Académique Documents
Professionnel Documents
Culture Documents
.B C !
.B B !
.B C .C B
.B D .B C .C D
( symmetry)
( the triangle inequality)
If a pseudometric space \ . is not a metric space it is because there are at least two points B C
for which .B C ! In most situations this doesn't happen; metrics come up in mathematics more
often than pseudometrics. However pseudometrics do occasionally arise in a natural way. Moreover,
many definitions and theorems actually only require using properties 1)-4). Therefore we will state our
results in terms of pseudometrics rather than metrics in situations where 5) is irrelevant.
61
Example 2.2
1) The usual metric on is .B C lB Cl Clearly, properties 1) - 5) are true. In fact,
1)-5) are chosen so that a metric imitates the usual distance function.
2) The usual metric on 8 is defined as follows: if B B" B# B8 and
3"
properties 1) - 5). But the details for verifying the triangle inequality are a little tricky, so we will go
through the steps. First, we prove another useful inequality.
Suppose E +" +# +8 and F ," ,# ,8 points in 8 Define
T A +3 A,3 # +3# # +3 ,3 A ,3# A#
8
3"
3"
3"
3"
which gives
3"
3"
3"
3"
8
3"
3"
B3 C3 # # B3 C3 C3 D3 C3 D3 #
8
3"
3"
3"
B3 C3 # # lB3 C3 llC3 D3 l C3 D3 #
8
3"
3"
3"
B3 C3 # # B3 C3 # "# B3 C3 # "# C3 D3 #
8
3"
3"
3"
3"
62
1) Let . be the usual metric on 8 and define . w B C "!!.B C. Then . w is also a metric
on In 8 . w , the usual distances are stretched by a factor of 100. It is as if we simply
changed the units of measurement from meters to centimeters and that change shouldn't matter in any
important way. In fact, it's easy to check that if . is any metric (or pseudometric) on a set \ and
!, then . w . is also a metric (or pseudometric) on \ .
8
3"
It is easy to check that .> satisfies properties 1) - 5) so 8 .> is a metric space. We call .> the taxicab
metric on 8 ( For 8 #, distances are measured as if you had to move along a rectangular grid of
city streets from B to C the taxicab cannot cut diagonally across a city block ).
3) If B B" B# B8 , C C" C# C8 are points in 8 define
. B C max lB3 C3 l 3 " # 8
Then 8 . is also a metric space. We will refer to . as the max metric on 8
When 8 " of course, . .> and . are exactly the same metric on .
We will see later that . w , .> . are all equivalent metrics on 8 for topological purposes.
Roughly, this means that whichever of these metrics is used in 8 exactly the same functions are
continuous and exactly the same sequences converge.
63
4) The unit sphere W " is the set of points in # that are at distance 1 from the origin. Sketch
the unit sphere in # using the metrics . .> . , and . w "!!.
Since there are only two coordinates, we will write a point in # in the usual way as B C
rather than B" B# .
For ., we get
W " B C . B C ! ! "
B C B# C# "
For . , we get
W " B C . B C ! ! "
B C maxlBl lCl "
Of course for the metric . w "!!., W " has the same shape as for the metric . , but the sphere is
reduced in size by a scaling factor of "!!.
Switching among the metrics . . w , .> . produces unit spheres in 8 with different sizes and shapes.
In other words, changing the metric on 8 may cause dramatic changes in the geometry of the space
for example, areas may change and spheres may no longer be round. Changing the metric can
also affect smoothness features of the space spheres may turn out to have sharp corners. But it turns
out, as mentioned earlier, that . . w , .> and . are equivalent for topological purposes. For
topology, size, geometrical shape, and smoothness don't matter.
64
When talking about 8 the usual metric . is the default that is, we always assume that 8 , or any
subset of 8 , has the usual metric . unless a different metric is explicitly stated.
Example 2.4 In each part, you should verify that . satisfies the properties of a pseudometric or metric.
1) For a set \ , define .B C ! for all B C \ We call . the trivial pseudometric on \ :
all distances are 0. (Under what circumstances is this . a metric?)
2) For a set \ , define .B C
! if B C
. We call . the discrete unit metric on \ .
" if B C
To verify the triangle inequality: for points B C D \ , .B D .B C .C D certainly is true if
B D ; and if B D , then .B D " and .B C .C D ".
Example 2.6
1) In , F% B! B! % B! %. More generally, F% B! in 8 is just the usual spherical
ball with radius % and center at B! (not including the boundary surface) If the metric .> is used in 8
then F% B! is the interior of a diamond-shaped region centered at B! . See the earlier sketches of
W " : in # ,.> , F" ! ! is the region inside the diamond-shaped W "
In \ ! " with the usual metric . , then F " ! ! "# , F" ! ! " F# ! ! "
#
B! if % "
. Therefore
if % "
\
every point B! in \ . is isolated. The same is true if we rescale and replace . by the metric .
where !.
4) Let G! " 0 !" 0 is continuous. For 0 , 1 G! " define
.0 1 ! l0 B 1Bl .B (*)
"
It is easy to check that . is a pseudometric on G! ". In fact . is a metric: if 0 1, then there must
be a point B! ! " where l0 B! 1B! l !. By continuity, l0 B 1Bl ! for B's near B! ,
that is, l0 B 1Bl ! on some interval + , ! " where B! + , (carefully explain why! ).
Let 7 minB+, l0 B 1Bl (why does 7 exist?). Then 7 !, so
.0 1 ! l0 B 1Bl .B + l0 B 1Bl .B + 7 .B 7, + !
"
65
G! " is a subset of the larger set ] 0 !" 0 is integrable. We can define a distance
function . on ] the same formula (*). In this case, . is a pseudometric on ] but not a metric. If
0 B ! for all B, and 1B
!
"
.0 1 ! l0 B 1Bl .B !
if B
if B
"
#
"
#
, then 0 1 but
"
This example shows how a pseudometric that is not a metric can arise naturally in analysis.
5) On G! " we can also define another metric . by
. 0 1 sup |0 B 1Bl B ! "
max |0 B 1Bl B ! "
(Replacing sup with max makes sense because a theorem from analysis says that the continuous
function l0 1l has a maximum value on the closed interval ! ".)
