FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Abstract. These are the notes prepared for the course MTH 405 to
be offered to graduate students at IIT Kanpur.

Contents
1. Basic Inequalities
2. Normed Linear Spaces: Examples
3. Normed Linear Spaces: Elementary Properties
4. Complete Normed Linear Spaces
5. Various Notions of Basis
6. Bounded Linear Transformations
7. Three Basic Facts in Functional Analysis
8. The Hahn-Banach Extension Theorem
9. Dual Spaces
10. Weak Convergence and Eberleins Theorem
11. Weak* Convergence and Banachs Theorem
12. Spectral Theorem for Compact Operators
References

1
3
5
6
9
15
17
20
23
25
28
30
31

1. Basic Inequalities
Exercise 1.1 : (AM-GM Inequality) Consider the set
An = {x = (x1 , , xn ) Rn : x1 + + xn = n, xi 0 every i},
and the function g : Rn R+ given by g(x1 , , xn ) = x1 xn . Verify:
(1) An is a compact subset of Rn , and g is a continuous function.
(2) Let z = (z1 , , zn ) Rn be such that maxxAn g(x) = g(z). Then
zi = 1 for all i.
(Hint. Let zp = min zi and zq = max zi for some 1 p, q n.
Define y = (y1 , , yn ) An by yp = (zp + zq )/2 = yq and yi = zi
for i 6= p, q. If zp < zq then
P g(y) > g(z).)
(3) Let x An . Set := n1 ni=1 xi and y := (x1 /, , xn /) An .
Then g(y) 1.
Q
P
Conclude that ( ni=1 xi )1/n n1 ni=1 xi .
Exercise 1.2 : (Characterization of differentiable convex functions) Let f :
[a, b] R be a differentiable function. For a < x < y < b, verify:
1

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

(1) If f is convex then f 0 is monotonically increasing.

Hint. Let h > 0 be such that x + h < y. Then
f (x + h) f (x)
f (y) f (x)
f (y) f (y h)

.
h
yx
h
(2) If f 0 is monotonically increasing then f is convex.
Hint. Let z = (1 t)x + ty. Then there exists c1 (x, z) and
c2 (z, y) such that
f (z) = f (x) + f 0 (c1 )(z x), f (y) = f (z) + f 0 (c2 )(y z).
Exercise 1.3 : (Characterization of twice differentiable convex functions)
Let f : [a, b] R be a twice differentiable function. Show that f is convex
(resp. strictly convex) if and only if f 00 0 (resp. f 00 > 0).
Hint. Necessary part follows from the last exercise. For sufficiency part,
use Taylors mean value theorem.
Remark 1.4 : The exponential is strictly convex.
Exercise 1.5 : Let a1 , a2 be positive numbers and let p1 , p2 > 0 be such
that p1 + p2 = 1. Prove that ap11 ap22 p1 a1 + p2 a2 . Equality holds iff a1 = a2 .
Hint. The logarithm log x1 is strictly convex.
Exercise 1.6 : (Youngs Inequality) Let p, q > 1 be conjugate exponents
(that is, 1/p + 1/q = 1). For positive numbers a, b prove that
ap bq
+ .
ab
p
q
p
q
Equality holds iff a = b .
Hint. Let a1 = ap , a2 = bq and p1 = 1/p, p2 = 1/q in the preceding exercise.
Exercise 1.7 : (Geometric Proof of Youngs Inequality) Let p, q > 1 be
conjugate exponents (that is, 1/p + 1/q = 1). Given positive real numbers
a b, consider
D1 := {(x, y) R2 : 0 x a, 0 y xp1 }
D2 := {(x, y) R2 : 0 y b, 0 x y q1 }.
Verify the following:
(1) The intersection of D1 and D2 is {(x, y) R2 : 0 x a , y = xp1 }
(Hint. Since (p 1)(q 1) = 1, y = xp1 iff x = y q1 ).
(2) The rectangle {(x, y) R : 0 x a, 0 y b} is contained in
the union D1 D2 .
Conclude that ab ap /p + bq /q. Equality holds iff ap = bq .
Exercise 1.8 : Assume the Holders (resp. Minkowskis) inequality for finite
sequences, and derive it for sequences.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Exercise 1.9 : (H
olders Inequality for measurable functions) Let p, q > 1
be conjugate exponents. Let f and g be Lebesgue measurable complexvalued functions. Then f g is measurable such that
Z
 Z
1/p Z
1/q


p
q
 f (x)g(x)dx
|f (x)| dx
|g(x)| dx
.


Hint. Let kf kp :=

|f (x)|p dx

1/p

, f = f /kf kp . By Youngs Inequality,

|f(x)||
g (x)| |f(x)|p /p + |
g (x)|q /q.
Now integrate both sides.
Exercise 1.10 : Prove Minkowskis Inequality for measurable functions.
2. Normed Linear Spaces: Examples
Throughout these notes, the field K will stand either for R or C.
Exercise 2.1 : For 1 p < , for x = (x1 , , xn ) Kn , consider

1/p
n
X
kxkp :=
|xj |p .
j=1

Show that

(Kn , k

kp ) is a normed linear space.

Hint. Minkowskis Inequality.

Exercise 2.2 : What goes wrong in the last exercise if 0 < p < 1 ?
Hint. If p = 1/2 then calculate the p-norms of (1, 1), (1, 0) and (2, 1).
Exercise 2.3 : How does a unit disc Dp := {(x, y) R2 : |x|p + |y|p < 1} in
(R2 , k kp ) look like ? Whether D makes sense ?
Hint. Dp is invariant under reflections along X and Y axes: (x, y) Dp
iff (x, y) Dp for , {1}. Now plot xp + y p = 1 for x, y 0.
Exercise 2.4 : For 1 p < , let lp stand for

X
p
l := {(an ) :
|an |p < }.
n=1

Show that

lp

P
p 1/p .
is a normed linear space with norm k(an )kp := (
n=1 |an | )

Exercise 2.5 : For a non-empty subset X of R (endowed with the subspace

topology), let b(X) denote the vector space of bounded functions f : X K,
and C(X) denote the vector space of continuous functions f : X K. Define
kf k := supxX |f (x)| for f C(X). Verify:
(1) k k defines a norm on b(X).
(2) k k defines a norm on Cb (X) := b(X) C(X).

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Exercise 2.6 : Discuss the last exercise in case X equals

(1) {1, , n}. What is C({1, , n}) ?
(2) the set N of positive integers. In this case, b(N) = Cb (N). This is
commonly denoted as l .
Exercise 2.7 : Let C (n) (0, 1) consist of functions f (t) on (0, 1) having n
continuous and bounded derivatives. Verify that
n
X
kf k := sup{
|f (k) (t)| : 0 < t < 1}
k=0

defines a norm on C (n) (0, 1).

We say that (Lebesgue) measurable functions f is equivalent to g (for
short, f g) if f and g agree outside a set of (Lebesgue) measure 0. Let [f ]

1/p
R
|f (x)|p dx
.
denote the equivalence class containing f, and let k[f ]kp :=
Exercise 2.8 : Verify the following:
(1) is an equivalence relation.
(2) If two measurable functions f, g are equivalent then k[f ]kp = k[g]kp .
For simplicity, we denote the equivalence class [f ] containing f by f itself.
Exercise 2.9 : For a measurable set X, let Lp (X) denote the set of all
(equivalence classes of) measurable functions f for which kf kp < . Show
that Lp (X) is a normed linear space with norm kf kp .
A complex-valued function f is said to be -essentially bounded if kkm,
is finite, where kkm, inf{M R+ : |(z)| M outside set of measure 0}.
Exercise 2.10 : If f g then kf k may be not be equal to kgk . Show
that kf km, = kgkm, .
Exercise 2.11 : For a measurable set X, let L (X) denote the set of all
(equivalence classes of) measurable functions f for which kf km, < .
Show that L (X) is a normed linear space with norm kf km, .
Exercise 2.12 : Let t, u, v, w be generators of R3 . For x R3 , define
kxk := inf{|a| + |b| + |c| + |d| : a, b, c, d R such that x = at + bu + bcv + dw}.
Show that (R3 , k k) is a normed linear space.
Hint. If kxk = 0 then there exists a sequence |an | + |bn | + |cn | + |dn | 0
such that x = an t + bn u + cn v + dn w. To see kx1 + x2 k kx1 k + kx2 k, note
{|a| + |b| + |c| + |d| : x1 + x2 = at + bu + cv + dw}
{|a1 + a2 | + |b1 + b2 | + |c1 + c2 | + |d1 + d2 | : xi = ai t + bi u + ci v + di w}.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

3. Normed Linear Spaces: Elementary Properties

Exercise 3.1 : (The norm determined by the unit ball)) Let (X, k k) be a
normed linear space. Let B(x0 , R) := {x X : kx x0 k < R} be the ball
centered at x0 and of radius R. Show that the norm is determined completely
by D(0, R): kxk = inf{R > 0 : x D(0, R)}.
Hint. If x D(0, R) then kxk < R. Thus kxk is at most the entity on
RHS. To see the reverse inequality, argue by contradiction.
We say that two norms k k and k k0 on a normed linear space X are
equivalent if the identity mapping I from (X, k k) onto (X, k k0 ) is a homoemorphism (that is, I is continuous with continuous inverse).
Remark 3.2 : A subset of X is open with respect to k k topology if and
only if it is open with respect to k k0 topology.
Exercise 3.3 : Prove that two norms k k and k k0 on a normed linear
space X are equivalent if and only if there exist positive scalars C1 and C2
such that C1 kxk kxk0 C2 kxk for all x X.
Exercise 3.4 : (Geometric interpretation of equivalence of norms) Show
that two norms k k and k k0 on a normed linear space X (with balls
B(x0 , r) and B0 (x0 , R)) are equivalent if and only B0 (0, 1) B(0, r) and
B(0, 1) B0 (0, R) for some positive constants r and R.
Exercise 3.5 : Show that the equivalence of norms is an equivalence relation. Conclude that all norms k kp on Kn are equivalent by verifying
kxk kxkp np kxk ..
Exercise 3.6 : (All norms on Kn are equivalent) Let k k be an arbitrary
norm on Kn . Verify:
(1) Let e1 , , en denote the standard basis of Kn . Then
n
X
kei k)kxk .
kxk (
i=1

(2) The function f : (Kn , k k ) R given by f (x) = kxk is continuous.

