Continuous Functions on Metric Spaces

Math 201A, Fall 2016

1 Continuous functions
Definition 1. Let (X, dX ) and (Y, dY ) be metric spaces. A function f :
X → Y is continuous at a ∈ X if for every  > 0 there exists δ > 0 such that
dX (x, a) < δ implies that dY (f (x), f (a)) < .
In terms of open balls, the definition says that f (Bδ (a)) ⊂ B (f (a)). In
future, X, Y will denote metric spaces, and we will not distinguish explicitly
between the metrics on different spaces.
Definition 2. A subset U ⊂ X is a neighborhood of a point a ∈ X if
B (a) ⊂ U for some  > 0.
A set is open if and only if it is a neighborhood of every point in the
set, and U is a neighborhood of x if and only if U ⊃ G where G is an open
neighborhood of x. Note that slightly different definitions of a neighborhood
are in use; some definitions require that a neighborhood is an open set, which
we do not assume.
Proposition 3. A function f : X → Y is continuous at a ∈ X if and only
if for every neighborhood V ⊂ Y of f (a) the inverse image f −1 (V ) ⊂ X is a
neighborhood of a.
Proof. Suppose that the condition holds. If  > 0, then V = B (f (a))
is a neighborhood of f (a), so U = f −1 (V ) is a neighborhood of a. Then
Bδ (a) ⊂ U for some δ > 0, which implies that f (Bδ (a)) ⊂ B (f (a)), so f is
continuous at a.
Conversely, if f is continuous at a and V is a neighborhood of f (a),
then B (f (a)) ⊂ V for some  > 0. By continuity, there exists δ > 0 such
that f (Bδ (a)) ⊂ B (f (a)), so Bδ (a) ⊂ f −1 (V ), meaning that f −1 (V ) is a
neighborhood of a.

1
Definition 4. A function f : X → Y is sequentially continuous at a ∈ X if
xn → a in X implies that f (xn ) → f (a) in Y .

Theorem 5. A function f : X → Y is continuous at a if and only if it is

sequentially continuous at a.

Proof. Suppose that f is continuous at a. Let  > 0 be given and suppose

that xn → a. Then there exists δ > 0 such that d(f (x), f (a)) <  for
d(x, a) < δ, and there exists N ∈ N such that d(xn , a) < δ for n > N .
It follows that d(f (xn ), f (a)) <  for n > N , so f (xn ) → f (a) and f is
sequentially continuous at a.
Conversely, suppose that f is not continuous at a. Then there exists
0 > 0 such that for every n ∈ N there exists xn ∈ X with d(xn , a) < 1/n
and d(f (xn ), f (a)) ≥ 0 . Then xn → a but f (xn ) 6→ f (a), so f is not
sequentially continuous at a.

Definition 6. A function f : X → Y is continuous if f is continuous at

every x ∈ X.

Theorem 7. A function f : X → Y is continuous if and only if f −1 (V ) is

open in X for every V that is open in Y .

Proof. Suppose that the inverse image under f of every open set is open. If
x ∈ X and V ⊂ Y is a neighborhood of f (x), then V ⊃ W where W is an
open neighborhood of f (x). Then f −1 (W ) is an open neighborhood of x and
f −1 (W ) ⊂ f −1 (V ), so f −1 (V ) is a neighborhood of x, which shows that f is
continuous.
Conversely, suppose that f : X → Y is continuous and V ⊂ Y is open. If
x ∈ f −1 (V ), then V is an open neighborhood of f (x), so the continuity of f
implies that f −1 (V ) is a neighborhood of x. It follows that f −1 (V ) is open
since it is a neighborhood of every point in the set.

Theorem 8. The composition of continuous functions is continuous

Proof. Suppose that f : X → Y and g : Y → Z are continuous, and g ◦ f :

X → Z is their composition. If W ⊂ Z is open, then V = g −1 (W ) is open,
so U = f −1 (V ) is open. It follows that (g ◦ f )−1 (W ) = f −1 (g −1 (W )) is open,
so g ◦ f is continuous.

Theorem 9. The continuous image of a compact set is compact.

