MAT 544 Real Analysis I Jason Starr

Stony Brook University Fall 2011

Problem Set 2, Due Thurs. 09/15/2011

MAT 544 Problem Set 2 Solutions

Problems.
Problem 1 A metric space is separable if it contains a dense subset which is finite or countably
infinite. Prove that every totally bounded metric space X is separable. Also give an example of a
bounded, separable metric space which is not totally bounded.
Problem 2 Recall from lecture that for a sequence (xn )n∈Z>0 with no convergent subsequence, the
infinite set S = {xn |n ∈ Z>0 } has a covering by balls (Br(x) (x))x∈S such that every Br(x) ∩ S = {x}.
(a) Prove that the covering (Br(x)/2 (x))x∈S has the same property as above, and further for every
pair of distinct points x, y of S, the balls Br(x)/x (x) and Br(y)/2 (y) are disjoint in X.
(b) Define a continuous function f : X → R which equals 0 on the complement of all of the balls
Br(x)/2 (x) and which is unbounded.
(c) Conclude that a metric space X is sequentially compact if and only if every continuous function
f : X → R is bounded.
Problem 3 Theorem 4.5.1 on p. 211 of the textbook proves that a continuous function f : X → Y
is uniformly continuous if X is compact. Give a second proof by applying Lemma 4.5.3 to the
inverse images of -balls of Y .
Problem 4 For a metric space X, recall the definition of the associated metric space,

C ∞ (X, R) := {f : X → R|f continuous and bounded },

kf k := least upper bound (|f (x)|)x∈X , d(f, g) := kf − gk.

Prove that for bounded, continuous functions (fn : X → R)n∈Z>0 and f : X → R, the sequence
(fn ) converges to f in this metric space if and only if the sequence of functions converges uniformly
to f . This justifies the names “uniform norm” and “uniform metric”.
Problem 5 Let X be a metric space and let (fn : X → R)n∈Z>0 be an equicontinuous sequence
of bounded, continuous functions which converge pointwise to the zero function. Recall this means
that for every  > 0 and for every x ∈ X, there exists δ = δ(x, ) > 0 such that dX (y, x) < δ implies
for all n ∈ Z>0 that |fn (y) − fn (x)| < .
(a) If X is compact, prove that the sequence converges uniformly.

1
MAT 544 Real Analysis I Jason Starr
Stony Brook University Fall 2011
Problem Set 2, Due Thurs. 09/15/2011

Hint. Show that for every  > 0 and for every x ∈ X, there exists a δ = δ(x, ) > 0 and a positive
integer N = N (x, ) such that dX (y, x) < δ implies for all n ≥ N that |fn (y)| < . Then apply
Theorem 4.5.3 to this covering of X by δ-balls.
(b) Let (xn )n∈Z>0 be a sequence in X with no convergent subsequence. With notation as above, find
an equicontinuous sequence of bounded, continuous functions (fx : X → R)x∈S such that fx is zero
on the complement of Br(x)/2 (x), and thus the sequence (fx )x∈S automatically converges pointwise
to the zero function, and yet the functions do not converge uniformly to the zero function.
Solutions to Problems.
Solution to (1) Let  > 0 be a real number. A subset S of a metric space (X, dX ) is -dense for
every x ∈ X, there exists y ∈ S with d(x, y) < . Note that if S is -dense, then for every real
δ > , also S is δ-dense. Also if S contains a subset Y , and if Y is -dense, then also S is -dense –
the element y ∈ Y in the definition above is also an element of S. A subset S is dense if for every
real number  > 0 it is -dense. Because every real number  > 0 is larger than some fraction 1/k
with k a positive integer, it is equivalent to say that S is (1/k)-dense for every positive integer k.
A metric space (X, dX ) is totally bounded if for every real number  > 0, there exists a finite subset
X ⊂ X which is -dense, i.e., for every x ∈ X there exists y ∈ X such that dX (x, y) < . As above,
it is equivalent to say that for every positive integer k, there exists a finite subset X1/k ⊂ X which is
(1/k)-dense. Assume that (X, dX ) is totally bounded. Using the countable axiom of choice, choose
one such finite subset X1/k for each positive integer k. Define S to be the union of these countably
many sets,
S = ∪∞k=1 X1/k .

