DMTH505 Measure Theorey and Functional Analysis PDF

MeasureTheoreyandFunctional
Analysis
DMTH505
MEASURE THEORY AND
FUNCTIONAL ANALYSIS
Copyright © 2012 Anuradha
All rights reserved
Produced & Printed by

EXCEL BOOKS PRIVATE LIMITED
A-45, Naraina, Phase-I,
New Delhi-110028
for
Lovely Professional University
Phagwara
SYLLABUS
Measure Theory and Functional Analysis
Objectives: This course is designed for the of analysis of various types of spaces like Banach Spaces, Hilbert Space, etc. and also
Sr. No Description
1 Differentiation and Integration: Differentiation of monotone functions,
Functions of bounded variation,
2 Differentiation of an integral, Absolute continuity
3 Spaces, Holder, Minkowski inequalities, Convergence and Completeness
4 Bounded linear functional on the Lp spaces, Measure spaces, Measurable
Functions, Integration
5 General Convergence Theorems, Signed Measures, Radon-Nikodym
theorem.
6 Banach spaces: Definition and some examples, Continuous linear
transformations, The Hahn-Banach theorem
7 The natural imbedding of N in N**, The open mapping theorem, The
closed graph theorem,
8 The conjugate of an operator, The uniform boundedness theorem, The
uniform boundedness theorem, Hilbert spaces : The definition and some
simple properties
9 Orthogonal complements, Orthonormal Sets, The conjugate space H*, The
Adjoint of an Operator, Self Adjoint Operators
10 Normal and Unitary Operators, Projections, Finite dimensional spectral
theory : the spectrum of an operator on a finite dimensional
Hilbert space, the Spectral theorem
CONTENTS
Unit 1: Differentiation and Integration: Differentiation of Monotone Functions 1
Unit 2: Functions of Bounded Variation 11
Unit 3: Differentiation of an Integral 27
Unit 4: Absolute Continuity 36
Unit 5: Spaces, Hölder 50
Unit 6: Minkowski Inequalities 65
Unit 7: Convergence and Completeness 72
Unit 8: Bounded Linear Functional on the L p-spaces 82
Unit 9: Measure Spaces 100
Unit 10: Measurable Functions 106

Unit 11: Integration 121
Unit 12: General Convergence Theorems 141
Unit 13: Signed Measures 158
Unit 14: Radon-Nikodym Theorem 166
Unit 15: Banach Space: Definition and Some Examples 174
Unit 16: Continuous Linear Transformations 183
Unit 17: The Hahn-Banach Theorem 201
Unit 18: The Natural Imbedding of N in N** 211
Unit 19: The Open Mapping Theorem 217
Unit 20: The Closed Graph Theorem 225
Unit 21: The Conjugate of an Operator 231
Unit 22: The Uniform Boundedness Theorem 238
Unit 23: Hilbert Spaces: The Definition and Some Simple Properties 244
Unit 24: Orthogonal Complements 253
Unit 25: Orthonormal Sets 262
Unit 26: The Conjugate Space H* 273
Unit 27: The Adjoint of an Operator 283
Unit 28: Self Adjoint Operators 294
Unit 29: Normal and Unitary Operators 305
Unit 30: Projections 316
Unit 31: Finite Dimensional Spectral Theory 329

Unit 1: Differentiation and Integration: Differentiation of Monotone Functions
Unit 1: Differentiation and Integration: Differentiation Notes
of Monotone Functions
CONTENTS
Objectives
Introduction
1.1 Differentiation and Integration
1.1.1 Lipschitz Condition
1.1.2 Lebesgue Point of a Function
1.1.3 Covering in the Sense of Vitali
1.1.4 Four Dini's Derivatives
1.1.5 Lebesgue Differentiation Theorem
1.2 Summary
1.3 Keywords
1.4 Review Questions
1.5 Further Readings
Objectives
After studying this unit, you will be able to:
 Understand differentiation and integration
 Describe Lipschitz condition and Lebesgue point of a function
 State Vitali’s Lemma and understand its proof.
 Explain four Dini’s derivatives and its properties
 Describe Lebesgue differentiation theorem.
Introduction
Differentiation and integration are closely connected. The fundamental theorem of the integral
calculus is that differentiation and integration are inverse processes. The general principle may
be interpreted in two different ways:
1. If f is a Riemann integrable function over [a, b], then its indefinite integral i.e.
x
F : [a, b] R defined by F (x) = f (t) dt is continuous on [a, b]. Furthermore if f is

a
continuous at a point xo [a, b], then F is differentiable thereat and F (xo) = f (xo).
2. If f is Riemann integrable over [a, b] and if there is a differentiable function F on [a, b] such
that F f (x) for x [a, b], then
x
f (t) dt = F (x) – F (a) [a x b].

a
LOVELY PROFESSIONAL UNIVERSITY 1

Notes 1.1 Differentiation and Integration
1.1.1 Lipschitz Condition
Definition: A function f defined on [a, b] is said to satisfy Lipschitz condition (or Lipschitzian
function), if a constant M > 0 s.t.
|f (x) – f (y)| M |x – y|, x, y [a, b].
1.1.2 Lebesgue Point of a Function
Definition: A point x is said to be a Lebesgue point of the function f (t), if
x n
1
Lim |f (t) f (x)|dt 0.
h 0 h x
1.1.3 Covering in the Sense of Vitali
Definition: A set E is said to be covered in the sense of Vitali by a family of intervals (may be open,
closed or half open), M in which none is a singleton set, if every point of the set E is contained in
some small interval of M i.e., for each x E, and > 0, an interval I M s.t. x I and (I) .
The family M is called the Vitali Cover of set E.
Example: If E = {q : q is a rational number in the interval [a, b]}, then the family I qi
1 1
where I qi q ,q ,i N is a vitali cover of [a, b].
i i
Vitali's Lemma
Let E be a set of finite outer measure and M be a family of intervals which cover E in the sense of
Vitali; then for a given > 0, it is possible to find a finite family of disjoint intervals {I k, k = 1, 2,
… n} of M, such that
m* E I
k 1
k < .
Proof: Without any loss of generality, we assume that every interval of family M is a closed
interval, because if not we replace each interval by its closure and observe that the set of end
points of I1, I2, …… In has measure zero.
[Due to this property some authors take family M of closed intervals in the definition of Vitali’s
covering].
Suppose 0 is an open set containing E s.t. m* (0) < m* (E) + 1 < we assume that each interval in
M is contained in 0, if this can be achieved by discarding the intervals of M extending beyond 0
and still the family M will cover the set E in the sense of Vitali.
Now we shall use the induction method to determine the sequence <Ik : k = 1, 2, … n> of disjoint
intervals of M as follows:
2 LOVELY PROFESSIONAL UNIVERSITY

Let I1 be any interval in M and let  1 be the supremum (least upper bound of the lengths of the Notes
intervals in M disjoint from I 1 (i.e., which do not have any point common with I 1).
Obviously  1 < as  1 m (0) < .
1
Now we choose an interval I 2 from M, disjoint from I1, such that  (I 2 )  1 . Let  2 be the
2
supremums of lengths of all those intervals of M which do not have any point common with I 2
or I2 obviously  2 < .
In general, suppose we have already chosen r intervals I 1, I2, … Ir (mutually disjoint). Let  r be
the supremums of the length of those intervals of M which do not have any point in common
with I
i 1
i (i.e., which do not meet any of the intervals I 1, I2, … Ir. Then  r m (0) < .
r r
Now if E is contained in i 1
I i , then Lemma established. Suppose I
i 1
i E . Then we can find
1
interval Ir+1 s.t.  (I r 1 )  r which is disjoint from I 1, I2, … Ir.
2
Thus at some finite iteration either the Lemma will be established or we shall get an infinite
1
sequence <Ir> of disjoint intervals of M s.t.  (I r 1 )  r and  r < , n = 1, 2, 3 ….
2
Note that <  r > is a monotonically decreasing sequence of non-negative real numbers.
Obviously, we have that I

i 1
r 0
r 1
 (  r ) m (0) hence for any arbitrary > 0, we can
find an integer N s.t.
1
 (I r )
r N 1
5 .
Let a set F = I
r 1
r .
The lemma will be established if we show that m* (F) < . For, let x F, then x Ir 1
r x is an
N
element of E not belonging to the closed set I

r 1
r an interval I in M s.t. x I and  (I) is so
N
small that I does not meet the I

r 1
r , i.e.
I Ir = , r = 1, 2, … N.
Therefore we shall have  (I)  N 2 (I n 1 ) as by the method of construction we take

1
 In 1 N .
2

Notes It also I IN+1 = , we should have  I  N 1 . Further if the interval I does not meet any of the
intervals in the sequence <Ir>, we must have
 I r , r
which is not true as  r 2  (I r 1 ) 0 as r .
I must meet at least one of the intervals of the sequence <Ir>. Let p be the least integer s.t. I
meets Ip. Then p > N and  (I)  p 1 2  (I p ). Further let x I as well x Ip, then the distance of x
from the mid point of Ip is at most
1 1 5
 (I) ( p ) 2  (I p )  (I p )  (I p )
2 2 2
Thus if Ip is an interval having the same mid point as I p but length 5 times the length of Ip, i.e.
 (J p ) 5 (I p ) . Then x Jp also.
Thus for every x F, an integer p > N s.t. x Jp
and  (J p ) 5 (I p ) . Also
F J
p N 1
p
m * (F)  (J p ) 5  (I p ) 5
p N 1 p N 1
5
and hence the Lemma holds good.
1.1.4 Four Dini's Derivatives
The usual condition of differentiability of a function f (x) is too strong. Here we are studying the
functions under slightly weaker condition (measurability). So why we define four quantities,
called as Dini’s Derivatives, which may be defined even at the points where the function is not
differentiable.
f(x h) f (x)
1. D+ f (x) = Lim , called upper right derivative
n 0 h
f (x h) f (x)
2. D+ f (x) = Lim , called lower right derivative
h 0 h
f (x h) f (x)
3. D– f (x) = Lim ,
h 0 h
f (x h) f (x)
or Lim , called upper left derivative
h 0 h
f (x h) f (x)
4. D– f (x) = Lim
h 0 h
f (x h) f (x)
or Lim , called lower left derivative
h 0 h

Notes
Notes
1. D+f (x) D+ f (x) and D f (x) D f (x)
If D+ f (x) = D+ f (x), then we conclude that right hand derivative of f (x) exists at the
point x and denoted by f (x+). Similarly if D– f (x) = D– f (x), we say that f (x) is left
differentiable at x and denote this common value by f (x–).
2. The function is said to be differentiable at x if all the four Dini’s derivatives are equal
but different than , i.e. if
D+ f (x) = D+ f (x) = D– f (x) = D– f (x)
and their common value is denoted by f (x).
Properties of Dini's Derivatives
1. Dini’s derivatives always exist, may be finite or infinite for every function f.
2. D+ (f + g) D+ f + D+ g with similar properties for the other derivatives.
3. If f and g are continuous at a point ‘x’, then
D+ (f . g) (x) f (x) D+ g (x) + g (x) D+ f (x).
4. D+ f (x) = – D+ (– f (x))
and D– f (x) = – D– (– f (x)).
5. If f is a continuous function on [a, b] and one of its derivatives (say D +) is non-negative on
(a, b). Then f is non-decreasing on [a, b] i.e.
f (x) f (y) whenever x y, y [a, b].
6. If f is any function on an interval [a, b], then the four derivatives if exist are measurable.
1.1.5 Lebesgue Differentiation Theorem
Statement: Let f : [a, b] R be a finite valued monotonically increasing function, then f is

differentiable. Also f : [a, b] R is L-integrable and
b
f (x) dx f (b) f (a) .
a
Proof: Define a sequence <fn> of non-negative functions, where f n : [a, b] R such that,
1
fn (x) = n f x f (x) , x [a, b] … (1)
n
and set f (x) = f (b), for x b.
By hypothesis, f : [a, b] R is an increasing function, therefore f n : [a, b] R is also an increasing

function and hence integrable in the Lebesgue sense.
Again from (i) we have
f {x (1/n)} f (x)
Lim
Lim fn (x) = 1/n , x [a, b] ,
n 0 (1/n)
= f (x), a.e.

Notes Thus, the sequence <fn> converges to f (x), a.e.
Using Fatou’s Lemma, we have
b b
f (x) dx Lim inf fn (x) dx … (ii)
a n a
b b
1
Again Lim inf fn (x) dx = Lim inf n f x f(x) dx
n a n a n
b b
1
= Lim inf n f x dx f(x) dx
n a n a
Putting t = x + (1/n), we get
b
1 b (1/n ) b (1/n )
f x dx = f (t) dt f (x) dx
a n a (1/n ) a (1/n )
[By the first property of definite integrals]
b b (1/n ) b
Lim inf fn (x) dx = Lim inf n f (x) dx f (x) dx
n a n a (1/n ) a
b (1/n ) a (1/n )
= Lim inf n f (x) dx f (x) dx … (iii)
n b a
Now extend the definition of f by assuming
f (x) = f (b), x [b, b + 1/n].
b (1/n )
b (1/n )
1
f (x) dx = f (b) dx f (b)
b b n
1
Also f (a) f (x), for x a, a , therefore
n
a (1/n ) a (1 /n )
1
f (x) dx f (a) dx f (a)
a a n
a (1/n )
1
– f (x) dx f (a)
a n
b b (1/n ) a (1/n )
(iii) Lim inf fn (x) dx = Lim

n
inf n f(b) dx f(x) dx
n a b a
1 1
Lim inf n f (b) f (a) f (b) f (a)
n n n
Thus from (ii), we get
b
f (x) dx f (b) – f (a)
a
f (x) is integrable and hence finite a.e. thus f is differentiable a.e.

Notes
Example: Let f be a function defined by f (0) = 0 and f (x) = x sin (1/x) for x 0. Find
D+ f (0), D+ f (0), D– f (0), D– f (0).
1
f(0 h) f(o) h sin 0
D+ f (0) = Lim Lim h
h o h h o h
1 1
= Lim sin 1, as 1 sin 1
h o h h
f(0 h) f(0) 1
Also D+ f (0) = Lim Lim sin 1
h o h h o h
1
( h)sin 0
f(0 h) f(0) h
D– f (0) = Lim Lim
h o 0 h h o h
1
= Lim sin 1
h o h
f(0 h) f(0) 1
and D– f (0) = Lim Lim sin 1
h o h h o h
Theorem: Let x be a Lebesgue point of a function f (t); then the indefinite integral
x
F (x) = F (a) + f (t) dt
a
is differentiable at each point x and F (x) = f (x).
Proof: Given that x is a Lebesgue point of f (t), so that
x h
1
Lim f (t) f(x) dt = 0 … (i)
h o h x
x h x h
1 1 1
Now f (x) dt = f (x) 1 dt f (x) [t]xx h
h x h x h
1
= f (x).h f (x)
h
x h
1
Thus f (x) = f (x) dt … (ii)
h x
x h x
Also F (x + h) – F (x) = f (t) dt f (t) dt
a a
x x h x x h
= f (t) dt f (t) dt f (t)dt f (t) dt
a x a x
x h
F(x h) F(x) 1
= f (t)dt … (iii)
h h x

Notes From (ii) and (iii) we have
x h x h
F(x h) F(x) 1 1
f(x) = f (t)dt f (x)dt
h h x h x
x h x h
1 1
= f (t) f (x) dt f (t) f (x) dt
h x h x
x h
F(x h) F(x) 1
Lim f (x) Lim f (t) f (x) dt 0 [Using (i)]
h o h h o h x
F(x h) F(x)
or Lim f (x) 0 … (iv)
h o h
Since modulus of any quantity is always positive, therefore
F(x h) F(x)
Lim f (x) 0 … (v)
h o h
Combining (iv) and (v), we obtain
F(x h) F(x)
Lim f (x) = 0
h o h
F(x h) F(x)
Lim = f (x)
h o h
F (x) = f (x).
Theorem: Every point of continuity of an integrable function f (t) is a Lebesgue point of f (t).
Proof: Let f (t) be integrable over the closed interval [a, b] and let f (t) be continuous at the
point xo.
f (t) is continuous at t = xo implies that > 0, a > 0 such that,
|f (t) – f (xo) | < , whenever |t – xo| < .
xo h xo h
|f (t) f(x o )|dt dt h whenever|h| .
xo xo
xo h
1
|f (t) f(xo )|dt … (i)
h xo
Now h 0 0. So from (i), we have
xo h
1
Lim |f (t) f(x o )|dt 0 … (ii)
h o h xo
xo h
1
Now Lim |f (t) f(x o )|dt
h o h xo
xo h
1
Lim |f (t) f(xo )|dt 0 [Using (ii)]
h o h xo

xo h Notes
1
or Lim |f (t) f(x o )|dt 0 [ Modulus of any quantity is always non-negative]
h o h xo
xo h
1
i.e. Lim |f (t) f(x o )|dt 0.
h o h xo
This shows that xo is a Lebesgue point of f (t).
1.2 Summary
 A function f defined on [a, b] is said to satisfy Lipschitz condition if a constant M > 0 such
that
| f (x) – f (y) M |x – y|, x, y [a, b].
 A point x is said to be a Lebesgue point of the function f (t), if

x h
1
Lim f (t) f (x) dt 0 =0
h o h x
 Let E be a set of finite outer measure and M be a family of intervals which cover E in the
sense of Vitali; then for a given > 0, it is possible to find a finite family of disjoint
intervals
{Ik, k = 1, 2, …, n} of M, such that
m* E I
k 1
k .
 Lebesgue differentiation theorem: Let f : [a, b] R be a finite valued monotonically

increasing function, then f is differentiable. Also
f : [a, b] R is L-integrable and
f (x) dx f (b) f (a) .

a
1.3 Keywords
Dinni’s Derivatives: These are the ways to define the quantities to judge the
measurability of the functions even at the points where it is not differentiable.
Fundamental Theorem of the Integral: The fundamental theorem of the integral calculus is that
differentiation and integration are inverse processes.
Measurable functions: An extended real valued function f defined over a measurable set E is said
to be measurable in the sense of Lebesgue if set
E (f > a) = {x E : f (x) > a} is measurable for all extended real numbers a.
Vitali's Lemma: Let E be a set of finite outer measure and M be a family of intervals which cover
E in the sense of Vitali; then for a given > 0, it is possible to find a finite family of disjoint
intervals {Ik, k = 1, 2, … n} of M, such that
m* E I
k 1
k < .

Notes 1.4 Review Questions
1. If the function f assumes its maximum at c, show that D+ f (c) 0 and D– f (c) 0.
2. Give an example of functions such that D + (f + g) D+ f + D+g.
3. Find the four Dini’s derivatives of function f : [0, 1]  R
such that f (x) = 0, if x 0, if x Q and f (x) = 1, if x Q.
4. Evaluate the four Dini’s derivative at x = 0 of the function f (x) given below:
1 1
ax sin 2 bx cos2 , x 0
f (x) = x x
1 1
px sin 2 qx cos2 , x 0
x x
and f (0) = 0, given that a < b, p < q.
5. Every point of continuity of an integrable function f (t) is a Lebesgue point of f (t). Elucidate.
Books J. Yeh, Real Analysis: Theory of Measure and Integration

Bartle, Robert G. (1976). The Elements of Real Analysis (second edition ed.)
Online links www.solitaryroad.com/c756.html

www.public.iastate.edu/.../Royden_Real_Analysis_Solutions.pdf

Unit 2: Functions of Bounded Variation
Notes
CONTENTS
Objectives
Introduction
2.1 Functions of Bounded Variation
2.1.1 Absolute Continuous Function
2.1.2 Monotonic Function
2.1.3 Functions of Bounded Variation – Definition
2.1.4 Theorems and Solved Examples
2.2 Summary
2.3 Keywords
Objectives

 Define absolute continuous function.
 Define monotonic function.
 Understand functions of bounded variation.
 Solve problems on functions of bounded variation.
Introduction
Functions of bounded variation is a special class of functions with finite variation over an
interval. In Mathematical analysis, a function of bounded variation, also known as a BV function,
is a real-valued function whose total variation is bounded: the graph of a function having this
property is well behaved in a precise sense. Functions of bounded variation are precisely those
with respect to which one may find Riemann – Stieltjes integrals of all continuous functions.
In this unit, we will study about absolute continuous function, Monotonic function and functions
of bounded variation.
2.1 Functions of Bounded Variation
A real-valued function f defined on [a,b] is said to be absolutely continuous on [a,b], if for an

arbitrary 0 , however small, a, 0, such that
n n
f br – f ar , wherever br ar ,
r 1 r 1

Notes a1 b1 a2 b2 ... an bn i.e., a i ’s and b i ’s are forming finite collection

where
ai ,bi : i 1,2,...,n of pair-wise disjoint (non-overlapping) intervals (or of disjoint closed

intervals).
Obviously, every absolutely continuous function is continuous.
Notes
 If a function satisfies f br – f ar , even then it is absolutely continuous.
 The condition br ar , means that total length of all the intervals must be
r 1
less than .
2.1.2 Monotonic Function
Recall that a function f defined on an interval I is said to be monotonically non-increasing, iff
x y f x f y , x, y I
and monotonically non-decreasing, iff
x y f x f y , x, y I
Also f is said to be strictly decreasing, iff
x y f x f y
and strictly increasing, iff
x y f x f y
2.1.3 Functions of Bounded Variation – Definition
Let f be a real-valued function defined on [a,b] which is divided by means of points
a x0 x1 x2 ... xn b.
Then the set P x0 , x1 , x 2 ,..., x n is termed as subdivision or partition of [a,b].
b n 1 b b
Let us take V f,P f xr 1 f xr , and V f,P sup V f,P for all possible subdivisions P of
a a a
r 0
b
[a,b]. (This is called total variation of f over [a,b] and also denoted by Ta f ).

b Notes
If V
a
f is finite, then f is called a function of bounded variation or function of finite variation
over [a,b].
Set of all the functions of bounded variation on [a,b] is denoted by BV [a,b].
Notes
If f is defined on R, then we define
a
V f lin V f .
a a
Some important observations about the functions of bounded variations.
Let f: [a,b] R and P be any subdivision of [a,b]. Then:

x
(i) f x f a V f , x [a, b]
a
a
(ii) V f 0
a
b b
(iii) P1 P2 V f,P1 V f,P2 , where P1 and P2 are any two subdivisions of [a,b].
a a
b b
(iv) V f,P V f , for all subdivisions P of [a,b].
a a
(v) For each 0, however small, at least one subdivision P’ of [a,b] such that
b b
V f,P' Vf.
a a
b
(vi) V f 0.
a
b c
(vii) a b c V f V f .
a a
Theorem 1: A monotonic function on [a,b] is of bounded variation.
Proof: Divide the interval [a,b] by means of points
a x0 x1 x2 ... xn b.
without any loss of generality, we can take f(x) as increasing function on [a,b]. Since if f is a
decreasing function, –f is an increasing function and so by taking –f = g, we see that g is an
increasing function and so we are allowed to consider only increasing functions. Thus
xr xr 1 f xr f xr 1
f xr 1 – f xr 0
f xr 1 – f xr f xr 1 – f xr ...(i)

Notes n 1 n 1
Now V f xr 1 – f xr f xr 1 – f xr [usin g (i)]
r 0 r 0
V f xn f x0 f b f a .
Now f is monotonic f b and f a are finite quantities.
V a finite quantity independent of the mode subdivision. Hence f is of bounded variation.
Notes
If f is a monotonic function on [a,b], then

b
T f f b f a
a
Theorem 2: Let V, P, N denote total, positive and negative variations of a bounded function f on
[a,b]; then prove that
V = P+N, and P–N= f(b)–f(a).
Proof: Let the interval [a,b] be divided by means of points
a x0 x1 x2 ... xn b.
n 1
v f xr 1 – f xr
r 0
If P denotes the sum of those differences f xr 1 – f xr which are +n for positive and –n for
negative, then obviously,
v = p + n, f(b) – f(a) = p – n ...(i)
Let P supp, V sup v,N supn, ...(ii)
where suprema are taken over all subdivisions of [a,b]. From (i), we have
v = 2p + f(a) – f(b), ...(iii)
v = 2n + f(b) – f(a). ...(iv)
Taking supremum in (iii) and (iv) and using (ii), we get
V = 2P + f(a) – f(b), ...(v)
V = 2N + f(b) – f(a). ...(vi)
By adding and subtracting, (v) and (vi) give
V = P+N and f(b) – f(a) = P–N.
Theorem 3: If f1 and f2 are non-decreasing functions on [a,b], then f 1–f2 is of bounded variation on
[a,b].
Proof: Let f = f1 – f2 defined on [a,b].
Then for any partition P a x 0 , x 1 ,..., x n b , we have

Notes
f xi – f xi 1 f1 xi – f1 xi 1 f2 xi – f2 xi 1
f1 b – f1 a f2 b – f2 a
as f1 and f2 are monotonically increasing.

b
V f f1 b f2 b f2 a f1 a , which is a finite quantity.
a
b
V f and hence f is of bounded variation.
a
Theorem 4: If f BV [a,b] and c a, b , then f BV [a,c] and f BV [c,b] . Also

b c b
V f V f V f
a a c
Proof: Since f BV[a,b] and [a,c] [a,b] we get
c b
V f V f
a a
f BV [a,c] and similarly f BV [c,b] .
Now if P1 and P2 are any subdivisions of [a,c] and [c,b] respectively, then P P1 P2 is a
subdivision of [a,b].
c b b b
V f ,P1 V f ,P2 V f ,P V f .
a c a a
But P1 and P2 are any subdivisions. So taking supremums on P 1 and P2, we get
c b b
V f V f V f . ...(1)
a c a
Now let P a x0 ,x1 ,x2 ,...,xn b be a subdivision of [a,b] and c xr 1 , xr
P1 x0 ,x1 ,x2 ,...,xr 1 ,c and
P2 c,xr ,xr 1 ,xr 2 ,...,xn are the subdivisions of [a,c] and [c,b] respectively.
b r 1 n
Now Va f,P f xi f xi 1 f xr f xr 1 f xi f xi 1
i 1 i r 1
r 1 n
f xi f xi 1 f xr f c f c f xr 1 f xi f xi 1
i 1 i r 1
r 1 n
f xi f xi 1 f c f xr 1 f xr f c f xi f xi 1
i 1 i r 1
c b c b
V f,P1 V f,P2 V f V f
a c a c
b c b
(i) and (ii) V f V f V f . ...(ii)
a a c

Notes
Notes
 This theorem enables us to define a new function (called variation function) say
x
V x V f , x [a, b] .
a
y x y
 If x > y in [a,b], then V f V f V f .
a a x
y
i.e. v(y) = v(x) + V f .
x
v x is an increasing function.
 If a c1 c2 ... cn b,then
b c1 c2 b
V f V f V f ... V f
a a c c
1 n
Corollary:
f BV[a, b] f BV[a, c],
f BV[c, b] for each c [a, b].
Theorem 5: If a function f of bounded variation in [a,b] is continuous at c [a, b], then the function
x
defined by v(x) = Va f , is also continuous at x = c and vice versa.
Proof: Suppose f is continuous at x = c. Hence for arbitrary /2 0, we can find a 1

such that
a c 1 x c or x c 1 f x f c /2 ...(i)
Also we know by remark (v) after the definition (2.1.3), for above , we can get a subdivision
P= a x0 , x1 , x 2 ,..., x n c of [a,c]
c c
s.t. V f V f,P ...(ii)
a a 2
Now choosing positive min[ 1 , c x n 1 ], we get that for any x such that c x c , we also
have x n 1 x xn .
c n 1
(ii) V f f xr f xr 1 f xn f xn 1
a
r 1 2
n 1
f xr f xr 1 f xn f x f x f xn 1
r 1 2

n 1 Notes
f xr f xr 1 f x f xn 1 f xn f x
r 1 2
x
V f + f c f x
a 2
c x
V f V f , [by (i)]
a a
c x
0 V f V f ,
a a
x s.t. c x c, we have v c v x
lt v x = v c .
x c 0
v x is continuous on the left at x = c.
Similarly considering the partition of [c,b], one can show that v(x) is right continuous also at
x = c and hence v(x) is also continuous at x = c.
Converse of the above Theorem
If v(x) is continuous at x c [a, b] so is f also at x – c.
Proof: Since v(x) is continuous at x = c, for arbitrary small 0, a 0 such that
v x v c ,x c ,c ...(i)
Now let c x c . Then by Note (ii) of Theorem 4, we get
x c x
V f V f V f
a a c
x
v x v c V f
c
x
v x v c V f f x f c
c
f x f c v x v c [by (i)] ...(ii)
Similarly, we can show that f c f x ,if c x c. ...(iii)
(ii) and (iii) show that f(x) is also continuous at x = c.
Theorem 6: Let f and g be functions of bounded variation on [a,b] ; then prove that f+g, f-g, fg and
f/g g x 0, x and cf are functions of bounded variation, c being constant.
Proof:
(i) Set f + g = h, then
h xr 1 h xr f xr 1 g xr 1 f xr g xr
where a = x0 < x1< x2<...<xn= b

Notes
f xr 1 f xr g xr 1 g xr
f xr 1 f xr g xr 1 g xr .
n 1 n 1 n 1
h xr 1 h xr h xr 1 h xr g xr 1 g xr .
r 0 r 0 r 0
b b b
or V h V f V g .
a a a
Now by hypothesis, f, g are functions of bounded variations.
b b
V f and V g are finite.
a a
b
V h a finite quantity.
a
Hence h = f + g is of bounded variation in [a,b].
(ii) Let h = f – g. Then as above,
h xr 1 h xr f xr 1 f xr g xr 1 g xr .
b b b
V h V f V g
a a a
b
V h a finite quantity.
a
Hence h = f – g is of bounded variation in [a,b].
(iii) Let h(x) = f(x).g(x). Then
h xr 1 h xr f xr 1 .g xr 1 f xr .g xr
f xr 1 .g xr 1 f xr g xr 1 f xr g xr 1 f xr g xr
g xr 1 f xr 1 f xr f xr g xr 1 g xr .
g xr 1 f xr 1 f xr f xr . g xr 1 g xr .
Let A = sup f x : x [a,b] ,
B = sup g x : x [a,b] ,
h xr 1 h xr B. f xr 1 f xr A. g xr 1 g xr .
n 1 n 1 n 1
h xr 1 h xr B. h xr 1 h xr A g xr 1 g xr .
r 0 r 0 r 0

b b b Notes
i.e. V h B.V f A.V g .
a a a
= a finite quantity.
Hence h(x) = f(x).g(x) is of bounded variation in [a,b].
(iv) First, we shall show that 1/g is of bounded variation, where g x 0, x [a, b].
Now, g x 0, x [a, b]
1 1
0, x [a, b].
g x
Again, we observe that
1 1 g xr g xr 1 1
2
g xr g xr 1
g xr 1 g xr g x r .g x r 1
n 1
1 1 1 n 1
2
g xr g xr 1
r 0 g xr 1 g xr r 0
b 1 1 b
V 2 a
V g a finite quantity.
a g
1
Hence g is of bounded variation in [a,b].
1
Now f and g are of bounded variation in [a,b].
1
f. is of bounded variation in [a,b] [by case (iii)]
g
f
is of bounded variation in [a,b].
g
b b
(v) Do yourself. Note that V cf cV f .
a a
Notes
Since BV [a,b] is closed for all four algebraic operations, it is a linear space.
Theorem 7: Every absolutely continuous function f defined on [a,b] is of bounded variation.
Proof: Since f is absolutely continuous on [a,b]; for 1, a 0

Notes n
s.t. f bi f ai 1,
i 1
whenever bi ai ,
i 1
and a a 1 b1 a2 b2 ... a n bn b.
Now consider another subdivision of [a,b] or say refinement of P by adjoining some additional
points to P in such a way that all the intervals can be divided into r parts each of total length less
than .
Let the r sub-intervals be [c0,c1], [c1,c2],...,[cr-1,cr] such that
a = c0, cr = b and (ck+1–ck) < , K 0,1,2,...,(r 1)
Obviously, f xi 1 – f xi 1 1, where xi 1 , xi . [ck , ck 1 ]

i
ck 1
or V f 1, [Using (i)]
ck
b c c c
1 2 r
Hence V f V f V f ... V f 1 1 1 ... 1 r finite quantity.
a c0 c1 cr 1
Hence, f is of bounded variation.
Notes
Converse of above theorem is not necessarily true. These exists functions of bounded
variation but not absolutely continuous.
Theorem 8: Jordan Decomposition Theorem
A function f is of bounded variation, if and only if it can be expressed as a difference of two

monotonic functions both non-decreasing.
Proof: Let f be the function of f :[a,b] R.
Case I. f BV[a,b]. Then we can write
f = v – (v – f), ...(i)
so that f x v x v x f x , x [a,b].
Now if x, y [a, b] such that x < y, then by the remark (ii) of theorem 4, we get
y x y
V f V f V f .
a a x
y
v y v x V f 0
x
v x v y and hence v is a non-decreasing function on [a,b].

Again, if x<y in [a,b], then as above Notes
y
v y v x V f f y f x f y f x
x
v y f y v x f x v f y v f x
v f is also a non-decreasing function on [a,b].
Thus (i) shows that f is expressible as a difference of two monotonically non-decreasing functions.
Case II. Set g (x) and h (x) be increasing functions such that f(x) = g(x) – h(x).
Divide the closed interval [a,b] by means of points
a = x0 < x1< x2<...< xn=b.
n 1
Let V f xr 1 f xr
r 0
Now, we have that
f xr 1 f xr g xr 1 h xr 1 g xr h xr
g xr 1 g xr h xr h xr 1
g xr 1 g xr h xr h xr 1
g xr 1 g xr h xr 1 h xr
Now, g(x) and h(x) are monotonically increasing functions, so that g xr 1 g xr 0
and h xr 1 h xr 0
g xr 1 g xr g xr 1 g xr
and h xr 1 h xr h xr 1 h xr .
Hence f xr 1 f xr g xr 1 g xr h xr 1 h xr
n 1 n 1 n 1
f xr 1 f xr g xr 1 g xr h xr 1 h xr
r 0 r 0 r 0
n 1
Now g xr 1 g xr g x1 g x0 g x2 g x1 ..... .... g xn g xn 1

r 0
g xn g x0
g b g a  xn b, x 0 a
n 1
Similarly, h xr 1 h xr h b h a .
r 0
n 1
Hence f xr 1 f xr g b g a h b h a .
r 0

Notes
since f is finite in [a,b] Now g b ,g a h b , h a are finite numbers.
n 1
f xr 1 h xr
r 0
b
V f .
a
f is a function of bounded variation. Alternatively, since g (x) and h(x) are both non-decreasing,
so by theorem 3, g(x) – h(x) and hence f(x) is of bounded variation.
Corollary: A continuous function is of bounded variation iff it can be expressed are as a difference
of two continuous monotonically increasing functions. It follows from the results of Theorems
5 and 8.
Theorem 9: An indefinite integral is a function of bounded variation, i.e. if f L[a, b] and F x is
indefinite integral of f x i.e. F (x) = f t dt, then F BV[a, b]. Also show that
a
x
b
V f f.
a
a
Proof: Since f L[a, b], also f L[a, b].
Let P = xi : i 0,1, 2,..., n be a subdivision of the interval [a,b]. Then
xi xi 1
n n
F xi F xi 1 f f
r 0 i 1 a a
xi xi
n n
f f
i 1 xi 1 i 1 x
i 1
f .
a
b
b
f BV[a, b] and V f,p f.
a
a
Further above result is true for any subdivision of P of [a,b]. Therefore taking supremum, we get
b b
f f.
a a

Notes
Example: A function f of bounded variation on [a,b] is necessarily bounded on [a,b] but
not conversely.
x b
Solution: If x [a, b], then f x f a V f
a
V f
a
b b
V f f x f a V f
a a
b b
f a V f f x V f f a
a a
f x is bounded on [a,b]
For the converse, define the function f on [0,1] by
0,if x 0
f x
x.sin ,if 0 x 1
x
since 0 x 1 and 1 sin 1, the function f is obviously bounded. Now consider the
x
partition
2 2 2 2 2
P= 0, , ,..., , , ,1 of [0,1]
2n+1 2n–1 7 5 3
Where n N. Then we get
1 2 2 2 2
V f,P f f 0 ... f f f 1 f
0 2n 1 3 5 3
2 n 2 2 2
1 0 ... 1 .1 0 1
2n 1 3 5 3
2 2 2 2
...
2n 1 3 5 3
1 1 1
4. ... .
3 5 2n 1
1
But we know that series is divergent. Therefore letting n we get that
2n 1
1 1
V f lt V f,P
0 n 0
f is not of bounded variation.

Notes
Example: Show that the function
x sin if 0 x 2
f x x is continuous
0,if x 0
without being of bounded variation.
or
show that there exists a continuous function without being of bounded variation.
Solution: We know that xlt0 f x 0 f 0
f x is continuous but not of bounded variation (see converse of above example.)
Hence the result.
Problem: Show that if f exists and is bounded on [a, b], then f BV [a, b].
Solution: According to given, let |f | M on [a, b].
Then for any X i – 1, xi [a, b], we get
f(xi ) f(xi 1 )
M |f(xi ) f(xi 1 )| M(xi xi 1 )
xi xi 1
for any partition P of [a, b],
b
V(f) M (x i xi 1 ) M(b a)
a
f B V [a, b].
Problem: Show that the function f defined as
1
f(x) x p sin for 0 x 1, f(o) 0, p 2.
x
is of bounded variation [0, 1].
1
(0 h)p sin 0
h
Solution: Note that RF (0) = Lim
h o h
1
= Lim h (p 1)
sin 0
h o h
1
( h)p sin 0
h
and Lf (0) lim 0
h o h

Notes
1 p 1 p 1 1
f (0) = 0 and f (x) = x cos px sin
x x2 x
1 1
f (x) = xp–2 px sin cos , for 0 < x 1
x x
f (x) is bounded for 0 x 1.
According to above problem, f BV [0, 1].
2.2 Summary
 A real-valued function f defined on [a,b] is said to be absolutely continuous on [a,b], if for

an arbitrary 0, however small, a, 0,s.t.
n n
f br f ar whenever br ar ,
r 1 r 1
where a1 b1 a2 b2 ... an bn
 A function f defined on an interval I is said to be monotonically non-increasing, iff
x y f x f y , x, y I.
and monotonically non-decreasing, iff x > y f(x), f(y) x, g I.
n 1
b b b
 Let V f,P f xr 1 f xr , and V f Sup V f,P for all possible subdivisions P of
a a a
r 0
b
[a,b]. If V
a
f is finite, then f is called a function of bounded variation over [a,b].
2.3 Keywords
Absolute Continuous Function: A real valued function f defined on [a, b] is said to be absolutely
continuous on [a, b], if for an arbitrary > 0, however small, a, > 0, such that
n n
|f(br ) f(ar )| , wherever (br ar )

r 1 r 1
where a1 < b1 a2 < b2 … an < bn i.e. a1’s and b1’s are forming finite collection {(a i, bi) : i = 1, 2,
…, n} of pair-wise disjoint intervals.
Continuous: A continuous function is a function f : X Y where the pre-image of every open

set in Y is open in X.
Disjoint: Two sets A and B are said to be disjoint if they have no common element, i.e. A B .

Notes Monotonic Decreasing Function: A monotonic decreasing function is a function that either
decreases or remains the same, never increases i.e. a function f(x) such that f(x 2) f(x1) for x2 > x1.
Monotonic Function: A monotonic function is a function that is either a monotonic increasing or

monotonic decreasing.
Monotonic Increasing Function: A monotonic increasing function is a function that either

increases or remains the same, never decreases i.e. a function f(x) such that f(x 2) f(x1) for x2 > x1.
1. Show that sum and product of two functions of bounded variation is again a function of
bounded variation.
2. Show that the function f defined on [0,1] by

x
x cos for 0 x 1
f x 2
0 for x 0
is continuous but not of bounded variation on [0,1].
3. Show that the function f defined on [0,1] as f(x) = x sin for x > 0, f(0)=0 is continuous but
x
is not of bounded variation on [0,1].
4. Define a function of bounded variation on [a,b]. Show that every increasing function on
[a,b] is of bounded variation and every function of bounded variation on [a,b] is
differentiable on [a,b].
5. Show that a continuous function may not be of bounded variation.
6. Show that a function of bounded variation may not be continuous.
7. If f is a function such that its derivative f’ exists and is bounded. Then prove that the
function f is of bounded variation.
Books Halmos, Paul (1950), Measure Theory, Van Nostrand and Co.
Kolmogorov, Andrej N.; Fomin, Sergej V. (1969). Introductory Real Analysis, New
York: Dovers Publications.
Online links www.ams.org
www.whitman.edu/mathematics/SeniorProjectArchive/.../grady.pdf

Unit 3: Differentiation of an Integral
Unit 3: Differentiation of an Integral Notes
CONTENTS
Objectives
Introduction
3.1 Differentiation of an Integral
3.2 Summary
3.3 Keyword
Objectives

 Define differentiation of an integral.
 Solve problems related to it.
Introduction
If f is an integrable function on [a, b], we define its indefinite integral to be the function F defined
on [a, b] by
F (x) = f (t) dt
a
Here, it is shown that the derivative of the indefinite integral of an integrable function is equal
to the integrand almost everywhere. We begin by establishing some lemmas.
3.1 Differentiation of an Integral
If f is an integrable function on [a, b] then f is integrable on any interval [a, x] [a, b]. The
function F given by
F (x) = f (t) dt c ,
a
where c is a constant, called the indefinite integral of f.
Lemma 1: If f is integrable on [a, b] then the indefinite integral of f namely the function F on
x
[a, b] given by F (x) = f (t) is a continuous function of bounded variation on [a, b].
a
Proof: Let xo be any point of [a, b].

Notes x xo
Then F(x) F(xo ) = f (t) dt f (t) dt

a a
x a
= f (t) dt f (t) dt
a xo
a x
= f (t) dt f (t) dt
xo a
= f (t) dt
xo
|f (t)|dt
xo
But f is integrable on [a, b]
|f| is integrable on [a, b]
[Since we know that measurable function f is integrable over iff |f| is integrable over E]
Given > 0, > 0 such that for every measurable set A [a, b] with m (A) < , we have
|f| by theorem, “if f is a non-negative function which is integrable over a set E, then
A
given > 0, there is a > 0 such that for every set A E with m (A) < , f .”
A
|f (t)|dt < , for |x – xo| < .

xo
x x
|F(x) – F (xo)| = f (t) dt |f (t)|dt <

xo xo
whenever |x – xo| < .
|F(x) – F(xo)| < wherever |x – x o| <
F is continuous at xo and hence in [a, b].
Now we shall show that F is a function of bounded variation.
Let P = {a = xo < x1 < x2 < … < xn = b} be a partition of [a, b].
Then
n n xi
F(x i ) F(xi 1 ) = f (t) dt

i 1 i 1 xi 1

n xi Notes
|f (t)|dt
i 1 x
i 1
= |f (t)|dt
a
Tab (F) |f (t)|dt

a
But |f| is integrable therefore.
|f|dt
a
Tab (F) <
F BV [a, b]
Hence the Proof.
Theorem 1: Let f be an integrable on [a, b].
If f(t)dt 0 x [a, b] then f = 0 a.e. in [a, b].

a
Proof: Let if possible, f 0 a.e. in [a, b].
Let f (t) > 0 on a set E of positive measure, then there exists a closed set F E with m (F) > 0.
Let A = (a, b) – F.
Then A is an open set.
Now f (t) dt f (t) dt

a A F
But f (t) dt 0
a
f (t) dt 0
A F
f (t) dt f (t) dt 0
A F
f (t) dt f (t) dt 0 f (t) dt f (t) dt

A F A F

Notes But f (t) > 0 on F with m (F) > 0 implies
f (t) dt 0
F
Therefore f (t) dt 0
A
Now, A being as open set, it can be expressed as a union of countable collection {(a n, bn)} of
disjoint open intervals as we know that an open set can be expressed as a union of countable
collection of disjoint open intervals.
bn
Thus f (t) dt f (t) dt

A n an
But f (t) dt 0
A
bn
f (t) dt 0
n an
bn
f (t) dt 0 for some n

an
an
either f (t) dt 0
a
bn
Or f (t) dt 0
a
In either case, we see that if f is positive on a set of positive measure, then for some x [a, b] we
have
f (t) dt 0.
a
Similarly if f is negative on a set of positive measure we have
f (t) dt 0.
a
But it leads to the contradiction of the given hypothesis. Hence our supposition is wrong.
f = 0 a.e. in [a, b].
Hence the proof.

Theorem 2: First fundamental theorem of calculus statement: If f is bounded and measurable on Notes
x
[a, b] and F (x) = f (t) dt + F (a), then F (x) = f (x) a.e. in [a, b].
a
Proof: Since every indefinite integral is a function of bounded variation, therefore F (x) is a
function of bounded variation over [a, b]. Thus F (x) can be expressed as a difference of two
monotonic functions and since every monotonic function has a finite differential coefficient at
every point of a set of non-zero measure, therefore F (x) has a finite differential coefficient a.e. in
[a, b]. Now F is given to be bounded;
|f| M (say) … (1)
F(x h) F(x)
Let fn (x) = .
h
1
with h = .
x
1
Then |fn (x)| = (F(x h) F(x)
h
x h x
1
= f (t) dt f (t) dt
h
a a
x h a
1
= f (t) dt f (t) dt
h
a x
a x h
1
= f (t) dt f (t) dt
h
x a
x h
1
= f (t) dt
h
x
But |f| M
x h
M M
|fn(x)| dt (x h x)
h h
x
M
|fn(x)| (h)
h
|fn(x)| M
Since fn (x) F (x) a.e.,

Notes then the bounded convergence theorem implies that

x x
F (x) dx = lim Fx (x) dx

h
a a
x
1
= lim [F(x h) F(x)] dx
h 0 h
a
x h a h
1 1
= lim [F(x)dx F(x) dx
h 0 h h
x a
= F (x) – F (a)
x
= f (t) dt, by hypothesis

a
x x
 F(x) f (t) dt F(a) F(x) F(a) f (t) dt

a a
or [F (t) f(t)] dt 0, x
a
F (x) – f (x) = 0 a.e. in [a, b]
Hence F (x) = f(x) a.e. in [a, b] by the theorem, “If f is integrable on [a, b] and f (t) dt 0, x [a, b]
a
then f = 0 a.e. in [a, b]”.
Hence F (x) = f (x) a.e. in [a, b].
Hence the proof.
Theorem 3: If f is an integrable function on [a, b] and if F(x) = f (t) dt + F(a) then F (x) = f (x) a.e.
a
in [a, b].
Proof: Without loss of generality, we may assume that f (x) 0 x
Let us define a sequence {fn} of functions
fn : [a, b] R, where
f(x) if f(x) n,
fn(x) =
n if f(x) n
Clearly, each f n is bounded and measurable function and so, by the theorem,
x
Let f be a bounded and measurable function defined on [a, b]. If F(x) = f (t) dt + F(a), then F (x)
a
= f(x) a.e. in [a, b]”, we have
x
d
fn fn (x) a.e.
dx
a

Also, f – fn > 0 n, and therefore, the function G n defined by Notes
G n (x) (f fn )
a
is an increasing function of x, which must have a derivative almost everywhere by Lebesgue

theorem and clearly, this derivative must be non-negative.
Since Gn (x) = (f fn )
a
x x
= f (t ) dt fn (t ) dt
a a
x x
f (t ) dt = G n (x) fn (t ) dt
a a
Now the relation
F (x) = f (t ) dt F(a) becomes

a
F (x) = G n (x) fn (t ) dt F(a) ,

a
F (x) = Gn (x) fn (x) a.e.
fn(x) a.e. n.
since n is arbitrary, we have
F (x) f(x) a.e.
b b
F (x) dx f (x) dx … (1)

a a
Also by the Lebesgue’s theorem, i.e. “Let f be an increasing real-valued function defined on
[a, b].
Then f is differentiable a.e. and the derivative f is measurable.
and f (x) dx f (b) – f (a)”, we have

a
F (x) dx F (b) – F (a) … (2)

a

Notes b
But F (x) = f (t) dt F(a)

a
F(b) – F(a) = f (x) dx

a
Therefore (2) becomes

b b
F (x) dx f(x) dx … (3)

a a
From (1) and (3), we get

b b
F (x) dx = f(x) dx
a a
b b
F (x) dx f(x) dx = 0
a a
F (x) f(x) dx = 0
a
since F (x) – f(x) 0 a.e., which gives that
F (x) – f(x) = 0 a.e. and
so F (x) = f(x) a.e.
3.2 Summary
 If f is an integrable function on [a, b] then f is integrable on any interval [a, x] [a, b]. The
function F given by
F (x) = f (t) dt c ,
a
where c is a constant, called the indefinite integral of F.
 Let f be an integrable on [a, b]. If f (t) dt 0 x [a, b] then f = 0 a.e. in [a, b].
a
3.3 Keyword
Differentiation of an Integral: If f is an integrable function on [a, b] then f is integrable on any

interval [a, x] [a, b]. The function F given by
F (x) = f (t) dt c ,
a
where c is a constant, called the indefinite integral of f.

3.4 Review Questions Notes
1. If f is an integrable function on [a, b] and if F (x) = f (t) dt F(a) then check whether
a
F (x) = f (x) is absolute continuous function in [a, b] or not.
2. If F is an absolutely continuous function on [a, b], then prove that F (x) = f (t) dt C where
a
f = F a.e. on [a, b] and C is constant.
Books Flanders, Harley. Differentiation under the Integral Sign

Frederick S. Woods, Advanced Calculus, Ginn and Company
David V. Widder, Advanced Calculus, Dover Publications Inc., New Edition
(Jul 1990).
Online links www.physicsforums.com > Mathematics > Calculus & Analysis

www.sp.phy.cam.ac.uk/~ alt 36/partial diff.pdf

Notes
Unit 4: Absolute Continuity
CONTENTS
Objectives
Introduction
4.1 Absolute Continuity
4.2 Summary
4.3 Keywords
Objectives
 Define Absolute Continuous function.
 Solve problems on absolute continuity
 Understand the proofs of related theorems.
Introduction
It may happen that a continuous function f is differentiable almost everywhere on [0,1], its
derivative f’ is Lebesgue integrable, and nevertheless the integral of f’ differs from the increment
of f. For example, this happens for the Cantor function, which means that this function is not
absolutely continuous. Absolute continuity of functions is a smoothness property which is
stricter than continuity and uniform continuity.
4.1 Absolute Continuity
A real-valued function f defined on [a,b] is said to be absolutely continuous on [a,b], if for an

arbitrary 0, however small, a, 0, such that
n n
r 1 r 1
where a1 b1 a2 b2 ... an bn i.e. a i ’s and b i ’s are forming finite collection
ai ,bi : i 1,2,...,n of pair-wise disjoint intervals.
Obviously, every absolutely continuous function is continuous.

Notes
Notes
1. If a function satisfied f br f ar , even then it is absolutely continuous.
2. The condition br ar means that total length of all the intervals must be less
r 1
than .
Theorem 1: Every absolutely continuous function f defined on [a,b] is of bounded variation.
Proof: Since f is absolutely continuous on [a,b]; for 1, a 0 such that
f bi f ai 1,
r 1
whenever bi ai ,
r 1
and a a1 b1 a2 b2 ... an bn b.
Now consider another subdivision of [a,b] or say refinement of P by adjoining some additional
points to P in such a way that all the intervals can be divided into r parts each of total length less
than .
Let the r-sub-intervals be c0 , c1 , c1 , c 2 ,..., cr 1 , cr such that
a c0 , cr b and c k+1 ck , k 0,1, 2,..., r 1
Obviously, f xi 1 f xi 1,
i
where xi 1 , xi ck , ck 1
c
or kV1 f 1,
ck
b c1 c2 cr
Hence V f V f V f ... V f 1 1 ... 1 r finite quantity.
a c0 c1 cr 1
Hence f is of bounded variation.

Notes
Notes
Converse of above theorem is not necessarily true. There exists functions of bounded
variation but not absolutely continuous.
Theorem 2: Let f(x) and g(x) be absolutely continuous functions, then prove that f x g x and
f x
f x .g x are also absolutely continuous functions. Hence show that if g x 0, x is also
g x
absolutely continuous function.
Proof: Given f x and g x are absolutely continuous functions on the closed interval [a,b],
therefore for each 0 , there exists 0 such that
f br f ar and
r 1
g br g ar ,
r 1
whenever br ar , for all the points a1 ,b1 ,a 2 ,b2 ,...,an ,bn such that
r 1
a1 b1 a2 b2 ... an bn .
n n n
(i) We have, f br g br f ar g ar f br f ar g br g ar .
r 1 r 1 r 1
Now if br ar , then
r 1
n n
f br f ar and g br g ar .
r 1
2 r 1
2
f br g br f ar g ar + = ,
r 1
2 2
whenever br ar .
r 1
This show that f x g x are also absolutely continuous functions over [a,b].

n
Notes
(ii) We have f br g br f ar g ar
r 1
f br g br f br g ar f br g ar f ar g ar
r 1
f br g br g ar g a r f br f ar
r 1
n n
f br g br g ar g a r f br f ar
r 1 r 1
n n
f br g br g ar g a r f br f ar .
r 1 r 1
Now every absolutely continuous function is bounded therefore f(x) and g(x) are bounded
in the closed interval [a,b].
Let f x K1 , g x K 2 , x [a, b].
Then we have
f br g br f ar g ar K1 K2 K1 K2 ,
r 1
Whenever br ar .
r 1
Setting K1 K2 *,
We have f br g br f ar g ar = *,
r 1
Whenever br ar ,
r 1
where a1 b1 a2 b2 ... an bn ;
Product of two absolutely continuous functions is also absolutely continuous.
(iii) We have g x 0 x [a, b] ; therefore
g x , where 0, x [a, b].

Notes
n n
1 1 g a r g br
Now, = < 2
,
r 1
g br g ar r 1
g br g a r
Whenever br ar . Setting 2
= *, we get
r 1
n
1 1
< *.
r 1
g br g ar
This show that 1 is absolutely continuous function over [a,b].

g x
Now f(x), 1 are absolutely continuous.

g x
1 is absolutely continuous.
f x .
g x
f x is also absolutely continuous over [a,b].

g x
Hence the theorem is true.
Note
By Theorem 1, its remark and above theorem it follows that set of all absolutely continuous
functions on [a,b] is a proper subspace of the space BV [a,b] of all functions of bounded
variation on [a,b].
Theorem 3: If BV[a, b], then f is absolutely continuous on [a,b], iff the variation function
x
v x V f is absolutely continuous on [a,b].
a
Proof: Case I: Given v(x) is absolutely continuous.
For arbitrary >0, 0 s.t.
n n
v br v a r < , whenever br ar .
r 1 r 1
x
Also, we know that f x f a V f
a
v x

n n
Notes
f br f ar = f br f a f a f ar
r 1 r 1
f br f a f a f ar
r 1
Now taking supremum over all collections of Pi of [ai, bi] for i = 2,...,n, we get
n bi
V f .
ai
r 1
bi ai bi
But V f V f V f
a a ai
bi bi ai
V f V f V f
a a a
i
b
i
V f v bi v ai
ai
v bi v ai <
i 1
v x is absolutely continuous.
Theorem 4: A necessary and sufficient condition that a function should be an indefinite integral
is that it should be absolutely continuous.
Proof: Condition is sufficient.
Let f(x) be an absolutely continuous function over the closed interval [a,b].
Therefore f is of bounded variation and hence we can express f(x) as
f(x) = f1(x) – f2(x)
where f1(x) and f2(x) are monotonically increasing functions and hence both are differentiable.
n n
v br v ar < whenever br ar
r 1 r 1
f is also absolutely continuous on [a,b].