(3) There exists a Kn such that kak = 1 such that kxk kak > 0
for every x Kn whenever kxk = 1.
(4) The norm k k is equivalent to k k .
Exercise 3.7 : (Inequivalent norms on K[x]) Let K[x] denote the vector
space of polynomials in x over K. Let p(x) K[x] be given by p(x) =
Pk
Pk
n

n=0 |cn ||an |. Verify:

n=0 an x . For c := {cn }n=0 , define kpkc :=
(1) k kc defines a norm on K[x].
(2) Let cn = 1/n and dn = n. Then k kc and k kd are not equivalent.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Exercise 3.8 : For 1 p < r , verify the following:

(1) kxkp kxkr for every finite or infinite sequence x.
(2) There does not exist K > 0 such that kxkp Kkxkr for every
infinite sequence x.
Conclude that lr ( lp .
Hint. For the first, let y =: x/kxkp which has norm equal to 1. Note that
1 = kykpp kykrr , and hence kykr 1. For the second part, find a sequence
x such that kxkr < but kxkp = (Try: xk = 1/k q ).
Exercise 3.9R : Let 1 p < r . For measurable function f on [0, 1],
let kf kp := [0,1] |f (x)|p dm, where dm denotes the normalized Lebesgue
measure. Verify:
(1) kf kp kf kr .
(2) There does not exist K > 0 such that kf kr Kkf kp for every
measurable function f on [0, 1].
Conclude that Lr [0, 1] ( Lp [0, 1].
Hint. Choose q > 1 such that p/r + 1/q = 1. Now apply Holders
inequality to f (x) = |x|p and g(x) = 1.
Exercise 3.10 : Let 1 p . Show that lp is separable iff p < .
Hint. Let {an } denote a countable dense subset of K. Then
{an1 e1 + + ank ek : nk N, k 1}
is dense in lp if p < . If p = then any two distinct points in the
uncountable set {x : xn = 0 or 1} are at distance 1.
Exercise 3.11 : Show that a proper subspace of a normed linear space has
empty interior.
Hint. If a proper subspace has non-empty interior then it contains a ball.
4. Complete Normed Linear Spaces
A sequence {xn } in a normed linear space X is Cauchy if kxn xm k 0
as n, m . We say that {xn } is convergent in X if there exists x X
such that kxn xk 0 as n .
Exercise 4.1 : If {xn } in a normed linear space X then show that there
exists a subsequence {xnk } of {xn } such that kxnk xnl k 2k for all l k.
Prove further that {xn } is convergent iff {xnk } is convergent.
A normed linear space X is said to be complete if every Cauchy sequence is
convergent in X. Complete normed linear spaces are also known as Banach
spaces.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Remark 4.2 : Let X be a normed linear space X with norm k k. Then X

is complete iff the metric space X with metric d(x, y) := kxyk is complete.
Exercise 4.3 : Verify the following:
(1) A Banach space is closed.
(2) A closed subspace of a Banach space is complete.
Remark 4.4 : If E is a subset of a Banach space X then the closure of the
linear span linspanE of E is also a Banach space.
Exercise 4.5 : Let X be a normed linear space with two equivalent norms
k k and k k0 . Show that (X, k k) is complete iff (X, k k0 ) is complete.
Exercise 4.6 : Let k k be an arbitrary norm on Kn . Show that the normed
linear space (Kn , k k) is complete.
Hint. In view Exercise 4.5, we need to check that (Kn , kk ) is complete.
Let us see an example of incomplete normed linear space.
Exercise 4.7 : Consider the vector space K[x] of polynomials p(x) in x with
P
the norm kpk := supx[0,1] |p(x)|. Let fk (x) = kn=0 (x/2)n be a sequence
of polynomials in K[x]. Verify:
(1) {fk } is Cauchy.
(2) There is no polynomial g(x) such that kfk gk 0 as k .
Hint. To see (2): uniform convergence implies point-wise convergence.
P
A series nn=0 xn in a normed linear space X is said to be convergent if
Pk
0 as k . We say that
there
Pn exists x X such that k n=0 xn xk P
n
x
is
absolutely
convergent
if
the
series
n=0 kxn k is convergent.
n=0 n
P
Exercise 4.8 : Show that the series n=1 en /n is convergent in l2 but not
absolutely convergent, where en l2 is the sequence with nth entry equal
to 1 and all remaining entries 0.
Remark 4.9 : If {ak :=

Pk

n=0 kxn k}

is Cauchy then so is {yk :=

Pk

n=0 xn }

kyk yl k |ak al |.
Exercise 4.10 : In a Banach space, show that every absolutely convergent
series is convergent.
P
Exercise 4.11 : Let fk be as in Exercise 4.7. Show that
k=0 fk is absolutely convergent, but not convergent.
Exercise 4.12 : Consider the vector space R[x] of polynomials in x is an
with
norm kpkc with c = {1, 1, , } (see Exercise 3.7). Consider f (x) :=
P the
xn
2
n=1 n . Verify the following:

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

(1) f (x) is absolutely convergent.

(2) For any g(x) K[x] of degree d and for any integer k > d,
k

k
X
xn
n=1

n2

g(x)k
(n)

1
.
(d + 1)2

(n)

Exercise 4.13 : Let (x(n) := {x1 , x2 , , }) be a Cauchy sequence in lp .

For  > 0, verify:
(n)

(1) (xi ) K converges to some xi for each i.

(2) For k 1, there exists n0 1 (independent of k) such that
k
X

(n)

|xi

xi |p  for all n n0 .

i=1

Pk

(3) For k 1, i=1 |xi |p ( + kx(n0 ) kp )p .

(4) The normed linear space lp is complete.
Exercise 4.14 : Show that the normed linear space l is complete.
Exercise 4.15 : Let X be a normed linear space with the property that
every absolutely convergent series is convergent. Let {xn } be a sequence in
a normed linear space X such that kxn xm k 2n for m n. Show that
{xn } is convergent.
P
Hint. Let x0 = 0 and yk = xk xk1 for k 1. Examine
k=1 yk .
Theorem 4.16. A normed linear space is complete iff every absolutely convergent series is convergent.
As an application let us see that Lp [0, 1] is complete.
P
Exercise 4.17 : (Riesz-Fischer) Let
n=1 fn be an absolutely convergent
p
series
< . Define the increasing sequence gn (x) :=
Pn in L [0, 1], where 1 pP

|f
(x)|
and
let
g(x)
:=
k
k=1
k=1 |fk (x)| [0, ]. Verify:
(1) g Lp [0, 1] (Hint. Show that there is M > 0 such that kgn kp M
for every n 1. Now apply Fatou Lemma).
(2) g(x) is finite for all x [0, 1] outside
a set E of measure zero. In
P
particular, for x
/ E, sn (x) := nk=1 fk (x) is convergent.
(3) Define the measurable function s by
(
limn sn (x) for x
/E
s(x) =
0 for x E.
Then |s(x)| g(x) for all x [0, 1], and hence s Lp [0, 1].
(4) ksn skp 0 as n (Hint. Lebesgue Convergence Theorem).

