FTAR – 1o S 2014/15

4 Countability axioms
Definition 4.1. Let X be a topological space

• X is said to be first countable if for any x ∈ X, there is a countable basis for the
neighborhoods of x.

• X is said to be second countable if there is a countable basis for the topology on X.

Of course, second countable ⇒ first countable. First countability is a mild assump-

tion, while having a countable basis gives much more information about a space.
A large class of first countable spaces are metric spaces: just take the balls at x with
radius 1/n, n ∈ N (or rational). We proved the following proposition for metric topolo-
gies:

Proposition 4.2. Let X be first countable, A ⊂ X, Y a topological space. Then

(i) x ∈ A if, and only if, there is a sequence xn ∈ A, n ∈ N, such that xn → x.

(ii) f : X → Y is continuous if, and only if, for any xn → x, xn , x ∈ X, then f (xn ) → f (x).

Examples 4.3. 1. Discrete spaces always first countable (are metrizable): {x} is a basis
at x, but second countable if, and only if, countable.

2. Rn is first and second countable:

]p1 , q1 [×...×]pn , qn [, pi , qi ∈ Q, i = 1, ..., n

is a countable basis for the standard topology.

3. An uncountable set with the co-finite / co-countable topology is not first countable:
if B is a basis, x ∈ X, have {x} = ∩B∈B B (it is T1 - see next section), if B is countable
then X \ {x} is countable union of finite/countable sets, hence countable.

4. R2 with the topology given by slotted discs: a basic neighborhhod of x ∈ R2 is

given by {x} ∪ B with B an open ball with straight lines through x removed: not
first countable.

5. RN the space of real sequences, with the product topology, is second countable: a
basis is given by sets of the form ΠUn such that Uni =]pi , qi [, pi , qi ∈ Q, i = 1, ..., k,
and Un = R, n , ni .

6. RρN the space of real sequences, with the uniform topology, is not second countable:
{0, 1}N is an uncountable discrete subset, as ρ(x, y) = 1 for any sequences x , y,
hence Bρ (x, 1) = {x} is open. It is first countable, as it is a metric space.

5
FTAR – 1o S 2014/15

7. RN the space of real sequence with the box topology is not first countable (hence
not metrizable), it does not satisfy (1) in Proposition 4.2: take A = {x = (xn ) : xn > 0},
show that the zero sequence is in A but there is no sequence in A converging to 0.

8. R J with J uncountable is not first countable in the product topology (hence not
metrizable) it does not satisfy (1) in Proposition 4.2: take A to be the subset of all
sequences x = (xα ) with xα = 1 for all but finite values of α, then the zero sequence
is A but there is no sequence in A converging to 0 (since J is uncountable, for any
sequence xn = (xnα ) ∈ A there must α ∈ J such that xnα = 1 for all n ∈ N).

Proposition 4.4. (i) subspaces of first / second countable spaces are first / second countable;

(ii) countable products of first / second countable spaces are first / second countable (in product
topology).

(iii) open, continuous images of first / second countable spaces are first / second countable.

Proof. Let B = {Bn }n∈N be a countable basis (at x ∈ A or for the whole topology of X).
Then {B ∩ A : B ∈ B} is also a basis.
Let X = Πn∈N Xn and Bn be a basis for Xn (at xn ∈ Xn or for the whole topology of Xn ).
Then the collection of open sets of the form ΠUn such that Uni ∈ Bni for i = 1, ..., p and
Un = Xn , n , ni forms a basis.
Let f : X → Y be a continuous, open map. Then if B is a basis, also { f (B) : B ∈ B} is a
basis. 

Moreover, one can prove that a product of first / second countable spaces is first
/ second countable if, and only if, it is a countable product. It follows from (iii) that
countability axioms are topological properties, ie, preserved by homeomorphisms (it is
however not true in general that the continuous image of a first / second countable space
is first / second countable).
It is easy to check that in a space with a countable basis:

(i) discrete1 subsets must be countable;

(ii) collections of disjoint open sets must be countable.

Recall that a subspace D ⊂ X is dense if D = X.

Proposition 4.5. If X has a countable basis, then

(1) X has a countable dense subset,

(2) any open cover of X has a countable subcover.

