Functional Analysis

Daniel Mao

Copyright © 2020 - 2022 Daniel Mao All Rights Reserved.

Contents

1 Linear Space 1
1.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Linear Subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Linear Span . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Linear Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Normed Linear Space 11

2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Equivalence of Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Finite-Dimensional Normed Linear Space . . . . . . . . . . . . . . . . . . . . 13
2.5 Dual Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Inner Product Space 17

3.1 Inner Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Inner Product Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4 Orthogonality 21
4.1 Orthogonal Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2 Orthogonal Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Orthogonal Complements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Orthogonal Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.5 Inequalities in Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 Sequence Space 27
5.1 p-norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 `p Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

i
ii CONTENTS

5.3 c0 Space and c00 Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.4 Hölder’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

6 Function Spaces 35
p
6.1 The L Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

7 Quotient Space 37
7.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.3 Quotient Spaces with Seminorms . . . . . . . . . . . . . . . . . . . . . . . . . 38
7.4 Quotient Spaces with Topologies . . . . . . . . . . . . . . . . . . . . . . . . . 39

8 Dual Space 41
8.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.3 Linear Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.4 The Dual Space Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.5 Annihilator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

9 Banach Space 45
9.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
9.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9.4 Direct Sums and Direct Products of Banach Spaces . . . . . . . . . . . . . . . 49
9.5 Unconditional Convergence in Banach Spaces . . . . . . . . . . . . . . . . . . 49
9.6 The Open Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

10 Hilbert Space 55
10.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
10.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
10.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
10.4 The Riesz Representation Theorem . . . . . . . . . . . . . . . . . . . . . . . . 56

11 Operators 59
11.1 Bounded Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
11.2 Operator Norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
11.3 Examples of Bounded Operators . . . . . . . . . . . . . . . . . . . . . . . . . 60
11.4 The Space of Bounded Linear Operators . . . . . . . . . . . . . . . . . . . . . 64
11.5 Invertible Bounded Linear Operators . . . . . . . . . . . . . . . . . . . . . . . 64

12 Matrix Space 67
CONTENTS iii

13 Balanced Set 69
13.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
13.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
13.3 Stability of Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
13.4 The Balanced Hull Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
13.5 The Balanced Core Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

14 Topological Vector Space 73

14.1 Preliminaries: Absorbing Set . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
14.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
14.3 Neighborhood Improvements . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
14.4 Properties of Topological Vector Spaces . . . . . . . . . . . . . . . . . . . . . 75
14.5 Operation on Sets in a Topological Vector Space . . . . . . . . . . . . . . . . 76
14.6 Finite-Dimensional Topological Vector Spaces . . . . . . . . . . . . . . . . . . 78

15 Continuous and Uniformly Continuous Functions 81

15.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
15.2 Extension of Continuous Linear Maps . . . . . . . . . . . . . . . . . . . . . . 81
15.3 Relation between the two Notions . . . . . . . . . . . . . . . . . . . . . . . . . 81

16 Complete Space 83
16.1 Cauchy Nets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
16.2 Complete Topological Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . 83

17 Locally Convex Space 85

17.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
17.2 Locally Convex Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
17.3 Relation to Other Topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
17.4 Continuity in Locally Convex Spaces . . . . . . . . . . . . . . . . . . . . . . . 91
17.5 Convergence in Locally Convex Spaces . . . . . . . . . . . . . . . . . . . . . . 92
17.6 Strong Operator Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
17.7 Weak Operator Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

18 The Hahn-Banach Theorem 95

18.1 The Extension Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
18.2 Separation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

19 Reflexive Banach Space 101

19.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
19.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
iv CONTENTS

20 Weak Topology 105

20.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
20.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
20.3 Theory on Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

21 Locally Compact Space 111

21.1 The F. Riesz’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

22 Adjoint Operator 113

22.1 Banach Space Adjoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
22.2 Hilbert Space Adjoint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
22.3 Properties of the Adjoint Operator . . . . . . . . . . . . . . . . . . . . . . . . 114
22.4 Normal Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

23 Convolution 117

24 Coercive Functions 119

24.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
24.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

25 Unclassified Results 121

Chapter 1

Linear Space

1.1 Definitions

1.2 Linear Subspaces

PROPOSITION 1.1. The singleton set consisting of the zero vector is a linear
subspace.

PROPOSITION 1.2. The intersection of a collection of subspaces is a subspace.

PROPOSITION 1.3. The sum of two subspaces is a subspace.

Proof. Let V be a vector space over field F. Let W1 and W2 be subspaces of V. I will show
that W1 + W2 is a subspace of V. Clearly W1 + W2 ⊆ V and W1 + W2 6= ∅. So there remains
only to show that W1 + W2 is closed under addition and scalar multiplication.
Closed Under Addition:
Let x and y be arbitrary elements of W1 + W2 . Then x = x1 + x2 for some x1 ∈ W1
and some x2 ∈ W2 ; and y = y1 + y2 for some y1 ∈ W1 and some y2 ∈ W2 . Since W1 is a
subspace, it is closed under addition. So x1 + y1 ∈ W1 . Since W2 is a subspace, it is closed
under addition. So x2 + y2 ∈ W2 . So

x + y = x1 + x2 + y1 + y2 = (x1 + y1 ) + (x2 + y2 ) ∈ W1 + W2 .

1
2 CHAPTER 1. LINEAR SPACE

So W1 + W2 is closed under addition.

Closed Under Scalar Multiplication:
Let k be an arbitrary element of F. Let x be an arbitrary element of W1 + W2 . Then
x = x1 + x2 for some x1 ∈ W1 and some x2 ∈ W2 . Since W1 is a subspace, it is closed
under scalar multiplication. So kx1 ∈ W1 . Since W2 is a subspace, it is closed under scalar
multiplication. So kx2 ∈ W2 . So

kx = k(x1 + x2 ) = kx1 + kx2 ∈ W1 + W2 .

So W1 + W2 is closed under scalar multiplication.

1.3 Linear Span

PROPOSITION 1.4. Linear spans are linear subspaces.

Proof. Let V be a vector space over field F. Let S = {x1 , ..., xk } be a subset of V where
k ∈ N. I will show that span(S) is a subspace of V. Clearly span(S) ⊆ V and span(S) 6= ∅. So
there remains only to show that span(S) is closed under addition and scalar multiplication.
Closed Under Addition:
k
X
Let a and b be arbitrary elements of span(S). Then a = λi xi for some λ1 , ..., λk ∈ F;
i=1
k
X
and b = µi xi for some µ1 , ..., µk ∈ F. Then
i=1

k
X k
X k
X
a+b= λi xi + µi xi = (λi + µi )xi .
i=1 i=1 i=1

So a + b ∈ span(S). So span(S) is closed under addition.

Closed Under Scalar Multiplication:
Let a be an arbitrary element of span(S). Let k be an arbitrary element of F. Then
Xk
a= λi xi for some λ1 , ..., λk ∈ F. Then
i=1

k
X k
X
ka = k λ i xi = (kλi )xi .
i=1 i=1

So ka ∈ span(S). So span(S) is closed under scalar multiplication.

1.4. LINEAR INDEPENDENCE 3

PROPOSITION 1.5 (The Linear Span Operator). Let V be a vector space over field
F. Then the linear span operator span has the following properties.

1. (Extensive) Let S be a subset of V. Then

S ⊆ span(S).

2. (Monotonic) Let S and T be subsets of V. Suppose that S ⊆ T . Then

span(S) ⊆ span(T ).

3. (Idempotent) Let S be a subset of V. Then

span(span(S)) = span(S).

Proof of (1). Let x be an arbitrary element of S. Define n := 1 ∈ N, λ1 := 1 ∈ F, and

Xn
x1 := x ∈ S. Then x = 1 · x = λ1 x1 = λi xi . So x ∈ span(S).
i=1

Proof of (3). From (1) we know that span(S) ⊆ span(span(S)). So there remains only
to show that span(span(S)) ⊆ span(S). Let x be an arbitrary element of span(span(S)).
n
X
Then ∃n ∈ N, λ1 , ..., λn ∈ F, and v1 , ..., vn ∈ span(S) such that x = λi vi . For each
i=1
i ∈ {1, ..., n}, since vi ∈ span(S), ∃ni ∈ N, ∃λi,1 , ..., λi,ni ∈ F, and ∃vi,1 , ..., vi,ni ∈ S such
Xni
that vi = λi,j vi,j . So
j=1

n
X ni
X ni
n X
X
x= λi λi,j vi,j = (λi λi,j )vi,j .
i=1 j=1 i=1 j=1

Note that ∀i, j ∈ {1, ..., n}, λi λi,j ∈ F and vi,j ∈ S. So x ∈ span(S). So span(span(S)) ∈
span(S). This completes the proof.

1.4 Linear Independence

4 CHAPTER 1. LINEAR SPACE

DEFINITION 1.6 (Linear Dependence and Linear Independence). Let V be a vector

space over field F. Let S be a subset of V. We say that S is linearly dependent if
n
X
∃x ∈ S, ∃n ∈ N, ∃x1 , ..., xn ∈ S \ {x}, ∃k1 , ..., kn ∈ F such that x = ki xi .
i=1
1.5. BASIS 5

We say that S is linearly independent if S is not linearly dependent.

PROPOSITION 1.7. A set that contains the zero vector is linearly dependent.

1.5 Basis

DEFINITION 1.8 (Basis - 1). Let V be a vector space. Let B be a subset of V. We

say that B is a basis for V if B is linearly independent and B is spanning for V.

DEFINITION 1.9 (Basis - 2). Let V be a vector space. Let B be a subset of V. We

say that B is a basis for V if B is a maximal linearly independent subset of V.

DEFINITION 1.10 (Basis - 3). Let V be a vector space. Let B be a subset of V.

We say that B is a basis for V if B is a minimal spanning subset of V.

PROPOSITION 1.11. Definitions 2 and 3 of basis are equivalent.

Proof. Let V be a vector space. Let S be a subset of V.

Forward Direction: Assume that S is a minimal spanning subset. I will show that S is a
maximal linearly independent set.
Assume for the sake of contradiction that S is not linearly independent. Then ∃x ∈ S
such that x ∈ span(S \ {x}). So span(S \ {x}) = span(S) = V. This contradicts to the
assumption that S is minimal. So S is linearly independent.
Assume for the sake of contradiction that S is not maximal. Then ∃y ∈ V \ S such that
S ∪ {y} is linearly independent. So y ∈
/ span(S). This contradicts to the assumption that S
is spanning. So S is maximal.
Backward Direction: Assume that S is a maximal linearly independent set. I will show
that S is a minimal spanning subset.
Assume for the sake of contradiction that S is not spanning. Then ∃x ∈ V such that
x∈
/ span(S). So S ∪ {x} is linearly independent. This contradicts to the assumption that
S is maximal. So S is spanning.
6 CHAPTER 1. LINEAR SPACE

Assume for the sake of contradiction that S is not minimal. Then ∃y ∈ S such that
S \ {y} is spanning. So y ∈ span(S \ {y}). This contradicts to the assumption that S is
linearly independent. So S is minimal.