Then . 0 1 % if and only if l0 B 1Bl % at every point B ! ", so we can picture F% 0 in
G! " . as the set of all functions 1 G! " whose graph lies entirely inside a tube of width
% containing the graph of 0 that is, 1 F% 0 iff 1 is uniformly within % of 0 on ! ". See the
following figure.
How are the metrics . and . from Examples 4) and 5) related? Notice that for 0 1 G! ":
.0 1 ! l0 B 1Bl .B ! maxB!" l0 B 1Bl .B ! . 0 1 .B . 0 1
"
"
"
We abbreviate this observation by writing . . It follows that F%. 0 F%. 0 : so, for a given
% ! the larger metric produces the smaller ball. (Note: the superscript notation on the balls
indicates which metric is being used in each case.)
The following below shows a function 0 and the graph of a function 1 F%. 0 F%. 0 . The graph
of 1 coincides with the graph of 0 , except for a tall spike: the spike takes the graph of 1 outside the %"
tube around the graph of 0 , but the spike is so thin that the .0 1 ! l0 B 1Bl .B
the total area between the graphs of 0 and 1 %.
66
sequence notation, then j# B5 B5 and B#5 converges. Thus, j# is the set of all square5"
5"
5"
5"
5"
5"
5"
5"
5"
+# B#5 #l+ll,l B#5 "# C5# "# ,# C5 (by the Cauchy-Schwarz inequality)
8
5"
5"
5"
5"
+# B#5 #l+ll,l B#5 "# C5# "# ,# C5 Q (all the series converge
5"
5"
5"
5"
because B C j#
Therefore the nonnegative series +B5 ,C5 # converges because it has bounded partial
5"
5"
# "#
5"
5"
5"
67
3"
. B C max .3 B3 C3 3 " 8
If each .3 is a metric, then so are . .> and . Notice that if each \3 and each .3 is the usual
metric on , then . .> and . are just the usual metric, the taxicab metric, and the max metric on 8
As we shall see, it turns out that these metrics on \ are all equivalent for topological purposes.
Definition 2.7 Suppose \ . is a pseudometric space and S \ . We say that S is open in \ .
if for each B S there is an % ! such that F% B S. (Of course, % may depend on B.)
For example,
i) The sets g and \ are open in any space \ .
ii) The intervals + ,, + , and are open in .
(Fortunately this terminology is consistent with the fact that these intervals were called open
intervals in calculus books. )
But notice that the interval + ,, when viewed as a subset of the B-axis in # , is not
#
open in . Similarly, is an open set in , but (viewed as the B-axis) is not open in # .
iii) The intervals + , + , and + , are not open in But the sets + , and + ,
are open in the metric space + , ..
Examples ii) and iii) illustrate that open is not an a property that depends just on the set E: whether
or not a set E is open depends on the larger space in which it is contained that is, open is a relative
term.
The next theorem tells us that the balls in \ . are the building blocks from which all open sets can
be constructed.
Theorem 2.8 A set S \ is open in \ . if and only if S is a union of a collection of balls.
Proof If S is open, then for each B S there is an %B ! such that F%B B S and
S BS F%B B
Conversely, suppose S BG F%B B for some indexing set G S We must show that if
C S, then F% C S for some % ! Since C S we know that C F%B! B ! for some B! G
Then .B! C $ %B! . Let % "# %B! $ ! and consider F% C If D F% C, then
.D B! .D C .C B! % $ "# %B! $ $ "# %B! "# $ "# %B! "# %B! %B!
so D F%B! B! Therefore F% C F%B! B! S.
Corollary 2.9 a) Each ball F% B is open in \ ..
b) A point B! in a pseudometric space \ . is isolated iff B! is an open set.
68
Theorem 2.11
(Conditions ii) and iii) say that the collection g. is closed under unions and closed under finite
intersections. )
Proof g is the union of the empty collection of open balls, and \ B\ F" B, so g \ g. .
69
Therefore all three metrics produce the same topology: g. g.> g. even though the balls are
different for each metric. It turns out that the open sets in \ . are the most important objects from a
topological point of view, so in that sense these metrics are all equivalent. (As mentioned above,
these metrics do change the shape and smoothness of the balls and therefore these metrics are not
equivalent for geometric purposes.)
Definition 2.13 Suppose . and . w are two pseudometrics (or metrics) on a set \ . We say that . and
. w are equivalent (written . . w if g. g. w , that is, if . and .w generate the same collection of open
sets.
Example 2.14
1) If . is the discrete unit metric on \ , then each singleton set B F" B is a ball, so each
B is open equivalently, every point B is isolated in \ . If E \ , then E +E + is open
because E is a union of balls. Therefore g. c \, called the discrete topology on \
If . w B C
"(
!
if B C
on a set \ , then . w ., where . is the discrete unit metric.
if B C
More generally, . . for any ! , and all of these metrics generate the discrete topology.
2) Let \ 8" 8 . Let . be the usual metric on \ and let . w be the discrete unit
metric on \ For each 8, F%. 8" 8" if we choose a sufficiently small %. Therefore, just as in part
1), every subset of \ is open in \ . But every subset in \ . w also is open, so . . w (even
though . and . w are not constant multiples of each other).
3) If . is the trivial pseudometric on a set \ , are there any other pseudometrics . w on \
for which . w . ?
3. The topology of
What do the open sets in look like? Since an % -ball in is an interval of the form (+ % + % , the
open sets are precisely the sets which are unions of open intervals. But we can say more to make the
situation even clearer. We begin by making precise the definition of interval.
70
Theorem 3.2 M is an interval iff M has one of the following forms (where + ,):
( , ( +, ( +, +,, +,, +,b), + , + ,, + , +, g
(*)
Proof It is clear that each of the sets in the list is convex and therefore is an interval.
Conversely, we need to show that every interval M has one of these forms. Clearly, if lMl "
then M g or M + If lMl # then the definition of interval implies that M must be infinite.
