Math 5000 Problem set 2 Due: Thursday September 28.

GIVEN EXTENSION IN OFFICE HOURS

1. Let BpSq be the metric space of bounded real-valued functions on the set S with the metric of
uniform convergence:
dpf, gq “ sup |f psq ´ gpsq|
sPS

Show that BpSq is separable if and only if S is finite (Hint: Think about balls centered at indicator
functions of subsets of S).

Solution

BpSq is separable if there exists a countable, dense subset. This means we have at most
countably many bounded, real-valued functions whose closure generates BpSq.

(ð)

Suppose S is finite, meaning we have finitely many input points. Then, we can find a countable
dense subset by choosing all of the functions with rational outputs.

This set of functions is countable because we have finitely many inputs and countably many
outputs for each input, which leads to a countable number of functions.

The set is dense because the rationals are dense in the reals, so our set of all rational outputs
for each input is dense in the set of all real outputs for each input— which corresponds to just the
set of all functions.

(ñ)

Now suppose we have that BpSq is separable, that is, there exists a countable set of real-valued
functions on S whose closure generates BpSq. We want to show that S must be finite.

Consider an indicator function for each subset of S, denoted X defined as follows:

" *
1, if x P X
fX pxq “
0, if x R X

Notice that the distance from fX to any other fX 1 is 1 if the sets aren’t equal (because there
exists some point x P X, x R X 1 where fX “ 1, fX 1 “ 0, and we want the supremum), and 0 if
they are equal.

1
 Consider the ball centered at fX with radius r ă 1: BpfX ; rq, then we have exactly the function
fX . So, for say r “ 12 , each element of the set of balls B “ tBpfX ; 12 q|X Ă Su is disjoint from
the others.

Suppose for contradiction that S is infinite. Since there exists a countable, dense subset, but
each ball in B is disjoint from the others, we can say that there is exactly one unique point from
the dense subset in each ball.

Now, note that because B is the same size as the set of all subsets of an infinite set, B must
have uncountably many elements.

However, this implies that we have an uncountably large subset of a countable subset, which is
impossible, so S must be finite.

2
2. Suppose f is a continuous function from the closed unit interval r0, 1s (with the standard
metric) to itself. Show that there exists a point t P r0, 1s such that f ptq “ t. (Keep in mind that
we haven’t talked about connectedness yet, so you can’t use the intermediate-value theorem).

Solution

Suppose we don’t have a fixed point, so f p0q ą 0 and f p1q ă 1. And let gpxq “ f pxq ´ x.

Then let A “ g ´1 p´1, 0q and B “ g ´1 p0, 1q where g ´1 denotes the pre-image of g.

Note both sets are open because we have the pre-image of an open set mapped by a continuous
function. Also, both sets are nonempty because we know that gp0q ą 0 and gp1q ă 0, so 1 P A
and 0 P B

We can write A Y B “ r0, 1s and A X B “ H (because we said f does not have a fixed point,
so g ‰ 0).

Consider sup B, the least upper bound of the pre-image of positive values of g. So for any
ϵ-interval surrounding sup B, we must have a point x P p´ϵ`sup B, sup B `ϵq such that gpxq ă 0.

If not, then gpbq ą 0 for every b P p´ϵ ` sup B, sup B ` ϵq (if g “ 0, then we have a
fixed-point).

This implies sup B is not the least upper bound for B (because sup B ´ ϵ is also an upper
bound) but this contradicts the definition of sup, so we must have a point x s.t. gpxq ă 0. Since
the interval also contains negative points, it must contain a point c where gpcq “ 0

3
3. (a) Show that if f is a continuous function from a compact metric space X to a metric space
Y , then the image of f is a compact subset of Y . (In particular, show that f is bounded, i.e., the
range of f is a bounded subset of Y ).