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Exercise 4.18 : Let X be a normed linear space with norm k k. Show that
X is complete iff the unit sphere SX in X is a complete metric space with
the metric d(x, y) := kx yk.
Hint. Since SX is closed, the necessary part is immediate. For sufficiency
part, let {xn } be a Cauchy sequence in X. By passing to a subsequence, we
may assume that xn 6= 0 for every n. Now consider two cases (1) inf kxn k = 0
and (2) inf kxn k > 0.
5. Various Notions of Basis
Recall that a subset of B of a vector space V is a basis if every x V
is a linear combination of finitely many elements in B. This notion of basis
from Linear Algebra is also known as Hamel Basis.
Exercise 5.1 : Let {sn } be a Schauder basis for an n-dimensional space X.
Show that the co-ordinate functionals i (i = 1, , n) are continuous.
Hint. Prove that the linear transformation T : X Kn given by T (x) =
(1 (x), , n (x)) is a homeomorphism.
Exercise 5.2 : Let X be a Banach space. For a sequence {xn } in X, let
Ym := linspan{x1 , , xm }. Verify:
(1) The complement T
X \ Ym of the closed set Ym is dense in X.
(2) The intersection X \ Ym is dense in X.
(3) linspan{xn } is a proper subspace of X.
(4) X can not a have countable Hamel basis.
Although any vector space has a Hamel basis, it is too big to have any
utility in case of Banach spaces. Thus the notion of Hamel basis is not
appropriate in the study of Banach spaces. This motivates another notion
of basis named after Schauder:
Definition 5.3 : A sequence {sn } in a Banach space X is said to be a
Schauder basis if for every x X there exists a unique sequence {cn (x)} of
P
scalars cn (x) depending on x such that kx kn=0 cn (x)sn k 0 as k .
Remark 5.4 : For finite-dimensional spaces, the notions of Schauder basis
and Hamel basis coincide.
Exercise 5.5 : Let {sn } be a Schauder basis for a Banach space X. Show
that the mapping cn : x cn (x) is well-defined and linear.
The mapping cn as given in the last exercise are known as the co-ordinate
functionals corresponding to the Schauder basis {sn }.
Exercise 5.6 : Let en denote the sequence with nth entry 1 and all remaining entries equal to 0. Show that {en } forms a Schauder basis for lp for
1 p < .

10

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Hint. To get uniqueness, use the definition of k kp . Note that your

argument fails in case p = .
Exercise 5.7 : Prove that a Banach space with a Schauder basis must be
separable. Conclude the following:
(1) lp is separable for 1 p < .
(2) l can not admit a Schauder basis.
We recall the notion of orthogonal vectors in Kn . Two vectors x and y in
are orthogonal if hx, yi = 0, where
(P
n
xi yi 2 if K = R;
hx, yi := Pni=1
i=1 xi yi 2 if K = C.

Kn

The orthogonal basis for Kn is nothing but a Hamel basis consisting mutually
orthogonal basis vectors. The well-known Gram-Schmidt process allows one
to construct an orthogonal basis from a given Hamel basis.
P
Exercise 5.8 : For x, y Kn , define hx, yi2 := ni=1 xi yi . Show that Kn is
an inner-product space with inner-product hx, yi2 iff K = R.
Exercise 5.9 : Let A Mn (K) be such that A is one-one. Prove that
hx, yiA := hAx, Ayi2 defines an inner-product on Kn .
Find A M2 (R) for which h(x1 , x2 ), (y1 , y2 )iA := x1 x2 + x1 y2 + x2 y1 +
2y1 y2 defines an inner-product on R2 .
If A Mn (K) then
(
transpose of A if K = R;

A :=
conjugate transpose of A if K = C.
Exercise 5.10 : Let A Mn (K) be such that A = A. Show that
hx, yi := hAx, yi2
defines an inner-product on Kn iff the eigen-values of A are positive.
Hint. Use Spectral Theorem for Symmetric/Self-adjoint matrices.
Exercise 5.11 : Let X denote
p an inner-product space with the innerproduct h, i and let k k =
h, i. Show that, for any x, y X, the
following hold true:
(1) (Parallelogram Law)
kx + yk2 + kx yk2 = 2kxk2 + 2kyk2 .
(2) (Polarization Identity)
(


1
kx + yk2 kx yk2 (x, y X) if K = R;
4


hx, yi = 1
2 kx yk2 + ikx + iyk2 ikx iyk2 if K = C.
kx
+
yk
4

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

11

(3) (Pythagorean Identity) If hx, yi = 0 then kx + yk2 = kxk2 + kyk2 .

Hint. Write kx + yk2 as hx + y, x + yi and then use properties of the
inner-product.
Here are two applications of the Parallelogram Law:
Exercise 5.12 : Suppose that n 2. Show that there does not exist
invertible matrix A in Mn (K) such that A maps {x Kn : kxk = 1} onto
{x Kn : kxk2 = 1}.
Exercise 5.13 : Let Y be a subspace of an inner-product space X and let
x X. Verify:
(1) There exists a Cauchy sequence {yn } in Y such that kx yn k
inf{kx yk : y Y } as n (Hint. Parallelogram Law).
(2) If {yn } converges to y Y then hx y, zi = 0 for every z Y
(Hint. Assume that kzk = 1, and note that kx yk kx wk,
where w := y + hx y, ziz.
Theorem 5.14. (Cauchy-Schwarz Inequality) Let X be an inner-product
p
h, i.
space with the inner-product h, i and the induced norm k k =
Then |hx, yi| kxkkyk for every x, y X. Moreover, equality holds iff x, y
are linearly dependent.
hy, yi
Proof. We may assume that hx, yi =
6 0. Letting t = hx,
yi in hy tx, y txi
0, we obtain Cauchy-Schwarz inequality. Equality holds iff hy tx, y txi =
0 iff y = tx.


Corollary 5.15. Let X p

denote an inner-product space with the inner-product
h, i. Show that k k = h, i defines a norm.
Proof. The only difficult part is the triangle inequality, which follows from
the Cauchy-Schwarz Inequality:
kx + yk2 = kxk2 + hx, yi + hy, xi + kyk2 (kxk + kyk)2 .

Remark 5.16 : Note that equality holds in the Cauchy-Schwarz Inequality
iff that holds in the triangle inequality.
Exercise 5.17 : Let X be an inner-product space and let 0 t 1. If
x, y X are linearly independent then ktx + (1 t)yk < 1 iff 0 < t < 1.
Hint. Use the equality part of Cauchy-Schwarz Inequality.
Corollary 5.18. The inner-product is jointly continuous.
Proof. Use the continuity of norm and either the Polarization Identity or
Cauchy-Schwarz Inequality.

Not all norms are induced by an inner-product.

12

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Exercise 5.19 : (Jordan and von Neumann) Let X denote a normed linear
space with the norm k k. If X satisfies the Parallelogram Law
p then the
expression hx, yi, as given in the Polarization Identity, satisfies h, i = kk
and defines an inner-product on X.
In other words, norm on any normed linear space is induced by an innerproduct if and only if it satisfies the Parallelogram Law.
Hint. We divide the verification into four steps:
(1) hx, yi = hy, xi (x, y X).
(2) hx/2, yi = 1/2hx, yi (x, y X).
(3) hx + y, zi = hx, zi + hy, zi = (x, y X).
Note that x + y + z = x + y/2 + y/2 + z.
(4) hx, yi = hx, yi for any C.
Use density of {m/2n : m Z, n N {0}} in R to conclude
that hx, yi = hx, yi for any real .
A Hilbert space is a complete inner-product space.
Exercise 5.20 : Verify the following:
(1) lp is a Hilbert space iff p = 2.
(2) Lp [0, 1] is a Hilbert space iff p = 2.
Hint. k kp satisfies the Paralleogram Law iff p = 2.
The Hardy space H 2 of the unit disc is a normed linear space of complexvalued functions f holomorphic on the unit disc D1 for which
Z 2
d
2
kf kH 2 := sup
|f (rei )|2
< .
2
0<r<1 0
Exercise 5.21 : For the Hardy space H 2 , verify the following:
(1) H 2 is an inner-product space endowed with the inner-product


1
hf, giH 2 :=
kf + gk2H 2 kf gk2H 2 + ikf + igk2H 2 ikf igk2H 2
4
(Hint. Verify that the norm on H 2 satisfies the Paralleogram Law).
(2) If {fn } is a Cauchy sequence in H 2 then
kfn fm kH 2
Rr
for any |z| r < R < 1 (Hint. By the Cauchy Integral Formula,
Z
f (w)
fn (z) fm (z) =
dw (|z| < R < 1).
|z|=R w z
|fn (z) fm (z)|

(3) If {fn } is a Cauchy sequence in H 2 then there exists a holomorphic

function f : D1 C such that kfn f k,K for every compact subset
K of the unit disc D1 (Hint. Weierstrass Convergence Theorem).
(4) H 2 is a Hilbert space.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

13

Remark 5.22 : In the discussion to follow, we confine ourselves to separable

Hilbert spaces. In particular, according to our definition, orthonormal basis
is necessarily (finitely or infinitely) countable. Even after this assumption,
practically, we do not loose much.
Exercise 5.23 : Find an example of a non-separable Hilbert space.
Let X be an inner-product space with inner product h, i. A subset E
of X is orthogonal if hx, yi = 0 for every x, y E. A subset E of X is
orthonormal if every element in E has unit norm and hx, yi = 0 for every
x, y E. We say that y is orthogonal to E if hx, yi = 0 for every x E.
Exercise 5.24 : Let X be a separable inner-product space. Any orthonormal subset of X is countable.

