1

The Metric Space 𝐶(𝐼)

Def. Let 𝑪(𝑰) = {bounded continuous functions 𝒇: 𝑰 ⊆ ℝ → ℝ}

Note: If 𝐼 is closed and bounded then any continuous function is bounded on 𝐼.

𝐶(𝐼) is a metric space with the distance defined as:

𝑑(𝑓(𝑥 ), 𝑔(𝑥 )) = sup |𝑓(𝑥 ) − 𝑔(𝑥 )|
𝑥∈𝐼
1. 𝑑(𝑓(𝑥 ), 𝑔(𝑥 )) = sup |𝑓(𝑥 ) − 𝑔(𝑥 )| ≥ 0 ; and
𝑥∈𝐼
𝑑(𝑓(𝑥 ), 𝑔(𝑥 )) = 0 implies 𝑓 (𝑥 ) = 𝑔(𝑥).

2. 𝑑(𝑓(𝑥 ), 𝑔(𝑥 )) = 𝑑(𝑔(𝑥 ), 𝑓(𝑥 )).

3. 𝑑(𝑓(𝑥 ), 𝑔(𝑥 )) ≤ 𝑑(𝑓(𝑥 ), ℎ(𝑥 )) + 𝑑(ℎ(𝑥 ), 𝑔(𝑥 )).

This is true because if 𝐴(𝑥) = 𝐵(𝑥) + 𝐸(𝑥) then by the triangle inequality:
|𝐴(𝑥)| ≤ |𝐵(𝑥)| + |𝐸 (𝑥)| for any 𝑥 ∈ 𝐼.

Thus we have: sup|𝐴(𝑥)| ≤ sup|𝐵(𝑥)| + sup|𝐸 (𝑥)|.

𝑥∈𝐼 𝑥∈𝐼 𝑥∈𝐼

Now let 𝐴(𝑥) = 𝑓(𝑥) − 𝑔(𝑥), 𝐵(𝑥) = 𝑓 (𝑥) − ℎ(𝑥), 𝐸 (𝑥) = ℎ(𝑥) − 𝑔(𝑥).

This gives us: 𝑑(𝑓(𝑥 ), 𝑔(𝑥 )) ≤ 𝑑(𝑓(𝑥 ), ℎ(𝑥 )) + 𝑑(ℎ(𝑥 ), 𝑔(𝑥 )).

Notice that a sequence of functions 𝑓𝑛 (𝑥) ∈ 𝐶(𝐼) converges to 𝑓(𝑥) with this
metric if given any 𝜖 > 0 there exists a 𝑁 ∈ ℤ+ such that if 𝑛 ≥ 𝑁 then
𝑑(𝑓𝑛 (𝑥 ), 𝑓(𝑥 )) = sup|𝑓𝑛 (𝑥 ) − 𝑓(𝑥 )| < 𝜖.
𝑥∈𝐼

This 𝜖 statement is equivalent to saying that |𝑓𝑛 (𝑥 ) − 𝑓 (𝑥 )| < 𝜖 for all 𝑥 ∈ 𝐼.

Thus convergence in 𝐶(𝐼) is the same as uniform convergence.
 2

We already know that if 𝑓𝑛 (𝑥) converges uniformly to 𝑓(𝑥) and all of the 𝑓𝑛 (𝑥)
are continuous then so is 𝑓(𝑥). Our goal is to show that 𝐶(𝐼) is complete with the
metric described above. Thus we must show that every Cauchy sequence in 𝐶(𝐼)
converges in 𝐶(𝐼). Later in this section we’ll see that if {𝑓𝑛 (𝑥)} is a Cauchy
sequence in 𝐶(𝐼) then {𝑓𝑛 (𝑥)} converges uniformly to a function 𝑓(𝑥). Since
each 𝑓𝑛 is continuous, 𝑓 must also be continuous. We will then show that since
each 𝑓𝑛 is bounded on 𝐼 and {𝑓𝑛 (𝑥)} and converges uniformly to a function 𝑓(𝑥),
then 𝑓(𝑥) is also bounded on 𝐼. Thus any Cauchy sequence in 𝐶(𝐼) converges to
a function in 𝐶(𝐼). Hence 𝐶(𝐼) is a complete metric space.

