Functional Analysis I

Unit I

Definition. Let X be a linear space over K (K = C or R). A norm on X is a

function k k : X → R such that for every x, y ∈ X and k ∈ K

1. kxk ≥ 0 and kxk = 0 if and only if x = 0

2. kx + yk ≤ kxk + kyk

3. kkxk = |k|kxk

Definition. A normed linear space (nls) is a linear space X with a norm k k on it.

Note : A nls X is a metric space with respect to the metric defined by d(x, y) = kx − yk
for all x, y ∈ X (verify)

Examples :

1. Let X = K. Then | | (absolute value) is a norm on K.

2. Let X = K n . Let p ≥ 1 be a real number.

For x = (x(1), x(2), . . . x(n)) ∈ K n , define
1
(
{|x(1)|p + |x(2)|p + · · · + |x(n)|p } p if 1 ≤ p < ∞
kxkp =
max{|x(1)|, |x(2)|, . . . |x(n)|} if p = ∞

Then clearly, kxkp is a norm on K n . (Verify)

3. Sequence Space
Define ∞
X
lp = {(x(1), x(2), . . .), x(j) ∈ K, |x(j)|p < ∞}
i=1

and
l∞ = {(x(1), x(2), . . .), x(j) ∈ K, sup |x(j)| < ∞}

That is, for 1 ≤ p < ∞, lp is the space of all p-summable scalar sequences and for
p = ∞, l∞ is the space of all bounded scalare sequences.
Clearly, lp , 1 ≤ p ≤ ∞ is a linear space with respect to component wise addition
and scalar multiplication.

1
 For x = (x(1), x(2), . . .) ∈ lp , define
1
(
{|x(1)|p + |x(2)|p + · · · } p if 1 ≤ p < ∞
kxkp =
sup{|x(1)|, |x(2)|, . . .} if p = ∞

Then kxkp is a norm for 1 ≤ p ≤ ∞. (Verify)

4. Function Space
Let T be a set and let B(T ) denote the set of all bounded scalar valued functions
on T . For x ∈ B(T ), define

kxk∞ = sup{|x(t)|, t ∈ T }

Then kxk∞ is a norm on B(T ). (Verify)

This norm is known as Sup norm.

5. Lp Space
Let m denote the Lebesgue measure on R and E be a measurable subset of R.
For 1 ≤ p < ∞, consider the set Lp (E) of all equivalance classes of p-integrable
R
1
functions on E. (p-integrable functions means E |x|p dm p < ∞)
For x ∈ LP (E), let
Z  p1
kxkp = |x|p dm
E
.
Then kxkp is a norm on Lp space.

For p = ∞, if m({t ∈ E, x(t) > α}) = 0, then x is known as essentially bounded

function and α is known as essential bound for |x| on E.
inf{α > 0} is known as essential supremum of |x| on E and is denoted by essupE |x|.
Consider the set L∞ (E) of all equivalent classes of essentially bounded functions.
Define kxk∞ = esssupE |x|
Then kxk∞ is a norm on L∞ (E) space.

Some Subspaces of lp Spaces

• C = {x ∈ l∞ , x(j) converges in K as j → ∞}

• C0 = {x ∈ C, x(j) converges to 0 in K as j → ∞}

• C00 = {x ∈ lp , all but finitely many x(j)0 s = 0}

Some Subspaces of B(T ) Space

• C(T ) = {x ∈ B(T ), x is continuous on T }

2
 • C0 (T ) = {x ∈ C(T ), for every  > 0, there is a compact set E ⊂ T such that |x(t)| <
 for all t ∈
/ E}

• Cc (T ) = {x ∈ C(T ), there is a compact set E ⊂ T such that x(t) = 0 for all t ∈

/
E}

Theorem. 5.2

(a) Let Y be a subspace of a normed space X. Then Y and its closure Ȳ are normed
spaces with the induced norm.

(b) Let Y be a closed subspace of a normed space X. For x + Y in the quotient space
X/Y , let
|||x + Y ||| = inf{kx + yk; y ∈ Y }.
Then ||| ||| is a norm on X/Y , called the quotient norm.

A sequence (xn + Y ) converges to x + Y in XY

if and only if there is a sequence (yn )
in Y such that (xn + yn ) converges to x in X.

(c) Let k kj be a norm on a linear space Xj , j = 1, 2, . . . , m. Fix p such that 1 ≤ p ≤ ∞.

For x = (x(1), x(2), . . . , x(m)) in the product space X = X1 × X2 × · · · × Xm , let
(
(kx(1)kp1 + · · · kx(m)kpm )1/p , if 1 ≤ p < ∞
kxkp =
max{kx(1)k1 , . . . , kx(m)km }, if p = ∞

Then k kp is a norm on X.
A sequence (xn ) converges to x in X if and only if (xn (j)) converges to x(j) in Xj
for every j = 1, 2, . . . m.

Proof. (a) Since Y is a subspace of X, it is easy to verify all the conditions of a nls on
Y and hence Y is a nls.

Now, to prove that Y is a nls, it is enough to prove that Y is subspace of X.

For that, let x, y ∈ Y .
Then there exists sequences (xn ) and (yn ) in Y such that xn → x and yn → y as
n → ∞.
Since Y is a subspace of X, k1 xn + k2 yn ∈ Y for all k1 , k2 ∈ K.
But we know that k1 xn + k2 yn → k1 x + k2 y as n → ∞.
That is, k1 x + k2 y is a limit point of a sequence in Y .
This implies that k1 x + k2 y ∈ Y .
Hence Y is a subspace of X.

