Linear Functional Analysis

by
Alex Hudyma

A project submitted to the Department of

Mathematical Sciences in conformity with the requirements
for Math 4301 (Honour’s Seminar)

Lakehead University
Thunder Bay, Ontario, Canada
copyright c (2015) Alex Hudyma
 Abstract

This project is designed to give readers a basic understanding on the topic of Linear
Functional Anaylsis. After covering preliminaries, we will examine different spaces and
their properties, bounded linear operators, duality, and finish with the Hahn-Banach
Theorem - a very powerful theorem that is a cornerstone of functional analysis and has
applications in other areas of mathematics as well.

i
 Acknowledgements

I would like to thank my supervisor, Monica Ilie, for her guidance with this project,
and the coordinator of Math 4301, Razvan Anisca. I would also like to thank the students
and faculty of the Lakehead University Department of Mathematical Sciences for a great
four years.

ii
 Contents

Abstract i
Acknowledgements ii
Chapter 1. Introduction 1
Chapter 2. Preliminaries 2
1. Linear Algebra 2
2. Metric Spaces 3
3. Lebesgue Integration 6
Chapter 3. Spaces 9
1. Normed Spaces 9
2. Banach Spaces 13
3. Inner Product Spaces 13
4. Hilbert Spaces 15
Chapter 4. Operators 17
1. Continuous Linear Transformations 17
2. The Space B(X, Y ) 18
3. Isometries, Isomorphisms, and Inverses 21
Chapter 5. Duality 22
1. Dual Spaces 22
2. Sublinear Functionals and Seminorms 23
3. The Hahn-Banach Theorem 25
Chapter 6. Concluding Remarks 28
Bibliography 29

iii
 CHAPTER 1

Introduction

This project will introduce the methods of linear functional analysis. Our basic goal
here is to perform analysis on infinite-dimension vector spaces. Because not all of our
linear algebra properties hold for infinite spaces, we extend ideas like the length of a
vector into an abstract setting: the norm. Much like in real and complex analysis, we will
give additional structures to certain spaces to yield special properties. Through different
examples, we will gain an understanding of normed and inner product spaces, bounded
linear operators, duality, and the Hahn-Banach theorem.

1
 CHAPTER 2

Preliminaries

1. Linear Algebra

Here, we will recall some essential concepts from Linear Algebra for use throughout
the project. We will be working with the sets R, C, and N. Note that R and C are both
algebraic fields, and when giving examples, theorems, or definitions that apply equally to
R and C, we will simply use F to denote either set.

Definition 2.1. A vector space over F is a non-empty set V with two operations,
vector addition and scalar multiplication, where vector addition is a function from V ×
V → V defined by (x, y) = x + y, and scalar multiplication is a function from F × V
→ V defined by (α, x) = αx. The following properties hold for any α, β ∈ F and any
x, y, z ∈ V .
(i) x + y = y + x, x + (y + z) = (x + y) + z;
(ii) there exists a unique 0 ∈ V such that for all x ∈ V , x + 0 = 0 + x = x;
(iii) for all x ∈ V , there exists a unique −x ∈ V such that x + (−x) = (−x) + x = 0;
(iv) 1x = x, α(βx) = (αβ)x;
(v) α(x + y) = αx + αy, (α + β)x = αx + βx.

Note that if F = R, then V is a real vector space, and if F = C, then V is a complex

vector space. For our purposes, we will usually just use the term “vector space,” as most
results about vector spaces will apply equally well to both the real and complex case.

Definition 2.2. Let V be a vector space, and U a non-empty subset of V . Then U

is a linear subspace of V if U is a vector space under the same vector addition and scalar
multiplication as V .

Rather than proving Definition 2.1 (i)-(v) for U , we can use the subspace test:
Let U be a non-empty subset of a vector space V . For all α ∈ F and x, y ∈ U , if
(i) x + y ∈ U ;
(ii) αx ∈ U
then U is a subspace of V . Note that the set {0} ⊂ V is a linear subspace.

Definition 2.3. Let V be a vector space, v = {v1 , v2 , . . . , vk } ⊂ V, k ≥ 1 be a finite

set, and A be an arbitrary non-empty subset of V .
2
Chapter 2. Preliminaries 3
(i) A linear combination of v is any vector x of the form:
x = α1 v1 + α2 v2 + · · · + αk vk ∈ V,
for any scalars α1 , α2 , . . . , αk .
(ii) The set v is linearly independent if:
α1 v1 + α2 v2 + · · · + αk vk = 0 ⇒ α1 = α2 = · · · = αk = 0.
(iii) The span of A (denoted SpA) is the set of all linear combinations of all finite
subsets of A. In other words, SpA is the intersection of all linear subspaces of V
containing A. By this definition, SpA is the smallest linear subspace of V that
still contains A.
(iv) Let v be linearly independent, and Spv = V . Then v is a basis of V . When V
has such a finite basis with k elements, every other basis of V will also have k
elements. We say that V is k-dimensional and write dimV = k.
(v) The set Fk is a vector space over F, and the set of vectors
ê1 = (1, 0, 0, . . . , 0), ê2 = (0, 1, 0, . . . , 0), . . . , êk = (0, 0, 0, . . . , 1)
forms a basis for Fk , known as the standard basis.
Definition 2.4. Let V and W be vector spaces over F. Then the Cartesian product
V × W is a vector space, where vector addition and scalar multiplication are defined as
follows: For any α ∈ F and (x1 , y1 ), (x2 , y2 ) ∈ V × W ,
(x1 , y1 ) + (x2 , y2 ) = (x1 + x2 , y1 + y2 ) and α(x1 , y1 ) = (αx1 , αy1 ).
Definition 2.5. Let V and W be vector spaces over F. Then a function T : V → W
is a linear transformation if, for all α, β ∈ F and x, y ∈ V ,
T (αx + βy) = αT (x) + βT (y).

We define a set of all linear transformations T : V → W , denoted by L(V, W ), which

is also a vector space. If V = W , then we will shorten L(V, V ) to just L(V ). A notable
element of L(V ) is the identity transformation on V , defined by IV (x) = x for x ∈ V .

2. Metric Spaces

Metric spaces are an abstract setting for the discussion of concepts from analysis
(such as convergence and continuity). The basic tool used here is the notion of a distance
function, or a metric, which we will define in this section.
Definition 2.6. A metric on a set M is a function d : M × M → R such that for all
x, y, z ∈ M :
(i) d(x, y) ≥ 0;
(ii) d(x, y) = 0 ⇔ x = y;
(iii) d(x, y) = d(y, x);
(iv) d(x, z) ≤ d(x, y) + d(y, z) (the triangle inequality).
Chapter 2. Preliminaries 4
If d is a metric on M , then we called the pair (M, d) a metric space. Unless the set M
consists of a single point, it can have more than one metric, and we often say “the metric
space M ” rather than (M, d).
Example 2.7. For an integer k ≥ 1, the function d : Fk × Fk → R defined by
k
!1/2
X
2
d(x, y) = |xj − yj | ,
j=1

is the standard metric on the set Fk . Thus we call Fk a metric space.

Definition 2.8. A sequence {xn } in a metric space M is convergent if for some x ∈ M
and every  > 0, there exists an N ∈ N such that
d(xn , x) <  for all n ≥ N.
If this is true, we write lim xn = x, or xn → x.
n→∞

Definition 2.9. A sequence {xn } in a metric space M is a Cauchy sequence if for

every  > 0, there exists an N ∈ N such that
d(xm , xn ) <  for all m, n ≥ N.
Definition 2.10. Let (M, d) be a metric space and A ⊂ M .
(i) A is bounded if there exists a number B > 0 such that d(x, y) < B for all x, y ∈ A.
(ii) A is open if for all x ∈ A, there exists an  > 0 such that Bx () = {y ∈ M :
d(x, y) < } ⊂ A. We call Bx () the open ball with radius  centered at x.
(iii) A is closed if the set M \ A is open.
(iv) A point x ∈ M is a closure point of A if for every  > 0, there exists a y ∈ A
with d(x, y) <  (we can also say, if there exists an {yn } ⊂ A such that yn → x).
(v) The closure of A, denoted Ā, is the set of all closure points in A.
(vi) A is dense in M if Ā = M .