Recall that:
Def. Let 𝑉 be a vector (linear) space. A real valued function ‖∙‖ on 𝑉 is called a
norm if for each 𝑣, 𝑤 ∈ 𝑋 and 𝛼 ∈ ℝ:

1. ‖𝑣 + 𝑤‖ ≤ ‖𝑣 ‖ + ‖𝑤‖ (Triangle inequality)

2. ‖𝛼𝑣 ‖ = |𝛼|‖𝑣 ‖ (positive homogenity)
3. ‖𝑣 ‖ ≥ 0 and ‖𝑣 ‖ = 0 if and only if 𝑣 = 0.

Ex. ℝ𝑛 is a normed linear space with 𝑣 =< 𝑎1 , … , 𝑎𝑛 >∈ ℝ𝑛 ; and

‖𝑣 ‖ = √𝑎12 + 𝑎22 + ⋯ + 𝑎𝑛2 .

Given any normed vector space 𝑉 we can always define a metric on 𝑉 by

𝑑 (𝑣, 𝑤) = ‖𝑣 − 𝑤‖.

Ex. 𝐶(𝐼) is a vector space. We can define a norm on 𝐶(𝐼) by

‖𝑓‖∞ = sup |𝑓(𝑥 )| , 𝑓 ∈ 𝐶(𝐼).
𝑥∈𝐼
 3

Def. Given a sequence {𝑣𝑛 } ⊆ 𝑉, a normed vector space, we say that {𝒗𝒏 }
converges to 𝒗 with respect to the norm ‖∙‖ on 𝑉 if {𝑣𝑛 } converges to 𝑣 with
respect to the metric 𝑑 (𝑣, 𝑤 ) = ‖𝑣 − 𝑤 ‖. That is, given any 𝜖 > 0 there exists a
𝑁 ∈ ℤ+ such that if 𝑛 ≥ 𝑁 then 𝑑(𝑣𝑛 , 𝑣 ) = ‖𝑣𝑛 − 𝑣 ‖ < 𝜖 .

Def. Given a normed vector space 𝑉, we say that 𝑽 is complete with respect to
‖∙‖ if 𝑉 is complete with respect to the metric 𝑑 (𝑣, 𝑤) = ‖𝑣 − 𝑤‖.

Def. A complete normed vector space is called a Banach space.

Ex. ℝ𝑛 is a Banach space with the standard norm on ℝ𝑛 .

Ex. 𝐶(𝐼) is a Banach space with 𝑑(𝑓 (𝑥 ), 𝑔(𝑥 )) = sup |𝑓(𝑥 ) − 𝑔(𝑥 )|. We
𝑥∈𝐼
will see this shortly.

Def. Let 𝑉 and 𝑊 be normed vector spaces. A linear transformation, 𝑇, from 𝑉 to

𝑊 is called an operator from 𝑽 to 𝑾. 𝑇 is called bounded if there is an 𝑀 ∈ ℝ
such that :
‖𝑇(𝑣 )‖𝑊 ≤ 𝑀‖𝑣 ‖𝑉 for all 𝑣 ∈ 𝑉.

3
Ex. Let 𝑇: 𝐶[0,3] → ℝ by 𝑇(𝑓) = ∫0 𝑓 (𝑥 )𝑑𝑥 .

𝑇 is bounded because:
3
‖𝑇(𝑓)‖ = ‖∫0 𝑓 (𝑥)𝑑𝑥 ‖ ≤ (3 − 0) sup |𝑓(𝑥 )| = 3 ‖𝑓‖∞ .
0≤𝑥≤3
 4

The set of bounded linear operators from 𝑉 to 𝑊, ℒ(𝑉, 𝑊 ), is also a normed

vector space. We can define a norm on ℒ(𝑉, 𝑊 ) by

‖𝑇‖ = inf{𝑀| ‖𝑇(𝑣 )‖𝑊 ≤ 𝑀‖𝑣 ‖𝑉 for all 𝑣 ∈ 𝑉}.

