Functional Analysis

Linear Space

A triple (𝑁, +,∙ ) is a vector space or a linear space over a field 𝔽 if

i. (𝑁, +)is an abelian group .

ii. For each 𝛼 ∈ 𝔽, 𝑣 ∈ 𝑁, there is a defined 𝛼𝑣 ∈ 𝑉 such that

a) 𝛼 (𝑣 + 𝑤) = 𝛼𝑣 + 𝛼𝑤 ,
b) (𝛼 + 𝛽)𝑣 = 𝛼𝑣 + 𝛽𝑣 ,
c) (𝛼𝛽)𝑣 = 𝛼 (𝛽𝑣 ) ,
d) 1 ∙ 𝑣 = 𝑣 , ∀ 𝛼, 𝛽 ∈ 𝔽, 𝑣, 𝑤 ∈ 𝑁 .

Examples :

1. ℝ is a linear space over ℝ .

2. ℂis a linear space over ℝ or ℂ .
3. ℝis not a linear space over ℂ (? ) .
4. ℝ𝑛 is a linear space over ℝ with respect to the operations
+∶ (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) + (𝑦1 , 𝑦2 , … , 𝑦𝑛 ) = (𝑥1 + 𝑦1 , 𝑥2 + 𝑦2 , … , 𝑥𝑛 + 𝑦𝑛 )
∙ ∶ 𝛼 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = (𝛼𝑥1 , 𝛼𝑥2 , … , 𝛼𝑥𝑛 )
Norm or length of a vector

Suppose 𝑁 is a linear space over 𝔽(ℝ or ℂ), a function ‖∙‖ ∶ 𝑁 ⟶ ℝ+ ∪ {0} is said to be a norm on 𝑁 if
i. ‖𝑥‖ ≥ 0 and ‖𝑥‖ = 0 ⇔ 𝑥 = 0 , ∀ 𝑥 ∈ 𝑁 .
ii. ‖𝛼 𝑥‖ = |𝛼|‖𝑥‖ , ∀ 𝛼 ∈ 𝔽, 𝑥 ∈ 𝑁 .
iii. ‖𝑥 + 𝑦‖ ≤ ‖𝑥 ‖ + ‖𝑦‖ , ∀ 𝑥, 𝑦 ∈ 𝑁 .

The pair (𝑁, ‖∙‖) is called a normed linear space .

Example :(ℝ, |∙|) is normed linear space . (Verify)

Note : Every normed linear space is a metric space .

Proof : Let (𝑁, ‖∙‖) be a normed linear space.

Define 𝑑 (𝑥, 𝑦) = ‖𝑥 − 𝑦‖ , ∀ 𝑥, 𝑦 ∈ 𝑁 .

Clearly, 𝑑 satisfies all the properties of a metric(Verify) and hence (𝑁, 𝑑 ) is a metric space.

Convergence in (𝑿, 𝒅)

// A sequence {𝑥𝑛 } in (𝑋, 𝑑 ) is said to be Cauchy if for every 𝜀 > 0, there exists an integer 𝑛0 > 0 such that
𝑑 (𝑥𝑛 , 𝑥𝑚 ) < 𝜀 , ∀ 𝑛, 𝑚 ≥ 𝑛0 .

Equivalently, a sequence{𝑥𝑛 } in (𝑁, ‖∙‖) is said to be Cauchy if for every 𝜀 > 0, there exists an integer 𝑛0 > 0 such
that ‖𝑥𝑛 − 𝑥𝑚 ‖ < 𝜀 , ∀ 𝑛, 𝑚 ≥ 𝑛0 .
 // A sequence {𝑥𝑛 } in (𝑋, 𝑑 ) is said to converge to 𝑥 ∈ 𝑋 if for every 𝜀 > 0, there exists an integer 𝑛0 > 0 such
that 𝑑 (𝑥𝑛 , 𝑥) < 𝜀 , ∀ 𝑛 ≥ 𝑛0 .

Equivalently, a sequence{𝑥𝑛 } in a normed linear space (𝑁, ‖∙‖) is said to converge to 𝑥 ∈ 𝑁 if for every 𝜀 > 0, there
exists an integer 𝑛0 > 0 such that ‖𝑥𝑛 − 𝑥‖ < 𝜀 , ∀ 𝑛 ≥ 𝑛0 .

Note : In any space, every convergent sequence is Cauchy, but a Cauchy sequence need not be convergent .

// A metric space in which every Cauchy sequence is convergent is called a complete metric space .

// A normed linear space which is complete as a metric is called a Banachspace .

Theorem : In a normed linear space, the following hold

i. Norm is continuous .
ii. Addition is jointly continuous.
iii. Scalar multiplication is continuous .

Proof : Let (𝑁, ‖∙‖) be a normed linear space .

Let {𝑥𝑛 } be a sequence in 𝑁 converging to 𝑥 in 𝑁 . 𝑖. 𝑒., {𝑥𝑛 } ⟶ 𝑥 .

To prove, {‖𝑥𝑛 ‖} ⟶ ‖𝑥‖ .

Now {𝑥𝑛 } ⟶ 𝑥 ⟹ 𝑑(𝑥𝑛 , 𝑥) ⟶ 0 as 𝑛 ⟶ ∞ or ‖𝑥𝑛 − 𝑥‖ ⟶ 0 as 𝑛 ⟶ ∞ .

Since
|‖𝑥‖ − ‖𝑦‖| ≤ ‖𝑥 − 𝑦‖ ,
we have |‖𝑥𝑛 ‖ − ‖𝑥‖| ≤ ‖𝑥𝑛 − 𝑥‖ ⟶ 0 as 𝑛 ⟶ ∞ .

⟹ {‖𝑥𝑛 ‖} ⟶ ‖𝑥 ‖ . Hence, norm is continuous.

Let {𝑥𝑛 } and {𝑦𝑛 } be two sequences in 𝑁 converging to 𝑥 and 𝑦 respectively in 𝑁 .

To prove, {𝑥𝑛 + 𝑦𝑛 } ⟶ 𝑥 + 𝑦 .

Consider
‖(𝑥𝑛 + 𝑦𝑛 ) − (𝑥 + 𝑦)‖ = ‖(𝑥𝑛 − 𝑥) + (𝑦𝑛 − 𝑦)‖

≤ ‖𝑥𝑛 − 𝑥‖ + ‖𝑦𝑛 − 𝑦‖ ⟶ 0
as𝑛 ⟶ ∞ .

⟹ {𝑥𝑛 + 𝑦𝑛 } ⟶ 𝑥 + 𝑦 .

Let {𝑥𝑛 } be a sequence in 𝑁 converging to 𝑥 in 𝑁 and {𝛼𝑛 } be a sequence in 𝔽(ℝ or ℂ) converging to 𝛼

in 𝔽 .

⟹ {‖𝑥𝑛 ‖} ⟶ ‖𝑥 ‖&{|𝛼𝑛 |} ⟶ |𝛼| .

Consider
‖𝛼𝑛 𝑥𝑛 − 𝛼𝑥‖ = ‖𝛼𝑛 𝑥𝑛 − 𝛼𝑛 𝑥 + 𝛼𝑛 𝑥 − 𝛼𝑥‖

= ‖𝛼𝑛 (𝑥𝑛 − 𝑥) + (𝛼𝑛 − 𝛼)𝑥‖

 ≤ |𝛼𝑛 |‖𝑥𝑛 − 𝑥‖ + |𝛼𝑛 − 𝑎|‖𝑥 ‖ ⟶ 0
as𝑛 ⟶ ∞ .

⟹ {𝛼𝑛 𝑥𝑛 } ⟶ 𝛼𝑥 .

Examples for normed linear spaces

1. 𝐶 [0,1], the set of all continuous real valued functions on [0,1] .

It is easy to verify that 𝐶 [0,1] forms a linear space over ℝ (Verify) with respect to the binary operations

∀ 𝑓, 𝑔 ∈ 𝐶 [0,1], 𝛼 ∈ 𝔽,we have

+ ∶ (𝑓 + 𝑔)(𝑥) = 𝑓 (𝑥) + 𝑔(𝑥)

∙ ∶ 𝛼 𝑓)(𝑥) = 𝛼 𝑓(𝑥) ,
( ∀ 𝑥 ∈ [0,1] .

Define ‖∙‖ on 𝐶 [0,1] by

1

‖𝑓‖ = ∫|𝑓 (𝑥)| 𝑑𝑥 .

0

i. ∀ 𝑓 ∈ 𝐶 [0,1], |𝑓(𝑥)| ≥ 0, ∀ 𝑥 ∈ [0,1] .

1

⟹ ∫|𝑓 (𝑥)| 𝑑𝑥 ≥ 0 or‖𝑓‖ ≥ 0 .

0
 1

‖𝑓‖ = 0 ⇔ ∫|𝑓 (𝑥)| 𝑑𝑥 ≥ 0 ⇔ |𝑓 (𝑥)| = 0 or 𝑓(𝑥) = 0, ∀ 𝑥 ∈ [0,1] ⇔ 𝑓 =0.

0

ii. ∀ 𝑓 ∈ 𝐶 [0,1], 𝛼 ∈ 𝔽,
1 1 1

‖𝛼𝑓‖ = ∫|(𝛼𝑓)(𝑥)| 𝑑𝑥 = ∫|𝛼 𝑓 (𝑥)| 𝑑𝑥 = |𝛼| ∫|𝑓 (𝑥)| 𝑑𝑥 = |𝛼|‖𝑓‖ .

0 0 0

iii. ∀ 𝑓, 𝑔 ∈ 𝐶 [0,1],
1 1 1

‖𝑓 + 𝑔‖ = ∫|(𝑓 + 𝑔)(𝑥)| 𝑑𝑥 = ∫|𝑓 (𝑥) + 𝑔(𝑥)| 𝑑𝑥 ≤ ∫(|𝑓 (𝑥)| + |𝑔(𝑥)|) 𝑑𝑥

0 0 0
1 1

= ∫|𝑓 (𝑥)| 𝑑𝑥 + ∫|𝑔(𝑥)| 𝑑𝑥

0 0
= ‖𝑓‖ + ‖𝑔‖ .

Hence, ‖∙‖ is a norm on𝐶 [0,1] .

If we define
1

𝑑 (𝑓, 𝑔) = ‖𝑓 − 𝑔‖ = ∫|(𝑓 − 𝑔)(𝑥)| 𝑑𝑥 ,

0
then𝑑 forms a metric on the normed linear space 𝐶 [0,1] . ( Verify )
 On 𝐶 [0,1] ,verify whether the function ‖∙‖ defined by

‖𝑓‖ = Sup |𝑓 (𝑥)|

𝑥∈[0,1]
forms a norm or not.

2. 𝑙2 𝑛 is a normed linear space where 𝑙2 𝑛 is the linear space of all 𝑛 −tuples (either ℝ𝑛 or ℂ𝑛 ) associated with
the norm
‖𝑥‖2 = (|𝑥1 |2 + |𝑥2 |2 + ⋯ + |𝑥𝑛 |2 )1/2 .

i. ∀ 𝑥 = (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) ∈ 𝑙2 𝑛 ,
1
𝑛 2

|𝑥𝑖 | ≥ 0 , ∀ 𝑖 ⟹ |𝑥𝑖 |2 ≥ 0, ∀ 𝑖 ⟹ (∑|𝑥𝑖 |2 ) ≥ 0 or‖𝑥‖2 ≥ 0 .

𝑖=1

Further,
1
𝑛 2 𝑛

‖𝑥‖2 = 0 ⇔ (∑|𝑥𝑖 |2 ) = 0 ⇔ ∑|𝑥𝑖 |2 = 0

𝑖=1 𝑖=1

⇔ |𝑥𝑖 |2 = 0 or𝑥𝑖 = 0, ∀ 𝑖 ⇔ 𝑥 =0.

ii. ∀ 𝑥 ∈ 𝑙2 𝑛 , 𝛼 ∈ 𝔽,
1 1 1
𝑛 2 𝑛 2 𝑛 2

‖𝛼𝑥‖2 = (∑|𝛼𝑥𝑖 |2 ) = (|𝛼| ∑|𝑥𝑖 |2 ) = |𝛼| (∑|𝑥𝑖 |2 ) = |𝛼 |‖𝑥‖2 .

𝑖=1 𝑖=1 𝑖=1
iii. Lemma(Cauchy’s inequality):

∀ 𝑥, 𝑦 ∈ 𝑙2 𝑛 ,
𝑛

∑|𝑥𝑖 𝑦𝑖 | ≤ ‖𝑥‖2 ‖𝑦‖2 .

𝑖=1
Proof :

Either if 𝑥 = 0 or 𝑦 = 0, then the above inequality holds trivially.

Assume 𝑥 ≠ 0 & 𝑦 ≠ 0.

For any two non-negative real numbers 𝑎 & 𝑏 ,

(𝑎 + 𝑏 )
𝑎 1/2 𝑏 1/2 ≤
2

For each 𝑖, making use of the above inequality for the non-negative real numbers
2 2
|𝑥𝑖 | |𝑦𝑖 |
𝑎𝑖 = ( ) &𝑏𝑖 = ( ) ,
‖𝑥‖2 ‖𝑦‖2
we get
|𝑥 | 2 |𝑦 | 2
|𝑥𝑖 | |𝑦𝑖 | ((‖𝑥‖𝑖 ) + (‖𝑦‖𝑖 ) )
2 2
≤ .
‖𝑥‖2 ‖𝑦‖2 2

Summing the 𝑛 inequalities, we get

∑𝑛𝑖=1|𝑥𝑖 𝑦𝑖 | (1 + 1)
≤ =1.
‖𝑥‖2 ‖𝑦‖2 2
Minkowski’sIneguality :
 ∀ 𝑥, 𝑦 ∈ 𝑙2 𝑛 ,
1
𝑛 2

‖𝑥 + 𝑦‖2 = (∑|𝑥𝑖 + 𝑦𝑖 |2 )
𝑖=1

𝑛 𝑛

⟹ (‖𝑥 + 𝑦‖2 )2 = ∑|𝑥𝑖 + 𝑦𝑖 |2 = ∑|𝑥𝑖 + 𝑦𝑖 ||𝑥𝑖 + 𝑦𝑖 |

𝑖=1 𝑖=1

𝑛 𝑛 𝑛

≤ ∑|𝑥𝑖 + 𝑦𝑖 |(|𝑥𝑖 | + |𝑦𝑖 |) = ∑|𝑥𝑖 ||𝑥𝑖 + 𝑦𝑖 | + ∑|𝑦𝑖 ||𝑥𝑖 + 𝑦𝑖 |

𝑖=1 𝑖=1 𝑖=1

≤ ‖𝑥‖2 ‖𝑥 + 𝑦‖2 + ‖𝑦‖2 ‖𝑥 + 𝑦‖2

⟹ ‖𝑥 + 𝑦‖2 ≤ ‖𝑥‖2 + ‖𝑦‖2 .

Hence, 𝑙2 𝑛 is a normed linear space.

3. 𝑙𝑝 𝑛 (1 ≤ 𝑝 < ∞) is a normed linear space where 𝑙𝑝 𝑛 is the linear space of all 𝑛 −tuples (either ℝ𝑛 or ℂ𝑛 )
associated with the norm
‖𝑥‖𝑝 = (|𝑥1 |𝑝 + |𝑥2 |𝑝 + ⋯ + |𝑥𝑛 |𝑝 )1/𝑝 .

i. ∀ 𝑥 ∈ 𝑙𝑝 𝑛 ,
 1
𝑛 𝑝

|𝑥𝑖 | ≥ 0 , ∀ 𝑖 ⟹ |𝑥𝑖 |𝑝 ≥ 0, ∀ 𝑖 ⟹ (∑|𝑥𝑖 |𝑝 ) ≥ 0 or‖𝑥‖𝑝 ≥ 0 .

𝑖=1

Further,
1
𝑛 𝑝 𝑛

‖𝑥‖𝑝 = 0 ⇔ |𝑝
(∑|𝑥𝑖 ) = 0 ⇔ ∑|𝑥𝑖 |𝑝 = 0
𝑖=1 𝑖=1

⇔ |𝑥𝑖 |𝑝 = 0 or𝑥𝑖 = 0, ∀ 𝑖 ⇔ 𝑥 =0.

ii. ∀ 𝑥 ∈ 𝑙𝑝 𝑛 , 𝛼 ∈ 𝔽,
1 1 1
𝑛 𝑝 𝑛 𝑝 𝑛 𝑝

‖𝛼𝑥‖𝑝 = (∑|𝛼𝑥𝑖 |𝑝 ) = (|𝛼| ∑|𝑥𝑖 |𝑝 ) = |𝛼| (∑|𝑥𝑖 |𝑝 ) = |𝛼 |‖𝑥‖𝑝 .

𝑖=1 𝑖=1 𝑖=1

iii. Holder’s inequality:

∀ 𝑥, 𝑦 ∈ 𝑙𝑝 𝑛 ,
𝑛
1 1
∑|𝑥𝑖 𝑦𝑖 | ≤ ‖𝑥‖𝑝 ‖𝑦‖𝑞 , where + = 1 .
𝑝 𝑞
𝑖=1

Lemma : For any two non-negative real numbers 𝑎 & 𝑏 ,

𝑎 𝑏 1 1
𝑎 1/𝑝 𝑏 1/𝑞 ≤ + , where + = 1 .
𝑝 𝑞 𝑝 𝑞
Proof of lemma:

Let 𝑓 (𝑡 ) = 1 − 𝑡 𝑘 + 𝑘(𝑡 − 1), where 0 < 𝑘 < 1, 𝑡 ≥ 1 .

