º Å Øö × Áò Ø ×

You might also like

You are on page 1of 14

2.

METRIC SPACES

1. Metri s
In this hapter we introdu e the on ept of a metri , whi h is a generalization of the everyday notion
of distan e. Then we pro eed to establish the on epts of open sets, limits, sequen es, and limit
points. These are ne essary in establishing the idea of topology (in this ase, metri topology) whi h
is needed in formulating the on ept of ontinuity later.
De nition: Let X be a non-empty set. A metri d is a mapping d : X  X ! R whi h satis es the
following onditions:

(1) d(x; y)  0; and d(x; y) = 0 () x = y (non-negativity);


(2) d(x; y) = d(y; x) (symmetry);
(3) d(x; z )  d(x; y) + d(y; z ) (triangle inequality):
The pair (X; d) onsisting of the set X and its metri d is alled a metri spa e. The metri that
an be de ned in a given set is never unique, as the examples would show below.
Example 1: Let X be an arbitrary set, and de ne a metri by:
(
d(x; y) =
0 x = y;
1 x 6= y:

Example 2: Let R n be the set of ordered n-tuples x := (x1 ; x2 ; : : : ; xn ). We de ne the Eu lidean


metri as
v
u
uXn
(4) d(x; y) := t (xk yk )2
k=1

Clearly, this satis es the non-negativity and the symmetry onditions. Only the triangle inequality
needs to be shown. Let
x = (x1 ; x2 ; : : : ; xn ); y = (y1 ; y2 ; : : : ; yn ) and z = (z1 ; z2 ; : : : ; zn )
be three distin t points in R n ; and let
ak := xk yk ; bk := yk zk k = 1; 2; : : : ; n:
Then the triangle inequality
v
u v
u v
u
uXn u n
X uXn
(5) t (xk zk )2  t (xk yk ) + t (yk
2 zk )2
k=1 k=1 k=1
1
2

be omes v v v
u
u n
X u
uXn u
uXn
(6) t (ak + bk )2  t ak2 + t bk2 :
k=1 k=1 k=1

Using the Cau hy-S hwarz inequality:


X
n 2 n
X n
X
(7) ak bk  ak2 bk2 ;
k=1 k=1 k=1
we obtain the following inequality:
n
X Xn Xn Xn
(ak + bk )2 = ak2 + 2 ak bk + bk2
k=1 k=1 k=1 k=1
v
u
n
X 2 n n
uX X 2 X n
 ak + 2t ak2 bk + bk2
k=1 k=1 k=1 k=1
v
u v
u !2
uXn u n
X
= t ak2 + t bk2 :
k=1 k=1
Taking the square roots, we obtain Eq.(6), and therefore Eq.(5). (R n ; d) is alled the n-dimensional
Eu lidean spa e.
Example 3: On the other hand, given the same set R n , we an make it into a di erent metri spa e
by de ning the metri to be:
n
X
(8) d(x; y) := jxk yk j 8x; y 2 R n :
k=1
It is an exer ise to verify that this de nition of a metri satis es all the requirements of being a
metri .
Example 4: Still given the same set R n ; we produ e yet another metri spa e by de ning the metri
to be:
(9) d(x; y) := max jxk yk j 8x; y 2 R n :
1kn

Example 5: In the set of omplex numbers C , if z = a + ib; we de ne


p p p
jz j := + jz j2 = + zz = + a2 + b2 :
Then we de ne the metri in the omplex plane C as:
(10) d(z1 ; z2 ) := jz1 z2 j 8z1; z2 2 C :

Example 6: The set l2 is the set of all in nite sequen es


x = (xk ) := (x1 ; x2 ; x3 ; : : : ; xk ; : : : )
of real numbers whi h satisfy the onvergen e ondition:
1
X
(11) xk2 < 1:
k=1
3

This set is made into a metri spa e by de ning the metri as:
v
u
uX1
(12) d(x; y) := t (xk yk )2 x; y 2 l2 :
k=1

