112

2.10 Normed Spaces of Operators and Dual Spaces

Let X and Y be normed spaces (both real or both complex). Consider the
set B(X, Y ) consisting of all bounded linear operators from X into Y . That
is, each such operator is defined on all of X and its range lies in Y . We aim
to show that B(X, Y ) can itself be made into a normed space.
First, B(X, Y ) becomes a vector space if we define
• The sum T1 + T2 of two operators T1, T2 ∈ B(X, Y ) by

(T1 + T2)x = T1x + T2x, ∀ x ∈ X.

• The scalar multiplication αT for T ∈ B(X, Y ) and a scalar α by

(αT )x = αT x, ∀ x ∈ X.

Theorem 4.15 (Normed Space of Operators). The vector space

B(X, Y ) of all bounded linear operators from a normed space X into a
normed space Y is itself a normed space with norm defined by
∥T x∥
∥T ∥ = sup = sup ∥T x∥.
x∈X ∥x∥ x∈X
x̸=0 ∥x∥=1

Theorem 4.16 (Completeness of B(X, Y )). If Y is a Banach space,

then B(X, Y ) is a Banach space.

Proof. We aim to show that B(X, Y ) is a Banach space under the operator
norm
∥T x∥
∥T ∥ = sup = sup ∥T x∥, T ∈ B(X, Y ).
x∈X ∥x∥ x∈X
x̸=0 ∥x∥=1
Step 1: Completeness of B(X, Y ).
 113

Let (Tn) be a Cauchy sequence in B(X, Y ). Then for every ε > 0, there
exists N ∈ N such that

∥Tn − Tm∥ = sup ∥Tn(x) − Tm(x)∥ < ε for all n, m ≥ N . (42)

x∈X
∥x∥=1

For all x ∈ X and n, m ≥ N , we have

∥Tnx − Tmx∥ = ∥(Tn − Tm)x∥

≤ ∥Tn − Tm)∥∥x∥

< ε∥x∥ (by (42)). (43)

Now for any fix x ∈ X and given ε0 > 0, we may choose ε = εx so that

ϵx∥x∥ < ϵ0

. Thus from (43), we get

∥Tnx − Tmx∥ < ϵ0 for all n, m ≥ N .

This shows that the sequence (Tnx) is a Cauchy sequence in Y . Since Y is

complete, (Tnx) converges in Y . Let T xn → y. Clearly, the limit y ∈ Y
depends on the choice of x ∈ X. Thus, we can define an operator T : X → Y
by
y = T x = lim Tnx.
n→∞

Step 2: Linearity of T .
For x, y ∈ X and α, β ∈ K (R or C), we have


T (αx + βy) = lim Tn(αx + βy) = lim αTnx + βTny = αT x + βT y,
n→∞ n→∞
114

where the linearity of Tn and the properties of limits were used. Thus, T is
linear.
Step 3: Boundedness of T . Sine Tmx → T x as m → ∞ and (43)
holds for every m ≥ N . Thus by letting m → ∞ in (43), we get

∥Tnx − T x∥ = ∥Tnx − lim Tmx∥

m→∞

= lim ∥Tnx − Tmx∥

m→∞

≤ ε∥x∥ for all n ≥ N . (44)

This shows that each Tn − T with n ≥ N is a bounded linear operator. Since

for each n, Tn is bounded, and B(X, Y ) is a vector space,

T = Tn − (Tn − T )

is bounded. Thus T ∈ B(X, Y ).

Step 4: Convergence of Tn to T in B(X, Y ). From (44), we have
∥Tnx − T x∥
≤ ε (for all 0 ̸= x ∈ X and n ≥ N )
∥x∥
∥Tnx − T x∥
=⇒ sup ≤ ε (for all n ≥ N )
x∈X ∥x∥
x̸=0

=⇒ ∥Tn − T ∥ ≤ ε (for all n ≥ N ).

Thus, Tn → T in the operator norm, and T ∈ B(X, Y ).

Since every Cauchy sequence in B(X, Y ) converges to a bounded linear
operator T ∈ B(X, Y ), the space B(X, Y ) is complete. Hence, B(X, Y ) is
a Banach space.
 115

Definition 4.19 (Dual Space of a Normed Space). Let X be a

normed space. The dual space (adjoint space or conjugate space) X ′ of a
normed space X is defined as the set of all bounded linear functionals on X,
with norm defined by

|f (x)|
∥f ∥ = sup = sup |f (x)|.
x∈X ∥x∥ x∈X
x̸=0 ∥x∥=1

Theorem 4.17 (Dual Space). The dual space X ′ of a normed space

X is a Banach space (whether or not X is complete).

Proof. We know that a linear functional on X maps X into R or C (the scalar

field of X). Now since R or C, taken with the usual metric, is complete, it
follows that X ′ is B(X, Y ), where Y = R or C. Hence, X ′ is a Banach
space. □

Recall that an isomorphism of a normed space X onto another normed

space X̃ is a bijective linear operator T : X → X̃ that preserves the norm:

∥T x∥ = ∥x∥, ∀ x ∈ X.

(Hence, T is isometric.) X and X̃ are called isomorphic normed spaces

(denoted as X ∼
= X̃).

Examples.

Example 4.14 (Space Rn). The dual space of Rn is isomorphic to Rn.