Case II: Given f is absolutely continuous on [a,b].
for a given 0, a 0 s.t.
f bi f ai < , ...(i)
i 1

Notes
for every finite collection P ai ,bi , i 1,2,...,n of pairwise disjoint sub-intervals of [a,b] such
n
that bi ai .
i 1
Now, let Pi xik 1 , bxik i ,k 1,...,m i be a finite collection of non-overlapping intervals of the
interval [ai,bi].
Then the collection x ki 1 , bx ki i : i 1, 2,..., n , k 1,..., mi is a finite collection of non-overlapping
sub-intervals of [a,b] such that
n m n
i
xik xik 1 bi ai .
i 1 k 1 i 1
n m
i
and hence by (i), f xik f xik 1 .

i 1 k 1
Hence f’(x) exists and f ' x f1' x f2' x

b
f x f1 b f2 b f1 a f2 a ,
a
f x is integrable also.
Now let F(x) be an definite integral of f’(x) i.e.
F(x) = F(a) + f t dt, x [a, b] ...(ii)

a
Using fundamental theorem of integral calculus,
We get
F’(x) = f’(x)
or F(x) = f(x) + constant (say c) ...(iii)
From (ii), we have F(a) = f(a),
Using this in (iii), we get c = 0 and hence F (x) = f(x).
Thus every absolutely continuous function f(x) is an indefinite integral of its own derivative.
Condition is necessary: Let f(x) be an indefinite integral of f(x) defined on the closed interval
[a,b], so that
F x f t dt f a , x [a,b] and f(x) is integrable over [a,b].

a
Corresponding to arbitrary small 0, let >0 be such that if m(A) ,then f ,

A

Now select 2n real numbers such that Notes
ai bi a2 b2 a3 ... an bn
n
n
such that A = U a i , bi and bi ai .
i 1
i 1
n n bi ai
Then F bi F ai f f
i 1 i 1 a a
n bi n bi
F f f .
i 1 ai i 1 ai A
Thus, we have shown that for arbitrary small 0, a 0 s.t. bi ai .

i 1
F bi F ai .
i 1
F is absolutely continuous.
Thus every indefinite integral is absolutely continuous.
Theorem 5: If a function f is absolutely continuous in an interval [a,b] and if f’(x) = 0. a.e. in [a,b],
then f is constant.
Proof: Let c [a, b] be arbitrary. If we show that f(c) = f(a), then the theorem will be proved.
Let E = x a, c : f '(x) 0 .
since c is arbitrary, therefore set E a, c . This implies any x E f' x 0.
Let , > 0 arbitrary. Now f ' x 0, x E an arbitrary small interval x,x h a,c
f x h f x
such that f x h f x h.
h
This implies that corresponding to every x E, an arbitrary small closed interval x,x h
contained in [a,c] s.t.
f x h f x h.
Thus the interval x,x h , x E, over E in Vitali’s sense. Thus by Vitali’s Lemma, we can
determine a finite number of non-overlapping intervals I k, where
Ik xk , y k k 1,2,3,...,n
such that this collection covers all of E except for a set of measure less than 0 where is pre-
assigned number which corresponds to occurring in the definition of absolute continuity of f.

Notes Suppose xk xk 1 ; then adjoining the points y 0, xn+1.
We have a y0 x1 y1 x2 y2 ... xn yn xn 1 c.
Now since f is absolutely continuous, therefore for above subdivision of [a,c], we have
n n
f xk 1 f yk , whenever xk 1 yk .
k 0 k 0
n n
(i) f yk f xk n yk xk n c a .
k 1 k 1
n n
Now f c f a f xk 1 f yk f yk f xk
k 0 k 1
n n
f xk 1 f yk f yk f xk
k 0 k 1
n c a
But , n and hence n c a are arbitrary small positive numbers. So letting 0, n 0
We get f(c) = f(a)
f x is a constant function.
Corollary: If the derivatives of two absolutely continuous functions are equivalent, then the
functions differ by a constant.
Proof: Let f and g be two absolutely continuous functions and f’ = g’ f g' 0 by above
theorem f – g = constant and hence the result.
Example: If f is an absolutely continuous monotone function on [a,b] and E a set of

measure zero, then show that f (E) has measure zero.
Proof: Let the function f be monotonically increasing. By the definition of absolute continuity of
f, for 0, 0 and non-overlapping intervals I n an ,bn such that
bn an f bn f an
or f bn f an
Now, E [a, b] E I n
f E f I n f In
m* f E m * f In f xn f xn ,

Notes
where f x n and f x n are the maximum and maximum values of f(x) in the interval [a n,bn].
Also note that xn xn bn a n
m* f E , being arbitrary.
m* f E 0 m f E 0.
Example: Give an example which is continuous but not absolutely continuous.
Solution: Consider the function f : F R, where F is the Cantor’s ternary set.
xk
Let x F x x1 x 2 x 3 ... , xk 0 or 2
K 1
3k
rk 1
Define f x , where rk xk .
K 1
2k 2
= 0. r 1, r2, r3.....
This function is continuous but not absolutely continuous.
(i) Note that this function is constant on each interval contained in the complement of the
Cantor’s ternary set.
For, let (a,b) be one of the countable open intervals contained in F c. Then in ternary
notation,
a = 0.a1a2...an–1 0 2 2 2
and b = 0.a1a2...an-1 2 0 0 0,
where ai = 0 or 2, for i n 1.
ai
f a 0.r1 ,r2 ,...,rn 1 0 1 1 1 1 ...,where ri ,
2
f b 0.r1 ,r2 ,...,rn 1 1 0 0 0 0 ...
But in binary notation
0.r1 ,r2 ,...,rn 1 0 1 1 1 1 ... 0.r1 ,r2 ,...,rn 1 1 0 0 0 0 ...
f a f b.
Thus, we extend the function f overall of the set [0,1] instead of F by defining
f x f b , x a, b F c . Thus, the Cantor’s function is defined over [0,1] and maps it
onto [0,1].
It is clearly a non-decreasing function.

Notes (ii) To show that f(x) is a continuous function. Note that if c', c'' F, then we have
c' 0. 2p1 2p 2 2p 3 ...

each pi ,q i 0 or =1
c'' 0. 2q 1 2q 2 2q 3 ...
1
If c' c'' , then pi = qi, for 1 i n 1 and hence
3n
1
f c' f c'' ...(i)
2n
as n , c' c'', f c' f c'' ,
Hence if c0 F and c n is a sequence in F such that cn c0 ,when n , then
f cn f c0 , when n .
Now let x 0 [0,1] and let x n be a sequence in [0,1] such that x n x0 as n .
Case I: Let x0 F x0 I,say a, b Fc
xn I and hence f xn f x f a
and hence f x n f x0 as n .
Case II: Let x0 F. Now for each n such that x n F,set x n c n and hence f xn f x0 .
If x n F, then an open int erval I Fc.
(i) if x n x 0 , then set cn as the upper end point of I.
(ii) If x 0 x n , then set cn as the lower end point of I.
in any case f xn f x0 as n .
But the sequence x n was any sequence satisfying the stated conditions.
f is a continuous function.
(iii) To show f(x) is not absolutely continuous. Note that f’(x) = 0 at each x Fc .
f ' x exists and is zero on [0,1] and is summable on [0,1].
We know that for f(x) to be absolutely continuous, we must have
f x f ' x dx f 0 .
0
Particularly, we must have
f 1 f 0 f ' x dx.
0

x Notes
But f(1) – f(0) = 1 and f ' 1 dx 0 as f ' x 0
0
f 1 f 0 f ' 1 dx 0 as f ' x 0
0
f x is not absolutely continuous.
Theorem 6: Prove that an absolutely continuous function on [a, b] is an indefinite integral.
Proof: Let f(x) be an absolutely continuous function in a closed interval [a, b] so that f (x) &
f (t) dt exists finitely x [a, b].

a
Let F(x) be an indefinite integral of f (x), so that
F(x) = f(a) f (t) dt, x [a, b] … (1)

a
We shall prove that F(x) = f(x).
Since an indefinite integral is an absolutely continuous function.
Therefore F(x) is absolutely continuous in [a, b].
Then from (1),
F (x) = f (x) a.e.
d
[F(x) f(x)] = 0.
dx
Integrating, we get
F(x) – f(x) = c (constant) … (2)
Taking x = a in (1), we get
F(a) = f(a) f (t) dt

a
F(a) – f(a) = 0
or F(x) – f(x) = 0 for x = a
Then from (2), we get c = 0.
Thus (2) reduces to
F(x) – f(x) = 0 a.e.
F(x) = f(x) a.e.
which shows that f(x) is indefinite integral of its own derivative.

Notes 4.2 Summary
 A real-valued function f defined on [a,b] is said to be absolutely continuous on [a,b], if for

an arbitrary 0, however small, a, 0, such that
n n
f br f ar , whenever br ar .
r 1 r 1
 Every absolutely continuous function is continuous.
 Every absolutely continuous function f defined on [a,b] is of bounded variation.
4.3 Keywords
Absolute Continuity of Functions: Absolute continuity of functions is a smoothness property

which is stricter than continuity and uniform continuity.
Absolute Continuous Function: A real-valued function f defined on [a,b] is said to be absolutely

continuous on [a,b], if for an arbitrary 0, however small, a, 0, such that
n n
r 1 r 1
where a1 b1 a2 b2 ... an bn i.e. a i ’s and b i ’s are forming finite collection
ai ,bi : i 1,2,...,n of pair-wise disjoint intervals.
1. Define absolute continuity for a real variable. Show that f(x) is an indefinite integral, if F
is absolutely continuous.
2. If f,g: [0,1] R are absolutely continuous, prove that f + g and fg are also absolutely
continuous.
3. Show that the set of all absolutely continuous functions on an interval I is a linear space.
4. If g is a non-decreasing absolutely continuous function on [a,b] and f is absolutely continuous

on [g(a), g(b)], show that fog is also absolutely continuous on [a,b].
5. If f is absolutely continuous on [a,b] and f ' x 0 for almost all x [a, b], show that f is
non-decreasing on [a,b].
Books Krishna B Athreya, N Soumendra Lahiri, Measure Theory and Probability Theory,
Springer (2006).

Aole Nielsen, An Introduction to Integration and Measure Theory, Wiley-Interscience, Notes

(1997).
N.L. Royden, Real Analysis (third ed.), Collier MacMillan, (1988).
Online links dl.acm.org
mrich.maths.org

Notes Unit 5: Spaces, Hölder
CONTENTS
Objectives
Introduction
5.1 Spaces, Hölder
5.1.1 Lp-spaces
5.1.2 Conjugate Numbers
5.1.3 Norm of an Element of Lp-space
5.1.4 Simple Version of Hölder's Inequality
5.1.5 Hölder's Inequality
5.1.6 Riesz-Hölder's Inequality

5.1.7 Riesz-Hölder's Inequality for 0 < p < 1
5.2 Summary
5.3 Keywords
Objectives
 Understand Lp-spaces, conjugate numbers and norm of an element of L p-space
 Understand the proof of Hölder’s inequality.
Introduction
In this unit, we discuss an important construction, which is extremely useful in virtually all
branches of analysis. We shall study about Lp-spaces and Hölder’s inequality.
5.1 Spaces, Hölder
5.1.1 L P-Spaces
The class of all measurable functions f (x) is known as L p-spaces over [a, b], if Lebesgue –
integrable over [a, b] for each p exists, 0 < p < , i.e.
|f|p dx , (p 0)
a
and is denoted by L p [a, b].

Unit 5: Spaces, Hölder
Notes
Note The symbol Lp is used for such classes when limits of integration are known and
mentioning of interval is not necessary.
5.1.2 Conjugate Numbers
1 1
Let p, q be any two n on-negative extended real numbers s.t. 1 , then p, q are called
p q
(mutually) conjugate numbers.
Obviously, 2 is self-conjugate number.
Also if p 2, then q 2. Further, if p = , then q = 1 1, are conjugate numbers.
Note Non-negativity p 1, q 1.
5.1.3 Norm of an Element of LP-space
The p-norm of any f Lp [a, b], denoted by f p

, is defined as
1
b p
f p
= |f|p ,0<p< .
a
Theorem 1: If f Lp [a, b] and g f, then g Lp [a, b].
Proof: Let be any positive real number.
{x [a, b] : g (x) > } = {x [a, b] : < g (x) f (x)} ( g f)
= {x [a, b] : f (x) > }

p
Again f L [a, b]
f is measurable over [a, b].
{x [a, b] : f (x) > } is a measurable set.
{x [a, b] : g (x) > } is a measurable set.

g is a measurable function over [a, b]
Again since g (x) f (x), x [a, b]
b b
|g|p dx |f|p dx ( |f|p L [a, b])

a a
or |g|p dx
a
Thus |g|p L [a, b].

Notes Thus we have proved that g is a measurable function over [a, b] such that
|g|p L [a, b]
Hence g Lp [a, b]
Theorem 2: If f Lp [a, b], p > 1, then f L [a, b]
Proof: f Lp [a, b] f is measurable over [a, b]
Let A1 = {x [a, b] : |f (x) 1}
and A2 = {x [a, b] : |f (x) 1}
Then [a, b] = A1 A2 and A1 A2 =
Using countable additive property of the integrals, we have

b
|f|dx = |f|dx |f|dx … (i)

a A1 A2
Now  |f (x)| 1, x A1
|f| |f|p on A1 as p > 1
|f|dx |f|p dx as f Lp [a, b] … (ii)

A1 A1
Now |f (x) |<|, x A2
Using first mean value theorem, we get
|f|dx m (A 2 ) = A finite quantity … (iii)

A2
Combining (ii) and (iii) and making use of (i), we get
|f|dx <
a
Thus f is a measurable function over [a, b], such that
|f|dx <
a
|f| L [a, b] and hence f L [a, b].

p p
Theorem 3: If f L [a, b], g L [a, b]; then f + g Lp [a, b]
Proof: Since f, g Lp [a, b] f, g are measurable over [a, b]
f + g is measurable over [a, b]
Let A1 = {x [a, b] : |f (x)| |g (x)|}
and A2 = {x [a, b] : |f (x)| < |g (x)|}
Then [a, b] = A1 A2 and A1 A2 =

Notes
Therefore |f g|p dx |f g|p |f g|p dx .
A1 A2
Again, |f + g|p (|f| + |g|)p (|g| + |g|)p on A2 and (|f| + |f|)p on A1
2P |g|p on A2 and 2p |f|p on A1
Integrating, we have
|f g|p 2 p |f|p
A1 A1
and |f g|p 2 p |g|p

A2 A2
Since f, g Lp [a, b] |f|p and |g|p

A1 A2
 |f g|p and |f g|p [by (i) and (ii)]

A1 A1
|f g|p dx f+g Lp [a, b]

a
5.1.4 Simple Version of Hölder's Inequality
1 1
Lemma 1: Let p, q > 1 be such that 1 , and let u and v be two non-negative numbers, at
p q
least one being non-zero. Then the function f : [0, 1] R defined by
1
q q
f (t) = ut + v(1 t ) , t [0, 1],
has a unique maximum point at
1
up q
s = … (1)
up vp
The maximum value of f is

1
p
max f (t) = (u vp )p … (2)
t [0 , 1]
Proof: If v = 0, then f (t) = tu, t [0, 1] (with u > 0), and in this case, the Lemma is trivial.
Likewise, if u = 0, then
1
f (t) = v(1 t q ) q , t [0, 1] (with v > 0), … (3)

Notes and using the inequality

1
(1 t q ) q 1, t (0, 1],
We immediately get f (t) < f (0), t (0, 1],
and the Lemma again follows.
For the remainder of the proof we are going to assume that u, v > 0.
Obviously f is differentiable on (0, 1) and the solutions of the equation (3)
f (t) = 0
Let s be defined as in (1), so under the assumption that u, v > 0, we clearly have 0 < s < 1.
We are going to prove first that s is the unique solution in (0, 1) of the equation (3).
We have
1
1 1
f (t) = u v (1 t q ) q q tq 1
… (4)
q
tq
= u v , t (0, 1)
1 tq
so the equation (3) reads
tq
u v = 0.
1 tq
Equivalently, we have
1
tq p
= u/v,
1 tq
tq
= (u/v)p,
1 tq
(u /v)p up
t2 = ,
1 (u /v)p up vp
Having shown that the “candidates” for the maximum point are 0, 1 and s let us show that s is the
only maximum point.
For this purpose, we go back to (4) and we observe that f is also continuous on (0, 1).
Since lim f (t) = u > 0 and

t 0
lim f (t) = –
t 1
and the equation (3) has exactly one solution in (0, 1), namely s, this forces
f (t) > 0 t (0, s)

and f (t) < 0 t (s, 1). Notes
This means that, f is increasing on [0, s] and decreasing on [s, 1], and we are done.
The maximum value of f is then given by
max f (t) = f (s),

t [0 , 1]
and the fact that f (s) equals the value in (2) follows from an easy computation.
5.1.5 Hölder's Inequality
Statement: Let a1, a2, …, an, b1, b2, …, bn be non-negative numbers. Let p, q > 1 be real number
1 1
with the property 1 . Then
p q
1 1
n
n p n q
a jb j a p
b q
… (5)
j j
j 1
j 1 j 1
Moreover, one has equality only when the sequences ap1 , , apn and bq , , bqn are proportional.
Proof: The proof will be carried on by induction on n. The case n = 1 is trivial.
Case n = 2.
Assume (b1, b2) (0, 0). (otherwise everything is trivial).
Define the number
b1
r = 1 .
bq1 bq2 q
Notice that r [0, 1] and we have
b2 1
1 = 1 rq q
bq1 b q q
2
1
Notice also that, upon dividing by bq1 bq2 q , the desired inequality
1 1
a1 b1 + a2 b2 a q1 a q2 p b1q bq2 q … (6)
reads
1 1
a 1 r + a2 1 r q q a q1 a q2 p … (7)
It is obvious that this is an equality when a 1 = a2 = 0. Assume (a1, a2) (0, 0), and set up the
function.
1
q
f (t) = a 1 t a 2 1 t q
,t [0, 1].

Notes We now apply Lemma (1) stated above, which immediately gives us (7).
Let us examine when equality holds.
If a1 = a2 = 0, the equality obviously holds, and in this case (a 1, a2) is clearly proportional to (b 1,
b2). Assume (a1, a2) (0, 0).
Again by Lemma (1), we know that equality holds in (7), exactly when
1
a p1 q
r =
a p1 a p2
1
b1 a p1 q
that is 1 = ,
bq1 bq2 q a p1 a p2
or equivalently
bq1 a p1
= .
bq1 bq2 a p1 a p2
Obviously this forces
bq1 a q1
q q = p ,
b 1 b 2 a 1 a p2
so indeed a p1 , a p2 and bq1 , bq2 are proportional.
Having proven the case n = 2, we now proceed with the proof of:
The implication: Case n = k case n = k + 1, start with two sequences (a 1, a2, …, ak, ak+1) and
(b1, b2, …, ak, bk+1).
Define the numbers

1 1
k p k q
p q
a = a j
and b = b j
.
j 1 j 1
Using the assumption that the case n = k holds, we have
1 1
k 1 k p k q
a jb j a pj bqj + ak+1 bk+1
j 1 j 1 j 1
= ab + ak+1 bk+1 … (8)
Using the case n = 2, we also have
1 1
ab + ak+1 bk+1 ap a kp 1
p bq bkq 1
q
1 1
k 1 p k 1 q
= a pj bqj , … (9)
j 1 j 1
so combining with (8) we see that the desired inequality (5) holds for n = k + 1.

Assume now we have equality. Then we must have equality in both (8) and in (9). Notes
On one hand, the equality in (8) forces ap1 , ap2 , , apk and bq1 , bq2 , , bkq to be proportional (since
we assume the case n = k). On the other hand, the equality in (9) forces ap , akp 1
and bq , bqk 1 to
be proportional (by the case m = 2). Since
k k
ap a pj and bq bqj ,
j 1 j 1
it is clear that ap1 , ap2 , , apk , apk 1

and bq1 , bq2 , , bkq , bqk 1
are proportional.
5.1.6 Riesz-Hölder's Inequality
Statement: Let p and q be conjugate indices or exponents (numbers) and f Lp [a, b], g Lq [a, b];
then show that
(i) f g L[a, b]
(ii) fg f p
g q
i.e.
1 1
p p q q
|fg| |f| |g|
with equality only when |f|p = |g|q a.e. for some non-zero constants and .
Lemma: If A and B are any two non-negative real numbers and 0 < < 1, then
A B1– A + (1 – ) B, with equality when A = B.
Proof: If either A = 0 or B = 0, then the result is trivial.
Let A > 0, B > 0
Consider the function
(x) = x – x, where 0 x< and 0 < <1
d 1 d2 1
x and ( 1)x .
dx dx 2
d
Now solving = 0, we get x = 1.
dx
d2
Also at x = 1, < 0 as 0 < < 1.
dx2
By calculus, (x) is maximum at x = 1, so
(x) (1) i.e. x – x l – . … (1)
A
Now, putting x = , we get
B
A A A
1 or A B 1
B B B

Notes Or A B1– – A (1 – ) B or
A B1– A + B (1 – ) … (2)
Obviously equality holds good only for x = 1, i.e. only when A = B.
Proof of Theorem
Note that when p = 1, q = , the proof of theorem is obvious. Let us assume that 1 < p < and
1<q< .
1
Now set = ;p>1 <1
p
1 1 1
Therefore  1
q =1– p q
Putting these values of and 1 – in (2), we get

1 1
A B
Ap Bq … (3)
p q
If one of the functions f (x) and g (x) is zero a.e. then the theorem is trivial. Thus, we assume that
f 0, g 0 a.e. and hence the integrals
b b
|f|p dx and |g|q dx

a a
are strictly positive and hence f p

> 0, g q
> 0.
f(x) g(x)
Set f (x) = , g(x)
f p g q
1 1
and A p = |f(x)|, B q g(x) .
Then (3) gives
p q
f(x) g(x)
|f(x) g(x)|
p q
Integrating, we get
b b b
1 1
|f(x) g(x)|dx |f(x)|p dx |g(x)|p dx
p q
a a a
b b
1 |f(x)|p 1 |g(x)|q
= b dx b dx
p q
a
|f|p dx a
|g|q dx
a a

b b Notes
p q
|f| dx |g| dx
1 a 1 a
= b b
p q
|f|p dx |g|q dx
a a
1 1
= 1.
p q
Hence |f(x) g(x)|dx 1.

a
Putting the values of f (x) and g (x), we get
b
|f(x) g(x)|dx
1 or fg f p
g q
… (4)
f p g q
a
Now f Lp [a, b], g Lq [a, b]
b b
|f|p dx and |g|q dx

a a
f p
and g q
Therefore, from (4), we have
fg 1
< fg L [a, b]
Also the equality will hold when A = B
i.e. |f (x)|p = |g (x)|q, a.e.
|f|p |g|q
i.e. if p = q , a.e.
f p
g q
q p
or if g q
|f|p = f p
|g|q , a.e.
or if we have got some non-zero constants ,
|f|p = |g|q, a.e.
Hence the theorem.
5.1.7 Riesz-Hölder's Inequality for 0 < p < 1
If 0 < p < 1 and p and q are conjugate exponents, and f Lp and g Lq, then
|fg| f p g q , provided |g|q 0.
(In this case, the inequality is reversed than that of the case for 1 p < .)
1 1
Proof: Conjugacy of p, q 1
p q

Notes
1 1
1
q p
p
1 p
q
p
p 1
q
1 p 1 1 1 1 1
If we take p and , then 1 and since 0 p 1 1 P 1,
P P Q p Q 0 P
1 p 1
i.e. 1< P < < and also 1 p 0 1 as 0 < p < 1 Q > 1.
Q q Q
P, Q are conjugate numbers with 1 < P < .
If we take |fg| = Fp and |g|q = GQ.
1 1 q p
q
p p Q q
Then fg = |fg |p |g|Q = |f| |g|
= |f|p.
f, g are non-negative measurable functions s.t.
Also f Lp and g Lq.
Applying the Hölder’s inequality for P, Q to the functions f and g, we get
|F G| F P
G Q
1 1
P Q
|f|p |F|P |G|Q as |fg| = fg = |f|p
p
p
q
|f|p |fg| |g|q
1 1
p p q q
|f| |fg| |g|
1
|fg|
p p
|f| 1 , provided |g|q 0
q q
|g|
1 1
p q
|fg| |f|p |q|q f p g q

Theorem 4: SCHWARZ or CAUCHY-SCHWARZ INEQUALITY statement: Let f and g be square Notes

integrable, i.e.
f,g L2[a, b]; then fg L [a, b] and fg f 2

g 2.
Proof: Let x [a, b] be arbitrary, then
[|f(x)| – |g(x)|]2 0
or 2|f(x)|.|g(x)| |f(x)2 + |g(x)|2.
On integrating, we get
b b b
2 |f(x)g(x)|dx |f(x)2 dx |g(x)|2 dx … (i)

a a a
b b
Now f, g L2[a, b] f and g are measurable over [a, b] and |f(x)|2 dx , |g(x)|2 dx .
a a
Using in (i), we get |f(x) g(x)|dx

a
Thus fg L [a, b].
Let a R be arbitrary. Then
( |f| + |g|)2 0
( |f| |g|) 2 0
a
b b b
2
or |f|2 dx 2 |fg|dx |g|2 dx 0
a a a
b b b
2
Write A = |f| dx, B 2 |fg|dx, C |g|2 dx
a a a
2
Then we have A+ B+C 0 … (ii)
Now, if A = 0, then f(x) = 0 a.e. in [a, b] and hence B = 0 and both sides of the inequality to be
proved are zero. Thus when A = 0, the inequality is trivial.
B
Again, let A 0. Writing =– in (ii), we get
2A
2
B B
A B C 0.
2A 2A
which gives B2 4AC.

Notes Now putting the values of A, B, C in last inequality, we have
b 2
b b
4 |fg|dx 4 |f|2 |g|2

a a a
b b 1/2 b 1/2
or |f(x) (g(x)|dx |f(x)|2 |g(x)|2

a a a
or fg f 2
g 2.
Note: The above theorem is a particular case of Hölder’s inequality.
Example: Let f, g be square integrable in the Lebesgue sense then prove f + g is also
square integrable in the Lebesgue sense, and f + g 2 f 2 + g 2.
Solution: By hypothesis f2 L [a, b], g2 L [a, b].
f2 , g 2 L [a, b] fg L [a, b]. [by Schwarz inequality]
Again (f + g)2 = f2 + g2 + 2fg L [a, b].
Hence (f + g) is square integrable, again, we have
b b b b
(f g)2 = f 2
g 2
2 fg
a a a a
b b b 1/2 b 1/2
f2 g2 2 f2 g2 (by Schwarz inequality)

a a a a
1/2 1/2 2
b b
2 2
= f g
a a
b 1/2
b 1/2 b 1/2
2
(f g) f 2
g 2
a
a a
or f+g 2
f 2
+ g 2.
Example: Prove that f + g 1

f 1
+ g 1.
Solution: We know that |f + g| |f| + |g|.
Integrating both the sides.
|f g| |f| |g|
f+g 1
f 1
+ g 1.

5.2 Summary Notes
 The class of all measurable functions f (x) is known as L p – space over [a, b], if Lebesgue-
integrable over [a, b] for each p exists, 0 < p < , i.e.
|f|p dx , (p 0)
a
 The p-norm of any f Lp [a, b], denoted by f p

, is defined as
1
b p
p
f p
|f| , 0 p
a
1 1
 Let p, q > 1 be such that 1 , and let u and v be two non-negative numbers, at least
p q
one being non-zero. Then the function f : [0, 1] R defined by
1
f(t) ut v 1 t q q , t [0, 1],
has a unique maximum point at

1
up q
s
up vp
 Let p and q be conjugate indices or exponents and f Lp [a, b], g Lq [a, b], then it is evident
that
(i) f, g L [a, b]
(ii) fg f p
g q
i.e.
1 1
p p q q
|fg| |f| |g|
5.3 Keywords
1 1
Conjugate Numbers: Let p, q be any two n on-negative extended real numbers s.t. 1 , then
p q
p, q are called (mutually) conjugate numbers.
Hölder's Inequality: Let a1, a2, …, an, b1, b2, …, bn be non-negative numbers. Let p, q > 1 be real
1 1
number with the property 1 . Then
p q
1 1
n
n p n q
a jb j a pj bqj
j 1
j 1 j 1
Moreover, one has equality only when the sequences ap1 , , apn and bq , , bqn are proportional.

Notes LP-Spaces: The class of all measurable functions f (x) is known as L p-spaces over [a, b], if Lebesgue
– integrable over [a, b] for each p exists, 0 < p < , i.e.
|f|p dx , (p 0)
a
p-norm: The p-norm of any f Lp [a, b], denoted by f p

, is defined as
1
b p
p
f p
= |f| ,0<p< .
a
1. If f and g are non-negative measurable functions, then show that in Hölder’s inequality,
equality occurs iff some constants s and t (not both zero) such that sfp + tgq = 0.
2. State and prove Hölder’s Inequality.
Books G.H. Hardy, J.E. Littlewood, G. Polya, Inequalities, Cambridge University Press,
(1934)
L.P. Kuptsov, Hölder inequality, Springer (2001)
Kenneth Kuttler, An Introduction of Linear Algebra, BRIGHAM Young University,
2007
Online links www.m–hiKari.com

www.math.Ksu.edu
www.tandfonline.com

Unit 6: Minkowski Inequalities
Unit 6: Minkowski Inequalities Notes
CONTENTS
Objectives
Introduction
6.1 Minkowski Inequalities
6.1.1 Proof of Minkowski Inequality Theorems
6.1.2 Minkowski Inequality in Integral Form
6.2 Summary
6.3 Keywords
Objectives
 Define Lp-space, conjugate numbers and norm of an element of L p-space.
 Understand Minkowski inequality.
 Solve problems on Minkowski inequality.
Introduction
In mathematical analysis, the Minkowski inequality establishes that the L p spaces are normed
vector spaces. Let S be a measure space, let 1 p and let f and g be elements of L p (s). Then
f + g is in Lp (s), we have the triangle inequality
f+g p
f p
+ g p
with equality for 1 < p < if and only if f and g are positively linearly dependent, i.e. f = g for
some 0. In this unit, we shall study Minkowski’s inequality for 1 p < and for 0 < p < 1. We
shall also study almost Minkowski’s inequality in integral form.
6.1 Minkowski Inequalities
Here, the norm is given by:
1/p
f p
= |f|p d
if p < , or in the case p = by the essential supremum
f = ess supx S |f (x)|.
The Minkowski inequality is the triangle inequality in L p(S). In fact, it is a special case of the
more general fact

Notes
f p
= sup |fg|d , 1/p + 1/q = 1
g 1
q
where it is easy to see that the right-hand side satisfies the triangular inequality.
Like Holder's inequality, the Minkowski inequality can be specialized to sequences and vectors
by using the counting measure:
n 1/p n 1/p n 1/p
|xk y k |p |xk |p |y k |p
k 1 k 1 k 1
for all real (or complex) numbers x 1, ..., xn, y1, ..., yn and where n is the cardinality of S (the
number of elements in S).
Thus, we may conclude that
If p > 1, then Minkowski's integral inequality states that
b 1 /p b 1 /p b 1 /p
|f(x) g(x)|p dx |f(x)|p dx |g(x)|p dx

a a a
Similarly, if p >1 and a k, bk > 0, then Minkowski's sum inequality states that
n 1/p n 1/p n 1/p

p p p
|a k bk | |ak | |bk |
k 1 k 1 k 1
Equality holds iff the sequences a 1, a2, ... and b1, b2, ... are proportional.
6.1.1 Proof of Minkowski Inequality Theorems
Theorem 1: State and prove Minkowski inequality. If f and g Lp (1 p < ), then f + g Lp and
f+g p
f p
+ g p.
or
Let 1 p . Prove that for every pair f, g Lp {0, 1}, the function f + g Lp {0, 1} and that
f+g p f p + g p. When does equality occur?
Or
Suppose 1 p . Prove that for any two functions f and g in L p [a, b]
1 1 1
b p b p b p
|f g|p |f|p dx |g|p dx

a a a
Proof: When p = 1, the desired result is obvious.
If p = , then
|f| f a.e.
|g| g a.e.

|f + g| f + |g| Notes
f + g a.e.
f+g f + g
Hence the result follows in this case also. Thus, we now assume that 1 < p < .
Since Lp is a linear space, f + g Lp.
1 1
Let q be conjugate to p, then 1.
p q
p
Now (f + g) L
q
(f + g)p/q L
1 1 1 1 1 p 1
Since p 1 1
q q p q p
1 q
(p 1)q p, |f g|p |f g|p
p
and therefore |f + g|p–1 Lp (f + g)p/q Lp because p – 1 = .
q
On applying Hölder’s inequality for f and (f + g) p/q, we get
p 1 1
p p p
q
|f| |f g| q dx |f|p ) dx |f g| q dx
p 1 1
p q
or |f| |f g| q dx |f|p ) dx |f g|p dx … (1)
Since g Lp, therefore interchanging f and g in (1), we get
p 1 1
p q
|g| |f g| q dx |g|p ) dx |f g|p dx … (2)
Adding, we get
p p 1 1 1
p p q
|f| |f g| q dx |g| |f g| q dx |f|p dx |g|p dx |f g|p dx … (3)
p 1
Now |f + g|p = f g f g
1 1 p p
But 1 1 p p 1
p q q q
p
q
|f + g|p = f g f g
p
|f| |g| |f g| q

Notes p p
or |f + g|p |f||f g| q |g||f g| q
Integrating, we get
p p
|f g|p dx |f| |f g| q |g| |f g| q dx … (4)
Using (3), relation (4) becomes
1 1
p p 1
|f g|p dx |f|p dx |g|p dx |f g|p dx q
1
Dividing each term by |f g|p dx q
, we get
1 1q 1
p
1
p
|f g|p dx |f|p dx |g|p dx
1 1 1 1
But 1 1
p q q p
1 1 1
p p p
So |f g|p dx |f|p dx |g|p dx
or f+g p
f p
+ g p
Hence the proof.
Note Equality hold in Minkowski’s inequality if and only if one of the functions f and g
is a multiple of the other.
Theorem 2: Minkowski’s inequality for 0 < p < 1. If 0 < p < 1 and f, g are non-negative functions
in Lp, then
f+g p
f p
+ g p.
Proof: For this proceed as in theorem Minkowski’s inequality and applying the Hölder’s
inequality for 0 < p < 1 for the functions f Lp and (f + g)p/q Lq, we get
1p 1q
|f||f g|p /q |f|p |f g|p /q )q
1p 1q
|f||f g|p /q |f|p |f g|p … (i)
Also g Lp, proceeding as above, we get
1p 1q
|g||f g|p /q |g|p dx |f g|p … (ii)

Adding these two, Notes
1 1 1
|f g|p /q (|f| |g|) |f|p p |g|p p |f g|p q … (iii)
1 1 p
Also 1 1 p
p q q
p p
|f + g|p = |f g| |f g| q |f| |g| |f g| q , as f 0, g 0
1
1 1 q
|f g|p = |f|p p
|g|p p
|f g|p
1
q
Dividing by |f g|p , we get
1 (1/q )
|f g|p f p
g p
1/q
|f g|p f p
+ g p
f+g p
f p
+ g p
6.1.2 Minkowski Inequality in Integral Form
Statement: Suppose f :  ×   is Lebesgue measurable and 1 p < . Then
1/p 1/p
h(x, y) dy dx h(x, y)p dx dy
Proof: By an approximation argument we need only consider h of the form
h(x, y) fj (x)1 fj (y), (x, y)   ,

j 1
where N is a positive integer, f j is Lebesgue measurable, and F j Ln, j = 1, … Ni and Fi Fj = if

1 i < j N. We use Minkowski’s inequality to estimate
p 1/p
1/p N N 1/p
h(x, y) dy dx = Fj fj (x) Fj fi (x)p dx

j 1 j 1
dx
But
p 1/p N 1/p N 1/p
h(x, y) dx dy = |h(x, y)|p dx |fj (x)|p dx

Fi Fj
j 1 j 1

Notes
Example: If <fn> is a sequence of functions belonging to L2(a, b) and also f L2 (a, b) and
Lim fn – f 2 = 0, then prove that
b b
f dx Lim fn2dx
2
a a
Solution: By Minkowski’s inequality, we get
fn 2 f2 fn – f 2
Lim fn 2 f Lim fn – f 2
=0
Lim fn 2 f =0 Lim fn 2
= f 2
b 1 /2 b 1 /2
b b
2 2
Lim (fn ) dx f dx Lim fn2 dx f 2dx .
a a a a
6.2 Summary
 The class of all measurable function f (x) is known as L p space over [a, b], if Lebesgue
integrable over [a, b] for each p exists, 0 < p < .
 If f and g Lp (1 p ), then f + g Lp and f + g p

f p
+ g p.
6.3 Keywords
Lp-space: The class of all measurable functions f (x) is known as Lp-space over [a, b], if Lebesgue-
integrable over [a, b] for each exists, 0 < p < , i.e.,
|f|p dx , (p 0)
a
Minkowski Inequality in Integral Form: Suppose f :  ×   is Lebesgue measurable and 1

p < . Then
1/p 1/p
h(x, y) dy dx h(x, y)p dx dy
Minkowski Inequality: Minkowski inequality establishes that the L p spaces are normed vector
spaces. Let S be a measure space, let 1 p and let f and g be elements of Lp (s). Then f + g is in
p
L (s), we have the triangle inequality
f+g p
f p
+ g p
with equality for 1 < p < if and only if f and g are positively linearly dependent.

1. If f, g are square integrable in the Lebesgue sense, prove that f + g is also square integrable
and
f+g 2
f 2
+ g 2.
2. If | < p < , then show that equality can be true, iff there are non-negative constants and
, such that f = g.
Books Books: Stein, Elias (1970). Singular Integrals and Differentiability Properties of
Functions. Princeton University Press.
Hardy, G.H.; Littlewood, J.E.; Polya, G. (1952). Inequalities, Cambridge

Mathematical Library (second ed.). Cambridge: Cambridge University Press.
Online links Mathworld.wolfram.com>Calculus and Analysis>Inequalities

Planet math.org/Minkowski In-equality.html

Notes Unit 7: Convergence and Completeness
CONTENTS
Objectives
Introduction
7.1 Convergence and Completeness
7.1.1 Convergent Sequence
7.1.2 Cauchy Sequence
7.1.3 Complete Normed Linear Space
7.1.4 Banach Space
7.1.5 Summable Series
7.1.6 Riesz-Fischer Theorem

7.2 Summary
7.3 Keywords
Objectives
 Understand convergence and completeness.
 Understand Riesz-Fischer theorem.
 Solve problems on convergence and completeness.
Introduction
Convergence of a sequence of functions can be defined in various ways, and there are situations
in which each of these definitions is natural and useful. In this unit, we shall start with the
definition of convergence and Cauchy sequence and proceed with the topic completeness of L p.
7.1 Convergence and Completeness
7.1.1 Convergent Sequence
Definition: A sequence <xn> in a normal linear space X with norm . is said to converge to an
element x X if for arbitrary > 0, however small, n0 N such that xn – x < , n > n0 .
Then we write lim x n x.

n

Unit 7: Convergence and Completeness
7.1.2 Cauchy Sequence Notes
Definition: A sequence <xn> in a normal linear space (X, . ) is said to be a Cauchy sequence if for
arbitrary > 0, n0 N s.t.
xn – xm < , n, m n0.
Definition: A normed linear space (X, . ) is said to be complete if every Cauchy sequence <xn> in
it converges to an element x X.
7.1.4 Banach Space
Definition: A complete normed linear space is also called Banach space.
7.1.5 Summable Series
Definition: A series u n in N1 is said to be summable to a sum u if u N1 and lim S n u , where

n
n 1
Sn = u1 + u2 + … + un
In this case, we write
u un .
n 1
Further, the series u n is said to be absolutely summable if un .

n 1 n 1
7.1.6 Riesz-Fischer Theorem
Theorem: The normed Lp-spaces are complete for (p 1).
Proof: In order to prove the theorem, we shall show that every Cauchy sequence in L p [a, b] space
converges to some element f in Lp-space. Let <fn> be one of such sequences in Lp-space. Then for
given > 0, a natural number n0, such that
m, n n0 fm – fn p
< ,
1
since is arbitrary therefore taking , we can find a natural number n 1 such that
2
1
for all m, n n1 f m – fn p
<
2
1
Similarly, taking , k N, we can find a natural number n k, such that
2k
1
for all m, n nk fm – f n <
p
2k

Notes 1
In particular, m > nk fm – fnk < .
2k
Obviously n1 < n2 < n3 … < nk < …
i.e. <nk> is a monotonic increasing sequence of natural numbers.
Set gk = fnk , then from above, we have
1
g2 – g 1 p
= fn 2 fn ,
2 p 2
1
g3 – g 2 = fn 3 fn ,
p 2 p 22
… … … … …
… … … … …
1
gk + 1 – g k p
= fnk fnk .
1 p 2k
… … … … …
… … … … …
Adding these inequalities, we get
1
gk 1 gk < 1 … (i)
k 1
p
k 1
2k
Thus gk 1
gk is convergent. Define g such that
p
k 1
g (x) = g 1 (x) gk 1
gk
p
if R.H.S. is convergent … (ii)
k 1
and g (x) = , if right hand side is divergent.
1
b b n p
p p
Now, |g(x)| dx = lim |g 1 (x) gk gk
n 1 dx
a a k 1
or g p
= lim g1 p
gk 1
gk (By Minkowski’s inequality)
n p
k 1
= g1 + gk 1
gk < g1 + 1, [by (i)]
p p p
k 1
g p
< g Lp [a, b].
Let E = {x [a, b] : g (x) = }.

Now we define a function f such that Notes
f (x) = 0, x E
and f (x) = g1 (x) + gk 1

g k , for x [a, b] but x E,
k 1
m 1
or f (x) = lim
m
g1 gk 1
gk , for x E
k 1
= lim g m (x)
m
Thus f (x) = 0, for x E and
f (x) = lim g m (x) for x E.

m
f (x) = lim
m
g m (x) a.e. in [a, b]
or lim g m f = 0 a.e. in [a, b] … (iii)

m
m 1
Also, gm (x) = g 1 gk 1
gk
k 1
m 1
|gm| |g1| + gk 1
gk
k 1
|g1| + gk 1
g k = g,
k 1
|gm| g, m N
lim g m (x) g
m
(iii) |f| g.
Again, |gm – f| |gm| + |f| g + g = 2g.
|gm – f| 2 g.
Thus there exists a function g Lp [a, b] s.t.
|gm – f| 2g, m
and lim g m f = 0 a.e. in [a, b] … (iv)

m
Applying Lebesgue dominated convergence theorem,
b b b
p p
lim gm f dx = lim g m f dx 0 dx 0 [Using (iv)]
m m
a a a

Notes 1
b p
p
lim gm f dx =0
m
a
lim g m f p
=0
m
lim fnm f = 0 as gm = fnm

m p
fnm f < .
p
Also fm fnm < .

p
fm – f p
= fm fnm fnm f
p
fm fnm fnm f
p p
<( + )= .
Hence lim fm f p
=0
m
or lim fm = f Lp [a, b].

m
This proves the theorem.
Alternative Statement of this Theorem
A convergent sequence <fn> in Lp-spaces has a limit in L p-space.
Or
Every Cauchy sequence <fn> in the Lp-space converges to a function in Lp-space.
Theorem: Prove that a normed linear space is complete iff every absolutely summable sequence
is summable.
Proof: Necessary part
Let X be a complete normed linear space with norm . and <fn> be an absolutely summable
sequence of elements of X
fn =M< ,
n 1
For arbitrary > 0, however small, no N
s.t. fn < , … (i)

n no
Now, if Sn = fi , then n m no, we get

i 1

n n
Notes
Sn – Sm p
= fi fi
i m 1 i m 1
fi
i no
Sequence <Sn> of partial sums is a Cauchy sequence
<Sn> converges.
Sequence <fn> is summable to some element S X.
But X is a complete space. Therefore <Sn> will converge to some element S X.
Sufficient part: Given that every absolutely summable sequence in the space X is summable.
To show that X is complete.
Let <fn> be a Cauchy sequence in X.
For each positive integer k, we can choose a number n k N such that
1
fn – fm < , n, m nk … (ii)
2k
We can choose these n k’s such that nk+1 > nk.
Then fn is a subsequence of <fn>.

k 1
Setting g1 = fn1 and gk = fnk fnk 1 , (k > 1), we get a sequence <gk> s.t. its kth partial.
Sum = Sk = g1 + g2 + … + gk = fn1 (fn 2 fn1 ) (fnk fnk 1 ) fnk .
1
Now, gk = fnk fnk , [by (ii)], k>1
1
2k 1
1
gk g1 = g1 + 1 (a finite quantity)
k 1 k 2
2k 1
The sequence <gk> is absolutely summable and hence by the hypothesis, it is a summable
sequence.
The sequence of partial sums of this sequence converges to some S X.
The sequence <Sk> converges and hence fnk converges to some f X.
Now, we shall show that the limit f n = f.
Again, since <fn> is a Cauchy sequence, we get that for each > 0, however small, n N s.t.
n, m > n .
fn – fm < .
2
Also since fnk f, n N such that k n ,

Notes
fnk f .
2
Choosing a number k, as large that k > n and nk > n , we get
fn f fn fnk fnk f ,
2 2
n > n , we obtain fn – f < , where is an arbitrary quantity.

fn f X and hence X is a complete space.
Theorem: Let {fn} be a sequence in L p, 1 p < , such that fn f a.e. and that f L p.
If lim fn p
= f p, then
n
lim fn f p
= 0.
n
Proof: Without any loss of generality, we may assume that each fn 0 a.e. so that f is also 0 a.e.
since the result in general case follows by considering f = f + – f–.
Now, let a and b be any pair of non-negative real numbers, we have
|a – b|p 2p (|a|p + |b|q),
1 p<
So, we get
2p (|fn|p + (|f|p) – |fn – f|p 0 a.e.
Thus, by Fatou’s Lemma and by the given hypothesis,
We get
p p p
2p 1
|f|p = lim 2 p fn f fn f
n
p p p
lim inf 2 p fn f fn f
n
p p p
= 2 p 1 lim fn 2p f lim inf fn f
n n
p p
= 2p 1
f lim sup fn f .
n
p
Since f it follows that
p
lim sup fn f 0.
n
p p
Therefore lim sup fn f lim inf fn f 0,
n n
p
So that lim fn f =0
n

1 Notes
p p
lim fn f dt =0
n
lim fn f p
= 0.
n
Theorem: In a normed linear space, every convergent sequence is a Cauchy sequence.
Proof: Let the sequence <xn> in a normed linear space N, converges to a point x o N1. We shall
show that it is a Cauchy sequence.
Let > 0 be given. Since the sequence converges to xo a positive integer mo s.t.
n mo xn – xo < /2 … (1)
Hence for all m, n mo, we have
x m – xn = xm – xo + xo – xn
x m – xo + x o – xn
< = by (1)
2 2
It follows that the convergent sequence <xn> is a Cauchy sequence.
Theorem: Prove that L [0, 1] is complete.
Proof: Let (fn) be any Cauchy sequence in L , and let
Ak = {x : |fk (x)| > fk },
Bm, n = {x : |fk (x) – fm (x)| > fk – fm }.
Then m (Ak) = 0 = m (Bm, n) (k, m, n = 1, 2, 3, …),
So that if E is the union of these sets, we have m (E) = 0.
Now, if x F = [0, 1] – E, then
|fk (x) fk
|fn (x) – fm (x) fn – f m 0 as n, m .
Hence the sequence (fn) converges uniformly to a bounded function on F.
Define f : [0, 1] R by
lim fn (x) if x F
f (x) = n
0, if x E
Then f L and fn – f 0 as n .
Thus L is
Hence proved.
7.2 Summary
 A sequence <xn> in a normal linear space X with norm . is said to converge to an element
x X if for arbitrary > 0, however small, no N s.t. xn – x < , n > no. Then we write
lim x n x.
n

Notes  A complete normed linear space is also called Banach space.
 The normed Lp-spaces are complete for (p 1).
 A convergent sequence <fn> in Lp-spaces has a limit in L p-space.
 A normed linear space is complete iff every absolutely summable sequence is summable.
 In a normed linear space, every convergent sequence is a Cauchy sequence.
7.3 Keywords
Banach Space: A complete normed linear space is also called Banach space.
Cauchy Sequence: A sequence <xn> in a normal linear space (X, . ) is said to be a Cauchy
sequence if for arbitrary > 0, n0 N s.t.
xn – xm < , n, m n0.
Complete Normed Linear Space: A normed linear space (X, . ) is said to be complete if every
Cauchy sequence <xn> in it converges to an element x X.
Convergence almost Everywhere: Let <fn> be a sequence of measurable functions defined over a
measurable set E. Then <fn> is said to converge almost everywhere in E if there exists a subset Eo
of E s.t.
(i) fn (x) f (x), x E – Eo.
and (ii) m (Eo) = 0.
Convergent Sequence: A sequence <xn> in a normal linear space X with norm . is said to
converge to an element x X if for arbitrary > 0, however small, n0 N such that xn – x <
, n > n0 .
Then we write lim x n x.

n
Normed Linear Space: A linear space N together with a norm defined on it, i.e., the pair (N, )
is called a normed linear space.
Summable Series: A series u n in N1 is said to be summable to a sum u if u N1 and lim S n u,

n
n 1
where
Sn = u1 + u2 + … + un
1. Prove that  np is complete.
2. Prove that the vector space L equipped with L

is a complete vector space.
3. Suppose f L is supported on a set of finite measure.
Then f Lp for all p < , and
f L
f L
as p .

Notes
If f Lp (p > 0), f 0 and fn = min (f, n) (n N), show that fn Lp and lim fn f p
= 0.
n
Books H.L. Royden, Real analysis.

Walter Rudin, Real and Complex Analysis, Third, McGraw-Hill Book Co., New
York, 1987.
Online links www.public.iastate.edu

www.scribd.com/doc/49732162/103.Convergence-and-Completeness.

Notes Unit 8: Bounded Linear Functional on the Lp-spaces
CONTENTS
Objectives
Introduction
8.1 Bounded Linear Functionals on L p-spaces
8.1.1 Linear Functional
8.1.2 Bounded Linear Functional
8.1.3 Bounded Linear Functional on L p-spaces
8.1.4 Norm
8.1.5 Continuous Linear Functional
8.1.6 Theorems
8.2 Summary
8.3 Keywords
Objectives
 Understand bounded linear functional on L p-spaces
 Understand related theorems.
 Solve problems on bounded linear functionals.
Introduction
In this unit, we obtain the representation of bounded linear functionals on L p-space. We shall
also study about linear functional, continuous linear functionals and norm of f  p . Further, we
*
shall prove important theorems on bounded linear functionals.
8.1 Bounded Linear Functionals on Lp-spaces
8.1.1 Linear Functional
Definition: Let N1 be a normed space over a field R (or C). A mapping f : N1 R (or C) is called a
linear functional on N 1 if f ( x + y) = f (x) + f (y), x, y N1 and , R (or C).
Definition: A linear functional f on a normed space N 1 is said to be bounded if there is a constant

k > 0 such that
|f (x)| k x , x N1 … (1)

Unit 8: Bounded Linear Functional on the L p-spaces
The smallest constant k for which (1) holds is called the norm of f, written f . Notes
|f(x)|
Thus f = sup :x 0 and x N 1 or equivalently
x
f = sup {|f (x)| : x X and x = 1}.
Also |f (x)| f x x N1.

p p
Definition: Let p R, p > 0. We define L = L [0, 1] to be the set of all real-valued functions on
[0, 1] such that
1
1 p
(i) f is measurable and (ii) f p
= |f| p < .
0
Note L1 or simply L denotes the class of measurable function f (x) which are also
L-integrable.
8.1.3 Bounded Linear Functional on L p-spaces
If x  p and f is bounded linear functional on  p , then f has the unique representation of the
form as an infinite series
f (x) = xk f(e k )
k 1
8.1.4 Norm
The norm of f  *p is given by
1
q
q
f = |f(ek )|
k 1
Likewise in finite dimensional case, the bounded linear functionals are characterised by the
values they assume on the set ek, k = 1, 2, 3, … .
8.1.5 Continuous Linear Functional
A linear functional f is continuous if given > 0 there exists > 0 so that
|f (x) – f (y)| whenever x – y .
8.1.6 Theorems
1 1
Theorem 1: Suppose 1 p < , and 1 , then, with  = Lp we have
p q
q
* = L ,

Notes in the following sense: For every bounded linear functional  on Lp there is a unique g Lq so
that
 (f) = f(x)g(x)d (x), for all f Lp

x
Moreover,  *
= g Lq
.
This theorem justifies the terminology where by q is usually called the dual exponent of p.
The proof of the theorem is based on two ideas. The first, as already seen, is Hölder’s inequality;
to which a converse is also needed. The second is the fact that a linear functional  on Lp, 1 p
< , leads naturally to a (signed) measure . Because of the continuity of  the measure is
absolutely continuous with respect to the underlying measure , and our desired function g is
then the density function of in terms of .
We begin with:
Lemma: Suppose 1 p, q , are conjugate exponents.
(i) If g Lq, then g Lq

sup fg .
f
Lp 1
(ii) Suppose g is integrable on all sets of finite measure and
sup fg = M <
f p
L 1
f simple
Then g Lq, and g Lq

= M.
For the proof of the lemma, we recall the signum of a real number defined by
1 if x 0
sign (x) = 1 if x 0
0 if x 0
Proof: We start with (i). If g = 0, there is nothing to prove, so we may assume that g is not 0 a.e.,
and hence g Lq
0 . By Hölder’s inequality, we have that
g sup fg .
Lq f
Lp 1
To prove the reverse inequality we consider several cases.
 First, if q = 1 and p = , we may take f (x) = sign g (x). Then, we have f L

1 and
clearly fg g L1
.
q 1
 If 1 < p, q < , then we set f (x) = |g (x)| q–1 sign g (x) g Lq
. We observe that
p p(q 1) p(q 1)
f Lp
g (x) d g Lq
1 since p (q – 1) = q, and that fg g Lq
.

 Finally, if q = and p = 1, let > 0, and E a set of finite positive measure, where |g (x)| Notes
g L
. (Such a set exists by the definition of g L
and the fact that the measure is -
finite). Then, if we take f (x) =  E (x) sign g (x)/ (E), where E denotes the characteristic
function of the set E, we see that f L1
1 , and also
1
fg |g| g .
(E)
E
This completes the proof of part (i).

To prove (ii) we recall that we can find a sequence {g n} of simple functions so that |gn (x) |g (x)|
q 1
while gn (x) g (x) for each x. When p > 1 (so q < ), we take fn (x) = |gn (x)|q–1 sign g(x) g n Lq
.
As before, fn Lp
1 . However
q
g n (x)
fn g gn ,
q 1 Lq
gn Lq
and this does not exceed M. By Fatou’s Lemmas if follows that |g|q Mq , so g Lq with
g Lq
M . The direction g Lq
M is of course implied by Hölder’s inequality. When p = 1 the
argument is parallel with the above but simpler. Here we take fn (x) = (sign g (x))  En (x), where
En is an increasing sequence of sets of finite measure whose union is X. The details may be left to
the reader.
With the lemma established we turn to the proof of the theorem. It is simpler to consider first the
case when the underlying space has finite measure. In this case, with  the given functional on
Lp, we can then define a set function by
(E) =  ( E),
where E is any measurable set. This definition make sense because E is now automatically in Lp
since the space has finite measure. We observe that
| (E)| C ( (E))1/p … (1)
1/p
where C is the norm of the linear functional, taking into account the fact that  E Lp
(E) .
Now the linearity of  clearly implies that is finitely-additive. Moreover, if {E n} is a countable
collection of disjoint measurable sets, and we put E  n 1 E n , EN*  n N 1 E n , then obviously
E  E*
N  E
n 1
n .
N
Then (E) E N* (E n ) . However EN* 0 , as N because of (1) and the
n 1
assumption p < . This shows that is countably additive and moreover (1) also shows us that
is absolutely continuous with respect to .
We can now invoke the key result about absolutely continuous measures, the Lebesgue-Radon
– Nykodin theorem. It guarantees the existence of an integrable function g so that (E) = g du

E

Notes
for every measurable set E. Thus we have (E)  Eg d . The representation (f) fg d
then extends immediately to simple function f, and by a passage to the limit, to all f Lp since the
simple functions are dense in L p, 1 p < . Also by lemma, we see that g Lq
 .
To pass from the situation where the measure of X is finite to the general case, we use an
increasing sequence {En} of sets of finite measure that exhaust X, that is, X =  n 1 E n . According
to what we have just proved, for each n there is an integrable function g n on En (which we can set
to be zero in Ecn ) so that
 (f) = fg n d … (2)
whenever f is supported in En and f Lp. Moreover by conclusion (ii) of the lemma g n Lp

 .
Now it is easy to see because of (2) that gn = gm a.e. on Em , whenever n m. Thus limn gn (x)
= g (x) exists for almost every x, and by Fatou’s lemma, g Lq  . As a result we have that
 (f) f g du for each f Lp supported in En, and then by a simple limiting argument, for all f
Lp supported in En. The fact that  g Lq is already contained in Hölder’s inequality and
therefore the proof of the theorem is complete.
Theorem 2: Let f be a linear functional defined on a normed linear space N, then f is bounded
f is continuous.
Proof: Let us first show that continuity of f boundedness of f.
If possible let f is continuous but not bounded. Therefore, for any natural number n, however
large, there is some point x n such that
|f (xn)| n xn … (1)
xn
Consider the vector y n so that
n xn
1
yn = .
n
yn 0 as n .
yn 0 in the norm.
Since any continuous functional maps zero vector into zero, and f is continuous f (yn) f (0) = 0.
1
But |f (yn)| = f(x n ) … (2)
n xn
It now follows from (1) and (2) that
|f (yn)|> 1, a contradiction to the fact that
f (yn) 0 as n .
Thus if f is bounded then f is continuous.

