Jiongmin Yong - Mathematical Analysis - A Concise Introduction-World Scientific Pub Co Inc (2020)

MATHEMATICAL
ANALYSIS
A Concise Introduction
MATHEMATICAL
ANALYSIS
jjpi;
*
.
Wmm
’
Siiii#
: -v
•
lit
fils wR
r JV
;
^-Y A
* V* .
'
fi fi fi
'
-
0m
' '
' • .
.. - .
M ^ isais
!
Are
JrS’ .
Mm
m
. .
fi,-
Jiongmin Yong
University of Central Florida, USA
World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONGKONG • TAIPEI • CHENNAI . TOKYO
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
Library of Congress Control Number: 2020055934
British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.
/
MATHEMATICAL ANALYSIS
Copyright © 2021 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
-
ISBN 978 981-122-163-7 ( hardcover)
-
ISBN 978-981-122 164-4 (ebook for institutions)
- - - -
ISBN 978 981 122 165 1 (ebook for individuals)
For any available supplementary material, please visit

https://www.worldscientific.eom/worldscibooks/10. l 142/11859#t=suppl
s
In the Memory of My Mother
4
:
i.
s
i
>
t
>
I
* !
)
i
?
1
J
'
l
f
;
Preface
Mathematical analysis serves as a common foundation for many research

areas of pure and applied mathematics. It is also an important and pow-
erful tool used in many other science fields, including physics, chemistry,
biology, engineering, finance, economics, to mention just a few. On the
other hand , it is a desire that after having taken a sequence of courses
on basic calculus and linear algebra, one could spend a reasonable length
of time ( ideally, say, one semester ) , to build an advanced base of analysis
sufficient for getting into various research fields other than analysis itself ,
and /or stepping into more advanced level (s) of analysis courses (such as
real analysis, complex analysis , differential equations, functional analysis,
stochastic analysis, etc. ). To meet such a need , I wrote this concise intro-
duction on mathematical analysis for a course that can be completed within
one semester.
From the author’s viewpoint , mathematical analysis should present a
theory for the functions between the best regular ones ( meaning analytic
functions which are treated in complex variables / analysis) and the worst
regular ones ( meaning measurable functions which are treated in real analy-
sis ). More precisely, this course discusses functions with various continuity,
differentiability, Riemann integrability, and relevant topics of limit , series,
and uniform convergence.
For the convergence and continuity of functions, the linear ( or algebraic )
structure of the underlying space is irrelevant . Therefore, we consider func-
tions and sequences in general metric spaces. As a preparation for that ,
some basic properties of metric spaces are discussed , such as completeness,
compactness, connectedness and separability, etc. For differentiability and
Riemann integrability of functions, the linear structure of the underlying
space is necessary. Therefore, we discuss the problems of differentiation
and integration in the m-dimensional Euclidean space.
Vll
vm Mathematical Analysis — A Concise Introduction
Series is an independent topic. Partially, it is an application of the
previous chapters if one views it from the limit of partial sums. On the
other hand , the study of the series convergence has lots of its own feature.
Power series and Fourier series make this part of theory very rich . Also, the
theory of series can even touch one of the most challenging open questions
in mathematical research: the Riemann hypothesis.
Real number system is traditionally included in the course of mathe-
matical analysis. Due to the time limit of the course, we decided to put it
in a self-contained appendix at the end of the book so that the interested
readers could still be able to read it separately, and it becomes optional for
those readers who are only interested in the main body of mathematical
analysis other than the real number system.
Finally, problems listed at the end of sections are complementary to the
material covered in the main body of the book , and those with asterisks
might be a little difficult. Readers could skip them at the first reading.
Jiongmin Yong
Orlando , USA
July, 2020
Contents
Preface Vll
1. Metric Spaces and Limits for Sequences 1

1 Metric Spaces 1
2 Sequences and Limits 10
3 Sets in Metric Spaces 21
4 Properties of Metric Spaces ... 31
2. Functions on Metric Spaces 45

1 Continuity 45
2 Properties of Continuous Functions 52
2.1 Compactness preserving and uniform continuity . 53
2.2 Connectedness preserving 58
3. Differentiation 61
1 Derivatives 61
2 Frechet Differentiability 76
3 Inverse Function and Implicit Function Theorems . ... 80
4 Higher Order Derivatives 89
4. Riemann Integrals 95
1 Definition of Integrals 95
2 Properties of Integrals 107
3 Further Theorems . . 112
IX
x Mathematical Analysis — A Concise Introduction
5. Uniform Convergence 121

1 Observations and Examples . . . . 121
2 Uniform Convergence 123
3 The Metric of Uniform Convergence 132
4 Limit Theorems 140
6. Series 151
1 Series of Numbers 151
2 Further Tests for Convergence . 157
3 Series of Functions 167
More Convergence Criteria . .
'
4 175
5 Power Series 179
5.1 A general consideration 179
5.2 Exponential and logarithmic functions 185
5.3 The sine and cosine functions 187
<
6 Fourier Series 192
6.1 Inner product 193
6.2 Trigonometric polynomial approximation 196
6.3 Fourier series . 198 t
Appendix The Real Number System 207

1 Natural Numbers 207
1.1 Peano axioms . 207
1.2 Addition and multiplication ... 209
1.3 Orders 214
2 Integers and Rationals 218
2.1 Integers 218
2.2 Rational numbers 226
3 Real Numbers . 236
3.1 Sequences of rational numbers . . 236
3.2 A construction of real numbers . . 237
3.3 Further properties of real numbers 243
3.4 Real exponentiation 252
4 Complex Numbers 256
Bibliography 257
Index 259
Chapter 1
Metric Spaces and Limits for

Sequences
1 Metric Spaces
Let us begin with some notions on set which is understood as a collection

of distinct objects.
Definition 1.1. ( i ) Let 5 be a set . If x is an element of 5, we denote
x E 5. If x is not an element of 5, we denote x £ S .
( ii) Let A and B be two sets. If
xE B Wx e A
we call A a subset of B , denoted by A C B.
( iii) Let A and B be two sets. We define the union, intersection and
difference of A and B as follows:
Au B = { x \ x e A, or x e B },
A D B = { x | x E A, and x E B } ,
A\ B = { x | x E A, and x £ B }.
If X is a fixed set and all the considered sets are in X , then for any S Cl,
the set Sc = X \ S is called the complementary set of S ( in X ). For
any family A\ ( A E A ) of sets, we can define the union (JAGA
intersection HAGA in the similar way as above.
^^
an(
(iv ) If a set S contains no element , it is called an empty set, denoted by

5 = 0. Otherwise, it is called a non- empty set
The following collects some basic results concerning the operations of
sets, whose proofs are straightforward .
Proposition 1.2. ( i ) Let A, B , C be subsets of X . Then
AU B = BU A A n B = B n A,
l
2 Mathematical Analysis — A Concise Introduction
and
iu(BnC)=(AuB)n(iu C ) ,
A n ( B i) C ) = ( A n B ) u ( A r C ) ,
A\ B = A n B c .
(ii) (de Morgan’s Law ) Let {A\ A £ A} be a family of subsets in
X . Then
(u aA = n
AGA AGA
(nHAGA
= u
AGA
We now introduce the following definition.
Definition 1.3. Let X be a non-empty set. A map d : X x X — R is

called a metric on X if the following hold:
( a ) ( positivity ) For any x , y X , d( x , y ) 0 and d ( x , y ) = 0 if and only
if x = y.
( b ) (symmetry ) For any x , y e X , d ( x , y ) = d ( y , x ).
( c ) (triangle inequality) For any x , y , z <E X , it holds:
d ( x , 2i ) d ( x , y ) + d ( y , z ). ( i.i )
In this case, ( X , d) is called a metric space. The metric d ( , •) is also called -
a distance on X .
We present some examples.

Example 1.4 (i) Let X. = R be the set of all real numbers,1 and
p{ x , y ) = \ x - y \ x , y G R.
Then a straightforward proof shows that /> ( - , ) is a metric on R, which is
*
called the standard metric on R , and ( R , p ) becomes a metric space.

( ii) Let X be a non-empty set . Define do : X x X —> R as follows:
x = y,
do ( x , y ) = ° x y.
1
^
It is easy to see that do ( , •) is a metric which is called the discrete metric on
*
X . Such a metric can always be defined on any non-empty set X . Hence,

for any non-empty set X , one can always define a metric so that it becomes
a metric space.
1See the Appendix for a theory of the real number system .
Metric Spaces and Limits for Sequences 3
Example 1.5. Let ( X , d ) be a metric space and Y Cl. Then the restric-
tion d |Y x Y of d on Y x Y is clearly a metric on Y; ( Y, d|y x y ) is called a
subspace of ( X, d ) ; and d | y x y is called the metric on Y induced by d . For
convenience, hereafter , we simply denote ( Y, d) instead of ( Y, d|y x y ) . Un-
der the standard metric , N = {1, 2, 3, • • • } ( the set of all natural numbers) ,
Z ( the set of all integers) , Q ( the set of all rational numbers ) , and interval
[a , b] are all subspaces of R.
To have more interesting examples, for any integer m 1, we define
Rm = { x = (x1, x2, • • • , xm ) | xk G R , 1 /c m}.

For any x = ( x1, #2 , - - , xm ) , t/ = (2/19 2/ 2 , • • , y m ) £ Rm and
* * A 6 R , we
introduce
x + y = (x1 + 2/1, x 2 -f y2 , - - - , xm + t/ m
), Ax = ( Ax1, Ax 2 , - - - , Axm ).
These operations are called addition and scalar multiplication, respectively,
for which the following eight axioms are satisfied:
' x + ?/ = y 4- x, Vx, y G Rm
( x + y ) + z = x + ( y 4- z ) Vx , ?/ , 2 G R m
x+0 = 0+x = x Vx G Rm
® + ( -x ) = 0 , Vx Rm
A ( x + y ) = Ax + Ay , Vx , y G Mm A G R ,
( A 4- p ) x = Ax f /ix ,
- Vx G Mm , A , fi G R ,
lx = x Vx G Rm
. A (/zx ) = ( Ay) x, Vx G Rm A , fi G R.
With the addition and scalar multiplication , Rm is called a linear space ( or
a vector space ) . Each x G Rm is called a vector. The addition and scalar
multiplication are called algebraic operations on Rm . Any properties of
Rm involving addition and scalar multiplication are referred to as algebraic
properties of Rm .
Now , for any x = (x1, x 2 , • • • , xm ) G Rm , we define
771 2.
II*IIP =
( EVr)' p e [i , oo )
( 1.2 )
=
k 1
max P = oo ,
^^
l /c ra
which is called the £ p- norm. We refer to ( Rm • ||p ) as a normed linear

space. The following proposition is concerned with the fp- norm.
Proposition 1.6 . L e t p e [ l , oo] . Then the above-defined £ p-norm satisfies

the following:
( i) ( positivity ) ||x ||p for all x £ Rm , and \ \ x \\ p = 0 if and only if
x = 0. ^0
( ii ) ( positive homogeneity ) || Ax ||p = |A|||x || p , for all x £ Rm and A £ R .
( iii) ( triangle inequality ) ||x -F y \\ p ||x ||p + \ \ y \\ p , for all x , y £ Rm .
From the definition of fp- norm , we easily see that (i) — ( ii) are true. To
prove ( iii ) , we first look at the cases p = 1, oo. One has
x + vh =
k =1
m
F + yk \
k 1
m
H= lxfcl + H=
m
k 1
= ll ^ ll 1 + IMIi .
and
x + y || oo = max |x* + {/fc|
max |xfe| + max |j/fc| = 11 x 11 «, + \\ y\\oo -

Now , we look at the case p = 2. If y = 0 , the proof is trivial. Thus, let
x, y £ R with 2/ 7^ 0. For any A £ R , one has
m m
=£
Taking
0 \\ x - X y\\ l
fc = l
(*fc - A / ) 2 = H II
1 2 - 2A
^
fc = l
x V + A2||y|| .
^
- m
A = 5
we obtain
0 n 0 1 m 0
AI ( £ *•*)
0 ||x||2 ~
+
which leads to
m
Y= V x, y
fc
^ l
xkyk < ||x|| 2 ||y ||2 e Rm . (1.3)
This is called the Cauchy- Schwarz inequality Then it follows that .

m m
II* + y||!= £ (** + = £ ( 1*k 2 + 2 xkyk + |yfe|2 )

|
=
k l k =1
m
2
= IMI 2 + 2 £ ®V+ I|y|l 2 < Ha; l 2 + 2 lla; ll 2 ||y ||2 + ||2/||2 = (||a;||2 +||2/||2 ) -
/c = l
Consequently, (iii) holds for p = 2.

For general p £ (1, 2 ) U ( 2, 00 ) , the situation is a little more subtle. We
need the following results which are interesting by themselves.
Proposition 1.7. Let p , q £ ( 1 , 00 ) such that = 1.

^^
( i) ( Young’s Inequality ) For any a , b £ (0, 00 ) , the following holds:
ap b«
.
ab — -b .
^ p —q (1.4 )
( ii) ( Holder’s Inequality ) For any x , y £ Rm , the following holds:

m
£
=
*V
k 1
IMIpIlyllg . (1.5)
(iii) ( Minkowski’s Inequality ) For any x , y £ Rm , the following holds:
\\ x + y \\ p ^ I ^ IIP + IIPIIP * ( 1.6 )

Proof , ( i) Let
./ x rp 1
fir ) = + - - r,
p
/
9 — r £ [0, 00 ) .
We try to find the minimum of / ( •) . To this end , 2 setting

_1
f' ( r ) = rp - 1 = 0,
we get r = 1. Clearly, by the first derivative test , one has
This is equivalent to
f (r ) / (1) =
^^ +
rp
P
1
f-
9
— 1 = 0,
r.
Vr £ ( 0 , 00 ) .
(1.7)
2 We assume that single variable calculus is known to the readers.
i
Taking r = ab 1 ~ p , one has
r p 1 a p b 1
=r p
h-=
q p +q
Consequently,
/ apb 1 ~ p _ aP_
+ -qJ) b
i
ab = ab 1 ~ p b
which gives (1.4 ) .
! P- < V p ^ _
v q
( ii) Let
£ wy
m m
M ,= {
i i
A=
k=l k= l
B = lly||« = ( El / l ) ? - 9
If AB = 0 , then x = 0 or y = 0 which leads to the conclusion trivially. We

now let AB 0. Denote
^ yk
bk = \ B \ k = 1, 2, • • • , m .
A ’ ’
Then by Young’s inequality,
( aky ( bky
akbk q
P
which results in
m m m m
l I l
AB k l,
=
Y
xky
k =1
pAp
k=1
k |p
+ qBQ El
=
k 1
y k iq
1 1
Thus
« l| |l" + « ll 2 l ? l 2 l 1^
X
/
/
P
+q L
-
m f
E= *V
k 1
AB = INIpIlylU
proving (1.5 ) .
(iii ) In the case that \\ x + y \\ p = 0 , the conclusion is trivial. We assume
\\ + 2/||p > 0 below. By Holder’s inequality,
x
m m
x + y \\ l
m
= E= lxfe +
k l
y k ip =5
m
fc l=
> + /r v +
fc
t/
fe
i
—
E
=
\*
k 1
k
+ y k |p 1
i i + Ei +
^ fe = i ^ ^ r ^! 1
m _ i m i
II*HP ( E i * + i
fe = l
® y k (p 1) ?
) 5
+ MiP ( E ix* + yk \ (p
fc = l
~ 1)9
) 5
m
( lNIP + IMIp ) ( Elxfe + yfelP) * = ( ll*llp + lly|| ) ll* + yllp
k=1
p
f
£
Dividing ||x + y \\ p > 0 in the above, we get
NIP + \\ v \\ p > \\ x + y \\ p = H ® + y \\ p .
q
This completes the proof.

Although in Proposition 1.7, the Holder’s and the Minkowski’s inequal-
ities are proved only for the cases p, q G ( l , oo ) with = 1, the cases
p = 1 ( then q = oc ) and p = oo ( then q = 1) are also true with very simple
proofs.
^^
Note that for any p G [1, oo ) , the following estimates hold:
i m i
INIloo = ( (EixV) 5
( m max |*Y "

max
^) (
k=l
= rap
i
x oo ?
®
Vx
IIP
G Rm .
Consequently, for any 1 p, q oo,
_i
m 9 ||x ||q ||x OO IMIp rap ||x||oo rap \\ x \\ q , Vx G Rm . (1.8)
p
This is referred to as the strong equivalence among all the
Rm ) .3
^ -norms (in
Now , we define
PP ( x , y ) = \\ x - y \\ p , x , p G Rm , p G [1, oo].
It is clear that properties ( a) and ( b ) in the definition of metric (see Def-
-
inition 1.3) are satisfied by pp ( , ) . The triangle inequality follows from
*
-
Minkowski’s inequality. Therefore, pp ( , •) is a metric on Rm , which is
called the metric induced by the p
^ -norm. We call p2 the Euclidean metric.
From (1.8) , we see that
ra * P q ( x , y ) Pp( x , y ) pq ( x , y ) , Vx, y G X. (1.9)
The above is referred to as the strong equivalence among all the metrics pp
induced by the -norms ( in Rm ) .4
^
Proposition 1.8. Let ( Xk , dk ) , k = 1, 2, • • • , n , be n metric spaces. Define
X = Xi x X 2 X • • • x Xn = { ( xi , X 2 , • • • , xn )
I Xk G Xk , 1 k nj ,
3 Note that in ( 1.8) , the dimension m of Rm is involved . Therefore, such an equivalence
is not expected for infinite dimensional spaces.
4 Similar to the p norm , the strong equivalence of all the metric
- p p only holds in finite-
^
dimensional normed linear spaces.
and
n
dk ( % k - V k )
J
^n ) ? ( ?/ l 5
Then ( X , d) is a metric space, called the product
* * ’ j
^/n ) )
^^
/c =1
of
) • ( 1.10 )
( X i , d i ) , • • • , ( Xn , dn ) .
The proof is obvious. We see that the conclusion of Proposition 1.8
remains true if (1.10 ) is replaced by
n 1
d( ( x xn ) , ( y i , ,V n ) ) = ( E dk ( X k , V k )P) ” 1 p < 00
fc = l
or
d ( ( xi , - - - , xn ) , ( y i , - - - , y n ) ) = sup dk ( x k , y k ).
Let us present the following additional example.
Example 1.9.5 For p £ [ l , oo ) , let
00
t? =|(a:fc ) = ( xk ) k i | xk R , /c > 1; E [x k ip
< OO }
^ k= l
and let
t°° = { ( xfc ) = ( xk ) > k 1 xk £ R , fc 1; sup \ xk \ < oo > .
We define
k
^ l
OO
PP { { xk ) , { yk ) ) = ( E ^*
k =1
- / ip)
p
V ( xk ) , ( yk ) e i p , p e [1, oo ) ,
and
Poo ( ( xk ) , ( yk ) ) = sup |xfe - ykI V( xfe ) , ( / ) s
k l
^ ^.
-
Then pp ( , •) is a metric on (p £ [l , oo] ) . In fact , again , properties ( a)
and ( b) in the definition of the metric ( Definition 1.3) hold for pp ( , ) in - -
f p . Now for the triangle inequality, we let N 1 and from the Minkowski’s
inequality, one has ^
( E i*
^^ ) ( E i** r) + ( E I »* IP ) lp
fc P
+ •
fc = l fc = l k =1
Letting N o o , one can rigorously show that
llx + y \\ p li ^ llp i II / IIP ?
^ “"
^
Hence, pp ( - , ) is a metric on
• lp.
5Since some results on series will be involved , those who axe not familiar with this could
skip this example until they have read Chapter 6.

We point out that unlike Rm , the spaces £ p are different for different
p G [1, oo ) , and all pp are not strongly equivalent.
To conclude this section , let us further look at the case p = 2. On Rm ,
we introduce a binary operation , called an inner product:
m
( x , y ) = x T y = Ylxky Vx, ?/ G Rm .
k=l
With such an inner product , Rm is called an inner product space. Clearly,

v ( x, y ) ( x , y ) satisfies the following conditions:
( x, y) = ( y, x ) Vx , y G Rm
( X x + y y , z ) = X ( x , z ) + /i ( y , z ) , VA , y e R , x , y , z £ Rm ( l. n )
( z , z ) 0, (x , x) = 0 x = 0.
Further , one sees that
||x||2 = y/ ( x , x ) Vx G Rm .
Therefore, we say that || • H 2 is the norm induced by the inner product ( * , ).
*
We call || • ||2 the Euclidean norm.
Exercises
1
1. Prove the de Morgan’s Law: Let { A\ | A G A} be a family of sets in
R. Then
( iMA = n A x
A A xeA
A? ( n ^)
A6A
= u ^-
AGA
2. Let d\ and d 2 be two metrics on the non-empty set X . Then for any
Ai , A 2 > 0, Aidi + A 2 2 is also a metric on X .
^
3. Define
x - y \\ p
d( x , y ) =1 Vx , y Rm
+ \\ x - y \\ P
where || • ||p is defined by (1.2 ) . Prove or disprove that d ( , •) is a metric on *
Rm .
-
4. Let d{ , •) be defined on R m as follows:
d( x, y ) = { k | xk yk , = , , , m} ,
K
^ 1
k
Vx = ( x , - - - , xm , — ( ) y
1 2 •
y
•
1
, • • , ym ) G Rm
where \ A\ stands for the number of elements in A. Then d ( • , •) is a metric
on Rm .
5. Let X be a non-empty set and let p : X x X -» [0 , oo ) be a map
satisfying the following:
p( x , y ) 0 , V x , y e X ; p( x , y ) = 0 x = y,
^
p( x , y ) p ( z , z ) + p( z , y ) Vrr , y , z e X .
>
Define
d ( x , y ) = m a x { p( x , y ) , p { y , x ) } Mx , y e X .
Then d ( • , •) is a metric on X .
6. Let ( X, d ) be a metric space and / : X -» X be a one-to-one map.
Define
d( x , y ) = d ( f ( x ) , f ( y ) ) x, y X .
Then d ( - , •) is a metric on X.
7*. Let and be defined in Example 1.9. Then £ l C
^ , and
2
^
2 Sequences and Limits
We introduce the following definition.

Definition 2.1. Let {xn }n sequence in a metric space ( X , d ) .
^ i be a F
( i) The sequence is said to be bounded if there exists an x G X and a
constant M > 0 such that
d ( xn , x ) M , n 1.
( ii) The sequence is said to be Cauchy if for any e > 0, there exists an
N 1 such that
d { X n i %m ) £ » n, m N .
^
(iii) The sequence is said to be convergent to x G X with respect to d if
for any £ > 0, there exists aniV l such that
^
d ( xn , x ) < e , n N.
^
^-
In this case, x is called a Zirait of the sequence {xn}n i The above is also
denoted by the following:
lim d ( xn , x ) = 0.
n —toc
When the metric d is clear from the context , we simply denote the above
by
lim xn = x .
—
n > oo
The sequence is said to be divergent if it is not convergent . M
From the above definition , we see that a sequence { xn }n i is bounded ,

Cauchy, convergent , or divergent , if and only if so is the sequence { xn }n k ^
for any fixed k 1. In other words , the above-mentioned properties remain ^
^
if we drop any first finitely many terms of the sequence. Likewise, by adding
finitely many terms at the beginning of the sequence, the above- mentioned
properties will not be changed. Hence, from now on , we will simply denote
the sequence by { xn } instead of { xn }n i , unless the initial term needs to
be emphasized. ^
The following result gives some basic properties of convergent sequences.
Proposition 2.2 . Let ( X , d ) be a metric space and { xn } C X be a se-

quence.
( i) If { xn } is convergent to x G X and to x' Gl , then x = x' .
( ii) If { xn } is convergent , then it is Cauchy.
( iii) If { xn } is Cauchy, then it is bounded .
Proof , (i) Suppose { xn } has two different limits x , x' G X . Then for
d ( x , x' )
£= > 0 , there exists an N 1 such that
2
^
d { xn , x ) + d ( xn , x' ) < £ , Vn
^ N.
Consequently, for any n
^ N,
0 < d ( x , x' ) d ( x , xn ) 4- d ( xn , x' ) < e = ^
a contradiction.
( ii) Let {xn } converge to x . Then for any e > 0, there exists an N
such that ^1
|
d ( xn , x ) < Vn N.
Consequently, for any m , n

^) Nd, (we ,have
d ( xn , xm xn + xm , x ) < e .
x ) d(
This shows that { xn } is Cauchy.
( iii) Let { xn } be Cauchy. Then for e = 1 > 0, there exists an iV
such that ^1
d ( xn , x N ) < e = 1 Wn
^ N.
Now letting
N -l
M =1+ d (, X k , X N )
k= l
one has
d(xn, XN ) M, Vn 1.
This completes the proof .
The following result is a little surprising, but the proof is obvious, which
is left to the readers.
.
Proposition 2.3 Let ( X, do ) be a discrete metric space and {xn } C X be
a sequence. Then { x n } is convergent to x E X if and only if there exists
an N 1 such that
Xfi —x Vn > N.
We point out that by changing the metric , the convergence of sequences
may be changed. For example, the sequence { }n > l converges to 0 under
the standard metric of R. However , this sequence is not convergent under
the discrete metric. Further , it is even possible that when the metric is
^
changed , the limit of a convergent sequence might be changed. Here is an
example.
Example 2.4 Let .

1 x = 0,
f ( x) = { X, (0, 1 ) *
0 X = 1.
Let X = [0, 1] and p( x , y ) = \ x — y \ be the standard metric. Define
d { x , y ) = p( f ( x ) , f ( y ) ) = \ f ( x ) - f ( y ) \ Vx , y e X .
Clearly,
d( x , y ) = \ f { x ) - f ( y ) \ = \ f ( y ) - f ( x ) \ = d{ x, y ) 0, Vx , y G [0, 1]
and since / ( •) is one-to-one, we have
d( x , y ) = | f ( x ) - f ( y )\ = 0 x = y.
Finally,
d( x , y ) = | / (x) - f ( y ) \ < |/ (x ) - f ( z ) \ + \ f ( z ) - f ( y ) \
= d ( x , z ) + d( z , y ) W x , y , z e [0 , 1].
Hence, d ( - , •) is a metric on X . Now , for sequence xn = we have
( n ) = -n -> o
p -, o
\ /
and
d (Vn . l ) =
n
"
/
=- n^
- - /U)
0.
Thus the sequence is convergent under p and d, but the limits are different!
Relevant to the above example, we introduce the following definition.
Definition 2.5. Let d , d be two metrics on X .
( i) d is said to be stronger than d if for any sequence {xn} C X , and
some x £ X ,
lim d ( xn , x ) = 0 => lim d ( xn , x ) = 0.
—
n > oo —
n » oo
In this case, d is said to be weaker than d.

( ii ) d and d are said to be equivalent if for any sequence {xn} C X , and
some x £ X ,
lim d ( xn , x ) =0 lim d ( xn , x ) = 0.
—
n > oo —
n > oo
It is clear that if two metrics d and d are strongly equivalent (see (1.9 ) ) ,
then they are equivalent . Hence, in Rm , convergence of sequences under
any pp ( p £ [1, oo] ) are all equivalent. We point out that equivalence of
metrics do not have to be strongly equivalent . ( Why? )
Proposition 2.6 . Let { xn } be a sequence in Mm with xn = (x , x , • • • ,

x ) , and x ( x1, x 2 , • • • , xm ) £ Rm . Then the following are equivalent :
— ^^
( i ) {xn } converges to x with respect to pp , for some p £ [1, oo];
( ii) {xn} converges to x with respect to pp , for every p £ [1, oo];
( iii ) For each 1 k m , the sequence {x£} is convergent to xk .
The proof is left to the readers.
Definition 2.7. Let ( X, d) be a metric space, and {xn}n be a sequence

in X. ^ i
—
( i) Let cr : N > N be a strictly increasing map. Then {x 7 ( n ) }n i is (
^
called a subsequence of {xn }n i . Sometimes, a subsequence is denoted by
^
^-
{ xnk }k 1
( ii) A point x G X is called a limit point of the sequence {xn}n if
there exists a subsequence {£< (n ) }n i such that ^ i
j
^
lim d ( xa /v n\ , x ) = 0.
n — oo
Proposition 2.8 . Let ( X , d ) be a metric space , and { xn } C X be a se-

quence. If {xn} is convergent to x , then any subsequence {xa ( n ) } of { xn }
is convergent to x as well. Conversely, if every subsequence {Xo- ( n )} of
{ xn } has a convergent sub-subsequence with all the limits being x , then
the sequence { xn } is convergent to x .
Proof . Suppose { xn } is convergent to x . Then for any e > 0, there
exists an N 1 such that
^ d ( xn , x ) < e , Vn N.
Now , let { X ( j i ) } be any subsequence of { xn } ( with cr : N
(j
^ —> • N being a
strictly increasing map ) . Noting cr ( n ) n , we have
( £ cr ( n ) ? 3C ) Vn N .
^ *
Thus , { xa ( n ) } is convergent to x .
^
Conversely, we prove by contradiction. Suppose { xn } is not convergent .
Then there exists an e > 0 and two subsequences {xai (n) } and {£ < 2 (n )} j
such that
( n ) , X ^- 2 ( n ) ) n 1.
^
Consequently, the limits of the convergent subsequences of {x Tl ( n ) } and (
{ £ <72 (n ) } will be different , a contradiction.

*
For the case X = R, we have some additional notions and results con- i0
cerning the limits. For convenience, we let c, c, and £°° be the sets of all
convergent , Cauchy, bounded sequences in R, respectively, and let £° be the
set of all general sequences in R . Then from Proposition 2.2, one has
c c c c r a0.
Let us now state the following result whose proof follows from the definition
of the limit easily.
Proposition 2.9 . Let { xn } , { yn } G c. Then
n
lim ( xn ± y n ) =
— oo — „ ± lim
n -
yn ,
oo
£n
*
n > oo
lim ( x n y n ) = ( lim x )( lim yn ) ,
n — oo —
lim xn
n — oo n oo
lim — = — n oo
i f lim y n
± 0.
Vn hm yn
— n oo
— n oo
The following gives a comparison theorem of sequences and their limits.
Proposition 2.10 . Let { xn } , { yn } G c and let
- Vn N0 ,
Xn
^ Vn
for some No
^ 1. Then lim yn .
— xn
lim ( 2.1 )
n > oo n — oo
Proof . Let
x = nlimoo xn , y = lim yn .
—* — n > oc
> 0, there exists an JV

Suppose x > y instead , then for
that
£ = ^ 1, such
This leads to ( noting

I xn - x \ + \ yn - y \ <
xn — yn 0)
0 < 2£ < x - 2/ = x - x n + x n - 2/n

^ n
^ N.
+ 2/n - 2/
|x - xn | + |?/n y \ < e , — N,
a contradiction.
The above result has a very useful corollary whose proof is obvious.
* Corollary 2.11. ( Squeeze Theorem ) Suppose { xn } ,{yn}, { zn } £ f sat- °
isfy
Vn NQ
for some Afo

Xn
^ yn
^
1. Further , suppose
— xn — n-+oc zn =
lim lim x.
n > oo
Then
lim yn = x .
—
n > oo
Next , we consider sequence with infinite limit.
Definition 2.12. Let { xn } G £ ° . If for any M > 0, there exists an N 1

such that
xn > Af ,
*
Vn
^ AT,
then we say that xn approaches to oo as n goes to infinity, denoted by

xn -» 00, or
— xn =
lim oo.
n >• oo
Likewise, if for any M > 0 , there exists an N
Xn < - M, Vn
Âr1, such that
then we say that xn approaches to — oo as n goes to infinity, denoted by
xn -> -oo, or
lim xn = — oo.
—
ra > oo
The following comparison of sequences is obvious. We leave the proof

to the readers.
Proposition 2.13. Let { xn } , { yn } G £ satisfying °
%n Vn Vn N0 ,
for some No . Then
^1 xn — y oo Vn 00
yn ->• -oo => xn — y — oo.

Monotone sequences are of special importance and they will play useful
roles below.
Proposition 2.14. Let { xn } G f
°.
( i ) Suppose there exists an M G R such that
Vn 1.
Then { xn } G c , and
lim xn = sup xn M,
n-> oo
where supn x xn
n l
^
is called the supremum ( or the least upper bound ) of the
^
sequence { xn }n i .
^
( ii ) Suppose there exists an M eR such that
Vn 1.
Then { xn } G c, and
•En
^1
^
lim xn = inf xn M,
—
n > oo
^
n l
where infn i xn is called the infimum ( or the greatest lower bound ) of the
^
sequence { xn }n i .6
^
6 The existence of supremum and infimum relies on the theory of the real number system ,
which will be presented in the Appendix.
F
Proof ( i) First of all, since { xn } is bounded above, by a result from the

,
Appendix,
b = sup xn 6 R
exists. Now , for any e > 0 , there exists an N 1 such that
XN > b — e .
By the monotonicity of {xn}, we have
xn 6, Vn
This implies
b — e < XN
^ N.
\ xn - b| < £ , Vn N.
Hence, we have the convergence of xn to 6.
(ii) The proof is similar and is left to the readers.
We know that c C c C °° , and c £°° . A natural question is what can
we say a little more about sequences in
^ ^
\cl We now investigate that .
Let {xn }n > i £ °°. Then for any k
following is well-defined:
^ ^ 1, { xn } ^ ^°°, and the
n k
Vk = nsupxn .
k
^
It is not hard to show that { yk } k is bounded and non-increasing:
^ i
Vk 1.
^ Vk Vk +1
Thus, by Proposition 2.14 , the following limit exists:
lim yk = kinfl yk = kinfl supxn = klim supxn = lim xn .
—
k too
^ n k
^ ^ yoo
n k n-+ oo— ^
This is called the limit superior ( limsup, for short ) of {xn}n
^-
i
Likewise, we may define

zk = ninfk xn , k 1
^
-
which is a non decreasing bounded sequence. Hence, the following limit
exists, by Proposition 2.14 again:
lim zk = sup zk = sup inf xn = klim inf xn = lim xn .
—
k > oo
^ k
^ l
^ k — ^
ln k
toon k n —> oo
This is called the limit inferior ( liminf , for short ) of {xn}n i -

^
The above shows that for any { xn } £ £°° , lim
xn and lim xn always
— —
exist , thanks to the theory of the real number system , which guarantees
n + oo n oo
the existence of supremum and infimum for bounded sequences (of real
numbers). Further , if { xn }n i is unbounded , lim xn and lim xn can still
be defined if they are allowed to be
^ ± oo.
—
n ± oo n
— oo
The following gives some relations among limsup, liminf , and limit .
Proposition 2.15 . For any sequence { xn } e £°° , the following hold :

lim
—
n too
xn lim
—
n too
xn , lim (
—
n > oo — xn = — lim— xn.
)
n oo
( 2.2)
Further , lim xn exists if and only if

—
n > oo
xn = nlimoo xn .
In this case ,
lim
n -> oo — ( 2.3)
lim xn = nlimoo xn = nlimoo xn .

n — oo
— —
In particular ,
- xn =
lim
n +oo
0 lim
n — oo
\ xn \ = 0. ( 2.4)
Proof . By definition , we have
Zk = ninfk xn sup xn = yk Vfc 1.

^ n k
^
Hence,
lim
n — oo
xn = klimoo Zk
— lim
fc — oo — oo xn.
= nlim r
This proves the first in ( 2.2 ) . The second in ( 2.2 ) can be proved similarly.
Next , suppose lim xn = x . Then for any e > 0, there exists an N 1
such that
n oo —
— £ < Xn < X + £ Vn
Hence 5
X
^ iV .
5 VA; iV.
Consequently,
x — ^ lim
— xn ^ — xn
£ lim
n oo n oo
x + £•
Since s > 0 is arbitrary, we obtain
—> oo xn — = n—tcc xn
lim xn = nlim x lim .
—
n > oo
Conversely, if
xn = nlim x,
— oo xn =
lim
—
n > oo
then for any e > 0, there exists an N 1 such that
| j/ fc - x \ < e ,
^
- x\ < ,
with
\ zk £V /c
^ iV,
x — £ < Zk = iinfk Xf>
^
xn < supk ^ = t/fc < x + £ , n
e
^ ^ k ^ N.
This leads to
Hence 5
xn — x \ < £ 5 Vn
^ N.
lim
n —> oo
xn = x .
We leave the proof of the rest of the conclusions to the readers.
.
Proposition 2.16 Let { xn } , { yn } G £°° .
( i) The following hold :
lim ( xn + yn )lim + nlim Vn
—
n » oo — n too —t oo
lim (xn + yn ) lim + nlim j/n .
n —> oo — n > oo —
» oo
( ii ) Let
Vn 1.
>0
Then
^ Vn
lim xn < nlim yn , lim xn lim yn .
—
n » oo —> oo —
n > oo n-* oo
The proof is left to the readers. Let us now recall the definition of limit
point (see Definition 2.7 (ii) ) .
.
Proposition 2.17 Let { xn } G °°. Then both lim
^x is a limit point — °°{ xn} and, then
lim xn are
limit points of { xn } . Further , if
n
of
>
— xn n i
n > oo
^
< x < nlim .
— oo xn
lim x n
—
n > oo )•
Exercises
1. Find an example showing that a bounded sequence does not have to
be Cauchy.
2. Let { xn }n
N such that ^ i be
Cauchy. Then either for any N
^ 1, there exists an
Vn N,
or there exists an n
^ Xn
Xn
N such that
>
^
xn Vn iV.
^ Xn
^
Is it possible that for any N 1 there always exist m , N such that
^ Vn iV?
Xn
^
Xn X n ,
i
^ 2
3. If { xn } , { yn } G c, then {max ( xn , yn ) } G c.
4. Let Co be the set of all real sequences that converge to 0. Then Co is a
linear space, i.e., for any {xn}, { yn } G Co and any a , /3 G R , { axn -F Pyn } G
c0 . Is the conclusion true for c ( the set of all Cauchy sequences ) ? Why?
5. Let {xn}, { yn } G i°° satisfy
Xn < Un ? n 1.
Then
lim
n — yoo
xn < nlim
— yoo
yn lim xn lim yn .
n —y oo —
n > oo
For each of the above, find examples that the equalities hold .
6. Let { xn } , { yn } 6 £°° . Then
n
lim
— oo
y
+ j/„ ) < nlim
( xn
—> oo
xn + nlim
— yoo
yn ,
lim ( xn + yn ) lim xn + lim yn .
n — y oo n » — oo
yoo n—
Further , for each of the above, find an example that the strict inequalities
hold .
7. Let { xn } G c and { yn } G -£°°. Then
lim yn .
— yoo xn + n — yoo
lim ( xn + 2/n ) = nlim
n — yoo
8. Let { xn }n
^ i , { yn }n^ i , { zn }n^ i £°° such that
Vn 1.
^ Un
Further , let the following hold:
Xn
lim xn = nlim zn = L lim xn = nlimoo zn = L.

—
n » oo » oo — —
n > oo — >
Then
lim = L, lim yn = L.
—
n > 00
—
n > oo
From the above, prove the Squeeze Theorem ( Corollary 2.11) .
9. Let 0 < a < 6, x\ = a > 0, yi = b > 0, and
%n Vn
^n + l — y/ XnVn
* Vn+1 2
2.
Then lim xn and lim yn exist and they are equal.
n > oo— n » oo —
10. Let lim xn = a. Then nlim
#1 H- # 2 “
hxn a.
•• • f~
“
n —> oo —> oo n =
11. Let XQ = 1, xn + i = 1 + j n 0. Show that lim
further , find the limit . ^ n
^ —
n > oo
exists and
12. Suppose { xn } £ £°°\ c. Show that there are two convergent subse-
quences of { xn } having different limits.
13. Let { xn } e c such that for some subsequence { xnk } k
lim xnk = x . Then lim xn = x . ^ i , it holds
—
k > oo n > oo —
14. Let ( X , do ) be a discrete metric space. Let xn G I be a sequence
-
converging to x under do Then there exists an N 1 such that
xn = x n N.
3 Sets in Metric Spaces
In this section , we look at some sets in metric spaces , and their properties.
Let us first introduce the following definition.
Definition 3.1. Let ( X , d ) be a metric space.
( i ) Let xo e X and r > 0. Define
B ( x0 , r ) = B( X ,d ) ( xo , r ) = { x e X \ d { x , x0 ) < r } ,
which is called the open ball centered at XQ with radius r .
(ii) A point xo £ E C X is called an interior point of E if there exists
aS > 0 such that
B ( xo , 6 ) C E . ( 3.1)
The set of all interior points of E is denoted by E° , which is called the
interior of E .
G X is called a boundary point of E if for any S > 0 ,
( iii) A point Xo
B ( xo , 5) fl E 0 B ( xo , S ) fl Ec 0 .
^ ^
The set of all boundary points of E is denoted by d E , which is called the
boundary of E .
( iv ) A point xo G X > 0,
is called an adherent point of E if for any 5
E n B ( x0 , 6 ) 0.
^
The set of all adherent points of E is denoted by E , which is called the
closure of E .
(v ) A point xo G X is called a limit point of E if it is an adherent point
of E\{x 0}.
( vi) A point xo G E is called an isolated point of E if there exists a 6 > 0
such that
(vii) A set E is said to be open if

E = E°
i.e., for any xo £ E , there exists a <5 > 0 such that ( 3.1) holds , which is also
equivalent to
E n dE = 0.
( viii) A set E is said to be closed if
E E — ( 3.2 )
which is equivalent to
d E C E. (3.3)
( ix ) A subset E of F is said to be dense in F if E D F .
(x) A set E is said to be disconnected if there are two non-empty sets
C? i , C?2 Q X such that
—
E Gi u G 25 Gi n (?2 G\ n c?2 = ( 3.4 )
If E is not disconnected we, say that E .
is connected In the case E = X,
we call that A is a disconnected metric space and connected metric space ,
respectively, if E = X is disconnected and connected , respectively.
For the connectedness above, we look at the following simple situation:

—
The set S = ( 1, 0 ) U (0, 1) is disconnected . In fact , by letting G\ = ( 1, 0 ) —
and G 2 = ( 0 , 1) , we see that the last two relations in ( 3.4 ) hold . On the
—
other hand , let us consider S = ( 1, 1) which should be not disconnected .
—
By letting G\ = ( 1, 0] and G 2 = ( 0 , 1) , although S = G\ U G 2 and
Gi n G 2 = 0, but Gi fl G2 = {0} 7^ 0, i.e., the last relation in (3.4 ) fails.
The point that we want to make here is that G\ fl G 2 = 0 is not enough
for disconnectedness.
Example 3.2. ( i) Any finite set A is closed , and all of its points are isolated
points.
( ii) The sets 0 , N , Z, and R are closed . All points in N and Z are
isolated , and all points in R are limit points.
( iii) The set Q is not closed and its closure Q = R. Therefore, Q is
dense in R . All points in Q are limit points of Q. There are no isolated
points in Q.
Proposition 3.3. Let ( X , d ) be a metric space.

(i) Let Si C 52 C X . Then
Si C $2, SI C S2° . ( 3.5 )
( ii) Let Si , S 2 C X be two sets. Then

{ Si n s2 ) c Si n s2, (S1 US2 ) = Si u s2 ( 3.6 )
and
( S i n S2 )° = S? nS£ , (Sius2 n s f u s2°. ( 3.7)
Proof . The proof of ( i) is clear. We now prove ( ii) . First of all,
Si n S2 c Si , S2 0 Si u S2 . ( 3.8)
Thus, by ( i) , we have
(Si n S 2 ) c Si n S2 S1 US2 C (S1 US2 ) .
Now , we prove the equality in (3.6). For any x G (Si U S 2 ) , there exists
a sequence { xn } C Si U S2 converging to x . Clearly, there must be a
subsequence of { xn } either in Si or in S2 . ( Why? ) Hence, x is either in Si
or in S 2 . Therefore,
(Si U S2 ) = Si u s2 .
Next , from ( 3.8) , by ( i) again , we have

( Si n S2 ) ° St n S2° S US C ( SiUS2 ) ° .
C
^ ^
We now prove the equality in ( 3.7). If there is an x E S£ fl S2 which is an
open set , then there exists a S > 0 such that
£ (*, £ ) c s? ns2° c Si ns2 .
This leads to x G ( Si fl S2 ) ° . Hence,
(Sins2 ) ° = Si° ns2° .
This completes the proof of ( ii).
Note that if we take
Si = ( 0, 1 ) , S2 = ( 1 , 2 )
then
Also, if we take
(Si nS2 ) = 0
^ { } = Si n S
1 2.
Si = ( 0 , 1 ] , ^ = [1, 2)
2
then
SI u S°2 = (0 , 1 ) U ( 1 , 2) = ( S 1 U S2 ) ° .
^( ( 0 , 2)
Hence, the two inclusions in ( 3.6 ) and 3.7) may be strict , in general.
Proposition 3.4. Let ( X , d ) be a metric space and E Cl. Then
E = E U <9£ = U dE . ( 3.9)
Proof . First of all, by the definition of adherent point , we have
E° U dE C E U dE C T.
To show
E C E° U dE ( 3.10 )
we take any xo E -S, i.e., Xo is an adherent point of E . If Xo G h° , we are
done. Suppose xo £ E° . Then by the definition of an adherent point , Xo
must be a boundary point of E . This proves ( 3.10 ) . Hence, ( 3.9 ) holds.
.
Proposition 3.5 Let ( X, d ) be a metric space.
( i) If { E\ }\e\ is a family of open sets in X , then UAGA is open. If
{FA}AG A is a family of closed sets, then DAG A is closed .
( ii) If , En are open sets in X , then E\ n E^ fl D En is open.

•• •••
IfFi , , Fn are closed , then F\ U U Fn is closed .

••• •••
( iii) For E C X , E° is the largest open set contained in E .

( iv ) For E C X , E is the smallest closed set containing E .
( v ) For E C X , E is open if and only if Ec is closed .
(vi) For E C X , E is closed if and only if for any { x n } C E , xn —> x
implies x G E .
The proof is straightforward and is left to the readers.

Note that the intersection of infinitely many open sets might not be
open , and the union of infinitely many closed sets might not be closed .
Here are counterexamples:
°° 1 1
71 =1
OO
1 '
( 0, 1 ) .
nJ =
-
=1 71
It is clear that the open ball B ( xo , r ) is an open set , and its closure
B ( x0 , r ) = { x X | d( x , x o ) r}
is closed , called the closed ball centered at XQ with radius r .
Example 3.6 . ( i ) Consider R 2 with Euclidean metric p2 - Let X = { { x , 0) |

x e R}. Then { X , p2 ) is a metric space. Let
E = { { x , 0 ) | -1 < x < 1} C X .
Then E is open in X , but it is not open in R 2 . ( Why? )
( ii) Let R be the metric space with the standard metric p. Let X =
—
( 1, 1). Then ( X , p ) is a metric space. Note that X is not closed in ( R , p ) .
But , X is always closed in ( X , p ) . ( Why? )

Definition 3.7. Let ( X , d ) be a metric space and Y CX , Let E C Y . We
say that E is relatively open ( resp. relatively closed) in Y if it is open ( resp.
closed ) in the metric space ( Y, d ) .
The following gives a representation of relative open sets and relative
closed sets.
Proposition 3.8. Let ( X , d ) be a metric space and E C Y Cl

(i) Set E is relatively open (resp. relatively closed ) in Y if and only if
there exists an open (resp. closed ) set V C X such that E V C\ Y . —
( ii) For any E C Y , if E and E stand for the closures of E in X and in
Y , respectively, then
E = E fl Y . (3.11)
Proof , ( i) Suppose E is relatively open in Y . Then for any x G E , there
exists a 5( x ) > 0 such that
B y { x , 5( x ) ) C E.
Let
v = U B*(*, <5(X)).
XEE
This is an open set in X . We now prove that E = V fl Y . First of all , for
any x G E ,
x G B Y ( X ,6( X ) ) C B X { X , S ( X ) ) C V ,
together with E C Y , we obtain
E c v n Y.
On the other hand , for any y G V D Y , since y E V , which implies that for
some x e E , y e B x ( x , 6 ( x ) ) . Since y e Y also, one actually has
yG B x { x , 5 ( x ) ) f ) Y = B Y { X , 5( X ) ) C E .
Hence , we obtain E — V fl Y .
Conversely, suppose E = V fl Y for some open set V m X . For any
x G E C V , since V is open in X , there exists a 5 > 0 such that
B x ( x , 5 ) C V.
Thus
BY ( X , 5 ) = B x ( x , 5 ) n Y CV n Y = E.
Consequently, x is an interior point of E in (Y , d ). Thus, E is open in (Y , d ) .
Now , let E be relatively closed in Y . Then Y \ E is relatively open in

Y . Thus, by what we have proved , there exists an open set V in X such
that
Y\ E =V n Y.
Hence, V c = X \ V is closed in X , and
E = Y \ (V n Y ) = Y n (V n Y ) c = Y n ( Vc u Y c ) = y n vc .
This proves the relatively closed case.
( ii) For any set E C Y , E is the smallest closed set in X containing E .
Now E D Y is a closed set in Y containing E . Clearly, it is the smallest one.
Hence, ( 3.11) holds.
The following result gives another interesting characterization of the
disconnectedness / connectedness , which will be useful later.
Proposition 3.9 . Let ( X, d) be a metric space and E C X . Then E is
disconnected if and only if there are two disjoint open sets V\, V2 C X such
that
E r \1 / 0, E r V :2 7^ 0, E C V1 U V2 . ( 3.12 )
Note that in the above, we do not assume if E is open or closed.
Proof . Sufficiency. Suppose there are two disjoint open sets Vi , V2 C X
such that ( 3.12 ) holds. Let
Gi = E n V i , (?2
—ED
Consequently,
E = E fl ( Vi U V2 ) = ( E n Vi ) U ( E D V2 ) = Gl u G2.
Further , since V\ fl V2 = 0,
Gl n G 2 = ( E n Vi ) n ( E n v2 ) = E n ( Vi n v2 ) = 0.
We claim that Gi fl G 2 = 0. In fact , for any x E G 2 = E fl V2 C V2 , since
V2 is open , there exists a 6 > 0 such that
B ( x , 6 ) C V2 .
On the other hand , if this x G Gi as well , then there exists a sequence
—
xn Gi = E n Vi such that xn > x . Hence, for n large and for the above
•
S > 0,
x n e G1 n B ( x , S ) c E n V\ n v2 c V\ n v2 = 0
2 8 Mathematical Analysis — A Concise Introduction
a contradiction. This shows that G\ D G 2 = 0. Likewise, Gi fl G2 = 0.

Therefore, E is disconnected .
Necessity. Suppose E is disconnected . Then there are non-empty sets
Gi , G 2 C X such that
.£/ — Gi u G , 2 Gi n G 2 — 0, Gi n G 2 = 0.
G1, xi G 2 . Thus,
For any x\ G
d ( xi , G 2 )
^
= d (xi , G2 ) = 2inf
GG 2
X
d ( xi , x 2 ) > 0 , Vxi G Gi .
Likewise
G2 .
<
^(^ , GI ) = d( x , Gi ) =
2 2
xi
inf d ( xi , X 2 ) > 0 ,
Gi
VX 2 G
Now , we define
xi
^
6 2
Both Fi and F2 are open and

E n Fl D Gi , £ n F2 D G 2 , Fl U F2 2 Gi U G 2 = £.
We now claim that Fi and F2 are disjoint . In fact , if there exists an x G
Fi DF2 , then there are x\ G Gi and X 2 G G 2 such that
^
x G J5 xi , Id (a: , G )) fl B ( ,
i
Without loss of generality, we assume d ( xi , G 2 )

2 X2
^ d ( x2 , Gi ) ).
d ( x 2 , Gi ). Then
d ( x 2 , Gi ) d ( xi , x 2 ) d ( xi , x ) + d ( x 2 , x )
1 1 2
- d ( xi , G 2 ) + -d ( x 2 , Gi ) - d ( x 2 , Gi ) ,
which is a contradiction. Therefore, Fi and F2 are disjoint .

Proposition 3.10. Let ( X , d ) be a metric space and E C X be a non-
empty set . Then E is disconnected as a set in X if and only if it is discon-
nected as a metric space , with the induced metric d .
Proof . As before, for any G C E , we let G and G be the closures of G

in X and in E , respectively.
Necessity. Suppose E is disconnected as a set in X . Then by definition ,
there are non-empty sets Gi and G2 such that
E = Gi U G2 J GI PI G 2 — 0, Gi n G 2 —0 *
The above implies that

= G\ n G% — G\ n G 2 n E = ( Gi n F ) n G 2 = Gi n G2
0 ?
XN
where G\ = G\ n E is the closure of G\ in E . Likewise,

0 = Gi n G2 — Gi n ( G 2 n E ) = Gi n G 2 ,
XS
with G 2 = G 2 fl E being the closure of G2 in E . Hence, by definition , E a

is disconnected metric space.
Sufficiency. Suppose E is a disconnected metric space. Then there are
non-empty sets Gi , G 2 C E such that
E = Gi U G 2 Gi nG 2 = 0 , GI n G2 = 0.
Then, by Proposition 3.8 ( ii) , we have
G\ — Gi n F, G2 —G 2 n E.
Consequently, noting Gi U G2 = E ,
Gi n G2 — Gi n E n G 2 — Gi n G2 — 0
and
—
Gi n G2 Gi n E n G 2 = G1 n G2 — 0 «
Hence, by definition , E is disconnected in X.

The following gives some interesting characterizations of disconnected-
ness of metric spaces.
.
Proposition 3.11 Let ( X , d ) be a metric space. Then the following are
equivalent :
(i) X is disconnected .
(ii ) There are two disjoint open sets VT , V 2 £ X = V\ U V2.
such that X
(iii) There are two disjoint closed sets Fi , F2 Cl such that X = F1 UF2 .
(iv ) There is a proper subset of X which is both open and closed in X .
Proof , ( ii) => (iii) . Since X = V\ U V2 with both V\ and V 2 open and
disjoint . Then
F1 = X \V1 = V2 , F2 = X \V 2 = Vi ,
which are both closed and disjoint , and
Fi U F2 = V2 U V\ = X .
( iii) => ( iv ) . When X = F1 UF2 with both F\ and F2 closed and disjoint.
Then Ff = F2 is open and also closed. This means that X has a proper
subset which is both closed and open.
( iv ) => (ii) . If G C X is a proper subset of X , which is both open and
closed , then G and Gc are two disjoint open sets and
X = G U Gc.
( i ) => ( iii) . By definition , there are disjoint non-empty subsets G\ , G 2
such
x — G\ u c?2 ) G1 n G 2 — 0, G\ n G2 —0 *
Then G\ n G2 =0. We claim that G\ = G\. In fact , if there exists an

x G Gi \ Gi G2 , which contradicts G\ fl G2 = 0. Hence, G\ is closed.
C
Similarly, G 2 is also closed . Thus, (iii) holds.
( iii) => (i) . If X = G\ U G 2 with G\ and G 2 both being closed and
disjoint , then
G\ n G 2
Hence, X is disconnected .
—G
1 n G2 — 0 G\ n G 2 —G nG \ 2 —0 *
Exercises
1. Let ( X, d ) be a metric space and E C X be a finite set. Then E is
closed.
2. Let ( X, do ) be a discrete metric space. Then any subset E of X is
both open and closed.
3. Let ( X, d ) be a metric space and E C X. Then E° is the largest
open set contained in E .
4. Let ( X, d ) be a metric space and E C X. Then F is the smallest
closed set containing E .
5. Let ( X, d ) be a metric space and E C X. Then F is closed if and
only if for any convergent sequence { xn }n i C F, the limit x of { xn }n i
is in F. ^ ^
6. Let ( Xfc , dfc ) be metric spaces k = 1, 2, • • • , m. Let X = Xi x • • • x Xm
with the product metric
m
---
d(( x , xm ) , ( y i , , t/m ) ) =5
^
fe = i
djfe ( xfc , 2/ fe ).
Let Ek C Xk , 1 k m. If each Ek is open ( resp. closed ) , then E =

E\ x • • • x Em is open ( resp. closed ) in ( X , d ) .
7. Show that for any open set G C R , there exists a finite or an infinite
sequence of disjoint open intervals (afc , £> /e ) , k 1 such that
G= UKA ).
Extend the above result to the case of Rm .
8. Let ( X, d ) be a metric space. Suppose G\ yG 2 £ -X . Then
"
X
inf d ( xi , X2 ) = _inf _ d ( xi , X 2 ) .
xi GGi , a;2 eG? 2 #I £ GI , X 2 CT 2
4 Properties of Metric Spaces
We now consider more properties of metric spaces.