The metri given by Eq.(12) is well-de ned be ause the inequality


(xk  yk )2  2 (xk2 + yk2 )
guarantees the onvergen e of the series
1
X
(xk yk )2 ;
k=1
given the onvergen e of the two series
1
X 1
X
x2 k and yk2 :
k=1 k=1
Moreover, for the same reason, if x and y belong to l2 , then so does the sum
x + y := (x1 + y1 ; x2 + y2 ; : : : ; xk + yk ; : : : ):
The metri (12) obviously satis es the rst two requirements for a metri . We need only to verify
the triangle inequality:
v
u v
u v
u
uX1 uX1 uX1
(13) t (xk zk )2  t (xk yk ) + t (yk
2 zk )2 :
k=1 k=1 k=1
First of all, the three series onverge for the same reason already stated above, x; y and z being in l2
by assumption. The example above on the n-dimensional Euxlidean spa e implies that the following
inequality holds for all n:
v
u v
u v
u
uXn uXn uXn
(14) t (xk zk )2  t (xk yk )2 + t (yk zk )2 :
k=1 k=1 k=1
Taking the limit n ! 1; we obtain Eq.(13).
Example 7: A real-valued fun tion f on the losed interval [a; b℄  R is said to be a bounded
fun tion if 9K 2 R su h that jf (x)j < K 8x 2 [a; b℄. If a fun tion is ontinuous over the nite
interval [a; b℄ then it is bounded there. Now onsider the set C[a; b℄ of all (bounded) ontinuous
fun tions on [a; b℄: We shall de ne the sum, the di eren e, the negative, and the s alar multiple of
fun tions in C[a; b℄ as follows. For f; g 2 C[a; b℄;
(15) (f + g)(x) := f (x) + g(x);
(16) (f g)(x) := f (x) g(x);
(17) ( f )(x) := ( 1)f (x);
(18) ( f )(x) :=  f (x) 2 R :
Note that the set C[a; b℄ is losed under these three operations. Moreover, in C[a; b℄ there is a zero
fun tion de ned by
(19) 0(x) := 0 8x 2 [a; b℄:
4

In anti ipation of future topi s, these make C[a; b℄ more than just a set of fun tions. It is a (ve tor)
spa e of fun tions. In this spa e, we de ne the norm of an element f as:
Zb
(20) kf k := jf (x)j dx
a
and the indu ed metri as:
Zb
(21) d(f; g) := kf gk = jf (x) g(x)j dx; f; g 2 C[a; b℄:
a

The rst two requirements for a metri are obviously satis ed by Eq.(21). We shall show that it also
satis es the triangle inequality:

Zb Zb
kf + gk = jf (x) + g(x)j dx  ( jf (x)j + jg(x)j ) dx
a a
Zb Zb
= jf (x)j dx + jg(x)j dx
a a
= kf k + kgk:

Example 8: The set of fun tions ontinuous in


 R , denoted by C(
); has another, more useful
metri indu ed by the norm:
(22) kf k := sup jf (x)j := supfjf (x)j : x 2
g
x2

(23) d(f; g) := sup jf (x) g(x)j:


x2

Let (X; d) be a metri spa e, and let A  X . The distan e from any point x 2 X to A is de ned as:
(24) d(x; A) := inf d(x; a);
a2 A
and the diameter of A is given by:
(25) d(A) := sup d(a; b):
a;b2 A

De nition: Let (X; dX ) and (Y; dY ) be metri spa es. A mapping f : X ! Y is alled an isometri
mapping if, for any two elements x1 ; x2 2 X; dY (f (x1 ); f (x2 )) = dX (x1 ; x2 ); i.e., the mapping f
preserves distan es.
Clearly, an isometri mapping f : X ! Y is inje tive. If it is also surje tive, then the two spa es
X and Y are said to be isometri and, as far as the theory of metri spa es is on erned, the two
spa es are identi al and indistinguishable, regardless of the intrinsi nature of their elements.
It should be obvious that isometry is an equivalen e relation in the lass of all metri spa es.
Hen eforth, whenever we speak of a metri spa e with a parti ular metri property, we a tually
mean not a parti ular metri spa e, but an entire equivalen e lass of isometri spa es. In other
words, by the term \metri spa es" we a tually mean the lass of all metri spa es modulo isometry.
5

2. Open Sets
Let (X; d) be a metri spa e, and x0 2 X . The open ball B (x0 ; r) with radius r and entered at x0
is de ned as:
(26) B (x0 ; r) := fx 2 X : d(x; x0 ) < rg:

r
x0

The open ball B (x0 ; r) is never empty, for it ontains at least its enter point. In Example (1), the
ball ontains exa tly one point - its enter. In R with its usual metri d(x; y) = jx yj, the ball
B (x0 ; ) is the open interval (x0 ; x0 + ).
De nition: Let A be a subset of a metri spa e (X; d). A point a 2 A is alled an interior point if
there exists a r > 0 su h that B (a; r)  A. The set A is alled an open set if ea h one of its points
is an interior point.
Example 9: The open interval (0; 1)  R is an open set. Every point in this interval is an interior
point (re all that the points 0 and 1 are not in the interval). On the other hand, the losed-open
interval [0; 1) is not an open set. The reason is that the point 0, whi h is in the interval, is not an
interior point of the interval. One annot onstru t an open ball (0 ; 0 + ) that will be totally
ontained in the interval [0; 1).
Example 10: The open disk in R 2 given by the equation x2 + y2 < 1 is an open set. On the other
hand the disk x2 + y2  1 is not an open set, for if we take a point of the disk, say, (1; 0), we annot
onstru t an open ball B ((1; 0); ) : (x 1)2 + (y 0)2 <  2 whi h is ompletely ontained in the
disk.
Theorem 1. In any metri spa e (X; d), both X and ? are open sets.