116

Justification: Let the dual space of Rn is denoted by (Rn)′, which is the

space of all bounded linear functionals f : Rn → R. Let {e1, e2, . . . , en}
be the standard basis for Rn. Then every x = (ξ1, ξ2, · · · , ξn) ∈ Rn can be
represented uniquely as

x = ξ1e1 + ξ2e2 + · · · + ξnen. (45)

Let f ∈ (Rn)′ be arbitrary. Then by applying linearity of f to (45), we get

f (x) = ξ1f (e1) + ξ2f (e2) + · · · + ξnf (en)

n
X
= ξk f (ek )
k=1
Xn
= ξk βk , where βk = f (ek ).
k=1

By the triangular inequality and Cauchy–Schwarz inequality, we have

n
X
|f (x)| ≤ |ξk βk |
k=1

n
! 21 n
! 12
X X
≤ ξk2 βk2
k=1 k=1

n
! 12
X
= ∥x∥ ξk2 .
k=1

Taking supremum over all x of norm 1, we get

n
! 21
X
∥f ∥ ≤ ξk2 . (46)
k=1

Claim: (Rn)′ ∼
= Rn .
 117

Define the map T : (Rn)′ → Rn by

T f = (β1, β2, · · · , βn) for all f ∈ (Rn)′ and βk = f (ek ), 1 ≤ k ≤ n.

Linearity of T : Let f, g ∈ (Rn)′ and a, b ∈ R. By definition of T , we have

T f = (β1, β2, . . . , βn), where βk = f (ek ),

and

T g = (γ1, γ2, . . . , γn), where γk = g(ek ).

Clearly af + bg ∈ (Rn)′. By definition, we get

T (af + bg) = (af + bg)(e1), (af + bg)(e2), · · · , (af + bg)(en)


= af (e1) + bg(e1), af (e2) + bg(e2), · · · , af (en) + bg(en)



= af (e1), af (e2), · · · , af (en) + bg(e1), bg(e2), · · · , bg(en)



= a β1, β2, · · · , βn + b γ1, γ2, · · · , γ1

= aT f + bT g.

Therefore, T is linear.
T is one-to-one: To prove that T is one-to-one, we show that if T f = T g,
then f = g.
Suppose T f = T g for f, g ∈ (Rn)′. By the definition of T ,

T f = (β1, β2, · · · , βn) and T g = (γ1, γ2, · · · , γn),

where βk = f (ek ) and γk = g(ek ) for k = 1, 2, . . . , n.

118

Since T f = T g, we have

f (ek ) = g(ek ) for all k = 1, 2, . . . , n.

Now, for any vector v = (v1, v2, · · · , vn) ∈ Rn, we can write v as a linear
combination of the standard basis vectors
Xn
v= vk ek , where vk ∈ R.
k=1

Using the linearity of f and g, we have

n
! n
X X
f (v) = f vk e k = vk f (ek ),
k=1 k=1

and similarly, !
n
X n
X
g(v) = g vk e k = vk g(ek ).
k=1 k=1
Since f (ek ) = g(ek ) for all k, it follows that

f (v) = g(v) for all v ∈ Rn.

Thus, f = g. Therefore, T is one-to-one.

Surjectivity of T : Let z = (α1, α2, . . . , αn) ∈ Rn be arbitrary. We need
to find fz ∈ (Rn)′ such that T (fz ) = (α1, α2, . . . , αn).
Define fz ∈ (Rn)′ by specifying its action on the standard basis vectors
{e1, e2, . . . , en} as follows

fz (ek ) = αk for k = 1, 2, . . . , n.

Then by definition of T , we have

T (fz ) = (α1, α2, · · · , αn).

 119

Therefore, T is onto.
T is an Isometry: To show that T is an isometry, we need to prove that
∥T f ∥ = ∥f ∥ for all f ∈ (Rn)′.
Recall that the Euclidean norm of T f is given by
v
u n
uX
∥T f ∥ = t βk2
k=1

where, by definition, T f = (β1, β2, . . . , βn), and βk = f (ek ).

For any f ∈ (Rn)′, we have
|f (x)|
∥f ∥ ≥ for all 0 ̸= x ∈ Rn
∥x∥
In particular, for x = (β1, β2, · · · , βn), we have
Pn 2
k=1 βk
∥f ∥ ≥ Pn 1/2
( k=1 βk2)
n
! 12
X
= βk2 . (47)
k=1

Thus, from (46) and (47), we get

v
u n
uX
∥f ∥ = t βk2 = ∥T f ∥.
k=1

Since ∥T f ∥ = ∥f ∥ for all f ∈ (Rn)′, the map T is an isometry.

Thus, we have shown that T is the linear operator, which is bijective and
norm-preserving. Therefore, T : (Rn)′ → Rn is an isometric isomorphism.
Hence, (Rn)′ and Rn are isomorphic. □
120

Example 4.15 (Space l1). The dual space of l1 is l∞.

Justification: A Schauder basis for l1 is {ek }, where ek = (δkj ) (1 in the

k-th place and 0 otherwise). Every x ∈ l1 has a unique representation
∞
X
x= ξk ek where x = (ξ1, ξ2, · · · ). (48)
k=1

Let the dual space of l1 is denoted by (l1)′. For any f ∈ (l1)′, since f is linear
and bounded, from (48), we have
∞
X
f (x) = ξk f (ek )
k=1
X∞
= ξk γk (49)
k=1

where γk = f (ek ) are uniquely determined by f .

Also, for each k, ∥ek ∥1 = 1 and by boundedness of f , we get

|γk | = |f (ek )| ≤ ∥f ∥∥ek ∥1 = ∥f ∥.

This holds for all k. Hence

sup |γk | ≤ ∥f ∥. (50)

k

This shows that (γk ) ∈ l∞.

Define the operator T : (l1)′ → l∞ by

T f = (γk ) where γk = f (ek ), k ∈ N and f ∈ (l1)′.

Clearly T is linear.