Hint. Suppose {x } is orthonormal. Then {B(x , 1/ 2)} is a collection

of disjoint open
balls in H. Now if {yn } is countable dense subset of H then
each B(x , 1/ 2) must contain at least one yn .
Theorem 5.25. (Orthogonal Decomposition of a Hilbert Space) Let H be a
Hilbert space and let Y be a closed subspace of H. Then every x H can be
written uniquely as y + z, where y Y and z is orthogonal to Y.
Proof. By Exercise 5.13(1), there exists a Cauchy sequence {yn } such that
kxyn k converges to the distance d between x and Y. Note that d is positive
since Y is closed subspace of H. Also, since H is complete and Y is closed,
{yn } converges to some y Y. By Exercise 5.13(2), x y is orthogonal to Y.
To see the uniqueness part, note that y1 +z1 = y2 +z2 implies y1 y2 = z2 z1
belongs to Y and orthogonal to Y. It follows that y1 = y2 and z1 = z2 . 
Example 5.26 : The orthogonal decomposition may not be unique in case
Y is not a closed subspace of H. For instance, consider H := l2 and c00 :=
2
{(x
n}. Then, for P
each k 1,
Pn many
Pn ) ek l 2: xn is non-zero for finitely
ek
ek
as k=1 k + z, where z :=
k=n+1 k is
k=1 k l can
Pn be edecomposed
k
orthogonal to k=1 k .
An orthonormal basis for X is a Schauder basis which is also an orthonormal subset of X.
Example 5.27 : Let en denote the sequence with nth entry equal to 1 and
all other entries 0. Then {en } forms an orthonormal basis for l2 .
Exercise 5.28 : Show that {z n } forms an orthonormal basis for the Hardy
space H 2 .
Exercise 5.29 : Let {en } be an orthonormal subset of an inner-product
space X. Verify the following:
P
2
2
(1) (Bessels Inequality)
n=0 |hx, en i| kxk .

14

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

(2) (Riemann-Lebesgue Lemma) limn hx, en i = 0.

Hint. Apply Pythagorean Identity to y := x
Pk
n=0 hx, en ien .

Pk

n=0 hx,

en ien and z :=

Exercise 5.30 : For n Z, let En (t) := eint . Verify:

2
(1) {En } is an
R orthonormal subset of L [0, 2]. R
(2) limn [0,2] f (t) cos(nt)dt = 0 = limn [0,2] f (t) sin(nt)dt for
any f L2 [0, 2].
Exercise 5.31 : Let {en } be an orthonormal basis of a Hilbert space H.
Verify the following:
P
P
(1) Every h H takes the form
n=0 hh, en ien (Hint. If h =
n=0 n en
then n = hh, en i).
(2) The co-ordinate functionals corresponding to the orthonormal basis
{en } are continuous.
P
2
(3) (Parsevals Identity) khk2 =
n=0 |hh, en i| .
Exercise 5.32 : (Maximal Orthonormal Set) Let {en } be an orthonormal
set in a Hilbert space H with the property: If x H such that hx, en i = 0
for all n then x = 0. Show that {en } is an orthonormal basis.
P
Hint. Examine hx
n=0 hx, en ien , em i.
Exercise 5.33 : For n Z, let En (t) := eint . RLet f L1 [0, 2] be such that
hf, En i2 = 0 for every n Z. Define F (t) := [0,t] f (s)ds. Verify:
(1) There is a scalar a K such that hF a, En i2 = 0 for every n Z
(Hint. Integration by Parts).
(2) There exists a sequence of trigonometric polynomials pn such that
kF a pn k 0 as n (Hint. Stone-Weierstrass Theorem).
(3) F = a a. e. (Hint. kF ak2 kF a pn k kF ak1 ).
(4) f is zero almost everywhere.
Conclude that {En } is an orthonormal basis for L2 [0, 2].
Exercise 5.34 : Show that, for every f L2 [0, 2], we have
Z
k
X
|f (s)
hf, eint ieins |2 ds 0 as k .
[0,2]

n=k

Scholium 5.35 : Every separable Hilbert space has an orthonormal basis.

Proof. This is an application of the Zorns Lemma: Suppose a partially
ordered set P has the property that every totally ordered subset has an upper
bound in P. Then the set P contains at least one maximal element.
Let P be the collection of all orthonormal subsets of H ordered by inclusion. If h 6= 0 then {h/khk} P, so that P is non-empty. If Q be a totally

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

15

ordered subset of P, then AQ A is an upper bound of Q in P. By Zorns

Lemma, P contains a maximal element B P. Since H is separable, by
Exercise 5.24, B is countable. We contend that B is an orthonormal basis
for H. To see that, let x H be such that hx, yi = 0 for every y B. If
x 6= 0 then B {x/kxk} P, which is impossible. It follows that x = 0,
and hence by Exercise 5.32, B is an orthonormal basis of H.

Exercise 5.36 : Show that there exists an isometric isomorphism from a
separable Hilbert space H onto l2 . In case H is L2 [0, 2], can you see what
will be the isometric isomorphism ?
Hint. For second part, recall the definition of Fourier transform.
6. Bounded Linear Transformations
A linear transformation T : X Y is said to be bounded if there exists
M > 0 such that kT xk M kxk for every x X.
Exercise 6.1 : Show that every linear transformation from a finite dimensional normed linear space is bounded.
Exercise 6.2 : Let X be a normed linear space with Hamel basis {en }
consisting unit vectors en . Define T en = nen and extend T linearly to X.
Verify that T is bounded iff X is finite-dimensional.
Exercise 6.3 : Use Riesz Lemma to show that if X contains an infinite
linearly independent subset {zn } then the unit sphere in X is not compact.
Hint. Let Zn := span{z1 , , zn }. Choose xn Zn+1 such that kxn k = 1
and d(xn , Zn ) 1/2.
Exercise 6.4 : For any subspace Y of H, show that Y := {x H :
hx, yi = 0 for all y Y } is a closed subspace of H.
Exercise 6.5 : (Orthogonal Projection) Let H be a Hilbert space with a
closed subspace Y. In view of Theorem 5.25, every x H may be uniquely
written as y + z, where y Y and z Y . Verify:
(1) P : H Y given by P x = y defines a bounded linear map.
(2) P : H Y given by P x = z defines a bounded linear transformation such that P + P = I.
Theorem 6.6. Let T : X Y be a linear transformation. Then the
following are equivalent:
(1) T is bounded.
(2) T is continuous.
(3) T is continuous at 0.
(4) T maps null-sequences to bounded sequences.

16

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

By a linear functional, we mean a linear transformation from a normed

linear space into K.
Exercise 6.7 : Let f be a linear functional be such that the kernel of f is
closed. Show that f is bounded.
Theorem 6.8. (Riesz Representation Theorem) Let H be a separable Hilbert
space and let f : H K be a bounded linear functional. Then there exists
a unique y H (representative of f ) such that f (x) = hx, yi for all x H.
Proof. Let {en } be an orthonormal basis for H as ensured by Scholium 5.35.
Pk
Pk
2
Consider yk :=
n=1 f (en )en H. Note that f (yk ) =
n=1 |f (en )| =
kyk k2 . Also, there is M > 0 such that |f (yk )| M kyk k. Combining last
two estimates, we obtain kyk k M for every k 1. In particular, y :=
P
n=1 f (en )en H. Check that f (x) = hx, yi.
To see the uniqueness part, note that hx, y1 i = hx, y2 i implies hx, y1 y2 i
for every x H, and hence for x := y1 y2 . This gives y1 = y2 .

Exercise 6.9 : Consider the Hilbert space Hk of trigonometric polynomials
spanned by the orthogonal set {eint : k n k} with k k2 norm. Find
representative of the linear functional f : Hk C given by f (p) = p0 (0).
Exercise 6.10 : Consider the Hilbert space L2 [0, 2]. Find representative
of the linear functional : L2 [0, 2] C given by (f ) = f (0).
Exercise
6.11 : Consider the linear functional f : c00 K given by f (x) =
P
k=1 xk /k. Verify:
p
(1) |f (x)| /6kxk2 .
(2) f does not have any representative in c00 .
Let T : X Y be a bounded linear transformation. The operator norm
of T is given by kT k := sup{kT xk : kxk = 1}. Since T is bounded, kT k is
finite. Further:
Exercise 6.12 : Show that kT k = inf{M R+ : kT xk M kxk}.
Exercise 6.13 : What is the operator norm of (1) a diagonal operator (2)
an isometry (3) an orthogonal projection (4) f appearing in RRT ?
Exercise 6.14 : Let A be an n n self-adjoint matrix with complex entries.
If 1 , , n are eigenvalues of A then show that kAk = max{|1 |, , |n |}.
Hint. If A, B are two n n matrices then kABk kAkkBk. Now apply
the Spectral Theorem.
Theorem 6.15. The set BL(X, Y ) of bounded linear transformations from
X into Y is a normed linear space with norm as the operator norm. If Y is
complete then so is BL(X, Y ).