The remainder of the proof uses the completeness property ( least upper bound property
of , and the argument falls into several cases:
Case I: M is bounded both above and below. Then M has a least upper bound and a
greatest lower bound: let + inf M and , sup M Of course + and , may or may not be in M .
a) if + , M ,
b) if + M but , M ,
c) if + M but , M ,
d) if + , M ,
we claim M
we claim M
we claim M
we claim M
+ ,
+ ,
+ ,
+ ,
Case II: M is bounded below but not above. In this case, let + inf M
a) if + M ,
we claim M +
b) if + M
we claim M +
a) if , M ,
Case III: M is bounded above but not below. In this case, let , sup M
we claim M ,
b) if , M ,
we claim M ,
Case IV: M is not bounded above or below. In this case, we claim M
The proof is similar in each case, using properties of sups and infs. To illustrate, We prove case Ic):
If B M then B sup M , Also, B inf M + and because + M we get
B +. So M + ,
We need to show that M + ,, so suppose B + ,. Then B + inf M
so B is not a lower bound for M . This means that there is a point D M such
that D B Then D B , where D , M , and M is convex, so B M
Therefore + , M , so M + ,
71
Note: We used the completeness property to prove Theorem 3.2. In fact, Theorem 3.2 is equivalent to
the completeness property. To see this
Assume Theorem 3.2 is true and that E is a nonempty subset of that has an upper
bound. Let M B B - for some - E Then M is an interval suppose
B C D where B D M . Then z - for some - E; therefore C - =9, by
definition of M , C M .
Since E g M must be infinite. An upper bound for E must also be an upper
bound for M . Since I is an interval with an upper bound, I must have one of the forms
( ,, ( , +,b), + , + ,, + ,, Then it's not hard to check that
E has a least upper bound, namely sup E ,
This is an observation I owe to Professor Robert McDowell
It is clear that an intersection of intervals in is an interval (why?). Also, a union of intervals need not
be an interval: for example ! " # $. But if every pair of intervals in a collection overlap, then
the union is an interval. The following theorem makes this precise.
Theorem 3.3 Suppose \ is a collection of intervals in . If M N g for all M N \ , then \ is
an interval. In particular, if \ g, then \ is an interval.
Open intervals in are open sets, and a union of any collection of open sets is open.
( ) Suppose S is open in . For each B S, there is an open interval (ball) M such that
B M S. Let KB M M is an open interval and B M S. Then B KB S.
By Theorem 3.3, KB is also an open interval (in fact, KB is the largest open interval containing B and
inside S; why? ). It is easy to see that there can be distinct points B C S for which KB KC . In fact,
we claim that if B C S , then either KB KC or KB KC g.
If KB KC g, then there is a point D KB KC . By Theorem 3.3, KB KC is
an open interval, a subset of S , and containing both B and C. Therefore KB KC
is a set in the collection whose union is KB . Therefore KB KC K B . Similarly,
KB KC KC , so KB KC
72
Removing any repetitions, we let W be the collection of the distinct intervals KB that arise in this way.
Clearly, S W and W is countable because the members of W are pairwise disjoint: for each
M W, we can pick a rational number ;M M , and these ;M 's are distinct. (More formally, the function
0 W given by 0 M ;M is one-to-one.)
We are now able to find the number of open sets in .
Corollary 3.5 There are exactly c open sets in .
Proof Let g. be the usual topology on . We want to prove that lg. l -
For each < , the interval < g. , so l g. l -
Let \ be the set of all open intervals in . Then l \ l - (why?. For each S g. , pick a sequence
M" M# M8 \ for which S
8" M8 . (We could also choose the M8 's to be pairwise disjoint,
but that is unnecessary in this argument the important thing here is that there are only countably
many M8 's.) Then we have a function 0 g. \ given by 0 S M" M# M8 . The function
0 is clearly one-to-one, so l g l l \ l - i! -
73
Exercises
E1
The following statements refer to a metric space \ .. Prove the true statements and give
counterexamples for the false ones. (The statements illustrate the danger of assuming that familiar
features of 8 necessarily carry over to arbitrary pseudometric spaces. )
a) F% B F% C implies B C (i.e., a ball can't have two centers)
b) The diameter of F% B must be bigger than %. The diameter of a set A in a metric
space is defined to be sup .B C B C E .
E2
The taxicab metric on # is defined by d> B" C" B# C# |B" B# | |C" C# |. Draw
the set of points in # , d> that are equidistant from ! ! and $ %
E3.
Hint: Let . be a metric on \ and suppose 0 is a function from the nonnegative real numbers
to the nonnegative real numbers for which:
0 ! ! B C 0 B 0 C, and
0 B C 0 B 0 C for all nonnegative B C Prove that .w B C 0 .B C is also a metric on
\ Then consider the particular function 0 B " B B
Note: For all B C in \ , d (B C " and . B C " that is, d and . are bounded metrics on
\ . Thus any metric d on \ can be replaced by an equivalent bounded metric that is, a bounded
metric generating the same topology. So boundedness is a property determined by the particular
metric, not the topology.
E4.
Suppose 0 . Let . be the usual metric on , and . w the usual metric on # . Define a
new distance function . ww on by . ww B C . w B 0 B C 0 C. Prove that . ww is a metric on
.
Must . ww be equivalent to . ? If not, can you describe conditions which will guarantee that
ww
. .?
E5.
Suppose a function . \ \ satisfies the conditions 1), 2), 3), and 5) in the definition
of a metric, but that instead of the triangle inequality, we have that for all B C D \
.B D .B C .C D
Prove that l\l "
E6.
Suppose E is a finite open set in a metric space \ .. Prove that every point of E is isolated
in \ .
E7.
Suppose \ . is a metric space and B \ Prove that the following two statements are
equivalent:
i) B is not an isolated point of \
ii) every open set containing B contains an infinite number of points.
74
E8.
The definition of an open set in \ . reads: S is open if for all B S there is an % ! such
that F% B S. In this definition, % may depend on B.
Suppose we define S to be uniformly open if there is an % ! such that for all B S,
F% B S that is, the same % works for every B S Uniformly open is not a standard term.
a) What are the uniformly open subsets of 8 ?
b) What are the uniformly open sets in \ . if . is the trivial pseudometric?
c) What are the uniformly open sets in \ . if . is the discrete unit metric?
E9.
Let p be a fixed prime number. We define the :-adic absolute value | |: (sometimes called
the :-adic norm) on the set of rational numbers as follows:
k
If 0 B , write B pnm for integers k,m,n, where p does not divide m or n, and define
|B|: :5 :"5 . (Of course, 5 may be negative. ). Also, define |!|: !
