Professional Documents
Culture Documents
with
Point-Set Topology
OONALD L. STANCL
MILDRED L. STANCL
ST. ANSELM COLLEGE
MANCHESTER, NEW HAMPSHIRE
Neither this book nor any part may be reproduced or transmi~ted in any
form or by any means, electronic or mechanical, including photocopying,
microfilming, and recording, or by any information storage and retrieval
system, without permission in writing from the publisher.
MARCEL DEKKER, INC.
270 Madison Avenue, New York, New York 10016
ln memory of
Bruce Amert Luzader
Minnie Farson Luzader
Frances W olff Stand
Preface
Vl
Preface
Preface
Vil
Contents
Preface
1
16
31
41
3 Topology
3.1 Topological Spaces
3.2 Open Sets and Closed Sets
57
4 Continuous Functions
4.1 Continuity
4.2 Connectedness and Compactness
75
41
51
57
66
75
90
107
107
116
ix
Contents
126
Metric Spaces
139
139
6.1
6.2
6.3
6.4
7
Sequences of Functions
134
149
156
165
177
177
190
Calculus
201
201
226
241
265
Bibliography
281
Index
283
1
Sets and Functions
ln this book we shall study various sets and functions, paying particular
attention to the set of real numbers and to functions whose values lie in
the real numbers. This introductory chapter presents some basic definitions and results concerning sets and functions which will be used
throughout the remainder of the text.
1.1
SETS
o
Figure 1.1
Element of a set.
Examples 1.1.2
N={l,2,3,4, ... }
= {: m E Z, n E Z, n -:f
1
o}
s=
-3
-2
-1
{!, 2, 3}
. 3
Sets
S={xlxeN,x<4}
or
S = {xeZ 1x 3 - 6x 2 + I Ix - 6 =O}
is called an open interval in R (see Figure 1.3). The real numbers a and b
are called the endpoints of (a, b). Note that the open interval (a, b) consists
of ali real numbers which are between a and b on the real line, but that the
endpoints a and b do not belong to (a, b). The open interval (O, 1) is
referred to as the open unit interval.
a
Figure 1.3
a~
b. The set
is called the closed interval in R with endpoints a and b (see Figure 1.4).
Note that the endpoints a and b belong to the closed interval [a, b]. Also,
if a = b, then [a, b] =[a, a] ={a}, so a closed interval may consist of a
single point. The interval [O, l] is the closed unit interval.
[a,b)={xeRla~x<b}
are half-open intervals in R with endpoints a and b (see Figure 1.5).
7. If a e R, then the sets
(-oo,a]={xeRlx~a}
are called closed rays in R (see Figure 1.6). The set Ris both an open ray
b
(a,b]
b
[a,b)
Sets
a
(a,+:x:)
a
(-:x:,a)
a
(a,+:x:)
a
(-:x:,a]
Figure 1.6
Rays.
R=(-oo,+oo)
Next we define the notion of a subset of a given set.
S C T
cf_
Examples 1.1.4
</JcS
ScS
and
S=S
NcZcQcR
Furthermore, N is a proper subset of Z, Z is a proper subset of Q, and Q
is a proper subset of R.
3. Let aeR and beR, with a< b; then
(a, b) is a proper subset of [a, b]
and
(b,
4.
subset
subset
5.
T=(O, 1)
Sets
(see Figure l.8b). If S n T = </J, we say that S and Tare disjoint sets.
3. The complement of T in S is the set S - T defined by
S - T = {x 1x e S and x ~ T}
SvT
union
(a)
SnT
intersection
(b)
S-T
complement
(e)
Bxamples 1.1.6
SuT={l,2,3,4}
SnT={3}
S-T={l,2}
T-S={4}
T = {(l, 3), (1, 4), (2, 3), (2, 4), (3, 3), (3, 4)}
and
T x S = {(3, 1), (3, 2), (3, 3), (4, 1), (4, 2), (4, 3)}
Sets
9
2. Let
R 2 =R xR={(x,y) lxeR,yeR}
(O, 1) n[l, 2] =
and
[l, 2] -[2, 3] = [l, 2)
[l, 2] - (O, l)
= [l, 2]
Note that (O, 1) and [l, 2] are disjoint, but that [l, 2] and [2, 3] are not
disjoint. The Cartesian product
[l, 2] x [2, 3] = {(x,y) 1xe[l,2], ye[2, 31}
is the subset of the Euclidean plane R 2 shown in Figure 1.10.
The algebraic rules which govern the set operations ofunion, intersection, and complementation are given in the following proposition.
y-axis
- - - - - - - , (x,y)
1
1
1
1
- - - - + - - - - - - - + - - - - - - - - x-axis
X
Figure 1.9
10
y
3
--~
--~
1
1
1
1
2
Proposition 1.1.7
hold:
l.
2.
3.
Idempotent laws
Commutative laws
Associative laws
4.
Distributive laws
S.
deMorgan's laws
SuS=S
SnS=S
SuT=TuS
SnT=TnS
Su(Tu V)= (Su T) u V
S n (T n V) = (S n T) n V
Su(Tn V)= (Su T) n(Su V)
S n (Tu V)= (S n T) u (S n V)
S - (Tu V) = (S - T) n (S - V)
S - (T n V) = (S - T) u (S - V)
11
Sets
again y eS u (T n V). Therefore if (S u T) n (S u V) =F </J then
[(Su T) n(Su V)] e [Su(Tn V)]
and
But the commutative and associative laws of Proposition 1.1.7 (together
with mathematical induction) imply that in such a union neither the order
of the sets nor the manner of their grouping matters. Thus we may write
Definition 1.1.8
12
set. We define the union and intersection of the family of sets {XIX 1 tX EA}
as follows:
IXE
n XIX= {x
IXE
1 XEXIX
The sets {XIX ltXEA} are mutually disjoint if and only if XIXnXp=</J
whenever tXEA, /JEA, and tX =F {3.
Examples 1.1.9 1. Let A = { 1, 2, 3}, and for each tX EA let XIX denote the
closed interval [tX, tX + 3] in R; then
IXE
and
IXE
Here the index set A is finite and we have taken the union and intersection
of finitely many closed intervals in R.
2. Let A= N, and for each nEN let Xn = {l, ... , n}; then
U XIX= neN
U Xn=N
n XIX = n Xn = { 1}
and
IXE
czeA
neN
Here the index set is infinite, and we have taken the union and intersection
of infinitely many sets, each of which was a subset of N.
3. For each n EN, let Xn be the half-open interval [n - 1, n) in R; then
U Xn =[O, +oo)
neN
and
l Xn=</J
neN
(see Exercise 15 of Section 1.1). Note that the sets {Xn} are mutually
disjoint and that lneN Xn = 0.
4. The previous example suggests that if {XIX 1tX EA} is a family of
mutually disjoint sets, then
IXE
XIX=</J
This is indeed true (see Exercise 16 in Section 1.1). However, the intersec-
Sets
13
tion may be empty even if no two of the sets {X" 1tX E A} are mutually
disjoint. To see this, let
for each n eN. No two of the sets {Xn 1n eN} are- disjoint, for if m ~ n
then Xm e Xn and hence Xn n Xm =F f/J. However, it is easy to check that
Xn=</J
neN
+ l) in R 2 ; then
Ifwe re)>resent R 2 by means ofthe Cartesian coordinate system ofExample 2 in Examples l. l .6, the above union is the line with slope l and
y-intercept l (see Figure l.l l).
Now that we can form unions and intersections of arbitrarily many
sets, we must generalize the distributive laws and deMorgan's laws of
Proposition l.l.7.
Proposition 1.1.10
{X" 1 tXEA}
{(r,r+ 1) 1 rER}
Figure 1.11
14
Yu ( (1 X"') =
ex e A
3.
Y-
U X"'=
ex e A
4.
Y - (1 X"' =
ex e A
(1 (
ex e A
Yu X"')
(Y-X"')
ex e A
U (Y -
X"')
ex e A
if
the
inclusion
also
holds
when
15
Sets
SuTuV
Sn(TuV)
SnT
S-T
T-S
TxV
2.
S- T
4.
5.
6.
7.
8.
Su(TnV)
S-(TnV)
and
V-(S-T)
VxT
3.
SnTnV
SnT
T- S
SuTuV
and
SnTnV
V -(Su T)
16
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
1.2 FUNCTIONS
ln arder to study sets more thoroughly, we need to be able to compare one
set with another. This means that we must have some method of "pairing" their elements and leads us to the concept of a function from one set
to another.
Let S and T be sets. A function from S to T is a subset
T such that
Definition 1.2.1
f of S x
l. For each x eS there is some y E T such that the ordered pair (x, y) Ef
2. If (x, y) and (x, y') are elements off, then y = y' in T.
The set S is called the domain of the function and the set T the range of
the function. If S = T, we refer to a function from S to S as a function
on S.
Note that according to this definition, a function from S to Tis a
17
Functions
= {(l,
and
Since each element of S appears as the first coordinate of some ordered pair belonging to f, and no element of S appears as the first coordinate of more than one ordered pair belonging to f, f is a function from S
to T. Note that not every element of the range T appears as the second
coordinate of an ordered pair inf; note also that the element pe T appears
as a second coordinate more than once.
The set g is also a function from S to T; in this case, every element of
T does appear as the second coordinate of some ordered pair belonging to
the function, and no element of T appears more than once.
The set h is not a function from S to T because it violates condition
(l) of the definition: 1 eS but there is no y e T such that ( l, y) eh. (However, h is a function from the set {2, 3} to T.)
The set k is not a function from S to T because the element l of
S appears as the first coordinate of more than one ordered pair belonging
to k.
2. Consider the subsetf of N x Q given by f= {(n, 1/n) 1neN}. We
claim thatf is a function from N to Q. This follows beca use if n eN, then
(n, l/n) e f, and if (n, y) and (n, y') both belong to f, then y = l/n = y'.
3. Let
f={(x,x 2 ) lxeR}
and
= {(x 2, x) 1 xeR}
18
Examples 1.2.3
f(l)
=(X
and
/(2) = /(3) =
This is the functionf of the first example in Examples l.2.2. The function
y
{(x,x) 1 XER, x;;., O}
19
Functions
f:S ~ T,
f(x) = y
A function as a mapping.
Figure 1.13
g(2) =
(l(
g(3) =
2. Let/: R-+R be given by/(x) = x 2 for all xeR. This is the function
f of the third example in Examples 1.2.2. Its graph is depicted in Figure
1.14.
3. Let/: R -+R be given by /(x) = lxl for ali xeR, where by definition,
if x~O
if x<O
This is the absolute value function introduced in Example 4 of Examples
1.2.2.
y
f(x) = x2
Figure 1.14
20
For each x in the domain, f(x) is defined and belongs to the range.
Iff(x) =y andf(x) =y', then y =y'.
Thus /: S - Tis onto if every element of its range gets "used," and it
is one-to-one if no two elements of its domain are assigned to the sarne
element of its range ( see Figures 1.15 and 1.16).
1. The function f of Example 1 in Examples 1.2.3 is not
a function of S onto T beca use y E T and there is no x E S such that
f(x) = y. Furthrmore,fis not one-to-one because/(2) = f3 and/(3) = f3.
However, the function g of the sarne example is both onto and oneto-one.
2. The function of Example 2 in Examples 1.2.3 is not a function
of R onto R beca use, for instance, - 1E R but there is no x E R
such that f(x) = - 1. It is not one-to-one because, for instance,
f ( - 1) = f (l) = 1.
3. The absolute value function of Example 3 in Examples 1.2.3
is neither onto .R nor one-to-one. (Why?) The square root function of
Example 4 in Examples 1.2.3 is onto [O, + oo) because if y e [O, + oo ),
then
Examples 1.2.5
y 2 e [O, +oo)
and
Functions
21
f is onto
f is not onto
Figure 1.15
Jx=Jz,
then
=z.
4. Let S and T be any nonempty sets and let e e T. Let /: S -+ T be
given by f(x) = e for all x e S. Such a function is called a constant function.
A constant function cannot be onto if its range contains more than one
element; it cannot be one-to-one if its domain contains more than one
element.
X
22
f is one-to-one
f is not one-to-one
-+
Functions
23
f(U)CT
Examples 1.2.7
given,by ,
f(a) = 3
f(b) = 4
f(c) = 3
24
f(x) = x 2 e [O, 4]; hence/([O, 2]) e [O, 4]. On the other hand, if xe [O, 4],
then Jxe [O, 2] andf(Jx) = xe [O, 4]; hence [O, 4] c/([O, 2]). Therefore
/([O, 2]) =[O, 4].
Now consider 1- 1([0, 4]) = {xeR 1x 2 e [O, 41}. If xe [ -2, 2], then
~learly f(x) = x 2 e [O, 4]; hence [
2] c/- 1([0, 4]). On the other hand,
2
2
if f(x) = x e [O, 4], then x = y' x e [ -2, 2], so 1- 1([0, 4]) e [ -2, 2].
Therefore/- 1([0, 4]) = [ -2, 2].
-'fo
Our next proposition tells us how set operations are affected by functions. lt states that taking preimages preserves containments, unions,
intersections, and complements, while taking images preserves containments and unions. Intersections and complements are not in general preserved by taking images.
Let S and T be sets and /: S
subsets of S and V1 and V2 be subsets of T.
Proposition 1.2.8
1.
-+
T. Let U1 and U2 be
If U1 e U2 , then/(U1) c/(U2).
If V1 e V2 , then/- 1(V1) c/- 1(V2).
25
Functions
=/d/J, we have
[f- 1(V1) nf- 1(V2)] cf- 1(V1 n V2 )
= </J. Therefore
Example 1.2.9 Let /: R -+R be given by f(x) = x 2 for ali xeR. Let
U 1 = [ - 2, O] and U2 =[O, 2]. We have U 1 n U2 = {O} and thus
whereas
f(U1) nf(U2) =[O, 4] n[O, 4] =[O, 4]
This shows that in general/(U1 n U2) # f(U 1) nf(U2 ). Of course Proposition 1.2.8 says that /( U 1 n U2) e [f( U1) n /( U2)], and this is certainly the
case here.
Now consider f(U 1 - U2) and/(U1) - /(U2)_. We have
f(U 1 - U2) = /([ -2, O))= (O, 4]
whereas
26
(g f) (x) = g(f(x))
Functions
27
Examples 1.2.13
f(l) =
f(2) =ex
f(3) =
and
p, y }, and
U = {u,
g(ex) = g(p) = u
v, w }.
g(y) = w
and
+ 1,
+l
Sincef and gare both one-to-one and onto, g ofis one-to-one and onto.
ln this case we can also form the composite functionf g. Thus we have
f g given by
(f g)(x) = f(g(x)) = f(x
+ 1) =
2(x
+ 1)
28
f-1:f(S)-+S
=X
and
(j
f- 1)(y) = Y
29
Functions
for ali xeR. Thereforef- 1(x) =x-1 for ali xeR. Note also that
(f-I of)(x) = f- 1(/(x)) = f- 1(X + 1) =(X+ 1) - 1 =X
as required.
3. Let /:[O, +oo)-+[O, +oo) be given by f(x)=Jx for ali
xe[O, +oo). Sincefis one-to-one and onto,f- 1:[0, +oo)-+[O, +oo).
Thus for ali x e [O, + oo ),
X=
30
Jx2+1)
f(x)
= {~
if X =/= 0
if X =0
U) S = T = R,f given by
if X ~0
if X< 0
f(x) = {: + 1
2.
3.
4.
/(2) =
/(3) =(X
and
g(l) = g(2) = }'
g(3) =
1- 1(V),
g-i(V).
5.
and
2x
h(x) = { X -1
if X ~t
if X <t
and
31
/3
/(2) = }'
/(3) =
(l(
/(4) =
/3
and
g(oc)
=V
g(/3) = z
g(y)
=u
1.3
Suppose we have two sets and we wish to compare the number of elements
which they contain. One way to do this is to pair the elements of one set
with those ofthe other in a one-to-one manner: if every element ofthe first
set is paired with an element of the second set, and if each element of the
second set is paired with some element of the first one, then the sets must
have the sarne number of elements. We make this notion precise with the
following definition.
32
ifzeN
ifzeZ -N
d-e
b _a (x - a)
+e
for all xe [a, b]. The functionfis one-to-one and onto (check this), and
thus any two closed intervals in R which contain more than one point are
equivalent. The sarne function may be used to show that any two open
intervals in R are equivalent.
Now we are ready to give a rigorous definition of finite and infinite
sets.
Definition 1.3.3 A set is infinite if it is equivalent to a proper subset of
itself. A set is finite if it is not infinite.
33
4. Let [a, b] be a closed interval in R, with a < b. Choose real numbers e and d such that a < e < d < b. Since the interval [e, d] is a proper'
subset of [a, b] and since any two closed intervals having more than one
point are equivalent by Example 4 of Examples 1.3.2, it follows that [a, b]
is infinite. Thus any closed interval having more than one point is infinite.
Similar arguments show that open intervals, half-open intervals, and rays
in R are infinite.
Intuitively, we may think of equivalent sets as being of the sarne
"size." Thus if two sets are equivalent, then either both of them must be
finite or both must be infinite. Furthermore, if Se T and Tis finite, then
S must be at least as "small" as T, and hence S must itself be finite. On
the other hand, if S e T and S is infinite, then T must be at least as
"large" as S and hence must be infinite. The following proposition and its
corollary prove these facts.
Proposition 1.3.5
1.
2.
Proof: 1. If the set S is equivalent to the infinite set T, then there exists
a one-to-one function f from S onto T and also a one-to-one function g
from T onto some proper subset V of T. Consider
See Figure 1.20. The function (f- 1 g) f is one-to-one and onto ( why?),
r-1
Figure 1.20
(f- 1 g)
34
if XES
if xET-S
Corollary 1.3.6
1.
2.
Example 1.3.7 The set Z of integers, the set Q of rational numbers, and
the set R of real numbers are ali infinite sets because each of them contains
the infini te set N.
There is more than one kind of infinity; that is, infinite sets differ in
the type or "intensity" of their infiniteness. As we shall see, the sets N, Z,
and Q ali possess the sarne type of infinity, that which is the least "intense." The set R of real numbers, on the other hand, has a different and
more "intense" type ofinfiniteness. Our next result is the first step toward
a classification of the different types of infinity.
Proposition 1.3.8 If Sisa nonempty set, then either there is some n EN
such that S is equivalent to {l, ... , n}, or else S conta.ins a subset equivalent to N.
Proof: Suppose that S is not equivalent to {1, ... , n}, for any n EN.
Since S is nonempty, we may choose an element x 1 eS. If S - {x 1} = </J,
35
then S = {x 1}; but since the set {x 1} is obviously equivalent to {l}, this
cannot occur. Hence S - {x1} =F </J.
Now suppose n ;;?; l and assume that we have selected n distinct elements x., ... , Xn of S. If S - {x., ... , xn} = </J, then S would be equivalent to {l, ... , n }. Therefore S - {x 1, , xn} =F </J, and we may select an
element Xn + 1 e S such that Xn + 1 ~ {xi> ... , Xn}. Thus by induction we
produce a subset {x 1, , Xm xn + 1, } of distinct elements of S, and
this subset is clearly equivalent to N by means of the functionf defined by
f(xn) = n for all n eN. D
Corollary 1.3.9 Every finite set is equivalent to {l, ... , n}, for some
neN. Every infinite set contains a subset equivalent to N.
36
denominators, as follows:
o
l' l'
-1
1 '
-1
2'
2'
-2
-3
l'
1 '
l'
2 '
2'
2
-1
3' 3'
3 '
3'
-1 2
o
4' 4' 4' 4'
-2 3
2 ' 2'
-2 3
3 ' 3'
-2 3
4 '
4'
' ...
-3
2
-3
3 ' ...
-3
4 ' ...
;1
o
-2
1,
2'
2
3'
3
-1
4
-2 3
,
2'
2
-2 3
3
-2
4'
. 3'
.
4'
-3
1 , ...
-3
.....
2
-3
...
3
-3
....
37
Proof: If Ris countable, then sois [O, l], by Proposition 1.3.12. Therefore it suffices to show that [O, I] is uncountable.
Each xe [O, l] has a decimal expansion of the form
where a;e{O, 1, ... , 9} for each i. Some numbers have two such expansions: for instance
1
2=
However, if a number does have two such expansions, ali the digits from
some p-J'lt on in one of them will be zeros, while ali the digits from some
point on in the other will be nines. Let us agree that if xe [O, l] has two
decimal expansions, we will always use the one which has a tail of nines.
With this agreement in force, we may consider that every xe [O, l] has
associated with it a unique decimal expansion.
Now assume that [O, l] is countable. Since [O, l] is infinite, it must be
denumerable, and hence we must be able to index the elements of [O, l] by
the natural numbers, thus:
38
Suppose that for each neNwe let O.an 1an2an 3 be the unique decimal
expansion of Xm so that
Xi
= O.a11a12a13
X2
= O.a21022023
uncountable.
The proof of the corollary will be left as an exercise.
There is much more to be said about set equivalence, finite and infinite sets, and the various types of infinity than we have been able to
present here. The reader interested in pursuing these topics may begin by
studying the appendix.
EXERCISES
1.
Prove:
(a) If set S is equivalent to set T, then set Tis equivalent to set S.
39
f(x) = 1 +
5.
6.
7.
8.
9.
10.
lxl
2
The Real Numbers
Let
xe R.
defined by
41
lxl, is
42
if X ~0
if X<
- y I See
ll=O
141=4
l-41=4
l5fil=5J2
l4-Il=3
II-41=3
Our first proposition lists some of the properties of absolute value and
distance in R.
lxl ~O.
lxl =O if and only if x =O.
X+ y
~0, SO
lx + YI = x + Y = lxl + IYI
lyl
IXI
~~
o
IX - yl
43
+ y < O,
so
lx
lx + Y 1 = x
If x
+ y <O,
then lx
+ YI =
+Y < x -
-(x + y)
lx 1 + IY 1
-x~x,so
lx + YI
-x - Y ~ x - Y
lxl + IYI
44
Examples 2.1.6
is a
tX
tX
tX ~ tX'.
or infimum, for S if
f3 is a
f3
{3'.
Proof: We prove the proposition only for the case of least upper
bounds.
45
Suppose ex 1 and ex 2 are both least upper bounds for the set S. Then by
(2) of Definition 2.1.7, ex 1 an upper bound and ex2 a least upper bound
implies that ex 2 ~ ex1> while ex2 an upper bound and ex 1 a least upper bound
implies that ex 1 ~ ex 2 Therefore ex 1 = ex2 D
Proposition 2.1.8 allows us to define the least upper bound and the
greatest lower bound of a set.