.
Definition 4.1 ( i ) A set X is said to be countable if there exists a bijection
—
f : N » X , i.e., / ( N ) = X and f ( £ ) = f ( n ) if and only in £ = n.
( ii) A set X is said to be at most countable if it is either finite or
countable.
( iii) A set X is said to be uncountable if it is not at most countable.
Practically, X is countable if and only if X can be written as follows:
X = {xi , x2 , 3, - - - }.
aj ( 4.1)
Proposition 4.2 ( i) . If X is at most countable, then so is any subset of
X.
( ii) If X and Y are at most countable , so are X UY and X x Y .
( iii ) Let X
oo ^ , k = 1, 2, ••• be a sequence of at most countable sets. Then
so is (J Xfc .
k=l
Proof , ( i) If X is finite, the conclusion is trivial. Let X be of form

( 4.1) . Let Y be a subset of X . In the case that Y is finite, it is of course at
most countable. In the case that Y is infinite, we can find a subsequence
{ x* ( n ) }n> l such that
Y — {^ ( n ) }n l *
(j
^
Thus, Y is countable.
(ii) We need only prove the case that
X = { x i , X 2 , x% , • Y = (yi , 3/2 , 2/3, - •
Clearly,
XUY = { x 1 , y 1 , x 2 , y2 , x 3 , y 3 , - - - }.
Hence, X U Y is countable. Also, one has
x X Y = { ( x i , y i ) , ( x i , y2 ) , ( x2 , y i ) , ( x i , y3 ) , ( x‘2 , y2 ) , ( x1 , y3 ) , - - - }. f
This implies that 1x 7 is countable.

(iii) Let
Xi = i = 1, 2 , 3, • • • .
Then
oo
[ j X i = { x\ , x\, x\ , x\, xl , x\ , - - - }.
i =1
oo
Therefore, ( J Xi is countable.
^
2 =1
For any x G R, let [x] be the largest integer that is less than or equal to
x . We define the sequence ao , ai , • • • as follows:
cto = [ x\ G Z, x - a0 e [0, 1) ;
oi = [10( x - a0 ) ] -
{0, l , - - , 9} =» x “
ao - e [° 10
»
a2 = 102 (x ai\l
e {0, 1, - - - , 9}
x — do — 1—0
&1 U2
102
G
^—
° i02 )
> »
n—1 n 1
an = [ ( — 53 ) ] °fc 1 °’ 10" ) ’
l 0" x
fc = 0
Hence, for any x G R , one has

^ 6 {0 , l , - " , 9} =>
fc = 0
10s
G
*= Ea10* fc do G Z, dk G {0 , 1, 2, • • • , 8, 9}, k
^ 1. ( 4.2 )
Clearly, for any given real number , the above representation is unique,
making use of the convention that any rational number is represented by
the corresponding repeating decimal number . Hence, we refer to ( 4.2 ) as
the canonical decimal representation of real numbers.
Proposition 4.3. For any a < b , the set [a , b] is uncountable.
Proof . Without loss of generality, we consider [0, 1]. Suppose [0, 1] is

countable. Thus, we let
[0 , 1] = {ai , a 2 , - - - }.
We use the canonical decimal representation for each an :
a\ = O.611&12 &13 * * *
5
a2 = 0.621622623 • • 5
as = O.631632633 • , • •
where bij G {0 , 1, 2 , • • • , 8, 9}. Now , we let

C = O.C1C2 C3 • • •
with
d e {0, 1, 2, - . . , 8, 9} \ {6 n }
c2 G {0, 1, 2 , • • • , 8, 9} \ {622}
C3 G {0 , 1 , 2 , • • • , 8, 9} \ {633}
Then
c (£ {a1, o2 , o3 , - - - } = [0, 1]
a contradiction. Hence, [0 , 1] is uncountable.
Definition 4.4. (i) A metric space ( X , d ) is said to be complete if every

Cauchy sequence in ( X , d ) is convergent in ( A, d ) . A normed linear space
X is called a Banach space if it is complete under the metric induced by
the norm. An inner product space is called a Hilbert space if it is complete
under the metric induced by the inner product .
(ii) A metric space ( X , d ) is said to be compact if every sequence in
( X , d ) has at least one convergent subsequence.7 A subset Y C X is said to
7Such a compactness definition only works for metric spaces , and it does not work for
general topological spaces. This notion is also referred to as the sequential compactness.
be compact if subspace ( Y, d ) is compact . Or , equivalently, for any sequence
in Y , there exists a convergent subsequence whose limit is in Y.
( iii ) A set Y in a metric space ( X, d ) is bounded if there exists a n x G l
and an r > 0 such that
Y C B( x , r ) .
A set Y is totally bounded if for any e > 0, there exists an integer N > 0
such that
N
YC 1J B( xn , e )
71 =1
for some , ,
X\ X 2 • • • , xn E X.
Example 4.5 . ( i) Let ( X, do ) be a discrete metric space. Then it is com-
plete.
( ii) Any finite dimensional normed linear space is complete.
(iii) Let
oo
{
P = (** ) *> 11
fc = 1
< OO j, l < p < o o .
We are able to show that £ p is complete. The detailed proof is left to the
readers.
Proposition 4.6. Let ( X, d ) be a metric space , and ( Y, d ) be a subspace

of ( X , d ) .
( i) If ( Y, d ) is complete , then Y is closed in X .
( ii) If ( X , d ) is complete, and Y is closed in X , then ( Y, d ) is also com-
plete.
The proof is straightforward .

The following result is called the completeness of real numbers.
Theorem 4.7. A sequence { xn } E £° is convergent if and only if it is

Cauchy, i.e., c c. —
Proof . We already know that if { xn } is convergent , then it is Cauchy
(see Proposition 2.2, ( ii )).
Conversely, let { xn }n i be Cauchy. Then it is bounded and for any

e > 0 , there exists an N 1 such that ^
^
\ x k - X j \ < e , VkJ N .
Hence
Consequently,
XN ~ £ xn XN +s Wn
^ N.
XN — £ lim
—
n > oo
xn lim
n -* oo
xn XN + £•
Thus
0 lim xn - lim xn 2e.
—
n » oo n —
foo
Since £ > 0 is arbitrary, we obtain that

lim xn = nlim
—¥ oc xn
,
71 —» OO
which leads to the convergence of {xn }, by Proposition 2.15.
We should point out that to be rigorous, one should go through the
theory of the real number system (see Appendix ) . As a matter of fact ,
Proposition 2.15 relies on the existence of the supremum and infimum ,
which is a result of the real number system. Actually, there is a one-to-one
correspondence between the set of real numbers and the set of the so-called
equivalent Cauchy sequences. See the Appendix for details.
For compactness, we have the following simple result .
Proposition 4.8 . Let ( X , d ) be a metric space.
( i) Every finite subset of X , including the empty set , is compact .
( ii ) If Yi , > 2 > * * * j Yn are compact sets of X , then Y\ U I 2 U ••• U Yn is
also compact .
( iii) If Yi , I 2 , • • • ,Yn are compact sets of X , then Y\ x Y2 x ••• x Yn is
compact in Xn = X x X x • • • x X .
—
Proof . The proofs of (i) ( ii ) are obvious. Let us prove ( iii) . We only
prove the case n = 2. For any sequence yn = (y , y%) E Y\ x Y2 , by
the compactness of Yi , we have a convergent subsequence {2/ 1 ( n ) }n 1 of
{Vn }n i and some y 1 G Y\ such that
^ ^ ^
^
n —y 00 .
Next , by the compactness of > 2 ? there exists a subsequence { y<2r 2 ( o-i ( n )) }n
of { yh ( n ) }n>1 and a y 2 e 2 such that ^ l
*
d ( yl 2 ( r i ( n ) ) ^ y 2 ) 0
(
77, —^ oo .
Consequently, ( yl , y 2 ) G Y\ x Y2 and
. Sa
d ( ( (ffi (n )) « ( « n (n)) ) » ( tfStf 2 ) )
wia
This proves the compactness of Y\ x F2 .
^0 n -» oo.
The following result gives deeper property of compact sets.

Proposition 4.9 . Let ( X, d) be a compact metric space. Then X is com-
plete and bounded ; and any Y C X is compact if and only if it is closed .
Proof . Suppose ( X , d ) is compact . If ( X, d ) is not bounded , then pick

any x\ G X , and we can find
x 2 G X \ B( xi , 1) .
Next , we can find
xs e X \ B ( x i , 1) U B( X 2 , 1) ,
since S(xi , 1) U 5(x 2 , 1) is a bounded set . We continue this procedure and
obtain a sequence xn with
—
n 1
xn & X \ (J B ( xk , 1) .
k =1
Then we see that
d ( xi , xk ) 1 Vi j.
,
Hence this sequence does not have a convergent ^
subsequence, a contradic-
tion , proving the boundedness of X .
Next , let { xn }n i be a Cauchy sequence in X . By the compactness
of X , there exists a^subsequence { xnk } k l converging to some limit point
^
x G X . Then we can show that the whole sequence is convergent to the
same limit x . Hence, X is complete.
The conclusion for Y is clear.
Note that the converse of the above is not true, i.e., bounded and closed
set is not necessarily compact. For example, if ( X, do ) is a discrete metric
space and Y C X is an infinite set. Then Y is bounded and closed , but not
compact . However , we have the following positive result.
Theorem 4.10 . ( Bolzano- Weierstrass ) Let {#n}n > i be a bounded se-
quence in Rm with the £ p -metric pp . Then it has a convergent subsequence.
Proof . We only prove the case m = 1. Since { xn }n i is bounded ,

L = lim xn exists. Then by Proposition 2.17, this L is a limit point of ^
n—¥ oo
{^n}n > i , i.e., for some subsequence xUk > L. — •
Corollary 4.11 . ( Heine—Borel ) Let Rm be the metric space with the

£ p -metric pp for some 1 p oo. Let E C Rm . Then E is compact if and
only if E is bounded and closed .
Proof . => It follows from Proposition 4.9.
<= It follows from Bolzano- Weierstrass Theorem ( Theorem 4.10 ) .

Theorem 4.12. ( Finite Open Cover ) Let ( X, d ) be a metric space and
Y C X . Then Y is compact if and only if it has the finite open cover
property: For any family of open sets {V\ | A E A } in X such that
IK
AGA
there exists a finite subset AQ C A such that
Y C {J V X .
AG Ao
Proof . Necessity. Suppose Y is compact and {V\ | A E A} is an open

cover of Y . Then for each y E F, there exists a A E A such that y E V\.
Since V\ is open , there exists a S > 0 such that B ( y , 5 ) C V . Now, we let
5( y ) — sup{<5 > 0 | B( y , S ) C V , for some A E A} > 0.
\
^
Define
S0 = inf{<5( j/ ) |y e Y }
We have two cases.
^ 0.
Case 1. 5Q = 0. Then for each n 1, we can find a yn eY such that
S( y n ) < -
Tl
Since Y is compact , there exists a subsequence ynk converging to some

y eY . Then for some Ao E A , we have a 5 E ( 0, S ( y ) ) such that
V e B ( y , S ) c vXo .
Since { ynk } converges to y , there exists a K > 0 such that
d{ ynk , y ) <
^ Vfc
^ K.
Consequently, for any z e B ( yn k , § ) , we have
8 8
d{ z, y ) d { z , ynk ) + d { ynk , y ) < - + - = 8.
This means that
B ( yn k i
l ) C B( y , 5 ) C V X o .
Hence
8 1
^ b\ y n k x) < -» 0
0
2 ^ rik
a contradiction , which means that this case will not happen .
Case 2. <5Q > 0. Then we must have
So
>y <%) VyG Y
This implies that for each y G Y , there exists a A ( j/ ) G A such that
B ( y , 50j ) c
<
VA ( w) .
Now , we construct a sequence. Pick any 2/1 G T. If
/
we are done. Suppose the above is not true. Then we can find
£
J/2 e Y \ VX ( y i ) c y \ B ( y i , y0 ).
Clearly,
<50
d { y2 , y i ) y-
if
Y C VA ( J/ I ) U VX ( y 2 )
"
-
we are done. Otherwise, we can find
2/3 G Y \ (yA (3/ ) U yA ( w ) ) C Y \ (s(|

l yi , ) U B (|
y2 , ) ) .
Thus
d { y3 , y i ) , d ( y3 , y2 ) *.
y
0
If the above procedure stops at a finite step , we are done. Otherwise, a

sequence yn can be constructed so that
Vn £
k =1 k =1
Hence
So
d( yn , yk ) >y VI k < n.
\
This implies that Y contains a sequence { y n }n
convergent subsequence, a contradiction. ^ i which does not have a
Sufficiency. By contradiction. Suppose Y is not compact. Then there
exists a sequence { y n } n i Q Y , which does not have convergent subse-
^
quences. In particular , { y n }n i is an infinite set . We claim that for any
— ^
y e y , there exists an e s( y ) > 0 such that
B ( y , e{ y ) ) n { y n n \ 1}
is a finite set. In fact , otherwise, by shrinking e > 0 , we will obtain a
subsequence of { y n }n i converging to y . Now , we have that
^
Y C (J B( y , e( y ) ) .
yer
Then by assumption , there is a finite open cover , i.e. , there are yi , • • • , y m
Y such that
m
{ V n }n > i £ Y C (J B( y k , e( y k ) ).
k =1
This leads to a contradiction since each B ( y k , e ( y k ) ) contains only finitely

many points of { y n } n> l -
Theorem 4.13. ( Finite Intersection) L e t ( X , d ) b e a metric space and
Y C X . Then Y is compact if and only if it has the finite intersection
property: If { F\ : A E A} is a family of closed sets in X such that for any
finite set AQ C A ,
yn ( n f t) * * .
AG Ao
> ( 4.3)
then
Y n( riftW ( 4.4 )
AGA
Proof . Necessity. Suppose Y is compact and { F\ | A G A} has the

property that for any finite set AQ C A , ( 4.3) holds, but
Then
Y n (!«)
AGA
- «
Y C ( n f> )
AGA
= u n-
AGA
Thus, { F£ | A G A} is an open cover of Y . Since Y is compact , there exists
a finite open cover , i.e. , there exists a finite set AQ C A such that
u ** = ( n
AGAO AGAO
f
> ) f
which implies that

Y n ( f| FX ) = 0 .
AGAQ A"
This is a contradiction.
Sufficiency. Suppose {VA | A G A} is an open cover of Y . Then we claim
that it must have a finite open cover of Y . If not , i.e., for any finite set
AQ C A, one has
0
^ YN, ( U vxy =, Yn ( n »?)
AGAQ
Since each Vf is closed by assumption we must have
AGAO
0
AGA AGA
which means that { V\ \ A G A} is not an open cover of Y a contradiction. ,
Theorem 4.14. ( Nested Compact Sets ) Let ( X , d ) be a metric space

and Kn be a sequence of nested non-empty compact sets:
Ki D K2 2 Ks D • • • .
Then
oo
n* ^
n =1
n 0. ( 4.5)
Moreover , if
diam ( ATn ) = sup d ( x , y ) —> 0,
x,yEKn
n —y oo ( 4.6)
then there exists a unique xGl such that
oo
n
n =1
k» = {*} ( 4.7)
Proof . Define
Vn = Kcn , n 1.
Then { Vn n 1} is a sequence of open sets in X . Now , if ( 4.5) is not
true, then
\J V n = \ j K c n = ( f ) KnY = X D K 1.
^ n l
^there exists^1an N
n l 71
By the compactness of > 1 such that

N N N c
K1 c \J v n = \ j K = [ f ] K n ) = K Nt i
n= l n= 1 n= l
which implies
0 —K '
jv n Ki C n Kft = 0
a contradiction. Hence, ( 4.5) holds.
Now , suppose ( 4.6) holds, then for any x, x E f ] 71 1 ifn , we have
^
d ( x, x ) diam ( Kn ) > 0, — n - oo.
>
Hence, ( 4.7) holds.
Next result is relevant to the total boundedness.
.
Theorem 4.15 L e t ( X , d ) b e a metric space and Y C X . Then Y is
compact if and only if Y is complete and totally bounded .
Proof . Necessity. Suppose Y is compact . Then Y has to be complete.
Next , for any e > 0, we have
Y C |J B{ x,e ).
xeY
Then by finite open cover theorem , we see that Y is totally bounded.
Sufficiency. Let { yn }n i Q Y be a sequence. For k = 1, Y is covered by
^
finitely many open balls of radius 1. Among them , there is a ball, denoted
by B ( x\ , 1) , containing a subsequence of { yn }n i We denote it by
£ ( xi , l ) n y,
^-
1. Vn
Next , since B { xi
Vn , I
^
, 1) fl Y is compact , it is covered by finitely many balls of
radius Then there exists an X2 E B ( x i , 1) and a subsequence { yn ,2 }n
of {2/ n , i }n i such that î
^
2/71 , 2
^
e H ( x2 , ) n y, n 1.
Continue this procedure, we can find { xk }k Q X and subsequences
{Vn ,k }n>1 such that î
1 1
xk G B ( x k - i ? fc-!) , yn , k G B { x k , ) n Y. n 1.
2 2 ^
Now , we let zk = V k ,k - Then { z k } k > l is a subsequence of { y n } n
that ^ i such
1
6 S (a:fe , ) D Y. k 1. ( 4.8)
Note that ^
j
'
-i i- i
1 1
< 2i _1 '
d ( x i , Xi+ j )
^ ^= ^
k 0 ^^
k =0
2^ + fe
This means that { x k }k

^ i is a Cauchy sequence. By ( 4.8) we see that
lim d ( x k , zk ) = 0.
—
k KXD
Hence, { zk }k i is also Cauchy. Then by the completeness of T , one has

^
the convergence of { zk }k i This proves the compactness of Y .
^-
The following example shows that in an infinite dimensional space,
bounded closed set is not necessarily compact , in general.
Example 4.16. Recall £ p ( p E [l , oo ) ) . Look at the closed unit ball cen-

tered at 0: r
oo
B ( 0 , 1) = ( xfc ) j G P |
k= l
\ xk \ P l }-
It is bounded and closed . We claim that it is not compact . In fact , let us
look at the following sequence:
Xn — {^ nk ) Vn 1
where
1 k = n,
5n k - n.
m , one has
0 k
^
Then for any n
^ —2 P .
Hence, the sequence { xn }n subsequence. This
proves our claim. ^ i does not have a convergent
Definition 4.17. Let ( X, d ) be a metric space. Then it is said to be sepa-

rable if there exists a countable set Xo such that
X0 = X.
In other words, X admits a countable dense subset.
Proposition 4.18. Let ( X, d ) be a compact metric space , Then X is
separable.
Proof . Since X is compact , it is totally bounded . Consequently, for any
n 1, there exists a finite set { x£ | 1 k Kn } such that
k= l
Now , let
Xo = {*£ 1 k Kn , n 1}
Then XQ is countable and it is dense in X . Hence, X is separable.
It is clear that if ( X, d ) is separable, then any subset Y C X is also
separable.
Example 4.19 . ( i ) If ( Xfc , d& ) ( 1 k n ) is separable, then X\ x X i< x

• • • x Xn is separable under the corresponding product metric.
( ii) Any subset Y of Mn is separable under the usual product metric of

^ -metric (1 p oo ) .
( iii) Let ( X , do ) be a metric space with the discrete metric, and X be
an uncountable set . Then ( X , d0 ) is not separable.
Exercises
1. Show that the set Q of all rational numbers is countable and the set
R \ Q of all irrational numbers is uncountable.
2. Prove the Bolzano-Weierstrass Theorem ( Any bounded sequence
in Rm contains a convergent subsequence) by the Nested Compact Sets
Theorem.
3. Prove the Heine-Borel Theorem by the Nested Compact Sets Theo-
rem.
4. Let 5 = l n n e N} U {0}. Then S is closed .
5. Show that any discrete metric space ( X, do ) must be complete.
6. Let ( X , d ) be a complete metric space and An C X satisfy
A-n+i C. Ani Vn 1
diam ( An ) = sup d ( x , y )
x,yeAn
—* 0, —
n > oo.
Let an G An . Then the sequence {an } converges. Find the difference

between this result and the Nested Compact Sets Theorem.
7. Show that Q fi [0 , 1] is totally bounded.
8. Show that £ p is separable for p G [1, oo ) and £°° is not separable.
9. Let ( X, do ) be a discrete metric space. Let E C X be a non-empty
subset . Then E is compact if and only if E is a finite set.
10. Let ( X, d ) be a metric space.
(i) Suppose Ki , •• , Xn C X are compact sets. Then is compact .

2 =1
( ii ) Suppose {X , A G A} is a family of compact subsets in X . Then
^
P
AGA
| K\ is compact.
11. Let K\ and K 2 be two disjoint compact sets in a metric space ( X, d ) .
Then
a=
XX - d(Xl X2 ) > 0.
K I ,X 2 K 2
>
Moreover there are x\ G K\ and x i G K 2 such that

<
a = d ( x i , x2 ) .
What happens if we only assume Xi and X2 to be closed ?
12. Let {xn} G £°° . Then the set C of all limit points of { xn } is non-
empty. Moreover , for any sequence { yn } C £ , if { yn } is convergent to y ,
then y G C .
Chapter 2
Functions on Metric Spaces
This chapter is devoted to the study of functions on metric spaces.
1 Continuity
We first give the following definition.

Definition 1.1 . Let ( X , dx ) and ( Y, dy ) be two metric spaces.
.
( i) Let E C X and / : E —> Y Let x0 be a limit point of E . We say
that f ( x ) is convergent to L e Y as x —» #o if for any e > 0, there exists a
•
5 > 0 such that

d Y { f { x ) , L ) < s, Vx G E , 0 < d x ( x , xo ) < 6.
We denote the above by
lim / (x) = EBxlimtxo f ( x ) = L .
X ± XQ —
( ii ) Let / : X —> Y . We say that / ( •) is continuous at XQ G X if for
every e > 0 , there exists a 5 > 0 such that
,
d y { f { x ) f { x0 ) ) <£ V d x ( x , x0 ) < S.
We say that / ( •) is continuous on X if it is continuous at every x G X.
The most important example of the above is the case ( X, d x ) = ( Mm , £>2 )
—
and ( y, dy ) = (R , p) . Any / : Rm >• R , with m > 1, is called a multi-
variable function. The following gives some equivalent conditions for a
function being continuous at a given point X Q .
Theorem 1.2 . Let ( X, dx ) and ( Y , d y ) b e metric spaces. Let f : X -» Y
and X Q G X . Then the following are equivalent :
45
(i) / ( •) is continuous at xo -
(ii) Whenever {xn }n i C X is convergent to xo under dx , the sequence
^
{/ £n )}n i is convergent to f ( x0 ) under d y .
(
^
(iii) For every open set V C Y containing f ( xo ) , there exists an open
set V C X containing XQ such that
f ( U ) C V.
The proof of the above result is straightforward. Such a result leads to

the following which is about the continuity of a function f : X -> Y o n X .
.
Theorem 1.3 L e t ( X , d x ) a n d ( Y , d y ) be metric spaces. Let f : X -> Y .
Then the following are equivalent :
( i) / ( •) is continuous on X .
(ii) Whenever { xn } C X is convergent to some xo under dx , the se-
quence { f { xn ) } is convergent to f ( xo ) under dy -
( iii) For every open set VC Y , f ~ x ( V ) is open in X , where
f -\V )
= {x G X | f ( x ) V }.
(iv ) For every closed set F in Y , f 1 { F ) is closed in X .
Proof , (i) => (ii). Let xn G X with xn > XQ . By the continuity of / ( •)
for any e > 0 , there exists a S > 0 such that
d y { f { x ) , f ( x0 ) ) <e d x { x , x0 ) < <5.
Also , there exists an N
^ 1( such, that
) 6,
This leads to
d x xn xo < n
^ N.
d Y ( f ( x n ) , f ( xo ) ) < £ n > N.
Hence, ( ii) holds.
_
(ii) => ( iii) . Let V C Y be an open set . If / 1( Vr ) is not open, then
there exists an xo e f ~ 1 (V ) and a sequence xn e f ~ 1 (V ) c converging to
xo Hence f ( xn ) —> f { xo ) . Since V is open, there exists an e > 0 such that
-
,
B y ( f ( xo ) e ) C V.
Therefore, there exists aniV l such that

^
f (Xn) B y ( f ( x ), e
0 ) Vn > N.
Functions on Metric Spaces 47
This implies that
*« e r ( Bj' ( / (*o ) , e) ) cr (
1 1
n
a contradiction .
( iii) => ( iv ) . For any closed set F C Y , F c is open. Thus,
r l ( Fc ) = r\F )c ,
is open , leading to that /-1 (F) is closed .
( iv ) => ( hi ) . The proof is the same as ( iii ) => ( iv ) .
( iii) => ( i) . Let xo X . For any e > 0, B y ( f ( xo ) , s) is open. Hence
f ( B y ( f ( x0 ) , e ) ) is open , and
~1
x0 e f 1
( BY ( f { x0 ) ,e ) ) .
Thus, there exists a 5 > 0 such that
B x ( x0 , S ) C f -1 ( BY ( f ( x0 ) , e) ) .
This means
f ( B x ( x0 , S ) ) C BY ( f ( x0 ) , e ) ,
.e.
i
d y { f { x ) , f ( x0 )) < e, V d x ( x , x0 ) < S.
Proposition 1.4. L e t ( X , d x ) , ( Y , d y ) , a n d ( Z , d z ) b e metric spaces. Let

—
f : X » y , g : V -> Z. If / ( •) is continuous at Xo e X and g ( ) i s
continuous at f ( xo ) , then go f : X Z is continuous at x$ . Consequently,
-
-
if / ( • ) is continuous on X and g ( ) i s continuous on Y , then ( g o / ) ( • ) is
continuous on X .
Now , we look at an important special case (Y , d y ) = ( M, p). Note that
in R , there is an additional structure the order structure. Due to such
a structure, we can say a little more on any bounded , not necessarily con-
—
tinuous function / : X » R. Suppose X Q e X and / ( •) is bounded in a ball
B ( xo , r ) , for some r > 0. For any S G (0, r] , we have the existence of
/ (so ; S ) = X inf f ( x), f + ( x0 ; S ) = sup / (*) •

B ( X O ,6 \
) { XO } X . B ( X O ,6 )\{ X Q }
Clearly, 8 f ( xo ; <5 ) is bounded and monotone non-increasing. Thus, the

following limit exists:
lim / ( x ) = lim / ( xo ; (5) = sup inf /( )•
) { xo } *
—
x txo < 4- 0
^
Likewise, the following is well-defined:
<5 > 0 x B ( x 0 ,5 \
lim / ( x ) = lim / + ( xo ; 5) = inf sup / (*) •

<5 > 0 cc B( x 0 , 5) \{ a;o }
We called the above the limit inferior (or simply liminf ) and limit supe -
-
rior (or limsup) of / ( •) as x -* xo Then we can introduce the following
definition.
.
Definition 1.5 Let ( X, d ) be a metric space and / : X > R be a bounded —
function.
(i ) We say that / ( •) is lower semi-continuous at xo if
/ ( x0 ) lim / ( x ) . (i.i )
—
a: > cco
( ii) We say that / ( •) is wpper semi- continuous at xo if
f ( x0 ) lim / ( x ) . (1.2 )
X
^XQ
( iii) If / ( •) is lower ( resp. upper ) semi-continuous at every point in X ,
we say that / ( •) is lower ( resp. upper ) semi-continuous on X .
To get some feeling about the lower and upper semi-continuous func-
tions, let us present the following example.
Example 1.6 Let .
1 x=0 -1 x=0
< p( x ) = V» (*) = < x O.
0 x O;
^ -
Then </? ( •) is upper semi-continuous, and ip( ) is lower semi-continuous. ^
The following gives a characterization of lower semi-continuous func-
tions.
.
Proposition 1.7 Let ( X, d ) be a metric space and / : X -> E be a
bounded function. Then the following are equivalent :
( i) / ( •) is lower semi-continuous on X .
( ii) For every xo £ X , for any £ > 0, there exists a 5 > 0 such that
f { x ) > f ( xo ) Vx £ B ( xo , 5) .
( iii) For every a £ R , the set
(/ > a ) = { x £ X | / (x) > a }
is open.
Proof , ( i) => ( ii ). By the lower semi-continuity of / ( •) at xo , we have

f ( x0 ) lim f ( x ) lim
X * Xo
—
inf — /( )•
540 XGB( XO ,5) \{ XO } *
Then for any e > 0, there exists a 6 > 0 such that
f { x0 ) < inf f(x) + £ f ( x) + £, Vz B( x0 , S )\{z0}.
x B ( xo ,6 \
) { xo }
The above is true, of course, for x = X Q . Thus, ( ii) holds.
( ii) (iii ) . Fix any a G R. For any xo G ( / > a ) , we have
f ( x0 ) > a.
Then by ( ii) , for £ = , there exists a S > 0 such that
f ( x0 ) - a / ( x0 ) + a > a
f ( x ) > / ( x0 ) -£ = f ( x0 ) - 2 2
Vx G B ( x o , S ) .
This implies that
F( £o , <5 ) Q { f > a ) .
Thus, ( / > a ) is open ; ( iii) holds.
(iii) => ( i) . For any xo G X and any e > 0,
x0 e ( / > / ( x 0 ) - e ) .
Then there exists a <5 > 0 such that
/ ( x ) > / ( x0 ) - e , Vx G B( XQ , 5) .
Hence
lim / (x ) / ( x0 ) - e ,
X
^XQ
for any e > 0 , which leads to the lower semi-continuity of / ( •) at xo -
Likewise, we have the following result for the upper semi-continuous
functions.
Proposition 1.8. Let ( X, d ) b e a metric space and f : X —> • R be a
bounded function. Then the following are equivalent :
( i) / ( •) is upper semi-continuous.
(ii) For every xo G X, any £ > 0, there exists a 5
for > 0 such that
f ( x ) < f { x0 ) + £ , Vx G L? ( xo , 5 ) .
( iii) For every a G R , t h e s e t
( f < a) = { x e X | / (x) < a}
is open .
The following is pretty obvious.
Corollary 1.9. Let ( X , d ) be a metric space and f : X > R be a bounded
function. Then / ( •) is continuous if and only if it is both lower and upper
—
semi-continuous.
Next , we look at two interesting functions.
Example 1.10. ( i ) The following is called the Dirichlet function:
1 x eQ
/ ( X) = 0 X e R \ Q.
Since for any x £ R ,
_ (x )
yx /
lim =1 hm f ( x ) = 0,
x — x — yx
we see that lim_ f ( x ) fails to exist for any x G l, which implies that the
X YX
—
Dirichlet function is nowhere continuous. It is upper semi-continuous at
every rational point , and lower semi-continuous at every irrational point .
( ii) The following is called the Riemann function:
I
/ (*) = {
if x —|£ Q, with q > 0 and p and q are co-prime,
o^ if x £ R \ Q.
Since for any x G l,
lim f ( x ) = 0,
( why? )
— x yx
we see that this function is continuous at every irrational point and is

discontinuous at every rational point .
We now introduce the following definition which is concerned with func-
tions from a metric space to itself.
Definition 1.11. Let ( X , d ) be a metric space, and / : X —> X . A point

x e X is called a fixed point of / ( •) if the following holds:
f {x) = X.
The following result is called the Contraction Mapping Theorem.
Theorem 1.12. Let ( X , d ) be a complete metric space and f : X —> X
satisfy the following:
d ( f ( x ) , f ( y)) ad ( x , y ) Vx , y e X, (1.3)
where a e ( 0 , 1 ) is a constant . Then / ( ) • admits a unique fixed point .
Proof . Pick any xo E X , define the sequence

£ 71
* = f ( %n —) l n 1. (1.4)
We have the following estimate:
n 1 — d ( xi , x0 ) 1.
d { xn , %n — ) ^ Oid { xn— , Xn — ) ^ ^
l 1 2 OL Vn
Consequently,
t i
— ) ^ = ^n+ — d(o: , x )
^^ ^^
/c l
d ( x n i X n+ £ ) d ( xnJtk , ^n + fc
^ fc= i
l
fc i
i 0
t
1 — cxf ^
— d( x , )
an
^^ ^ afc d (xi , xo ) = a n ^
1 £0
1- a
fc = 0
OLU
— a d (xi , x ) —> 0 0 u t — 00 .
This means that {xn} is Cauchy. By the completeness of ( X , d ) , one has
—
xn y x for some x G X . Then / ( x ) = x. Now , if there exists another
x £ X such that / ( x ) = x, then
d ( x, x ) = d ( / (x) , / (x ) ) ad ( x , x ) .
Since a < 1, we must have x = x , proving the uniqueness.

In the above, (1.4) is called a Picard iteration.
Let us consider two functions f : X -> Y and g : X -» Z . We can define
their direct sum f ® g : X -» Y x Z by the following:
( f ® 9 ) ( x ) = ( f ( x ) , g ( x )) Vx X.
The following can be proved easily.
Proposition 1.13. L e t ( X , d x ) , ( Y , d y ) , and ( Z , dz ) be metric spaces. Let

—
f : X > Y and g : X Z.
-
( i) / ( •) and g ( ) are both continuous at xo £ X (resp. on X ) if and only
if / 0 <7 is continuous at xo ( resp. on X ).
( ii ) In the case that Y = Z = R with the standard metric, if / ( •) and
-
g ( ) are continuous at xo ( resp. on X ) , the functions / ( •) ± g ( ) , f ( ) g ( ) , - - -
- -
/ ( •) M') > / ( ') Vg ( ) = m a g ( ) } , a n d / ( •) A g ( ) = min{/ ( •) , £? ( •)} are -
continuous at XQ ( resp. on X ). For f ( ’ ) / g ( ) , the continuity holds in its -
domain.
Exercises
1. Prove Proposition 1.4.
2. Let ( X, d x ) and (F, dy ) be two metric spaces.
( i ) Suppose ( X , dx ) is discrete. Then any function f : X —> Y must be

continuous.
( ii) Suppose ( X, d x ) = ( Mm , p z ) and (F, dy ) are discrete. Then f : X -A

Y is continuous if and only if / is a constant function.
—
3. Let / : Mm » R e , with / ( # ) = ( /i ( x ) , , / ( x )) , for all x e Mm .
^
* * *
Then / ( •) is continuous if and only if each /* ( •) is continuous, 1 i L

4. Let ( X, d ) be a metric space. Then the metric d : X x X —> R is
continuous.
5. Let ( X, dx ) and ( F, d y ) be two metric spaces and / : X —> Y such

that
dY { f ( x) J ( y )) L d x ( x, y ) Vx, y e X ,
for some L > 0. Then / ( •) is continuous.

6. Let G C Mm be a domain. Then G is pathwise connected, i.e., for
any G G, there exists a continuous function 7 : [0 , 1] > G such that —
7( 0) = x and 7(1) = y .
7. Find a counterexample showing that a connected set in Rm might
not be pathwise connected .
8. Prove Proposition 1.13.
—
9. Let f\ : R > R ( A E A ) be a family of continuous functions. Let
F ( x ) = sup f x ( x ) x G M.
AGA
Show that F( - ) is lower semi-continuous.
2 Properties of Continuous Functions
In this section , we will present several important properties of continuous

functions.
2.1 Compactness preserving and uniform continuity

The following result is concerned with the compactness preserving property
of continuous functions.
Theorem 2.1. Let ( X , d x ) and ( Y , d y ) be metric spaces. Let f : X
be continuous.
—> Y
(i) Let K C X be compact . Then
I
f ( K ) = {/ (*) x e K }
is compact in Y .
(ii ) Let y = R , and ( X , d x ) be compact . Then / ( •) is bounded and it
attains its maximum and minimum.
Proof , ( i) Let { y n }n i he any sequence from f ( K ) . Then there is a
sequence { x n }n i C K such that ^
^
f { x n ) = Vn n 1.
Since K is compact , there exists a subsequence { x n k } k
x e K . Then by the continuity of / ( •) , we have ^ i such that x U k —>
Vnk = f ( x n k ) -* f ( x ) f ( K ),
which shows the compactness of f { K ) .
(ii) See the next theorem.
The next theorem generalizes ( ii) of the above theorem .
.
Theorem 2.2 L e t ( X , d ) be a compact metric space and f : X ^ R be —
lower (resp. upper ) semi-continuous. Then / ( •) attains its minimum ( resp.
maximum ) on X .
Proof . Let xn G X be a minimizing sequence , i.e.,

lim f ( x n ) = inf f ( x ) .
n —t o o x£ X
Since X is compact , there exists a subsequence x U k such that
lim d ( x n k , x ) = 0,
—
k » oo
for some x E X . Then by the lower semi-continuity of / ( •) , one has

inf f { x ) f {x) lim f ( x n ) = inf f ( x ) .
x£ X —
n > 00 xGX
Thus, x is a minimum of / ( •).

The case of upper semi-continuous function can be proved similarly.
We now present an interesting example.

. .
Example 2.3 Let X = £ l Let B = B ( 0, 1) be the closed unit ball in X
centered at 0. Define h : B > R as follows:
oo
— •
h{ x ) — 5=> ( |* | —
fc l
2 fe
l) + 5 Vz = (:xk ) G B 5 ( 2.1 )
where a+ — a V 0 = max{a , 0}. Note that for any x = ( xk ) G B,

oo
Ei *fci ^ -
= k 1
Thus, there are at most one non- zero terms in the series ( 2.1) defining /i ( - ).
-
Now , we show that h( ) is continuous on B . To see that , we let x = ( xk ) G B
be given, and let xn = ( x£ ) G B such that
oo
E= \ xn Xk \ -> 0.
Pl ( xn , x ) =
k 1
~
We may assume that there is a X 1 such that

- 1
^
I*f c iI <
i
4
Vfc K.
Then , there exists an 1 such that
^ 1
N, k
|a£|
Hence 5
| - xk \ + \ xk \ |0i (xn , z) + \ xk \ < - 5
^ K.
K
+
Kxn ) = Efc ( 2 lXn \ ~ !) 5 n
^ iV,
k= l
with K
^ 1 independent of n ^ iV. Consequently,
if
Ih( xn ) - h{ x)| E= (2 IX« I ~ x ) + ( 2|zfe| - 1) +

k 1
if oo
E=
fe l
2fcix
- zfe| 2 KE= l ^
fc l
n - = 2 KPl ( xn , x ) -> 0 .
-
This proves the continuity of h ( ) at such a point x. Next , for xn = ( Snk ) k ^ i 5
we have
h ( xn ) = n —> oo , as n oo.
-
Therefore, /i ( ) is continuous and unbounded on the bounded and closed
set B . We now define
1
9{ x ) = 1 h( x ) x e B.
+
Then g ( - ) is continuous and
inf g ( x ) = 0, g( x ) > 0, x E B.
X B
This means that continuous function g ( - ) does not attain its minimum over
the bounded and closed set B.
Recall from Chapter 1, Example 4.16 that in i 1 , the closed unit ball
5(0, 1) is not compact . Hence , the above example tells us that if the domain
space X is not compact , a continuous function defined on a bounded and
closed set might not be bounded and might not achieve its minimum and
maximum .
Next , let us look at another example which will lead us to another issue.
Example 2.4. ( i) Prove the continuity of f ( x ) = x 2 on [0 , 1].
Let XQ .
E [0 , 1] One has
\ x 2 — XQ \ = \ x — X o \\ x + X o \ 2 \ x — X o \ .
Hence , for any e > 0 , by taking 5 = § > 0, one has
If ( x ) - f ( xo ) | = \ x 2 - X Q \ 2 \ x - x0| < 25 = s , V|x — xo | < 6.
We note that S =|is independent of X Q .
( ii ) Prove the continuity of g ( x ) = - on ( 0 , 1].
Letxo E (0, 1]. For \ x — xo \ < we have x > Now , for \ x — xo \ < — ’
2
\ g( x ) - g { x0 ) \ =
1
-
1 \ x - x0 \
Jb Xo XXQ x0
Thus, for any e > 0 , by taking 5 = min { —

2 ’ — },
2 we have
\ g ( x ) - g ( x0 ) \ = \ X X o \ ^ —^ \ x — xo| < e,
X 2
XQ
—
-
V| x XQ | < S.
XQ
-
This proves the continuity of g ( ) on (0, 1] . In this proof , 5 not only depends
-
on > 0, but also depends on xo We claim that for this function , one
cannot find a S > 0 that is independent of XQ . In fact , if there existed a
5 = 5 ( e ) independent of xo , then , say, for e = 1, we would have a 5 > 0

such that
1
\ 9( x ) - g { - ) I
1 1 |nx — 1| < I Vx — —n < 8.
x l/n X
But this is impossible , since we may take x = -~~T with n

so that
lb X
^ 1 large enough
2 1 n+1
0< < 8.
n 1 n n( n 1) — —
Then for such an x
1>
|nx — 1| n— 1
2
n+1
Vn 1
X
n— 1
2
which is a contradiction.
The above examples show that sometimes, even for one variable func-
tions, continuity could be quite different. The following notion reveals such
a difference.
Definition 2.5. Let ( X, dx ) and ( Y, dy ) be metric spaces. A map / : X —>

Y is said to be uniformly continuous on X if for any e > 0, there exists a
5 = 5 ( e ) > 0 , only depends on £, such that
dy ( / (*) > A*') ) <e Vx , x' G xr
X , dx ( x , ) < 5.
The following result gives uniform continuity of continuous functions in
metric spaces.
Theorem 2.6 . Let ( X , dx ) &nd (Y , dy ) be metric spaces, with ( X , d x )

being compact . Let f : X —» Y be continuous. Then / ( •) is uniformly
continuous.
Proof . Fix any e > 0. For each XQ E X , by the continuity of / ( •) , there
exists a 5 ( X Q )
> 0 such that
dy ( / (*) , / (*o ) ) |
< X E J3 ( xo , 5( xo ) ) .
Then for any x , x' E B ( X Q , <5 ( XO ) ) , we have

d y ( f ( x ) , f ( x') ) d Y ( f ( x ) , f ( x0 ) ) + dy ( / ( x0 ) , / ( x') ) < £.
Clearly,
XC 1J B ( x0 , 6 ( x0 ) / 2 ) .
XQ& X
By compactness of X , we can find xi , • • • , xn G X such that
XC \J B ( xi , S ( xi ) / 2 ) .
i=1
Take
5= = min 5 ( x i ) / 2 > 0.
Now , let x , x' £ X with dx ( x , x' ) < S . There must be some 1 i n such
that
x G B ( xi 5( xi ) / 2).
)
Then
d x ( x’ , X i ) d x ( x' , x ) + d x ( x , X i ) < 5 +
Hence, x, x' G j9 ( x* , <5 ( x* ) ) which implies
dy ( / (x ) , / (*') ) < e.
This proves the uniform continuity of / ( ) . •
Let us provide a different proof of the above result .

Proof . Suppose / ( •) is not uniformly continuous. Then there exists an
£o > 0 such that for any n 1, there are two points xn , yn G X satisfying
d y { f ( x n ) , f { y n j) > £o , d x( xn, yn ) <
Since X is compact , there exists a subsequence { xnk }k
that for some x, ^ i of {xn} ^ n i such
lim dx { xnk 5 x )
k — ¥ oo
= 0.
Consequently,
dx ( ynk , x )
^ dx { xnk , 2 /nfc ) + dx (x nic x ) ? <—
nk
+ dx{x nk 5 —
x ) > 0.
Then
o < £o < lim d y ( f ( xnk ) , f ( ynk ) )
K OOr
felm ( d y ( f ( xnk ), f ( x ) ) + d y ( f ( x ) , f (y n k ) ) ) = 0
a contradiction. ^
It is not hard to see that in Example 2.3, the continuous function h ( - )
defined on the closed unit ball is not uniformly continuous. Hence, the
compactness condition of X cannot be replaced by boundedness and com-
pleteness in the above theorem.
2.2 Connectedness preserving

We recall the definition of connected / disconnected sets ( Definition 3.1 ( x) ,
and its equivalent condition in Proposition 3.9 of Chapter 1). We present
the following connectedness preserving property of continuous functions.
Theorem 2.7. Let ( X , dx ) and ( Y , dy ) be two metric spaces, and f : X > —

V be continuous. Let E C X be connected . Then f ( E ) is also connected .
Proof . Suppose f ( E ) is disconnected. Then by Proposition 3.9 of Chap-

ter 1, there are disjoint non-empty open sets U and V such that
f(E) cu u v, f ( E ) nu , f ( E ) nv 0.
Since / ( •) is continuous, / 1 ( U ) and

^ X
/ ( )
0
V are open in
^
X. Further ,
E c r\u u v ) = r\u ) u r\v ).

Moreover , / 1 ( U ) and / 1 (V ) are disjoint , and
E n f -\u ) E n f -\V )
^ 0,
This means that E is disconnected , a contradiction .

^ 0.
The above result has a well-known corollary which is presented here.
Corollary 2.8 . ( Intermediate Value Theorem ) L e t ( X , d x ) be a met-

ric space and f : X -» R be continuous. Let E C X be connected , and
a , b G E with
m < m.
Then for any y G ( / ( a ) , f { b ) ) , there exists a c G E such that
/ (c) = y.
Proof . To prove our conclusion , we first claim that any set G C R is

connected if and only if it is an interval. In fact , for any x , y G G , we claim
that [ x , y\ C G. By contradiction , if there exists a z G ( x , y )\ G , then
—
[ ( oo , z ) n G ] U [ { z , oo ) D G\ = G .
—
We see that ( oo , z ) C\ G and ( z , oo ) D G are two disjoint non-empty open
sets in G , which means that G is disconnected , a contradiction. Now , let
-hoc.
a = inf G ^ — oo, b = sup G
Then the above result implies that ( a , b ) C G. We then have the following
situations,
( a, b ) , a , b $ G,
r
[a , 6) a <= G , b £ G,
( a , 6] a G , 6 e G,
k
[« > ^
CL , b E G .
Thus, G is an interval. Conversely, of course, if G is an interval , then , it is

clearly connected .
Now , since E is connected , by Theorem 2.7, f ( E ) is connected in R ,
which must be an interval. Hence, ( / ( a ) , / ( 6) ) C f ( E ) . Our conclusion
then follows.
Exercises
1. Let ( X , d x ) and (Y , d y ) be two metric spaces. Assume that ( X , d x )
is compact , and / : X —> Y is bijective and continuous. Then f ~ l : Y » X —
is also continuous. What if ( X , d x ) is not compact ? Why?
2. Let ( X, d ) be a compact metric space and / : X —> X be a continuous
function such that
d( f { x ) , f ( y ) ) < d( x , y ) V x, y e X , x 2/ .
Then / has a unique fixed point.
^
3. Let ( X , d ) be a compact metric space and /, g : X > R be continuous — •
functions such that
/ ( z ) 7^ 0( z) Vx G X.
Then there exists an e > 0 such that
\ f ( x ) - g{ x )\ e, V x s X.
4. Show that lim

— ( sinj) = 1 and lim—>
x fO x 0
. 1
sin —X - -1.
. 1
5. Find lim (sin a
—> o Lv
x
—x; sin
6. Let / : R —> R and A\ C R ( A E A ) be a family of sets. Then
r\u a>)= u r ^) .
AGA AGA
1
/
_i
( n a\) = AGA
AGA
n
—
7. Suppose / : [a , b] > R is continuous such that for any £ E Q D [a , b]
— -
/ ( £) 0 Then f ( x ) = 0 for all x E [a , b].
8. Let / : [a , 6] R be continuous. Let a < x\ < X2 < ••• < xn < b.
Then there exists a £ E [a , b] such that
f ( x i ) + / (X2 ) + + / (S n )
/( ) = n
9. Let / : [a , 6] -» R be continuous such that / ( x ) 0 for all x E [a , b] ,
and there exists a £ E [a , 6] such that / (£) > 0. Then / ( x ) > 0 for all ^
x E [a , 6].
—
10. Suppose / : [a , 6] » R is an increasing function and for any y E
•
[ / (a ) , / ( 6) ] , there exists a £ E [a , 6] such that / ( £) = t/ . Then / ( •) is

continuous on [a , 6] . What happens if / ( •) is not monotone? Why?
11. Let / : ( a , 6) R be lower semi-continuous such that
lim / ( x ) = lim / ( x ) = oo.
£
—> a +0 ——
x >6 0
Then there exists a £ E ( a , b ) such that
/ (£) < f ( x ) Vx E ( a , 6) .
12. Let SCR ( not necessarily bounded ) and f , g : 5 -» R be uniformly
continuous. Then / d are uniformly continuous on 5. Is / •# also uniformly
=^
continuous? Why?
13. Let 5 be a bounded set in R ( not necessarily closed ) and f : S R
be a uniformly continuous function. Then / is bounded on S . What if S
is not bounded , why?
—
14. Let / : ( a , b ) > R be continuous. Then / ( •) is uniformly continuous
•
on (a , b ) if and only if both lim / ( x ) and lim / ( x ) exist and finite.

x — —
x >b ~
15. Prove or disprove that the following functions are uniformly contin-
uous in their domains:
(i) f ( x ) = sin A on (0, TT ] ;
(ii) / (x ) = sinx on ( — oo , oo ) ;
( iii) / ( x ) = x 2 on ( — oo, oo ) .
16. Suppose / : R R is uniformly continuous. Then there exists a
constant K > 0 such that
1 / 0*01 K {\ + |ar|) Vx E R.
Chapter 3
Differentiation
1 Derivatives
Throughout of this chapter , G C Mm will be a domain ( a connected and

open set ). Let / : G —» M and xo G be given . We would like to look at
the behavior of / ( •) near the point Xo ( by which we mean in some small
neighborhood of XQ ) beyond the limit lim / ( x ) and the continuity. To this
end , we introduce the following definition.
X
—
+ XQ
Definition 1.1. Let / : G —> R.