PROOF: Sin e X is the full spa e, then it ontains every ball entered at any of its points, hen e it
is open. The empty set ? is open sin e, having no point, it trivially satis es the de nition of the
open set. :)
If we take R, together with its usual metri d(x; y) = jx yj as our metri spa e, then the interval
[0; 1) is not an open set. However, if our metri spa e onsists of the interval [0; 1) and the metri
d(x; y), then this interval, being the full spa e, is open. The essential point here is that a set is
neither open nor not-open on its own, but only in the ontext of the spe i metri spa e being
onsidered. In other words, the set property of being \open" is not an intrinsi property of the set.
It depends partly on the ambient metri spa e.
Example 11: In a metri spa e, every open ball is an open set. For, let x 2 B (x0 ; r). then
d(x; x0 ) < r. Now let  := r d(x; x0 ). Then B (x; )  B (x0 ; r).
6

Theorem 2. Let(X; d) be a metri spa e. Then an arbitrary union of open sets is open; while any
nite interse tion of open sets is open.
PROOF: Let fO g be an arbitrary lass of open sets labeled by the index , and let:
[
O := O :

Let x 2 O. Then x must be in at least one O . Sin e the latter is open by hypothesis, then there
exists an open ball B entered at x and totally ontained in this O  O. Hen e B must be totally
ontained in O also. Sin e x is an arbitrary point of O, O must be open.
Now let U be the nite interse tion of the sets O :
n
\
U := O
=1
If U = ?, then U is open. We assume then that U is not empty. Let x 2 U . Then x must
be in ea h of the O , for = 1; 2; : : : ; n. Moreover, sin e ea h of these n sets are open, then
9B (x; r ); = 1; 2; : : : ; n, su h that B (x; r1 )  O1 ; B (x; r2 )  O2 ; : : : ; B (x; rn )  On . Now, let
r be the smallest of the n radii, r = min(r1 ; r2 ; : : : ; rn ). Then the ball B (x; r) entered at x, with
radius r is ontained in ea h of the O , hen e it is ontained in U . Sin e x is an arbitrary point in
U , then U is open. :)
Note that the theorem would fail for an in nite interse tion of open sets. To see this, onsider the
lass of intervals fIk = ( 1=k; 1=k) : k = 1; 2; 3; : : : g. If we take the in nite interse tion,
1
\
I := Ik = f0g
k=1
whi h is a singleton, and is not an open set in R.
Let A be an arbitrary subset of the metri spa e (X; d). The set of all the interior points of A is
denoted by int(A), or AÆ . The following properties of interiors are easy to show:
1) AÆ  A, and is the maximal open subset of A in the sense that any other open subset of A is
ontained in AÆ ;
2) A is open () A = AÆ ;
3) AÆ equals the union of all the open subsets of A.
Example 12: If I = [a; b℄, then I Æ = (a; b). On the other hand, if J = (a; b); then J Æ = (a; b) = J .
Example 13: The set Z has no interior, i.e., ZÆ = ?.

3. Closed Sets
De nition: Let (X; d) be a metri spa e, and A  X . A point x 2 X is alled a limit point of A if,
for every  > 0, the open ball B (x; ) ontains at least one point of A - other than x, in ase x 2 A.
Example 14: Consider R with its usual metri , and onsider the interval [0; 1). Every point in the
interval is a limit point of the interval. But in addition, [0; 1) has a limit point whi h is not in it -
the point 1.
7