1
𝑓 ′ (𝑡 ) = 𝑘 (1 − )>0
𝑡 1−𝑘

⟹ 𝑓is an increasing function and hence 𝑓 (𝑡 ) ≥ 𝑓 (1) = 0.

⟹ 1 − 𝑡 𝑘 + 𝑘 (𝑡 − 1) ≥ 0 or𝑡 𝑘 ≤ 𝑘𝑡 + 1 − 𝑘 , 𝑡 ≥ 1, 0 < 𝑘 < 1 .

𝑎 1
Case 1 : If 𝑎 ≥ 𝑏, taking 𝑡 = & 𝑘 = , we arrive at the proof of lemma.
𝑏 𝑝
𝑏 1
Case 2 : If 𝑎 < 𝑏, taking 𝑡 = & 𝑘 = , we arrive at the proof of lemma.
𝑎 𝑞

Proof of Holder’s inequality :

For each 𝑖, taking

𝑝 𝑞
|𝑥𝑖 | |𝑦𝑖 |
𝑎𝑖 = ( ) &𝑏𝑖 = ( ) ,
‖𝑥 ‖𝑝 ‖𝑦‖𝑞
we get
|𝑥𝑖 | |𝑦𝑖 | 1 |𝑥𝑖 |𝑝 1 |𝑦𝑖 |𝑞
≤ + .
‖𝑥 ‖𝑝 ‖𝑦‖𝑞 𝑝 ‖𝑥‖𝑝 𝑝 𝑞 ‖𝑦‖𝑞 𝑞

Summing the 𝑛 inequalities, we get

 𝑛
1 1
∑|𝑥𝑖 𝑦𝑖 | ≤ ‖𝑥‖𝑝 ‖𝑦‖𝑞 , where + = 1 .
𝑝 𝑞
𝑖=1

Minkowski’sIneguality :

∀ 𝑥, 𝑦 ∈ 𝑙𝑝 𝑛 ,
1
𝑛 𝑝

‖𝑥 + 𝑦‖𝑝 = (∑|𝑥𝑖 + 𝑦𝑖 |𝑝 )
𝑖=1

𝑛 𝑛
𝑝
⟹ (‖𝑥 + 𝑦‖𝑝 ) = ∑|𝑥𝑖 + 𝑦𝑖 |𝑝 = ∑|𝑥𝑖 + 𝑦𝑖 |𝑝−1 |𝑥𝑖 + 𝑦𝑖 |
𝑖=1 𝑖=1

𝑛 𝑛 𝑛

≤ ∑|𝑥𝑖 + 𝑦𝑖 |𝑝−1 (| 𝑥𝑖 | + |𝑦𝑖 |) = ∑|𝑥𝑖 ||𝑥𝑖 + 𝑦𝑖 |𝑝−1 + ∑|𝑦𝑖 ||𝑥𝑖 + 𝑦𝑖 |𝑝−1

𝑖=1 𝑖=1 𝑖=1

1 1
𝑛 𝑞 𝑛 𝑞

≤ ‖𝑥 ‖𝑝 (∑(|𝑥𝑖 + 𝑦𝑖 |𝑝−1 )𝑞 ) + ‖𝑦‖𝑝 (∑(|𝑥𝑖 + 𝑦𝑖 |𝑝−1 )𝑞 )

𝑖=1 𝑖=1

1𝑝 1𝑝
𝑛 𝑝𝑞 𝑛 𝑝𝑞

= ‖𝑥‖𝑝 (∑|𝑥𝑖 + 𝑦𝑖 |𝑝 ) + ‖𝑦‖𝑝 (∑|𝑥𝑖 + 𝑦𝑖 |𝑝 )

𝑖=1 𝑖=1

𝑝/𝑞 𝑝/𝑞
= ‖𝑥 ‖𝑝 ‖𝑥 + 𝑦‖𝑝 + ‖𝑦‖𝑝 ‖𝑥 + 𝑦‖𝑝
 𝑝/𝑞
Dividing by ‖𝑥 + 𝑦‖𝑝 , we get

‖𝑥 + 𝑦‖𝑝 ≤ ‖𝑥‖𝑝 + ‖𝑦‖𝑝 .

Hence, 𝑙𝑝 𝑛 is a normed linear space.

Exercise :Verify whether𝑙∞ 𝑛 is a normed linear space where 𝑙∞ 𝑛 is the linear space of all 𝑛 −tuples
(eitherℝ𝑛 or ℂ𝑛 ) associated with the function

‖𝑥‖∞ = max{|𝑥1 |, |𝑥2 |, … |𝑥𝑛 |} .

Note:For a sequence of functions {𝑓𝑛 } from 𝑋 into ℝ, {𝑓𝑛 }converges pointwise to a function 𝑓 from 𝑋 into ℝ if for
each 𝑥 ∈ 𝑋, ∃ an integer 𝑛0 > 0 such that |𝑓𝑛 (𝑥) − 𝑓(𝑥)| < 𝜀, ∀ 𝑛 ≥ 𝑛0 . Here 𝒏𝟎 = 𝒏𝟎 (𝒙, 𝜺) .

If for all points 𝑥 ∈ 𝑋, the same integer 𝑛0 > 0 works, then we say {𝑓𝑛 }converges uniformly to 𝑓on .
In this case, 𝒏𝟎 = 𝒏𝟎 (𝜺) .

𝒍𝒑 𝒏 (𝟏 ≤ 𝒑 < ∞)is complete and hence a Banach space

𝑛
Every element 𝑥 ∈ 𝑙𝑝 𝑛 can be regarded as a function 𝑥: {1,2, … 𝑛} ⟶ 𝑙𝑝 such that 𝑥 = (𝑥 (1), 𝑥 (2), … 𝑥 (𝑛)) .

𝑖. 𝑒., 𝑓𝑚 ∈ 𝑙𝑝 𝑛 ⟹ 𝑓𝑚 = (𝑓𝑚 (1), 𝑓𝑚 (2), … 𝑓𝑚 (𝑛)) .

Let {𝑓𝑚 } be a Cauchy sequence in 𝑙𝑝 𝑛 .

For every 𝜀 > 0, there exists an integer 𝑛0 > 0 such that 𝑑 (𝑓𝑚 , 𝑓𝑚′ ) < 𝜀 , ∀ 𝑚, 𝑚′ ≥ 𝑛0 .

⟹ ‖𝑓𝑚 − 𝑓𝑚′ ‖ < 𝜀 , ∀ 𝑚, 𝑚′ ≥ 𝑛0 .

1
𝑛 𝑝

⟹ (∑|𝑓𝑚 (𝑖) − 𝑓𝑚′ (𝑖)|𝑝 ) < 𝜀 , ∀ 𝑚, 𝑚′ ≥ 𝑛0 .

𝑖=1

⟹ ∑|𝑓𝑚 (𝑖) − 𝑓𝑚′ (𝑖)|𝑝 < 𝜀 𝑝 , ∀ 𝑚, 𝑚′ ≥ 𝑛0 .

𝑖=1

⟹ |𝑓𝑚 (𝑖) − 𝑓𝑚′ (𝑖)|𝑝 < 𝜀 𝑝 , ∀ 𝑚, 𝑚′ ≥ 𝑛0 .

⟹ |𝑓𝑚 (𝑖) − 𝑓𝑚′ (𝑖)| < 𝜀 , ∀ 𝑚, 𝑚′ ≥ 𝑛0 , ∀ 𝑖 .

⟹ {𝑓𝑚 (𝑖)}is a Cauchy sequence in ℝ which is complete , ∀ 𝑖 .

⟹ {𝑓𝑚 (𝑖)}converges to some point 𝑓(𝑖) , 𝑖 = 1,2, … 𝑛 .

Then, {𝑓𝑚 (𝑖)} converges uniformly to 𝑓 (𝑖).

⟹ ∀ 𝑖, for every 𝜀 > 0, there exists an integer 𝑛0 > 0 such that |𝑓𝑚 (𝑖) − 𝑓 (𝑖)| < 𝜀 , ∀ 𝑚 ≥ 𝑛0 .

𝜀𝑝 𝜀𝑝
⟹ for > 0, we have |𝑓𝑚 (𝑖) − 𝑓𝑚′ (𝑖)|𝑝 < , ∀ 𝑚 ≥ 𝑛0 , 𝑖 = 1,2, … 𝑛 .
𝑛 𝑛
Adding the 𝑛 inequalities, we get
𝑛
𝜀𝑝
∑|𝑓𝑚 (𝑖) − 𝑓 (𝑖 )|𝑝 < 𝑛 = 𝜀 𝑝 , ∀ 𝑚 ≥ 𝑛0 .
𝑛
𝑖=1

1
𝑛 𝑝

⟹ (∑|𝑓𝑚 (𝑖) − 𝑓 (𝑖)|𝑝 ) < 𝜀 , ∀ 𝑚 ≥ 𝑛0 .

𝑖=1

⟹ ‖𝑓𝑚 − 𝑓‖ < 𝜀 , ∀ 𝑚 ≥ 𝑛0 .

⟹ 𝑑(𝑓𝑚 , 𝑓) < 𝜀 , ∀ 𝑚 ≥ 𝑛0 ,where𝑓 ∈ 𝑙𝑝 𝑛 .

⟹ 𝑙𝑝 𝑛 is complete and hence a Banach space .

Note :𝐶 [0,1] is the set of all continuous and bounded real valued functions on [0,1] and 𝐵[0,1] is the set of all
bounded real valued functions on [0,1] .

Theorem :𝐶 [0,1] is a closed subspace of 𝐵[0,1].

Proof : Clearly 𝐶 [0,1] is a subspace of 𝐵[0,1] (Verify) .

To prove𝐶 [0,1] is closed.

Let 𝑓 be limit point of 𝐶 [0,1] .

Then for every 𝜀 > 0, (𝑆𝜀⁄3 (𝑓) − {𝑓}) ∩ 𝐶 [0,1] ≠ ∅ .

Let 𝑓0 ∈ (𝑆𝜀⁄3 (𝑓) − {𝑓}) ∩ 𝐶 [0,1] .

⟹ 𝑓0 ∈ (𝑆𝜀⁄3 (𝑓) − {𝑓})and𝑓0 ∈ 𝐶 [0,1] .

⟹ 𝑓0 ∈ 𝑆𝜀⁄3 (𝑓)&𝑓0 ≠ 𝑓 .

⟹ ‖𝑓0 − 𝑓‖ < 𝜀⁄3 .

𝑖. 𝑒., |(𝑓0 − 𝑓)(𝑥)| < 𝜀⁄3 or|𝑓0 (𝑥) − 𝑓 (𝑥)| < 𝜀⁄3 , ∀ 𝑥 ∈ [0,1] .

Now 𝑓0 ∈ 𝐶 [0,1] or 𝑓0 is bounded and continuous on [0,1] ⟹ for every 𝜀 > 0 (𝜀⁄3 > 0), ∃ 𝛿 > 0
such that |𝑓0 (𝑥) − 𝑓0 (𝑦)| < 𝜀⁄3 whenever |𝑥 − 𝑦| < 𝛿 .

Let 𝑥0 ∈ [0,1] be arbitrary. Consider

|𝑓 (𝑥) − 𝑓 (𝑥0 )| = |𝑓 (𝑥) − 𝑓0 (𝑥) + 𝑓0 (𝑥) − 𝑓0 (𝑥0 ) + 𝑓0 (𝑥0 ) − 𝑓 (𝑥0 )|

≤ |𝑓 (𝑥) − 𝑓0 (𝑥)| + |𝑓0 (𝑥) − 𝑓0 (𝑥0 )| + |𝑓0 (𝑥0 ) − 𝑓 (𝑥0 )|

< 𝜀⁄3 + 𝜀⁄3 + 𝜀⁄3 = 𝜀 , whenever|𝑥 − 𝑥0 | < 𝛿 .

⟹ 𝑓is continuous at 𝑥0 ∈ [0,1] and hence on [0,1] .

 Further ∀ 𝑥 ∈ [0,1] ,

|𝑓 (𝑥)| = |𝑓 (𝑥) − 𝑓0 (𝑥) + 𝑓0 (𝑥)| ≤ |𝑓 (𝑥) − 𝑓0 (𝑥)| + |𝑓0 (𝑥)|

< 𝜀⁄3 + 𝑘 ( ∵ 𝑓0 is bounded)

⟹ 𝑓 ∈ 𝐶 [0,1]and hence 𝐶 [0,1] is a closed subspace of 𝐵[0,1] .

Examples of Banach Spaces

1. 𝐵(𝑋 ), the set of all bounded scalar valued functions on 𝑋 . 𝑖. 𝑒., 𝐵(𝑋 ) = 𝐵(𝑋, ℝ)or 𝐵(𝑋 ) = 𝐵(𝑋, ℂ) .

Under the operations

+ ∶ (𝑓 + 𝑔)(𝑥) = 𝑓 (𝑥) + 𝑔(𝑥)
∙ ∶ (𝛼 𝑓)(𝑥) = 𝛼 𝑓 (𝑥) , ∀ 𝑥∈𝑋 ,
𝐵(𝑋 )is a linear space. ( Verify )

Define
‖𝑓‖ = sup{|𝑓 (𝑥)|: 𝑥 ∈ 𝑋 } = sup{|𝑓 (𝑥)|} , ∀ 𝑓 ∈ 𝐵 (𝑋 ) .
𝑥∈𝑋

We verify if the above function is a norm.

i. Since |𝑓 (𝑥)| ≥ 0, ∀ 𝑥 ∈ 𝑋 , sup{|𝑓 (𝑥)|} ≥ 0 ⟹ ‖𝑓‖ ≥ 0 , ∀ 𝑓 ∈ 𝐵 (𝑋 ) .

𝑥∈𝑋
 Further,
‖𝑓‖ = 0 ⟺ sup{|𝑓 (𝑥)|} = 0 ⟺ |𝑓 (𝑥)| = 0 or 𝑓(𝑥) = 0, ∀ 𝑥 ∈𝑋 ⟺𝑓 =0.
𝑥∈𝑋

ii. ∀ 𝑓 ∈ 𝐵(𝑋 ), ∀ 𝛼 ∈ 𝔽,

‖𝛼𝑓‖ = sup{|(𝛼𝑓)(𝑥)|} = sup{|𝛼 𝑓(𝑥)|} = |𝛼| sup{|𝑓 (𝑥)|} = |𝛼 |‖𝑓‖ .

𝑥∈𝑋 𝑥∈𝑋 𝑥∈𝑋

iii. ∀ 𝑓, 𝑔 ∈ 𝐵(𝑋 ),
‖𝑓 + 𝑔‖ = sup{|(𝑓 + 𝑔)(𝑥)|} = sup{|𝑓 (𝑥) + 𝑔(𝑥)|}
𝑥∈𝑋 𝑥∈𝑋

≤ sup{|𝑓 (𝑥)| + |𝑔(𝑥)|}

𝑥∈𝑋

= sup{|𝑓 (𝑥)|} + sup{|𝑔(𝑥)|} = ‖𝑓‖ + ‖𝑔‖ .

𝑥∈𝑋 𝑥∈𝑋

Hence, (𝐵(𝑋 ), ‖∙‖) is a normed linear space .

To show that (𝐵(𝑋 ), ‖∙‖) is complete.

Let {𝑓𝑛 } be a Cauchy sequence in 𝐵(𝑋 ) .

Then for every 𝜀 > 0, there exists an integer 𝑛0 > 0 such that 𝑑 (𝑓𝑚 , 𝑓𝑛 ) < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

⟹ ‖𝑓𝑚 − 𝑓𝑛 ‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

⟹ sup{|(𝑓𝑚 − 𝑓𝑛 )(𝑥)|} < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

𝑥∈𝑋

⟹ sup{|𝑓𝑚 (𝑥) − 𝑓𝑛 (𝑥)|} < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

𝑥∈𝑋
 ⟹ |𝑓𝑚 (𝑥) − 𝑓𝑛 (𝑥)| < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 , ∀𝑥 ∈𝑋.

⟹ {𝑓𝑛 (𝑥)}is Cauchy in ℝ or ℂ for each 𝑥 ∈ 𝑋 .

Since ℝ&ℂ are complete, {𝑓𝑛 (𝑥)} ⟶ 𝑓 (𝑥), for some 𝑓: 𝑋 ⟶ ℝ or ℂ, ∀ 𝑥 ∈ 𝑋 .

𝑖. 𝑒., |𝑓𝑛 (𝑥) − 𝑓(𝑥)| ⟶ 0, as 𝑛 ⟶ ∞, ∀𝑥 ∈𝑋.

⟹ sup{|𝑓𝑛 (𝑥) − 𝑓 (𝑥)|} ⟶ 0, as 𝑛 ⟶ ∞ .

𝑥∈𝑋

⟹ ‖𝑓𝑛 − 𝑓‖ ⟶ 0, as 𝑛 ⟶ ∞ or{𝑓𝑛 } ⟶ 𝑓 .

To show 𝑓 ∈ 𝐵(𝑋 )

Since 𝑓 = lim 𝑓𝑛 ,
𝑛→∞
‖𝑓‖ = ‖ lim 𝑓𝑛 ‖ = lim ‖𝑓𝑛 ‖ .
𝑛→∞ 𝑛→∞

Since 𝑓𝑛 ′s are all bounded, 𝑓 is also bounded. 𝑖. 𝑒., 𝑓 ∈ 𝐵(𝑋 ) .

Hence, 𝐵(𝑋 ) is complete and a Banach space.

