l\1ETRIC SPACES

E. COPSON
.;;
Regi'U8 Profe880T"of Mathematios in the
University of Andrew8
1968
Published by the Syndics of the Cambridge University Press
Bentley House, P,O, Box 9.2, 200 Euston Road, London, N.W.
Amerioan Branoh: 32 East 57th Street, New York, N.W.I0022
Cambridge University Press 1968
Library of Congress Ca.talogue Card Number: 68-18343
Standard Book Number: 521 04722 6
Printed in Great Britain
at the Unive:rsity Hooae, Cambridge
(Brooke Crutohley, University Printer)
I i
CHAPTER 1
IN'I'RODUCTION
1.
A collection of objects occurring
thus member xl> a second member
indefinitely. For every positive integer k, there is a corresponding
kthmember of the sequence. The members of such a sequence need
not be all different. We can have a sequence all of whose members
are the same; such a sequence is called a constant sequence.
If {k'l'J is a strictly increasing sequence of positive integers,
the sequence {x",,} is called a subsequence of {x
n
}. The definition
implies that {x
n
} is a subsequence of itself.
A numbers is said to converge
value of e, all but a finite U,LUUVL
between x - e and x + e.
converges, every subsequence I'nrltTf',l'fU""
to sequence of real numbers which PlUllm,rmlQ
null-sequence. Thus if {x
n
} converges
null -sequence.
It is often convenient to represent real numbers by points on
a line, and to speak of the point of abscissa x simply as the point x.
The distance between the points x and y is Ix-yl. To say that
the sequence of real numbers {x
n
} converges to x is thus the same
thing as saying that the sequence of points {x
n
} converges to the
point x, or that the distance between the point Xn and the point
x tends to zero as n -+ 00.
In frequently speak of the complex UU,I,llUUL
complex plane. That the sequence
"'''"rc,T',",,><' to Z means that the distance
the point Z in the complex
geometrical language is very
geometry is used. All we need
is between two points satisfies
inequality IZ1-Zzl ~ IZ1-Zsl+lzs-Z21.
[1] eMS
2 :?tDJTRIC Sl'ACE
As the complex plane is merely a geometrical picture, we are
not compelled use Euclidean formula IZI for dis-
tance between two In extended complex plane, that
is, the ordinary complex plane with an added ideal point at
infinity, it is sometimes more convenient use the' distance'
between the points Zl Z2 the eX"DrE:SSl.on
21
Z
1-
Z
21
.J{(1 + IZlI2) (1 + IZ212)}
This is known as the chordal distance since it is the length of the
chord joining the points corresponding to Zl and Z2 on the
Riemann sphere whose stereographic projection the extended
complex plane. This 'distance' satisfies the triangle inequality;
and when we are dealing with convergence t,o a finite limit, it does
not matter which definition of 'distance) we use,
There are, however, important differences. The complex
plane is unbounded, that is, we can find pairs of points whose
distances apart are as as like. But the extended complex
plane with the chordal definition of 'distance' is bounded, since
no pair of points is at a 'distance' apart which exceeds 2. The
'distance' from to the point infinity is 2/.J(1 IzI2). say
that the sequence {zn} tends to the limit 00 is equivalent to saying
that, for every positive number e ( < 2), all but a finite number of
members of the sequenee are at 'distance' less than 6 the
point at infinity; it is readily seen that this condition reduces to
4 I}l
IZnl>
for all but a finite number of values of n, as it should since we
can choose 6 ( 2) so (46-
2
.qt is large we please.
A sequence might be a sequence {xn(t)} of real functions, each
continuous on an interval a ::;;; t ::;;; b. If this sequence converged
for value in the interval, its limit x(t) would be a funetion
defined on the same interval; {xn(t)} is then said to be pointwise
convergent. The sequence is said to converge uniformly to x(t)
on the interval for every positive value of e, there exists an
integer no, depending e but not on such whenever
n ~ no, the inequality IXn(t) -x(t)1 < 6
are
dis-
that
.t at
nce'

the
Lded
!ity;
loes
)lex
lOse
)lex
ince
rhe
say
ring
Irof
the
s to
we
!tch
ged
ion
rise

an
ver
INTRODUC'l'ION
of the interval;
x(t) is itself continuous,
another definition of uniform
depends on the idea
that, for every
supremum Mn. This
Ixn(t)-x(t)1
3
satisfied,
and that, for every positive value of e, there exists a number to
of the interval such that
Ixn(t
o
) - x(to) I > Mn -e.
The sequence {xn(t)} converges uniformly to x(t) if and only if
to zero as n 00.
alternative definition of uniform pn,nrc'r'O',p]1('A
idea in a geometrical
on a t b, a 'point';
hl'ir.""'An two 'points' x(t) and
definition which satisfies
sequence of continuous
continuous function to saying
that the 'distance' from the 'point' xn(t) to the 'point' x(t)
tends to zero as n 00.
We might have defined the 'distance' between the 'points'
x(t) and y(t) as
When the sequence of' points' {xn(t)} converges to x(t) with this
definition of 'distance', the sequence of functions {x.,,(t)} is said
mean of order 2 to
'A.UIUU'" convergence in mean
example, {ntf(1 +n2t2)}
zero, but is not uniformly
the properties of sets
geometrical property is the '-'A,'C>V'JU"V
pair of 'points' is topology.
It had its beginnings in the theory of linear sets of points, which
arose in the 19th century from a discussion of the theory of
1'2
)
f
INTRODUCTION 5
of all real numbers between 0 and 1 is not countable: the real
numbers between and 1 cannot arranged a sequence,
We said that a set is countable if its members can be arranged
as a sequence, But a sequence is not a set, not even if all the mem-
bers the sequence are distinct. reason for this is that set
is a collection of distinct entities, considered merely as a collec-
tion; but in sequence the order important. example, from
the set of positive integers, we can construct many different
sequences; 2,3, 5 ... and 1, 2,5, 7, 4, 9, 11,6, . are two
different sequences containing all the positive integers.
Sets are usually denoted by capital letters, members of a set
by small letters. The notation a means that is a member of
the set A; aA that a is not a member of A. The empty set-
there only one empty e t ~ i s denoted 0.
A 'finite set can be defined by writing down explicitly the
elements which are members. Tbus {I, 3, 5,7} a set with
members, the odd natural numbers less than 8. Again {O} is the
set whose only member is the number zero; it is not the empty
set -it has member.
Certain capital letters are reserved to denote particular sets.
Some these are
N: the set of natural numbers 0, 1,2,3, ...
the set of real numbers.
R2: the set of all ordered pairs (x, y) of real numbers.
R3; the set of ordered triples (x, z) of real numbers.
O[a, b]: the set of all functions x(t) continuous on a ~ t ~ b.
Sometimes is convenient to define a by stating property
which is possessed by all the members of the set and by no
element which does not belong to the set. Thus set 3} might
be defined as the set of all real 'numbers x which satisfy the
equation x
2
- 5x + 6 = 0; we should then denote the set by
{xeR: x
2
_ 5x+ 6 = OJ. Again the set {I, 5, 7} could defined
as {xeN: x == 1 (mod 2), x < 8}. The set {xeN: x == 0 (mod 2),
== 1 (mod is the empty set, since there is no natural number
which is both even and odd.
6 METHIC SPACES
Sets and subsets
The sets A and B are said to be equal, A = B, if every member
of belongs to and every member of B belongs to A.
'1'he set A is said to be contained in B if every member of A
belongs to B. We then write A s; B. In this case, we may
say that B contains A, and write B 2 A; or again that A is a
subset of B. If A = B, then A S Band IJ 2 A; and conversely.
The set A is said to properly contained in B member
of A belongs to B and there is at least one member of B which
not belong t.o A. We write A B or B ~ A, and
that A is a proper subset of B.
Some authors write A c B whenever A is a subset of B, not
a proper subset B, aud do not the notation
S,2.
Evidently,
if A c and sO, then S; 0;
if A s Band B c 0, then A cO;
A c and S 0, then cO;
if A c Band B c 0, then A c O.
The empty set 0 is a subset of every set. For if not, thoro
would exist a set A of which 0 is not a subset; this would mean
that there is a member of 0 which does not belong to This
impossible, since 0 has no members.
4. The complement of a set
Let us now consider sets which are all subsets of some fixed
set E. The complement of a subset A with respect to E is the set
of members of E which do belong to We shall denote
by A'. The following results are evident fi'om the definition:
(i) E' )0;
(ii) 0' = E;
(iii) (A')' A;
(iv) A s; B if and only if B' SA'.
This notation is satisfactory so as are considering
INTRODUCTION 7
complements with respect to one fixed set E. When complements
with respect to several sets occur, other notations are sometimes
used, such as CEA, or E-A.
5. Finite unions and Intersections
The union of a family of subsets of the fixed set E is defined to
be the set of members of E, each of which belongs to at least one
subset of the family. The intersection of a family of subsets is
defined to be the set of members of E, each of which belongs to
all the subsets of the family. In either case, the family may con-
sist of a finite or of a infinite number of subsets of E; and if it
consists of an infinite number of subsets, the family is not neces-
sarily countable.
The union A v B of two subsets A and B of the fixed set E is
the subset of E which consists of those members of E which
belong to A or to B or to both.
The intersection A f'\ B is the subset of E which consists of
those members of E which belong to A and to B. If A f'\ B = 0,
A and B have no common member and are said to be disjoint.
The intersection A f'\ B' of A and the complement of B with
respect to E evidently consists of the members of A which do not
belong to B. It is called the relative complement of B with respect
to A.
If A, B, 0 are three subsets of E, then (A v B) v 0 = A v (B v 0).
For if xe (A v B) v 0, then xeA v B, or xe 0, or x belongs to
both. IfxeO,xeBvOand thereforexeA v (BvO). IfxeA vB,
then x belongs to A or to B or to both; if xeA, xeA v (B v 0);
if xeB, then xeBv 0 and therefore xeA v (BvO). Thus if
xe(AvB)vO, then xeAv(BvO),
and so (A v B) v 0 s;; A v (B v 0).
Similarly A v (B v 0) s;; (A v B) v O. Hence the result. We may
therefore leave out the brackets and write A v B v 0 for this
triple union. And, evidently, the triple union does not depend on
the order of the sets A, B, O.
If n is any positive integer and A
l
, A
2
, , An are subsets of E,
the set
8
METRIC SPACES
consists of the members of E which belong to at least one of the
subsets Ai' It is denoted by
n
U
i=l
If A. three subsets then (A " B) A " (B " C),
The proof is similar to that given above. Again we may leave out
the brackets, and write simply A" B" O. The intersection
consists members which belong the subsets
AV
A
2' ,An
If A and B are any two subsets of E, then (A u B)' = A' "B'
and (A" B)' = A' u B'.
If x (A u B)', xA u B,
Similarly , and so
Again, r.B', xA'
xA u is, x (A
result follows.
Hence x A and therefore x A',
B'. Thus ; A' "B',
B. Hence
uB)'. The
The second result can be proved in the same way. Alternatively,
it can be proved by taking complements. Remembering that
A" = A, have A uB)" = . Writing
A' = 0, A" = 0', B D'; and so
O'uD'
6. Infinite unions and intersections
Let S be any set, finite or infinite; it might be the positive
integers 0 and 10, be all the integers,
it might all the numbers between l. Suppose
that, to each member of IX there exists
subset of E which we denote by.A.
a
We then have a family of
subsets of E with index set S. We denote the union and inter-
section of the family by
Evidently
INTRODUCTION 9
The union countable family CO'untable i8 countable.
A countable family sets can arranged as a sequence
<IJ
.AI' A
2
, A
3
, Let Am = {am l' am 2' am 3' }. Then U Am consists
, " m=l
all am,n' and these can arranged sequence
If the sets are not disjoint, there will be repetitions in this
sequence.
00
the \:Jl".U!t:l.'!UJ:! of U
m=l
form a subsequence this
sequence when repetitions are omitted. Hence U Amis countable.
m=I
It follows from this that the set of all rational numbers is
countable. have secn that the consisting of all rational
numbers that ~ 1 is countable. follows that the
set AI' consisting of all rational numbers x such that 1 < x ~ 2
is countable since they are in one-one correspondence with the
members Similarly for An' set of all rational numbers
such that x ~ n for any or n.
The set of all rational numbers is
AOvA! vA_
1
vAllvA_2v ... ;
hence the
7. The algebra of sets
Let A, B, 0 be subsets of a given set E. Then the operations of
intersection complementation have the following
properties:
(a) A vA' = E,
(b) AvA = A,
Av A,
(d) A" 0 = 0,
A"A'= 0.
A "A = A.
"E =
AvE=E.
(e) A vB = BvA,
A"B=B"A.
(A u 0 =.A vO),
"B) u A" (B
(g) A ; A v B, A ;;;;2 A"
(h) (A v B) " 0 = (A" 0) v (B" 0),
(A B) vO =
0)" (B C).
10 METRIC SPACES
These properties have been arranged
pair be deduced from other by
of the previous section, viz. (A v B)' = A' f'I B' and its equiva-
lent (A f'I B)' = A' vB'. Thus it is necessary only to prove half
these formulae; fact, of them are obvious from
definitions of and intersection.
In (U),itisevidentthatA A vB.Hence
A';;2
B)' = B'.
ReplaceA',B'byG,D. ThenG;2 0 f'lD.
To prove the first formula in (h), note that, if xe (A v B) f'I 0,
then belongs Au Band O. But A v B, belongs to A
or to or to hence belongs f'I 0 or 8 f'I 0
both. 'I'hus xe (A f'I 0) v (B f'I G), and so
(A u B) f'I 0 c:; (A f'I 0) v (B f'I 0).
Again, xe (A v (B x belongs to A or to
or to both. If x e A f'I 0, x belongs to A and to 0; therefore x e A vB
and xeO and so xe (A v B) f'I 0; and similarly if xeB f'I O. Hence
(A f'I (B f'I (A v B) , which completes proof.
To obtain the second in (h), take complement
of the first formula, which gives
B)' v 0' (A f'l0
/
) (B f'I 0)'.
Hence (A' f'I B') v 0' = (A' V 0') f'I (B' V 0')
or, replacing A', B', 0' by their complements,
B)v (AvO) vO).
8. Equivalence relations
Two natural numbers a b may connected by one
many relations; for example, a < b, a == b (mod 3), a is a divisor
of b, a is a multiple of b, and so on. All these are examples of
what called binary relations. Such relations occur in
other branches In plane geometry, the property
that a point a is conjugate to a point b with respect to a fixed
INTRODUCTION 11
conic is a binary relation. They also occur in every-day life: on
the set of all human beings, we have, for example the binary
relations' a is the father of b' or 'a is of the same nationality as b',
have a given relation often write
that the element a the given
element b of E.
obvious what a relation difficult to
of defining a relation introducing the
fLTi'I'WI"wn '/lTC/WILC/, of two sets. Let X taking
first a member x of X and then a member y of Y, we obtain an
ordered pair (x.y). The set of all such ordered pairs is called the
Cartesian product of X and Y, and is denoted by X x Y.
Now suppose we have a relation,.., defined on a set E. This
means that for each x belonging to E, there is at least one y
belonging to E such that x,.., y; in other words, for each x which
belongs to E, there are one or more ordered pairs (x, y) such that
Ex,.., y. But the set pairs is
snbset of the Cartesian Thus a
E is a non-empty
relation defined said to be
all a belonging to symmetric
"'"' a. It is said to be b, b ,.., c
implies a,.., c. For example, on the set N of natural numbers,
the relation < is transitive but not reflexive or symmetric; the
relation 'is a divisor of' is reflexive and transitive, but it is not
symmetric; the relation of 'congruence (mod3)' is reflexive,
symmetric and transitive. A binary relation which is reflexive,
symmetric, and transitive is called an equivalence relation.
If ,.., is an equivalence relation on E, the set of all members
,.., a is called an let us de-
by a. For examplc, ""',''''''''" .... ,
congruence (mod 3) on thc
equivalence classes
... }, {1,4,7,1O, ..
equivalence
numbers;
If ,.., is an equivalence relation on a set E, a ,.., b implies a = 0,
and conversely.
METRIC SPACES
Suppose that a '" b. Let c a. Then c f'oJ a. Since the relation is
transitive, this implies thut c f'oJ so C Hence a ;
But a f'oJ b implies b f'oJ a since the relation is symmetric, and
hence b :;; a. Therefore a = b.
Now suppose that a = b. By the reflexive law, b "" b and so
b b. But this implies that b a, that is, b f'oJ a.
a of E, we mean a collection of non-empty subsets
of E such that each member of E belongs to one and only one
subset of the collection.
If f'oJ is an equivalence relation on a set E, two equivalence classes
are either identical or have no common member; the collection of all
equivalence classes is a partition of ]j).
If a and b have a common member c, then c f'oJ a and c "'" b.
the and transitive a b, henee a =
Thus if a and b are different, they have no common member.
Let d any member of Sinee d..., dEll. Henee every
member of E belongs to an equivalence class; and since two
equivalence classes cannot overlap, it belongs to precisely one
equivalence class.
9. The axioms of the system of real numbers
There are various ways of defining the real number system.
can define a real number as a decimal form which may terminate,
recur or not recur, as is done, for example in P. Dienes's book
TILe TGlylor Series (Oxford, 1931). One can define a realllumher
as a Dedekind section of the rational numbers. One can define
a real number as an equivalence class equivalent Cauchy
sequences of rational numbers, the Cantor construction. The
interesting thing is it does not matter which definition
adopt. Each method generates an Archimedean ordered field
with the supremum property, and any two such fields are iso-
morphic. In popular language, this means that, whatever
finition of the real number system is adopted, any arithmetical
operation leads to the same answer.
A field is defined to be a set S of elements which may be com-
bined by two operations, ealled addition and multiplication,
and these operations satisfy the following basic laws.
INTRODUCTION 13
(i) The laws of closure. Addition and multiplication are
well defined. This means that, for every ordered pair a and b
of members of S, there are unique elements a + b and a x b
belonging
(ii)
ofS,
(iii)
c of S,
CIJ'lI'/ifflij'LULI,'(Ve laws. For every pair of members
b+a, axb = b
laws. For any three band
(b+c), (axb)x
(iv) The existence of zero and unity. S contains unique and
distinct elements 0 and 1 such that, for every member a of S,
a + 0 = a, a x 1 = a.
(v) The existence of additive and muUiplicative inverses.
For every a belonging to S, there exists a unique member of S,
denoted that a+ (-a) = O. For belong-
ing to S, unique member of S, such
that a x
(vi) law
+c) = (axb)+(ax
Briefly, commutative additive omit the
element 0, it is also a commutative multiplicative group; and it
satisfies the distributive law (vi).
A field S is said to be ordered if it contains a set P of' positive'
members with the properties:
(i) P is closed with respect to addition and multiplication.
(ii) If a is an element of the field S, then either a = 0, or
aPor
When this satisfied, the relations
(defined to be b + (-
with 0 and 1
n",eii """,fl field.
said to have the supre'rn'U'rn
subset of S, which is
14 ME SPACES
A subset A of S is said to be bounded above if there is an ele-
ment k in such x < for every member x of Such
element is called an upper bound. If k is an upper bound, any
element" of S greater than k is an upper bound. The point of the
definition that the set of upper bounds of A has a member,
which is called the 8upremurn or least upper bound of A, and
which is denoted by sup A or sup x or sup {x:xeA}. Ifkl = supA,
e.4
then kl and every member greater than an upper bound,
but no member of S less than kl is an upper bound. An equivalent
form of the definition is:
ordered field has the supremium property if and only
every non-empty subset of S which is bounded below has an in-
fimum in S.
infimum of subset is its greatest lower bounp-, and
denoted by infA or infx or inf{x:xeA}.
::I'.4
ordered field of rational numbers does not have
supremum property. For example, the subset A of Q, consisting
of the partial sums of the series
is bounded above, each partial sum is less than But
is no rational number which is the supremum of A.
These are the axioms of a real number system. A detailed
discussion be found for example Birkhoff and
Survey of Modern Algebra (New York, 1953).
10. Properties of the real numbers
By a system of real numbers, we mean an ordered field with
the supremum property. analysis, we use of
representations to which we have referred; it does not matter
which. We often do not need to use a representation; we only need
certain basie properties which follow from the axioms and do
depend on a concrete representation. .
Let R be a system of real numbers. The positive integers in R
arc defined the obvious by 1 = + 2 3, and
INTIWDUCTION
S of real numbers x such that an> x for at most a finite number
of values of n. The set S is bounded below since {an} is a bounded
sequence; H be infimum of We that there
subsequence of {an} which converges to H.
For every positive value of e, an > H + e for at most a finite
number of values of but > H - for an infinite number
values of n. Let kl be the least positive integer such that a
k1
lies
between H L Let be the least positive integer greater than
kl that aka lies between H i. Next ka be the least
positive integer greater than k2 such that aka lies between H 1.
And on indefinitely, by definition of II. In way,
construct a strictly increasing sequence {k
n
} positive integers
such that, for every positive integer n,
1 1
H - - < a
k
< H + - .
n n n
Hence the {a
kn
} converges to H,
No subsequence can converge to a limit greater than H.
The number which is denoted by lim sup or lim aU)
n-+-cx;)
sometimes called the limes superior of the sequence {an}.
By a similar argument, we can prove the existence of a real
number h the property that no subsequence can converge
to a limit less than h and that there exists a subsequence which
converges to h. The number k, which is denoted by lim inf an

or by lim an' is sometimes called the Umes inferior of the sequence.
_<Xl
If the sequence {an} converges to a limit a, we have H = h = a)
since every subsequence converges a. Conversely H
the sequence is convergent.
A sequence of real numbers is said to be a Oauchy 8equence if,
for every positive value of there exists a positive integer
such that
whenever m n
A necessary and sufficient conditionfor a seq'uence to be convergent
is that it be a Oauchy sequence.
result is known as Owuchy's l)rinciple convergence.
2 eMS
18 METRIC SPACES
{an} a sequence of rmtl numlJers converges to
real number a. Then, for every positive value of e, there exists
a positive integer no such that
-al <
whenever n ;;:. no. It follows that, if m > n ;;:. no,
ani
The sequence is therefore a Cauchy sequence.
let be a Cauchy sequence, so that, for every
of e, there exists a positive integer no such that
lam-ani < e,
whenever > n no' This implies that laml < lanol +e,
that the sequence is bounded. It follows that the sequence con-
tains a subsequence {ak
n
} which converges a a.
k
m
m, we have
whenever > n ~ Makcm-+ then
la-ani ~ 6,
whenever n ~ no, which means that the seqnence {an} converges
to The condition thus sufficient.
In the terminology to be introduced in Chapter IV, the real
number system is complete,
12. Functions and m a p p i n ~ s
the of functions a complex variable, meet
idea of a conformal mapping. We say, for example, that the
half-plane rez > 0 is mapped conformally onto the unit disc
Iwl 1 by relation
z-
W=--
z+ l'
When we say that this is mapping, we mean that relation
associates with each point z of the half-plane a single well-defined
point w of the disc. (That this mapping conserves angle is of no
interest in present context). This idea mapping call
extended easily to abstract sets.
lNTRODUOTION 19
Let E and F be two non-empty sets (possibly even the same
Bya j of into F, we mean a which
sociates with element of E a well-defined element
of F, which is denoted by j{a) and called the image of a under j.
This mapping is denoted by !:E -+ F. The set E is called the
U01nU't,n of set of points which
uu.U>i'i'JPofpeints ofEis the range of the The range
may be the whole of F or a proper subset of F; it may even be a
single point of F, in which case the mapping is called a constant
mapping. In example drawn from complex theory,
domain thehalf-planere z > range disc Iwl
since each point of the disc is the image of a point of the half-
plane.
elementary calculus, often use different reIlre!!entatlOn
function-its graph. graph = x
2
out all
points (x, y) of the Cartesian plane for which the ordinate y is the
square of the abscissa x-it picks out all the ordered pairs (x, y)
for which y in the given relation to x. The graph of the
function is subset the Carteshm plane.
Given two abstract sets E and F, the Cartesian product E x F
was defined as the set of all ordered pairs (x,y) where xE,
YEP, A function or mapping/: E -+ P can be defined as a subset
ExF the property that, for each aE, contains
actly one element (a,j(a)).
This definition of function of mapping is not restricted to be
merely a generalization of a function of one variable, The domain
the could well Cartesian product
x Y. The mapping j: X Y -+ F then associates with each
ordered pair (x,y) where XX, y Ya single well-defined
element of F.
the of functions of a the
OC)
f(z) of a power series anz
n
is an analytic function regular on its
o
disc of convergence Izi < R. If 0 < R < 00, it is sometimes
to continue j(z) tmalytically outside the By
mean tbtt is to find analytic function
regular in a region of which Izl < R is a proper subset; and on
the disc of convergence, g(z) = j(z). The process of analytical
20 METRIC SP.AC
continuation extends the domain of definition of analytic
functionf(z).
A similar may arise in the mapping of one abstract set
into another. Let E and F be abstract sets, and let El be a proper
subset of E. Suppose f:E
1
is a mapping of El F.
If there is a mapping g: E ~ F such that f(x) g(x) for every x
belonging Ev g called an extentJion off, f the restriction
of to E
1