Example 15: In R 3 with its usual Eu lidean metri , every point in the open ball B (x; r) is a limit
point of the ball. In addition, every point in the 2-dimensional sphere S (x; r) := fy : d(y; x) = rg is
also a limit point of the open ball, although these points do not belong to the open ball.
Example 16: In R with its usual metri , the set f1,3g has no limit point.
De nition: Let (X; d) be a metri spa e. A set B  X is losed if it ontains all its limit points.
Example 17: In R with its usual metri , every losed interval [a; b℄ is a losed set.
Example 18: The set Z  R has no limit point; hen e it ontains all its limit points, and therefore
losed.
Example 19: In the metri spa e C[a; b℄ with the sup-metri , the set of fun tions bounded by K ,
ff (x) : jf (x)j  K g, is a losed set.
Theorem 3. Let (X; d) be a metri spa e. Then both X and ? are losed sets.
PROOF: Sin e X is the full spa e, it ontains all points and therefore ontains its limit points. ?
has no limit points; therefore it ontains all of them, hen e it is losed. :)
The on epts of open sets and losed sets are not \opposite" on epts, in the sense that they are
mutually ex lusive. A set may be both open and losed - like ? and the full spa e X - or may be
neither. The next theorem shows the relationship between open and losed sets.
Theorem 4. Let (X; d) be a metri spa e, and let A  X . Then A is losed () A is open.
PROOF: We rst prove suÆ ien y ()).
Let A be losed. If A = ?, then it is open by the pre eding theorem. Now suppose A 6= ?. Sin e
A is losed, it ontains all its limit points. So, for any x 2 A , x is not a limit point of A. Hen e,
for some r > 0, the ball B (x; r) ontains no point of A and therefore B (x; r)  A . Sin e x was
arbitrarily hosen, A must be open.
Now we prove ne essity (().
Assume A is not losed. Then it has at least one limit point, all it x, whi h is in A . Be ause
x is a limit point of A, then for any r > 0, the ball B (x; r) must in lude at least one point of A.
Therefore, for any radius whatsoever, the ball B (x; r) annot be totally ontained in A . Hen e A
is not open. :)
Theorem 5. The interse tion of an arbitrary number of losed sets is losed; the union of a nite
number of losed sets is losed.
PROOF: This is a dire t onsequen e of Theorem 2 and De Morgan's rule in set algebra. :)
This theorem will not hold for an in nite union of losed sets, as will be shown: Consider the set of
intervals indexed by n, In = [ 1+ n1 ; 1 n1 ℄; n = 1; 2; 3; : : : . The in nite union is I = [n In = ( 1; 1)
whi h is open.
De nition: Let (X; d) be a metri spa e, and let A  X . The losure of A, denoted by A, is the
union of A and all its limit points.
Example 20: In C , if B is the open unit disk fz : jz j < 1g, then its losure is B = fz : jz j  1g.
The following statements hold for the losure:
1) A is a losed superset of A, and is ontained in every losed superset of A - i.e., A is the minimal
element of the lass of losed supersets of A;
8

2) A =interse tion of all the losed supersets of A;


3) A = A () A is losed.

Another important on ept is that of the boundary point.


De nition: Let (X; d) be a metri spa e. A point x 2 X is a boundary point of A  X if every
open ball entered on x interse ts both A and A . The olle tion of all boundary points of A is
alled the boundary of A, denoted by A.
Example 21: Let R with its usual metri be the metri spa e. Consider the set A = (a; b℄. The
omplement A is then ( 1; a℄ [ (b; 1). The boundary points of A are a 2 A and b 2 A. Both
points are also limit points of A.
The boundary of A satis es the following statements.
1) A is losed;
2) A = A \ A ;
3) A is losed i it ontains its boundary.
De nition: A point x 2 A  (X; d) is alled an isolated point of A if 9 r0 > 0 su h that the ball
B (x; r) ontains no other points of A whenever r  r0 .
Hen e, A ontains 1) limit points of A whi h are in A, 2) limit points of A whi h are not in A, and
3) isolated points of A.
De nition: Let (X; d) be a metri spa e. A  X is said to be dense (or everywhere dense) in X
if A = X . On the other hand, B  X is said to be nowhere dense if B Æ = ? - i.e., its losure has
empty interior.
In the language of open balls, A being dense in X means that every open ball entered at any point
of X ne essarily ontains a point of A. Now, if A  X , then every ball entered at any point of A
must ontain a point of A.
Example 22: The set Q of rational numbers is dense in R [see Dieudonne℄.
De nition: A metri spa e (X; d) is said to be separable if it has a ountable dense subset.
Example 23: R is separable, sin e Q is dense in R, and is ountable.
Example 24: The omplex plane C is separable.
Example 25: The spa es R n and C n are separable.
The property of separability be omes very important later when we take up Hilbert spa es later
be ause it is ru ial in expressing a ve tor as a dis rete sum. Almost all, if not all, Hilbert spa es
you will en ounter in physi s are separable.
Example 26: The spa e of all bounded in nite sequen es of real numbers, x = (x1 ; x2 ; x3 ; : : : ; xk ; : : : ),
denoted by l1, where the metri is de ned by:
d(x; y) := sup jxk yk j;
k
is not separable. Consider the subset E  l1 whi h onsists of all sequen es of 1 and 0. We know
that E is un ountable (Theorem 5 of the previous hapter). A ording to the metri above, the
distan e between any two points in E is equal to 1. Now suppose we surround ea h point of E with
an open ball of radius 21 . Then we obtain an un ountable family of pairwise disjoint balls. Now we
9