Since every set X1/k is finite, the union S is either finite or countably infinite. And for every positive
integer k, S is (1/k)-dense since already the finite subset X1/k of S is (1/k)-dense. Therefore S is
dense.
The converse does not hold, essentially because a countably infinite set S which is (1/k)-dense may
contain no finite subset Y which is also (1/k)-dense, e.g., the set of integers inside R is 1-dense but
contains no finite subset which is 1-dense. There do exist separable, bounded metric spaces which
are not totally bounded. The closed unit ball in any infinite dimensional, separable Banach space
gives an example (for more on this, see Problem Set 3). Here is an even easier example that uses
the same trick from the solution to Problem 3 from the online solutions to Problem Set 1. Recall
that given a metric space (X, dX ) and a positive real number , we define a new function

d0X : X × X → R, d0X (x, y) := min(dX (x, y), ).

It is straightforward to check that this is again a metric. Also the topology induced by dX equals
the topology induced by d0X . In particular, a subset S ⊂ X is dense for dX if and only if it is
dense for d0X , and thus X is separable for dX if and only if it is separable for d0X . Moreover, total
boundedness is equivalent to the existence of finite subsets of X which are δ-dense for all real
numbers δ > 0 which are sufficiently small (in fact, just for a sequence of such numbers δ which
converges to 0). And for δ ≤ , a subset of X is δ-dense for dX if and only if it is δ-dense for d0X .

2
MAT 544 Real Analysis I Jason Starr
Stony Brook University Fall 2011
Problem Set 2, Due Thurs. 09/15/2011

Thus X is totally bounded for dX if and only if it is totally bounded for d0X . (In fact, more generally
separableness and total boundedness are preserved upon replacing the metric by any “equivalent”
metric, i.e., another metric such that the identity map is bi-Lipschitz).
However one property of the metric space is not the same for both dX and d0X : whether or not
(X, dX ) is bounded, the metric space (X, d0X ) is always bounded, in fact equal to the 2-ball about
any element. So for a metric space (X, dX ) that is separable yet not totally bounded, the metric
space (X, d0X ) is separable and bounded yet not totally bounded. Now R with its standard metric,
dR (s, t) = |t − s|, is a metric space which is separable since the countably infinite subset Q is dense
in R. Yet R is not totally bounded since it is not even bounded. Therefore (R, d0R ) is a metric space
which is separable and which is not totally bounded, yet which is bounded.
Solution to (2) First of all, for an element x ∈ S, let us recall the proof that there exists r(x) > 0
with Br(x) (x) ∩ S = {x}. By way of contradiction, suppose for every integer k > 0, the ball
B1/k (x) contains an element xn of the sequence different from x, i.e., 0 < dX (x, xn ) < 1/k. We
will construct a subsequence (xmr )r=1,2,... which converges to x. Define m1 to be an integer n as
above so that 0 < d(x, xm1 ) < 1/1. And by way of induction, assume we have constructed integers
m1 < m2 < ... < mr−1 such that for k = 1, . . . , r − 1 we have 0 < dX (x, xmk ) < 1/k. Let  equal the
minimum of 1/(r − 1) and the finitely many positive real numbers dX (x, xn ) with n = 1, . . . , mr−1
and with xn 6= x. By the Archimedean property, there exists an integer k such that 1/k < . And
by hypothesis, there exists an integer mr such that 0 < dX (x, xmr ) < 1/k. Note, in particular, that
1/k < 1/(r − 1) so that k ≥ r. And since 0 < dX (x, xmr ) < dX (x, xn ) for all n = 1, . . . , mr−1 with
dX (x, xn ) > 0, it follows that mr equals no integer n with n = 1, . . . , mr−1 , i.e., mr−1 < mr . So by
induction, and by the countable axiom of choice, there exists a subsequence (xmr )r=1,2,... converging
to x. This contradicts that the sequence contains no convergent subsequence. Therefore for every
s ∈ S there exists a real number r(x) > 0 such that Br(x) (x) ∩ S equals {x}.
Solution to (a). This follows from the triangle inequality. By way of contradiction, assume that
there exists z ∈ Br(x)/2 (x) ∩ Br(y)/2 (y). Then by the triangle inequality we have
r(x) r(y) r(x) + r(y)
dX (x, y) ≤ dX (x, z) + dX (z, y) < + = .
2 2 2
By trichotomy for real numbers, either r(y) ≤ r(x) or r(x) ≤ r(y), i.e., either (r(x) + r(y))/2 is
≤ r(x) or ≤ r(y). Thus either dX (x, y) < r(x) or dX (x, y) < r(y). The first conclusion contradicts
that Br(x) (x) ∩ S equals {x} and the second conclusion contradicts that Br(y) (y) ∩ S equals {y}.
Either way there is a contradiction. Therefore B≤r(x)/2 (x) and B≤r(y)/2 (y) are disjoint. Form the
subset
U := ∪x∈S Br(x)/2 (x).
Since it is a union of open balls, U is an open set. And by the argument above, it is a disjoint
union
U = tx∈S Br(x)/2 (x).