PROPOSITION 1.12. All three definitions of basis are equivalent.

THEOREM 1.13. Let V be a vector space over field F. Let S1 be a linearly inde-
pendent subset of V. Let S2 be a spanning subset of V. Suppose that S1 ⊆ S2 . Then
∃B ⊆ V such that S1 ⊆ B ⊆ S2 and that B is a basis for V.

Proof. Let F be the family of all linearly independent subsets S of V such that S1 ⊆ S ⊆ S2 .
I claim that there is a maximal element B of F. Let C be an arbitrary chain of elements of
[
F. Define a set U for C by U := C. Clearly S1 ⊆ U ⊆ S2 . So there remains only to
C∈C
show that U is linearly independent.
X n
Let n ∈ N, v1 , ..., vn ∈ U , and λ1 , ..., λn ∈ F be such that λi vi = 0. For each
[ i=1
i ∈ {1, ..., n}, since vi ∈ U and U = C, ∃Ci ∈ C such that vi ∈ Ci . Since C is a chain,
C∈C
∃i0 ∈ {1, ..., n} such that ∀i ∈ {1, ..., n}, Ci ⊆ Ci0 . So ∀i ∈ {1, ..., n}, vi ∈ Ci0 . Since
Ci0 ∈ C ⊆ F, Ci0 is linearly independent. So λ1 , ..., λn = 0. So U is linearly independent.
It follows that there is a maximal element B of F. Since B ∈ F, B is linearly independent.
Since B is maximal in F, S2 ⊆ span(B). So span(B) = V. So B is a basis for V.

COROLLARY 1.14 (The Replacement Theorem). Let V be a vector space. Let B

be a basis for V. Let S be a linearly independent subset of V. Then ∃B 0 ⊆ B such that
B 0 ∪ S is a basis for V. i.e., S replaces B \ B 0 .

PROPOSITION 1.15. All bases have equal cardinality.

PROPOSITION 1.16. Every spanning subset of a vector space can be reduced to a

basis for the space.
1.6. DIMENSION 7

PROPOSITION 1.17. Every linearly independent subset of a vector space can be

extended to a basis for the space.

THEOREM 1.18. Every vector space has a basis.

1.6 Dimension

DEFINITION 1.19 (Dimension). Let V be a vector space. Let B be a basis for V.

We define the dimension of V, denoted by dim(V), to be the cardinality of B.

1.6.1 Finite Dimensional Linear Space

PROPOSITION 1.20. Let V be a finite dimensional vector space. Let S be a subset

of V. Then

1. If |S| > dim(V), then S is not linearly independent.

2. If |S| < dim(V), then S is not spanning for V.

Proof of Statement 1. Suppose that |S| > dim(V).

Assume for the sake of contradiction that S is linearly independent.
Since dim(V) is the cardinality of the maximal linearly independent subset of V,
we have dim(V) ≥ |S|. Note that |S| > dim(V). So dim(V) > dim(V).
This leads to a contradiction.
So S is not linearly independent.

Proof of Statement 2. Suppose that |S| < dim(V).

Assume for the sake of contradiction that S is spanning for V.
Since dim(V) is the cardinality of the minimal spanning subset of V,
we have dim(V) ≤ |S|. Note that |S| < dim(V). So dim(V) < dim(V).
This leads to a contradiction.
So S is not spanning for V.
8 CHAPTER 1. LINEAR SPACE

PROPOSITION 1.21. Let V be a finite dimensional vector space. Let S be a subset

of V. Suppose that |S| = dim(V). Then S is linearly independent if and only if S is
spanning for V.

Proof of Direction 1. Assume that S is linearly independent. I will show that S is spanning
for V. Since S is linearly independent, by the Replacement Theorem, there is a subset B 0
of some basis B for V such that S ∪ B 0 is a basis for V. Since S ∪ B 0 is a basis for V,
|S ∪ B 0 | = dim(V). So |S| = |S ∪ B 0 |. So B 0 = ∅. So S is a basis for V.

Proof of Direction 2. Assume that S is spanning for V. I will show that S is linearly in-
dependent. Assume for the sake of contradiction that S is not linearly independent. Then
∃S 0 ( S that is linearly independent. Then |S 0 | < |S| = dim(V). So |S 0 | is not span-
ning for V. This contradicts to the assumption that S is spanning for V. So S is linearly
independent.

1.6.2 Finite Dimensional Subspace

PROPOSITION 1.22. Let V be a vector space. Let W be a subspace of V. Then

1. dim(W) ≤ dim(V).

2. If dim(W) = dim(V) and are both finite, then W = V.

Proof of (1). Any linearly independent subset of W is a linearly independent subset of V.

Any spanning subset of V is a spanning subset of W.

Proof of (2). Let B be a basis for W. Then |B| = dim(W) and hence |B| = dim(V). Since
B is a basis, it is linearly independent. Since B is linearly independent and |B| = dim(V), B
is a basis for V. Since B is a basis for W and V, we have W = span(B) and V = span(B).
So W = V.

PROPOSITION 1.23. Let V be a vector space over field F. Let W1 and W2 be

finite dimensional subspaces of V. Then we have

dim(W1 + W2 ) = dim(W1 ) + dim(W2 ) − dim(W1 ∩ W2 ).

1.6. DIMENSION 9

Proof. Since W1 and W2 are subspaces of V, W1 ∩W2 is a subspace of V. Let k := dim(W1 ∩

W2 ). Then k ≤ min{dim(W1 ), dim(W2 )}. Let m := dim(W1 ) − k and n := dim(W2 ) − k.
Let B = {v1 , ..., vk } be a basis for W1 ∩ W2 . Extend B to a basis B1 = {v1 , ..., vk , x1 , ..., xm }
for W1 . Extend B to a basis B2 = {v1 , ..., vk , y1 , ..., yn } for W2 .
Part 1: Show that B1 ∪ B2 = {v1 , ..., vk , x1 , ..., xm , y1 , ..., yn }.
Assume for the sake of contradiction that {x1 , ..., xm } ∩ {y1 , ..., yn } =
6 ∅. Say ∃v ∈
{x1 , ..., xm } ∩ {y1 , ..., yn }. Then {v1 , ..., vk , v} is a linearly independent subset of W1 ∩ W2 .
However, this contradicts to the assumption that k = dim(W1 ∩ W2 ). So {x1 , ..., xm } ∩
{y1 , ..., yn } = ∅. So
B1 ∪ B2 = {v1 , ..., vk , x1 , ..., xm , y1 , ..., yn }.
Part 2: Show that B1 ∪ B2 is linearly independent.
Let λ1 , ..., λk , α1 , ..., αm , β1 , ..., βn ∈ F be such that
k
X m
X n
X
λi vi + αi xi + βi yi = 0.
i=1 i=1 i=1

I will show that λ1 , ..., λk , α1 , ..., αm , β1 , ..., βn = 0. From the above we get
k
X m
X n
X
λ i vi + αi xi = (−βi )yi .
i=1 i=1 i=1

Note that the LHS ∈ span(B1 ) = W1 . So RHS = LHS ∈ W1 . Note that RHS ∈ span(B2 ) =
W2 . So RHS ∈ W1 ∩ W2 = span(B). So ∃µ1 , ..., µk ∈ F such that
n
X k
X
RHS = (−βi )yi = µi vi .
i=1 i=1

That is,
k
X n
X
µi vi + βi yi = 0.
i=1 i=1
Since B2 is linearly independent, we get µ1 , ..., µk , β1 , ..., βn = 0. So
k
X m
X n
X
λi vi + αi xi = (−βi )yi = 0.
i=1 i=1 i=1

Since B1 is linearly independent, we get λ1 , ..., λk , α1 , ..., αm = 0. So B1 ∪ B2 is linearly

independent.
Part 3: Show that B1 ∪ B2 is a spanning subset of W1 + W2 .
Clearly span(B1 ∪ B2 ) ⊆ W1 + W2 . So there remains only to show that W1 + W2 ⊆
span(B1 ∪ B2 ). Let x be an arbitrary element of W1 + W2 . Then x = x1 + x2 for some
x1 ∈ W1 and some x2 ∈ W2 . Then ∃λ1 , ..., λk , α1 , ..., αm ∈ F and ∃µ1 , ..., µk , β1 , ..., βn ∈ F
such that
k
X m
X k
X n
X
x1 = λi vi + α i xi and x2 = µi vi + βi yi .
i=1 i=1 i=1 i=1
10 CHAPTER 1. LINEAR SPACE

So
k
X m
X k
X n
X
x = x1 + x2 = λ i vi + αi xi + µi vi + β i yi
i=1 i=1 i=1 i=1
k
X m
X n
X
= (λi + µi )vi + αi xi + βi yi ∈ span(B1 ∪ B2 ).
i=1 i=1 i=1

So W1 ⊆ W2 ⊆ span(B1 ∪ B2 ). So B1 ∪ B2 is a spanning subset of W1 + W2 .

Part 4:
By part 2 and part 3, B1 ∪ B2 is a basis for W1 + W2 . So

dim(W1 + W2 ) = |B1 ∪ B2 | = k + m + n. (1)

On the other hand, we have

dim(W1 ) + dim(W2 ) − dim(W1 ∩ W2 ) = |B1 | + |B2 | − |B| = k + m + n. (2)

From (1) and (2), we get

dim(W1 + W2 ) = dim(W1 ) + dim(W2 ) − dim(W1 ∩ W2 ),

as desired.

COROLLARY 1.24. Let V be a vector space. Let W1 and W2 be finite dimensional

subspaces of V. Then we have

dim(W1 + W2 ) ≤ dim(W1 ) + dim(W2 ).

Chapter 2

Normed Linear Space

2.1 Definitions

DEFINITION 2.1 (Seminorm). Let X be a vector space over field F. We define a

seminorm on X, denoted by ν, to be a map from X to R that satisfies the following
conditions.

1. Nonnegativity: ∀x ∈ X, ν(x) ≥ 0.

2. Absolute Homogeneity: ∀λ ∈ F, ∀x ∈ X, ν(λx) = |λ|ν(x).

3. Subadditivity: ∀x, y ∈ X, ν(x + y) ≤ ν(x) + ν(y).

Note that the nonnegativity condition is not necessary since it can be implied by the
other two.
The idea behind the seminorm is that we are trying to give our vector space a notion of
“length” of vectors.

