Giovanni Leoni, Topology, Fall 2009, Carnegie Mellon University

Monday, August 24, 2009

1 Metric Spaces
De…nition 1 A metric on a set X is a map d : X X ! [0; 1) such that

(i) d (x; y) d (x; z) + d (z; y) for all x; y; z 2 X,

(ii) d (x; y) = d (y; x) for all x; y 2 X,
(iii) d (x; y) = 0 if and only if x = y.

A metric space (X; d) is a set X endowed with a metric d. When there is no

possibility of confusion, we abbreviate by saying that X is a metric space.

Example 2 Here are some of the most important examples of metric spaces.

(i) In R we have d (x; y) := jx yj.

(ii) In RN , N 1, for x = (x1 ; : : : ; xN ) and y = (y1 ; : : : ; yN ),
q
2 2
d (x; y) := (x1 y1 ) + : : : + (xN yN ) :

(iii) Following what we do in RN , given two metric spaces (X; dX ) and (Y; dY ),
we can construct a metric in X Y , taking
q
2 2
d ((x1 ; y1 ) ; (x2 ; y2 )) := d1 (x1 ; x2 ) + d1 (y1 ; y2 ) :

(iv) Given a nonempty set X, the metric

1 if x 6= y;
d (x; y) :=
0 if x = y
is called the discrete metric.
(v) Consider the interval [a; b] R and take

C ([a; b]) := ff : [a; b] ! R : f continuousg

with the metric

d (f; g) := max jf (x) g (x)j :
x2[a;b]

Note that here we are using the Weierstrass theorem. More generally, if
K RN is compact, then we can take

C (K) := ff : K ! R : f continuousg

with the metric

d (f; g) := max jf (x) g (x)j :
x2K

1
(vi) Consider the interval [a; b] R and take

C ([a; b]) := ff : [a; b] ! R : f continuousg

with Z b
d (f; g) := jf (x) g (x)j dx;
a
where the integral is the Riemann integral. Is this a metric?
If we take instead

R ([a; b]) := ff : [a; b] ! R : f bounded and Riemann integrableg

with Z b
d (f; g) := jf (x) g (x)j dx;
a
then do we have a metric? This is called a pseudometric. This will moti-
vate quotient spaces.

2
 Wednesday, August 26, 2009

Example 3 (vii) Consider C ((0; 1)) := ff : (0; 1) ! R : f is continuousg. Con-

sider Kn := n1 ; 1 n1 . Then
1
[
Kn = (0; 1) :
n=1

De…ne
1 maxx2Kn jf (x) g (x)j
d (f; g) := max : (1)
n 2n 1 + maxx2Kn jf (x) g (x)j
Then d is a metric and d (fn ; f ) ! 0 ! 0 if and only if fn ! f uniformly
on compact sets.
N
More generally if SR1 is an open set, construct an increasing sequence
of compact sets such n=1 Kn = and put on C ( ) a metric as in (1).
(viii) Given a set X let

Bb (X) = `1 (X) := ff : X ! R : f is boundedg

with
d1 (f; g) := sup jf (x) g (x)j :
x2X

For the next example we need the notion of in…nite sums.

De…nition 4 Given a set X and a function f : X ! [0; 1] the in…nite sum

X
f (x)
x2X

is de…ned as
( )
X X
f (x) := sup f (x) : Y X; Y …nite :
x2X x2Y

Proposition 5 Given a set X and a function f : X ! [0; 1], if

X
f (x) < 1;
x2X

then the set fx 2 X : f (x) > 0g is countable, say, fxn gn and

X X
f (x) = f (xn ) ;
x2X n

where the right-hand side is either a …nite sum or a series. Moreover, f does
not take the value 1.

3
Proof. De…ne X
M := f (x) < 1:
x2X

For k 2 N set Xk := x 2 X : f (x) > k1 and let Y be a …nite subset of Xk .

Then X
1
number of elements of Y f (x) M;
k
x2Y

which shows that Y cannot have more than bkM c elements, where b c is the
integer part. In turn, Xk has a …nite number of elements, and so
1
[
fx 2 X : f (x) > 0g = Xk
k=1

is countable.
Exercise 6 Let f : [a; b] ! R be an increasing function. Prove that the set of
discontinuity points of f is countable.
Exercise 7 Given a nonempty set X and two functions f; g : X ! [0; 1].
(i) Prove that
X X X
(f (x) + g (x)) f (x) + g (x) :
x2X x2X x2X

(ii) If f g, then X X
f (x) g (x) :
x2X x2X

Example 8 Given a nonempty set X and 1 p < 1, we de…ne the space

( )
X p
`p (X) := f : X ! [ 1; 1] : jf (x)j < 1 :
x2X

Consider the metric

! p1
X p
dp (f; g) := jf (x) g (x)j :
x2X

In the particular case in which X = N, then a function

f : N ! [ 1; 1]
n 7! f (n) =: an
is just a sequence fan gn and so we have
( 1
)
X p
p
` (N) := fan gn R: jan j < 1 :
n=1

4
We usually write `p := `p (N).
Given a number 1 p 1, the Hölder conjugate exponent of p is the
number 1 q 1 de…ned as
8 p
< p 1 if 1 < p < 1;
q := 1 if p = 1;
:
1 if p = 1:

Note that, with an abuse of notation, we have

1 1
+ = 1:
p q

In the sequel, the Hölder conjugate exponent of p will often be denoted by p0 .

Proposition 9 (Young’s inequality) Let 1 < p < 1, and let q be its Hölder
conjugate exponent. Then
1 p 1 q
ab a + b
p q
for all a; b 0.

5
 Friday, August 28, 2009
Proof. If a = 0 or b = 0, then there is nothing to prove. Thus, assume that
a; b > 0. Since the function t 2 [0; 1) 7! ln t is concave and p1 + 1q = 1, we have

1 p 1 q 1 1
ln a + b ln ap + ln bq = ln ab;
p q p q
that is
1 p 1 q
a + b ab:
p q

Theorem 10 (Hölder’s inequality) Let X be a set, let 1 p 1, and let

q be its Hölder conjugate exponent. Given f; g : X ! [ 1; 1], then
! p1 0 1 q1
X X p
X q
jf (x) g (x)j jf (x)j @ jg (y)j A
x2X x2X y2X

if 1 < p < 1, while

!
X X
jf (x) g (x)j jf (x)j sup jg (y)j
y2X
x2X x2X

if p = 1. In particular, if f 2 `p (X) and g 2 `q (X) then f g 2 `1 (X).1

P p P q
Proof. Assume that 1 < p < 1. If x2X jf (x)j = 0 or y2X jg (y)j = 0,
then f (x) g (x) = 0 for all x 2 X and so there is nothing to prove. Thus assume
that both sums are positive. If one of them is in…nite, then the right-hand side
is 1 and so there is nothing to prove. Hence in what follows we consider the
case in which both sums are …nite belong to (0; 1).
If we apply Young’s inequality with
jf (x)j jg (x)j
a= 1 and b = 1 ;
P p p P q q
y2X jf (y)j y2X jg (y)j

we get
p q
jf (x) g (x)j 1 jf (x)j 1 jg (x)j
P + P :
P 1
p p P q
1
q
p y2X jf (y)jp q z2X jg (z)jq
y2X jf (y)j z2X jg (z)j

By Exercise 7, taking the sum on both sides, we obtain

P P p P q
x2X jf (x) g (x)j 1 x2X jf (x)j 1 x2X jg (x)j
P + P
P 1
p p P
1
q q p y2X jf (y)jp q z2X jg (z)jq
y2X jf (y)j z2X jg (z)j
1 1
= + = 1:
p q
1 Here we de…ne 0 1 to be 0.

6
This gives the desired result for 1 < p < 1.
If p = 1 and q = 1, then
jf (x) g (x)j jf (x)j sup jg (y)j ;
y2X

and we can now sum both sides.

Theorem 11 (Minkowski’s inequality) Let X be a set, let 1 p < 1, and
let f; g : X ! [ 1; 1] be two functions. Then,
! p1 ! p1 ! p1
X p
X p
X p
jf (x) + g (x)j jf (x)j + jg (x)j :
x2X x2X x2X

In particular, if f; g 2 `p (X), then f + g 2 `p (X).

P p P p
Proof. If x2X jf (x)j = 1 or x2X jg (x)j = 1 then the right-hand side
of Minkowski’s inequality is 1, and so there is nothing to prove. Thus assume
that both sums are …nite.
We consider …rst the case 1 < p < 1. By the convexity of the function
t 2 [0; 1) 7! tp , for any a, b > 0, we have
p
p a+b 2p p 2p p
(a + b) = 2p a + b = 2p 1
(ap + bp ) ;
2 2 2
and so by Exercise 7,
!
X p
X p
X p p
p 1
jf (x) + g (x)j (jf (x)j + jg (x)j) 2 (jf (x)j + jg (x)j ) ;
x2X x2X x2X

which shows that f + g 2 `p (X). To prove Minkowski’s inequality, we observe

that
X p
X p 1
jf (x) + g (x)j = jf (x) + g (x)j jf (x) + g (x)j
x2X x2X
X p 1
X p 1
jf (x)j jf (x) + g (x)j + jg (x)j jf (x) + g (x)j :
x2X x2X

By applying Hölder’s inequality, we get

! p1 ! 10
X p
X p
X (p 1)p0
p

jf (x) + g (x)j jf (x)j jf (x) + g (x)j

x2X x2X x2X
! p1 ! 10
X p
X (p 1)p0
p

+ jg (x)j jf (x) + g (x)j

x2X x2X
0 ! p1 ! p1 1 ! 10
X p
X p
X p
p
@ jf (x)j + jg (x)j A jf (x) + g (x)j ;
x2X x2X x2X

7
 P p
where we have used the fact that (p 1) p0 = p.PIf x2X jf (x) + g (x)j = 0,
p
then there is nothing to prove, thus assume that x2X jf (x) + g (x)j 2 (0; 1).
P p
1
p0
Hence, we may divide both sides of the previous inequality by x2X jf (x) + g (x)j
to obtain
! p1 ! p1 ! p1
X p
X p
X p
jf (x) + g (x)j jf (x)j + jg (x)j ;
x2X x2X x2X

where we have used the fact that p1 + p10 = 1.

If p = 1, Minkowski’s inequality follows from the triangle inequality.

8
 Monday, August 31, 2009
Minkowski’s inequality shows that `p (X) is a metric space.

Corollary 12 Let X be a nonempty set and let 1 p < 1. Then `p (X) is a

metric space.

Proof. De…ne ! p1
X p
kf kp := jf (x)j :
x2X

By Minkowski’s inequality, we have that kf + gkp kf kp + kgkp , and so, since

d (f; g) = kf gkp ;

it follows that

d (f; g) = kf gkp = kf h gkp kf hkp + kh gkp

= d (f; h) + d (h; g) :

De…nition 13 A pseudometric on a set X is a map :X X ! [0; 1) such

that

(i) (x; y) (x; z) + (z; y) for all x; y; z 2 X,

(ii) (x; y) = (y; x) for all x; y 2 X,
(iii) (x; x) = 0 for all x 2 X.

The pair (X; ) is called a pseudometric space.

Example 14 (i) Consider the set

R ([a; b]) := ff : [a; b] ! R : f is Riemann integrableg

and de…ne
Z b
(f; g) := jf (x) g (x)j dx; f; g 2 R ([a; b]) :
a

Then is a pseudometric on R ([a; b]).

(ii) If X is a nonempty set and f : X ! R is any function, then the function

f (x; y) := jf (x) f (y)j ; x; y 2 X;

is a pseudometric on X.

9
(iii) Given a nonempty set X, consider the space RX of all functions f : X ! R
(see Example 123). Fix x0 2 X and de…ne

x0 (f; g) := jf (x0 ) g (x0 )j ; f; g 2 RX :

Then x0 is a pseudometric on RX .

(iv) Given 1 p < 1 consider the set

( Z )
b
p p
L ([a; b]) := f : [a; b] ! R : f is Lebesgue measurable and jf (x)j dx < 1
a

and de…ne

Z ! p1
b
p
p (f; g) := jf (x) g (x)j dx ; f; g 2 Lp ([a; b]) :
a

Then p is a pseudometric on Lp ([a; b]).

Given a nonempty set X, a set R X X is called an equivalence relation

if

(i) (Re‡exivity) (x; x) 2 R for every x 2 X.

(ii) (Symmetry) For all x; y 2 X, if (x; y) 2 R, then (y; x) 2 R.

(iii) (Transitivity) For all x; y; z 2 X, if (x; y) 2 R and (y; z) 2 R, then

(x; z) 2 R.

We write x y if (x; y) 2 R. Given x 2 X, the equivalence class determined

by x is given by
[x] := fz 2 X : x zg :
We de…ne the quotient space

Y =X := f[x] : x 2 Xg

Given a pseudometric space, we can de…ne an equivalence relation on X.

Given x; y 2 X, we say that x y if (x; y) = 0. Consider the quotient space
Y = X . In Y we can de…ne

d ([x] ; [y]) := (x; y) ; [x] ; [y] 2 Y: (2)

Exercise 15 Let (X; ) be a pseudometric space.

(i) Prove that the function d de…ned in (2) is well-de…ned.

(ii) Prove that d is a metric.

10
1.1 Topological Properties of Metric Spaces
Let (X; d) be a metric space. Given r > 0 and x0 2 X, the ball of center x0
and radius r is the set

B (x0 ; r) := fx 2 X : d (x0 ; x) < rg :

Given a metric space (X; d) a subset U X is open if for every x 2 U there

exists r > 0 such that B (x; r) U .

Proposition 16 Let (X; d) be a metric space. Then for every x0 2 X and

r > 0 the ball B (x0 ; r) is open.

Proof. Given x 2 B (x0 ; r), we have that r1 := r d (x0 ; x) > 0. Let y 2

B (x; r1 ). Then

d (x0 ; y) d (x0 ; x) + d (x; y) < d (x0 ; x) + r1 = d (x0 ; x) + r d (x0 ; x) = r;

which shows that B (x; r1 ) B (x0 ; r). Hence, B (x0 ; r) is open.

The main properties of open sets are given in the next proposition.

Proposition 17 Let (X; d) be a metric space. Then

(i) ; and X are open.

(ii) If Ui X, i = 1; : : : ; n, is a …nite family of open sets of X, then U1 \
\ Un is open.
S
(iii) If fU g 2 is an arbitrary collection of open sets of X, then 2 U is
open.

Proof. To prove (ii), let x 2 U1 \ \ UM . Then x 2 Ui for every i = 1; : : : ; n,

and since Ui is open, there exists ri > 0 such that B (x; ri ) Ui . Take r :=
min fr1 ; : : : ; rn g > 0. Then

B (x; r) U1 \ \ Un ;

which shows that U1 \ \ UnSis open.

To prove (iii), let x 2 U := 2 U . Then there is 2 such that x 2 U
and since U is open, there exists r > 0 such that B (x; r) U U . This
shows that U is open.
Properties (i)–(iii) will be used to de…ne topological spaces.

11
 Wednesday, September 2, 2009

Remark 18 The intersection of in…nitely many open sets is not open in gen-
1 1
eral. Take X = R and Un := n ; n for n 2 N. Then

1
\ 1 1
; = f0g ;
n=1
n n

which is not open.

The following characterization of open sets is a corollary of the previous

propositions.

Corollary 19 Let (X; d) be a metric space. A nonempty set U X is open if

and only if it is given by the union of balls.

Proof. By property (iii) of Proposition 17 and by Proposition 16 the arbitrary

union of balls is open. Conversely, if U is open, then for every x 2 U there is a
ball B (x; rx ) U and so [
U= B (x; rx ) :
x2U

Exercise 20 Prove that in RN (with the Euclidean metric) every nonempty

open set can be written as union of countably many balls.

Example 21 If we take a di¤ erent metric in RN , then the previous exercise is

no longer true. Indeed, consider the discrete metric in RN ,

1 if x 6= y;
d (x; y) :=
0 if x = y:

Then
fxg if 0 < r 1;
B (x; r) =
RN if r > 1:
Consider the open set U = RN n f0g. Then by the previous corollary, we can
write [
U= B (x ; r ) :
2
N N
Since R n f0g =
6 R , necessarily r 1 for all 2 , and so
[
U= fx g :
2

Thus, the number of balls is uncountable.

12
 Given a metric space (X; d) and a set E X, a point x 2 E is called an
interior point of E if there exists r > 0 such that B (x; r) E. The interior
E of a set E X is the union of all its interior points.
The proof of following proposition is left as an exercise.

Proposition 22 Let (X; d) be a metric space and let E X. Then

(i) E is an open subset of E,

(ii) E is given by the union of all open subsets contained in E; that is, E is
the largest (in the sense of union) open set contained in E,

(iii) E is open if and only if E = E ,

(iv) (E ) = E .

Exercise 23 Let (X; d) be a metric space.

(i) Prove that if E1 ; : : : ; En are subsets of X, then

(E1 ) \ \ (En ) = (E1 \ \ En ) ;

(E1 ) [ [ (En ) (E1 [ [ En ) :

(ii) Show that in general (E1 ) [ [ (En ) 6= (E1 [ [ En ) .

(iii) Let fE g 2 be an arbitrary collection of sets of X. What is the relation,
T T S
if any, between 2 (U ) and 2 U ? And between 2 (U )
S
and 2 U ?
(iii) Think of some nontrivial condition on fE g 2 that guarantees
!
\ \
(U ) = U :
2 2

Given a metric space (X; d), a subset C X is closed if its complement

X n C.
The main properties of closed sets are given in the next proposition.

Proposition 24 Let (X; d) be a metric space. Then

(i) ; and X are closed.

(ii) If Ci X, i = 1; : : : ; n, is a …nite family of closed sets of X, then C1 [
\ Un is closed.
T
(iii) If fC g 2 is an arbitrary collection of closed sets of X, then 2 C
is closed.

13
 The proof follows from Proposition 17 and De Morgan’s laws. If fE g 2
is an arbitrary collection of subsets of a set X, then De Morgan’s laws are
!
[ \
Xn E = (X n E ) ;
2 2
!
\ [
Xn E = (X n E ) :
2 2

Given a metric space (X; d), a sequence fxn g X converges (strongly) to

x 2 X if
lim d (xn ; x) = 0:
n!1

Exercise 25 (Uniqueness of the limit) Given a metric space (X; d), prove
that if fxn g X converges to x and y in X, then x = y.

Given a metric space (X; d), a subset C X is sequentially closed if for

every sequence fxn g C such that fxn g converges to some x 2 X, then x
belongs to C.

Proposition 26 Let (X; d) be a metric space and let C X. Then C is closed

if and only if C is sequentially closed.

Proof. Step 1: Assume that C is closed and let fxn g C be such that fxn g
converges to some x 2 X. We need to show that x belongs to C. If not, then
x 2 X n C. Since X n C is open, there exists r > 0 such that B (x; r) X n C.
But then, taking " = r there exists nr 2 N such that d (xn ; x) < r for all n nr ,
which implies that xn 2 B (x; r) X n C for all n nr . This contradicts the
fact that fxn g C.
Step 2: Assume that C is sequentially closed. We need to show that X n C is
open. Let x 2 X nC. We claim that there exists r > 0 such that B (x; r) X nC.
If not, then for every r > 0 we can …nd y 2 C such that y 2 B (x; r). Taking
r = n1 we can …nd xn 2 C such that d (xn ; x) < n1 ! 0, which shows that fxn g
converges to x. Since C is sequentially closed, it follows that x 2 C, which is a
contradiction.

14
 Friday, September 4, 2009
Given a metric space (X; d) and a set E X, the closure of E, denoted E,
is the intersection of all closed sets that contain E; in other words, the closure
of E is the smallest (with respect to inclusion) closed set that contains E. It
follows by Proposition 24 that E is closed.
The proof of following proposition is left as an exercise.

Proposition 27 Let (X; d) be a metric space and let C X. Then C is closed

if and only if C = C.

Exercise 28 Let (X; d) be a metric space, let x0 2 X, and let r > 0.

(i) Prove that B (x0 ; r) fx 2 X : d (x0 ; x) rg.

(ii) Prove that in general it may happen that

B (x0 ; r) 6= fx 2 X : d (x0 ; x) rg :

Proposition 29 Let (X; d) be a metric space, let E X, and let x 2 X. Then

x 2 E if and only if B (x; r) \ E 6= ; for every r > 0.

Proof. Let x 2 E and assume by contradiction that there exists r > 0 such
that B (x; r) \ E = ;. Since B (x; r) is open and B (x; r) \ E = ;, it follows
that X n B (x; r) is closed and contains E. By the de…nition of E we have that
E X n B (x; r), which contradicts the fact that x 2 E.
Conversely, let x 2 X and assume that B (x; r) \ E 6= ; for every r > 0. We
claim that x 2 E. Indeed, if not, then x 2 X n E, which is open. Thus, there
exists B (x; r) X n E, which contradicts the fact that B (x; r) \ E 6= ;.
The previous proposition leads us to the de…nition of accumulation points.
Given a metric space (X; d) and a set E X, a point x 2 X is an accumula-
tion, or limit, point of E if for every r > 0 the ball B (x; r) contains at least one
point of E di¤erent from x. The set of all accumulation points of E is denoted
acc E.

Remark 30 Note take if x 2 X is an accumulation point of E, then by taking

r = n1 , n 2 N, there exists a sequence fxn g E with xn 6= x for all n 2 N such
that d (xn ; x) < n1 ! 0. Thus fxn g converges to x. Conversely, if there exists a
sequence fxn g E with xn 6= x for all n 2 N such that d (xn ; x) ! 0, then x is
an accumulation point of E.

It turns out that the closure of a set is given by the set and all its accumu-
lations points.

Proposition 31 Let (X; d) be a metric space and let E X. Then

E = E [ acc E:

In particular, a set C X is closed if and only if C contains all its accumulation

points.

15
Proof. Let x 2 E and assume by contradiction that x 2 = E [ acc E. Since
x2 = acc E, then there exists a ball B (x; r) that contains no other point of E
other than x, but since x 2
= E, it follows that B (x; r) X n E. This contradicts
Proposition 29.
Conversely, let x 2 E [ acc E. If x 2 E, then since E E, there is nothing
to prove. If x 2 acc E, then the result follows from Proposition 29.

Exercise 32 Let (X; d) be a metric space.

(i) Prove that if E1 ; : : : ; En are subsets of X, then

E1 \ \ En E1 \ \ En ;
E1 [ [ En = E1 [ [ En :

(ii) Show that in general E1 \ \ En 6= E1 \ \ En .

(iii) Let fE g 2 be an arbitrary collection of sets of X. What is the relation,
T T S
if any, between between 2 U and 2 U ? And between 2 U
S
and 2 U ?

(iii) Think of some nontrivial condition on fE g 2 that guarantees

[ [
U = U :
2 2

16
 Wednesday, September 9, 2009

De…nition 33 Given a metric space (X; d), a set E X is said to be dense if

E = X. The metric space (X; d) is separable if there exists a sequence fxn g X
that is dense in X.

Example 34 We discuss separability of some of the examples introduced before.

(i) RN is separable, since QN is dense in RN .

(ii) Given a nonempty set X with discrete metric

1 if x 6= y;
d (x; y) :=
0 if x = y;

X is separable if and only if X is countable. Why?

(iii) Using uniform continuity, one can show that piecewise a¢ ne functions are
dense in C ([a; b]). By approximating a piecewise a¢ ne function with one
with rational slopes and endpoints, it follow that C ([a; b]) is separable.
(iv) `1 = `1 (N) is not separable (exercise).

(v) `p = `p (N), 1 p < 1, is separable. Take

E = fx = (r1 ; : : : ; rn ; 0; : : :) : ri 2 Q; i = 1; : : : ; n; n 2 Ng :

If x = fxn g 2 `p and " > 0, let n" 2 N be so large that

1
X p "p
jxn j :
n=n" +1
2

Using the density of Q in R we may …nd r1 ; : : : ; rn" such that

n"
X p "p
jxn rn j :
n=1
2

Then y = (r1 ; : : : ; rn" ; 0; : : :) belongs to E and

n" 1
! p1 1
X p
X p "p "p p
dp (x; y) = jxn rn j + jxn j + = ":
n=1 n=n" +1
2 2

(vi) If X is uncountable, then `p (X), 1 p < 1, is not separable (exercise).

17
1.2 Completeness
A Cauchy sequence in a metric space is a sequence fxn g X such that

lim d (xn ; xm ) = 0:
n;m!1

A set E in a metric space X is bounded (in the metric sense) if it is contained

in some ball.

Exercise 35 Let (X; d) be a metric space and let fxn g X.

