Академический Документы
Профессиональный Документы
Культура Документы
Analysis
DMTH505
MEASURE THEORY AND
FUNCTIONAL ANALYSIS
Copyright © 2012 Anuradha
All rights reserved
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 23: Hilbert Spaces: The Definition and Some Simple Properties 244
of Monotone Functions
CONTENTS
Objectives
Introduction
1.2 Summary
1.3 Keywords
Objectives
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
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
Definition: A function f defined on [a, b] is said to satisfy Lipschitz condition (or Lipschitzian
function), if a constant M > 0 s.t.
x n
1
Lim |f (t) f (x)|dt 0.
h 0 h x
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) .
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:
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).
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.
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
I Ir = , r = 1, 2, … N.
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
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.
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
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
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
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.
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.
f {x (1/n)} f (x)
Lim
Lim fn (x) = 1/n , x [a, b] ,
n 0 (1/n)
= f (x), a.e.
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
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 )
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
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 )
1 1
Lim inf n f (b) f (a) f (b) f (a)
n n n
b
f (x) dx f (b) – f (a)
a
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
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
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
F(x h) F(x)
Lim f (x) 0 … (v)
h o h
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.
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
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
1.2 Summary
A function f defined on [a, b] is said to satisfy Lipschitz condition if a constant M > 0 such
that
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
m* E I
k 1
k .
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
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 < .
1. If the function f assumes its maximum at c, show that D+ f (c) 0 and D– f (c) 0.
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
5. Every point of continuity of an integrable function f (t) is a Lebesgue point of f (t). Elucidate.
Notes
Unit 2: Functions of Bounded Variation
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
2.4 Review Questions
2.5 Further Readings
Objectives
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.
f br – f ar , wherever br ar ,
r 1 r 1
Notes
The condition br ar , means that total length of all the intervals must be
r 1
less than .
x y f x f y , x, y I
x y f x f y , x, y I
x y f x f y
x y f x f y
a x0 x1 x2 ... xn 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
a
V f lin V f .
a 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
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 .
Notes
Theorem 2: Let V, P, N denote total, positive and negative variations of a bounded function f on
[a,b]; then prove that
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,
where suprema are taken over all subdivisions of [a,b]. From (i), we have
Theorem 3: If f1 and f2 are non-decreasing functions on [a,b], then f 1–f2 is of bounded variation on
[a,b].
Notes
f xi – f xi 1 f1 xi – f1 xi 1 f2 xi – f2 xi 1
f1 b – f1 a f2 b – f2 a
b
V f and hence f is of bounded variation.
a
c b
V f V f
a a
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
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:
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.
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
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.
v x v c ,x c ,c ...(i)
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
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:
h xr 1 h xr f xr 1 g xr 1 f xr g xr
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
b b
V f and V g are finite.
a a
b
V h a finite quantity.
a
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
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 .
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.
(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
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.
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 .
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
Notes
Converse of above theorem is not necessarily true. These exists functions of bounded
variation but not absolutely continuous.
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
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
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).
n 1
Let V f xr 1 f xr
r 0
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
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
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.
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
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]
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
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
Notes
Example: Show that the function
x sin if 0 x 2
f x x is continuous
0,if x 0
or
show that there exists a continuous function without being of bounded variation.
Problem: Show that if f exists and is bounded on [a, b], then f BV [a, b].
f(xi ) f(xi 1 )
M |f(xi ) f(xi 1 )| M(xi xi 1 )
xi xi 1
b
V(f) M (x i xi 1 ) M(b a)
a
f B V [a, b].
1
f(x) x p sin for 0 x 1, f(o) 0, p 2.
x
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
2.2 Summary
n n
f br f ar whenever br ar ,
r 1 r 1
where a1 b1 a2 b2 ... an bn
x y f x f y , x, y 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
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.
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.
1. Show that sum and product of two functions of bounded variation is again a function of
bounded variation.
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].
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.
www.whitman.edu/mathematics/SeniorProjectArchive/.../grady.pdf
CONTENTS
Objectives
Introduction
3.2 Summary
3.3 Keyword
Objectives
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.
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
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
Notes x xo
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
[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
x x
Then
n n xi
n xi Notes
|f (t)|dt
i 1 x
i 1
= |f (t)|dt
a
|f|dt
a
F BV [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.
But f (t) dt 0
a
f (t) dt 0
A F
f (t) dt f (t) dt 0
A F
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
But f (t) dt 0
A
bn
f (t) dt 0
n an
bn
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
f (t) dt 0.
a
But it leads to the contradiction of the given hypothesis. Hence our supposition is wrong.
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(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
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
x x
or [F (t) f(t)] dt 0, x
a
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
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].
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
G n (x) (f fn )
a
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
fn(x) a.e. n.
since n is arbitrary, we have
F (x) f(x) a.e.
b b
Also by the Lebesgue’s theorem, i.e. “Let f be an increasing real-valued function defined on
[a, b].
Notes 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
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
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
F (x) = f (t) dt c ,
a
1. If f is an integrable function on [a, b] and if F (x) = f (t) dt F(a) then check whether
a
2. If F is an absolutely continuous function on [a, b], then prove that F (x) = f (t) dt C where
a
Notes
Unit 4: Absolute Continuity
CONTENTS
Objectives
Introduction
4.2 Summary
4.3 Keywords
Objectives
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.
f br f ar whenever br ar ,
r 1 r 1
Notes
Notes
2. The condition br ar means that total length of all the intervals must be less
r 1
than .
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 .
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
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
Proof: Given f x and g x are absolutely continuous functions on the closed interval [a,b],
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].
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 ;
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
1 is absolutely continuous.
f x .
g x
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
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.
Let f(x) be an absolutely continuous function over the closed interval [a,b].
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 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].
n m n
i
xik xik 1 bi ai .
i 1 k 1 i 1
n m
i
f x f1 b f2 b f1 a f2 a ,
a
f x is integrable also.
We get
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
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
F bi F ai .
i 1
F 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 .
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.
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
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.
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].
m* f E , being arbitrary.
m* f E 0 m f E 0.
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.....
(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 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].
Notes (ii) To show that f(x) is a continuous function. Note that if c', c'' F, then we have
1
If c' c'' , then pi = qi, for 1 i n 1 and hence
3n
1
f c' f c'' ...(i)
2n
f cn f c0 , when n .
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 .
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 f ' x dx f 0 .
0
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
Proof: Let f(x) be an absolutely continuous function in a closed interval [a, b] so that f (x) &
d
[F(x) f(x)] = 0.
dx
Integrating, we get
F(a) – f(a) = 0
n n
f br f ar , whenever br ar .
r 1 r 1
4.3 Keywords
f br f ar whenever br ar ,
r 1 r 1
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.
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).
mrich.maths.org
CONTENTS
Objectives
Introduction
5.1.1 Lp-spaces
5.2 Summary
5.3 Keywords
Objectives
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.
5.1.1 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
Notes
Note The symbol Lp is used for such classes when limits of integration are known and
mentioning of interval is not necessary.
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.
Note Non-negativity p 1, q 1.
1
b p
f p
= |f|p ,0<p< .
a
b b
or |g|p dx
a
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]
Now |f (x)| 1, x A1
|f| |f|p on A1 as p > 1
|f|dx <
a
|f|dx <
a
Notes
Therefore |f g|p dx |f g|p |f g|p dx .