Prove that l l: behaves the way an absolute value (norm) should that is, for all B C
a) l Bl: ! and l Bl: ! iff B !
b) l BCl: l Bl: l Cl:
c) l B Cl: l Bl: l Cl:
l l: actually satisfies a stronger inequality than the inequality in part c). Prove that
d) l B Cl: max l Bl: l Cl: lBl: l Cl:
Whenever we have an absolute value (norm), we can use it to define a distance function:
for B C let .: B C lB Cl:
e) Prove that .: is a metric on , and show that .: actually satisfies an inequality
stronger than the usual triangle inequality, namely:
for all B C D , .: B D max.: B C .: C D
f) Give a specific example for B C D : for which
.: B D max.: B C .: C D
(Hint: It might be convenient to be able to refer to the exponent 5 associated with a particular B. If
k
B pnm , then 5 roughly refers to the number of :'s that can be factored out of B so we can call
5 / B Prove that / (a b) min / + / , whenever a,b with a,b 0 and a b. Note
that strict inequality can occur here: for example, when p 3, / (8) / (2) 0, but
/ (8 2) / (6) 1. )
g) Suppose : # Calculate .# #8 ! What are lim .# #8 ! and lim .# %8 !?
8
75
J is closed in \ .
\ J is open in \ .
for all B \ J , there is an % 0 for which F% B \ J
for all B \ J , there is an % 0 for which F% (x) J g.
The closed sets are completely determined by the open sets and vice-versa so that in any space \ ,.
the collection of closed sets and the collection of open sets, g. , contain exactly the same information.
The close connection between the closed sets and the open sets is illustrated in the following theorem.
Theorem 4.2
(Conditions ii) and iii) say that the collection of closed sets is closed under intersections and finite
unions.)
Proof If we take complements, then these statements follow from the corresponding properties of
open sets. Since g and \ are open, the complements \ g \ and \ \ g are closed.
Suppose J is closed for each E. Then \ J is open for each E, so
E \ J is open, and therefore the complement \ E \ J
E \ \ J E J is closed.
The proof of iii) uses the fact that a finite intersection of open sets is open.
Exercise: Give an example to show that an infinite union of closed sets need not be closed.
Example 4.3
1) In , the interval ! " is closed since its complement ! " ! " is
open. Equivalently, we can say that ! " is closed because for each B ! " there
is an % ! for which F% B ! " B % B % ! " g
2) A set can be neither open nor closed: for example, consider the following subset of :
! ", and .
76
3) A set can be both open and closed these terms are not mutually exclusive. Such sets in
\ . are called clopen sets. For example, g and \ are clopen in every pseudometric space \ ..
Sometimes there are other clopen sets and sometimes not.
In the space \ ! " $ % with the usual metric ., the set ! " is clopen. But in , for
example, g and are the only clopen subsets (This fact is not too hard to prove but it is also not
obvious the proof depends on the completeness property least upper bound property
in . We will prove this fact later when we need it in Chapter V.)
4) In any pseudometric space \ ., the set B \ .+ B % J is a closed set.
To see this, suppose C J , Then .+ C $ % Let %" $ % !. Then F%" C J g
(If D F%" C J , then we would have .+ C .+ D .D C % %" % $ % $ ,
which is false.)
B \ .+ B % is called the closed ball centered at B with radius %.
For example, ! " B . "# B lB "# l "# is the closed ball centered at "#
5) Let \ ! " # & with the usual metric . .
! " and # & are both clopen in \ ..
$ % is neither open nor closed in \ .
Notice again that open and closed are not absolute terms: whether a set E is open (or closed)
depends on what larger space \ that E lives in.
6) Let . be the discrete unit metric, so that g. is the discrete topology: every subset of \ is
open. Then every subset of \ is clopen.
7) Let . be the trivial pseudometric on \ Then F% B \ for every B \ and every % !.
Therefore a union of balls must be either \ or g (in the case of the union of an empty collection of
balls), so g. g \, which is called the trivial topology on \ . Since g and \ must be open in any
space \ ., the trivial topology is the smallest possible topology on \ .
In \ . the only closed sets are g and \ .
Using the open and closed sets in \ ., we can define some useful operators on subsets of \ An
operator creates a new subset of \ from an old one.
Definition 4.4 Suppose \ . is a pseudometric space and E \
We will omit the subscript \ when the context makes clear the _space \ in which the operations are
being performed. Sometimes int E and cl E are denoted E and E respectively. Some books use the
77
notation `E for Fr E, but the symbol `E has a different meaning in algebraic topology so we will
avoid using it here.
Theorem 4.5
1) a) int E is the largest open subset of E (that is, if O is open and O E, then
O int E ).
b) E is open iff E int E (since int E E, the equality is equivalent to E int E.
c) B int E iff there is an open set S such that B S E
iff b % 0 such that F% B E.
Informally, we can think of 1c) as saying that the interior of E consist of those
points comfortably inside E surrounded by a small cushion, that is, points not
on the edge of E.
2) a) cl E is the smallest closed set containing E (that is, if J is closed and J E,
then J cl E)
b) E is closed iff E cl E (since E cl E the equality is equivalent
to cl E E)
c) B cl E iff for every open set S containing B, S E g
iff for every % !, F% B E g.
Informally, 2c) states that -6E consists of the points in \ that can be approximated
arbitrarily closely by points from within the set E
3) a) Fr E is closed and Fr E Fr \ E.
b E is clopen iff Fr E g.
c) B Fr E iff for every open set S containing B, S E g and S \ E g
iff for every % ! F% B E g and F% B \ E g.
Informally, 3c) states that Fr E consists of those points in \ that can be
approximated arbitrarily closely both by points from within E and by points from
outside E.
Proof
1) a) Clearly int E E. If S open and S E, then S is one of the sets whose union is
int E, so S int E.
b) int E is open (it is a union of open sets), so if int E E, then E is open.
Conversely, if E is open, then E is the largest open subset of E, so E int E
c) Since int E is a union of open sets, it is clear that B int E iff B S E for some
open set S Since a open set is open iff it is a union of %-balls, the remainder of the
assertion is obviously true.