Conversely, let f is bounded. Then for any sequence (x n), we have Notes
|f (xn)| k xn n = 1, 2, … and k 0.
Let xn 0 as n then
f (xn) 0
f is continuous at the origin and consequently it is continuous everywhere.
This completes the proof of the theorem.
Theorem 3: If L is a linear space of all n-tuples, then
(i)  np *  nq
(ii)  n1 *  n
(iii)  n *  n1
Proof: Let (e1, e2, …, en) be a standard basis for L so that any x = (x 1, x2, …, xn) L can be written
as x = x1e1 + x2e2 + … + xnen.
If f is a scalar valued linear function defined on L, then we get
f (x) = x1 f (e1) + x2f (e2) + … + xnf (en) … (1)
f determines and is determined by n scalars y i = f (ei).
Then the mapping

n
y = (y1, y2, … yn) f where f (x) = xi y i is an isomorphism of L onto the linear space L of all
i 1
function f. We shall establish (i) – (iii) by using above given facts.
(i) If we consider the space L =  np (1 p < ) with the pth norm, then f is continuous and L
*
represents the set of all continuous linear functionals on  np so that L =  np .
Now for y f as an isometric isomorphism we try to find the norm of y’s.
For 1 < p < , we show that
 np * =  np .
For x  np , we have defined,
1
n p
p
x = |xi |
i 1
n n
Now |f (x)| = xi yi |x i ||y i |

i 1 i 1
By using Holder’s inequality, we get
1 1
n n p n q
|xi y i |  p
|xi | q
|y i | so that
i 1 i 1 i 1

Notes 1 1
n q n p
q p
|f (x)|  |y i | |x i |
i 1 i 1
Using the definition of norm for f, we get
1
n q
q
f  |y i | … (2)
i 1
Consider the vector, defined by
|y i |q
xi = , yi 0 and xi = 0 if yi = 0 … (3)
yi
1 1
n n p p
p
|y i |q
Then, x = |xi |p … (4)
i 1 i 1
|y i |
Since q = p (q – 1) we have from (4),
1
n q
q
x = |y i | … (5)
i 1
n n
|y i |q
Now f (x) = xi yi yi … (4)
i 1 i 1
yi
= |y i |q (By (3))
i 1
So that
|y i |q = |f (x) f x … (6)
i 1
From (5) and (6) we get,
1
n 1
p
|y i | q f
i 1
1
n q
|y i |q f … (7)
i 1
Also from (2) and (7) we have
1
n q
q
f = |y i | , so that
i 1

y f is an isometric isomorphism. Notes
Hence  np * =  nq .
(ii) Let L =  n1 with the norm defined by
x = |xi |
i 1
Now f defined in (1), above is continuous as in (i) and L here represents the set of continuous
linear functional on  n1 so that
L = 1 * .
n
We now determine the norm of y’s which makes y f an isometric isomorphism.
Now, f (x) = xi yi
i 1
|xi ||y i |
i 1
n n
But |xi ||y i | max. |y i | |x i | so that

i 1 i 1
f (x) max. |y i | |x i |.
i 1
From the definition of the norm for f, we have
f = max. {|yi| : i = 1, 2, …, n} … (8)

Now consider the vector defined as follows:
If |yi| = max |y i | , let us consider the vector x as

1 i n
|y i |
xi = when|y i | max |y i | and x i 0 … (9)
yi 1 i n
otherwise
From the definition, xk = 0 k i, so that we have
yi
(x) = =1
y
Further | f (x)| = (x i y i ) = |yi|.

i 1
Hence |yi| = |f (x)| f x

Notes |yi| f or max {|yi|} [ x = 1]
f … (10)
From (8) and (10), we obtain
f = max. {|yi|} so that
f is an isometric isomorphism of L to 1 * .
n
y
Hence  n1 *  n .
(iii) Let L =  n with the norm
x = max {|xi| : 1, 2, 3, …, n}.
Now f defined in (1) above is continuous as in (1). Let L represents the set of all continuous
linear functionals on  n so that
L = n * .
Now we determine the norm of y’s which makes y f as isometric isomorphism.
n n
|f (x)| = xi yi |x i ||y i | .
i 1 i 1
n n
But |xi ||y i | max (|xi |) |y i |

i 1 i 1
Hence we have
n n
|f (x)| |y i | ( x ) so that f |y i | … (11)

i 1 i 1
Consider the vector x defined by
|y i |
xi = when yi 0 and x i = 0 otherwise. … (12)
yi
|y i |
Hence x = max 1.
|y i |
n n
and |f (x)| = |x i y i | |y i |
i 1 i 1
Therefore
n
|y i | = |f (x)| f x = f .
i 1
|y i | f … (13)
i 1

It follows now from (11) and (13) that Notes
f = |y i | so that y f is an isometric isomorphism.

i 1
Hence,  n *  n1 .
Note We need the signum function for finding the conjugate spaces of some infinite
dimensional space which we define as follows:
If is a complex number, then
sgn if 0
| |
0 if 0
(i) |sgn | = 0 if = 0 and | sgn | = 1 if 0
(ii) sgn = 0 if = 0 and sgn = =| |, if 0.

| |
Theorem 4: The conjugate space of  p is  q , where
1 1
= 1 and 1 < p < .
p q
or  *p q .
Proof: Let x = (xn)  p so that |x n |p … (1)

n 1
Let en = (0, 0, 0, …, 1, 0, 0, …) where 1 is in the n th place.
en  p for n = 1, 2, 3, … .
We shall first determine the form of f and then establish the isometric isomorphism of  p onto
*
q .
By using (en), we can write any sequence
(x1, x2, … xn, 0, 0, 0, …) in the form xk e k and

k 1
x xk ek (0, 0, 0, , xn 1 , xn 2 , ).
k 1

Notes 1
n p
p
Now x xk ek |xk | … (2)
k 1 k n 1
The R.H.S. of (2) gives the remainder after n terms of a convergent series (1).
1
p
Hence |xk |p 0 as n . … (3)
k n 1
From (2) and (3) if follows that
x = xk ek … (4)
k 1
Let f  *p and Sn = xk e k then

k 1
Sn x as n (Using (4))
Since f is linear, we have
f (Sn) = xk f(e k ) .
k 1
Also f is continuous and Sn x, we have
f (Sn) f (x) as n .
f (x) = xk f (e k ) … (5)
k 1
which gives the form of the functional on  p .
Now we establish the isomeric isomorphism of  *p onto  q , for which proceed as follows:
Let f (ek) = k
and show that the mapping
T :  *p  q given by
T (f) = ( 1, 2
, …, k
, …) is an isomeric isomorphism of  *p onto  q .
First, we show that T is well defined.
For let x  p , where x = ( 1, 2

, …, , 0, 0, …) where
n
| |g
k
1
sgn k , 1 k n
= n k
k 0
| k| = | k|q–1 for 1 k n.
1)p 1 1
| k|p = | k |(q = | k|q.  q p(q 1) q
p q

Notes
Now k k k | k |q 1 sgn k | k |q 1
sgn k
k k
= | k|q = | k|p … (7)
(Using property of sgn function)
1
n p
p
x = | k|
k 1
1
n p
x = | k |p
k 1
1
n q
q
= | k | … (8)
k 1
Since we can write
x = k e k , we get
k 1
n n
f (x) = k f (e k ) k k
k 1 k 1
f (x) = | k |q (Using (7)) … (9)

k 1
We know that for every x p
| f (x) | f x ,
which upon using (8) and (9), gives
1
n n p
|f (x)| | k | q
f | k
q
|
k 1 k 1
which yields after simplification.
1
n p
q
| k | f … (10)
k 1
Since the sequence of partial sum on the L.H.S. of (10) is bounded; monotonic increasing, it
converges. Hence
1
n p
q
| k | f … (11)
k 1
So the sequence ( k) which is the image of f under T belongs to  q and hence T is well defined.

Notes We next show that T is onto  q .
Let ( k)  q , we shall show that is a g  *p such that T maps g into ( k).
Let x  p so that
x = xk ek
k 1
We shall show that
g (x) = xk k is the required g.

k 1
Since the representation for x is unique, g is well defined and moreover it is linear on  p . To
prove it is bounded, consider
n n
|g (x)| = k xk k xk
k 1 k 1
1 1
n p n q
p q
xk k (Using Hölder’s inequality)
k 1 k 1
1
n q
q
|g (x)| x | k|
k 1
g is bounded linear functional on  p .
Since ek  p for k = 1, 2, …, we get
g (ek) = k
for any k so that
Tg = ( k) and T is on  *p onto  q .
We next show that
Tf = f so that T is an isometry.
Since Tf  q , we have from (6) and (10) that

1
q
q
= Tf f
k
k 1
Also, x p x= xk e k . Hence
k 1
f (x) = xk (e k ) xk k
k 1 k 1

Notes
|f (x)| |xk || k|
k 1
1 1
n q p
q p
| k | xk (Using Hölder’s inequality)
k 1 k 1
1
n q
or |f (x)| | k | q
x x p .
k 1
Hence we have
1
|f(x)| q
sup | |q = Tf (Using (6))
x 0 x k
k 1
which upon using definition of norm yields
f Tf … (13)
Thus f = Tf (Using (12) and (13))
From the definition of T, it is linear. Also since it is an isometry, it is one-to-one and onto
(already shown). Hence T is an isometric isomorphism of  *p onto  q , i.e.,
 *p q .
1 1
Theorem 5: Let p > 1 with 1 and let g Lp (X). Then the function defined by
p q
F (f) = fg d for f L p (X)

X
is a bounded linear functional on L p (X) and
F = g q
… (1)
Proof: We first note that
F is linear on L p (X). For if f1, f2 Lp (X), then we get
F (f1 + f2) = (f1 f2 )g d f1g d f2g d

X X X
= F (f1) + F (f2)
So that
F (f1 + f2) = F (f1) + F (f2)
and F ( F) = fg d F(f) .
X
Now |F (f)| = fg d |fg|d … (2)

X X

Notes Making use of Hölder’s inequality, we get
1 1
p q
|fg|d |f|p d |g|q d
X X X
= f p
g q
… (3)
From (2) and (3) it follows that
|F (f)| f p
g q.
F|f|
Hence sup :f Lp(X) and f 0 g q
f p
F g q
(Using definition of the norm) … (4)
Further, let f = |g|q–1 sgn g … (5)
Since sgn g = 1, we get
|f|p = |g|p(q–1) = |g|q ( p (q – 1) = q)
1 1
p q
Thus, f Lp (X) and |f| d p = |g| d q … (6)
X X
1
p
q /p
But |g| d q
|g|p d = g q
X X
which implies on using (6) that
q /p
f p
= g q
… (7)
Now F (f) = fg d |g|q 1 g sgn g d

X X
q
= |g|q d g q
X
q
Hence g q
g = F (f) F f p.
and this on using (7) yields that
q q /p
g q
= F (f) F g q
q q /p
g q
= g q
F … (8)
( g 0)

From (4) and (8) it finally follows that Notes
F = g q.
Approximation by Continuous Function
Theorem 6: If f is a bounded measurable function defined on [a, b], then for given > 0, a
continuous function g on [a, b], such that
f–g 2
<
Proof: Let F (x) = f(t) dt where x [a, b].

a
x h x
Then |F (x + h) – F (x)| = f(t) dt f(t) dt

a a
x h x h
= f(t) dt f(t) dt
x x
Mh, where |f (x)| M, x [a, b].

Taking h < , and Mh < 1
, we get
|x + h – x| < | F (x + h) – F (x) | < 1
F (x) is continuous on [a, b].
x h
Let Gn (x) = n f(t) dt : x [a, b] and n N;

x
1
then Gn (x) = n F x F(x) ( F (x) is continuous on [a, b]
n
Gn (x) is continuous on [a, b] n)
Again, since F (x) = f (t) dt, x [a, b] .

a
F (x) = f (x) a.e. in [a, b].
F(x (1/n) F(x)

Now, Lim G n (x) = Lim
n n 1/n
F(x h) F(x) 1
= Lim ,h
h 0 h n
= F (x) = f (x) a.e. in [a, b]
2
and hence Lim G n (x) f(x) = 0.
n

Notes x (1 / n ) x (1 / n )
Also |Gn (x)| = f (t) dt n |f (t)|dt M.

x x
Hence |Gn (x)| M, n N and x [a, b].
[Gn(x) – f (x)]2 (M + M)2 = 4M2, x [a, b].
On applying Lebesgue bounded convergence theorem, we get
b b
Lim (G n f)2 = Lim (G n f)2 0

n
a a
2
Lim G n f 2
=0
n
Lim Gn f 2
n
or Lim f G n =0
n
for given > 0, no N, such that n no
f – Gn 2
<
Particularly for n = no.
f G no <
2
f–g 2
< (Taking G no = g)
Thus there exists a continuous function G no (x) g (x)
x (1 / n o )
= no f(t) dt, x [a, b] ,

x
which satisfies the given condition.
8.2 Summary
 A linear functional f on a normed space N 1 is said to be bounded if there is a constant

k > 0 such that
|f (x)| k x , x N1
 If x  p and f is bounded linear functional on  p , then f has the unique representation of

the form as an infinite series.
f (x) = xk f (e k )
k 1
 The norm of f  *p is given by

1
q
q
f = |f (e k )|
k 1

8.3 Keywords Notes
Bounded Linear Functional on L p-spaces: If x  p and f is bounded linear functional on  p , then

f has the unique representation of the form as an infinite series
f (x) = xk f(e k )
k 1
Bounded Linear Functional: A linear functional f on a normed space N 1 is said to be bounded if

there is a constant k > 0 such that
|f (x)| k x , x N1
Continuous Linear Functional: A linear functional f is continuous if given > 0 there exists >
0 so that
|f (x) – f (y)| whenever x – y .
Linear Functional: Let N1 be a normed space over a field R (or C). A mapping f : N 1 R (or C)
is called a linear functional on N 1 if f ( x + y) = f (x) + f (y), x, y N1 and , R (or C).
Norm: The norm of f  *p is given by

1
q
q
f = |f(ek )|
k 1
1. Account for bounded linear functionals on L p-space.
2. State and prove different continuous linear functional theorems.
3. Describe approximation by continuous function.
4. How will you explain norms of bounded linear functional on L p-space?

5. What is Isometric Isomorphism?
Books Rudin, Walter (1991), Functional Analysis, Mc-Graw-Hill Science/Engineering/

Math
Kreyszig, Erwin, Introductory Functional Analysis with Applications, WILEY 1989.
T.H. Hilderbrandt, Transactions of the American Mathematical Society . Vol. 36,
No. = 4, 1934.
Online links www.math.psu.edu/yzheng/m597k/m597kLIII4.pdf

www.public.iastate.edu/…/Royden_Real_Analysis_Solutions.pdf

Notes Unit 9: Measure Spaces
CONTENTS
Objectives
Introduction
9.1 Measure Space
9.1.1 Null Set in a Measure Space
9.1.2 Complete Measure Space
9.1.3 Measurable Mapping
9.2 Summary
9.3 Keywords

Objectives
 Define measure space.
 Define null set in a measure space.
 Understand theorems based on measure spaces.
 Solve problems on measure spaces.
Introduction
A measurable space is a set S, together with a non-empty collection, S, of subsets of S, satisfying

the following two conditions:
1. For any A, B in the collection S, the set1 A – B is also in S.
2. For any A1, A2, … S, Ai S.
The elements of S are called measurable sets. These two conditions are summarised by saying
that the measurable sets are closed under taking finite differences and countable unions.
9.1 Measure Space
Measurable Space: Let  be a -algebra of subsets of set X. The pair (X, ) is called a measurable
space. A subset E of X is said to be -measurable if E .
(a) If is a measure on a -algebra  of subsets of a set X, we call the triple (X, , u) a measure
space.
(b) A measure on a -algebra  of subsets of a set X is called a finite measure if m (X) < . In
this case (X, , ) is called a finite measure space.

Unit 9: Measure Spaces
(c) A measure on a -algebra  of subsets of a set X is called a -finite measure if there exists Notes
a sequence (En : n ) in  such that n  En = X and (En) < for every n . In this case
(X, , ) is called a -finite measure space.
(d) A set D  in an arbitrary measure space (X, , ) is called a -finite set if there exists a
sequence (Dn : n ) in such that Un  Dn = D and (Dn) < for every n .
Lemma 1: (a) Let (X, , ) be a measure space. If D  is a -finite set, then there exists an
increasing sequence (F n : n ) in  such that lim Fn D and (Fn) < for every n  and there
n
exists a disjoint sequence (G n : n ) in such that n 

Gn = D and (Gn) < for every n .
(b) If (X, , ) is a -finite measure space then every D  is a -finite set.
Proof 1: Let (X, , ) be a measure space. Suppose D  is a -finite set. Then there exists a
sequence (Dn : n ) in  such that Un  Dn = D and (Dn) < for every n . For each n ,
let Fn U kn 1 D k . Then (Fn : n ) is an increasing sequence in  such that lim Fn U n  Fn
n
n n
Un  Dn D and (Fn ) k 1
Dk
k 1
(D k ) for every n .
Let G1 = F1 and Gn = Fn\ U kn 11 Fk for n 2. Then (Gn: n ) is a disjoint sequence in  such that
U n  Gn U n  Fn D as in the proof of Lemma “let (E n : n ) be an arbitrary sequence in an
algebra of subsets of a set X. Then there exists a disjoint sequence (F n: n ) in such that
N N
(1) E F
n 1
n
n 1
n for every N ,
and
(2)  E  F ”.
n 
n
n 
n

n 1
(G1) = (F1) < and (G n ) Fn Fk (Fn ) for n 2. This proves (a).
k 1
2. Let (X, , ) be a -finite measure space. Then there exists a sequence (En : n ) in  such
that Un  En = X and (En) < for every n . Let D . For each n , let Dn = D En.
Then (Dn : n ) is a sequence in such that Un  Dn = D and m (Dn) (En) < for every
n . Thus D is a -finite set. This proves (b).
9.1.1 Null Set in a Measure Space
Definition: Given a measure on a -algebra  of subsets of a set X . A subset E of X is called a null

set with respect to the measure if E  and (E) = 0. In this case we say also that E is a null set
in the measure space (X, , ). (Note that is a null set in any measure space but a null set in a
measure space need not be .)
Theorem 1: A countable union of null sets in a measure space is a null set of the measure space.
Proof: Let (En : n ) be a sequence of null sets in a measure space (X, , ). Let E = Un 
En. Since
 is closed under countable unions,
we have E .

Notes By the countable subadditivity of on ,
we have (E) n 
(En) = 0.
Thus (E) = 0.
This shows that E is a null set in (X, , ).
9.1.2 Complete Measure Space
Definition: Given a measure on a -algebra  of subsets of a set X. We say that the -algebra 
is complete with respect to the measure if an arbitrary subset E0 of a null set E with respect to
is a member of  (and consequently has  (E0) = 0 by the Monotonicity of ). When  is
complete with respect to , we say that (X, , ) is a complete measure space.
Example: Let X = {a, b, c}. Then = { , {a}, {b, c}, X} is a -algebra of subsets of X. If we
define a set function on by setting ( ) = 0, ({a}) = 1, ({b, c}) = 0, and (X) = 1, then is
a measure on . The set {b, c} is a null set in the measure space (X, , ), but its subset {b} is not
a member of . Therefore, (X, , ) is not a complete measure space.
9.1.3 Measurable Mapping
Let f be a mapping of a subset D of a set X into a set Y. We write D (f) and R (f) for the domain of
definition and the range of f respectively. Thus
D (f) = D X,
R (f) = {y Y : y = f (x) for some x D (f)} Y.
For the image of D (f) by f, we have f (D (f)) = R (f). For an arbitrary subset E of y we define the
preimage of E under the mapping f by
F–1 (E) : = {x X : f (x) E} = {x D (f) : f (x) E}.
Notes
1. E is an arbitrary subset of Y and need not be a subset of R (f). Indeed E may be disjoint
from  (f), in which case f –1 (E) = . In general, we have f (f –1 (E)) E.
2. For an arbitrary collection C of subsets of Y, we let f –1 (C) : = {f–1 (E) : E C}.
Theorem 2: Given sets X and Y. Let f be a mapping with D (f) X and  (f) Y. Let E and E be
arbitrary subsets of Y. Then
1. f–1 (Y) = D (f),
2. f–1 (EC) = f–1 (Y\E) = f–1 (Y)\f–1 (E) = D (f) \ f–1 (E),
3. f–1 (U 
E )=U 
f–1 (E ),
4. f–1 ( 
E )= 
f–1 (E ).
Theorem 3: Given sets X and Y. Let f be a mapping with D (f) X and R (f) Y. If  is a -algebra
of subsets of Y then f–1 () is a -algebra of subsets of the set D (f). In particular, if D (f) = X then
f–1 () is a -algebra of subsets of the set X.

Proof: Let  be a -algebra of subsets of the set Y. To show that f–1 () is a -algebra of subsets of Notes
the set D (f) we show that D (f) f–1 (); if A f–1 () then D (f)\A f–1 (); and for any sequence
(An : n ) in f–1 () we have Un An f–1 (B).
1. By (1) of above theorem, we have D (f) = f–1 (Y) f–1 (B) since Y .
2. Let A f–1 (). Then A = f–1 () for some B . Since BC  we have f–1 (BC) f–1 (). On the
other hand by (2) of above theorem, we have f–1 (BC) = D (f)\f–1 (B) = D (f)\A. Thus D (f)\A
f–1 ().
3. Let (An : n ) be a sequence in f–1 (). Then An = f–1 (Bn) for some Bn  for each n .
Then by (3) of above theorem, we have
A  f
n 
n
n 
1
(B n ) f 1
B
n 
n f 1 ( ) ,
since B
n 
n ( ) .
Measurable Mapping
Definition: Given two measurable spaces (X, ) and (Y, ). Let f be a mapping with D (f) X and
 (f) Y. We say that f is a / measurable mapping if f–1 (B) for every B , that is, f–1 ()
.
Theorem 4: Given two measurable spaces (X, ) and (Y, ). Let f be a /-measurable mapping.
(a) If , is a -algebra of subsets of X such that , , then f is 1/-measurable.
(b) If 0 is a -algebra of subsets of Y such that 0 , then f is /0-measurable.
Proof: (a) Follows from f–1 ()  1 and (b) from f–1 (B0) f–1 () .
Composition of two measurable mappings is a measurable mapping provided that the two
measurable mappings from a chain.
Theorem 5: Given two measurable spaces (X, ) and (Y, ), where = () and  is arbitrary
collection of subsets of Y. Let f be a mapping with D (f)  and  (f) Y. Then f is a /-
measurable mapping of D (f) into Y if and only if f –1 () .
Proof: If f is a /-measurable mapping of D (f) into Y, then f–1 ()  so that f–1 () .
Conversely if f–1 () , then (f–1 () () = . Now by theorem,
“Let f be a mapping of a set X into a set Y. Then for an arbitrary collection C of subsets of Y, we
have (f–1 ()) = f–1 ( ().”
(f–1 () = f–1 ( ()) = f–1 (). Thus f–1 ()  and f is a /- measurable mapping of D (f).
Theorem 6: If X0 is a thick subset of a measure space (X, S, ), if S0 = S X0, and if, for E in S, 0
(E
X0) = (E), then (X0, S0, 0) is a measure space.
Proof: If two sets, E1 and E2, in S are such that E 1 X0 = E2 X0, then (E1 E2) Xo = 0, so that
(E1 E2) = 0 and therefore (E1) = (E2). In other words 0 is indeed unambiguously defined on
S0.
Suppose next that {Fn} is a disjoint sequence of sets in S 0, and let En be a set in S such that
F n = En X0, n = 1, 2, … .

Notes
If E n En {E i : 1 i n}, n 1, 2, , then
E n En X 0 = Fn Fi : 1 i n Fn
= Fn Fn = 0,
So that E n E n = 0, and therefore
n 1 0 (Fn ) = n 1 0 (E n ) n 1 0 (E n ) n 1 E n
=  n 1
En 0  n 1
Fn
In other word 0
is indeed a measure, and the proof of the theorem is complete.
9.2 Summary
 Let  be a -algebra of subsets of a set X. The pair (X, ) is called a measurable space. A
subset E of X is said to be -measurable if E .
 If is a measure on a -algebra of subsets of a set X, we call the triple (X, , ) a measure
space.
 A subset E of X is called a null set with respect to the measure if E  and (E) = 0.
 Two measurable spaces (X, ) and (Y, ). Let f be a mapping with D (f) X and  (f) Y.
We say that f is a /-measurable mapping if f–1 (B)  for every B , that is f–1 () .
9.3 Keywords
Complete Measure Space: Given a measure on a -algebra  of subsets of a set X. We say that
the -algebra  is complete with respect to the measure if an arbitrary subset E 0 of a null set
E with respect to is a member of  (and consequently has (E0) = 0 by the Monotonicity of ).
When  is complete with respect to , we say that (X, , ) is a complete measure space.
Measurable Mapping: Given two measurable spaces (X, ) and (Y, ). Let f be a mapping with D
(f) X and  (f) Y. We say that f is a / measurable mapping if f–1 (B) for every B , that
is, f–1 () .
Measurable Space: A measurable space is a set S, together with a non-empty collection, S, of

subsets of S.
Null Set in a Measure Space: A subset E of X is called a null set with respect to the measure if
E  and (E) = 0. In this case we say also that E is a null set in the measure space (X, , ).
Sigma Algebra:  is sigma algebra which establishes following relations:
(i) Ak  for all k implies A k 1

k 
(ii) A implies AC 
(iii) 

1. Let  be a -algebra of subsets of a set X and let Y be an arbitrary subset of X. Let

 = {A Y : A }. Show that is a -algebra of subsets of Y.
2. Let (X, , ) be a measure space. Show that for any E 1, E2  we have the equality:
(E1 E2) + (E1 E2) = (E1) + (E2).
Books Paul Halmos, (1950). Measure Theory. Van Nostrand and Co.
Bogachev, V.I. (2007), Measure Theory, Berlin : Springer
Online links planetmath.org/measurable space.html

mathworld.wolfram.com > Calculus and Analysis > Measure Theory

Notes Unit 10: Measurable Functions
CONTENTS
Objectives
Introduction
10.1 Measurable Functions
10.1.1 Lebesgue Measurable Function/Measurable Function
10.1.2 Almost Everywhere (a.e.)
10.1.3 Equivalent Functions
10.1.4 Non-Negative Functions
10.1.5 Characteristic Function
10.1.6 Simple Function

10.1.7 Step Function
10.1.8 Convergent Sequence of Measurable Function
10.1.9 Egoroff's Theorem
10.1.10 Riesz Theorem
10.2 Summary
10.3 Keywords
Objectives

 Understand measurable functions.
 Define equivalent functions and characteristic function.
 Describe Egoroff's theorem and Riesz theorem.
 Define simple function and step function.
Introduction
In this unit, we shall see that a real valued function may be Lebesgue integrable even if the
function is not continuous. In fact, for the existence of a Lebesgue integral, a much less restrictive
condition than continuity is needed to ensure integrability of f on [a, b]. This requirement gives
rise to a new class of functions, known as measurable functions. The class of measurable functions
plays an important role in Lebesgue theory of integration.

Unit 10: Measurable Functions
10.1 Measurable Functions Notes
10.1.1 Lebesgue Measurable Function/Measurable Function
Definition: Let E be a measurable set and R* be a set of extended real numbers. A function
f : E R* is said to be a Lebesgue measurable function on E or a measurable function on E iff the
set
E (f > ) = {x E : f (x) > } = f–1 { , )} is a measurable subset of E R.
Notes
1. The definition states that f is a measurable function if for every real number , the
inverse image of ( , ) under f is a measurable set.
2. The measure of the set E (f > ) may be finite or infinite.
3. A function whose values are in the set of extended real numbers is called an extended
real valued function.
4. If E = R, then the set E (f > ) becomes an open set.
Example: A constant function with measurable domain is measurable.
Sol: Let f be a constant function defined over a measurable set E so that f (x) = x E.
Then for any real number ,
E, if c
E (f > ) =
, if c
The sets E and are measurable and hence E (f > ) is measurable i.e. the function f is measurable.
Theorem 1: Let f and g be measurable real valued functions on E, and c is a constant. Then each of
the following functions is measurable on E.
(a) f c (b) cf
(c) f+g (d) f–g
(e) |f| (f) f2
(g) fg (h) f/g (g vanishes no where on E)
Proof: Let be an arbitrary real number.
(a) Since f is measurable and
E (f ± c > ) = E (f >  c),
the function f ± c is measurable.
(b) To prove c f is measurable over E.
If c = 0, then cf is constant and hence measurable because a constant function is measurable.

Notes Consider the case in which c 0, then
E f if c 0
c
E (cf > ) =
E f if c 0
c
Both the sets on R.H.S. are measurable.
Hence E (cf > ) is measurable and so cf is measurable c R.
(c) Before proving f + g is measurable, we first prove that if f and g are measurable over E then
the set E (f > g) is also measurable.
Now f > g a rational number r such that
f (x) > r > g (x)
Thus E (f > g) =  [(E (f

r Q
r) (E(g r))]
= an enumerable union of measurable sets.
= measurable set, since Q is an enumerable set.
Now, we shall prove that f + g is measurable over E. Let a be any real number.
Now E (f + g > a) = E (f > a – g) … (1)
Again, g is measurable
cg is measurable, c is constant.
( We know that if f is a measurable function and c is constant then cf is measurable)
a + cg is measurable a, c R
a – g is measurable by taking c = –1,
since f and a – g are measurable
E (f > a – g) is measurable.
E (f + g > a) is a measurable set.
f + g is a measurable function.
(d) To prove that f – g is measurable. Before proving f – g is measurable, we first prove that if
f and g are measurable over E then the set E (f > g) is also measurable.
Now f > g a rational number r, such that f (x) > r > g (x).
Thus E (f > g) =  [(E (f

r Q
r) (E(g r))]
= an enumerable union of measurable sets.
= measurable sets, since Q is an enumerable set.
Now we shall prove that f – g is measurable over E.
Let a be any real number.

Now E (f – g > a) = E (f > a + g) Notes
since g is measurable.
cg is measurable, c is constant.
a + cg is measurable a, c R
a + g is measurable by taking c = 1,
since f and a + g are measurable
E (f > a + g) is measurable.
E (f – g > a) is a measurable set.
f – g is a measurable function.
(e) To prove |f| is measurable.
We have
E if 0
E (|f| > ) =
[E (f )] [E(f )] if 0
[because we know that |x| > a x > a or x < –a]
since f is measurable therefore E (f > ) and E (f < – ) are measurable by definition.
Also we know that finite union of two measurable sets is measurable.
E (f > ) E (f < – ) is measurable.
E (|f| > ) is measurable.
|f| is measurable.
(f) To prove f2 is measurable.
E if 0
We have E (f2 > ) =
E (|f| )] if 0
But E (|f| > ) = [E(f )] [E(f )], if 0 ( |x| > a x > a or x < – a)
E if 0
E (f2 > ) =
[E (f )] [E(f )] if 0
But f is measurable over E.
E (f ) and E (f ) are measurable sets.
[E (f )] [E(f )] is measurable.
( union of two measurable sets is measurable)
E (f2 > ) is measurable because both the sets on RHS are measurable.
f2 is measurable over E.
(g) To prove fg is measurable.
Clearly, f + g and f – g are measurable functions over E.

Notes (f + g)2, (f – g)2 are measurable functions over E.
(f + g)2 – (f – g)2 is a measurable function over E.
1
(f g)2 (f g)2 is a measurable function over E.
4
fg is a measurable function over E.
(h) To prove f/g is measurable.
Let g vanish nowhere on E, so that
g (x) 0 x E.
1
exists.
g
1
Now we shall show that is measurable.
g
We have
E(g 0) if 0
1 1
E [E(g 0)] E g if 0
g
1
[E(g 0)] E(g 0) E g
1
Also finite union and intersection of measurable sets are measurable. Hence E is
g
measurable in every case.
1
Since f and are measurable.
g
1
(f) is measurable over E.
g
f
is measurable over E.
g
10.1.2 Almost Everywhere (a.e.)
Definition: A property is said to hold almost everywhere (a.e.) if the set of points where it fails to
hold is a set of measure zero.
Example: Let f be a function defined on R by
0, if x is irrational
f (x) =
1, if x is rational
Then f (x) = 0 a.e.

10.1.3 Equivalent Functions Notes
Definition: Two functions f and g defined on the same domain E are said to be equivalent on E,
written as f ~ g on E, if f = g a.e. on E, i.e. f (x) = g (x) for all x E – E1, where E1 E with m (E1)
= 0.
Theorem 2: If f, g : E R (E M) such that g (E).
Proof: Let be any real number and let E 1 = E (f > ) and E2 = E (g > )
Then E1 E2 = (E1 – E2) (E2 – E1)
= {x E : f (x) g (x)}
so that by given hypothesis we have
m (E1 E2) = 0.
This together with the fact that E1 is measurable
E2 is measurable.
Hence g (E).
10.1.4 Non-negative Functions
Definition: Let f be a function, then its positive part, written f+ and its negative part, written f –1, are
defined to be the non-negative functions given by
f+ = max (f, 0) and f–1 = max (–f, 0) respectively.
Note f = f+ – f–1
and |f| = f+ + f–1
Theorem 3: A function is measurable iff its positive and negative parts are measurable.
Proof: For every extended real valued function f, we may write
1
f+ = [f + |f|]
2
1
and f –1 = [|f| – f]
2
But f is measurable then |f| is measurable and hence positive and negative parts of f i.e. f + and
f– are measurable.
Conversely, let f+ and f–1 be measurable.
Since f = f+ – f–1
Since we know that if f and g are measurable functions defined on a measurable set E then f – g
is measurable on E.
Here f+ – f–1 is measurable.
and hence f is measurable.

Notes Theorem 4: If f is a measurable function and f = g a.e. then g is measurable.
Proof: Let E = {x : f (x) g (x)}.
Then m (E) = 0
Let be a real number.
{x : g (x) > } = {x : f (x) > } {x E : g (x) > } – {x E : g (x) }
since f is measurable, the first set on the right is measurable i.e. {x : f (x) > } is measurable.
The last two sets on the right are measurable since they are subsets of E and m (E) = 0.
Thus, {x : g (x) > } is measurable.
So, g is measurable.
Example: Give an example of function for which f is not measurable but |f| is measurable.
Sol: Let k be a non-measurable subset of E = [0, 1).
Define a function f : E R by
1 if x k
f (x) =
1 if x k
The function f is not measurable, since E (f > 0) (=k) is a non-measurable set. But |f| is measurable
as the set
E if 1
E (|f| > ) = if 1 is measurable
10.1.5 Characteristic Function
Definition: Let A be subset of real numbers. We define the characteristic function A

of the set A as
follows:
1 if x A
A
(x) = 0 if x A
Note The characteristic function A

of the set A is also called the indicator function of A.
Theorem 5: Show that the characteristic function A

is measurable iff A is measurable.
Proof: Let A
be measurable.
Since A = {x : A
(x) > 0} is measurable.
But A
is measurable, therefore the set {x : A
(x) > 0} is measurable.
A is measurable.
Conversely, let A be measurable and be any real number.

if 1
then E ( A
> ) = A if 0 1
E if 0

Every set on R.H.S. is measurable. Notes
Therefore E ( A
> ) is measurable.
Hence A
is measurable.
Note The above theorem asserts that the characteristic function of non-measurable sets
are non-measurable even though the domain set is measurable.
10.1.6 Simple Function
A real valued function is called simple if it is measurable and assumes only a finite number of
values.
If is simple and has the values 1

, 2
,… , then
n
= i Ai
i 1
where Ai = {x : (x) = i
}
and Ai Aj is a null set.
Thus we can always express a simple function as a linear combination of characteristic function.
Notes
(i) is simple A is are measurable.
(ii) sum, product and difference of simple functions are simple.
(iii) the representation of as given above is not unique.
But if is simple and { 1, 2

, ……, } is the set of non-zero values of f, then
n
= i Ai
i 1
}
This representation of is called the Canonical representation. Here A is are disjoint

and is are distinct and non-zero.
(iv) Simple function is always measurable.
10.1.7 Step Function
A real valued function S defined on an interval [a, b] is said to be a step function if these is a
partition a = xo < x1 … < xn = b such that the function assumes one and only one value in each
interval.

Notes
Notes
(i) Step function also assumes finite number of values like simple functions but the sets
{x : S (x) = Ci} are intervals for each i.
(ii) Every step function is also a simple function but the converse is not true.
1, x is rational
e.g. f : R R such that f (x) = 0, x is irrational
is a simple function but not step as the sets of rational and irrational are not intervals.
Theorem 6: If f and g are two simple functions then f+ g is also a simple function.
Proof: Since f and g are simple functions and we know that every simple function can be
expressed as the linear combination of characteristic function.
f and g can be expressed as the linear combination of characteristic function.
f= i Ai
i 1
and g= j Bj
j 1
where A is and B js are disjoint.

Ai = {x : f (x) = i
}
Bj = {x : g (x) = j}
The set Ek obtained by taking all intersections Ai Bj from a finite disjoint collection of measurable
sets and we may write
f = ak Ek
k 1
and g = bk Ek
k 1
where n = mm .
n n
f+ g = ak Ek bk Ek
k 1 k 1
= ( ak bk ) Ek
k 1
which is a linear combination of characteristic functions, therefore it is simple.

n Notes
Similarly fg = ak bk Ek
k 1
which is again a linear combination of characteristic function, therefore fg is simple.
Theorem 7: Let E be a measurable set with m (E) < and {fn} a sequence of measurable functions
converging a.e. to a real valued function defined on E. Then, given > 0 and > 0, there is a set
A E with m (A) < and an integer N such that |fn (x) – f (x)| < for all x E – A and all n N.
Proof: Let F be the set of points of E for which f n f. Then m (F) = 0 and f n (x) f (x) for all x
E – F = E1 (say). Then by the previous theorem for the set E 1, we get A1 E1 with m (A1) < and
an integer N such that
|fn (x) – f (x) | < for all n N and x E1 – A1.
We get the required result by taking
A = A1 F since m (F) = 0 and E – A = E1 – A1
Note Before proving this theorem first prove the previous theorem.
10.1.8 Convergent Sequence of Measurable Function
Definition: A sequence {fn} of measurable functions is said to converge almost uniformly to a

measurable function f defined on a measurable set E if for each > 0 there exists a measurable
set A E with m (A) < such that {fn} converges to f uniformly an E – A.
10.1.9 Egoroff's Theorem
Statement: Let E be a measurable set with m (E) < and {fn} a sequence of measurable functions
which converge to f a.e. on E. Then, given > 0 there is a set A E with m (A) < with that the
sequence {fn} converges to f uniformly on E – A.
Proof: Applying the theorem, “Let E be a measurable set with m (E) < and {fn} a sequence of
measurable function converging a.e. to real valued function f defined on E. Then given > 0 and
> 0 there is a set A E with m (A) < and an integer N such that
|fn (x) – f (x) | < for all x E – A and all n N”
with = 1, = /2, we get a measurable set
A1 E with m (A1) < /2 and a positive integer N1, such that
|fn (x) – f (x)|< 1 for all x N1
and x E1, where E1 = E – A1.
Again, taking = 1/2 and = n/22,
we get a measurable set A2 E1 with m (A2) < /22, and a positive integer N2 such that
1
|fn (x) – f (x)| < n N2 and x E2 where E2 = E1 – A2, and so on.
2

Notes At the pth stage, we get a measurable set
n
Ap Ep–1 with m (Ap) < and a positive integer Np such that
2p
1
|fn (x) – f (x)| <  n Np and x Ep where
p
Ep = Ep – 1 – Ap.
Let A = Ap ,
p 1
then m (A) m(A p )

p 1
n
But m (Ap) <
2p
n
m (A) <
p 1
2p
1 1
1 a 2 2 =1
But is a G.P. series so S =
2p 1 r 1 1
p 1 1
2 2
m (A) <
Also, E–A =E– A

p
p
=  (E
p
Ap )
=  (E
p
p 1 Ap )
= E
p
p
Let x E – A. Then x Ep p and so
1
|fn (x) – f (x)| < , n Np.
p
1
Let us choose p such that < , we get
p
|fn (x) – f (x)| < x E – A and n Np = N

Notes
Note Egoroff’s theorem can be stated as: almost every where convergence implies almost
uniform convergence.
10.1.10 Riesz Theorem
Let {fn} be a sequence of measurable functions which converges in measure to f. Then there is a
subsequence fnk which converges in measure to f a.e.
Proof: Let { n} and { n} be two sequences of positive real numbers such that n
0 as n and
n .
n 1
Let us now choose an increasing sequence {n k} of positive integers as follows.
Let n be a positive integer such that
m x E fn (x) f(x)
Since fn f in measure for a given 1

> 0 and 1
> 0, a positive integer n1 such that
m x E, fn1 (x) f(x) 1 1 , n1 n
Similarly, let n2 be a positive number such that m x E, fn 2 (x) f(x) 2 2 , n2 n 1 and so
on.
In general let nk be a positive number such that
m x : x E, fnk (x) f (x) k k and that n k nk 1 .
We shall now prove that the subsequence fnk converges to f a.e.
Let Ak =  x:x
i k
E, fni (x) f (x) i ,k N and A A
k 1
k .
Clearly, {Ak} is a decreasing sequence of measurable sets and m (A 1) < .

Therefore, we have
m (A) = lim m(A k )

k
But m (Ak) i 0 as k .
i k
Hence m (A) = 0.
Now it remains to show that {fn} converges to f on E – A. Let xo E – A.

Notes Then xo A k o for some positive integer ko.
or xo {x E : |fn (x) – f (x)| k

}, k ko
which gives |fn (xo) – f (xo)| < k

,k ko
But k
0 as k .
Hence lim fnk (xo ) f(xo ) .

k
10.2 Summary
 Let E be a measurable set and R* be a set of extended real numbers. A function f : E R* is

said to be a Lebesgue measurable function on E or a measurable function on E iff the set E
(f > ) = {x E : f (x) > } = f–1 {( , )} is a measurable subset of E R.
 A property is said to hold almost everywhere (a.e.) if the set of points where it fails to hold
is a set of measure zero.
 Two functions f and g defined on the same domain E are said to be equivalent on E, written
as f ~ g on E, if f = g a.e. on E, i.e. f (x) = g (x) for all x E – E1, where E1 E with m (E1) = 0.
 f+ = max (f, 0) and f –1 = max (–f, 0)
|f| = f+ + f–1
 Let A be subset of real numbers. We define the characteristic function A

of the set A as
follows:
1, if x A
A
(x) = 0, if x A
 A real valued function is called simple if it is measurable and assumes only a finite
number of values.
10.3 Keywords
Almost Everywhere (a.e.): A property is said to hold almost everywhere (a.e.) if the set of points
where it fails to hold is a set of measure zero.
Characteristic Function: Let A be subset of real numbers. We define the characteristic function
A
of the set A as follows:
1 if x A
A
(x) = 0 if x A
Egoroff’s Theorem: Let E be a measurable set with m (E) < and {fn} a sequence of measurable
functions which converge to f a.e. on E. Then given n > 0 there is a set A E with m (A) < n such
that the sequence {fn} converges to f uniformly on E – A.
Equivalent Functions: Two functions f and g defined on the same domain E are said to be
equivalent on E, written as f ~ g on E, if f = g a.e. on E, i.e. f (x) = g (x) for all x E – E1, where E1
E with m (E1) = 0.

Haracteristic Function: Let A be subset of real numbers. We define the characteristic function A
Notes
of the set A as follows:
1 if x A
A
(x) = 0 if x A
Lebesgue Measurable Function: A function f : E R* is said to be a Lebesgue measurable function

on E or a measurable function on E iff the set
E (f > ) = {x E : f (x) > } = f–1 { , )} is a measurable subset of E R.
Measurable Set: A set E is said to be measurable if for each set T, we have
m* (T) = m* (T E) + m* {T Ec)
Non-negative Functions: Let f be a function, then its positive part, written f+ and its negative
part, written f–1, are defined to be the non-negative functions given by
f+ = max (f, 0) and f–1 = max (–f, 0) respectively.
Riesz Theorem: Let {fn} be a sequence of measurable functions which converges in measure to f.
Then there is a subsequence fnk which converges to f a.e.
Simple Function: A real valued function is called simple if it is measurable and assumes only
a finite number of values.
If is simple and has the values 1

, 2
,… , then
n
= i Ai
i 1
}
and Ai Aj is a null set.
Step Function: A real valued function S defined on an interval [a, b] is said to be a step function
if these is a partition a = xo < x1 … < xn = b such that the function assumes one and only one value
in each interval.
Subsequence: If (xn) is a given sequence in X and (nk) is an strictly increasing sequence of positive
integers, then x nk is called a subsequence of (xn).
1. If f is a measurable function and c is a real number, then is it true to say that cf is measurable?
2. A non-zero constant function is measurable if and only if X is measurable comment.
3. Let Q be the set of rational number and let f be an extended real-valued function such that
{x : f (x) > } is measurable for each Q. Then show that f is measurable.
4. Show that if f is measurable then the set {x : f (x) = } is measurable for each extended real
number .
5. If f is a continuous function and g is a measurable function, then prove that the composite
function fog is measurable.

Notes 6. Show that
(i) A B A B
(ii) A B A B A B
(iii) Ac
1 A
Books Dudley, R.M. (2002). Real Analysis and Probability (2 ed.). Cambridge University
Press
Strichartz, Robert (2000). The Way of Analysis. Jones and Bortlett.
Online links mathworld.wolfram.com>calculus and Analysis > Measure theory

planetmath.org/Measurable functions.html
zeta.math.utsa.edu/~ mqr 328/class/real2/mfunct.pdf

Unit 11: Integration
Unit 11: Integration Notes
CONTENTS
Objectives
Introduction
11.1 Integration
11.1.1 The Riemann Integral
11.1.2 Lebesgue Integral of a Bounded Function over a Set of Finite Measure
11.1.3 The Lebesgue Integral of a Non-negative Function
11.1.4 The General Lebesgue Integral
11.2 Summary
11.3 Keywords
Objectives
 Define the Riemann integral and Lebesgue integral of bounded function over a set of
finite measure.
 Understand the Lebesgue integral of a non-negative function.
 Solve problems on integration.
Introduction
We now come to the main use of measure theory: to define a general theory of integration. The
particular case of the integral with respect to the Lebesgue measure is not, in any way, simpler
the general case, which will give us a tool of much wider applicability.
11.1 Integration
11.1.1 The Riemann Integral
Let f be a bounded real valued function defined on the interval [a, b] and let a = x 0 < x1 < … < xn= b
be a sub-division of [a, b].
Then for each sub-division we can define the sums
S = (xi xi 1 ) M i
i 1
and s = (xi xi 1 ) m i
i 1

Notes
where Mi = sup f (x) ,
xi 1 x xi
mi = inf f (x)
xi 1 x xi
We then define the upper Riemann integral of f by
R f (x) dx inf S
a
where the infimum is taken over all possible sub-divisions of [a, b].
Similarly, we define the lower Riemann integral
R f (x) dx sup S
a
The upper integral is always at least as large as the lower integral, and if the two are equal, we
say that f is Riemann integrable and we call this common value the Riemann integral of f.
It will be denoted by
R f (x) dx
a
Note By a step function we mean function s.t.
(x) = i
x [xi–1, xi]
for some sub-division of [a, b] and some set of constant i

then
b x1 xn
(x) dx = (x) dx (x) dx

a xo xn 1
x1 x2 xn
= 1 dx 2 dx n dx
xo x1 xn 1
= 1
(x1 – x0) + 2
(x2 – x1) + … + n
(xn–1 – xn)
= i (xi xi 1 ) … (1)
i 1
with this in mind, we see that
R f (x) dx = inf p (f)

a

n Notes
= inf M i (xi xi 1 )
i 1
= inf (x) dx for all step functions

a
(x) f (x)
Similarly R f (x) dx = sup Lp (f)

a
= sup m i (x i xi 1 )
i 1
= sup (x) dx for all step function

a
(x) f (x).
11.1.2 Lebesgue Integral of a Bounded Function over a Set of Finite

Measure
Characteristic Function
The function E
defined by
1 if x E
(x) =
E 0 if x E
is called the characteristic function of E.
Simple Function
A linear combination (x) = i Ei (x) is called a simple function if the sets Ei are measurable.
i 1
This representation of is not unique.
However, a function is simple if and only if it is measurable and assumes only a finite number
of values.
Canonical Representation
If is simple function and { 1, 2

, …, n
} the set of non-zero values of , then
= i Ei ,
i 1
where Ei = {x : (x) = i
}.

Notes This representation of is called the canonical representation. Here Ei’s are disjoint and i
’s are
finite in number, distinct and non-zero.
Elementary Integral
Definition: If vanishes outside a set of finite measure, we define the elementary integral of by
n
(x) dx i mE i when has the canonical representation.

i 1
= i Ei .
i 1
We sometimes abbreviate the expression for this integral . If E is any measurable set, we
define the elementary integral of over E by E .

E
If E = [a, b], then the integral will be denoted by .

[a , b] a
Theorem 1: Let and be simple functions which vanish outside a set of finite measure, then
(a b ) = a b and if a.e., then
Proof: Since and are simple functions.
Therefore these can be written in the canonical form
= i Ai
and = Bj Bj
j 1
where {Ai} and {Bj} are disjoint sequences of measurable sets and
Ai = {x : (x) = i
}
and Bj = {x : (x) = j}
The set Ek obtained by taking all intersections Ai Bj form a finite disjoint collection of measurable
sets. We may write
= ak Ek and
k 1
= bk Ek (where N = mm )
k 1

N N Notes
Now a +b = a ak Ek b bk Ek
k 1 k 1
= (aa k bbk ) Ek
k 1
which is again a simple function.
Since = ai m Ei
i
(a b ) = (aa k bbk ) m Ek (by definition)

k 1
N N
= a ak m Ek b bk m Ek
k 1 k 1
= a b
Now since a.e.
– 0 a.e.
We have proved that
(a b ) = a b
Put a = 1, b = –1 in the first part, we get
( ) =
Since – 0 a.e. is a simple function, by the definition of the elementary integral, we have
( ) 0
Theorem 2: Riemann integrable is Lebesgue integrable.
Proof: Since f is Riemann integrable over [a, b], we have
b b b
inf 1 (x) dx sup 1 (x) dx R f (x) dx

1 f f
a a a
where 1
and 1
vary over all step functions defined on [a, b].

Notes Since we know that every step function is a simple function,
b b
sup 1 (x) dx sup (x) dx

1 f f
a a
b b
and inf 1 (x) dx

inf (x) dx
f
1 f
a a
where and vary over all the simple functions defined on [a, b]. Thus from the above relation,
we have
b b b b
R f (x) dx sup (x) dx inf (x) dx R f (x) dx

f f
a a a a
b b
sup (x) dx inf (x) dx

f f
a a
b b
f (x) dx R f (x) dx
a a
Note The converse of this theorem is not true i.e.
A Lebesgue integrable function may not be Riemann integrable
e.g. Let f be a function defined on the interval [0, 1] as follows:
1, if x is rational
f (x) = 0, if x is irrational
Let us consider a partition p of an interval [0, 1].
U (p, f) = Mi xi
i 1
= 1 x1 + 1 x2 + …… + 1 xn = 1 – 0
= 1.
f dx = inf U (p, f) = 1 – 0 = 1.
0
f dx = sup L (p, f)
0
= sup {0 x1 + 0 x2 + …… + 0 xn}
=0

Notes
Thus f dx f dx
The function is not Riemann integrable.
Now for Lebesgue Integrability
Let A1 be the set of all irrational numbers and A 2 be the set of all rational numbers in [0, 1].
The partition P = {A1, A2} is a measurable partition of [0, 1] and mA 1 = 0, mA2 = 1.
L (p, f) = inf f(x) mA 1 inf f(x) mA 2

A1 A2
= 0 mA1 1 mA 2 = 1.
U (p, f) = sup f(x) mA1 sup f(x) mA2

A1 A2
= 0 mA1 1 mA 2 = 1.
sup L (p, f) = 1 inf U(p, f)

p p
f is Lebesgue integrable over [0, 1].
Theorem 3: If f and g are bounded measurable functions defined on the set E of finite measure,
then
(1) (af bg) a f b g

E E E
(2) If f = g a.e., then f g

E E
(3) If f g a.e., then f g

E E
Hence f |f|
(4) If A and B are disjoint measurable set of finite measure, then
f f f
A B A B
Proof of 1: Result is true if a = 0
Let a 0.
If is a simple function then so is a and conversely.
Hence for a > 0
af = inf a
a af
E E

Notes
= inf a ( a > 0)
f
E
= inf a
f
E
= a inf
f
E
= a f
E
Again if a < 0,
af inf a
= a af
E E
= sup a ( a < 0)
f
E
= sup a
f
E
= a sup
f
E
= a f
E
Therefore in each case
af = a f … (i)
E E
Now we prove that
(f g) = f g
E E E
Let 1 and 2
be two simple functions such that 1
> f and 2
g, then 1
+ 2
is a simple
function and 1
+ 2 f + g.
or f+g = 1
+ 2
(f g) ( 1 2 )
E E
But 1 2 = 1 2
E E E

Notes
(f g) 1 2
E E E
Since inf 1 = f
f 1
E E
and inf 2 = g
g 2
E E
(f g) f g … (2)
E E E
On the other hand if 1

and 2
are two simple functions such that 1
< f and 2
g. Then 1
+ 2
is
simple function and
1
+ 2
f + g,
or f+g 1
+ 2
(f g) ( 1 2 )
E E
But ( 1 2 ) = 1 2
E E E
(f g) 1 2
E E E
Since sup 1 = f
f 1
E E
and sup 2 = g
f 2
E E
(f g) f g … (3)
E E E
(f g) = f g
E E E
(af bg) = af bg
E E E
= a f b g from (i)
E E
Proof of 2: Since f = g a.e.
f – g = 0 a.e.

Notes Let F = {x : f (x) g (x)}
Then by definition of a.e., we have mF = 0 and F E.
(f g) = (f g) (f g) (f g)
E F (E F ) F E F
= (f – g) mF + (f – g) m (E – F)
= (f – g) . 0 + 0 . m (E – F) [ mF = 0 and f – g = 0 over E – F]
=0
(f g) = 0
E
f g =0 f g
E E E E
Note Converse need not be true
e.g. Let the functions f : [–1, 1] R and g : [–1, 1] R be defined by
2 if x 0
f (x) =
0 if x 0
and g (x) = 1 x.
1 1
Then f(x) dx = 2 = g(x) dx

1 1
But f g a.e.
In other words, they are not equal even for a single point in [–1, 1].
Proof of 3: f g a.e.
f–g 0 a.e.
Let be simple function,
=f–g
0 [ f – g = 0 a.e.]
0
E
(f g) 0
E
f g 0
E E

Notes
f g
E E
Since f |f|
f |f| … (1)
E E
Again – f |f|
f |f|
E E
or |f| f … (2)
E E
From (1) and (2) we get
|f| f |f|
E E E
f |f|
E E
Proof of 4: It follows from (3) and the fact that 1 mE .