Solution to (b) For every element x in S, let n(x) denote the smallest integer n such that xn
equals x. And define δ(x) = min(r(x)/3, 1/n(x)). Since δ(x) < r(x)/2, the closed ball B≤δ(x) (x) is

3
MAT 544 Real Analysis I Jason Starr
Stony Brook University Fall 2011
Problem Set 2, Due Thurs. 09/15/2011

contained in the ball Br(x)/2 (x). So by the solution to (a), for x, y ∈ S with x 6= y, B≤δ(x) (x) and
B≤δ(y) (y) are disjoint, i.e., the following union is a disjoint union

C = ∪x∈S B≤δ(x) (x) = tx∈S B≤δ(x) (x).

Claim 0.1. The subset C of X is closed, i.e., for every sequence (zm )m=1,2,... in C which converges
to a limit z∞ ∈ X, the limit z∞ is in C.

Claim 0.1 will follow from a second claim.

Claim 0.2. For every sequence (zm )m=1,2,... in X (not necessarily entirely in C), the subset {zm |m =
1, 2, . . . } intersects only finitely many of the closed balls B≤δ(x) (x).

Claim 0.2 is proved by contradiction. If the sequence intersects infinitely many balls, then this gives
a subsequence (zmk ) each of which is contained in a ball Bδ(xnk ) (xnk ) for a distinct xnk . But since
the radii δ(x) limit to 0, it follows that the sequences (zmk )k=1,2,... and (xnk )k=1,2,... are equivalent in
the sense of equivalence of Cauchy sequences. Since (zmk )k=1,2,... converges to z∞ , the same holds
for (xnk )k=1,2,... . But this contradicts the hypothesis that no subsequence of (xn )n=1,2,... converges.
Therefore, by contradiction, the set {zm |m = 1, 2, . . . } intersects only finitely many of the balls
Bδ(x) (x), i.e., Claim 0.2 is proved. Next suppose that (zm )m=1,2,... is a sequence entirely contained
in C, i.e., every zm is contained in some ball B≤δ(x) (x). By Claim 0.2, the sequence intersects only
finitely many balls. So there must be a single ball B≤δ(x) (x) which contains an infinite subsequence
(zmk )k=1,2,... . The subsequence (zmk ) still converges to z∞ . And by Problem 3 from Problem Set 1,
the set B≤δ(x) (x) is closed. Therefore the limit z∞ is contained in B≤δ(x) (x), and thus also contained
in C. This proves Claim 0.1, C is closed.
Since C is closed, its complement V := X \ C is open. Observe that the intersection of the open
sets U and V is the disjoint union


U ∩ V = tx∈S Br(x)/2 (x) \ B≤δ(x) (x) .