show that any set dense in l1 is ne essarily un ountable. Let M be su h a set dense in l1. Then
ea h of the ball we have onstru ted above must ontain at least one point of M . Therefore M must
be un ountable, hen e l1 annot be separable.

4. Convergen e and Completeness


This se tion will deal with onvergen e of sequen es su h as (xk ) := (x1 ; x2 ; x3 ; : : : ; xk ; : : : ) in a
metri spa e (X; d).
De nition: A sequen e (xn ) in a metri spa e (X; d) is said to be Cau hy if, for ea h  > 0, there
exists a positive integer N su h that m; n > N ) d(xm ; xn ) < .
Note that the de nition does not say that m and n are onse utive integers! The inequality must
hold even if, say, m = 106, while n = 109, provided both are greater than N .
Example 27: The sequen e (xn ) = (1=n) in R is a Cau hy sequen e.
De niton: A sequen e (xn ) in a metri spa e (X; d) is said to be onvergent if there is a point
x 2 X alled the limit of the seqeuen e, su h that:
1) for ea h  > 0, there exists a positive integer N su h that n > N ) d(xn ; x) <  or, alternatively,
2) for ea h open ball B (x; ); xn 2 B (x; ) whenever n > N .
We also say that the sequen e (xn ) onverges to x. We often denote this by limn!1 xn = x or, more
simply, xn ! x.
Example 28: The expression of any quantity S as an in nite sum makes sense only if the latter
onverges.
1
X
(27) S= xk ; xk 2 R
k=1

and in fa t we take onvergen e as synonymous with existen e of the sum. For example, some in nite
sums and the fun tions that they de ne are:
2 3 1 k
X
(28) exp x := 1 + 1!x + x2! + x3! + : : : = xk!
k=0
3 5 7 1
X 2k+1
(29) sin  :=  3! + 5! 7! + : : : = ( 1)k (2xk + 1)!
k=0

What we mean here by P the onvergen e of the sum is this: taking Eqn. (27), for example, we take
the partial sums sn := nk=1 xk ; n = 1; 2; 3; : : : Then we form the sequen e of su h partial sums
(sn ) := (s1 ; s2 ; s3 ; : : : ), ea h element being a real number. When we say that the sum onverges, it
is this sequen e whi h whi h we a tually mean to onverge to the limit S - i.e., limn!1 sn = S .

Theorem 6. Every onvergent sequen e (xn ) is Cau hy


10

PROOF: Let  be given. Sin e (xn ) is onvergent (to x), then 9N su h that n > N ) d(xn ; x) < =2.
Similarly, m > N ) d(x; xm ) < =2. By the triangle inequality, d(xn ; xm ) < d(xn ; x) + d(x; xm ) <
=2 + =2 = . :)
The onverse of this theorem is not true. Consider the half-line I = (0; 1) with the usual metri of
R as the metri spa e, and onsider the sequen e S = (1=n); n = 1; 2; 3; : : : in it. S is Cau hy as
may be veri ed, but it is not onvergent be ause the point 0 that the sequen e \wants" to onverge
to is not in I . This example illustrates the fa t that \ onvergen e" is not an intrinsi property of a
sequen e. It is partly determined by the underlying metri spa e.