CHAPTER 2
METRIC SPACES
13. The definition of a metric space
Let E be any set. Let p(x, y) be a function defined on the set
Ex E of all ordered pairs (x, y) of members of E, and satisfying
the following conditions:
(i) p(x,y) is a finite real number for every pair (x,y) of
ExE;
(ii) p(x, y) = 0 if and only if x = y;
(iii) p(y,z) ~ p(x,y)+p(x,z), where x, y, z are any three
elements of E.
Such a function p(x, y) is called a metric on E; it is a mapping of
E x E into R. A set E with metric p is called a metric apace. The
members of E are frequently called 'points', and the function
p(x,y) the 'distance' from the 'point' x to the 'point' y.
Different choices of metric on a given set E give rise to different
metric spaces. For example, the metric space consisting of all
complex numbers with metric Ix - yl is not the same as that con-
sisting of all complex numbers with metric
2l x -yl .
.J{l + IxI2}.J{1 + IYI2}'
and both are different from that consisting of all complex
numbers with metric
Ix-yl
l+lx-YI"
14. Properties of the metric
H, in (iii), we put z = y, we get
2p(x,y) ;;;= p(y,y) = 0
so that p(x, y) is a rum-negative real function. H we put z = x,
we obtain
p(y,x) ~ p(x,y) +p(x, x) = p(x,y).
[21 ]
22 METRIC SPACES
But x and yare arbitrary points, a similar argument gives
p(x, y) p(y, x).
Hence p(x, y) = p(y, . thus p(x, is function.
Using the symmetry property, we obtain from (iii)
p(y, z) p(y, x) + p(x, z)
or changing the symbols,
(iv) p(x, y) p(x, z) + p(z, y).
If.x, y, z are points in the complex plane with the usual Euclidean
metric, this inequality asserts that the sum of the lengths of two
sides of tl'iangle is not less than length of the third side.
For this reason, (iii) or its equivalent (iv) is called the Triangle
Inequality.
Since
p(x,y)-p(x,z) p(z,y) = p(y,z),
and p(x,z)-p(x,y) p(y,z),
we
(v) p(y,z) ;a. Ip(x,y)-p(x,z)l.
15. Bounded and unbounded metric spaces
Let M he a metric space consisting of a set E and a metric p
on E. If there exists positive number k such that p(x, y) :S;; k for
every pair of points x and y of E, we say that M is a bounded
metric space. A metric space which is not bounded is said to be
unbou1Uled; that case, p(x, takes values as large as we please.
If the metric space M consisting of a sot E with metric p is
unbounded, we can define in many ways a bounded metric
space Ml consisting of the same set E with a different metric.
For example, it suffices to put
(
X ) = p(x,JI>-
Pl ,y l+p(x,y),
That this function does satisfy the conditions for a metric is
proved follows.
(i) Since p 0, PI (x, y) = if and only if p(x, y) = 0, that is,
if and only if x = y.
METRIC SPACES
(
") '( ) ( ) p(x, y) p(x, z)
II Pl X, Y +Pl X,Z = 'f+p(x,y) + i +p(x,z)
p(x,y) +p(x,z)
~ ~ ~ ~ ~ ~ ~ ~
l +p(x,y)+p(x,z)
= l/{l+ P(X,y):P(X,Z)}
~ 1/{1+ P(:,Z)}
= Pl(y,
23
Since 0 ~ Pl ~ 1, Ml is a bounded metric space. The metric Pl is
important because is closely related the metric p. As we
shall see in 61, the family of open sets in M is identical with the
family open sets in 111
1
; the two metrics arc then said to
equivalent.
16. The diameter of a subset of a metric space
Let A be a subset of a metric space M. Consider the set of non-
negative real numbers
{p(x, y) :xeA, yeA}.
If this set is bounded, it has a supremum, denoted by 8(A), which
is called the diameter of A. If the set is unbounded, it is conven-
tional to write o(A) + 00, and to say that the set has an infinite
diameter. It can be easily shown that a necessary and sufficient
condition for a non -empty set A to consist of a single point is that
8(A) = O.
The diameter 8(M) of the whole space is finite if M is bounded,
but is infinite if M is unbounded. For every subset A of M,
8(A) ~ 8(M).
17. The distance between two subsets of a metric space
I.et A and
real numbers
be two subsets of a metric space M. The set
{p(x,y):xeA, yeB},
is bounded below by zero. Its infimum, denoted by d(A,B),
24 METRIa SPACES
is called the distance between the subsets A and B. Since
p(x,y) = p(y,x),d(A,B) = d(B, A).
The equation d(A, B) 0 does imply that A und B
points in common. For example, if A and B are the subsets x > 0
and x < 0 of the real line with the usual metric, d(A, B) = 0,
yet A B have no common
If A consists of a single point a,
(l(A, B) inf{p(a,y):ye
This is called the distance of the point a from the subset B, and
is denoted by d(a, B). The equation d(a, B) = 0 does not imply
that a belongs
Lastly, if A and B consist of single points a and b respectively,
the distance between the subsets A and B reduces to p(a, b).
18. Subspaces
Let }YI be a metric space consisting of a of. points E
metric p; and El be proper subset The metric p
function defined on Ex E. Let PI be its restriction to the subset
El x E
1
, that is,
y) = y)
whenever x and y are points of E
1
Then
is the ind'uced Thc EI
a subspace of
19. Examples metric spaces
is a metric on E
I
:
metric Pl is
The simplest metric is the discrete metric which can be used
to metrize any set E. It is defined as follows: if x and yare any
two of E,
p(x, x) = 0, p(x, y) = 1 (x *' y).
Its properties very special, but it is sometimes useful.
space discrete metIic bounded.
The set R of all real numbers with metric p(x, y) = Ix - YI is a
metric space. It an unbounded metric space which is
the real line.
The extended real number system consists of all real numbers
together with two ideal elements and with property
METRIC SPACES 25
that -00 < x < 00 for every real number x. Ifwe define a function
f(x) by
x
f(x) = llXl' f( +00) = 1, f( -(0) = -1,
the p(x, y) = If(x)
extended set of real
is called the extended
metric we could
real number We could apply the
resulting
extended
Another
possible metric is
p(x,y) = Itan-1x-tan-1yl,
where -11T < tan-
1
x < t1Twhen -oo<x< 00, tan-
1
( +00) = t1T,
tan -1 ( - 00) = - t1T. Again we get a bounded metric space, since
no two points are at a distance which exceeds 1T.
The metric space 0 of all' complex numbers with metric
Ix-yl is also an unbounded space; it is called the complex plane
or the The extended system
consists numbers together number,
infinity. representation of the provided
by the projection of the + = 1
onto taking the point vertex of pro-
jection.
the point (g, we get on
= 0 the point of affix g + i1)
z =
corresponding to the point of affix z on the plane we get the point
1),
point at infinity,
The chordal distance
pair of points
2/x-y/

vertex of
the points
eXWlllueu complex
satisfies the condition for a metric. The resulting metric space is
denoted by O.
METRIC 51' ACES 27
with equality if and only if
~ (r 2, ... ,
that is, if and only if at' = kb
r
for every value of r, k being inde-
pendent of r.
Now write
= C
r
,
= dr'
where c
r
and d
r
are not negative,
1 1
= p' = q'
where p and q are greater than unity and
1
-+ 1.
P
It follows that, if c
r
~ 0, d, ~ 0,
with equality if and only if ~ = k d ~ for every value of r, k being
independent of r. result known as Holder's Inequality.
The particular case p = called Inequality.
Now consider
wherea'j' b, ~ 0
n
~ (ar+br)P,
1
>1.
1 1
-=1--
q .
as
n
Then I: (at' + br)P I: ar(a,. + br)P-l + br(a, + br)P-l
1 1 1
(
n ) lip (n )l/q
+ 1 bf 1 (a
r
+ br)q(p-l)
28
METRIC SPACES
Butq(p -I) = p, and I_q-l = p-l. Hence
with
(
'" ) lip ('" ) lip ('" ) lip
:pa,.+b,.)p +
if a,. = kb,. for every
known as Minko'Wski's
inequalities reduce
a,.b,.)2
1 1
{f(a,.+b,.)2t
The corresponding results for integrals can be proved in the
same way. They are that, iff;;:: 0, g;;:: 0, p > I, q = p/(P-I),
then
f: fgdx {J: (f(XPdXf'P {J: (g(Xadxr'a,
{J: (f+ {J: (f(x)}Pdxf'P +
whenever
be finite
UIIJOl',lttli::\ oxist. The limits ex,
21. Some sequence spaces
may
The set of all sequences x = {x",} of real or complex numbers
can be turned into a metric space by taking
where !:.A1I
function OC""'''Uu"
15, Ix-
( )
A Ix", - Ynl
p x, y = 1 '" 1 + Ix", - Ynl '
convergent series of
conditions for a
co
Ix-YI does. < the space
1
The set sequences x = {x
ll
}
numbers can be metrized by
p(x,y) = sup{lxn-Ynl:n = 1,2,3, ... }.
This
since
complex
METRIC SPACES 29
But this metric space is unbounded, since for every positive
number k, p(kx, ky) = kp(x, y).
A more difficult example of a sequence space consists of all
real or complex that I x n l p
fixed real value sequence
space
p(x, y) = {f lx'll.
show that p(x, y) number for
every pair x, y of points ofthe set. By Minkowski's inequality,
Hence
for
and
Since
convergent, and
points of the set.
are the same point.
finite value
y) = 0 if
{f lx'll. -YnIPf'P ~ {flxnIPf'P +{f IYnIPr'P,
the sequence {xn-Yn} belongs to the set whenever {xn} and {Yn}
do. Let a = {an}, b = {b
n
}, C = {cn} be three points of the set.
Putting xn = an -b
n
, Yn = an -cn in the last inequaltiy, we have
{f lan-bnIPf'P +
p(a, b)+p(a, c),
30 TRIO SPACES
EXERCISES
1. E is the set of all ordered pairs x = (xu X2) of real numbers. Prove
that p(X, y) = max {Ix
l
- YII, IX2 - Yall is a metric on E.
2. the set all functions x(t) continuous on the bounded closed
interval a oE; t oE; b, Prove that
p(x,y) = sup{lx(t)-y(t)l:tE:[a,b]}
is a metric onE.
3. E is the set of all differentiable fWlCtions x(t) whose uerivatives are
continuous on the bounded closed interval a oE; t oE; b, Prove that
p(x,y) Ix(a) + sup {Ix'(t) :tE:[a,
is a metric on E.
4. p(x, y) is a motric on Bot E. Is p2(X, y) a metric? Is min {p(x, I}
a metric
5. lathefunction
p(x,y) = I y I
-1+.J(I+y
2
)
a metric on the real line? If p(x, +oo),p(x, -oo),p( +00, -00) are defined
as limits, is the flllction a metric on the extended real line ?
6. a metric space metric
metric Pa(x
a
, Va). Prove that
(i) PI(XI, Yl) + P2(X2, Y2),
(ii) max {Pl(X1 , Y2)}'
(iii) ",{pi(xl , YI) + Ya)}
are metrics on M a.
is a metric space with
7. A and B arc nonempty subsots of a. motric space. Prove that, if
A c: B, then SeA) oE; 8(B).
8. and Bare non.empt,y subsets of a metric space, Prove if
An B isnotempty, then8(A B) oE; +S(B),
CUAPTER 3
OPEN AND CLOSED SETS
22. Spheres
If a is a point of a metric space M with metric p, the set of all
points x of M such that p(a, x) < r, where r > 0, is called. a sphere
with centre a and radius r or a spherical neighbourhood of the
point a of radius r. Although the word 'sphere' is used in ele-
mentary sense 'spherical surface'
usage as the set of points
in a metric rarely. Some authors call
neighbourhood translation of the French
but there advantage in this, since
hollow like or solid like a golf balL
is warned sphere and ball, will be
literature.
In the theory of functions of a complex variable, the 'sphere'
I z - a I < r is often called a disc. Again, there is no reason for
using the term 'disc' for a spherical neighbourhood in a metric
space. Although no question of dimensionality arises, a word
suggesting a flat object might be misleading. There is, in fact, no
really satisfactory word; so we content ourselves by using
'sphere'.
The
N(a; r),
hood'.
If Mis
and radius r will be UGJl.i.U',uU
as an abbreviation for
metric Ix-yl, N(a; r) is
< x < a+r. But if M
real numbers x::::; I, N(!; 1) is the
but N(O; < !. as also is N(-h;
two results queer; but it must be remembered in
them, M consists only of real numbers for which 0 ::::; x ::::; I.
The Euclidean plane consists of all ordered pairs
x = (Xl' x2), Y = (YI' Y2)' .
[31 ]
32
of numbers
METRIC SPACES
metric
p(x, y) = .J{(X
I
- YI)2 + (X2 - Y2)2},
and the spheres are ordinary circular discs. But if we impose on
the set R x R of all ordered pairs of real numbers the metric
p(x, y) = max{lx1-Yll, IX2 -Y21h
the spheres are squares I Xl - < r l I x
2
- < r, And if we
impose the discrete metric on the same set, each sphere ofradius
not exceeding unity consists of a single point, every sphere of
radius exceeding unity the whole metric space.
The set of all points X of M such that p(a, x) ~ r, where r > 0
is calIed a closed 8phere with centre a and radius r, and be
denoted by K(a; r).
23. Open sets
If is a subset a metric space M, the a of A is said to
be an interior point of A if a is the centre of a sphere which con-
sists only of points of A, The set of interior points of A is
called the interior of A and is denoted by lIlt A or ..4. If every
point of A is an interior point of A, A is called an open set. For
example, the line, the set of points x whieh h < k
is an open set, and is usually denoted by (h, k). Again, the set of
points of the real for which h is also open: it is denoted by
(h, +00) .
. A 8phere an open set.
H b is a point of the sphere N(a; r), then p(a, b) = r' < r. If x
belongs to N(b; r r'), we have
p(a, x) ~ p(a, b) + p(b, x) < r' + (r-r') = r
so that r - S; r). Hence is an interior point of
N(a; r), which proves the result. This includes the result for the
real since the interval (a- a+r) the sphere N(a,r),
The empty set is an open set.
To prove this, we have to show that rp contains no point which
is not an interior point; is true since 1> contains points.
If M is a metric space, M is open.
TIlls is obvious, since evory is point M,
OPEN AND ED SETS 33
A subset 0 of a metric space M is an open set if and only if it is
the union of a family of spheres.
The set is open set,
family spheres. Conversely,
spheres is empty, and so is open.
it is the union of
union of empty
If 0 is a non-empty open set, each point of 0 is the centre of a
sphere contained in and 0 union the family of all
such Conversely, if 0 is union non-empty family
Y of spheres, and if b is any point of 0, b lies in a sphere N(a; r)
belonging to Y. If b is not the centre of this sphere, then it is the
centre sphere . r') contained in r), and hence
contained O. Thus, either b is an interior of 0,
which was to be proved.
The union of a family of open sets in a metric space is open.
Each set of family is family of SDJ[lOI:es.
Hence union of tho family union of a family
of spheres, which proves the result.
The intersection of a finite number of open sets in a metric space
l,s open,
The 'finite' is essential in enwlciation. For consider
the metric space consisting of the real line with metric Ix-yl.
The interval 0 < x < 1 + lin is open for every positive integer n.
But the intersection this infinite is the interval
o < x ~ 1, which is not an opeu set.
Let 0
1
and O
2
be two open sets. If 0
1
or O
2
is empty, or if
0
1
and O
2
are disjoint, 0
1
1"'10
2
is the empty set which is open. So
we have to consider the when none of 0
1
, O
2
) n O
2
is empty.
Let 0
1
1"'1 O
2
= O. If a is any point of 0, it is a point of the open
set 0
1
, Thus a is the centre of a sphere N(a; contained in 0
1
;
similarly the of a N(a; contained O
2
,
l'he sphere N(a; r) where r is the smaller of r
1
and r
2
, is therefore
contained in 0
1
and in O
2
, and this implies that N(a; r) is con-
tained in 0, This proves that a is interior of 0 and hence
that 0 is The for the intersection finite number
of open sets follows by induction.
The interior of any subset A of a metric space M is the largest
open set contained in
3
34
METRIC SPACES
A h,1S no interior points, its interior is empty set which
is open, and A contains no other open set.
If the interior of A is not empty, let a be interior point
that there is a r
1
) contained A. If b any
of N(a; r
1
), there is a sphere N(b; r
2
) contained in N(a; r
1
) and
hence in Therefore b is also an interior point of A, The interior
A is therefore union of a of spheres, so is
open set.
If B is any open set contained in B is a subset of the interior
Hence the interior of is the largest open subset of A.
Adherent points
If A is a subset of a metric space M, the point a of M is said
to be an adherent point of A if every sphere with centre a contains
point Adherent are two
An adherent point a of a subset A of M is called an isolated
point if there is a sphere with centre a which contains no point
other than a itself. If every of A an isolated
A is called an isolated set.
An adherent point of A which is not an isolated point is called
u point of accumulat'ion of A. If a is point of accumulation
every sphere with centre a contains a point of A distinct from a,
and this implies that every sphere with centre a contains an
infinite number of points A. For the N(a; contuins
a point a
1
of A distinct from a. Ifp(a, a
1
) = r
1
, the sphere N(a; r
1
)
contains a point a
2
of A distinct from a and a
1
And so on in-
definitely, should be noted that point accumulation
is not necessarily a point of A.
The point a in a metric space M is an adherent point of the subset
and if d(a, A) =
Since d(a,A) = inf{p(x,a):xA}, the equation d(a, A) = 0
implies that every sphere N(a; r) contains a point of A; hence a
adherent point of A. if a is adherent point
A, it is either an isolated point of A or a point of accumulation.
In either case, d(a,A) = O.
36 METRIC SPACES
and that c does belong to '1'here-
fore ; Au
generally, closure of the of any family
of subsets of M is the union of their closures. For an infinite
family, it is not necessarily the case that the closure of the union
is the union of the closures of the subsets of the family. If Acx is
an family subsets of M, index set we can
general is
UAcx ; 00.4:.
os os
To prove that An B ; An B, we observe that An B ; A.
Henco A n B ; A, similarly A B Therefore
AnB;A
More generally, the closure of the intersection of any family,
finite or infinite, of subsets of a metric space is contained in the
intersection of their closures. It should be noted that it can hap-
A nBc For A and may be that
numbers,
numbers, A n B = {O; but A = B = R.
26. Interior, and Frontier Points
sometimes write A - for
of subset A. closure of
ment of A, but A _I the complement of the closure.
if
closure
comple-
a is an interior point of a subset A if there exists a sphere
N(a; r) contained in A. The interior of A, denoted by IntA
or A, the set of interior points an open
exterior A if it
A I. The P'!:I"VT'U'"
interior
of all
exterior points of A, and is therefore the interior of A'. The
exterior of A, denoted by Ext A, is therefore also an open set, and
ExtA IntA', ExtA' IntA.
Thc a is an ,n"','""'O''' point of and only is a
OPEN AND CLOSED SETS 37
N(a; r) of A, that and only
exterior of A
It follows that
exterior
space M,
ExtA = A-'.
IntA = A'-'.
interior of not nelceSiSalrIl}
(ExtA uIntA), = (ExtA)' n (IntA)'
fill up the
= A-" nA'-" = A-nA'-,
set An A' necessarily it is the frontier
and is denoted Fr A (or sometimes by A ' A and A'
the same frontier. The frontier of A is the complement
the open set lnt A u Ext A, and this, as we shall see later, implies
that the frontier of A is a closed set.
The boundary of a set A, denoted by BdA, is the part of the
frontier of A belongs to ; hence
A nA'-,
= A'n
Evidently, if A is closed, its frontier is its boundary.
The boundary of a set contains no non-empty open set.
For if BdA contained a non-empty open set 0, 0 would be
contained in IntA is largest open of A, 0
,'muId be contained Int A. impossible
o s;; Bd A S;; ]'r A S;; (Int A)'.
Hence there is no non-empty open set contained in BdA. The
frontier of a set does not necessarily have this property; for
OA,a,LLllJ"", on the line, the of the rational
numbers is the line.
27. Dense subsets
A subset A of a metric space M is said to be everywhere dense
if M is the closure of A. exterior of everywhere
inM,A to be nowhere dense in M.
38 METRIC SPACES
The 8Uhset A of a metric space M is nowhere dense in M if and
only if the closure of A has no interior points.
If A nowhere the of Ext is M, that
A-'-= M.
Taking complements, A-'-' = 0
and so lnt 0.
And conversely.
A metric space M is said to be separable if there is a countable
subset everywhere dense in M. For example, the real line is
separable; the closure of the countable set rational numbers
is the real1ine.
A subset A of a metric space M is said to be dense with respect
to another subset B if B A. Evidently every subset is dense
with to itself. This is the same saying every
subset is A subset A of a metric space M is
dense-in-itself when every point' of A is a point of accumulation
of A. A subset which is both closed and dense-in-itselfis said to
be perfect.
28. Closed sets
Just used letter 0 denote an set (un ensemble
ouvert), so we shall use the letter F to denote a closed set ('un en-
semble ferme). There are other conventions; this seems a con-
venient one.
It from definition in a metric space every
finite set, the empty set and the whole space are closed sets. It will
be noticed that M and 0 are both open and closed. Sometimes
there are other subsets of M with this property. For example,
consider metric space M consisting of complex numbers
such that Iz - 11 < lor Iz + 11 < I,'with metric p(zv Z2) = IZl - z21.
Then Iz-II < 1 is an open subset of M. By a theorem we shall
prove shortly, the complement of IZ-II < 1 with respect to M
is closed. 'l'his complement is Iz I < 1 which is also Thus
Iz+ 11 < I, and similarly Iz-11 < 1, are subsets of M which are
both open and closed. Naturally Iz + 11 < 1 is not a closed set
40
METRIC SPACES
of all nUIhbers x such that o and positive integers,
with the metric Ix-yj, we have
N(O; 1) = R: - ~ x ~
K(O; 1) = {xR:-l:e.; x ~ O}u{I},
so that the closure of N(O; 1) is a proper subset of K(O; 1).
The 'intersect'ion of a family of closed sets in a metric space is
closed.
Let be a of sets with index set S. The comple-
mellt their intersection,
(nFa), = U F ~ ,
E.S E.S
is open, being the union of a family of open sets. lIenee n Fa is
closed. Similarly we can show that
The 'union oj finite n2tmber closed in a metric space is
closed.
The word 'finite' is essential here. For example, if {Fn} is
the sequence of closed subsets lin x ~ the line, their
union
QO
U Fn
n=l
is the set 0 < x ~ 1 which is not a closed subset of the real line.
The closure of a subset A of a metric space M is the intersection
of all closed sets containing A.
There are closed sets containing A; for instance A and M.
Let F be the intersection of all closed sets containing A; then F
IS closed. But A ; F, As;; = F. Hence every closed set
which contains A contains A. But is a closed set containing A,
while F, being the intersection of all closed sets containing A,
IS contained in each of them. particular F s;; Therefore
A = P. The closure of A is thus the least closed set containing A.
29. Subspaces a metric space
IfMI is a subspace of M, a set which is open inM
I
is not neces-
sarily open in a set which is dosed in is not necessarily
closed M. For example, if M the real line R the usual
AND CLO ETS 41
metric and Jf
1
the subspace 0 ~ x ~ 1, the subset 0 ~ x < t
is open in Ml but not in M.
In order that B s; Ml b,e open in the subspace Ml of a metric space
s'ullicient that a set 0,
other words, open sets in intersections
with the open sets of M.
If a is a point of M
1
, denote by N1(a; r) the set of points of Ml
such that p(a, x) < r. The sets N1(a; r) are the spheres of Ml since
the metric of Ml is the induced metric. Evidently
r) = M1f"l
B be a subset open reJative M
1
Then allY
point of B, there is a sphere N1(x; r(x)) contained in B, where the
notation indicates that the radius r(x) depends on x. We then
have
B = U N
1
(xj r(x)) = Ml n U N(x; r(x)) = Ml nO,
xB xB
= UN(x;
xB
being the union of a family of spheres of M, is open with respect
to M. This proves the condition is necessary.
Conversely, let B = 0 n Ml where 0 is open relative to M.
If x is any point of B, x is a point of 0, and so there exists a
N(x; r(x)) containod in O. Hence
~ ( x ;
nN(x;
so that x is an interior point of the subset B of the subspace M
1