( i ) Let xo G , and let u G 5 ( 0 ) = { u e Mm | ||tt|| = 1}. Suppose
G
/'(*o ; «) si =
^
lim
f ( x0 + Su ) - f ( xo )
( 1.1 )
o (5
exists. Then we say that / ( •) is directionally differentiable at xo in the

direction u , and /' ( xo ; u ) is called a directional derivative of / ( •) at xo in
the direction of u. If f ' ( xo ] u ) exists for all u E 5]n ( 0 ) , we say that / ( •) is
Gateaux differentiable at xo , and call the map u f ' ( xo ] u ) the Gateaux
derivative of / ( •) at XQ .
( ii) If for xo E G, f' ( xo ; u) exists for every u E 5 ( 0 ) , and there exists
a row vector , denoted by ( V / )( xo ) , such that ^
f ' ( x0 ; u ) = ( V / ) ( x 0 ) u , Vtz E 5im ( 0 ) ( 1.2)
-
then ( V / ) ( xo ) is called the gradient of / ( •) at xo If there exists a function ,
denoted by V / : G —> R l x m such that (1.2 ) holds for all xo E G , then we
call V / the gradient field of / ( •) .
( iii) If for xo G G, there exists a row vector, denoted by / ( xo )
X E Mlx m
such that
lim \ f ( xo + y ) ~ f ( xo ) - /S ( ZQ ) 2/ 1
=0 (1.3)
IMI -m liyll
61
then we say that / ( •) is Frechet differentiable at xo , and f x ( xo ) is called

-
the Frechet derivative of / ( •) at xo In the case m = 1, we usually denote
f x ( xo ) by f' ( xo ) .
From the above definition , we see that if / ( •) is directionally differen-
tiable at #o in the direction u G Sf O) , then
/'(x0 ; An ) = lim
f ( x0 + S X u ) - / ( x 0 ) ^
=\f' ( x0 ; u ) , VA 0. (1.4)
<5 —
>0 5
Consequently, if / ( •) is Gateaux differentiable at xo , then the Gateaux
derivative u ^ /'( xo ; u ) can be extended to a map from Rm to R ,
which is positively homogeneous. However, this map might not be linear.
Whereas , if / ( •) admits a gradient ( V / )( xo ) , then the map u »->• /'( xo ; u ) =
( V / ) ( xo ) u is linear. The following example shows that Gateaux differen-
tiability does not necessarily imply the existence of a gradient .
Example 1.2. Let m = 2 and
xy
{ x, y ) ± ( 0, 0 )
f ( x, y ) = yjx + y21
0 ( x , y ) = ( 0, 0 ) .
For any ( a , /3) 6 K2, we have
f ( 6 a , 6 /3 ) - f ( 0 , 0 )
/'( (0, 0) ; (a , /3)) = slim
io 5 /
y a2 + /32
which is positively homogeneous in (a , /3 ) , but it is not linear. Hence,
( V / ) ( 0 , 0 ) does not exists , although / ( •) is Gateaux differentiable at ( 0 , 0 ).
As a matter of fact , even in the case m = 1, Gateaux differentiability
does not implies the existence of a gradient. An easy counterexample is
f ( x ) = \ x \ at x = 0. We have
/'( 0; A ) = |A| A G R,
which is not linear in A. Thus, ( V / )( 0 ) fails to exist .
Next , we have the following result.
—
Proposition 1.3. L e t f : G >• R and XQ G . I f / ( •) i s Frechet differen -
tiable at X Q , then / ( •) admits a gradient ( V / )( XQ ) . Moreover ,
( V / ) (*<> ) = /* (so ) . (1.5 )
Consequently, for any u G 5jn (0) , the directional derivative of /( )
• in the
direction u at XQ is given by
f ( xo ; u ) = f x ( x0 )u . ( 1.6 )
Differentiation 63
Proof . For any u E S ]a ( 0) , by (1.3) , we have
lim l /
( x0 + Su) - / (sp ) - <5/s (soH _ (1.7)
<5 +0 5
which implies that
f { x0 + Su ) - / ( z0 )
lim = fx ( xo ) u. (1.8)
<5 +0 8
Thus, by definition , we see that the gradient ( V f )( xo ) exists at xo and
(1.5) holds.
From Definition 1.1, we see that for m = 1, the existence of a gradient
and the Frechet differentiability are equivalent. However , we claim that the
existence of a gradient does not necessarily imply the Frechet differentia-
bility, for m 2. To show this , let us first present the following result .
^ . —
Proposition 1.4 I f f : G » R is Frechet differentiable at xo , then it is
continuous at XQ .
Proof . By definition , we have that
lim /
| ( x ) - f ( xo ) - f x ( xo )(x - x0 )| _ |-Rl (s, X0 )|
— lim
— = (1.9)
where
x yxQ x - X o \\ x
^ x0 ||x XQ||
R l ( x , X o ) = f ( x ) - f ( x0 ) - f x ( x0 )( x - xo ) . ( 1.10 )
Thus
( 1.11 )
|i?i ( x , x0 )| '
l|x Xoll ll
lim + /x (x0 )||
< X X Q || = 0,
X >£o
— —
proving the continuity of / ( •) at x o -
Note that (1.10 ) can also be written as
f { x ) = f ( xo ) + f x { xo ) { x - X o ) + R i ( x , x0 ) ( 1.12 )
with the remainder term R i ( x , xo ) given by (1.10) . We refer to (1.12 ) as a
first order Taylor expansion. We will come back to this a little later .
The above result will help us to see that having gradient does not nec-
essarily imply the Frechet differentiability. Here is an example.
.
Example 1.5 Let m = 2 and
x3 y
/ ( x, y ) = x6 + y 2
(x , y ) ± (0 , 0 ) , (1.13)
0 ( x , y ) = ( 0 , 0) .
Then for any (a , /3) £ R 2 with /3 ,

/ ( to, <5/3) - (0 0)
^
/ ,
0 we have
52 /3 as
no < a6 + /?2 =
lim
540
2
5
~
54 °
and for any (a , 0 ) £ R
lim / ( to, 0) - / ( 0, 0)
<540 5
Thus, we have
( V / )(0, 0 ) = ( 0, 0) .
On the other hand , by taking a sequence {( xn , yn )} approaching to ( 0, 0)
along the curve y = ax3 for some a £ R , we see that
lim f ( xn , y n ) =
71
—
» 00
Hence, for different choices of a £ R , the above limit is different . This ^

—1 + Gr
means that / ( x, £/ ) does not have a limit at ( 0, 0 ) , which implies that / ( •, •)

is not continuous at (0 , 0 ) . By Proposition 1.4, / ( •, •) must not be Frechet
differentiable at ( 0 , 0 ) .
Note that in Definition 1.1, various differentiabilities were introduced

without using the coordinates of the space. Now , in the case that one uses
the coordinates, some more interesting features will be revealed .
Let e* £ Rm be the vector whose i-th component is 1 and all others
are 0. We call {ei , • • • , em } the standard orthonormal basis for Rm . If / ( •)
admits a directional derivative at Xo in the directions of ±e* , and
— e i ) = - f ( x0 ; e i ) (1.14 )
then we denote
f x‘ ( x 0 ) = e i ) = lim
f ( x0 +
(1.15) Sei ) - f { x 0 )
5 « 4.0
and call it the ( first order ) partial derivative of / ( •) at xo with respect to
x \ Note that when (1.14) holds, we have
lim / ( x0 + S e { ) ~ f ( x0 )
lim / ( x0 + 5e< ) - f ( x0 )
(1.16 )
<5 —^0+ 6 6^ - 0- 5
Thus,
f i x 0 + Sei ) - f i x0 )
f x* i x0 ) = lim (1.17)
5 0 < -> 5
Differentiation 65
Further , if m = 1, we simply denote

f ' { xo ) = f x i { x0 ) (1.18)
and call it the derivative of / ( •) at x$ .
The following result gives a representation of gradient as well as direc-
tional derivatives in terms of partial derivatives.
Proposition 1.6 . Let f : G —» R , xo £ G , and / ( • ) admit gradient

-
( V / ) ( xo ) at #0 Then / ( •) has partial derivatives f x i ( xo ) , 1 i m.
In this case, under the standard orthonormal basis of Rm , the gradient
( V / ) ( xo ) admits the following representation:
( V / )( x0 ) = , f x m { xo ) ) - (1.19)
Consequently, for any direction u = ( u l , , vm )T e Si"(0) ,
m
f ' ( x0 ; u ) =
^
i=1
2 u l f x i ( x ).
0
Proof . Taking u = ± e* in (1.1) , we see that f x i ( xo ) exists for each

( 1.20 )
1 < i m . Next , for any u e Sf O ) , we have

^ m
y
u = ' u7 ej
' ,
( 1.21 )
i= 1
with u l — u - C i (1 i < m ) . Then
m m
yV ( V / ) (a:o )ei = '
( V / ) (a;0 )« =
i= l
^
i= l
u i f x i ( xo )
/ u 1\ ( 1.22 )
( f x1 ( ^0 ) 5
* ? fx (ô ))
171
;
\Um /
This implies the representation (1.19) of ( V / )( x0 ) under the standard or-
thonormal basis of Rm . Then (1.20 ) follows from (1.2 ) and (1.19 ) .
Let us go back to look at Example 1.2. It is clear that
/* (0, 0) = /y ( 0 , 0) = 0,
but V / (0 , 0) does not exist . Thus, one could have existence of all first order
partial derivatives without the existence of the gradient .
The following result collects some basic properties of the gradients.
Proposition l . T. Let f , g : G -» R admit gradients at XQ .
E G Then
( i) for -
any A , /i E R , A / ( •) + pg ( ) admits a gradient at Xo and
[V ( A / + W)] ( x0 ) = A ( V / )( x0 ) + / i( Vff )(xo, ) . (1.23)
- -
( ii) the product f { ) g { ) admits a gradient at XQ and
[V ( /s)] (xo ) = g ( x0 ) ( V / ) (x0 ) + f ( xo )(V g ) ( xo ) . (1.24)
1
( iii) the reciprocal / ( ) admits a gradient at XQ , provided f ( xo ) / 0,
and in this case,
( V / ) ( xo )
[ vcr : )] ( x ) =
1
0
f ( xo ) 2
(1.25)
The proof is straightforward . We also have similar result with the gra-
dient replaced by any first order partial derivative. We omit the details
here.
Next , we look at the vector-valued function case. Similar to the scalar
case, we introduce the following.
Definition 1.8 . Let / : G —> R^ .
xo E G, and let u E Sf O ) . Suppose
( i) Let
/'(x0 ; u ) = 510
lim
^
/ (x0 + S u ) - f ( xQ ) (1.26 )
5
exists. Then we say that / ( •) admits a directional derivative f' ( xo ; u ) at xo
in the direction of u. If f' ( x y, u ) exists for all u E 5]7l (0 ) , we say that / ( •)
is Gateaux differentiable.
(ii) If for xo £ G, there exists an ( i x m ) matrix, denoted by ( V / ) ( xo ) £
R^ xm such that
/'(x0 ; u ) = ( V / ) ( x0 ) w , Vu E 5 (0 ) (1.27 ) r
then ( V / ) ( xo ) is called the gradient of / ( •) at xo. If there exists an ( x m )
matrix-valued function , denoted by V / : G —> Rm such that (1.2 ) holds for
•
^
all XQ £ G, then we call V / the gradient field of / ( ) . •
( iii) If for xo £ G, there exists an ( £ x n ) matrix, denoted by f x { xo ) £

R^ xm such that
lim 11 / QPQ + y ) ~ f ( xo ) - f x ( xo ) y \\ _0 (1.28)
IMI -+0
then we say that / ( •) is Frechet differentiable at xo Further , if / ( •) is -
Frechet differentiable in a neighborhood of XQ £ G and f x (’ ) is continuous
in that neighborhood , we say that / ( •) is continuously Frechet differentiable
near XQ , or simply / ( •) is G1 near XQ .
Differentiation 67
The following result links the (directional, Gateaux, and Frechet ) differ-
entiability of a vector-valued function with its scalar function components.
Proposition 1.9 . L e t f = ( f 1,• •• , fe ) : G -» and XQ .
E G Then
( i) / ( •) is directionally differentiable at XQ in the direction u E 5]n ( 0 ) if
and only if so is each / * ( •) , and in this case ,
T
^
f' ( x0 ; u ) = ( ( / '( xo ; «) , • • , ( feY ( xo ; « ) )
( ii) / ( •) is Gateaux differentiable at XQ if

- (1.29)
and only if so is each / * ( •) .
( iii) / ( •) admits a gradient at XQ if and only if so is each / * ( •) , and in
this case ,
( V / )( xo ) = R£ x m (1.30 )
( v /* )( x0 ) y
( iv) / ( •) is Frechet differentiable at XQ if and only if so is each / * ( •) , and
in this case ,
/ /i (x0 )
fx ( x0 ) = G Ri x m (1.31)
\ /i (*o )
Note that under standard orthonormal bases of Rm and R/ , we can
write more explicitly that
( f i x (xo ) /ia ( x0 )
/xi ( Xo ) /Ja (x0 )
• ••
• • •
-
/i ( x0 ) N
/xm ( x0 )
/ ( XQ )
X G R^ x m . (1.32 )
J
^
\ /*i (x0 ) / ( x0 ) • • • f £ m ( x0 )
The above is called the Jacobian matrix of / ( •) with respect to x ( at xo ) .
In the case f = m , the Jacobian matrix f x ( xo ) E Rmxm is a square matrix,
and we may define its determinant det [/^ ( xo ) ] which is called the Jacobian
( determinant) of / ( •) with respect to x ( at xo ) . Sometimes, one denotes
0( / V - , / m )
det [/x ( x0 ) ] =
^ (x0 ) =
Next , we look at the derivative of composition of functions.
Theorem 1.10. ( Chain Rule ) (i ) Let g : G »
^( x1, • • • , xm )
be directionally dif - —
(1.33)
ferentiable at XQ E G in the direction of u E 5 ( 0 ) and f : Re Rk

fl
be Frechet differentiable at g ( X Q ) . Then / 0 9 : G -t lk is directionally

differentiable at XQ G in the direction of u e S ” ( 0) , and
( / 0 9 )' ( x 0 ; u ) = fy ( g ( x0 ) ) g' { x0 ; u ) . (1.34 )
( ii ) Let g : G -> be Gateaux differentiable (resp. admitting a gra-
dient , Frechet differentiable ) at XQ G and f : > Rfc be Frechet
differentiable at g ( x 0 ) . Then f og : G -> Rk is Gateaux differentiable ( resp.
—
admitting a gradient , Frechet differentiable ) at xo G , and
( / 0 g )' ( xo ; u ) = fy ( g ( x0 ) ) g' ( xo ; u ) Vu G Rm . (1.35)
( resp. [V ( fog ) ] ( x0 ) = fy ( g ( xo ) ) CVg )( x0 ) (1.36)
( f ° g ) x ( x0 ) = fy ( g ( x0 ) ) gx { xo ) ) (1.37)
Proof . We only prove ( i) . By the Frechet differentiability of / ( •) at
g ( x 0 ) , we know that
f ( y ) = f ( a ( x0 )) + fy ( g { xo ) ) { y - g ( x0 )) + Ri { y, g ( x0 ) ) (1.38)
with
,. I|fli (y; g(g0)) ll n
.
»-*(*0 ) \\ y - g ( xo ) \\
(1.39)
Now , by taking y = g ( xo + 5u) , we have

f ( g ( xo + Su ) ) - f ( g ( x 0 ) )
(1.40 )
= fy ( g ( xo ) ) [ g ( xo + Su ) - p(x0 )] + Ri ( g ( x0 + 5u); g ( x0 )).
Hence
lim / (g(*0 + <5u ) ) - f { g ( x0 ) ) - fy ( g { xo ) )g' ( xo\ u )

si 0 5
rg ( o w ) -5(a;o ) 7?i (g (a;o + <5w ) ;g (ô ))
= olimil )
fy ( g ( xo ) ) ^ +
5
^ -p'(x0 ;u ) 5
g ( x0 + 5u ) - g ( x0 )
ll /y ( ( ô ))|| lim - g' ( x0 ;u )
^ diu (5
(
\\ Ri 0 +
,. g ( x Suy , g ( xo ) ) \\ g ( x0 + 5u ) - g ( x0 ) \\
<5- 0 \\ g ( x0 + Su ) - g ( x0 ) \\ 6 =
^
This proves (1.34 ) .
We note that in the above, / ( •) is assumed to be Frechet differentiable.
A natural question is whether we can relax this condition . In other words,
naively, we expect the following: If / ( •) has a gradient at g ( xo ) and g ( ) -
Differentiation 69
is Frechet differentiable at xo , then it seems that the following should be

true:
[ V ( / ° s ) ] (io ) = ( V / )(ff ( x0 ))3x ( x0 ) . (1.41)
Unfortunately, the above is not true in general. Here is a counterexample.
Example 1.11. Let / ( • ) be defined by (1.13) . Then , from Example 1.5,

we know that V / ( 0 , 0) exists and / ( •) is not Frechet differentiable at (0, 0 ).
Now, let
g ( z ) = { z , z 3 )T z R.
Then
V / (s( 0))fl* ( 0 ) = V / ( 0, 0 ) Q = 0.
Whereas
1
2’
o
^z = .
z 0
0
which is not directionally differentiable at ( 0 , 0 ) in any direction. Hence,
(1.41) does not hold .
The following is usually referred to as the Fermat Theorem.
Theorem 1.12. ( Fermat ) Suppose / : G > R attains a local maximum—

( or minimum ) at point XQ G G at which / ( •) admits a gradient . Then
( V / ) (*o ) = 0. (1.42)
Proof . Suppose / ( •) attains a local maximum at xo By definition , this
means that there exists an open ball B ( xo , <r ) C G such that
-
/ 0*0 f ( xo ) Vx G B ( xo , a ) . (1.43)
Thus, for any u G
^^ (O) , we have f ( x 0 + S u ) - f ( xo)
( V / )( x0 ) • u = lim
54.0 S < 0. (1.44)
Consequently, one must have (1.42 ) .

The case of local minimum can be proved similarly.
Any point io 6 G satisfying (1.42 ) is called a critical point. Thus,
a critical point is a candidate for a local maximum or a local minimum.
Note that in the above, we have only assumed the existence of a gradient
( V / ) ( XQ ) of / ( •) at xo ( not the Frechet differentiability ).
Theorem 1.13. ( Cauchy Mean Value Theorem) Let a , b e G C Rm ,
with a 7 b , such that
^
— {\a -f- (1 — A )b | A G [0, 1]} C G.
[a , b] (1.45)
Let f : [a , b] —» and g : [a , b] —> R be continuously Frechet differentiable.
Then for any w G R^ , there exists a £ G (a, b ) = [a , 6] \ {a , 6} ( depending
on w ) such that
9 x ( 0 ( b ~ a ) [ f ( b ) - / ( a ) ] • w = \g ( b ) - g ( a ) ] [ f x ( £ ) ( b - a ) ] w . (1.46 )
Note that if
then (1.46) can be written as

9x ( £) ( b a)
^ 0,
[.f ( b) - / ( « )] ' { [9 ( b ) ~ 9 ( a ) ] w } . (1.47)

9 x (0 ( b - a )
There are several useful special cases of the above,
(i ) g ( x ) = ( b — a ) • x. In this case, with a =/= 6,
gx( x ) = ( b — a) T
Vx G Rm ; 9 ( b ) - 9( a ) = \\ b — a\\ .
2
Thus, (1.46 ) is equivalent to the following:

l f ( b ) - f ( a )] w = [ f x ( £) ( b ~ a ) ] • w .
• (1.48 )
From this , by taking w = - / ( « ) ( assuming that f ( b )
— fWu — f ( a ) ^ 0) , we
have
f ( b ) - / (a )
ll / W - / (o) ll = [/* (0(6 - o)] - 11 / (6) - / (o)|| (1.49 )
which leads to
11 / ( 6) — / (o)|| < ||/x (0(6 — o)||. (1.50 )

(ii) g ( x ) = ( b — a ) • x and i — 1. In this case, by taking w — 1, (1.48)
becomes
f ( b) ~ f ( a ) = f x ( 0 ( b - o ). (1.51)
This is called the Lagrange Mean Value Theorem for multi-variable func -
tions. A further special case of this is G = [ a , b ] and f ( a ) = f ( b ) , which
leads to the well-known Rolle’s Theorem. We also point out that for the
Differentiation 71
case m = £ = 1, since the Frechet differentiability is equivalent to the

existence of a derivative, one has (1.51) under the condition that / ( •) ad-
mits a derivative at every point on the interval ( a , 6) , without assuming the
continuity of the derivative.
( iii) G = [ a , b\ C R and = 1. In this case, by taking w = 1, we see
that (1.46 ) is equivalent to the following:
^
m m - f (a) ' (1.52)
9' iO 9( b ) - g ( a )
This is also referred to as the Cauchy Mean Value Theorem.
Proof of Theorem 1.13. First , if
9( b) ~
9( a ) = 0, (1.53)
we let
h( t ) = g { a + t ( b - a ) ) - g ( a ) t e [0, 1] . (1.54 )
—
Then h : [0, 1] > R is continuous and Frechet differentiable. Moreover ,
h { 0) = h( 1) = 0. Thus , by Rolle’s Theorem ,1 there exists a 6 £ ( 0 , 1) such
that
0 = h' ( 6 ) = gx ( a + 0 ( b — a ) ) ( 6 — a ).
• (1.55)
Hence, with £ = a 4- 0 ( b — a ) £ (a , 6) , both sides of (1.46) are zero.
Next , let (1.53) fail. For any w G R£ , let
h( t ) = f ( a + t( b - a ) ) - f (a) -w
m ~
/ ( «) ] *
— \g ( a + t ( b - a ) ) - g ( a ) ] t e [0 , 1] .
(1.56 )
9( b) ~
9( a )
Then h : [0 , 1] -* R is continuous and admits derivatives on (0 , 1) . More-
over , we again have h { 0 ) = h(1) = 0. Thus, by Rolle’s Theorem , there
exists a 6 e (0 , 1) such that
0 = h! ( 0 ) = [ f x ( a + 6 { b - a ) ) ( b - a ) ] w - (1.57)
_ [/ ( & ) - f ( a )\ w gx ( a 9 { b - a ) ) • ( b - a ) .
+
9( b) ~ g( a )
Thus, if we let £ = a + 9 { b — a ) e ( a , b ) , then (1.46) follows.
xWe assume that Rolle’s Theorem is known to the readers.
—
Note that by definition , we see that if / : G > R is Frechet differentiable
at X Q G G, then near xo , / ( X ) can be approximated by a linear function:
/ ( x ) = f (x0 ) + / (x0 ) • ( x - x0 ) + R ( x; x0 ) = L (x ; x0 ) + i? (x ; x0 )
X
where
L ( x; x 0 ) = / ( x0 ) + f x ( xo ) • ( x - x0 ) ,
k
R ( x\ X o ) = f { x ) - f ( xo ) - f x { xo ) • ( x - Xo )
with
lim l -R ( o;; ar0 ) | =

—
x cco || > 0 ||X — XQ ||
—
Whereas, by the Lagrange mean value theorem , we have
/ (x ) = / ( xo ) + f x ( x0 + 6 ( x - x0 )) • (x - Xo ) (1.58)
in which the remainder term does not appear , and as a trade-off , f x ( xo ) is
replaced by f x ( x0 + 0 (x - x 0 )) .
As an application of the above Theorem 1.13, we present the following
result.
.
Proposition 1.14 L e t f : G > — b e Frechet differentiable. Then
f x ( x ) = 0, x eG (1.59)
if a n d only if / ( x ) is a constant .
Proof . Sufficiency is clear . We now prove the necessity. Since G is a

domain (connected and open in R^ ) , it must be pathwise connected (see
Problem 6 in Chapter 2, Section 1) , for any two points xo , x G G, we can
find a finite sequence xi , • • • , x & G G so that Xk = x and the segments
[ x u X i+1] C G , 0 < i ^ k - 1.
By (1.50 ) , we obtain
||/ ( xi ) - / ( x0 ) || \\ f x (0 ( x l - ®o ) || = 0.
Use the same argument recursively, we obtain that
f ( x ) = f ( x0 ) VXGG
proving our claim.
Differentiation 73
The following is another very interesting application of the Cauchy mean

value theorem.
Theorem 1.15. ( l’Hopital’s Rule ) Let f , g : [ a , b\ -» R be continuous.
Let XQ £ [a , b ] such that
f { xo ) = g ( x o ) = 0. (1.60 )
-
(i ) Suppose / ( ) and g ( ) have derivatives f ' ( xo ) and g' ( xo ) at XQ such
that g' ( xo ) 0. Then there exists a 6 > 0 such that
^ g{ x ) ± o, 0 < \x — xo\ < S , (1.61)
and
lim / 0*0 = f ( xo ) ' (1.62 )
g{ x ) g' ( x 0 )
(ii ) Suppose XQ = a and there exists a 5 > 0 such that f ' ( x ) and g' ( x )
exist for all x £ ( xo , xo + 5) and
Then
g\x )
^ o, x G ( ZQJZO + 5) . (1.63)
lim / 0*0 lim

fix )
(1.64 )
x -> x , £ s0*0 x- * x,i 9' { x )
provided the limit on right-hand side exists.
(iii) Suppose XQ = b and there exists a S > 0 such that /'( x ) and g' ( x )
exist for all x £ ( XQ (5, XQ ) and
g' { x ) + o, x G (x 0 — S, )- XQ (1.65)
Then
f (x) fix)
lim lim ( 1.66 )
X -+ X O 9 ( x ) X
—
^r X o 9' { x ) ’
( iv ) Suppose xo £ ( a, b ) and there exists a 5 > 0 such that /'(x ) and
—
g' ( x ) exist for all x £ ( xo <5, xo + 5) \ {xo} and
g' ( x ) ± o, x G ( x0 - <5, x0 + S )\{x0}. (1.67)
Then
f (x) f (x)
lim ( 1.68 )
x ^Txo g ( x ) g' { x )
P r o o f , (i) Without loss of generality, we assume that g' ( xo ) > 0. For

£ = Y9-- > 0, there exists a 6 > 0 such that
g ( x ) - g ( x0 )
X — Xo
~
9 (*o ) < —g 0 < |x xo| < 5, —
which leads to ( note g ( xo ) = 0)
g( x ) g x0 )
\
X — Xo >
^
2
0 < \x — XQ \ < 5.
Hence
g( x ) o, 0 < \x — XQ| < 6.
Then
Xo )
/ 0*0 / 0*0 - f ( xo ) X
— Xp /'(*o )
x
^ xo g [ x ) —
g ( x ) g ( xo ) X
lim
— ¥ X o g( x )
X
- g( x 0 )
XQ
g!( x 0 )
'
( ii) First of all, by (1.63) , we claim that (1.61) holds for the given S > 0.
In fact , if there exists an x G ( xo , X o + 5) such that g ( x ) = 0, then applying
Rolle’s Theorem on [xo , x] , we have some y G ( xo , x ) such that g' ( y ) = 0
which contradicts the assumption. Hence, our claim holds. Now , by Cauchy
Mean Value Theorem , for any x G ( xo , xo + $) , there exists £ G ( xo , x ) such
that
/ 0*0 = / 0*0 - / (so ) f (g) '
5( x ) g ( x ) - g ( x0 ) = g' ( £ )
Hence
lim / (*) lim m lim /'(*)

X - j s0*0
+X , a; — yx+
o s'(0 X - J 9' 0*0
+X ,
provided the right-hand side exists.

The proofs of (iii) and ( iv ) are similar.
The above result is very useful for finding the limit of the expression
as x —> xo when lim f i x ) = lim o ( x ) = 0.
a:->- xo .
Exercises
1. Let Rnxm be the set of all ( n x m ) real matrices. Let A G Rnxn and
f i x ) = xT .Ax , for all x G Rn . Show that / ( •) is Frechet differentiable on
Rn . Find f x{ x ) .
Differentiation 75
771 J.
2. Let / ( x ) = ||x UP = ( X> r)
/c = l
for any x G Rm , with p G [1, oo ) .
Then / ( •) is Frechet differentiable at any x G Rm \{0}. Discuss the situation

at x = 0.
sinx
3. Suppose it is known that lim
x = 1. Derive the following formulas
by definition:
(sinx )' = cosx , ( cosx )' = - sinx.
4. Let
1
/ (*) =
x|asin
\ x\ P
x
^ O,
Find conditions on a , 0 G R so that

0 x — 0.
/( ) • has derivative at x = 0.
—
5. Let / : R » R be differentiable, ( i) If / ( x ) is an even function , then
/' ) is an odd function; (ii) If / ( x ) is an odd function, then /'(x ) is an
( x
even function.
6. Let f , g : R —> K. Suppose / (x) is differentiable at xo( and <?(x ) is
not differentiable at Xn . What about the differentiability of f x ) + q ( x ) and
f { x ) g ( x ) at x0 ? Why?
7. Let
1 1
/ (a:) = 22n
x= —
2n
n 1
0 otherwise.
Prove or disprove the differentiability of / ( x ) at x = 0. Is / ( x ) continuous
at x = 0? Why?
8. Let
1
/ (x) = <
x 2 sin
•
—x x / 0,
0 x = 0.
Show that /' (x ) exists but not continuous.
9*. Let
xpyq
{ x, y ) ± ( 0 , 0)
f ( x , y ) = |x|" + \ y \ m
0 { x , y ) = ( 0, 0 ) ,
where m , n , p , q > 0. Find conditions on m , n , p , q such that f { x , y ) is

Gateaux differentiable, has a gradient , and Frechet differentiable, respec-
tively.
10. Prove Theorem 1.15 parts ( iii) and ( iv ) .
—
11. Let /, g : R > R be continuous and differentiable with
lim f ( x ) = lim g ( x ) = 0.
—
x ± oc —
x too
Let
/'(*) = L
9' i.x ) ± 0, xGM
x
lim
—
^ oo 9' { x )
for some L G R. Show that
lim / 0*0 = L.
—
x > oo 9( x )
2 Frechet Differentiability
In the previous section , we have seen that , in general, existence of a gradient

(which implies the existence of all first order partial derivatives ) does not
implies the Frechet differentiability. In this section , we would like to study
the following problem: If / ( •) admits all the first order partial derivatives
at xo , under what additional condition (s) , will / ( •) be Frechet differentiable
at xo ? We have the following result .
Theorem 2.1 . Let f : G —> R admit all the first order partial deriva-
tives fxi (1 i m ) in a neighborhood of xo G G , and all these partial
derivatives are continuous at xo - Then / ( •) is Frechet differentiable at x$ .
Proof . Let us first look at the case m = 2. We write x = ( £, T? ) t , and

T
XQ = ( £O 5 ?7O ) - Since £ i-» / ( £ , 770 + 77 ) admits a derivative in a neighborhood
of £0 ( for 1771 small enough ) , by Lagrange Mean Value Theorem , there exists
a 9 G ( 0 , 1) such that , for |£| small,
f ( o+& ,no+v ) - f ( Zo ,‘no +v ) = /« (& + &;, % + »?) £
On the other hand , since 77 H-» / (£0 * 77 ) admits a derivative at 770 , one has
f ( Zo , rio + v ) - / (6o , 77o ) = /*» (£<) , f 7o ) »7 + R( v,Vo ) ,
with
lim ! # (?/; >7o) I =0

M “>0 M
Differentiation 77
Consequently,
+ x ) - f ( xo ) = fit0 + t , Vo + v ) ~ fito ,Vo )

f { x0
= fdto + &t , VO + v )t + fdto , Vo )V + Riv \ Vo )
= ifdxo ) , fdxo ) ) - x + R( x , x0 , d ) ,
with
R{ x , x0 , 6 ) = [ fdto + Ot ,Vo + v ) - fdto ,Vo ) ] t + R{ v ; vo ) -

Clearly, ( note x = (t , v )T and the continuity of fd‘ ) )
lim
|| R ( x , xo , 0) || l -R (^?; »7o ) ||
IMHO 11*11 ^ IlSo ||x||
+ lim \\ fdto + ^t , vo + v ) - fdto , vo ) \\ \\ t \\ ...
0
IMHO I*
This proves the Frechet differentiability of / ( •) at #o = (£O 577O ) T -
In the above, we have used the continuity of the partial derivative
A (& *l ) near ( £oiVo ) , and the Frechet differentiability of 77 » >• / (£o , v ) at —
Vo -
Next , we use induction. Suppose our conclusion holds for the case of
G C Rm- 1 . Then we write x = (£ , 77 ) and xo = ( o , Vo ) with £ , £o R and
_
77 , 770 G Rm . Now , / (£, 77) 1S continuous near (£o > f 7o ) > and by induction
1
^
hypothesis , 77 h-* / (£o > f ? ) is Frechet differentiable. Hence, copying the above
proof line-by-line, we can get the proof of our conclusion.
A careful observation of the above proof suggests that we actually have

a little more general result .
Theorem 2.2. Let
r = {ii ,
A = Ol , '
— , 4} c {1, 2, • , m } ,
••
• • • , j m-k } = {1, 2 , • • , m} \ r .
•
( 2.1 )
Write x = (£, rj) where £ = ( xn , • • • , xlk ) T and r] = ( x-?1 , • • • , xim k ) T . Let ~
f : G —> R admit all the first order partial derivatives at XQ = ( £o ,Vo ) ;

For each i G T , fxi ( x ) exists in a neighborhood of xo and continuous; and
—
the map rj 1 > f ( £o , v ) JS Frechet differentiable at 770 - Then / ( • ) is Frechet
differentiable at XQ .
Proof . Our condition says that /$ (£, rj) is continuous near (£0 , 770 ) , and
r) i-» / ( £0 , 77) is Frechet differentiable. Hence, the above proof can be ap-
plied.
We point out that in the above Theorem 2.2, if we only assume that
£ • > / ( £, 770 ) and 77 1-» / ( £0 , 77 ) are Frechet differentiable at £0 and 770 ,
-
separately, then we might not have the Frechet differentiability of / ( •) at
XQ = ( £0 , 770 ) . In fact , Examples 1.2 and 1.5 all gives
/* (0, 0) = /v (0, 0 ) = 0.
Thus , x n-» / ( # , 0 ) and differentiable at 0. But
y i-> / ( 0 , 7/ ) are all Frechet
the corresponding functions f { x , y ) are not Frechet differentiable at (0, 0) .
However , we should point out that the conditions imposed in the above
theorems are only sufficient for the Frechet differentiability. The following
example illustrate this point .

1
( x 2 + y 2 ) sin fay ) ^ 0,
7
ffay ) = x2 + y2
0 f a y ) = 0.
Then we have
1 2x 1
2 x sin cos 2 ( x, y ) ± 0,
fx( x, y) = x2 + y2 x 2 A- y2 x 4- y 2 ’
0 ( x , y ) = 0,
and
fv( x, y) = <
sin —x +1 y
g2 2 — ,+
X2
7
2y
y o2
COS
x 2
1
+ y2
~ ( x, y )
^ 0,
0 ( x, y ) = .
0
Clearly, both f x and f y are discontinuous at ( 0, 0 ) . However ,
I f ( x , y ) - / ( 0, 0 ) - ( 0, 0 ) ( x, y )|
.
•
lim
|| ( x ,y ) ||-> 0 ii (* i/) ii
1
lim x / +
x2 y2 sin =0
x 2 + y2 >0— x2 + y2
which means that /( )• is Frechet differentiable at ( 0, 0 ) with V / ( 0, 0 ) =
(0 , 0 ) .
Differentiation 79
Despite of the above example, Theorem 2.1 is still one of the most com-
monly used sufficient conditions for the considered function to be Frechet
differentiable. We now introduce the following definition.
—
Definition 2.4. Let G C Rm . Function / : G > R is said to be C 1 (on
G ) if / ( •) is continuous on G, admits all the first order partial derivatives
on G, and all of these partial derivatives are continuous on G and can be
continuously extended to dG.
By Theorem 2.1, we see that if / : G -» R is G1, then it is Frechet

-
differentiable on G, and the Frechet derivative /z ( ) , which admits a rep-
resentation (1.19 ) , in terms of partial derivatives, is continuous. From now
on , we will only consider Frechet differentiability. Thus, for simplicity, we
will omit the word “Frechet ” in Frechet differentiability. Also , we will not
distinguish f x ( x ) and ( V / )( x ) hereafter when the function is differentiable.
Exercises
1. Let || • = || • ||2 be the ^2-norm in Rm . Define

\ x T y\ p x 0, or y 0,
f { x> v ) =
0
^
x = y = 0,
^
with x , y G Rm . Discuss the differentiability of / ( x, y ) at ( x , ?/ ) = (0, 0 ) for
different p > 0.
2. Let / : Rm —> R be Lipschitz continuous, i.e., the following holds:
<
\ f ( x ) - f ( y ) \ £||z - i/|| Vx , y G R m
for some L > 0. Suppose / ( x0 ) = 0. Then x / (x ) 2 is Frechet differen-
tiable at XQ .
3. Let / : Rm —> R be convex, i.e.
/ ( Ax + (1 - A ) y ) A f ( x ) + (1 - A ) f ( y ) Vx , y e Km , A S (0, 1).
Suppose / ( ) is also differentiable. Then
•
f ( x ) + f x ( x )( y - x ) f ( y ) Vx , y Km .
4. Let / : R 2 —> R be defined by the following:
xsin y
f ( x, y ) = y
y ^ 0,
7
x y = 0.
Prove or disprove that / ( x , y ) is Frechet differentiable at ( 0 , 0 ).
3 Inverse Function and Implicit Function Theorems
Let A e Rmxm and b e Rm . Consider

f ( x ) = A x 4- 5, x G Rm .
If A~ l exists, then / ( •) admits an inverse function
= A 1{ y - b) ,
~
y e Rm
with
r ) (i/ ) = ^-
( 1
y
1
= [/x (r1(i/ ))]- 1 . ( 3.1)
We now would like to extend the above to nonlinear case. The following is
called the Inverse Function Theorem.
Theorem 3.1. Let f : G Rm be C 1 and xo G G such that det [ /x (#o ) ]

0 . Then there exist open sets U C G containing xo and V C Rm containing ^
f ( xo ) such that f ( U ) = V , and there is an inverse function /-1 : V —> U
of / ( • ) , which is C 1 in V with
( f-
_1 _
X i v ) = [/x ( / ( y ))] 1, Vy e V. ( 3.2)
Proof . Replacing / ( •) by [ f x ( xo ) ] 1 / ( ) ? necessary, we may assume
*
^
that f x ( xo ) = I ( the identity matrix of order m ). Now, by (1.50 ) , we have
Ik - *11 - 11 / 0*0 - / (*) I < I / (*) ~ x~ L / (* ) - x ] \\

IIfx( x + 0( x - x ) ) - I x — X
for some 9 G (0, 1) . Consequently,
{1 - I f x ( x + 9 ( x - x ) ) - I \\ } ||x - z|| ||/ (s) - / ( x ) ||.

-
Since f x ( ) is continuous , and f x ( xo ) = / , we can find a <5 > 0 such that
1
l /« (*WII < 2
\/ x e B( x o , 6 ) .
Hence, for some 6' G ( 0, 1)
1
- || x - x|| I f { x ) - / ( x ) || \ \ f x ( x + 0' ( x - x ) ) \\ \\ x - x \\
( 3.3)
— X
Vx, x G B( XQ , 6 ) .
Differentiation 81
The above implies that / ( •) is one-to-one on B ( X Q , 5 ) . Further , since

d B ( xo , 5 ) is compact , we have from ( 3.3) that
S
a ~
min
x £ d B ( x o ,6 )
|| / (x ) - / ( xo ) || l-2 mm
x d B ( x o ,6 )
\\ x - Xo - > 0. ( 3.4 )
Let
( f ( x0 ) ) .
{ y e R m \ \\ y - ttx 0)
Now , for any fixed y G V , we want to find a unique x G B ( xo , S ) such that

\\ <
f ( x ) = y . To this end , we consider continuous function h( x ) = \\ f ( x ) y \\ 2

on the closed ball B ( xo , S ) . It attains its minimum at some x G B ( xo , S ) .
^ }= B « ( 3.5)
—
We claim that x d B ( xo , S ) . In fact , if x G d B ( xo , (5) , then by ( 3.4 )
I> ||/ ( zo ) - y \\ = V h( xo ) > y

/ h( x )
= I I f i x ) - y \\
which is a contradiction. This means that x
Hence, by Fermat ’s theorem ,
I / (*) - f ( x0 )|| - ||y - f ( x0 )|| > cr -
G ^
B ( X Q , 5 ) ( an open ball ).
0= hx ( x ) = 2 f x ( x ) ( f ( x ) - y ) .
Consequently, by the invertibility of f x ( x ) (since x G B ( xo , 5 ) ) , we obtain
that
f {x) = y
It follows from ( 3.3) that such an x must be uniquely determined by y .
_
Hence, y X defines the inverse / 1 ( ) of / ( •) . If we let U = f ~ 1 (V ) , then
*
by the continuity of / ( • ) , U must be open , and f ( U ) = V , f : U ) V is —

one-to-one and onto.
Next , we prove the differentiability of / -1 ( - ) . To this end , we take any
x G U . By the differentiability of / ( •) , we have
f ( x ) = f ( x ) + f x { x ) ( x - x ) + R( x , x )
with
lim ll -RfoaQll = Q
||cc -®||-> 0 \\ X — X \ \
Denote y = f ( x ) and y
above becomes
— f ( x ) , with x
^x( so that y
^ y ) . Then the
y - y - f x( f 1
( y ) )[ f 1
(y) - f 1(
, y ) } = R( f 1
( y ), f 1
( y)) -
By applying — [ f x ( f 1 ( 2/ ) ) ] 1 on the both sides of the above, one has

f ~\y ) - r\y ) - [ f x ( f { y ) ) ] { y - y ) ~l ~l
= - lfx { r\y ) ) ] lRU \y ) , f Hv ) ) = R( y v ) -

~ ~ ~
>
Note that (taking into account ( 3.3) )
|| [/x ( / - 1 ( y )) ] - 1 i?( / - 1 ( y ) , / - 1 ( y ))||

lim
v -> yII 2/ — 2/ I y -+v \\ y - y \\
|fl( ) X — X
wifAr' mr 1 xlimy x l *> * l
— I * *11 l / *) - / (*) l
-
<
(
, )
i f z V - Hv ) ) ] - 1 lim iw* * in = 0.
x -+ x x - x \\
-
This means that /-1 ( ) is differentiable and ( 2.1) holds, which further im-
-
plies that /-1 ( ) is C 1 .
The following is called the Implicit Function Theorem.
Theorem 3.2 . Let G C Rk x Re , and f : G »

£ xe
be C 1 such that —
f ( x o i U o ) = 0 a n d f y ( x0 , y o ) G R is invertible. Then there exists a 5 > 0
and a unique map g : B ( x0 , S ) > Re such that —
9 ( x0 ) = 2/o , ( 3.6 )
and
f ( x , g ( x ) ) = 0, Wx G B( XQ ,6 ). ( 3.7 )
Moreover ,
9x ( x ) = - f y ( x , g ( x ) )
1 fx
( x,g{ x ) ), x G B ( xo , 5 ) . ( 3.8)
Proof . Let
F{ x, y ) = ( x, f { x, y ) ) ( x, y) G G.
Then F : G > Rfe — +£ is differentiable, such that *
I
F( x , y ) ( X 1 y ) — 0
fx( x, y ) fy( x, y )
( x, y) G.
Thus,
det [F( X 3/ ) ( XQ , 2/O )] = d e t [ f y ( x0 , y0 ) }
Also
)
^ 0.
( u0 , v0 ) = F ( x0 , y o ) = ( xo , f ( x 0 , y o ) ) = (xo , 0 ). ( 3.9 )
Differentiation 83
Hence, by Theorem 3.1, there exist open sets U containing ( #o , 2/o ) and
V containing ( xo , 0) , such that F : U —> V is one-to-one and onto , and
F ~ l : V —* U is C 1 . Now , if we denote
( x , y ) = F 1 ( u , v ) = ( < p( u , v ) ,tl>( u , v ) ) , ( u, v ) e V, ( 3.10)
then
( u,v ) = F o F ( u , v ) = F ( ip ( u , v ) , { u , v ) )
1
^
= Mu , v ) , f ( < p ( u , v ) , ip ( u , v ) ) ) ( u , v ) e V.
This means that
x = ip ( u , v ) = u , V (u, v ) GV , ( 3.11)
which implies that
V = f ( u , i/>( u , v ) ) , V ( u, v ) e V. ( 3.12 )
Now , by letting
g ( x ) = 'ip ( x , 0 ) Vx G B ( X Q , 8 ) ( 3.13)
for some S > 1
0 , we have that g ( - ) is C , and
/ (x , g ( x ) ) = / (x , ( x , 0 )) =
f ( u (u 0 ) ) = ( 3.14)
which gives
^
( 3.7) . Also, from ( 3.9) ( 3.10) , (
-
^) , and (
3.11
, , 0,
3.13) , we have
Vo = vo ) = 0) = g { x0 ) ,
which gives ( 3.6 ) . Finally, by chain rule,
0 = f x ( x , g ( x ) ) + f y( x , g{ x ) )g x( x ).
Hence, ( 3.8) follows.
Let us present an example.
Example 3.3. Suppose G C Rm and / : G —> R. Write x = ( £ , xm ) G Rm .
Let XQ = (ZOIXQ1) £ G such that
/xm ( xo ) 7^
and
/ (x0 ) = / ( £o , ô" ) = c.
Then there exists a <5 > 0 and a unique g : H m — (Co £) £ Rm
1
?
1 -* R such
that
_
g ( Zo ) = aft , / ( £ , <? ( £ )) = 0 , Bm 1 ( 0 , S ) .
G
The function xm = <? (£) is called the level set surface whose graph coincides ^
with the following:
/ X (c) = ( / = c) = { x e G | / (x ) = c}.
Next , let G C Rm be a domain , and / : G -» Rm be C l . By Proposition
1.14, we know that
f x ( x ) = 0, WxeG ( 3.15)
if and only if / ( •) is a constant vector. Now , we would like to ask what will
dx ---
d ( x1 ,
-
happen if instead of ( 3.15) , we have the following condition:
s (x ) 9( /1, . . . , / ) ( x ) 0, Vx G G.
det [ f x ( x ) ] = /
, xm ) =
The above implies that for each x G G , there exists a non- zero vector
( 3.16)
£ ( x ) G Rm such that
f x ( x )Tt ( x ) = 0. ( 3.17)
It is natural to imagine that the components of / ( •) should have some
special relations. To get some feeling , let us look at the following situation.
Suppose there exists a C 1 function 6 : Rm —» R such that
f ( / (*)) = o , xGG ( 3.18)
and
@y( y ) 7 ^ yG / (G)
then , by the chain rule, we have
f X ( x )T 6 y ( f ( x ) ) T =0 xG, G
which is of the form ( 3.17 ) with £ (:r ) = 0 y ( f ( x ) ) T . In particular , if

%) = T y , Vye Mm
* for some A G Rm \ {0},
then ( 3.18) means that / * , • • • , f m are linearly dependent , and ( 3.17 ) holds
with £ ( x ) = A . Inspired by this, when ( 3.18) holds for some C1 map # ( •) ,
we say that the components /1 ( ) , • • • , / m ( - ) are functionally dependent on
*
G. Hence, the above means that when , / m ( ) are functionally

*
dependent , then ( 3.16) holds. The following gives a kind of converse of

such a statement.
Theorem 3.4. L e t G C Rm b e a bounded domain and let f : G -A R m be -
C1 on G . Suppose ( 3.16 ) holds in a neighborhood B ( xo , 5 ) C G o f some
point X Q G G. Moreover ,
r a n k [ f x ( x0 )\ = m - 1,
meaning that if we write the ( m x m ) matrix f x ( x ) = (ai , • • • , am ) , then
the set {ai , - - - , am} is linearly dependent , and there exists at least one
subset of it containing ( m — 1) vectors that is linearly independent . Then
there exists a neighborhood of xo and a C 1 function 6 : Rm —> R such that •
( 3.18 ) holds near XQ .

Differentiation 85
Proof . Without loss of generality, we assume that

3(.fV - - , /"1- _
1)
( x ) / 0, x e B ( xo , 5 ) .
d ( x , • • • , xm )
1 1
Denote
£ = (x1, - - - xm 1) T
"
Co = - *)
¥? ( x ) = </> (£, xm ) = ,/ — (x) )
m 1 T
.
_1 _1 - _ 1 be
Let F : Rm xRx Rm > Rm defined by
f & xm u ) = tp ( £ , xm ) - u , with u0 = ¥> ( x0 ) = </> (£() , x£* ).
Clearly,
F ( £o , x
, u0 ) = 0.
Consider the equation F ( £ , xm , u ) = 0 (for £) . Since
Fsfa , xV , uo ) = ve ( to > * o ) = <Pt ( xo )
is invertible, by the Implicit Function Theorem , we have a locally defined Cl
_
—
function g : O » Rm 1 with O C Rm being a neighborhood of ( XQ1 , V? ( XO )) ,
such that
F( fl ( xm u ) , xm , u)
• = ( p( g ( xm , u ) , xm ) — u = 0, V ( xm , u ) e O . ( 3.19)
On the other hand , for any fixed xm near XQ1, the map £ H-> <£> (£ , xm )
satisfies the condition of the inverse function theorem near £o - Therefore,
we see that the above obtained function u - i > g ( xm , u ) is the inverse of
C• > £< ( , xm ) . Namely,

“
^
5( xm , V3(e, xm ) ) = C , V(£ , xm ) near (£0 , x ) x0 .
Now , differentiating ( 3.19) with respect to x m ^=
we get (suppressing ( xm , u ) )
( 3.20 )
0 = ( p$ gxm + v(xm , u ) G e> ,

which leads to
9 xm
Next , we differentiate the map
— — (Fc ) xm 1(
Pxm
M -
V( xm , u ) G O,
fm ( g ( xm , u ) , xrn ) to get:
[ fm ( 9 ( xm
_
= f% n ~ / (< ) V* det [/ ] = 0.
This means that h{ u ) =/ m( ^ "> x
# ( xm , u ) , xm ) is independent of xm . Then by
defining
0 ( f ) = h( f \ • • • , / m — )-/
1 m ,
and taking into account ( 3.20 ) , we obtain that in a neighborhood of XQ
9 ( f ( x ) ) = h( < p ( x ) ) - f m ( x ) = r {9{ x M x ) ),x ) - r ( x )

= / m (f , *m ) - / m (*) = 0,
proving our conclusion.
We now would like to look at some geometric feature of the gradient for
scalar-valued functions. To this end , we first introduce more notions.
.
Definition 3.5 Let / : G > R be given. — •
( i ) Let xo G / -1 ( c) and u G Rm be a unit vector. Then the straight line

x = XQ 4- t u ( t G M ) is a tangent line of the level set /-1 ( c) at XQ if
p( x p + h u , f ^ (c))
lim =0 ( 3.21)
h — () h
where
p{ x , f x (c)) = y e finf-' ic ) ||i/ - a|| ( 3.22 )
is the distance from x to / -1 ( c) . In this case, u is called a tangent vector

_
of / 1 ( c) at xo .
( ii) Let Xo G /-1 ( c) . A plane, denoted by TXo is called the tangent plane
_
of / 1 ( c) at xo if any straight line on this plane passing through xo is a
tangent line of f ~ 1 { c ) at xo .
According to the above definition , if the level set f ~ 1 ( c ) is given by a
circular cone with the vertex xo , then / -1 ( c) does not have tangent lines
at xo , since ( 3.21) cannot hold for any direction. Hence, it does not have a
tangent plane either. Note that the above definition can be extended to a
more general case by replacing /-1 ( c) by any surface S .
The following result is concerned with tangent vectors.
Proposition 3.6 . Let f : G -» R and g : [0 , T ) -» G be C l such that
f ( g ( t ) ) = c, t G [0, T]. ( 3.23)
/ - 1 ( c) at g { t ) G / -1 (c).
Then g ( t ) =
^ g ( t ) is a tangent vector of
Proof . Let to [0 , T] be fixed and xo = g ( to ) G / -1 ( c) . We know that

x = xo + t g { to ) (t G R ) is the tangent line of the curve g ( ) at g ( to ) = xo -
Hence, one has
-
lim \\ g ( t0 + h ) - x0 - hg ( t0 ) \\ = Q
h-¥ o h
Differentiation 87
By ( 3.22 ) , we obtain
p( x0 + hg ( to ) , /
_1
lim
( c))
lim \\ g ( t0 + h ) - x0 - hg ( t0 ) \\ = Q
h Q
^ - h h-* 0 h
This proves our claim .
Next result gives the existence of tangent planes , for which we need to
use the implicit function theorem.
—
Theorem 3.7. Let f : G > R be C 1 , and let xo E / -1 ( c) such that
V / ( xo ) 7 0. Then the plane passing through xo with normal vector V / ( xo )
^ _
is the tangent plane of / 1(c) at XQ . The equation for this tangent plane
is given by
V / ( xo ) ( x - x0 ) = 0. ( 3.24 )
Proof . It suffices to show that any vector u _L V / ( xo ) T must be a
tangent vector of /-1 ( c) at xo - We now prove this. Consider the following
equation (for v £ R ) :
F (t , v) T
= f (x0 + t u + vV / ( x0 ) ) - c = 0.
Clearly,
F( 0, 0) = / ( x 0 ) - c = 0,
and
(t , v ) = ( 0, 0 )
= V / (ar0 ) V / (a;o ) T = ||V / (cc0 ) || 2 > 0.
Hence, by implicit function theorem ( Theorem 3.2 ) , one can find a function
- —
v ( ) defined on [ 5, S ] for some 5 > 0 , such that v (0 ) = 0 and
F ( t , v(t ) ) = / (x0 4- t u 4- v (t ) V / (x0 ) T ) = 0 t E [-5, 5].
(
Differentiating the above with respect to £ , and evaluating at t = 0, we

have ( noting u _L V / ( XQ ) t )
*
0
- -
V / (*o ) u + l | V / ( xo )||2 u ( 0 ) ||V / ( xo ) ||2 u (0 ).
Thus, it is necessary that i;( 0 ) = 0. Then with g ( t ) = xo+tu -\- v ( t ) V / ( xo ) T
one has
f ( g{t ) ) = c, t [ -5, 5]
< ( 3.25)
5( 0) = x0 , 5( 0) = u.
Hence, by Proposition 3.6, we see that u is a tangent vector of / ~ 1 (e) at
-
X Q This proves our theorem.
The above theorem is very useful in finding tangent planes for the level
surface of given function / ( •) .
To conclude this section , we introduce the following notions.
-
Definition 3.8. Let g : G -A R be C 1 and S C Rm be a surface. Let XQ G S
and N ( xo ) be a unit normal vector of S at xo , by definition , N ( XQ ) is a
normal vector of the tangent plane of S at xo. The directional derivative of
-
g ( ) at point xo in the normal direction N ( XQ ) is called the normal derivative
-
of g ( ) on the surface S ( with respect to N ( xo )) , denoted by
According to the above definition , it is clear that
d9
dn 0 =
( x ) Vg ( x ) N ( x ) .
0 0 ( 3.26)
Further , if the surface S is closed and given by /-1 (c) for some / : G —> R
and c G R ( for example, a sphere ) , then the unit normal vector N ( xo ) is
usually taken to be the one pointing outside of the enclosed domain , which
is called the outward unit normal vector. Thus, if the “inside” domain is
( )
characterized by ( / < c) , then l|VV//s( oo ) || is the outward unit normal vector
*
and
V (so ) T
= Wg { xo ) |V/ ( ) ||
dg
(xo ) ( 3.27)
dn l / *o '
Exercises
—
1. Let / : Rm > Rm be C 1 such that for each x G Rm , /^ ( x ) G Rmxm
is an invertible matrix. Then for any open set G C Rm , f ( G ) is open.
2. Let / , g , h : R —» R be continuously differentiable such that / ( •) is
•
non-decreasing, <? ( x ) > 0 for all x G [0, oo ). Moreover ,

/ (1) + p (1) + h( 1) = 0.
Then there exists a (5 > 0 such that for some continuously differentiable
—
function : (1 <5, 1 + <S ) — R , the following holds:
x / Mx )) + g ( x ) i p ( x ) + h( x ) = 0, x G (1 - (S, 1 + (S ) .
3. Let ( f —
: Rm > R be a C1 convex function such that ipx ( x )
^ 0 for all
x G Rm . Let
|
G = x G Rm | <p( x ) |
ô 7^ 0.
Let x G dG. Then there exists a k = 1, 2, • • • , m, a 5 > 0, and a C1 function
0 : T C R 771 1 -> R , with
"
T — {( ^1 > *
* * *
—
> •I' k 1 ) J ’
* *
?
^m ) |
*
(Xl , j
^ /e — 15 ^ /c + l
* 5 ' > ) G
^ (0) + xj ^
Differentiation 89
such that
* * *
5 Xfc — 1 5 @ { X \ 5 * * *
5
^ /c — 1 ^ /e+1
* j 5
* * *
5 £771 ) 5
*
^- /e+ l 5 ? *£ m) —O 5
^ (^ — • 15 * * *
5 *£ f c 1 5 *£ /c + l * * *
5 E ?7l )
*
^^
Give a geometric interpretation of the above result .
4 Higher Order Derivatives
—
For a differentiable function / : G > R, fx : G —> Rm is a vector-valued
function. We may further consider the differentiability of fx ( ) - This leads '
to the second order derivatives for the original function / ( •) . Since fx ( - )

is not just a vector-valued function , it is the gradient of a function , there-
fore, we expect to have some additional feature of the ( partial) deriva-
tive /gradient of /x ( ) - A natural question is: For i j , is it true that
*
fxix^ ( ^0
^
* f x x i ( ^)
^ *
^ ( 4.1)
Unfortunately, the above might fail in general. Here is an example.

x2 - y2\
xy ( x, y ) ± 0,
f ( x, y) = x2 + y2 / ( 4.2)
0 ( x , y ) = 0.
A straightforward calculation shows that:
/* ( 0, 2/ ) = -2/ , fy { x , 0 ) = x . ( 4.3)
Consequently,
/*y ( 0, 0 ) = -l = /y*(0, 0).
Thus , for this function , ( 4.1) fails.
^ l ( 4.4 )
The following gives some sufficient conditions under which ( 4.1) holds.
—
Theorem 4.2 . ( Clairaut ) Let f : G > R such that all first order partial
derivatives fxi ( •) exist in a neighborhood of xo E G. Moreover , all of these
partial derivatives are differentiable at XQ . Then ( 4.1) holds.
Proof . It suffices to prove the case m

that we are interested in. Define
— 2. Let (a, b ) E G be the point
i p{ x )= f ( x , b + h ) - f ( x , b) , ( 4.5)
V’( y ) = f ( a + h, y ) - f ( a , y ) .
Clearly, </? ( •) and ( ) are differentiable. Hence, by Lagrange Mean Value
^ »
Theorem (see Theorem 1.13, special case ( ii)) , there exist 0i , 02 £ ( 0 , 1)

such that
( p ( a+ h ) - ( p ( a ) = p' ( a +0ih ) h = [ fx ( a+6\ h , b+ h ) — fx ( a+6 ih , b ) } h , ( 4.6)
(
and
ip ( b 4- h ) ~ { b ) = 'ip' ( b + 02 h ) h = [ fy ( a + h , b + 02 h ) - fy ( b + h , b ) ] h. ( 4.7)
^
Now , by the differentiability of fx at ( a , b ) , we further have
- tp{ a ) = { [ fx ( a , b ) + fxx { a , b )61 h + fxy ( a , b ) h + £ i \ h\

tp { a + h )
[ fx ( a , b ) + fxx ( a , b )91 h + s'1 \ h|

~
\ /i ( 4.8)
= fxy ( a , b ) h2 + (ei + e[ ) \ h\ h
where £ i , ei ^ 0 as \ h \ -A- 0. Likewise, by the differentiability of fy at (a , 6) ,
we have
tp ( b + h ) ~ i/ j ( b ) = fyx ( a , b ) + ( e2 + e'2 ) \ h\ h, ( 4.9 )

with
^
£2 ) 2 —
0 as \ h\ > 0. On the other hand ,
tp ( a + h ) - = f ( a + h , b + h ) - f ( a + h , b ) - f ( a , b + h ) + f ( a , b)
= i/ j ( b + h ) - ip( b ) .
( 4.10)
Consequently,
fxy ( a , b ) - fyx ( a , b ) = ( e2 + e'2 - £ i - ei )

Hence, our conclusion follows.
Combining the above with Theorem 4.1, we obtain the following:
^ -> 0 . ( 4.11)
Corollary 4.3. Suppose f : G -* R admits all the first and second or-
der partial derivatives in a neighborhood of xo G G and all of them are
continuous at XQ . Then ( 4.1) holds at x = XQ .
.
Definition 4.4 Let G C Rm . Function / : G » R is said to be C 2 on —
G if / ( •) is continuous on G, admits all the first and second order partial
derivatives and all of these derivatives are continuous on G and can be
continuously extended to dG .
Differentiation 91
If / ( •) is C2 , we denote
rc1 x1 ( ^ ) /X 1 X 2 ( X ) * * *
/x ; 1 xrn 0r ) \
f x 2 X 1 ( X ) f :X 2 X 2 ( ^ )
*
* * *
/x ; 2 xrn ( x )
fxx { X ) — ( 4.12 )
\ fxrnx1 { x ) fxmx 2 { x ) * * * fxTnxrn { x ) J

which is an ( m x m ) symmetric matrix. We call the above matrix the
Hessian of function / ( •) .
We may further define higher order Frechet derivatives and higher order
partial derivatives.
Definition 4.5. Let G C Rm . Function / : G -» R is said to be Ck on G

if / ( •) is continuous on G , admits all the partial derivatives up to order k ,
and all of these partial derivatives are bounded and continuous on G, and
can be continuously extended to dG . Further , if / ( •) is Ck on G for any fc ,
we say that / ( •) is G°° on G.
Theorem 4.6 . ( Taylor ) Let f : [a , b] -» R be Gfe+1 and X Q G ( a , 6) . Then
for any x G [a , 6] , there exists a £ between XQ and x such that
f ( x ) = f { xo ) /'(*0 ) (a; - xQ ) + /"(*o ) (x - x ) 2 H

+ 1! ! 0
2
( fe +1) ( ( 4.13)
f ( k ) ( xo )
ft!
( x - x0 ) fc + /( ft 1 )0! (x - x0 ) fc+1.
+
Proo/. Fix an x ( a , 6) . Let
G
/ (*) = m + nt
)
(x - 1 ) + (x rw — t ) 2
H h
/ ( fe ) (0 ( x - t ) fc
1! 2! ft!
m = ( x - t )k+i .
Then by Cauchy Mean-Value Theorem , we have some £ between XQ and x
such that
/'(C ) = / pr ) - / (xo )
'
ff '(C) 5(*) - 5(*o )
Note that
r
f i t ) = ( t ) - /'(« ) + n t ) (x - 1) - n t ) (x - t ) H
1!
( x - t ) fc - 1 / W ( t )
1!
(s - t )*
_ 1 / (fc+1) ( t ) (x - t ) fe
( ft - 1)! ( ft - 1)! + ft!
( fe +1)
/ W { x - t )k ,
ft!
and
g' ( t ) = -( k + l ) ( x - t ) k .
Thus
/ (fc +1) (M f (0 m - f ( xo ) f ( x ) - / ( xo )
( fc + 1)! <?'(£ ) g( x ) - g ( x0 ) -sMo )
1 /'(xo ) (
— (x — Xo ) +i I-/ MO - (7(zo ) +
fc 1 1!
x - x0 )
n2*! o ) (x - x ) 0
2
H h
/ fc ) (x0 ) ( x - x0 ) fe
) •
ife!
Hence, we obtain ( 4.6) .
Now , we look at the multi-variable case.
Corollary 4.7. Let f : G -A- M be Cfe+1 in a neighborhood of xo G G.
Then
k
/ (x ) = X= ! H= [ ^
i 0 |a| i
“ ) 01
/ (xo ) (® - x0 )a
( 4.14)
^ yi? / ( xo + 0 (x — X0 ) )(x — X0 ) “
Q
+
|a|= fc + l
for some 0 G (0 , 1) , where a = (ai , c*2 , ' ‘ ? am ) £ ( N U {0}) m is called a

*
multi-index, with |a | = a\ H h am , a! = aiia ! • • • am ! and

^
n >(x) = 9“ i
- - - a“^ v (x) ? Vo = ( oi , • • • , om ) .
In particular ,
f (x) = / ( x0 ) + /x (x0 ) • ( x - Xo )
( 4.15 )
4 ( x - X0 ) T /xx ( xo + 0 ( x - x0 ))( x - Xo ).
”
Proof . Fix XQ G G and for any x near xo , define

h( t ) = f ( x 0 + t ( x - x0 ) ) t G [ — 5, 1 + 5]
for some (5 > 0. Then /i ( - ) G Ck+ . Applying the above Taylor Theorem ,
1
we have
h" ( 0 ) / *0 ( 0)
( /i + (0) (fc 1)
M1) - M0) = M (0) + 2! + • • • + * fcT + ( fc l )! ’
~~ ~
+
for some 6 G ( 0, 1). The above leads to ( 4.14) by some careful calculation.
Differentiation 93
The above result leads to a refinement of the Fermat ’s theorem.