Note that if x ∈ A, then x is a closure point of A (for every  > 0, d(x, x) = 0 < ),
hence A ⊂ Ā. However, the converse is not necessarily true, as we see in the following
theorem.
Theorem 2.11. Let (M, d) be a metric space and A ⊂ M .
(i) Ā is closed and is the smallest closed set that contains A.
(ii) A is closed, if and only if A = Ā.
(iii) A is closed if and only if the following implication holds:
if {xn } is a sequence in A and x ∈ M such that xn → x, then x ∈ A.
(iv) x ∈ Ā if and only if inf{d(x, y) : y ∈ A} = 0.
(v) A is dense if and only if for any element x ∈ M , there exists a sequence {yn } in
A such that yn → x.
Chapter 2. Preliminaries 5
We can extend the idea of a continuous function from real analysis to describe general
metric spaces.

Definition 2.12. Let (M, dM ) and (N, dN ) be metric spaces, and f : M → N .

(i) f is continuous at a point x ∈ M if for all  > 0, there exists a δ > 0 such that,
for all y ∈ M ,

dM (x, y) < δ ⇒ dN (f (x), f (y)) < .

(ii) f is continuous on M if it is continuous at each point of M .

(iii) f is uniformly continuous on M if for all  > 0, there exists a δ > 0 such that,
for all x, y ∈ M ,

dM (x, y) < δ ⇒ dN (f (x), f (y)) < .

Definition 2.13. A metric space (M, d) is complete if every Cauchy sequence in

(M, d) is convergent. Furthermore, a set A ⊂ M is complete in (M, d) if every Cauchy
sequence in A converges to an element of A.

Example 2.14. For k ≥ 1, the space Fk with the standard metric is complete.

Definition 2.15. Let (M, d) be a metric space. A set A ⊂ M is compact if every

sequence in A contains a subsequence converging to an element of A. A set A ⊂ M is
relatively compact if Ā is compact.

Theorem 2.16. Let (M, d) be a metric space and A ⊂ M . Then:

(i) if A is complete, then A is closed;
(ii) if M is complete, then A is complete if and only if A is closed;
(iii) if A is compact, then A is closed and bounded;
(iv) every closed and bounded subset of Fk is compact (Bolzano-Weirstrass theorem).

Theorem 2.17. Let (M, d) be a compact metric space and f : M → F be continuous.

Then f is bounded. Particularly, if F = R, then the supremum and infimum of f (x) exist
and are finite, and f attains the values inf{f (x) : x ∈ M } and sup{f (x) : x ∈ M }.

Definition 2.18. Let (M, d) be a compact metric space. Then CF (M ) denotes the
set of continuous functions f : M → F. We will define the uniform metric on CF (M ) by

d(f, g) = sup{|f (x) − g(x)| : x ∈ M }.

Note that most properties of CF (M ) hold for both real and complex cases, so generally,
we will omit the subscript and just write C(M ). Furthermore, if M is a bounded interval
[a, b] ⊂ R, we will write C[a, b]. Also note that (C(M ), d) is a complete metric space.

Theorem 2.19. The metric space C(M) is complete.

Chapter 2. Preliminaries 6
3. Lebesgue Integration

So far we have seen only the uniform metric on C[a, b], but there are other metrics
Rb
that are defined by integrals. If a f (x)dx is the usual Riemann integral of f ∈ C[a, b],
then for 1 ≤ p < ∞, the function dp : C[a, b] × C[a, b] → R defined by
Z b 1/p
p
dp (f, g) = |f (x) − g(x)| dx
a

is a metric on C[a, b]. However, (C[a, b], dp ) is not a complete metric space, due to
some drawbacks of the Riemann integral. Thus, we will introduce here the Lebesgue
Integral, which is more powerful and works on a wider class of functions. We will look at
metric spaces which are complete using the metric dp , and will introduce the notion of a
“measure” of a set (AKA the “length;” for example, the bounded interval [a, b] has length
b − a). Note that many sets such as R have “infinite measure,” and so we introduce the
+
extended sets of real numbers: R = R ∪ {−∞, ∞} and R = [0, ∞) ∪ {∞}. The standard
algebraic operations apply, with the products 0 · ∞ = 0 · (−∞) = 0, and the operations
∞ − ∞ and ∞/∞ forbidden (like dividing by zero in R).
Definition 2.20. A σ-algebra is a collection, Σ, of subsets of X satisfying:
(i) ∅, X ∈ Σ;
(ii) S ∈ Σ ⇒ X \ S ∈ Σ;
∞
S
(iii) Sn ∈ Σ, n = 1, 2, . . . , ⇒ Sn ∈ Σ.
n=1

A set S ∈ Σ is said to be measurable.

Definition 2.21. Let X be a set and Σ be a σ-algebra of subsets of X. A function
+
µ : Σ → R is a measure if it has the following properties:
(i) µ(∅) = 0;
(ii) µ is countably additive. In other words, if Sj ∈ Σ, j = 1, 2, . . . , are pairwise
disjoint sets, then
∞
! ∞
[ X
µ Sj = µ(Sj ).
j=1 j=1

The triple (X, Σ, µ) is called a measure space.

Definition 2.22. Let (X, Σ, µ) be a measure space. Then a set S ∈ Σ with µ(S) = 0
is said to have measure zero. Also, a given property P (x) is said to hold almost everywhere
if the set {x : P (x) is false} has measure zero.
Definition 2.23. Let X = N, Σc be the class of all subsets of N, and for any S ⊂ N,
define µc (S) to be the number of elements of S. Then Σc is a σ-algebra and µc is a
measure on Σc called the counting measure on N. Note that the only set of measure zero
in (N, Σc , µc ) is the empty set.
Chapter 2. Preliminaries 7
Definition 2.24. Let X = R, ΣL be a σ-algebra, and µL be a measure on ΣL , such
that any finite interval I = [a, b] ∈ ΣL and µL (I) = `(I). Then ΣL is called the Lebesgue
measure and the sets in ΣL are Lebesgue measurable.