As x was any point of B, B is open relative to M
1
, and so the
condition is sufficient.
order that every B of the M
1
, which
necessary and that Ml be
open in that the condition necessary.
If B is open inMl' B = 0 nM
1
where 0 is open in M.lfffI
1
is
open in M, then B, being the intersection of two open sets, is
open in M. Thus the condition is sufficient.
In order that B s; be closed in the subspace Ml of the metric
JJ!, it is necessary and sufficient exist a set closed
such that B
42
B is closed
is closed in Ml
that
SPACES
B' "M
I
is open in M
I
,
exists a set 0, open in
O"Mp
If this condition is satisfied, we have B u M ~ = 0' v M ~ by
taking complements. Hence
B = B" MI = (B u M ~ ) "M
I
= (0' u MD "M
I
= 0' "M
I
,
so that B is the intersection ofM
l
and the closed set 0'.
Conversely, let B = F "M
I
, where F is a closed subset of M.
Then B' = F' u Mi, and so
B' "M;. = (F' u M ~ ) "M;. = F' "M
I
,
where F' is open
In order that
in Mv be closed
inM.
Ml is closed
closed in M
I
,
in M, B, being
condition is satisfied,
the subspace M
I
, which
nece881aru and sufficient that
condition is necessary.
is closed in M. If Ml
two closed sets, is closed
Thus the condition is sufficient.
The closure of a subset B of the subspace Ml of a metric space M
with respect to MI is B "MI'
If a is an adherent point of B with respect to Mv every sphere
inMI' NI(a; r) say, contains a point of B. But
~ ( a ; r) = N(a; r) "M
I
; N(a,r).
Hence a is an adherent point of B with respect to M. Thus
aeB "MI' The respect to Ml is "V>lV,,",ll1<OU
B "MI'
Conversely,
B with respect
B with respect
evidently an adherent
MI is contained in the
"Ul,llP'OU'"'' the proof.
EXERCISES
1. N(a; r) is 1:1. sphere in the metric space M. A is a subset of M which
intersects N(a; r) and has diameter less than r. Prove that A S N(a; 2r).
2. Prove that the diameter of a. subset of a metric space is equal to the
diameter of its closure.
AND OLO TS 43
3. M is a metric space consisting of a set E with metric p. Ml is the
metric space coni'Jisting of the same set E with metric Pl' The metrics p
and Pl are said to be equivalent if the open sets of M are the open sets of
Ml and conversely. Prove that the metrics p andpJ(l +p) are equivalent.
the set of all pairs x =
= ./{(x
l
- -Ya)2}. The "Uli.UCRIU
max {/xl Yan
is also a metric on the same set. Prove that p and Pl are equivalent.
S. A is a set of points in a metric space M; B is the set of all points of
accumulation of A. Prove that B is closed.
E, 0 are subsets metric space B,
dense with respect Prove that
Prove that the results are equivalont:
(i) A is everywhere dense in the metric space M.
(ti) The only closed set which contains A is M.
(ill) The only open set disjoint from A is the empty set.
(iv) A intersects every non.empty open set.
(v) A intersects every sphere.
Show that a. subset a metric is nowhere and
every sphere M eontains a from points
Show that any of a separable metric space is
10. A and B are disjoint closed subsets of a metric space M. Show that
U = {xM:d(x,A) < d(x,B)}, V = {xM:d(x,B) < d(x,A)}
are disjoint open sets containing A and B respectively.
is an open metric space any subset
a"A,
o A.
Deduce that, if 0 "A is empty, so also is 0 " .if.
12. If A and B are two subsets of a metric space M,
Fr(AuB) ; FrA uFrB.
that, if .if are disjoint, =FrAu
Prove that the c;vl,uIJu:a.u.o ofthe
is the closure
Show that
(i) FrA ; FrA,
(ti) Fr Int A ; Fr A,
(ill) A = IntA u FrA.
Prove that
B) ExtA B.
Prove
metric
44
METRIC SPACES
15. Prove that the frontier of a subset A of a metric space M is empty
if and only if A is both open and closed.
16. Prove that an open set contains none of its frontier points.
18. If A
defined by
open sets in a metric space
Fr(01" Os)
(01 "FrO
s
) u (Os"
metric space M, the
IntA, {J(A) = IntA.
are
Provethatcxa(A) s; a(A),{J{J(A) ;2 {J(A).
If 0 is an open set, F a closed set, prove that 0 s; a(O), F;2 {J(F).
Deduce that, for any set A, cxa(A) = a(A),{J{J(A) = {J(A).
19. A and B are nonempty subsets of a bounded metric space M with
metric P; d(x,A) and d(x, B) are the distances of a point x of M from
A and B respectively. If x is a given point of M, show that, for every
positive number e, there exists a point b of B and a point a of A such that
Deduce that,
Hence show
+e, p(a, b) ~ d(b,
{d(x,A):xe:B}, then
... PAB+d(x,B) +2e.
third non-empty subset
PAC'" PAB +PBC.
is the family of all closed non-empty subsets of M. Show that
max (PAB,PBA) is a metric on .
20. M is any set, and P is a real valued non -negative function such that,
for all members x, y, z of M, p(x, y) = 0 if and only if x = yand
Show that p
Show that,
closed, and
Prove also that,
closed spheres
21. The
Ms is the set
p(x,y) .... max (P(x,z),p(y, z.
space, every sphere and
sphere K(a; r) is closed.
8) have a point in common, one of these
the other.
the set E1 with metric
The metric space
space
to be
the set E1 x Ea with metric
p(x, y) = max {Pl(X1, Yl),P2(X2, Ya)}
AND CLOSED SETS
where x = (xl,x
a
), y = (YvYa) are any two pointsofE
l
X E !fa = (al,a.)
is any point ofMl x MIand ifr > O,provethat
r) = r) x N
2
(a
2
;
r) = r) x K
S
(a
2
;
Deduce that, if Al is open in Mv As open in M
a
, then Al x AI is open in
MlXMa
Show subset of f l , A 2 a of M
2
, t,ho closure
that a Ilecessary and sufficient condition
Al be closed inMl,andA2 be closed
CHAPTER 4:
COMPLETE METRIC SPACES
30. Convergent sequences
The sequence of points {an} in a metric space M is said to con-
verge to point M th.e distance a) tends to zero as
n -+ 00, that is, if for every positive value of 6 there exists an
integer no, depending on e, such that
0:::; p(an,a) < 6,
whenever n ~ no' The point a is caned the limit of the sequence.
say the limit, beeause sequenee can converge to two
for if {an} converged to a and to b, we should have
o p(a,b) p(a
n
, +p(a
n
, -+0
as n -+ 00, and so p(a, b) = 0 which is impossible since a and b
distinct.
particular, = a all but finite number of values
the sequence {an} converges to a.
If {k
n
} striotly increasing sequence of positive integers, so
k
n
;<: and k
1
-+ 00 as -+ 00, the sequence { a ~ is called a
subsequence of {an}; and if an -+ a, ak,. -+ a.
\Ve that sequenee is not a set. The elements a
sequence are ordered and are necessarily distinct;
elements in a set are not ordered and are distinct. For example,
could define sequence by a
2n
b, a
2n
+1 = c, b c;
values taken an form a set 'with only two members.
If a is a point of accumulation of a subset A of a metric space M,
is a sequence {an} of points all distinct a, wMch
converges to a.
Every sphere with centre a contains a point of A distinct from
(which mayor not belong A). with any point a
1
of A, and let r
1
= min (l,p(a,a
1
)). 'rhen the sphere N(a; '1)
contains apointa
2
of A, distinct from a. Let r
2
= min (t,p(a,a
2
));
sphere N(a; contains a aa of distinet from a.
[46 ]
COMPLETE METRIC SPACES 47
N(a; ra) a point
of A, indefinitely. way, we
construct a sequence of distuUJt POUlts {an}, all distillct from a,
such that, if r n = min (l/n,p(a, an)), the sphere N(a; r n) contaills
the poillt a
Ml
of A. Thus p(a,a
n
) < r
n
-
l
l/(n-l), and so
p(a, an) tends to zero as n -+ 00. Hence the sequence {an} of poillts
A converges the poillt accumulation a.
If a and b two point8 of a space and a 8equence
which converges to b, p(a, b
n
) converges to p(a, b).
By the triangle illequality,
Ip(x,y)-p(x,z)1 p(y,z).
we put x = b, z = b
n
have
jp(a, b) - p(a,
Sillce p(b, b
n
) converges to zero, p(a, b
n
) converges to p(a, b).
If a and b are two points of a metric 8pace, and if {an} and {b
n
}
are 8equences which converge to a and b re8pectively, p(a
n
, b
n
)
converges to
For
Ip(a, b) Ip(a, b)
p(a, an) + p(b, b
n
)
and so p(a
n
, b
n
} converges to p(a, b).
Cauchy's principle of 4'1l1nVt'rO
One of the fundamental theorems of analysis is Cauchy's
Principle of Convergence, which enables one to prove the
vergence of a sequence of real numbers without knowillg
hand what should This prillciple was proved
11 runs foHows:
The neCe8sary sufficient cond'ition for the convergence of a
8equence {an} ol real number8 is t}/,O,t,lor every p08iti've value of e,
there exi8t8 an integer n
l
(e) such that I am - ani < e whenever

A sequence of real numbers {an} which satisfies this condition
called a 8equence.
48
METRIC SPACES
The definition of a Cauchy sequence can be extended to com-
plex sequences, to sequences of real functions, and, indeed,
to a sequence of points in any metric space. The sequence of
points {an} in p is called a Cauchy "'''Rluvl.Hjv
if, for every there exists an integer
that p(a
m
, an) n ~ n
1

For a sequence 'fnet'ric space M to be convergecnt,
necessary that Oauchy sequence.
If the sequence of M converges to the
of M, then, for every positive value of 10, there exists an integer
no(e)suchthatp(a,a
n
) < ewhenevern ~ no' Henceifm > n ~ no,
o p(am,a
n
) ~ p(a
m
, a) +p(a, an) < 210.
The sequence is thus a Cauchy sequence; the occurrence of 210
instead of 10 is irrelevant.
Ina general metric space, a Oauchy sequence is not necessarily
convergent. .
For example, space consisting of all
numbers with sequence of partial sums
series
is a Cauchy sequence of rational numbers, but it does not con-
verge to a rational number. Or again, take the metric space which
consists of all real numbers x such that 0 < x < I; the sequence
{lin} is a Cauchy sequence, but it does not converge to a point of
the space.
A metric space M is said to be complete if every Cauchy
sequence of points of M converges to a point of M.
32. Convergent
A bounded
sequences
numbers either converges
sequence condition ensures
it does not that it converges.
tion is different in a general metric space. A bounded sequence
may either converge or oscillate finitely or not converge at all.
The Cauchy sequence condition ensures that it does not oscillate;
COMPLETE METRIC SPACES
the space is
which the
complete, there
converges.
be no
49
the
If a Oauchy sequence of points in a metric space contains a con-
vergent sUbsequence, the sequence is convergent.
Suppose {an} is a Cauchy sequence in a space M with metric p.
every value of e, exists an llt,egE3r
p(a
m
, an) whenever m
sequence a subsequence
to the point a of M, we have
p(akm' an) < e
{k
n
} is a
"""",,o,"H6 m ~ 00,
p(a,a
n
) ~
whenever n ;:: n1(e); hence the sequence {an} converges to a.
33, Cantor's tn1ttllw'''''''I''t,
theoreln
necessary and condition metric space be
complete is that every nest {Kn} of closed spheres, such that the dia-
meter of Kn tends to zero as n ~ 00, has a non-empty intersection.
A closed sphere K(a; r) is the set of points x such that
p(x,a) ~
closed a sequence closed spheres, of
is a proper of its predecessor,
Let the sphere Kn have centre xn and radius r n' Since
Km ; Kn ; K
no
'
centres xn belong to
rn ~ 0 as for every
IS we can the integer no that r no < Hence
p(xm' xn) < e whenever m > n ;:: no, and so the sequence {xn}
of centres of the nest {Kn} is a Cauchy sequence.
If M is complete, the sequence {x,J converges to a point x
of each closed contains a finite of
the sequence {xn} and hcnee contains x.
50 lIIETRIC SPAC
'"
n Kn contains x and sois not empty; it consists of the single
n=1
point x. Thus the condition is necessary.
Next suppose that condition is i:li:H,l1>l,II;1U
Then there a Cauchy
not niWnrt'l'tu> to a By the condition,
find a strictly increasing sequence {k
n
} of positive integers such
that
whenever m > k
n

Let the closed ybelongs
Hence p(y, x
kn
) < 1/2
n
so that y is an interior point of Kn. There-
fore Kn for positive integer
hypothesis the {Kn} has a <LV," VtCU
There therefore, Z which U V { ' V U ~ ' ' ' '
spheres of the nest. Since the radius of Kn tends to zero as n ~ 00,
z is the limit of the subsequence {Xkn} of {x
n
}. Hence the Cauchy
sequence converges to z, contrary to the hypothesis that {xn}
was a Cauchy sequence which did not converge to a point of M.
The that not complete untenable,
so the condition is
By similar argument, we can prove Cantor's
theorem, that if {Fn} is a sequence of non-empty closed sets in a
complete metric space M such that FI ::::I Fa ::::I Fa ::::I , and if the
diameter of Fn tends to zero as n tends to infinity, the intersection
00
nFn
n=1
is non-empty and consists of a single point.
34. Baire's category theorem
A A of a metric space M,
said of the first if it is
family
52 METRIC SPACE
need to distinguish between them. Thus we may say that it is
possible to embed field Q of rational numbers in the field R
ofroal numbors in suoh a way that Q is dense in R, the embedding
being isometric. We now show that it is possible to perform th(
same sort of completion any metric space.
We start by considering the set of all Cauchy sequences oj
points of a metric space with metric where M not com
plete. On this set, we set up the relation {xn} '"" {Yn} between tW(
Cauchy sequences as meaning that p(x
n
, Yn) tends to zero a ~
n --+ For example, {xn} {xn+p} for every fixed integer p; 01
again, if {xn} converges to x, and if Y
n
= x for all x, {xn} '" {Yn}.
This relatiun is an equivalence relation. The equivalence classes
form a set which we denote by M*. By an appropriate choice of
the defInition of distance, we can make 111* into a metric space,
which turns out to be complete.
The equivalence classes which are the elements of M* are of
two tJ1.Jes. element of first type an equivalence class of
convergent Cauchy sequences; if {xn} is a Cauchy sequence which
converges a x of we say that the equivalence class
which contains {xn} (and, of course, also contains the constant
sequence all of whose members are equal to x) corresponds to the
point x of An element of the second type is an equivalence
class which does not correspond to any element of M; it consists
of equivalent Cauchy sequences which not converge to any
point of M.
It cOllvenient to denote elements of M* by bold letters,
such as a, b or x. If {an} and {b
n
} are Cauchy sequences contained
in the equivalence classes a and b respectively, the triangle
inequality gives
Ip(a
m
, - p{a
fp
b
n
)/
,.;; Ip(am,bm)-p(ambm)1 + Ip(an,bm)-p(an,bn)1
~ p(a
m
, an) + p(b
m
,
But since {an} and {b
n
} are Cauchy sequences, there exists, for
every positive value of a positive integer n1(e) such that
p(a
m
, an) < e, p(b
m
, b
n
) < e whenever m > n ;::: n
1
Hence
COMPLETE METRIC SPACES 53
whenever m nv that b
n
)} a Cauchy sequence of
real numbers which is therefore convergent. Moreover the limit
to whichp(a
n
, tends does not depend the particular Cauchy
sequences chosen from a, b. if two other
Cauchy sequences belonging to a and b respectively,
b;J - p(a
n
,
::::; -p(a
n
, + Ip(a
n
, -p(a
n
, bn)1
an) + b
1
J
o
as n tends to the same limit asp(a
n
, b
n
). This
limit,whichdcpcnds ona b,we byp*(a, b).
We note that, a and correspond to elements and b M,
we can use the representations of a and b by constant sequences
and take = b; it follows that
p*(a, = p(a, b).
This function p*(a,
on Evidently if a
satisfies all the conditions for a metric
b, p*(a, = O. Conversely, if
p*(a, b) = 0,
limp(a
n
, b
n
) =
for any Cauchy sequences {an} and {b
n
} which belong to a and b
respectively; hence "" {b
n
} and so b.
p*(a, b) = p*(b, a)
since p(a
n
, b
n
) = p(b
n
an). Lastly if {en} is a Cauchy sequence
belonging to elemcnt c of
p(b
n
, en) p(a
n
, b
n
) + p(a
n
, en)
and p*(b, p*(a, b)+p*(a, c)
the triangle inequality of M*.
Let us denote by A * the subset of elements of M* which cor-
respond to of 'The restriction p* to is a metric
on which is a subspace of If the points a, b of A * eor-
respond to the points a and b of M, we have proved that
p*(a,b) p(a,b),
so that A * is isometric to M.
54 METRIC SPACES
The subspace A * is dense in M*. To prove this we must show
that every point of M*, which is not a point of A *, is a point of
accum ulation of A * .. Let a be of M* which is point
of A *, let {an} Cauchy sequence of of j}1 belonging
to the equivalence class a. We define a sequence {an} of points
of A *, where the equivalence class a
k
contains the constant
Cauchy sequence whose elements are Then
p*(a, a
k
) limp(an> ak)'
But since {an} is a Cauchy sequence of points of M, there exists,
for every positive of e, integer such that
o :s; p(a
n
, ak) < e,
whenever n > k ~ n
1
; hence
o :s; a/o:) :s; e,
whenever ~ n
1
. 110inta which nota ofA*,
is the limit of a sequence of points {an} of A *, and so is a point of
accumulation of A *.
It remains to show that every Cauchy sequence points of
M* converges to a of M*. prove that every Cauchy
sequence of points of A * converges to a point of M* and complete
the proof by using the property that A * is dense in M* .
. If {an} a Cauchy sequence points and if is the
corresponding of of M, we hlwe
p*(am> an) = p(a
m
, an)
and so is a sequence in M. equivalence class
containing the sequence {an} point a .. M*. Now {b
n
} is a
Cauchy sequence in the equivalence class b,
p*(a, b) = limp(am,b
m
).
If we b to be point A *, which contains con-
stant sequence all of whose members are an' then
p*(a, an) = limp(a
m
, an).
But since {un} is a sequence of of M, there exists,
for every positive value of e, an integer nl(e) such that
o :s; p(a
m
, an) < e,
COMPLETE METRIC SPACES
whenever m > n n
1
It follows that
=so; an) :::;;
55
whenever n n
1
, and hence that the sequence {an} of points of A *
converges to point of M*.
Now let {zn} be a Cauchy sequence points of M*. Then,
for every positive value of e, there exists an integer n2(e) such
that p*(zm, zn) e whenever > n na' Keep e fixed. Since
A * is dense ill M*, we can find, for every integer n, a point an
of A * such that p*(a
n
, zn) < e. It follows that, if m > n n
2
, .
p*(a
m
, =so; zn)+p*(zm> +P*(Zn' an) <
so that {an} is a Cauchy sequence of points of A*. By what we
have proved, {an} converges to a a M*. But
p*(am,z,J ::::; < 2e,
whenever m > n n
z
' Make m -;.. 00; then
p*(a,zn)
whenever n n
2
, and so {zn} converges to a.
36. Some complete metric spaces
The real line with metric x
2
1 is complete. So also is tho
complex plane with metric - z21. }'or if is Cauchy
sequence of complex numbers, for every positive value of
e, there exists an no that IZm - < e, whenever
m > no' we denote the real and imaginary of zn
by xn and Yn, we have
IXm < IYm- < e,
whenever m > n no' Hence {xn} and {yn} are Cauchy sequences
of real numbers, which therefore converge to real numbers x and Y
respectively. lIenee the sequence {zn} converges to X+
The argument used here can be used to prove the following
more general result. If X and Yare complete metric space8 with
metrics PI and the product space X x with metric
p((x
1
, Yl)' (x
2
, Y2 = ,J{(Pl(XV X
2
)2 + P2(Yl' Y2))2}
is complete. ]Tom this follows by induction a Euclidean
space of any finite nurnber of dimensions is complete.
56
denote the
UU.UHJULO such that
1
that
metric on this
a Cauchy
every positive
METRIC SPACES
all sequences = {xn} of COlnplex
is convergent. have seen
it is complete.
on l2, where ..
of 6, there exists an integer no sueh
foHows that,
<
whenever k > l no. But since the complex plane is complete,
x! tends to a limit xn as k -+ 00.
Now, if k > l no and if m is any positive integer, we have
If we make k tend to infinity in this finite sum, we obtain
m
IX
n
6
2
,
n=1
every positive un;v,"""" m. Henee
terms
when l
{f IXnl2t {f Ixn-
x
tl
2
t