3
(b) Clearly X
Y
is a linear space with respect to the addition (x1 + Y ) + (x2 + Y ) =
x1 +x2 +Y and the scalar multiplication k(x1 +Y ) = kx1 +Y for all x1 +Y, x2 +Y ∈ X
Y
and k ∈ K.

Now we prove that |||x + Y ||| defined by |||x + Y ||| = inf{kx + yk; y ∈ Y } is a norm
on X
Y
.

X
Clearly, by the definition, |||x + Y ||| ≥ 0 for all x + Y ∈ Y
.

Now, suppose that |||x + Y ||| = 0. This implies that inf{kx + yk; y ∈ Y } = 0.
Then, there exists a sequence (yn ) ∈ Y such that x + yn → 0 as n → ∞
=⇒ yn → −x =⇒ −x ∈ Y =⇒ x + Y = 0 as n → ∞.

Now, let x1 + Y, x2 + Y ∈ X Y
and  > 0.
Then, there exist y1 , y2 ∈ Y such that

kx1 + y1 k < inf {kx1 + yk, y ∈ Y } +
2

kx2 + y2 k < inf{kx2 + yk, y ∈ Y } +
2

Therefore, kx1 + y1 k < |||x1 + Y ||| + 2
and kx2 + y2 k < |||x2 + Y ||| + 2 .
Hence
kx1 + x2 + y1 + y2 k = k(x1 + y1 ) + (x2 + y2 )k
≤ kx1 + y1 k + kx2 + y2 k
< |||x1 + Y ||| + |||x2 + Y ||| + 
Since y1 + y2 ∈ Y , we get
|||x1 + x2 + Y k| ≤ |||x1 + Y ||| + |||x2 + Y ||| + 
Since, this is true for every  > 0, we get
|||x1 + x2 + Y k| ≤ |||x1 + Y ||| + |||x2 + Y |||
X
Now, let x + Y ∈ Y
and k(6= 0) ∈ K.
Then
|||k(x + Y )||| = |||kx + Y |||
= inf {kkx + yk, y ∈ Y }
y
= inf{|k| x + , y ∈Y}
k
y
= |k| inf{kx + y1 k, y1 = ∈ Y }
k
= |k| |||x + Y |||

4
 X
Hence ||| ||| is a norm on Y
.

X
Let (xn + Y ) be a sequence in Y
.

First suppose that (yn ) be a sequence in Y such that xn + yn → x in X.

Then,

|||(xn + Y ) − (x + Y )||| = |||(xn − x) + Y |||

= inf{kxn − x + yn k, yn ∈ Y }
≤ kxn − x + yn k
≤ kxn + yn − xk
X
This implies that xn + Y → x + Y in Y
as n → ∞.

Conversely assume that xn + Y → x + Y in X Y

as n → ∞.
Now, |||(xn + Y ) − (x + Y )||| = |||xn − x + Y ||| = inf{kxn − x + yk, y ∈ Y }.
Choose yn ∈ Y such that kxn − x + yn k < |||xn − x + Y ||| + n1 , n = 1, 2, . . .
=⇒ kxn − x + yn k < |||(xn + Y ) − (x + Y )k| + n1 for all n.
=⇒ kxn − x + yn k → 0 as n → ∞
=⇒ xn − x + yn → 0 as n → ∞.
=⇒ xn + yn → x in X as n → ∞.
(c) Let X = X1 × X2 × · · · × Xm
For x = (x(1), x(2), . . . , x(m)) ∈ X, define
 1
 Pm p p
kx(j)k j if 1 ≤ p < ∞
kxkp = j=1

1≤j≤m kx(j)k if p = ∞
max

Then clearly, kxkp is a norm on X for all p such that 1 ≤ p ≤ ∞. (Verify)

Now, if a sequence (xn ) in X converges to x in X if and only if xn − x → 0 as

n→∞
if and only if kxn − xkp → 0 as n → ∞
if and only if kxn (j) − x(j)kj → 0 as n → ∞ for j = 1, 2, . . . m
if and only if xn (j) − x(j) → 0 as n → ∞ in Xj for j = 1, 2, . . . m
if and only if xn (j) → x(j) as n → ∞ in Xj for j = 1, 2, . . . m

This completes the proof of Theorem 5.2

Lemma. 5.3 (Riesz Lemma)
Let X be a normed linear space, Y be a closed subspace of X and Y 6= X. Let r > 0
be a real number such that 0 < r < 1. Then, there exist some xr ∈ X such that kxr k = 1
and r < dist(xr , Y ) ≤ 1.

5
Proof. Consider x ∈ X but x ∈ / Y . (This is possible as Y 6= X).
We know that d = dist(x, Y ) = inf{d(x, y), y ∈ Y }
Since Y is closed, d > 0.
dist(x,Y )
Again, since r < 1, there exists an element y0 ∈ Y such that kx − y0 k < r
.
x−y0
Define xr = kx−y 0k
.
Clearly, kxr k = 1.
Now, dist(xr , Y ) = dist(x−y0, Y )
kx−y0 k
= dist(x,Y
kx−y0 k
)
> r.
Hence dist(xr , Y ) > r.
Also,

dist(xr , Y ) = inf{d(xr , y), y ∈ Y }

= inf{kxr − yk, y ∈ Y }
≤ kxr − 0k (since 0 ∈ Y )
≤ kxr k
≤ 1

Hence r < dist(xr , Y ) ≤ 1.