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

17

Remark 6.16 : Let X 0 denote the normed linear space of bounded linear
functionals on a normed linear space X. Then X 0 is always complete.
Exercise 6.17 : Let H be a separable Hilbert space. Define : H H 0
by (y) = fy , where fy (x) = hx, yi (x H). Show that is an isometric
isomorphism. Conclude that the norm on H 0 is induced by an inner-product.
In the following two exercises, we need the fact that every bijective bounded
linear operator on l2 has bounded inverse (This will be proved later!)
Exercise 6.18 : Define D : l2 l2 by Den = 1/nen . Show that D is
one-one with dense range, but not surjective.
Exercise 6.19 : Let B : l2 l2 be defined by Be1 = 0 and Ben = en1 for
n 2. Show that B I is injective with dense range, which is not surjective.
Hint. Verify: (1) Bx = x implies x = 0. (2) {en }P ran(B). (3) If there
n
2
is > 0 such that k(B I)xk kxk then try x =
n=1 (1 1/k) en l .
The following exercise shows how the information about operators can be
used to know more about spaces.
Exercise 6.20 : Suppose that there exists an injective operator T B(K)
such that ran(T ) ( K is dense in H. Prove that there exist a Hilbert space
H and closed subspaces M and N of H such that M + N ( H is dense in
H and M N = {0}.
Hint. Let H := K K (with inner-product h(x y, x0 y 0 i = hx, x0 i +
hy, y 0 i), M := {x T x : x K} and N := K {0}.
Exercise 6.21 : Let X be a normed linear space. Let M be a closed linear
subspace of X, let N be a finite-dimensional subspace, and let x X. Prove:
(1) M + Kx is a closed subspace of X.
(2) M + N is a closed subspace of X.
Conclude that any finite-dimensional subspace of X is closed in X.
Hint. If yn + n x is a Cauchy sequence in M + Kx then so is n :
kyn + n x (ym + m x)k |n m |d(M, x).
7. Three Basic Facts in Functional Analysis
A mapping T : X Y is said to be open if T (U ) is open in Y for every
open subset U of X.
Exercise 7.1 : Show that translation and dilation are open mappings.
Exercise 7.2 : Show that every linear open map is surjective.
If T has continuous inverse then clearly T is open. What is surprising
is that this is true even if T is not invertible. This is the content of the

18

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

open mapping theorem. Before we state and prove it, let us see a handy
characterization of linear open mappings.
Lemma 7.3. Let T : X Y be a linear transformation. Then the following
are equivalent:
(1) T sends the open unit ball to an open subset of Y.
(2) T is an open mapping.
(3) There exists c > 0 such that for each y Y there corresponds x X
with the properties kxk ckyk and T x = y.
Proof. (1) implies (2): Let U be an open subset of X. Let y = T x T (U ).
x
Then B(x, R) U for some R > 0. Then B(0, 1) UR
:= { yx
R : y U }.
Since T (B(0, 1)) is open in Y, so is R T (B(0, 1)) + T x = T (B(x, R)) in T (U ).
(2) implies (3): Note that 0 T (B(0, 1)) is open, and hence B(0, R)
y
B(0, R) for any
T (B(0, 1)) for some R > 0. Thus for every y Y , r kyk
y
0 < r < R. Then there exists x0 B(0, 1) such that T x0 = r kyk
. Check that
1
T x = y, where x = x0 kyk/r. Check that kxk R kyk, so that (3) holds
with c = R1 .
(3) implies (1): For any r > 0, check that B(y0 , rc1 ) T (B(x0 , r)),
where y0 = T x0 .

Exercise 7.4 : Let X, Y be Banach spaces and let T : X Y be a surjective
bounded linear transformation and let y Y be a unit vector. Verify:
(1) Y =
n=1 T (B(0, n)).
(2) There is an integer k 1 such that T (B(0, k)) contains a non-empty
open set W.
(3) Let y0 W be such that B(y0 , R) W for some R > 0. If z
B(y0 , R) then there exists {un } B(0, 2k) such that T un z y0 .
(4) There exists x1 X such that kx1 k 2k
R and ky T x1 k < 1/2.
1
(5) There exists a sequence {xn } such that kxn k 2k
R 2n1 and ky
n
(T x1 + + T xn )k 1/2 .
(6) The sequence {x1 + + xn } converges to some x X. Moreover,
kxk 4k/R and T x = y.
The preceding exercise and Lemma 7.3 immediately give the following:
Theorem 7.5. (Open Mapping Theorem) Every bounded linear transformation from a Banach space onto a Banach space is open.
The following is often known as the Bounded Inverse Theorem (for short,
BIT).
Corollary 7.6. (Algebraic invertibility implies topological invertibility) A
bijective bounded linear transformation is a homeomorphism.
Example 7.7 : Let (an ) P
be a sequence of positive real numbers. Define
k ka on l by k(bn )ka := n an |bn |. Note that k ka satisfies all conditions

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

19

of a norm except that k ka is [0, ]-valued function. We contend that for

no (an ), k(bn )ka < iff (bn ) l . We prove this by contradiction.

P Note that k ka defines a norm, which makes l Pcomplete. In particular,

n an < . It follows that k(bn )ka k(bn )k
n an . Thus the identity
transformation from l onto the Banach space (l , k ka ) is continuous.
By Bounded Inverse Theorem, there exists M > 0 such that k(bn )k
M k(bn )ka forPall (bn ) l . Letting (bn ) = en , we obtain 1 M an , which
implies that n an = .
Exercise 7.8 : Use BIT to show that C[0, 1] is incomplete in the k kp norm
for any 1 p < .
Corollary 7.9. (Closed Graph Theorem) Let X and Y be Banach spaces.
Let T : X Y be a linear operator, which is closed in the sense that if
xn x and T xn y then T x = y. Then T is continuous.
Proof. Define the norm |x| on X by |x| := kxk+kT xk (x X). Then (X, ||)
is complete. Apply BIT to identity mapping from X onto (X, | |).

Exercise 7.10 : Define a P
linear operator D by Den = nen . P
Extend D

2 |x |2 < } by setting Dx =
linearly to {x = (xn ) l2 :
n
n
n=1
n=1 nxn .
Show that D is closed but not continuous. What goes wrong with CGT ?
Exercise 7.11 : (Multiplication Operators) Let Lp [0, 1] be such that
f Lp [0, 1] whenever f Lp [0, 1]. Define a linear operator M : Lp [0, 1]
Lp [0, 1] by M (f ) = f. Show that M is a bounded linear operator.
Hint. Closed Graph Theorem.
Exercise 7.12 : Show that, up to equivalence of norms, the sup norm is the
only norm on C[0, 1], which makes C[0, 1] complete and which also implies
the point-wise convergence.
Exercise 7.13 : Show that, up to equivalence of norms, the k kp norm
is the only norm on Lp [0, 1], which makes Lp [0, 1] complete and which also
implies the point-wise convergence almost everywhere of a subsequence.
Exercise 7.14 : Suppose X is a Banach space, Y is a normed linear space,
and F B(X, Y ). For n 1, let
Vn := {x X : there exists T F for some kT xk > n}.
Verify the following:
(1) Vn is an open subset of X.
(2) If Vn is dense for every n 1, then there exists a dense subset E of X
such that supT F kT xk = for all x E (Hint. Take E := n Vn ).
(3) If there is an N 1 such that VN is not dense then supT F kT k < .

20

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

The following is often referred to as the Uniform Boundedness Principle

(for short, UBP).
Theorem 7.15. Suppose X is a Banach space, Y is a normed linear space,
and F B(X, Y ). Then only one of the following holds true:
(1) supT F kT k < .
(2) There exists a dense set E X such that supT F kT xk = for all
x E.
Proof. Use the last exercise.

Corollary 7.16. If {Tn } is a sequence of bounded linear operators from

a Banach space X into a normed linear space Y such that limn Tn x
exists for every x X. Then the linear operator T : X Y defined by
T x := limn Tn x (x X) is a bounded linear operator.
Proof. Apply UBP to F := {Tn } to conclude that supn kTn k < . For given
 > 0, choose N 1 (depending on x) such that kTn x T xk < . Then, for
any unit vector x X, kT xk  + supn kTn k.