DEFINITION 2.2 (Norm). Let X be a vector space over field F. We define a norm
on X, denoted by ν, to be a seminorm on X that satisfies the additional condition:

∀x ∈ X, µ(x) = 0 ⇐⇒ x = 0.

2.2 Properties

11
12 CHAPTER 2. NORMED LINEAR SPACE

PROPOSITION 2.3. Proper subspaces of a normed linear space has empty interior.

Proof. Let X be a normed linear space. Let M be a proper subspace of X. Assume for the
sake of contradiction that M has non-empty interior. Then ∃x0 ∈ M and ∃r > 0 such that
ball(x0 , r) ⊆ M where ball(x0 , r) denotes the open ball centered at point x0 with radius
r
r. Let x be an arbitrary point in X. Define a point y(x) as y(x) := x0 + x. Then
2kxk
2kxk r
x = (y − x0 ). It is easy to verify that ky − x0 k = < r. So y ∈ ball(x0 , r). So
r 2
2kxk
y ∈ M. Since y, x0 ∈ M and M is a subspace, we get (y − x0 ) ∈ M. That is, x ∈ M.
r
So ∀x ∈ X, x ∈ M. So M = X. This contradicts to the assumption that M is a proper
subspace of X. So M has empty interior.

PROPOSITION 2.4. Closed proper subspaces of a normed linear space are nowhere
dense.

Proof. Let X be a normed linear space. Let M be a closed proper subspace of X. Since M
is closed, cl(M) = M. So cl(M) = M is a closed proper subspace of X. Since cl(M) is a
proper subspace of X, int(cl(M)) = ∅. So M is nowhere dense.

PROPOSITION 2.5. Let X be a normed linear space over field F. Then X is

complete if and only if the closed unit ball X1 is complete.

Proof. For one direction, assume that (V, k · k) is complete. We are to prove that (B(0, 1), k ·
kV ) is complete. Since (B(0, 1), k·kV ) is a closed subspace of (V, k·k) and (V, k·k) is complete,
(B(0, 1), k · kV ) is also complete. For the reverse direction, assume that (B(0, 1), k · kV ) is
complete. We are to prove that (V, k · kV ) is complete. Let {xi }i∈N be an arbitrary Cauchy
sequence in (V, k·kV ). Since {xi }i∈N is Cauchy in (V, k·kV ), {xi }i∈N is bounded in (V, k·kV ).
Let λ be a positive upper bound for {kxi kV }i∈N . Since {xi }i∈N is Cauchy in (V, k · kV ),
{xi /λ}i∈N is Cauchy in (B(0, 1), k · kV ). Since {xi /λ}i∈N is Cauchy in (B(0, 1), k · kV )
and (B(0, 1), k · kV ) is complete, {xi /λ}i∈N converges in (B(0, 1), k · kV ). Since {xi /λ}i∈N
converges in (B(0, 1), k · kV ), {xi }i∈N converges in (V, k · kV ). Since any Cauchy sequence
in (V, k · kV ) converges in (V, k · kV ), (V, k · kV ) is complete.

2.3 Equivalence of Norm

2.4. FINITE-DIMENSIONAL NORMED LINEAR SPACE 13

DEFINITION 2.6 (Equivalence of Norm). Let X be a vector space over field F. Let
k · k1 and k · k2 be two norms on V . We say that k · k1 and k · k2 are equivalent if

∃c1 , c2 > 0, ∀v ∈ X, c1 kvk1 ≤ kvk2 ≤ c2 kvk2 .

Or equivalently,
c1 kvk2 ≤ kvk1 ≤ c2 kvk2 .

PROPOSITION 2.7. The equivalence of norms is an equivalence relation.

PROPOSITION 2.8. Let X be a vector space. Let k · k1 and k · k2 be two norms

on X. Then k · k1 and k · k2 are equivalent if and only if they generate the same metric
topology.

Proof. Convergence to 0 is equivalent under either k · k1 or k · k2 . i.e., equivalent norms give

rise to the same set of sequences that are convergent. Convergence of sequences defines the
topology.

PROPOSITION 2.9. Let X be a vector space. Let k · k1 and k · k2 be two norms on

X. Let ι be the identity map from (X, k · k1 ) to (X, k · k2 ). Suppose that k · k1 and k · k2
are equivalent Then the identity map ι is continuous, and in fact, a homeomorphism
between (X, k · k1 ) and (X, k · k2 ).

2.4 Finite-Dimensional Normed Linear Space

PROPOSITION 2.10. A finite-dimensional linear manifold of a normed linear space

is closed.

THEOREM 2.11. Let X be a finite-dimensional normed linear space. Let S be a set

14 CHAPTER 2. NORMED LINEAR SPACE

in X. Then S is norm-compact if and only if S is closed and bounded.

THEOREM 2.12. All norms on a finite-dimensional normed linear space are equiv-
alent.

Proof Approach 1. Let X be a finite-dimensional normed linear space. Let k · k1 and k · k2

be two norms on X. Let B1 denote the closed unit ball under norm k · k1 . Then B1 is closed
and bounded. Since B1 is closed and bounded and X is finite-dimensional, B1 is compact.
Since B1 is compact and k · k2 is continuous, the set V := {kxk2 : x ∈ B1 } ⊆ R+ is compact.
Since V is a compact subset of R+ , it is bounded. So ∃c1 > 0 such that ∀x ∈ B1 , kxk2 ≤ c1 .
i.e., ∀x ∈ X, if kxk1 ≤ 1, then kxk2 ≤ c1 . So kxk2 ≤ c1 kxk1 . So k · k2 ≤ c1 k · k1 . Similarly,
∃c2 > 0 such that k · k1 ≤ c2 k · k2 . So k · k1 and k · k2 are equivalent.

Proof Approach 2. Let k · kp be an arbitrary p-norm on V and k · k be an arbitrary norm

on V . Let B be the standard basis for V . Say B = {e1 , e2 , . . . , en }. Let v be an arbitrary
vector in V .
n
X n
X n
X
kvk = k vi e i k ≤ kvi ei k = |vi |kei k
i=1 i=1 i=1
n
! p1 n
!1− p1
X X p
≤ |vi |p kei k p−1

i=1 i=1
n
!1− p1
X p
= kei k p−1 kvkp
i=1

:= c1 kvkp .

LEMMA 2.13 (Riesz’s Lemma). Let X be a normed linear space. Let Y be a closed
proper subspace of X. Let α be an element of the open interval (0, 1). Then there is
some x0 ∈ X \ Y such that kx0 k = 1 and ∀y ∈ Y, we have kx0 − yk ≥ α.

Proof. Since Y ( X, X \ Y 6= ∅. Let x be an element of X \ Y. Let d denote the distance

/ Y and Y is closed, d > 0. Since α < 1, α−1 > 1.
d := inf{kx − yk : y ∈ Y}. Since x ∈
Since d > 0 and α−1 > 1, we get dα−1 > d. So ∃y0 ∈ Y such that kx − y0 k ≤ dα−1 . Define
x − y0
a point x0 by x0 := . Since y0 ∈ Y and x ∈ X \ Y, we get x0 ∈ X \ Y. It is clear
kx − y0 k
2.5. DUAL NORMS 15

that kx0 k = 1. Let y be an arbitrary element of Y. Then

x − y0 kx − (y0 + kx − y0 ky)k
kx0 − yk = k − yk =
kx − y0 k kx − y0 k
d
≥ , since y0 + kx − y0 ky ∈ Y
kx − y0 k
d
≥ , since kx − y0 k ≤ dα−1
dα−1
= α.

That is, kx0 − yk ≥ α.

THEOREM 2.14. A normed linear space is finite-dimensional if and only if its closed
unit ball is norm-compact.

Proof. Let X be a normed linear space. Let X1 denote the closed unit ball in X.
Forward Direction: Assume that X is finite-dimensional. I will show that X1 is norm-
compact. Since X is finite-dimensional, being compact is equivalent to being closed and
bounded. Clearly X1 is closed and bounded. So X1 is compact.
Backward Direction: Assume that X1 is norm-compact. I will show that X is finite-
dimensional. Assume for the sake of contradiction that X is infinite-dimensional. Since
dim(X) = ∞, X 6= ∅. Let x1 be an element of X such that kx1 k = 1. Define a subspace Y1
by Y1 := span{x1 }. Then Y1 is a finite-dimensional subspace of X. Since dim(Y1 ) < ∞,
Y1 is closed. Since dim(Y1 ) < ∞ and dim(X) = ∞, Y1 ( X.. So Y1 is a closed proper
subspace of X. By the Riesz’s Lemma, there is some x2 ∈ X \ Y1 such that kx2 k = 1
1
and kx2 − x1 k ≥ . Define a subspace Y2 by Y2 := span{x1 , x2 }. Similarly, Y2 is a
2
closed proper subspace of X. By the Riesz’s Lemma, there is some x3 ∈ X \ Y2 such that
1 1
kx3 k = 1 and both kx3 − x1 k ≥ and kx3 − x2 k ≥ . In general, define a subspace Yn
2 2
by Yn := span{xi }ni=1 . Then Yn is a closed proper subspace of X. By the Riesz’s Lemma,
1
there is some xn+1 ∈ X \ Yn such that kxn+1 k = 1 and ∀i = 1..n, we have kxn+1 − xi k ≥ .
2
Define a sequence x ∈ XN by x := (xn )n∈N . Then by the construction of x, we have that
1 1
x ∈ X1 and that ∀i, j ∈ N, kxi − xj k ≥ . Since ∀i, j ∈ N, kxi − xj k ≥ , x contains no
2 2
convergent subsequence. Since x ∈ X1 and x contains no convergent subsequence, X1 is not
compact. This contradicts to the assumption that X1 is compact. So X is finite-dimensional.

2.5 Dual Norms

16 CHAPTER 2. NORMED LINEAR SPACE

DEFINITION 2.15 (Dual Norm). Let (V, k · k) be an normed vector space. We

define the dual norm of k · k, denoted by k · k◦ , to be a function given by

|v · w|
kvk◦ := max v · w = max .
kwk=1 kwk6=0 kwk

PROPOSITION 2.16. Dual norms of norms are indeed norms.

PROPOSITION 2.17. Let (V, k · k) be a normed vector space. Let v, w be vectors

in the space. Then
|v · w| ≤ kvk · kwk◦ .

PROPOSITION 2.18. Let p be an arbitrary number in [1, +∞). Then the dual
norm of the p-norm k · kp is the q-norm k · kq where q is such that satisfies

1 1
+ = 1.
p q
Chapter 3

Inner Product Space

3.1 Inner Products

3.1.1 Definitions

DEFINITION 3.1 (Inner Product). Let V be a vector space over field F. We define
an inner product on V , denoted by h·, ·i, to be a scalar-valued function defined on
V × V such that

1. Positive Definiteness:

∀x ∈ V, hx, xi ≥ 0 and hx, xi = 0 ⇐⇒ x = 0.