{f IXnl2t
is cOIlvergemt.
METRIC SPACES
= (X1,XZ,X
a
, ) is a
Xl) ~ e,
whenever sequence { x ~ therefore NunTP,rm'"
to the point x of 12 and so 12 is complete.
A similar argument, with the same conclusion, can be applied
when the metric is { 00 }l/P
p(x, y) = ~ IX
n
- Ynl
p
,
1
wherep> l.
37. Some complete function spaces
The space consists of all functions
every
that
[a, b] with metric
sup Ix(t)-y(t)l,
te: [a,b]
If {xn(t)} is a Cauchy
there exists a positive mt,egl3r
IXm(t)-xn(t)1 < e,
whenever m > n ~ no, for all tE [a, b]. This is the Cauchy con-
dition for uniform convergence. But a uniformly convergent
sequence of continuous functions tends to a continuous limit.
Therefore the Cauchy sequence {xn(t)} converges to a point of
O[a,b], so that the space is complete.
The subspace, consisting of all polynomials defined on [a, b]
with the complete, since the
formly of polynomials is not nOljOS13arl
a polynomial.
The
p(x,
of all functions x(t) which
points of a closed interval
by taking
Ix(a)-y(a)l+ sup Ix'(t)-y'(t)l.
te:[a,b]
This again is a complete metric space. If {xn(t)} is a Cauchy se-
quence, {xn(t)} converges uniformly to a continuously differen-
tiable functionf(t) and x ~ ( t ) } converges uniformly tof'(t).
58 METRIC SPACES
If, however, we metrize the space of functions continuous on
[a,b] by taking b t
p(x,Y) = {fa Ix(t)-Y(t)12 dt} ,
Then,ifm > n,
space is not complete,
continuous functions
= 0,
= nt, 0:::;; t
= 1, lin:::;; t :::;; 1
{p(x
m
,x
n
)}2 = (m-n)2 t
2
dt+ (1-nt}2dt
f
Ilm fl/n
o 11m
(m-n}2 1
= 3m2n < 3n < e
sequence is thus a Cauchy
Cauchy sequence "011V#"'l'"",,rI
convergence being with l'AQnA{t,
f ~ l I
X
n(t}-x(t}1
2
dt
metric;
This would imply that the integral with any limits between 1
also tends to zero. In particular
But
of n.
which
Again, if a > 0,
:::;; 0, and so this unp!n'",
continuous function
when t:::;; 0.
COMPLETE METRIC SPACES 59
n-+ 00. If choose n > Ifa,
{Il- -+0
as n -+ 00. As the integral is independent of n, it vanishes; and
since x(t) is continuous, x(t) = 1 for t ~ a.
But a can near to zero we please. we have a
continuous which when t ~ which is
to unity t > O. Therefore the Cauchy sequence does
not converge to a point of the space, and so the space is not com-
plete.
The completion of this metric space is the space L2 of functions
whose squares integrable ill sense of
EXERCISES
1. The subspace A of a metric space M is complete. Prove that A is
closed. F is a closed subset of a complete metric space M. Prove that the
subspace F is complete.
M consists of all ordered pairs
numbers with
y) = max
Prove that M is complete.
,x") of real
3. A space M consists of all sequences x = {X,.} of real numbers, where
only a finite number of members of each sequence is nonzero. Prove that
p(x,y) =
metric on
Show that the sequence {x .. } where
is a Cauchy sequence. Deduce that the space is not complete.
A space M consists of all bounded sequences x = {x,,} of real numbers.
Prove that
p(x,y) =
is a metric on M and that M is complete.
5. A space M consists of all convergent sequences x = {x,,} of real
numbers. Prove that
metric on
60 METRIC SPACES
6. A space M consists of all sequences x = {x,,} of real numbers. Prove
that
00 1 Ix -y I
p(x,y) =:E- " "
1 n!l + Ix,,-y,,1
is a metric on M and that M is complete.
7. {X,,} is a sequence of complete metric spaces, the metric of X" being
p". Their Cartesian product is the set of all sequences x = (x,,} where
X"E:X ... Prove that
p(x, y) = 1 p,,(x .. , y,,)
1 n! l+p,,(x .. ,y .. )
is a metric on the Cartesian product, and that the Cartesian product space
is complete.
S; The extended complex plane a has metric
, 2Iz-z'l
p(z,z) = .J(1 + IzI2).J(1 + Iz'12)'
2
p(z,oo) = .J(1 + IzI2)'
Prove that a is complete.
9. M is the set of functions x(t) continuous on every finite interval.
Show that
where
~ 1 p,,(x,y)
p(x,y) = 1 n! l+p,,(x,y)'
P .. (x,y) = sup Ix(t) -y(t)1
t(-n.n]
is a metric on M and that M is complete.
10. M is a complete bounded metric space. :F is the metric space
defined in Ex. 19 of Chapter 3 whose members are the closed nonempty
00 00
subsets of M. Prove that:F is complete by considering n U F
r
, where
n-l r=n
{F .. } is a Cauchy sequence in :F.
11. M is the set of all analytic functions of the complex variable z,
regular on the unit disc Izl < 1 such that
sup i If(reOi)lldO < 00.
O";r<1 -"
Prove that p(f,g) = sup { ~ i u(reOi)_g(reOi)lldO}i
O";r<1 21T -"
is a metric on M.
If Izl < r < 1, prove that
If(z) I r-"Izl { 2 ~ i" If(reOI)12dOt-
Hence show that the space M with metric p is complete.
COMPLETE METRIC SPACES 61
12. The metric space Ml is the set El with metric PI; the space M!
is the sot E. metric The metric spaee Ml X is the El x E'j,
with motric
Prove that, in order that sequence {X"} = x MI
be convergent, is necessary and sufficient that both limits
lim = Xl> lim x: = x.
n-+CXl
should exist, then
n-+CXl
(ii) Prove that, in order that a sequence {X"} be a Cauchy sequence
in M., it necessary and sufficient {x:} Cauchy
sequences.
(iii) Prove that Ml x Ma is complete if and only if Ml and M2 are
complete.
CHAPTER 5
CONNECTED SETS
38. Separated
Two sets A and B in a metric space M are said to be separated
if neither has a in common with closure of the other,
that A n B ,A n B . If A B are separated, they
are disjoint since .A nBs; .A n B = 0; but two disjoint sets are
not necessarily separated.
In order that the two sets A and B may be separated, it is 8ujficient,
but necessary, d(A, O.
For if d(A,B) ... k > 0, the equation
k = inf{p(a,b);aeA,bEB}
implies that p(a, k for pair points a, belonging
A and B respectively. If a is a point of A, it is not an adherent
point of B, since the sphere N(a; ik) contains no point of B.
Therefore A n 11 0 ; similarly An B
The condition not necessary. Two can zero dis-
tance apart, and yet be separated. For example, on the real line,
the sets x < 0 and x> 0 are separated but at zero distance; on
the hand, sets x x ~ 0 disjoint not sepa-
rated,
39. Properties of separated sets
If A, Bl where and B separated, then Al
Bl are separated.
SinceB
1
s; B, Bl S; B. Hence Al n Bl S; A
1
n B S; An B = 0;
similarly Al n Bl = 0. Hence Al and BI are separated.
Two closed sets separated and they are disjoint.
This follows at once from the fact that, if FI and 1 ~ are closed
sets, F,. n P
2
= PI n F2 = Fl n F
2
, since PI = F
I
, P
2
= F
2

Two open sets are separated if and only if they are disjoint.
Any two separated sets are disjoint. have to that two
[62 ]
CONNECTED SETS
O
2
are separated. If they
n O
2
, 0
1
n O
2
is not
there is a point a whieh
open, there is a sphere
in 0
1
, But since a is an adherent point of O
2
, N(a; r) contains a
point of O
2
, Hence 01n O
2
is not empty, which is impossible since
0
1
and O
2
are disjoint.
IftMopen set 0 is the union of two separfJ,ted sets A and B, then
A and B are open.
We may suppose A and B are not empty. For if A = {(1, then
B = 0, and A and B are both open.
no
point
larly
If the
A andB are
of A. Then there exists
of O. Since a is not an
N(a; rl)' where r
1
r,
consists entirely of points
interior point of A; hence
'union of two separated sets
Since F = A v B is closed, we have
AvB =AvB = AvB.
Hence
A = An (AvB) = An (AvB) = (AnA) v (AnB) = Av {(1 = A
so that A is closed. Similarly B is closed.
it is
special
C,D
connected sets
space M is said to be UitSIOO1t'fU;;Cbeu
non-empty separated sets,
'two'; for if A = Bv
cach pair is separated,
L1 = Bv (CvD),
where Band C v D are separated. A subset of M which is not
disconnected is said to be connected. '"
A metric space M is connected if and only if the only non-empty
subset of M which is both open and closed is M itself.
64 METltIC SPACES
If M is disconnected, it is the union of two separated sets
Band O. Since M closed, and are closed. Since and are
the complements of 0 and B, B and 0 are also open. Hence a
disconnected metric space is the union of two non-empty
disjoint sets which are both open closed.
If M is the only non -empty subset of M which is both open and
closed, M is not the union of two non-empty disjoint sets which
are both and closed, so M not disconnected. Hence 211
is connected.
If M is connected, and if A is a non-empty subset which is both
open and closed, A' both closed open. rl'hus the union
of two disjoint sets A and A' which are both open and closed.
Therefore A and A' are separated, which is impossible if A and A I
are both non-empty. But by definition, non-empty. Hence
A' = 0 and A = M. The only non-empty subset of M which is
both open and closed is M.
subset of the metric space M connected if only the
only non-empty subset of A which is both open and closed with
to the subspace A is itself.
the A discomlected, = B v where
BnO=13nO= 0.
Hence An 13 = (Bv 0) n 13 = (B n B) u (0 n B) = B, similarly
A n {) = O. The set B. being the intersection of A and the closed
set is closed with respect the 8n bspace and its comple-
ment with respect to A, is open with respect to A. Similarly 0 is
closed and B is open with respect to A. Thus a disconnected sub-
set of M the union of two disjoint subsets B and which are
both open and closed with respect to the subspace A.
The result now follows from the preceding result by regarding
A a space in its mvn right.
A subset A of the metric space M is disconnected if and only if
A ~ 0
1
U O
2
, where 0
1
and Oa are disjoint open sets such that
A and n02 not
This is merely another way of expressing the previous result.
There is a corresponding result with 0
1
and O
2
replaced by dis-
closed sets and RI. such that A. n and.A Fa are not
empty.
CONNEC'XED SE'XS 65
,Properties connected
The closure of a connected set is connected,
Let A be a connected subset of a metric space M, If A is
disconnected, FI v wherc PI and F2 arc disjoint non-
empty closed sets, But
= A n = (A PI) v n F
2
)
expresses A as the union of two disjoint non-empty sets closed
with to connected subspace which impossible,
Hence connected,
If A is a connected subset of a metric space M, and if B is a subset
that s;; B A, then B is connected,
Suppose that B is disconnected, Then there exist disjoint
closed sets FI and F2 such that B s;; FI ~ , where B n FI and
nF
2
are not B A and nF
2
not
empty, But since A s;; B, A s:: lJ s;; FI ] ~ = j ~ V F
2
, 'his is
impossible since A, being the closure of a connected set, is
yet FI An are empty Hence B is
connected,
If two connected sets are not separated, their union is connected,
Let A and B two connected sets which are separated,
If A v B is not connected, A vB = v D where and Dare
non-empty separated sets, Then
A = A n (A v B) = An (0 v D) = (A n 0) v (A n D),
which expresses A as the union two separated sets, whieh is
impossible since is connected, wlless of A and n D
is empty, Suppose that AnD is empty, so that A s;; 0, Similarly
B s; or B s;; D. If B ; A vB s; 0, hence D is empty;
contrary to hypothesis, 'I'herefore B s;; D, But this gives
AnBS;;OnD= 0, AnBs;;CnD= 0,
which is impossible since A and B are not separated; hence A v B
is connected,
If two connected sets have non-empty intersection, their union is
connected,
5
eMS
METRIC SPACES
A and B be two connected sets, whose intersection A " B
is empty. Since A" B := A" 13 and A s; A the
result is an immediate consequence of the one we have just
proved. We here different proof of the weaker result which
is capable extensioll to any family connected sets.
Let 0 = Au B, and let D be any non-empty subset of 0 which
is both open and with respect the subspace O. must
intel'sect A or B or both; suppose it intersects A, so that D" A
is not empty. Then D" A is a subset of the subspace A which
is both open and with respect the subspace Since
A is connected, D" A = A, and hence .A s; D.
Now A" B is not empty, and therefore D" B is not empty.
By the argument, D. Therefore = A s; D.
But D was a subset of 0 by definition; hence D = O.
We have thus proved that the only non-empty subset of G
which is both open closed with to subspaee 0 is
o itself, and this implies that C = A u B is connected.
The generalization is:
{A",: a is a farnily connected sets whose intersection is
not empty,
is connected,
Denote the union of the family of sets by G, and let D be any
non-empty subset of 0 which is both open and closed with respect
to subspace O. proof then follows preeisely just
given.
If every pair of points in a subset A of a metric space M lies in a
connected subset of A, then A connected.
Suppose A is disconnected. Then there exist disjoint open
0
1
and O
2
of M such that A s; 0
1
U O
2
and A " 0
1
, A " O
2
are not
Let be point A " b any of" O
2
,
By hypothesis, there is a connected subset B of A which contains
both a and b. But B s; 0
1
U O
2
, and B n 0
1
and B" O
2
are not
since they contain the points a b respectively. Tlus is
impossible since B is connected. Hence A is not disconnected.
If X is a connected space with metric Pv Y a connected space with
metrie P2 the Oartesian product X x with metric .J(Pi + is
connected.
CONNECTED SETS 67
The proof of this depends on the fact that a continuous map-
ping a connected set into connected The result also
true metric x Y is or Max (Pl,P2)' Sec
42. Components
Let and b be points subset A metric M.
We "''"rite a,.. b ifthere is a connected subset of A which contains
both a and b. This relation connecting a pair of points of A is an
equivalence relation. It is reflexive, since the set consisting of a
single a is connected. It evidently symmetric. also
transitive; for if a b, there is a connected subset Al of A
which contains a and b, and if b ,.. c, there is a connected subset
A2 which contains band c. But the set Al v A2 is connected
(since n A2 is empty) contains both a and hence
a,.. c.
The family of equivalence classes defined by this relation is
called a partition of A into components. Each component is a con-
nected and every point of A member exactly com-
ponent. The component containing a is the union of all connected
subsets of A which contain a, and so is the largest connected set
containing a. If A is a connected set, it has only one component,
A itself,
Since the metric space M is a subset of we can M
itself into components. The components of a metric space are closed
sets.
Let a component of M, 0 is connected, so is 0.
Let a point b a point 0. Then connected set 0
contains both a and b. But by definition 0 is the greatest con-
nected set containing a, and so b belongs to O. Hence 0 ; 0;
thus closed.
43. Totally disconnected sets
In Euclidean plane, 'sphere' a connected set.
One naturally asks whether true every metric space.
The answer is that it is not. A simple example is provided by the
set Q of rational numbers with p(x,y) = Ix-yl. The sphere
N(l; with centre and radius in Q is set of an rational
68 METRIC SPACES
numbers such that x < 2. can this into two
separated sets in many ways. For instance, we can take the sub-
sets for which 0 < X
2
< 2 and 2 < X
2
< 4. The sphere N(l; 1)
is thus connected.
If every connected subset of A in metric spnee reduces
to a single point, that is, if no two points of A lie in a connected
subset of A, we say that A is totally disconnected. The set of
rational numbers totally disconnected; also is set of
irrational numbers. Another is by any
set and constructing a metric space with the discrete metric for
which p(x,x) = 0 and p(x, y) = 1 when x =1= y.
44. Intervals on the real line
The real line is the set of real numbers with metric Ix-yl. If
a and b are two finite real numbers, the set of real numbers x
such that :;:; x :;:; a bounded closed denoted [a, b];
it is called a closed interval. The clo"ed interval [a, a] consi"ts of a
siIigle point. If we omit the end points, we get a < x < b, which is
a bounded open set, denoted by (a, b); it is called a bounded open
interval. There are other bounded intervals which neither
open nor closed; are a < x :;:; b, denoted by (a, b], and
a:;:; x < b denoted by [a, b).
Since there are no points + 00 or - 00 on the real line (though
there are the extended real the > a, denoted by
(a, + (0) .1Il unbounded open interval of real line ; similarly
the set x < a, denoted by ( - 00, a) is an unbounded open interval.
But the unbounded set x;;::: a, denoted by [a, +(0) is closed
since its complement 00, a) open. And similarly for 00, ] .
45. The connected subsets of the real line
The real line is connected.
If the line R disconnected, it is the union A of two
non-empty disjoint closed sets. Let a
l
belong to A, b
I
to B.
If ex is the midpoint of (a
v
bI)' ex, being a point of R, belongs to A
or to B. the former case, take b) and it (a
2
, ; in the
latter, (aI' ex) call it b
2
). the process in-
CONNECTED SETS
definitely, We get sequence of intervals {(an' with
properties:
(i) {an} is a bounded increasing sequence of real numbers,
(Ii) {b
n
} is a bounded decreasing sequence of real numbers,
(iii) b
n
- an = (b
1
- a
1
)/2n-l,
(iv) for each integer n, A, b
n
69
It follows that {an} and {b
n
} converge to the same limit c. And
since A and B are closed, c belongs to both A and B. This is
since and B disjoint. Hence R is connected,
The non-empty connected s'ubsets of the real line are intervals.
An interval of the form [a, a] is evidently connected. The
argument used for the realIine can applied to an interval
any of the types of 44, hut the interval must be regarded as a
subspace in the proof: closure is relative to the subspace, not
R.
Conversely, let A be a connected subset of R. Let b = supA,
a = infA, where b = +00 when A is unbounded above, a=-oo
when A unbounded below. Suppose that a point 6, where
a < 6 < b, did not belong to A. Then A; 0
1
u O
2
, where 0
1
and
O
2
are the disjoint open sets {xER:x < S} and {xER:x > S};
moreover n 0
1
and A n are empty, But implies
that the subset A of R is disconnected, contrary to hypothesis.
Hence, if A is connected, every point x such that
infA < x < sup A
belongs to Thus A is an intervaL The particular typo of inter
depends on whether either of or sup A belong to A,
The structure of open and closed sets on the real line
Since every point of an open set on the real line is an interior
point of tho set, overy open set is the union of a family of open
intervals (' spheres which may overlap. A more precise result
that
Every set on the real line is the union of a finite countable
of disjoint open intervals.
Every set can be partitioned in a unique way into separated
pOluerUiS. The containing particular point
70 METRIC SPACES
the ,n">!I.1".,""'" connected set containing
the real line
the intervals.
open set on the real component
intervals since they are they may
abut. If two component intervals abut at a point a, a cannot
belong to either set, since otherwise the two component intervais
would not be separated. Again, if a component interval with a
frontier point b has no component interval abutting on it at b,
b cannot belong to the component, since b, being a point of 0,
must be an interior point of O. Thus the components of 0 are
separated open intervals of one of the forms (a,b), (-oo,b),
(a, epen intervals are if only if they
every
able,
open set is the union of disjoint
complete the proof) show that
disjoint open intervals or count-
Let first the case when bounded;
suppose in the open interval this interval
is of unit length, for every positive integer n, the number of
components oflength ~ Ijn cannot exceed n. It follows that the
number of components oflength ~ lin and less than Ij(n-l) is
finite. The components of the open set 0 can now be arranged
as a sequence in order of their length-groups, taking first those
of lengths ~ t ordered from left to right, then those that are of
lengths < t and ~ 1 again ordered from left to right, and so on.
There are only a finite number to arrange at each stage. There
number of But if 0
number of components,
sequence means that
they can be
countable
the general case to case when 0
the transformation + t. This is
a correspondence po:iri.ts of the
infinite interval - 00 < x < 00 and the points of the open inter-
val (0,1); and the open intervals of the real line correspond to
open intervals of (0,1).
Every closed set on the real line is the complement oj a finite or
countable Jamily oj disjoint open intervals.
CONNECTED SETS 71
Since closed set is the complement of open set, the result,
is evident. Isolated points in a closed set are points where two of
the disjoint open intervals abut.
EXERCISES
1. If the sets A and B are separated, and if the sets B and 0 are sepa-
rated, prove that B and A v 0 are separated.
If A and B closed sets, prove A n and n A' sepa
rated.
A is oonnected subset of a oormected metric space The comple.
mont of A is the muon of two separated sots B and O. Prove that A vB
and A v a are connected. Show also that, if A is closed, so also are A v B
and A va.
4. A and B are two closed sets in a metric space M. Show that, if
A v B and A n B are connected, then A is connected.
A and B are two sets in a metric space lit and neither A B nor
A' n B is empty. Prove that, if B is connected, (Fr A) n B is not empty.
Deduce that, if M conneoted, every subset of 11, other than 1.'1;1 and 5,
has least one frontier point.
6. Prove that a subset A of a metric space M is connected if and only if
there exists no pair of closed sets and sucbbhat
A s;F1vF
a
, FlnFanA = 121, FlnA =1= 121, FanA =1= 121.
C[a,
metric
is the space all fWlCtions continuous 011
p(x,y) = sup /x(t)-y(t)/.
tli, b)
Prove that the space is connec-ted.
~ t with
8. If {A,,: a, e:S} is a family of connected sets and if one set of the family
intersects all t,he others, prove that
is eormectod.
72
CHAPTER 6
CO]!'Il} ACTNESS
47. Uniform COllltiltlut the Heine-Borel Theorem
Let !(x) the real variable x COIltirmo
each point
value of e , of [a, b], it is possible
open interval N(g; 8) such that 1!(x)-!(g)1 < e whenever the
point x of [a, b] lies in N(g; 8); indeed for each S there are an
infinite number of such open intervals since, if N(s; 8
1
) is one
such interval, so also is N(s; 8) for every 8 < 8
1
The infinite
family {N(g; 8): SE [a, b]} of all these open intervals correspond-
ing to all the points g of [a, b] is called an infinite open covering
of [a, b]; every belongs to at least one open
of the family.
The Heine-Borel asserts that, from
open covering select a finite number
vals of the covers [a, b]. Every
belongs to open intervals of this
covering. ]'rom this follows the uniform continuity property,
that, for every positive value of e, there exists a positive number
depending on e, such that 1!(x
1
) - !(x
z
) I < e whenever the
distance between the points Xl and X
z
is less than We return
to this in a more general context later.
Although the Heine-Borel Theorem arose in connexion with
uniform continuity, it haanothing whatever to do with continu-
ous functions, that, from any infinite
open intervals which covers a bounded
interval of possible to choose a
intervals, which also covers
bounded
Another bounded closed
the real line, closely related to the Heine-Borel
is the Bolzano-Weierstrass Theorem, which asserts that any
infinite set of points contained in a closed interval [a, b] .
[72 ]
74 METRIC SPACES
choose from this family finite open of JJI consisting of
F' and {na: ae8
1
} where 8
1
is a finite subset of the index set 8.
The cOITesponding family {Oa: aeB
1
} is a .finite family sets
which are open with respect to F and cover F. Hence the sub-
space F has the Heine-Borel property.
49. Sequentially compact metric spaces
A metric space is said to be sequentially compact if every se-
qnence of points of the space contains convergent subsequence.
A subset A of a metric space is said to be sequentially compact
if the subspace A is sequentially compact: we need only con-
sider spaces.
A metric space is sequentially compact if and only if every infinite
sub8et has point of accumulation.
Let A be any infinite subset of a sequentially compact metric
space M. Let {arl} be any sequence of distinct points of A. Since
M is sequentially compact, this sequence contains a convergent
subsequence {ak
n
} whose limit is a, say. Then a a point of
accumulation of .A.
Conversely, let M be a metric space with the property that
every infinite subset a point of accumulation. Let {an} be
any sequence of points of M. If any point occurs infinitely often
in the sequence, the sequence contains a constant subsequence
which is convergent. If not, we may assume all the members of
{an} are distinct. The A of points which are the values of an
(n = 1,2,3, ... ) is an infinite set, and so has at least one point of
accumulation. If a is a point of accumulation of A, we can choose
a sequence of points of.A which converges to a, and this sequence
is a subsequence of {an}. Hence the result that a metric space is
sequentially compact and if every infinite subset has a
point of accumulation-the Bolzano-Weierstrass property.
compact metric space is compact.
We have to show that every infinite subset of a compact metric
space has a point of accumulation. If A is an infinite subset of a
compact metric space M and if A. has no point of accumulation,
each point of M is the centre of a sphere which contains no point
of save possibly the ccntre the sphere. These spheres form
COMPACTNESS 75
an infinite open covering Since M is compa,ct, we
from this family of spheres, a finite number which also cover M.
Since each sphere contains at most one point of A, A is a finite
set, contrary to hypothesis. Hence every infinite subset of M
has a point of accumulation.
A compact sub8et metric space i8 bounded closed.
Let A be a subset of a metric space M, such that the subspace
is compact. Since there nothing to prove if .if is a finite
we assume that A is an infinite set, and, by the preceding result,
we may suppose it to be sequentially compact.
If .A is unbounded, we can find pairs of points of .A at arbi-
trarily large distances apart. This enables us to construct a
sequence of points A which possesses no convergent subse-
quence. Starting with any point a
1
of A, we choose a point a
2
such that p(a
1
, a
2
) > ; we then choose a point such that
p(al> as) > 1+p(al>a
2
),
and so on. For every value of n, an is chosen so that
an) 1 + p(av an-I)'
and this implies that if m > n,
p(a
l
, am) > 1 + p(a
l
, an).
Hence
From this it follows that {a,J eontains no convergent subsequenee,
which is impossible. Hence A is bounded.
If a is any point of A there exists a sequence of points {an} of A
which converges to and every subsequence {an} converges
to a. But since A is a compact subspace, {an} has a subsequence
which converges to point A. Hence a belongs A, and there-
fore A is closed.
50. Totally bounded spaces
A metric space is said to totally bounded if, every
positive value of e, M can be covered by a finite number of
spheres of radius e. A totally bounded space is evidently
76 METRIC SPACES
A sequentially compact metric space is totally bounded and com-
plete.
Let a
l
be any point of a sequentially compact metric space M.
Fix e if possible a point not less
than 6 choose a point as at
so on. For every value
an from is not less than e. We
must that, if M is bounded
enough, terminates at the first stC:l}--tJler
point at a distance not less than 6 from a
l