Note that the sets of measure zero in (R, ΣL , µL ) are the sets A with the following
property: for any  > 0, there exists a sequence of intervals Ij ⊂ R, j = 1, 2, . . . , such that
∞
[ X∞
A⊂ Ij and `(Ij ) < .
j=1 j=1

Definition 2.25. A function φ : X → R is simple if it has the form

k
X
φ= αj χSj ,
j=1

for some k ∈ N where αj ∈ R and Sj ∈ Σ, j = 1, . . . , k. If φ is non-negative and simple,

then the integral of φ over X with respect to µ is defined by
Z Xk
φdµ = αj µ(Sj ).
X j=1
+
We do allow µ(Sj ) = ∞ here, and use the algebraic rules of R from the beginning of
this section to evaluate (since φ is non-negative, we will not encounter any problems like
∞ − ∞). The value of the integral may be ∞.
Definition 2.26. Let (X, Σ, µ) be a measure space. Then a function f : X → R is
measurable if, for every α ∈ R,
{x ∈ X : f (x) > α} ∈ Σ.
Definition 2.27. If f is non-negative and measurable then the integral of f is defined
to be Z Z 
f dµ = sup φdµ : φ is simple and 0 ≤ φ ≤ f .
X X
R
Definition 2.28. If f is measurable and X |f |dµ < ∞, then f is integrable, and the
integral of f is defined to be
Z Z Z
f dµ = +
f dµ − f − dµ,
X X X
where f ± (x) = max {±f (x), 0}. The set of F-valued integrable functions on X will be
denoted by L1 (X)

Now we would like to define a metric on L1 (X), and the obvious choice is
Z
d1 (f, g) = |f − g|dµ,
X
1
for all f, g ∈ L (X). However, this doesn’t fit criteria (ii) of Definition 2.6 for being a
metric. Unfortunately, L1 (X) has functions f, g such that f = g almost everywhere, but
Chapter 2. Preliminaries 8
f 6= g (which may happen on a set of measure zero). Hence we may have d1 (f, g) = 0,
but f 6= g. So, we will regard “equivalent” any two functions f, g ∈ L1 (X) which are
equal almost everywhere. We will define an equivalence relation ≡ on L1 (X) by
f ≡ g ⇐⇒ f (x) = g(x) almost everywhere for all x ∈ X.
This partitions L1 (X) into a space of equivalence classes, which we will denote by L1 (X).
The space L1 (X) is a vector space and we see that d1 (f, g) = 0 if and only if f ≡ g. Thus
d1 meets all criteria to be a metric for L1 (X), which we will use from now on instead of
L1 (X). We should also note that when we say the “function” f ∈ L1 (X), we mean some
representative of the appropriate equivalence class.
Example 2.29. Take the counting measure (N, Σc , µc ). Any function f : N → F can
be seen as an F-valued sequence {an } (with an = f (n) for n ≥ 1) such that every sequence
P∞ function. It follows, then, that {an } is integrable with respect
is a measurable
1
to µc if
and only if n=1 |an | < ∞. The space of these sequences will be denoted by ` .
Definition 2.30. Let (X, Σ, µ) = (Rk , ΣL , µL ) for some k ≥ 1. If f ∈ L1 (X), then f
is Lebesgue integrable.
Theorem 2.31. If I = [a, b] is a bounded interval and f : I → R is bounded and
Riemann integrable on I, then f is Lebesgue integrable on I, and the values of the two
integrals are the same. Particularly, continuous functions on I are Lebesgue integrable,
and we use the same notation as the Riemann integral.
Definition 2.32. We define the space Lp (X) as follows:
Z
L (X) = {f : f is measurable and ( |f |p dµ)1/p < ∞} when 1 ≤ p < ∞.
p
X

We also define the corresponding sets L (X) by identifying functions in Lp (X) which
p

are equal almost everywhere, and considering corresponding spaces of equivalence classes.
Example 2.33. For 1 ≤ p < ∞, the function dp : Lp (X) × Lp (X) → R defined by
Z 1/p
p
dp (f, g) = |f − g| dµ
X
p
is the standard metric on L (X).
Example 2.34 (Counting Measure). Let 1 ≤ p ≤ ∞ and (X, Σ, µ) = (N, Σc , µc ).
Then the space Lp (N) consists of the set of sequences {αn } in F with the properties:
∞
!1/p
X
p
|αn | < ∞ for 1 ≤ p < ∞,
n=1

sup {|αn | : n ∈ N} < ∞ for p = ∞.

These measure spaces will be denoted `p .
Theorem 2.35. Let 1 ≤ p ≤ ∞. Then the metric space Lp (X) is complete, and in
particular, the sequence space `p is complete.
 CHAPTER 3

Spaces

1. Normed Spaces

In R2 and R3 , we can easily picture the length of a vector. However, when it comes to
other, possibly infinite-dimensional vector spaces, we assign a set of axioms to describe
the length of a vector in these spaces. These axioms define the “norm” of a vector as we
see in the next definition.
Definition 3.1. Let X be a vector space over F. Then a norm on X is a function
k·k : X → R such that for all x, y ∈ X and α ∈ F,
(i) kxk ≥ 0;
(ii) kxk = 0 if and only if x = 0;
(iii) kαxk = |α|kxk;
(iv) kx + yk ≤ kxk + kyk (the triangle inequality).
Definition 3.2. If X is a vector space with a norm on it, then X is a normed vector
space (also just called a normed space).
Example 3.3. The function k·k : Fn → R defined by
n
!1/2
X
k(x1 , x2 , . . . , xn )k = |xj |2
j=1

is the standard norm on Fk .

Example 3.4. Let M be a compact metric space and recall CF (M ), the set of con-
tinuous F-valued functions defined on M . The function k·k : CF (M ) → R defined by
kf k = sup{|f (x)| : x ∈ M }
is the standard norm on CF (M ).
Definition 3.5. If X is a vector space with norm k·k, and d : X × X → R is a metric
defined by d(x, y) = kx − yk, then d is called the metric associated with the norm.
Theorem 3.6. Let X be a vector space over F with norm k·k, {xn } ∈ X and {yn } ∈ X
be sequences such that xn → x and yn → y in X, and {αn } ∈ F be a sequence such that
αn → α in F. Then we have the following properties:
(i) |kxk − kyk| ≤ kx − yk;
9
Chapter 3. Spaces 10
(ii) lim kxn k = kxk;
n→∞
(iii) lim (xn + yn ) = x + y;
n→∞
(iv) lim αn xn = αx.
n→∞

Proof.

(i) From the triangle inequality, kxk = k(x − y) + yk ≤ kx − yk + kyk. Rearranging,

we get kxk − kyk ≤ kx − yk. Similarly for y, −ky − xk ≤ kxk − kyk. However,
note that ky − xk = k(−1)(x − y)k = kx − yk. Thus we have

−kx − yk ≤ kxk − kyk ≤ kx − yk,

and so |kxk − kyk| ≤ kx − yk.

(ii) We know that xn → x, and from (i), |kxk − kxn k| ≤ kx − xn k for all n ∈ N.
So, letting n → ∞ we get |kxk − kxn k| ≤ 0, which means |kxk − kxn k| = 0 or
lim kxn k = kxk.
n→∞
(iii) We know that xn → x and yn → y. By the triangle inequality, k(xn + yn ) − (x +
y)k = k(xn − x) + (yn − y)k ≤ kxn − xk + kyn − yk. Again, letting n → ∞,
k(xn + yn ) − (x + y)k ≤ 0, hence lim (xn + yn ) = x + y.
n→∞
(iv) Since {αn } is convergent, there exists a K > 0 such that |αn | ≤ K for all
n ∈ N. Furthermore, we see that kαn xn − αxk = kαn (xn − x) + (αn − α)xk ≤
|αn |kxn − xk + |αn − α|kxk ≤ Kkxn − xk + |αn − α|kxk. Now since αn → α and
xn → x, letting n → ∞ gives kαn xn − αxk ≤ 0, hence lim αn xn = αx.
n→∞

Definition 3.7. Let X be a vector space and k·k1 and k·k2 be two different norms
on X. Then k·k1 is equivalent to k·k2 if there exists an M, m > 0 such that for all x ∈ X,

mk·k2 ≤ k·k1 ≤ M k·k2

Note that the notion “equivalent to” defines an equivalence relation on the set of all
norms on X.