2. Sesqui-Linearity:

∀x, y, z, w ∈ V, hx + y, z + wi = hx, zi + hy, zi + hx, wi + hy, wi, and

∀a, b ∈ F, ∀x, y ∈ V, hax, byi = abhx, yi.

3. Conjugate Symmetry:
∀x, y ∈ V, hx, yi = hy, xi.

DEFINITION 3.2 (Induced Norm). Let X be an inner product space over field K.
We define the norm induced by h·, ·i, denoted by k · k, to be a function from X to R+
given by
p
kxk := hx, xi

17
18 CHAPTER 3. INNER PRODUCT SPACE

3.1.2 Examples of Inner Products

DEFINITION 3.3 (Standard Inner Product). For V = Fn , we define the standard

inner product by
n
X
hx, yi := xi yi .
i=1

DEFINITION 3.4 (Frobenius Inner Product). For V = Fn×n , we define the Frobe-
nius inner product by
hM1 , M2 i := tr(M2∗ M1 ).

DEFINITION 3.5. Let V be the space of continuous scalar-valued functions on

[0, 2π]. We define the inner product on V by
Z 2π
1
hf, gi := f (x)g(x)dx.
2π 0

3.1.3 Properties

PROPOSITION 3.6. Let V be a finite dimensional inner product space. Let B be

a basis for V . Let x and y be vectors in V . Then

x = y ⇐⇒ ∀b ∈ B, hx, bi = hy, bi.

3.2 Inner Product Space

DEFINITION 3.7 (Inner Product Space). Let X be a vector space over field F. Let
h·, ·i be an inner product on X. We define an inner product space to be the pair
(X, h·, ·i).

3.3 Inequalities
3.3. INEQUALITIES 19

THEOREM 3.8 (Minkowski).

n
! p1 n
! p1 n
! p1
X X X
p p p
|xi + yi | ≤ |xi | + |yi |
i=1 i=1 i=1

PROPOSITION 3.9 (Cauchy-Schwarz Inequality). Let V be an inner product space.

Then
∀x, y ∈ V, |hx, yi| ≤ kxk · kyk

PROPOSITION 3.10 (Triangle Inequality). Let V be an inner product space. Then

∀x, y ∈ V, kx + yk ≤ kxk + kyk

PROPOSITION 3.11 (Parallelogram Law). Let X be an inner product space. Then

∀x, y ∈ X, kx + yk2 + kx − yk2 = 2kxk2 + 2kyk2 .

Proof.

kx + yk2 + kx − yk2 = hx + y, x + yi + hx − y, x − yi
= hx, xi + hx, yi + hy, xi + hy, yi
+ hx, xi − hx, yi − hy, xi + hy, yi
= 2hx, xi + 2hy, yi
= 2kxk2 + 2kyk2 .

That is,
kx + yk2 + kx − yk2 = 2kxk2 + 2kyk2 .
20 CHAPTER 3. INNER PRODUCT SPACE
Chapter 4

Orthogonality

4.1 Orthogonal Sets

DEFINITION 4.1 (Orthogonality). Let V be an inner product space. We say that

points x and y in V are orthogonal if hx, yi = 0.

DEFINITION 4.2 (Orthogonal Set). Let X be an inner product space. Let S be a

subset of X. We say that S is orthogonal if

∀x, y ∈ S : x 6= y, hx, yi = 0.

i.e., if any two distinct vectors are orthogonal.

DEFINITION 4.3 (Orthonormal Set). Let X be an inner product space. Let S be

a set in the space. We say that S is orthonormal if S is orthogonal and ∀x ∈ S,
kxk = 1 where k · k is the norm induced by the inner product.

PROPOSITION 4.4. Orthogonal sets are linearly independent.

4.2 Orthogonal Bases

21
22 CHAPTER 4. ORTHOGONALITY

DEFINITION 4.5 (Orthogonal Basis). Let V be an inner product space. Let S be a

set in the space. We say that S is an orthogonal basis for V if it is an ordered basis
for V and orthogonal.

DEFINITION 4.6 (Orthonormal Basis). Let X be an inner product space. Let S be

a set in the space. We say that S is an orthonormal basis for X if it is the maximal,
with respect to inclusion, in the collection of all orthonormal sets in the space.

PROPOSITION 4.7. Let V be an inner product space. Let S = {v1 , ..., vn } be an

orthogonal subset of V where each vi is non-zero. Then
n
X hy, vi i
∀y ∈ span(S), y= vi .
i=1
kvi k2

THEOREM 4.8 (Gram-Schmidt Process). Let V be an inner product space. Let

S = {x0 , ..., xn } be a linearly independent subset of V . Then the set T = {y0 , ..., yn }
given by y0 := x0 and
i−1
X hxi , yj i
∀i ∈ {1, ..., n}, yi := xi − yj
j=1
kyj k

is an orthogonal subset of V consisting of non-zero vectors such that span(S) =

span(S 0 ).

PROPOSITION 4.9. Let V be an inner product space and S = {v0 , v1 , . . . , vn } be

0
an orthogonal subset of V . Then the set S derived from the Gram-Schmidt process is
exactly S.

THEOREM 4.10 (Parseval’s Identity). Let V be a finite-dimensional inner product

4.3. ORTHOGONAL COMPLEMENTS 23

space. Let B = {v1 , ..., vn } be an orthogonal basis for V . Then

n
X
∀x, y ∈ V, hx, yi = hx, vi ihy, vi i.
i=1

PROPOSITION 4.11. Let H be a Hilbert space. Let E be an orthonormal set in

H. Then E is an orthonormal basis for H if and only if
X
∀x ∈ H, x= hx, eie.
e∈E

4.3 Orthogonal Complements

DEFINITION 4.12 (Orthogonal Complement). Let X be an inner product space.

Let S be a non-empty subset of V . We define the orthogonal complement of S,
denoted by S ⊥ , to be a set given by

S ⊥ := {x ∈ X : ∀s ∈ S, hx, si = 0}.

i.e., the set of all points in X that are orthogonal to all vectors in S.

PROPOSITION 4.13. Let V be a finite-dimensional inner product space. Then

1. V ⊥ = {OV }
⊥
2. {OV } = V

PROPOSITION 4.14. Orthogonal complements are always linear subspaces.

PROPOSITION 4.15. Let V be an inner product space and W be a subspace of V

with basis β. Then a vector in V is also in W ⊥ if and only if it is orthogonal to all
vectors in β.
24 CHAPTER 4. ORTHOGONALITY

PROPOSITION 4.16 (Extension). Let V be an n-dimensional inner product space

and S = {v1 , v2 , . . . , vk } be an orthogonal subset of V . Then S can be extended to an
orthogonal basis B = {v1 , v2 , . . . , vk , vk+1 , . . . , vn } for V .

PROPOSITION 4.17. Let V be an inner product space. Then

1. S ⊆ T implies T ⊥ ⊆ S ⊥ for any subsets S and T of V .

2. S ⊆ (S ⊥ )⊥ for any subset S of V .

PROPOSITION 4.18. Let V be a finite-dimensional inner product space and W be

a subspace of V . Then

1. W = (W ⊥ )⊥

2. V = W ⊕ W ⊥

PROPOSITION 4.19. Let V be a finite-dimensional inner product space and W1

and W2 be subspaces of V . Then

1. (W1 + W2 )⊥ = W1⊥ ∩ W2⊥

2. (W1 ∩ W2 )⊥ = W1⊥ + W2⊥

4.4 Orthogonal Projection

DEFINITION 4.20 (Orthogonal Projection). Let V be a vector space. Let W be a

finite-dimensional subspace of V . Let x be a vector in V . We define the orthogonal
projection of x on W , denoted by (x), to be the vector u in W such that x = u + v
where v is another vector in W ⊥ .

4.5 Inequalities in Hilbert Spaces

4.5. INEQUALITIES IN HILBERT SPACES 25

THEOREM 4.21 (Bessel’s Inequality). Let H be a Hilbert space. Let E be an

orthonormal set in the space. Then
X
∀x ∈ H, |hx, ei i|2 ≤ kxk2 .
e∈E

PROPOSITION 4.22. Let H be a Hilbert space. Let E be an orthonormal set in

X
H. Let x be a point in the space. Then the net hx, eie converges in H.
e∈E

Proof. Let F be the collection of all finite subsets of E, partially ordered by inclusion. Define
X
for each F ∈ F a vector yF as yF := hx, eie. Let ε be an arbitrary positive number.
e∈F
Since E is an orthonormal set, the set {e ∈ E : hx, ei =
6 0} is countable. Let {ei }i∈N denote
X∞
the set. By the Bessel’s inequality, ∃N ∈ N such that |hx, ei i|2 < ε2 . Define a set
i=N +1
F0 as F0 := {e1 , ..., eN }. Let F and G be arbitrary elements in F such that F0 ≤ F and
F0 ≤ G. Then
2
X X
kyF − yG k2 = hx, eie − hx, eie
e∈F \G e∈G\F
X
= |hx, ei|2
e∈F ∪G\F ∩G
X X
= |hx, ei|2 − |hx, ei|2
e∈F ∪G e∈F ∩G
X∞
≤ |hx, ei i|2
i=N +1

< ε2 .

So {yF }F ∈F is Cauchy. Since H is complete and {yF }F ∈F is Cauchy, {yF }F ∈F converges.

26 CHAPTER 4. ORTHOGONALITY
Chapter 5

Sequence Space

5.1 p-norms

DEFINITION 5.1 (p-norm). Let V be a finite-dimensional normed vector space over

field F. Let B = {b1 , ..., bn } be a basis for V where n = dim(V ). Let v be a vector in
a normed vector space. For p ∈ [1, +∞), we define the p-norm of v, denoted by kvkp ,
to be the number given by

n
! p1
X
kvkp = |(vB )i |p .
i=1

DEFINITION 5.2 (Infinity Norm - 1). Let X = Kn where K is a field and n ∈ N.

We define the infinity norm on X, denoted by k · k∞ , to be a function given by

kvk∞ := max{|vi |}ni=1 .

DEFINITION 5.3 (Infinity Norm - 2). Let X = KN . We define the infinity norm
on X, denoted by k · k∞ , to be a function given by

kvk∞ := sup |vi |.

i∈N

27
28 CHAPTER 5. SEQUENCE SPACE

DEFINITION 5.4 (Infinity Norm - 3). Let X = C([0, 1], C). We define the infinity
norm on X, denoted by k · k∞ , to be a function given by

kf k∞ := sup |f (x)|.
x∈[0,1]

PROPOSITION 5.5. Let X := C([0, 1], C). Let x be an arbitrary number in [0, 1].
Define a function νx on X by νx (f ) := |f (x)|. Define a function ν on X by ν(f ) :=
sup |f (x)|. Then νx is a seminorm on X for each x and ν is a norm on X and we have
x∈[0,1]
ν = sup ν.
x∈[0,1]

PROPOSITION 5.6. p-norms are indeed norms.