Proposition 4.8. Let / : G —» R be C 2 in a neighborhood of xo E G .
( i) Suppose that / ( •) attains a local maximum ( resp. local minimum )
at XQ . Then
f x ( xo ) = o, ( 4.16 )
and
£,r f x x ( xo )£ o, ( resp. 0), e Mm . ( 4.17)
( ii) Suppose that ( 4.16 ) holds and
C f x x { xo )£ < 0, ( resp. > 0) , V£ e Rm \ {0}. ( 4.18 )
Then xo is a local maximum ( resp. minimum ) of / ( ) . •
P r o o f , ( i) Suppose / ( •) attains a local maximum at Xo - Then by Fer-

mat’s theorem and ( 4.15) , we have
1
f ( x ) = f ( xo ) + { x - X 0 )T f x x ( x o + 9 ( x - x0 ) )( x - X o )
~ f {xo)
for all x near xo , which implies that
(T f x x ( x 0 + 9 { x - X 0 ) ) £
0, E Mm .
^
Then ( 4.17) follows from the continuity of f x x { ) ( by sending x -» xo ). '
Similarly, we can prove the case that xo is a local minimum of / ( ). •
( ii) From ( 4.15) , we see that when ( 4.16 ) and ( 4.18 ) hold , we have
( making use of the continuity of fXx ( ) ) '
f {x) f { xo ) ( 4.19)
-
for all x near xo This means that xo is a local maximum of / ( • ). The
conclusion for a local minimum can be proved similarly.
Exercises
1. Construct a smooth function ( p : Rm —> R such that
1
< p( x ) = 0 l| x||2 > 2.
2. Let / : Rm -> R be C 2 . Then / ( •) is convex , i.e.,
f ( X x + (1 - A ) y ) 4 A f ( x ) + (1 - A ) f ( y ) Vx, y E R m A e ( o, i ) ,
if and only if the Hessian f x x ( x ) is positive semi-definite, denoted by
/** 0*0 0, for any x G Rm .
—
3. ( Darboux Theorem ) Let / : [a , b] > R have derivatives on [a , 6] .
Then for every y between /'( a ) and f ' ( b ) , there exists a c G [a , b] such that
/'(c) = y . (Hint: Consider g ( t ) = f ( t ) - y t .)
—
4. Let / : R > R satisfy the following: For some a > 1,
|/ (®) - / ( j/ ) l < K \ x - y \ a V x , y e R.
Then / x = / 0 ) for all x G R
( ) ( .
5. Let / : R R be differentiable and lim yoo /
'(x ) = 0. Then for any
a G R fixed , lim [/ (x + a ) — / ( x )l = 0.
x —
X
—¥ OQ L
6. Let a , 6 G R.
J
( i) Show that if n is even , equation xn -f a x -f b = 0 has at most two

distinct real roots;
( ii) Show that if n is odd , equation xn + a x -\- b = 0 has at most three
distinct real roots.
7. Suppose ao , ai , • • • , an G R such that
ai fln i an 0.
ao + T + • • • + n n +1 =
Then the following equation
anxn + an-ixn 1
H h aix + a 0 =0
has a real root in ( 0, 1) .
8. Let -F( x ) = f ( x ) g ( x ) with / ( x ) and g ( x ) being n-th time differ-
^
entiable. Find a formula of F ( x ) in terms of / ( x ) and g ( x ) and their
derivatives. ( Hint: Using induction. )
9. Let / ( •) be C 2 on ( a , b ) and let /"( x ) 0. Then for any x G ( a , b )
and h > 0 with x + 2h e (a , 6) , ^
/ ( x + 2 h ) - f ( x ) - fc "(
/'(® ) = 2h / 0
for some G ( x , x + 2 /i ). Further ,
^ sup | /'( X )|2 < 4 (VxGsup |/ ( x ) |) ( sup //
xe [a ,b\ [a,6] x [a , 6]
10. Let
m= 6 a? X ± o,
0 x = 0.
Find the Taylor expansion of / ( •) at x = 0 . What conclusion can you draw
from it ?
Chapter 4
Riemann Integrals
1 Definition of Integrals
We recall Z , the set of all integers, and let Am be the set of all arrays a :
aJ ) | of R , ji Z, 1 i s$ m } C Rm , ( 1.1 )
satisfying the following conditions:
j e Z, 1 z m,
—a|| = max= —
lim a ]
j -+ oo
oo lim a?
— — ai\=< , .
j too
+1
oo 1 z m
( 1.2 )
sup
;
We call ||a || the mesh size of a. For any a

[o
^ G
oo
4m , define
*
m
Qa ={ Q= FNW + ] 1
) a }. ( 1.3)
=
t l
Clearly,
Rm = |J Q
Q Qa (1.4)
Q° f ]Q = 0, V Q, Q Qa Q , Q.
We call Qa the rectangular partition of Rm associated with the array a .
^
m
J—J [af , af + ] , diameter, denoted by diam ( Q ) , is defined

"
1
For Q = G Qa its
i 1
by the following:
m i
diam ( Q ) = { ]T}(<*f +1 - of ) 2}' < \/m ||a|| ( 1.5)

i= 1
95
and its volume (in Rm ) , denoted by | Q|, is defined by

m
\Q \ == ji +1
- aJ‘K ||a |r . (1.6)
i l
If a , fi G Am such that a C /3, we call fi a refinement of a. It is clear that
WfiW 11 « Va , /3 G 4m , aC. fi . v (1.7)
Also, if a , /3 G *4m , then a U fi is a refinement of both a and /3. Thus,
^ ,.by defining(1.8)
a U /3|| min{||a || , || /3||} Va , /3 G m
Because of this, we see that for any sequence {an} C Am
fin — fc = i
n 1 (1.9 )
we obtain a sequence { fin } £ such that
/3 + fim n 1.fin ( 1.10)

&n
^ n l)
In what follows, whenever saying ||a || —> 0, we always mean that there •
exists a sequence {an} Am such that an an + for all n 1 and

||an || > 0.
-
C
^ C i
Now , for a bounded domain G C Rm , by our construction , for any

a G .4m , G and 9G will be covered by finitely many subsets of Qa , respec-
tively. We will denote Qa ( G ) and Qa ( dG ) to be the smallest subsets of Qfi
covering G and <9G, respectively. Also, we will denote
Let
Q“’° (G )
- Qa ( G )\ Qa ( dG ) .
Qa ( G ) = { Qae | 1
* < 4} -
Q“ ° ( G ) = {Q? | 1 < * < 4} ( 1.12)
where 4
^ 4- Thusea, (0G) = {Q
? 4+ (1.13)
Note that Qa ,° ( G ) covers most part ( but not necessarily all ) of G°. If we
let
G? = G f ) Q e , 1 l ion OL G Am (1.14)
then
Ga = 1J Q C G° C G C |J Q = Ga (1.15)
-
Q Qa ° ( G ) Q e e° ( G )
Riemann Integrals 97
and
| GQ \ GQ| = £
Q e Q* ( d G )
\Q \ . (1.16 )
The following assumption will be important later .

( HI ) Let G C Rm be a bounded domain such that
\1 dG \1 = Wlim \ Q \ = 0. (1.17)
|-+ o ^
QeQa (dG )
Condition (1.17) is referred to as the zero volume condition of the bound-

ary dG of G in Rm . Now , we present a simple result concerning bounded
domains , whose proof is pretty obvious.
Proposition 1.1. Let G C Rm be a bounded domain. If we denote

V$ [ G ) =
Q Q a(G ) ^
l l> V- ( G) = E
Q e Q a >° ( G )
101, -
(1 18)
then
V° ( G ) V «(G ), Va Am (1.19)
and
V£ ( G)
Consequently,
V+ ( G )
^
V (G ) V° ( G ) V a , (3 e A m , a C /3. (1.20 )
lim 1/ “ ( G ) = 14 ( G ) VC (G) = limnV“ (G ) . ( 1.21 )
Further , if (1.17) holds, then

MG ) = MG ) = | G | ( 1.22 )
which is called the volume of G .
Note that if condition (1.17) were not assumed for a bounded domain
G C Rm , we would not be able to define the volume \ G \ as (1.22 ) .
Now, let G C Rm be a given bounded domain satisfying ( HI ) , and let

/ :G —
» R be a bounded function ( not necessarily continuous ) . For any
a G Am , recalling (1.11)— (1.15) , we let
r A ( ) sup f ( x )
“ / =x G«
1 t a G A m. (1.23)
' —
£ (V / )
A? inf f (V x )
x G G“ '
Define simple functions
tec toe
/? (*) = 'E w f W * )’ /- (*) = £
t= 1 t= l
w)i«? (*), xeGa
(1.24)
ea
X
^
/+ (*) = A“ ( ) 1(?? (x ) ’ /- (x ) =
t= 1 ^ t= l
Note that if Qf and Q%, are next to each other , then

(*) x e G“ .
^t ,t' = Qe Q% 7^ 0-
Therefore, rigorously speaking , according to the above notation , for any
x
^ >
/? (*) = A? ( / ) + Ag ( / )
and it might be even more complicated if the point x is in the intersection of
more than two rectangular sets (since Rm could have the dimension m > 2 ).
But , instead , we define
/? (x) = A? ( / ) VA? ( / )
and misuse the notation in (1.24) for simplicity ( and convenience). Like-
wise, for any x e we define
/? (*) = W) AA? ( / ).
The case of x being in the intersection of more than two rectangular sets
can be treated similarly.
Now , we have
/f (x ) / (x ) < f+ ( x ) X G a C G°
(1.25)
f-(x) f (x) /“ (x ) x e G C G a.
Further , let us define
tec tec
$? ( / ) = £ A? ( / ) IQ? I ,
^ (/) = Ew) io? i
£= i £=1
(1.26 )
5?( / ) = EA
=
^ i ^-
“( /) i “ i 5? ( / ) = EW)|Q?|.
*= i
We call S+ ( / ) and 5+ ( / ) the upper Riemann sums, and 52 ( / ) and 52 ( / )

the /ower Riemann sums of / ( •) on Ga and G, respectively. We have the
following result whose proof is straightforward .
Proposition 1.2. Let G C Rm be a bounded domain and let f : G R
be a bounded function. Then
S$ { f ) > S Z ( f ) , $? ( / ) > 3? ( / ) Va .4m (1.27)
and
'
s$ ( f ) > S$ { f ) S t ( f ) S* ( f ) V a , f3 & A m , a c p. (1.28)
\ s? ( /) > 5? ( /) > # ( / ) > £ ( / )
Consequently,
lim 5T
“ (/)
|| a||-> 0
= 5+ ( / ) S_ ( / ) = ||olim
|| —> 0
5“ ( / ) ,
(
(1.29)
lipns ( / ) = 5+ ( f ) s- ( f ) = ||alim
||„ a ||->- 0 ? ||—> 0
§* ( f ) .
Note that if M > 0 is a bound of / ( •) , then

ia e<*
|S? (/) - 5? ( /) l = E A“ ( / ) i <3“ i E \Q f \
*=*a + l
and
|5f ( / ) - 5“ ( / )| = E+ A“ ( / ) iQ“ i E i «“ i -
&= l id
*a (
""
l
Hence
|S+ ( / ) - S+ ( f )| = lim |5“ ( / ) - S+ ( / )| M|9G|,
(1.30 )
13-U ) - S- ( f ) I = IM-+01ISf ( / ) - OT) | M \ dG \ .
Consequently, under ( HI ) , one has

S+ ( / ) = $+ ( / ) > £- ( / ) = £- ( / )
We refer to 5+ ( / ) and 5_ ( / ) as the 2/pper and lower Riemann integrals
of / ( •) , respectively. The above result shows that if G C Rm is a domain
satisfying ( HI ) , then any bounded function on G admits upper and lower
Riemann integrals. Then the following definition becomes very natural.
Definition 1.3 If .
S+ ( f ) = S+ ( f ) = £_ ( / ) = S- ( f ) (1.31)
then we say that / ( •) is ( Riemann) integrable on G , and denote
S+ { f ) = S- { f ) = f f ( x )d x , (1.32 )
JG
call it the definite ( Riemann) integral of / ( •) on G . Function / ( •) is called
an integrand. The set of all bounded Riemann integrable functions on G
is denoted by L ( G ) . When G = ( a , 6) for some a , b E R with a < 6, we
usually denote ^
[
JG
f ( x )d x = [
Ja
f { x )d x (1.33)
and call a and b the lower and upper limits of the definite Riemann integral.
Note that if we define
(1.34)
t=i
with x*I e Gf , then
S? ( / K S f ( / ) < S? ( / ) . (1.35)
Thus, in the case / ( •) is Riemann integrable, one has
l i m S (/) =
IHI -+0 * JG
f f ( x )d x . (1.36)
The following result shows that the set of Riemann integrable functions is
very large.
Theorem 1.4. Let G C Rm satisfy ( HI ) . Then any continuous function

f : G -» E is Riemann integrable on G.
Proof . Since / ( •) is continuous on the compact set G, it is uniformly

continuous. Thus, for any e > 0 , there exists a 5 > 0 such that
0 A“ ( / ) - A? ( / ) < e, Vl < * < 4, a G *4m , ||a || < 5.
Next , we let
M = max \ f ( x ) \ . (1.37)
xeG
Then the following holds:
eQ toe
o S%U ) - 5f ( / ) = E [ A? ( / ) A? ( / )
=
£ 1 £= l
Hence, it follows that

0 < 5+ (/ ) - 3_ ( / ) = || lim
|| —> 0
o:
§" ( / ) - lim S Z ( f ) C e ,
*
|| ||— >- 0 ^ o
(1.38)
for some constant C > 0. Similarly,
0 S+ ( f ) — S- ( f ) = lim 5? ( / ) - lim 5“ ( / )
—-
||o:|| > 0 —-
|| o || > 0 < Ce. (1.39)
Combining the above with (1.30 ) , we have the Riemann integrability of / ( •)
since e > 0 is arbitrary.
Example 1.5. Let G C Mm satisfy ( HI ) . Then the function f ( x ) = 1 is
Riemann integrable on G and
d x.
G
This is the volume of G .
We point out that if / ( •) is not continuous on G , the above theorem

might not be true. Here is an example.
Example 1.6. Let G — (0, 1) and

1 x e QH [0 , 1] .
/ (*) = (1.40 )
* 6 [0, 1] \ Q.
Q
Recall that this is the Dirichlet function. For this function , we see that
A“ ( / ) = 1, A? ( / ) = 0, VI i 4, a G Am .
Hence
5? ( / ) = 1 , S“ ( / ) = 0, Va 6 Am .
Consequently, (1.31) fails. This example also shows that a bounded function
is not necessarily in L ( G ) .
^
Despite of the above example, the following result tells us that some
discontinuous functions are still Riemann integrable.
Theorem 1.7. Let G C Rm satisfy ( HI ) . Let f : G > M be bounded and —
denote
D( f ) — {x G G | / ( ) is discontinuous at x }.
• (1.41)
Suppose that
\ D ( f ) \ = ||alim Y" | Q| = 0. (1.42)
|K 0
Q
^
Q a ( D( f ) )
11
Then / ( ) is Riemann integrable.

•
Condition (1.42) means that the set at which / ( •) is discontinuous has
a zero volume in Rm . Note that Dirichlet function (see Example 1.6 ) does
not satisfy condition (1.42 ) .
The proof of Theorem 1.7 is essentially the same as that of Theorem 1.5
in which we replace dG by dG U D ( f ) . The details are left to the readers.
The following corollary is very useful.
Corollary 1.8. Let G C Rm be a bounded domain satisfying ( HI ) and / ( •)

be Riemann integrable on G . Then for any rectangular domain Q D G , the
function / ( •) defined by
’
/ (*) x e G,
m= 0 x G Q\ G ,
(1.43)
is Riemann integrable on Q , and
f { x )d x = f ( x )d x . (1.44)
Q G
Proof . Using the same technique used in the above proof , treating dG
as a part of D ( f ) , the discontinuous set of / ( •) , we are able to obtain the
conclusion.
The above result shows that when discussing Riemann integrals, one
can assume that the bounded domain G is a rectangular , as long as ( HI )
holds.
Next result gives a nice approximation of bounded Riemann integrable
functions, which will be useful later .
Proposition 1.9 . Let G C Rm be a bounded domain satisfying ( HI ) . Let
f : G -* R be bounded and Riemann integrable. Then for any e > 0, there
are continuous functions g ± : G » R such that — •
9- { x ) f{x) g% { x ) WxeG (1.45)

and
[p+ ( z ) - ge_ ( x ) } d x < s . (1.46)

G
Proof . First of all, for any a' < a < b < b' , we define
V;a' ,a ,6,6 (s) —
/ -1
1 [a ' , a ) ( s ) + 1[a , b] ( 5 ) +
jy — §
1 ( 6 , 6' ] ( s ) s G R.
o'
—
Then il)a' ,a ,b ,b' : R > R is a continuous function. Now , for any rectangular
£Li “
Q = II [ i > &i ] > we let
771
Q' = UteAl a'i < ai < bi < b'i , 1 i m
and define
—
i 1
771
X = ( x1, • • • xm ) G Rm.
Clearly,
—
i 1
C ( Rm , R ) and
/
? >Q , Q G
'
1 x G <2 ,
Q Q . , w= 0 X G Rm \ Q'.
For given rectangular Q , let Q', Q" be rectangulars satisfying
Q" cQ° cQc (Q')°.
We have
1Q » (*) (X) lQ (x ) , (x ) (x) , x Rm
which leads to
,Q Q,Q
^1 Q/
G
o<
4 ipQ Q , { x )dx - \ Q \ |Q' \ Q|
0 |Q| - / Q
( x )dx < |Q \ Q"|.
Thus
(K
/v.
JQ'
[ >Q Q '
{x) -
^ „ { x )] - ^ \ '\
Q Q
dx Q Q \ + \ Q\ Q" \ .
—
Now , let / : G » [0 , M] be bounded non negative and Riemann integrable.
Then for any e > 0 , there exists an a G Am such that
ea
o /f (*) = £ W )!«? (*) < / (*)
£ =1
101 *
< EA? ( / ) iof (*) = /+ (*x XGG

*= I ^
and
0 5+ ( / ) - 5 ( / ) = _ [ [/£ (*) - /f (x)]dx

JG
Ia
= £= [a
£ 1
+ ( / ) - A “ ( / ) ]|Q?| <
Let us take
m" c ( o? )° c o f c i m r .
Define
ta
.
f
9+ ( x ) = W)
( )
' * 9- ( x ) = *
(*) •
*=i &= i
Then
0 < / “ (a:) f(x) < /+ (x ) £+ ( x ) M, xeG
and
[?+ (*)
E {A?(/ ) /
1 Qf 'CQgy ( x )
dx — X “ ( / ) J[ Rm
(Q ? )" ,Q?
<= i
E {A?(/) [ /J ?
felL *
L m
W no
r Q n ,( Q
? Y (*)
dx — a
ea
+ Au f )\m - [ V( Q f
”
) , Q ? (*)ds
f ] + £[A? ( /) - A? (/)|Q?|}
J 1 IRm £= 1
to
E= { A?(/)KQ?) \ Q? I + A?(/)IQ? \ (Q?n} + s?(/ )
/
- sf ( / )
* 1
M E { ic^)' \ Q? l + IQ? \ (Q? )"l } + 1 -

i i
=
Then by taking ( Q f Y , ( Q f )" suitably, we obtain our conclusion
For general case, we have / ( •) = / + ( •) / ( •) with both / + ( •) and / ( •) — “ “
being non- negative. From the above, we can find ( non- negative) continuous
- -
functions g+ ( - ), g t ( ) , 9+ ( ) , g Z { ) such that -
9- { x ) < f+(x) 5+ (^) > 9- { x ) f~{x) g+ ( x ) xeG
and
[ \g l ( x ) - g ± { x )\d x + [ [ff + ( z) - g _ ( x )\d x < e.
Then we have
g ~ ( x ) = g t { x ) - g+ ( x ) / + (x ) - / ( a:)
= / 0*0 < 0+ 0*0 - 9- ( x ) = 0+ 0*0 .
and
[ g+ ( x ) - g- ( x ) ] dx = / _
[;g ( x ) - g ( x ) - gt ( x ) + g+ ( x ) ] dx < e.
Mm J Em +
Next , we define the Riemann integral of possibly unbounded functions
on possibly unbounded domains.
Definition 1.10 . Let G C Rm be an unbounded domain with the property

that
lim \ dGM ( x0 ) \ = 0, Vx o G Rm ( 1.47 )
M -¥ O O
where
— {x G G | |x — xo| < M } = G C B( xo, M).

GM ( XO ) \
Let / : G —> R be a possibly unbounded function. For — iVi , AT > 0, define 2
/ 0*0 Nx < f ( x ) < N 2 ,

M$ (*) Ni f(x)
^ Nu x eG
N2 , fix) > N , 2
which is a bounded function defined on G . Suppose [/] ( ) is Riemann

integrable on GM { % O ) > for any xo G Rm , M , — N\ , N 2 > 0; and for any
£ > 0 , xo G Rm , there exist Mo , No > 0 such that
^ *
[
J G M ( XO)
[ f } Nl ( X ) d x - [
'G
[ /] jv' ( x )dx < £
C M' ( X O) ( 1.48 )
VM, M' M0
Then the following limit exists which is called the Riemann integral of the
function / ( •) over G :
lim
M , \ N 1 \ , N 2 - yoo
t
JGm
[ f ] 1N 1 ( x )dx = j f ( x )dx -
JG
( 1.49 )
^
In what follows, we denote LR ( G ) to be the set of all Riemann integrable
functions / : G —> R , in the above sense, which could be unbounded and
not necessarily continuous. Note that we should allow |iVi |
.
Xo G Rm to be arbitrary This will exclude the following function from
N 2 and
^
being Riemann integrable:
1
x 7 0,
^
/ (*) = { x
0, x = 0.
In fact , for this function , we have
rM
/-
[ f } - N ( x ) d x = 0.
J M
According to the above definition , if Go C G are two domains, both satisfy
(1.47) , then / ( •) LR ( G ) implies / ( •)|Go e LR { G0 ) , where / ( •)|Go is the
restriction of / ( •) on Go . Note that it might not be the case if Go is just
a ( possibly disconnected ) sub-open set of G (see Problem 7 at the end of
this section ) .
To conclude this section , we introduce the following definition .
Definition 1.11. Let G C Rm be a domain satisfying (1.47). Let / : G > —
R . If both / ( •) and |/ ( )|p are Riemann integrable on G, for some p 1,
*
we say that / ( •) is Lp - Riemann integrable; in the case p = 1, we say that

/ ( •) is absolutely Riemann integrable.
—
It is possible that a function / : G > R is not Riemann integrable but
-
| / ( ) |p is Riemann integrable. Here is a simple example.
1 x e Q fi [0, 1]
/ (*) =
i -i . x G [0, 1] \ Q.
Hence, the above definition , excludes the above function to be Lp- Riemann
integrable. We denote LPR
( G ) to be the set of all Lp-Riemann integrable
functions on G. Then it is clear that
L\{ G ) C LR ( G ) , (1.50 )
which means that absolute Riemann integrability implies Riemann integra-
bility but not vice versa.
Exercises
1. Let G C Rm be a bounded domain with \ dG \ = 0 and / : G —> (0 , oo )
be a continuous function. Then is Riemann integrable on G.
2. Let / : [a , b] -» [0, oo ) be continuous. Then
f f ( x )d X 0.
Ja
3. Let /, g : [a , b] -> R be continuous. Then
IJ f ( x ) g { x )d x ( J\ f ( x )\2 dxy ( j \ ( x )\ dxy .

g 2
( Hint: Consider f
Ja
[ f ( x ) + Ap (s)] 2 d x 0. )
4. Show that when (1.48) holds, the limit (1.49 ) exists.

5. Let
1
f(x) = { q
0
x=
x
^ e On [0, 1] , q
[R \ Q
Then / ( •) is Riemann integrable on [0 , 1]. What is the value of
] n [ , ].
0 l
^ o,
f * f ( x ) d x?
—
6. Let / : [a , b ] » R be Riemann integrable. Let oi , a 2 , • • > an [a , b] *
be finitely many different points, and ci , C2 , • • • , cn G M be arbitrarily given.

Define
= 1 , 2, - - , n ,
9( x ) =
Ci
/ (*)
,
x
x ^= a u i
d i , i = 1 , 2, - - , n .
•
Prove or disprove that g ( - ) is Riemann integrable.

7.* Let
sinx
x e (0 , oo ) ,
/ (® ) = x
1 x = 0.
Show that / ( •) e LR ( [0 , OO )) \ L ) ( [0, oo )) . Moreover , there exists an open
^
set Go C [0, oo ) such that JGQ f ( x ) d x = oo.
2 Properties of Integrals
The following result collects some basic properties of Riemann integrable

functions.
Proposition 2.1. Let G C R m be a bounded domain satisfying ( HI ) ,
( i) For any E LR { G ) , and c £ l ,
/ [c f { x ) + g ( x )\d x — c / f ( x ) d x + / g ( x ) d x ( 2.1)
JG JG JG
Consequently, LR ( G ) is a linear space. The same conclusion also holds for
LpR ( G ) .
( ii) I f 6 LR ( G ) and
f (x) g( x ) VxeG ( 2.2 )
then
/
JG
f ( x )d x
JG
/ g ( x ) d x. ( 2.3)
Thus, the Riemann integral preserves the order of R. Consequently, for any
/ ( •) LlR( G ) ,
G
f ( x )d x
JG
[ \ f ( x )\ d x. ( 2.4)
( iii) L e t - - -
e L% ( G ) . Then f ( ) g ( ) , | / ( )|, / ( •) V $( - ) , and / ( •) A
# ( •) are i n L ( G ) , where recall that aWb = max{a , 6} and aAb = min{a , 6}.
^
(iv ) Let G = G\ (J with both G\ and G 2 satisfying ( H I ) , a n d
Gif|G 2 = 0. ( 2.5)
L]j( G ) . Then / ( •)
Let / ( ) •
f f ( x )d x =
^
L Gj ) , / ( ) e L ( G2 ) , and
f f ( x )d x + f
^f ( x )d x . ( 2.6)
JG J G\ J (? 2
Proof . We leave the proofs of ( i) and ( ii) to the readers.

(iii) For any a £ Am , we define /"( ) and <?“ ( •) as in (1.24 ) . Then
f-( x ) / (x ) f+ ( x ) , g°L ( x ) s( x) g'f ( x ) X G . ( 2.7)
This yields
/- (*) V 5“ ( x ) < / (x ) V ff (x) /+ (x) V g% ( x ) x e G a. ( 2.8)

Hence
L,
f - ( x ) V g1 ( x ) < Vs) lGf ( x )
<= i ( 2.9)
L,
5<=IA“ (/ v 5)1 “ (ar) < /"(x) Vff ^ (x)
i
G x e G“ .
This leads to
f * ( x ) V t f L ( x )d x SZ ( f \f g ) S$ ( f V g )
G« ( 2.10 )
/ /+ 0*0 V
^ ( x ) cte.
By the Riemann integrability of / ( •) and g ( ) , for any e - > 0, there exists a

/3 6 Am such that
[ f ( x )d x St ( f ) S+ ( / ) / / (x)dx + £
h [ g( x )d x [ g ( x )d ( 2.11 )
—£ ~ S“ (p ) S+ (# ) X +£
JG JG
m,
a 2 /?.
Let
Va G
^
0 /*“ ( •) = /? ( ) - /- ( •) + 3“ ( ) • ( 2.12)
Then by (ii) ,
0 < / h a ( x )d x
// f + ( x )d x -
J
[
Ga
f “ ( x )d x + [
JGQ
g+ { x ) d x -
/ g°i ( x ) d x
( 2.13)
= S? ( ) - W ) + - Sf (5 ) 4e .
Observe the following:
m* ) = m + [/? (*) - /- (*)] < /“( ) + * (*) ®

o
and
p+ ( x ) = g - ( x ) + [ g% { x ) 9- ( x ) ] g° ( x ) + h a ( x ) .
Therefore
f% ( x ) V g% ( x ) <!
/ (I) VJ (I) + /i“ ( x ) .
Hence
o s° ( f V g ) - S Z ( f V g )
[ [ f% ( x ) V g+ ( x ) - f “ { x ) \/ g° ( x ) ] d x
JG <*
[
JGQ
h a ( x )d x 4e .
Since e > 0 is arbitrary, we obtain the Riemann integrability of / V g .

The Riemann integrability of / A # follows from that of [ ( / ) V ( # ) ]. — — —
Then we further obtain the Riemann integrability of / + = / V 0 and
~
—
f = f AO. Hence, |/ | = / + / is Riemann integrable as well.
”
Now , we show that / ( •) <? ( •) L ( G ) . To this end , we first let
r^
£
/ = /+ 9=9
+ ~
9
with
/ + = / V 0, /- = -( / AO) ,
<?+ = <? V 0, g - = - { g AO ) .
We have the Riemann integrability of f ± and g ± . Further ,
fg = ( f + - f ~ ) ( g + - 9 ~ ) = f + 9+ ~ f + 9 ~ - f ~ 9 + + f ~ 9 ~ - _
Hence, the Riemann integrability of f g is implied by that of / + g + , / + <? ,
f ~ g + , and f ~ g ~ . Therefore, it suffices to prove our conclusion for the case
—
that f , g : G > [0, M] , for some M > 0.
-
Now , for any a e A m , we define f ± ( ) and g± ( ) as in (1.24 ). Then
(comparing with ( 2.7))
-
0 /"( x ) f (x) f%( x )
x e G a. ( 2.14 )
0 gt(x) g+ ( x ) .
g( x )
This implies (since all the involved terms are non-negative)
f - ( x ) g° ( x ) f ( x )g{ x ) f % ( x ) g% ( x ) , x e G a. ( 2.15 )
By the Riemann integrability of / ( ) and (? ( ) , for any e > 0, there exists a
/3 A m such that ( 2.11) holds. Also, by ( 2.12 ) ,
0 < /? (x)5“ (x) - /« (x )5“ (*)
= /£ (*) [# (*) - <£ (*)] + £ (*) [/? (*) - /“ (*)]
M [# ( x ) - <?“ (*)] + M [/ “ ( x ) - /“ (x)] = M h a ( x ) .
Hence, using ( 2.13)
0 < S+ ( f g ) [ [ f+ ( x )g° ( x )
— S- ( f g ) < JGa \
f * ( x ) gZ ( x ) d x
M h a ( x )d x 4Me.
Since £ > 0 is arbitrary, we obtain the Riemann integrability of / ( •) # ( •) .

-
(iv ) By ( iii) , we have that / ( •)1GI ( ) and / ( )1G 2 ( - ) are absolutely Rie-
*
mann integrable, and

f (x) = / ( x ) lGl ( x ) + / ( X ) 1G 2 ( X ) x e G.
Then (i) applies to get ( 2.6 ) .
Note that the above, ( iii) implies that Lg> (G ) C L ( G ). Combining
with (1.50) , we have ^
LR { G ) C Li (G) C Lfl (G ). ( 2.16 )
Now , we briefly look at vector-valued function case / : G -» Re . We

have the following definition.
—
Definition 2.2 . Let / : G > R ^ , / ( •) = ( /1 ( ) ? , / ^ ( ) ) We say that * * * * *
-
/ ( • ) is Riemann integrable on G if each component /* ( •) G LR ( G ) . In this
case, we denote
T
G
f { x )dx = ( j\
f x )dx , ... G
We call it the definite Riemann integral of function / ( •) on G . The set of all

Riemann integrable Revalued functions is denoted by LR ( G ] R£ ) . Similar
to scalar functions , we denote ( G', R^ ) to be the set of all functions
LPR
-
/ ( ) G LR( G ; Re ) such that |/ ( )| e LR { G ) .
p
It is possible to use the upper and lower Riemann sums to define the
Riemann integral of vector-valued functions. We leave the details to the
interested readers. Note that the definite integral of an Revalued function
/ ( •) is a vector in R^ . For Riemann integrable vector-valued functions, we
have the following result .
Proposition 2.3. Let G C Rm be a bounded domain satisfying ( HI ) .
( i) Both LR ( G ; R ^ ) and ( G ; R^ ) are linear spaces ( for all p 1 ).
LPR
L G ; Re ) ,
^
( ii) For any / ( •) G
^ f ( x )d x [ ll / (*) l|d»- ( 2.17)
G JG
( iii) Let G = G\ (JG2 with both G\ and G 2 satisfying ( HI ) , and ( 2.5) .
Let / ( •) G L G ; Re ) . Then / ( •) G L Gy, Re ) , / ( •) G L 1R ( G 2 ; Rf ) , and
^ ^
L f ( x )d x =
J
[
G\
f ( x )d x + f
J G2
Proof . The proofs of ( i ) and (iii) are straightforward. We now look at

f { x )d x . ( 2.18)
( ii ) . By definition ,
I2 e r 2
fG r (x) ( j dfj dx
G
JG
f ( x )d x
I
\ J2
I
=
JG
\
f x )dx
( x ) (yfG f\ y )dv ) dx =
— Z 1
fG [f ( x ) fG f ^ f\y )
( y ) dv d x
[ 11 / 0*0 I [
JG JG
f ( y )d y d x = I [ f { ) \II J[G l / Mll ^x.
II J G
y dy
Thus, ( 2.17) follows.
Exercises
1. Suppose / : [a , b] -» [0 , oo ) is continuous. Then
What if
/
Ja
fdx
/ is piecewise continuous?
=0 V x G [a , 6].
2. Let / : G -> M be a uniformly continuous function on a bounded

domain G. Then / ( •) is Riemann integrable on G . ( Note: G is open.)
3. Let f ( x ) = sin which is defined on ( 0, 1) . Is it uniformly continuous
^
on (0 , 1) ? Is it Riemann integrable on ( 0 , 1) ? Why? What conclusion that
you can draw from your conclusions of the above?
4. Let / : ( a , b ) —» R be a bounded monotone function. Then /( )
• is
Riemann integrable.
—
5*. Let /, g : [a , b] » M be continuous functions. Then
b
j\ { x )\ Pdx
1
\ f ( x ) g ( x ) \ dx ( f
where p , q G ( l , oo ) , j +\ = 1.
6* . Let / : [a, b\ —> be continuous and p G (1, oo ) . Then
b
f ( x ) dx
P
< J[a \\ f ( x ) \\ pdx .
3 Further Theorems
In this section, we present some more results on Riemann integrals. The

first result is the absolute continuity of Riemann integral, which is the prop-
erty of the Riemann integral stated in the following proposition.
Proposition 3.1. Let G C Rm satisfy (1.47) , and / ( •) G LR ( G ) . Then

for any e > 0, there exists a 5 > 0 such that for any countable family
{Ga , a e A } of disjoint subdomains of G , provided
| IJ G« | < <s^ \ dGa \ = 0, VaG A, ( 3.1)
the following holds

f ( x ) dx < e . ( 3.2 )
Riemann Integrals 1 1 3
P r o o f . First of all, if / ( •) is bounded by M , the result is trivial. In fact ,

we need only to take 5 = Now , for general case, we write f = f + f ~ —
with / + = / V 0, / = ( — / ) V 0. By definition , for any xo E Mm and e > 0,
“
there exist M , N2 > 0 such that ( recall GM { %O ) = G fl B ( xo , M ) )
0 < J[G f + { x )d x - J[CG M ( XO )

[ f + ) o 2 ( x )d x
L G\G M { X O )
/ + ( x )d x + [JGM ( f + { x ) - [ f + ] o ( x ) ) d x < 4j .
{ XO ) V
Let S = 4 j - . Then for any countable family {Ga , a £ A } of disjoint
2
/
^
subdomains of G satisfying ( 3.1) , denote Go = UaeAGa . We have
L f + ( x )d x — JG ( f + ( x ) - [ / + ]02 ( x ) d x + ) JG [ f + } o 2 ( x )d x
LG Q\G M ( X Q )
( / + (*) [ f + ] oa ( x ) ) <b
-
+
GQDGM ( xo )
(/ + (*) Lf +]or ( )) d +|
- 2
® ®
L t ( f + { x ) - [ f + ]o
/+ { x ) d x + )
(x) dx + <
^
2
.
J G\G M ( X O ) J G M ( XO ) V ' 4j 2
We can do the same thing for / ( ) . Then combining them , we obtain our
“
•
conclusion.
Note that in Rm , any family of disjoint non-empty open sets must be
countable. ( Why? )
The following is a one-dimensional case.
Corollary 3.2. L e t / : [a , 6] -> R be Riemann integrable. Then for any
e > 0, there exists a 5 > 0, such that for any family of disjoint intervals
( a* , bi ) C [a, b\ , i > 1, as long as
TP* - Oi ) < 5
i l
,
the following holds:

^
bi
f { x )d x < e.
Next result is called the Fundamental Theorem of Calculus which con-

sists of two parts, the differential form and the integral form.
Theorem 3.3. Let f : [a , b] —> M he Riemann integrable.
—
( i) Let F : [d , 6] > R be deSned by
F( x ) = [
Ja
f ( s )d s ( 3.3)
-
Then F ( ) is continuous on [a , b\ . Further , if / ( •) is continuous at xo E ( d , b ) ,
-
then F ( ) has derivative F' ( xo ) a t xo a n d
F\x 0 ) = f ( x0 ) . ( 3.4)
( ii) Let / ( •) have at most countably many discontinuous points and
—
F : [a , b\ > R be an anti- derivative / ( •) , i.e., at every continuous point XQ
F' ( x0 ) = f { x o ).
Then the following Newton- Leibniz formula holds:
[ f ( x )d x = F ( b ) - F ( a ). ( 3.5)
Ja
-
Proof , ( i) The continuity of F( ) follows from Corollary 3.2. Now , let
xo E (a, b ) at which / ( ) is continuous.
• Then for any e > 0, there exists a
S > 0 such that
\ f ( x ) - f ( x0 ) \ < s , V|x — xo \ < 5 , x e [a , b].

Next , for any x e [a , b] with \ x — X Q \ < 5 , one has
pC
1 J/x
%
|F ( x ) - F ( X Q ) - f { x0 )( x - z0 )| = f ( s ) d s - f ( x0 )( x - xo )
1 0
f | f ( s ) - f ( xo ) \ d s < e\ x - x0 \ .
)
f { s ) - f ( x0 ) d s
dxo '
-
Hence, F ( ) has derivative F' ( xo ) at xo and ( 3.4 ) holds.
(ii) We only prove the case that / ( •) is bounded and has finitely many
discontinuous points listed as follows:
a — do ^ d /c
— i d /e d / j-f- i = b.
By mean-value theorem , we have
F ( a i ) - F { d i-1) = F' ( £i ) ( a i - a* _ i ) = / ( & ) (!

< * - a,_ i )
i 1, 2, • • • , k + 1, —
for some & G ( a* _ i , ai ) . Hence,

k +1 fc + i
saM ) = T ] xG [ inf_ , ) / (x)(ai - at _ i )

t =l
cii i di
2 =1
/ (& )(«» - a »- i )
fc + 1
£ [nO - = F ( b ) - F ( a)
=
2 1
fc +1
sup f ( x )( d i — d i - i ) = S“ ( / ) ,
2 = 1 X G [fflj — 1 )
where a = {ao , ai , • • • , afc , afc + i} G A1 . Also,
S“ ( / )
Ja
f f { x )d x £% ( f ) .
Thus
6
F( 6) - F( o) - [
Ja
f { x )d x ( / ) - £“ ( / ) .
Note that the left- hand side is independent of a. Hence, letting ||a|| > 0, — •
we obtain our conclusion.

The general case is left to the readers.
The following result is called the integration by parts.
Proposition 3.4. Let F , G : [ a , b ] —> R have derivatives such that F' ( ) -
-
and G'( ) are Riemann integrable on [a , b\ . Then
f
Ja
F { x )G' ( x ) d x = F ( b )G { b ) - F ( d )G ( d ) -
Ja
f F ( x )' G ( x ) d x .
Proof . By the Newton-Leibniz formula, we have
F ( b ) G ( b ) - F ( a )G ( a ) = f F ( x )G ( x ) d x
Ja ^
[ F' ( x )G ( x ) d x + f F ( x )G' ( x ) d x .
Ja J a
Then our conclusion follows.
Riemann integral has two mean-value theorems. We now present the
First Mean- Value Theorem.
Theorem 3.5. Let G C Rm be a bounded domain satisfying ( HI ) . Let
f , g : G -» R be continuous, with g ( x ) 0 for all x G G . Then there exists
a £ e G such that
[
JG
f ( x )g ( x )d x = / (£) J[G g ( x ) d x . ( 3.6 )
-
Proof . Since g ( ) is continuous and non- negative, thus if
/ g ( x )d x = 0
JG
we must have
g ( x ) = 0, x e G.
For such a case, ( 3.6 ) is obviously true. We now assume that
/ g ( x )d x > 0.
JG
By the continuity of / ( ) on the compact set G , we may let
•
A = min f ( x ) A = ma x f ( x ).
xeG XEG
Then
f ( x )g( x ) A g { x )
Ap ( x ) x e G.
^ ^
Consequently, by Proposition 2.1, ( ii ) , one has
A / g ( x )d x / f ( x )g ( x )d x A / g ( x )d x .
JG JG JG
This implies that
f ( x ) g ( x )d x
< A.
/
JG
g { x )d x
Now , by the intermediate value theorem ( Corollary 2.8 of Chapter 2 ) , we

can find a £ G G such that
/ f { x ) g { x )d x
/ )=« JG
g ( x )d x
G
which is equivalent to ( 3.6 ).

Let us look at a couple of special cases of the above.
( i) If g ( x ) = 1, then ( 3.6 ) reads
[
JG
f ( x ) d x = f ( £ ) \G \ . ( 3.7)
In particular , when G = [a , 6] , one has
[ f ( x )d x — f ( £ )( b — a ) . ( 3.8)
Ja
( ii) If G = [a , 6] , then ( 3.6 ) reads

/» 6 pb
/a / f ( x ) g ( x )d x = / (0 . / 5(a:) ete ( 3.9)
• / a
Next , we present the Second Mean-Value Theorem of Riemann integrals.

Theorem 3.6. Let g : [a , b] -> R be continuous and f : [a , b] —» R be C 1
such that
f x ) > 0,
\ Vx G ( a , fr ).
Then there exists a £ G ( a , 6) such that
[ f ( x ) g ( x )d x = f ( a ) f g( x )d x + f ( b ) f g ( x )d x . ( 3.10 )
Ja Ja J£
Proof . Let
G( x ) = J fa g ( s )d s x G [a , b } .
Using integration by parts, together with ( 3.6) , one has ( note G { a ) = 0)

[ f ( x ) g { x )d x = J af f ( x )G' ( x ) d x = f ( b )G ( b ) - f f ' ( x )G ( x ) d x
Ja J a
= f { b )G { b ) - G (0 J [
a
f ( x )d x = f ( b ) [G ( b ) - G (0] + / (o)G (O
= / (a )
J 9 ( x )d x + f ( b )
J g { x )d x ,

The following result is called Fubini’s Theorem which says that , under
proper conditions, multiple integral is equal to iterated integrals, and the
order of iterated integrals can be exchanged .
Theorem 3.7. ( Fubini ) Let G = G\ x G with G\ C Rmi and G 2 C Rm 2

being bounded domains satisfying
^
\ 9Gi \ = 0, in Rmi ; \ dG2 \ - 0, in Rm 2 .
Let f : G -» R be continuous. Then
j
j
Lr
f ( x )dx =
J
«/ Crl
[j
«/ Lr 2
f ( x i , x2 )dx2 dxi
( 3.11)
f ( x\ , X 2' jdX y dx 2 '
J G1
where the right-hand sides are called iterated integrals.
Proof . First , since / ( •) is continuous on G = G\ x G 2 which is com-
pact , it is uniformly continuous. Therefore, the following integrals are well-
defined and the following defined maps are continuous:
f ( x i , x 2 ) dx 2 x2 f { x i , x2 )dx\.
X\
G1
^ G2
Hence, we have the existence of the iterated integrals:
Gi G2
f ( x i , x 2 ) dx 2 dx1,
l J G1
f ( x\ , X 2 )dxi dx 2 .
Now , we show ( 3.11) . For any a1 G .Ami and a2 G .4m 2 , let

/ 1/
J G2
a = (ai , a 2 ) = 115 * * * , <*1lmi 5 m2 )I

- aL ) 4 < = 1 2}.
|
Next , similar to the discussion at the beginning of Section 1, we let Qai ( G i )

be the smallest subsets of Qai covering Gi , and let
,
_ r*air vo ^t a
G# = <5 , 1 i £Oti 5 * = 1, 2,
^ ^ e ^ Ia .
2
~
Ul , /c X / 1 k ai ? 1 2
Then
f ( xl , x dx1 = lim f ( xlk . T , jdx , |Gfi|

Gi G2 ||QI ||-> 0
lim
— -
||ai || > 0 ||a 2||-> 0 Z^
M
A
’*
2
’*' ' 1,fcM 2/ 1
Y f { x )d x . [
lim
IMI -+0 k ,e
^
f ( xkt ) \Gke \ =
This proves a half of ( 3.11) . The other half can be proved similarly.
By some small modifications, we could relax the conditions assumed in
the above theorem. We leave the details to the readers. See problem 7
below.
Exercises
1. Let / : [0, 1] —> R • be continuous such that
l
/(1) = 0, x f ( x ) 2 d x = 1.
o
Then
and
s
? f ( x ) f' ( x )d x = -2 ,
J
( |/'(x)| dx) ( x f ( x ) dxSj 4.
2
J 4 2
2. Let / : [a , b ] —» R be differentiable. Then

•
/ ( &) - m = ( f ' ( a + £ ( 6 — J ( 6 — a ).
3. Let g : [0 , 1] —> ( 0, oo ) be continuous. Suppose the following holds:
an f xng ( x )d x + an _
Jo
— g ( x )d x H i
A n 1
+ Ul / x g { x ) d x 4- ao / =0
7o Jo
Then the equation anxn + an _ i # n — 1 H h aix -f ao = 0 has a solution in
[0, 1] .
4. Let g : [a , 6] —» R be continuous and / : [a , 6] —> R be C1. Further ,
•
/ ( ) has a unique maximum at 77 G (a , 6 ) . Then there are points G (a , 77 )

•
and £2 ( 77, 6) such that
[
Ja
f ( x ) g ( x )d x = f ( a ) j
Ja
g { x ) d x + / ( 77 ) f
J £1
g ( x )d x + f ( b ) (
J £2
g ( x )d x.
5. Let / : [0, 00 ) —> R such that

•
lim f ( x ) = 77i,
—
x > 00
for some 771 G R. Let l > 0 fixed. Find lim f ( s )d s .

x > 00 —
6. Let Gfc C Rmfc be bounded domains, k = 1, 2 such that
|9G 2 | = 0 , in Kms .
Let / : G\ x G2 -> M be continuous. Define
^ 1) [
= JG
2
f ( xi x2 ) dx 2
) xi G G1.
Then F : Gi -> M is continuous.
7. Let Gi C Rmi and G2 C Mm 2 be bounded sets with
\ 9G\\ = 0, |SG?2| = 0,
in Rmi and in Rm 2 , respectively. Let / : G\ x G 2 —> R • be bounded and let
D(/) = { ( xi , X 2 ) G Gi x G2 | / ( ) is discontinuous at (xi , X 2 )},

•
D ( / ( xi , - ) ) = {x 2 G G2 | / ( xi , - ) is discontinuous at X2 },
£) ( / ( • X 2 ) ) = {xi G G 2 | / ( • X 2 ) is discontinuous at xi }.
5 5
Suppose
\ D ( f ) 1 = 0, in Rmi + m 2 ,
|D ( / (x1 - )) | = 0,
; in R 2,
Vxi G Gi ,
| £> ( / ( - . *2 ) ) | = 0 , in Rmi Vx 2 G G2 .
Then Fubini Theorem holds.
8. Let G C Rm be a bounded domain with \ dG \ = 0 in let /, g :

G —> R be continuous. Let pG (1, 00 ) . Then
i
p
+
9. Let Gi C Rmi and G 2 Q Mm 2 be bounded sets with
1 0, \dG2\ = o ,
^1 =
in Rmi and in Rm 2 , respectively. Let / : Gi x G2 -* M be continuous. Then
for any p G [1, 00 ) ,
( /Gi f
«/ «/ G2
f { x i , x 2 ) d x 2 dxi ) < [ i fGi
P
^ G2
”
^- 2
Chapter 5
Uniform Convergence
1 Observations and Examples
Let us begin with some observations.
•Observation 1. Suppose {anm}n , m i is a sequence with double indices.

^
We may expect or hope that
lim lim anm lim lim anm • ( i .i )
- —
n too m + oc — —
ra > oo n > oo
•Observation 2. Suppose {/n ( )}n is a sequence of continuous func-

^
*
i
tions defined on [a , 6] , n 1, such that
lim /n ( x ) = f { x ) e [a , b] . X
— 71 » 00
We may expect or hope that / ( ) is continuous. In other
• words, for any
#0 ,
£ [n , b ] we hope
lim lim f n ( x ) = Xlim / (x ) = / (x0 ) = nlim lim /n (x ). (1.2 )
— —
x > xo n » oo ± XQ —> oo X - XQ
^
• Observation 3. Suppose {/n ( )}n i is a sequence of differentiable
*
^
functions defined on [a , b] and converges to some function / ( •) . Then we
may expect or hope that / ( •) is differentiable and
f ' ( x ) = lim fn ( x ). (1.3)
This is equivalent to the following:
71
—
> 00
f n( x + h ) ~ f n( x ) f n( x + h ) ~ f n{ x )
lim lim = 7ilim lim (1.4 )
— —
h » 0 71 HX ) h > oo h > 0 — — h
• Observation 4 • Suppose { f n ( ) } n^ i is a sequence of Riemann inte-
'
grate functions on [a , b] and /n ( - ) is convergent to / ( •) . We may expect or

hope that / ( •) is Riemann integrable and
b
f { x ) d x = lim
71 —^ OO /
Ja
fn { x ) dx . (1.5)
121
m m
Y ] f n( x /„ ( x A x ,
where
rpTfl
lim lim
->
771 00 71
—» 00
1 = 1'
)Ax
1=1'
lim lim
—
71 fOO 771
—
» 00
is a partition of [a , 6] ( labeled by rn ) , and Ax

^ ^ ( 1.6 )
=x -
Xi -1 *
Let us look at the following example.