This completes the proof of the lemma.
Lemma. 5.4 Let X be a normed space and Y be a subspace of X.
(a) For x ∈ X, y ∈ Y and k ∈ K, we have

kkx + yk ≥ |k| dist(x, Y ).

(b) Let Y be finite dimensional. Then Y is complete. In particular, it is closed in X.

Let {y1 , y2 , . . . ym } be a basis for Y and (xn ) be a sequence in Y . If

xn = kn,1 y1 + kn,2 y2 + · · · kn,m ym , n = 1, 2, . . . ,

then (xn ) converges to x = k1 y1 + k2 y2 + · · · km ym if and only if kn,j → kj and

n → ∞ for each j = 1, 2, . . . m.
Also, (xn ) is bounded if and only if (kn,j ) is bounded for each j = 1, 2, . . . , m.
Proof. (a) If k = 0, then the proof is obvious. (since kyk ≥ 0).
If k 6= 0, then
y
kkx + yk = k(x + )
k
y
= |k| kx + y1 k, y1 = ∈ Y
k
≥ |k| dist(x, Y )

(b) Here, we use Mathematical Induction in the dimension m of Y .

6
First, we consider the case of m = 1.
Then Y = {ky, k ∈ K} with y 6= 0.
If (xn )is a Cauchy sequence in Y with xn = kn y, then
kxn − xp k = kkn y − kp yk = k(kn − kp )yk = |kn − kp | kyk
=⇒ |kn − kp | = kxnkyk
−xp k

Hence (kn ) is a Cauchy sequence in K.

Since K is complete, (kn ) converges in K.
Suppose, kn converges to some k in K.
=⇒ kn y → ky as n → ∞.
=⇒ xn → x for some x ∈ Y .
=⇒ (xn ) converges in Y .
Hence, every Cauchy sequence in Y converges in Y .
Therefore, Y is complete and hence is closed.
Now, assume that every m − 1 dimensional subspace of X is complete.

Let Y be an m-dimensional subspace of X and let (xn ) be a Cauchy sequence in

Y . (Here, we have to prove that (xn ) converges in Y ).
Let Z = span{y2 , y3 , . . . ym }.
Then dim(Z) = m − 1 and hence is complete by inductive assumption.
Here, we can express as xn = kn y1 + zn for some kn ∈ K and zn ∈ Z, n = 1, 2, . . ..
Consider

kxn − xp k = kkn y1 + zn − (kp y1 + zp )k

= k(kn − kp )y1 + (zn − zp )k
≥ |kn − kp | dist(y1 , Z) by part (a)

kxn − xp k
=⇒ |kn − kp | ≤
dist(y1 , Z)
(This is possible, as y1 ∈
/ Z and Z is closed implies dist(y1 , Z) > 0 ).
=⇒ (kn ) is a Cauchy sequence in K.
Therefore, kn → k for some k ∈ K.
Now, Consider

kzn − zp k = kxn − kn y1 − (xp − kp y1 )k

= kxn − xp + (kp − kn )y1 k
≤ kxn − xp k + k(kp − kn )y1 k
≤ kxn − xp k + |kn − kp | ky1 k

Hence (zn ) is a Cauchy sequence in Z.

Since Z is complete, (zn ) converges to some z ∈ Z.
Now, kn → k and zn → z implies that kn y1 → ky1 + z as n → ∞.
=⇒ xn → ky1 + z as n → ∞.

7
Clearly, ky1 + z ∈ Y .
Hence (xn ) converges in Y .
That is every Cauchy sequence in Y converges in Y .
Therefore, Y is complete and closed, in this case also.

Hence by Mathematical Induction, every finite dimensional space is complete and

closed.
Now, let {y1 , y2 , . . . , ym } be basis for Y .
Let xn = kn1 y1 + kn2 y2 + · · · + knm ym ∈ Y, n = 1, 2, . . .
Suppose that knj → kj as n → ∞, for each j = 1, 2, . . . , m, then
let x = k1 y1 + k2 y2 + · · · + km ym . Then x ∈ Y .
Consider
kxn − xk = kkn1 y1 + kn2 y2 + · · · + knm ym − (k1 y1 + k2 y2 + · · · km ym )k
= k(kn1 − k1 )y1 + (kn2 − k2 )y2 + · · · + (knm − km )ym k
≤ |kn1 − k1 | ky1 k + |kn2 − k2 | ky2 k + · · · + |knm − km | kym k
→ 0 as n → ∞
=⇒ xn → x as n → ∞

Conversely, suppose that xn → x as n → ∞ where xn = kn1 y1 +kn2 y2 +· · ·+knm ym ∈

Y, n = 1, 2, . . . and x = k1 y1 + k2 y2 + · · · + km ym ∈ Y .
Then,
kxn − xk = kkn1 y1 + kn2 y2 + · · · + knm ym − (k1 y1 + k2 y2 + · · · km ym )k
= k(kn1 − k1 )y1 + (kn2 − k2 )y2 + · · · + (knm − km )ym k
Define Yj = span{yi , i = 1, 2, . . . , m, i 6= j}, j = 1, 2, . . . m.
Then
kxn − xk = k(kmj − kj )yj − yk, y ∈ Yj
≥ |knj − kj | dist(yj , Yj ) by part (a)
Since yj ∈
/ Yj and Yj is closed, dist(yj , Yj ) > 0.
Therefore, |knj − kj | ≤ dist(y1j , Yj ) kxn − xk → 0 as n → ∞.
Hence knj → kj for all j.