PROPOSITION 5.7. For any vector v in Rn , we have

lim kvkp = kvk∞ .

p→∞

i.e.,
n
! p1
X
p
lim |vi | = max{|vi |}ni=1 .
p→∞
i=1

Proof. Let p be an arbitrary number in [1, +∞). Let k be an arbitrary index in {1, .., n}.
Then
Xn
|vk | ≤ ( |vk |p )1/p = kvkp .
i=1
So
max{|vk |} = kvk∞ ≤ kvkp .

So
lim kvkp ≥ kvk∞ . (1)
p→∞

On the other hand, note that

n n
X
p
X |vi | p
( |vi | )/kvkp∞ = ( )
i=1 i=1
kvk∞
5.2. `P SPACE 29

decreases as p increases. So it is bounded above. Say

n
X
( |vi |p )/kvkp∞ ≤ C
i=1

for some C ∈ R. Then

n
X
( |vi |p )1/p = kvkp ≤ C 1/p kvk∞ .
i=1

So
lim kvkp ≤ lim C 1/p kvk∞ = kvk∞ . (2)
p→∞ p→∞

From (1) and (2) we get

lim kvkp = kvk∞ .
p→∞

PROPOSITION 5.8. Let p and q be numbers in [1, +∞]. Let v be a vector in Rn .

Then if p ≤ q,
1 1
kxkq ≤ kxkp ≤ n p − q · kxkq .

5.2 `p Space

DEFINITION 5.9 (`p Space). We define the `p space to be the set of all scalar
sequences x such that kxkp is finite, equipped with the p-norm k · kp .

DEFINITION 5.10 (Weighted `p Space). Let (ri )i∈N be a sequence of positive

integers. We define the weighted `p space to be the set given by
X
`p := (xi )i∈N ∈ KN : ri |xi |p < +∞.


i∈N

PROPOSITION 5.11. For p ∈ [1, +∞), (`p , k · kp ) is complete.

30 CHAPTER 5. SEQUENCE SPACE

Proof.
Let {xn }n∈N be an arbitrary Cauchy sequence in `p .
Since {xn }n∈N is Cauchy in `p , ∀ε > 0, ∃N (ε) ∈ N such that ∀m, n > N , we have
kxm − xn kp < ε.
Since kxm − xn kp < ε and |x(i) (i) (i) (i)
m − xn | ≤ kxm − xn kp for any i∈ N, |xm − xn | < ε for
any i∈ N.
Since for any i∈ N and any positive number ε, there exists an integer N (ε) such that for
any indices m, n > N , we have |x(i) (i) (i)
m − xn | < ε, by definition, {xn }n∈N is Cauchy in F.
Since {x(i) (i)
n }n∈N is Cauchy in F and F is complete, {xn }n∈N converges.
(i) (i)
Let x0 = x(i)
n . Let x0 = {x0 }i∈N .

∞
X (i) 1
kx0 kp = ( |x0 |p ) p
i=1

5.3 c0 Space and c00 Space

DEFINITION 5.12 (c0 Space). We define c0 to be

c0 := {xn }n∈N ∈ RN : lim xn = 0 .


n→∞

DEFINITION 5.13 (c00 Space). We define c00 to be

c00 := (xn )n∈N ∈ RN : ∃N ∈ N, ∀n > N, xn = 0 .



i.e., the set of all eventually zero sequences of real numbers. i.e., the set of all sequences
with finite support.

PROPOSITION 5.14. The c00 is not complete in (`1 , k · k1 ).

1
Proof. Define a sequence of vectors (xi )i∈N by xji := for j ∈ {1..i} and xji := 0 for j > i.
j2
Then (xi )i∈N converges to something that is not in c00 .
5.3. C0 SPACE AND C00 SPACE 31

PROPOSITION 5.15. The closure of c00 in the space (Rω , d1 ) is `1 .

Proof. For one direction, we are to prove that cl(c00 ) ⊆ `1 . Let x be an arbitrary element
in cl(c00 ). Since x ∈ cl(c00 ), there exists another element y ∈ c00 such that d1 (x, y) < 1.
Let N ∈ N be such that ∀n > N , yn = 0. Then

d1 (x, y) < 1
X
⇐⇒ |xn − yn | < 1
n∈N
N
X X
⇐⇒ |xn − yn | + |xn − yn | < 1
n=1 n>N
N
X X
⇐⇒ |xn − yn | + |xn | < 1
n=1 n>N
N
X X
=⇒ |xn | − |yn | + |xn | < 1
n=1 n>N
N
X
X
=⇒ |xn | − |yn | + |xn | < 1
n=1 n>N
N
X N
X X
=⇒ |xn | − |yn | + |xn | < 1
n=1 n=1 n>N

X N
X
⇐⇒ |xn | − |yn | < 1
n∈N n=1

X N
X
⇐⇒ |xn | < 1 + |yn |.
n∈N n=1
X
Since |xn | is bounded, x ∈ `1 .
n∈N
For the reverse direction, we are to prove that `1 ⊆ cl(c00 ). Let x be an arbitrary element
in `1 . For i ∈ N, define xi = {xij }j∈N as xij = xj for j ≤ i and xij = 0 for j > i. Then
∀i ∈ N, xi ∈ c00 . Then

lim d1 (xi , x)
i∈N
X
= lim |xij − xj |
i∈N
j∈N
X
= lim |xij − xj |
i∈N
j>i
X
= lim |xj |
i∈N
j>i
32 CHAPTER 5. SEQUENCE SPACE

= 0.

That is, lim d1 (xi , x) = 0. So lim xi = x. So x ∈ cl(c00 ).

i∈N i∈N

PROPOSITION 5.16. The closure of c00 in the space (Rω , d∞ ) is c0 .

Proof. For one direction, we are to prove that cl(c00 ) ⊆ c0 . Let x be an arbitrary element in
cl(c00 ). Let ε be an arbitrary positive real number. Since x ∈ cl(c00 ), there exists another
element y in c00 such that d∞ (x, y) < ε. That is, ∀j ∈ N, |xj − yj | < ε. Since y ∈ c00 ,
∃N ∈ N such that ∀j > N , yj = 0. So ∀j > N, |xj | < ε. That is,

∀ε > 0, , ∃N ∈ N, ∀j > N, |xj | < ε.

By definition of convergence of limits, lim xj = 0. So x ∈ c0 .

j∈N
For the reverse direction, we are to prove that c0 ⊆ cl(c00 ). Let x be an arbitrary element
in c0 . For i ∈ N, define xi as xij = xj for j ≤ i and xij = 0 for j > i. Then ∀i ∈ N, xi ∈ c00 .
Let ε be an arbitrary positive real number. Since x ∈ c0 ,

∃N ∈ N, ∀j > N, |xj | < ε/2.

Let i > N . Then

d∞ (xi , x)
= sup |xij − xj |
j∈N

= sup |xij − xj |
j>i

= sup |xj |
j>i

≤ ε/2 < ε.

That is,
∀ε > 0, ∃N ∈ N, ∀i > N, d∞ (xi , x) < ε.
By definition of convergence of sequences, lim xi = x. So x ∈ cl(c00 ).
i∈N

 X
PROPOSITION 5.17. Let A := {xn }n∈N ∈ c00 : xn = 0 . Then A is a subset
n∈N
of `1 and is closed in (`1 , d1 ). i.e. cl(A) = A in (`1 , d1 ).
5.3. C0 SPACE AND C00 SPACE 33

Proof. Let x = {xi }i∈N be a sequence in `1 , where each xi = {xij }j∈N is an element in A,
that converges in (`1 , d1 ). Say lim xi = x∞ .
i→∞
First I claim that x∞ ∈ c00 .
X
Now I claim that x∞j = 0. i.e. x
∞
∈ A. Since x∞ ∈ c00 ,
j∈N

∃N ∈ N, ∀j > N, x∞
j = 0.

N
X N
X
Define yi := xij . Define y∞ := x∞
j . It is easy to see that lim yi = y∞ . Assume for
i∈N
j=1 j=1
the sake of contradiction that y∞ 6= 0. i.e. {yi }i∈N does not converge to 0. Then

∃ε0 > 0, ∀M ∈ N, ∃i0 > M, |yi0 − 0| = |yi0 | ≥ ε0 . (1)

Since lim xi = x∞ ,
i→∞
∃M0 ∈ N, ∀i > M0 , d1 (xi , x∞ ) < ε0 . (2)

Consider statement (1) for a particular M , M0 , we have

∃i0 > M0 , |yi0 | ≥ ε0 . (3)

That is,
N
X
xij0 ≥ ε0 . (3’)
j=1

Consider statement (2) for a particular i, i0 , we have

d1 (xi0 , x∞ ) < ε0 . (4)

From statement (4) we can derive:

d1 (xi0 , x∞ ) < ε0
X
⇐⇒ |xij0 − x∞j | < ε0
j∈N
N
X X
⇐⇒ |xij0 − x∞
j |+ |xij0 − x∞
j | < ε0
j=1 j>N
X
=⇒ |xij0 − x∞
j | < ε0
j>N
X
⇐⇒ |xij0 − 0| < ε0
j>N
X
⇐⇒ |xij0 | < ε0
j>N
34 CHAPTER 5. SEQUENCE SPACE
X
=⇒ xij0 < ε0
j>N
X
=⇒ xij0 < ε0
j>N

X N
X
⇐⇒ xij0 − xij0 < ε0
j∈N j=1
N
X
⇐⇒ 0 − xij0 < ε0
j=1
N
X
⇐⇒ xij0 < ε0 .
j=1

This contradicts to statement (3’). So the original assumption that y∞ 6= 0 is false. i.e.
X
y∞ = 0. It follows that x∞
j = 0. This completes the proof.
j∈N

5.4 Hölder’s Inequality

THEOREM 5.18 (Hölder’s Inequality). Let X = Rn for some n ∈ N. Let x = (xi )ni=1
and y = (yi )ni=1 be vectors in X. Then ∀p, q ∈ (1, +∞) : 1/p + 1/q = 1, kxyk1 ≤
kxkp kykq . i.e.,
n
X n
X Xn
|xi yi | ≤ ( |xi |p )1/p ( |yi |q )1/q .
i=1 i=1 i=1
Chapter 6

Function Spaces

6.1 The Lp Norm

! p1
Z b
p
kf kp = |f (x)| dx .
a

35
36 CHAPTER 6. FUNCTION SPACES
Chapter 7

Quotient Space

7.1 Definitions

DEFINITION 7.1 (Quotient Space). Let V be a vector space. Let W be a subspace

of V. We define a quotient space of V mod W, denoted by V/W, to be a set given
by {v + W : v ∈ V} with addition and scalar multiplication defined by

(v1 + W) + (v2 + W) := (v1 + v2 ) + W and

κ(v + W) := (κv) + W.