(i) Prove that if fxn g converges to some x 2 X, then fxn g is a Cauchy

sequence.
(ii) Prove that if fxn g is a Cauchy sequence, then the set E := fxn : n 2 Ng
is bounded.

Exercise 36 Let (X; d) be a metric space and let fxn g X.

(i) Prove that if fxn g is a Cauchy sequence and if a subsequence fxnk g of

fxn g converges to some x 2 X, then fxn g converges to x.

(ii) Prove that if there exists x 2 X such that

n for o
every subsequence fxnk g of
fxn g there exists a further subsequence xnkj that converges to x, then
fxn g converges to x.

18
 Friday, September 11, 2009
A metric space X is said to be complete if every Cauchy sequence is conver-
gent.

Proposition 37 Let (X; d) be a complete metric space and let C X. Then

C is closed if and only if (C; d) is a complete metric space.

Proof. Assume that C is closed and let fxn g C be a Cauchy sequence. Then
fxn g is a Cauchy sequence in X, and since X is complete, there exists x 2 X such
that fxn g converges to x. Using Proposition 26 we have that C is sequentially
closed, and so x 2 C, which proves that (C; d) complete. Conversely, assume
that (C; d) is a complete metric space. To prove that C is closed, again by
Proposition 26 it is enough to show that C is sequentially closed. Let fxn g C
be a sequence converging to some x 2 X. Then by the previous exercise, fxn g
is a Cauchy sequence in X, and in turn also in C. Since (C; d) is complete, it
follows that fxn g converges to some y 2 C. By the uniqueness of the limit, we
have that x = y 2 C, which shows that C is sequentially closed.

Example 38 We discuss completeness of some of the examples introduced be-

fore.

(i) RN is complete.
(ii) Given a nonempty set X with the discrete metric, then X is complete.
Why?

(iii) Given a nonempty set X, `p (X), 1 p < 1, is complete. To see this, let
ffn g `1 (X) be a Cauchy sequence. Let " > 0 and …nd n" 2 N so large
that ! p1
X p
jfn (x) fm (x)j = dp (fn ; fm ) "
x2X

for all n; m n" . Since for every …xed x 2 X,

0 1 p1
X p
jfn (x) fm (x)j @ jfn (y) fm (y)j A "
y2X

for all n; m n" , this implies that the sequence of real numbers ffn (x)g
is a Cauchy sequence in R and so there exists

lim fn (x) = f (x) 2 R:

n!1

Fix a …nite subset Y X. Since

X p
jfn (y) fm (y)j "p
y2Y

19
 for all n; m n" , letting n ! 1 gives
X p
jf (y) fm (y)j "p
y2Y

for all m n" . This holds for every …nite subset Y X. Hence, taking
the supremum over all …nite subsets Y X gives
X p p
jf (y) fm (y)j = [dp (f; fm )] "p
y2X

for all m n" ; that is dp (f; fm ) ! 0 as m ! 1. Moreover, by

Minkowski’s inequality,
! p1 ! p1 ! p1
X p
X p
X p
jf (x)j = jf (x) fn" (x)j jf (x) fn" (x)j
x2X x2X x2X
! p1 ! p1
X p
X p
+ jfn" (x)j "+ jfn" (x)j < 1;
x2X x2X

which implies that f 2 `p (X).

(iv) Given a nonempty set X, the space `1 (X) is complete. To see this, let
ffn g `1 (X) be a Cauchy sequence. Let " > 0 and …nd n" 2 N so large
that
sup jfn (x) fm (x)j = d1 (fn ; fm ) "
x2X
for all n; m n" . This implies that for every …xed x 2 X, the sequence of
real numbers ffn (x)g is a Cauchy sequence in R and so there exists
lim fn (x) = f (x) 2 R:
n!1

Since
jfn (x) fm (x)j "
for all n; m n" , letting n ! 1 gives
jf (x) fm (x)j "
for all m n" . This holds for every x 2 X. Hence, taking the supremum
over all x 2 X gives
sup jf (x) fm (x)j = d1 (f; fm ) "
x2X

for all m n" ; that is d1 (f; fm ) ! 0 as m ! 1. Moreover, for every

x 2 X,
jf (x)j jf (x) fn" (x)j sup jf (x) fn" (x)j+sup jfn" (x)j "+sup jfn" (x)j ;
x2X x2X x2X

which implies that f 2 `1 (X).

20
 (v) To prove that C ([a; b]) is complete, it is enough to show that C ([a; b]) is
a sequentially closed subset of the complete metric space `1 ([a; b]) (see
Propositions 26 and 37). Let ffn g C ([a; b]) be such that

d1 (f; fn ) ! 0

as n ! 1 for some function f 2 `1 ([a; b]). We claim that f is continu-

ous. Fix x0 2 [a; b] and " > 0. Then there exists n" 2 N such that

sup jf (x) fn (x)j = d1 (f; fn ) "

x2X

for all n n" . Since fn" is continuous at x0 there exists > 0 such that

jfn" (x) fn" (x0 )j "

for all x 2 [a; b] with jx x0 j . Hence, for all x 2 [a; b] with jx x0 j

,

jf (x) f (x0 )j jf (x) fn" (x)j + jfn" (x) fn" (x0 )j + jfn" (x0 ) f (x0 )j
" + " + ";

which shows that f is continuous.

(vi) The space C ([a; b]) with the metric
Z b
d (f; g) := jf (x) g (x)j dx;
a

where the integral is the Riemann integral, is not complete (Exercise).

(vii) Prove that C ((0; 1)) is complete.

21
 Monday, September 14, 2009

Example 39 The following example shows that if a sequence of functions con-

verges uniformly on every compact interval of an open interval, then it does not
necessarily converge uniformly in the open interval. For n 2 N consider the
function
fn (x) = xn ; x 2 R:
Then 8
>
> 1 if x > 1;
<
1 if x = 1;
lim fn (x) =
n!1 >
> 0 if 1 < x < 1;
:
does not exist if x 1:
Thus the sequence of functions ffn g converges pointwise in the set E = ( 1; 1]
to the discontinuous function

1 if x = 1;
f (x) =
0 if 1 < x < 1:

Since the functions fn are continuous and f is discontinuous, there cannot be

uniform convergence in ( 1; 1]. Consider now the interval [ a; a], where 0 <
a < 1. Then
n
sup jfn (x) f (x)j = sup jxn 0j = sup jxj = an ! 0
x2[ a;a] x2[ a;a] x2[ a;a]

as n ! 1, since 0 < a < 1. Hence, ffn g converges uniformly to 0 in every

[ a; a] ( 1; 1). Note that
n
sup jfn (x) f (x)j = sup jxn 0j = sup jxj = 1 9 0;
x2( 1;1) x2( 1;1) x2( 1;1)

so there is no uniform convergence in ( 1; 1).

Theorem 40 (Baire category theorem) Let (X; d) be a complete metric space.

Then the intersection of a countable family of open dense sets in X is still dense
in X.

Proof. Let fUn g X be a countable family of dense open sets. We consider

here only the case in which the family fUn g is in…nite. The case in which the
family is …nite is simpler and is left as an exercise. Let
1
\
E := Un :
n=1

We claim that E = X. Fix x0 2 X. We claim that x0 2 E. To see this, in view

of Proposition 29, it is enough to show that for

B (x0 ; r) \ E 6= ;

22
for every r > 0. Fix r > 0. Since U1 is dense, x0 2 X = U1 , and so by
Proposition 29, the open set B (x0 ; r) \ U1 is nonempty. Let x1 2 B (x0 ; r) \ U1 .
Since B (x0 ; r) \ U1 is open, there exists 0 < r1 < 1 such that

B (x1 ; 2r1 ) B (x0 ; r) \ U1 : (3)

Inductively, assume that xn 2 X and 0 < rn < n1 have been chosen. Since Un+1
is dense, xn 2 X = Un+1 , and so by Proposition 29, the open set B (xn ; rn ) \
1
Un+1 is nonempty, and so there exist xn+1 2 X and 0 < rn+1 < n+1 such that

B (xn+1 ; 2rn+1 ) B (xn ; rn ) \ Un+1 : (4)

By induction we can construct two sequences fxn g and frn g such that (4) holds
and 0 < rn < n1 for all n 1. Note that, by construction (see (3) and (4)), for
every n 2 N,

B (xn+1 ; 2rn+1 ) B (xn ; rn ) B (xn ; 2rn ) B (x1 ; 2r1 ) B (x0 ; r) \ U1 :

(5)
Hence, if n; m > k, then xn ; xm 2 B (xk ; rk ), so that
2
d (xn ; xm ) d (xn ; xk ) + d (xk ; xm ) < rk + rk < :
k
By letting k ! 1, we conclude that fxn g is a Cauchy sequence. Since X is a
complete metric space, there exists x 2 X such that d (xn ; x) ! 0 as n ! 1.
We claim that x 2 B (x0 ; r) \ U` for every ` 2 N. Indeed, …x k 2 N. Then for
all n > k we have that xn 2 B (xk ; rk ), and so,

d (xk ; x) d (xk ; xn ) + d (xn ; x) < rk + d (xn ; x) :

Letting n ! 1 we conclude that d (xk ; x) rk < 2rk . It follows that x 2

B (xk ; 2rk ) B (xk 1 ; rk 1 ) \ Uk for all k 2 N (where we set r0 := r). In turn,
by (3) and (4), we have that x 2 B (x0 ; r) \ E holds and the proof is complete.

De…nition 41 Let (X; d) be a metric space. A set E X is called

(i) nowhere dense if the interior of its closure is empty.

(ii) meager if it can be written as a countable union of nowhere dense sets.

Note that if U is open and dense, then its complement is closed and nowhere
dense. Hence, we have the following.

23
 Wednesday, September 16, 2009

Corollary 42 Let (X; d) be a nonempty complete metric space. If

[
X= Cn ;
n=1

where Cn is closed for every n 2 N. Then at least one Cn has nonempty interior.
In particular, every complete nonempty metric space is not meager.

Proof. If every Cn has empty interior, then Cn is nowhere dense. Hence,

Un := X n Cn is open and dense. By De Morgan’s laws,
[ [ \
X= Cn = (X n Un ) = X n Un ;
n=1 n=1 n=1
T
which implies that n=1 Un is empty. This contradicts Baire’s theorem.

Theorem 43 There exists a continuous f : [0; 1] ! R that is nowhere monotone.

Proof. Let I be a closed interval of [0; 1] and let

II := ff 2 C ([0; 1]) : f is increasing in Ig ;

DI := ff 2 C ([0; 1]) : f is decreasing in Ig

The sets II and DI are closed. De…ne

MI := II [ DI :

Then MI is closed. Moreover, MI has empty interior (why?). Consider the

sequence fIn gn of closed intervals I1 = 0; 12 , I2 = 12 ; 1 , I3 = 0; 31 , I4 =
1 2 2 1
3 ; 3 , I5 = 3 ; 1 , I6 = 0; 4 , etc... and let Mn := MIn . Then

1
[
M := Mn
n=1

is a meager set. By the previous corollary

1
[
C ([0; 1]) 6= Mn :
n=1
S1
Any function f 2 C ([0; 1]) n ( n=1 Mn ) is nowhere monotone.

1.3 Completion of a Metric Space

In this subsection we show that every metric space can be completed.

24
De…nition 44 Given two metric spaces (X; dX ) and (Y; dY ), a function f :
X ! Y is called an isometry if

dY (f (x1 ) ; f (x2 )) = dX (x1 ; x2 )

for all x1 ; x2 2 X.

Note that f is one-to-one (injective). If f is also onto, then the metric spaces
(X; dX ) and (Y; dY ) are called isometric.

Exercise 45 Let (Y; d) be a metric space and let E X be a dense set with
the property that every Cauchy sequence fyn g E converges to some point in
Y . Prove that Y is complete.

Theorem 46 Given a metric space (X; dX ), there exists a complete metric

space (Y; dY ) and an isometry f : X ! Y such that f (X) is dense in Y . The
space Y is unique up to isometries, that is, if (Y 0 ; dY 0 ) is a complete metric
space having a dense subset isometric to X, then Y and Y 0 are isometric.

Proof. 2 Let Y := ffxn g X : fxn g is a Cauchy sequenceg. Given two Cauchy

sequences fxn g ; fzn g X, we say that they are equivalent, and we write
fxn g fzn g, if
lim dX (xn ; zn ) = 0:
n!1

It can be checked that is an equivalence relations; that is,

(i) (Re‡exivity) fxn g fxn g for every Cauchy sequence fxn g X.

(ii) (Symmetry) For all Cauchy sequences fxn g ; fzn g X, if fxn g fzn g,
then fzn g fxn g.
(iii) (Transitivity) For all Cauchy sequences fxn g ; fzn g ; fwn g X, if fxn g
fzn g and fzn g fwn g, then fxn g fwn g.

Given a Cauchy sequence fxn g X the equivalence class determined by

fxn g is given by
[fxn g] := ffzn g 2 Y : fxn g fzn gg :
We de…ne
Y := f[fxn g] : fxn g 2 Yg
and dY : Y Y ! [0; 1) as

dY ([fxn g] ; [fzn g]) := lim dX (xn ; zn ) :

n!1

Step 1: We claim that dY is well-de…ned. We begin by showing that the limit

exists in R. By the triangle inequality applied twice,

dX (xn ; zn ) dX (xn ; xm ) + dX (xm ; zm ) + dX (zm ; zn ) ;

2 I did not prove this in class, so I will not ask for the proof in the written exams, but please

read it because it is important.

25
and so (interchanging the roles of xn and yn )

jdX (xn ; zn ) dX (xm ; zm )j dX (xn ; xm ) + dX (zm ; zn ) :

Letting n; m ! 1 and using the fact that fxn g ; fzn g are Cauchy sequences in
X gives
lim jdX (xn ; zn ) dX (xm ; zm )j = 0;
n;m!1

which shows that fdX (xn ; zn )g is a Cauchy sequences in R. Thus, there exists
limn!1 dX (xn ; zn ) 2 [0; 1).
It remains to show that this limit does not depend on the choice of the
representatives in the equivalence class. Thus, let fx0n g ; fyn0 g 2 Y be such that
fxn g fx0n g and fzn g fzn0 g, so that

lim dX (xn ; x0n ) = lim dX (zn ; zn0 ) = 0:

n!1 n!1

By the triangle inequality,

dX (xn ; zn ) dX (xn ; x0n ) + dX (x0n ; zn0 ) + dX (zn0 ; zn ) ;

and so (interchanging the roles of xn and yn )

jdX (xn ; zn ) dX (x0n ; zn0 )j dX (xn ; x0n ) + dX (zn0 ; zn ) :

Letting n ! 1 and using the fact that fdX (xn ; zn )g is a convergent sequence,
we obtain
lim dX (xn ; zn ) = lim dX (x0n ; zn0 ) ;
n!1 n!1

which shows that dY is well-de…ned.

Step 2: Next we prove that dY is a metric. Given Cauchy sequences fxn g ; fzn g ; fwn g
X, we have that

0 = dY ([fxn g] ; [fzn g]) = lim dX (xn ; zn ) if and only if fxn g fzn g ;

n!1

which is equivalent to say that [fxn g] = [fzn g].

Since dX is a metric,

dY ([fxn g] ; [fzn g]) = lim dX (xn ; zn ) = lim dX (zn ; xn ) = dY ([fzn g] ; [fxn g]) :
n!1 n!1

Finally, since dX (xn ; zn ) dX (xn ; wn ) + dX (wn ; zn ), we have

lim dX (xn ; zn ) lim dX (xn ; wn ) + lim dX (wn ; zn ) ;

n!1 n!1 n!1

that is,

dY ([fxn g] ; [fzn g]) dY ([fxn g] ; [fwn g]) + dY ([fwn g] ; [fzn g]) :

26
Step 3: We construct an isometry f : X ! Y . For every x 2 X de…ne
b := [fxn g], where xn := x for all n 2 N. De…ne
x

f :X!Y
b
x 7! x

We claim that f is an isometry. To see this, note that

x; zb) = dY ([fx; : : : ; x; : : :g] ; [fz; : : : ; z; : : :g])

dY (f (x) ; f (z)) = dY (b
= lim dX (x; z) = dX (x; z) :
n!1

Step 4: We prove that f (X) is dense in Y . Let [fxn g] 2 Y and let " > 0.
Since fxn g is a Cauchy sequence, there exists n" 2 N such that

dX (xn ; xm ) "

for all n; m 2 N with n; m d

n" . Consider f (xn" ) = x n" 2 Y . Then

d
dY ([fxn g] ; x n" ) = lim dX (xn ; xn" ) ";
n!1

which proves that f (X) is dense in Y .

Step 5: We prove that Y is a complete metric space. In view of the previous
exercise and Step 4, it su¢ ces to show that every Cauchy sequence fyk g con-
tained in f (X) converges in Y . Let fyk g be a Cauchy sequence in f (X). Then
for each k 2 N we have that

ck = [fxk ; : : : ; xk ; : : :g]
yk = x

for some xk 2 X. By Step 3, we have that

lim dX (xl ; xk ) = lim dY (xbl ; x

ck ) = 0
l;k!1 l;k!1

and so the sequence fxn g X is a Cauchy sequence. Hence, [fxn g] 2 Y . We

ck g is a Cauchy
claim that fyk g converges to [fxn g]. Fix " > 0. Since fyk = x
sequence, there exists k" 2 N such that

dY (xbl ; x
ck ) = dX (xl ; xk ) "

for all l; k 2 N with l; k k" . Then for every k k" ,

ck ) = lim dX (xn ; xk )
dY ([fxn g] ; x ";
n!1

which proves that fyk g converges to [fxn g].

Step 6: (sketch) Finally, we prove uniqueness. Assume that (Y 0 ; dY 0 ) is a
complete metric space having a dense subset X 0 isometric to X. Given y 0 ; w0 2
Y 0 by the density of X 0 in Y 0 , we can …nd two sequences fyn0 g ; fwn0 g X0
0 0 0 0 0 0
such that dY 0 (yn ; y ) ! 0 and dY 0 (wn ; w ) ! 0. In particular, fyn g ; fwn g are

27
Cauchy sequences in X 0 . Since X 0 is isometric to X and X is isometric to f (X),
there exist fyn g ; fwn g X corresponding to fyn0 g ; fwn0 g such that

dY (f (yn ) ; f (ym )) = dX (yn ; ym ) = dY 0 (yn0 ; ym

0
);
dY (f (wn ) ; f (wm )) = dX (wn ; wm ) = dY 0 (wn0 ; wm
0
);
dY (f (yn ) ; f (wn )) = dX (yn ; wn ) = dY 0 (yn0 ; wn0 ) ;

which implies that ff (yn )g ; ff (wn )g are Cauchy sequences in Y and so, by
the completeness of Y , they converge to some y; w 2 Y , respectively. De…ne
g (y 0 ) := y, g (w0 ) := w.
By the triangle inequality,

jdY 0 (y 0 ; z 0 ) dY 0 (yn0 ; wn0 )j dY 0 (yn0 ; y 0 ) + dY 0 (wn0 ; w0 ) ;

jdY (y; z) dY (f (yn ) ; f (wn ))j dY (f (yn ) ; y) + dY 0 (f (wn ) ; w)

and so

dY 0 (y 0 ; z 0 ) = lim dY 0 (yn0 ; wn0 ) = lim dY (f (yn ) ; f (wn )) = dY (y; z) :

n!1 n!1

Thus, Y and Y 0 are isometric.

Example 47 If we take Q with the metric d (x; y) := jx yj and we complete

it, we obtain R.

Given a metric space (X; d) and a function f : X ! R, the support of f is

the closed set
supp f := fx 2 X : f (x) 6= 0g:

Exercise 48 Consider an open interval I R and consider the space

Cc (I) := ff : I ! R continuous, supp f is a compact set of Ig

with the metric

d (f; g) := max jf (x) g (x)j :
x2I

Prove that Cc (I) is not complete and …nd its completion.

28
 Friday, September 18, 2009

1.4 Limits
De…nition 49 Consider two metric spaces (X; dX ) and (Y; dY ) and a function
f : E ! Y , where E X. Given x0 2 acc E, if there exists y 2 Y with the
property that for every " > 0 there exists = (x0 ; ") > 0 such that

dY (f (x) ; y) < "

for all x 2 E with 0 < dX (x; x0 ) < , we write

y = lim f (x)
x!x0

that y is the limit of f as x approaches x0 .

Note that x0 need not belong to E.

Proposition 50 Consider two metric spaces (X; dX ) and (Y; dY ) and a func-
tion f : E ! Y , where E X. Given x0 2 acc E and y0 2 Y the following are
equivalent.

(i) y0 = limx!x0 f (x).

(ii) f (xn ) ! y0 for every sequence fxn g E n fx0 g converging to x0 .

Proof. Assume that there exists y0 = limx!x0 f (x) and let fxn g E n fx0 g
converge to x0 . Fix " > 0 and …nd = ("; x0 ) > 0 such that

dY (f (x) ; y0 ) < "

for all x 2 E with dX (x; x0 ) < . Since xn ! x0 , there exists n" 2 N such that
dX (xn ; x0 ) < for all n n" and so

dY (f (xn ) ; y0 ) < "

for all n n" , which shows that f (xn ) ! y0 .

Conversely, assume that (ii) holds and assume by contradiction that (i) does
not hold. Then there exists " > 0 such that for every there exists x 2 E n fx0 g
with dX (x; x0 ) < such that

dY (f (x) ; y0 ) ":
1 1
Take = n for n 2 N and …nd xn 2 E n fx0 g with dX (xn ; x0 ) < n such that

dY (f (xn ) ; y0 ) ":

This contradicts (ii).

In the special case in which Y = R or Y = [ 1; 1] all the standard theorems
about the sum, product, quotient of limits continue to hold with the standard
modi…cations. We omit the details.

29
1.5 Continuity
De…nition 51 Consider two metric spaces (X; dX ) and (Y; dY ) and a function
f : E ! Y , where E X. Given x0 2 E, the function f is said to be continuous
at x0 if for every " > 0 there exists = ("; x0 ) > 0 such that

dY (f (x) ; f (x0 )) < "

for all x 2 E with dX (x; x0 ) < . The function f is said to be continuous if it

is continuous at every point of E.

Remark 52 Note that if x0 2 E is an isolated point of E; that is, if there

exists r > 0 such that BX (x0 ; r) \ E = fx0 g, then f is continuous at x0 (take
= r). Thus, it is enough to check the continuity of a function f : E ! Y at
points x0 2 E \ acc E, in which case continuity at x0 is equivalent to saying that
there exists the limit
lim f (x) = f (x0 ) :
x!x0

The space of all continuous functions f : X ! Y is denoted C (X; Y ).

If Y = R we write C (X). The space of all continuous bounded functions
f : X ! Y is denoted Cb (X; Y ). It is a metric space with the distance

d1 (f; g) := sup dY (f (x) ; g (x)) :

x2X

Proposition 53 Consider two metric spaces (X; dX ) and (Y; dY ) with X non-
empty. Then the metric space (Cb (X; Y ) ; d1 ) is complete if and only if (Y; dY )
is complete.

Proof. If (Y; dY ) is complete, then reasoning as in Example 38, one can show
that (Cb (X; Y ) ; d1 ) is a closed subset of the space of all bounded functions
f : X ! Y . We omit the details.
Conversely, assume that (Cb (X; Y ) ; d1 ) is complete and let fyn g Y be a
Cauchy sequence. De…ne the functions

fn (x) = yn ; x 2 X:

Then fn is bounded and continuous, and so it belongs to Cb (X; Y ). Moreover,

d1 (fn ; fm ) = sup dY (fn (x) ; fm (x)) = dY (yn ; ym ) ! 0

x2X

as n; m ! 1, and so ffn g is a Cauchy sequence in Cb (X; Y ). Since (Cb (X; Y ) ; d1 )

is complete, there exists f 2 Cb (X; Y ) such that fn ! f in Cb (X; Y ). Fix
x0 2 X. Then

0 dY (yn ; f (x0 )) = dY (fn (x0 ) ; f (x0 )) sup dY (fn (x) ; f (x)) = d1 (fn ; f ) ! 0
x2X

as n ! 1. Hence, yn ! f (x0 ) in Y , which shows that Y is complete.

The proof of the next proposition follows from Proposition 50.

30
Proposition 54 Consider two metric spaces (X; dX ) and (Y; dY ) and a func-
tion f : E ! Y , where E X. Given x0 2 E \ acc E, the following are
equivalent.

(i) f is continuous at x0 .
(ii) f is sequentially continuous at x0 ; that is, f (xn ) ! f (x0 ) for every
sequence fxn g E that converges that to x0 .

Exercise 55 Consider two metric spaces (X; dX ) and (Y; dY ) and a function
f : X ! Y . Prove that the following are equivalent.