2
Proof. Suppose that f : X → Y is continuous and X is compact. If {Gα :
α ∈ I} is an open cover of f (X), then {f −1 (Gα ) : α ∈ I} is an open cover of
X, since the inverse image of an open set is open. Since X is compact, it has
a finite subcover {f −1 (Gαi ) : i = 1, 2, . . . , n}. Then {Gαi : i = 1, 2, . . . , n} is
a finite subcover of f (X), which proves that f (X) is compact.

2 Uniform convergence
A subset A ⊂ X is bounded if A ⊂ BR (x) for some (and therefore every)
x ∈ X and some R > 0. Equivalently, A is bounded if

diam A = sup {d(x, y) : x, y ∈ A} < ∞.

Definition 10. A function f : X → Y is bounded if f (X) ⊂ Y is bounded.

Definition 11. A sequence (fn ) of functions fn : X → Y converges uniformly
to a function f : X → Y if for every  > 0 there exists N ∈ N such that
n > N implies that d (fn (x), f (x)) <  for all x ∈ X.
Unlike pointwise convergence, uniform convergence preserves bounded-
ness and continuity.
Proposition 12. Let (fn ) be a sequence of functions fn : X → Y . If each
fn is bounded and fn → f uniformly, then f : X → Y is bounded.
Proof. By the uniform convergence, there exists n ∈ N such that

d (fn (x), f (x)) ≤ 1 for all x ∈ X.

Since fn is bounded, there exists y ∈ Y and R > 0 such that

d (fn (x), y) ≤ R for all x ∈ X.

It follows that

d(f (x), y) ≤ d (f (x), fn (x)) + d (fn (x), y) ≤ R + 1 for all x ∈ X,

meaning that f is bounded.

Theorem 13. Let (fn ) be a sequence of functions fn : X → Y . If each fn is
continuous at a ∈ X and fn → f uniformly, then f : X → Y is continuous
at as.

3
Proof. Let a ∈ X. Since (fn ) converges uniformly to f , given  > 0, there
exists n ∈ N such that

d (fn (x), f (x)) < for all x ∈ X,
3
and since fn is continuous at a, there exists δ > 0 such that

d (fn (x), fn (a)) < if d(x, a) < δ.
3
If d(x, a) < δ, then it follows that

d (f (x), f (a)) ≤ d (f (x), fn (x)) + d (fn (x), fn (a)) + d (fn (a), f (a)) < ,

which proves that f is continuous at a.

Since uniform convergence preserves continuity at a point, the uniform
limit of continuous functions is continuous.

Definition 14. A sequence (fn ) of functions fn : X → Y is uniformly

Cauchy if for every  > 0 there exists N ∈ N such that m, n > N implies
that d (fm (x), fn (x)) <  for all x ∈ X.

A uniformly convergent sequence of functions is uniformly Cauchy. The

converse is also true for functions that take values in a complete metric space.

Theorem 15. Let Y be a complete metric space. Then a uniformly Cauchy

sequence (fn ) of functions fn : X → Y converges uniformly to a function
f :X →Y.

Proof. The uniform Cauchy condition implies that the sequence (fn (x)) is
Cauchy in Y for every x ∈ X. Since Y is complete, fn (x) → f (x) as n → ∞
for some f (x) ∈ Y .
Given  > 0, choose N ∈ N such that for all m, n > N we have

d (fm (x), fn (x)) <  for all x ∈ X.

Taking the limit of this inequality as m → ∞, we get that

d (f (x), fn (x)) ≤  for all x ∈ X

for all n > N , which shows that (fn ) converges uniformly to f as n → ∞.

4
3 Function spaces
Definition 16. The metric space (B(X, Y ), d∞ ) is the space of bounded
functions f : X → Y equipped with the uniform metric

d∞ (f, g) = sup {d (f (x), g(x)) : x ∈ X} .