If the process did not terminate, we should have constructed
a sequence {an} with the property that no two points of the se-
quence are at a distance less than 6. But since M is sequentially
compact, this sequence would contain a convergent subsequence,
which is impossible.
For every positive value of e, the process gives a finite set
of points such that the where
n = 1, M, which is therefore bounded.
Such points is called an sequentially
possesses an e-net value
sequentially compact is com-
plete, we have to show that every Cauchy sequence converges to
a point of M.
Let {an} be a Cauchy sequence, so that, for every positive value
of 6, there exists an integer k such that p(a
m
an) < e, whenever
m > n ~ k. This Cauchy sequence contains a subsequence {akJ
which converges to a point a of M, that is
Since
have
whenever
lim p(akn,a) = o.
1r+OO
llR;n:,,,.,,1111!!. sequence of integers,
p(a
n
, ak
m
) + p(a
km
,
Make m ~ 00. Then
o p(an,a) ~ e,
We now
whenever n ~ k. Therefore the Cauchy sequence {an} converges
toa.
78 METRIC SPACES
between the two kinds of compactness when we are dealing with
metric spaces. In the proof we need the following lemma.
IJ the Jamily oj open sets {O a: ex: e S} i8 an infinite open covering
oj the compact metric 8pace a p08itive
number every 8phere oj radiu,8 in at least
one of
',",,,"ULUI.., false. Then, no small e is,
of radius e contained of the sets Oa.
every positive integer a sphere
N(a
n
; lin) which is not contained in any of the sets Oa. But
since M is sequentially compact, the sequence {an} of centres
of this sequence of spheres contains a subsequence {a
kn
} which
converges to a point a of M.
Since {O a: ex: e S} covers M, this point a lies in at least one of the
open sets of the family; suppose it lies in 0
0
The point a is an
interior point of 0
0
, and so there exists a integer no such
that 0
0
The sphere contains all but
UU.U.!I,}"'.I of members of the particular
anI of the subsequence,
Hence
s;; N(a
nl
; llno) S;;
> no, con-
which contradicts the definition of the sequence {an}. Hen(:e
the lemma is not false.
A 8equentially compact metric space is compact.
Let {Oa: ex:eS} be any infinite open covering of a sequentially
compact metric space M. If e is the positive number whose exis-
tence was proved in the lemma, every sphere of radius e is con-
tained of the open sets sequen-
tially totally bounded, so e, we can
find an ... , ak, with the nl'rlnA,'t;v
k
M s;; UN(a
r
;
r=l
By the lemma, each sphere N(a
r
; e) is contained in an open set
o a,. of the family. Therefore
k
M s;; U Oa,.,
r=l
COMPACTNESS
SO that M has a finite open covering chosen from the infinite
open covering {O/X:aeS}. Hence M is compact.
To sum up, we defined a compact metric space as one which has
Heine-Borel property. We now proved that a metric
space is compact if and only one of the following proper-
(i) it possesses the Bolzano-Weierstrass property;
(ii) it is sequentially compact;
(iii) it is complete and totally bounded.
Compactness and the intersection property
A quite different formulation of compactness depends on the
finite intersection property. A family {F/X: aeS} of closed sets,
where the index set S is infinite, is said to have the finite inter-
section property if, for every finite subset So of S,
is not empty .
.A metric 8pace M i8 compact iJ and only iJ, Jor every infinite
Jamily {F .. : a e S} oj cl08ed 8et8 with the finite inter8ection property,
not empty.
Let M be and let e S} be any of closed
sets with the finite intersection property. If n F/X is empty,
/X(OS

/X (0,";
Hence covering Since M
compact, there exists a finite So of S such
U
''(OS.
Taking complements,
which is since the of closed has the finite
intersection property. Hence n is not empty.
/X (OS
80 METRIC SPACES
compact. Then an infinite
S} which contains covering.
subsetSoofS, U O",isapropersubsetofM.Hence
"'S.
n ~ is not empty; the family { O ~ : a E is a family of closed
"'S,
sets with the finite intersection property. But
and so
u 0", = M,
"'s
n ~ = 0.
"'s
Therefore if M is not compact, it is not true that every infinite
family with the finite interseetion has a
non-empty Hence the result.
53. The
It is
product of two ",""m""".
result that a rectangle
~ x ~ b, a ~ Y
spaces
in the Euclidean plane has the Bolzano-Weierstrass property and
so is compact. This is a particular case of a more general result
concerning compact metric spaces.
If X i8 a compact metric 8pace with metric PI and if Y i8 a compact
metric 8pace with metric P2' the Oartesian product space with metric
"j (pi + ~ ) i8 compact.
We have to show that, if {xn} is any sequence of points in X,
{Yn} any points in Y, the sequence of points
in X x convergent subsequence.
'dVLUjJ<."V, {xn} contains a which
of X. Since Y is emnp.ami
which converges to
h"""nnonno {XIJ of {xn}
value of 6, there n' such
that PI(X, x",) < 6/"j2 when n> n', and an integer nil such that
P2(Y'Yln) < 6/";2 when n> n". Hence if no = max (n',n"). and
COMPACTNESS 83
whenever It
1
-t
2
1 < 8, for every point x of M. We prove M com-
pact by showing that every sequence {xn} of points of M contains
a convergent subsequence which converges to a point of M.
We recall that ordinary uniform con-
vergence.
With any fixed
M> b-a. Divide
positive integer k so that
by points
< tk = b.
Then if t and t' are any two points of a closed interval [ti-v tiJ, we
have
for every member of the sequence.
Let Ti be the mid-point of [t
i
-
1
, til. Then corresponding to each
member Xn of the sequence, we have a point xn in the k-dimen-
sional Euclidean space Rk with co-ordinates
Since IXn(t)1
quence and so
In n. We call T.nC'T'OT(U'O
,Xn(Tk))
{xn} is a bounded se-
subsequence {x",}, where
nr.c"h"TA integer no so that
whenever m > n no, and for i = 1, 2, 3, ... , k.
Let t be any point of [a, b]. Suppose that it lies in [ti-vti]
so that
Ixn(t)-Xn(Tj)1 < e
for every value of n. Therefore, if m > n no
Hence {xl,,(t)}
to a point x(t)
-x/n(Tj)1
+ IX/n(Ti) - x",(t)\
which converges uniformly
closed, x(t) belongs to M.
6-2
84 METRIC SPACES
EXERCISES
1. Give an example of a separable metric space which is not compact.
2. Give an example of a complete metric space which is not compact.
3. Prove that the extended complex plane with metric
21z1-z21
:J{i + IZlI2) .j(i + IZ212)
is compact.
4. A and B are two compact subsets of a metrIc spaco M. Prove that
A v B and A " B are compact.
5. A and B are non-empty sets in a metric space and B is compact.
Prove that d(A, B) = 0 if and only if A" B is not empty.
6. A is a compact set in a metric space M and b is a point of A'. Prove
t,hat there a point a of A such that b) d(A, b).
7. A is a compact set of diameter d(A). Prove that there exists a pair of
x and y of A such that p(x,y) = d(A),
8. A and B are disjoint compact sets in a metric space M. Prove that
d(A, > Show also thatChere exist disjoint open sets 0
1
and such
thatA c 01,B cO
2
,
9. Prove that metric space is eompact if and only it the
following property: from any infinite family of closed sets whose inter-
section is empty, is to ehoose finite family whose intersection
is empty.
10, {FCI:CtS} an infinite family closed sets with the finite inter-
section property, and one set of the family is compact. Prove that n Fa
not empty. (H_B
11. Prove that, from any infinite open covering of a separable metric
space one can extract countable open covering.
12, Prove that a separable metric space is compact if, from every
countable open covering, one can extract a finite open covering.
CHAPTER 7
FUNCTIONS AND MAPPINGS
56. Functions defined on an abstract set
This chapter is concerned the extension of idea of
continuity to a function defined on an abstract set and taking
values in an abstract set, possibly the same set. We recall the
definition a function or Chapter
Let and E2 be sets. mapping of EI
into E
2
, we mean a relation which associates with each element
a of EI a single well-defined element of E
2
, denoted by f(a) and
called image of under mapping The f of
EI into is denoted by f: El E
2
The EI is called the
domain of the mapping; the set {f(a): aeE
l
} of all points of E2
which are images of points of El is called the range of f. The
range be a subset E
2
; it even single
point, which case mapping called constant mapping.
A function is, by definition, one-valued.
Another way of picturing a mapping is a generalization of the
graph elementary calculus. Consider Cartesian product
El x E2 which is the of all ordered y) x eEl'
yeE
2
A mapping or functionf:E
l
-+ E2 is a particular sort of
subset of El x E
2
; a subset with the property that, amongst the
ordered (x, y) which form function, each of El
occurs once and once only. subset f of El x E2 is a
function with domain El and range in E2 then has the following
properties:
(a) eachxeEltherecorrespondsay such that y)ef:
(b) If (Xl,Yl) and (X2'Y2) are in/and Yl '* Y2' then Xl '* x2
In a mappingf of El into E
2
, the range may be the whole of E2
or may be a proper subset of E
2
If the range is the whole of E
2
,
the is said be a or mapping onto
E
2
Every 'onto' mapping is 'into' but not all 'into' mappings
are 'onto'.
IfAisa subset of E
1
, the set {f(x):xeA}isdenoted by f(A) and
86 METRIC SPACES
called the rna ppingf. If B subset of
set {x e points whose images
Bj it is called inverse of B and is denoted by f-I(B).
Ifno point of EI has an image in B,j-l(B) is empty. In particular,
if B consists of a single point, its inverse image may be empty,
may consist of one point or may consist of more than one point.
'I'his implies is, in not a or function,
since a mapping function onc-valued.
If the image of each point of E2 empty
consists of a single point of E
I
, the mapping f:E
I
- E2 is said to
be an injection or a one-to-one mapping. Iff:E
I
_ E2 is an injec-
tion,f(x
l
) = f(x
2
) implies Xl = x
2
;andx
I
=1= x
2
impliesf(x
l
) =l=f(x
2
)
an 1-1 is a off(E
I
)
A mapping -+ E2 whieh both an and a sur-
jection is called a bijection; every point x e EI has a unique image
in E
2
, every point yeE
2
has a unique inverse image in E
l
.
An important instance of a bijection is the schlicht function of
complex variable theory.
Letfbe a of El g a mapping E2 into
mapping into Esis by writing g(y),where
y = f(x) for every point x of E
I
, or h(x) = g(f(x. This mapping is
called the compo8ition or product of the mappings f and g, and
is denoted by go f.
Properties mappings
In this we list here without proofs properties of
mapping f:E
l
- E
2
. A, All A2 are subsets of E
I
, and CE1A
denotes the complement orA with respect to E
I
. B, Bv B2 are
subsets of E
2
, and C E
z
B denotes the complement of B with respect
E2
(i) f(Al) ;
(ii) = f(Al)
(iii) f(AI n A
2
) ; f(AI) nf(A
2
}.
(iv) f-l(B
I
vB
2
) =f-l(B
l
) vf-I(B
2
).
(v) f-l(B
l
n B
2
) = f-l(B
l
} nf-I(B
2
).
(vi) =
(vii) If then
(viii) If ; B.
e
'.
l!'UNCTIONS AND
more precise results
f(A) holds for
E A if and only iff is
87
special types
(x) f(AI n A
2
) = f(AI) nf(A2) holds for all pairs of subsets
AI' A2 of E if and only iff is an injection.
(xi) If B = f(A). then f-I(B) = A holds for all subsets A
of E if and only iff is an injection.
(xii) If A = f-I(B), then f(A) = B holds for all subsets B
of E2 if and only iff is a surjection.
58, I"I".u,,,,,<> and sequence spaces
interest in analysis
or both are sequence
every point is a
if the numbers
a real space.
in which the
sequence space
{xn} of real or
space is called
The general sequence space, denoted by 8, is the space of all
sequences {xn}; it can be metrized by taking
(
X ) - ~ A IXn-Ynl
p ,Y - t 11,1+ IXn-Ynl'
where ~ A n is a convergent series of positive terms. The space
m of bounded sequences; each point x = {X1/,} is a bounded
sequence, and
p(x, Y) = sup
{
00 }l/P
p(x, y) = f IX
n
- Y1/,lp ;
it, too, is complete.
all convergent
null sequences.
lp (p ;a>; 1) of
with metric
If A is an infinite matrix, of which the element in the nth
row and kth column is An,k (n = 1,2,3, ... ;k = 1,2,3, ... ), we
88
can define
by
METHIC SPACES
A (x) of a sequence space
at any rate when the infinite series are meaningful. In particular,
if the matrix A is row-finite, i.e. each row of the matrix contains
only a finite number of non-zero elements, the mapping maps the
general sequence space 8 into itself since no question of conver.-
gence arises. Mappings of this type arise in the theory of summa-
bility of infinite series. For example, if
1
(k ~ n), An,k = 0 (k > n),
the equation
+X
2
+ ... +X
n
);
C into itself. But it may U l l " l n ~ m
not; in this case lim
n-+<Xl
Cesaro limit or (C, 1) limit of the sequence {x
n
}.
Another type of mapping maps a function space into a se-
quence space. The function space C[ -11,11] consists of all functions
x(t) .continuous on the closed interval [ -11,11] of the real line with
metric
p(X,y) = sup{lx(t)-y(t)l: te[ -11,11]}.
Ifwewrite
x(t) dt,
e-fltri1dt (n = 1,2,3, ...
the function onto the sequence
x = {xn:n = 0,1,2,3, ... }
of Fourier coefficients. Thus we have a mapping of C[ -11,11]
into a sequence spage, e.g. into the space Co of null sequences,
since xn tends to zero ali n ~ 00.
FUNCTIONS AND MAPPINGS 89
More generally, if x(t) L2[ -1T, 1T], the space of all functions
whose squares are integrable over [ -1T, 1T] in the Lebesgue sense
with metric
by Parseval's Theorem, the sequence x = :E'ourier
coefficients belongs to l2. In this case we have a mapping of
L2[ - 1T, 1T] into l2. This mapping is a bij ection; for, by the Riesz-
Fischer Theorem, if x = {x
n
} is any point of l2, there exists a
function of L2[ - with Fourier coefficients
{x
n
}, provided regard all which only on
of zero as identicaL lIIoreover this pre-
serves distance; if x(t) and y(t) are mapped onto x and y respec-
tively,
theory falls outside the scope book.
59. Limits and continuity
One way of introducing the ideas of limit and continuity of
functions defined on an abstract space and taking
abstract space introduce of distance.
we restrict attention to spaces and consider the
mapping of one metric space into another metric space.
Let f: X ~ Y be a mapping of a metric space X with metric
PI into a metric space Y with metric P2' If A is a subset of X, x
any point of A, a any point of A, we say that f(x) tends to the
b as x a on A if, every positive of e,
~ ( a ; 8) such that ~ ( a ; 8)
a) implies f(x)N
2
(b; suffixes and N2
indicate that the spheres refer to the metrics PI and Pa respec-
tively. This is the straightforward translation into the new con-
text of a well-known situation in the theory of functions of two
variables when consider the a function as the
(x, y) moves to the origin a curve
the case is the whole metric space ~ l ( , that
J(x) tends to the limit b as x tends to a if, for every positive value
90 METRIC SPACES
of e, there exists a sphere N1(a; 8) such that xeN1(a; 8) (x =1= a)
UllIJU,,,,, that eN
2
(b; In the case mentioned,
reduces the definitien of the limit of a funetion f(x,
of two real variables as (x, y) moves up to the origin in any
manner.
particular, if b = the f: X -'>" said to
continuous
The mapping f: X -+ Y of a metric space X into a metric space
Y is continuous at the point a of X if and only if,for every sequence
of points which converges to corresponding sequence
converges to f(a).
Iffis continuous at a, for every positive value of e there exists
a sphere N1(a; 8) such thatxeNl(a; 8)impliesthatj(x)eN2(f(a); e).
But if {x
n
} is sequence converging a, there an integer
such that no that N
1
(a; 8) hence
f{x
n
) e e). Hence the sequence {f(xnH converges to f(a).
The condition is thus necessary.
To prove the condition sufficient, suppose that it is satisfied
that f continuous at a. there a positive
number e such that, no matter how small 8 may be, xeN
1
(aj 8)
does not imply that f(x) eN
2
(f(a); e). In this case we can find a
sequence {xn} of points of X such that P2(f(x
n
),!(a)) e, with
Pl(X
V
1
Pl(X
2
, a) < Min (!,Pl(X
V
a))
Pl(X
a
, a) < Min (1, Pl(X
2
, a))
Pl(xn ,a) <
sequence converges to a; hence the sequence
converges to This impossible since P2(f(X
n
), j(a))
all n. The assumption that J is not continuous at a is thus false.
Hence the result.
Continuous mappings
A mappingJ:X -+ Y of a metric space X into a metric space
Y is said to be continuous if it is continuous at each point of X.
FUNCTIONS AND MAPPINGS 91
The mapping f: X ~ Y of a metric space X into a metric space
is contimtO'U8 if and if the image every open 0
contained is open.
If f(X) is the range of f, the inverse image of an open set 0
contained in Y is the inverse image of 0 nf(X). But 0 nf(X) is
open with respect to There thus no generality in
assuming f(X) that is, that f is mapping X
onto Y.
Let 0 be a non-empty open set contained in Y, and let A be
inverse image. A empty of Y the
image of X. of A, it unique
b which belongs to O. Since 0 is open, there is a sphere N
2
(bj 6)
contained in O. Because f is continuous at a, there is a sphere
N
1
(aj8) image contained N;(bje) and therefore OJ
hence Nl (a; s;; A. every a of A the centre a
sphere contained in A, and so A is open.
Conversely suppose that the inverse image of every non-
empty open set in Y is open. Let a any point of X, and let
= f(a). every value the inverse image the
open sphere N
2
(b; e) is open and contains a. There is therefore a
sphere Nl(a; 8) contained in the inverse image of N
2
(b; e). Thus
for every positive value of e, we can choose 0 so that the image
Nl(a; 8) contained N
2
(b; e). this is condition that
f is continuous at a. a was any point of the mapping
f: X ~ Y is continuous.
The mapping f: X ~ Y of a metric space X into a metric space
is continWJU8 if and if the image every set
contained is closed.
This follows from the preceding result by taking complements .
. Homeomorphisms and equivalent metrics
If the mapping f: X ~ Y is a bijection, f-
1
: y ~ X is also a
mapping. If, in additionfis continuous on X, andf-l continuous
on Y, the mapping is a homeomorphism. For example, in
complex variable theory, the mapping
z-c
w=-1 -,
-cz
92 METRIC SPACES
where lei < l,ofthecloseddisclzl ~ 1 onto the closed disc Iwl ~ 1
is a homeomorphism.
Let X and Y be two metric spaces consisting of the same under-
lying set E with metrics PI and P2 are said
to be equivalent if of X onto Y, defined
by i(x) = x for aU Since i and its
inverse are continuous, mapped onto open
sets in X; open open sets in Y. In
other words, the same as the open sets
of Y.
The family of open sets of a metric space is called its topology.
If we metrize a set E with equivalent metrics, we get two metric
spaces with the same topology. Properties of a metric space which
depend only on its open sets and can be completely freed from the
idea of distance are called topological properties. The definitions
of closed set, point of accumulation, closure, interior, exterior,
frontier, continuity, eompactness, connectivity
can all be expressed So a property of a
metric space these is a topological
property.
The conditions
for every point a value of 6, there
exist positive numbers 8
1
and 8
2
such that
pz(a,x) < 6 whenever Pl(a,x) < 8
11
Pl(a,x) < 6 whenever P2(a,x) < 8
2