1
A subset A ⊂ X is discrete if the induced subspace topology is discrete, ie, {x} is open in A, x ∈ A.

6
FTAR – 1o S 2014/15

Proof. Let B = {Bn }n∈N be a countable basis. To prove (1), pick xn ∈ Bn and let D =
{xn : n ∈ N}. Then for any x ∈ X and any neighborhood U of x, there is Bn ⊂ U, hence
D ∩ U , ∅ and x ∈ D, that is D is dense.2
To prove (2), let U = {Uα } be an open cover of X and for each n ∈ N, let Un = {U ∈
U : Bn ⊂ U}. If Un , ∅, pick Un ∈ Un and let U 0 = {Un }n∈N ⊂ U. Then for x ∈ X, exists
U ∈ U and Bn ∈ B such that x ∈ Bn ⊂ U, so Un is defined and x ∈ Bn ⊂ Un , and we
conclude that U 0 covers X.


A space X that satisfies (1) is said to be separable and a space X that satisfies (2) is said
to be Lindeloff. These conditions are usually weaker than being second countable, for
instance Rl is separable and Lindeloff but not second countable. However, if we a have
a metric:

Proposition 4.6. If X is metrizable, then

second countable ⇔ separable ⇔ Lindeloff.

Proof. If X is separable, and D is a dense countable set, then B = {B(a, 1/n) : a ∈ D} is

countable and is a basis for the topology of X: for any open U, x ∈ U, there is n ∈ N such
that B(x, 1/n) ⊂ U and there is a ∈ B(x, 1/(2n)), as D is dense. Then B(a, 1/(2n)) ⊂ U.
If X is Lindeloff, take a countable subcover of the collection B(x, 1/n), show it is a
basis. 

If X compact, it is Lindeloff. If it is compact and metrizable, then it is second countable

(can see directly: since for any n ∈ N it can be covered by a finite number of ball of
radius 1/n. The family of all those balls is a countable basis).

Proposition 4.7. (i) open subspaces of separable spaces are separable; closed subspaces of
Lindeloff are Lindeloff.

(ii) countable products of separable spaces are separable (in product topology).

(iii) continuous images of separable/ Lindeloff spaces are separable/ Lindelof.

Examples 4.8. 1. Rl
– first countable: [x, x + q[, q ∈ Q is a basis at x ∈ R
– not second countable: if B is a basis, let Bx ∈ B such that Bx ⊂ [x, x + 1[. Then
x , y ⇒ Bx , B y , as x = min Bx , min B y , hence B is uncountable.
– separable: Q is dense;
– Lindeloff.
2
This provides a way of constructing a dense subset from any basis: just pick one element from each
basic set.

7
FTAR – 1o S 2014/15

2. Rl × Rl : first countable and separable (as Rl is, not second countable (as Rl × {0} is
a subspace), not Lindeloff:
Let L = {(x, −x) : x ∈ R} then L is closed and L ∩ [x, b[×[−x, c[= {(x, −x)} (ie, the
subspace topology in L is discrete; since L uncountable, Rl2 not second countable).
Cover Rl2 by Rl2 \R and [a, b[×[−a, c[, a, b, c ∈ R, then there is no countable subcover,
hence not Lindeloff.
(Also shows: L closed, not separable, and L ⊂ Rl2 separable.)

3. The ordered square I02 = [0, 1] × [0, 1] with the dictionary order:
– it is Lindeloff - compact; but [0, 1]×]0, 1[ not Lindeloff (open), since the collection
{x}×]0, 1[, x ∈ [0, 1], has no countable subcover
– not separable
– not second countable: the collection of disjoint open sets {x}×]0, 1[, x ∈ [0, 1] is
uncountable (or {x} × {1/2} is a discrete uncountable set).

5 Separation axioms
Definition 5.1. Let X be a topological space, x, y ∈ X.

• X is T1 : if for x, y ∈ X, x , y there exist neighborhoods U of x and V of y such that

y < U and x < V.

• X is Hausdorff (or T2) if for x, y ∈ X, x , y there exist neighborhoods U of x and V

of y such that U ∩ V = ∅.

• X is regular if X is T1 3 and for x ∈ X, B ⊂ X closed with x < B, there exist open sets
U, V such that x ∈ U, B ⊂ V and U ∩ V = ∅.

• X is normal if X is T1 and for A, B ⊂ X closed with A ∩ B = ∅, there exist open sets

U, V such that A ⊂ U, B ⊂ V and U ∩ V = ∅.