Solution

Consider any open cover of f pXq, then we know from lecture that the pre-image f ´1 of each
open set in the open cover must be open because f is continuous. Compactness of X implies
that this collection of sets in the pre-image (that form an open cover of X) has a finite subcover.
Applying f to each of these sets gives a finite subcover of f pXq, so f pXq is a compact subset of
Y , which implies that f pXq is a bounded subset of Y .

(b) Prove that a continuous real-valued function on a compact metric space assumes its maxi-
mum and minimum values.

Solution

Since f pXq is compact when X is compact and f is continuous (by part (a)), we can say that
f pXq is closed and bounded on the real line. Let m “ infpf q. Then, for positive ϵ, we have

f pXq X pm, m ` ϵq ‰ H

(because m ` ϵ must be contained in f pXq or m wouldn’t be the infimum).

Since f is closed, we can write that m P f , so f contains its infimum, so it attains its minimum.
We can repeat the same process with the supremum to show that f also assumes its maximum.

4
 (c) Prove that a metric space X is compact if and only if every continuous real-valued function
on X is bounded.

Solution

(ñ)

Suppose that X is compact, then we showed in part (a) that an arbitrary continuous function
f takes X to a compact subset of output metric space Y , implying that f is bounded.

Since we showed this for arbitrary f , it must be true for all f and in the particular case where
the output subset is on the real line.

(ð)

Suppose that X is not compact. Then we claim that not every continuous f is bounded on
the real line.

Consider a sequence xn with no convergent subsequence (because X is not compact) and the
function given in office hours:
f pxq “ inf dpx, xn q ` n
nPN

Then, for all x, y we have dpy, xq ` dpx, xn q ě dpy, xn q by triangle inequality.

So, for all n:

dpy, xq ` dpx, xn q ě inf dpy, xn q “ f pyq ´ n
nPN

And this implies

f pyq ´ n ď dpy, xq ` f pxq ´ n

So we’ve shown that

|f pyq ´ f pxq| ď dpy, xq

Which actually proves Lipschitz continuity.

Obviously, we have f pxn q ě n since distance is non-negative, so the function is continuous but
unbounded.

5
4. Let pX, dq be a metric space and let S be a non-empty set. Let BpS, Xq be the set of bounded
functions from S to X.

(a) Show that

ρpf, gq “ sup dpf psq, gpsqq
sPS

defines a metric on BpS, Xq. (This metric is called the metric of uniform convergence on S.)

Solution

• ρpf, gq ě 0 with 0 iff f “ g (shown below), and since d is a metric, we know ρ is real-valued.

• ρpf, f q “ supsPS dpf psq, f psqq, and by definition dpf, f q “ 0, so ρpf, f q “ 0, as desired.

• ρpf, gq “ supsPS dpf psq, gpsqq “ supsPS dpgpsq, f psqq because d is a metric.

• dpf, hq ď dpf, gqq ` dpg, hq ùñ dpf, hq ď supsPS dpf, gq ` supsPS dpg, hq

This implies
sup dpf, hq ď sup dpf, gq ` sup dpg, hq
sPS sPS sPS

(b) Show that BpS, Xq is complete if and only if X is complete.

Solution

(ñ)

We want to show that if all Cauchy sequences of functions converge, then all Cauchy sequences
in X converge.

For sequence of points xn P X, consider the sequence of constant functions fn psq “ xn for
all s. Then any Cauchy sequence of functions converges (by supposition), but this sequence of
functions is exactly a Cauchy sequence of points in X, which also converges.

(ð)

Given that all Cauchy sequences in X converge, we want to show Cauchy sequences of functions
converge.

You can ignore the following proof— this direction was proven in lecture as Thm
1.16 (although mine is fairly similar)

6
 Let’s define our limit function as we would pointwise: that is, for all fixed s, a Cauchy sequence
of functions fn converges pointwise to limit function f if we have

lim dpfn psq, f psqq “ 0

nÑ8

Existence of f is guaranteed by completeness of X

Now, it suffices to show that any Cauchy sequence converges uniformly.