2) Exercise
78
int
int g
cl
cl
Fr g
Fr
cl ! " ! "
Fr Fr Fr g
In any space \ ., it is obviously true that intint E int E and clcl E cl E. But this need not be
true for Fr, as the last example shows. ( It is true that FrFrFr E FrFr E in any space \ But
this is not a useful fact, and it is also not interesting to prove.)
2) \ ! # (with the usual metric)
cl\ ! " ! "
cl\ " # " #
79
3"
3R "
%
#R
%#
# .
for 3 " R
.+ C +3 C3 # +3 C3 # +3 C3 #
3"
3"
3R "
+3 C3 # C3# R
R
3"
3R "
%#
#R
%#
#
%# %
Therefore + F% C E
We also claim that int E g To prove this, we need to show that if B B3 E
then no ball centered at B is a subset of E. To see this, pick any % ! and choose
an irrational C" such that lC" B" l % Modify B by changing B" : define
C C" B# B8 Then .B C lB" C" l % and C E so
F% B
E.
What is Fr E ?
5) In any pseudometric space \ ., F% + is a subset of the closed set
B \ .+ B %. Therefore cl F% + B \ .+ B %
But these two sets are not necessarily equal: sometimes the closed ball is larger
than the closure of the open ball F% + ! For example, suppose . is the discrete unit
metric on a set \ where l\l " Then
+ F" + cl F" + \ B \ .+ B "
Definition 4.7 Let \ . be a pseudometric space and suppose that H \ We say that H is dense
in \ . if cl H X. The space \ . is called separable if it possible to find a countable dense set
H in X. (Note the spelling: separable, not seperable. )
Since separable is defined in terms of the closure and the closure operator depends only on the
topology, not the particular metric that generates the topology. Therefore separability is a topological
property: if \ . is separable and . w . , then \ . w is also separable.
More informally, H is dense in \ means that each point B \ can be approximated arbitrarily
closely by a point from H. A countable dense set in \ (if one exists) is a small set which can be
used to approximate any point in \ arbitrarily closely.
80
Example 4.8
1) 8 is separable because 8 is a countable dense set in 8 in particular, is a
countable dense set in , so is separable. is an example of an uncountable dense subset of .
2) Any countable space \ . is separable, because \ is dense in \ .
3) Suppose g. is the discrete topology on \ . Then \ . is separable iff \ is
countable (because any proper subset of \ is closed and therefore not dense).
If g. is the trivial topology and \ g, then every nonempty subset H is dense
(why?). Therefore \ . is separable because, for example, each singleton set B is dense.
4) The set E j# is dense in j# ( see Example 4.6(4) ). This set E is an
uncountable dense set because every sequence of rationals B3 with lB3 l "3 is in E (why?), and there
are - such sequences (why?). However, j# . is separable. Can you find a countable dense set H?
The computation in Example 4.6.4 might give you an idea.
Definition 4.9 For nonempty subsets E F in a pseudometric space \ ., we define the distance
between them by
distE F inf .+ , + E and , F
Although it is an abuse of notation . we usually abbreviate distE F by .E F Going one step
further, we also write .+ F as an abbreviation for .+ F
Of course, if + F , then .+ F ! But the converse may not be true. For example, let
F " "# 8" Then .! F ! even though ! F Similarly, if E is the C-axis in #
and F B C # C B" , then E and F are disjoint, closed sets but .E F !
The distance from a point to a set can be used to describe the closure of a set.
Theorem 4.10
.B E !
81
Then B cl E iff
Exercises
E10.
E11.
E12. An infinite union of closed sets need not be closed. However if infinitely many closed sets are
spread out enough from each other, their union is closed. Parts a) and b) illustrate this.
a) Suppose that for each 8 J8 is a closed set in and that J8 8 8 ". Prove that
8" J8 is closed in .
b) More generally: suppose for E, each J is a closed set in \ . and that for each
point B \ there is an % ! such that F% B has nonempty intersection with at
most finitely many J 's. Prove that E J is closed in \ .. (Notice that b a. Why?)
E13.
a) Give an example of #- subsets of all of which have the same closure Do the same in # .
b) Prove or disprove: there exist #- subsets of # such that any two have different closures.
Can you do the same in ?
82
E16. Let \ . be a pseudometric space. Suppose that for every % !, there exists a countable
subset D% of X with the following property: aB \ bC H% such that .B C %. Prove that
\ . is separable.
E17. Suppose that \ is an uncountable set and . is any metric on \ which produces the discrete
topology. (Such a metric d does not have to be a constant multiple of the discrete unit
metric: see Example 2.14.2). Show that for some % ! there is an uncountable subset E of \ such
that .B C % for all B C E
E18. Let \ . be an infinite metric space. Prove that there exists an open set Y such that both Y
and \ Y are infinite.
(Hint: Consider a non-isolated point, if one exists.)
E19. A metric space \ . is called extremally disconnected if the closure of every open set is
open. ( Note: extremally is the correct spelling; the word is not the same as the everyday word
extremely. )
Prove that if \ . is extremally disconnected, then the topology g. is the discrete topology.
83
E20. Since dist provides a measure of distance between nonempty subsets of \ ., one might
ask whether c \ g dist is a metric (or pseudometric) space. Is it?
84
5. Continuity
Suppose + E and that 0 E In elementary calculus, the set E is usually and interval,
and the idea of continuity at a point + in E is introduced very informally. Roughly, it means that if B is
a point near + in the domain, then 0 B is near 0 + In advanced calculus or analysis, the idea of
continuity of 0 at + is defined carefully. The intuitive version of continuity stated in terms of
nearness is made precise by measuring distances:
0 is continuous at + means: a % ! b $ ! such that
if B E and lB +l $ , then l0 B 0 +l %
We say that 0 is continuous if 0 is continuous at every point + in its domain E.
An important thing to notice is that the definition is made using the distance function in :
.B + lB +l and .0 B 0 + l0 B 0 +l. Since we have a way to measure distances in
pseudometric spaces, we can make an a completely similar definition of continuity for functions from
one pseudometric space to another.
