Urysohn's Lemma: 22M:132 Fall 07 J. Simon

22M:132
Fall 07
J. Simon
Urysohn’s Lemma
(Relates to text Sec. 32-36)
Introduction. Saying that a space X is normal turns out to be a very strong

assumption. In particular, normal spaces admit a lot of continuous functions:
Theorem (Urysohn’s Lemma). If A, B are disjoint closed sets in a normal space
X, then there exists a continuous function f : X → [0, 1] such that ∀a ∈ A, f (a) = 0
and ∀b ∈ B, f (b) = 1.
This lemma has several “big” applications:
(1) Urysohn Metrization Theorem (Section 34) If X is a normal space
with a countable basis (i.e. second-countable), then we can use the abundance
of continuous functions from X to [0, 1] to assign numerical coordinates to
the points of X and obtain an embedding of X into Rω . From this we see
that every second-countable normal space is a metric space.
(2) Tietze Extension Theorem (Section 35) (This theorem is a useful
technical tool, rather than a big conceptual climax like the Metrization
theorem.) Suppose A is a subset of a space X and f : A → [0, 1] is a
continuous function. If X is normal and A is closed in X, then we can find
a continuous function from X to [0, 1] that is an extension of f .
(3) Embedding manifolds in RN (Section 36) A space X is called a
topological n-manifold if each point x ∈ X has an open neighborhood U(x)
such that U is homeomorphic to an open n-ball.
Usually people insist that X be Hausdorff, sometimes 2nd-countable,
sometimes “paracompact” [defined later in the text]. Right now our text
assumes X is Hausdorff and has a countable basis, so we will assume that as
part of the definition of “manifold”. The spaces Rn , S n , any open subsets of
Rn are n-manifolds; the cartesian product of an n-manifold with a
k-manifold is a (n+k)-manifold; so there are lots of manifolds. If f : Rn → R
is a smooth function [see 22M:133], then for almost all a ∈ R, the solution
set W = {w ∈ Rn : f (w) = a} is a (n-1)-manifold. So solution sets of
equations often are manifolds. What about the other way around? Is every
manifold actually realizable as a solution set in some Rn ? Maybe an easier
question... Is every manifold homeomorphic to a subspace of some Rn ?
Using Urysohn’s lemma to develop the tool called partitions of unity, we
obtain the following theorem.
Each compact n-manifold is homeomorphic to a subspace of some RN .
c Simon, all rights reserved

J. page 1
Of course the “N” depends on X; it is impossible to embed a
high-dimensional manifold in a low-dimensional RN ; in fact, generally
N n. The embedding theorem in our text does not attempt to control the
dimension N; however, the machinery of smooth manifolds (22M:133 or
22M:200) allows one to embed each compact n-dimensional smooth manifold
in R2n .
Proof of Urysohn’s Lemma. The text gives a detailed proof. So this discussion
offers a different approach to some parts, and includes more comments and
motivation for what we do. But the actual proof is close to the text’s.
Somehow, we have to associate to each point x ∈ X a number, f (x). Furthermore,
our function f has to be continuous [otherwise the proof would be trivial and the
theorem would have no meaningful content], send set A to 0, and B to 1. All we
know about X is the hypothesis that X is normal. Here is the plan: We are going
to define a certain [large] collection of open sets in X; then we will decide for each
x ∈ X what f (x) should be by looking at which of these open sets do, or do not,
contain x.
Let D be the set of dyadic rationals in [0, 1], that is D = {0, 1, 12 , 14 , 43 , 81 , 83 , 84 , 78 , . . . }.
We will construct a sequence of open sets Uq in X, indexed by q ∈ D.
First, let U1 = X. Now we get to the real work:
Since X is normal, there exist disjoint open neighborhoods U(A) and V (B). Note
that the existence of V disjoint from U tells us that Ū ∩ B = ∅, i.e. Ū ⊆ (X − B).
Let U0 be this neighborhood U(A). For all the subsequent sets Uq that we will
define, we will have A ⊆ Uq ; and for all q < 1, Uq ∩ B = ∅ .
The closed set Ū0 is contained in the open set X − B. Since X is normal, there
exists an open set (call it U1/2 ) such that
Ū0 ⊆ U1/2 ⊆ Ū1/2 ⊆ (X − B) .
We continue inductively: interpolate U1/4 between U0 and U1/2 ; interpolate U3/4
between U1/2 and X − V ; then define U1/8 , U3/8 , etc.
We get a sequence of open sets Uq such that
(1) For each q ∈ D, A ⊆ Uq .
(2) B ⊆ U1 and for each q < 1, B ∩ Uq = ∅.
(3) For each p, q ∈ D with p < q, we have Ūp ⊆ Uq .
We now define f : X → [0, 1] by
f (x) = inf {q | x ∈ Uq }
for each x ∈ X.