Definition 2.1.9 Let S be a nonempty subset of R. If they exist, the
unique least upper bound of S is denoted by sup S and the unique greatest
lower bound of S by inf S.
Examples 2.1.10 1. We claim that sup (O, 1) = 1 and inf(O, 1) =O.
Since we know that 1 is an upper bound for (O, 1), to show that 1 is the
least upper bound it suffices to show that if ex is any upper bound for (O, 1)
then 1 ~ ex. Suppose the contrary; that is, suppose that ex is an upper
bound for (O, 1) such that ex < 1. Since ex is an upper bound for (O, 1), we
must have O< ex. But then x =ex+ ( 1 - ex)/2 is in (O, 1) and ex < x, thus
contradicting the fact that ex is an upper bound for (O, 1). Therefore ex < 1
is impossible, so sup (O, 1) = 1. A similar argument shows that
inf (O, 1) =O. Note that in this example sup S and inf Sare not elements
of the set S.
2. The argument of the previous example shows that sup [O, l] = 1
and also that inf[O, l] = 1. Note that here both sup S and inf S are
elements of S.
3. Let S = {I/n 1 n e N}. We claim that sup S = 1: for 1 is certainly
an upper bound for S, and no real number ex < 1 can be an upper bound
for S because 1 e-S-:"Note that sup S e S.
4. If S is not bounded above, then of course sup S cannot exist; if S
is not bounded below, then inf S cannot exist. Hence sup N, sup Z, and
sup Q do not exist, nor do inf Z and inf Q. However, inf N = 1.
46
Theorem 2.1.11
rational number.
Proof: We must show that if x and y are real numbers with x < y, then
there is some rational number r such that x < r < y. If x <O and y >O,
we may take r = O, so that it suffices to prove the theorem for the cases
O~ x < y and x < y ~O.
Assume first that O< x <J'. By the Archimedean Property for R,
there exist natural numbers m and n such that mx < 1 and n(y - x) > 1.
Let k be the larger of m and n; k is a natural number, and since kx > l
and k(y - x) > 1, we have I/k < x and l/k < (y - x). Again by the
47
Since
1
p
-<x<k
k
But then
q+l q 1
1
x<--=-+-<x+-<x+y-x=y
J2,
48
.J2
But then reQ and r 2 < 2, so reS, and thus sup S < r contradicts the fact
that sup S is an upper bound for S. Therefore sup S =
.J2.
.J2,
Next we remove from C 1 its open middle thirds ( 1/9, 2/9) and (7/9, 8/9) to
obtain
-[O, !]9
C2 -
2] 1]
[~ !] [~3' 9 u [~9'
u 9' 3 u
J[ J
J[ J
1 u 27'
2 9
l u [29' 27
7 u 27'
8 3
l u [23' 27
19] u [20
7]
C3 = [ O, 27
27' 9
u
[2627' l J
[9'825]
27
u
49
2
3
E J
_1_
9
E:3 B
E
_g_
9
_1_
2
3
BB
E
--9
B E3
o
C3
Figure 2.2
(See Figure 2.2.) Continuing in this manner, for each n eNwe define a set
Cn. The Cantor ternary set C is then given by
C=
Cn
neN
Thus the Cantor ternary set C consists of exactly those elements of [O, l]
which remain after ali open middle thirds have been removed. Another
way of writing the Cantor set is to let
C =[O, l] -
LJ
Sn
neN
50
(a) lx - 21=3
(b) lx - 21<3
(e) l3x+5l~6
(d) 12-xl ~ 4
(e) IS+ 2xl ~ 6
(f) lx+ll<lx-31
2. (a) Prove 1, 2, 3, and 4 of Proposition 2.1.3.
(b) Prove Corollary 2.1.4.
3.
4.
5.
6.
7.
8.
9.
12.
13.
14.
51
52
(v12
Figure 2.3
If r e Q and S,
Proposition 2.2.2
cut.
____
_______
,_ _,/
v
\...._
'-------.
V
..... ,..
Figure 2.4
Dedekind cuts.
53
+y
54
Theorem 2.2.6
Property.
Proof: Let X be a nonempty subset of R which is bounded above. We
will show that X has a least upper bound.
Every real number xeX is a Dedekind cut and hence is a nonempty
proper subset of Q which satisfies properties ( 1) and (2) of Definition
2.2.1. Let
S = (qeQ 1 qex for some xeX}
Thus
S=
Ux
xeX
55
2.
Prove that two rational Dedekind cuts S, and Sq are equal if and only
if r = q.
If x and y are real numbers (Dedekind cuts), define
x
3.
4.
-x
{qeQ l -q~x}
-x
5.
xy = Ou{qp 1 qe x, pe y, O~ q, O~p}
(a) Prove that xy is a Dedekind cut.
(b) Prove that xO =O.
56
3
Topology
Thus far we have been concerned mainly with combining sets by means of
set operations and comparing them using functions. ln so doing we have
been looking at sets primarily from an externai point of view, and have
not examined their internai structures in any detail. lt is important to
consider the internai structures ofsets, however. For instance, the set R of
real numbers is more than just an uncountable collection of elements:
certain subsets of R (such as the open intervals) have important properties
of their own, and we cannot understand the real numbers thoroughly
unless we examine these subsets and their properties. ln this chapter we
will study the internai structures of sets by considering certain collections
oftheir subsets known as topologies. We begin by introducing the concept
of a topological space and then explore such spaces with the aid of continuous functions. Ali our general results about topology can be applied to
the set R of real numbers, and beginning in this chapter and throughout
the rest of the text, the reader should always give thought to what our
general definitions and results say about R.
3.1
TOPOLOGICAL SPACES
58
Topology
Definition 3.1.1 A topological space is a pair (X, r), where X is a set and
r is a collection of subsets of X such that
l. Xer and 0er.
2. If A is an index set and Grxe 't" for each oieA, then UrxeA Grxe r.
3. If A is a finite index set and Grxe for each oieA, then lrxeA Grxe.
Such a collection r of subsets of X is called a topology on X, and X is then
referred to as the space, or underlying point set, of r. Also, the elements of
X are called the points of the topological space. A subset G of X is an open
set in the topology r if G e r; a subset F of X is a closed set in the topology
r if its complement X - Fe r.
Notice that 0 and X are subsets of X which are both open and closed in
the topology. Also, in this example every subset of the topological space
is either open, closed, or both open and closed.
2.Let X= {x,y,z} and let
1
{0, {x}, X}
't"4
The collections r 1 and r 2 are topologies on X(check this), but r 3 and r 4 are
Topological Spaces
59
{x}u{y} = {x,yH-r3
while -r4 is not a topology on X because
{x, y} n {x, z}
{x} ~'r4
Notice that there are subsets of Xwhich are neither open nor closed in the
topology -r 1
3. Let X be any set and let -r = {S 1 S is a subset of X}; then -ris a
topology on X, called the discrete topo/ogy on X. Every subset ofthe space
is both open and closed in the discrete topology.
4. Let X be any set and let -r = {0, X}; then -r is a topology on X,
called the indiscrete topology on X. ln the indiscrete topology, the empty
set and the entire space are the only open sets and also the only closed
sets.
Assigning a topology to a set is called topo/ogizing the set. There are
always at least two ways to topologize a set, for we may always assign it
either the discrete topology or the indiscrete topology.
Now consider the set R of real numbers. We can of course topologize
R by assigning it either the discrete or the indiscrete topology, but these
are not very interesting. Instead, we will use the open intervals as the basis
for a topology on R. This topology, called the natural topology on R,
will be of fundamental importance to us throughout the remainder of this
text.
Topology
60
Now let A be an index set and suppose that G~ E r for all r:t. e A. If
= </), then U~EA G~ E r by what we have just proved, so let us
assume that U~EA G~ # </). If x E U~EA G~, then there is some /3 e A such
that x E Gp. But since Gp E r, there is some open interval
lx e Gp e U~EA G~ such that x E lx, and therefore the set U~EA G~ E r.
Finally, let A be a finite index set and suppose that G~ E r for all rx e A.
Again, if n~EA ~ = </), we are done, so assume that n~EA ~ # </) and let
xen~EA~ Then XEG~ for each r:t.E, and since G~Er, for each r:t.Er
there is some open interval (a~, b~) e G~ such that x E (a~, b~). Because A
is a finite set, the set {~ 1 rx E A} is finite and therefore has a largest
element: let a = max {~ 1 rx E A} be this largest element. Similarly, let
b = min {b~ 1 rx e A}. Then xe (a, b) e (a~, b~) for all r:t. e A, and hence
(a, b) is an open interval contained in n~EA ~ such that XE (a, b). Therefore n~EA ~E r, and we are done. D
U~EA G~
Topological Spaces
61
= {G e
62
Topology
Proof: Let xeG. Since G is open in the natural topology, there is some
open interval lx such that x elx e G. Let Jx denote the union of all subsets
of G which contain x and are either open intervals or open rays. lt is easy
to check that Jx must itself be either an open interval or an open ray. (See
Exercise 18 of Section 1.1, Chapter 1.)
We claim thatif xeG andyeG, then either Jx nJy = </J or Jx = JY. For
if Jx nJy # </J then Jx uJY must be either an open interval or an open ray,
and since x eJx u JY and (Jx u Jy) e G, the definition of Jx tells us that
(Jx uJy) e Jx. Similarly, (Jx uJy) e JY. Therefore Jx = JY whenever
Jx n JY # </J. Clearly G = UxeG Jx, and thus we have established that G is
a union ofs'ets each ofwhich is either an open interval or an open ray, and
that the distinct sets in the union are mutually disjoint. But any union of
mutually disjoint subsets of R is countable, because each set in the union
must contain a rational number which does not belong to any of the other
sets in the union, and hence there is a one-to-one correspondence between
the sets in the union and a subset of the rational numbers. D
Topological Spaces
Proof:
63
0=Yn0e'.
Let A be an index set and suppose that G ~e ' for ali oc e A; then for
each oc e A we have G~ = Y n G,,, for some G,,, e Thus
u G~ u
=
a.e A
(YnG,,,) = Yn
a.e A
(u G,,,)
a.e A
ds
la.eA
Examples 3.1.9 1. Let X= {x, y, z} and "t" = {0, {x, y }, X}. Let
Y = {y, z }. The subspace topology induced on Y by is the topology
' =
{0, {y }, Y}.
EXERCISES
1.
Let X= {x, y }. Find ali topologies on X. For each topology, find ali
the open sets, ali the closed sets, ali the sets which are both open and
closed, and ali the subsets of X which are neither open nor closed in
the topology.
64
Topology
2.
3.
4.
5.
6.
7.
8.
9.
10.
Prove that ris a topology on R,. that it is neither the discrete nor the
indiscrete topology, and that it is smaller than the natural topology.
11. Considering the natural numbers N and the rational numbers Q as
subsets of R,
(a) Show that the natural topology on R induces the discrete topology on N;
(b) Describe the open sets in the subspace topology induced on Q
by the natural topology on R.
12. Let R have the open ray topology ofExercise 10 above. Describe the
open sets in the subspace topologies induced on N and on Q by the
open ray topology.
13. Let X be a set, Y a nonempty subset of X, and r 1 and r 2 topologies
on X with r 1 # r 2 Is it possible for r 1 and r 2 to induce the sarne
subspace topology on Y? Justify your answer.
14. Let r 1 and r 2 be topologies on a set X and define
r 1 rH 2 = {G e X 1G e r 1 and G e r 2 }
and
Topological Spaces
15.
16.
65
Fa. effe.
for some real numbers h and k and some positive real number r.
(a) Let r 1 = {G e R 2 G = f/J or G is a union of open rectangles}.
Prove that r 1 is a topology on R 2 . This topology is called the
product topology on R 2 .
(b) Let r 2 = {G e R 2 G = f/J or G is a union of open discs}. Prove
that r 2 is the product topology of part (a).
Let X be a set, r be a topology on X, and x e X. A subset V of X is
a neighborhood of x if there is some open set G e r such that G e V
and xeG.
(a) Let X= {x, y, z} and r = {f/J, {x }, {x, y }, X}. Find ali neighborhoods of each element of X.
(b) Suppose X= R and ris the natural topology. Show that Vis a
neighborhood of x if and only if there is some open interval I
such that xe/ and I e V.
(e) Suppose r is the discrete topology. For any xeX, describe ali
neighborhoods of x. Do the sarne if r is the indiscrete topology.
(d) Prove that a subset G of X is open in r if and only if G is a
neighborhood of each of its points.
17.
66
Topology
1.
Ua.eA
2.
If A is finite,
X.
67
To prove the converse, let G be a subset of X such that for each xeG
there is some open set Hx such that xeHx e G. But then G = UxeG Hx is
a union of open sets and hence is itself open. D
We now have two characterizations for open sets: a global one which
says that a set G is open if and only if it belongs to the topology, and a
local one which says that a set G is open if and only if we can surround
each of its points with an open set which is itself contained in G.
Now we tum our attention to closed sets. We have a global characterization of closed sets: a subset F of the topological space X is closed in X
ifand only ifits complement X - Fis open in X. We would like to obtain
a local characterization of closed sets. ln order to do so, we must introduce the concept of a limit point of a set. This concept is important not
only for closed sets, but for ali sets in a topological space.
Definition 3.2.3 Let X be a topological space and let S be a subset of X.
A pointxeXis a limit point of Sifwhenever G is open in Xand xeG, then
(GnS)-{x}#0
The set of ali limit points of S, called the derived set of S, is denoted
by
s.
S = {y, z}
Note that S is not a closed set and that S <t;. S.
Now consider the subset T = {y, z} of X. The element x is nota limit
point of T beca use {x} is an open set containing x which does not contain
a point of T different from x. Furthermore, y is not a limit point of T
because {x, y} is an open set containing y which does not contain a point
68
Topology
t= {z}
Note that Tis closed and that te T.
2. Let R have the natural topology and consider the set Z of integers
as a subset of R. For any xeR we can always find an open interval which
contains x and does not contain any element of Z, except possibly x if
xeZ. Therefore no element of R can be a limit point of Z, and thus
Z=</J
Now consider the set Q of rational numbers as a subset of R. If xeR
and G is any open set containing x, then there is some open interval (a, b)
such that xe(a, b) e G. But then there is a rational number q such that
a< q < x, and hence qe(Q nG) - {x }. Therefore every xeR is a limit
point of Q, and thus
Q=R
3. Let R have the natural topology and let S be a nonempty subset of
R which is bounded above. If sup S rtS and (a, b) is an open interval which
contains sup S, then (a, b) must contain a point of S, for otherwise the
endpoint b would be a least upper bound for S which is less than sup S.
Hence if sup S S then sup S is a limit point of S. A similar result holds
for greatest lower bounds.
4. If R has the natural topology and S =(a, b) is an open interval and
T = [a, b] a closed interval in R, then
S =[a, b]
and
=[a, b]
We leave the proof to the reader. (See Exercise 7 at the end of this
section.) Note that S is not closed and does not contain ali its limit points,
while Tis closed and does contain ali its limit points.
Now we use the concept of limit point to obtain a local characterization of closed sets. The characterization says that a set is closed if and only
if it contains ali its limit points; it will appear as a corollary to the next
proposition.
69
Proof: If F contains ali its limit points, then F =Fui, which is closed
by the proposition. Conversely, suppose F is closed but does not contain
all its limit points. ln this case there is some x E X such that x F and x
is a limit point of F. But then xeX - F, which is an open set containing no point of F, and this contradicts the fact that x is a limit point
of F. D
Now that we have considered open sets and closed sets, what about
sets which are neither open nor closed? For example, consider the halfopen interval (a, b] in R. This set, which is neither open nor closed in the
natural topology, contains a "largest open set," namely (a, b), and is
contained in a "smallest closed set," namely [a, b]. Thus if we wished we
could study (a, b] by analyzing the open set (a, b), the closed set [a, b], and
the "boundary set" {a, b }. This sort of dissection of a set can be done in
any topological space, provided we define "largest open set," "smallest
closed set," and "boundary set" properly.
Definition 3.2.7
1.
70
Topology
The remarks preceding the definition show that for any subset S of a
topological space X, the interior, closure, and boundary of S exist and are
unique.
Examples 3.2.9
T= {y}
and
V= {x, z}
S=S
S=X
and
bd S= {y, z}
T=f/J
f'= {y, z}
and
bd T= {y, z}
V= {x}
V=X
and
bd V= {y, z}
S= {x}
We have
2. Let R have the natural topology and let S be any of the intervals
(a, b), [a, b], (a, b], or [a, b); then
S =(a, b)
S= [a, b]
and
bd S= {a, b}
71
Z=0
Z=Z
and
bdZ=Z
Q=0
Q=R
and
bdQ=R
and
1.
2.
3.
ScS cS.
S is open in X if and only if S = S.
S is closed in X if and only if S = S.
Corollary 3.2.11
l.
2.
3.
Snbd S =Sn(S-S) =0
Conversely, if S n bd S = 0, then S n (S - S) = 0, and since S - S is
contained in S - S, it follows that S - S = 0. But this implies that
S = S and thus that S is open. O
Definition 3.2.12 Let X be a topological space and let S be a subset of X.
If S = X, we say that S is dense in X.
Examples 3.2.13 1. Let X= {x, y, z} and r = {0, {x }, {x, y }, X}. Then
{x} is dense in X but {y} is not dense in X.
Topology
72
1.
2.
3.
Proof: We prove (1) and (2), leaving (3) as an exercise. (See Exercise 18
at the end of this section.)
Let us put T = {xeS 1xeG for some open set G e S}: then in arder
prove ( l) we must show that S = T. If .S = </J, then the only o pen set
contained in S is </J, and T must be empty also. On the other hand, if T = <b
then S cannot contain a nonempty open set and hence S = </J. Therefore
( 1) holds if either S or Tis empty, so we assume that these sets are not
empty.
If x e S then x e T, since S is an o pen set contained in S. Therefore
S e T. Conversely, if x e T, then there is some o pen set G such that x e G
and G e S. But then G e S, because S is the largest open set contained
in S. Therefore Te S, and we have proved (1).
Now we prove (2). Since Se S, every limit point of S is also a limit
point of S. (See Exercise 8 at the end of this section.) Since S is closed, it
contains ali its limit points. Therefore S contains ali the limit points of S,
and it follows that (S u S) e S. On the other hand, since S u S is a closed
set containing S, we must have Se (S u S), and we are done. O
la.eA
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
73
74
Topology
(a) Prove that (S n T) = Sn T.
(b) Prove that Su T = Su T.
(e) Give examples to show that in general (Su T) # Su T and
SnT#SnT.
21.
4
Continuous Functions
CONTINUITY
Continuous Functions
76
y
Figure 4.1
f((a,b)) = [c,d)
The continuous image of an open interval need not be an open set.
77
Continuity
in the natural topology, namely, the open intervals? As Figure 4.1 shows,
a function from R to R which is continuous according to our intuitive
definition need not carry open intervals onto open intervals, or even onto
open sets. Thus in Figure 4.1 the image under the continuous function f
of the open interval (a, b) is the half-open interval [e, d). Hence we
may conclude that the continuous image of an open set is not always an
open set.
Now suppose we consider preimages under continuous functions. If
we begin with an open set in the range of a continuous function and take
its preimage, we will obtain a set in the domain of the function. Figure 4.2
shows a continuous function f from R to R. Note that in the figure the
preimage under f of the open interval (a, b) in the range is a union ofopen
intervals, and hence an open set, in the domain. This is always so for a
continuous function from R to R: under such a function, the preimage of
an open set is an open set. On the other hand, if a function from R to R
is not continuous, then there will be some open set in its range whose
preimage under the function is not open in its domain. For example,
Figure 4.3 shows a function g from R to R which is not continuous: note
that the preimage under g of the open interval (a, b) is not an open set.
The above remarks suggest that if we want continuity in topological
spaces to be a generalization of our intuitive notion of continuity on R,
.............
--?:
1
1
X
Figure 4.3 The preimage of some open interval under a discontinuous function
will not be an open set.
Continuous Functions
78
f(x) = tX
f(y) = /(z) =
and
g(x) =
g(z)
tX
g(y) =
We have
and
Therefore the preimage under f of every open set in Y is an open set in X,
so f is a continuous function from X to Y. For g, on the other hand,
and
g- 1({tX})={x,z}
and since {tX} is open in Y but its preimage under g is not open in X, g is
a discontinuous function from X to Y.
2. Continuity depends on the topology as well as on the function. A
function which is continuous for one topology may not be continuous for
another. For instance, suppose we define X, Y, u,f, and g as in Example
1 above, but change the topology on X by setting r = {0, {x, z }, X};
then f is a discontinuous function from X to Y and g a continuous one.
( Check this.)
3. Every constant function is continuous. To see this, suppose X and
Yare topological spaces and/: X-+ Yis given by f(x) =e for all xeX. If
Continuity
79
if CEG
Thereforef- 1(G) is always open in X, sofis continuous.
4. Let /: R -+ R be given by f(x) = x + 1 for ali x E R. Let us show
that f is continuous. It will suffice to show that under f the preimage of
every open interval is an open set in the natural topology. (Why?) But this
isso, for if (a, b) is an open intervalf- 1((a, b)) =(a - l, b - 1).
5. Let XQ: R-+ R be defined as follows:
if XEQ
ifxeR-Q
The function XQ is called the characteristic function of the rationals. It is
discontinuous because every open interval (a, b) contains both rational
and irrational numbers and hence
if x E [0, l]
if x ~ [0, l]
Thenfis discontinuous, sincef- 1((!, ~)) =[O, l], which is not an open set.
However, if we restrict f to the subset [O, l] supplied with the subspace
topology induced by the natural topology, the restriction of f to [O, l] is
a constant function, which is continuous by virtue of Example 3 above.
The definition of a continuous function from one topological space to
another is given in terms of open sets, but of course there must be an
equivalent formulation in terms of closed sets.
Proposition 4.1.3 Let X and Y be topological spaces. A function
/: X-+ Y is continuous if and only if whenever F is a closed set in Y its
preimage 1- 1(F) is a closed set in X.
