Functional Analysis

NPTEL

Functional analysis is abstract brach of mathematics.

Consider a set (X, +, ∗), X ∈ Rn it satisfies certain properties (with addition and multiplication defined)
then this set is Vector/Linear space.
In general requirements are
1. we can add two elements
2. We can multiply two elements
3. We can multiply with a real number.
If some metric is defined on this space like (X, d), X ∈ Rn , d ∈ R, then it is called as Metric space.
1. |X| ≥ 0, |X| = 0 iff X = 0
2. |αX| ≤ α|X|
3. |X + Y| ≤ |X| + |Y|
4. If |X – Y| = 0 ⇒ X = Y
5. |X – Z| ≤ |X – Y| + |Y – Z|
Triangular inequality sum of the two sides of the triangle can never be less than third side.

If metric is length in 3D case it is Eucledian space.

If lenght generalized for a given vector space it is Normed/Banach space.
If inner product defined on the vector space it is Inner product/Hilbert space.

Acutually Hilbert space is used for functions and sequences.

Completeness means there are no voids in it.
Here set of elements, elements may be real numbers, complex numbers, functions, bounded
functions, sequences

Metric Space: A pair (X, d) is called metric space,if X is a set of elements and d is a metric defined
on set X as d : X ∗ X → R+ and if it follows certain conditions as, for x, y, z ∈ X,
1. d(x, y) ≥ 0
2. If d(x, y) = 0 ⇒ x = y
3. d(x, y) = d(y, x)
4. d(x, y) ≤ d(x, z) + d(y, z)
Above conditions are called as axioms of metric space.

1
 Triangular inequality can be generalized upto n– terms as

d(x, y) = d(x, x1 ) + d(x1 , x2 ) + .... + d(xn , y)

Eg.1 Real line X = R with metric d(x, y) = |x – y| where x, y ∈ R

p
Eg.2 Euclidean metric spcae X = R2 with metric d(x, y) = (x1 – y1 )2 + (x2 – y2 )2 , where (x1 , x2 ), (y1 , y2 ) ∈
R2
Here we need Minkowski’s inequality to prove traingular inequality.
p
Eg.3 Eucledian space Rn with metric d(x, y) = (x1 – y1 )2 + ... + (xn – yn )2 , where x, y are n– element
tuples.
p
Eg.4 Unitary space Cn (for n = 1 it is a set of complex numbers) with metric d(x, y) = |x1 – y1 |2 + ... + |xn – yn |2 ,
where each xi , yi ∈ C and x, y ∈ Cn .
Does the Eucledian metric is only thing possible on R2 , lets take

d1 (x, y) = |x1 – y1 | + |x2 – y2 |

This also forms (R2 , d1 ) metric space.

d1 (x, y) = |x1 – y1 | + |x2 – y2 |,

= |x1 – z1 + z1 – y1 | + |x2 – z2 + z2 – y2 |,
≤ |x1 – z1 | + |z1 – y1 | + |x2 – z2 | + |z2 – y2 |,
≤ |x1 – z1 | + |x2 – z2 | + |z1 – y1 | + |z2 – y2 |,
≤ d1 (x, z) + d1 (z, y)

Eg.5 Sequence Space (l∞ , d): let X be the set of all bounded sequences of real or complex numbers,

X = {x = (x1 , ..., xn , ..); xi ∈ R or C}

It is bounded then |xj | ≤ Mx ∀ j

We define metric as (x, y ∈ l∞ )
d(x, y) = sup |xi – yi |
i

d(x, y) = sup |xi – yi |,

i
= sup |xi – zi + zi – yi |,
i
≤ sup |xi – zi | + sup |zi – yi |,
i i
≤ d(x, z) + d(z, y).

Eg.6 Function space C[a, b], let X be the set of all real valued (X : R → R) functions (f(t), g(t), h(t), ...)
which are function of independent real variable t ∈ R and well defined and continuous on given interaval
[a, b].