It is straightforward to check that d∞ is a metric on B(X, Y ); in partic-

ular, d∞ (f, g) < ∞ if f , g are bounded functions.
Theorem 17. Let Y be a complete metric space. Then (B(X, Y ), d∞ ) is a
complete metric space.
Proof. If (fn ) is a Cauchy sequence in B(X, Y ), then (fn ) is uniformly
Cauchy, so by Theorem 15 it converges uniformly to a function f : X → Y .
By Proposition 12, f ∈ B(X, Y ), so B(X, Y ) is complete.
Definition 18. The space C(X, Y ) is the space of continuous functions f :
X → Y , and (Cb (X, Y ), d∞ ) is the space of bounded, continuous functions
f : X → Y equipped with the uniform metric d∞ .
Theorem 19. Let X be a metric space and Y a complete metric space.
Then (Cb (X, Y ), d∞ ) is a complete metric space.
Proof. By Theorem 13, Cb (X, Y ) is a closed subspace of the complete metric
space B(X, Y ), so it is a complete metric space.

4 Continuous functions on compact sets

Definition 20. A function f : X → Y is uniformly continuous if for ev-
ery  > 0 there exists δ > 0 such that if x, y ∈ X and d(x, y) < δ, then
d (f (x), f (y)) < .
Theorem 21. A continuous function on a compact metric space is bounded
and uniformly continuous.
Proof. If X is a compact metric space and f : X → Y a continuous function,
then f (X) is compact and therefore bounded, so f is bounded.
Let  > 0. For each a ∈ X, there exists δ(a, ) > 0 such that

d (f (x), f (a)) < for all x ∈ X with d(x, a) < δ(a, ).
2
5
 
Then Bδ(a,) (a) : a ∈ X is an open cover of X, so by the Lebesgue covering
lemma, there exists η > 0 such that for every x ∈ X there exists a ∈ X with
Bη (x) ⊂ Bδ(a,) (a). Hence, d(x, y) < η implies that x, y ∈ Bδ(a,) (a), so

d (f (x), f (y)) ≤ d (f (x), f (a)) + d (f (a), f (y)) < ,

which shows that f is uniformly continuous.

For real-valued functions, we have the following basic result.

Theorem 22 (Weierstrass). If X is compact and f : X → R is continuous,

then f is bounded and attains its maximum and minimum values.

Proof. The image f (X) ⊂ R is compact, so it is closed and bounded. It

follows that M = supX f < ∞ and M ∈ f (X). Similarly, inf X f ∈ f (X).

5 The Arzelà-Ascoli theorem

Definition 23. A family of functions F ⊂ C(X, Y ) is equicontinuous if for
every a ∈ X and  > 0, there exists δ > 0 such that if x ∈ X and d(x, a) < δ,
then d (f (x), f (a)) <  for every f ∈ F.

Definition 24. A family of functions F ⊂ C(X, Y ) is uniformly equicon-

tinuous if for every  > 0, there exists δ > 0 such that if x, y ∈ X and
d(x, y) < δ, then d (f (x), f (y)) <  for every f ∈ F.

Theorem 25. If X is a compact metric space and F ⊂ C(X, Y ) is equicon-

tinuous, then F is uniformly equicontinuous.

Proof. Let F ⊂ C(X, Y ) be equicontinuous, and suppose that  > 0. For

each a ∈ X, there exists δ(a, ) > 0 such that if x ∈ X and d(x, a) < δ(a, ),
then d (f (x), f (a)) < /2 for every f ∈ F. Then {Bδ(a,) (a) : a ∈ X} is
an open cover of X, so by the Lebesgue covering lemma there exist η > 0
such that for every x ∈ X we have Bη (x) ⊂ Bδ(a,) (a) for some a ∈ X. If
d(x, y) < η, then x, y ∈ Bδ(a,) (a), so

d (f (x), f (y)) ≤ d (f (x), f (a)) + d (f (a), f (y)) <  for every f ∈ F,

which shows that F is uniformly equicontinuous.

6
 For simplicity, we now specialize to real-valued functions. In that case,
we write C(X) = C(X, R). If X is compact, then C(X) equipped with
the sup-norm kf k∞ = supx∈X |f (x)|, and the usual pointwise definitions of
vector addition and scalar multiplication, is a Banach space.

Definition 26. A family of functions F ⊂ C(X) is pointwise bounded if for

every x ∈ X there exists M > 0 such that |f (x)| ≤ M for all f ∈ F.

Definition 27. A subset A ⊂ X is of a metric space X is precompact (or

relatively compact) if its closure Ā is compact.