R
dt
Exercise 7.17 : Let f L1 [, ] with kf k1 := [,] |f (t)| 2 . Let f(k) =
R
ikt dt for k Z. Verify the following:
2
[,] f (t)e
R
Pn
dt
ikx =

(1)
k=n f (k)e
[,] f (t)Dn (x t) 2 , where
Dn (t) :=

n
X
k=n

eikt =

sin((n + 1/2)t)
.
sin(t/2)

Pn

1
k=1 k

(2) kDn k1
(Hint. |Dn (t)| 2| sin((n + 1/2)t)|/|t|).
P
(3) If n : C[, ] C is given by n (f ) := nk=n f(k) then n is a
bounded linear functional with kn k = kDn k1 .
P
ikt

(4) There exists f C[, ] such that the Fourier series

k= f (k)e
diverges at any given point in [, ].
8. The Hahn-Banach Extension Theorem
In this section, we discuss the norm-preserving extension problem, that
is, the problem of extending a given linear functional defined on a subspace
linearly and continuously to the given space such that norm is preserved:
Let Y be a subspace of X, and let f : Y K be a bounded linear functional.
We wish to find a linear functional g : X K such that
(1) g(y) = f (y) for every y Y, and
(2) kgk = kf k.
If this happens then we say that g is a norm-preserving extension of f.
Remark 8.1 : (1) above implies that kf k kgk.
Let us see a couple of special cases in which the norm-preserving extension
problem can be solved.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

21

Exercise 8.2 : If Y is a dense subspace of X then any bounded linear

functional f : Y K extends uniquely to a bounded linear functional
g : X K such that kgk = kf k.
Exercise 8.3 : If Y is a subspace of a Hilbert space H and f : Y K then
there exists a linear functional g : X K such that g(y) = f (y) for every
y Y, and kgk = kf k.
Hint. g := f PY , where PY is the orthogonal projection of H onto Y .
To treat the general case, let us first solve the problem for a subspace of
co-dimension 1 over real field.
Exercise 8.4 : Let Y be a subspace of the normed linear space X over R
such that X \ Y is a 1 dimensional space spanned by x1 . Let f : Y R be
a bounded linear functional. Verify the following:
(1) For y1 , y2 Y ,
f (y2 ) kf kky2 + xk f (y1 ) + kf kky1 + xk.
(2) For y Y, there exists R (independent of y) such that
f (y) kf kky + x1 k f (y) + kf kky + x1 k
(Hint. Take sup over left, and then inf over right in (1). Any
between the sup and inf works).
(3) | + f (y)| kf kky + xk for every y Y.
(4) Define the functional g : Y + Rx1 R by g(y + x1 ) = f (y) + .
Then g1 is a well-defined linear functional such that kgk = kf k.
Exercise 8.5 : Let X be a normed linear space and let Y be a subspace of
X (over R). Given a bounded linear functional f : Y R, consider
P := {(Z, g) : g : Z R is a norm-preserving bounded linear extension of f }.
Verify the following:
(1) P is a non-empty partially ordered set with order defined by
(Z1 , g1 ) (Z2 , g2 ) if Z1 Z2 and g1 (z) = g2 (z) for all z Z1 .
(2) Suppose Q := {(Zi , gi ) P : i I} is a totally ordered subset of P.
Then (Z, g) is a upper bound of Q in P, where
[
Z :=
Zi , and g(z) = gi (z) if z Zi .
iI

(3) If (Z, g) is the maximal element of P then Z = X.

(4) There exists a norm-preserving bounded linear extension of f.
Exercise 8.6 : Let X be a normed linear space and let Y be a subspace of
X (over C). Let f : Y C be a bounded linear functional. Verify:

22

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

(1) There exists a bounded R-linear functional u : Y R such that

f (y) = u(y) iu(iy) for all y Y and kf k = kuk ((Hint. For y Y,
find R such that |f (y)| = f (ei y) = u(ei y) kukkyk).
(2) Let v : X R be a norm-preserving extension of u (as guaranteed
by the previous exercise). Then g(x) = v(x) iv(ix) is a normpreserving extension of f .
We combine last two exercises to obtain Hahn-Banach extension Theorem.
Theorem 8.7. Let X be a normed linear space and let Y be a subspace of
X (over K). Then, for any bounded linear functional : Y K, there
exists a bounded linear functional : X K such that
(y) = (y) (y Y ) and kk = kk.
Corollary 8.8. If x 6= y X then there exists a bounded linear functional
g : X K such that g(x) 6= g(y).
Proof. Consider the one-dimensional subspace Y spanned by xy and define
f (t(x y)) = tkx yk. Now apply HBT.

Exercise 8.9 : Show that
kxk = sup{|f (x)| : f is a bounded linear functional such that kf k 1}.
Corollary 8.10. If M is a closed subspace of X and x X \ M then there
exists a bounded linear functional g : X K of unit norm such that g = 0
on M and g(x) = d(x, M ).
Proof. Consider the subspace Y := M +Kx, and define f (y+x) = d(x, M )
for y M and K. Now apply HBT.

Exercise 8.11 : Let {Mk }k1 be a countable collection of closed subspaces
of H. Show that
_
{k1 Mk } =
Mk .
k1

Hint. Consider the two closed subspaces S1 and S2 of H:

_
S1 :=
Mk , S2 := {k1 Mk } .
k1

Verify that S1 S2 . Suppose, we have the strict inclusion, S1 ( S2 , and

apply the previous corollary.
Exercise 8.12 : Show that polynomials are dense in the Hardy space of the
unit disc.
Hint. Consider Mk = z k H 2 . By the Identity
Theorem, k1 Mk = {0}.
W

Apply the last exercise to conclude that k1 Mk = H 2 . Since Mk is the

vector space of polynomials of degree less than k, we are done.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

23

Remark 8.13 : Apart from an application of HBT, this is not the best
solution.
Exercise 8.14 : Let M be a closed subspace of X. Show that
\
M=
ker f,
f F

where F is the space of all bounded linear functionals f on X such that

M ker f.
The last exercise may be used to give a proof of Runges Theorem from
C-analysis [2, Chapter III, Section 8].
9. Dual Spaces
Let X be a normed linear space. The dual space X 0 of X is defined as the
normed linear space of all bounded linear functionals f : X K. We have
seen that X 0 is a Banach space with norm kf k := sup{|f (x)| : kxk 1}
(compare this with the conclusion of Exercise 8.9).
Exercise 9.1 : X 6= {0} if and only if X 0 6= {0}.
Hint. Let x X be non-zero. Then f (x) = kxk is a bounded linear
functional on Kx. Now apply HBT.
Exercise 9.2 : For p < , find dual space of (1) (Kn , k kp ) (2) lp .
Hint. Let
P q be such that 1/p + 1/q = 1. Define F ((yn )) = y , where
y ((xn )) = n xn yn . Check that ky k kykq . We will see in the class that
ky k = kykq if y lq . If (lp )0 then = y with y = ((e1 ), (e2 ), , ). If
p = 1 then kyk kkkei k1 = kk. Otherwise, let zm = (y1 , , ym , 0, , )
and note that kzm k = kzm kq . Since kzm k kk, we obtain kykq =
limm kzm kq kk.
Exercise 9.3 : Let {f1 , f2 , , } be a countable dense subset of the unit
sphere in X 0 . Let Q be a dense subset of K. Verify the following:
For
all n 1, there is xn X such that kxn k = 1 and |fn (xn )| > 1/2.
Pm
{ i=1 ki xi : ki Q} is a countable dense subset of Y := linspan{xn }.
Y is dense in X if and only if the following is true: For every f X 0
such that f (y) = 0 for all y Y implies f = 0.
(4) Let f X 0 be such that f (y) = 0 for all y Y . Then f = 0 (Hint.
Suppose kf k = 1 and find n 1 such that kfn f k < 1/2. However,
|fn (xn )| = |fn (xn ) f (xn )| kfn f k).

(1)
(2)
(3)

Conclude that X is separable whenever so is X 0 .

Remark 9.4 : The dual of l can not be l1 .

24

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Exercise 9.5 : Let Y be a dense subspace of X. Define F : Y 0 X 0 by

F (f ) = g, where g is the unique continuous extension of f to X (discussed
in the class). Show that F is an isometric isomorphism.
Remark 9.6 : For p < , the dual of (c00 , k kp ) is lq .
Let us calculate the dual of the Banach space c0 of sequences converging
to 0 (with sup norm).
Exercise 9.7 : Verify the following:
P
0
(1) For y l1 , define fy : c0 K by fy (x) =
j=1 xj yj . Then fy (c0 )
such that kfy k kyk1 .
(2) If xj = yj /yj for 1 j n and 0 otherwise. Then fy (x) kyk1 .
(3) kfy k = kyk1 .
(4) Every f (c0 )0 is of the form fy for some y l1 .
Remark 9.8 : The dual of (c00 , k k ) is l1 .
Exercise 9.9 : Show that the dual of L2 [0, 1] is L2 [0, 1] itself.
Hint. Riesz Representation Theorem for Hilbert spaces.
Let us try to understand the dual of L .
Proposition 9.10. (Dual of L is bigger than L1 ) Let gR L1 [0, 1]. Then
the linear functional g : L [0, 1] C given by g (f ) := [0,1] f (t)g(t)dt is
bounded. Furthermore, the mapping F : L1 [0, 1] (L )0 given by F (g) =
g is injective but nor surjective.
Proof. We prove only that F is not surjective. Consider the subspace Y :=
C[0, 1] of X := L [0, 1] and the bounded linear functional : Y C
given by (f ) = f (0). By HBT, there exists : X C such that (f ) =
(f ) (f Y ) and kk = Rkk. Suppose that there exists g L1 [0, 1] such
that = g . But then [0,1] f (t)g(t)dt = 0 for every f Y such that
f (0) = 0. Let f C[0, 1]. Given  > 0, let f C[0, 1] be a function such
that f (0) = 0, f = f on [, 1], and supt[0,] |f (t)f (t)| 2kf k . But then
R
R
| [0,1] f (t)g(t)dt| 2kf k [0,] |g(t)|dt. Since  is arbitrary, by Dominated
R
Convergence Theorem, we get [0,1] f (t)g(t)dt = 0 for every f C[0, 1]. By
R
R
similar argument, one can see that [0,t] g(t)dt = 0, and hence [0,t] g(t)dt = 0.
Differentiating both sides, we obtain g(t) = 0 almost eveywhere. But then
, and hence = 0, which is absurd.

