as
zero.
The sum
of this series
is
11
discontinuous for every rational
value of x of the form p/2n, where p
is
an odd integer prime
to n.
With Riemann's
definition the restrictions
which Dirichlet had
placed upon the function f(x) were considerably relaxed.

this Biemann contributed much, and the numerous writers
To
who
have carried out similar investigations since his time have still
further widened the bounds, while the original idea that every
continuous function admitted of such an expansion has been
shown to be false. Still it remains true that for all practical
purposes, and for all ordinary functions, Dirichlet's investigation
established the convergence of the expansions.
Simplih'cations
have been introduced in his proof by the introduction of the
Second Theorem of Mean Value, and the use of a modified form

of Dirichlet's Integral, but the method which he employed
the basis of most rigorous discussions of Fourier's Series.
is still
The nature of the convergence of the series began to be examined after the discovery by Stokes (1847) and Seidel (1848) of
the property of Uniform Convergence. It had been known since
Dirichlet's time that the series were, in general, only conditionally
convergent, their convergence depending upon the presence of
both positive and negative terms.
showed
Heine
Dirichlet's
that,
if
the
It
function
Conditions in the interval
was not
is
(
finite
TT,
TT),
1870 that
and satisfies
till
the Fourier's
Series converges uniformly in any interval lying within an

interval in which f(x) is continuous.
This condition has, like
the other conditions of that time, since been somewhat modified.
In the last thirty or forty years quite a large literature has

arisen dealing with Fourier's Series.
The object of many of the
sufficient conditions to be
has
been
to
determine
investigations
satisfied
by the function
f(x), in order that its Fourier's Series
TT, TT), or at
converge, either throughout the interval (
It appears that the convergence
particular points of the interval.
or non-convergence of the series for a particular value of x
really
may
depends only upon the nature of the function in an arbitrarily

small neighbourhood of that point, and is independent of the
general character of the function throughout the interval, this
general character being limited only by the necessity for the
existence of the coefficients of the series.
These memoirs
12
associated chiefly with the names of Du Bois-Reymond, Lipschitz,

Dini, Heine, Cantor, Jordan, Lebesgue, de la Valle'e Poussin,
Hobson and Young have resulted in the discovery of sufficient
conditions of wide scope, which suffice for the convergence of the

series, either at particular points, or, generally, throughout the
interval.
The necessary and sufficient conditions for the con-
vergence of the series at a point of the interval, or throughout

any portion of it, have not been obtained. In view of the general
character of the problem, this is not surprising. Indeed it is not
improbable that no such necessary and sufficient conditions may
be obtainable.
In
many
of the
works referred
to above, written after the
discovery by Lebesgue (1902) of his general theory of integration, series whose terms did not exist under the old definition of
the integral are included in the discussion.
The fact that divergent series may be utilised in various
ways
widened the
in analysis has also
field of investigation,
and
indeed one of the most fruitful advances recently made arises

from the discussion of Fourier's Series which diverge. The word
"
sum,"
when
applied to a divergent series, has, of course, to be
but all methods of treatment agree in this, that"

the same process is applied to a convergent series the
"sum," according to the new definition, is to agree with the
"
"
sum obtained in the ordinary way. One of the most important
defined afresh
when
"
methods of " summation is due to Cesaro, and in its simplest

form is as follows
Denote by s n the sum of the first u terms of the series
:
j
Let
,
When
that
its
"
Lt Sn =
sum
"
S,
we say
convergent, then
so that the
"
9l
n
--.
that the series
is
"
suimnable," and
is S.
It is not difficult to
is
show that
if
the series
Lt Sn = Lt
-oo
sn
/'->-/;
condition of consistency
"
is satisfied.
[Cf.
102.]
was the
Fejer
first
13
to consider this sequence of Arithmetic
'Means,
*lt
If or
*1?2
He
the Fourier's Series.
that the sequence
is
at every point in
+ +
2
'
established the remarkable theorem *
convergent, and
TT, TT)
its limit is
where f(x + 0) and f(x
only conditions attached to f(x) being that,
be integrable in
TT, TT),
and
that, if it is
if
0) exist, the
it shall
bounded,
unbounded,
f(x)dx
shall be absolutely convergent.
Later,
Hardy showed
that
if
a series
summable by this method, and the general term u n is of the

order 1/n, the series is convergent in the ordinary sense, and thus
the sum -S (= Lt 8n ) and the sum s ( = Lt s n ) will be the same.
is
->CO
'/1->OQ
102.]
[Cf.
.Hardy's theorem, combined with FejeYs, leads at once to a

proof of the convergence of Fourier's Series, and it can also
new
be applied to the question of its uniform convergence. Many of

obtained by earlier investigators follow directly from
rthe results
the application to Fourier's Series of the general theory of
summable
series.!
Recent investigations show that the
now
"Series,
frequently
called
coefficients in Fourier's
Fourier's
Constants, have
im-
[portant properties, independent of whether the series converges

or not. For example, it is now known that if f(x) and <f>(x]
iare
two
tan
bn
functions, bounded and integrable in (

are Fourier's Constants for f(x), and a
TT,
f
,
TT),
an
',
and a
bn
'
those
for 0(a5), the series
converges, and
its
sum
1 f""
is
TTJ _-
Math. Ann., Leipzig, 58, 1904.
Chapman, Q.
(^''.
'->),
12, 1913.
J.
f(x)(j>(x)dx.
Math., London, 43, 1912
To
this theorem,
Hardy, London, Proc. Math. Soc.
14
and
to the results
which follow from
it,
much
attention has
recently been given, and it must be regarded as one of the most

important in the whole of the theory of Fourier's Series.*
The question
a
finite
He proved
1885.1
an arbitrary function by
was examined by Weierstrass in
f(x) is a continuous and periodic
of the approximation to
that
if
function, given the arbitrary small positive quantity e, a finite

Fourier's Series can be found in a variety of ways, such that the
absolute value of the difference of its sum and f(x) will be less
any value of x in the interval. This theorem was

by Picard, and it has been generalised in recent
memoirs by Stekloff and Kneser.
In the same connection, it should be noted that the application
of the method of least squares to the determination of the
than
for
also discussed
coefficients of a finite trigonometrical series leads to the Fourier
This result was given by Topler in 1876.1

As
of
Fourier's
Series
deal
a
with
applications
really only
coefficients.
many
finite
number
From
of terms, these results are of special interest.

the discussion of the Fourier's Series it was a natural
step to turn to the theory of the Trigonometrical Series
+ (ajCos x -f Z^sin x) + b
cos 2x
+6
sin 2x)
no longer the Fourier coefficients.

where the coefficients
The most important question to be answered was whether such.an
expansion was unique in other words, whether a function could
be represented by more than one such trigonometrical series.
This reduces to the question of whether zero can be represented
by a trigonometrical series in which the coefficients do not all
The discussion of this and similar problems was carried
vanish.
on chiefly by Heine and Cantor, from 1870 onwards, in a series
are
of papers which gave rise to the modern Theory of Sets of Point:

another instance of the remarkable influence Fourier's Series hav
In this place it wilJ
had upon the development of mathematics.
||
*Cf. Young, London, Proc. R. Soc. (A), 85, 1911.

\-J.
+
I!
Math., Berlin, 71, 1870.
Topler, Wien, Anz. Ak. Wiss., 13, 1876.
Bibliography,
p. 305.
Van Vleck, "The Influence of Fourier's Series upon the Development of Mathe-
matics," American Association for the Advancement of Science (Atlanta), 1913.

See also a paper with a similar title by Jourdain, Scientiu, Bologna (Ser. 2), 22,
1917.
be sufficient to state that Cantor showed that
all
15
the coefficients
of the trigonometrical series must vanish, if it is to be zero for

all values of x in the interval (-TT, TT), with the exception of
those which correspond to a set of points constituting, in the

tb
language of the Theory of Sets of Points, a set of the n order,
for which points we know nothing about the value of the series.
REFERENCES.
BURKHARDT,
"
Entwicklungen
n-ach oscillirenden
Functioneri," Jahresber.
D. Math. Ver., Leipzig, 10, 1901.
GIBSON,
"On
the History of the Fourier Series," Edinburgh, Proc. Math.
Soc., 11, 1893.
RIEMANN,
"
Uber
die Darstellbarkeit einer Function .durch erne trigonoGes. Wiss., 13, 1867; also Werke, pp.
metrische Reihe," Gottingen, Abh.

fcl
3-250.
SACHSE, Versuch einer Geschichte der Darstellung willkilrlicher Functionen

durch trigonometrische Rei/ien, Dias. Gottingen, 1879 also Z*. Math., Leipzig,
25, 1880, and, in French, Bui. sci. math., Paris (Ser. 3), 4, 1880.
;
CHAPTER
RATIONAL AND IRRATIONAL NUMBERS

THE SYSTEM OF REAL NUMBERS
1.
Rational Numbers.
The question
of the convergence of
Infinite Series is only capable of satisfactory treatment when

the difficulties underlying the conception of irrational number
have been overcome.
For
this reason
we
shall first of all give a
short discussion of that subject.

The idea of number is formed
by a series of generalisathe
The operations
with
positive integers.
begin
of addition and multiplication upon these numbers are always
possible; but if a and b are two positive integers, we cannot
determine positive integers x and y, so that the equations
tions.
We
a = b + x and
a=by
greater than
b,
is
a multiple of
b.
are satisfied, unless, in the
and, in the second case, a
is
first case,
To overcome this difficulty fractional and negative numbers are

introduced, and the system of rational numbers placed at our
disposal.*
The system of rational numbers is ordered, i.e. if we have two

numbers a and b of this system, one of them is greater
than the other. Also, if a>6 and b>c, then a>e, when a, b
and c are numbers of the system.
Further, if two different rational numbers a and b are given,
we can always find another rational number greater than the
different
one and
less
than the other.
It follows
from
this that
between
* The reader who wishes an extended treatment of the

system of rational
numbers is referred to Stolz und Gmeiner, Theoretische Arithmetik, Leipzig,
1900-1902, and Pringaheim, Vorlesnnycn ilber Zalilen- und Fnnktionenlehre,
Leipzig, 1916.
16

any two
different rational
numbers there are an
infinite
1*7
number
of rational numbers.*
The introduction
2.
bers
may
be justified
and negative rational numfrom two points of view. The fractional

of fractional
numbers are necessary for the representation of the subdivision

of a unit magnitude into several equal parts, and the negative
numbers form a valuable instrument for the measurement of
magnitudes which may be counted in opposite directions. This
taken as the argument of the applied mathematician.
is the argument of the pure mathewith
whom
the
notion
of number, positive and negative,
matician,
may be
On the
other hand there
integral and fractional, rests upon a foundation independent

of measurable magnitude, and in whose
eyes analysis is a
scheme which deals with numbers only, and has no concern
se
per
with measurable quantity.
It is possible to
found mathe-
matical analysis upon the notion of

positive integral number.
Thereafter the successive definitions of the different kinds of num-
and inequality among these numbers, and of the

four fundamental operations, may be presented abstractly.!
ber, of equality
3.
Irrational Numbers.
The extension
of the idea of
number
from the rational to the irrational is as natural, if not as easy, as

is that from the
positive integers to the fractional and negative
rational numbers.
Let a and b be any two positive integers. The equation x b = a
cannot be solved in terms of positive integers unless a is a perfect
6
th
To make the
power.
solution possible in general the irrational
numbers are introduced. But it will be seen below that the system
of irrational numbers is not confined to numbers which arise as
the roots of algebraical equations whose coefficients are
integers.
So much for the desirability of the extension from the abstract
side.
We
*
From the concrete the need for the extension is also evident.
have only to consider the measurement of any quantity to
When we
that there
set is less
is
say that a set of things has a finite number of members, we mean

a positive integer w, such that the total number of members of the
than
n.
When we
say that it has an infinite number of members, we mean that it has

unot a finite number. In other words, however
large n may be, there are more
members of the set than n.
tCf. Hobson, London, Proc. Math. Soc. (Ser. 1), 35,
p. 126, 1913; also the
author's Theory of Functions of a Real Variable,
13.
p.
C.I
same
18
which the property of unlimited divisibility is assigned, e.y. a

Take any segment of this
straight line L produced indefinitely.
line as unit of length, a definite point of the line as origin or
zero point, and the directions of right and left for the positive
and negative senses. To every rational number corresponds a
-2-1
Fm.
l.
on the line. If the number is an integer, the point

obtained by taking the required number of unit segments one
If it is a fraction
after the other in the proper direction.
p/q,
it is obtained by dividing the unit of length into q equal parts
definite point
is
and taking p of these to the right or left according as the sign is

These numbers are called the measures of
positive or negative.
the corresponding segments, and the segments are said to be
commensurable with the unit of length. The points corresponding to rational numbers may be called rational points.
There are, however, an infinite number of points on the line
L which are not rational points. Although we may approach
them as nearly as we please by choosing more and more
rational points on the line, we can never quite reach them in
The simplest example is the case of the points cointhis way.
one end of the diagonal of a square, the sides of
with
ciding
which are the unit of length, when the diagonal lies along the
line L and its other end coincides with any rational point.
Thus, without considering any other case of incommensurability, we see that the line L is infinitely richer in points than
the system of rational numbers in numbers.
Hence it is clear that if we desire to follow arithmetically all
the properties of the straight line, the rational numbers are
insufficient, and it will be necessary to extend this system by the
creation of other numbers.
Returning to the point of view of the pure mathematician,

now describe Dedekind's method of introducing the
irrational number, in its most general form, into analysis.*
4.
we
shall
*Dedekind (1831-1916) published his theory in Mctit/b-it nml 1,-rnlioimlf

Zahfen, Braunschweig, 1872 (English translation in Dedekind's Essays on Number,
Chicago, 1901).

all
19
Let us suppose that by some method or other we have divided

the rational numbers into two classes, a lower class A and an
upper class B, such that every number a of the lower class is less
than every number ,8 of the upper class.
When this division has been made, if a number a belongs to
number
the class A, every
number
does so
/3
less
if
/3
also.
Three different cases can arise

(I)
than a does so also; and

number greater than
belongs to the class B, every
The lower class can have a greatest number and the upper
class
no
smallest number.
This would occur
every number
less
if,
than
for example,
5 in the
we put
lower
class,
the
and
number
we put
if
5 and
in the
numbers greater than ^5.

upper
(II) The upper class can have a smallest number and the
lower class no greatest number.
This would occur if, for example, we put the number 5 and
all the numbers greater than 5 in the upper class, while in the
lower class we put all the numbers less than 5.
class all the
It
impossible that the lower class can have a greatest

class a smallest number n, in the
is
number m, and the upper
.same division of the rational
numbers
m and
numbers
for
between the rational
there are rational numbers, so that our hypothesis that the two classes contain all the rational numbers is
contradicted.
But a third case can

(III)
arise
The lower class can have no greatest number and the

upper class no smallest number.
For example,
squares in
numbers to
let
us arrange the positive integers and their

so that the squares are underneath the
two rows,
which they correspond.
tion in its lowest terms
is
Since the square of a fraca fraction whose numerator and deno-
minator are perfect squares,* we see that there are not rational
numbers whose squares are
It
'2
1 2 3
4' 5
2, 3, 5, 6, 7, 8, 10, 11, ...
3
6 7 8 9 10 11
4...
12 13 14 15 16
.,
a formal proof of this statement is needed, see Dedekind, foe. cil., English
Hardy, Course of Pure Mathematics (2nd Ed.), p. 6.
translation, p. 14, or
20
However, there are rational numbers whose squares are as near

numbers as we please. For instance, the numbers
these
2,
1-5,
142, 1-415, 1-4143,
1,
1-4,
1-41, 1-414, 1-4142, ...,
...,
form an upper and a lower set in which the squares of the terms
than 2, and the squares of the terms in the
are
than
2.
We can find a number in the upper
greater
upper
set and a number in the lower set such that their squares differ
from 2 by as little as we please.*
Now form a lower class, as described above, containing all
negative rational numbers, zero and all the positive rational
numbers whose squares are less than 2 and an upper class
containing all the positive rational numbers whose squares are
greater than 2. Then every rational number belongs to one class
Also every number in the lower class is less than
or the other.
every number in the upper. The lower class has no greatest
number and the upper class has no smallest number.
in the lower are less
5.
When by any means we have
obtained a division of
all
the
numbers into two classes of this kind, the lower class

no
greatest number and the upper class no smallest
having
create a new number defined by this division.
We
we
number,
and
we
that
it
is
than
an
irrational
it
call
number,
say
greater
all the rational numbers of its lower class, and less than all the
rational numbers of its upper class.
Such divisions are usually called sections.^ The irrational
number ^/2 is defined by the section of the rational numbers
described above. Similar sections would define the irrational
numbers 4/3, 4/^> etc. The system of irrational numbers is
given by all the possible divisions of the rational numbers into a
lower class A and an upper class B, such that every rational
number is in one class or the other, the numbers of the lower
class being less than the numbers of the upper class, while the
lower class has no greatest number, and the upper class no
rational
smallest number.
In other words, every irrational number is defined by its

"
"
It may be said to
section (A, B).
correspond to this section.
*Cf. Hardy, he.
f French, coupure;
Y., p. 8.
German,
Schnift.

The system
of rational
make up
together
21
numbers and irrational numbers
the system of real numbers.
The rational numbers themselves "correspond" to divisions of rational

numbers.
For instance, take the rational number m. In the lower class A put
the rational numbers less than m, and
m itself.
the rational numbers greater than m.

of the rational numbers.
Then
In the upper class B put all

corresponds to this division
Extending the meaning of the term section, as used above in the definition
number, to divisions in which the lower and upper classes
have greatest or smallest numbers, we may say that the rational number m
corresponds to a rational section (A, B),* and that the irrational numbers
correspond to irrational sections. When the rational and irrational numbers
are defined in this way, and together form the system of real numbers, the
real number which corresponds to the rational number m (to save confusion
it is sometimes called the rational-real number) is
conceptually distinct from
m. However, the relations of magnitude, and the fundamental operations
of the irrational
for the real
numbers, are defined
in
such a way that this rational-real number

it is usually denoted by the
has no properties distinct from those of m, and
same symbol.
We
have extended our

6. Relations of Magnitude for Real Numbers.
must now arrange the system of real numbers
conception of number.
in order
i.e. we must say when two numbers are equal or unequal to,
We
greater or less than, each other.
In this place we need only deal with cases where at least one of the
numbers
An
is
irrational.
irrational
number
is
never equal to a rational number.
They are
always different or unequal.

Next, in 5, we have seen that the irrational number given by the section
(A, B) is said to be greater than the rational number m, when m is a member
of the lower class A, and that the rational number m is said to be greater than
the irrational
1
number given by the
section (A, B),
when
is
member
of
the upper class B.
Two
numbers are equal, when they are both given by the same
They are different or unequal, when they are given by different
irrational
section.
sections.
The
number a given by the

number a' given by the section
irrational
irrational
numbers
of the class B'.
a certain
number
Now
of the class
section (A, B)
(A', B'),
when
is greater than the

the class
contains
has no greatest number. But

belongs to the class B', all the numbers of
the class
if
number m could correspond to two sections: the one named in

and that in which the lower class A contains all the rational numbers
less than m, and the upper class B, in and all the rational numbers
greater than m.
To save ambiguity, one of these sections only must he chosen.
*The
'the text,
rational
22
number also belong to B'. The class A thus contains an

number of members of the class B', when a>a'.
a real number a is greater than another real number a', then a' is less
greater than this

infinite
If
than
a.
>, =, < is used in dealing with real

numbers as in dealing with rational numbers.
The real number ft is said to lie between the real numbers a and y, when
one of them is greater than ft and the other less.
With these definitions the system of real numbers is ordered. If we have
two different real numbers, one of them is greater than the other and if we
have three real numbers such that a > ft and ft > y, then a > y.
These definitions can be simplified when the rational numbers themselves
It will be observed that the notation
are given
Between any two
7.
numbers.
of rational
different rational
numbers there
and
a, a' there
are
are rational, the property is known.

and a' irrational, let us assume a>a'.
a'
is
an
infinite
number
Between any two different real numbers

of rational number*.
If a
5.
similar property holds for the system of real
numbers, as will now be shown

(I)
end of
sections, as explained at the
by
If a is rational
an
Let
infinite
a'
number
be given by
the section (A', B'). Then the rational number a is a member of the upper
class B', and B' has no least number.
Therefore an infinite number of
members
of the class B' are less than
5 that there are
of
and
less
than
an
infinite
number
than the rational number
the case
when
There remains the case when a and
(II)
number a
is
a!
a'
are both irrational.
by the
section
(A', B').
greater
Let a be
Also let
A of a contains an infinite number of members

and these numbers are less than a and greater than a.
class
A similar proof applies to

The
the irrational
a.
given by the section (A, B) and
class B' of a'
from the definitions
numbers greater than a
a.
A similar proof applies to
a>o/.
Then the
It follows
a.
of rational
the case
when
a<
of the
a'.
which has just been proved can be made more general

Between an// two different real numbers there are an in-finite number of
result
irrational numbers.
be the two given numbers, and suppose a<a'.

Take any two rational numbers ft and ft', such that a<ft<ft'< a. If we
can show that between ft and ft' there must be an irrational number, the
theorem is established.
Let i be an irrational number. If this does not lie between ft and ft', by
For we
adding to it a suitable rational number we can make it do so.
can find two rational numbers m, n, such that m<i<n and (n m) is
The number ft-m + i is irrational, and lies between
less than (ft - ft).
Let
a, a'
ft
and
ft'.
23
We
shall now prove a very imof

the
of
real * numbers, which will be
portant property
system
used frequently in the pages which follow.
Dedekind's Theorem.
8.
the system of real numbers is divided into two classes

B, in such a ^vay that
(i) each class contains at least one number,
If
and
number belongs to one class or the other,

number in the lower class A is less than every
number in the uyyper class B
then there is a number a such that
every number less than a belongs to the lower class A, and
every number greater than a belongs to the upper class B.
The separating number a itself may belong to either class.
(ii)
every
(iii)
every
A and B.
two classes e.g. A' and B' such that every
rational number is in one class or the other, and the numbers in
the lower class A' are all less than the numbers in the upper
Consider the rational numbers in
These form
class B'.
As we have seen
(i)
in
three cases, and only three, can arise.
4,
The lower class A' can have a greatest number

upper class B' no smallest number.
The
rational
number
is
the
number
b,
and
a of the theorem.
clear that every real number a less than

class A, since in is a member of this class.
it is
number
the
For
belongs to the
Also every real
greater than m, belongs to the class B.
This
is
evident
b is rational, since b then belongs to the class B', and B' is part
If b is irrational, we can take a rational number n between
of B.
if
m and
(ii)
b.
Then n belongs
to B.
and therefore
b does so also.
The upper class B' can have a smallest number

the lower class A' no greatest number.
It follows, as above, that the rational

a of our theorem.
number
is
the
and
number
The lower class A' can have no greatest number and the
upper class B' no smallest number.
Let
be the irrational number defined by this section (A', B').
(iii)
'
Every
rational
* It will
property.
number
less
than
belongs to the class A,
be observed that the .system of rational numbers does not possess this
24.
and every
rational
number greater than
belongs
to
the
class B.
We
have yet to show that every irrational number less than in

belongs to the class A, and every irrational number greater than
in to the class B.
But this follows at once from 6. For if m is an irrational

number less than m, we know that there are rational numbers
between
does so
and
in'.
These belong to the
class A,
and therefore
also.
similar
argument applies
to the case
when m'>m.
In the above discussion the separating number a belongs to

the lower class, and is rational, in case (i); it belongs to the
upper class, and is again rational, in case

and may belong to either class, in case (iii).
9.
now
been
The Linear Continuum.
(ii);
it
Dedekind's Axiom.
is
irrational,
We
return
to the straight line L of 3, in which a definite point

has
taken as origin and a definite segment as the unit of length.
We
have seen how to effect a correspondence between the

The
rational numbers and the "rational points" of this line.
"
rational points
"
are the ends of segments obtained

o
by marking
FIG. 2.
off
on the
from
line lengths equal to multiples or sub-multiples
of the unit segment,
and the numbers are the measures of the
corresponding segments.
Let OA be a segment incommensurable with the unit segment.
The point A divides the rational points of the line into two
classes,
such that
all
the points of the lower class are to the left
The lower class has no last

first point.
no
point,
We then say that A is an irrational point of the line, and
that the measure of the segment OA is the irrational number
defined by this section of the rational numbers.
of all the points of the upper class.
and the upper
Thus
to
any point
class
of the line
corresponds a real number, and
to different points of the line correspond different real numbers.

There remains the question To evert/ real number does there
correspond a point of the line
'i

For
all
rational
numbers we can answer the question
When we
affirmative.
25
the irrational
turn to
in the
numbers, the
question amounts to this // all the rational points of the line

are divided into two classes, a lower and an upper, so that the
:
lower class has no last point

is there one,
this
and only
separation
one,
and
the
point on
upper
the line
no first point,
which brings about
class
The existence of such a point on the line cannot be proved.

The assumption that there is one, and only one, for every section
of the rational points is nothing less than an axiom by means of
which we assign its continuity to the line.
This assumption is Dedekind's Axiom of Continuity for the
In adopting it we may now say that to every point P of
the line corresponds a number, rational or irrational, the
line.
the segment OP, and that to every real number

a
corresponds
point P of the line, such that the measure of OP
is that number.
The correspondence between the points of the line L (the linear
continuum] and the system of real numbers (the arithmetical
continuum) is now perfect. The points can be taken as the
images of the numbers, and the numbers as the signs of the
measure of
In consequence of this perfect correspondence, we may,

points.
in future, use the terms number and point in this connection as
identical.
The Development of the System
10.
is
instructive to see
we have
how
of Real
Numbers.
It
the idea of the system of real numbers,
The irrational numbers,

it, has grown.*
in
do
modern
arithmetical theory to the
belonging
realm of arithmetic, arose from the geometrical problems which
required their aid. They appeared first as an expression for the
ratios of incommensurable pairs of lines.
In this sense the Fifth
Book of Euclid, in which the general theory of Ratio is
developed, and the Tenth Book, which deals with Incommensurable Magnitudes, may be taken as the starting point of
the theory. But the irrationalities which Euclid examines are
only definite cases of the ratios of incommensurable lines, such
as
described
as they
Enc.
Cf
d.
" Irrationalzahlen
Pringsheim,
math. Wis*., Bd. L, Tl. I., p. 49
u.
Konvergenz unendlicher Prozesse,"
et seq.,
1898.
26
as
may
be obtained with the aid of ruler and compass; that
to say, they depend on square roots alone.

The idea that
the ratio of any two such incommensurable lines determined
is
a definite (irrational) number did not occur to him, nor to any

of the mathematicians of that age.
Although there are traces in the writings of at least one of
the mathematicians of the sixteenth century of the idea that
every irrational number, just as much as every rational number,

possesses a determinate and unique place in the ordered sequence
of numbers, these irrational numbers were still 'considered to
from certain cases of evolution, a limitation which is

due
to the commanding position of Euclid's methods in
partly
and
Geometry,
partly to the belief that the problem of finding
the nth root of an integer, which lies between the n^ powers of
arise only
two consecutive
integers,
was the only problem whose
solution
could not be obtained in terms of rational numbers.
The introduction
of the
methods of Coordinate Geometry by
Descartes in 1637, and the discovery of the Infinitesimal Calculus

by Leibnitz and Newton in 1G84-7, made mathematicians regard
this question in
another
light, since
the applicability of
number
magnitude is a fundamental postulate of Coordinate

"
The view now prevailed that number and quantity
Geometry.
were the objects of mathematical investigation, and that the two
were so similar as not to require careful separation. Thus
to spatial
number was applied
to quantity without any hesitation, and,

where
existing numbers were found inadequate to
conversely,
new
ones were created on the sole ground that
measurement,
have a numerical measure."
must
every quantity
It was reserved for the mathematicians of the nineteenth
century notably Weierstrass, Cantor, Dedekind and Heine to
*
Until their writings

establish the theory on a proper basis.
appeared, a number was looked upon as an expression for the
the measurement of a line by another which was

regarded as the unit of length. To every segment, or, with the
natural modification, to every point, of a line corresponded a
result of
definite
number, which was either rational or irrational
the term irrational
number was meant a number
*Cf. Russell, Principles of Mathematics,
Cli.
XIX.,
and by
by an
defined
p. 417,

infinite set
of arithmetical operations
(e.g.
27
infinite decimals or
The
justification for regarding such an

unending sequence of rational numbers as a definite number was
considered to be the fact that this system was obtained as the
continued fractions).
equivalent of a given segment by the aid of the same methods of

measurement as those which gave a definite rational number for
other segments. However it does not in any way follow from
this that, conversely, any arbitrarily given arithmetical representation of this kind can be regarded in the above sense as an
irrational
number that
;
is
to say, that
we can
consider as evident
the existence of a segment which would produce by suitable

measurement the given arithmetical representation. Cantor *
has the credit of
first pointing out that the assumption that a

segment must correspond to every such sequence is
neither self-evident nor does it admit of proof, but involves an
actual axiom of Geometry.
Almost at the same time Dedekind
si lowed that the axiom in
question (or more exactly one which is
to
first
it)
equivalent
gave a meaning, which we can comprehend,
to that property which, so far without any sufficient definition,
had been spoken of as the continuity of the line.
definite
To make the theory of number independent of any geometrical

axiom and to place it upon a basis entirely independent of
measurable magnitude was the object of the arithmetical theories
associated with the names of Weierstrass, Dedekind and Cantor.
The theory of Dedekind has been followed in the previous pages.
Those of Weierstrass and Cantor, which regard irrational
numbers as the limits of convergent sequences, may be deduced
from that of Dekekind. In all these theories irrational numbers
appear as new numbers, to each of which a definite place in
the domain of rational numbers is assigned, and with which we
can operate according to definite
of arithmetic for these
rules.
The ordinary operations
numbers are defined in such a way as to
be in agreement w4th the ordinary operations upon the rational

numbers. They can be used for the representation of definite
quantities, and to them can be ascribed definite quantities, according to the axiom of continuity to which we have already
referred.
Math. Ann., Leipzig,
5, p.
1'27,
1872.
28
KEFERENCES.
BROMWICH, Theory of
Infinite Series,
DEDEKIND,
und
Stetigkeit
App. L, London, 1908.
irrationale Zahlen, Braunschweig, 1872 (English
translation in Dedekind's Essays on Number, Chicago, 1901).
DE LA VALLEE
d Analyse,
1
POUSSIN, Cours
T.
I.
(3 ed.), Introduction,
1,
Paris, 1914.
DINI, Fondamenti per la Theorica delle Funzioni di Variabili Reali,
1-9,
Pisa, 1878.
GOURSAT, Cours
d Analyse,
T.
I.
<3
ed.),
HOBSON, Theory of Functions of a Real
Ch.
I.,
Paris, 1917.
Variable, Ch.
PRINGSHEIM, Vorlesungen iiber ZahlenAbsch. L, Kap. I.-IIL, Leipzig, 1916.
I.,
Cambridge, 1907^
und Funktionenlehre, Bd.
I.,
RUSSELL, Principles of Mathematics, Vol. L, Ch. XXXIV., Cambridge, 1903.

STOLZ u. GMEINER, Theoretische Arithmetik, Abth. II., Absch. VII., Leipzig,
1902.
TANNERY, Introduction a
la The'oric des Fonctions, T.
I.
(2
ed.),
Ch.
I.,
Paris, 1914.
And
PRINGSHEIM,
Knc.
d.
" Irrationalzahlen
u.
Konvergenz unendliclier Prozesse
math. Wiss., Bd. L, Tl. L, Leipzig, 1898.
"
CHAPTER
INFINITE SEQUENCES
II
AND
SERIES
Aggregates. We are accustomed to speak of the

positive integral numbers, the prime numbers, the integers which
These are all examples of infinite sets
are perfect squares, etc.
of numbers or sets which have more than a finite number of terms.
In mathematical language they are termed aggregates, and the
theory of such infinite aggregates forms an important branch of
11.
Infinite
modern pure mathematics.*

The terms of an aggregate are
all
different.
Their number
In the latter case the aggregates are

may
called
infinite
usually
aggregates, but sometimes we shall refer to
them simply as aggregates. After the discussion in the previous
chapter, there will be no confusion if we speak of an aggregate
be
finite or infinite.
an aggregate of numbers. The

We associate with each number the
.two notions are identical.
point of which it is the abscissa. It may happen that, however
of points on a line instead of
far
on.
we go along
the
In this case
line,
there are points of the aggregate further
we say that
it
extends to infinity.
An aggregate
the right, or bounded above, when

there is no point of it to the right of some fixed point. It is
said to be bounded on the left, or bounded below, when there
said
is
is
to be
no point of
bounded on
it
to the left of
some fixed
point.
The aggregate
Cantor may* be taken as the founder of this theory, which the Germans call
Menye-Lehre. In a series of papers published from 1870 onward he showed its
importance in the Theory of Functions of a Real Variable, and especially in
the rigorous discussion of the conditions for the development of an arbitrary
function in trigonometric series.
'
Reference may be made to the standard treatise on the subject by

Grace Chisholm Young, Theory of Set* of Points, Cambridge, 1906.
29
W.
H. and
INFINITE SEQUENCES
30
AND SERIES
numbers greater than zero is bounded on the left.

The aggregate of rational numbers less than zero is bounded on
the right.
The aggregate of real positive numbers less than
unity is bounded above and below in such a case we simply
say that it is bounded. The aggregate of integral numbers is
of rational
unbounded.
The Upper and Lower Bounds of an Aggregate. When

aggregate (E)* is bounded on the rig Jit, there is a number
12.
an
which possesses
the following properties
no. number of (E) is greater than

hoivever small
We
ilie
positive
of (E) greater than

can arrange all the
number
may
be, tJiere is
a number
Me.
numbers
real
in
two
classes,
and B,
relative to the aggregate.

number x will be put in the class
if one or more numbers of (E) are
greater than x. It will be put
in the class B if no number of (E) is greater than x.
Since the
classes,
is bounded on the right, there are members of both

and any number of the class A is smaller than any
number
of the class B.
aggregate
M
M
By Dedekind's Theorem (8) there is a number

separating
the two classes, such that every number less than
belongs to
to the class B.
the class A, and every number greater than
We
shall
In the
now show
For suppose there
there
is
of our theorem.
number
number of (E) greater than M.
such a number M+h (A,>0). Then the
that this
first place,
is
the
is 110
number M-^-^h, which is also greater than J/, would belong to

the class A, and
would not separate the two classes A and B.
In the second place, whatever the positive number e may be, the
It follows from the w ay
number
e belongs to the class A.
in which the class A is defined that there is at least one number
of (E) greater
than
This number
is
e.
called the
upper bound of the aggregate
(E).
may belong to the aggregate. This occurs when the aggregate

But when the aggregate
contains a finite number of terms.
It
of terms, the upper bound need not

example, consider the rational numbers whose
contains an infinite
belong to
*
it.
This notation
term ensemble.
For
is
number
convenient, the letter
being the
first letter of
the French
AND SERIES
INFINITE SEQUENCES
31
squares are not greater than 2. This aggregate is bounded on

the right, its upper bound being the irrational number ^2, which
does not belong to the aggregate.
On the
other hand, the aggre-
gate of real numbers whose squares are not greater than 2 is

But
also bounded on the right, and has the same upper bound.
^/2 belongs to this aggregate.

If the upper bound
of the aggregate (E) does not belong to
an
number of terms of the aggregate
must
be
infinite
there
it,
M and M
however small the positive number e may

If there were only a finite number of such terms, there
be.
would be no term of (E) between the greatest of them and M,
which is contrary to our hypothesis.
It can be shown in the same way that when an aggregate (E)
is bounded on the left, there is a number in possessing the followbetween
ing properties
e,
no number of (E)
smaller than
is
however small the positive number

+ e.
of (E) less than
The number
defined in this
As above,
of the aggregate (E).
way
it
may be, there is

is
a number
called the lower
may, or
may
not,
bound
belong to
the aggregate when it has an infinite number of terms. But

when the aggregate has only a finite number of terms it must
belong to
13.
it.
Limiting Points of an Aggregate.

l
2'
3"" n"
Consider the aggregate
infinite number of points of this aggregate in any

however small, extending from the origin to the right.
Such a point round which an infinite number of points of an
There are an
interval,
* of the
called a limiting point
aggregate.
More definitely, a will be a limiting point of the aggregate (E)
be, there is in (E}
if, however small the positive number e
aggregate cluster,
is
may
a point other than a whose distance from a
is less
than
e.
If
there be one such point within the interval (a e, a e), there

will be an infinite number, since, if there were
only n of them,
*
Sometimes the term point of condensation
^'accumulation
is used
French, point-limite, point
(Jerman, Hdv/wngftpunkt^ Verdichtunyspnnkt.
;
INFINITE SEQUENCES
32
AND SERIES
and u n were the nearest to a, there would not be in (E) a point

other than a whose distance from a was less than |a aj.* In
that case a would not be a limiting point, contrary to our
hypothesis.
An
may have more than one limiting point. The

numbers between zero and unity form an aggregate with
an infinite number of limiting points, since every point of the
segment (0, 1) is a limiting point. It will be noticed that some
of the limiting points of this aggregate belong to it, and some,
namely the irrational points of the segment and its end-points,
do not.
aggregate
rational
In the example at the beginning of this section,
111
X 3"" n"
the lower bound, zero, is a limiting point, and does not belong to
the aggregate. The upper bound, unity, belongs to the aggregate,
and is not a limiting point.
The
which
set of real
is
numbers from
to
1,
inclusive, is
an aggregate
identical with its limiting points.
Weierstrass's Theorem, An infinite aggregate, bounded

above and beloiv, lias at least one limiting point.
14.
Let the infinite aggregate (E) be bounded, and have

and lower bounds.
M and m
for its upper
We
can arrange
the aggregate (E).

class
A when
It will
x.
Since
an
all
the real numbers in two classes relative to
infinite
number x will be said to belong to the

number of terms of (E) are greater than
be said to belong to the class
m belongs to
the class
and
in the contrary case.
M to the class B, there are
members of both classes. Also any number in the class A is less

than any number in the class B.
separating the
By Dedekind's Theorem, there is a number
JUL
two
classes.
However small the
e
positive number e may be, /u.
to
the
class
B.
Thus
the
C
+
belongs to the class A, and
interval (jm
e, M + e) contains an infinite number of terms of the
JUL
aggregate.
*
It isûsual to denote the difference
between two
numbers a and b, taken

modulus of (-?>). With
real
or
positive, by |a-6|, and to call it the absolute ralne
this notation |# + 2/|^|a'| + |y|, and |#y|
|#|ty|-
INFINITE SEQUENCES
AND SERIES
is a limiting point.
Hence
As will be seen from the example
33
/UL
coincide with
An
of
13, this
M or m.
infinite aggregate,
point
when unbounded, need not have a
may
limit-
ing point e.g. the set of integers, positive or negative. But if

the aggregate has an infinite number of points in an interval of
;
finite length,
then
must have
it
at least one limiting point.
Convergent Sequences. We speak of an infinite sequence

numbers
u^ u^ u^ ... u^ ...
when some law is given according to which the general term un
15.
of
may
be written down.
u lt u2
and
The sequence
UB
...
have the limit A, when, by

the
difference between A and u n
indefinitely increasing n,
and
becomes,
thereafter remains, as small as we please.
said
is
to
be convergent
to
so fundamental that
it is well to put it more

The sequence is said to be convergent and
when, any positive number e having been
as small as we please, there is a positive integer v such that
This property
is
precisely, as follows
to have the limit A,
:
cJiosen,
\A-u n
<
e,
provided that n
v.
For example, the sequence

1
2'
3'
has the limit zero, since l/n

greater than 1/e.
The notation that
is
I
**:
less
is
employed
than
for all values of
in this connection is
and we say that as n tends to infinity, u n has the limit A*

The letter e is usually employed to denote an arbitrarily small
positive number, as in the above definition of convergence to a
limit as n tends to infinity.
Strictly speaking, the words as
small as we please are unnecessary in the definition, but they are
inserted as making clearer the property that is being defined.
We shall very frequently have to employ the form of words
which occurs
*
in this definition, or
The phrase " u n tends

c. I
to the limit
A
c
words analogous
as n tends to infinity
"
to them,
is
also used,
and
AND SERIES
INFINITE SEQUENCES
34
the beginner is advised to make himself familiar with them by

formally testing whether the following sequences are convergent
or not
:
(a)
\,
-1,
(1) 1,
A
If
1,....
1,
1,....
+ 1,
(c)
1,
(d)
1,
-1,
+1+1
1,
-1,....
sequence cannot converge to two distinct limits
this
were
possible, let
<
'
....
Then there
and B.
are
only
a finite number of terms of the sequence outside the interval

This
e, A + e), since the sequence converges to the value A.
(A
contradicts the statement that the sequence has also the limit B,
for we would only have a finite number of terms in the interval
of the
same length with
as centre.
The
application of the test of convergency contained in the

Thus it will
definition involves the knowledge of the limit A.
frequently be impossible to use it. The required criterion for

the convergence of a sequence, when we are not simply asked to
test
whether a given number is or is not the limit,

fundamental general principle of convergence
in the
and
necessary
limit to the sequence
is that
contained
*
sufficient condition /or the existence of a
a positive integer
that \u n+p
v exists sucli
small as we please when n
More exactly
is
:
=^
v,
for every
un
becomes as
\
positive, integer p.
A necessary and sufficient condition for the existence of a

limit to the sequence
...
M!
is that, if
we
any
positive
please, there shall be
\u n+p
un
<
e,
M-2,
number
has been chosen, as small as

a positive integer v such tJtaf
e
when n^v, for
every positive integer p.
* This is one of the most

important theorems of analysis. In the words of
Pringsheim, "Dieser Satz, mit seiner Ubertragung auf beliebige (z.B. stetige)
Zahlenmengen von du Bois-Reymond als das 'attgemeine Convergenzprinzip'
bezeichnet (Allg. Funct.-Theorie, pp. 6, 260) istder eigentliche Fundamental *atz

der gesamten Analysis und sollte mit geniigender Betonung seines fundamental
Characters an der Spit/e jedes rationellen Lehrbuches der Analysis stehen,
loc, tit.,
Enc. d. math. Wiss.
p. 66.
INFINITE SEQUENCES
We
AND SERIES
show that the condition
shall first of all
35
is
necessary.
the sequence converges, this condition
secondly
that, if this condition is satisfied, the sequence converges; in
other words, the condition is sufficient.
is satisfied;
if
i.e.
The condition is necessary.

Let the sequence converge to the limit A.
Having chosen the arbitrary positive number e, then take
We know that there is a positive integer v such that
(i)
A.
I
But
when n 2
Je,
u n \^\ u n+p
un+p
<
v.
- un ) = (un+p -A) + (A- u n
(u n+p
Therefore
un
<
n ^ v>
if
-f
ir
ie.
un
).
Je,
for every ppsitive integer p,
<
(ii)
The condition
is sufficient.
We
must examine two cases first, when the sequence contains

an infinite number of terms equal to one another; second, when
it
does not.
(a) Let there be
Then,
\\u n +.t
an
infinite
number
of terms equal to A.
if
u.n
<
e,
when n^v, and p
we may take u n+p = A

I
is
any
positive integer,
some value of p, and we have
for
A un
<
e,
when n ^ v.
Therefore the sequence converges, and has A for its limit.

Let there be only a finite number of terms equal to one
(/;)
another.
Having chosen the arbitrary
We know
Therefore
all
number
a positive integer
\un
lie
is
un <--, when n^-N,

follows that we have
\un+p
It
that there
positive
UN
<
le,
e,
then take
?,e.
N such that
for every positive integer p.
when n^-N.
the terms of the sequence
within the interval whose end-points are
u^ê and
AND SERIES
INFINITE SEQUENCES
36
infinite number of distinct terms in tin's

Otherwise we would have an infinite number of
terms equal to one another.
Consider the infinite aggregate (E) formed by the distinct
-.
-.
terms in
....
There must be an
sequence.
1*2,
ft/j,
Tliis
aggregate
limiting point
Ivj,
bounded and must have at least one

within, or at an end of, the above interval.
is
(Cf.14.)
There cannot be another limiting point A', for if there were,
we could choose e equal to |- \A A'\ say, and the formula
\u n+p
shows that
u n <e, when n~v,
the terms of the sequence
all
U-^
except a
J
\A
for every positive integer p,
finite
A'\.
number, would
This
is
U.2 >
impossible
>
within an interval of length

A, A' are limiting points of the
lie
if
aggregate.
Thus the aggregate (E) has one and only one limiting point A.
We shall now show that the sequence
u lt
converges to
We
Therefore
|
...
n tends to oc
u n A = (u n ?%) 4- (ux A
u.n
A\^\ u n u x + u y A
as
have
us
u.2)
.
<
-i<?
).
when n ^ N,
when n = N.
>
e,
A for its
We have therefore proved this theorem
A necessary and sufficient condition for the
Thus the sequence converges, and has
limit.
the sequence
u^
is that, to the
u^
arbitrary positive
v such that
%j
number e,
convergence of
...
there shall correspond
a positive integer
\Un+p
It
is
Un\
<
when n^v,for
easy to show that the above condition
following
In order that
the
seuence
may
*be
replaced by the
AND SERIES
INFINITE SEQUENCES
37
converge, it is necessary and sufficient that, to the arbitrary positive

shall correspond a positive integer n such that
e, there
may
number
\u n+p
- un
< e for every positive integer p.
the sequence converges, this condition is satisfied by n = v.

Further, if this condition is satisfied, and is an arbitrary positive number,
to the number
there corresponds a positive integer n such that
It
is
clear that
if
But
un
\u n +p
u n +tf - Un +ts
<
u n+p - un +
'f
<
u, l+J/
- un
when p', p"

Therefore the condition in the text
i,
is
are any positive integers.

and the sequence
also satisfied,
converges.
16.
and
Divergent
Sequences.*
Oscillatory
u^
sequence
u^
u^
When
the
...
does not converge, several different cases arise.
may have
the terms
the property that

if any
positive number A, however large, is chosen, there is
a positive integer v such that
(i)
In the
first place,
u n >A, when n = v.
In this case
diverges to
we say
+ oo
that the sequence
and we write
is
divergent, and that
it
this
Lt u n =
+ oo
/.
(ii)
it*
In the second place, the terms
any negative number
be, there is
a positive integer
un
In this case
to
diverges
we say
oo
to
+ oo
or to
is
oc
A may
is
divergent, and that
it
this
un =
>-X
of a sequence
when n
the property that
when n = v.
that the sequence
and we write
Lt
The terms
i>
<A,
II
value,
may have
chosen, 'however large

such that
is
may
all
be very large in absolute

may not diverge
very large, yet the sequence

.
sufficient illustration of this is

n
is (
l) n.
given by
the sequence whose general term
*
In the first edition of this book, the term divergent was used as meaning
merely not convergent. In this edition the term is applied only to the case of
oc
and sequences which oscillate infinitely are placed
divergence to + oo or to
,
among the
oscillatory sequences.
INFINITE SEQUENCES
38
After some value of
AND SERIES
the terms must
all
have the same
sign,
the sequence is to diverge to + GO or to

cc
the sign being
in
the
first
and
in
the
second.
alternative,
positive
negative
if
When
(iii)
+x
or
the sequence does not converge,

oo
to
it is
said
and does not diverge
to oscillate.
An
oscillatory sequence is said to oscillate finitely, if there is

such that \u n \<A, for all values of n;
positive number
and
it is
said
to oscillate infinitely
when
there is
For example, the sequence whose general term

finitely; the sequence whose general term is
no such number.
is (
l)
n
l)
n oscillates
oscillates
infinitely.
We may
distinguish between convergent and divergent sethat a convergent sequence has a finite limit,
saying
by
quences
i.e. Lt u n =A, where A is a definite number
a divergent sequence
;
><*>
has an infinite limit,
i.e.
Lt
un = + oc
or Lt
un =
oo
n >oo
H >co
must be remembered that the symbol co and the terms

and tend to infinity, have purely conventional
meanings. There is no number infinity. Phrases in which the
term is used have only a meaning for us when we have previously,
But
it
infinite, infinity
by
definition, attached a
meaning
to them.
say that n tends to infinity, we are using a short

and convenient phrase to express the fact that n assumes an
endless series of values which eventually become and remain
When we
So far we
greater than any arbitrary (large) positive number.
have supposed n, in this connection, to advance 'through integral
This restriction will be removed later.
values only.
similar remark applies to the phrases divergence to
IQ
oc
and oscillating
infinitely, as well
as
-f oo
or
to our earlier
an infinite number, infinite sequence and

In each case a definite meaning has
aggregates.
been attached to the term, and it is employed only with that
use of the terms
infinite
meaning.
much of our work might be simplified by the

oo
and by assuming the
new numbers +00,
the
line which we have
corresponding points upon
It is true that
introduction of
existence of
used as the domain of the numbers. But the creation of these

numbers, and the introduction of these points, would be a matter
for separate definition.
INFINITE SEQUENCES
Monotonic Sequences.
17.
AND SERIES
39
// the terms of the sequence
satisfy either of the following relations
or
^ u = u.
i^
= un
. . .
the sequence is said to be monotonic.
In the
first case,
the terms never decrease, and the sequence

in the second case, the
be called monotonic increasing

terms never increase, and the sequence
may
may
be called monotonic
decreasing*
Obviously,
when we
are concerned with the convergence or
divergence of a sequence, the monotonic property,

need not enter till after a certain stage.
The
such
if
exist,
convergence or divergence are extremely simple

in the case of monotonic sequences.
tests for
^^
the sequence
u ^ u^
is monotonic increasing, and its terms are all less than some fixed
number B, the sequence is convergent and has for its limit a
number ft such that u n = ft = B for every positive integer n.
If
aggregate formed by distinct terms of the

bounded
sequence.
by u^ on the left and by B on the
Thus
it
must
have
an upper bound ft (cf. 12) equal to or
right.
less than B, and, however small the positive number e may be
there will be a term of the sequence greater than ft e.
Let this term be u v
Then all the terms after u v .\ are to the
If any of them coincide
e and not to the right of ft.
right of /3
with ft, from that stage on the terms must be equal.
Thus we have shown that
Consider
the
It is
|/3
un |<e, when
and therefore the sequence
The following
test
may
n^Lv,
convergent and has ft for

be proved in the same way
is
its limit.
u lt u 2 u^, ...
If the sequence
monotonic decreasing, and its terms are
,
is
fixed
number A, then
*The words
words
than some
and has for
and steadily decreasing are sometimes employed

and when none of the terms of the sequence are equal, the
steadily increasing
in this connection,
all greater
the sequence is convergent
in the stricter sense are added.
its
AND SERIES
INFINITE SEQUENCES
40
limit a
number a such
un ~a = A
that
for every positive
integer n.
It
an immediate consequence of these theorems that a
is
monotoiiic sequence either tends to a limit or diverges to
to
Let A^ A.2
18.
+ x>
or
A3
an
he
...
infinite set of intervals, each lying
entirely within the preceding, or lying within it and having

with it a common end-point ; also let the length of
n tend to
zero as
tends
Then
to infinity.
there is one,
and only one,

as an internal
point which belongs to all the intervals, either

point of all, or from and after a definite stage, as a
end-point of all.
An
Let the representative interval
common
be given by
a*3Z*..
a^ = a = a
&! = b = b
Then we have
and
Thus the sequence
a,
a.2
and a n = a
...
as
,...(1)
-J
an
a
for every positive integer
a,
of end-points
î,
has a limit, say
a
FIG.
17).
bn
bz
b,
3.
Also the sequence of end-points

&!,
has a
(
limit,
say
ft,
and
b2
6n
...(2)
b,...,
/8
for
every positive integer
17).
Now
less
it is
than
clear that,
under the given conditions,
/3
cannot be
a.
Therefore, for every value of n,

bn
Lt
But
It follows that a
-an > /3
(&
aw )
0.
= 0.
.*
This result also follows at once from the fact that,

- &) = a 26, Theorem I. )
(Cf
(a,,
j8.
then Lt
if
Lt a n
and Lt
b n = (3,
AND SERIES
INFINITE SEQUENCES
common
Therefore this
satisfies
the sequences (1) and (2)
limit of
the inequalities
a n = a = b n for every positive integer

and thus belongs to all the intervals.
Further, no other point (e.g. y) can
satisfy
n,
e& M
= y~&M
values of n.
all
Since
we would have
same time
at the
Lt a n = y
Lt b n =
and
n>v>
n^-v)
which
is
Infinite Series.
Let
u lt
be
an
and
let
be
formed.
8ly
S2
If
infinite sequence,
the sequence
convergent
and has
infinite series
and
It
tfyis
y,
y = a.
impossible unless
The Sum of an
19.
is
41
uz
u3
...
the successive
Ss
...
the limit S, then
ni+
sums
^+^+
S
.
sum
is called the
of the
series is said to be convergent.
must be carefully noted that what we
infinite series is
limit, the limit of the
call
sum
of
the
sum
n terms
of the
of
tends to infinity. Thus we have no right to assume without

proof that familiar properties of finite sums are necessarily true
as
for
sums such as
$.
When Lt $ = + oc
or Lt
Sn =
oo
infinite series is divergent, or diverges to
may
If
we
shall say that the
n >co
/<,->ao
<x>
or
as the case
be.
S n does not tend
to a limit, or to
terms in
16.
+ cc
or to
co
then
it
according to the definitions of these

In this case we shall say that the series oscillates
oscillates finitely or infinitely
finitely or infinitely.*
* Cf.
footnote, p. 37.
INFINITE SEQUENCES
42
The conditions obtained
AND SERIES
15 for the convergence of a

sequence allow us to state the criteria for the convergence of the
in
series in either of the following
ways
(i) The seri& converges and has S for its sum, if, any
ositive number e having been chosen, as small as we please,
there is a positive integer v such that
I
(ii)
necessary
$n <C
when n^lv.
e,
and
sufficient condition for the convergence
if any positive number e has been chosen,

small as we please, there shall be a positive integer v such
of the series is
"as
tJiat,
that
Sn
\Sn+p
<
e,
when n^v, for
every positive integer
It is clear that, if the series converges,
Lt
u n = 0.
This
p*
is
con-
It is a necessary condition for

not a sufficient condition e.g. the series
tained in the second criterion.
convergence, but
is
it is
Lt u n = 0.
divergent, though
If
we denote
un+i
un+z +
~^~
+ UH+P
or
>
SK+P
&n by p Rn
>
the above necessary and sufficient condition for convergence of

the -series may be written
\
Rn
<
e,
when n^v,
u^ -f u 2
Again, if the series

converges
and has S for
its
converges
and has S
for
)t.
Lt
Therefore
P> ^
*As remarked
number
in
. . .
sum.
.
Sn+p = S.
(p
R n = S-Sn
)
this condition can be replaced by: To

must correspond a positive integer n such *ht
Sn+p - S n
e for every positive integer p.
15,
there
|
>.
the series
sum,
its
fixed, it is clear that
Lt
positive
+ u% -f
$ p = Sn + p R n
For we have
Also keeping
for every positive integer
<
the arbitrary
AND SERIES
INFINITE SEQUENCES
Tims
if
we
write
Rn
for the
sum
of the series
Un+l~^~ U n+z+
S=Sn -\-R n
we have
The
first criterion
"
now
for convergence can

\ftn\
<C
G>
43
be put in the form
when n ^ v.
R n is usually called the remainder of

p R n a partial remainder.
the series after
terms,
and
20.
Series
whose Terms are
all Positive.
Let
be
tJtis
series
The sum of n terms of

or
it
limit,
diverges to + oc
the
successive
sums
positive,
whose terms are
series either tends to
Since the terms are
all
all positive.
|form a monotonic increasing sequence, and the theorem stated

above follows from 17.
When a
series
whose terms are
all positive is convergent, the
we obtain when we take the terms in any order we

also convergent and has the same sum.
series
is
This change of the order of the terms
is
please
to be such that
there will be a one-one correspondence between the terms of the

The term in any assigned place in the
old series and the new.
one series is to have a definite place in the other.
Let
Then the aggregate
Su
( f/),
&|,
which corresponds
8ZJ
$3
...
to the sequence
its upper bound 8 is the sum of the series.

Let (/') be the corresponding aggregate for the series obtained by taking the terms in any order we please, on the
understanding we have explained above.
Every number in
is
bounded and
INFINITE SEQUENCES
44
AND SERIES
than S. In addition, if A is any number less than

must be a number of (U) greater than A, and a fortiori
a number of (t^) greater than A. The aggregate (U') is thus
bounded on the right, and its upper bound is S. The sum of
the new series is therefore the same as the sum of the old.
(V)
is lens,
S, there
It follows that if the series
whose terms are
changing
all positive, diverges, the series

the order of the terms must also diverge.
we obtain by
The following theorems may be proved at once by the use

(19):
of the second condition for convergence

If. the series
is
convergent
from
and
all its
terms are positive, the series we obtain
this, either
only a part of its terms,

or (2) by replacing certain of its terms by others, eithei
positive or zero, which are respectively equal or in
(1) by keeping
ferior to them,
or (3) by changing the signs of some of
its
terms,
are also convergent.

21.
Absolute and Conditional Convergence.
metrical series, whose properties

to the class of series
we
The trigono-
shall investigate later, belong

is due to the presence of
whose convergence
both positive and negative terms, in the sense that the series
would diverge if all the terms were taken with the same sign.
series
and negative terms is said to be

when the series in which all the terms
ivith positive
absolutely convergent,
are taken with the same sign converges.

In other words, the series
is
absolutely convergent
when
KI +
the series of absolute values
W + KI + -
is
convergent.
It is obvious that an absolutely convergent series is also convergent in the ordinary sense, since the absolute values of the
partial remainders of the original series cannot be greater than
INFINITE SEQUENCES
those
series
AND SERIES
45
There are, however, convergent

of the second series.
which are not absolutely convergent
:
i+
e.g.
which
Series in
of both positive
-J
+i
is
convergent.
. .
is
divergent.
depends upon the presence

said to be conditionally
terms
are
negative
the convergence
and
convergent.
The reason for this
name
same sum, even although we

are taken
that, as
is
absolutely convergent series
we
shall
now
prove, an
remains convergent, and has the

alter the order in which its terms
while a conditionally convergent series
may
con-
verge for one arrangement of the terms and diverge for another.
Indeed we shall see that we can make a conditionally convergent
series
we
have any sum we please, or be greater than any number

by changing the order of its terms. There is
care to name-,
nothing very extraordinary in this statement. The rearrangement of the terms introduces a new function of n, say S'n
Sn as
reason why
instead of the old function
There
is
no a priori
the
of the first
terms.
this function S' n should
tends to infinity, or, if

should be the same as the limit of 8n *
a limit as
sum
it
has a
limit,
have
that this
<
Absolutely Convergent Series. The sum of an absolutely

convergent series remains the same when the order of the*
terms is changed.
Let (S) be the given absolutely convergent series; ($') the
series formed with the positive terms of (S) in the order in
22.
which they appear
(S") the series
formed with the absolute
values of the negative terms of (S), also in the order in whicli
they appear.
in (S') or (S") is limited, the
Îf the number of terms either
theorem requires no proof, since we can change the order of
the terms in the finite sum, which includes the terms of (S) up to
the last of the class which is limited in number, without altering
its sum, and we have just seen that when the terms are of the
same sign, as in those which follow, the alteration in the order
in the convergent series does not affect its sum.
*Cf. Osgood, Introduction
to Infinite, Series, p. 44, 1897.
INFINITE SEQUENCES
46
Let
Be the
sum
AND SERIES
of the infinite series formed
by the absolute
values of the terms of (8).

Let Sn be the sum of the
In this sum
Let
Let
let n'
tirst n terms of (S).

terms be positive and n' negative.
Sn be the sum- of these n' terms.

Sn be the sum of the absolute values
>
taking in each case these

appear in (8).
8 = 8,-8
Then
Now,
of these n" terms,

terms in the order in which they
as
Sn 8a
increases
>,
never diminish.
increases without limit, the successive values of
Thus, as
Sn S form two
H
>,
monotonic sequences such as we have examined in 17,

whose terms do not exceed the fixed number 2. These sequences,
therefore, tend to fixed limits, say, S' and 8".
Thus
= S'-8".
Lt
infinite
(Sn )
TO
Hence
the
sum
>co
of the absolutely convergent series (S) is equal
the difference between the sums of the two infinite series

formed one with the positive terms in the order in which they
to
appear,
and
the other
with the absolute values of the negative

which they appear in (S).
terms, also in the order in
Now any alteration in the order of the terms of (8) does

not change the values of S' and S" since we have seen that in
the case of a convergent series whose terms are all positive we do
;
sum by rearranging the terms. It follows that

remains
(S)
convergent and has the same sum when the order of
its terms is changed in any way we please, provided that a onenot alter the
one correspondence exists between the terms of the old series

and the new.
We
add some other results with regard to absolutely convergent series which admit of simple demonstration
Any series whose terms are either equal or inferior in
absolute value to the corresponding terms of an absolutely
:
is also absolutely convergent.

absolutely convergent series remains absolutely convergent
we suppress a certain number of its terms.
convergent series
An
when
INFINITE SEQUENCES
//
AND SERIES
1^
+ 11.,+ ...,
Vi
+ V2+'~
47
'
U and
V,
sums are equal
to
sums are
are two absolutely convergent series whose

the series
and
(i^
are aiso absolutely convergent
UV
and
their
respectively.
The sum of a condi23. Conditionally Convergent Series.

tionally convergent series depends essentially on the order of its
terms.
Let (S) be such a series. The positive and negative terms
must both be infinite in number, since otherwise the series
would converge absolutely.
Further, the series formed by the positive terms in the order
in which they occur in (S), and the series formed in the same
way by the negative terms, must both be divergent.
Both could not converge, since in that case our series would
be equal to the difference of two absolutely convergent series,
some of whose terms might be zero, and therefore would be
absolutely convergent ( 22). Also (S) could not converge, if
one of these series converged and the other diverged.
We can therefore take sufficient terms from the positive
sum exceed any positive number we care

way we can take sufficient terms from
the negative terms to make the sum of their absolute values
exceed any number we care to name.
terms to
make
to name.
their
In the same
Let a be any positive number.

First take positive numbers from (8) in the order in
which
they appear, stopping whenever the sum is greater than a.

Then take negative terms from (S), in the order in which they
appear, stopping whenever the combined sum is less than a.
Then add on as many from the remaining positive terms as
will
make
the
sum exceed
a,
stopping
when
the
sum
first
exceeds
and then proceed to the negative terms and so on.

In this way we form a new series (S') composed of the same
terms as (S), in which the sum of n terms is sometimes greater
than a and sometimes less than a.
;
AND SERIES
INFINITE SEQUENCES
48
Now
the series ($) converges.

Then, with the usual notation,
\u n
Let
its
terms be u lt
u.,,
?
;i
<e, when n^v.
Let the points B v and A v (Fig. 4) correspond to the sums

obtained in ($'), as described above, when v groups of positive
terms and v groups of negative terms have been taken.
a
FIG.
4.
A v ) and (Bv a) are each less than

it is clear that (a
since each of these groups contains at least one term of (8),
v ), (B v
a) are at most equal to the absolute value of
(a
Then
\u v \,
and
the last term in each group.

Let these 2v groups contain. in
The term
when
u' n in (8'),
>
all v
n'
terms.
= v,
is
less in absolute
we proceed from A v the sums

Thus,
the interval (a e, a + e), when n'=i/.
In other words, |$V a\ < e, when ri^v.
than
if
e.
Lt S' n
Therefore
n'
similar
argument holds
lie
value
within
= a.
for the case of a negative
now we
any value we
series
to
number,
begin with the negative
have thus established the following theorem
If a conditionally convergent
arrange the order of the terms as
series converge to
>
>oo
the only difference being that

terms of the series.
We
S'n
is
make
given,
the
sum
we can so
of the new
care to name.
REFERENCES.
BROMWICH, loc. cit., Ch.
DE LA VALLEE POUSSIN,
I.
-IV.
loc. cit.,
T.
I.
(3
ed.),
Introduction,
2,
Ch. XI.,
1, 2.
GOURSAT,
loc.
cit.,
T.
I.
(3
ed.),
Ch.
HARDY, Course of Pure Mathematics
I.
and VIII.
(2nd. Ed.), Ch. IV.
and VIII., Cam-
bridge, 1914.
PRINGSHEIM,
loc.
cit.,
Ed.
I.,
Absch.
I.,
Kap.
III.,
V.
Absch.
II.,
Kap.
I.-III.
STOLZ
u.
GMEINER,
loc.
cit.,
Abth.
II.,
Absch. IX.
And
PRINGSHEIM,
d.
"
Irrationalzahleo u. Konvergenz unendlicher Prozesse," Enc.
math. Wiss., Bd.
I.,
Tl.
I.,
Leipzig, 1898.
CHAPTER
III

LIMITS AND CONTINUITY
24. The Idea
when we speak
of a Function.
of a
In Elementary Mathematics,
we usually mean a real
function of x,
^Jx, logo?,
operations, e.g. x
sin" ^.
In some cases, from the nature of the operations, the
range of the variable x is indicated. In the first of the ai>ove
expression obtained
by
certain
in the
is unlimited
in the second, x =
and in the last = x = 1
"
In Higher Mathematics the term " function of x has a much
more general meaning. Let a and b be any two real numbers,
where b
a.
If to every value of x in the interval a = x = b
there corresponds a (real) number y, then we say that y is a
function of x in the interval (a, b), and we write y =f(x).
Sometimes the end-points of the interval are excluded from
x <[ b. In this case
the domain of x, which is then given by a
examples, the range

third x
>
>
<
the interval
is
said to be
open at both ends
when both ends
An interval
a = x = b) it is said to be closed.
may be open at one end and closed at the other (e.g. a<^x^b).
Unless otherwise stated, when we speak of an interval in the
rest of this work, we shall refer to an interval closed at both
are included
(i.e.
And when we say that x lies in the interval (a, b),

we mean that a = x = b, but when x is to lie between a and b,
ends.
and not
the
to coincide with either,
open interval (a,
*In Ch.
either,
II.,
we have
we
shall
say that x
lies
b).*
when a
point x lies between a and b, and does not coincide with

referred to it as within the interval (a, b).
This form of words is
convenient, and not likely to give rise to confusion.

c.
in
49
50
insider the aggregate formed by the values of a function

If this aggregate is bounded
/(.'), given in an interval (a, b).
that
the
we
function
(cf.
11),
say
f(x) is bon ndrd, in the interval.
('<
The numbers
and m, the upper arid lower bounds of the

aggregate (cf. 12), are called the upper and loiver bounds of
the function in the interval. And a function can have an upper
bound and no lower bound, and vice versa.
The difference (M m) is called the oscillation of tlie function
in the interval.
should be noticed that a function
It
in
may
be determinate
an interval, and yet not bounded in the interval.