Corollary :𝐶 (𝑋 ), the collection of all continuous, bounded, scalar valued functions on 𝑋 is a Banach space.
Theorem : Let 𝑁 be a normed linear space and 𝑆 = {𝑥 ∈ 𝑁 ∶ ‖𝑥 ‖ = 1}. Then, 𝑁 is a Banach space if and only if
𝑆 is complete.

Proof : If suppose 𝑁 is a Banach space.

Then 𝑁 is complete.

Since 𝑆 is a closed subset of 𝑁, 𝑆 is complete.

Conversely, suppose 𝑆 is complete.

Let {𝑥𝑛 } be a Cauchy sequence in 𝑁 .

1
Case 1 : For each positive integer 𝑘, ∃ 𝑛𝑘 > 0 such that ‖𝑥𝑛𝑘 ‖ < .
𝑘

As 𝑘 ⟶ ∞, ‖𝑥𝑛𝑘 ‖ ⟶ 0 .

Thus we have a subsequence {𝑥𝑛𝑘 } of {𝑥𝑛 } such that {𝑥𝑛𝑘 } ⟶ 0 .

⟹ {𝑥𝑛 } ⟶ 0 ∈ 𝑁 .

⟹ 𝑁is complete.
1
Case 2 : Suppose there is a 𝑘 > 0 such that ‖𝑥𝑛 ‖ > , ∀ 𝑛.
𝑘

Define
𝑥𝑛
𝑦𝑛 = , ∀ 𝑛.
‖𝑥𝑛 ‖
Then ‖𝑦𝑛 ‖ = 1, ∀ 𝑛 ⟹ 𝑦𝑛 ∈ 𝑆, ∀ 𝑛 ⟹ {𝑦𝑛 } is a sequence in 𝑆 .

Now
𝑥𝑛 𝑥𝑚 𝑥𝑛 ‖𝑥𝑚 ‖ − 𝑥𝑚 ‖𝑥𝑛 ‖
‖𝑦𝑛 − 𝑦𝑚 ‖ = ‖ − ‖ = ‖ ‖
‖𝑥𝑛 ‖ ‖𝑥𝑚 ‖ ‖𝑥𝑛 ‖‖𝑥𝑚 ‖

𝑥𝑛 ‖𝑥𝑚 ‖ − 𝑥𝑚 ‖𝑥𝑚 ‖ + 𝑥𝑚 ‖𝑥𝑚 ‖ − 𝑥𝑚 ‖𝑥𝑛 ‖

= ‖ ‖
‖𝑥𝑛 ‖‖𝑥𝑚 ‖

‖𝑥𝑛 − 𝑥𝑚 ‖ |‖𝑥𝑚 ‖ − ‖𝑥𝑛 ‖|

≤ +
‖𝑥𝑛 ‖ ‖𝑥𝑛 ‖

𝜀
Since {𝑥𝑛 } is Cauchy in 𝑁, for every 𝜀 > 0 ( > 0), there exists an integer 𝑛0 > 0 such that
2𝑘

𝜀
‖𝑥𝑛 − 𝑥𝑚 ‖ < , ∀ 𝑛, 𝑚 ≥ 𝑛0 .
2𝑘
𝜀 𝜀
⟹ ‖𝑦𝑛 − 𝑦𝑚 ‖ < 𝑘+ 𝑘 = 𝜀.
2𝑘 2𝑘

⟹ {𝑦𝑛 }is a Cauchy sequence in 𝑆 which is complete.

𝑥𝑛
⟹ {𝑦𝑛 } ⟶ 𝑦 ∈ 𝑆 or { } ⟶ 𝑦 ∈ 𝑆.
‖𝑥𝑛 ‖

Since {𝑥𝑛 } is a Cauchy in 𝑁, {‖𝑥𝑛 ‖} is Cauchy in ℝ, which is complete.

⟹ {‖𝑥𝑛 ‖} ⟶ 𝛼 ∈ ℝ .
 𝑥
Now, {𝑥𝑛 } = {‖𝑥𝑛 ‖ ‖𝑥𝑛 ‖} ⟶ 𝛼 𝑦 ∈ 𝑁 .
𝑛

𝑖. 𝑒., 𝑁is complete and hence a Banach space.

If 𝑁 is a normed linear space and 𝑀 is a closed subspace of 𝑁, then 𝑁⁄𝑀 is a normed linear space with
respect to :
+ ∶ ( 𝑥 + 𝑀 ) + (𝑦 + 𝑀 ) = (𝑥 + 𝑦 ) + 𝑀
∙ ∶ 𝛼 (𝑥 + 𝑀) = (𝛼𝑥) + 𝑀 , ∀ 𝑥 + 𝑀, 𝑦 + 𝑀 ∈ 𝑁⁄𝑀 , 𝛼 ∈ 𝔽
and
‖𝑥 + 𝑀‖ = inf{‖𝑥 + 𝑚‖ ∶ 𝑚 ∈ 𝑀} .

𝑁⁄ is a linear space over 𝔽 ( Verify )

𝑀
Normed linear space

i. ∀ 𝑥 ∈ 𝑁, 𝑚 ∈ 𝑀 , ‖𝑥 + 𝑚‖ ≥ 0

⟹ inf{‖𝑥 + 𝑚‖ ∶ 𝑚 ∈ 𝑀} ≥ 0 𝑜𝑟 ‖𝑥 + 𝑀‖ ≥ 0 , ∀ 𝑥 + 𝑀 ∈ 𝑁⁄𝑀
Further, 𝑥 + 𝑀 = 𝑀 ( Zero of 𝑁⁄𝑀 )
⟹ ‖𝑥 + 𝑀‖ = ‖𝑀 ‖ = 0 (Since 0 ∈ 𝑀)

Conversely, if ‖𝑥 + 𝑀‖ = 0, 𝑖. 𝑒., inf{‖𝑥 + 𝑚‖ ∶ 𝑚 ∈ 𝑀} = 0

⟹ ∃ a sequence {𝑚𝑘 } in 𝑀 such that ‖𝑥 + 𝑚𝑘 ‖ ⟶ 0 , as 𝑘 ⟶ ∞ .

⟹ 𝑥 + 𝑚𝑘 ⟶ 0 ,as𝑘 ⟶ ∞ .

⟹ −𝑥is a limit point of 𝑀, which is closed.

 ⟹ −𝑥 ∈ 𝑀 ⟹ 𝑥∈𝑀 ⟹ 𝑥+𝑀 =𝑀.

ii. ∀ 𝑥 + 𝑀 ∈ 𝑁⁄𝑀 , 𝛼(≠ 0) ∈ 𝔽 ,

‖𝛼(𝑥 + 𝑀)‖ = ‖(𝛼𝑥) + 𝑀 ‖ = inf{‖𝛼𝑥 + 𝑚‖ ∶ 𝑚 ∈ 𝑀}

𝑚
= inf{‖𝛼(𝑥 + 𝑚′ )‖ ∶ 𝑚′ ∈ 𝑀} , 𝑤ℎ𝑒𝑟𝑒 𝑚′ = ∈𝑀
𝛼

= |𝛼| inf{‖𝑥 + 𝑚′ ‖ ∶ 𝑚′ ∈ 𝑀} = |𝛼|‖𝑥 + 𝑀‖ .

iii. ∀ 𝑥 + 𝑀, 𝑦 + 𝑀 ∈ 𝑁⁄𝑀 ,
‖(𝑥 + 𝑀) + (𝑦 + 𝑀)‖ = ‖(𝑥 + 𝑦) + 𝑀‖ = inf{‖(𝑥 + 𝑦) + 𝑚‖ ∶ 𝑚 ∈ 𝑀}

= inf{‖(𝑥 + 𝑦) + 𝑚′ + 𝑚′′ ‖ ∶ 𝑚′ , 𝑚′′ ∈ 𝑀}

≤ inf{‖𝑥 + 𝑚′ ‖ + ‖𝑦 + 𝑚′′ ‖ ∶ 𝑚′ , 𝑚′′ ∈ 𝑀}

= inf{‖𝑥 + 𝑚′ ‖ ∶ 𝑚′ ∈ 𝑀} + inf{‖𝑦 + 𝑚′′ ‖ ∶ 𝑚′′ ∈ 𝑀}

= ‖𝑥 + 𝑀‖ + ‖𝑦 + 𝑀‖ .

𝑁⁄ is complete when 𝑁 is complete

𝑀
Let {𝑥𝑛 + 𝑀} be a Cauchy sequence in 𝑁⁄𝑀 .
1 1
For 𝜀 = , choose 𝑥1 + 𝑀, 𝑥2 + 𝑀 in 𝑁⁄𝑀 such that ‖(𝑥1 + 𝑀) − (𝑥2 + 𝑀)‖ < .
2 2
𝑁 1
Choose 𝑥3 + 𝑀 in ⁄𝑀 such that ‖(𝑥2 + 𝑀) − (𝑥3 + 𝑀 )‖ < 2 .
2
.
.
.
1
Choose 𝑥𝑛+1 + 𝑀 in 𝑁⁄𝑀 such that ‖(𝑥𝑛 + 𝑀) − (𝑥𝑛+1 + 𝑀)‖ < 𝑛 .
2
.
.
.

Choose 𝑦𝑛 ∈ 𝑥𝑛 + 𝑀, ∀ 𝑛.
1
Then ‖𝑦𝑛 − 𝑦𝑛+1 ‖ < , ∀ 𝑛.
2𝑛

If 𝑚 < 𝑛, then ‖𝑦𝑚 − 𝑦𝑛 ‖ = ‖𝑦𝑚 − 𝑦𝑚+1 + 𝑦𝑚+1 − 𝑦𝑚+2 + 𝑦𝑚+2 − ⋯ 𝑦𝑛 ‖

≤ ‖𝑦𝑚 − 𝑦𝑚+1 ‖ + ‖𝑦𝑚+1 − 𝑦𝑚+2 ‖ + ⋯ + ‖𝑦𝑛−1 − 𝑦𝑛 ‖

1 1 1
< + + ⋯ +
2𝑚 2𝑚+1 2𝑛−1
1 1 1
= (1 + + + ⋯)
2𝑚 2 22
1
As 𝑚 ⟶ ∞, ‖𝑦𝑚 − 𝑦𝑛 ‖ = ⟶ 0.
2𝑚−1

⟹ {𝑦𝑛 }is a Cauchy sequence in 𝑁 .

If 𝑁 is complete, then {𝑦𝑛 } ⟶ 𝑦 ∈ 𝑁 .

We claim {𝑥𝑛 + 𝑀} ⟶ 𝑦 + 𝑀 .

Consider ‖(𝑥𝑛 + 𝑀) − (𝑦 + 𝑀)‖ = ‖(𝑥𝑛 − 𝑦) + 𝑀‖

 = inf{‖(𝑥𝑛 − 𝑦) + 𝑚‖ ∶ 𝑚 ∈ 𝑀}

= inf{‖(𝑥𝑛 + 𝑚) − 𝑦‖ ∶ 𝑚 ∈ 𝑀}

≤ ‖𝑦𝑛 − 𝑦‖ ⟶ 0, as 𝑛 ⟶ ∞ .

⟹ {𝑥𝑛 + 𝑀} ⟶ 𝑦 + 𝑀 ∈ 𝑁⁄𝑀 .

Hence, 𝑁⁄𝑀 is complete whenever 𝑁 is complete.

Continuous Linear Transformations

// A transformation 𝑇 ∶ 𝑁 ⟶ 𝑁 ′ is linear if 𝑇(𝛼𝑥 + 𝛽𝑦) = 𝛼𝑇(𝑥) + 𝛽𝑇(𝑦) , ∀ 𝑥, 𝑦 ∈ 𝑁 , ∀ 𝛼, 𝛽 ∈ 𝔽 .

// A transformation 𝑇 ∶ 𝑁 ⟶ 𝑁 ′ is continuous at a point 𝑥 ∈ 𝑁 if for every sequence {𝑥𝑛 } in 𝑁 converging to

𝑥, the sequence {𝑇(𝑥𝑛 )} converges to 𝑇(𝑥) in 𝑁 ′ .

// A transformation 𝑇 ∶ 𝑁 ⟶ 𝑁 ′ is bounded if ‖𝑇(𝑥)‖ ≤ 𝑀, ∀ 𝑥 ∈ 𝑁, where 𝑀 is some positive constant .

Theorem : If 𝑇 ∶ 𝑁 ⟶ 𝑁 ′ is a linear transformation, then the following are equivalent :

i. 𝑇 is continuous ;
ii. 𝑇 is continuous at the origin 𝑖. 𝑒., 𝑥𝑛 ⟶ 0 ⟹ 𝑇(𝑥𝑛 ) ⟶ 𝑇(0) = 0 ;
iii. there exists a real number 𝐾 > 0 with the property ‖𝑇(𝑥)‖ ≤ 𝐾 ‖𝑥‖ , ∀ 𝑥 ∈ 𝑁 ;
iv. if 𝑆 = {𝑥 ∶ ‖𝑥‖ ≤ 1 } is the closed unit sphere in 𝑁, then 𝑇(𝑆) is a bounded set in 𝑁 ′ .

Proof :
 (𝑖) ⟹ (𝑖𝑖)

𝑇is continuous on 𝑁 ⟹ 𝑇 is continuous at the origin (∵ 0 ∈ 𝑁) .

(𝑖𝑖) ⟹ (𝑖)

Suppose 𝑇 is continuous at the origin.

Let 𝑥 ∈ 𝑁 be arbitrary and let {𝑥𝑛 } be a sequence in 𝑁 converging to 𝑥.

Then 𝑥𝑛 ⟶ 𝑥 or𝑥𝑛 − 𝑥 ⟶ 0 .

⟹ 𝑇(𝑥𝑛 − 𝑥) ⟶ 𝑇(0) = 0 or 𝑇(𝑥𝑛 ) − 𝑇( 𝑥) ⟶ 0 or 𝑇(𝑥𝑛 ) ⟶ 𝑇( 𝑥) .

⟹ 𝑇is continuous at 𝑥 ∈ 𝑁.

Since 𝑥 ∈ 𝑁 was arbitrary, 𝑇 is continuous on 𝑁.

(𝑖𝑖) ⟹ (𝑖𝑖𝑖)

Suppose 𝑇 is continuous at the origin.

If (𝑖𝑖𝑖 ) is not true .𝑖. 𝑒., ∀ 𝑛 > 0, ∃ 𝑥𝑛 ∈ 𝑁such that ‖𝑇(𝑥𝑛 )‖ > 𝑛 ‖𝑥𝑛 ‖ .

‖𝑇(𝑥𝑛 )‖
⟹ >1.
𝑛 ‖𝑥𝑛 ‖

Set
𝑥𝑛
𝑦𝑛 = , ∀𝑛.
𝑛 ‖𝑥𝑛 ‖
Clearly,
‖𝑇(𝑥𝑛 )‖
𝑦𝑛 ⟶ 0 as 𝑛 ⟶ ∞ but‖𝑇(𝑦𝑛 )‖ = ↛ 0 as 𝑛⟶∞.
𝑛 ‖𝑥𝑛 ‖

⟹ 𝑇is not continuous at the origin.

𝑖. 𝑒., ~(𝑖𝑖𝑖 ) ⟹ ~(𝑖𝑖)or(𝑖𝑖) ⟹ (𝑖𝑖𝑖 ) .

(𝑖𝑖𝑖) ⟹ (𝑖𝑖)

If there exists a real number 𝐾 > 0 such that ‖𝑇(𝑥)‖ ≤ 𝐾 ‖𝑥 ‖ , ∀ 𝑥 ∈ 𝑁 .

If {𝑥𝑛 } is a sequence in 𝑁 such that 𝑥𝑛 ⟶ 0 as 𝑛 ⟶ ∞ .

Then ‖𝑇(𝑥𝑛 )‖ ≤ 𝐾 ‖𝑥𝑛 ‖ ⟶ 𝐾 ∙ 0 = 0 as 𝑛 ⟶ ∞ .

⟹ 𝑇(𝑥𝑛 ) ⟶ 0 as 𝑛 ⟶ ∞ or𝑇 is continuous at the origin.

(𝑖𝑖𝑖 ) ⟹ (𝑖𝑣 )

If there exists a real number 𝐾 > 0 such that ‖𝑇(𝑥)‖ ≤ 𝐾 ‖𝑥 ‖ , ∀ 𝑥 ∈ 𝑁 .

If 𝑦 ∈ 𝑇(𝑆) , then 𝑦 = 𝑇(𝑥), where 𝑥 ∈ 𝑆 = {𝑥 ∶ ‖𝑥‖ ≤ 1 } .

⟹ ‖𝑦‖ = ‖𝑇(𝑥)‖ ≤ 𝐾 ‖𝑥‖ ≤ 𝐾 , ∀ 𝑦 ∈ 𝑇 (𝑆 ) .

⟹ 𝑇(𝑆)is bounded.

(𝑖𝑣 ) ⟹ (𝑖𝑖𝑖)
 If 𝑇(𝑆) is bounded. 𝑖. 𝑒., ‖𝑇(𝑥)‖ ≤ 𝐾 , ∀ 𝑥 ∈ 𝑆.