Definition 5.1 Suppose 0 \ ] where \ . and ] = are pseudometric spaces If + \ and
0 \ ] , we say 0 is continuous at a if a% ! b$ ! such that: if .+ B $ , then
=0 B 0 + %
Notice that the sets \ and ] may have two completely unrelated metrics . and =: . measures
distances in dom0 and = measures distances in ] . But the idea is exactly the same as in calculus:
continuity of 0 at + means, roughly, that points B near + in the domain \ have images 0 B near
0 + in ] .
Theorem 5.2 Suppose \ . and ] = are pseudometric spaces If + \ and 0 \ ] , then the
following statements are equivalent:
1)
2)
3)
4)
0 is continuous at a
a% ! b$ ! such that 0 F$ + F% 0 +
a% ! b$ ! F$ + 0 " F% 0 +
aR ] : if 0 (a) int R , then a int 0 " [ N ].
Proof It is clear that conditions 1) - #) - 3) are just equivalent restatements the definition of continuity
at + phrased in terms of images and inverse images of balls. Condition 4), however, seems a bit
strange. We will show that 3) and 4) are equivalent.
3) 4) Suppose 0 + int R By definition of interior, there is an % ! such that
F% 0 + R , so that 0 " F% 0 + 0 " R By 3), we can pick $ ! so that
+ F$ + 0 " F% 0 + 0 " R . Since the $ -ball at + is an open subset of 0 " R , we get that
+ int 0 " R , as desired.
4) 3) Suppose % ! is given. Let R F% 0 + Then R is open and 0 + R
int R We conclude from 4) that + int 0 " R int 0 " F% 0 + Therefore for some $ !
F$ + int 0 " F% 0 + 0 " F% 0 +, so 3) holds.
85
With this new terminology, we can restate the equivalence of 1) and 4) in Theorem 5.2 as:
0 is continuous at + iff
whenever R is a neighborhood of 0 + in ] , then 0 " R is a neighborhood of + in \ .
This tells us something very important. Interiors (and therefore neighborhoods of points) are defined in
terms of the open sets (without needing to mention the distance function), so the neighborhoods of a
point B depend only on the topology, not on the specific metric that generates the topology. Therefore
whether or not 0 is continuous at + does not actually depend on the specific metrics but only on the
topologies in the domain and range. In other words, continuity at + is a topological property.
For example, the function sin is continuous at each point + in , and this is remains true if
we measure distances in the domain with, say, the taxicab metric .> and distance in the range with the
max-metric . since these are both equivalent to the usual metric . on .
We now define 0 is a continuous function in the usual way.
Definition 5.5 Suppose \ . and ] = are pseudometric spaces. We say that 0 \ ] is
continuous if 0 is continuous at each point of \ .
Theorem 5.6 Suppose 0 \ ] where \ . and ] = are pseudometric spaces.
The following are equivalent:
86
1) 0 is continuous
2) if S is open in ] , then 0 " S is open in \
3) if J is closed in ] , then 0 " J is closed in \ .
Proof 1) 2) Suppose S is open in ] and that B 0 " S. Since S is a neighborhood of 0 B and
0 is continuous at B, we know from Theorem 5.2 that 0 " S is a neighborhood of B. Therefore
0 " S is a neighborhood of each of its points, so 0 " S is open.
2) 3) If J is closed in ] , then ] J is open. By 2), 0 " ] J \ 0 " J is open
in \ so \ \ 0 " J 0 " J is closed in \ .
3) 2) Exercise
2) 1) Suppose + \ and that R is a neighborhood of 0 + in ] , so that
0 + int R R By 2), 0 " int R is open in \ , and + 0 " int R 0 " R . Therefore
0 " R is a neighborhood of + Therefore 0 is continuous at +. Since + was an arbitrary point in \ , 0
is continuous.
Notice again that Theorem 5.6 shows that continuity is completely described in terms of the open sets
(or equivalently, the closed sets), and the proof of the theorem is phrased entirely in terms of open
(closed) sets, without any explicit mention of the pseudometrics on \ and ] . Replacing . and = with
equivalent pseudometrics would not affect the continuity of 0 .
Theorem 5.7 Suppose \ ., ] = and ^ > are pseudometric spaces and that 0 \ ] and
1 ] ^ If 0 is continuous at a \ and g is continuous at 0 (a) ] , then 1 0 is continuous at
a. (Therefore, if 0 and 1 are continuous, so is 1 0 )
Proof If R is a neighborhood of 10 +, then 1" R is a neighborhood of 0 +, because 1 is
continuous at 0 +. Since 0 is continuous at +, 0 " 1" R is a neighborhood of +. But
0 " 1" R 1 0 " R so 1 0 is continuous at +.
Example 5.8
1) Suppose 0 is given by 0 B
" if B !
Then R ( "# $# is a
! if B !
neighborhood of 0 ! " but 0 " R ! is not a neighborhood of ! Therefore
0 is not continuous at 0. To see the same thing using slightly different language:
0 is not continuous at ! because, choosing % "# there is no choice of $ ! such
that 0 F$ ! F% 0 ! F% "
87
" if B
. Then 0 is not continuous
! if B
at any point + (why?). However 0 l 1 is continuous at every point
of , because 1 is a constant function.
There is a curious old result called Blumberg's Theorem which states:
For any 0 , there exists a dense set H such that
0 lH 1 H is continuous.
Blumberg's Theorem is rather difficult to prove, and not very useful.
4) Suppose 0 1 \ where \ . is a pseudometric space. Since these
functions are real-valued, it makes sense to define functions 0 1 0 1 0 1 and
0
1 in the obvious way. For example, 0 1B 0 B 1B, where the on
the right is ordinary addition in .
If 0 and 1 are continuous at a point + \ then the functions 0 1 0 1 and
0 1 are also continuous at +; and 01 is continuous at + if 1+ 0. The proofs are
just like those given in calculus where \
For example, consider 0 1: given % !, then (because 0 1 are continuous at +)
we can find $" ! and $# ! so that
if .B + $" then l0 B 0 +l #% and
if .B + $# then l1B 1+l #%
Let $ min $" $# Then if .B + $ , we have
l0 1B 0 1+l l0 B 0 + 1B 1+l
l0 B 0 +| |1B 1+l #% #% %
You can find at the other proofs in an analysis textbook.
3"
so 0 F$ B F% 0 B.