/(0) = 0,
E.g. let
and
when
f(x)=X
5>.0.
Then f(x) has a
definite value for every x in the interval

a
is any given positive number.
where
But f(x) is
0<#Ea,
not bounded in this interval, for we can make f(x) exceed any
number we care to name, by letting x approach sufficiently near
to zero.
Further, a bounded function need not attain its upper and

and in need not be members of
lower bounds in other words,
the aggregate formed by the values of f(x) in the interval.

E.g. let
/(0)
= 0,
and
f(x) =
I-x when 0<aj<l.
This function, given in the interval
bound
zero,
Lt
25.
but not
f(x).
its
(0, 1),
upper bound unity.
attains
>
its
lower
In the previous chapter we have dealt with the
.'>
u lt u 2 u s .... In other words,

a
where n is a positive
with
function
we have been dealing
<j>(n),
integer, and we have considered the limit of this function as
when
limit
We
pass
TI->OO of a sequence
now
to the function of the real variable x
limit of f(x) when xâ.

Differential Calculus rests
The idea
and the
familiar enough.
The
But for our purpose we
is
upon it.
must put the matter on a precise arithmetical footing, and a
definition of what exactly is meant by the limit of a function
of x, as x tends to a definite value, must be given.
f(x] is said to have the limit b as x tends to a, wlien, any
positive number e having been chosen, as small as we please,
AND CONTINUITY
LIMITS
n positive number
such
which
a
x
=
\x
of for
In other words \f(x) b\ must be
there is
that \f(x)
tj
<
rl nes
the interval (a
When
this
ri,
= b, for
Lt f(x)
is
b\<ê, for all
v\.
than
less
a + rf) except the point
condition
51
for all points in
a.
we employ
satisfied,
the notation
the phrase the limit of f(x), as x tends to a, is
x->a
and we say that f(x) converges to b as x tends to

One advantage of this notation, as opposed
b.
a.
to
Lt /(#) = &,
x=a
the fact that we say nothing about

equal to a. In the definition it will be
observed that a statement is made about the behaviour of f(x)
is
that
it
brings out
what happens when x
!for all
is
<
values of x such that
\x
The
a\ ^.rj.
first
inserted expressly to exclude x = a.

inequalities
Sometimes x tends to a from the right hand only
of these
is
or from the left
hand only
<
(i.e.
In these cases, instead of
<
and
is
Lt f(x)
>
a),
a).
tj,
we have <(#
and
Lt
f(x).
a)
17
a;->o-0
The
(i.e.
y (left hand), in the definition.

x)
notation adopted for these right-hand and left-hand limits
(right hand)
The
<(a
x
x
assertion that Lt f(x)
=b
thus includes
x Xi
Lt f(x)= Lt f(x) = b.
,
X-+0-Q
x-+a+Q
It is convenient to use
/(a + 0) for Lt f(x) when this limit

x-+a+0
exists,
and similarly f(a
0) for
Lt f(x) when
-
this limit exists.
x->
When
f(x) has not a limit as x-^-a, it may happen that it

+ oo or to oo in the sense in which these terms
diverges to
were used in
16.
Or, more
precisely, it
iy positive number A, however large, is

$ponds to it a positive number such that
that if
there
correchosen,
may happen
r\
f(x)> A, when
In
this case
we say
that Lt
< \x
f(x) = + oo
a\
E>/.
A >rt
it
may happen that if any negative number A is
however
chosen,
large A may be, there corresponds to it a positive
number such that f(x)< A, when
|aj
a\ ^tj.
Again,
t]
<
52
III ////x rust' ire
.svn/
Lt f(x) =
thut
oc
jc->a
The modifications
When
when/(a0) =
Lt f(x) does not
exist,
.>'-a
to
+ 70,
or to
oo, as JC-H, it is said to oscillate as #->a.
oscillates finitely if /(#) is
that point.*
are obvious.
00
and when f(x) does not diverge
bounded
It
some neighbourhood of
there is no neighbourhood
in
It oscillates infinitely if
a in which f(x] is bounded. (Cf. 16.)

The modifications to be made in these definitions when x->a
only from the right, or only from the left, are obvious.
of
26.
// Lt
Some General Theorems on

f(x) = a and
I.
The Limit of a Sum.
Lt g(x) = fr then Lt [f
Let the positive number
Then
Limits.
we
be chosen, as small as
numbers
to e/2 there correspond the positive
when
>; 15
1].2
please.
such that
0< a -a
|
Thus,
if
>/
is
not greater than
\f(x)+g(x)-a-p\^\f(x)-a
<
<
Therefore
|
e,
or
>; 2
|^)-/8|,
^,
when
when
Lt [/
.r->ff
This result can be extended to the sum of any number of

functions.
The Limit of a Sum is equal to the Sum of the
Limits.
*f(x)
when
is
there
said to satisfy a certain condition in the neighbourhood of

is a positive number h such that the condition is satisfied
0<|:e-a|</t.
Sometimes the neighbourhood is meant to include the point x
this case it is defined by \x -a\^h.
The corresponding theorem

proved in the same way, and
t
is
the previous chapter.
x a
when
itself.
In
for functions of the positive integer ?., as ?i->oo ,

useful in the argument of certain sections of
is
AND CONTINUITY
LIMITS
The Limit
II.
=
// Lt f(x) a
of a Product.
53
and Lt
z>a
x>a
x->a
f(x) = a'+</>(x)
Then
= /3,
= a{3.
then Lt [f(x)g (a?)]
Let
#(>;)
Lt
x >.
000 =
and
ff(
and
Lt
./;>
= al3 + P<j>(x) + a\ls(

f(x)(j(x)
Also
From Theorem
our result follows
I.
Lt
if
Since <(#) tends to zero as ic-x- and ^-(a?) tends to zero as

But if a formal
a?-xt, a proof of this might appear unnecessary.
run
as
follows
it
could
is
required,
proof
:
Given the arbitrary position number
e,
we
have, as in L,
K-sA> when 0<|a;-a|î

I
Thus,
if
//
is
< v/*>
not greater than ^ L or

<ê,
|^(ic)^r(ic)|
lll
< |ft-a|3s^.
wlien
;._,
< |x
when
ct|
iy
Lt
Therefore
This result can be extended to any number of functions.

is equal to the Product of the Limits.
The
mimit of a Product
The Limit
III.
(i)
// Lt
/(aj)
of a Quotient.
= a^0,-
then Lt
This follows easily on putting f(x) =

the expression
If
(ii)
(f>(x)
+a
and examining
Lt f(x) = a, and Lt g(x) = p^0, then Lt
This follows from
II.
and
III.
(i).
This result can obviously be generalised as above.

IV.
The Limit of a Function of a Function.
Let
=6
<(.*)
Lt
j:
7%cw
>
are given that Lt f(u)
a;
f(b).
Lt
>
^(^)J-
f[<Mx)].
FUNCTIONS OF A S1NCLE VARIABLE
54
Therefore to the arbitrary positive number

such that
i)
number
there corresponds a positive
/W(*)] -/(&)!<,
en !<(*)- &|<7h ..................... (1)
= b.
<f>(x)
Also we are given that Lt

x-*-a
Therefore to this positive number

there corresponds a positive number
such that
wl
û
...................... (2)
|<K*)-&l<fi,
->) l
7,
0<\x-a\^
Combining
and
(1)
sponds a positive number

I
number
to the arbitrary positive
(2),
?/
there corre-
such that
/[*(*)]-/(&)!<, when 0<l#-a|^7j.
Thus
Lt/[^)] =/(&)=/[
Lt
<(*')].
EXAMPLES.
1.
If
?i
a positive integer, Lt
is
,v
= 0.
x*Q
2.
If M
is
a negative integer, Lt
is odd or even.
according as ?i
n=
[If ?i=0, then #
3.
and Lt
arÔ
V' + V'
5.
Lt r n = rt n
6.
If
if
?i
is
"---." =
+ &*-!*' + *. / '>n
any positive or negative
unle
6.
integer.
P(x)
Lt P(x) = P(d).
then
7 J (- V ) =
Let
and
>a
m+
a-itf*-
(? (^
The "
8.
If
Lt
the same as Lt f(x+a).
/(.f) exists, it is
x>a'
9.
If
/(#)
and
<
Lt
x
x >0'
<t7(.#)
for
/(.r)
= a,
'
a
If
Lt/(a;)=0, then Lt
x >a
11.
If
gr
|/(#)|
hold.
/t,
/3.
=0, and conversely.
xi
Lt/(#) = J>0, then Lt
The converse does not
a-h<.v<a +
Lt (*) = j8
x >a
>a
then
10.
.*,*"=-
>0
Lt
x
7.
=+x;
Lt
ac
4.
= !.]
and Lt #
.f
|/(o?)|
= K|.
or
+x
LIMITS
Let
12.
/(.?)
AND CONTINUITY
be defined as follows
/()=# sin 1
when ^-^0^
/a?,
"
/(0) =
Then Lt f(x}=f(a)
Lt
27.
X
f(x).
55
for all values of a.
meaning
of the term
(or to
OD )
precise definition of the
>-^3
"
the limit of f(x)

needed.
when x tends
+ oo
to
"
+ oo
f(x) is said to have the limit b as x tends to

positive number e having been chosen, as small as
a positive number
there is
also
if,
any
please,
X suck that
- b |<
\f(x)
When
we
is
e,
this condition is satisfied,
when x > X.
we write
Lt f(x) = b.
Lt f(x) = b,
similar notation,
x > -oo
when
oo
and the
f(x) has the limit b as x tends to
term can be obtained by substituting
"a negative number
X" and "x
X" in the corresponding
places in the above.
is
used
precise definition of the
When
it is
Lt f(x)
From
is
clear that only positive values of
Lt
used instead of
x are
in question the notation
/(.')
x tends
the definition of the limit of f(x) as
to
oo, it
follows that
Lt /(#) = &
X
carries with
Lt
it
And, conversely,
>+:
Lt
if
then
Lt
"
*_*+*,
Similarly
we have
a;*-
The
\x
Lt /(.*)=
oo
Lt /(\3',
->
4- oo
or
when
- cc
Lt /(.r)=+oo
or
- oo
x tends
to
modifications in the above definitions
Lt /(#) =
(i)
'
*->+00
and
will
(ii)
be obvious, on referring to
And
25.
oscillation, finite or infinite, as
as before.
+oc or to -
oo
is
treate^
")()
28.
necessary and sufficient condition for the existence

x tends to a. The general principle of
of a limit to f (x) as
convergence.*
A necessary and sufficient condition for the existence of a

limit to f(x) as x tends to a, is that, when any positive number
e has been chosen, as small as we
please, there shall be a positive
number
rj
such that |/(X')-/<X)I

\x"-a\
\x'-a\ ^ n
<
for which
<
The condition
(i)
for all value* of x, x"
is necessary.
Let Lt f(x) = b.
x>a
Let
Then
be a positive number, as small as we please,

to e/2 there corresponds a positive number tj such that
|
Now
let x', x"
Then
f(x)
<
a
\
rj.
be any two values of x satisfying
\f( x "}
/(#')
/(#")
<
(ii)
< \x
when
~,
The condition
~~
f(x')
e.
is sufficient.
Let
e 1?
e2
e3
...
be a sequence of positive numbers such that

e n+1
<en
and
Lt
n
Let
= 0.
><*>
^,
% ...
be corresponding positive numbers such that
\f(")-f(')\
On,
when
< l^'-aK |c'-a|

(= 1,2,
tj
n]
3, ...).)'
<
e n we can obviously assume that

Then, since e n+1
Now take e x and the corresponding r
In the inequalities (1) put x
a-\-^ and x" = x.
,
tj
>;
Then we have
<
*
|/(^)
/( a + //i)| <C
See footnote, p. 34.
î>
when
< |a?--a| <^
i/ 1
LIMITS
AND CONTINUITY
57
Therefore
f(a + ni ) -e,< f(x)
< f(a +
In Fig. 5 f(x) lies within

centre at /(a 4-^), when
<
+ e, when
the interval A
,
7l )
<
<
a|
|aj
>/
x - a |<
(2)
of length 2e 1? with
A,
\
FIG.
Now
take
e2
5.
and the corresponding
?/ 2
remembering that
<;
We
have, as above,
/(a + ^ 2 )
Since
rj 2
e2
tj l
<[/(a;X/(a + ^ 2 ) + e2 when 0<|.

,
the interval for
the interval for x in
(2),
Therefore, in Fig.
which
lies
common
An
end-point.
could be cut
off,
(3)
>/ 2 )
now
f(x)
entirely within A lt or
in (3) cannot extend beyond

is in the open interval
l
a;
and /(a +
6,
a|<>/ 2
lies
within the interval A.2
lies
within
it
and has with
it
overlapping part of
in virtue of (2).
A,
î
FIG.
In this
way we
A
.
6.
obtain a series of intervals

-"!
~i
-^2'
'
each lying entirely within the preceding, or lying within it and

having with it a common end-point; and, since the length of
-4 n
2en we have Lt A n
0, for we are given that Lt e n = 0.
If
and
we denote
/3 1} /3 2 , /3 S
n >
>c*
the end-points of these intervals by a 1? a.,, a.i? ...

...
where /3n
a n> then we know from 18 that
>
Lt
n
Denote
We
>-oo
Lt
n
ft,.
><
common limit by a.
now show that a is the
this
shall
limit of f(x) as x->a.
58
We
can choose
where
e is
en
in the sequence
e 1? e 2 , e 3 , ...
Then we
</()< j8i>
an = a
^pn
But
2e,,
<
e,
when
0<|-a|<;
7n
&n - a n
Therefore |/(aj)-o|<
<
<
2* M
when
e,
Lt
It follows that
< \x
a matter of
a\
<
*/
= a.
/(a;)
we have not obtained

< \x-a\^
a| < e, when
< \x a\ < rj n
in the above, hut when
However, we need only take r; smaller than this
As
so that
positive number.
have, as above in (2) and (3),
any given
fact,
|/(#)-
>/
?;
inequalities used in our definition of a
and we obtain the
limit.
In the previous section we have supposed that x tends to

a from both sides. The slight modification in the condition for
convergence when it tends to a from one side only can easily
29.
be made.
Similarly, a necessary and sufficient condition for the existence
00, is that, if the 'positive
of a limit to f(x) as x tends to
number e has been chosen, as small as we please, there shall be a
number
positive
X such that
\f(x") -f(x)
Lt
In the case of
f(x),
<
we
e-,
when
x"
>x'^ X.
have; in the same way, the condition
X >-co
l/(") -/(aO
The conditions
to
+ oo
or to
oo
<*> when
a?<x'^-X.
for the existence of a limit to f(x) as x tends

can, of course, be deduced from those for the
existence of a limit as x tends to -f
or to
0.
Actually the argument given in the preceding section is simpler

when we deal with +00 or oo ,* and the case when the variable
tends to zero from the right or left can be deduced from these
two, by substituting x =
stitute x
* Cf.
= a-\
when
it
tends to a,
we must
sub-
Bd. I., p. 27, Leipzig, 1907.

Osgood, Lehrbuch der Funktionentheorie,
also be established in this way.
15
can
of
convergence
of
principle
The general
LIMITS
AND CONTINUITY
59
The function f(x) is said to be

Continuous Functions.
when x = x if f(x) has a limit as x tends to X from
30.
continuous
and each of these limits is equal to /(# ).

Thus f(x) is continuous when x = x if, to the arbitrary posi-.
such that
tive number e, there corresponds a positive number
either side,
tj
1/0*0 -/fao)
<
when
*,
x - xo
n-
defined in. an interval (a, 6), we shall say that it

f(x)
is continuous in the interval (a, 6), if it is continuous for every
When
is
and b (a <#<&)> and if f(a + 0) exists

and f(b 0) exists and is equal to f(b).
convenient to make a slight change in our
value of x between a
and
to /(a),
is
equal
In such cases
it is
definition of continuity at a point, and to say that f(x) is continuous at the end-points a and 6 when these conditions are
satisfied.
from the definition of continuity that the sum or

product of any number of functions, which are continuous at a
The same holds for the
point, is aJso continuous at that point.
It follows
quotient of two functions, continuous at a point, unless the
denominator vanishes at that point

function of a continuous function
(cf.
26).
continuous
also a continuous function
is
26 (IV.)).
(cf.
The polynomial
P(x) = a x
is
continuous for
The
is
of
all
+ a^n - +
values of
+ a n . & + an
x.
rational function
continuous in any interval which does not include values

x making the polynomial Q(x) zero.
The functions sin a;, cos a, tana;, etc. and the corresponding
functions sin'
a;,
cos" 1 a;, tan'
a;,
etc.
are continuous except, in
certain cases, at particular points.

e
continuous everywhere; logo;
is
terval x
>
is
continuous for the in-
0.
Properties of Continuous Functions.* We shall now

prove several important theorems on continuous functions, to
31.
* This section follows

closely the
(3*ed.),
8.
treatment given by Goursat (he.
cit. ),
T.
I.
60
which reference will frequently be made later. It will be seen

that in these proofs we rely only on the definition of continuity
and the results obtained in the previous pages.
THEOREM I. Let f(x) be continuous in the interval (a, b)*,

and let the positive number e be chosen, as small as we please.
Then the interval (a, b) can always be broken up into a finite
number of partial intervals, such that \f(x) /(&")] <C e, when
x and x" are any two points in the same partial interval.
Let us suppose that this is not true. Then let c = (a + 6)/2.
At least one of the intervals (a, c), (c, b) must be such that it
is impossible to break it
up into a finite number of partial intervals which satisfy the condition named in the theorem.
Denote
by (a l} b^ this new interval, which is half of (a, b). Operating
on (ttj,, 6 X ) in the same way as we have done with (a, b), and
then proceeding as before, we obtain an infinite set of intervals
such as we have met in the theorem of
a,
end-points
a lf
a.,,
18.
The sequence
converges, and the sequence
...
of
of
end-
... also
converges, the limit of each being the
Also
each
of
the intervals (a n b n ) has the property
same, say
we have ascribed to the original interval (a, b). It is im-
points
6,
b lt
6.7
a.
possible to break
which
it
up into a
satisfy the condition
number
finite
named
of partial intervals
in the theorem.
Let us suppose that a does not coincide with a or b. Since the

function f(x) is continuous when x = a, we know that there is a
such that |/(#) /()| <C e/2 when \x a\
positive number
Then the
Let us choose n so large that (b n a n ) is less than
a + 77), for we
interval (a n b n ) is contained entirely within (a
know that a n = a = b n Therefore, if x and x" are any two
points in the interval (a n b n \ it follows from the above that
rj
*].
//.
;,
\f( x ')-f(x")
Thus we have
|
= !/(*')-/
f(x'} -f(x")
and our hypothesis leads

*
<
to a contradiction.
In these theorems the continuity of f(x)
interval
(a^x^b),
e,
as explained in
30.
is
supposed given in the closed
AND CONTINUITY
LIMITS
61
There remains the possibility that a might coincide with either

a or b. The slight modification required in the above argument
is
obvious.
Hence the assumption that the theorem is untrue leads in

every case to a contradiction, and its truth is established.
COROLLARY
of
(a, b)
Theorem
Then
Let
I.
into
a,
partial
x lt x 2
...
xn _ l
intervals
b be a
mode
of subdivision
the
conditions of
satisfying
I.
\f(x)
\^\f(a)\ + \f(x)-f(a)\
<
/() +
< (x - a) ^ (x
when
e,
- a).
Therefore
In the same
\f(x)
way
IÎ/
<\f( X l)\ +
<\f(a) |+
Therefore
I
/(,)!< !/()!+
2e.
Proceeding in the same
we
obtain from the
t]t
way
for each successive partial interval,
interval
\f(x)\<\f(a)\ + n e when
,
Thus we
see that in the
It follows that
0<(x-xn _ )^(b-x n _ \
l
whole interval
a function which
is
(a, b)
continuous in a given
interval is bounded in that interval.
COROLLARY
II.
Let us suppose the interval
(a, b)
divided up
partial intervals (a, a?j), (x lt X 2 ), ... (x n _ lt b), -such that

\f(x')f(x")\ <e/2 for any two points in the same partial inter-
into
val.
Let
rj
be a positive
number smaller than
the least of the
numbers (x l a), (x2 ic 1 ), ... (b x n _ l ). Now take any two points

x and x" in the interval (a, 6), such that \x
If these
x"\^tj.
two points belong to the same partial interval, we have
f
\f(x')-f(x")\<
f /-2.
62
On
the other hand,
they must
interval,
if
lie
they do not belong to the same partial

two consecutive partial intervals. In
in
this case it is clear that

\f(x)-f(x")\ <e/2 e/2 = e.
the
number
e having been chosen, as small as
Hence,
positive
we please, there is a positive number such that \f(x') f(x")\ < e,

when x, x' are any two values of x in the interval (a, b) for
rj
which
We
(a, b).
\x'
x"\
</.
started with the assumption that f(x) was continuous in

It follows from this assumption that if x is
any point in
this interval,
a positive
and
number
e
rj
any arbitrary positive number, then there
is
such that
l/( aj/ )-/( aj
)l<* when \n'-x\^ n
To begin with, we have no justification for supposing that the

same
could do for all values of x in the interval. But the
>/
theorem proved in this corollary establishes that this is the case.

This result is usually expressed by saying that f(x) is uniformly
continuous in the interval (a, b).
We have thus shown that a function which is continuous in
.
an
interval is also uniformly continuous in the interval.
THEOREM II. If f(a) andf(b) are unequal and f(x) is continuous in the interval (a, b), as x passes from a to b, f(x) takes
at least once every value between f(a) andf(b).
suppose that /(a) and f(b) have different signs,
and
0.
We shall show that for at least one
e.g. f(a)
f(b)
value of x between a and b, f(x) = 0.
First, let us
<
From
>
the continuity of f(x),
we
see that
it is
negative in the
neighbourhood of a and positive in the neighbourhood of b.

Consider the set of values of x between a and b which make
Let X be the lower bound of this aggregate.
b.
From the definition of the lower bound
x<\. But Lt f(x) exists and
negative or zero in a
f(x) positive.
Then
f(x)
is
<X<
.c-^-A.
is
equal to /(X).
Therefore /(X)
is
also negative or zero.
/(X) cannot be negative. For if /(X) =

number, then there is a positive number
But
m, m
r\
being a positive
such that
l/0)-/(X)|<m, when \x-\\^

continuous when x = \. The function
t
since f(x)
is
then be negative for the values of x in
(a, b)
f(x) would
between X and X-f
>;
AND CONTINUITY
LIMITS
63
and X would not be the lower bound of the above aggregate.

We must therefore have /(X) = 0.
be any number between f(a) and /(&), which may
Now let
be of the same or different signs. The continuous function
has opposite signs when x a and x = b. By the
(j>(x)=f(x)
we have
case
just discussed,
(f>(x)
vanishes for at least one value
x between a and b, i.e. in the open interval

Thus our theorem is established.
of
if
Again,
we show
bounded
it is
that
THEOREM
it
continuous -in
is
f(x)
above that
(a, b),
(a, b).
we know from
in that interval.
Corollary I.
In the next theorem
attains these bounds.
III.
Iff(x)
is
continuous in the interval
(a, b),
and
m are its upper and
M,
lower bounds, then f(x) takes the value

the value in at least once in the interval.
M and
We
shall
Let
least
show
at least once in the interval.

that f(x) =
the upper bound of f(x) is equal to M, for at
first
(a +6)/2
one of the intervals
;
interval,
we
Theorem
I.,
(a, c), (c, b).
Replacing
by
(a, b}
this
In this way, as in
it, and proceed as before.
of
intervals
we obtain an infinite set
(a, b), (a 15 bj,
bisect
tending to zero in the limit, each lying entirely

within the preceding, or lying within it and having with it a
common end-point, the upper bound of f(x) in each being M.
(a.2
b 2 ),
...
common
limit of the sequences a, a lt a 2 ... and

show
that /(X) = M.
^
where /*,>0. Since f(x) is continuous
For suppose f(\) =
at x = X, there is a positive number ^ such that
Le^X
b,
62
be the
. .
We
shall
Mh,
when Is-XlS*
Thos/(oj)<if-|,
Now
interval
take
(
n,
when
so large that (b n
an)
b n ) will be contained
will be less than
Combining
this
bn )
The
X + >y).
would then be
wholly within (\
The upper bound of f(x) in the interval (a w

different from M, contrary to our hypothesis.
q.
rj,
theorem with the preceding we obtain the
following additional result
THEOREM IV. // f(x) is continuous in the

and M, in arc its upper and lower bounds, then
interval
it
(a, b),
takes at least
64
once in this interval
M and
M,m,n</
/If i-olucs
rrrri/
ralue betiveen
in.
Also, since the oscillation of a function in an interval was

defined as the difference between its upper and lower bounds
and
24),
(cf.
we can
into a finite
Theorem
(x lt
xj,
...
(x n _ lt
known
b),
than any given
as uniform continuity.
Some
Continuity in an Infinite Interval.
some
is
x2 ),
the oscillation of f(x) is less
section can be extended to the case
a similar change can be made in the statement of the
property
32.
least once
afresh as follows
I.
(a, b), then we can divide

intervals
of partial
number
(a,
in each of which
positive number.
And
state
If f(x) is
(a, b)
bounds at
since the function attains its
in the interval,
definite positive
Let u = a\x.
With
We thus
^j. 1
< u^. 1, associate
sponding points in .r^rt, and to w =

obtain a function of
?<,
.?'
where
exists.
>-00
<u
have
of the results of the last
continuous in
is
number, and Lt f(x)
When x ^ o we
the values of u in
when /(#)
the values of /(#) at the corre-
assign Lt /(#).
X *oo
which
is
continuous in the closed interval
(0, 1).
Therefore it is bounded in this interval, and attains its bounds J7, m.

Also it takes at least once every value between
and wi,as u passes over the
interval
(0, 1).
Thus we may say that f(x) is bounded in the range* given by

new "point" #=oo at which f(x) is given the value Lt /(.r).
the
xâ
and
Also f(x) takes at least once in this range its upper and lower bounds,
and every value between these bounds.
#2
oo ).
It does not
For example, the function
9 is continuous in (0,
-
Cl
upper bound
attain
its
x = QO
as defined above.
unity
~r~ 3C**
when .r>0, but
it
takes this value
when
When f(x) is defined for

Functions.
x \^h), and
the neighbourhood of X Q (e.y.
\x
at
XQ
is
continuous
then
+
0)=f(xQ 0)=/(z ),
/(x)
f(x
On the other hand, when f(x) is defined for the neighbourhood
Discontinuous
33.
of x
*
/(oo
<
and
XQ
It
)
and
is
for
it
may
be also for x
while f(x)
is
not continuous at
convenient to speak of this range as the interval
Lt
(a,
),
and to write
AND CONTINUITY
LIMITS
x
it is
natural to say that f(x)
65
discontinuous at XQ) and to
is
x a point of discontimiity of f(x).

Points of discontinuity may be classified as follows
call
I-
/(#o
+ 0)
common
and f(xQ
value
and be
exist
may
0)
from f(x
different
is
or
),
if
equal.
f(x)
is
their
If
not defined
we have a point of discontinuity there.

= (x - XQ) sin 1 j(x - XQ \ when x < XQ
f(x)
+ 0)=f(xQ -0)=0, and if we give f(x9 any value
for XQ then
,
Ex.
Here f(x
zero, or if
we
leave /(.r
undefined,
XQ
is
other than
a point of discontinuity of f(x).
s*.
Then x
0) may exist and be unequal.
f(x +0) and f(x
a point of discontinuity of f(x), whether f(x ) is defined or not.
II.
is
Ex.
/(^) =
Here
/(.?
__1__,
+ 0) =
In both these cases f(x)

discontinuity at x
and /(# -0) = 1.

said to have an
is
And
when
the same term
ordinary or simple
is
applied
when
the
an end-point of the interval in which f(x) is given,

+0),
and/(z
0), exists and is different from f(x ), if f(x)
or/(#
is defined for XQ
point x
is
III. f(x)
side,
and
may have the limit + oo

may oscillate on one side
section the cases in
which there
arranged as follows
(i)/(z
is
+ 0)=/(%-0)=+oo
no
(or
f(x
+ Q)= + 00
x
Ex.
Ex.
f(x} = I/(x- .r
In these cases
we say
XQ
),
on either
Take
These
in this
may
be
-oo).
X>XQ
0)=
ao
(or
/(a^-0)= +00
or
f(xQ + 0)
f(x) =x
as x-*x
+00).
when
-oo)|
0) exists
oscillation.
and f(xQ
= \x-x
(or
f(x
oo
(or
oo
or the other.
f(x) = \l(x-xtf, when
Ex.
(ii)
or
it
(or
-oo
exists
when x >
when x =
that the point x
is
an
infinity of f(x),
and the same term is used when x is an end-point of the interval

in which f(x) is given, and/(z + 0), or/(#
oo
0), is +00 or
,
It is usual to
say that f(x) becomes infinite at a point x of
the kind given in

c. i
(i),
and that f(xQ )

E
-f oo (or
oo).
But
this
66
must be regarded as simply a short way of expressing the

that f(x) diverges to +00 (or to
GO ) as X->XQ
fact
It will be noticed that tan
tan
JTT is
On
not defined.
tan(^7r
x has an
infinity at
i?r,
but that
the other hand,
and
0)= +00
tan(j7r
0)
ce
IV. When f(x) oscillates at x on one side or the other, x is

said to be a point of oscillatory discontinuity.
The oscillation
is finite when f(x) is bounded in some
neighbourhood of XQ it is
;
infinite
bounded
when
(cf.
Ex.
there
is
no neighbourhood of XQ in which
f(x) is
25).
(i)
(ii)
f(x) = sin
f(x)
1 l(x
= 1 l(x - #
-x
)
),
sin !/(#
In both these examples x
-#
),
when x ^ #
when x :< XQ
a point of oscillatory discon-
is
The first oscillates finitely at x the second oscillates

The same remark would apply if the function had
infinitely.
tinuity.
been given only for one side of x
The
infinities defined in III. and the points at which f(x)

oscillates infinitely are said to be points of infinite
discontinuity.
Monotonic Functions. The function f(x), given in

înterval (a, b), is said to be monotonic in that interval if
34.
the
when a = x' <x" = b'

f(x) =f(x"\ when a^x' < x" ^ b.
'
either
or
In the
(i)
(ii)
first case,
f(x')=f(x"),
the function never decreases as x increases
and it is said to be monotonic increasing in the second case, it

never increases as x increases, and it is said to be monotonic
;
decreasing.*
The monotonic character
of the function
points of the interval, and in this case
it is
may
fail at
said to be
the end-
monotonic
in the open interval.

The properties of monotonic functions are very similar to those
of monotonic sequences, treated in
17, and they may be estab-
same way
(i) // f(x) is monotonic increasing when x^-X, and f(x) is
less than some fixed number A when x = X, then Lt f(x) exists
and is less than or equal to A.
lished in precisely the
The
footnote, p. 39, also applies here.
LIMITS
AND CONTINUITY
67
// f(x) is monotonic increasing when x^X, and f(x) is

greater than some fixed number A ivhen x^X, then Lt f(x)
exists and is greater than or equal to A.
(ii)
monotonic increasing in an open interval

(a, b),
greater than some fixed number A in that
f(x)
then
interval,
/(a + 0) exists and is greater than or equal
open
(iii)
is
If f(x)
and
to
is
A.
monotonic increasing in an open interval

than some fixed number A in that open
(a, b),
exists
and is less than or equal to A.
interval, thenf(b
0)
These results can be readily adapted to the case of monotonic
decreasing functions, and it follows at once from (iii) and (iv)
that if f(x) is bounded and monotonic in an open interval, it
can only have ordinary discontinuities in that interval, or at
(iv)
is
// f(x)
and
f(x) is less
its ends.
may be worth observing
that if f(x) is monotonic in a closed

result
same
the
follows, but that if we are only given
interval,
that it is monotonic in an open interval, and not told that it
is bounded, the function may have an infinity at either end.
It
E.g. f(x)
= l/x
is
monotonic in the open interval
(0, 1),
but
not bounded.
At
is
first one might be inclined to think that a function which

bounded and monotonic in an interval can have only a finite
number of points of discontinuity in that

The following example shows that this
Let
let
/() = !,
when?
/(#) = -,
when
^2
interval.
is
not the case
<flJ
= l;
<x^^\
and, in general,
(n being any positive integer).

Also let /(0) = 0.
Then f(x)
monotonic in the interval (0, 1).

This function has an infinite number of points of discontinuity,
is
namely at x =
^ (n
being any positive integer).
68
Obviously there can only be a

continuity at which the
to k,
where k
35.
/>),
of points of dis-
is
interval.
Inverse Functions. Let the function /(.r), defined in the interval

be continuous and monotonic in the stricter sense* in (a, b).
For example,
to
number
be greater than or equal

any fixed positive number, if the function is
monotonic (and bounded) in an
(ft)
finite
jump would
B as
a:
Then
value of
let
y=f(x) be continuous and continually
passes from a to
to every value of
x in (a, b). [ 31,
This value of x
y in (J, B) there corresponds one and only one

Theorem II.]
a function of y
is
increase from
b.
say
<(,?/),
which
is
itself
continually
increasing in the interval (A, B).
FIG. 7.
The function
We
shall
<$>(y] is
now show
called the inverse of the
that $(y]
f unction /(ar).
a continuous function of y
is
in the interval
which it is defined.
For let y be any number between A and /?, and ,r the corresponding
value of x. Also let e be an arbitrary positive number such that # - e and
in
.r
+ eliein(a,
b) (Fig. 7).
Let y -rj l and 3/0 + ^/2 be the corresponding values of y.

Then, if the positive number rj is less than the smaller of i^ and
(l
clear that
\X-XQ
Therefore
Thus
<
j
<(?/) is
<
(?/)-</> (3/0)
continuous at
;/
when \y-yQ \^t).

when l# - #o
<
>
?;.2
it is
*7-
similar proof applies to the end-points A and B, and it is obvious that

the same argument applies to a function which is monotonic in the stricter
sense
and decreasing.
*
Cf footnote,
p. 39.
LIMITS
cos"
The functions sin"
and COS.T, where O X
1
.?;,
AND CONTINUITY
.*
= =
etc.,
?TT,
thus arise as the inverses of the functions
and
so on.
1
namely sin~ y, where
The
where
etc.
is
then
symbol y
O^y^l,
Q=y I,
replaced by
the usual symbol x for the independent variable.
In the same way log x appears as the inverse of the function e*.
There is a simple rule for obtaining the graph of the inverse function
f~ (x) when the graph of f(x) is known. f~ (x) is the image of f(x) in the
lineyx. The proof of this may be left to the reader.
The following theorem may be compared with that of 26 (IV.)
Let f(u) be a continuous function, tnonotonic in the stricter sense, and let
In the
place these appear as functions of y,
first
<
cos" 1 ?/,
Then Lt
exists
and
is
equal
to b.
strict proof of this result may be obtained, relying on the

property
proved above, that the inverse of the function f(u) is a continuous function.
The theorem is almost intuitive, if we are permitted to use the graph
:
of f(u).
The reader
familiar with
is
its
application in finding certain

cases it is shown
where logarithms are taken. In these

that Lt log u = log 6, and it is inferred that Lt u = b*
limiting values,
Let the bounded f unction /(^)fgiven in the interval
36.
interval can be divided
this
intervals, in each of
f(x)
is
up
into
(a, 6),
number
finite
of
be such that
open partial
which
monotonic.
Suppose that the points

la?i, ,r2
...
xn _i divide
this in-
terval into the open intervals

(a,
in
^), (#,,
#2 ),
which /(&)
is
(#_!, 6),
monotonic.
...
Then we know that/(#) can

only have ordinary discontinuities, which can occur at
the points a, x l x%
,vn -i 6,
also at any number of
points within the partial
,
. .
and
ntervals.
34.)
I.
Let us take
case
where f(x)
first
is
the
X,
con-
tinuous
at a, .i'j, ... #,,_i, 6, and alternately monotonic

increasing and
To make matters clearer, we shall assume that there are
decreasing.
only three of these points of section, namely #,, .r2 x^f(x) being monotonic
-increasing in the first interval (a, .r x ), decreasing in the second, (x l #2 ),
and so on (Fig. 8).
,
Cf Hobson,
Tri<jono>n<try, (3rd Kd.), p.
l.SO.

It is obvious that the intervals
in this case
may
be regarded as closed,
the monotonic character of /(#) extending to the ends of each.

Consider the functions F{x), G(x) given by the following scheme
0(x)
xl
/(*!)-/(*)
It
is
F(x) and G{x) are monotonic increasing

and thai f(x) = F(x)-G(x) in (a, b).
clear that
interval (a,
b),
If f(x) is decreasing in the first partial interval,
we
the closed
in
start with
when a^x^x,,
G(x)=f(a)-f(x),
and proceed as before, i.e. we begin with the second line of the above
diagram, and substitute a for # n etc.
Also, since the function f(x) is bounded in (a, b), by adding some number
to both F(x) and G(x), we can make both these functions positive if, in the
F(.v)=f(a\
original discussion, one or both were negative.

It is clear that the process outlined above applies equally well to
partial
intervals.
We
have thus shown that when
the bounded function f(x), given in the

such that this interval can be divided up into a finite number
interval (a,
b), is
of partial
intervals, f{x) being alternately
tonic decreasing in these intervals,
monotonic increasing and monoat their ends, then we can
and continuous
express f(x) as the difference
of two (bounded) functions,

which are positive and mono-
-X
tonic increasing in the inter-
val (a,
b).
There remains the case

when some or all of the
II.
points a, x\, x, ... #M _i, b

are points of discontinuity
of f(x), and the proviso that
the function
x,
x monotonic
is
alternately
increasing
and
'.
decreasing is dropped.
We can obtain from f(x)
function <$>(x), with the same monotonic properties as/(.r) in the open
intervals (a, .^), (x, x%), ... (# M _i, b), but continuous at their ends.
FIG.
new
partial
The process is obvious from Fig. 9.

the curve fixed, move the second up or
We
need only keep the
down
till its
first
part of
end-point (x lt /(# i:f 0))
AND CONTINUITY
LIMITS
71
coincides with (x^ f(x\ -0)), then proceed to the third curve
down to the new position of the second, and so on.
and move
it
up
or
If the values of f(x) at a,
x lt #2
...
/fo + 0)
/(a + 0),
#n_i,
or
b are not the
/fa-0),
same as
etc.,
we must
treat these points separately.

In this way, but arithmetically,* we obtain the function <(#) defined as
follows
a^x<x
In
At
In
.r
#2
And
We
<
,
(x) =/(#),
supposing for clear ness /(a) =/(a + 0).
(*)=/(*) + 04.
(#) =/(#) + a : + a 2
<X#)=/(#) + a.i + a 2 + a 3-
so on,
being definite numbers depending on
a si
-2
<
< x < #2
At
a i> a
lt
*!
can
now apply the theorem proved above

<(#) = $(#)-*(#) in
write
<&(x) and "V(x) being positive

It follows that
f(x\Q\
/(#i), etc.
to the function $(x)
and
(a, b),
and monotonic increasing
in this interval.
In
a^x<x
At
Xl
In
And
If
= *(*)-(*).
= 3>(x)-V(x)-a.,.
- a2
=
f(x) <(#) ^(x)
/(#)
f(x)
<x< #
l ,
ci!
so on.
any
of
the terms
the positive terms

that
we
04,
a2
...
are negative
leave with ^{x}.
f(x)
= F(x)-G(x)
Thus
in (a,
we put them with

we obtain, as
finally
<$>(x)\
before,
6),
where F(x) and G(x) are positive and monotonic increasing

We have thus established the important theorem
in this interval.
If the bounded function f(x\ given in the interval (a, 6), is such that this
be divided up into a finite number of open partial intervals,
in each of which f(x) is monotonic, then we can express f(x) as the difference of
two {bounded) functions, which are positive and monotonic increasing in the
interval can
interval (a, b).

Also it will be seen
from the above discussion that the discontinuities of

F(x) and G(x\ which can, of course, only be ordinary discontinuities, can
occur only at the points where f(x) is discontinuous.
It should, perhaps, also be added that, while the monotonic properties
ascribed to f(x) allow it to have only ordinary discontinuities, the number
of points of discontinuity may be infinite ( 34).
So far we have dealt

Functions of Several Variables.
Vith
If to every value of x
functions
of
a
variable.
only
single
in the interval a = x = b there corresponds a number y, then we
37.
It will be noticed that in the proof the curves
as illustrations.
and diagrams are used simply
72
have said that y is a function of x in the interval

have written y =f(x).
The extension
two variables
to functions of
To every pair of values of x and
y,
is
(a, 6),
and we
immediate
such that
a number z. Then z is said to be a function

domain, and we write z=f(x, y).
If we consider x and y as the coordinates of a point in a
plane, to every pair of values of x and y there corresponds a
point in the plane, and the region defined by a = x = a, b=y = b'
let
there correspond
of x and y in
this
will be a rectangle.
In the case of the single variable, it is necessary to distinguish
x
between the open interval (a
6) and the closed interval
< <
=x=
two dimensions, it is well to

In the former
distinguish between open and closed domains.
the boundary of the region is not included in the domain in the
(a
in the case of
So,
b).
latter it is included.
In the above definition we have taken a rectangle for the

A function of two variables may be
variables.
domain of the
defined in the same
way
domain
for a
of
which the boundary
is
a curve C: or again, the domain may have a curve C for its

external boundary, and other curves, C', C", etc., for its internal
boundary.
function of three variables, or any number of variables, will

For three variables, we can still draw
be defined as above.
and refer
upon the language
of geometry,
contained within a surface S, etc.
of
We
shall
two
variables.
now
function
defined,
if
is
refer briefly to
said to be
some properties
bounded
the set of values of
forms a bounded aggregate.
z,
in the
domain
f(x, y)
said
to
in
which
it is
The upper and lower bounds,

24.
y ),
(x, y) tends to (x
we
as
as
small
been
chosen,
having
have the limit
when, any positive number e

please, there is a positive number
all values of (x, y)
functions
of
for all the points of this domain,
M and m, and the oscillation, are defined as in

is
domain as
to the
v\
as
such that
\f(x,
y)
l\
<
/or which
and
x-x +
for
AND CONTINUITY
LIMITS
73
In other words, \f(x, y) I must be less than e for all points in

the square, centre (x y ), whose sides are parallel to the coordinate axes and of length 2>/, the centre itself being excluded
,
from the domain.
A necessary and sufficient condition for the existence of a limit

y) as (x, y) tends to (x y ) is that, to the arbitrary
positive number e, there shall correspond a positive number
-suck that \f(x', y')f(x", y")
/
e, where (x', y') and (x" y") are
two
than
other
any
(x y ) in the square, centre (x
points
y ),
to f(x,
<
whose sides are parallel to the coordinate axes and of length 2

The proof of this theorem can be obtained in exactly the same
17.
as in the one-dimensional case, squares taking the place
way
of the intervals in the preceding proof.

function f(x, y) is said to be continuous
y = y*>
when x = x and
if f( x y) has the limit /(b 2/o) as (x, y) tends to

(x
y Q ).
Thus, f(x, y) is continuous when x x and y = y if, to the
>
arbitrary positive number e, there corresponds a positive number

such that \f(x, y)-f(x y )\<e for all values of (x, y) for
rj
,
which \x-x \^ n and \y-y
\^>,.
In other words, \f(x, y)f(x

points in the square, centre (x
)\
z/
),
must be less than e for

whose sides are parallel
all
to
the coordinate axes and of length 2>;.*

It is convenient to speak of a function as continuous at a
Also when a
y Q ) instead of when x = x and y = y
function of two variables is continuous at (x, y), as defined above,
point (x
for every point of a domain, we shall say that

function of (x, y) in the domain.
It is easy to see that
centre at (XQ y
,
The
),
f(x, y) is said to
If
rj
<
substitute for the square, with

same centre.
would then read as follows
have the limit
the arbitrary positive
which
number
such that \f(x,y)

- xtf + (y - 2/ ) 2
(x
I
e,
(x, y) tend to (x
there corresponds
y Q ), if, to
a positive
for all values of
(x,
y) for
according to this definition

converges according to the former defini,
circle), it
as
l\<e
^n
a function converges at (X Q y
(based on the
a continuous
referred to above, a circle with the
definition of a limit
number
we can
it is
* There
are obvious changes to be made in these statements when we are
dealing
with a point (# , y ) on one of the boundaries of the domain in which the function
74
tion (based on the square)

both cases are the same.
Also continuity at (x
\f( x >
y)f(o>
2/o)
for
att
Every function, which
the limits in
would now be defined as follows
f(x, y) is continuous at (xQ

there corresponds
e,
number
And
and conversely.
is
y ), if, to the arbitrary positive

a positive number
such that
values of (x, y) for which
,
rj
continuous at (# y ) under this

y ) under the former definition,
,
definition, is continuous at (x
and conversely.
It
is
important to notice that
if
a function of x and y is continuous with

it is also continuous when con-
respect to the two variables, as defined above,

sidered as a function of x alone, or of y alone.
For example, let/(#, y} be defined

?
tf(*i
/)
=
y2-
/
?
wn en
2>
as follows
at least one of the variables
is
not zero,
t/(0,0) = 0.
Then /(#, y) is a continuous function of x, for all values of x, when y

has any fixed value, zero or not and it is a continuous function of ?/, for all
values of y, when x has any fixed value, zero or not.
;
But
it is
not a continuous function' of
the origin, since /(.r, y)

For,
is
we put x=rcos
if
of
pendent
However,
r,
6,
y = rsin
and varies from -
0,
to
domain which includes
when #=0 and y = 0.
we have/(#, y)=sin 20, which
is
inde-
+ 1.
a continuous function of
it is
y) in any
(x,
not continuous
(x,
y) in any domain which does
not include the origin.

On the other hand, the function defined by
(f(.
x y}
i
,/
r?/
y when
at least one of the variables
is
not zero,
l/(0,0) = 0,
is
(x,
y) in any domain which includes the origin.
The theorems
as to the continuity of the sum, product and, in certain

cases, quotient of two or more continuous functions, given in
30, can be
readily extended to the case of functions of two or more variables.
continuous function of one or more continuous functions
In particular
we have
the theorem
also continuous.
Let u = <j>(x,y\v = ty(x,y) be continuous at (#,

Let z=f(u, v) be continuous in
is
(u, v) at
),
and
let
uQ =
).
Then z=f[<f>(x, y\ "fC^ #)] is continuous in (x, y) at (r y ).

Further, the general theorems on continuous functions, proved in
hold, with only verbal changes, for functions of two or more variables.
,
31,
LIMITS
For example
75
If a function of two variables

it is
AND CONTINUITY
is
continuous at every point of a closed domain,
uniformly continuous in the domain.
In other words, when the positive number e has been chosen, as small as we
a positive number 77 such that f(x', y') -/(#", y") < *, when
(.v, y') and (x", y") are any two points in the domain for which
please, there is
REFERENCES.
DE LA VALLÊ
GOURSAT,
HARDY,
POUSSIN, he.
loc. cit.,
loc. cit.
T.
I.
cit.,
(3 ed.),
T.
Ch.
I.
3.
(3 ed.), Introduction,
I.
(2nd Ed.), Ch. V.
OSGOOD, Lehrbuch der Funktionentheorie, Bd.
I.,
Tl. I.,'Kap.
I.,
Leipzig,
1907.
PIERPONT, Tlwory of Functions of Real Variables, Vol.
I.,
Ch. VI.-VII.
Boston, 1905.
And
PRINGSHEIM, "Grundlagen der allgenieineu Punktionenlehre," Enc.
Bd. II., Tl. I., Leipzig, 1899.
Wiss.,
d.
math.
CHAPTER
IV

38.
In the usual elementary treatment of the Definite Integral,
defined as the limit of a sum, it is assumed that the function of x

considered may be represented by a curve. The limit is the
area between the curve, the axis of x and the two bounding
ordinates.
For long this demonstration was accepted as sufficient. To-day,

however, analysis is founded on a more solid basis. No appeal
is made to the intuitions of
geometry. Further, even among the
many which
continuous functions of analysis, there are

be represented graphically.
f(x) = x sin
E.g. let
when x 5
0,
cannot
l_
and
Then f(x)
is
continuous for every value of
differential coefficient
It
is
continuous at
.v
= 0,
Also
|/(#)
it is
functions
-/(O)
<
e,
<
when
continuous when .rÔ, since
[cf.
but
it
has not a
because
!/(*)-/(<>) !=!/(*)
and
x,
when x = 0.
^1*1;
x
\
it is
?;,
if
17
<
e.
the product of two continuous
30].
It has not a differential coefficient at
.i'
= 0,
because
and
sin I/A has not a limit as A >0.

It has a differential coefficient at every point
points
/'(tf)=sm---cos-.
76
where
#<0, and
at such

More curious
77
Weierstrass discovered a function, which
still,
is
continuous for every value of x, while it has not a differential

This function is defined by the sum of
coefficient anywhere.*
the infinite series
>
a being a positive odd integer and b a positive number less than

1 + ITT.!
unity, connected with a by the inequality ab
Other examples of such extraordinary functions have been
>
given since Weierstrass's time. And for this reason alone it

would have been necessary to substitute an exact arithmetical treatment for the traditional discussion of the Definite
Integral.
Riemann
treatment.
The
was the first to give such a rigorous arithmetical

The definition adopted in this chapter is due to him.
limitations imposed
we
as
upon the integrand f(x)
will be indicated
proceed.
In recent years a more general definition of the integral has
been given by Lebesgue,

and extended by others, notably
de la Vallee Poussin and W. H. Young, the chief object of
Lebesgue's work being to remove the limitations on the integrand
required in Riemann's treatment.
The Sums S and
39.
given in the interval
s.
||
Let f(x) be a bounded function,
(a, b).
* It seems
Weierstrass
impossible to assign an exact date to this discovery.
himself did not at once publish it, but communicated it privately, as was his
habit, to his pupils

in 1874.
and
Du Bois-Reymond
friends.
quotes
it
in a
paper published
< <
Hardy has shown
>
that this relation can be replaced by

a
1
1, &
Trans. Amer. Math. Soc., 17, 1916].
An interesting discussion of
Weierstrass's function is to be found in a paper, " Infinite Derivates," Q. J. Math.,
f
and
ab^l
London,
[cf.
47> P- 127, 1916,
by Grace Chisholm Young.

' '
Uber die Darstellbarkeit] einer Function durch eine

trigonometrische Reihe." See above, p. 9.
But the earlier work of Cauchy and Dirichlet must not be forgotten.
In his
Cf
classical paper,
de la Vallee Poussin,
Lebesgue, Lemons sur I'Integration, Paris, 1904
And papers by Bliss and Hildebrandt in
;
Integrates de Lebesgue, Paris, 1916.
Bull Amer. Math.

||
Soc., 24, 1917.
The argument which follows
Conrs
is
taken, with slight modifications, from Goursat's
d' Analyse, T. I. (S'e'd.), pp. 171 ft seq.
78
Suppose this interval broken up into

(a, a^),
where
Let
(X,^),
<x <#
l
...
...
partial intervals
0' 7l _i,
b),
< a^ <
6.
in be the upper and lower bounds of f(x) in the whole

inr those in the interval (x r _ lt x r ), writing a = x Q
interval,
r
y
and
and
= xn
S=M (x -a) + M (x -x )+...+Mn (b-x n _

- a^) + 4- mn (b - xn .^.
s=m
(^ - a)+m
Let
and
2 (a; 2
To every mode
of
subdivision of (a, b) into such
sum S and a sum
intervals, there corresponds a
partial
such that
8^8.
The sums S have a lower bound, since they are all greater than
m(b a), and the sums s have an upper bound, since they are all
than M(b a).
Let the lower bound of the sums
of the sums s be J.
less
We shall now
be J, and the upper bound
show that I=J.
Let
a,
x lt
xz
xn _ l
...
be the set of points to which a certain 8 and s correspond.

(xn _ 19 b)
Suppose some or all of the intervals (a, ajj), (x l} x 2 ),
to be divided into smaller intervals, and let
y lt
a,
2/ 2 ,
...
y k _ lt
x lt
y k)
y k+l
...
. lf
x,,
y,,...b
be the set of points thus obtained.

The second mode of division will be called consecutive to the
first,
when
Let 2,
cr
it is
obtained from
be the sums for the
it
in this
new
way.
division.
Compare, for example, the parts of S and

the interval
2 which come from
(a, a^).
Let M\, m\ be the upper and lower bounds of

M'2> m' 2 in (y ly y 2 and so on.
The part of S which comes from (a, x^ is then
f(x) in (a, yj,
),
But the numbers M\, M' 2 ... cannot exceed r

Thus the part of 2 which we are considering is
,
to
M (x
l
at
most equal
a).
Similarly the part of 2 which comes from (x l} x 2 )

equal to JH2 (a;2 a^), and so on.
is
at most
i

Adding these
79
we have 2 =
results
we obtain
Consider now any two modes
or
Similarly
s.
of division of (a,
6).
Denote them by
a,
and
x l}
a,
lt
x.2 , ...
...
x m _ l}
y n -i,
with sums
b,
is
b,
consecutive to
^S
Also, since (3)
is
'
=2
But
2=0-.
Therefore
S^s'
and
mode
s,
(1)
s'
(2)
of division (3),
(1),
and
consecutive to
'
with sums
On superposing these two, we obtain a third

consecutive to both (1) and (2).
Let the sums for (3) be S and o-.
Then, since (3)
S and
cr
s.
s'.
(2),
and
and
<r
^s.
Thus the sum S

not less
arising from any mode of division of (a, b) is

than the sum arising from the same, or any other, mode
of division.
It follows at once that
I = J.
For we can find a sum s as near 7 as we please, and a sum S

(not necessarily from the same mode of division) as near J as we
If /> J, this would involve the existence of an s and an
please.
$ for which
The argument
of this section will offer less difficulty,
if
the reader follow
for an ordinary function represented by a curve, when the

will refer to certain rectangles associated with the curve.
it
40.
to
Darboux's Theorem.
J and
finitely,
I,
when
in such a
The sums
S and
sums S and
tend respectively
the points of division are multiplied indeway that all the partial intervals tend to zero.
Stated more precisely, the theorem reads as follows

If the positive number e is chosen, as small as we please, there
:
a positive number
such that, for all modes of division in
which all the partial intervals are less than or equal to y, the
sum S is greater than J by less than e, and the sum s is smaller
''han I by less than e.
is
tj
Let
be any positive
number
as small as
we
please.
80
Since the sums S and s have J and / for lower and upper
bounds respectively, there is a mode of division such that the
sum 8 for it exceeds J by less than ie.
Let this mode of division be

a,
a2
a{,
...
ap _ lt
Then
with sums
6,
s l ....... (1)
S^J+fc.
be a positive number such that

of (1) are greater than
Let
Let
S and
r]
-all
the partial interval
tj.
a=x
x2
x lt
...
x n _ ly
= xn
with sums
S.2
and
.%, ...(2)
be any mode of division such that

(x r
The mode
a,
is
ajj,
x2
x r -i) =
#3
Oj,
a.2 ,
#4
consecutive to (1) and
Then, by
when r=l,
r],
of division obtained
39,
x n _ lt
...
2, ...
n.
by superposing (1) and (2), e.g.

b, with sums 83 and s3
...(3)
,
(2).
we have
^S
But
S3
Therefore
<
Further,
denoting the upper bound of f(x) in the interval (x, x"),

and the symbol S standing as usual for a summation, extending
in this case to all the intervals (o? r -i, x r ) of (2) which have one of
the points a l} a 2 ... ap - l as an internal point, and not an endJtf (a;', ai")
From
the fact that each of the partial intervals of (1) is

greater than >y, and that each of those of (2) does not exceed q,
we see that no two of the a's can lie between two consecutive o;'s
point.
of (2).
There are at most (p 1) terms in the summation denoted

by S. Let \f(x) have A for its upper bound in (a, 6).
We can rewrite $2 83 above in the form
\
xr )-M(ak x r )}(xr -a b )].

But {M(x r _ l) x r )-M(x r . 1) a k )} and {M(x r _ lt x r )-M(a t ,x r )}
are both positive or zero, and they cannot exceed 2 A.
+ {M(x
r . lt
& - S ^ 2.4 2(o;, - a
Therefore
summation having at most (p

most equal to
the
at
r_
81
J,
1) terms,
and
V-i)
(#,.
being
tj.
Thus
8t
Therefore
since
we have
$3
seen that
So far the only restriction placed upon the positive number

has been that the partial intervals of (1) are each greater than
We can thus choose rj so that
r\
*/.
With such a
choice of
Thus we have shown
$2 < J+c.
tj,
for any mode of division such that

than or equal to a
tJiat
the greatest of the partial intervals is less
certain positive
by
than
less
number
rj,
dependent on
e,
the
sum S
exceeds
e.
Similarly for s and / and it is obvious that we can make the

same satisfy both S and s, by taking the smaller of the two to
which we are led in this argument.
;
rj
41.
come
The Definite Integral of a Bounded Function.

to
function f(x), given in an interval
We now
the definition of the definite integral of a bounded
bounded function
(a, b).
given in the interval (a, b), is said

to be integrable in that interval, when the lower bound J of the
sums S and the upper bound / of the sums s of 39 are equal.
The common value of these bounds I and J is called the
f(x),
definite integral off(x) between the limits
a and
b,
and
is
written
Cb
f(x)dx.
a
Jb f(x)dx
than the sum
*
The bound J
and the bound /
of the
or less than the
of the
sum
cannot be greater
corresponding to any
sums S is usually called the upper

sums s the lower integral.
integral
of
f(x)
82
mode
These form approximations by excess
of division of (a, b).

to the integral.
and defect
We
can replace the sums
=m
a)
(x l
by more general expressions,

Let
(a,
fljj),
(x lt
...
x z ),
fr
...
...
(av-!,
as follows
be any values of x in the partial intervals
x r ),
...
(# n _i, 6) respectively.
The sum
/te)(^-)+/(^)(^-*i)+...+/(^)(6-^-i) ....... (1)

obviously lies between the sums S and s for this mode of division,
since we have mr =f(gr ) =
r for each of the partial intervals.
But, when the number of points of division (x r ) increases indefinitely in such a way that all the partial intervals tend to
inereiore tne
sum
dx.
(i) nas tne same limit.

shown that, for an integrable function f(x),
Thus we have
the
fb f(x)
sum
_o
Cb
has the definite integral
Ja
f(x)dx for
when
its limit,
the
number
of points of division (x r ) increases indefinitely in such a

that all the partial intervals tend to zero, l
n being
2
values of x in these partial intervals*
.
In particular, we may take

for the values of ^, ^2 ... n
,
a,x l ,x 2 ,... x n _^ or x l x z
,
way
any
...
xn ^
b,
Necessary and Sufficient Conditions for Integrability.

is a necessary and sufficient condition
of
the
the
bounded function f(x) given in the
for
integrability
42.
Any
one of the following
interval (a, b)
When any
positive number e has been cliosen, as small as

be a positive number */ sucli that
there
shall
please,
mode
division
of (a, b) in which all the partial
of
for every
I.
Ss<ê
we
intervals are less than or equal to

* We may substitute
rj.
in the above,' for /(])?/(
intermediate between (M^, m^),

in the partial intervals.
(Mz w 2
,
),
etc.,
2)*
/()> any
values
/x 2
... /*
the upper and lower bounds of f(x)
We
8s <
have
J must
And
Thus the condition
Further,
if
I=J,
For, given e,
such that $
which
in
sÎ.
be equal to
is sufficient.
the condition
is satisfied.
by Darboux's Theorem, there
/<
tie
and /
< Je for
S-s = (S-J) + (I-s),

$
Therefore
When any
II.
/.
is
a positive number
every mode of division
the partial intervals are less than or equal to
all
But
we
as stated above.
and
/-/<.
Therefore
rj
e,
S^J
But
83
<
since
r\.
/=/.
e.
number e has been chosen, as small as

a mode of division of (a, b) such that
positive
S-8<6.
been proved in I. that this condition is sufficient. Also
For we are given /=/, as f(x) is integrable, and
necessary.
we have shown that in this case there are any number of modes
s
e.
of division, such that $
It has
it is
<
Let w,
III.
o~
be
There shall be a mode of
any pair of positive numbers.
division of (a, b) such that the sum of the lengths of the partial intervals in
which the oscillation is greater than or equal to M shall be less than cr*
This condition
number
e,
is
take
sufficient.
For, having chosen the arbitrary positive
a"d
2(W=m)
W=
2(6^)'
m are the
upper and lower bounds respectively of /(#) in (a, b).

Then there is a mode of division such that the sum of the lengths of the
partial intervals in which the oscillation is greater than or equal to w
Let the intervals (.rr _ l5 xr ) in which the oscillation
shall be less than <r.
is greater than or equal to to be denoted by l) r
and those in which it is
less than o> by a?,., and let the oscillation (Mr - m r ) in (.rr _], xr ) be denoted
where M,
by o>r
Then we have,
.
for this
mode
of division,
*Cf. Pierpont, Theory of Functions of Real Variables, Vol.
I.,
498.
84
Also the condition
there
is
is necessary.
For, by II., if /(./) is integrable in (<',
wo-.
of division such that S s
Using /),., dr as above,
Therefore
wcr
and
43.
/>),
<
mode
*2/} r
> wS/)
<tr.
Integrable Functions.
If f(x) is continuous in (a, b), it is integrable in (a, b).

In the first place, we know that f(x) is bounded in the interval,
I.
since
it is
continuous in
(a, b) [cf.
31].
we know
that, to the arbitrary positive number e, there

a
corresponds
positive number >; such that the oscillation of f(x)
is less than e in all partial intervals less than or equal to >/
Next,
[cf.
31].
Now we
wish to show that, given the arbitrary positive

of division such that $ s<e
is a mode
e,
with
the
[$42, II.]Starting
given e, we know that for e/(b a)
such that the oscillation of f(x) is
there is a positive number
less than e/(b
a} in all partial intervals less than or equal to
If we take a mode of division in which the partial intervals
then for it we have
are less than or equal to this
r
number
there
r\
>/.
t],
Therefore f(x)
II.
is
// f(x)
In the
is
integrable in
(a, b).
(a, b), it is integrable in (a, b).

note that the function, being given in the
and being monotonic, is also bounded. We
monotonic in
first place,
we
closed interval (a, b),

shall take the case of a monotonic increasing function, so that
we have
f(a} ^f( x ^f( x 2 )

.
for the
mode
of division given
a,
x lt
...
^f(xn .
by
x^
...
x n _^
b.
Thus we have
S=f(x
Therefore,
)(x 1 -a)+f(x,)(x 2 -x l )
if all
...
the partial intervals are less than or equal to

since
/(,) -/(),
are none of
it
/(a^ -/(*,),
. .
/(&) -/(*_,)
them negative.
follows that
Thus/($c)
85
is
integrable in (a,
<
e.
b).
The same proof
applies to a monotonic decreasing function.

seen that a monotonic function, given in (a, 6), can
We
have
have
only
ordinary discontinuities, but these need not be
number
(cf.
We
34).
finite in
are thus led to consider other cases in
is integrable, when discontinuities

of the function occur in the given interval.
simple test of
integrability is contained in the following theorem
which a bounded function
bounded function
points of discontinuity in
III.
(a, b) can
sum of which is
number of intervals
is integrable
in
(a, b),
be enclosed
when
all its
in a finite
the
less than any
arbitrary
number.
positive
Let e be any positive number, as small as we please, and let
the upper bound of \f(x) in (a, b) be A.
\
hypothesis we can enclose all the points of discontinuity

of f(x) in a finite number of intervals, the sum of which is
less than /4>A.
By our
The part of 8 s coming from these

multiplied by their sum.
On
the other hand, f(x)
is
intervals
continuous in
all
is,
at most,
2A
the remaining
(closed) intervals.
We
can, therefore, break
number
of
S-s< Je(cf.
this part of (a, b) into a finite
I.).
Thus the combined mode

is
up
of partial intervals such that the corresponding

portion
such that for
it
<
of division for the
whole of
(a, 6)
e.
Hence f(x) is integrable in (a, 6).