E
Proof of 5: f = f A B
A B
Now A B = A B A B
where A and B are disjoint measurable sets i.e.
A B =
f = f( A B ) f A B
A B
= f A f B 0 [ A B = and m ( ) = 0]
= f f
A B
11.1.3 The Lebesgue Integral of a Non-negative Function
Definition: If f is a non-negative measurable function defined on a measurable set E, we define
f = sup h ,
h f
E E

Notes where h is a bounded measurable function such that
m {x : h (x) 0} < 
Theorem 4: Let f be a non-negative measurable function. Show that f 0 implies f = 0 a.e.
Proof: Let be any measurable simple function such that
f.
Since f = 0 a.e. on E
0 a.e.
(x) dx 0
E
Taking supremum over all those measurable simple functions f, we get
f dx 0 … (1)
E
Similarly let be any measurable simple function such that f
Since f = 0 a.e.
0 a.e.
(x) dx 0
E
Taking infimum over all those measurable simple functions f, we get
f dx 0 … (2)
E
f dx = 0.
E
Conversely, let f dx = 0
E
1
If En = x : f (x) , then
n
1
f dx > En (x) dx
n
E E
1 1
But En (x) dx = m En (By definition)
n n
E

1 Notes
f dx > m En
E
n
1
Or m En < f dx
n E
But f dx < 0
E
1
m En < 0
n
m En < 0
But m En 0 is always true
m En = 0
But {x : f (x) > 0} =  E n

n 1
and m En = 0
m  En =0
n 1
m {x : f (x) > 0} = 0
f = 0 a.e. on E
Theorem 5: Let f and g be two non-negative measurable functions. If f is integrable over E and
g (x) < f (x) on E, then g is also integrable over E, and
(f g) = f g.
E E E
Proof: Since we know that if f and g are non-negative measurable functions defined on a set E,
then
(f g) = f g
E E E
Since f = (f – g) + g,
therefore we have
f = (f g g) (f g) g … (1)
E E E E
Since the functions f – g and g are non-negative and measurable. Further, f being integrable over
E, f < (by definition)

E

Notes Therefore, each integral on the right of (1) is finite.
In particular, g < ,
E
which shows that g is an integrable function over E.
Since f = (f g) g
E E E
f g = (f g) .
E E E
11.1.4 The General Lebesgue Integral
For the positive part f+ of a function f, we define
f+ = max (f, 0)
and negative part f– by f–
f – = max (–f, 0)
and that f is measurable if and only if both f + and f– are measurable.
Note f = f+ – f–
and |f| = f+ + f–
Definition: A measurable function f is said to be Lebesgue integrable over E if f+ and f– are both
Lebesgue integrable over E. In this case, we define f f f .

E E E
Theorem 6: Let f and g be integrable over E, then
(a) The function of f is integrable over E, and cf c f.

E E
(b) Sum of two integrable functions is integrable i.e. the function f + g is integral over E, and
(f g) = f g
E E E
(c) If f g a.e., then f g.

E E
(d) If A and B are disjoint measurable sets contained in E, then
f = f f.
A B A B

Proof: (a) If c 0, then Notes
(cf)+ = cf+
(cf)– = cf
and if c < 0, then
(cf)* = (–c) . f
(cf)– = (–c) . f+
Since f is integrable so f + and f– are also integrable and conversely. Hence the result
follows.
(b) In order to prove the required result first of all we show that if f 1 and f2 are non-negative
integrable functions such that f = f1 – f2, then
f = f1 f2 … (1)
E E E
Since f = f+ – f–,
Also f = f1 – f2,
then f+ – f– = f1 – f2
f+ + f2 = f1 + f– … (2)
Also we know that if f and g are non-negative measurable functions defined on a set E,
then
(f g) = f g
E E E
Then from (2), we get
f f2 = f1 f
E E E E
f f = f1 f2 … (3)
E E E E
But f is integrable so f+ and f– are integrable i.e.
f = f f
Therefore (3) becomes
Hence f = f1 f2
E E E
which proves (1).
Now, if f and g are integrable functions over E, then
f+ + g+, f– + g– and f + g = (f+ + g+) – (f– + g–)
and also integrable functions over E.

Notes Furthermore f and g are integrable, it implies that |f| and |g| are integrable.
( A measurable function f is integrable over E if and only if |f| is integrable over E.)
Thus |f| + |g| is integrable over E.
( (f g) f g and by the definition of integrable)

E E E
Since |f + g| |+|f| + |g|
which shows that f + g is integrable.
Hence sum of two integrable functions is integrable.
Thus (f g) = (f g ) (f g )
E E E
= f g f g
E E E E
= f f g g
E E E E
= f g
E E
(c) f g a.e.
f–g 0 a.e.
g–f 0 a.e.
(g f) 0
E
Since g = f + (g – f) and f, g – f are integrable over E.
Then by the given hypothesis (g – f) – = 0 a.e.
then (g f) = 0,
E
(Since we know that if f = 0 a.e. then f = 0)

E
g f (g f) f (g f) (g f)
E E E E E E
becomes
g = f (g f) 0 f (g f) ( (g – f) 0)
E E E E E

Notes
f
E
g f f g
E E E E
(d) f = f A B
A B
= f A f B
= f f.
A B
Example: Let f be a non-negative integrable function. Show that the function F defined
by
F (x) = f (t) dt is continuous on R.
Solution: Since f is a non-negative integrable function, then given > 0 there is a > 0 such that for
every set A R with m (A) < , we have
f <
A
If xo R, then x R with |x – xo| < , we have
f (t) dt <
xo
f (t) dt f (t) dt <

xo
f (t) dt f (t) dt <

xo
x xo
f (t) dt f (t) dt <
|F (x) – F (xo)| <
Hence F is continuous at xo. Since xo R is arbitrary, F is continuous on R.

Notes 11.2 Summary
 Let and be simple functions which vanish outside a set of finite measure, then
(a b ) a b and if a.e., then
 A Lebesgue integrable function may not be Riemann integrable.
 Let A1 be the set of all irrational numbers and A2 be the set of all rational numbers in [0, 1].
 If f is a non-negative measurable function defined on a measurable set E, we define
f sup h
h f
,
E E
where h is a bounded measurable function such that
m {x : h (x) 0} <
 Let f and g be two non-negative measurable functions. If f is integrable over E and g (x) <
f (x) on E, then g is also integrable over E, and
(f g) f g
.
E E E
11.3 Keywords
Canonical Representation: If is simple function and { 1, 2

, …, } the set of non-zero values of
n
, then
= i Ei ,
i 1
where Ei = {x : (x) = i
}.
Characteristic Function: The function E

defined by
1 if x E
(x) =
E 0 if x E
is called the characteristic function of E.
Elementary Integral: If vanishes outside a set of finite measure, we define the elementary
n
integral of by (x) dx i mE i when has the canonical representation.

i 1
= i Ei .
i 1
Lebesgue Integrable: A measurable function f is said to be Lebesgue integrable over E if f + and f–
are both Lebesgue integrable over E. In this case, we define f f f .

E E E

n Notes
Simple Function: A linear combination (x) = i Ei (x) is called a simple function if the sets
i 1
Ei are measurable.
Simple Function: A linear combination (x) = i Ei (x) is called a simple function if the sets Ei
i 1
are measurable.
This representation of is not unique.
However, a function is simple if and only if it is measurable and assumes only a finite number
of values.
The Lebesgue Integral of a Non-negative Function: If f is a non-negative measurable function

defined on a measurable set E, we define
f = sup h ,
h f
E E
where h is a bounded measurable function such that
m {x : h (x) 0} < 
The Riemann Integral: Let f be a bounded real valued function defined on the interval [a, b] and
let a = x0 < x1 < … < xn= b be a sub-division of [a, b].
Then for each sub-division we can define the sums
S = (xi xi 1 ) M i
i 1
and s = (xi xi 1 ) m i
i 1
where Mi = sup f (x) ,

xi 1 x xi
mi = inf f (x)
xi 1 x xi
1. Prove that af a f real number a.

E E
2. If f is bounded real valued measurable function defined on a measurable set E of finite
measure such that a f (x) b, then show that amE f bmE.

E
3. If f and g are non-negative measurable functions defined on E  then prove that
(a) cf c f, c 0
E E

Notes
(b) (f g) f g
E E E
(c) f 0 f 0 a.e.
(d) If f g a.e. then f g

E E
4. If f is integrable over E, then show that |f| is integrable over E, and f |f|.
E E
5. Show that if f is a non-negative measurable function then f = 0 a.e. on E iff f = 0.

E
6. If f = 0 and f (x) 0 on E, then f = 0 a.e.

E
Books Erwin Kreyszig, Introductory Functional Analysis with Applications, John Wiley &
Sons Inc., New York, 1989
Walter Rudin, Real and Complex Analysis, Third McGraw Hill Book Co., New York,
1987
R.G. Bartle, The Elements of Integration and Lebesgue Measure, Wiley Interscience,
1995
Online links www.maths.manchester.ac.uk

www.uir.ac.za

Unit 12: General Convergence Theorems
Unit 12: General Convergence Theorems Notes
CONTENTS
Objectives
Introduction
12.1 General Convergence Theorems
12.1.1 Convergence almost Everywhere
12.1.2 Pointwise Convergence
12.1.3 Uniform Convergence, Almost Everywhere (a.e.)
12.1.4 Bounded Convergence Theorem
12.1.5 Fatou’s Lemma
12.1.6 Monotone Convergence Theorem
12.1.7 Lebesgue Dominated Convergence Theorem
12.2 Summary
12.3 Keywords
Objectives
 Understand bounded convergence theorem.
 State and prove monotone convergence theorem and Lebesgue dominated convergence
theorem.
 Solve related problems on these theorems.
Introduction
Convergence of a sequence of functions can be defined in various ways and there are situations
in which each of these definitions is natural and useful. In this unit, we shall study about
convergence almost everywhere, pointwise and uniform convergence. We shall also prove
bounded convergence theorem and monotone convergence theorem which are so useful in
solving problems on convergence. The dominated convergence theorem is one of the most
important results of Lebesgue’s integration theory. It gives a general sufficient condition for the
validity of proceeding to the limit of a sequence of functions under the integral sign. It is an
invaluable tool to study functions defined by integrals.
12.1 General Convergence Theorems
12.1.1 Convergence almost Everywhere
Let <fn> be a sequence of measurable functions defined over a measurable set E. Then <f n> is said
to converge almost everywhere in E if there exists a subset E 0 of E s.t.

Notes (i) fn(x) f (x), x E – E0 ,
and (ii) m (E0) = 0.
12.1.2 Pointwise Convergence
Let <fn> be a sequence of measurable functions on a measurable set E. Then <fn> is said to
converge “pointwise” in E, if a measurable function f on E such that
fn(x) f (x) x E or
lt fn (x) = f (x)
n
12.1.3 Uniform Convergence, Almost Everywhere (a.e.)
Let <fn> be a sequence of measurable functions defined over a measurable set E. Then the
sequence <fn> is said to converge uniformly a.e. to f, if a set E0 E s.t.
(i) m (E0) = 0 and
(ii) <fn> converges uniformly to f on the set E – E0.
12.1.4 Bounded Convergence Theorem
Theorem 1: State and Prove: Bounded Convergence Theorem
Statement: Let {fn} be a sequence of measurable functions defined on a set E of finite measure, and
suppose that there is a real number M such that |fn(x)| M n and x. If f (x) = lim fn (x) for each
n
x in E, then
f = lim fn
n
E E
Proof: Since f (x) = lim fn (x) and fn is measurable on E

n
E
f is also measurable on E
Let > 0 be given
Then measurable set A E with mA < and a positive integer N such that
4M
|fn(x) – f (x)| < n N and x E–A

2mE
fn f = (fn f)
E E E
(fn f)
E

Notes
= (fn f) (fn f) as (E – A) A=
E A A
1 fn f
2mE
E A A
m(E A) 2M 1
2mE
A
mE 2M mA as m (E – A) mE
2mE
< 2M
2 4M
=
2 2
Thus fn f <
E E
But was arbitrary
lim fn = f
n
E E
12.1.5 Fatou’s Lemma
If {fn} is a sequence of non-negative measurable functions and fn(x) f (x) almost everywhere on
a set E, then
f lim fn
n
E E
i.e. f lim inf fn

n
E E
Proof: Since integrals over sets of measure zero are zero.
Without loss of generality, we may assume that the convergence is everywhere. Let h be
a bounded measurable function with h f and h (x) = 0 outside a set E E of finite
measure.
Define a function hn by
hn(x) = Min. {h(x), fn(x)}
then hn(x) h(x) and hn(x) fn(x)
hn is bounded by the boundedness of h and vanishes outside E as x E–E h(x) = 0

hn(x) = 0 because

Notes Since hn = h or hn = fn
hn is measurable function on E
If hn = h, then hn h
If hn = fn < h f
then fn h as fn f
hn h
Thus hn h
Since hn (x) h (x) for each x E and {hn} is a sequence of bounded measurable functions
on E
By Bounded Convergence Theorem
h h h h lim h n
n
E E E E E E
as E = (E – E ) E & (E – E ) E =
lim h n
= n
E
lim fn as h n fn
n
E
lim fn as E E
n
E
h lim fn
n
E E
Taking supremum over all h f, we get
sup h h lim fn
n f n
E E E
f f lim fn
n
E E E
Remarks:
(1) If in Fatou’s Lemma, we take
1, n x n 1
fn(x) =
0, otherwise
with E = R
then f lim fn
n
E E
Thus in Fatou’s Lemma, strict inequality is possible.

(2) fn 0 x E is essential for Fatou’s Lemma Notes
However, if we take
1 2
n, x
fn(x) = n n
0, otherwise
with E = [0, 2]
Then f lim fn .
n
E E
12.1.6 Monotone Convergence Theorem
Statement: Let {fn} be an increasing sequence of non-negative measurable functions and let
f = lim fn . Then
n
f lim fn
n
Proof: Let h be a bounded measurable function with h f and h (x) = 0 outside a set E E of finite
measure
hn(x) = Min. {h(x), fn(x)}
then hn(x) h(x) and hn(x) fn(x)
hn is bounded by the boundedness of h and vanishes outside E as
x E–E h(x) = 0 hn(x) = 0 because fn(x) 0
Since hn = h or hn = fn
then fn h as fn f
hn h
Thus hn h
Since hn(x) h(x) for each x E and {hn} is a sequence of measurable functions on E
h h h h lim h n
n
E E E E E E
as E = (E – E ) E & (E – E ) E = .
= lim h n
n
E

Notes
= lim fn
n
E
lim fn as E E
n
E
h lim fn
n
E E
sup h lim fn
n
E E
f lim fn … (1)
n
E E
Since {fn} is monotonically increasing sequence and f n f
fn f
fn f
lim fn f … (2)
n
From (1) and (2), we have
f lim fn lim fn f
n n
f lim fn
n
Theorem 2: Let {un} be a sequence of non-negative measurable functions, and let f = un .

h 1
Then f un
h 1
Proof: Let fn = u1 + u2 + … + un = uj
j 1
then fn f
i.e. lim fn f
n
Let h be a bounded measurable function with h f and h(x) = 0 outside a set E E of finite
measure.
hn(x) = Min. {h (x), fn(x)}

then hn (x) h(x) and hn(x) fn(x) Notes
hn is bounded by the boundedness of h and vanishes outside E as
x E–E h(x) = 0 hn(x) = 0 because fn(x) 0.
Since hn = h or hn = fn
If hn = fn < h < f
then fn h as fn f
hn h
Thus hn h
Since hn(x) h(x) for each x E and {hn} is a sequence of measurable function on E
h h h h lim h n
n
E E E E E E
as E = (E – E ) E & (E – E ) E =
= lim h n
n
E
= lim fn
n
E
lim fn as E E
n
E
h lim fn
n
E E
sup h lim fn
h f n … (1)
E E
f lim fn
n
E E
Since {fn} is monotonically increasing sequence and f n f
fn f
fn f
lim fn f … (2)
n

Notes From (1) and (2), we have
f lim fn lim fn f
n n
f lim fn
n
= lim fn
n
= lim uj
n
j 1
= lim
n
uj
j 1
= uj
j 1
Hence f = un
n 1
Theorem 3: Let f be a non-negative function which is integrable over a set E. Then given >0
there is a > 0 such that for every set A E with mA < , we have
f <
A
Proof: If f is bounded function on E
Then positive real number M such that
|f (x)| M x E
For given > 0 = such that for every set A E with mA < , we have
M
f M M mA M. M
M
A A
i.e. f
A
Thus the result is true if f is a bounded function. So assume that f is not a bounded function on E.
Define a function fn on E by
f(x) if f(x) n
fn (x)
n otherwise

Then each fn is bounded and Notes
fn f at each point
Since {fn} is an increasing sequence of bounded functions such that f n f on E
By the monotone convergence theorem
lim fn f
n
E E
For given >0 a positive integer N such that
fn f for n N
2
E E
f fN
2
E E
f fN
2 2
E E
(f fN )
2
E
Choose
2N
If mA < , then we have
f (f fN ) fN
=
A A
(f fN ) fN
=
A A
(f fN ) N as fN N
E A
< N mA
2
< N
2
< N
2 2N
=
2 2

Notes
f
A
12.1.7 Lebesgue Dominated Convergence Theorem
Theorem 4: State and prove Lebesgue dominated convergence theorem
Statement: Let g be an integrable function on E and let {f n} be a sequence of measurable functions

such that |fn| g on E and lim fn = f a.e. on E. Then
n
f lim fn
n .
E E
Proof: Since we know that if f is a measurable function over a set E and there is an integrable
function g such that |f| g, then f is integrable over E. So clearly, each f n is integrable over E.
Also lim fn = f a.e. on E.

n
and |fn| g a.e. on E
|f| g a.e. on E.
Hence f is integrable over E.
Let { n} be a sequence of functions defined by n

= fn + g. Clearly, n
is a non-negative and integrable
function for each n.
Therefore, by Fatou’s Lemma, we have
(f g) lim (fn g)
n
E E
f lim fn … (1)
n
E E
Similarly, let { n} be a sequence of functions defined by n = g – fn. Clearly n

is a non-negative
and integrable function for each n. So, given by Fatou’s Lemma, we have
(g f) lim (g fn )
n
E E
g f g lim fn
n
E E E E
f lim fn
n
E E
f lim fn … (2)
E E
Hence from (1) & (2), we get
f = lim fn lim fn
E E E

Notes
But lim fn = lim fn lim fn
E E E
Hence f = lim fn .
E E
Corollary: Let {un} be a sequence of integrable functions on E such that u n converges a.e. on
n 1
E. Let g be a function which is integrable on E and satisfy ui g a.e. on E for each n. Then
i 1
u n is integrable on E and un un .
n 1 E n 1 n 1 E
Proof: Let ui fn .
i 1
Applying Lebesgue Dominated Convergence Theorem for the sequence {f n}, we get
un un
E n 1 n 1 E
Corollary: If f is integrable over E and {E i} is a sequence of disjoint measurable sets such that
Ei E , then
i 1
f f
E i 1 E
i
Proof: Since {Ei} is a sequence of disjoint measurable sets, we may write.
f= f Ei
i 1
The function f. Ei is integrable over E since f Ei |f| and |f| is integrable over E. Moreover
f Ei |f|, n N
i 1
Thus the conditions of above corollary are satisfied and hence
f = f Ei
E E i 1

Notes
= f Ei
i 1 E
= f
i 1 E
i
nx
Example: Show that the theorem of bounded convergence applies to fn(x) = ,0
1 n 2 x2
x 1.
nx
Sol: fn(x) =
1 n 2 x2
1
=
1
nx
nx
1
= 2
1
nx 2
nx
1
2
1 1
Thus a number such that |fn(x)| .
2 2
Hence it satisfies the conditions of bounded convergence theorem.
Now
1 1
nx
lim fn (x) dx = lim dx
n 1 n 2 x2
0 0
1
= lim log(1 n 2 x2 )  form
n 2n
[1/(1 n 2 x 2 )] 2nx 2
= lim [Using L’Hospital Rule]
n 2
nx 2
= lim
n 1 n 2 x2
1 2
x
= lim n 0
n 1
x2
n2
1 1
nx
and lim fn (x) dx = lim dx
n n 1 n 2 x2
0 0

1 Notes
= (0) dx 0
0
1 1
lim fn (x) dx = lim fn (x) dx

n n
0 0
This verifies the result of bounded convergence theorem.
Example: Use Lebesgue dominated convergence theorem to evaluate lim fn (x) dx ,

n
0
where
n 3 /2 x
fn(x) = , n = 1, 2, 3, … 0 x 1.
1 n 2 x2
n 3 /2 x
Solution: fn(x) =
1 n 2 x2
1 n 3/2 x2
=
x 1 n 2 x2
1
g(x), (say)
x
fn(x) g (x)
and g (x) L (0, 1],
Hence by Lebesgue Dominated Convergence Theorem.

1 1
lim fn (x) dx = lim fn (x) dx

n n
0 0
1
n 3 /2 x
= lim dx
n 1 n 2 x2
0
1
1 x
= lim dx
n n 1
0 x2
n2
= 0 dx = 0.
0
Example: If (fn) is a sequence of non-negative function s.t. fn f and fn f for each n, show
that
f lim fn

Notes Solution: From the given hypothesis it follows that
lim fn f … (1)
Also by Fatou’s Lemma, we have
f lim fn … (2)
Then from (1) and (2), we get
f lim fn lim fn f.
Hence f = lim fn lim fn lim fn .
n n
x
Example: If > 0, prove that Lim 1 x 1dx e x .x 1dx , where the integrals are
n n
o o
taken in the Lebesgue sense.
n n
x x x
Solution: If fn(x) = 1 .x 1
0 , then fn(x) g(x), where g(x) = e–x.x –1 recall Lim 1 e
n n n
Also g(x) L[0, ], hence by Lebesgue dominated convergence theorem, we get
Lim fn (x) dx = Lim fn (x) dx

n n
o o
n
x 1
= Lim 1 .x dx
n n
o
= e x .x 1
dx
o
Example: Show that if > 1,
1
x sin x
dx 0(n 1 ) as n .
1 (nx)
o
Solution: Consider the sequence <fn(x)> s.t.
nx sin x
fn(x) = , n = 1, 2, ………
1 (nx)
Obviously since > 1, and x [0, 1]

Notes
nx sin x
1
1 (nx)
If (x) = 1, x, then | fn (x) | (x), x.
Hence by dominated convergence theorem, we get
1 1 1
nx sin x nx sin x
Lim dx Lim dx (0) dx 0
n 1 (nx) n 1 (nx)
0 0 0
1
x sin x
Lim n dx =0
n 1 (nx)
0
1
x sin x
dx = 0 (n–1).
1 (nx)
0
2 2
n 2 xe n x
Example: Show that Lim dx = 0, if a > 0, but not for a = 0.
n 1 x2
a
putting nx u
Solution: If a > 0, and du ndx , we get
2 2 2
n 2 xe n x 2 ue u du ue u
dx ( na , ) du
1 x2 1 u 2 /n 2 1 u 2 /n 2
a na o
2
u.e u u2
Also ( na , ) u.e L [0, ]
1 (u 2 /n 2 )
2
u.e u
and Lim ( ne , ) 0 as ( , ) 0.
n 1 u 2 /n 2
Hence by Lebesgue dominated convergence theorem, we obtain
2 2 2
n 2 xe n x u.e u
Lim dx Lim ( na , ) du
n 1 x2 n 1 u 2 /n 2
a a
2
u.e u
= Lim ( na , ) du 0 dx 0 .
n 1 u 2 /n 2
a o
Now when a = 0,
2 2 1 2 2
n 2 xe n x n 2 xe n x
dx dx
1 x2 1 x2
o 0

Notes 1
1 n 2 x2
n 2 x.e dx (putting 1 in place of x2)
2
0
1 n 2 x2
1
1
= e
4 0 4
12.2 Summary
 Bounded Convergence Theorem: Let {fn} be a sequence of measurable functions defined on

a set E of finite measure, and suppose that there is a real number M such that |f n(x)| < M
n and all x. If f (x) = lim fn (x) for each x in E, then
n
f lim fn
n
E E
 Monotone Convergence Theorem: Let {fn} be an increasing sequence of non-negative

measurable functions and let f = lim fn . Then
n
f lim fn
n
 Lebesgue Dominated Convergence Theorem: Let g be an integrable function on E and let {f n}

be a sequence of measurable functions such that |fn| g on E and lim fn = f a.e. on E. Then
n
f lim fn
n
E E
12.3 Keywords
Convergence almost Everywhere: Let <fn> be a sequence of measurable functions defined over a
measurable set E. Then <fn> is said to converge almost everywhere in E if there exists a subset E0
of E s.t.
(i) fn(x) f (x), x E – E0 ,
and (ii) m (E0) = 0.
Convergence: Refers to the notion that some functions and sequence approach a limit under
certain conditions.
Fatou’s Lemma: If {fn} is a sequence of non-negative measurable functions and f n(x) f (x) almost
everywhere on a set E, then
f lim inf fn
n
E E
Pointwise Convergence: Let <fn> be a sequence of measurable functions on a measurable set E.

Then <fn> is said to converge “pointwise” in E, if a measurable function f on E such that
fn(x) f (x) x E or
lt fn (x) = f (x)
n

Uniform Convergence, Almost Everywhere (a.e.): Let <fn> be a sequence of measurable functions Notes
defined over a measurable set E. Then the sequence <f n> is said to converge uniformly a.e. to f,
if a set E0 E s.t.
(i) m (E0) = 0 and
(ii) <fn> converges uniformly to f on the set E – E0.
1. Show that we may have strict inequality in Fatou’s Lemma.
2. Let <fn> be an increasing sequence of non-negative measurable functions, and let f = lim f n.
Show that f lim fn .
Deduce that f u n , if un is a sequence of non-negative measurable functions and

n 1
f un .
n 1
3. State the Monotone Convergence theorem. Show that it need not hold for decreasing
sequences of functions.
4. Let {gn} be a sequence of integrable functions which converge a.e. to an integrable function
g. Let {fn} be a sequence of measurable functions such that |f n| gn and {fn} converges to f
a.e.
If g lim g n
n
then prove that f lim fn .

n
5. State and prove monotone convergence theorem.
Books G.F. Simmons, Introduction to Topology and Modern Analysis, New York: McGraw
Hill, 1963.
H.L. Royden, Real analysis, Prentice Hall, 1988.
Online links dl.acm.org

math.stanford.edu
www.springerlink.com

Notes Unit 13: Signed Measures
CONTENTS
Objectives
Introduction
13.1 Signed Measures
13.1.1 Signed Measure: Definition
13.1.2 Positive Set, Negative Set and Null Set
13.1.3 Hahn Decomposition Theorem
13.1.4 Hahn Decomposition: Definition
13.2 Summary
13.3 Keywords
Objectives
 Define signed measure.
 Describe positive and negative and null sets.
 Solve problems on signed measure.
Introduction
We have seen that a measure is a non-negative set function. Now we shall assume that it takes
both positive and negative values. Such assumption leads us to a new type of measure known as
signed measure. In this unit, we shall start with definition of signed measure and we shall prove
some important theorems on it.
13.1 Signed Measures
13.1.1 Signed Measure: Definition
Definition: Let the couple (X, ) be a measurable space, where  represents a -algebra of
subsets of X. An extended real valued set function
: [– , ]
defined on  is called a signed measure, if it satisfies the following postulates:
(i) assumes at most one of the values – or + .
(ii) ( ) = 0.

Unit 13: Signed Measures
(iii) If <An> is any sequence of disjoint measurable sets, then Notes
A
n 1
n
n 1
(A n ),
i.e., is countably additive.
From this definition, it follows that a measure is a special case of a signed measure. Thus, every
measure on  is a signed measure but the converse is not true in general, i.e. every signed
measure is not a measure in general.
If – < (A) < , for very A , then we say that signed measure is finite.
13.1.2 Positive Set, Negative Set and Null Set
Definition
(a) Positive Set: Let (X, ) be a measurable space and let A be any subset of X. Then A X is
said to be a positive set relative to a signed measure defined on (X, ), if
(i) A , i.e. A is measurable.
(ii) (E) 0, E A s.t. E is measurable.
Obviously, it follows from the above definition that:
(i) every measurable subset of a positive set is a positive set,
(ii) is a positive set w.r.t. every signed measure.
Also for A to be positive. (A) 0 is the necessary condition, but not in general sufficient
for A to be positive.
(b) Negative Set: Let (X, ) be a measurable space. Then a subset A of X is said to be a negative
set relative to a signed measure defined on measurable space (X, ) if
(i) A  i.e., A is measurable.

set A is negative w.r.t. , provided it is positive w.r.t. – .
(c) Null Set: A set A X is said to be a null set relative to a signed measure defined on
measurable space (X, ) is, A is both positive and negative relative to .
Thus, measure of every null set is zero.
Now, we know that a measurable set is a set of measure zero, iff every measurable subset of it
has measure zero. Thus, if A X is a null set relative to then (E) = 0, measurable subsets
E A. In other words.
A is a null set (E) = 0, measurable subsets E A.
Theorem 1: Countable union of positive sets w.r.t. a signed measure is positive.
Proof: Let (X, ) be a measurable space and let be a signed measure defined on (X, ). Let <An>
n
be a sequence of positive subsets of X, let A =  A and let B be any measurable subset of A.

i 1
i
Set n = B An A Cn 1  A C1 , n N.

Notes where A Cn (n 1, 2, 3 n 1) denotes complement of An(n = 1, 2, 3, … n – 1) with respect to X.
Now, we know that complement of a measurable set is also measurable so that each
A Cn (n 1, 2, 3 n 1) is measurable relative to . Again, intersection of countable collection of
measurable sets is also measurable. Hence B n is a measurable subset of the positive set An. Thus
(Bn) 0 (by the definition of positive set) … (i)
Obviously, the set Bn are disjoint and
if B = B
n 1
n , we get … (ii)
(B) = (B n ) … (iii)
n 1
In view of (i).
(B) 0.
Thus, we have
(1) A is measurable for
An is a positive set An is a measurable set
countable union A
n 1
n
is measurable,
A= A
n 1
n is measurable
(2) (B) 0, B A s.t. B is a measurable set.
Hence A is a positive set, by definition.

Theorem 2: Let (X, ) be a measurable space and let be a signed measure defined on (X, A). If B
is a measurable set with finite negative measure i.e., – < (B) < 0, then prove that B contains a
negative set A B with the property (A) < 0.
Proof: If B is itself a negative set, then we may take A = B and theorem is done. Therefore
consider the case when B is not a negative set. Then there must exist a measurable subset E 1 B
and a smallest positive integer n 1, s.t.
1
(E1) >
n1
 B = (B – E1) E1 and (B – E1) E1 = ,
(B) = (B – E1) + (E1) … (i)
or (B – E1) = (B) – (E1) … (ii)
Since (B) is finite, (i) implies that (B – E1) and (E1) are finite. Again (B) < 0, (ii) implies that
(B – E1) < 0.

Now, the set B – E1 is either negative or contains a subset of positive measure. If the set B – E 1 is Notes
a negative set, then we may take A = B – E1 and the theorem is done. So, suppose that B – E1 is not
a negative set. Then there must exist a measurable subset E 2 of B – E1 and a smallest positive
number n2 with a property
1
(E2) > .
n2
Since B = (B – E1 E2) (E2 E2),
and (B – E1 E2 ) (E1 E2 ) = ,
we have (B) = (B – E1 E2) + (E2 E2 )
or (B – E1 E2) = (B) – (E2 E2 )
= (B) – (E1) – (E2).
As before, (B – E1 E2) > 0 [ (B) < 0, (Er) > 0 for r = 1, 2]
Thus, B – E1 E2 is a set of negative measure, which is either a negative set or contains a subset
of positive measure. If B – E1 E2 is a negative set, then the theorem is done by taking B = A – E1
E2. Otherwise we repeat the above process.
On repeating this process, at some stage we shall get either a negative subset A B s.t. (A) < 0
or a sequence <Er> of disjoint measurable sets and a sequence <n r : r N> of positive integers s.t.
r 1
E
1
Er B– n and < (Er) <
n 1
nr
In first case, we have nothing to do. In the latter case, let
A =B– E
n 1
n or B = A E
n 1
n … (iii)
Then as before, it follows that
(B) = (A) + (E n ) .
n 1
1
> (A) + … (iv)
k 1
nk
[ change of suffix is in material]
Since (B) is finite and assumes at most one of the values – and , it follows from (iv) that
1
(A) is finite and the series is convergent.
k 1
nk
1
Then (A) < (B) –
k 1
nk
= a finite negative number

( (B) is a finite negative number)

Notes or (A) < 0.
Again, we know that difference of two measurable sets is measurable and enumerable union of
measurable sets is measurable therefore it follows from (iii) that A is a measurable set.
Now we shall prove that A is a negative set. Let E A be an arbitrary measurable set.
Since A =B– E
n 1
n ,
E =B– E
n 1
n .
Since nk , we can choose k so large that
1
(E) nk 1
Letting nk , we obtain
(E) 0.
Thus we have
(1) A is measurable.
(2) (E) 0, E A s.t. E is measurable.
Hence A is a negative set.
13.1.3 Hahn Decomposition Theorem
Theorem 3: Let be a signed measure on a measurable space (X, ). Then there exists a positive
set P and a negative set Q s.t.
P Q = and P Q = X.
Proof: Let (X, ) be a measurable space and let be a signed measure defined on a measurable
space (X, ). Since, by definition, assumes at most one of the values + or collection of all
negative subsets of X w.r.t. and let  be a collection of all negative subsets of X w.r.t. and let
k = inf { (E) : E )
(i) that there exists a sequence <En> in  such that
Lim (E n ) = k.
n
Let Q = E
n 1
n .
Since  is a family of negative sets, < E n> is a sequence of negative sets. Again, we know by
remark of theorem 1 that countable union of negative sets is negative, it follows that Q is a
negative subset of X so that
(Q) K.

Now, Q – En is a subset of Q, it follows that Notes
(Q – En) 0.
Since (Q – En) En =
and Q = (Q – En) En,
we have (Q) = (Q – En) + (En)
(Q) (En), n N and En B.
Therefore (Q) K. … (iii)
(ii) and (iii) (Q) = K – < k. … (iv)
Now we shall show that p = Q C, the complement of Q w.r.t. is a positive subset of X. Suppose
not, i.e. P is negative. Then E P s.t. E is measurable and (E) < 0. Now we know that if
– < (E) < 0, we get a negative set A E s.t. (A) < 0.
A, Q are distinct negative subsets of X
A Q is negative set
(A Q) K [using (i)]
(A) + (Q) K,
(A) + K K, [using (iv)]
(A) 0,
a contradiction, for (A) < 0
P = QC is a positive subset of X
Q is a negative subset of X.
Thus X = P Q, P Q= .
13.1.4 Hahn Decomposition: Definition
A decomposition of a measurable space X into two subsets s.t. X = P Q, P Q= ,
where P and Q are positive and negative sets respectively relative to the signal measure , is
called as Hahn decomposition for the signed measure . P and Q are respectively called positive
and negative components of X.
Notice that Hahn decomposition is not unique.
13.2 Summary
 Let the couple (X, ) be a measurable space, where represents a -algebra of subsets of
X. An extended real-valued set function
:  [– , ]
(i) assumes at most one of the values – or + .
(ii) ( ) = 0.
(iii) If <An> is any sequence of disjoint measurable sets, then is countably additive.

Notes  Let (X, ) be a measurable space and then A X is said to be a positive set relative to a
signed measure defined on (X, ) if
(i) A is measurable
 Let (X, ) be a measurable space. Then A X is said to be negative set relative to a signed
measure if
(i) A is measurable
 A X is said to be a null set relative to a signed measure defined on measurable space

(X, ) is: A is both positive and negative relative to .
13.3 Keywords
Hahn Decomposition: Definition: A decomposition of a measurable space X into two subsets s.t.
X = P Q, P Q = .
Negative Set: Let (X, ) be a measurable space. Then a subset A of X is said to be a negative set
relative to a signed measure defined on measurable space (X, ) if
(i) A  i.e., A is measurable.
Null Set: A set A X is said to be a null set relative to a signed measure defined on measurable
space (X, ) is, A is both positive and negative relative to .
Positive Set: Let (X, ) be a measurable space and let A be any subset of X. Then A X is said to
be a positive set relative to a signed measure defined on (X, ), if
(i) A , i.e. A is measurable.
Signed Measure: Let the couple (X, ) be a measurable space, where  represents a -algebra of
subsets of X. An extended real valued set function
: [– , ]
(i) assumes at most one of the values – or + .
(ii) ( ) = 0.
x2
1. If (E) = xe dx, then find positive, negative and null sets w.r.t. . Also give a Hahn
E
decomposition of R w.r.t. .
2. State and prove Hahn decomposition theorem for signed measures.
3. If is a measure and 1, 2 are the signed measures given by 1

(E) = (A E), 2
(E) = (B
E), where (A B) = 0, show that 1 2
.

4. Show that if 1
and 2
are two finite signed measures, then so is a 1 + b 2 where a, b are real Notes
numbers.
Books Bartle, Robert G., The Elements of Integration, New York – London – Sydney: John
Wiley and Sons
Cohn, Donald L. (1997) [1980], Measure Theory (reprint ed.), Boston – Based –
Stuttgart: Birkhauser Verlag
Online links www.maths.bris.ac.uk

www.planetmath.org/signedmeasure.html

Notes Unit 14: Radon-Nikodym Theorem
CONTENTS
Objectives
Introduction
14.1 Radon-Nikodym Theorem
14.1.1 Absolutely Continuous Measure Function
14.2 Summary
14.3 Keywords
Objectives
 Define Absolutely continuous measure function
 State Radon-Nikodym theorem
 Understand the proof of Radon-Nikodym theorem
 Solve problems on this theorem
Introduction
In mathematics, the Radon-Nikodym theorem is a result in measure theory that states that given
a measurable space (X, ), if a -finite measure on (X, ) is absolutely continuous with respect to
a -finite measure on (X, ), then there is a measurable function f on X and taking values in [0, ],
such that for any measurable set A.
The theorem is named after Johann Radon, who proved the theorem for the special case where
the underlying space is R N in 1913, and for Otto Nikodym who proved the general case in 1930.
In 1936 Hans Freudenthal further generalised the Radon-Nikodym theorem by proving the
Freudenthal spectral theorem, a result in Riesz space theory, which contains the Radon-Nikodym
theorem as a special case.
If Y is a Banach space and the generalisation of the Radon-Nikodym theorem also holds for
functions with values in Y, then Y is said to have the Radon-Nikodym property. All Hibert
spaces have the Radon-Nikodym property.
14.1 Radon-Nikodym Theorem
14.1.1 Absolutely Continuous Measure Function
Let (X, ) be a measurable space and let and be measure functions defined on the space (X, A).
The measure is said to be absolutely continuous w.r.t. if
(A) = 0 or | | (A) = 0, A  (A) = 0, and is denoted by  .

Unit 14: Radon-Nikodym Theorem
Notes
Notes
 If is -finite, the converse is also true.
 If and are signed measures on (X, ), then  if | |  | |.
Radon-Nikodym Theorem
Let (X, , ) be a -finite measure space. If be a measure defined on A s.t. is absolutely

continuous w.r.t. , then there exists a non-negative measurable function f on s.t.
(A) = f d , A .
A
The function f is unique in the sense that if g is any measurable function with the property
defined as above, then f = g almost everywhere with respect to .
Proof: To establish the existence of the function f, we shall use the following two Lemmas:
Lemma 1: Let E be a countable set of real numbers. Let for each a E there is a set Fa  s.t.
Fa Fb, whenever b < a i.e. <Fn> is a monotonic decreasing sequence of subsets of  corresponding
to the sequence <an> of real numbers in E. Then a measurable extended real valued function f
on X s.t.
f (x) a, x Fa ,
and f (x) a, x (X – Fa).
Proof: Let f (x) = inf {a : x Fa } x X and let, conventionally
inf {empty collection of real numbers} =
Now, x Fa f(x) a
x Fa x Fa for every b < a
f (x) a
Now, f (x) < a x Fb for some b < a
or {x : f (x) < a} = [F ] .

b a
b
Also x Fb f (x) b < a for some b < a.

Hence f is measurable.
Again, by definition of f, we observe that
f (x) a, x Fa ,
and f (x) a, x Fa .
Thus f is the required function.
Lemma 2: Let E be a countable set of real numbers. Let corresponding to each a E, there is a set
Fa  s.t.
(Fa – Fb) = 0 whenever b > a.
Then there exists a measurable function f with the property
x Fa f (x) a a.e.
and x (X – Fa) f (x) > a a.e.

Notes
Proof: Let P = {F
b a
a Fb } .
Evidently (P) = 0.
Let Fa = Fa P.
This Fa Fb = (Fa – Fb) – P = for a < b.
In view of Lemma 1, it follows that a measurable function f s.t.

f (x) a, x Fa
and f (x) a, x X – Fa
Thus we have
x Fa f(x) a a.e.,
except for x P.
x X Fa f(x) a a.e.,
Proof of the main theorem
At first, suppose that is finite.
( – a ) is a signed measure on  for each rational number a.
Let (Pa, Qa) be a Hahn decomposition for the measure ( – a ).
Let PO = X and QO = .
By the definition of Hahn decomposition theorem,
Pa Qa = X,
and Pb Qb = X.
Therefore, Qa – Qb = Qa – (X – Pb)
= Qa Pb.
Thus, ( – a ) (Qa – Qb) 0 … (i)

Similarly, we can prove that
( – b ) (Qa – Qb) 0 … (ii)
Let a < b, then from (i) and (ii), we have
(Qa – Qb) = 0.
Therefore, by Lemma (ii)
f (x) a, a.e. x Pa
and f (x) a, a.e. x Qa ,
where f is measurable
Since QO = , it follows that f is non-negative
Again, let A  be arbitrary.
Qr 1 Qr
Define Ar = A
no no

Notes
Qr
A = A– .
no
Evidently, A = A A
r o
r ,
where A is disjoint union of measurable sets.
(A) = (A ) + (A r ) .
r o
Qr 1 Qr
Obviously Ar
no no
r r 1
f(x) , x Ar
no no
r r 1
(A r ) fd (A r ) [by first mean value theorem]
no no
Ar
r r 1
Again (A r ) (A r ) (A r ) , we have
no no
1 1
(A r ) (A r ) fd (A r ) (A r ) … (iii)
no no
Ar
Now, if (A ) > 0, then g (A ) = 0, [ ( – a ) A is positive, a]
and (A ) = 0 if (A ) = 0 [  )]
In either case, (A ) = fd .
Ar
Adding the inequalities (iii) over r, we get
1 1
(A) (A) fd (A) (A r ).
no no
A
Since no is arbitrary and (A) is assumed to be finite, it follows that
(A) fd A .
A
To show that the theorem is true for -finite measure , decompose X into a countable union of
Xi of finite -measure. Applying the same argument as above for each X i, we get the required
function.
To show the second part, let g be any measurable function satisfying the condition,
(A) = fd A .
A

Notes For each n N, define
1
An= x X : f(x) g(x) 
n
1
and Bn = x X : g(x) f(x) .
n
1
Since f (x) – g (x) , x A n , we have by first mean value theorem
n
1
(f g)d (A n )
n
An
1
fd gd (A n )
n
An An
1 1
(An) – (An) (A n ) or 0 (A n )
n n
(An) 0.
Since (An) is always greater than equal to zero, we have (An) = 0.
Similarly, we can show that
(Bn) 0.
If C = {x X : f (x) g (x)}
= (A
n 1
n  B n ),
then (C) = 0 f = g a.e. on X w.r.t. .
Hence the theorem.
Theorem 1: If 1, 2
are -finite signed measures on (X, A) and 1
 , 2
, then
d( 1 2 ) d 1 d 2 d 1 d( 1 )
and
d d d d d
Proof: Since 1
, 2
are -finite and 1
 , 2
 , we have that 1
+ 2
is also -finite and
1
+ 2  .
Now for any A ,
( 1 + 2) (A) = 1
(A) + 2
(A)
d 1 d 2 d 1 d 2
= d d d
d d d d
A A A
d 1 2 d 1 d 2
d d
d d d
A A

Notes
d( 1 2 ) d 1 d 2
d d d
Prove the other result yourself.
Theorem 2: If is a -finite signed measures and is a -finite measure s.t.  , show that
d| | d
.
d d
+ –
Proof: Let = – with Hahn decomposition A, B.
d d d d
Then on A, and on B,
d d d d
d d d d( ) d| |
d d d d d .
Theorem 3: If be a -finite signed measure and , be -finite measures on (X, A) s.t.  ,

 : then show that
d d d
d d d
+ –
Proof: Since we may write = – and
d d( ) d d( )
, .
d d d d
we need to prove the above result for measures only.
d d
If f and g , (f, g are non-negative functions as obtained in Radon-Nikodym Theorem),
d d
then we need to prove that
(F) = fg d .
F
Let be a measurable simple function s.t.
= ai Ei ,
i 1
then d a i (E i F)
F i 1
= ai gd g d .
i 1 Ei F F
Let < > be a sequence of measurable simple function which converges to f, then
n
(F) = fd lim n d .
F F

Notes
= lim n gd fgd as n g fg
F F
d d du
= fg .
d d d
14.2 Summary
 Let (X, ) be a measurable space and let r and m be measure functions, defined on the space
(X, A). The measure is said to be absolutely continuous w.r.t. if
 Let (X, , ) be a -finite measure space. If Y be a measure defined on A s.t. is absolutely

continuous w.r.t. , then there exists a non-negative measurable function f s.t.
(A) = fd , A 
A
14.3 Keywords
Absolutely Continuous Measure Function: Let (X, ) be a measurable space and let and be
measure functions defined on the space (X, A). The measure is said to be absolutely continuous
w.r.t. if
Radon-Nikodym Theorem: Let (X, , ) be a -finite measure space. If be a measure defined on

A s.t. is absolutely continuous w.r.t. , then there exists a non-negative measurable function f
on s.t.
(A) = f d , A .
A
1
d d
1. Show that ,
d d
where and are -finite signed measures and  ,  .
2. If (E) = fd , , where fd exists, then find | | (E).

E E
3. State and prove Radon-Nikodym theorem.

14.5 Further Readings Notes
Book G.E. Shilov and B.L. Gurevich, Integral, Measure and Derivative: A Unified Approach,
Richard A. Silverman, trans. Dover Publications, 1978.
Online links www.math.ksu.edu/nnagy/real-an/4-04-rn.pdf

mathworld.woltram.com
www.csun.edu
pioneer.netserv.chula.ac.th

Notes Unit 15: Banach Space: Definition and Some Examples
CONTENTS
Objectives
Introduction
15.1 Banach Spaces
15.1.1 Normed Linear Space
15.1.2 Convergent Sequence in Normed Linear Space
15.1.3 Subspace of a normed Linear Space
15.1.5 Banach Space
15.2 Summary
15.3 Keywords
Objectives
 Know about Banach spaces.
 Define Banach spaces.
 Solve problems on Banach spaces.
Introduction
Banach space is a linear space, which is also, in a special way, a complete metric space. This
combination of algebraic and metric structures opens up the possibility of studying linear
transformations of one Banach space into another which have the additional property of being
continuous. The concept of a Banach space is a generalization of Hilbert space. A Banach space
assumes that there is a norm on the space relative to which the space is complete, but it is not
assumed that the norm is defined in terms of an inner product. There are many examples of
Banach spaces that are not Hilbert spaces, so that the generalization is quite useful.
15.1 Banach Spaces
15.1.1 Normed Linear Space
Definition: Let N be a complex (or real) linear space. A real valued function n : N R is said to
define, a norm on N if for any x, y N and any scalar (complex number) , the following
conditions are satisfied by n:
(i) n (x) 0, n (x) = 0, x = 0;
(ii) n (x + y) n (x) + n (y); and

Unit 15: Banach Space: Definition and Some Examples
(iii) n ( x) = | | n (x) Notes
It is customary to denote n (x) by
n (x) = x (read as norm x)
With this notation the above conditions (i) – (iii) assume the following forms:
(i) x 0, x = 0 x = 0;
(ii) x+y x + y ; and
(iii) x = x .
A linear space N together with a norm defined on it, i.e., the pair (N, ) is called a normed linear
space and will simply be denoted by N for convenience.
Notes
1. The condition (ii) is called subadditivity and the condition (iii) is called absolute
homogeneity.
2. If we drop the condition viz. x = 0 x = 0, then is called a semi norm (or pseudo
norm) or N and the space N is called a semi-normed linear space.
Theorem 1: If N is a normed linear space and if we define a real valued function d : N × N R by

d (x, y) = x – y (x, y N), then d is a metric on N.
Proof: We shall verify the conditions of a metric
(i) d (x, y) 0, d (x, y) = 0 x–y =0 x = y;
(ii) d (x, y) = x – y = (–1) (y – x) = |–1| y – x = y – x = d (y, x);
(iii) d (x, y) = x – y = x – z + z – y (z = N)
x–z + z–y = d (x, z) + d (z, y)
Hence, d defines a metric on N. Consequently, every normed linear space is automatically a

metric space.
Notes
1. The above metric has the following additional properties:
(i) If x, y, z N and is a scalar, then
d (x + z, y + z) = (x + z) – (y + z) = x – y = d (x, y).
(ii) d ( x, y) = x– y = (x – y)
= | | x – y = | | d (x, y).
2. Since every normed linear space is a metric space, we can rephrase the definition of
convergence of sequences by using this metric induced by the norm.

Notes 15.1.2 Convergent Sequence in Normed Linear Space
Definition: Let (N, ) be a normed linear space. A sequence (x n) is N is said to converge to an

element x in N if given > 0, there exists a positive integer n o such that
xn – x < for all n no.
If xn converges to x, we write Lim x n x.

n
or xn x as n
It follows from the definition that
xn x xn – x 0 as n
Theorem 2: If N is a normed linear space, then
x y x – y for any x, y N
Proof: We have
x = (x – y) + y
x–y + y
x – y x–y … (1)
Using (1), we have
–( x – y ) = y – x y–x
But y–x = (–1) (x – y) = |–1| x – y
Therefore
–( x – y ) x – y so that
x – y – x–y … (2)
x y x–y
15.1.3 Subspace of a Normed Linear Space
Definition: A subspace M of a normed linear space is a subspace of N consider as a vector space

with the norm obtain by restricting the norm of N to the subset M. This norm on M is said to be
induced by the norm on N. If M is closed in N, then M is called a closed subspace of N.
Theorem 3: Let N be a normed linear space and M is a subspace of N. Then the closure M of M is
also a subspace of N.
(Note that since M is closed, M is a closed subspace).
Proof: To prove that M is a subspace of N, we must show that any linear combination of
element in M is again in M. That is if x and y M , then x + y M for any scalars and .

Notes
Since x, y M , there exist sequences (xn) and (yn) in M such that
xn x and yn y,
By joint continuity of addition and scalar multiplication in M.
xn + yn x + y for every scalars and .
Since xn + yn M, we conclude that
x+ y M and consequently M is a subspace of N.
Notes
1. The scalars , can be assumed to be non-zero.
For if = 0 = , then
x+ y=0 M M
2. In a normed linear space, the smallest closed subspace containing a given set of
vectors S is just the closure of the subspace spanned by the set S. To see this, let S be
the subset of a normed linear space N and let M be the smallest closed subspace of N,
containing S. We show that M = [S] , where [S] is the subspace spanned by S.
By theorem, [S] is a closed subspace of N and it contains S.
Since M is the smallest closed subspace containing S, we have
M [S] .
But [S] M and M = M , we must have
[S] M = M so that [S] M.
Hence [S] = M.
Definition: A normed linear space N is said to be complete if every Cauchy sequence in N

converges to an element of N. This means that if xm – xn 0 as m, n , then there exists x
N such that
xn – x 0 as n .
15.1.5 Banach Space
Definition: A complete normed linear space is called a Banach space.
OR
A normed linear space which is complete as a metric space is called a Banach space.

Notes In the definition of a Banach space completeness means that if
xm – x n 0 as m, n , where (x n) N, then
ax N such that
xn – x 0 as n .
Note A subspace M of a Banach space B is a subspace of B considered as a normed linear

space. We do not require M to be complete.
Theorem 4: Every complete subspace M of a normed linear space N is closed.
Proof: Let x N be any limit point of M.
We have to show that x M.
Since x is a limit point of M, there exists a sequence (x n) in M and xn x as n .
But, since (xn) is a convergent sequence in M, it is Cauchy sequence in M.
Further M is complete (xn) converges to a point of M so that x M.
Hence M is closed.
Theorem 5: A subspace M of a Banach space B is complete iff the set M is closed in B.
Proof: Let M be a complete subspace of a Banach space M. They be above theorem, M is closed
(prove it).
Conversely, let M be a closed subspace of Banach space B. We shall show that M is complete.
Let x = (xn) be a Cauchy sequence in M. Then
xn x in B as B is complete.
We show that x M.
Now x M x M ( M being closed M= M)
Thus every Cauchy sequence in M converges to an element of M. Hence the closed sequence M
of B is complete. This completes the proof of the theorem.
Example 1: The linear space R of real numbers or C of complex numbers are Banach
spaces under the norm defined by
x = |x|, x R (or C)
Solution: We have
x = |x| > 0 and x = 0 |x| = x=0
Further, let z1, z2 C and let z 1 and z 2 be their complex conjugates, then
|z1 + z2|2 = (z1 + z2) (z 1 z2 )
= z1 z1 z1 z2 z2 z1 z2 z2
2 2
z1 2 z1 z 2 z2

2 2 Notes
= z1 2 z1 z 2 z2  z 1 , z2 z1 , z 2 z1 z 2
= (|z1| + |z2|)2
|z1 + z2| |z1| + |z2|
or z1 + z2 z1 + z2 ( x = |x|)
Also x = | x| = | | |x| = | | x
Hence all the conditions of normed linear space are satisfied. Thus both C or R are normed linear
space. And by Cauchy general principle of convergence, R and C are complete under the matrices
induced by the norm. So R and C are Banach spaces.
Example 2: Euclidean and Unitary spaces: The linear space R n and Cn of all n-tuples (x1,
x2 …, xn) of real and complex numbers are Banach spaces under the norm
1/2
n
x = |xi |2
i 1
[Usually called Euclidean and unitary spaces respectively].
Solution: (i) Since each |xi| 0, we have
x 0
and x = 0 |x i |2 = 0 xi = 0, i = 1, 2, …, n
i 1
(x1, x2, … xn) = 0
x=0
(ii) Let x = (x1, x2, …, xn)
and y = (y1, y2, … yn) be any two numbers of Cn (or Rn). Then
2 2
x+y = (x1, x2, …, xn) + (y1, y2, … yn)
2
= (x1+ x1), (x2 + y2), …, (xn + yn)
= |xi y i |2
i 1
|x i y i |(|xi | |y i |)
i 1
n n
|xi y i ||xi | |xi yi ||y i |

i 1 i 1
Usually Cauchy inequality for each sum, we get
2 1 1 1
n n 2 n 2 n 2
2 2 2
x+y 2
= |xi yi | |xi | |xi yi | |y i |
i 1 i 1 i 1 i 1

Notes = x+y x + x+y y
= ( x + y ) ( x + y ).
If x + y = 0, then the above inequality is evidently true.
If x + y 0, we can divide both sides by it to obtain
x+y x + y .
1 1
n 2 n 2
2 2
(iii) x = | xi | | | |x i |
i 1 i 1
=| | x .
This proves that Rn or Cn are normed linear spaces.
Now we show the completeness of Cn (or Rn).
Let < x1, x2, … xn > be a Cauchy sequence in Cn (or Rn). Since each xm is an n-tuple of complex (or
real) numbers, we shall write
xm = x1 , x 2 ,  , x n
(m ) (m ) (m )
So that x(m
k
)
is the kth coordinate of xm.
Let > 0 be given, since <xm> is a Cauchy sequence, there exists a positive integer m o, such that
,m mo xm x
2 2
xm x
x(m
i
)
x(i  ) 2
… (1)
i 1
x(m)
i x(i  ) 2
(i = 1, 2, ……, n)
x(m)
i x(i  )
Hence x(m
i
)
m 1
is a Cauchy sequence of complex (or real) numbers for each fixed but
arbitrary i.
Since C (or R) is complete, each of these sequences converges to a point, say 2 i in C (or R) so that
Lim x(m
i
)
= zi (i = 1, 2, …, n) … (2)
m
Now we show that the Cauchy sequence <xm> converges to the point z = (z1, z2, ……, zn) Cn (or
Rn).
To prove this let  in (1). Then by (2) we have
n
2
x(m
i
)
zi 2
i 1

xm – z 2
< 2 Notes
xm – z <
It follows that the Cauchy sequence <xm> converges to z Cn (or Rn).
Hence Cn or Rn are complete spaces and consequently they are Banach spaces.
15.2 Summary
 A linear space N together with a norm defined on it, i.e. the pair (N, ) is called a normed
linear space.
 Let (N, ) be a normed linear space. A sequence (x n) in N is said to converge to an element

x in N if given > 0, there exists a positive integer n o such that
xn – x < for all n no.
 If N is a normed linear space, then
x y x y for any x, y N.
 A normed linear space N is said to be complete if every Cauchy sequence in N converges

to an element of N.
 A complete normed linear space is called a Banach space.
15.3 Keywords
A Subspace M of a Normed Linear Space: A subspace M of a normed linear space is a subspace of

N consider as a vector space with the norm obtain by restricting the norm of N to the subset M.
If norm on M is said to be induced by the norm on N. If M is closed in N, then M is called a closed
subspace of N.
Banach Space: A complete normed linear space is called a Banach space.
Complete Normed Linear Space: A normed linear space N is said to be complete if every Cauchy
sequence in N converges to an element of N. This means that if xm – xn 0 as m, n , then
there exists x N such that
xn – x 0 as n .
Normed Linear: A linear space N together with a norm defined on it, i.e., the pair (N, ) is called
a normed linear space and will simply be denoted by N for convenience.
1. Let N be a non-zero normed linear space, prove that N is a Banach space {x : x = 1} is

complete.
2. Let a Banach space B be the direct sum of the linear subspaces M and N, so that B = M N.
If z = x + y is the unique expression of a vector z in B as the sum of vectors x and y in M and
N, then a new norm can be defined on the linear space B by z = x + y .
Prove that this actually is a norm. If B symbolizes the linear space B equipped with this
new norm, prove that B is a Banach space of M and N are closed in B.

