Math 55a, Fall 2004

Fifth Assignment, Solutions

Adapted from Andrew Cotton, George Lee, and Tseno Tselkov

Note: These solutions often use this simple fact proved in class: Suppose
that X is a topological space and that S ⊂ X, x ∈ X. Then x 6∈ clos(S) ⇐⇒
∃ open neighborhood Ux of x such that S ∩ Ux = ∅. (The restriction “open”
may also be removed from this condition.) Also, any open set U contain-
ing a point x is an open neighborhood of that point, because x ∈ U = int(U ).

Problem 1.    
T-2 T-3
• Every compact space X is . (Thus, every compact Haus-
T-3 T-4
dorff space is normal.)

 Suppose we have disjointsets S, T ⊂ X, where S is closed and

T contains one point t ∈ X
. We must prove there are disjoint sets U ⊃ S
T is closed
 
T-2
and V ⊃ T. Because X is , for each s ∈ S there exist open disjoint
T-3
sets Us 3 s and Vs ⊃ T. The {Us } cover S, which is a closed subset of a
compact space and thus compact itself. Thus finitely many of the {Us } —
say, Us1 , Us2 , . . . , Usn — cover S. Then consider the sets
n
[ n
\
U= Usi and V = V si .
i=1 i=1
U is the union of open sets and V is the finite intersection of open sets, so
they are both open themselves. Next, S ⊂ U (since U covers S) and T ⊂ V
(since T is in each Vsi ). And finally, U and V are disjoint — each u ∈ U lies
in some Usi , and thus not in Vsi and not in V. This completes the proof.

• Every metric space (X, d) is normal.

First Solution:
Given two disjoint, non-empty, closed subsets S, T ⊂ X, for each point
s ∈ S there exists an open set Us containing s and disjoint from T. Then in
this open set, there must be an epsilon-ball B(s, s ) ⊂ Us also disjoint from

T. Thus for any t ∈ T, we must have s ≤ d(s, t). Now, write Bs = B s, 2s
and define [
US = Bs .
s∈S
As a union of open balls, US is open. Define the values t and the open
set UT similarly. We claim that US ∩ UT = ∅. Suppose not. Then ∃x
 –2–

such that d(x, s) < 2s and d(x, t) < 2t for some s ∈ S, t ∈ T . But then
d(s, t) ≤ d(x, t) + d(x, s) < min(s , t ) ≤ d(s, t), a contradiction. Thus US
and UT are indeed disjoint open sets containing S and T , respectively.

Second Solution:
Let S and T be two non-empty, disjoint, closed subsets of X. For x ∈ X
define d(x, S) = infs∈S d(x, s). If d(x, S) = 0, then we can find x1 , x2 , . . . in
S such that xk ∈ d(x, k1 ) for k ∈ N. Because S is closed, the limit x of {xk }
must be in S as well. Next, suppose that  > 0 is fixed. Then if d(x, x0 ) < ,
for each s ∈ S we have |d(x, s) − d(x0 , s)| < . Thus
d(x0 , S) ≥ infs∈S (d(x, s) − ) = d(x, S) − 
and similarly d(x, S) ≥ d(x0 , S) − . Therefore by the δ- definition of conti-
nuity, x 7→ d(x, S) is continuous.
Define d(x, T ) analogously; as above, x 7→ d(x, T ) is continuous, and if
d(x, T ) = 0 then x ∈ T.
Now consider the function f : X → R such that
d(x, S)
f (x) = ,
d(x, S) + d(x, T )
This function is defined, because we can never have d(x, S) = d(x, T ) = 0
since S ∩T = ∅. Note that this function assumes the value 0 on S and 1 on T
and is continuous (because the functions d(x, S) and d(x, T ) are continuous,
and because addition and division on the reals are continuous). Then by the
converse of Urysohn’s lemma (which is easier and done in class right before
the lemma) we get that X must be normal.

Problem 2.

We prove the statements slightly out of the order, in order to use (c) in the
proof of (b).

(a) Suppose that a convergent net (S, D, ≥) converged to distinct points x

and y. Because X is Hausdorff, there exists disjoint open neighborhoods
Nx of x and Ny of y. Because the net converges to x, there exists n0 ∈ D
such that S(n) ∈ Nx for all n ≥ n0 . Similarly, there exists n1 ∈ D such
that S(n) ∈ Ny for all n ≥ n1 . Letting n equal the upper bound of the set
{n0 , n1 }, we find that S(n) ∈ Nx ∩ Ny = ∅, a contradiction.