Lemma 28. Let X be a complete metric space. A subset A ⊂ X is precom-

pact if and only if it is totally bounded.

Proof. If A is precompact, then the compact set Ā is totally bounded, so

A ⊂ Ā is totally bounded.
Conversely, suppose that A is totally bounded. If  > 0, then A has a
finite (/2)-net E, and every x ∈ Ā belongs to B̄/2 (a) for some a ∈ E. It
follows that E is a finite -net for Ā, so Ā is totally bounded. Moreover, Ā
is a closed subset of a complete space, so Ā is complete, which shows that Ā
is compact and A is precompact.

Theorem 29 (Arzelà-Ascoli). Let X be a compact metric space. A family

F ⊂ C(X) of continuous functions f : X → R is precompact with respect
to the sup-norm topology if and only if it is pointwise bounded and equicon-
tinuous. Furthermore, F is compact if and only if it is closed, pointwise
bounded, and equicontinuous.

Proof. Since C(X) is complete, a subset is complete if and only if it is closed.

It follows that F is compact if and only if it is closed and totally bounded.
From Lemma 28, F is precompact if and only if it is totally bounded. Thus,
it suffices to prove that F is totally bounded if and only if it is pointwise
bounded and equicontinuous.
First, suppose that F is pointwise bounded and equicontinuous, and let
 > 0. We will construct a finite -net for F.
The family F is equicontinuous and X is compact, so F is uniformly
equicontinuous, and there exists δ > 0 such that

d(x, y) < δ implies that |f (x) − f (y)| < for all f ∈ F.
3

7
Since X is compact, it has a finite δ-net E = {xi : 1 ≤ i ≤ n} such that
n
[
X= Bδ (xi ).
i=1

Furthermore, since F is pointwise bounded,

Mi = sup {|f (xi )| : f ∈ F} < ∞ for each 1 ≤ i ≤ n.

Let M = max{Mi : 1 ≤ i ≤ n}. Then [−M, M ] ⊂ R is compact, so it has a

finite (/6)-net F = {yj ∈ R : 1 ≤ j ≤ m} such that
m
[
[−M, M ] ⊂ B/6 (yj ).
j=1

Here, B/6 (yj ) is an open interval in R of diameter /3.

Let Φ be the finite set of maps φ : E → F . For each φ ∈ Φ, define

Fφ = f ∈ F : f (xi ) ∈ B/6 (φ(xi )) for i = 1, . . . , n .

Since f (xi ) ∈ [−M, M ] and {B/6 (yj ) : 1 ≤ j ≤ m} covers [−M, M ], we have

[
F= Fφ .
φ∈Φ

Let f, g ∈ Fφ . Then for each xi ∈ E, we have f (xi ), g(xi ) ∈ B/6 (yj ) for
some yj ∈ F , so

|f (xi ) − g(xi )| < .
3
Furthermore, if x ∈ X, then x ∈ Bδ (xi ) for some xi ∈ E, so d(x, xi ) < δ, and
the uniform equicontinuity of F implies that

|f (x) − g(x)| ≤ |f (x) − f (xi )| + |f (xi ) − g(xi )| + |g(xi ) − g(x)| < .

Thus, kf − gk∞ < , so the diameter of Fφ is less than , and Fφ ⊂ B (fφ )

for some fφ ∈ F. Hence, [
F⊂ B (fφ )
φ∈Φ

is totally bounded.

8
 Conversely, suppose that F ⊂ C(X) is precompact. Then F is bounded,
so there exists M > 0 such that kf k∞ ≤ M for all f ∈ F, which implies that
F is pointwise (and, in fact, uniformly) bounded.
Moreover, F is totally bounded, so given  > 0, there exists a finite (/3)-
net E = {fi : 1 ≤ i ≤ n} for F. Each fi : X → R is uniformly continuous, so
there exists δi > 0 such that d(x, y) < δi implies that |fi (x) − fi (y)| < /3.
Let δ = min{δi : 1 ≤ i ≤ n}. If f ∈ F, then kf − fi k∞ < /3 for some
fi ∈ E, so d(x, y) < δ implies that

|f (x) − f (y)| ≤ |f (x) − fi (x)| + |fi (x) − fi (y)| + |fi (y) − f (y)| < ,

meaning that F is uniformly equicontinuous.

