Math 535 - General Topology

Fall 2012
Homework 5 Solutions

Problem 1. Let X be a first-countable topological space, (xn )n∈N a sequence in X, and y ∈ X

a cluster point of this sequence. Show that there is a subsequence (xnk )k∈N that converges to y.

Solution. Since X is first-countable, the point y has a countable neighborhood basis V1 ⊇

V2 ⊇ . . ..
Since y is a cluster point of (xn ), there is an index n1 ≥ 1 satisfying xn1 ∈ V1 . Then there is an
index n2 > n1 satisfying xn2 ∈ V2 . Repeating the argument, one can find an index nk > nk−1
satisfying xnk ∈ Vk for all k ∈ N.
We claim that the subsequence (xnk )k∈N converges to y. Let V be a neighborhood of y. By
definition of neighborhood basis, there is some m ∈ N satisfying Vm ⊆ V . For all index k ≥ m,
we have
xnk ∈ Vk ⊆ Vm ⊆ V
which proves convergence xnk → y.

Remark. This shows that a first-countable compact space is always sequentially compact.

1
Problem 2. Consider the discrete space with two elements {0, 1} and consider the space

X := {0, 1}P(N) ∼
Y
= {0, 1}
S∈P(N)

with the product topology. Here P(N) denotes the power set of N, i.e. P(N) = {S | S ⊆ N} is
the set of all subsets of N.
One can view X as the set of all functions f : P(N) → {0, 1}. In this viewpoint, the canonical
projection pS : X  {0, 1} corresponds to evaluation at S, i.e. sending the function f ∈ X to
f (S) ∈ {0, 1}.
For each n ∈ N, consider the element fn ∈ X whose components are
(
1 if n ∈ S
fn (S) =
0 if n ∈
/ S.

In other words, fn is the function “Is n in there?”

Show that the sequence (fn )n∈N in X has no convergent subsequence.

Solution. Given any subsequence (fnk )k∈N , we will show that it does not converge by finding
a subset S ⊆ N such that the sequence (fnk (S))k∈N in {0, 1} does not converge. Indeed, the
projection pS : X  {0, 1} is continuous and therefore sends convergent sequences to convergent
sequences.
Take the set S = {n2 , n4 , n6 , . . .} = {n2k | k ∈ N} ⊂ N. Then the sequence pS (fnk ) = fnk (S) is
(0, 1, 0, 1, 0, 1, . . .), more precisely
(
1 if k is even
fnk (S) =
0 if k is odd

because the numbers nk are all distinct.

The sequence (fnk (S))k∈N in {0, 1} does not converge, since the only convergent sequences in
the discrete space {0, 1} are the sequences that are eventually constant.

Remark. By Tychonoff’s theorem, X is compact, so that the sequence (fn )n∈N has a cluster
point, even though it has no convergent subsequence. Thus X is an example of compact space
which is not sequentially compact.

2
Problem 3. Let X be a topological space and consider the set of all continuous functions on
X with values in [0, 1]
C := {f : X → [0, 1] | f is continuous}.
Consider the set [0, 1]C ∼
Q
= f ∈C [0, 1] of all functions from C to [0, 1], endowed with the product
topology. Consider the map

ϕ : X → [0, 1]C
x 7→ (f (x))f ∈C

so that ϕ(x) is “evaluation at x”.

a. Show that ϕ is continuous.

Solution. A map into a product is continuous if and only if all of its components are conti-
nuous. For any index f ∈ C, the f th component of ϕ is pf ◦ ϕ = f

ϕ pf
X / C
[0, 1] / / [0, 1]
6

pf ◦ϕ=f

which is continuous, by definition of C.

b. Show that the closure of the image ϕ(X) ⊂ [0, 1]C is a compact Hausdorff space. (Feel free
to assume the axiom of choice!)

Solution. Since [0, 1] is Hausdorff, the product [0, 1]C is Hausdorff, and so is the subspace
ϕ(X) ⊂ [0, 1]C .
Since [0, 1] is compact, the product [0, 1]C is compact, by Tychonoff’s theorem. Since ϕ(X) is
closed in [0, 1]C , it is itself compact.

c. Show that ϕ is injective if and only if points of X can be separated by functions, i.e. for
any distinct points x, y ∈ X, there is a continuous function f : X → [0, 1] satisfying f (x) = 0
and f (y) = 1.
This property is sometimes called functionally Hausdorff or completely Hausdorff.