In particular, a bounded function, with only a finite number
of discontinuities in (a, b), is integrable in this interval.
The
discontinuities referred to in this theorem III. need not
be ordinary discontinuities, but, as the function

cannot be infinite discontinuities.
is
bounded, they
86
IV. // a bounded, function is integrable in each of the partial

intervals (a, c^), (a 1? a 2 ),
interval (a, b).
(ap _ l5
...
b), it is
integrable in the whole
Since the function
that
mode
vals, there is
integrable in each
of division for each
of
is
than
s for it is less
where
e/p,
these
(e.g.
e is
inter-
a r -\, a r ), such
any given
positive
number.
Then S
the combined
s for
interval (a, b)
is less
than
Therefore the function
mode
of division of the whole
e.
is
integrable.
clear that if a bounded function
is such that the interval (a, b) can be broken up into a finite
From
the above results
number of open partial

is
it is
intervals,
monotonic or continuous, then
in each of which the function

integrable in (a, b).
it is
V. // the bounded function f(x) is integrable in

|
/(a?)
is also
|
integrable in
E.g. let /(#)
=!
/(*')=
Then
for rational values of
-1
42
the condition II. of
in (0,
but f(x)
is
1),
in (0,
1).
not integrable, for it is obvious that

not satisfied, as the oscillation is 2 in any
is
small.
bounded function f(x) is integrable

number of points in any partial interval
If the
44.
infinite
is
however
for irrational values of
|/(^)| is integrable,
interval,
not greater than
is
be remarked that the converse does not hold.
may
and
then
(a, b).
This follows at once, since S s for \f(x)\

8 for f(x) for the same mode of division.
It
(a, b),
in
(a, b),
of
(a,
there are an
b) at
which
f (x)
continuous.*
Let (>!>
o> 2
>
o>
3 ...
be an infinite sequence of positive numbers, such that
Lt w M = 0.
Let
(a, /?)
Then, by
be any interval contained in

42, III., there is a
of the partial intervals in which

equal to w : is less than (/3-a).
If
we remove from
(i> &) such that

than J) l
<
(a, 6)
/?).
such that a
â< <
ft
b.
these partial intervals, the remainder must cover

can thus choose within (a, /?) a new interval
We
- 04) <
(/?!
at least part of (a,
(a, b)
of division of (a, 6) such that the sum

the oscillation of f(x) is greater than or
mode
(/3
- a) and
the oscillation in (04, fa)
is
less
Proceeding in the same way, we obtain within (a 1}

*
Of. Pierpont, loc.
cit.,
Vol. L,
508.
ft-,)
new
interval

(a 2 ,
such that OSg
/8 2 )
than
<o 2
And
Thus we
%) and
Oj)<^(/3j
87
the oscillation in
(<x 2
/2 2 )
ess
so on.
find
an
J 15 J 2
infinite set of intervals
>
eac ^ contained entirely

M-QC, and the
within the preceding, while the length of J,, tends to zero as

oscillation of /(#) in ^4 W also tends to zero.
By
lies
the theorem of
within
Let
all
18,
the set of intervals defines a point
which
be any positive number, as small as we please.

in the sequence w 1} o> 2 ... a number <o r less than e.
be the corresponding interval (a r J3r ), and 77 a positive number
Then we can choose

Let A r
(e.g. c)
the intervals.
smaller than
(c
- a r) and
(J3 r
- c).
7
/
|/(#)-/(c)| <e, when |#-e|
and therefore we have shown that/(.r) is continuous at <\
Since this proof applies to any interval in (, 6), the interval (a, /?) contains
an infinite number of points at which f(.v) is continuous, for any part of
(a, /?), however small, contains a point of continuity.
Then
Some Properties
45.
of the Definite Integral.
establish some of the properties of

bounded in (a, b) and integrable.
I.
If f(x)
interval
is integrable
(a, /3)
in
contained in
f*
I
We
shall
now
f(x) dx, the integrand being
(a, b), it is
also integrable in
any
(a, b).
From 42, I. we know that to the arbitrary positive number e

there corresponds a positive number r\ such that the difference
$ s
e for every mode of division of (a, b) in which all the
<
partial intervals are less than or equal to rj.

can choose a mode of division of this
We
kind with (a, ft) as
ends of partial intervals.
Let 2, a- be the sums for the mode of division of (a, /3)
included in the above.
Then we have
Thus f(x)
0^2-o-^*S-s<
integrable in (a,
is
the value
/3) [
e.
42, II.].
of the
If
integrable function f(x) is altered
a finite number of points of (a, 6), the function <f>(x) thus
obtained is integrable in (a, b), and its integral is the same as
II.
at
that of f(x).
can enclose the points to which reference is made in a
finite number of intervals, the sum of which is less than
e/4.4,
We
where
bound
of
is
any given
<j>(x)
in (a,
b).
positive number,
and
is
the upper
88
The part of S s for (f>(x), arising from these intervals, is at

most 2A multiplied by their sum, i.e. it is less than ic.
On the other hand, f(x) and <j>(x), which is identical with f(x)
in the parts of (a, b) which are left, are integrable in each of
these parts.
Thus we can obtain a mode of division for the whole of them

which will contribute less than Je to S s, and, finally, we have
a mode of division of (a, b) for which S s
e.
<
Therefore
<j>(x)
is
integrable in (a,
b).
Ib
(f>(x)dx=
it.
f(x)dx.
<(,
Iba <f>(x)dx
the limit of
the intervals (a, x^, (x lt x.2 ), ... (x n _ l b) tend to zero, and

n are an y va l ues f x i n these intervals.
f
when
'
is
We may
put /(), /(&), .../(&) for <(), 0(),
this sum, since in each interval there are points at
...
in
^(f )
which 0(&)
and f(x) are equal.

In this
way we
obtain a
sum
Jbaf(x)
III. It follows
of the
form LtZ/(
.)(#,.
%r-i),
dx.
immediately from the definition of the integral,

b), so also is Cf(x\ where G is any
that if f(x) is integrable in (a,

constant.
Again, if fi(x)
and
f.,(x)
are integrable in
(a, b), their
sum
is
also integrable.
For, let S, s S' s' and 2, a- be the sums corresponding

same mode of division for/ô;), fz (x) and/1 (a?)+/2 (o5).
Then it is clear that
;
and the
Also
result follows.
it is
easy to show that

b
[
Ja
and
Cf(x)dx=C\J a'f(x)dx,
to the
89
IV. TJie product of two integrable functions fi(x), f^( x )
i-
integrable.
To begin with, let the functions /1 (o5),/2 (a5) be positive in (a, 6).
Let M r mr M' r m'r
r m,. be the upper and lower bounds of
;
fi( x )ifz( x ) an d/i(#)/2(X) i n th e partial interval (x r _ lf x r ).

Let S, s S', s and 2, <r be the corresponding sums for a certain
mode of division in which (x r _i, x r ) is a partial interval.
;
Then
clear that
it is
m = M M'
M,.
A fortiori,
m m' = M (M' m' + iri,(M m

M _ ^ < M(Mf _ m^ + M ^ Mf _ m^
r
M are the upper bounds
where M,
Multiplying this inequality
sponding
results,
It follows that
is
interval,
by
r)
ot f^x),
(x r
#,._,)
fz (x)
in (a,
r ).
b).
and adding the
corre-
we have
2 - a- = M(S' - s) + M'(S - s).
o-
integrable in (a,
If the
tends to zero, and the product of /i(X),
/2 (#)
b).
two functions are not both positive throughout the

we can always add constants c l and c2 so that fi(x) + c lt
,
f2 (x) + c.2 remain
positive iu (a,
6).
The product
is
(/iO) + ^(Mx) + c 2 ) =/; (x)f.2 (x) + Cl f2 (x) + cj, (x) + Cl c 2

then integrable.
But
c l 2 (x)-\-c 2fl (x)

c l c., is integrable.
It follows that f1 (x)fz (x) is
integrable.
On combining these results, we see that if f^x),

are integrable functions, every polynomial in
?i
46.
integrable function
f.2 (x) ...
/(#)
Properties of the Definite Integral
(continued).
a
I-
Ja
f(x)dx=-(Jb f(x)dx.
In the definition of the sums
*
[of.
is,
that a
s,
and of the
was
less
however, unnecessary, and will
now
Jbaf(x)dx,
restriction
we assumed
S and
than
definite
b.
This
be removed.
This result can be extended to any continuous function of the n functions
Hobson,
loc. cit;, p. 345].
90
If
>
6,
we take
as before the set of points

a,
X n _i,
...
x^)
cc.p
o,
and we deal with the sums

s
=m
a)
(x l
The new sum S

to the
sum
+ m.
+ mn (b
^ H _ 1 )- J
when they have a common

It is thus clear that,
s in (1) follows,
common
defined as the
#11
bounds of S and
of the
definite integral is
and the
value of these bounds,
value.
with this extension of the definition
we have
f
(v>
Ja
a, b
^ n _2>
# H _ 15
0,
41,
x^)
obtained from
The existence
of
(x z
equal in absolute value, but opposite in sign,
is
f(x)dx=-\J b f(x)dx*
being any points of an interval in which f(x)
bounded and
is
integrable,
II.
Let
c be
any point
bounded and
of
an
interval
(a, b)
in ivkich f(x)
is
integrable.
*
Then
Ja
f(x)
.
= P/O) dx + { V'O) dx
J
Ja
Consider a mode of division of (a, b) which has not c for a

point of section. If we now introduce c as an additional point
of section, the
sum S
But the sums S
is certainly not increased.

for (a, c) and (c, 6), given
division, are respectively not less
Thus every mode
than
Ja
by
this
f(x) dx and
of division of (a, b) gives a
Jc
mode
of
f(x) dx.
sum S not less than
[f(x)dx+[Jc f(x)dx.
Ja
It follows that
b
f(x)dx^ Ja
[ f(x)dx+ [ f(x)dx.
Jc
a
But the modes
possible
mode
of division of (a, c)
of division of (a,
6).
and
(c,
And we
b) together
form a
can obtain modes
* The results proved in

42-45 are also applicable, in some cases with slight
verbal alterations, to the Definite Integral thus generalised.

of division of (a,
and
c)
(c, 6),
91
the sums for which differ from
fb
Jcaf(x)
dx and
Jc
f(x) dx, respectively,
as little as
by
we
please.
sign of inequality in the above relation

must be deleted, so that we have
It follows that the
fV(a) dx
Ja
f'/(a)
Ja
dx+ \
Jc
f(x) dx.
If c lies on (a, b) produced in either direction, it is easy to

show, as above, that this result remains true, provided that f(x)
is integrable in (a, c) in the one case, and (c, b) in the other.
47.
If f(x)
= g(x), and both functions are integrable in
(a, b),
then Pf(x)dx^Pg(x)dx.
0(0) =/()-</() SO/
Let
Then
sum s,
<f>(x)
integrable in (a,
is
Therefore
J*
COROLLARY
We
and obviously, from the
dx - P g(x) dx ^ 0.
If f(x) is integrable in
I.
have seen
f(x)
b),
43 that
in
if
f(x)
is
(a, 6),
then
integrable in (a,
6),
so also
is|/(a)|.
And
The
-\f(x)\^f(x)^\f(x)\.
result follows
COROLLARY
in
(a, b).
II.
If f(x)
from the above theorem.
Let f(x) be integrable

is
continuous at
in
and never negative

and /(c)>0, then
(a, b)
[
Ja
f(x)dx>0.
We have seen in 44 that if f(x) is integrable in (a, 6), it must

have points of continuity in the interval. What is assumed here
is that at one of these
points of continuity f(x) is positive.
Let this point c be an internal point of the interval (a, b), and
not an end-point.
Then there is an interval (c, c"), where
92
tt<c'<c<V'<6, such
that f(x)>k for every point of (c,

k being some positive number.
c"),
dx = 0.
Jc'a f(x)
> k in (c,
And, since f(x)
c"),
f(x) dx
^ k(c" - c) > 0.
Jc'
Also, since /(a;)
Adding these
results,
in
(c", 6),
we have
Jn
The changes
in the
f(x)dx^0.
f(x) dx
argument when
>
c is
0.
an end-point of
(a, b)
are slight.
COROLLARY III. Let f(x) = g(x), and both be integrable in

At a point c in (a, b), let f(x) and g(x) both be con-
(a, b).
Then
tinuous, andf(c)^>g(c).
Ja
f(x)dx^>\ g(x)dx.
Ja
This follows at once from Corollary
II.
by writing
<f>(x)=f(x)-g(x).
By the aid of
may be obtained
If f(x)
the theorem proved in
44, the following simpler result
> g(x\
and
both are inteymble In (a,
Z>),
then
For, if f(x) and g(x) are iutegrable in (a, 6), we know that f(je)-g(x) is
integrable and has an infinite number of points of continuity in (a, b).
At any one of these points f(x)-c/(x) is positive, and the result follows
from Corollary
II.
Let <j>(x), \^(x) be

48. The First Theorem of Mean Value.
two bounded functions, integrable in (a, b), and let \fs(x) keep
the same sign in this interval e.g. let \fs(x) = in (a, b).
Also let M, in be the upper and lower bounds of <j>(x) in (a, b).
Then we have, in (a, b),
;
m = (p(x) = M,
and multiplying by the factor
\fs(x),
which
m\Is (x) ^(f>(x)\!s (x)

It follows
from
Ja
<f>(x)\fs(x)
^ M\fs
not negative,
(x).
47 that
m f ^ (x) dx ^ [
since
is
Ja
is
(x)
^ (x) dx ^ M\J V (x)
also integrable in (a,
b).
clr,
93
fb
dx = /UL\
Jba $ (x) \}s (x)
Jo.
where
/x
is
dx
some number satisfying the relation mîu. = M.

argument applies also to the case when
It is clear that the
If
IJL
(f>(x)
is
continuous in
(a, b),
we know
that
takes the value
it
some value of x in the interval (cf. 31).

We have thus established the important theorem
for
are two bounded functions, integrable in (a, b),

and \[s(x) keeping the same sign in the
continuous
being
If
(f>(x), \js(x)
(j)(x)
interval, then
f*
r&
(
x)
Ja
x ) ax
(f )
Ja
x ) dx
>
in a.= x = b.
Further, if (f>(x) is not continuous in (a, b), we replace <(f)
where
= = M, m, being the
by
satisfies the relation
bounds of <fi(x) in (a, b).
This is usually called the First Theorem of Mean Value.
where (
is
some
JUL,
As a
definite value of x
/UL
particular case,
<j)(x)dx
when
= (b
</>
(x) is
continuous,
where a =
u)0(),
fj.
...
It will be seen
from the corollaries to the theorem in
^ ^
we can
47 that in certain
< <
b by a
b.*
replace a
f
However, for most applications of the theorem, the
ment in the text is sufficient.
cases
more general
state-
49. The Integral considered as a Function of its Upper Limit.

Let f(x) be bounded and integrable in (a, b), and let
X
F(x)=\Ja f(x)dx,
where x
Then
is
any point
if (x
h) is
in (a,
b).
also in the interval,
f(x)dx.
fx+h
X
F(x + h)-F(x) = nh,

where m = /x = M, the numbers M, m being the upper and lower
Thus
bounds of/(#)
It follows
in (x,
x + h).
that F(x)
is
*Cf. Pierpont, he.
a continuous function of x in
cit.,
Vol.
I.,
pp. 367-8.
(a, b).
94
Further,
if
f(x)
continuous in
is
(a, b),
where x = g= x + h.
F(x + h) - F(x) = hf(g),
When
h tends to
zero,
/() has
Lt
Therefore
Thus ivhen f(x)

in
is
the limit f(x).

-
continuous in
=f(x).
(a, b),
f(x)
Ja
dx
is
continuous
and has a differential coefficient for every value of

this differential coefficient being equal to f(x).
(a, b),
(a, b),
x in
is one of the most important theorems of the Calculus.

shows that every continuous function is the differential co-
This
It
a continuous function, usually called its primitive, or

indefinite integral.
It also gives a means of evaluating definite integrals of continuous functions. If f(x) is continuous in (a, 6) and
efficient of
F(x)=[ f(x)dx,
then
we know
or other,
d
that
we have
Suppose
-j-F(x)=f(x).
that,
obtained a continuous function
by some means
(j>(x)
such that
d
must then have F(x) = <j,(x) + C,
We
/
7 \
since
,-
(*&
J*.
(F(x)
<j>(x))
=Q
in
(a, b).*
To determine the constant

vanishes at x
Thus we have
Ja
50.
we
C,
use the
fact
that F(x)
= a.
f(x)
dx =
(x)
</>
(a).
The Second Theorem of Mean Value.
We now
come
to
a theorem of which frequent use will be made.
and therefore integrable, in (a, b)

and let \fs(x) be bounded and integrable, and not change sign
more than a finite number of times in (a, b)j
I.
Let
<j>(x)
*Cf. Harcty,
be inonotonic,
toe. cit., p.
tThis restriction on
condition imposed upon
Hobson,
loc. cit., p.
228.
\f/(x)
For a proof in which the only

bounded and integrable in (a, 6), see
can be removed.
\^(x] is
that
360, or Goursat,
it
is
toe. cit.,
T.
I., p.
182.
95
rf
fia
<l>(x)\!s(x)dx
= (l>(a)\
Ja
where g -is some definite value of x in a = x = b.

For clearness we shall take <j>(x) monotonic increasing in
The modifications in the proof for a monotonic decreasing
(r/., fr).
function will be obvious.
we assume that \!s(x) does not change

number of times in (a, b), we can take
Since
finite
a2
a ly
such that
\fs(x)
more than a
keeps the same sign in the partial intervals

(a
K,
aj),
()
\/r (iK)
t&5
=V
f/;
a
First
where
0(r-i)
Therefore
(a)
(as)
of
(a n _!,
an ).
rr
I
Theorem
...
2 ),
71
9!)
Now, by the
an =
a n _ lt
...
sign
(ir) T/^ (05)
d.
r _i
Mean
Value,
= v* = $(,)
cioj
/uL r
[F(
r)
- F(a
._ l )] >
Ja,._]
where we have written F(x)=
Ja
\js(x)dx.
Thus we have
Since F(a) = 0,
(1) in the form
we may add on
the term $(a)F(a), and rewrite
aO<te
= *() [*()-/,]"'
(2)
The multipliers of F(a), F(a

must be negative or zero.
),
...
^(6) in all but the last line
06
We may
therefore replace the
some
is
least of F(a),
or other.
But, since.
in (a,
+ 1)
terms by
number intermediate between the

F(a l ) ... F(a n _ ), F(b), or coinciding
definite
greatest and
with one
of these (n
M[0 (<<)-
-e.by
where
sum
F(x)=
we know
\fs(x)dx,
that F(x)
is
continuous
b).
Therefore
value of x in a
It follows
may
= x^b
from
be taken as F(g ), where
(2)
is
some
definite
31).
(cf.
that
c&
(fa
V,(a)<:Za
+ 0(&) J
\ls(x)dx,
where ^ is some definite value of x in a = x = b.

Thus the theorem stated above is proved.
Jba f(x)dx
the value of f(x) at a finite
Now
(f)(x)
Also
exist.
is
number
monotonic in
we may
give
(a, 6),
0(.T)
is
unaltered
of points in
if
we change
(a, b).
and therefore 0(
+ 0),
respectively, without altering its monotonic character or
We
thus * obtain the result
II.
Let
and
(j>(x)
let \[s (x)
be
be
monotonic,
bounded and
0(6
0)
these values at a'^ft and x = b,
changing
and
therefore integrable, in
and
integrable,
in (a,
more than a finite number of times

Then
not change
(a, b)
its
sign
b).
Fb
fba
where
*
<j>(x)\ls(x)dx
is
some
= (j>(a + 0)\ff \lr(x)da+<j>(b-Q)\

J
Ja
definite value of
\fs(x)dx,
ij
x in a = x = b.
This and the other theorems which follow could be obtained at once by
suitable changes in the terms </>(a) and <J>(b) on the right-hand side of (2).
making

Also
it is
and 0(6
that
we
clear that
0),
could in the same
respectively,
A = 0(a + 0)
and
way
by any numbers
B = 0(6
0)
97
replace 0(tt
and B, provided
in the case of the monotonic
increasing function; and J.

0(&+0), .BÊ 0(6 0) in the case
of the monotonic decreasing function.
We thus obtain, with the same limitation on 0(#) and ^(#)
as before,
III.
(x)
Ja
dx + B ^(x) dx,
^ (x) dx = A Ja
{ ^ (x)
Jf
\
A = (j>(a-}-0) and -Bi^ 0(6 0),

creasing, and A = (a + 0), B= 0(6
where
monotonic in0), if <f)(x) is monotonic

some
value
decreasing,
being
definite
of x in a = x = b.
in I., II., III. need not, of course, be the same,
The value of
and in III. it will depend on the values chosen for A and B.
Finally, as in III., we may take .4=0 and 5 = 0(6), when
0(#) = and is monotonic increasing in (a, 6).
Thus, with the same limitations on \^(#) as before, when
0(a?) i=0 and is monotonic increasing in (a, 6), we have
if
(f>(x)
is
(/>
rb
IV.
Ja
where
is
some
Again, when
may
take
(f>(x)\/s(x)dx
= (j>(b)\
\j,(x)dx,
Jf
in a x = b.
monotonic decreasing in
definite value of x
0(#)
A = </>(a)
and is
and
B = 0,
same limitations on ^(x) as
(a, b),
we
obtaining in this way, with UK*
before,
Jba (t>(x)\ls(x)dx
= (j>(a)\f
Ja
\f,(x)dx,
where is some definite value of x in a = x = b.

Theorems IV. and V. are the earliest form of the Second
Theorem of Mean Value, and are due to Bonnet,* by whom they
were employed in the discussion of the Theory of Fourier's
Series.
The other Theorems
I.,
II.,
III.
can be deduced from
Bonnet's results.
Theorem
I.
was given by Weierstrass
in his lectures
and
Du
Bois-Reymond,t independently of Bonnet.
Theorem
II. is
the form in which
use the Second Theorem of

*
Bruxelles,
Mdm.
cour.
Acad.
Mean
we
shall
roy., 23, p. 8, 1850; also /.
p. 249, 1849.
t.7.
Math., Berlin, 69,
c. i
p. 81,
most frequently
Value.
1869; and 79,
p. 42, 1875.
Math., Paris, 14
98
INTEGRAND BOUNDED.
INTERVAL INFINITE.
INFINITE INTEGRALS.
In the definition of the ordinary integral
51.
in the preceding sections
f
J
f(x) dx,
and
(i
chapter, we have supposed

in the interval of integration
of
this
that the integrand is bounded

which extends from one given point a to another given point b.
We proceed to extend this definition so as to include cases in which
the interval increases without limit,

the integrand has a finite number of infinite discontinuities.*
(i)
(ii)
I.
Integrals to +
o>
f (x) dx.
Ja
Let f(x) be bounded
where a
and
is fixed
and
b is
integrable in the interval
any number
greater than
(a, b)
a.
We
Px
fa
dx as Lt
f(x)
We
speak of
x>x J a
exists, j
f(x) dx in this case as
an
f(x) dx,
when
infinite integral,
this
limit
and say that
it
converges.
On
the other hand,

p*
infinite
integral
I
'
Ja
tion of divergence to
EX.1.
when
Ja
f(x}dx tends
to
QC
as .r->cc,
f(x}dx diverges to x, and there
is
we say
that the
a similar defini-
oo
of
Ja
f(x)
d.c.
[V*^=l; Ji
P^f=2.
ff*
Jo
f e-x
For
Jo
dx= Lt \*-*da:= Lt (l- c -*)=l.

x >ao J o
x >oo
And
Ex.2.
*d*=
For
*For the
definition of the
Lt
e*dx= Lt e*-l = oc.
term "infinite discontinuities," see
33.
more convenient to use this notation, but, if the presence of the

variable x in the integrand offers difficulty, we may replace these integrals
fit
is
by f f(t)dt and Lt
Ja
*->
Ja

And
99
Lt
"
Similarly
"
- dv = -
log
Ji
oo
Js
- oo
-^
l-#
==
oc
These integrals diverge to oo or

as the case may be.
Finally, when none of these alternatives occur, we say that the
integral
f(x)dx
sin
xdx
oscillates finitely.
xsinxdx
oscillates infinitely.
f (x) dx.
Jb-co
When
infinite
16 and 25.
oscillates finitely or infinitely, as in
Ex.3.
where b
bounded and integrable in the interval (a, b),

is any number less than b, we define the
f(x) is
and a
is fixed
Cb
}b-
f(x)
dx as Lt
when
f(x) dx,
X > -CO J X
05
this limit exists.
,b
We
speak of
f(x)dx as an
infinite integral,
and say that
it
converges.
J-*>
The
cases in which
f(x) dx
is
said to diverge to
oo
or to
- oo
or to
./-ao'
oscillate finitely or infinitely, are treated as before.
Ex.1.
Ex.
2.
I
e~ x dx diverges to
oo
x dx diverges
to
rsinh
-x
90
sin.rcfo? oscillates finitely.
.'-oo
III.
x sin xdx
Integrals from
oo to oo
If the infinite integrals
vergent
Since
and
^s
equal
oscillates infinitely.
to tfieir
J -
J -
f (x) dx.
CO
f(x)dx and
oo
sum.
P/fa) dx = \f(x) dx + f'/H dx,

Ja
Ja
Ja
Ja
f(x)dx are both

f(x)
fco
dx
< a < x,
is con-
100
/*>
two
follows that, if one of the

converges, the other does.
it
Also
Ja
f(x)
integrals
f(x)
Ja
dx
/(./:)
Ja
integrals
f(x)dx=[
J
f(x)
a
co
or
dx + ["/(a?) dx.
<a<,
Jx
J["
dx
dx =
and, if one of the two

converges, the other does.
dx or
dx
f(x)
a
oo
f(x)dx.
f(x)dx+[
Ja
"
Thus
~s;
f(x)
[ f(x) dx = Jx
["/<#) dx+ ["/() <&>
Ja
Similarly,
Also
dx +
f(x)
f
J-oo
and the value
f(x)
f(x)
dx
Ja
f"
of
J -
dx =
f(x)
dx +
is
f/{aj)
<fce,
Ja
J-oo
independent of the point a used
CO
in the definition.
r ~' ,=7r, J-oo

r*-+b-*rr***.
Jo
EX.
J.ool+A^
A necessary
52.
and
sufficient condition for the
convergence
f(x)dx.
off
Ja
C
Let
F(x)=\
f(x)dx.
The conditions under which F(x)

27 and 29.
have been discussed in
integral
we
are thus able to say that
f(x)dx
is
when, any positive number
II.
further
and has
the
value
J,
we\
provided that x = X.
necessary
of the integral
.
infinite
e having been chosen, as small as

such that
a positive number
-[f(x)dx
Ja
And
have a limit as x->vc
In the case of the
convergent
fee
a'
please, there is
shall
and
sufficient condition
p
Ja
f(x)dx
is that,
when any
for
the convergent
positive
number
'
101
has been chosen, as small as we please, there shall be a positive

f*"
such that
number
f(x)dx\<e
for all values of
x',
x" for which x'">x'
= Ji.
/co
We
have seen in
51 that
if
Ja
r/(x) dx = Jf
Ja
"
/(a?)
a
dx + ^f(x) dx,
J
number
f
Jx
if
These
dx converges, to the arbitrary
f(x)
number
that
Also,
<
a.
fee
a
positive
dx converges, then
f(x)
/(*)
dx
X such
when x =
<,
this condition is satisfied, the integral converges.
results,
and the others which
immediately to the
follow, can be extended
infinite integral
f(x)dx.
f (x) dx.
53.
J
positive
integrand f(x)
is
r.c
when x>a,
it is
clear that
Ja
f=o
ai verge to oo
I.
If the
Integrand Positive.
"-
f(x)
f(x)dx
dx must
is
a monotonic
in-
either converge or
It will converge if there is
Jf(x)dx<A
a
when x>a, and in
It will diverge to oo if there is
a positive number
this case
Ja
such that
f(x)dxÂ.
no such number.
These statements follow from the properties of monotonic

functions
34).
Further, there
is
an important
vergence of integrals
when
"
comparison test
the integrand
"
for the con-
is positive.
II. Let f(x), g(x)

be two functions which are
bounded and integrable in the arbitrary interval (a,
let
g(x)=f(x) when x = a.
Jco
a
g (x) dx
Then, if
is convergent,
Ja
f(x)dx
and
b).
Also
is convergent,
/OD
/c
Ja
positive,
g (x) dx =
Ja
f(x) dx.
it
1C 2
For from
we know
47
g(x)dx^\J a f(x)dx, when
Ja
Therefore
J[a
that
</
(x)
dx
<
Ja
>
a.
f(x) dx.
Ja
/X
Then, from I,
05
poo
g(x)dx^\J a f(x)dx.
HI- If 9 (%)=/(&)> and
|CO
Ja
f(x)dx diverges, so also does
JCQ
a
g(x)<to*
This follows at once, since

Ja
g(x)dx^\J a f(x)dx.
7
aX
where
{00
We
have
Ja
^g,
fxa #
Thus,
when n > 1
And, when
n {^-"-a
= log x
,
1,
log a,
Lt
Xn
Ja
*
n=f=l,
"
-
x^<o} a
i.e.
when n = 1
a:^>3o J a
Lt
-"},
^=^
when
= 00,
diverges.
Since the relative behaviour of the positive integrands f(x) and

these conditions may be expressed in terms of limits
only as x-+<x>
When
fj(x)jf(x)
has a limit as x-oc
y(x) cfo converges,
if
if /
f(x)
y(x)/f(x) has a limit, not zero, or diverges, as a:^<x
dx
matters
diverges.
f(x) dx converges.
.'a
When
rj(x)
:
.'
y(x)
dx
diverges,
/""sin
3.
since
efa? converges,
%j-
Ja
sin 2 #
=-
^_ V"
1
-g,
103
when
#â > 0.
_
fte
The integral
Absolute Convergence.
54.
when
be absolutely convergent
the arbitrary interval (a,
f(x) is
and
b),
Ja
Ja
f(x)
dx
is
said
to
bounded and inteyrable in
\f(x)
dx is convergent.
Since
f(x)
dx
\f(x)
dx, for x"
>x'â
(cf.47,Cor.
it
follows from
that
52, II.
if
Ja
\f(x)
dx converges,
L),
so also does
p<
Ja
f(x)dx.
But the converse

type
may
converge,
is
not true.
and
An
infinite integral of this
yet not converge absolutely.
For example, consider the integral
The Second Theorem
Mean Value
of
50)
shows that
this
integral converges.
For we have
(**"
Jt
since
1 f*
dx = -,
,
sm x dx-\-
Jx'
1 f*"
sin
x dx,
where
But
I
sin
x dx
Therefore
Thus
provided that
and
sin
x dx are each
^^^
--
(x
less
than or equal to
<e, when x">x' = X,

fr.S2.ifa
X
IJ^'
4
2.
104
sin
Therefore
value
its
if
dx converges, and we
Joo
To prove
dx
this, it is
diverges.
only necessary to consider the integral

since
is
any
nn sin
!
have
on putting x = (r
But the
sin
,
\
'
dx.
&
1)7r
71
sin^/
dx=\
-r
-dy,
'
-f y.
TT
Isinz
series
r?r
Jo(f-l)7r + 2/
%
1
V7r
._
J(r-l)ir
\sinx-
J(r-l)n
"
dx =
"
rir
Therefore
JO
But
positive integer.
I7
We
'cfaj,
88 that
is
But the integral
where n
shall find in
a;
dx>- If smydy
rTrJo
77
on the right hand diverges to

nn
mu
c
Therefore
T
Lt
i.
fi i
oo
as
n >co JO
But when
Lt
Therefore
When
infinite integrals of this type converge, but do not conthe convergence must be due to changes of sign
absolutely,
verge
in the integrand as #->oo
.
55.
I.
(a, b)
The M-Test
for the
Let f(x) be bounded
where a>0.
If
Convergence of
and
f(x)dx.
Ja
integrable in the arbitrary interval
there is
a number
/u.
greater than
such
105
MO
that xf(x)
bounded when xâ, then
is
f(x)dx converges
absolutely.
Here \x f(x) \<Â, where

and x = a.
fl
Thus
some
is
definite positive
number
l/(z)l<^.
-
^ converges.
Jdx
a
It follows that
fee
a
Let f(x)
II.
interval
(a, 6),
or equal to
|/0) dx converges.
I
Ja
f(x)
dx converges, and the convergence
is
absolute.
be bounded and integrable in the arbitrary

less than
where a>0. // there is a number
such that x*f(x) has a positive lower bound
//,
/.CO
when x = a, then
Ja
f(x)dx diverges
Here we have, as
It follows that
Ja
when x = a.
=f(x)
diverges to
It follows that
Jo
III.
to
equal
f(x)
oo
when
/*
dx diverges
Let f(x) be bounded
interval
before,
,
But
to oo
and
'
!.
to
integrable in the arbitrary
less than or
where a>0. If there is a number
such that x*f(x) has a negative upper bound
(a, 6),
jj.
when x = a, then
F"
I
Ja
f(x)
This follows from
II.,
dx diverges
oo
to
for in this case
must have a positive lower bound when x = a.

But,
if
Lt
(af/(o?)) exists, it
follows that arf(x)
is
bounded
in
./_> to
# = a;
will
also,
either
number X, x*f(x)
lower bound, when this limit is
by properly choosing the

have a positive
positive
106
upper bound, when this limit is negative

that
x
X.
provided
Thus, from I.-IIL, the following theorem can be immediately
positive, or a negative
deduced
Let f(x) be bounded

ivhere
(a, b),
If there
and
a>0.
EX-)MB
a number
is
dx
f(x)
/m
greater than
such that Lt (xrf(x))
converges.
..
// there
a number
is
jm
than
less
or
Ja
and
the
We
it
same
make very
shall
as the "/x-test."
ing the integral
Ja
such
f(x)dx diverges;
+ oo
true if xrf(x) diverges to
is
to
equal
or to
oo
as
frequent use of this test, and refer to

clear that we are simply compar-
It is
f(x)
dx with the
integral
Ja
OP"
and deducing
the convergence or divergence of the former from that of the

latter.
Ex.1.
rf
r Y^-^
6 ^ +
dx
:>
<j!
diverges,
It should be noticed that the

,,
to the integral
t
^^
x* x y^
2 V
2
.r
If
Ja
+ c29 /
=7962
theorems of this section do not apply
dx.
(j>(x)
dx
<j>(x) \/s(x)
For we have
|
number and x
poo
Ja
f (x) dx.
bounded when x = a, and integrable in
is
f*a
then
Lt
c-MoA
Further Tests for the Convergence of
56.
I.
/""sin
since
a.
\]s
(x)
dx converges
the
absolutely,
is absolutely convergent.
(.#)
\<A, where A
is
some
definite positive
107
Also
^(x)\dx<A\Jx' \^(x}\dx,
</>(x)
Jx'
when
x"
>
/GO
we
Since
are given that
dx converges, the
\\Is(x)
result
follows.
(
'
Ex.
1.
X ^dx,
Ja
Ja
^dx
converge
absolutely,
when n and a
are
positive.
/oo
2.
e~ ax cos
Ja
bxdx converges
absolutely,
when a
is
positive.
converges absolutely.
Let <j)(x) be monotonic and bounded when x = a. Let \/s(x)

bounded and integrable in the arbitrary interval (a, b),
and not change sign more than a finite number of times in the
II.
be
Jco
a
\fs(x)dx converge.
dx
</>(&) \js(x)
converges.
Jco
a
This follows from the Second Theorem of
Jjc"
x'
<j)(x)\ls(x)dx
= (/>(x')\f
J x'
\ls(x)dx+(j>(x")\
less
Value, since
Cx"
where a<^x' = =x".

But (f>(x ) and <j>(x") are each
!
Mean
\fs(x)dx,
than some definite positive
number A.
Also
we can
choose
X so that
\Is(x)dx
and
are each less than e/2^1, when x">x'

number, as small as we please.
It follows that
= X,
positive
<e, when
and the given integral converges.

EX.
i.
r<e~ xr
sin.r
dx converges.
.'o
2.
Ja
(1
- e~x )
X^-
dx converges when a>0.
and
is
any given
108
Let
III.
be
<f>(x)
monotonic and bounded when x = a, and
Lt
Let
bounded and integrable in the arbitrary interval

not change sign more than a finite number of times
\js(x) be
and
(a, 6),
in the interval.
Also
let
Ja
Then
Ja
As
dx
\js(x]
be
bounded ivhen x>a.
(j>(x)\p-(x)dx is convergent.
above, in
II.,
we know
Cx"
(j>(x)\ls(x)dx
that
= (j>(x']
\js(x)dx,
where a<x' = g=x"

Cx
But
Ja
\fs(x)dx
when x>a, where
is
some
definite
positive number.
And
\[,(x)dx
\!,(x)dx
j*
<2A.
Similarly
Lt
Also
<
.,> GO
Therefore, if e is any positive number, as small as

such that
there will be a positive number
we
\i) ( x )\
<^7~4
>
when x = X.
It follows that
p"
<j)(x) \fs(x)
dx <e, when x'">x' = X,
Jx'
/
and
Ja
Ex.1.
<j>(x)\ls(x)
f^SlQ.V
I
Ja
dx,
dx converges.
r^COS.V
I
Ja
Xn
T
dx converge when n and a are
converges.
positive.
please,
109
The Mean Value Theorems for the Infinite Integral.

The First Theorem of Mean Value.
<b(x) be bounded when x^.a, and integrable in the arbitrary
57.
Let
interval
,*)
Let ^{x} keep the same sign in x^-a,
and
Ja
ty(x)dx converge.
Then
m^p^M,
where
upper and lower bounds of
the
<f>(x)
x^~a
in
being
and m.
We
and,
w = ^(#) = M,
have
if
m }(x) ^ <(*) +
m I* ^(x)dx^.
Therefore
Ja
But,
when #j:a,
VT(#)> 0,
by
56,
^ M^(x).
when x=â.
(* <j>(x)\jr(.v)dx^M {"-^(^dx,
Ja
Ja
<^(x)^(x}dx converges, and we are given
I.,
Ja
(of)
that
-^r(x)dx converges.
Thus we have from these
inequalities
m r ^(x) dx ^
Ja
In other words,
where
58.
Ja
Ja
</>(.r)
^(x) dx
tj>(x)\(<(x)dx**iil
ÂfT
^(.r)
Ja
d.,>.
^(.)-)d,v,
-'a
m^p^M.
The Second Theorem of Mean Value.
LEMMA.
Let
Ja
Then F(x)
f(x) dx be a convergent integral,
continuous when
is
and F(x) =
Jx
f(x) dx (x^.a).
#^, and bounded in the interval
(<(,
takes at least once in that interval every value between its upper
bounds, these being included.
it
The continuity
of
oc
).
and
Also
lower
F(x) follows from the equation
Further, Lt F(x} exists and
is zero.
ar->oo
It follows
from
32 that F(x) is. bounded in the interval (, oo ), as defined

are its upper and lower bounds, it takes at least
in that section, and, if M,

once in (a, oo ) the values
M and m and every value between
J/,
m.
bounded and monotonic when x

a.
Let ^(x) be bounded and integrable in the arbitrary interval (, 6), and not
Also let
change sign more than a finite number of times in the interval.
Let
Ja
(f>(x) be
\Jj(x)dx converge.
110
Then
a
g QD
Suppose <(.?) to be monotonic increasing.
We apply the Second Theorem of Mean Value to the arbitrary interval
.
,'>).
Then we have
where a
Add
<^<
6.
B=
to both sides
poo
<
^ (.r) cfa,
( oc )
observing that <(so) exists, since </>(#)

does not exceed some definite number (
is
monotonic increasing in .râ and
34).
/.CO
Also Lt .Z?=0and
b
>ao
Then J5+
<j>(x)^(x)dx converges
56, II.].
Jci
<>xxdx
fb
f*
^(a?)dar4-^(ao)
^(.r)rf.
J^>
J&
F, .................................................... (1)
^(.r)G?.r,
f"
and
F={<^(x)-^(6-0)}[ $(x)dx.
Now we know
Let
J/,
from the above
Lemma
that
m be its upper and lower bounds.
Then
^[
and
wi^
Therefore
{<^(6
-0)
^(x}dx^M,
$(x)dx^M.
*Cf. Pierpont,
loc. cit.,
Vol.
I.,
\^(x)
dx is bounded in (a, x
).

Adding
these,
we
111
see that
î U+ F^{<(oc )- <( + 0)} J/.

M.
Therefore tf + F=/x{<jf>(oo ) <(a + 0)}, where
Insert this value for U+ V in (1), and proceed to the limit when
m^p^
00
^?)f(#>ck~$(a+0)j ^)cfo+/z'W )-<M+
Then
|
where
uf
Lt
)}>
/x.
6-J.oo
This limit must
it
exist, since
the other terms in (1) have limits
Also, since
m^p
follows that
m^-
But
when
J/,
\js(x)dx takes the value // at least once in the interval (a, oo
Thus we may put

Therefore
JMO
///=
we have
\^(x) dx,
^ ^
where a
).
'
oo
finally
f
J
where a
^ ^
^'
oo
we might have used the other forms I. and

Second Theorem of Mean Value and obtained corresponding
It is clear that
In the preceding sections

/.GO
the infinite integrals

the integrand f(x)
is
Ja
we have
f(x)dx,
bounded
results.
INTEGRAND INFINITE.
INFINITE INTEGRALS.
Jba f(x)dx.
50, of the
III.,
J -co
in
dealt with
f(x)dx and
any arbitrary
J-co
f(x) dx,
interval,
when
however
large.
further extension of .the definition of the integral
so as to include the case in
which f(x) has a
is
required
number
finite
of
33) in the interval of integration.

take the case when a is the only point of infinite
infinite discontinuities (cf.
First
we
discontinuity in (a,
and
integrable
in
b).
The integrand f(x)
the
On this understanding,
->0,
we
arbitrary
if the integral
define the infinite integral
is
interval
f
\
supposed bounded
(a
f*
\
Ja+
f(x)
Ja
where
dx has a limit a
f(x)
b),
dx as Lt
t^^Ja + t
f(x) dx.
112
when
Similarly,
continuity in
the
(a, b),
arbitrary interval
point b
and
is the
only point of infinite dis-
bounded and integrable in ///'

where a<^b
<^b,we define the in-
f(x) is
(a, b
g),
f&-f
Jba
Again,
f(x)dx as Lt
when
f(x)dx,
this limit exists.
f-X)Ja
when a and
b are both
points of infinite discontinuity,
{baf(x)
f
infinite integrals
Ja
f(x)
dx and
dx as
the
sum
Jc
when
f(x) dx,
these integrals
as defined above, c being a point between a and

This definition is independent of the position of
exist,
a and
a)
where
a<c'<
Finally,
b.
between
we have
since
b,
of the
f(x) efoj+
f(x) dx,
51, III.).
(cf.
there be
let
dx =
a finite number of points of infinite disxn

Let these points be x lt x 2
(a, b).
continuity in the interval
where a = x 1 <^x2
the
f(x)dx by
<xn =
...
We
b.
. . .
define the infinite integral
equation
\ f(x)
Ja
when
dx =
Ja
f(x)dx + J[ f(x) dx +
xl
on
the integrals
the
right-hand
. .
+ JT
f(x) dx,
xn
exist,
according
to
the
definitions just given.

It should be noticed that with this definition there are only to
be a finite number of points of infinite discontinuity, and f(x) is
to be
bounded
in
any
of these points as an
one
partial interval of (a, b), which has not
internal point or an end -point.
This definition was extended by Du Bois-Reymond, Dini and Harnack to

which the integrand has an infinite number of points of
infinite discontinuity, but the case given in the text is amply sufficient for
certain cases in
our purpose. The modern treatment of the integral, due chiefly to Lebesgue,*
has rendered further generalisation of Riemann's discussion chiefly of
historical interest.
It is convenient to
speak of the
infinite integrals of this
the succeeding section as convergent, as

*
Cf footnote,
.
p. 77.
we
did
when one
and
or other

of the limits of integration was infinite,
and oscillatory are employed as before.
Some
113
and the terms divergent
writers use the term proper integral for the ordinary integral
fb
I
f(.v)dv,
when f(x)
bounded and integrable
is
in the interval (a,
6),
and
Jd
improper integral for the case when it has points of infinite discontinuity in
(, 6), reserving the term infinite integral for
30
x
f
fa
f
/(*)<** or
j(x}dx,
/(.<
Ja
J-co
J-co
French mathematicians refer to both as inte'grales ge'ne'ralise'es
refer to both as uneigentliche Integrate, to distinguish them from
Germans
eigentliche
Integrate or ordinary integrals.
60.
ff(x)dx.
Ja
f(x)dx.
J -co
f*
J -
Let f(x) have infinite discontinuities at a

in
any
interval,
however
f(x)dx.
<
finite
number of points
large.
For example, let there be infinite discontinuities only at

x lt x 2 ... xn in x = a, f(x) being bounded in any interval (c, b),
where
Let
,
Then we have,
as above (59),
f(x)dx= Ja
\ 'f(x)dx+ \*f(x)dx+...+ Jf f(x)dx+ [ f(x)dx,
Jc
Ja
Ja-j
n
\
jr
where x n <c<b, provided that the integrals on the right-hand

side exist.
rt
It will be noticed that the last integral
integral, f(x) being
pac
If the integral
rJ
c,
in
is
an ordinary
(c, b).
f(x)dx also converges, we define
the infinite
the equation:
It is clear that this definition is
since
f(x) dx
f(x)dx by
of
independent of the position
we have
f(x)dx,
f(x)dx=[
J x n f(x)dx+[
J
c'
where # H
114
we may
Also
write the above in the form

l
+ J\x f(x)dx.
f(x)dx=\*
Jx
Ja f(x)dx+[*'f(x)<fo+.:.
n
Jf
E-co
fa
f(x)dx,
as.
before, as the
f(x)
dx
sum
of
GO'
the integrals
J -<x
f(x)dx and
Ja
f(x).
It is easy to show that this definition is

position of the point a.
{ba
independent of the
It is clear that
f (x) dx.
need only discuss the case when there is a point of

continuity at an end of the interval of integration.
If
x=a
the only point of infinite discontinuity,
is
we
infinite dis-
we have
f(x)dx = Lt
J\
when
this limit exists.
from the
It follows at once,
I.
f(x)dx,
The integral
Ja
f(x)
dx
definition, that
is
convergent
and has
the value
e having been chosen, as small as we

a positive number such that
when, any positive number

please, there is
r\
-P
Ja+t
And
further
II.
f(x)dx
Ja
and
f(x)
dx
sufficient condition for the convergence

is that, if
been chosen, as small as
number
r\
if
we
any
positive
number
e lias
a positive
such that
+"
Also,
provided that
necessary
of the integral
f(x)
dx <e, when
this infinite integral
j a,
f(x)
0<f <f =
n-
dx converges, we have
f(x)
a'
dx=
\*f(x) dx-}- \f(x) dx,

Jx
Ja
a<x<b.

f(x)
{6a
number
positive
dx converges,
e,
115
to the arbitrary
number
such that
rx
Ja
f(x)dx <e, when 0<(o?
he absolutely convergent, if f(x) is

the arbitrary interval (a
b),
to
in
\
f(x)
dx
a)
17.
fbn f(x)
bounded and
dx
is
integrable
<<&
where
said
and
converges.
(i>
a
It follows
from
II.
ordinary convergence.
that absolute convergence carries with it

But the converse is not true.
infinite
An
integral of this kind may converge, but not converge absolutely,*

as the following example shows.
>
An
It
example of such an integral
is
clear that
suggested at once by
is
sinl
[converges,
fo<to
Jo
but not absolutely, for this integral
y substituting
l/.r
for
$ 54.
is
reduced to
.r.
--x-converges,
fi
Jba
-.
if
0<?i<l.
a)
(K
For we have
{7
^-\x
a
==
H
T
-^- =
Therefore Lt
\~
a)
Also the integral diverges
From
|>f
we
this
obtain
n {(^~~~^)^
~ a^ "
(6
,
1l
}'
when 0<?i<l.
when n = 1.
results
which correspond
to
those
55.
III.
Let f(x)
interval (a + g,
and
petween
fba f(x)
*
be
b),
bounded and integrable in the arbitrary

where 0<f <b a. If there is a number
/JL
such that (x
dx converges
Cf.
43,
V.
a)' -f(x) is
absolutely,
47, Cor.
I.
and
54.
bounded when a<^x = b
116
Again,
IV. Let f(x) be bounded and integrable in the

arbitrary
interval (&+(, b), where
// there is a number
IUL
greater than or equal to 1 such that (x aYf(x] has a
positive lower bound when a<^x
b, or a negative
upper
<<&-,.
=
diverges to
Jba f(x)dx
-co in
And
the first case,
and
to
the second case.
finally,
V. Letf(x) be bounded
(a
+00 in
where
b),
If there
is
and integrable in
the arbitrary interval
0<<6-a.
a number
between
jm
and
a )**/(#)
such that Lt (x
.r->a+0
Jba f(x)dx
If
there is
Lt (x
converges absolutely,
a number
ay/(x)
exists
//
greater than or equal to 1 such that
and
not zero, then
is
the
same
true if (x
is
Ja
.r->a+0
and
f*
aYf(x) tends
f(x)dx diverges;
to -foo
or to
a>
as
x->a + 0.
We
speak of this
shall
Jba f(x)dx,
when x = a
test as the /x-test for the infinite integral
a point of infinite discontinuity.
is
It is
clear that in applying this test we are simply asking ourselves

the order of the infinity that occurs in the integrand.
The
results can be readily adapted to the case
when
the upper
limit b is a point of infinite discontinuity.

Also, it is easy to show that
VI. If
<p(x)
.is
in
(a, b),
and
fb
Jba \ls(x)dx
converges absolutely then

,
convergent.
The
we
(p(x)\fr(x)dx is absolutely
J<i
(Cf.
56, I.)
most of the cases which

But it would not be difficult to develop in detail
which correspond to the other tests obtained for the
tests
given in III.- VI. will cover
shall meet.
the results
ICO
a
No
when
f(x)dx.
special discussion is required for the integral
a certain
number
p
I
Ja
f(x)dx,
of points of infinite discontinuity occur

in (a,
or for
b),
Ja
in
f(x)dx,
These integrals
60.
f(x)dx, and
oo
we have
the types for which
f(x)dx, as deh'ned
J -
oo
sum
reduce to the
all
117
of integrals of
already obtained the required
criteria.
We add some examples illustrating the points to which we

have referred.
Ex.
(i)
1.
Prove that
converges and that
,
j
diverges.
Let
Lt
Then
The
//-test
thus establishes the convergence of
/o
(ii)
Let
/(#):
Then
Lt xf(x)=\.
->o
Therefore the integral divei'ges by the same
Ex.
2.
Prove that
J'
Jo
The
integral
is
^dx converges,
an ordinary
Therefore
3.
the integral
Prove that
The integrand has
We have
finite integral if
converges
0<n <1.
0.
is
It diverges
when
converges.
infinities at
when 0<n<l.
,v
= and
.f=l.
thus to examine the convergence of the two infinite integrals
''O
where a
The
when
Also
Ex.
test.
some number between
/x-test is sufficient in
and
each case.
where we have written f(x) =
1.
118
4Ex.
Show
4.
that
log sin
Jo
The only
from the
infinity is at
xdx
converges and
is
#=0, and the convergence
equal to
^7rlog2.
of the integral follows
/x-test.
Further,
log sin
Jo
t*
x dx = 2
log sin
dx
2.v
Jo
An
4n
= 7rlog2 + 2/
\og8inxdjj + 2l
\ogcosxdx
4-
= TT log 2 + 41
log sin
Jo
,-"
But
log sin
dx = 2
A*
x dx.
/!
log sin
x dx.
,4-
log sin & do: =
Therefore
J
From
this result it is easy to
TT
are equal to
Ex.
- cos #) do;
and
log
+ cos &) da-
2.
e~
Gamma
Function
dx.
JQ
Then the integrand
is
bounded
in
0<.vâ, where a
need only consider the convergence of
The
x
(1
Letw^l.
(i)
of e
log
Jo
Discuss the convergence or divergence of the

x n ~l
5.
integral
TT
integrals
-TT
log ( 1
2.
show that the convergent
~n
Jo
log
/u,-test
is
of
e~
n~ l
is
arbitrary,
and we
dx.
55 establishes that this integral converges, since the order
greater than any given power of

as follows
Or we might proceed
x.
Since
=
2!
when x > 0,
e*
>
xr
(r
= any
positive integer),
ton A
,H
-,-
r-n+1
t'
But whatever
w,
may
be,
we can
choose
so that
r-n + l>l.
It follows that, whatever n may be,

~
/
e~xx n l dx converges.
Let <n<\.
~l
x
has an infinity at #=0.
In this case e~ jf
(ii)
The
/
e~
/x-test
xn
~l
shows that
dx converges.
e~ xx
n~ l
dx converges, and we have
just
shown that

Therefore
er*af*~~
119
dx converges.
Jo
(iii)
LetttÔ.
In this case e~x.vn -1 has an infinity at # = 0, and the

oc
e~xx n ~ l dx
to
diverges
Ex.
Discuss the integral
6.
Since Lt (xr log.v) = 0,
shows that
jtx-test
xn ~ l log x dx*
JQ
when r>0,
the integral
an ordinary
is
integral,
x->0
when n > 1
Also we know
.
that
It follows that
\ogxdx = xLt>0
JO
Again,
Lt
(xv-
x x*
0<w<
And when
1,
log x}
we can
{x(\ -logjp)-
Lt (#M+
1}=
-L
=
log x) 0,
choose a positive
if /x
number
/x
> 1 - n.
less
than
which
satisfies this condition.
Therefore
Finally,
<n_
xn ~ l log x dx converges, when
we have
Lt (x x x~ l log x
\
Therefore
Jn
= Lt
.t-"
|
log x
= oo when
,
0.
-~
&M l log ju dx diverges, when ?i^0.
REFERENCES.
DE LA VALLÊ
POUSSIN,
toe. cit.,
T.
e
(3 ed.), Ch. VI.
I.
DINI, Lezioni di Anolisi Infinitesimale, T.

Pisa, 1909.
GOURSAT,
loc. cit.,
T.
I.
(3 eU), Ch.
KOWALEWSKI, Grundziige der
IV. and
II., 1*
e
T. II. (2 ed.), Ch. II.
Parte, Cap.
I.
and VII.,
XIV.
Differential- u. Integralrechming,
Kap. XIV.,
Leipzig, 1909.
LEBESGUE, Lecons sur V Integration, Ch. I. and

OSGOOD, loc. cit., Bd. L, Tl. I., Kap. III.
PIERPONT,
loc. cit.,
Vol.
I.,
II.,
Paris, 1904.
Ch. V.
STOLZ, Grundziige der Differential-
u.
Integralrechnung, Bd.
I.,
Absch. X.,
Leipzig, 1893.
And
BRUNEL,
"
Bestimmte Integrate," Enc.
d.
math. Wiss., Bd.
II., Tl.
L, Leipzig,
1899.
* This
integral can be reduced to the
vergence or divergence follows from Ex.
Gamma
5.
Function integral, and its conAlso see below, Ex. 5, p. 120.
120
EXAMPLES ON CHAPTER
1.
Show
that the following integrals converge
am*
r
1
Jo
+cosx+e*
r <*-iy* <**,
Ji l+x+xs + smx
l+x
Jo
2.
IV.
\M<fJ.
I*.*-,
l-x*
Jo
Discuss the convergence or divergence of the following integrals

/
Ja
(x-a\!(b-x)
x+l
Jo
dx,
Jo
^ *^~1 dx,
x-1 dx,
Show
3.
f
Jo
where
0<c<l,
Jo
sin"
cos"
dO.
Jo
that the following integrals are absolutely convergent
sin
--j-,
x
e- a2x2 coshvdj,
e~ *
l
Jo
Jo
>*jW
f $h
3
a polynomial of the wi th degree, and Q(x) a polynomial of the

degree, w^wi + 2, and a is a number greater than the largest root of
where
th
?i
-/
(.v) is
4.
Let /(a;) be defined in the interval 0<^t'^
and
so on, the values being alternately positive
Show
that the infinite integral
as follows
and negative.
f(x)dx converges, but not absolutely.
Jo
Using the substitution x = e~ show that

m.v
ll
5.
>
>
and n
1.
converges, provided that in
And by means of a similar substitution, show that
m <0 and n > -
converges, provided that

6.
Show
X
that
/
*
when /xÔ,
>
*
q
V
1+
1.
converges when /x>0 and that
it
diverges
the lower limit a of the integral being some
number greater than
unity.
Deduce that if there is a number

then
*a
f(x)
'
//,
> 0, such that
Lt {^(log x)l+ *f(x) }
dx converges, and give a corresponding
exists,
test for the divergence of

this integral, f(x) being

where b a.
>
(a, 6),
Show
- -
that
and
1.
integrating
[*
I
dx converges,
cos
x log x dx by
cos ^ log x ax = si n
diverges.
parts,
x log a?
*/l
Deduce that
cos x log x dx converges,
Jo
On
we obtain
Cx sin
I
xj
*R
jl
i#7
cos # log x dx oscillates infinitely.
Also show that
8.
any arbitrary interval
in
On
vs
2
121
fx"
integrating
and
is
equal to
*-- dx.
Jo
coBJ^dx by parts, we obtain

1
1
-*
where x" >
x'
> 0.
cos x*dx.
9.
Let f(x) and
certain
number
for the
two
g(.v)
be bounded and integrable in
Cb
Prove
(a, 6),
except at a
of points of infinite discontinuity, these points being different

functions.
that
rb
f(x)g(x)dx converges,
if
Cb
\f(x}\dx
and
\g(x)\dx
converge.
10.
Let f(x) be monotonic when
Then the
is
series
/(a) +/(
xâ, and Lt
+ 1 ) +/(a + 2) +
convergent or divergent according as
Prove that
for all
Ja
f(x) = 0.
. . .
f(x}dx converges or diverges.
values of the positive integer n,
A, so show that
converges to a value between i(?r +
1)
and
CHAPTER V
THE THEORY OF INFINITE SERIES, WHOSE TERMS ARE
We
62.
shall
now
consider some of the properties of series
whose terms are functions of
We
and the terms of the

of x in some interval,
When we
series are
e.g.
we
that
of x
we wish
the
sum
of
insert this value of x in the different terms
of the series
(iii)
what value
we then
that
infinite series
(x) + Uz(x)+...
settle for
the series
(ii)
(a, b).*
(x) + u
to be understood
(i)
supposed to be given for values
speak of the sum of the
u
it is
x.
denote such a series by
that
we then
find the
that
we then
find the limit of this
sum
Sn (x)
of the first
terms:
and
(iv)
On
is
as
^->x keeping
,
the time at the value settled upon.
this understanding, the series
said
sum,
all
sum
if,
to be
convergent for the value x, and to have f(x) for its

x having been first inserted in the different
this value of
in
24, when we say that x lies in the interval (a, b) we mean
In some of the results of this chapter the ends of the interval
that
When this is so, the fact that we are dealing
are excluded from the range of x.
with the open interval (a
x<.b) will be stated.
As mentioned
a^x^b.
<
tCf. Baker, Nature, 59,
p. 319, 1899.
122
123
terms of the series, and any positive number e having been

chosen, as small as we please, there is a positive integer v such that
Sn (x)\ <e, when n = v.
\f(x)
Further,
A
if
necessary and sufficient condition for convergence is that,

any positive number e has been chosen, as small as ive please,
there shall be
a positive integer
such that
$n (#)|< e, when n = v,
\Sn+p (x)
when we
similar convention exists
are dealing with other limiting
In the definition of the differential coefficient of f(x) it

stood that we first agree for what value of x we wish to know f'(x)
processes.
+ /i)
then calculate/^) and /(.-
for this value of
is
;
under-
that
we
then obtain the value of
the limit of this fraction as

ti
Again, in the case of the definite integral

that
we
/(#, a)o?^,
it is
understood
insert in f(x, a) the particular value of a for which we wish the

we proceed to the summation and limit involved in the
integral before
integration.
We
shall write, as before,
where f(x) is the sum of the series, and we shall

remainder after n terms.
As we have seen in 19, R n (x) is the sum of the
Also
we
call
R H (x)
the
series
shall write
Rn (x) = Sn+p (x) - Sn (x),
and call this a partial remainder.

With this notation, the two conditions
when n^v,
(x)\<e> wnen n = v,
\Rn (x)\<
(i)
(ii)
\p
for convergence are
>
A series
and not
may converge
for every value of
for the end-points a or
E.g. the series
1 -f
converges and has -
When
.v=l,
it
for its
in the
open interval a <.<;<
b.
x + x2 +
-
. .
sum, when - 1<# <
diverges to +00
when x= -
1, it
1.
oscillates finitely.
fr
THEORY OF INFINITE
124
63.
SERIES,
WHOSE TERMS
The Sum of a Series whose Terms are Continuous Funcx may be discontinuous. Until Abel * pointed out that
tions of
the periodic function of x given by the series
which represents x in the interval

7r<X7r,
at the points x = (2r+l)w, r being any integer,
is
discontinuous
was supposed
a
of functions,
series
by
convergent
continuous in a given interval, must itself be continuous in
that interval. Indeed Cauchy f distinctly stated that this was
it
that a function defined
the case, and later writers on Fourier's Series have sometimes

tried to escape the difficulty by asserting that the sums of these
trigonometrical series, at the critical values of x, passed continuously from the values just before those at the points of
discontinuity to those just after.
:{:
This mistaken view of the

different errors.
The
increases, the curves
first
sum
of such series
was due
y = Sn (x)
must approach more and more
nearly to the curve y =f(x), when the sum of the series

an ordinary function capable of graphical representation.
curves y
series, but
two
to
consisted in the assumption that, as

is
f(x),
These
Sn (x) we shall call the approximation curves for the

we shall see that cases may arise where the approxi-
mation curves, even for large values of n,

from the curve y=f(x).
It is true that, in
very considerably
a certain sense, the curves
y = Sn (x)
(i)
differ
and
(ii)
y=f(x)
approach towards coincidence but the sense is that, if we choose

any particular value of x in the interval, and the arbitrary small
positive number e, there will be a positive integer v such that, for
;
this value of x, the absolute value of the difference of the ordi-
nates of the curves
(i)
and
(ii)
will be less
than
when n = v.
not the same thing as saying that the curves

coincide geometrically.
They do not, in fact, lie near to each
Still this is
other in the neighbourhood of a point of discontinuity of /(#)>

and they may not do so, even where f(x) is continuous.
*Abel, J. Math., Berlin,
1, p. 316, 1826.
t Cauchy, Cours d' Analyse,

(Ser. 2), T. III., p. 120.
I
Cf. Sachse, loc.
cit.
re
Partie, p. 131, 1821.
Donkin, Acoustics,
p. 53, 1870.
Also (Em-res de Cauchy,
ARE FUNCTIONS OF A SINGLE VARIABLE

The following examples and diagrams
Ex.
1.
illustrate these points
125
Consider the series
Here
(n-l).r+l
n.v+1'
and
Thus,
U S (.r)=l;
when x > 0,
>-oo
when
,v
=Q
Lt
n (.r)
= 0,
since
Sn (0)=0.
The curve y =/(#), wlien x 0, consists of the part of the line y = 1 for which
>0, and the origin. The sum of the series is discontinuous at .r=0.
Now
examine the approximation curves
This equation
may be
written
As n
more
increases, this rectangular hyperbola (cf. Fig. 10) approaches more and
If we reasoned from the shape of the
closely to the lines ?/ = ], .r = 0.
-2
FIG. 10.
approximate curves, we should expect to find that part of the axis of y for
which 0<^<1 appearing as a portion of the curve y=-f(x) when x
0.
As Sn (x) is certainly continuous, when the terms of the series are continuous, the approximation curves will always differ very materially
the curve y=f(x\ when the sum of the series is discontinuous.
from
THEORY OF INFINITE
126
Ex.
2.
WHOSE TERMS
SERIES,
Consider the series
*(*) = r-r^5 - vT7T-^3-
where
In this case
and
Lt
jj
Thus the sum
all
values of
.r.
of this series
is
continuous for
all
values of
see that the approximation curves differ very materially

y =/(.?) in the neighbourhood of the origin.
.r, but we shall

from the curve
1-0
The curve
has a
maximum
at (l/n ^)
t
and a minimum at
I/n,
- 1)
(cf.
Fig.
1).
The
points on the axis of x just below the maximum and minimum move in
towards the origin as n increases. And if we reasoned from the shape of the
y = <Sw (.r), we should expect
curves
to find the part of the axis of
y from
to | appearing as a portion of the curve #=/(#).
Ex.
3.
Consider the series
w (') =
where
iV
Here
and
The sum
(cf.
Lt
of the series
Fig. 12),
is
for all values of x.
again continuous, but
which have a maximum at
(l/\/
>
-the
approximation curves
M ) an ^ a niinimum at

(-l/*Jri\
- J>/%),
bourhood
of the origin.
axis of
differ
y should appear
127
very greatly from the curve y=f(x) in the neighIndeed they would suggest that the whole of the
as part of y=f(x).
1-0
These remarks dispose of the assump64. Repeated Limits.

tion referred to at the beginning of the previous section that the
approximation curves y = Sn (x), when n is large, must approach
closely to the curve y =f(x), where f(x) is the sum of the series.
The second
error alluded to above arose from neglect of the
convention implied in the definition of the sum of an infinite

series whose terms are functions of x.
The proper method of
finding the sum has been set out in 62, but the mathematicians
to whom reference is now made proceeded in quite a different
manner. In finding the sum for a value of x, say x at which
,
a discontinuity occurs, they replaced a? by a function of n, which

converges to x as n increases. Then they took the limit when
9?ôo of
Sn (x) in
its
new
In this method x and n approach

Lt
and the value of
Sn (x) may
form.
their limits concurrently,
{.e^Kr ,tt->co}
Indeed, by
quite well differ from the actual sum for x = x
choosing the function of n suitably, this double limit may be
made to take any value between /(# -f 0) and/(# 0).
.
THEORY OF INFINITE
128
For instance,
in the series of
WHOSE TERMS
SERIES,
63, Ex.
-T,
lJ
1,
+ ..., aSO,
,&
we have seen that x = is a point of discontinuity.