This norm is often called the operator norm.

3
Ex. Let 𝑇: 𝐶[0,3] → ℝ by 𝑇(𝑓) = ∫0 𝑓(𝑥 )𝑑𝑥 . Then

3
‖𝑇 ‖ = inf{𝑀| ‖∫0 𝑓 (𝑥 )𝑑𝑥 ‖ ≤ 𝑀‖𝑓‖∞ for all 𝑓 ∈ 𝐶[0,3]}.

From the previous example we know that ‖𝑇 ‖ ≤ 3. However, we also

3
know that 𝑓 (𝑥 ) = 1 ∈ 𝐶[0,3] and 𝑇(𝑓) = ∫0 1𝑑𝑥 = 3 = 3 ‖𝑓‖∞ , so
‖𝑇 ‖ = 3.
 5

Ex. Describe 𝑁𝑟 (𝑓), a neighborhood of radius 𝑟 centered at 𝑓 ∈ 𝐶[0,1].

𝑁𝑟 (𝑓) = {𝑔 ∈ 𝐶 [0,1]| sup |𝑓 (𝑥 ) − 𝑔(𝑥)| < 𝑟}

0≤𝑥≤1

= {𝑔 ∈ 𝐶 [0,1]| 𝑓(𝑥 ) − 𝑟 < 𝑔(𝑥 ) < 𝑓(𝑥 ) + 𝑟}

𝑦 = 𝑓 (𝑥 ) + 𝑟

0 1

𝑦 = 𝑔(𝑥)

𝑦 = 𝑓(𝑥)

𝑦 = 𝑓 (𝑥) − 𝑟

𝑦 = 𝑔(𝑥) is in 𝑁𝑟 (𝑓) if the graph of 𝑦 = 𝑔(𝑥) lies between the dotted green
curves given by 𝑦 = 𝑓 (𝑥 ) + 𝑟 and 𝑦 = 𝑓 (𝑥 ) + 𝑟 when 0 ≤ 𝑥 ≤ 1.
 6

Ex. Let 𝐸 ⊆ 𝐶[0,1], with 𝐸 = {𝑓 ∈ 𝐶 [0,1]| sup |𝑓 (𝑥 )| < 1}. Notice that
0≤𝑥≤1
𝐸 = 𝐵1 (𝑔(𝑥 ) = 0), the ball of radius 1 around 𝑔(𝑥 ) = 0 in 𝐶[0,1].
a. Is 𝐸 open in 𝐶[0,1]? If so, prove it.
b. If 𝑔 ∈ 𝐶[0,1] and sup |𝑔(𝑥 )| = 1, is 𝑔 a limit point of 𝐸 ?
0≤𝑥≤1
c. Is 𝐸 bounded in 𝐶[0,1]?
d. Is 𝐸 totally bounded in 𝐶[0,1]?

a. Yes, 𝐸 is open in 𝐶[0,1]. To prove this we must show that every

element of 𝐸 is an interior point. That is, given any 𝑓 ∈ 𝐸 , there exists
a neighborhood of 𝑓, 𝑁𝑟 (𝑓 ), such that 𝑁𝑟 (𝑓) ⊆ 𝐸 .

𝑁𝑟 (𝑓) = {𝑔 ∈ 𝐶 [0,1]| 𝑑(𝑓, 𝑔) < 𝑟}

= {𝑔 ∈ 𝐶 [0,1]| sup |𝑓 (𝑥 ) − 𝑔(𝑥 )| < 𝑟}
0≤𝑥≤1

1
(𝐾 + 1)/2
𝐾
1−𝐾
𝑦 = 𝑓(𝑥) +
2

1−𝐾
𝑦 = 𝑓 (𝑥 ) −
2 𝑦 = 𝑓(𝑥)

−1

Since |𝑓 (𝑥)| is a continuous function on [0,1] it attains its maximum

value. Let sup |𝑓(𝑥 )| = 𝐾 < 1.
0≤𝑥≤1
1−𝐾
Choose 𝑟 = .
2
1−𝐾
Then: 𝑁1−𝐾 (𝑓) = {𝑔 ∈ 𝐶 [0,1]| sup |𝑓 (𝑥 ) − 𝑔(𝑥 )| < }.
2 0≤𝑥≤1 2
 7

Claim: 𝑁1−𝐾 (𝑓) ⊆ 𝐸 .