A1 A2
Integrating, we have
|f g|p 2 p |f|p
A1 A1
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],
1
up q
s = … (1)
up vp
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
(1 t q ) q 1, t (0, 1],
For the remainder of the proof we are going to assume that u, v > 0.
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
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).
lim f (t) = –
t 1
and the equation (3) has exactly one solution in (0, 1), namely s, this forces
This means that, f is increasing on [0, s] and decreasing on [s, 1], and we are done.
and the fact that f (s) equals the value in (2) follows from an easy computation.
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.
Case n = 2.
b1
r = 1 .
bq1 bq2 q
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).
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
bq1 a q1
q q = p ,
b 1 b 2 a 1 a p2
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).
1 1
k 1 k p k q
a jb j a pj bqj + ak+1 bk+1
j 1 j 1 j 1
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
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
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
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)
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
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(x) g(x)
Set f (x) = , g(x)
f p g q
1 1
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
b
|f(x) g(x)|dx
1 or fg f p
g q
… (4)
f p g q
a
b b
f p
and g q
fg 1
< fg L [a, b]
|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.
If 0 < p < 1 and p and q are conjugate exponents, and f Lp and g Lq, then
(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
1 1 q p
q
p p Q q
Then fg = |fg |p |g|Q = |f| |g|
= |f|p.
|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
[|f(x)| – |g(x)|]2 0
On integrating, we get
b b b
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
( |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
b 2
b b
b b 1/2 b 1/2
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.
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
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.
|f g| |f| |g|
f+g 1
f 1
+ g 1.
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
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],
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
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
CONTENTS
Objectives
Introduction
6.2 Summary
6.3 Keywords
Objectives
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.
1/p
f p
= |f|p d
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:
|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).
b 1 /p b 1 /p b 1 /p
Similarly, if p >1 and a k, bk > 0, then Minkowski's sum inequality states that
Equality holds iff the sequences a 1, a2, ... and b1, b2, ... are proportional.
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
1 1 1
b p b p b p
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 < .
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
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)
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)
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
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)
1p 1q
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
1/p 1/p
p 1/p
1/p N N 1/p
But
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
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 < .
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
1/p 1/p
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.
CONTENTS
Objectives
Introduction
7.3 Keywords
Objectives
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.
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 .
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.
Sn = u1 + u2 + … + un
u un .
n 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
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
… … … … …
… … … … …
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
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) = }.
f (x) = 0, x E
m 1
or f (x) = lim
m
g1 gk 1
gk , for x E
k 1
= lim g m (x)
m
f (x) = lim
m
g m (x) a.e. in [a, b]
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.
|gm – f| 2 g.
|gm – f| 2g, m
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
fnm f < .
p
fm – f p
= fm fnm fnm f
p
fm fnm fnm f
p p
<( + )= .
Hence lim fm f p
=0
m
Or
Theorem: Prove that a normed linear space is complete iff every absolutely summable sequence
is summable.
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
n n
Notes
Sn – Sm p
= fi fi
i m 1 i m 1
fi
i no
<Sn> converges.
Sufficient part: Given that every absolutely summable sequence in the space X is summable.
1
fn – fm < , n, m nk … (ii)
2k
Setting g1 = fn1 and gk = fnk fnk 1 , (k > 1), we get a sequence <gk> s.t. its kth partial.
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.
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
Notes
fnk f .
2
fn f fn fnk fnk f ,
2 2
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
1 p<
So, we get
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
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)
x m – xn = xm – xo + xo – xn
x m – xo + x o – xn
< = by (1)
2 2
|fk (x) fk
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
A normed linear space is complete iff every absolutely summable sequence is summable.
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.
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 .
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.
where
Sn = u1 + u2 + … + un
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
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
8.4 Review Questions
8.5 Further Readings
Objectives
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
*
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).
|f (x)| k x , x N1 … (1)
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
Note L1 or simply L denotes the class of measurable function f (x) which are also
L-integrable.
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
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, … .
8.1.6 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
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:
sup fg = M <
f p
L 1
f simple
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
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
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) .
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
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)
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.
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
f (yn) 0 as n .
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
(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
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
(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 .
np * = np .
1
n p
p
x = |xi |
i 1
n n
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
1
n q
q
f |y i | … (2)
i 1
|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 |
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
1
n 1
p
|y i | q f
i 1
1
n q
|y i |q f … (7)
i 1
1
n q
q
f = |y i | , so that
i 1
Hence np * = nq .
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
Now, f (x) = xi yi
i 1
|xi ||y i |
i 1
n n
f (x) max. |y i | |x i |.
i 1
|y i |
xi = when|y i | max |y i | and x i 0 … (9)
yi 1 i n
otherwise
yi
(x) = =1
y
f … (10)
f is an isometric isomorphism of L to 1 * .
n
y
Hence n1 * 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 * .
n n
|f (x)| = xi yi |x i ||y i | .
i 1 i 1
n n
Hence we have
n n
|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
Hence, n * n1 .
Note We need the signum function for finding the conjugate spaces of some infinite
dimensional space which we define as follows:
sgn if 0
| |
0 if 0
1 1
= 1 and 1 < p < .
p q
or *p q .
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 .
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
x = xk ek … (4)
k 1
Sn x as n (Using (4))
f (Sn) = xk f(e k ) .
k 1
f (Sn) f (x) as n .
f (x) = xk f (e k ) … (5)
k 1
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 .
| |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)
1
n p
p
x = | k|
k 1
1
n p
x = | k |p
k 1
1
n q
q
= | k | … (8)
k 1
x = k e k , we get
k 1
n n
f (x) = k f (e k ) k k
k 1 k 1
| f (x) | f x ,
1
n n p
|f (x)| | k | q
f | k
q
|
k 1 k 1
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.
Let x p so that
x = xk ek
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 (ek) = k
for any k so that
Tg = ( k) and T is on *p onto q .
Tf = f so that T is an isometry.
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
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 = g q
… (1)
= F (f1) + F (f2)
So that
and F ( F) = fg d F(f) .
X
1 1
p q
|fg|d |f|p d |g|q d
X X X
= f p
g q
… (3)
|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)
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
q /p
f p
= g q
… (7)
q
= |g|q d g q
X
q
Hence g q
g = F (f) F f p.
q q /p
g q
= F (f) F g q
q q /p
g q
= g q
F … (8)
( g 0)
F = g q.
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
<
x h x
x h x h
= f(t) dt f(t) dt
x x
x h
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)
F(x h) F(x) 1
= Lim ,h
h 0 h n
2
and hence Lim G n (x) f(x) = 0.
n
Notes x (1 / n ) x (1 / n )
b b
2
Lim G n f 2
=0
n
Lim Gn f 2
n
or Lim f G n =0
n
f – Gn 2
<
f G no <
2
f–g 2
< (Taking G no = g)
x (1 / n o )
8.2 Summary
|f (x)| k x , x N1
f (x) = xk f (e k )
k 1
f (x) = xk f(e k )
k 1
|f (x)| k x , x N1
Continuous Linear Functional: A linear functional f is continuous if given > 0 there exists >
0 so that
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).
CONTENTS
Objectives
Introduction
9.2 Summary
9.3 Keywords
Objectives
Introduction
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.
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.
(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
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).
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 .
we have (E) n
(En) = 0.
Thus (E) = 0.
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.
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,
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
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.
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
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.
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,
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 () .
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, , ).
(ii) A implies AC
(iii)
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
CONTENTS
Objectives
Introduction
10.2 Summary
10.3 Keywords
Objectives
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.
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
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.