Continuous Functions
80
1(F))
F= F
But f(S) e F implies that S cf- 1(F) = S. We have thus shown that
Se S, and since S e S always holds, we conclude that S = S. Therefore
S is a closed set in X. D
Continuity
81
where x ranges over ali elements of1- 1(G). This shows thatf- 1(G) is a
union of open sets in X and hence is itself open in X, and therefore that
f is continuous. D
Corollary 4.1.6 Let S be a subset of R and let S have the subspace
topology induced by the natural topology on R. A function f: S-+ R is
continuous if and only if it satisfies the following property for ali x E S:
82
Continuous Functions
1. Let
X= {x, y, z}
r = {0, {x }, X}.
and
Let
f(y) =
and
f(z) =y
ifx >O
if X
~0
1((~. 3;))={l}c(c,d)
A similar argument holds if x < O.
4. Let XQ be the characteristic function of the rationals. Thus
if XE Q
ifxeR-Q
83
Continuity
Since every open interval (a, b) contains both rational and irrational numbers, it follows that
XQ((a, b)) = {O, 1}
for every open interval (a, b). Therefore if x is rational, so that xQ(x) = 1,
then there can be no open interval (a, b) such that
xQ((a, b)) e(!,~)
-!, !)
Proposion 4.1.9 Let X, Y, and Wbe topological spaces and letf: X-+ Y
and g: Y-+ W. Iff is continuous at xe X and g is continuous atf(x) e Y,
then g f is continuous at x.
Proof: Let G be an open set in W such that (g f)(x) is an element of G.
Since g is continuous at f(x), there is some open set H in Y such that
f(x) e H and g(H) e G. Since f is continuous at x, there is some open
set K in X such that x e K and f(K) e H. But then K is open, x e K, and
\Since
(g f)(K) = g(f(K)) e g(H) e G
0
we have g f continuous at x. O
0
Continuous Functions
84
1.
2.
3.
4.
5.
Proposition 4.1.12
1.
2.
3.
4.
The
The
The
The
Proof: We prove (1) and (2), leaving (3) and (4) as exercises. (See
Exercises 9 and 10 at the end of this section.)
To prove ( 1) it will suffice to show that if (rf)(x) = r f(x) is an
element of an open interval (e, d), then there is some open interval (a, b)
such that xe (a, b) and (rf)(S n(a, b)) e (e, d). (Why?) lf r =O, then
(rf)(x) =O, and if Oe (e, d), we may take (a, b) to be any open interval
containing x and obtain (rf)(S n (a, b)) ={O} e (e, d). Thus ( 1) is true if
r=O.
Now suppose that r >O. lf r -f(x) E (e, d), then/(x) E (c/r, d/r), and
since f is continuous at x, there is some open interval (a, b) such that
x E (a, b) and f(S n ((a, b)) e (e /r, d/r). But this in tum implies that
(rf)(S n (a, b)) e (e, d)
Continuity
85
Hence (1) holds if r >O, and a similar argument proves the case where
r <0.
such that
Ct
<f(x) <d1
and
Therefore f + g is continuous at x.
Now consider f - g: by part ( 1) of the proposition, ( - l)g is continuous at x, and therefore by what we have just proved, so is
f + ( - l)g = f - g. D
Let S be any subset of R and let S have the subspace
topology induced by the natural topology on R. Iff: S-+R and g: S-+R
are continuous functions, then
Corollary 4.1.13
86
3.
4.
Continuous Functions
The function fg: S-+ R is continuous;
The functionf/g: S-+R is continuous if g(x) #O for ali
xE
S.
Definition 4.1.14 A functionf from a topological space X to a topological space Y is a homeomorphism from X to Y if it is continuous and its
inverse function J- 1 exists and is a continuous function from Y to X. If
there is a homeomorphism from X to Y, we say that X is homeomorphic
to Y.
Implicit in the definition of a homeomorphism is the fact that both the
function and its inverse must be one-to-one and onto. Furthermore,
if a function is a homeomorphism from X to Y, then its inverse is a
homeomorphism from Y to X. Note also that if X and Y are homeomorphic topological spaces, then they are equivalent sets as defined in Chapter 1, Section 1.3. Therefore sets which are not equivalent cannot be
homeomorphic, no matter how they are topologized. For instance, R and
Q cannot be homeomorphic because they are not equivalent. However,
equivalent sets may not be homeomorphic even if they are topologized in
what appear to be similar ways. For instance, the results of the next
section will allow us to prove that R with the natural topology and [O, l]
with the subspace topology induced by the natural topology are not
homeomorphic even though they are equivalent sets. (See Exercise 15,
Section 4.2 of this chapter.)
Our final proposition in this section compares the topologies on
homeomorphic spaces.
Proof:
Continuity
87
The import of Proposition 4.1.15 is that if X and Y are homeomorphic topological spaces, then there is a one-to-one correspondence between their open sets (and between their closed sets). Thus homeomorphic
spaces are topologically indistinguishable from one another, being topologically identical in much the sarne way that equivalent sets are set-theoretically identical.
EXERCISES
1.
2.
3.
4.
5.
if xe S
ifxeX-S
Continuous Functions
88
6.
7.
10.
11.
12.
13.
14.
15.
16.
= f(g
- g(x))
+ g(x)(f -
/(x))
89
Continuity
17.
18.
f(x) = {
22.
23.
-1
1
ifx <O
ifx ~o
=X
by py((x, y))
=y
and
90
Continuous Functions
(The functions Px and Py are the projections of the plane onto the
x- and y-axes, respectively.) Prove that
1
1
1
1
1
e
1
------,-1
X
a
f((a,b))
= [e,+ :xi)
Figure 4.4 The continuous image of an open interval need not be an interval.
91
2. Let X= {x,y, z} and r = {0, {x}, {x,y}, {x, z}, X}. Let
S = {y, z}. The subset S is disconnected in X, since the sets G = {x, y} and
H = { x, z} are open sets in X such that G n S and H n S are nonempty
and disjoint and
S = (G nS) u(H nS)
(Note that the sets G and H are not disjoint in X: the definition only
requires that G n S and H n S be disjoint, not that G and H be disjoint.)
S is disconnected in X
92
Continuous Functions
Now let T = {x, y }; then Tis connected in X, for if
T= (G n T) u(H n T)
.j2.)
and
H=(fi, +oo)
Then G and H are open sets, G n Q and H n Q are nonempty and disjoint,
and
Q = (GnQ)u(HnQ).
4. The empty set cannot be disconnected (why not?) and therefore is
a connected subset of every topological space.
We will soon prove that intervals and rays are connected subsets of R,
and that in fact these are the only nonempty connected subsets of R.
Our first proposition concerning the phenomenon of connectedness
provides us with an alternate definition of disconnectedness, one which is
often easier to use than that given in Definition 4.2.1.
Proposition 4.2.3 Let X be a topological space. A subset S of X is disconnected in X if and only if S contains a nonempty proper subset T such that
Tis both open and closed in the subspace topology induced on S by the
topology on X.
Proof:
for some sets G and H such that G, H are open in X and G n S, H n Sare
nonempty and disjoint. Let T = G n S. Then T is a nonempty proper
subset of S, T is open in the subspace topology on S because it is the
intersection of S with the open set G, and Tis closed in the subspace
topology on S because its complement in S is H n S, which is also open
in the subspace topology on S.
93
94
Continuous Functions
y
f(x)
1
1
- - i1- - - 1
f(y)
1
1
1
_ _ _ ,_ _
1
---r-----
1
1
(-oo, t) and (t, +oo) are open in R and (-oo, t)nf(S), (t, +oo)nf(S)
are disjoint sets which are nonempty because f(x) is an element of one of
them andf(y) is an element of the other. If t ~f(S), we have
f(S) = (( - oo, t) nf(S)) u ((t, + oo) nf(S))
J2
J2.
95
S={xeJlx:s:;;z<y}
Then x e S, so S is nonempty, and since S is bounded above by y we
conclude that s = sup S exists. Note that x :s:;; s :s:;; y, so se/, and hence
either s eJ or se/ - J. We shall show that neither of these cases can occur,
and therefore that I must be connected.
Suppose se/ - J: Se J implies Se J, and since s = sup Ses, we
must have s eJ. But J is closed in the subspace topology on /, so se/ and
s eJ together imply that s eJ. Therefore we have s ef:J and s eJ, which is
impossible.
Now suppose that s eJ: then s ef: I - J, and thus s =F y, so s < y. Consider the interval (s, y] e/. If J n (s, y] is not empty, then s =F sup S; hence
J n (s, y] = f/J and this implies that (s, y] e (/ - J). Since s is a limit point
of (s, y], it is therefore a limit point of I - J, and since I - J is a closed set,
it follows that se/ - J. Thus we have sef:/ -J and se/ - J, and we have
shown that I is connected.
To finish the proof we must demonstrate that if I is a nonempty
connected subset of R, then it must be an interval ora ray. Ifwe can find
xel, yel, and zeR - / such that x < z <y, then
/=((-oo,z)n/)u((z, +oo)n/)
is disconnected in R, a contradiction. Therefore we conclude that the
connected set I has the property that whenever x e/, y e/ and x <y, then
[x, y] e /. But any subset of R which has this property is either an interval
ora ray. (See Exercise 7 of Section 2.1, Chapter 2.) D
Corollary 4.2. 7 The set R of real numbers contains no nonempty proper
subset which is both open and closed in the natural topology on R.
96
Continuous Functions
Se
U G,,_
a.e A
U Gp
PeB
97
U (-n,, n
1)
1 ~i~k
But if m = max {n1o ... , nd this would imply that R = ( -m, m), which is
impossible. Since we have produced an open covering of R which has no
finite subcovering, it follows that R is not compact.
4. Open intervals are not compact in R: an open covering of (a, b)
which has no finite subcovering is {(a+ (l/n), b) 1nen}. Similarly, rays
are not compact in R. We leave the proofs of these facts as an exercise.
(See Exercise 11 at the end of this section.)
We will soon show that every closed interval is compact in R. Before
doing so, however, let us prove the important fact that continuous functions preserve compactness.
Proposition 4.2.11 Let f be a continuous function from a topological
space X to a topological space Y. If S is a compact subset of X, then/(S)
is a compact subset of Y.
u-
But then
Se
LJ 1- 1(Gix)
1 ~i~n
implies that
/(S) e
LJ
G"',
1 ~i~n
Therefore the open covering {G"' 1 tX e A} for /(S) has a finite subcovering,
namely {G"'' ... , G"'J. O
Now we show that closed intervals are compact subsets of R.
98
Continuous Functions
is a finite subcovering for [a, y], and this implies that y eS, contradicting
the fact that s = sup S. Therefore s < b is impossible, so s = b, and
Example 4.2.13 Let R have the open ray topology. The interval [O, l] is
neither open nor closed in the open ray topology (why?), but it is compact. To see this, suppose that {G" 1cxeA} is an open covering for [O, l]
99
( 1 )
( 1 )
r~~~~~J\-.~~~-..
'
100
Continuous Functions
LJ
Gx
xeS
and because S is compact the open covering {Gx 1 xeS} has a finite
subcovering, say {Gx 1, , GxJ Therefore
and
Note that l 1 ,,.;,,.n Hx, is an open set.
For each i, l ~ i ~ n, Gx, n Hx, = 0 implies that
Hence
We have thus shown that for each y e X - S there exists an open set which
101
inf/(S) exist. Note that inf/(.sj ~f(x) ~ supf(S) for ali xeS. Sincef(S)
is closed in R, it contains ali its limit points, and from this it follows that
supf(S) and inf/(S) are elements of f(S). But this means there exist
elements s and t in S such thatf(s) = inf/(S),f(t) = supf(S), and we are
done. D
Continuous Functions
102
y
f(t)
1
1
1
1
1
1
- - ---1--- --1
f(s)
1
1
1
s
1
1
1
X
Figure 4.8
on S.
Note that in the preceding corollary both closedness and boundedness are necessary. For example, letf(x) = x for ali xeR. This function is
certainly continuous on R, but it attains neither its maximum nor its
minimum on the closed (but unbounded) set R; similarly, it attains neither its maximum nor its minimum on the bounded (but not closed) set
(O, l).
Proposition 4.2.20
103
EXERCISES
1.
2.
3.
4.
5.
6.
7.
8.
9.
f(x)
10.
11.
12.
=X.
104
Continuous Functions
13.
14.
15.
16.
17.
18.
19.
20.
=..::0
O.
105
YI 1 y e E}
for ali
XER
YI 1yeF}
for ali
XER
and
and consider f
- g.)
5
Sequences and Series
ln the previous chapter we utilized continuous functions to study topological spaces, paying particular attention to the space R of real numbers. ln
this chapter we will continue this process, but concentrating our attention
now on the continuous functions known as sequences. The first section of
the chapter examines the properties of sequences in general topological
spaces, while the second considers sequences in the space of real numbers.
The third section is devoted to the special type of real sequence known as
an infinite series, and the fourth examines the relationship between real
sequences and functional limits.
5.1
SEQUENCES
108
Sequences are important for what they tell us about the range space
X, and hence it is common practice to suppress ali mention ofthe function
f and focus attention only on the image (i.e., the terms) of the sequence.
We shall adopt this convention and in addition shall use the notation {xn}
for a sequence. Therefore rather than speaking of
"the sequence f: N-+ X given by f(n) = Xn for ali n eN"
we will refer to
"the sequence {xn} in X"
Our next definition states what convergence means for a sequence in
a general topological space.
Definition 5.1.2 Let {xn} be a sequence in the topological space X and let
xeX. If for every open set G containing x there is some meN such that
Xn eG for ali n ;;?; m, then we say that the sequence converges to x, and that
x is a limit of the sequence. If the sequence has a limit in X, it is a
convergent sequence; if it has no such limit, it is a divergent sequence.
109
Sequences
G
{x 0 } converges to x
Figure 5.1
Convergence of sequences.
110
Corollary 5.1.5' Every convergent sequence in R (with the natural topology) has a unique limit.
Proposition 5.1.4 allows us to use the limit notation in Hausdorff
spaces.
Definition 5.1.6 Let X be a Hausdorff space. lf the sequence {xn} converges to xeX, we write Limn--+co Xn =x.
Next we show that continuous functions preserve the limits of sequences. This does not depend on whether or not the spaces involved are
Hausdorff.
Sequences
111
n--+oo
Lim Xn =X
n--+
oo
112
such that
Xk E (X -
~' X + ~)
and
for l
~k ~n
cause/(x) ~/(S) by assumption. We have thus proved that every open set
which contains f(x) must contain a point of /(S) distinct from f(x) and
this establishes that/(x) is a limit point of /(S). D
= {_
if
if
>0
X ~0
113
Sequences
if X< 0.
Proposition 5.1.11
1.
2.
+ Yn}
n-+oo
n-+ oo
+n-+OO
Lim Yn
n-+ oo
3.
n-+OO
4.
n-+oo
n-+OO
If Yn =I= O for ali n eN and Limn-+ 00 Yn =FO, the sequence {xn/Yn} converges and
00 Xn
. [Xn]
Limn-+
L im
- =
--Limn-+oo Yn
n-+ oo Yn
Proof:
We shall prove ( 1) and part of ( 2), lea ving the rest as an exercise.
114
Lim Xn =X
implies that
n--+oo
implies that
n--+oo
oo
+ Yn is an element
Sequences
115
contains,x there is some meNsuch that xneG for ali n ?;;m. It would be
convenient to have a characterization of convergent sequences which
would allow us to examine a particular sequence and determine whether
or not it is convergent without having to predetermine its limit, if any.
Such a characterization exists for sequences of real numbers, and we will
develop it in the next section.
EXERCISES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
116
12.
Xn =X
n-+
13.
14.
5.2
REAL SEQUENCES
bounded sequence.
= min {X 1, ... , Xm _ ., X
1}
117
Real Sequences
and
b
Since a ~ Xn ~ b for ali n EN, the image {Xn 1n EN} of the sequence is a
bounded subset of R. D
Proposition 5.2.2 tells us that a convergent sequence in R must be
bounded, but of course the converse is not true: a bounded sequence need
not be convergent. (For example, the sequence {( - l)n} is bounded but
not convergent in R.) However, the converse is true for the class of
sequences known as monotonic sequences.
Definition 5.2.3 A sequence {xn} in R is monotonically increasing if
Xn ~ Xn + 1 for ali n EN, and monotonically decreasing if Xn ~ Xn + 1 for ali
n EN. A sequence is referred to as monotonic if it is either monotonically
increasing or monotonically decreasing.
Example 5.2.4 The sequence {2n} is monotonically increasing, the sequence {l/n} is monotonically decreasing, and the constant sequence {l}
is both monotonically increasing and monotonically decreasing; therefore
these are ali monotonic sequences. On the other hand, the sequence
{( - l)n} is neither monotonically increasing nor monotonically decreasing, and hence is not monotonic.
Proposition 5.2.5 Every monotonically increasing sequence in R which is
bounded above converges to its least upper bound and every monotonically decreasing sequence in R which is bounded below converges to its
greatest lower bound.
Proof: Let {xn} be a monotonically increasing sequence which is
bounded above, and let x = sup {xn 1n EN}; then x is either an element or
a limit point of the set {xn 1n EN}. We will show that in either case the
sequence converges to x.
Suppose x is an element of {xn 1 n EN}, so that x = xm for somem EN.
Since x is an upper bound for the set and the sequence is monotonically
increasing, it follows that xn = x for ali n ~ m. But this surely implies that
the sequence converges to x.
Now suppose that x is a limit poiilt of {xn 1 n EN}, and let (a, b) be an
open interval which contains x. Because x is a limit point, (a, b) contains
some element xm of the set, and since x is an upper bound for the
118
oo
Xn =
+oo
and say that the sequence diverges to plus infinity. If {xn} is a monotonically decreasing sequence which is unbounded below, we write
Lim
Xn = -00
n-+oo
Real Sequences
Proposition 5.2.8
1.
119
= sup {xn
1n
~ k}
we have
!, ... } = 4
1 1
}
1
sup { 3,
4, = 3
4, !, ... } = O
t2 = inf {4, !, ... } = O
t 1 = inf {1,
t3 =
120
converge to the sarne limit and that the original sequence {xn} also converges to this limit.
2. Let {xn} = {l, -1, 4, -2, -3, ... }. This sequence is bounded
above, so the sequence {sd exists, and we have
t.
{ sk} = { l,
4, 4, !, !, }
Note that Limk--.oo sk =O. Since {xn} is not bounded below, the sequence
{ tk} does not exist.
Now we are ready to define the concepts of the limit superior and
limit inferior of a real sequence. These concepts will be useful in our
characterization of convergent sequences in R.
2.
tk
k--+00
k-+OO
1
-k
=O
k-+OO
tk
= Lim O= O
k-+oo
Note that Lim sup Xn = Lim inf Xn =O and that {xn} also converges to O.
Real Sequences
121
= sup {( - 1)n 1 n ~ k} = 1
tk
= inf {( - l)n 1n ~ k} = -1
Thus
= Lim 1 = 1
k-+OO
and
Lim inf Xn = Lim
tk
= Lim - 1 = - 1
k-+oo
k-+oo
Note that Lim sup Xn =F Lim inf Xn and that the sequence {xn} diverges.
3. Let {xn} = {l, -1, !, -2, t, -3, ... }. By virtue of Example 2 of
Examples 5.2.9,
Lim sup Xn = Lim sk = O
k-+00
=-
00
4. Consider the sequence {xn} = {( - l)nn }. This sequence is unbounded above and unbounded below, so
Lim sup Xn =
+ oo
and
Lim inf Xn = -
00
+ oo
It is bounded below, however, and since tk = k, for ali keN, the monotonically increasing sequence {td diverges to + oo. Therefore we have
tk
+ oo
00
and
122
Lim SUp Xn
Proof: Suppose first that {xn} is bounded above and l:...:low. It is easy
to check that
t; =
k-+CO
j-+CO
Lim sup Xn
Real Sequences
123
Now we are ready to prove the theorem which characterizes convergent sequences in R. The theorem not only tells us exactly which real
sequences are convergent but also what the limit of a convergent sequence
must be.
Theorem 5.2.13
if and only if
Lim sup Xn
= Lim inf Xn = X
and
Lim inf Xn = Lim
k--+
tk =
sup {tk} =
00
both existas finite real numbers. Note that for ali keN,
and
t ;;:;: t1r
If Limn .... 00 Xn = x < s, then there exists an open interval (a, b) such that
xe(a, b) and b <s. But then there is some meNsuch that xne(a, b) for
ali n ~ m, and therefore
sm
which contradicts the fact that s is a lower bound for the sequence {sd.
Thus we must have x ;;:;: s. A similar argument shows that x ~ t, and hence
124
contains x. Since
x
sk
k~oo
For each of the following real sequences, prove convergence or divergence. If the sequence converges, find its limit.
{n: 1}
(b) {n:1}
(d)
(a)
(e)
3.
{2n-:~2}
Let reR, r ~O. Under what circumstances does the sequence {rn}
converge, and what is its limit? Prove your answers.
Let {Xn} be a real sequence. Show that
Lim
k~oo
4.
(c) {n 2 -n}
2.
{2nn ++ l}
Xn
=O
k~oo
lxn 1 = O
Real Sequences
5.
6.
7.
125
t;: 1}
(c)
t 2n~ 1}
9.
10.
11.
Let {xn} be a real sequence. A sequence {Yn} is said to be a subsequence of {xn} if both of the following conditions hold:
(i) For every m eN, there is some n eN such that Ym = xn;
(ii) If Ym = Xn and Ym+ 1 = xk, then n < k.
Prove that if {xn} has a subsequence which is bounded below by b
then Lim sup Xn ~ b, while if {xn} has no subsequence which is
bounded below, then Lim sup xn = - oo. State and prove a similar
result for Lim inf Xn.
Let {xn} be a bounded real sequence. Let Lim sup Xn = s and
Lim inf xn = t, and let e be any positive real number. Prove:
(a) There are infinitely many values of n for which xn > s - e;
(b) There are infinitely many values of n for which xn < t +e;
(c) Xn < s +e for all but finitely many values of n;
(d) xn > t - e for all but finitely many values of n.
Prove that every bounded real sequence has a convergent subsequence by showing that if {xn} is such a sequence, then it has a
subsequence which converges to Lim sup xn. State and prove a similar result for Lim inf Xn.
Let {xn} and {Yn} be bounded real sequences, with Xn ~Yn for ali
n eN. Show that
Lim sup Xn
Lim sup Yn
Lim inf Xn
Lim inf Yn
and
12.