Example 1.1. ( i ) Let
n
^ 71771
— n+m
n , ra 1. (1.7)
Then
Thus, in general 1.1) fails ,

—
71
,(
lim lim a71771
tOO 771 —» 00
=0 1 = lim lim a71771 •
— —
771 KX> 71 > 00
( ii ) Let
nx
/« (z ) = nx T x G [0, 1] , n 1. ( 1.8 )
+
Then each /n ( ) is continuous and
*
0 x = 0,
lim f n ( x ) =(\
. — > 00 Z ( 0, 1] ,
0
which is not continuous. Hence, (1.2 ) is not true in general,
( iii) Let
1
/n (x ) = X log ( nx + l ) XG [0, 1] . (1.9 )
n
Then
1 nx
/n ( z ) = 1 - nx 4- 1 nx -f 1 ’
/
and
lim f n ( x ) = x, x <E [0, 1] .
71
—
> 00
But ,
1 x G ( 0, 1]
( lim f n( x ) ) = lim f'n ( x ) =
-
n +oo / 71
—
> 00 Q x = 0.
Therefore, (1.3) is not guaranteed in general.
Uniform Convergence 123
( iv ) Let
1
n * e (0, -
n
),
f n ( x ) = nl (0 i ) ( x ) = <
0 x [o , 1] \ (0 , 1) .
fi
Then
l
f n ( x )d x = 1 , Vn 1
o
and
lim /n (x ) = 0 , Vx G [0 , 1] .
—
n > oo
Hence
I l
lim f n ( x ) d x .
n^ o o j/o
0=
lim f n( x )d x = 1
This means that (1.5) is not true in general.

^ Q —
n » 00
From the above example, we see that all 4 observations listed at the
beginning of this section could be false.
2 Uniform Convergence

Definition 2.1. Let ( X , d x ) and (Y , d y ) be two metric spaces and f n :
X -» Y be a sequence of functions ( n 1) . Also, let / : X -» Y be another
given function.
( i ) We say that /n ( ) pointwisely converges to / ( •) if for any given x G l ,
*
the following is true: For any e > 0 , there exists aniV = N ( s , x ) 1 ( may
depend on x ) such that
dvifnix ), f ( x ) ) < £, Vn N.
( ii ) We say that /n ( ) uniformly converges to / ( •) if for any e > 0 , there

*
exists an iV = N ( e ) > 0, ( independent of x G X ) such that

d Y ( f n( x ) J ( x ) ) < 6 , Vn N,x G X.
- ^
( iii) We say that {/n ( )}n i is uniformly Cauchy if for any e > 0, there
>
exists a n N = N ( e ) 1 (independent of x ) such that
d y ( f n( x ) 1 f m( x ) ) < £, Vn , m N, x G X.
Example 2.2. (i) Let X = [0 , 1] and Y = R with the standard metric. Let
nx
f n{ x ) = n 1.
nx A 1 -
Then as Example 1.1, ( ii) , one has
x 0,
lim f n ( x ) =
^-> oo
<
1
0
^
x = 0.
We claim that the above convergence is not uniform. In fact , for any x
in order to have ^,
O
nx 1
I f n ( x ) - 1| = 1 - n x A- 1 n x A- 1 < £ Vn N , ^
we need
Nx >
1
1=
1 — £
£ £
N>
1 — £
£X
Hence, as x is getting smaller and smaller , N has to be getting larger and
larger . Therefore, the convergence is not uniform. Now , if instead of [0, 1] 5
we let
1
X=
m {
m = 1, 2, • • U {0 } >
with the metric induced from the standard metric of R , then consider the
restriction f n x = f n • X -A Y , we have
\
n
fn( ~ = an m t f n ( 0) = 0,
m n A- m
which is the sequence appears in Example 1.1, ( i ) . Hence, the above argu-
ment shows that
lim an m
—
n > oo —
n > oo m
is not uniform in m. Consequently, we can show that the convergence
lim anm = 0
m —* oo
is not uniform in n , either ,
( ii) Let
f n{ x ) — X — —n log (nx 4- 1) x G [0, 1] , n 1.

Then
1 log ( n + 1)
\ f n( x ) - x\ - log ( nx + 1)
n n <
-> 0 n -» oc
uniformly in x G [0 , 1] . That is /n ( x ) converges to f ( x ) = x uniformly.

However ,
nx
1 x G [0, 1] .
nx + 1 nx + 1
-
From part ( i) , we see that f'n ( ) is not uniformly convergent on [0, 1].
( iii) Let
fn { x ) = nl (0ii ) (x ) , xG [0, 1] .
We see that /n ( - ) convergent pointwise on [0, 1] , but not uniformly. In fact ,
since the limit function is f ( x ) = 0, in order /n ( - ) to converge to / ( •) , it
-
is necessary that /n ( ) is uniformly bounded . But the above /n ( - ) is not
uniformly bounded .
The following result gives us some more feeling about the pointwise
convergence and the uniform convergence.
Proposition 2.3. Let f : R -» R , and for any a G M, define the shift

function as follows:
/ a (z ) = f ( x - a ) x G E.
Then / ( •) is ( uniformly ) continuous on R if and only if for any sequence
-
an converging to 0 , the sequence / an ( ) of shift functions is ( uniformly )
convergent to / ( •) .
Proof Suppose / ( •) is continuous. Then for any an — ¥ 0,
lim f a n ( x ) = nlim f ( x - an ) = f ( x ).
n —> oo —> oo
Conversely, if / ( •) is discontinuous at x, then there is a sequences an con-
\
vergent to 0, such that for some e > 0 ,
< \ f ( x — a n ) — f ( x ) \ = | /° (x) - / (x)|.

e n
This means that / ( - ) is not convergent to / ( ) , a contradiction .

Gn •
Next , we look at the uniform continuity case. First , suppose / ( •) is

uniformly continuous. Let a n be an arbitrary sequence that is convergent
to 0. By the uniform continuity of / ( •) , for any e > 0, there exists a
S = 6 ( e ) > 0 , independent of x , such that
\ f { x ) - f { x ) \ < £ , Vx , x e R , \ x — x \ < 5.
Now since a n —* 0 , there exists an iV 1, depending on 5 > 0 such that
\d n \ ^ Vn N .
"
Hence, for any x £ R , by taking x = x — a n , one has

^
^
\P n( x) f {x)I = I f ( x) - f ( x)I < £,
~
proving the uniform convergence of / an ( - ) .

Conversely, suppose / ( •) is not uniformly continuous. Then there exist
two sequences xn , a n such that for some £ > 0, Q
\ f ( x n — a n ) - f ( x n ) I ^ e0 n 1.
Then , it leads to that fan ( •) is not uniformly convergent to / ( ) , a contra- •
diction.
We now present the following result which will be used below.
Theorem 2.4. Let ( X , d x ) and ( Y , d y ) be metric spaces, and E C X ,
fn, f : E Y , n 1 . Suppose /n ( - ) converges to / ( ) uniformly. Then •
{ /n ( - ) }n i is uniformly Cauchy. Conversely, if { /n ( - ) Wi is uniformly

^
Cauchy and (Y , d y ) is complete , then { f n ( - ) } n i is uniformly convergent .
^
-
Proof . Suppose f n ( ) is uniformly convergent to / ( •) . Then for any
e > 0, there exists an N = N ( e ) 1 such that
X.
d Y ( f n( x ) , f ( x ) ) < £
Consequently, for any n , m N , one has
Vn
^ N, x G
^
( f n{ x ) , f ( x ) ) X.
d Y ( f n( x ), f m( x ) ) < dY + d Y ( f m ( x ) , f ( x ) ) < 2e, V x
Thus, { /n ( - ) }n is uniformly Cauchy.
^ i
Conversely, let { /n ( ) }n i be uniformly Cauchy. Hence, for any e > 0,

^
*
there exists an N 1 such that

d Y ( f n( x ) , f m( x ) ) < £ Vn , m N , Vx £ X .
Then for each x £ X , { f n ( x ) } n i is a Cauchy sequence in Y . Since (Y , d y )
is complete, we may define ^
f (x) = nlim f n( x ) x e X.
—> oo
•
Consequently, for any x G X , we can find an N = N ( e, x ) N such that

N ( s , x ).
^
d Y ( f n( x ) , f { x ) ) < e, Vn
This leads to that for any N
d Y ( f n{ x ) , f ( x ) ) d Y ( f n( x ) , f f f ( £tX ) { x ) ) + d y ( x ) , f ( x ) ) < 2s .
This shows that /n ( ) converges to / ( •)
* uniformly.
The following is a main result of this section.
Theorem 2.5. L e t ( X , d x ) &n d ( Y , d y ) b e metric spaces, and E C X ,

-
f n , f : E -± Y , n l . Suppose f n ( ) converges to / ( •) uniformly. Then the
^
following statements are true.
( i) Suppose xo is a limit point of E , and
lim f n( x ) = Ln , n
E B x^ xo- ^1
for some Ln G Y . Then , { Ln }n
i is Cauchy in Y . In the case that ( Y , dy )
^
is complete , there exists some L eY such that
lim lim f ( x )
Ln = L = EBx
n-» oo —t x o
which amounts to saying that
lim lim f n( x ) = lim lim f n { x ) .
n^ ooE 3 x -+ x0 EBx -+ xo n-¥ oo
( ii) Suppose for each n 1, /n ( - ) is continuous. Then / ( •) is continuous.

Proof , ( i ) By Theorem 2.4 , for any e > 0, there exists an N = N ( s) 1
such that
d Y ( f n( x ) , f m ( x ) ) < £ , Vn , m N , x E.
Also, for xo , a limit point of E , for any n
^ G
1, there exists a Sn = 5 ( S , X Q
n ) > 0 such that
d y ( fnipt ) i L f f ) , Vx E D B x ( x0 , Sn ) .
^ £
Then for any n , m > N , we choose x n m G B x ( xo , Sn A Sm ) H E t o get the

following:
d y ( Ln , Lm ) nm ), m ( nm ) )
^ d y (+Tn , fn(/{ Xnrn( xnm) ) )d, Lm
m
d y (/ (
*
) 3,
^ n
£
/ ^
which shows that { Ln}n i is Cauchy. Consequently, if ( Y, d y ) is complete,
we have some L e Y so that ^
lim d y ( Ln , L ) = 0.
n — K x)
Then for any e > 0, there exists an N

^ N such that
dy ( Ln,L) < £, Vn N.
Hence, for any x G E fl Bx ( xo , (5 ) , we have
^
d y ( f ( x ) , L) d Y ( f ( x ) , f f i ( x ) ) + d Y ( f f r { x ) , L f i ) + d Y ( L f r , L ) < 3e,
which proves our claim.
(ii) It follows from ( i ) with L n = f n { xo ) and L — f ( xo ) .
Uniform convergence of a continuous function sequence is a sufficient
condition for the limit function to be continuous, and it is not necessary.
Here is a simple example.
Example 2.6. Let

nx x [0 , n-1] ,
f n( x ) = 2 — nx x [n ~ \ 2n-1] ,
0 x [2n-1, 1] .
-
Then /n ( ) is continuous and f n -> 0 pointwise, but not uniform. Whereas,
the limit function / ( •) is continuous.
Thus, a natural question is what kind of convergence for the sequence of

continuous functions will exactly lead to the continuity of the limit function?
We now introduce the following notion.
Definition 2.7. Let ( X , d ) be a metric space and f n , f : X * R. We say —

that /n ( ) is quasi-uniformly converges to / ( •) if /n ( ) » / ( •) pointwise on
* *
—
X , and for any £ > 0, iV > 1, there are at most countably many open sets
{ Gi }i l , and corresponding indices such that
^
XC \J Gi i
^l
i
and
|/n, (x) - / ( x )| < £ , Vx e G i, i 1.

We have the following interesting result .

Theorem 2.8 . Let ( X, d ) be a compact metric space and fn : X -» R be a
sequence of continuous functions convergent pointwise to some f : X > R. —
-
Then f n ( ) converges to / ( •) quasi- uniformly if and only if f is continuous.
Proof . Sufficiency. For any e > 0 and iV > 1, by the uniform continuity
of / ( •) on the compact metric space X , there exists a S£ > 0 such that
l / (*) - f ( y ) \ < £ > Vx, y G X, d ( x , y ) < S£ .
By the pointwise convergence of /n ( ) to / ( ) , for each x
« • X , there exists
an Nx N such that
I I N J X ) - f i x )| < £.
Further , by the continuity of / NX { •) > there exists an ax G (0, 5 ) such that
£
\ fNx ( y) fNx { x )\ < £,

- Vy X , d ( x , y ) < ax .
Hence, for any y G X , d ( x , y ) < ax ,
.
! / ( ?/) - f N ( v ) I < I f i v ) - / (*) I + 1 / (*) - /JV,(*) |
+ I //Vx (x) - /Arx (y)| < 3e.
Clearly, the family {Bax ( x ) x G X} is an open cover of X. By the
compactness of X , we may assume that
k k
X C \J Baxi { Xi ) = \J Gi .
=1
i i =1
Also, let rii = Nx . . Then
f i v ) I < 3e,
\ f n, ( y ) - V y e X n G i,
- —
This means that /n ( ) > / ( •) quasi- uniformly.
Necessity. Fix any x G X . For any £ > 0 , by the definition of quasi-
uniform convergence, there exists an open set G containing x and an N
such that
\ f N { x ) - f ( x )\ < £,
and
\ f N ( y ) - f ( y )\ < s , V y e G.
By the continuity of /JV , there exists a 5 > 0 such that
\ f N ( x ) - f N ( y ) \ < £ , Vy G , d { x , y ) < 6 .
Then for any y G , d {y , x ) < 5 ,

\ f ( x ) - f (v ) \ < \ f ( x ) - f N ( x ) \ + \ f N ( x ) - f N ( y ) \ + \ f N ( y ) - f ( y ) \ < 3e.
Thus, / ( •) is continuous at x .
We have seen that uniform limit of continuous function sequence on a
compact set is continuous (Theorem 2.5, ( ii)) ; the pointwise limit of con-
tinuous function sequence on a compact set is continuous if and only if the
convergence is quasi- uniform ( Theorem 2.8) . Now, we would like to know
when a pointwise convergent continuous function sequence is convergent
uniformly. Here is such a result.
Theorem 2.9 . ( Dini ) Suppose ( X , d ) is a compact metric space and fn :
X -> R is a sequence of continuous functions. Suppose
= f {x)
lim f n ( x ) x e X,
—
n > oo•
with f : X —> R being continuous. Further , suppose that

fn ( % ) fn-1-1 ( E ) * x e x. ( 2.1 )
Then the convergence of /n ( ) to / ( •) is uniform.
*
^ *
Proof . Let
9n { x ) = f i x ) - /„ ( x ) x e G. ( 2.2 )
Then each <?„ ( •) is continuous,
9n ( x ) s„+ i (a:) 0, Vx X, n 1 ( 2.3)
and
lim 3„ ( x ) - 0. ( 2.4 )
n oo—
Now , for any e > 0, let
On { e ) = 9n 1( { -oo, e)) = {x X \ g n ( x ) < e}. ( 2.5)
Then by the continuity of gn ( - ) , 0n (£ ) is open. Moreover, by ( 2.3) , we have
On ( e ) C On + i (e) Vn 1. ( 2.6)
On the other hand , by ( 2.4 ) , for each x G X , there exists an N such that
g n ( x ) < e, Vn N, ( 2.7)
which means that x G On ( e ) . Hence,
OO
x c ( j on( s ). ( 2.8 )
n= l
Now , by the compactness of X , we have a finite subset of { On ( e ) | n 1}

covering X . Let N be the largest index of this finite set . Then by ( 2.6 ) , we
must have
0 N ( e ) 2 X.
This means that
o < e,
f ( x ) - f n( x ) = gn( x ) gN ( x ) Vn
^ iV, x E X.
-
Hence, the convergence of f n ( ) to / ( •) is uniform.
It is clear that condition ( 2.1) can be replaced by
fn( x ) f n+l ( x ) x E X.
We also note that in Dini’s Theorem , X can be any compact metric space,
in particular , it could be any bounded and closed set in Rm , and it does
not have to be the closure of some domain.
Exercises
1. Let ( X, d ) be a metric space and f n : X > R be a continuous function —
- —
for each n 1 such that f n ( ) > / ( •) uniformly for some / : X > R.
^ —
Suppose xn E E such that xn x e E . Then
lim f n ( xn ) = f ( x ) .
n too—
2. Let ( y, d ) be a metric space and a nm G 7, n , m ) l , such that
lim anm
—
n > oo
=6 7715 uniformly in m
^ 1,
and
771
lim a71771
—» 00 —^ 71 *
Then cn is Cauchy in y and in the case that Y is complete, there exists a

c such that
71
lim lim anm
—» —*
00 771 oo — cn — c — lim
= lim n > oo
— = lim lim a
— — 771 OO 771 > 00 71 > 00
nm’
3. Let ( X, d ) be a metric space and / , pn : X —> M such that f n —> f

n •
and g n — y g uniformly, for some /, g : X —> R . •
( i) Show that for any a , /3 E R , a /n + (3gn -» a / -f (3g uniformly.

( ii ) Prove or disprove that if an , /3n E R with an —» a and /3n —» /3 for
some a , /3 G R , then an /n + f3ngn a / + fig uniformly.
( iii) Show that in general “ fngn fg uniformly” is not true. But it is

true if both f n and g n are uniformly bounded .
4. Let ( X , d x ) and ( Y , d y ) be metric spaces and f n : X —> Y be
—
uniformly convergent to some / : X > Y . Let g : Y -A- R be Lipschitz
continuous. Prove or disprove that g o f n converges to g o f uniformly.
5. Let
1 1
f n( s ) =
0 0 <x<n -f 1
or x > —
n
1 1
• 2*
—x
n -fl ^ n
sin
^ ^
Show that fn (’ ) converges to a continuous function , but the convergence is
not uniform.
6. Let / : [0, 1] -> 1 be a continuous function. Show that there exists a
—
sequence of step functions fk : [0 , 1] >• R such that //c ( ) converges to / ( •) -
uniformly. Is the result true if the domain [0, 1] is replaced by ( 0 , 1) ? Why?
7. Let ( X, d ) be a metric space which is totally bounded , not necessarily
complete, and f n : X R be a sequence of uniformly continuous functions.
Suppose
f n( x ) f n+\ { x ) x G X, n > 1
and
lim f n { x ) = f ( x ) xeX
n —¥ oo
with some uniformly continuous function / : X —> R. Show that the
convergence is uniform. What if each /n ( ) is just continuous? Give your
*
reason.
8. Let ( X , d ) be a compact metric space and f n : X -+ R converge
to —
/ : X > R uniformly. Then f n converges to / quasi- uniformly. Is the
converse true? Why?
3 The Metric of Uniform Convergence
We now look at the uniform convergence of functions from a different view-

point .
Proposition 3.1. L e t ( X , d x ) a n d (Y , d y ) b e complete metric spaces, let

B ( X ; Y ) = { f : X -> Y \ / ( ) is bounded } ,
•
—
C ( X ; Y ) { / ( •) G B ( X ; Y ) | / ( ) is continuous } ,
•
and define
doo ( f , g ) = sup d Y ( f ( x ) , g { x ) ) ,
x X
(i) doc i s a metric on B ( X ; Y ) , which is called the sup norm metric or

L°° - metric. A sequence { f n }n 1 m B ( X ; Y ) converges to / ( •) G B( X \ Y )
^
in the metric d 00 if and only if f n ( ) converges to / ( •) uniformly.
m
( ii) The set C ( X ; Y ) is complete as a subspace of B ( X ; Y ) , i.e., any

Cauchy sequence {/n ( - )}n i in C ( X ; Y ) is convergent to some / ( •) G
C ( X ; Y ) under doo - ^
Proof . The proof of ( i) is straightforward.
(ii ) Let {/n ( )}n i be a Cauchy sequence in C ( X ; Y ) . Then by (i) , /n ( )
*
^ -
converges to some / ( •) G B ( X ; Y ) . By Theorem 2.5 ( ii ) , / ( •) G C ( X ; Y ) .
We now introduce a new notion .

.
Definition 3.2 Let ( X , d x ) and (Y , d y ) be metric spaces. A family T C
C ( X \ Y ) is said to be equicontinuous at xo G X if for any e > 0, there
exists a <5 = S ( xo , e ) > 0 , independent of the particular / ( •) G X , such that
d Y ( f ( x ) , f { xo ) ) < £ V d x ( x , x 0 ) < S , / ( •) G T .
The family T is said be uniformly equicontinuous if for any e > 0, there
exists a S = 6 ( e ) > 0 such that
d v ( f ( x ), f ( x ) ) < £ V d x ( x , x ) < 6 , / ( •) 6 7.
The following result collects some interesting relations among some rel-
evant notions.
Theorem 3.3. L e t ( X , d x ) be a compact metric space and (Y , d y ) be a
—
complete metric space. Let fn : X > Y be continuous such that /n ( )
• *
converges to / ( •) pointwise on X . Then the following are equivalent:

-
( i ) The sequence { f n { ) } is equicontinuous on X .
( ii) Function / ( •) is continuous on X and the convergence /n ( - ) » / ( •) —
is uniform on X .
Proof , ( i) => (ii): Since ( X , d x ) is compact , the equi-continuity of
{/n ( )} is uniform in x G X . That is, for any e > 0, there exists a 5 > 0
*
( only depending on e > 0 ) such that the following holds:

d Y { f n { x ) , f n { x' ) ) < £ , Vx, x' G X, d x ( x , x' ) < 5, n 1.
—
Letting n > oo , we obtain
d Y ( f ( x ) , f ( x0 ) ) e. (3.1)
Thus , / ( •) is continuous.
Next , by the compactness of X , for the above S > 0 , there are finitely
many xi , • • • , X k such that
k
X C
i =1
and
d Y ( f n ( x ) , f n( X i ) ) < £ , Vx e B ( x i , 5 ) 1 i k Vn 1. ( 3.2 )
Further , by the pointwise convergence /n ( - ) —> / ( •) , for the above e > 0,
d Y ( f n( X i ) , f ( X i ) ) < £ Vn N, 1 i < fc. ( 3.3)

Consequently, for any j/ 6 l, letting 7/ G B ( x i , S ) for some 1 i fc , we
have ( noting ( 3.1) , ( 3.2 ) , and ( 3.3))
d Y ( f n( y ) , f ( y ) )
^ d Y+( df nY( (yf) (, xf ni ){,X/i () y) )+) d<Y3(ef,n( X i )Vn, f ( X>i )AT) .
This means that the convergence /n ( ) > / ( •) is uniform. - —
( ii) ( i ): Let Xo G X . For any e > 0, there exists a 5 > 0 such that
d Y ( f ( x ) , f ( xo )) < £ , Vx B ( xo , 5 ) ,
and there exists an N 1 ( independent of x E X ) such that
^
{ f n( x ), f { x ) ) , N, e X.
dY <£ Vn X
Then for any n iV , and x i? ( xo , 5) ,
d Y { f n ( x ) , f „( xo ) )
d Y ( f n ( x ) , f ( x ) ) + d Y ( f ( x ) , f ( x0 ) )
+ d Y ( f ( x o ) , f n ( x o ) ) < 3e.
—
Also, for each n = 0, 1, • • • , iV 1, there exists a 6n > 0 such that
d Y ( f n ( x ) , f n ( x0 ) ) < £ x G ( xo , 5) .
^
Then by taking
5 = min{5, <50 , $i , • • • , <5AT _ I}
we see that for any x G B ( X Q , S )

d Y ( f n { x ) , f n { xo ) ) < 3e
- -
proving the equicontinuity of {/n ( )} near xo Then by the compactness of
X , we obtain ( i) .
We see that in the above, ( ii ) => (i) does not need the compactness of
X.
The following result can be regarded as an extension of Bolzano-
Weierstrass Theorem to a family of functions , which gives a criterion for
compact subsets in C( X ; Y ) .
Theorem 3.4. ( Arzela-Ascoli ) Let ( X , dx ) and ( Y, d y ) be compact met-

—
ric spaces. Let T { f\ : X — Y, A G A} be a family of functions. Suppose
.
that T is equicontinuous Then there exists a sequence { f\n ( •) } Q. which
converges to some continuous function f : X -» Y uniformly on X .
Proof . Since X is compact , by Proposition 4.18 of Chapter 1, there

exists a countable set Xo = {xn , n ) 1} C I which is dense in X , i.e.,
Xo = X .
Now , for x\ G Xo , since { /A ( # I ) | A G A} is a sequence in the compact met-
ric space Y, there exists a subsequence { f\k ( x\ ) } C Y that is convergent to
some point denoted by f ( xi ) G Y. Next , since £ Y is a sequence
in the compact metric space Y , we can further extract a subsequence of
{ f\k ( x 2 ) } which is convergent to some point f ( x 2 ) G Y. By relabeling, we
may assume that { f\k ( x 2 ) } itself is convergent to some point denoted by
/ (x 2 ) . Continue this procedure, we obtain a sequence { f\k ( ) } Q T such
that
-
lim
—
fe > 00
h k ( xn ) = f ( xn ) VXyj G X0 , -
( 3 4)
where f ( xn ) is some point in Y depending on xn . Next , we claim that f\k ( •)
is uniformly convergent on X . In fact , for any e > 0, by the equicontinuity
of T , we have a S > 0, independent of k 1, such that
d v( fxk ( x ), fxk ( x)) < £ X , X G X, d x { x , x ) < J, fc 1 .
By the compactness of X , there are points X i, - , x n G X0 , with n de-
* *
pending on 5 > 0, such that
XC
1
IJ= B { X i , 5 ).
1
By ( 3.4 ) , there exists an N 1, depending on e > 0 ( and n

eventually depends on e > 0 ) , such that ^ ^ 1, which
fXj {Xi ), fXk ( Xi )) < £ j,k
Now , for any x
dy(
G
,
X , let X i G X 0 such that

^ N, 1 i n. ( 3.5)
d x ( x i , x ) < S.
Then for j , k N,
d Y ( fXj ( x ) , f X j (x, ) ) + d Y ( f X j ( x t ) , f X k (a;* ) )

d y ( h j ( x ) , /A , ( X ) )
+ d Y ( fXk { x i ) , f X k ( x ) ) < 3e.
-
This shows that { f\k ( ) } is Cauchy in { C ( X ] Y ) , doo ) . Hence, we may
denote
f ( x ) = lim fXk { x ) uniformly in x G X .
k toc —
Then, by Theorem 3.3, it is necessary that / ( ) is continuous.
•
A well-known form of Arzela-Ascoli Theorem is the following corollary.

Corollary 3.5. L e t C C R n b e a compact set , and f n : G -A- Rm b e a s e-
quence of functions. Then { /n ( - ) }n i has a subsequence which is uniformly
convergent if the sequence is uniformly bounded and equicontinuous.^
Next result gives an approximation of continuous functions by polyno-
mials.
Theorem 3.6. ( Weierstrass ) If f : [a , b ] -> R is continuous. Then there
exists a sequence of polynomials Pn ( ) such that *
—
n > oo
-
lim ll -Pn ( ) - / ( Olloo = zeSUP
[a , &]
\ pn ( x ) - f ( x ) \ = 0.
Proof . First , without loss of generality, we assume that [a , b ] = [0 , 1].
Further , we assume
/ (0) = / (1) = 0.
In fact , if necessary, instead of / ( ) on [0 , 1] , we may consider
•
g ( x ) = f ( x ) - / (0) - x [ f ( 1) - / (0)] X G [0 , 1].
If such a g ( ' ) can be approximated by polynomials pn {‘ ) uniformly, i.e.
—
n * oo
-
lim ||g ( ) - Pn ( )||oo - = 0.
Then
lim
71
— OO
! ! / ( •) - / (0) - x [/ ( l ) - / (0)] - Pn ( - )||oo = 0,
-
and pn ( ) + / ( 0) + x [ / (1) - / ( 0 ) ] is a polynomial.
Next , we redefine / ( •) to be zero outside of [0, 1] . Then / ( •) is bounded
and uniformly continuous on M. Let
M = sup | / ( x ) |.
zG [0 , l ]
Define
Qn ( x ) = Cn { 1 - Z 2 )" n
^1
with cn being chosen so that
Consider
i: Qn ( x ) dx = 1.
ip ( x ) = (1 - x 2 ) n - 1 4- nx 2 , xG [0, 1] .
We have </? (0 )= 0, and
_
/ (x ) = — n ( l — x 2 ) n 12x + 2nx = 2nx [l — (1 — x 2 ) n * ]
( 0.
Hence, £> ( x ) 0 for all x G [0 , 1] , which implies
(
(1 - x 2 ) n 1 - nx 2 , x G [0, 1] .
Consequently,
/* (1 — x )ndx
l I
1
-1
(1 - x2 )ndx = 2
0
(1 — x )ndx 2
2 2
^ /* (
l
4 1
— nx )dx = 2 x —
^ ^^ x3
2
2 1 >
This yields
0 3y
^
Thus, for any given 5 G (0, 1) ,

Cn — J^ (1 — x ) ndxSj
2
< y / n.
/n ( l - x 2 ) n / ( l - ^2 ) n
which implies
Qn { x ) \ \
^ 5 x 1,
lim Qn ( x ) = 0, uniformly in x G [5, 1].

71
—
KX)
Note that f ( x ) = 0 for x £ [0, 1] . Thus, we have

/* 1 rl
— x
Pn { x ) =
J f ( x + t )Qn ( t ) dt =
J f ( x + t ) Q n ( t )d t
f ( t )Qn ( t - X )d t , x e ( 0, l ] .
0
Clearly, Pn ( - ) is a polynomial. Now , for any > 0, there exists a S > 0
such that
|/ ( x ) - f ( x ) \ < | Vx, x e [0, 1] , \ x - x \ < S.
Then for any x e [0, 1] ,
\ Pn { x ) - f ( x ) I = I J \ ( +t ) f x ~ f ( x ) ] Q n ( t )d t
:
/\
(J
f (x+t) - f ( x ) \ Q n ( t )d t
«c 2 M Qn ( t ) dt + Qn ( t ) dtj +|
J Q n ( t )d t
= 4M i: Qn ( t ) dt +
4M \/n ( l - <52 )" + ^

|< e
J Qn ( t ) dt
as long as n large enough. This proves the theorem .

The following gives an application of the above theorem.
Example 3.7. Let / : [a , b] —> R • be continuous such that for any polyno-
mial p( - ) ,
[
f ( x ) p( x ) d x = 0
Ja
Then / ( •) = 0. To show this, we choose a sequence of polynomials p n ( •)
such that
1
11 / ( 0 - ftiOllooC - It
Vn 1.
By our condition , we have
f \ f ( x ) \ 2d x = f f ( x ) [ f ( x ) - p n( x )]d x
Ja da
n — —
M(
J
[
a
l / (*) ll / (») — p n ( x ) \ d x — b a) 0.
Thus, / ( •) = 0.
A generalization of the above Weierstrass Theorem is the following.
.
Theorem 3.8 ( Miintz-Szasz ) Let 0 = Ao < Ai < • • • . In C [0, 1] , con-
sider the following subset :
11( A ) = span { xXk | k = 0, 1, 2, • • • }.
Then 11( A ) is dense in C [ 0, 1] if and only if
o
° 1
E Xri - * '
The proof of this theorem is beyond the scope of this book and we omit
it here. Weierstrass Theorem is a special case of the above with Ak k . —
Exercises
—
1. Let fn : [0 , 1] » R be a sequence of continuously differentiable
•
functions such that

/n ( 0 ) = 0, | / («)| < 1 ;
Vn 1, x G ( 0, 1) .
-
Suppose further that f n ( ) is convergent pointwise to some function / ( •) .
Then /n ( - ) converges to / ( •) uniformly.
2. Let f n ( x ) = xn , x G [0 , 1]. Show that for any S G ( 0, 1) , the sequence
f n is equicontinuous on [0, 5] , but the sequence is not equicontinuous on
[0, 1] .
3. Let fn : [a, b ] -* R be continuous and uniformly bounded. Define
px
0nO*O = fn { s )ds , x G [a , 6] , n 1.
-
Then pn ( ) is equicontinuous.
4. Let / : [0, 1] -» R be continuous such that
[
x k f ( x )d x = 0 ,
Jo
Vfc 5.
^
Then / ( x ) = 0 for all x G [0, 1].
5. Let / : [0, 1] -* R be continuous such that
[o
x2k f { x ) d x = 0 , VJb l .
^
Then / ( x ) = 0, for all x G [0, 1] .
6. Let K : [0, 1] x [0 , 1] R be continuous. Let f n : [0 , 1] —> Rn be
continuous and uniformly bounded . Define
9n ( x ) =
/ *< x , y ) f n { y )d y ,
Then the sequence {<?n ( )} is uniformly bounded and equicontinuous.
*
x G [0, 1] , n 1.
4 Limit Theorems
In this section , we consider the order exchanges. The first result is the
order exchange of the limit and integration.
Theorem 4.1 . Let G C Mm be a bounded domain with \ dG \ = 0 , and

f n : G -A- R b e a sequence of Riemann integrable functions. Suppose that
fn ( •) uniformly converges to / ( •) on G . Then / ( •) is Riemann integrable
and
lim
— [
n » oo J G
fn ( x )dx = [
JG
f ( x )d x = [ lim
—
J G n > oo
f n ( x )d x .
-
Proof . Since /n ( ) converges to / ( •) uniformly, for any e > 0 , there
exists an iV 1 such that
^
\ f n ( x ) ~ / (® ) | < £ , Vn > N, x e G.
This implies that
f n( x ) ~ £ < f { x ) < f n( X ) + £ , N, Vn x G G.
Consequently, for any a G
^
Am , (see Chapter 4, (1.26 ) )
S - ( f n ) - e \Ga \ <S ~
(f) < S+ ( / ) < S+ ( /„ ) + e|Ga |
which leads to
0 5+ ( / ) - S - { f ) < S+ ( f n ) - S ~
( f n ) + 2e|G“|.
Hence
lim S+ ( / ) - S- ( / ) < 2e|G|.
I a|| >0 L
| “ J
Since e > 0 is arbitrary, one sees that /( )
• is Riemann integrable. The
above also implies
I[
1 JG
f n ( x )d x —[ f ( x )d x 2e \ G \ n N,
JG
The following is a consequence of the above theorem and the Dini’s
Theorem.
Corollary 4.2. ( Levi’s Lemma ) Let G C Rm be a bounded domain with
\ dG \ = 0. Suppose fn : G -» R is a sequence of continuous functions such
that
lim f n ( x ) = f { x ), xeG
—
n > oo
with f : [a , b ] —> M being continuous, and

f n( x ) /n + lW x e G, n 1.
Then
lim
n^°° /
JG
f n ( x )d x =
JG
[ f ( x ) d x.
Proof . By Dini’s Theorem , the convergence of /n ( ) to / ( •) is uniform *
(since G is compact ) . Hence, Theorem 4.1 applies.

The following is another corollary.
Corollary 4.3. ( Fatou’s Lemma) Let G C Rm be a bounded domain
with \ dG \ = 0. Suppose f n : G -± [0, oo ) is continuous such that for each
k l,
>
gk ( x ) = ninfk f n ( x ) x eG
^
is continuous on G , and
lim f n { x ) = f ( x) x eG
—
n » oo
is also continuous. Then
lim f n ( x ) d x lim / f n ( x )d x.
Gn —¥ oo n-*oo JG
Proof . Since
9i ( x ) Pfc ( x ) 0*+ i (a;) xeG

and
lim gk ( x ) = lim f n { x ) = f ( x ) x e G.
—
k too
—
n » oo
By Corollary 4.2, we have
— J/G
lim
k >oo
g k ( x )d x ==
JG / f ( x )d x.
On the other hand
5fc ( z ) /fe (x ) Vz e G
which leads to
/
JG
g k ( x )d x /
JG
f k { x )d x.
Hence
[
JG
f ( x ) dx = lim
oo /
JG
Qk ( x ) dx lim / fn ( x )dx ,
n-» oo JG
proving our result .
We know that in Example 1.1, ( iv ) ,
fn { x ) = nl (0 i ) (x ) , x G [0, 1]
which pointwisely converges to 0. But the convergence is not uniform.
Therefore, the order exchange of limit and integration is not necessarily
true. On the other hand , the uniform convergence of /n ( ) to / ( •) is not -
necessary for the order exchange of limit and integration. Here is an exam-
pie.
Example 4.4 Let .

f n ( x ) = 1(0, JL ) (® ) x G [0, 1] .
Then
lim fn ( x ) = 0, x G [0, 1]
—
n > oo
but the convergence is not uniform . However ,

l
'
0
fn ( x ) dx =
1
n
>0=
From the above, we may ask can one relax the condition that ensures
f 0dx .
the order exchange of the limit and integration? The answer is yes and it
can be stated as follows.
.
Theorem 4.5 ( Arzela ) Let G C Rm be a bounded domain with \ dG \ = 0.
Let fn : G -» R be Riemann integrable and uniformly bounded . Suppose
lim /„ ( x ) = f ( x ) x G G
—
n > oo
for some Riemann integrable function f : G -* M. Then
— JG[ Ifn { x ) - f ( x )\d x = 0,

lim
n too
( 4.1)
which leads to
n —* JG[ fn( x )dx = JG[ f ( x )d x

lim
oc
( 4.2 )
Proof . First , let fn : G -» R be a decreasing sequence of bounded

functions ( not necessarily Riemann integrable) such that
lim f n ( x ) = 0, \/x e G .
71 —> OC
By the proof of Proposition 1.9 of Chapter 4 , for any e > 0 and any n 1
there exists a continuous function gn£ : G > R such that —
0 gen ( x ) f n( x ) VXGG
and
S~ ( f n ) < JGgSn ( x )dx + ( 4.3)
For each n 1, set
= min { (x) ,
^ - - - , c£ (x)} x eG.
Then
0 h* ( x ) fl£ ( x ) f n( x ) x eG
and { hn£ }n i is decreasingly approaching to 0 on G as n -> oo. Hence, by
^
Dini’s Theorem , the convergence is uniform . Consequently,
Now , we claim that

lim
—
71 HX)
L hn£ ( x ) dx = 0.
o < $- ( /„ )
In fact , for each i
L
= 1, 2, • • , n , one has
£
hn ( x )dx + e ( l - 2 ") n 1. ( 4.4)
o < s£ ( z ) = + gen ( x ) - gi ( x ) £
5i ( a?) + max{gf ( x ) , • • • , fl£ (x )} - g f ( x )

71
— 1
< g t ( x ) + Y2 { max{|
ff ( x ) ' ••• gn£ ( x ) } - g £j ( x ) )
3= 1
which leads to
71
— 1
0 ffn ( x ) hn ( x ) + ( max{0j(x ) , - - - g n£ ( x ) } - 9 j ( x ) )- ( 4.5)
3=1
On the other hand , by

max{ ( x ) , • • • , s£ ( x )} m a x{ f j ( x ) , • • • , /„ ( x )} = /, (x ) ,
^ xeG
one has
S
^^ IG (
Thus, by ( 4.3) ,
^ maX (X) ’ " ' ’ 9n ( X ) } ~
9 j ( x ) } dx + /
JG
g £j ( x ) dx .
Jf ( ^ - ,
J^ gej
max
Consequently, combining ( 4.5)-( 4.6 ) , we obtain

x ),
( x )dx <
gn£ ( x ) } - g £ ( x ) dx
| T , 1 ^< j n.
( 4.6)
.
r
—
n 1 «
L gn£ { x ) dx
JG K ( x )dx + ' 52
3 ^
^ j =
JG h£n ( x ) dx + e l ( _
1
2n 1 /
Hence, ( 4.4) holds and we obtain

lim S ( /n )
—
n > oc
= 0.
Now , we prove our theorem. Without loss of generality, we assume that
0 fn ( x ) M, Vx G G, n 1,
and / ( x ) = 0 (otherwise, we consider | /n ( x ) — / ( x )| which is Riemann
integrable) . We define
pn ( x ) = SUp fn+k ( x ) x G G, n
^ 1.
Note that pn ( - ) is not necessarily Riemann integrable. Clearly,
0 f n( x ) Pn( x ) x G G, n 1,
and {Pn ( , )}n i is a decreasing sequence of bounded functions that converges
^
to 0 pointwise. Hence, by what we have proved ,
lim 5 ( pn ) = 0.
71
—» 00
Consequently,
0 lim fn ( x ) d x lim S ( pn ) = 0,
—
n > oo J Q —
n > oo
proving the theorem.

With the above Arzela’s Theorem , we could relax the conditions in the
Levi’s Lemma and Fatou ’s Lemma. Details are left to the readers. Note
that the condition that / ( •) being Riemann integrable cannot be removed
as the pointwise limit of a sequence of Riemann integrable functions might

be not Riemann integrable. Here is an example.
Example 4.6 . Let Q fl [0 , 1] = {r^ | k 1 }. Define
1 x {rk | 1 k n}
f n( x ) = < otherwise.
Then each /„ ( •) is Riemann integrable, and
1 i Q n [0, l],
lim f n ( x ) = f {x) = <
1
^ 00 0 * e [0, l] \ Q.
This is a Dirichlet function and it is not Riemann integrable.
Next , we consider the situation of differentiation . We first look at the
following example.

sin n x
f n{ x ) = x e [0, 1] , n 1.
s/ n
Then
lim f n { x ) = 0 = f { x ), uniformly in x G [0, 1].
— n > oo
On the other hand
f n( x ) = y/ n c o s n x , x e [o , l ]
and f' ( x ) = 0. Hence,
( lim
—
\ n » oo
f n{ x ) )
/ —»oo fn { x )
= nlim
is not true,
( ii) Let
—+x , 2 x
^ 1? f ] ? n
^ 1.
Then
lim f n ( x ) = \ x \ = f ( x) x e [-1, 1]
and
—
n > oo
X 0 x = 0,
f n( x ) = -> g { x ) = \
A +a\
;2 >
sgn ( x ) X 7
^ 0.
Thus, the limit function ( ) / • is not differentiable at x = 0 , and of course
( nlim f n( x ) ) = nlim f n( x )
is not true.
\ —> oo / —> c 50
However , here is a positive result.
Theorem 4.8 . Let fn : [a , b] R be a sequence of continuously differen -
tiable functions such that
lim fn ( x ) = g ( x ) uniformly in x £ [a , b\
—
n > oo
for -
some g ( ) , and there exists an XQ E [a , b] such that
lim f n ( xo ) = L
n-¥ oo
holds. Then there exists a differentiable function /( )
• such that
lim /„ (x ) = f ( x ) uniformly in x [a , b] ,
—
n > oo
and
f' ( x ) = g ( x ) x e [a , b\ .
Proof . We have
fn { x ) = fn ( xo ) +
J XQ
t fn { t ) dt .
Now , passing to the limit on the right-hand side, making use of Theorem
4.1, we have
f ( x ) = L+ g { t ) dt .
J XQ
Then our conclusion follows.
The above amounts to saying that under conditions of Theorem 4.8,
lim
i
—> oo
/' (x) = f' { x ) = (V nlim
—> oo f n { x )/) .
We now look at the multi-variable case.
-
Proposition 4.9 . Let f n : G -A R be a sequence of differentiable func-
-
tions. Suppose that V /n ( ) is uniformly bounded on G . Then { /n ( )} JS *
equicontinuous.
Proof . By our assumption , we may let
l|V /ra (x)|| < M, Vx e G . ( 4.7)

Hence, by the mean value theorem , we have
I f n ( x ) - f n ( y ) I = || V /„ (x + 0 ( y - x ) ) (x - y)|| ( 4.8)
< M \\ x - y \\ \/ x , y e G.
-
From this , we see immediately that {/n ( )} l s equicontinuous on G.
Theorem 4.10. Let G C Rm be a bounded domain and fn : G —> R •
be a sequence of C 1 functions that converges to a function / : G » M. —

-
Moreover , the sequence ( V /n ( )} is uniformly bounded and convergent .
Then the following are equivalent :
-
(i) ( V /n ( )} is equicontinuous on G .
— -
( ii) / ( •) is C 1 , V /n : G > Mm converges to V / ( ) uniformly, and f n ( )
converges to / ( •) uniformly as well.
•
-
Proof , ( i) => ( ii) : Since {V /n ( - )} is equicontinuous, for any e > 0, there
exists a 6 > 0 such that
—
l|V /n (z ) V /„ ( y ) || < e , V x , y e G , \\ x - y \\ < 6. ( 4.9)
Now , for a given xo G G, suppose B ( xo , S ) C G (if necessary, we may shrink
S > 0 ) . By the mean value theorem , we have
f n ( x 0 -f y ) - f n ( xo ) = V /n ( x0 + 0ny ) • y , 9n e ( 0 , 1) . ( 4.10 )
Thus
||/n (a:o + y ) ~ f n ( xo ) - V /„ (a;0 ) • y||
IMI
||V /„ (a:o + 0n y ) - V /„ ( x0 ) || < e, 6 ( 0 , 1 ) , V|| y || < S .
On the other hand , by the uniform boundedness and equi-continuity of
{V /n ( - )}, we may let V /nfc ( ) converge to some <? ( •) . Then passing to the
*
limit along this subsequence in the above, we obtain

||/ (ô + y ) ~ f ( x0 ) - g ( x 0 ) • y \\
<e V || tf || < <5-
This implies that / ( •) is differentiable at xo and

V / ( xo ) = g ( x o ) .
Since every subsequence of {V /n ( xo )} admits a sub-subsequence converging
to the same point V f ( xo ) , one obtains that the whole sequence ( V /n ( xo )}
is convergent for all X o G G. Further , by the equi-continuity of ( V /n ( )} *
again , we obtain the continuity of V / ( - ) . Then by Theorem 3.3, one has

that the convergence V /n ( - ) —> V / ( - ) must be uniform . Since V /n ( ) is -
uniformly bounded , by Proposition 4.9 , {/n ( )} l s equicontinuous. Hence,
*
it follows from Theorem 3.3 again that the convergence of /n ( - ) to / ( •) is

uniform.
(ii) => ( i): This follows immediately from Theorem 3.3.

The above result essentially says that under certain conditions, one has
v [Lnlim> oo /„ (x|
) = lim V /„ ( x ) . ( 4.11)
— J n —> oo
To conclude this section , we present an interesting result .
Theorem 4.11. ( Riemann-Lebesgue Lemma) Let / ( •) E C ( [ a , &] ; R ) .
Then the following hold :
[h
lim / / ( x ) s i n nxdx = 0 , ( 4.12 )
71 - 00
^ Ja
fb
lim / f ( x ) cos nxdx = 0 ( 4.13)
71
—
>00 J a
Note that the integrands in the above are not convergent as n —> oo.
Thus, the above result gives us a quite different flavor. Actually, we have
a more general result of which the above is a direct consequence.
Theorem 4.12. ( Extended Riemann-Lebesgue Lemma) Let / ( •) E

-
C ( [a , 6] ; R ) and g ( ) E C( R ; R ) such that
g ( x + T ) = g( x ) Vx G M .
Then
b
lim
n-+oo
f ( x ) g ( n x )d x =T( —
1
J g ( x ) d x^)
J f { x ) dx . ( 4.14)
Proof . First , we assume that /( )• is continuously differentiable and

T
g ( x )dx = 0.
o
Let
Jo
g ( t ) dt
G( x ) = [
Vx E R.
Then G ( - ) is continuous on R and periodic with period T. Hence, it is

bounded. We obtain
Ja
f f { x ) g { n x ) dx = —n Jaf f { x ) dG { n x )
= ln [ f ( b )G ( nb ) - f ( a )G( na ) ] - I f /'
Ja
( x )G ( n x ) dx -> 0.
Next , we take a sequence of polynomials pk( - ) such that
b
f { x ) g ( n x )d x
/
rb
11 / ( 0 - p* ( - ) lloo < .
Since g ( - ) is continuous and periodic, we may let M — || ( - )||oo Then
p k ( x ) g ( n x )d x
^ ob
^ -
+ / \ g ( x ) \ \ f ( x ) - Pk ( x ) \ d x
a Ja Ja
M(b - a)
j / p k ( x )g ( n x )d x + k
' Ja
Hence,
fb M(b — a)
lim I f ( x ) g ( n x )d x /
-* J a
OO TO k
Since k could be arbitrarily large, we obtain
lim i rb
—
n too / f ( x ) g ( n x )d x
Ja
= 0.
Finally, for the general case, from the above, we have
1
lim g{nx ) - - d x = 0.
n —
*oo T
Then our result follows.
Let us look at the following example.
Example 4.13 Let . / ( •) £ C ( [a , 6] ; R ). Then
lim
*oo J a
TO - / *
/ (*) i sin n x \ d x = —^2 J/afb f ( x ) d x
Since g { x ) = | sinx| has period 7r , and
Jor
7T
1 1
n
sinxl d x = —
* Jo
sin x d x —2
= 7r .
Similarly,
n
lim
— f
oo J a
f { x ) I cosnx|dx = —f
^ Ja
f ( x )d x .
Exercises
1. Let f n : (a , 6) -> R be a sequence of continuous functions such that
|/n (*)| M, Vn 1, x £ (a , 6) ,
for some constant M > 0, and
lim fn[ x ) = f ( x ) x G ( a, b )
n — y oo
—
for some / : (a , b ) > R. Moreover , for any small e > 0, the above conver-
•
—
gence is uniform on [ a 4- e , b e ] . Then / ( •) is continuous on ( a , b ) and
b
n
lim
— oo a
f n ( x )d x = Ja
[ f ( x )d x.
2. Let / : [0, oo ) —> [0, oo ) be continuous monotone increasing. Find

nx
n
lim
— yoc 1+ nx ) d x.
3. Let f n , f , g • R —>
R be Riemann integrable ( not necessarily
bounded ) , n 1. Suppose that
^| g( x ), /„ (x ) R.
Then
/« (z )| n
lim
—
yoc = f {x) X G
n — JR[ Ifn { x ) — / ( x ) |dx = 0.

lim
foo
4. State the Levi’s and Fatou’s Lemmas under weaker conditions and
prove them using Arzela’s Theorem.
5. Let /n , / : [a , b } —> [0, oo ) be continuous such that
inf
n m
( /» - l / n - /1 )
^
is continuous for each m
^ 1 and rb pb
lim f n( x ) = f ( x ) , x e [ a , b\ ; lim / fn { x ) dx = J/a f ( x )dx
n —
yoc n —
yoo
Ja
Then
6
r
lim |/n (a:) - / ( x ) |dx = 0.
n- yoo
a
Chapter 6
Series
1 Series of Numbers
We first introduce the following definition.

oo
Definition 1.1. ( i) Let { an } E £° . A formal summation = Ylan is
n =0
called a series.
<
( ii ) For a series
^ an , the following is called a partial sum of the series:
Y=
SN =
71
^
0
N
^ O.
( iii) A series Ylan is said to be convergent if the sequence { SN } N O of
partial sums is convergent , i.e., there exists an L E R such that for any ^
s > 0, one can find an AQ 0 having the following property
^ N
=| —
| SN L\
=0 ^
an L < e
71
ViV N0 .
In this case, L is called the sum of the series and denote L =

is not convergent , it is said to be divergent. ^2 an. If Ylan
From the above, we see that a series is convergent if and only if the
sequence of its partial sums is convergent. Hence, we have the following
result whose proof is left to the readers.
Proposition 1.2 . ( i) Series an is convergent if and only if
^ Q
^ for any £ > 0 ,
V# > P N.
n= p
n
^£
151
(ii) If Un is convergent , then it is necessary that the general term an

approaches to 0, i.e.,
lim an = 0.
n — oo
oo
oo
(iii) Series
^
n =0
an is convergent .
an is convergent if and only if for any fixed m 1 senes
—
n m
oo
We see that the convergence of a series
^
an depends only on its “tail” ;
71
any first finitely many terms are irrelevant to the convergence. If the series
is convergent , the first finitely many terms only possibly make the sum of
=0
the series different . Therefore, if we are only interested in the convergence

of a series , then it suffices to simply write which has been done
already in the above, unless the sum of the series is an interested part of
the problem.
oo
r
xn is convergent for |x| < 1 and it
Example 1.3. ( i ) Geometric series
is divergent for |x | 1. In fact , for any xGl, ^^

=0
71
N
1 — XiV +1
SN = J=>
71 0
" 1 x — 0.
r
If |x| < 1, one has

1
lim SN = -1
! -+ oo — X
oo
xn
Hence,
^=0
71
1 —— , for |x| < 1.
X *
On the other hand , if |x| 1, then xn does not go to zero as n > oo. —
Hence, the series is divergent . A special case is x = 1. Thus, series —
XX l ) n is divergent.
“
oo
1
( ii) Harmonic series
71 ='
1
n —
is divergent . For this series, although the
general term
^ converges to 0, but the speed is not fast enough. We look
Series 153
at the following:
1 1 1 1 1 1 1 1
1+
2 +3+4+5+6+7+8+9 "'
1 1 1 1\
> 1 + 5 + i + l + (8 + 8 + 8 + 8 + ' "
( )
l l
- 1 + + H = o o.
’
In this case
In general,
Si = l a - l +i St > l + l Sg > !+ -, •• •
S2 k > 1 + fc 0.
2
Thus
lim SN = oo.
—
N too
The divergence of the harmonic series tells us that the general term
—
a n 0 as n > oo is not sufficient for the convergence of the series. Fur-
—>
thermore, we have the following example concerning the Cauchy sequence
of real numbers.
Example 1.4. Let m > 1 be fixed . There exists a sequence {sn }n i E
such that for any e > 0 and any ra 1, there exists an ^
l such that
^^
Is* - Si| < e, Vn fc, £ n + m, n N . ( i .i )
But {sn }n i is not convergent. This means that a sequence satisfies the
^
above condition might not be a Cauchy sequence.
Such a sequence can be constructed in the following way: Let
ajm+1 — O' jm+2 — * * *
— &( j +l) m — ( j +11 )m j > 0.
Then define
k
Sk = £ 0 11 * k 1.
i= l
It is clear that
, 1 1 1
+ +- 1.
2+3
Sjm - 1 + • •• • 5 Vj
J
Thus, S k —> oo. However , for any n 1 , we may let j m < n ( jf + l ) m .
<
Then for any n /c < £ n + m , we have ^
*
4 . ^ 0 < - Sk <
1
J
Thus, (1.1) holds.
\
A particular case of the above says that if for any e > 0, there exists an
N 1 such that
l ^n+ l Sn | < £5 \/ n
^ N,
we do not necessarily have the Cauchy property of such a sequence {sn }.
Definition 1.5. ( i) Series Y an is said to be absolutely convergent if Y \ an \

is convergent.
( ii ) Series Y an is said to be conditionally convergent if it is convergent

but not absolutely convergent .
Proposition 1.6 . ( i) I f Y a n Is absolutely convergent , then it is conver-

gent .
( ii) If there exists a K 0 such that
an 0, Vn K,
00
then
^71 =0
an is convergent if and only if there exists an M > 0 such that
N
VN 1.
71=0
Further , in this case , the series must also be absolutely convergent .
(iii ) ( Comparison Test ) Suppose there exists an N e N such that
I an | ^ bn n N, 5
and Y bn is convergent . Then Y an is absolutely convergent .

( iv ) ( Ratio Comparison Test ) Let Y an and Y bn be two series of
positive numbers. Suppose that there exists an N
^ 1 such that
&71+1
an
^ bn+
71 1
Vn N. ( 1.2 )
Then the convergence o f Y b n implies that o f Y a n •

( v ) ( Alternative Terms Test ) Suppose { a n } is a decreasing sequence.
Then £]( — 1 ) nan
is convergent if and only if an —» 0.
( vi ) There exists a conditionally convergent series.
Series 155
Proof , ( i ) For any e > 0 , there exists an N 1 such that

q q
IYlan
= n p n= p
i°n i < 6 \/q > p N.
This proves our claim ,

( ii ) For any N > K ,
N
SN = SK + y; CLn -
n = /C'-f 1
It is clear that {SW }jv .fc:+ i is a non-decreasing sequence. Hence, it is

^
convergent if and only if SN is bounded from above.
( iii) For any e > 0 , there exists an N
Q
^ 1 such that
<£ Vq > p N.
n p —
Then
Q Q
ia" i n <£ q> N.

n p — n p —
This leads to the absolute convergence of an .
( iv ) Without loss of generality, we let N = 1. Then
d2 &3 bn ai
On — CL i •
<2 i a2 a n- 1 ^ aih '
b2 bn — 1
,
b1
bn
Hence, by the convergence of Y bn we see that Y an is convergent.
( v ) Note that
52/0 + 2 =
*
^ 2k + ( —1) 2 /c+ l a2 k+l + ( “ 1) 2 /c + 2 ^ 2 /e + 2
=S 2k —( &2 k + l ~
^2/c+2 ) 52/C , V /c 0
and
52/0 +1 = *52/ -1 + ( — l ) 2 /ca 2 /c + ( —1) 2 /c + l ^2 /c + l

* 0
= *52/ -1 + Cl 2k — ^2/e + l *52/0-1,

0 VA: 0.
V
“
Also
52/c + 2 = *52 /e + l + &2 /c+ 2
* 52/e + l
* *52/e-l * * * 5,
* l
and
&2k + l $2 k $ S - So
~ ~
^ &2 k + l
^
o is decreasing , bounded below , and {S
^ 2k 2 /c 2 *
Hence, {S2 /c }fc

^
ing, bounded above. Therefore,
} 2 /C +1 k
^ o is increas-
lim S2 k
AC — OO
= S, lim S2 k +1 = S .
—
k > oo
Further
| *W
_
^ 2 /e | = ^2fc + l Vfc 0
which implies that S = S if and only if an —> 0.
OO
(vi) Let an = —^ Then an is conditionally convergent .