This result, and the proof, also applies to complex-valued functions in
C(X, C), vector-valued functions in C(X, Rn ), and — with a stronger hy-
pothesis — to functions that take values in a complete metric space Y : If
X is compact, then F ⊂ C(X, Y ) is precompact with respect to the uni-
form metric d∞ if and only if F is equicontinuous and pointwise precompact,
meaning that the set {f (x) : f ∈ F} is precompact in Y for every x ∈ X.

6 The Peano existence theorem

As an application of the Arzelà-Ascoli theorem, we prove an existence result
for an initial-value problem for a scalar ODE. (Essentially the same proof
applies to systems of ODEs.)

Theorem 30. Suppose that f : R × R → R is continuous and u0 ∈ R. Then

there exists T > 0 such that the initial-value problem
du
= f (u, t),
dt
u(0) = u0

has a continuously differentiable solution u : [0, T ] → R.

Proof. For each h > 0, construct an approximate solution uh : [0, ∞) → R

as follows. Let tn = nh and define a sequence (un )∞
n=0 recursively by

un+1 = un + hf (un , tn ) .

9
Define uh (t) by linear interpolation between un and un+1 , meaning that

(t − tn )
uh (t) = un + (un+1 − un ) for tn ≤ t ≤ tn+1 .
h
Choose L, T1 > 0, and let

R1 = (x, t) ∈ R2 : |x − u0 | ≤ L and 0 ≤ t ≤ T1 .


Then, since f is continuous and R1 is compact,

M = sup |f (x, t)| < ∞.

(x,t)∈R1

If (uk , tk ) ∈ R1 for every 0 ≤ k ≤ n − 1, then

n−1
X n−1
X
|un − u0 | ≤ |uk+1 − uk | = h |f (uk , tk )| ≤ M tn .
k=0 k=0

Thus, (un , tn ) ∈ R1 so long as M tn ≤ L and tn ≤ T1 . Let

 
L T
T = min T1 , , N= ,
M h

and define R ⊂ R1 by

R = (x, t) ∈ R2 : |x − u0 | ≤ L and 0 ≤ t ≤ T .


Then it follows that (un , tn ) ∈ R if 0 ≤ tn ≤ T and 0 ≤ n ≤ N . Moreover,

|un+1 − un | = h|f (un , tn )| ≤ M h for every 0 ≤ n ≤ N .

It is clear that the linear interpolant uh then satisfies

|uh (s) − uh (t)| ≤ M |s − t| for every 0 ≤ s, t ≤ T .

To show this explicitly, suppose that 0 ≤ s < t ≤ T , tm ≤ s ≤ tm+1 , and

10
tn ≤ t ≤ tn+1 . Then
n−1
X
|uh (t) − uh (s)| ≤ |uh (t) − uh (tn )| + |uh (tk+1 ) − uh (tk )|
k=m+1

+ |uh (tm+1 ) − uh (s)|

n−1
X
≤ (t − tn ) |f (un , tn )| + h |f (uk , tk )|
k=m+1

+ (tm+1 − s) |f (um , tm )|
≤ M [(t − tn ) + (tn − tm+1 ) + (tm+1 − s)]
≤ M |t − s|.

In addition, |uh (t)| ≤ L for t ∈ [0, T ].

By the Arzelà-Ascoli theorem, the family {uh : h > 0} is precompact in
C([0, T ]), so we can extract a subsequence (uhj )∞ j=1 such that hj → 0 and
uhj → u uniformly to some u ∈ C([0, T ]) as j → ∞.
To show that u is a solution of the initial-value problem, we note that,
by the fundamental theorem of calculus, u ∈ C 1 ([0, T ]) is a solution if and
only if u ∈ C([0, T ]) and u satisfies the integral equation
Z t
u(t) = u0 + f (u(s), s) ds for 0 ≤ t ≤ T .
0

Let χ[0,t] be the characteristic function of the interval [0, t], defined by
(
1 if 0 ≤ s ≤ t
χ[0,t] (s) =
0 if s > t

Then the piecewise-linear approximation uh satisfies the equation

N
X −1 Z tk+1
uh (t) = u0 + χ[0,t] (s)f (uh (tk ), tk ) ds.
k=0 tk

Let  > 0. Since f : R → R is continuous on a compact set, it is uniformly

continuous, and there exists δ > 0 such that

|x − y| < M δ and |s − t| < δ implies that |f (x, s) − f (y, t)| < /T .