De nition: A metri spa e in whi h every Cau hy sequen e is onvergent is said to be omplete.
Example 29: R is omplete.
The ompleteness of R gives rise to the ompleteness of C with the metri (10) of Example (1). Let
(zn ) = (an + ibn) be a Cau hy sequen e of omplex numbers. Then (an ) and (bn ) must also be
Cau hy sequen es of real numbers, sin e jam an j  jzm znj; and jbm bnj  jzm znj. Therefore
it follows that 9a; b su h that an ! a and bn ! b. Now, de ne z := a + ib. Then we have
jzn z j = j(an + ibn) (a + ib)j
= j(an a) + i(bn b)j
 jan aj + jbn bj;
and sin e sin e the right side approa hes zero, so does the left side, and therefore zn ! z .
The ompleteness of R n also follows from Example (4). For, let
(xp ) = ( (x1p ; x2p ; : : : ; xnp ) ); p = 1; 2; 3; : : :
be a Cau hy sequen e in R n . Then, for any  > 0; 9N su h that
n
X
(xk p xkq )2 < 2 whenever p; q > N:
k=1
From this, it follows that
jxkp xkq j < ; k = 1; 2; : : : ; n for p; q > N:
This means that the sequen e (xk1 ; xk2 ; xk3 ; : : : ; xkp ; : : : ) is a Cau hy sequen e for ea h k. By Example
(4), there exists a real number xk su h that xkp ! xk as p ! 1 for k = 1; 2; : : : ; n. If we now de ne
x := (x1 ; x2 ; : : : ; xn ) 2 R n
then we have that xp ! x as p ! 1.
Example 30: Consider the spa e C 2 [a; b℄, the same set as in Example (1), but with the metri
de ned by:
Z b 1=2
(30) d(f; g) := jf (x) g(x)j2 dx ;
a
the so- alled L2 metri . Eq. (30) obviously satis es the rst two requirements of a metri . That it
satis es the triangle inequality an be established with the help of S hwarz's inequality:
Z b 2 Zb Zb
(31) f (x)g(x) dx  f 2 (x) dx g2 (x) dx:
a a a
11

This spa e is not omplete. We shall exhibit a Cau hy sequen e in this spa e whi h fails to onverge.
Consider the sequen e ('n (x)) de ned over the interval [ 1; 1℄ with ea h 'n (x) given by:
8
< 1
> 1x 1
n
n x n
'n (x) = nx 1 1
>
: 1  x  1:
1 n

This is a Cau hy sequen e in the the spa e C 2 [ 1; 1℄, sin e


Z1
j'n (x) 'm (x)j2 dx  minf2m; ng :
1
But the sequen e ('n (x)) annot onverge to any fun tion in C 2 [ 1; 1℄. To see this, onsider rst a
dis ontinuous fun tion (
(x) =
1 for x < 0;
1 for x  0:
Now, let f be an arbitrary fun tion in C2 [ 1; 1℄. Using S hwarz's inequality (still valid for pie e-wise
ontinuous fun tions)
Z 1 1=2 Z 1 1=2 Z 1 1=2
jf (x) (x)j2 dx  jf (x) 'n (x)j2 dx + j'n (x) (x)j2 dx :
1 1 1
By the ontinuity of f , the left side annot be zero. Also,
Z1
lim
n!1
j'n (x) (x)j2 dx = 0:
1
Therefore, it follows that
Z1
jf (x) 'n (x)j2 dx
1
annot onverge to zero, hen e 'n annot onverge to f as n ! 1.
We have used the terms limit and limit points. Is there a onne tion between the two? There
is, if you onsider that the points of a sequen e also make up a set. For example, the sequen e
(1; 1; 1; 1; : : : ) is onvergent, and its limit is 1. The members of this sequen e, however, make up the
singleton f1g, whi h has no limit point. On the other hand, the sequen e (xn ) = (1; 21 ; 13 ; : : : ; n1 ; : : : )
has 0 as its limit, and 0 is also a limit point of the set f1; 21 ; 13 ; : : : g.
Theorem 7. The limit of a sequen e is a limit point of its set of points if the sequen e has in nitely
many distin t points.

PROOF: Let (X; d) be the underlying metri spa e, and let (xn ) be a onvergent sequen e with
limit x. Suppose x is not a limit point of the set fxn g. Then there exists an open ball B (x; )
whi h ontains no point of the sequen e other than x itself. But sin e (xn ) onverges to x, then all
the elements xn for some value of n onwards must lie inside the ball B (x; ). Hen e they must all
oin ide with x, and sin e there is only a nite number of points of the sequen e (xn ) outside of the
open ball, the sequen e has only a nite number of distin t points. :)
12

De nition: Let (X; d) be a metri spa e. A subset Y  X , together with the metri d whi h it
inherits from X , form the metri spa e (Y; d) alled a metri subspa e or subspa e for short, of (X; d).
Theorem 8. Let (Y; d) be a subspa e of a omplete metri spa e (X; d). Then Y is omplete i it
is losed.
PROOF: We prove ).
Assume Y is a omplete subspa e of X . Then every Cau hy sequen e in Y onverges to a point
in Y . Let y 2 X be a limit point of Y , taken as a subset of X . We shall show that Y is losed
by showing that, in fa t, y 2 Y .For ea h positive integer n, the open ball B (y; n1 ) must ontain an
element yn of Y . Clearly, the sequen e (yn ) onverges to y in X , and is Cau hy in Y . Sin e Y is
omplete by assumption, then y 2 Y , whi h makes Y losed.
We now prove (.
We assume Y is losed, and show that it is omplete. Let (yn ) be a Cau hy sequen e in Y . Then
it is also Cau hy in X , and sin e X is omplete by hypothesis, the sequen e onverges to a point
y 2 X . We need to show that y 2 Y also. If the sequen e (yn ) has only a nite number of distin t
points, y is that point whi h is in nitely repeated in the sequen e, and hen e is in Y . If, on the
other hand, (yn ) has in nitely many distin t points, then by Theorem (7), y is a limit point of the
set fyn g  Y , hen e it is a limit point of Y itself, and sin e Y is losed, y 2 Y . :)