Now, if (knj ) is bounded for each j = 1, 2, . . . m, say |knj | ≤ αj , n = 1, 2, . . .

then,
kxn k = kkn1 y1 + kn2 y2 + · · · + knm ym k
≤ kkn1 y1 k + kkn2 y2 k + · · · + kknm ym k
≤ |kn1 |ky1 k + |kn2 |ky2 k + · · · + |knm |kym k
≤ α1 ky1 k + α2 ky2 k + · · · + αm kym k

8
 =⇒ (xn ) is bounded.

Conversely, assume that (xn ) is bounded.

Then,

kxn k = kkn1 y1 + kn2 y2 + · · · + knm ym k

= kknj yj − yk, y ∈ Yj
≥ |knj | dist(yj , Yj ) as discussed above

Therefore,
kxn k
|knj | ≤
dist(yj , Yj )
This implies that (knj ) is bounded.
This completes the proof of the lemma.

Note : An infinte dimensional subspace of a nls need not be closed.

For example, X = l∞ , Y = C00 .
Then Y is a subspace of X, but it is not closed in X. (Explain the details)

Theorem. 5.5 Let X be a normed space. The following conditions are equivalent.

(a) Every closed and bounded subset of X is compact.

(b) The subset {x ∈ X : kxk ≤ 1} of X is compact.

(c) X is finite dimensional.

Proof. (a) =⇒ (b)

Suppose that every closed and bounded subset of X is compact.
Since the subset {x ∈ X : kxk ≤ 1} is closed and bounded, it is compact.

(b) =⇒ (c)
Suppose that the subset {x ∈ X : kxk ≤ 1} of X is compact.
If possible, let X be infinite dimensional.
Let {z1 , z2 , . . .} be an infinite linear independent subset of X.
Define Zn = span{z1 , z2 , . . . zn }, n = 1, 2, . . ..
Then Zn is finite dimensional and hence it is closed subspace of Zn+1 (by previous theo-
rem)
Also, Zn 6= Zn+1
Therefore, by Riesz Lemma, there exists an element xn ∈ Zn+1 such that kxn k = 1 and
1
2
< dist(xn , Zn ).
Hence, (xn ) is a sequence in {x ∈ X, kxk ≤ 1}.
Also, it has no convergent subsequence, because, kxn − xm k > 12 for all m 6= n.
=⇒ {x ∈ X, kxk ≤ 1} cannot be compact.
This is a contridiction to the statement (b).

9
Hence our assumption is wrong.
Hence X is finite dimensional.

(c) =⇒ (a)
Suppose that X if finite dimensional.
Let {y1 , y2 , . . . , ym } be a basis for X.
Let E be a closed and bounded subset of X.
(To prove that E is compact, we have to show that every bounded sequence
in E has a subsequence which converges in E.)
Let (xn ) be a bounded sequence in E.
Therefore, there exists knj ∈ K such that xn = kn1 y1 + kn2 y2 + · · · + knm ym , n = 1, 2, . . . .
Since (xn ) is bounded, (knj ) is also bounded (by previous theorem) for j = 1, 2, . . . m.
By Bolzano-Weirstrass theorem for K, (knj ) has a convergent subsequence, say, (knp j ).
That is (knj ) has a convergent subsequence, say, (knp j ).
Therefore, by previous theorem, (xnp ) also converges in X.
Since (xnp ) ∈ E and E is closed, the limit point of (xnp ) is also contained in E.
Therefore, (xnp ) also converges in E.
That is the sequence (xn ) has a subsequence which converges in E.
Hence E is compact.

This completes the proof the theorem.

Theorem. 5.6 Let X be a normed space.

(a) If E1 is open in X and E2 ⊂ X, then E1 + E2 is open in X.

(b) If E be a convex subset of X. Then the interior E o of E and the closure E of E

are also convex. If E o 6= φ, then E = E o .

(c) Let Y be a subspace of X. Then Y o 6= φ, if and only if Y = X.

Proof. (a) Let x ∈ X and x1 ∈ E1 .

Since E1 is open, there exist some r > 0 such that U (x1 , r) ⊂ E1 .
Now, U (x1 + x, r) = U (x1 , r) + x ⊂ E1 + x
That is some nhd of every every element of E1 + x is contained in E1 + x.
Hence E1 + x is open in X for all x ∈ X.
Now, E1 + E2 = ∪{E1 + x2 , x2 ∈ E2 }, the union of open sets.
Hence E1 + E2 is also open.

(b) To prove that E o is convex

Let 0 < t < 1.
Since E is convex, tx + (1 − t)y ∈ E for all x, y ∈ E.
Therefore, tE o + (1 − t)E o ⊂ E.
Also tE o + (1 − t)E o is an open subset of E.
But E o is the largest open set contained in E.

10
 Hence tE o + (1 − t)E o ⊂ E o .
This implies that E o is a convex subset of E.

To prove that E is convex

Let x, y ∈ E.
Then there exist sequences (xn ) and (yn ) in E such that xn → x and yn → y as
n → ∞.
Since E is convex, this implies that txn + (1 − t)yn ∈ E for all t ∈ (0, 1).
=⇒ txn + (1 − t)yn → tx + (1 − t)y as n → ∞.
Hence tx + (1 − t)y ∈ E. (since limit points of sequences in E is contained in E).
Therefore, E is convex.