This brings us to an elementary, but nonetheless fundamental, fact about continuous functions.
Let (A, dA ) and (B, dB ) be metric spaces (or even just topological spaces). Let (Ai )i∈I be an open
covering of A. For every i ∈ I, let fi : Ai → B be a function. Assume that for every i, j ∈ I, the
restrictions of fi and fj to Ai ∩ Aj are equal. Then there exists a unique function f : A → B whose
restriction to every open subset Ai equals fi .
Claim 0.3. The function f : A → B is continuous if and only if for every i ∈ I the restriction to
the open subset Ai , fi : Ai → B, is continuous (where Ai has the topology induced from A).

Of course the restriction of a continuous function to a subset is continuous. So if f is continuous,

then so is every restriction fi . Conversely suppose that every fi is continuous. To prove that f is
continuous, we must prove that for every open W ⊂ B, the preimage f −1 (W ) is open in A. Since
A equals the union ∪i∈I Ai , also f −1 (W ) equals ∪i∈I (f −1 (W ) ∩ Ai ), i.e., it equals ∪i∈I fi−1 (W ). For

4
MAT 544 Real Analysis I Jason Starr
Stony Brook University Fall 2011
Problem Set 2, Due Thurs. 09/15/2011

every i ∈ I, since fi is continuous, fi−1 (W ) is open in Ai . And also Ai is open in A. Therefore

fi−1 (W ) is open in A. And every union of open subsets is an open subset. Therefore ∪i∈I fi−1 (W )
is open in A, which proves Claim 0.3.
Because of Claim 0.3, to construct a continuous function f : X → R, it is equivalent to construct
continuous functions fU : U → R and fV : V → R which are equal on U ∩ V . Similarly, since U
is the disjoint union of the opens Br(x)/2 (x), to construct a continuous function fU : U → R, it is
equivalent to construct for every x ∈ S, a continuous function fx : Br(x)/2 (x) → R. First of all,
define fV : V → R to be identically zero. This is trivially continuous. Next for every x ∈ S, define
fx : Br(x)/2 (x) → R by
  
1
 n(x) 1 − δ(x)
 dX (x, z) , dX (z, x) < δ(x),
fx (z) = 0, dX (z, x) = δ(x),

0, δ < dX (z, x) < r(x)/2


Observe that on the closed ball B≤δ(x) , fx is expressed as n(x)(1 − dX (x, z)/δ(x)). Since the metric
function dX is continous, fx is continuous on the closed ball B≤δ(x) (x). And on the (relatively)
closed subset Br(x)/2 (x) \ Bδ(x) (x) of Br(x)/2 , fx is the constant function 0, which is also continuous.
And on the intersection of these sets, i.e., on the subset {z ∈ Br(x)/2 (x)|dX (x, z) = δ(x)}, the
functions agree.
This brings us to the closed counterpart of Claim 0.3. As before, let (A, dA ) and (B, dB ) be metric
spaces (or topological spaces). Let J be a finite index set, and let (Cj )j∈J be a collection of closed
subsets of A such that A = ∪Cj . For every j ∈ J, let fj : Cj → B be a function. Assume that for
every i, j ∈ I, the restrictions of fi and fj to Ci ∩ Cj are equal. Then there exists a unique function
f : A → B whose restriction to every subset Cj equals fj .
Claim 0.4. The function f : A → B is continuous if and only if for each of the finitely many
elements j ∈ J the restriction to the closed subset Cj , fj : Cj → B, is continuous (where Cj has
the topology induced from A).