We now address the question: if a metri spa e (X; d) is not omplete, is there an operation by
whi h we an omplete it? In the most elementary ase as, for example, the spa e I = (0; 1℄, all
that is needed is to adjoin the point 1; then the spa e I = [0; 1℄ is omplete. In general, a metri
spa e (X; d) whi h is not omplete an be isometri ally imbedded in one whi h is omplete. Hen e
we have the ompletion theorem whi h says that any metri spa e X has a ompletion X .
5. Continuous Mappings
We re all the de nition of ontinuity in elementary al ulus: A fun tion f : R ! R is ontinuous at
x0 if, for any given  > 0; 9 Æ > 0 su h that jf (x) f (x0 )j <  whenever jx x0 j < Æ. We generalize
this de nition to metri spa es other than the real numbers.
De nition: Let (X; dX ) and (Y; dY ) be metri spa es. A mapping f : X ! Y is ontinuous at a
point x0 2 X if either of the two equivalent onditions are satis ed:
1) For ea h  > 0; 9 Æ > 0 su h that dX (x; x0 ) < Æ ) dY (f (x); f (x0 )) < ;
2) For ea h open ball B (f (x0 ); ) in Y , there exists an open ball B (x0 ; Æ) in X su h that f [B (x0 ; Æ)℄ 
B (f (x0 ); ).
The mapping f is simply said to be ontinuous if it is ontinuous at all points of its domain.
Theorem 9. Let (X; dX ) and (Y; dY ) be metri spa es. A mapping f : X ! Y is ontinuous at x
i xn ! x ) f (xn ) ! f (x).
PROOF: We prove ne essity; we assume that f is ontinuous at x, and we show as a onsequen e
that if xn ! x, then f (xn ) ! f (x).
Constru t the open ball B (f (x); ) entered at f (x). By our assumption on the ontinuity at x of
f; 9 B (x; Æ) su h that f [B (x; Æ)℄  B (f (x); ). Sin e xn ! x, then all the xn 's for n greater than
some N lie in B (x; Æ). Then by the set in lusion above, all the f (xn )'s must lie in B (f (x); ), whi h
means that f (xn ) ! f (x), whi h is what we wanted to prove.
13

Now we prove suÆ ien y; we show that if xn ! x implies f (xn ) ! f (x), then f is ontinuous at x.
Assume that f is not ontinuous at x. Then by the de niton of ontinuity at a point, there is an
open ball B (f (x); ) whi h has the property that the image under f of every open ball B (x; Æ),
whatever Æ may be, an never be ontained in it. Now onsider a sequen e of nested open balls,
B (x; 1); B (x; 12 ); B (x; 13 ); : : : ; B (x; n1 ); : : : . Form a sequen e (xn ), where xn 2 B (x; n1 ) while
f (xn ) 62 B (f (x); ). Then obviously, we have xn ! x, while the sequen e (f (xn )) does not onverge
to f (x). :)
The pre eding theorem and the de nition of ontinuity of a mapping give us the obvious extension:
Theorem 10. Let (X; dX ) and Y (dY ) be metri spa es. A mapping f : X ! Y is ontinuous i
xn ! x ) f (xn ) ! f (x):

What these two theorems in e e t simply say is that ontinuous mappings are those mappings whi h
are onvergen e preserving - i.e., the image of every sequen e onvergent to a point is also onvergent,
and onverges pre isely to the image of the point. This is, in a tual situations, a more useful riterion
in determining the ontinuity of a mapping instead of the formal de nitions involving the 's and
the Æ's.
The next theorem is equally useful, and in those ases where one does not have a metri spa e at
all, but only a topologi al spa e (as in the ase of Lie groups, for example), this theorem in fa t
be omes the de nition of ontinuous mappings.
Theorem 11. Let (X; dX ) and (Y; dY ) be metri spa es. A mapping f : X !Y is ontinuous i
f 1[O℄ is open in X whenever O is open in Y .