DEFINITION 7.2 (The Canonical Quotient Map). Let X be a vector space. Let M
be a linear manifold in X. We define the canonical quotient map on X with respect
to M, denoted by qM , to be a function from X to X/M given by

qM (x) := [x] = x + M

7.2 Properties

PROPOSITION 7.3. Let V be a finite dimensional vector space over field F. Let
W be a finite dimensional subspace of V. Then we have

dim(V/W) = dim(V) − dim(W).

37
38 CHAPTER 7. QUOTIENT SPACE

Proof. Let BW = {w1 , ..., wm } be a basis for W. Extend BW to a basis BV = {w1 , ..., wm , v1 , ..., vn }
for V. Now I claim that B := {v1 + W, ..., vn + W} is a basis for V/W. Clearly B is linearly
independent and span(B) ⊆ V/W. So there remains only to show that V/W ⊆ span(B).
Let x + W be an arbitrary element of V/W where x ∈ V. Then ∃λ1 , ..., λm , µ1 , ..., µn ∈ F
such that
m
X n
X
x= λi wi + µi vi .
i=1 i=1

Then
Xm n
X X m  X n 
x+W =( λi wi + µi vi ) + W = ( λi wi ) + W + ( µi vi ) + W
i=1 i=1 i=1 i=1
 X n   Xn
= 0+W + ( µi vi ) + W = ( µi vi ) + W
i=1 i=1
n
X
= µi (vi + W) ∈ span(B).
i=1

So V/W ⊆ span(B). So B is a basis for V/W. So

dim(V/W) = |B| = n.. (1)

On the other hand, we have

dim(V) − dim(W) = |BV | − |BW | = m + n − m = n. (2)

From (1) and (2), we get

dim(V/W) = dim(V) − dim(W),

as desired.

7.3 Quotient Spaces with Seminorms

DEFINITION 7.4 (Seminorm on Quotient Spaces). Let (X, k · k) be a normed linear

space. Let M be a linear manifold of X. We define a seminorm on X/M to be a
function from X/M to R given by

p(x + M) := inf{kx + mk : m ∈ M}.

7.4. QUOTIENT SPACES WITH TOPOLOGIES 39

PROPOSITION 7.5. Seminorms on quotient spaces are indeed seminorms.

PROPOSITION 7.6. A seminorm on a quotient space X/M is a norm if and only if

M is closed.

PROPOSITION 7.7 (Quotient maps are contractive). Let X be a normed linear

space. Let M be a closed linear subspace of X. Then

∀x ∈ X, kx + MkX/M ≤ kxkX .

PROPOSITION 7.8. Let X be a normed linear space. Let M be a closed linear

subspace of X. Let q denote the canonical quotient map from X to X/M. Then q is a
continuous under the norm topology.

Proof. Since q is contractive, q is continuous.

7.4 Quotient Spaces with Topologies

DEFINITION 7.9 (Quotient Toplogy). Let (V, T ) be a topological vector space. Let
W be a closed subspace of V. We define the quotient topology Tq on the quotient
space V/W as
Tq := G ⊆ V/W : q −1 (G) ∈ T .


PROPOSITION 7.10. The canonical quotient map is a continuous map.

PROPOSITION 7.11. The canonical quotient map is an open map.

40 CHAPTER 7. QUOTIENT SPACE
Chapter 8

Dual Space

8.1 Definitions

DEFINITION 8.1 (Linear Functional). Let X be a vector space over field K. We

define a linear functional on X to be a linear map from X to K.

DEFINITION 8.2 (Algebraic Dual). Let X be a vector space over field K. We define
the algebraic dual of X, denoted by X# . to be the vector space of all linear functionals
on X.

DEFINITION 8.3 (Topological Dual). Let X be a topological vector space over field
K. We define the topological dual of X, denoted by X∗ , to be the vector space of all
continuous linear functionals on X.

PROPOSITION 8.4. Let X be a normed linear space. Then there exists a contractive
map from X to its double dual X∗∗ .

8.2 Examples

41
42 CHAPTER 8. DUAL SPACE

EXAMPLE 8.5. (c0 (N))∗ is isometrically isomorphic to `1 (N).

EXAMPLE 8.6. (`1 (N))∗ is isometrically isomorphic to `∞ (N).

8.3 Linear Functionals

PROPOSITION 8.7. Let V be a vector space over field K. Let g, f1 ..fn ∈ V # where
n
\
n ∈ N. Then g ∈ span{fi }ni=1 if and only if ker(fi ) ⊆ ker(g).
i=1

n
\
Proof. Forward Direction: Assume that g ∈ span{fi }ni=1 . I will show that ker(fi ) ⊆
i=1
n
X
ker(g). Since g ∈ span{fi }ni=1 , ∃λ1 , ..., λn ∈ K such that g = λi fi . Let x be an arbitrary
i=1
n
\
element of ker(fi ). Then
i=1

n
X n
X
g(x) = λi fi (x) = λi · 0 = 0.
i=1 i=1

n
\
That is, g(x) = 0. So x ∈ ker(g). So ker(fi ) ⊆ ker(g).
i=1
n
\
Backward Direction: Assume that ker(fi ) ⊆ ker(g). I will show that g ∈ span{fi }ni=1 .
i=1
Assume without loss of generality that {fi }ni=1 are linearly independent. Define a set N by
n
\
N := ker(fi ). Then dim(V/N ) ≤ n. It follows that dim(V/N )# ≤ n. Define for each
i=1
i = 1..n a function Fi : V/N → K by Fi (x + N ) := fi (x). Then clearly each Fi is linear
and hence Fi ∈ (V/N )# . Since f1 , ..., fn are linearly independent, F1 , ..., Fn are linearly
independent. Since (V/N )# contains a linearly independent set of size n, dim(V/N )# ≥ n.
So dim(V/N )# = n. So {Fi }ni=1 is a basis for (V/N )# . Define a function G : V/N → K by
G(x + N ) := g(x). Then clearly, G is linear and hence G ∈ (V/N )# . Since G ∈ (V/N )#
n
X
n #
and {Fi }i=1 is a basis for (V/N ) , ∃λ1 , ..., λn ∈ K such that G = λi Fi . Then ∀x ∈ V,
i=1
8.4. THE DUAL SPACE OPERATOR 43

we have
n
X n
X
g(x) = G(x + N ) = λi Fi (x + N ) = λi fi (x).
i=1 i=1
n
X
So g = λi fi and hence g ∈ span{fi }ni=1 .
i=1

PROPOSITION 8.8. Let V be a topological vector space over field K. Let ρ ∈ V # .

Then ρ ∈ V ∗ if and only if ker(ρ) is a closed set.

Proof. Forward Direction: Assume that ρ ∈ V ∗ . I will show that ker(ρ) is closed. Notice
{0} is closed in K. Since ρ ∈ V∗ , ρ is continuous. So ρ−1 ({0}) is closed. Note that
ρ−1 ({0}) = ker(ρ). So ker(ρ) is closed.
Backward Direction: Assume that ker(ρ) is a closed set. I will show that ρ ∈ V ∗ . If
ρ = 0, then we are done. Otherwise, assume that ρ 6= 0. Define a map ϕ : V/ ker(ρ) → K by
ϕ(x + ker(ρ)) := ρ(x). Then clearly ϕ is linear. Since dim(V/ ker(ρ)) = 1 and dim(K) = 1,
ϕ is continuous. Let q denote the canonical quotient map from V to V/ ker(ρ). Then q is
continuous. Note that ρ = ϕ ◦ q. So ρ is continuous.

8.4 The Dual Space Operator

PROPOSITION 8.9. Let V be a vector space. Suppose that V ∗ is separable. Then

V is also separable.

REMARK 8.10. Note that `1 (N) is separable but its dual `∞ (N) is not. So the
converse of the above is false.

8.5 Annihilator

DEFINITION 8.11 (Annihilator). Let X be a normed linear space. Let M be a

subset of X. We define the annihilator of M, denoted by M0 , to be the subset of X∗
44 CHAPTER 8. DUAL SPACE

given by M0 := {x∗ ∈ X∗ : x∗ |M = 0}.

DEFINITION 8.12 (Pre-Annihilator). Let X be a normed linear space. Let N be a

subset of X∗ . We define the pre-annihilator of N, denoted by 0 N, to be the subset
of X given by 0 N := {x ∈ X : x̂|N = 0}.

PROPOSITION 8.13. The annihilator operator does not distinguish a set from its
closure.

PROPOSITION 8.14. Annihilators are weakly closed.

PROPOSITION 8.15. Pre-annihilators are normed closed.

PROPOSITION 8.16. Let X be a Banach space. Let M be a closed subspace of

X. Let q denote the canonical quotient map. Define a map Θ from (X/M)∗ to M0 by
Θ(ξ) := ξ ◦ q. Then Θ is an isometric isomorphism.

PROPOSITION 8.17. Let X be a Banach space. Let M be a closed subspace of X.

Define a map Θ from X∗ /M0 to M∗ by Θ(x∗ + M0 ) := x∗ |M .
Chapter 9

Banach Space

9.1 Definition

DEFINITION 9.1 (Banach Space). Let (X, k · k) be a normed linear space. Let d be
the metric induced by k · k. We say that X is a Banach space if (X, d) is a complete
metric space. i.e., we define a Banach space to be a complete normed linear space.

9.2 Examples

EXAMPLE 9.2. (C([0, 1], F), k · k∞ ) is a Banach space.

EXAMPLE 9.3 (Disc Algebra). Define D := {z ∈ C : |z| < 1}. Define A(D) := {f ∈
C(D) : f |D is holomorphic }. Define k · k∞ by kf k∞ := sup |f (z)|. Then (A(D), k · k∞ )
z∈D
is a Banach space.

EXAMPLE 9.4. Let (X, Ω, µ) be a measure space. Let p be a number in [1, +∞).
Define
Z
p
L (X, µ) := span{f : X → [0, +∞] | f is measurable and |f |p < +∞}.
X

45
46 CHAPTER 9. BANACH SPACE

Define an equivalence relation on Lp (X, µ) by f ≡ g if and only if

µ({x ∈ X : f (x) 6= g(x)}) = 0.