(i) f is continuous.
1
(ii) f (U ) is open for every open set U Y.
1
(iii) f (C) is closed for every closed set C Y.

Next we introduce the notion of uniform continuity.

De…nition 56 Consider two metric spaces (X; dX ) and (Y; dY ) and a function
f : E ! Y , where E X. The function f is said to be uniformly continuous if
for every " > 0 there exists = (") > 0 such that

dY (f (x) ; f (x0 )) < "

for all x; x0 2 E with dX (x; x0 ) < .

31
 Monday, September 21, 2009

Remark 57 To negate uniform continuity it is enough to …ns two sequences

fxn g ; fzn g E such that

lim dX (xn ; zn ) = 0
n!1

and dY (f (xn ) ; f (zn )) 9 0 (so either the limit does not exist or it exists but it
is not zero).

Example 58 The function f (x) = x, x 2 R, is uniformly continuous, while

the function g (x) = x2 , x 2 R, is not. To see this, take " = for the function
f . To prove that g is not uniformly continuous, consider the two sequences
xn = n + n1 and zn = n. Then xn zn = n1 ! 0, while
2
1 1
f (xn ) f (zn ) = n+ n2 = 2 + ! 2 6= 0;
n n

which implies that g is not uniformly continuous, by the previous remark.

Simple examples of uniformly continuous functions are Lipschitz and Hölder’s

continuous functions.

De…nition 59 Consider two metric spaces (X; dX ) and (Y; dY ) and a function
f : E ! Y , where E X.

(i) The function f is said to be Lipschitz continuous if there exists L > 0

such that
dY (f (x1 ) ; f (x2 )) LdX (x1 ; x2 )
for all x1 ; x2 2 E. The number

dY (f (x1 ) ; f (x2 ))
Lip (f ; E) := sup L
x1 ;x2 2E; x1 ;6=x2 dX (x1 ; x2 )

is called the Lipschitz constant of f . It is also denoted Lip f . The function

f is called a contraction if Lip f < 1.

(ii) The function f is said to be Hölder continuous with exponent 2 (0; 1)

if there exists L > 0 such that
a
dY (f (x1 ) ; f (x2 )) L (dX (x1 ; x2 ))

for all x1 ; x2 2 E.

Remark 60 Consider two metric spaces (X; dX ) and (Y; dY ) and a function
f : E ! Y , where E X.

32
 (i) If f is Lipschitz continuous with Lipschitz constant L, then to see that
it is uniformly continuous,
p given " > 0, it is enough to take = L" . The
function f (x) = x, x 2 [0; 1], is uniformly continuous, but not Lipschitz.
Indeed, in the case X = Y = R, if f is Lipschitz, then
f (x) f (y)
L
x y
for all x; y 2 E, x =6 y. In particular, if f is di¤ erentiable at x, then
letting y ! x in the previous inequality gives,

jf 0 (x)j L:
p
Hence, the derivative is bounded (where it exists). For f (x) = x, x 2
1
[0; 1], we have that f 0 (x) = 2px for x 2 (0; 1), which is not bounded.
Similarly, if X = RN and Y = R, then if f is Lipschitz and admits a
@f @f
partial derivative @x i
at some point x, then @x i
(x) L.

(ii) If f is Hölder continuous with exponent 2 (0; 1) and constant L > 0,

then to see that it is uniformly continuous, given " > 0, it is enough to
1 1
take = L" . Indeed, if x1 ; x2 2 E with dX (x1 ; x2 ) < L" , then

a "
dY (f (x1 ) ; f (x2 )) L (dX (x1 ; x2 )) < L = ":
L
The Weierstrass nowhere di¤ erentiable function is an example of a uni-
formly continuous function that is not Hölder continuous of any 2 (0; 1).
(iii) If f is Lipschitz continuous with Lipschitz constant L > 0 and if E is
bounded, then f is Hölder continuous of any exponent 2 (0; 1). To see
this, let E BX (x0 ; r). Then for all x1 ; x2 2 E, we have
a 1 a
dY (f (x1 ) ; f (x2 )) LdX (x1 ; x2 ) = L (dX (x1 ; x2 )) (dX (x1 ; x2 ))
a 1 a
L (dX (x1 ; x2 )) (2r) ;

where in the last inequality we have used the fact that dX (x1 ; x2 )
dX (x1 ; x0 ) + dX (x0 ; x2 ) < r + r, since E BX (x0 ; r). If E is unbounded,
then this is no longer true. Indeed, the function f (x) = x, x 2 R, can-
not be Hölder continuous of any exponent 2 (0; 1). To see this, take
x1 = x > 0 and x2 = 0, then we cannot have an inequality of the type

x = jf (x) f (0)j Lxa ;

because as x ! 1, x goes faster than xa .

Let (X; d) be a metric space. If x 2 X and E X, the distance of x from

the set E is de…ned by

dist (x; E) := inf fd (x; y) : y 2 Eg ;

33
while the distance between two sets E1 ; E2 X is de…ned by

dist (E1 ; E2 ) := inf fd (x; y) : x 2 E1 ; y 2 E2 g :

Exercise 61 Let (X; d) be a metric space and let E X be a nonempty set.

(i) Fix x0 2 X. Prove that the function

x 2 X 7! d (x; x0 )

is Lipschitz continuous with Lipschitz constant one.

(ii) Prove that the distance function

x 2 X 7! dist (x; E)

is Lipschitz continuous with Lipschitz constant one.

(ii) Characterize the points x 2 X such that dist (x; E) = 0.

Proposition 62 Let (X; dX ) be a metric spaces, let (Y; dY ) be a complete met-

ric spaces, and let f : E ! Y be uniformly continuous, where E X. Then f
can be extended uniquely to a uniformly continuous function g : E ! Y .

34
 Wednesday, September 23, 2009
Proof. Step 1: We begin by showing that if fxn g E is a Cauchy sequence,
then ff (xn )g is a Cauchy sequence in Y . Fix " > 0. By the uniform continuity
of f there exists = (") > 0 such that
dY (f (x0 ) ; f (x00 )) < " (6)
for all x0 ; x00 2 E with dX (x0 ; x00 ) < . Since fxn g X is a Cauchy sequence,
there exists n" 2 N such that
dX (xn ; xm ) <
for all n; m n" , and so
dY (f (xn ) ; f (xm )) < " (7)
for all n; m n" , which shows that ff (xn )g is a Cauchy sequence in Y .
Step 2: For x 2 E de…ne g (x) := f (x). Fix x 2 E n E. By Remark 30 there
exists a sequence fxn g E with xn 6= x for all n 2 N such that fxn g converges
to x. In particular, fxn g is a Cauchy sequence, and so by the previous step
ff (xn )g is a Cauchy sequence in Y . Since Y is complete, ff (xn )g converges
to some element y 2 Y . Note that for any continuous extension h : E ! Y of
f to E we must have h (x) = y (this will show uniqueness). Thus, we de…ne
g (x) := y. To make sure that g is well-de…ned, we need to verify that g (x)
does not depend on the particular sequence fxn g converging to x. Thus, let
fzn g E be another sequence converging to x. Then by the triangle inequality
we have
dX (xn ; zn ) dX (xn ; x) + dX (x; zn ) < +
2 2
for all n 2 N su¢ ciently large, say n n1 . Hence, by (6),
dY (f (xn ) ; f (zn )) < "
for all n n1 . Then, since ff (xn )g converges to y,
dY (y; f (zn )) dY (y; f (xn )) + dY (f (xn ) ; f (zn )) "+"
for all su¢ ciently large, which shows that ff (xn )g converges to y. Thus, g (x)
is well-de…ned.
Step 3: It remains to show that g is uniformly continuous. Let x0 ; x00 2 E be
such that dX (x0 ; x00 ) < and consider two sequences fx0n g ; fx00n g E converging
to x0 and x00 , respectively. Then for all n su¢ ciently large we have that
dX (x0n ; x00n ) dX (x0n ; x0 ) + dX (x0 ; x00 ) + dX (x00 ; x00n ) < ;
and so by (6),
dY (f (x0n ) ; f (x00n )) < "
for all n su¢ ciently large. Letting n ! 1, we obtain
dY (g (x0 ) ; g (x00 )) ";
which shows that g is uniformly continuous.

35
Theorem 63 (Banach’s contraction principle) Let (X; d) be a nonempty
complete metric space and let f : X ! X be a contraction. Then f has a
unique …xed point; that is, there is a unique x 2 X such that f (x) = x.

1.6 Application to Ordinary Di¤erential Equations

An important application of Banach’s contraction principle is the existence of
solutions of ODE. Consider the initial value problem

u0 (t) = f (t; u (t)) ;

u (t0 ) = u0 :

Here, I R is an open interval, t0 2 I and f : I R ! R is continuous. Assume

that f satis…es the following Lipschitz condition

jf (t; z1 ) f (t; z2 )j L jz1 z2 j

for all t 2 I, z1 ; z2 2 R. Then we can prove short time existence of solutions.

Consider the space X = C ([t0 ; t0 + T ]), where T will be chosen later and con-
sider the operator

F : C ([t0 ; t0 + T ]) ! C ([t0 ; t0 + T ])

given by Z t
F (g) (t) = u0 + f (s; g (s)) ds
t0

for g 2 C ([t0 ; t0 + T ]) and t 2 [t0 ; t0 + T ]. It is clear that F is well-de…ned,

since the function on the right-hand side is continuous. Let’s prove that F is a
contraction. Take g1 ; g2 2 C ([t0 ; t0 + T ]). Then
Z t
jF (g1 ) (t) F (g2 ) (t)j = [f (s; g1 (s)) f (s; g2 (s))] ds
t0
Z t
jf (s; g1 (s)) f (s; g2 (s))j ds
t0
Z t Z t
L jg1 (s) g2 (s)j ds L max jg1 (y) g2 (y)j ds
t0 y2[t0 ;t0 +T ] t0
LT d1 (g1 ; g2 ) ;

and so taking the maximum over all t 2 [t0 ; t0 + T ], we get

d1 (F (g1 ) ; F (g2 )) LT d1 (g1 ; g2 ) :

If we take T so small that LT < 1 and t0 +T 2 I, then F is a contraction. By Ba-

nach’s contraction principle there exists a unique …xed point u 2 C ([t0 ; t0 + T ]),
that is, Z t
u (t) = F (u) (t) = u0 + f (s; u (s)) ds
t0

36
for all t 2 [t0 ; t0 + T ]. Since u is continuous, the right-hand side is of class C 1 ,
and so u is actually of class C 1 . By di¤erentiating both sides, we get that u is
a solution of the ODE. Moreover, u (t0 ) = u0 . Since any other solution of the
initial value problem is a …xed point of F , we have uniqueness.

37
 Friday, September 25, 2009

1.7 Continuity, continued

Next we prove Banach’s contraction theorem.
Proof. Step 1: Let’s …rst prove uniqueness. Assume that x1 and x2 are …xed
points of f . Then

dX (x1 ; x2 ) = dX (f (x1 ) ; f (x2 )) LdX (x1 ; x2 ) ;

which implies that

(1 L) dX (x1 ; x2 ) 0:
Since L < 1, we have that dX (x1 ; x2 ) = 0, and so x1 = x2 .
Step 2: To prove existence, …x x0 2 X and de…ne inductively

x1 := f (x0 ) ; xn+1 := f (xn ) :

We claim that fxn g is a Cauchy sequence. Indeed, note that

dX (x1 ; x2 ) = dX (f (x0 ) ; f (x1 )) LdX (x0 ; x1 )

and by induction

dX (xn ; xn+1 ) = dX (f (xn 1) ; f (xn )) Ln dX (x0 ; x1 ) :

Hence, for every m; n 2 N, by the triangle inequality

n+m
X1 n+m
X1
dX (xn ; xn+m ) dX (xi ; xi+1 ) dX (x0 ; x1 ) Li
i=n i=n
1
X Ln
dX (x0 ; x1 ) Li = dX (x0 ; x1 ) :
i=n
1 L

Letting n ! 1, we have that fxn g is a Cauchy sequence. Since the space

is complete, there exists x 2 X such that fxn g converges to x. But by the
continuity of f ,
x xn+1 = f (xn ) ! f (x) ;
which shows that f (x) = x.
Another important …xed point theorem is the following.

Theorem 64 (Brouwer’s …xed point theorem) Let K RN be a non-

empty compact convex set and let f : K ! K be a continuous function. Then
there exists x 2 K such that f (x) = x.

In the next two subsections we prove that a continuous function preserves

two important notions: connectedness and compactness.

38
1.8 Connectedness
De…nition 65 Let (X; d) be a metric space.

(i) A set E X is disconnected if there exist two nonempty disjoint open

sets U; V X such that

E U [ V; E \ U 6= ;; E \ V 6= ;:

(ii) A set E X is connected if it is not disconnected.

Next we show that continuous functions preserve connectedness.

Proposition 66 Consider two metric spaces (X; dX ) and (Y; dY ) and a con-
tinuous function f : X ! Y . Then f (E) is connected for every connected set
E X.

Proof. Let E X be a connected set and assume by contradiction that f (E)

is disconnected. Then there exist two nonempty disjoint open sets U; V X
such that
f (E) U [ V; f (E) \ U 6= ;; f (E) \ V 6= ;:
1 1
By continuity, f (U ) and f (V ) are open,
1 1 1 1
E f (U ) [ f (V ) ; E\f (U ) 6= ;; E\f (V ) 6= ;;

which shows that E is disconnected.

De…nition 67 A set C RN is convex if for every x; y 2 C, the segment of

endpoints x and y, precisely,

ftx + (1 t) y : t 2 [0; 1]g ;

is contained in C.

Theorem 68 A set C R is connected if and only if it is convex.

39
 Monday, September 28, 2009
Proof. Step 1: Assume that C is convex. We claim that C is connected. If
not, then there exist two nonempty disjoint open sets U; V R such that
C U [ V; C \ U 6= ;; C \ V 6= ;:
Let x 2 C \ U and y 2 C \ V . Without loss of generality, we may assume that
x < y. By convexity, the interval [x; y] is contained in C. Let
z := sup (U \ [x; y]) :
Then x z y. Since x 2 U and y 2 V and U and V are open, we can …nd
> 0 such that x + < z < y . We will show that z 2= U [ V . Indeed, if
z 2 U , then, since since U is open, we can …nd r > 0 such that (z r; z + r) U
and taking r < y z, we have that [z; z + r) U \ [x; y], which contradicts the
de…nition of z. Hence, z 2 = U but since
z 2 [x; y] C U [ V;
this implies that z 2 V . Since V is open, we can …nd r1 > 0 such that
(z r1 ; z + r1 ) V and taking r1 < z x, we have that (z r; z] V \ [x; y],
which contradicts the de…nition of z. This shows that C is connected.
Step 2: Assume that C is connected, let x; y 2 C, with, say, x < y. We claim
that the interval [x; y] is contained in C. If not, then there exists x < z < y
such that z 2
= C. De…ne
U := ( 1; z) ; V := (z; 1) :
Then U and V are open, disjoint, both intersect C and their union covers C.
This contradicts the fact that C is connected.
We now introduce another notion of connectedness, which is simpler to verify.
De…nition 69 Given a metric space (X; d), a continuous path, or curve, is a
continuous function : I ! X, where I R is an interval. The set (I) X
is called the range of the path. If I = [a; b], the points (a) and (b) are
called endpoints of the path. A set E X is called pathwise connected if for
all x; y 2 E there exists a continuous path with endpoints x; and y and range
contained in E.
Proposition 70 Let (X; d) be a metric space and let E X be pathwise con-
nected. Then E is connected.
Proof. We claim that E is connected. If not, then there exist two nonempty
disjoint open sets U; V X such that
E U [ V; E \ U 6= ;; E \ V 6= ;:
Let x 2 C \ U and y 2 C \ V . By hypothesis there exists a continuous path
: [a; b] ! X such that (a) = x, (b) = y and ([a; b]) E. By Proposition
66 and Theorem 68, we have that ([a; b]) is connected. On the other hand,
([a; b]) E U [ V; x2 ([a; b]) \ U; y2 ([a; b]) \ V;
which is a contradiction.

40
Proposition 71 Let (X; d) be a metric space and let E X be a connected
set. Then E is connected.

Proof. If not, then there exist two nonempty disjoint open sets U; V X such
that
E U [ V; E \ U 6= ;; E \ V = 6 ;:
By Proposition 29, it follows that E \ U 6= ; and E \ V 6= ;, which contradicts
the fact that E is connected.

41
 Wednesday, September 30, 2009
The next example and exercise show that in RN a connected set may fail to
be pathwise connected, unless the set is open.

Example 72 Let E R2 be the set given by

E1 = (x; y) 2 R2 : x = 0; 1 y 1 ;
1
E2 = (x; y) 2 R2 : x > 0; y = sin ;
x
E = E1 [ E2 :

The set E2 is connected since it is the image of (0; 1) through the continuous
function

g : (0; 1) ! R2
1
x 7! x; sin
x

and since E = E2 , it follows by Proposition 71 that E is connected. We claim

that E is not pathwise connected. Indeed, assume by contradiction that there
exists a continuous curve : [0; 1] ! E such that (0) = 1 ; 0 and (1) =
(0; 0). Let
t0 = inf ft 2 [0; 1] : (t) 2 E1 g :
Then ([0; t0 ]) contains at most one point of E1 , while ([0; t0 ]) contains E1 .
Indeed, ([0; t0 ]) contains all points x; sin x1 with 0 < x < 1 . Indeed, if there
1
existed 0 < x0 < such that ([0; t0 ]) x0 ; sin x10 2
= ([0; t0 ]), then the open
sets ( 1; x0 ) R and (x0 ; 1) R would disconnect. In particular, ([0; t0 ]) is
not closed, which contradicts the fact that a continuous function sends compact
sets into compact sets (we will prove this later).

De…nition 73 In the Euclidean space RN , a polygonal path is a continuous

path : [a; b] ! RN for which there exists a partition a = t0 < t1 < < tn = b
with the property that : [xi 1 ; xi ] ! RN is a¢ ne for all i = 1; : : : ; n, that is,

(t) = ci + tdi for t 2 [xi 1 ; xi ] ;

for some ci ; di 2 RN .

Exercise 74 Let O RN be open and connected and let x0 2 O.

(i) Prove that the set

U := fx 2 O : there exists a polygonal path with

endpoints x and x0 and range contained in Og

is open and nonempty.

42
 (ii) Prove that the set
V := fx 2 O : there does not exist a polygonal path with
endpoints x and x0 and range contained in Og
is open.
(iii) Prove that O is pathwise connected.
Exercise 75 Prove that the set R2 n Q2 is connected.
Exercise 76 Let (X; d) be a metric space and let E1 ; E2 X be two connected
sets. Prove that if there exists x 2 E1 \ E2 , then E1 [ E2 is connected.
Next we show that if a set is not connected, we can decompose it uniquely
into a disjoint union of maximal connected subsets.
Proposition 77 Let (X; d) be a metric space and let E X. Assume that
[
E= E ;
2
T
where each E is a connected set. If 2 E is nonempty, then E is connected.
Proof. We claim that E is connected. If not, then there exist two nonempty
disjoint open sets U; V X such that
E U [ V; E \ U 6= ;; E \ V 6= ;:
Since each E is connected, we must have that either E U or E V . On
the other hand, if 6= , then E \E is nonempty, while U \V is empty. Thus,
all E either belong to U or to V . This contradicts the fact that E \ U 6= ; and
that E \ V 6= ;.
Let (X; d) be a metric space and let E X. For every x 2 E, let Ex be
the union of all the connected subsets of E that contain x. Note that Ex is
nonempty, since fxg is a connected subset. In view of the previous proposition,
the set Ex is connected. Moreover, if x; y 2 E and x 6= y, then either Ex \Ey = ;
or Ex = Ey . Indeed, if not, then again by the previous proposition the set
Ex [ Ey would be connected, contained in E, and would contain x and y, which
would contradict the de…nition of Ex and of Ey . Thus, we can partition E into
a disjoint union of maximal connected subsets, called the connected components
of E.
Proposition 78 Let (X; d) be a metric space and let C X be a closed set.
Then the connected components of C are closed.
Proof. Let C be a connected component of C. Then C C C = C. By
Proposition 71, C is connected, and so by the maximality of C , C = C ,
i.e., C is closed.
Exercise 79 Prove that if U RN is open, then the connected components of
U are open. Is this still true for open subsets of arbitrary metric spaces?

43
 Friday, October 02, 2009

1.9 Compactness
De…nition 80 Let (X; d) be a metric space.

(i) A set K X is compact if for every Sopen cover of K, i.e., for every
collection fU g of open sets such that U K, there exists a …nite
subcover (i.e., a …nite subcollection of fU g whose union still contains K).
(ii) A set K X is sequentially compact if for every sequence fxn g K,
there exist a subsequence fxnk g of fxn g and x 2 K such that xnk ! x as
k ! 1.

(iii) A set K X is totally bounded if for every " > 0 there exist x1 ; : : : ; xm 2
K such that
[m
K B (xi ; ") :
i=1

(iv) A set E X is relatively compact (or precompact) if its closure E is

compact.

Example 81 Let (X; d) be a metric space. A …nite set K X is compact,

sequentially compact, and totally bounded.

The following theorem is one of the main results of this subsection.

Theorem 82 Let (X; d) be a metric space and let K X. The the following
are equivalent.

(i) K is sequentially compact.

(ii) K is complete and totally bounded.
(iii) K is compact.

We begin with a preliminary result.

Lemma 83 Let (X; d) be a metric space and let K X be compact. Then K

is closed.

Proof. It is enough to show that X n K is open. Fix x 2 X n K. For every

y 2 K consider Uy := B (y; ry ) and Vy := B (x; ry ), where r := d(x;y) 4 . Then
fUy gy2K is an open cover of K, and so there exist y1 ; : : : ; ym 2 K such that
m
[
K Uyi =: U:
i=1

44
 Tm
Let V := i=1 Vyi . Then V is open and does not intersect U . In particular,
V X n K. This shows that every point of X n K is an interior point, and so
X n K is open.
We now turn to the proof of Theorem 82.
Proof of Theorem 82. (i))(ii) Assume that K is sequentially compact. We
claim that (K; d) is complete. To see this, let fxn g K be a Cauchy sequence.
Since K X is sequentially compact, there exist a subsequence fxnk g of fxn g
and x 2 K such that xnk ! x as k ! 1. Since fxn g K is Cauchy sequence,
for every …xed " > 0 there exists n" 2 N such that
"
d (xn ; xm ) (8)
2
for all n; m 2 N with n; m n" . On the other hand, since xnk ! x as k ! 1,
there exists exists k" 2 N such that
"
d (xnk ; x) (9)
2
for all k 2 N with k k" . Fix k 2 N so large that nk maxfn" ; nk" g. Then for
all n 2 N with n n" we have that
" "
d (xn ; x) d (xn ; xnk ) + d (xnk ; x) + ;
2 2
which implies that d (xn ; x) ! 0 as n ! 1. Hence, (K; d) is complete.
Next we show that K is totally bounded. Assume by contradiction that K is
not totally bounded. Then there exists "0 > 0 such that K cannot be covered by
a …nite number of balls of radius "0 . Fix x1 2 K. Then there exists x2 2 K such
that d (x1 ; x2 ) "0 (otherwise B (x1 ; "0 ) would cover K). Similarly, we can …nd
x3 2 K such that d (x1 ; x3 ) "0 and d (x2 ; x3 ) "0 (otherwise B (x1 ; "0 ) and
B (x2 ; "0 ) would cover K). Inductively, construct a sequence fxn g K such
that d (xn ; xm ) "0 for all n; m 2 N with n 6= m. The sequence fxn g cannot
have a convergent subsequence, which contradicts the fact that K is sequentially
compact.
(ii))(i) Assume that K is complete and totally bounded. Let fxn g K.
We want to prove that a subsequence of fxn g is A Cauchy sequence and then use
the completeness of K. For every k 2 N let Bk be a …nite cover of K with balls
of radius 21k and centers in K. Since B1 covers K there exists a ball B1 2 B1
such that xn 2 B1 for in…nitely many n 2 N. Since B1 covers K \B1 there exists
a ball B2 2 B2 such that xn 2 B1 \ B2 for in…nitely many n 2 N. Inductively,
for every k 2 N we may …nd a ball Bk 2 Bk such that xn 2 B1 \ \ Bk for
in…nitely many n 2 N.
Let n1 2 N be the …rst of the integers n 2 N such that xn 2 B1 , let n2 2 N be
the …rst of the integers n 2 N such that n > n1 and xn 2 B1 \ B2 . Inductively,
for every k 2 N let nk 2 N be the …rst of the integers n 2 N such that n > nk 1
and xn 2 B1 \ \ Bk . We claim that the subsequence fxnk g is a Cauchy
sequence. Indeed, if k; ` 2 N with k; ` m, then xnk ; xn` 2 Bm , and so
1
d (xnk ; xn` ) !0
2m

45
as m ! 1. Thus, fxnk g is a Cauchy sequence and since K is complete, fxnk g
converges to a point in K.