Let us see the dual of L1 [0, 1].
Exercise 9.11R: Let g L [0, 1]. Define the linear functional g : L1 [0, 1]
C by g (f ) := [0,1] f (t)g(t)dt. For (Lp [0, 1])0 , verify the following:
(1) g (L1 [0, 1])0 .

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

25

(2) For a measurable subset of [0, 1], define () = ( ). Then is

a countably additive measure with the property: If is of Lebesgue
measure 0 then () = 0 (Hint. |()| = |( )| kkk k1 =
kk(Lebesgue measure of )).
(3) RThere exists a Lebesgue measurable function g such that () =
g(t)dt for every Lebesgue measurable subset of [0, 1] (This follows
from the Radon-Nikodym Theorem).
(4) kg k = kgk an hence g L [0, 1] (Hint. For  > 0, let A = {x
[0, 1] : |g(x)| > kk + } and let f = A (
g /g). Calculate kf k1 and
examine g (f )).
(5) = g (Hint Check it for simple measurable functions and apply
Dominated Convergence Theorem).
The total variation V (g) of a function g : [0, 1] K is defined as
sup
P

n
X

|g(ti ) g(ti1 )|

i=1

, where sup is taken over all partitions P : {t0 < t1 < < tn } of [0, 1]. We
say that g is of bounded variation if its total variation V (g) is finite.
Recall that the vector space B[0, 1] of bounded linear functionals on [0, 1]
is a normed linear space endowed with the sup norm.
Exercise 9.12 : Let (C[0, 1])0 . Verify the following:
R
(1) If g is of bounded variation then g (f ) := [0,1] f (t)dg(t) defines a
bounded linear functional on C[0, 1]. Moreover, kg k V (g).
(2) There exists a bounded linear functional : B[0, 1] K such that
(f ) = (f ) for all f C[0, 1] and kk = kk.
(3) Define g : [0, 1] K by g(0) = 0 and g(t) = ((0,t] ). Then
V (g) kk (Hint. There exists a real such that |g(ti )g(ti1 )| =
ei (g(ti ) g(ti1 )) = (ei (ti1 ,ti ] )).
P
(4) For f C[0, 1], consider sn := nr=1 f (r/n)((r1)/n,r/n] . Then sn
converges uniformly to f (Hint. Given  > 0, choose n 1 such
that |f (s) f (t)|
P <  for all x, y such that |x y| < 1/n. Note that
sn (t) f (t) = nr=1 (f (r/n) f (t))((r1)/n,r/n] (t)).
(5) = g for g of bounded variation (Hint. (sn ) converges to (f )).
10. Weak Convergence and Eberleins Theorem
As we have seen that the unit ball in an infinite-dimensional normed linear
space can never be compact (in the norm topology). The question is whether
the unit ball is compact in some topology weaker than norm topology. To
answer this, we introduce a new convergence, which relies on the structure
of the dual space.
We say that a sequence {xn } in X converges weakly to x X if x0 (xn )
0
x (x) for every x0 X 0 .

26

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Remark 10.1 : Every convergent sequence is weakly convergent.

Exercise 10.2 : Limit of a weakly convergent sequence is unique.
Hint. HBT.
Exercise 10.3 : Let {xk } be a sequence in Kn . Show that every weakly
convergent sequence is convergent.
Exercise 10.4 : Show that {en } converges weakly to 0 in lp for 1 < p < .
Exercise 10.5 : For n Z, consider the function En (t) = eint . Show that
{En } converges weakly to 0 in L2 [0, 2].
Exercise 10.6 : Show that any orthonormal sequence in a Hilbert space
converges weakly to 0.
None of the sequences discussed in last three exercises converges in norm.
It is natural to know whether there exists an infinite-dimensional normed
linear space in which weak convergence is equivalent to normed convergence.
Theorem 10.7. (Schurs Lemma) A sequence {xn } in l1 is weakly convergent iff {xn } is convergent.
Proof. Suppose there exists a weakly convergent sequence {xn }, which is not
convergent. Replacing xn by xn x, we may assume that {xn } converges
weakly to 0, and kxn k 9 0. Given  > 0, there exists a subsequence of {xn },
denoted by {xn } itself, such that kxn k1 5 for all n 1. By the weak
convergence of {xn }, xn (j) 0 as n (j 1).
SetPn0 = 1 = m0 . Let n1 be the smallest integer greater than n0 such
0
that m
= |xn1 (1)| < . Let m1 be the smallest integer bigger
j=1 |xn1 (j)| P
than m0 such that
j=m1 +1 |xn1 (j)| < . Inductively define {nk } and {mk }
of natural P
numbers such that nk is the smallest integer greater than nk1
mk1
satisfying j=1
|x (j)| < , and mk be the smallest integer greater than
Pnk
mk1 such that j=mk +1 |xnk (j)| < . Define y l by y(1) = 1 and y(j) =
xnk (j)/|xnk (j)| if mk1 < j mk . Clearly, fy (l1 )0 . Since kyk = 1,
|fy (xnk ) kxnk k1 | < 4.
It follows that |fy (xnk )| kxnk k1 4 . However, since {xn } converges
weakly to 0, so does {xnk }. This contradiction completes the proof.

Exercise 10.8 : Show that {en } is not weakly convergent in l1 .
For a normed linear space X, consider the dual X 00 of the dual X 0 of
X. Consider the mapping JX : X X 00 given by JX (x)(x0 ) = x0 (x). By
Exercise 8.9, kJX (x)k = supkx0 k1 |JX (x)(x0 )| = supkx0 k1 |x0 (x)|| = kxk.
Thus JX is a bounded linear isometry which embeds X into X 00 .
We will refer to JX as the canonical embedding of X into X 00 .

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

27

Remark 10.9 : In general, JX is not surjective. For example, if X = l1

then X 0 = l , and X 00 = (l )0 6= l1 . This shows that Jl1 is not surjective.
Example 10.10 : Consider a Hilbert space H. Recall that any g H 0
is given by the inner-product h, gi for some g H. Thus the canonical
embedding JH is given by JH (f )(g) = hf, gi.
Exercise 10.11 : A weakly convergent sequence is bounded.
Hint. It suffices to check that {JX xn } is bounded. Note that for every
x0 X 0 , {(JX xn )(x0 ) = x0 (xn )} is convergent, and hence bounded. Now
apply UBP to {JX xn }.
Exercise 10.12 : Let 1 p < . Show that the following are equivalent:
(1) A sequence {xn } in weakly convergent to x in lp .
(2) {xn } is bounded and xn (j) x(j) for every integer j 1.
0 be of the form f for
Hint. To see that (2) implies (1), let f (lp )P
y
y lq . Given  > 0, there exists N 1 such that jN |y(j)|q q . Note
P
p 1/p kyk +  sup kx xk . Now
that |fy (xn ) fy (x)| ( N
q
n
p
j=1 |xn (j) x(j)| )
PN
p
1/p
choose n sufficiently large, so that ( j=1 |xn (j) x(j)| ) < .

We say that a Banach space X is reflexive if JX is surjective.

Example 10.13 : The spaces lp and Lp [0, 1] are reflexive for 1 < p < .
The spaces l1 and L1 [0, 1] are not reflexive.
Exercise 10.14 : Suppose X is reflexive. Then X is separable if and only
if X 0 is separable.
Hint. In view of Exercise 9.3, it suffices to check that if X is separable
then so is X 0 . Note that X 00 is separable if so is X. Now apply Exercise 9.3.
Exercise 10.15 : Show that C[0, 1] is not reflexive.
Hint. For t [0, 1], define Et : C[0, 1] K by Et (f ) = f (t). Then
Et (C[0, 1])0 . Choose a continuous function f0 with kf0 k = 1 such that
for s 6= t, f0 (s) = 1 and f0 (t) = 0. It follows that that kEt Es k =
supkf k=1 |f (t) f (s)| 1. This shows that (C[0, 1])0 is not separable.
Exercise 10.16 : Let X be a normed linear space and let Y be a subspace.
Define : X 0 Y 0 by (x0 ) = x0 |Y . Show that is a surjective linear
isometry.
Exercise 10.17 : Let X be a Banach space and let Y be a closed subspace
of X. Let y 00 Y . Verify the following:
(1) For every y 0 Y 0 , there is x0 X 0 such that x0 |Y = y 0 .

28

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

(2) Define x00 X 00 by x00 (x0 ) := y 00 (x0 |Y ). Let x X be such that

JX (x) = x00 . Then x Y (Hint. HBT).
(3) JY (x) = y 00 . (Hint. JY (x)(y 0 ) = y 0 (x) = x0 |Y (x) = x0 (x) =
JX (x)(x0 ) = x00 (x0 ) = y 00 (x0 |Y ) = y 00 (y 0 ).).
(4) If X is reflexive then so is Y
The the main result of this section is the following result due to Eberlein.
The proof will be presented in the next section.
Theorem 10.18. If X is reflexive then every bounded sequence has weakly
convergent subsequence.
Corollary 10.19. Let X be a reflexive space. If the norm convergence is
equivalent to the weak convergence then X is finite-dimensional.
Proof. Suppose weak convergence implies norm convergence. Let {xn } be a
bounded sequence. By Eberleins Theorem, {xn } has a weakly convergent
subsequence, and hence by assumption, {xn } has a convergent subsequence.
But then the unit sphere in X is (sequentially) compact. Hence, by Exercise
6.3, X must be finite-dimensional.