Notes 15.5 Further Readings
Books Bourbaki, Nicolas (1987), Topological Vector Spaces, Elements of Mathematics, Berlin:
Springer-Verlag.
Beauzamy, Bernard (1985), Introduction to Banach Spaces and their Geometry (Second
revised ed.), North-Holland.
Online links mathword.wolfram.com?Calculus and Analysis>Functional Analysis

homepage.ntlworld-com/ivan.wilde/notes/fal/fal.pdf
www.math.ucdavis.edu/~ hunter/book.chs.pdf

Unit 16: Continuous Linear Transformations
Unit 16: Continuous Linear Transformations Notes
CONTENTS
Objectives
Introduction
16.1 Continuous Linear Transformation
16.1.1 Continuous Linear Functionals Definition
16.1.3 Norm of a Bounded Linear Functional
16.1.4 Equivalent Methods of Finding F
16.1.5 Representation Theorems for Functionals
16.2 Summary
16.3 Keywords
Objectives
 Understand continuous linear transformation
 Define bounded linear functional and norm of a bounded linear functional
 Understand theorems on continuous linear transformations.
Introduction
In this unit, we obtain the representation of continuous linear functionals on some of Banach
spaces.
16.1 Continuous Linear Transformation
16.1.1 Continuous Linear Functionals Definition
 Let N be a normed linear space. Then we know the set R of real numbers and the set C of
complex numbers are Banach spaces with the norm of any x R or x C given by the
absolute value of x. Thus with our previous notations, (N, R) or (N, C) denote respectively
the set of all continuous linear transformations from N into R or C.
 We denote the Banach space (N, R) or (N, C) by N* and call it by the conjugate space (or
dual space or adjoint space) of N.
 The elements of N* will be referred to as continuous linear functionals or simply functionals

on N.

Notes
Note The conjugate space (N*)* of N* is called the second conjugate space of N and shall
be denoted by N**. Also note that N** is complete too.
Theorem 1: The conjugate space N* is always a Banach space under the norm
f(x)
f = sup : x N, x 0 … (i)
x
= sup f(x) : x 1
= inf k, k 0 and f(x) k x x
Proof: As we know that if N, N are normed linear spaces, (N, N ) is a normed linear space. If
N is a Banach space, (N, N ) is Banach space. Hence (N, R) or (N, C) is a Banach space because
R and C are Banach spaces even if N is not complete.
Theorem 2: Let f be a linear functional on a normed linear space. If f is continuous at x o N, it

must be continuous at every point of N.
Proof: If f is continuous at x = xo, then
xn xo f (xn) f (x)
To show that f is continuous everywhere on N, we must show that for any y N,
yn y f (yn) f (y)
Let yn y as n
Now f (yn) = f (yn – y + xo + y – xo)
since f is linear.
f (yn) = f (yn – y + xo) + f (y) – f (xo) … (1)
As yn y y n – y + xo xo by hypothesis
Also f is continuous, f (yn – y + xo) f (xo) … (2)
From (1) and (2), it follows that
f (yn) f (y) as n .
f is continuous at y N and consequently as it is continuous everywhere on N.
Hence proved.
A linear functional on a normed linear space N is said to be bounded, if there exists a constant k
such that
f (x) K x x N.

Notes
Note
We may find many K’s satisfying the above condition for a given bounded function. If it
is satisfied for one K, it is satisfied for a K 1 > K.
Theorem 3: Let f be a linear functional defined on a normed linear space N, then f is bounded
f is continuous.
Proof: Let us first show that continuity of f boundedness of f.
If possible let f is continuous but not bounded. Therefore, for any natural number n, however
large, there is some point x n such that
|f (xn)| n || xn|| … (1)
xn
Consider the vector, yn = so that
n xn
1
yn = .
n
yn 0 as n
yn 0 in the norm.
Since any continuous functional maps zero vector into zero and f is continuous f (y n) f (0) = 0.
1
But |f (yn)| = f (xn) … (2)
n xn
It now follows from (1) & (2) that |f (yn)| > 1, a contradiction to the fact that f (yn) 0 as n .
Thus if f is bounded, then f is continuous.
Conversely, let f is bounded. Then for any sequence (x n), we have
|f (xn)| K || xn||  n = 1, 2, …, and K 0.
Let xn 0 as n then
f (xn) 0 f is continuous at the origin and consequently it is continuous everywhere.
Note The set of all bounded linear function on N is a vector space denoted by N*. As in the
case of linear operators, we make it a normed linear space by suitably defining a norm of
a functional f.
16.1.3 Norm of a Bounded Linear Functional
If f is a bounded linear functional on a normed space N, then the norm of f is defined as:
f(x)
|| f|| = sup … (1)
x 0 x

Notes We first note that the above norm is well defined. Since f is bounded, we have
|f (x)| M || x||, M 0.
f(x)
Let M be the set of real numbers M satisfying this relation. Then the set ;x 0 is
x
bounded above so that it must possess a supremum. Let it be f . So f is well defined and we
must have
f(x)
f x 0.
x
or |f (x)| f x .
Let us check that defined by (1) is truly a norm on N*:
If f, g N*, then
f(x) g(x)
f g = sup
x 0 x
f(x) g(x)
sup sup
x 0 x x 0 x
f g f g .
Similarly, we can see that f f .
16.1.4 Equivalent Methods of Finding F
If f is a bounded linear functional on N, then
|f (x)| M x ,M 0.
(I) f = inf {M : M M } where M is the set of all real numbers satisfying
|f (x)| M x ,
Since f M and M is the set of all non-negative real numbers, it is bounded below by
zero so that it has an infimum. Hence
f inf {M : M M} … (2)
f(x)
For x 0 and M M we have M. Since M is the only upper bound then from
x
definition (2), we have
f (x)
M sup = f for any M M.
x 0 x

Notes
Since M is bounded below by f , it has an infimum so that we have
inf M = inf {M : M M} f … (3)

M M
f = inf {M : M M}
(II) f = sup f (x)

x 0
Let us consider x 1. Then
f (x) f x f .
Therefore, we have
sup f (x) f . … (4)

x
Now by definition,
f (x)
f = sup
x 0 x
It follows from the property of the supremum that, given > 0, an x N such that
f (x )
>( f ) … (5)
x
Define
x
x . Then x is a unit vector.
x
Since x 1 x 1 , we have
1
sup f(x) f(x) f (x ) ( f ) [by (2)]
x 1 x
Hence > 0 is arbitrary, we have
sup f(x) > f … (6)

x 1
sup f(x) = f .
x 1
(III) f = sup f(x) .

x 1
Consider x = 1, we have
f (x) f x f

Notes So that
sup f(x) f … (7)

x 1
Now consider
f (x)
f = sup
x 0 x
By supremum property, given > 0, x 0
Such that |f (x )| > (||f|| – ) ||x ||
x
Define x .
x
Since f is continuous in ||x|| 1 and reaches its maximum on the boundary ||x|| = 1,
We get
1 .
sup f (x) f (x) f (x ) f
x 1 x
sup f (x) f .
x 1
The arbitrary character of yields that
sup f (x) f
x 1
… (8)
Hence from (7) and (8), we get
f = sup f (x) .
x 1
Note If N is a finite dimensional normed linear space, all linear functions are bounded
and hence continuous. For, let N be of dimension n so that any x N is of the form
n
i xi , where x1, x2, …, xn is a basis of N and 1

, 2
, …, n
are scalars uniquely determined
i 1
by the basis.
f (x) = i f (xi) so that

i 1
| f (x)| i |f (xi)| … (1)

i 1

We have from (1) by using the notation of the Zeroth norm in a finite dimensional space, Notes
| f (x)| x 0
f(xi ) … (2)
i 1
If f(x i ) = M, then from (2), we have

i 1
| f (x)| M x 0
.
Hence f is bounded with respect to 0

.
Since any norm on N is equivalent to 0

, f is bounded with respect to any norm on N.
Consequently, f is continuous on N.
16.1.5 Representation Theorems for Functionals
We shall prove, in this section, the representation theorems for functionals on some concrete
Banach spaces.
Theorem 4: If L is a linear space of all n-tuples, then (i)  np *  nq .
Proof: Let (e1, e2, …, en) be a standard basis for L so that any x = (x 1, x2, …, xn) L can be written
as
x = x1e1 + x2e2 + … + xnen.
If f is a scalar valued linear function defined on L, then we get
f (x) = x1 f (e1) + x2 f (e2) + … + xn f (en) … (1)
f determines and is determined by n scalars
yi = f (ei).
Then the mapping
y = (y1, y2, …, yn) f
where f (x) = xi y i is an isomorphism of L onto the linear space L of all function f. We shall
i 1
establish (i) – (iii) by using above given facts.
(i) If we consider the space
L =  p (1
n
p< ) with the pth norm, then f is continuous and L represents the set of all
continuous linear functionals on  np so that
L = p * .
n
Now for y f as an isometric isomorphism we try to find the norm for y’s.
For 1 < p < , we show that
 np * =  nq .

Notes For x  np , we have defined
1
n p
p
x = xi
i 1
n n
Now |f (x)| = xi yi xi yi
i 1 i 1
By using Hölder’s inequality, we get
1 1
n n p n q
p q
xi yi xi yi
i 1 i 1 i 1
so that
1 1
n q n p
q p
|f (x)| yi xi
i 1 i 1
Using the definition of norm for f, we get
1
n q
q
f yi … (2)
i 1
Consider the vector, defined by
q
yi
xi = , yi 0 and xi = 0 if yi = 0 … (3)
yi
Then
1
1 p
n n q p
p
p yi
x = xi … (4)
i 1 i 1
yi
Since q = p (q – 1) we have from (4),
1
n p
q
x = yi … (5)
i 1
Now
n n q
yi
|f (x)| = xi yi yi
i 1 i 1
yi
n
q
= yi , (By (3))
i 1

So that Notes
n
q
yi = |f(x)| f x … (6)
i 1

1
n 1
p
q
yi f
i 1
1
n q
q
yi f … (7)
i 1
Also from (2) and (7), we have

1
n q
q
f = yi , so that
i 1
y f is an isometric isomorphism.
Hence  np *  nq .
(ii) Let L =  n1 with the norm defined by x xi .

i 1
Now f defined in (1), above is continuous as in (i) and L here represents the set of continuous
linear functional on  n1 so that
L =  n1 * .
We now determine the norm of y’s which makes y f an isometric isomorphism.

Now,
|f (x)| = xi yi
i 1
xi yi
i 1
n n n
But xi y i max. y i x i so that f (x) max. y i xi .

i 1 i 1 i 1
From the definition of norm for f, we have

f = max. yi : i 1,2, ,n … (8)
Now consider the vector defined as follows:
If |yi| = 1max
i n
yi , let us consider vector x as

Notes
yi
xi = when |yi| = 1max yi
yi i n
and xi = 0 otherwise.
From the definition, xk = 0 k i. So that we have
yi
x = =1
y
Further |f (x)| = (x i y i ) = |yi|

i 1
Hence |yi| = |f (x)| f x
|yi| f or max. {|yi|} [ ||x|| = 1]
||f|| … (10)
||f|| = max. {|yi|} so that
y f is an isometric isomorphism of L to  n1 * .
Hence  n1 *  n .
(iii) Let L =  n with the norm
x = max {|xi| : i = 1, 2, 3, …, n}.
Now f defined in (1) above is continuous as in (1).
Let L represents the set of all continuous linear functionals on  n so that
L = n * .
Now we determine the norm of y’s which makes y f as isometric isomorphism
n n
|f (x)| = xi yi xi yi .
i 1 i 1
n n
But xi yi max( x i ) yi
i 1 i 1
Hence we have
n
| f (x) | yi x so that
i 1
f yi … (11)
i 1

Consider the vector x defined by Notes
yi
xi = when yi 0 and x i = 0 otherwise. … (12)
yi
yi
Hence x max 1.
yi
n n
and |f (x)| = xi yi yi .
i 1 i 1
Therefore yi f(x) f x f .
i 1
yi f … (13)
i 1
It follows now from (11) and (13) that f yi so that y f is an isometric

i 1
isomorphism.
Hence,  n *  n1 .
Theorem 5: The conjugate space of  p is  q , where
1 1
1 and 1 < p < .
p q
or  *p q .
p
Proof: Let x = (xn)  p so that xn . … (1)
n 1
Let  n = (0, 0, 0, …, 1, 0, 0, …) where 1 is in the mth place.
en  p for n = 1, 2, 3, …
We shall first determine the form of f and then establish the isometric isomorphism of  p onto
*
q .
By using (en), we can write any sequence (x 1, x2, …, xn, 0, 0, 0, …) in the form xk e k and
k 1
x xk e k = (0, 0, 0, …, x , x , …).
n+1 n+2
k 1

Notes 1
n p
p
Now x xk ek xk … (2)
k 1 k n 1
The R.H.S. of (2) gives the remainder after n terms of a convergent series (1).
1
p
p
Hence xk 0 as n … (3)
k n 1
x= xk ek . … (4)
k 1
Let f  *p and s n xk e k then

k 1
sn x as n . (Using (4))
since f is linear, we have
f (sn) = xk f(e k ) .
k 1
Also f is continuous and sn x, we have
f (sn) f (x) as n
f (x) = xk f(e k ) … (5)

k 1
which gives the form of the functional on  p .
Now we establish the isometric isomorphism of  p onto  q , for which we proceed as follows:
*
Let f (ek) = k
and show that the mapping
T : p
*
 q given by … (6)
, …) is an isometric isomorphism of  p onto  q .

*
T (f) = ( 1, 2
, …, k
First, we show that T is well defined.
For let x  p , where x = ( 1, 2

, …, n
, 0, 0, …)
g 1
k
sgn k , 1 k n
where = n k
k
0
| k| = | k|q – 1 for 1 k n.
(q 1)p
1 1
| k|p = = | k|q.  q p(q 1) q
k p q

q 1 q 1 Notes
Now k k
= k k sgn k k k sgn k
k k
= | k|q = | k|p (Using property of sgn function) … (7)
1
n p
p
x = k
k 1
1
n q
q
= k … (8)
k 1
Since we can write
x = k e k , we get
k 1
n n
f (x) = k f (e k ) k k
k 1 k 1
n q
f (x) = k ( Using (7)) … (9)

k 1
We know that for every x p
| f (x)| f x ,
which upon using (8) and (9), gives
1
n n p
q q
|f (x)| k f k
k 1 k 1
which yields after simplification.
1
n p
q
k f … (10)
k 1
since the sequence of partial sums on the L.H.S. of (10) is bounded, monotonic increasing, it
converges. Hence
1
n q
q
k f … (11)
k 1
so the sequence ( k) which is the image of f under T belongs to  q and hence T is well defined.
We next show that T is onto  q .
Let ( k)  q , we shall show that there is a g  *p such that T maps g into ( k).

Notes Let x  p so that
x = xk ek .
k 1
We shall show that
g (x) = xk k is the required g.

k 1
Since the representation for x is unique, g is well defined and moreover it is linear on  p . To
prove it is bounded, consider
n
|g (x) | = k xk k xk
k 1 k 1
1 1
p q
p q
xk k (Using Hölder’s inequality)
k 1 k 1
1
q
q
|g (x)| x k .
k 1
g is bounded linear functional on  p .
since ek  p for k = 1, 2, …, we get
g (ek) = k
for any k so that
Tg = ( k) and T is on  p onto  p .
*
We next show that
Tf f so that T is an isometry.
Since Tf  q , we have from (6) and (10) that
1
q
q
k = || Tf || || f || … (12)
k 1
Also, x  p x xk e k . Hence
k 1
f (x) = xk (e k ) xk k .
k 1 k 1
|f (x)| xk k
k 1

1 1
Notes
q p
q p
k xk (Using Hölder’s inequality)
k 1 k 1
1
q
q
or |f (x)| k x x p .
k 1
Hence, we have
1
f (x) q
q
sup k = Tf (Using (6))
x 0 x k 1
which upon using definition of norm yields.
f Tf … (13)
Thus f = Tf (Using (12) and (13))

From the definition of T, it is linear. Also since it is an isometry, it is one-to-one and onto.
Hence T is an isometric isomorphism of  p onto  p , i.e.

*
 *p q
Theorem 6: Let N and N be normed linear and let T be a linear transformation of N into N . Then
the inverse T–1 exists and is continuous on its domain of definition if and only if there exists a
constant m > 0 such that
m x T (x) x N. … (1)
–1
Proof: Let (1) holds. To show that T exists and is continuous.
Now T–1 exists iff T is one-one.
Let x1, x2, N. Then
T(x1) = T(x2) T (x1) – T(x2) = 0
T (x1 – x2) = 0
x1 – x2 = 0 by (1)
x 1 = x2
Hence T is one-one and so T –1 exists. Therefore to each y in the domain of T –1, there exists x in N
such that
T (x) = y T–1 (y) = x … (2)
Hence (1) is equivalent to
1
m T–1(y) y T–1(y) y
m
T–1 is bounded T–1 is continuous converse.
Let T–1 exists and be continuous on its domain T(N). Let x be an arbitrary element in N. Since
T–1 exists, there is y T(N) such that T–1 (y) = x T(x) = y.

Notes Again since T–1 is continuous, it is bounded so that there exists a positive constant k such that
T–1 (y) k y x k T(x)
1
m x T(x) where m = > 0.
k
Theorem 7: Let T : N N be a linear transformation. Then T is bounded if and only if T maps

bounded sets in N onto bounded set in N .
Proof: Since T is a bounded linear transformation,
T (x) k x for all x N.
Let B be a bounded subset of N. Then
x k1k x B.
We now show that T(B) is bounded subset of N .
From above we see that
T (x) k1 x B.
T (B) is bounded in N .
Conversely, let T map bounded sets in N into bounded sets in N . To prove that T is a bounded
linear transformation, let us take the closed unit sphere S [0, 1] in N as a bounded set. By
hypothesis, its image T (S[1, 0]) must be bounded set in N .
Therefore there is a constant k 1 such that
T (x) k1 for all x S [0, 1]
x
Let x be any non-zero vector in N. Then S[0,1] and so we get
x
x
T k1
x
T (x) k1 x .
Since this is true for x = 0 also, T is a bounded linear transformation.
16.2 Summary
 Let N be a normed linear space. Then we know the set R of real numbers and the set C of
complex numbers are Banach spaces with the norm of any x R or x C be the absolute
value of X. (N, R) or (N, C) denote respectively the set of all continuous linear
transformations from N into R or C.
 A linear functional on a normed linear space N is said to be bounded, if there exists a

constant k such that
|f (x)| k x x N.

 If f is a bounded linear functional on a normed space N, then the norm of f is defined as: Notes
f (x)
f = sup
x 0 x
16.3 Keywords
Bounded Linear Functional: A linear functional on a normed linear space N is said to be bounded,
if there exists a constant k such that
f (x) K x x N.
Continuous Linear Transformations: Let N be a normed linear space. Then we know the set R of
real numbers and the set C of complex numbers are Banach spaces with the norm of any x R or
x C given by the absolute value of x. Thus with our previous notations, (N, R) or (N, C)
denote respectively the set of all continuous linear transformations from N into R or C.
Norm of a Bounded Linear Functional: If f is a bounded linear functional on a normed space N,

then the norm of f is defined as:
f(x)
|| f|| = sup
x 0 x
Second Conjugate: The conjugate space (N*)* of N* is called the second conjugate space of N .
1. Prove that the conjugate space of  1 is  ,
i.e.  *1  .
2. Prove that the conjugate space of co is  1 .
or co* 1
1 1
3. Let p > 1 with = 1 and let g Lq (X).
p q
Then prove that the function defined by
F (f) = fg d for f Lp (X)

X
is a bounded linear functional on Lp (X) and
F = g q
4. Let N be any n dimensional normed linear space with a basis B = {x 1, x 2, ..., x n}. If
(r1, r2, ..., rn) is any ordered set of scalars, then prove that, there exists a unique continuous
linear functional f on N such that
f (xi) = ri for i = 1, 2, …, n
5. If T is a continuous linear transformation of a normed linear space N into a normed linear

space N , and if M is its null space, then show that T induces a natural linear transformation
T of N/M into N and that T = T .

Notes 16.5 Further Readings
Books JB Conway (1990), A Course in Functional Analysis.

E Hille (1957), Functional Analysis and Semigroups.
Online links pt.scribd.com/doc/86559155/14/Continuous-Linear-Transformations

www.math.psu.edu/bressan/PSPDF/fabook.pdf

Unit 17: The Hahn-Banach Theorem
Unit 17: The Hahn-Banach Theorem Notes
CONTENTS
Objectives
Introduction
17.1 The Hahn-Banach Theorem
17.1.1 Theorem: The Hahn-Banach Theorem – Proof
17.2 Summary
17.3 Keywords
Objectives
 State the Hahn-Banach theorem
 Understand the proof of the Hahn-Banach theorem
 Solve problems related to it.
Introduction
The Hahn-Banach theorem is one of the most fundamental and important theorems in functional
analysis. It is most fundamental in the sense that it asserts the existence of the linear, continuous
and norm preserving extension of a functional defined on a linear subspace of a normed linear
space and guarantees the existence of non-trivial continuous linear functionals on normed linear
spaces. Although there are many forms of Hahn-Banach theorem, however we are interested in
Banach space theory, in which we shall first prove Hahn-Banach theorem for normed linear
spaces and then prove the generalised form of this theorem. In the next unit, we shall discuss
some important applications of this theorem.
17.1 The Hahn-Banach Theorem
17.1.1 Theorem: The Hahn-Banach Theorem – Proof
Let N be a normed linear space and M be a linear subspace of N. If f is a linear functional defined
on M, then f can be extended to a functional f o defined on the whole space N such that
fo = f .
Proof: We first prove the following lemma which constitutes the most difficult part of this
theorem.

Notes Lemma: Let M be a linear subspace of a normed linear space N let f be a functional defined on M.
If xo N such that xo M and if Mo = M + [xo] is the linear subspace of N spanned by M and xo, then
f can be extended to a functional fo defined on Mo s.t.
fo = f .
Proof: We first prove the following lemma which constitutes the most difficult part of this
theorem.
Lemma: Let M be a linear subspace of a normed linear space N let f be a functional defined on M.
If xo N such that xo M and if Mo = M + [xo] is the linear subspace of N spanned by M and xo, then
f can be extended to a functional fo defined on Mo s.t.
fo = f .
Proof: The lemma is obvious if f = o. Let then f 0.
Case I: Let N be a real normed linear space.
Since xo M, each vector y in Mo is uniquely represented as
y=x+ xo, x M and R.
This enables us to define
fo : M o R by
fo (y) = fo (x + xo) = f (x) + ro,
where ro is any given real number … (1)
We show that for every choice of the real number ro, fo is not only linear on M but it also extends
f from M to Mo and
fo = f .
Let x1, y1 Mo. Then these exists x and y M and real scalars and such that
x1 = x + xo and y1 = y + xo,
Hence, fo (x1 + y1) = fo (x + xo + y + xo)
= fo (x + y + ( + ) xo)
= f (x + y) + ( + ) ro, ro is a real scalar … (2)
Since f is linear M, f (x + y) = f (x + y) … (3)
From (2) and (3) it follows after simplification that
fo (x1 + y1) = f (x) + ro + f (y) + ro
= fo (x + xo) + fo (y + xo)
= fo (x1) + fo (y1)
fo (x1 + y1) = fo (x1) + fo (y1) … (4)
Let k be any scalar. Then if y Mo, we have
fo (ky) = fo [k (x + xo)]
= fo (kx + k xo)
But fo (kx + k xo) = f (kx) + k ro
= k f (x) + k ro

Hence fo (ky) = k [f (x) + ro] = k fo (y) … (5) Notes
From (4) and (5) it follows that f o is linear on M o.
If y M, then = 0 in the representation for y so that
y = x.
Hence fo (x) = f (x) x M.
fo extends f from M to Mo.
Next we show that
fo = f .
If = 0 this is obvious. So we consider when 0. Since M is a subspace of Mo we then have
fo = sup {|fo (x)| : x Mo, x 1}
sup {|fo (x)| : x M, x 1}
= sup {|f (x)| : x M, x 1} ( fo = f on M)
= f .
Thus, fo f … (A)
So our problem now is to choose ro such that fo f .
Let x1, x2 M. Then we have
f (x2) – f (x1) = f (x2 – x1)
|f (x2 – x1)|
f (x2 + xo) – (x1 + xo)
f ( x2 + xo + – (x1 + xo) )
= f x2 + xo + f x1 + xo
Thus – f (x1) – f x1 + xo – f (x2) + f x2 + xo … (6)

Since this inequality holds for arbitrary x 1, x2 M, we see that
sup f (y) f y xo int f (y) f y xo

y M
y M
Choose ro to be any real number such that
sup f (y) f y xo ro
y M
inf f (y) f y xo
y M
From this, we get for all y M
sup {– f (y) – f ( y + xo )} ro
inf {– f (y) + f ( y + xo )}
x
Let us take y = in the above inequality, we have

Notes
x x ro
sup f f xo
y M
x x
inf f f xo … (7)
y M
If > 0 the right hand side of (7) becomes
1 1
ro f (x) f x xo which implies that
f (x) + ro = fo (x + xo) f x + xo
If z = x + xo Mo then we get from above
|fo (z)| f z … (8)
If < 0, then from L.H.S. of (7) we have
x x
f f xo ro
1 1 1 1
f (x) f x xo ro, since < 0, .
f (x) + ro f x+ xo
fo (z) f z for every z Mo … (9)
Replacing z by –z in (9) we get
– fo (z) f z , since fo is linear on Mo … (10)
Hence we get from (9) and (10).
|fo (z)| f z … (11)
Since f is functional on M, f is bounded.
Thus ( ) shows that fo is a functional on M o.
Since fo = sup {|fo (z)| : z Mo, z |}, it follows from ( ) that
fo f
We finally obtain from (A) and (B) that
fo = f
This power the lemma for real normed linear space.
Case II: Let N be a complex normed linear space.
Let N be a normed linear space over C and f be a complex valued functional on a subspace M of
N.
Let g = Re (f) and h = Im (f) so that we can write
f (x) = g (x) + i h (x). We show that g (x) and h (x) are real valued functionals.
f (x + y) = f (x) + f (y)

g (x + y) + i h (x + y) = g (x) + i h (x) + g (y) + i h (y) Notes
= g (x) + g (y) + i (h (x) + h (y))
Equating the real and imaginary parts, we get
g (x + y) = g (x) + g (y)
and h (x + y) = h (x) + h (y)
If R, then we have
f ( x) = g ( x) + i h ( x)
Since f is linear
f ( x) = f (x) = g (x) + i h (x)
f ( x) = g (x) and h ( x) = h (x)

(equating real and imaginary parts)
g, h are real linear functions on M.
Further |g (x)| |f (x)| f x
If f is bounded on M, then g is also bounded on M.
Similarly h is also bounded on M.
Since a complex linear space can be regarded as a real linear space by restricting the scalars to be
real numbers, we consider M as a real linear space. Hence g and h are real functional on real
space M.
For all x in M we have
f (i x) = i f (x) = i {g (x) + i h (x)}
or g (i x) + i h (i x) = – h (x) + i g (x)
Equating real and imaginary parts, we get
g (i x) = – h (x) and h (i x) = g (x)
Therefore we can express f (x) either only by g or only h as follows:
f (x) = g (x) – i g (i x)
= h (i x) + i h (x).
Since g is a real functional on M, by case I, we extend g to a real functional g o on the real space Mo
such that go = g . For x Mo, we define
fo (x) = go (x) – i go (i x)
First note that fo is linear on the complex linear space M o. Such that fo = f on M.
Now fo (x + y) = go (x + y) – i go (i x + i y)
= go (x) + go (y) – i go (i x) – i go (i y)
= fo (x) + fo (y).
Now for a, b R, we have
fo ((a + i b) x) = go (ax + i bx) – i go (– bx + i ax)
= a go (x) + b go (i x) – i (–b) go (x) – i a go (i x)
= (a + ib) {go (x) – i go (i x)}

Notes So that
fo ((a + i b) x) = (a + i b) fo (x)
fo is linear on M o and also go = g on M.
fo = f on M.
Now have to show that fo = f on Mo.
Let x Mo and fo (x) = rei
|fo (x) | = r = e–i rei = e–i fo (x) … (12)
Since fo (x) is linear,
e–i fo (x) = fo (e–i x) … (13)
So we get from (12) and (13) that
|fo (x) | = r = fo (r e–i x).
Thus the complex valued functional fo is real and so it has only real part so that
|fo (x) | = go (e–i x) |go (e–i x)|
But |go (e–i x)| go e–i x
= go x ,
We get |fo (x)| go x
Since go is the extension of g, we get
go x = g x f x
Therefore
|fo (x) f x so that from the definition of the norm of fo, we have
fo f
As in case I, it is obvious that f fo
Hence fo = f .
Theorem: The generalized Hahn-Banach Theorem for Complex Linear Space.
Let L be a complex linear space. Let p be a real valued function defined on L such that
p (x + y) p (x) + p (y)
and p ( x) = | | p (x) x L and scalar .
If f is a complex linear functional defined on the subspace M such that |f (x)| p (x) for x M,
then f can be extended to a complex linear functional to be defined on L such that |fo (x)| p (x)
for every x L.
Proof: We have from the given hypothesis that f is a complex linear functional on M such that
| f (x) p (x) x M.

Let g = Re (f) then g (x) |f (x)| p (x). Notes
So by the generalised Hahn-Banach Theorem for Real Linear space, can be extended to a linear
functional go on L into R such that go = g on M and g o (x) p (x) x L.
Define fo (x) = go (x) – i go (i x) for x L as in the Hahn-Banach Theorem, fo is linear functional on

L such that fo = f on M.
To complete the proof we have to prove that
|fo (x)| p (x) x L.
Let x L and fo (x) = r ei , r > o and real. Then
|fo (x)| = r = e– rei = e–i fo (x)
= fo (e–i x).
Since r = fo (e–i x), fo is real so that we can take
|fo (x)| = r = fo (e–i x) = go (e–i x) … (1)
Since go (x) p (x), go (e–i x) p (e–i x) for x L.

But p (e–i x) = |e–i | p (x) so that go (e–i x) p (x) … (2)
It follows from (1) and (2) that
|fo (x)| p (x)
Corollary 1: Deduce the Hahn-Banach theorem for normed linear spaces from the generalised
Hahn-Banach theorem.
Proof: Let p (x) = f x for x N.
We first note that p (x) 0 for all x N.
Then for any x, y N, we have
p (x + y) = f x+y
f ( x + y )
= f x + f y
= p (x) + p (y)
p (x + y) p (x) + p (y)
Also p ( x) = f x =| | f x = | | p (x).
Hence p satisfies all the conditions of the generalized Hahn-Banach Theorem for
Complex Linear space. Therefore a functional fo defined on all of N such that fo = f on M and
|fo (x)| p (x) = f x x N.
fo f … (3)
Since fo is the extension of f from a subspace M, we get
f fo … (4)
fo = f

Notes
Notes Let L be a linear space. A mapping p : L R is called a sub-linear functional on L if

it satisfies the following two properties namely,
(i) p (x + y) p (x) + p (y) x, y L (sub additivity)
(ii) p ( x) = p (x), 0 (positive homogeneity)
Thus p defined on L in the above theorems is a sub-linear functional on L.
Some Applications of the Hahn-Banach Theorem
Theorem: If N is a normed linear space and x o N, xo 0 then there exists a functional f o N* such
that
fo (xo) = xo and fo = 1.
Proof: Let M denote the subspace of N spanned by xo, i.e.,
M = { xo : any scalar}.
Define f : M F (R or C) by
f ( xo) = xo .
We show that f is a functional on M with f = 1.
Let x1, x2 M so that
x1 = 1
xo and x2 = 2
xo. Then
f (x1 + x2) = f ( 1
xo + 2
xo)
=( 1
+ 2
) xo
But ( 1
+ 2
) xo = 1
xo + 2
xo
= f (x1) + f (x2)
Hence f (x1 + x2) = f (x1) + f (x2) … (1)
Let k be a scalar (real or complex). Then if x M, then x = xo.
Now f (kx) = f (k xo) = k xo = k f (x) … (2)
If follows from (1) and (2) that f is linear.
Further, we note that since x o M with = 1, we get

f (xo) = xo .
For any x M, we get, | f (x)| = | | || xo|| = xo = x
|f (x) | = x
f is bounded and we have
|f(x)|
sup = 1 for x M and x 0.
x
So by definition of norm of a functional, we get
f = 1.

Hence by Hahn-Banach theorem, f can be extended to a functional f o N* such that fo (M) = f (M) Notes
and fo = f = 1, which in particular yields that
fo (xo) = f (xo) = xo and fo = 1.
Corollary 2: N* separates the vector (points) in N.
Proof: To prove the cor. it suffices to show that if x, y N with x y, then there exists a f N*
such that f (x) f (y).
Since x y x–y 0.
So by above theorem there exists a functional f N* such that
f (x – y) = f (x) – f (y) 0
and hence f (x) f (y).
This shows that N* separates the point of N.
Corollary 3: If all functional vanish on a given vector, then the vector must be zero, i.e.
if f (x) = 0 f N* then x = 0.
Proof: Let x be the given vector such that f (x) = 0 f N*.
Suppose x 0. Then by above theorem, there exists a function f N* such that
f (x) = x > 0, which contradicts our supposition that
f (x) = 0 f N*. Hence we must have x = 0.
17.2 Summary
 The Hahn-Banach Theorem: Let N be a normed linear space and M be a linear subspace of
N. If f is a linear functional defined on M, then f can be extended to a functional f o defined
on the whole space N such that
fo = f
 If f is a complex linear functional defined on the subspace M such that |f (x)| p (x) for
x M, then f can be extended to a complex linear function f o defined on L such that
| fo (x) | p (x) for every x L.
17.3 Keywords
Hahn-Banach theorem: The Hahn-Banach theorem is one of the most fundamental and important
theorems in functional analysis. It is most fundamental in the sense that it asserts the existence
of the linear, continuous and norm preserving extension of a functional defined on a linear
subspace of a normed linear space and guarantees the existence of non-trivial continuous linear
functionals on normed linear spaces.
Sub-linear Functional on L: Let L be a linear space. A mapping p : L R is called a sub-linear

functional on L if it satisfies the following two properties namely,
(i) p (x + y) p (x) + p (y) x, y L (sub additivity)
(ii) p ( x) = p (x), 0 (positive homogeneity)

Notes Thus p defined on L in the above theorems is a sub-linear functional on L.
The Generalized Hahn-Banach Theorem for Complex Linear Space: Let L be a complex linear
space. Let p be a real valued function defined on L such that
p (x + y) p (x) + p (y)
and p ( x) = | | p (x) x L and scalar .
1. Let M be a closed linear subspace of a normed linear space N and x o is a vector not in M.
Then there exists a functional f o N* such that
fo (M) = 0 and fo (xo) 0
2. Let M be a closed linear subspace of a normed linear space N, and let x o be a vector not in
M. If d is the distance from x o to M, then these exists a functional f o N* such that
fo (M) = 0, fo (xo) = d, and fo = 1.
3. Let M is a closed linear subspace of a normed linear space N and x o N such that xo M.
If d is the distance from xo to M, then there exists a functional fo N* such that fo (M) = 0, fo
1
(xo) = 1 and fo = .
d
4. Let N be a normed linear space over R or C. Let M N be a linear subspace. Then

M =N f N* is such that f (x) = 0 for every x M, then f = 0.
5. A normed linear space is separable if its conjugate (or dual) space is separable.
Books Walter Rudin, Functional Analysis (2nd ed.). McGraw-Hill Science/Engineering/

Math 1991.
Eberhard Zeidler, Applied Functional Analysis: Main Principles and their Applications,
Springer, 1995.
Online links mat.iitm.ac.in

www.math.ksu.edu
mizar.uwb.edu.pl/JFW/Vol5/hahnban.html

Unit 18: The Natural Imbedding of N in N**
Unit 18: The Natural Imbedding of N in N** Notes
CONTENTS
Objectives
Introduction
18.1 The Natural Imbedding of N into N**
18.1.1 Definition: Natural Imbedding of N into N**
18.1.2 Definition: Reflexive Mapping
18.1.3 Properties of Natural Imbedding of N into N**
18.2 Summary
18.3 Keywords
Objectives
 Define the natural imbedding of N into N**.
 Define reflexive mapping.
 Describe the properties of natural imbedding of N into N*.
Introduction
As we know that conjugate space N* of a normed linear space N is itself a normed linear space.
So, we can find the conjugate space (N*)* of N*. We denote it by N** and call it the second
conjugate space of N. Likewise N*, N** is also a Banach space. The importance of the space N**
lies in the fact that each vector x in N given rise to a functional F x in N** and that there exists an
isometric isomorphism of N into N**, called natural imbedding of N into N**.
18.1 The Natural Imbedding of N into N**
18.1.1 Definition: Natural Imbedding of N into N**
The map J : N N** defined by
J (x) = Fx x N,
is called the natural imbedding of N into N**.
Since J (N) N**, N can be considered as part of N** without changing its basic norm structure.
We write N N** in the above sense.

Notes 18.1.2 Definition: Reflexive Mapping
If the map J : N N** defined by
J (x) = Fx x N,
is onto also, then N (or J) is said to be reflexive (or reflexive mapping). In this case we write
N = N**, i.e., if N = N**, then N is reflexive.
Note Equality in the above definition is in the sense of isometric isomorphism under the
natural imbedding. Since N** must always be a complete normed linear space, no
incomplete space can be reflexive.
18.1.3 Properties of Natural Imbedding of N into N**
I. Let N be a normed linear space. If x N, then

x = sup {|f (x)|: f N* and f = 1}.
Using natural imbedding of N into N**, we have for every x N,
Fx (f) = f (x) and Fx = x .
Now, Fx = sup {|Fx (f)|} sup {|f(x)|, f N*}

f 1 f 1
therefore, x = sup {|f (x)|: f N*, f = 1}.
II. Every normed linear space is a dense linear subspace of a Banach space.
Let N be a normed linear space. Let
J:N N** be the natural imbedding of N into N**.
The image of the mapping is linear subspace J (N) N**. Let J(N) be the closure of N(N)
in N**.
Since N** is a Banach space, its closed subspace J(N) is also a Banach space. Hence if we
identity N with J(N), then J(N) is a dense subspace of a Banach space.
Theorem 1: Let N be an arbitrary normal linear space. Then each vector x in N induces a functional
Fx on N* defined by
Fx(f) = f(x) for every f N* such that Fx = x .
Further, the mapping J : N N** : J (x) = Fx for every x N defines and isometric isomorphisms
of N into N**.
Proof: To show that Fx is actually a function on N*, we must prove that F x is linear and bounded
(i.e. continuous).
We first show Fx is linear.

Let f, g N* and , be scalars. Then Notes
Fx ( f + g) = ( f + g) x = f (x) + g (x)
= Fx (f) + Fx (g)
Fx is linear
Fx is bounded.
For any f N*, we have
|Fx (f)| = |f (x)|
f x … (1)
Thus the constant x is bounded (in the sense of a bounded linear functional) for F x. Hence Fx is
a functional on N*.
We now prove Fx = x
We have Fx = sup {|Fx (f)| : f 1}
sup { F x : f 1} (Using (1))
Hence Fx x … (2)
To prove the reverse inequality we consider the case when x = 0. In this case (2) gives
Fo = 0 = 0.
But Fx = 0 always. Hence Fo = 0 i.e. Fx = x for x = 0.
Not let x 0 be a vector in N. Then by theorem (If N is a normal linear space and x o N, xo 0,
then there exists a functional f o N* such that
fo (xo) = xo and fo = 1.)
a functional f N* such that
f (x) = x and f = 1.
But Fx = sup {|Fx (f)| : f 1}
= sup { f (x) : f = 1}
and since x = f (x) sup {|f (x)| : f = 1}
we conclude that Fx x … (3)
[Note that since f (x) = x 0 we have f (x) = |f (x)|]
From (2) and (3); we have
Fx = x … (4)
Finally, we show that J is an isometric isomorphism of N into N**. For any x, y N and scalar.
Fx+y (f) = f (x + y) = f (x) + f (y)
= Fx (f) + Fy (f)
Fx+y (f) = (Fx + Fy) f … (5)
F x+y = Fx + Fy … (6)
Further, F x (f) = f ( x) = f (x) = ( Fx) (f)

Notes Hence F x
= Fx … (7)
Using definition of J and equations (6) and (7) we get
J(x + y) = Fx+y = Fx + Fy = J (x) + J (y) … (8)
and J( x)
= F x = Fx = J (x) … (9)
(8) and (9) J is linear and also (4) shows that J is norm preserving.
For any x and y in N, we have
J (x) – J (y) = Fx – Fy = Fx–y = x – y … (10)
Thus J preserve distances and it is an isometry. Also (10) shows that
J (x) – J (y) = 0 J (x – y) = 0 x–y=0
i.e. J (x) = J (y) x = y so that J is one-one.
Hence J defines an isometric isomorphism of N into N**. This completes the proof of the theorem.
Example 1: The space  np (1 p < ) are reflexive.

Solution: We know that if 1 p < , then
 np * =  np .
But  nq * =  np
Hence  np * * =  np
Similarly we have  n1 * * =  n1 for p = 1
and  n * * =  n for p =
So that  np spaces are reflexive for 1 p< .
Example 2: The space  p for 1 < p < are reflexive.
Sol: We know that if  * p  p and  * q p
 * *q p .
 p are reflexive for 1 < p < .
A similar result can be seen to hold for L p (X).
Example 2: If N is a finite dimensional normed linear space of dimension m, then N* also

has dimension m.
Solution: Since N is a finite dimensional normed linear space of dimension m then {x 1, x2, …, xm}
is a basis for N, and if ( 1 2 … m) is any set of scalars, then there exists a functional f on N such
that f (xi) = i, i = 1, 2, …m.

To show that N* is also of dimension m, we have to prove that there is a uniquely determined Notes
basis (f1, f2, …, fm) in N*, with fj (xi) = y.
By the above fact, for each i = 1, 2, …, m, a unique f j in N* exists such that fj (xi) = . We show now
ij
that {f1, f2, …, fn} is a basis in N* to complete our proof.
Let us consider 1
f1 + 2
f2 + …… m
fm = 0 … (1)
For all x N, we have 1

f1 (x) + 2
f2 (x) + …… + m
fm (x) = 0.
We have f (x j ) 0
j j j ij i for i = 1, 2, …, m, when x = x i.
j j
f1, f2, …… fm are linearly independent in N*.
Now let f (xi) = i

.
Therefore if x = i x i , we get
d
f (x) = 1
f (x1) + 2
f (x2) + …… + m
f (xm) … (2)
Further fj (x) = f (x1) + …… + ifj (xi) + …… +
1 j m j
f (xm)
fj (x) = j
f (x) = 1
f1 (x) + 2
f2 (x) + …… + m
fm (x)
= ( 1f + f + …… +
2 2 m
fm) (x)
(f1, f2, …… fm) spans the space.
N* is m-dimensional.
18.2 Summary
 The map J:N N** defined by
J (x) = Fx x N,
 If the map J : N N** defined by
J (x) = Fx x N,
is onto also, then N (or J) is said to be reflexive. In this case we write N = N**, i.e., if N = N**,
then N is reflexive.
 Let N be an arbitrary normal linear space. Then each vector x in N induces a functional F x
on N* defined by Fx (f) = f (x) for every f N* such that Fx = x .
18.3 Keywords
Natural Imbedding of N into N**: The map J : N N** defined by
J (x) = Fx x N,

Notes Reflexive Mapping: If the map J : N N** defined by
J (x) = Fx x N,
is onto also, then N (or J) is said to be reflexive (or reflexive mapping).
1. Let X be a compact Hausdorff space, and justify the assertion that C (X) is reflexive if X is
finite.
2. If N is a finite-dimensional normed linear space of dimension n, show that N* also has

dimension n. Use this to prove that N is reflexive.
3. If B is a Banach space, prove that B is reflexive B* is reflexive.
4. Prove that if B is a reflexive Banach space, then its closed unit sphere S is weakly compact.
5. Show that a linear subspace of a normed linear space is closed it is weakly closed.
Books G.F. Simmons, Introduction to topology and Modern Analysis. McGraw-Hill,

Kogakusha Ltd.
J.B. Conway, A Course in Functional Analysis. Springer-Verlag.
Online links www.mathoverflow.net/…/natural-embedding

www.tandfonline.com

Unit 19: The Open Mapping Theorem
Unit 19: The Open Mapping Theorem Notes
CONTENTS
Objectives
Introduction
19.1 The Open Mapping Theorem
19.1.1 Lemma
19.1.2 Proof of the Open Mapping Theorem
19.2 Summary
19.3 Keywords

Objectives
 State the open mapping theorem.
 Understand the proof of the open mapping theorem.
 Solve problems on the open mapping theorem.
Introduction
In this unit, we establish the open mapping theorem. It is concerned with complete normed
linear spaces. This theorem states that if T is a continuous linear transformation of a Banach
space B onto a Banach space B , then T is an open mapping. Before proving it, we shall prove a
lemma which is the key to this theorem.
19.1 The Open Mapping Theorem
19.1.1 Lemma
Lemma 1: If B and B are Banach spaces and T is a continuous linear transformation of B onto B ,
then the image of each sphere centered on the origin in B contains an open sphere centered on
the origin in B .
Proof: Let Sr and Sr respectively denote the open sphere with radius r centered on the origin in B
and B .
We one to show that T (Sr) contains same Sr .
However, since T (Sr) = T (r S1) = r T (S1), (by linearity of T).
It therefore suffices to show that T (S1) contains some Sr for then S , where = r2, will be contained
in T (Sr). We first claim that T(S 1 ) (the closure of T (S1)) contains some S r .

Notes If x is any vector in B we can by the Archimedean property of real numbers find a positive
integer n such that n > x , i.e., x Sn,
therefore
B = Sn 1
n
and since t is onto, we have
B = T (B)
=T S
n 1
n
=  T (S )
n 1
n
Now B being complete, Baire’s theorem implies that some T S n0 possesses an interior point
Z0. This in turn yields a point y 0 T S n0 such that y0 is also an interior point of T S n0 .
Further, maps j : B B and g : B B
defined respectively by j (y) = y = y – y0 and g (y) = 2 n0 y
where no is a non-zero scalars, are homeomorphisms as shown below f is one-to-one and onto.
To show f, f–1 are continuous, let yn B and yn y in B.
Then f (yn) = yn – y0 y – y0 = f (y)
and f–1 (yn) = yn + y0 y + y0 = f–1 (y)
Hence f and f–1 are both continuous so that is a homeomorphism.
Similarly g : B B : g (x) = 2n0y is a homeomorphism for, g is one-to-one, onto and bicontinuous

for n0 0.
Therefore we have
(i) f (y0) = 0 = origin in B is an interior point of f T S n .
(ii) f T S n0 = f T S n0
= T S n0 y0
T S 2n  y0 T S n0
0
(iii) T S 2n = T 2 n0 S 1 2n 0 T S 1
0
= g (T(S 1 )) g (T(S 1 ))
= 2 n0 T(S 1 )

Combining (i) – (iii), it follows that origin is also an interior point of (T(S 1 )) . Consequently, Notes
there exists > 0 such that

S T(S 1 )
This justifies our claims.
We conclude the proof of the lemma by showing that
S /3 T(S 1 ) , i.e., S T(S 3 )
Let y B such that y < . Then y T(S 1 ) and therefore there exists a vector x 1 B such that
x1 < 1, y – y1 < /2 and y1 = T (x1)
We next observe that
S /2 T(S1/2 ) and y – y1 S /2
Therefore there exists a vector x 2 B such that
1
x3 , y y1 y2 and y2 = T (x2)
2 22
Continuing this process, we obtain a sequence (x n) in B such that
1
xn , yn T(x n ) and y (y 1 y2  yn )
2n 1
22
Let sn = x1 + x2 + … + xn, then
sn = x1 + x2 + … + xn
x1 + x2 + … + xn
1 1 1
< 1 
2 22 2n 1
1
2 1
2n
<2
Also for n > m, we have
s n – sm = sm+1 + … + xn
xm+1 + … + xn
1 1
< 
2m 2n 1
1 1
1
2m 2n m
= 1 (summing the G.P.)
1
2
1 1
=
2m 1 2n m 1
0 as m, n

Notes Thus (sn) is a Cauchy sequence in B and since B is complete, a vector x B such that
sn x and therefore
x = lim s n lim s n 2 3,
n n
i.e., x S3.
It now follows by the continuity of T that
T (x) = T lim s n
n
= lim
n
T(s n )
= lim (y 1 y2  yn )
n
=y
Hence y T (S3)
Thus y S y T (S3). Accordingly
S T (S3)
This completes the proof of the lemma.
Note If B and B are Banach spaces, the symbol S (x; r) and S (x; r) will be used to denote
open spheres with centre x and radius r in B and B respectively. Also Sr and S r will denote
these spheres when the centre is the origin. It is easy to see that
S (x; r) = x + Sr and Sr = r S1
For, we have
y S (x; r) y–x <r
z < r and y – x = z, z Sr
y = x + z and z < r
y x + Sr
Thus S (x; r) = x + Sr
x
and Sr = {x : x < r} = x : 1
r
= {r . y y < 1}
= r S1
Thus Sr = r S1
Now we prove an important lemma which is key to the proof of the open mapping theorem.

19.1.2 Proof of the Open Mapping Theorem Notes
Statement: If T is a continuous linear transformation of a Banach space B onto a Banach space B ,

then T is an open mapping.
Proof: Let G be an open set in B. We are to show that T (G) is an open set in B
i.e. if y is any point of T (G), then there exists an open sphere centered at y and contained in T (G).
y T (G) y = T(x) for some x G.
x G, G open in B there exists an open sphere S (x; r) with centre x and radius r such that
S (x; r) G.
But as remarked earlier we can write S (x; r) = x + Sr, where Sr is open sphere of radius r centered
at the origin in B.
Thus x + Sr G … (1)
By lemma (2) (prove it),
T (Sr) contains some S r1 . Therefore
S (y; r1) = y + S r1
y + T (Sr)
= T (x) + T (Sr)
= T (x + Sr)
T (G), (Using (1))
since x + Sr = S (x; r) G.
Thus we have shown that to each y T (G), there exists an open sphere in B centered at y and
contained in T (G) and consequently T (G) is an open set.
Theorem 1: Let B and B be Banach spaces and let T be an one-one continuous linear transformation
of B onto B . Then T is a homeomorphism.
In particular, T–1 is automatically continuous.
Proof: We know that a one-to-one continuous open map from B onto B is a homeomorphism.
By hypothesis T : B B is a continuous one-to-one onto mapping.
By the open mapping theorem, T is open. Hence T is a homeomorphism. Since T is

homeomorphism, T– exists and continuous from B to B so that T– is bounded and hence
T–1 (B , B).
Cor. 1: Let B and B be Banach spaces and let T (B, B ). If T : B B is one-to-one and onto, there
are positive numbers m and M such that
m x T (x) M x .

Notes Proof: By the theorem,
T : B B is a homeomorphism. So that T and T –1 are both continuous and hence bounded. Hence
by theorem,
Let N and N be normed linear spaces. Then N and N are topologically isomorphic if and only
if there exists a linear transformation T of N onto N and positive constants m and M such that
m x T (x) M x , for every x N.
constants m and M such that
m x T (x) M x .
Theorem 2: If a one-to-one linear transformation T of a Banach space B onto itself is continuous,

then its inverse T–1 is continuous.
Proof: T is a homeomorphism (using theorem of B onto B. Hence T–1 is continuous.
Note The following examples will show that the completeness assumption in the open
mapping theorem and theorem can neither be omitted in the domain of definition of T nor
in the range of T.
Example 1: Let C [0, 1] be the set of all continuous differentiable function on [0, 1]. We
know that C [0, 1] is an incomplete space with the norm
f = sup {|f (x)| : 0 x 1}
But it is complete with respect to the norm
f = f + f .
Now let us choose B = [C [0, 1], ] and N = [C [0, 1], ].
Consider the identity mapping I : B N. The identity mapping is one-to-one onto and continuous.
I–1 is not continuous. For, if it were continuous, then it is a homeomorphism. Mapping of a
complete space into an incomplete space which cannot be. Hence I does not map open sets into
open sets.
Thus the open mapping theorem fails if the range of T is not a Banach space.
Example 2: Let B be an infinite dimensional Banach space with a basis { i : i I} with

i
= 1 for each i I. Let N be the set of all functions from I to C which vanish everywhere except
a finite member of points in I. Then N is a linear space under addition and scalar multiplication.
We can define the norm on N as
f = |f (i)|, i I.
Then N is an incomplete normed linear space. Now consider the transformation
T:N B defined as follows.
For each f N, let T (f) = f (i) i

.

Then T is linear and Notes
T (f) |f(i)| i
i I
= |fi | = f for every f N.

i I
Hence T is bounded transformation from N to B . It is also one-to-one and onto. But T does not
map open subsets of N onto B . For, if it maps, it is a linear homeomorphism from N onto B
which cannot be since N is incomplete.
Theorem 3: Let B be a Banach space and N be a normed linear space. If T is a continuous linear
open map on B onto N, then N is a Banach space.
Proof: Let (yn) be a Cauchy sequence in N. Then we can find a sequence of positive integer (n k)
such that nk < nk + 1 and for each k
1
y nk y nk
1
2k
Hence by theorem: “Let N and N be normed linear spaces. A linear map T : N N is open and
onto if and only if there is a M > 0 such that for any y N , there is a x N such that Tx = y and
x M y .”
For y nk 1
yn N , there is a n k B and a constant M such that
T (xk) = y nk 1
y n and x k y nk 1
yn .
By on choice y nk y nk is convergent so that x nk is convergent. Since B is a Banach

1
k 1 k 1
space there is a x B such that

n
x = Lim x nk
n
k 1
Since T is continuous T(x k ) T (x) as n

k 1
But T(xk ) y nk y n1 so that

1
k 1
y nk 1
y n1 T(x) y nk 1
y n1 T(x)
Since (yn) is a Cauchy sequence such that every subsequences is convergent, (yn) itself converges
and yn y n1 + T (x) in N.
Hence N is complete. Consequently, N is a Banach space.
Example: Let N be complete in two norms 1 and 2

respectively. If there is a number
a > 0 such that x 1 a x 2 for all x N, then show that the two norms are equivalent.
Solution: The identity map
i : (N, 2
) (N, 1
) is an one-one onto map.

Notes Also x 1
a x 2
i is bounded … (1)
i is continuous.
Hence by open-mapping theorem, i is open and so it is homeomorphism of (N, x 2

) onto
(N, x 1). Consequently i is bounded as a map from (N, x 1) (N, x 2)
Since i–1 (x) = x, a, b s.t. x 2

b 1
… (2)
(1) & (2) imply that the norms are equivalent.
19.2 Summary
 If B and B are Banach spaces and T is a continuous linear transformation of B onto B , then
the image of each sphere centered on the origin in B contains an open sphere centered on
the origin in B .
 The open mapping theorem : If T is a continuous linear transformation of a Banach space

B onto a Banach space B , then T is an open mapping.
19.3 Keywords
Banach Space: A normed space V is said to be Banach space if for every Cauchy sequence
n n 1
V then there exists an element V such that lim n .
n
Homeomorphism: A map f : (X, T) (Y, U) is said to be homeomorphism if
(i) f is one-one onto.
(ii) f and f–1 are continuous.
Open Sphere: Let xo X and r R+. Then set {x X : p (xo, x) < r} is defined as open sphere with
centre xo and radius r.
1. If X and Y are Banach spaces and A : X Y is a bounded linear transformation that is

bijective, then prove that A–1 is bounded.
2. Let X be a vector space and suppose 1

, and 2
are two norms on X and that T1 and T2
are the corresponding topologies. Show that if X is complete in both norms and T 1 T2 ,
then T1 = T2.
Books Walter Rudin, Functional Analysis, McGraw-Hill, 1973.