If we put x=p/n in the
expression for Sn (x), and' then let
??->ac p remaining fixed, we can make
Lt
Sn (x) ifoke any
,
{*-*),->>}
value between
have $nOV^) =
and
according to our choice of p. A For
we
rv\
jT|
which
is
to 1 as
Sx
Lt
which passes from
independent of n, and
increases from
to oo
be seen that the matter at issue was partly a question

of words and a definition.
The confusion can also be traced, in
some cases, to ignorance of the care which must be exercised in
It will
any operation involving repeated

here with two limiting processes.
If the series is
convergent and
limits, for
sum
its
we
are really dealing
is f(x),
then
f(x)= Lt Sn (x),
n
and the
a limit,
limit oif(x) as
is
x tends to X Q assuming that there

,
Is
such
given by
Lt f(x)= Lt
If
>-oo
we may
[Lt
8n (x)] ...................... (1)
use the curve as an illustration, this
is
the ordihate
towards which we move as we proceed along the

curve y = f(x), the abscissa getting nearer and nearer to X Q but
not quite reaching x
According as x approaches x from the
of the point
the limit given in (1) will be/(o; + 0) or/(#

0).
Now/(cc ), the sum of the series for x = x is, by definition,
right or
left,
Lt [8*(xJ],
>-ao
and since we are now dealing with a
definite
number
of con-
tinuous functions,
a continuous function of x in the
interval
concerned.
Thus
S n (x) is
with which we are
Sn (xQ )= Lt Sn (x).

There f orflJJhe sum of the series for x = x
Lt
4W\
The tw>yexpressions
in
[Lt
>co
(
may
129
be written
Sn (x)]
(2)
x >x9
and
(2)
need not be the same.
are so onljrisvvhen f(x) is continuous at x
They
Uniform Convergence.* When the question of changing

oltwo limiting processes arises, the principle of uniform
65.
;the order
convergent, ^hich we shall now explain for the case of infinite

whrwe terms are functions of x, is fundamental. What is
this principle will be seen most clearly by returning
series
involved^
to the
sej^
""
Sn (x) = 1 -
Sn (x) = l, when #
Lt
<
when x>0, and
/2 n (0)
= 0.
the^rbitrary positive number e is chosen, less than unity,

x is taken, it is clear that l/(iia; + l)<e for a
son^positive
if
positive/>c only
j
If
:and
If
A-
= 0-l,
0-01, 0-001,
...
10-", respectively, l/(n.r +
>10
And when e=
4
,
103 10,
,
...,
10<+ 3
l)<
only
when
>.
and .v=10~ p n must be greater than 10 p+(
As we approach the origin we have tp take more and more terms of

make the sum of n terms differ from the sum of the series by
series to
than a given number.

contribute 1% of the sum.
.r=10~ K
n>
The inequality
shows that when
When
e is
any given
the
first
million
if
the
less
terms do not
positive
number
less
than unity,
The simplest treatment of uniform convergence will be found in a paper by

Osgood, Bull. Amer. Math. Soc., 3, 1896.
c. i
THEORY OF INFINITE
130
WHOSE TERMS
SERIES,
and x approaches nearer and nearer to zero, the smallest positive

integer which will make R n (x), R n+l (x), ... all less than e increases
without
limit.
There is no positive integer v which will make R v (x), R v+l (x),

all less than this e in x = 0, the same v
serving for all values of x
in this range.
.
On
the other hand there
is a positive integer v which will

if
this
the
condition,
a, where
satisfy
range of x is given by x
a is some definite positive number.
Such a value of
would be the integer next above
Our
series is said to converge uniformly in

not converge uniformly in x
0.
xâ,
lj/a.
but
it
does
We
turn
now
to the series
and define uniform convergence
Let the series
* in
an interval as follows
u2 (x) + us (x) + ...

l (x) +
= =
x b and its
converge for all values of x in the interval a
be f(x).
It is said to converge uniformly in that interval, if,
any positive number e having been chosen, as small as we please,
sum
a positive integer
there is
such
that, for all values of
It is true that, if the series converges,

(a, b)
in the
n^vA
\f(x)-S n (x)\<e, when
interval,
Rn (x) |<e for each
x in
when n = v.
The additional point in the definition of uniform convergence

is that, any positive number e having been chosen, as small as
we please, the same value of v is to serve for all the values
of x in the interval.
For
this integer v
we must have
\R v+ i(x)\, ...
no matter where x lies in (a, 6).
\R,(x)\,
all less
than
e,
of uniform convergence was discovered independently by Stokes

Phil. Soc., 8, p. 533, 1847) and Seidel (cf. Miinchen, Abh.
Trans.
(cf. Cambridge,
Ak. Wiss., 5, p. 381, 1848). See also Hardy, Cambridge, Proc. Phtt. Soc. 19,
*The property
p. 148, 1920.
< <
x b, or the
can also have uniform convergence in the open interval a
a<x^b, a^x<b; but, when the terms are continuous
in the closed interval, uniform convergence in the open interval carries with it
uniform convergence in the dosed interval (cf. 68).
f
We
half -open intervals

The series does not converge uniformly
enow that for some positive number (say
)ositive integer v which will make
R,(x)\,
ill
less
than
for every x in
\Rrn(x)\,
in
e
(a,
)
131
b)
there
if
is
we
no
...
(a, b).
It will be seen that the series
on verges uniformly in any interval a = x = b, wr here a, b are any

given positive numbers.
It may be said to converge infinitely slowly as x tends to zero,
we get nearer and still nearer to the origin,

we cannot fix a limit to the number of terms which we must
take to make R n (x)\<e. It is this property pf infinitely slow
in the sense that, as
(e.g. X Q ) which prevents a series converging

an interval (X Q 8, x +S) including that point.
Further, the above series converges uniformly in the infinite
înterval x=a, where a is any given positive number.
It is sometimes necessary to distinguish between uniform convergence in an infinite interval and uniform convergence in a
fixed interval, which may be as large as we please.
The exponential series is convergent for all values of x, but it
does not converge uniformly in the infinite interval x = 0.
For in this series R n (x) is greater than x n /n\, when x is positive.
Thus, if the series were uniformly convergent in x^-Q, x n /nl
would need to be less than e when n
v, the same v serving for
all values of x in the interval.
But it is clear that we need only take x greater than (v\e) v to
make R n (x) greater than e for n equal to v.
convergence at a point
uniformly in
However, the exponential series is uniformly convergent in the

(0, b), where b is fixed, but may be fixed as large as we
interval
please.
For take
for x
c greater
than
= c.
We know that
the series converges
^(cXe, when n^v.

R n (x)<R n (c), when Q=x^b<c.
Therefore
But
Therefore
b.
R n (x)<ê,
values of x in
(0, b).
when n^v, the same
serving for
all
132
THEORY OF INFINITE
From
interval
b, b),
we
be fixed as large as
Prove that the
1.
WHOSE TERMS
the uniform convergence of the exponential series in the

(0, b), it follows that the series also converges uniformly
in the interval
Ex.
SERIES,
where
in both cases b is fixed, but
may
please.
series
converges uniformly to 1/(1 -x) in
Ex.
Prove that the
2.
series
(l-.r) + .r(l
converges uniformly to
Ex.
Prove that the
3.
Ex.
1 in
(1
Prove that the
4.
=x
series
- x)
in
^x ^
1.
series
converges uniformly in the infinite interval
Ex.
Prove that the
5.
series
converges uniformly in the interval (0, ft), where b is fixed, but may be fixed
as large as we please, and that it does not converge uniformly in the infinite
interval
xÔ.
necessary and sufficient condition for Uniform Convergence. When the sum f(x) is known, the above definition
66.
often gives a convenient means of deciding whether the convergence is uniform or not.
When
the
sum
is
not known, the following
to the general principle of convergence (15),
Let
Ui(x}
+ u, (x) + Us(x}+
2
test,
is
corresponding
more
suitable.
...
an
infinite series, whose terms are given in the interval (a, b).
A necessary and sufficient condition for the uniform convergence of the series in this interval is that, if any positive
number e has been chosen, as small as we please, there shall be a
positive integer v such that, for all values of x in the interval,
be
R n (x)\<ê, when n^v,for

(i)
condition
Tlie
is necessary.
Let the positive number

m take e/2.
Since the series
iteger
v,
133
be chosen, as small as
uniformly convergent, there
is
is
we
please,
a positive
such that
\f(x)-Sn (x)\<~,wheun^v,
same v serving for
the series.
Let
vi",
all
values of x in
(a, 6),
f(x) being the
be any two positive integers such that
- 8*(x) S
Then
8,,.(x) -f(x)
n">n'=v.
+ }f(x) - S,,(x)
sum
for every positive

v,
\Sn+p (x) Sn (x)\<^, when n
the
same
for
all
v
values
of
x in (a, b).
iteger p,
serving
Thus
The condition
(ii)
We know
is sufficient.
that the series converges,
when
this condition is
ttisfied.
Let
sum
its
be f(x).
the arbitrary positive number

Again
iere is a positive integer v such that
let
SH+}l (x)
le
same
Sn (x)
v
< 9, when n =
serving for
all
v,
be chosen.
Then
values of x in (a,
b).
Thus
Also
S,(aO-|=/() = &(a) + 2-
Therefore
But
\Sn (x)
It follows that,
;ified
id this
Sv+p (x)=f(x).
Û
when n
is
greater than or equal to the value
above,
holds for
Thus the
all
values of x in
series converges
(a, b).
uniformly in this interval.
THEORY OF INFINITE
134
WHOSE TERMS
SERIES,
Weierstrass's M-Test for Uniform Convergence.

The
following simple test for uniform convergence is due to
67.
Weierstrass
The
series
+u
(x)
will converge uniformly in

of positive constants
such
(x)
+ us (x) +
(a, b),
no matter what value x
that,
un (x)
= Mn
. .
a convergent
if there is
have in
may
(a, 6),
convergent, with the usual notation,
when n = v,
U/U^)l =
But
Thus
p R n (x)\^Mn
<
e,
when n =
v,
the inequality holding for all values of x in (a, 6).

Thus the given series is uniformly convergent in
For example, we know that the series
is
series
for every positive integer n.
M + M<> + M% +
Since the series

is
convergent,
when
is
any given
(a, b).
number
positive
less
than
unity f
.
It follows that the series
is
uniformly convergent in the interval

Ex.
Show
1.
uniformly convergent for

is
a, a).
<
that the series
+ .y2 cos 20 + ^
xcos
is
any
cos
interval (#
30+ ...
A\\ where
-l<^
()
<<v l <l and
any given number.
Ex.
2.
Show
that the series
x cos
X
,vcos$+ - cos 20 +
and
are uniformly convergent for
number
+ x2 cos 20 + x3 cos 30 + ...
less
than unity.
all
^-
values of
cos 30+...
0,
when x
|
is
\
any given positive

This theorem leads at once to the following result
series is absolutely
when
the
135
and uniformly convergent in an interval,
the absolute values of its terms
in the interval do not exceed
corresponding terms of a convergent series whose terms are
all positive
and independent
of x.
Uniform Convergence does not imply
It should be noted that
Absolute Convergence.
E.g.
the series
not absolutely convergent

the infinite interval x
0*
is
but
it is
uniformly convergent in
Also a series
may
be absolutely convergent without being
uniformly convergent.
(i- x )+ x (i- x ) + x z(i- x )+.,.
E.g.
is
absolutely convergent in the interval
but
it is
68.
=x=
l,
where
not uniformly convergent in this interval.
Uniform Convergence of Series whose Terms are Con-
tinuous Functions of
In the previous sections dealing with
x.
uniform convergence the terms of the series have not been

We shall now
assumed continuous in the given interval.
prove some properties of these series
added.
when
this condition
I.
Uniform convergence implies continuity in
If
the terms of the series
are continuous in
(a, b),
and
the
is
sum.
the series converges
f(x) in this interval, then f(x)

in (a, b).
uniformly to
a continuous function of x
is
Since the series converges uniformly, we know that, however

small the positive number e may be, there is a positive integer v,
such that
\f(x)-Sn (x)\<'3
the same
serving for
ft/i (<')
<
all
when
values of x in
for all values of
.r.
n^ v
(a, b).
THEORY OF INFINITE
136
SERIES,
WHOSE TERMS
Choosing such a value of n, we have
R n (x)
where
Since
<
Sn (x)
continuous in
is
for all values of
e/:*,
the
sum
of
in (a,
b).
continuous functions,
it
is
also
tj
such
(a, b).
Thus we know, from
31 that there
is
a positive number
that
when
xx
x, x'
are
any two values
of
in the interval (a, b) for
which
tj.
f(x')=Sn (x) + R n (x'),
But
where
\R n (x'\<
/3.
f(x) -f(x) = (Sn (x) Sn (x) + R n (x) - R n (x).
Also
Thus
\f(x')-f(x)\^\Sn (x')-8n (x)\+
<e, when
Therefore f(x)
is
\x
continuous HI
Rn (x')\+ R n (x)
x ^rj.
(a, b).
If a series, whose terms are continuous functions, has a

discontinuous sum, it cannot be uniformly convergent in an
interval which contains a point of discontinuity.
For if the series were uniformly convergent, we have just seen
II.
that
its
sum must
be continuous in the interval of uniform
convergence.
III.
Uniform convergence
the continuity of the
sum
is
of a
thus a sufficient condition for

series of continuous functions.
a necessary condition; since different non-uniformly

convergent series are known, which represent continuous
It is not
functions in the interval of non-uniform convergence.

For example, the series discussed in Ex. 2 and Ex. 3 of
are uniformly convergent in x = a
0, for in both cases
63
>
*0)l<^=^
Thus
of x.
(x)
<
e,
when
xâ>0.
when 7i>l/ae, which
is
-independent
137
But the interval of uniform convergence does not extend up to

and include # = 0, even though the sum is continuous for all
values of
This
x.
clear
is
in
where
Rn (x) = I
Ex.
2,
<
when n = v,
/y*
m
for if it is
g-^,
-J- n"X
the same
serving for all

values of x in x^Q, the statement is shown to be untrue by
= l/2, and thus R n (x)\ê,
=
v (x)
pointing out that for x l/v,
asserted that
\R n (x)
e,
when n^v,
right through the interval,
Similarly in Ex.
3,
where
fv\
R n (x) =
if
Q
e<l/2.
/Y>
-^
if it is
asserted that
< e, when w = the same v serving for all3 values of x in3

R v (x) = JV*'
we
need only point out that for #=l/v
x=Q>
Thus \R n (x)\<e, when n = v, right through the interval, if
Rn(%)
i/,
/i/
There
both
in
is,
cases,
a positive
jR,i(l/?n)<e<l/2, when n = v, but
for
integer
which
than
this integer is greater
1/m.
Thus
clear that the convergence
it is
becomes
infinitely
slow
as x->0.
IV. If the terms of the series are continuous in the closed

interval (a, 6), and the series converges uniformly in a<#<6,
then it must converge for x = a and x = b, and the uniformity
of the convergence will hold for the closed interval (a,b).

Since the series is uniformly convergent in the open interval
,
we
have, with the usual notation,
\Sm (x)-Sn (x)
<,
when m>nî/,
............ (1)
the same
v serving for every x in this open interval.

Let m, n be any two positive integers satisfying this relation.
Since the terms of the series are continuous in the closed
interval (a, 6), there are positive numbers ^ and i/ 2 say, such that
,
i#m()-Sm(a)|<~, when 0^(aj-a)îy

\Sn (x)-Sn (a)\<l, when
Choose a positive number

let
^ (x
a) =
rj.
>/
O^-a)^.
not greater than
or
tj 2
and
THEORY OF INFINITE
138
Then
WHOSE TERMS
SERIES,
Sm (a)-Sn (a)\
^\Sm (a)-Sm (x)\ + Sm (x)-Sn (x)\+\Sn (x)-Sn (a)
<3+3+3
<e,
similar
when
m>n^v
(3)
see that the series converges for x = a

and, combining (1), (2) and (3), we see that the con-
(2)
and x = b,
m>n^v
when
\Sm (b)-Sn (b)\<,
From
(2)
argument shows that

and
(3)
we
dition for uniform

convergence in the closed interval (a, b)
is
satisfied.
If the
terms of the series
M 1 (.r) +
are continuous ip (a,
(a, /3), where a <; a
6),
2 (.r)
+ w3 (j;) + ...
and the
< ft < 6,
series converges uniformly in

the series need not converge for x
E.g. the series
converges uniformly in (-a,
a),
+ 2.r + 3^2 +
every interval
=a
x=b.
or
. . .
where a<l, but

1 or x=\.
it
does not converge for
x=
However we shall see that in the case of the Power Series, if it converges
x = a or x = b, the uniform convergence in (a-, (3) extends up to a or 6, as
for
the case
But
may be. (Cf. 72.)

this property is not true in general.
The
series of continuous functions
converge uniformly for every interval (a, (3) within (, 6), and converge
of uniform convergence does not extend up
to and include the point a or b.
may
for
xa or #=6, while the range
x + ^x
E.g. the series
formed from the logarithmic
3
by taking two consecutive

convergent when
^v
-\-i-x
+ ^x
1&
(1)
series
positive terms
and then one negative term,
is
-Kx^l, and its sum,
when #=1, is f log 2.*

convergent when x < 1, and
Further, the series (2) is absolutely

the sum is not altered by taking the terms in any other order.
It follows that when x\<\ the sum of (1) is log (1 +x\ and
\
sum is | log 2.
Hence (1) is discontinuous
at
x = \ and therefore the
22.)
when #=1
its
interval of uniform
convergence does not extend up to and include that point.

*Cf. Hobson, Plane Trigonometry (3rd Ed.),
therefore
(Cf.
p. 251.

69.
Uniform Convergence and the Approximation Curves.
Then we have,
be uniformly convergent in
when
\Sm (x)-Sn (x)\<e,
In particular,
all
values of
Sm (x] - Sv (x)
m>n^v,
in the interval.
<
when
e,
m > v,
and we
shall suppose v the smallest positive integer

condition for the given e and every x in the interval.
Plot the curve
Let a
(a, 6).
as before,
the same v serving for
a strip
139
which
y = Sv (x) and the two
a of breadth
2e,
whose central
will satisfy this
S v (%)
parallel curves y
line is y
v (x).
(Fig. 13.)
forming
=S
FIG. 13.
All the approximation curves y = Sm (x\ m>v, lie in this strip, and the
curve #=/(#), where f(x) is the sum of the series, also lies within the strip*
or at most reaches its boundaries. (Cf.
66 (ii).)
Next choose
e less than e, and let the corresponding smallest

positive integer
condition for uniform convergence be v'. Then v' is greater
the
satisfying
than or equal to v. The new curve y = Sv '(x) thus lies in the first strip, and
new
breadth 2e', formed as before, if it goes outside the first

have this portion blotted out, for we are concerned
only with the region in which the approximation curves may lie as m
the
strip in
increases
or'
strip
any
of
part, can
from the value
In this way,
if
we
v.
take the set of positive numbers
where Lt
>e'>e"...,
and the corresponding positive integers

v
we
obtain the set of strips
_v
o-',
o-'
(K)
= O,
THEORY OF INFINITE
140
Any
one,
SERIES,
WHOSE TERMS
strip lies within, or at most reaches the boundary of the preceding

their breadth tends to zero as their number increases.
and
Further, the curve y=f(x)

the strips.
lies
within, or at most reaches, the boundary of
This construction, therefore, not only establishes the

continuity of the sum
of the series of continuous functions, in an interval of uniform
convergence,
but
shows that the approximation curves, as the number
it
may be
increase,
of the terms
used as a guide to the shape of the curve for the sum right
through the interval.*
70.
sufficient Condition for Term by Term Integration of a

whose Terms are Continuous Functions of x. When the
Series

sum, f(x), is continuous in (a, b).
integrable between x and x lt when a
is
(a, b),
But
it
we have
seen that
It follows that f(x) is
its
<#! = &.
does not follow, without further examination, that the
series of integrals
fxi
fXl
x
Jx
convergent, and, even
is
fXi
ul (x)dx+\ u2 (x)dx+\ u3 (x)dx+...

J jc
if it
be convergent,
f*i
X
it
does not follow,
f(x) dx.
The geometrical treatment of the approximation curves in 69

suggests that this result will be true, when the given series is
uniformly convergent, arguing from the areas of the respective
curves.
We
shall
now
demonstration
state the
theorem more precisely and give
u9 (x\ u3 (x),... be continuous

u (x) + u z (x) + us (x)+...
Let the functions u^x),
and
let
its
the series
in
(a, b),
be
Then
where
*
f(x)dx=\
(a, b)
and have f(x) for
its
sum.
u (x)dx+\ u2 (x)dx+
l
a^x<
Of course the argument of
graphically represented.
this section applies only to such functions as can be
141
Let the arbitrary positive number e be chosen.

Since the series is uniformly convergent, we may put
where
|
e
Rn (x) |<^
o ~- a
when n = v,
the same
v serving for all values of x in (a,

Also f(x) and Sn (x) are continuous in
6).
and therefore
(a, b)
integrable.
Thus we have
f(x)
Jx
dx =
I
Jx
Sn (x) dx +
|
Jx
R n (x) dx,
where a =
Rn (x)dx
Therefore
x
^-X
b
a
<e, when n = v.
l
But
r^W*) dx=Y^ u r (x) dx.
Jx
\
\Jxo
C*l
f*l
f(x}dx-Y\ u r (x)dx <e, when n^v.
Therefore
Jxo
JA,
f(x)
fx\
X
COROLLARY
and
Let u^x),
I.
u 2 (x), us (x),...
be
continuous in
dx.
(a, b)
the series
to f(x) in (a,
the series of integrals
converge uniformly
Then
fu (x)dx+\
l
J .>o
*0
b).
uz (x)dx+\ u3 (x)dx+
J
...
.'o
fx
f(x)dx in
(a, b),
when a = xQ <x = b.
_Jx
This follows at once from the argument above.
COROLLARY
(a,b)
and
II.
Let u^x),
ii.2
(x),
the series
converge uniformly
to f(x)
in
(a, b).
u3 (x),
...
be
continuous in
THEORY OF INFINITE
142
Also
let
and
g(x) be hounded
integrable in (a,b).
co
f.tr
x
where a = x
WHOSE TERMS
SERIES,
r.e
f(x)g(x)dx=y]\J* u n (x)(j(x)dx,
<x =
b,
and the convergence of the
uniform in (a, b).

Let the arbitrary positive number
the upper bound of \g(x) in (a, b).
series of integrals
is
let
M be
converges uniformly to f(x) in
(a, 6),
be chosen, and
00
Since the series
û
n (x)
we may put
where
\R n (%)
the same
v serving for
Therefore we have
\*
Jx
< 17/7{0--a)v
values of x in
all
f(x)g(x)dx =
wnen n = v,
r~
J.U
(a, b).
X
\
Jx
Sn (x)g(x)dx+ f
Jzo
n (x)g(x)dx,
where
Thus
f(x)g(x)dx-
R n (x)g(x)dx
Sn (x)g(x)dx
And
,(<%(*) <*x
<^
e,
when n ^
i/,
which proves our theorem.

It is clear that these integrations
times as
Ex.
we
many
To prove that
+
I log(l-2?/cos.r ?/
We
can be repeated as
wish.
know
)rf/r
= 0, when
|y|<l,*
= TT log
when y
2
?/
'
>
that
~ COS
'
l-2j/cos.r+#
9
2
=- cos x - y cos 2.r - y
cos
3./-
+ ................ (1)
when \y\<I.
* It follows
and
from Ex.
^ respectively.
4,
118, that
we may
replace the symbols
<,
>
by

Also the series
(1)
143
converges uniformly for any interval of y within
(-1,1) (67).
i.-y
Therefore
dy =
Therefore J log (l-2y cos x + if)
"
^^
C S
=-|î^y
But the series () converges uniformly

number less than unity ( 67).
when
'
;/
<l
'
'
when |y|<l ............. (2)

when y
for all values of x,
is
some
positive
Thus
\ ( log (1
- 2y cos x +3/ 2 ) dx = - f) ^-
Jo
Therefore
But
t cos ?^ d.r,
njQ
2
when
J \og(l-2ycoax+y )dx=0,
( log ( 1
Therefore
71.
log(l
A suflficient
<
-23/cos^+y )c?.r- = 7rlog?/

2
:i
\y\<l.
T Qogy + log fl - - cos.r +
- 2y cos x +# 2 ) dr =
Condition for
when y
\J J
dx.
when |y|>l.
Term by Term
Differentiation.
If the series
converges in
coefficient,
(a, b)
and
continuous in
each of
its
and
(a, b),
terms has a differential

if the series of differential
u'
coefficients
converges uniformly in (a,b), then f(x), the

series,
has a differential
sum
of the original
coefficient at every point of (a,
b),
and
Let
Since this series of continuous functions converges uniformly
b), we can integrate it term by term.
in (a,
Thus we have
ci
f*l
f-
J XQ
where
</>(x)dx=\
Jr-o
f*i
(Vi
u '(x)dx+\ u2'(x)dx+\
J.r
J.r
l
"
ft
a^x^^x^b.
^x^
1
Therefore
0(^
But
and
Therefore
0() dx =/(.^) -f(x Q
).
THEORY OF INFINITE
144
Now
XQ = x
put
SERIES,
and
Then, by the First Theorem of
where x =
WHOSE TERMS
x1 =
Mean
Value,
rn U
e
Therefore
/"(a;)
But
Lt
since 0(#) is continuous in (a,
b).
Therefore f(x) has a differential coefficient f(x) in
and
a;
(a,
It must be remembered that the conditions for continuity, for

term by term differentiation and integration, which we have
obtained are only sufficient conditions.
They
are not necessary
We
have imposed more restrictions on the functions

than are required. But no other conditions of equal simplicity
have yet been found, and for that reason these theorems are of
conditions.
importance.
It should also be noted that in these sections
been dealing with repeated limits
(cf.
64),
we have again
and we have found
that in certain cases the order in which the limits are taken
may
be reversed without altering the result.

In term by term integration, we have been led to the equality,
in certain cases, of
f*i
Jx n
Lt Sn (x)dx
and
><
7i
Similarly in term
by term
Lt
>w
fî
Sn (x)dx.
Jx
differentiation
we have found
that,
in certain cases,
Lt
-.
Lt
h->QL n ->,\
li
and
Lt
M->OO
Lt
U-
are equal.
72.
The Power
Series.
The properties
of the
Power
Series
are so important, and it offers so simple an illustration of the results we have

just obtained, that a separate discussion of this series will now be given.

I.
is
+ a^x + a.,j; 2 + ...
the series
//'
x=x^
convergent for
that
<
|.*-|
\x
absolutely convergent for every value of
in
it
such
\.
Since the series
is
convergent for
such that
anX
<^
x=x
there
= c<l,
if
is
when n >
But
Therefore
145
a positive
number
'
are less than the corresponding terms of the convergent series
J/{l+c + <*+...},
and our theorem
series does not converge for
If the
II.
follows.
value of x such that
#
|
This follows from
that
|
,v
\
|,
it
It follows
III.
occur
>
>
I.,
it
does not converge for any
since
the series converges for a value of
if
must converge
from
# = .r
._
and
I.
for
II.
x = .r
.v,
such
that only the following three cases can
(i)
The
series converges for
E.g.
(ii)
The
= and
1 + 1!
.r
no other value of
x.
series converges for all values of x.
E.g.
(iii)
There
is
some positive number p such
converges, and,
when \x\>p the

t
E.g.
The
-p<x<p
that,
when x\<p the

t
series
series does not converge.
tf-^+Ja3 -....
called the interval of convergence of the series.

Also it is convenient to .say that, in the first case, the interval is zero, and, in
the second, infinite. It will be seen that the interval of convergence of the
following three series is ( 1, 1)
interval
is
But
it
should be noticed that the
of the interval
first of
these does not converge at the ends

and the third
the second converges at one of the ends
converges at both.
In the Power Series there cannot be first an interval of convergence, then
an interval where the convergence fails, and then a return to convergence.
THEORY OF INFINITE
146
SERIES,
WHOSE TERMS
Also the interval of convergence is symmetrical with regard to the origin.

shall denote its ends by L', L.
The series need not converge at L' or L,
but it may do so and it must converge within L'L.
We
IV. The
V. The
series is absolutely convergent in the
and uniformly
series is absolutely
==-r
=p
&,
where 8
is
open interval
- p<x<p.
convergent in the closed interval

less than p.
any assigned positive number
FIG. 14.
we have only to remark that if

is a point ,r n
where
between JV and the nearer boundary of the interval of convergence, there are values of x for which the s.eries converges, and thus by I.
To prove
IV.,
-p<x^<p,
it
X=XQ
converges absolutely for
\L
M'
O
FIG.
15.
To prove V., let M', Jf correspond to x = p + 8 and ,r=p B respectively.

now choose a point N (say .r ) between
and the nearer boundary L.
The series converges absolutely for xx^, by IV.
We
Thus, with the usual notation,

I
a nxQn
n
+
\a nx
But
n+l
an+l x
n+1 .r
n+1
|
+ ...<, when
+ ...
v.
than the above for every point in M'M, including the ends M', M.
It follows that our series is absolutely and uniformly convergent in the
closed interval (M', M)* And the sum of the series is continuous in this
is less
closed interval.
It remains to
examine the behaviour of the series at the ends of the

and we shall now prove Abel's Theorem t
interval of convergence,
VI. If the series converges for either of the ends of the interval of convergence,
the interval of uniform convergence extends up to and includes that point, and the
continuity of f(x\ the
sum of
the series, extends
up
to
and
includes that point.
Let the series converge for x=p.

Then, with the usual notation,
1)
and
\
})
Rn (p) = a n pn + a n+l p
Rn (p)\<e, when
*When
+l
+ ... + an+p _
n+
P-\
n^-v, for every positive integer p.
the interval of convergence extends to infinity, the series will be
absolutely convergent for every value of x, but it need not be uniformly convergent in the infinite interval. However, it will be uniformly convergent in any
interval ( - b, b), where b is fixed, but may be fixed as large as we please.
E.g. the exponential series converges uniformly in
be arbitrarily great, but not in an infinite interval
may
f/. Math., Berlin,
1, p. 311, 1S26.
any fixed
[cf.
65],
interval,
which

RH
But
JG
= a,,v H + a nl ^n+l +
and the factors (-
when 0<.r</3.
Thus
\pj
. . .
+ a il+
(-)
,...(-)
\p/
\p/
^.v'
-1
l+
are all positive
Mx) = iRn(p)c
147
and decreasing,
where c='-.
P
This expression
But
-c" +1
may be
c"* 1
-^* 2
...
- 1 are
c" +p
and
all positive,
when n v.
when n
n - n+1
n+l - c n+2
c
p/? M (.-) < [(c
) + (c
)+
are all less than
Therefore,
|
the form
re- written in
i/,
. . .
(C
H+P ~-
- c"*''- ) + c'"^1
<C"
<
e,
provided that
But we started with
R H (p)
Thus we have shown that the
Combining
[convergent
number
less
than
p.
And
< x < p.
< when
e,
series is
this result with V.,
in the interval
we
see that the series
-p + 8^x=p, where
it
v.
follows that /(*') the
tinuous in this closed interval.
In particular, when the series converges at
.r
= p,
Lt
-
In the case of the logarithmic
series,
the interval of convergence is Further, when = !, the series converges.

.?-
It follows
from Abel's Theorem that
i.e.
Iog2 = l
- +
-...
is
sum
is
now uniformly
any assigned positive

of the series, is con-
THEORY OF INFINITE
148
SERIES,
WHOSE TERMS
Similarly, in the Binomial Series,
when
And
is
< .?;< 1.
known *
it is
that
+m+
when
conditionally convergent
~^ +
. . .
l<wi<0, and
absolutely convergent
when m>0.
Lt (l+.r)"
Hence
2!
both these
in
On
cases.
the other hand,
we put x = 1 in the
The uniformity
if
which does not converge.
is
for the interval
than
-l^x^l,
where
series for (1 +.r)~ 1 ,
we
get a series
of the convergence of the series
any given positive number
is
less
1.
VII. Term by term differentiation and integration of the Power Series.

We have seen in V. that the Power Series is uniformly convergent in any
of
closed interval M', M[ p +
8] contained within its interval
8^x^p
convergence
Z',
p < x < p].

70 that the
series may be integrated term by term in the

from
and the process may be repeated any number of times.
now show that a corresponding result holds for differentiation.
We
From the theorem proved in 71, it is clear that we need only show that
It follows
M\
interval M',
shall
the interval of convergence
is
(-p<x<p) of
also the interval of convergence of the series
a1 + 2
To prove
is
this, it will
convergent when x
|
When x
|
Then the
< p,
let
.r
+ 33#2 +
. . .
be sufficient to show that
p.
p.
2
,
a>
convergent, because the ratio of the n term to the preceding has for
limit
which
is less
than
1.
P
If
we
multiply the different terms of this series by the factors

\
a \\P-
Of. Chrystal, Algebra, Vol. II. p. 131.
its

which are
than some fixed number, since (by IV.) the series
all less
convergent,
is
convergent when x
it is
If this series
< p.
show that this

were convergent
to
hold for the series

,,,
and
which we thus obtain, namely
clear that the series
is
We have yet
149
also for the series

|
a^
+g
+
x=
,*,
when x
last series diverges
for
a^
2
1
\x l
where
|,
+ 3 !,*,
+
> p.
same would
+...,
a^r,
\x\ |>/J, the
. . .
since the terms of the latter are not greater than those of the former.
But this is impossible, since we are given that the interval of convergence
of the original series is
Thus the
and the
< x < p.
a^ +
a^ + a^x2 + a3x3 +
'! + 2%? + 3% -f
series
. . .
obtained by differentiating
vergence,* and
...
series
it
term by term, have the same interval of con-
it
follows from
f(' V)
71 that
when x is any point in the open interval

p < x < p.
Also this process can be repeated, as in the case of integration, as often
;is
we
Extensions of Abel's Theorem on the Power Series.
73.
I.
please.
We have
converges, the
is
72 that
seen in
Power
if
the series
Series
uniformly convergent, when
Oâ-^1 and that,

= + a vv + 2 +
= a + a + 2 -+....
/(ar)
;
f(x)
Lt
. .
if
x >1-0
The above theorem
.>6-
of Abel's is a special case of the following
CO
Let the series
%a n converge, and
ay
a,,
a2
...
a sequence of positive numbers
be
such that
^ a < 04 < a
when /fÔ,
. . .
and if /() = !
Then
the series
2a
a n e-*nt, we have Lt
<e
~ *"'
is
t-*+o
Consider the partial remainder p R n (t) for the series

*
If
we know that Lt
ratio-test for convergence, since in each series

,
,
if
|*1
<
Lt
Sa Me~ a
exists, this result follows
n+l'
Lt
uniformly convergent,
= â n
f(t}
" (.
immediately from the
THEORY OF INFINITE
150
WHOSE TERMS
SERIES,
RH (t) = a n e - -f u n +i& ~ a + + + a n+p -ie - +p= ir H - a* + ( rn - !-/> - a+i + + p rn - _ rn )e --u-i,

= n + +i, ... these being the partial remainders for
where irn =an
Then
<
ie
1*
. .
2 ''n
series
. . .
(p
l}
the
,.
Rewriting the expression for

/>fln(0
But the
= \r n (e -
series
2a,
R n (t), we
obtain, as in
72, VI.,
- e-
o-nt
is
convergent.
Therefore, with the usual notation,
/,n|<, when w^v,
Also the numbers e--nt,
>
when
when
/4(0 \<e[(e~
<
<
And

...
e-^+sf,
are
all
positive
and decreasing,
0.
It follows that,
,,
e -*+i,
ee a
a-nt
> 0,
-e- *.+i<) + (e -
e,
this also holds
+i* - e -
//+=<)
. . .
-j-
+;;- 1^]
<
when ?t v,
when = 0.
CO
Therefore the series
2a
tl
e- a u
is
uniformly convergent when t^.0.
Let
its
sum be /(*).
Lt /(0=/(0) = S
Then
<->+o'
.*
o
00
II.
In Abel's Theorem and the extension proved above, the series
are
supposed convergent. We proceed to prove Bromwich's Theorem dealing

with series which need not con verge, t In this discussion we shall adopt the
following notation
...+ a n
and we write $
6>
0) s l>
for the Arithmetic
*^j
shown
It can be
t
(cf.
=
n
102) that,
then, with the above notation,
* If
of the first n terms of the sequence
S 2i
Ihus
Mean
ttj
if
the series
Lt S n = S.
are functions of x and the series
]#* converges and

o
its
sum
But the converse does not
2 a n converges uniformly
is
hold.
to F(x) in
a given interval.
Then
it
follows from the above argument that Lt

t
>+o
2ae- a
converges uniformly to F(x) in this interval.
\Mafh. Ann. Leipzig, 65, p.

Amer. Math. Soc., 25, p. 258,
,
350, 1908.
1919.
See also a paper by C. N. Moore in Bull.

The sequence
of Arithmetic
Means may converge, while the sum
151
n fails to
2)
converge.*
series
â n converge to
A Iso let u n be a function of t
S.
Sn
Let the sequence of Arithmetic Means
Bromwich's Theorem.
for the
with the following proper-
w hen t>0
ties,
(a}
''
i n A% <
(/?)
Lt nu n =0,
any ? ositive intc9* rs ; A", a positive\

J
\number independent ofp, q and t
Pt q
A't (
'
?=!.
Lt
(y)
->+o
OS
00
Then
the series 2, a, u n converges

t
when t>0, and Lt
We
have
,w M = (7
?f
+ (o"i
=o
(T
- 2o- + o- =
Thus
'2t a, l u n
<-+3
--
- So-o)^! + (o- 2 -
( 'o
+ (r
2o-!
= S.
etc.
+ (w
)w 2 4-
2cr, t _
+ o-
(t
.,)
UH
Therefore
M+1 -0-,^,,+..,
cr,,_,<f M+
(1)
But
sequence of Arithmetic Means
tlie
converges, and Lt
*3l-"
<2>
*1>
Sn = S.
It follows that
KM < (ft + ljC' for

I
Also from
(/?)
is
number
(<r n
(7,
not
than \S\
less
such
that
w.
clear that
it is
Lt
n *
there
every integer
M w+1 )= Lt
(cr n ?
M+2 )=
u >
Further, the series
2 (^ 4- 1
A'-V
j
/(
Lt
(<r n .
n xo
M n+1 )
................... (2)
converges, since, from
(a),
the series
o
00
2w A
I
*If
w n converges.
I
a,,
-.1)",
nÔ,
it is
obvious that Lt
/t
^n = 4, but
the seriesâ,,
is
not con-
>
00
But
vergent.
see Hardy's
Theorem,
102, II.
When
Lt S H = S, the
series
2>
is
n >eo
often said to be " summable (Cl)" and
For a
its sum (C 1) is said to be S.
method of treating series, due to Ce.saro, reference may be made
Also see
to Whittakcr and Watson, Course of Modern Analysis, p. 155, 1920.
101-103 and 108.
below,
discussion of this
A 2 ?t,
is
written for (u n - 2 H+1
2}A*tfn are
j
+ ?/+;>).
Since all the terms in the series
positive, this condition (a) implies the convergence of this series.
THEORY OF INFINITE
152
Also
Therefore the series

It follows
from
<r w
A 2 M M < C(n+l)'\ A 2 wn
<r n
A% n converges absolutely.
and
(1)
(2)
=1
special case
<r H = n+l
we have
|.
that
oo*
Taking the
WHOSE TERMS
SERIES,
at =
= 0,
. . .
M=2 (tt+l)A
and
fi
tt.
(4)
Thus, from
(3)
and
(4),
Su = 2 (o"n
2o a n u n
Therefore, to the arbitrary positive

integer v such that
n+
Thus
|o-,,|
<(M + 1)(7,
It follows,
o-,,
S)
?t n
(5)
Lt -^- = S.
_*, W+l
Now
Also
(n +
-S <
l
(n +
1)
S
|
<
number
when n
r?,
e,
(M
+ 1 when w^ r.
),
for every positive integer,
from these inequalities and
v.
w - Sun |^| "sVn - (n + l)/S)A% H

"
and
C.
|<S|
that
(5),
+ 2f (/
1
(w
+ l)lA% H
i+D|A% M + ^|;(w
4A v
But
(6)
'
<2A^by
And
Lt A"w w =0,
It follows that, v being fixed, there

,,
Thus from
(6), (7)
is
a positive
number
?;
such that
idw^v-l
and
(7)
(a)
Lt u n =l,l>y(y).
since
(8),
we
(8)
see that
a n un - Su \<^
+ ^<f, when (Xtîj.
Lt
Therefore
-X-(-0
And,
since,
Lt
finally,
from
Lt
(y),
?<
= 1-
--+o
III. Let the sequence of Arithmetic Means for the series
and
Lt
let
un
be e~
2a n u n =S.
nt
(or e~'
l2f
).
Then
the series
â n un
â n
converges,
converge to S,
when
>(),
and

This follows at once from Bromwich's Theorem above,
(a), (/?) and (y) of that theorem.
if
153
e~ nt (or e~ n2t ) satisfy
the conditions
It
obvious that
is
that (a)
(i)
(ft)
and
(y) are satisfied, so it only
remains to establish
is satisfied.
Let
tt
A2
Then
= e-'",
*>0.
= un - 2w H+1 4- u ll+
*-" (!--'),
Therefore
A2
,,
is
positive.
i>A%
Also
Therefore
2>|A%n =e~
f
,
and the condition
(a) is satisfied.
(ii)
Let
= <?-"%
/>0.
In this case
where
M M = e-<"+^ (4(w + 0) 2 * 2 - 20,
0<0<2.*
Therefore the sign of A-M M depends upon that of (4(n +

It follows that
it is
0)*t'
2/).
positive or negative according as
and
Therefore
value of
But
it
A% n cannot change sign
more than three times
for
any positive
t.
follows at once from the equation
that a positive number, independent of

than this number for all values of
is less
t,
can be assigned such that
ft
n.
Hence K can be chosen so that the condition (a) is satisfied, provided that
sum of any sequence of terms, all of the same sign, that we can choose
from 2wA 2 M, is less in absolute value than some fixed positive number for
the
all
values of
Let
t.
2wA 2 M w
be the
sum
of such a set of consecutive terms.
Then we have
This follows from the fact that
where
0<0<2,
(CL Goursat,
provided that f(x), f'(x), f"(x) are continuous from x to x + 2h.
loc. cit.,
T.
I.,
22.)
THEORY OF INFINITE
154
which
differs
SERIES,
WHOSE TERMS
from
r2t
r(e~
- e -('"H>
<)
by at most unity.
But, when n is a positive integer and t>0,
< n (e-" - e-i'W) = 2n (rc + 0) te-<"+<
2'
2<
...
(0
< B< 1
Therefore, for the set of terms considered,
Then the argument above shows that the condition

IV. Let the sequence of Arithmetic Means for the
Then
the series 2X,.v
will converge
when
(a) is satisfied.
series
â n
converge to S.
0<#<1, and
Lt
x >1-0
This follows from the
first
part of III. on putting x = e~*.

00
V. Let the terms of the
series 2,a lt be
functions of x,
Arithmetic Means for this series converge uniformly
S(x) in
a^x^b.
Lt
Then
f
>+
and
to the
â n u n converges uniformly
the sequence
of
bounded function
to
S{x) in
this in-
terval, provided that u n is a function of t satisfying the conditions (a), (/?) and
(y) of Bromwich's Theorem, when t>0.
This follows at once by making slight changes in the argument of II.
The theorems proved in this section will be found useful in the solution
problems in Applied Mathematics, when the differential equation, which

corresponds to the problem, is solved by series. The solution has to satisfy
of
certain initial arid
boundary conditions.
What we
really need is that, as
we approach
the boundary, or as the time tends to zero, our solution shall

have the given value as its limit. What happens upon the boundaries, or at
the instant
= Q,
is
not discussed.
(See below
123.)
Integration of Series. Infinite Integrals. Finite Interval. In

70 we dealt only with ordinary finite integrals. We shall
now examine the question of term by term integration, both when the
integrand has points of infinite discontinuity in the interval of integration,
74.
the discussion of
supposed finite, and when the integrand is bounded in any finite interval,
but the interval of integration itself extends to infinity. In this section we
shall deal with the first of these forms, and it will be sufficient to confine the
discussion to the case
(a, b),
I.
when the
say x = b.
infinity occurs at
Let Wi(#), u z (x\ ...be continuous in
uniformly
to
f(&) in
(a, b).
(a, b)
and
one end of the interval
the series
~i( (x) converge

lt

A Iso let g (x) have an
infinite discontinuity at
roo
convergent.*
and
g (x) dx
155
be absolutely
rb
f(x) g (x) dx = ^,\ u n (x) g (x) dx.

1
From
f(x)
the uniform convergence of
Ja
2 un {x) in
(a,
we know
b\
that
its
sum
continuous in (, b\ and thus bounded and integrable.
is
Also
Ja
f(x) g (x) dx
is
absolutely convergent, since \g(x)dx\s so (61,VL).

Ja
Let
Then, having chosen the positive number
where
|
Rn (x) < ^, when n^ v,

A
\
Jm
we
all
please,
we may put
values of
in (a,
b).
rb
"b
But
as small as
e,
f(x)g(x)dx and
Ja
Sn (x)g(x)dx both
exist.
rb
It follows that
Ja
Rn (x)g(x)dx also
and that
exists,
Sn (x)g(x)dx+ \
/V(*)0(*><fc- Ja
l
Ja
Ja
Thus we
Rn (x)g(x)dx.
see that
l"R
Ja
<e, when w
which proves that the
rb
sum
is
/
Ja
Ex.
rb
series
n (x)g(x}dx
u n (x)q(x)dx
is
v,
convergent and that
its
'
f(x)g(x}dx.
This case
is
illustrated
by
___
n+l
n
2-2
log 2
^7r
I
It is clear that this proof also

applies
discontinuities in (a, b)
and
g(x)dx
Ja
t Cf.
Carslaw, Plane Trigonometry
is
when
2
,
using the series for ^n^.

i'2
.7(0;)
has a finite number of infinite
absolutely convergent.
('2nd Ed.), p. 279.
THEORY OF INFINITE
156
Here the
series for log(l
SERIES,
WHOSE TERMS
iuO^x^l,
+x) converges uniformly

n
log x dx
and
Jo
converges absolutely (as a matter of fact log x

while log A>ac as x->0.
0<^^1),
On
cases
always of the same sign in
the other hand, we may still apply term by term integration in certain
the above conditions are not satisfied, as will be seen from the
when
following theorems
II.
is
Let
WI(A'),
u%(x\
converge uniformly
Further,
let
to
be continuous
...
and
positive
f(x) in the arbitrary interval (a,
and
a),
g(x) be positive, bounded and integrable in
A*)*(*)n(x)g(x)d
ft
f(x)g(x)dx converges.
In other words, we are given that the repeated limit
Lt
71
f6-f
->0 J a
û
Lt
H->CC
Since the functions Ui(x\ u%(x\
...
from the convergence of
(x)'\g(x)dx exists.
are
as well as g(-v\ in
all positive,
(,
a),
f(x)g(x)dx there follows at once the convergence
of
b
l ur (x)g(x)dx
(r
= l,
2, ...).
Ja
Again,
Then
/(#) = u^ (x) + u^(x) +...+ un (x) +
let
Rn (x)g(x)dx also
RH (x).
converges, and for every positive integer n
-'a
b
Jl
a
R
f(x)g(x)dx-\
UrW<j(x)dx=\a n (x)g(x)dx.
1 **
n
But from the convergence
of
it
f(x)g(^v)dx
follows that,
arbitrary positive number e has been chosen, as small as

be a positive number such that
0<
A fortiori,
and
this holds for all positive integers n.

(a, b
J)
series
............ (1)
2wM (.r) converges
we
when
the
please, there will
f(x)g(x)dx<.
R n (x)g(x}dx<\,
0<f
*
Jb-g
Let the upper bound of g(x) in
The
............................. (2)
be M.
uniformly in
(a,
&-)

Keeping the number
will be a positive
we have chosen above, there
157
integer v such that
A>
a)
0<flii(a')<oi//A
ZM(o
x
the same v serving for all values of

6 "*
when n
in (a, b
=v
>
).
&"
Thus
<^,
Combining these
results (2)
and
(3),
when
n^v......................... (3)
we haVe
when n^v.
0<J R n (x)g(x)dx<t,
Then, from
(1),
0<
rf(*)a(*) dx
Ja
-2
I
roo
f(x) g (.r) rf.r
u r (x)g (.r) c?.r < e, when w
Ja
= 21
1
v.
r&
UT (>v) ff (#) dx.
'a
The other alternative, stated in the enunciation, may be treated in the

same way.
Bromwich has pointed out that in this case where the terms are all
is substantially the
positive, as well as the multiplier g(x\ the argument
same as that employed in dealing with the convergence of a Double Series of
theorem
positive terms, and the same remark applies to the corresponding
in $ 75.*
Ex.
Show
1.
that
-^
<**
= if xn \^xdx t
Ex.
an
The conditions imposed upon the terms u^.v), ^2(^)5
('*)
may be removed, and the following more general theorem stated
."7
are positive,
oo
III. Let Ui(x),
2 (#), ...
be continuous
and
the series
*\u n (x)\
*
formly in the arbitrary interval (,

* Cf Bromwich,
.
a),
where a <a<-&.
Infinite Series, p. 449,
Also
and Mess. Math.
converge uni00
let *2,u n (x)f(x).
Cambridge, 36,
p.
1906.
fThe
the
72,
interval
first
VI.
we
(0, 1)
has to be broken up into two parts, (0, a) and

I. and in the second Theorem II.
Or we
use Theorem
(a, 1).
In
may apply
THEORY OF INFINITE
158
Further,
let
g(x) be bounded
and
WHOSE TERMS
integrable in (a, a).
f /(*)?(*)<**= if un (x)g(X)dx,
Then
Ju
1 -'a
provided that either the integral
2/
SERIES,
Ja
^\u n (x}\.\g(x)\dx
or the series
\u n (x)\g(x)\dx converge.
This can be deduced from
un g = {un + \u n
since that
II.,
using the identity
}{g + \g\}-{n n + \u n \}\g\-\u n \{g + \g\\ + \un
theorem can be applied to each term on the right-hand
Ex.
1.
Show
that
Ex.
2.
Show
that
dx = 2( - 1)"
\g\,
side.
x" log x dx
12'
when p + 1 > 0.*

75.
Integration of Series.
For the second form

the theorems proved in
I.
Interval Infinite.
Infinite Integrals.
of infinite integral
we have
results corresponding to
74.
Let u\(x\ U 2 (x), ...be continuous and bounded in x^.a, and
ûn (x) converge uniformly to
f(x) in
x^La.
Further,
integrable in the arbitrary interval (a, a), where

large as we please
and
let
Ja
g(x)dx converge
Too
Then
Ja
The proof
of this
ao
g(x) be bounded
a<a, and a may
and
be chosen as
absolutely.
r*>
f(x)g(x)dx=^\Ju u n (x)g(x)dx.
1
theorem follows exactly the same
EX.
let
let the series
lines as
I.
of
74.
ffr^f^^L**-
This follows from the fact that the series
converges uniformly to \fx in #

* If
^>>0, Theorem
III.
1.
can be used at once.
0>j9> - 1, the interval has to be broken up into two parts (0, a) and (a, 1).
In the first we use Theorem I. and in the second Theorem III. Or we may apply
If
72, VI.

But
/
by term integration when
often necessary to justify term
it is
is
\g(.v)\dc
2t n(#) can only be shown
divergent or
159
either
to converge uniformly
* (I
in the arbitrary interval (a, a),
where a can be taken as large as we
please.
important cases are included in the following theorems, which

correspond to II. and III. of 74
Many
00
II. Let MI(#),
u<) (x),
...be continuoiis
and
positive
and
the series
verge uniformly to f(x) in the arbitrary interval (a, a), where a
ûn (x)
may
con-
be taken as
large as we please.
Also
let
g(x) be positive, bounded and integrable in (,
f(x}g(x)dx =
Then
un (x}g(x}
a).
dx,
rao
,-QO
prov ided that either the integral
I
"
Ja
f(x\g(x)dx or
the series
2 â un (x)g(x)dx
/
converge.
Let us suppose that

In other words
we
f(x)g(x)dx converges.
are given that the repeated limit

a
Lt
[Lt
-> x a n
2ur(x)]ff(x)djt; exists.
> 00 1
00
Since the terms of the series
A'^,
'a
ûr (x~)
are
all positive,
as well as
g(x\
f(x)y(x)dx there follows at once that
f Ur (x)g(x)dx
(r=l,2,
in
of
...)
-'n
Then
/(#)= Ul (x) + u,(x) + ... + u n (x) +
let
Again
/
Ja
R n (.r)g(x)d.r also
I
Ja
"/(*)*(*)<**-
lt n (x}.
converges, and for every positive integer
if M#)0(.r)d>= f R(x)g(*)dr
1
But from the convergence
-a
of
->
Ja
we
this holds for all positive integers n.

this choice of a, let the upper bound of g(.r) in (a, a) be
With
00
The
series
2wn (.r)
converges uniformly in
(a, a).
when
the
please, there will
0< /^ Rn (.v}g(x)dx<^>
A fortiori,
and
........... (1)
f(.v)g(.r)d.r, it follows that,
arbitrary positive number e has been chosen, as small as

be a positive number a such that
M.
THEORY OF INFINITE
160
Keeping the number
we have chosen
WHOSE TERMS
SERIES,
above, there will be a positive
integer v such that
all
<
Thus
and
values of
in (a, a).
Rn (x}g(x) dx <
when n ^ v.
|,
this holds for all values of n.
Combining these two
0<J
and from
we have
results,
Rn (x)g(x}dx<t, when
n~^.v,
(1),
u r (x)g(x)dx<e, when
0<|Ja f(x)g(x)dx-^l
Ja
n^.v.
f f(x)g(x}d.v =
Therefore
''a
The other
^l
Ja
u r (x)g(x)dx.
oc
e- ax
Ex.
same way.
alternative can be treated in the
2.
=2
wshbxdx
e' ** cosh bx dx
Jo
/<*>
7,2n
e-
=2^
tue
if
.i* rf,v
0<|6|<a.
2nlJo
Further, the conditions imposed upon u^(x\ u.2 (x\ ... and g(x), that they
be positive, may be removed, leading to the theorem
shall
00
III. Let Ui(x\
u 9 (x),
...be continuous
and
the series 5)|w(#)| converge unii
formly in the arbitrary interval

Also
please.
let
(a, a),
where a
may
be taken as large as u*e
2 un (x) =f(.v).
i
Further,
let
g(x) be bounded and integrable in
(a, a).
Then
ao
2
roo
.V
w )i(^)
fl(-v)
^'
or the series
r*>
2Ja
/
un (x) \g(x) dx
\
converges.
This
is
deduced, as before, from the identity
u n g = {un+\u n \}{g + \g\}-{u n + \u n \}\g\-\u n \{g + ff \} + \u n \g\,

since the Theorem II. can be applied to each term in the right-hand side.
\
Ex.
1.
Show
that
e~ax cos bx dx
& ^^ f
e- axyr n dx,
if
<
b
|
< a.

Show
that
Ex.
2.
76.
Certain cases to which the theorems of
by the following test
to
75 do not apply are covered
Let u v (x\ u z (x\ ...be continuous in
uniformly
161
xâ, and let
f(x) in the arbitrary interval
(a, a),
the series
where a
may
2 u n (x) converge
be taken as large
as we please.
Further,
let the
integrals
u l (x)dxt
converge,
and
M! (x) dx +
converge uniformly in
the series
converges,
and
x^-a.
the integral
! f(x)
Ja
Sn (x) =
f(x) dx converges.
dx =
dx +
Ja
u 2 (v) dx+....
MI (x)
dx +
Also we know that Lt Sn (x)

X
now show
Ja
Then we know that Lt Sn (.v)
shall
M! (x)
-'a
Ja,
We
u.2 (x)dx+...
'
Ja
of integrals
A Iso
Let
etc.,
the series of integrals
Ja
Then
u z (x)dx,
\
* (i
u^(x) dx +...+ I* u n (x) dx.

Ja
exists in
exists,
z^La, and we denote
and we denote
it
>
by
it
by
F(x).
G(ri).
that Lt F(x) and Lt G(ri) both exist, and that the
two
limits are equal.

From this result our
theorem as to term by term integration will follow
at once.
I.
To prove Lt F(x)
x
Since Lt
n
notation,
exists.
*ao
Sn (x)
>
converges uniformly to F(x) in
we have
F(x)
/ \
- Sn (x)
<-
when n
xâ,
v,
the same v serving for every x greater than or equal to a.

Choose some value of n in this range.
Lt Sn (x) = G(ri).
Then we have
Therefore
we can
choose
X so that
C -,
c. i
when
x"
> x'
with the usual
THEORY OF INFINITE
162
But
WHOSE TERMS
\F(x")-F(x')\
F(x")
- S n (J
<e, when
Thus F(x) has a limit as
II.
SERIES,
G (n)
To prove Lt
")
.r->oo
- $n(x) + Sn (J) - F(x')

|
exists.
n-^-ao
Since Lt
n (.r)
n >
converges uniformly to F(x) in

I
the same v serving for every
But
we can
choose
when n">n'^v,
"(#)-'(#) <|>
6
Therefore
.r
G(n')
x greater than or equal

Lt Sn
X*m (x)=G(n\
X X
lt
to a.
such that
- Sn ,(x) <-, when x ^ X

3
|
\0(n")-SnH (x)\<l
Then, taking a value of x not
less
> a,
x^X
when
than X^ or JT2
>a.
\G(n")-G(n'}\
w
|
(7(n
<,
- Sn ,,(x) + |^(.r) &,(#) + ^(^) 6-'(n')

|
when n">n'^.v.
Therefore Lt G(n) exists.

n
III.
>x>
To prove Lt F(x)= Lt G(n).
Since Lt G(n) exists,
we
can choose v l so that
when n^.v
<-,
3
00
Again
2
1
fx
u r (x)dx converges uniformly to F(x) in x
a.
Ja
Therefore
we can choose
v2 so that
Cx
!oo
2)
ur (x)dx <-, when n
v2 ,
the same v 2 serving for every # greater than or equal to

Choose v not less than v l or v2
r=
From
the convergence of the integral
Ja
i/
2 u r (x) dx, we can choose X so that

1
<-, when x^.

f
But
u r (x)dx-F(x}
Ja
to
2,u r (x)d.v +
1
163
w,(^)^'c
i>+lja
jsJVw*
<
e,
when
^^
00
Lt / (^) = S
T
Therefore
x-*<a
IV. But
we
r= Lt
Ja
are given that the series
converges uniformly to/(#) in any arbitrary interval

Therefore we have
(a, a).
ul
j*f(.v)dx=j
(x)dv+j* Ui2 (x)d.v+...
in
x^
Thus, with the above notation,
f(x)dx converges, and from
III. that
r
REFERENCES.
BROMWICH, loc. cit., Ch. VII.- VIII. and App. III., 1908.
DE LA VALLÊ POUSSIN, loc. tit., T. I. (3 e ed.), Ch. XI.
DINI, Lezionidi Analisilnfinitesimale, T.
GOURSAT,
HOBSON,
loc. cit.,
loc. cit.,
KOWALEWSKI,
OSGOOD,
STOLZ,
(3
6d.),
Ch.
II., 1'
Parte, Cap. VIII., Pisa, 1909.
II.
Ch. VI.
Bd.
IOC. tit.,
loc. cit.,
I.
loc. cit.,
loc. cit.,
PlERPONT,
T.
Bd.
Kap. VII., XV.
I.,
Vol.
I.,
Tl.
I.,
II.,
Ch. V.
Kap.
III.
Absch. X.
And
PRINGSHEIM, "Grundlagen der allgemeinen Funktionenlehre," Enc.
math. Wiss., Bd.
II., Tl. I.,
Leipzig, 1899.
d.
THEORY OF INFINITE
164
SERIES,
WHOSE TERMS
EXAMPLES ON CHAPTER
V.
UNIFOKM CONVERGENCE.
Examine the convergency
1.
of the series
*>
v
and by
its
means
-V*
Prove that the
2.
non-uniform convergence upon the
illustrate the effect of
continuity of a function of
x represented by an
2
.2
~2
is
convergent for
infinite series.
series
all
+
.
values of
.r,
but
is
not uniformly convergent in an
interval including the origin.
Determine whether the
3.
is
all
values of
Find for what values of x the
4.
Find also whether the

(i)
(ii)
5.
series
.r.
series
2^
converges where
series is
uniformly convergent through an interval including 4-1

continuous when x passes through the value +1.
Discuss the uniformity or non-uniformity of the convergence of the
series
whose general term
is
6.
_ \-(\+x) l-(l+.r)
Un
n
~l+(l+x)
+ + ...
Let
n~ 1
</(,
be an absolutely convergent series of constant terms, and
be a set of functions each continuous

comprised between certain fixed limits,
where A
B are
constants.
Show
in the interval
let
a^.i?^p, and each
that the series
%/oW+i/i(*)+
represents a continuous function of
.r
in the interval
a^.r^

Show
7.
and continuous
is finite
Show
8.
a series
= =
&'
b,
that
series
Examine whether the
for all values of x.
is
by the
that the function defined
165
series is
such values.
all
u^(x] + u v (x) +
if
. . .
of functions each continuous and having no roots

and if
un (x)
in the interval
where
series is uniformly cony, v do not depend on *, then the given

vergent in this interval.
this test to the series
Apply
where
0<a<
1.
THE POWER
SERIES.
rx
9.
From
sin -1
the equation
show that the
series for sin"

1 A-
*+g
is
tan
Show
that
3 a*
~l
n sin "Vi?
'^
this hold
2,
dx = ^^r
,
and
-,
due
27i
|*|
-1
in the
ju
< 1.
T
+ l)-
and
td from the integr
integration by parts
of -
2.4...2w
and tan~U*
cuet
Jo
(2?t
when |*| = 1.^
x-x -
and integrate term by term, when

Also show that
Jo
when * <
. . .
substitute the series for sin" 1 .*;

1
sintan 1 * ,
ain"
* 7,
f*
p tan/"*
f*
Jo
-f-
we may
integrals
-v
-.
also valid
x does
of
obtain Gregory's Series,
11.
"~3K
tan -1 *=
the equation
Within what range
fa.
-^
is
+~
-5
Prove that the expansion
From
10.
*=
tan
-1
- and
= 2(-l)'V^
.
o?*
Jo
TTv>'
J
(2w + l)
'
'--
prove that these
series also represent
f v/(l
I'^-U.,
-.r2 )
and
.'o
*Whena=l,
Lt
n
Goursat,
/oc. c/<,, p.
-? =
l,
.'o
but Raabe's test
(cf.
Bromwich,
loc.
cit., p.
"?*u
404)
shows that the
series converges for this value of x.
35, or
THEORY OF INFINITE
166
12.
If
|
Show
< 1,
that the result also holds for ./;=!, and deduce that
-3 -S + + -
If
=0-43882....
!-
x \< 1, prove that
Does the result hold for x

14.
WHOSE TERMS
prove that
13.
SERIES,
1 ?
Prove that
Express the integrals
as infinite series.
15.
Show
that
-1- i+^ + ^-J,--
A-
INTEGEATION AND DIFFEEENTIATION OF SERIES.

16.
>
y)>
Prove that the
where
a, (3,
series
y and
^(ae-' laX
c are positive
/Se-nfa)
is
and c<y.
Verify that
17.
x between
If it be given that for values of

_
TT
cosh ax = 2
sum air
(I
-
and
acos.v a cos 2.i^

=
^ + -^
j-
-j
TT,
1
...
prove rigorously that
sum ct.v = 2 sinh avr

.
TT
f
-{
sin
;
x
.,
2 sin 2 A'
^^-.-'
>
3 sin 3^
...,
^
...
j

oo
18.
Show
equal to
19.
that
if
- 2^ 2)
When
f(x)
"I
= ^^
f r
2\2
167
j-,,
then
a ^ values of
has a differential coefficient
it
x.
a stands for a positive number, then the series

r
a- r
a~ r
(-l)
|1
r\
a~ Zr +x*
\
o r\ ar^+x?
1
are uniformly convergent for all values of
F(x) respectively,
20.
Find
all
the values of
for
e* sin
Does
which the
x -f e2* sin
and
if
2.v
Show
f(.v)
and
series
. . .
converges.
converge uniformly for these values
of x can the series be differentiated term by term ?
21.*
sums are
their
a -r
it
For what values
that the series
can be integrated term by term between any two finite limits. Can the
and QO ?
function defined by the series be integrated between the limits
If so, is the value of this integral given by integrating the series term by
term between these limits

22.
is
If each of the
terms of the series
satisfies
x.a>Q
in
and
if
the series
the Jf-test (67), then the original series
term from a to
23.
Show
oo
may be
integrated term by
that the series
can be integrated term by term between any two positive finite limits.
and GO ?
this series be integrated term by term between the limits
Can
Show
that the function defined by the series cannot be integrated between these
limits.
24.
Show
that the function defined by the series
can be integrated from

to oo
term integration of the series.
*
and that
its
value
Ex. 21-27 depend upon the theorems of
is
given by the term by
74-76.
THEORY OF INFINITE SERIES
168
25.
Prove that
11
r#-i
l+.#
a,
Explain the nature of the
a+2
a+1
difficulties
involved in your proof, and justify
the process you have used.