46
 Monday, October 5, 2009
Proof. (ii))(iii) Assume that K is complete
S and totally bounded. Let fU g
be a collection of open sets such that U K. As in the previous part, for
every k 2 N let Bk be a …nite cover of K with balls of radius 21k and centers
in K. We want to prove that there exists k 2 N such that every ball in Bk is
contained in some U . Note that this would conclude the proof. Indeed, for
every B 2 Bk …x one U containing B. Since Bk is a …nite family and covers
K, the subcover of fU g just constructed has the same properties.
To …nd k, assume by contradiction that for every k 2 N there exists a ball
B xk ; 21k 2 Bk that is not contained in any U . Since fxk g K, there exist
aS subsequence xkj of fxk g and x 2 K such that xkj ! x as j ! 1. Since
U K, there exists a such that x 2 U . Since U is an open set, there
exists r > 0 such that B (x; r) U . Let j 2 N be so large that d xkj ; x < 2r
and 2k1j < 2r . Then
1
B xkj ; kj B (x; r) U ;
2
which is a contradiction.
(iii))(ii) Assume that K is compact. By the previous lemma, K is closed,
and so sequentially closed by Proposition 26. We claim that K is sequentially
compact. To see this, assume by contradiction that there exists a sequence
fxn g K that has no subsequence converging in K. Then for every m 2 N
the number of n 2 N such that xn = xm is …nite (otherwise, if xn = xm for
in…nitely many n 2 N, then this would be a convergent subsequence). More-
over, the set C := fxn : n 2 Ng has no accumulation points. Indeed, if C
had an accumulation point, then since K is sequentially closed, there would
a subsequence of fxn g converging to K. Since C has no accumulation point,
it follows, in particular, that C is closed. Similarly, for every m 2 N the sets
Cm := fxn : n 2 N; n mg are closed. Moreover, Cm+1 Cm and by what we
said before,
1
\
Cm = ;: (10)
m=1

For every m 2 N the set Um := X n Cm is open, Um+1 Um and by (10) and

De Morgan’s laws
1 1 1
!
[ [ \
Um = (X n Cm ) = X n Cm = X:
m=1 m=1 m=1

In particular, fUm gm is an open cover of K. By compactness, it follows that

there m 2 N such that
m
[
K Um = Um = X n Cm ;
m=1

which implies that K \ Cm = ;. This is a contradiction, since Cm is nonempty

and contained in K.

47
Exercise 84 Let R := [ 1; 1] be the extended real line. Prove that there is a
metric d on R that makes R compact.

Exercise 85 Let (X; d) be a compact metric space. Prove that X is separable

and complete.

Exercise 86 Let (X; d) be a metric space and let

Y := fC X : C is closed and boundedg :

Given C1 ; C2 2 Y , de…ne

dY (C1 ; C2 ) := sup fdist (x; C2 ) : x 2 C1 g + sup fdist (y; C1 ) : y 2 C2 g :

Prove that (Y; dY ) is a metric space.

Proposition 87 Let (X; d) be a metric space and let K X be compact. Then

(i) K is closed and bounded (in the metric sense),

(ii) if C K is closed, then C is compact.

Proof. (i) Let K be compact. We have already proved in Lemma 83 that K is

closed. To prove that K is bounded, assume by contradiction that it is not. Fix
x0 2 K. Then we may construct a sequence fxn g K such that d (xn ; x0 ) ! 1
as n ! 1. On the other hand, since K is sequentially compact by Theorem
82, there exist a subsequence fxnk g of fxn g and x 2 K such that xnk ! x as
k ! 1. Then
d (xnk ; x0 ) d (xnk ; x) + d (x; x0 ) M
for all k 2 N and for some M 2 R, which is a contradiction.
(ii) Let fU g 2 be an open cover of C. Then fU g 2 [ fX n Cg is an
open cover of K. By the compactness of K there exist U 1 ; : : : ; U n such that

U 1
[ [U n
[ (X n C) K C:

Then
U 1
[ [U n
C;
which shows that C is compact.

Exercise 88 This exercise shows that in RN a closed and bounded set is com-
pact.

(i) Prove that an interval [a; b] R is compact.

(ii) Prove that the Cartesian product of two compact metric space is compact.
(iii) Prove that a rectangle [a1 ; b1 ] [aN ; bN ] RN is compact.

De…nition 89 A family fE g 2 of subsets of a set X has the …nite intersec-

tion property if every …nite subfamily has nonempty intersection.

48
 A decreasing sequence of nonempty sets has the …nite intersection property.

Exercise 90 Let (X; d) be a metric space. Prove that a set K X is compact

if and only if for every family fC g 2 of closed subsets of K with the …nite
intersection property, \
C 6= ;:
2

The importance of compactness is explained by the following results.

Theorem 91 (Weierstrass) Let (X; d) be a metric space, let K X be com-

pact, and let f : X ! R be a continuous function. Then there exists x0 ; x1 2 K
such that
f (x0 ) = min f (x) ; f (x1 ) = max f (x) :
x2K x2K

49
 Wednesday, October 7, 2009
First proof. This proof only uses compactness. Let
t := inf f (x) :
x2K

If the in…mum is not attained, then for every x 2 K we may …nd t < tx < f (x).
Then the family of open sets
Ux := fy 2 X : f (y) > tx g ; x 2 K;
is an open cover for the compact set K, and so we may …nd a …nite cover
Ux1 ; : : : ; Uxl of the set K. But then for all x 2 K,
f (x) min txi > t = inf f (w) ;
i=1;:::;l w2K

which contradicts the de…nition of t.

The proof for the existence the maximum is similar.
Remark 92 Note that to prove the existence of a minimum we only used a
weaker form of continuity, namely that the set fy 2 X : f (y) > tg is open for
all t 2 R. A function satisfying this property is called lower semicontinuous.
Second proof. This proof uses sequential compactness. Construct a sequence
of real numbers t < tn such that tn ! t as n ! 1 (if t 2 R we can take
tn = t + n1 , while if t = 1, take tn = n). By the de…nition of in…mum, for
every n 2 N we may …nd xn 2 K such that
t < f (xn ) < tn :
Letting n ! 1, by the squeezing theorem, we get
lim f (xn ) = t: (11)
n!1

Since fxn g K, and K is sequentially compact by Theorem 82, there exist a

subsequence fxnk g of fxn g and x 2 K such that xnk ! x as k ! 1. Using the
continuity of f and (11), we get
t = lim f (xn ) = lim f (xnk ) = f (x) ;
n!1 k!1

which shows that the in…mum is a minimum.

The sequence fxn g constructed in the previous proof is called a minimizing
sequence.
Remark 93 Note that to prove the existence of a minimum in the second proof
we only used a weaker form of continuity, namely that the set
lim inf f (xj ) f (x)
j!1

for all sequences fxj g converging to x 2 R. A function satisfying this property

is called sequentially lower semicontinuous.

50
 In view of the previous theorem if (X; d) is a compact metric space, then
C (X) is a metric space with the distance

d1 (f; g) := max jf (x) g (x)j :

x2X

If the metric space (X; d) is not compact, then we cannot use the distance d1 .
However, we can consider a smaller space, precisely

Cc (X) := ff 2 C (X) : supp f is a compact setg

with the metric d1 .

Exercise 94 Let (X; d) be a metric space. Characterize the completion of

Cc (X). It is denoted C0 (X) and is called the space of functions vanishing
at in…nity.

Remark 95 A typical application of the Weierstrass theorem is the following.

Let (X; d) be a metric space and let f : X ! R be a continuous function.
Assume that f is bounded from below, so that

` = inf f (x) > 1

x2X

and that for a given x0 2 X,

lim f (x) = 1:
d(x;x0 )!1

We would like to know if f has a minimum. By the de…nition of limit, we can

…nd R > 0 such that f (x) > ` for all x 2 X such that d (x; x0 ) R. Thus,

` = inf f (x) = inf f (x) :

x2X x2B(x0 ;R)

If X = RN with the euclidean distance, then we know that B (x0 ; R) is compact

and so by the Weierstrass theorem, f has a minimum in B (x0 ; R) and we
are done. Unfortunately, for all reasonable in…nite-dimensional metric spaces
B (x0 ; R) is neither compact, nor sequentially compact. We will see in functional
analysis that the way around this is to put a weaker topology on the metric space
in such a way that B (x0 ; R) becomes compact.

Theorem 96 Let (X; dX ) and (Y; dY ) be two metric spaces, let K X be

compact, and let f : K ! Y be a continuous function. Then f is uniformly
continuous.

Proof. Fix " > 0. By continuity for every x 2 K there exists x = x (") > 0
such that
dY (f (x) ; f (z)) < " (12)

51
for all z 2 K with dX (x; z) < x . The family B x; 2x x2K is an open cover
for the compact set K, and so we may …nd a …nite cover x1 ; : : : ; xm such that
m
[ xi
K B xi ; :
i=1
2

Let
xi
:= min > 0:
i=1;:::;m 2
Sm xi
Let x; z 2 K be such that dX (x; z) < . Since K i=1 B xi ; 2 , there
xi
exists we may …nd xi such that x 2 B xi ; 2 , and so by the triangle inequality

xi xi xi
dX (z; xi ) dX (z; x) + dX (x; xi ) < + + = xi ;
2 2 2
which shows that z 2 B (xi ; xi ). Hence, by (12),

dY (f (x) ; f (z)) dY (f (x) ; f (xi )) + dY (f (xi ) ; f (z)) < 2":

Next we show that continuous functions preserve compactness.

Proposition 97 Consider two metric spaces (X; dX ) and (Y; dY ) and a con-
tinuous function f : X ! Y . Then f (K) is compact for every compact set
K X.

52
 Friday, October 09, 2009
Proof. Let fU g 2 be an open cover of f (K). By continuity, f 1 (U ) is
open for every 2 , and so f 1 (U ) 2 is an open cover of K. Since K
l
is compact, we may …nd U 1 ; : : : ; U l such that f 1 (U i ) i=1 cover K. In
turn, U 1 ; : : : ; U l cover f (K). Indeed, if y 2 f (K), then there exists x 2 K
such that f (x) = y. Let i = 1; : : : ; m be such that x 2 f 1 (U i ). Then
y = f (x) 2 U i .

De…nition 98 Consider two metric spaces (X; dX ) and (Y; dY ) and a bijective
function f : X ! Y . Then f is a homeomorphism if f and f 1 are continuous.

Proposition 99 Consider two metric spaces (X; dX ) and (Y; dY ) and a bijec-
tive function f : X ! Y . If X is compact and f is continuous, then f 1 is
continuous. In particular, f is a homeomorphism.

Proof. For every closed set C X, we have that C is compact by Proposition

87, and so f (C) is compact by Proposition 97. Again by Proposition 87 we
have that f (C) is closed. Let g := f 1 . We have shown that g 1 (C) is closed
for every closed set C X. Thus, by Exercise 55, g is continuous.
Note that without compactness the result is false.

Example 100 Let X = (0; 1) [ [2; 3] and de…ne

x if x 2 (0; 1) ;
f (x) =
x 1 if x 2 [2; 3] :

Then f : (0; 1) [ [2; 3] ! (0; 2] is continuous, bijective, but the inverse function
is discontinuous. The problem here is the fact that (0; 1) [ [2; 3] is not compact
nor connected.

1.10 The Ascoli–Arzela Theorem

Next we show that in an arbitrary metric space, closed and bounded sets are
not compact.

Example 101 Let X := C ([0; 1]). The sequence of functions

fn (x) = xn ; x 2 [0; 1]

is bounded in C ([0; 1]), but no subsequence converges uniformly to a continu-

ous function. This shows that BX (0; 1) is closed and bounded but not com-
pact. Hence, Bolzano–Weierstrass theorem fails for in…nite dimensional metric
spaces.

De…nition 102 Let (X; dX ) and (Y; dY ) be metric spaces. A family F of func-
tions f : X ! Y is said to be equicontinuous at a point x0 2 X if for every
" > 0 there exists = (x0 ; ") > 0 such that

dY (f (x) ; f (x0 )) "

53
for all f 2 F and for all x 2 X with d (x; x0 ) . The family F of functions
f : X ! Y is said to be (uniformly) equicontinuous if for every " > 0 there
exists > 0 such that
dY (f (x) ; f (y)) "
for all f 2 F and for all x; y 2 X with d (x; y) .

De…nition 103 Let (X; d) be a metric space. A family F of functions f : X !

R is said to be pointwise bounded if for every x 2 X there exists Mx > 0 such
that
jf (x)j Mx
for all f 2 F.

Example 104 The sequence of functions

fn (x) = xn ; x 2 [0; 1] ;

is pointwise bounded but not equicontinuous at x = 1. To see this, …x 0 < " < 1.
We want to …nd > 0 such that 1 xn " for all 1 x < 1. We
1=n 1=n
have (1 ") x. So for each n the best is 1 n = (1 ") , that is,
1=n
n =1 (1 ") ! 0 as n ! 1. Hence, no works for all n.

Example 105 Consider two metric spaces (X; dX ) and (Y; dY ) and a family F
of functions from X into Y . If there exist 2 (0; 1] if there exists L > 0 such
that
a
dY (f (x1 ) ; f (x2 )) L (dX (x1 ; x2 ))
for all x1 ; x2 2 X and for all f 2 F, then the family F is equicontinuous. The
sequence of functions
xn
fn (x) = ; x 2 [0; 1] ;
n
is pointwise bounded and equicontinuous at x = 1. Indeed,

fn0 (x) = xn 1
; x 2 [0; 1] ;

so that maxx2[0;1] xn 1 = 1, which shows that the sequence ffn g is equi-

Lipschitz (take L = 1). Hence, it is equicontinuous.

54
 Monday, October 12, 2009

Theorem 106 (Ascoli–Arzelà) Let (X; d) be a separable metric space and let
F C (X) be a family of functions. Assume that F is pointwise bounded and
equicontinuous. Then every sequence in F has a subsequence that converges
uniformly on every compact subset of X to a continuous function g : X ! R.

Proof. Without loss of generality, we may assume that F has in…nite many
elements, otherwise there is nothing to prove. Since X is separable, there exists
a countable set E X such that X = E.
Step 1: Let G F be an in…nite set. We claim that G contains a sequence
ffn g such that the limit limn!1 fn (x) exists in R for all x 2 E. The proof
makes use of the Cantor diagonal argument. Write E = fxk gk . Since the set

ff (x1 ) : f 2 Gg

is bounded in R, by the Bolzano–Weierstrass theorem we can …nd a sequence

ffn;1 gn G for which there exists the limit

lim fn;1 (x1 ) = `1 2 R:

n!1

Since the set

ffn;1 (x2 ) : n 2 Ng
is bounded in R, again by the Bolzano–Weierstrass theorem we can …nd a se-
quence ffn;2 gn ffn;1 gn for which there exists the limit

lim fn;2 (x2 ) = `2 2 R:

n!1

By induction for every k 2 N, k > 1, we can …nd a subsequence ffn;k gn

ffn;k 1 gn for which there exists the limit

lim fn;k (xk ) = `k 2 R:

n!1

We now consider the diagonal elements of the in…nite matrix, that is, the se-
1
quence ffn;n gn . For every …xed xk 2 E we have that the sequence ffn;n (xk )gn=k
is a subsequence of ffn;k (xk )gn , and thus it converges to `k as n ! 1. This
completes the proof of the claim. Set fn := fn;n and de…ne g : E ! R by

g (x) := lim fn (x) 2 R; x 2 E: (13)

n!1

Step 2: Let K X be compact and …x " > 0. By equicontinuity, there exists

> 0 such that
jf (x) f (y)j " (14)
for all f 2 F and for all x; y 2 X with d (x; y) . Since K is compact, we may
cover it with a …nite number of balls B y1 ; 2 ,. . . , B yM ; 2 . Since E is dense,

55
for every i = 1; : : : ; M there exists zi 2 B yi ; 2 \ E. Using (13), we have that
there exists an integer n" 2 N such that

jfn (zi ) fm (zi )j " (15)

for all i = 1; : : : ; M and for all n; m 2 N with n; m n" . Fix x 2 K. Then x

belongs to B yi ; 2 for some i. In particular,

d (x; zi ) d (x; yi ) + d (y; zi ) < + = :

2 2
Using (14) and (15), we have that

jfn (x) fm (x)j jfn (x) fn (zi )j+jfn (zi ) fm (zi )j+jfm (zi ) fm (x)j "+"+"

for all n; m 2 N with n; m n" , which shows that the sequence ffn (x)g is a
Cauchy sequence in R. Hence, there exists

g (x) := lim fn (x) 2 R:

n!1

Moreover, since
jfn (x) fm (x)j 3"
for all x 2 K and all n; m 2 N with n; m n" , letting m ! 1, we conclude
that
jfn (x) g (x)j 3"
for all x 2 K and all n 2 N with n n" , or, equivalently,

sup jfn (x) g (x)j 3"

x2K

for all n 2 N with n n" , which shows that ffn g converges to g uniformly on
K. In turn, g restricted to K is continuous.
Step 3: Is g de…ned everywhere? Yes, for every x 2 X, take K to be the
singleton fxg. Is g continuous? Yes, this follows from (14).

56
 Wednesday, October 14, 2009

Corollary 107 Let (X; d) be a compact metric space. Then F C (X) is

relatively compact if and only if is bounded and equicontinuous. In particular,
F C (X) is compact if and only if is closed, bounded, and equicontinuous.

Proof. If F is bounded and equicontinuous, then by the previous theorem it

follows that the closure of F is sequentially compact, and so by Theorem 82
the closure of F is compact. Conversely, assume that F C (X) is relatively
compact. Then by Proposition 87, the closure of F is bounded. It remains to
show that F is equicontinuous. Assume, by contradiction, that this is not the
case. Then there exist " > 0, ffn g F, and fxn g ; fyn g X such that

jfn (xn ) fn (yn )j > "

1
and d (xn ; yn ) n . Since X is compact (and so sequentially compact), there
exist a subsequence fxnk g of fxn g and x0 2 X such that d (xnk ; x) ! 0 as k !
1.
n Inoturn, since ffnk g F, again by Theorem 82, there exist a subsequence
fnkj of ffnk g and f0 2 C (X) such that dC(X) fnkj ; f0 ! 0 as j ! 1. By
your homework, fnkj xnkj ! f (x) and fnkj ynkj ! f (x). Hence,

" < fnkj xnkj fnkj ynkj ! jf (x) f (x)j = 0;

which is a contradiction.

Theorem 108 Let 1 p < 1. Then F Lp RN is relatively compact if

and only if F is bounded in Lp RN and for every " > 0 there exist > 0 and
R > 0 such that Z
p
jf (x + h) f (x)j dx "
RN

for all f 2 F and h 2 RN with jhj < and

Z
p
jf (x)j dx "
RN nB(0;R)

for all f 2 F.

1.11 The Stone–Weierstrass Theorem

Another important application of compactness is given by the Stone–Weierstrass
theorem. We begin with some results that are of interest in themselves.

Theorem 109 (Dini) Let (X; d) be a compact metric space and let ffn g
C (X) be a sequence of functions such that

fn (x) fn+1 (x) (16)

57
for every x 2 X and for all n 2 N, and

lim fn (x) =: f (x) 2 R: (17)

n!1

If f : X ! R is continuous, then the convergence is uniform.

Proof. Fix " > 0. By (17) for every x 2 X there exists nx 2 N such that

0 f (x) fn (x) "

for all n nx . Since the functions f and fnx are continuous at x, there exists
B (x; rx ) such that
0 f (y) fnx (y) 3"
for all y 2 B (x; rx ). By (16), we have that

0 f (y) fn (y) 3" (18)

for all y 2 B (x; rx ) and for all n nx . By the compactness of X we may …nd
a …nite number of balls that cover X, say B (x1 ; rx1 ), . . . , B (xm ; rxm ). Let
n" := max fnx1 ; : : : ; nxm g. Then for all y 2 X and n n" we have that y is
contained in one of the balls B (xi ; rxi ), i = 1; : : : ; m, and so (18) holds. In
turn,
max jf (y) fn (y)j 3"
y2X

for all n n" , which gives the desired result.

Example 110 If f is not assumed to be continuous, then the result is false.

Take
fn (x) = xn ; x 2 [0; 1] :
Then fn fn+1 but the convergence is not uniform.
Similarly, if X is not compact, then the result fails. Take X = R and
8
< 1 if x n;
fn (x) = 1 + n x if n < x < n + 1;
:
0 otherwise.

Then fn fn+1 , fn (x) ! f (x) = 1, which is a continuous function, but there

is no uniform convergence.
p
Exercise 111 Let f (x) := x, x 2 [0; 1]. For n 2 N and x 2 [0; 1] de…ne
recursively
1
p0 (x) := 0; pn (x) := pn 1 (x) + x p2n 1 (x) :
2
(i) Prove that each pn is a polynomial.
(ii) Prove that jpn (x)j 1 and for all n 2 N and x 2 [0; 1].

58
(iii) Prove that fpn g converges uniformly to f in [0; 1].

Theorem 112 (Stone) Let (X; d) be a compact metric space and let F
C (X) be a family of functions such that

(i) F separates points; that is, if x; y 2 X with x 6= y, then there exists f 2 F

such that f (x) 6= f (y),
(ii) F contains the constant functions,
(iii) F is an algebra; that is, if f; g 2 F and t 2 R, then f + g, f g, and tf
belong to F.

Then F is dense in C (X).

Proof. Step 1: We claim that F satis…es properties (i)-(iii). We only need

to prove property (iii). Given f; g 2 F and t 2 R, by Remark 30 there exist
ffn g ; fgn g F such that d1 (fn ; f ) ! 0 and d1 (gn ; f ) ! 0. By prop-
erty (iii), fn + gn , fn gn , and tfn belong to F. Since d1 (fn + gn ; f + g) ! 0,
d1 (fn gn ; f g) ! 0, and d1 (tfn ; tf ) ! 0 (exercise), it follows again by Remark
30, that f + g, f g, and tf belong to F. It remains to show that F. = C (X).

59
 Monday, October 19, 2009
Proof. Step 2: We prove that if f belongs to F, then so does jf j. Since X is
compact, by the Weierstrass theorem f is bounded by some constant M > 0.
De…ne
jf (x)j
g (x) := ; x 2 X:
M
Then g (x) 2 [0; 1]. In view of (iii), it su¢ ces to show that g belongs to F. By
the previous exercise there exists a sequence p of polynomials fpn g that converges
uniformly in [0; 1] to the function h (t) := t, t 2 [0; 1]. De…ne
!
2
f (x)
gn (x) := pn ; x 2 X:
M
r
2
f
Then gn converges uniformly in X to the function M = g. Since F is an
algebra, we have that gn 2 F. Hence, using the fact that F is closed, it follows
that g belongs to F.
Step 3: We prove that if f; g belong to F, then so do max ff; gg and min ff; gg.
It is enough to observe that
1
max ff; gg = [f + g + jf gj] ;
2
1
min ff; gg = [f + g jf gj] :
2
Step 4: We prove that if x; y 2 X with x 6= y and ; 2 R, then there exists
g 2 F such that g (x) = and g (y) = . To see this, use property (i) to …nd
f 2 F such that f (x) 6= f (y) and de…ne
(f (z) f (y)) (f (x) f (z))
g (z) := ; z 2 X:
(f (x) f (y))

Step 5: We are now ready to prove that F = C (X). Let f 2 C (X) and " > 0.
By the previous step, for every x; y 2 X there exists a function gx;y 2 F such
that gx;y (x) = f (x) and gx;y (y) = f (y). De…ne
Ux;y := fz 2 X : gx;y (z) < f (z) + "g ;
Vx;y := fz 2 X : gx;y (z) > f (z) "g :
By the continuity of gx;y and f we have that Ux;y and Vx;y are open sets contain-
ing x and y. Since fUx;y gx2X is an open cover of X, it follows by compactness
that there exist x1 ; : : : ; xmy 2 X such that
my
[
Uxi ;y = X: (19)
i=1

De…ne n o
gy := min gx1 ;y ; : : : ; gxmy ;y :

60
Then gy belongs to F by Step 3 and by (19) and the de…nition of Uxi ;y and
Vxi ;y ,

gy (z) < f (z) + " for all z 2 X; (20)

my
\
gy (z) > f (z) " for all z 2 Vy := Vxi ;y : (21)
n=1

Since Vy is open and contains y, the family fVy gy2X is an open cover of X.
Again by compactness, there exist y1 ; : : : ; yn 2 X such that
n
[
Vyi = X:
i=1

De…ne
g := max fgy1 ; : : : ; gyn g :
Then g belongs to F by Step 3 and by (20) and (21),

f (z) " < g (z) < f (z) + " for all z 2 X:

Hence maxz2X jf (z) g (z)j ". Since g 2 F, we may …nd h 2 F such that
maxz2X jh (z) g (z)j ", and thus, by the triangle inequality, maxz2X jf (z) h (z)j
2". This concludes the proof.