Theorem 3.8. Let X be a finite-dimensional vector space with norm k·k and basis
Pn P 1/2
n 2
{e1 , e2 , . . . , en }. Define another norm on X by k j=1 λj ej k1 = j=1 |λj | . Then
the norms k·k and k·k1 are equivalent.
Chapter 3. Spaces 11
n
!1/2
X
Proof. Let M = kej k . Since {e1 , e2 , . . . , en } forms a basis for X, we have
j=1
M > 0. Additionally,
Xn n
X
k λj ej k ≤ kλj ej k
j=1 j=1
n
X
= |λj |kej k
j=1

n
!1/2 n
!1/2
X X
≤ |λj |2 kej k2
j=1 j=1
n
X
= Mk λj ej k1 .
j=1

Now define a function f : Fn → F by f (λ1 , . . . , λn ) = k nj=1 λj ej k, and take S =

P

{(λ1 , λ2 , . . . , λn ) ∈ Fn : nj=1 |λj |2 = 1}. Now S is compact, so there exists (µ1 , µ2 , . . . , µn ) ∈

P

SPsuch that m = f (µ1 , µ2 , . .P . , µn ) ≤ f (λ1 , λ2 , . . . , λn ) ∈ S. If we have m = 0, then

n n
k j=1 µj ej k = 0, meaning j=1 µj ej = 0. However, this contradicts the fact that
{e1 , e2 , . . . , en } is a basis for X, thus m > 0. Furthermore, by the construction of k·k1 , if
y y
kxk1 = 1 then kxk ≥ m. So for y ∈ X \{0} (so that = 1), we’d have ≥ m,
ky1 k ky1 k
and kyk ≥ mkyk1 . For the case y = 0 we obviously have kyk ≥ mkyk1 . Hence k·k and
k·k1 are equivalent. 
Lemma 3.9. Let X be a vector space and k·k1 , k·k2 , and k·k3 be norms on X such
that k·k2 is equivalent to k·k1 and k·k3 is equivalent to k·k2 . Then:
(i) k·k1 is equivalent to k·k2 ;
(ii) k·k3 is equivalent to k·k1 .

Proof. From the definition of equivalency, there exists an M, m > 0 such that
mk·k1 ≤ k·k2 ≤ M k·k1 , and there exists a K, k > 0 such that kk·k2 ≤ k·k3 ≤ M k·k2
for all x ∈ X. So:
(i) Letting N = M1 and n = m1 gives N, n > 0 such that N k·k2 ≤ k·k1 ≤ nk·k2 for
all x ∈ X. Thus k·k1 is equivalent to k·k2 .
(ii) Letting L = KM and l = km gives L, l > 0 such that lk·k1 ≤ k·k3 ≤ Lk·k1 for
all x ∈ X. Hence k·k3 is equivalent to k·k1 .

Corollary 3.10. If k·k and k·k2 are any two norms on a finite-dimensional vector
space X, then they are equivalent.
Chapter 3. Spaces 12
Proof. Let {e1 , e2 , . . . , en } be a basis for X, and let k·k1 be the norm defined in
Theorem 3.8. Then by Theorem 3.8, both k·k and k·k2 are equivalent to k·k1 , and by
Lemma 3.9, k·k2 is equivalent to k·k. 
Theorem 3.11. Let X be a vector space and k·k and k·k1 be equivalent norms on X
such that d and d1 are metrics defined by d(x, y) = kx − yk and d1 (x, y) = kx − yk1 . Then
(X, d) is complete if and only if (X, d1 ) is complete.

A proof of this theorem can be found in [1]. Based on this idea, when we have two
equivalent norms on the same metric space, we may use whichever is easier to work with
to show that the space is complete.
Lemma 3.12. Let X be a finite-dimensional vector space over F with basis {e1 , e2 , . . . , en }.
Pn P 1/2
n 2
If k·k1 is a norm on X defined by k j=1 λj ej k1 = j=1 |λj | , then X is a complete
metric space.

Proof. Let {xm }∞ m=1 be a Cauchy sequence

Pn in X, and let  > 0. Then each element
of the sequence can be written as xm = j=1 λj,m ej for some λj,m ∈ F. Since {xm }∞
m=1 is
Cauchy, there exists an N ∈ N such that for k, m ≥ N ,
n
X
|λj,k − λj,m |2 = kxk − xm k21 ≤ 2 .
j=1

So we get |λj,k − λj,m |2 ≤ 2 for all k, m ≥ N and 1 ≤ j ≤ n. Thus {λj,m }∞ m=1 is a Cauchy
sequence in F. Now since F is a complete metric space, there exists a λj ∈ F such that
λj,m → λj as m → ∞. By the definition of sequence limits, there exists an Nj ∈ N such
that for all m ≥ Nj ,
2
|λj,m − λj |2 ≤ .
n
Now let N0 = max(N1 , N2 , . . . , Nn ) and write x = nj=1 λj ej . Then for m ≥ N0 ,
P

n n
X X 2
kxm − xk21 = 2
|λj,m − λj | ≤ = 2 .
j=1 j=1
n

Hence by definition, lim xm = x, and since {xm }∞

m=1 was arbitrary, X is complete.
m→∞

Corollary 3.13. If k·k is any norm on a finite dimensional space X, then X is a
complete metric space.

Proof. Let {e1 , e2 , . . . en } be a basis for X and let k·k1 be the norm on X defined
in Lemma 3.11. Then by Corollary 3.10, the norms k·k and k·k1 are equivalent, and by
Lemma 3.12 the metric space X under the norm k·k1 is complete. Thus by Theorem 3.11,
X with norm k·k is also complete. 
Chapter 3. Spaces 13
2. Banach Spaces

When we look at infinite-dimensional vector spaces, we find that there may be two
norms which are not equivalent. So, we may not assume that theorems from the previous
section hold, and will now see what we can say about infinite-dimensional vector spaces.
Theorem 3.14. Let X be a normed vector space and S a linear subspace of X. Then
S̄ is a linear subspace of X.

Proof. Take x, y ∈ S̄ and α ∈ F. Now since x and y are in S̄, by Definition 2.10
(iv), there exist sequences {xn } and {yn } in S such that lim xn = x and lim yn = y. We
n→∞ n→∞
know that S is a linear subspace, and so xn + yn ∈ S for all n ∈ N. Hence
x + y = lim (xn + yn ) ∈ S̄.
n→∞
Similarly, αxn ∈ S for all n ∈ N, so
αx = lim αxn ∈ S̄.
n→∞

Thus by the subspace test, S̄ is a linear subspace of S (and consequently, of X). 

Definition 3.15. A Banach space is a normed vector space which is complete under
the metric associated with the norm.
Theorem 3.16. It is convenient to list a few general spaces which are Banach spaces.
(i) Any finite-dimensional normed space is a Banach space.
(ii) If M is a compact metric space, then CF (M ) is a Banach space.
(iii) If (X, Σ, µ) is a measure space, then Lp (X) is a Banach space (for 1 ≤ p ≤ ∞).
(iv) `p is a Banach space (for 1 ≤ p ≤ ∞).
(v) If X is a Banach space and Y a linear subspace of X, then Y is a Banach space
if and only if Y is closed in X.

Proof.
(i) This follows from Corollary 3.13.
(ii) This follows from Theorem 2.19.
(iii) This follows from Theorem 2.35.
(iv) This is a special case of part (iii) (by taking counting measure on N).
(v) This proof can be found in [1].