Jean Diendonne, Treatise on Analysis, Volume II, Academic Press (1970).
Online links euclid.colorado.edu/ngwilkin/files/math 6320…/OMT_CGT.pdf

people.sissa.it/nbianchin/courses/…/lecture05.banachstein.pdf

Unit 20: The Closed Graph Theorem
Unit 20: The Closed Graph Theorem Notes
CONTENTS
Objectives
Introduction
20.1 The Closed Graph Theorem
20.1.1 Graph of Linear Transformation
20.1.2 Closed Linear Transformation
20.1.3 The Closed Graph Theorem - Proof
20.2 Summary
20.3 Keyword

Objectives
 State the closed graph theorem.
 Understand the proof of the closed graph theorem
 Solve problems based on the closed graph theorem.
Introduction
Though many of the linear transformations in analysis are continuous and consequently bounded,
there do exist linear transformation which are discontinuous. The study of such kind of
transformation is much facilitated by studying the graph of transformation and using the graph
of the transformation as subset in the Cartesian product space to characterise the boundedness of
such transformations. The basic theorem in this regard is the closed graph theorem.
20.1 The Closed Graph Theorem
20.1.1 Graph of Linear Transformation
Definition: Let N and N be a normed linear space and let T : N N be a mapping with domain
N and range N . The graph of T is defined to be a subset of N × N which consists of all ordered
pairs (x, T (x)). It is generally denoted by G T.
Therefore the graph of T : N N is
GT = {(x, T (x) : x N}.
Notes GT is a linear subspace of the Cartesian product N × N with respect to coordinate-

wise addition and scalar multiplications.

Notes Theorem 1: Let N and N be normed linear spaces. Then N × N is a normed linear space with
coordinate-wise linear operations and the norm.
1
p p p
(x, y) = x x , where x N, y N
and | p < . Moreover, this norm induces the product topology on N × N , and N × N is
complete iff both N and N are complete.
Proof:
(i) It needs to prove the triangle inequality since other conditions of a norm are immediate.
Let (x, y) and (x , y ) be two elements of N × N .
Then (x, y) + (x , y ) = (x + x , y + y )
1
p p p
x x y y
1
p p p
= x x y y
1 1
p p p p p p
= x y x y
(By Minkowski’s inequality)

= (x, y) + (x , y ) .
This establishes the triangular inequality and therefore N × N is a normed linear space.
Furthermore (xn, yn) (x, y) xn = x and yn = y. Hence theorem on N × N induces the

product topology.
(ii) Next we show that N × N is complete N, N are complete.
Let (xn, yn) be a Cauchy sequence in N × N . Given > 0, we can find a n o such that
(xn, yn) – (xm, ym) < m, n no. … (1)
(xn – xm) < and yn – ym < m, n no
(xn) and (yn) are Cauchy sequences in N and N respectively.
Since N, N are complete, let
xn xo N and yn yo N in their norms,
i.e. (xn – xo) < and yn – ym < m, n no. … (2)
since xo N, yo N , (xo, yo) N×N.
Further (xn, yn) – (xo, yo) < n no (using (2))
(xn, yn) (xo, yo) in the norm of N × N and (xo, yo) N×N.
N × N is complete.
The converse follows by reversing the above steps.

Notes
Notes The following norms are equivalent to above norm
(i) (x, y) = max { x , y }
(ii) (x, y) = x + y (p = 1 in the above theorem)
20.1.2 Closed Linear Transformation
Definition: Let N and N be normed linear spaces and let M be a subspace of N. Then a linear
transformation
T:M N is said to be closed
iff xn M, xn x and T (xn) y imply x M and y = T (x).
Theorem 2: Let N and N be normed linear spaces and B be a subspace of N. Then a linear
transformation T : M N is closed its graph GT is closed.
Proof: Let T is closed linear transformation. We claim that its graph G T is closed i.e. GT contains
all its limit point.
Let (x, y) be any limit point of GT. Then a sequence of points in GT, (xn, T (xn), xn M, converging
to (x, y). But
(xn, T (xn)) (x, y)
xn, T (xn) – (x, y) 0
(xn – x), T (xn) – y 0
xn – x + T (xn) – y 0
xn – x 0 and T (xn) – y 0
xn x and T (xn) y ( T is closed)
(x, y) GT. (By def. of graph)
Thus we have shown that every limit point of G T is in GT and hence GT is closed.
Conversely, let the graph of T, G T is closed.
To show that T is closed linear transformation.
Let xn M, xn x and T (xn) y.
Then it can be seen that (x, y) is an adherent point of G T so that
(x, y) G T . But G T = GT ( GT is closed)
Hence (x, y) GT and so by the definition of G T we have x M and y = T (x).
Consequently, T is a closed linear transformation. This completes the proof of the theorem.
20.1.3 The Closed Graph Theorem – Proof
If B and B are Banach spaces and if T is linear transformation of B into B , then T is continuous
Graph of T (GT) is closed.
Proof: Necessary Part:
Let T be continuous and let GT denote the graph of T, i.e.
GT = {(x, T (x) : x B} B×B.

Notes
We shall show that G T = GT.
Since GT G T always, it suffices to show that G T GT.
Let (x, y) G T . Then there exists a sequence (x n, T (xn)) in GT such that
(xn, T (xn)) (x, y)
xn x and T (xn) y.
But T is continuous T (xn) T (x) and so y = T (x)
(x, y) = (x, T (x)) GT
GT GT
Hence GT = G T i.e. GT is closed.
Sufficient Part:
Let GT is closed. Then we claim that T is continuous. Let B1 be the given linear space B renormed
by 1
given by
x 1
= x + T (x) for x B.
Now T (x) x + T (x) = x 1.
T is bounded (continuous) as a mapping from B 1 to B .
So if B and B1 have the same topology then T will be continuous from B to B . To this end, we have
to show that B and B1 are homeomorphic.
Consider the identity mapping
I : B1 B defined by
I (x) = x for every x B1 .
Then I is always one-one and onto.
Further I (x) = x x + T (x) = x 1
I is bounded (continuous) as a mapping from B 1 onto B.
Therefore if we show that B 1 is complete with respect to 1

, then B1 is a Banach space so by
theorem.
“Let B and B be Banach spaces and let T be one-one continuous linear transformation of B onto
B . Then T is a homeomorphism. In particular, T –1 is automatically continuous.”
I is homeomorphism. Therefore to complete the proof, we have to show that B 1 is complete

under the norm 1
.
Let (xn) be a Cauchy sequence in B1. Then
xn – xm 1
= xn – xm + T (xn – xm) 0 as m, n
(xn) and (T (xn)) are Cauchy sequences in B and B respectively.
Since B and B are complete, we have
xn x in B and T (xn) T (x) in B … (1)

Since GT is closed, we have Notes
(x1 T (x)) GT and if we take
y = T(x); then (x, y) GT.
Now xn – x 1
= xn – x + T (xn – x)
= xn – x + T (xn) – T(x)
= xn – x + T (xn – y) 0 as n . (Using (1))
Hence, the sequence (xn) in B1 x B1 and consequently B1 is complete.
Theorem 3: Let B and B be Banach spaces and let T : B B be linear. If GT is closed in B × B and
if T is one-one and onto, then T is a homeomorphism from B onto B .
Proof: By closed graph theorem, T is continuous.
Let T = T–1 : B B. Then T is linear.
Further (x, y) GT (y, x) GT .
GT is closed in B × B.
T is continuous (By closed graph theorem)
T is a homeomorphism on B onto B .
Theorem 4: Let a Banach space B be made into a Banach space B by a new norm. Then the
topologies generated by these two norms are the same if either is stronger than the other.
Proof: Let the new norm on B be . Let is stronger than . Then a constant k such that
x k x for every x B.
Consider the identity map
I:B B.
We claim that G1 is closed.
Let xn x in B and xn y in B .
Then x k x x B, I (xn) = xn y in also.
Since a sequence cannot converge to two distinct points in , y = x. Consequently G1 is closed.
Hence closed graph theorem, I is continuous. Therefore a k s such that
x = I(x) k x for every x B. Hence is stronger than . Hence two topologies are
same.
20.2 Summary
 Let N and N be a normal linear space and let T : N N be a mapping with domain N and
range N . The graph of T is defined to be a subset of N × N which consist of all ordered
pairs (x, T (x)). It is generally denoted by G T.

Notes  Let N and N be normed linear spaces and let M be a subspace of N. Then a linear
transformation T : M N is said to be closed iff xn M, xn x and T (xn) y imply
x M and y = T (x).
 If B and B are Banach spaces and if T is linear transformation of B into B , then T is

continuous Graph of T (GT) is closed.
20.3 Keyword
Closed Linear Transformation: Let N and N be normed linear spaces and let M be a subspace of
N. Then a linear transformation
T:M N is said to be closed
iff xn M, xn x and T (xn) y imply x M and y = T (x).
1. If X and Y are normed spaces and A : X Y is a linear transformation, then prove that
graph of A is closed if and only if whenever x n 0 and Axn y, it must be that y = 0.
2. If P is a projection on a Banach space B, and if M and N are its range and null space, then
prove that M and N are closed linear subspaces of B such that B = M N.
Books Folland, Gerald B, Real Analysis: Modern Techniques and their Applications (1st ed.),
John Wiley & Sons, (1984).
Rudin, Walter, Functional Analysis, Tata McGraw-Hill (1973).
Online links euclid.colorado.edu/ngwilkin/files/math6320.../OMT_CGT.pdf

mathworld.wolfram.com

Unit 21: The Conjugate of an Operator
Unit 21: The Conjugate of an Operator Notes
CONTENTS
Objectives
Introduction
21.1 The Conjugate of an Operator
21.1.1 The Linear Function
21.1.2 The Conjugate of T
21.2 Summary
21.3 Keywords
Objectives
 Understand the definition of conjugate of an operator.
 Understand theorems on it.
 Solve problems relate to conjugate of an operator.
Introduction
We shall see in this unit that each operator T on a normed linear space N induces a corresponding
operator, denoted by T* and called the conjugate of T, on the conjugate space N*. Our first task is
to define T* and our second is to investigate the properties of the mapping T T*.
21.1 The Conjugate of an Operator
21.1.1 The Linear Function
Let N* be the linear space of all scalar-valued linear functions defined on N. Clearly the conjugate
space N* is a subspace of N*. Let T be a linear transformation T of N* into itself as follows:
If f N+, then T (f) is defined as
[T (f)]x = f (T (x))
Since f (T (x)) is well defined, T is a well-defined transformation on N +.
Theorem 1: Let T : N+ N+ be defined as
[T (j)] x = f (T (x)), f N+, then
(a) T (j) is a linear junction defined on N.
(b) T is a linear mapping of N + into itself.

Notes (c) T (N*) N* T is continuous, where T is a linear transformation of N into itself which is
not necessarily continuous.
Proof:
(a) x, y N and , be any scalars. Then
[T (f)] ( x + y) = f (T ( x + y))
Since T and f are linear, we get
f (T ( x + y)) = f (T (x) + f (T (y))
= [T (f)] (x) + [T (f)]y
part (a).
(b) Let f, g N+ and , be any scalars. Then
[T ( f + g) (x)] = ( f + g) (T (x)) = [T (f)] (x) + (T (g)] (x)
T is linear on N+
part (b)
(c) Let S be a closed unit sphere in N. Then we know that T is continuous T (S) is bounded
f (T (S)) is bounded for each f N*.
By definition of T , f (T (S)) is bounded if and only if [T (f)] (S) is bounded for each f in
N* = T (f) is in N* for each f in N*.
T (N) N*
part (c)
Note: Part (c) of the above theorem enables us to restrict T to N* iff T is continuous. Hence
by making T continuous we define an operation called the conjugate of T by restricting T
to N*. We see it below.
21.1.2 The Conjugate of T
Definition: Let N be normed linear space and let T be a continuous linear transformation of N into
itself (i.e. T is an operator). Define a linear transformation T* of N* into itself as follows:
If f N*, then, T* (f) is given by
[T* (f)] (x) = f (T (x))
We call T* the conjugate of T.
Theorem 2: If T is a continuous linear transformation on a normed linear space N, then its

conjugate T* defined by
T* : N* N* such that
T* (f) = f.T where
[T* (f)] (x) = f (T (x)) f N* and all x N
is a continuous linear transformation on N* and the mapping T T* given by
: (N) (N*) such that

(T) = T* for every (N) Notes
is an isometric isomorphism of b (N) into b (N*) reverses products and preserves the identify
transformation.
Proof: We first show that T* is linear
Let f, g N* and , be any scalars
then [T* ( j + g)] (x) = ( j + g) (T (x))
= ( j) T (x) + ( g) T (x)
= [j (T (x))] + [g (T (x))]
= [T* (j)] (x) + [T* (g)] (x)
= [ T* (j) + T* (g)] (x) x N
Hence T* ( f + g) = T* (f) + T* (g)
T* is linear on N*.
To show that T* is continuous, we have to show that it is bounded on the assumption that T is
bounded.
T* = sup { T* (f) : f 1}
= sup { [T* (f)] (x) : f 1 and x }
= sup { f (T (x))|: f 1 and x }
= sup { f T x : x 1 and x } … (1)
T* is a bounded linear transformation on N* into N*. Hence by application of Hahn-

Banach theorem, for each non-zero x in N, a functional f N* such that
f = 1 and f (T (x)) = T (x) … (2)
T(x)
Hence T = sup :x 0
x
f T(x)
= sup : f 1, x 0 (by (2))
x
T * (f) (x)
= sup : f 1, x 0 (by (1))
x
T * (f) x
sup : f 1, x 0
x
= sup T * (f) : f 1 T* … (3)
T = T* . … (4)

Notes Now we show that
: (N) (N*) given by

(T) T * … (5)
for every T (N) is an isometric isomorphism which reverses the product and preserve the
identity transformation.
The isometric character of follows by using (5) as seen below:
(T) = T* = T .
Next we show that is linear and one-to-one. Let T, T 1 (N) and , be any scalars. Then
( T+ T1 ) = ( T + T)* by (3)
But [( T + T1)* (f)] (x) = f ( T + T1) (x)
= f ( T (x) + T1 (x) )
Since f is linear, we get
[( T + T1)* (f)] = f (T (x)) + f (T1 (x))
= [T* (f) (x) + T *1 (f) (x)]
= [T * (f)] [T * 1 (f)] (x)
x N. Hence we get
[( T + T1)* (f)] = [T* (f)] + [T *1 (f) ]
= T* T *1 (f)
Hence ( T+ T1)* = T* + T1* … (6)
Therefore ( T+ T1 ) = ( T + T1)* = T* + T *1 = (T) + (T1)
is linear.
To show is one-to-one, let (T) = (T1)
Then T* = T *1
T* T *1 =0
Using (6) by choosing = 1, = – 1 we get
(T – T1)* = 0 T – T1 = 0 or T = T1.
is one-to-one.
Hence is an isometric isomorphism on (N) onto (N*).
Finally we show that reverses the product and preserves the identity transformation.
Now [(T T1)* (f)] (x) = f ((T T1) (x))
= f (T (T1 (x))
= [T* (f)] [T1 (x)], since T1 (x) N and T* (f) N*.

= [ T *1 (T* (f))] (x) Notes
= [( T *1 T*) (f)] (x)
Hence, we get
(T T1)* = T *1 T* so that
(T T1) = (T T1)* = T *1 T.
reverses the product.
Lastly if I is the identity operator on N, then
[I* (f)] (x) = f (I (x)) = f (x) = (I f) (x).
I* = I so that (I) = I* = I
preserves the identity transformation.
Theorem 3: Let T be an operator on a normal linear space N. If N N* in the natural imbedding,

then T** is an extension of T. If N is reflexive, then T** = T.
Proof: By definition, we have
(T*)* = T**
Using theorem 2, we have T* = T .
Hence T** = T* = T .
By definition of conjugate of an operator
T:N N, T* : N* N*, T** : N** N**.
Let J ; x Fx be the natural imbedding of N onto N** so that
Fx (f) = f (x) and J (x) = Fx.
Further, since T** is the conjugate operator of T*, we get
T** (x ) x = x (T* (x )) where x N* and x N**
T** (x ) x = T** (J (x)) x .
Using the definition of conjugate, we get
T** (J (x)) x = J (x) (T* (x )).

By definition of canonical imbedding
J (x) (T* (x )) = T* (x ) x.
Again T* (x ) (x) = x (T (x)) (By definition of conjugate)
Now x (T (x)) = J (T (x))x (By natural imbedding)
Hence T** (J (x))x = J (T (x))x .
T** . J = J T
and so T** is the norm preserving extension of T. If N is reflexive, N = N** and so T** coincides
with T.

Notes This completes the proof of the theorem.
Theorem 4: Let T be an operator on a Banach space B. Then T has an inverse T–1 T* has an
inverse (T*)–1, and
(T*)–1 = (T–1)*
Proof: T has inverse T–1 TT–1 = T–1 T = I
By theorem 2, the mapping :T T* reverse the product and preserves the identity
(TT–1)* = (T–1 T)* = I*
(T–1)*T = T* (T–1)* = I
(T*)–1 exists and it is given by (T*)–1 = (T–1)*. This completes the proof of the theorem.
21.2 Summary
 Let N+ be the linear space of all scalar-valued linear functions defined on N. Clearly the
conjugate space N* is a subspace of N+. Let T be a linear transformation T of N into itself
as follows:
T (f) x = f (T (x))
 Let N be a normed linear space and let T be a continuous linear transformation of N into
itself. Define a linear transformation T* of N* into itself as follows:
If f N+, then T (f) is given by
T (f) x = f (T (x))
21.3 Keywords
The Conjugate of T: Let N be normed linear space and let T be a continuous linear transformation
of N into itself (i.e. T is an operator). Define a linear transformation T* of N* into itself as
follows:
If f N*, then, T* (f) is given by
[T* (f)] (x) = f (T (x))
The Linear Function: Let N* be the linear space of all scalar-valued linear functions defined on N.
Clearly the conjugate space N* is a subspace of N*. Let T be a linear transformation T of N* into
itself as follows:
[T (f)]x = f (T (x))
Since f (T (x)) is well defined, T is a well-defined transformation on N +.
1. Let B be a Banach space and N a normed linear space. If {T n} is a sequence in B (B, N) such
that T(x) = lim Tn (x) exists for each x in B, prove that T is a continuous transformation.

2. Let T be an operator on a normed linear space N. If N is considered to be part of N** by Notes

means of the natural imbedding. Show that T** is an extension of T. Observe that if N is
reflexive, then T** = T.
3. Let T be an operator on a Banach space B. Show that T has an inverse T –1 T* has an inverse
(T*)–1, and that in this case (T*) –1 = (T–1)*.
Books James Wilson Daniel, The Conjugate Gradient Method for Linear and Non-linear
Operator Equations.
G.O. Okikiolu, Special Integral Operators: Poisson Operators, Conjugate Operators and
related Integrals. Vol. Okikiolu Scientific and Industrial Organization, 1981.
Online links epubs.siam-org/sinum/resource/1/sjnamm/v9/i|/p165_s|

www.ima.umm.edu

Notes Unit 22: The Uniform Boundedness Theorem
CONTENTS
Objectives
Introduction
22.1 The Uniform Boundedness Theorem
22.1.1 The Uniform Boundedness Theorem – Proof
22.2 Summary
22.3 Keywords
Objectives
 State the uniform boundedness theorem.
 Understand the proof of this theorem.
 Solve problems related to uniform boundedness theorem.
Introduction
The uniform boundedness theorem, like the open mapping theorem and the closed graph
theorem, is one of the cornerstones of functional analysis with many applications. The open
mapping theorem and the closed graph theorem lead to the boundedness of T –1 whereas the
uniform boundedness operators deduced from the point-wise boundedness of such operators.
In uniform boundedness theorem we require completeness only for the domain of the definition
of the bounded linear operators.
22.1 The Uniform Boundedness Theorem
22.1.1 The Uniform Boundedness Theorem – Proof
If (a) B is a Banach space and N a normed linear space,
(b) {Ti} is non-empty set of continuous linear transformation of B into N, and
(c) {Ti (x)} is a bounded subset of N for each x B, then { Ti } is a bounded set of numbers, i.e.
{Ti} is bounded as a subset of (B, N)
Proof: For each positive integer n, let
Fn = {x B : Ti (x) n i}.
Then Fn is a closed subset of B. For if y is any limit point of F n, then a sequence (xk) of points of
Fn such that
xk y as k

Unit 22: The Uniform Boundedness Theorem
Tixk Tiy as k (By continuity of Ti) Notes
Tixk Tiy as k (By continuity of norm)
Tiy = lt Ti xk
k
n i ( xk Fn)
y Fn.
Thus Fn contains all its limit point and is therefore closed. Further, if x is any element of B, then
by hypothesis (c) of the theorem a real number k 0 s.t.
Tix k i
Let n be a positive integer s.t. n > k. Then
Tix <n i
so that x Fn.
Consequently, we have B = F .
n 1
n
Since B is complete, it therefore follows by Baire’s theorem that closure of some Fn, say F no Fno ,
possesses an interior point x o. Thus we can find a closed sphere S o with centre xo and radius r o
such that So Fno .
Now if y is any vector in Ti (So), then
y = Ti so
where so So Fno .
y = Ts
i o no.
Thus norm of every vector in Ti (So) is less than or equal to no. We write this fact as Ti (So) no.
So xo
Let S = . Then S is a closed unit sphere centred at the origin in B and
ro
So xo
Ti (S) = Ti
ro
1
= Ti (S o ) Ti (x o )
ro
1
Ti (S o ) Ti (x o )
ro
2n o
i
ro
2n o
Hence Ti i
ro

Notes 22.1.2 Theorems and Solved Examples
Theorem 1: If B is a Banach space and (f i (x)) is sequence of continuous linear functionals on B such
that (|fi (x)|) is bounded for every x B, then the sequence ( fi ) is bounded.
Proof: Since the proof of the theorem is similar to the theorem (1), however we briefly give its
proof for the sake of convenience to the readers.
For every m, let Fm B be the set of all x such that |fn (x)| m n
|fn (x)| m n
Now Fm is the intersection of closed sets and hence it is closed.
As in previous theorem, we have
B= F
m 1
m . Since B is complete. It is of second category. Hence by Baire’s theorem, there is a
xo Fm and a closed sphere S [x o, ro] such that |fn (x)| m n.
Let x be a vector with x ro.
Now fn (x) = fn (x + xo – xo)
= fn (x + xo) – fn (xo)
|fn (x)| |fn (x + xo)| + |fn (xo)| … (1)
Since x + xo – xo = x < ro, we have (x + ro) S [xo, ro]
|fn (x + xo) | m … (2)
Also we have |fn (xo)| k n … (3)
From (1), (2) and (3), we have for x S [xo, r].
|fn (x)| < (m + k) n.
ro x
Now for x B, consider the vector .
x
x ro x x fn (x) m k
Then |fn (x)| fn (m k) so that .
ro x ro x ro
In other words,
m k
f .
ro
Example 1: Show that the completeness assumption in the domain of (Ti) in the uniform
boundedness theorem cannot be dropped.
Solution: Consider N= space of all polynomial x
= x (t) = a n t n , an 0
n 0
for finitely many n’s.

It we define the norm on N as Notes
x = max {|an|, n = 0, 1, 2, …}
then N is an incomplete normed linear space.
n 1
Now define fn (x) = a k , n = 1, 2, …

k 0
The functions {fn} are continuous linear functional on N.
If we take x = a0 + a1t + …… + am tm then
|fn (x)| (m + 1) max {|ak|} = (m + 1) x ,
so that (|fn (x)|) is point-wise bounded.
Now consider x = 1 + t + t2 + … + tn–1. Then x = 1 and from the definition of |f n (x)| = n.
fn (x)
Hence fn = n.
x
( fn ) is unbounded.
Thus if we drop the condition of completeness in the domain of (T i), the uniform boundedness
theorem is not true anymore.
Theorem 2: Let N be a normed linear space and B be a Banach space. If a sequence (Tn) (B, N)
such that T (x) = lim Tn (x) exists for each x in B, then T is a continuous linear transformation.
Proof: T is linear.
T ( x + y) = lim Tn ( x + y)
= lim {Tn ( x) + Tn ( y)|}
= lim Tn (x) + lim Tn (y)
= T (x) + T (y) for x, y B and for any scalars and .
since lim Tn (x) exists, (Tn (x)) is a convergent sequence in N. Since convergent sequences are
bounded, (Tn (x)) is point-wise bounded.
Hence by uniform bounded theorem, ( Tn ) is bounded so that a positive constant such that
Tn n.
Now Tn(x) Tn x x .
Since Tn (x) T (x), we have
T (x) x
T is bounded (continuous) linear transformation. This completes the proof of the theorem.
Corollary 1: If f is a sequence in B* such that f (x) = lim fn (x) exists for each x B, then f is
n
continuous linear functional on B.
Example 2: Let (an) be a sequence of real or complex numbers such that for each x = (x n)
co, a n x n converges. Prove that |a n | .

n 1 n 1

Notes n n
Solution: For every x co, let fn a i x i . Since each a i x i is a finite sum of scalars, (f n) is
i 1 i 1
sequence of continuous linear functional on c o. Let f (x) = lim

n
fn (x) lim
n
a i x i . By cor. 1, f (x)
i 1
exists and bounded. f = |a n |. Since f is bounded, |a n | < .

n 1 n 1
Theorem 3: A non-empty subset X of a normed linear space N is bounded f (X) is a bounded set
of numbers for each f in N*.
Proof: Let X be a bounded subset of N so that a positive constant 1

such that
x 1
x X … (1)
To show that f (X) is bounded for each f N*. Now f N* f is bounded.
2
> 0 such that |f (x)| 2
x x N … (2)
It follows from (1) & (2) that
|f (x)| 1 2
x X.
f (X) is a bounded set of real numbers for each f N*.
Conversely, let us assume that f (X) is a bounded set of real numbers for each f N*.
To show that X is bounded. For convenience, we exhibit the vectors in X by writing X = {x i}. We
now consider the natural imbedding J from N to N** given by
J : xi Fxi
From the definition of this natural imbedding, we have
Fx (f) = f (x) for each x N.
Hence our assumption f (X) = {f (xi)} is bounded for each f N* is equivalent to the assumption
that Fxi (f) is bounded set for each f N*.
Since N* is complete Fxi is bounded subset of N** by uniform boundedness theorem.
That is, Fxi is a bounded set of numbers. Since the norms are preserved in natural imbedding,
we have Fxi = xi for every xi X.
Therefore ( xi ) is a bounded set of numbers. Hence is bounded subset of Ni.
Theorem 4: Let N and N be normed linear space A linear transformation.
T:N N is continuous for each f N*, f o T N*.
Proof: We first note that f o T is linear. Also f o T is well defined, since T (x) N for every x N
and f is a functional on N so that f (T (x)) is well defined and f o T N*. Since T is continuous and
f is continuous, f o T is continuous on N.

Conversely, let us assume that f o T is continuous for each f N*. To show that T is continuous Notes
it suffices to show that
T(B) = {Tx : x N, B = x 1} is bounded in N .
For each f N*, f o T is continuous and linear on N and so (f o T) B = f (T(B)) is bounded set for
every f N*, where we have considered the unit sphere B with centre at the origin and radius 1.
Since any bounded set in N can be obtained from B, T (B) is bounded by a non-empty subset X of
a normed linear space N of bounded f (X) is a bounded set of number for each f in N*.
22.2 Summary
 Uniform Boundedness Theorem: If (a) B is Banach space and N a normed linear space,
(b) {Ti} is non-empty set of continuous linear transformation of B into N and (c) {Ti (x)} is a
bounded subset of N for each x B, then { Ti } is a bounded set of numbers, i.e. {Ti} is
bounded as a subset of (B, N).
 If B is a Banach space and (f i (x)) is sequence of continuous linear functionals on B such that
(|fi (x)|) is bounded for every x B, then the sequence ( Ti ) is bounded.
22.3 Keywords
Imbedding: Imbedding is one instance of some mathematical structure contained within another
instance, such as a group that is a subgroup.
Uniform Boundedness Theorem: The uniform boundedness theorem, like the open mapping
theorem and the closed graph theorem, is one of the cornerstones of functional analysis with
many applications.
1. If X is a Banach space and A X*, then prove that A is a bounded set if and only if for every
x in X, Sup {|f (x)| : f A} < .
2. Let  be a Hilbert space and let  be an orthonormal basis for . Show that a sequence {hn}
in satisfies <hn, h> 0 for every h in if and only if sup { hn : n 1} < and <hn, e>
0 for every e in .
Books Bourbaki, Nicolas, Topological vector spaces, Elements of mathematics, Springer (1987).
Diendonne, Jean, Treatise on Analysis, Volume 2, Academic Press, (1970).
Rudin, Walter, Real and Complex Analysis, McGraw-Hill, 1966.
Online links www.jstor.org/stable/2035429

www.sciencedirect.com/science/article/pii/S0168007211002004

Notes Unit 23: Hilbert Spaces: The Definition and

Some Simple Properties
CONTENTS
Objectives
Introduction
23.1 Hilbert Spaces
23.1.1 Inner Product Spaces
23.1.2 Hilbert Space and its Basic Properties
23.1.3 Hilbert Space: Definition
23.1.4 Examples of Hilbert Space
23.2 Summary
23.3 Keywords
Objectives
 Define inner product spaces.
 Define Hilbert space.
 Understand basic properties of Hilbert space.
 Solve problems on Hilbert space.
Introduction
Since an inner product is used to define a norm on a vector space, the inner product are special
normed linear spaces. A complete inner product space is called a Hilbert space. We shall also see
from the formal definition that a Hilbert space is a special type of Banach space, one which
possesses additional structure enabling us to tell when two vectors are orthogonal. From the
above information, one can conclude that every Hilbert space is a Banach space but not conversely
in general.
We shall first define Inner Product spaces and give some examples so as to understand the
concept of Hilbert spaces more conveniently.
23.1 Hilbert Spaces
23.1.1 Inner Product Spaces
Definition: Let X be a linear space over the field of complex numbers C. An inner product on X is
a mapping from X × X C which satisfies the following conditions:

Unit 23: Hilbert Spaces: The Definition and Some Simple Properties
(i) ( x + y, z) = (x, z) + (y, z), x, y, z X and , C. (Linearity property) Notes
(ii) (x, y) = (y, x) (Conjugate symmetry)
(iii) (x, x) 0, (x, x) = 0 x=0 (Non-negativity)
A complex inner product space X is a linear space over C with an inner product defined on it.
Notes
1. We can also define inner product by replacing C by R in the above definition. In that
case, we get a real inner product space.
2. It should be noted that in the above definition (x, y) does not denote the ordered pair
of the vectors x and y. But it denotes the inner product of the vectors x and y.
Theorem 1: If X is a complex inner product space then
(a) ( x – y, z) = (x, z) – (y, z)
(b) (x, y + z) = (x, y) + (x, z)
(c) (x, y – z) = (x, y) – (x, z)
(d) (x, 0) and (0, x) = 0 for every x X.
Proof: (a) ( x – y, z) = ( x + (– ) y, z)
= (x, z) + (– ) (y, z)
= (x, z) – (y, z).
(b) (x, y + z) = ( y z, x) ( y, x) ( z, x)
= (y, x) (z, x)
= (x, y) (x,z)
(c) (x, y – z) = (x, y + (– ) z) = (x, y) ( ) (x,z)
= (x, y) (x,z)
(d) (0, x) = (0 , x) = 0 ( , x) = 0, where is the zero
element of x and (x, 0) (0, x) 0 0.
Further note that (x, y + z) = (x, |y + 1|z) = 1 (x, y) + 1 (x, z)
Hence (x, y + z) = (x, y) + (x, z).

Notes
Notes
1. Part (b) shows an inner product is conjugate linear in the second variable.
2. If (x, y) = 0 x X, then y = 0. If (x, y) = 0 x X, it should be true for x = y also,

so that (y, y) = 0 y = 0.
Example 1: The space  n2 is an inner product space.
Solution: Let x = (x1, x2, ……, xn), y = (y1, y2, ……, yn)  n2 .
Define the inner product on  n2 as follows:

n
(x, y) = xi yi
i 1
Now
n
(i) ( x + y, z) = xi yi zi
i 1
n n
= xi zi yi zi
i 1 i 1
= (x, z) + (y, z)
(ii) (x, y) = xi yi
i 1
= (x1 y1 x2 y2  xn yn )
= (x1 y1 x2 y2  xn yn )
= x1 y1 x2 y2  xn yn
= (y, x)
n
(iii) (x, x) = xi xi
i 1
n
2
= xi 0
i 1
Hence (x, x) 0 and (x, x) = 0 xi = 0 for each i, i.e. (x, x) = 0 x = 0.
(i) – (iii)  n2 is a inner product space.
23.1.2 Hilbert Space and its Basic Properties
By using the inner product, on a linear space X we can define a norm on X, i.e. for each x X, we
define x = (x,x) . To prove it we require the following fundamental relation known as
Schwarz inequality.

Theorem 2: If x and y are any two vectors in an inner product space then Notes
|(x, y)| x y … (1)
Proof: If y = 0, we get y = 0 and also theorem 1 implies that |(x, y)| = 0 so that (1) holds.
Now, let y 0, then for any scalar C we have

2
0 x– y = (x – y,x – y)
But
(x – y, x – y) = (x, x) – (x, y) – ( y, x) + ( y, y)
= (x, x) – (x, y) – (y, x) + (y, y)

2
= x – (y, x) – (x, y) + | |2 y 2
2
x – (y, x) – (x, y) +| |2 y 2
= (x, y, x – y)
2
= x– y 0 … (2)
(x, y)
Choose = 2 , y 0, y 0.
y
We get from (2)
2
2 (x, y) (x y) (x, y) (x, y) 2
x 2 2 (x, y) 4 y 0
y y y
2
2 |(x, y)|2 |(x, y)|2 (x, y)
x 2 2 2 0
y y y
2 |(x, y)|2
x 2 0
y
2 2
x y |(x, y)|2
or |(x, y)| x y .
Theorem 3: If X is an inner product space, then (x, x) has the properties of a norm, i.e.
x = (x, x) is a norm on X.
Proof: We shall show that satisfies the condition of a norm.
(i) x = (x, x) x 2
= (x, x) 0 and x = 0 x = 0.
(ii) Let x, y X, then

2
x+y = (x + y, x + y)
= (x, x) + (x, y) + (y, x) + (y, y) … (1)
2 2
= x + (x, y) + (x, y) + y
= x 2
+ 2Re (x, y) + y 2
[ (x, y) (x, y) 2 Re(x, y)]
2 2
x + 2|(x, y)| + y [ Re (x, y) |(x, y)|]

Notes x 2
+2 x y + y 2
[using Schwarz inequality]
= ( x + y )2
Therefore x+y x + y
2
(iii) x = ( x, x)
= (x, x)
2 2
=| | x
,x = x
(i)-(iii) imply that x = (x, x) is a norm on X. This completes the proof of the theorem.
Note Since we are able to define a norm on X with the help of the inner product, the inner
product space X consequently becomes a normed linear space. Further if the inner product
space X is complete in the above norm, then X is called a Hilbert space.
23.1.3 Hilbert Space: Definition
A complete inner product space is called a Hilbert space.
Let H be a complex Banach space whose norm arises from an inner product which is a complex
function denoted by (x, y) satisfying the following conditions:
H1 : ( x + y, 2) = (x, 2) + (y, 2),
H2 : (x, y) = (y, x), and
H3 : (x, x) = x 2,
for all x, y, z H and for all , C.
23.1.4 Examples of Hilbert Space
1. The space  n2 is a Hilbert space.
We have already shown in earlier example that  n2 is an inner product space. Also  n2 is a
Banach space. Consequently  n2 is a Hilbert space. Moreover  n2 , being a finite dimensional,
hence  n2 is a finite dimensional Hilbert space.
2.  2 is a Hilbert space.
Consider the Banach space  2 consisting of all infinite sequence x = (x n), n = 1, 2, … of
2
complex numbers such that xn with norm of a vector x = (xn) defined by x =
n 1
2
xn .
n 1

We shall show that if the inner product of two vectors x = (x n) and y = (yn) is defined by Notes
(x, y) = x n y n , then  2 is a Hilbert space.

n 1
We first show that inner product is well defined. For this we are to show that for all x, y in  2 the
infinite series x n y n is convergent and this defines a complex number.

n 1
By Cauchy inequality, we have

1 1
n n 2 n 2
2 2
xi yi xi yi
i 1 i 1 i 1
1 1
2 2
2 2
xn yn .
n 1 n 1
n
2 2
Since xn and yn are convergent, the sequence of partial sum x i y i is a monotonic
n 1 n 1 i 1
increasing sequence bounded above. Therefore, the series x i y i is convergent. Hence

n 1
x n y n is absolutely convergent having its sum as a complex number.

n 1
Therefore (x, y) = x n y n is convergent so that the inner product is well defined. The condition
n 1
of inner product can be easily verified as in earlier example.
Theorem 4: If x and y are any two vectors in a Hilbert space, then

2 2 2
(x + y) + x–y =2( x + y 2)
Proof: We have for any x and y

2
(x + y) = (x + y, x + y) (By def. of Hilbert space)
= (x, x + y) + (y, x + y)
= (x, x) + (x, y) + (y, x) + (y, y)

2 2
= x + (x, y) + (y, x) + y … (1)
2
x–y = (x – y, x – y)
= (x, x – y) – (y, x – y)
2
= (x, x) – (x, y) – (y, x) + y … (2)
Adding (1) and (2), we get

2 2 2 2 2
x+y + (x – y) =2 x +2 y =2( x + y 2)

Notes Theorem 5: In a Hilbert space the inner product is jointly continuous i.e.,
xn x, yn y (xn, yn) (x, y)
Proof: We have
|(xn, yn) – (x, y)| = |(xn, yn) – (xn, y) + (xn, y) – (x,, y)|
= |(xn, yn – y) + (xn – x, y)|
(by linearity property of inner product)
|(xn, yn – y)| + |(xn – x, y)| [ | + | | | + | |]
x n + y n – y + xn – x y [By Schwarz inequality]
Since xn x and y n y as n .
Therefore yn – yn 0 and xn – x 0 as h . Also (xn) is a continues sequence, it is bounded

so that xn M n.
Therefore
| (xn, yn) – (x, y) | 0 as n .
Hence (xn, yn) (x, y) as n .
Theorem 6: A closed convex set E in a Hilbert space H continuous a unique vector of smallest
norm.
Proof: Let = inf { e ; e E}
To prove the theorem it suffices to show that there exists a unique x E s.t. x = .
Definition of yields us a sequence (xn) in E such that
Lim xn = … (1)
n
xm xn
Convexity of E implies that E . Consequently
2
xm xn
xm + xn 2 … (2)
2
Using parallelogram law, we get

2 2 2 2
xm + xn + x m – xn = 2 xn + 2 xn
2 2 2 2
or x m – xn = 2 xm + 2 xn – x m – xn
2 2 2
2 xm + 2 xn –d (Using (2))
0 as m, n (Using (1))
2
xm – x n 0 as m, n
(xn) is a CAUCHY sequence in E.
x E such that Lim x n x , since H is complete and E is a closed subset of H, therefore

n
E is also complete and consequently (x n) is in E is a convergent sequence in E.

Notes
Now x = Lim
n
xn
= Lim
n
xn ( norm is continuous mapping)
= .
Uniqueness of x.
Let us suppose that y E, y x and y = .
x y
Convexity of E E
2
x y
… (3)
2
Also by parallelogram law, we have
2 2 2 2
x y x y x y
=
2 2 2 2 2
2 2 2 2
x y 2 x y
=
2 2 2 2
2
< .
So that
2
x y
< , a result contrary to (3).
2
Hence we must have y = x.
Example: Give an example of a Banach space which is not an Hilbert space.

Solution: C [a, b] is a Banach space with supremum norm, i.e. if x C [a, b] then
x = Sup {|x(t)| : t [a, b]}.
Then this norm does not satisfy parallelogram law as shown below:
t a
Let x(t) = 1 and y (t) = . Then x = 1, y = 1
b a
t a
Now x (t) + y (t) = 1 + so that x + y = 2
b a
t a
x (t) – y (t) = 1 – so that x–y =1
b a
2
Hence 2 ( x – y 2) = 4, and x + y 2
+ x–y 2
=5
2 2 2 2
So that x + y + x–y 2 x +2 y .
C [a, b] is not a Hilbert space.

Notes 23.2 Summary
 Let X be a linear space over the field of complex numbers C. An inner product on X is a
mapping from X × X C which satisfies the following conditions:
(i) ( x + y, z) = (x, z) + (y, z) x, y, z X and , C.
(ii) (x, y) = (y, x)
(iii) (x, x) 0, (x, x) = 0 x=0
 A complete inner product space is called a Hilbert space.
23.3 Keywords
Hilbert Space: A complete inner product space is called a Hilbert space.
Inner Product Spaces: Let X be a linear space over the field of complex numbers C. An inner
product on X is a mapping from X × X C which satisfies the following conditions:
(i) ( x + y, z) = (x, z) + (y, z), x, y, z X and , C. (Linearity property)
(ii) (x, y) = (y, x) (Conjugate symmetry)
(iii) (x, x) 0, (x, x) = 0 x=0
1. For the special Hilbert space  n2 , use Cauchy’s inequality to prove Schwarz’s inequality.
2. Show that the parallelogram law is not true in  n2 (n > 1).
3. If x, y are any two vectors in a Hilbert space H, then prove that

2 2 2
4 (x, y) = x + y – x–y + i x + iy – i x – iy 2.
4. If B is complex Banach space whose norm obeys the parallelogram law, and if an inner
product is defined on B by
2 2 2
4 (x, y) = x + y – x–y + i x + iy – i x – iy 2,
then prove that B is a Hilbert space.
Books Bourbaki, Nicolas (1987), Topological vector Spaces, Elements of Mathematics, BERLIN:
Springer – Verlag.
Halmos, Paul (1982), A Hilbert space Problem Book, Springer – Verlag.
Online links www.math-sinica.edu.tw/www/file_upload/maliufc/liu_ch04.pdf

mathworld.wolfram.com>Calculus and Analysis > Functional Analysis

Unit 24: Orthogonal Complements
Unit 24: Orthogonal Complements Notes
CONTENTS
Objectives
Introduction
24.1 Orthogonal Complement
24.1.1 Orthogonal Vectors
24.1.2 Pythagorean Theorem
24.1.3 Orthogonal Sets
24.1.4 Orthogonal Compliment: Definition
24.1.5 The Orthogonal Decomposition Theorem or Projection Theorem
24.2 Summary
24.3 Keywords
Objectives
 Define Orthogonal complement
 Understand theorems on it
 Understand the Orthogonal decomposition theorem
 Solve problems related to Orthogonal complement.
Introduction
In this unit, we shall start with orthogonality. Then we shall move on to definition of orthogonal
complement. Let M be a closed linear subspace of H. We know that M is also a closed linear
subspace, and that M and M are disjoint in the sense that they have only the zero vector in
common. Our aim in this unit is to prove that H = M M , and each of our theorems is a step in
this direction.
24.1 Orthogonal Complement
24.1.1 Orthogonal Vectors
Let H be a Hilbert space. If x, y H then x is said to be orthogonal to y, written as x y, if

(x, y) = 0.
By definition,
(a) The relation of orthogonality is symmetric, i.e.,
x y y x

Notes For, x y (x, y) = 0
(x, y) 0
(y, x) = 0
y x
(b) If x y then every scalar multiple of x is orthogonal to y i.e. x y x y for every scalar
C.
For, let be any scalar, then
( x, y) = (x, y)
= .0
=0
x y x y.
(c) The zero vector is orthogonal to every vector. For every vector x in H, we have
(0, x) = 0
0 x for all x H.
(d) The zero vector is the only vector which is orthogonal to itself. For,
2
if x x (x, x) = 0 x =0 x=0
Hence, if x x, then x must be a zero vector.
24.1.2 Pythagorean Theorem
Statement: If x and y are any two orthogonal vectors in a Hilbert space H, then
2 2 2
x+y = x–y = x + y 2.
Proof: Given x y (x, y) = 0, then we must have
y x i.e. (y, x) = 0
2
Now x+y = (x + y, x + y)
= (x, x) + (x, y) + (y, x) + (y, y)

2 2
= x +0+0+ y
2 2
= x + y
2
Also, x–y = (x – y, x – y)
= (x, x) – (x, y) – (y, x) + (y, y)

2 2
= x –0–0– y
2 2
= x + y
2 2 2 2
x+y = x–y = x + y
24.1.3 Orthogonal Sets
Definition: A vector x is to be orthogonal to a non-empty subset S of a Hilbert space H, denoted

by x S if x y for every y in S.

Two non-empty subsets S1 and S2 of a Hilbert space H are said to be orthogonal denoted by Notes
S1 S2, if x y for every x S1 and every y S2.
24.1.4 Orthogonal Compliment: Definition
Let S be a non-empty subset of a Hilbert space H. The orthogonal compliment of S, written as S

and is read as S perpendicular, is defined as
S = {x H:x y y S}
Thus, S is the set of all those vectors in H which are orthogonal to every vectors in H which are
orthogonal to every vector in S.
Theorem 1: If S, S1, S2 are non-empty subsets of a Hilbert space H, then prove the following:
(a) {0} = H (b) H = {0} (c) S S {0}
(d) S1 S2 S2 S1 (e) S S
Proof:
(a) Since the orthogonal complement is only a subset of H, {0} H.
It remains to show that H {0} .
Let x H. Since (x, 0) = 0, therefore x {0} .
Thus x H x {0} .
H {0} .
Hence {0} = H
(b) Let x H. Then by definition of H, we have
(x, y) = 0 y H
Taking y = x, we get
2
(x, x) = 0 x =0 x=0
Thus x H x=0
H = {0}
(c) x S S.
Then x S and x S
Since x S , therefore x is orthogonal to every vector in S. In particular, x is orthogonal to

x because x S.
2
Now (x, x) = 0 x =0 x = 0.
0 is the only vector which can belong to both S and S .
S S {0}
If S is a subspace of H, then 0 S. Also S is a subspace of H. Therefore 0 S . Thus, if S is

a subspace of H, then 0 S S . Therefore, in this case S S = {0}.

Notes (d) Let S1 S2, we have
x S2 x is orthogonal to every vector in S2
x is orthogonal to every vector in S1 because S1 S2.
x S1
S2 S1
(e) Let x S. Then (x, y) = 0 y S.
by definition of (S ) , x (S ) .
Thus x S x S .
S S .
Theorem 2: If S is a non-empty subset of a Hilbert space H, then S is a closed linear subspace of

H and hence a Hilbert space.
Proof: We have
S = {x H : (x, y) = 0 y S} by definition. Since (0, y) = 0 y S, therefore at least 0 S and

thus S is non-empty.
Now let x1, x2 S and , be scalars. Then (x 1, y) = 0, (x2, y) = 0 for every y S.
For every y S, we have
( x1 + x2, y) = (x1, y) + (x2, y)
= (0) + (0)
=0
x1 + x2 S
S is a subspace of H.
Next we shall show that S is a closed subset of H.
Let (xn) S and xn x in H.
Then we have to show that x S.
For this we have to prove (x, y) = 0 for every y S.
Since xn S , (xn, y) = 0 for every y S and for n = 1, 2, 3, …
Since the inner product is a continuous function, we get
(xn, y) (x, y) as n
Since (xn, y) = 0 n, (x, y) = 0
x S.
Hence S is a closed subset of H.

Now S is a closed subspace of the Hilbert space H. Notes
So, S is complete and hence a Hilbert space. This completes the proof of the theorem.
Theorem 3: If M is a linear subspace of a Hilbert space H, then M is closed
M=M .
Proof: Let us assume that M = M ,
M being a subspace of H.
by theorem (2), (M ) is closed subspace of H.
Therefore M = M is a closed subspace of H. Conversely, let M be a closed subspace of H. We

shall show that M = M .
We know that M M .
Now suppose that M M .
Now M is a proper closed subspace of Hilbert space M . a non-zero vector zo in M such that
zo M or zo M .
Now zo M and M gives zo M M … (1)
Since M is a subspace of H, we have
M M = {0} … (2)
(by theorem 1 (iii))
From (1) and (2) we conclude that z = 0, a contradiction to the fact that z o is a non-zero vector.
M M can be a proper inclusion.
Hence M = M .
Cor. If M is a non-empty subset of a Hilbert space H, then M = M .
Proof: By theorem (2), M is a closed subspace of H. So by theorem (3),
M = (M ) =M .
Theorem 4: If M and N are closed linear subspace of a Hilbert space H such that M N, then the
linear subspace M N is closed.
Proof: To prove: M + N is closed, we have to prove that it contains all its limit point.
Let z be a limit point of M + N,
a sequence (zn) in M + N such that zn z in H.
Since M N, M N = {0} and M + N is the direct sum of the subspace M and N, zn can be written
uniquely as
zn = xn + yn where xn M and yn N.
Taking two points z m = xm + ym and zn = xn + yn, we have
zm – zn = (xm – xn) + (ym – yn).
Since xm – xn M and ym – yn N, we get
(xm – xn) (ym – yn)

Notes So, by Pythagorean theorem, we have

2 2
(xm – xn) + (ym – yn) = xm – xn + ym – yn 2.
But (xm – xn) + (ym – yn) = zm – zn so that

2 2 2
z m – zn = xm – xn + ym – yn … (1)
Since (zn) is a convergent sequence in H, it is a Cauchy sequence in H.

2
Hence zm – zn 0 as m, n … (2)
Using (2) in (1), we see that

2 2
xm – xn 0 and ym – yn 0
So that (xn) and (yn) are Cauchy sequence in M and N.
Since H is complete and M and N are closed subspace of a complete space H, M and N are
complete.
Hence, the Cauchy sequence (xn) in M converges to x in M and the Cauchy sequence (y n) in N
converges to y in N.
Now z = lim zn = lim (xn + yn)
= lim xn + lim yn
But lim xn + lim yn = x + y M+N
Thus, z =x+y M+N
M + N is closed.
24.1.5 The Orthogonal Decomposition Theorem or Projection Theorem
Theorem 5: If M is a closed linear subspace of a Hilbert space H, then H = M M.
Proof: If M is a subspace of a Hilbert space H, then we know that M M = {0}.
Therefore in order to show that

H=M M , we need to verify that
H=M+M.
Since M and M are closed subspace of H, M + M is also a closed subspace of H by theorem 4.
Let us take N = M + M and show that N = H.
From the definition of N, we get M N and M N. Hence by theorem (1), we have
N M and N M .
Hence N M M = {0}.
N = {0}
N = {0} = H … (1)
Since N = M + M is a closed subspace of H, we have by theorem (3),
N =N … (2)
From (1) and (2), we have
N = M + M = H.

Since M M = {0} and Notes
H =M+M ,
we have from the definition of the direct sum of subspaces,
H =M M.
Theorem 6: Let M be a proper closed linear sub space of a Hilbert space H. Then there exists a
non-zero vector zo in H such that zo M.
Proof: Since M is a proper subspace of H, there exists a vector x in H which is not in M.
Let d = d (x, M) = inf { x – y } : y M}.
Since x M, we have d > 0.
Also M is a proper closed subspace of H, then by theorem: “Let M be a closed linear subspace of
a Hilbert space H. Let x be a vector not in M and let d = d (x, m) (or d is the distance from x to M).
Then there exists a unique vector y o in M such that x – yo = d.”
There exists a vector yo in M such that
x – yo = d.
Let zo = x – yo. We then here
z o = x – yo = d > 0.
zo is a non-zero vector.
Now we claim that Zo M.
Let y be an arbitrary vector in M. We shall show that z o y. For any scalar , we have
zo – y = x – yo – y = x – (yo + y).
since M is a subspace of H and y o, y M,
yo + M M.
Then by definition of d, we have
x – (yo + y) d
Now zo – y = x – (yo + y) d = zo
2 2
zo – y zo
or (zo – y, zo – y) – (zo, zo) 0
or (zo, zo) – (zo, y) – (y, zo) + (y, y) – (zo, zo) 0
or (z o , y) (z o , y) (y, y) 0 … (1)
The above result is true for all scalars .
Let us take (z o , y) .
Putting the value of , in (1), we get
2 2
(z o , y) (z o , y) (z o , y) (z o , y) (z o , y) (z o , y) y 0
or –2 |(zo, y)|2 + 2
|(zo, y)|2 y 2
0

Notes or |(zo, y)|2 { y 2

– 2} 0 … (2)
The above result is true for all real suppose that (zo, y) 0. Then taking positive and so small
that y 2 < 2, we see from (2) that |(zo, y)|2 { y 2 – 2} < 0.
This contradicts (2).
Hence we must have (zo, yo) = 0 zo y, y M.
zo M.
24.2 Summary
 Let H be a Hilbert space. If x, y H then x is said to be orthogonal to y, written as x y, if

(x, y) = 0.
 If x and y are any two orthogonal vectors in a Hilbert space H, then

2 2 2
x+y = x–y = x + y 2.
 Two non-empty subsets S1 and S2 of a Hilbert space H are said to be orthogonal denoted by
S1 S2, if x y for every x S1 and every y S2.
 Let S be a non-empty subsets of a Hilbert space H. The orthogonal compliment of S,

written as S and is read as S perpendicular, is defined as
S = {x H:x y y S}
 The orthogonal decomposition theorem: If M is a closed linear subspace of a Hilbert space

H, then H = M M .
24.3 Keywords
Orthogonal Compliment: Let S be a non-empty subset of a Hilbert space H. The orthogonal

compliment of S, written as S and is read as S perpendicular, is defined as
S = {x H:x y y S}
Orthogonal Sets: A vector x is to be orthogonal to a non-empty subset S of a Hilbert space H,

denoted by x S if x y for every y in S.
Two non-empty subsets S1 and S2 of a Hilbert space H are said to be orthogonal denoted by S 1
S2, if x y for every x S1 and every y S2.
Orthogonal Vectors: Let H be a Hilbert space. If x, y H then x is said to be orthogonal to y,

written as x y, if (x, y) = 0.
Pythagorean Theorem: If x and y are any two orthogonal vectors in a Hilbert space H, then
2 2 2
x+y = x–y = x + y 2.
1. If S is a non-empty subset of a Hilbert space, show that S = S .
2. If M is a linear subspace of a Hilbert space, show that M is closed M=M .
3. If S is a non-empty subset of a Hilbert space H, show that the set of all linear combinations
of vectors in S is dense in H S = {0}.

4. If S is a non-empty subset of a Hilbert space H, show that S is the closure of the set of all Notes
linear combinations of vectors in S.
5. If M and N are closed linear subspace of a Hilbert space h such that M N, then the linear
subspace M + N is closed.
Books Halmos, Paul R. (1974), Finite-dimensional Vector Spaces, Berlin, New York
Paul Richard Halmos, A Hilbert Space Problem Book, 2nd Ed.
Online links Itcconline.net/green/courses/203/…/orthogonal complements.html

www.math.cornell.edu/~andreim/Lec33.pdf
www.amazon.co.uk

Notes Unit 25: Orthonormal Sets
CONTENTS
Objectives
Introduction
25.1 Orthonormal Sets
25.1.1 Unit Vector or Normal Vector
25.1.2 Orthonormal Sets, Definition
25.1.3 Examples of Orthonormal Sets
25.1.4 Theorems on Orthonormal Sets
25.2 Summary
25.3 Keywords
Objectives
 Understand orthonormal sets
 Define unit vector or normal vector
 Understand the theorems on orthonormal sets.
Introduction
In linear algebra two vectors in an inner product space are orthonormal if they are orthogonal
and both of unit length. A set of vectors from an orthonormal set if all vectors in the set are
mutually orthogonal and all of unit length.
In this unit, we shall study about orthonormal sets and its examples.
25.1 Orthonormal Sets
25.1.1 Unit Vector or Normal Vector
Definition: Let H be a Hilbert space. If x H is such that x = 1, i.e. (x, x) = 1, then x is said to be
a unit vector or normal vector.
25.1.2 Orthonormal Sets, Definition
A non-empty subset { ei } of a Hilbert space H is said to be an orthonormal set if
(i) i j ei ej, equivalently i j (ei, ej) = 0
(ii) ei = 1 or (ei, ej) = 1 for every i.