Exercise 113 (Weierstrass) Let K RN be a compact set. Prove that every

continuous function f : K ! R is the uniform limit in K of a sequence of
polynomials.

Corollary 114 Let (X; d) be a compact metric space. Then C (X) is separable.

Proof. Since X is separable by Exercise 85, there exists a sequence fxn g X

such that fxn g = X. For every n de…ne

fn (x) := d (x; xn ) ; x 2 X:

Then fn is continuous. We claim that ffn g separates points. Indeed, assume

the contrary. Then there exist x; y 2 X such that fn (x) = fn (y) for every
n 2 N. By density we may …nd a subsequence fxnk g X of fxn g such that
xnk ! x. Hence,

d (y; xnk ) = fnk (y) = fnk (x) = d (x; xnk ) ! 0

as k ! 1. Thus, xnk ! y. By the uniqueness of limits, it follows that x = y.

This proves the claim.
De…ne f0 := 1 and for every k 2 N and n1 ; : : : ; nk 2 N0 de…ne

fn1 ;:::;nk (x) := fn1 (x) fnk (x) ; x 2 X:

61
Consider the family F given by all …nite linear combinations of functions of the
form fn1 ;:::;nk . Then F satis…es the hypotheses of Stone’s theorem, and so F is
dense in C (X). On the other hand, the family F 0 given by all …nite rational
linear combinations of functions of the form fn1 ;:::;nk is countable. For every
C (X) and " > 0 we may …nd g 2 F such that

d1 (f; g) ":

Since g is a …nite linear combinations of functions of the form fn1 ;:::;nk , using
the density of the rationals in the real, we may …nd h 22 F such that

d1 (h; g) ":

This shows that F 0 is dense in C (X) and, in turn, that C (X) is separable.

Exercise 115 Prove that Cb (R) is not separable.

62
 Wednesday, October 21, 2009

2 Topological Spaces
De…nition 116 Let X be a nonempty set. A collection P (X) is a topology
if the following hold.

(i) ;; X 2 .

(ii) If Ui 2 for i = 1; : : : ; M , then U1 \ : : : \ UM 2 .

S
(iii) If fU g 2 is an arbitrary collection of elements of , then 2 U 2 .

The pair (X; ) is called a topological space and the elements of are open
sets. For simplicity, we often apply the term topological space only to X.

Example 117 (i) Given a nonempty set X, the smallest topology consists of
f;; Xg, while the largest topology contains all subsets as open sets.
(ii) Given a metric space (X; d) the family of open sets is a topology.

(iii) Given a topological space (X; X) and an onto function f : X ! Y , the

family of sets
1
Y := E Y : f (E) 2 X

is a topology on Y . It is called the quotient topology (relative to f and

X ). We now consider an important special case. Given a nonempty set
X, let be an equivalence relation. We de…ne

Y =X := f[x] : x 2 Xg

and consider the projection of X onto Y

P :X!Y
x 7! [x]

If X is a topological space with topology , we can consider in Y the

quotient topology (relative to P and ), precisely,
1
Y := E Y : P (E) 2 X :

Note that [
1
P (E) = fx 2 X : [x] 2 Eg = [x] ;
[x]2E

that P 1 (E) is is given by the union of the equivalence classes belonging to

E. Thus, an open set in the quotient topology is a collection of equivalence
classes whose union is an open set of X.

63
 As Example 117(i) shows, a set X can have more than one topology. Note
that if is any topology on the set X, then the inclusions

f;; Xg P (X)

hold. If 1 and 2 are two topologies on X, we say that 1 is weaker, or coarser,

than 1 if 1 2.

Remark 118 Given a family of topologies f g 2 on a set X, the family of

sets \
:= fU X : U 2 for every 2 g
2
S
is still a topology on X, while in general 2 is not.

In view of the previous remark, we can construct topologies starting from

any family of subsets of X.

Proposition 119 Let X be a set and let F be a family of subsets of X. Then

there exists a unique, smallest topology containing F. Moreover, consists
of X, ;, …nite intersections of elements of F and arbitrary unions of …nite
intersections of elements of F.

The family F is called a subbase for and is said to be generated by F.

Proof. Let f g 2 be the family of all topologies that contain F. Note that
this family is nonempty since P (X) is one such topology. Then
\
:=
2

is still a topology, contains F, and it is the smallest such topology. Moreover,

it is unique in view of its de…nition. By the properties of a topology, we have
that contains X, ;, all …nite intersections of elements of F and all arbitrary
unions of …nite intersections of elements of F. On the other hand, let 0 be
the family that consists of X, ;, all …nite intersections of elements of F and
arbitrary unions of …nite intersections of elements of F. Then 0 is a topology.

Example 120 (i) In R we can consider the family F = f(a; 1) : a 2 Rg [

f( 1; b) : b 2 Rg. The smallest topology containing F is the standard
one.
(ii) Given two topological spaces (X; X ) and (Y; Y ), by considering the family
F = fU V : U 2 X ; V 2 Y g, the smallest topology containing F on
X Y is called the product topology.

Given a point x 2 X, a neighborhood 3 of x is an open set U 2 that contains

x. Given a set E X, a neighborhood of E is an open set U 2 that contains
E.
3 In some texts the de…nition of neighborhood is di¤erent.

64
 Given a topological space (X; ) and a set E X, a point x 2 E is called
an interior point of E if there exists a neighborhood U of x such that U E.
The interior E of a set E X is the union of all its interior points.
The proof of following proposition is left as an exercise.

Proposition 121 Let (X; ) be a topological space and let E X. Then

(i) E is an open subset of E,

(ii) E is given by the union of all open subsets contained in E; that is, E is
the largest (in the sense of union) open set contained in E,

(iii) E is open if and only if E = E ,

(iv) (E ) = E .

We now introduce the notion of a base for a topology. Let (X; ) be a

topological space. A family of open sets of X is a base for the topology if
every open set U 2 may be written as the union of elements of .

65
 Friday, October 23, 2009
Given a topological space (X; ) and a point x 2 X, a family x of neighbor-
hoods of x is a local base at x if every neighborhood of x contains an element
of x .

Proposition 122 Let X be a nonempty set and let P (X) be a family of

sets. Then is a base for a topology if and only if

(i) it contains the empty set;

(ii) for every x 2 X there exists B 2 such that x 2 B,

(iii) for every B1 ; B2 2 with B1 \ B2 6= ; and for every x 2 B1 \ B2 there

exists B3 2 such that x 2 B3 and B3 B1 \ B2 .

Proof. Assume that is a topology and that is a base for . Then every
open set is a union of sets of . In particular, the empty set and X can be
written as union of sets of , and so (i) and (ii) hold. To prove (iii), note that if
B1 ; B2 2 , then B1 \ B2 is open, and so it can be written as union of elements
of , say [
B1 \ B2 = B

Hence, if x 2 B1 \ B2 , then there is a B such that x 2 B B1 \ B2 .

Conversely, let = fB g 2 be a family of sets satisfying (i)-(iii) and let
be given by arbitrary unions of elements of . We need to show that is a
topology. By property (i), we have that the empty set belongs to , while by
property (ii) we have that [
X= B ;
2

and so X 2 .
If Ui 2 for i = 1; : : : ; M , then we can write
[
Ui = B
2 i

for some i . Then

M [
\
U1 \ : : : \ UM = B :
i=1 2 i

If x 2 U1 \ : : : \ UM , then there exist i 2 i such that x 2 B 1 \ : : : \ B M . By

property (iii) and an induction argument we may …nd Bx 2 such that x 2 Bx
and Bx B 1 \ : : : \ B M . Hence
[
U1 \ : : : \ UM = Bx 2 :
x2U1 \:::\UM i

66
Finally, given an arbitrary collection fU g 2 of elements of , since each U
S
is a union of elements of , we have that 2 U is a union of elements of ,
and so it belongs to .
Thus, is a topology. The fact that is a base for follows from the
de…nition of .

Example 123 We give some examples of bases.

(i) Given a metric space (X; d), consider the family

= fB (x; r) : x 2 X; r > 0g :

Then properties (i) and (ii) are satis…ed, and so by the previous proposition
we obtain a topological space (X; ) by taking as open sets arbitrary unions
of open balls.
(ii) Given a pseudometric space (X; ), we can construct a topology on X
by taking for every x 2 X the local base of all pseudoballs centered at x
and of radius r > 0,
fy 2 X : (x; y) < rg :

(iii) Given a nonempty set X, we recall that RX denotes the space of all func-
tions f : X ! R. We construct a local base for a topology. For every
f 2 RX let r > 0 and let Y X be a …nite subset. Consider

B (f ; r; Y ) := g 2 RX : jg (x) f (x)j < r for all x 2 Y :

The family = B (f ; r; Y ) : f 2 RX ; r > 0; Y X …nite is a base for

a topology on RX . See Example 14.
(iv) Given a set X and a family F of subsets of X, let be the smallest
topology containing F. Then a basis for is given by …nite intersections
of elements of F.

Exercise 124 Let (X; ) be a pseudometric space. Let Y = X be as in

Exercise 15.

(i) Prove that a set U X is open (with respect to the topology X) if and
only if the set P (U ) is open (with respect to the metric d).
(ii) Prove that the topology induced by the metric d is the quotient topology.

Exercise 125 For every (x; y) 2 R2 consider the family of rectangles [x; x + r)
[y; y + t], where r; t > 0. Consider the family

= [x; x + r) [y; y + t] : (x; y) 2 R2 ; r; t > 0 :

Prove that is a base for a topology on R2 that is not the canonical topology
on R2 .

67
Exercise 126 Given a nonempty set X, consider two metrics d1 : X X !
[0; 1) and d2 : X X ! [0; 1). What are the relations, if any, among the
following properties?

(i) d1 and d2 are equivalent, that is, for every sequence fxn g X and every
x 2 X,

lim d1 (xn ; x) = 0 if and only if lim d2 (xn ; x) = 0:

n!1 n!1

(ii) For every x 2 X and r > 0 there exist r1 ; r2 > 0 such that

Bd2 (x; r1 ) Bd1 (x; r) Bd2 (x; r2 ) :

(iii) d1 and d2 generate the same topology.

A set C X is closed if its complement X n C is open. The next proposition

follows from De Morgan’s law.

Proposition 127 Let (X; ) be a topological space. Then

(i) ; and X are closed.

(ii) If Ci X, i = 1; : : : ; n, is a …nite family of closed sets of X, then C1 [
\ Un is closed.
T
(iii) If fC g 2 is an arbitrary collection of closed sets of X, then 2 C
is closed.

Let (X; ) be a topological space. The closure E of a set E X is the

smallest closed set that contains E. A subset E of a topological space X is
said to be dense if its closure is the entire space, i.e., E = X. We say that a
topological space is separable if it contains a countable dense subset.

Proposition 128 Let (X; ) be a topological space and let E X. Then x 2 E

if and only if E \ U is nonempty for every neighborhood U of x.

Proposition 129 Let (X; ) be a topological space and let fE g be an arbi-

trary family of subsets of X. Then
[ [
U U

and equality holds if the family fE g is …nite.

The proof of following proposition is left as an exercise.

Proposition 130 Let (X; ) be a topological space and let C X. Then C is

closed if and only if C = C.

68
 Monday, October 26, 2009
A point x0 2 X is an accumulation point for a set E X if for every open set
U that contains x0 there exists x 2 E \U , with x 6= x0 . The set of accumulation
points of E is denoted acc E. Exactly as in the case of metric spaces we have
the following proposition.

Proposition 131 Let (X; ) be a topological space and let E X. Then

E = E [ acc E:

In particular, a set C X is closed if and only if C contains all its accumulation

points.

2.1 Inadequacy of Sequences

A topological space (X; ) is a Hausdor¤ space if for any x; y 2 X with x 6= y
we may …nd two disjoint neighborhoods of x and y.

Proposition 132 Let (X; d) be a metric space and let be the topology deter-
mined by d. Then (X; ) is a Hausdor¤ space.

Proof. If x 6= y, then B x; d(x;y)

2 and B y; d(x;y)
2 are disjoint neighborhoods
of x and y, respectively.
Given a topological space (X; ) and a sequence fxn g, we say that fxn g
converges to a point x 2 X if for every neighborhood U of x we have that
xn 2 U for all n su¢ ciently large. Note that unless the space is Hausdor¤, the
limit may not be unique.

Proposition 133 (Uniqueness of limits) Let (X; ) be a Hausdor¤ space.

If fxn g converges to x and to y, then x = y.

Proof. Assume that x 6= y. Then there exist two disjoint neighborhoods U and
V of x and y, respectively. Since fxn g converges to x we have that xn 2 U for
all n su¢ ciently large. This implies that xn 2
= V for all n su¢ ciently large, and
so fxn g cannot converge to y.
Given a topological space (X; ), a subset C X is sequentially closed if
for every sequence fxn g C such that fxn g converges to some x 2 X, then x
belongs to C.

Proposition 134 Let (X; ) be a topological space and let C X be a closed

set. Then C is sequentially closed.

Proof. Assume that C is closed and let fxn g C be such that fxn g converges
to some x 2 X. We need to show that x belongs to C. If not, then x 2 X n C.
Since X n C is open, there exists a neighborhood U of x such that U X n C.
But then xn 2 U for all n su¢ ciently large, which implies that xn 2 X n C for
all n large. This contradicts the fact that fxn g C.

69
 Given a topological space (X; ) and a set E X, the sequential closure of
E is the set
seq
E := fx 2 X : there exists fxn g E converging to xg :

Proposition 135 Let (X; ) be a topological space and let E X set. Then
seq
E E E:

Proof. Since E is closed, by the previous proposition it is sequentially closed.

seq
Hence, E E.
seq
The next example shows that strict inclusion in E E is possible.

Exercise 136 Let R[0;1] ; be as in Example 123 and consider the subset E
of R[0;1] that consists of all functions f that take value zero at a …nite number
of x 2 [0; 1] and that take value 1 otherwise.

(i) Prove that the function f0 0 belongs to E.

seq
=E
(ii) Prove that f0 2 .

(iii) Prove that there is no metric on R[0;1] compatible with the topology .

De…nition 137 Let (X; ) be a topological space.

(i) The space X satis…es the …rst axiom of countability if every x 2 X admits
a countable local base.

(ii) The space X satis…es the second axiom of countability if it has a countable
base.

Example 138 Consider an uncountable set X with the discrete topology .

Then X satis…es the …rst axiom of countability, since for every x 2 X, the
singleton fxg is a local base. However, X does not have a countable base, since
singletons belong to and are uncountable.

Example 139 A metric space (X; d) satis…es the satis…es the …rst axiom of
countability (take as a local base at x 2 X the family of balls B x; n1 n2N ,
but not necessarily the second (see the next exercise).

Exercise 140 Let (X; ) be a topological space satisfying the second axiom of
countability. Prove that X is separable.

Proposition 141 Let (X; ) be a topological space satisfying the …rst axiom of
countability and let E X. Then
seq
E = E:

70
 seq
Proof. In view of Proposition 135, it remains to show that E E. Let
x 2 E and let fBn gn be a local base at x. By replacing Bn with B1 \ \ Bn ,
we may assume that fBn gn is a decreasing sequence. By Proposition 128 we
have that E \ Bn is nonempty, and so there exists xn 2 E \ Bn . We claim that
the sequence fxn g converges to x. Indeed, let U be a neighborhood of x. Since
fBn gn is a local base at x, there exists n 2 N such that Bn U . Using the fact
that fBn gn is a decreasing sequence, we have that

xn 2 Bn Bn U

for all n n, which shows that fxn g converges to x.

71
 Wednesday, October 28, 2009

2.2 Limit, Limsup, and Liminf

De…nition 142 Let (X; X ) and (Y; Y ) be two topological spaces and let f :
E ! Y , where E Y . Given x0 2 acc E, if there exists y0 2 Y with the
property that for every neighborhood V Y of y0 , there exists a neighborhood
U X of x0 such that
f (x) 2 V
for all x 2 U \ (E n fx0 g), we write
y0 = lim f (x)
x!x0

and we say that y0 is the limit of f as x approaches x0 .

Note that x0 need not belong to E.
Remark 143 In applications it is enough to take V in a local base of y and U
in a local base of x0 .
Proposition 144 Let (X; X ) and (Y; Y ) be two topological spaces with Y a
Hausdor¤ space and let f : E ! Y , where E Y . Given x0 2 acc E, if there
exist limx!x0 f (x) = y1 and limx!x0 f (x) = y2 , then y1 = y2 .
Proof. Assume by contradiction that there y1 6= y2 . Then there exist two
disjoint neighborhoods V1 and V2 of y1 and y2 , respectively. In turn, by the
de…nition of limit, there exist two neighborhoods U1 and U2 of x0 such that
f (x) 2 V1 for all x 2 U1 \ (E n fx0 g) and f (x) 2 V2 for all x 2 U2 \ (E n fx0 g).
Since x0 2 acc E, it follows by Proposition 128 that there exists x 2 U1 \ U2 \
(E n fx0 g). But then f (x) 2 V1 \ V2 , which contradicts the fact that V1 and V2
are disjoint.
We have seen that for metric spaces, when working with limits, it is enough
to work with sequences. For topological spaces, this is no longer true, although
we have one implication.
Proposition 145 Let (X; X ) and (Y; Y ) be two topological spaces and let f :
E ! Y , where E X. Given x0 2 acc E, if there exists limx!x0 f (x) = y0 ,
then f (xn ) ! y0 for every sequence fxn g E n fx0 g that converges that to x0 .
Proof. Let fxn g E n fx0 g converge to x0 . Given a neighborhood V Y of
y0 , there exists a neighborhood U X of x0 such that
f (x) 2 V
for all x 2 U \ (E n fx0 g). Since xn ! x0 , there exists n" 2 N such that xn 2 U
for all n n" and so
f (xn ) 2 V
for all n n" , which shows that f (xn ) ! f (x0 ).
The converse is false in general.

72
Proposition 146 Let (X; X ) be a topological space satisfying the …rst axiom
of countability, let (Y; Y ) be a topological space, and let f : E ! Y , where
E X. Given x0 2 E \ acc E, if there exists y0 2 Y such that f (xn ) ! y0
for every sequence fxn g E n fx0 g that converges that to x0 , then there exists
limx!x0 f (x) = y0 .

Proof. We claim that limx!x0 f (x) = y0 . If not, then there exists a neighbor-
hood V Y of y0 such that for every neighborhood U X of x0 there exists
x 2 U \ (E n fx0 g) such that f (x) 2 = V . Since (X; ) satis…es the …rst axiom
of countability, there exists a countable local base fBn gn at x0 . By selecting
a subsequence, we may assume that Bn+1 Bn for every n 2 N. Then for
every n we may …nd xn 2 Bn \ (E n fx0 g) such that f (xn ) 2 = V . Since fBn gn
is a decreasing local base at x0 , the sequence fxn g converges to x0 . By hy-
pothesis, f (xn ) ! y0 , and so f (xn ) 2 V for all n su¢ ciently large, which is a
contradiction.
In the special case in which Y = R or Y = [ 1; 1] all the standard theorems
about the sum, product, quotient of limits continue to hold with the standard
modi…cations. We omit the details.

De…nition 147 Let (X; ) be a topological space, let f : E ! R, where E R,

and let x0 2 acc E. The limit inferior of f as x tends to x0 is de…ned as

lim inf f (x) := sup inf f (x) ;

x!x0 U 2 (x0 ) x2U \(Enfx0 g)

while the limit superior of f as x tends to x0 is de…ned as

lim sup f (x) := inf sup f (x) ;

x!x0 U 2 (x0 ) x2U \(Enfx0 g)

where (x0 ) stands for the collection of all neighborhoods of x0 .4

Remark 148 If U and V are two neighborhoods of x0 with U V , then

inf f (x) inf f (x) ;

x2V \(Enfx0 g) x2U \(Enfx0 g)

and since we are interested in the supremum over all neighborhoods of x0 , we

can neglect V . Thus, we can focus on “small” neighborhoods of x0 (this is the
analogous of " > 0 small in the standard de…nition of limits). In particular, we
could replace (x0 ) with a local base at x0 in the de…nition of lim inf . A similar
x!x0
reasoning works for lim sup. This is why in a metric space we only consider
x!x0
balls.
4 In several books, sup
A2 (x0 ) inf x2A\E f (x) is used as a de…nition for the limit inferior
lim inf x!x0 f (x). The de…nition we use here is in accordance with the de…nition of limit, in
which the value of the function at the point x0 plays no role. In particular, with our de…nition
we recover the fact that the limit exists at x0 if and only if the limit inferior and superior
coincide, while with the other de…nition one would get that the limit inferior and superior at
x0 coincide if and only if the function f is continuous at f0 .

73
Theorem 149 Let (X; ) be a topological space, let f : E ! R, where E R,
and let x0 2 acc E. Then

lim inf f (x) lim sup f (x) : (22)

x!x0 x!x0

Moreover there exists lim f (x) if and only if equality holds in (22), and in this
x!x0
case the limit coincides with the common value in (22).

Proof. Let U and V be two neighborhoods of x0 . Then

inf f (x) inf f (x)

x2U \(Enfx0 g) x2U \V \(Enfx0 g)

sup f (x) sup f (x) :

x2U \V \(Enfx0 g) x2V \(Enfx0 g)

Taking the supremum over all U 2 (x0 ) gives

lim inf f (x) sup f (x) :

x!x0 x2V \(Enfx0 g)

Taking the in…mum over all V 2 (x0 ) gives (22).

74
 Friday, October 30, 2009
Proof. To prove the second part of the theorem assume that there exists
lim f (x) = `. I will consider only the case ` 2 R and leave the cases ` = 1
x!x0
and ` = 1 as an exercise. By the de…nition of limit, for every " > 0 there
exists a neighborhood U" of x0 such that

` " f (x) `+"

for all x 2 U" \ (E n fx0 g). Hence,

` " inf f (x) sup inf f (x) = lim inf f (x)

x2U" \(Enfx0 g) U 2 (x0 ) x2U \(Enfx0 g) x!x0

lim sup f (x) inf sup f (x) sup f (x) ` + ":

x!x0 U 2 (x0 ) x2U \(Enfx0 g) x2U" \(Enfx0 g)

Letting " ! 0+ , we conclude that

lim inf f (x) = lim sup f (x) = `:

x!x0 x!x0

Conversely, assume that

lim inf f (x) = lim sup f (x) = L

x!x0 x!x0

for some L 2 [ 1; 1]. Again we consider the case L 2 R and leave the cases
L = 1 and L = 1 as an exercise. Fix " > 0. By the de…nition of lim inf f (x)
x!x0
and lim sup f (x) there exist two neighborhoods U" and V" of x0 such that
x!x0

L " inf f (x) ;

x2U" \(Enfx0 g)

sup f (x) L + ":

x2V" \(Enfx0 g)

Taking U := U" \ V" , we have that for all x 2 U \ (E n fx0 g),

L " f (x) L + ";

which shows that there exists lim f (x) = L.

x!x0

Exercise 150 Let (X; ) be a topological space, let f : E ! R and g : E ! R,

where E R, and let x0 2 acc E. Assume that one of the two functions is
bounded. Prove that

lim inf f (x) + lim inf g (x) lim inf (f (x) + g (x)) lim sup f (x) + lim inf g (x)
x!x0 x!x0 x!x0 x!x0 x!x0

lim sup (f (x) + g (x)) lim sup f (x) + lim sup g (x)
x!x0 x!x0 x!x0

75
and that in general all inequalities may be strict. Prove that if there exists
lim f (x) = ` 2 R, then
x!x0

lim inf f (x) + lim inf g (x) = lim inf (f (x) + g (x)) ;
x!x0 x!x0 x!x0

lim sup (f (x) + g (x)) = lim sup f (x) + lim sup g (x) :
x!x0 x!x0 x!x0

The proof of the following corollary is left as an exercise.

Corollary 151 (Cauchy) Let (X; ) be a topological space, let f : E ! R,

where E R, and let x0 2 acc E. Then a necessary and su¢ cient condition for
lim f (x) to exist in R is that for every " > 0 there exists a neighborhood U" of
x!x0
x0 such that
jf (x1 ) f (x2 )j "
for all x1 ; x2 2 U" \ (E n fx0 g).