11
For 0 < h < δ and tk ≤ s ≤ tk+1 , we have |uh (s) − uh (tk )| < M δ and
|s − tk | < δ, so for 0 ≤ t ≤ T we get that
Z t
uh (t) − u0 − f (uh (s), s) ds
0
N −1
X k+1Z t
≤ χ[0,t] (s) |f ((uh (s), s) − f (uh (tk ), tk )| ds
k=0 tk
N −1 Z tk+1
 X
< χ[0,t] (s) ds
T k=0 tk

< .

Thus,
Z t
uh (t) − u0 − f (uh (s), s) ds → 0 uniformly on [0, T ] as h → 0.
0

Since uhj → u uniformly as j → ∞, it follows that

Z t


u0 + f uhj (s), s ds → u(t) uniformly on [0, T ] as j → ∞.
0

However, the uniform convergence uhj → u and the uniform continuity of

f imply that f (uhj (t), t) → f (u(t), t) uniformly. Then, since the Riemann
integral of the limit of a uniformly convergent sequence of Riemann-integrable
functions is the limit of the Riemann integrals, we get that
Z t Z t


u0 + f uhj (s), s ds → u0 + f (u(s), s) ds as j → ∞
0 0

for each t ∈ [0, T ]. Hence, since the limits must be the same,
Z t
u(t) = u0 + f (u(s), s) ds,
0

so u ∈ C 1 ([0, T ]) is a solution of the initial-value problem.

7 The Stone-Weierstrass theorem

Definition 31. A subset A ⊂ X is dense in X if Ā = X.

12
Definition 32. A family of functions A ⊂ C(X) is an algebra if it is a linear
subspace of C(X) and f, g ∈ A implies that f g ∈ A.
Here, f g is the usual pointwise product, (f g)(x) = f (x)g(x).
Definition 33. A family of functions F ⊂ C(X) separates points if for every
pair of distinct points x, y ∈ X there exists f ∈ F such that f (x) 6= f (y).
A constant function f : X → R is a function such that f (x) = c for all
x ∈ X and some c ∈ R.
Theorem 34 (Stone-Weierstrass). Let X be a compact metric space. If
A ⊂ C(X) is an algebra that separates points and contains the constant
functions, then A is dense in (C(X), k · k∞ ).
Proof. Suppose that f ∈ C(X) and let  > 0. By Lemma 35, for every
pair of points y, z ∈ X, there exists gyz ∈ A such that gyz (y) = f (y) and
gyz (z) = f (z).
First, fix y ∈ X. Since f − gyz is continuous and (f − gyz )(z) = 0, for
each z ∈ X there exists δ(z) > 0 such that
gyz (x) < f (x) +  for all x ∈ Bδ(z) (z).
The family of open balls {Bδ(z) (z) : z ∈ X} covers X, so it has a finite
subcover {Bδi (zi ) : 1 ≤ i ≤ n} since X is compact. By Lemma 37, the
function
gy = min{gyzi : 1 ≤ i ≤ n}
belongs to Ā. Moreover, gy (y) = f (y) and
gy (x) < f (x) +  for all x ∈ X.
Next, consider gy . Since f − gy is continuous and (f − gy )(y) = 0, for
each y ∈ Y there exists δ(y) > 0 such that
gy (x) > f (x) −  for all x ∈ Bδ(y) (y).
Then {Bδ(y) (y) : y ∈ X} is an open cover of X, so it has a finite subcover
{Bδ(yj ) (yj ) : 1 ≤ j ≤ m}. By Lemma 37, the function
g = max{gyj : 1 ≤ j ≤ m}
belongs to Ā. Furthermore,
f (x) −  < g(x) < f (x) +  for all x ∈ X,
so kf − gk∞ < , which shows that Ā is dense in C(X).