26.
By
expansion in powers of
/
Jo
Jo
e~ x (l
e~ ax )
os
tan" 1 (a sin A-)
sin
if
prove that,
a,
< 1,
= log(l + a),
=^7rsinh~ 1 a,
examining carefully the legitimacy of term by term integration

/
27.
/,
,,\
_'
Assuming that
V
show that
Jo
when a > 0.
\""
e~ axJJJ)x)dx=-n-s
x/( a
rrro
+^")
in each case.
CHAPTER
VI.
DEFINITE INTEGRALS CONTAINING AN ARBITRARY
PARAMETER.
Continuity of the Integral.
77.
Finite Interval.
In the
let a, a be constants.
<jt(x, y)dx
fa'
a
Then the integral will be a function of y*
The properties of such integrals will be found to correspond
very closely to those of infinite series whose terms are functions
of a single variable.
Indeed this chapter will follow almost the
same
were
I.
lines as the preceding one, in

treated.
//
y) is
(/>(x,
a = x = a,
then
fa'
a
which such
<j>(x,y)dx is
=y=
(x,
infinite series
y) in the region
a continuous function of y in
the interval
V).
(b,
Since
in
37,
function of
Thus
y)
<t>(x,
is
also a continuous function of
it is
(x, 7/)t,
as defined
x and a continuous
y.
<j>(x,
y) is integrable
with regard to
x.
Let
*
As already remarked in 62, it is understood that, before proceeding to the

limit involved in the integration, the value of y, for which the integral is required,
is to be inserted in the integrand.
f
When
a function of two variables x, y
variables, as defined in
It will be noticed that
37,
we speak
we do
is
continuous with respect to the two
of it as a continuous function of (x, y).

not use the full consequences of this continuity in
the following argument.
169
DEFINITE INTEGRALS CONTAINING
170
We
know that, since 0(#, y) is a continuous function of (x, y)

in the given region, to any positive number e, chosen as small
as we please, there corresponds a positive number Y\ such that
the same
Let
|^(aJ,y+Ay)-^(oJ,y))<e, when Ay
serving for all values of x in (a, a).*
satisfy this condition, and write
rj
Ay
fa'
0(
Ja
/(y + Ay) -/(y)
Then
[0 (a, y + Ay)
["
Ja
(a,
y)]
cfo.
Therefore
<(&'
Thus /(y)
II.
If
=y=
is
</>(x,
a continuous function of (x, y) in a = x = a',

\{s(x) is bounded and integrable in (a, a'), then
$(x, y)\[s(x)dx is
fa'
a
fa'
a
The
tions
same notation as
/(y + Ay)-/(y)=
[<t>(x,y
in
I.,
+ by)-<l>
\[s(x)
in (a,
a').
\f(y + by)-f(y)\<M(a'-a)e, when
Thus f(y)
is
continuous in
(6, 6').
Differentiation of the Integral.

fa'
a
tion of (x y) in
}
T
Ja
M be the upper bound of
Then
the
two integrable func-
integrable.
Also, with the
78.
(6, &').
<j>(x,y)\/s(x)dx.
integral exists, since the product of
is
Let
(b, &').
y) is
and
b',
when
0e>
(p(x,
y)dx, where
a = x = a,
=y=
(j>(x,
6',
y) is a continuous func-
and
exists
and
satisfies
same condition.
This follows from the theorem on the uniform continuity of a continuous
function
(cf
37, p. 75).
AN ARBITRARY PARAMETER
Then f'(y)
Since
to
any
and
exists
is
is
s
positive
number
e,
Let
>/
Ay
serving for
dx.
Ja
y) in the given region,
(x,
chosen as small as
responds a positive number
the same
to
equal
tj,
we
please, there cor-
such that, with the usual notation,
values of x in
all
171
(a, a').
satisfy this condition.
Then
/(y + Ay)-/(y) =
'
0(a,
Ja
Thus we have
_'
rf
Ay
<(a'
And
this establishes that
a)e,
Lt
exists
Ay
equal to
II.
in
Ja
Let
I., t(/ncZ
^- dx at any point
oy
/(y)=
Then f(y)
find, as
<f>(x,
y)
-^
and --
oy
bounded and integrable in
exists
and
is
to
equal
|
\fs(x)
(a, a).
dx.
Ja ^-\fs(x)
oy
I
in (a, a') be
M.
above, that
'-*, when
And
the result follows.
and
is
b').
Let the upper bound of
Then we
in (6,
$(x,y)\fs(x)dx 'where
Ja
^(cc) is
when
are as
172
The theorems -of
fa'
a
<j>(x,y)
show that
this section
if
we have
F(x, y) dx, where F(x, y)
is
to differ-
of the
form
and these functions
satisfy the conditions

of differentiation under
or (j>(x,y)\ls(x),
named
above, we may put the symbol

the integral sign. In other words, we may reverse the order of
the two limiting operations involved without affecting the result.
we
be noticed that so far
It will
The
integrals.
are dealing with ordinary

is finite, and the function
interval of integration
has no points of infinite discontinuity in the interval.

79.
Integration of the Integral.
fa'
a
<f>(x,
<f>(x,
a^x = a', b = y =
(x, y) in
integrable in (a, a).
function of
and
y)\fs(x)dx, where
Then
Jyo
where y
f(y)dy = J dx \
n
</>(x,
and
a continuous
i/s(x) is
bounded
y)\!s(x)dy,
Jy
y are any two points in
b',
y) is
Jy
(b, b').
y)dy.
0(oj,
3cb
Then we know that
= <f>(x,
y)
easy to show that <fr(x, y)

x a, b y
y) in the region a
and
(x,
^-
it is
Ja'
a
From
78,
we know
is
= =
= =
49],
f
$(x,
that
^
3*
(a'
=
Also g'(y)
is

*
Therefore
Jy
where y
cj(y)
dy = T dy
Jy
J\a
<f>
(x,
(b, b').
(6, b')
y)
by
^ (x) dx,
77.
Thus
dy
173
(f)(x,y)ifs(x)dx
J.rt
7/0
=
\
dx
(V
\
Ja
[y
dx\
Jy
Thus we have shown that we may invert the order

tion with respect to x and y in the repeated integral
dx
Ja
when
F(x, y) dy,
the integrand satisfies the above conditions
ticular, since
function of
= I, when
\[s(x)
we may put
y) in the region with
(x,
of integra-
F(x, y)
is
and in para continuous
which the integral
deals.
In the preceding sections of this chapter the intervals

and (b, b') have been supposed finite, and the integrand
bounded in a=x a, b^.y^-b'. The argument employed does
80.
(a, a')
not apply to infinite integrals.

For example, the infinite integral
f(y)=\Joo
converges
when y = Q, but
f(y) = l when y>0,
and
Similarly
Jo
it
discontinuous at
is
2/
= 0,
since
= 0.
/(0)
~
sin Try e x 8in2 ^
dx
all values of y, but it is discontinuous for every

=
and
positive
negative integral value of y, as well as for y 0.
converges for
tl
Under what
conditions then,
Ja
if
in
may
be asked, will the infinite
/v
F(x, y)
dx
when b = y = b',
And when can we
convergent
(b, b') ?
it
/>
integrals
the sign of integration
and
Ja
F(x, y) dx,
define continuous functions of

differentiate
y
and integrate under
174
In the case of infinite
we have met with
series,
the same
We
questions and partly answered them [cf.
68, 70, 71].
to
discuss
them
for
both
of
infinite
The
proceed
types
integral.
discussion requires the definition of the form of convergence of
which corresponds
infinite integrals
to
uniform convergence in
infinite series.
81.
first
Uniform Convergence of
with the convergent
Ja
where F(x,y)
number
a'
bounded
is
Infinite
Integrals.
We
deal
infinite integral
F(x, y) dx,
in the region
a=x=a',
= y^b', the
being arbitrary.
MO
I.
The integral
F(x, y)
Ja
dx
is
said to converge uniformly
to its
value f(y) in the interval (b, b'), if, any positive number e
having been chosen, as small as we please, there is a positive
number
X such that
l*v
Ja
F(x,y)dx <e. when x = X,
X serving for every y in (b, b').

And, just as in the case of infinite series, we have a useful test
for uniform convergence, corresponding to the general
principle
the same
of convergence
II.
15)
necessary
and
sufficient
convergence of the integral
Ja
F(x,
condition for the uniform
y)dx in
that, if any positive number e has been

we please, there shall be a positive number
I
I
the same
the interval
(b, b') is
chosen, as small as
X such that
F(x,y)dx <e, when x">x' = X,
serving for every y in
The proof that
II.
(b, b').
forms a necessary and
uniform convergence of the integral, as defined in I.,

follows exactly the same lines as the proof in 66 for the corresponding theorem in infinite series.
for the
will be seen that if
it
Further,
Ju
F(x,
175
y)dx converges uni-
formly in (b, b'), to the arbitrary positive number

such that
sponds a positive number
there corre-
dx
F(x, y)
the same
The
<
(b, b').
and theorem given above correspond exactly to
definition
those for the series
(x)
+u
= =
uniformly convergent in a x b
RH (x ) "C e, when n = v
and
|
Rn (x)
the same
82.
<
e,
when n = v,
...
namely,
serving for every value of x in the interval
Uniform Convergence of
We now
(x)+
when x = X,
e,
(a, b).
Infinite Integrals (continued).
consider the convergent infinite integral
F'x y)dx,
y
fa'
a
where the interval (a, a') is finite, but the integrand is not
bounded in the region a = x = a',b = y = b'. This case is more
complex than the preceding, since the points of
infinite discon-
tinuity can be distributed in more or less complicated fashion

over the given region. We shall confine ourselves in our
definition, and in the theorems which follow, to the simplest
is also the most important, where the integrand
has
F(x, y)
points of infinite discontinuity only on certain lines
x = a lt a 2 ... a n (a'^-a l
...
a n =a'), and is bounded in the
2
case,
which
<
<
given region, except in the neighbourhood of these lines.

This condition can be realised in two different ways.
be at isolated points, or they

right along the lines.
infinities
may
E.g.
(i)f
JO
(ii)
The
be distributed
7
>J\
xv- l e- x dx,
Jo
In the
may
when 0=.i/<l.
of these integrals there is a single infinity in the

given region, at the origin in the second, there are infinities
right along the line x = from the origin up to but not including
first
176
In the definitions and theorems which follow there

for
any
distinction
between the two
no need
the convergent integral
first of all,
Consider,
is
cases.
where F(x, y) has points of infinite discontinuity on x = a', and

bounded in a x a' g, b^y b', where a<a'
O'.
For this integral we have the following definition of uniform
= =
is
convergence
fa'
a
F(x, y) dx is said to converge uniformly to
value f(y) in the interval (b, &'), if, any positive number e
having been chosen, as small as we please, there is a positive
number r such that
its
^f (x,y)dx\<e,
1
\f(y)-\
Ja
the same
>;
And, from
(b, b').
this definition, the following test for
vergence can be established as before

II.
necessary
and
any
number
positive
the same
*/
J a' -
for the uniform con-
F(x, y) dx in the interval

e
uniform con-
ta'
a
if
when
(6, b') is that,
has been chosen, as small as we

such that
rj
a positive number
**F(x,y)dx <e, when

'
0<?<? =
*,
(b, b').
we
see that if this infinite integral is uniformly conthere will

in
vergent
(b, b'), to the arbitrary positive number e
that
such
correspond a positive number r\
Also
II
the same
*}
F(x,y)dx <e, when
(b, b).
and the above condition, require obvious modifications when the points of infinite discontinuity lie on x = a
instead of x = a.
The
definition,
And when they
on
lie
lines
x = a^ a2
177
a n in the given
can be broken up into
...
region, by the definition of the integral it

several others in which F(x, y) has points of infinite discon-
tinuity only at one of the limits.

In this case the integral is said to converge uniformly in (b, >'),
when each of these integrals converges uniformly in that interval.
And, as before, if the integrand F(x, y) has points of infinite

=
discontinuity on x a lt a 2 ... a n and we are dealing with the
,
J*
F(x, y) dx, this integral must be broken up into several
Ct
integrals of the preceding type, followed

form discussed in 81.
The
when
integral
now
is
by an integral
of the
said to be uniformly convergent in (6, b')

it has been divided are each
the integrals into which
uniformly convergent in this interval.
The simplest
Tests for Uniform Convergence.
83.
F(x, y) dx, taking the
J*
a
first
test for
type of infinite integral, corresponds to Weierstrass's 3f-test

uniform convergence of infinite series ( 67).
for the
I.
Let F(x,y) be bounded in
tegrable
Then
in
the integral
if there is
(ii)
For,
(iii)
by
and, from
f*
I
J a
ju.(x)
(i)
/m(x),
independent of
when x = a;
y)\ = ]u.(x), when x = a
y,
(b, b').
(b, b'),
such that
and
= y = b'\
/j.(x)dx exists.
and
(iii),
for every y in
in-
= Q,
\F(x,
Ja
a' is arbitrary,
F(x, y) dx will converge uniformly in
a function
(i)
and
where
(a, a'),
a^x=a', b^.y = b' and
(ii),
there
when x"*>x' = a and
is
=y=b
f
,
a positive number X, such that
Cx"
when
x'
These conditions will be satisfied if x n F(x, y) is bounded

when x = a, and b = y' = b' for some constant n greater than 1.
M
c. i
178
COROLLARY.
Let F(x, y) = (f>(x, y)\Js(x\ where
bounded in xâ and b = y = b', and integmble in
(a, a'),
where
dx
is
a'
arbitrary, for every y in
formly convergent in
Ex.
1.
Ex.
2.
e~ xy
-fj^>
Then
be absolutely convergent.
r\fs(x)
-*
e y
Ja
is
(J>(x,y)
the interval
(b, b').
Also
let
F(x, y) dx is uni-
(b, ?/).
dx converge uniformly
dx converges uniformly
in 7/^?/n
0< y^j=.
in
Y,
>0.
where Fis an arbitrary
positive number.
converge
y where n
uniformly for all values of
> 0.
II. Let
(x, y) be bounded inx = a, b = y = &', and a monotonic
function of x for every y in (b,b ). Also let ty(x) be bounded
and not change sign more than a finite number of times in
f
the arbitrary interval (a, a'),*
Then
let
Ja
\^(x)
dx
^(x) dx converges uniformly in
<f>(x,y)
Ja
and
exist.
(b, b').
This follows immediately from the Second Theorem of

Value, which gives, subject to the conditions
named
Mean
above,
rtj.//
fy.ll
x'
<j>(x,
where
But
=
y) \[s(x) dx
satisfies
<f>(x,
y)
is
(f>(x,
y)
J x'
dx + <j>(x", y)
\^(x)
JI
^(x) dx,
< x' = = x".
bounded
in
x = a and
=y=
b' t
(j>(x",
y)
and
converges.
Thus
I
it
follows from the relation
<j>(x,y)\!s(x)dx
\
\fs(x)dx
that
</>(x,
"(*''"
y) \fs(x)dx converges uniformly in
T/ri
Jf
(6, &').
Jtt
* This condition
Mean
is
borrowed from th
enunciation in the Second Theorem of
50.
If a more general proof had been given, a
Value, as proved in
corresponding extension of II. would have been treated here.
It is evident that
*/s(x,
y\
if
"
Ex.
e~ xy
dx,
JO
e~ xy
Ja
be replaced by
may
y)dx converges uniformly
\[r(x,
ja
theorem
in this
^(x)
179
in
(6, b').
-c?.y(<x>0) converge uniformly in
0.
?/
a monotonic function of x for each y in

(b, b'), and tend uniformly to zero as x increases, y being kept
constant.
Also let ^s(x) be bounded and integrable in the arbitrary interval (a, a'), and not change sign more than a finite
Ze
III.
0(&, y) be
number of times in such an
interval.
Then
(/>(x,y)\/s(x)dx is
Ja
let
Further,
bounded in xâ, without converging as x -> oo
\ls(x)dx be
(b,
b).
Mean Value,
be replaced by
This follows at once from the Second Theorem of

as in II.
Also
\js(x,y),ii
will be seen that
it
~V
Ex.
1.
~dx
2.
y<-y
It
e~** cos
xdx
Jo
and
bounded
x = a and b = y = b'*
''
and
s
''.;'
y=
dr both converge uniformly
in ?/^
<
<0.
can be
left to
the reader to enunciate and prove similar
theorems for the second type of
The most
in
may
-00
/
er-^sin.rrf.r,
Jo
Ex.
is
\js(x,y)dx
^(x)
infinite integral
I,
Ja
F(x,y)dx.
useful test for uniform convergence in this case

I. above.
is
that
corresponding to
Ex.
1.
Ex.
2.
x*- l dx,
^^d.v
ap- l e-*dv converge uniformly in
converges uniformly in
0<^^ y
J"
a
F(x, y)
is
integral in
Later
we
We
F(x, y) dx.
bounded in the region a = x = a',
>-
>yi^/
<l.
consider, to begin with, the infinite integral
arbitrary.
shall
Ja
b
F(x, y) dx, where
= y = b\
shall return to the other
a'
form of
which the region contains points of
In Examples
and 2
of
being
infinite
infinite
continuity.
*
now
*x
88 illustrations of this theorem will be found.
dis-
180
{00
a
function of
of the
is
and
(a,
(x,
form
\/s(x) is
F(x,y)dx, where F(x,y)

y)
is either
in a = x = a', b = y = &',
(f>(x,
y)\fs(x), wlieve
and
bounded^
<j>(x,
a'
a continuous
being arbitrary, or
y) is continuous as above,
<>').
/CO
Also
let
Ja
Then f(y)
F(x,y)dx converge uniformly in
(6, V}.
a continuous function of y in (b, ?/).

Let the positive number e be chosen, as small as we please.
Then to e/3 there corresponds a positive number X such that
is
ii:
the same
F(x,y)dx
X serving for every y in
But we have proved

|
F(x, y)
dx
is
when x =
^,
3'
(6, b').
77 that, under the given conditions,
in
continuous in y in
Therefore, for some positive

x
(b, b').
number
rj,
f*
F(x,y + &y)dx-\ F(x,y)dx

Ja
when \Ay ^y.

<^,
&
Also
Thus
f(y)=\
F(x,y)dx+{JX F(x,y)dx.
Ja
f(y + Ay) -f(y) = [
y + Ay) dx
^F(x,
+f
Jx
Also
F(x, y
'
we have
J\ A
fee
^F(x,
+ &y)dx-\
"'
Jx
F(x,y)dx
^(,
Therefore, finally,
<e, when
Thus f(y)
is
continuous in
(b, b').
a
J*
Let F(x, y) satisfy the
F(x, y)dx.
same conditions as in
da;]
F(x,y)dx.
'
'
y)
84.
? dy f F(x, y)dx={ dx P
J
J(l
Jy
Then
181
F(x, y) dy,
J]fe
where y
Let
(b, b').
F(x,y)dx.
f(y)=(
Ja
Then we have shown that under the given conditions f(y)

continuous, and therefore integrable, in (b, b').
Also from 79, for any arbitrary interval (a,
we
be taken as large as
T dx f
Jy
Ja
2/
2/
x),
is
where x can
please,
F(x,
y)dy= f dy
[*
being any two points in
F(x, y)dx,
J't
J.",,
(b, b').
Therefore
[ dx J\*
Ja
provided
3/0
F(x,y)dx = Lt Jfcty J[V(a;,y)<fo,

a
* >QO y
we can show
that the limit on the right-hand exists.
But
y
f
Jy
dy [
Ja
F(x, y)
dx=
Thus we have only
{* dy \ > Jy
u
F(x,y)dx = 0.
Of course we cannot reverse the order

and write this as
of these limiting pro-
cesses
'
for
we have not shown
that this inversion would not alter the
result.
But we are given that

in
F(x, y) dx
is
uniformly convergent
(b, b').
Let the positive number e be chosen, as small as we please.

Then take e/(b' b). To this number there corresponds a positive
number
such that
F(x,y)dx <T7^7, when x = X,

the same
X serving for every y in
(b, b').
182
It follows that
dy\ F(x,y)dx
when x =
y lie in (b, b').

In other words,
if
Lt
[
a:->
And from
LJy
dy\J x F(x,y)dx\=0.
the preceding remarks this establishes our result.

/-CO
Differentiation of the Integral
86.
F(x, y)dx.
Ja
Let F(x, y) be either a continuous function of (x, y) in the

x a', b^y b', a being arbitrary, or be of the form
region a
as above, and \fs(x) is
</>(x,y)\fr(x), where </>(x,y) is continuous
= =
bounded and integrable in
the arbitrary interval (a,
^
F(x, y) have a partial differential coefficient ^
fies the same conditions.
let
Then, if the integral

00
ferential coefficient at every point in
from
(b, b'),
(b, b'),
(b, b').
Let
85,
where y Q y l are any two points in (b,

Let
y = y and y
,
fy+by
y
Therefore
the
84 that, on the assumption named above,
Then, by
and
and
is
satis-
f(y) has a dif-
T^f
dy
Also
dx converges uniformly in
We know
which
F(x, y) dx converges to f(y),
^jf
a').
T7I
g(y) dy
g(
C^
J
b'\
[F(x,
y + ky}- F(x,
y)] dx.
<
Ay =f(y + Ay) -f(y\

foo
a
F(x, y) dx.
183
Thus
But Lt g()=g(y),
since g(x) is continuous.
Ay >0
It follows that f(y) is difFerentiable,
and that
/W-J.**
where y
87.
is
any point
in
(6, 6').
Properties of the Infinite Integral
F(x, y)dx.
The
Jft
84-86 can be readily extended to the second type of

It will be sufficient to state the theorems
results of
infinite integral.
The steps in the argument are in each case

in the preceding discussion.
to
those
As before, the
parallel
region with which we deal is
without proof.
JO,'
{a'
a
F(x, y) dx.
F(x, y) dx, where F(x, y) has points of infinite
x = a i ,a.1 ... a n ) between x = a

and x a', and is either a continuous function of (x, y), or the
product of a continuous function 0(#, y) and a bounded and
integrable function ty(x\ except in the neighbourhood of the
discontinuity on certain lines
said
(e.g.,
lines.
(V
Then, if
F(x,
y)dx
is
P dy
Then
F(x, y) dx = f dx
(*
Ja
I.
F(x, y) dy,
(b, b).
fa'
a
Let f(y) =
is
F(* y) dx.
same conditions as in
where y Q y are any two points in

,
f(y)
(b, b').
fa'
a
Let F(x, y) satisfy the
(b, b'),
F(x, y) dx.
F(x, y) dx, where F(x, y) has points of infinite
184
discontinuity on certain lines (e.g.,x = a l a 2 ... an ) between x = a

and x = a, and is either a continuous function of (x, y), or the
,
product of a continuous function <j>(x, y) and a bounded and

integrable function \js(x), except in the neighbourhood of the
said lines.
Also
let
F(x, y) have a partial differential coefficient^- which

same conditions.
,
satisfies the
fa'
let
Further,
F(x, y) dx
Ja
uniformly in
Then f'(y)
Ja
-fip
^- dx converge
dy
(b, b').
and
exists
is
equal to
dx in
(b, b').
Applications of the preceding Theorems.
88.
^
dv=^.
x
u
ll
Ex.1. To prove
Jo
(i)
|V
and
converge,
Let
e~<^
F(a)=
(aÔ).
^djc
This integral converges uniformly when aiÔ. (Cf.

For e~^\x is a monotonic function of x when .f>0.
Mean
Thus, by the Second Theorem of

rx"
g-ax
Jx'
'
jx'
where
0< x
Value,
e~ a r
e~ox'
83, III.)
bl
X"
"
/''"
Jf
^ ^ ^ x".
Therefore
x e
t
Jx'
^^\\\xdx\^ - X
<4
since
x'
-\-
n~
since a
= 2 for
siujedx
/
\
ainxdx
<-
smxdv
J.,:'
',
all
values of
p and q,
JP
0.
or
It follows that
^ dx
e-**
<
e,
when
x"
>x
^ X,
*'
J x'
> 4/e, and this holds for every a greater than or equal to 0.
provided that
from 84
This establishes the uniform continuity of the integral, and
0.
a
in
F(a) must be continuous,
^(0)= aLt
/
>0 J9
Thus
i.e.
Jo
<-
dx= Lt
n->o Jo
e~
';''/',
Again, the integral
(ii)
is
Thus, by
â >
0.
again an example of
is
83, III.
> 0,
when a
86,
e-
uniformly convergent in a
This follows as above, and
185
=
da
Jo
- "^
dx
.#
(cos
.r
e~ mx \nxdx.
But
(.
\
+ a sin x)=
(a
-x
Therefore
- *
sin
x dx = -
Jo
1
=
a1 + 1
Thus
And
1
^( a ) = - tan- a + 1,
= 0.*
jF(a)
Lt
since
a
It follows
from
that
(i)
dx=-.
Jo
Ex.
To prove
2.
F cos cur
"^
1+^^=2*-*
TT
*
If a formal proof of this
Let the arbitrary positive
e-**
is
e
sinew;
/"*
'
= TT.,,1 -
"
<!
(u
a)
2(
required, we might proceed as follows

be chosen, as small as we please.
such that
rf-t
J(TT^)
.1
J,
Since
.1'
in
aÔ,
there
is
_ aj
J.v
and
this
TJ
But
e~ ax
I
I
sinu;
J*t
where x =^
i-
r
dxê***"
ff
wax dx + e~ mV
aX f-*mnx
7
J*
dx,
^ X.
And we know
that
;
<TT for
c~*
Therefore
I
Thus we can choose
J*n
q>p>0
dx
(cf
p. 202).
< 2ire-*:
so large that
te|<l, when a^^l.

It follows that
-^
r e^^daj^
+
|
^^|
-</
i.
Let
The
is
integral
continuous for
uniformly convergent for every a, so that, by 84, /(a) is

values of a, and we can integrate under the sign of
all
integration (85).
Let
(ii)
(" f(a)da.
<(a)=
Then <(a)
is
a continuous function of a for

j
Also
r dx
<p(a)=
Jo
all
values of a (49).
a cos
l~
I
1
ax ,
da,
+x
j
dx.
(iii)
lies
we know
Again,
that /'(a) will be equal to
Jo
~x
1
if
-T-
within an interval of uniform convergence of this integral.
But
r,
l+A' 2
is
monotonic when x
-,
l+x*
= 0.
1,
Of
x'
and Lt
x -,. x
Thus we have, when x">x'
where
^ = x".
It follows that
P" - X
X'
-^
1+^2
Jx'
sin cue a.*;=
rr-
a(l+^
f
Therefore
IJy
Thus
JO
(iv)
Now
'^jSinaxdx
X"
~t~
is
MM
2
fa*
/
X"
sin.
)Jox-'
_^L_2
1+^
si
uniformly convergent when
aâ
>0, and
187
Lt <(a) = 0,
Therefore
a-*0
and
Also
<f>'(a)
is
continuous in
<x
0,
and
(j>'(0)
= ->
A - B - - = 0.
Therefore
.1=0
It follows that
and
^=-|-
Thus
=
<(<x)
And
/( a ) = <'(a) = |e-
Both these
Ex.
(i)
3.
when u>0.
|(l-e-"),
results obviously hold for a
Gamma
The
To prove T(n)
as well.
r*
Function T(n)
is
when a>0.
N may be and however near zero n^ may

e~xxn
~l
w>0, and
ff~*aP~*dxt
^=n^n >
n
its derivatives.
0,
however large
be.
dx has
to be
examined for convergence
When 0<w<l, the integrand becomes

In this case we break up the integral into
only at the upper limit.
#=0.
e-*tf
-1
n
( e-*x
dx+
Jo
Take
first
-l
infinite at
dx.
ji
the integral
~x
xn - 1 dx.
Jo
When < x < 1
.t-"-
^^ ~
^ g-^'ow
e^x"-1
Therefore
1
,
if
n j^ n Q > 0.
if
n^n^X).
from the theorem which corresponds to

converges uniformly when n
n^ > 0.
83, I. that
It follows
r e~ xn ~
x
Again consider
When
^>1,
e- xxn
Therefore
It follows as
for
above from
dx
xn - l ^x"-\
-l
if
ê~xx N -\
83,
I.
e^.c'
-1
dx
> 0.
0<n^N.
if
that
e~ xx 1l
~l
0<n^N.
Combining these two
results,
when JV^?i^?i >0, however

may be.
we see
that
large ^V
r
/
may
e^ x"'~
]-
be and however near zero n
(>
30
(ii)
To prove T'(n)=
e~*x n 1 log x dx,
Jo
We
know
that
Lt (x log x)
0,
n>0.
when r>0,
x>0
so that the integrand has an infinity at # =
when O<TI<
I.
for positive values of
n only
188
But when 0<#<1,

Therefore
e~
x n - l ^va^~ l
x n - 1 log x\^.xn - 1 \logx |,
n^
n^
if
if
And we
have seen [Ex.
119] that
6, p.
.i**-
log
a;
dx converges when
io<l.
It follows as above that
Also for
e~ xx n
~l
log
e~
x n~ 1 log x dx
is
x dx, we proceed as follows

1
x"-*
N.
xX-\ if < n
:
When
x>
Therefore
e~xxn
~l
log #
^ tf'V -
log .*
lo
since
But/
e~ xx x dx
Therefore
e~
is
jc
g^<l when
convergent.
n~ l
log
.r
dx
uniformly convergent when (X/iIfi/V.
is
,-*
Combining these two

convergent when
zero
n may
We are
we
results,
see that
e~*xn
Jo
N^.n^.n >0 however

t
log x
dx
is
uniformly
large N may be and however near
be.
thus able to state, relying on

x
F(w) = (" e- x
It can
.*
86, 87, that
n~l
log
x dx
for
be shown in the same way that the successive derivatives of T(n)
can be obtained by differentiating under the integral sign.
Ex.
interval of y
(i)
To prove
4. (i)
Since
{e.g.
2y cos x + y'*) dx
log (I
b=y = b')
and (ii)
uniformly convergent for any
deduce that
2
I
2
2y cos .r -f-^ = (y
cos^')
+ sin%,
We
consider
log 2.
l)
m = 0,
x
1,
2, etc.,
and x=ir
and
in the
y=
y=
first
\sir
this expression is positive for
x = nnr and y
It follows that the integrand becomes infinite at

\.
1, and in the other when
one case when
log sin x dx
Jo
values of x, y, unless when

for these values it is zero.
all
to
is
the infinity at # = 0.
As
the integrand is bounded in any strip O^x^. A", where A'<TT, for any
interval of y which does not include y = 1, we have only to examine the integral
/
in the
log
(1
neighbourhood of y=l.
1 +h, where h
a and a
Put y
to be fixed
more
Cx
Since
definitely later.
I
log(l
2y cos x + y
is
some
2
)
dx
positive
number
less
than unity,
it is
we need only
clear that
Take a value
of a(0
discuss the convergence of the integral
<a < 1) such
-^
that
-<1.
2(1
Then
Now
since
COS
be seen that, when h

|
"^
I
ct.
~1
27T-^y
a,
0<l-cos.r:5l -cos.r+
provided that
2(1+ A)
ft)
let
It will
a)
>Q
i>-
>
2(1
189
2(1 \_/\
>
(Xx^ft.
Therefore, under the same conditions,
But the
/x-test
log(l
-cos.r +
2 (l
log(l -cos .r)

|.
shows that the integral

cos
log(l
.?')(/.?'
converges.
It follows that
log
converges uniformly for h
And
therefore
a.
'
'
83, I.]
[Cf.
d.r
*\
log (1
- cos .r +
.'o
2# cos
.r
+ y 2 ) dr
Jo
converges uniformly for any interval (/>, b') of y.

The infinity at .r = 7r can be treated in the same way, and the uniform
convergence of the integral
n
log ( 1
2y cos x + y
dx
Jo
is
thus established for any interval

(ii)
Let
/(?/)
(ft,
log (1
b')
of y.
- 2y cos x +y 2 ) dx.
Jo
We know from
70 that
y(^) = 0,
and
when |y|<l,
f(y) = TT log y
2
,
when y
|
>
1.
But we have just seen that the integral converges uniformly for
interval of y.
It follows from
87, L, that
l= Lt
and
/(-!)= Lt .%)=0.
any
finite
190
But
/'(I )
f log 2 (1 - cos x) dx
Jo
= 2?r log 2
-f-
logsin|(ir
,7T
= 2?r log 2 + 4
"
log sin
/o
.r
cr.
ff
f*
Thus
log sin
From /( - 1) = 0, we
find in the
f*
ls
xdx=\ir log 2.*
same way that
logcosxdx= - |TT log 2,
Jo
a result which, of course, could have been deduced from the preceding.
rco
-oo
The Repeated Integral
89.
dx
It is not easy to deter-
f (x, y) dy.
J a,
mine general conditions under which the equation

r*>
r*>
V
dx /(.r, y)dy=\ dy \ f(.r, y) dx
00
Ja
-'b
Jm
is satisfied.
The problem
infinite series
what similar
is
closely analogous to that of term by term integration of an

infinite limits.
shall discuss only a case some-
We
between
to that in infinite series given in
Let f(x, y} be a continuous function of
(.y,
76.
y) in
xâ. y^b, and
let the
integrals
(i)
respectively, converge
/(*,
(ii)
y)<fcr,
uniformly in the arbitrary intervals
b=y^
A ho
let the
integral
(iii)
f-y
dx
Ja
Then
and
Since
f(x, y}
are equal.
we
and
f(x,y)dy
dx\
Jb
are given that
/
-
interval
Jb
dy
=2: b.
the integrals
Ja
exist
a^xî a.
b',
|
Jb
f(x,y)dx converges uniformly in the arbitrary
dy f /(#, y) dx=
Ja
Ja
fl
b^.y^b', we know from

Jb
dy f f(x, y) d.r
85 that
Ja
dx \
Jb
f(x, y) dy,
when y > b.
It follows that
f ^r/(*,y)e***=
Ja
Jb
Lt
f(x,y)dy,
rdx\'
Jb
y>v>Ja
provided that the limit on the right-hand side

* This
integral
exists.
was obtained otherwise
in Ex. 4, p. 118.
To prove the
191
existence of this limit, it is sufficient to show that to the

number e there corresponds a positive number
such that
arbitrary positive
dy
<
But from the uniform convergence
of
dx
/
I
== b>
we can
Ja
/(#, y)
when
e,
choose the positive
when
Jb
f(x, y)
.................. (1)
b.
dy
uniformly convergent in the arbitrary interval (a, ').

Therefore we can choose the positive number Y so that
Y serving for every

P dx\fy "
we have
the same
Thus
But
it is
in (a, X).
f(x,y)dy
J(l
Jyl
< "|,
when y">y'^Y.
d^J(x,y}dy
'?/'
Therefore from
f*dx['
f
-a
y>
(1)
and
f(x,y)dy +
(2)
dx^f(x,y)dy- f dx^f(x,y)dy.
we have
O
We
............. (2)
clear that
=
I
x^X,
X serving for every y greater than or equal to
Also we are given that

is
^ Y.
X such that
number
\^dx^f(x,y}dy\<^
the same
y" >y'
Jy'
<
when
e,
y" >y'
have thus shown that
f(x,y)dx= Lt
y
= Y.
dxl f(r,y)dy
>-x./a
(3)
.'6
It remains to prove that
dx
fry
I
r*
f(x,
y)dy=j
Let the limit on the left-hand side be
/*
dx
|
/(#, y) dy.
I.
Then
being any positive number, as small as

positive number Yl such that
I-
Ja
we
f(x,y)dy
dx\
Jb
(4)
Also, from the uniform convergence of

/
Ja
dx \
Jb
/(.r, ?/)
dy,
please, there is a
when y^b,
192
we know that there
is
a positive
'
('/(*,
the same
y}
d>/
number
X such
that
y} dy
:|,
Ijdx l'f(x,
when X'^X,
serving for every y greater than or equal to

Choose a number X' such that X'^.X >a.
Then, from the uniform convergence of

interval,
we know
that there
is
f(x,y)dy in any arbitrary
C
:
3(Z' -a)'
the same
F2
serving for every
that
wheny = r*
in (a, X').
Thus
Now
6.
number F2 such
a positive
f(x\y)dy\<* when
take a
Equations
number
(4), (5)
Y greater than F
and
and
(6)
(5)
hold for this
y>Y
(6)
F,.
number
F.
But
f(*,y)<fy\
5 + l+5
from
(4), (5)
f*'dxf*f(x
and
y) dy
j*dx[f(x
y) dy
(6),
<.
This result holds for every number ^Y' greater than or equal to X.
Thus we have shown that
Also,
1= Lt
x>x
from (3), we have
a
under the conditions stated in the theorem.

It must be noticed that the conditions we have taken are sufficient^ but not
For a more complete discussion of the conditions under which the
necessary.
integrals
when they both

de
la
exist, are equal, reference
Vallee Poussin,* to
whom
should be made to the works of
the above treatment
is
His investigations are contained in three memoirs, the
due.
Ann.
and the third in
first in Bruxelles,
Soc. sclent., 17; the second in /. math., Paris, (Ser. 4) 8, 1892;

J. math., Paris, (Ser. 5) 5.
See also Brornwich, London, Proc. Math. Soc. (Ser.
A valuable dis-
2), 1, 1903.
cussion of the whole subject
193
also given in Pierpont's Theory of Functions

is dealt with in Hobson's Theory of
is
The question
of a Real Variable.
Fa actions of a Real
but from a more
Variable,
difficult standpoint.
REFERENCES.
III.
BROMWICH, loc. cit., App.

DE LA VALISE POUSSIN,
e
T. II. (2 ed.), Ch. II.
loc. cit.,
DINI, Lezioni di Analisi Infinitesimale, T. II.
Parte, Cap. IX., X., Pisa,
1909.
GOURSAT,
OSGOOD,
loc. cit.,
Tl.
I.,
Vol.
Bd.
loc. cit.,
e
(3 6d.), Ch. IV.-V.
I.
Bd.
loc. cit.,
PIERPONT,
STOLZ,
T.
loc. cit.,
I.,
III.
Kap.
Ch. XIII.-XV.
I.,
Absch. X.
I.,
And
BRUNEL,
" Bestimmte
Integrate,"
Em.
d.
math. Wiss., Bd.
II., Tl. I.,
Leipzig,
1899.
EXAMPLES ON CHAPTER
VI.
r*
1.
Prove that
e~
nsti
dx
is
-jQ7"a
2.
ig
Prove that
e~ xy
J9
e~ ax
that
''
---dx
is
dx
is
aÔ, when
aâ
>0, and that
>0.
y=v
yÔ, when
>0.
>0, and
/*
3.
Prove that
e~ ax .^~ l cos
dx
is
>'o
a^
>0, when n^.1, and in the interval

Prove that
4.
aâ
e~
ax
^~
sin
x dx
>0, when n^.0, and
Using the
in the interval
fact that
JQ
-dx
l
I
dx
which does not include y =

6.
Show
that
(i)
[o
(1
is
is
and
/=yo<0
?
,
show that
is
'
5.
aÔ, when Q<n<l.
a^.0, when
y^-y
a.
- e~*)
^' dx,
- COB fear)
c. i
<0,
uniformly convergent in any interval of y
(ii)
are uniformly convergent
l<w<l.
foryÔ.
N
^.
194
Discuss the uniform convergence of the integrals
7.
?L^dx.
(iv)
(v)
log x
Jo
8. Show that differentiation under the integral sign is allowable in the

following integrals, and hence obtain the results that are given opposite each
:
dx
9.
Establish the right to integrate under the integral sign in the following
integrals
(i)
(ii)
e~ ax cos
(iii)
e~'
(iv) I
10.
e~ax dx', interval
t:t
b.v
dx
sin b.vd.r
xa d.v
Assuming that
>0.
interval
f(
interval
a^
e~ ax sin
^ >bxdx=
Deduce that
(i)
'
=a
interval a
â
>0, or any interval
of
b.
>0, or any interval
of
6.
1.
>0, show that
T Llil^si,
^
Jo
(ii)
11.
Show
=
|T^cfo ^,
b>0.
that the integrals
/
/
.o
t>"
.'o
can be differentiated under the integral sign, either with regard to a or

and hence obtain the values of
-X
-00
I
xe~"x cos bx o?.r,
.re~
n;r
Jo
Jo
Jo
^-""cosbxdv,
Jo
x-e-
sin
bx dx,
6,
ff
r
Let
12.
Show
/(/)=
that /'(?/) =
sin
195
.*;?/
e~*
Jo
dx
for all values of y,
and deduce that
Let
13.
Show
On
From
-21
that f(y)=
integrating
.re"* sin
show
this result,
Also show that
that/(?/)
r/a:/
= | N/7re~
e-*
8in
U=
I*
^ =^
rfar
Jo
Let
assuming that
"
14.
6~*GQ62sydy=* [f(y)dy
and deduce that
?/.
found that
parts, it will be
by
for all values of
%xydx
V=
e-^x"- 1 cos b.v dx,
I" e~^di/.
Jo
e~ axxn
^ sin bx dx,
where a>0, n>0.
Make
the following substitutions
Then show that
From
these
it
= rsii\6,
follows that
~'lW
u = T(n)coan6,
Thus
U=T(n)
if
= r(w)sinw0.
F=r(
0<?Kl, Lt U= .'0
(****
a *0
Lt
aÔ
And
~
-
-r^
Deduce that
Also show that,
where -\ir
r= /"V^si
J%
deduce that
mr
[Compare Gibson,
15.
Treatise on the Calculus, p. 471.]
Prove that
r*>
where
is
rx
/*>
cte/
|o
e-*y
*inxdy=\ dy
any positive number.
/-x
j
e~ xy sin.r^-,
CHAPTEE
VII
FOURIER'S SERIES
90.
We
Trigonometrical Series and Fourier's Series.
have
already discussed some of the properties of infinite series whose

terms are functions of x, confining our attention chiefly to those
whose terms are continuous functions.

The trigonometrical series,
where a
a lt b v etc., are constants, is a special type of such series.

Let f(x) be given in the interval ( TT, TT). If bounded, let it
if unbounded, let the infinite
be integrable in this interval
,
integral
f(x)
dx be
J -77*
(f(x)
cos nx' dx
f(x') smnx' dx'
and
J -77*
-77
exist for all values of n.
The trigonometrical
the coefficients d a 1?
,
a =
Then
(C1,
VI.)
series (1) is called a

?>
lf
etc.,
are given
Fourier s
Series,
when
by
f(x) cos nx' dx,

ZTTJ-,,.'
If
an =
TTj-Tr
and these
77
f(x') cos nx' dx'
coefficients
are
bn
77
-1 f
f(x') sin nx' dx',
7TJ-T7*
called
/9\
Fourier's
Constants for the
function f(x).
The important thing about the

f(x)
the
Fourier's Series
sum
is
that,
when
very general conditions in the interval ( TT, TT),

of this series is equal to f(x), or in special cases to
satisfies
0)],
when x
lies in this interval.
196
FOURIER'S SERIES
we assume
If
interval
that the arbitrary function f(x), given in the

can be expanded in a trigonometrical series of
TT, TT),
197
(1), and that the series may be integrated term by

term after multiplying both sides by cos nx or sin nx, we obtain
the form
these values for the coefficients.
For, multiply both sides of the equation
f(x)
by
= aQ + (a
cos
x + frj
sin x) -f (a 2 cos
and integrate from
cos nx,
Then
mx cos nx dx =
cos
when
?)i, ?i
mx cos nx dx = 0,
sin
J -TT
. . .
TT.
nx dx = TTU n
f(x) cos
since
to
TT
2x + 6 2 sin 2x) -f
n-
and
are different integers,

cos 2 nx
dx = TT.
Thus we have
If"
a n=
And
7TJ
/(x^cosnxdx, when
?i
= l.
.jr
similarly,
^,i
=TTj
_//(#')
sin 7m-
Inserting these
alucs in the series (3), the result
f(x)
=J
^~J
/(.,')
+ 7T- S f
dx
-ir
may be written
os ^(^'
/(') cos
^'
^'.
-fl-
-TTâ^TT ....... (4)

This
is
the Four'n
r's N' ries
If the arbitrary function,
in other words, if
for /(a
given in
f(x)=f(
/'(./') r/,/j' -f-
TrJo'
x V]
cos iwc
TT, TT),
is
if
CTT
JV
I
f(x') cos
nxdx,
OÔ^TT ....... (5)
it is an odd function
i.e. if /(#)=
f( x)
the Fourier's Series becomes the Sine Series:
- v
the Fourier's
Jo
.
Again,
an even function^
%
"'
1 (V
--
).
when 0<>;<7r
.c)
Series becomes the Cosine Series
/(^) =
,c)
77
sin
='2^
^
i
when
??./
/'(.r')
Jo*
sin nx'dx',
TT ........
(6)
FOURIER'S SERIES
198
The expansions in (5) and (6) could have been obtained in the
same way as the expansion in (3) by assuming a series in cosines
only, or a series in sines only, and multiplying by cos me or
to TT.
sin nx, as the case may be, integrating now from
we take
if
Further,
the interval
1,
instead of
1)
TT,
find the following expansions, corresponding to (4), (5)
f(x) = 7
ri
I
f(x) dx
(sJo
TT),
and
Jo
f(x') cos
'-
m-
x'dx,
OSzSl
y\
O^
>
y /j
ci
+ yi ^] cos '-j- x
we
(6)
(8)
r^
ci r~i
Dili
^y*
,L/
I
I
Tf'Y '^ olil

QITI
î// ^
1
/y'/"//^'
i//
(AjJU
\J
"^^^
~^~~
/^
i//
^Q\
\"/
7
(/
However, this method does not give a rigorous proof of these

very important expansions for the following reasons
:
(i)
We
have assumed the possibility of the expansion of the
function in the series.
We
have integrated the
term by term.
if the convergence of
the series were uniform, since multiplying right through
by cosnx or sinnx does not affect the uniformity; but
this property has not been proved, and indeed is not
generally applicable to the whole interval in these ex-
(ii)
series
This would have been allowable
pansions.
(iii)
The
discussion does not give us
any information
as to the
behaviour of the series at points of discontinuity, if such

arise, nor does it give any suggestion as to the conditions
to
which f(x) must be subject
Another method of obtaining the
may
if
the expansion
coefficients,
is
to hold.
due to Lagrange,*
be illustrated by the case of the Sine Series.
Consider the curve
y = <LI sin x + a 2 sin 2# +
We
+ a n _j sin (n
I ) x.
can obtain the values of the coefficients

1,
. . .
Lagrange, (Eurrcs, T.
I., p.
<!*,,
-..
",i-l,
533; Byerly.
F<mrii:r'* Series, etc., p. 30.
FOURIER'S SERIES
so that this curve shall pass
199
through the points of the curve
'y =/(*),
at which the abscissae are
27T
7T
._,
-,
n
way we
In this
find
a lt
to the limit as ?i->oo,
a.7 ,
...
...
-1
7T
(n-I)-.
n
a n-1
we have
as functions of n.
Proceeding
the values of the coefficients in
the infinite series

f(x)
But
=a
sin
x + a 2 sin 2x -f
from a finite number of equations to an

requires more complete examination before the
this passage
infinite
number
results can be accepted.
satisfactory method of discussing the possibility o

an
arbitrary function f(x), given in the interval
expressing
The most
TT, TT)
by the corresponding
a
+ (04 cos x +
>!
sin x)
Fourier's Series,
+ (a
cos 2x
+b
is
to take the series
cos 2x)
. . .
where the constants have the values given in (2), and sum the
terms up to (an cosnx + b n smnx) then to find the limit of this
sum, if it has a limit, as ??,->x
In this way we shall show that when f(x) satisfies very general
;
conditions, the Fourier's Series for f(x) converges to f(x) at every

that it converges to
TT, TT), where f(x) is continuous
point in (
;
i[/(#-f 0)-H/(-~0)] at every point of ordinary discontinuity;

also that it converges to t[/("- T+0)-4'/(x ())] at &=7r,
<
when
these limits exist.
Since the series

is
periodic in x with period 2?r, when the

TT), it is also known for every value of x.
is
known
in
If it is
more convenient
defined as
x,
(0, 2?r),
sum
to take the interval in which f(x) is

the values of the coefficients in the correspond-
ing expansion would be
an =
i r 27r
'
7T J o
It
f(x) cos nx'dx',
bn
=1
1*-*
7T J o
f(x) sin nx'dx,
n^l.
need hardly be added that the function f(x) can have
different analytical exp-essions in different parts of the given

interval.
And in particular we can obtain any number of such
FOURIER'S SERIES
200
expansions which will hold in the interval (0, TT), since we can
give f(x) any value we please, subject to the general conditions
we
shall establish, in the interval
0)
TT,
The following
discussion of the possibility of the expansion

of an arbitrary function in the corresponding Fourier's Series
depends upon a modified form of the integrals by means of which
Dirichlet * gave the first rigorous proof that, for a large class of
With the help
functions, the Fourier's Series converges to fix).
of the Second
Theorem
of
Mean Value
sum
the
of the series can
be deduced at once from these integrals, which
we
shall call
Dirichlet's Integrals.
Dirichlet's Integrals (First Form).
91.
rt
Lt
ft,
>C# J
sin
TT
/ULX
rt v
+ 0),
cfo=o/(
W
J(x)- *
U.
Lt
,,
><Jft
sin
si
f(x)-
where
When we
apply the Second Theorem of Mean Value to the

si n x ,
f
dx,
0<0<<e,
x
h'
'
integral
we
see that
b'
'
f*
sin
b ib
h'
where
-dx = r
x
Casino?
x dx + .
'
c
.
sin
x dx,
<>
= = c.
c>
Thus
I
sin x
b'
It follows that the integral
smx dx
Jo
is
convergent.
The
Its value has
been found in
^&
sil
integrals
f"
Jo
and
can be transformed, by putting

sin x 7
W.
Ja
/mx
88 to be
x, into
sin
ir
dx,
respectively.
*J. Math., Berlin,
p. 152, 1S37.
4, p. 157, 1829,
and Dove's i>pertorium dcr
Pliyxik, Bel. I.,
FOURIER'S SERIES
201
It follows that
a
f
r
Lt
,
sin /mx
-
M ôoJo
r
Lt
,
and
jj.>
sin
/ULX
Ja
7
=
dx
sin
" M' 1
*^
T
Lt
dx=
JO
co
sin
dx=\
M~*" ^
These results are special cases of the theorem that, wlien f(x)
satisfies certain conditions, given beloiv,
Lt
In the discussion of this theorem
we
shall, tirst of all,
assume
that f(x) satisfies the conditions we have imposed upon (j>(x) in

the statement of the Second Theorem of Mean Value ( 50) viz.,
;
bounded and monotonic (and therefore integrable)

the interval with which we are concerned.
it is
to be
01 Y\
It is clear that -
////*
- satisfies
in that theorem.
^(x)
I.
shall
the conditions imposed upon
bounded and integrable, and does
It is
not change sign more than a
We
in
finite
remove some of the
number of times
restrictions placed
in the interval.
upon f(x)
later.
Consider the integral
From
,,
the Second Theorem of

,
sin /mx
f(x)where
is
/(6-0)
-dx=f(a + 0)\Ja
some
Since f(x)
is
Mean Value
sin /mx
b
,
_^+/(6-0)
Jf
-dx,
= =
/(ct-f 0)
and
exist.
And we have seen that the

hand are zero as /* -> oo
limits of the integrals
on the right-
It follows that,
under the conditions named above,

S1
O
II.
JJLX
x in a = x b.
monotonic in a x
b, the limits
definite value of
sin
~^d
= 0, when 0<a<6.
FOURIER'S SERIES
202
Put
/(a?)
inoiiotonic
Then
= 0(aj)+/( + 0). The

in Q^xâ.
+ 0)
limit /(
exists, since f(x) is
monotonic and
</>(#) is
Also
a
s,
sin
We
i*/
As
yu
:r-<fo=/( + 0)
/<
/x->oo
shall
the
first
now show
integral
sin
u#
Jo
Jo
limit
\ir.
that the second integral has the limit zero.

is sufficient to show that, to the
arbitrary
number
positive
when
f
Jo
such that
v.
JO
FQ
Let us break up the interval

(a, a), where a is chosen so that
We
on the right-hand has the
To prove this, it
positive number e, there corresponds a
there
tt
f
--=cfa+
^
f"
(0,
two
a) into
parts, (0, a)
and
this, since we are given that 0( + 0) = 0, and thus

a positive number a such that
can do
is
0(a?)
1< e/2w, when 0<>â.
Then, by the Second Theorem of

a
f
JQ
sin juix
<}>(x)- X
.
Mean
a
dx = (j>(a
-,
0)
Value,
sin
/u.x
Jg
dx,
+ 0) = 0,
^ being some definite value of x in

sinx
icÔ,
But, in the curve
y=
since 0(
= x = a.
tV
the successive waves have the same breadth and diminishing

and TT is greater than that
amplitude, and the area between
between TT and 2?r in absolute value that between TT and 2?r is

greater than that between 2?r and 3-rr, and so on since sin x goes
through the same set of values in each case, and 1/x diminishes
:
as x increases.
rp,
Thus
I**
sin
Jo
whatever positive value x

q
( sin x ^
Also
Jp
and each
than
TT
0<p<(/.
dx=
(* sin
Jo
may
ax =
of the integrals
for
x
- j
(
\
dx<
TT,
have.
sin
a;
%i '
7
ax
Jo
sin
on the right-hand
.x
ax,
is
positive
and
less
FOURIER'S SERIES
Therefore
I
It follows that
dx
Jp
<
TT,
203
when
=p <
q.
-dx
d>(x)
and this is independent of /x.

But we have seen in I. that
Lt
Therefore, to the arbitrary positive
a positive number v such that
a
/-xSin/,
Ja
number
dx
00*0- *
e/2,
when
|,
there corresponds
/x
t/.
Also
If"
i
Jo
sin
<(#)
JU.X
rta;
"
a
f
(j>(x)
Jo
si
7
dx
Therefore
when n = v.
Thus
Lt
>&>
fJL-.
JO
And, Hnally, under the conditions named above,
In the preceding section we have assumed that f(x) is

bounded and monotonic in the intervals (0, a) and (a, b). We
92.
shall
now show
that these restrictions
may
be somewhat relaxed.
Dirichlet's Integrals still hold when f(x) is bounded, and

the interval of integration can be broken up into a finite number
I.
of open partial intervals, in each of which f(x) is monotonic*

This follows at once from the fact that under these conditions
we may
*
write
This condition
f( x )
is
= F(x)-G(x),
sometimes, but
shall have only a finite
number
of
less exactly,
expressed by the phrase

in the interval.
maxima and minima
f(x)
FOURIER'S SERIES
204
where F(x) and G(x) are

creasing in the interval
of
positive, bounded, and monotonic inwith which we are concerned [cf. 36].
This result can be obtained, as follows, without the use of the theorem
36
:
Let the interval
(0,
a) be
(0,
broken up into the n open intervals,

(a 1?
!>,
rt
2 ),
...
(_!,
),
which f(x) is bounded and monotonic.

Then, writing a =0 and a n = a, we have
in each of
The first integral in this

the limit zero when //, =>x
sum has
the limit -/(+0), and the others have
It follows that,
under the given conditions,
Lt
The proof
that,
under the same conditions,
Lt
is
dx= 0, 0<a<6,
/(#)
practically contained in the above.

It will be seen that we have used the condition that the
intervals
is
is finite,
equal to the
II.
as
sum
we have
relied
number of
partial
upon the theorem that the limit of a
sum
of the limits.
The integrals
hold for certain cases where a finite
still
number of points
in
of infinite discontinuity of f(x) (as defined

33) occur in the interval of integration.
We shall suppose that, when arbitrarily small neighbourhoods

of these points of infinite discontinuity are excluded, the remainder of the interval of integration can be broken up into a
finite
number of open partial
bounded and monotonic.

Further, ive shall assume
intervals, in each of
which f(x)
is
that the infinite integral
f( x )
"'' '*
absolutely convergent in the interval of integration, and that

x = is not a point of infinite discontinuity.
may take first the case when an infinite discontinuity
We
sin Hoc
'
Jb
only tnere.
Since
know
(a, b).
we
that
And
are given that

I
f(x)
Ja
dx
f(x) dx
also
is
and
dx,
f(x)
absolutely convergent,
converges, for
this convergence is uniform.
sm/x#
x
we
FOURIER'S SERIES
205
To the arbitrary positive number e there corresponds a

number which we take less than (b a), such that
positive
rj,
'6
'6-
and the same

But
,
serves for
r\
.sin/uix
P"
I.
.sm/uiX
/(*)
'/-/
,,
'6-7;
above,
6
TJ.
Lt
It follows that there is

6-1,
'^/
sm/xo;
and
(3),
-dx
f(x)~
(1), (2)
(fossO.
a positive number
\ OllJ. AJU>^
//
xSin/ucc
/()
Ja
>oo
/i
From
/x.
f-^+1
/(^-^<&=JJo
And, by
values of
all
we have
<|,
such that
when n = *
(3)
at once
sn IULX
u:
<
Thus we have shown
Lt
/m->
Ja
that,
e,
when
/*
i/.
with the conditions described above,
wdÔ,
&
0<a<6.
argument applies to the case when an infinite disoccurs

at the lower limit a of the integral, and only
continuity
similar
there.
When
there is an infinite discontinuity at a and at

the
result follows from these two, since
there,
When
b,
and only
discontinuity occurs between a and b we

proceed in the same way; and as we have assumed that the
number of points of infinite discontinuity is finite, we can break
an
infinite
up the given interval into a definite number of partial

which we can apply the results just obtained.
Thus, under the conditions stated above in II.,
to
I*'H/()- -dx = 0, when 0<a<b.

fx->
intervals,
FOURIER'S SERIES
206
Further,
we have assumed
that x =
is
not a point of infinite
Thus the interval (0, a) can !><

discontinuity of f(x).
into two intervals, (0, a) and (a, a), where f(x) is
(0, a),
and
satisfies
the conditions given in
I.
broken up
bounded in
of this section in
(0, a).
It follows that
and we have just shown that

/u.->
Therefore, under the conditions stated above in
II.,
Lt
>OD
/J.
we
The
Dirichlet's Conditions.
93.
tained in
results
which we have ob-
91, 92 can be conveniently expressed in terms of what
shall call Dirichlet's Conditions.
function f(x) will be said to satisfy Dirichlet's Conditions
an interval (a, b), in which it is defined, when it is subject to

one of the two following conditions
(i) f(x) is bounded in (a, b), and the interval can be broken
up into a finite number of open partial intervals, in each
of ivhich f(x) is monotonic.
in
(ii)
f(x) has a finite
in
number of points of infinite discontinuity

when arbitrarily small neighbour-
the interval, but,
is bounded in
remainder of the interval, and this can be broken up
into a finite number of open partial intervals, in each of
which f(x) is monotonic. Further, the infinite integral
hoods of these points are excluded, f(x)

the
fba f(x)
dx
We may now
Whenf(x)
an'd
(a, b)
is to be absolutely convergent.
say that
satisfies Dirichlet's
respectively,
U
U
jU.->X
and
u.
("
JO
> cc J a
where
Conditions in the intervals
<-<&, andf( + Q)
exists,
(0,
then
a)
FOURIER'S SERIES
207
from the properties of monotonic functions (cf. 34)

the -points, if any, where f(x) becomes infinite, or
at
that except
oscillates infinitely, a function which satisfies Diriehlet's ConIt follows
can only have ordinary discontinuities.*

But we have not assumed that the function f(x) shall have
ditions, as defined above,
only a finite number of ordinary discontinuities. A bounded

function which is monotonic in an open interval can have
an
infinite
[cf.
34].
number
of ordinary discontinuities in that interval
Perhaps it should be added that the conditions which Dirichlet

himself imposed upon the function f(x) in a given interval (a, b)
were not so general as those to which we have given the name
Dirichlet's
Conditions
He
contemplated at
functions, continuous, except at a finite

continuities,
and with only a
finite
first
only bounded
number of ordinary disnumber of maxima and
Later he extended his results to the case in which
minima.
there are a finite
number
of points of infinite discontinuity in the
interval, provided that the infinite integral
p
f(x)
dx
is
absolutely
convergent.
somewhat
If the
a function of bounded variation, due to

introduced, the statement of Dirichlet's Conditions can be simpliBut, at least in this place, it seems unadvisable to complicate the
Jordan,
fied.
discussion
94.
difficult idea of
is
by further reference
to this class of function.
Dirichlet's Integrals (Second Form).
Lt
ju.->oo
where
In the discussion of Fourier's Series the integrals which we
meet are slightly different from Dirichlet's Integrals, the
properties of which we have just established.
shall
These conditions can be further extended so as -to include a finite number

points of oscillatory discontinuity in the neighbourhood of which the
function is bounded (_e.<j. sinl/(x--c) at x = c], or of continuity, with an infinite
of
number of maxima and minima in their neighbourhood [e.g. (x-c)sinl/(x-c)

at x = c],
This generalisation would also apply to the sections in which Dirichlet's
Conditions are employed.
FOURIER'S SERIES
208
The second type of integral and this is the one which Dirichlet
himself used in his classical treatment of Fourier's Series is
a
,.
/()
Jo
sin
sin
-,
fj,x
dx,
sin
^x
^
J/f(x) smx
'
a;
dx,
where
We
When
in
now prove
shall
that
f(x) satisfies Dirichlet' s Conditions (as defined
the intervals (0, a)
and f( + 0)
exists,
and
(a, b)
in 93)
where
0<a</;<7r,
respectively,
then
Lt
/x->oo
f
Jo
and
Let us suppose that f(x) satisfies the first of the two conditions
given in 93 as Dirichlet's Conditions
f(x) is bounded, and the intervals (0, a) and (a, b) can be
broken up into a finite number of open partial intervals, in each
:
which f(x)
of
Then, by
is
36,
monotonic.
we can
write
f(x)=F(x)-G(x),
where F(x), G(x) are positive, bounded and monotonic increasing
in the interval with which we are concerned.
s
X
H*
= F(x) -A- - <?(*)' -^
Then
/(*) sm x
sm x
sm x J x
L
But xlsmx
(0,
a) or (a,
b),
is
bounded, positive and monotonic increasing in
when 0<a<6<7r.*
/y
/y
Thus F(x)~.
and G(x) sin x
sin x
and monotonic increasing in the
will
both be bounded, positive
interval
(0,
a) or (a,
b),
provided that
91 that
It follows from
case
may be,
=!/(+<>),
and
Lt ["/()
x> J a
iu.-
=
^^dx
Sin
We assign to z/sinx
0,
when
3?
the value
at
x = 0.
0<<6<7r.
as the
FOURlfeR'S SERIES
let
We
the second of the conditions given in
let f(x) satisfy
Next,
and
/( +
93,
0) exist.
can prove, just as in
92, II., that
=
f(x)dx
S1UX
Lt
dx
fba f(x)
that
when
'
ôoJa
we know
209
absolutely convergent, and
is
x/smx
is
P f(x)
It follows that
'
)a
in (a,
b).
dx
smx
is
absolutely convergent; and the preceding proof [ 92, II.] applies
to the neighbourhood of the point, or points, of infinite discon/y
tinuity,
when we
write f(x)
sin
Lt
Also, for the case
we need
and
P"
_^
Jo
in place of f(x).
'
dx,
f(x)^^
sin
a;
only, as before, break
where f(x)
(a, a),
we have already
is
up the interval (0, a) into (0, a)

bounded in (0, a), and from the results
obtained in this section the limit
found as
is
stated.
If it is desired to obtain the second
the use of the theorem of

Let/(.r) be positive,
(i)
Then
But, by
may
proceed as follows
bounded and monotonic increasing
so also,
is
sm.r
Dirichlet's Integrals without
form of
36, the reader
and
<h(x) = /x
f(x)-.
'
is
sin.r
so also,
in (0, a)
and (a,
b).
0<<6<7r.
91,
Lt
M
>ao
Therefore
ra
Lt
n
-ao
Jon
Also
C
Ja
~.
^ dx=
sin ujc
f(x)
Therefore
sin
Jo
f.
f(v)
y
^ dx-
sin LLK
.
sm x
Lt
Sin
(ii)
f
I
Jo
gi n
M^' j
*/
\
dx
f(x).
sm x
positive, bounded and monotonic decreasing.