3. Inner Product Spaces

Recall that in R3 , we can calculate the angle θ between two vectors using the scalar
product. Much like we did with the length of a vector and the norm, we can extend this
idea to other spaces with higher dimensions by finding a set of axioms that the scalar
product satisfies, and using them to form a definition for the general case.
Chapter 3. Spaces 14
Definition 3.17. Let X be a real vector space. An inner product on X is a function
(·, ·) : X × X → R such that for all x, y, z ∈ X and α, β ∈ R,
(i) (x, x) ≥ 0;
(ii) (x, x) = 0 if and only if x = 0;
(iii) (αx + βy, z) = α(x, z) + β(y, z);
(iv) (x, y) = (y, x).
Pk
Example 3.18. The function (·, ·) : Rk × Rk → R defined by (x, y) = n=1 xn yn is
the standard inner product on Rk .
Note that the above definition and example only apply to real vector spaces. We have
a separate definition for complex inner products that is slightly different.
Definition 3.19. Let X be a complex vector space. An inner product on X is a
function (·, ·) : X × X → C such that for all x, y, z ∈ X and α, β ∈ C,
(i) (x, x) ∈ R and (x, x) ≥ 0;
(ii) (x, x) = 0 if and only if x = 0;
(iii) (αx + βy, z) = α(x, z) + β(y, z);
(iv) (x, y) = (y, x).
Pk
Example 3.20. The function (·, ·) : Ck × Ck → C defined by (x, y) = n=1 xn y n is
the standard inner product on Ck .
Definition 3.21. If X is a real or complex valued vector space with an inner product
(·, ·), then X is an inner product space.
We notice that since the definitions are so similar, we will continue to generalize
theorems and ideas of inner product spaces to F, differentiating between complex and
real inner products when necessary. Generally, an inner product can be defined on any
finite-dimensional vector space.
Example 3.22. Let X be a k-dimensional vector space with basis {e1 , e2 , . . . , ek },
and x, y ∈ X such that x = kn=1 λn en and y = kn=1 µn en . Then the function (·, ·) :
P P

X × X → F defined by (x, y) = kn=1 λn µn is an inner product on X. Furthermore, if

P
{e1 , e2 , . . . , ek } is the standard basis on X, then we obtain a standard inner product.
Definition 3.23. The norm k·k : X → R defined by kxk = (x, x)1/2 on the inner
product space X is known as the induced norm of the inner product (·, ·).
So every inner product space can be regarded as a normed space using the induced
norm. However, it is not true that every norm is induced by an inner product.
Theorem 3.24. Let X be an inner product space and x, y ∈ X. Then the Cauchy-
Schwarz inequality says that
|(x, y)| ≤ kxkkyk.
Lemma 3.25. Let X be an inner product space and let u, v ∈ X. If (x, u) = (x, v) for
all x ∈ X, then u = v.
Chapter 3. Spaces 15
4. Hilbert Spaces

When discussing normed spaces, we noted that completeness is an important property

(if a normed space is complete, we call it a Banach space). We can also define a complete
inner product space as having special characteristics.
Definition 3.26. A Hilbert space is an inner product space which is complete with
respect to the metric associated with the norm induced by the inner product.
Theorem 3.27. Again, we will list some general spaces which are Hilbert spaces.
(i) Any finite-dimensional inner product space is a Hilbert space.
(ii) L2 (X) with the standard inner product is a Hilbert space.
(iii) `2 with the standard inner product is a Hilbert space.
Proof.
(i) This follows from Corollary 3.13.
(ii) This follows from Theorem 2.35.
(iii) This follows from Theorem 2.35.

Example 3.28. Fk with the standard inner product is a Hilbert space.
Theorem 3.29. If H is a Hilbert space and Y ⊂ H is a linear subspace, then Y is a
Hilbert space if and only if Y is closed in H.
Proof.
(⇒) Let H and Y be as described and assume that Y is a Hilbert space. Then by
definition, Y is complete, and by part (i) of Theorem 2.16, Y is closed.
(⇐) Assume that Y ⊂ H and Y is closed. Since H is a Hilbert space, it is complete.
Hence Y is complete (and a Hilbert space) by part (ii) of Theorem 2.16. 
Theorem 3.30. Let Y be a closed linear subspace of a Hilbert space H. For any x ∈ H,
there exists a unique y ∈ Y and z ∈ Y ⊥ such that x = y + z, and kxk2 = kyk2 + kzk2 .

A proof of Theorem 3.30 can be found in [1]. We will now give a couple definitions
and theorems here for use later in the project.
Definition 3.31. Let X be an inner product space. Then the vectors x, y ∈ X are
orthogonal if (x, y) = 0.
Definition 3.32. Let X be an inner product space with A ⊂ X. The orthogonal
complement of A is the set
A⊥ = {x ∈ X : (x, a) = 0 for all a ∈ A}.

So we see that the set A⊥ is the set of vectors in X which are orthogonal to every
vector in A (if A = ∅, then A⊥ = X).
Chapter 3. Spaces 16
3 ⊥
Example 3.33. If X = R and A = {(a1 , a2 , 0) : a1 , a2 ∈ R}, then A = {(0, 0, x3 ) :
x3 ∈ R}, since by definition, x ∈ A⊥ if and only if for any a = (a1 , a2 , 0) we have
(x, a) = x1 a1 + x2 a2 = 0. If we want this equality to hold for every a ∈ A, we must have
x1 = x2 = 0.
 CHAPTER 4

Operators

1. Continuous Linear Transformations

Since we’ve covered some properties of normed spaces, we can now talk about functions
mapping from one normed space into another. The simplest to work with of these maps
will be linear (recall the definition of a linear transformation: a function T : V → W
such that for all α, β ∈ F and x, y ∈ V , T (αx + βy) = αT (x) + βT (y)), and the most
important will be continuous. We will give examples, as well as look at the space of all
continuous linear transformations. First however, we will give alternate characterizations
of continuity for linear transformations.
Theorem 4.1. Let X and Y be normed linear spaces and T : X → Y be a linear
transformation. Then the following are equivalent:
(i)T is uniformly continuous;
(ii)T is continuous;
(iii)T is continuous at 0;
(iv) there exists a real number k > 0 such that kT (x)k ≤ k whenever x ∈ X and
kxk ≤ 1;
(v) there exists a real number k > 0 such that kT (x)k ≤ kkxk for all x ∈ X.

While the first two implications are quite obvious, the proof of this theorem can be
found in [1].
Example 4.2. Define T : C[0, 1] → F by T (f ) = f (0). Then if f ∈ C[0, 1], we get
kT (f )k = |f (0)| ≤ sup{|f (x)| : x ∈ [0, 1]} = kf k.
So using part (v) of Theorem 4.1 with k = 1, we see that T is continuous, and also that
this was a much easier way of proving continuity than the epsilon-delta methods that we
are used to.
Theorem 4.3. Let X be a finite-dimensional normed space, Y be any normed linear
space, and T : X → Y be a linear transformation. Then T is continuous.

While we omit the proof here, it can be found in [1].

Definition 4.4. Let X and Y be normed linear spaces and T : X → Y be a linear
transformation. T is bounded if there exists a real number k > 0 such that kT (x)k ≤ kkxk
for all x ∈ X.
17
Chapter 4. Operators 18
Notice that this definition is part (v) of Theorem 4.1, hence we can use the words con-
tinuous and bounded interchangeably when talking about linear transformations. However,
this use of the word bounded is different than our conventional definition for functions
from R → R. For example, the linear transformation T : R → R defined by T (x) = x is
bounded by Definition 4.4, but obviously not bounded in the usual sense.
Definition 4.5. Let X and Y be normed linear spaces. Then the set of all continuous
linear transformations from X to Y is denoted B(X, Y ). Elements of this set are called
bounded linear operators (or sometimes just operators).
Lemma 4.6. Let X and Y be normed linear spaces, and S, T ∈ B(X, Y ) with kS(x)k ≤
k1 kxk and kT (x)k ≤ k2 kxk for all x ∈ X and some real numbers k1 , k2 > 0. Let λ ∈ F.
Then
(i) k(S + T )(x)k ≤ (k1 + k2 )kxk for all x ∈ X;
(ii) k(λS)(x)k ≤ |λ|k1 kxk for all x ∈ X;
(iii) B(X, Y ) is a linear subspace of L(X, Y ), and so B(X, Y ) is a vector space.
Proof.
(i) For x ∈ X,
k(S + T )(x)k = kS(x) + T (x)k ≤ kS(x)k + kT (x)k ≤ k1 kxk + k2 kxk = (k1 + k2 )kxk.
(ii) For x ∈ X,
k(λS)(x)k = |λ|kS(x)k ≤ |λ|k1 kxk.
(iii) First, B(X, Y ) is not empty because we can always define We see that (S + T )
and (λS) are bounded by Theorem 4.1 (v), and so are elements of B(X, Y ).
Thus parts (i) and (ii) satisfy the subspace test, so B(X, Y ) is a linear subspace
of L(X, Y ) (and hence is a vector space).