Since fn is Cauchy, we have that @s P S, @ϵ ą 0, DN P N s.t.

m, n ą N ùñ dpfn , fm q ă ϵ

Just taking the pointwise limit as m Ñ 8 implies

dpfn , f q ă ϵ

for all s, so
sup dpfn , f q ď ϵ
sPS

You can ignore the above proof— this direction was proven in lecture as Thm 1.16

7
 (c) Show that if S is itself a metric space, then the space BCpS, Xq of bounded continuous
functions from S to X is a closed subspace of BpS, Xq.

Solution

We want to show that for all positive r,

BCpS, Xq “ tf P BpS, Xq| Bpf ; rq X BCpS, Xq ‰ Hu

We will be done if we can show that arbitrary f satisfying this property must also be continuous,
that is, for any sequence sn Ñ s in S, we must have that f psn q Ñ f psq in X

Suppose f is a function in BpS, Xq s.t. for all positive r, Bpf ; rq X BCpS, Xq ‰ H.

Then, there exists a function g P BCpS, Xq s.t.

sup dpf psq, gpsqq ă r

sPS

for all r.

We’re going to rewrite this equivalently as

ϵ
sup dpf psq, gpsqq ă
sPS 3

for all ϵ ą 0

We can also write for all ϵ ą 0

ϵ
dpf psn q, gpsn qq ă
3
because we got rid of the supremum, so this expression is still less than all positive ϵ (because we
actually reduced the distance)

Since g P BCpS, Xq, @s P S, @ϵ ą 0, Dδ ą 0 s.t.

ϵ
dpsn , sq ă δ ùñ dpgpsn q, gpsqq ă
3

Applying triangle inequality:

ϵ ϵ ϵ
dpf psn q, f psqq ă dpf psn q, gpsn qq ` dpgpsn q, gpsqq ` dpgpsq, pf psqqq “ ` ` “ϵ
3 3 3

8
 8
ÿ
5. Write ℓp for the space of real sequences px1 , x2 , . . .q for which |xk |p is finite, with the norm
k“1
8
ÿ
defined by }x}pp “ |xk |p . Prove that:
k“1

(a) ℓp is a Banach space.

Solution

Proposition 1.42 shows that ℓp is a Banach space; however, a section earlier on asks the reader
to verify that ℓp is a normed vector space.

If we want to verify that ℓp is a vector space, just notice that we can represent each element of
the sequence as a corresponding entry of a infinite-dimension real-valued vector. By this correspon-
dence, “sequence” addition is done exactly the same as vector addition— element-wise, and we
can define scalar multiplication, again, element-wise. This is a structure-preserving map between
sequences in ℓp and vectors in R8 , so I’m not going to check all of the vector space axioms.

Now, let’s check that

8
ÿ
}x}pp “ |xk |p
k“1

is actually a norm.

• First, we know the norm must be real-valued because each component of the sum is real-
valued. Since we’re dealing with the sum of absolute values, we know that }x}pp ą 0. This
also implies that }x}pp “ 0 iff each component of the sum is 0, or, in other words, x “ 0

• Now, consider
8
ÿ 8
ÿ 8
ÿ
}cx}pp “ |cxk |p “ |c|p |xk |p “ |c|p |xk |p “ |c|p }x}pp
k“1 k“1 k“1

• For the triangle inequality, we apply Minkowski’s inequality, from which it follows immediately.

9
 (b) If 1 ď p ă r ă 8 then ℓp is a proper dense subset of ℓr .

Solution

We want to show that for any sequence xn P ℓr , there exists a sequence yn P ℓp such that the
induced metric gives dpyn , xn q ă ϵ for any positive ϵ (so in any ball of any size, the intersection is
nonempty).