Unit 25: Orthonormal Sets
Thus a non-empty subset of a Hilbert space H is said to be an orthonormal set if it consists of Notes
mutually orthogonal unit vectors.
Notes
1. An orthonormal set cannot contain zero vector because 0 = 0.
2. Every Hilbert space H which is not equal to zero space possesses an orthonormal
set.
x
Since 0 x H. Then x 0. Let us normalise x by taking e = , so that
x
x 1
e = x = 1.
x x
e is a unit vector and the set {e} containing only one vector is necessarily an
orthonormal set.
xi
3. If {xi} is a non-empty set of mutually orthogonal vectors in H, then {e i} = is an
xi
orthonormal set.
25.1.3 Examples of Orthonormal Sets
1. In the Hilbert space  n2 , the subset e1, e2, …, en where ei is the i-tuple with 1 in the ith place
and O’s elsewhere is an orthonormal set.
For (ei, ej) = 0 i j and (ei, ej) = 1 in the inner product x i y i of  n2 .

i 1
2. In the Hilbert space  2 , the set {e1, e2, …, en, …} where en is a sequence with 1 in the n th place
and O’s elsewhere is an orthonormal set.
25.1.4 Theorems on Orthonormal Sets
Theorem 1: Let {e1, e2, …, en} be a finite orthonormal set in a Hilbert space H. If x is any vector in
H, then
n
2
2
(x, e i ) x ; … (1)
i 1
further,
n
x (x, e i ) e i ej for each j … (2)

i 1
Proof: Consider the vector

n
y = x (x, e i ) e i
i 1

Notes We have y 2
= (y, y)
n n
= x (x, e i ) e i , x (x, e i ) e i
i 1 i 1
n n
= (x, x) (x, e i ) (e i , x) (x, e j ) (x, e j )

i 1 j 1
n n
(x, e i ) (x, e i ) (e i , e j )
i 1 j 1
n n n
2
= x (x, e i ) (x, e i ) (x, e i ) (x, e j ) (x, e j ) (x, e i )
i 1 j 1 i 1
On summing with respect to j and remembering that (e i, ej) = 1, i = j and (ei, ej) = 0, i j
n n n
2 2 2 2
= x x, e i x, e i (x, e i )
i 1 i 1 i 1
n
2 2
= x (x, e i )
i 1
n
2
2 2
Now y 0, therefore x – (x, e i ) 0
i 1
n
2
2
(x, e i ) x
i 1
result (1).
Further to prove result (2), we have for each j (1 j n),
n n
x (x, ei ) ei , e j = (x, e j ) (x, e i ) e i , e j

i 1 i 1
= (x, e j ) (x, e i ) (e i , e j )
i 1
= (x, ej) – (x, ej) [ (ei, ej) = 1, i j 0, i = j]
=0
Hence x (x, e i ) e i e j for each j.

i 1
Note The result (1) is known as Bessel’s inequality for finite orthonormal sets.

Theorem 2: If {ei} is an orthonormal set in a Hilbert space H and if x is any vector in H, then the Notes
set S = {ei : (x, ei) 0} is either empty or countable.
Proof: For each positive integer n, consider the set
2
2 x
Sn = e i : (x, e i ) .
n
If the set Sn contains n or more than n’ vectors, then we must have

2
2 x
(x, ei ) >n x … (1)
ei S n
n
By theorem (1), we have
2
2
(x, ei ) x … (2)
ei S n
which contradicts (1).
Hence if (2) were to be valid, Sn should have at most (n – 1) elements. Hence for each positive n,
the set Sn is finite.
Now let ei S. Then (x, ei) 0. However small may be the value of |(x, e i)|2, we can take n so
large that
2
x
|(x, ei)|2 > .
n
Therefore if ei S, then ei must belong to some Sn. So, we can write S = S

n 1
n .
S can be expressed as a countable union of finite sets.
S is itself a countable set.
If (x, ei) = 0 for each i, then S is empty. Otherwise S is either a finite set or countable set.

Theorem 3: Bessel’s Inequality: If {ei} is an orthonormal set in a Hilbert space H, then |(x, ei)|2
x 2 for every vector x in H.
Proof: Let S = {ei : (x, ei) 0}.
By theorem (2), S is either empty or countable.
If S is empty, then (x, ei) = 0 i.
So if we define |(x, ei)|2 = 0, then
|(x, ei)|2 = 0 x 2.
Now let S is not empty, then S is finite or it is countably infinite.
If S is finite, then we can write S = {e1, e2, …, en} for some positive integer n.
In this case, we have

n
|(x, ei)|2 = |(x, e i )|2 x 2

… (1)
i 1
which represents Bessel’s inequality in the finite case.

Notes If S is countable infinite, let S be arranged in the definite order such as {e 1, e2, …, en, …}.
In this case we can write
|(x, ei)|2 = |(x, e n )|2 … (2)

n 1
The series on the R.H.S. of (2) is absolutely convergent.
Hence every series obtained from this by rearranging the terms is also convergent and all such
series have the same sum.
Therefore, we define the sum |(x, ei)|2 to be |(x, e n )|2 .

n 1
Hence the sum of |(x, ei)|2 is an extended non-negative real number which depends only on S
and not on the rearrangement of vectors.
Now by Bessel’s inequality in the finite case, we have
|(x, e i )|2 x 2
… (3)
i 1
For various values of n, the sum on the L.H.S. of (3) are non-negative. So they form a monotonic
increasing sequence. Since this sequence is bounded above by x 2, it converges. Since the
sequence is the sequence of partial sums of the series on the R.H.S. of (2), it converges and we
have ei S,
|(x, ei)|2 = |(x, e i )|2 x 2
n 1
Note: From the Bessel’s inequality, we note that the series |(x, e n )|2 is convergent series.
n 1
Corollary: If en S, then (x, en) 0 as n .
Proof: By Bessel’s inequality, the series |(x, e n )|2 is convergent.

n 1
Hence |(x, en)|2 0 as n .
(x, en) 0 as n .
Theorem 4: If {ei} is an orthonormal set in a Hilbert space H and x is an arbitrary vector in H, then
{x – (x, ei) ei} ej for each j.
Proof: Let S = {ei : (x, ei) 0}
Then S is empty or countable. [See theorem (2)]
If S is empty, then (x, ei) = 0 for every i.
In this case, we define (x, ei) ei to be a zero vector and so we get
x– (x, ei) ei = x – 0 = x.

Hence in this case, we have to show x ej for each j. Notes
Since S is empty, (x, ej) = 0 for every j.
x ej for every j.
Now let S is not empty. Then S is either finite or countably infinite. If S is finite, let
S = {e1, e2, …, en} and we define
(x, ei) ei = (x, e i ) e i ,

i 1
and prove that x (x, ei ) ei ej for each j. This result follows from (2) of theorem (1).
n 1
Finally let S be countably infinite and let
S = {e1, e2, …, en, …}
Let sn = (x, e i ) e i
i 1
2
m
For m > n, sm – sn 2
= (x, e i )e i
i n 1
m
2
= (x, e i )
i n 1
2
By Bessel’s inequality, the series (x, e n ) is convergent.
n 1
2
Hence (x, e i ) as m, n .
i n 1
2
sm – sn 0 as m, n .
(sn) is a Cauchy sequence in H.
Since H is complete sn s H. Now s H can be written as
s = (x, e n ) e n
n 1
Now we can define (x, ei) ei = (x, e n ) e n .

n 1
Before, completing the proof, we shall show that the above sum is well-defined and does not
depend upon the rearrangement of vectors.
For this, let the vector in S be arranged in a different manner as
S = {f1, f2, f3, …, fn, …}

Notes n
Let sn = (x, fi ) fi
i 1
As shown for the case above for (s n), let
sn s in H where we can take
s = (x, fn ) fn .
n 1
We prove that s = s . Given > 0 we can find n 0 such that n n0 .
2
x, e i < 2
, sn – s < , sn s < … (1)
i no 1
For some positive integer m0 > n0, we can find all the terms of s n in smo also.
Hence s mo s no contains only finite number of terms of the type (x, e i) ei for i = n0 + 1, n0 + 2, …
2 2
Thus, we get s mo s no x, e i so that we have
i no 1
s mo s no < … (2)
Now s –s = s s mo s mo s no s no s
s sm0 sm0 s n0 s n0 s
< + + =3 (Using (1) and (2))
Since > 0 is arbitrary, s – s = 0 or s = s .
Now consider
(x – (x, ei) ei, ej) = (x – s, ej)
But (x – s, ej) = (x, ej) – (s, ej)
= (x, ej) – (lim sn, ej) … (3)
By continuity of inner product, we get
(lim sn, ej) = lim (sn, ej) … (4)
Using (3) in (4), we obtain
(x – (x, ei) ei, ej) = (x, ej) – lim (sn, ej)
If ej S, then
(sn, ej) = (x, e i ) e i , e j =0

i 1
lim (s n , e j ) = 0
n

Hence (x – (x, ei) ei, ej) = (x, ej) = 0 since ej S. Notes
If ej S, then (sn, ej) = (x, e i ) e i , e j … (5)

i 1
If n > j, we get (x, e i ) e i , e j = (x, ej) … (6)

i 1
From (5) & (6), we get
lim (s n , e j ) = (x, ej).

n
So, in this case
(x – (x, ei) ei, ej) = (x, ej) – (x, ej) = 0
Thus (x – (x, ei) ei, ej) = 0 for each j.
Hence x – (x, ei) ei ej for each j.

Theorem 5: A Hilbert space H is separable every orthonormal set in H is countable.
Proof: Let H be separable with a countable dense subset D so that H = D .
Let B be an orthonormal basis for H.
We show that B is countable.
For x, y B, x y, we have
2 2 2
x–y = x + y =2
Hence the open sphere
1 1
S x; z: z x = as x B are all disjoint.
2 2
1
Since D is dense, D must contain a point in each S x, .
2
Hence if B is uncountable, then B must also be uncountable and H cannot be separable contradicting
the hypothesis. Therefore B must be countable.
Conversely, let B be countable and let B = {x1, x2, …}. Then H is equal to the closure of all finite
linear combinations of element of B. That is H = L(B) . Let G be a non-empty open subset of H.
n
Then G contains an element of the form a i x i with ai C. We can take ai C. We can take ai
i 1
to be complex number with real and imaginary parts as rational numbers. Then the set
D = ai xi , n 1, 2, , a i rational
i 1
is a countable dense set in H and so H is separable.

Notes Theorem 6: A orthonormal set in a Hilbert space is linear independent.
Proof: Let S be an orthonormal set in a Hilbert space H.
To show that S is linearly independent, we have to show that every finite subset of S is linearly
independent.
Let S1 = {e1, e2, …, en} be any finite subset of S.
Now let us consider
e +
1 1
e +…+
2 2 n n
e =0 … (1)
Taking the inner product with e k (1 k n),
n n
iei , ek = i (e i , e k ) … (2)
i 1 i 1
Using the fact that (ei, ek) = 0 for i k and (ek, ek) = 1, we get
i (e j , e k ) = k
… (3)
i 1
It follows from (2) on using (1) and (3) that
(0, ek) = k
k
=0 k = 1, 2, …, n.
S1 is linearly independent.
Example: If {ei} is an orthonormal set in a Hilbert space H, and if x, y are arbitrary vectors
in H, then (x, ei ) (y, ei ) x y .
Solution: Let S e i : (x, e i )(y, ei ) 0
Then S is either empty or countable.

If S is empty, then we have
(x, e i )(y, e i ) = 0 i
and in this case we define
2
(x, ei ) (y, ei ) to be number 0 and we have 0 x y 2.
If S is non-empty, then S is finite or it is countably infinite. If S is finite, then we can write

S = {e1, e2, …, en} for some positive integer n.
In this case we define
(x, ei ) (y, ei ) = (x, e i ) (y, e i )

i 1

1 1
Notes
n 2 n 2
2 2
(x, e i ) (y, e i ) (By Cauchy inequality)
i 1 i 1
2 2
x y (by Bessel’s inequality for finite case)
(x, e i ) (y, e i ) x y … (1)

i 1
Finally let S is countably infinite. Let the vectors in S be arranged in a definite order as
S = {e1, e2, …, en, …}.
Let us define
(x, ei ) (y, ei ) = (x, e n ) (y, e n ) .

i 1
But this sum will be well defined only if we can show that the series (x, e n ) (y, e n ) is
n 1
convergent and its sum does not change by rearranging its term i.e. by any arrangement of the
vectors in the set S.
Since (1) is true for every positive integer n, therefore it must be true in the limit. So
(x, e n ) (y, e n ) x y … (2)

n 1
From (2), we see that the series (x, e n ) (y, e n ) is convergent. Since all the terms of the series
n 1
are positive, therefore it is absolutely convergent and so its sum will not change by any
rearrangement of its terms. So, we are justified in defining
(x, ei ) (y, ei ) = (x, e n ) (y, e n )

n 1
and from (2), we see that this sum is x y .
25.2 Summary
 Two vectors in an inner product space are orthonormal if they are orthogonal and both of
unit length. A set of vectors from an orthonormal set if all vectors in the set are mutually
orthogonal and all of unit length.
 Examples of orthonormal sets are as follows:
(i) In the Hilbert space  n2 , the subset e1, e2, …, en where ei is the i-tuple with 1 in the ith
place and O’s elsewhere is an orthonormal set.
n
For (ei, ej) = 0 i j and (ei, ej) = 1 in the inner product xi y i of  n2 .

i 1

Notes (ii) In the Hilbert space  2 , the set {e1, e2, …, en, …} where en is a sequence with 1 in the
nth place and O’s elsewhere is an orthonormal set.
25.3 Keywords
Orthonormal Sets: A non-empty subset { ei } of a Hilbert space H is said to be an orthonormal set

if
(i) i j ei ej, equivalently i j (ei, ej) = 0
(ii) ei = 1 or (ei, ej) = 1 for every i.
Unit Vector or Normal Vector: Let H be a Hilbert space. If x H is such that x = 1, i.e. (x, x) =
1, then x is said to be a unit vector or normal vector.
1. Let {e1, e2, …, en} be a finite orthonormal set in a Hilbert space H, and let x be a vector in H.
n
If 1
, 2
, …, n
are arbitrary scalars, show that x i ei attains its minimum value
i 1
i
= (x, ei) for each i.
2. Prove that a Hilbert space H is separable every orthonormal set in H is countable.
Book Sheldon Axler, Linear Algebra Done Right (2nd ed.), Berlin, New York (1997).
Online links www.mth.kcl.ac.uk/~jerdos/op/w3.pdf

mathworld.wolfram.com
www.utdallas.edu/dept/abp/PDF_files

Unit 26: The Conjugate Space H*
Notes
CONTENTS
Objectives
Introduction
26.1 The Conjugate Space H*
26.1.1 Definition
26.2 Summary
26.3 Keywords
Objectives
 Define the conjugate space H*.
 Understand theorems on it.
 Solve problems related to conjugate space H*.
Introduction
Let H be a Hilbert space. A continuous linear transformation from H into C is called a continuous
linear functional or more briefly a functional on H. Thus if we say that f is a functional on H, then
f will be continuous linear functional on H. The set H,C of all continuous linear functional on
H is denoted by H* and is called the conjugate space of H. The elements of H* are called continuous
linear functional or more briefly functionals. We shall see that the conjugate space of a Hilbert
space H is the conjugate space H* of H is in some sense is same as H itself. After establishing a
correspondence between H and H *, we shall establish the Riesz representation theorem for
continuous linear functionals. Thereafter we shall prove that H * is itself a Hilbert space and H is
reflexive, i.e. has a natural correspondence between H and H ** and this natural correspondence
is an isometric isomorphism of H onto H**.
26.1 The Conjugate Space H*
26.1.1 Definition
Let H be a Hilbert space. If f is a functional on H, then f will be continuous linear functional on

H. The set H,C of all continuous linear functional on H is denoted by H * and is called the
conjugate space of H. The conjugate space of a Hilbert space H is the conjugate space H * of H is in
some sense is same as H itself.

Notes 26.1.2 Theorems and Solved Examples
Theorem 1: Let y be a fixed vector in a Hilbert space H and let fy be a scalar valued function on
H defined by
fy x x, y x H.
Then fy is a functional in H * i.e. fy is a continuous linear functional on H and y fy .
Proof: From the definition
fy : H C defined as fy x x, y x H.
We prove that fy is linear and continuous so that it is a functional.
Let x1 ,x2 H and , be any two scalars. Then for any fixed y H,
fy x1 x2 x1 x2 , y
x1 , y x2 , y
fy x1 fy x2
fy is linear.
To show fy is continuous, for any x H
fy x x, y x.y ...(1)
(Schwarz inequality)
Let y M. Then for M > 0
fy x M x so that fy is bounded and hence fy is continuous.
Now let y = 0, y 0 and from the definition fy = 0 so that fy y .
Sup fy x
Further let y 0. Then from (1) we have y.
x
Hence using the definition of the norm of a functional,
we get fy y ...(2)
Further fy sup fy x : x 1 ...(3)
y
Since y 0, is a unit vector.
y
From (3), we get
y
fy fy ...(4)
y

Notes
y y 1
But fy ,y y, y y ...(5)
y y y
Using (5) in (4) we obtain
fy y
fy y
Theorem 2: (Riesz-representation Theorem for Continuous Linear Functional on a Hilbert Space):

Let H be a Hilbert space and let f be an arbitrary functional on H *. Then there exists a unique
vector y in H such that
f = fy, i.e. f(x) = (x,y) for every vector x H and f y.
Proof: We prove the following three steps to prove the theorem.
Step 1: Here we show that any f H * has the representation f = fy.
If f = 0 we take y = 0 so that result follows trivially.
So let us take f 0.
We note the following properties of y in representation if it exists. First of all y 0, since

otherwise f = 0.
Further (x,y) = 0 x for which f(x) = 0. This means that if x belongs to the null space N(f) of f, then
(x,y) = 0.
y N f .
So let us consider the null space N(f) of f. Since f is continuous, we know that N(f) is a proper
closed subspace and since f 0,N f H and so N f 0 .
Hence by the orthogonal decomposition theorem, ay0 0 in N f . Let us define any

arbitrary x H.
z f x y0 f y0 x
Now f z f x f y0 f y0 f x 0
z N f .
Since y 0 N f , we get
0 z, y0 f x y0 f y0 x, y 0
f x y0 , y0 f y 0 x, y 0
Hence we get
f x y0 , y0 f y0 x,y0 0 ...(3)

Notes 2
Noting that y 0 , y 0 y0 0, we get from (3),
f y0
f x 2 x, y 0 ...(4)
y0
We can write (4) as
f y0
f x x, 2 y0
y0
f yo
Now taking y o as y, we have established that there exists a y such that f(x) = (x,y) for x H.
yo
Step 2: In this step we know that
f y
If f =0, then y = 0 and f y hold good.
Hence let f 0. Then y 0.
From the relation f(x) = (x,y) and Schwarz inequality we have
f x x, y x y.
f x
sup y.
x 0 x
Using definition of norm of f, we get from above
f y ...(5)
Now let us take x = y in f(x) = (x,y), we get
2
y y, y f y f y
y f ...(6)
(5) and (6) implies that
f y.
Step 3: We establish the uniqueness of y in f(x) = (x,y). Let us assume that y is not unique in
f(x) = (x,y).
Let for all x H, y 1 , y 2 such that
f(x) = (x,y1) = (x,y2)
Then (x,y1) – (x,y2) = 0
(x,y1 – y2) = 0 x H.

Let us choose x to be y1 – y2 so that Notes
2
y1 y 2 , y1 y 2 y1 y 2 0
y1 y 2 0
y1 y2
y is unique in the representation of f(x) = (x,y)
Note The above Riesz representation theorem does not hold in an inner product space
which is not complete as shown by the example given below. In other words the
completeness assumption cannot be dropped in the above theorem.
Example: Let us consider the subspace M of l2 consisting of all finite sequences. This is the
set of all scalar sequence whose terms are zero after a finite stage. It is an incomplete inner
product space with inner product
x, y x n y n x, y M
n 1
Now let us define
xn
f x as x xn M.
n 1
n
Linearity of f together with Hölder’s inequality yields

2
2 xn 1 2
f x xn
n 1
n n 1
n2 n 1
2 2
2
x, x x ,
6 6
2
1
since .
n 1
n2 6
f is a continuous linear functional on M.
We now prove that there is no y M such that
f x x, y x M.
Let us take x = en = (0,0,....,1,0,0,.....) where 1 is in n th place.

Notes 1
Using the definition of f we have f(x) = .
n
Suppose y yn M satisfying the condition of the theorem, then
f x x, y xn yn y n as x en .
1
Thus Riesz representation theorem is valid if and only if y n 0 for every n.
n
Hence y yn M.
no y M such that f(x,y) = (x,y) for every x H.
the completeness assumption cannot be left out from the Riesz-representation theorem.
Theorem 3: The mapping :H H* defined by :H H* defined by y fy where

fy(x) = (x,y) for every x H is an (i) additive, (ii) one-to-one, (iii) onto, (iv) symmetry, (v) not
linear.
Proof:
(i) Let us show that is additive, i.e.,
y1 y2 y1 y 2 for y 1 , y 2 H.
Now from the definition y1 y2 fy1 y2
Hence for every x H, we get
fy x x, y 1 y2 x, y 1 x, y 2
1 y2
fy 1 x fy x
2
fy y1 y2 fy fy y1 y2
1 y2 1 2
(i) is one-to-one. Let y1 ,y2 H
Then y 1 =fy and y 2 =fy . Then

1 2
y 1 = (y 2 ) fy = fy
1 2
fy x = fy x x H. ...(1)
1 2
fy x = x,y 1 and fy x x, y 2
1 2
from (1), we get
x, y1 x, y2 x, y1 x, y2 0
x, y1 y2 0 x H ...(2)
Choose x = y1 – y2 then from (2) if follows that (y 1 – y2,y1 – y2) = 0

y1 y 2
2
0 Notes
y1 y2
is one-to-one.
(iii) is onto: Let f H* . Then y H such that
f(x) = (x,y)
since f(x) = (x,y) we get
f = fy so that y = fy = f.
Hence for f H * , a pre-image y H. Therefore is onto.
(iv) is isometry; let y1 ,y2 H, then
y1 y2 fy fy
1 2
fy f y2
1
But fy f y2 fy y1 y2 (By theorem (1))

1 1 y2
Hence y1 y2 y1 y2 .
(v) To show is not linear, let y H and be any scalar. Then ,y f y. Hence for any
x H, we get
f y (x) (x, y) (x, y) fy (x)
f y fy
y y
is not linear. Such a mapping is called conjugate linear.
Note: The above correspondence is referred to as natural correspondence between H and

H* .
Theorem 4: If H is a Hilbert space, then H* is also an Hilbert space with the inner product defined
by
(fx, fy) = (y,x) … (1)
Proof: We shall first verify that (1) satisfies the condition of an inner product.
Let x, y H and , be complex scalars.
(i) We know (see Theorem 3) that
f y fy
f y fy fy .
Now fx fy ,fz fx f y ,fz ... (2)

Notes
But f x f y , fz z, x, y (by (1))
Now z, x y z, x z, y
fx , fz fy , fz … (3)
From (2) and (3) it follows that fx fy , fz fx , fz fy , fz
(ii) fx , fy y, x x, y fx , fy .
2 2
(iii) fx , fx x, x x fx so fx , fx 0 and fx 0 fx 0.
i iii implies that (1) represents an inner product. Now the Hilbert space H is a complete
normed linear space. Hence its conjugate space H * is a Banach space with respect to the norm
defined on H*. Since the norm on H* is induced by the inner product, H* is a Hilbert space with
the inner product (fx, fy) = (y,x)
Cor. The conjugate space H** of H* is a Hilbert space with the inner product defined as follows:
If f ,g H * , let Ff and Fg be the corresponding elements of H** obtained by the Riesz representation
theorem.
Then (Ff,Fg) = (g,f) defines the inner product of H**.
Theorem 5: Every Hilbert space is reflexive.
Proof: We are to show that the natural imbedding on H and H ** is an isometric isomorphism.
Let x be any fixed element of H. Let Fx be a scalar valued function defined on H* by Fx(f) = f(x) for
every f H * . We have already shown in the unit of Banach spaces that Fx H * * . Thus each vector
x H gives rise to a functional F x in H**. Fx is called a functional on H * induced by the vector x.
Let J : H H ** be defined by J x Fx for every x H.
We have also shown in chapter of Banach spaces that J is an isometric isomorphism of H into H **.
We shall show that J maps H onto H**.
int o
Let T1 : H H * defined by
T1 x fx ,fx y y,x for every y H.

int o
and T2 : H * H ** defined by
T2 fx Ffx ,Ffx f f, fx for f H* .
Then T2.T1 is a composition of T2 and T1 from H to H**. By Theorem 3, T 1,T2 are one-to-one and
onto.
Hence T2.T1 is same as the natural imbedding J.
For this we show that J(x) = (T 2.T1)x for every x H.
Now (T2.T1)x = T2(T1(x)) = T2(fx) = Ffx .

By definition of J, J(x) = Fx. Hence to show T2.T1 = J, we have to prove that Fx = Ffx . Notes
For this let f H * . Then f = f y where f corresponds to y in the representation

Ffx (f) (f, fx ) (fy , fx ) (x, y) .
But (x,y) = fy(x) = f(x) = Fx(f).
Thus we get Ffx (f) Fx (f) for every f H* .
Hence the mapping Ff and Fx are equal.

x
T2.T1 = J and J is a mapping of H onto H**, so that H is reflexive.
Notes
1. Since Fx Ffx x H (From above theorem)
Fx ,Fy Ffy ,Ffy fy , fx x, y by using def. of inner product on H** and by the
*
def. of inner product on H .
2. Since an isometric isomorphism of the Hilbert space H onto Hilbert space H **,
therefore we can say that Hilbert space H and H** are congruent i.e. they are equivalent
metrically as well as algebraically. We can identify the space H ** with the space H.
26.2 Summary
 Let H be a Hilbert space. If f is a functional on H, then f will be continuous linear functional

on H. The set H,C of all continuous linear functional on H is denoted by H * and is called
conjugate space of H. Conjugate space of a Hilbert space H is the conjugate space H * of H.
 Riesz-representation theorem for continuous linear functional on Hilbert space:
Let H be a Hilbert space and let f be an arbitrary functional on H *. Then there exists a
unique vector y in H such that f = fy, i.e. f(x) = (x,y) for every vector x H and f y.
26.3 Keywords
Continuous Linear Functionals: Let N be a normal linear space. Then we know that the set R of
real numbers and the set C of complex numbers are Banach spaces with the norm of any
x R or x C given by the absolute value of x. We denote the BANACH space N,R or N,C
by N*.
The elements of N* will be referred to as continuous linear functionals on N.
Hilbert space: A complete inner product space is called a Hilbert space.
Let H be a complex Banach space whose norm arises from an inner product which is a complex
function denoted by (x,y) satisfying the following conditions:

Notes x y,z x,z y,z

H1:
H2: x, y y,x
2
H3: x,x x
for all x, y,z H and for all , C.
Inner Product: Let X be a linear space over the field of complex numbers C. An inner product on
X is a mapping from X × X C which satisfies the following conditions:
(i) ( x + y, z) = (x, z) + (y, z) x, y, z X and , C.
(ii) (x, y) = (y, x)
(iii) (x, x) 0, (x, x) = 0 x=0
Riesz-representation Theorem for Continuous Linear Functional on a Hilbert Space: Let H be a

Hilbert space and let f be an arbitrary functional on H *. Then there exists a unique vector y in H
such that
f = fy, i.e. f(x) = (x,y) for every vector x H and f y.
The Conjugate Space H*: Let H be a Hilbert space. If f is a functional on H, then f will be
continuous linear functional on H. The set H,C of all continuous linear functional on H is
denoted by H* and is called the conjugate space of H. The conjugate space of a Hilbert space H is
the conjugate space H* of H is in some sense is same as H itself.
1. Let H be a Hilbert space, and show that H* is also a Hilbert space with respect to the inner
product defined by (fx, fy) = (y, x). In just the same way, the fact that H* is a Hilbert space
implies that H** is a Hilbert space whose inner product is given by (F f, Fg) = (g, f).
2. Let H be a Hilbert space. We have two natural mappings of H onto H**, the second of
which is onto: the Banach space natural imbedding x Fx, where fx (y) = (y, x) and
Ffx (f) (F, fx ). Show that these mappings are equal, and conclude that H is reflexive. Show
that (Fx, Fy) = (x, y).
Books Hausmann, Holm and Puppe, Algebraic and Geometric Topology, Vol. 5, (2005)
K. Yosida, Functional Analysis, Academic Press, 1965.
Online links www.spot.colorado.edu
www.arvix.org

Unit 27: The Adjoint of an Operator
Notes
CONTENTS
Objectives
Introduction
27.1 Adjoint of an Operator
27.2 Summary
27.3 Keywords
Objectives
 Define the adjoint of an operator.
 Understand theorems on adjoint of an operator.
 Solve problems on adjoint of an operator.
Introduction
We have already proved that T gives rise to an unique operator T* and H* such that (T*f) (x) =
f(Tx) f H * and x H. The operator T* on H* is called the conjugate of the operator T on H.
In the definition of conjugate T* of T, we have never made use of the correspondence between H
and H*. Now we make use of this correspondence to define the operator T* on H called the
adjoint of T. Though we are using the same symbol for the conjugate and adjoint operator on H,
one should note that the conjugate operator is defined on H*, while the adjoint is defined on H.
27.1 Adjoint of an Operator
Let T be an operator on Hilbert space H. Then there exists a unique operator T* on H such that
(Tx,y)= (x,T*y) for all x, y H
The operator T* is called the adjoint of the operator T.
Theorem 1: Let T be an operator on Hilbert space H. Then there exists a unique operator T* on H
such that
(Tx,y)= (x,T*y) for all x, y H ...(1)
Proof: First we prove that if T is an operator on H, there exists a mapping T* on H onto itself
satisfying
(Tx,y)= (x,T*y) for all x, y H. ...(2)

Notes Let y be a vector in H and fy its corresponding functional in H*.
Let us define
int o
T* : H H * by
T* : fy fz ...(3)
Under the natural correspondence between H and H*, let z H corresponding to fz H * . Thus
starting with a vector y in H, we arrive at a vector z in H in the following manner:
y fy T * fy fz z, ...(4)
where T* : H* H * and y fy and z fz are on H to H*
under the natural correspondence. The product of the above three mappings exists and it is
denoted by T*.
Then T* is a mapping on H into H such that
T * y z.
We define this T* to be the adjoint of T. We note that if we identify H and H* by the natural
correspondence y fy , then the conjugate of T and the adjoint of T are one and the same.
After establishing, the existence of T*, we now show (1). For x H, by the definition of the
conjugate T* on an operator T,
T * fy x fy Tx ...(5)
By Riesz representation theorem,
y fy so that
fy Tx Tx, y ...(6)
Since T* is defined on H*, we get
T * fy x fz x x,z ...(7)
But we have from our definition T*y = z ...(8)
T * fy x Tx, y ...(9)
T * fy x x, T * y ...(10)
From (9) and (10), we thus obtain
Tx, y x,T * y x, y H.

Notes
Note The relation Tx, y x, T * y can be equivalently written as
T * x, y x,Ty since
T * x, y y, T * x = Ty, x = x, Ty = x, Ty
T * x, y x, Ty .
Example: Find adjoint of T if T is defined on  2 as Tx 0, x1 , x 2 ,... for every x xn 2.
Let T* be the adjoint of T. Using inner product in  2 , we have
T * x, y = x, Ty
since Ty 0, y1 , y 2 ,... , we have
T * x, y x, Ty = x n+1 y n Sx, y ,
n=1
where S x x 2 , x 3 ,...
Hence T * x, y Sx, y for every x in  2 .
Since T* is unique, T*=S so that we have
T* x x 2 , x 3 , x 4 ,... .
Theorem 2: Let H be the given Hilbert space and T* be adjoint of the operator T. Then T* is a
bounded linear transformation and T determine T* uniquely.
Proof: T* is linear.
Let y 1 , y 2 H and , be scalars. Then for x H, we have
x,T * y1 y2 Tx, y1 y2
But Tx, y1 y2 Tx, y1 Tx, y 2
Tx, y1 x,T * y 2
x, T * y1 x, T * y 2 .
Hence for any x H,
x,T * y1 y2 x, T * y1 x, T * y 2

Notes
x, T * y 1 T * y2 .
T* is linear.
T* is bounded
for any y H, let us consider
2
T*y T * y, T * y
TT * yy
TT * y y using Schwarz inequality
T T*y y
2
Hence T * y T T*y y 0 ...(1)
If T * y 0 then T * y T y because T y 0
Hence let T * y 0.
Then we get from (1)
T*y T y.
since T is bounded,
T M so that
T*y M y for every y H.
T* is bounded.
T* is continuous.
Uniqueness of T*.
Let if T* is not unique, let T’ be another mapping of H into H with property
Tx, y = x,T * y x, y H.
Then we have
Tx, y = x, T'y ...(2)
and Tx, y = x,T * y ...(3)
x,T'y = x,T * y x, y H
x, T'y T * y =0

Notes
x, T' T * y =0 x H
T ' T * y = 0 for every y H
Hence T'y = T * y for every y H.
T T *.
Notes
1. We note that the zero operator and the identity operator I are adjoint operators. For,
(i) x,0 * y 0x, y 0, y 0 x,0 = x,0y
so from uniqueness of adjoint 0* = 0.
(ii) (x,Iy) = (Ix,y) = (x,y) = (x,Iy)
so from uniqueness of adjoint I*=I.
2. If H is only an inner product space which is not complete, the existence of T*

corresponding to T in the above theorem is not guaranteed as shown by the following
example.
Example: Let M be a subspace of L2 consisting of all real sequences, each one containing
only finitely many non-zero terms. M is an incomplete inner product space with the same inner
product for  2 given by
x, y = xn yn ...(1)
n=1
For each x M, define
xn
T x ,0,0,...... ...(2)
n=1
n
Then from the definition, for x,y M,
xn
T x, y y1 .
n=1
n
Now let en 0,0,...,1,0,... where 1 is in the n th place.

Then using (3) we obtain
e n ( j) 1
Ten ,e 1 1. 1. .
j n

Notes
Now we check whether there is T* which is adjoint of T. Now e n ,T * e1 T * e1 .e n , where the
R.H.S. gives the component wise inner product. Since T * e 1 M, T * e 1 .e n cannot be equal to
1
n 1, 2,...
n
there is no T* on M such that
T e n ,e1 e n ,T * e1
Hence completeness assumption cannot be ignored from the hypothesis.
Notes
1. The mapping T T * is called the adjoint operation on H.

2. From Theorem (2), we see that the adjoint operation is mapping T T * on H
into itself.
Theorem 3: The adjoint operation T T* on (H) has the following properties:
(i) T1 T2 * T *1 T 2* (preserve addition)
(ii) T1 T2 * T *2 T *1 (reverses the product)
(iii) T * T* (conjugate linear)
(iv) T* T
2
(v) T*T T
Proof: (i) For every x, y H, we have
(x, (T1 + T2)*y = ((T1 + T2) x, y) (By def. of adjoint)
= (T1x + T2x, y)
= (T1x, y) + (T2x, y)
= (x, T1*y) + (x, T2*y)
= (x, T1*y + T2*y)
= (x, (T1* + T2*)y)
(T1 +T2)* = T1* + T2* by uniqueness of adjoint operator
(ii) For every x, y H, we have
x T1T2 * y T1T2 x, y
T1 T2 x , y

Notes
T2 x,T1 * y
x, T2* , T1* y
x, T2* T1* , y
Therefore from the uniqueness of adjoint operator, we have T1 T2 * T2* T1* .
(iii) For every x,y H, we have
x, T *y T x, y Tx , y
Tx, y
x, T * y x, T*y
x T* y .
Therefore from the uniqueness of adjoint operator, we have
T* T *.
(iv) For every y H we have

2
T*y T * y,T * y
TT * y, y
2
 T*y TT * y, y is a real number 0
TT * y, y
TT * y y By Schwarz inequality
T T*y y  Tx T x
2
Thus T * y T T*y y y H
T*y T y y H. ...(1)
Now T * Sup T * y : y 1
from (1), we see that if y 1 then T * y T
T* T ...(2)
Now applying (2) from the operator T* in place of operator T, we get
T* * T*

Notes
T** T*
T T* T * * T ...(3)
T T* .
(v) We have T * T T* T
T T  T* T
2
T ...(4)
Further for every x H, we have
2
Tx Tx, Tx
T * Tx, x
T * T x, x
T*T x x (By Schwarz inequality)
2
T*T x
Then we have
2 2
Tx T*T x x H ...(5)
Now T sup Tx : x 1
2 2
T sup Tx : x 1
2
sup Tx : x 1
From (5) we see that
2
if x 1 , then Tx T*T .
2
Therefore, Sup Tx : x 1 T*T
2
T T*T . ...(6)

2
T*T T .

Cor: If Tn is a sequence of bounded linear operators on a Hilbert space and Notes
Tn T, then T n* T *.
We have
T *n T* Tn T *
Tn T (By properties of T*)
Since Tn T as n
Tn* T * as n .
Theorem 4: The adjoint operation on H is one-to-one and onto. If T is a non-singular operator

on H, then T* is also non-singular and
1 1
T* T *.
Proof: Let : H H is defined by
T T * for every T H .
To show is one-to-one, let T1 , T2 H . Then we shall show that T1 T2 T1 T2 .
Now T1 T2
T *1 T *2
T *1 * T *2 * (using Theorem 4. prop (iv))
T1 T2
is one-to-one.
is onto:
For T* H , we have on using Theorem 4 (iv),
T * = T* * =T.
Thus for every T* H , there is a T* H such that
T* T is onto.
Next let T be non-singular operator on H. Then its inverse T 1 exists on H and

1
TT T 1T I.

Notes Taking the adjoint on both sides of the above, we obtain

1
TT * T 1T * I * .
By using Theorem 4 and note (2) under Theorem 2, we obtain

1 1
T * T* T* T * I.
T * is invertible and hence non-singular.

Further from the above, we conclude
1 1
T* T *.
Note From the properties of the adjoint operation T T * on H discussed in Theorems

(3) and (4), we conclude that the adjoint operation T T * is one-to-one conjugate linear
mapping on H into itself.
Example: Show that the adjoint operation is one-to-one onto as a mapping of (H) into
itself.
Solution: Let : (H) (H) be defined
(T) = T* T (H)
We show is one-to-one and onto.
is one-one:
Let T1, T2 (H). Then
(T1) = (T2) T1* = T2*
(T1*)* = (T2*)*
T1** = T2**
T1 = T2
is one-to-one.
is onto:
Let T be any arbitrary member of (H). Then T* (H) and we have (T*) = (T*)* = T** = T. Hence,
the mapping is onto.
27.2 Summary
 Let T be an operator on Hilbert Space H. Then there exists a unique operator T* on H such
that Tx,y x,T * y for all x,y H. The operator T* is called the adjoint of the operator T.

 The adjoint operation T T * on H has the following properties: Notes
(i) T1 T2 * T *1 T 2*
(ii) T1 T2 * T *2 T *1
(iii) T * T*
(iv) T* T
2
(v) T*T T
27.3 Keywords
Adjoint of the Operator T: Let T be an operator on Hilbert space H. Then there exists a unique
operator T* on H such that
(Tx,y)= (x,T*y) for all x, y H
Conjugate of the Operator T on H: T gives rise to an unique operator T* and H* such that (T*f) (x)
= f(Tx) f H * and x H. The operator T* on H* is called the conjugate of the operator T on H.
1. Show that the adjoint operation is one-to-one onto as a mapping of H into itself.
2
2. Show that TT * T .
3. Show that O*=O and I*=I. Use the latter to show that if T is non-singular, then T* is also
1 1
non-singular, and that in this case T * T *.
Books N.I. Akhiezer and I.M. Glazman, Theory of Linear Operators in Hilbert Space, Vol. II,
Pitman, 1981.
K. Yosida, Functional Analysis, Academic Press, 1965.
Online links www.math.osu.edu/  gerlach.1/math.BVtypset/node 78.html.
sepwww.standford.edu/sep/prof/pvi/conj/paper_html/node10.html.

Notes
Unit 28: Self Adjoint Operators
CONTENTS
Objectives
Introduction
28.1 Self Adjoint Operator
28.1.1 Definition: Self Adjoint
28.1.2 Definition: Positive operator
28.2 Summary
28.3 Keywords
Objectives
 Define self adjoint operator.
 Define positive operator.
 Solve problems on self adjoint operator.
Introduction
The properties of complex number with conjugate mapping z z motivate for the introduction
of the self-adjoint operators. The mapping z z of complex plane into itself behaves like the
adjoint operation in H as defined earlier. The operation z z has all the properties of the
adjoint operation. We know that the complex number is real iff z z . Analogue to this
characterization in H leads to the motion of self-adjoint operators in the Hilbert space.
28.1 Self Adjoint Operator
28.1.1 Definition: Self Adjoint
An operator T on a Hilbert space H is said to be self adjoint if T*=T.
We observe from the definition the following properties:
(i) O and I are self adjoint  O* O and I* I
(ii) An operator T on H is self adjoint if
Tx, y x,Ty x, y H and conversely.

If T* is an adjoint operator T on H then we know from the definition that Notes
Tx, y x,T * y x, y H .
If T is self-adjoint, then T = T*.
Tx, y x,Ty x, y H.
Conversely, if Tx, y x,Ty x, y H then we show that T is self-adjoint.
If T* is adjoint of T then (Tx, y) = (x, T*y)
We have x,Ty x,T * y
x, T T* y 0 x, y H
But since x 0 T T* y 0 y H
T = T*
T is self adjoint.
(iii) For any T H , T T * and T * T are self adjoint. By the property of self-adjoint operators,
we have
T T* * T* T**
T* T
T T*
T T* * T T*,
and T * T * T*T** T*T
T*T* * T * T.
Hence T T * and T * T are self adjoint.
Theorem 1: If An is a sequence of self-adjoint operators on a Hilbert space H and if An

converges to an operator A, then A is self adjoint.
Proof: Let An be a sequence of self adjoint operators and let An A.
A n is self adjoint A n* A n for n 1,2,...
We claim that A A *
Now A A* A A n A n A n* A n* A *
A A* A An An A *n An A *
An A An An An A  A n* An
An A 0 An A

Notes = 2 An – A
0 as n
A A* 0 or A A* 0 A A*
A is self-adjoint operator.
Theorem 2: Let S be the set of all self-adjoint operators in (H). Then S is a closed linear subspace
of (H) and therefore S is a real Banach space containing the identity transformation.
Proof: Clearly S is a non-empty subset of (H), since O is self adjoint operator i.e. O S.
H , since O is self adjoint operator i.e. O S.
Let A1 ,A 2 S, We prove that A1 A2 S.
A1 ,A 2 S A *1 A1 and A 2* A2 ...(1)
For , R , we have
A1 A2 * A1 * A2 *
A *1 A *2
A1 A2  , are real numbers, ,
A1 A 2 is also a self adjoint operator on H.
A 1 ,A 2 S A1 A2 S.
S is a real linear subspace of H .
Now to show that S is a closed subset of the Banach space H . Let A be any limit point of S.
Then a sequence of operator An is such that A n A. We shall show that A S i.e. A A *.
Let us consider
A A* A An An A*
A An An A*
A An An A *n A *n A*
A An An A n* A *n A*
An A An A n An A * An S A n* An
An A 0 An A

Notes
2 An A 0 0, T * T and T T
0 as An A
A A* 0 A A* 0
A A* A is self adjoint
A S
S is closed.
Now since S is a closed linear subspace of the Banach space H , therefore S is a real Banach
space. ( S is a complete linear space)
Also I* I the identity operator I S.
Theorem 3: If A1 ,A 2 are self-adjoint operators, then their product A1 ,A 2 is self adjoint
A1 ,A2 A2 ,A1 (i.e. they commute)
Proof: Let A1 ,A 2 be two self adjoint operators in H.
Then A *1 A 1 ,A *2 A2 .
Let A1 ,A 2 commute, we claim that A1 ,A 2 is self-adjoint.
A 1 ,A 2 * A *2 A *1 A 2A 1 A 1A 2
A 1 ,A 2 * A 1A 2
A 1A 2 is self adjoint.
Conversely, let A 1A 2 is self adjoint, then
A 1A 2 * A 1A 2
A *2 A *1 A1A 2
A1 ,A 2 commute
Theorem 4: If T is an operator on a Hilbert space H, then T = T 0 Tx, y 0 x, y H.
Proof: Let T = 0 (i.e. zero operator). Then for all x and y we have
T x, y Ox, y O, y O.
Conversely, Tx, y O x, y H
Tx, Tx O x, y H (taking y = Tx)
Tx O x, y H

Notes T O i.e. zero operator.

Theorem 5: If T is an operator on a Hilbert space H, then
Tx, x 0 x in H T O.
Proof: Let T = O. Then for all x in H, we have
Tx, x Ox, x 0, x 0.
Conversely, let Tx, x 0 x, y H. Then we show that T is the zero operator on H.
If , any two scalars and x, y are any vectors in H, then
T x y , x y Tx Ty, x y
Tx, x y Ty , x y
Tx,x Tx, y Ty,x Ty,x
2 2
Tx,x Tx, y Ty,x Ty,x
2 2
T x y , x y Tx,x Ty, y Tx, y Ty,x ...(1)
But by hypothesis Tx, x 0 x H.
L.H.S. of (1) is zero, consequently the R.H.S. of (1) is also zero. Thus we have
Tx, y Ty,x 0 ...(2)
for all scalars , and x, y H.
Putting 1, 1 in (2) we get
Tx, y Ty,x 0 ...(3)
Again putting i, 1 in (2) we obtain
i Tx, y i Ty,x 0 ...(4)
Multiply (3) by (i) and adding to (4) we get
2i Tx, y 0 x, y H
Tx, y 0 x, y H
Tx, Tx 0 x, y H (Taking y = Tx)
Tx 0 x, y H

T = 0 (zero operator) Notes
Theorem 6: An operator T on a Hilbert space H is self-adjoint.
Tx,x is real for all x.

Proof: Let T* =T (i.e. T is self adjoint operator)
Then for every x H, we have
Tx, Tx x, T * x x, Tx Tx, x
Tx, x equals its own conjugate and is therefore real.
Conversely, let Tx, x is real x H. We claim that T is self adjoint i.e. T*=T.
since Tx, x is real x H,
Tx, x Tx, x x, T * x T * x, x
Tx,x T * x,x 0 x H
Tx T * x,x 0 x H
T T * x, x 0 x H
T – T* = 0 [ if (Tx, x) = 0 T = 0]
T = T*
T is self adjoint.
Cor. If H is real Hilbert space, then A is self adjoint
Ax, y Ay,x x, y H.
A is self adjoint for any x, y H.
Ax, y x,A * y A * y,x .
since H is real Hilbert space A * y, x A * y, x so that Ax, y Ay, x  A* A
Theorem 7: The real Banach space of all self-adjoint operators on a Hilbert space H is a partially
ordered set whose linear and order structures are related by the following properties:
(a) If A 1 A 2 then A1 +A A2 +A for every A S;
(b) If A 1 A 2 and 0, then A1 A2 .
Proof: Let S represent the set of all self-adjoint operators on H. We define a relation on S as
follows:
If A 1A 2 S, we write A 1 A 2 if A 1 x, x A2 , x x in H.
We shall show that ' ' is a partial order relation on S. ' ' is reflexive.

Notes Let A S. Then
Ax, x Ax, x x H
Ax, x Ax, x x H
By definition A A.
' ' on S is reflexive.
' ' is transitive.
Let A 1 A 2 and A 2 A 3 then
A 1 x, x A 2 x, x x H.
and A 2 x, x A 3 x, x x H.
From these we get
A 1 x, x A 3 x, x x H.
and A 2 x, x A 3 x, x x H.
From these we get
A 1 x, x A 3 x, x x H.
Therefore by definition A 1 A 3 and so the relation is transitive.
' ' is anti-symmetric.
Let A1 A2 and A2 A1 then to show that A1 A2 .
We have A 1 A2 A 1 x,x A 2 x,x x H.
Also A 2 A1 A 2 x, x A 1 x, x x H.
From these we get
A 1 x,x A 2 x,x x H.
A 1 x A 2 x, x 0 x H.
A 1 A 2 x, x 0 x H.
A1 A 2 0
A1 A2
' ' on anti-symmetric.
Hence ' ' is a partial order relation on S.
Now we shall prove the next part of the theorem.
(a) We have A 1 A2 A 1 x,x A 2 x,x x H.

A 1x,x Ax,x A 2 x,x Ax,x x H. Notes
A1 A 2 x,x A1 A x,x x H.
A1 A 2 A 2 A, by def. of .
(b) We have A 1 A2 A 1x,x A 2 x,x x H
A 1x,x A 2 x,x x H  0
A 1 x, x A 2 x, x x H
A1 x,x A2 x,x x in H
A1 A 2 ,by def. of ' '.
28.1.2 Definition – Positive Operator
A self adjoint operator on H is said to be positive if A 0 in the order relation. That is
if Ax, x 0 x H.
We note the following properties from the above definition.
(i) Identity operator I and the zero operator O are positive operators.
Since I and O are self adjoint and

2
Ix, x x, x x 0
also (Ox, x) = (0, x) = 0
I, O are positive operators.
(ii) For any arbitrary T on H, both TT* and T*T are positive operators. For, we have
TT * * T * * T* TT *
TT * is self adjoint
Also T * T * T* T* * T*T
T * T is self adjoint
Further we see that
2
TT * x, x T * x, T * x T*x 0
2
and T * Tx, x Tx, T * *x (Tx, Tx) Tx 0
Therefore by definition both TT* and T*T are positive operators.
Theorem 8: If T is a positive operator on a Hilbert space H, then I+T is non-singular.
Proof: To show I+T is non-singular, we are to show that I+T is one-one and onto as a mapping of
H onto itself.
I+T is one-one.
First we show I T x 0 x 0

Notes We have I T x 0 Ix Tx 0 x Tx 0
Tx x
2
Tx, x x, x x
x
2
0  Tx, x 0
2
x 0
2 2
x 0  x is always 0
x 0
I T x 0 x 0.
Now I T x I T y I T x y 0
x y 0 x y
Hence I+T is one-one.
I+T is onto.
Let M = range of I+T. Then I+T will be onto if we prove that M=H.
We first show that M is closed.
For any x H, we have

2 2
I T x x Tx
x Tx,x Tx
x,x x,Tx Tx,x Tx,Tx

2 2
x Tx Tx,x Tx,x
x
2
Tx
2
2 Tx, x  T is positive T is self-adjoint Tx,x real
x
2
 T is positive Tx,x 0
Thus x I T x x H
Now let I T x n be a CAUCHY sequence in M. For any two positive integers m,n we have
xm xn I T xm xn
I T xm I T xn 0,
since I T x is a CAUCHY sequence.
xm xn 0
x n is a CAUCHY sequence in H. But H is complete. Therefore by CAUCHY sequence x n in

H converges to a vector, say x in H.

Notes
Now Lim I T x n I T lim x n I T is a continuous mapping
I T x M range of I T
Thus the CAUCHY sequence I T x n in M converges to a vector I T x in M.
every CAUCHY sequence in M is a convergent sequence in M.
M is complete subspace of a complete space is closed.
M is closed.
Now we show that M = H. Let if possible M H.
Then M is a proper closed subspace of H.
Therefore, a non-zero vector x 0 in H s.t. x 0 is orthogonal in M.
Since I T x0 M, therefore
x0 M I T x0 , x 0
x0 Tx0 ,x0 0
x0 , x0 Tx 0 , x 0 0
2
x Tx 0 , x 0 0
2
x Tx 0 , x 0
x
2
0  T positive Tx0 ,x0 0
2
x 0
2
x 0 x 0
x 0
a contradiction to the fact that x0 0.
Hence we must have M = H and consequently I+T is onto. Thus I+T is non-singular.
Cor. If T is an arbitrary operator on H, then the operator I+TT* and I+T*T are non-singular.
Proof: We know that for an arbitrary T on H, T*T and TT* are both positive operators.
Hence by Theorem (8) both the operators I+TT* and I+T*T are non-singular.
28.2 Summary
 An operator T on a Hilbert space H is said to be self adjoint if T*=T.
 A self adjoint operator on H is said to be positive if A 0 in the order relation. That is if

Ax, x 0 x H.

Notes 28.3 Keywords
Positive Operator: A self adjoint operator on H is said to be positive if A 0 in the order

relation. That is
if Ax, x 0 x H.
Self Adjoint: An operator T on a Hilbert space H is said to be self adjoint if T*=T.
1. Define a new operation of “Multiplication” for self-adjoint operators by
A 1A 2 A 2 A 1
A1  A2 , and note that A1  A2 is always self-adjoint and that it equals A1A 2
2
whenever A1 and A2 commute. Show that this operation has the following properties:
A1  A 2 A 2  A1 ,
A1  A2 A3 A1  A2 A1  A3 ,
A1  A2 A1  A2 A1  A2 ,
and A  I I  A A. Show that
A1  A2  A3 A 1  A 2  A 3 whenever A1 and A3 commute.
2
2. If T is any operator on H, it is clear that Tx,x Tx x T x ; so if H 0 , we have
2
sup Tx, x / x : x 0 T . Prove that if T is self-adjoint, then equality holds here.
Books Akhiezer, N.I.; Glazman, I.M. (1981), Theory of Linear Operators in Hilbert Space
Yosida, K., Functional Analysis, Academic Press
Online links www.ams.org/bookstore/pspdf/smfams-14-prev.pdf-UnitedStatesmath

world.wolfram.com

Unit 29: Normal and Unitary Operators
Notes
CONTENTS
Objectives
Introduction
29.1 Normal and Unitary Operators
29.1.1 Normal Operator
29.1.2 Unitary Operator
29.1.3 Isometric Operator
29.2 Summary
29.3 Keywords
Objectives
 Understand the concept of Normal and Unitary operators.
 Define the terms Normal, Unitary and Isometric operator.
 Solve problems on normal and unitary operators.
Introduction
An operator T on H is said to be normal if it commutes with its adjoint, that is, if TT*=T*T. We
shall see that they are the most general operators on H for which a simple and revealing
structure theory is possible. Our purpose in this unit is to present a few of their more elementary
properties which are necessary for our later work. In this unit, we shall also study about Unitary
operator and Isometric operator.
29.1 Normal and Unitary Operators
29.1.1 Normal Operator
Definition: An operator T on a Hilbert space H is said to be normal if it commutes with its

adjoint i.e. if TT* = T*T
Conclusively every self-adjoint operator is normal. For if T is a self adjoint operator i.e. T*=T
then TT* =T*T and so T is normal.
Note A normal operator need not be self adjoint as explained below by an example.

Notes
Example: Let H be any Hilbert space and I : H H be the identity operator.
Define T = 2iI. Then T is normal operator, but not self-adjoint.
Solution: Since I is an adjoint operator and the adjoint operation is conjugate linear,
T* = –2iI* = –2iI so that
TT* =T*T =4I.
T is a normal operator on H.
But T = T* T is not self-adjoint.
Note If T H is normal, then T* is normal.
since if T* is the adjoint of T; then T**= T.
T is normal TT*=T*T
Hence T*T**=T*T=TT*=T**T* so that T*T**=T**T
T* is normal if T H.
Theorem 1: The limit T of any convergent sequence (T k) of normal operators is normal.
Proof: Now T *k T * Tk T * Tk T
T *k T * as k since Tk T as k .
Now we prove TT* T * T so that T is normal.
TT * T * T TT * Tk T *k Tk T *k T k* Tk T *k T TT * TT * Tk Tk* Tk Tk* Tk* Tk
T *k Tk TT *
TT * T * T TT * Tk T k* T *k Tk TT * ...(1)
 Tk is normal i.e. Tk T*k T *k Tk
since Tk T as T k* T*, R.H.S. of (1) 0
TT * T * T 0
TT* T*T
T is normal.
Theorem 2: The set of all normal operators on a Hilbert space H is a closed subspace of H
which contains the set of all set-adjoint operators and is closed under scalar multiplication.
Proof: Let M be the set of all normal operators on a Hilbert space H. First we shall show that M
is closed subset of H .