Define a space Lp (X, µ) := Lp (X, µ)/ ≡. Then Lp (X, µ) is a Banach space when
equipped with the norm
Z 1/p
k[f ]kp := |f |p .
X

EXAMPLE 9.5. Let PC [0, 1] denote the set of all polynomials with complex coeffi-
cients. For each p ∈ [1, +∞), define a norm
Z 1 1/p
p
kf kp := |f | .
0

For p = +∞, define a norm

kf k∞ := sup |f (z)|.
x∈[0,1]

9.3 Properties

PROPOSITION 9.6. Let (X, k · k) be a normed vector space over field F. Then
(X, k · k) is a Banach space if and only if every absolutely summable series in X is
summable.

Proof. Forward Direction: Suppose that X is a Banach space. I will show that any
X
absolutely summable series in X is summable. Let
xn be an absolutely summable series.
n∈N
X n
X
i.e., kxn k < +∞. Define for each n ∈ N a vector yn as yn := xi . Let ε > 0 be
n∈N i=1
∞
X
arbitrary. Then ∃N ∈ N such that ∀n > N , kxi k < ε. Let n > m > N be arbitrary.
i=n
Then
n
X m
X n
X
kyn − ym k = k xi − xi k = k xi k
i=1 i=1 i=m+1
Xn ∞
X
≤ kxi k < kxi k
i=m+1 i=m+1
9.3. PROPERTIES 47

< ε.

That is, kyn − ym k < ε. So (yn )n∈N is Cauchy. Since X is a Banach space and (yn )n∈N is
X
Cauchy, it converges. So xn is summable.
n∈N
Backward Direction: Suppose that every absolutely summable series in X is summable.
I will show that X is a Banach space. Let (yn )n∈N be an arbitrary Cauchy sequence in X.
1
Then ∀n ∈ N, ∃Nn ∈ N such that ∀k, l ≥ Nn , kyk − yl k < n . Assume that N1 < N2 < ....
2
Define x1 := yN1 . Define for each n ∈ N a vector xn+1 as xn+1 := yNn+1 − yNn . Then

∞ ∞ ∞
X X X 1
kxn k = kx1 k + kyNn+1 − yNn k < kx1 k + n
n=1 n=1 n=1
2

= kx1 k + 1 < +∞.

X
So xn is absolutely summable. By assumption, it is summable. i.e., (yn )n∈N converges.
n∈N
Since any Cauchy sequence in X converges, X is complete and hence a Banach space.

PROPOSITION 9.7 (Quotient Spaces of Banach Spaces are Banach Spaces). Let X
be a Banach space. Let M be a closed subspace of M. Then the quotient space X/M
is also a Banach space.

Proof. Proof Approach 1.

Let (q(xn ))n∈N be an arbitrary Cauchy sequence in X/M. We are to prove that it
converges.

Proof. Proof Approach 2.

X
Let q denote the canonical quotient map. Let q(xn ) be an arbitrary absolutely
n∈N
summable series in X /M. Since kq(xn )k is defined to be kq(xn )k := inf{kxn +mk : m ∈ M},
1
∃mn ∈ M such that kxn + mn k < kq(xn )k + n . Then
2
∞ ∞   X∞ ∞
X X 1 X 1
kxn + mn k = kq(xn )k + n = kq(xn )k +
n=1 n=1
2 n−1 n=1
2n
∞
X
= kq(xn )k + 1 < +∞.
n=1
48 CHAPTER 9. BANACH SPACE
X X
So (xn + mn ) is absolutely summable. Since X is a Banach space, (xn + mn ) is
n∈N n∈N
X
summable. Say (xn + mn ) = x• . Then
n∈N

∞
X ∞
X N
X N
X
q(xn ) = q(xn + mn ) = lim q(xn + mn ) = lim q( (xn + mn ))
N →∞ N →∞
n=1 n=1 n=1 n=1
N
X
= q( lim (xn + mn )) = q(x• ).
N →∞
n=1
X
So q(xn ) is summable. Since any absolutely summable series in X/M is summable,
n∈N
X /M is complete.

PROPOSITION 9.8. Let X be a normed linear space. Let M be a closed subspace

of X. If M and X/M are both complete, then X is a Banach space.

Proof. Let (xn )n∈N be an arbitrary Cauchy sequence in X. We are to prove that it converges.
Let q denote the canonical quotient map. Since (xn )n∈N is Cauchy in X, (q(xn ))n∈N is
Cauchy in X/M. Since X/M is a Banach space and (q(xn ))n∈N is Cauchy, (q(xn ))n∈N
converges. Say lim q(xn ) = q(x• ) for some x• ∈ X. By definition of norms in the quotient
n∈N
1
space, for n ∈ N, we can choose mn ∈ M such that kx• − xn − mn k ≤ kq(x• ) − q(xn )k + .
n
So
1
lim kx• − xn − mn k ≤ lim kq(x• ) − q(xn )k + lim = 0 + 0 = 0.
n∈N n∈N n∈N n
So (xn + mn )n∈N converges to x• . So (xn + mn )n∈N is Cauchy. Since (xn )n∈N and (xn +
mn )n∈N are both Cauchy, (mn )n∈N is Cauchy. Since M is a Banach space and (mn )n∈N is
Cauchy, (mn )n∈N converges. Say lim mN = m• . So
n∈N

lim xn = lim ((xn + mn ) − mn ) = lim (xn + mn ) − lim mn

n∈N n∈N n∈N n∈N

= x• − m• .

So (xn )n∈N converges. Since any Cauchy sequence in X converges, X is a Banach space.

PROPOSITION 9.9 (Dual Spaces of Banach Spaces are Banach Spaces). Let X be
a Banach space. Then the dual space X∗ is a also a Banach space.
9.4. DIRECT SUMS AND DIRECT PRODUCTS OF BANACH SPACES 49

PROPOSITION 9.10. Any Banach space with a Schauder basis has to be separable.

9.4 Direct Sums and Direct Products of Banach Spaces

DEFINITION 9.11. Let (X, k · kX ) and (Y, k · kY ) be two Banach spaces over field
K. Let p ∈ [1, +∞). We define

X ⊕p Y := {(x, y) : x ∈ X, y ∈ Y}

and
k(x, y)kp := (kxkpX + kykpY )1/p .

For p = +∞, we define

X ⊕∞ Y := {(x, y) : x ∈ X, y ∈ Y}

and
k(x, y)k∞ := max(kxkX , kykY ).

• Note that the norms behave similarly to what the p norm would do.

• We can similarly define the direct sum of finitely many Banach spaces.

PROPOSITION 9.12. k·, ·kp is a norm on X ⊕p Y.

PROPOSITION 9.13. X ⊕p Y is complete with respect to k·, ·kp .

9.5 Unconditional Convergence in Banach Spaces

DEFINITION 9.14 (Unconditional Convergence). Let X be a Banach space. Let

(xλ )λ∈Λ be a set of vectors in X. Let F be the collection of all finite subsets of Λ,
50 CHAPTER 9. BANACH SPACE
X
partially ordered by inclusion. Define a net (yF )F ∈F on F by yF := xλ . We say
λ∈F
X
that the series xλ is unconditional convergent if the net (yF )F ∈F converges.
λ∈Λ

PROPOSITION 9.15 (Equivalent Formulations of Unconditional Convergence). Let

X be a Banach space. Let (xn )n∈N be a sequence of vectors in X. Then the following
conditions are equivalent.
X
1. For any permutation π of N, xπ(n) converges.
n∈N
X
2. For any subsequence indexing (kn )n∈N , xkn converges.
n∈N
X
3. ∀ε > 0, ∃µ ∈ N such that for all finite subsets F of N\{1..µ}, we have k xn k <
n∈F
ε.

4. ∃y ∈ X such that ∀ε > 0, there is a finite subset F0 of N such that for all finite F
X
such that F0 ⊆ F ⊆ N, we have k xn − yk < ε.
n∈F
X
5. For any sequence (αn )n∈N ∈ {±1}N , αn xn converges.
n∈N
X
6. For any bounded sequence (βn )n∈N , βn xn converges.
n∈N

Proof. Proof of (1) =⇒ (5).

X
Assume that for any permutation π of N, xπ(n) converges. We are to prove that for
n∈N
X
any sequence (αn )n∈N ∈ {±1}N , αn xn converges. Assume for the sake of contradiction
n∈N
X
N
that there is some (αn )n∈N ∈ {±1} such that αn xn diverges. i.e., ∃ε0 > 0 such that
n∈N
∀N ∈ N, ∃kN > lN > N such that
lN
X
k αn xn k ≥ ε0 . (*)
n=kN

For N = 1, find k1 and l1 . For N = l1 , find k2 and l2 . In general, for N = ln , find kn+1
and ln+1 . Then we have k1 < l1 < k2 < l2 < .... For each n, there is an mn ∈ [kn , ln ]
and a permutation πn of [kn , ln ] such that πn (i) ∈ [kn , mn ] if αi = 1 and πn (i) ∈ (mn , ln ] if
αi = −1. Define a permutation π of N as π(i) := i if ∀n ∈ N, i ∈
/ [kn , ln ]; and π(i) := πn (i)
9.5. UNCONDITIONAL CONVERGENCE IN BANACH SPACES 51
X
if i ∈ [kn , ln ]. By assumption, for π, xπ(n) converges. So for ε0 , ∃N ∈ N such that
n∈N
j
X
∀j > i > N , k xn k < ε0 /2. So
n=i

lN
X mN
X lN
X
k αn xn k = k αn xn + αn xn k
n=kN n=kN n=mN +1
mN
X lN
X
=k xn − xn k
n=kN n=mN +1
mN
X lN
X
≤k xn k + k xn k
n=kN n=mN +1

< ε0 /2 + ε0 /2 = ε0 .

That is,
lN
X
k αn xn k < ε0 . (**)
n=kN

Notice (*) and (**) contradict. So the assumption that there is some (αn )n∈N ∈ {±1}N such
X X
that αn xn diverges does not hold. i.e., for any sequence (αn )n∈N ∈ {±1}N , αn xn
n∈N n∈N
converges.

Proof. Proof of (5) =⇒ (2).

X
Assume that for any sequence (αn )n∈N , αn xn converges. We are to prove that for any
n∈N
X
subsequence indexing (kn )n∈N , xkn converges. Let (kn )n∈N be an arbitrary subsequence
n∈N
X X
indexing. Consider (αn )n∈N be given by αn := 1 for all n ∈ N. Then αn xn = xn
n∈N n∈N
converges. Consider (αn )n∈N be given by αn := 1 for n ∈ {ki }i∈N ; and αn := −1 for
X X X
n∈/ {ki }i∈N . Then αn xn = xn − xn converges Notice
n∈N n∈{ki }i∈N n∈{k
/ i }i∈N
 
X 1X 1 X X
xkn = xn +  xn − xn  .
2 2
n∈N n∈N n∈{ki }i∈N n∈{k
/ i }i∈N
X
So xkn converges.
n∈N

Proof. Proof of (2) =⇒ (3).

