Metric and metrizable spaces

These notes discuss the same topic as Sections 20 and 21 of Munkres book; some
notions (Symmetric, -metric, -spaces ...) are not discussed in Munkres book.
1. Symmetric, -metric, and metric
Denition 1. Let X be a set. A function d : X
2
R is called a symmetric on X
if d satises the following two axioms:
(1) (x, y X) d(x, y) 0 and d(x, y) = 0 i x = y;
(2) (x, y X) d(x, y) = d(y, x) (the axiom of symmetry.
If, in addition, d satises the following axiom of triangle
(3) (x, y, z X) d(x, z) d(x, y) + d(y, z),
then d is called a metric on X. A function d that satises axioms (1) and (3) is
called a -metric.
If d is a metric (symmetric, -metric) on a set X then the pair (X, d) is called
a metric (resp. symmetric, -metric) space.
Examples 2. (1) X = R; d(x, y) = |x y| is the standard metric on R;
(2) X = R
n
; d(x, y) =

k=1
(x(k) y(k))
2
is the Euclidean metric on R
n
;
1
(3) X = R
n
; d(x, y) =
n

k=1
|x(k) y(k)| is the NY-metric on R
n
;
(4) X = R
n
; d(x, y) = max
1kn
|x(k) y(k)| is the maximum metric on R
n
;
(5) Consider the set X from Example 22 in the notes Countability conditions
and convergence. For x, y X, put
d(x, y) = d(y, x) =
_
_
_
1
m
if one of x, y is b
m
and the other is c;
1
n
if one of x, y is a
m,n
and the other is b
m
;
1 in all remaining cases.
Then d is a symmetric on X which does not satisfy axiom 3 and thus is not
a metric;
(6) X is any non empty set;
d(x, y) =
_
0 if x = y;
1 if x = y
is a metric on X.
Denition 3. Let X be a set, x X, d a (sym)metric on X and > 0. The -ball
around x is
O

(x) = {y X : d(x, y) < }.

Exercise 4. Sketch -balls for Examples 2; in cases 2, 3, 4, take n = 2.
Denition 5. A topology T on the set X is said to be generated by the metric d
on X if the family {O

(x) : x X, > 0} is a base for T . If a topology T on a

set X can be generated by some metric on X then we say that the space (X, T ) is
metrizable.
A topology T on the set X is said to be generated by the symmetric d on X if
the family if the following two conditions are equivalent for U X:
1
Here and in the next two examples, n N, x = x(1), ..., x(n), y = y(1), ..., y(n).
1
2
U T ;
for every x U there is > 0 such that O

(x) U.
If a topology T on a set X can be generated by some symmetric on X then we say
that the space (X, T ) is symmetrizable.
2
Examples 6. The metric d from Example 2(6) generates the discrete topol-
ogy on X.
The metric d from Example 2(1) generates the standard topology on R.
The metrics d from Examples 2(2,3,4) all generate the standard topology
on R
n
.
The semymetric d from Example 2(5) generates the topology on X that
we discussed in Example 22 in the notes Countability conditions and con-
vergence, and then mentioned again when discussing quotient spaces and
quotient mappings. Recall that this topology is not rst-countable.
2. Properties of metrizable spaces; more on countability conditions
Proposition 7. Every metizable space is Hausdor.
3
Proposition 8. Every metizable space is rst countable.
Exercise 9. Give an example of a metrizable space which is not second countable.
Denition 10. Let (X, T ) be a topological space. A subset D X is called dense
in X if D = X. If X has a dense countable subset then X is called separable.
Denition 11. Let X be a set. A family U of subsets of X is called a cover of X
4
if