Let T be any limited point of M. Then to show that T M i.e. to show that T is a normal operator Notes
on H.
Since T is a limited point of M, therefore a sequence Tn of distinct point of M such that

Tn T. We have
T n* T* Tn T* Tn T 0.
T n* T* 0 T n* T *.
Now, TT * T * T TT * Tn T * Tn T *n T*T TT * Tn T n* Tn T *n T*T
TT * Tn T * Tn T n* T n* Tn T n* Tn T*T
TT * Tn T *n Tn T n* T *n Tn T *n Tn T*T
TT * Tn T *n 0 T *n Tn T*T
 Tn M Tn is a normal operator on H i.e. Tn T*

n T*n Tn and 0 0
TT * Tn T *n T *n Tn T*T 0 since Tn T and T n* T*
Thus, TT * T * T 0 TT * T * T 0
TT* T * T T is normal operator on H.
T M and so M is closed.
Now every self adjoint operator is normal. Therefore the set M contains the set of all self-adjoint
operators on H.
Finally, we show that M is closed with respect to scalar multiplication i.e. T M
T M, is any scalar.
In other words, we are to show that if T is a normal operator on H and is any scalar, then T
is normal operator on H. Since T is normal, therefore TT* =T*T.
We have T * T *.
Now T T * T T* TT * .
Also T * T T T T*T ( ) (TT*)
T T * T * T
T is normal.
Theorem 3: If N1, N2 are normal operators on a Hilbert space H with the property that either
commutes with the adjoint of the other, then N1 + N2 and N1N2 are also normal operators.
Proof: Since N1, N2 are normal operators, therefore
N 1 N*1 = N*1 N1 and N2 N*2 N*2 N2 ...(1)

Notes Also by hypothesis, we have N 1 N* 2 =N*2 N1 and N2 N*1 … (2)

N*1 N2
we claim that N1 + N2 is normal.
i.e. (N1 + N2)(N1 + N2)* = (N1 + N2)*(N1 + N2) ...(3)
since adjoint operation preserves addition, we have
(N1 + N2)(N1 + N2)* = (N1 + N2) N*1 +N* 2
N1N*1 N1N 2 N 2 N*1 N 2 N*2 ...(4)
N1N*1 N*2 N1 N*1N2 N*2 N2
= N*1 N* 2 N 1 N 2
= N1 N2 * N1 N2 (using (1) and (2))
N1 N2 N1 N2 * N1 N2 * N1 N2
N1 N2 is normal.
Now we show that N1N2 is normal i.e.
N 1N 2 N 1N 2 * N 1N 2 * N 1N 2 .
L.H.S.= N 1N 2 N 1N 2 * N 1N 2 N * 2 N * 1
N 1 N 2 N * 2 N *1
N 1 N * 2 N 2 N *1
N 1N * 2 N 2 N * 1
N * 2 N 1 N *1 N 2
N * 2 N 1N * 1 N 2
N * 2 N * 1 N 1N 2
N 1N 2 * N 1N 2
N 1N 2 N 1N 2 * N 1N 2 * N 1N 2
N1N2 is normal.
Theorem 4: An operator T on a Hilbert space H is normal T*x Tx for every x H.
Proof: We have T is normal TT* T * T
TT * T * T 0

Notes
TT * T * T x, x 0 x
TT * x; x T * Tx, x x
T * x, T * x Tx, T * *x x
2 2
T*x Tx x T * * T
T*x Tx x.
2
Theorem 5: If N is normal operator on a Hilbert space H, then N N2 .
Proof: We know that if T is a normal operator on H then
Tx T*x x ...(1)
Replacing T by N, and x by Nx we get
NNx N * Nx x
N2 x N * Nx x ...(2)
Now N2 Sup N 2 x : x 1
Sup N * Nx : x 1 (by (2))
N*N
2
N
Theorem 6: Any arbitrary operator T on a Hilbert space H can be uniquely expressed as

T T1 iT2 where T1 ,T2 are self-adjoint operators on H.
T T* 1
Proof: Let T1 and T2 T T*
2 2i
Then T1 iT2 T ...(1)
1 *
Now T *1 T T*
2
1
T T* *
2
1
T* T**
2
1 1
T* T T T* T1
2 2

Notes T *1 T1
T1 is self-adjoint.
1 *
Also T * 2 T T*
2i
1
T T* *
2i
1
T* T**
2i
1 1
T* T T T* T2
2 2i
T *2 T2
T2 is self-adjoint.
Thus T can be expressed in the form (1) where T 1,T2 are self adjoint operators.
To show that (1) is unique.
Let T = U1 + iU2, U1,U2 are both self-adjoint
We have T* U 1 iU 2 *
U *1 iU 2 *
U *1 iU * 2
U * 1 iU * 2 U 1 iU 2
T T* U 1 iU 2 U 1 iU 2 2U,
1
U1 T T* T1
2
and T T* U 1 iU 2 U 1 iU 2 2iU 2
1
U2 T T* T2
2i
expression (1) for T is unique.
Note The above result is analogous to the result on complex numbers that every complex
number z can be uniquely expressed in the form z = x + iy where x, y are real. In the above
theorem T =T1 + T2, T1 is called real part of T and T 2 is called the imaginary part of T.

Theorem 7: If T is an operator on a Hilbert space H, then T is normal its real and imaginary Notes
parts commute.
Proof: Let T1 and T2 be the real and imaginary parts of T. Then T 1, T2 are self-adjoint operators
and T = T1 + i T2.
We have T* = (T1 + iT2)* = T1* + (iT2)*
= Ti* + i T2*
= Ti* – iT2*
= T1 – iT2
Now TT* = (T1 + iT2) (Ti – iT2)
= T12 + T22 + i (T2T1 – T1T2) … (1)
and T*T = (Ti – iT2) (T1 – iT2)
= T12 + T22 + i (T1T2 – T2T1) … (2)
Since T is normal i.e. TT* = T*T.
Then from (1) and (2), we see that
T12 + T22 + i (T2T1 – T1T2) = T12 + T22 + i (T1T2 – T2T1)
T2T1 – T1T2 = T1T2 – T2T1
2T2T1 = 2T1T2
T2T1 = T1T2 T1, T2 commute.
Conversely, let T1, T2 commute
i.e. T1T2 = T2T1, then from (1) and (2)
We see that
TT* = T*T T is normal.
Example: If T is a normal operator on a Hilbert space H and is any scalar, then T – I is

also normal.
Solution: T is normal TT* = T*T
Also (T – I)* = T* – ( I)*
= T* – I*
= T* – I.
Now (T – I) (T – I)* = (T – I) (T* – I)
= TT* – I – T* + | |2I … (1)
Also (T – I)* (T – I) = (T* – I) (T – I)
= T*T – I* – T + | |2I … (2)

Notes Since TT* = T*T, therefore R.H.S. of (1) and (2) are equal.
Hence their L.H.S. are also equal.
(T – I) (T – I)* = (T – I)* (T – I)
T – I is normal.
29.1.2 Unitary Operator
An operator U on a Hilbert space H is said to be unitary if UU* =U*U =I.
Notes
(i) Every unitary operator is normal.
(ii) U* = U-1 i.e. an operator is unitary iff it is invertible and its inverse is precisely equal
to its adjoint.
Theorem 8: If T is an operator on a Hilbert space H, then the following conditions are all
equivalent to one another.
(i) T*T = I.
(ii) (Tx,Ty) = (x,y) for all x, y H.
(iii) Tx x x H.
Proof: (i) (ii)
(Tx,Ty) = (x,T*Ty) = (x,Iy) = (x,y) x and y.
(ii) (iii)
We are given that
Tx, Ty x, y x, y H.
Taking y = x, we get
2 2
(Tx,Tx) = (x,x) Tx x
Tx x x H.
(iii) (i)
Given Tx x x
2 2
Tx x
Tx, Tx x, x
T * Tx, x x, x

Notes
T * T I x, x O x H
T*T I O
T*T I
29.1.3 Isometric Operator

Definition: An operator T on H is said to be isometric if Tx Ty x y x, y H.
Since T is linear, the condition is equivalent to Tx x for every x H.
For example: let e 1 ,e 2 ,...,e n ,... be an orthonormal basis for a separable Hilbert space H and
T H be defined as T x1e 1 x 2 e 2 ... x 1e 2 x 2 e 3 ... where x xn .
2 2 2
Then Tx xn x
n 1
T is an isometric operator.
The operator T defined is called the right shift operator given by Te n = en+1.
Theorem 9: If T is any arbitrary operator on a Hilbert space H then H is unitary it is an

isometric isomorphism of H onto itself.
Proof: Let T is a unitary operator on H. Then T is invertible and therefore T is onto.
Further TT* = I.
Hence Tx x for every x H. [By Theorem (7)]
T preserves norms and so T is an isometric isomorphism of H onto itself.
Conversely, let T is an isometric isomorphism of H onto itself. Then T is one-one and onto.
Therefore T–1 exists. Also T is an isometric isomorphism.
Tx x x
T*T = I [By Theorem (7)]
1 1
T*T T IT
1 1
T * TT T
1
T*I T
TT* I T * T and so T is unitary.

Notes
Note If T is an operator on a Hilbert space H such that Tx x x H and T is definitely

an isometric isomorphism of H onto itself. But T need not be onto and so T need not be
unitary. The following example will make the point more clear.
Example: Let T be an operator on l2 defined by T x1 ,x2 ,... 0,x1 ,x2 ,...
Tx x x I2 .
T is an isometric isomorphism of l 2 into itself.
However T is not onto. If y1 , y2 ,... is a sequence in l2 such that y1 0, then no sequence in l2

whose T-image is y1 , y2 ,... . Therefore T is not onto and so T is not unitary.
29.2 Summary
 An operator T on a Hilbert space H is said to be normal if it commutes with its adjoint i.e.
if TT* = T*T. Conclusively every self adjoint operator is normal.
 The set of all normal operators on a Hilbert space H is a closed subspace of H which
contains the set of all set-adjoint operators and is closed under scalar multiplication.
 An operator U on a Hilbert space H is said to be unitary if UU* = U*U =I.
 An operator T on H is said to be isometric if Tx Ty x y x, y H , since T is linear,
the condition is equivalent to Tx x for every x H.
29.3 Keywords
Normal Operator: An operator T on a Hilbert space H is said to be normal if it commutes with

its adjoint i.e. if TT* = T*T.
Unitary Operator: An operator U on a Hilbert space H is said to be unitary if UU* = U*U = I.
Isometric Operator: An operator T on H is said to be isometric if Tx Ty x y x, y H.

Since T is linear, the condition is equivalent to Tx x for every x H.
1. If T is an operator on a Hilbert space H, then T is normal its real and imaginary part
commute.
2. An operator T on H is normal T*x Tx for every x.
3. The set of all normal operators on H is a closed subset of H which contains the set of all
self-adjoint operators and is closed under scalar multiplication.
4. If H is finite-dimensional, show that every isometric isomorphism of H into itself is

unitary.
5. Show that the unitary operators on H form a group.

29.5 Further Readings Notes
Books Arch W. Naylor, R Sell George, Linear Operator Theory in Engineering and Sciences,
New York Springer, (1982).
Paul Garret, Operators in Hilbert Spaces, 2005.
Online link www.maths.leeds.ac.uk/nkisilv/

Notes
Unit 30: Projections
CONTENTS
Objectives
Introduction
30.1 Projections
30.1.1 Perpendicular Projections
30.1.2 Invariance
30.1.3 Orthogonal Projections
30.2 Summary
30.3 Keywords
Objectives
 Define perpendicular projections.
 Define invariance and orthogonal projections.
 Solve problems on projections.
Introduction
We have already defined projections both in Banach spaces and Hilbert spaces and explained
how Hilbert spaces have plenty of projection as a consequence of orthogonal decomposition
theorem or projection theorem. Now, the context of our present work is the Hilbert space H, and
not a general Banach space, and the structure which H enjoys in addition to being a Banach space
enables us to single out for special attention those projections whose range and null space are
orthogonal. Our first theorem gives a convenient characterisation of these projections.
30.1 Projections
30.1.1 Perpendicular Projections
A projection P on a Hilbert space H is said to be a perpendicular projection on H if the range M

and null space N of P are orthogonal.
Theorem 1: If P is a projection on a Hilbert space H with range M and null space N then M N P
is self-adjoint and in this case N M .
Proof: Let M N and z be any vector in H. Then since H M N, we can write z uniquely as
z x y,x M, y N.

Thus Pz P x y Notes
Px Py
Px x y N
Pz,z x,z  Pz = P x+y x,P being projection on H
x, x y
x, x x, y
2
x
and Pz*,z z,Pz
x y,x x,x x,y

2
x .
Hence Pz,z Pz*,z z H
P P * z,z 0 z H
P P* 0 i.e. P P*
P is self adjoint.
Further, M N N M
If N M , then N is a proper closed linear subspace of the Hilbert space M and therefore a
vector z0 0 M s.t. z0 N.
Now z0 M and z0 N and H = M N.
z0 H z0 0, a contradiction.
Hence N M
Conversely, let P* P,x, y be any vectors in M and N respectively. Then
x, y Px, y
x,P * y x,Py
x,0 0
M N.
Theorem 2: If P is the projection on the closed linear subspace M of H, then
x M Px x Px x.
Proof: We have, P is a projection on H with range M then, to show x M Px x.

Notes Let Px = x. Then X is in the range of P because Px is in the range of P.
Px x x M.
Conversely, let x M. Then to show Px = x.
Let Px = y. Then we must show that y = x.
We have
Px y P Px Py P 2x Py
Px Py  P 2 =P
P x y 0
x y is a in null space of P.
x y M .
x y z,z M .
x y z.
Now y Px y is in the range of P.
i.e. y is in M. Thus we have expressed
x y z, y M,z M .
But x is in M. So we can write x = x+0, x M,0 M
But H M M .
Therefore we must have y = x, z = 0
Hence x M Px x.
Now we shall show that Px = x Px x.
If Px = x then obviously Px x .
Conversely, suppose that Px x .
We claim that Px = x. We have

2 2
x Px I P x ...(1)
Now Px is in M. Also P is the projection on M.
I P is the projection on M .
I P x in M .
Px and I P x are orthogonal vectors.
Then by Pythagorean theorem, we get
2 2 2
Px I P x Px I P x ...(2)

From (1) and (2), we have Notes
2 2 2
x Px I P x
2
I P x 0  by hypothesis Px x
I P x 0
x Px 0
Px 0
Theorem 3: If P is a projection on a Hilbert space H, then
(i) P is a positive operator i.e. P 0

(ii) 0 P 1
(iii) Px x x H.
(iv) P 1.
Proof: P, projection on H P2 P,P * P.
Let M = range of P.
(i) Let x H. Then
Px,x PPx,x
2
Px,P * x Px,Px Px 0
Px, x 0 x H.
P is a positive operator i.e. P 0.
(ii) P is a projection on H I – P is also a projection on H.
I –P 0. (by (i))
P I
But P 0 , consequently 0 P 1.
(iii) Let x H. If M is the range of P, then M is the range of (I – P).
Now Px is in M and (I – P)x is in M .
Therefore Px and (I – P)x are orthogonal vector. So by Pythagorean theorem, we have
2 2 2
Px I P x Px I P x
2 2 2
x Px I P x  Px+ I–P x x

Notes 2 2
x Px
Px x.
(iv) We have P sup Px : x 1
But Px x x H (by(iii))
sup Px : x 1 1
Hence P 1
Example: If P and Q are the projections on closed linear subspaces M and N of H. Show
that PQ is a projection PQ = QP. In this case, show that PQ is the projection on M N.
Solution: Since P and Q are projections on H, therefore P 2 = P, P* = P, Q2 = Q, Q* = Q. Also it is given

that M is range of P and N is the range of Q.
Now suppose PQ is projection on H. Then to prove that PQ = QP.
Since PQ is a projection on H.
(PQ)* = PQ
Q* P* = PQ
QP = PQ ( Q* = Q, P* = P)
Conversely, let PQ = QP. We shall show that PQ is a projection on H.
We have (PQ)* = Q*P* = QP = PQ.
Also (PQ)2 = (PQ) (PQ) = (PQ) (QP)
= PQ2P = PQP
= QPP = QP2
= QP = PQ
Thus (PQ)* = PQ and (PQ)2 = PQ.
PQ is a projection on H.
Finally we are to show that PQ is the projection on M N, i.e. we are to show that range of PQ
is M N.
Let R (PQ) = range of PQ.
Let x M N x M, x N we have
(PQ) (x) = P (Qx) = Px [ N is range of Q and x N Qx = x]
=x [ M is range of P and x P]
(PQ)x = x
x R (PQ)
x M N x R (PQ)

M N R (PQ) Notes
Now let x R (PQ). Then (PQ)x = x
Now (PQ) x = x
P [(PQ) x] = Px
[P (PQ)] x = Px
(P2Q) x = Px
(PQ) x = Px
But (PQ) x = x.
We have Px = x x M i.e. the range of P.
Also PQ = QP
x R (PQ) (PQ) x = x
(QP)x = x Q [(QP)x] = Qx
(Q2P)x = Qx (QP)x = Qx
But (QP)x = x, Qx = x x N.
Thus x R (PQ) x M and x N
x M N
R(PQ) M N
Hence R(PQ) = M N.
Example: Show that an idempotent operator on a Hilbert space H is a projection on H

it is normal.
Solution: P is an idempotent operator on H i.e. P2 = P.
Let P be a projection on H. Then P* = P. We have
PP* = P* P* [taking P* in place of P in L.H.S.]
= P* P [ P* = P]
P is normal.
Conversely, let PP* = P*P.
Then to prove that P* = P.
For every vector y H, we have
(Py, Py) = (y, P* Py) = (y, PP*y) [ P*P = PP*]
= (P*y, P*y) [ (P*)* = P]
From this we conclude that
Py = 0 P*y = 0.
Now let x be any vector in H.
Let y = x – Px. Then

Notes Py = P(x – Px) = Px – P2x = Px – Px = 0
0 = P*y = P*(x – Px) = P*x – P*Px
P*x = P*Px x H
P* = P*P
Now P = (P*)* = (P*P)* = P*P = P*
P is a self adjoint operator.
Also P2 = P.
Hence P is a projection on H.
30.1.2 Invariance
Definition: Let T be an operator on a Hilbert space H and M be a closed subspace of H. Then M

is said to be invariant under T if T M M. If we do not take into account the action of T on
vectors outside M, then T can be regarded as an operator on M itself. The operator T on H induces
on operator T M on M such that TM(x) = T(x) for every x M. This operator TM is called the
restriction of T on M.
Further, let T be an operator on Hilbert space H. If M is a closed subspace of H and if M and M

are both invariant under T, then T is said to be reduced by M. If T is reduced by M, we also say
that M reduces T.
Theorem 4: A closed linear subspace M of a Hilbert space H is invariant under the operation
T M is invariant under T*.
Proof: Let M is invariant under T, we show M is invariant under T*.
Let y be any arbitrary vector in M . Then to show that T*y is also in M i.e. T*y is orthogonal to
every vector in M.
Let x be any vector in M. Then Tx M because M is invariant under T.
Also y M y is orthogonal to every vector in M.
Therefore y is orthogonal to Tx i.e.
(Tx,y) = 0
(x,Ty) = 0
T*y is orthogonal to every vector x in M.
T * y is in M and so M is invariant under T*.
Conversely, let M is invariant under T*. Thus to show that M is invariant under T. Since M is
a closed linear subspace of H invariant under T*, therefore by first case M is invariant
under T.
But M M M and T * * T** T.

Hence M is invariant under T. Notes
Theorem 5: A closed linear subspace M of a Hilbert space H reduces on operator M is invariant

under both T and T*.
Proof: Let M reduces T, then by definition both M and M are invariant under T*. But by
theorem 4, if M is invariant under T then M i.e. M is invariant under T*. Thus M is invariant
under T and T*.
Conversely, let M is invariant under both T and T*. Since M is invariant under T*, therefore M
is invariant under T * * = T (by theorem 4). Thus both M and M are invariant under T.
Therefore M reduces T.
Theorem 6: If P is the projection on a closed linear subspace M of a Hilbert space H, then M is

invariant under an operator T TP PTP.
Proof: Let M is invariant under T.
Let x H. Then Px is in the range of T, Px M TPx M.
Now P is projection and M is the range of P. Therefore TPx M TPx will remain unchanged
under P. So, we have
PTPx = TPx
PTP = TP (By equality of mappings)
Conversely, let PTP = TP. Let x M. Since P is a projection with range M and x M , therefore
Px = x
TPx = Tx
PTPx =Tx  PTP TP
PTPx = TPx  TPx Tx
But P is a projection with range M.
P TPx TPx TPx M Tx M

Since TPx = Tx.
Thus x M Tx M
M is invariant under T.
Theorem 7: If P is the projection on a closed linear subspace of M of a Hilbert space H, then M

reduces an operator TP PT.
Proof: M reduces T M is invariant under T and T*.
TP PTP and T * P PT * P
TP PTP and T * P * PT * P *
TP PTP and P * T * * P * T * *P *
TP PTP and PT PTP  P is projection P* P. AlsoTT* T

Notes Thus M reduces T.
TP PTP and PT PTP ...(1)
Now suppose M reduces T. Then from (1), TP = PTP and PT = PTP. This gives TP = PT.
Conversely, let TP = PT
PTP =P2T (Multiplying both sides on left by P.)
or PTP = PT P2 P
similarly multiplying both sides of TP = PT on the right of P, we get
TP2 = PTP or TP = PTP. Thus
TP = PT TP = PTP and PT = PTP.
Therefore from (1), we conclude that M reduces T.
Theorem 8: If M and N are closed linear subspace of a Hilbert space H and P and Q are the
projections on M and N respectively, then
(i) M N PQ O. and
(ii) PQ O QP O.
Proof: Since P and Q are projections on a Hilbert space H, therefore P* = P, Q* = Q.
We first observe that
PQ O PQ * O * Q * P* O *
QP O.
Therefore to prove the theorem it suffices to prove that
M N PQ O.
First suppose M N. If y is any vector in N, then M N y is orthogonal to every vector in M.
so y M .Consequently N M .
Now, let z be any vector in H. Then Qz is the range of Q i.e. Qz is in N.
Consequently Qz is in M which is null space of P.
Therefore P(Qz) = O.
Thus PQz = O z H
PQ = O
Conversely, let PQ = O and x M and y N.
since M is the range of P, therefore Px = x. Also N is the range of Q. Therefore
Qy = y
We have (x,y) = (Px, Qy) = (x,P*Qy)

Notes
= (x,PQy)  P* P
= (x,Oy) PQ O
= (x,O) = O
M and N are orthogonal i.e. M N.
30.1.3 Orthogonal Projections
Definition: Two projections P and Q on a Hilbert space H are said to be orthogonal if PQ = O.
Note: By theorem 8, P and Q are orthogonal iff their ranges M and N are orthogonal.
Theorem 9: If P1, P2, ... Pn are projections on closed linear subspaces M 1, M2, ... Mn of a Hilbert
space H, then P = P1 + P2 + ... + Pn is a projection the Pi's are pair-wise orthogonal (in the sense
that Pi Pj 0,i j).
Also then P is the projection on M = M1 + M2 + ... + Mn.
Proof: Given that P1, P2, ... Pn are projections on H.
Therefore Pi2 Pi Pi* for each i = 1, 2, ...,n.
Let P = P1 + P2+ ... + Pn. Then P* = (P1 + P2+ ... + Pn)* = P1* ... Pn*
= P1 + P2 + ... + Pn = P.
Sufficient Condition:
Let Pi Pj O,i j. Then to prove that
P is a projection on H. We have
P2 = PP = (P1 + P2+ ... + Pn) (P1 + P2+ ... + Pn)
= P12 P22 ... Pn2 Pi Pj 0,i j
= P1 + P2+ ... + Pn
=P
Thus, P* = P = P*.
Therefore P is a projection on H.
Necessary Condition:
Let P is a projection on H.
Then P2 = P = P*.
We are to prove that Pi Pj 0 if i j.
We first observe that if T is any projection on H and z is any vector in H, then
(Tz, z) = (T Tz, z) = (Tz, T*z)
= (Tz, Tz)
2
= Tz ...(1)

Notes Now let x belongs to the range of some Pi so that Pix = x. Then
2 2
x Pi x
n
2 2 2
Pj x P1 x ... Pn x
j 1
Pj x, x [Using (i)]
j 1
P1 x, x ... Pn x, x
P1 P2 ... Pn x
Px, x
Px
2
[by (1)]
x
2
...(2)
Thus we conclude that sign of equality must hold throughout the above computation. Therefore
we have
n
2 2
Pi x Pj x
j 1
2
Pj x O if j i
Pj x O, j i
Pj x O, j i
x is in the null space of Pj ,i j
x M j ,if j i
x is orthogonal to the range Mj of every Pj with j i.
Thus every vector x in the range P i(i = 1,...,n) is orthogonal to the range of every P j with j i.
Therefore the range of P i is orthogonal to the range of every P j with j i. Hence
Pi Pj O, i j [By theorem (8)]
Finally in order to show that P is the projection on M M 1 M 2 ... M n
We are to show that R(P) = M where R(P) is the range of P.
Let x M. Then x x1 x2 ... xn
2 2
where xi Mi , 1 i n. Now from (2), we observe that x Px if x is the range of some P i.
xi M, i.e. the range of Pi .

2 2 Notes
Pi x xi Pi x xi
Px i xi
xi the range of P.
xi R P , for each i 1, 2,..., n
x1 x2 ... x n R P .
x R P .
Then x M x R P
M R P ...(3)
Now suppose that x R P . Then
Px = x
P1 P2 ... Pn x x
P1x P2 x ... Pn x x
But P1 x M1 ,P2 x M2 ,...,Pn x Mn .
x M1 M2 ... M n and so R P M ...(4)
Hence from (3) and (4), we get
M = R(P)
30.2 Summary
 A projection P on a Hilbert space H is said to be a perpendicular projection on H if the

range M and null space N of P are orthogonal.
 Let T be an operator on a Hilbert space H and M be a closed subspace of H. Then M is said

to be invariant under T if T M M.
 Let T be an operator on Hilbert space H, if M is closed subspace of H and if M and M are

both invariant under T, then T is said to be reduced by M.
 Two projections P and Q on a Hilbert space H are said to be orthogonal if PQ = O.
30.3 Keywords
Invariance: Let T be an operator on a Hilbert space H and M be a closed subspace of H. Then M

is said to be invariant under T if T M M.
Orthogonal Projections: Two projections P and Q on a Hilbert space H are said to be orthogonal
if PQ = O.
Perpendicular Projections: A projection P on a Hilbert space H is said to be a perpendicular

projection on H if the range M and null space N of P are orthogonal.

1. If P and Q are the projections on closed linear subspaces M and N of H, prove that PQ is a
projection PQ QP. In this case, show that PQ is the projection on M N.
2. If P and Q are the projections on closed linear subspaces M and N of H, prove that the
following statements are all equivalent to one another:
(a) P Q;
(b) Px Qx for every x;
(c) M N;
(d) PQ P;
(e) QP = P.
3. If P and Q are the projections on closed linear subspaces M and N of H, prove that Q P is
a projection P Q. In this case, show that Q – P is the projection on N M .
Books Borbaki, Nicolas (1987), Topological Vector Spaces, Elements of mathematics, Berlin:
Springer – Verlag
Rudin, Walter (1987), Real and Complex Analysis, McGraw-Hill.
Online links www.math.Isu.edu/ ~ sengupta/7330f02/7330f02proiops.pdf
Planetmath.org/....OrthogonalProjections OntoHilbertSubspaces.html.

Unit 31: Finite Dimensional Spectral Theory
Unit 31: Finite Dimensional Spectral Theory Notes
CONTENTS
Objectives
Introduction
31.1 Finite Dimensional Spectral Theory
31.1.1 Linear Operators and Matrices on a Finite Dimensional Hilbert Space
31.1.2 Similar Matrices
31.1.3 Determinant of an Operator
31.1.4 Spectral Analysis
31.1.5 Spectrum of an Operator
31.1.6 Spectral Theorem

31.1.7 Spectral Resolution of an Operator
31.1.8 Compact Operators
31.1.9 Properties of Compact Operators
31.2 Summary
31.3 Keywords
Objectives
 Understand finite dimensional spectral theory.

 Describe spectral analysis and spectral resolution of an operator.
 Define compact operators and understand properties of compact operators.
 Solve problems on spectral theory.
Introduction
The generalisation of the matrix eigenvalue theory leads to the spectral theory of operators on
a Hilbert space. Since the linear operators on finite dimensional spaces are determined uniquely
by matrices, we shall study to some extent in detail the relationship between linear operators in
a finite dimension Hilbert spaces and matrices as a preliminary step towards the study of
spectral theory of operators on finite dimensional Hilbert spaces.

Notes 31.1 Finite Dimensional Spectral Theory
31.1.1 Linear Operators and Matrices on a Finite Dimensional Hilbert

Space
Let H be the given Hilbert space of dimension n with ordered basis B = {e 1, e2, …, en} where the
ordered of the vector is taken into consideration. Let T (H) (the set of all bounded linear
operators). Since each vector in H is uniquely expressed as linear combination of the basis, we
n
can express Tej as Tej = ij ie , where the n-scalars 1j

, ,…
2j nj
are uniquely determined by Tej.
i 1
Then vectors Te1, Te2, …, Tej determine uniquely the n2 scalars ij, i, j = 1, 2, … n. These n2 scalars
determine matrix with ( i1, i2, …, in) as the ith row and ( 1j, 2j, … nj) as its jth column. We
denote this matrix by {T} and call this matrix as the matrix of the operator T with respect to the
ordered basis B.
11 12  1n

Hence  = [ ij] = 21 22 2n

n1 n2  nn
We note that
(i) [0] = 0, which is the zero matrix.
(ii) [I] = I = [ ij], which is a unit matrix of order n. Here ij

is the Kronecker delta.
Definition: The set of all n × n matrices denoted by A n is complex algebra with respect to
addition, scalar multiplication and multiplication defined for matrices.
This algebra is called the total matrices algebra of order n.
Theorem 1: Let B be an ordered basis for a Hilbert space of dimension n. Let T (H) with (T) =
[ ij], then T is singular [ ij] is non-singular and we have [ ij]–1 = [T–1].
Proof: T is non-singular iff there exists an operator T –1 on H such that
T–1 T = T T–1 = I … (1)

–1
Since there is one-to-one correspondence between T and [T ],
(1) is true [T–1 T] = [TT–1] = [I]
from (2) [T–1] [T] = [T] [T–1] = [I] = [ ij]
so that [T–1] [ ij] = [ ij] [T–1] = [ ij], [T] = [ ij].
[ ij] is a non-singular and [ ij]–1 = [T–1].
31.1.2 Similar Matrices
Let A, B are square matrix of order n over the field of complex number. Then B is said to be
similar to A if there exists a n × n non-singular matrix C over the field of complex numbers such
that
B = C–1 AC.

This definition can be extended similarly for the case when A, B are operators on a Hilbert space. Notes
Notes
1. The matrices in An are similar iff they are the matrices of a single operator on H
relative to two different basis H.
2. Similar matrices have the same determinant.
31.1.3 Determinant of an Operator
Let T be an operator on an n-dimensional Hilbert space H. Then the determinant of the operator
T is the determinant of the matrix of T, namely [T] with respect to any ordered basis for H.
Following we given properties of a the determinant of an operator on a finite dimensional

Hilbert space H.
(i) det (I) = 1, I being identity operator.
Since det (I) = det ([I]) = det ([ ij]) = 1.
(ii) det (T1 T2) = (det T1) (det T2)
(iii) det (T) 0 [T] is non-singular
det ([T]) 0.
Hence det (T) 0 [T] is non-singular.
31.1.4 Spectral Analysis
Definition: Eigenvalues
Let T be bounded linear operator on a Hilbert space H. Then a scalar is called an eigenvalue of
T if there exists a non-zero vector x in H such that Tx = x.
Eigenvalues are sometimes referred as characteristic values or proper values or spectral values.
Definition: Eigenvectors
If is an eigenvalue of T, then any non-zero vector x H such Tx = x, is called a eigenvector

(characteristic vector or proper vector or spectral vector) of T.
Properties of Eigenvalues and Eigenvectors
Notes If the Hilbert space has no non-zero vectors then T cannot have any eigenvectors
and consequently the whole theory reduces to triviality. So we shall assume that H 0
throughout this unit.
1. If x is an eigenvector of T corresponding to the eigenvalue and is a non-zero

scalar, then x is also an eigenvector of T corresponding to the same eigenvalue .
Contd...

Notes Since x is an eigenvector of T, corresponding to the eigenvalue 0 and Tx = x.
0 x 0
Hence T ( x) = T(x) = ( x)
Therefore corresponding to an eigenvalue there are more than one eigenvectors.
2. If x is an eigenvector of T, then x cannot correspond to more than one eigenvalue

of T.
If possible let 1
, 2
be two eigenvalues of T, ( 1 2
) for eigenvector x. Then
Tx = 1
x and Tx = 2
x
1
x= 2
x
( 1
– 2
)x = 0
1
– 2
=0 ( x 0)
l
= 2
3. Let be an eigenvalue of an operator T on H. If M is the set consisting of all

eigenvectors of T corresponding to together with the vector 0, then M is a non-
zero closed linear subspace of H invariant under T.
By definition x M Tx = x … (1)
By hypothesis 0 M and 0 vector satisfies (1).
M = {x H : Tx = x} = {x H : (T – I) x = 0}
Since T and I are continuous, M is the null space of continuous transformation T – I.

Hence M is closed.
Next we show that if x M , then Tx M . If x M then Tx = x.
Since M is a linear subspace of H, x M x = Tx M.
M is invariant under T.
Definition: Eigenspace
The closed subspace M is called the eigenspace of T corresponding to the eigenvalue .
From property (3), we have proved that each eigenspace of T is a non-zero linear subspace of H
invariant under T.
Note It is not necessary for an operator T on a Hilbert space H to possess an eigenvalue.
Example: Consider the Hilbert space  2 and the operator T on  2 defined by

T (x1, x2, …) = (0, x1, x2, …)
Let be a eigenvalue of T. Then a non zero vector
y = (y1, y2, …) in  2 such that Ty = y.

Now Ty = y T (y1, y2, …) = (y1, y2, …) Notes
(0, y1, y2, …) = ( y1, y2 …)
y1 = 0, y2 = y1, ……
Now y is a non-zero vector y1 0
y1 = 0 = 0.
Then y2 = y1 y1 = 0 and this contradicts the fact that y is a non-zero vector. Therefore T cannot
have an eigenvalue.
31.1.5 Spectrum of an Operator
The set of all eigenvalues of an operator is called the spectrum of T and is denoted by (T).
Theorem 1: If T is an arbitrary operator on a finite dimensional Hilbert space H, then the spectrum
of T namely (T) is a finite subset of the complex plane and the number of points in (T) does
not exceed the dimension n of H.
Proof: First we shall show that an operator T on a finite dimensional Hilbert space h is singular
if and only if there exists a non-zero vector x H such that Tx = 0.
Let a non-zero vector x H s.t. Tx = 0. We can write Tx = 0 as Tx = T0. Since x 0, the two distinct
elements x, 0 H have the same image under T. Therefore T is not one-to-one. Hence T –1 does
not exist. Hence it is singular.
Conversely, let T is singular. Let no non-zero vector such that Tx = 0. This means Tx = 0 x=
0. Then T must be one-to-one. Since H is finite dimensional and T is one-to-one, T is onto, so that
T is a non-singular, contradicting the hypothesis that T is singular. Hence there must be non-
zero vector x s.t. Tx = 0.
Now if T is an operator on a finite dimensional Hilbert space H of dimension n. Then A scalar

(T), if there exists a non-zero vector x such that (T – I)x = 0.
Now (T – I)x = 0 (T – I) is a singular.
(T – I) is singular det (T – I) = 0. Thus (T) satisfies the equation det (T – I) = 0.

Let B be an ordered basis for H. Thus det (T – I)
= det ([T – I]B)
But det ([T – I]B) = det ([T]B – [I]B)
Thus det (T – I) = det ([T]B – [ ij]).
So det (T – I) = 0 det ([T]B – [ ij]) = 0 … (1)
If [T]B = [ ij] is a matrix of T then (1) gives
11 12  1n
22 22  2n
… (2)
  
n1   nn
The expression of determinant of (2) gives a polynomial equation of degree n in with complex
coefficients in the variable . This equation must have at least one root in the field of complex
number (by fundamental theorem of algebra). Hence every operator T on H has eigenvalue so

Notes that (T) . Further, this equation in has exactly n roots in complex field. If the equation has
repeated roots, then the number of distinct roots are less than n. So that T has an eigenvalue and
the number of distinct eigenvalue of T is less than or equal to n. Hence the number of elements
of (T) is less than or equal to n. This completes the proof of the theorem.
Example: For a two dimensional Hilbert space H, let B = {e 1, e2} be a basis and T be an
operator on H given by the matrix
11 12
A = … (1)
21 22
(i) If T is given by Te1 = e2 and Te2 = – e2, find the spectrum T.
(ii) If T is an arbitrary operator on H with the same matrix representation, then
T2 ( 11
+ 22
)T+( 11 22
– 12 21
)I=0
Sol:
(i) Using the matrix A of the operator T, we have
Te1 = 11
e1 + 21
e2 = e2 so that 11
= 0 and 21
=1
Te2 = 12
e1 + 22
e2 = –e1 so that 12
= –1 and 22
=0
11 12 0 1
Hence [T] = .
21 22 1 0
For this matrix, the eigenvalue are given by the characteristic equation
1
=0
1
2
+1=0 = i so (T) = { i}.
(ii) Let us consider the eigenvalues of A, which are given by
11 12
=0
21 22
2
–( 11
+ 22
) +( 11 22
– 12 21
)=0 … (2)
Since (2) is true for , we can take
T = I … (3)
T2 – ( 11
+ 22
)T+( 11 22
– 12 21
)I=0 … (4)
The operator T on H having as an eigenvalue satisfies equation (4).
Theorem 2: If T is an operator on a finite dimension Hilbert space, then the following statements
are true.
(a) T is singular 0 (T)

–1
(b) If T is non-singular, then (T) (T–1)
(c) If A is non-singular, then (ATA–1) = (T)
(d) If (T) and if P is polynomial then P ( ) (P (T)).

Proof: Notes
(a) T is singular a non-zero vector x H such that Tx = 0 or Tx = 0. Hence T is singular

0 is the eigenvalue of T i.e. 0 (T).
(b) Let T be non-singular and (T).

–1
Hence 0 by (a) so that exists. Since is an eigenvalue of T, a non-zero vector
x H s.t. Tx = x.
Premultiplying by T–1 we get
T–1 Tx = T–1 ( x)
1
T–1 (x) = x for x 0
–1
Hence ( (T–1))
–1
Conversely, if is an eigenvalue of T –1 then ( –1 –1
) = is an eigenvalue of (T –1)–1 = T.
Hence (T).
(c) Let S = ATA–1. Then we find S – I.
Now S – I = ATA–1 – A ( I) A–1
= A (T – I) A–1
det (S – I) = det (A(T – I) A–1)
= det (T – I)
is an eigenvalue of T det (T – I) = 0.
Hence det (T – I) = 0 det (S – I) = 0
S and T have the some eigenvalues so that
(ATA–1) = (T).
(d) If = (T), is an eigenvalue of T. Then a non-zero vector x such that Tx = x.

2
Hence T (Tx) = T ( x) = Tx = x.
Hence if is an eigenvalue of T, then 2 is an eigenvalue of T 2. Repeating this we get that

if is an eigenvalue of T, then n is an eigenvalue of T n for any positive integer n.
Let P (t) = a0 + a1t + … + amtm, a0, a1, ……, am are scalars.
Then [P (T)]x = (a0I + a2T + …… + amTm)x

m
= a0x + a1 ( x) + …… + am ( x)
m
= [a0 + a1 ( ) + …… + am ]x
m
Hence P( ) = a0 + a1 + …… + am is an eigenvalue of P (T).
This if (T), then P ( ) (P (T)).
31.1.6 Spectral Theorem
Statement: Let T be an operator on a finite dimensional Hilbert space H with 1, 2, …, m as the

eigenvalues of T and with M 1, M2, …, Mm be then corresponding eigenspaces. If P 1, P2, …, Pm are
the projections on the spaces, then the following statements are equivalent.

Notes (a) The Mi’s are pairwise orthogonal and span H:
(b) The Pi’s are pairwise orthogonal and P 1 + P2 + … + Pm = I and T = 1

P1 + 2
P2 + … + m
Pm.
(c) T is normal operator on H.
Proof: We shall show that
(i) (ii) (iii) (i)
(i) (ii)
Assume that Mi’s are pairwise orthogonal and span H. Hence every x H can be represented
uniquely as
x = x1 + x2 + … + xm … (1)
where xi Mi for i = 1, 2, …, m
by hypothesis Mi’s are pairwise orthogonal. Since P i’s are projections in M i’s Pi’s are pairwise
orthogonal, i.e. i j PiPj = 0.
If x is any vector in H, then from (1) for each i,
Pi(x) = Pi (x1 + x2 + … + xm)
= Pix1 + Pix2 + … + Pixm … (2)
Since Pi is the range of M i, Pi xi = xi.
For i j Mj Mi since xj Mj for each j we have
xj Mi for j i.
Hence xj M i (null space of Pi)
xj Mi Pixj = xi
from (2) we get
Pix = xi … (3)
Since I is the identity mapping on H, we get
Ix = x1 + x2 + … + xm … (by (1))
= P1x + P2x + … + Pmx … (by (3))
= (P1 + P2 + … + Pm) x x H.
This show that I = P1 + P2 + …… + Pm.
For every x H, we have from (1)
T (x) = T (x1 + x2 + … + xm)
= Tx1 + Tx2 + … + Txm
Since xi Mi Txi = xi
Tx = 1 1
x + x + …… +
2 2 m m
x … (4)
= 1
P1x + 2
P2x + …… + m
Pmx … (5)
T = 1
P1 + 2
P2 + …… + m
Pm
(ii) (iii)

Let T = 1P1 + 2
P2 + …… + m
Pm, where Pi’s are pairwise orthogonal projections and to show that Notes
T is normal.
Since Pi’s are projection and P *i Pi and Pi2 Pi … (6)
Further we have PiPj = 0 for i j
Since adjoint operation is conjugate linear, we get
T* = ( 1P1 + 2
P2 + …… + m
Pm)*
= 1 P *1 2 P *2 m P *m
= 1 P1 2 P2 m Pm .
Now TT* = ( 1P1 2 P2 m Pm ) ( 1P1 2 P2 m Pm )
= | 1 |2 P12 | 2 |2 P22 | 1 |2 Pm2 ( PiPj = 0, i j)
= | 1|2P1 + | 2|2P2 + …… + | m
|2Pm … (by (6))
Similarly T*T can be found s.t.
T*T = | 1|2P1 + | 2|2P2 + …… + | m|2 Pm
Hence T*T = TT* T is normal.
(iii) (i)
Let T is normal operator on H and prove that M i’s are pairwise orthogonal and M i’s span H.
We know that if
T is normal on H its eigenspaces Mi’s are pairwise orthogonal.
So it suffices to show that Mi’s span H.
Let M = M1 + M2 + …… + Mm
and P = P1 + P2 + …… + Pm
Since T is normal on H, each eigenspace Mi reduces T. Also Mi reduces T PiT = TPi for each Pi.
TP = T (P1 + P2 + …… + Pm)
= TP1 + TP2 + …… + TPm
= P1T + P2T + …… + PmT
= (P1 + P2 + …… + Pm) T
= PT
Since P is projection on M and TP = PT, M reduces T and so M is invariant under T. Let U

be the restriction of T to M . Then U is an operator on a finite dimensional Hilbert space
M and Ux = Tx x M . If x is an eigenvector for U corresponding to eigenvalue then
x M and Ux = x.
Tx = x and so x is also eigenvector for T.
Hence each eigenvector of U is also an eigenvector for T. But T has no eigenvector in M . Hence
M M = {0}. So U is an operator on a finite dimensional Hilbert space M and U has no
eigenvector and so it has no eigenvalue.

Notes M = {0}.
For if M {0}, then every operator on a non-zero finite dimensional Hilbert space must have an
eigenvalue.
Now M = {0} M = H.
Thus M = M1 + M2 + …… + Mm = H and so Mi’s span H.
This complete the proof of the theorem.
31.1.7 Spectral Resolution of an Operator
Let T be an operator on a Hilbert space H. If there exist distinct complex numbers 1

, 2
, …, m
and non-zero pairwise orthogonal projections p 1, p2, …, pm such that
T= 1
p1 + 2
p2 + … + m
pm
and p = p1 + p2 + … + pm, then the expression
T= 1
p1 + 2
p2 + … + m
pm for T is called the spectral resolution for T.
Note We note that the spectral theorem coincides with the spectral resolution for a
normal operator on a finite dimensional Hilbert space.
Theorem: The spectral resolution of the normal operator on a finite dimensional non-zero Hilbert
space is unique.
Proof: Let T = 1
p1 + 2
p2 + … + m
pm
be a spectral resolution of a normal operator on a non-zero finite dimensional Hilbert space H.

Then 1, 2, …, m are distinct complex numbers and p i s are non-zero pairwise orthogonal
projections such that p1 + p2 + … + pm = 1. We establish that 1 + 2, …, m are precisely the distinct
eigenvalues of T.
To this end we show first that for each i, i is an eigenvalue of T. Since p i 0 is a projection, a
non-zero x in the range of p 1 such that pix = x
Let us consider
Tx = ( 1p1 + 2
p2 + … + m
pm)x
= ( 1p1pi + 2
p2pi + … + ip2i + … + m
pmpi)x
So pi’s are pairwise orthogonal p ipj = 0 for i j and p2i = pi, we have Tx = ipix = ix by pix = x.
i
is an eigenvalue of T.
Next we show that each eigenvalue of T is an element of the set ( 1, 2, …, m

). Since T is an
operator on a finite dimensional Hilbert space, T must have an eigenvalue.
If is an eigenvalue of T then Tx = x = Ix.
( 1
p1 + 2
p2 + … + m
pm) x = (p1 + p2 + … + pm) x
{( 1
– ) p1 + ( 2
– ) p2 + … + ( m
– ) pm} x = 0 … (2)
Since p2i = pi and pi pj = 0 for i j operating with pi throughout (2), we get
( i – ) pix = 0 for i = 1, 2, …, m.

If i
for each i, pix = 0 for each i. Notes
Hence pix + p2x + … + pmx = 0
(p1 + p2 + … + pm) x = 0
Ix = 0
x = 0, a contradiction to the fact that x 0. Hence must be equal to i for some i. This in the
spectral resolution (1) of T, the scalar i are the precisely the eigenvalue of T.
If the spectral resolution is not unique.
Let T = 1
Q1 + 2
Q2 + … + Qn
n
… (3)
be another revolution of T. Then 1, 2, … n is the same set of eigenvalues of T written in

different order. Hence writing the eigenvalues in the same order as in (1) and renaming the
projections, we can write (3) as
T = 1
Q1 + 2
Q2 + … + m
Qm … (4)
To prove uniqueness, we shall show that p i in (1) and Qi in (4) are some.
Using the fact p2i = pi, pipj = 0 i j, we can have
T0 = I = p1 + p2 + … + pm
T1 = 1
p1 + 2
p2 + … + m
pm
T2 = 2
1
p1 + … + 2
m
pm
and Tn = n
1
p1 + … + n
m
pm for any positive integer n. … (5)
Now if g (t) is a polynomial with complex coefficient in the complex variable t, we can write
g (T) as
g (T) = g ( 1) p1 + g ( 2) p2 + … + g ( m
) pm
= g ( j ) pj … (by 5)
j 1
Let i
be a polynomial such that i
( i) = 1 and i
( j) = 0
if i j
Taking i
in place of g, we get
m m
(T) = i ( j ) pj ij pj pi
i
j 1 j 1
Hence for each i, let pi = i (T) which is a polynomial in T. The proof is complete if we show the
existence of i over the field of complex number.
(t 1) (t i 1 ) (t i 1 )...(t m )
Now i (t)
( i 1 )....( i i 1 ( i
) i 1) ( i m )
satisfies our requirements i.e. i
( i) = 1 and i
( j) = 0 if i j
Repeating the above discussion for Qi’s we get in a similar manner Q i = i

(T) for each i.
pi = Qi for each i.

Notes 31.1.8 Compact Operators
Definition: A subset A in a normed linear space N is said to be relatively compact if its closure A
is compact.
A linear transformation T of a normed linear space N into a normed linear space N is said to be
a compact operator if it maps a bounded set of N into a relatively compact set in N , i.e.
T:N N is compact of every bounded set B N, T(B) is compact in N .
31.1.9 Properties of Compact Operators
1. Let T : N N be a compact operator. Then T is bounded (continuous) linear operator. For,

let B be a bounded set in N. Since T is compact, T(B) is compact in N . So T(B) is complete
and totally bounded in N . Since a totally bounded set is always bounded, T(B) is bounded
and consequently T(B) is bounded, since a subset of a bounded set is bounded.
T is a bounded linear transformation and it is continuous.
2. Let T be a linear transformation on a finite dimensional space N. Then T is compact

operator. For, N is finite dimensional and T is linear T(N) is finite dimensional. Since any
linear transformation on a finite dimensional space is bounded. T(B) is bounded subset of
T(N) for every bounded set B N. Now if T(B) is bounded so is T(B) and is closed. T(N) is
finite dimensional, any closed and bounded subset of T(N) is compact, so that T(B) is
compact, being closed and bounded subset of T(N).
3. The operator O on any normed linear space N is compact.
4. If the dimension of N is infinite, then identity operator I : N N is not compact operator.
For consider a closed unit sphere.
S = {x N: x 1} then S is bounded.
Since N is a infinite dimensional.
I (S) = S = S is not necessarily compact.
Hence I : N N is not compact operator. But I is a bounded (continuous) operator.
Theorem: A set A in a normed linear space N is relatively compact every sequence of points
in A contains a convergent sub sequence.
Proof: Let A is relatively compact.
Since A A , every sequence in A is also sequence in A . Since A is compact, such a sequence

in A contains a convergent subsequence. Hence every sequence in A has a convergent
subsequence.
Conversely, let every sequence in A has a convergent subsequence.
Let (yn) be a sequence of points in A . Since A is dense in A , a sequence (xn) of points of A s.t.
1
xn yn … (1)
n

Notes
and x nk x A … (2)
we can find a (y nk ) of (yn) s.t.
y nk x = y nk x nk x nk x
y nk x nk x nk x
0 as n .
A is compact.
31.2 Summary
 If T is an arbitrary operator on a finite dimensional Hilbert space H, then the spectrum of

T namely (T) is a finite subset of the complex plane and the number of points in (T) does
not exceed the dimension n of H.
 Let T be bounded linear operator on a Hilbert space H. Then a scalar is called an eigenvalue
of T if there exists a non-zero vector x in H such that Tx = x.
 The closed subspace M is called the eigenspace of T corresponding to the eigenvalue .
 The set of all eigenvalues of an operator is called the spectrum of T. It is denoted by (T).
 The spectral resolution of the normal operator on a finite dimensional non-zero Hilbert
space is unique.
 A subset A in a normed linear space N is said to be relatively compact if its closure A is

compact.
31.3 Keywords
Eigenspace: The closed subspace M is called the eigenspace of T corresponding to the eigenvalue .
Eigenvalues: Let T be bounded linear operator on a Hilbert space H. Then a scalar is called an
eigenvalue of T if there exists a non-zero vector x in H such that Tx = x.
Eigenvalues are sometimes referred as characteristic values or proper values or spectral values.
Eigenvectors: If is an eigenvalue of T, then any non-zero vector x H such Tx = x, is called a

eigenvector.
Similar Matrices: Let A, B are square matrix of order n over the field of complex number. Then
B is said to be similar to A if there exists a n × n non-singular matrix C over the field of complex
numbers such that
B = C–1 AC.
Spectrum of an Operator: The set of all eigenvalues of an operator is called the spectrum of T and
is denoted by (T).
Total Matrices Algebra: The set of all n × n matrices denoted by An is complex algebra with
respect to addition, scalar multiplication and multiplication defined for matrices.
This algebra is called the total matrices algebra of order n.

1. If T (H) is a self-adjoint operator, then (T) = {m, M} where m, M are spectral values.
2. If T is self-adjoint operator then (T) is the subset of the real line [– T , T ].

–1
3. Let R (T) = (T – I) for a T B (X, X). Prove that R T 0 as .
4. Prove that the projection of a Hilbert space H onto a finite dimensional subspace of H is
compact.
Books Walter Rudin, Real and complex analysis, Third, McGraw-Hill Book Co., New York,
1987.
Erwin Kreyszig, Introductory functional analysis with applications, John Wiley &
Sons Inc., New York, 1989.
Online links www.math.washington.edu

chicago.academia.edu
www.math.ethz.ch/~ Kowalski/spectral-theory.pdf

LOVELY PROFESSIONAL UNIVERSITY
Jalandhar-Delhi G.T. Road (NH-1)
Phagwara, Punjab (India)-144411
For Enquiry: +91-1824-300360
Fax.: +91-1824-506111
Email: odl@lpu.co.in

DMTH505 Measure Theorey and Functional Analysis PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

DMTH505 Measure Theorey and Functional Analysis PDF

Загружено:

Авторское право:

Доступные форматы

MeasureTheoreyandFunctional

Produced & Printed by

Unit 1: Differentiation and Integration: Differentiation of Monotone Functions 1

Unit 2: Functions of Bounded Variation 11

Unit 3: Differentiation of an Integral 27

Unit 4: Absolute Continuity 36

Unit 5: Spaces, Hölder 50

Unit 6: Minkowski Inequalities 65

Unit 7: Convergence and Completeness 72

Unit 8: Bounded Linear Functional on the L p-spaces 82

Unit 9: Measure Spaces 100

Unit 10: Measurable Functions 106

Unit 12: General Convergence Theorems 141

Unit 13: Signed Measures 158

Unit 14: Radon-Nikodym Theorem 166

Unit 15: Banach Space: Definition and Some Examples 174

Unit 16: Continuous Linear Transformations 183

Unit 17: The Hahn-Banach Theorem 201

Unit 18: The Natural Imbedding of N in N** 211

Unit 19: The Open Mapping Theorem 217

Unit 20: The Closed Graph Theorem 225

Unit 21: The Conjugate of an Operator 231

Unit 22: The Uniform Boundedness Theorem 238

Unit 24: Orthogonal Complements 253

Unit 25: Orthonormal Sets 262

Unit 26: The Conjugate Space H* 273

Unit 27: The Adjoint of an Operator 283

Unit 28: Self Adjoint Operators 294

Unit 29: Normal and Unitary Operators 305

Unit 30: Projections 316

Unit 31: Finite Dimensional Spectral Theory 329

Unit 1: Differentiation and Integration: Differentiation Notes

1.1 Differentiation and Integration

1.1.1 Lipschitz Condition

1.1.2 Lebesgue Point of a Function

1.1.3 Covering in the Sense of Vitali

1.1.4 Four Dini's Derivatives

1.1.5 Lebesgue Differentiation Theorem

1.4 Review Questions

1.5 Further Readings

After studying this unit, you will be able to:

 Understand differentiation and integration

 Describe Lipschitz condition and Lebesgue point of a function

 State Vitali’s Lemma and understand its proof.

 Explain four Dini’s derivatives and its properties

 Describe Lebesgue differentiation theorem.

F : [a, b] R defined by F (x) = f (t) dt is continuous on [a, b]. Furthermore if f is

f (t) dt = F (x) – F (a) [a x b].

LOVELY PROFESSIONAL UNIVERSITY 1

Notes 1.1 Differentiation and Integration

1.1.1 Lipschitz Condition

|f (x) – f (y)| M |x – y|, x, y [a, b].

1.1.2 Lebesgue Point of a Function

Definition: A point x is said to be a Lebesgue point of the function f (t), if

1.1.3 Covering in the Sense of Vitali

The family M is called the Vitali Cover of set E.

2 LOVELY PROFESSIONAL UNIVERSITY

Obviously  1 < as  1 m (0) < .

Obviously, we have that I

find an integer N s.t.

element of E not belonging to the closed set I

small that I does not meet the I

Therefore we shall have  (I)  N 2 (I n 1 ) as by the method of construction we take