2
 Here metric defined as
d(f, g) = max |f(t) – g(t)|
t∈[a,b]

d(f, g) = max |f(t) – g(t)|,

t∈[a,b]
= max |f(t) – h(t) + h(t) – g(t)|,
t∈[a,b]
≤ max |f(t) – h(t)| + max |h(t) – g(t)|,
t∈[a,b] t∈[a,b]
≤ d(f, h) + d(h, g).

Another metric defined as area difference between two curves

Z b
d(f, g) = |f(t) – g(t)|dt
a

This function space doamin can be generalized to a set (Rk ) instead of the closed interval [a, b]
Eg.7 Sequence Space let X be set of all bounded and unbounded sequences of complex numbers (X : R →
C),with the metric
∞
X 1 |xi – yi |
d(x, y) =
2i 1 + |xi – yi |
i=1

x is always (≤ 1) metric is dominated by 1 which is a convergence series.

As 1+x 2i

t 1 t 1
f(t) = , ḟ(t) = – = ≥0
1+t 1 + t (1 + t)2 (1 + t)2

f(t) is always increading fucntion, hence

|a + b| ≤ |a| + |b|,
f(|a + b|) ≤ f(|a| + |b|),
|a + b| |a| + |b|
≤ ,
1 + |a + b| 1 + |a| + |b|
|a| |b| |a| + 2|a||b| + |b|
≤ + = ,
1 + |a| 1 + |b| 1 + |a| + |b| + |a||b|

Lets take a = xi – zi and b = zi – yi

|xi – zi + zi – yi | |xi – zi | |zi – yi |

≤ + ,
1 + |xi – zi + zi – yi | 1 + |xi – zi | 1 + |zi – yi |
∞ ∞ ∞
X 1 |xi – yi | X 1 |xi – zi | X 1 |zi – yi |
≤ + ,
2i 1 + |xi – yi | 2i 1 + |xi – zi | 2i 1 + |zi – yi |
i=1 i=1 i=1
d(x, y) ≤ d(x, z) + d(z, y).

Anotehr metric defiend as, replace 2i → μi

∞
X 1 |xi – yi |
d(x, y) =
μi 1 + |xi – yi |
i=1

3
 Eg.8 Discrete metric space: let X be set of points with metric

0 if x = y
d(x, y) =
1 if x 6= y

Eg.9 lp space: let X be a set of infinite sequences (x1 , x2 , ....xn ...) with a condition ∞ p
P
i=1 |xi | < ∞.

If xi ’s are real it is Real lp space

If xi ’s are complex it is Complex lp space
For this mketric defiend as
"∞ #1
X p
p
d(x, y) = |xi – yi | , p≥1
i=1

Conjugate Exponents:
If α and β are real positive numbers then if p ≥ 1 then define q which satisfies p1 + q1 = 1, then

αp βq
αβ ≤ +
p q

1 1
+ = 1,
p q
q+p
= 1,
pq
p + q – pq = 0,
q
p(1 – q) + q = 0 → p = ,
q–1
p
p + q(1 – p) = 0 → q = ,
p–1
pq 1
pq = →q–1= .
(p – 1)(q – 1) p–1

p, q are called as conjugate exponents.

1
Proof: Consdier a function y = tp–1 ⇒ t = y p–1 = yq–1
Consider (α, β) as ordered pair and (αβ) is area of that box, now we can write areas relation as
Z α Z β
p–1
αβ ≤ t dt + yq–1 dy,
0 0
tp α yq β
 
≤ + ,
p 0 q 0
αp βq
≤ + .
p q

Holder’s Inequality: Lets define sequences as

∞
X ∞
X
|x̄i |p = 1, |ȳi |q = 1
i=1 i=1

4
 Lets take α = |x̄i | and β = |ȳi | then
αp βq
αβ ≤ + ,
p q
|x̄ |p |ȳ |q
|x̄i ||ȳi | ≤ i + i ,
p q
∞ ∞ p ∞
X X |x̄i | X |ȳi |q 1 1
|x̄i ȳi | ≤ + = + = 1.
p q p q
i=1 i=1 i=1