71=1
The following collects some basic properties of convergent series.
Proposition 1.7. (i) Suppose Ylan and are convergent . Then
2 an + X] ^ ’ (ca „ ) = c
OO
+K) —
^ n
^ _^ OO
^ 71 5 Vc e R.
case
( ii )
^
n= 0
an is convergent if
OO
and only
OO
^
if
n= k
an /c is convergent . In this
71=0
a« - 5=Za —
71 fc
71 fc *
( iii ) Let an —> 0. Then

OO
a. n-j_ i ) — Q-o

^^
71 =0
( fln *
Exercises f
1. Let Xlan be convergent . Must l) nan also be convergent ? Why?

2. Let CLn be convergent with all an > 0. Is it possible that X} is
convergent ? Why? How about XI ? Why?
f -
3. Let X an and X n be convergent. Is
How about XI ^
if all bn 0? Why? ^^ Ylanbn convergent ? Why?
Series 157
oo n
4. Let
n = l'
^
an be convergent. Let bn = —nl }
fc = l
^
a /c . Is Y \ convergent ?
^ bn
Why?
5. Let an 0, and an -» 0. Let m > 1 be fixed . Define
bmk+ j = ( 1) k > 0, 0 j m — 1.
Is convergent ? Why?
2 Further Tests for Convergence
We now present some further tests for the convergence of series.

Proposition 2.1. ( Cauchy Test ) Let
Vn 1.
oo
tin
^ 1
oo
Then
^
n= l
an
Proof . Let
is convergent if and only if
^
fc = 0
2fca 2 fe is convergent .
iV if
n= l
We now prove the following by induction:
T/f =
^
/c = 0
2^ 02 ^ ViV 1, K 0.
52 K+i
_ i TK 252 K V/f 0. ( 2.1 )
For K = 0, we need
S\ To 25i ,
which is the following true statement :
ai cii 2ai .
Suppose our claim ( 2.1) is true for some K . Then , for K -\- 1, we need to
prove
S
* 2K + 2 !
_ TK +1 252 K +1 . ( 2.2 )
Note that
T -j- i
^ = TK + 2K +1a 2 K+i . ( 2.3)
* Also
2 k +1 2 /C +1
S2 K + 1 = S 2 K +
^^
n= 2 K + 1
an
^ 52 K +
^^
n =2 K + l
a2 K + i = *S2 K + 2Ka 2 /c + i .
Hence, by induction hypothesis, one has
2 S 4- 2 +la
2 S 2 K +\ K
TK 4- 2 + K 1
= TK+I ( 2.4 )
^ 2K
^
Similarly, ( noting ( 2.3) , and induction hypothesis)
2 K +i CI 2 K + I *
2 K +2 1
_ 2k +2 -1
_ l = S2 + _1 + y
S 2 K+2
=S
K 1
n= 2 K +1
4- 2 x +1a 2 ^ + i
^ 52K + I - I 4-
K +
n = 2 K -f 1
4- 2 1 a2 K + i
'
^ & 2 k +1
-
( 2.5)
^ +i -i TK = TK + I
Then , combining ( 2.4 ) and ( 2.5 ) , we obtain ( 2.2 ) . This proves ( 2.1) . Hence,
TK is bounded from above if and only if SN is bounded from above. This
leads to our proposition.
The above result has a very interesting corollary.
oo
i
Corollary 2.2. Let q > 0 be a real number. Then
n= l '
—
nq
is convergent if
and only if q > 1.
Proof . Let
1
&n — nq 71 1.
Then
1
Cl 2 k
—( 2k )q
Vfc
^ O.
oo
1
Hence, by Proposition 2.1, we see that z} nq is convergent if and only if
n='
1
—
E
=
oo
*
k 0
1
( 2k )q -E oo
=
k 0
is convergent . For such a geometric series, it is convergent if and only if
1
( 2 q ~' ) k
<7 > 1.
The following is another relevant test .
Theorem 2.3. ( Cauchy’s Integral Test ) Let f : ( 0 , oo ) -» ( 0 , oo ) be
continuous and non-increasing on [ N , oo ) for some N N. Let
an = f ( n) n 1.
Then an is convergent if and only if
b
lim
—
HX )
f f ( x )d
JN
X < OO. ( 2.6 )
Series 159
Proof . For any n N , we have

an+1 = f ( n + 1) f (x) / (n ) = an , Vx G [n , n + 1].
Thus
& iV + l H h ttn +1 f { x )d x avH

^ h an -
Consequently, our conclusion follows.
The above Corollary 2.2 can also be regarded as a consequence of the
above Cauchy’s integral test .
Let us briefly mention an interesting problem relevant to the series in
Corollary 2.2.1 To this end , we recall (from the Appendix ) C , the set of all
complex numbers. We look at the following series:
°o
( 2.7)
n=1
where s G C. Write s = a + it with cr, t G R. Then
( cr + it ) 1
~9
\ n \ = |n "
| = |n ~ a ||e^ logn | =
na
Thus, series ( 2.7) is absolutely convergent for a = Res > 1. One can show
that the above function admits an analytic extension on the complex plan
C , still denote it by £ ( s ) , except for the point 5 = 1, with
limy ( 5 - 1) C( « ) = 1 .
s —
1
Such an extension is called the Riemann zeta function. One can show that
function £ ( •) satisfies the following Riemann’s functional equation:
C (s) = 2s 7Ts 1
sin ( y )r( l - sK ( l - s) s e C\ { 1}, ( 2.8)
-
where T ( ) is called the Gamma function defined by the following:
roc
Jo
e X x9 l d x.
T( s) = /
Note that in the above functional equation ,
and
lim sin
s-* 2n ( y ) r <l - . )#0 n 0
C (-2n ) = 0, n 1.
1Since this part of the material is not closely related to the following material , the first
time reader of the book could skip this part .
Thus, all negative even numbers are called the trivial zeros of ( ( •). The
well- known Riemann hypothesis says that:
All the non-trivial zeros of ( ( • ) lie on the line Res =
This is a very important and challenging conjecture. It is open as of today.
oo
— . Let
1
Example 2.4. Consider series
1 1 1
^
n= 0
1
+d _
1
^
n
Sri “
+ + + '" n 0 ( 2.9 )
0! l! 2! k =0
f e!
since each term in the sum is positive, we see that Sn is increasing. We
now show that Sn is bounded from above. In fact , for any n > 2, we have
1 1
— <
n! 2 1 ’
n
( 2.10 )
—
Thus
i hr
Sn < 1 + 1 + + • • • + 2n ~1 = 1 + 1 - — < 3,
i i-(
~
5
Vn
^ l. ( 2.11)
Hence, sequence { SN}N I is convergent .

^
Proposition 2.5 . ( Cauchy’s Root Test ) For series Ylan , let
a = lim
n —too \ an \ .
Then
(i) a < 1 , series is absolutely convergent ;
(ii) a > 1 , series is divergent ;
( iii) a = 1 , the test fails.
Proof . In the case that 0 < a < 1, we can find an N

. __
. i , , _L l —o a +1 ^ 1 suchNthat
.
\ an \ n sup |a m |
m n
< cH g 2
<1 Vn
9
Hence
^
00
°°
^
1
£ 1 — E w t =£+ ( K|i )” < £ KI + =£+ (

n N n N n N n N
) < 00.
^ l
In the case a > 1, we have a subsequence ank such that

^ l
1
a n k \ n k > 0C
a l a 4- 1
k 1. —
2 5
Series 161
Hence
+1 »
ank ( 0
) 15 Vfc
^1 5
which implies that \ an \ does not go to 0. Consequently, Yhan is divergent.

Finally, in the case a = 1, the series could be convergent and also could
be divergent . For example,
00
1 1 \ n
lim
—
n » oo n2 =1
71 =1
and
oo
1\ i
y - = oo lim
—
n > oo n = 1.
n= l
Theorem 2.6. ( d’Alembert’s Ratio Test ) For series ]Can , let an

for all n N with some N £ N and iet
^ ^ 0
n l| |®n l |
a = lim l ^ + /3 = lim &+
71 —» 00 |®Tl | 71 —» 00 TI
Then
( i ) a < 1, the series is absolutely convergent ;
( ii) /3 > 1, the series is divergent ;
( iii ) a = 1, or /3 = 1, the test faiis.
Proof . In the case that a < 1, we have N m such that
l& +l | | + 11 1-a 1+ a
^ N.
Ti
sup <a+ <1 Vn
® 7l | 1 ^771 2 2
Hence
771
^ 71
|ara |
/1 + a \ n- w Vn iV.
This leads to
^ ^ 2 )
oo
laJv |
E la« l E i ®ni + y an
n N = TV 4- 1
^ 71
V
E KI + =E- (4^
n N
^ 77 TV fl
r" M < OO.
Next , for the case /3 > 1, we have N m such that
|«TI +1|
inf l I ra + , *
& l 1+
«TI > m
^ n I &m \ 2
>1 Vn
^ N.
Hence
|«TI | N,
and
^
lim |an | = oo .
—
n > oo •
Consequently, the series is divergent .

Finally,
°o 1 1/ ( n + 1) 2
n2 < °°
lim
n-> oo 1/ n2
=1
71 = 1
and
OO
1
l/ (n + l )
vi
'n
= 00 lim
—
71 fOO 1/ n = 1.
=1
71
Thus, in the case a = 1 or /3 = 1, the test fails.

We have seen that when
«71+1
( 2.12)
«71 =
lim 1
>- 71
OC
—
the d ’Alembert ’s ratio test fails. The following gives some further test when
( 2.12 ) holds.
Theorem 2.7. ( Raabe’s Test ) * Suppose that
«71+1 |\ 1
t = lim
—
n > oo M 1
«71 1 /
( 2.13)
exists (including ± oo ). Then an > 1, and *
not absolutely convergent if t < 1. ^ is absolutely convergent if t

r
•
Proof . First of all, applying the Taylor’s formula to / ( x ) = (1 + x ) p ,

for some p > 0 , at x = 0, we get f
p( p + \ ) x 2
—
(1 -f x ) ~ p = 1 p x +
2 ( 1 + 02 + 2
6 (0, 1 ) .
+
«
Hence, letting x
1 -
1+ -
p ^
= , one has
=i- n + + * i
-
P(P + 1)
> 0. ( 2.14 )
n n2 2 (1 + i ) p+ 2
Series 163
Now , suppose t > 1 ( allowing t = oo ) . Let pG ( l , t ) . Then there exists an

N 1 such that
^ p < n( l
!)
Qn
" +1 n Z N,
'
tin
which , together with ( 2.14) , implies that
-P (n + l ) ~ p
tin-f 1
tin
< 1- -np < ( i + -nJ)
\ n~ P
Wn
^ N.
p
Hence, by Proposition 1.6 (iv ) , and the convergence of 2n
we obtain the convergence of J2 an . ^ ( n te P > 1)
°
Now , let t < 1. We can take p 6 ( t , 1) . Since
may find some N 1 large enough so that
An
^ 0 is bounded, we
^
—n > n 1 \
^n+1 ) •
! n
^ TV,
which , together with ( 2.14 ) , implies
An - _ (n + l ) ~p
®n -H 1
Qn
> 1 - nP
.
n2 K )
Hence, by Proposition 1.6 ( iv ) again , and the convergence of \ an \ will lead
P
to the convergence of Y n ~ p which is not the case since p < 1. Therefore,

^ ^
2 \ an \ must be divergent.
^ If in the above, t — 1, then the test fails. In this case, one has
& n +1 1
n
The following says more for such a case.
Theorem 2.8 . ( Gauss’ Test ) * Let

^71+1 = i - £ + An ( 2.15 )
n nq ’
where <7 > 1 and { An } is bounded . Then an is absolutely convergent
ifp > 1, and not absolutely convergent ( either divergent or conditionally
* convergent ) if p 1.
Proof . By ( 2.15 ) , we see that
lim n (1— ^71+1 = P- ( 2.16 )

71
—> 00 V
^ 71
Thus, the conclusions for p 7 1 follows from the Raabe’s test.

^
We now look at the case p = 1. In order to consider this case, we first
apply the Taylor’s formula to f ( x ) = log ( l x ) at x = 0 to get —
Taking x =
Next , let
„
l g( l - *
we have
log ( l -
n
> = -* - 2(1
n
An
nz ’
^
9
5
An —
0 e ( 0, 1 ) .
1
2( 1 - ) 2 S ' ( 2.17)
1
bn = ( n 1) log ( n 1) n 2.
— —
Then
bn+1 ( n - l ) log ( n - 1) _ ( n - l ) [logn + log ( l - - )]
bn nlogn nlogn
1 ( n - l ) / l An
|
n nlogn \ n n2
1 1
H
Bn
n n log n nz ^ /
with
1 An ( jt 1)
Bn =
logn nlogn
which is bounded. Then we have
^n -4-1 n -fl 1
^ An Bn
&n bn nlogn nq nz
when n is large enough. Since bn is divergent , we have the divergence of
El an •
For any finite sequence {an } =1 of real numbers, let r : {1, 2 , • • • , m} ->
{1, 2, • • • , m} be any bijection. Then
771 m
an — ar ( n ) ‘
n= l n= l
We call {ar ( n ) }71m=1 a rearrangement of {an } = x . The above means that the
sum of a rearrangement of a finite series is the same as that of the original
(finite) series. The question is whether this is true for infinite series.
A
j\n
y —— /
Example 2.9 . Consider series
gent . Note that
1 i 1 A
=
71
^
1
i lAl
which is conditionally conver-
2 + 4 + - - + 2N
'
k =1 k =1
N —> oo.
•
Series 165
Hence, there exists an N i 1 such that
±2-k > 2
Ni 1
iV2
V 2ik > l, V ?
fe = i fe = Ni + i
In general,
;vi + i
1
E —
2k
> i i 1.
= +
fe AT< l
Then we make the following rearrangement :

1 1
5 -1
2’ 2iVi
1 1 1
2( Nx + 1)
?
2A 2 ’
^ 3’
1 1 1
2 ( JVi + 1) 2 iVi-|_ i 5
( 2i + 1) ?
The resulting series is divergent .
Actually, we have the following result.

oo
Proposition 2.10. Let
^
OO
=171
an be absolutely convergent , and r : N -> N
be any bijection. Then

^—
n 1
aT ( n )
oo
JS a so absolutely convergent , and
^
oo
an ar (n ) *
n =1 n= l
Proof . Let
oo
| an | — L.
=1
71
For any n
^ 0, let r (n) = max{r ( k ) | 1 k < n}.
Then
»*>
71 T ( TI ) OO
y ia ( / ) i y M y L.
k — ^
1 ^,
^r c
k= 1 ^ =
71 1 ^ \ an \ —
\
oo
Hence,
^
n= l
|ar ( n ) | is convergent , and
—
oo
L
oo
^—^
n 1
|ar ( n ) |
^ L.
oo
By switching the positions of
Consequently, our conclusion follows.

^< ^
n=l
|
L
and
L'.
n= l
|ar ( n ) |, we see that
By the similar idea of Example 2.9, we have the following result .

oo
Proposition 2.11. Let an be conditionally convergent . Then for any
n=1
oo
—
L G R , there exists a bijection r : N > N such that
and
oo
^
n=1
ar { n ) i S convergent
Qr ( n )
n=l
We leave the detailed proof to the interested readers.
Exercises
1. Let Yhan and 2 bn be two series of positive terms. Suppose that
there exists an iV 1 such that
& n +1^ &
^
<^ n + l Wn N.
tin bn
Then the divergence of 2 an implies the divergence of 2 bn .
^
2. Let an = np
1 ^
Find necessary and sufficient conditions on p, q
^
( lnn )q
such that the series an is convergent.
^ 22, a 2n+ 1
3. Show that for any a 6 R ,
4. Suppose Y2 an and
^—
n 1
n\
bn are convergent .
is convergent.
Then Y2 anbn is absolutely *

9
convergent .
5. Is£ ( -n 1+)"1* convergent ? absolutely convergent? Why?
2
6. Let Y2 an be convergent and p > Then Y2 7$ 1S absolutely con- f
vergent .
is convergent . Is Y2 an necessarily convergent ? Why?
7. Suppose
^2 an
Series 167
3 Series of Functions
Let ( X , d ) be a metric space and f n : X —» R be a sequence of functions

oo
f n { x ) . For any given
n 1. We now consider the following type series:
x e X , the above is a usual series of real numbers. To study the above
series, we define the partial sum as follows:
^
71 =1
N
~
We have the following definition.

SN ( x ) =
^—
n 1
2 f n { x) x e X.
.
Definition 3.1 If for each x G X , S N { X ) is convergent to some number ,
denoted by S( x ) , then we say that the series is pointwise convergent, and
write
oo
S ( x ) = lim SN ( x )
N - oo >
=V
^ '
fn( x ) I 6l
=
n l
If the convergence is uniform in x G X , we say that the series is uniformly

convergent.
.
Definition 3.2 For any bounded function / : X -» R , we define the sup
norm of / ( • ) by the following:
sup | / ( x ) |.
ll / lloo = xex
The following is the simplest criterion for the convergence of function
series .
.
Theorem 3.3 ( Weierstrass M-Test ) Let ( X , d ) be a metric space and
let /n ( - ) be a sequence of bounded real- valued continuous functions on X
such that
oo
< OO. (3- 1)
71 =1
oo
Then the series
-
S( ) on X . ^
71 =1
/n ( x ) converges uniformly to some continuous function
Proof . Since
N oo
Ifn ( x ) \ ||/n ||oo = M < OO xe X
=1
71 71 =1
we see that /n (x) is absolutely converges for each x e X . Next , let SV ( ) -
be the partial sum sequence. Then SJV ( ) is continuous, and for any N 1 - ^
and ra 1, we have
-/V + 771 OO
]T || /„||oo y ll /nlloo -> 0 N oo.

7l = JV + l ^ n= N + l
-
This means that SV ( ) is uniformly Cauchy. Hence, it converges to S( ) -
-
uniformly with S ( ) being continuous.
The following is an interesting application of the above Weierstrass At -
test , which gives an example of a continuous function that is nowhere dif-
ferentiable.
Example 3.4. Consider the following series:
/ (x ) = y —2" ^
71 =1
cos (8n 7rx )
i e l.
Since
cos (8n 7rx ) 1
Vn 1, x e R ,
2n 2n ’
by Weierstrass M -test , we see that the series is uniformly convergent . Thus,
/ ( •) is continuous. Now , we want to show that / ( • ) is nowhere differentiable.
To this end , for any m , j 1, we evaluate
^
+ +H £
|/ ( ) -/ (
771 —1
71 =1
2-”
b -«» bM
n
COS
0 + 1) ?r ) - ( j&r )
gTn n —
71=1
+ 2— 771
cos ( O '
+ 1)*-) — cos 0* )
oo
_n
+ ]T 2 cos (8n-m ( j -f l ) 7r ) — COS (8"- +)
= 771+1
71
2-2 rn _
771
0
£
-1 " n 7T
m
771
— 1
— = 2 2 -
" ‘
•
Zs §
Q 771 71 §771 / ^
71 =1 71 =1
_m m
7T 4 -
= 2•2 -
8m 4 - r » (2 - I K" > 0.
Series 169
Now, for any xo G R, we can find m, jf 1 such that

j 8 mxo < j + 1,
x0 G gm 3 gm +1 X '
’
If / ( •) is differentiable at xo , then for any e > 0, there exists an mo 1
such that for m mo with a proper j ,
v
(2 - f )2-" « KXH (iy
3 +1
v
( | /'(xo ) | + e) 8-m .
^
This leads to a contradiction.
^
(|/'(a?o )| + e) ( r xo ) + ( | /'(®o ) | + e) ( X
°
“
8m )
The following results are concerned with the so-called integration term-
by-term and differentiation term- by-term for the series.
Theorem 3.5 . ( i ) Let G CM771 be a bounded domain with \ dG \

oo
= 0.
Let fn : G -» R be a sequence
uniformly convergent . Then
« oo
of continuous
oo n
functions and
^
n=l
/n (x ) be
=£
J £ fn ( x )dx
J fn( x )dx
.
»
(ii ) Let fn : G R be a sequence of continuously differentiable func-
oo
V /n ( x ) be uniformly convergent , and there exists an xo
tions. Let
^
n= l
oo
fn ( xo ) is convergent . Then
G G
such that
^
77 =1
V
OO
( £ / (*) ) = £ V/ (*) .
»
OO
»
71 =1 71 =1
-
Proof , ( i ) Let 5AT ( ) be the sequence of partial sums. Now , in the case
- -
that Sjv ( ) converges to S( ) uniformly, we have
[ S{ x )d x lim / Sjsr { x ) dx
JG —
N > oo JG
which is what we want to prove.
( ii ) Since V / ( x ) = ( fx1 ( # ) , • • • , /xm (# ) ) > it suffices to prove the case
m = 1 . Thus , we let G = ( a , b ) . By assumption , S'N ( ) converges to some -
-
g ( ) uniformly and SN ( XO ) is convergent. Hence, we have
,
S( x ) = 5 ( ar0 ) + /
J Xo
g ( t )dt .
Consequently,
S' ( x ) = 3(1) x e [a , 6]
which is exactly what we want .
Note that if we use Arzela Theorem (Theorem 4.5 of Chapter 5 ) , the
above can be improved .
To conclude this section , we look at limit theorems for sequences of
number series.
Theorem 3.6 . Let {anm}n , m be a double-indexed sequence of real num-

bers. ^ i
( i ) ( Dominated Convergence Theorem) Let there exist sequences

{ bn }n 1 and { cn }n i of real numbers such that
^ ^
n, m
00 00
bn
^ &nm
^ On , 1
cn are convergent . Suppose

and
^ ^
n— 1
bn and
=
n 1
lim a nm
—
m toc — On , Vn
^ 1.
co
Then
^ n= 1
an is convergent and
co CO CO
z anm =zY =zY2' mlim

V' '
lim
—
m > 00
n= l n= l —too anm
n=l
•
( ii) ( Levi’s Lemma) Suppose

Vn , m 1,
and
0
^ &nm
^ an (m+ l )
lim a nm
m —> oo — On i n 1.
Series 171
Then
oo oo
lim
m-> oo
flnm — lim
—
m > oo
®nm
-
77 =1' n = l'
( iii) ( Fatou’s Lemma ) Suppose there exists a sequence { bn }n of real
numbers such that î
Vn , m
OO
& nm
^ bni 1J
and bn Is convergent . Then

n= 1
oo oo
=1 m > oo ^—
77 77 =1
lim a nm
Proof , (i) For any £ > 0, there exists an N 1 such that

lim
m —> oo
X anm -
^ *
^
IX] bn + 1 Y2 „ <
n= k n=k
c e5
Then
t i t
y
oo
-e <
^ ^ ^ ^^
n=k
bn
n= k
&nm
^— n k
<£ 5
anm is convergent uniformly in m. Letting m —» oo in the above

Hence,
one obtains ^
71=1
t
5
—£
£
^ — Unn k
5
oo
This shows that an is convergent. Moreover
77 =1
oo oo N N oo
x>
oo
nm XA | | X) ® nm + 1C anm + y~] Qn
77 =1 =
77 1 = 77 1 77 =1 fe = AT + l 77 = jV -fl
AT
y
77
Hence , by sending m -> oo , we obtain our conclusion ,

=1 ^ 1^ 77777
^ | I77 “ "
^.
£
( ii) Since
»
0 m, n
^ &777T7
^^ 77 )
^1 5
S
'
we have
oo oo
If we have
^^ ^^
n= l
(trim
^ n= l
&n *
oo
lim
m —> oo
anm = OO ,
n— 1
our conclusion is obvious. We now assume that
oo
lim
m-> oo Yl anm = S <
11
OO .
n=
Then for any TV
^1 N oo ,r
Sending m —> oo, one has

E
n= l
«nm
^ ^^
n= l
Q>nm
^
AT
an
n 1 — ^
Consequently,
oo
y
Then we have the following situation:
n= l ^ ^5 * *
0 Vn 15
OO OO
^ dnm
^^
y
with
^ ^
n=l
0 and
n= l
an convergent . Hence, ( i ) applies.
( iii ) First of all, if the right- hand side is infinite, the conclusion is trivial.
Now , let the right-hand side be finite. By
bn Vm , n 1
one has
^ &nm ?
bn Vm , n
and
&nm ~
^0 ^1
an — bn = mlimoo a nm - bn 05 Vn 1. «
—*
Series 173
Hence, for any N 15

N N
—
° < E= n l
( &n
TV
bn )
^^
n=1
ilm (®nm
m- yoo
oo
bn )
y ~
y . (a
lim
m — yoo n 1
—
oo
^V] ( a nm 6„ )
oo
a„ m - Y] bn < oo.
m
lim
— yoo n = l
^ nm -M
= mlim
— yoo ,
n= l n= l
Then we may let iV > oo to get — •
oo oo oo
x^
n= l
(a
" bn
) lim V', a nm
—
m oo > n= l
!=>
n l
•
This implies that the series on the left- hand side is convergent . Hence,
oo oo oo oo
V ^nm -
^
n= l m—
,
lim amn
y oo
This proves our conclusion.

^\
n= l
bn ) b
"
^^ ^ ^
n= l
n lim
m — yoo n l
=
The following example shows that if conditions assumed in the above

are absent , the corresponding conclusions might fail .

1, n= m
® nm = snm 0, n / m.
oo
Then there exists no cn controlling anm from above such that is
n=1
convergent . In this case,
oo
—1
Whereas 5
^^—
n 1
&nm 5 Vra 1.
m
lim
— oo
y
&nm — 0, Mn 1.
Thus
oo oo
m—
lim
yoo
anm 1
^ 0= Y" mlim
— 1 yoo ^ nm -
n= l n= l
\
( ii ) Let
—m n = mJ
& nm — o,
5
n m.
^ oo
Then there is no bn controlling a nm from below such that
gent . In the case, ^
n= l
bn is conver-
lim anm =0
m — y oo
whereas,
oo
£= —
n 1
&nm .
ni
Hence ?
oo oo
lim } a nm = -oo < o = V] lim anm
m y oo
—
n= l n = l ^00
This example also serves as a counterexample for Levi’s Lemma since the
*
monotonicity condition fails.
Exercises
1. Let
oo
sin nx
x G R.
n= l
—
Show that / : M > M is continuous for p > 1 and is continuously differen-
tiable for p > 2 .
2. Let
n =0
( i ) Show that f ( x ) is well-defined on [0, 2].
( ii) Find all x G [0, 2] at which f ( x ) is continuous, continuously differ-
entiable, respectively.
(iii) Find all x G [0, 2] at which f ( x ) is infinite time differentiable.
= 04 ) n ^ i
n)
3. Let x k G i e.,
OO 1_
n*fciip = ( £= i4 r)
n l
n) p
< 00 5 k i.
Series 175
Suppose { xk }k 1 is Cauchy under the fp- norm || • ||p , i.e., for any e > 0,
^
there exists an iV 1 such that
^
\\ xk xm \\ p < e , Vfc, m N.
Show that there exists an x = ( x ^ ) ^-
n i G £ p such that
lim %k ~ x \ \ p = 0.
—
k too
4 More Convergence Criteria
In this section , we will present some more convergence tests for series. A
major motivation is to determine if the following type series are convergent:
oo
sin nx
^=
n 1
n
( 4.1)
It is clear that since

sin nx
n
<-
1
n
V ±n = o c
n '1
—
Weierstrass M-test fails. We will introduce more powerful tests to deter-
mine the convergence of such kind of series.
We first consider series of the form: ]Tan &n , where an , bn G R. The
following lemma is crucial.
Lemma 4.1. ( Abel’s Lemma ) Let
and
CL\
^-^ ^
02 ( 4.2 )
k
1 k <n 5 ( 4.3)
i= 1
for some M > 0. Then
n
~
5= akbk < Af (|ai| + 2|o„|) .
Proof . Let
^
k 1
( 4.4)
k
Bk = YJK
=
i 1
fc 1.
\
Then ( with BQ = 0 )
bk = Bk - Bk -1, 1 k m.
Hence
n n n n
akh —
^^ ^^ —
/c = l
n
=
fc l
n 1
(-^fe I )
fe = l
n 1
cikBk
—
^^
fc = l
ctkBk -1
( 4.5)
^^ ^ ^ =
/c l
akBk
fc = l
Qfc + i - fc
^— ® nBn 4-
^
k =1
& /c + i ) - fc *
^
Consequently, by ( 4.2 )-( 4.3) , we have
n n 1 —
^^ ^ — fc = l
dfc 6fc |ttn ||J3n | 4
n 1
~
^^
fc = l
|cifc & fc + i || - fc |
^ r
proving ( 4.4 ) .
M
^ \ an \ 4-
M ( |ai | 4- 2 |an | )
^
/c = l
( a* - & k +1
)
— M (|< 2n | 4~ d\ on )
>
Clearly, if we replace ( 4.2 ) by the following:

( 4.6)
we will still have ( 4.4 ).
d\
^< ^ ^^
22 * * * n j
Theorem 4.2 . ( Abel’s Criterion) Let ]C 6n be convergent and { an } be

monotone and bounded sequence. Then Yhanbn is convergent .
Proof . First of all by the convergence of 6n , for any e > 0 there exists
an N 1 such that
n +m
I 5Z bk <
/e = n + l
e5 Vra > 1,5 N.
Then by Lemma 4.1, assuming \ an \ K for all n > 1

n+m
| __
^^
fc = n +1
CLkh
^ ^ (|^ + | 4 2|
n l “
<2n } 7n |)
^ 3Ks.
Hence, an 6n is convergent.
Theorem 4.3. ( Dirichlet’s Criterion) Suppose there exists an M > 0
*
such that
n
| fc = i
bk | Vn 1 ( 4.7)
an converges to 0 monotonically. Then 6n is convergent .

and
^ an
Series 177
Proof . By the convergence of an , for any e > 0, there exists an N

such that ^1
\ "n |
Q £ •> Vn > N.
On the other hand , by ( 4.7)
m n -j- m n n+m n
|y | y E= 6*| + | EH
k=1 ^ bn+k —
Hence, by Lemma 4.1, one has
n+ m
k =1
m
bk
k=1 ^ bk
k l = ^ 2M -
k 1
—|y
^^
k =n
^ kbk ^ ri+kbn + /c
This gives the convergence.
k =1 ^ 2 M (|an + i | + 2|an + m|) < 6M£.
Example 4.4. Consider series ( 4.1) . We saw that Weierstrass M-test does
not work. We now show that for each x
n
\
yy sin kx M (x ) 5
fc = l
for some M ( x ) > 0. If x = 2m7r , we have
n
*
yy sin kn = 0.
fc = l
Next , we note that
sin a sin /3 = -
Thus, for x 2m7r ,
^
2 sin
^^
x
sin x 4- sin 2x H
3x
= cos —2 — cos T + C“ T — cos
4 cos
(2n
3x
f sin nx
— l )x — cos ( 2n 4 l )x = cos —x — cos

5x
+ . .. — cos
( 2n — l ) x
2
( 2n -f l ) x
2 2 2 2
Consequently,
n
yy sin kx sin ||
= M (x ).
k= l
Then , by Dirichlet ’s criterion , we have the ( pointwise) convergence of
oo
sin nx
. The same idea also applies to
n
=
n l
oo oo oo oo
sin nx cos nx sin nx cos nx
E=
71 1
np 5
E=
71 0
np 5
E=
n 2
( In n ) p 5 E= (In n )
n 2
p
%
with p > 0.
\
We now look at the following type series: Y an { x ) bn ( x ).

.
Theorem 4.5 ( Abel’s Criterion) Let ( X , d ) be a metric space and
Y bn ( x ) be convergent uniformly on X , and {an (x )}n i be monotone and
uniformly bounded on X . Then Yan { x ) bn ( x ) is convergent uniformly.
->
Proof . By the uniform convergence of Ybn ( x ) , for any £ > 0 there

exists an N
^ 1 such
+ n m
that
M*) < £ J Vra > 1, x e X, n N.

k = n+ l
Then by Lemma 4.1, assuming |an ( x ) |

n+ m
K for all n
^ 1 and x,
ak { x ) bk ( x ) e (|on + i (ar ) | + 2|an + m ( x )|) < 3-ftTe.
k = n+ l
Hence, Y a n ( x ) bn ( x ) is convergent uniformly.

Theorem 4.6 . ( Dirichlet’s Criterion) Let ( X , d ) be a metric space.
Suppose there exists an M > 0 such that
f
n
y bk ( x )
and
^
k= l
an ( x ) converges to 0 monotonically and

M x E X , Vn 15
uniformly on X . Then the

series Y an ( x ) bn ( x ) Is uniformly convergent.
Proof . By the convergence of an ( x ) , for any e > 0 , there exists an N 1
such that
\ an ( x ) \ < s , Vn N, Vz X.
On the other hand >
m n+ m n
^ vbn
k=1
+k ( x )
k= l
M ®) - J2= h( x )
k 1
n+ m n
Y=2 + | bfc (x ) 2 M .
Hence, by Lemma 4.1, one has
k l
bk ( x )
^
=k 1 ^
r
- 2Af ( |an _ _ ( x ) | K 2|an +

^^= ^
- ffcip ) bn+ { pc ) ) < QMe .
'
m (x) |
k 1
Qn
^k f i -
Series 179
This gives the convergence.
Exercises
1. Let
oo
sin nx
/(X) = £
=
71
„r 1
x e ( 0 , 27r ) .
Show that / ( •) is continuous for any p > 0 and is continuously differentiable

for any p > 1. ( Compare with Problem 1 of Section 3. )
oo
sin nx
2. Discuss the convergence of the series
^
n= l \/n + x
4 4
3. Consider series
series is convergent.
“^
n= 1
rx ( x + n ) in
L n
. Find all values of x such that the
5 Power Series
We now look at a special case of function series:

oo
£> (x
n =0
n - x0 ) n . (5.1)
Such a series is called a power series.
5.1 A general consideration
Definition 5.1 . Number 0 is called the radius of convergence for series

(5.1) if
OO
y an{ x - x o ) is convergent \ x — xo | < R ,

-
^
n
(5.2 )
n= 0 is divergent , \ x — xo \ > R.
In the above case, we call ( xo — R, xo -f R ) the interval of convergence.
Note that no condition is given for |x — xo| = R in the above definition.
Proposition 5.2 . The radius of convergence R is given by
( lim | |i ) .
— an
R= (5.3)
n too /
Thus, for the case R > 0 , the following function is well-defined :
oo
f (x) = Ylan
=
n 0
( X ~
Xo )
n
x e ( xo — o + R) >
oo
Moreover , < r < R , the series — xo )
^
n
for any 0 an ( x is convergent
n= 0
uniformly on [ xo — r , X o + r ] to / ( ) which is continuous and differentiable
•
with
oo
y nan ( x - X o ) — ( xo - r , x0 + r ) .
^
n 1
f \x) = x 6
71 =1
Further, for any — R<a<b< + R,
-
XQ XQ
7* 0
/ y an ( x - xo) dx = f : - (° - fl)’?1..
la
'
^
n =0 n =0
Proof . We first show that for any x with r = |x XQ | > -R , the series —
is divergent . In fact , by the definition of R , we can find a subsequence n
such that for 0 < £ = ri-?rf * there exists a K 1 so that
’
i 1 1 r-R 1
R Rr r
k
^ K.
Then ( note \ x — xo \ = r )
(i - ) r = 1
1
|anfc | nfc \ x - x0| >
Hence, the series is divergent .
E k
^ K.
Next , we prove that the series is convergent uniformly on [ xo — r , xo + r]
for any r G ( 0, R ) . By the definition of R , for any 0 < £ < £—
when n is large, we have
( 1 + Re )r
^ ,
x
0 < \ an \ r < +e r R G< 1.
Hence, by Weierstrass M -Test , the series is uniformly convergent on [ xo
)
—
r , x0 + r ] .
The proof of the rest of the conclusions are straightforward .
The following result is concerned with the continuity of the power series
at the end point (s) of the interval of convergence.
Theorem 5.3. ( Abel ) Let
oo
y an { x - xo )
^
f (x) = n
n =0
Series 181
oo
anRn is convergent , then
with the radius of convergence R G ( 0 , oo ) . If
/ ( ) is left-continuous at x = xo + R , i.e.,
•
oo oo
^
n =0
f ( x0 + R ) .
n =0 n= 0
Likewise , the series is convergent at x

if = XQ — R , then / ( ) is right-
•
continuous at x = XQ R.
Before proving the above, let us make a simple remark. If the series
—
Y an ( x xo ) n is uniformly convergent on [ xo — R, xo + il] , then the con-
^
clusion follows from the exchange of the order of the limit and summation.
Now , we only have uniform convergence on [ xo r, xo + r] for any 0 < r < R —
and the convergence at x = XQ + R. Therefore, the conclusion is not obvious.
Proof . Without loss of generality, we may assume that xo = 0 and

R = 1. Then we need only to prove that
oo oo
JS 2=>?_ n 0
*" = £ “" = s-
n= 0
Let
N -l
SN = ( Y a")
n =0
- 55 AT 15 So = -5.
Our condition implies

lim SN 0, *$n + l n 0.
AT — oo
Hence, for any x G ( 1, 1) —

oo oo oo oo
n+ l Sn ) xn = Y Sn+^ U
- E
n =0
oo
n =0
oo
n 0
— oo
n=0
Y= Sn+ iXn
n 0
-
n =0
S'n + iXn+1 So = Y= Sn+ ( xn - ^ + ) + S.
n 0
l
n 1
—
Since 5n > 0, for any e > 0, there exists anA
•
^ ^ l such that
\ Sn\ < e , Vn
^ N.
For any x G [0, 1) , we have xn +1 xn . Consequently,
oo oo
I Y— anxn - S = | — S„+ (x + - x")

^
n 0
^
n 0
i
n 1
N oo
< £ |S„+ |(x" - x"+ ) + E z ( xn - xn+ )

i
1 1
71 = 0 = + n N l
N
El Sn+ \ ( xn - xn+ ) + exN + .

i
1 1
=
n 0
Now , in the above, we let x 1. Then
OO
limj £ «„x” - s k £ .
=0 71
Since e > 0 is arbitrary, we see that

oo oo
l i m V" anXn
£-> l
^
71=0
—s—y 71
^
=0
Qn -
Definition 5.4. Let E C R and / : E > R , xo be an interior point of E . —

We say that / ( •) is real analytic at x = XQ if there exists an r > 0 such that
oo
fix) = E= an{ x —
71 0
XQ ) " ? Vx G ( XQ — r, XQ + r ) C E.
If E is open, and / ( •) is real analytic at every point of E , we say that / ( •)
is real analytic on i£.
—
Proposition 5.5. Let f : E > R be real analytic at xo having the following
•
power series representation:

oo
/ (x ) = E= ° (* - «o )
71 0
»
B
? x e (x0 - r, XQ + r ).
Then / ( •) is infinitely differentiable at xo , i.e., for any fc 1, ( xo ) exists.

Moreover ,
f { n ) ( xo )
n! 5
djin 0. — >
Consequently, / ( •) admits the following Taylor expansion:
-
/ <* > E= 71 0
^ ( x - Xo )" 5 X e (x0 - r, x0 + r ) J
Series 183
and if / ( •) admits the following power series representation:

oo
/ (x ) = Ylbn ( x ~ x )n
° x G ( XQ — r , Xo + r)
n— 0
then it is necessary that
tin — Vn 1.
Proo/. Without loss of generality, we let XQ = 0. Then by definition ,

there exists a power series such that
oo
/ (*) = E= anxn .
71 0
By the uniform convergence of the power series in any compact subsets of

—
( r , r ) , one can obtain
z
71 ='0
n! —
x e ( r, r ) .
This proves the differentiability of / ( •) . The rest conclusions are clear .
Example 5.6. Let

0 x 0
f(x) = 1
e x > 0.
Then
0 x 0,
/ ( fc ) (x )
i4 ( x ) e
~
^, x>0
with Rk ( •) being a rational function . Hence in particular ,
/ ( fc) (0) = 0 fc 0.
Then for any <5 > 0,
y
“
- mxn = ^ /
71= 0
f {n)
n!
0 (x ) , xe ( 0, <5).
This means that / ( •) is infinitely differentiable at x = 0, but not real

analytic.
Next result is about the product of two power series.
Proposition 5.7. Let / , g : ( #o — r, X o f r ) -- 4 R be real analytic with
power series representations:
oo oo
f {x) = XM^ - ZO )"» 5(z ) = X— ~ 1
n =0 n 0
—
which are convergent in ( #o r , xo -F r ) , for some r > 0. Then the product
—
/ ( •) # ( •) is also real analytic on (xo r, xo + r ) with power series expansion:
oo
f ( x )g ( x ) = X= ° ( x
k 0
k ~ x )k 1
°
where
k
—
Ck
^^
771 =0
tlmbk — m i
Proof . Without loss of generality, we let xo = 0. For any 5 £ ( 0 , r ) i

— —
consider x £ [ r + 5, r 5] . Since the power series for / ( •) and g ( ) are -
uniformly and absolutely convergent on [ r 4- 5 , -fr <J] , for any £ > 0 , — —
oo ^ oo
X I°» I i*r + X i 6” i M ” < e \ x\ r - 5.

n = ./V +1 n = AT + l
Now i
- )(£*
N M N M 7V + M
(£ “) = £ £
Further ,
77
N+ M
=0
« *
777 =0
»*
n =0 m = 0
anbmxn+m =
^
/e = 0
X Cfex'' - f ( x ) g ( x )
fc = 0
N+M N oo M oo
Y=2 ckxk - ( x= anxn +=X+

k 0 77 0 77 TV 1
a » x" ) ( X bnXn + X
77 =0 77 = M +1
I
N oo M oo
| Xa"xn|(
77 =0 77
X
= M +1
IM M") + 1 x "xi (
77 =0
6
77
X
= TV +1
l°» IWn ) |
oo OO
+( X KIN") ( X IMM")
77 = + TV 1 n =M+ l
( 1 / 0*01 + e) e + ( Is 0*0 l + e)e + £2
< e [ |xmax
| r - <5
( 1 / 0*01 + lfl (*) l ) ] + 3£2 -
^
Series 185
Hence
oo
f ( x )g ( x ) = ckXk
k =0
5.2 Exponential and logarithmic functions

In this subsection , we construct exponential and logarithmic functions by
power series.
Example 5.8. Let
EM - E~ ^
n= 0
rn
(5.4 )
First of all, Stirling’s formula says

. . n\ en
lim =L (5.5)
n-> oo nnv7=2n 7T
Applying the above, we calculate
1
lim lim = 0.
n — oo ( n!) i —
n > oo
Hence, the radius of convergence for the series (5.4 ) is R = oo. In other
words, the series converges for all x G R. Further , it is clear that the
convergence is uniform in any compact sets, and the series of formal term-
by-term derivatives has the same form and therefore converges uniformly
in any compact sets. Hence, we have
E\x ) = E{ x) Vx G R.
Likewise, we can integrate term- by-term to get
f E ( x )d x = E ( b ) - E ( a ).
Ja
Further , for any x, y G R,
oo
> - = /
_°°_
E=
n 0
T
4
-n
r
\ /
_°°
m 0
. 'll 171\
E= *i E= +E=
r2' ( x + y )
k
- _°°_
k 0n m k
xnym
n\m\
EE E
— = n!( /c — n)! =
k 0n 0 k 0
k\ = E ( x + y ).
Thus
We also have
E( x ) =
Kf )r o.
1 = E ( 0) = E ( x - x ) = E ( x ) E ( -x )
leading to
E ( x ) > 0,
1 Vx G R.
E ( -x ) =
E( x )
Define the Euler’s number by the following:
2, 1
e = E ( 1) =
n\ £
n 0 —
Then , by induction , for any integer m > 1 ,
E( m) = E { l)m = em
and
e= E
which leads to
1
em .
i
E
m =
Consequently, for any rational number r =|> 0
-
*(?) «’ *( -!)
1
—e
Finally, by the continuity of the map x i-» E( x ) , we obtain that
_£
« .
£(x ) = e* Vx G R .
This function is called the exponential function.
Example 5.9 . Consider the following series:
Since
*> -t <=r -
n= 1
(* !)“ •
— (;) * = 1
lim
n > oo
Series 187
we obtain that the radius of convergence is equal to 1. Thus, the above

series is convergent for x G ( 0, 2 ) , and the function </? ( •) is real analytic on
( 0, 2 ) . Further , in this interval,
oo oo
—
n 1 1 1
n =1
Consequently, by the chain rule,
n =0
1 — (1 — x) x‘
1
Me*)'] = <fi'(ex)ex = -ex = 1 ° = <^ (1) = 0.
¥> ( e )
Hence
if(ex ) = x , x G (0, 2) .
-
This means that tp( ) is an inverse function of ex. We denote
( p( x)(0 , 2 ) .
= lnx, xG
We can extend it to ( 0 , oo ) as an inverse function of x t-> ex. This function
is called the logarithmic function.
5.3 The sine and cosine functions

In this subsection , we construct sine and cosine functions by power series.
First of all , we recall the set C of all complex numbers:
C
(
— { a -- ib | u, b G R},
j i 2 = -1,
with the usual rules of addition , multiplication , etc. Also, the conjugate of
a 4- ib is defined by
a + ib = a — ib
and the norm of a + ib is given by
|a 4- ib\ = \ Ja2 + b2 .
Next , we extend the exponential function ex from R to C:
00
rn
e* = z eC.
n== 0
It is clear that the series is convergent for all z G C. Now , we define:
eix + e-ix — (-l ) mx 2m
CM = 2
1
2n
^
=0
(ix ) n -f ( ix ) n
n! ^
E
m=0
( 2m )!
eix _ e XX
1 (ix )n — ( — ix ) n “ ( l)ma + — ;2m 1
«(*) = = E ( 2m + 1 ) !
2* 2i
^
n=0
n\
=
m 0
X G R.
—
Then (7, 5 : R > R are real analytic on R, and the following holds:
elx = C ( x ) + i S ( pc ) , Vx G R. ( 5.6 )
It is seen from the definition that
<7( x ) 2 + 5(x ) 2 = |ei:c | 2 = = c° = 1, Vx G R,
—
C'(x ) = 5( x ) , 5'( x ) = C' (x ) , Vx G R, ( 5.7)
<7( 0 ) = 1, 5( 0 ) = 0.
We claim that there exists an x > 0 such that
C ( x ) = 0.
If this is not the case, then from (7( 0) = 1 > 0, we must have
C ( x ) > 0,
Consequently,
Vx
^ 0.
S'( x ) = C ( x ) > 0 , Vx G R .
-
Hence , S ( ) is strictly increasing. By 5( 0 ) = 0, we must have
5( x ) > 0, Vx > 0.
Now , for any 0 < x < y , one has
f
S ( x )( y - x ) <
JX
S { t )d t = C ( x ) - C ( y )
^ l.
This will lead to a contradiction since 5( x ) > 0 and we may let y large.
Now , we let x > 0 be the smallest positive number such that C ( x ) = 0.
Denote
7r = 2 x.
Note that , at the moment , 7r is just a notation. Then by (5.7)
( ) = S(al) = ±l.
s| ( 5.8 )
Since
(7( x ) > 0, xe 0 , ) = [0, x )
| ( 5.9 )
by 5( 0) = 0 and 5'(x ) = (7( x) , we see that
5( x ) > 0 , X (o, j] ( 5.10)
which leads to
0 < 5( x ) < 1 X |).
(0, ( 5.11 )
Series 189
Thus,
e 2
Consequently,
e** + l = 0, e2 = 1.
Therefore
ez+2*i = ez V z e C. (5.12 )
Then
C { x + 2TR ) + i S ( x + 2TT ) = eix+2ni = e I X = C ( x ) + i S { x ) , VxeR
which yields
C( x -f 27T ) = C( x) S( x + 2TT ) = S( x ) Vx G R. (5.13)
Next , we claim that
VO < t < 2TT . (5.14 )
Suppose otherwise, there exists a t G (0, 27r ) such that
ei£ = 1.
We write
eu
4
cQ + iS0 == x + iy
with x, y G R. Then
1= = ( x + ii/ ) 4 = x4 + 4x3 (iy ) + 6 x2 ( i y ) 2 + 4x ( zt/ ) 3 + (i ?/ ) 4
= x4 — 6x 2 ?/ 2 + y4 -f 4i x y ( x2 — y 2 ) .
Hence, it is necessary that
x y ( x 2 - y 2 ) = 0.
On the other hand , since|G (0 , ) , by ( 5.9) and ( 5.10 ) , we have
0
^
< x , y < 1.
Thus , we must have
x2 - y 2 = 0, x2 + »2 = l,
which implies
1
x2 = y2 = - '
2
As a result , one has
6 1
1 = x 4 - 6x V + y4 =
j~ 4 + 4 = -1
a contradiction. This proves our claim (5.14 ) .
Further , we show that for any z £ C with \ z \ = 1, there exists a unique
t £ [0 , 27T ) such that
eu = z . (5.15 )
In fact , for any 0 t\ < < 27T,
e» ( t 2 -*l ) = e*t2 e-iti
which leads to
!
^
eUl eit .
êxistence
2
This gives the uniqueness. For the , we write

= x -Hy ,
2 x , y £ R, x 2 + y2 = \ z \ 2 = 1.
Case 1. x , y 0. On [0 , ] , C ( - ) decreases from 1 to zero (see (5.7) and
^
(5.8)) . Hence, for x £ [0, 1] , there exists a t £ [0, such that
C ( t ) = x.
Then by (5.7) and (5.10 ) , we obtain
5(f ) = Vi - c(f ) 2 = y ,
leading to
elt — x + iy — z ,
giving the existence for this case (see (5.6 )) .
.
Case 2 x < 0 and y 0. In this case,
-iz = 2/ 4- i (-x ) , y , -x 0.
Hence, there exists at £ [0, §] such that
eil = — i z .
Then
z= t 4-
\ e [0, r ).7
Case 3. y < 0. In this case,

— z = — x — iy = — x 4- i ( — y ) - y > o.
Series 191
Hence, by Cases 1 and 2 , there exists a t G [0 , 7r ) such that

z= . — eu
Then
_ _ eHt+ r ) .
= eit
2:
7
This completes the proof of (5.15) .

Finally, we identify ir .
To this end , we define a curve
7 0 = elt ,
( t G [0 , 27r ]
which is the unit circle in C. The arc length is given by
/" 2 TT /* 2 TT
/
JO * /
/
IT (*)I =
JO
dt = 27r .
This coincides the usual definition of 7T .
We denote
cosx = C ( x ) sinx = S( x ) Vx G R
and call them cosine function and sine function, respectively. These are
trigonometric functions.
Definition 5.10. Let T > 0. A function / : R -» C is periodic with period
T, or T - periodic , if
/ ( x + T) = / ( x ) x G R.
From the above, we see that cos x and sin x are periodic functions with
period 27r.
Exercises
1. Find a rational function / ( x ) such that
oo
/ (*) = £ nx
2
X < 1.
oo
n 1
—
anxn. Find a rational function / (x ) such that
2. Consider series
^
71 =0
f (x ) =
OO
x| <
=0
71
where i? is the radius of convergence for the series , to be determined , for
each of the following cases:
( i) Let an = 2n -f 1, n 0.
( ii) Let an + i = an + an _ i for n 1, ao = ai = 1.
( iii ) Let an + i = aan + ficin- i f r n 1 and ao , ai , a , /3 G M are given.
° ?
6 Fourier Series
We first introduce the following definition.
Definition 6.1. The following is called a trigonometric polynomial:

N
f (x) = ao +
^
where an , bn G C. If all coefficients
=1
71
2 cos nx +
an , 6n
^ sin nx ) 1
G
x6R (6.1)
R , the above polynomial / ( •)

is said to be real
Note that
cos nx 1/1 1 emx ginx 1 i cos nx
sin nx 2V i i— —
e mx ?
e — mx 1 —2 sin nx
( 6.2)
Therefore, the above (6.1) can also be equivalently written as

N
c
/ (*) = cneinx 5 x G R5 (6.3)
—
n= N
for some cn G C. It is clear that for / ( •) of form (6.3) to be real-valued for

all x if and only if
cn — Vn = 0, ±1, =L 2, • • • , =bN .
Note that the domain of the above function is R (with possibly complex
valued ) .
In this section , we will present a theory that represent functions in terms

of the following type series:
oo
71
E
=— OO
cneinx )
or
oo oo
^=0
71
an cos nx +
71 =1
We call the above type series trigonometric series, and when the coefficient
sin nx.
r
cn , or an and bn are properly obtained , they are called Fourier series.

Series 193
6.1 Inner product

Let C( [a , 6] ; C ) be the set of all continuous functions / : [a , 6] -» C .
Definition 6.2. For any / ( •) , <? ( •) £ C ( [a , 6] ; C ) , the following is called an
inner product of / ( •) and # ( •):
-
( / 5) =
b
f ( x ) g ( x ) dx ,
Proposition 6.3. The inner product ( • , •) satisfies the following:

V /, 5 C( M ; C ) .
( i ) ( Hermitian property ) For any / ( •) , g ( ) - G C( [a , 6] ; C ) ,
< / , 2> =
( ii) ( Positivity ) For any / ( •) G C ( [a , 6] ; C ) ,
< /, / ) > 0, and ( /, / ) = 0 / ( •) = 0.