If 𝑥 = 0, then 𝑇(𝑥) = 𝑇(0) = 0 ⟹ 0 = ‖𝑇(𝑥)‖ = 𝐾 ‖𝑥‖ = 𝐾 ∙ 0

𝑥
If 𝑥 ≠ 0, then 𝑧 = ‖𝑥‖ ∈ 𝑆 .
𝑥 ‖𝑇(𝑥)‖
‖𝑇(𝑧)‖ ≤ 𝐾 ⟹ ‖𝑇 ( )‖ = ≤ 𝐾 ⟹ ‖𝑇(𝑥)‖ ≤ 𝐾 ‖𝑥‖ .
‖𝑥‖ ‖𝑥 ‖

Theorem : The set 𝐵(𝑁, 𝑁 ′ ) of all continuous linear transformations of 𝑁 into 𝑁 ′ is a linear space with respect to

+ ∶ (𝑇 + 𝑆)(𝑥) = 𝑇(𝑥) + 𝑆(𝑥)

∙ ∶ (𝛼 𝑇)(𝑥) = 𝛼 𝑇(𝑥) , ∀ 𝑥 ∈ 𝑁 , 𝛼 ∈ 𝔽, 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ )

Further, ‖𝑇‖ = sup{‖𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } defines a norm on 𝐵(𝑁, 𝑁 ′ ) and 𝐵(𝑁, 𝑁 ′ ) is complete when
𝑁 ′ is complete.

Proof :
 We first show that 𝐵(𝑁, 𝑁 ′ ) is a linear space over 𝔽, where 𝑁 and 𝑁 ′ are normed linear spaces over 𝔽.

For 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ), ∀ 𝑥, 𝑦 ∈ 𝑁 & ∀ 𝛼, 𝛽 ∈ 𝔽,

(𝑇 + 𝑆)(𝛼𝑥 + 𝛽𝑦) = 𝑇(𝛼𝑥 + 𝛽𝑦) + 𝑆(𝛼𝑥 + 𝛽𝑦)

= 𝛼𝑇(𝑥) + 𝛽𝑇 (𝑦) + 𝛼𝑆(𝑥) + 𝛽𝑆(𝑦)

 = 𝛼 (𝑇(𝑥) + 𝑆(𝑥)) + 𝛽 (𝑇(𝑦) + 𝑆(𝑦))

= 𝛼 (𝑇 + 𝑆)(𝑥) + 𝛽 (𝑇 + 𝑆)(𝑦)

⟹ 𝑇 + 𝑆is linear.

Let {𝑥𝑛 } be a sequence in 𝑁 converging to 𝑥 in 𝑁.

𝑖. 𝑒., lim 𝑥𝑛 = 𝑥 .
𝑛→∞

Then
lim (𝑇 + 𝑆)(𝑥𝑛 ) = lim (𝑇(𝑥𝑛 ) + 𝑆(𝑥𝑛 ))
𝑛→∞ 𝑛→∞

= lim 𝑇(𝑥𝑛 ) + lim 𝑆(𝑥𝑛 )

𝑛→∞ 𝑛→∞

= 𝑇(𝑥) + 𝑆(𝑥) = (𝑇 + 𝑆)(𝑥)

⟹ 𝑇 + 𝑆is continuous.

Hence, 𝑇 + 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ) .

For any scalar 𝜆, 𝑇 ∈ 𝐵 (𝑁, 𝑁 ′ ),

(𝜆𝑇)(𝛼𝑥 + 𝛽𝑦) = 𝜆 𝑇(𝛼𝑥 + 𝛽𝑦)

= 𝜆 ( 𝛼𝑇(𝑥) + 𝛽𝑇(𝑦))

= 𝜆 𝛼 𝑇 (𝑥 ) + 𝜆 𝛽 𝑇 (𝑦 )
 = 𝛼 (𝜆𝑇)(𝑥) + 𝛽 (𝜆𝑇)(𝑦)

⟹ 𝜆𝑇is linear.

Let {𝑥𝑛 } be a sequence in 𝑁 converging to 𝑥 in 𝑁.

𝑖. 𝑒., lim 𝑥𝑛 = 𝑥 .
𝑛→∞

Then
lim (𝜆𝑇)(𝑥𝑛 ) = lim 𝜆 𝑇(𝑥𝑛 )
𝑛→∞ 𝑛→∞

= 𝜆 lim 𝑇(𝑥𝑛 )
𝑛→∞

= 𝜆 𝑇(𝑥) = (𝜆𝑇)(𝑥)

⟹ 𝜆𝑇is continuous.

Hence, 𝜆𝑇 ∈ 𝐵(𝑁, 𝑁 ′ ) .

Therefore, 𝐵(𝑁, 𝑁 ′ ) is closed under addition and scalar multiplication.

Linear space properties of 𝐵(𝑁, 𝑁 ′ )

i. ∀ 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ), 𝑇 + 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ).

ii. ∀ 𝑇, 𝑆, 𝑅 ∈ 𝐵(𝑁, 𝑁 ′ ), [(𝑇 + 𝑆) + 𝑅](𝑥) = (𝑇 + 𝑆)(𝑥) + 𝑅(𝑥)

= (𝑇(𝑥) + 𝑠(𝑥)) + 𝑅(𝑥) = 𝑇(𝑥) + (𝑆 + 𝑅)(𝑥)
 = [𝑇 + (𝑆 + 𝑅)](𝑥) , ∀ 𝑥 ∈ 𝑁 .
⟹ [(𝑇 + 𝑆) + 𝑅] = [𝑇 + (𝑆 + 𝑅)] .

iii. 0 ∶ 𝑁 ⟶ 𝑁 ′ ∶ 0(𝑥) = 0, ∀ 𝑥 ∈ 𝑁 is the additive identity of 𝐵(𝑁, 𝑁 ′ ) .

Since (𝑇 + 0)(𝑥) = 𝑇(𝑥) + 0(𝑥) = 𝑇(𝑥) + 0 = 𝑇(𝑥), ∀ 𝑥 ∈ 𝑁 .
⟹ 𝑇 + 0 = 𝑇,similarly0 + 𝑇 = 𝑇 .

iv. ∀ 𝑇 ∈ 𝐵(𝑁, 𝑁 ′ ) , − 𝑇 = (−1) 𝑇 ∈ 𝐵(𝑁, 𝑁 ′ )is its additive inverse.

Since [𝑇 + (−𝑇)](𝑥) = 𝑇(𝑥) + (−𝑇)(𝑥) = 𝑇(𝑥) − 𝑇(𝑥) = 0 = 0(𝑥), ∀ 𝑥∈𝑁.
⟹ 𝑇 + (−𝑇) = 0 .

v. ∀ 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ), (𝑇 + 𝑆)(𝑥) = 𝑇(𝑥) + 𝑆(𝑥) = 𝑆(𝑥) + 𝑇(𝑥) = (𝑆 + 𝑇)(𝑥), ∀ 𝑥 ∈ 𝑁 .

⟹ 𝑇 + 𝑆 = 𝑆 + 𝑇, ∀ 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ) .

Hence, (𝐵(𝑁, 𝑁 ′ ), +) is an abeliangroup.

Scalar multiplication in 𝐵(𝑁, 𝑁 ′ )satisfies :

∀ 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ), ∀ 𝛼, 𝛽 ∈ 𝔽,

i. [𝛼(𝑇 + 𝑆)](𝑥) = 𝛼 (𝑇 + 𝑆)(𝑥) = 𝛼[ 𝑇(𝑥) + 𝑆(𝑥)] = 𝛼 𝑇(𝑥) + 𝛼 𝑆(𝑥)

= (𝛼𝑇)(𝑥) + (𝛼𝑆)(𝑥) = (𝛼𝑇 + 𝛼𝑆)(𝑥) , ∀ 𝑥 ∈ 𝑁 .
⟹ 𝛼(𝑇 + 𝑆) = 𝛼𝑇 + 𝛼𝑆 .

ii. (𝛼 + 𝛽)𝑇 = 𝛼𝑇 + 𝛽𝑇
iii. (𝛼𝛽)𝑇 = 𝛼(𝛽𝑇)
iv. 1∙𝑇 =𝑇.

 We now show that ‖𝑇‖ = sup{‖𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } defines a norm on 𝐵(𝑁, 𝑁 ′ ) .

 i. Since ‖𝑇(𝑥)‖ ≥ 0, ∀ 𝑥 ∈ 𝑁, we havesup{‖𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } ≥ 0 .

⟹ ‖𝑇‖ ≥ 0, ∀ 𝑇 ∈ 𝐵(𝑁, 𝑁 ′ ) .
𝑥
If 𝑥 ≠ 0, let 𝑦 = ‖𝑥‖ so that ‖𝑦‖ = 1.

𝑥
Clearly, ‖𝑇‖ ≥ ‖𝑇(𝑦)‖ ⟹ ‖𝑇‖ ≥ ‖𝑇 ( )‖
‖𝑥‖
⟹ ‖𝑇(𝑥)‖ ≤ ‖𝑇‖‖𝑥‖ .

Further, ‖𝑇‖ = 0 ⟺ sup{‖𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } = 0.

⟺ ‖𝑇 (𝑥)‖ = 0, ∀ 𝑥 (≠ 0) ∈ 𝑁 ⟺ 𝑇(𝑥) = 0, ∀ 𝑥 (≠ 0) ∈ 𝑁 ⟺ 𝑇 =0.

ii. For 𝛼 ∈ 𝔽, ‖𝛼𝑇‖ = sup{‖(𝛼𝑇)(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } = sup{‖𝛼 𝑇(𝑥)‖ ∶ ‖𝑥 ‖ ≤ 1 }

= |𝛼| sup{‖𝑇 (𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } = |𝛼|‖𝑇‖ .

iii. ∀ 𝑇, 𝑆 ∈ 𝐵(𝑁, 𝑁 ′ ), ‖𝑇 + 𝑆‖ = sup{‖(𝑇 + 𝑆)(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 }

= sup{‖𝑇(𝑥) + 𝑆(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 }

≤ sup{‖𝑇(𝑥)‖ + ‖𝑆(𝑥)‖ ∶ ‖𝑥 ‖ ≤ 1 }

= sup{‖𝑇 (𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } + sup{‖𝑆(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 }

= ‖𝑇‖ + ‖𝑆‖ .

Hence, (𝐵(𝑁, 𝑁 ′ ), ‖∙‖) is a normed linear space.

 To prove 𝐵(𝑁, 𝑁 ′ ) is complete when 𝑁 ′ is complete.

Let {𝑇𝑛 } be a Cauchy sequence in 𝐵(𝑁, 𝑁 ′ ) .

Then for every 𝜀 > 0, there exists an integer 𝑛0 > 0 such that ‖𝑇𝑛 − 𝑇𝑚 ‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

∀ 𝑥 ∈ 𝑁, ‖𝑇𝑛 (𝑥) − 𝑇𝑚 (𝑥)‖ = ‖(𝑇𝑛 − 𝑇𝑚 )(𝑥)‖ ≤ ‖𝑇𝑛 − 𝑇𝑚 ‖‖𝑥‖ .

If ‖𝑥 ‖ ≤ 1, ‖𝑇𝑛 (𝑥) − 𝑇𝑚 (𝑥)‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

⟹ {𝑇𝑛 (𝑥)}is a Cauchy sequence in 𝑁 ′ .

If 𝑁 ′ is complete, {𝑇𝑛 (𝑥)} ⟶ 𝑇(𝑥) ∈ 𝑁 ′ .

𝑖. 𝑒., 𝑇 ∶ 𝑁 ⟶ 𝑁 ′ is such that 𝑇(𝑥) = lim 𝑇𝑛 (𝑥) .

𝑛→∞

𝑇is linear

∀ 𝑥, 𝑦 ∈ 𝑁, ∀ 𝛼, 𝛽 ∈ 𝔽,

𝑇(𝛼𝑥 + 𝛽𝑦) = lim 𝑇𝑛 (𝛼𝑥 + 𝛽𝑦) = lim (𝛼 𝑇𝑛 (𝑥) + 𝛽 𝑇𝑛 (𝑦))

𝑛→∞ 𝑛→∞

= 𝛼 lim 𝑇𝑛 (𝑥) + 𝛽 lim 𝑇𝑛 (𝑦) = 𝛼 𝑇(𝑥) + 𝛽 𝑇(𝑦)

𝑛→∞ 𝑛→∞

⟹ 𝑇is linear .

𝑇is continuous ( ⟺ 𝑇 is bounded )

‖𝑇(𝑥)‖ = ‖ lim 𝑇𝑛 (𝑥)‖ = lim ‖𝑇𝑛 (𝑥)‖

𝑛→∞ 𝑛→∞
 ≤ lim ‖𝑇𝑛 ‖‖𝑥‖ = ‖𝑥‖ lim ‖𝑇𝑛 ‖
𝑛→∞ 𝑛→∞

Since |‖𝑇𝑛 ‖ − ‖𝑇𝑚 ‖| ≤ ‖𝑇𝑛 − 𝑇𝑚 ‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 ,

{‖𝑇𝑛 ‖}is a Cauchy sequence in ℝ, which is complete.

⟹ {‖𝑇𝑛 ‖} ⟶ 𝐾 ∈ ℝ or lim ‖𝑇𝑛 ‖ = 𝐾 ∈ ℝ .

𝑛→∞

⟹ ‖𝑇(𝑥)‖ ≤ 𝐾 ‖𝑥‖ .

It remains to show that 𝑇 = lim 𝑇𝑛 .

𝑛→∞

Since {𝑇𝑛 } is Cauchy, we have

‖𝑇𝑛 (𝑥) − 𝑇𝑚 (𝑥)‖ ≤ ‖𝑇𝑛 − 𝑇𝑚 ‖‖𝑥 ‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 , if‖𝑥‖ ≤ 1 .

Fix 𝑛, let 𝑚 → ∞, we get ‖𝑇𝑛 (𝑥) − 𝑇(𝑥)‖ < 𝜀 , ∀ 𝑛 ≥ 𝑛0 , if‖𝑥‖ ≤ 1 .

⟹ ‖𝑇𝑛 − 𝑇‖ = sup{‖(𝑇𝑛 − 𝑇)(𝑥) = 𝑇𝑛 (𝑥) − 𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } < 𝜀 , ∀ 𝑛 ≥ 𝑛0 .

⟹ {𝑇𝑛 } ⟶ 𝑇 ∈ 𝐵(𝑁, 𝑁 ′ ) .

Hence, 𝐵(𝑁, 𝑁 ′ ) is complete when 𝑁 ′ is complete.

Note : Replacing 𝑁 ′ by 𝑁, we get 𝐵(𝑁) = 𝐵(𝑁, 𝑁) .

𝐵(𝑁)is a Banach space if 𝑁 is Banach. The elements of 𝐵(𝑁) are called continuous linear operators on 𝑁 .

Exercise :
 𝑁
1. If 𝑁 is a normed linear space and if 𝑀 is a closed subspace of 𝑁, then = { 𝑥 + 𝑀 ∶ 𝑥 ∈ 𝑁 } is a normed
𝑀
𝑁
linear space and is complete when 𝑁 is complete. The map 𝑇 ∶ 𝑁 ⟶ ∶ 𝑇(𝑥) = 𝑥 + 𝑀 is a continuous
𝑀
linear transformation and ‖𝑇‖ ≤ 1 .

𝑇is linear

∀ 𝑥, 𝑦 ∈ 𝑁, ∀ 𝛼 ∈ 𝔽,
𝑇 (𝑥 + 𝑦 ) = (𝑥 + 𝑀 ) + (𝑦 + 𝑀 ) = 𝑇 (𝑥 ) + 𝑇 (𝑦 )
𝑇(𝛼𝑥) = (𝛼𝑥) + 𝑀 = 𝛼 (𝑥 + 𝑀) = 𝛼 𝑇(𝑥)
𝑇is continuous ( at origin )

For every {𝑥𝑛 } in 𝑁 converging to 0, ‖𝑇(𝑥𝑛 ) − 𝑇(0)‖ = ‖(𝑥𝑛 + 𝑀) − (0 + 𝑀)‖

= ‖(𝑥𝑛 − 0) + 𝑀 ‖ = ‖𝑥𝑛 + 𝑀‖

= inf{‖𝑥𝑛 + 𝑚‖ ∶ 𝑚 ∈ 𝑀}

Since 𝑥𝑛 ⟶ 0, as 𝑛 ⟶ ∞, 𝑥𝑛 + 𝑚 ⟶ 𝑚, as 𝑛 ⟶ ∞ &‖𝑥𝑛 + 𝑚‖ ⟶ ‖𝑚‖, as 𝑛 ⟶ ∞ .

Therefore, ‖𝑇(𝑥𝑛 ) − 𝑇(0)‖ ⟶ inf{‖𝑚‖ ∶ 𝑚 ∈ 𝑀, 0 ∈ 𝑀} = 0 .

⟹ 𝑇(𝑥𝑛 ) ⟶ 𝑇(0) as 𝑛 ⟶ ∞ .

Therefore, 𝑇 is continuous at the origin and hence on 𝑁 .

‖𝑇‖ ≤ 1

‖𝑇‖ = sup{‖𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 }

 = sup{‖𝑥 + 𝑀 ‖ ∶ ‖𝑥‖ ≤ 1 }

≤ sup{‖𝑥‖ ∶ ‖𝑥 ‖ ≤ 1 } = 1 ∵ ‖𝑥 + 𝑀‖ = inf{‖𝑥 + 𝑚‖ ∶ 𝑚 ∈ 𝑀, 0 ∈ 𝑀}

2. If 𝑇 is a continuous linear transformation of a normed linear space 𝑁 into a normed linear space 𝑁 ′ with the
null space 𝑀, then 𝑇 induces a natural linear transformation 𝑇 ′ of 𝑁⁄𝑀 into 𝑁 ′ such that ‖𝑇 ′ ‖ = ‖𝑇‖ .
Given, 𝑇 ∶ 𝑁 ⟶ 𝑁 ′ is linear and continuous and 𝑀 = { 𝑥 ∈ 𝑁 ∶ 𝑇(𝑥) = 0 of𝑁 ′ } .