Now lets define actual sequences which satisfies infinite summation → 1 as

xi yi
x̄i = P 1 , ȳi = P 1
∞ ∞
[ i=1 |xi |p ] p [ i=1 |yi |q ] q

It gives Holders inequality as

∞
"∞ #1 " ∞ #1
X X p X q
p q
|x̄i ȳi | ≤ |xi | |yi |
i=1 i=1 i=1

Minkowski Inequality :Let x, y ∈ lp are infinite sequences, then

∞
!1 ∞
!1 ∞
!1
X p X p X p
p p p
|xi + yi | ≤ |xi | + |yi |
i=1 i=1 i=1

Proof: Consider zi = xi + yi
|zi |p = |xi + yi ||zi |p–1 ,
≤ |xi ||zi |1–p + |yi ||zi |p–1

To use Holders inequality z ∈ lq where (p – 1)(q – 1) = 1, now if we can show that |zi |1–p is finite then
we can continue
 q
|zi |p–1 = |zi |q(p–1) ,
 
1
q(p – 1) = + 1 (p – 1) = –1 + 1 + p = p
p–1
As lp is finite, we can use Holder’s inequality.
|zi |p ≤ |xi ||zi |1–p + |yi ||zi |p–1 ,
n
X n
X n
X
|zi |p ≤ |xi ||zi |1–p + |yi ||zi |p–1 ,
i=1 i=1 i=1
" n #1 " n #1 " n #1 " n #1
X p X q X p X q
≤ |xi |p |zi |p + |yi |p |zi |p ,
i=1 i=1 i=1 i=1
"
∞
#1 " ∞ #1 " ∞ #1
X p X p X q
p p p
≤ |xi | + |yi |  |zi | ,
i=1 i=1 i=1
"∞ #1– 1 "∞ #1 "∞ #1
X q X p X p
|zi |p ≤ |xi |p + |yi |p ,
i=1 i=1 i=1

5
 It gives final result from zi = xi + yi and p = 1 – q1

∞
!1 ∞
!1 ∞
!1
X p X p X p
|xi + yi |p ≤ |xi |p + |yi |p
i=1 i=1 i=1

Using conjugate exponents, Holder’s and Minkowski’s inequality we should prove the metric as
∞
!1
X p
p
d(x, y) = |xi – yi |
i=1

Proof: Consider |xi – yi | = |xi – zi + zi – yi |, then we can write as

∞
!1 ∞
!1
X p X p
|xi – yi |p ≤ |xi – zi + zi – yi |p
i=1 i=1
∞
!1 ∞
!1
X p X p
p p
≤ |xi – zi | + |zi – yi | ,
i=1 i=1
d(x, y) ≤ d(x, z) + d(z, y)

Subspace : Let (X, d) be a metric space and Y ⊂ X be a non-empty subset of X, suppose d̄ is the
restriction of d on Y × Y as
d̄(y1 , y2 ) = d(y1 , y2 )

Eg.1 If A is subspace of l∞ , which consists of all sequences which are 0/1 only then, Discrete metric and l∞
metric both are equivalent.

0 if x = y
d̄(x, y) = ,
1 if x 6= y
d(x, y) = sup |xi – yi |
i

Let (X, d) be a metric and x0 ∈ X and r > 0, then

Open Ball : defined as
B(x0 , r) = {x ∈ X : d(x0 , x) < r}

Closed Ball : defined as

B̃(x0 , r) = {x ∈ X : d(x0 , x) ≤ r}

Sphere : defined as
S(x0 , r) = {x ∈ X : d(x0 , x) = r}

Eg. Define open ball of continuous functions C[a, b] with d(x0 , x) < 1, let x0 is identically zero in [a, b],
then
B(x0 , 1) = {x ∈ C[a, b] : d(x0 , x) < 1}

For Eucledian Rn space, sphere never can be null set. But for general metric spaces it can be null set.
Eg. Consider discrete metric space

S(x0 , 1) = {x ∈ X : d(x0 , x) = 1} =
6 φ

B(x0 , 1) = {x ∈ X : d(x0 , x) = 1} = φ

6
 A point x0 ∈ D ⊂ X is said to be an interior point we can form a open ball around x0 with in D.
Set of all interior points defined as D0 0r int(D)
If every point of D can form an open ball then it is Open Set.