Let {xn} and {Yn} be bounded real sequences, and consider the
sequence {xn + Yn}
126
+ Ynl
Lim sup Xn
+ Lim sup Yn
Lim inf[xn
+ Ynl
Lim inf Xn
+ Lim inf Yn
and
(b) Give examples to show that equality need not hold in (a).
5.3
INFINITE SERIES
sm =Xi
+ X2 + ... + Xm = L
xk
k-1
The real number Sm is the mth partia/ sum of the sequence {xn} The
sequence
of m th partia! sums of {Xn} is called the infinite series with n th term Xn and
is denoted by
Infinite Series
127
{ - 1, o,
~::"= 1 ( -
I-
n= 1 2n
L - = 1 -2mk= 1 2k
so ~::"= 1 (1 /2n) is the sequence {1 - (1 /2m)}
a. ....}.
~. ~
L Xn=X
n= 1
S2 =I+t>I
S4 = S2 + t + ~ > S2 + 2(~) = S2 + t
S8 = S4 + ! + ~ + t + i > S4 + 4(!) = S4 + t
S 16 = S8 + ~ + + i<; > S8 + 8(1<;) = S 8 + t
128
and so on. Thus the sequence of partia! sums is clearly unbounded, and
hence divergent, and therefore the harmonic series diverges.
Now we prove some results about infinite series. Our first result will
completely characterize the type of series known as a geometric series.
Definition 5.3.5
L arn = a + ar + ar + ar +
n=O
2
lrl
lrl
Proposition 5.3.6 If
< 1, the geometric series
a/( 1 - r). If a =I O and
~ 1, the series diverges.
Proof:
:r.;:o= 0 arn
converges to
ric series is
S = a(l - rm)
m
1-r
and the proposition follows from this fact and the examination of a few
special cases. The details are left to the reader. (See Exercise 1 at the end
of this section.) D
The geometric series are virtually the only important series for which
it is possible to determine at a glance whether or not a particular series is
convergent. Other types of series generally require more sophisticated
tests to determine whether or not they converge, and there are many such
tests, some of them quite specialized. ln the remainder of this section, we
presenta few ofthe more general convergence tests; others are given in the
exercises at the end of the section.
Our first test is one which is freque'ltly useful in establishing the
divergence of a series.
Proposition 5.3.7
If the series
:r.;:o= 1 xn converges,
Proof: If the series :r.;:o= 1 xn converges, then its sequence of partia! sums
{Sm} = {S 1, S2 , S3 , } converges to some real number x. But then the
sequence {Tm} ={O, S 1, S 2 , } also converges to x. (Why?) Since
129
Infinite Series
Xm = sm - Tm for ali meN, we have
m-+oo
m-+oo
m-+OO
m-+OO
Proposition 5.3.7 provides a divergence test for series because it implies that if Limn .... oo Xn #O, then the series ~:'= 1 Xn must diverge. Note
that the converse of the proposition is not true: a series ~n = 1 Xn may
diverge even though Limn .... oo Xn =O, as the example of the harmonic
series shows.
Example 5.3.8 The
series~:'= 1 (n
Lim n + 1 =1
n-+ 00
The next result may also be thought of as providing a test for divergence, though not a particularly easy one to apply. It is included here
because it is needed to prove the more useful comparison test.
The next result, known as the comparison test, is perhaps the most
useful of ali convergence tests.
~:'= 1 Yn
are in-
130
Examples 5.3.11
n-+co
2.
I
> l, then I
If Lim Xn +
n-+oo
Xn
xn converges;
n=l
Xn diverges.
n=1
k=I
Xm+k
Infinite Series
131
converges (by comparison wth the geometric series). But if ~k= 1 Xm+k
converges, so does ~;::'= 1 xn. (See Exercise 3 at the end of this section.)
2. If Limn--+oo (xn+ ifxn) > l, then there is some meN such that
(xn + if xn) > l for ali n ~ m, and thus
Xm < Xm + 1 < Xm + 2 <
l. The series
~;::'= 1 (
~;::'= 1
Xn+t
L" 2n+t/(n + l) L" 2 n
2
Li. m
-- =
im
= im - - =
n--+OO Xn
n--+OO
2n;n
n--+OO n + l
EXERCISES
1.
2.
3.
4.
(a)
00
(-Itn+l
n=I
(b)
n~I ]n
(c)
I 1
n= 1 n.
<d)
3( 5) 1 -
n=I
oo n3
2n
(e)
n~t
(f)
n~I l +~n-1
oo n3
132
5.
~::"= 1
(an/10n), where
ane{O, 1, 2, ... , 9}
then
6.
~::"=
Yn = y, then
00
(Xn
n=I
+ Yn) = X + Y
(b) Prove that if ~::"= 1 xn = x and ris any real number, then
00
n=I
7.
rxn = rx
(a)
Xn
Prove this.
I Xn converges;
> 1, then I Xn diverges.
n=
11
11
< 1, then
n= 1
1
lnfinite Series
133
10. The root test for convergence of a nonnegative series may be stated
as follows:
Let I::'= 1 xn be a series of nonnegative real numbers.
00
( i) If Lim
Xn
converges;
Xn
diverges.
n= t
n-+oo
00
(ii) If Lim
n= t
n-+oo
00
L-;;
n~I
oo
n=ln
11.
(n+l)n
2n+1
Xn
=O.
R-+00
I. (- l)n +
n= 1
1n
+1
n
(-l)n+I lnn
n+1
n=I
n= 1
I.-1n(ln n)
n=2
134
5.4
FUNCTIONAL LIMITS
=X0
if for every open set G containing x 0 there exists an open set H containing
e such thatf(x) eG for ali xeS n H - {e}.
Suppose that Limx-+cf(x) = x 0 The definition says that if we surround the point x 0 with an open set G, then there must be an open set H
about e such thatf carries the elements of S n H (with the possible exception of e, if e eS) into the set G. Therefore by choosing G to be smaller and
smaller, we can force the functional values ofjto approach x 0 more and
more closely. Note that the value of f at e is immaterial, for we never
considerf(c); indeed,fneed not even be defined ate, for we do not require
that e be an element of the domain S off, but only that it be a limit point
of S. (It is necessary to require that e be a limit point of S in order to
ensure that the open set H contains some point ofthe domain S.) We also
remark that use of the phrase "the limit of f as x approaches e" in the
definition is justified, for the limit of f as x approaches e is unique if it
exists. As usual, this depends on the fact that R is a Hausdorff space, and
we leave the proofto the reader. (See Exercise l at the end ofthis section.)
Examples 5.4.2
l. Let f: R
-+
f(x) =
R be given by
{~
if x = l
if X # l
135
Functional Limits
We claim that Limx .... t f(x) = 1. To prove this, we must show that if G is
an open set which contains 1, then there exists an open set H which
contains 1 and such thatf(x) eG for all xeH - { 1}. But sincef(x) = x for
all x =F 1, we may take H = G. Note that Limx .... tf(x) = 1 even though
f(l) =I= 1.
2. Let f: (O, 1) -+R be given by f(x) = 2x + 1 for all xe(O, 1). We
claim that Limx_,J(x) = 2c + 1 for all ce[O, l]. To see this, let G be an
open set containing 2c + 1. There is some open interval (a, b) contained in
G such that 2c + 1 e(a, b). If H =((a - 1)/2, (b - 1)/2), then H is open,
ceH, andf(x)e (a, b) e G for all xe(O, 1) nH - {e}. Note that if e= O
ore= 1, thenf(c) is not defined, but Limx_,J(x) exists nonetheless.
3. Let f: R -+ R be given by
f(x) = { _
ifx >O
if X ~0
neN.
Proof: Suppose Limx_,J(x) = x 0 and let {xn} be a sequence ofpoints of
S which converges to e and such that xn =F e for all n eN. If G is an open
set containing x 0 , there is an open set H containing e such thatf(x) eG for
ali x e S n H - {e}. Since Limn .... 00 Xn = e, there is some m e N such that
Xn eH for all n ~ m. Since Xn =F e for all n eN and since {xn} is a sequence
136
Limx_,J(x) = f(c).
Proof: If f is continuous at e, then for every sequence {xn} in S which
converges to e, we must have {J(xn)} converging to f(c) by Proposition
5.1.7. But then Proposition 5.4.3 shows that Limx_,J(x) = f(c). Conversely, if Limx_,J(x) = f(c), then for every open set G containing f(c)
there is an open set H containing e such that f(x) eG for ali
xeSnH-{c}. But sincef(c)eG, we conclude that for every open set
G containing f(c) there is an open set H containing e such that f(x)
eG for ali xeSnH, and this is the definition of continuity of fat the
point e. D
Functional Limits
137
2.
3.
4.
For all reR, Lim rf(x) exists and Lim rf(x) = r Limf(x);
X-+C
X-+C
X-+C
Lim (/ + g)(x) exists and Lim (/ + g)(x) = Limf(x) + Lim g(x);
X-+C
X-+C
X-+C
X-+C
Lim (fg)(x) exists and Lim (fg)(x) = [Limf(x)][Lim g(x)];
x-+c
x-+c
X-+C
X-+C
If Lim g(x) =F O, then Lim (f/g)(x) exists and
X-+C
X-+C
(L)
Lim
(x) = L~mx .... cf(x)
x-+c g
L1mx .... c g(x)
EXERCI SES
1.
2.
(e) Lim x
X-+I
3.
2+
X
+1
x1
-
6
Metric Spaces
Throughout this text we have been using the real number system R as
both an example of a topological space and as a guide toward further
topological generalization. ln this latter connection, there are several important properties of R which we have not yet considered from a topological point of view, and foremost among these is the property that
between any two points of R there is a well-defined distance. (Recall that
the distance between the real numbers x and y is defined to be lx - yl,
the absolute value of their difference.) ln this chapter we will extend
the concept of distance between points to sets other than R by means
of functions known as metrics. We will see that if it is possible to define
a metric function on a set, then it is possible to topologize the set in
a very interesting and fruitful manner, thus creating a topological
space called a metric space. Metric spaces share many of the properties
of the real number system R and are of considerable importance in
analysis.
6.1
Suppose x and y are real numbers. Then (see Corollary 2.1.4, Section 2.1
of Chapter 2)
139
140
1.
Metric Spaces
The distance between x and y is a nonnegative real number:
lx:-YI ~o
2.
lx-yl =0
3.
if and only if x
=y
lx -yl = IY-xl
4.
lx - YI ~ lx - zl + IY - zl
These properties of distance in R motivate the following definition of a
distance function, or metric, on an arbitrary set M.
Definition 6.1.1
Mif
1.
2.
3.
4.
+ p((y, z))
141
p(x,y) =
-+ R
as follows:
if X# y
ifx =y
The function pisa metric on the Euclidean plane R 2 . We leave the proof
of this fct as an exercise. (See Exercise 2 at the end of this section.) This
metric is known as the product metric on R 2 .
5. Let M = {bounded sequences in R}, and define p: M x M -+R as
follows: if {xn}eM and {Yn}eM, then
p( {xn}, {yn})
= sup
Metric Spaces
142
for ali n e N, and it is easy to check that this in tum implies that
sup {lxn - Ynl 1 n eN} ~ sup {lxn - znl 1 n eN} + sup {IYn - znl 1 n eN}
Thus
Therefore p is indeed a metric on M. The set M of ali bounded sequences
in R is commonly denoted by / 00 , and the metric of this example is called
the supremum metric on / 00
If M is a set, p is a metric on M, and x is a point of M, it is natural
to consider the subset of M consisting of ali points which lie within a
specified distance of x.
P(x, e)
= {y e M 1 p(x, y)
< e}
,,,,_- .....
/
f. ' \
I
\
...--,
I
_,,
'
......
~(X,f.)
Figure 6.1
G
~(X,f.)
143
open bali consists of ali points of M which lie within the interior of the
circle, while the closed bali consists of the interior points together with
those which lie on the circumference of the circle.
Examples 6.1.4 1. Let p be the absolute value metric on R, let x e R, and
let e be a positive real number. We have
Hence for any xeR and any positive real number e, the open bali f3(x, e)
is just the open interval (x - e, x +e). Similarly, the closed bali p(x, e) is
just the closed interval [x - e, x +e].
2. Let M be a set and let p be the discrete metric on M, so that
p(x, y) = 1 if x ~ y in M, and p(x, y) = Oif x = y in M. If e e R, e > 1, we
have
f3(x, e)= {yeM 1p(x,y) <e}= M
Similarly, if e~ 1, then p(x, e)= M, while if O< e< 1, then P(x, e)= {x }.
3. Let p be the product metric on R 2 of Example 4 ofExamples 6.1.2.
Let (h, k) be a point in R 2 , and let e be a positive real number; then
which is the set of ali points in the plane which lie in the interior of the
circle with center at (h, k) and radius e; that is, f3((h, k), e) is the open disc
with center (h, k) and radius e. ~ilarly, the closed bali P((h, k), e) is the
closed disc with center (h, k) and radius e.
The preceding examples suggest a connection between a metric on a
set and a certain topology on the set. For instance, consider the absolute
value metric on R. It is clear that a subset G of R is open in the natural
topology on R if and only if for each xeG there is some positive real
number e such that f3(x, e) = (x - e, x +e) e G. (Why?) Thus we could
use the natural metric on R to define the natural topology on R: we could
define a subset G of R to be open in the natural topology if for every xeG
there is some positive real number e such that f3(x, e) e G. ln this manner,
given any set M with a metric p, we may use p to define a topology on M.
144
Metric Spaces
p
p
145
p(x, z)
But p(x, y) <e implies that y ep(x, e). Therefore P(z, '7) e P(x, e), and we
have proved that every open bali is an open set in the metric topology.
Now we prove that every closed bali if(x, e) is closed in (M, p). If
if(x, e) = M, then surely it is closed, so assume that if(x, e) =F M and consider M - if(x, e). lt will suffice to show that M - if(x, e) is open. If
zeM - if(x, e), then p(x, z) >e. Choose '1 = p(x, z) - e. Again, '1 is a positive real number, and we claim that P(z, '7) e M - if(x, e). This follows
because if yep(z, 'J), then p(y, z) < 'J, and by the triangle inequality,
p(x, z) ~ p(x, y) + p(y, z). Therefore
p(x, y)
146
Metric Spaces
and
p(x, 1) =M
EXERCISES
1.
4.
5.
6.
7.
147
8.
9.
10.
11.
lx-yl
ifl(x, y) = l + lx - y
12.
:r.:=
00
p( {xn}, {Yn})
L lxn - Ynl
n=I
Metric Spaces
148
13. We shall use this exercise to construct the metric space (Rn, Pn)
known as Euclidean n-space. The readeris asked to fill in the details.
Let n eN, and set Rn = {(x1o ... , xn) 1xkeR, 1 ~ k ~ n }. Thus, Rn is
the set of ali n-tuples of real numbers.
(a) Prove the Cauchy-Schwartz inequality for Rn: if (x1o ... , xn)
and (y 1, , Yn) are elements of Rn, then
+ Yk)2
]1/2 [
~ k~1xk2
]1/2 [
+ k~1Yk2
xn) and
]1/2
Hint: Expand
149
00
(e) Prove that if {xn} and {Yn} are elements of 12, then {xn + Yn} is
an element of 12. Next show that the Minkowski inequality
00
n~l (Xn
+ Yn)2
]1/2 [
00
~ n~l Xn 2
]1/2 [
00
+ n~l Yn 2
]1/2
holds in /2.
(d) If {xn}e/2, define the norm of {xn}, denoted by llxn li. as follows:
Prove that
(i) llxnll ~O for all {xn}e/2;
(ii) llxn li =O if and only if Xn =O for all n eN;
(iii) llcxnll =lei llxnll for all {xn}e/2 and ali ceR;
(iv) llxn + Ynll ~ llxnll + llYnll for all {xn}e/2 and {Yn}e/2.
Show that the function p defined on /2 x / 2 by
15.
is a metric on 12.
Let (M, p) be a compact metric space. Prove that M contains a
countable dense subset.
6.2.
150
Metric Spaces
implies
P2(f(x),f(y)) < 6.
151
tions," and when we use Proposition 6.2.l or Corollary 6.2.2 to show that
a function is continuous ata point, we will be doing " - e proofs." The
following examples illustrate the technique of - e proofs.
Examples 6.2.3 1. Let /: R-+ R be given by f(x) = 3x + 2 for ali x eR.
We use Cotollary 6.2.2 to show thatfis continuous at x for ali xeR.
Given any real number e > O, we must find a real number > Osuch
that lx - YI < implies lf(x) - f(y)I <e. But if lx - YI <, then
+ 2) -(3y + 2)1=3lx -
YI < 3
= 3
(i)
=e
xy
lx-yl
--<xy
xy
Metric Spaces
152
From this last inequality we see that we must choose in such a way that
the quotient /xy is less than e. lt will not help us to choose to be a small
positive real number if at the sarne time we force the product xy to be
small, for then the quotient /xy may be very large. To keep this from
happening, we will select a preliminary which forces xy to be bounded
below and then adjust this preliminary to obtain /xy <e.
Suppose we choose ~ x/2. If x and y are in the interval (O, l] and
lx - YI <~ x/2, then x/2 < y < 3x/2 and therefore xy > x(x/2) =
x 2/2. Choosing ~ x/2 thus forces xy to be bounded below by x 2/2. Now
let us see what effect this has on lf(x) - f(y)i: since xy > x 2/2, we have
lx -yl
2
lf(x) -f(y)I = ~ < xy < x2;2 = x2
Clearly, if we now adjust so that it is less than or equal to ex 2 /2 as well
as less than or equal to x/2, then we will have lf(x) - f(y)I <e. Therefore
we let = min {x/2, ex 2/2}, and conclude that if x and y are in (O, l] with
lx - YI <, then xy > x 2/2 and hence
153
implies
lx-yl
--<e
xy
If x is any point of (O, l) such that O< x <, and if y = x/2, then
X
lx-yl=-<
2
Therefore
lx-yl l
--=-<e
xy
X
But this is impossible, for no matter how is chosen, l/x will be arbitrarily
large for x sufficiently near zero. Thus f cannot be uniformly continuous
on (O, l].
3. Let aeR, O< a< l, and let/: [a, l]-+ R be given by f(x) = l/x for
ali x e[a, l]. We claim that f is uniformly continuous on [a, l]. For any
e> O, choose = ea 2 If x and y are points in [a, l], then xy > a 2 ; if also
Metric Spaces
154
lx - YI < = ea 2, then
lf(x) - f(y)I = lx - YI < (ea2) < (ea2) =e
xy
xy
a2
Let = min {(t 1)/2, ... , (tn)/2}. Clearly is a positive real number, and
we claim that if x and y are points of M 1 such that p 1(x, y) <, then
P2(f(x),f(y)) < e.
If xeM1o then xe{J(tk, (tk)/2), for some k, 1 ~ k ~ n, and hence
155
If yeM1 and p 1(x, y) <, then p 1(x, y) < (tk)/2. By the triangle inequality, we have
Jx,
3.
f(x)
4.
{~
ifx ;60
if X =0
Metric Spaces
156
5.
6.
=JX,
7.
clx - YI
6.3
Proof: If x and y are distinct points in a metric space (M, p) and we let
e= p(x, y), then p(x, e/2) and p(y, e/2) are nonempty disjoint open sets
which separate the points x and y. D
Corollary 6.3.2
unique limit.
157
Proposition 6.3.3
M, then
Limn~
11-
Proposition 6.3.3 is often used as the definition of a convergent sequence in a metric space. Similarly, Corollary 6.3.4 is often used as the
definition of a convergent sequence in R.
lt is possible to do more with sequences in metric spaces that it is
in more general topological spaces because in metric spaces there is
an explicit measure of the distance between points. Thus, in a manner of
Metric Spaces
158
Cauchy sequence.
Proof: Suppose {xn} converges to x in the metric space (M, p). By Proposition 6.3.3, for every positive real number e there is somem eN such that
p(x, xn) < e/2 for all n ;;:;: m. But then for ali n ;;:;: m and k ;;:;: m we have
p(xm xk) ~ p(xm x) + p(xk, x) < e/2 + e/2 =e, which shows that {xn} is a
Cauchy sequence. D
If { xn} is a convergent sequence in R, then for every e > O
there is some me N such that lxn - xk 1 < e for ali n ;;:;: m and k ;;:;: m.
Corollary 6.3.8
J2.
J2.
Definition 6.3.10
We have just seen that Q with the absolute value metric is not a
complete metric space, so not all metric spaces are complete. It is a very
159
important fact that R with the absolute value metric is a complete metric
space, and we will prove this. Before we can doso, however, we need the
following. result.
Proposition 6.3.11
Let a= max {lxd, ... , lxml}: for ali neN, lxnl < 1 +a, and hence {xn} is
bounded. D
Let meN and choose reN such that r > max {t, m}. Then r > m and
If is, - xPI ~ e/2 for ali p ~ r, then s, cannot be the supremum for the set
{xn 1n ~ r }; but this is the definition of s,. Therefore there is some
p ~ r > m such that Is, - xPI < e/2. Thus we have p > m and
Metric Spaces
160
Now we are ready to show that the Cauchy sequence {xn} converges
tos. Since {xn} is a Cauchy sequence, there is some meN such that
for ali n
m and k
sequence.
Complete metric spaces have many properties which incomplete ones
do not. For instance, suppose that {if(xm en) 1n eN} is a collection of
closed balis in a metric space such that the real sequence {en} converges to
O and if(xn+ 1, Bn+ 1) e if(xm en) for ali neN. If the metric space is complete, then the intersection of these nested closed balis must consist of
exactly one point, while if it is not complete the intersection may be
empty.
Theorem 6.3.14
Proof: We will show that the sequence {xn} of center points of the closed
balis is a Cauchy sequence in M; since M is complete, this will imply that
161
the sequence {xn} converges to a point x of M, and the point x will be the
unique element of lneN P(xm en).
Let e be a positive real number. Since Limn .... 00 en =O in R, there is
some m eN such that O< en < e/2 for ali n ~ m. But if n ~ m and k ~ m
then P(xm en) e P(xm, em) and P(xk, ek) e P(xm, em), and thus
But then p(x, e) n P(xm en) = 0 for ali n ~ m, and this in tum implies that
p(x, xn) ~e for ali n ~ m, which contradicts the fact that {xn} converges
to x. Hence X E lneN P(xm en).