6 ?,
PROOF: Assume f is ontinuous, and let O be an open set in Y . We assume that f 1 [O℄ =
otherwise it is trivially open. Let x 2 f 1 [O℄. Then f (x) 2 O and be ause O is open, there exists
an open ball B2 := B (f (x); )  O. The de nition of ontinuity then implies that there exists open
ball B1 := B (x; Æ) su h that f [B1 ℄  B2  O. This means that B1  f 1 [O℄, and sin e x was
hosen arbitrarily, then f 1 [O℄ is open.
Now assume the onverse: f 1 [O℄ is open in X whenever O is open in Y . Let x be an arbitrary
point in X , and onsider the open ball B2 := B (f (x); ). This is an open set ontaining f (x), and
therefore its inverse image f 1 [B2 ℄ is also an open set (by our hypothesis) whi h ontains x. Hen e
there is an open ball B1 := B (x; Æ)  f 1 [B2 ℄. Clearly, f [B1 ℄  B2 . Thus, f is ontinuous at x by
our de nition. Sin e x was arbitrarily hosen, f is ontinuous on X .:)
A note on some nuan es of ontinuity. We de ned the mapping f : X ! Y as ontinuous at x 2 X
whenever dX (x0 ; x) < Æ ) dY (f (x0 ); f (x)) < ; for some . One may ask, for a given , what
determines the size of the orresponding Æ? The answer is: for the given f under dis ussion, the size
of Æ is de nitely dependent on . However, it may also depend on x (as an example, onsider the
fun tion f (x) = 1=x) - i.e., Æ = Æ(x; ). But for some fun tions, the size of Æ is determined solely by
. In otherwords, on e  is given and a Æ is found, that same Æ works throughout the entire X , so
that Æ = Æ(). In that ase, we say that the fun tion f is uniformly ontinuous.
De nition: Let f : X ! Y be a bije tive mapping between two metri spa es. If both f and f 1
are ontinuous, then f is said to be a homeomorphi mapping or a homeomorphism (between X and
Y ). Metri spa es X and Y are said to be homeomorphi if there exists a homeomorphism between
them.
14

If two spa es are homeomorphi , then they have the same topologi al stru ture. What this means is
that open sets in one spa e is arried to open sets in the other spa e by the homeomorphism f or
its inverse. From the point of view of topology, the two spa es are indistinguishable. The relation
of being homeomorphi is an equivalen e relation.

De nition: Let (X; d) be a metri spa e. A mapping A : X ! X is alled a ontra tion mapping
if 9; 0 <  < 1, su h that
(32) d(Ax1 ; Ax2 )  d(x1 ; x2 )
for any two points x1 ; x2 2 X .
The following theorem is one of the most useful theorems. In parti ular, this is used to establish the
existen e theorem in ordinary di erential equations.
Theorem 12 (Fixed Point Theorem). Let (X; d) be a omplete metri spa e. Every ontra tion
mapping of X onto itself has a unique xed point.
PROOF: Let x0 be an arbitrary point in X , and denote by A : x 7! Ax the ontra tion mapping.
Let
(33) x1 := Ax0 ; x2 := Ax1 = A2 x0 ; : : : ; xn := Axn 1 =    = An x0 ; : : :
The sequen e (xn ) is Cau hy for, assuming expli itly that m < n, then
d(xm ; xn ) = d(Am x0 ; An x0 )  m d(x0 ; xn m )
 m [d(x0 ; x1 ) + d(x1 ; x2 ) +    + d(xn m 1 ; xn m )℄
 m d(x0 ; x1 )[1 +  + 2 +    + n m 1 ℄ < m d(x0 ; x1 ) 1 1  :
Sin e  < 1, the expression on the right, and therefore the distan e between xm and xn , an be
made arbitrarily small by hoosing n suÆ iently large.
By the ompleteness of X , the Cau hy sequen e (xn ) has a limit, x = limn!1 xn . Now, sin e A is
ontinuous (see lemma after this),
Ax = A nlim
!1 xn = nlim
!1 Axn = nlim !1 xn+1 = x;
whi h proves the existen e of the xed point. We prove next the uniqueness of the xed point.
Suppose there are two xed points,
Ax = x; and Ay = y:
Then, by Eq. (32), we have
d(x; y)  d(x; y):
But sin e  < 1, this an only be satis ed if d(x; y) = 0. Therefore x = y. :)
Lemma 1. Every ontra tion mapping is ontinuous.
PROOF: Exer ise. :)

You might also like