(c) First assume that Y = X.

Then Y o = X o = X 6= φ.
Therefore, Y o 6= φ.

Conversely, assume that Y o 6= φ.

We know that Y ⊆ X............(1)
Take x(6= 0) ∈ X.
Let a ∈ Y o .
Since Y o is open, there exists some r > 0 such that U (a, r) ⊂ Y .
Then  
rx rx r
a− a+ = = <r
2kxk 2kxk 2
rx
Therefore, a + 2kxk ∈ U (a, r) ⊂ Y .
Hence x ∈ Y . (Since Y is a subspace and a ∈ Y ).
Hence X ⊆ Y .............(2)
From (1) and (2), we get X = Y .
This completes the proof of the theorem.

Definition. The set {x ∈ X, kxk = 1} is known as unit sphere.

Definition. A nls X is said to be strictly convex if for x, y ∈ X, x 6= y with

kxk = 1 = kyk, then kx + yk < 2.

Notes :
x+y
1. If x and y are points in a unit sphere, then the midpoint 2
is not a point on the
unit sphere.

2. If n ≥ 2 in K n , then k k1 and k k∞ are not strictly convex. (Show by some

examples)

3. In K n , k k2 is strictly convex. (Verify)

11
Definition. A metric d is stronger than a metric d0 if every open subset X with respect
to d0 is also an open subset of X with respect to d. A matric d is equivalent to a metric
d0 if and only if each metric is stronger than the other.

Definition. Let k k and k k0 be two norms associated with matrices d(x, y) = kx − yk

and d0 (x, y) = kx − yk0 on a linear space X. Then k k is said to be stronger than k k0 if
the metric d is stronger than the metric d0 . k k is said to be equivalent to k k0 if the
metric d is equivalent to the metric d0 . k k and k k0 are said to be comparable if one of
them is stronger than the other.

Note : k k is stronger than k k0 whenever kxn − xk → 0 we also have kxn − xk0 → 0 as

n → ∞.

Theorem. 5.7 Let k k and k k0 be norms on a linear space X. Then the norm k k is
stronger than the norm k k0 if and only if there is some α > 0 such that kxk0 ≤ αkxk for
all x ∈ X.
Further, the norm k k is equivalent to the norm k k0 if and only if there are α > 0 and
β > 0 such that βkxk ≤ kxk0 ≤ αkxk for all x ∈ X.
Proof. First assume that k k is stronger than k k0 .
Define the metrics d and d0 such that d(x, y) = kx − yk and d0 (x, y) = kx − yk0 .
Therefore, every open set with respect to d0 is also an open set with respect to d.
Hence, there exist some r > 0 such that {x ∈ X, kxk < r} ⊂ {x ∈ X, kxk0 < 1}
Let x(6= 0) ∈ X and  > 0.
Now,
rx rkxk r
= = <r
(1 + )kxk (1 + )kxk 1+

Hence 0
rx
<1
(1 + )kxk

=⇒ rkxk0 < (1 + )kxk

Since this is true for all  > 0, we get rkxk0 ≤ kxk or kxk0 ≤ 1r kxk
=⇒ kxk0 ≤ αkxk where α = 1r .

Conversely assume that there exist some α > 0 such that kxk0 ≤ αkxk for all x ∈ X.
Let (xn ) be a sequence in X such that d(xn , x) = kxn − xk → 0 as n → ∞.
Then d0 (xn , x) = kxn − xk0 ≤ αkxn − xk → 0 as n → ∞.
=⇒ d is stronger than d0 .
=⇒ k k is stronger than k k0 .

Now assume that k k is equivalent to k k0 .

Therefore, k k is stronger than k k0 and k k0 is stronger than k k.
If and only if there exist some a > 0 and b > 0 such that kxk0 ≤ akxk and kxk ≤ bkxk0

12
for all x ∈ X.
If and only kxk ≤ bkxk0 ≤ bakxk for all x ∈ X.
If and only 1b kxk ≤ kxk0 ≤ akxk for all x ∈ X.
If and only βkxk ≤ kxk0 ≤ αkxk for all x ∈ X where β = 1
b
and α = a.

This completes the proof of the theorem.

Theorem. 6.1 If X is a finite dimensional normed space and Y is a normed space then
every linear map from X to Y is continuous. Conversely, if X is infinite dimensional
and Y 6= {0}, then there is a discontinuous linear map from X to Y .
Proof. Let X be finite dimensional and F : X → Y be linear.
If X = {0}, then there is nothing to prove. (Since F is linear and hence F (0) = 0)
If X 6= 0, let {a1 , a2 , . . . , am } be a basis for X.
Let (xn ) be a sequence in X which converges to x ∈ X.
Therefore, there exist scalar sequences (knj ) ∈ K such that xn = kn1 a1 +kn2 a2 +· · · knm am
and scalars k1 , k2 , . . . km ∈ K such that x = k1 a1 + k2 a2 + · · · + km am .
Now, xn → x if and only if knj → kj for j = 1, 2, . . . , m. (By Lemma 5.4(b))
Now,

F (xn ) = F (kn1 a1 + kn2 a2 + · · · knm am )

= kn1 F (a1 ) + kn2 F (a2 ) + · · · knm F (am )
→ k1 F (a1 ) + k2 F (a2 ) + · · · km F (am ) as n → ∞
→ F (k1 a1 + k2 a2 + · · · + km am )
→ F (x)

That is xn → x if and only if F (xn ) → F (x) as n → ∞.