The numbers 8
1
and 8
2
depend, in general, not only on 6 but also
on a. If there exist positive numbers aI' and a
z
, depending only
on 6 and such that all E, the two metrics
are said to be ?J/YJ)Hmrm,/lI ''''''iff,'lm"" "'ii,',
A sufficient, but
be uniformly
numbers ex and fJ
for all pairs of points x and y of E.
If PI is a metric on E, so also is
C;UI"nu""vu for two metrics to
should exist positive
y)
P
Pl.
2 = l+Pl'
FUNCTIONS AND MAPPINGS 93
Pl(X,y) < e that P2(X,y) < e, and P2(X,y) < 6/(1+6)
implies Thus PI P2 are uniformly
lent, but satisfy condition the T ... ,..,,,nm
paragraph. When we are dealing with properties which can be
expressed entirely in terms of open sets, with topological proper-
ties, it does not matter which metrie we use. metric
the advantage of being bounded.
Again, consider two metric spaces Ml and M
2
, consisting of
setsE
1
andE
2
withmetricsPl andp2respectively. We canmetrize
x E2 various If x x
2
), y Y2) are two
points of E
2
, the funetions
p(x, y) = Max{Pl(x1'Yl),P2(X2'Y2)}'
p'(x, Pl(Xl' +P2(X2'Y2)'
p"(x, y) = .j{Pi(Xl' Yl) + ~ ( X 2 ' Y2)}
are all metrics on El x E
2
Since P :s:; p" :s:; P' :s:; 2p, they are uni-
formly equivalent metrics. If we dealing with topological
properties the metrized produet space E 1 it does not
matter which metric we use; the first is usually the easiest to
handle.
The continuous mapping connected set
If f(x) is a real function of the real variable x, continuous on
a bounded closed interval [a,b], it is bounded on [a,b], and it
attains on b] its supremum M, infimum and every value
between 'In and M. Thus Y = f(x) maps the bOWlded closed inter-
val [a, b] onto the bounded closed interval Em, M]. Now a bounded
closed interval is (i) a connected set of the rea] (ii) a compact
set on the line; tre two corresponding generalizations,
one relating to connected sets, the other to compact sets in a
metric space.
If f: X Y i8 a continuous 'fftUm'lJ't'Itu of a space X into
metric Y, tILe irn,age of connected subset is
connected.
Let A be a connected subset of X, and consider the continuous
mapping ~ f(A) subspace A onto subspace
Y. Let any non-empty subset off(A) is both open
94 METRIC SPACES
closed with to f(.A). continuous, is
non-empty subset the subspace which is both and
closed with respect to A. But, since A is connected, the only
non-empty subset which is both open and closed with respect to
.A is.A itself, and so f-I(B) = .A. The mapping f:.A f(.A) is a
surjection, and so f(.A) = B. Therefore the only non-empty subset
subspacef(A) which is both and closed respect
is the whole subspace f(.A); f(.A) is conneeted.
an application of this result, we prove the following theorem
concerning product spaces.
Let the set EI with metric PI be a connected metric space M
I
; let
the set E 2 with metric pz be a connected metric space M 2' The metric
x M
2
, of the set with metric
= Max
is connected.
Let (ai' a
2
), (b
l
, b
2
) be any two points of MI x M
2
The mapping
f:M
I
x M
2
, defined by f(XI) = (Xl' b
2
) maps MI onto the sub-
set x {b
2
} of Ml x The mapping is continuous; if Xl
are any two of M
I
,
p(f(xl),J(YI)) (PI(X
I
, b
2
)) = PI(J;l'
The mapping conserves distance; it is isometric.
Since MI is connected, its image
f(M
I
) =M
I
eonneeted. Similarly X M2 is conneeted. But these
xM
2
have point (ai' eommon. Hence
is connected.
sub-
set
The set Ea1,b
B
contains the points (aI' a
2
) and (b
l
, b
2
). Therefore
cOIUlected of MI which contains a
2
)
\eU"""'.'",,, any other (b
v
b
2
). the component which
(aI' a
2
) is whole space M
2
, and so is
connected.
It is evident that the proof would be unaltered if we had used
instead of P either of the uniformly equivalent metrics p', pH of
FUNCTIONS AND MAPPINGS 95
Connected the Euclidean plane
The theorem of 62 enables us to prove the following result
which characterizes connected open sets in the Euclidean plane.
A non-empty open set in the Euclidean plane is connected if and
only ifevery pair of its points can be joined by a polygonal arc which
the set.
convenient represent of the ,vith
rectangular cartesian co-ordinates (x, y) by the complex number
Z = x+iy, so that the metric is P(ZlIZ2) = /Zl-Z2/' The sphere
N(a; r) is the disc Iz -a/ < r with centre a and radius r. A poly-
gonal arc is a finite chain of straight segments.
prove the 0 be any u v " ~ m c l l
cOlmected of O. Let
subset of 0 consisting of points which can be joined to a by a
polygonal arc; let O
2
be the complement of 0
1
with respect to 0,
the set of points which cannot be joined to a by a polygonal arc.
0
1
and O
2
are disjoint sets; we show that they are open sets.
is any 0
1
, there N(a
1
; contained
since 0 is every point disc can joined
to a
1
by a segment lying in 0 and can therefore be joined
to a by a polygonal arc. Every point of the disc belongs to 011
and hence a
1
is an interior point of 01" As was any point of 0
1
,
0
1
is open.
let a
2
be of O
2
,
contained If any point to
011 we could join a
2
to a by a polygonal aJ:C consisting of the poly-
gonal arc from a to aa, followed by the segment from aa to a
2

Thus a
2
would belong to 0
1
which is impossible since 0
1
and O
2
are
disjoint. Since every point of O
2
is an interior point of O
2
, O
2
is
have thus expressed the eonnected the
two disjoint sets 011 is impossible unless
one of the sets 0
1
, O
2
is empty. Since OIls not empty, 021sempty;
and every point of 0 can be joined to any point of 0 by a poly-
gonal arc lying in O.
To prove the condition sufficient, let 0 be an open set, every
whose points be joined polygonal arc O.
not connected, the union disjoint 0
1
96 METRIC SPACES
of 01' a
2
any of Oz. Since
polygonal are in 0, one
must be with ends b
z
belonging
0
1
and O
2
respectively; call this open segment L, its closure L.
Then, on L, z = f(t), where
f(t) = b
1
(I-t)+b
z
t (0 t I).
,VltlerllvlV f is a continuous mapping open interval
elosod interval and a continuous mapping of
ontoL.
Consider the subsets II' 12 of (0, I) defined by
II = {t:f(t)e0
1
}, I
z
= {t:f(t)e0
2
}
II I), 1
1
1'1
and 02 are
Now 0
1
1'1 Land
2
1'1 L are open with respect to L. Hence their
inverse images II and 12 are open with respect to (0, 1). We have
thus expressed the open interval (0,1) as the union of two dis-
sets, which impossible since is connected,
of 11 and 12
empty, 01 consists of the point {b
1
},
inverse image consists of the single point t = 0. But 0
1
n L is
open with respect to L, and therefore its inverse image is open
with respect to [0, 1], which is impossible. Hence II is not empty;
12 is not assumption is not COIlnelet(IU
proved to be
proof will also with the modifications
Euclidean space of any finite number of dimensions.
A similar theorem is:
A non-empty open set in the Euclidean plane is connected if and
every pair of can be joined arc which
is defined set of points is the continuous
image of a bounded closed interval of the real line. "
64. Continuous mappings of a compact metric space
(;u"Ii""b'lt'ltU'tUS mapping compact metric
x uniformly continuous.
FUNCTIONS MAPPINGS 97
To say that j is uniformly continuous means that, for every
positive value of e, there exists a positive number /1 such that,
if Xl and X
z
are any two points at a distance less than
11, the distance between their imagesj(x
l
) andj(x
z
) is less than e.
The proof follows the corresponding proof for real functions of
a real variable.
Let a be any point of X. Then, for every positive value of e,
we can choose a positive number 8(e, a) such that PI (x, a) < 8(e, a)
implies P2(f(x),f(a)) < In fact there are an infinite number of
such numbers 8(e, a); for if we find one such, any smaller positive
number will do just as well.
The family spheres
{Nl(a; !8(e,a)):aeX}
is an open covering of X. Since compact, we choose from
this covering a finite sub covering
{N1(a
r
; !:8,):r 1,2,. Ie},
where al' a
2
, , a
k
are points of X and 01' = 8(6, a,). Let o(e)
be the least of the numbers 81> 8
2
, , 8
k

is any of X, it lies at least one sphere of the finite
subcovering, say in Nl(a
S
; 18
8
), so that
P2(!(x
l
),J{a
S
< e.
If xz.is any point such that Pl(X
l
, x
z
) < !o(e), x
2
lies in the sphere
Nl(aS ; 8s), and hence (f() f( ))
P2 x
z
, as < 6.
Therefore P2(f(X
l
),!(X
2
)) < 26,
whenever PI (Xl' x
2
} < !8(e).
Hence if /1(e) i8(1e), Pl(X
l
, x
2
) < ./1(e) implies that
< e.
Iff: X -7 Y is a contin'uous 'mapping of a compact metric space
X onto a metric space Y, Y is compact.
Let {Yn} be any sequence of of For Yn' can
choose a point xn of X such thatf(xn) = Yn' since the mapping is
a surjection. But X is compact, and so the sequence {x
n
} contains
98 METRIC SPACES
a convergent subsequence {Xk,J; the continuity off implies that
the corresponding subsequence {Ykn} is convergent. Thus every
sequence of points of Y contains a convergent subsequence. This
proves that Y is compact.
If f: X - Y is a continuous mapping of a metric space X into
a metric space Y, the image of any compact subset of X is compact.
When we say that A is a compact subset of X we mean that
the subspace consisting of the points of A with induced metric
is a compact space. Thus we are concerned with the continuous
mapping of a compact space A onto a space f(A). By the pre-
ceding result, f(A) is compact.
If f: X - Y is a continuous mapping of a compact metric space
X into a metric space Y, the image of any closed set in X is a closed
set in Y.
Since X is compact, a set P closed in X is compact. Its image
f(P) is therefore compact and so is bounded and closed.
If f:X - Y is a continuous bijection of a compact metric X
space onto a metric space Y, then f is a homeomorphism.
All we have to prove is that f-
1
: Y _ X is continuous. This
is true since f = (f-1 )-1 maps every closed set in X onto a closed set
in Y.
65. Extension theorems
Let X and Y be abstract sets, and let A be a proper subset of X.
If f is a mapping of A into Y, a mapping g: X _ Y was called
an extension of f if f(x) = g(x) for every point x of A; and f is
then the restriction of g to A.
If f and g are two continuous mappings of a metric space X into
a metric space Y, the set of points x belonging to X such that
f(x) = g(x) is closed.
It suffices to show that the set of points A at whichf(x) =1= g(x)
is open. Let ae A, and let ps(f(a), g(a = 3k; by hypothesis
k > O. Since f and g are continuous, for every positive value of e,
there exists a positive number 8(e) such that
whenever
Ps(f(x),f(a < e, P2(g(X), g(a < e
Pl(x,a) < 8(e),
FUNCTIONS AND MAPPINGS 99
and therefore
3k = P2(J(a), g(a)) ~ P2(f(a),f(x)) + Pa(f(x), g(x +P2(g(X), g(a
< Pa(f(x), g(x)) + 2e.
Take e = k; then, whenever Pl(X, a) < 8(k), Pa(f(x), g(x)) > k.
Hence the sphere ~ ( a ; 8(k)) belongs to A. Every point of A is
therefore an interior point of A, and so A is open.
IJ J and g are two continU0'U8 mappings oj a metric space X into a
metric space Y, and iJ J(x) = g(x) at all points oj a set everywhere
dense in X, thenJ= g.
We know that the set F of points at which J(x) = g(x} is
closed. H J(x) = g(x) on a set A everywhere dense in X, A s; F,
and therefore As; F = F. But A = X; hence F = X, and so
J = g everywhere on X. Thus a continuous mapping J: X ~ Y
is determined uniquely by the valuesJtakes on a set everywhere
dense inX.
H a is an adherent point of a subset A of a metric space X, we
said that J(x) tends to the limit 1 as x tends to a on A if, for every
positive value of e, there exists a sphere N
1
(a; 8) such that
J(x) E Na(l; e} for all points x * a belonging to A" ( a ; 8).
We then write
lim J(x) = 1.
:l:-+a, :l:e:.<f
H aEA and 1 =J(a}, we say thatJ(x) is continU0'U8 at a on A.
Let A be a set everywhere dense in a metric space X, and let
J:A ~ Y be a mapping oj A into a metric space Y. In order that
there should exist a continuous mapping g: X ~ Y coinciding with
JonA, it is necessary and sufficient that,Jorevery point a of X,J(x)
should tend to a limit as x tends to a on A. The extension g oj J
Jrom A to X is unique.
The uniqueness follows from the previous result. We have to
prove the condition necessary and sufficient.
The condition is necessary. For if J has such a continuous
extension g, then for every point aEA = X, we have
g(a) = Iimg(x) = lim g(x}.
:l:-+a :l:-+a, :l: e:.<f
7-2
100 METRIC SPACES
But g(x) = !(x) on A. It follows that
lim f(x)
a:--> a, a: .A
which proves condition necessary.
Next suppose the condition satisfied. ]j'or every point a of X,
which mayor may not belong to A, we define g(a} by
g(a} =
as may since a an adherent point of A. the definition
of limit, every positive value of there exists a positive
number 8 such that
f(x)N2(g(a); Ie},
whenever x and Ar. 8}.
If b is any point of N;.(a; 8), g(b) is the limit of !(x) as x -+ b on
A N1 (a; Therefore
g(b}f(A" N;.(a; 8)} S; N2(g(a); ie} c N2(g(a); e),
which implies that is continuous X.
It remains to show thatf(x) = g(x) on A. Let a be a fixed point
of A. Since g(x) is continuous, for every positive value of e, there
exists a N;.(a; in which P2(g(X), g(a Since
f(x) -+ g(a) as x -+ a on A,
there exists sphere N;.(a; such that, A" 8
2
),
P2(f(X}, g(a < e. Hence if 8 is the smaller of 8
1
and 8
2
,
P2(f(X), g(;:c <
whenever xN1(a; 8). As e is arbitrary, this implies that
g(x) f(x)
on A. Thus g is a continuous extension off.
f be a uniformly continuous mapping of a A, everyu,here
dense in the 1netric space X, into a complete metric space Then
there exists a unique continuous mapping g: X -+ Y, coinciding with
f on and uniformly continuous.
FUNCTIONS AND MAPPINGS 101
By previous to prove the existence of g, we to
show that, for every point aX,j(x) tends to a limit as x --* a
on A.
Let be any point of X. Let be any sequence of of
A which converges a. Since uniformly continuous A,
for every positive value of e, there exists a positive number
8(e) such that P2(J(X), j(x')) < e whenever x and x' belong to
A and x') < lJ. Since {x,,} is a convergent sequence, we can
choose positive no such that Pl(X
m
, < 8, whenever
m > n no; hence P2(J(X
m
),J(x,,J) < e, whenever m > n no'
This shows that {j(X'I1)}iS a Cauchy sequence in the complete
metric space Y and hence it converges to a limit which we denote
byg(a).
Next we prove that!(x) converges to g(a) as x --* a on A. Let
x be any point of A such that PI (x, a) < !8. Since {x
n
} converges
to a, we can choose an integer 11,1 so that Pl(a, x
n
} < !8, whenever
n 11,1' Hencepl(x, < 8 and thereforeP2(f(x),f(x'I1)) < when-
evern?J; n
1
.Maken--*oo,rememheringthatp2(b,y)isacontinuQus
function of y. We then have P2(J(X), g(a)) e, whenever xA
and PI (x, a) < !8. This proves thatf(x) tends to the limit g(a) as
x--*a
To that g uniformly continuous, a and at two
points of X. Let {x,,} be a sequence of points of A converging to a;
let be a sequence of points of A converging to a'. Suppose
that < 18, can choose positive n
2
that
!8,Pl(X;" < -18, n2
< 8
whenever m > n it follows that < Make
m --* then 00. This P2(g(a), e, whenever
Pl(a,a') < 18, so that g is uniformly continuous.
If we omit the condition that the continuity of the mapping
of the ever,ywhere dense the metric space X, Y is
uniform, the conclusion that! continuous extension X
does not necessarily follow. A simple example is as follows. Let
A be the set 0 < Ixl Ion the real line: it is everywhere dense in
[-1, Let!(x) = (1TjX) on Thenf(x) continuous A;
but it not continuous since can find of
FUNCTIONS A.ND MAPPINGS 103
If
= piJfI,J),
the sequence {fn(x)} converges uniformly tof(x) on X if and only if
N
n
as n 00.
Let N
n
-+ O. Then, for every positive value of e, there exists an
integer nl(e) such that 0 < ~ ~ f n < e whenever n ~ n
l
. Hence
whenever n n
l

Pa(fn,J) < e,
this Implies that
P2(!n(x),J(x < e,
whenever n ;;,: n
l
, for all points x of X, so that the sequence
converges uniformly to f(x).
Conversely, let convergenee be uniform. 'then, for every
positive value of e, there exists an integer nl(e) such that
P2(fn(x),J(x)) < ie,
whenever n ~ nl(e) for an points x of X. Hence
N
n
= Pa(fn,J) = sUP{P2(fn(x),f(x:xeX} ~ !e < e,
whenever n nl(e). 'rherefore Nfl 0 as -+ 00.
In other words, uniform convergence on X is equivalent to
ordinary convergenee in Z.
67. Uniform convergence and continuity
The limit of a uniformly convergent sequence of continuous
functions of real variahle is continuous. The same is true in
metric space.
If {j':L(X)} is a seq1tence of continuous /unctions defined on a
metric space X and taking values a metric space Y, and if
sequence converges uniformly to f(x) on X, thenf(x) is continuous.
If convergenee is uniform, then, every positive value
of e, we can find an integer no such that P2(fn(x),f(x < e when-
ever ~ no all belonging X. m any fixed
teger ~ no' Then fm(x) is continuous. Hence if a is any fixed
point oiX, we can find a positive number 8 such that
P2(Jm(X),Jm(a)) < e whenever xeN1(a; 8).
104 METRIC SPACES
Therefore
P2(f(x),f(a ~ P2(J(X),fm(X + P2(!m(x),fm(a + P2(fm(a),J(a < 3,
where xNI(a; 8), and so f(x) is continuous at a. As a was any
point of X, the result follows.
68. Real functions on a metric space
A real function on a metric space X is a mapping f: X ~ R of
X the real line. simple example of a function
PI(x,a) where a is any fixed point of X. The function PI(x,a)
is continuous on X; in fact, it is uniformly continous. For if
Xl x
2
are two of
/p(x
l
,a)-Pl(x
2
,a)/ ~ Pl(X
1
,X
2
) < e,
whenever x
2
) More generally, if A is fixed non -empty
subset of X, d(x, A) is a uniformly continuous real function on X.
In the simplest case, the metric space X is the real line, and the
results of this section then reduce to known results in theory
of functions of a real variable.
IJ J is a real Junction on a metric space X and continuous on a
compact subset A of f is uniJormly continuous and bounded
onA and attains its supremurn and infimum on A.
Since the subspace A is compact, uniform continuity follows
from previous result. The image J(A) is a compact on the
real line; it is therefore bounded and closed, and so contains
its supremum and infimum.
is a Junction on a metric X contin/uous on
connected sub8et A of X, the image f(A) i8 an inte'fval.
The image J(A) is a connected set of the real line and so is an
interval. If and b any points of f(A), every point b e ~
tween a and b belongs to J(A).
IJ J and g are two continuous mappings oj a metric space X
into real the of poi'nls at which J(x) g(x) is open.
Let A be the set of points of X at which f(x) > g(x). If A is
empty, it is open. If A is not empty let a be any point of A, so
that J(a) > Choose a real number so that J(a) > g(a).
Since 11 = (b, + oo) is an open set, the inverse image f-l( 11) is
an open set containing a; onit,J(x) > b. Similarly if 12 = (-oo,b),
the inverse
g(x) < b.
is an open
dently
AND MAPPINGS
an open set containing
" g-I(IJ
on which f(x) > b >
f-I(II) "g-I(I
2
) s A.
Thus a, being a point of an open subset of A, is an interior point
of A. Hence A is open.
It follows that the set of points on whichf(x) g(x), being the
complement of A, is closed.
If f and g are two continuous mappings of a metric space X
into the real line, and if f(x) g(x) on a set A everywhere dense in
on a closed set F s;;;
F = F, which proves
X, thenf(x)
We know
that A s
If{fn}
sequence of real functions,
on a
onX. (Dini's
and if the sequence ran'/lJl'r'flI'R
each point of X, the convergence
Suppose that the sequence is increasing. For every positive
value of e and for each point a of X, there exists an integer n
such that 0 f(a) - fn(a) < e; n depends on e and on a. But since
f andfn are continuous, there exists a sphere N(a; 8) such that
/f(x) - f(a)/ < e, /fn(x) - fn(a)/ < e,
whenever x N( a; 8) ; the radius 8 depends on a, e and n and there-
fore on a that 0 f(x) - fn(x) <
xN(a;
There
all the
point a
r