If 𝑥 is a limit point of 𝑀, then there is a sequence {𝑥𝑛 } in 𝑀 such that {𝑥𝑛 } ⟶ 𝑥 .

Since 𝑇 is continuous, {𝑥𝑛 } ⟶ 𝑥 ⟹ {𝑇(𝑥𝑛 )} ⟶ 𝑇(𝑥) .

𝑥𝑛 ∈ 𝑀 = Null space of 𝑇 ⟹ 𝑇(𝑥𝑛 ) = 0, ∀ 𝑛.

⟹ lim 𝑇(𝑥𝑛 ) = 0 or 𝑇 ( lim 𝑥𝑛 ) = 𝑇(𝑥) = 0 .

𝑛→∞ 𝑛→∞

⟹ 𝑥∈𝑀.

Therefore, 𝑀 is closed and hence 𝑁⁄𝑀 exists.

From the previous example, ∃ a continuous linear transformation
𝑆 ∶ 𝑁 ⟶ 𝑁⁄𝑀 ∶ 𝑆(𝑥) = 𝑥 + 𝑀 and‖𝑆‖ ≤ 1 .
Define 𝑇 ′ ∶ 𝑁⁄𝑀 ⟶ 𝑁 ′ by 𝑇 ′ (𝑥 + 𝑀) = 𝑇(𝑥) .
𝑖. 𝑒., 𝑇(𝑥) = 𝑇 ′ (𝑆(𝑥)) .

𝑇 ′ is linear

∀ 𝑥+𝑀, 𝑦 + 𝑀 ∈ 𝑁⁄𝑀 , ∀ 𝛼 ∈ 𝔽,
 𝑇 ′ [(𝑥 + 𝑀) + (𝑦 + 𝑀)] = 𝑇 ′ [(𝑥 + 𝑦) + 𝑀]

= 𝑇 (𝑥 + 𝑦 ) = 𝑇 (𝑥 ) + 𝑇 (𝑦 )

= 𝑇 ′ (𝑥 + 𝑀 ) + 𝑇 ′ (𝑦 + 𝑀 )

𝑇 ′ [𝛼(𝑥 + 𝑀)] = 𝑇 ′ [(𝛼𝑥) + 𝑀]

= 𝑇(𝛼𝑥) = 𝛼𝑇(𝑥) = 𝛼 𝑇 ′ (𝑥 + 𝑀)

‖𝑇 ′ ‖ = ‖𝑇‖

‖𝑇 ′ ‖ = sup{‖𝑇 ′ (𝑥 + 𝑀)‖ ∶ ‖𝑥 + 𝑀‖ ≤ 1 } = sup{‖𝑇(𝑥)‖ ∶ ‖𝑥 + 𝑀‖ ≤ 1 }

≤ sup{‖𝑇(𝑥)‖ ∶ ‖𝑥 ‖ ≤ 1 }

= ‖𝑇‖

‖𝑇‖ = sup{‖𝑇(𝑥)‖ ∶ ‖𝑥 ‖ ≤ 1 } = sup{‖𝑇 ′ (𝑆(𝑥))‖ ∶ ‖𝑥‖ ≤ 1 }

≤ sup{‖𝑇 ′ ‖‖𝑆(𝑥)‖ ∶ ‖𝑥 ‖ ≤ 1 }

≤ sup{‖𝑇 ′ ‖‖𝑆‖‖𝑥‖ ∶ ‖𝑥‖ ≤ 1 }

= ‖𝑇 ′ ‖‖𝑆‖ sup{‖𝑥‖ ∶ ‖𝑥‖ ≤ 1 }

≤ ‖𝑇 ′ ‖ ∵ ‖𝑆‖ ≤ 1
3. Let (𝐵, ‖ ∙ ‖) be a Banach space such that 𝐵 = 𝑀⨁𝑁, where 𝑀 and 𝑁 are subspaces of 𝐵. If 𝑧 ∈ 𝐵, then 𝑧 =
𝑥 + 𝑦.
We define ‖𝑧‖1 = ‖𝑥‖ + ‖𝑦‖ .Further, if 𝑀 and 𝑁 are closed in 𝐵, then 𝐵1 = (𝐵, ‖ ∙ ‖1 ) is a Banach space.

‖ ∙ ‖1 is a norm on 𝐵

For any 𝑧 ∈ 𝐵, 𝑧 = 𝑥 + 𝑦 (unique), where 𝑥 ∈ 𝑀, 𝑦 ∈ 𝑁 .

i. Since (𝐵, ‖ ∙ ‖) is a normed linear space, ‖𝑥 ‖ ≥ 0, ‖𝑦‖ ≥ 0 .

⟹ ‖𝑧‖1 = ‖𝑥‖ + ‖𝑦‖ ≥ 0 .

Further,
‖𝑧‖1 = 0 ⟺ ‖𝑥 ‖ + ‖𝑦‖ = 0 ⟺ ‖𝑥‖ = 0 &‖𝑦‖ = 0 ⟺ 𝑥 =0&𝑦 =0 ⟺𝑧 =0.

ii. ∀ 𝑧 ∈ 𝐵, 𝑧 = 𝑥 + 𝑦, ∀ 𝛼 ∈ 𝔽,

‖𝛼𝑧‖1 = ‖𝛼(𝑥 + 𝑦)‖1 = ‖𝛼𝑥 + 𝛼𝑦‖1 = ‖𝛼𝑥‖ + ‖𝛼𝑦‖ = |𝛼|‖𝑥‖ + |𝛼|‖𝑦‖

= |𝛼|(‖𝑥‖ + ‖𝑦‖) = |𝛼|‖𝑧‖1

iii. ∀ 𝑧1 , 𝑧2 ∈ 𝐵,
‖𝑧1 + 𝑧2 ‖1 = ‖(𝑥1 + 𝑦1 ) + (𝑥2 + 𝑦2 )‖1 = ‖(𝑥1 + 𝑥2 ) + (𝑦1 + 𝑦2 )‖1

= ‖𝑥1 + 𝑥2 ‖ + ‖𝑦1 + 𝑦2 ‖

≤ ‖𝑥1 ‖ + ‖𝑥2 ‖ + ‖𝑦1 ‖ + ‖𝑦2 ‖

= ‖𝑧1 ‖1 + ‖𝑧2 ‖1
Hence, 𝐵1 = (𝐵, ‖ ∙ ‖1 ) is a normed linear space .

If suppose 𝑀 and 𝑁 are closed.

Let {𝑧𝑛 } be a Cauchy sequence in 𝐵1 .Then 𝑧𝑛 = 𝑥𝑛 + 𝑦𝑛 , where 𝑥𝑛 ∈ 𝑀, 𝑦𝑛 ∈ 𝑁 .

For every 𝜀 > 0, there exists an integer 𝑛0 > 0 such that ‖𝑧𝑛 − 𝑧𝑚 ‖1 < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

𝑖. 𝑒., ‖𝑥𝑛 + 𝑦𝑛 − (𝑥𝑚 + 𝑦𝑚 )‖1 < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

⟹ ‖𝑥𝑛 − 𝑥𝑚 ‖ + ‖𝑦𝑛 − 𝑦𝑚 ‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

⟹ ‖𝑥𝑛 − 𝑥𝑚 ‖ < 𝜀 &‖𝑦𝑛 − 𝑦𝑚 ‖ < 𝜀 , ∀ 𝑚, 𝑛 ≥ 𝑛0 .

⟹ {𝑥𝑛 }is Cauchy in 𝐵 and hence in 𝑀 , {𝑦𝑛 } is Cauchy in 𝐵 and hence in 𝑁 .

Since 𝑀 and 𝑁 are closed subspaces of 𝐵 which is complete, 𝑀 and 𝑁 are complete.

⟹ {𝑥𝑛 } ⟶ 𝑥 ∈ 𝑀 &{𝑦𝑛 } ⟶ 𝑦 ∈ 𝑁 or‖𝑥𝑛 − 𝑥‖ ⟶ 0 &‖𝑦𝑛 − 𝑦‖ ⟶ 0 𝑎𝑠 𝑛 ⟶ ∞.

If 𝑧 = 𝑥 + 𝑦, ‖𝑧𝑛 − 𝑧‖1 = ‖(𝑥𝑛 + 𝑦𝑛 ) − (𝑥 + 𝑦)‖1 = ‖(𝑥𝑛 − 𝑥) + (𝑦𝑛 − 𝑦)‖1

= ‖𝑥𝑛 − 𝑥‖ + ‖𝑦𝑛 − 𝑦‖ ⟶ 0 𝑎𝑠 𝑛 ⟶ ∞.

𝑖. 𝑒., ‖𝑧𝑛 − 𝑧‖1 ⟶ 0 𝑎𝑠 𝑛⟶∞.

⟹ {𝑧𝑛 } ⟶ 𝑧 ∈ 𝐵1 .

Hence, 𝐵1 = (𝐵, ‖ ∙ ‖1 ) is complete and hence a Banach space.

Functionals
Let 𝐵(𝑁, 𝑁 ′ ) denote the collection of all continuous linear transformations from a normed linear space
𝑁 into a normed linear space 𝑁 ′ .

𝐵(𝑁, 𝑁 ′ ) is a normed linear space with respect to the norm ‖𝑇‖ = sup{‖𝑇(𝑥)‖ ∶ ‖𝑥‖ ≤ 1 } and is
complete when 𝑁 ′ is complete.

If 𝑁 ′ = 𝑁, 𝐵(𝑁, 𝑁 ′ ) = 𝐵(𝑁) is the space of all continuous linear operators on 𝑁 .

If 𝑁 ′ = 𝔽, 𝐵(𝑁, 𝔽) = 𝑁 ∗ is the Banach space of all functionals on 𝑁 (Dual space) .

// A continuous linear transformation from a normed linear space 𝑁 into a scalar field 𝔽 is called a
functional on 𝑁 .

The Hahn-Banach Theorem

Every functional 𝑓 on a linear subspace 𝑀 of a normed linear space 𝑁 can be extended to a functional 𝐹 on the
whole space 𝑁 with ‖𝐹 ‖ = ‖𝑓‖ .

Lemma : Let 𝑀 be a subspace of a normed linear space 𝑁 and let 𝑓 be a functional on 𝑀. If 𝑥0 ∈ 𝑁, with 𝑥0 ∉ 𝑀,
then 𝑀0 = 𝑀 + (𝑥0 ) is a subspace of 𝑁 and 𝑓 can be extended to a functional, say 𝑓0 on 𝑀0 such that ‖𝑓0 ‖ = ‖𝑓‖ .

Proof of lemma :
Case (i) : Let 𝑁 be a real normed linear space.

Given 𝑀 ⊆ 𝑁 and 𝑓 ∶ 𝑀 ⟶ ℝ is linear and continuous. Without loss of generality, we assume ‖𝑓‖ = 1.

Let 𝑥0 ∈ 𝑁, with 𝑥0 ∉ 𝑀 and 𝑀0 = 𝑀 + (𝑥0 ) .

𝑖. 𝑒., 𝑀0 = {𝑚 + 𝛼𝑥0 ∶ 𝑚 ∈ 𝑀, 𝛼 ∈ ℝ }

𝑀0 is a subspace of 𝑁

Let 𝑧1 , 𝑧2 ∈ 𝑀0 & 𝛼, 𝛽 ∈ ℝ ⟹ 𝑧1 = 𝑚1 + 𝛼1 𝑥0 , 𝑧2 = 𝑚2 + 𝛼2 𝑥0 , where 𝑚1 , 𝑚2 ∈ 𝑀 & 𝛼1 , 𝛼2 ∈ ℝ.

Consider 𝛼 𝑧1 + 𝛽 𝑧2 = 𝛼 (𝑚1 + 𝛼1 𝑥0 ) + 𝛽 (𝑚2 + 𝛼2 𝑥0 )

= (𝛼𝑚1 ) + (𝛼𝛼1 ) 𝑥0 + (𝛽𝑚2 ) + (𝛽𝛼2 ) 𝑥0

= (𝛼𝑚1 + 𝛽𝑚2 ) + (𝛼𝛼1 + 𝛽𝛼2 )𝑥0 ∈ 𝑀0

⟹ 𝑀0 is a subspace of 𝑁.

Define 𝑓0 ∶ 𝑀0 ⟶ ℝ ∶ 𝑓0 (𝑥 + 𝛼𝑥0 ) = 𝑓 (𝑥) + 𝛼𝑟0 , where 𝑟0 = 𝑓0 (𝑥0 ) .

𝑓0 is well defined and if 𝛼 = 0 , then 𝑓0 (𝑥) = 𝑓 (𝑥), ∀ 𝑥 ∈ 𝑀 ⟹ 𝑓0 |𝑀 = 𝑓 .

𝑓0 is continuous

For every 𝑧 ∈ 𝑀0 and a sequence {𝑧𝑛 } in 𝑀0 converging to 𝑧, we have

𝑧𝑛 = 𝑥𝑛 + 𝛼𝑛 𝑥0 ⟶ 𝑧 = 𝑥 + 𝛼𝑥0 , where 𝑥, 𝑥𝑛 ∈ 𝑀, 𝛼, 𝛼𝑛 ∈ ℝ, ∀ 𝑛.
⟹ 𝑥𝑛 ⟶ 𝑥 & 𝛼𝑛 ⟶ 𝛼 .

⟹ 𝑓(𝑥𝑛 ) ⟶ 𝑓 (𝑥) & 𝛼𝑛 ⟶ 𝛼 ∵ 𝑓 is continuous.

⟹ 𝑓0 (𝑥𝑛 + 𝛼𝑛 𝑥0 ) = 𝑓(𝑥𝑛 ) + 𝛼𝑛 𝑟0 ⟶ 𝑓(𝑥) + 𝛼 𝑟0 = 𝑓0 (𝑥 + 𝛼𝑥0 ) = 𝑓0 (𝑧).

⟹ 𝑓0 is continuous.

𝑓0 is linear

Let 𝑧1 , 𝑧2 ∈ 𝑀0 & 𝛼, 𝛽 ∈ ℝ ⟹ 𝑧1 = 𝑚1 + 𝛼1 𝑥0 , 𝑧2 = 𝑚2 + 𝛼2 𝑥0 , where 𝑚1 , 𝑚2 ∈ 𝑀 & 𝛼1 , 𝛼2 ∈ ℝ.

Consider 𝑓0 (𝛼 𝑧1 + 𝛽 𝑧2 ) = 𝑓0 [ 𝛼 (𝑚1 + 𝛼1 𝑥0 ) + 𝛽 (𝑚2 + 𝛼2 𝑥0 )]

= 𝑓0 [(𝛼𝑚1 + 𝛽𝑚2 ) + (𝛼𝛼1 + 𝛽𝛼2 )𝑥0 ]

= 𝑓(𝛼𝑚1 + 𝛽𝑚2 ) + (𝛼𝛼1 + 𝛽𝛼2 ) 𝑟0

= 𝛼𝑓(𝑚1 ) + 𝛽𝑓 (𝑚2 ) + 𝛼(𝛼1 𝑟0 ) + 𝛽 (𝛼2 𝑟0 )

= 𝛼[𝑓(𝑚1 ) + 𝛼1 𝑟0 ] + 𝛽 [𝑓(𝑚2 ) + 𝛼2 𝑟0 ]

= 𝛼𝑓0 (𝑚1 + 𝛼1 𝑥0 ) + 𝛽𝑓0 (𝑚2 + 𝛼2 𝑥0 )

= 𝛼𝑓0 (𝑧1 ) + 𝛽𝑓0 (𝑧2 )

⟹ 𝑓0 is linear.

Hence, 𝑓0 is a functional on 𝑀0 .
For 𝑥1 , 𝑥2 ∈ 𝑀, 𝑓(𝑥2 ) − 𝑓(𝑥1 ) = 𝑓 (𝑥2 − 𝑥1 ) ≤ |𝑓 (𝑥2 − 𝑥1 )|

≤ ‖𝑓‖ ‖𝑥2 − 𝑥1 ‖

= ‖𝑥2 − 𝑥1 ‖ ∵ ‖𝑓‖ = 1 .

𝑖. 𝑒., 𝑓 (𝑥2 ) − 𝑓 (𝑥1 ) ≤ ‖𝑥2 − 𝑥1 ‖ = ‖(𝑥2 + 𝑥0 ) − (𝑥1 + 𝑥0 )‖

≤ ‖𝑥2 + 𝑥0 ‖ + ‖𝑥1 + 𝑥0 ‖

⟹ −𝑓 (𝑥1 ) − ‖𝑥1 + 𝑥0 ‖ ≤ −𝑓 (𝑥2 ) + ‖𝑥2 + 𝑥0 ‖ , for every 𝑥1 , 𝑥2 ∈ 𝑀 .

Set
𝑎 = sup{−𝑓 (𝑥) − ‖𝑥 + 𝑥0 ‖} , 𝑏 = inf {−𝑓 (𝑥) + ‖𝑥 + 𝑥0 ‖}
𝑥∈𝑀 𝑥∈𝑀

We have 𝑎 ≤ 𝑏 .

Choose 𝑟0 such that 𝑎 ≤ 𝑟0 ≤ 𝑏 .