For the next definition, recall that if X and Y are normed spaces then the Cartesian
product X × Y is also a normed space.
Definition 4.7. If X and Y are normed spaces and T : X → Y is a linear transfor-
mation, the graph of T is the linear subspace of X × Y defined by
G(T ) = {(x, T x) : x ∈ X}.

2. The Space B(X, Y )

We just saw that B(X, Y ) is a vector space, so we will now look to define a norm. It is
important to note that proofs here may have three different norms from different spaces
in the same equation. However, we will use the symbol k·k as usual for all three norms,
as it will be easy to determine which norm we are referring to based on the space that
the element is from. We should also note that for a normed linear space X, the set of
bounded linear operators from X to X will simply be denoted B(X).
Chapter 4. Operators 19
Lemma 4.8. Let X and Y be normed spaces and define k·k : B(X, Y ) → R by kT k =
sup {kT (x)k : kxk ≤ 1}. Then k·k is a norm on B(X, Y ), so B(X, Y ) is a normed space.

Proof. Before checking the axioms of a norm for k·k, we must note the following
consequence of Theorem 4.1:
sup{kT (x)k : kxk ≤ 1} = inf{k : kT (x)k ≤ kkxk ∀ x ∈ X},

and so kT (y)k ≤ sup{kT (x)k : kxk ≤ 1}kyk ∀ y ∈ X.

Now we can proceed checking the axioms of a norm:
(i) Obviously, kT k = sup {kT (x)k : kxk ≤ 1} ≥ 0 for all T ∈ B(X, Y ).
(ii) Note that the zero transformation is defined by T (x) = 0 for all x ∈ X. Then
kT k = 0 ⇐⇒ kT xk = 0 ∀ x ∈ X
⇐⇒ T x = 0 ∀ x ∈ X
⇐⇒ T is the zero transformation.
(iii) Since kT (x)k ≤ kT kkxk, we know that k(λT )(x)k ≤ |λ|kT kkxk for all x ∈ X by
Lemma 4.6 (ii), hence kλT k ≤ |λ|kT k. If λ = 0, then clearly kαT k = |λ|kT k and
we are done. So assume λ 6= 0. Then
kT k = kλ−1 λT k ≤ |λ−1 |kλT k ≤ |λ|−1 |λ|kT k = kT k.
So we get kT k = |λ|−1 kλT k, and multiplying both sides by |λ|−1 gives
kλT k = |λ|kT k.
(iv) By the construction of kT k, we see that the triangle inequality holds as follows:
k(S + T )(x)k ≤ kS(x)kkT (x)k
≤ kSkkxk + kT kkxk
= (kSk + kT k)kxk.
So kS + T k ≤ kSk + kT k.
Therefore k·k is a norm, known as the norm of T . 

Lemma 4.9. Let X be a normed linear space with W a dense subspace of X, and
let Y be a Banach Space with S ∈ B(X, Y ). If x ∈ X and {xn }, {yn } are sequences
in W such that lim xn = lim yn = x, then {S(xn )} and {S(yn )} both converge, and
n→∞ n→∞
lim S(xn ) = lim S(yn ).
n→∞ n→∞

Proof. Since xn → x, {xn } is a Cauchy sequence, and because we have

kS(xn ) − S(xm )k = kS(xn − xm )k ≤ kSkkxn − xm k,
Chapter 4. Operators 20
we see that {S(xn )} is also a Cauchy sequence. Now since Y is a Banach space, it is
complete, hence {S(xn )} converges. Furthermore, because of the condition lim xn =
n→∞
lim yn = x, we must have lim (xn − yn ) = 0. Now because
n→∞ n→∞

kS(xn ) − S(yn )k = kS(xn − xm )k ≤ kSkkxn − xm k,

we get lim (S(xn ) − S(yn )) = 0, hence lim S(xn ) = lim S(yn ). 
x→∞ n→∞ n→∞

In the next theorem, we explore what conditions are required for B(X, Y ) to be a
Banach Space.
Theorem 4.10. If X is a normed linear space and Y is a Banach space, then B(X, Y )
is also a Banach Space.

Proof. By the Lemma 4.8, we know that B(X, Y ) is a normed space, so we must
show that it is complete. Let {Tn } be a Cauchy sequence in B(X, Y ). Then {Tn } is
bounded, so there exists an M > 0 such that kTn k ≤ M for all n ∈ N. Take x ∈ X. Since
we have
kTn x − Tm xk = k(Tn − Tm )(x)k ≤kTn − Tm kkxk,
and {Tn } is Cauchy, it follows that {Tn x} is a Cauchy sequence in Y . Now since Y is
a Banach space, it is complete, and hence {Tn x} is convergent. So we define a transfor-
mation T : X → Y by T (x) = lim Tn (x). We want to show that T is a bounded linear
n→∞
operator, and that {Tn } converges to T . First showing that T is linear, we see
T (x + y) = lim Tn (x + y) = lim (Tn x + Tn y) = lim Tn x + lim Tn y = T x + T y
n→∞ n→∞ n→∞ n→∞

and T (αx) = lim Tn (αx) = lim αTn x = α lim Tn x = αT (x),

n→∞ n→∞ n→∞
so T is a linear transformation. Furthermore, kT xk = limn→∞ kTn xk ≤ M kxk, so it
follows that T is bounded. Hence T ∈ B(X, Y ). Lastly, we want to show lim Tn = T .
n→∞

So take  > 0. There exists N ∈ N such that for m, n ≥ N , kTn − Tm k < . Then for
2

any x with kxk ≤ 1 and m, n ≥ N, kTn x − Tm xk ≤ kTn − Tm kkxk < kxk. Now since
2

T (x) = lim Tn (x), there exists an N1 ∈ N such that for m ≥ N1 , kT x − Tm xk < . Thus
n→∞ 2
for n ≥ N and m ≥ N1 ,
 
kT x − Tn xk ≤ kT x − Tm xk + kTn x − Tm xk < + kxk ≤ .
2 2
So we get kT − Tn k ≤  when n ≥ N , thus lim Tn = T . Since {Tn } was an arbitrary
n→∞
Cauchy sequence, B(X, Y ) is complete. 
Definition 4.11. Let X, Y , and Z be normed linear spaces, T ∈ B(X, Y ) and
S ∈ B(Y, Z). Then the composition of S and T , called the product of S and T , will be
denoted by ST .
Chapter 4. Operators 21
Given the conditions from the definition, if X, Y , and Z are not all the same, the
fact that we can define ST does not necessarily mean that we can define T S. However,
if X = Y = Z, both products will be defined, but generally T S 6= ST . Another notation
point to make is that if X is a normed space and T ∈ B(X), the product T T will be
denoted T 2 , and the product of T with itself n times will be denoted T n .

3. Isometries, Isomorphisms, and Inverses

In this section we discuss different classifications that may be given to an operator.