-
( iii ) ( Linearity ) For any / ( •) , # ( •) , ft ( ) G C( [a , 6] ; C ) and a , /3 G C, it
fioids
(a / + fig , h ) = a ( /, ft ) + /? ( # , ft ) ,
( /, a + /3ft ) = a ( /, # ) + /? ( /, ft ) .
^
We now define
( yi /(x) i 2dx)
5
-
V/ ( ) C ( M ; C ).
This is called the L2- norm on C ( [a , 6] ; C ) , for which one has the following
result whose proof is straightforward as well.
Proposition 6.4. The L2 -norm || • ||2 satisfies the following:

( i ) ( Non-degeneracy ) For any / ( •) G C ( [a , 6] ; C ) ,
ll / ( - ) ll 2 > 0, and < =» / ( • ) = 0.
( ii) ( Cauchy- Schwarz Inequality ) For any / ( •) , g ( ) G C ( [ a , b\ ; C ) -
\ { f , 9 ) \ < \\ f h \\ g h -
( iii ) ( Triangle Inequality ) For any / ( •) , <? ( •) G C ( [a , 6] ; C )
11 / + 0 I 2 < 11 / H 2 + iisib.
( iv ) (Pythagorean Theorem) For any / ( •) , # ( •) e C ( [a , b] ; C ) , with

( f , g ) = 0 ( which is called orthogonal )
11 / -Mil = 11 / 111 + Nil -
( v ) ( Positive Homogeneity ) For any / ( •) e C( [a , 6] ; C ) and c G C,
Ik/ lb = |c| ||/ || 2.
We define
<k ( f > 9 ) = I / - 5I 2 = ( / |/ ( x ) - fl (a;)|2dx
- -
V/ ( ) , fl ( )
) 5
C( [o, 6] ; C) .
Then cfeO » • ) is a metric on C( [a , 6] ; C) .
Warning: C ( [a , b\ ; C) is not complete under the metric ( T ) The -
completion of C ( [a , 6] ; C ) under L 2- norm ( with the integral being Lebesgue ^
integral ) is a Hilbert space , and denoted by L2 ( a , b ; C ) .
—
Definition 6.5 . Let 0n : [ a , b ] > C be a sequence of continuous complex
functions. Then {0n} is called an orthogonal system of C ( [ a , b ] ; C ) if
If , in addition ,
(0ra 0n )
j
— /
J a
</>m ( x ) 4>n ( x )dx = 0, m 7 n. ^
b
( 0n ? ) — \ j n { x ) \ 2 dx = 1
() Vn
we call {0n} an orthonormal system.
The following example will give us the orthogonal system that we will
use below.
Example 6.6. ( i) Consider C ( [ — r , r] ; C ) .
7 7
For m , n eZ
<l, efa > = — le-
with m n , we have
F
J 7r
1
dx = -- e ~lx
2
^
= 0.
— 7T
> - /— —
*
ei ( m-n ) xdx = i ( m1 ^
— n ) ei
l imx ginx (m n)x
=0
^ •/ 7T — 7T
and
(evmx ^
imx
> -/ '
ei { m ~m )
xdx =
J
f— 7T
d x = 27r.
Series 195
Hence, { einx n e Z} is an orthogonal system of C ( [ — r , r]; C). If we define

7 7
1 ginx
en (x ) =
X/27T
xe[ — r , r] ,
7 7 n e Z, (6.4)
-
then {en ( ) | n G Z} is an orthonormal system of C( [ re , 7r] ; C ). —
—
( ii ) Consider C ( [ 7r , 7r] ; R ) . If m , n e N U {0} with m
^ n , then
7T
( cosmx , cosnx ) = cos m x cos nxdx

— 7T
J ^ cos( m + n ) x + cos( m — n)x

^ dx =0
and
7T 77
1 + cos 2mx
( cosmx , cosmx ) = cos2 mxdx = dx = 7r .
— 7T
— 7T 2
If m , n e N with m n , then
^ 7T
(sinmx , sinnx ) = sin mx sin nxdx

1
— IT
and
2
cos( m
7T
— n ) x — cos( m + n )x
,
7r
1
^ dx =0
— cos 2mx dx =
(sin mx, sin mx ) = sin 2 m x d x = 7r .
— 7T — 7T 2
Finally, for any n 1, m 0,
7T
(sinnx , cosmx ) = sin n x cos m x d x

— 7T
J
^^ sin ( n + m ) x + sin ( n
Therefore, {1, cosnx , sinnx n G N} is an orthogonal system of C ( [

R ) . Further ,
— m)x
^ dx = 0.
— r,
7 7T ];
1 cos nx sin nx
n 1 (6.5)
X/TT ’
5
\/27r x / 5r
form an orthonormal system of C ( [ — r, r]; R ).
7 7
6.2 Trigonometric polynomial approximation
Recall from Chapter 5, Theorem 3.6, we know that any / ( •) E C( [a , 6] ; R )
can be uniformly approximated by polynomials. The following gives an-
other type of approximation of continuous functions.
Theorem 6.7. ( Weierstrass) Let / ( •) E C ( [ 7r , 7T] ; C) . Then there exists
a sequence of trigonometric polynomial Tn ( - ) such that
—
lim
—
n > oo
•
—
! ! / ( ) Tn ( - )||oo = 0.
Proof . Let
1
$N ( X ) =
n=- ( N -1)
Note that
N -l N -l iV - l
2
—
^ | E en
v 7r
n =0
E
/r27=r n^\
, ra = 0
e i( n m ) x
n ,m =0
n— m E &
= ATeo 4- ( N — l ) ei 4- ( N — 2) e2 4- • • • 4- 2e./v-2 4- C N - I
4- { N — l ) e _ i 4- ( N — 2 ) e _ 2 H h 2e _ ( -2) + e _ ( jv _ i ) r
^ - ( iV - l )
N -l N -l N -l
E ^ — n)e„+E
n =0 n= l
- n) e-«= E
n =0
“ ) errf E—
n= 1 ^ ~
Inl ) e«
iV- l
= y: ( N - \ n\ ) en = V T K N Q N .
n= — N — )
( l
On the other hand , when 0 < | x| 7r ,

N -l N -l • ATr
1 eiNx - 1 1 , ( i V- l ) x
?= \
^ ^=>
mx Sin -FT
en e 2
n=0 V 7T
n 0 V 2n eix —1 sin
2
Hence
iV- i 2 • 2 Nx
i
1 sin 1
0 $N ( x ) = N | E e"( x) 2
2iW sin| 2
27r sin|
( 6.6 )
n =0
Also
AT
MO) = -
Further ,
Jr_
N~ l
/r $Ar ( a;)da; = E N\ i
/
VV1 6
inx
dx = 1.
* -*
• n = ( iV
— — l)
N J 2 ,K n
p"
Series 197
Next , by the continuity of / ( •) , we have
1 / 0*01 M, Vx G [ — 7r , 7r] 5
and for any s > 0, there exists a 6 > 0 such that
1 / 0*0 — f ( y )\ < e, \ x - y \ < S.
By (6.6 ) , for the above 5 > 0, there exists an iV
1
^ 1 such that
0 $N ( x ) 2 £5 \/ 6 |x | 7r .
2 NTT sin|
Now, we extend / ( •) periodically outside of [ — r,7 TT ] , and define
j +-
X K
( / * $w ) 0*0 =J f f ( t )$ N ( x - t )d t = f (x -r)
^ N ( r )d r
— TT J - TT -\- X
= (J
'
f TT ~\~ X
+ r
JTT
+X
'
) f(x - T )$ N ( r ) d T
TT +X
--
TT \ X
7T
f ( x - T )& N ( r )d T , x G R.
— 7T
Clearly, the above is a trigonometric polynomial. We now estimate
1 / 0*0 - / * $JV 0*0 I = / 0*0 [ f ( x - t )® N ( t ) d t

J — TT
7T
| / ( x ) - / ( x - y ) |$ jv ( y ) dy
——
/•
7T
’
O 5
/* < /* 7T
/— 2M e d y + / e$ jv ( j/ ) d 2/ + / 2M s d y
J TT J —S JS
*
4Me ( 7r — 5 ) + e ( 4M 7r + l ) e .
(
This proves our theorem.
Corollary 6.8. Suppose / ( •) G C( [ — 7r , 7T ] ; C) satisfying the following:

*
( /, e„ ) = 0, Vn G Z.
Then / ( •) = 0.
Proof . We can find a sequence Tm ( - ) of trigonometric polynomials of

the following form:
Nm
—
such that
Tm (^)
n ^^
=- Nm
Cmn &n { % )
1
11 / ( 0 - Tm ( - ) lloo < -
m
Vm
^ 1.
Then
11 / 0111 = ( f J - Tm ) = J[ \ f ( x )\ \ f ( x ) - Tm ( x )\d x
— 7T
27 T
2^ 11 / 110011 / - T,m oo —
< m ll / lloo 0.
Hence, one must have / ( •) = 0.

The above result tells us that if / ( •) e C ( [ 7r , 7r] ; M ) is perpendicular —
to every en ( ) , then it is necessary that / ( •) = 0. This means that the
«
orthonormal system { en } is complete , in some sense.
6.3 Fourier series

Now , if / ( •) is given by (6.3) , then it is continuous and periodic with period
2 n. We observe the following:
N c
— j^ ei( n-m ) xdx = 2ncm.
f—
J 7T
/ (*) c imx
dx = 12
=- N
Cn
J ~'K *
Hence
f— / (*) e — m x dx =
1 1
C
"
~ ( /, en ) — iV n < iV. ( 6.7)
2TT J 7r V 7T
^
In other words, such an / ( •) admits the following representation:
N
1
/ 0*0 = /27r
\ v^-JV
n=
( f i e n ) en -
Now , for any / : [ — 7r , 7T] —* C with | / ( - )| G LR ( [ , ] ; R ) , we define
7T 7r *
7r
/ («) = ( /, e„ ) = f ( x )e nGZ ( 6.8 )

02 J —
^ 7T
Series 199
which is called the n-th Fourier coefficient of / ( • ) . We write

oo oo
1
/ ( •) ~ X / ( n ) e« X / ( n ) einx
n =— oo V 2TT n =— oo
indicating the relation between / ( •) and the coefficients of the trigonometric
series. The right- hand side of the above is called the Fourier series of
/ ( •) . Note that / ( •) is not necessarily bounded , not necessarily continuous.

Hence, we usually might not have
oo
/ ( •) = X—
n = oo
(6.9)
However , we have the following result.

Theorem 6.9 . ( Fourier ) Let / ( •)
oo
G —
C( [ 7r , 7r ] ; C) . Then its Fourier series
f { n ) en converges in L 2 -norm to / ( •) , i.e. >
—
n = oo
N 2
^
lim f ( U ) en
—
N too
n =- N
9
( 6.10)
n 7r N 2
^
— ^/- X ^
lim / (x ) “ f ( n ) en ( x ) d x = 0
N > oo 7r
n=- N
Further , if
oo
X—
n= oo
I / MI < OO 5 ( 6.11 )
oo
then the Fourier series y f { n ) en converges to / ( •) uniformly, i.e.y
n =— oo
AT
lim
—
TV > OC
/~ X—
n= 'N
^
Z ( n ) e" OO
= 0.
Proof . For any £ > 0, by Weierstrass Theorem , there exists a trigono-

metric polynomial
No
!( •) = £
n =- 7V0
such that
,
1
l / - 7 || < £. 0O
\
Now , for any N > No , let

N
FN ( X ) = E—
n= N
f ( n ) en ( x ) .
Then for any \ m\ < N , we have

N
— FN , em ) = ( /, em ) — 2 f ( n ) ( en , em ) = f ( m ) - f ( m ) = 0.
(f
This implies that

=- n
^ N
No
( / - F n , FN - T ) — (f — Fn, E
n=- N 0
( f ( n ) - cn )en )
- ( No +1) N
+ (/ — Fff ,(
n= N
E— + E ) / (»)Cn ) = 0.
n= N0+ l
Consequently, r
r
\\ f - T \\l = \\ f - FN + FN - T \\l
= 1 / - *V |||+ H -fjv - T||l + 2 ( / - FN , FN - T)
= 11 / — Tiv| || + l -FV — T||l .
r
Hence
l / - TAr||Ull / - T||l < 27re.
This proves
|| - || 0.
—*oo / FJV 2 =
Jim
TV
Now , suppose (6.11) holds, then by Weierstrass M-test , we see that the
Fourier series is convergent uniformly to some function F( ) , which implies -
that
N 2 N 2
EN Rn )en - F 47T 92 E f ( n )en —F -* 0 5 r
n
—— 2
n ——
uniformly, as iV —* oo. By the uniqueness of the L 2 -limit , we must have
N
oo
-
F ( ) = / ( •) , proving our theorem.
Because of the above result , for any / ( •) £ C ( [ — r,
7 7T ] ; C ) , we may write
co
Js £— *">'
*
I
'
/ 00 = E—
n= oo
f ( n ) en ( X ) -
v n= oo
mx
Series 201
which is understood as ( 6.10) . Further , if / ( •) £ C ( [ 7r , 7r] ; M ) and the —

Fourier series is uniformly convergent on [a , 6] C [ — 7r , 7r] , then the above
equality holds for x £ [a , 6]. Moreover , we have the following result.
Theorem 6.10. ( Plancherel ) For any / ( •) £ C ( [

oc
— r,7 ]; C ) , the series
7T
^
2
| / ( n )| is convergent and the following Parseval’s equality holds:
n
—— oo
oo
I /I 2 =
n
E
= —
l / (") l -
oo
2
( 6.12 )
Proof . For any £ > 0, by Fourier Theorem , there exists an NQ 1 such

that
N
f - E / ( n ) e« ||2 £ VJV No .
n=- N
This leads to
TV
E—
n= N
f ( n ) en I / I 2 + £.
V On the other hand , since {en} is an orthonormal system , it is clear that
E / ( n ) e« = E i /(n) i 2-
\ =
n
— N =—
n N
Thus
( Il / l|2 - e ) 2 < E
=—
n N
l / (») l 2 < ( ll / lla + ® )2.
Consequently,
TV
(||/ ||2 - £ ) 2 < lim
— E— _ l / WI
2
A T >0 0 z ' < ( l / ll 2 + e) 2 -
n= N
Since £ > 0 is arbitrary, we obtain the Parseval’s equality.
Corollary 6.11. Let f : [ — 7r , 7T ] > R he Riemann integrable with | / ( - ) | 2

"
also being Riemann integrable. Then ( 6.10 ) holds.

Proof . We can find a sequence of continuous functions f m ( - ) such that

lim
—
m > oo !! / ( ) — / m ( - ) ||2 = o.
•
From the above theorem , we have

oo
n
E I /"»
=— oc
- /* (») I 2 = n /m ( - ) - /* ( ) Hi o m,
^ » oo.
Note that
r'
| / ( n ) - / ( n ) K -i
m r
J- l/ 0 /
TT
m
( - (x )|dx
1
27|
( jr (x ) - / (x ) i &) " 5
< — T
2 -> 0.
By Fatou’s Lemma (Theorem 3.6, (iii )) , we obtain

oo
E— i / m (o - / (oi 2 < ii /mo - / oiii

n = oo
Hence
/* 7T .
^ 2
11
^ (0 -
N
^VCOIIi = 27r _ r
l
/
J 7
n
S—
—
oo
N
[/’m ( n ) - / ( n ) ] e — i n x dx r
E— l /mN - / (n )|2 < E m

l / (n) - / (n) l 2 l! / m
( ' ) - / ( ) lli
*
-
Also
n= N n = oo
—
F
OO OO
n
E
=— oo
l / (») l 2 =
n
E (2 | /
=—oo
m
( n )|2 + 2| / m ( n ) - / ( n )|2 )
= 211 / ( Oil!+ 2||/ ( - ) - / ( Olla
m 4||/ (.) H 1 + 6||/ m (.) - / Olli < oo.
Then /*•
11 / 0 - 0112 ^
^ 36||(11 /0-- roiil++ iiro- OIl 3 + PTO - «) ll 3)
/o /mOll 2 l /mO W ) lli
^
Consequently,
f ( - ) Fn ( - ) \\U 6 || / 0 - / m 0 ll 2 -
- \\
lim ~
N + 00
Therefore, one must have ( 6.10 ) .
f
Series 203
The above result tells us that when / ( •) , together with | / ( ) | being -

Riemann integrable, we can write ( 6.9 ) , as long as it is understood in the
sense of ( 6.10 ) .
From the relation ( 6.2 ) , we see that for any / ( •) E C( [ — r, r]; R ), one
7 7
may write (see ( 6.8) )
oo ,
1 1 °° r 7T
E
— /,
E(
f { x) = f ( n )em x f ( y )e ginx
„= — oo 27T J
n = oo ^
Note that
7T
/ ( 2/ ) e em x + e — inx
— 7T
7T
f ( v )( e — my
+ einy )dySJ cos nx 4- ( J
*
f ( y )i ( e ~ iny - einy ) dy ) sin nx
Hence
cos nx 4- 2 ( J f ( y ) sin nydy' sin nx . j
00
/ (*) =
1
27T i E=
i
+ 7T “
n l
cos nx
1 °°
+- V4 sin nx
7T
^ “
n= l
oo oo
=a ° + nX= l ] an cos nx + E
n= l
s*n nx
with
/•*
7T
1 1
a
,/ (</ ) <% an — 7T J/- pi
f ( y ) cos nydy ,
° 2 rr L 7T (6.13)
1
/ ( / ) sin nydy , n 1.
7T
— 2 7T
From the above, we can restate Theorems 6.9 and 6.10 as follows.
— —
Theorem 6.12. Let f : [ 7r , 7T ] > R , together with | / ( - )|2 be Riemann
integrable, let an , bn be defined by ( 6.13) . Then the following are true:
( i) The ParsevaTs equality holds:
oo
2na 20 + n J2 { a l + bl ) = ll / ll!- ( 6.14)
n=l
1
oo
( ii) The Fourier series do + an cos nx + bn sin nx ) converges to / ( •)
n= l
in L2 -norm , i.e.,
N
lim
iV —> 00
|(ao + ] n= 1
^ ( an cos n • +bn sin n •
W« IL = o.
oo
+ l &nl ) < oo and / ( ) <£ C ( [ -n , 7r]; R ) , then
( iii) If
^
2 (|
=
n 1
0n|
N
•
f
lim
N —> oo
(a 0 +
n= l
an cos n • +6n sin n / OO
=o
i.e., the Fourier series converges to / ( •) uniformly.
It is interesting that in order for a trigonometric series to be a Fourier

series of some continuous function , ( 6.14) has to be true, in particular , the /
series on the left-hand side must be convergent.
Exercises
1. Let / : [ — 7r , 7r] -> R be continuous. Let
1 7r
i rn
GO = 727TT " f ( x )d x, ak =- f ( x ) cos fcxdx ,
1
—
' 7T
7r 7T
— 7T
bk = - / (x ) sin /cxdx, k 1.
* J -7 T
—
Let 7r < c < x < 7r . Show that
r
J /m *
= a0 ( x - c) +
OO
—^ is convergent , where bk is defined as in

2. Show that the series 7
fc = i
the above problem for some continuous function / ( ) on [ — r, ]. • 7 7T
3. State and prove the results of Fourier series for C( [d , 6] ; R ) , i.e.,

replace [ — 7r , 7r] by general [a , b\ .
4. Let / : [ — , ] —> R be continuously differentiable with / (
7T 7r — TT ) =
/ 7r ) . Suppose
(
00
/ ( x) ~ do 4-
^^
k =1
dfc cos kx + bk sin kx ^j .
Series 205
Show that
°o
~ fe —^
f' ( x )
—
oo .
: r sin
^=i
7
kx
^ a /c cos + bk sin .
is convergent on [0 , 2-7r] , and is convergent

5*. Show that
^—
—
k 2
In k
uniformly on any [<5, 2 n 5] for any small 5 > 0. Is this a Fourier series of
some continuous function on ( 0 , 27r ) ? Why?
\
’S
y
\
Appendix
The Real Number System
1 Natural Numbers
1.1 Peano axioms

Let us begin with some basic notions.
• A collection of certain type objects is called a set.1 If A is a set , and

an object a is in A , we denote a A. In this case, we call a an element
in A. If a is not an element of A , we denote a £ A. If A contains at least
one element , we say that A is non-empty. If A contains no element , we say
that A is empty, denoted by A = 0.
•Let A and B be two sets. Denote A\ B , called the difference of A and

B , to be the set that contains all the elements in A but not in B . If any
element of A is in B , we say that A is a subset of B , denoted by A C B ,
or equivalently, B D A. In the case that A is a subset of B and B is also a
subset of A, we say that A and B are equal , denoted by A = B . If A and
B are different , by which we mean that at least one element in A is not in
B , or at least one element in B is not in A , then we denote A B . In the
case that A C B and A
^ ^
B , we call A a proper subset of B , denoted by
A c B.
•Let A and B be two sets. Let / ( ) be a rule that assigns each element
•
ae A a unique element b G B , denoted by / (a ) = b. We call such an / a

map from A to B , denoted by / : A -> B .
Now , let N be a non-empty set . On N, we introduce a binary relation

called equal, denoted by “=” , which has the following properties:
1We are satisfied with the definition of set at such
a level and we have no intention to
introduce axioms for that .
207
• ( Reflexivity) For all n G N, n = n;
• (Symmetry) For all n , m G N, if n = m , then m = n;
• (Transitivity) For all n, m, f G N , if n = m and m = £ , then n = £ .
We now introduce the Peano Axioms for the set N.
. —
Axiom 1 A map a : N » N is defined such that for all n , ra G N,
•
n — m if and only if a ( n ) = <r ( ra ) . The map a is called a successor

function, and for every n G N , cr ( n ) G N is called the successor of n.
Axiom 2. There is a special element in N, denoted by 0 , such that for

all n e N, a ( n ) 0.
^
Axiom 3. ( Principle of Mathematical Induction ) Let K be a
non-empty set having two properties: (i) 0 G K , and ( ii) n G K implies
cr ( n ) G X. Then AT D N.
Let N satisfy the above Peano Axioms, and let N = N \ {0}. We call N
the natural number set and any element in N is called a natural number.
Another way of stating Axiom 3 is as follows.

Axiom 3'. If P ( n ) is a statement depending on n G N such that
• P(0) is true, and

• for every n G N, if P { n ) is true, so is P( cr ( n) ) ,
then P( n ) is true for every n G N.
We will denote
<J (0) = 1, <T ( 1 ) - 2, <J ( 2 ) = 3, tr ( 3 ) = 4, •• •
and so on. By Axioms 1 and 2, we see that
K = {0, 1 , 2, - - - } = {0 , <7 (0 ) , <7 ( <r ( 0) ) , • • • } c N .
On the other hand , let P( n ) be the statement “n G K ” . Then P( 0 ) is true

and if P { n ) is true, P ( a ( n ) ) is true. Hence, by Axiom 3', P ( n ) is true for
all n G N , which means NCL Therefore, we obtain N = K , i.e. ,
N = {0, 1 , 2 , - - - }, N = {1 , 2, 3, - . . }. (i.i)
The Real Number System 209
1.2 Addition and multiplication

In this subsection , we introduce two binary operations in N. First , let us
introduce the following.
Definition 1.1. For any given me N, define

0 + m = m. ( 1.2)
Inductively, suppose n + m is defined . Then we define
cr ( n )+ m = cr ( n + m ) . (1.3)
The above defined operation “+” is called the addition on N.
For the addition defined above, we have the following result .
Lemma 1.2. ( i) For any n e N,
n + 0 = n. (1.4 )
( ii) For any n , m e N,
n + <r ( m ) = cr ( n + m ) . (1.5)
Proof , (i) Use induction on n.
For n = 0: By (1.2 ) , we have
0 + 0 = 0.
Now , suppose n + 0 = n. By (1.3) and induction hypothesis,
cr ( n ) + 0 = o( n + 0 ) = cr ( n) ,
proving (1.4 ) .
( ii) Fix m , using induction on n.
For n = 0: By (1.2 ) , one has
<r ( 0 + m ) = cr ( ra ) = 0 + cr ( m ) .
Hence, the statement holds for n = 0. Now suppose
n + cr ( ra ) = cr ( n + m ) .
Then the above, together with (1.3) , gives
cr ( n ) + cr ( ra ) = a ( n + cr ( m ) ) <j ( a{ n + m ) ) = a (cr ( n ) + m ).
This completes the induction.
Combining the above lemma with the definition of addition , we have
0+n = n -f 0 = n VneN ( 1.6 )
and
< j( n ) -f m = n + a ( m ) = cr ( n + ra ) , Vn , m G N. (1.7)
The above leads to the following result on the addition of the elements in
N.
Proposition 1.3. For any n , m , £ £ N,
( i) ( Commutativity) n + m = m -f n.
= n + (m + <) .
( ii) ( Associativity) ( n + m ) + f
( iii) ( Cancellation Property ) n + m = n + £ if and only if m = i .
Proof , ( i ) Fix an m G N, we use induction on n.
For n = 0: It follows from (1.6 ) .
Now , suppose
n+m =m+n Vra e N.
Then it follows from (1.7) that
o ( n ) + m = o( n + m ) = cr ( ra + n ) = m -f cr ( n ) .
This proves ( i) .
( ii) Fix m, < G N, we use induction on n.
For n = 0: By the definition of addition , we have
0 + ( m -F £ ) = m 4- 1 — (0 -f m ) + L
Now , suppose
( n + m ) + ( = n + ( m + <) .
By the definition of addition , and induction hypothesis, we have
(cr ( n ) + m ) 4- 1 = cr ( n + m ) + £ = cr ( ( î + m ) -f ^)
•
= < ( n + ( m + ) ) = <r ( n ) + ( m + ).
j
^ ^
Hence, ( ii) follows.
(iii) Use induction on n.
For n = 0: By definition of addition , one clearly has
= 0+f
0+m m = £.
Thus, the statement holds for n = 0. Now , suppose that
n + m= n +l m = t. ( 1.8 )
We want to show that
a ( ri) + m = a ( n ) -f t m = l. (1.9)
By (1.3) , the left- hand side of (1.9) implies
cr ( n + m ) = cr ( n ) + m = a ( n ) + £ = a ( n + ) . ^
By Axiom 2, one must have the left- hand side of (1.8) which leads to m = t
by induction hypothesis (1.8). Conversely, if m = £ , then the left- hand side
of (1.8) holds ( by induction hypothesis) . Hence, by Axiom 1,
a( m ) = a( £ ) a( n + ra ) = <j ( n + ( ).
Consequently, by (1.3)
a( n ) + m = a ( n + m ) = a ( n + £) = cr ( n ) + £ ,
proving our claim .
From Lemma 1.2 , we also have the following interesting corollary.
Corollary 1.4. For any n £ N,
a { n ) = n + 1. ( 1.10)
—
Proof . By (1.4 ) (1.5 ) , we have
o( n ) = o( n -F 0 ) = n -f cr ( 0 ) = n + 1.
This proves our conclusion.
Because of the above corollary, from now on , we will use n + 1 instead
of o{ n) . We now introduce another binary operation on N.
Definition 1.5. Let m e N. Define

0 x m=0 (l.n )
Suppose n x m i s defined . Then define
( n + 1) x m = ( n x m ) 4- m . ( 1.12)
The above defined operation “ x ” is called the multiplication in N. Sim-

ilar to Lemma 1.2, we have the following lemma.
Lemma 1.6 . (i) For any n G N,
nx 0 = 0 nxl = n. (1.13)
(ii) For any m , n £ N,
n x ( m + 1) = ( n x m ) + n. (1.14)
Proof , ( i ) We prove the first equality in (1.13) by induction.
For n = 0: By definition , 0 x 0 = 0.
Suppose n x 0 = 0. Then by definition and induction hypothesis,
( n + 1) x 0 = ( n x 0) + 0 = 0 + 0 = 0.
This proves the first equality in (1.13) .
Now , we prove the second equality in (1.13) by induction.
For n = 0: By (1.11) , we have
0 x 1 = 0.
Suppose the conclusion holds for n. Then by definition and induction hy-
pothesis, we have
(n + l ) x l = (n x l ) + l = n + l
proving the second equality in (1.13) .
( ii) Fix m , use induction on n.
For n = 0: We need to show that
0 x ( m -f 1) = (0 x m ) + 0,
which is true since both sides are 0.
Now suppose
n x ( m + 1) = ( n x m ) 4- n.
Then for n + 1, by definition of multiplication , the induction hypothesis,
and Proposition 1.3,
( n 4- 1) x ( m + 1) = n x ( m + 1) -f ( m + 1) = ( ra x m ) + n + ( m + 1)
= ( n x m ) + m + ( n + 1) = ( n + 1) x m + ( n + 1) .
Hence, our conclusion follows.

Prom the definition of multiplication and the above proposition , we have
0xm =mx0=0 lxm =m x l =m Vra G N. (1.15)
Now , we are able to prove the following result .

Proposition 1.7. For any n , m , £ G N,
( i ) ( Commutativity) n x m = m x n.
( ii) ( Distributivity) n x ( m + £ ) — n x m + n x £.
( iii ) ( Associativity ) ( n x m ) x £ = n x ( m x £ ) .
Proof , ( i) Fix m , and use induction on n.
For n = 0: It is trivial since both sides are zero (see (1.15) ).
Suppose
n x m = m x n.
Then by (1.12 ) and (1.14) , one has
( n 4- 1) x m — ( n x m ) 4- m = ( m x n) + m = m x ( n 4- 1).
This proves ( i ) .
(ii ) Fix n , m G N and use induction on £ .
For £ = 0: We need to show that
n x ( m + 0 ) = n x m + n x 0.
This is true since both sides are equal to n x m. Suppose
nx ( m + £ ) = n x m + n x £ .
Then by (1.14 )
n x ( m + ( £ + 1) ) — n x ( (m 4 £ ) + l ) = n x ( m 4- £ ) + n
-
= n x m + n x £ + n = n x m + n x ( + 1) . ^
This proves ( ii) .
(iii) Fix m , £ G N and use induction on n.
For n — 0: We need to show
(0 x m ) x £ = 0 x ( m x £ )
which is true since both sides are zero.
Now suppose
( n x m ) x £ = n x ( m x £ ).
Then by the proved ( ii) ,
( ( n + 1) x m ) x £ = ( ( n x m ) -f m ) x £ = ( n x m ) x £ + m x £
= n x ( m x £ ) -f m x £ = ( n + 1) x ( m x £ ) .
1.3 Orders
We now look at another important feature of elements in N, which is de-
scribed by the following binary relation.
Definition 1.8. Let n , ra G N. We say that n m (or m n ) if there

exists an £ G N such that
n = m + £,
and we say that n > m (or m < n ) if
n > m, m.
n
^
In particular , when n > 0 , we say that n is positive.
Let us now present the following simple lemma.
Lemma 1.9. ( i) Let n G N. Then , n 0; and n > 0 if and only if n ± 0.
( ii) For any n G N, there exists a unique me N such that n = m + 1.
( iii ) Let n , m G N. Then n > m if and only if n = m + £ , for some £ > 0;
and if and only ifn m + l .
^
(iv ) Let n , m G N. Then n m if and only if n -f 1 > m.
( v ) Let n, m G N. Then n + m > 0 if and only if either n > 0 or m > 0.
Proof , ( i) For any n G N, since n = 0 + n , by definition , n 0; Further ,

if n 7 0 , then by definition , n > 0. Conversely, if n > 0, by definition
^ ^
n / 0.
( ii) By Axiom 1, if m G N exists such that n = m + 1, it must be unique.
We use induction on n .
For n — 1: We may take m = 0 since 1 = 0 + 1.

Now , suppose n > 0 and there exists a unique m E N such that n =
m + 1. Then , for n + 1, it suffices to take £ = m + 1 to get
n+l = ( ra + l ) + 1 = + 1.
^
This completes the induction .
( iii) By definition , n > m if and only if n = m + £ for some £ £ N and
m. Hence, £ 0 , or £ > 0. ( Otherwise, n = m. )
n
^Next, ^
suppose n = m + £ with £ > 0. Since £ > 0, by (ii) , one has
some I G N. Then
£ = ^ + 1 for
n= m+ £ = m + ( £ + 1) = ( m + 1) + £ ,
which means n m + 1.
Conversely, if n m + 1, then there exists an £ e N such that
n = ( m + 1) + £ = m + ( £ + 1) ,
and + 1) is positive, which gives n > m.
( iv ) The proof is similar to that of ( iii ) , and it is left to the readers.
( v ) Suppose n + m > 0 , but n = m = 0. Then n + m = 0 + 0 = 0,
a contradiction. Conversely, if , say, n > 0, then we show n + m > 0 by
induction on m .
For m = 0: In this case
+ m = n + 0 = 0 + n = n > 0.
7i
Now , suppose that n + m > 0. We want to show that n + ( m + 1) > 0. In

fact , by the associativity of addition , one has
= (n + m ) + 1,
n + ( m + 1)
which cannot be 0 ( by Axiom 2 ) . Hence, n + ( m + 1) > 0. This completes
the induction.
The following proposition collects the three most important properties
of the relation introduced in Definition 1.8.
Proposition 1.10 . Let n , m , G N. Then
^
( i) ( Reflexivity ) n n.
( ii) ( Transitivity ) If n m , m > £ , then n £.
( iii ) ( Anti-symmetry) If n
^ m and m ^ n, then n = m.
Proof , ( i ) Since n = n + 0 with 0 G N, we have n
( ii ) Under our conditions ,
^ n.
n= m+ m m = i + £,
for some m , t G N. Then
n = m -\- m = £ + £ + m.
Since m + i G N , we see that n by definition.
( iii) If n ra , then together with n ra , we have n > m and thus there
^
exists an ra G N such that
n = m + ra.
Likewise, there exists an n G N such that
ra = n + n = ra + ra + n.
By the cancellation property of addition (see Proposition 1.3, ( iii)) we
obtain
n+m=0
with both n > 0 and m > 0. This contradicts Lemma 1.9, (v ) .
Proposition 1.11. ( Trichotomy ) Let n , m G N. Then exactly one of the

following holds:
n > m, n = m, n < m. (1.16)
Proof . First , we show that for any n , m G N, at most one of relation
in (1.16 ) can hold . In fact , if n > m , then both n = m and n < m fail; if
n < m , then both n = m and n > m fail; and if n = m , then both n > m
and n < m fail.
Next , we show that at least one of the three is true. To show that , we
fix m and use induction on n.
For n = 0: For any m G N, m 0 since m = m -f 0 with 0 G N. So
^
either m = 0 , or m > 0 , i.e., either m = n or m > n , with n 0. —
Now , suppose the statement holds for n G N , i.e., one of the three
relations holds between n and m.
In the case n > m , we have ( n + 1) > n > m , leading to ( n + 1) > m.
In the case n = m , we have ( n + 1) > n = m leading to ( n + 1) > m.
In the case n < ra , by ( iii) of Proposition 1.9, we have ( n + 1) ra ,

which means m = ( n 4- 1) + £ for some < G N. If £ = 0 , then ra = ( n + 1).
If £ > 0 , then
( n + 1) < ra.
Either case leads to a completion of the proof .
The following proposition collects some compatibility properties of the
addition and multiplication with the order and the equality.
.
Proposition 1.12 Let n , m , £ G N.
( i) ( Order Preservation under Addition ) n > m if and only if n + £
ra + £.
( ii) ( Order Preservation under Multiplication ) If £ > 0, then n > m if
and only if n x t > m x £ . In particular ( taking m = 0 ) , n x £ > 0 if and
only if n , £ > 0.
( iii) ( Cancellation Property ) If £ > 0 , then n x £ = m x £ if and only if
n = m.
Proof , (i) By definition, n m if and only if
n = m + m, (1.17)
for some me N. Then , by the associativity and commutativity of addition ,
n + £ = ( m 4- fh ) + £ = ( m 4- £ ) + fh .
Hence, n + £
such that ^ m + £. Conversely, by n + £ ^ m -1- £ , there exists an fh EN
n -f £ = ( m + £ ) + m = ( m + m ) + £ •
By the cancellation property of addition (see Proposition 1.3) , ( iii) , one
obtains (1.17) , which means n m.
^
( ii) Suppose £ > 0 and n > m. Then by Lemma 1.9, ( ii)-(iii) and
Definition 1.8, £ = £ 4- 1 for some I G N , and n = m + fh for some fh G N.
Hence,
n x £ = ( m + m ) x ( £ + 1) = m x ( ft + 1) + m x { £ 4- 1)
—m x £\- - fh x £ + fh.
Since fh x £ e N and fh > 0, by Lemma 1.9, (v ) , one has fh x£+ fh > 0.

Therefore, by Definition 1.8, n x £ > m x £ .
Conversely, let n x £ > m x £ , with £ > 0. We claim that n > m.
Otherwise, by Proposition 1.11, one has n m. Then there exists an
h G N such that m = n + n . Hence,
m x £ = ( n + n ) x £ = n x £ + n x £.
This implies n x £ m x £ , contradicting n x £ > m x t .
( iii) If instead , n > m , then by ( ii) , we have
n x £ > m x £,
a contradiction. Likewise, n < m is also not possible. Hence, we must have
n = m. The other direction is obvious.
Next , we look at another interesting result on N.
Proposition 1.13. ( Euclidean Algorithm ) L e t n G N, q G N. Then
there exist m , r G N such that
n = m x q + r, 0 r < q. (1.18)
Proof . Fix q > 0, use induction on n.
For n = 0: We take m = 0 and r = 0. Thus , (1.18) holds.
Suppose for n the following holds:
n = m x q + r, for some 0 r < q.
Then for n + 1, we have two cases:
( i) q = r + 1. For this case,
n -\- l = ( m x q + r ) + l = m x q + q = ( m -\- l ) x q .
( ii) q > r + 1. For this case,
= ( rax ? + r ) + l = rax <7 + ( r + l ) .
n+l (
Hence, we obtain the result .
2 Integers and Rationals
2.1 Integers
Recall that on the set N , we have a binary operation which maps

every given pair ( n , m ) to a number n + m G N. Interestingly, 0 is a special
element in N that has the property
n+0 =0+n = n V n G N.
We call 0 the unit of addition in N. Further , for n ra , one can find an

t G N so that n = m + L However , this cannot be done if m > n . In
^
particular , for m > 0, one could not find an l G N such that
0 = m+ L
Therefore, we would like to find a bigger set containing N so that the above
is possible. To this end , let us introduce the following set
NxN = { ( m , n ) m , n GN}
which is called the Cartesian product of N with itself . On N x N , we intro-
~
duce a relation as follows:
( m, n ) ~ ( m', n') m + n' = m' + n. ( 2.1 )
We have the following result concerning this relation.
Proposition 2.1. The relation ~ is an equivalence relation on N x N, i.e.,
( i) ( Reflexivity) ( m , n )~ (m, n ) for any (m , n ) E N x N;
( ii ) (Symmetry) ( m , n ) ~ ( m', n') if and only if ( m', n') ~ ( m , n ) ;
( iii) ( Transitivity ) ( m , n ) ~ ( m', n' ) and ( m', n' ) ~ ( m", n") implies
that ( m , n ) ~ ( m", n") .
Proof , (i ) For any m , n E N, by the reflexivity of “=” in N, we have
m + n = m + n.
This leads to
( m , n) ~ ( m, n) .
( ii) If ( m , n ) ~ (m', n') , then
m + n' = m' 4- n ,
which , by the symmetry of “=” in N , is the same as
m' + n = m + n' .
Hence, ( m', n') ~ (m, n ).
( iii) Suppose ( m , n ) ~ ( m', n') and ( m', n') ~ ( m", n") . Then
m + n' = m' + n , m' -f n" = m" 4- n'.
This leads to
m + n' + m' + n" = m' + n + m" + n'.
By the cancellation property of addition ( Proposition 1.3, (iii ) ) , we obtain
m + n" = m" + n.
Thus, ( m , n ) ~ ( m", n") .
Having the above, we now introduce the following definition .
Definition 2.2. For any ( m , n ) G N x N , let

nn © n = { ( m', n') G N x N | ( m', n' ) ~ ( m , n )} ( 2.2 )
which is called the equivalent class of ( m , n ) under the equivalent relation
~ defined by ( 2.1) . Any element ( m' , n f ) G m 0 n i s called a representative
of its equivalent class. Let
Z = { m Q n | (m, n) G N x N}.
Any element m © n G Z is called an integer.
Lemma 2.3. Let m , n , m' , n' G N. Then
( m, n) ~ (m', n') mQn = m' Q n' . ( 2.3)
In particular , for any m , n , t G N,
( m + ) © ( n + I) = m © n .
^ ( 2.4 )
Proo/. Necessity. For any ( m", n") G m © n , we have
( m", n") ~ (m, n ) ~ ( m', n').
Thus, ( m" , n" ) G m' © n', leading to
mQn C m' O n' .
The other inclusion direction is similar.
Sufficiency. Since
(m, n ) E mQn= m' Qn ',
we obtain ( m , n ) ~ (m', n').
The last conclusion is obvious.
Let
Z0 = {m © 0 m G N} C Z.
If we define a map : N —> Z by•
( p ( m ) = m Q 0, Mm G N
then the image <p ( N ) = Zo Moreover , ( f ( m ) (/2 ( 777') if and only if m m' .
- ^
Hence, N can be identified as ZQ , a subset of Z. Because of this, hereafter , ^
we directly write
71 © 0 — -
77 VnS N. ( 2.5)
In particular , one has
(0 © 0 ) = 0, ( 1 © 0 ) = 1.
Next , we extend the addition and multiplication from N to Z.
Definition 2.4. For any ( m Q n ) , ( p Q q ) E Z, define addition and multi-
plication of them as follows:
( m © n ) -F (p © q ) = ( ra + p ) © ( n + <? ) ,
( m Q n ) x ( p Q q ) = ( m x p + n x q ) Q ( m x q + p x n ).
The following proposition is necessary before going further.
Proposition 2.5 . Addition and multiplication of integers are well-defined .
Proof . Let
m © n = m! © n' p e q = p' O q' ,
which means (777 , 77 ) ~ ( m' , nr ) and (p, q ) ~ ( p' , q' ) - Thus,
77i + n' = m' + - 77 p + q' = p’ + q .
Add the above together , we have
( m + n') + ( p + q' ) = ( m' + n ) + ( p' + q )
which is the same as
( m + p ) + (77' + q' ) = ( 777' + p' ) + ( n + q ) .
Hence
( m + p) © (n + q) = ( 777' + p') © ( 77' + q' ) .
By definition , this means
= ( 777' © n') + (p' © </ ).
( TTT, © 77) + ( p © q )
Therefore, the extended addition is well-defined on Z.
Now , for multiplication , we show the following: If
777 © 77 = m' © n',
then for any p Q q GZ,
( m © n ) x ( p © <7 ) = ( ra' © n') x (p © q ). ( 2.6 )

( m x p + n x q ) Q ( m x q + n x p)
= ( ra' x p + n' x q ) Q ( ra' x q + n' x p )
m x p + n x q + m/ x q + n' x p = m' x p + n' x q + m x q + n x p.
By the rules of addition and multiplication on N, we have
( m -f n! ) x p + ( m' + n ) x q = ( m! + n ) x p + ( m + n' ) x q .
Since ra + n' = m' + n , the above holds true which means that ( 2.6 ) holds.
Therefore, the extended multiplication is well-defined on Z.
Let us look at the following special cases:
( n © 0 ) + ( ra © 0 ) = ( n + m ) © 0, Vn , m e N.
( n © 0 ) x ( m © 0) = ( n x m ) © 0,
Hence, besides that N can be identified as the subset Zo of Z, the additions
and multiplications defined on N and Z are compatible.
The following definition gives a new feature of the integer set Z that the
set N does not have.
Definition 2.6 . ( Negation of integers) For any ( m © n ) G Z, define
— ( ra © n ) = n © m,
which is called the negation of ( ra © n ) . In particular , if n = n © 0 with
n G N ( which is also called a non- negative integer from now on ) , we denote
— ( n © 0 ) = (0 © n ) = — n ,
and call it a non- positive integer. If a G Z is non- positive and is different
from 0, we call it a negative integer, denoted by a < 0. Any natural number
is called a positive integer.
We now collect some laws of addition and multiplication for integers.

Proposition 2.7. ( Algebraic Laws for Integers ) Let a , b , c G Z. Then
CL -*c b —b CL ,
( A + 6 ) -+- c = ci + ( 6 + c) ,
—— —
62. f 0 = 0 I- tX = G ,
— —
a + ( a ) = ( a ) -f a = 0 ,
a x b = b x a,
( a x I)) x c = a x ( & x c) ,
axl = lxa = a ,
a x ( 6 + c) = a x f r + a x c.
P r o o f . To prove the first , we write a

definition of addition , we have
— ra © n , b = p © <?. Then use
a + b = ( m Q n) + ( p Q q) = (m + p) Q ( n+ q )
= ( p + m ) Q ( q + n ) = ( p Q q ) + ( m Q n ) = b + a.
The proofs of others are similar , which are left to the readers.
According to the above, we have
( 0 © 1) = 1 —
and
(-1) x n = n x (-1) = ( n © 0 ) x (0 © 1)
= ( n x 0 + 0 x 1) © (n x 1 + 0 x 0) = (0 © n) = — n.
Further , for any ra © n G Z, we have
m©n= — n) © 0 = m — n
( ra m n
^ ( 2.7)
0 © ( n — m ) = — (n — m), m < n.
Hence, the following representation holds:
Z = {. . . , -3, -2, -1, 0 , 1, 2, 3, . . }. * ( 2.8 )
We now define subtraction on Z by the following:
a - b = a + ( -b ) Va , 6 G Z,
—
and call a b the difference between a and b. Note that if a = m © n and
b = p © q for some m , n , p, # G N , then
a — 6 = ( m © n ) — ( p © <?) = (m © n) + ( q © p) = ( m + g ) © ( n + p ) .
\
The following proposition collects some more laws for integers.
Proposition 2.8. Let a , 6, c G Z, with c 0. Then
a — b= 0
^ a = b. ( 2.9 )
—
( a ) x b = a x ( 6) — = — ( a x 6) . ( 2.10 )
a x b=0 a = 0, or 6 = 0. ( 2.11)
a x c= b x c a = b. ( 2.12 )
Proof . Let n , m , p, q G N such that
a = m © n, b = p Q q.
For ( 2.9 ) , we look at the following:
0©0 =0 = a — 6 = ( ra © n ) — (p © / ) = ( ra © n ) + ( — (p © q ) )
(
= ( m © n ) + ( q © p) = ( m + q ) © ( n + p).
This is equivalent to
m + q = n + p,
from which , one has
a = m Q n = p Q q = b.
Hence, ( 2.9 ) holds. Next , for ( 2.10) , we observe the following:
( — a ) x b = ( — ( m © n ) ) x ( p Q q ) = (n © m) x ( p G q )
—
= ( n x p -j- m x q ) Q ( m x p -i- n x q ) = ( m O n ) x ( q Q p ) = a x ( 6) .
Also we note that
-( a x 6 ) = ~ ( ( m © n ) x (p © g ) )
—
= ( ( m x p + n x q ) © (n x p + m x q ) )
= {n x p+ m x q) Q (m x p+ n x q)
= ( n © m ) x (p © ) = ( a ) x 6.
^ —
This proves ( 2.10 ) .
Now , we look at ( 2.11) . Suppose both a and b are non- zero and a x b = 0.
Then we have the following cases:
•a , b > 0. By Proposition 1.12, (ii) , a x b > 0, a contradiction.

•a > 0 and b < 0. Then — b > 0 and by Proposition 1.12, together with
( 2.10 ) , one has
0 < a x ( b) = — — (a x b ) .
This implies a x 5 / 0, a contradiction.
Other two cases ( i.e., a < 0, b > 0; and a < 0 , b < 0 ) can be treated
similarly. Thus, the necessity of ( 2.11) holds. The sufficiency is obvious.
We now prove ( 2.12 ). If a x c = b x c and c / 0, then we claim that
—
a b = 0. If not , then
^
0 7 (a — b ) x c = ( a + ( — 6) ) x c = a x c + ( — b ) x c
= a x c - ( b x c ).
Hence, by ( 2.9) , a x c 6 x c, a contradiction. Consequently, a — b = 0.
^
Thus, the necessity of ( 2.12 ) holds. The sufficiency is obvious.
Note that for any a , b E N,
o. — b — CL -f ( — 6) = { CL 0 0 ) K ( 0 0 b) CL 0 b.
" “
—
Further , for any a , b G N with a 6, we have
^
a = 6 + c,
for some c G N, and
—
a b = ( b + c ) Qb = cQ 0 = c.
Because of such a compatibility, from now on , we replace © by — , for
notational simplicity.
Now , we extend the orders “ > ” and from N to Z.
Definition 2.9. Let a , b G Z. We say that a > b (or b < a ) if a — b G N,
—
and a b (or b a ) if a b G N.
^
The following proposition is concerned with the order on Z , whose proof
is left to the readers.
Proposition 2.10 . Let a , b , c G Z.
(i) ( Addition preserves order ) a > b if and only if a + c > b + c .
(ii ) ( Positive multiplication preserves order ) Let c > 0. Then a > b if
and only if a x c > b x c.
( iii) ( Negation reverses order ) a > b if and only if — a < — b.
( iv ) ( Order is transitive) If a > b and b > c , then a > c .
( v ) ( Order trichotomy) Exactly one of a > b , a < b , and a = b holds.
To conclude this subsection , let us point out one fact: The principle of
induction does not apply to the integers. Namely, if P ( n) is a statement
pertaining to n G Z , if P( 0 ) holds and whenever P ( n ) holds, P ( n+ 1 ) holds.
This does not imply that P ( n ) holds for all n G Z. Here is a simple example:
Let P ( n ) be the following statement: “n is non- negative” . Clearly, P ( 0 )
is true and when P ( n ) holds, P { n 4- 1) holds. But , “ P ( n ) is true for all
n G Z” is not true.
2.2 Rational numbers
—
Note that for any a , 6 G Z, both a 4 b and a b are in Z. Such a property
is described by saying that Z is closed under addition and subtraction . On
the other hand , for any a , 6 G Z, a x 6 G Z, and integer 1 has an interesting
property that
=a x l =a
lxa Va G Z.
We call 1 the unit of multiplication in Z. Now , for any positive integers a , b
with a > 6, sometimes one can find an integer c such that a = bxc. In this
case, we say that a is a multiple of b , b is a divisor of a , and a is divisible
by 6; for example a = 6, b = 2, then c = 3. But sometimes one could not
do that , for example, a = 5, b = 2, i.e., there exists no c G Z such that
5 = 2 x c. Therefore, we would like to further extend the set Z. To this
end , we introduce a relation on the set Z x ( Z \ {0}) as follows:
( a, b) ~ (c, d ) if a x d = c x b. ( 2.13)
We claim that ~ is an equivalent relation, i.e., the following hold:
Reflexivity: ( a , b ) ~ (a , 6) , since a x b = a x b.
Symmetry: ( a , b ) ~ ( c, d ) implies ( c, d ) ~ ( a , b ) since
a x d = c x b => cx b — a x d .
Transitivity: ( a , b ) ~ ( c , d ) , ( c , d ) ~ ( e, / ) implies ( a , 6 ) ~ ( e , / ) , since

a x d — c x b cx f = ex d ,
imply
axdxcxf = c x b x e x d.
0, by cancellation property, the above implies
Since d
^ axcxf = c x b x e.
This leads to the following: either c = 0, or a x / = e x b. In the case, c = 0,

we must have a = e = 0 since d 0. Hence, in either case, we always have
^
a x f = e x b,
which gives ( a , b ) ~ ( e, / ) .
Now , for any ( a , b ) G Z x (Z \ {0}) , we define corresponding equivalence
class
a / b = { ( c , d ) G Z x ( Z \ { 0} ) | ( c , d ) -( a , 6) }
and let
|
Q = a / 6 | (a , 6 ) G Z X ( Z \ {0}) .}
Any element in Q is called a rational number. Next , we extend the addition ,
multiplication and negation from Z to Q.
Definition 2.11 . For any a / b , c / d G Q , we define the addition , multiplica-

tion , and negation as follows:
( a / b ) -f ( c / d ) = ( a x d -\- b x c ) / ( b x d )
( a / b ) x ( c/ d ) = ( a x c ) / ( b x d ) ,
- ( a / b ) = ( -a / b ) .
Proposition 2.12. T h e above defined operations are well-defined .
Proof . We show the conclusion for the addition. Let a / b , a! / b', c / d G Q

with a / b = a! / b' . Then
a x b' = a' x b.
It suffices to show that
( a x d + c x 6) / ( 6 x d ) = ( a / b ) + ( c / d )
= ( a' / b' ) + ( c/ d ) = ( a' x d -\- c x b' ) / ( b' x d ) .
Or equivalently,
( a x d + c x b ) x ( b' x d ) = ( a' x d 4- c x b' ) x ( b x d ) .
This is the same as ( making use of Proposition 2.7)
a x b' x d x d + c x b x b' x d = a' x b x d x d + c x b x b' x d ,
a x b' x d x d = a' x b x d x d .
Since d 7^ 0, the above is equivalent to
a x b' = a' x b,
which is true.
The proofs of the other two are similar.
The following result collects some interesting facts that reveals the re-
lation between Z and Q.
Proposition 2.13. For any a , b G Z,
( a /1) 4- ( b / 1) — (a 4- b ) / l
( a /1) x ( b / 1 ) = (a x b) /1
— ( « /!) = (-a ) / l .
0,
and in the case b
^ ( b / b ) = (1/1) , ( 0 / 6) = ( 0 /1)
(a / 6) = (1/1) a b. —
(a / 6) = ( 0 /1) «= a=0
( a / 6) + ( 0/1) = a / 6,
( a / 6) x (0/1) = (0/1) .
Proof . It is clear that
( a /1) + { b / 1 ) = (a x l + 6 x l ) / ( l x l ) = (a 4 6) / l .
The others are similar .
Because of the above, from now on , we identify
a /1 = a , Va G Z. ( 2.14)
In particular ,
0/1 = o, 1/1 = 1. ( 2.15)
Hence, from ( 2.14) , Z becomes a subset of Q:
ZcQ. ( 2.16 )
For any a , & G Z \ {0}, a / b / 0. In this case, we define the reciprocal of

a / b by
{ a / b) 1
= b / a.
One can check that this operation is also well-defined . Note that 0 does
not have its reciprocal.
Proposition 2.14. ( Algebraic Laws for Rationals ) Let £, 77 , £ G Q.
Then
£ + T7 = ?7 + £,
( £ + r] ) + C = £ + ( V + 0 »
£ + 0 = 0 + £ = £,
£ + (-£) = (-£) + £ = 0,
£ x 77 = 77 x
( £ X ??) x C = £ x ( 7? X 0 ,
f,
£ xl = lx
^
£ X (7? + C) = £ X 77 + X C.
If £ 7^ 0, then £ 1
^
is defined and non- zero as well. Moreover ,
£ x r‘ = r ' x £ = 1.
Proof . We prove the second equality. For £, 77 , £ G Q, let
£ = a / ^ , V = c / d , C = e / /, a , c e G Z , 6, d , / G Z \ {0}.
Then
( £ + h ) + C = ( ( a / 6 ) + ( c/ d ) ) + ( e / / )
= ( (a x d + i) X c) / ( 6 x d ) ) + (e / / )
= ( (a x d 4- 6 x c) x / -f 6 x d x e) / (6 x d x / ) ,
and
£ + ( h + C ) = (a / 6 ) + ( ( c/ d ) + ( e / / ) )
= ( a / 6) + ( (c x / + d x e ) / (d x / ) )
= (a x d x / + 6 x ( c x / + d x e ) ) / ( 6 x d x / ) .
Hence,
( + ^7 ) + C = C + ( ^7 + 0 -
The others can be proved similarly.
For notational simplicity, from now on , for any £, 77 G Q , we write
£ X T] = £ 77 = £7? ,
unless the symbol “ x ” or needs to be emphasized .
Now , for any £ , rj G Q with
^Z/v , tv x -
rj 0 we define the quotient £ / 77 as follows:
= ( 2.17)
The above is also denoted by 77 or £— , The operation -f is called the
division. Clearly, if £ = £ Vi then *
C
£ = C + n = C /v = -v rj = c -s- e = c/e = C
The following extends the order from Z to Q.
.
Definition 2.15 ( i) £ G Q is positive, denoted by £ > 0, if there are
a , 6 G N such that £ = a / 6; £ G Q is negative, denoted by £ < 0 , if £ is —
positive; £ G Q is non- negative, denoted by £ 0, if £ > 0 or £ = 0; £ G Q
>
—
is non- positive, denoted by £ 0 , if £ is non-negative.
^
£ > 77 if £ — 77 > 0, and £
( ii ) Let £ , rj G Q. We say that
^ 77 if £ — 77 ^ 0.
We have the following result which is comparable with that for integers
(see Proposition 2.10 ).
.
Proposition 2.16 Let £, T/ , £ G Q. Then
( i) ( Addition preserves order ) £ > 77 if and only if £ -1- £ > 77 4- £.
(ii ) ( Positive multiplication preserves order ) Let £ > 0. Then £ > 77 if
and only if £ • £ > rj • £.
( iii) ( Negation reverses order ) £ > 77 if and only if — £ < -77.
( iv ) ( Order is transitive) If £ > 77 and 77 > £ , then £ > £.
( v ) (Trichotomy of rationals ) Exactly one of the three statements holds:
£ < 77, £ = 77, and £ > 77.
_1
(vi) If £ > 77 > 0, then 0 < £ < rj ~ l .
Proof . We only prove ( vi) , leaving the others to the readers. By £>
77 > 0, we may assume that
£ = a/ b, 77 = c/ d,
for some a , 6, c, d G N, and
0<£ — 77 = a / 6 — c/d = a / 6 + ( — c) /d
= [ad -f ( — c) 6] /6d = (ad — 6c) /6d.
Thus ad — bc > 0. Now,_we look at

— £ = d / c — b / a = ( ad — bc ) / ac > 0.
rj
~l 1
It is obvious that £-1 > 0 . This proves (vi) .