Finally we show that x is the only element in lneN P(xm en). To see
this, suppose that y is also a point in this intersection; then
for ali n e N. But O ~ p(x, y) < 2en for all n e N and Limn .... 00 en = O together imply that p(x, y) =O and hence that y = x. O
Corollary 6.3.15
162
Metric Spaces
The proofs of Corollaries 6.3.15 and 6.3.16 will be left to the reader.
(See Exercise 7 at the end of this section.)
Example 6.3.17 As was previously noted, the conclusion of Theorem
6.3.14 need not hold if (M, p) is not a complete metric space. As an
example, consider the metric space (O, 1) with the absolute value metric.
The sequence {l/(n + 1)} is a Cauchy sequence in (O, 1) which does not
converge to a point of the space, so the space is not complete. Note that
the collection of closed balis {if(O, l/(n + 1)) 1n eN} satisfies the hypotheses of Theorem 6.3.14, but that the intersection of these closed balls is
empty.
Proof: Since S is bounded, there exists a closed interval [a, b] such that
Se [a, b]. Let x 1 be the midpoint of [a, b] and let 6 1 = (b - a)/2; then
Se [a, b] = if(x 1, 6 1). Because S is infinite, at least one of the intervals
[a, xi], [x 1, b] must contain an infinite subset of S. Choose one of the
intervals [a, x 1], [x 1, b] which does contain an infinite subset of S; if x 2 is
the midpoint of the chosen interval and
6 2).
Note that
Now bisect the closed interval if(x2 , 6 2) and argue as before. ln this manner we obtain for each n eN a closed interval if(xno 6n) which contains an
infinite subset of S and such that 6n = (b - a) /2n and
if(xn + 1o 6n + 1) e if(xno 6n). Therefore by Theorem 6.3.14 there is a unique
point x e lneN if(xno 6n). The point x is a limit point of S, because if G is
any open set containing x then there is some n e N such that if(xno 6n) e G
and if(xno 6n) contains an infinite subset of S. D
Note that the conclusion of the Bolzano-Weierstrass theorem fails to
hold if either of the hypotheses is contravened: thus, a finite subset of R
163
{~}
(b)
{n
(c)
2.
3.
4.
5.
6.
7.
8.
2:
1}
t2: l}
if(xk, e)
l~k~n
164
Metric Spaces
Bn
=O
n--+<X>
13.
14.
Xn
=O for all n ~ m}
If p( {xn}, {Yn}) = sup {lxn - Ynl 1 n eN} for ali {xn} and {Yn} in M,
then (M, p) is a subspace of the metric space / 00 , and hence is itself
a metric space. Show that (M, p) is not complete. (Hint: consider
the sequence {sn} in M, where for each n e N, sn is the sequence
{l, 1/2, ... , I/n, O, O, ... }.)
Let S be a subset of R and let f: S -+ R. A point se S is a fixed point
of fif f(s) = s. Supposef satisfies a Lipschitz condition with positive
165
l.
2.
3.
166
Metric Spaces
implies that {lxn - znl} converges to zero (why?) and hence that
{xn}R{zn}
Note that [{O}]= [{l/n}l = [{an}], where {an} is any Cauchy sequence in
Q which converges to zero.
As the preceding examples suggest, if R is an equivalence relation on
a set S, then either xRy, in which case [x] = [y] and the elements x and
y are representatives of the sarne class, or x.Ry, in which case [x] n [y]
= f/J. Thus an equivalence relation on a set S partitions the set; that is, it
defines a collection of mutually disjoint subsets of S whose union is S.
(The mutually disjoint subsets are the distinct equivalence classes of S
under R.)
167
Definition 6.4.4 Let (M1> p 1) and (M2 , p 2 ) be metric spaces and letfbe a
function from M 1 onto M 2 lf p 1(x, y) = p(f(x),f(y)) for ali xEM1,
yEM1, thenfis called an isometry and M 1 and M 2 are said to be isometric
spaces.
It is easy to show that every isometry is a homeomorphism. (See
Exercise 2 at the end of this section.) Indeed, since an isometry preserves
distances, isometric spaces are identical as metric spaces; hence, if
/: M 1 -+M2 is an isometry, then we may think of M 2 as a "copy" of M 1
Now we are ready to begin the process of completing a metric space.
We wish to prove that every metric space is isometric to adense subset of
a complete metric space. Thus if (M, p) is any metric space, we must
construct from it a complete metric space (M*, p*) and then produce a
functionf: M -+M* such thatfis an isometry from (M, p) to (f(M), p*)
andf(M) is dense in M*. The basic idea of the proof is to construct M*
as a set of equivalence classes of Cauchy sequences in M.
Definition 6.4.5 Let (M, p) be a metric space and let C be the set of ali
Cauchy sequences in M. Define a relation R on C by setting {xn}R{yn} if
the sequence of nonnegative real numbers {p(xm Yn)} converges to zero.
Proposition 6.4.6 The relation R of Definition 6.4.5 is an equivalence
relation on C.
The proof of the Proposition is similar to that of Example 5 of Examples 6.4.2, and is left to the reader. (See Exercise 3 at the end of this
section.)
Definition 6.4.7 Let (M, p) be a metric space and let R be the equivalence
relation of Definition 6.4.5. If C is the set of ali Cauchy sequences in M
and {xn}EC, denote the equivalence class of {xn} under R by [xnl Let M*
denote the set of distinct equivalence classes of C under R, and define
p*: M* x M*-+R as follows: if[xn]EM* and [Yn]EM*, then
P*([xn], [yn]) = Lim p(Xm Yn)
n--+
oo
in R.
Our first task is to show that p*([xn], [yn]) exists and is well-defined.
Proposition 6.4.8
For ali [xn] EM* and [Ynl EM*, p*([xn], [YnD exists and
Metric Spaces
168
its value does not depend on the Cauchy sequences {xn}, {Yn} which are
chosen as representatives for the equivalence classes [xnl, [Ynl
Proof: To show that p *([xn], [yn]) exists for ali [xnl e M* and [Ynl e M*,
it will suffice to show that {p(xm Yn)} is a Cauchy sequence in R. (Why?)
By the triangle inequality in (M, p), we have
and
and
(2)
for ali neN and meN. But because {xn} and {Yn} are Cauchy sequences
in (M, p ), for every positive real number e there is some k e N such that
n ~ k and m ~ k imply that
and
Hence the inequalities (1) and (2) together show that
n-+OO
Also, since p *([xn], [yn]) and p *([sn], [tn]) exist by the first part of this
169
proof, the sequences {p(xm Yn)} and {p(sm tn)} converge in R. By the
triangle inequality,
Thus
n--+oo
n--+oo
n--+
oo
Hence
n--+
oo
n--+oo
Proof: We will show that the triangle inequality holds, leaving the remainder of the proof to the reader. (See Exercise 4 at the end of this
section.)
Let [xn], [yn], and [zn] be elements of M*. By the triangle inequality for
(M,p),
Therefore
oo
n--+
oo
n--+oo
Metric Spaces
170
We have now constructed from the metric space (M, p) a new metric
space (M*, p *) whose points are equivalence classes of Cauchy sequences
in M. We still must show that (M, p) is isometric to a dense subset of
(M*, p*) and that (M*, p*) is a complete metric space. Note that if xeM
then {x} is the constant sequence ali of whose terms are equal to x, and
[x] EM* is the equivalence class of this sequence.
Proposition 6.4.10 The functionf: M --+M* defined by f(x) = [x] for ali
xeM is an isometry from (M, p) onto (f(M), p*), and f(M) is a dense
subset of (M*, p *).
Proof:
~e
Now we are ready to begin the proof that the metric space (M*, p*)
is complete.
Proposition 6.4.11 If {xm *} is a sequence in (M*, p*), then for each
meN there is a Cauchy sequence {xnm 1 neN} in (M, p) such that
Xm * = [xnml and p(xnm' Xkm) < 1/m for ali n EN and k EN.
Proof: For each meN, xm *is a point of M*, and hence is an equivalence class of Cauchy sequences in (M, p); let {xnm} denote a representative of the class which determines xm *, so that Xm * = [xnml Since {Xnm} is
a Cauchy sequence, there is some tm eN such that p(xnm' xkm) < I/m for ali
n ~ t and k ~ tm. Consider the Cauchy sequences {Xnm 1 n EN} and
{xnm n ~ tm}: it is easy to show that these two sequences in (M, p) are
equivalent under the equivalence relation R of Definition 6.4.5 ( see Exercise 5 at the end of this section), and hence they define the sarne point,
namely xm * , in M*. If we now renumber the terms of the sequence
171
{xnm 1n ~ tm} so that its first term is Xin rather than x,mm we obtain a
sequence which satisfies the requirements of the proposition. D
Proposition 6.4.12 Let {xm *} be a Cauchy sequence in (M*, p*), and for
each m eNlet {xnm} be a Cauchy sequence in (M, p) such that Xm * = [xnm1
and p(xnmXkm) < 1/m for ali neN and keN. Then the sequence
{xim 1 meN} is a Cauchy sequence in (M, p).
Proof: The existence ofthe sequences {xnm} is guaranteed by the previous proposition; the sequence {xim 1m eN} consists of the first terms of
ali these sequences. We must prove that {xim} is a Cauchy sequence in
(M,p).
If e is a positive real number, then, since {xm *}is a Cauchy sequence,
there is some Pi eN such that p*(xm *, x, *) < e/3 for ali m ~Pi and t ~Pi
Choose p eN such that p > max {p., 3/e }: then 1/p < e/3 and
p*(xm *, x, *) < e/3 for ali m ~ p and t ~ p. Now,
in R. Since p*(xm, x,) < e/3 for m ~ p and t ~ p, it follows that there must
be q eN such that p(Xnm Xn1) < e/3 for ali n ~ q.
Suppose m ~ p and t ~ p and consider p(xim Xi 1). By the triangle
inequality,
But
1 1 6
p(Xim Xqm) <- <- <3
Metric Spaces
172
We have thus shown that for any positive real number e there is some
k eN such that m ~ k and t ~ k imply p(x 1m, xm,) <e.
The Cauchy sequence {x 1m} in (M, p) defines a point x* = [x 1ml in
M*, and we claim that the Cauchy sequence {xm *} in (M*, p*) converges
to x*. ln arder to prove this, we must show that if e is a positive real
number, then there is some keN such that p*(x, *, x*) <e for t ~ k.
However, by the result proved in the preceding paragraph, there is some
keN such that p(x 1m, xm,) < e/2 for m ~ k and t ~ k. Therefore for t ~ k
we have
p*(x, *, x*) = p*([xm,l, [X1mD = Lim p(xmt X1m) <-28 <
n-+OO
[J
173
used to construct the completion of a metric space can, with some modification, be used to construct R from Q. (Modification is required because
several of our definitions and proofs made use of real numbers, and of
course if we intend to construct R from Q, we cannot use real numbers in
so doing.) Thus, we begin with the number system Q, and assume that the
operations of arithmetic, the order relation <, and absolute value have
been defined in Q. We then define the concept of an open interval in Q and
allow the open intervals to generate the natural topology on Q in the usual
manner. Imitating the proof of Proposition 6.3.3, we show that a sequence
{xn} in the topological space Q converges to x e Q if and only if for every
qeQ, q >O, there is somem eN such that lxn - xi< q for ali n ~ m. Next
we define {xn} to be a Cauchy sequence in Q if for every q e Q, q > O, there
issomemeNsuch that lxn - xkl < qforall n ~ m, k ~ m. Nowitispossible
to define a relation R on the set of ali Cauchy sequences in Q by setting
{xn}R{yn} if the sequence {lxn - Ynl} converges to O in Q; as Example 5
of Examples 6.4.2 shows, this relation is an equivalence relation. Finally,
we define the set of real numbers to be the set of ali equivalence classes of
rational Cauchy sequences under the relation R.
Once the construction of R outlined above has been accomplished, it
is possible to define the usual operations of arithmetic, the order relation
<, and the concepts of convergent sequence and Cauchy sequence in R.
lt then requires only minor modifications of the relevant proofs of this
section to show that Q is isometric to a dense subset of R and that every
Cauchy sequence in R converges. We do not intend to pursue this topic any
further, but the interested reader may consult any one of a number of texts
on advanced real analysis. Incidentally, if R has been constructed both by
means of Dedekind cuts and as equivalence classes of rational Cauchy
sequences, then it becomes necessary to show that the two constructions
yield the sarne number system. This can be done by defining a one-to-one
function from the set of all equivalence classes of rational Cauchy sequences onto the set of ali Dedekind cuts, which preserves the arithmetic
operations and ordering. (An excellent source for the complete details of
the construction ofthe reais by both methods is C. Goffman, Real Analysis,
Holt, Rinehart and Winston.)
EXERCISES
1.
2.
174
3.
4.
5.
6.
Metric Spaces
Prove Proposition 6.4.6.
Prove Proposition 6.4.9.
Prove that the Cauchy sequences {xnm 1neN} and {xnm 1n ~ tm} in
the proof of Proposition 6.4.11 are equivalent.
Let Mbe adense subset ofthe metric space (M1, p 1) and M' be adense
subset of the metric space (M2 , p2 ). Prove that if (M, p 1) is isometric
to (M', p2 ), then (M1, p 1) is isometric to (M2 , p2 ). (Hint: if f is an
isometry from M to M', extendf to M 1 by defining
(c) Show that if {xn} and {Yn} are Cauchy sequences in Q, then sois
{XnYn}
(d) Define multiplication of real numbers ( equivalence classes of
rational Cauchy sequences) as follows: if {xn} and {Yn} are
Cauchy sequences in Q, with equivalence classes [xn] and [Yn],
175
then
7
Sequences of Functions
7.1
ln this section we define two types of convergence for sequences of functions and show how these preserve or fail to preserve certain properties.
We begin with the definition of pointwise convergence for a sequence of
functions.
Definition 7.1.1 Let (M, p) be a metric space, letfbe a function from M
to R, and for each n e N let f,,: M--+ R. We say that the sequence offunc-
tions {fn} converges pointwise to the function f if for each x e M the sequence of real numbers {f,,(x)} converges to the real number f(x).
177
178
Sequences of Functions
Thus {fn} converges pointwise to f if Limn .... ,xJ(xn) = f(x) for all
xeM.
Examples 7.1.2 1. For each n eN, definefn: R-+ R by f,,(x) = x/n for all
xeR (see Figure 7.1). Since Limn--+cxJ(xn) = Limn--+co x/n =O for ali
xeR, the sequence offunctions {x/n} converges pointwise to the function
f defined by f(x) = O for ali x E R.
2. For each neN, definefn: R-+R by f,,(x) =xn for ali xeR (see
Figure 7.2). If x > 1, Limn--+cofix) = Limn--+co xn does not exist, so the
sequence {xn} does not converge pointwise.
3. Pointwise convergence of a. sequence of functions depends on the
metric space as well as on the functions. To see this, letf,,: [O, l] -+R be
given by f,,(x) = xn for ali n eN and x e [O, l] (see Figure 7.3). These are
the sarne functions as in the previous example, but here
Lim f,,(x) = Lim xn =
n-+oo
n-+oo
ifxe[O, 1)
if X= 1
hence the sequence {xn} converges pointwise to the functionf: [O, l] -+R
Figure 7.1
{f,.(x)}
= {x/n}
on R.
179
Sequences of Functions
180
defined by
f(x) =
{~
ifxe[O,l)
if X= 1
4. Let f,,: (O, + oo)-+ R be defined by f,,(x) = 1+x/(1 + nx) for all
n e N and all x > O ( see Figure 7.4). Since
X
l<l+--<l+-=l+l+nx
nx
n
we have
1
H-+00
+ nx)]}
on (O,
+ oo ).
181
= {~
if X =/= 0
ifx =0
Sequences of Functions
182
l
n
---<1 +nx
for ali xe(O, +oo), if we choose meN such that m > 1/e, then n ';i:;m
implies
X
183
lfn(x) - f(x)I
= lxl/n <
for ali x e R. But this cannot be true for ali x e R, for if n = m and x = m,
then lxl/m = 1.
2. Example 2 of Examples 7.1.4 shows that the sequence
{1 + x /( 1 + nx)} converges uniformly on (O, + oo) to the function f
defined by f(x) = 1 for ali xe(O, +oo).
3. Uniform convergence depends on the metric space as well as on the
functions. For instance, let fn: [O, l]--+ R be given by fn(x) = x/n for ali
n e N and ali x e [O, 1]. These are the sarne functions as in Example 1
above, but here the sequence {x/n} converges uniformly to the zero function on [O, l]. To see this, suppose e >O is given and choose m > 1/e. Since
O~ x ~ 1, n ~ m implies that
X
1=-~-~-<e
X
1 1
lfix)-f(x)I= I--0
n
n n m
for ali xe[O, l], and hence {x/n} converges uniformly on [O, l].
4. The sequence of functions {1/(1 + nx 2)} of Example 5, Examples
7.1.2, does not converge uniformly on R to the function f defined by
f(x) =
{~
if X =/=
ifx =O
Sequences of Functions
184
5. Let a eR, a> O, and for each n eN definef,.(x) = 1/( 1 + nx 2) for ali
+ oo ). The sequence {ln} converges uniformly to the zero function
+ oo ), because if me N, m > ( 1 - e)/a 2 , then n ~ m implies
x e [a,
on [a,
lfn(x) -
+ oo ).
the other hand, Examples 1 and 4 of Examples 7.1.6 show that pointwise
convergence does not imply uniform convergence. Thus uniform convergence is a stronger condition than pointwise convergence.
Application of the.definition of uniform convergence requires knowledge ofthe function/which is the limit of the sequence Un} lt would be
convenient to have a characterization of uniformly convergent sequences
which does not utilize the limit functionf The next result, known as the
Cauchy criterion for uniform convergence, provides such a characterization.
Proposition 7.1.7 (Cauchy Criterion) Let (M, p) be a metric space and
let f,.: M--+ R for ali n e N. The sequence Un} converges uniformly to a
function from M to R if and only if for every positive real number e there
is somem eN such that n ~ m and k ~ m imply lf.(x) - ,h(x)I <e for all
xeM.
Proof:
2 + 2 =e
185
for ali xeM. We claim that Un} converges uniformly to this functionf
If e eR, e >O, then there is some m eN such that n:;:::: m and k:;:::: m
imply lfn(x) - .fk(x)I < e/2 for ali xeM. For fixed k, k:;:::: m, and fixed
xeM, consider the sequence {IJ,,(x) - .fk(x)l neN}: since
and
for n:;:::: m
we have
Thus if k:;:::: m, then lf(x) - .fk(x)I <e for ali xeM, which shows that ifn}
converges uniformly to f on M. D
Suppose {fn} is a sequence of functions which converges to f and f,,
has property P for each n e N. Will the limit function f necessarily have
property P? The answer is often no if the convergence is pointwise, but yes
if it is uniform. Thus uniformly convergent sequences of functions are
important because properties shared by the functions ofthe sequence tend
to be preserved in the limit, whereas this is frequently not the case for
pointwise convergence. We will illustrate by showing that the properties
ofboundedness aIJ.d continuity are preserved by uniform convergence, but
not by pointwise convergence.
First we must define what is meant by a bounded function from a
metric space to R.
Definition 7.1.8 If (M, p) is a metric space andfis a function from M
to R, thenf is bounded on M if f(M) is a bounded subset of R.
Proof:
Sequences of Functions
186
such that n ~ m implies IJ(x) - fn(x)I < 1 for all x eM. Since fm is
bounded on M, there is some reR such that lfm(x)I < r for all xeM.
Therefore
IJ(x)I ~ IJ(x) - fm(x)I
+ lfm(x)I < 1 + r
{~
fn(x) =
ifx ~n
if k e N and k - 1 ~ x < k
If x e [O,
We conclude this section by showing that uniform convergence preserves continuity while pointwise convergence need not e!~ so.
Proposition 7.1.11 Let (M, p) be a metric space and let {f,,} be a sequence of continuou~ functions from M to R. If {f,,} converges uniformly
on M to the functionf, thenfis continuous on M.
implies
<3+3+3=e
which proves that the limit function f is continuous at x. O
187
f(x)
= {~
if X =/=
ifx =0
OI
[O, l]
nx
+ oo)
nx
2.
Sequences of Functions
188
4.
Let
5.
6.
7.
8.
9.
if xeQ
if x~Q
11.
189
(See Exercise 10, Section 5.2, Chapter 5.) Similarly, the sequence
{!1m(x2 )} is boqnded and therefore has a convergent subsequence
U2m(X2)} = {/21(X2),J;2(X2), .}
Sequences of Functions
190
Definition 7.2.1
llf
11
11
>O;
M to R;
The proof of Proposition 7.2.2 will be left to the reader. ( See Exercise
l at the end of this section.)
Now suppose that for all/eB(M) and geB(M), we define
p*(f,g) = llJ-gll
then p* will be a metric on B(M). ln fact, not only is (B(M), p*) a metric
space, it is a complete metric space.
191
to an element of B(M).
If Un} is a Cauchy sequence in B(M), then for every positive real
number e there is some me N such that n ~ m and k ~ m imply
But
192
Sequences of Functions
metric spaces. We will not pursue this topic any further, but merely
remark that function spaces are of considerable importance in higher
analysis.
Now we tum our attention to the uniform approximation of functions. Suppose Un} is a sequence of functions which converges uniformly to f on M. Then for any e e R, e > O, there is some m e N such that
n ~ m implies lf(x) - fn(x) 1 <e for ali x eM. Therefore by making e
small enough we can find a function f m which is arbi trarily elo se to f on
M. For this reason we say that the limit function f is uniformly approximated by the functions Un}. There are many approximation theorems of
the following form: let f be a function ha ving property P; then f can be
uniformly approximated by functions having property P'. This is the
sarne as saying that if f has property P, then there is a sequence of
functions having property P' which converges uniformly to f We will
prove two important approximation theorems of this type for continuous functions on R.
Definition 7.2.6 Let S be a subset of R and letfbe a function from S to
R. We say that f is a step function on S if S is a finite union of mutually
if X~ -1
if-l<x<!
1
2
l 2~X ~ 1
if X> 1
Thenf is a step function on R. See Figure 7.6.
Our first approximation theorem says that a function which is continuous on a closed interval can be uniformly approximated there by
step functions.
Proposition 7.2.8 If/:[a, b]-+ R is continuous, then f can be uniformly
approximated on [a, b] by step functions.
Proof: We must produce a sequence {f,.} of step functions which converges uniformly to f on [a, b].