2.3 Continuity
De…nition 152 Let (X; X ) and (Y; Y ) be two topological spaces, and let f :
E ! Y , where E Y . Given x0 2 E, the function f is said to be continuous
at x0 if for every neighborhood V Y of f (x0 ), there exists a neighborhood
U X of x0 such that
f (x) 2 V
for all x 2 U \ E. The function f is said to be continuous if it is continuous at
every point of E.

Remark 153 Note that if x0 2 E is an isolated point of E; that is, if there

exists a neighborhood U X of x0 such that U0 \E = fx0 g, then f is continuous
at x0 (take U = U0 ). Thus, it is enough to check the continuity of a function
f : E ! Y at points x0 2 E \ acc E, in which case continuity at x0 is equivalent
to saying that there exists the limit

lim f (x) = f (x0 ) :

x!x0

The space of all continuous functions f : X ! Y is denoted by C (X; Y ).

If Y = R we write C (X). If Y is a metric space, the space of all continuous
bounded functions f : X ! Y is denoted Cb (X; Y ). It is a metric space with
the distance
d1 (f; g) := sup dY (f (x) ; g (x)) :
x2X

The proof of the next two propositions follows from Propositions 145 and 146.

Proposition 154 Let (X; X ) and (Y; Y ) be two topological spaces and let f :
E ! Y , where E X. Given x0 2 E \ acc E, if f is continuous at x0 , then
f is sequentially continuous at x0 ; that is, f (xn ) ! f (x0 ) for every sequence
fxn g E that converges that to x0 .

76
Proposition 155 Let (X; X ) be a topological space satisfying the …rst axiom
of countability, let (Y; Y ) be a topological space, and let f : E ! Y , where
E X. Given x0 2 E \ acc E, if f is sequentially continuous at x0 , then f is
continuous at x0 .

Exercise 156 Consider two topological spaces (X; X ) and (Y; Y ) and a func-
tion f : X ! Y . Prove that the following are equivalent.

(i) f is continuous.
1
(ii) f (U ) is open for every open set U Y.
1
(iii) f (C) is closed for every closed set C Y.
1
(iv) f (B) is open for every set B Y in a base (or subbase) of Y .

Exercise 157 Consider three topological spaces (X; X ), (Y; Y ), and (Z; Z )
and two continuous functions f : X ! Y and g : Y ! Z. Prove that g f :
X ! Z is continuous.

De…nition 158 Given two topological spaces (X; X ) and (Y; Y ), a function
f : X ! Y is said to be a homeomorphism if f is one-to-one, onto, and
continuous and f 1 : Y ! X is continuous. In this case, the topological spaces
(X; X ) and (Y; Y ) are said to be homeomorphic.

Given a topological space (X; ), a topological property is a property that if

possessed by X, is possessed by all spaces homeomorphic to X.

De…nition 159 Given two topological spaces (X; X ) and (Y; Y ), a function
f : X ! Y is said open if f (U ) is open for every open set U X. Similarly,
f : X ! Y is said closed if f (C) is closed for every closed set C X.

Thus, if f : X ! Y is f is one-to-one and onto, then f is a homeomorphism

if and only if f is continuous and open (or continuous and closed).

2.4 Lower Semicontinuity

De…nition 160 Let (X; ) be a topological space and let f : E ! R be a func-
tion, where E X.

(i) The function f is said to be lower semicontinuous at a point x0 2 E if

either x0 is an isolated point or x0 2 acc E and

lim inf f (x) f (x0 ) :

x!x0

The function f is said to be lower semicontinuous if it is lower semicon-

tinuous at every point of E.

77
 (ii) The function f is said to be upper semicontinuous at a point x0 2 E if f
is lower semicontinuous at x0 . The f is said to be upper semicontinuous
if it is upper semicontinuous at every point of E.

We recall x0 2 E is an isolated point of E if there exists a neighborhood

U X of x0 such that U0 \ E = fx0 g.

Example 161 Given the function

sin x1 if x 6= 0;
f (x) =
a if x = 0;

we have that
lim inf f (x) = 1 and lim sup f (x) = 1:
x!0 x!0

Thus f is lower semicontinuous at 0 if and only if a 1, f is upper semicon-

tinuous at 0 if and only if a 1, while if 1 < a < 1, then f is neither lower
nor upper semicontinuous at 0.

The following characterizations of lower semicontinuity will be of use in the

sequel. We recall the de…nition of epigraph and of limit inferior.

De…nition 162 Let (X; ) be a topological space and let f : E ! R, where

E R. The epigraph of f is the set

epi f := f(x; t) 2 E R : f (x) tg :

Proposition 163 Let X be a topological space and let f : X ! R. Then the

following are equivalent:

(i) fx 2 X : f (x) tg is closed for every t 2 R;

(ii) epi f is closed;
(iii) f is lower semicontinuous.

Proof. Step 1: We prove that (i) is equivalent to (ii). Assume that (i) holds
and let
D := (X R) n epi f = f(x; t) 2 X R : f (x) > tg : (23)
Fix (x0 ; t0 ) 2 D and let 0 < " < f (x0 ) t0 . Then the set
1
U := f ((t0 + "; 1])

is open and contains x0 , and so U (t0 "; t0 + ") is a neighborhood of (x0 ; t0 ).

If (x; t) 2 U (t0 "; t0 + ") then f (x) > t0 + " > t, which implies that
(x; t) 2 D. Hence D is open and its complement, epi f , is closed.
Conversely, assume that epi f is closed (and so the set D de…ned in (23) is
open), and for t0 2 R consider the set

U := fx 2 X : f (x) > t0 g :

78
If x0 2 U then for any …xed 0 < "0 < f (x0 ) t0 the pair (x0 ; t0 + "0 ) belongs
to the open set D, and so we may …nd a neighborhood U0 X of x0 and
0 < " "0 such that

U0 (t0 + "0 "; t0 + "0 + ") D:

In particular, if x 2 U0 then f (x) > t0 + "0 " > t0 , and so U0 U , which

shows that U is open and in turn proves (i)
Step 2: We show that (i) is equivalent to (iii). Let f satisfy (i) and assume by
contradiction that there exists x0 2 X such that

f (x0 ) > lim inf f (x) :

x!x0

Fix f (x0 ) > t > lim inf x!x0 f (x). Then the open set

Ut := fx 2 X : f (x) > tg (24)

belongs to (x0 ), and so

lim inf f (x) = sup inf f (x) inf f (x) t:

x!x0 U 2 (x0 ) x2U nfx0 g x2Ut nfx0 g

which is a contradiction.
Conversely, assume that (iii) holds and …x t 2 R. If the set Ut de…ned in
(24) is nonempty, …x any x0 that belongs to it. Since

lim inf f (x) f (x0 ) > t;

x!x0

by the de…nition of limit inferior there exists U 2 (x0 ) such that

inf f (x) > t;

x2U nfx0 g

which, together with the fact that f (x0 ) > t, implies that U Ut . Hence (i)
holds.

79
 Monday, November 2, 2009

2.5 Compactness
De…nition 164 Let (X; ) be a topological space.

(i) A set K X is compact if for every open S cover of K, i.e., for every
collection fU g of elements of such that U K, there exists a …nite
subcover (i.e., a …nite subcollection of fU g whose union still contains K).
(ii) A set K X is sequentially compact if for every sequence fxn g of ele-
ments of K, there exists a subsequence converging to an element of K.
(iii) A set E X is relatively compact (or precompact) if its closure E is
compact.

Remark 165 Consider a nonempty set X and let 1 and 2 be two topologies
on X with 1 2 . If K X is compact with respect to 2 , then K X is
compact with respect to 1 . The less open sets we have, the easier it becomes for
a set to be compact. This remark will be important in Functional Analysis and
it is related to Remark 95.

Proposition 166 Let (X; ) be a Hausdor¤ space, let K X be a compact set

and let x0 2 X n K. Then there exist two disjoint open sets U and V such that
U contains K and x0 2 V .

Proof. Since X is Hausdor¤, for every x 2 K there is a neighborhood Ux of x

and a neighborhood Vx of x0 such that Ux \ Vx = ;. Since fUx gx2X is an open
cover of K, by the compactness of K there exist x1 ; : : : ; xm 2 K such that

U := Ux1 [ [ Uxm K:

The open set

V := Vx1 \ \ V xm
is a neighborhood of x0 and does not intersect U .

Proposition 167 Let (X; ) be a topological space and let K X be a compact

set.

(i) If C K is closed, then C is compact.

(ii) If X is a Hausdor¤ space, then K is closed.

Proof. The proof of (i) is the same as the proof of Proposition 87. It remains
to prove (ii). By the previous proposition for every x0 2 X n K there exists
a neighborhood U of x0 that does not intersect K. This shows that x0 is an
interior point of X n K and, in turn, that X n K is open.
Note that if the topology is not Hausdor¤ then compact sets may not be
closed.

80
Example 168 Given a nonempty set X endowed with the smallest topology,
any nonempty set strictly contained in X is compact but not closed.

We have seen that for metric spaces, compactness and sequential compact-
ness are the same. For topological spaces, these two notions are not related.
[0;1]
Exercise 169 Consider the space [0; 1] of all functions f : [0; 1] ! [0; 1]
with the topology introduced in Example 123. We will see later that this space
[0;1]
is compact (see Tychono¤ ’s theorem). Prove that [0; 1] is not sequentially
compact.

81
 Wednesday, November 4, 2009

Exercise 170 Let X be the family of all nonempty countable subsets of R.

Given E; F 2 X, de…ne
8
< min f1; dist (E n F; F )g if F $ E;
(E; F ) := 0 if F = E;
:
1 otherwise.

(i) Prove that (X; ) is a pseudometric space.

(ii) Let fEn g X be a sequence such that En Q for every n 2 N. Prove

that fEn g converges to
1
[
E := En :
n=1

(iii) Prove that X is sequentially compact.

(iv) For every E 2 X, prove that the set P (E) n f;g X is open in the
topology determined by .
(v) Prove that X is not compact.

Exercise 171 Let X be the …rst uncountable ordinal with its well-ordering.
De…ne
0 if x y;
(x; y) :=
1 if x < y:

(i) Prove that (X; ) is a pseudometric space.

(ii) Prove that X is sequentially compact. Hint: prove that every sequence
contains a monotone subsequence.
(iii) Prove that X is not compact.

Proposition 172 Let (X; X ) be a topological space and let K X be a com-

pact set. Then every in…nite set of K has an accumulation point.

Proof. Let E K be an in…nite set and assume by contradiction that E

has no accumulation points. Then E is closed. Moreover, for every x 2 E
there exists a neighborhood Ux of x such that Ux \ E = fxg. The family of sets
fUx gx2E [fX n Eg covers the compact set K. Hence, there exist x1 ; : : : ; xm 2 E
such that
Ux1 [ [ Uxm [ (X n E) K E:
Since X n E does not intersect E and each Uxi intersects E only at the point xi ,
it follows that E cannot have more than m elements, which is a contradiction.

82
Proposition 173 Let (X; X ) be a topological space satisfying the …rst axiom
of countability and let K X be closed and compact. Then K is sequentially
compact. In particular, if X is also Hausdor¤ , then every compact set is se-
quentially compact.

Proof. Consider a sequence fxn g K and E := fxn : n 2 Ng. By the previ-

ous proposition, either the set E is …nite, in which case one element is repeated
in…nitely many times (so we have a convergent subsequence) or it has an accu-
mulation point x0 . Since (X; ) satis…es the …rst axiom of countability, there
exists a countable local base fBk gk at x0 . By selecting a subsequence, we may
assume that Bk+1 Bk for every k 2 N. Since x0 is ana accumulation point
of E, for every k we may …nd nk 2 N such that xnk 2 Bk \ E. Since fBk gk is
a decreasing local base at x0 , the sequence fxnk g converges to x0 . Since K is
closed and E K, it follows that x0 2 acc E E K = K, which shows that
x0 2 X.
We now show that Weierstrass’s theorem continues to hold for lower semi-
continuous functions. This fact alone explains the importance of this class of
functions in minimization problems.
Let (X; ) be a topological space, let E X and let f : E ! R. The
function f is said to be sequentially lower semicontinuous at some x0 2 E if

lim inf f (xn ) f (x0 )

n!1

for every sequence fxn g E such that xn ! x0 as n ! 1.

Theorem 174 (Weierstrass) Let (X; ) be a topological space, let K X be

compact (respectively sequentially compact), and let f : X ! R be a lower semi-
continuous (respectively sequentially lower semicontinuous, that is) function.
Then there exists x0 2 K such that

f (x0 ) = min f (x) :

x2K

Proof. The proof is the same of the ones we gave for Weierstrass theorem in
metric spaces.

Theorem 175 The Cartesian product of two compact topological spaces is a

compact topological space.

Proof. Let (X; X ) and (Y; Y ) be two compact topological spaces. Assume by
contradiction that X Y is not compact. Then there is an open cover W of
X Y with the property that no …nite subfamily of W covers X Y .
Step 1: We claim that there is x0 2 X such that for every neighborhood U of
x0 , no …nite subfamily of W covers U Y . If this is not the case, then for all
x 2 X there is a neighborhood Ux of x and a …nite subfamily of W that covers
Ux Y . Since fUx gx2X is an open cover of X, by the compactness of X there
exist x1 ; : : : ; xm such that

Ux1 [ [ Uxm = X:

83
For each xi , i = 1; : : : ; m, …nd a …nite subfamily Wi of W that covers Uxi Y.
Then the subfamily

fW : W 2 Wi for some i = 1; : : : ; mg

is …nite and covers X Y , which is a contradiction. This proves the claim.

84
 Friday, November 6, 2009
Proof. Step 2: We claim that there is y0 2 Y such that for every neighborhood
U V of (x0 ; y0 ), no …nite subfamily of W covers U V . If this is not the case,
then for all y 2 Y there is a neighborhood Uy Vy of (x0 ; y) and a …nite
subfamily of W that covers Uy Vy . Since fVy gy2Y is an open cover of Y , by
the compactness of Y there exist y1 ; : : : ; y` such that

Vy1 [ [ Vy` = Y:

For each yi , i = 1; : : : ; `, …nd a …nite subfamily Wi0 of W that covers Uyi Vyi .
Then the subfamily

fW : W 2 Wi0 for some i = 1; : : : ; mg

is …nite and covers (Uy1 \ \ Uy` ) Y , which contradicts Step 1. This proves
the claim.
Step 3: Let x0 2 X and y0 2 Y be given as in Steps 1 and 2, respectively. Since
W is an open cover of X Y , there exists W 2 W such that (x0 ; y0 ) 2 W . But
then we can …nd neighborhoods U and V of x0 and y0 such that U V W,
which contradicts Step 2. This completes the proof.
If X is a topological space, we denote by Cc (X) the space of all continuous
functions f : X ! R whose support is compact.

2.6 Product Topology

Given a collection fX g 2 of sets, we de…ne the product of the collection as
( )
Y [
X := f : ! X : f ( ) 2 X for every a 2 :
2 2
Q
Remark 176 Note that, if the collection if …nite, then 2 X reduces to the
usual Cartesian product. Q
On the other hand, if X = X for all 2 , then 2 X is simply the
space of all functions f : ! X. It is sometimes written X .

For each 2 there is a projection onto X given by

Y
: X !X
2
f 7! f ( )

If
Q each X is endowed with a topology , we de…ne the product topology on
2 X as the smallest topology that makes each projection continuous. More
1
precisely, since is continuous if and only if (V ) is open for every open
set V of X , is the smallest topology that contains the family
1
F= (V ) : V 2 ; 2 : (25)

85
Proposition 177 Let f(X ; )g 2 be a collection of topological spaces. Then
a base for the product topology is given by all sets of the form
\
1
(V ) ;
2 0

where 0 is a …nite subset of and V is an open set of X , or, equivalently,

by all sets of the form Y
V ;
2

where V is an open set of X and V = X for all but …nitely many 2 .

Proof. In view of Proposition 119 and of Example 123(iii), a base for the
topology is given by …nite intersections of elements of F, precisely,
\
1
(V )
2 0

where is a …nite subset of and V is an open set of X . Now, for 2 0 ,

0
( )
[
1
(V ) = f : ! X : f ( ) 2 X for every a 2 n f g and f ( ) 2 V
2
Y Y
= X V X :
2 ; < 2 >

Hence, \ Y
1
(V ) = V ;
2 0 2

where V := X for all a 2 n 0.

Exercise 178 Given a nonempty set X, consider the space X R of all functions
f : X ! R. What is the relation between the topology introduced in Example
123 and the product topology?

Why this funny topology?

Q Why not take the smallest topology that contains
all sets of the form 2 V , where V is an open set of X ? This topology is
called the box topology. If is …nite, then the product topology and the box
topology coincide. However, if is in…nite, then the box topology has too many
open sets, and so a lot of properties that are true for the product topology fails
for the box topology (compactness, connectedness, continuity of the projections,
etc.).

Exercise 179 Let f(X ; )g 2 be aQcollection of topological spaces. Prove

that for each 2 the projection : 2 X ! X is continuous and open
but not closed.

86
Theorem 180 Let (X; ) and (X ; ), 2 , be topological spaces and let
Y
f :X! X :
2

Then f is continuous if and only if f : X ! X is continuous for every

2 .

Example 181 Consider the set RN = fg : N ! Rg and the function

f : R ! RN
x 7! (x; x; : : :)

For every n 2 RN we have that n f : R ! R is the function n f (x) = x,

which is continuous. Hence, f is continuous with respect to the product topology.
However, f is not continuous with respect to the box topology. Indeed, consider
the open set

1 1 1 1
B = ( 1; 1) ; ;
2 2 n n
1
Y 1 1
= ; :
n=1
n n

If f were continuous, then the set f 1 (B) would be open. Note that 0 2 f 1 (B),
and so f 1 (B) should contain an interval ( ; ) for some > 0. Hence,
f (( ; )) B, but then taking projections, we get

1 1
( n f ) (( ; )) = ( ; ) ;
n n

for all n 2 N, which is a contradiction.

87
 Monday, November 9, 2009
Proof of Theorem 180. Let f be continuous. Since is continuous by
Exercise 179, it follows that f is continuous. Conversely, assume that each
f is continuous.
Since a subbase for the product topology is given by all sets of the form
1
(V ), where V is an open set of X , by Exercise 156, it su¢ ces to show
that f 1 1
(V ) is open in X. But
1 1 1
f (V ) = ( f) ((V )) ;

which is open in X, since f is continuous, again by Exercise 156. Hence, f

is continuous.

Exercise 182 Let (X; d) and (X ; d ), 2 , be metric spaces and let

Y
f :X! X :
2

Prove f is continuous if and only if that f is continuous for every 2

and for every x 2 X there exists a ball B (x; r) such that f : B (x; r) ! X
is constant for all but …nitely many 2 .

To prove the next theorem we need trans…nite induction. Consider an ar-

bitrary set X and a binary relation on X. We call (X; ) a linear ordering
if

(i) is transitive on X:

for all x; y; z2 X, if x y and y z then x z;

(ii) trichotomy holds:

for all x; y2 X we have that x y or y x or x = y;

(iii) is irre‡exive:

for all x 2 X the property x x does not hold.

We say that x 4 y if x y or x = y.
We say that (X; ) is a well-ordering if it is a linear ordering and

for every E X with E 6= ; there exists x 2 E

such that x 4 y for all y 2 E;

The element x is called the -least element of E. The usual proof by induction
may be extended to well-orderings in the following way.

88
Proposition 183 (Proofs by induction) Let (X; ) be a well-ordering. Let
P (x) be a statement about a variable x. Suppose that for all y 2 X,
if P (x) holds for all x y then P (y) holds.
If P (x0 ) is true for the -least element x0 of X, then P (y) holds for all y 2 X.
Another fundamental assumption of mathematics is that for every set X,
there exists a binary relation on X such that (X; ) is a well-ordering. This
is the axiom of choice (AC) in set theory.
Theorem 184 (Tychono¤’s theorem) Wright. Q Let f(X ; )g 2 be a col-
lection of compact topological spaces. Then 2 X is compact.
Proof. By the axiom of choice, we may assume that is well-ordered; that is,
there is an order relation such Q that every subset of has a smallest element.
Assume by contradiction
Q that 2 X is not compact. Then there is an open
cover
Q W of 2 X with the property that no …nite subfamily of W covers
2 X .
We claim that for every 2 there exists x 2 X with the property that
if W is any open set containing
Y Y
fx g X ;
2 ; 2 >

then no …nite subfamily of W covers W . We use Proposition 183. Let 0 2

be the least element of . Reasoning as
Q in Step 1 of the proof of Theorem 175
(with X replaced by X 0 and Y by 2 > 0 X ) we can prove that there
exists x 0 2 X 0 such that if W is any open set containing
Y
fx 0 g X ;
2 > 0

then no …nite subfamily of W covers W .

Assume next that x 2 X have been chosen for all < . We claim that
there exists x 2 X with the if W is any open set containing
Y Y
fx g X ;
2 ; 2 >

then no …nite subfamily of W covers W . If this is not the case, then for all
x 2 X there is an open set Wx containing
Y Y
fx g fxg X ;
2 ; < 2 >

and a …nite subfamily of W that covers Wx . Without loss of generality, we may

assume that Wx has the form
Y Y
Wx = U ;x U ;x X ;
2 ; < 2 >

89
where U ;x is open in Xa for every 2 with and U ;x = X for all but
…nitely many 2 with . Since fU ;x gx2X is an open cover of X , by
the compactness of X there exist x1 ; : : : ; xm such that

U ;x1 [ [U ;xm =X :

For each xi , i = 1; : : : ; m, …nd a …nite subfamily Wi of W that covers Wxi .

Then the subfamily

fW : W 2 Wi for some i = 1; : : : ; mg

is …nite and covers

Y Y
(U ;x1 \ \U ;xm ) X X ;
2 ; < 2 >

which is an open set covering

Y Y
fx g X :
2 ; < 2
Tm
This contradicts the way the points x have been chosen. Indeed, since i=1 U ;xi =
X , for all but …nitelyTmany 2 with , let 0 be the maximum 2
m
with for which i=1 U ;xi 6= X . Then we have contradicted the choice
of x 0 . This proves the claim. Thus we have constructed a family fx g 2 .
The function
[
f: ! X
2
7! x

Then no element of W covers f . Indeed, if W 2 W covers f , then it must

contains an open set of the form
Y
Ug
2

where U is a neighborhood of x for every 2 and U ;x = X for all but

…nitely many 2 , which contradicts the the way the points x have been
chosen.
N
Exercise 185 Prove that [0; 1] = ff : N ! [0; 1]g with the box topology is not
compact.

Exercise 186 Let f(X ; )g 2 be a collection of nonempty

Q topological spaces,
and let E X be nonempty for every 2 . Fix g 2 2 E and consider
the set
( )
Y
E := f 2 E : f ( ) = g ( ) for all but …nitely many 2 :
2

90
Prove that Y
E= E :
2

Exercise 187 Prove that in RN = ff : N ! Rg with the box topology, the set

U := ff : N ! R : f is boundedg

is both open and closed, so that RN is not connected.

Exercise 188 Let f(X ; )g 2 be a collection of topological spaces and let

E X be nonempty sets for every 2 .

(i) Prove that

Y Y
E = E :
2 2

(ii) Is it true that !

Y Y
E = (E ) ?
2 2
Q
(iii) Prove that 2 E is closed if and only if E is closed for every 2 .

Q Let f(X ; )g 2 be a collection of Hausdor¤ topological spaces.

Exercise 189
Prove that 2 X is a Hausdor¤ space.

Example 190 2[0;1] with the product topology is compact but not sequentially
compact.

91
 Wednesday, November 11, 2009

2.7 Compacti…cation
In studying a noncompact topological space X it is often useful to construct
a space that contains X and that is compact. The extended real numbers are
such an example.
The simplest type of compacti…cation is given by adding one point to X.

Theorem 191 (Alexandro¤) Let (X; ) be a topological space. Let 1 denote

a point not in X and consider the set X 1 := X [ f1g. Let 1 be the collection
of all subsets U X 1 such that either U is an open set of X or 1 2 U and XnU
is a closed compact set of X. Then (X 1 ; 1 ) is a compact topological space.
Moreover, (X 1 ; 1 ) is a Hausdor¤ space if and only if (X; ) is Hausdor¤ and
locally compact.

Proof. Step 1: We prove that (X 1 ; 1 ) is a topological space. We begin by

observing that if U belongs to 1 if and only if

(i) U \ X belongs to
(ii) if 1 2 U , then X n U is compact set of X.