2

We have to show that for any 𝑔 ∈ 𝑁1−𝐾 (𝑓), 𝑔 is also in 𝐸 . That is,
2

we need to show that sup |𝑔(𝑥 )| < 1.

0≤𝑥≤1

1−𝐾
𝑔 ∈ 𝑁1−𝐾 (𝑓) ⟹ sup |𝑔(𝑥 ) − 𝑓(𝑥 )| <
2 0≤𝑥≤1 2

𝐾−1 1−𝐾
< 𝑔(𝑥 ) − 𝑓(𝑥 ) < ; for all 0 ≤ 𝑥 ≤ 1.
2 2

𝐾−1 1−𝐾
𝑓(𝑥 ) + < 𝑔(𝑥 ) < 𝑓 (𝑥 ) +
2 2

Since sup |𝑓 (𝑥 )| = 𝐾 , − 𝐾 ≤ 𝑓(𝑥) ≤ 𝐾; for all 0 ≤ 𝑥 ≤ 1.

0≤𝑥≤1

Thus we have:
𝐾−1 𝐾−1 1−𝐾 1−𝐾
−𝐾 + ≤ 𝑓(𝑥 ) + < 𝑔(𝑥 ) < 𝑓(𝑥 ) + ≤𝐾+
2 2 2 2

𝐾+1 𝐾−1 1−𝐾 𝐾+1

−( 2
) ≤ 𝑓(𝑥 ) + 2
< 𝑔(𝑥 ) < 𝑓(𝑥 ) + 2
≤ 2

But since 0 ≤ 𝐾 < 1 we have:

𝐾+1 𝐾−1 1−𝐾 𝐾+1
−1 < − ( ) ≤ 𝑓(𝑥 ) + < 𝑔(𝑥 ) < 𝑓(𝑥 ) + ≤ < 1.
2 2 2 2

These inequalities hold for all 0 ≤ 𝑥 ≤ 1, so sup |𝑔(𝑥 )| < 1, and 𝑔(𝑥) ∈ 𝐸.
0≤𝑥≤1

Thus 𝑁1−𝐾 (𝑓) ⊆ 𝐸 and 𝐸 is an open set in 𝐶[0,1].

2
 8

b. If sup |𝑔(𝑥 )| = 1, let’s show 𝑔(𝑥) is a limit point of 𝐸 .

0≤𝑥≤1
To be a limit point we must show that every neighborhood of 𝑔(𝑥),
𝑁𝑟 (𝑔), intersects 𝐸 in some point other than 𝑔.

1
𝑟
1−
2

𝑦 = 𝑔(𝑥) + 𝑟

𝑦 = 𝑔 (𝑥 ) − 𝑟 𝑦 = 𝑔(𝑥)
𝑦 = ℎ(𝑥)

𝑟
−1
2
−1

We can always construct a function ℎ ∈ 𝐶[0,1] such that

𝑟
ℎ(𝑥 ) = 𝑔(𝑥 ) if |𝑔(𝑥 )| < 1 −
2
𝑟 𝑟
=1− if 𝑔(𝑥 ) ≥ 1 −
2 2
𝑟 𝑟
= −1 if 𝑔(𝑥 ) ≤ − 1.
2 2

1
Now if we take 𝑟 < , then ℎ(𝑥) ∈ 𝐸 and 𝑁𝑟 (𝑔).
2

1 1
If 𝑟 ≥ 2 , take ℎ(𝑥) defined with 𝑟 = 2 , then ℎ(𝑥) ∈ 𝐸 and 𝑁𝑟 (𝑔).