Always have:

normal ⇒ regular ⇒ Hausdorff ⇒ T1

but the reverse implications NOT true in general (Hausdorff, not regular: RK ; T1, not
Hausdorff: RK /K; regular, not normal: Rl2 , or R J , J uncountable)

3
In R with topology given by ] − ∞, a[, a ∈ R (is T0, not T1, not Hausdorff) any two closed sets intersect

8
FTAR – 1o S 2014/15

Remark 5.2. 1. In T1 spaces: x < {y}, all x , y.

X is T1 ⇔ {x} is closed, x ∈ X
⇔ {x} = ∩x∈U open U = ∩B∈Bx B, Bx basis of nbhds of x.

In T1 spaces, x is limit point of A iff any nbhd of x intersects A in infinitely many

points. (If F = (A \ {x}) ∩ U is finite, hence closed, then X \ F) is open nbhd of x that
does not intersect A.)

2. In Hausdorff spaces, there is unicity of limits of sequences and of extensions of

continuous functions defined on dense subsets. Moreover:

X is Hausdorff ⇔ Λ = {(x, x) : x ∈ X} is closed in X × X

(note that U ∩ V = ∅ iff (U × V) ∩ Λ = ∅, so X is Hausdorff iff (x, y) < Λ for x , y)

that is, Λ ⊂ Λ.

Examples 5.3. 1. X infinite with the cofinite topology is T1 but not Hausdorff.4

2. RK Hausdorff, not regular: K is closed cannot be separated from 0.

3. Rl is normal, Rl2 not normal (but regular).

4. if J uncountable R J not normal (hard to prove: see Ex. 32.9 [Munkres] - see that
N J not normal)

The following result gives equivalent definitions of regular and normal spaces, that
we will use quite often:

Proposition 5.4. Let X be a T1 space.

(i) X is regular ⇔ x ∈ X, U nbhd of x, there exists nbhd W of x such that W ⊂ U.

(ii) X is normal ⇔ A ⊂ X closed, U ⊃ A open, there exists open W ⊃ A such that W ⊂ U.

Proof. (i) U nbhd of x ⇔ B = X \ U is closed, x < B; W nbhd of x such that W ⊂ U ⇔ the

open V = X \ W ⊃ X \ U = B and U ∩ V = ∅.


Proposition 5.5. 1) Products: if ∀α ∈ J, Xα is T1 / Hausdorff / regular, then X = Πα∈J Xα is T1

/ Hausdorff / regular.

2) Subspaces: if X is T1 / Hausdorff / regular and A ⊂ X, then A is T1 / Hausdorff / regular.

4
It is the smallest T1 topology on X - check.

9
FTAR – 1o S 2014/15

For normal spaces, not true: Rl2 not normal, Rl normal so 1) fails and [0, 1] J normal,
]0, 1[ J not normal, so 2) fails. However, if A ⊂ X is closed and X is normal, then A is
normal. In fact, the image of a normal space under a closed continuous map is normal.
It is not true in general that images under continuous functions of Hausdorff / regular
spaces are Hausdorff / regular, not even open or closed images. However, it is easy to
see that any of the separation axioms are topological, in that they are preserved by
homeomorphisms.
Even if regular is usually weaker that normal, if we add a countability axiom then
the two notions coincide.
Theorem 5.6. Let X have a countable basis (ie, second countable). Then X regular ⇔ X normal.
Proof. Need to show that X regular and second countable then X is normal. Let B be a
countable basis for X and A, B ⊂ X be closed.
For each x ∈ A, by regularity, can take a neighborhood U of x such that U ∩ B = ∅.
Since any such U contains a basis element and B is countable, we obtain a countable
covering of X by sets Un such that Un ∩ B = ∅, n ∈ N.
We can do the same for the set B to obtain a covering of B by open sets Vn such that
Vn ∩ A = ∅, n ∈ N. Let
U = ∪n∈N Un , V = ∪n∈N Vn .
Then U, V open, and U ⊃ A, V ⊃ B, but U and V need not be disjoint. Now let

Un0 = Un \ ∪nk=1 Vk , Vn0 = Vn \ ∪nk=1 Uk

and U0 = ∪Un0 , V = ∪Vn0 . Then U0 , V 0 open, U0 ⊃ A, V 0 ⊃ B and u0 ∩ V 0 = ∅. 

Note that we proved in fact that if X is Lindeloff and regular, then it is normal.