Hence F is continuous.
If X is infinite dimensional, we can can define a linear map from X to Y which is not
continuous.
Complete this portion as a home work.
Theorem. 6.2 Let X and Y be normed linear spaces and F : X → Y a linear map.
Then, the following are equivalent:
(a) F is bounded on some closed ball about 0
(b) F is continuous at 0
(c) F is continuous on X
(d) F is uniformly continuous on X
(e) kF (x)k ≤ αkxk ∀x ∈ X and some α > 0.
X
(f ) The zero space Z(F ) of F is closed in X and the linear map F̄ : Z(F )
−→ Y defined
by F̄ (x + Z(F )) = F (x), x ∈ X, is continuous.

13
Proof. Here, we prove the theorem in the following order.

(a) =⇒ (e)
Let kF (x)k ≤ β for all x ∈ U (0, r), r > 0.
If x = 0, then F (x) = 0.
rx
If x 6= 0, then, kxk = rkxk
kxk
= r.
rx
Hence ∈ U (0, r).
kxk  
rx
Therefore, F kxk ≤ β.
r
=⇒ kxk kF (x)k ≤ β.
=⇒ kF (x)k ≤ β kxk
r
.
β
=⇒ kF (x)k ≤ αkxk where α = r
> 0.

(e) =⇒ (a)
Let kF (x)k ≤ αkxk for every x ∈ U (0, r), α > 0.
Since x ∈ U (0, r), kxk ≤ r.
Therefore, kF (x)k ≤ αr
=⇒ kF (x)k ≤ β, β = αr.
Hence F is bounded on U (0, r)

(e) =⇒ (b)
Let kF (x)k ≤ αkxk for every x ∈ X, α > 0.
Suppose that (xn ) be a sequence in X such that xn → 0 as n → ∞.
Therefore, kF (xn )k ≤ αkxn k → 0 as n → ∞.
=⇒ F (xn ) → 0 as n → ∞.
That is xn → 0 =⇒ F (xn ) → 0 as n → ∞.
Hence F is continuous at x = 0.

(b) =⇒ (e)
Let F be continuous at x = 0.
If possible, suppose that there exist no such α such that kF (x)k ≤ αkxk for every
x ∈ U (0, r), α > 0.
It now follows
 that,
 for each n, we can find xn such that kF (xn )k > nkxn k
xn
=⇒ F nkxn k >1
xn kxn k
Now, nkxn k
= nkx nk
= n1 → 0 as n → ∞.
 
xn
But, F nkx nk
> 1 and hence not converges to 0 as n → ∞.
That is F is not continuous at x = 0, which is a contridiction.
Therefore, kF (x)k ≤ αkxk for every x ∈ X for some α > 0.

(b) =⇒ (c)
Let F be continuous at 0.

14
Let (xn ) be a sequence in X such that xn → x as n → ∞.
Therefore,xn − x → 0 as n → ∞.
=⇒ F (xn − x) → F (0) = 0 as n → ∞.
=⇒ F (xn ) − F (x) → 0 as n → ∞.
=⇒ F (xn ) → F (x) as n → ∞.
Hence F is continuous on X.

(c) =⇒ (b)
Let F be continuous on X.
Hence F is continuous at every point of X.
In particular, F is continuous at x = 0.

(e) =⇒ (d)
Let kF (x)k ≤ αkxk for every x ∈ X for some α > 0.
Suppose x, y ∈ X.
Then x − y is also in X.
Hence kF (x − y)k ≤ αkx − yk
=⇒ kF (x) − F (y)k ≤ αkx − yk
Hence F is uniformly continuous on X.

(d) =⇒ (e)
Let F be uniformly continous on X.
Since every uniformly continuous function is continuous, F is continuous on X.

(c) =⇒ (f )
Let F be continuous on X.
Then Z(F ) = F −1 ({0}) is closed in X since {0} is closed in Y .
X
Hence, Z(F )
is a nls with respect to the quotient norm k| k|.
X
Consider the map F : Z(F )
→ Y given by F (x + Z(F )) = F (x) for all x ∈ X.
Clearly, it is well-defined. (Verify)
Since F is continous, by (e), there exists an α > 0 such that kF (x)k ≤ αkxk for all
x ∈ X.
Let x ∈ X. Take some z ∈ Z(F ).
Then kF (x + Z(F ))k = kF (x + z + Z(F ))k = kF (x + z)k ≤ αkx + zk
Since this is true for all z ∈ Z(F ), we get
kF (x + Z(F ))k ≤ α inf{kx + zk, z ∈ Z(F )} = αk|x + Z(F )k|
X
By condition (e), this implies that F is continuous on Z(F )
.

(f ) =⇒ (c)
X
Let the zero space Z(F ) of F is closed in X and the linear map F̄ : Z(F )
−→ Y defined
by F̄ (x + Z(F )) = F (x), x ∈ X, is continuous.

15
Then

kF (x)k = kF (x + Z(F ))k

≤ αk|x + Z(F )k| for some α > 0
≤ α inf{kx + zk, z ∈ Z(F )}
≤ αkx + 0k
≤ αkxk for all x ∈ X

Hence F is continuous on X by condition (e).