Definition 4.12. Let X and Y be normed linear spaces and T ∈ L(X, Y ). If
kT (x)k = kxk for all x ∈ X, then T is called an isometry.
One can see it is easy to find the norm of this type of operator. Note that on every
normed space, there is at least one isometry, as we see next.
Example 4.13. If X is a normed space and I is the identity linear transformation on
X, then I is an isometry since for all x ∈ X, I(x) = x. Hence kI(x)k = kxk.
Definition 4.14. If X and Y are normed linear spaces and T is an isometry from
X onto Y , then T is called an isometric isomorphism and X and Y are isometrically
isomorphic.
If two spaces are isometrically isomorphic, we can think of them as having essentially
the same structure concerning vector space and norm properties. Next we will try and
generalize the idea of an inverse to an infinite-dimensional setting.
Definition 4.15. Let X and Y be normed linear spaces. An operator T ∈ B(X, Y )
is invertible if there exists an S ∈ B(Y, X) such that ST = IX and T S = IY . We say that
S is the inverse of T and denote it by T −1 .
Definition 4.16. Let X and Y be normed linear spaces. If there exists an invertible
operator T ∈ B(X, Y ), then X and Y are isomorphic, and we say T is an isomorphism.
It is apparent that if T ∈ B(X, Y ) is an isomorphism, then T −1 ∈ B(X, Y ) is also an
isomorphism. Also, spaces that are isomorphic are similar in structure, as we see next.
Lemma 4.17. If two normed linear spaces X and Y are isomorphic, then:
(i) dimX < ∞ ⇐⇒ dimY < ∞ (in which case, dimX = dimY );
(ii) X is separable ⇐⇒ Y is separable;
(iii) X is complete ⇐⇒ Y is complete.
Lemma 4.18. If X and Y are normed linear spaces and T ∈ B(X, Y ) is invertible,
then for all x ∈ X,
kT xk ≥ kT −1 k−1 kxk.
Proof. For all x ∈ X we have kxk = kT −1 (T x)k ≤ kT −1 kkT xk, hence multiplying
both sides by the inverse of kT −1 k gives kT xk ≥ kT −1 k−1 kxk, as desired. 
 CHAPTER 5

Duality

1. Dual Spaces

Here we will study another space of functions known as the dual space.
Definition 5.1. Let X be a normed space. Linear transformations from X to F are
called linear functionals. The space B(X, F) is called that dual space of X and is denoted
by X 0 .
Corollary 5.2. If X is a normed vector space, then X 0 is a Banach space.

Proof. We know that the space F is complete, hence by Theorem 4.10, X 0 = B(X, F)
is a Banach space. 
Definition 5.3. We define the Kronecker delta, denoted δjk . For any integers j, k,

1, if j = k
δjk =
0, if j 6= k
Theorem 5.4. If X is a finite dimensional normed linear space with basis {v1 , v2 , . . . , vn },
then X 0 has a basis {f1 , f2 , . . . , fn } such that fj (vk ) = δjk for 1 ≤ j, k ≤ n. In particular,
dimX 0 = dimX.

Proof. Let x ∈ X. Since {v1 , v2 , P . . . , vn } is a basis for X, we know there exist unique
scalars α1 , α2 , . . . , αn such that x = nk=1 αk vk . For j = 1, . . . , n, we define a function
fj : X → F by fj (x) = αj for x ∈ X. It can be verified that fj is a linear transformation
such that fj (vk ) = δjk . Also, fj is continuous by Theorem 4.3, and so fj ∈ X 0 . Now to
showP that {f1 , f2 , . . . , fn } is a basis for X 0 , suppose that β1 , β2 , . . . , βn are scalars such
that nj=1 βj fj = 0. Then
n
X n
X
0= βj fj (vk ) = βj δjk = βk , 1 ≤ k ≤ n,
j=1 j=1

and so {f1 , f2 , . . . , fn } is linearly independent.

Now take an arbitrary f ∈ X 0 and let γj = f (vj ), j = 1, . . . , n. Then
n
X n
X
γj fj (vk ) = γk δjk = γk = f (vk ), 1 ≤ k ≤ n,
j=1 j=1
Pn
thus f = j=1 γj fj , since {v1 , v2 , . . . , vn } is a basis for X. 
22
Chapter 5. Duality 23
0
Theorem 5.5 (Riesz-Frechet Theorem). Let H be a Hilbert space and f ∈ H . Then
there exists a unique y ∈ H such that f (x) = fy (x) = (x, y) for all x ∈ H. Furthermore,
kf k = kyk.

Proof. First, we will prove the existence of y. If f (x) = 0 for all x ∈ H, then y = 0
works. Otherwise, Kerf = {x ∈ H : f (x) = 0} is a proper closed subspace of H, so by
Theorem 3.30, (Kerf )⊥ 6= {0}. Hence there exists a z ∈ (Kerf )⊥ such that f (z) = 1.
z
Now we define y = (since z 6= 0), and take an arbitrary x ∈ H. Then since f is
kzk2
linear,
f (x − f (x)z) = f (x) − f (x)f (z) = 0,
so x−f (x)z ∈ Kerf . But z ∈ (Kerf )⊥ , so (x−f (x)z, z) = 0. Thus (x, z)−f (x)(z, z) = 0,
so by rearranging we see (x, z) = f (x)kzk2 , and therefore
z
f (x) = (x, ) = (x, y).
kzk2
Now if kxk ≤ 1, then using the Cauchy-Schwarz inequality we get
|f (x)| = |(x, y)| ≤ kxkkyk ≤ kyk,
y
so f (x) ≤ kyk. Furthermore, if x = then kxk = 1 and
kyk
|f (y)| (y, y)
kf k ≥ |f (x)| = = = kyk.
kyk kyk
Hence we also have f (x) ≥ kyk, and so f (x) = kyk.
Next, we will prove uniqueness. If y and w are such that
f (x) = (x, y) = (x, w)
for all x ∈ H, then we must have (x, y − w) = 0 for all x ∈ H. So by Lemma 3.25, we
have y − w = 0, meaning that y = w. Hence y is unique. 

Theorem 5.5 gives a representation of all elements of the dual space of a general Hilbert
space. So, in a sense, H0 can be identified with H.

2. Sublinear Functionals and Seminorms

If we have a vector space X, many times in functional analysis we will have a linear
functional fW : W → F defined on a subspace W ⊂ X, but to use this functional, we
need it to be defined on the whole set X. So, we will now see if we can extend the domain
from W to X, and get the following definition.
Definition 5.6. Let X be a vector space, W a linear subspace of X, and fW a linear
functional on W . Then a linear functional fX on X is an extension of fW if fX (w) = fW (w)
for all w ∈ W .
Chapter 5. Duality 24
Definition 5.7. Let X be a real vector space. A sublinear functional on X is a
function p : X → R such that:
(i) p(x + y) ≤ p(x) + p(y) x, y ∈ X.
(ii) p(αx) = αp(x) x ∈ X, α ≥ 0.

Note that from this definition, it is clear that if p is a sublinear functional on X, then

p(0) = 0,
−p(−x) ≤ p(x) for all x ∈ X,
(2.1) −p(y − x) ≤ p(x) − p(y) ≤ p(x − y) for all x, y ∈ X.

In addition, if we know that p satisfies p(−x) = p(x) for all x ∈ X, then p(x) ≥ 0 for all
x ∈ X, and (1.1) becomes:

|p(x) − p(y)| ≤ p(x − y) for all x, y ∈ X.

Example 5.8. We will list some examples here of sublinear functionals.

(i) If f is a linear functional on X, then it is sublinear.
(ii) If f is a non-zero linear functional on X, then the functional p(x) = |f (x)| is
sublinear (but it is not necessarily linear).
(iii) If X is a normed space then p(x) = kxk is sublinear.
(iv) If X = R2 and p(x1 , x2 ) = |x1 | + x2 , then p is sublinear.