Next , we introduce the following notion.
Definition 2.17 (i) For any . £ E Q, define its absolute value |£| by the
following:
if 0,
if C < 0.
^
( ii) For any £, rj E Q, define the distance between £ and rj by the follow-
ing:
d(£ , v ) =
The following result is concerned with the absolute value and distance.
.
Proposition 2.18 Let £, 77 , £ E Q. Then
( i) |£| 0 and |£| = 0 if and only if £ = 0.
( ii) |f + r?| |f | + |T?|.
(iii) For 77 0, — 77 £ 77 if and only if |£| 77. In particular ,
-|£l < £ < l£ l -
( iv ) |f \ = |f | \ r)\ . In particular, \ - f | = |f|.
• T} •
( v ) d (£ , 77 ) 0 and d (£, 77 ) = 0 if and only if £ = 77.

( vi ) d (f , 7? ) = dfof ).
( vii) d (f , f ) d (f , r? ) + d ( »?, 0 -
Next , we introduce the following notion called the exponentiation.
.
Definition 2.19 ( i ) For any a E N, define
0a = 0.
( ii) Let £ G Q \ {0}. Define
*° = i , e-
n f1
= £" £, • V n e N,
and define
rB = i /r Vn N.
We have the following laws for the exponentiation.

Proposition 2.20 . Let £ , rj E Q\ {0}.
( i ) For any a , 6 E Z, the following hold :
£ £6
0
*
£ 0 +6 ( n b = eb , (Z - v) a
= C - va loi = IO -
(ii) For a E Z \ {0}, £a = r]a if and only if £ = rj .
( iii) For n E N,
n
£ > r] > 0 C > r j n > 0, rfn > £
“
> 0.
Proof , ( i) For the first equality, we first show that £ m • £n = £m + n for all
£ 0 and ra , n E N. To this end , we let m E N be fixed and use induction
^
in n.
Forn = 0: £ m • £° = £m . 1 = £ m = £m + 0.
Suppose the conclusion holds for n. Then for n + 1, by definition and
induction hypothesis, we have
n+ l
= r -r -e = e 771 + 71 ra + n +1
This means that £a • £6 = £a + b holds for a , b E N.

Now , if a = — n, 6 = — m with n, m E N , then
e - ? = cn - z — _= ( Vo _( l /r ) = i / (o r )
m • •
n+m
£— (n + m )
£ 0 +6
= ra and 6 = — n, then we may let , say, m > n. Consequently,

Next , if a
£ £ 6 £ . £ — £ n £ ly/ n — £ ~ n — £ <*+6
0
>
771 71 771
— <
7i #
^
77l
The case ra n can be treated similarly. Hence, the first equality holds.
For the second equality, first let a E Z and b = n E N (still let £ 0 ).
Then we use induction on n. The case n = 0 is trivial. Suppose the equality ^
holds for n. Then for n + 1, we have, making use of the first equality,
(£0 ) 71+1
_ (£ ) 0 71 e
(£0 ) —£ an . £a _£ +0
071
_£ ( O TI +1)
Thi§ completes the induction. In the case that b = — n with n E N , we have

( o b = (e°rn = i / (on = i /e = r £ an an 06
For the third equality, we first let a = n 6 N. Use induction on n. The

—
case n 0 is trivial. Suppose the statement is true for n. Then for n 4- 1,
we have ( using the commutativity of multiplication )
( *
*?)
71 +1
= ( t - v )n - ( t; - v ) = c - vn - t; - v = z 71 +1 v
^
71 +1
This proves the third equality for the case that a = n G N. Next , let a = n —
with n G N. Then by what we have proved
K v ) ~n = i / (
• • »? ) n = i / (r •
vn ) = i /r w = rn rn -
• ?
Hence, the third equality holds.

For the fourth equality, if £ 0 , it is trivial. Let now £ < 0. Then
— —
£ = 1 £| = ( 1) • |£|. Hence, by the proved third equality,^
e = [(-i ) + i ]° = (-i )° + r .
Consequently,
ie° i = | ( - i ) - iei ° | = ier

0
The proofs of the rest of the conclusions are left to the readers.
Next , we look at an interesting issue of rational numbers.
Proposition 2.21. (i) (Interspersing integers by rationals ) For any £ G Q ,
there exists a unique integer , denoted by [£] such that [£] £ < [£] + 1.
(ii) (Interspersing of rationals by rationals) If £ , r) G Q with £ < rj . Then
there exists a £ G Q such that £ < £ < rj .
Proof , ( i ) Let £ = n / q with n , q G N , q > 0. Then by Euclidean

algorithm , there exist r, m G N with 0 r < q such that
n = mq + r.
Hence
n mq .
1
Multiplying q > 0, one gets
1
£ = n / q = ( mq 4 r ) q *
= m -f rq 1 m.
Also, since r < q , there exists a k G N , k > 0 , such that
q = r 4- k .
Hence
n -f k = mq -f r + k = ( m + l ) p.
This gives
m + 1 = ( n + fc ) / p = n / p + /t / (z > n / <7 =£ m.
Now , if £ < 0, then — £ > 0. Hence, from the above, we have some m G N
such that
m —£ < m + 1.
Consequently,
— ( m + 1) < £ — m.
Hence , we can always find a p G Z such that
< P + 1.
Now , suppose there exists another ( different ) p G Z such that
P < £ < P + 1.
Without loss of generality, let p < p. Then p + 1 p. Consequently,
£ < p + l < p < £,

a contradiction. Hence, p is unique, and we denote [£] = p.
(ii) Note that 1/ 2 G Q and it is positive. Hence,
£ / 2 < p / 2.
Consequently,
e = e / 2 + e / 2 < e / 2 + p/ 2 < p/ 2 + p / 2 = p .
Clearly, for £ = e / 2 + p / 2 G Q, we have £ < £ < p.
For any (a , 6) G Z x ( Z \ {0}) , we may always find a representative

( a', 6') G a / 6 such that 6' G N. Therefore, hereafter , without loss of gener-
ality, we may let
Q = { m / n | m = 0, ±1, ±2, • • • , n = 1, 2 , 3, - - • }. ( 2.18)
Let L be a horizonal straight line on which we define an order as follows:

For any x , y G L,
x<y if and only if x is on the left of y .
Fix a point , label it 0 and call it the origin. Pick a unit length. For each
n G N , find a point on the line L , label it n , which is on the right of the
origin with the distance of n units from the origin. Also, find a point , label
—
it n , which is on the left of the origin with the distance of n units from
the origin. Then Z is labeled on L . For any rational number of form m / n
with m G N , n G N , by Euclidean algorithm , we have
m = nq + r, 0 r<n q e N.
We mark m / n on the line L as follows: equally divide the line segment of
L between the points q and q -P 1 into n parts with the partition points:
q = Co < £L < £2 < * * * < £n = q + 1-
Then mark m / n at £r . For negative rational numbers , we can do the similar
thing. As a result , we see that
N c Z c Q C L, ( 2.19 )
We call the points on L corresponding to those in N , Z , and Q the natu-
ral number, the integer, and the rational number points, respectively. By
Proposition 2.21 , Q is dense in L , in the sense that between every two
rational number points , there exists a rational number point . A natural
question is: Do rational numbers occupy the whole line L? Equivalently, is
the following true:
Q = L?
The following result answers the question negatively.
.
Proposition 2.22 There does not exist a rational number £ e Q for which
£2 = 2. This means that the side length of a square with area 2 is not a
rational number .
Proof . Suppose there exists a rational number £ = p / q with p, q G N ,
q 7 0, and they are co - prime , i.e. , the greatest common divisor of p and q
^
is 1, such that
2 = £2 = P7?2 -
This implies
2 q 2 = p2 .
Since p and q are co-prime, the above tells us that p is divisible by 2. Hence,
we may write p = 2p. Then the above leads to
q2 = 2P2 .
Similar to the above argument , we must have that q is divisible by 2. Hence,
both p and q are divisible by 2 , contradicting our assumption.
The above result shows that L \Q 0. We call any x L \ Q an
irrational point. ^ G
3 Real Numbers
We have seen that although the set Q of rational numbers is closed under
addition , subtraction , multiplication , and division , there is still a big unsat-
isfactory point: Q L In another word , there are ( a lot of ) points on the
^
line L that are not rational points. Therefore, we want to further extend
the set Q of rational numbers.
3.1 Sequences of rational numbers
Definition 3.1 . A set of ordered rational numbers £i , £2 , £3 , • • • is called a

sequence of rational numbers, denoted by {£n}n i - The set of all sequences
of rational numbers is denoted by ( Q). ^
^°
• For any sequence {£ } ^ -^° (Q) , it is allowed to have £
n n i £ n = for
some n / m.
Sequence could be a finite one. For convenience hereafter, if £1, £2 ? • • ,
• *
is a finite sequence of rational numbers, we define £n = £ , , for all n > k .

*
Therefore, we will only consider infinite sequences.
^°
Definition 3.2. ( i) A sequence {£n}n i £ ( Q) is said to be bounded if
there exists a rational number M > 0 such that ^
|£ | < M, Vn 0.
n
^
The set of all bounded sequences of rational numbers is denoted by
^°° (Q).
(ii ) A sequence {£n}n i ° (Q) is said to be Cauchy if for any rational
^ ^
number e > 0, there exists a n i V e N such that
V j, k > N.
The set of all Cauchy sequences of rational numbers is denoted by c(Q ) .
( iii) Two sequences { £n }n 1, {Vn }n 1 ° ( Q ) are said to be equivalent,
denoted by {£n}n > 1 ~ ^ ^ ^
{ Vn }n i , if for any rational number e > 0, there
exists an N G N such that ^
| Vn| Vn N.
^£ >
The following result is basic for the sequences of rational numbers, whose
proof is left to the readers.
Proposition 3.3. ( i ) Every Cauchy sequence is hounded .
( ii) { £n }nzu {nn }n> i G f ° ( Q) are equivalent , then {£nW 1 G C( Q)
If
and only if { r]n }n i G c ( Q) .
^
( iii) Let 1, {r?n }n i G c ( Q ) . Then
^
{£n b ^7n }n l {£n ’ ?7n}n l G c( Q) .
^ ^
"
?
(iv ) {£n}n i G c( Q ) is bounded away from zero , i.e., there exists a

If
rational number ^5 > 0 such that
Vn l
then { nl}n l G c( Q ) .
^
^ ^
(v ) If feWi , fcWi , feWi » fcWi G f °° ( Q ) such that
Then
{£n}n l
^ ~ {f ?n}n l ~ ^ *
{£n H ?7n }n l ~ {£n b V n }n l

^ ^
" “
{£n ’ Rn }n l ~ {£n ' n }n l

^
and in addition , {£n}n i is bounded away from zero, then {£n}n l must
if ^ ^ )
also be bounded away from zero and ^ ^

{£n ^ }w l
( vi) The relation defined on c(Q) is an equivalent relation , i .e., it ^ *
possesses the reflexivity, symmetry and transitivity.
3.2 A construction of real numbers

We now introduce the following.
Definition 3.4. ( i ) For any {£n}n G c(Q ) , define the corresponding
equivalent class by î
Let
lhn £n —
^
R=
{??n }n
Any element in R is called a real number.

^
{ lim £
l G
n
c ( Q) |\j]n }n
| {£„}> i
^ 1 ~ {£n }n ^ l
j
c( <Q> ) .
^ *
(ii ) For any r £ Q, define £n = r for all n 1. Then {£n}n G c (Q )

which defines a real number. Thus , in such a sense, ^ i
Q C R. (3- 1)
Any element in R \ Q is called an irrational number.
It is clearly that
{£n }n
^ ~ l 1 lim £n = lim rjn .
Hence, we may identify lim £n with any { r)n}n i which is equivalent to
^
{£n}n i - In particular , lim £n is identified with the sequence {£n }n i We
^
have the following basic result of reflexivity, symmetry, and transitivity for ^-
the real numbers with respect to the equality.
Proposition 3.5. Let {fn }n

^ , { Vn }n^ u {CnWi
i c(Q ) and denote
x = limfn , y = lim r]n , z = limCn -

Then
x —x \
x= y y = x;
x = y, y = z x = z.
Proof . Let us prove the transitivity. Let
x = lim £n , y = lim r)n , z = lim £n .

By x — y and y = z , we know that
{£n}n
^ l ^ {^?n }n
^ l5 1 ~ {Cn} ^ n l*
Thus, {Cn }n
^ ~ {Cn}n ^ , which means x = z.
i i
The above actually shows that the real numbers are well-defined. We
now introduce algebraic operations on R.
.
Definition 3.6 For any x , y G R with
x = lim £n , y = limr7n ,
for some {Cn }n i 5 { n}n £ c( Q) , define addition, multiplication and
negation as follows: ^ ^ ^ i
x+y = lim (fn + r/n ) , x • y = lim (fn • 77^ ) , -x = lim (-£n ) . ( 3.2 )
In addition , if {£n}n i is bounded away from zero, we say that x is non-zero,
^
denoted by x 0, and define the reciprocal of x by
^ (3.3)
According to the properties of the Cauchy sequence, the addition , mul-

tiplication , negation , and reciprocal of real numbers are all well-defined.
Then we define subtraction and division by
x - y = x + ( -y ) x / y = xy - l (V + 0) .
We have results for real numbers similar to Proposition 2.14 for rational
numbers. Since the statement is almost the same, except that Q is now
replaced by R , we shorten the statement below.
Proposition 3.7. All the algebraic laws hold for real numbers.
.
Definition 3.8 ( i) A sequence {£n}n i £ ° (Q) is said to be positively
^
^
bounded away from zero if there exists a positive rational number 5 > 0
and an N e N such that
Vn N.
( ii) A real number x is positive, denoted by x > 0 if there exists a

sequence {£n}n i c( Q) , positively bounded away from zero such that
^
x = lim £n .
If x > 0 or x = 0, we say that x is non- negative, denoted by x 0. If
— x > 0 , we say x is negative, denoted by x < 0 , and if x 0, we say that —
x is non- positive, denoted by x 0.
( iii) For any x E R , define its absolute value \ x \ by the following:
x x 0,
x
—x
^
x<0 .
( iv ) For any x , y e R , define x > y (or y < x ) if x — y > 0, and x y

(or y x ) if x > y or x = y .
( v ) For any x , y £ R , define
x y V, x y,
max ( x , y ) =
X
y x < y,
min ( x, y ) = <
x
^
x< y .
It is interesting that the following hold:
x + y + \ x — y\
m a x( x , y ) =
2 Vx , y G R.
x + y - \ x - y|
min ( x , y ) =
2
Thus, both maps ( x , y ) »-> max ( x, y ) and ( x , y )

• min ( x, y ) can be written
in terms of the absolute value.
.
Lemma 3.9 Let {Cn}n
^ , {Vn }n^ i
1 c ( Q ) such that for some N G N ,
Cn Vn N.
Then
^ Tjn
lim Cn lim rjn .

In particular , if
then
Cn > 0, \/ n
^ N,
lim Cn 0.
Proof . We prove the second conclusion. The first can be obtained from
considering Cn - rjn .
Suppose x = lim Cn < 0. Then by definition , there exists a positive
rational number 5 and a sequence {Cn}n i such that
^
x = limCn , < S , Vn ^ 1.
Cn
i , for £ = 5 / 2 , there exists an N
Since {Cn}n
^ ~ {Cn}n ^
i N such that
Cn | < £ = 5 / 2 ,
This implies
|Cn ~ \/n
^ N.
Cn < Cn + £
a contradiction.
^ — 5 + 5 / 2 — — 5 / 2 < 0,
( Vn >N
Note that {Cn}n i c( Q) with Cn > 0 , for all n 1 is not enough to

^
guarantee that x = lim £n is positive. For example, for Cn n-1, n 1, ^ —
one has limCn 0. — ^
The following is comparable with Proposition 2.16 ( which is for rational
numbers ).
Proposition 3.10 . Let x , y , z e R . Then

( i) ( Addition preserves order ) x > y if and only if x + z > y + z .
( ii) ( Positive multiplication preserves order ) Let z > 0. Then x > y if
and only if x • z > y • z .
(iii) ( Negation reverses order ) x > y if and only if — x < — y.

( iv ) ( Order is transitive) If x > y and y > z , then x > z .
( v ) ( Trichotomy of reals ) Exactly one of the three statements holds:
x < y , x = y , and x > y .
( vi) If x > 0 , then x ~ l > 0. If x > y > 0 , then x ~ l < y ~ l
-
Proof . We only prove ( vi ) . The proof for the other parts is left to the
readers. Since x > 0 , there exists a sequence {£n}n i c( Q ) such that
x = lim £n and for some rational numbers M, 5 > 0 , ^
<5, Vn 1.
1
Then we know that {£ “
}n c( Q) and
^ i
n 1.
This implies that
1 1
> 0.
x
Now, let x > y > 0. Then we have sequences {£n}n
= lim
^ { yn }n c( Q)
such that for some rational numbers M, 5 > 0 , one has î >
î
^ ^
Hence, for some N N,
niRn
^ G
Rn
^^ 5 Vn
^ 1.
> Vn + S / 2 6, Vn TV.
Consequently,
1
£n 1 < (% + V2) 1
< 1
Vn iV.
Hence
M " 1
( y + 5/ 2 )-1 < y-1.
This proves ( vi ) .
We now extend the exponentiation from Q to R.
Definition 3.11. For any x G R \ {0}, define

x° =l xn+1 = xn • x Vn
^0
and
x
‘ n
= ( xn ) ~l
= l / xn .
The above defines xa for any x G R \ {0} and any a G Z. The following
lemma is useful below.
Lemma 3.12. Let x = lim £n G R \ {0} with {£n}n > l c(Q) , which is
away from zero. Then for any G G Z, {£n}n i c(Q) and
^
x° = lim £“ .
Proof . First , we let a = m G N and use induction on m. For m = 0 , it
is trivial. Suppose the conclusion holds for m. Then for m + 1 we
m +1
= X m • x = lim (C ) • lim (£n ) = lim (C • Cn ) = Hm (C+1)
£
—
Now , for a = m with me Z, we have
= (a; )-1 = lim (C )-1 = lim (£

x° = ~m
) = lim £.
We now present the following result which is comparable with Proposi-
tion 2.20.
Proposition 3.13. Let x , y G R \ {0} and a , 6 G Z . Then the following

hold :
(i ) xa xb = xa+b , { xa ) b = xab , ( x y )a = x° ya , \ xa \ = |x|a .
• • •
( ii) If a ^ 0, then xa = ya if and only if x = y .

7
( iii ) For n G N,
x>y>0 xn > yn > 0, y ~ n > x ~ n > 0.
Proof . Let us prove the first equality in ( i) to get the flavor. Other
proofs are left to the readers. Let {£n}n i £ c( Q) such that
^
x = lim £n .
Then by Proposition 2.20 , we know that
Sn > n
_
pa ’ Spb _ pa -\- b
#
Sn
Thus , by Lemma 3.12,
xa - x b = lim (£ )
• lim (
^ ) = lim(£ £ ) = lim(£“+ b) = x“+ b
3.3 Further properties of real numbers

We now look at some further properties of the set R. Recall that there is
no rational number £ such that £ 2 = 2. Therefore , we would like to see if
there exists a real number x such that x 2 = 2. Before going further , we
need to make some preparations.
Proposition 3.14. ( i) ( Bounding of reals by integers ) For anyx G R , there

exists a unique integer , denoted by [x] such that
[x] X < [ x ] + 1. ( 3.4)
( ii) ( Archimedean property) For any real numbers x , e > 0, there exists
a natural number N such that Ne > x .
( iii) ( Density of rational numbers ) For any real numbers x < y , there
exists a rational number q such that x < q < y .
Proof , ( i) Let x > 0. There exists a sequence {£n }n c(Q) such that
^ i
= limfn .
x
Since Cauchy sequence is bounded , we can find rational numbers q , r and
a natural number N such that
0<q Vn N.
^ ^
Then , by Proposition 2.21, we have
[q] <Q r < [r ] + 1.
Hence
[q] < x < [r ] + 1.
Now , comparing [ q ] + 1 with x, there are two possibilities:
[q ] + 1 > or [ q\ + 1 x.
,
If the former holds, then we take [ x ] = [ q\ if the latter holds, we further
compare [ q ] + 2 with x . Since [r ] G N , finitely many comparisons will give
us an integer p such that
p x < p -f 1.
Clearly, such a p is unique. Hence, we call p = [x].
The case x < 0 can be treated similarly.
( ii) Since x / e > 0 is a given a real number , there exists a natural number
N such that
x / e < N.
Then
x < Ne .
( iii) Since y - x > 0, there exist sequences {£n}n { r]n }n £ c (Q )
such that ^ i?
î
x = lim £n , y = lim rjn ,
and for some positive rational number 5 > 0 ,
1.
Then
Tin
^5 < ? Vn
TJn .
Also, since both {£n}n
ând 4{ r5/}4 <
"
}n n i
T )n
^ 5/ 4
are Cauchy, we may assume that
^ i
^ N.
\ n ~
N \ AVU ~
VN \ < <5 / 4, Vn
Hence
< + S/4 - 6 + <5 / 4 < r)n + 5/ 4 - 5 + 5/ 4 = rjn - (5 / 2 < r/n .
Therefore,
N
^ rjN
x = lim £n £ N + <5 / 4 r] N - 35/ 4 limryn = p.

Consequently, we may take
35 5
1/
This proves ( iii) .

«=
^
2
"
T + )) £ VN -
2
£N
4’
.
Definition 3.15 ( i) Let E Cl . An M G R is called an ?/pper hound for
£7 if
x Af , V X G E.
When such an M exists, £7 is said to be bounded from above. If no such an
MG R exists, E is said to be unbounded from above.
R. A real number M is called a least upper bound of E if
( ii) Let E C
M is an upper bound of E and for any upper bound M' of E, it holds:
M M'.
Example 3.16. (i) Let R + = {x G R | x 0}. Then R + does not have an

upper bound. Thus, R + is unbounded from above. ^
R | 0 x 1}. Then [0, 1] has an upper bound ,
( ii) Let [0, 1] = { x G
and 1 is a least upper bound of [0 , 1] .
( iii ) Any M G R is an upper bound of empty set 0.
The following result is very important .
Theorem 3.17. Let E C R.

( i) E can have at most one least upper hound .
(ii) If E has an upper bound , then it has a least upper bound .
Proof , ( i) Let M\ and M2 be two least upper bounds. Since they are
upper bounds of E , we must have
M2 ,
Hence, Mi
Mi M2
^ Mi .
= M2 .
( ii) Without loss of generality, we may assume that there exists an
#o E such that xo > 0. In fact , take any xo £ E and define
EQ = {x + |x0| + 1 | x G E }.
If we can find a least upper bound so of Eo , then E has a least upper bound
s given by the following:
s= so - |x01 - 1.
We now fix an xo G E with xo > 0. Let M xo > 0 be an upper bound
^
^
of E . For any n G N such that n > n = [ ] + 1, one has
—xo1 < Lxo— -I + 1 < n,

which means that xo > 1/ n. Thus , 1/ n is not an upper bound of E . Now ,
by Proposition 3.14 , (ii) ( the Archimedean property ) , there exists a K G N,
such that
K/n M. ( 3.5)
We claim that there exists an mn G N such that
0 < mn K ,
246 — A Concise Introduction
Mathematical Analysis
with mn / n being an upper bound of E and ( mn — 1 ) / n not being upper

bound of E . To show this, we begin with K . By ( 3.5) , we know that K / n
is an upper bound of E . If ( K — l ) / n is not an upper bound of E , then
we let mn = K . Suppose ( K — 1) / n is still an upper bound of E . Then
we check if ( K — 2 ) / n is an upper bound of E . If not , we let mn — K — 1.
Otherwise, we continue the procedure to check if ( K 3) / n is an upper
bound of E , and so on. Since 1/ n is not an upper bound of E , the above
—
procedure will stop at some mn with 0 < mn K .
Now , for the sequence {ran}n for any n, n' N ( with N n large)
we have ^n >
mn > ( mn' - 1) 5
mn > mn - 1
>
n n ' n ' n
since the left-hand sides are upper bounds of E and the right- hand sides
are not . Thus
mn mn > > 1 1
•
n n' n' N>
and
mn mn> < -
1 1
n n' n N'
Hence, { mn / n }n is Cauchy. Let
^n
s = rlim = vlim
mn - 1
n n
We claim that s is the least upper bound of E . In fact , since for any x G E,
mn
x
^ —n Vn n.
This leads to
.rlim mn
x
^ n = s.
On the other hand , since [s not an upper bound of E , for any upper 1
bound M' of E , we have

mn - - < M'
n
Vn
^ n.
Hence
s = lim
mn - - < M\
n
This shows that s is the least upper bound of E .
Because of the above result , we hereafter denote the least upper bound
of E by sup(.E ) , call it the supremum of E . If E has no upper bound , we
let
sup ( -E ) = 4- oo ,
and in the case E = 0 , we define
—
sup (.E ) = oo ,
since any M e R is an upper bound of 0.
Now , we are ready to prove the following proposition.
Proposition 3.18 . There exists a positive x e R such that x 2 = 2.
Proof . Let
E = { y e R | y 0 , y 2 < 2}.
Since 0 e E , E 0 . Also, if y > 2 , one has
^ y 2 > 4 > 2.
Hence, 2 is an upper bound of E . Now , let
x = sup ( E ) e M.
We claim that x 2 = 2. In fact , suppose x2 < 2. Let 0 < e < 1 be small.
Then ( note that 2 is an upper bound of E , x 2 )
( x 4- e ) 2 = x 2 + 2 sx 4- s2 x 2 4- 4e 4- e x2 4- 5e.
Thus, by choosing e > 0 small, we have
( x 4- e ) 2 < 2.
Hence, x 4- £ E , contradicting the definition of x .
Now , suppose x 2 > 2. Then again let 0 < e < 1 be small. We have
— — —
( x s ) 2 x2 2ex 4- e 2 x2 — 2e:r x 2 4e , —
since x 2. Hence, by choosing e > 0 small, we will have
^ —
( x e )2 > 2 .
—
—
But then , x e is an upper bound of E (since if there exists any e E such
—
that x £ y , then ( x e ) 2 < y 2 < 2 , a contradiction ) . This contradicts
the definition of x again. Hence, x 2 = 2 .
The following result is useful.
Theorem 3.19. Let E C R have an upper bound , and be R . Then
b = sup ( -E ) if and only if
x b, Vx e E ( 3.6)
and for any e > 0, there exists an x e E such that
x > b — £. ( 3.7)
Proof . Sufficiency. Let M = sup( E ) . By our assumption, b is an upper
bound of E , we must have
M
b
Next , for any £
^
> 0, since there exists an x
.
G E such that ( 3.7 ) holds,

—
which means that b e is not an upper bound of E , whereas M is an upper
bound of E . Hence, one must have
b—£ <M b.
Since e > 0 is arbitrary, we obtain b = M .
Necessity. If b = sup ( E) , then it is an upper bound of E . Therefore,
( 3.6 ) holds. Also, since b is the least upper bound which means that for
—
any e > 0, b e is not an upper bound of E . This implies that there exists
an x G E such that ( 3.7) holds.
Symmetrically, we may define lower bound of a set E C R and the
greatest lower bound for E . We have the following result.
Theorem 3.20. Let E CM.
( i) E can have at most one greatest lower bound .
( ii ) If E has a lower bound , then it has a greatest lower bound .
The proof is very similar to that of Theorem 3.17. One may also do the
following. Let
— E — { — x G | x G E }.
—
Then E has an upper bound if E has a lower bound . Therefore, there
—
exists a unique least upper bound sup ( E ) . Then we let
K = — sup ( — E ) .
One can check that K is the unique greatest lower bound of E . Hereafter ,
we denote the greatest lower bound of E by inf ( E) , and called the infimum
of E . Similar to Theorem 3.19, we also have the following result .
Theorem 3.21. Let E C R have a lower bound , and a E R. Then a =

inf ( E ) if and only if
x a VxeE
and for any £ > 0, there exists an x G E such that
x <a+£ •
Proposition 3.18, together with the results concerning the supremum

leads to some more general result . We now establish such kind of result .
.
Lemma 3.22 Let x E R , x > 0 and let n E N, n 2. Then the set
^
En { x ) = {y G K y 0, yn x } I
is non-empty and hounded above .
Proof . Since 0 E En ( x ) , we have the non-emptiness of En ( x ) . Further ,
we claim that E has an upper bound. We look at two cases.
Case 1. x 1. Then we must have
I/ 1 < Vy E Fn ( x ) .
In fact , if the above is not the case, i.e., there exists a ye En ( x ) such that
y > 1, then x yn > 1, a contradiction .
^
Case 2. x > 1. Then we must have
y x V2/ E En ( x ) .
In fact , if the above is not the case, i .e., there exists a y E En ( x ) such that
y > x > 1.
Then yn 1 > 1. Thus
yn = yn l y > y > x ,
which contradicts y E En ( x ) .
Combining the above result with Theorem 3.17, we see that the following
definition makes sense.
Definition 3.23. Let x E R , x > 0, and n E N, n ) 2. Define
xn- = sup{ y E R | y 0, yn x},
j
which is called the n -th root of x. Also, we define

- Jn- , -ix - l 1
X (X n ) = X n
The following result is concerned with the n-th root operation .
Proposition 3.24. Let x , y E R , x, y > 0, and n , m E N, n , m 1. Then
the following hold :
(i) y = x if and only if yn — x . Thus, ( x i ) n = x .
( ii ) x > y i f and only if xi > y - In particular , xi > 0.
1 J ,J_ _ _ 1
= Xn .
. v
(iii) =x 71, (X n m
( iv ) For n m, if x > 1, then xi < x ; if x < 1, then xi > x

if ^
x = 1 , then xi = 1 for all k E N.
; and
Proof , ( i) Suppose y = i « . We show that yn = x. If this is not the
case, then either yn < x or yn > x . In the first case, for e G ( 0 , 1) small
enough , we have
n
n\ - l y n- k < X
( : y + e )n = yn + £
k
k =1
fc!( n — fc)! )
provided e > 0 small enough , where n! = n • ( n — 1) • • • 2 • 1 is called the

factorial of n. Hence,
zn
Consequently,
y + e G {O 0 | O 0,
^ x } = En ( x ) .
y+e sup (£„ ( x ) ) = x = y, »
a contradiction. Now , in the second case, for e G ( 0, 1) small enough ,
{ y - e )n = yn + £
n\ k - l y n- k _
Hence, y — e £ En ( x ) . This leads to
^
fe=i
( -l ) k
—
fc!( n fc )!
£ >x
y-£ sup ( En ( x ) ) = X =y
a contradiction again. Hence, yn = x .
Conversely, if yn = x, then En ( x ) which implies
yG
y sup ( En { x ) ) = X " .
On the other hand , for any z G En ( x ) ,
zn x= yn .
By Proposition 3.13, ( iii) , we must have z y . This means that y is an
upper bound of En ( x ) . Hence,
Hence, y
x
i
-. = sup ( En ( x ) ) y.
= sup .En (x) = x n

xi > y . Then by Proposition 3.13, ( iii) , we have
( ii) First , let
*
x = (a? )n > ( yi )n = y -
Conversely, let x > y . Thus ,
En{ y ) = { z K|z 0, y} C En { x ) .
This implies that x « is an upper bound of En ( y ) . Hence,

_
l L
_
yn .
^ Xn
We claim that the strict inequality holds in the above. If not , then yi = Xn
i
which implies that y = x , a contradiction.

The proofs of ( iii ) and ( iv ) are left to the readers.
With the above results, we now introduce the following definition.
Definition 3.25. Let x e R , x > 0 and let £ G Q with £ = a / b for some
a , b £ Z , b 0 . Define
^ x ^ = (x* ) ° .
The following shows that the above defined rational exponent of a real
number is well-defined .
Lemma 3.26 . Let a , a', 6, 6' £ Z, 6, 6' > 0 such that

a a'
b=V'
Then for any x £ R , x > 0,
(x * )a = (x £ )a' .
Proo/. Consider three cases.
Case 1. a = 0. Then a! — 0. Therefore,
*
(* ) » = X° = (** )“'.
Case 2. a > 0. Then a' > 0. From
ab' — a!b,
we have
y = xTU
1 /
— \
(X b7 ) a
i i
2/
1
= x T b = (x b ) 7 .
/ lv 1
Hence, by Proposition 3.24, (i) ,
Therefore
a
2/ = x
^ = xb .
= a/ar' =
^
(* )°'
—
y ao'
—
Case 3. a < 0. Then ( a ) / b = ( a' ) / b' . Since
= (2/°' )° =
^ )a.
—a, — a' > 0, we obtain

*
( x ) - ° = { x v ) ~ a' .
r
\
Taking reciprocal, we are done.

The following collects some results for the rational exponents of real
numbers.
Proposition 3.27. Let x , y G R with x , y > 0 and let £, rj E Q. Then

(i) x* • x11 = x*+T? , (x*) 77 = x^ , x £ = 1/x ^ .
“
( ii ) If £ > 0, then x > y if and only if x > y ^ ^ . In particular, x% > 0.

if
( iii) If x > 1 then x
and only if £ < 77 .
^ > xv if and only if £ > rj . If x < 1, then x& > xv
3.4 Real exponentiation
^
We have defined x for any x G R , x > 0 and any £ G Q. We now would
like to define xy for any x , y e R , x > 0. Let us begin with the following
lemma.
Lemma 3.28. Let x, y E R with x > 0 , and y = lim 77n with { rjn }n i e
c( Q) . Then {x^ jn i c( R ) . Furthermore , if { ( n }n>i then ~ ^
^ ^
{ x jn i c( R ) and ^
— ^ —> oo x .
= nlim
lim x n 77
n t oo
Proof . Let x > 1. Since {r/n}n is Cauchy, it is bounded by, say,

M > 0: î
|»7n | M, Vn > 1.
Also, since lim x
—-
n > oo
= 1, for any £ > 0, there exists ai^ ) l such that
~M
\ x* -\\ < ex
and there exists an N E N such that
Consequently, by assuming,
\V n
, we have
~
Vm \ <
^
: Vn , m
^ N.
^ rjn rjm
Hence, { xVn }
^ =x | X ^
^ ^
1 | xM | x — 1| <
Vm. X »M
is Cauchy, and it is convergent .

Vra , N. £
n i
^
Next , suppose « } > ~ {Vn } > - Let rn = - £ . Then rn -> 0. rjn
n n i n l n
We claim that
— xrn = 1.
lim
n > oo
( 3.8)
If this is the case, then by xVn = x ^ xrn , we have

lim xVn = lim x^ lim xTn — lim x^ .
n n
n —» oo n—+oc —> oo —> oo n n
Now , we prove ( 3.8) . To this end we find a sequence kn of natural numbers

such that
1
\ pn \ Vn 1.
^ kn i
Then ( noting x > 1)

1\ 1 \ \r n |rn |
X X
=x “
1 X
By Squeeze Theorem ( Corollary 2.11 of Chapter 1) , we have

lim xTn
—
n ¥ oo = 1.
Because of the above result , we now introduce the following real expo-
nentiation of real numbers.
Definition 3.29 . Let x , y G R with x > 0 . Let y = limp with rjn G Q,

n 1. We define:
^ x 27 x^ . ( 3- 9 )
= nlim
—
> oo
The following gives some properties of real exponents of positive real
numbers.
Theorem 3.30. Let x , y > 0 and p, q G R. Then

(i) xp eR , xp > 0 .
(ii ) xp • xq = xp+q , ( xp ) q — xpq .
( iii) x “ p
= l / xp .
( iv ) If p > 0 , then x >y if and only i f xp > yp .
(v ) x > 1, then xp > x9 if and only i f p > q ; i f x < 1, then xp > x9 if
If
and only i f p < q .
Proof . We leave the proof of (i)-( iii ) to the readers.

( iv ) Since p > 0, we have pn G Q , bounded away from 0 such that
lim pn
—
n too
— p, lim
n-> oo
—pn = —p .
Consequently,
xp = lim xPn , 2/
p
= nlim yPn
—
n > oo — foo
It is known that
X
—> oo x ,
= nlim ^ y = nlim
—
> oo
x > y, <=> xPn > yPri , Vn 1.

Hence
— yPn = yp.
xp = lim xPn
n —too ^ limn too
We claim that
xp > yp . ( 3.10 )
If not , i.e., xp = yp , then
x — —
xp
£
— n —> oo yp ) = y ,
= nlim
oo
1
lim ( xp ) p n(
•
1
Pn
contradicts x > y . Hence, ( 3.10 ) holds. The other direction can be proved
in the same way.
( v ) Let x > 1. We only prove the case that xp > 1 if and only if p > 0
(i.e. , 9 = 0 ) . If p > 0 , we can find pn , 5 e Q such that
p = lim pn , Pn S > 0, Vn 1.
—
n » oo ^
Then
xp = lim o?Pn x5 > 1.
5n —
KX
Conversely, let £p > 1. We claim that p > 0. First of all p = 0 is impossible.

If p < 0, then we have pn , 5 e Q such that
p = lim pn Pn 8 < 0, Vn 1.
—
n >oo
Then
xp x5 < 1.
—> oo xPn
= nlim
This proves the special case. The other cases can be proved similarly.
We recall that the horizontal straight line L has the property that QcL
by marking every rational number on L. Also, we have defined a natural
order on L: Any two points x , y e L satisfying x < y if and only if x is on the
left of y . On the other hand , one has Q C R , by identifying any r G Q with
an equivalent class of Cauchy sequence for which a representative {£n }n > i
is given by £n = r for all n 1. A natural question is what is the relation

^
between R and L? To look at this, we pick any x E R \ Q, let {£n }n i £ c( Q)
such that ^
x = lim
—y oo £n .
= nlim
Now , we construct { r]n } E c( Q) , which is increasing, and {£n } E c( Q) ,
which is decreasing, such that both are equivalent to {£n }. Since {£n } is
Cauchy, for any k 1, there exists an Nk E N such that
^ 1
1 ?» I < fc( fc + 1) ’ Vi, j Nk .
We may take N\ < < Ns < • • • . Thus,
1 1
$Nk ~
fc ( fc + l ) & rfc +1
< \ < + fc ( fc
+ l) ’
This implies
1 1 1 1
< 64 +1 + k 1 < N + k 1.
^
< CNfc + 1
,
k +1 + " ‘ Vfc
Hence, by taking
1 1
Vk = ~ T, Cfc = +£ k 1
we see that {%} is increasing and { ( k } is decreasing. Further
lim
—
k >oo
= klim
rjk
—>oo £n = x.
—t oo ( k = nlim
Thus, it is easy to see that {r?n } ~ {£n } ~ { ( n }. Now , r]k , ( k E <Q> are
well- marked on L and
2
Cfc - Vk = -> o.
^
Hence, there exists a unique point on L, denoted by x , such that
Vi < m < •• < •
Rk < x < ( k < • • < C2 < Ci >
* Vfc 1.
Then we naturally identify the point x E L with the real number x E R.
This shows that RCL On the other hand , any point on L must be a limit
point of a rational sequence. Therefore, by the property that R is closed
under limit process, we obtain that L C R. Hence, in such a way, we finally
obtain
NCNcZcQcR = L. ( 3.11)
The equality R = L explains the completeness of the real number set R
perfectly: The horizonal straight line L is exactly covered by R.
4 Complex Numbers
Before concluding this appendix of the real number system , let us ask the
following question: Is the real number system R perfect ? To answer this
question , we look at the following equation:
2
X + 1 = 0. ( 4.1)
We could just say the equation does not have a root in R . Therefore, we
encounter an equation of second order that does not have ( real ) solutions.
Recall the procedure of developing R from N to Z , then to Q and finally to
R , we could ask if it is possible to further extend R . This can be done by
introduce the root i of equation ( 4.1) . Thus, i is a symbol that satisfies
i 2 = -l .
Then we introduce the following set:
C = { a + bi a , be R }.
\
any element a -f bi e C is call a complex number, with a and bi called the
real part and the imaginary part, respectively. Also, we define the conjugate
of ( a + bi ) by
a + bi = a bi —
and the modulus
/
|a + bi \ = \ a2 4- b2 .
On C, we introduce the following operations , which are extensions of the
corresponding ones from R :
• Addition: ( a -f bi ) 4- ( c 4- di ) = ( a + c) 4- ( b + d )i .
—
•Subtraction: ( a -f bi ) — (c + di ) = ( a — c ) 4- ( b d )i .
• Multiplication: ( a 4 bi ) (c 4- di ) = ( ac — bd ) + ( be + ad )i.
- •
( a -f bi ) ( c — di )
•Division: c +
a bi
if c2 + d 2 > 0.
+ di c2 + d 2
Further , it is possible to define exponentiation . We omit that here. To
conclude, we state the following theorem called the fundamental theorem
of algebra whose proof will not be presented here.
Theorem 4.1. Every non-constant single-variable polynomial with com-
plex coefficients has at least one complex root . Consequently, every such a
polynomial of degree n has exactly n complex roots, counting multiplicity.
From this result , we see that the set C is algebraically closed , which is
perfect!
Bibliography
[ i] J . M . Almira, On Muntz type theorems , I. Surv. Apporx . Theory, 3 ( 2007) ,

152-194.
[2] C. Arzela, Sulla integrazione per serie , Atti Acc. Lincei Rend ., Rome , 1
(1885) , 532-537, 596-599.
[3] R. S. Borden , A Course in Advanced Calculus , Dover , New York , 1998.
[4] D. S. Bridges, Foundations of Real and Abstract Analysis, Springer-Verlag ,
New York , 1998.
[5] J . Dieudonne, Foundations of Modern Analysis , Academic Press, New York ,
1969.
[6] J . J . Duistermaat and J . A. C. Kolk, Multidinmensional Real Analysis I :
Differentiation; II: Integration , ( Translation from Dutch by J . P. van Braam
Houckgeest ) , Cambridge Univ. Press, 2004.
[7] W. Fleming, Functions of Several Variables , 2nd Ed . , Springer-Verlag, 1977.
[8] D . J . H . Garling , A Course in Mathematical Analysis, Vol. I: Foundations
and Elementary Real Analysis , Cambridge Univ. Press, 2013.
[9] L. H. Loomis and S. Sternberg , Advanced Calculus, Jones and Bartlett ,
Boston , 1990.
[10] H. Lou , Mathematical Analysis: Key Points , Difficulties and Extensions ,
High Education Press, Beijing , China, 2020 ( in Chinese) .
[11] W . A. J . Luxemburg, Arzela’s dominated convergence theorem for Riemann
integral , Amer. Math . Monthly, 78 (1971) , 970-979.
[ 12] W. Rudin , Principle of Mathematical Analysis , 3rd Ed ., McGraw- Hill, 1976.
[13] B. S. W. Schroder , Mathematical Analysis: A concise introduction , Wiley-
Interscience, 2008.
[14] T. Tao, Analysis I ,II , 3rd Ed . , Hindustan Book Agency, 2014.
257
[15] A. E. Taylor and W. R. Mann , Advanced Calculus, 3rd Edition , John Wiley
& Sons , New York , 1983.
[16] J . K . Truss, Foundations of Mathematical Analysis, Clarendon Press-Oxford,
1997.
If
/
Index
n-th root , 249 equi- , 133

continuous, 45
absolute value, 231, 239 convergence
addition , 3, 209 absolute, 154
adherent point , 22 conditional , 154
algebraic operation, 3 pointwise , 123
algebraic property, 3 quasi- uniform, 128
analytic extension , 159 uniform, 123
anti-derivative, 114 criterion
Abel’s, 176 , 178
bijection , 31 Dirichlet ’s , 176 , 178
binary operation , 209 critical point , 69
binary relation, 207
equal , 207 de Morgan’s Law , 2
boundary, 22 dense, 22
boundary point , 22 derivative, 65
Chain rule, 67
Cartesian product , 219 directional , 61, 66
closed ball , 25 Frechet , 62
closed set , 22 Gateaux , 61, 66
closure, 22 gradient field, 61
co- prime, 235 normal , 88
complex number partial, 64
conjugate, 256 diameter , 95
imaginary part , 256 difference , 223
modulus, 256 differentiability
complex numner Fechet , 62
real part , 256 differentiable
connected metric space, 22 Frechet , 66
connected set , 22 direct sum, 51
continuity directionally differentiable, 61
259
disconnected metric space, 22 simple, 98

disconnected set , 22 sine, 191
distance, 2 , 231 successor , 208
divergence, 151 trigonometric, 191
divisible, 226 uniformly continuous, 56
division , 239 upper semi-continuous, 48
divisor , 226 functionally dependent , 84
domain , 61
boundary Gateaux differentiable , 61
zero volume condition , 97 gradient , 61, 66
volume, 97 gradient field , 66
greatest lower bound , 16 , 248
equality
Parseval’s, 201 Hessian , 91
equation
Riemann’s functional , 159 inequality
equi-continuity, 133 Cauchy-Schwarz , 5, 193
uniform , 133 Holder ’s, 5
equivalence class, 220 Minkowski’s, 5
equivalence relation , 219 Young’s, 5
Euclidean algorithm , 218 infimum, 16, 248
Euclidean metric , 7 inner product , 9, 193
Euclidean norm, 9 integer , 220
Euler’s number , 186 negative, 222
exponentiation, 231 non- positive, 222
positive, 222
finite intersection property, 39 integrable
finite open cover property, 37 Lp-Riemann integrable, 106
first order Taylor expansion , 63 absolute Riemann , 106
fixed point , 50 integral
Fourier coefficient , 199 absolute continuity, 112
function definite Riemann, 100
continuous, 45 integrand , 100
cosine, 191 lower Riemann , 99
Dirichlet , 50, 101 Riemann definite, 111
exponential , 186 upper Riemann , 99
Gamma, 159 integration by parts, 115
inverse, 80 interior point , 22
level set , 86 interval of convergence, 179
logarithmic, 187 isolated point , 22
lower semi-continuous, 48 iterated integrals, 118
multi- variable, 45
periodic, 191 Jacobian , 67
Riemann , 50 Jacobian matrix , 67
Riemann zeta, 159
shift , 125 l’Hopital’s rule, 73
Index 261
least upper bound , 16 , 244 number

Lemma complex , 256
Abel’s, 175 irrational , 237
extended Riemann-Lebesgue, 148 negative, 230, 239
Fatou ’s, 141, 171 non-negative, 230 , 239
Levi’s, 140, 170 non-positive, 230, 239
Riemann-Lebesgue, 148 positive, 230, 239
level set , 86 rational , 227
level set surface, 83 real , 237
limit , 10
infinite, 15 open set , 22
limit inferior , 17, 48 orthogonal system, 194
limit point , 14, 22 orthonormal system, 194
limit superior, 17, 48 outward unit normal vector , 88
lower bound , 248
lower limit , 100 partial derivative, 64
lower semi-continuous, 48 first order , 64
partial sum , 151
map, 207 partition
mesh size, 95 rectangle, 95
metric, 2 refinement , 96
L°°- , 133 pathwise connected, 52
discrete, 2 Peano Axiom, 208
equivalent , 13 Picard iteration , 51
Euclidean, 7 point
induced , 3 adherent , 22
standard , 2 boundary, 22
strong equivalence, 7 critical, 69
stronger , 13 interior , 22
sup norm, 133 isolated , 22
weaker , 13 limit , 14, 22
metric space Principle of Mathematical Induction ,
compact , 33 208
complete, 33
multi-index , 92 quotient , 230
multiple, 226
multiplication , 212 radius of convergence, 179
real analytic, 182
natural number , 208 real number
negation , 222 canonical decimal representation ,
Newton-Leibniz formula, 114 33
norm real trigonometric polynomial , 192
L2 , 193
- reciprocal , 228, 238
p
-norm, 4 relatively closed , 25
^
Euclidean , 9 relatively open , 25
strong equivalence, 7 Riemann hypothesis, 160
Riemann integrable, 100 closure , 22
Riemann integral , 105 compact , 34
Riemann sum complementary, 1
lower , 99 connected , 22
upper , 99 countable, 31
dense, 22
scalar multiplication , 3 difference, 1, 207
sequence, 10, 236 disconnected , 22
bounded , 10, 236 domain , 61
Cauchy, 10, 236 empty, 1, 207
convergence, 10 equal , 207
divergence, 10 interior , 22
equivalent , 236 intersection , 1
equivalent Cauchy sequence, 35 natural number , 208
limit , 10 non-empty, 1, 207
limit point , 14 open , 22
minimizing , 53 open ball , 21
monotone, 16 proper subset , 207
subsequence , 13 relatively closed , 25
uniformly Cauchy, 123 relatively open , 25
series , 151 subset , 1, 207
absolute convergence , 154 totally bounded , 34
Cauchy test , 157 uncountable, 31
Cauchy’s integral, 158 union , 1
Cauchy’s root test , 160 space
conditionally convergence, 154 Banach space, 33
convergence, 151 connected metric, 22
d’Alembert ’s ratio test , 161 disconnected metric, 22
divergence, 151 Hilbert space, 33, 194
Fourier , 192 inner product , 9
geometric, 152 linear , 3
Harmonic, 152 metric, 2
partial sum, 151, 167 normed linear , 4
pointwise convergence, 167 product , 8 s
power , 179 separable, 43
rearrangement , 164 vector , 3
sum , 151 standard orthonormal basis, 64
trigonometric, 192 Stirling’s formula, 185
uniform convergence , 167 subspace, 3
Weierstrass M-test , 167 subtraction, 223, 239
set , 1, 207 successor , 208
at most countable, 31 supremum, 16, 247
boundary, 22
bounded , 34 tangent line, 86
closed , 22 tangent plane, 86
closed ball , 25 tangent vector , 86
Index 263
test Inverse Function , 80

alternative terms, 154 Lagrange Mean Value, 70
comparison , 154 Miintz-Szasz , 139
Gausss, 163 Nested Compact Sets, 40
Raabe, 162 Plancherel , 201
ratio comparison , 154 Pythagorean, 194
Theorem Rolle, 70
Abel , 180 Second Mean-Value, 117
Arzela’s, 142 squeeze, 15
Arzela- Ascoli, 135 Taylor , 91
Cauchy Mean Value, 70, 71 Weierstrass, 136 , 196
comparison , 15 trichotomy, 216
Contraction Mapping , 50 trigonometric polynomial , 192
Darboux , 94
Dini , 130 uniformly continuous, 56
Dominated Convergence, 170 unit of addition, 219
Fermat , 69 unit of multiplication , 226
First Mean-Value, 115 upper bound , 244
Fourier , 199 upper limit , 100
Fubini, 117 upper Riemann , 99
fundamental theorem of algebra, upper semi-continuous, 48
256
Fundamental Theorem of Calculus volume, 96
113
Implicit Function , 82 zero volume condition , 97
CPSIA information can be obtained
at www.ICGtesting.com
Printed in the USA
BVHW 011149060621
608889BV00002B/256 9 789811 221637

Jiongmin Yong - Mathematical Analysis - A Concise Introduction-World Scientific Pub Co Inc (2020)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Jiongmin Yong - Mathematical Analysis - A Concise Introduction-World Scientific Pub Co Inc (2020)

Uploaded by

Copyright:

Available Formats

MATHEMATICAL

Library of Congress Control Number: 2020055934

British Library Cataloguing-in-Publication Data

For any available supplementary material, please visit

Mathematical analysis serves as a common foundation for many research

1. Metric Spaces and Limits for Sequences 1

2. Functions on Metric Spaces 45

5. Uniform Convergence 121

Appendix The Real Number System 207

Metric Spaces and Limits for

Let us begin with some notions on set which is understood as a collection

(iv ) If a set S contains no element , it is called an empty set, denoted by

Definition 1.3. Let X be a non-empty set. A map d : X x X — R is

We present some examples.

called the standard metric on R , and ( R , p ) becomes a metric space.

X . Such a metric can always be defined on any non-empty set X . Hence,

Rm = { x = (x1, x2, • • • , xm ) | xk G R , 1 /c m}.

which is called the £ p- norm. We refer to ( Rm • ||p ) as a normed linear

Proposition 1.6 . L e t p e [ l , oo] . Then the above-defined £ p-norm satisfies

x + y || oo = max |x* + {/fc|

max |xfe| + max |j/fc| = 11 x 11 «, + \\ y\\oo -

This is called the Cauchy- Schwarz inequality Then it follows that .

II* + y||!= £ (** + = £ ( 1*k 2 + 2 xkyk + |yfe|2 )

Consequently, (iii) holds for p = 2.

Proposition 1.7. Let p , q £ ( 1 , 00 ) such that = 1.

( ii) ( Holder’s Inequality ) For any x , y £ Rm , the following holds:

(iii) ( Minkowski’s Inequality ) For any x , y £ Rm , the following holds:

\\ x + y \\ p ^ I ^ IIP + IIPIIP * ( 1.6 )

We try to find the minimum of / ( •) . To this end , 2 setting

we get r = 1. Clearly, by the first derivative test , one has

If AB = 0 , then x = 0 or y = 0 which leads to the conclusion trivially. We

This completes the proof.

( m max |*Y "

skip this example until they have read Chapter 6.

With such an inner product , Rm is called an inner product space. Clearly,

We call || • ||2 the Euclidean norm.

We introduce the following definition.

From the above definition , we see that a sequence { xn }n i is bounded ,

Proposition 2.2 . Let ( X , d ) be a metric space and { xn } C X be a se-

Consequently, for any m , n

Example 2.4 Let .

Hence, d ( - , •) is a metric on X . Now , for sequence xn = we have

In this case, d is said to be weaker than d.

Proposition 2.6 . Let { xn } be a sequence in Mm with xn = (x , x , • • • ,

Definition 2.7. Let ( X, d) be a metric space, and {xn}n be a sequence

Proposition 2.8 . Let ( X , d ) be a metric space , and { xn } C X be a se-

{ £ <72 (n ) } will be different , a contradiction.

The following gives a comparison theorem of sequences and their limits.

Proposition 2.10 . Let { xn } , { yn } G c and let

> 0, there exists an JV

This leads to ( noting

0 < 2£ < x - 2/ = x - x n + x n - 2/n

for some Afo

Next , we consider sequence with infinite limit.

Definition 2.12. Let { xn } G £ ° . If for any M > 0, there exists an N 1

then we say that xn approaches to oo as n goes to infinity, denoted by

The following comparison of sequences is obvious. We leave the proof

yn ->• -oo => xn — y — oo.

Proof ( i) First of all, since { xn } is bounded above, by a result from the

exists. Now , for any e > 0 , there exists an N 1 such that

Likewise, we may define

This is called the limit inferior ( liminf , for short ) of {xn}n i -

Proposition 2.15 . For any sequence { xn } e £°° , the following hold :