193
-1
-1
-2
-3
b-a)
[ a, a +-n-
b-a
2(b-a))
l n2= [ a +
--,a+--n
n
l nn
=[
a+
(n - I)(b-a)
'
b]
Then for each n e N, [a, b] = Ui . _ k .._ n Ink and the intervals Ink are mutually
disjoint. For each k, 1 ~ k ~ n, let Xnk denote the midpoint of the interval
Ink For each neN, define/,,: [a, b]-+R by
if X E [nk> 1 ~ k
Sequences of Functions
194
y
See Figure 7.7. Clearly each f,, is a step function on [a, b]. We will show
that the sequence {/,,} converges uniformly to f on [a, b].
Suppose e is a positive real number. Since f is continuous on the
compact set [a, b] it is uniformly continuous there, and hence there exists
a positive real number such that whenever x and y are in [a, b] with
lx -yl <
<>
then
lf(x) -f(y)I <e
Choose meNsuch that (b - a)/m <. For n;;:;: m, xElnk and yelnk imply
b-a
n
b-a
m
lx - YI < - - ~ - - <
<>
195
n ! = n(n - 1)(n - 2) 1
O!= 1
(n) =
k
for O ~ k
n!
k!(n -k)!
L
n
k=O
(n)
k
( k)2
xk(I -x)n-k X - -
1
~4n
The proof is left to the reader. (See Exercise 9 at the end of this
section.)
Next we define the polynomial functions which we will use to approximate a given continuous function.
Definition 7.2.10
Let /: [O, l]
-+ R.
-196
Sequences of Functions
n = 1: Bn
n = 2: Bn
G)
(1-x)O +xi =x
x 0 (1 - x) 1/(0) +
C)
x 1(1 - x)f(l)
+G)x 2(1-x)f(l)
= (1 -
x) 20 + 2x(1 - x)! + x 2 I = 4x + !x 2
1 (n - 1)
Bn =;;x + -n- x2
Proof: We first prove the theorem for the case where [a, b] =[O, l].
Let /: [O, l]-+ R be continuous. We will show that the sequence of
Bernstein polynomials for f converges uniformly to f on [O, l]. Since
it will suffice to show that if e is a positive real number, then there is some
me N such that n ;;:;: m implies
197
kto
(~)xk(I-x)n-kk(x)-1(~)1
=
-x)n-qk(x)
-1(~)1
where p ranges over ali values of k such that lx - (k/n)I <, and q ranges
over all values of k such that lx - (k/n)I ~. Because lx - (p/n)I < for
each p, we have ll(x) - l(p/n)I < e/2 for each p, and thus
k~O k
+ (1-x))n = 1
Since ll(x)I < r for all xe[O, l], we have ll(x) -l(q/n)I < 2r, and hence
L (n) xq( 1 - x)n - qll(x) - j fj_)I < 2r L (n) xq(I - x)n-q (x - q/n) 2
q q
J\.n
q q
(x-q/n) 2
198
Sequences of Functions
(2r) 7~ (n)
2
(2') _r
and r/2n 2 < e/2 if l/n < 2e/r. Thus if we choose meN such that
1/m < min {, 2e/r}, what we have done shows that neN, n ~ m implies
Therefore the theorem is proved for the case where [a, b] =[O, l].
Now suppose /: [a, b] -+ R is continuous. Define
h(x) = f(a
+ (b -
a)x)
for ali x e [O, l]. Then h is a continuous function on [O, l], since it is the
composite ofthe continuous function g(x) =a+ (b - a)x with the continuous function f By what we have already shown, for every positive real
number e there is some polynomial P(x) such that lh(x) - P(x) 1 < e for ali
xe[O, l]. But then
p
(X -a)
b-a
is a polynomial and
2.
3.
199
6.
7.
Prove that (C(a, b), p*) is a complete metric space. (Hint: use the
result of Exercise 3 of Section 6.3.)
Find a sequence of step functions which converges uniformly to f on
[O, l] if
(a) f(x) = x for ali xe[O, l];
(b) f(x) =x 2 for ali xe[O, l].
A function g: [a, b]-+ Ris piecewise linear on [a, b] if there exist points
t0 , t1o . .. , tn and real numbers r0 , r1o ... , rno s0 , s1o . .. , sn such that
and
8.
9.
k=O
and show that the three terms on the right-hand side reduce to x 2,
-2x 2 , and x/n + [(n - l)x 2]/n, respectively.)
8
Calculus
ln this chapter we will study the basic concepts of calculus, the definite
integral and the derivative. We will not be concerned with techniques of
integration and differentiation, but rather with the theory of integration
and differentiation. We begin with a section devoted to the definition of
the Riemann integral and integrable functions, proceed to consider the
properties of the Riemann integral, and conclude with an examination of
the derivative and its properties.
8.1
ln this section we will define the integral of calculus, which is known as the
Riemann integral, and discuss briefly the question of the Riemann integrability of certain functions. Before we can define the integral, we need
some preliminary results about partitions of intervals and Riemann sums.
201
Xn
= b
Calculus
202
are called subintervals of the partition P. If P and P' are both partitions
of [a, b] and P is a proper subset of P', we say that P' is a refinement
of P.
Example 8.1.2
and
, {o'4'2'
I I I i}
J2'
p =
The sets P, P', and P" are ali partitions of [O, l]. The partition P' is a
refinement of P and P" is also a refinement of P; however, P' is not a
refinement of P", nor is P" a refinement of P'.
Now we define the upper and lower Riemann sums of a bounded
function over a partition.
Definition 8.1.3
={a= x0 , X1o
, Xn
= b}
U(f,P)=
k=I
L( f, P) =
L
k=l
203
Figure 8.1
Note that sincef is bounded on [a, b] both U(f, P) and L(f, P) exist
as finite real numbers for any partition P of [a, b]. Furthermore, it is clear
from the definition that L(f, P) ~ U(f, P) for any partition P of [a, b].
Iff is a continuous function on [a, b], then U(f, P) is the sum of the
areas of the rectangles whose bases are the subintervals [xk _ 1, xk] and
whose heights are equal to the maximum values which f attains on the
subintervals (see Figure 8.1). Similarly, if f is continuous on [a, b], then
L(f, P) is the sum of the areas of the rectangles whose bases are the
subintervals [xk _ 1, xk] and whose heights are equal to the minimum values whichf attains on the subintervals (see Figure 8.2). Notice, however,
that Definition "8.1.3 does not require thatf be continuous on [a, b], but
only that it be bounded there.
Examples 8.1.4 1. Let e eR and letfbe the constant function defined by
f(x) =e for ali xe[a, b]. If P ={a= x0 , x., ... , Xn = b} is any partition
of [a, b], then
n
U(f, P) =
c(xk -xk_ 1)
c(xk...- xk_ 1)
k=I
and
n
L(f, P) =
L
k=I
204
Calculus
y
b
L(f,P) is the sum of the areas of the rectangles
2. Let f: [a, b] -+R be given by f(x) = x for all xe[a, b]. For each
Then
U(f, Pn) =
f
k=I
b- a
=--na+
n
(b - a) 2
= b- a
k=I
k=I
~2a)2 n(n; 1)
1
1
=-(b 2 -a 2) +-(b-a) 2
2n
a+ (b -2a)2
k
n
k=I
205
Similarly,
ifO ~x < !
{~
if!~x ~ 1
U(f, P) =
O(xk -xk_ 1)
+L
l(xk -xk-1)
k=i
k=l
=Xn -X;-1 =
-X;-1
L(f, P)
O(xk - xk_ 1)
+ L
l(xk - xk_ 1)
k=j+l
k=l
= Xn - xi = 1 - xi
For any partition P of [O, l] we will have L(f, P) =O, because every
subinterval of P must contain a point x such that f(x) =O. Now let us
consider upper Riemann sums. We claim that for every positive real
number e we can find a partition P. of [O, l] such that O< U(f, PJ <e.
We may assume that e< 1, so that there is some meN such that
1
e 1
0<--<-<m+ 1 2 m
Calculus
206
It is possible to choose eR, >O, such that <e/( 4m) and also such
that
for l
m unless n = k
(Why?) Let
el
l
l
l
}
p = { o, -2 -m - , -m + , - . - , - . + , ... ' 1 - , l
mmWe have
sup {/(x) 1x e [
~ - , ~ + ]} = ~
~k ~m
=--.1
m
e
-2+
(ml)
L -k 2 + l
k=2
<-+
L - =-+-=e
2 k= 1 2m 2 2
5. Let x denote the characteristic function of the rationals restricted
to [O, l], so that
x(x) =
{~
if x e [O, l] is rational
if x e [O, l] is irrational
and
inf {x(x)} = O
207
L(f, P)
L(f, P')
U(f, P')
U(f, P)
Proof: We have already observed that, as a consequence of the definition of upper and lower Riemann sums, the lower Riemann sum over a
given partition is less than or equal to the upper Riemann sum over the
sarne partition. Therefore
L(f, P')
U(f, P')
and
L(f, P')
U(f, P')
U(f, P)
Let us first consider the special case in which P' has been obtained by
adding one point to P; say
P ={a= x0 ,
X1o , Xn
= b}
and
P'
={a= x0 , x 1,
, X;-1'
y, X;,
. . . , Xn
= b}
lt is clear that
and
sup{f(x) lxe[y,x;]}~sup{f(x) lxe[x;_ 1,xJ}
Calculus
208
Hence
n
U( f, P') =
k=I
k-Fi
+ sup {f(x)
+ sup {f(x)
k=I
k"" i
+ sup {f(x)
+ sup {f(x)
k=I
= U(f, P)
U(f, P)
and
L(f, P)
L(f, P')
We leave the details of the induction argument to the reader. (See Exercise 3 at the end of this section.) D
L(f, P u P')
U(f, P u P')
U(f, P') D
Thus for a given bounded functionf, every lower Riemann sum is less
209
is bounded below and therefore has a greatest lower bound, and that
{L(f, P) 1 Pisa partition of [a, b]}
sup L(f, P)
p
and
inf U(f, P) = inf {U(f, P) 1P is a partition of [a, b]}
p
sup L(f, P)
p
inf U(f, P)
p
sup L(f, P)
p
Proof:
inf U(f, P)
p
If
But then, again by virtue of Definition 8.1. 7, there is some partition P'
such that
L(f, P) > U(f, P')
Calculus
210
Since
sup L(f, P)
p
inf U(f, P)
p
always holds, we may single out those bounded functions f for which
sup L(f, P) = inf U(f, P)
p
on [a, b], thenf is Riemann integrable on [a, b], and the Riemann integral
of f on [a, b], denoted by
is defined by
b f dx =
Ja
does not stand for any specific quantity, but is merely part of the notation
for the Riemann integral. When dealing with a more general integral than
the Riemann integral, the dx is replaced by another, more appropriate,
symbol. Thus the function of the dx is to tell us that we are dealing with
the Riemann integral of f rather than one of the more general integrais
off
211
The reader should note that the Riemann integral as defined above
has no connection with derivatives. As we shall see in Section 8.3, it is
indeed true that for some Riemann integrable functions f, it is possible to
calculate the value of
by finding a function whose deriva tive is/ and evaluating this function at
a and b. However, this is nota definition of the Riemann integral, but a
theorem about it. Furthermore, there are many Riemann integrable functions for which this theorem does not hold.
We have not yet demonstrated that there are any Riemann integrable
functions. The first example below shows that every constant function
from a closed interval to R is Riemann integrable.
Examples 8.1.10 1. Let ceR and/be defined on [a, b] by f(x) =e for ali
x e [a, b]. From Example 1 of Examples 8.1.4,
r r
f dx =
e dx = c(b - a)
2. Let/be defined on [a, b] by f(x) = x for ali x e [a, b]. From Example 2 of Examples 8.1.4, we know that for each n e N we can find a
partition Pn such that
and
212
Calculus
Therefore
~ -21 (b 2 -
a 2)
~ sup
L(f, P)
p
and since
sup L(f, P)
p
inf U(f, P)
p
ibf ib
dx =
x dx = -1 (b 2 - a 2)
{~
ifO~x <!
if!~x ~ 1
and
L(f, P) = 1 - xk =
213
Therefore
and
Again, since
sup L(f, P)
p
inf U(f, P)
p
l
fdx=-
lt follows that
5. Let
214
Calculus
to [O, l]. From Example 5 of Examples 8.1.4, for every partition P of [O, l]
we have U(x, P) = 1 and L(x, P) = O. Thus
inf U(x, P) = 1
p
so the function
sup L(x, P) = O
and
Let us consider the question of the integrability of a real-valued function f defined on a closed interval [a, b]. What must we do to show thatf
is Riemann integrable on [a, b]? We must first prove that it is bounded on
[a, b], for if it is not, we cannot form its Riemann sums and it certainly is
not Riemann integrable. Having established thatf is bounded on [a, b], it
then suffices to show that
inf U(f, P)
p
sup L(f, P)
p
since this, together with the result of Proposition 8.1.8, implies that
inf U(f, P)
p
= sup
L(f, P)
p
and hence thatfis Riemann integrable on [a, b]. The task ofshowing that
inf U(f, P)
p
sup L(f, P)
p
Proof:
i~f U(f, P) + ~
215
s~p
L(f, P) -
U(f, P')
and
L(f, P.)
i~f U(f, P) + ~
and
U(f, P.)
L(f, P")
Hence
U(f, P.) <
-2
But if/is Riemann integrable on [a, b] then infp U(f, P) = suppL(f, P),
so
U(f, P) -L(f, P) <
2 +2 =e
then
inf U(f, P)
p
sup L(f, P)
p
+e
Thus
inf U(f, P) < sup L(f, P)
p
+e
sup L(f, P)
p
Calculus
216
from above the graph of the function to the area between the graph and
the x-axis from x =a to x = b (see Figure 8.1), its lower Riemann sums
are approximations from below the graph to this sarne area (see Figure
8.2), and these approximations can be made to come arbitrarily close to
the true value ofthe area, and hence to each other, by taking the partition
fine enough. Therefore for a continuous functionf, it should be the case
that
inf U(f, P) = sup L(f, P)
p
and thus thatf is Riemann integrable on [a, b]. We now prove this.
Proposition 8.1.12 Every continuous real-valued function defined on a
closed interval [a, b], where a< b, is Riemann integrable on [a, b].
Proof: Letfbe a continuous real-valued function defined on [a, b]. Since
/is continuous on the co~pact set [a, b], it is both bounded and uniformly
continuous on [a, b]. Since f is uniformly continuous on [a, b], for every
positive real number e there is some positive real number such that
YI <
~a
and
Thus
n
L
k=I
217
n, and therefore
n
h(x)
-1
1
= {~
ifO~x ~t
ift<x~l
if x is irrational, x e [O, l]
if x is rational, x e [O, 1]
Calculus
218
2.
3.
4.
Let neN and let Pn ={O, l/n, 2/n, ... , (n -1)/n, l}. Write out the
upper and lower Riemann sums U(f, Pn), L(f, Pn), U(g, Pn),
L(g, Pn), U(h, Pn), and L(h, Pn).
ln Proposition 8.1.5, show that L(f, P) ~ L(f, P').
Carry out the induction argument which completes the proof of
Proposition 8.1.5.
Using only the definition of the Riemann integral and the identity
2
m(m + 1)(2m + 1)
+ 22 ++m 2 =
--6- - -
5.
f(x) = {
-1
ifO ~x <
r 1
3
1 3~X<5
iq~x ~ 1
6.
Let Qn[O, l] = {q 1, q2 ,
}.
Xndx
7.
8.
219
{~
if x is irrational, x E [a, b]
if x is rational, x E [a, b]
9.
gdx
(Hint: prove for the case where g differs from f at only one point,
then use induction.)
10. Let f be a step function defined on [a, b], where a < b. (Thus
[a, b] is a union of mutually disjoint intervals / 1, , Im and
f(x) = ck for ali xelk.) Prove thatf is Riemann integrable on [a, b]
and evaluate
11.
~f(x) ~
h.(x)
220
Calculus
and
12.
This result is sometimes used as the definition of Riemann integrability for a function f
Let P ={a = x0 , xt> . .. , Xn = b} be a partition of [a, b], where
a < b, and let f: [a, b]--+ R be bounded. Define
n
R(f, P) =
L f(yk)(xk -
xk _ 1)
k=l
(Hint: P = P. u {y 1,
Yn} is a refinement of P. )
a
!I
Xo
Xn-2
li
Yn-1
Yn b
li
Xn
221
This result is sometimes used as the definition of Riemann integrability for a function f
We utilize Exercises 13-16 to prove that a function is Riemann integrable on a closed interval if and only if it is bounded there and its set of
discontinuities within the interval is a set of measure zero.
13. If I is an interval in R, then I =(a, b), I =[a, b], I =(a, b], or
I = [a, b), and the length of I is defined to bem(/) = b - a. A subset
S of R is said to be a set of measure zero if for every positive real
number e there exists a collection of open intervals {/"' 1 tXeA} such
that (i) A is a countable set; (ii) Se UoceA /"'; and (iii)
I:oceA m(l") <e. Note that the collection of open intervals {!"' 1 tX e A}
can be either finite or denumerable. Prove:
(a) If S is a finite subset of R, then S is a set of measure zero;
(b) If S is the union of countably many sets of measure zero, then
sisa set ofmeasure zero. (Hint: if s = Un Sm then Sn e
Ink
where I:k m(lnk) < e/2n.)
(e) If S is a countable subset of R, then S is a set of measure zero;
(d) If T is a subset of S and S is a set of measure zero, then T is a
set of measure zero;
(e) If S contains an open interval, then S is not a set of measure
zero.
14. Letfbe a bounded function on [a, b], where a < b. If I is any interval
contained in [a, b], define w1 (I), the oscillation of f over /, by
uk
wJ(l)
15.
Prove that
(a) lffis continuous at t, then w1 (t) =O;
(b) Iff is discontinuous at t, then w1 (t) >O;
(e) If Sn = {te(a, b] 1W1(t);;;.:: l/n}, then Sn is ciosed. (Hint: if ris a
limit point of Sm any open interval I containing r must contain
a point s of Sm and then w1 (1) ;;;.:: w1 (s) ;;;.:: l/n.)
Letfbe a real-valued function defined on [a, b], where a < b. Prove
that if f is Riemann integrable on [a, b], then the set of discontinuities of f within [a, b] is a set of measure zero.
222
Calculus
(a) Let Sn ={te [a, b] 1 mj(t) ~ l/n }. Show that Sn is a set of
measure zero:
(i) There is a partition P of [a, b] such that
U( f, P) - L( f, P) < 2n
m1([xk- i xk])(xk - xk _ i) = U( f, P) - L( f, P)
k=i
and then use this and the result of (i) above to show that
(v) Use the results of (iii) and (iv) above to prove that Sn is a
set of measure zero.
223
16.
(b) Show that the set of discontinuities offwithin [a, b] is the union
of the sets Sm n e N, and then conclude that the set of discontinuities of f is a set of measure zero.
Letfbe a real-valued function defined on [a, b], where a <b. Prove
that iff is bounded on [a, b] and its set of discontinuities on [a, b] is
a set of measure zero, then f is Riemann integrable on [a, b].
(a) Show that we may assume that wf([a, b]) >O.
(b) Choose n eN such that (b - a)/n < e/2, and define
m(J.i)
U( f, P) - L( f, P) < -
Calculus
224
1 ~j
that
17.
R(f, g, P) =
k~I
'
{~
if ~X< t
ift~x~ 1
g on [O, l].
225
ifO~x<!
if!~x~ 1
{~
(d) Let/(x) = 1 for all xe[a, b] and let g(x) be bounded on [a, b].
Prove that/ is Riemann-Stieltjes integrable with respect to g on
[a, b] and that
r
f
dg = g(b) - g(a)
18.
if they exist.
(a) If f and g are bounded functions on [a, b], prove that f is
Riemann-Stieltjes integrable with respect to g on [a, b] if and
only if for every positive real number e there is a partition P. of
[a, b] such that whenever P', P" are refinements of P., then
IR(f, g, P') - R(f, g, P")I <e
Calculus
226
~y
imply g(x)
g(y)
whenever
lx - y 1 <
2.
and
L(rf, P) = r L(f, P)
and
L(rf, P) = r U(f, P)
Proof: We prove the first part of ( 1), lea ving the remainder of the proof
as an exercise (See Exercise_ l at the end of this section.)
227
then
n
U(rf, P) =
k=I
n
.=
sup{rf(x) lxe[xk_.,xk]}(xk-xk-1)
k=I
n
rsup{f(x) lxe[xk-lxk]}(xk-xk-1)
k=I
n
= r
k=I
= r U(f, P) D
Corollary 8.2.2
a< b, then
reR, r;;:;: O implies
1.
2.
and
and
228
Calculus
rfdx =r rfdx
and
b f dx
Ja
Thus we have
sup L(rf, P) = r
p
b f dx
Ja
and
inf U(rf, P) = r
p
b f
Ja
dx
and the proposition is proved when r ~ O. The proof for the case r < O is
similar. D
Next we prove that the integral of a sum (difference) is the sum
( difference) of the integrais.
Proposition 8.2.4 Iff and g are functions which are Riemann integrable
on [a, b], then f + g and f - g are Riemann integrable on [a, b], and
r
r
r(/- r r
(f+g)dx= rfdx+
gdx
g dx
g) dx
f dx -
Proof: It will suffice to prove the proposition for f + g, since the conclusion for f - g will then follow from Proposition 8.2.3 and the fact that
f-g = f + ( - l)g.
229
Since f and g are Riemann integrable on [a, b], they are bounded
there, and thus f + g is also bounded there. Because f is Riemann integrable on [a, b], there is for every positive real number e a partition P. of
[a, b] such that U(f, P.) - L(f, P.) < e/2, and it follows that
U(g, P;)
+ g(x)
b
6
<]a gdx +2
p] it is clear that
P; such that
1xe[tX,
+ sup {g(x)
1 xe[tX,
P]}
U(f + g, P)
U(f, P)
+ U(g, P)
ln particular, then,
U(f + g, P. u P;)
U(f, P. u P;)
+ U(g, P. u P;)
But
U(g, P.uP;)
Jb f dx +26
r
r r
~ U(g, PJ<
Therefore
f dx
gdx
+~
g dx +e
b f dx + b g dx +e
]a
Ja
Calculus
230
Hence
inf U(f +g, P)
P
~ b fdx + b gdx
Ja
Ja
Since the preceding inequality holds for every pair of functions which
are Riemann integrable on [a, b], it must hold for -fand -g, which are
Riemann integrable on [a, b] by Proposition 8.2.3. Therefore
infU(-(f +g),P) =infU(-f-g,P)
P
-f
~ b
Ja
-fdx+
b
Ja
-gdx
Jdx-f gdx
-sup L(f + g, P)
p
Thus
- sup L(f + g, P)
P
~ - b f
Ja
dx -
b g dx
Ja
or
sup L(f + g, P)
(**)
~ b fdx + b g dx
Ja
Ja
= sup (f + g, P) = b f
p
Ja
dx
+ b g dx
Ja
231
Since U(f, Pa) "'."'" L(f, P) and U(f, Pb) - L(f, Pb) are nonnegative, it follows that
and
Hencefis Riemann integrable on [a, e] and on [e, b] by virtue of Proposition 8.1.11.