11. Weak* Convergence and Banachs Theorem
A sequence {x0n } in X 0 is said to be weak* convergent to some x0 X 0 if
for every x X, x0n (x) x0 (x) as n .
Remark 11.1 : If X is a Banach space then a weak* convergent sequence
is bounded. This follows from UBP.
Example 11.2 : Let {x0n } be a sequence in a Hilbert space H. Thus there
exists yn H such that x0n (x) = hx, yn i. It follows that {x0n } is weak*
convergent if and only there exists y H such that hx, yn i hx, yi.
Recall that the dual of (c0 , k k ) is isometrically isomorphic to l1 : For
every f (c0 )0 , there exists unique y l1 such that

f (x) = fy (x) =
x(j)y(j).
j=1

Moreover, kfy k = kyk.

Exercise 11.3 : Let fyn (c0 )0 for yn l1 . Show that {fyn } is weak*
convergent to fy iff yn (k) y(k) for all k 1 and sup kyn k1 < .
P
Hint. For fixed m 1, check that | m
j=1 x(j)yn (j)| kxk supn kyn k1 .
Letting n , we obtain
m
X
|
x(j)y(j)| kxk sup kyn k1 .
j=1

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

29

Now let m to get kfy k supn kyn k1 < . It follows that y l1 . Let
 > 0 and x c0 . Choose N 1 large enough so that |x(j)| <  for j N.
Note that for large n 1, |yn (j) y(j)| < /N.
|fyn (x) fy (x)|

N
X

|x(j)||(yn (j) y(j))| +

j=1

|x(j)||(yn (j) y(j))|

j=N

kxk  + (kyn k1 + kyk1 ).

The following is a special case of the Banach-Alagou Theorem.
Theorem 11.4. If X is a separable then every bounded sequence in X 0 has
a weak* convergent subsequence.
Proof. Suppose that X has a countable dense subset {xk }. Let M :=
supn kx0n k. Since |x0n (x1 )k M kx1 k, {x0n (x1 )} is a bounded sequence in
K. By the Bolzano-Weierstrass Theorem, {x0n (x1 )} has a convergent subsequence {x0n1 (x1 )}. Since |x0n1 (x2 )k M kx2 k, {x0n1 (x2 )} is a bounded
sequence in K. Again by the Bolzano-Weierstrass Theorem, {x0n1 (x2 )} has a
convergent subsequence {x0n2 (x2 )}. Inductively, for k 1, {x0nk1 (xk )} has
a convergent subsequence {x0nk (xk )}.
For fixed k 1, consider the sequence {x0nn (xk )}. Note that x0nn (xk )
belongs to the convergent sequence {x0nk (xk )} for n k. Hence {x0nn (xk )}
is also convergent. To complete the proof, let x X. Given  > 0, choose k
large enough so that kx xk k < . For m, n 1, note that
|x0mm (x) x0nn (x)| |x0mm (x) x0mm (xk )| + |x0mm (xk ) x0nn (xk )|
+ |x0nn (xk ) x0nn (x)|
2M  + |x0mm (xk ) x0nn (xk )|.
Thus {x0nn (x)} is a Cauchy sequence. It follows that x0nn converges to a
linear functional x0 . Finally, note that kx0 k M.

Remark 11.5 : If X is separable then the closed unit ball in X 0 is weak*
sequentially compact.
Now we are in a position to prove Theorem 10.18.
Proof of Eberleins Theorem. Suppose that X is reflexive, and let {xn } be a
bounded sequence in X. Note that Y = {xn } is closed and separable subspace of X. By Exercise 10.17, Y is reflexive, and hence Y 0 is separable. By
11.4, the bounded sequence {JY (xn )} in Y 00 has a weak* convergent subsequence {JY (xnk )}. Now if x0 X 0 then x0 (xnk ) = x0 |Y (xnk ) = JY (xnk )(x0 |Y )
converges as desired.


30

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

12. Spectral Theorem for Compact Operators

Let X, Y be Banach spaces. A linear operator T from X into Y is said to
be compact if for every bounded sequence {xn } in X, {T xn } has a convergent
subsequence.
Remark 12.1 : If S is compact and T is bounded then ST, T S are compact.
Exercise 12.2 : Show that every finite-rank linear mapping is compact.
Proposition 12.3. Compact operators form a closed subspace of B(X, Y ).
Proof. Given  > 0, find N 1 such that kT TN k < . Now find a
convergent sequence {TN xnk } using compactness of TN . Check that {T xnk }
is a Cauchy sequence.

Exercise 12.4 : Let {an } be a bounded sequence. Show that the diagonal
operator on l2 with diagonal entries a1 , a2 , is compact iff {an } c0 .
Hint. If T is a diagonal operator with diagonal entries b1 , b2 , , then
kT k = supn |bn |. The sufficient part follows from this. WLOG, assume
that {an } is bounded from below. To see the necessary part, note that
kDen Dem k2 = a2n + a2m > c > 0 if {an }
/ c0 .
Rx
Exercise 12.5 : Consider (T f )(x) = 0 f (y)dy (x [0, 1]) and
Z k/n
n1
X
f (y)dy (x [0, 1]).
(Tn f )(x) =
[k/n,(k+1)/n) (x)
k=0

Verify the following:

(1) T B(L2 (0, 1), L2 (0, 1)).
(2) Tn B(L2 (0, 1), L2 (0, 1)) has n-dimensional range.
(3) kT Tn k n1/2 .
(4) T is compact.
Exercise 12.6 : If T is compact then show that the closure of {T x : kxk
1} is sequentially compact.
Exercise 12.7 : Consider the linear T : L2 (0, ) L2 (0, ) given by
Z
1 x
(T f )(x) =
f (t)dt (x (0, )).
x 0
Show that T is not compact.
Hint. Consider fn (t) = n if 0 < t 1/n2 , and 0 otherwise. Note that
kT fn k 1 for all n. However, hT fn , gi 0 for all g L2 (0, ).
Theorem 12.8. Show that if T is compact and the range R(T ) is closed
then the closed unit ball in R(T ) is sequentially compact. In particular, R(T )
is finite-dimensional in this case.

FUNCTIONAL ANALYSIS: NOTES AND PROBLEMS

Proof. Use Open Mapping Theorem.

31

Exercise 12.9 : Let be a non-zero complex number and T : X Y be

compact. Then T |ker(T ) is compact with closed range ker(T ).
We say that T has an eigenvalue if ker(T ) is non-zero.
Exercise 12.10 : Show that the eigenspace ker(T ) of a compact operator
corresponding to non-zero eigenvalue is finite-dimensional.
A bounded linear operator on a Hilbert space H is normal if T T = T T .
We say that T is self-adjoint if T = T.
The operator of multiplication by L on L2 is normal. In fact,

M = M , where (z)
= (z). Note that M is self-adjoint iff is realvalued.
Exercise 12.11 : If N is normal then so is N I for any scalar . Use
this to deduce that the eigenspaces corresponding to distinct eigenvalues are
orthogonal.
Exercise 12.12 : Consider the linear operator T on L2 [0, 1]:
Z 1
Z x
yf (y)dy + x
(1 y)f (y)dy (x [0, 1]).
(T f )(x) = (1 x)
0

Verify the following:

(1) T B(L2 (0, 1), L2 (0, 1)) is compact and self-adjoint (Hint. Use
Exercise 12.5 and Remark 12.1).
(2) If T f = f then for some integer n 1, f (x) = c sin(nx) for some
scalar c and = 1/n2 2 (Hint. Calculate second derivative of T f ).
Theorem 12.13. Let T be a normal operator on an infinite-dimensional
Hilbert space. If T is compact then there exists an orthonormal basis {en }
of ker(T ) and a sequence {n } of complex numbers (possibly repeated) such
that T en = n en . Moreover, the following hold:
(1) For each n 1, ker(T n ) is finite-dimensional.
(2) n 0 as n .
(3) {n } has no accumulation
P point except 0.
(4) For any x H, T x =
n=1 n hx, en ien .
References
[1]
[2]
[3]
[4]
[5]

B. Bollab
as, Linear Analysis: An Introductory Course, Cambridge Univ. Press, 1999.
J. Conway, A Course in Functional Analysis, Springer-Verlag, New York 1997.
S. Kumaresan, A Problem Course in Functional Analysis, private communication.
B. Limaye, Functional Analysis, New Age International Limited, New Delhi, 1996.
Milan Miklavcic, Applied Functional Analysis and Partial Differential Equations,
World Scientific, Singapore, 1998.
[6] M. Schechter, Principles of Functional Analysis, American Mathematical Society,
Providence, Rhode Island, 2002.