Then for some value of c the function c -f(x) is positive, bounded
and monotonic increasing.
Let/(^) be
c. i
FOURIER'S SERIES
210
Also
a
r
I
M 8 " 1 /*^
CIX = C
j-i
[X]
sin
Jo
Using
(iii)
If f(x) is
Jo
s\n ax ,
u-3?
i*
...
SIII.T
.sin//./'
'
X]
(I,''.
sin.'
Jo
the result follows,
(i),
bounded and monotonic increasing, but not
positive all the
by adding a constant we can make it positive, and proceed as

and a similar remark applies to the case of the monotonic
(ii)
time,
in
decreasing function,
(iv)
When
is
f(x)
number
follows from
(v)
And
bounded and the interval can be broken up into a finite

which it is monotonic, the result
of open partial intervals in

(i)-(iii).
f(x) has a finite number of points of infinite discontinuity, as

stated in the second of Dirichlet's Conditions, so far as these points
if
are concerned the proof
is
similar to that given above.
Proof of the Convergence of Fourier's Series. In the

opening sections of this chapter we have given the usual
elementary, but quite incomplete, argument, by means of which
95.
the coefficients in the expansion
f(x) = a
+ (a
cos x
+ ^ sin x) + (a
cos % x
+b
sin 2x)
. .
are obtained.
We now
return to this question, which
we approach
in quite
a different way.
We take the Fourier's Series
a
where the
an
+ (a
cos
# + ^ sin x) + (a 2 cos 2x + b.2 sin 2x) +
coefficients are
If
~
f(x')cosnxdx',
find the
bn
We
shall
TJ^TJ-^
this
sum
point in
when
n
I
f(x'}^mnxdx'.
J
of the terms of this series
sinnx, and we then examine whether
73-J-7T
sum
given by
77
7TJ-7T
We
this
up
to
cosnx and
sum has a
limit as
prove that, when f(x) is given in the interval

satisfies Dirichlet's Conditions in that interval,
and
has a limit as n-^cc. It is equal to f(x) at any

7r<^x<^7r, where f(x) is continuous; and to
there is
an ordinary discontinuity
at the point;
and
to
FOURIER'S SERIES
7r
and
TT
f(
+ O)
0)]
at
X=TT,
211
when
the
limits /(TT
0)
exist.
Let
Sn (x) = a +
(ttj_
cos
re
+ &! sin
n cos wcc
+ 6n sin TI#),
where a a 1? & 1? etc., have the values given above.

Then we find, without difficulty, that
,
8n (x) =
^7T
J*
J(&) [1+ 2 cos (a;'
- a;) +
+ 2 cos ?i
(a?'
- x)] da?'
7T
-x)
If'
sm J(x -.T
2x/-;//(^flEi^l}
sm
o?
Thus
*
sin
sn
on changing the variable by the substitutions x' x= T2a.

If
7r<^x<^7r and /(#) satisfies Dirichlet's Conditions in the
interval
TT,
TT),
/(xH=2a) considered as functions of a in the
integrals of (1) satisfy Dirichlet's Conditions in the intervals

and (0, J-TT ^x) respectively, and these functions of
(0, \-7r-\-\x)
a have limits as^t->0, provided that at the point x with which

we are concerned f(x + Q) and f(x 0) exist.
from 94 that, when x

and
+ 0) both exist,
f(x
f(x-0)
It follows
Lt
lies
between
?r
and
?r
and
&*=
/() at a point where f(x) is continuous.

TT.
have yet to examine the cases x =
In finding the sum of the series for x = TT, we must insert this
giving the value
We
value for x in
Thus
8n (x)
before proceeding to the limit.
f /(* - 2)
si
TT J o
since the second integral in (1)
is zero.
FOURIER'S SERIES
212
It follows that
^//
/(-TT
,sin(2
>
2a)
TT J o
aa
sin
si
sma
where
any number between
is
and
TT.
We
can apply the theorem of 94 to these integrals, if f(x)

satisfies Dirichlet's Conditions in (
TT, TT), and the limits f(ir
0),
/(-7T + 0) exist.
Thus we have
A
x=
S7r =
Lt
similar discussion gives the
which
TT,
is
same value
sum
for the
at
otherwise obvious since the series has a period
27T.
Thus we have shown that when

the arbitrary function f(x)

Conditions in the interval ( TT, TT), and
= If*
a,ft
7T J -ir
/(#') cos nx' dx,
bn
77
=1
7T J _7T
f(x') sin
n x' dx,
Fourier s Series
j
cos x
+ 6j sin
ic)
+ (a., cos 2# + &
sin
2^c)
+ 0) +/( - 0)]
7r<C^<7r where f(x + Q) and
i [/(
at every point in
and at
TT it
converges to
f(x
0) .exist;
x=
TT
/(
+ O)
*If the reader refers to

interval
101, he will see that,
if
f(x)
is
defined outside the
by the equation /(.r + 27r)=/( a. ), we can replace

1
-TT,
.
(a
TT)
'-Y(a;-2tt)""
In this form
(
we can apply
TT,
TT),
-sin a
7rJ
interval
0) exist*
andf(7r
"da + -/
the result of
TT
'
f(x + 2a)^
(1)
by
(2n+l)a da.
sin a
94 at once to every point in the closed
except points of infinite discontinuity.
FOURIER'S SERIES
There
of course,
is,
no reason
why
by the same analytical expression
213
the arbitrary function should be defined

[cf. Ex. 2 below].
in all the interval
Also it should be noticed that if we first sum the series, and then let x
approach a point of ordinary discontinuity .r we would obtain f(.v + Q) or
On the
./(''o- )> according to the side from which we approach the point.
other hand, if we insert the value .r in the terms of the series and then sum
,
we
the series,
obtain |[/(.^) + 0)-f/(^ -0)].
We
have already pointed out more than once that when we speak of the
sum of the series for any value of x, it is understood that we first insert this
value of x in the terms of the series, then find the sum of n terms, and
finally obtain the limit of this sum.
Ex.
Find a
1.
7T
represent
What
2 sinh
is
the
and
series of sines
x
cosines of multiples of
in the interval -TT<;C<TT.
TT
sum
of the series for
x'
TT
Here
IT ?
-2 siirh TT
FIG.
/**
/
e* cos n.v dx,
Ki.
Integrating by parts,
(
1 -f
n2 )
e*
cos n.v d.v = (c w
Wu = Lz
Therefore
Also we find
Similarly,
bn
7T
2x11111^'
7r<.i'<7r.
- e~ w ) cos rnr.
/
.
^.
(l
l+*
Therefore
when
x which
sin ?^*
dx
1.
will
FOURIER'S SERIES
214
When # =
sum
the
TT,
of the series
is
^Trcoth
TT,
since
- TT + 0) +/(TT - 0) = TT coth TT.
/(
In Fig. 16, the curves
;...)+(- T^cos8*+
are
drawn
for the interval
r^8in3*)j
TT, TT).
It will be noticed that the expansion
we have obtained converges very
slowly, and that more terms would have to be taken to bring the approximation curves [;/=,(.*)] near the curve of the given function in -7r<x<7r.
At .v= TT, the sum of the series is discontinuous. The behaviour of the
approximation curves at a point of discontinuity of the sum is examined in
Chapter IX.
Ex.
Find a
2.
and cosines of multiples

- TT < .v < TT, when
series of sines
represent /(./) in the interval
<l
=-
v TTX COS )iX

:
7
o
n
1
(* TTX sin
TrJo
Therefore
TT
When
f(.v)
< <
A"
.^=
TT
J-^(COS
nx ax=
?27T
will
),
7T
~-
4n
cos mr.
,r
TT.
the
sum
of the series
_ i
~7^
Find a
series of sines
1
represent x + x in the interval
Here
/
(.<:
f*
I
TTJjr
and, after integration
by
an =
which reduces to
Qn
/>
and we obtain the well-known
r >
QO ~T" 5"
3-
and cosines
of multiples of
n
./-')
(/.*=-=-/
x-d.>:=
TTJ
(x + x~] cos
??A'
dx
2
-
/''
I
-,
x- cos
nx (/.',
7TJO
part?,
r,
cos
TTJx
we
find that
4
?i
Also,
7r<.r<7r.
47Tj-ir
1
an = -
is ^7r
^
3.
dx =
= ~ + - - cos + sin - sin 2.; +

^
[^
^
fl?J
result,
Ex.
x which
*-
Here
when
of
?ITT.
(x-\-3P)&innxdx=\
TT.'U
""'
)
.''sin n.i'<li\
x which
will
FOURIER'S SERIES
215
Therefore
7T
when
.1:
When x=
4(
-^-
< <
TT
+ -cos# +
o\
x + x2 =
the
TT
sum
of the series
and
f(x) in
[
^2
The Cosine
96.
TT,
V)
satisfies
to
it
the results of
*"
leads to a
f(x')dx'
1 f
-n
1 f
95.
If*
I
f(x')dx',
2 f*
f(x) cos nx'dx' leads to a n = -\ f(x)cos nx'dx,

TTj
7TJ -
Thus the
7TJ
bn
Define
x)=f(x). The function

Dirichlet's Conditions in this
the equation /(
clear that in this case

1
an =
and
Let f(x) be given in the interval
and we can apply
it is
tt
Series.
= x<^0 by
thus defined for
But
and we obtain the well-known
is ?r 2 ,
satisfy Dirichlet's Conditions in that interval.
7r
interval,
...
TT.
result that
(0, TT),
1
1
\
/ 1
\
-sin x ) -41 ^cos2^ + -sin2^ )+
4
JB '--'/.'
\3r
/
f(x)sm nx'dx
leads to 6 n =
sine terms disappear
0.
from the Fourier's Series for
this function.
Also,
from the way
we have
in
which /(,) was defined in
i [/( +
and
+/( -
)J
7r
<!0
=/( + 0),
[/(-TT -f 0)+/V-0)]=/( T -0),
provided the limits /( + 0) and/(7r 0) exist.

In this case the sum of the series for x =
is
/(
+ 0),
and
for
= 7ritis/(7r-0).
when f(x) is an arbitrary function satisfying
Conditions in the interval (0, TT), the sum of the
It follows that,
Dirichlet's
Cosine Series
1 fn
7T J
"
f(x)dx+~y]cosnx\
7T
J
^p
f(x') cos
nx'dx'
+ 0)+/(a;-0)]
a?icZ TT where f(x + Q) and f(x
every point between
0) exist
and, when /( + 0) a??cZ /(TT 0) exis^, ê sum is /( + 0) at x =
is.
equal
and
/(TT
to
J[/(
0) at
X = TT.
FOURIER'S SERIES
210
Thus, when f(x) is continuous and satisfies Dirichlet's Conditions in the interval (0, TT), the Cosine Series represents it in
this closed interval.
Ex.
Find a
1.
the interval
(0,
series of cosines of multiples of
x which
will represent
TT).
Here
1 (""
/
xdx = ^TJ\
IT JO
and
ou
2 r*
/
x cos nx dx
if- (cos
V
W 2 7T
IT JO
*-Mn
Therefore
Since the
cos.r
sum
TrL
of the series
is
cos3^' +
zero at
In Fig.
17, the lines
17.
y=x.
^
'
'
I,
).
O^.r^TT,
# = 0, we have again
and the approximation curve
7T_4 /
y = 2 TrV 008
...
mr -
J.
32
are drawn.
FIG. 18.
in
FOURIER'S SERIES
how
It will be seen
y=
in the
217
closely this approximation curve approaches the lines
whole interval.
Since the Fourier's Series has a period 2?r, this series for unrestricted
x represents the ordinates of the lines shown in Fig. 18, the part
from the interval (-TT, TT) being repeated indefinitely in both directions.
values of
The sum
Ex.
2.
is
continuous for
Find a
all
values of x.
f(x) in the interval
x which
will represent
where
(0, TT),
f(x) = |TT (TT
#), *TT
< x^
IT.
Here
2 /*"
an = ~
and
2 /*
dx + ~
,
TTX cos w.r
=i
which gives
an =
^r
TT; O
^ cos nx dx + 1
(?r
TT(TT
cos
.r)
Also
vanishes
/'(.'-
when n
In Fig. U), the lines
^^ dx,
1-
FIG. 19.
Thus
x) cos u? ao:
is
odd or a multiple of
?/=^7r.r,
4".
M7T
FOURIER'S SERIES
218
and the approximation curves

2
V = TF T - \ cos 2# ~ r
are drawn.
It will be noticed that the approximation curves, corresponding to the
terms up to and including cos 60;, approaches the given lines closely, except
at the sharp corner, right through the interval (0,
For unrestricted values of x the

shown in Fig. 20, the part from
TT).
series represents the ordinates of the lines

to TT being repeated indefinitely in both
TT
directions.
The sum
Ex.
3.
is
continuous for
Find a
all
values of
in the interval (0,
TT),
Here
(#) cos
f(.v)
Fig. 21, the
the
at
will represent f(x)
?w dx =
7T.
cos
??.;
dx = - -
= \TT - [cos x - 1 cos 3. + 1 cos o.v -...],
= \TT,
since, when X
From the values
Iii
<X^
Also
x which
where
\1T
Thus
x.
sum
#=
^x^
of the Fourier's Series
and
.>;
= -, we
TT,
is
i[/(ir+0)4-/(Jir
have the well-known result,
10)].
graph of the given function, and the approximation curves
y = ITT - cos x + i cos 3.^,

y = ITT - cos A- + cos 3*' - i cos 5^,
^x^
TT,
are drawn.
The points .r=0 and X T^ are points of continuity in the sum of the series
the point X=\TT is a point of discontinuity.
The behaviour of the approximation curves at a point of discontinuity,
:
when n is large, will be treated fully in Chapter IX. It will be sufficient

now that it is proved in 117 that just before x = îr the approxima-
to say
FOURIER'S SERIES
219
tion curve for a large value ofji will have a minimum at a depth nearly 0'14
that it will then ascend at a steep gradient, passing near the
below i/ =
point (^TT, |TT), and risiifg to a maximum just after X=$TT at a height nearly
:
0'14 above
^TT.
0)
(3)
FIG. 21
Ex.
4.
Find a
in the interval (0,
Also

TT),
where
x which
will represent/(.r)
FOURIER'^ SERIES
220
"
Ti
=1
Here
-,
71
rir
=
Thus
sin
when n
vanishes
2 /3
~- [cos x f(x) =
And
-,
|J^
+ sill *W7T|
[sillAwTT
L
3?i
dx = Q,
/-TT
n-sj o
2
SU.
C*
-1
,
cfa;
is
iwr cos fyiir.
even or a multiple of
cos 5^- + } cos 7# - -fa cos
3.
1 l.r +...],
^^^
The points ^-=0 and ,v = ir are points of continuity in the sum

The points x=îr and X^TT are points of discontinuity.
Fig. 22 contains the
TT.
of the series.
graph of the given function, and the approximation
curves
3
y = ~~ [cos x
\ cos
y = ^- [cos .v
i cos 5^ +
2 /S
y
97.
(0, TT),
-\- [cos .r
The Sine
I cos
5.i;],
cos
7.^'],
5x + } cos 7x -
Series.
and satisfy
Again
^ cos
let f(x)
1 1^
be given in the interval

Define
Dirichlet's Conditions in that interval.
-7r^<0
f(x) in
by the equation /(-#)= -f(x). The function thus defined for (
TT, TT) satisfies Dirichlet's Conditions in
and we can apply
this interval,
But
it is
1 f
~
to
it
the results of
95.
clear that in this case
TTJ-jr
f( x ')
sin
'
nx'dx' leads to b n =
f
\
f(x') sin
nx'dx,
TTJO'
an =
and that
when n ^ 0.
Thus the cosine terms disappear from the Fourier's
Since
all
b l sin
vanish
Series.
when x =
x + 6 2 sin 2x +
and X
the
sum
of the series
is zero at
TT,
these points.
It follows that, ivhen f(x) is an
arbitrary function satisfying
Dirichlet's Conditions in the interval (0, TT), the sum
of the Sine
Series,
77
77
sin
nx
equal
to
f(x') sin
nxdx,
Jo"
is
f
I
i-
/(a?
+ 0) +f(x - 0)]
FOURIER'S SERIES
221
(2)
(3)
I!
Fia. 22.
(4)
FOURIER'S SERIES
222
and TT where f(x + 0) and f(x

and X = TT, the sum is zero.
noticed that, when f(x) is continuous at
at every point between
0) exist
and, when x =
be
It will
the end-
points x =
and x = TT, the Cosine Series gives the value of the

function at these points. The Sine Series only gives the value
of f(x) at these points
Ex.
Find a
1.
if
f(x)
is
zero there.
series of sines of multiples of
x which
will represent
in
the interval 0<.r<vr.
Here
bn
=-
x sin nx dx = ( -
1 )"~
7T/0
Therefore
At x
tf
the
x = 2 [sin x
sum
is
\ sin 2# +
sin
3# -...],
0^.
discontinuous.
FIG. 23.
In Fig.
23, the line
^t
and the approximation curve

y = 2 [sin # i sin 2^ + ^ sin 3.r
T = # = ^j
J sin
4^'
+ ^ sin 5.r],
TT
^x^
TT,
are drawn.
The convergence of the series is so slow that this curve does not approach
- TT and TT nearly as closely as the corresponding approximation
y=x between
FOURIER'S SERIES
223
y=x.
curve in the cosine series approached

If n is taken large enough
the curve y = Sn (x) will be a wavy curve oscillating about the line y x
from - TT to +TT, but we would be wrong if we were to say that it descends
/
-TT to the end of y = x, and again descends from

X=TT at a steep gradient. As a matter of fact the
summit of the first wave is some distance below y = x at x= -TT, and the
summit of the last wave a corresponding distance above y=x at X = TT when
at a steep gradient from
the other end of y = x to
is
large.
this question
we
To
&=
return in Chapter IX.
~7
FIG
Since the Fourier's Series has a period 27r, this series for unrestricted
values of x represents the ordinates of the lines shown in Fig. 24, the part
from the open interval ( TT, TT) being repeated indefinitely in both directions.
The points
Ex.
2.
STT, ...
TT,
Find a
in the interval
Here
&
"""
=-
=J
,
Thus
/>
At
these the
x which
ITTX sin
nx dx +
a?
nx dx +
sin
JO
is
zero.
/*(.*)
(TT
\Tr(ir-x}<s\Knxdx
- x) sin nx dx>
"/iff
.
TTJlir
11TT
sin-
= sin x -
f(x)
Fig. 25 contains the lines
sin 3.r
y=
y~-
and the approximation curves

1
y = sin x -
sin 3.r
+ ^ sin 5^7,
sum
will represent
O^.^^TT where
TTJQ
which gives
are points of discontinuity.
+ -^ sin 5.r - ....
FOURIER'S SERIES
224
be noticed that the last of these curves approaches the given lines
except at the sharp corner, right through the interval.
It will
closely,
FIG. 25.
For unrestricted values
shown
of
in Fig. 26, the part
both directions.
The sum is continuous for
x the
series represents the ordinates of the lines
from
all
TT
to +TT being repeated indefinitely in-
values of
x.
FIG. 26.
Ex.
3.
Find a
in the interval (0,
x which

TT),
where
*<ny
f(x) =
TT,
/(*)== 0.
\TT
< x<
TT,
j
will represent f(x)
FOURIER'S SERIES
Here
bn
225
shinxd.f
Jto
1
2
mr
SWTT
= -sm
r- am
.
when
Therefore 6 n vanishes
And
is
?i
a multiple of
>sin5.r-
4.
= sin .r- sin 2^7 + A sin 3^+
graph of the given function, and the approximation
curves
sin
sin
2.t',
y = sin x sin 2# + \ sin 3.r,

y = sin .r sin 2.r & sin 3.r + 1 sin 5#,
I.,,
-I-
.r =
JTT
The points
and x=ir are points of discontinuity in the sum of the
The behaviour of the approximation curves for large values of n at
these points will be examined in Chapter IX.
series.
Ex.
4.
Find a series of sines of multiples of

TT), where
which
.r
will represent f(.v)
in the interval (0,
/(0)=/0r) = 0;
Also
bn =
Here
/(M-jTr;
sin n.f
( /./
/(fir)= -Jir.
sin
Jo
7i.r
dx
Jjn-
2
- cos J W7T - cos 5W7T -f cos
= s[1
NTT]
;
3T6
Therefore
And
/(#)
vanishes
when n
is
odd or a multiple of
= sin 2.f + i sin 4.r + 1 sin 8.r +
The points ^=0,

sum of the series.
.r
= JTT, x
$7r
and
.t'
= 7r
graph of the given
sin
1 QJ-;
. . .
function,,
y = sin 2.r,
= sin2.r + |sin
y = sin 2.r + sin 4.# + j sin 8#,
= sin + i sin 4.v + ^ sin + 5 sin 10*',
?/
c, i
4./',
2.1'
8.t'
.r
"^ TT.
are points of discontinuity in the
rvs
7/
6.
and the approximation
FOURIER'S SERIES
226
II
0)
:|
i
(2)
(3)
FTG. 27.
x.
FIG. 28.
(4)
FOURIER'S SERIES
228
When
Other Forms of Fourier's Series.
98.
function
given in the interval
is
interval to
In this
by the substitution
TT, TT)
way we may deduce
those already obtained
ii
the arbitrary
we can change
1,1),
the following expansions from
',
V cos ^fx[ f(x
-l^x^l,
'+~
"i
,
,x
sm
n-JT
-y
x ,-,,
ax
^x<
Jo
t-
(#
this
= 7rx/l.
cos^xdx,
^
I,
...(1)
.(2)
_
...................... (3)
O^x^l
n
When /(#) satisfies Dirichlet's Conditions in ( I, I), the sum of

the series (1) is equal to o[/(^ + ^)+/(^~^)] a ^ every point in
l<^x<^l where /(z + 0) and f(x
sum
and at 0;=
the limits
its
/(-Z + 0)
and
exist.
/(Z-0)
When
/(a?) satisfies Dirichlet's
the series (2)
is
equal to ^[f(x
0<^x<^l where /(#+0) and f(x

is
0) exist;
) + /V-)]> wn en
K/(- +
is
/( + 0), at x = l
When
its
sum
is
Conditions in
(0, I),
the
sum
of
0)+f(x 0)] at every point in

its sum
0) exist; and at x =
f(l
0),
when
these limits exist.
f(x) satisfies Dirichlet's Conditions in
(0, I)
the
sum
of
is equal to %[f(x + Q)+f(x

Q)] at every point in
where /(x+0) and f(x 0) exist; and at x = and
the series (3)
0<<Z
x=l
its
It is
sum
is zero.
sometimes more convenient to take the interval in which
the arbitrary function
is
given as
for
We may deduce the

from that already found
(0, 27r).
corresponding series for this interval

TT, TT).
Consider the Fourier's Series

*
y
7r^-'J_
P F(x) dx + ^TPj-jr
where F(x)
F(x') cos
Conditions in
U = TT + X, U' = TT + X and
Then we obtain the series for f(u),
Let
oo a**' +
for the interval (0,
2-Tr).
TT
n (x' - x) dx,
7r
TT, TT).
f(u) = F(u
7r).
FOURIER'S SERIES
On changing u
2 'r
^TTJo
'f(x')
dx
into x,
we have
+ ^- S 'O
/<X) cos
229
the series for/(&),
<X - *) d'
= x = --*
-(4)
The sum of the series (4) is J[/( + 0)+/(x-0)] at every

and 2-jr where f(x + 0) and /(a; 0) exist; and
point between
at a; = and a; = 2?r its sum is
when
In
these limits exist.

(4), it is
assumed that f(x)
in the interval (0,
Conditions
27r).
Again, it is sometimes convenient to take the interval in which

We can deduce the correspondthe function is defined as (a, 6).
ing series for this interval from the result just obtained.
Taking the series
i
f27r
=-
F(x') dx'
we
.,
a)#
jF(o;')
cos
,,
n (x - x) dx,
f(u) = F\
and
r27r(it
Q^X^ZTT,
a)l
aû^b we have the series for f(u\

2
2??7T
f
^n +
2 /(* cos rzr ~ u ^^
for the interval
If
/
= a-f (6
,
?t-
pa*-
JO
write
Then
i
Tf"
7rJ()
&
'
/( tc
'
On changing u into
a?,
6 ')
we obtain
the series for f(x) in
a^x^b,
namely,
is J[/( a; + ^)+/( a; '~^)] a ^ very point

where
in a<ix<^b
f(x + Q) and f(x 0) exist; and at # =
=
and x b its sum is i[/(ot + 0)+/(6 0)], when these limits exist.
Of course f(x) is again subject to Dirichlet's Conditions in the
The sum
of the series (5)
interval
(a-, 6).
The corresponding Cosine

i
tively,
Series
f ft
2
r
and Sine Series
-- (x
Vcosrb.a
TITT
are, respec-
a)
(6)
and
agagft....... (7)
FOURIER'S SERIES
230
Ex.
1.
Show
that the series

4 /
is
equal to
Ex.
2.
when
Show
Ex.
Show
3.
3
is
2c/
4.
Sm
7r?tt
represents
TTX
ira
(TT
to zero
TWT f_,
~3l
2 ( y3-î)
vt
Show
ZTTX
when a<.x<.L
that the series
equal to
Ex.
0<#<.
when 0<.i'<a and
equal to
Vi+Vo-f-yo
TTX
that the series
ca
is
sm
when -
2ft7r
-3-
mrx
sin -
l<x<
i/c
+( 2
--,
o
v2
when
~3<^<3'
v3
when
-<#<.
that the series
x) in the interval
As has been
99. Poisson's Discussion of Fourier's Series.
mentioned in the introduction, within a few years of Fourier's
discovery of the possibility of representing an arbitrary function
by what is now called its Fourier's Series, Poisson discussed the
subject from a quite different standpoint.
He began with
the equation
-52
= 1 4- 2
Vr
cos
n (x - x),
where |r|<l, and he obtained, by integration,

ff
1 f
r2
If
= ^pA7TJ
77
n
f(x ) dx' + ^ y] r
_.
f
"
71
J-T
/(a?')
cos
TI (x
x) dx'.
Poisson proceeded to show that, as r->l, the integral on the

left-hand side of this equation has the limit f(x), supposing f(x)
continuous at that point, and he argued that f(x) must then be the
sum of the series on the right-hand side when r = 1 Apart from
.
the incompleteness of his discussion of the questions connected
FOURIER'S SERIES
231
with the limit of the integral as r-l, the conclusion he sought

draw is invalid until it is shown that the series does converge
when r = 1, and this, in fact, is the real difficulty. In accordance
to
with Abel's Theorem on the Power Series (72), if the series

converges when r=l, its sum is continuous up to and including
r=l. In other words, if we write
F(r, &) =
we know
that,
if
Vr
f(x) dx
x}dx,
f(x')Gosn(x'
F(I, x) converges, then
Frx
Lt
But we have no right
to assume,
Lt F(r,x\
r-M
that F(\,x) does converge.

Poisson's method, however, has a definite value in the treatment
of Fourier's Series,
and we
series
100.
is
and integrals
in the previous pages of this book.
The
Poisson's Integral.
called Poisson's Integral.

shall assume that/(.r)
We
(
TT, TT),
now give a presentation of it on

we have followed in the discussion of
shall
the more exact lines which
is
integral
either
or that the infinite integral

I
Now we know
that
-n
f(x)dx
in the interval
"
r2
is
= 1 + 2 2 rn cos nO
when |r'<l, and that this series is uniformly convergent

when r has any given value between - 1 and + 1.
for
any interval
of 6,
*
It follows that
^ /
ZTTJ - n 1
- 2?
cos (x
-dx' = l,
-x) + r'
and that
at
TTT
27T J -
l-7'2
l-2/-cos(.r'-^)+r
2/l
(1)
nuder the limitations above imposed upon

74,1.)
f(.v).
(Cf.
70, Cor. II.,
and
FOURIER'S SERIES
232
Now
sum
let
number x between
TT and TT for which we
what is the same thing, the value of Poisson's
us choose a
of this series, or,
27T/- ,
wisli the
Integral,
l-
Denote this sum, or the integral, by F(r, .).

Let us assume that, for the value of x chosen,
Lt [/(* +
/
<)
*o
+/(*- 01
exists.
Also, let the function
<!>(#')
be defined when - TT^X'
</,>=/>-*
^== TT
by the equation
Lt [/
t
>o
Then
F(r,- x)
- 1 I* [f(x + 1)
- 0]
+/(.
Lt
t^-o
[f(.v
+ t)+f(.v-t)'\
exists.
Let the arbitrary positive number e be chosen, as small as

to e/2 there will correspond a positive number
such that
?/
when (Xt^r).
The number
beyond
Then
77
fixed
upon
will be such that
we
please.
(x-i^ # + ?/) does not go
TT, TT).
1,
27T
It follows that
Then
1-r 2
FOURIER'S SERIES
Also,
I
when 0<r<l,
/ c*-*
233
/*"
But
And
provided that
It follows that
1
Combining
(4)
and
(6), it
has been chosen, as small as

|
F(r, x)
will
we
be seen that when any positive number

number p such that
please, there is a positive
- 4 Lt [/(* +
+/(.r
*)]
<
*,
when/3^r<l, provided that for the value of x considered Lt [f(x+t)+f(x
t)]
exists.
We
have thus established the following theorem

TT, TT), be bounded and integrable, or have an
(
Then for any value of
absolutely convergent injinite integral, in this range.
x in
which
.
Lfc [/(jF + 0+/(ar
0]
:
Let f(x\ given in the interval
-<*<>
exists, Poisson's
>o
Integral converges to that limit as r->l
from
below.
In particular, at a point of ordinary discontinuity of f(x), Poissoris Integral

converges to
4[/(* + 0)+/(.t-0)]
and, at a point where f(x) is continuous,

It has already been pointed out that
converges to f(x).
no conclusion can be
it
drawn from
this
as to the convergence, or non-convergence, of the Fourier's Series at this

But if we know that the Fourier's Series does converge, it follows
point.
from Abel's Theorem that

Integral converges as r->l.
it
must converge
to the limit to
which Poisson's
FOURIER'S SERIES
234
We
have thus the following theorem
any function, given in
is
If f(x)
- TT,
(
TT),
which
bounded and
either
is
rir
integrable, or has
an
absolutely convergent infinite integral
-n
any point x in
must be equal
TT<X<TT
f(x) dx, then, at
at which the Fourier 's Series is convergent,
its
sum
to
Lt [/(# + 0+/(*-01
<->o
provided that
With
this limit exists.
certain obvious modifications these theorems can be
to the points
TT and TT as well as
points between
It follows immediately from this theorem that
If
TT
and
made
to apply
TT.
Constants are zero for a function, continuous in the

then the function vanishes identically.
all the Fourier's
interval
TT, TT),
the constants vanish but the function only satisfies the conditions
ascribed to /(.*') in the earlier theorems of this section, we can only infer that
the function must vanish at all points where it is continuous, and that at
If
points where Lt
t
if
Further,
number
p,
>o
f(x + 1) +/(#
(a, b} is
- t)],
an interval
exists, this limit
in
must be
zero.
which f(x) is continuous, the same

e, may be chosen to serve all the
corresponding to the arbitrary
values of x in the interval
(a, b) ; for this is true, first of the number TJ in

in (5), and thus finally of p.
It follows that Poisson's Integral converges as r->\ uniformly to the value
(3),
then of
f(x) in any interval (a, b) in which f(x) is continuous*

This last theorem has an important application in connection with the
approximate representation of functions by

101.
trigonometrical series, t
Fear's Theorem.!
Let f(x) be given in the interval

be integrable; if
unbounded,
terms
Also
finite
<v
TT,
TT).
If bounded,
the infinite integral
let
Denote by
sn the
sum
let it
f(x)dx
of the
i
'
let
Sn (x) =
*It is assumed in this that /(a - 0) =/() =/(o + 0) and/(6-0)=/(6)=/(6 + 0).

Also f(x) is subject to the conditions given at the beginning of this section.
Of.
107.
tCf. Picard, Traite d' Analyse, (2e ed.), T. I., p. 275, 1905
Princeton, N.J. (Ser. 2), 7, p. 102, 1906 Hobson, loc. cit.,
;
JCf. Math. Ann.. Leipzig, 58,
p. 51, 1904.
Bocher, Ann. Math.,
p. 722.
FOURIER'S SERIES
Then
235
at every point x in the interval
f(x + 0) and
7r<x<7r
at
which
0) exist,
f(x
^
Lt Sn (x) =
With the above

ti
notation,
1
cos
Jj
x f* j/f{~/\
27rJ _ n
I
n (X
'
~~
X ^~ COS ( n + 1 K X/ ~~ x~~^
cos(V
x)
M
(AriL>
Therefore
Sn(x) = =
f*
2n7r)
_n
,'
X)
00871(0?'
IGOS(XX)^dx
f(x )
j
d
if
is
/(x)
defined outside the interval
by the equation
TT, TT)
/(* + 2,r) =/(*).

Dividing
7r
(x,
+ #),
the
in the second,
Sn (x) =
Now
and
/(aj
of
range
and substituting
we
integration
x'
=x
into
2a in the
TT
first,
+ X,
and
x)
and
x'
obtain
v
f(x
suppose that x
da +
2a) sin 9
is
?l
a point in
f(x + 2a) Sln

.
TT,
at
TT)
-.
c?a ....... (2)
which
0) exist.
be any positive number, chosen as small as we please.

Then to e there corresponds a positive number rj chosen less
Let
than
JTT
such that
|/(a
+ 2a)|-/(aj + 0)|<6 when
Also
-\
r^\^
sm
^/
--- jf(x
+ 0}\
mr
Jo
da
sm
............... (3)
*
is
This discussion also applies when the upper limit of the integral on the
any positive number less than \ic.
left
FOURIER'S SERIES
236
Putting
<r
..
= J + cos 2a + cos 4a +
. . .
+ cos 2 ( n
) a,
-cos 2a)
4(1
Thus
since all the terms
except the
first
on the right-hand side disappear on integration
in each of the
/2 =
It follows that
fsin
cr's.
J/(aj
7i-a
2
rjg sin a
,
7ro
<Ae
.......................................................... (4)
l/.

defined
f(x)
da
<
Further,
+ 0).
outside
the
I
|
J -7T
f(x) dx' converges, and we have

\
interval
TT, TT)
by the equation
*+%r)-A*>
Let
Jf"-n
\f(x')\dx'
= TrJ,
say.
Then we have
Als
I
7*
<
rl/(*+0)l
...........................
(6)
FOURIER'S SERIES
Combining these
Now
results, it follows
from
237
(3) that
be a positive integer such that
let v
^{/+|/(,
+ 0)|}<e
(7)
Then
<
e,
when
71
i/.
In other words,
1
nr
7rJ
when f(x + 0)
exists.
In precisely the same

in
way we
TA - f
Lt
/(%
7
^^TiTrJo*
when f(x
find that
sin 2 7i
2a);
sm92 a
da = ^f(x
,
/.,
0),
0) exists.
Then, returning to
U
when /(ic0)
(2),
we have
&(*)=* [/(* + <+/(! -0)],
exist.
This proof applies also to the points X=TT, when /(TT 0)

and /( TT 4- 0) exist. Since we have defined f(x) outside the
interval
TT,
TT)
by the equation f(x + 27r)=f(x),
it
is
clear
(- 7r + 0)=/( 7r + 0) and f(- 7r -0)=f( 7r - 0).

In this way we obtain
+ 7r
Lt
when
/(
7T
COROLLARY.
+ 0)
and
If f(x)
is
/(TT
0) exist.
continuous in a^.v^=b, including the end-points,

is chosen, the same 17 will do for all
when the arbitrary positive number

values of x from a to
including the
?>,
end-points.
Then, from
that the sequence of arithmetic means

#!>
converges uniformly to the
sum f(x)
*^2>
'^35
in the interval (a,
Z>).
(7), it
follows
FOURIER'S SERIES
238
It
x=b
assumed
is
this statement that /(.r)
in
as well as in
the interval
i.e.
(a, 6);
is continuous at x = a and
f(a-0)=f(a)=f(a + 0), and
Before applying this

102. Two Theorems on the Arithmetic Means.
very important theorem to the discussion of Fourier's Series, we shall prove
two theorems regarding the sequence of arithmetic means for any series
In this connection we adopt the notation
THEOREM
converges
We are
If the
I.
and
its
sum
+ u^ + w 3 +
Lt Sn = Lt s n
series
is s,
u^
then
n
. . .
><x
s.
>ao
given that the series

tt
converges, and that
its
sum
is
+ M 2 + M 3 +...
5.
Thus, to the arbitrary positive number e there corresponds a positive

v,
integer v such that
\s-sn \<e, when n
Thus, with the usual notation,

p Rn
Also
it
=
|
sn+p
follows,
-sn \< 2e, when n ^ v,
from the definition
of
Sn
that,
when
n> v,
n-\
But
are
all
positive
and decreasing
\U v +i\,
are all less than 2e

It follows, as in
Thus
Therefore
v
when
,?
1
and
+ U v+a
n>
\,
...
\U v +i+U v+S +...+U
v.
50, that
Sn-liiix + UvlV 1
I
But
\U v+1
when
-H
+ ... +wJl--)
\
n n <2e,
nj
71
>v.
- S , |< 2, +
being fixed, we can choose the positive integer

+(i/-l)?^|
<
e,
when
"
71
N so that
N>
v.
FOURIER'S SERIES
-
Therefore
< 3e,
when
n^
\Sn -s\<4c, when

Lt Sn =s.
n^
239
But
And
Therefore
Thus
THEOREM
II.*
Let the sequence of arithmetic means
Sn
of the
series
(1)
<
to S.
converge
Then, if a positive integer n exists such that

\
where
its
is
sum is S.
With the same notation
We
un
when n > ?i
< K/n,
some positive number independent of

as above, let
have to prove that
Lt
n
tn
?i,
the series (1) converges
and
= s n -'S.
vao
If t n has not the limit zero as ?i->oo, there must be a positive number h
such that there are an infinite number of the terms t n which satisfy either
>h or (\\)tn <-h.

We shall show that neither of
(\)t n
Take the former
to be true,
Then
<rn
Lt
and
these hypotheses can be admitted.
and
/n=Sn -S,
o-n
fn=0.
Thus, to the arbitrary positive

integer
n-^
such that
\(rn
let
number
e,
/n\<f, when 71^71,.
Choose this number e so that hlK+2<%h 2 /fLe.

Now let n be any positive integer greater than both N O and n such that
tn
>h.
Then, when r=0,
1, 2, ...
un+r
< K/n.
FlO. 29.
Now
plot the points
diagram.
*
Since
t n+r+
r,
whose coordinates are
i-tn+r =u n+r +i, the slope
of
(r, t n+r )
the line
in a Cartesian
Pr Pr +i
is
less
This theorem was published by Hardy in London, Proc. Math. Soc. (Ser. 2), 8,
This proof, due to Littlewood, is given in Whittaker and
pp. 302-304, 1910.
Watson, Modern Analysis,

cit.
T. II.,
151.
p.
157,
1920.
Cf. also
de
la Vallee Poussin,
loc.
FOURIER'S SERIES
240
than tan" 1 A'/?*.

Therefore the points P Plt
above the line /=A-.vtan0, where ta.u0=K/n.
Let PK be the last of the points P Plt P2 ... which lies to the
A cot 0=hnjK.
upon, the line x = h cot #, so that K
in absolute value
P^
...
left of,
lie
or
Draw rectangles as shown in the diagram. The area of these rectangles

exceeds the area of the triangle bounded by the axes and y = h-xta.n 0.
Thus we have
"M+K
But
~ Vn-l=t n + tn + l + ... + tn K >

+
|o-n+K -0- n _ 1 |^|orn+/c
+ |o-n_
(K + 2w
Therefore
1) e
But
M COt 6=%h nfK.

2
|<[(?l + K) + (-l)],
SmC6
?l
>?!.
> %h n/K
2
K^hcotO=hn/K.
Ar^/^^ K > WnlKc. - (2w -
Thus
).
+ 2] > 1.
2
h/K - A /A'e + 2 < 0.
% [A/ A'-
Therefore
But
Therefore
/ A'e
<T
But we have assumed that there are an infinite number of values of n

t n >h.
The hypothesis (i) thus leads to a contradiction and a precisely similar
argument shows that the hypothesis (ii) does the same.
such that
Thus
Lt
*n
=0,
n ><
audit follows that
Lt
n
sn
Lt
n
*-oo
Sn =S.
><x>
COROLLARY.
Let
be
+ M 2 (.r)-fw3 (.r)+
tt 1 (-*0
whose terms are functions of x,
and
(1)
the sequence of arithmetic

means Sn (x) for the series converge uniformly to S(x) in an interval (a, b).
Then, if a positive integer n Q exists such that
series
let
'
u n (x)
< K/n,
independent of n and
when n > n Q
and
serves for all values of x

in the interval, the series (1) converges uniformly to S(x) in (a, 6).
For, with the notation of the theorem just proved, if t n (x) does not tend
where
is
x,
the
same n
uniformly to zero through (a, b), there must be a positive number A, independent of x and n, such that an infinite sequence of values of n can be found
which t n (xn )>h or t n (x n )<-h for some point xn in the interval; the
value of xn depends on the particular value of n under consideration.
then find, as in the original theorem, that both these hypotheses are
for
We
inadmissible.
103.
We
shall now use Fejer's

Fej&r's Theorem and Fourier's Series.*
to establish the convergence of Fourier's Series under the limitations
Theorem
imposed in our previous discussion

!
Cf.
Whittaker and Watson,
that
is
loc. cit., p. 167,
we
shall
1915,
show that
and the note on
p.
263 below.
FOURIER'S SERIES
Whenf(x)
241
TT, TT),
and
'
'
bn = - f
')
n
f(af) sin nx'dx' (n
cos not dx,
),
TTJ -7T
szm o/ the
series
-TT<X<TT
at ever2/ point in
~0)]
x=
/(.r-0) e#ul; and at
TT
sum
the
is
where
/(.?'
+ 0) and
^[/( -7r + 0)+/(7r-0)], when these
limits exist.
First, let /(.r)
I.
be bounded in
- TT,
rr)
and otherwise
satisfy Dirichlet's
Conditions in this interval.

interval (-TT, TT) can be broken up into a finite number (say p)
open partial intervals in which f(x) is monotonic, and it follows at once
from the Second Theorem of Mean Value that each of these intervals contributes to \a n or |6J a part less than 4J//W7T, where \f(.v)\<Min (-TT, TT).
Then the
of
Thus we have
where
is
a n cos nx + 6 n sin nx
< 8pM/nv,
independent of n.
from Fejer's Theorem, combined with Theorem
It follows
II. of
102,
that the Fourier's Series

)
converges, and
at
which
its
f(xQ)
sum
is
exist,
&[/(.^
2 cos
2.7-+
+ 0)+/(*i'-0)]
b<>$
at every point in
and at x=* ir its sum

and /(TT - 0) exist.
is
-TT<X<TT
i[/( -7r + 0)+/(7r-0)J,
provided that /( TT + 0)
II. Next, let there be a
finite number of points of infinite discontinuity

arbitrarily small neighbourhoods of these points are
excluded, let f(x) be bounded in the remainder of the interval, which can be
broken up into a finite number of open partial intervals in each of which /(.r)
in (-TT,
is
TT),
when
but,
monotonic.
And
/*T
in addition
the infinite integral
let
f(x')dx'
be
case, let x be a point between -TT and TT at which /(.r + 0) and
-0) exist. Then we may suppose it an internal point of an interval
(, &), where 6-a<7r, and f(x) is bounded in (a, b) and otherwise satisfies
In this
/'(.
Dirichlet's Conditions therein.
Let
7ra,/=
irb n'=
and
/(.r')cos nx'dx'
7a
'
/'(.')
sin nx'dx'
'
while
27r
Then, forming the arithmetic means for the series

o'
c. i
+ (i'cos .r + 6/sin
.r)
+ (.2 'cos 2.T + 62'sin

Q
2.r)
. . .
FOURIER'S SERIES
242
we
have, with the notation of
where
But
101,
and ^(&-#) are each positive and less than |TT.

be seen that the argument used in Fejer's Theorem with regard
(#-)
it
will
to the integrals
/(^2a)
J
than
'
when the upper
applies equally well

less
sin 2 OTa
rfr
nirJQ
sin 2 a
da
limits of the integrals are positive
and
JTT.*
Therefore, in this case,
Sn (x] = \[f(x + 0) +/(.* - 0)].
Lt
And,
as the
'
terms
(a n cos
satisfy the condition of
ao
its
sum
Lt
is
11
But
n ')
+ 2{(n -
')
'
sin nx)
102, it follows that the series
II. of
/cos .r + fc/sin x) + (
converges and that
Theorem
nx + b n
2 'cos
%x + 6 2 'sin
2.r)
,,(.r).
>30
cos ?i.r + (6 n
- &') sin ?z.r}
sma
-a)
94 both of these integrals vanish in the limit as

It follows that the series
By
converges, and that its
')
+ 2{(
sum
- 6 /)sin
n')
cos ??^ + (6
is
defined by the equation f(x + 2ir)
J(
nx]
is zero.
But we have already shown that
converges, and that
its
sum
is
Cf. footnote, p. 235.
Outside the interval
TT,
ir)f(x)
f(x).
FOURIER'S SERIES
It follows,
by adding the two
series,
that
+ 6,
converges, and that
at
its
sum
sin nx)
is
any point between -TT and
TT
at which these limits exist.
When
the limits /( Tr + 0) and/(7r

of the sum of the series for x=
equ ation /( x + 2;r ) =/( x).

We can then treat x=
243
TT
0) exist,
TT
we can reduce
the discussion
above argument, using the
to the
as inside an interval (a,
b), as'
above.
REFERENCES.
BOCHER, "Introduction to the Theory of Fourier's
Princeton, N.J. (Ser.
DE LA VALLEE
2), 7,
Series,"
Ann. Math.,
1906.
POUSSIN,
T. II. (2" ed.), Ch. IV.
loc. cit.,
DINI, Seine di Fourier e altre rappresentazioni analytiche delle funzioni di
una
variabile reali, Cap. IV., Pisa, 1880.
FOURIER, Theorie analytique de la chaleur, Ch. III.

e
GOURSAT, loc. cit., T. I. (3 d.), Ch. IX.
HOBSON, loc. cit., Ch. VII.
e
JORDAN, Cours d> Analyse, T. II. (3 6d.), Ch. V.
LKBESGUE, Lemons sur les se'ries trigonome'triquex, Ch. If., Parrs, 1906.
NEUMANN, Uber d. nach Kreis-, Kugel- u. Cylinder- Functionen
schreitenden Entwckelungen,
RIEMANN,
II.,
fort-
Leipzig, 1881.
loc. cit.
WEBER-RIEMANN, Die
Physik, Bd.
Kap.
I. (2.
Aufl.),
partiellen Differential-gleichungen der mathematischen

Braunschweig, 1910.
WIIITTAKER AND WATSON, Modern Analysis, Ch. IX., Cambridge, 1915 and
1920.
EXAMPLES ON CHAPTER
1.
and
In the interval
-I
in the interval
p
2.
fi
f(x) =
0<.r<-,
.r
2?r^
\_M_I
The function f(x)
< <
is
I,
/(.?')
- .r,
3
=x
.Gir.v
VII.
-l>
_1_
I&ir.r
defined as follows for the interval (0,
f(x] = %x,
when ^ x IfE \TT,

f(x) = \ir, when J7r<.^<7r,
= (7T-.r), when 7r^.r^7r.
/(.<)
Show
that
\zn
1)"
TT)
FOURIER'S SERIES
244
3.
Kxpand/(.r)
in a series of sines of multiples of
Tr.r/or,
given that
= m.t.;
f(.v)
when OÂ'îrt,
f(as)m(ax\ when la^Sxâ.
4.
Prove that
2?i7r.r
S111
oo
< <L
when
.r
and
5.
of
Obtain an expansion in a mixed series of sines and cosines of multiples
x which
and gives
6.
is
its
Show
zero between
- ?r and
and
0,
is
equal to
that between the values
expansions hold
/ sin
= -2 sin WITT
7W.tr
TT
cosh
-TT and
mx
2w
m cos
2/1
2sin2.r
wcos#
cos
TT,
Express
.r
+TT of
.r
the following
+ T9J
1
.r
3sin3#
wcos3^ ~
7?icos2.r
-m +~z3- -7ft
w cos 2/r ~ m cos 3j?
f
9-"^â
22
for values of x between -TT and

and a series of cosines of multiples of .r.
Prove that the locus represented by
sin
9
2
TT
as the
sum
of a constant
w# sin ny=Q
is two
systems of lines at right angles dividing the plane of
squares of area TT-.
8.
and
between
7.
e*
values at the three limits.
.r,
into
Prove that
2 c3
"
represents a series of circles of radius c with their centres on the axis of x

at distances 2c apart, and also the portions of the axis exterior to the circles,
one
circle
9.
having
its
polygon
is
centre at the origin.

inscribed in a circle of radius
a,
and
is
such that the
alternate sides beginning at $ = subtend angles a and ft at the centre of

the circle. Prove that the first, third, ... pairs of sides of the polygon may
be represented, except at angular points, by the polar equation
770
27T0
FOURIER'S SERIES
245
Where
/8
p C08
n*<f>
fj.
Find a similar equation to represent the other
sides.
Investi10.
regular hexagon has a diagonal lying along the axis of
gate a trigonometrical series which shall represent the value of the ordinate
of any point of the perimeter lying above the axis of x.
11. If
./;.
0<.r<27r, prove that

TT
~'
x) _ sin x
a (ir
sinh
"
2
12.
2 sin
3 sin &p
2.t-
Prove that the equation in rectangular coordinates

2
4A /
TTJ;
27r.y
ZTTX
represents a series of equal and similar parabolic arcs of height h and span
2K standing in contact along the axis of x.
The
arcs of equal parabolas cut off by the latera recta of length 4a are
arranged alternately on opposite sides of a straight line formed by placing
the latera recta end to end, so as to make an undulating curve. Prove that
13.
the equation of the curve can be written in the form

3
TTX
1
3irx
7T y
--H
--h c-i1J sin 4TTX
f- = sin 7- + -T.3 sin
5
4a
64a
4a 3
4a
.
14. If circles be drawn on the sides of a square as diameters, prove that

the polar equation of the quatrefoil formed by the external semicircles,
referred to the centre as origin, is
where a
is
the side of the square.
15. On the sides of a regular pentagon remote from the centre are
described segments of circles which contain angles equal to that of the
pentagon ; prove that the equation to the cinquefoil thus obtained is
7rr=5atan
a being the radius of the
16.
and
In the interval
in the interval
circle
circumscribing the pentagon.
0<.r<
-
f(.v)=x*
FOURIER'S SERIES
246
Express the function by means of a series of sines and also by means of a
^. Draw
figures
represented by the two series respectively for

lie
between
17.
and
A point
What
I.
moves
sums
are the
showing the functions
values of
all
x not
of the series for the value
#= 5 ?
2
with a velocity which
in a straight line
restricted to
is initially
and which receives constant increments each equal to u at equal intervals

Prove that the velocity at any time t after the beginning of the motion is
1
+~Tru-^l
2 -sm
lit
v = -u +
2
T
and that the distance traversed

Ut f
2T
18.
A curve
is
2mr
T
1,
UT
UT
2^7T
+T)+ T2~2p^ cos ~7~'-
formed by the positive halves of the
and the negative halves
- (4n - l)a) 2 +!/* = a 2
n being an integer. Prove that the equation

by Fourier's method is
Having given the form

2
y = TT
roi
_,
[See Ex> 4 above>]
circles
of the circles
(.v
19.
is
(*
n= \n
T.
of the curve
2 sin rxJo
for the complete curve obtained
y=f(x\
trace the curves
f(t) sin rt dt,
y= 7T- 2 sin(2r - 1U-JO
/(0 sil
and show what these become when the upper limit

20.
Prove that
for all values of
between Oand
is
instead of
- the value of the series
a
rirb
,, 1
- sin
Z,
-j.
is
zero for
is
-j
all
values of x between
for all values of
21.
Find the sum
sm
nrat
TTTX
j- sin
j
L
L
,
and
x between b-at and
at
at +
and between at + b aud

b,
when
Z,
and
6<;r2
of the series
1
* sin
2?i7T.
sin(2n-l)iro7
=
7^
2n-l
'
and hence prove that the greatest integer in the positive number x
2
[See Ex. 1, p. 230, and Ex. 4 above.]
represented by x+u-8v
.
is
FOURIER'S SERIES
22. If
-r,
the value of
and
a, z
the rectangular coordinates of a point which moves so
y, z are
y=
that from
y = x the value of z
2
2
K(a -y ), show that
to
z is
may be
find the values of
-> ?('^)
1
Ap
/(#) = ^
and
prove
when
#,
'î
find the values
>
^* ^^
when
<^^
2.t'
TT,
Ex.
[Cf.
*S'
1, p.
(su\x
2 /
->
3.*;
4-
-]-
sin G.r 4-
;\
+5
sin4.v-f^
f(x)=
If
is
246
siu
^3 = sin 2# 4of Slt S.2 and
the assigned interval.
show that f(x)
(^r)'
when
^ = sin x + ^ sin
24
sill
OÊ#<TT,
where
and
and from y = x to y = a
x and y between
-.r 2 ),
for all values of
for the different types of terms.
/(#) = |j sin
23. If
that,
is /<(a a
expressed by a series in the form
-J
and
247
sin
-1-
^sin6.r-...,
<
bx+ ...
sin UU- 4-
. . .
separately for values of
x lying within
230.]
sm 3>t
sm 5j/ -
sin2#
sin
r+:
("
continuous between
show that /'(->') has a sudden change
and
of value
3./-
and that /(TT-O)=
TT,
- at the
point
^.
1.
Also
[See Ex. 1,2,
pp. 222-3.]
25.
Let
/(#)=*
2)i-\
when O^^'^TT.
Show that
also that
Draw
above.]
~2S
8n
^_
+--
/( + 0)=/0r-0)=
'"
H
/
2;i
-i-TT,
AO)-/['lr)-/ahr)-/[v;)-0,
the graph of f(x} in the interval
(0, TT).
See Ex.
1, p.
230,
and Ex.
CHAPTER
VIII
THE NATURE OF THE CONVERGENCE OF FOURIER'S

SERIES
104.
The Order of the Terms.
Before entering upon the
discussion of the nature of the convergence of the Fourier's

Series for a function satisfying Dirichlet's Conditions, we shall
show that
in
determined
easily.
cases the
certain
order of the terras
may
be
// f(x) is bounded and otherwise satisfies Dirichlet's Conditions in the interval (

TT, TT), the coefficients in the Fourier's
is
Series for f(x) are less in absolute value than K/n, ivhere
some positive number independent of n.
I.
Since
tra n
-n
f(x) cos
nx dx,
and the interval may be broken up into a h'nite number of open

partial intervals (c r c r+1 ) in which f(x) is monotonic, it follows,
from the Second Theorem of Mean Value, that
,
7rtt M
=S
("
c r+l
Jcr
f(x) cos
nx dx
r
cr
where f r
is
+0)|
some
Thus
'
cosnxdx+f(c r+l
definite
n
number
0)
in (c r c r+l ).
,
|<~n
n
where
bound
is
number
the
of \f(x)
of partial intervals
in the interval (
TT, TT).
Therefore
is
some positive
<
an
K/n,
number independent
\
where K
and
248
of n.
is
the upper
FOURIER'S SERIES
And
similarly
we
249
obtain
\l\<K/n.
We may
speak of the terms of this series as of the order l/n.

the terms are of the order l/n, the series will, in general,
When
be only conditionally convergent, the convergence being due -to

the presence of both positive and negative terms.
II.
// f(x)
is
bounded and continuous, and otherwise satisfies

in
TT<< TT, while /(TT 0) =/( TT + 0),
bounded and otherwise satisfies Dirichlet's Consame interval, the coefficients in the Fourier's
Dirichlet's Conditions
and
if f'(x) is
ditions in the
Series for f(x) are less in absolute value than K/ri2 where
,
some positive number independent of

In this case
TT, TT)
x=
TT
n.
we can make f(x) continuous in
by giving to it the values /(

and TT respectively.
Then
ira n
=
\
J -If
=i
is
the closed interval
and
Tr-fO)
0) at
/(TT
f(x) cos nx dx
= --1
i f*
~r
f(x) sin
n\_
nx\
_]_.
nj
,n
nx d&
/'(#) s i n
77
f
I
f(x) sin nx dx.
But we have just seen that with the given conditions
ff'(x)
-n
is
of the order l/n.

It follows that
|
/(TT
a^
nx dx
< Kjn\
some positive number independent of n.

similar argument, in which it will be seen that the condition
0)=/( Tr + O) is used, shows that
where
sin
is
Since the terms of this Fourier's Series are of the order

it
follows that
it is
absolutely convergent, and
convergent in any interval.

The above result can be generalised as follows
f(x)
and
its
differential
coefficients,
up
to
bounded, continuous and otherwise satisfy

ditions in the interval
7r<^x<^7r, and
/'V-O),
[r-0,
1, ...
l/7i
also uniformly
If the function
the
th
I)
(p
Dirichlet's
(p-l)],
are
Con-
THE NATURE OF THE CONVERGENCE OF
250
and
th
if the p differential coefficient is
satisfies Dirichlet's Conditions in the
bounded and otherwise

same interval, the coefficients in the Fourier s Series for f(x) will be less in absolute
value than K/n p+1 where
is some positive number inde-
n.
pendent of
into a
(
TT, TT) can be broken up
open partial intervals (-TT, Cj), (c n c 2 ), ... (c m TT), in
each of which /(.>*) and f'(x) are bounded, continuous and otherwise satisfy
Also let f"(x) be bounded and otherwise satisfy
Dirichlet's Conditions in the whole interval.
In this case /(-Tr + 0), ... f(c r 0), ... /(vr-O) and /'(-vr + 0), ...
Again suppose that the interval
105.
number
certain
of
f(c r
...
0),
/'(TT
7m n =
Also
Cl
when r^m.
0) exist,
/(.*) cos
Integrating by parts,
we
nx dx +
'f(x) cos
nx dx -f
. . .
4-
f(x) cos nx dx.
find that
'
(1)
A B =2 sin
where
TT
and
TT^H'
M {f(c
r
f'(<v)
sin
- o) -f(c r + 0)}
nx dx.
J -it
Similarly, from the equation

TT& U
we
Cl
f(x) sin nx dx +
J Cl
'f(x) sin
nx dx +
. . .
f(x) sin nx dx,
Jcm
find that
*-+
where
7T^ rt =
and
7T
U '=
- 2 cos ?tc r {/(c r - 0) -/(c r + 0)} + cos mr\f( - 7T + 0) -/(TT
0)[
f'(x)cQ$nxdx.
In the same
way we
a.',^.-^,
?i
w
'
get
6,,'
where
'
and
7ra u
=-
"=
J
7r6 n
"=
-n
J -IT
COS HC r {f'(cr
f"
/"(*') si' 1
nx
- 0) -/'(Cr + 0)} -H COS 717T {/( - 7T + 0) -/(IT - 0)},
FOURIER'S SERIES
^.^-<,
u"
n
n*
an =
Thus
bn
'
Â n
= Bn +
7r|Z?n |, TT\
'
bn
"
^-^'
~^
This result includes that of

Tz-J.4,,1,
251
104 as a special case.
It is also clear that
are at most equal to the sum of the "jumps"

TT as points
as tne case ma 7 be, including the points x=
A H and
'\
7r|Z?,/|
in f(ji) and /'(*')>

of discontinuity, when /(TT
- 0)
- TT + 0), and
/(
"
"
f(ir 0) +f'(
are of the order l/n.
- TT + 0).
And
under the given conditions a n and b n

The expressions for a n and b n can be used in determining the points at
which the sum of a given Fourier's Series may be discontinuous.
Thus,
we are given the
if
sin
and assume that

series (cf.
x + ^ sin 3# + i
and
(1)
(2)
Hence,
Since this
is
true for
all
need to be multiples of TT
= Q and c 2 c3 ... do not
C!
,
Since this
Thus we
is
cl
satisfied
if
a,,'
= <)=&'.
are the points of discontinuity,
Substituting
. . .
cos mr}l%n.
=(!- cos ?i7r)/2,
c2 ,
...
bn = ( 1
above would be
An=
if c
bx +
see that
an = 0,
So that
sin
a Fourier's Series, and not simply a trigonometrical
it is
we
90),
series
... would
integral values of n, the numbers c l9
interval
lie
in
the
...
since
c
c
-7r<.r<7r,
and,
lt
2
._,,
exist.
in the equation for J5n ,
true for all integral values of n,
see that,
if
the series
is
a Fourier's Series with ordinary discon-
Also, since a n = b u = 0,
we should expect f(.v) to be a constant (say k) in the open interval (0, TT),
and - k in the open interval ( - TT, 0).
tinuities,
As
f(.i-)
has discontinuities at
a matter of fact, the series
is
.v
= and x
'
'
TT.
the Fourier's Series for/(.f), where
#=
J7T
111
f( X ) =
- fr
in
- 7T < .V < 0.
Discussion of a case in which f (x) has an infinity in ( - TT, TT).

have seen that Dirichlet's Conditions include the possibility of f(x)
106.
We
having a certain number of points of
infinite discontinuity in the interval,
subject to the condition that the infinite integral
convergent.
f(x)
dx
is
absolutely
252
Let us suppose that near the point #

/(#)
is
such that
where -7r<^' <7r, the function
r)
<ft(.
(tf-tfo)*'
where
<(#
OO<1
and <(.r) is monotonic

0) do not both vanish.
and
to the right
In this case the condition for absolute convergence

Then, in determining a n and 6 n where
left
of
^'
while
is satisfied.
cos
nx da
and
J-TT
we break up
the interval into

(-TT, a),
(a,
.r
),
(#
/3),
and
(ft
TT),
where (a, ft) is the interval in which /(#) has the given form. In ( - rr, a)
and (ft TT) it is supposed that f(x) is bounded and otherwise satisfies
Dirichlet's Conditions, and we know from
98 that these partial intervals
give to a n and b n contributions of the order l/n.
The remainder of the integral, e.g. in a n is given by the sum of
,
/*<>-
Lt
(X
.'a
rP
&(x)
6->0
-'-codftxcbe
^)
and
these limits being known to exist.

take the second of these integrals,
We
of
<h(x)
^
Lt
6->0
J.r
+ 5 (#
and apply
~ ô)
to it the
Second Theorem
Mean Value.
Thus we have
we
Putting ?t(.r-^ )=y,
But
/"^
cos
^+ -
(?/
w,r,,
yv
Ja
Also when
obtain
b are positive,
are both less than definite
numbers independent of a and i, when 0< v < 1.

may be, and whatever value 8 may have,
Thus, whatever positive integer n

subject to
0<8</3-#
where K'
is
^^^, cos 1IX dx
some positive number independent
of n
and
FOURIER'S SERIES
253
It follows that
Lt
A similar argument applies
to the integral
rx -S
"
d>(#)
--
cos
T
nx d.r.
It thus appears that the coefficient of coan.v in the Fourier's Series for the
1
is some
given function f(x) is less in absolute value than A'/w "", where
positive number independent of n

regard to the coefficient of sin nx.
It
and a corresponding result holds with
easy to modify the above argument so that

ir.
the infinity occurs at
is
when
it
will
apply to the case
The Uniform Convergence of Fourier's
107.
shall deal, first of all,
We
Series.
with the case of the Fourier's Series for
TT, TT) and otherwise satisfies

f(x), when f(x) is bounded in (
Later we shall discuss the case where a
finite
number
of points of infinite discontinuity are admitted.
It is clear that the Fourier's Series for f(x)
convergent in
since
tinuity
any
interval
cannot be uniformly
which contains a point
of discon-
uniform convergence, in the case of series

terms are continuous, involves continuity in the sum.
;
Let f(x) be
satisfy
bounded in
Dirichlet's
the interval
Conditions
in
that
TT, TT),
and
interval.
whose
otherwise
Then
the
Fourier's Series for f(x) converges uniformly to f(x) in any

interval which contains- neither in its interior nor at an end
any point of discontinuity of the function.