each point a of X,
open covering of X. But
finite set of points all
spheres form a finite
values of nand 8 C01Teeiponding
If x is any point of X, it belongs to at least one sphere of this
finite family. Suppose it belongs to the rth. Then if n n
r
,
o f(x) - fn(x) f(x) - fn,.(x) < 3e.
106
lIiETRIC SPACES
Let no be the greatest of nlJ n
2
, , n/c. It follows that for every
positive value of e and for all XX,
whenever n ;;:: no. The convergence is therefore uniform.
69. The extension of real continuous functions
In this section we prove the Extension Theorem of Tietze
and Urysohn, thatiJJis a bounded continuous realfunction defined
on a closed S ~ t b 8 e t F oj a metric space X, there exists a continuous
real function g defined on X which coincides with Jon F.
The proof depends two lemmas.
LEMMA 1. For every pair di8joint closed sub8ets A and B a
metric space X, there exists a continuous real Junction g such that
g(x) = - for XEA, g(x) + for X -1 < g(x) 1 Jar
xA' n B'.
'1'he space X is the union of A, B and A' n B'. The function
(x) = d(x, A) - d(x, B)
g d(x,A)+d(x,B)
is continuous on X and has the desired properties.
LEMMA 2. Iff is a continuous real Junction defined on the closed
subset F oj a metric space X such that If(x) I 3, there exists a
continuous real function g defined on X such that Ig(x)1 ~ 1 for all
xdl, If(x)-g(x)1 ~ 2Jorallxd!"and Ig(x)1 < IJorall xF'.
Let A be the set of points of P such that 3 ~ f(x) ~ - Then
A is a closed subset of the closed subspace P, and so is a closed
subset of X. Similarly if B is the set of points of If' such that
1 ~ f(x) 3, B is closed. Moreover A and B are disjoint. The
function g(x) of Lemma 1 satisfies the stated conditions.
'1'here no merit in the constant 3. If If(x)i M, we can find
g(x) so that
Ig(x)1 ~ iM on P, If(x) - g(x)1 ~ 1M on P, Ig(x)1 < iM on P'
merely by multiplication everywhere by the factor 1M.
To prove the extension theorem, we construct a Cauchy
FUNCTIONS AND MAPPINGS 107
sequence {Yn(:X:)} ofren! functions, continuous on whose limit
has the desired property.
I ~ e t ~ on By I.emma we find real
Yl(X), continuous on X, such that
If(x) Yl(x)i ~ ill! (x
IY1(X)j ~ iM (xeX).
lVIaking ! - Yl play the role of f in Lemma 2, we next fuid a
real function Y2 - Yl' continuous on X, such that
If(x)-Y2(x)i " (i)2M (xF),
ig2(X) Yl(X)/ ~ !(i)2M X).
And so on. An easy induction shows that this leads to a sequence
functions {gn(x)}, each continuous on such that
If(x)-Yn(x)1 ~ (i)nM (xEF),
IYn(x) - gn-l (x) I ::;; !(i)n M (x eX).
follows that, if m n,
m m
IYm(x)-Yn(x)! ::;; ~ !Yr(X)-gr-l(X)! ::;; l M ~ t i t < (i)nM.
nTl n+1
Hence for every positive value of e, there exists an integer no
such that IYm(x) -Yn(x)i < whenever m n ~
For every xeX, the sequence {gn(x)} is a Cauchy sequence of
numbers and is convergent. The limit g(x) therefore
exists for all XEX, and y(x) = f(x) on F. :Moreover, the Cauchy
sequence is a uniformly convergent sequence of real functions,
eaeh continuous on ,hence y(x) is continuous on whieh
proves the theorem.
The restriction that f be bounded on can removed
considering for example the function tanhf(x).
EXERCISES
1. A and E are sets;jis a mapping of A into E, g is a mapping of E into
A, If the mappinggojof A into A is a surjection and the mappingjog of
into B injection, prove that! and are both bijections.
108 TRIC SPACES
2. The identity mapping i : X ~ X is defined by i(x) = x for every
point x of X. If f: X ~ X is any mapping, prove that = i of = f.
Show also that, iff: ~ X bijeetion,foj-l =f-1 0 f = and that there
is then no mapping g:X X such thatfog = gof = i other thanf-1.
3. Prove thatfi'l continuous mapping of the metric space X the
metric space Y if and only iff-1(B) S;; f-
1
(11) for every subset B of Y.
Prove that f is continuous mapping of the metric space X the
metdc space Y if and only if f-
1
CInt B) s: Intj-I(D) for every subset B
ofY.
5. j:X is a mapping of a metric space X into a metric space Y,
and A and B are open subsets of X. Prove that, iff is continuous on A and
on B, it is continuous on Au B. Is this result true if A und B arc elosed
subsets? Is the result true for the union of an infinite number of open
sets, (ii) an infinite number of closed sets?
j:X is a mllpping of a metric space X into a metric space Y.
The point a is an adherent point of a subset A of X, and! is continuous
at a. Prove thatf(a) is an adherent point off(A).
Prove that j is a eontinuous mapping if and only
for overy subset A X.
7. Prove that a necessary and sufficient condition for the bijection
f:X Yto a homeomorphism thatj(.4) =j(A) for every subset A
ofX.
8. fis the continuous mapping of the real line into the real line defined
by j(:I:) = . Find (i) an open 0 such that is open, (ii) a
closed set F such thatf(F) is not closed, (iii) a set A such thatf(A) c: f(A).
If f is continuous mapping the metric space X into the metric
space Y and ifg is a continuous mapping of Y into Z,prove that gofis a
continuous mapping of X into Z. Show that the result also holds if
'continuous I is replaeed by , uniformly continuous'.
10. f is a real valued function defined on a metric space X and continu-
ous an subset 0 of X. Iff LS positive the point a prove tha.t
there is a sphere N(a;r) on whichj is positive.
11. A is a non-empty subset of a metric space X, and d(x,A) is the
distance of the point:/: from A. Prove that d(x, A) uniformly continuous.
A and B are separa.ted non-empty subsets of X. Show that
{x:d(x,A) < d(x,B)}, {x: B) < d(x,A)}
are disjoint open sets containing A and B respectively, and that the set
d(x,A) = d(x,
is closed.
FUNCTIONS AND MAPPINGS 109
12. f(x,y) is a continuous real function of two real variables x and y.
Prove that the curve in the Euclidean plane whose equation isf(x,y) =
is a closed set.
13. Prove that
-I"t"Iofthe
plane.
14. M is the subset
Prove that
of the closed interval
disc Izl " 1 of the complex
= 0, x(I) = 1, and
is a continuous mapping of Minto [0, 1]. By considering the sequence
{tn} of points of M, show that M is not compact.
15. f is a real continuous function defined on a metric space X. At
each point of a subset A of X,f(x) is either equal to + 1 or to - 1. Show
that A is connected, if and only if every such f(x) is constant on A. De-
duce that, if {A,,: 1% sets in X such that n A"
is not empty, then U ,,S
"
16. {F,,} is a sequcnce
such that F"+l S;; F"
and not empty. Let/:
00
compact metric space X
00
that F = n F" is closed
1
CUll wLioHllJU" mapping of X into a metric
space Y. If Y nf(F,,), prove that there is a subsequence {F k,.} of {F,,} and
1
that x"" Fk,.,f(xk,.) = y, and
{Xk .. } converges to a point x. Prove that xF" for every integer n, and
00
deduce that nf<F .. ) S;;f(F).
1
17. M is a metric space consisting of a set E with metric p. M x M is
the set Ex E with
PlXl,
Prove that the
uniformly continuous.
X 2),P(Yl'YZ)}'
defined by f(x,y) = p(x,y) is
18. f:X -+Xis a metric space X withmetricp
into itself, and a
o
and X. Sequences {a,,} and {b,,}
are defined by a" = f(a"-l) , b" = f(b"-l)' Prove that there is an increasing
sequence {k,,} of positive integers such that {a" .. } and {bTc,,} are convergent.
If the mapping is such that
p(a, b) EO; p(f(a),f(b,
10 TRIO SPACES
show that, for every positive value of e, there exists an integer no such
that, whenever n > no,
< e,
where l" = k .... Hence prove that
p(a
1
, b
l
) ~ p(az,., bz,.) ~ p(a
o
, b
o
>+ 2e.
Deduce that, for every pair of points of X, p(a,b) = p(f(a),j(b.
'l11e sequence {az,.} con verges to ao' Prove that the sequence {az,.-l} also
convergent sequence limit a', thatj(a')
CHAPTER 8
SOME APPLICATIONS
70. Fixed point theorems
This chapter is concerned with applications of the theory of
metric spaces to algebra and analysis. All depend on fixed point
theorems.
To illustrate what is meant by a fixed point property, let us
consider the homographic transformation
az+
w=--
cz+d'
where a, b, c, d are real or complex numbers such that ad-be =1= O.
This transformation is a of the extended complex plane
onto itself. A point whose position is not changed by the trans-
formation is called a fixed point. 'rhere are two fixed points,
whose affixes satisfy the quadratic equation
ez
2
+ (d-a)z-b = 0;
if c = 0, one of the fixed points is the point at infinity. '!'he same
transformation occurs in projective geometry; the fixed points
of a homography on a straight line or conic are caned self-
corresponding points or united points.
The problem of solving an algebraic equation J(z) 0 call be
expressed as a fixed point problem. The relation
w = z+J(z)
maps the complex plane into itself; and the fixed points are the
zeros of!(z). Again, consider thc system of n lincar equations in
nvariables
or, in vector notation, Ax = c,
[111 ]
2,3, ... ,n)
112
METRIC SPACES
where A is the square matrix with a,,8 in its rth row and sth
column, and x and c are column-vectors
Denote by I the unit matrix with 6,,8 in its rth row and sth column,
where 6,,8 = 0 or 1 according as r =1= 8 or r = 8. Then the trans-
formation
y = (I - A) x + c,
maps the space of vectors into itself; if we put y = x to get the
fixed points, we find that they satisfy Ax = c. H A is not a
singular matrix, there is one fixed point; but if A is singular,
either there is no fixed point or an infinite number.
71. Contraction m a p p i n ~ s
It is often possible to prove the existence of a fixed point under
a given mapping by showing that the mapping is a contraction
mapping. An example of this occurs in Cauchy'S method of
successive approximations for proving the existence of a unqiue
solution of a first-order ordinary differential equation. A more
elementary example, which occurs in the first chapter of Brom-.
wich's Introduction to the Theory of Infinite Series, is the numeri-
cal solution of a real algebraic equation x = f(x). This is the
problem of finding the fixed points of the mapping y = f(x) of
the real line into itself. The method is to construct a sequence
{an} of real numbers, defined by aMI = f(a
n
); if the sequence
converges to the limit IX, then IX is a root of the equation x = f(x).
Letf(x) be continuous in a closed interval [a, b] which contains
the root IX, and letfmap [a,b] into [a,b]. Hf(x) is differentiable
in (a, b), the condition I!,(x)j ~ k, where k < 1, ensures that the
method leads to the solution. For, if Xl and X
z
are any two points
of [a,b], we have f(xl)-f(x
z
) =!,(g)(xz-x
z
),
where g lies between Xl and x
z
, and so
If(xl)-f(xz)1 ~ klxl-xzl
It follows
and therefore
APPLICA TIONS
/f(a
n
) - f(a)l klan - ,
kn-Ila
l
-al
Since 0 < k < 1, an a as n 00.
The condition is sufficient. For
lan+1-anl klan -an_II
and, so, if m > n,
lam-ani lam-am-ll + lam- I -am- 2 1 + ... + lan+l-anl
(k
n
-
1
k
n
+ ... +k
m
-
2
)1a
2
-a
l
!
Since 0 sequence {an} is a Cauchy QAn!HPlnPA
numbers converges to a limit a.
wheref(x) is a root ofx =f(x).
The provided a continuous
the closed interval [a, b] of the real line into itself. If YI and Y2
are the images of the points Xl and X
2
of the closed interval,
IYI-Y21 klxl -x21,
where 0 < k < 1. The distance between YI and Y2 is definitely
less than the distance between Xl and x
2
; this is why the mapping
is called a contraction mapping.
A condition such as
is called
but it docs
continuous
klx
l
-x
2
1 (k>
condition. It implies thatf(x)
is differentiable.
a Lipschitz condition.
72. A fixed for metric spaces
Let No be a 8phere N(a; 8) in the complete metric 8pace X. Let f
be a contraction mapping of No into X which 8ati8fies the Lipschitz
condition
8 CMS
METRIC SPACE
Jor every pair of points Xl and X
2
of No, k being a constant such that
0< k < 1. Then if p(a,J(a)) < 8(1-k), there is a unique point a
in No such that a = f(a).
with the = f(a), we sequence
the relation
Xn+l = f(xn)
We show that each point of the sequence belongs to No and that
it is a Cauchy sequence.
The point Xl belongs to No, since
p(a, p(a,j(a)) < 8(1- k) o.
that Xl 'X2'
UtalllUU. Ifs ~ r,
points of xr+l = f(xr)
p(xs' XS+l) = P(f(XS-l),J(XS)) ~ k p ( X ~ l ' xs)
~ kSp(a,x
1
) < kB8(1-k).
Since 0 < k < 1,
xr+l) ~ p(a,
1+
X
2
) + ...
... + ld") 8(1
so that xr+l belongs to No. By induction {xn} is a sequence of
points of No.
Next remembering that 0 < k < 1, we find that, if m > n,
xm) ~ p(xn' + p(xn+l' xn+2) xm)
But since k
n
-+ 0 as n -+ 00, for every positive value of 6 we can
find a positive integer no such that 8kno < 6. Hence if m > n ~ no,
p(xn' xm) < 6. The sequence {xn} is thus a Cauchy sequence; since
X is complete, this sequence converges to a point a of X. But
ex) = limp(a, lim (1 +k+
= p(a,x1)/(1-k) < 8;
hence a is a point of No.
Lastly
p(xn,f(a)) = p(f(Xnl),f(a)) ~ kp(x
n
-_
ll
a) -+ 0
Therefore = 0, and This proves
a fixed point.
116 METRIC SPACES
the real line into itself, where /(x)If'(x) continuous on a
closed interval [a,b] and differentiable on the open interval
b). Xl and X
2
are any two points of [a, b] which map into YI'
f /(x1) !(X2)}
Yl - Y2 Xl X
2
-If' (Xl) - f' (X
2
)
f(s)f"@
= (Xl - X2) {f' (S)}2 '
where g lies between Xl and x
2
Hence if
j
f(X)tJX) I :;;. k < 1
{f'(X)}2 ....
[a, the mapping a contraction mapping [a, onto a
closed interval of the real line. It follows from the theorem that
Xo is point <;>f b) such that If(xo)/!'(xo)1 8(1 k), the map-
ping has a unique fixed point ex in (xo - {j, Xo + (j) " [a, b], and that
the Ne"\\1;on-Raphson sequence converges to ct.
74. Systems of linear equations
Consider system of linear equations in unknowns, which
it is convenient to write in the form
n
(r= 1,2, ... ,11.).
8=1
The constants a,., 8 and c,. and the unknowns x,. are complex num-
This system can written vector form
x = Ax+c
or (l-A)x = c,
where XII' , and c = [c
I
, c
2
, , are column vec-
tors, 1 is the unit matrix and A is the matrix with a".8 as the
element in rth row and 8th column. is well-known that the
system has a unique solution if and only if the matrix 1 - A is non-
It of interest to what the fixed point theorem tens
us about the problem.
The set of all column vectors can be turned into metric
space in various ways. If we write
Xl = [xl, ... , X2 = ... ,x!],
SOME APPLICATIONS 117
where the superscripts are merely labels, a simple metric is
,X2) = 1,2,. .. ,
and the resulting metric space is complete. We wish to show that,
wider certain conditions, the .LLLa'.FIJu'6
y = f(x), where = Ax+c,
from this metric space into itself has a fixed point. We have

p(Xl, x
2
)max{ lar.sl: r = 1,2, ... ,n}.
B-1
Hence, if, for every value of r,
the mapping is a contraction mapping.
Let a any point this metric space, which is not
point. Thenp(a,f(a)) where o. If we choose
6> Llj(l-k),
the fixed theorem shows that the has a nnU.,'"IA
fixed point in N(a; ilS 6 can be as large as we please, the
mapping has precisely one fixed point, and so the system of
equations has a unique solution.
Actually is not result. The condition
lar.sl} < 1
known imply that the roots lie in 1.
Hence the latent roots of! - A lie in Iz - 11 < 1. Thus z = 0 is not
a latent root of I-A, so that I-A is non-singular. But there
are matrices A, such that I-A is non-singular, which not
satisfy the condition.
118 METRIC SPACES
A different result is obtained if we use the metric p(Xl, X2)
defined by 11.
{p(Xl,X2)}2 =
2=1
The resulting metric space is again complete. We now have
11. 11.
I: la1',sI2. Ixt-X:12
1'-18=1 t=l
= {p(Xl, la
r
,sI2,
r,s
Henee if
L lar.sl
ll
k
2
< 1
",8
the mapping is a contraction mapping, and the proof of the
existence of a unique fixed point follows as before,
This too, is not a good result. It can beshownthat,if All A2' ... , An
are the latent roots of A,
1i\112+ IA212+ ... + IAnl
2
la,.,sl:!
",s
Hence, if k < 1, all the latent roots of A lie in Izi < I, which
implies as before that 1-A is non-singular; but not all nop.-
singular matrices satisfy the condition.
75. Infinite systems of linear equations
Turn now to the infinite system of linear equations
co
xr=;:sa,.,sxs+c,. (r 1,2,3, ... )
8-1
which we write in vector form
x = Ax+c.
We are now working in a sequence-space, and the results depend
on the particular complete metric sequence-space we choose.
SOME APPLICATIONS
the complete metric
complex numbers
119
bounded se-
In order that the equation may have a meaning, the infinite
series
must converge for every positive integer r and for every point
x of m. Now for every fixed value of r, we can take the bounded
sequence defined by
x _Iar.sl
s-
ars
X
B
= 1 (a,., a
must therefore be COl:1V{)rgenL Thus, in order
LU.\O'q;U.LLU;:.;, ....... for every point necessary that
the
00
lar.sl
6=1
be convergent for every positive integer r; and the condition
00
sup lar,sl < 00
,. B=1
then ensures that Ax belongs to m.
We
infinite matrix
mapping y = f(x) of
f(x) = Ax+

p(Xl, X2). sup I a,., sl }.
3, ... }
120 METIUC SPA.CES
Hence there exists a positive nnmber k, between 0 and 1,
such that
00
lar,sl k
8=1
for every positive integer r, the mapping is a contractiollma pping.
The existence of a unique fixed point then follows as in the finite
case. Thus if c belongs to the space m of bounded sequences and
if there exists a positive Ilumber k (0 < < such that
00
lar.sl k (r = 1,2,3, ... )
s=1
there exists a unique bounded sequence x which satisfies the
equation x = Ax + c.
Next suppose that we are seeking a solution in the complete
metric space l2, which consists of sequences x = {xn} for which
is convergent, the metric being defined by
00
{p(Xl,X2)}2 =
1
In order that the equation may have a meaning, x, c and Ax
must to l2, the first place, Ax must exist; hence the
series
must converge for every positive integer r and for every point x
of 12. As before, the series must be absolutely convergent. And it
00
is known that l: lar,sllxsl is convergent whenever l: Ix
s
l
2
is con-
8=1
00
vergent if and only if 1: la
r
.
s
l
2
is convergent. Thus Ax exists if
s=1
and only if each of the series
2: la
r

s
l
2
(r = 1,2,3, ... )
11=1
is convergent. If y Ax, we then have
00 00 00
IYrj2 l: jar.sl
ll
2: Ix,11l = {P(X,O)}2 2: lar.sli! (r = 1,2,3, ... ).
,=1 t=l
SOME APPLICATIONS 121
00 00
Hence, la
r
,s!2 Is convergent, also belongs
r=1
Suppose now that the infinite matrix A satisfies the condition
that L lar.Blllis convergent; let the sum of this double series be k
2
,
Consider the mapping y = f(x) of l2 into [2, where f(x) = Ax + c.
We then have
Ixl-xW = k2{p(Xl,X2)}2
r,8 t
and so p(f(Xl), f{x2 kp{Xl, X2).
Therefore, lar.sll! 1, the mapping is a contraction mapping.
The existence of a fixed point then follows as in the finite case.
76. Systems of ordinary
order
equations the first
fixed point theorem enables to prove, under suitable
conditions, existence of a set of functions
{xr(t): r = 1,2,3" ,n},
which satisfy a system of real first-order differential equations
=
with the initial conditions xr = a
r
when t = to, for r = 1,2, ... ,n.
Using the notation vectors, we write
x for (Xl' X
2
, ... , x
n
),
F;.(X, t) for F;.(Xl> X
2
, " x
n
' t), and
F(x, t) for {FI (Xl' t), F
2
(x, t), ... , Fn(x, tn.
The system then becomes
dx
dt
F(x, t),
122 METRIC SPACES
initial condition = a. We also
IxJ2 = X ~ , IF(x, t)i2 = {F,.(x, t)}2.
r=1 r=1
If we denote the set of all x by X we can regard F as a mapping
domain and are both in mapping which depends
parameter we denote of all real t
a mapping domain is T and whose is
in X. When we say that F is continuous, we mean that each com-
ponent F,.(x
1
, X
2
, ... , X
n
, t) of F is a continuous function of n + 1
variables in the ordinary sense.
F(x, t) be in the closed - tol :;;; y, 8
it satisfies conditi01't
K being a positive constant. Then, if y is sufficiently small, the
differential equation
dx
dt = F(x,
has in It-tol < y a unique solution which satisfies the condition
x = a when t = to'
We may suppose that to and all the numbers a
r
are zero. Since
continuous y, Ixl :;;; bounded; let the
supremum of may also that y(M 8;
for if not, we may replace y by a smaller number without chang-
ing 8, K or M, though M will then possibly be an upper bound and
not the supremum. If we write Ky = k, we have My < 8(1-k)
0< k < L
the differential equation has x(t) which satisfies
initial condition, each xr(t) is differentiable and thc;reiore
continuous. But since
~ r = F,.(x(t) , t),
Fr is continuous, each dxrldt continuous. Therefore
Hence
ishes
APPLICATIONS
dll1erentl;a.l equation has a solution
integral equation
x(t) = f: F(X(T), T) dT.
Conversely, ifx(t) is a continuous solution of this integral equa-
tion, dx(t)/dt exists and is equal to F(x(t), t). Thus solving the
differential equation is equivalent to solving the integral equa-
tion.
Let 0 be the space of all x(t) continuous on - y t y, with
metric
This space
respect
ordinary
0=(0,
sup /x(t)1
No into
{IX1(t)_X2(t)/: -y t
if {xP(t)} is a sequence
x{!(t) is uniformly
Let No be the sphere
the set of all points x(t)
show that the mapping
y(t) = f: F(X(T), T) dT
has a fixed point.
First, this is a mapping of No into No. For if 0 t y,
and
Therefore
/y(t)/2 = {f:
{f: T)/ dTf
,=:f:
jtf
2
t
2
M2y2;
t O. Hence /y(t)1
8(1-k) < 8,
as required. Incidentally we have proved thatp(O, f(O)) < 8(1-k)
which is one of the conditions of the fixed point theorem.
The mapping is also a contraction mapping. For if yl(t) and
124
the images
we have
METRIC SPACES
points Xl(t) of No, and if
I: {ft [F,.(xl(r), r) r)] ar}2
.. -1 0
i; t r 1F,.(xl(r),r)-F,.(x2(r),r)12dr
.. -1 Jo
Ixl(r) - X
2
(T)12
< K2t
2
sup Ixl(r)-x2(7}jl:l
K2y2{p(Xl, X2)}2 = k2.{p(xl, X2)}2
and similary if - /' :;;; t 0, Hence
kp(Xl, x
2
),
k < 1.
The conditions of the fixed point theorem are thus satisfied.
The mapping has a unique fixed point a(t), and so
F(a(r), 7)
of differential equations
dx
dt = F(x,t)
has therefore the unique solution x = a(t) on 'l' t /" which
"",",,1;;:1110;;;: the initial = 0 when
be noted of order n
dnx (dX d
2
x dn-1X)
dtn = F x, dt' dt2 ' ... , dtn-1 , t
can he expressed as a of first-order equations by writing
d
n
-
1
X=
XII' ... , dtn- 1
The system is
dx,.
(Ii = Xr+l
(1' = 1,2, ... ,11.-1),
SOME APPLICATIONS
125
theorem of tWs o,,'C ",v.u. therefore an existenee
theore.m for a differential equation of any
77. Boundary value problems
Let F(x, y, t) be a real continU0U8 function. If (Xl' Yl, t), (x
2
, Y2, t)
two points X ~ 8, - 0 ~ t ~
a condition
where Po, Pl depend only on 8. Then, for all sufficiently small values
of the real constant A, the differential equation
(
dx
t\F x, dt'
has a unique solution, which is continuously differentiable in
o t ~ I, where it satisfies the conditions Ix(t)1 ~ 8, Ix'(t)l ~ 8,
and which vanishes where t = 0 and t = L
we speak differentiability closed interval
understand the derivatives one-sided at
end-points. Since F is continuous on Ixl ~ 8, /yl ~ 8, 0 ~ t ~ 1,
it is bounded; let M be its supremum. Evidently M will depend
on 8.
If this boundary value problem has a continuously differen-
solution x(t), fmIction F(x(t), continuous
Hence we may the expression each side of
dHrerential equation and obtain
x'(t) = A+A f;F(x(v),X'(v),V)dV,
x(t) = At (t-v)F(x(v), v)dv,
where A and B are constants of integration. But since x(t)
vanishes when t = 0, and t = I, we have
x'(v), v) B=O.
126 METRIC SPACES
It follows that
x(t) = - A(1- t) vF(x(v), x'(v), v) dv
;\t It
1
v) F(x(v), x'(v), dv.
Conversely I if x(t) is a continuously differentiable solution of this
integral equation, it solution of the boundary value problem,
Now consider the metric space A, which consists of all real
functions which vanish at t = 0 and are continuously differenti-
able on [0, with metric
A subspace of complete metric space Q1[0, and
itself complete. For if {xn(t)} is a Cauchy sequence, the sequence
of continuous function converges uniformly in the ordi-
nary sense a continuous function and this implies that
{xn(t)} converges uniformly in the ordinary sense to z(r) dr,
which belongs to A.
Now consider the mapping/of A into itself defined by
- t) vF(x(v), x'(v), v)
-At Ie
l
(I-v) F(x(v), x'(v), v) dv,
We have to show that it has a fixed point for all sufficiently small
values of A. Since
y'(t) = AI: VF(x(v),X'(v),V)dV-Af (l-v)li'(x(v),x'(v),v)dv,
we
ly'(t)1 IAIM vdv+ IAIM It
l
(l-v)dv = tIAIM(I-2t+2t2)
and y) = sup{ly'(t)i.te[O, ll"IM.
If No is the sphere N(O; 8), / maps NO into NO. provided that
IAI 281M,
let YI =
ofN;,. We