A similar argument shows that the projection 1B B8 is also continuous,
and an argument only slightly more complicated would show, for example, that the
projection function 2 j# $ given by 2B B$ B* B"" is continuous.
88
Example 5.10
1) Suppose 0 \ . ] =, where . is the trivial pseudometric on \ and ] is
any metric space. We already know that if 0 is constant, then 0 is continuous.
If 0 is not constant, then there are points + , \ for which 0 + 0 ,
Since = is a metric, we can pick disjoint open sets Y and Z in ] with
g (+ 0 " Y )
0 + Y and 0 , Z Then 0 " Y
\ (, 0 " Y )
"
Since g. g \, we see that 0 Y is not open so 0 is not continuous.
So, in this situation: 0 is continuous iff 0 is constant.
2) Suppose . is the usual metric and = is the discrete unit metric on . Let
3 . = is the identity map 3B B For every open set S in . the
image set 3S is open in =, but this is not the criterion for continuity: in fact, 0 is
not continuous at any point. The criterion for continuity is that the inverse image of
every open set must be open.
Example 5.10.2 leads us to a definition.
89
8 terms
90
"
"
"
"
7
07
8 0 8 8 70 8 7 - 8 - 8
7 terms
So far, we have shown that a function 0 satisfying must have the formula 0 B -B for every
positive rational B 7
8
Since
! 0 ! 0 7
8
0
7
8
7
8
7
8
7
07
8 8 - -
07
8 we get that
7
8
In calculus, another important is the idea of a convergent sequence (in or 8 ). We can generalize
the idea of a convergent sequence in 8 to any pseudometric space.
Definition 5.15 A sequence B8 in \ . converges to B \ if any one of the following (clearly
equivalent) conditions holds:
1) a% ! bR such that if n N, then B8 F% B
(that is, if the sequence of numbers .B8 B ! in )
2) if S is open and B S X, then bR such that if n N, then B8 S
3) if W is a neighborhood of B, then bR such that if n N, then B8 W.
If B8 converges to \ , we write B8 B
Condition= 2) and 3) describe the convergence of sequences in terms of open sets (or neighborhoods)
rather than directly using distances. Therefore replacing . with an equivalent metric . w does not affect
which sequences converge to which points: sequential convergence is a topological property.
If a sequence B8 has a certain property T for all 8 some R , we say that B8 eventually has
property T . For example, the sequence (0,3,1,7,7,7,7,...) is eventually constant; the sequence
" $ & # & ' ( ) 8 8 " is eventually increasing. Using this terminology, we can
give a completely precise definition of convergence by saying: B8 converges to B if B8 is
eventually in each neighborhood of B.
91
Theorem 5.18 tells us something very important about the role of sequences in pseudometric spaces.
The set E is closed iff E cl E. But E cl E is always true, so we can say E is closed iff
cl E E. But that is true iff the limits of convergent sequences +8 from E must also be in E.
Therefore a complete knowledge about what sequences converge to what points in \ . would let
you determine which sets are closed (and therefore, by taking complements, which sets are open). In
other words, all the information about which sets in \ . are open or closed? is revealed by the
convergent sequences. We summarize this by saying that in a pseudometric space \ . sequences
are sufficient to describe the topology.
92
Sequences are sufficient to determine the topology in a pseudometric space, and continuity is
characterized in terms of open sets, so it should not be a surprise that sequences can be used to decide
whether or not a function 0 between pseudometric spaces is continuous.
93
Theorem 5.22 Suppose \ . and ] = are pseudometric spaces, and that 0 \ ] . Then
0 is continuous at + \ iff 0 B8 0 + for every sequence B8 +.
Proof Suppose 0 is continuous at + and consider any sequence B8 +. If [ is a neighborhood of
0 + then 0 " [ is a neighborhood of +, so B8 is eventually in 0 " [ . This implies that 0 B8
is eventually in [ , so 0 B8 0 +
Conversely, if 0 is not continuous at +, then a% ! b$ ! 0 F$ B F% 0 +,
that is, b% ! a$ ! 0 F$ B
F% 0 +.
"
For this % and $ 8 we have that 0 F " B
F% 0 +. So for each 8 we can choose a point
8
"
8
!),
but 0 B8
0 + (because =0 B8 0 + % for all 8). Therefore 0 is not continuous at +.
Note: the first half of the proof is phrased completely in terms of neighborhoods of B! and
0 B! that part of the proof is topological. However the second half makes explicit use
of the metric. In fact, as we shall see later, the second half of the proof must involve a little
more than just the open sets.
Notice also that the second half of the proof makes uses the Axiom of Choice (the function
B chooses B8 for each 8.
The following theorem and its corollaries are often technically useful. Moreover, they show us that a
pseudometric space \ . has lots of built-in continuous functions these functions can be defined
using pseudometric . .
Theorem 5.23
In a pseudometric space \ .:
if B8 B and C8 C then .B8 C8 .B C.
Proof Given % !, pick R large enough so that 8 R implies both .B8 B #% and .C8 C #%
Since .B8 C8 .B8 B .B C8 .B8 B .B C .C C8 , we get that
if 8 R ,
.B8 C8
%
#
.B C
%
#
.B C %
.B C
%
#
.B8 C8
%
#
.B8 C8 %
94
Proof In Theorem 5.23, let C8 be the constant sequence where C8 + for each 8.
Suppose + \ .. Define 0 \ by 0 B .B + Then 0 is continuous.
Corollary 5.25
Theorem 5.26 If E is a nonempty subset of the pseudometric space \ ., then the function
0 \ defined by 0 B .B E is continuous.
Proof We show that 0 is continuous at each point B! \ .
inequalities are true for any B \ :
Let + E.
.+ B! .+ B .B B!
.+ B .+ B! .B! B
Apply inf+E to each inequality to get
.E B! .E B .B B! or
.E B .E B! .B! B or
.E B! .E B .B B!
.E B .E B! .B B!
so for all B \ ,
l.E B .E B! | .B B! . In other words,
for all B \ , l0 B 0 B! | .B B!
95
Exercises
E22.
The order of the quantifiers is important. What functions are described by each of the following
modifications of the definition:
a) a $ ! b % ! such that if B E and lB +l $ , then l0 B 0 + %
b) a % ! a $ ! such that if B E and lB +l $ , then l0 B 0 + %
c) b % ! b $ ! such that if B E and lB +l $ , then l0 B 0 + %
In each case, what happens if the restriction ! is dropped on either % or $ ?