∀ 𝑥 ∈ 𝑀, − 𝑓 (𝑥) − ‖𝑥 + 𝑥0 ‖ ≤ 𝑎 ≤ 𝑟0 ≤ 𝑏 ≤ −𝑓 (𝑥) + ‖𝑥 + 𝑥0 ‖
𝑥
If 𝛼 ≠ 0, ∈𝑀.
𝛼

𝑥 𝑥 𝑥 𝑥
⟹ −𝑓 ( ) − ‖ + 𝑥0 ‖ ≤ 𝑎 ≤ 𝑟0 ≤ 𝑏 ≤ −𝑓 ( ) + ‖ + 𝑥0 ‖
𝛼 𝛼 𝛼 𝛼

1 ‖𝑥+𝛼𝑥0 ‖ 1 ‖𝑥+𝛼𝑥0 ‖
⟹ − 𝑓 (𝑥 ) − |𝛼|
≤ 𝑎 ≤ 𝑟0 ≤ 𝑏 ≤ − 𝑓 (𝑥) + |𝛼|
𝛼 𝛼

1
Adding 𝑓 (𝑥) , we get
𝛼
 ‖𝑥 + 𝛼𝑥0 ‖ 1 ‖𝑥 + 𝛼𝑥0 ‖ |𝑓 (𝑥) + 𝛼 𝑟0 | ‖𝑥 + 𝛼𝑥0 ‖
− ≤ 𝑓 (𝑥) + 𝑟0 ≤ (or) ≤
|𝛼| 𝛼 |𝛼| |𝛼| |𝛼|

⟹ |𝑓0 (𝑥 + 𝛼𝑥0 )| ≤ ‖𝑥 + 𝛼𝑥0 ‖ , ∀ 𝑥 ∈ 𝑀.

Now
‖𝑓0 ‖ = sup { |𝑓0 (𝑥 + 𝛼𝑥0 )| ∶ ‖𝑥 + 𝛼𝑥0 ‖ ≤ 1 }

≤ sup { ‖𝑥 + 𝛼𝑥0 ‖ ∶ ‖𝑥 + 𝛼𝑥0 ‖ ≤ 1 } = 1

⟹ ‖𝑓0 ‖ ≤ 1.

But 𝑓0 |𝑀 = 𝑓 ⟹ ‖𝑓0 ‖ = ‖𝑓‖ on 𝑀 (or) ‖𝑓0 ‖ = 1 on 𝑀 .

Therefore, ‖𝑓0 ‖ = 1 on 𝑀0 .

Hence, ‖𝑓0 ‖ = ‖𝑓‖ .

Case (ii) : Let 𝑁 be a complex normed linear space.

Given 𝑀 ⊆ 𝑁 and 𝑓 ∶ 𝑀 ⟶ ℂ is linear and continuous. Without loss of generality, we assume ‖𝑓‖ = 1.

For 𝑥 ∈ 𝑀, 𝑓 (𝑥) ∈ ℂ ⟹ 𝑓 (𝑥 ) = 𝑔 (𝑥 ) + 𝑖 ℎ (𝑥 ), where 𝑔, ℎ ∶ 𝑀 ⟶ ℝ.

𝑔 and ℎ are continuous

For every 𝑥 ∈ 𝑀 and 𝑥𝑛 ⟶ 𝑥 in 𝑀, 𝑓 (𝑥𝑛 ) ⟶ 𝑓(𝑥) in ℂ .

𝑖. 𝑒., 𝑔(𝑥𝑛 ) + 𝑖 ℎ(𝑥𝑛 ) = 𝑔(𝑥) + 𝑖 ℎ(𝑥) , for large 𝑛 .

Equating real and imaginary parts, 𝑔(𝑥𝑛 ) ⟶ 𝑔(𝑥) & ℎ(𝑥𝑛 ) ⟶ ℎ(𝑥).

Hence, 𝑔 and ℎ are continuous .

𝑔 and ℎ are linear

∀ 𝑥, 𝑦 ∈ 𝑀, ∀ 𝛼, 𝛽 ∈ ℝ, 𝑓(𝛼𝑥 + 𝛽𝑦) = 𝑔(𝛼𝑥 + 𝛽𝑦) + 𝑖 ℎ(𝛼𝑥 + 𝛽𝑦)

⟹ 𝛼𝑓(𝑥) + 𝛽𝑓 (𝑦) = 𝑔(𝛼𝑥 + 𝛽𝑦) + 𝑖 ℎ(𝛼𝑥 + 𝛽𝑦)

⟹ 𝛼[𝑔(𝑥) + 𝑖 ℎ(𝑥)] + 𝛽 [𝑔(𝑦) + 𝑖 ℎ(𝑦)] = 𝑔(𝛼𝑥 + 𝛽𝑦) + 𝑖 ℎ(𝛼𝑥 + 𝛽𝑦)

⟹ 𝛼𝑔(𝑥) + 𝛽𝑔(𝑦) + 𝑖 [𝛼ℎ(𝑥) + 𝛽ℎ(𝑦)] = 𝑔(𝛼𝑥 + 𝛽𝑦) + 𝑖 ℎ(𝛼𝑥 + 𝛽𝑦)

Equating the real and imaginary parts,

𝑔(𝛼𝑥 + 𝛽𝑦) = 𝛼𝑔(𝑥) + 𝛽𝑔(𝑦) & ℎ(𝛼𝑥 + 𝛽𝑦) = 𝛼ℎ(𝑥) + 𝛽ℎ(𝑦)

⟹ 𝑔 & ℎ are linear.

𝑖. 𝑒., 𝑔 & ℎ are functionals on 𝑀.

We have 𝑓 (𝑖 𝑥) = 𝑔(𝑖 𝑥) + 𝑖 ℎ(𝑖 𝑥)

But 𝑓(𝑖 𝑥) = 𝑖 𝑓(𝑥) = 𝑖(𝑔(𝑥) + 𝑖 ℎ(𝑥)) = −ℎ(𝑥) + 𝑖 𝑔(𝑥)

⟹ 𝑓 (𝑥 ) = 𝑔 (𝑥 ) − 𝑖 𝑔 (𝑖 𝑥 ) , ∀ 𝑥∈𝑀
Now
‖𝑔‖ = sup { |𝑔(𝑥)| ∶ ‖𝑥‖ ≤ 1 } = sup { |Re 𝑓 (𝑥)| ∶ ‖𝑥‖ ≤ 1 } ≤ sup { |𝑓 (𝑥)| ∶ ‖𝑥 ‖ ≤ 1 } = ‖𝑓‖ = 1

⟹ ‖𝑔‖ ≤ 1 .

𝑖. 𝑒., 𝑔 ∶ 𝑀 ⟶ ℝ is a functional such that ‖𝑔‖ ≤ 1 .

From case (i), 𝑔 has an extension say 𝑔0 to 𝑀0 (𝑖. 𝑒., 𝑔0 |𝑀 = 𝑔) such that 𝑔0 ∶ 𝑀0 ⟶ ℝ is a functional with
‖𝑔0 ‖ = ‖𝑔‖ ≤ 1 .

Define 𝑓0 ∶ 𝑀0 ⟶ ℂ ∶ 𝑓0 (𝑥) = 𝑔0 (𝑥) − 𝑖 𝑔0 (𝑖 𝑥), ∀ 𝑥 ∈ 𝑀0 .

For 𝑥 ∈ 𝑀, 𝑓0 (𝑥) = 𝑔0 (𝑥) − 𝑖 𝑔0 (𝑖 𝑥) = 𝑔(𝑥) − 𝑖 𝑔(𝑖𝑥) = 𝑓 (𝑥) .

𝑖. 𝑒., 𝑓0 |𝑀 = 𝑓 .

𝑔0 is continuous ⟹ 𝑓0 is continuous .

𝑓0 is linear

∀ 𝑥, 𝑦 ∈ 𝑀0 , 𝑓0 (𝑥 + 𝑦) = 𝑔0 (𝑥 + 𝑦) − 𝑖 𝑔0 (𝑖 (𝑥 + 𝑦))

= 𝑔0 (𝑥) + 𝑔0 (𝑦) − 𝑖 (𝑔0 (𝑖 𝑥) + 𝑔0 (𝑖 𝑦))

= 𝑔0 (𝑥) − 𝑖 𝑔0 (𝑖 𝑥) + 𝑔0 (𝑦) − 𝑖 𝑔0 (𝑖 𝑦)

= 𝑓0 (𝑥) + 𝑓0 (𝑦)

∀ 𝑥 ∈ 𝑀0 , 𝛼 ∈ ℝ, 𝑓0 (𝛼𝑥) = 𝑔0 (𝛼𝑥) − 𝑖 𝑔0 (𝑖 (𝛼𝑥))

 = 𝛼𝑔0 (𝑥) − 𝑖 𝑔0 (𝛼 (𝑖𝑥)) = 𝛼𝑔0 (𝑥) − 𝑖 𝛼𝑔0 (𝑖 𝑥) = 𝛼 𝑓0 (𝑥)

Now
𝑓0 (𝑖 𝑥) = 𝑔0 (𝑖 𝑥) − 𝑖 𝑔0 (𝑖 (𝑖 𝑥)) = 𝑔0 (𝑖 𝑥) − 𝑖 𝑔0 (− 𝑥) = 𝑔0 (𝑖 𝑥) + 𝑖 𝑔0 ( 𝑥) = 𝑖 (𝑔0 (𝑥) − 𝑖 𝑔0 (𝑖 𝑥)) = 𝑖 𝑓0 (𝑥)

∀ 𝛾 = 𝛼 + 𝑖 𝛽 ∈ ℂ, 𝑓0 (𝛾 𝑥) = 𝑓0 [(𝛼 + 𝑖 𝛽)𝑥] = 𝑓0 (𝛼𝑥 + 𝑖 𝛽𝑥) = 𝑓0 (𝛼𝑥) + 𝑓0 (𝑖 𝛽𝑥 )

= 𝛼 𝑓0 (𝑥) + 𝑖 𝛽 𝑓0 (𝑥) = (𝛼 + 𝑖 𝛽)𝑓0 (𝑥) = 𝛾𝑓0 (𝑥)

𝑖. 𝑒., 𝑓0 ∶ 𝑀0 ⟶ ℂ is linear and continuous and hence a functional.

Now, to prove ‖𝑓0 ‖ ≤ 1 on 𝑀0

If 𝑥 ∈ 𝑀0 , ‖𝑥‖ ≤ 1 and 𝑓0 (𝑥) is real, then 𝑓0 (𝑥) = 𝑔0 (𝑥) ⟹ |𝑓0 (𝑥)| = |𝑔0 (𝑥)|

⟹ sup { |𝑓0 (𝑥)| ∶ ‖𝑥‖ ≤ 1 } = sup { |𝑔0 (𝑥)| ∶ ‖𝑥‖ ≤ 1 }

𝑖. 𝑒., ‖𝑓0 ‖ = ‖𝑔0 ‖ ≤ 1 ⟹ ‖𝑓0 ‖ ≤ 1 .

If 𝑓0 (𝑥) is a complex number, let 𝑓0 (𝑥) = 𝑟 𝑒 𝑖𝜃 , 𝑟 > 0 .

⟹ |𝑓0 (𝑥)| = 𝑟 = 𝑒 −𝑖𝜃 𝑓0 (𝑥) = 𝑓0 (𝑒 −𝑖𝜃 𝑥) = 𝑔0 (𝑒 −𝑖𝜃 𝑥)

⟹ |𝑓0 (𝑥)| = 𝑟 = |𝑓0 (𝑒 −𝑖𝜃 𝑥)| = |𝑔0 (𝑒 −𝑖𝜃 𝑥)| ≤ ‖𝑔0 ‖ ‖𝑒 −𝑖𝜃 𝑥‖

≤ ‖𝑔0 ‖ |𝑒 −𝑖𝜃 | ‖𝑥‖ ≤ 1 ∙ 1 ∙ 1 = 1

𝑖. 𝑒., |𝑓0 (𝑥)| ≤ 1, ∀ 𝑥 ∈ 𝑀0 with ‖𝑥‖ ≤ 1 .

⟹ ‖𝑓0 ‖ = sup { |𝑓0 (𝑥)| ∶ ‖𝑥‖ ≤ 1 } ≤ 1 .

But ‖𝑓‖ = 1 and 𝑓0 |𝑀 = 𝑓 ⟹ ‖𝑓0 ‖ = ‖𝑓‖ = 1 .

Proof of the theorem :

Define
𝒫 = { 𝑔 ∶ 𝑑𝑜𝑚 𝑔 ⟶ ℝ or ℂ is a functional such that 𝑀 ⊆ 𝑑𝑜𝑚 𝑔 ⊆ 𝑁, 𝑔|𝑑𝑜𝑚 𝑔 = 𝑓, ‖𝑔‖ = ‖𝑓‖ }

The set 𝒫 ≠ ∅ (∵ 𝑓 ∈ 𝒫 ).

On 𝒫, we define a relation ≤ by 𝑔1 ≤ 𝑔2 in 𝒫 if and only if 𝑑𝑜𝑚 𝑔1 ⊆ 𝑑𝑜𝑚 𝑔2 .

≤ is a partial order on 𝒫

i. For any 𝑔 ∈ 𝒫, 𝑑𝑜𝑚 𝑔 ⊆ 𝑑𝑜𝑚 𝑔 ⟹ 𝑔≤𝑔.

𝑖. 𝑒., ≤ is reflexive .

ii. For 𝑔1 , 𝑔2 ∈ 𝒫, 𝑔1 ≤ 𝑔2 and 𝑔2 ≤ 𝑔1 ⟹ 𝑑𝑜𝑚 𝑔1 ⊆ 𝑑𝑜𝑚 𝑔2 and 𝑑𝑜𝑚 𝑔1 ⊆ 𝑑𝑜𝑚 𝑔2 .

⟹ 𝑑𝑜𝑚 𝑔1 = 𝑑𝑜𝑚 𝑔2 ⟹ 𝑔1 = 𝑔2 . ∵ 𝑔1 & 𝑔2 are extensions of the same function.

𝑖. 𝑒., ≤ is anti-symmetric .

iii. For 𝑔1 , 𝑔2 , 𝑔3 ∈ 𝒫, 𝑔1 ≤ 𝑔2 and 𝑔2 ≤ 𝑔3 ⟹ 𝑑𝑜𝑚 𝑔1 ⊆ 𝑑𝑜𝑚 𝑔2 and 𝑑𝑜𝑚 𝑔2 ⊆ 𝑑𝑜𝑚 𝑔3 .

⟹ 𝑑𝑜𝑚 𝑔1 ⊆ 𝑑𝑜𝑚 𝑔3 ⟹ 𝑔1 ≤ 𝑔3 .
 𝑖. 𝑒., ≤ is transitive , and hence a partial order on 𝒫.

(𝒫, ≤) is a POSET.

Let 𝐶 = {𝑔𝑖 }𝑖∈Δ be any chain in 𝒫.

𝑖. 𝑒., every two elements in 𝐶 are comparable with the relation ≤ .

Consider ⋃𝑖 𝑔𝑖 ,

𝑑𝑜𝑚 (⋃ 𝑔𝑖 ) = ⋃ 𝑑𝑜𝑚(𝑔𝑖 ) .
𝑖 𝑖
Since

𝑑𝑜𝑚(𝑔𝑗 ) ⊆ ⋃ 𝑑𝑜𝑚(𝑔𝑖 ) = 𝑑𝑜𝑚 (⋃ 𝑔𝑖 ) , for each 𝑗 ∈ Δ ,

𝑖 𝑖
we have
𝑔𝑗 ≤ ⋃ 𝑔𝑖 , for each 𝑗 ∈ Δ .
𝑖
⟹ ⋃𝑖 𝑔𝑖 is an upper bound for the chain 𝐶.

Also, ⋃𝑖 𝑔𝑖 is an extension of 𝑓 with the same norm as that of 𝑓. 𝑖. 𝑒., ⋃𝑖 𝑔𝑖 ∈ 𝒫.

Since 𝐶 was an arbitrary chain in 𝒫, every chain in 𝒫 has an upper bound and hence by Zorn’s lemma, 𝒫 has a
maximal element say 𝐹.

𝑖. 𝑒., 𝐹 is an extension of 𝑓 (𝑖. 𝑒., 𝐹|𝑀 = 𝑓) with 𝑀 ⊆ 𝑑𝑜𝑚 𝐹 ⊆ 𝑁, ‖𝐹 ‖ = ‖𝑓‖ .

We claim 𝑑𝑜𝑚 𝐹 = 𝑁 or 𝐹 is the required functional on 𝑁 with ‖𝐹 ‖ = ‖𝑓‖.

If suppose 𝑑𝑜𝑚 𝐹 ≠ 𝑁.

Choose an element 𝑥1 ∈ 𝑁 such that 𝑥1 ∉ 𝑑𝑜𝑚 𝐹.

Set 𝑀1 = 𝑑𝑜𝑚 𝐹 + (𝑥1 ) .

Then 𝑀1 is a subspace of 𝑁 and 𝑀 ⊆ 𝑑𝑜𝑚 𝐹 ⊆ 𝑀1 .

From the lemma, 𝐹 can be extended to a functional on 𝑀1 , say 𝐹1 having the same norm as that of 𝐹.

𝑖. 𝑒., 𝐹1 |𝑑𝑜𝑚 𝐹 = 𝐹 and ‖𝐹1 ‖ = ‖𝐹 ‖ .

Since 𝐹|𝑀 = 𝑓 and ‖𝐹 ‖ = ‖𝑓‖, we have 𝐹1 |𝑀 = 𝐹 and ‖𝐹1 ‖ = ‖𝐹 ‖ = ‖𝑓‖ .

𝑖. 𝑒., 𝐹1 ∈ 𝒫 and 𝐹 ≤ 𝐹1 .