SOME APPLICATIONS
(I-v) Gdv,
where G = F(XI(V), v) - F(x
2
(v), v).
Therefore
IG(v)1 Jlolx1(v) - X
2
(v) 1 + Jlllxi(v) - I-
x2(v)j = If:
sup dt
:s:; sup
:s:; (Po x
2
)
::;; (Ito P'I)P(XI'X2) {J>dv+ Ie
l
(1
follows that
= !(Po+ JlI)P(XI' x
2
) 1"1 (1- 2t + 2t2)
lIAlvlo+,uI)P(X1 ,
127
two
Therefore, if 1"1 < 2j(PO+,uI)' the mapping f of No into No is a
contraction mapping.
The conditions of the fixed point theorem are thus satisfied if
. (28
< nun M'
integral and the boundary value
problem have a unique continuously differentiable solution for
all sufficiently small values of 1"1.
If the boundary conditions were x(O) = co' x'(O) = C
1
, the
could be through modifications,
reduce the to the solved here by x
cIt-CO'
128 METRIC SPACES
78. Fredholm's integral equation
Let ifJ(s) be continuous on the interval a ::::;: s ::::;: b. Let K(s, t) be
contin'uous on square a ::::;: 8 b, a ~ b. Then for suffi-
ciently small 'lJal'ues of 1il.1, the integral equation
x(s) (s) + AI: K(s, x(t) dt
has a unique solution, continuous on [a, b].
Let X be complete metric space O[a,b] all continuous
functions x(s) with metric
p(x
t
,x
2
) = sup{lx
1
(s)-x
2
(s)l:a::::;: s::::;: b}.
Since the expression the the integral
equation is continuous on [a, b],
y(s) (s) + K(s, x(t)dt
is also a point of X. If we write y = f(x), f is a mapping of X
into We have to prove that, for all sufficiently small values
of lill, the mapping has a unique fixed point.
In the general theorem, take No to be the sphere N(O; ~ ) , where
a can chosen as large as we Since ifJ and are continu-
ous, they are bounded; let lifJ(s)1 =:;; m,jK(s,t)l ::::;: M. Then, if
Yl = !(x
1
), Y2 = !(x
2
) ,
IYl(s)-Y2(s)j = lilI: K(S,t){X1(t)-X2(t)}
dt
l
lill M(b a)p(xv x
2
),
and
p(f(X
1
),J(X
2
= sup IYl(s)-Y2(s)1 =:;; lill M(b-a)p(x
1
,X
2
)
Therefore if choose any positive number k < k and
consider values of lill such that lill M(b - a) ::::;: k, the mapping is
a contraction mapping.
Again, since 1(0) =
p(O,f(O = sup lifJ(s)1 ::::;: m.
If we suppose chosen so that > - k), second con-
dition of the theorem is satisfied.
It follows that there is a unique point IX in No such that IX = f(IX).
SOr.'l:E APPLICATIONS 129
8 can large as please, means for every suffi-
ciently small value of 11\1, the integral equation has a unique
continuous solution x{s) = a{s).
In particular, for all sufficiently small values of 11\1, only
solution homogeneous
x{s) = 1\ J: K{s,t)x{t)dt
x{s) == }'or if this were not the solution of non-
homogeneous equation would not unique; could to it
any solution of the homogeneous equation.
Volterra's equation
Let rp(8) continuous on the 'interval a 8 b. Let t) be
continuous on the triangle a t s b. Then for all values of 1\,
the equation
= rp{s) K{s, dt (a b)
has a unique continuous solution.
This differs from the Fredholm equation in that the upper
limit of the integral is variable and that there is no restric-
tion on the value of
Let X be the complete metric space Ora, b] of all continuous
functions x{s) with metric
x
2
) {IXi{S) b}.
Since the expression on the right-hand side of the integral
equation is continuous on [a,b],
y{s) qS(s)+1\ f: t)x{t)
is also a point of X. If we write y =f(x), f is a mapping of X
into itself.
In the fixed point theorem, take No to be the sphere N(O; 8)
where 8 be chosen large please, Since rp and are
continuous, they are bounded; let 1(s)1 m, t)1 'l'hen
IYl(s)-Yz{s)1 = 11\ J: K{S,t){X1{t)-X2(t)} dt l
11\\.Lll(b-a)p(xv
9
eMS
130 METRIC SPACES
just as in the Fredholm case. Thusf is a contraction mapping
for sufficiently small values litl.
Denote by In) the composition
fofo .. , of
n factors, and In)(x) = lin), It
duction that
n-l fb
'!In) = rp(s) + r= Ar a K,(s, t) (t) dt
where
and
t) K(s,
Kr+l(S, t} = f
a t 8 b.
Now IK1(s, t)1 M. Hence
readily proved by ill-
itn Kn(s, x(t) dt,
u}K,.(u, du
IK
2
(s,t)i = If K1(S,U)K1(U,t)dUI M2(S-t),
generally,

IKr(s, t)1 (n-l)! .
Using this bound for IKrl. we have
Therefore
SOME APPLICATIONS
= sup {1!A
n
)-

n.
131
But 1"-lnMn(b-a)n/n! tends to zero as Hence we can
choose n so that 1"-lnMn(b-a)fi
k = -'---'--n--':!'------=-
lies between 0 and 1. Therefore, with such a choice of n,
contraction mapping,
= sup \ (8) + :;:
"m{I+:gl"-lr
which is finite. We can therefore choose 8 so large that
p(O,j<n)(o < 8(1- k).
The conditions of the fixed point theorem are satisfied. The
mapping y = j<n)(x) has therefore a unique fixed point a(8) in No.
As we saw in 72 it follows that a(8) is a unique fixed point
of the mapping y = f(x). Hence VolteITa's integral equation has
a solution, irrespective value of A.
80. theorem with
Let metric 8pace
8phere T be a metric 8pace
8phere f be a continU0U8 ma.1nl1,mn
X which sati8fies the Lip8chitz condition
Pl(j(Xl , t)f(x
2
, t " kpl(X
1
, x
2
)
let Nl be tke
let Na be the
into
for every pair of point8 (Xl' t) and (Xa, t) of x Na, k being a constant
132 METRIC SPACES
8uchtltat 0 < k L t)) 8(1- k),
there exists uniq'ue mapping a(t) of N2 into such that
a(t) = f(a(t), t),
and a(t) is continuous on
The metric of X x T is anyone of the equivalent metrics of
61.
Apart from the reference to the parameter t, this theorem is
identical with that of 72, so that there is no need to prove the
existence of a unique a(t) for each point t of It only remains to
prove that a(t) is continuous.
As in 72, we form a sequence {xn} now defined by relations
Xl = f(a, t), xn+1 = f(x
1P
t),
where each Xn depends on the parameter t; and the proof shows
that {xn(t)} is uniformly convergent toa(t) on the closed sphere
K2(b; 11)' where 11 is any number less than 1. Since !(x, t) is
continuous, Xl is continuous; it follows by induction that each
member of the sequence is continuous on K2(b; 11)' Therefore a(t),
being the uniform limit of a sequence of continuous functions
r
is continuous on K2(b; Yl) for every Yl < y, and hence on N
2
.
81. The inverse function theorem
A simple consequence of the fixed point theorem with a para
meter is the inverse function theorem of elementary analysis.
One form of this theorem is as follows.
Let F(x) be a real function of the real variable X which satisfies
the condition
for every pair of points of the interval lx-a! < 8. Let .F(a) = b.
Then iflt-b! < m8, the equation F(x) = t 008 a unique solution
x = a(t), where a = a(b), and a(t) is continuous on !t-hl- <mao
We apply the fixed point theorem to
f(x,t) =
SOME APPLICATIONS 133
where.:\ is a real constant to be suitably determined, the metric
spaces X and T both being the real line with the usual metric.
The condition satisfied by F(x) implies that F(x) is continuous.
Hence/is a continuous mapping of X x T into the real line.
The condition
p(J(x
l
, t),J(xa, t ~ kpl(X
l
, xa)
becomes
IXI - Xa - ':\{F(Xl) - F(xa)} I ~ klxl - xal
or
or, again,
1
1_.:\F(Xl)-F(xa)1 ~ k
Xl-Xa
1-k .:\F(xl)-F(xa) ~ I+k.
Xl-Xa
This inequality is satisfied if we take
1-k l+k _ M
T=m, .:\ - ,
which gives
so that 0 < k < 1. The other condition
becomes
Pl(a,/(a, t < 8(1- k)
It-bl < m8,
which is one of the assumptions.
The conditions of the fixed point theorem are thus satis-
fied. It follows that, if It- bl < mtt, where b = F(a), the equation
F(x) = t has a unique solution x = Gt(t), where Gt(t) is continuous.
82. The implicit function theorem
This theorem is concerned with the circumstances under
which a set of n equations in n unknowns Xl' X
Z
' , X" of the form
.F,.(xl,x
a
.... ,x,,; tl,t
a
, ... ,t
m
) = 0 (r = 1,2, ... ,n)
involving m parameters t
l
, ta ... , t
m
, can be solved. It is convenient
to use vector notation. Let us regard (Xl' XII' ... , X,,) as the co-
ordinates of a point x in an n-dimensional Euclidean space X and
(t
l
til' ... , tm) as the co-ordinates of a point t in an m-dimensional
134 METRIC SPACES
Euclidean space T. 'l'he product space X x T is an (m+n)
dimensional Euclidean space. We write the set of equations as
F,.(x; t) = 0 = 1,2,. n)
or F(x; t) = 0,
where F is a point of X, depending on x and on t.
Suppose that each component F,. of:F is continuously differen-
tiable on an open set S of X x T, Let
_ O(Fl' F
2
, , Fn}
- O(Xl' X
2
, , xn)
be the Jacobian ofF with respect to x. If J =F 0 at a point (XO; til)
of S and if F(xO; to) = 0, the equation F(x; t) 0 has a unique
80lution x = <I(t) which is continuously differentiable on a sphere
It tOIl"
We may suppose XO = 0, to = O. Let A(x; t) be the Jacobian
matrix which has oF,./ox
s
i.n its rth row and 8th column; then
J detA. hypothesis, the matrix, A.
o
= A(O, 0) non-
singular and so has an inverse Aill. If A
o
is not the unit matrix,
consider the vector
The Jacobian matrix of G is Ail
l
A which reduces to the unit
matrix when x = 0, t O. Since G 0 if and only if F = 0, it ..
suffices to consider a system of equations:F = 0 whose Jacobian
matrix the unit matrix when x = 0, t = O. Let us apply the fL'l:ed
point theorem with a parameter t to the mapping f: S X, where
f(x; t) = x-F(x; t).
A fixed point of this mapping is a solution of the equation
F(x; t) O.
The first component of the vector f(x'; t) - f(x; t) is
fl(x'; t)-fl(x; t) = x ~ -xl-{Fl(x'; t)-Fl(X; t)}
n
= ~ -x
l
- ~ {Fl(Xr; t) -Fl(xr+l, tn,
1'=1
where Xl = X', X
n
+
l
= X
but
APPLICA'l'IONS
if 1 < r < n+ 1. Since F1 is differentiable,
F
1
(x'; t)-F1(X"+I; t) = (x;-x,)Fl,,(;';t),
lies between and x; and denotes
Hence
n
fl(X'; t)-f1(X; t) = c,(x;-xr)'
,=1
(r> 1).
{F
1
.,(x; t). ...,n}isthe of the Jacobian
matrix A(x; t). By hypothesis, each element of A is continuous,
and A tends to the unit matrix as x and t tend to zero. Therefore
each of the coefficients C
v
c
2
' , c
n
tends to zero as x, x' and t
zero. For positive value there exist
y
1
and8
1
Icrl < ewhenever < 81> Ix'i
Y 1. Therefore
Ifl(x'; t)-fl(X; t)1 < e!x'-x!"n,
whenever Ixl < 8
1
, Ix'i < 81> It I < Yl
for value of r, there positive numbers
8, such that
If,(x'; t)-f,(x; t)! < elx' -xl"n,
whenever Ix! < 8,., Ix'i < 8" !tl < Yr. Hence
f(x; t)1 < ne
whenever Ix! < 8 It I < Y we choose
that k = ne < 1, the mapping f is a contraction mapping.
Next
p(O,f(O; t) = IF(O; t)1 -+ 0
as Itl-+ O. We may therefore suppose that Y was chosen so small
t) < p(1
whenever It! < Y, the second condition of the fixed point
theorem.
136 METRIC SPACE
It follows that the equation x = f(x; t), that is, F(x; t} = 0,
has a unique solution X = a.(t) which continuous It I y
and satisfies the condition cx(O) = O.
It remains to prove that I%(t} continuously differentiable.
Let t and t+Llt be two points such that It I < ')', It+LltI < y.
Write x = cx(t), x+Llx = tt(t+ Llt). Then F(X+TLlx; t+TLlt)
vanishes when T = and when T 1. Consider each component
of this vector separately.
By mean value theorem, exists a number
Tr (0 < Tr < 1)
such that
dT F,.(x + T Llx; t + T Llt)
vanishes when Since is differentiable,
11 oF. m oF.
:E c,......!.ll.xs + :E r Llts 0,
8= 1 uX
8
8= 1 () 8
each derivative at xr = X + TrLlx, tr = t+T,.Llt.
This a system n linear equations for Ll:?:2' , ,
the determinant of the system has Fr.s(x
r
; tr} as the element in
the rth row and sth column, where F,.,s(x; t) denotes oF,./8:?:,q.
'l
'
his determinant is a continuous function of the position of the
n points (xr; tr) on an open set 8 in X x T. Its value when the.
points coincide at 0) unity sincc then reduccs to the
Jacobian. Therefore, if y and 8 are small enough, the determinant
bounded from zero. We can therefore solve equations
the components Llx
r
, in the form
m

8=1
where the coefficients c
r

8
continuous funct,ions of x, t,
andLlt.
In particular, if the only non-zero component of Llt
8

Llxr
Tt = C,.,s
8
and c
r

8
tends to a continuous limit as -+ O. Therefore OlXr/ot
B
exists and is continuous.
CHAPTER 9
DEVELOPMENTS
83. Banach spaces
The applications of the theory of metric spaces in Chapter 8
were all in the fields of classical algebra and analysis. We turned
various sets of 'points' into metric spaces by defining a 'dis-
tance'; but in each case, we assumed that the space had
in addition a structure which was not part of the general
theory One line of
require abstract set always
structure.
Let additive Abelian group,
properties:
y of elements of X
defined' x+y, and x+y = y+x.
(ii) Addition is associative; (x + y) + z = x + (y + z).
(iii) There is a unique neutral element (J such that x + (J = x
for every element x.
(iv) Every element x has a unique additive inverse, denoted
by -x, such that x+ (-x) = (J.
An additive Abelian group X is called a real vector 8pace if it
admits the operation of multiplication by real numbers IX, fJ, ... ,
an operation called scalar multiplication, with the following
properties:
there is
(ii) lX(fJX) = (lXfJ) x.
(iii)
(iv)
These -l)x = -x and that
A complex vector 8pace is similarly defined; the scalar multi-
pliers IX, fJ, ,)" ... are then complex numbers. When we speak of a
vector space, we may mean a real vector space or a complex
vector space according to the context. A simple example of a
[ 137]
l'IIETRIC SPA
real vector space is the set of all ordered triples of real numbers
(xv X
2
, xa) with the rules
(Xl' X2, Xa) + (Yll Y2' Ya) = (Xl + Yll X2 + Y2' Xa + Ya)
this is ordinary vector algebra without scalar and vector pro-
ducts .
.A normed vector 8pace is a vector space on which is defined, for
each point X of X, number called the norm of X and
,such that
Ilxll 0 = 0 if and = O.
(il) For each scalar llluitiplier IX and for each point x of X,
IIX+yll
example, in the possible norm

A normed vector space can be metrized by taking
y) = Ilx-
which has -Y, 0). In
lar, p(x, 0) = IIxll so that the norm of x is the' distance of x from'
the origin'. If a normed vector space is complete with respect to
the metric IIx-ylI, it is called a Banach space. The complex
plane with norm Izi is a Banach space.
space of sequences x = {x
n
} of
space if
eomplex numbers
and () is the sequence every member of which is zero. The follow-
sequence-spaces :Banach spaces:
The set m bounded sequences = {x
n
} with
FURTHER DEVELOPMENTS 139
ao
(il) The set 7:P of all sequences x = {xn} for which Ixnlp is
1
convergent, where p 1, with norm
b] of functions
space if we use
by a constant.
on [a, b], be-
UL"UllIJAV',U' of addition,
norm
Ilxll = sup{lx(t)l:te[a,b]},
O[ a, b] is a Banach space.
All these spaces were used in earlier chapters; but there was
no need to introduce the term' Banach space' as no use was made
of the vector space property.
84. spaces
product is
vector algebra on a real Eucli-
are two types of product,
Ifx = (xvx2,xa)
In particular, x.x is the square of the length of the vector x.
In order to avoid confusion between multiplication of a vector
by a scalar and the scalar product of two vectors, the scalar
product of two vectors is often called their inner product.
vector space. A function (x,y)
of (x,y) of points of inner product
if
linear in x.
the complex conjugate
O.
(iv) (x,x) = 0 if and only if x = O.
By (i), we mean that
(ax
I
+/Jx
2
,y) = cx(x
1
,y)+/J(x
2
,y)
140 METRIC SPACES
If we denote the conjugate of a, complex number z by Z, (ii) then
gives
(x, aYl f J y ~ = (rxYl+PY;;X)
= a(YI' x) + P(Y2' x)
+P(x,
In particular
(ax,y) = a(x,y), (x,ay) = ?i(x,y);
hence, putting a = 0,
y) = 0,
Moreover it can be shown that
fJ) = o.
l(x,y)12 ~ (x, x) (y,y)
with if and if x and are linearly dependent.
A space on an product been defined is
called an inner product space. A trivial example is the set of all
ordered triples x = (Xl' x
2
' x
s
) of complex numbers with
y) = X
2
Y2+
'rhe set of all complex valued functions x(t), continuous on a
closed interval [a,b] of the real line, is a complex vector space;
if we
y) = ~
y(t) dt,
it becomes an inner product space.
If, on mner space, write
(x,x) Ilx112,
Ilxll satisfies all the conditions for a n ~ ; conditions (i) and (ii)
are evident, and ~
Yl12 = (x x+y) x) + (x, (y,x) y)
= Il xI1
2
+2rl(x,y)+ IIyl12
~ IIxl12 21(x,y)1 IIyl12
~ II xii 211
x
llllyll IIyl12
and so
Ilx+yll ~ II
x
ll + lIyll,
which is condition
FURTHER DEVELOPMENTS 141
An inner product space can be regarded as a metric space with
metric p(x, y) defined by
p(x,y) = Ilx-yll.
If the inner complete with respect
space is the space
<Xl
x = {x
n
} such that Ix
n
l
2
is e0I1Verge,nt.
1
is a vector space a{Xn} = {ax
n
}, {Xn} + {Yn} {Xn + Yn}.
It is an inner product space if we define
<Xl
(x,y) = ~ X n Y n '
1
From this, :we derive the norm
And it respect to the metric I I x ~
David Hilbert in his work
forms in variables which he applied
of integral
Another space of measurable J.lUHjI)UWC"
such that Ix(t)12 is integrable in Lebesgue's sense over the closed
interval [a,b]. We do not distinguish between two equivalent
functions, two functions which differ only on a set of zero
measure, so that a 'point' of the space is really an equivalence
class of equivalent functions. With the obvious definitions of
addition, multiplication by a scalar constant and zero, this is a
vector space. The expression
satisfies
the norm
= f: X(t) y(t) dt
an inner product. From
J f: Ix(t)1
2
dt;
and the space is complete with respect to the metric II x - yll .
142 METRIC SPACES
85, General topology
In 83-84, we indicated two lines of development of the
theory of metric spaces. In both, the abstract space from which
we started was given greater specialization. Another line of
development aims at freeing the subject from the idea of dis-
tance altogether.
In Chapter 3 we defined open and closed sets, which depended
on the metric used. In some of the subsequent work, we could
have dispensed with distance and used only closed and open sots.
For example, we could have defined the continuity of a mapping
f: -+ Y by the condition that the inverse image evcry open
subset of Y is open.
Let X be an abstract space, and lot (f) be a family of subsets
of X with the following properties:
(i) The empty set 0 and X itself are in (9.
The intersection of any two members of is (f).
(iii) The union of any number of members of (f) is in (f).
The family (!) is called a topology for X. The set X with a topo-
is called a topological space; and tho members of the family
(f) are called the open sets of the topological space. General
topology is the study of the propertics of topological spaces.
A metric space is a topological space, since the open sets of a
metric space possess the above properties. A topological space is
said to be metrizable if a metric can defined on it such that the
open sets determined by the metric are precisely the open sets
which constitutothe topology. General topology is generaliza-
tion of the theory of metric spaces since there are topological
spaces which are not metrizable.
accumulation, point
adherent point, 34
Arzela's theorem,
Harre's category theorem, 50
Banach space, 137
Bolzano-Weierstrass property, 74
boundary, 37
bounded metric space, 22
Cantor's intersection theorem, 49
Cartesian product space, 11, 19, 30,
44, 55, 61, 80, 93, 94
category theorem, 50
Cauchy se<luelnCIS,
Cauchy's m'3Qllla:IIL,V,
Cauchy's principle "nn"""""",,,,
47
closed sets, 35,
closed sphere, 32,
closure, 35, 42,
compactness, 73,
complement, 6,
complete metric
completion of a metric space, 51
connected sets, 63, 93, 94, 95
connected components, 67
continuity, 89
contraction mappings, 112
convergent sequences, 1, 16, 46, 48
countable sets, 4, 9
covering, open, 73
dense subsets, 37
diameter of a subsot,
differential equations,
Dini's theorem,
distance between
distance of a point
disjoint sets, 7,
empty set, 4
equations, numerical
equivalence relations,
equivalent metrics, 43, 81, 91
extension theorems, 98, 106
exterior point, 36
everywhere dense property, 37
HIJIIlelillr. 37
functions, 18, 85
UL"<';'llUU spaces, 3, 57, 81, 88
Heine-Borel property, 72
Hilbert space, 139
Holder's inequality, 26
homeomorphisms, 91
implicit function theorem, 133
induced metric, 24
infimum, 14
integral equations, 123, 126, 128, 129
interior point, 32
of a set, 32, 36
intersection, 7, 8, 33, 40
InT,"l""UR, 68
89
mappings, 18, 85
metric space, 21
Minkows!ri's inequality, 26
norm, 138
nowhere dense property, 37
open sets, 32
real functions, 104
real numbers, 12, 68, 81
separable space, 38, 77
sets, 62
tie'CjUl3Il(le spaces, 28, 56, 87, 118
sequentially compact spaces, 74
sphores, 31, 32, 39
Bubspaces, 24, 40
supremum, 14
topology, 142
bounded spaces, 75
unbounded metric space, 22
uniform continuity, 96, 100
uniform convergence, 2, 102
nnion, 7, 8, 33, 40, 65
finite intersection property, 79
fixed point theorems, 113, 131 vector spaces, 137
[143 ]