E23. Suppose 0 What does each of the following statements tell us about 0 ?
(In this exercise, interval means bounded open interval + ,; and 0 takes M into N
means that 0 M N )
a) For every interval M containing + and every interval N containing ,, 0 takes M into N .
b) There exists an interval N containing , and there exists an interval M containing + such that
0 takes M into N .
c) There exists an interval N containing , such that for every interval M containing +, 0 takes M
into N .
d) There exists an interval N containing , such for every interval M containing +, 0 does not
take M into N .
e) For every interval M containing +, there exists an interval N containing , such that 0 takes M
into N .
f) There exists an interval M containing + such that for every interval N containing ,, 0 takes M
into N
E24. (The Pasting Lemma, easy version) The two parts of the problem give conditions when a
collection of continuous functions defined on subsets of \ can be united ( pasted together) to
form a new continuous function. In \ . suppose the sets S E are open and that J" J8
8 are closed.
a) Suppose that functions 0 S ] = are continuous and that, if " , then
0 lS S" 0" lS S" (that is, 0 and 0" agree where their domains overlap).
Then E 0 0 E S ] is continuous.
b) Suppose that for each 3 " 8 03 J3 ] = is continuous and that, if 3 4, then
03 l(J3 J4 ) 04 lJ3 J4 (that is, 03 and 04 agree where their domains overlap).
Then 0 83" 03 83" J3 ] is a continuous function.
c) Give an example to show that b) may be false for an infinite collection of functions 03
3 defined on closed subsets of \ , even if the domains J3 are pairwise disjoint.
96
E26.
"
!
if B E
if B E
E28. Suppose B8 is a sequence in \ , that B is some point in \ and that 0 B8 0 B for every
0 G\. Prove that B8 B or give a counterexample to show the statement is false.
E29. Let . be the usual metric on . Find a metric . w on such that B8 ! with respect to . iff
B8 ! with respect to . w , but . w is not equivalent to . .
E30. a) Suppose E is a closed set in the pseudometric space \ . and that B! E. Prove that
there is a continuous function 0 \ ! " such that 0 lE ! and 0 B! " (Hint: Consider the
function distance to the set E.)
b) Suppose E and F are disjoint closed sets in \ .. Prove that there exists a continuous
.BE
function 0 \ such that 0 lE ! and 0 lF ". Hint: Consider .BE .BF
c) Using b) (or by some other method) prove that if E and F are disjoint closed sets in \ .,
then there exist open sets Y and Z such that E Y F Z and Y Z g Can Y and Z always be
chosen so that cl Y cl Z g ?
E31. E function 0 \ . (] ,=) is called an isometry between \ and ] if 0 is onto and, for all
B C \ , .B C =0 B 0 C. If such an 0 exists, we say that \ . and ] = are isometric to
each other. If 0 is not onto, we say 0 is an isometry of \ into ] , or that 0 is an isometric embedding
of \ . into ] =. Let and # have their usual metrics.
a) Prove that there is no isometry between and # .
b) Let + . Prove that there are exactly two isometries from onto which hold the point +
fixed (that is, for which 0 + +).
c) Give an example of a metric space which is isometric to a proper subset of itself.
97
E32. Use convergent sequences to prove the theorem that two continuous functions 0 and 1 from
\ . into a metric space are identical if they agree on a dense set in \ .
E33. Suppose 0 . Then we can define J # by J B B 0 B, so that ranJ is
the graph of 0 .
a) Prove that the following statements are equivalent:
i) 0 is continuous
ii) J is continuous
iii) The sets B C C 0 B and B C C 0 B are both closed in # .
b) Prove that if 0 is continuous, then its graph is a closed set in # . Give a proof or a
counterexample for the converse.
E34.
98
Chapter II Review
Explain why each statement is true, or provide a counterexample.
1
8
1
7
lBC l
"lBCl
12. Let .> denote the taxicab metric .> B C 83" lB3 C3 | on 8 . 8 is dense in (8 .> .
13. If F is a countable subset of , then F is dense in .
14. If E ! " and cl E ! ", then int E g.
15. Let E 81 : 8 }.
metric.
The discrete unit metric produces the same topology on E as the usual
99
|B C |
1 |BC|
and let +8
n#
n# + 1
. In ., (a8 ) 1 .
22. Suppose . is the usual metric on and ." is another metric with the property that
for every sequence <8
<8 & with respect to ." if and only if <8 & with respect to .
Then ." .
23. Suppose E F \ .. If cl E cl F , int E int F , and Fr E Fr F , then E F .
24. Suppose G! " has the metric .0 1 ! | f g | and define F G! " . by
"
F0 ! 0 . Then F is continuous.
"
25. Let . be the trivial pseudometric on . In . each real number < is the limit of a sequence of
irrational numbers.
26. Suppose 0 and 1 are continuous and that 0 1. Then there must exist a point
: where 0 : 1: and a point ; where 0 ; 1;
27. In a pseudometric space \ ., cl E int E if and only if E is clopen.
28. If \ . is a metric space in which every convergent sequence is eventually constant,
then g. c \ )
29. Let B \ . Suppose B8 is a sequence such that 0 B8 0 B for every continuous
0 \ . Then B8 B.
30. Let . 0 1 sup |0 B 1Bl B ! " for 0 1 G! ".
Let 0 G! " be the function given by 0 B B # and let 08 be a sequence such that
08 0 in G! " .
Then there is an R such that 08 B B for all B ! " and all 8 R .
31. If 0 is continuous, then the graph of 0 is a closed subset of # .
32. If the graph of a function 0 is closed subset of # , then 0 is continuous.
33. The are exactly - different metrics . on for which g. the usual topology on .
34. In , the interval # " can be written as a countable intersection of open sets.
35. For any 0 then there is a continuous function 1 such that 1l 0
36. Let . be the usual metric on and . w the discrete unit metric. Suppose 0 . . w ) is
continuous. Then 0 is constant.
37. If l\l " then there are infinitely many different metrics . on \ for which g. is the discrete
topology.
100
For B , let
101
102