3. A function whose values are in the set of extended real numbers is called an extended
real valued function.
Sol: Let f be a constant function defined over a measurable set E so that f (x) = x E.
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
E f if c 0
c
E (cf > ) =
E f if c 0
c
(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, we shall prove that f + g is measurable over E. Let a be any real number.
Again, g is measurable
cg is measurable, c is constant.
a + cg is measurable a, c R
E (f > a – g) is measurable.
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).
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,
E (f > a + g) is measurable.
f – g is a measurable function.
We have
E if 0
E (|f| > ) =
[E (f )] [E(f )] if 0
|f| 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
[E (f )] [E(f )] is measurable.
E (f2 > ) is measurable because both the sets on RHS are measurable.
f2 is measurable over E.
1
(f g)2 (f g)2 is a measurable function over E.
4
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
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.
0, if x is irrational
f (x) =
1, if x is rational
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.
Proof: Let be any real number and let E 1 = E (f > ) and E2 = E (g > )
= {x E : f (x) g (x)}
m (E1 E2) = 0.
E2 is measurable.
Hence g (E).
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
Note f = f+ – f–1
and |f| = f+ + f–1
Theorem 3: A function is measurable iff its positive and negative parts are measurable.
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.
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.
Then m (E) = 0
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.
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
1 if x A
A
(x) = 0 if x A
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.
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.
A real valued function is called simple if it is measurable and assumes only a finite number of
values.
= i Ai
i 1
where Ai = {x : (x) = i
}
Thus we can always express a simple function as a linear combination of characteristic function.
Notes
= i Ai
i 1
where Ai = {x : (x) = i
}
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= i Ai
i 1
and g= j Bj
j 1
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
n Notes
Similarly fg = ak bk Ek
k 1
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
Note Before proving this theorem first prove the previous 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
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
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
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) <
= (E
p
Ap )
= (E
p
p 1 Ap )
= E
p
p
1
|fn (x) – f (x)| < , n Np.
p
1
Let us choose p such that < , we get
p
Notes
Note Egoroff’s theorem can be stated as: almost every where convergence implies almost
uniform convergence.
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
m x E fn (x) f(x)
on.
Let Ak = x:x
i k
E, fni (x) f (x) i ,k N and A A
k 1
k .
But m (Ak) i 0 as k .
i k
Hence m (A) = 0.
But k
0 as k .
10.2 Summary
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| = f+ + f–1
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
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
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.
= i Ai
i 1
where Ai = {x : (x) = i
}
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?
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.
(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.
CONTENTS
Objectives
Introduction
11.1 Integration
11.2 Summary
11.3 Keywords
11.4 Review Questions
Objectives
Define the Riemann integral and Lebesgue integral of bounded function over a set of
finite measure.
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
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].
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
R f (x) dx inf S
a
where the infimum is taken over all possible sub-divisions of [a, b].
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
(x) = i
x [xi–1, xi]
b x1 xn
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
n Notes
= inf M i (xi xi 1 )
i 1
(x) f (x)
= sup m i (x i xi 1 )
i 1
(x) f (x).
Characteristic Function
The function E
defined by
1 if x E
(x) =
E 0 if x E
Simple Function
A linear combination (x) = i Ei (x) is called a simple function if the sets Ei are measurable.
i 1
However, a function is simple if and only if it is measurable and assumes only a finite number
of values.
Canonical Representation
= 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
= i Ei .
i 1
We sometimes abbreviate the expression for this integral . If E is any measurable set, we
Theorem 1: Let and be simple functions which vanish outside a set of finite measure, then
= 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
Since = ai m Ei
i
N N
= a ak m Ek b bk m Ek
k 1 k 1
= a b
– 0 a.e.
(a b ) = a b
( ) =
Since – 0 a.e. is a simple function, by the definition of the elementary integral, we have
( ) 0
b b b
where 1
and 1
vary over all step functions defined on [a, b].
b b
b b
where and vary over all the simple functions defined on [a, b]. Thus from the above relation,
we have
b b b b
b b
b b
f (x) dx R f (x) dx
a a
1, if x is rational
f (x) = 0, if x is irrational
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
Let A1 be the set of all irrational numbers and A 2 be the set of all rational numbers in [0, 1].
= 0 mA1 1 mA 2 = 1.
= 0 mA1 1 mA 2 = 1.
Theorem 3: If f and g are bounded measurable functions defined on the set E of finite measure,
then
Hence f |f|
f f f
A B A B
Let 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
af = a f … (i)
E E
(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
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
f – g = 0 a.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
2 if x 0
f (x) =
0 if x 0
and g (x) = 1 x.
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.
=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
|f| f |f|
E E E
f |f|
E E
Proof of 5: f = f A B
A B
Now A B = A B A B
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
f = sup h ,
h f
E E
m {x : h (x) 0} <
f.
Since f = 0 a.e. on E
0 a.e.
(x) dx 0
E
f dx 0 … (1)
E
Since f = 0 a.e.
0 a.e.
(x) dx 0
E
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
m En = 0
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
In particular, g < ,
E
Since f = (f g) g
E E E
f g = (f g) .
E E E
f+ = max (f, 0)
and negative part f– by f–
f – = max (–f, 0)
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
(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
f = f f.
A B A B
(cf)+ = cf+
(cf)– = cf
(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
f f2 = f1 f
E E E E
f f = f1 f2 … (3)
E E E E
f = f f
Hence f = f1 f2
E E 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) = (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
then (g 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
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
f (t) dt <
xo
x xo
Let and be simple functions which vanish outside a set of finite measure, then
Let A1 be the set of all irrational numbers and A2 be the set of all rational numbers in [0, 1].
f sup h
h f
,
E E
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
= i Ei ,
i 1
where Ei = {x : (x) = i
}.
1 if x E
(x) =
E 0 if x E
Elementary Integral: If vanishes outside a set of finite measure, we define the elementary
n
= i Ei .
i 1
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.
However, a function is simple if and only if it is measurable and assumes only a finite number
of values.
f = sup h ,
h f
E E
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].
S = (xi xi 1 ) M i
i 1
and s = (xi xi 1 ) m i
i 1
mi = inf f (x)
xi 1 x xi
(a) cf c f, c 0
E E
Notes
(b) (f g) f g
E E E
(c) f 0 f 0 a.e.
4. If f is integrable over E, then show that |f| is integrable over E, and f |f|.
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
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
12.4 Review Questions
12.5 Further Readings
Objectives
State and prove monotone convergence theorem and Lebesgue dominated convergence
theorem.
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.
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.
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
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.
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
f is also measurable on E
Then measurable set A E with mA < and a positive integer N such that
4M
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
lim fn = f
n
E E
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
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
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
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
sup h h lim fn
n f n
E E E
f f lim fn
n
E E E
Remarks:
1, n x n 1
fn(x) =
0, otherwise
with E = R
then f lim fn
n
E E
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
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
Define a function hn by
Since hn = h or hn = fn
hn is measurable function on E
If hn = h, then hn h
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
fn f
fn f
lim fn f … (2)
n
f lim fn lim fn f
n n
f lim fn
n
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.
Define a function hn by
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 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
fn f
fn f
lim fn f … (2)
n
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
|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
fn f at each point
lim fn f
n
E E
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
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
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.
|f| g a.e. on E.