Theorem 5.7. Let X be compact Hausdorff. Then X is normal.

Proof. We have seen that in a Hausdorff space compact (disjoint) sets are separated
by open sets. Since any closed subset of X is compact, as X is compact, normality
follows. 

Corollary 5.8. Any locally compact Hausdorff space is regular.

Proof. It is a subspace of a normal, hence regular, space hence it is regular. (Alternatively,
any locally compact Hausdorff has a basis of neighborhoods with compact closure: given
x ∈ X and U a neighborhood of x, there exists open B, with x ∈ B and B ⊂ U, ie., X is
regular.) 
A locally compact Hausdorff space needs not be normal: e.g., take ]0, 1[ J ⊂ [0, 1] J with
J uncountable. Then ]0, 1[ J is locally compact Hausdorff as it is open in [0, 1] J , which is
compact Hausdorff (hence normal). But ]0, 1[ J  R J not normal.

10
FTAR – 1o S 2014/15

Theorem 5.9. Let X be metrizable. Then X is normal.

Proof. Let d be a metric inducing the topology on X, and A, B ⊂ X be closed, A ∩ B = ∅.
Then for each a ∈ A and b ∈ B, since A ∩ B = ∅ and B ∩ A = ∅, we can find a > 0 and
b > 0 such that B(a, a ) ∩ B = ∅ and B(b, b ) ∩ A = ∅. Let

U = ∪a∈A B(a, a /2), V = ∪b∈B B(b, b /2).

Then U, V open, A ⊂ U, B ⊂ V and U ∩ V = ∅: for a ∈ A, b ∈ B, d(a, b) ≥ max(a , b ), but if

there were z ∈ B(a, a /2) ∩ B(b, b /2) then
a + b
d(a, b) ≤ d(a, z) + d(z, b) < < max(a , b ).
2

Examples 5.10. 1. Rn is normal.

2. Any discrete space is normal - metrizable.

3. RN is normal - metrizable (any countable product of metrizable spaces is metriz-

able).

4. R J with the uniform topology is normal (not with product!)

6 Urysohn’s Lemma and applications

6.1 Urysohn’s Lemma
One of the most important features of normal spaces is that normality is the suitable
condition to prove the very useful Urysohn’s Lemma.
Theorem 6.1 (Urysohn’s Lemma). Let X be a normal space, A, B ⊂ X closed and disjoint.
Then there exists continuous

f : X → [0, 1] s.t. f (A) = 0, f (B) = 1.

(Of course, [0, 1] can be replaced by any interval [a, b].) Before we get to the proof,
note that, for a T1 space, the existence of such a function f clearly yields normality:
U = f −1 ([0, [ and V = f −1 (]1 − , 1] are open disjoint (if  ≤ 1/2) sets that contain A and
B, respectively.
To prove this theorem, we have to use the the topology on X to construct ’from
scratch’ such a continuous function f . The main idea is to construct a sequence of
ordered neighborhoods of A that will be the level sets of f .
We will use the alternative characterization of normal spaces: in a normal space,
given A closed and U ⊃ A open, there exists an open V ⊃ A such that V ⊂ U.

11
FTAR – 1o S 2014/15

Proof. We outline the proof (see the details in [Munkres]).

Step 1) First we construct a sequence Uq , for q ∈ [0, 1] ∩ Q such that Uq is open and

p < q ⇒ Up ⊂ Uq .
We do this as follows: let U1 = X \ B and by normality, take U0 open such that A ⊂ U0
and U0 ⊂ U1 . Let P = {qn : n ∈ N, qn ∈ [0, 1] ∩ Q}, for instance P = {1, 0, 21 , 13 , 23 , 14 , 34 , 15 , 52 , ...}.
Then since U0 ⊂ U1 , U0 closed, U1 open, we can take U 12 open such that U0 ⊂ U 12 and
U 12 ⊂ U1 . We then proceed to define U 13 in between U0 and U 12 , then U 23 in between U 12
and U1 , etc.

• U0 open such that A ⊂ U0 and U0 ⊂ U1 ;

• U 12 open such that U0 ⊂ U 12 and U 12 ⊂ U1 ;

• U 13 open such that U0 ⊂ U 13 and U 13 ⊂ U 12 ;

• U 23 open such that U 12 ⊂ U 13 and U 32 ⊂ U1 ;

• ...