8
ÿ
Notice that |xk |r is finite, by so divergence theorem, the individual terms must go to 0 as
k“1
k Ñ 8. This means that for some K P N, we can bound the sum as such
8
ÿ
|xk |r ă ϵ
k“K

Then, let yi “ xi for 1 ď i ď K and, for i ą K, let each yi “ 0. This means that dpxk , yk q ă ϵ.

The sum of yn must converge because we only have a finite number of non-zero terms.

(c) ℓ1 is not dense in ℓ8 . (Bonus: What is the closure of ℓ1 in ℓ8 ?)

Solution
8
ÿ
ℓ1 is the the space of real sequences px1 , x2 , . . .q for which |xk | is finite, with the norm
k“1
8
ÿ
defined by }x}1 “ |xk |.
k“1

Notice that ℓ1 is just the space of sequences whose sum is absolutely convergent.

The closure of ℓ1 consists of the sequences psn q of which for all r ą 0,

Bppsn q; rq X ℓ1 ‰ H

where distance is defined as the norm-induced metric.

We can just choose a constant sequence, say p1, 1, 1, ...q, then this sequence is clearly in ℓ8 but
not in ℓ1 . If we were to suppose that there is a nonempty intersection for a ball of any nonnegative
radius, we could consider the ball of radius 1{2 surrounding p1, 1, ..., 1q. Clearly, we end up with
a sequence that still diverges in ℓ1 , so the intersection cannot be nonempty for all r, so ℓ1 is not
dense in ℓ8

10
6. Cpr0, 1sq is the Banach space of continuous real-valued functions on the interval r0, 1s, endowed
with the sup norm.

Solution

(a) For a set of points 0 ď x1 ă x2 ă x3 ¨ ¨ ¨ ă xm ď 1, let L be the functional on Cpr0, 1sq

defined by
ÿm
Lpf q “ ak f pxk q
k“1

where the ak ’s are real numbers. Show that L is continuous with norm
m
ÿ
}L} “ |ak |
k“1

It suffices to show that the RHS is bounded, since bounded operators are continuous. This
means that the supremum of the map from the unit ball is finite.

So consider an arbitrary function f P Cpr0, 1sq where the norm of f is less than 1, so for the
whole interval, |f pxq| ă 1. Thus,

ˇ ˇ
ˇÿm ˇ ÿ m ÿm
}Lpf q} “ |Lpf q| “ ˇ ak f pxk qˇ ď
ˇ ˇ
|ak | |f pxk q| ď |ak | ă 8
ˇk“1 ˇ k“1 k“1

pnq pnq
(b) For each integer n ą 0, let txk unk“1 be points in r0, 1s and let tak unk“1 be positive
numbers such that ż1 n
ÿ pnq pnq
f ptq dt “ ak f pxk q
0 k“1

for all polynomials f of degree at most n. Show that

ż1 n
ÿ pnq pnq
f ptq dt “ lim ak f pxk q
0 nÑ8
k“1

for all f P Cpr0, 1sq.

(Hint: You may use something like the Weierstrass approximation theorem, that polynomials
are dense in Cpr0, 1sq. Lemma 1.43 might also be useful.)

11
 Solution

Consider arbitrary f P Cpr0, 1sq. We want to show that for all positive ϵ, there exists a natural
number N s.t. ˇż ˇ
ˇ 1 ÿn ˇ
pnq pnq ˇ
n ą N ùñ ˇ f ptq dt ´ ak f pxk qˇ ă ϵ
ˇ
ˇ 0 k“1
ˇ

By density of polynomials in Cpr0, 1sq, we have a polynomial p P Cpr0, 1sq s.t.

dpp, f q ă ϵ

Now, we can extend the degree of p. If p has degree N , then for n ą N , there exists a sequence
pnq pnq
of points txk unk“1 P Cpr0, 1sq and a sequence of positive reals tak unk“1 s.t.
ż1 n
ÿ pnq pnq
f ptq dt “ ak ppxk q
0 k“1

as given.