(f g) lim (fn g)
n
E E
f lim fn … (1)
n
E E
(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
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
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
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
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
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],
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
lim fn f … (1)
f lim fn … (2)
f lim fn lim fn f.
n n
x
Example: If > 0, prove that Lim 1 x 1dx e x .x 1dx , where the integrals are
n n
o o
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
n
x 1
= Lim 1 .x dx
n n
o
= e x .x 1
dx
o
1
x sin x
dx 0(n 1 ) as n .
1 (nx)
o
nx sin x
fn(x) = , n = 1, 2, ………
1 (nx)
Notes
nx sin x
1
1 (nx)
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
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
f lim fn
n
E E
f lim fn
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.
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
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.
2. Let <fn> be an increasing sequence of non-negative measurable functions, and let f = lim f n.
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
Books G.F. Simmons, Introduction to Topology and Modern Analysis, New York: McGraw
Hill, 1963.
H.L. Royden, Real analysis, Prentice Hall, 1988.
CONTENTS
Objectives
Introduction
13.2 Summary
13.3 Keywords
13.4 Review Questions
Objectives
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.
Definition: Let the couple (X, ) be a measurable space, where represents a -algebra of
subsets of X. An extended real valued set function
: [– , ]
(ii) ( ) = 0.
A
n 1
n
n 1
(A n ),
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.
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
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
(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 .
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.
Proof: Let (X, ) be a measurable space and let be a signed measure defined on (X, ). Let <An>
n
Set n = B An A Cn 1 A C1 , n N.
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
if B = B
n 1
n , we get … (ii)
(B) = (B n ) … (iii)
n 1
In view of (i).
(B) 0.
Thus, we have
countable union A
n 1
n
is measurable,
A= A
n 1
n is measurable
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
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
and (B – E1 E2 ) (E1 E2 ) = ,
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
A =B– E
n 1
n or B = A E
n 1
n … (iii)
(B) = (A) + (E n ) .
n 1
1
> (A) + … (iv)
k 1
nk
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
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 .
1
(E) nk 1
Letting nk , we obtain
(E) 0.
Thus we have
(1) A is measurable.
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 )
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.
(Q – En) 0.
Since (Q – En) En =
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 is negative set
(A Q) K [using (i)]
(A) + (Q) K,
(A) 0,
P = QC is a positive subset of X
Q is a negative subset of X.
Thus 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.
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
: [– , ]
(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
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
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
Signed Measure: Let the couple (X, ) be a measurable space, where represents a -algebra of
subsets of X. An extended real valued set function
: [– , ]
(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. .
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
CONTENTS
Objectives
Introduction
14.2 Summary
14.3 Keywords
Objectives
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.
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
Notes
Notes
Radon-Nikodym Theorem
(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
Notes
Proof: Let P = {F
b a
a Fb } .
Evidently (P) = 0.
Let Fa = Fa P.
Thus we have
x Fa f(x) a a.e.,
except for x P.
x X Fa f(x) a a.e.,
Let PO = X and QO = .
Pa Qa = X,
and Pb Qb = X.
Therefore, Qa – Qb = Qa – (X – Pb)
= Qa Pb.
(Qa – Qb) = 0.
f (x) a, a.e. x Pa
where f is measurable
Qr 1 Qr
Define Ar = A
no no
Notes
Qr
A = A– .
no
Evidently, A = A A
r o
r ,
(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
and (A ) = 0 if (A ) = 0 [ )]
In either case, (A ) = fd .
Ar
1 1
(A) (A) fd (A) (A r ).
no no
A
(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
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.
(Bn) 0.
If C = {x X : f (x) g (x)}
= (A
n 1
n B n ),
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 .
( 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
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 .
d d d
d d d
+ –
Proof: Since we may write = – and
d d( ) d d( )
, .
d d d d
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
= 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
(A) = fd , 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 .
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
(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 .
1
d d
1. Show that ,
d d
where and are -finite signed measures and , .
Book G.E. Shilov and B.L. Gurevich, Integral, Measure and Derivative: A Unified Approach,
Richard A. Silverman, trans. Dover Publications, 1978.
CONTENTS
Objectives
Introduction
15.2 Summary
15.3 Keywords
Objectives
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.
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:
With this notation the above conditions (i) – (iii) assume the following forms:
(i) x 0, x = 0 x = 0;
(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.
(iii) d (x, y) = x – y = x – z + z – y (z = N)
Notes
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.
or xn x as n
xn x xn – x 0 as n
x y x – y for any x, y N
Proof: We have
x = (x – y) + y
x–y + y
x – y x–y … (1)
–( x – y ) = y – x y–x
Therefore
–( x – y ) x – y so that
x – y – x–y … (2)
x y x–y
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.
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,
Notes
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.
M [S] .
Hence [S] = M.
xn – x 0 as n .
OR
A normed linear space which is complete as a metric space is called a Banach space.
xm – x n 0 as m, n , where (x n) N, then
ax N such that
xn – x 0 as n .
Hence M is closed.
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.
xn x in B as B is complete.
We show that x 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
Further, let z1, z2 C and let z 1 and z 2 be their complex conjugates, then
= 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
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
x 0
and x = 0 |x i |2 = 0 xi = 0, i = 1, 2, …, n
i 1
x=0
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
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
= ( x + y ) ( x + y ).
x+y x + y .
1 1
n 2 n 2
2 2
(iii) x = | xi | | | |x i |
i 1 i 1
=| | x .
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).
n
2
x(m
i
)
zi 2
i 1
xm – z 2
< 2 Notes
xm – z <
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.
x y x y for any x, y N.
15.3 Keywords
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.
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.
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.
CONTENTS
Objectives
Introduction
16.2 Summary
16.3 Keywords
Objectives
Introduction
In this unit, we obtain the representation of continuous linear functionals on some of Banach
spaces.
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.
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
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.
xn xo f (xn) f (x)
yn y f (yn) f (y)
Let yn y as n
since f is linear.
As yn y y n – y + xo xo by hypothesis
f (yn) f (y) as 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.
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
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 .
Let xn 0 as n then
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.
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 .
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 .
|f (x)| M x ,M 0.
|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
f = inf {M : M M}
f (x) f x f .
Therefore, we have
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
sup f(x) = f .
x 1
Consider x = 1, we have
f (x) f x f
Notes So that
Now consider
f (x)
f = sup
x 0 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
sup f (x) f
x 1
… (8)
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
by the basis.
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
| f (x)| M x 0
.
We shall prove, in this section, the representation theorems for functionals on some concrete
Banach spaces.
Proof: Let (e1, e2, …, en) be a standard basis for L so that any x = (x 1, x2, …, xn) L can be written
as
yi = f (ei).
where f (x) = xi y i is an isomorphism of L onto the linear space L of all function f. We shall
i 1
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.
np * = nq .
1
n p
p
x = xi
i 1
n n
Now |f (x)| = xi yi xi yi
i 1 i 1
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
1
n q
q
f yi … (2)
i 1
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
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 q
q
yi f … (7)
i 1
y f is an isometric isomorphism.
Hence np * nq .
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 * .
|f (x)| = xi yi
i 1
xi yi
i 1
n n n
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.
yi
x = =1
y
||f|| … (10)
From (8) and (10), we obtain
y f is an isometric isomorphism of L to n1 * .
Hence n1 * n .
L = n * .
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
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
isomorphism.
Hence, n * n1 .
1 1
1 and 1 < p < .
p q
or *p q .
p
Proof: Let x = (xn) p so that xn . … (1)
n 1
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
sn x as n . (Using (4))
f (sn) = xk f(e k ) .
k 1
f (sn) f (x) as n
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)
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
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)| f x ,
1
n n p
q q
|f (x)| k f k
k 1 k 1
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.