By (i), …nite intersections and arbitrary unions of elements of 1 intersect

X in open sets. If U1 ; U2 2 1 and 1 2 U1 \ U2 , then by De Morgan’s laws

X n (U1 \ U2 ) = (X n U1 ) [ (X n U2 ) ;

and since both X n U1 and X n U2 are closed and compact, so is their union.
This shows that …nite intersections of elements of 1 are still in 1 . If fU g 2
is an arbitrary family of elements of 1 and if
[
12 U ;
2

then
S 1 2 U for some 2 S . Hence, X n U is closed and compact. Since
2 U \ X is open, X n 2 U is closed and since
[
Xn U X nU ;
2
S S
by Proposition 167, X n 2 U is compact. Hence, 2 U belongs to 1 .
The set X 1 belongs to 1 , since X 1 n X 1 = ; is closed and compact, while
the empty set belongs to 1 by (i). Thus, (X 1 ; 1 ) is a topological space.
Step 2: We prove that (X 1 ; 1 ) is compact. Let fU g 2 be a family of
elements of 1 such that [
X1 = U :
2

92
Then 1 2 U for some 2 . Hence, X n U is closed and compact. Since
[
X nU U \ X;
2

there exist 1; : : : ; m 2 such that

m
[
X nU U i
\ X:
i=1

The …nite family fU 1 ; : : : ; U m ; U g covers X 1 .

Step 3: Finally, we show that (X 1 ; 1 ) is a Hausdor¤ space if and only if
(X; ) is Hausdor¤ and locally compact. Assume that (X 1 ; 1 ) is a Hausdor¤
space. Then (X; ) is a Hausdor¤ space in view of property (i). To prove that it
is locally compact, let x 2 X. Since (X 1 ; 1 ) is a Hausdor¤ space, there exist
two disjoint neighborhoods U and V of x and 1, respectively. Then X n V is
closed and compact and U X n V . Hence, U X n V , and so U is compact
by Proposition 167.
Conversely, assume that (X; ) is Hausdor¤ and locally compact. Let x; y 2
X 1 be two distinct points. If neither of them is 1, then x; y 2 X, and so since
(X; ) is a Hausdor¤ space, there exist two disjoint neighborhoods U X and
V X of x and y, respectively. By (i), U and V belong to 1 . If y = 1,
since (X; ) is locally compact, choose a neighborhood U of x such that U is
compact. Then X1 n U belongs to 1 . Hence, U and X1 n U are two disjoint
neighborhoods of x and 1.

Exercise 192 Prove that the circle is the one-point compacti…cation of (0; 1).

Exercise 193 Describe the one-point compacti…cation of the following sets of

R2 with the usual topology.

(i) f(x; y) : x 2 (0; 1] ; y = 0g

1 1
(ii) n; n : n2N

(iii) (x; y) : x2 + y 2 < 1

(iv) (x; y) : x2 + y 2 < 1 [ f(0; 1)g

(v) f(x; y) : 1 x 1g.

93
 Friday, November 13, 2009
Given a topological space (X; ) and a function f : X ! R, the support of
f is the closed set
supp f := fx 2 X : f (x) 6= 0g:
Exercise 194 Let (X; ) be a topological space and consider the space
Cc (X) := ff : X ! R continuous, supp f is a compact set of Xg
with the metric
d (f; g) := max jf (x) g (x)j :
x2X

The completion of Cc (X) is the space C0 (X) of functions vanishing at in…nity.

Prove that f 2 C0 (X) if and only if for every " > 0 there exists a closed compact
set K X such that
jf (x)j < "
for all x 2 X n K.
Exercise 195 Let (X; ) be a topological space. Prove that g 2 C (X 1 ) if and
only if
(g c)jX = f
for some f 2 C0 (X) and c 2 R. Show also that
kg ckC(X 1 ) = max kf kC0 (X) : (26)

2.8 µ
Stone–Cech’
s Compacti…cation
Given a topological space (X; ), a compacti…cation of X is a pair (h; Y ), where
(Y; Y ) is a compact topological space and h : X ! Y is one-to-one, continuous,
h (X) is dense in Y and h 1 : h (X) ! X is continuous; that is, h is a home-
omorphism of X into a dense subset of Y . A compacti…cation (h; Y ) of X is
Hausdor¤ if the space Y is Hausdor¤.
Note that if (X 1 ; 1 ) is the one-point-compacti…cation of X, then (i; X 1 ) is
a compacti…cation, where I : X ! X 1 is the identity function. For noncompact
spaces, the one-point-compacti…cation is a way the smallest compacti…cation.
We now construct the largest.
Given a topological space (X; X ) consider the space Cb (X) of all real-valued
continuous bounded functions f : X ! R. For every f 2 Cb (X) there exists
tf > 0 such that f (x) 2 [ tf ; tf ] for all x 2 X. Consider
Y
Y0 := [ tf ; tf ] = fg : Cb (X) ! R : g (f ) 2 [ tf ; tf ] for every f 2 Cb (X)g :
f 2Cb (X)

By Tychono¤’s theorem, this space is compact with the product topology Y0 .

Consider the evaluation map e : X ! Y0 de…ned as follows: for every x 2 X,
e (x) : Cb (X) ! R is the function
e (x) (f ) := f (x) 2 [ tf ; tf ] ; f 2 Cb (X) :

94
We claim that the function e is continuous. To see this, note that for every
f 2 Cb (X) the projection f is given by

f : Y0 ! [ tf ; tf ]
g 7! g (f )

and so f e : X ! [ tf ; tf ] is the function f itself, since for every x 2 X,

( f e) (x) = f (e (x)) = e (x) (f ) = f (x) : (27)

Since f is continuous, it follows from Theorem 180 that e is continuous. De…ne

(X) := e (X) Y0 . Since Y0 is compact and (X) is closed, we have that
(X) is compact (see Proposition 167). Thus we can consider the pair (e; (X)).
We want to see when this pair is a compacti…cation of X. We are missing two
properties. We need e to be one-to-one and we need e 1 : e (X) ! X to be
continuous.
The evaluation map e is one-to-one if for every x; y 2 X with x 6= y, we
have that e (x) 6= e (y). This is equivalent to say that there exists f 2 Cb (X)
such f (x) = e (x) (f ) 6= e (y) (f ) = f (y). This property says that bounded
continuous functions separate points. Assume that this is the case. Then we
can consider the inverse function e 1 : e (X) ! X.
Note that e 1 : e (X) ! X is continuous if and only if e : X ! e (X) is
open.

De…nition 196 A topological space (X; ) is completely regular if for every

x0 2 X and every closed set C X that does not contain x0 , there exists a
continuous function f : X ! [0; 1] such that f (x0 ) = 1 and f = 0 on C.

Exercise 197 Prove that a metric space is completely regular.

95
 Monday, November 16, 2009
Assume that X is completely regular. Fix an open set U X and let x0 2 U .
Since X is completely regular, there exists f0 2 Cb (X) such that f0 (x0 ) = 1
and f0 = 0 on X n U . The set Vf0 := f0 1 ((0; 1)) is open in X, contains x0 and
Vf0 U . Moreover,

e (Vf0 ) = fg 2 Y0 : g (f0 ) > 0g \ e (X)

Y
= Af \ e (X) :
f 2Cb (X)

where Af = [ tf ; tf ] if f 6= f0 and Af0 = (0; tf0 ], which is open in [ tf0 ; tf0 ].

Thus, e (Vf0 ) is an open neighborhood of e (x0 ) in e (X).
Thus we have proved the following.

Theorem 198 (Stone–Cech) µ Let (X; ) be a completely regular topological

space such that bounded continuous functions separate points. Then (e; (X))
is a compacti…cation of X.
µ
An important property of the Stone–Cech compacti…cation of X is that
every bounded continuous function f1 : X ! R can be uniquely “extended” to
a continuous function F1 : (X) ! R. More precisely, f1 e 1 : e (X) ! R can
be uniquely “extended” to a continuous function F1 : (X) ! R. To see this,
note that the projection mapping

f1 : (X) ! [ tf1 ; tf1 ]

g 7! g (f1 )

is continuous. Moreover, if g 2 e (X), then there exists x 2 X such that g = e (x)

and so
f1 e 1 (g) = f1 e 1 (e (x)) = f1 (x) :
On the other hand, by (27),

f1 (g) = f1 (e (x)) = e (x) (f1 ) = f1 (x) :

Hence, f1 e 1 (g) = f1 (g) for all g 2 e (X). Thus, if we identify X with

e (X), then f1 can be considered as a continuous extension of f1 .
Note that this extension is actually unique. Indeed, we have the following.

Exercise 199 Let (Y; Y ) and (Z; Z ) be topological spaces, with Z Hausdor¤
and let f : E ! Z be a continuous function, where E Y . Then there is at
most one extension of f to E.

2.9 Normal Spaces

De…nition 200 A topological space (X; ) is a normal space if for every pair
of disjoint closed sets C1 ; C2 X there exist two disjoint open set U1 ; U2 such
that U1 C1 and U2 C2 .

96
Remark 201 Note that U1 \ C2 = ;. Indeed, assume by contradiction that
there exists x 2 U1 \ C2 . Since U2 C2 , we have that x 2 U2 , which implies
that U2 is a neighborhood of x. Using the fact that x 2 U1 and Proposition 128,
we conclude that U1 \ U2 is nonempty, which is a contradiction.

The next result shows that a metric space (X; d) is a normal space.

Proposition 202 Let (X; d) be a metric space and let be the topology deter-
mined by d. Then (X; ) is a normal space.

Proof. If C1 ; C2 X are disjoint and closed, then the open sets

U1 := fx 2 X : dist (x; C1 ) < dist (x; C2 )g ;

U2 := fx 2 X : dist (x; C1 ) > dist (x; C2 )g

are two disjoint neighborhoods of C1 and C2 , respectively.

Exercise 203 Prove that every compact Hausdor¤ space is normal.

97
 Wednesday, November 18, 2009

Exercise 204 Let (X; ) be a topological space and let ff g 2 be a (possi-

bly uncountable) family of lower semicontinuous (respectively sequentially lower
semicontinuous) functions, f : X ! R. Assume that the function

f+ := sup f
2

is real-valued.

1. Prove that f+ is lower semicontinuous.

2. Prove that if is a …nite set, then then the function

f := min f

is still lower semicontinuous.

3. Prove that if is in…nite and f is real-valued, then f may not be lower
semicontinuous.

Exercise 205 Let (X; ) be a topological space and let E X.

(i) Prove that the characteristic function of E, de…ned by

1 if x 2 E;
E (x) :=
0 otherwise,

is lower semicontinuous if and only if E is open.

(ii) Prove that E is sequentially lower semicontinuous if and only if E is
sequentially open (that is, X n E is sequentially closed).
(iii) Prove that there exist sequentially lower semicontinuous functions that are
not lower semicontinuous.

The next theorem gives an important characterization of normal spaces.

Theorem 206 (Urysohn’s lemma) A topological space (X; ) is normal if

and only if for all disjoint closed sets C1 ; C2 X there exists a continuous
function f : X ! [0; 1] such that f 1 in C1 and f 0 in C2 .

Lemma 207 Let (X; ) be a normal space and let Let C U X, where C is
closed and U is open. Then there exists an open set V X such that

C V V U:

98
Proof. Since the sets C and X n U are closed and disjoint, by Remark 201,
we may …nd an open set V X such that V C and V \ (X n U ) = ;. In
particular, V U.
We now turn to the proof of Urysohn’s lemma.
Proof of Urysohn’s lemma. Step 1: Assume that X is normal. Let C1 ; C2
1
X be two disjoint closed sets. Set r0 := 0 and r1 := 1 and let frn gn=2 be an
enumeration of the rationals numbers in (0; 1). By the previous lemma applied
to C1 and X n C2 there exists an open set V0 X such that

C1 V0 V0 X n C2 :

Again by the previous lemma, this time with C1 and V0 , there exists an open
set V1 X such that
C1 V1 V1 V0 ;
and so,
C1 V1 V1 V0 V0 X n C2 :
Inductively, assume that given n 2 N there exist open sets Vr1 ; : : : ; Vrn X such
that if ri < rj , then Vrj Vri . Consider rn+1 . Since r0 = 0 < rn+1 < r1 = 1,
one of the numbers r1 ; : : : ; rn , say ri , will be the largest below rn+1 , and one,
say rj , will be the smallest greater than rn+1 . Then Vrj Vri , and so by the
previous lemma we may …nd an open set Vrn+1 X such that

V rj Vrn+1 Vrn+1 Vri :

Thus, by induction, we can construct a sequence of open sets fVr gr2[0;1]\Q with
the properties that C1 Vr , Vr X n C2 , and for all r; s 2 [0; 1] \ Q with r < s
we have Vs Vr . For r; s 2 [0; 1] \ Q de…ne the functions

r if x 2 Vr ; 1 if x 2 Vs ;
fr (x) := gs (x) :=
0 otherwise, s otherwise,

and
f := sup fr ; g := inf gs :
r2[0;1]\Q s2[0;1]\Q

Then f is lower semicontinuous, g is upper semicontinuous, and 0 f 1. If

x 2 C1 V1 , then, since V1 Vr for all r 2 [0; 1] \ Q, we have that x 2 Vr for
all r 2 [0; 1] \ Q, and so fr (x) = r and

f (x) = sup fr (x) = sup r = 1;

r2[0;1]\Q r2[0;1]\Q

while if x 2 C2 , then, since V0 X n C2 and Vr V0 for all r 2 [0; 1] \ Q, we

have that x 2= Vr for any r 2 [0; 1] \ Q, and so fr (x) = 0 and

f (x) = sup fr (x) = sup 0 = 0:

r2[0;1]\Q r2[0;1]\Q

99
To conclude the …rst part of the proof, it remains to prove that f is continuous.
It is enough to show that f = g. Given x 2 X, if fr (x) > gs (x) for some
r; s 2 [0; 1] \ Q, then, necessarily, r > s, x 2 Vr and x 2= Vs . But s < r implies
Vr Vs , which is a contradiction. Hence, fr gs for all r; s 2 [0; 1] \ Q, and
so, taking …rst the supremum over all r and then the in…mum over all s, we
conclude that f g. Now assume by contradiction that f (x) < g (x) for some
x 2 X. Then by the density of the rationals numbers we may …nd r; s 2 [0; 1]\Q
such that
f (x) < r < s < g (x) :
Since f (x) < r, it follows that x 2
= Vr . On the other hand, since s < g (x), we
have that x 2 Vs . This contradicts the fact that Vs Vr and completes the …rst
part of the proof.
Step 2: Assume that for all disjoint closed sets C1 ; C2 X there exists a
continuous function f : X ! [0; 1] such that f 1 in C1 and f 0 in C2 . We
claim that X is normal. Indeed, let C1 ; C2 X be disjoint closed sets and let
1 1
f : X ! [0; 1] be as above. Then the sets f 1 2; 2 and f 1 12 ; 32 are
open, disjoint, and contain C1 and C2 , respectively. This concludes the proof.

Exercise 208 Let (X; ) be a normal space.

(i) Prove that given a proper set closed set C X, there exists a continuous
function f : X ! [0; 1] such that f 0 in C and f > 0 in X n C if and
only if C is a G set.
(ii) Let C1 ; C2 X be two disjoint closed sets. Prove that there exists a
continuous function f : X ! [0; 1] such that f 1 in C1 , f 0 in C2 ,
and 0 < f < 1 in X n (C1 [ C2 ) if and only if C1 ; C2 are G sets.

100
 Friday, November 18, 2009
To prove the next theorem, we need a few facts about the induced topology.
Next we introduce the notion of induced topology.

De…nition 209 Given a topological space (X; ) consider a subset E X.

Consider the family of sets

E := fU \ E : U 2 g :

Then (E; E ) is a topological space. The topology E is called the relative, or

induced, topology on E. An element of E is called a relatively open set.

Exercise 210 Consider a topological space (X; ) and a subset E X. For

E
every F E, let F denote the closure of F with respect to the topology E .
Prove that
E
F = F \ E:

Remark 211 It follows from the previous exercise that if E is closed (with
E
respect to ), then F is closed (with respect to ), since it is given by the
intersection of two closed sets (with respect to ).

Remark 212 Let (X; X ) and (Y; Y ) be two topological spaces, and let f :
E ! Y , where E Y . We recall that given x0 2 E, the function f is said to
be continuous at x0 if for every neighborhood V Y of f (x0 ), there exists a
neighborhood U X of x0 such that

f (x) 2 V

for all x 2 U \ E. Since U \ E 2 E , we have that f is continuous at x0 if and

only if f : (E; E ) ! Y is continuous at x0 .

Theorem 213 (Tietze’s extension theorem) A topological space (X; ) is

normal if and only if for every closed set C X and every continuous function
f : C ! R there exists a continuous function F : X ! R such that F (x) = f (x)
for all x 2 C: Moreover, if f (C) [a; b], then F may be constructed so that

F (C) [a; b] .

Proof. Step 1: Assume that X is normal, let C X be a closed set and let
f : C ! [ 1; 1] be a continuous function. Then the sets f 1 31 ; 1 and
f 1 1; 13 are disjoint closed subsets of C with respect to the induced
topology C but since C is closed, it follows by the previous exercise that they
are actually closed with respect to . Hence, we may apply Urysohn’s lemma to
1 1 1 1 1
…nd a continuous function f1 : X ! 3 ; 3 such that f1 3 in f 3; 1
1
and f1 3 in f
1
1; 13 . We claim that

2
jf f1 j on C:
3

101
Indeed, if f (x) 2 1; 13 , then f1 (x) = 31 ; if f (x) 2 13 ; 1 , then f1 (x) = 31 ;
1 1 1 1
while if f (x) 2 3 ; 3 , then so f1 (x) 2 3; 3 .
1 2
Repeat this construction with f f1 in place of f and (f f1 ) 9; 1
1 2 1 1
and (f f1 ) 1; 9 in place of f 1 3 ; 1 and f 1 1; 3 , re-
2 2 2
spectively, to …nd a continuous function f2 : X ! 9 ; 9 such that f2 9 in
1 2 2 1 2
(f f1 ) 9 ; 1 and f 2 9 in (f f 1 ) 1; 9 . As before, we can
prove that
2
2
j(f f1 ) f2 j on C:
3
Inductively for every n 2 N we can construct a continuous function
" #
n 1 n 1
1 2 1 2
fn : X ! ; (28)
3 3 3 3

such that n
2
jf f1 fn j on C: (29)
3
De…ne
1
X
F (x) := fn (x) ; x 2 X:
n=1

Note that by (28),

1
X 1
1X
n 1
2
jfn (x)j = 1; (30)
n=1
3 n=1 3

and so F is well-de…ned and takes values in [ 1; 1]. In turn, since the series
converges, it follows from (29) that for every x 2 C,
m
X
jf (x) F (x)j = f (x) lim fn (x)
m!1
n=1
Xm m
2
lim f (x) fn (x) lim = 0;
m!1
n=1
m!1 3

and so F = f on C. It remains to show that F is continuous. Fix x 2 X and

" > 0 and …nd n" so large that
1
X n 1
2 "
: (31)
n=n" +1
3 2

Since f1 ; : : : ; fn" are continuous at x, for every n = 1; : : : ; n" there exists a

neighborhood Un of x such that if y 2 Un , then
"
jfn (y) fn (x)j : (32)
2n"

102
 T n"
Take U := n=1 Un . Then by (30), (31), and (32), for every y 2 U ,
n"
X n"
X 1
X 1
X
jF (y) F (x)j = fn (y) fn (x) + jfn (x)j + jfn (y)j
n=1 n=1 n=n" +1 n=n" +1
n"
X 1
X n 1
2 2 " "
jfn (y) fn (x)j + n" + = ";
n=1
3 n=n 3 2n" 2
" +1

which shows that F is continuous at x. Note that the proof continues to work
if in place of f : C ! [ 1; 1] we have f : C ! [a; b], with the only change that
in this case F : X ! [a; b].
Step 2: Assume that X is normal, let C X be a closed set and let f : C ! R
be a continuous function. Since ( 1; 1) is homeomorphic to R, we can construct
an homeomorphism g : R ! ( 1; 1). Consider the function h := g f : C !
[ 1; 1]. Since h is continuous, by Step 1 there exists a continuous function
H : X ! [ 1; 1] such that H = h on C. The problem is that H can take values
1 and 1. To avoid this, let C1 := H 1 (f 1; 1g). Then C1 and C are closed
and disjoint, and so by Urysohn’s lemma we may …nd a continuous function
h1 : X ! [0; 1] such that h1 0 in C1 and h1 1 in C. Then H1 := Hh1 is
continuous, H1 = H = h on C and H1 : X ! ( 1; 1). Since g 1 is continuous,
the function F := g 1 H1 is continuous and real-valued and for x 2 C,
1 1 1
F (x) = g (H1 (x)) = g (h (x)) = g (g f (x)) = f (x) :

Step 3: Assume that or every closed set C X and every continuous function
f : C ! R there exists a continuous function F : X ! R such that F (x) = f (x)
for all x 2 C. Assume also that if f (C) [a; b], then F (C) [a; b]. Let
C1 ; C2 X be two disjoint closed sets. Then C1 [ C2 is closed. De…ne f := 1
in C1 and f := 0 in C2 . Then f : C1 [ C2 ! [0; 1] is continuous, and so
by hypothesis there exists a continuous function F : X ! [0; 1] such that
F (x) = f (x) for all x 2 C1 [ C2 . Thus, we are in a position to apply Urysohn’s
lemma (or repeat Step 2 of its proof) to conclude that X is normal.

103
 Monday, November 23, 2009
Next we study partitions of unity for normal spaces.

De…nition 214 Let X be a topological space and let F be a collection of subsets

of X. Then

(i) F is point …nite if every x 2 X belongs to only …nitely many U 2 F,

(ii) F is locally …nite if every x 2 X has a neighborhood meeting only …nitely
many U 2 F,
(iii) G P (X) is a re…nement of F if
[ [
G= F
G2G F 2F

and every element of G is contained in some element of F.

Theorem 215 A topological space (X; ) is normal if and only if for every
point …nite open cover fU g 2 of X there exists another open cover fV g 2
of X with the property that V U .

Proof. Assume that (X; ) is normal and let fU g 2 be a point …nite open
cover of X. By the axiom of choice, we may assume that is well-ordered;
that is, there is an order relation such that every subset of has a smallest
element. Let 0 2 be the least element of . We claim that for every 2
there is an open set V with the properties that

C V V U ;

where 00 1 0 11
[ [
C := X n @@ V A[@ U AA :
< >

To construct V we use trans…nite induction on . De…ne

[
C 0 := X n U :
> 0

Then C 0 U 0
and is closed. By Lemma 207 there exists an open set V 0
such that
C 0
V 0
V 0
U 0:
Suppose that V has been chosen for every 2 with < and de…ne
00 1 0 11
[ [
C := X n @@ V A[@ U AA :
< >

104
To prove that C U , …x x 2 C . Then

x2
= V for any < and x 2
= U for any > : (33)

Since fU g 2 is a point …nite cover, x belongs to only …nitely many U ,

say, U 1 ; : : : ; U m . Without loss of generality, we may assume that m =
max f 1 ; : : : ; m g. Then x 2= U for > m . Hence, by (33), we have that
m . We claim that m = . Indeed, if m < , then by (33), x 2
= V m but
! !!
[ [
x2C m =Xn V [ U ;
< m > m

which contradicts the fact that C m V m . This shows that m = , so that

x2U .
By Lemma 207 there exists an open set V such that

C V V U :

Thus, by Proposition 183 we have constructed a family of open sets fV g 2 .

It remains to show that it covers X. Fix x 2 X. Since fU g 2 is a point
…nite cover, x belongs to only …nitely many U , say, U 1 ; : : : ; U m . As before
assume that m = Smax f 1 ; : : : ; m g. Then x 2
= U for > m . There S are now
two cases. If x 2 < m V , then there is nothing to prove. If x 2= < m V m,
then x 2 C m and so x 2 V m . This shows that fV g 2 is an open cover of X.
Conversely, assume that every point …nite open cover fU g 2 of X there
exists another open cover fV g 2 of X with the property that V U . We
claim that (X; ) is normal. Let C1 ; C2 X be two disjoint closed sets. Then
fX n C1 ; X n C2 g is a point …nite open cover of X. Hence, there exist two open
sets V1 and V2 such that

X = V1 [ V2 ; V1 X n C1 ; V2 X n C2 :

The sets XnV1 and XnV2 are open, disjoint, and contain C1 and C2 , respectively.

As a corollary of the previous theorem we can show that in a normal space

every locally …nite open cover admits a partition of unity subordinated to it.

De…nition 216 If (X; ) is a topological space, a partition of unity on X is a

family f'i gi2 of continuous functions 'i : X ! [0; 1] such that
X
'i (x) = 1
i2

for all x 2 X. A partition of unity is locally …nite if for every x 2 X there exists
a neighborhood U of x such that the set fi 2 : U \ supp 'i 6= ;g is …nite. If
fUj gj2 is an open cover of X, a partition of unity subordinated to the cover
fUj gj2 is a partition of unity f'i gi2 such that for every i 2 , supp 'i Uj
for some j 2 .