J. page 2
The function f is defined because each point of X is contained in some Uq , at least
in U1 = X. By condition (1), f is 0 on set A. And by condition (2), f is identically
1 on set B. (Note: We are not claiming that f is 0 only on A, or 1 only on B. In
general, the zero set and “1-set” of f will be larger than just A and B.) It remains
to show that f is continuous. This is Step 4 in the text’s proof.
We first establish two mini-lemmas:
A. If f (x) > q then x ∈
/ Ūq .
B. If f (x) < q then x ∈ Uq .
For each x ∈ X, let D(x) = {q | x ∈ Uq }. So f (x) = inf D(x). The numbers q and
the sets Uq are ordered the same way. So if q ∈ D(x) and q ′ > q, then q ′ ∈ D(x).
The infimum f (x) might be the smallest element of D(x) or it might be a lower
limit point that it not itself in D(x).
Proof of A. If f (x) > q then there must be some gap between q and D(x); in
particular, there exists some q ′ such that q < q ′ < f (x). But q ′ < f (x) =⇒ x ∈
/ Uq′ ,
and then Ūq ⊆ Uq′ =⇒ x ∈ / Ūq .
Proof of B. If f (x) < q then there exists q ′ ∈ D(x) such that f (x) < q ′ < q, in which
case q ∈ D(x), so x ∈ Uq .
We now can show that f is continuous. We need to show that the pre-image of each
sub-basic set (a, 1] or [0, a) is open in X. Suppose first that f (x) ∈ (a, 1]. Pick some
q with a < q < f (x). We claim that the open set W = X − Ūq is a neighborhood of
x that is mapped by f into (a, 1]. First, by (A), f (x) > q =⇒ x ∈ W , so W is a
neighborhood of x. If y is any point of W , then f (y) must be ≥ q > a; otherwise, if
f (y) < q, then, by (B), y ∈ Uq ⊆ Ūq .
The argument is simpler for f −1 [0, b). Suppose f (x) < b and pick q such that
f (x) < q < b. By (B), x ∈ Uq . We claim that the neighborhood Uq is mapped by f
into [0, b). Suppose y is any point of Uq . Then q ∈ D(y), so f (y) ≤ q < b.
HOMEWORK in Section 33
Due Wednesday Dec 12
Page 212
#1 (i.e. do you understand the proof of Urysohn’s Lemma)
#2 (cute - see sample Final Exam problems below for another way to ask the same
questions)
#3 (practice with metrics)
#7 (a review of LCH spaces. To warm up for this, you might first prove that a
compact Hausdorff space is completely regular. Also remember our old friend the
“pasting lemma”.)

J. page 3
Problem. (Sample problem for final exam)
Write a one-page {summary, outline, sketch} of a proof of Urysohn’s Lemma.
Problem. (Sample problem for final exam)

a. Suppose X is a countable space. Prove X has the Lindelöf property.
b. There exists [free gift, not done in our class] a space X that is countable,
Hausdorff, and connected. Use this space to show that the theorem “regular
+ Lindelöf =⇒ normal” cannot be modified to say that “Hausdorff +
Lindelöf =⇒ regular.
(end of handout)

J. page 4

Urysohn's Lemma: 22M:132 Fall 07 J. Simon

Uploaded by

Copyright:

Available Formats

Urysohn's Lemma: 22M:132 Fall 07 J. Simon

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Urysohn's Lemma: 22M:132 Fall 07 J. Simon

Uploaded by

Copyright:

Available Formats

22M:132

Introduction. Saying that a space X is normal turns out to be a very strong

c Simon, all rights reserved

c Simon, all rights reserved

c Simon, all rights reserved

Problem. (Sample problem for final exam)

c Simon, all rights reserved

You might also like