Let ( k) q , we shall show that there is a g *p such that T maps g into ( k).
x = xk ek .
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 (ek) = k
for any k so that
Tg = ( k) and T is on p onto p .
*
Tf f so that T is an isometry.
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
f Tf … (13)
*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.
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
1
m T–1(y) y T–1(y) y
m
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
1
m x T(x) where m = > 0.
k
This completes the proof of the theorem.
x k1k x B.
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 .
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 .
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.
|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.
f(x)
|| f|| = sup
x 0 x
Second Conjugate: The conjugate space (N*)* of N* is called the second conjugate space of N .
i.e. *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 = 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
CONTENTS
Objectives
Introduction
17.2 Summary
17.3 Keywords
Objectives
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.
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 .
fo : M o R by
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,
= fo (x + y + ( + ) xo)
= fo (x + xo) + fo (y + xo)
= fo (x1) + fo (y1)
fo (ky) = fo [k (x + xo)]
= fo (kx + k xo)
= k f (x) + k ro
y = x.
fo = f .
= f .
Thus, fo f … (A)
|f (x2 – x1)|
f ( x2 + xo + – (x1 + xo) )
= f x2 + xo + f x1 + xo
sup f (y) f y xo ro
y M
inf f (y) f y xo
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
1 1
ro f (x) f x xo which implies that
f (x) + ro = fo (x + xo) f x + xo
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 f
fo = f
Let N be a normed linear space over C and f be a complex valued functional on a subspace M of
N.
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) = g (x) + g (y)
If R, then we have
f ( x) = g ( x) + i h ( x)
Since f is linear
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.
or g (i x) + i h (i x) = – h (x) + i g (x)
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).
Notes So that
fo ((a + i b) x) = (a + i b) fo (x)
fo = f on M.
Thus the complex valued functional fo is real and so it has only real part so that
= go x ,
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
Hence fo = f .
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)
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.
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.
= fo (e–i x).
Corollary 1: Deduce the Hahn-Banach theorem for normed linear spaces from the generalised
Hahn-Banach theorem.
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)
f fo … (4)
fo = f
Notes
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.
M = { xo : any scalar}.
Define f : M F (R or C) by
f ( xo) = xo .
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)
|f (x) | = x
|f(x)|
sup = 1 for x M and x 0.
x
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
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.
f (x – y) = f (x) – f (y) 0
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.
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.
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)
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
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
5. A normed linear space is separable if its conjugate (or dual) space is separable.
CONTENTS
Objectives
Introduction
18.2 Summary
18.3 Keywords
18.4 Review Questions
Objectives
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**.
J (x) = Fx x 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.
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.
II. Every normed linear space is a dense linear subspace of a Banach space.
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
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).
Fx ( f + g) = ( f + g) x = f (x) + g (x)
= Fx (f) + Fx (g)
Fx is linear
Fx is bounded.
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
Hence Fx x … (2)
To prove the reverse inequality we consider the case when x = 0. In this case (2) gives
Fo = 0 = 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
f (x) = x and f = 1.
= sup { f (x) : f = 1}
Fx = x … (4)
Finally, we show that J is an isometric isomorphism of N into N**. For any x, y N and scalar.
= Fx (f) + Fy (f)
F x+y = Fx + Fy … (6)
Notes Hence F x
= Fx … (7)
and J( x)
= F x = Fx = J (x) … (9)
(8) and (9) J is linear and also (4) shows that J is norm preserving.
Hence J defines an isometric isomorphism of N into N**. This completes the proof of the theorem.
np * = np .
But nq * = np
Hence np * * = np
and n * * = n for p =
* *q p .
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)
We have f (x j ) 0
j j j ij i for i = 1, 2, …, m, when x = x i.
j j
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)
N* is m-dimensional.
18.2 Summary
J (x) = Fx x N,
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
J (x) = Fx x N,
J (x) = Fx x N,
1. Let X be a compact Hausdorff space, and justify the assertion that C (X) is reflexive if X is
finite.
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.
CONTENTS
Objectives
Introduction
19.1.1 Lemma
19.2 Summary
19.3 Keywords
Objectives
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.
19.1.1 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 .
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
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 .
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.
Therefore we have
(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
Let y B such that y < . Then y T(S 1 ) and therefore there exists a vector x 1 B such that
S /2 T(S1/2 ) and y – y1 S /2
1
x3 , y y1 y2 and y2 = T (x2)
2 22
1
xn , yn T(x n ) and y (y 1 y2 yn )
2n 1
22
sn = x1 + x2 + … + xn
x1 + x2 + … + xn
1 1 1
< 1
2 22 2n 1
1
2 1
2n
<2
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)
S T (S3)
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.
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).
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)
S (y; r1) = y + S r1
y + T (Sr)
= T (x) + T (Sr)
= T (x + Sr)
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.
Proof: We know that a one-to-one continuous open map from B onto B is a homeomorphism.
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 .
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 .
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 = f + f .
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.
f = |f (i)|, i I.
T (f) |f(i)| i
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 .
x = Lim x nk
n
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.
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.
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 .
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
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.
CONTENTS
Objectives
Introduction
20.2 Summary
20.3 Keyword
Objectives
After studying this unit, you will be able to:
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.
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 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.
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
This establishes the triangular inequality and therefore N × N is a normed linear space.
Let (xn, yn) be a Cauchy sequence in N × N . Given > 0, we can find a n o such that
(xn, yn) (xo, yo) in the norm of N × N and (xo, yo) N×N.
N × N is complete.
Notes
Definition: Let N and N be normed linear spaces and let M be a subspace of N. Then a linear
transformation
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.
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.
xn x and T (xn) y.
GT GT
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.
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.
I : B1 B defined by
“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.”
xn – xm 1
= xn – xm + T (xn – xm) 0 as m, n
Now xn – x 1
= xn – x + T (xn – x)
= xn – x + T (xn) – T(x)
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 .
GT is closed in B × B.
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.
I:B B.
We claim that G1 is closed.
Let xn x in B and xn y in B .
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).
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
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).
CONTENTS
Objectives
Introduction
21.2 Summary
21.3 Keywords
Objectives
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*.
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))
Notes (c) T (N*) N* T is continuous, where T is a linear transformation of N into itself which is
not necessarily continuous.
Proof:
[T (f)] ( x + y) = f (T ( x + y))
part (a).
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
By definition of T , f (T (S)) is bounded if and only if [T (f)] (S) is bounded for each f in
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.
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:
T* : N* N* such that
is an isometric isomorphism of b (N) into b (N*) reverses products and preserves the identify
transformation.
= ( j) T (x) + ( g) T (x)
= [j (T (x))] + [g (T (x))]
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}
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
T = T* . … (4)
for every T (N) is an isometric isomorphism which reverses the product and preserve the
identity transformation.
(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)
= f ( T (x) + T1 (x) )
x N. Hence we get
= T* T *1 (f)
is linear.
To show is one-to-one, let (T) = (T1)
Then T* = T *1
T* T *1 =0
(T – T1)* = 0 T – T1 = 0 or T = T1.
is one-to-one.
Finally we show that reverses the product and preserves the identity transformation.
= f (T (T1 (x))
Hence, we get
(T T1)* = T *1 T* so that
(T T1) = (T T1)* = T *1 T.
I* = I so that (I) = I* = I
(T*)* = T**
Hence T** = T* = T .
J (x) (T* (x )) = 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.