105
 Monday, November 30, 2009

Theorem 217 Let (X; ) be a normal space and let fU g 2 be a locally …nite
open cover of X. Then there exists a partition of unity subordinated to it.

Proof. Since a locally …nite cover is point …nite, by the previous theorem there
exists another open cover fV g 2 of X with the property that V U . Note
that V 2
remains locally …nite.
By Lemma 207, there exists an open set W such that

V W W U :

Since the closed sets sets V and X n W are disjoint, there exists a continuous
function f : X ! [0; 1] such that f 1 in V and f 0 in X n W . Hence,

fx 2 X : f (x) > 0g W

and so,
supp f W U :
De…ne X
f (x) := f (x) ; x 2 X:
2

Since fV g 2 covers X, for every x 2 X there exists 2 such that x 2 V ,

and so f (x) = 1. Thus, f > 0. Moreover, since V 2
is locally …nite for
every x 2 X there exists a neighborhood U of x that intersects only …nitely
many V . Thus, f reduces to a …nite sum in U . In particular, f < 1 in U and
f is continuous in x. By the arbitrariness of x, we have that f is continuous.
Hence the function
f (x)
' (x) := ; x 2 X;
f (x)
is well-de…ned and continuous. Since supp ' = supp f U , the family
f' g 2 is a locally …nite partition of unity subordinated to fU g 2 .

2.10 Metrization
A topological space (X; ) is metrizable if its topology can be determined by a
metric. The metrizability and the normability of a given topology depend on
the properties of a base (see Theorems 218 and 224 below).
In view of and Theorem 219, in order for topological space (X; ) to be
metrizable it is necessary that (X; ) be Hausdor¤ and normal. Thus in the
next two theorems, without loss of generality, we will assume that these two
properties are satis…ed.

Theorem 218 (Urysohn’s metrization theorem) A topological space (X; )

is metrizable and separable if and only if X is Hausdor¤ , normal, and it has a
countable base.

106
Proof. If (X; d) is a metric space, then by Propositions 132 and 202 it is Haus-
dor¤ and normal. Moreover, if (X; d) is separable, then there exists a sequence
fxn g X which is dense in X. The countable family of balls B xn ; k1 k;n2N
is a base for the topology determined by d.
Conversely, assume that (X; ) is a Hausdor¤ normal space with a countable
base B = fBn gn . We will show that X is homeomorphic (in the topological
sense) to a subset of `2 , which is separable.
Step 1: We claim that every closed set is a G , or, equivalently, that every
open set is an F set. Fix an open set U X. Fix x 2 U . Since X is Hausdor¤,
the singleton fxg is closed, and fxg U . By Lemma 207 there exists an open
set V X such that
fxg V V U:
Since B is a base, there exists Bx 2 B such that x 2 Bx V , and so

fxg Bx Bx V U:

This shows that [

U= Bn ;
Bn U

that is, that U is an F set.

Step 2: By the previous step and Exercise 208 for each element Bn in the base
B there exists a continuous function 'n : X ! [0; 1] with the property that

'n (x) = 0 for x 2 X n Bn ; 'n (x) > 0 for x 2 Bn : (34)

De…ne
1 'n (x)
n (x) := q ; x 2 X:
n 2
1 + ('n (x))
Then n is continuous. Moreover,
X X1 2 X1
2 1 ('n (x)) 1
( n (x)) = < 1;
n n=1
n2 1 + ('n (x))2 n=1
n2

which shows that for every …xed x 2 X, f n (x)gn belongs `2 . Hence, the map

f : X ! `2
x 7! f n (x)gn

is well-de…ned.
We claim that f is one-to-one. To see this, let x; y 2 X with x 6= y. Since
X is Hausdor¤, there exists Bn such that x 2 Bn and y 2 X n Bn . It follows
by (34) that n (x) > 0, while n (y) = 0. Hence, f (x) 6= f (y) and the claim is
proved.

107
 Wednesday, December 2, 2009
Proof. Next we claim that f is continuos. Fix x0 2 X and " > 0 and …nd
n" 2 N such that
1
X 1 "2
:
n=n +1
n2 8
"

Since each n , n = 1; : : : ; n" , is continuous at x0 and we have a …nite number

of them, there exists a neighborhood V U of x0 such that
"
j n (x) n (x0 )j p
2n"
for all x 2 V and n = 1; : : : ; n" . Then for x 2 V ,
n"
X 2 "2 "2
( n (x) n (x0 )) n" = ;
n=1
2n" 2

while
1
X 1
X h i
2 2 2
( n (x) n (x0 )) 2 ( n (x)) + ( n (x0 ))
n=n" +1 n=n" +1
X1
1 1 4"2 "2
2 2
+ 2 = :
n=n" +1
n n 8 2

Hence,
d2 (f (x) ; f (x0 )) "
for all x 2 V , which shows continuity at x0 .
Finally, we prove that f 1 : f (X) ! X is continuous. Let y0 2 f (X), then
there exists x0 2 X such that f (x0 ) = y0 . Consider a neighborhood U of x0 .
Since B is a base, there exists Bn in B such that Bn U so that n (x0 ) > 0.
Let := n (x0 ). If d2 (f (x) ; f (x0 )) < , then
1
! 21
X 2
j n (x) n (x0 )j ( n (x) n (x0 ))
n=1
= d2 (f (x) ; f (x0 )) < = n (x0 ) ;
which implies that n (x) > 0, and so, by (34), that x 2 Bn U . This shows
that f 1 is continuous at y0 = f (x0 ).
Next we drop the separability. The next theorem was …rst proved by Stone
in 1948. The present proof is due to Ornstein (see also a proof of M.E. Rudin).
Theorem 219 Let (X; d) be a metric space. Then every open cover fU g 2
of X admits a point …nite re…nement.
Proof. Let fU g 2 be an open cover of X. By the axiom of choice we may
assume that is well-ordered. Fix 2 . A chosen ball (with respect to ) is
a ball B x; 2nx1+1 such that

108
 (i) B x; 2n1x U ,
(ii) nx is the smallest integer for which (i) holds,
(iii) B x; 2n1x U for some < .

Step 1: Let B := B x; 2nx1+1 : B x; 2nx1+1 is a chosen ball and de…ne

the open set
[
V := U n B:
B2B

We claim that fV g 2 is still an open cover of X. Indeed, assume by con-

tradiction that there exists x 2 X that is not covered by fV g 2 . Let U
be the …rst element that contains x (recall that the set f 2 : x 2 U g has
a minimum). Then B (x; r) U for some r > 0. Since x 2 = V and x is not
in any chosen ball (with respect to ) in view of (iii), it follows that x must
be
n an accumulation o point of chosen balls, namely, there exist two sequences
B xk ; 2nx1k +1 B and yk 2 B xk ; 2nx1k +1 such that yk ! x as k ! 1
(the balls could be repeated). Note that nxk cannot approach in…nity along a
subsequence, not relabeled, since this would imply that

1
B xk ; nx k 1 B (x; r) U
2

for in…nitely many k and this would contradict (ii). Hence,

1 1
min = n0 +1
k 2nxk +1 2
1
for some n0 2 N. Let k 2 N be so large that d (x; yk ) < 2n0 +1
, then

1 1 2 1
d (x; xk ) d (x; yk ) + d (y; xk ) < + nx +1 = nx ;
2n0 +1 2 k 2nxk +1 2 k
that is, x 2 B xk ; 2n1xk . But then by (iii), it follows that x must belong to U
for some < , which contradicts the choice of and proves the claim.

109
 Friday, December 4, 2009
Step 2: Next we prove that for every x 2 X there exists a …nite number of
V that contain x. By the previous step there exists V such that x 2 V .
By construction, this means that U is the …rst element to contain some ball
B x; 21m . Note that if U is the …rst element to contain some ball B x; 21m
and U is the …rst element to contain some ball B x; 21n , and if, say, n > m,
then . Hence, the family such that V contain x form a descending
sequence. Since is well-ordered, only a …nite number are distinct.
Theorem 220 Let (X; d) be a metric space. Then every point …nite open cover
fV g 2 of X admits a locally …nite re…nement.
Proof. We construct a re…nement of fV g 2 that is locally …nite. For every
x 2 X let
1
rx := sup fr > 0 : B (x; r) V for some 2 g :
2
If rx = 1 for some x, then one could construct a re…nement of fV g 2 given
by sequence of balls fB (x; n)gn2N , which is a locally …nite. Thus, assume that
rx < 1 for all x 2 X. For every 2 , let W be the union of all balls B x; r2x
such that V is the …rst open set in the cover fV g 2 to contain B (x; rx ). By
construction, fW g 2 is still an open cover of X. Moreover, W V U
for every 2 , and so fW g 2 is a re…nement of fV g 2 . It remains to
show that fW g 2 is locally …nite.
We claim that if W \ B x; r8x 6= ;, then x 2 V . Indeed, assume the
r
contrary for some x 2 X. Then there exists y 2 W such that B y; 2y \
rx
B x; 8 6= ; such that x 2 = V . Since y 2 W , we have that B (y; ry ) V . In
particular, x 2
= B (y; ry ), and so
ry rx
ry < d (x; y) < + ;
2 8
ry rx
which implies that 2 < 8 . Hence,
ry rx rx rx rx
d (x; y) < + < + = ;
2 8 8 8 4
5ry
that is, y 2 B x; r4x . In turn, B y; 2 B (x; rx ). By the de…nition of rx ,
this implies that B (y; 5ry ) belongs to some V , which contradicts the de…nition
of ry . Hence, the claim holds. Thus, for every x 2 X, the only open sets W
that intersect B x; r8x are those for which V contains x. By the previous step
these V are …nite and the proof is complete.
De…nition 221 Let X be a topological space and let F be a collection of subsets
of X. Then F is -locally …nite if
1
[
F= Fn ;
n=1

where each Fn is a locally …nite collection in X.

110
Corollary 222 Let (X; d) be a metric space. Then every open cover fU g 2
of X admits a locally …nite partition of unity subordinated to it.

Corollary 223 Let (X; d) be a metric space. Then X admits a -locally …nite
base.

Proof. For every x 2 X and n 2 N consider the ball B x; n1 . Then B x; n1 x2X;n2N

is a base. Fix n 2 N and consider the open cover of X, B x; n1 x2X .
By the previous theorems, there exists a locally …nite open re…nement Vn of
B x; n1 x2X .
It turn out that the previous condition is also su¢ cient for metrizability.

Theorem 224 (Nagata–Smirnov’s metrization theorem) A topological space

(X; ) is metrizable if and only if it is Hausdor¤ , normal, and has a -locally
…nite base.

END OF THE COURSE

111
 Monday, December 32, 2009

Lemma 225 Let fE g 2 be a locally …nite family of sets. Then

[ [
E = E :
2 2

In particular, the union of a locally …nite family of closed sets is closed.

Proof. De…ne [
E := E :
2

Indeed, by Proposition 129, we have that

[
E E :
2

To prove the opposite inclusion, let x 2 E. Since fE g 2 is locally …nite,

there exists a neighborhood U of x that intersects only …nitely many E , say,
E 1 ; : : : ; E m . Let V be a neighborhood of x. Since x 2 E, by Proposition 128
we have that (V \ U ) \ E is nonempty. By the choice of U , this implies that
m
[
(V \ U ) \ E i
6= ;:
i=1

Hence,
m
[ m
[
y2 E i = E i;
i=1 i=1

where we have used again Proposition 129. This concludes the proof.
We now turn to the proof of Nagata–Smirnov’s metrization theorem.
Proof. If (X; d) is a metric space, then by Propositions 132 and 202 it is
Hausdor¤ and normal. In Theorem 219, we will see that X has a -locally …nite
base.
Conversely, assume that X is Hausdor¤, normal, and has a -locally …nite
base, that is, a base of the form
1
[
B= Bn ;
n=1

where each Bn = fBn; g 2 n is locally …nite, then X is metrizable. We will

show that X is homeomorphic (in the topological sense) to a subset of an `2
metric space.
Step 1: We claim that every closed set is a G , or, equivalently, that every
open set is an F set. Fix an open set U X. As in Step 1 of the proof of

112
Urysohn’s metrization theorem, for every x 2 U there exists Bx 2 B such that
x 2 Bx Bx U . For every n 2 N, let
[
Cn := Bx :
Bx 2Bn

By Lemma 225, Cn is closed. Moreover, Cn U . Since

1
[
U= Cn ;
n=1

it follows that U is an F set.

Step 2: By the previous step and Exercise 208 for each element Bn; in the
base B there exists a continuous function ' ;n : X ! [0; 1] with the property
that
' ;n (x) = 0 for x 2 X n B ;n ; ' ;n (x) > 0 for x 2 B ;n : (35)
De…ne
1 ' ;n (x)
;n (x) := q P ; x 2 X:
n 1+ (' ;n (x))
2

Note that since each Bn is locally …nite, for every x 2 X there exists a neigh-
borhood U of x such that ' ;n = 0 in U for all except at most …nitely many.
P 2
Thus, the in…nite sum (' ;n ) reduces to a …nite sum in U . This shows that
;n is continuous. Moreover,

X X1 1
1 X X
2
2 (' ;n (x)) 1
( ;n (x)) = P 2 < 1;
;n n=1
n2 1+ (' ;n (x)) n=1
n2

which shows that for every …xed x 2 X, f ;n (x)g ;n belongs to the `2 space
( )
X
2 2
` := fa ;n g ;n : a ;n < 1
;n

with metric ! n1
X 2
d2 (a; b) := (a ;n b ;n ) ;
;n

where a = fa ;n g ;n and b = fa ;n g ;n . Hence, the map

f : X ! `2
x 7! f ;n (x)g ;n

is well-de…ned.
We claim that f is one-to-one. To see this, let x; y 2 X with x 6= y. Then
there exist B ;n such that x 2 B ;n and y 2 X n B ;n . It follows by (35) that
;n (x) > 0, while ;n (y) = 0. Hence, f (x) 6= f (y) and the claim is proved.

113
 Next we claim that f is continuos. Fix x0 2 X and " > 0 and …nd n" 2 N
such that
1
X 1 "2
:
n=n +1
n2 8
"

By local …niteness, for every n = 1; : : : ; n" we may …nd a neighborhood of x0

that intersects …nitely many B ;n in Bn . By intersecting this …nite number of
neighborhoods, we obtain a neighborhood U of x that intersects …nitely many
B ;n for n = 1; : : : ; n" , say, say, B 1 ;n1 ; : : : ; B m ;nm . Since each i ;ni
is con-
tinuous at x0 and we have a …nite number of them, there exists a neighborhood
V U of x0 such that
"
j i ;ni (x) i ;ni
(x0 )j p
2m
for all x 2 V and i = 1; : : : ; m. Then for x 2 V ,
n" X
X m
X
2 2 "2 "2
( ;n (x) ;n (x0 )) = ( i ;ni
(x) i ;ni
(x0 )) m = ;
n=1 i=1
2m 2

while
1
X X 2
1
X Xh 2 2
i
( ;n (x) ;n (x0 )) 2 ( ;n (x)) + ( ;n (x0 ))
n=n" +1 n=n" +1
X1
1 1 4"2 "2
2 2
+ 2 = :
n=n" +1
n n 8 2

Hence,
d2 (f (x) ; f (x0 )) "
for all x 2 V , which shows continuity at x0 .
Finally, we prove that f 1 : f (X) ! X is continuous. Let y0 2 f (X),
then there exists x0 2 X such that f (x0 ) = y0 . Consider a neighborhood U
of x0 . Since B is a base, there exists B ;n in B such that B ;n U so that
;n (x 0 ) > 0. Let := ;n (x 0 ). If d 2 (f (x) ; f (x 0 )) < , then
1 X
! 21
X 2
j ;n (x) ;n (x0 )j ( ;n (x) ;n (x0 ))
n=1
= d2 (f (x) ; f (x0 )) < = ;n (x0 ) ;
which implies that ;n (x) > 0, and so, by (35), that x 2 B ;n U . This
shows that f 1 is continuous at y0 = f (x0 ).

2.11 Paracompact Spaces and Partitions of Unity

Theorem 217 leaves open the question of the existence of a partition of unity
subordinated to an arbitrary open cover of X. This problem brings us to the
de…nition of paracompact spaces.

114
De…nition 226 A topological space (X; ) is paracompact if it is Hausdor¤
and if for every an open cover fU g 2 of X, there exists an open cover of X
that is a locally …nite re…nement of fU g 2 .

An important class of paracompact spaces is given by metric spaces. This

follows from Theorems 219 and 220.
Next we show that paracompact spaces are normal spaces.

Proposition 227 Let (X; ) be a paracompact space. Then (X; ) is a normal

space.

Proof. Step 1: Let C X be a closed set and let x 2 X n C. We claim that

there exist two disjoint neighborhoods of x and C. Since X is a Hausdor¤ space,
for every y 2 C, there exist two disjoint neighborhoods Vx and Vy of x and y.
Note that since Vx \ Vy = ;, by Proposition 128, we have that x 2= Vy . Then the
sets Vy , y 2 C, and X n C form an open cover of X. Since X is paracompact,
there exists a locally …nite re…nement fU g 2 . De…ne
[
U := U :
2 : U \C6=;

Then U is open and contains C. Moreover, by the previous lemma,

[
U= U :
2 : U \C6=;

Since by construction each U such that U \ C 6= ; is contained in some Vy ,

it follows that U Vy , and so x 2
= U . In turn, x 2
= U . The sets X n U and U
are open, disjoint, and contain x and C, respectively.
Step 2: Let C1 ; C2 X be two closed disjoint sets. By the previous step, for
each y 2 C2 we may …nd an open set Vy such that Vy \ C1 = ;. De…ne U as in
the previous step with C replaced with C2 . Then exactly as before we can show
that C1 \ U = ;, and so the sets X n U and U are open, disjoint, and contain
C1 and C2 , respectively. This shows that X is normal.
Next we present several characterizations of paracompact spaces.

Theorem 228 (Michael) Let (X; ) be a normal space. Then the following
are equivalent.

(i) (X; ) is paracompact.

(ii) For every open cover of X there exists a locally …nite re…nement (not
necessarily open).
(iii) For every open cover of X there exists a closed, locally …nite re…nement.
(iv) For every open cover of X there exists a -locally …nite open re…nement.

115
Proof. (i)=)(ii) There is nothing to prove.
(ii)=)(iii) Assume (ii) and let fU g 2 be an open cover of X. By Theorem
215 there exists an open cover fV g 2 of X with the property that V U
for every 2 . Now apply (ii) to fV g 2 to …nd a locally …nite re…nement E
of fV g 2 . Let C := E : E 2 E . Then C is still a re…nement of fU g 2 . It
remains to show that C is still locally …nite. Let x 2 X. Since E is locally …nite,
there exists a neighborhood U of x that intersects only …nitely many elements
of E, say, E1 ; : : : ; Em . If E \ U 6= ;, then by Proposition 128, E \ U 6= ;, and
so E is one of the E1 ; : : : ; Em . Thus, C := E : E 2 E is locally …nite.
(iii)=)(i) Assume (iii) and let fU g 2 be an open cover of X. By (iii)
there exists a closed locally …nite re…nement C of fU g 2 . Hence, for every
x 2 X there exists a neighborhood Vx of x that intersects only …nitely many
elements of C. Since fVx gx2X is an open cover of X, we can apply (iii) once
more to …nd a closed locally …nite re…nement K of fVx gx2X . For every C 2 C
de…ne KC := fK 2 K : K \ C = ;g and set
[
DC := X n K:
K2KC
S
Since KC is locally …nite, by Lemma 225 the set K2KC K is closed, and thus
DC is open. Moreover, since K is a cover of X, it follows from the de…nition
of KC that DC contains C. Thus, the family fDC gC2C is an open cover of
X. Moreover, if K 2 K, then K intersects DC if and only if K intersects C.
We claim that fDC gC2C is locally …nite. To see this, …x x 2 X. Since K
is locally …nite there exist a neighborhood U of x that intersects only …nitely
many elements of K, say, K1 ; : : : ; Km . On the other S hand, if U \ DC 6= ;
for some C 2 C, then there is y 2 U such that y 2 = K2KC K. Since K is a
cover, y belongs to some K 2 = KC . Thus, U \ K 6= ; for some K 2 = KC , which
implies that one of the K1 ; : : : ; Km intersects C \ U . This means that one of
the K1 ; : : : ; Km intersects DC . Since each Ki is contained in some Vz and Vz
intersects only …nitely many C in C, we have that U \ DC 6= ; for …nitely many
C 2 C, which shows that fDC gC2C is locally …nite.
Since C is re…nement of fU g 2 , for each C 2 C we may …nd C 2 such
that C U C . Finally, de…ne
V := fDC \ U C
: C 2 Cg :
We claim that is locally …nite open re…nement of fU g 2 . Since the open set
DC \ U C contains C and C is a cover of X, the family V is an open re…nement
of fU g 2 . Since fDC gC2C is locally …nite, then so is V.
It remains to show that (iv)=)(ii). Assume (iv) and let fU g 2 be an
open cover of X. By (iv) there exists an open re…nement of the form
1
[
V= Vn ;
n=1

where each Vn is locally …nite. For every n 2 N and for every V 2 Vn de…ne
VV := fU 2 V : U 2 Vk for some k < ng

116
and set [
EV := V n U:
U 2VV

Then EV V and fEV gV 2V is a cover of X. Note that the family fEV gV 2V

is locally …nite. Indeed, given x 2 X, let n 2 N be the …rst integer such that
x belongs to some V in the family Vn . Then V is a neighborhood of x that
does not intersect any EU for U 2 Vk with k > n (since V has been removed
from each such EU ). Thus, V can only intersect sets EU such that U 2 Vk with
k n. Since each Vk is locally …nite for k n, we may …nd a neighborhood
Wk of x that intersects only …nitely many U in Vk ., and, in turn, only …nitely
many of the corresponding EU . The neighborhood
n
\
V \ Wk
k=1

of x intersects …nitely many EU . This shows that fEV gV 2V is locally …nite and
concludes the proof.
The importance of paracompact spaces comes from the following theorem.

Theorem 229 (Michael) Let (X; ) be a normal space. Then the following
are equivalent.

(i) (X; ) is paracompact.

(ii) For every open cover of X there exists a locally …nite partition of unity
subordinated to it.
(iii) For every open cover of X there exists a partition of unity subordinated to
it.

Proof. Step 1: Assume that (X; ) is paracompact and let fU g 2 be an open

cover of X. Since (X; ) is paracompact there exists a locally …nite re…nement
of fU g 2 and (X; ) is a normal space, by the previous proposition. We are
now in a position to apply Theorem 217 to …nd locally …nite partition of unity
subordinated to the re…nement, and in particular, to fU g 2 .
Step 2: Assume that every open cover fU g 2 of X admits a locally …nite
partition of unity f'i gi2I subordinated to it. Fix any two such fU g 2 and
f'i gi2I . For every n 2 N and every i 2 I consider the set

1
Vi;n := x 2 X : 'i (x) > :
n

Then Vn;i is open. We claim that fVi;n gi2I;n2N is an open cover of X that is a
locally …nite re…nement of fU g 2 . Indeed, if x 2 X, then
X
'i (x) = 1;
i2I

117
and so there exists i 2 I such that 'i (x) 2 (0; 1]. It follows that 'i (x) > n1 for
all n 2 N su¢ ciently large, and so x 2 Vi;n for all n 2 N su¢ ciently large. This
shows that fVi;n gi2I;n2N is an open cover of X. Moreover, since Vi;n supp 'i
and f'i gi2I is subordinated to fU g 2 , each supp 'i (and in turn each Vi;n )
is contained in some U . Thus, fVi;n gi2I;n2N is a re…nement of fU g 2 . We
claim that fVi;n gi2I;n2N is -locally …nite. Fix x0 2 X and n 2 N. Write
X X
1= 'i (x0 ) = 'i (x0 ) ;
i2I i2I0
P
where I0 := fi 2 I : 'i (x0 ) > 0g is countable. Since the series i2I0 'i (x0 ) is
convergent, there exists a …nite subset I1 I0 such that
X 1
'i (x0 ) > 1 :
2n
i2I1

By continuity and the fact that I1 is …nite, we may …nd an open neighborhood
U of x0 such that
X 1
'i (x) > 1
n
i2I1

for all x 2 U . Note that if i does not belong I1 , then U cannot intersect Vi;n
(the sum would be greater than one). Hence, U intersect only …nitely many Vi;n
(recall that n is …xed). The result now follows from the previous theorem.

118