Contradicts the maximality of 𝐹 in 𝒫.

Therefore, 𝑑𝑜𝑚 𝐹 = 𝑁 and 𝐹 is the required functional (extension of 𝑓) on 𝑁 with ‖𝐹 ‖ = ‖𝑓‖.

Consequences of Hahn-Banach Theorem

Theorem : Let 𝑁 be a normed linear space and 𝑥0 (≠ 0) ∈ 𝑁, then there is a 𝑓0 ∈ 𝑁 ∗ such that 𝑓0 (𝑥0 ) = ‖𝑥0 ‖ and
‖𝑓0 ‖ = 1.

Proof : Let 𝑀 = (𝑥0 ) = { 𝑘𝑥0 ∶ 𝑘 is a scalar } .

The set 𝑀 is a subspace of 𝑁.

Define 𝑓 ∶ 𝑀 ⟶ ℝ or ℂ by 𝑓 (𝑘𝑥0 ) = 𝑘 ‖𝑥0 ‖, where 𝑘 is a scalar.

In particular if 𝑘 = 1, 𝑓(𝑥0 ) = ‖𝑥0 ‖ .

𝑓 is linear and continuous

Let 𝑥, 𝑦 ∈ 𝑀, 𝛼, 𝛽 ∈ 𝔽, then 𝑥 = 𝑘1 𝑥0 , 𝑦 = 𝑘2 𝑥0 , for some scalars 𝑘1 & 𝑘2 .

Consider 𝑓 (𝛼𝑥 + 𝛽𝑦) = 𝑓[𝛼 (𝑘1 𝑥0 ) + 𝛽 (𝑘2 𝑥0 )] = 𝑓 [(𝛼𝑘1 )𝑥0 + (𝛽𝑘2 )𝑥0 ]

= 𝑓 [(𝛼𝑘1 + 𝛽𝑘2 )𝑥0 ] = (𝛼𝑘1 + 𝛽𝑘2 )‖𝑥0 ‖

= 𝛼(𝑘1 ‖𝑥0 ‖) + 𝛽 (𝑘2 ‖𝑥0 ‖) = 𝛼𝑓 (𝑘1 𝑥0 ) + 𝛽𝑓 (𝑘2 𝑥0 )

= 𝛼𝑓(𝑥) + 𝛽𝑓 (𝑦)

Let 𝑦𝑛 ⟶ 𝑦 in 𝑀 as 𝑛 ⟶ ∞.

𝑦𝑛 = 𝑘𝑛 𝑥0 , 𝑦 = 𝑘𝑥0 where 𝑘𝑛 , 𝑘 ∈ 𝔽, ∀ 𝑛.

Then 𝑓 (𝑦𝑛 ) = 𝑓(𝑘𝑛 𝑥0 ) = 𝑘𝑛 ‖𝑥0 ‖ ⟶ 𝑘 ‖𝑥0 ‖ = 𝑓 (𝑘𝑥0 ) = 𝑓 (𝑦), as 𝑛 ⟶ ∞.

⟹ 𝑓 is continuous and hence is a functional on 𝑀.

Now, ‖𝑓‖ = sup { |𝑓(𝑥)| ∶ ‖𝑥 ‖ ≤ 1 } = sup { |𝑓 (𝑘𝑥0 )| ∶ ‖𝑘𝑥0 ‖ ≤ 1 }

= sup { |𝑘 |‖𝑥0 ‖ ∶ |𝑘 |‖𝑥0 ‖ ≤ 1 } = 1.

By Hahn-Banach theorem, there is a functional 𝑓0 on 𝑁 such that 𝑓0 |𝑀 = 𝑓 and ‖𝑓0 ‖ = ‖𝑓‖ .

Now, 𝑥0 ∈ 𝑀 ⟹ 𝑓0 (𝑥0 ) = 𝑓 (𝑥0 ) = ‖𝑥0 ‖ and ‖𝑓0 ‖ = ‖𝑓‖ = 1.

Corollary : 𝑁 ∗ separates vectors in 𝑁.

Proof : Let 𝑥, 𝑦 ∈ 𝑁 with 𝑥 ≠ 𝑦. Then 𝑥 − 𝑦 ≠ 0.

From the above result, there is a functional 𝑓0 ∈ 𝑁 ∗ such that 𝑓0 (𝑥 − 𝑦) = ‖𝑥 − 𝑦‖ ≠ 0.

⟹ 𝑓0 (𝑥) − 𝑓0 (𝑦) ≠ 0 or 𝑓0 (𝑥) ≠ 𝑓0 (𝑦).

Therefore, 𝑁 ∗ separates vectors in 𝑁.

Theorem : If 𝑀 is a closed linear subspace of a normed linear space 𝑁 and 𝑥0 ∉ 𝑀, then there exists 𝑓0 ∈ 𝑁 ∗ such
that 𝑓0 (𝑀) = {0} and 𝑓0 (𝑥0 ) ≠ 0.
𝑁
Proof : Consider 𝑇 ∶ 𝑁 ⟶ ∶ 𝑇(𝑥) = 𝑥 + 𝑀 , ∀ 𝑥 ∈ 𝑁.
𝑀

This 𝑇 is a continuous linear transformation on 𝑁 with ‖𝑇‖ ≤ 1.

∀ 𝑥 ∈ 𝑀, 𝑇(𝑥) = 𝑥 + 𝑀 = 𝑀, zero element of 𝑁⁄𝑀 or 𝑇(𝑀) = {𝑀} = {zero of 𝑁⁄𝑀} .
If 𝑥0 ∉ 𝑀, 𝑇(𝑥0 ) = 𝑥0 + 𝑀 ≠ 𝑀, zero element of 𝑁⁄𝑀 .
𝑖. 𝑒., 𝑇(𝑥0 ) is a non-zero element of 𝑁⁄𝑀 .
𝑁 ∗
From the previous theorem, there exists a functional 𝑓 ∈ ( ) , such that 𝑓(𝑥0 + 𝑀) = ‖𝑥0 + 𝑀‖ and
𝑀
‖𝑓‖ = 1.

𝑁 𝑁
Combining 𝑇 ∶ 𝑁 ⟶ & 𝑓∶ ⟶ ℝ or ℂ, we get 𝑓 ∘ 𝑇 ∶ 𝑁 ⟶ ℝ or ℂ.
𝑀 𝑀
 Since 𝑓 and 𝑇 are linear and continuous, 𝑓 ∘ 𝑇 is linear and continuous.

⟹ 𝑓0 = 𝑓 ∘ 𝑇 ∈ 𝑁 ∗ and 𝑓0 (𝑀) = (𝑓 ∘ 𝑇)(𝑀) = 𝑓(𝑇(𝑀)) = 𝑓 ({𝑀}) = {𝑓(𝑀)}

= {𝑓 (0 + 𝑀)} = {‖0 + 𝑀‖} = {‖𝑀‖} = {0} .

𝑓0 (𝑥0 ) = (𝑓 ∘ 𝑇)(𝑥0 ) = 𝑓(𝑇(𝑥0 )) = 𝑓 (𝑥0 + 𝑀) = ‖𝑥0 + 𝑀‖ ≠ 0.

Example : Let 𝑀 be a closed subspace of a normed linear space 𝑁 and let 𝑥0 ∉ 𝑀. If 𝑑 = 𝑑 (𝑥0 , 𝑀), then there exists
1
a functional ℎ ∈ 𝑁 ∗ such that ℎ(𝑀) = 0, ℎ(𝑥0 ) = 1, ‖ℎ‖ = .
𝑑

Given 𝑀 is a closed subspace of 𝑁 and 𝑥0 ∉ 𝑀.

From the above result, there exists a functional 𝑓0 ∈ 𝑁 ∗ such that 𝑓0 (𝑀) = {0}, 𝑓0 (𝑥0 ) = ‖𝑥0 + 𝑀‖ ≠ 0,
𝑁 𝑁
where 𝑓0 = 𝑓 ∘ 𝑇, 𝑓 ∶ ⟶ ℝ or ℂ with ‖𝑓‖ = 1, 𝑇 ∶ 𝑁 ⟶ ∶ 𝑇(𝑥) = 𝑥 + 𝑀 .
𝑀 𝑀

|𝑓0 (𝑥)| = ‖(𝑓 ∘ 𝑇)(𝑥)‖ = ‖𝑓(𝑇(𝑥))‖ ≤ ‖𝑓‖‖𝑇(𝑥)‖ = ‖𝑓‖‖𝑥 + 𝑀 ‖

= ‖𝑥 + 𝑀‖ ≤ ‖𝑥‖

⟹ ‖𝑓0 ‖ = sup { |𝑓0 (𝑥)| ∶ ‖𝑥‖ ≤ 1 } ≤ sup { ‖𝑥‖ ∶ ‖𝑥‖ ≤ 1 } = 1 .

Also, ‖𝑓0 ‖ = sup { |𝑓0 (𝑥)| ∶ ‖𝑥 ‖ ≤ 1 } ≥ sup { |𝑓(𝑥 + 𝑀)| ∶ ‖𝑥 + 𝑀‖ ≤ 1 } = 1 .

Combining, ‖𝑓0 ‖ = 1 .

Let 𝑑 = 𝑑 (𝑥0 , 𝑀) = inf { ‖𝑥0 − 𝑥‖ ∶ 𝑥 ∈ 𝑀 } = inf { ‖𝑥0 + 𝑚‖ ∶ 𝑚 ∈ 𝑀 } = ‖𝑥0 + 𝑀 ‖ ≠ 0.

 𝑓0
Define ℎ = .
𝑑

𝑓0 (𝑀) 𝑓0 (𝑥0 ) ‖𝑥 +𝑀‖

Then ℎ is a linear functional on 𝑁 such that ℎ(𝑀) = = 0 & ℎ(𝑥0 ) = = ‖𝑥0 =1.
𝑑 𝑑 0 +𝑀‖

‖𝑓0 ‖ 1
Also, ‖ℎ‖ = = .
𝑑 𝑑

// A linear transformation 𝑇 from a normed linear space 𝑁 into a normed linear space 𝑁 ′ is said to be an isometric
isomorphism if 𝑇 is one-one and an isometry.

// 𝑁 is said to be imbedded into 𝑁 ′ if there is an isometric isomorphism of 𝑁 into 𝑁 ′ .

// If the isometric isomorphism is onto, onto the second dual 𝑁 ∗∗ of 𝑁, then we say that 𝑁 is a reflexive normed
linear space.

// Let 𝑁 be a normed linear space over 𝔽(ℝ or ℂ). Then 𝑁 ∗ = { 𝑓 ∶ 𝑁 ⟶ ℝ or ℂ ∶ 𝑓 is linear & 𝑐𝑜𝑛𝑡𝑖𝑛𝑢𝑜𝑢𝑠 } is
a normed linear space over 𝔽 called the dual space of 𝑁 or the first dual of 𝑁.
The first dual of 𝑁 ∗ is called the second dual of 𝑁 and is denoted by 𝑁 ∗∗ . This space 𝑁 ∗∗ is non-empty.

Theorem : For each 𝑥 ∈ 𝑁, we can get an element 𝐹𝑥 ∈ 𝑁 ∗∗ in a natural way.

Proof : 𝑁 is a normed linear space with 𝑁 ∗ & 𝑁 ∗∗ as its first and second duals respectively.

For each 𝑥 ∈ 𝑁 & 𝑓 ∈ 𝑁 ∗ , 𝑓 (𝑥) is a scalar.

Define 𝐹𝑥 ∶ 𝑁 ∗ ⟶ ℝ or ℂ by 𝐹𝑥 (𝑓) = 𝑓 (𝑥), ∀ 𝑓 ∈ 𝑁 ∗ .

𝐹𝑥 is linear

∀ 𝑓, 𝑔 ∈ 𝑁 ∗ and 𝛼, 𝛽 ∈ 𝔽, 𝐹𝑥 (𝛼𝑓 + 𝛽𝑔) = (𝛼𝑓 + 𝛽𝑔)(𝑥) = 𝛼𝑓 (𝑥) + 𝛽𝑔(𝑥) = 𝛼𝐹𝑥 (𝑓) + 𝛽𝐹𝑥 (𝑔) .

𝐹𝑥 is continuous (𝐹𝑥 (𝑆) is bounded )

Let 𝑆 be an unit sphere in 𝑁 ∗ . 𝑖. 𝑒., 𝑆 = { 𝑓 ∈ 𝑁 ∗ ∶ ‖𝑓‖ ≤ 1 }

Now 𝐹𝑥 (𝑆) = { 𝐹𝑥 (𝑓) ∶ 𝑓 ∈ 𝑆 } = { 𝑓 (𝑥) ∶ ‖𝑓‖ ≤ 1 }.

|𝐹𝑥 (𝑆)| = |𝑓 (𝑥)| ≤ ‖𝑓‖‖𝑥‖ ≤ 1 ∙ ‖𝑥 ‖ = ‖𝑥‖ , ∀ 𝑓 ∈ 𝑆

⟹ 𝐹𝑥 (𝑆) is bounded or 𝐹𝑥 is continuous.

Hence, 𝐹𝑥 ∈ 𝑁 ∗∗ .

Theorem : There is a natural imbedding of 𝑁 into 𝑁 ∗∗ obtained by the isometric isomorphism

Φ ∶ 𝑁 ⟶ 𝑁 ∗∗ ∶ Φ(𝑥) = 𝐹𝑥 .

Proof : Given 𝑁 is a normed linear space over 𝔽(ℝ or ℂ).

From the previous result, for each 𝑥 ∈ 𝑁, we can get an element 𝐹𝑥 ∈ 𝑁 ∗∗ ∶ 𝐹𝑥 (𝑓) = 𝑓 (𝑥), ∀ 𝑓 ∈ 𝑁 ∗ .
Define Φ ∶ 𝑁 ⟶ 𝑁 ∗∗ ∶ Φ(𝑥) = 𝐹𝑥 , ∀ 𝑥 ∈ 𝑁.

Φ is linear

∀ 𝑥, 𝑦 ∈ 𝑁 and 𝛼, 𝛽 ∈ 𝔽, Φ(𝛼𝑥 + 𝛽𝑦) = 𝐹𝛼𝑥+𝛽𝑦

𝐹𝛼𝑥+𝛽𝑦 (𝑓) = 𝑓 (𝛼𝑥 + 𝛽𝑦) = 𝛼𝑓 (𝑥) + 𝛽𝑓 (𝑦) = 𝛼𝐹𝑥 (𝑓) + 𝛽𝐹𝑦 (𝑓) = (𝛼𝐹𝑥 + 𝛽𝐹𝑦 )(𝑓), ∀ 𝑓 ∈ 𝑁 ∗ .

⟹ 𝐹𝛼𝑥+𝛽𝑦 = 𝛼𝐹𝑥 + 𝛽𝐹𝑦 = 𝛼Φ(𝑥) + 𝛽Φ(𝑦).

Φ is one-one

For some 𝑥, 𝑦 ∈ 𝑁, if 𝑥 ≠ 𝑦.

Since 𝑁 ∗ separates the vectors in 𝑁, there exists 𝑓 ∈ 𝑁 ∗ such that 𝑓 (𝑥) ≠ 𝑓(𝑦).

⟹ 𝐹𝑥 (𝑓) ≠ 𝐹𝑦 (𝑓) , for some 𝑓 ∈ 𝑁 ∗ .

⟹ 𝐹𝑥 ≠ 𝐹𝑦 or Φ(𝑥) ≠ Φ(𝑦) .

Therefore, Φ is an isomorphism.

Φ is norm-preserving

For 𝑥 ∈ 𝑁,
‖Φ(𝑥)‖ = ‖Fx ‖ = sup { |𝐹𝑥 (𝑓)| ∶ ‖𝑓‖ ≤ 1 } = sup { |𝑓 (𝑥)| ∶ ‖𝑓‖ ≤ 1 }

≤ sup { ‖𝑓‖‖𝑥‖ ∶ ‖𝑓‖ ≤ 1 } = ‖𝑥 ‖ sup { ‖𝑓‖ ∶ ‖𝑓‖ ≤ 1 } = ‖𝑥‖ .

⟹ ‖Φ(𝑥)‖ ≤ ‖𝑥‖ − − (1)

If 𝑥 = 0, 𝐹𝑥 = 𝐹0 .

⟹ 𝐹𝑥 (𝑓) = 𝐹0 (𝑓) = 𝑓 (0) = 0, ∀ 𝑓 ∈ 𝑁 ∗ .

⟹ Φ (𝑥 ) = Φ (0) = 0 ⟹ ‖Φ(𝑥)‖ = ‖𝑥‖

If 𝑥 ≠ 0, then by a consequence of Hahn-Banach theorem, there exists 𝑓0 ∈ 𝑁 ∗ such that 𝑓0 (𝑥) = ‖𝑥‖ and
‖𝑓0 ‖ = 1 .

Now, ‖Φ(𝑥)‖ = ‖Fx ‖ = sup { |𝐹𝑥 (𝑓)| ∶ ‖𝑓‖ ≤ 1 } ≥ |𝐹𝑥 (𝑓0 )| = |𝑓0 (𝑥)| = ‖𝑥‖ .

Hence, ‖Φ(𝑥)‖ ≥ ‖𝑥‖ − − (2)

Combining (1) & (2), ‖Φ(𝑥)‖ = ‖𝑥‖ . 𝑖. 𝑒., Φ is norm-preserving.

For 𝑥, 𝑦 ∈ 𝑁, 𝑑(Φ(𝑥), Φ(𝑦)) = ‖Φ(𝑥) − Φ(𝑦)‖ = ‖Φ(𝑥 − 𝑦)‖ = ‖x − y‖ = 𝑑 (𝑥, 𝑦)

⟹ Φ is an isometry.