This completes the proof of the theorem.

Corollary. 6.3

a) Let X and Y be two normed linear spaces and let F : X → Y be a linear map.
Then F is a homeomorphism if and only if there are α > 0 and β > 0 such that

βkxk ≤ kF (x)k ≤ αkxk

for all x ∈ X. In case there is a linear homeomorphism from X onto Y , X is

complete if and only if Y is complete.

b) Let X and Y be nls with X finite dimensional. Then every bijective linear map from
X to Y is a homeomorphism. All norms on X are equivalent and X is complete in
each. If dimX = n, then there is a linear homeomorphism from K n onto X.

Proof. a) First assume that F is a linear map from X to Y such that βkxk ≤ kF (x)k ≤
αkxk for all x ∈ X, for some α > 0 and β > 0.
Then F is injective. (Verify)
Then F is continuous. (By theorem 6.2)
The inverse map F −1 : R(F ) → X is linear. (By theorem 2.4(a))
If x ∈ X, then F (x) ∈ Y .
Let F (x) = y. Then x = F −1 (y).
Now βkxk ≤ kF (x)k for all x ∈ X =⇒ βkF −1 (y)k ≤ kyk
=⇒ kF −1 (y)k ≤ β1 kyk for all y ∈ Y
Hence F −1 is continuous.
Therefore, F is a homeomorphism.

Conversely, assume that F is a linear homeomorphism from X to Y .

Then F and F −1 are continuous.
Hence, there exists a > 0 and b > 0 such that kF (x)k ≤ akxk and kF −1 (y)k ≤ bkyk
for all x, y ∈ X.
=⇒ kF (x)k ≤ akxk and kxk ≤ bkF (x)k for all x ∈ X.

16
 Hence, kxk ≤ bkF (x)k ≤ bakxk for all x ∈ X.
=⇒ 1b kxk ≤ kF (x)k ≤ akxk for all x ∈ X.
1
=⇒ βkxk ≤ kF (x)k ≤ αkxk for all x ∈ X where β = b
and α = a.

Now, suppose that there is a linear homeomorphism from X onto Y .

Then (xn ) is a Cauchy sequence in X if and only if F (xn )) is a Cauchy sequence in
Y.
Also (xn ) converges in X if and only if (F (xn )) converges in Y .
Hence X is complete if and only if Y is complete.

b) Let X and Y be nls with X finite dimensional.

Suppose F is a bijective linear map from X to Y .
Since every linear map defined on a finite dimensional space is continuous, F is
continuous.
Since F is bijective and Y is finite dimensional, F −1 is also continuous and linear.
Hence F is a homeomorphism from X to Y .

Now, let k k and k k0 be two norms on a finite dimensional linear space X.

We know that the identity map I : (X, k k) → (X, k k0 ) is linear and bijective.
Hence, it is a homeomorphism.
Therefore, by (a) part, there exist α, β > 0 such that βkxk ≤ kI(x)k0 ≤ αkxk for
every x ∈ X.
=⇒ βkxk ≤ kxk0 ≤ αkxk for every x ∈ X.
Hence k k and k k0 are equivalent norms.
Hence all norms on X are equivalent.

Now, let dimX = n.

Let {a1 , a2 , . . . an } be a basis of X.
Let x ∈ X.
Then there exists scalars k1 , k2 , . . . kn ∈ K such that x = k1 a1 + k2 a2 + · · · kn an .
Define a map F from K n to X as F ((k1 , k2 , kn )) = x
Then clearly, F is well-defined, linear and bijective.
Hence F is a homeomorphism.
Since F is a homeomorphism and K n is complete, X is also complete.

This completes the proof of the corollary.

Corollary. 6.4 Let X and Y be normed spaces and F : X −→ Y be a linear map such
that the range R(F ) of F is finite dimensional. Then F is continuous if and only if the
zero space Z(F ) of F is closed in X.
In particular, a linear functional f on X is continuous if and only if Z(f ) is closed
in X.

17
Proof. Let X and Y be normed spaces and F : X −→ Y be a linear map such that the
range R(F ) of F is finite dimensional.
Let {y1 , y2 , . . . ym } be a basis for R(F ).
Let xj ∈ X such that F (xj ) = yj , j = 1, 2, . . . m.

X
Now we show that Z(F )
is finite dimensional
X
For that, we show that {x1 + Z(F ), x2 + Z(F ), . . . xm + Z(F )} is a basis for Z(F )
.

Spanning property
X
Take some x + Z(F ) ∈ Z(F )
Then x ∈ X and F (x) ∈ R(F ).
Therefore,
F (x) = k1 y1 + k2 y2 + · · · km ym where k1 , k2 , . . . km ∈ K.
= k1 F (x1 ) + k2 F (x2 ) + · · · + km F (xm )
= F (k1 x1 + k2 x2 + · · · km xm )

=⇒ F (x) − F (k1 x1 + k2 x2 + · · · km xm ) = 0
=⇒ F (x − k1 x1 − k2 x2 − · · · km xm ) = 0
=⇒ x − k1 x1 − k2 x2 − · · · km xm ∈ Z(F )
=⇒ x − k1 x1 − k2 x2 − · · · km xm + Z(F ) = Z(F )
=⇒ x + Z(F ) = k1 x1 + k2 x2 + · · · km xm + Z(F )
=⇒ x + Z(F ) = k1 (x1 + Z(F )) + k2 (x2 + Z(F )) + · · · km (xm + Z(F ))
Hence x+Z(F ) can be expressed as a linear combination of {x1 +Z(F ), x2 +Z(F ), . . . xm +
Z(F )} .
X
Therefore {x1 + Z(F ), x2 + Z(F ), . . . xm + Z(F )} is a spanning set for Z(F )
.