To conclude:

As we take the limit as N goes to infinity, for all n ą N , we have

ˇż ˇ ˇż
ˇ 1 ÿn ˇ ˇ 1 ż1 ˇ ˇż 1 ˇ
pnq pnq ˇ
ˇ ˇ ˇ
ˇ f ptq dt ´ ak f pxk qˇ “ ˇˇ f ptq dt ´ pptq dtˇˇ ď ˇˇ pf ptq ´ pptqq dtˇˇ ă ϵ
ˇ
ˇ 0 k“1
ˇ 0 0 0

12
7. (a) You have no doubt seen the function
#
1 if x is rational
f pxq “
0 if x is irrational
and you know that f is discontinuous at every point of R. To warm up, prove this.
Solution

A function is discontinuous at every point if for every x0 P R, we can find a sequence where
xn Ñ x0 but f pxn q ↛ f px0 q

(ñ)

So let’s define a sequence for arbitrary rational x0 as follows. For any fixed ϵ ą 0, define x1 as
any irrational point in the interval
p´ϵ ` x0 , x0 ` ϵq
and define each following xk as any irrational point in the interval
´ ϵ ϵ¯
´ ` x0 , x 0 `
k k

We are able to construct a sequence this way because of the density of irrationals in the reals.
Now, notice that xn Ñ x0 .

However, f pxn q “ 0 for each point in the sequence. And we know that f px0 q “ 1 because x0
is rational. As such, f pxn q ↛ f px0 q, so f is discontinuous at each rational.

Irrationals

We’re going to follow a similar procedure as earlier.

Define a sequence for arbitrary irrational x0 as follows. For any fixed ϵ ą 0, define x1 as any
rational point in the interval
p´ϵ ` x0 , x0 ` ϵq
and define each following xk as any rational point in the interval
´ ϵ ϵ¯
´ ` x0 , x 0 `
k k

We are able to construct a sequence this way because of the density of rationals in the reals.
Now, notice that xn Ñ x0 .
13
 However, f pxn q “ 1 for each point in the sequence. And we know that f px0 q “ 0 because x0
is rational. As such, f pxn q ↛ f px0 q, so f is discontinuous at each irrational.

As such, we conclude f is discontinuous at each point of R

(b) Here’s a (perhaps familiar) twist on that one. Let
#
1{q if x is rational and x “ p{q in lowest terms
gpxq “
0 if x is irrational

Show that g is continuous at x if and only if x is irrational.

Solution

It suffices to show that g is continuous at each irrational but discontinuous at each rational.

Irrationals

Suppose x0 is an arbitrary irrational in R. Then we want to show that @ϵ ą 0, Dδ ą 0 such that

|x ´ x0 | ă δ ùñ gpxq ă ϵ

For all irrational x, this is true regardless of δ, so we only need to consider rational x.

For rational x, we think it would make our lives much easier if we could show that q must
increase the closer we get to x0 . This makes a lot of sense based on decimal expansions and
approximations of irrationals, so we make the claim that for all ϵ ą 0, sufficiently small δ, large M
and rational x,
1
|x ´ x0 | ă δ ùñ |gpxq| ď ăϵ
M

To show this claim is true, first consider M ą 1ϵ (so M can get infinitely large as ϵ gets infinitely
small). Notice that if we choose small δ surrounding x0 , then we can only get a finite number of
denominators q ă M when every rational is in reduced form. This is because our lower and upper
bounds pigeonhole the number of possible numerators we can have for a particular denominator.
As such, we can just choose δ small enough that every denominator q is greater than or equal to
M in the δ-interval.

As such, we have shown our claim is true, which further implies that gpxq is continuous at
irrationals.

14
Rationals

Suppose x0 is an arbitrary rational in R. For any fixed ϵ ą 0, define x1 as any irrational point in
the interval
p´ϵ ` x0 , x0 ` ϵq
and define each following xk as any irrational point in the interval
´ ϵ ϵ¯
´ ` x0 , x 0 `
k k
1
Then, xn Ñ x0 , but gpxn q Ñ 0 ‰ q
for any q ‰ 0, so gpxq is discontinuous at each rational.