Now we show that
If P is any partition of [a, e] and P' any partition of [e, b], then
f dx
~ U(f, P u P') =
U(f, P)
+ U(f, P')
b f
]a
dx
232
Calculus
or
The preceding inequality is valid for ali functions which are Riemann
integrable on [a, b], hence is valid for -f, and thus
b
]a
rc
-fdx
or
(**)
provided the integral on the right-hand side of the equation exists. With
these definitions established, the equality
233
of Proposition 8.2.5 holds for ali real numbers a, b, and e, as long as ali
three of the integrais exist. (See Exercise 2 at the end of this section.)
We return to our study of the properties ofthe Riemann integral. Our
next result is often useful when it is necessary to bound an integral.
rldx~r gdx
Prool:
Let h = g -
r r
hdx=
gdx-f ldx
supL(h,p) =
P
~O
b hdx ~O
]a
234
Calculus
Since -lf(x)I ~f(x) ~ lf(x)I for ali xe[a, b], it follows from Proposition 8.2.6 that
and thus
Next we state the First Mean Value Theorem for Integrais, which we
will need in the next section to prove the Fundamental Theorem of Calculus. The proof of the First Mean Value Theorem is left to the reader.
(See Exercise 7 at the end of this section.)
Proposition 8.2.8 ( First Mean Value Theorem for Integrais) If f is a
real-valued function defined and continuous on [a, b], there is some
e E (a, b) such that
f dx =f(c)(b - a)
b f
Ja
b fn dx
n-+ oo Ja
dx = Lim
235
Xndx =0
and the sequence {Xn} converges pointwise (but not uniformly) to x, the
characteristic function ofthe rationals restricted to [O, l]. But as we showed
in Example 5 of Examples 8.1.10, x is not Riemann integrable on (O, l].
2. This example will show that even if {f,,} is a sequence of Riemann
integrable functions which converges pointwise to a Riemann integrable
function f, it need not be the case that
b f
Ja
b f,, dx
n-+ oo ]a
dx = Lim
For each n eN, letfn be the function whose graph over ( 1/n, l] is zero
and whose graph over [O, l/n] is the isosceles triangle with altitude n (see
Figure 8.4). The defining equations for fn are
ifO ~ x ~ 1/2n
if 1/2n < x ~ 1/n
if 1/n <X~ l
Each functionf,, is continuous on [O, l], hence Riemann integrable there,
and using the results of Propositions 8.2.3, 8.2.4, and 8.2.5 and ofExamples
1 and 2 of Examples 8.1.10, we have
f,, dx = 2n 2
ll/2n
x dx
+ 2n il/n
1 dx - 2n 2
l/2n
21( 1 ) (1 1)
1) +o (1--n1)
-2n 21(1
- --2 n
4n
=2n - - - 0 +2n - - 2 4n 2
n 2n
2
=-
il/n x dx +
l/2n
il
l/n
Odx
Calculus
236
y
Jo f, dx =
1
Lim ! = !
2 2
n-+oo
Now let us show that uniform convergence does preserve integrability, and that under uniform convergence the limit of the integrais is the
integral of the limit.
237
for ali x E [a, b]. -Because fm is Riemann integrable on [a, b], there is a
partition P = {x0 , . . . , xn} of [a, b] such that
U(fm, P) - L(fm, P) <
Now,
8
+ 3(b -
a)
for all x E [a, b]. Since this last inequality must also hold on any subinterval of [a, b], it follows that
= U(fm, P)
= U(fm, P) +3
-xk_ 1)
Calculus
238
Similarly,
L(f, P)
3e
L(fm, P) -
Therefore
+ 3+ 3 < 3+ 3+ 3 =
b f
Ja
b f,, dx
n-+ oo Ja
dx = Lim
Since Un} converges uniformly tof on [a, b] there is for every positive real
number e some me N such that n ~ m implies
r r r(/- r
f dx -
fn dxl
=1
f,,) dxl
li - fnl dx
8 -dx
rb-
J
0
b -a
rfdx
and we are done. O
8 -(b-a) =e
=-
b-a
239
EXERCISES
1.
If a< b and
exists, define
3.
4.
5.
6.
7.
and
p=
240
Calculus
Prove that
a.(b - a)
8.
dx
~ /J(b -
a)
(b) Prove the First Mean Value Theorem for Riemann integrais.
Let f be continuous on [a, b] and let
F(t) =
9.
fdx
for ali t E [a, b]. (Note that F(a) =O by definition.) Prove that F is a
continuous function on [a, b].
(a) Letfbe continuous on [a, b] withf(x) ~O for ali n E [a, b]. Prove
that if f(x 0 ) > O for some x 0 e [a, b], then
.
rfdx>O
(b) Letf, g be continuous on [a, b] withf(x) ~ g(x) for ali xe [a, b].
Prove that if f(x 0 ) < g(x0 ) for some x 0 E [a, b], then
rfdx <
10.
11.
gdx
241
The Derivative
r r
r r
(a)
(b)
rf dg = r
f dg
f d(rg) = r
f dg
r(/
(e)
(d)
+g)dh = rfdh
gdh
+ rfdh
r r r
r r
f dg =
(f)
f dg
f(b)g(b) - f(a)g(a) =
f dg
f dg
g df
Limf(x) = x 0
X-+C
242
Calculus
x eS, O<
lx - e 1<
imply
{>
IJ(x) - x0 I < e
The proof of the proposition will be left to the reader. (See Exercise
1 at the end of this section.)
Now let us define the derivative of a function at a point.
Definition 8.3.2 Let/be a real-valued function whose domain is a subset
S of R, and let e e S be a limit point of S. The deriva tive off at e, denoted
by /'(e), is defined by
f'(c) = Lim /(x) - f(c)
x--+c
-e
Note that we consider the derivative off at e only for those points e
of the domain S which are also limit points of S. (If S is an interval or ray,
as will frequently be the case, then every point of S is a limit point of S,
and we may consider the derivative off ate for ali ceS.) However, even
if e e S is a limit point of S, the deriva tive off at e need not exist, for the
limit as x approaches e of the function
f(x) -f(c)
x-c
may not exist. Ifj'(c) does exist for ali ceS, we say that/is dijferentiable
on S. The function
f(x) -f(c)
x-c
The Derivative
243
Examples 8.3.3 1. Let S be any subset of R, let reR, and let f be the
constant function defined by f(x) = r for ali xeS. If ceS is a limit point
of S, then
O
. f(x) - f(c) L' r - r L'
Lim
= im--= im-x-+c
X - e
X-+C X e X-+C X - e
But since x # e in the limit
--=0
x-c
so
f'(c) = Lim O
X-+C
provided this limit exists. But the constant function g(x) = O is certainly
continuous at e, so by Proposition 5.4.4 of Chapter 5,
Lim O= Lim g(x) = g(c) =O
x-+c
X-+c
Thus we have shown that ifjis a constant function on S and ceSis a limit
point of S, then f'(c) exists and f'(c) =O. Therefore we may say that if
f(x) = r on S, thenf'(x) =O for ali xeS such that x is a limit point of S.
2. Let S be any subset of R, and letf be the identity function on S,
defined by f(x) = x for ali xeS. If ceS-is a limit point of S, then
f'(c) = Lim f(x) - f(c) = Lim x - e
x-+c
X -
X-+C
X -
x-c
--=l
x-c
and therefore the continuity of the constant function g(x) = 1 implies that
f'(c) = Lim 1 = 1
X-+C
244
Calculus
3. Let/(x) = x 2 for ali xeR. If ceR, then
. f(x) - f(c) L' x 2 - c 2 L' (x + c)(x - e)
Llffi
= lffi
= lffi - - - - X-+C
X-C
X-C
X-+C
X-C
X-+C
+ c)(x -
x-+c
e)
x - e
L'
lffi X
+e)
x-+c
X-+C
x- e
x-+c
x- e
x-+c
However,f'(O) does not exist. To see this, supposef'(O) does exist; then
f'(O) = Lim f(x) - /(O)
x-+0
X -
But no matter what the value of , there are real numbers y and z such
that - < y < O< z < , and hence such that
11~1_f(o)I=1-1-f'(O)I<1
The Derivative
245
and
Jx,
X -
x~c
Jx - Jc = Lim Jx - Jc Jx + Jc
x~c
.jX + Jc
X -
= L1m
X -
-e
x~c (X - c)(.jX
+ jC)
.
1
= L1m - - - x~c
.jX + Jc
Jc> 2Jc
Limx~cCJx +
=
by continuity of the function
g(x) =
+
at e. Therefore if f(x) =
and x >O, then f' =
l/2Jx. However,f'(O) does not exist, because
But
Jx Jc
Jx
X -
Jx = Lim-1x~o Jx
x~o X
Calculus
246
continuous at e.
Proof:
f(x) - f(c)
f(x) - f(c) (x - e)
x-c
Therefore
Lim (f(x) - /(e)) = Lim f(x) - f(c) Lim (x - e)
x-+c
x- e
x-+c
x-+c
provided that
f(x) - f(c)
L i. m
---x-+c
-e
and
Lim (x -e)
X-+C
by virtue of the continuity ofthe linear function g(x) = x - e and Proposition 5.4.4 of Chapter 5.-Thus
Lim (f(x) - /(e)) = f (e) O =O
X-+C
and hence
Limf(x) = f(c)
x-+c
247
The Derivative
are functions which are continuous at every real number but which do not
have a derivative at any real number. (See Exercise 24 at the end of this
section.)
We next state some of the well-known rules of differentiation. The
proofs will be left to the reader. (See Exercises 7 and 8 at the end of this
section.)
2.
3.
4.
+ g'(c),
( [_)'(e) =
g
+ f(c)g(c);
f (c)g(c) - g'(c)f(c)
[g(c)]2
for ali x in the domain of f such that x =F e. The following identity holds
for ali x in the domain of f, including x =e:
[u(x)
+f
(c)](x - e)
f(x) - f(c)
g(y~ =~~~(e)) -
g'(f(c))
248
Calculus
for ali y in the domain of g such that y =F f(c). The following identity holds
for ali y in the domain of g, including y =/(e):
( **)
[-r(y)
+ g'(f(c))](y -
Note that
Lim cr(x) = Lim f(x) - f(c) - Limf'(c) = f'(c) - f'(c) = O
x-+c
X-+C
X -
X-+C
and
Lim -r(y) = Lim g(y) y-+f(c)
y -
y-+f(c)
~~~(e)) e
Lim g'(f(c))
y-+f(c)
= g'(f(c)) - g'(f(c)) =O
Using the identities ( *) and ( ** ), we have
(g f)(x) - (g f)(c) = g(f(x)) - g(f(c))
0
= [-r(/(x))
+ g'(f(c))](f(x) -
= [-r(/(x))
+ g'(f(c))][cr(x) + f'(c)](x
f(c))
- e)
Thus if x =F e, then
(g 0 f)(x) - (g 0 f)(c)
x-c
Hence
Lim (g f)(x) - (g 0 f)(c) = Lim [-r(/(x)) + g'(f(c))] Lim [cr(x) +/'(e)]
x-+c
x- e
x-+c
x-+c
provided the limits on the right-hand side of the equation exist. However,
Limx-+c cr(x) =O implies that Limx-+c [cr(x) +f'(c)] = f'(c). Furthermore,
since /'(e) exists, f is continuous at e, and therefore Limx_...J(x) = f(c);
that is,f(x) approaches/(c) as x approaches e. Thus
Lim (f(x)) =
X-+C
Lim -r(/(x))
f(x)-+f(<)
249
The Derivative
(See Exercise 2 at the end ofthis section.) But since Limy .....f(c) r(y) =O, we
have Limf(x)-+f(c) r(f(x)) =O. Therefore
Lim [r(.[(x))
+ g'(f(c))] = g'(f(c))
X-+C
desired result.
Next we are going to consider very briefly the well-known connection
between the derivative and relative maxima and minima of a function.
Definition 8.3.7 Let f be a real-valued function defined on the open
interval (a, b). The functionfhas a relative maximum at cE (a, b) if there
is some positive real number such that
f(x)
~f(c)
for ali
xE(c -, c +)
for ali
xE(c - , c +)
Calculus
250
Let = min {y, 17 }. Then
( )
xe (e - , e+ ), x =F e
implies
1f(x) - f(c) -
x-c
so that
1
';J::; f'(c) =
lf'(c)I
so that
1
';J::;
-f'(c) = lf'(c)I
Proof: If f(x) =O for ali x E [a, b] then surely f'(c) =O for all e E (a, b)
and we are done, so we assume thatfis not identically zero on [a, b]. Since
f is continuous on [a, b], it attains its maximum at some point e E [a, b] and
its minimum at some point d E [a, b]. Since f is not identically zero on [a, b]
at least one of f(c), f(d) must be nonzero. For definiteness, suppose
f(c) =F O; then e =F a and e =F b, so e E (a, b), and beca use
/(e)= max {/(x) 1 xe[a, b]}, the functionfcertainly has a relative maximum at e. But f is differentiable on (a, b), so f'(c) exists, and therefore
f'(c) =O by Proposition 8.3.8. D
251
The Derivative
Define
g(x) = f(x) -
f(b) -f(a)
b _ a (x - a) - f(a)
Apply the Law of the Mean on [a, x], for all x E (a, b]. D
We leave it to the reader to finish the proofs of the preceding corollaries. (See Exercise 17 at the end of this section.)
Now we tum to an examination of the relationship between the
derivative and the Riemann integral. The key theorem in this regard is
the following one, which says that the derivative of the integral is the
integrand.
Theorem 8.3.13 Letfbe a real-valued function defined and continuous
on [a, b]. If F is defined on [a, b] by
F(x) =
lx
f dx
Calculus
252
Let e e [a, b], and consider the difference quotient
F(x) -F(c)
x-c
If x > e, we have
F(x) -F(c)
X - e
=-1-[ Ja
x fdx - cfdx]
X -
Ja
=-1- J
x fdx
X -
f dx =f(y)(x -e)
= -f(y)(c - x) = f(y)
for some y e (x, e). Hence for ali x e [a, b], x #e, there is some point y
between x and e such that
F(x) - F(c) = f(y)
x-c
The Derivative
253
f dx = G(b) - G(a)
Proof: Let Fbe as in Theorem 8.3.13. Since F'(x) =f(x) = G'(x) for ali
xe [a, b], by Corollary 8.3.12 there is some reR such that F(x) = G(x) + r
for all xe[a, b]. But then G(a) + r = F(a) =O, so r = -G(a), and hence
f dx = F(b)
= G(b)
+ r = G(b) -
G(a). D
i2
x412 =-(2
l
x 3 dx = 4
15
-1 4 ) = 4
l
fdx=-
but the Fundamental Theorem of Calculus does not apply here, because
Calculus
254
g(b)
f dx =
ib
g(a)
( / 0 g)g' dx
Proof: The continuity of f, g, and g' implies the existence of the integrais. Let F be any function such that F' = f on g([a, b]). ( Such a function
F exists by Theorem 8.3.13.) Let G = F g. The function G is defined on
[a, b] and by the Chain Rule
0
g(b)
g(a)
= G(b) - G(a) =
r(/
g)g' dx D
g(b)
f(x) dx =
g(a)
ib
f(g(x))g'(x) dx
it is evident that the right-hand side involves the substitution of g(x) for
x(x 2 + 1) 5 dx
255
The Derivative
Letting/(x) = x 5 and g(x) = x 2 + 1 for ali xe[O, l], we have
li
o
li
2o
1ig(I) fdx=-112x
(f 0 g)g'dx=-
2g(O)
21
dx
But if F(x) = x 6 /6, then F'(x) = x 5 (see Exercise 9 at the end of this
section) and hence
112 x
1x612
dx=-2 6
21
4
i3 X dx
Jo JXTI
If we let/(x) = x/jXTI and g(x) = x 2 - 1, then
1
3
~dx
o V .. '
ig-1(3)
g - l(O)
fdx =
12 (f
g)g' dx =
= 2 Jir2 (x 2 - 1) dx = 2
12x2l
--2xdx
1
(x33 _ x ) 12 = '83
1
We conclude this section with a discussion of the behavior of sequences of differentiable functions. Thus suppose {f,,} is a sequence of
functions which converges to the function f on [a, b]. As we did for
integrais, we can ask two questions: if fn is differentiable on [a, b] for ali
n eN, is the limit functionf differentiable on [a, b]? Iffis differentiable on
[a, b], isf = Limn .... cxJ~? We might expect the answers to these questions
will be no if the convergence is only pointwise, but yes if it is uniform.
Unfprtunately, this is not the case: even uniform convergence is not
enough to ensure that the limit function is differentiable, or, if it is, that
the derivative of the limit is the limit of the derivatives.
Examples 8.3.17 1. This example will show that a sequence of differentiable functions may converge uniformly to a nondifferentiable function.
Let f be the absolute value function, restricted to the interval [ -1, l].
According to the Weierstrass Approximation Theorem of Section 7.2,
Chapter 7, there is a sequence of polynomials which converges uniformly
Calculus
256
for ali x e [O, l] ( see Exercise 7 at the end of this section). But the sequence
{f,,} converges uniformly to the zero function on [O, l], whereas the sequence of deriva tives {!~} does not converge to the deriva tive of the zero
function (which is again the zero function), because for x = 1,
Lim f,,(l) =F f' (1) = O
n-+
oo
Proposition 8.3.18 Let Un} be a sequence offunctions defined and differentiable on [a, b]. If the sequence of deriva tives {f,,} converges uniformly
on [a, b], if the sequence Un(c)} converges for some e e [a, b], and if f,, is
continuous on [a, b] for ali n eN, then {f,,} converges to a differentiable
functionf on [a, b], and
Limf,, =f
n-+OO
on [a, b].
Proof: Suppose that {f,,} converges uniformly to the function g on
[a, b]. By Proposition 8.2.10,
Lim
n~co
jx f~ dx = jx g dx
Jc
Jc
The Derivative
257
for ali x E [a, b] and the Fundamental Theorem of Calculus implies that
lx f~ dx
= fn(x) - fn(c)
oo
+ Limfn(c)
oo
n-+
for.-all x E [a, b]. Thus Limn-. 00 fn(x) exists for ali x E [a, b], so the sequence
{!~ l_'converges pointwise on [a, b]. Let Limn-. 00 fn = f; then
f(x) =
!-_!~fn(x) =
g dx
+f(c)
for ali x E [a, b]. Since {f,,} is a sequence of continuous functions which
converges uniformly to g, g must be continuous, and it follows from
Theorem 8.3.13 and ( *) above that f is differentiable and f' = g on the
closed interval with endpoints e and x, for ali x E [a, b]. Hencefis differentiable on [a, b] and
f =g =
Limf,,
n-+oo
on [a, b]. O
We should remark that uniform convergence of the sequence of
derivatives {f,,} is indeed stronger than uniform convergence of the sequence {fn}, for it can be shown that the former implies the latter,
whereas we know from Examples 8.3.16 that the latter does not imply the
former. Furthermore, Proposition 8.3.17 remains true if we remove the
hypothesis that each derivative function f,, be continuous. We assumed
the continuity of the derivatives in order to produce a relatively straightforward proof based on the Fundamental Theorem of Calculus. A proof
which does not require the continuity of the derivatives is outlined in
Exerci se 23 below.
EXERCISES
1.
2.
Limx-.c g(f(x)) =
258
Calculus
3.
4.
5.
+ h) -
h--+0
This result is often used as the definition of the deriva tive off at e.
For each of the following, find the derivative without using any
differentiation formulas.
(a) f(x) = 3x + 2 for ali xeR;
(b) f(x) = x 3 for ali x eR;
(e) f(x) = 1/x for ali xe(O, +oo);
(d) f(x) = 1/x 2 for ali x e(O, + oo ).
(a) Let
f(x) = {
.
-1
O
1
if X <0
if X= 0
if X> 0
1/3
for ali xeR, x ~O, and show thatf'(O) does not exist.
Letf be a real-valued function defined at ceR. Show that if f'(c)
exists, it is the slope of the tine tangent to the graph off at the point
6.
9.
+ g(x) -
g(c) f(c))
x-c
259
The Derivative
1O.
+ oo ). Prove that
f'(x) = qxq- I
for ali xe(O, +oo). (Hint: if q=m/n, then [f(xW=xm.)
12. Letfbe a real-valued function which is differentiable on [a, b]. Prove
that even though f' need not be continuous on [a, b], it is still true
thatf' has the intermediate value property; that is, show thatf' takes
on every value betweenf'(a) andf'(b). (Hint: if e is betweenf'(a) and
f'(b), consider g(x) = f(x) - ex.)
13. Letfbe a real-valued one-to-one function defined on [a, b] and let g
be its inverse function. Prove that iff is continuous at e E [a, b] and
g'(f(c)) exists and is nonzero, thenf'(c) exists and
1
f'(c) = g'(f(c))
14.
15.
~y
imply
f(x)
~f(y)
(b) Prove that if f'(x) >O for ali x E (a, b), then
16.
imply
f(x) <f(y)
(e) State and prove results analogous to (a) and (b) if/'(x) ~O and
f'(x) <O on (a, b)
(a) Prove the First Derivative Test for relative maxima:
Letfbe a real-valued function which is continuous at cER and
suppose there exists a positive real number such that f is
differentiable on (e - , e +), except possibly at e. If f'(x) >O
for x E (e - , e) andf'(x) <O for x e(c, e +), thenfhas a relative maximum at e.
(b) State and prove a First Derivative Test for relative minima.
260
Calculus
17.
Fill in the details of the proofs for Corollaries 8.3.l O, 8.3. l l, and
8.3.12.
Prove the Cauchy form of the Law of the Mean:
18.
f'
. f- = L.im---;
L im
X-+Q
20.