As before, if f is continuous then its restriction to every subset is continuous, so each fj is continu-
ous. Conversely suppose that every fj is continuous. To prove that f is continuous, we must prove
that for every closed subset D ⊂ B, the preimage f −1 (D) is closed in A. Since A equals the union
∪j∈J CC , also f −1 (D) equals ∪j∈J (f −1 (D) ∩ Cj ), i.e., it equals ∪j∈J fj−1 (D). For every j ∈ J, since
fj is continuous, fj−1 (D) is closed in Cj . And also Cj is closed in A. Therefore fj−1 (D) is closed in
A. And every finite union of closed subsets is a closed subset. Therefore ∪j∈J fj−1 (D) is closed in
A, which proves Claim 0.3.
In particular, applying Claim 0.4 to the pair of (relatively) closed subsets B≤δ(x) (x) and Br(x)/2 (x) \
Bδ(x) (x) of Br(x)/2 (x), it follows that fx is continuous on Br(x)/2 . Since U is the disjoint union
over x ∈ S of the open sets Br(x)/2 , there is a unique function fU : U → R whose restriction to
every subset Br(x)/2 (x) equals fx . And by Claim 0.3, the function fU is continous on U . And the
restriction of both fU and fV to


U ∩ V = tx∈S Br(x)/2 (x) \ B≤δ(x) (x)

5
MAT 544 Real Analysis I Jason Starr
Stony Brook University Fall 2011
Problem Set 2, Due Thurs. 09/15/2011

is the zero function. So the restrictions are equal on U ∩ V . Therefore there is a unique function
f : X → R whose restrictions to U and V are fU and fV respectively. And by Claim 0.3 once more,
f is a continuous function.
The final claim is that f is unbounded. Indeed, for every x ∈ S, f (x) = fx (x) = n(x). Since the
set S is infinite, the set of distinct, positive integers {n(x)|x ∈ S} is also infinite, hence unbounded.
So f : X → R is a continuous, unbounded function on X.
Solution to (c) As proved in lecture, the image of a compact metric space under a continuous
map is compact. By the Heine-Borel theorem, the compact subsets of R are precisely the closed,
bounded subsets of R. Hence a continuous function from a compact metric space to R has image
equal to a closed, bounded subset of R, and thus the function is bounded. On the other hand, the
example above proves that on every non-compact metric space there exists a continuous, unbounded
function to R. Therefore a metric space is compact if and only if every continuous function from
the metric space to R is bounded.
Solution to Problem 3 Let (X, dX ) be a metric space. Let f : X → R be a continuous function.
Let 0 > 0 be a real number. For every x ∈ X, the ball B0 (f (x)) is an open subset of R. Since f is
continuous, the inverse image Uf,0 ,x := f −1 (B0 (f (x))) is an open subset of X containing x. Thus
the collection (Uf,0 ,x )x∈X is an open covering of X. By Lemma 4.5.3, if X is compact then there
exists a real number δ > 0 such that for every y ∈ X, there exists x ∈ X with Bδ (y) ⊂ Uf,0 ,x . For
every z ∈ Bδ (y), since z is in Uf,0 ,x , we have |f (z) − f (x)| ≤ 0 . By the triangle inequality we have

|f (z) − f (y)| ≤ |f (z) − f (x)| + |f (y) − f (x)| < 0 + 0 = 20

Therefore, choosing 0 = /2, it follows that for every pair of elements y, z ∈ X with dX (y, z) < δ,
|f (z) − f (y)| < . Since this holds for every  > 0, f is uniformly continuous.
Solution to Problem 4 This is simply a matter of parsing the definitions carefully. No solution
will be written up.
Solution to Problem 5
Solution to (a) Fix a real number  > 0 and an element x ∈ X. Since (fn (x)) converges to 0,
there exists an integer N = N (x, ) such that n ≥ N implies that |fn (x)| < /2. Since the sequence
(fn )n≥N is equicontinuous, there exists a real number δ = δ(x, ) > 0 such that for all y ∈ Bδ(x,) (x),
|fn (y) − fn (x)| < /2. Then by the triangle inequality,

|fn (y)| ≤ |fn (y) − fn (x)| + |fn (x)| < /2 + /2 = .