We can show by induction that if we have a family Uq defined for the first n elements
of P then also Uqn+1 is well-defined: just take its immediate predecessor p and sucessor q
in Pn = {q0 , q1 , ..., qn }. Since p < q, Up ⊂ Uq ; let Uqn+1 such that Up ⊂ Uqn+1 and Uqn+1 ⊂ Uq .
Then Up ⊂ Uq for all p < q, for all p, q ∈ Pn+1 , that is, for all p, q ∈ Q ∩ [0, 1].
For convenience, we define Uq = ∅ for q ∈ Q, q < 0, and Uq = X, for q ∈ Q, q > 1.
Then Up ⊂ Uq for all p < q, p, q ∈ Q.
Step 2) For any x ∈ X, the set {p ∈ Q : x ∈ Up } is non-empty and bounded from below,
hence we can define
f (x) = inf{p ∈ Q : x ∈ Up }.
Then f (A) = 0, and f (B) = 1.
Step 3) Show that f is continuous. Note that
(1) x ∈ Ur ⇒ f (x) ≤ r: since if x ∈ Ur then x ∈ Us , for s > r, hence f (x) ≤ r;
(2) x < Ur ⇒ f (x) ≥ r: since for s < r, Us ⊂ Ur , hence x < Us , s < r and f (x) ≥ r.
Let x0 ∈ X such that f (x0 ) ∈ ] a, b [. Take p, q ∈ Q such that x0 ∈ ] p, q [⊂] a, b [. Then, p <
f (x0 ) < q yields that x0 ∈ Uq \ Up (by (2), f (x) < q ⇒ x ∈ Uq and by (1), f (x) > p ⇒ x < Up ).
Moreover,
x ∈ Uq ⇒ f (x) ≤ q, x < Up ⇒ f (x) ≥ p.
It follows that for any x ∈ U = Uq \ Up open, f (x) ∈ [p, q] hence f (U) ⊂] a, b [ and f is
continuous. 

12
FTAR – 1o S 2014/15

Corollary 6.2. Let X be a normal space, A closed and U ⊃ A open. Then there exists continuous

f : X → [0, 1] s.t. f (A) = 1, f (x) = 0, x < U.5

Proof. Just take B = X \ U in Urysohn’s lemma. 

It follows from Urysohn’s Lemma that X is normal ⇔ X is T1 and for A, B ⊂ X closed,

there is f : X → [0, 1] continuous such that f (A) = 0 and f (B) = 1.

6.2 Completely regular spaces

Definition 6.3. X is completely regular if X is T1 and for x ∈ X, B ⊂ X closed with x < B,
there if f : X → [0, 1] continuous such that f (x) = 0, f (B) = 1.

A completely regular space is regular: given x ∈ X and A closed with x < A, let
U = f −1 ([0, 1/2[ and V = f −1 (]1/2, 1]). We always have

normal ⇒ completely regular ⇒ regular

but the reverse implications are also NOT true in general: Rl2 is completely regular
(being the product of completely regular spaces) but not normal. Examples of regular
spaces that are not completely regular are more sophisticated (see Tychonoff ’cork screw’
example, or [Munkres] Ex. 33.11).
This new axiom is better behaved than normality:

Proposition 6.4. 1) Products: if ∀α ∈ J, Xα is completely regular, then X = Πα∈J Xα is

completely regular.

2) Subspaces: if X is completely regular and A ⊂ X, then A is completely regular.

It follows that any subspace of a normal space is completely regular. In particular,

any subspace of a compact, Hausdorff space is completely regular. In particular, any
locally compact Hausdorff space is completely regular. We see now that completely
regular spaces can always be embedded in a compact Hausdorff space, which leads
to a characterization of spaces that have a compactification, and is a step in proving
Urysohn’s metrization theorem.
First note that we can give equivalent definition of a completely regular space using
neighborhoods, instead of closed sets: X is completely regular if, and only if, X is T1 and
given x0 ∈ U open, there exists a continuous function f : X → [0, 1] such that f (x0 ) = 1
and f (x) = 0, x < U.

5
f is said to have support in U.

13
FTAR – 1o S 2014/15

Theorem 6.5 (Embedding theorem). Let X be T1. If there is a family of continuous functions
fα : X → [0, 1] such that for x0 ∈ U open, there exists α such that fα (x0 ) > 0 and fα (X \ U) = 0,
then
F : X → [0, 1] J , F(x) = ( fα (x))α∈J
is an embedding.