As in part (a), we were able to choose irrationals in this way because of density of irrationals
in reals.

(c) For any function h : R Ñ R and any x P R define

jmpph, xq “ lim` supt|hpx1 q ´ hpx2 q| ḟor x1 , x2 P px ´ δ, x ` δqu
δÑ0

Show that h is continuous at x if and only if jmpph, xq “ 0.

Solution

(ð)

Suppose jmpph, xq “ 0.

A function h is continuous at x if for all ϵ ą 0, there exists a δ ą 0 such that

|x ´ x0 | ă δ ùñ |hpxq ´ hpx0 q| ă ϵ

Notice that
lim supt|hpxq ´ hpx0 q| ḟor x0 P px ´ δ, x ` δqu
δÑ0`

ď lim` supt|hpx1 q ´ hpx2 q| ḟor x1 , x2 P px ´ δ, x ` δqu “ 0

δÑ0

implies the left side is equal to 0. And the left side is exactly the same as:
" ˆ ˙*
δ δ
lim sup |hpxq ´ hpx0 q| ḟor x0 P x ´ , x `
δÑ0` 2 2
This implies that
|x ´ x0 | ă δ ùñ |hpxq ´ hpx0 q| ă ϵ
15
for all ϵ and sufficiently small δ. (Since lim sup ą lim and lim sup goes to 0)

(ñ)

Suppose jmpph, xq “ a, a ‰ 0.

A function h is discontinuous at x if there exists an ϵ ą 0 s.t. for every δ ą 0

|x ´ x0 | ă δ ùñ |hpxq ´ hpx0 q| ě ϵ

Notice that
b “ lim` supt|hpxq ´ hpx0 q| ḟor x0 P px ´ δ, x ` δqu
δÑ0

ď lim` supt|hpx1 q ´ hpx2 q| ḟor x1 , x2 P px ´ δ, x ` δqu “ a

δÑ0

However, b is non-zero. Suppose it was, then that means that there’s a delta small enough s.t.
every point in the interval maps to the same value as hpxq, but then the right side couldn’t be
nonzero.

And the left side is exactly the same as:

" ˆ ˙*
δ δ
lim sup |hpxq ´ hpx0 q| ḟor x0 P x ´ , x `
δÑ0` 2 2

This implies that

|x ´ x0 | ă δ ùñ |hpxq ´ hpx0 q| ě b
for all δ ą 0

16
 (d) In contrast to part (b), show that there is no function that is continuous at x if and only
if x is rational. (Problem 4 from the last problem set might be helpful).

Solution

By Problem 4 of the last problem set, the set of rational numbers cannot be expressed as the
R
intersection of countably many open subsets of . We claim that the set of all continuous points
can be expressed as the intersection of countably many open subsets, thus the rationals cannot be
the set of continuous points.

Consider the set

# +
Un “ xP R |Dδ ùñ sup |f paq ´ f pbq| ă
1
n
a,bPBpx;δq

We claim that this set is open. To show this, we can consider arbitrary x0 P Un and prove that
there exists an open ball around x0 that must be contained in the ball around x.

So, take x0 P Un . Then, let r “ δ ´ dpx0 , xq. Then Bpx0 ; rq Ă Un because

1
sup |f paq ´ f pbq| ď sup |f paq ´ f pbq| ă
a,bPBpx0 ;rq a,bPBpx,δq n

Now, let’s take the countable intersection of these open subsets.

Ş
Then, part (c) tells us that f is continuous at each x if and only if limδÑ0` Un goes to 0.
This is true because we’re forcing n1 infinitely small as n grows infinitely large.

As such, all continuous points are in the intersection of countably many open subsets of R, so
the rationals cannot be equal to the set of continuous points.

17