Definition 5.9. Let X be a real or complex vector space. A seminorm on X is a

real-valued function p : X → R such that:
(i) p(x + y) ≤ p(x) + p(y) for all x, y ∈ X;
(ii) p(αx) = |α|p(x) for all x ∈ X and α ∈ F.

Notice the distinctions between the definition of a sublinear functional and seminorm.
We have that if X is real, then a seminorm p is a sublinear functional (but the converse
is not necessarily true). Also notice that if we add the condition p(x) = 0 implies x = 0,
then the seminorm p meets the criteria of a norm.
Also, the definition implies that if p is a seminorm, then

p(0) = 0,

p(−x) = p(x) for all x ∈ X,

p(x) ≥ 0 for all x ∈ X, and

|p(x) − p(y)| ≤ p(x − y) for all x, y ∈ X.

Chapter 5. Duality 25
3. The Hahn-Banach Theorem

We will state some definitions and lemmas here for use in our proof of the Hahn-Banach
theorem. This theorem allows us to construct extensions of functionals on general spaces,
and show the existence of elements in a dual space.
Definition 5.10. Suppose that M is a non-empty set, and ≺ is an ordering on M.
Then ≺ is a partial order on M if
(i) x ≺ x for all x ∈ M;
(ii) x ≺ y and y ≺ x ⇒ x = y;
(iii) x ≺ y and y ≺ z ⇒ x ≺ z;
and we say M is a partially ordered set.
In addition, if ≺ is defined for all pairs of elements (so for any x, y ∈ M, either x ≺ y
or y ≺ x holds), then ≺ is a total order and M is a totally ordered set.
Furthermore, if M is a partially ordered set, then y ∈ M is a maximal element of M
if y ≺ x ⇒ y = x. If N ⊂ M, then y ∈ M is an upper bound for N if x ≺ y for all x ∈ N .
Example 5.11. We use partial and total orders in mathematics all the time.
(i) The usual ordering, ≤ on R is a total order.
(ii) A partial order on R2 is given by (x1 , x2 ) ≺ (y1 , y2 ) ⇐⇒ x1 ≤ y1 and x2 ≤ y2 .
(iii) If S is an arbitrary set and M is the set of all subsets of S, then set inclusion
(A ⊂ B for A, B ∈ M) is a partial order on M.
Lemma 5.12. Let X be a real vector space with a proper linear subspace, W . Let p be
a sublinear functional on X and fW be a linear functional on W such that fW (w) ≤ p(w)
for all w ∈ W . Suppose that z1 ∈
/ W , and let
W1 = Sp{z1 } ⊕ W = {αz1 + w : α ∈ R, w ∈ W }.
Then there exists a ξ1 ∈ R and fW1 : W1 → R that satisfies
(3.1) fW1 (αz1 + w) = αξ1 + fW (w) ≤ p(αz1 + w), α ∈ R, w ∈ W.
Since fW is a linear functional, fW1 is also linear, and for w ∈ W we have fW1 (w) =
fW (w), so fW1 is an extension of fW .
Proof. For any u, v ∈ W , we have
fW (u) + fW (v) = fW (u + v) ≤ p(u + v) ≤ p(u − z1 ) + p(v + z1 ),
which gives
fW (u) − p(u − z1 ) ≤ −fW (u) + p(v + z1 ).
Hence
ξ1 = inf {−fW (v) + p(v + z1 )} > −∞,
v∈W
and
−ξ1 + fW (u) ≤ p(u − z1 ), ξ1 + fW (v) ≤ p(v + z1 ), u, v ∈ W.
Chapter 5. Duality 26
Now multiplying the first inequality by a β > 0 gives
−βξ1 + fW (βu) ≤ p(β(u − z1 )),
and if we let α = −β and w = βu, we get the desired inequality (3.1) when α < 0.
Similarly, the second inequality gives (3.1) when α > 0, and for α = 0, (3.1) follows
immediately from the way p is defined. 
Lemma 5.13 (Zorn’s Lemma). Let M be a non-empty, partially ordered set such that
every totally ordered subset of M has an upper bound. Then there exists a maximal
element in M.
Theorem 5.14 (The Hahn-Banach Theorem). Let X be a real vector space, with a
sublinear functional p defined on X. Suppose that W is a linear subspace of X, and fW
is a linear functional on W satisfying
fW (w) ≤ p(w), w ∈ W.
Then fW has an extension fX on X such that
fX (x) ≤ p(x), x ∈ X.

Proof. Let E denote the set of linear functionals f on X that satisfy the following:
(i) f is defined on a linear subspace Df such that W ⊂ Df ⊂ X;
(ii) f (w) = fW (w) for w ∈ W ;
(iii) f (x) ≤ p(x) for x ∈ Df .
So E is the set of all extensions f on fW to general subspaces Df ⊂ X which satisfy
the hypothesis of the theorem on their domain. We will apply Zorn’s lemma to our set
E, and then show that the maximal element of E is our desired functional. So first let us
verify that E satisfies the criteria of Zorn’s lemma. We know that fW ∈ E, so E 6= ∅. Now
define a relation ≺ on E as follows: for any f, g ∈ E,
f ≺ g ⇐⇒ Df ⊂ Dg and f (x) = g(x) for all x ∈ Df .
So f ≺ g iff g is an extension of f . One can verify that ≺ is a partial order on E (though
it is not a total order, since there can be functionals f, g ∈ E which are both extensions
of fW but neither is an extension of the other). Now, take a totally ordered set G ⊂ E.
By definition, G is totally ordered iff for any two arbitrary functionals f, g ∈ G, one is an
extension of the other. We will construct an upper bound for G in E. Define a set
[
ZG = Df .
f ∈G

Now using the total ordering of G, it can be verified that Zg is a linear subspace of X.
So we will define a linear functional fG on ZG as follows. Given a z ∈ ZG , there exists a
ξ ∈ E such that z ∈ Dξ . Next define fG (z) = ξ(z) (this definition does not depend on ξ,
since if η is a functional in G with z ∈ Dη , we must have ξ(z) = η(z) by the total ordering
of G). Again, we can verify fG is linear by the total ordering of G. Also, since ξ ∈ E, we
Chapter 5. Duality 27
have fG (z) = ξ(z) ≤ p(z), and if z ∈ W then fG (z) = fW (z). Thus fG ∈ E and f ≺ fG
for all f ∈ G, and we have that fG is an upper bound for G.
Now since G was an arbitrary, totally ordered subset of E, we conclude by Zorn’s
lemma that E has a maximal element, fmax . Now suppose that the domain Df max 6= X.
Then by Lemma 5.12, fmax has an extension which also lies in E. However, this would
contradict the maximality of fmax in E. So we must have Df max = X, and hence fX = fmax
is our desired extension. 
 CHAPTER 6

Concluding Remarks

In this paper, we have given a basic introduction to the branch of mathematics known
as linear functional analysis. We discussed the fundamental ideas of normed spaces, inner
product spaces, operators, dual spaces, and the Hahn-Banach Theorem. These concepts
have applications in many other areas of mathematics, including applied studies such as
quantum mechanics and differential equations.
There is plenty of opportunity for further study in this field. While I only proved
the Hahn-Banach theorem for real vector spaces, there are other versions with different
hypotheses, including a version for normed spaces. This is a tool used in many proofs
in all areas of mathematics. Staying in functional analysis, some further topics of study
include weak convergence, the adjoint of an operator, compact operators, and spectral
theory.

28
 Bibliography

[1] Bryan P. Rynne and Martin A. Youngston. Linear Functional Analysis. Springer, second edition,
2008. 12, 13, 15, 17

29