Proof. First, F is continuous: in the product topology F is continuous iff πα ◦ F = fα is

continuous for all α ∈ J.
Then F is injective: if x , y, take a neighborhood U of x such that y < U (X is T1). Then
there is α ∈ J such that fα (x) > 0 and fα (y) = 0, hence (F(x))α , (F(y))α ⇒ F(x) , F(y).
Now see that F is open: let U ⊂ X be open, x0 ∈ U and z0 = F(x0 ) ∈ F(U). We show
that there is a neighborhhod of z0 contained in F(U), hence F(U) is open.
Take α such that fα (x0 ) > 0, ie, (z0 )α > 0, and fα (X \ U) = 0. Let W = π−1
α (]0, 1]) ∩ F(X),
open in F(X). Then z0 ∈ W and

z ∈ W ⇔ πα (z) > 0 ∧ z ∈ F(X) ⇒ z = F(x) ∧ πα (F(x)) = fα (x) > 0 ⇒ x ∈ U.

Hence, W ⊂ F(U). 

Since the conditions of the embedding theorem are equivalent to X being completely
regular, and by Tychonoff’s theorem, [0, 1] J is compact, it follows:

Corollary 6.6. X is completely regular if, and only if, X is homeomorphic to a subspace of a
compact, Hausdorff space.

In particular, any metrizable (hence normal) space embeds in a compact, Hausdorff

space.

Remark 6.7. We saw in section 3 that a compactification of a space X is a compact

Hausdorff space Y such that X ⊂ Y and X = Y. There can be many compactifications to
a given (necessarily completely regular) space. We saw that if X is locally compact, then
there is a minimal compactification, adding one point ’at infinity’.
Now any embedding h : X → Z with Z compact, Hausdorff, defines a compactifi-
cation of X: take X0 = h(X)  X and Y0 = h(X), which is compact, Hausdorff (being
closed). Let A be a set such that there is a bijection k : A → Y0 \ X0 and define Y = X ∪ A.
Then define H : Y → Y0 such that H(x) = h(x), x ∈ X, h(x) = k(x), x ∈ A. If we endow Y
with the topology given by U open iff H(U) open, we get that H is a homeomorphism,
Y is compact Hausdorff and X ⊂ Y is a subspace. Moreover, H(X) = H(X) = h(X) = Y0 ,
hence X = Y. (Note that H : Y → Z is an embedding that extends h.)
The results above show that

X is completely regular iff it has a compactification.

14
FTAR – 1o S 2014/15

6.3 Urysohn’s Metrization theorem

We will now see Urysohn’s metrization theorem. Recall that a countable product of
metric spaces is always metrizable: if X = Πn∈N Xn and dn (a) = min{dn (a), 1} is the
standard bounded metric associated to the metric dn on Xn , then we can define a metric
(inducing the topology) on X for instance by

dn (xn , yn )
D(x, y) := sup .
n∈N n
Theorem 6.8 (Urysohn’s metrization theorem). Let X be regular and have countable basis.
Then X is metrizable.
(Note that metrizable ⇒ regular, but not necessarily second countable. eg discrete
uncountable space, or the space of sequence RN in the uniform topology.)
Proof. We will construct an embedding of X in [0, 1]N with the product topology, which
yields that X, being a subspace of a metrizable space, is also metrizable.
We showed in Theorem 5.6 that a regular, second countable space is normal. We will
use Urysohn’s lemma, and the countable basis, to construct a countable family satisfying
the conditions of the Embedding theorem.
Let B = {Bn }n∈N be a countable basis for the topology on X. Let x0 ∈ X, U a
neighborhood of x0 and Bn ∈ B such that x0 ∈ Bn ⊂ U. By regularity, we can take Bm such
that Bm ⊂ Bn ⊂ U. Now apply Urysohn’s lemma (in fact, Corollary 6.2) to Bm ⊂ Bn to get
that there is a continuous function
gn,m : X → [0, 1], gn,m (Bm ) = 1, gn,m (X \ Bn ) = 0,
in particular, gn,m (x0 ) = 1 and gn,m (x) = 0, x < U.
We obtain in this way a countable family of function satisfying the conditions of
the Embedding theorem. Hence X can be embedded in [0, 1]N , and therefore it is
metrizable. 