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)*
By theorem 2, the mapping :T T* reverse the product and preserves the identity
(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:
If f N+, then T (f) is defined as
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:
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:
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))
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.
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.
CONTENTS
Objectives
Introduction
22.1 The Uniform Boundedness Theorem
22.1.1 The Uniform Boundedness Theorem – Proof
22.1.2 Theorems and Solved Examples
22.2 Summary
22.3 Keywords
22.4 Review Questions
22.5 Further Readings
Objectives
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.
(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)
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
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
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 .
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
This completes the proof of the theorem.
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
B= F
m 1
m . Since B is complete. It is of second category. Hence by Baire’s theorem, there is a
= fn (x + xo) – fn (xo)
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
This completes the proof of the theorem.
Example 1: Show that the completeness assumption in the domain of (Ti) in the uniform
boundedness theorem cannot be dropped.
= x (t) = a n t n , an 0
n 0
x = max {|an|, n = 0, 1, 2, …}
n 1
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)
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 .
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
Example 2: Let (an) be a sequence of real or complex numbers such that for each x = (x n)
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
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*.
x 1
x X … (1)
2
> 0 such that |f (x)| 2
x x N … (2)
|f (x)| 1 2
x X.
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
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*.
That is, Fxi is a bounded set of numbers. Since the norms are preserved in natural imbedding,
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
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.
CONTENTS
Objectives
Introduction
23.2 Summary
23.3 Keywords
23.4 Review Questions
Objectives
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.
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:
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.
Proof: (a) ( x – y, z) = ( x + (– ) 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)
= (x, y) (x,z)
Notes
Notes
1. Part (b) shows an inner product is conjugate linear in the second variable.
Solution: Let x = (x1, x2, ……, xn), y = (y1, y2, ……, yn) n2 .
(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
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
Proof: If y = 0, we get y = 0 and also theorem 1 implies that |(x, y)| = 0 so that (1) holds.
But
(x – y, x – y) = (x, x) – (x, y) – ( y, x) + ( y, y)
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.
(i) x = (x, x) x 2
= (x, x) 0 and x = 0 x = 0.
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.
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:
H3 : (x, x) = x 2,
We have already shown in earlier example that n2 is an inner product space. Also n2 is a
2. 2 is a Hilbert space.
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
We first show that inner product is well defined. For this we are to show that for all x, y in 2 the
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
Therefore (x, y) = x n y n is convergent so that the inner product is well defined. The condition
n 1
= (x, x + y) + (y, x + y)
= (x, x – y) – (y, x – y)
2
= (x, x) – (x, y) – (y, x) + y … (2)
Notes Theorem 5: In a Hilbert space the inner product is jointly continuous i.e.,
Proof: We have
|(xn, yn) – (x, y)| = |(xn, yn) – (xn, y) + (xn, y) – (x,, y)|
Since xn x and y n y as n .
Therefore
Theorem 6: A closed convex set E in a Hilbert space H continuous a unique vector of smallest
norm.
To prove the theorem it suffices to show that there exists a unique x E s.t. x = .
Lim xn = … (1)
n
xm xn
Convexity of E implies that E . Consequently
2
xm xn
xm + xn 2 … (2)
2
0 as m, n (Using (1))
2
xm – x n 0 as m, n
Notes
Now x = Lim
n
xn
= Lim
n
xn ( norm is continuous mapping)
= .
Uniqueness of x.
x y
Convexity of E E
2
x y
… (3)
2
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
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 .
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:
23.3 Keywords
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:
1. For the special Hilbert space n2 , use Cauchy’s inequality to prove Schwarz’s inequality.
Books Bourbaki, Nicolas (1987), Topological vector Spaces, Elements of Mathematics, BERLIN:
Springer – Verlag.
Halmos, Paul (1982), A Hilbert space Problem Book, Springer – Verlag.
CONTENTS
Objectives
Introduction
24.2 Summary
24.3 Keywords
Objectives
Understand theorems on it
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.
By definition,
x y y x
(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.
( 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
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.
y x i.e. (y, x) = 0
2
Now x+y = (x + y, x + y)
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.
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:
(d) S1 S2 S2 S1 (e) S S
Proof:
Thus x H x {0} .
H {0} .
Hence {0} = H
(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
S S {0}
x S1
S2 S1
by definition of (S ) , x (S ) .
Thus x S x S .
S S .
Proof: We have
= (0) + (0)
=0
x1 + x2 S
S is a subspace of H.
Next we shall show that S is a closed subset of H.
(xn, y) (x, y) as n
x S.
So, S is complete and hence a Hilbert space. This completes the proof of the theorem.
M=M .
M being a subspace of H.
We know 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 .
M M = {0} … (2)
From (1) and (2) we conclude that z = 0, a contradiction to the fact that z o is a non-zero vector.
Hence M = M .
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.
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.
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.
= lim xn + lim yn
M + N is closed.
H=M+M.
N M and N M .
Hence N M M = {0}.
N = {0}
N = {0} = H … (1)
N =N … (2)
N = M + M = H.
H =M+M ,
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.
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.”
x – yo = d.
z o = x – yo = d > 0.
zo is a non-zero vector.
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).
yo + M M.
x – (yo + y) d
Now zo – y = x – (yo + y) d = zo
2 2
zo – y zo
or (z o , y) (z o , y) (y, y) 0 … (1)
Let us take (z o , y) .
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
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.
zo M.
24.2 Summary
S = {x H:x y y S}
24.3 Keywords
S = {x H:x y y 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.
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.
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.
CONTENTS
Objectives
Introduction
25.2 Summary
25.3 Keywords
25.4 Review Questions
Objectives
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.
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.
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
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.
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.
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.
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
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
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).
n n
= (x, e j ) (x, e i ) (e i , e j )
i 1
=0
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.
2
2 x
Sn = e i : (x, e i ) .
n
2
2
(x, ei ) x … (2)
ei S n
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
If (x, ei) = 0 for each i, then S is empty. Otherwise S is either a finite set or countable set.
|(x, ei)|2 = 0 x 2.
If S is finite, then we can write S = {e1, e2, …, en} for some positive integer n.
Notes If S is countable infinite, let S be arranged in the definite order such as {e 1, e2, …, en, …}.
Hence every series obtained from this by rearranging the terms is also convergent and all such
series have the same sum.
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.
|(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,
n 1
Note: From the Bessel’s inequality, we note that the series |(x, e n )|2 is convergent series.
n 1
(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 = x – 0 = x.
x ej for every j.
Now let S is not empty. Then S is either finite or countably infinite. If S is finite, let
and prove that x (x, ei ) ei ej for each j. This result follows from (2) of theorem (1).
n 1
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 .
s = (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.
Notes n
Let sn = (x, fi ) fi
i 1
s = (x, fn ) fn .
n 1
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
Now consider
If ej S, then
lim (s n , e j ) = 0
n
For x, y B, x y, we have
2 2 2
x–y = x + y =2
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
To show that S is linearly independent, we have to show that every finite subset of S is linearly
independent.
e +
1 1
e +…+
2 2 n n
e =0 … (1)
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
(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 .
(x, e i )(y, e i ) = 0 i
2
(x, ei ) (y, ei ) to be number 0 and we have 0 x y 2.
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)
Finally let S is countably infinite. Let the vectors in S be arranged in a definite order as
Let us define
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
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
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.
(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
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
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.
Book Sheldon Axler, Linear Algebra Done Right (2nd ed.), Berlin, New York (1997).