As before the bounded function f(x), satisfying
Conditions in
TT,
TT), is
Dirichlet's
defined outside that interval
equation
->_\
<TI)
by the
_ f(Wr
\
./
Then we can express f(x) in any interval e.g. ( 27r, 2x) as

the difference of two functions, which we shall denote by F(x)
and G(x), where F(x) and G(x) are bounded, positive and
monotonic increasing.
where f(x)
is
continuous
They
[
are also continuous at
all
points
36].
Let f(x) be continuous at a and 6* and at all points in

a<X&, where, to begin with, we shall assume 7r<a and
Also let x be any point in (a, b}.
*Thu8/(a+0)=/(a)=/(a-0)
and /(6+0)=/(&)=>/(6-0).
254
Then with the notation

1 f-
S n (x) = ^-
27Tj _ n
/(#)
j
of
95,
sini(2
.
sin J (a;
-ax
x)
m=2
where
sin
Thus
x
,(g)
= -1 f
**
r/
+ o2a)
\
7^(^
ma
m
-da
sin
>
SHI a
TTJ-ATT
ma
da .................... (1)
sma
sin
We
shall
now
discuss the first integral in (1),
ma
"
71
-n,
F(x + 2a)
Ji7r
rr/
+ o2a)
\
J^(ic
f
Let
JUL
sin
da,
JTT
sm m
"
71
da
sin a
f\
TJ/
F(x
2a)
ma
da.
sma
sin
be any number such that
Then
f
ff
sin
ma
-,
(2)
We
can replace ^(^ + 0) by
AT
Now we know
that
f**"
mor 2n + l
is
continuous at
integer.
I^^mx)
Also (F(x + 2a)
any
x.
ma =
da i?r,
sma
sin
an odd positive
Thus
increasing in
is
-.
Jo
since
^(cc), since F(x)
,F(a;)}
interval
is
(3)
boi/nded, positive
and
jA-
is
and monotonic
also bounded, positive
and
monotonic increasing in 0<a=&7r.
*We have replaced the limits -TT, TT bjr -TT + .T, ?r + ar in the integral before
changing the variable from x' to a b} the substitution x' = x + 2a. Cf. 101.
r
FOURIER'S SERIES
we can apply
Therefore
the integral
255
,sinma
-
Mean Value
to
da,
f/>(a)
Jo
where
<f,
(a)
(F(x + 2a)
- J-(a;)
It follows that
juj
where =
/x.
But we know that
sn ma
wl ' t
I
sin a
IJn
-da
7
(91.)
<TT.
Therefore
(4)
/3 =
Finally
where
But,
= ^ JTT.
r
ju
if
--
sma
rsin77ia
where
da =
fx
sm ma da,
sm ma aa + -: If*.
.
7: 1
8lH0J|
sin^)J x
0^=sin
Therefore
'J/i'
(Za
sin a
f*
< m {cosec
1
< m cosec
+ cosec
6.
It follows that
lit'
Sill /X
m sm
/m
some positive number, independent of m, and depending on the upper bound of \f(x) in ( TT, TT).
Combining (3), (4) and (5), we see from (2) that
where
is
If
Trjo
EV
ma
sma
o .sin
F(x + 2a)
-.
da
^(
4/f
(6)
256
similar
argument applies
to the integral
-F(x-2a)-.
'
sin a
,,
f^
but in this case
it
sin?7ia
da,
has to be remembered that F(x
2ot) is
mono-
to JTT.
tonic decreasing as a increases from
The corresponding result for this integral is that
mTr sm
(7)
/x.
K as before being some positive number independent of m, and

depending on the upper bound of f(x) in ( TT, TT).
the same in (6)
Without loss of generality we can take
|
and
(7).
From
(6)
and
(7)
we
obtain at once
,
F(x + 2a)'
Similarly
we
ma da F(x)
sma
sin
'
find that
sn ma
(9)
Thus, from
(1),
in-jr
Now F(x) and

x = a and x = b.
(r(a;)
sin
................. (10)
/u
are continuous in a<^
and
also
when
Thus, to the arbitrary positive number e, there will correspond

a positive number Q (which can be taken less than JTT) such that
/UL
when
/UL
a<x<b.
yu
[Cf.
the same
32.]
/x
serving for
all
values
of
in
FOURIER'S SERIES
we know
Also
to JTT as
/m
Choose
ment
that
/u.
passes from
/i
cosec
/x
increases continuously from unity
to \ir.
as above, less than
of (1) to (10).
257
This
is
|TT,
and put
/tt
= /u
in the argu-
allowable, as the only restriction
was that it must lie between and JTT.

upon
Then the terms on the right-hand side of (10), not including
fj.
IGK/mTrsmiuLQ, are together
less
than 8e for
all
values of x in
(a, b).
it
So far nothing has been said about the number m, except that
is an odd positive integer (2n + I).
Let ?i be the smallest positive integer which satisfies the
inequality
As-
K,
choose
and
JU.Q
Then
i$j{
(i.e.
it
are independent of x, so also

2n 1) so that n n
nQ
We now
follows from (10) that

|
the same
is
nQ
w (#)-/(a)|<9e,
serving for every
In other words,
x in a
we have shown
n>n
when
0>
^ x ^ b.
that the Fourier's Series con-
verges uniformly to f(x), under the given conditions, in

interval (a, b).*
the
If/(-7r + 0)=/(7r-0), we can regard the points

TT,
27r, etc., as points
at which f(x\ extended beyond (-TT, TT) by the equation f(x 4- 2ir)=/(.r), is
continuous, for we can give to/(7r) the common value of/( TT + O) and
* It
may
special case
help the reader to follow the argument of this section

:
E.g. /(a?) =0,
/(*) = !,
Then we have
Interval.
If
0<a
continuous.
c. i
^x^
0<x^
- TT
0,
7r.
if
we take a
258
The argument
which - TT or
of which f(x)
of the preceding section will then apply to the case in

an end-point of the interval (a, b) inside and at the ends
TT
is
is
continuous.
The Uniform Convergence of
108.
argument
Fourier's Series (continued).
By
employed in the preceding section, it can be proved

bounded or not, satisfies Dirichlet's Conditions in the
and/(.r) is bounded in the interval (', I') contained within
similar to that
that when /(#),

interval (-TT, TT),
(-TT, TT), then the Fourier's Series for f(x) converges uniformly in any
interval (a, b) in the interior of (', &'), provided f(x) is continuous in (a, b\
including its end-points.

But, instead of developing the discussion on these lines,
how the question can be treated by Fejer's Arithmetic
and we
we shall now show

Means (cf. 101),
shall prove the following theorem

Letf(x\ bounded or not, satisfy Dirichlet's Conditions in the interval ( -TT, TT),
and let it be continuous at a and b and in (a, 6), where TT a and bîr.
Then the Fourier's Series for f(x) converges uniformly to f(x) in any interval
:
(a +
8,
contained within
8)
Without
(a, b).
we may assume
loss of generality
could be treated as the
sum
of
two such
27r< =
Let
7rrt n
If(of) dx',
Ja
'=
f(x') cos nx' dx
Upl-
1
'
and
7rbn
f(x') sin nx'dx',
Ja
Since f(x)
(a, 6),
is continuous at a and
and we can use the corollary to
the sequence of Arithmetic
Means
'
+2
converges uniformly to f(x) in

|
But/(#)
is
bounded
and
Fejer's
cos
(, 6), it is also bounded in

Theorem ( 101) and assert that
in
nx + bn
'
sin nx)
(a, b).
'
an cos nx + 6 n sin nx
in (a, 6)
'
for the series
(<*>
'
Also
for a greater interval
a^?r,
intervals.
and
^ (a n
'
+ bn
'
)^.
Conditions therein.
Thus we can write

f(x) = F(x) G(x\
where F(x) and G(x) are bounded, positive and
functions in (a, b). It follows that

Mean Value to the integrals
and we deduce at once that

where
in (a,
K
6).
is
(a,/
we can apply
monotonic increasing
+ bn 2 )^ < JT/,
'
some positive number depending on the upper bound
of
|
f(x)
FOURIER'S SERIES
Then we know, from the
corollary to
Theorem
259
102, that the series
II.
QC
OQ'
'
+ 2 ( an cos nx + bn
sin nx)
converges uniformly to f(x) in (, 6).

Let us now suppose x to be any point in the interval (a + 8, 6-8) lying
within the interval (a, 6)..
With
the usual notation
103, that
It follows, as in
QO
-a
')
+ 2 {(n - #') cos nx + (b n - bn
')
sin w.
is
equal to
/(.r)
being defined outside the interval
Now
(say
w)
We
/(.r) is
-TT,
TT)
by the equation
supposed to have not more than a
of infinite discontinuity in (-TT,
TT),
can therefore take intervals 2y 15 2y 2
and
finite
jT
number
of points
\f(x')\dx' converges.
2y m enclosing these points,
the intervals being so small that

I
/(#') dx'
|
<
2e sin $8,
[r
= 1,
2,
m]
being any given positive number.

,
da,
.r,
as already stated, being a point in (a + 8, 6 -8).

a passes from \(x-a) to ^TT, we may meet some or all of the
As
of discontinuity of the given function in/(# 2a).

centres of the corresponding intervals y 1? y 2 , ... y m
Also the smallest value of (x
Thus
v
[ /(.r-2a)
jy,/
a)
is 8.
8in(2M + 1)a <fa

sin a
<-4j,
sm|8
2 sin
<.
Jf
Vr
points
Let these be taken as the
|/(.r-2a)|rfa
w
260
When these intervals,
such of them as occur, have been cut out, the integral
-sma
-
,
ft
o vsin(2/i+l)a
^
da
fix - 2a)
most consist of (ra + 1) separate integrals (lr ). [r=l, 2, ... w + l.J

In each of these integrals (7r ) we can take /(# 2a) as the difference of two
bounded, positive arid monotonic increasing functions
will at
and
F(x-2a)
Then, confining our attention to
see that
sin
/
I
si
,J
G(x-2a).
we
(/,.),
/
(F- #) {cosec 8/2 - (cosec 8/2 - cosec a)} sin (2w + l)a da
where
/!
= cosec 8/2 Fain (2w + 1) a c?u,

/
J2 = cosec 8/2 # sin (2w + 1 ) a rfa,

J
,/3 =
J^=
7^{cosec 8/2
- cosec a} sin (2w + 1
)a da,
G {cosec 8/2 - cosec a} sin (2ra + l)a da.
But we can apply the Second Theorem
Mean Value to each of these

which multiplies sin(2w + l)a is
of
integrals, since the factor in each integrand
monotonic.
IF
It follows that
where
is
Ir
<
cosec J8,
some positive number independent
only on the values of /(.r) in (-TT,

been removed from that interval.
Thus
'
f,
f(x
TT),
- --
+l)a
7- ,
- o2a)vSin(2?z
da
^
(2w + l)> A' cosec ^8.

Since this choice of n is independent of
<
n and
of
when the
,-
.r,
and depending
intervals 2y,, 2y 2
(m-+ l)K
y
.,
v
...
have
+ me
when
to zero as w->oo
Similarly
we
when x
lies in
.r,
the interval
find that
ft
converges uniformly to zero

Thus the series
(a
-a
')
the integral converges uniformly

(
+ S, 6 - 8).
'- da
# + o2a \sin(2w+l)a
i-:
when x
lies in this interval.
+ 2{(a w - a n') cos ^,r + (b n - b n
converges uniformly to zero in (a + 8, 6-8).
')
sin nx\
FOURIER'S SERIES
But we have shown that the
series
+ 2 (n' cos nx + bn
o'
261
sin 72^)
converges uniformly tof(x) in (a, b).

Since the sum of two uniformly convergent series converges uniformly,
see that
a +
cos #+ 6 M sin nx)
we
2K
converges uniformly tof(.v) in (a + 5, 6

Differentiation
109.
seen
that,
and Integration of Fourier's
Series.
We
have
Dirichlet's Conditions, the Fourier's Series
satisfies
uniformly convergent in the interior of any interval in which f(.v)

We may therefore integrate the series term by term within
for/(.v) is
is
when f(x)
-.8).
continuous.
such an interval, and equate the result to the integral of f(x) between the
same limits. And this operation may be repeated any number of times.
With regard to term by term differentiation, a similar statement would

obviously be untrue. We have shown that in certain general cases the order
In the first alternative, it is clear that the series
of the terms is l/n or ]/w 2
.
we would
obtain by term by term differentiation would not converge and

the same remark applies to the other, when second differential coefficients
;
are taken.
This difficulty does not occur in the application of Fourier's Series to the
Conduction of Heat. In this case the terms are multiplied by a factor
2
as it increases the
(e.g. e-KnWt/a" ), which may be called a convergency factor,
rapidity of the convergence of the series, and allows term by term differentiation both with regard to
is
If f(x)
finite
x and
continuous in (-TT,
number of points of
t.
TT)
and f( - TT) =/(TT),
Fourier's Series fop f'(x\ whether it converges or not,

differentiation of that corresponding tof(x).
Let
and
ofj,
&!,..., o
')
while f'(x) having only
an
b^.^be
the
is
Fourier's
given by term
Constants
f'(x).
Then
27ra
2ira
[' f(x)dx,
-n
Tn= J -IT f(-v

/
cos
'= f* f'(x)dx,
J -TT
nx dx,
7ra M'=
f'(x) cos
J -7T
nx dx,
etc.
Integrating the expression for

an
a,/
= nbn
we
by
we
parts,
= - {/(TT) -/( - TT)} cos mr + 7T
TT
since
see that
f(x) sin nx dx'
J-ir
are given /( TT) =/( -TT).
Also a '=0,
()
Similarly, starting with
tt,
infinite discontinuity is absolutely integrally the
&,/
and integrating by
bn
'=-na.
parts,
we
see that
to
term
for f(x)
262
Thus the Fourier's Series for f'(x] is that which we obtain on differentiating
the series for/(^) term by term.
On the other hand, with the same assumptions, except that /(TT)^/( -TT),
the terms in the series for /'(.) are given by
a"
^ l/pn
b n '=
'
na n
J(
Thus the
Fourier's Series for f'(x) does not agree with that which
obtain on term by term differentiation of the series for /(.*)*
More General Theory
110.
in this
chapter and
the'
we
The questions treated

looked at from another point of
of Fourier's Series.
preceding
may be
view.
We may
start with the Fourier's Series

,...(1)
where
27ra
[" f(x')dx',
J -7T
7ran
J-T
f(x')cosnx'dx',
irb n
J -
7T
the only condition imposed at this stage being that,

is
bounded,
integral
it
J -7T
We
for
shall be integrable in (-TT,
f(x} dx
TT),
and,
if
if
the arbitrary function
unbounded, the
infinite
shall be absolutely convergent.
then proceed to examine under what conditions the series (1) converges
X=XQ
difficult to show that the behaviour of the series at x depends only

on the nature of'the function f(x) in the neighbourhood of XQ
Also a number of criteria have been obtained for the convergence of the
It
is
not
# to | [/(# + 0) +/(# ~ 0)]i when these limits exist.t

Further, it has been shown that, if (a, b) be any interval contained in
(
TT, TT), such that f(x) is continuous in (a, 6), including the end-points, the
answer to the question whether the Fourier's Series converges uniformly in (a, 6),
series at
or not, depends only
upon
the nature off(x) in
an
interval
(a',
b'\
which includes
and exceeds it in length by an arbitrarily small amount.

can be shown that, when f(x) has only a finite number of points of
(a, b) in its interior,
Again
infinite
where
it
discontinuity, while
TT
â< ^
/^
TT,
is
f(x) dx converges absolutely, then
f(x) dx,
represented by the convergent series obtained by
*See
also Gibson, Edinburgh, Proc.
j Of.
de
Math.
la Vallee Poussin, lac. cit., T. II.,
Soc., 12, p. 47,
137-143.
et seq.,
1894.
FOURIER'S SERIES
263
integrating the Fourier's Series (1) term by term, no assumption being

regards the convergence of the original Fourier's Series.
The reader who wishes
made
as
to pursue the study of Fourier's Series on these
lines is referred to the treatises of
Hobson and de
la
Vallee Poussin.
KEFERENCES.
BOCHER, loc. cit.
DE LA VALLEE 'PoussiN, loc,
HOBSON, loc. cit., Ch. VII.
LEBESGUE, Lemons sur
Note.
is
e
T. II. (2 ed.), Ch. IV.
les series trigonome'triques,
WHITTAKER AND WATSON,
108
cit.,
loc. cit.,
Ch. III., Paris, 1906.
Ch. IX.
The treatment of the uniform convergence of Fourier's Series in

founded on the proof given by Whittaker and Watson on'pp. 169-170
of the 1915 edition of their
Modern Analysis.
In the 1920 edition the authors have made considerable changes both in
the proof of Fourier's Theorem (9. 42) and in the discussion of the uniform
convergence of Fourier's Series
(9. 44).
CHAPTER IX
PHENOMENON
We
have seen in 104 that, when f(x)

continuous, and otherwise satisfies Dirichlet's
111.
GIBBS'S
IN FOURIER'S SERIES*
is
bounded and
Conditions in
7r<#<7r, /(TT 0) being equal to /( 7r + 0), and f(x) is

bounded and otherwise satisfies Dirichlet's Conditions in the
same interval, the coefficients in the Fourier's Series for f(x) are
of the order l/?i 2 and the series is uniformly convergent in any
,
interval.
In this case the approximation curves
y = 8n (x)
in the interval
TT
<x<
TT
will nearly coincide
with
=/<),
when n is taken large enough.

As an example, let f(x) be the odd function
as follows
The
defined in
?r,
TT)
=_
Fourier's Series for f(x) in this case

sin
^ sin 3# -f ^ sin 5&
is
the Sine Series

...
o"
which
is
uniformly convergent in any interval.
The approximation curves

y = sin x,
y = sin x
y = sn a;
* This
chapter is
o-^
sin ox,
02
F2
founded on a paper by the author in The American Journal
of Mathematics, 39, p. 185, 1917.
264
GIBBS'S
PHENOMENON
IN FOURIER'S SERIES
265
are given in Fig. 30, along with y =f(x), for the interval (0, TT),
it will be seen how
closely the last of these three approaches
and
the
sum
of the series right through the interval.
(3)
FIG. 30.
Again,
let
f(x) be the corresponding even function
i7r<oj<7r.
f(x)= M-aa;),
The
Fourier's Series for f(x) in this case

2
jgT
which
is
2J22
cos 2x
+g
cos
6x+
is
the Cosine Series
cos
10x+ ...
again uniformly convergent in any interval.
r,
26G

y = T\7r
y = --igTr
J cos 2x,
1 COS %X
T\- COS 6x,
are given in Fig. 31, along with y =/(#), for the interval (0, TT),
and again it will be seen how closely the second of these curves
approaches y =f(x) right through the interval.

y
(2)
FIG. 31.
It will be noticed that in
both these examples for large values

nearly agrees with that pf y=f(x),
the slope of y = 8n (x)

except at the corners, corresponding to
of
x=
JTT,
where
f'(x) is
This would lead us to expect that these series

be differentiated term by term, as in fact is the case.
discontinuous.
may
112.
When
the function f(x)
f(\
J(X)
X,
is
given by the equation
_^^ ^
the corresponding Fourier's Series
<
7r<^J<./7r,
is
the Sine Series
^ sin 2# + ^ sin 3# ...}.

x for all values in the open interval
TT.
zero when x =
2{shi
The sum
of this series is
7r<j<7r, and
it is
This series converges uniformly in any interval ( 7r+(5, TT <5)

contained within ( TT, TT) (cf. 107), and in such an interval, by
taking n large enough, we can make the approximation curves
=x
as closely as we please.
it was wrongly believed that, for large values of n,
each approximation curve passed at a steep gradient from the
oscillate
about y
Until 1899
GIBBS'S
point
till
IN FOURIER'S SERIES
a point near ( TT,

x approached the value TT,
TT,
PHENOMENON
0) to
TT),
267
and then oscillated about
when
the curve passed at a

to (TT, 0). And to those who
steep gradient from a point near (TT, TT)

did not properly understand what is
infinite series, the difference
meant by the sum of an

between the approximation curves
y=sn (x),
for large values of n,
siderable difficulty.
and the curve
In Fig. 32, the line y
= x and
y = 2(sin x
2x +
sin
-J-
y=
Lt 8n (x) offered con-
the curve
sin
3x
j-
sin 4x
+ ^ sin 5x)
are drawn, and the diagram might seem to confirm the above
view of the matter namely, that there will be a steep descent
y
FIG. 32.
near one end of the
line, from the point (

TT, 0) to near
and
a
point
TT),
corresponding steep descent near
other end of the line. But it must be remembered that
convergence of this series is slow, and that n = 5 would
(
TT,
count as a large value of n.
the
the
the
not
268
113. In -1899 Gibbs, in a letter to Nature*

pointed out that
the approximation curves for this series do, in fact, behave in
quite a different way at the points of discontinuity
TT in the
He stated, in effect, that the curve y = Sn (x), for large

values of n, falls from the
TT, 0) at a steep gradient to a
point (
sum.
point very nearly at a depth 2 f

Jo
dx below the axis
of x> then
above and below y = x close to this line until x

approaches TT, when it rises to a point very nearly at a height
oscillates
7*
/ dx above the axis of x and then falls rapidly to (TT, 0).

CS1T1
The approximation curves, for large values of n, would thus
approach closely to the continuous curve in Fig. 33, instead of
the straight lines in Fig. 32, for the interval
TT, TT).
FIG. 33.
His statement was not accompanied by any proof.
Though
the remainder of the correspondence, of which his letter formed

a part, attracted considerable attention, this remarkable observation
passed practically unnoticed for several years.
In 190G
Bocher returned to the subject in a memoir f on Fourier's

*
Nature, 59,
Series,
p. 606, 1899.
\Ann. Math., Princeton, N.J. (Ser. 2), 7, 1906.

See also a recent paper by Bocher in /. Math., Berlin, 144, 1914, entitled
" On Gibbs's Phenomenon."
Reference should also be made to Runge, Theorie u. Praxis der Reihen,
certain series is there discussed, and the nature of
pp. 170-180, Leipzig, 1904.
the jump in the approximation curves described ; but no reference is made to
Gibbs, and the example seems to have been regarded as quite an isolated one.
GIBBS'S
PHENOMENON
IN FOURIER'S SERIES
269
and greatly extended Gibbs's result. He showed, among other

things, that the phenomenon which Gibbs had observed in the
case of this particular Fourier's Series holds in general at ordinary
*
points of discontinuity. To quote his own words
:
If f(x) has the period 2?r and in any finite interval has no
discontinuities other than a finite number of finite jumps, and
if it has
a derivative which in any
finite interval has
no
dis-
continuities other than a finite number of finite discontinuities,

then as n becomes infinite the approximation curve y Sn (x]
approaches uniformly the continuous curve

(a) the discontinuous curve y =f(x),
(b)
an
infinite
parallel to the axis
made up
of
number of straight lines of finite lengths

of y and passing through the points a lt a 2 ...
,
the axis of x where the discontinuities off(x) occur.

If a is
the
line
between
the
in
extends
one
these
points,
question
of
lany
on
two points whose ordinates are
DP,
DP,
ichere
is the
magnitude of the jump in f(x) at
a,t
and
Ct,+2TT
FIQ. 34.
This theorem
is
the jumps at a lt
a.2
illustrated in Fig. 34, where the amounts of

Until
are respectively negative and positive.
loc. cit., p. 131.
He.
270
Gibbs made his remark, it was supposed that the vertical lines
extend merely between the points whose ordinates are/(a0).
The series in the examples of Chapter VII. converge so slowly that
the approximation curves of Figs. 21-23, 27, 28 are of little use in this
connection.
The series on which Bocher founded his demonstration of

and other extensions of Gibbs's theorem is
114.
this
sin
x+
which, in the interval

as follows
sin 2x
(0, %TT)
represents the function f(x) defined
y(0) =/(2ir)
+ ^ sin 3x +
= 0,
f(x)=(7T-x),
In this case
8n (x) = sin x + ~ sin 2x +...+
Jx (cos
+ cos 2a +
nx
sin
+ cos no.) da
smja
maxima and minima
The properties of the

of S n (x) are not so
easy to obtain,* nor are they so useful in the argument, as those of
Sm
In his memoir Bocher dealt with the
maxima and minima of

and he called attention more than once to the fact that
the height of a wave from the curve y=f(x) was measured
Rn (x),
This point has

parallel to the axis of y.
in some expositions of his work.
been
lost sight of
In the discussion which follows, the series

2(sinaj-f isin3^
is
employed.
In the interval
as follows
TT,
TT) it
represents the function f(x), defined
/( TT) =/(0) =/(TT) = 0,

f(x)=-7T,
+ ism5^+...)
-7T<a,'<0,
/(*0 = |7r, 0<><7r.
Gronwall has discussed the properties of Sn (x) for this series, and deduced
Phenomenon for the first wave, and some other results. Cf. Math.
Ann., Leipzig, 72, 1912. Also Jackson, Palermo, Rend. Circ. mat., 32, 1911.
Gibbs's
PHENOMENON
GIBBS'S
The sum
of the first
A number
=
y Sn (x) are
IN FOURIER'S SERIES
terms of the series
is
denoted by
271
Sn (x),
of interesting properties of the turning points of
obtained by quite simple methods, and all the

features of Gibbs's Phenomenon, as described by Bocher, follow
from these properties.

of course, not an unusual thing that the approximation
curves y Sn (x), for a series whose terms are continuous functions
of x, differ considerably, even for large values of n, from the
curve y= Lt Sn (x) = S(x). They certainly do so in the neigh-
-at once
It
is,
n >oo
bourhood of a point where the sum of the series is discontinuous,

and they also may do so in the neighbourhood of a point where
the sum is continuous. This question we have already examined
in Chapter III.
But, in passing, it may be remarked that when
f(x) satisfies Dirichlet's Conditions, the approximation curves for
the corresponding Fourier's Series will, for large values of n,
differ only slightly from the curve y =/(#) in the neighbourhood
of points at which f(x) is continuous, for we have shown that the
Fourier's Series converges uniformly in any interval contained
within an interval in which f(x)
is
continuous.
The existence of maxima (or minima) of Sn (x), the abscissae of

which tend towards a as n increases (a being a value of x for
which the sum of the series is discontinuous), while their
ordinates remain at a finite distance from /(a + 0) and f(a 0), is
the chief feature of Gibbs's Phenomenon in Fourier's Series. And
most remarkable that its occurrence in Fourier's Series
remained undiscovered till so recent a date.*
it is
THE TRIGONOMETRICAL SUM

.
2
115.
If
we
define f(x)
by the equations
-7r<o;<
*Cf. also
Weyl
(i)
"Die Gibbs'sche Erscheinung
Funktionen," Palermo, Rend. Circ. mat., 29, 1910 ;

Erscheinung und verwandte Konvergenzphanomene,"
in der Theorie der

(ii)
Kugel-
" Uber die Gibbs'sche
ibid., 30, 1910.
272
- ?r ^ x ^ TT
the Fourier's Series for f(x) in the interval
As
stated above, we denote

including the term in sin (2n - 1
by Sn (x)
Then
this
sum up
Since
We
8n(x)
an odd function, we need only consider the
is
<# <
TT.
proceed to obtain the properties of the
Since, for
maxima and minima
Sn (x).
any integer m,
+ x ) = sin (2m
f
sin
it
and
sin a
of this function
I.
to
)x.
interval
is
1 ) ( JTT
(2m
follows from the series that
and when x =
and x = TT
S n(x)
is
1) ( JTT
x'\
symmetrical about X =
^TT,
it is zero.
When < x < TT, $n(#) w positive.

From (I) we need only consider < a? ^ Jw.
II.
f*sin2na a==
în^"^
sin a
J
J
2^
/^J
sin a
We
have
sn
272-
The denominator
increases in the
successive
in the integrand is positive and continually

of integration.
By considering the
waves
interval
in the
graph of sin a cosec
--,
the last of which
2t)ii
or
may
not be completed,
may
it
is
clear that the integral
positive.
III.
The turning points of y = Sn (x) are given by
TT
->
We
= 2?r
^(n-i)
= n-
\,...
T (numma).
have
31
=
y
J
f
I
The
2?i-l
3?r
TT
sin
2na
sm
result follows at once.
da,
and
dy = sm
-^
sin x
s
dx
:
is
GIBBS'S
PHENOMENON
IN FOURIER'S SERIES
273
As we proceed from x = to X = ^TT, the heights of the

maxima continually diminish, and the heights of the minima
IV.
continually increase,
being kept jixed.
/--
2-fl
2-
The curve y = Sn (x\ when n =

FIG. 35.
maxima in the interval

/2m + l \
and Sn
TT), m being
Consider two consecutive
/2m
namely,
1
TT
integer less than or equal to |(n
/2m
/2m+l
We
1).
\ _
rC 2 " 1
Tr 2 " '
1
sing
r(2+i)
JTT,
a positive
have
- 1 )'
sin a
sin
<x=
si
na
C
**"
HJ^-^'sin^'
v
sin^-
The denominator
in both integrands is
positive and it con1
a
tinually increases in the interval (2m
) TT
(2m 1) TT ;
also the numerator in the first is
continually negative and
^ ~
the second continually positive, the absolute values for

elements at equal distances from (2m I)TT and 2m?r
being the
same,
in
c. i
274
Thus the result follows.

to examine the sign of
where
is
a positive integer
maximum
V. The first
its
Similarly for the minima,
less
than or equal to \n.

of x =
to the right
height continiudly diminishes as
to infinity, its limit is
we have
is at
x=
When n
increases.
and
tends
/-,
nx
dx.
Jo
The curves
when % = 1,
y=Sn (x\
2, 3, 4,
and
6.
FIG. 36.
We
have
f 2^
2na
If*.
-aa ~^~\
sm a cosec
sin
^
2?*
Thus
271
.
sin
in a
/I
I
\2?i
sma
TT- aa.
2?i
+2
a
\
-- =---s cosec s--
a
cosec 5
2-i
2n
<ia.
2?i
2/
Since a/sin a continually increases from 1 to oo as a passes

to TT, it is clear that in the interval with which we have
,
from
to deal
27i
COSeC
27i
7;
COSeC
rr
rr
> 0.
^
GIBBS'S
PHENOMENON
IN FOURIER'S SERIES
Sn (x)
is
when
Sn (
275
Thus
But, from
(I),
It follows that
The value
positive
tends to a limit as
by the method used by
of this limit can be obtained
Sm
2
smja
Bocher for the integral
da*
Jo
tends to infinity.
But
is
it
readily
obtained from the definition of the Definite Integral.

2
I?
'
3-7T
sm
271271
(27l-l)7T
/sm77it ;
= 22
;
7*1=1
--T'
4
^
m=
W*A
8nA=f
/i^
Therefore
= 2^ f* sin -da?
a;
=27i
VI.
The
The r
sin
a?
a?
f T^n -cto.
7
-tZa;=
a?
Jo
wave
result obtained in (V) fop the first
case of the following

th
f*
05
is
a special
maximum
the right of
to
x=
is at
x 2r -
^
= 2r
^
1
TT,
Ail/
height continually diminishes as n increases, r being

When n tends to infinity, its limit is
kept constant.
1 )sin 0) ,
and
its
p-
-dx,
Jo
which
is greater
The rth
than
minimum
i TT.
to the right
height continually increases as
of x =
is at
man ^.
these theorems
sin a
/ 1
(
\Zn
consider
a
cosec ~
*Ann. Math., Princeton, N.J.

p. 649.
we
Tr,
and
its
first
irt
dx,
which
is less
the integral
r^ cosec
Zn
(Ser.
m TI
J*
" IT
/v
increases, r being kept constant.

1
To prove
x 2r =
2),
7,
p.
124.
Also Hobson,
loc.
cit.,
276
being a positive integer less than or equal to
^<
"'2ti
<^ 7r * U
^ ne
F(a) =
Then
in this interval.
n t erva l
1,
so that
integration.
cosec
2
2?i
2?i
271
^
+
2
cosec =-^-5
2+.2
>
(Of. (V).)
Further,
"
COS
= a~2 {02 cos
cosec 2 (j>-\js2 cos
= a/(2n + 2)
where
and
But
And
COSec2
\/s
2S
\/r
cosec2
CO8
CO8ec2
<
i/^},
= a/2n.
2
-(0 cos
cosec 2 0)
the right-hand side of the equation will be seen to be

0<<<i7r and the lower
negative, choosing the upper signs for

for J7T<0<7T.
2
cosec 2 0- diminishes as
to TT.
increases from
from the expression for ^(a), that ^ (a)>0,.and
F(a) increases with a in the interval of integration.
Therefore
<
cos

It follows,
The curve
= sin
cosec
<*
cosec
<x
'
thus consists of a succession of waves of length TT, alternately

above and below the axis, and the absolute values of the ordinates at points at the same distance from the beginning of each
wave continually
It
increase.
follows that,
when
is
7/1
equal to
2, 4,
...
,2(n
1),
the
integral
l
is
negative
*
.
and,
/ 1
sm a s- cosec
when
,r-
is
- ^ 1-^ cosec
equal to
1, 3, ...,
2n
1,
this integral
is positive.
th
Returning to the maxima and minima, we have, for the r
=
maximum to the right of x 0,
PHENOMENON
GIBBS'S
IN FOURIER'S SERIES
277
-J-tfn+ifcsr-i)
Zr-l
2r-l
f
an
sin2rca
sm
f 2(n+i)
da
sm
/I
--
-a
a( =r- cosec
(<*-*>-.
sin
I
--'da
un2'(n4l)a
^
a
--
\,
=
^
s aa.
tn 2n
2
\2n
2?i + 2/
+ ^cosec
Therefore, from the above argument, Sn (x 2r _ ) ^> Sn+l (x.2r _ ).
Also for the rth minimum to the right of x = 0, we have
)
and Sn (x 2r )<Sn+l (x.

By an argument similar
2l .).
to that at the close of (V)
cr
Lt
*^
JO
7l->oo
we have
r*si ux
j *
- dx*
Sn (x r =
TJ.
It is clear that these limiting values are all greater than JTT for
the maxima, and positive and less than JTT for the minima.
GIBBS'S
PHENOMENON FOR THE SERIES
2 (sin
116.
Gibbs's
-J-
sin
3x+
-sin
ox+ ...).
of
115
From
the Theorems I.-VI.

Phenomenon for the series
2 (sin x
the features of
all
+ ^ sin 3x + \ sin 5x -f ...).,.,
TT
^x^
TT,
follow immediately.
It is obvious that
^x^
we need only examine the interval

and that a discontinuity occurs at x = 0.
For large values of n, the
where
(#)
=2
^sin
at a steep gradient
x+
sin So?
. . .
from the origin to
very near, but above, the point
0,
si n
its first
C w sin
/
is
+ ~^T
T
( 2/?1
~ !)
TT,
#)
maximum, which
\
dx (115,
j
V.).
The
curve, then, falls at a steep gradient, without reaching the axis

T
of x( 115, II.), to its first minimum, w hich is very near, but
0,
x dx](
(f^'sin
Jo
I
*For the
p. 129.
values of
/'
}dx,
see
115, VI.).
It
then oscillates
Ann. Math., Princeton, N.J.
(Ser. 2), 7,
278
y = %7r, the heights (and depths) of

we proceed from x = to
X = ^TT ( 115, IV.); and from x\ir to X -TT, the procedure is
^ x ^ TT being symmetrical
reversed, the curve in the interval
about x = JTT ( 115, I.).
above and below the
line
the waves continually diminishing as
th
wave to the right of
(or lowest) point of the r
will, for large values of n, be at a point whose abscissa is
The highest
x=
T'TT
115, III.)
and whose ordinate

sin
a?
By
increasing
close as
we
dx
is
very nearly
115, VI).
^ x 5= TT
the curve for
sin
sin
^T. <><J
We may
(i)
can be brought as
please to the lines
state these results
If e is
any
a positive integer
more
positive number,
dx.
definitely as follows
as small as
we
please, there is
such that
|j7T->Sfn(^)|<e,
when n^v, C^X^^TT.
This follows from the uniform convergence of the Fourier's

Series for f(x), as defined in the beginning of this section, in an
PHENOMENON
GIBBS'S
interval
which does not
Qj
first
maximum to
dx
as
tends to infinity, there
is
sin
Let v" be the integer next greater .than
than
/y
positive integer v" such that
is less
the right of x
-*-\
JTTo
maximum
abscissa of the first
an
107).
(cf.
Since the height of the
(iii)
279
include, either in its interior or at
end, a discontinuity of f(x)
(ii)
IN FOURIER'S SERIES
=-.
Then the
# = 0, when n^v",
to the right of
e.
It follows
from
and (iii) that, if v is the greatest of the

and v", the curve y = Sn (x), when n = v,
(i), (ii)
positive integers v, v"

behaves as follows
:
It rises at a steep gradient
which
is
above
Jo
Vy
from the origin to
its first
dx and within the rectangle
^*" Cj
^^ rf ^^^
JLs
f)
V/
^ ^
,, .
^^^
(/
^.
sin
\
I
Jo
xJ
Cvw
|^
After leaving this rectangle, in which there may be

tions about y = JTT, it remains within the rectangle
eO<7r
Finally,
it
maximum,
e,
many oscilla-
JTT
enters the rectangle
f sin x
Jo
and the procedure
The
in the first region is repeated.*
cosine series
-
which represents
cos
x - = cos 3x +
in the interval
be treated in the same
way
0^.x<^
cos
and
5x+
|
...
in the interval
as the series discussed in this article.
^<rc^7r,
can
280
GIBBS 's PHENOMENON FOR FOURIER'S SERIES IN GENERAL.

Let f(x) be a function with an ordinary discontinuity
satisfies Dirichlet's Conditions in the interval
117.
when x = a, which
Denote as usual by /(a-fO) and f(a 0) the values towards

which f(x) tends as x approaches a from the right or left. It
will be convenient to consider f(a + Q) as greater than f(a
0) in
the description of the curve, but this restriction is in no way
necessary.
Let
tx-a =
Then
(f)(x
(f>(x
2&n2r
a)=
a}=
JTT,
0(4-0)=
0(0)=
Now
and
0(-0)=
J-TT,
and
<j)(x)
put
let /(a),
Then
at x
when
when
\ir,
\/s(a
when x = a, be
defined as
+ Q)=\ls(a
= \ls(a) = 0, and
Q)
M/( a + 0)+/(a
^(ft)
is
0)}.
continuous
= a.
The following distinct steps in the argument are numbered for

the sake of clearness
:
(i)
Since
\[s(x) is
continuous at x = a and
positive number, as small as we
that
IÔ)I<4>
If ^ is not originally less
please, a
when |-a| =
than
c,
we can
\[s(a)
number
= Q,
r\
if e is
exists such
i?.
choose this part of
tj
for our interval.
can be expanded in a Fourier's Series,* this series

contained
x
being uniformly convergent in an interval a = ^ /3
within an interval which includes neither within it nor as an
(ii)
i/r(aj)
end -point any other discontinuity of f(x) and ^(x

x = a (cf. 108).
*If f(x) satisfies Dirichlet's Conditions,
(j)(x-a) that \l/(x) does so also.
it
is
clear
a) than
from the definition of
PHENOMENON
GIBBS'S
IN FOURIER'S SERIES
281
Let 8n (x), n (x a} and <rn (x) be the sums of the terms up to

and including those in sini^ and cosnx in the Fourier's Series
(j)
for f(x), (J)(x a)

of (i), as small as
and
Then
\!s(x).
please, there exists a positive integer
ffn\
x ) ~~ Y\ X )
that
the same
^ 74
serving for every x in a
i/
being the positive number

v such
we
when n
^ ^
/3.
<
/3,
oj
i/ ,
Also
in
ic
|
a ^
Now
|
(iii)
right of
ic
if
=a
if
tj,
<a
even, the
is
at a
is
<a<a
r\
+ -;
first
and
4-
and n >
maximum
if
>/
is
in
n (o;
a) to the
at a-\
it is
odd,
</>
/'.
In
=..
either case there exists a positive integer v" such that the height
maximum
of the first
f'sinic
Jo
(iv)
a+
tj,
between
lies
ax and f'sino;
,
Jo
This
first
TTC
dx+^-r-f-,
maximum
<
provided that
Let v" be the
first
will
have
7^ when n ^ v
-&
TTT
2{/(aH-0)
its
/(a
0)}
abscissa between a
and
>;.
which
positive integer
satisfies this in-
equality.
+ < ^
x
s n (x) converges uniformly
(v) In the interval a
tj
/3,
to f(x).
Therefore a positive integer y(iv) exists such that, when
\f(x)-s n (x)\<
the same
i/
lv)
Now, from
serving for every x in this interval.

the equation defining \/s(x),
(a
It follows
greater than
+ 0)-/(a-0)
from
(i)-(v) that if
v, v"
v" and
i/
the interval
iv)
,
t/
is
the
the curve
first
a<x^/3, behaves as follows

When x = a, it passes through a point whose
j of
J (/(a
+ 0) +/(-
a point within
positive integer
in
/,
y = sn (x), when TI^
0)),
ordinate
within
and ascends at a steep gradient
of
{/(a+ o )+ /( a -o)}
is
r
+ /i^tfco>
7T
Jo
^^
^
to
THE APPROXIMATION CURVES
282
This
may
be written
and, from Bocher's table, referred to in
we have
115,
then oscillates about y = f(x) till x reaches a + rj, the character

of the waves being determined by the function
n (x
a), since
It
<j>
the term
o-n
(x)
only adds a quantity
passing beyond x = a +
And on
less
within, the strip of width 2e enclosing
the other side of the point
to the ordinate.
the curve enters, and remains
>;,
x = /3.
On
than
yf(x)
a a similar
from x = a+t] to
set of circumstances
can be established.
D=f(a + 0)-f(a-0),
Writing
first
wave
D
where P,
and right
to the left
sin x
J,
the crest (or hollow) of the

a tends to a height
of x
ax.
REFERENCES.
BOCHER,
"
Introduction to the Theory of Fourier's Series," Ann. Math.,
Princeton, N.J. (Ser.
2), 7,
1906.
GIBBS, Scientific Papers, Vol.
II., p.
And
in the text.
the other papers
Note.
named
London, 1906.
Y
w = l,
258,
2r-l
6, 11 and 16 can be obtained as lantern slides from the firm of

Martin Schilling, Berlin. The last of these illustrates Gibb's Phenomenon
even more clearly than Fig. 35 above.
for
CHAPTER X
FOURIER'S INTEGRALS
118. When the arbitrary function f(x) satisfies Dirichlet's
Conditions in the interval ( I, 1), we have seen in 98 that the
sum
of the series
7
(aO^+jSf' f(x')cos ^(x-x)dx

%[f(x + 0)+f(x 0)] a ^ every point in
.........
(1)
is
equal to
x=l
where /(#+0) and f (x 0) exist; and that at

is i[/(-^ + )+./V- () )]> when these limits exist.
Corresponding results were found for the series
1
^J
ft
>\
>
2
,
n7r
n-rr
*\
IV n
^V~\
/,
yVsin-y-ic
I
J
fr^
I
in the interval (0,
..
IV H
/(a;) sin -7-05
-I
its
sum
f(x)dx+^coti-j-x] f(x)cos-j-xdx,
and
l<x<l
.........
(2)
dx,
n\
(3)
I).
Fourier's Integrals are definite integrals

arbitrary function in an unlimited interval.
which represent the
They
are suggested,
but not established, by the forms these series appear to take as

tends to infinity.
If
is
please,
taken large enough,
7r/l
and we may neglect the
assuming that
i..
f(x)
dx
is
may
we
be made as small as
first
convergent.
term in the
series (1),
Then we may write
717T
cos
ag
^L
Aa
f(x') cos
Aa (x'
x) dx'
+ Aa
r
J -i
-I
where Aa = 7r/.
283
A a (x'
cos 2
x) dx' -f-
>
FOURIER'S INTEGRALS
284
this sum has a limit as l->x> an

Assuming
assumption
which, of course, would have to be defended if the proof were to
be regarded as in any degree complete, its value would be
that
da
TTJo
and
f(x) cos a (x
J-co
x)
dx,
would follow that
it
da I
J-
TTJo
-00<><00,
when
these limits exist.
way we
In the same
are led to the Cosine Integral and Sine

and the Sine Series
Integral corresponding to the Cosine Series

2
00
aa
f(
"
da
O
when
si
ax sin axdx!
f(x) sin
Jo
these limits exist.
must be remembered that the above argument

a proof of any of these results. All that it does
It
not
is
is-
to
suggest the possibility of representing an arbitrary function /(#),

given for all values of x, or for all positive values of x, by these
integrals.
shall
We
now show that this representation is possible,

out
in
our proof the limitation the discussion imposes
pointing
the
upon
arbitrary function.
119.
of
Let the arbitrary function f(x), defined for all values
x, satisfy Dirichlet's
in addition
let
J -00
f(x)
Conditions in any finite interval,
dx
and
Then
i
r 30
da
TfJo
J -oo
f(x) cos a (x' x) dx
= \ [ f(x + 0) +f(x - 0)],
at every point where f(x

0) exist.
0) and f(x
fixed
the
value of x for which
Having
upon
we wish
to
evaluate the integral, we can choose a positive number a greater

than x such that f(x) is bounded in the interval a = x^==b,
FOURIER'S INTEGRALS
where
b is arbitrary,
|00
and
285
\f(x)\dx converges, since with our
Jri
definition of the infinite integral only a finite

of infinite discontinuity were admitted ( 60).
f*
It follows that
number
of points
x) dx'
f(x') cos a (x'
Ja
converges uniformly for every a, so that this integral

tinuous function of a ( 83, 84).
Therefore
by
a con-
x)
da
exists.
JO
Also,
x)dx'
da\ f(x')cosa(x'
is
85,
50
ftf
da
But
'f
f(x') cos a (x'
Jo
x'
x) dx'
xâ #>0
And
Jn
in
/(^
f2
dx' \ f(x') cos a (x
-a;
do/ also converges.

(a?/
</v
we have chosen
since
"
f /(a? 8myaj'-as
It follows that
Jo'
dx converges.
il
a'
Ja
x'â,
Therefore
a?)
<&'
Ja
converges uniformly for every q.

Thus, to the arbitrary positive number
positive number A>a, such that
the same
serving for every value of

94 that
there corresponds a
q.
But we know from
Lt
*
,,
Lt
f(u + x)
smqu'.,
du
5~>oo
= 0,
since
(a
x,
f(u + x)
x),
satisfies
Dirichlet's
and both these numbers are
positive.
FOURIER'S INTEGRALS
286
Thus we can choose the
number Q
positive
W^S^'^^Ki
and
It follows 'from
(2)
when q^Q.
^
Lt
rm
Urns
da
J
But,
by
da
â
,,
/(')
sin
--
<7
#)
(#'
J^- JC
-'do:'
= 0,
JU
/(V) cos a (x'- a)cfcc' =
................... (4)
87,
I
Jx
Jo
poo
q^Q .......... (3)
(3) that
-,
Ja
when
*0
*}
so that
/(a/) cos a (x
=
x) dx'
d^'
Jx
/{a;')
cos a (x
a;)
cZa
Jo
a~x
J
Letting g->oo
f
JQ
we have
+ O),
dafVÔcosaK-^^^S/^
J*
exists.
Adding
and
(4)
JQ
when f(x + 0)
(5),
............
(5)
we have
f(x')eosa(x'-x)dx'
da\
Jx
= lf(x + 0),
............ (6)
exists.
Similarly, under the given conditions,
f
J
when f(x
Adding
7T f
JQ
da\J-<
.......... (7)
0) exists.
(6)
da f
and
J -co
(7),
we
obtain Fourier's Integral in the form
f(x) cos a (x
for every point in

exist.
f(x')cosa(x-x)dx=~f(x-0),
6
oq
- x) dx' = \ [f(x + 0) +f(x - 0)]
< x < oo
where f(x + 0) and f(x
0)
FOURIER'S INTEGRALS
More general conditions
120.
287
In this section we shall show
for f(x).
that Fourier's Integral formula holds if the conditions in any finite interval
remain as before, and the absolute convergence of the infinite integral
I
J-,
f(%) dx
replaced by the following conditions
is
For some positive number and to the right of it, f(x') is bounded and monoand Lt /(#') =
and for some negative number and to the left of it,f(x')
tonic
bounded and monotonic and Lt
is
/(.?/)
().
Having fixed upon the value of x for which we wish to evaluate the
we can choose a positive number a greater than .r, such that/(.r')
bounded and monotonic when x'â and Lt f(x') = 0.
integral,
is
vao
x'
/oo
Consider th e integral
By
/(.*')
Mean
- x)dx' =f(A'}
/(.r')cos a(.r'
a< 4'
where
cos a (of
Value,
u(^ - .r) dx' +f(A")
cos
- x) dx'.
"cos a(x'
- x~)dx,
^^
Thus
'A'
Therefore
"A"
''"
JA'
But we are given that Lt

It follows that
Jd
f(x') cos a.(x
uniformly convergent for a
is
tinuous function of a in
Also,
i
by
da
Ja
But
than
x'
>0, and
this integral represents a con-
a*=9 u .*
85,
f)o
^g
x) dx'
f(.v')
cos a(x'
- .r) dx =
xâ-x>Q in
r
I
Ja
00
dx'
/"2
I
f(x) cos a(x' - x) da
Jq n
the interval
x'â,
since
we have chosen a greater
x.
A,,d
both converge.
* These extensions are due to Pringsheim. Cf. Math. Ann.,
Leipzig, 68, p. 367,
Reference should also be made to a paper by
1910, and 71, p. 289, 1911.
W. H. Young in Edinburgh, Proc. R. Soc., 31, p. 559, 1911.
288
FOURIER'S INTEGRALS
Therefore, from
Now
(1),
consider the integral
From
the Second
Theorem
of
Mean
Value,
^,
where a
.....
(3).
< A' ^Â".

sin
.Also
/*
'<*-*>
?o(y -*) rfy = f
j'-x
-X
JA'
!LL dUf
U
the limits of the integral being both positive.
~ *>
'
sin
Therefore
And
similarly,
Thus, from
(3),
^"?^&(_=.)
<,.
<7r
rf
(Cf.
91.)
*
............. ........ (4)
It follows that
is
an(^
84
by
it is
/M Ea^z) d y =
Lt
Thus
since the integral vanishes
Also from
<?o
when
<?
continuous in this range.

0|
,...(5)
=0.
(2),
(6)
But we have already shown that the

is
integral on the right-hand side of (6)
q^ 0.

Proceeding as in 119, (2)
and
Lt
q
Thus, from
(6),
/"*
.'0
> oo Ja
dal
(3),* it follows that
/(
-V
/(.r')cos a(.v'
- .v) d.v =
......................... (7)
Ja
Or we might use the Second Theorem
Infinite Integral.
&
of
Mean Value
as proved in
58 for the
FOURIER'S INTEGRALS
But we know from
289
119 that
,
................... (8)
when
this limit exists.
L (*
Thus
TT JO
when
daC f(x')co*a(x'-x}dx' = %f(x + Q),
.................. (9)
JX
this limit exists.
Similarly,
under the given conditions, we find that

(10)
when
this limit exists.
Adding
/(#0)
and
(9)
our formula
(10),
proved for every point at which
is
exist.
Other conditions for
121.
We
f(x).
now show
shall
Integral formula also holds when the conditions at

section are replaced by the following
00
that Fourier's
of the previous
For some positive numler and to the right of it, and for some negative
number and to the left of it, f(x') is of the form g(x')cos(Xx' + /x),
where g(x') is bounded and monotonic in these intervals and has the
(I.)
limit zero as
x^
1
oo
Also, (11.)
and
dx' converge.
We have shown in 120 that when g(x) satisfies the conditions named
above in (I.), there will be a positive number a greater than x such that
But,
if
is
any positive number,

f
JQ
r\
=
=i
=
Ja
,-co
g(x } cos a (x' - .v) dx' +
r 00
rA
da
J-X
da
g(x') cos
Ja
*v>
a.(x'
f<n
J\
- x) dx' +
da
C^
Ja
q(x') cos (a
'
da
+ X.)(x' - x) dx' + i
Ja
Jo
/-CO
o?a
g(x') cos a(x'
- x) dx'
g(x')cosa(x' -x)dx'
-co
Therefore
refore
/-00
- ,v) dx'
da
-A
J\
Ja
q (x') cos a (x'
'
,,cc
JQ
g (.^' ) cos a (#'
c. i
da
Ja
g(x')coaa(x'-x)dx' + %l
Ja
7-^-X
=i
da I g(x')cosa(x'-x)dx'
Ja
*<
da
J/
-x
g(x') cos a.(x'
.r ) c?r'
/*
/*
7o
Ja
-co
da
Ja
<7(^')cos (a
A)(.r'
- x)dx'.
FOURIER'S INTEGRALS
290
Again we know, from
$ 120, that
00
f
Ja
is
uniformly convergent for a

i'î
It follows that
0.
I"'*
da
g(.v')sin a(x'
Ja
JA
exists, for Aj
= AO >
-A
> A > 0.
Also
da
Ja
J\o
g (x' ) sin a (x
x) dx'
l
dx' ( g(x') sin a(x'
- x) da
(by
84)
since both integrals converge.
/""
converges
and
follows that
it
-v'
^LJ
-
Ja
also converges, so that
is
we know
uniformly convergent for A
Thus
rgw
Lt
that
from
(2) that,
r da f,(.r-)8iu
JO
a(.r'
and therefore continuous.
0,
**(*-*)*<= rîdx'.
-x
x'
AO-Ô Ja
It follows
-,)<W-
Jflt
we
find that,
Therefore, from (2) and
'^
^'-
(~
Jd
and
T
da
Ja
!*
JO
d*.
(3)
'^
(3),
X
-*
(*)"***^
with the conditions imposed upon g(s!\*
00
JA.
x'-x
Ja
when A>0,
J<?
Also, as before,
o?.r'
.^
.t?'
/"
g(x')sma(x'-x)dx'=\
Ja
/\
g(x')
COS
MX'-X)> dx
j
......... (4)
................. (5)
daT q(x')sinM-x)dx'= ("$$&-&*
& v
'
Ja
-rt
can be obtained at once from the Second Theorem of Mean Value, as

58 for the Infinite Integral ; but it is easy to establish the result, as
proved
in
119, without this theorem.
* This
in
FOURIER'S INTEGRALS
r
rA.
Again, since
we have
exists,
da
g (x ) sin a (x - x) dx'
r
da
291
00
g(x')$ii\a(x'
-x)dx' =
(6)
-CO
da
/.CO
/.CO
- x) dx',
g(x') sin a(x'
/CO
tnat
it
follows
/.OO
,-CO
^
rfa
c?a/
^(.^')sin a(.r'-^)o?.r'
</(#')
sin a(x'
x)dx'
Q.
Thus
(x')$\v(a- \)(x' -x)dx' = 0.
Therefore
da f g(x')siu \(x' -x)cosa(.v' -x)dx' = Q.
Multiply
(1)
,...(7)
Ja
Jo
o
by cos(A.r + /x) and
(7)
sin(A.r + /i)
by
and
subtract.
It follows that
/.CO
y.%
And
in the
c?a|^
same way, with the conditions imposed upon

/
These
#(y)cos(Ay + /x)cosa(.r'-#)cfo/ =
g(.v')cos(\x'
daj
and
results, (8)
(9),
may be
written
Ja
da
JO
..................... (10)
/"-'
J-
= 0,
/(.r')cosa(.r'-.r)rf.r'
co
when /(.r')= g(x') cos (A.r' + /x) in

But we know that, when/(.r)
(a, oo
and
oo
-a').
Conditions in
.Q)l
when
these limits exist.
and
we have
+ ij,)cosa(x'-x)dx' = Q ............... (9)
g(x'),
da P/(.r')cosa(.r'-.r)rfy = 0, }
J0
r
................ (8)
119
[Cf.
-a',
),
......... (11)
(5).]
we
see that Fourier's Integral formula holds, when

the arbitrary function satisfies the conditions imposed upon it in this section.
It is clear that the results just established still hold if we
replace
cos(X.r + u) in (I.) by the sum of a number of terms of the type
Adding
(10)
It can be
by an
proved
(11),
* that the theorem
infinite series
is
also valid
when
this
sum
is
replaced
2 an cos (A.?'
when 2^n converges
absolutely and the constants
An
so far arbitrary, tend
to infinity with n.
*
Cf Pringsheim, Math. Ann., Leipzig, 68,

.
p. 399.
FOURIER'S INTEGRALS
292
Fourier's Cosine Integral and Sine Integral. In the

f(x) is given only for positive values of x, there are
122.
case
when
two forms of Fourier's Integral which correspond

Series and Sine Series respectively.
I.
In the
first place,
consider the result of
to the Cosine
119,
when
f(x)
has the same values for negative values of x as for the corre0.
sponding positive values of x: i.e. f( x)=f(x), x
>
/<
Then
da
r>
r 80
x)] dx'
80
2 C
ca
and
pCC
interval, while
f(x) cos a# cos ax'dx.
from
x,
+ x) + cos a (x
Jo
7T J
values of
f(x') [cos a (x
da\
TT J
It follows
x) dx
f(x) cos a (x
J-
TT J
= -1
CO
4, op
119 that when f(x) is defined for positive

satisfies Dirichlet's Conditions in any finite
f(x)dx converges absolutely, then
Jo
2
00
f"
^Jo
c2a
/(a;')
cos ax cos ax'dx
a every point where

the
f(x-}-0)
value of the integral
Also
infinity
it
= J [/(#+ 0) +/(#
0)],
Jo
is f(
from
follows
and f(x0)
+ 0),
exist,
and
'when x =
if this limit exists.
120 and 121 that the condition at
be replaced by either of the following:

For some positive number and to the right of it, f(x')
shall be bounded and monotonic and Lt f(x) =
may
(i)
X'->CG
or,
(ii)
For some positive number and
to the right
of
it,
f(x')
form g(x) cos(X + //), where g(x) is

bounded and monotonic and Lt g(x ) = 0. Also
shall be of the
j
r J0(*o dx
,
II.
'
must
converge.
In the. next place, by taking f(
x)=f(x), x^>0, we see
that, ivJien f(x) is defined for positive values of x,
Dirichlet's Conditions
in any finite interval, while
converges absolutely, then
da
if
Jn
Jo
and
f(x')sinaxsinaxdx
satisfies
f(x) dx
FOURIER'S INTEGRALS
at every point where f(x-\-Q)
and f(x
293
0) exist,
and when & =
the integral is zero.
Also
it
120 and 121 that the condition at
follows from
infinity may be replaced by one or other of those given under I.

It should be noticed that, when we express the arbitrary
function fix) by any of Fourier's Integrals, we must first decide

for what value of x we wish the value of the integral, and that
x must be inserted in the integrand before the
this value of
integrations indicated are carried out
(cf.
62).
123. Fourier's Integrals.

Sommerfeld's Discussion. In many of the
problems of Applied Mathematics in the solution of which Fourier's Integrals
occur, they appear in a slightly different form, with an exponential factor
a 2t added.
In these cases we are concerned with the limiting value
(e.g. e)
'
of the integral
and, so far as the actual physical problem

integral for t = is not required.
was shown,
It
first of all
is
by Sommerfeld,*
concerned, the value of the
that,
when
the limit on the
right-hand side exists,
Lt
t~*4) 7T Jo
Ja
when
when x=a,
when x = b,
the result holding in the case of any integrable function given
interval (a, b).
The
it
case
when the
interval
infinite
is
was
also treated
in
the
by Sommerfeld, but
much greater detail by Young. t It will be

that, when the arbitrary function satisfies the
has been recently examined in
sufficient in this place to state

conditions at infinity imposed in
for
an
120-122, Sommerfeld's result
still
holds
infinite interval.
However,
it
should be noticed that
integral
r*
we cannot deduce
the value of the
.-*
da
from the above

for
This would require the continuity of the function
= 0.
We
ment
*
results.
have come across the same point in the discussion of Poisson's

[Cf.
Sommerfeld, Die wUlkiirlichen Fnnctionen in der mathematischen Physik, Diss.
niiigslMTg, ISiM.
treat-
99.]
W. H. Young,
loc. cit.,
Edinburgh, Proc.
/?.
Soc., 31, 1911.
FOURIER'S INTEGRALS
294
REFERENCES.
de la chaleur, Oh. IX.
FOURIER,
e
JORDAN, loc. cit., T. II. (3 ed.), Ch. V.
LEBESGUE, Lecons sur les series trigonome'triques, Ch.
NEUMANN, loc. cit., Kap. III.
The'orie analytique
WEBER-RIEMANN, loc. cit., Ed. I. (2.

the papers named in the text.
III., Paris, 1906.
13-20.
Aufl.),
And
EXAMPLES ON CHAPTER
1.
zero
Taking
when
/(.r)
<.r,
as
in Fourier's
Cosine Integral
X.
when 0<.r<l, and
as
show that
1
= 2-
r sin a cos cur
(0<O;
da,
<
a cos our
sin
-li
2.
By
x
considering Fourier's Sine Integral for e~P (/3>0), prove^hat
/"*
a sin cur T
TT ~
e P^
T^T- act. =
a
jn
;
'
and
in the
same way, from the Cosine Integral, prove that

00
f
cos ax , _ TT __
3.
Show
/""
/
COS
z
equal to x
is
to
Show
when
that
(Ox\dO
I
TT Jo
4.
2/3
<x2-f/?
that the expression
2a2
is
Jo
=.y<a, and
-/
sin
qx
-î
^ cos y
"w
a*J
to zero
- + tan a
when .r>a.
^^-
dq
the ordinate of a broken line running parallel to the axis of x from .r =

x=a, and from x=b to ^-=oc , and inclined to the axis of x at an angle a
between x = a and x = b.
5.
Show
Integral,
that /(a?)
and
= -7
V
satisfies
Show
Integral,
120 for Fourier's
verify independently that

cos
6.
the conditions of
*l
that f(x) =
ax cos
satisfies
ax'
j-
= -T- when x > 0.
the conditions of
and verify independently that

30
7rJ
_.
rta
r*
/
J_
sin
sin./x cos a (.r - .r )//.<'T = --.''
0?
121
for Fourier's
APPENDIX
PRACTICAL HARMONIC ANALYSIS AND PERIODOGRAM

ANALYSIS
1.
We
Let y = f(x) be a given periodic function, with period 27r.

have seen that, for a very general class of functions, we may
represent f(x)
by
its
Fourier's Series
I
cos x -f
sin #
where a
2
6.
cos 2x-|sin
are the Fourier's Constants for f(x).

We may suppose the range of x to be O X^ZTT. If the period
COS
is a, instead of 27r, the terms
nz are replaced by J 2mrx/a,
sin
sin
,
av
a.2
...
bv
b.,,
...
and the range becomes 0ô;^.

However, in many practical applications, y is not known
analytically as a function of x, but the relation between the
dependent and independent variables is given in the form of a
curve obtained by continuous observations. Or again, we may
only be given the values of y corresponding to isolated values of
x,
Jn
the observations having been made at definite intervals.
we may suppose that a continuous curve is drawn
the latter case
through the isolated points in the plane of x, y. And in both

cases the Fourier's Constants for the function can be obtained
1
by mechanical means. One of the best known machines for the

purpose is Kelvin's Harmonic Analyser.*
2. The practical questions referred to above can also be treated
*
Such mechanical methods are described
in the
handbook entitled Modern
Instruments and Methods of Calculation, published by Bell

with the Napier Tercententary Meeting of 1914.
295
& Sons
in connection
APPENDIX
296
for Fourier's Infinite Series a trigonometrical

with only a limited number of terms.
Suppose the value of the function given at the points
where ??ia = 27r.
0, a, 2a, ... (m
I) a,
by substituting
series
Denote these points on the interval

^0
and the corresponding values
2/i>2/ 2
2/o>
Let
If
ym-i-
can determine these
when = 0,
a
...
-f
+an cos wart
2n+l
1 , 2,
2%-f-l equations giving the values of
are as follows
y by
>
Sn (xg ) = y,,
by
1>
Sn (x) = a + a cosx + a2 cos2x+
2n+l=m, we
The
of
(0, 2-Tr)
*** *^wi
*^l' *^2'
constants so that
2w.
. . .
a 1?
...
&,
frp
...
bn
cos xl
+a
cos rx -f
-f a n cos wo^
a x cos x2n +...+,. cos nc 2n +...+

6 X sin
XM +
+b
sin
Adding these equations, we
rx., n
cos
rac2n
+ b n sin nx. u
. . .
see that
=0
since
and
+ cos ra + cos 2ra +

sin ra + sin 2m +
...
+ cos 2nra = 0,
...
+sin2wa = 0,
(2n+l)a = 27r.
when
Further,
we know
that
+ cos 2n* cos 2sa

+ cos 2nra cos 2nsa = 0,
1 -h cos ra cos sa
s=/= r,
= l,2, n\
=
Is
1,2,... wj
cos ra sin sa + cos 2ra sin 2sa
+
And
cos 2 ra
by
+ cos
It follows that, if
the third
cos rx%,
fr
+ cos 2nra sin %nsa = 0,

2
2ra
. . .
we multiply
etc.,
+ cos
2nra = J (2n + 1 ).
the second equation by cosra 15
and add, we have

cos rsa
APPENDIX
Similarly,
we
297
find that
\ (2n+
A trigonometrical
br
^y
sin rsa.
s=l
series of
(2w+ 1) terms has thus been formed,
whose sum takes the required values at the points

where (2w+l) = 27r.
0, a, 2a, ... 2na,
It will be observed that as n-^-<x> the values of
a lt
...
and
b lt 6 2
reduce to the integral forms for the coefficients, but as remarked in 90, p. 199,
this passage from a finite number of equations to an infinite number requires
more
careful handling
if
the proof
is
made
to be
rigorous.
For purposes of calculation, there are advantages in taking

an even number of equidistant points instead of an odd number.
3.
Suppose that to the points

0, a, 2a, ...
we have
(2n
where na = 7r,
l)a,
the corresponding values of y,

2/o> 2/i> 2/ 2 >-
In this case
we can
2/^-1-
obtain the values of the 2n constants in
/the expression
a
+ cos x+ a.
+ &! sin x+
!
so that the
sum
cos 2sc+
b.2
sin 2x4-
+ a n _ cos (nl)x + a n cos nx]

/'
+ b n _ siu(nI)x
l
shall take the values
yQ y v
,
...
y-2H -i
at these
2n
points in (0, 2-rr).

It will be found that
2n-l
S
=o
2n -
^yscosxsa,
if
sin rsa
Runge
gave a convenient scheme for evaluating these con-
This and a similar

stants in the case of 12 equidistant points.
table devised by Whittaker for the case of 24 equidistant points
* Z*.
Math., Ldpzig, 48, 1903, and 52,
pp. 147-164, Leipzig, 1904.
190.").
Also Theorie
u.
Praxis der Reihen,
APPENDIX
298
will be
found in Tract No. 4 of the Edinburgh Mathematical
Tracts.*
4.
This question
be looked at from another point of view.
may
Suppose we are given the values of

2/0>
y, viz.
y,n-l,
2/2,
?/!
corresponding to the points

0,
2a,
a,
Denote these values of
Let
(m
...
= tt
x,
ma = 27r.
where
l)a,
as before,
by
acosz6fcos3;
j
sin
a; -f-
6 2 sin 2s;
...
...
acosnx}
+ b n sin rac
'
./
For a given value of n, on the understanding that m>2?i + l,t

the 2n + I constants a a 1} ...a n \,...b n are to be determined
,
so that
$(#)
shall
as
approximate
as
closely
y m -\ at a; a^, ... x,,^.

The Theory of Least Squares shows that the
mation will be obtained by making the function
possible
to
2/o> 2/i>
regarded as a function of a
The conditions
for a
a 1?
an
...
6P
...
?;
a minimum.
in this case, are
minimum,
closest approxi-
m-l
V t/ ^
^^^
71 V
*/ /
wi.-1
Sn (X ))cOSpX = Q
Vs
(y-
il
- 8*
=0
sin p^'
2 above, that these equations lead to

It will be found, as in
the following values for the coefficients
:
* Carse
and Shearer,
Course in Fourier's Analysis and Periodogram Analysis,
Edinburgh, 1915.
flf
SH (x)
m <2.+ l,
we can
shall be equal to
stants than equations.

so that this condition
choose the constants in any number of ways so that

y l9 ... ym _ l at o? x lt ... xm _ l for there are more con-
And
if
m = 2n +
is satisfied-
1,
we can
choose the constants in one
way
APPENDIX
ma r = V y
299
1
1
cos rsa
odd.
But
if
is
r=l,
3=0
even, the coefficient a r
(when r = ^m)
is
given by
the others remaining as above.