U = X.
A space (X, T ) is called Lindelof if for every cover U of X consisting of open
sets
5
there exists a countable subfamily of U that still covers X.
Theorem 12. Every second countable space is both separable and Lindelof.
Examples 13. (1) The Sorgenfrey line is both separable and Lindelof but (as
we already know) not second countable.
(2) The one-point Lindelocation of a discrete space. Let D be an uncount-
able set, and p D. Put X = D{p}. Topologize X declaring the points of
D isolated while a basic neighborhood of p takes the form O
C
= {p}(D\C)
where C is an arbitrary countable subset of D. Then X is Lindelof but non
separable and thus non second countable.
(3) (R, T
fin
) is separable but non Lindelof and thus non second countable.
Theorem 14. For a metrizable space X, the following conditions are equivalent:
(1) X is second countable;
(2) X is separable;
(3) X is Lindelof.
6
2
The reason for the dierence in the denitions is that, in the case of a metric d, the balls
O(x) have to be open in the topology generated by d while in the case of a symmetric d they do
not have to.
3
In fact, much better than Hausdor, we will see this later
4
Or we say that U covers X
5
briey: for every open cover
6
The proof will be discussed in class.
3
Proposition 15. Every subspace of a metrizable space is metrizable.
Proposition 16. Every nite or countable
7
product of metrizable spaces is metriz-
able.
A product of uncountably many factors each of which consists of more than one
point is (not rst countable therefore) non metrizable.
3. Properties (and more examples) of symmetrizable spaces
Proposition 17. Every symmetrizable space is T
1
.
Proposition 18. Every symmetrizable space is sequential.
Examples 19. (1) Niemytzky plane (see Hw 3, Part 1, Problem 1) is sym-
metrizable.
(2) Lets return to Example 2, (5). This space is symmetrizable, hence se-
quential, but, as we already know, it is not Frechet. So a symmetrizable
space
4. Just a little about -metrizable spaces
The Sorgenfrey line is -metrizable
8
but not symmetrizable.
9
A real life example of a -metric space which is not symmetric: in a city, measure
the distance between points by the time needed to travel; consider a city with some
one-way routes.
Another example: traveling in the mountains.
A real life example of a symmetric space which is not -metric: measure the
distance between cities by the time needed to travel by air. Assume that if there
is a ight from A to B, then there is one from B to A that takes the same time.
Now suppose there is no direct ight from A to C, but there is one from A to B,
and one from B to C, and at B you have to wait.
5. One more generalization of metric spaces; -spaces
10
Denition 20. A function d : X
2
R that satises axiom 1 from Denition 1 is
called a o-metric on X. Then the pair (X, d) is called an o-metric space. -balls are
dened like in Denition 3. If the topology T on X is generated by the o-metric d
in the way described in Denition 5 for symmetrics, then the pair (X, T ) is called
an o-metrizable (or weakly rst-countable
11
) space.
Example 21. The Alexandro-Urysohn -space. For each r P, x a subset S
r
of
Q which can be enumerated as a sequence converging to r. Consider the topology
T

on R such that the points of Q are isolated in T

while a basic neighborhood of

a point r P takes the form {r} (S
r
\ F) where F is an arbitrary nite subset of
S
r
. Let us call the pair (R, T

) an AU--space.
7
Countable products will be discussed later
8
Very easy to see
9
The proof that I know uses the Baire property (discussed in Chapter 8 of Munkres book).
10
This section is optional
11
Guess why
4
Let E denote the standard topology on R. Then T

E, so the identity mapping

i : R R is a condensation of an AU--space onto the real line with the standard
topology. Note that Q is an open discrete subspace of (R, T

), and P is a closed
discrete subspace of (R, T

). (R, T

) is rst-countable, hence o-mtrizable. With

an accurate choice of the sequences S
r
, (R, T

) can be made symmetrizable. Since

(R, T

) contains an uncountable discrete subspace, (R, T

) is not second countable.

Moreover, since (R, T

) contains an uncountable closed discrete subspace, (R, T

)
is not even Lindelof.
12
Since (R, T

) is separable but not second countable, (R, T

)
is not metrizable.
Example 22. The general denition of a -space. Let N be a countably innite
set. A family R of countably innite subsets of N is called almost disjoint if for
every two distinct r
1
, r
2
R, the intersection r
1
r
2
is nite.
13
So let R be an
almost disjoint family of innite subsets of N. The set (R) = NRis topologized
by declaring the points of N isolated while a basic neighborhood of r R takes
the form {r} (r \ F) where F is an arbitrary nite subset of r.
14
A general space has properties similar to those of the AU--space: N is open
and discrete in (R) while R is closed and discrete. (R) is always rst countable.
(R) is second countable i Ris countable (and in this case (R) is homeomorphic
to the discrete union of countably many copies of the space convergent sequence).
Denition 23. A function f : X R is bounded if there is a constant M such
that |f(x)| < M for every x X. A topological space is pseudocompact if every
continuous function f : X R is bounded.
Denition 24. Let N be a countably innite set. An almost disjoint family R
consisting of innite subsets of N is called maximal almost disjoint
15
if R is not a
proper subset of any larger almost disjoint family consisting of innite subsets of
N.
One can prove using the axiom of Choice that every almost disjoint family can
be extended to a maximal almost disjoint family.
Proposition 25. (R) is pseudocompact i R is a mad family.
(R) with a mad family R is not symmetrizable except when R is nite.
6. Bounded metrics
It is natural to call a metric d on a set X bounded if there is a constant M such
that d(x, y) < M for every x, y X.
Notation 26. Let d be a metric on a set X. For every x, y X, put
d(x, y) = min{d(x, y), 1}.
Then d is called the standard bounded metric corresponding to d.
12
It is a good exercise to show that a closed discrete subspace of a Lindelof space can be at
most countable.
13
Thus the family {Sr : r P} in Example 21 is almost disjoint because sequences converging
to dierent points may have only nitely many common points. This also shows that an almost
disjoint family of subsets of a countable set may have cardinality as big as c (while a disjoint
family of subsets of a countable space is at most contable).
14
So r R is, at the same time, a point of (R), and a subset of (R).
15
Abbreviated as mad
5
Proposition 27. (1) d is a metric.
(2) The topology on X generated by d is the same as the topology generated by d.
So every metrizable topology can be generated by a bounded metric. Here is
another way:
Exercise 28. Let d be a metric on a set X. For every x, y X, put