Therefore, Φ ∶ 𝑁 ⟶ 𝑁 ∗∗ is an isometric isomorphism. 𝑖. 𝑒., 𝑁 is imbedded into 𝑁 ∗∗ .

The Open Mapping Theorem

If 𝑇 is a continuous linear transformation from a Banach space 𝐵 onto a Banach space 𝐵′ , then 𝑇 is an open
mapping.

Lemma : If 𝐵 and 𝐵′ are Banach spaces, then under a continuous linear transformation the image of each open
sphere centered at the origin in 𝐵 contains an open sphere centered at the origin in 𝐵′ .

Proof of lemma :

Consider 𝑆𝑟 = { 𝑥 ∈ 𝐵 ∶ ‖x‖ < 𝑟 }, an open sphere centered at the origin of radius 𝑟 in 𝐵.

Similarly, 𝑆𝑟′ = { 𝑦 ∈ 𝐵′ ∶ ‖y‖ < 𝑟 }, an open sphere centered at the origin of radius 𝑟 in 𝐵′ .

𝑇(𝑆𝑟 ) = { 𝑦 ∈ 𝐵′ ∶ 𝑦 = 𝑇(𝑥) , 𝑥 ∈ 𝑆𝑟 }

 𝑇(𝑆𝑟 ) = 𝑇(𝑟𝑆1 ) = 𝑟 𝑇(𝑆1 )

x
Let 𝑦 ∈ 𝑇(𝑆𝑟 ) ⟺ 𝑦 = 𝑇(𝑥), 𝑥 ∈ 𝑆𝑟 ⟺ 𝑦 = 𝑇(𝑥), ‖x‖ < 𝑟 ⟺ 𝑦 = 𝑇 (𝑥 ) , ‖ ‖ < 1
r

𝑥
⟺ 𝑦 = 𝑇(𝑟𝑡 ), ‖t ‖ < 1 where 𝑡= ⟺ 𝑦 = 𝑇(𝑟𝑡 ), 𝑡 ∈ 𝑆1 ⟺ 𝑦 ∈ 𝑇(𝑟𝑆1 )
𝑟
𝑥
⟺ 𝑦 = 𝑟𝑇(𝑡 ), ‖t ‖ < 1 where 𝑡= ⟺ 𝑦 = 𝑟𝑇(𝑡 ), 𝑡 ∈ 𝑆1 ⟺ 𝑦 ∈ 𝑟𝑇(𝑆1 )
𝑟

 𝑥 + 𝑆𝑟 = { 𝑥 + 𝑡 ∶ 𝑡 ∈ 𝑆𝑟 } = { 𝑥 + 𝑡 ∶ ‖t‖ < 𝑟 } = { 𝑦 ∶ ‖y − x‖ < 𝑟 } = 𝑆𝑟 (𝑥), an open sphere centered at

𝑥 with radius 𝑟 .

Similarly, 𝑦 + 𝑆𝑟′ is an open sphere centered at 𝑦 with radius 𝑟 in 𝐵′ .

  𝑘 𝑆𝑟 = 𝑆𝑘𝑟 , 𝑘 > 0

𝑘 𝑆𝑟 = { 𝑘 𝑥 ∶ 𝑥 ∈ 𝑆𝑟 } = { 𝑘 𝑥 ∶ ‖x‖ < 𝑟 } = { 𝑘 𝑥 ∶ 𝑘 ‖x‖ < 𝑘𝑟 } = { 𝑡 ∶ ‖t‖ < 𝑘𝑟 } = 𝑆𝑘𝑟

We now show that it is sufficient to prove the lemma for 𝑟 = 1

Suppose the lemma holds for 𝑟 = 1. 𝑖. 𝑒., ∃ 𝜀 > 0 such that 𝑆𝜀′ ⊆ 𝑇(𝑆1 ) .

⟹ 𝑟 𝑆𝜀′ ⊆ 𝑟 𝑇(𝑆1 ) ⟹ ′
𝑆𝑟𝜀 ⊆ 𝑇(𝑟𝑆1 ) ⊆ 𝑇(𝑆𝑟 ) .

Proof of lemma for 𝑟 = 1

We have
𝐵 = ⋃ 𝑆𝑛 .
𝑛∈ℕ

⟹ 𝑇(𝐵) = 𝑇 (⋃ 𝑆𝑛 ) = ⋃ 𝑇(𝑆𝑛 )
𝑛∈ℕ 𝑛=1

∞
′
⟹ 𝐵 = ⋃ 𝑇(𝑆𝑛 )
𝑛=1

Since 𝐵′ is complete, by Baire’s category theorem, 𝐵′ is of second category.

Then ∃ an integer 𝑛0 such that 𝑇(𝑆𝑛0 ) is not nowhere dense.

 𝑖. 𝑒., 𝑖𝑛𝑡 ( ̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 )) ≠ ∅ .

 𝑇(𝑆𝑛0 ) ≠ ∅

Let 𝑦0 ∈ 𝑖𝑛𝑡 ( ̅̅̅̅̅̅̅̅̅

𝑇(𝑆𝑛0 )) ⟹ 𝑦0 ∈ 𝑈 ⊆ ̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) , where 𝑈 is an open set.

⟹ 𝑦0 ∈ 𝑈 ⊆ 𝑇(𝑆𝑛0 ) ∪ 𝑏𝑑𝑟𝑦 (𝑇(𝑆𝑛0 ))

If 𝑦0 ∈ 𝑇(𝑆𝑛0 ), then 𝑇(𝑆𝑛0 ) ≠ ∅ .

If 𝑦0 ∈ 𝑏𝑑𝑟𝑦 (𝑇(𝑆𝑛0 )) , then 𝑈 ∩ 𝑇(𝑆𝑛0 ) ≠ ∅ ⟹ 𝑇(𝑆𝑛0 ) ≠ ∅ .

Hence, 𝑇(𝑆𝑛0 ) is non-empty. Let 𝑦0 ∈ 𝑇(𝑆𝑛0 ) .

⟹ 𝑦0 = 𝑇(𝑥0 ) , where 𝑥0 ∈ 𝑆𝑛0 or ‖x0 ‖ < 𝑛0 .

 ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅)
𝑇(𝑆𝑛0 ) − 𝑦0 ⊆ 𝑇(𝑆 2𝑛0

𝑆𝑛0 − 𝑥0 = { 𝑥 − 𝑥0 ∶ 𝑥 ∈ 𝑆𝑛0 }

⟹ ‖𝑥 − x0 ‖ = ‖𝑥 + (−x0 )‖ ≤ ‖𝑥‖ + ‖x0 ‖ < n0 + n0 = 2n0

⟹ 𝑥 − 𝑥0 ∈ 𝑆2𝑛0 ⟹ 𝑆𝑛0 − 𝑥0 ⊆ 𝑆2𝑛0 ⟹ 𝑇(𝑆𝑛0 − 𝑥0 ) ⊆ 𝑇(𝑆2𝑛0 )

⟹ 𝑇(𝑆𝑛0 ) − 𝑇(𝑥0 ) = 𝑇(𝑆𝑛0 ) − 𝑦0 ⊆ 𝑇(𝑆2𝑛0 )

 ⟹ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅)
𝑇(𝑆𝑛0 ) − 𝑦0 ⊆ 𝑇(𝑆 2𝑛0

 ̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) − 𝑦0 = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) − 𝑦0

Let 𝑧 ∈ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) − 𝑦0

⟺ ∃ a sequence {𝑇(𝑡𝑛 ) − 𝑦0 } in 𝑇(𝑆𝑛0 ) − 𝑦0 such that 𝑇(𝑡𝑛 ) − 𝑦0 ⟶ 𝑧 as 𝑛 ⟶ ∞.

⟺ 𝑇(𝑡𝑛 ) ⟶ 𝑧 + 𝑦0 as 𝑛 ⟶ ∞

⟺ 𝑧 + 𝑦0 ∈ ̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) ⟺ 𝑧 ∈ ̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) − 𝑦0

 𝑦 ⟶ 𝑦 − 𝑦0 is a translation by −𝑦0 from 𝐵′ to 𝐵′ and a translation is a homeomorphism.

Since 𝑦0 is an interior point of ̅̅̅̅̅̅̅̅̅

𝑇(𝑆𝑛0 ) , 𝑦0 − 𝑦0 is an interior point of ̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) − 𝑦0 .

⟹ 0′ is an interior point of ̅̅̅̅̅̅̅̅̅

𝑇(𝑆𝑛0 ) − 𝑦0 = ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑇(𝑆𝑛0 ) − 𝑦0

⟹ ̅̅̅̅̅̅̅̅̅̅) .
0′ is an interior point of 𝑇(𝑆 2𝑛0

⟹ ̅̅̅̅̅̅̅̅̅̅) , for some 𝑎 > 0 .

0′ ∈ 𝑆𝑎′ ⊆ 𝑇(𝑆 2𝑛0

⟹ ̅̅̅̅̅̅̅̅̅̅) = ̅̅̅̅̅̅̅̅̅̅̅̅
0′ ∈ 𝑆𝑎′ ⊆ 𝑇(𝑆 𝑇(2𝑛0 𝑆1 ) = 2𝑛0 ̅̅̅̅̅̅̅
𝑇(𝑆1 )
2𝑛0

1
⟹ 0′ ∈ 𝑆′ = 𝑆 ′ 𝑎 ⊆ ̅̅̅̅̅̅̅
𝑇(𝑆1 )
2𝑛0 𝑎 2𝑛0
 𝑎
⟹ 0′ ∈ 𝑆𝑏′ ⊆ ̅̅̅̅̅̅̅
𝑇(𝑆1 ) , where 𝑏 = >0.
2𝑛0

 Let 𝑦 ∈ 𝑆𝑏′ , then 𝑦 ∈ ̅̅̅̅̅̅̅

𝑇(𝑆1 )
𝑏
⟹ ∃ 𝑦1 ∈ 𝑇(𝑆1 ) such that ‖𝑦 − 𝑦1 ‖ <
2

Then 𝑦1 = 𝑇(𝑥1 ) where 𝑥1 ∈ 𝑆1 or ‖𝑥1 ‖ < 1 .

1 1 ̅̅̅̅̅̅̅̅
⟹ 𝑦 − 𝑦1 ∈ 𝑆𝑏′ = 𝑆𝑏′ ⊆ ̅̅̅̅̅̅̅
𝑇(𝑆1 ) = 𝑇 (𝑆1 )
2 2 2
2

𝑏
⟹ ∃ 𝑦2 ∈ 𝑇 (𝑆1 ) such that ‖𝑦 − 𝑦1 − 𝑦2 ‖ <
2 22

1
Then 𝑦2 = 𝑇(𝑥2 ) where 𝑥2 ∈ 𝑆1 or ‖𝑥2 ‖ < .
2 2

Continuing the process like this, at the nth stage we have

𝑏
𝑦𝑛 ∈ 𝑇 (𝑆 1 ) such that ‖ 𝑦 − 𝑦1 − 𝑦2 − … 𝑦𝑛 ‖ <
2𝑛−1 2𝑛

1
where 𝑦𝑛 = 𝑇(𝑥𝑛 ) , 𝑥𝑛 ∈ 𝑆 1 or ‖𝑥𝑛 ‖ < .
2𝑛−1 2𝑛−1

Define 𝑋𝑛 = 𝑥1 + 𝑥2 + … + 𝑥𝑛 , 𝑌𝑛 = 𝑦1 + 𝑦2 + … + 𝑦𝑛 .
𝑏
Then ‖𝑦 − 𝑌𝑛 ‖ < ⟶ 0 as 𝑛 ⟶ ∞ .
2𝑛
⟹ 𝑌𝑛 ⟶ 𝑦 as 𝑛 ⟶ ∞ or lim 𝑌𝑛 = 𝑦 .
𝑛→∞

{𝑋𝑛 } is Cauchy

If 𝑛 > 𝑚, ‖𝑋𝑛 − 𝑋𝑚 ‖ = ‖𝑥𝑚+1 + 𝑥𝑚+2 + … 𝑥𝑛 ‖ ≤ ‖𝑥𝑚+1 ‖ + ‖𝑥𝑚+2 ‖ + … ‖𝑥𝑛 ‖

1 1 1
≤ + + … +
2𝑚 2𝑚+1 2𝑛−1
1 1 1
= (1 + + + …)
2𝑚 2 22

1
= ⟶ 0 as 𝑚 ⟶ ∞
2𝑚−1

For large 𝑛 & 𝑚, ‖𝑋𝑛 − 𝑋𝑚 ‖ ⟶ 0 .

𝑖. 𝑒., {𝑋𝑛 } is Cauchy in 𝐵 which is complete.

⟹ {𝑋𝑛 } ⟶ 𝑥 ∈ 𝐵 or 𝑥 = lim 𝑋𝑛
𝑛→∞

⟹ ‖𝑥‖ = ‖ lim 𝑋𝑛 ‖ = lim ‖𝑋𝑛 ‖ = lim ‖𝑥1 + 𝑥2 + … + 𝑥𝑛 ‖

𝑛→∞ 𝑛→∞ 𝑛→∞

≤ lim ( ‖𝑥1 ‖ + ‖𝑥2 ‖ + ⋯ ‖𝑥𝑛 ‖ )

𝑛→∞
1 1 1
< lim ( 1 + + 2 + … + 𝑛 ) = 2
𝑛→∞ 2 2 2

⟹ ‖𝑥‖ < 2 or 𝑥 ∈ 𝑆2 .
 Also
𝑇(𝑥) = 𝑇 ( lim 𝑋𝑛 ) = lim 𝑇(𝑋𝑛 ) = lim 𝑇(𝑥1 + 𝑥2 + … + 𝑥𝑛 )
𝑛→∞ 𝑛→∞ 𝑛→∞

= lim (𝑇(𝑥1 ) + 𝑇(𝑥2 ) + … + 𝑇(𝑥𝑛 ))

𝑛→∞

= lim (𝑦1 + 𝑦2 + … + 𝑦𝑛 ) = lim 𝑌𝑛 = 𝑦

𝑛→∞ 𝑛→∞

Therefore 𝑥 ∈ 𝑆2 ⟹ 𝑦 = 𝑇(𝑥) ∈ 𝑇(𝑆2 ) .

Thus we have proved 𝑦 ∈ 𝑆𝑏′ ⟹ 𝑦 ∈ 𝑇(𝑆2 ) .

1 1
⟹ 𝑆𝑏′ ⊆ 𝑇(𝑆2 ) ⟹ 𝑆′ ⊆ 𝑇(𝑆2 )
2 𝑏 2

𝑏
𝑖. 𝑒., 𝑆𝜀′ ⊆ 𝑇(𝑆1 ) , where 𝜀 = .
2

𝑖. 𝑒., the lemma is true for 𝑟 = 1 and hence is true for all values of 𝑟 > 0 .

Proof of the theorem :

Given 𝑇 ∶ 𝐵 ⟶ 𝐵′ is linear and continuous.

Let 𝐺 be an arbitrary open set in 𝐵. Then, 𝑇(𝐺 ) ⊆ 𝐵′ .

Let 𝑦 ∈ 𝑇(𝐺 ) be arbitrary.

Then 𝑦 = 𝑇(𝑥), for some 𝑥 ∈ 𝐺.

Since 𝐺 is open, ∃ 𝑥 + 𝑆𝑟 in 𝐵 such that 𝑥 ∈ 𝑥 + 𝑆𝑟 ⊆ 𝐺.

⟹ 𝑇(𝑥) ∈ 𝑇(𝑥 + 𝑆𝑟 ) ⊆ 𝑇(𝐺 ) ⟹ 𝑦 ∈ 𝑇(𝑥) + 𝑇(𝑆𝑟 ) ⊆ 𝑇(𝐺 ) ⟹ 𝑦 ∈ 𝑦 + 𝑇(𝑆𝑟 ) ⊆ 𝑇(𝐺 )

By lemma, ∃ 𝜀 > 0 such that 𝑆𝜀′ ⊆ 𝑇(𝑆𝑟 ) .

⟹ 𝑦 ∈ 𝑦 + 𝑆𝜀′ ⊆ 𝑦 + 𝑇(𝑆𝑟 ) ⊆ 𝑇(𝐺 ) .

⟹ 𝑦 is an interior point of 𝑇(𝐺 ).

Since 𝑦 ∈ 𝑇(𝐺 ) was arbitray, all the points of 𝑇(𝐺 ) are interior points and hence 𝑇(𝐺 ) is open in 𝐵′ .

Hence, 𝑇 ∶ 𝐵 ⟶ 𝐵′ is an open map.

Corollary : A continuous one-one linear transformation from a Banach space 𝐵 onto a Banach space 𝐵′ is a
homeomorphism.

Exercise : Let (𝐵, ‖ ∙ ‖) and (𝐵1 , ‖ ∙ ‖1 ) be two normed linear spaces. A function

𝑑2 (𝑧1 , 𝑧2 ) = max{ 𝑑 (𝑥1 , 𝑥2 ), 𝑑1 (𝑦1 , 𝑦2 ) } , ∀ 𝑧1 = (𝑥1 , 𝑦1 ), 𝑧2 = (𝑥2 , 𝑦2 ) ∈ 𝐵 × 𝐵1

where 𝑑 & 𝑑1 are metrics induced by ‖ ∙ ‖ & ‖ ∙ ‖1 on 𝐵 & 𝐵1 respectively, is a metric on 𝐵 × 𝐵1 .