Corollary 6.9. X regular and second countable ⇔ X metrizable and separable .

Proof. Recall that if X is metrizable, then second countable ⇔ separable. 

Corollary 6.10. Let X be compact Hausdorff. Then X metrizable ⇔ second countable.

Example 6.11. Recall that a n-manifold M is a Hausdorff, second countable space where
each point has a neighborhood diffeomorphic to an open subset of Rn .
Then, since the closed ball in Rn is compact (Heine-Borel’s theorem), we can show
that M is locally compact, hence regular. It follows that any n-manifold is metrizable.
Another nice application of Urysohn’s Lemma is a short proof that any compact
n-manifold can be embedded in RN , for some N (see section 36 - [Munkres]).

15
FTAR – 1o S 2014/15

6.4 Tietze extension theorem

Our final application of Urysohn’s lemma is an extension theorem. Extending contin-
uous functions is a useful tool in many applications of topology. A consequence of
Urysohn’s lemma is the following important theorem, that says that in normal spaces,
real functions on a closed subset can be always extended to the whole space.6

Theorem 6.12 (Tietze extension theorem). Let X be a normal space, A ⊂ X be closed. Then

(i) f : A → [a, b] continuous can be extended to g : X → [a, b] continuous;

(ii) f : A → R continuous can be extended to g : X → R continuous.

In fact, this theorem is equivalent to Urysohn’s lemma, in that, in one direction, we

use it in the proof, and conversely, if (i) or (ii) holds then f (A) = 0, f (B) = 1 for disjoint,
closed A, B ⊂ X can be extended to X, in particular, assuming X is T1, then it is normal

Proof. The main idea is to construct a sequence sn of continuous functions, which con-
verges uniformly to some function g with g = f on A. Uniform convergence guarantees
that g is continuous.
Let us consider the case (i), can assume that [a, b] = [−1, 1]. (Then (ii) will follow,
identifying R ] − 1, 1[ and using (i).) h i
Let B1 = f −1 ([−1, − 31 ]) and C1 = f −1 13 , 1 . Then B1 , C1 are closed and disjoint, hence
by Urysohn’s lemma, there exists
 
1 1 1 1
g1 : X → − , , g1 (B1 ) = − , g1 (C1 ) = .
3 3 3 3

In particular:

(i) |g1 (x)| ≤ 13 , for x ∈ X,

h i h i h i
(ii) | f (a) − g1 (a)| ≤ 32 , for a ∈ A: note that [−1, 1] = −1, − 13 ∪ − 13 , 13 ∪ 13 , 1 and each of
these sub-intervals has length 2/3, so it suffices to show that f (a) and g1 (a) are in
the same sub-interval. Now have 3 cases:
– a ∈ B1 : then f (a) ∈ [−1, − 31 ] and g1 (a) = − 31 ;
h i
– a ∈ C1 : then f (a) ∈ 13 , 1 and g1 (a) = 31 ;
h i
– a ∈ A \ (B1 ∪ C1 ): then f (a), g1 (a) ∈ − 13 , 13 .

Let s1 = g1 .
6
It is clear that on open sets that is not always possible, e.g, f (x) = 1/x on R+ .

16
FTAR – 1o S 2014/15

h i
Now apply the step above to the function f − g1 : A → − 23 , 32 to obtain a function
h i
g2 : X → − 92 , 29 such that
 2
2
| f (a) − g1 (a) − g2 (a)| ≤ , a ∈ A.
3
Let s2 = g1 + g2 .   n−1  n−1 
We can proceed by induction to define functions gn : X → − 13 23 , 13 23 such
Pn
that, writing sn = i=1 gi , we get
 n
2
| f (a) − sn (a)| ≤ , a ∈ A.
3
In particular, for a ∈ A, sn (a) → f (a), that is, sn converges pointwise to f in A.
 n−1
Moreover, since |gn (x)| ≤ 31 23 , by the comparison test for infinite series, for each
Pn
x ∈ X, sn (x) = i=1 gi (x) converges and we define
∞
X
g(x) = gn (x) = lim sn (x).
n→∞
n=1

We have that g = f on A. Need only show that g is continuous, which will follow once
we show that the convergence is uniform (exercise). 

In fact, the fact that the sequence sn converges uniformly (can be proved directly but)
follows from the fact that it is a Cauchy sequence and that R is complete. We will see these
notions next.

17