d(x, y) =
Arctan d(x, y). Show that

d is a metric on X, and that

d generates the same
topology on X as d.
7. Metrics on finite products
Examples 29. Let n N and for each k {1, ..., n} let (X
k
, T
k
) be a topo-
logical space and d
k
a metric on X
k
that generates the topology T
k
. For x =
x(1), ..., x(n), y = y(1), ..., y(n) P =

1in
X
i
put
d
Euclidean
(x, y) =

_
n

k=1
(x(k) y(k))
2
;
d
NY
(x, y) =
n

k=1
|x(k) y(k)|; d
max
(x, y) = max
1kn
|x(k) y(k)|.
Then each of d
Euclidean
, d
NY
, d
max
(is a metric and) generates the Tychono product
topology on P.
8. Metrics on infinite products
Denition 30. Let for each a A, (X
a
, T
a
) be a topological space such that the
topology T
a
is generated by the metric d
a
. For x, y P =

aA
X
a
, put
d(x, y) = sup
aA
d
a
(x(a), y(a)).
Then d is called the uniform metric on P and the topology generated by d is called
the uniform topology.
The uniform topology on P is stronger than the Tychono product topology but
weaker than the box product topology.
Proposition 31. Let for each n N, (X
n
, T
n
) be a topological space such that the
topology T
n
is generated by the metric d
n
. For x, y P =

nN
X
n
, put
D(x, y) = sup
nN
1
n
d
n
(x(n), y(n)).
Then D generates the Tychono product topology on P.
9. Mappings of metric spaces
Proposition 32. Let (X, d
X
) and (Y, d
Y
) be two metric spaces. A mapping f :
X Y is continuous (with respect to the topologies generated on X and Y by d
X
and d
Y
) i for every x
0
X and every > 0 there exists > 0 such that for every
x X, if d
X
(x, x
0
) < then d
Y
(f(x), f(x
0
)) < .
6
Denition 33. Let (X, d
X
) and (Y, d
Y
) be two metric spaces. A mapping f :
X Y is called uniformly continuous (with respect to the metrics d
X
and d
Y
) if
for every > 0 there exists > 0 such that for every x
0
, x X, if d
X
(x, x
0
) <
then d
Y
(f(x), f(x
0
)) < .
Example 34. The mapping f : R R is continuous (with respect to the standard
topology on R) but not uniformly continuous (with respect to the standard metric
on R.
Denition 35. Let (X, T ) be a topological space. A sequence of functions {f
n
:
n N} (where f
n
: X R) converges pointwise to a function f : X R if for
every x X, lim
n
f
n
(x) = f(x) (that is for every x X for every > 0 there is
N(, x) such that for all n > N(, x), |f
n
(x) f(x)| < ). In this case we say that
f is the pointwise limit of the sequence {f
n
: n N}.
A sequence of functions {f
n
: n N} (where f
n
: X R) converges uniformly
to a function f : X R if for every > 0 there is N() such that for all n > N(),
for every x X, |f
n
(x) f(x)| < ).In this case we say that f is the uniform limit
of the sequence {f
n
: n N}.
Denition 35 corresponds to the convergence in the sense of pointwise or uniform
topology on R
X
.
16
Example 36. The pointwise limit of a sequence of continuous functions is not
necessarily continuous.
Indeed, put
f
n
(x) =
_
_
_
0 if x 0;
nx if 0 x
1
n
;
1 if x
1
n
for n N and f(x) =
_
0 if x 0;
1 if x > 0.
Proposition 37. The uniform limit of a sequence of continuous functions is a
continuous function.
16
Note that typically R
X
is not sequential, so convergent are not enough to describe its
topology.