Urysohns Theorem
Urysohns Theorem
The separation axioms introduced in the last chapter can be seen as a tool to constructing closer and
closer approximations of the class of metrizable spaces. However, even normal spaces, i.e. spaces that
satisfy the of the strongest of these axioms need not be metrizable. For example, take the real line
R with the arrow topology (4.8). One can show that it is a normal space (exercise), but by Exercise
5.15 this space is not metrizable. The Urysohn Lemma, which is the main result of this chapter, shows
however that normal spaces retain some useful properties of metrizable spaces. Recall that in the last
chapter we have seen that for any metric space X , and any pair of disjoint closed sets in X we can find
is a continuous function f : X → [0, 1] which maps one set to 0 and the other set to 1. The Urysohn
lemma says that the same property holds for any normal space:
10.1 Urysohn Lemma. Let X be a normal space and let A, B ⊆ X be closed sets such that A ∩ B = ∅.
There exists a continuous function f : X → [0, 1] such that A ⊆ f −1 ({0}) and B ⊆ f −1 ({1}).
10.2 Lemma. Let X be a topological space. Assume that for each r ∈ [0, 1] ∩ Q we are given an open
set Vr ⊆ X such that V r ⊆ Vr 0 if r < r 0 . There exists a continuous function f : X → [0, 1] such that if
x ∈ Vr then f(x) ≤ r and if x 6∈ V1 then f(x) = 1.
X
V1
V8/9
V3/5
V1/2
V0
66
10. Urysohn Lemma 67
is a subbasis of the topology on [0, 1], so it will suffice to show that for any a ∈ [0, 1] the sets f −1 ([0, a))
and f −1 ((a, 1]) are open in X .
We have: [
f −1 ([0, a)) = Vr
r<a
Indeed, if x ∈ X r V r for some r > a then x 6∈ Vr . This gives f(x) ≥ r > a, and so x ∈ f −1 ((a, 1]).
Conversely, assume that x ∈ f −1 ((a, 1]). Then f(x) > a so there exist r > a such that x 6∈ Vr Take
r 0 ∈ [0, 1] ∩ Q such 0
S that a < r < r. Since V r ⊆ Vr we get that x 6∈ V r , or equivalently x ∈ X r V r .
0 0 0
Therefore x ∈ r>a X r V r .
Since the sets X r V r are open it follows that f −1 ((a, 1]) is an open set.
10.3 Lemma. Let X be a normal space, let A ⊆ X be a closed set and let U ⊆ X be an open set such
that A ⊆ U. There exists an open set V such that A ⊆ V and V ⊆ U.
X
U
Proof. Exercise.
10. Urysohn Lemma 68
Proof of Urysohn Lemma 10.1. We will show that for each r ∈ [0, 1] ∩ Q there exists an open set
Vr ⊆ X such that
1) A ⊆ V0
2) B ⊆ X r V1
3) if r < r 0 then V r ⊆ Vr 0 .
X
V1
V8/9
V3/5
V1/2
V0
A B
By Lemma 10.2 this will give a continuous function f : X → [0, 1] such that f(x) ≤ r for all x ∈ Vr and
f(x) = 1 for all x 6∈ V1 . By 1) we will get then that f(x) = 0 for all x ∈ A and by 2) that f(x) = 1 for
all x ∈ B.
Construction of sets Vr proceeds as follows. Since the set [0, 1] ∩ Q is countable we can arrange its
elements into a sequence:
[0, 1] ∩ Q = {r0 , r1 , r2 , . . . }
We can assume that r0 = 0 and r1 = 1. We will construct the sets Vrk by induction with respect to k.
Take Vr1 = X r B. Since Vr1 is open and A ⊆ Vr1 by Lemma 10.3 there exists an open set V such that
A ⊆ V and V ⊆ Vr1 . Define Vr0 = V .
Next, assume that we have already constructed sets Vr0 , . . . , Vrn . We obtain the set Vrn+1 as follows.
Let rp be the biggest number in the set {r0 , . . . , rn } satisfying rp < rn+1 , and let rq be the smallest
number in {r0 , . . . , rn } satisfying rn+1 < rq . Since rp < rq we have V rp ⊆ Vrq . By Lemma 10.3 there
exists an open set V such that V rp ⊆ V and V ⊆ Vrq . We set Vrn+1 := V .
One can ask whether an analog of Urysohn Lemma holds for regular spaces: given a regular space X ,
a point x ∈ X , and a closed set A ⊆ X such that x 6∈ A is there a continuous function f : X → [0, 1]
such that f(x) = 0 and f(A) ⊆ {1}? It turns out that this is not true, but it provides motivation for one
more separation axiom:
10.4 Definition. A topological space X satisfies the axiom T31/2 if X satisfies T1 and if for each point
x ∈ X and each closed set A ⊆ X such that x 6∈ A there exists a continuous function f : X → [0, 1]
such that f(x) = 1 and f|A = 0.
A space that satisfies the axiom T31/2 is called a completely regular space or a Tychonoff space.
10. Urysohn Lemma 69
While Definition 10.4 may seem a bit artificial at the moment, there is a different context which makes
the class of completely regular spaces interesting. We will get back to this in Chapter 18.
10.5 Note. By Urysohn Lemma every normal space is completely regular. Also, if X is a completely
regular space then X is regular. Indeed, for a point x ∈ X and a closed set A ⊆ X such that x 6∈ A let
f : X → [0, 1] be a function as in Definition 10.4. Let U = f −1 ([0, 12 )) and let V = f −1 (( 12 , 1]). Then the
sets U, V are open in X , A ⊆ U, x ∈ V , and U ∩ V = ∅.
T1
T2
T3
T31/2
T4
antidiscrete spaces
No area of this diagram is empty: there exist regular spaces that are not completely regular and there
exist completely regular spaces that are not normal.
Exercises to Chapter 10
E10.1 Exercise. Let RAr denote the set of real numbers with the arrow topology (4.8). Show that this
space is normal.
E10.3 Exercise. By Corollary 9.19 metric spaces satisfy a stronger version of the Urysohn Lemma
10.2: for any pair of disjoint, closed subsets A, B in a metric space X there exists a continuous function
f : X → [0, 1] such that A = f −1 ({0}) and B = f −1 ({1}). One can ask if the same is true for all normal
spaces. The goal of this exercise is to resolve this question.
set A ⊆ X is called a Gδ -set if there exists a countable family of open sets U1 , U2 , . . . such that
a) A T
A= ∞ n=1 Un . Let X be a topological space and let A, B ⊆ X be disjoint, closed subsets such that
there exists a continuous function f : X → [0, 1] with A = f −1 ({0}) and B = f −1 ({1}). Show that both
A and B are Gδ -sets.
10. Urysohn Lemma 70
Note: One can also show that the converse holds: if X is a normal space and A, B are closed, disjoint
Gδ -sets in X then such function f exists (see Exercise 11.4).
Notice that as a consequence the space X described in part b) gives another example of a space which
is normal but not metrizable.