Solution. Consider the equivalent statements

ϕ : X → [0, 1]C is injective.
⇔ For any distinct points x, y ∈ X, we have ϕ(x) 6= ϕ(y), i.e. there is some index f ∈ C such
that the f th components ϕ(x)f = f (x) and ϕ(y)f = f (y) are distinct.
⇔ For any distinct points x, y ∈ X, there is a continuous function f : X → [0, 1] satisfying
f (x) 6= f (y) (WLOG f (x) = 0 and f (y) = 1, rescaling the function if needed).

3
d. While we’re at it, show that a functionally Hausdorff space is always Hausdorff.

Solution. Let x, y ∈ X be distinct points. Pick a continuous function f : X → [0, 1] that

separates x and y, say f (x) = 0 and f (y) = 1. Then f −1 [0, 31 ) and f −1 ( 32 , 1] are open in X
(since f is continuous), disjoint, and satisfy x ∈ f −1 [0, 31 ) and y ∈ f −1 ( 32 , 1].

4
Problem 4. (Willard Exercise 17F.1) A topological space X is countably compact if every
countable open cover of X admits a finite subcover. (In particular, compact always implies
countably compact, but not the other way around in general.)
Show that X is countably compact if and only if every sequence in X has a cluster point.
Hint: Recall that compactness can be described in terms of closed sets. Countable compactness
has a very similar description in terms of closed sets, which could be useful here.

Solution. (⇒) Assume X is countably compact and let (xn )n∈N be a sequence in X. Consider
the n-tail of the sequence
Tn := {xk | k ≥ n} ⊆ X
and note that the collection of all tails {Tn }n∈N has the finite intersection property, therefore
T does the collection of their closures {Tn }n∈N . Since X is countably compact, the intersection
so
n∈N Tn 6= ∅ contains a point y.

We claim that y is a cluster point of the sequence. Let V be a neighborhood of y and N ∈ N

any index. The condition y ∈ TN guarantees that there is some xn ∈ TN ∩ V , i.e. some point
xn ∈ V of the sequence, with index n ≥ N .
(⇐) Let {Cn }n∈N be a countable collection of closed subsets of X with the finite intersection
property. For every n ∈ N, pick a point
n
\
xn ∈ Ci 6= ∅
i=1

and let y ∈ X be a cluster point of the sequence (xn )n∈N .

T
We claim y ∈ n∈N Cn . Let V be a neighborhood of y. Since y is a cluster point of the
sequence, for all index n ∈ N, there is some index m ≥ n satisfying xm ∈ V . The inequality
m ≥ n guarantees
\m
xm ∈ Ci ⊆ Cn
i=1

T yields xm ∈ V ∩ Cn 6= ∅. We conclude y ∈ Cn = Cn , which holds for all n ∈ N, i.e.

which
y ∈ n∈N Cn .

5
Problem 5. (Munkres Exercise 27.2) Let X be a metric space and A ⊆ X a subset. Define
the -neighborhood of A as the set

B (A) := {x ∈ X | d(x, A) < }.

a. Show that the -neighborhood of A is

[
B (A) = B (a)
a∈A

i.e. the union of all open balls of radius  around points a ∈ A.

Solution. (⊇) Let a ∈ A. For any x ∈ B (a), we have

d(x, a) <  ⇒ d(x, A) ≤ d(x, a) < 

which
S implies x ∈ B (A). Since this inclusion B (a) ⊆ B (A) holds for all a ∈ A, we obtain
a∈A  (a) ⊆ B (A).
B
(⊆) Let x ∈ B (A), which means d(x, A) < . By definition of d(x, A) = inf a∈A d(x, a) there is
some 0 <  and a0 ∈ A satisfying
d(x, a0 ) = 0 < .
S
This implies x ∈ B (a0 ) ⊆ a∈A B (a).

b. Assume A is compact and let U ⊆ X be an open set containing A. Show that some
-neighborhood of A is contained in U , i.e. B (A) ⊆ U for some  > 0.