In some
cases, it is sufficient to find the ternis
+a
The values
most
of
cr
and
a,
6n
cos
which
up
to cos.*:
and
sin
jc t
viz.
x + &! sin x.
will
make
this expression
approximate
closely to
.'/o>
at
0,
are then given
by
?
.
/i>
a.,
^>
2a,
...
#-!
when
(i -!)-,
wio.=2?r,
=0
tn-1
=2
?/cossa,
1=0*
Jwi6 t
=2
=0
Tables for evaluating the coefficients in such cases
have been constructed by
Turner.*
5.
In the preceding sections we have been dealing with a set

known to have a definite period. The graph for
of observations
the observations would repeat itself exactly after the lapse of

the period and the function thus defined could be decomposed
;
by the methods just described.

But when the graph of the observations is not periodic, the
into simple undulations
function
may
yet be represented by a
sum
with
of periodic terms
The
other.
each
whose periods are incommensurable

gravitational attractions of the heavenly bodies, to which the
tides are due, are made up of components whose periods are not
commensurable.
* Tables
for
But
in the tidal
graph of a port the component
Harmonic Ancdysis, Oxford University
Press, 1913.
APPENDIX
300
due to each would be resolvable into simple undulations. A

method of extracting these trains of simple waves from the
record would allow the schedule of the tidal oscillations at the
port to be constructed for the future so far as these components
are concerned.
The usual method
of extracting from a graph of length L, a

that
itself
part
repeats
periodically in equal lengths A is to cut
up the graph into segments of this length, and superpose them
by addition or mechanically. If there are enough segments, the

sum thus obtained, divided by the number of the segments,
approximates to the periodic part sought the other oscillations
of different periods may be expected to contribute a constant to
;
the sum, namely the

6.
sum
of the
The
hidden
principle of this
periodicities in a
mean part
method
set
of
of each.
also used in searching for
is
observations
taken
over
Suppose that a period T occurs in these

observations and that they are taken at equal intervals, there
considerable time.
being n observations in the period T.

Arrange the numbers in rows thus
UQ,
Add
16j,
1^2>
the vertical columns, and let the sums be
U U
U U^ U
t,
lt
...
U,_,,
U,,_,.
... UH _ lt
In the sequence
Un the component of
z
the
variable parts of the
and
T
will
be
m-fold,
period
multiplied
other components may be expected to disappear, as these will
,
enter with different phases into the horizontal rows, and the
rows are supposed to be numerous. The difference between the
Ult U2 ... Un . 2 Un -i
greatest and least of the numbers U
,
furnishes a rough indication of the amplitude of the component

of period T, if such exists and the presence of such a period is
indicated by this difference being large.
;
Let y denote the difference between the greatest and least of

U Z7 1>% Z72
^-2 f-i corresponding to the trial
"
If
a function of x, we obtain a curve
is
as
x.
period
plotted
y
the numbers
APPENDIX
301
This curve will have peaks at the values of x

corresponding to the periodicities which really exist. When
the presence of such periods is indicated by the curve, the
of periods."
then analysed by the methods above described.

This method was devised by Whittaker for the discussion of
statistics are
the periodicities entering into the variation of variable stars.*

It is a modification of Schuster's work, applied by him to the
discussion of the statistics of sunspots and other cosrnical
"
phenomena.! To Schuster, the term periodogram analysis" is
"
"
due, but the curve of periods referred to above is not identical
with that finally adopted by Schuster and termed periodograph
(or periodogram).
For numerical examples, and for descriptions of other methods

of attacking this
problem, reference
Edinburgh Mathematical Tract, No.
may
4,
be
made
to
the
already cited, and to
Schuster's papers.
*
Monthly Notices, H.A.S., 71, p. 686, 1911.

See also a paper by Gibb, " The Periodogram Analysis of the Variation of
SS Cygni," ibid., 74, p. 678, 1914.

The following papers may be mentioned
j"
Cambridge, Trans. Phil. Soc., 18, p. 108, 1900.

London, Phif. Trans. /?. Soc., 206 (A), p. 69, 1906.
London, Proc. R. Soc., 77 (A), p. 136, 1906.
APPENDIX
II
BIBLIOGRAPHY*
FOURIERS SERIES AND INTEGRALS
TREATISES
Th6orie analytique de la chaleur. Paris, 1822.
English Translation (Freeman). Cambridge, 1872.
FOURIER.
German
Translation (Weinstein).
Traite de Mecanique (2 e
POISSON.
Theorie mathematique de
6d.).
Berlin, 1884.
Paris, 1833.
la chaleur.
Paris, 1835.
Serie di Fourier e altre rappresentazioni analitiche delle funzioni di

una variabile reale. Pisa, 1880.
DINI.
NEUMANN.
tJber die nach Kreis-, Kugel-,
schreitenden Entwickelungen.
und Cylinder-Functionen
fort-
Leipzig, 1881.
BEAU. Analytische Untersuchungen im Gebiete der trigonometrischen

Beihen und der Fourier'schen Integrate. Halle, 1884.
3LY.
An
Elementary Treatise on Fourier's Series and Spherical,

and Ellipsoidal Harmonics, with Applications to Problems
Mathematical Physics. Boston, 1893.
Cylindrical,
in
Theorie analytique de
POINCARE".
FRISCHAUF.
Vorlesungen
liber
la
propagation de
Kreis-
la chaleur.
Paris, 1895.
und Kugel-Fnnctionen Reihen.
Leipzig, 1897.
BURKHARDT. Entwicklungen nach oscillirenden Fimctionen.

D. Math. Ver., Leipzig, Bd. X., Heft. II., 1901.
CARSLAW.
Jahresber.
Fourier's Series and Integrals and the Mathematical Theory of

London, 1906.
the Conduction of Heat.
LEBESGUE.
HojRfioig-
Lesoiis sur les series trigonometriques.
The Theory
Fourier's Series.
*
The abbreviated
Paris, 1906.
of Functions of a Real Variable
Cambridge,
titles
and the Theory
of
1907.
for the Journals are taken from the International
Catalogue of Scientific Literature.
302
APPENDIX
303
A course in Fourier's Analysis and Periodogram

(Edinburgh Mathematical Tracts, No. 4.) Edinburgh, 1915.
In many other works considerable space is given to the treatment of
The following may be specially
Fourier's Series and Integrals.
mentioned
CARSE and SHEARER.

Analysis.
DE LA VALLEE
Cours d'Analyse, T.
POUSSIN.
GOURSAT.
Cours d'Analyse, T. I. (3
Cours d'Analyse, T. II. (3 e
JORDAN.
ed.).
ed.).
Theorie und Praxis der Reihen.
RUNGE.
SERRET-HARNACK.
Lehrbuch der
II. (2
ed.).
Paris, 1912.
Paris, 1918.
Paris, 1913.
Leipzig, 1904.
Differential-
und Integral-Rechnung, Bd.
Leipzig, 1915.
II. (6 u. 7 Aufl.).
WEBER.
Die partiellen DifFerential-gleichungen der mathematischen Physik,

(2 Aufl.).
Braunschweig, 1910.
work
is
also
known as the fifth edition of Riemann's lectures
[This
Bd.
I.
entitled Partielle Differential-gleichungen und deren Anwendungen auf

physikalischen Fragen. Braunschweig, 1869. It is referred to in the
text as Weber-Riemann.]
WHITTAKER and WATSON.

[This
Course of Modern Analysis. Cambridge, 1920.

work with the same title.]
the third edition of Whittaker's
is
MEMOIRS AND PAPERS.

FOURIER.
Fourier's first paper was communicated to the Academy in 1807.

discussion of the subject was presented in his Memoire sur la
A fuller
propagation de la chaleur, communicated in 1811, and not printed till

1824-6 (Paris, Mem. Acad. sci., 4, 1825, and 5, 1826). This memoir is
practically contained in his treatise (1822).
Many other papers bearing
this subject and upon the Mathematical Theory of the Conduction
Heat were published by him about the same time. A list of these
upou
of
writings
given in Freeman's Edition, pp. 10-13.
is
Fourier.
See also (Euvres de
Paris, 1888.
fonctions par les series de

(i) Sur la maniere d'exprimer les
quantites periodiques, et sur 1'usage de cette transformation dans les
resolutions de differents problemes. J. 6c. poly tech., Paris, 11, 1820.
POISSON.
(ii)
Memoire sur
les integrates definies et
sur la sommation des series.
J. ec. polytechn., Paris, 12, 1823.

(Cf. also
CAUCHY.
Paris,
(ii)
1827.
(i)
Sur
Sur
Mem.
Sur
le
les integrates definies.
developpement des
Acad.
sci., 6,
les residus
Cf. (Euvres de
J. ec. polytechn., 9, 10,
and
11.)
fonctions en series periodiques.
1826.
de fonctions exprimees par integrales
Cauchy
(Ser. 2), T.
definies.
7.
(iii) Memoire sur 1'application du Calcul des Residus aux questions de

physique mathematique. 1827. Cf. (Euvres de Cauchy (Ser. 2), T. 15.
(iv) Second memoire sur 1'application du Calcul des Residus aux
questions de physique mathematique. Paris, Mem. Acad. sci., 7, 1827.
APPENDIX
304
Sur
POISSON.
Acad.
calcul
le
sci., 7,
numerique des integrales
definies.
Paris,
Mem.
1827.
Uber die Darstellung beliebiger Functionen mittelst Reihen die

nach den Sinussen und Cosinussen der Vielfachen eines Winkels fortschreiten.
Berlin, Abb. Ak. Wiss., 1827.
DIRKSEN.
Sur la convergence des series trigonometriques qui servent a

representer une fonction arbitraire entre des limites donnees. J. Math.,
DIRICHLET.
Berlin, 4, 1829.
DIRKSEN
Uber die Convergenz einer nach den Sinussen und Cosinussen der
(i).
Vielfachen eines Winkels fortschreitenden Reihe.
J. Math., Berlin, 4,
1829.
(ii).
Uber
die
Summe
einer nach den Sinussen
und Cosinussen der
Vielfachen eines Winkels fortschreitenden Keihe.
Berlin,
Abh. Ak.
Wiss., 1829.
DIRICHLET.
(i)
Die Darstellnng ganz willkiirlicher Functionen durch SinusDove's Repertorium der Physik., Bd. I., 1837.
und Cosinus-Reihen.
(ii) Sur les series dont le term general depend de deux angles, et qui
servent a exprimer des fonctions arbitrages entre des limites donnees.
J.

(iii)
finies
Sur 1'usage des integrales definies dans

on infinies. J. Math., Berlin, 17, 1837.
On Fluctuating
HAMILTON.
Functions.
la
sommation des
series
Dublin, Trans. R. Irish Acad.,
19, 1843.
Demonstration simple du theoreme de Fourier.
BONNET.
Paris,
3,
On
BOOLE.
Nouv. ami. math.,
1844.
the Analysis of Discontinuous Functions.
Dublin, Trans. R.
Irish Acad., 21, 1848.
STOKES.
On
the Critical Values of the
Sums
of Periodic Functions.
Cam-
bridge, Trans. Phil. Soc., 8, 1849.
BONNET,
(ii)
(i)
Sur quelques integrales
Sur
definies.
la theorie generate des series.
J.
math., Paris, 14, 1849.

Mem. cour. Acad.
Bruxelles,
roy., 23, 1850.
DUHAMEL.
Sur la discontinuity des valeurs des
la reconnaitre.
RIEMANN.
Uber
J.
series, et
sur les
moyens de
math., Paris, 19, 1854
die Darstellbarkeit einer Function durch eine trigonome(Also Gottingen, Abh.

Gottingen, Hab. Schrift, 1854.
trische Reihe.
Ges. Wiss., 13, 1868.
De explicatione per series trigonometricas instituenda functiunius variabilis arbitrarium et praecipue earum quae per variabilis
spatium finitum valorum maximorum et minimorum numerum habent
LIPSCHITZ.
onum
infinitum, disquisitio.
J.
APPENDIX
Du
305
BOIS-REYMOND. Uber die aUgemeinen Eigenschaften der Klasse von

Doppel-Integralen zu welcher das Fourier'sche Doppel-Integral gehort.
J. Math., Berlin, 69, 1878.
Uber trigonometrische Reihen. J. Math., BerUn, 71, 1870.

CANTOR, (i) Uber einen die tngonometrischen Reihen betreffenden Lehrsatz.
HEINE.

Beweis dass eine fur jeden reellen Werth von x durch eine trigonometrische Reihe gegebene Function sich nur auf einzige Weise in dieser
J.
(ii)
Form

Notiz zu diesem Aufsatze. J. Math., Berlin, 73, 1871.
Uber trigonometrische Reihen. Math. Ann., Leipzig,
darstellen lasst.
(iii)
(iv)
Du
1871.
4,
BOIS-REYMOND. Die Theorie der Fourier'schen Integrale und Formeln.

Zur Integration der Differential-Gleichung ^-r,-f^-^ = 0.
PRYM.
J.
Math.
Berlin, 73, 1871.
Uber
CANTOR.
die
Ausdehnung
Uber
SCHWARZ.
der trigono-
eines Satzes aus der Theorie
metrischen Reihen.
5, 1872.
die Integration der Differential Gleichung
Aw = 0.
J.
Math.,
Berlin, 74, 1872.
THOMAE.
iiber Fonrier'sche
Bemerkung
Reihen.
Zs. Math., Leipzig, 17,
1872.
ASCOLI.
Du
Uber trigonometrische Reihen.
BOIS-REYMOND.
Ges. Wiss.
GLAISHER.
On
t)ber
die Fourierschen
Reihen.
6,
1873.
Gottingen, Nachr.
1873.
Fourier's
Double Integral Theorem.
Mess.
Math.,
Cam-
bridge, 2, 1873.
Du
BOIS-REYMOND. "Dber die sprungweisen Werthanderungen analytischer

Functionen. Math. Ann., Leipzig, 7, 1874.
Einige Zweifel an der allgemeinen Darstellbarkeit einer willperiodischer Function einer Variablen durch eine trigono-
SCHLAFLI.
kurlicher
metrische Reihe.
(Cf. also.
Uber
Bern. Universitats-Program, 1874.
die partielle Differential- Gleichung -~-
K^-^.
J.
Math.,
Berlin, 72, 1870.)
ASCOLI.
Sulla serie di Fourier.
Du
BOIS-REYMOND.
Du
BOIS-REYMOND.
Reihe
Ann. mat., Milano
(Ser. 2), 6, 1875.
Allgemeine Lehrsatze iiber den Giiltigkeitsbereich der

Integralformehi die zur Darstellung willkiirlicher Functionen dienen.
GENOCCHI. Intorno ad alcune serie. Torino, Atti Ace. sc., 10, 1875.
(i)
Beweis dass die coefficienten der trigonometrischen

00
/(.?)
= 2 (aP cos px -f bp sin px\

p=0
C.
APPENDIX
306
Werthe
die
a ) cos ^ a da
>
= -~1
bP
/( a ) sin P a rf
haben, jedesmal wenn diese Integrale endlich und bestimmt sind.

chen, Abh. Ak. Wiss., 12, 1876.
Miin-
iiber die Convergenz und Divergenz der Fourier'(ii) Untersuchungen

schen Darstellungsformeln. Miinchen, Abh. Ak. Wiss., 12, 1876.
(iii) Zusatze zu dieser Abhandlung.
Notiz iiber eine bemerkenswerthe Eigenschaft der periodischen

Wien, Anz. Ak. Wiss., 13, 1876.
TOPLER.
Reihen.
GLAISHER.
Note on Fourier's Theorem.
ASCOLI.
Nuove
(i)
Mess. Math., Cambridge,
ricerche sulla serie di Fourier.
Roma, Rend.
6,
1877.
Ace. Lincei
(Ser 3), 2, 1878.

di
(ii) Sulla rappresentabilita
una funzione a due variabili per serie

doppia trigonometrica. Roma, Rend. Ace. Lincei (Ser. 3), 4, 1879.
APPEL. Sur une theoreme concernant les series trigonometriques. Arch.
Math., Leipzig, 64, 1879.
BONNET.
Sur
la
formule qui sert de fondement a une theorie des series

Bui. sci. math., Paris (Ser. 3), 3, 1879.
trigonometriques.
Du
Determination de la valeur limite d'une integrate.
BOIS-REYMOND.
sci.
math., Paris (Ser.
HOPPE.
Bemerkung
3), 3,
Bui.
1879.
iiber trigonometrischen Reihen.
Arch. Math., Leipzig,
64, 1879.
Versuch einer Geschichte der Darstellung willkurlicher Functionen
SACHSE.
einer Variablen durch trigonometrischen Reihen.

Diss., Gottingen, 1879.
Also Zs. Math., Leipzig, 25, 1880, and in French, Bui. sci. math., Paris
(Ser. 3), 4, 1880.
ASCOLI.
Sulla serie trigonometriche a due variabili.

3), 8, 1880.
Roma, Rend. Ace.
Lincei (Ser.
CANTOR,
(i)
Bemerkung
iiber
trigonometrische
Math.
Reihen.
Ann.,
Leipzig, 16, 1880.

(ii)
Fernere Bemerkung iiber trigonometrische Reihen.
Math. Ann.,
Leipzig, 16, 1880.
Du
BOIS-REYMOND.
Zur Geschichte der trigonometrischen Reihen.
Tub-
ingen, 1880.
BOUSSINESQ.
Coup
d'ceil
sur la theorie
des series trigonometriques et sur

J. math., Paris (Ser. 3),
une raison naturelle de leur convergence.

7, 1881.
Du
BOIS-REYMOND.
Uber Darstellungsfunctionen.
18, 1881.
HARNACK.
Uber
die trigonometrische Reihe
licher Functionen.
JORDAN.
Sur
la serie
und
die Darstellung willkur-

de Fourier. Paris, C. R. Acad.
sci.,
92, 1881.
APPENDIX
307
SHARP.
On
ASCOLI.
Una osservanza relativa ad un teorema contenuto nella sua memoria
Fourier's Theorem.
Sulla rappresentabilita di
trigonometrica.
HALPHEN.
Sur
HARNACK.
(i)
Q. J. Math., London, 17, 1881.
una funzione a due
Milano, Rend.
la serie
1st.
de Fourier.
variabili per serie
doppia
lomb., 15, 1882.
Paris, C.R. Acad. sci., 95, 1882.
Theorie de la serie de Fourier.
Bui.
sci.
math., Paris (Ser.
3),
1882.
6,
(ii) Vereinfachung der Beweise in der Theorie der Fourier'schen Reihen.

HOBSON.
On
Fourier's Theorems.
LINDEMANN.
tlber das Verhalten der Fourier'schen Reihe an SprungMath. Ann., Leipzig, 19, 1882.
stellen.
Sur
POINCARE".
VELTMANN.
CANTOR.
Du
Mess. Math., Cambridge, 11, 1882.
les series
trigonometriques.
Die Fouriersche Reihe.
Sur
les series
Paris, C.R. Acad.
sci.,
95, 1882.
Zs. Math., Leipzig, 27, 1882.
trigonometriques.
Acta Math., Stockholm,
2, 1883.
BOIS-REYMOND.
tJber die Integration der trigonometrischen Reihe.
GILBERT.
Demonstration simplifiee des formules de Fourier.
Bruxelles, Ann.
Soc. scient., 8, 1884.
HOLDER.
Zur Theorie der trigonometrischen Reihen.
24, 1884.
ALEXANDER,
(i)
Failing
Cases
Edinburgh, Proc. Math. Soc.,

(ii)
Boole's
and other proofs
Edinburgh, Proc. Math. Soc.,
HOLDER.
of
Fourier's
Double
3,
of Fourier's
Uber eine neue hinreichende Bedingung

Uber das
Theorem.
Double Integral Theorem.
1885.
Function durch die Fourier'sche Reihe.
KRONECKER.
Integral
3, 1885.
f iir
die Darstellbarkeit einer
Berlin, SitzBer. Ak. Wiss., 1885.
Dirichlet'sche Integral.
Berlin, SitzBer. Ak. Wiss.,
1885.
Sur
POINCARK.
les
series
Uber
die
trigonometriques.
Paris, C.R. Acad.
sci.,
101,
1885.
WEIERSTRASS.
Darstellbarkeit sogenannter willkiirlicher FuncBerlin, SitzBer. Ak. Wiss., 1885.
tionen durch die Fouriersche Reihe.
A symbolic proof of Fourier's Double

ALEXANDER.
ARZELA.
Integral Theorem.
Sopra una certa estensione di un teorema relative

Roma, Rend. Ace. Lincei (Ser. 4), 1, 1886.
alle serie trigono-
metriche.
(i) A proof of Fourier's Series for Periodic Functions.

Math., Cambridge, 16, 1887.
ALEXANDER,
(ii)
Extensions of Fourier's trigonometric Series Theorem.
Cambridge, 16, 1887.

c. i
u2
Mess.
Mess. Math.,
APPENDIX
308
JOHNSON.
proof of Fourier's Series Theorem.
Mess. Math., Cambridge,
16, 1887.
HARNACK.
Uber Cauchy's zweiten Beweis fur die Convergenz der Fourier' schen Reihen und eine damit verwandte altere Methode von Poisson.
BOURLET.
Sur la multiplication des series trigonometriques.

Paris (Ser. 2), 13, 1889.
Sur la representation approchee des fonctions.

112, 1891.
PICARD.
sci.,
SOMMERFELD.
sci.
math.,
Paris, C.R. Acad.
Die willkurlichen Functionen in der mathematischen Physik.
Konigsberg, 1891.
Diss.,
DE
Bui.
LA VALL^B POUSSIN.
Sur
SEQUIER.
Sur une demonstration des formules de Fourier
Bruxelles, Ann. Soc. scient., 15, 1892.
generalisê.
la serie
de Fourier.
Nouv. aim. math., Paris
(Ser. 3), 11, 1892.
Einige Mittheilungen aus dem Gebiete der trigonometrischen Reihen

und der Fourier'schen Integrale. Prog. Gymn. Sorau, No. 87, 1893.
BEAU.
DE LA
VALLISE POUSSIN.
Ann. Soc.
Bruxelles,
Sur quelques applications de
de Poisson.
1'integrale
scient., 17, 1893.
the History of the Fourier's

(i) On
Math. Soc., 11, 1893.
GIBSON,
Edinburgh, Proc.
Series.
(ii) A proof of the Uniform Convergence of Fourier's Series, with notes

on the Differentiation of the Series. Edinburgh, Proc. Math. Soc., 12,
1894.
BEROER. Sur le developpement de quelques fonctions discontinues en

de Fourier. Upsala, Soc. Scient. Acta (Ser. 3), 15, 1895.
On
WILLIAMS.
the Convergence of Fourier's Series.
Phil. Mag..
series
London
(Ser. 5), 41, 1896.
Les rapports de Fanalyse et de
POINCARE\
sci.,
gen.
On
PRESTON.
London
la
physique mathematique.
Rev.
Paris, 18, 1897.
the general extension of Fourier's Theorem.
Phil.
Mag.,
(Ser. 5), 43, 1897.
BRODEN.
Uber das
MAROLLI.
Di una
classe di funzioni finite e continue
serie di Fourier nel
punto a =0.
52, 1899.
non
sviluppibabili in
Giorn. mat., Napoli, 37, 1899.
Demonstrazione semplice delle sviluppibilita in serie di Fourier di

una funzione angolare finita e ad una sol valore. Roma, Rend. Ace.
VOIGT.
Lincei (Ser.
FE.TR.
Sur
5), 8,
1899.
les fonctions
bornees et integrables.
Paris, C.R. Acad. sci.,
131, 1900.
LERCH.
Remarque
24, 1900.
sur la serie de Fourier.
Bui.
sci.
math., Paris (Ser.
2),
APPENDIX
FORD.
of showing the convergence of Fourier's series and of

New York, N.Y., Bull. Amer. Math. Soc.,
method
Dini's
309
other allied developments.

7, 1901.
HURWITZ.
Sur
STACKEL.
les series
Uber das
(i)
Paris, C.R. Acad.
de Fourier.
sci.,
132, 1901.
Leipzig, Ber. Ges. Wiss.,
53, 1901.
(ii)
Uber
Leipzig
(3.
Sur
FE.TER.
die Convergenz der trigonometrischen Reihen.

2, 1901.
Arch. Math.,
Reihe),
la differentiation
de la
de Fourier.
serie
Paris, C.R. Acad. sci.,
134, 1902.
HURWITZ. Sur quelques applications geometriques des

Ann. sci. EC. norm., Paris (Ser. 3), 19, 1902.
STEKLOFF.
(i)
series
de Fourier.
Paris, C.Jl. Acad. sci.,
Sur certaines egalites remarquables.
135, 1902.
Paris, C.R.
Sur la rapresentation approchee des fonctions.

135, 1902.
(ii)
sci.,
Acad.
en series
(iii) Sur quelques consequences de certains developpements
Paris, C.R. Acad.
analogues aux developpements trigonometriques.
sci.,
135, 1902.
(iv)
Remarque
relative
ma
Note
Paris, C. R. Acad.
des fonctions.
sci.,
Sur la representation approchee

135, 1902.
..
Uber
HURWITZ.
die Fourier'schen
Konstanten integrierbarer Functionen.
Uber Fouriersche
KRAUSE.
Reihen mit zwei
veranderlichen
Grossen.
Leipzig, Ber. Ges. Wiss., 55, 1903.
Sur
LEBESGUE.
les
series
Ann.
trigonometriques.
sci.
EC. norm., Paris
(Ser. 3), 20, 1903.
Some
MACDONALD.
Applications of Fourier's Theorem.
London, Proc. Math.
Soc., 35, 1903.
An
STEPHENSON.
extension of
the Fourier
Method
of
Expansion in Sine
Series.
Uber das Restglied trigonometrischer Reihen.
BIERMANN.
Wien, SitzBer.
Ak. Wiss., 113, 1904.
Zwei neue Beweise fiir den Fundamental- Satz der Fourier'schen

MonHfte. Math. Phys., Wien, 15, 1904.
FISCHER.
Konstanten.
FEJJ^R.
Untersuchungen
liber Fourier' sche
Reihen.
58, 1904.
KNESER.
(i)
discher
Die Fourier' sche Reihe und die angenaherte Darstellung perioFunctionen durch endliche trigonometrische Reihen.
Arch.
Math., Leipzig
in
(3.
Reihe), 7, 1904.
Untersuchungen iiber die Darstellung willkurlicher Functionen

der mathematischen Physik. Math. Ann., Leipzig, 58, 1904.
(ii)
APPENDIX
310
Sur certaines egalites generates communes a plusieurs series de

dans 1'analyse.
St. Petersburg, Mem.
STEKLOFF.
fonctions souvents employees

Ac. Sc. (Ser. 8), 15, 1904.
On
HOBSON.
the failure of the convergence of Fourier's Series.
London,
Proc. Math. Soc. (Ser. 2), 3, 1905.
KNESER.
Beitrage sur Theorie der Sturm-Liouville'schen Darstellung

will-
kurlicher Funktionen.
LEBESGUE.
Sur une condition de convergence des
(i)
Acad.
Paris, C.R.
Sur
(ii)
sci.,
la divergence et la
Paris, C.R.
Fourier.
Acad.
de Fourier.
convergence non-uniforme des
sci.,
series
de
141. 1905.
Recherches sur la convergence des
(iii)
series
140, 1905.
series
Math. Ann.,
de Fourier.
Leipzig, 61, 1905.
SEVERINI.
BOCHER.
Venezia, Atti
Sulla serie di Fourier.
1st. ven., 64,
Introduction to the theory of Fourier's Series.
1905.
Ann. Math., Prince-
ton, N.J. (Ser. 2), 7, 1906.
FATOTJ.
(i)
Sur
integrables.
le
developpement en
Paris, C.R. Acad.
serie
trigonometrique des fonctions non-
142, 1906.
sci.,
Acta Math., Stock-
(ii) Series trigonometriques et series de Taylor.

holm, 30, 1906.
FEJR.
Sur
la serie
de Fourier.
Paris, C.R. Acad.
sci.,
142, 1906.
(Also paper in Hungarian in Math. Termt. Ert., Budapest, 24, 1906.)
The development of the theory of the transfinite numbers.

Arch. Math., Leipzig (3. Reihe), 10, 1906.
(See also 14, 1909 ; 16, 1910; and 22, 1913.)
JOURDAIN.
On an
LEES.
extension of the Fourier method of expanding a function in

and cosines. Mess. Math., Cambridge, 35, 1906.
series of sines
BRENCKE. On the convergence and

Ann. Math., Princeton, N.J. (Ser.
BUHL.
Sur
les
differentiation of trigonometric series.

3), 8, 1907.
nouvelles formules de sommabilite.
Bui.
sci.
math., Paris
(Ser. 2), 31, 1907.
FEJR.
Uber
die Fouriersche Reihe.
HOBSON. On the uniform convergence

Math. Soc. (Ser. 2), 5, 1907.
MOORE.
New
Note on Fourier's constants.
London, Proc.
York, N.Y., Bull. Amer. Math.
Soc., 13, 1907.
Uber
PRINGSHEIM.
die Fouriersche Integraltheorem.
Jahresber. D. Math.
Ver., Leipzig, 16, 1907.
RIESZ.
Les
BUHL.
Sur
series trigonometriques.
la
146, 1908.
Paris, C.R. Acad. sci., 145, 1907.
sommabilite des series de Fourier.
Paris, C.R. Acad.
sci.,
APPENDIX
311
DE LA VALLEY
POUSSIN. Sur 1'approximation des fonctions d'une variable

de leurs derivees par des polynomes et des suites limitees de
Fourier.
Bruxelles, Bui. Acad. roy., 1908.
reelle et
HARDY.
Some
points in
On certain
the integral calculus, XXIII. -V.
Mess. Math., Cambridge, 38,
oscillating cases of Dirichlet's integral.
1908.
ORLANDO.
Sulla formula integrale di Fourier.

(Ser. 5), 17, 1908.
YOUNG.
note on trigonometrical
series.
Roma, Rend. Ace.
Lincei
Mess. Math., Cambridge, 38,
1908.
BROMWICH.
FEJER.
Note on Fourier's Theorem.

Beispiele stetiger
(i)
Math. Gaz., London,
5, 1909.
Funktionen mit divergenter Fourierreihen.
J.
Eine
(ii)
stetige
Funktion
deren
Reihe
Fouriersche
divergiert.
Palermo, Rend. Circ. mat., 28, 1909.
GRAZIANI.
Sulla
(i)
Lincei>6er.
5), 18,
Roma, Rend. Ace.
formula integrale di Fourier.

1909.
Funzioni rappresentabili con la formula integrale di Fourier.
(ii)
Rend. Ace. Lincei
Roma,
(Ser. 5), 18, 1909.
ORLANDO. Jûove osservazioni sulla formula integrale

Rend. Ace. Lincei (Ser. 5), 18, 1909.
Roma,
di Fourier.
DE LA VALLÊ
POUSSIN. Un nouveau cas de convergence des series de

Palermo, Rend. Circ. mat., 31, 1910.
Fourier.
FEJ^R.
Lebesguesche Konstanten und divergente Fourierreihen.
J.
Math.,
Berlin, 138, 1910.
LEBESGUE.
Sur la representation trigonometrique approchee des fonctions

& une condition de Lipschitz. Paris, Bui. soc. math., 38,
satisfaisant
1910.
PLANCHEREL.
Contribution a
1'
etude de la representation d'une fonction

Palermo, Rend. Circ. mat., 30,
arbitrarie par des integrates defmies.

1910.
On
PRASAD.
Series.
the present state of the theory and application of Fourier's

Calcutta, Bull. Math. Soc., 2, 1910.
Uber neue
PRINGSHEIM.
Integralformel.
RIESZ.
Giiltigkeitsbedingungen
fur
die
Fouriersche
Untersuchungen
liber
Systeme integrierbarer Funktionen.
Ann., Leipzig, 69, 1910.

Rossi. Intorno ad un caso note vole negli integrali doppi di Fourier.
fis. mat. sc. nat., Pavia, 21, 1910.
YOUNG,
(i)
On
Math.
Riv.
the conditions that a trigonometrical series should have the

London, Proc. Math. Soc. (Ser. 2), 9, 1910.
Fourier form.
(ii)
On
the integration of Fourier Series.
(Ser. 2), 9, 1910.
London, Proc. Math. Soc,
APPENDIX
312
CARATHÔDORY. Uber den Variabilitatsbereich der Fourier' schen Konstanten

von positiven harmonischen Funktionen. Palermo, Rend. Circ. mat., 32,
1911.
The
DIXON.
multiplication of Fourier Series.
Cambridge, Proc.
Phii. Soc.,
16, 1911.
Sur le singularites de la serie de Fourier des

Ann. sci. EC. norm., Paris (Sen 3), 28, 1911.
FEJER.
HARDY.
and the theory
Fourier's double integral
fonctioris continues.
of divergent integrals.
Cambridge, Trans. Phil. Soc., 21, 1911.
NEUMANN.
Zur Theorie des logarithmischen Potentials.
Aufsatz VIII.
Nachtrag zu der Abhandlung Neue Giiltigkeitsbedingungen

PRINGSHEIM.
fur die Fouriersche Integralformel.
Uber die Summation divergenter Fourierscher Reihen.

SCHECHTER.
Hfte Math. Phys., Wien, 22, 1911.
Mon-
Nouvelle methode d'analyse harmonique par la summation

algebrique d'ordonees determinees. Paris, C.R. Acad. sci., 153, 1911.
THOMPSON.
Uber die Fouriersche Entwickelung positiver Funktionen.
TOPLITZ.
Rend.
YOUNG,
(i)
On
the convergence of a Fourier Series and of
London, Proc. Math. Soc.
On
(ii)
Palermo,
Circ. mat., 32, 1911.

its allied series.
(Ser. 2), 10, 1911.
the nature of the successions formed by the coefficients of a

Fourier Series.
On
(iii)
the Fourier Constants of a Function.
London, Proc. R.
Soc.,
85, 1911.
(iv) On a class of parametric integrals and their application in the
theory of Fourier Series. London, Proc. R. Soc., 85, 1911.
London, Proc. R. Soc.,
(v) On a mode of generating Fourier Series.
85, 1911.
On
(vi)
Edinburgh, Proc. R. Soc., 31,
Fourier's repeated integral.
1911.
(vii)
On Somnierfeld's form
of Fourier's repeated integral.
burgh, Proc. R. Soc., 31, 1911.

(viii) Uber eine Summationsmethode
f iir
Edin-
die Fouriersche Reihe. Leipzig,
Ber. Ges. Wiss., 63, 1911.
Konvergenzbedingungen fur die verwandte Reihe einer FourierMiinchen, SitzBer. Ak. Wiss., 41, 1911.
(ix)
schen Reihe.
On
CHAPMAN.
Fourier's
GRONWALL.
the general theory of summability, with applications to

series.
Q.J. Math., London, 43, 1912.
and other
(i)
Uber
die Gibbs'sche Erscheinung
Summen sin x + \ sin 2x + ...+(ii)
New
On a theorem
of Fejer's
sin nx.
und
die trigonometrischen
and an analogon
to Gibbs'
York, N.Y., Trans. Amer. Math. Soc., 13, 1912.
phenomenon.
APPENDIX
313
HARDY.
Oscillating Dirichlet's Integrals.
JACKSON.
On approximation by trigonometric sums

New
Q.J. Math., London, 43, 1912.
and polynomials.
Uber das Poissonsche Integral und iiber die partiellen Ableitungen zweiter Ordnung des logarithmischen Potentials. J. Math.,
LICHTENSTEIN.
Berlin, 141, 1912.
NEUMANN.
Zur Theorie
des
logarithmischen
Potentials.
Aufsatz
IX.
Remarks concerning Fourier's Theorem as applied to physical

Phil. Mag., London (Ser. 6), 24, 1912.
Uber die Konvergenz einer Fourierschen Reihe.
Gottingen,
EAYLEIGH.
problems.
THOMAS.
Nachr. Ges. Wiss., 1912.
YOUNG,
On
(i)
Math. Soc.
On
(ii)
successions of integrals
and Fourier
Series.
London, Proc.
(Ser. 2), 11, 1912.
multiple Fourier Series.
(Ser. 2),
11, 1912.
On
(iii)
a certain series of Fourier.
(Ser. 2),
11, 1912.
(iv)
Series.
(v)
Note on a certain functional reciprocity in the theory of Fourier

Sur la generalisation du theoreme de Parsival.
Paris, C.R. Acad.
155, 1912.
sci.,
On the convergence of certain series involving the Fourier constants
(vi)
of a function.
(vii)
London, Proc. R.
Soc., 87, 1912.
On classes of summable functions and their Fourier Series. London,
Proc. R. Soc., 87, 1912.
On
(viii)
the multiplication of
London, Proc. R.
FEJER.
Sur
(i)
les
successions of
Fourier
Constants.
Soc., 87, 1912.
polynomes trigonometriques.
Paris,
C.R. Acad.
sci.,
157, 1913.
(ii)
Uber
reihe.
die
Bestimmung des Sprunges der Funktion aus ihrer
Fourier-
J. Math., Berlin, 142, 1913.
paper
(Also
in
Hungarian in
Math. Termt. Ert., Budapest,
31,
1913.)
On
FILON.
a symbolic proof of Fourier's Theorem.
Math. Gaz., London,
7, 1913.
HARDY,
calcul
(ii)
(i)
"
"
InfinitarOscillating Dirichlet's Integrals, an essay in the
of Paul du Bois-Reymond.
J.
Math., London, 44, 1913.
Q.
On
the summability of Fourier's Series.
London, Proc. Math.
Soc. (Ser. 2), 12, 1913.
HARDY and LITTLEWOOD.

sommable.
Sur
la serie
Paris, C.R. Acad.
sci.,
de Fourier d'une fonction a carre

156, 1913.
APPENDIX
314
JACKSON,
On
(i)
the approximate representation of an indefinite

integral
and the degree of convergence

N.Y., Trans. Amer. Math. Soc.,
(ii) On the accuracy of trigonometric interpolation.

Trans. Amer. Math. Soc., 14, 1913.
KUSTERMANN.
New
York, N.Y.,
Uber Fouriersche Doppelreihen und das Poissonsche Doppel-
Sur
convergence des series trigonometriques de Fourier.
la
C.R. Acad.
MOORE,
York,
Diss., Miinchen, 1913.
integral.
LUSIN.
New
of related Fourier's Series.

14, 1913.
sci.,
Paris,
156, 1913.
On convergence factors in double series and the double Fourier

New York, N.Y., Trans. Amer. Math. Soc., 14, 1913.
On the summability of the double Fourier's Series of discontinuous
(i)
Series.
(ii)
functions.
NEUMANN.
Zur Theorie der Fourierschen Reihen.
Leipzig, Ber. Ges. Wiss.,
65, 1913.
Sopra un nuovo aspetto della formula

Roma, Rend. Ace. Lincei (Ser. 5), 22, 1913.
ORLANDO.
integrate de Fourier.
STEINHAUS.
Sur le developpement du produit de deux fonctions en une

de Fourier. Krakow, Bull. Intern. Acad., 1913.
serie
VAN
The
VLECK.
mathematics.
YOUNG,
On
(i)
Math. Soc.
On
(ii)
influence of Fourier's Series
Amer. Ass. Adv.
upon the development
of
Sc. (Atlanta), 1913.
the Fourier Series of bounded functions.
London, Proc.
(Ser. 2), 12, 1913.
the determination of the summability of a function by means

of its Fourier constants.
On the mode of oscillation of a Fourier Series and of its allied series.
(iii)

(iv)
On
(vi)
On
(Ser. 2), 12, 1913.
the usual convergence of a class of trigonometric series.

(v) On the formation of usually convergent series.
London, Proc.
R. Soc., 88, 1913.
Fourier Series and functions of bounded variation.
London,
Proc. R. Soc., 88, 1913.

(vii)
form.
On
the condition that a trigonometrical series should have a certain
London, Proc. R.
YOUNG, W. H. and
G. C.
Soc., 88, 1913.
On
the theorem of Riesz-Fischer.
London, 44, 1913.

BERNSTEIN. Sur la convergence absolue des
C.R. Acad. sci., 158, 1914.
BOCHER.
On
Gibbs's Phenomenon.
J.
Q.J. Math.,
series trigonometriques.
CAMP.
Lebesgue Integrals containing a parameter, with applications.

FATON.
Sur
la
convergence absolue des series trigonometriques.
Bui. soc. math., 41, 1914.
Paris,
New
Paris,
APPENDIX
315
Uber konjugierte trigonometrische Reihen.
FEJER.
Math., Berlin, 144,
J.
1914.
GRONWALL.
la serie
a
(i) Sur quelques methodes de sommation et leur application
de Fourier. Paris, C.R. Acad. sci., 158, 3914.
(ii) On the summability of Fourier's

Amer. Math. Soc 20, 1914.
New York,
Series.
N.Y., Bull.
KRYLOFF. Sur 1'equation de fermeture pour

Nouv. ann. math., Paris, 14, 1914.
les series
NEUMANN.
der
Uber
die
Dirichletsche
Theorie
trigonometriques.
Reihen.
Fourierschen
Leipzig, Abh. Ges. Wiss., 33, 1914.
POL.
Sur
les
On
YOUNG.
sci.,
London, Proc. R.
finitely.
FEKETE.
transformation de fonctions qui font converger leur series de

Paris, C.R. Acad.
Fourier.
Untersuchung
158, 1914.
whose Cesaro
summations
oscillate
(Hungarian.)
Math.
partial
Soc., 89, 1914.
liber Fouriersche
Reihen.
Terrnt. Ert., Budapest, 33, 1915.
GRONWALL. On Approximation by trigonometric sums.

Bull. Amer. Math. Soc., 21, 1015.
Zur Poissouschen Sumniierung.
GROSS.
New
York, N.Y.,
Ak. Wiss., 124,
Wieii, SitzBer.
1915.
KUSTERMANN.
Proof of the convergence of Poisson's integral for nonNew York, N.Y., Bull. Amer. Math.
absolutely integrable functions.

Soc., 21, 1915.
Some
RAYLEIGH.
calculations in illustration of Fourier's Theorem.
London,
Proc. R. Soc., 90, 1915.
Fourier Series and
BOSE.
mathematical analysis.
BROWN.
Fourier's Integral.
its
influence on
some
of
the developments of
Calcutta, Bull. Math. Soc.,
8,
1916.
Edinburgh, Proc. Math. Soc., 34, 1916.
Uber das Vorwiegen des ersten Koefficienten in der Fourierentwickelung einer konvexen Funktion. Math. Ann., Leipzig, 77, 1916.
FRANK.
GRONWALL.
Uber einige Sunmiationsmethoden und ihre Anwendung auf

die Fouriersche Reihe.
Jahresber. D. Math. Ver., Leipzig, 25, 1916.
HAHN.
Uber Fejers Summierung der Fourierschen Reihe.
Math. Ver., Leipzig,

PRIVALOFF.
(i)
Sur
Jahresber. D.
25, 1916.
la derivation
des series de -Fourier.
Palermo, Rend.
Circ. mat., 41, 1916.

(ii)
Sur la convergence des
C.R. Acad.
SIERPINSKI.
sci.,
Un
series trigonometriques conjuguees.
Paris,
162, 1916.
exemple
elementaire
presque partout une derivee nulle.
d'une function croissante qui a

Giorn. mat., Napoli, 54, 1916.
APPENDIX
316
YOUNG,
Sur
(i)
convergence des series de Fourier.
la
C.R
Paris,
Acad.
163, 1916.
sci.,
Les
(ii)
series trigonoraetriques et les
C.R. Acad.
Sur
(iii)
C.E. Acad.
moyennes de Cesaro.
Paris,
conditions de convergence des series de Fourier.
Paris,
163, 1916.
sci.,
les
sci.,
163, 1916.
Sur un precede de sornmation des
ANGELESCO.
Paris, C.R. Acad.
series
trigonometriques.
165, 1917.
sci.,
CAMP. Multiple integrals over infinite fields and the Fourier multiple
Amer. J. Math., Baltimore, Md., 39, 1917.
CARSLAW.
trigonometrical
Amer.
Series.
J.
sum and
the Gibbs'
Phenomenon
integral.
in Fourier's
Math., Baltimore, Md., 39, 1917.

00
Dart.
+ 2(a n cos A + &,,sm A, s)
Sugli sviluppi in serie i
(i
.e
dove
radici delle equazioni trascendente F(z) coairz + F^z) sin

mat., Milano (Ser. 3), 26, 1917.
FEJR.
Fourierreihe
HARDY.
und Potenzreihe. MonHfte. Math.
HARDY and LITTLEWOOD.

series
de Taylor.
On
HART.
Sur
la
Ann.
On
XLV.
a point
convergence des series de Fourier et des
Paris, C.R. Acad.

series.
trigonometric
= 0.
7T2
sono
Phys., Wien, 28, 1917.
Notes on some points in the integral calculus.
in the theory of Fourier Series.
le A, t
sci.,
165, 1917.
Ann. Math., Princeton, N.J.
(Ser.
2),
18, 1917.
JACKSON. Note on representations of the partial sum of a Fourier's

Ann. Math., Princeton, N.J. (Ser. 2), 18, 1917.
The Influence
JOURDAIN.
of Fourier's
Series.
Theory of the Conduction of Heat on

Scientia, Bologna (Ser. 2),
the development of pure mathematics.

22, 1917.
KUSTERMANN.
J.
Fourier's Constants of functions of several variables.
Amer.
Math., Baltimore, Md., 39, 1917.
Sur
LUSIN.
la notion
On
POLLARD.
de
1'integrale.
Ann. mat., Milano
from Fejer's theorem concerning

Proc. Math. Soc. (Ser. 2), 15, 1917.
Series
VAN VLECK.
YOUNG,
(i)
On
On
New
London,
York, N.Y.,
Soc., 23, 1917.
the order of magnitude of the coefficients of a Fourier Series.
London, Proc. R.
(ii)
their summability.
Current tendencies of mathematical research.
Amer. Math.
Bull.
(Ser. 3), 26, 1917.
the deduction of criteria for the convergence of Fourier's
Soc., 94, 1917.
ordinary convergence of restricted Fourier Series.
London,
Proc. R. Soc., 94, 1917.

(iii)
Series.
On
the mode of approach to zero of the

London, Proc. R, Soc., 94, 1917.
coefficients of a
Fourier
APPENDIX
suite de conditions
Sur une nouvelle
(iv)
de Fourier.
series
Sur
(v)
C.R. Acad.
Sur
(vi)
Paris, C.R. Acad.
la theorie
sci.,
317
de
sci.,
pour
la
convergence des series de Fourier.
la
Paris,
164, 1917.
Paris, C.R. Acad.
theorie des series trigonometriques.
la
convergence des
164, 1917.
165, 1917.
sci.,
CARATHEODORY.
Tiber
die Fourierschen Koefficienten der nach
Math.
integrierbaren Funktionen.
Uber
CARLEMAN.
Riemann
Zs., Berlin, 1, 1918.
Acta
die Fourierkoefficienten einer stetigen Funktion.
Math., Stockholm, 41, 1918.
GEIRINGER.
MonHfte. Math. Phys., Wien,
Trigonometrische Doppelreihen.
29, 1918.
The Fourier Harmonic Analysis
LARMOR.
"Ober die
RIESZ.
schrjinkter
YOUNG,
(i)
its
illustrations.
practical scope, with optical
(Ser. 2), 16, 1918.
Funktion von be-
Fourierkoemcienten eiuer stetigen
Math.
Schwankung.
On non-harmonic
Zs., Berlin, 2, 1918.
Trigonometrical Series.
London, Proc. R.
Soc.,
95, 1918.
On
(ii)
the Cesaro convergence of restricted Fourier Series.
London,
Proc. R. Soc., 95, 1918.
LANDAU.
Bemerkungen zu
einer Arbeit
von Herrn Carleman tJber die

Math. Zs., Berlin, 5, 1919.
:
Fourierkoefficienten einer stetigen Funktion.
MOORE.
Applications of the theory of summability to developments of

orthogonal functions. New York, N.Y., Bull. Amer. Math. Soc., 25, 1919.
YOUNG,
(i)
On
the convergence of the derived series of Fourier Series.
On
(ii)
(Ser. 2) 17, 1919.
restricted Fourier Series
and the convergence
of
power
series.
(Ser. 2), 17, 1919.
and Fourier Series.

(iii) On the connection between Legendre Series
(iv) On non-harmonic Fourier Series.
(Ser. 2), 18, 1919.
Note on a method of proof in the theory of Fourier's

York, N.Y., Bull. Amer. Math. Soc., 27, 1920.
JACKSON.
New
NEDER.
tiber die Fourierkoefficienten der Funktionen
Schwankung.
STEINHAUS.
(i)
Math.
Series.
von beschrankter
Zs., Berlin, 8, 1920.
Bemerkungen zu der Arbeit der He.rrn Neder t)ber die

der Funktionen von beschriinkter Schwankung.
:
Fourierkoefficienten
Math.
(ii)
Math.
Zs., Berlin, 8, 1920.
On
Fourier's coefficients of
Soc. (Ser. 2), 19, 1920.
bounded
functions.
London, Proc.
LIST OF
AUTHORS QUOTED
The numbers
refer to pages.
Gibson, 15, 195, 262.
Abel, 124, 146.
Gmeiner
Baker, 122.
Gronwall, 270.
Bliss, 77.
Bocher,
234, 243, 263, 268-271, 275,
8,
282.
Hardy, 13, 19, 20, 48,

Harnack, 112.
Bonnet, 97.
7.
Boussinesq,
Bromwich,
28, 48, 150, 157, 163, 165,
192, 193.
Burkhardt,
Heine, 11, 12, 14,26.

Hildebrandt, 77.
12, 17, 28, 69, 89, 94, 138, 163,
193, 234, 243, 263, 275.
2, 15.
Jackson, 270.
Jordan, 12, 207, 243, 294.
Jourdain, 14.
Byerly, 198.
Cantor, 12, 14, 15, 26, 27, 29.
Carse and Shearer, 298.
Carslaw, 155, 264.
Kelvin, Lord,
Kneser, 14,
8, 9, 77, 124.
6.
Kowalewski, 119, 163.
Cesaro, 12, 151.
Chapman,
75, 77, 94,1130, 239.
Hobson,
Brunei, 119, 193.
Cauchy,
and Gineiner).
163, 165, 193, 243.
1, 3.
Bernoulli,
(see Stolz
Goursat, 28, 48, 59, 75, 77, 94, 119, 153,
13.
Lagrange, 2, 3, 5, 6, 198.
Laplace, 6.
Lebesgue, 10, 12, 77, 112, 119, 243,
Chrystal, 148.
Clairaut, 3, 5.
D'Alembert, 1, 3.
Darboux, 4, 6, 7.
Dedekind, 18, 19,
Delambre, 6.
263, 294.
Legendre,
23, 25-28.
6.
Leibnitz, 26,
Lipschitz, 12.
De
la Vallee Poussin, 10, 12, 28, 48, 75,

77, 119, 163, 192, 193, 239, 243, 262,
Littlewood, 239.
263.
Moore, 150.
Descartes, 26.
Dini, 12, 28, 112, 119, 163, 193, 243.
Dirichlet, 4, 8, 9, 11, 77, 200, 208.
Neumann,
Newton,
243, 294.
26.
Donkin, 124.
Du Bois-Reymond,
12, 34, 77, 97, 112.
Osgood,
45,
58,
75,
119,
129,
163,
193.
Euclid, 25, 26.

Euler, 1-5.
Picard, 14, 234.

Pierpont, 75, 83, 86, 93, 110, 119, 163,
Fejer, 13, 234, 258.

Fourier, 2-9, 243, 294.
193.
Poincare, 7.
Poisson, 7, 8, 230, 231, 293.
Pringsheim, 16, 25, 28, 34, 48, 75, 163,
287, 291.
Gibb, 301.
Gibbs, 268-270, 282.
318
LIST OF
Raabe, 165.
Riemann, 5, 9-11,
Runge, 268, 297.
AUTHORS
319
Topler, 14.
15, 77, 112, 243.
Turner, 299.
Russell, 26, 28.
Van
Sachse, 15, 124.

Schuster, 301.
Watson (see Whittaker and Watson).

Weber-Riemann, 243, 294.
Vleck, 14.
Seidel, 11, 130.
Weierstrass, 14, 26, 27, 77, 97, 134.
Shearer (see Carse and Shearer).
Weyl, 271.
Sommerfeld, 293.
Whittaker, 297, 301.
Stekloff, 14.
Stokes, 11, 130.
Stolz, 119, 163, 193.
Whittaker and Watson, 151, 239, 240,
Stolz
and Gmeiner,
Tannery,
28.
243, 263.
16, 28, 48.
Young (Grace
Young, W. H.,
293.
Chisholm), 29, 77.

10, 12, 14, 29, 77, 287,
GENERAL INDEX
The numbers
Abel's theorem on the power series, 146
Absolute convergence, of
Absolute value, 32.
series,
44
refer to pages.
extensions
of, 149.
of integrals, 103.
Aggregate, general notion of, 29 ; bounded above (or on the right), 29 ; bounded
below (or on the left), 29 ; bounded, 30
upper and lower bounds of, 30 ;
limiting points of, 31 ; Weierstrass's theorem on limiting points of, 32.
;
Approximation curves for a
See also Gibbs 's phenomenon.
series, 124.
Bocher's treatment of Gibbs's phenomenon, 268.
Bounds (upper and lower),

f(x, y) in a domain, 72.
Bromwich's theorem, 151.
an aggregate, 30
of
of f(x) in
an
interval,
50
of
method of summing series (CI), 151, 238-240.

Change of order of terms, in an absolutely convergent
Cesaro's
convergent
series,
45
in a conditionally
series, 47.
Closed interval, definition of, 49.
Conditional convergence of series, definition of, 45.

Continuity, of functions, 59 ; of the sum of a uniformly convergent series of
continuous functions, 135 ; of the power series (Abel's theorem), 146 ; of
/**
/
f(x)dx when f(x)
is
bounded and
integrable,
93
of ordinary integrals
.'a
of infinite integrals involving a single

involving a single parameter, 169
parameter, 179, 183.
Continuous functions
theorems on, 60
of two variables,
integrability of, 84
73
non-differentiable, 77.
;
Continuum, arithmetical, 25
linear, 25.
of series, 41
of functions, 51 ; of integrals, 98.
Convergence, of sequences, 33
See also Absolute convergence, Conditional convergence, and Uniform convergence.
Cosine integral (Fourier's integral), 284, 292.
;
Cosine series (Fourier's series), 197, 215.
Darboux's theorem, 79.

Dedekind's axiom of continuity, 24.
Dedekind's sections, 20.
Dedekind's theory of irrational numbers, 18.
Dedekind's theorem on the system of real numbers, 23.
integrals containing an arbitrary parameter (Chapter VI.) ; ordinary
integrals, 169 ; continuity, integration and differentiation of, 169 ; infinite
integrals, 173 ; uniform convergence of, 174 ; continuity, integration and
differentiation of, 179.
Definite
320
GENERAL INDEX
321
Definite integrals, ordinary (Chapter IV.); the sums S and s, 77; Darboux's
theorem, 79 ; definition of upper and lower integrals, 81 ; definition of, 81 ;
necessary and sufficient conditions for existence, 82 ; some properties of, 87 ;
first theorem of mean value, 92 ; considered as functions of the upper limit, 93 ;
second theorem of mean value, 94. See also Dirichlet's integrals, Fourier's
integrals, Infinite integrals and Poisson's integral.
Differentiation, of series, 143 ; of power series, 148 ; of ordinary integrals, 170
of infinite integrals, 182 ; of Fourier's series, 261.
Dirichlet's conditions, definition of, 206.

Dirichlet's integrals, 200.
See also Infinite discontinuity

Discontinuity, of functions, 64 ; classification of, 65.
and Points of infinite discontinuity.
Divergence, of sequences, 37 ; of series, 41 ; of functions, 51 ; of integrals, 98.
FejeYs theorem, 234.

Fejer's theorem and the convergence of Fourier's
series, 240, 258.
Fourier's constants, definition of, 196.

Fourier's integrals (Chapter X.); simple treatment of, 284; more general conditions for, 287 ; cosine and sine integrals, 292 ; Sommerfeld's discussion
of, 293.
and VIII.) ; definition of, 196 Lagrange's treatproof of convergence of, under certain conditions, 210 ; for
even functions (the cosine series), 215 ; for odd functions (the sine series),
220 ; for intervals other than ( - TT, TT), 228 Poisson's discussion of, 230
orde~ of the terms in, 248
uniform convergence
Fejer's theorem, 234, 240
more general theory of, 262.
differentiation and integration of, 261
of, 253
Fourier's series (Chapters VII.
ment
of,
198
Functions of a single variable, definition of, 49 bounded in an interval, 50 upper

and lower bounds of, 50 oscillation in an interval, 50 limits of, 50 continuous, 59
monotonic, 66 ; inverse, 68 ; integrable,
discontinuous, 64
84 ; of bounded variation, 207.
;
Functions of several variables, 71.

General principle of convergence, of sequences, 34
Gibbs's
phenomenon
of functions, 56.
in Fourier's series (Chapter IX.)
264.
Hardy's theorem, 239.
Harmonic analyser (Kelvin's), 295.

Harmonic analysis (Appendix I.), 295.
Improper
'
integrals, definition of, 113.
See Aggregate.
Infinite discontinuity.
See Points of
Infinite aggregate.
infinite,
discontinuity.
Infinite integrals (integrand function of a single variable), integrand bounded and

interval infinite, 98 ; necessary and sufficient condition for convergence of,
100
with positive integrand, 101
absolute convergence of, 103
/x-test for
convergence of, 104 other tests for convergence of, 106 mean value theorems
;
for, 109.
Infinite integrals (integrand function of a single variable), integrand infinite, 111 ;

yu-test and other tests for convergence of, 114 ; absolute convergence of, 115.
Infinite integrals (integrand function of two variables), definition of uniform convergence of, 174 ; tests for uniform convergence of, 174 ; continuity, in-
tegration
and
Infinite sequences
differentiation of, 179.
and
series.
See Sequences and Series.
Infinity of a function, definition of, 65.
Integrable functions, 84.
GENERAL INDEX
322
Integration of integrals
ordinary, 172 ; infinite, 180, 183, 190.
Integration of series (ordinary integrals), 140 ; power series, 148
:
261
Fourier's series,
(infinite integrals), 154.
open, closed, open at one end and closed at the other, 49.
Inverse functions, 68.
Interval
See Numbers.
Irrational numbers.
Limits, of sequences, 33
repeated, 127.
of functions, 50
of functions of
two variables, 72
Limiting points of an aggregate, 31.
Lower
Mean
value theorems of the integral calculus ; first theorem (ordinary integrals),

;
(infinite integrals), 109 ; second theorem (ordinary integrals), 94 ; (infinite
92
integrals), 109.
See Absolute, value.
Modulus.
Monotonic functions, 66
admit only ordinary discontinuities, 67
integrability
of, 84.
Monotonic in the
stricter sense, definition of, 39.
M-test for convergence of series, 134.

/i-test for
convergence of integrals, 104, 116.
Neighbourhood of a
point, definition of, 52.
Numbers (Chapter I. );
rational, 16
irrational, 17 ; Dedekind's theory of irrational,
18 ; real, 21 ; Dedekind's theorem on the system of real, 23 ; development of
the system of real, 25. See also Dedekind's axiom of continuity, and Dedekind's
;
sections.
Open
interval, definition of, 49.
Ordinary or simple discontinuity, definition of, 65.

Oscillation of a function in an interval, 50
of a function of two variables in a
;
domain,
72.
Oscillatory, sequences,
Partial remainder ( P
Rn
38
),
series,
41
functions, 52
definition of, 42
integrals, 99.
(pR n (x)), definition
of, 123.
Periodogram analysis, 299.

Points of infinite discontinuity, definition of, 66.
Points of oscillatory discontinuity, definition
of, 66.
Poisson's discussion of Fourier's series, 230.

Poisson's integral, 231.
Power
series ; interval of convergence of, 145 ; nature of convergence

Abel's theorem on, 146 ; integration and differentiation of, 148.
Proper integrals, definition
123.
limits, 127.
Repeated integrals, (ordinary), 172

Sections.
146;
of, 113.
Rational numbers and real numbers. See Numbers.

Remainder after n terms (JRJ, definition of, 43 (R n (x)),
Repeated
of,
See Dedekind's
(infinite), 180, 183, 190.
sections.
limit of, 33
Sequences ; convergent, 33
necessary and sufficient condition for
convergence of (general principle of convergence), 34 ; divergent and oscilmonotonic, 39.
latory, 37
;
GENERAL INDEX
323
definition of sum of an infinite, 41 ; convergent, 41 ; divergent and oscilfor convergence of, 42 ; with
latory, 41 ; necessary and sufficient condition
44 ; definition
positive terms, 43 ; absolute and conditional convergence of,
of sum, when terms are functions of a single variable, 122 ; uniform convergence of, 129 ; necessary and sufficient condition for uniform convergence
-test for uniform convergence of, 134 ; uniform
Weierstrass's
of, 132 ;
and continuity of, 135 ; term by term differentiation and integra-
Series
convergence
tion of, 140.

series,
Power
See also Differentiation of

series
and Trigonometrical
series,
series.
Simple (or ordinary) discontinuity, definition

Sine integral (Fourier's integral), 284,
Sine series (Fourier's series), 197, 220.
Summable series (CI), definition

Sums S and s, definition of, 77.
Trigonometrical
of, 65.
29'2.
of, 151.
series, 196.
Uniform continuity of a function, 62.

Uniform convergence, of series, 129;
Upper
Fourier's series, Integration of
of integrals, 174.
Weierstrass's non-differentiable continuous function, 77.
Weierstrass's Jf-test for uniform convergence, 134.

Weierstrass's theorem on limiting points to a bounded aggregate, 32.
^
GLASGOW
FEINTED AT THE UNIVERSITY PRESS BY ROBERT MACLBHOSE AND CO. LTD.
14 DAY USE
RETURN TO DESK FROM WHICH BORROWED
LOAN DEPT.
This book
or
is due on the last date stamped below,
on the date to which renewed.
Renewed books
are subject to immediate recall.
llNotfBQAR
FLb
^lS
REC'D LD
flCU&W
REC'D CD
5Jun'60BM
MAY
270ct'58H
2 3 1060
Ul
ÂN-2-
(C8481slO)476B^
RL-.-D
>_ji
MAK23
1961
General Library
University of California
Berkeley
YC 91320
UNIVERSITY OF CALIFORNIA LIBRARY

Introduction To TH 00 Cars Rich

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Introduction To TH 00 Cars Rich

Загружено:

Авторское право:

Доступные форматы

INTRODUCTION TO THE

THEORY OF* FOURIER'S

Sc.D. (CAMBRIDGE), D.Sc. (GLASGOW), F.R.S.E.

SECOND EDITION, COMPLETELY REVISED

MARTIN'S STREET, LONDON

GLASGOW: PRINTED AT THE UNIVERSITV PRESS

Heat Conduction, that it has seemed advisable to

concerned with the Theory

Integrals, with special reference to Fourier's

Mathematical Theory of the Conduction of Heat.

and integrals. With these questions is bound up

The theory of series whose terms are functions of a

of Fourier's Series in Chapter VII. depends

the Second Theorem

There, and elsewhere throughout the book,

of the argument. In the same chapter the work of Poisson is

to Fejer's work, both in the proof of the fundamental theorem

in the discussion of the nature of the convergence of Fourier's

devoted to Gibbs's Phenomenon, and the

for which f(x) is defined can be broken up into a

and finite in number. Such functions will satisfy most of the

of Lebesgue, has introduced into the Theory of Fourier's

Theory of Functions of a Real Variable. My purpose has been to

has been given as a

course in the Infinitesimal Calculus

revision of the proofs and for many valuable suggestions, I

Mathematics in the University

EATIONAL AND IRRATIONAL NUMBERS

Relations of Magnitude for Real

The Linear Continuum.

The Upper and Lower Bounds

Limiting Points of an Aggregate

Infinite Set of Intervals

19-23. Infinite Series

FUNCTIONS OF A SINGLE VARIABLE. LIMITS AND

Some Theorems on Limits

Sufficient Conditions for the Existence of a

30-32. Continuous Functions

Functions of Several Variables

Possibility of Expressing a Function as the Difference of

The Sums S and

Necessary and Sufficient Conditions for Integrability

in (a, &) has an Infinite Number of Points

of the Definite Integral

The Second Theorem

Bounded Integrand. Infinite Interval

59-61. Infinite Integrals.

Definite Integral considered as a Function of its

Series of Continuous Functions

Extensions of Abel's Theorem on the Power Series

64. Repeated Limits

74-76. Integration of Series.

DEFINITE INTEGRALS -CONTAINING AN ARBITRARY

Continuity of the Integral

Differentiation of the Integral

Integration of the Integral

80-83. Infinite Integrals

Continuity of the Integral ( F(x,y)dx

Integration of the Integral

Differentiation of the Integral/

87. Properties of the Infinite Integral

Applications of the Preceding Theorems

The Repeated Integral ( dx