Notes
Unit 26: The Conjugate Space H*
CONTENTS
Objectives
Introduction
26.1.1 Definition
26.2 Summary
26.3 Keywords
Objectives
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**.
26.1.1 Definition
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.
fy : H C defined as fy x x, y x H.
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.
fy x x, y x.y ...(1)
(Schwarz inequality)
Sup fy x
Further let y 0. Then from (1) we have y.
x
we get fy y ...(2)
y
Since y 0, is a unit vector.
y
y
fy fy ...(4)
y
Notes
y y 1
But fy ,y y, y y ...(5)
y y y
fy y
fy y
So let us take 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 .
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
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
f y
f x x, y x y.
f x
sup y.
x 0 x
f y ...(5)
2
y y, y f y f y
y f ...(6)
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).
(x,y1 – y2) = 0 x H.
2
y1 y 2 , y1 y 2 y1 y 2 0
y1 y 2 0
y1 y2
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
xn
f x as x xn M.
n 1
n
2 2
2
x, x x ,
6 6
2
1
since .
n 1
n2 6
f x x, y x M.
Notes 1
Using the definition of f we have f(x) = .
n
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.
the completeness assumption cannot be left out from the Riesz-representation theorem.
Proof:
y1 y2 y1 y 2 for y 1 , y 2 H.
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
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
x, y1 x, y2 x, y1 x, y2 0
x, y1 y2 0 x H ...(2)
y1 y 2
2
0 Notes
y1 y2
is one-to-one.
f(x) = (x,y)
f = fy so that y = fy = f.
y1 y2 fy fy
1 2
fy f y2
1
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 fy
y y
is not linear. Such a mapping is called conjugate linear.
Theorem 4: If H is a Hilbert space, then H* is also an Hilbert space with the inner product defined
by
Proof: We shall first verify that (1) satisfies the condition of an inner product.
f y fy
f y fy fy .
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)
(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.
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.
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
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.
By definition of J, J(x) = Fx. Hence to show T2.T1 = J, we have to prove that Fx = Ffx . Notes
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 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*.
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:
H2: x, y y,x
2
H3: x,x x
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:
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)
www.arvix.org
Notes
Unit 27: The Adjoint of an Operator
CONTENTS
Objectives
Introduction
27.2 Summary
27.3 Keywords
Objectives
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.
Let T be an operator on Hilbert space H. Then there exists a unique operator T* on H such that
Theorem 1: Let T be an operator on Hilbert space H. Then there exists a unique operator T* on H
such that
Proof: First we prove that if T is an operator on H, there exists a mapping T* on H onto itself
satisfying
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)
under the natural correspondence. The product of the above three mappings exists and it is
denoted by T*.
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)
y fy so that
fy Tx Tx, y ...(6)
T * fy x fz x x,z ...(7)
T * fy x Tx, y ...(9)
T * fy x x, T * y ...(10)
Tx, y x,T * y x, y H.
Notes
T * x, y x,Ty since
T * x, y y, T * x = Ty, x = x, Ty = x, Ty
T * x, y x, Ty .
T * x, y = x, Ty
T * x, y x, Ty = x n+1 y n Sx, y ,
n=1
where S x x 2 , x 3 ,...
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.
x,T * y1 y2 Tx, y1 y2
Tx, y1 x,T * y 2
x, T * y1 x, T * y 2 .
x,T * y1 y2 x, T * y1 x, T * y 2
Notes
x, T * y 1 T * y2 .
T* is linear.
T* is bounded
2
T*y T * y, T * y
TT * yy
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.
T*y T y.
since T is bounded,
T M so that
T* is bounded.
T* is continuous.
Uniqueness of T*.
Tx, y = x,T * y x, y H.
Then we have
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 *.
This completes the proof of the theorem.
Notes
1. We note that the zero operator and the identity operator I are adjoint operators. For,
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
xn
T x ,0,0,...... ...(2)
n=1
n
xn
T x, y y1 .
n=1
n
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
T e n ,e1 e n ,T * e1
Notes
(iv) T* T
2
(v) T*T T
= (T1x + T2x, y)
= (T1x, y) + (T2x, y)
x T1T2 * y T1T2 x, y
T1 T2 x , y
Notes
T2 x,T1 * y
x, T2* , T1* y
x, T2* T1* , y
x, T *y T x, y Tx , y
Tx, y
x, T * y x, T*y
x T* y .
T* T *.
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
T* T ...(2)
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)
2
Tx Tx, Tx
T * Tx, x
T * T x, x
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
2
if x 1 , then Tx T*T .
2
Therefore, Sup Tx : x 1 T*T
2
T T*T . ...(6)
Tn T, then T n* T *.
We have
T *n T* Tn T *
Since Tn T as n
Tn* T * as n .
1 1
T* T *.
T T * for every T H .
Now T1 T2
T *1 T *2
T1 T2
is one-to-one.
is onto:
T * = T* * =T.
T* T is onto.
1 1
T* T *.
Example: Show that the adjoint operation is one-to-one onto as a mapping of (H) into
itself.
(T) = T* T (H)
is one-one:
(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.
(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
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.
sepwww.standford.edu/sep/prof/pvi/conj/paper_html/node10.html.
Notes
Unit 28: Self Adjoint Operators
CONTENTS
Objectives
Introduction
28.2 Summary
28.3 Keywords
Objectives
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.
Tx, y x,T * y x, y H .
Tx, y x,Ty x, y H.
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*,
T*T* * T * T.
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.
A1 ,A 2 S A *1 A1 and A 2* A2 ...(1)
For , R , we have
A1 A2 * A1 * A2 *
A *1 A *2
A 1 ,A 2 S A1 A2 S.
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)
Then A *1 A 1 ,A *2 A2 .
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.
A 1A 2 * A 1A 2
A *2 A *1 A1A 2
A1 ,A 2 commute
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 O x, y H
Tx, x 0 x in H T O.
Tx, x Ox, x 0, x 0.
T x y , x y Tx Ty, x y
Tx, x y Ty , x y
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)
L.H.S. of (1) is zero, consequently the R.H.S. of (1) is also zero. Thus we have
2i Tx, y 0 x, y H
Tx, y 0 x, y H
Tx 0 x, y H
Tx, Tx x, T * x x, Tx Tx, x
Conversely, let Tx, x is real x H. We claim that T is self adjoint i.e. T*=T.
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.
Ax, y Ay,x x, y H.
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:
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.
Ax, x Ax, x x H
Ax, x Ax, x x H
By definition A A.
A 1 x, x A 2 x, x x H.
and A 2 x, x A 3 x, x x H.
A 1 x, x A 3 x, x x H.
and A 2 x, x A 3 x, x x H.
A 1 x, x A 3 x, x x H.
Also A 2 A1 A 2 x, x A 1 x, x x H.
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
A1 A 2 x,x A1 A x,x x H.
A1 A 2 A 2 A, by def. of .
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
if Ax, x 0 x H.
(i) Identity operator I and the zero operator 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
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
I+T is onto.
Let M = range of I+T. Then I+T will be onto if we prove that M=H.
x Tx,x Tx
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,
xm xn 0
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
M is closed.
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
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
if Ax, x 0 x H.
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 ,
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
Notes
Unit 29: Normal and Unitary Operators
CONTENTS
Objectives
Introduction
29.2 Summary
29.3 Keywords
Objectives
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.
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.
Solution: Since I is an adjoint operator and the adjoint operation is conjugate linear,
T is a normal operator on H.