X-+a
lx
f dx
for ali xe[a, b]. Iffis continuous at ce[a, b], then F'(c) exists and
F'(c) = f(c).
21.
fg' dx
22.
= f(b)g(b) - f(a)g(a) -
f' g dx
(b)
I
l
x 2Jx3+} dx
X+ l
------dx
(2x 2 +4x+l) 3
The Derivative
23.
261
(e)
(d)
J2
xJx+}dx
13
(x - 1)3 dx
g(x) 11 x
E [a,
b]})
(a) If x E [a, b], apply the Law of the Mean to f,, - fm on [e, x] ( or
[x, e]) and take least upper bounds to conclude that
llfn - m li ~ (b -
a) llJ~
- fm li + lfn(c) -
fm(c)I
llJ~ -J',,, li
k1o then
(e) Let g = Limn .... co~ on [a, b]. There is some k 2 EN such that
implies
262
Calculus
Let k = max {k1, k 2 }. Since/~(y) exists, there is some >O such
that
0< lx-yl <
implies
fk(x) - fk(y)
x-y
-!:' ( )1 < ~
k
24.
for all keN. (Thus sk is the kth partia! sum of the sequence of
functions Un}.) If the sequence {sd converges pointwise (uniformly) to a function f on S, we say that the series offunctions
00
L. ln
n=I
converges pointwise ( uniformly) to the function f on S, and write
00
L. ln=!
n=I
(a) Prove the Weierstrass M-Test:
if ~::'= 1 Mn is a convergent series of nonnegative constants
and lfn(x)I ~ Mn for ali neN and ali xeS, then
00
L. ln
n=I
converges uniformly on S.
263
The Derivative
for
xe[ -!, !]
f1(X
+ 1) = f1(X)
Xn =
+4-1-n
lfm(Xn) - fm(c)I =
or
Xn =
{~-l -n
_4-1-n
ifm ~n
ifm >n
Appendix
Cardinal Numbers
266
Appendix
onto. ln such a situation S is equivalent to a subset of T and Tis equivalent to a subset of S, and intuition tells us that then S and T should be
equivalent. This important fact is indeed true; it is known as the
Schroeder-Bernstein Theorem.
(Schroeder-Bernstein) Let S and T be sets. If S is equivalent to a subset of T and T is equivalent to a subset of S, then S is
equivalent to T.
Theorem A.1
Figure A.l
S - g(X)
Figure A.2
267
Cardinal Numbers
AcBcT
implies
U VcX'
Ve"Y
implies
Appendix
268
then
and
and hence h(s 1) = h(s2) is impossible. Thus if h(s 1) = h(s2), then either s 1
and s2 are both elements of g(X), in which case the fact that g - 1 is
one-to-one implies that s 1 = s2 , or s 1 and s2 are both elements of S - g(X),
in which case the fact that f is one-to-one implies that s 1 = s2 Therefore
h is one-to-one.
Finally, suppose te T. If te X, then g(t) eg(X) and thus
t = g - 1(g(t)) = h(g(t))
= f(s) = h(s)
Examples A.2
1. The function f: R
-+ ( -
1, 1) defined by
f(x)
= 1 + lxl
269
Cardinal Numbers
Proposition A.3
X12
X13
X1.4
X21
X22
X23
X24
X31
X32
X33
X34
X41
X42
X43
X44
Since the sets Xn are mutually disjoint, the elements in the array are ali
distinct. Furthermore, every element of U,.eN Xn appears in the array. We
now define a function g: N-+ Xn by counting along diagonais ofthe array:
14
270
Appendix
Thus
g(l)
=X11,
g(2)
=X21>
g(3)
=X12
g(4)
=X31' ...
Since the function g is clearly one-to-one and onto, UneN Xn is equivalent to N and hence is countable.
Now we prove the proposition in its full generality. Suppose that A
is any countable index set and that Y"' is a countable set for every a e A.
We wish to show that UixeA Y"' is countable. Since A is countable, and
nonempty because it is an index set, A is equivalent to some subset of N.
This subset may be a proper subset of N, but in any case each a e A
corresponds to a unique neN and we may write Y"' = Yn. But ym being
countable, is equivalent to a subset (which may be proper) of the set Xn
used above, and thus UixeA Y"' is equivalent to a subset of UneN Xn. Since
we have shown that UneN Xn is countable, UixeA Y"' is countable by
Proposition A.3. D
Corollary A.5 The countable union of denumerable sets is a denumerable set.
set.
Cardinal Numbers
271
and that in fact, ISI = ~o for any denumerable set S. Furthermore, since
every infinite set contains a denumerable subset, ~o is the smallest infinite
cardinal number. Note also that ITI <~o for any finite set T.
4. We define IRI =e. (The letter e stands for continuum; the real
numbers form a continuum of points.) Note that e is an infinite cardinal
number and that ~o < e. (Why?)
Our first result concerning cardinal numbers shows that the arder
relation < behaves as expected.
Proposition A.9
1.
2.
Let
Definition A.10
1.
Let
272
2.
Appendix
Let S and
=IS x
TI
Examples A.11
1. ln this example we show that addition and multiplication of finite cardinal numbers is the sarne as addition and multiplication
of nonnegative integers. For instance, for any cardinal number oc, we have
oc + O= oc and oc O= O (here O is the cardinal number of the empty set),
because if S is a set with cardinal number oc, then
oc +o =
is u 01 = 1s1 = oc
and
(l(
o = IS
01 = 101 = o
Now suppose that oc and p are nonzero finite cardinais; then there exist
natural numbers n and m such that
oc
= 1{ 1,
2, ... , n} 1
and
p=
1{
1, 2, ... , m} 1
oc +
A similar argument shows that ocp = nm. (See Exerci se 3 at the end of this
appendix.)
2. The arithmetic of infinite cardinais is quite different from that of
finite cardinais. For instance, ~o+ ~o = ~0 To see this, let S and T be
disjoint denumerable sets; then
ISl=~o.
and
~o+~o= ISuTI
Cardinal Numbers
273
is equivalent to R. Hence
But since [O, ll is clearly equivalent to the subset {(x, O) 1x e[O, 11} of
[O, ll x [O, ll, we conclude that e ~e e. On the other hand, we may
represent each x e[O, ll by a unique decimal expansion, as in the proof of
Theorem 1.3.13, and use this to define a functionf: [O, ll x [O, ll-+[O, ll
as follows:
if x = O.a 1a 2
let
f(x, y) = O.a 1b 1a 2b2
Our next definition shows how to raise a cardinal number to a cardinal number power.
Definition A.12 Let S and T be sets. The power set T 8 is the set of ali
functions from S to T. If ex and f3 are cardinal numbers and S and T are
sets such that ISI =ex, ITI = /3, then 1311. = IT8 I.
Example A.13 Let S = {a, b} and T = {O, 1}. There are exactly four
functions from S to T, namely f, g, h, and k, defined by
f(a) = f(b) =O
g(a) = 1, g(b) =O
h(a) =O, h(b) = 1
k(a) = k(b) = 1
Therefore
as expected.
Appendix
274
Notice that in the preceding example the set S has four distinct subsets, and each of these consists of the elements of S which one and only
one of the functions from S to {O, 1} carries onto 1. Thus
S= {xeS 1k(x) = l}
Clearly, there is a one-to-one correspondence between the power set
{O, 1}8 and the set of ali subsets of S, and therefore the power set {O, 1}8
is equivalent to the set of ali subsets of S. This is so for every set S: the
power set {O, 1} 8 is equivalent to the set of ali subsets of S. Before we
prove this it will be convenient. to define the characteristic function of a
set.
Let S be a set and let T be a subset of S. The characteristic function of Tis the function XT: S-+ {O, 1} defined by
Definition A.14
if xeT
if xeS-T
Let S be a set. The power set {O, 1}8 is equivalent to the
set of ali subsets of S.
Proposition A.15
Proof: Let Y' denote the set of ali subsets of S. For each TeY', the
characteristic function XT is an element of {O, 1} 8 , and
T{xeS 1XT(X) = l}
Define a function /: Y' -+ {O, 1} 8 as follows:
f(T) = XT
We will show thatfis one-to-one and onto and hence that Y' is equivalent
to {O, 1}8 .
Suppose that/(T) = f(V) for subsets Tand Vof S; then XT = Xv. and
Cardinal Numbers
275
Example A.17 We claim that 21{ =e. Let .A denote the set of ali subsets
of N. We know that
binary expansion.
If xe[O, l] has the binary expansion x = O.a 1a 2 , define
f(x) = {neN 1an "#O}
Clearly f is a function from [O, l] to .A, and if f(x) = f(y), then x and y
have the sarne nonzero terms in their binary expansions and hence have
the sarne binary expansions, so x = y. This shows thatfis one-to-one and
thus that [O, l] is equivalent to a subset of .A.
Now we show that .A is equivalent to a subset of [O, l]. Each xe [O, l]
has a ternary expansion x = O.b 1b2 , where bne{O, 1, 2} for ali neN. If
x has two such expansions, then in one of them every term after some kth
term is a 2. Define a function g: .A -+ [O, 1] as follows: for every subset M
of N, g(M) = O.b 1b2 , where O.b 1b2 is the ternary expansion with
ifneM
ifneN-M
276
Appendix
If M and M' are subsets of .;V such that M # M', then the ternary
expansions g(M) and g(M') are different in the sense that their terms are
not identical. If g(M) and g(M') represent the sarne real number in [O, l],
then every term in one of them past some kth term must be a 2. But this
is impossible, for by the way g is defined, neither g(M) nor g(M') can have
a 2 in its ternary expansion. Therefore g is one-to-one, so .;V is equivalent
to a subset of [O, l], and the Schroeder-Bernstein Theorem now implies
that .;V is equivalent to [O, l]. O
If tX is a finite cardinal number then of course tX < 2"'. As the preceding
example suggests, this inequality remains true for infinite cardinais. This
result is known as Cantor's Theorem.
tX, tX
< 2"'.
P;oof: Let S be a set such that tX = ISI; then 2"' is the cardinal of !/, the
set of ali subsets of S. Therefore it will suffice to show that for every set
S, S is equivalent to a subset of !/, but that S cannot be equivalent to!/.
Let f: S-+ !/ be given by f(x) = { x} for ali x E S. Since f is obviously
one-to-one, S is equivalent to a subset of !/.
We show that S cannot be equivalent to!/ by proving that ifjis any
function from S to!/, thenf cannot be onto. Letfbe a function from S
to!/, so thatf(x) is a subset of S for ali xeS. Let T be the subset of S
defined by T = { x E S 1 x f(x)}. The set Te!/, but there is no x E S such
thatf(x) = T. To see this, supposef(x) = Tand consider the cases xET,
x T: if x E T, then by the definition of T, x f(x) = T, while x T = f(x),
then by the definition of T, x E T. Since each possibility leads to a contradiction, f cannot be onto, and we are done. O
The first three of the cardinais in this chain are cardinais of what might be
called "naturally-occurring" sets: ~o is the cardinal of the natural numbers, e the cardinal of the reais, and 2c the cardinal of the set of ali graphs
which can be drawn in the Euclidean plane. (See Exercise 15 at the end of
this appendix.) To our knowledge, no one has ever given an example of a
Cardinal Numbers
277
naturally-occurring set which has a cardinal greater than 2c. For this
reason it is sometimes said that as far as infinite cardinal numbers are
concerned, we can only count up to three.
Let us again consider the chain
Are there cardinal numbers between the ones in this chain? To be more
specific, is there a cardinal number between ~o and e? This leads us to a
statement of the famous Continuum Hypothesis:
Continuum Hypothesis.
oc <e
The Continuum Hypothesis was a famous unsolved problem in mathematics for over 50 years. ln order to disprove it, one would have to
exhibit a set with cardinal greater than ~o but less than e; in other words,
a set "larger" than Q but "smaller" than R. Since no such set could be
found, the Continuum Hypothesis seemed reasonable, but ali efforts to
prove it failed. P. Cohen finally settled the matter by showing that in fact
the Continuum Hypothesis is independent of the other axioms of set
theory. This means that it cannot be proved as a theorem of set theory,
nor can it be disproved. If we wish it to. be true, we must take it as an
axiom of set theory; if we wish it to be false, we must take its denial as an
axiom.
We conclude this appendix with a paradox. Let A be the set of ali
cardinal numbers. For each oceA let S"- be a set such that oc = IS"-1' and
consider Y' = urJ.EA s(J.. Each set s(J. is obviously equivalent to a subset of
Y' and therefore
for ali oc e A. Therefore jY' 1 is the largest cardinal number. But this contradicts Corollary A.19, so we have a paradox in the theory of cardinal
numbers. What is wrong? Notice that at the beginning of our construction
of the largest cardinal we said, "let A be the set of ali cardinal numbers."
ln fact, the collection of ali cardinal numbers is nota set; it is too large to
be a set, and it cannot be treated as if it were one. Thus our construction
Appendix
278
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
and
Cardinal Numbers
279
Bibliography
ANALYSIS
Bartle, Robert G. ( 1964). The Elements of Real Analysis. John Wiley and
Sons: New York.
Gelbaum, Bernard R. and Olmsted, John M. H. (1964). Counterexamples
in Analysis. Holden-Day: San Francisco.
Goffman, Casper. (1961). Real Functions. Holt, Rinehart and Winston:
New York.
Randolph, John F. ( 1968). Basic Real and Abstract Analysis. Academic
Press: New York.
Royden, H. L. ( 1963). Real Analysis. The Macmillan Company: New
York.
Rudin, Walter. ( 1964). Principies of Mathematical Analysis. McGrawHill: New York.
TOPOLOGY
Dugundji, James. ( 1970). Topology. Allyn and Bacon: Boston.
Kelley, John L. (1964). General Topology. D. Van Nostrand Company:
Princeton, New Jersey.
Moore, Theral O. (1964). Elementary General Topology. Prentice-Hall:
Englewood Cliffs, New Jersey.
281
282
Bibliography
lndex
Absolute value, 41
Absolute value function, 18
Absolute value metric, 140
Algebraic number, 279
Alternating series, 133
Approximation theorems, 192-199
Archimedean Property, 46
Arzela-Ascoli Theorem, 189
Bernstein polynomials, 195
Binomial Theorem, 195
Bolzano-Weierstrass Theorem, 162
Boundary of a set, 70
Bounded function, 185
Bounded sequence, 116
Bounded set, 43
Boundedness Test, 129
Cantor ternary set, 48-50
Cantor's Diagonalization Procedure, 37-38
Cantor's Theorem, 276
Cardinal numbers, 270-278
arithmetic of, 271-276, 278
finite and infinite, 270
set of ali, 277
Cartesian product of sets, 7
Cauchy Criterion, 160, 184
Cauchy Sequence, 158
Cauchy-Schwartz inequality, 148, 149
Chain Rute, 247
Characteristic function, 79, 87, 274
Closed bali, 142
Closed set, 58, 66-74
Closure of a set, 70
Compactness, 96-102
and uniform continuity, 154
283
lndex
284
Dense subset, 71
Denumerable set, 35, 270
Derivative, 241-263
of a function at a point, 242
Derived set, 67
Difference quotient, 242
Differentiability, 242
and continuity, 246
of a sequence of functions, 255-256, 261
Differentiation, rules of, 247-248
Disconnected set, 91
Discontinuous function, 78
Discrete metric, 141
Discrete topology, 59
Disjoint sets, 7
Divergence, of a sequence, 108, 118
of a series, 127
Domain of a function, 16
Empty set, 1
Equicontinuity, 188
Equivalence class, 165
Equivalence relation, 165
Equivalent sets, 32
Euclidean n-space, 148
Euclidean plane, 9
Extreme Value Theorem, 101
Finite Intersection Property, 104
Finite set, 11, 31-39
Finite subcovering, 96
First Derivative Test, 259
First Mean Value Theorem for Integrais,
234
Fixed point, 164
Function, 16-31
absolute value, 18
and preservation of set operations, 24
bounded, 185
composite, 26
constant, 21
continuous, 78
continuous everywhere, differentiable
nowhere, 263
derivative of, 242
difference quotient of, 242
differentiable, 242
discontinuous, 78
domain of, 16
graph of, 18
image under, 22
inverse, 27-28
one-to-one, 20
onto, 20
oscillation of, 221
piecewise linear, 199
preimage under; 22
range of, 16
restriction of, 25
Riemann integrable, 210
Riemann-Stieltjes integrable, 224
sequences of, 177-199
space, 191
step, 192
Function spaces, 190-192, 198-199
Functional limits, 134-137
and continuity, 136
and sequences, 135-136
properties of, '!37
Fundamental Theorem of Calculus, 253
Geometric series, 128
Graph of a function, 18
Greatest lower bound, 44
Greenbaun, N. N., vii
Harmonic series, 127
Hausdorff space, 99
Heine-Borel Theorem, 101
Hilbert space, 149
Homeomorphism, 86
Image, 22
Index set, 11-12
lndiscrete topology, 59
Infimum, 44
lnfinite series, 126-133, 262
alternating, 133
boundedness test, 129
comparison test, 129
convergence tests for, 129-131, 132-133,
262
convergent, 127
divergence test, 128
divergent, 127
geometric, 128
285
Index
harmonic, 127
integral test, 133
nth partia! sum, 126
of functions, 262
p-series, 130
partia! sum of, 126
ratio test, 130
root test, 133
Infinite set, 11, 31-39
lntegers, 2
Integrability theorems, 214-217, 219, 224,
228-233, 239
Integral Test, 133
lntegration by Parts Theorem, 260
lntegration by Substitution Theorem, 254
Interior of a set, 70
Intermediate Value Theorem, 93
Intersection of sets, 7
Interval, 3-4
partition of, 201
continuous image of, 96, 102
Inverse function, 27-28
lrrational numbers, 39
lsometry, 167
L'Hospital's Rute, 260
100 ' 142
/ 1, 147
Law of the Mean, 251
Cauchy form, 260
Least upper bound, 44
Limit inferior, 120
and convergence, 123
Limit of a function, 134
and sequences, 135
properties of, 137
Limit of a sequence, 108
Limit point, 67
Limit superior, 120
and convergence, 123
Lipschitz condition, 156
Local continuity, 81
Lower Riemann sum, 202
properties of, 207-210
Lower bound, 43
Measure zero, 221
Metric, 140
2-adic, 146--147
absolute value, 140
discrete, 141
natural, 140
product, 141
supremum, 142, 191
Metric space, 144-175
complete, 158
completion of, 165-175
continuity in, 149-156
Euclidean n-space, 148
Hilbert space, 149
/ 00 ' 142
/ 1, 147
sequences in, 156--165
uniform continuity in, 152-155
Metric topology, 144
Minkowski inequality, 148, 149
Monotonic sequence, 117
Monotonically increasing function, 226
Mutually disjoint sets, 12
Natural metric, 140
Natural numbers, 2
Neighborhood, 65
Nested Intervals Theorem, 161
Norm, 149, 190
Normal space, 105
Nul! set, 1
One-to-one function, 20
Onto function, 20
Open bali, 142
Open covering, 96
Open ray topology, 62, 98-99
Open set, 58, 66--74
in the natural topology, 59-60, 61
Ordered pair, 7
Oscillation of a function, 221
p-adic valuation on Q, 146--147
p-series, 130
Partia! sum, 126
Partition, of an interval, 201
of a set, 166
Piecewise linear function, 199
Pointwise convergence, 177-182
Power set, 273
286
Preimage, 22
Product metric, 141
Product topology, 65
Range of a function, 16
Ratio Test, 130
Rational numbers, 2
completion of, 172-173
countability of, 35-36, 270
dense in reais, 46, 72
incompleteness of, 48, 158
Rays, 4-5
Real tine, 2
Real numbers, 2, 41-56
Archimedean Property, 46
as a complete metric space, 159-160
as Dedekind cuts, 53-56
as equivalence classes of Cauchy sequences, 173, 174-175
completeness of, 45-46, 54, 159
construction of, 51-56, 172-173
distance between, 42
natural topology on, 59
sequences of, 116-126
series of, 126-133
triangle inequality for, 42
uncountability of, 37
Refinement of a partition, 202
Regular space, 104
Relation on a set, 165
Relative maximum, minimum, 249
Restriction of a function, 25
Riemann integral, 210-226
and continuity, 216
and differentiation, 251
and sequences of functions, 234-237, 239
First Mean Value Theorem, 234
integrability theorems, 214-217, 219-224,
228-233, 239
integration by parts, 260
integration by substitution, 254
properties of, 226-241
Riemann sum, 220
upper, lower, 202
Riemann-Stieltjes integral, 224
integrability theorems, 225-226
properties of, 241
Riemann-Stieltjes sum, 224
Index
Rolle's Theorem, 250
Root Test, 133
Schroeder-Bernstein Theorem, 266
Separation axioms, 104-105
Sequence, 107-126, 156-165, 177-199
and continuity, 111-112
bounded, 116
Cauchy, 158
Cauchy Criterion, 160
convergent, 108
divergent, 108
in Hausdorff space, 110
in a metric space, 156-165
limit inferior of, 120
limit of, 108
limit superior of, 120
monotonic, 117
of functions, 177-199
of real numbers, 116-126
Sequence of functions, 177-199, 234-238,
255-257
and differentiability, 255-257, 261
and integrability, 234-237, 239
Cauchy Criterion, 184
equicontinuous, 188
pointwise convergence of, 177
uniform convergence of, 182-183
uniformly bounded, 188
Series, of functions, 262
of real numbers, 126-133
Set, 1-16
boundary of, 70
bounded, 43, 163
Cantor ternary, 48-50
Cartesian product, 7
characteristic function of, 274
closure of, 70
compact, 96
complement of, 7
connected, 91
containment, 5
countable, 35-37, 269-270
denumerable, 35, 270
derived, 67
disconnected, 91
disjoint, 7
element of, 1
287
lndex
empty, 1
equivalence, 32
finite, 11, 31-39
identities, 10, 13-14
index, 11-12
infinite, 11, 31-39
interior of, 70
intersection, 7
limit point of, 67
of measure zero, 221
operations, 7
partition of, 166
power, 273
proper subset of, 5
relation on, 165
subset of, 5totally bounded, 163
unbounded,43
uncountable, 35
union, 7
Step function, 192
Subsequence, 125 '
Subset, 5
Subspace topology, 62
Supremum, 44
Supremum metric, 142, 191
Supremum norm, 190
Topological space, 57-65
closed s'et in, 58
Hausdorff space, 99
neighborhood in, 65
point of, 58
subspace of, 62
normal space, 105