So 𝑔(𝑥) is a limit point of 𝐸 .

c. Yes, 𝐸 is bounded because for any 𝑓 ∈ 𝐸, 𝑑 (𝑓, 𝑔(𝑥 ) = 0) < 1.

 9

d. No, 𝐸 is not totally bounded.

3
Let 𝑓𝑛 (𝑥 ) be a continuous function that is 0 for 0 ≤ 𝑥 ≤ or
2𝑛+2
3 3 1
≤ 𝑥 ≤ 1 and rises linearly to 𝑓𝑛 (𝑥 ) = at 𝑥 = . So 𝑓𝑛 ∈ 𝐸.
2𝑛+1 4 2𝑛

𝑦 = 𝑓3 (𝑥)
𝑦 = 𝑓2 (𝑥)
3
4

3
Then 𝑑 (𝑓𝑛 , 𝑓𝑚 ) = if 𝑛 ≠ 𝑚. 𝑦 = 𝑓1 (𝑥)
4
1
Thus if we take 𝜖 = , then no
4
finite number of elements in 𝐸
1 1 1
1
0 1
8 4 2
with balls of radius will cover {𝑓𝑛 }
4

where 𝑛 = 1,2, …. This is because

1
each ball of radius can contain at most one of the 𝑓𝑛′ 𝑠.
4

We can see this by assuming that 𝑓𝑛 , 𝑓𝑚 ∈ 𝐵1 (𝑔) for some 𝑔 ∈ 𝐶 [0,1], and 𝑛 ≠ 𝑚.
4

Then by the triangle inequality we get:

3 1 1 1
= 𝑑 (𝑓𝑛 , 𝑓𝑚 ) ≤ 𝑑 (𝑓𝑛 , 𝑔) + 𝑑 (𝑔, 𝑓𝑚 ) < + =2
4 4 4
which is a contradiction. Thus 𝐵1 (𝑔) can contain at most one 𝑓𝑛 .
4

1
Thus no finite number of balls of radius will cover {𝑓𝑛 } and thus no
4
1
finite number of balls of radius will cover 𝐸 .
4
 10

Theorem: 𝑓𝑛 (𝑥) converges uniformly to 𝑓(𝑥) on 𝐼 if and only if for all 𝜖 > 0 there
exists an 𝑁 ∈ ℤ+ , such that for all 𝑥 ∈ 𝐼, if 𝑛, 𝑚 ≥ 𝑁 then |𝑓𝑛 (𝑥 ) − 𝑓𝑚 (𝑥 )| < 𝜖 .

(As we will see shortly this means, if {𝑓𝑛 (𝑥 )} ⊆ 𝐶 (𝐼 ), then {𝑓𝑛 (𝑥 )} converges
to 𝑓(𝑥) ∈ 𝐶(𝐼), if and only if {𝑓𝑛 (𝑥) } is a Cauchy sequence in 𝐶(𝐼)).

Proof: Assume that 𝑓𝑛 (𝑥) converges uniformly to 𝑓(𝑥) on 𝐼.

By the triangle inequality we have:
|𝑓𝑛 (𝑥 ) − 𝑓𝑚 (𝑥 )| ≤ |𝑓𝑛 (𝑥 ) − 𝑓(𝑥 )| + |𝑓 (𝑥 ) − 𝑓𝑚 (𝑥 )|

Since 𝑓𝑛 (𝑥) converges uniformly to 𝑓(𝑥) on 𝐼, there exists 𝑁 ∈ ℤ+ such that if

𝜖
𝑛 ≥ 𝑁 then |𝑓𝑛 (𝑥 ) − 𝑓 (𝑥 )| < for any 𝑥 ∈ 𝐼.
2
𝜖
And, of course, if 𝑚 ≥ 𝑁 then |𝑓𝑚 (𝑥 ) − 𝑓(𝑥 )| < for any 𝑥 ∈ 𝐼.
2
Thus if 𝑚, 𝑛 ≥ 𝑁 then we have for any 𝑥 ∈ 𝐼:
𝜖 𝜖
|𝑓𝑛 (𝑥 ) − 𝑓𝑚 (𝑥 )| ≤ |𝑓𝑛 (𝑥 ) − 𝑓(𝑥 )| + |𝑓 (𝑥 ) − 𝑓𝑚 (𝑥 )| < + = 𝜖.
2 2