Linearly independent
X
Suppose that k1 (x1 + Z(F )) + k2 (x2 + Z(F )) + · · · km (xm + Z(F )) = Z(F ) in Z(F )
.
This implies that k1 = k2 = · · · = km = 0. verify

X
Hence {x1 + Z(F ), x2 + Z(F ), . . . xm + Z(F )} is a basis for Z(F )
.
X
That Z(F )
is finite dimensional.

X
Now, if Z(F ) is closed, then Z(F )
is finite dimensional nls.
X
In that case, F : Z(F ) → Y defined by F (x + Z(F )) = F (x), x ∈ X is continuous. (by
theorem 6.1)
By thereom 6.2, this is possible if and only if F is continuous.
Therefore, F is continuous if and only if Z(F ) is closed.

18
Now, if f is a linear functional, then R(f ) ⊆ K.
Hence dimR(f ) ≤ 2 and hence R(f ) is finite dimensional.
Hence condition of the theorem is satisfied.
Therefore, f is continuous if and only if Z(f ) is closed.

This completes the proof the corollary.

Definition. Let X and Y be two nls. The set of all bounded linear maps from X to
Y is denoted by BL(X, Y ).
In the above definition, if Y = X, the map is known as an operator on X and
BL(X, X) is usually denoted by BL(X) and is known as bounded operators on X.
If Y = K, BL(X, Y ) is denoted by X 0 and is known as bounded linear functionals
on X.
Notes :
1. If F ∈ BL(X, Y ), F 6= 0, then kF (x)k ≤ αkxk for all x ∈ X and some α > 0.

2. A linear map F ∈ BL(X, Y ) is bounded below if there exists some β > 0 such that
βkxk ≤ kF (x)k for all x ∈ X.

3. A linear mpa F is a homeomorphism if and only if it is bounded and bounded

below.
Theorem. 6.6 Let X and Y be normed spaces. For F ∈ BL(X, Y ), define

kF k = sup{kF (x)k : x ∈ X, kxk ≤ 1}.

Then k k is a norm on BL(X, Y ), called the operator norm.

For all x ∈ X, we have kF (x)k ≤ kF k kxk.
In fact,
kF k = inf{α ≥ 0; kF (x)k ≤ αkxk for all x ∈ X}.
Also, if X 6= {0}, then

kF k = sup{kF (x)k : x ∈ X, kxk < 1} = sup{kF (x)k : x ∈ X, kxk = 1}.

Proof. Clearly, BL(X, Y ) is a linear space under the pointwise opertions

(F + G)(x) = F (x) + G(x) and (kF )(x) = k F (x) for all x ∈ X and k ∈ K.

(verify)
Define kF k = sup{kF (x)k : x ∈ X, kxk ≤ 1}.
Clearly k k is a norm on BL(X, Y ). (verify)

To prove that kF (x)k ≤ kF k kxk for all x ∈ X.

If X = {0}, then the proof is trivial.
If X 6= {0}, then

19
If F = 0 then F (x) = 0 for all x ∈ X. Hence, kF k = 0 and kF (x)k = 0.
Hence the result is proved in the case of F = 0.

If F 6= 0 then take a non zero x ∈ X.

x
In that case, kxk = 1 and
 
x
kF (x)k = F kxk
kxk
 
x
= kxk F
kxk
≤ kF k kxk (by the definition of kF k)

Hence kF (x)k ≤ kF k kxk for all x ∈ X.

Proof of remaining part

Let α0 = inf{α ≥ 0; kF (x)k ≤ αkxk for all x ∈ X}.
β = sup{kF (x)k : x ∈ X, kxk = 1}
and γ = sup{kF (x)k : x ∈ X, kxk < 1}.
Then clearly, β ≤ kF k and γ ≤ kF k................(1)

Let x(6= 0) ∈ X, r ∈ R with 0 < r ≤ 1.

Since F is linear,
 
kxk rx
kF (x)k = F
r kxk
kxk
≤ sup {kF (z)k , z ∈ X, kzk = r}
r
If r = 1, then kF (x)k ≤ kxk 1
β
=⇒ α0 ≤ β.......................(2)

If r < 1, then
kxk kxk
kF (x)k ≤ sup{kF (z)k, z ∈ X, kzk < 1} = γ
r r
Letting r → 1, we get kF (x)k ≤ γ kxk for all x ∈ X
=⇒ α0 ≤ γ.....................(3)

(Now, we show that kF k ≤ α0 )

For that, consider α ≥ 0 such that kF (x)k ≤ αkxk for all x ∈ X.
Taking sup over all x ∈ X with kxk ≤ 1,

sup{kF (x)k, x ∈ X, kxk ≤ 1} ≤ sup{αkxk, x ∈ X, kxk ≤ 1}

20
 =⇒ kF k ≤ α
Since α0 is the infimum of all such α’s, we get
kF k ≤ α0 .................(4)
From (1), (2), (3) and (4), we get kF k = α0 = β = γ.

This completes the proof.

21