(c) Suppose by way of contradiction that some net (S, D, ≥) in B converges

to a point x 6∈ clos(B). Because x 6∈ clos(B), there exists an open neighbor-
hood Nx of x disjoint from B. Then by the definition of convergence, there
is some n0 ∈ D such that S(n0 ) ∈ Nx ; but since (S, D, ≥) is a net in B, we
have S(n0 ) ∈ B as well. Thus Nx and B are not disjoint, a contradiction.
Therefore our original assumption was false, and no convergent net in B
converges to a point outside clos(B).
 –3–

Now suppose that x ∈ clos(B) — that is, every neighborhood of x contains

some element of B. Let D be the set of these neighborhoods (whether open
or not), where given sets A1 , A2 ∈ D we have A1 ≥ A2 ⇐⇒ A1 ⊂ A2 . Also,
for each A ∈ D there is some point bA ∈ A ∩ B; using the axiom of choice,
for each A we may define S(A) = bA to be one such point. Then (S, D, ≥) is
a net because (D, ≥) is clearly anT order relation, and because given finitely
many subsets {Ai } in D, the set Ai is an upper bound of {Ai }. Then given
any neighborhood Nx of x, for all N ≥ Nx we have S(N ) ∈ N ⊂ Nx so that
(S, D, ≥) converges to x. Thus every point in clos(B) is the limit of some
net in B. This completes the proof.

(b) If B is closed, then from (c) no net in B converges to a point not in

clos(B) = B. If B is not closed then there is a point x ∈ clos(B)\B ⊂ X \B;
so from (c), some net in B converges to x.

(d) Note: We assume that D is nonempty, which is probably what Prof.

Schmid wanted anyways.
Suppose that some subnet (T, E, ≥E ) = (S ◦ F, E, ≥E ) of (S, D, ≥D ) con-
verges to x; we prove that x is a cluster point of (S, D, ≥D ). Suppose we
have d ∈ D and a neighborhood Nx of x. Since (T, E, ≥E ) is a subnet, we
know that there exists k0 ∈ E such that k ≥E k0 ⇒ F (k) ≥D d. And
since (T, E, ≥E ) converges to x, we know there exists k1 ∈ E such that
k ≥E k1 ⇒ T (k) = S(F (k)) ∈ Nx . But by the definition of a net, {k0 , k1 }
has some upper bound k; then letting n = F (k), we have both n ≥ d and
S(n) ∈ Nx . Therefore x is indeed a cluster point of (S, D, ≥D ), as desired.

Now suppose that x is a cluster point of a net (S, D, ≥D ). Let

E = {(U, d) | U a neighborhood of x, d ∈ D s.t. S(d) ∈ U } .
Also let F : E → D be defined by F (U, d) = d, and write (U1 , d1 ) ≥E (U2 , d2 )
iff U1 ⊂ U2 and d1 ≥D d2 .
It’s easily shown that ≥E is reflexive, transitive, and antisymmetric (be-
cause both ⊂ and ≥D are). Furthermore, T given any finite subset
{(U1 , d1 ), . . . , (Un , dn )} ⊂ E, let U = ni=1 Ui and let m ∈ D be an upper
bound of {d1 , . . . , dn } with respect to ≥D . Then since x is a cluster point,
∃d ≥D m ≥D d1 , d2 , . . . , dn such that S(d) ⊂ U, so that (U, d) ∈ E is an
upper bound of {(U1 , d1 ), . . . , (Un , dn )} with respect to ≥E .
Thus, (E, ≥E ) is a directed system and (T, E, ≥E ) = (S ◦ F, E, ≥E ) is
a net in X. We already know T = S ◦ F ; and given n0 ∈ D, then letting
k0 = (X, n0 ) we have (U, d) ≥E k0 ⇒ F (U, d) = d ≥D n0 by the definitions
of F and ≥E . Thus (T, E, ≥E ) is actually a subnet of (S, D, ≥D ). And given
any neighborhood U of x in X, because x is a cluster point ∃d ∈ D such that
S(d) ∈ U. Writing n0 = (U, d), we know that n = (U 0 , d0 ) ≥E n0 ⇒ T (n) ∈
U 0 ⊂ U, so (T, E, ≥E ) converges to x. Therefore if x is a cluster point of a
net, some subnet converges to x. This completes the proof.