Now assume for all 𝜖 > 0 there exists an 𝑁 ∈ ℤ+ , such that for all 𝑥 ∈ 𝐼,

if 𝑛, 𝑚 ≥ 𝑁 then |𝑓𝑛 (𝑥 ) − 𝑓𝑚 (𝑥 )| < 𝜖 .

For each 𝑥 ∈ 𝐼, {𝑓𝑛 (𝑥 )} is a Cauchy sequence of real numbers and thus

converges to a real number 𝑓(𝑥).

So lim 𝑓𝑛 (𝑥 ) = 𝑓(𝑥) (this is a pointwise limit).

𝑛→∞
 11

Now we must show that 𝑓𝑛 (𝑥) converges uniformly to 𝑓(𝑥).

By assumption, there exists an 𝑁 ∈ ℤ+ , such that for all 𝑥 ∈ 𝐼, if 𝑛, 𝑚 ≥ 𝑁

then |𝑓𝑛 (𝑥 ) − 𝑓𝑚 (𝑥 )| < 𝜖.

This is true for all 𝑚 ≥ 𝑁, so let 𝑚 go to ∞. So we have:
there exists an 𝑁 ∈ ℤ+ , such that for all 𝑥 ∈ 𝐼, if 𝑛, 𝑚 ≥ 𝑁

then |𝑓𝑛 (𝑥 ) − 𝑓(𝑥 )| ≤ 𝜖.

and 𝑓𝑛 (𝑥) converges to 𝑓(𝑥) uniformly.

Now let’s see why a set of bounded uniformly convergent continuous functions
must converge to a bounded continuous function. Suppose |𝑓𝑛 (𝑥 )| ≤ 𝑀𝑛 for all
𝑥 ∈ 𝐼 and each 𝑛. How do we know that as 𝑛 goes to infinity, 𝑀𝑛 doesn’t go to
infinity?
By the previous theorem we know that any Cauchy sequence in 𝐶(𝐼), {𝑓𝑛 (𝑥)},
converges to uniformly to some 𝑓(𝑥) on 𝐼 (which must be continuous since all of
the 𝑓𝑛′ 𝑠 are). Thus we have for all 𝜖 > 0 there exists an 𝑁 ∈ ℤ+ , such that for all
𝑥 ∈ 𝐼 , if 𝑛 ≥ 𝑁 then |𝑓(𝑥 ) − 𝑓𝑛 (𝑥 )| < 𝜖.
In particular, |𝑓 (𝑥 ) − 𝑓𝑁 (𝑥 )| < 𝜖 for all 𝑥 ∈ 𝐼. Thus we have:

−𝜖 < 𝑓 (𝑥 ) − 𝑓𝑁 (𝑥 ) < 𝜖
𝑓𝑁 (𝑥 ) − 𝜖 < 𝑓(𝑥 ) < 𝑓𝑁 (𝑥 ) + 𝜖
−𝑀𝑁 − 𝜖 ≤ 𝑓𝑁 (𝑥 ) − 𝜖 < 𝑓(𝑥 ) < 𝑓𝑁 (𝑥 ) + 𝜖 ≤ 𝑀𝑁 + 𝜖

Thus |𝑓 (𝑥 )| ≤ 𝑀𝑁 + 𝜖 and 𝑓(𝑥 ) is bounded.

Hence any Cauchy sequence in 𝐶(𝐼) must converge to a bounded continuous

function, 𝑓(𝑥), thus 𝑓(𝑥) ∈ 𝐶(𝐼) and 𝐶(𝐼) is complete.