T is normal TT*=T*T
T* is normal if T H.
Proof: Now T *k T * Tk T * Tk T
T *k T * as k since Tk T as k .
T *k Tk TT *
TT * T * T TT * Tk T k* T *k Tk TT * ...(1)
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.
T n* T* Tn T* Tn T 0.
T n* T* 0 T n* 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
Thus, TT * T * T 0 TT * T * T 0
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.
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 * .
T T * T * T
T is normal.
This completes the proof of the theorem.
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.
= N*1 N* 2 N 1 N 2
N1 N2 N1 N2 * N1 N2 * N1 N2
N1 N2 is normal.
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.
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 .
Tx T*x x ...(1)
NNx N * Nx x
N2 x N * Nx x ...(2)
Now N2 Sup N 2 x : x 1
N*N
2
N
T T* 1
Proof: Let T1 and T2 T T*
2 2i
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.
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
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.
= Ti* + i T2*
= Ti* – iT2*
= T1 – iT2
2T2T1 = 2T1T2
We see that
= T* – I*
= T* – I.
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.
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.
(iii) Tx x x H.
(ii) (iii)
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
For example: let e 1 ,e 2 ,...,e n ,... be an orthonormal basis for a separable Hilbert space H and
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.
Further TT* = I.
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
1 1
T*T T IT
1 1
T * TT T
1
T*I T
Notes
Tx x x I2 .
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.
29.3 Keywords
1. If T is an operator on a Hilbert space H, then T is normal its real and imaginary part
commute.
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.
Books Arch W. Naylor, R Sell George, Linear Operator Theory in Engineering and Sciences,
New York Springer, (1982).
Notes
Unit 30: Projections
CONTENTS
Objectives
Introduction
30.1 Projections
30.1.2 Invariance
30.2 Summary
30.3 Keywords
Objectives
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
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
x, x y
x, x x, y
2
x
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.
z0 H z0 0, a contradiction.
Hence N M
x, y Px, y
x,P * y x,Py
x,0 0
M N.
This completes the proof of the theorem.
x M Px x Px x.
Px x x M.
Conversely, let x M. Then to show Px = 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.
x y z, y M,z M .
But H M M .
Hence x M Px x.
If Px = x then obviously Px x .
I P is the projection on M .
I P x in M .
2 2 2
Px I P x Px I P x ...(2)
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
This completes the proof of the theorem.
(iii) Px x x H.
(iv) P 1.
Let M = range of P.
Px,x PPx,x
2
Px,P * x Px,Px Px 0
Px, x 0 x H.
I –P 0. (by (i))
P I
But P 0 , consequently 0 P 1.
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.
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.
Since PQ is a projection on H.
(PQ)* = PQ
Q* P* = PQ
QP = PQ ( Q* = Q, P* = P)
= PQ2P = PQP
= QPP = QP2
= QP = 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 x M N x M, x N we have
=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 (PQ) x = x
P [(PQ) x] = Px
[P (PQ)] x = Px
(P2Q) x = Px
(PQ) x = Px
But (PQ) x = x.
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.
x M N
R(PQ) M N
Hence R(PQ) = M N.
= P* P [ P* = P]
P is normal.
Py = 0 P*y = 0.
P*x = P*Px x H
P* = P*P
Also P2 = P.
Hence P is a projection on H.
30.1.2 Invariance
Theorem 4: A closed linear subspace M of a Hilbert space H is invariant under the operation
T 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.
(Tx,y) = 0
(x,Ty) = 0
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.
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.
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
Conversely, let PTP = TP. Let x M. Since P is a projection with range M and x M , therefore
Px = x
TPx = Tx
Thus x M Tx M
M is invariant under T.
TP PTP and T * P PT * P
TP PTP and T * P * PT * P *
TP PTP and P * T * * P * T * *P *
Now suppose M reduces T. Then from (1), TP = PTP and PT = PTP. This gives TP = PT.
Conversely, let TP = PT
or PTP = PT P2 P
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.
PQ O PQ * O * Q * P* O *
QP O.
M N PQ O.
so y M .Consequently N M .
Therefore P(Qz) = O.
Thus PQz = O z H
PQ = O
Qy = y
Notes
= (x,PQy) P* P
= (x,Oy) PQ O
= (x,O) = 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).
Let P = P1 + P2+ ... + Pn. Then P* = (P1 + P2+ ... + Pn)* = P1* ... Pn*
= P1 + P2 + ... + Pn = P.
Sufficient Condition:
P is a projection on H. We have
= 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*.
= (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 M j ,if 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.
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.
2 2 Notes
Pi x xi Pi x xi
Px i xi
xi the range of P.
x1 x2 ... x n R P .
x R P .
Then x M x R P
M R P ...(3)
Px = x
P1 P2 ... Pn x x
P1x P2 x ... Pn x x
M = R(P)
30.2 Summary
30.3 Keywords
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;
(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
Planetmath.org/....OrthogonalProjections OntoHilbertSubspaces.html.
CONTENTS
Objectives
Introduction
31.2 Summary
31.3 Keywords
Objectives
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.
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
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
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.
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].
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.
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.
det ([T]) 0.
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
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.
0 x 0
Hence T ( x) = T(x) = ( x)
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
By definition x M Tx = x … (1)
M = {x H : Tx = x} = {x H : (T – I) x = 0}
M is invariant under T.
Definition: Eigenspace
From property (3), we have proved that each eigenspace of T is a non-zero linear subspace of H
invariant under T.
y1 = 0, y2 = y1, ……
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.
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.
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
T2 ( 11
+ 22
)T+( 11 22
– 12 21
)I=0
Sol:
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}.
11 12
=0
21 22
2
–( 11
+ 22
) +( 11 22
– 12 21
)=0 … (2)
T = I … (3)
T2 – ( 11
+ 22
)T+( 11 22
– 12 21
)I=0 … (4)
Theorem 2: If T is an operator on a finite dimension Hilbert space, then the following statements
are true.
Proof: Notes
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).
= A (T – I) A–1
= det (T – I)
is an eigenvalue of T det (T – I) = 0.
(ATA–1) = (T).
(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.
xj Mi for j i.
xj Mi Pixj = xi
Pix = xi … (3)
Ix = x1 + x2 + … + xm … (by (1))
= (P1 + P2 + … + Pm) x x H.
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.
T* = ( 1P1 + 2
P2 + …… + m
Pm)*
= 1 P *1 2 P *2 m P *m
= 1 P1 2 P2 m Pm .
= | 1|2P1 + | 2|2P2 + …… + | m
|2Pm … (by (6))
(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
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)
= (P1 + P2 + …… + Pm) T
= PT
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.
T= 1
p1 + 2
p2 + … + m
pm
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
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.
( 1
p1 + 2
p2 + … + m
pm) x = (p1 + p2 + … + pm) x
{( 1
– ) p1 + ( 2
– ) p2 + … + ( m
– ) pm} x = 0 … (2)
( i – ) pix = 0 for i = 1, 2, …, m.
If i
for each i, pix = 0 for each i. Notes
(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.
Let T = 1
Q1 + 2
Q2 + … + Qn
n
… (3)
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.
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
pi = Qi for each i.
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.
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.
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).
S = {x N: x 1} then S is bounded.
Theorem: A set A in a normed linear space N is relatively compact every sequence of points
in A contains a convergent sub sequence.
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)
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
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 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.
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.
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.
1. If T (H) is a self-adjoint operator, then (T) = {m, M} where m, M are spectral values.
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.