Solution. Since U is open, for every a ∈ A, there is some radius (a) > 0 satisfying B(a) (a) ⊆
U . Consider the open cover [
A⊆ B (a) (a)
2
a∈A

of A. Since A is compact, there is an open subcover

A ⊆ B (a1 ) (a1 ) ∪ . . . ∪ B (an ) (an ).

2 2

Take  := min{ (a21 ) , . . . (a2n ) }. For any a ∈ A and x ∈ B (a), the point a is in one of the open
balls B (ak ) (ak ) and we obtain
2

(ak ) (ak ) (ak )

d(x, ak ) ≤ d(x, a) + d(a, ak ) <  + ≤ + = (ak )
2 2 2
which implies x ∈ B(ak ) (ak ) ⊆ U .
Since x was arbitrary, the inclusion B (a) ⊆ U holds. Since a was arbitrary, the inclusion
B (A) ⊆ U holds, using part (a).

6
Alternate solution. In the case U = X, any  > 0 works. In the case U 6= X, the com-
plement U c is a non-empty closed subset, disjoint from A. Consider the function f : A → R
defined by
f (a) = d(a, U c ).
Then f is continuous (c.f. HW 6 Problem 3a). Since A is compact, f is bounded and achieves
its bounds, in particular its lower bound f (a0 ) = m ≥ 0. However, we have m > 0, since f is
strictly positive on A:
f (a) = 0 = d(a, U c ) ⇒ a ∈ U c = U c
but A and U c are disjoint, i.e. A ∩ U c = ∅.
We claim Bm (A) ⊆ U , in other words Bm (A) ∩ U c = ∅. Any point x ∈ U c satisfies

d(a, x) ≥ d(a, U c ) = f (a) ≥ m

for all a ∈ A, which implies d(x, A) ≥ m, i.e. x ∈

/ Bm (A).

7
Problem 6. In all parts of this problem, let f : X → Y be a uniformly continuous map
between metric spaces.

a. Show that f sends Cauchy sequences to Cauchy sequences. In other words, if (xn )n∈N is a
Cauchy sequence in X, show that (f (xn ))n∈N is a Cauchy sequence in Y .

Solution. Given  > 0, we want to find N ∈ N so that d(f (xm ), f (xn )) <  for all m, n ≥ N .
Since f is uniformly continuous, there is a δ > 0 guaranteeing d(f (x), f (x0 )) <  whenever
d(x, x0 ) < δ.
Since the sequence (xn )n∈N is Cauchy, there is some N ∈ N guaranteeing d(xm , xn ) < δ for all
m, n ≥ N .
Putting these two facts together, we conclude d(f (xm ), f (xn )) <  for all m, n ≥ N .

b. Assuming moreover that f is a homeomorphism and Y is complete, show that X is com-

plete.

Solution. Let (xn )n∈N be a Cauchy sequence in X. Then (f (xn ))n∈N is a Cauchy sequence in
Y , since f is uniformly continuous. Since Y is complete, this sequence converges: f (xn ) → y ∈
Y.
Since the inverse f −1 : Y → X is continuous, it sends convergent sequences to convergent
sequences, and we obtain
f −1 (f (xn )) → f −1 (y)
i.e. xn → f −1 (y).

c. Find an example where f is a homeomorphism and X is complete, but Y is not complete.

(Don’t forget to show that your example f is uniformly continuous.)

Solution.
Consider f = arctan : R → − π2 , π2 . Then f is continuous, and its inverse

tan : − π2 , π2 → R is also continuous, i.e. f is a homeomorphism.

Moreover, f is everywhere differentiable and its derivative f 0 (x) = 1+x
1 0
2 is bounded by |f (x)| ≤

1. Therefore f is Lipschitz continuous, and in particular uniformly continuous.

Finally, note that R is complete, whereas the open interval Y := − π2 , π2 is not, since the

sequence yn = π2 − n1 is Cauchy but does not converge in Y .

Remark. Part (b) implies that if two metric spaces are uniformly equivalent, then one is complete
if and only if the other is complete. In other words, completeness depends on more than just
the topology, but at most on the uniform type.