For fixed  > 0, the collection of balls {Bδ(x,) (x)|x ∈ X} is an open covering of X. Since X is
topologically compact, there is a finite subcover, i.e., there is a finite subset X ⊂ X such that
the finitely many sets {Bδ(x,) (x)|x ∈ X } cover X. Denote by N be the maximum of the positive
integers N (x, ) as x varies in the finite set X . For every y ∈ X, there exists x ∈ X such that
y ∈ Bδ(x,) (x). Thus, by the inequality above, for all n ≥ N (x, ), |fn (y)| < . In particular, this
holds for all n ≥ N . So for every n ≥ N , for every y ∈ X, |fn (y)| < . By definition of the uniform

6
MAT 544 Real Analysis I Jason Starr
Stony Brook University Fall 2011
Problem Set 2, Due Thurs. 09/15/2011

norm, kfn k ≤ . Thus for every real number  > 0, there exists an integer N > 0 such that n ≥ N
implies that kfn k ≤ . This precisely says that (fn ) converges to 0 in the uniform metric.
Solution to (b) Notations here are as in the solution to Problem 2. For every x ∈ S, define
gn(x) to be the function which agrees with f on the ball Br(x)/2 (x) and which equals 0 outside of
B≤δ(x) (x). And for integer n > 0 which are not of the form n(x), define gn to be identically zero.
The function gn(x) is continuous by the same arguments as in the solution to Problem 2. Notice
that gn(x) is nonzero only on Bδ(x) (x). By Problem 2(a), for every y ∈ X, there exists at most one
x ∈ S such that y ∈ Br(x)/2 (x). And if so, then for all n > n(x), gn is zero on y. Thus (gn (y))n=1,2,...
converges to 0, i.e., the sequence (gn ) converges pointwise to 0.
For essentially the same reason, (gn ) is also equicontinuous. Let  > 0 be a real number. Let y be
an element of X. We need to prove that there exists δ = δ(y, ) such that for every n = 1, 2, . . . ,
for every z ∈ Bδ(y,) (y), |gn (z) − gn (y)| < . There are two cases depending on where y lies. If y
is in U , then since U is open, there exists δ = δ(y) such that Bδ(y) (y) is in U . Thus every gn is
identically zero on Bδ(y) (y). So for every n = 1, 2, . . . , for every z ∈ Bδ(y) (y), |gn (z)−gn (y)| = 0 < .
Next, if y is in some subset Br(x)/2 (x), then since Br(x)/2 (x) is open, there exists δ 0 = δ(y) such
that Bδ0 (y) ⊂ Br(x)/2 (x). Hence for x0 ∈ S with x0 6= x, by Problem 2(a), Bδ0 (y) (y) is disjoint
from Br(x0 )/2 (x0 ). So for n 6= n(x), gn is identically zero on Bδ0 (y) (y). Thus for n 6= n(x) and for
z ∈ Bδ0 (y) (y), |gn (z) − gn (y)| = 0 < . Finally, since gn(x) is continuous at y, there exists δ 00 (y, ) > 0
such that z ∈ Bδ00 (y,) (y) implies that |gn(x) (z) − gn(x) (y)| < . Taking δ(y, ) = min(δ 0 (y), δ 00 (y, ),
for every n = 1, 2, . . . , and for every z ∈ Bδ(y,) (y), |gn (y) − gn (z)| < . Thus in both cases, (gn ) is
equicontinuous at y. So the sequence is equicontinuous.
So (gn ) is an equicontinuous sequence of functions which converges to 0 pointwise. Yet the sequence
does not converge to 0 uniformly. Indeed gn(x) (x) equals n(x), so that kgn(x) k ≥ n(x). Since the
integers n(x) are unbounded, the sequence (kgn k) is unbounded, hence does not converge to 0.