13
 To complete the proof of the Stone-Weierstrass theorem, we prove several
lemmas.

Lemma 35. Let A ⊂ C(X) be a linear subspace that contains the constant
functions and separates points. If f ∈ C(X) and y, z ∈ X, then there exists
gyz ∈ A such that gyz (y) = f (y) and gyz (z) = f (z).

Proof. Given distinct points y, z ∈ X and real numbers a, b ∈ R, there exists

h ∈ A with h(y) 6= h(z). Then g ∈ A defined by
 
h(x) − h(y)
g(x) = a + (b − a)
h(z) − h(y)

satisfies g(y) = a, g(z) = b. If y 6= z, choose gyz as above with a = f (y) and

b = f (z); if y = z, choose gyy = f .
The next lemma is a special case of the Weierstrass approximation theo-
rem.

Lemma 36. There is a sequence (pn ) of polynomials pn : [−1, 1] → R such

that pn (t) → |t| as n → ∞ uniformly on [−1, 1].

Proof. Define a sequence of polynomials qn : [0, 1] → R by the recursion

relation
1
t − qn2 (t) ,


qn+1 (t) = qn (t) + q0 (t) = 0.
2
We claim that for every n ∈ N, we have
√
qn−1 (t) ≤ qn (t) ≤ t for all t ∈ [0, 1].

Suppose as an induction hypothesis that this inequality holds for some

n ∈ N. Then qn+1 (t) ≥ qn (t) and
√ √ 1
t − qn2 (t)


t − qn+1 (t) = t − qn (t) −
2
√  1 √ 
= t − qn (t) 1 − t + qn (t)
2
≥ 0.
√ √
so qn (t) ≤ qn+1 (t) ≤ t. Moreover, q1 (t) = t/2, so q0 (t) ≤ q1 (t) ≤ t, and
the inequality follows by induction.

14
 For each t ∈ [0, 1] the real sequence (qn (t)) is monotone increasing and
bounded from above by 1, so it converges to some limit q(t) as n → ∞.
Taking the limit of the recursion relations
√ as n → ∞, we find that q 2 (t) = t,
and since q(t) ≥ 0, we have q(t) = t.
According to Dini’s theorem, a monotone increasing sequence of contin-
uous functions that converges pointwise √ on a compact set to a continuous
function converges uniformly, so qn (t) → t uniformly on [0, 1].
Finally, defining the polynomials pn : [−1, 1] → R by pn (t) = qn (t2 ), we
see that pn (t) → |t| uniformly on [−1, 1].
Lemma 37. Let X be a compact metric space, and suppose that A ⊂ C(X)
is an algebra. If f, g ∈ A, then max{f, g}, min{f, g} ∈ Ā.
Proof. Since
1 1
max{f, g} = (f + g + |f − g|) , min{f, g} = (f + g − |f − g|)
2 2
it is suffices to show that |f | ∈ Ā for every f ∈ A.
If f ∈ A, then f is bounded, so f (X) ⊂ [−M, M ] for some M > 0.
From Lemma 36, there exists a sequence (pn ) of polynomials that converges
uniformly to |t| on [−1, 1]. Then pn (f /M ) ∈ A and pn (f /M ) → |f |/M
uniformly as n → ∞, since pn is uniformly continuous on [−1, 1]. It follows
that |f | ∈ Ā.
Definition 38. A metric space is separable if it has a countable dense subset.
Corollary 39. Let X ⊂ Rn be a closed, bounded set and P the set of
polynomials p : X → R. Then P is dense in (C(X), k · k∞ ). Moreover, C(X)
is separable.
Proof. A closed, bounded set in Rn is compact, and the set P of polynomials
is a subalgebra of C(X) that contains the constant functions and separates
points, so P is dense in C(X). Any polynomial can be uniformly approxi-
mated on a bounded set by a polynomial with rational coefficients, and there
are countable many such polynomials, so C(X) is separable.
In particular, we have the following special case.
Theorem 40 (Weierstrass approximation). If f : [0, 1] → R is a continuous
function and  > 0, then there exists a polynomial p : [0, 1] → R such that
|f (x) − p(x)| <  for every x ∈ [0, 1].

15