A point x0 ∈ X(observe not subset D) is said to be an boundary point we can form a open ball
around x0 which has intersection with both D and its compliment D0 .
Set of boundary points is called as boundary of D(∂D).
If set D contains boundary points also in it, then it is Closed Set.

ε– Nbd:An open ball B(x0 , ε) is called an ε neighbourhood of x0 .

By nbd of x0 means any subset of X, which contains x0 in it.
Set changes its category (open→close) (close→open)based on definition we use for norm.
Theorem: Every open ball is an open set.

Proof: Consider a point in an open ball z ∈ B(x0 , r)

Define an open ball around z as B(z, r1 ), Let r1 = r – d(x0 , z), consider a third point y ∈ B(z, r1 )

d(x0 , y) ≤ d(x0 , z) + d(z, y),

≤ r – r 1 + r1 ,
≤ r.

Theorem : Every closed ball is a closed set.

Def : Let (X, d) be a metric space then
– φ and X are open sets
– Finite or infinite unions of open sets (countable or uncountable) is an open set
– Finite intersection of the open sets is an open set
Def : Let (X, d) be a metric space then
– φ and X are closed sets sets
– Finite or infinite intersection of closed sets (countable or uncountable) is a closed set
– Finite union of the closed sets is a closed set
Limit point/Accummulation poinjt : Let (X, d) be a metric space and M ⊂ X. A point x0 is
(may in M, may not in M) is said to be a limit point if evary neighbourhood of x0 atleast contain one
point on M other than x0 .

Boundary point has extra condition that it should have M0 elements also.
Dense Set : Let (X, d) be a metric space and D ⊂ X ,D called as dense set if its closure(D) is X it
self.
Closure: Closure of a set means union of set and all its limit points.
Eg. X–Real line, D–set of rational numbers , then D = R

7
 Continuous functions : let (X, d) and (Y, d̃) be two different metric spaces, a mapping T : X → Y
called continuous if
∃ ε – nbd of Tx0 ∈ Y for δ – nbd of x0 ∈ X
d(x0 , x) < ε; d̃(Tx0 , Tx) < δ

Other definition : Inverse image of an open set ∈ Y is also an open set ∈ X.

Convergence :Let (X, d) be sequence metric space, we say xn converges to x ∈ X if

lim d(xn , x) = 0
n→∞

x sholud be inside X,other wise we say it does not converge.

Eg. Consider X = (0, 1] sequence n1 , as n → ∞ sequence goes to 0, but 0 ∈
/ X, so it does not converge.

Semi-metric space :if d(x, y) = 0 and x 6= y, then this is semi-metric space.

Equivalence Relation :Let R be relation between xRy, then if its
– Reflexive : xRx
– Symmetric : xRy = yRx
– Transitive : xRy, yRz given xRz can be found
To make semi metric space (X, d) as metric space, define an equivalence relation as

xRy ⇒ d(x, y) = 0

It divides semi-metric space space into number of disjoint (Ex , Ey ....) sets w.r.t. any point x ∈ X

Here
d(x, x0 ) = 0, d(y, y 0 ) = 0

Union of all these disjoint sets forms a metric space with new metric defined as

d̃(Ex , Ey ) = d(x, y)

Lebesgue integrable function,if it satisfies given condition as L[a, b]

Z b
|f(x)|dx < ∞
a

Eg. L[0, 1] is semi-metric space as d(f, g) = 0 results even for f 6= g.

Separable Metric Space :A metric space (X, d) is said to be separable is it has countable subsets
and dense w.r.t. X.
Countable means for every natural number we can assign a sunset.

Eg. R, C are separable metric spaces.