X
Assume that for any subsequence indexing (kn )n∈N , xkn converges. We are to prove
n∈N
X
that ∀ε > 0, ∃µ ∈ N such that for all finite subsets F of N \ {1..µ}, we have k xn k < ε.
n∈F
52 CHAPTER 9. BANACH SPACE

Assume for the sake of contradiction that ∃ε0 > 0 such that ∀µ ∈ N, there is some finite
X
subset F of N \ {1..µ} such that k xn ≥ ε0 . For µ = 1, find F1 ⊆ N \ {1..µ} finite. For
n∈F
µ = max{F1 }, find F2 ⊆ N\{1..µ} finite. In general, for µ = max{Fn }, find Fn+1 ⊆ N\{1..µ}
finite. Then we have that the Fn ’s are disjoint. Define a subsequence indexing (kn )n∈N as
[ X
(kn )n∈N := Fn . By assumption, for (kn )n∈N , xkn converges. So for ε0 , ∃N ∈ N such
n∈N n∈N
that ∀j > i > N ,
j
X
k xkn k < ε0 . (*)
n=i
So for N , there is some finite subset F of N \ {1..µ} such that
X
k xn k ≥ ε0 .
n∈F

Notice F = {kn }jn=i

N
N
for some iN and jN . So (*) and (**) contradict. So the assumption
that ∃ε0 > 0 such that ∀µ ∈ N, there is some finite subset F of N \ {1..µ} such that
X
k xn ≥ ε0 does not hold. i.e., ∀ε > 0, ∃µ ∈ N such that for all finite subsets F of
n∈F
X
N \ {1..µ}, we have k xn k < ε.
n∈F

Proof. Proof of (3) =⇒ (1).

Assume that ∀ε > 0, ∃µ ∈ N such that for all finite subsets F of N \ {1..µ}, we have
X X
k xn k < ε. We are to prove that for any permutation π of N, xπ(n) converges.
n∈F n∈N
Assume for the sake of contradiction that there is some permutation π of N such that
X lN
X
xπ(n) diverges. i.e., ∃ε0 > 0 such that ∀N ∈ N, ∃lN > kN > N such that k xπ(n) k ≥
n∈N n=kN
ε0 . Let µ be an arbitrary element of N. Define N as N := max{π −1 (n)}µn=1 . For N , find
lN
X
lN > kN > N such that k xπ(n) k ≥ ε0 . Define a set F as F := {π(n)}ln=k
N
N
. So
n=kN
X lN
X
F ⊆ N \ {1..µ}. Then k xn k = k xn k ≥ ε0 . So ∃ε0 > 0 such that ∀µ ∈ N, there
n∈F n=kN
X
is some finite subset F of N \ {1..µ} such that k xn k ≥ ε0 . This contradicts to the
n∈F
X
assumption. So the assumption that there is some permutation π of N such that xπ(n)
n∈N
X
diverges does not hold. So for any permutation π of N, xπ(n) converges.
n∈N

9.6 The Open Mapping Theorem

(bug)
9.6. THE OPEN MAPPING THEOREM 53

LEMMA 9.16. Let X and Y be Banach spaces. Let T be an element of B(X, Y).
Suppose that Y1 ⊆ cl(T Xm ) for some m ≥ 1. Then Y1 ⊆ T X2m .

Proof. Let y be an arbitrary element of Y1 . Then y ∈ cl(T Xm ). So ∃x1 ∈ Xm such that

ky − T x1 k < 1/2. So y − T x1 ∈ Y1/2 . Since Y1 ⊆ cl(T Xm ), we have Y1/2 ⊆ cl(T Xm/2 ).
So y − T x1 ∈ cl(T Xm/2 ). So ∃x2 ∈ Xm/2 such that ky − T x1 − T x2 k < 1/4. In general,
Xn
suppose that we have y − T xi ∈ Y1/2n for some n ∈ N. Since Y1 ⊆ cl(T Xm ), we
i=1
n
X
have Y1/2n ⊆ cl(T Xm/2n ). So y − T xi ∈ cl(T Xm/2n ). So ∃xn+1 ∈ Xm/2n such that
i=1
n+1
X X
ky − T xi k < 1/2n+1 . Then T xn = y. Define a sequence x• in X by x• := (xn )n∈N .
i=1 n∈N
X m X
Since ∀n ∈ N, xn ∈ Xm/2n−1 , we have kxn k ≤
= 2m. So x• is absolutely
2n−1
n∈N n∈N
summable. Since X is a complete space and x• is absolutely summable, x• is summable.
X
Define a point x in X by x := xn . Then
n∈N

X n
X n
X n
X X
kxk = k xn k = k lim xi k = lim k xi k ≤ lim kxi k = kxn k ≤ 2m.
n→∞ n→∞ n→∞
n∈N i=1 i=1 i=1 n∈N

So x ∈ X2m . Now
X n
X n
X n
X X
Tx = T xn = T lim xi = lim T xi = lim T xi = T xn = y.
n→∞ n→∞ n→∞
n∈N i=1 i=1 i=1 n∈N

So ∀y ∈ Y1 , ∃x ∈ X2m such that T x = y. So Y1 ⊆ T X2m .

THEOREM 9.17 (The Open Mapping Theorem). Let X and Y be Banach spaces.
Let T be a surjective bounded linear map from X to Y. Then T is an open map.

Proof. Notice
[ [ [
Y = TX = T Xn = T Xn ⊆ cl(T Xn ).
n∈N n∈N n∈N

Since Y is complete, by the Baire Category Theorem, ∃m ∈ N such that int(cl(T Xm )) 6=

∅.

not finished
54 CHAPTER 9. BANACH SPACE

THEOREM 9.18 (The Inverse Mapping Theorem). Let X and Y be Banach spaces.
Let T be a bijective bounded linear map from X to Y. Then T is a homeomorphism.

THEOREM 9.19 (The Closed Graph Theorem). Let X and Y be Banach spaces.
Let T be a linear map from X to Y. Suppose that the graph G(T ) := {(x, T x) : x ∈ X}
is closed in X ⊕1 Y. Then T is bounded.
Chapter 10

Hilbert Space

10.1 Definition

DEFINITION 10.1 (Hilbert Space). We define a Hilbert space, denoted by H, to

be a complete inner product space.

10.2 Examples

EXAMPLE 10.2. Let (X, µ) be a measure space. Then L2 (X, µ) is a Hilbert space
with inner product given by
Z
hf, gi := f (x)g(x)dµ(x).
X

X
EXAMPLE 10.3. `2 (K) = (xi )i∈N ∈ K∞ : |xi |2 < +∞ is a Hilbert space with


i∈N
inner product given by
X
h(xi )i ∈ N, (yi )i∈N i := xn yn .
i∈N

10.3 Properties

55
56 CHAPTER 10. HILBERT SPACE

PROPOSITION 10.4. Let H be a Hilbert space. Let S be a non-empty set in the

space. Then S ⊥⊥ = clspan(S).

Proof. For one direction, we are to prove that clspan(S) ⊆ S ⊥⊥ .

For the reverse direction, we are to prove that S ⊥⊥ ⊆ clspan(S). Assume for the sake
of contradiction that ∃x ∈ S ⊥⊥ with x 6= 0 such that x ∈
/ clspan(S). Say x = m1 + m2
for some m1 ∈ clspan(S) and some m2 ∈ clspan(S)⊥ . Note that clspan(S)⊥ = S ⊥ . So
m2 ∈ S ⊥ . Since x ∈ S ⊥⊥ and m2 ∈ S ⊥ , we should have hx, m2 i = 0. However,

hx, m2 i = hm1 + m2 , m2 i
= hm1 , m2 i + hm2 , m2 i
= 0 + hm2 , m2 i
> 0, since m2 6= 0.

This leads to a contradiction. So S ⊥⊥ ⊆ clspan(S).

PROPOSITION 10.5 (Stability of Hilbert Spaces Under Quotients). Let H be a

Hilbert space. Let M be a closed subspace of H. Then the quotient space H/M is
again a Hilbert space.

10.4 The Riesz Representation Theorem

THEOREM 10.6 (The Riesz Representation Theorem). Let H be a Hilbert space

over field K ∈ {R, C}. Suppose that H 6= {0}. Then for any ϕ ∈ H∗ , ∃y ∈ H such that

∀x ∈ H, ϕ(x) = hx, yi.

Proof. Define for each y ∈ H a function βy ∈ H∗ by βy (x) := hx, yi. We are to prove that
H∗ = {βy : y ∈ H}. It is easy to verify that each βy is linear and bounded. So ∀y ∈ H,
βy ∈ H∗ . i.e., {βy : y ∈ H} ⊆ H∗ . Define a map Θ from H to H∗ as Θ(y) := βy . It is easy
to verify that Θ is linear.

kΘ(y)k = kβy k = sup{βy (x) : kxk = 1}

= sup{hx, yi : kxk = 1}
10.4. THE RIESZ REPRESENTATION THEOREM 57

≤ sup{kxkkyk : kxk = 1}
= kyk.

That is, kΘ(y)k ≤ kyk. So kΘk ≤ 1. On the other hand, consider an arbitrary point y0 ∈ H
with y0 6= 0:
 
kΘ(y)k
kΘk = sup : y 6= 0
kyk
kΘ(y)k
≥
kyk y=y0
kΘ(y0 )k
=
ky0 k
kβy0 k
=
ky0 k
 
1 kβy0 (y)k
= sup : y 6= 0
ky0 k kyk
1 kβy0 (y0 )k
≥
ky0 k ky0 k
1 ky0 k2
≥
ky0 k ky0 k
= 1.

That is, kΘk ≥ 1. So kΘk = 1. So Θ is isometric. It immediately follows that Θ is injective.

Now it remains to prove that Θ is surjective. Let ϕ ∈ H∗ . If ϕ = 0, then ϕ = Θ(0) and
we are done. Otherwise, let M := ker(ϕ). Then we have codim M = dim M⊥ = 1. Take
e ∈ M⊥ such that kek = 1. Let P denote the orthogonal projection onto M. Then 1 − P
is the orthogonal projection onto M⊥ .

x = P x + (1 − P )x = P x + hx, eie.

So for x ∈ H,
ϕ(x) = ϕ(P x) + hx, eiϕ(e) = hx, ϕ(e)ei = βy (x)

where y := ϕ(e)e. Hence ϕ = βy . So Θ is surjective. This completes the proof.

58 CHAPTER 10. HILBERT SPACE
Chapter 11

Operators

11.1 Bounded Operators

DEFINITION 11.1 (Bounded Operator). Let X and Y be normed linear spaces.

Let T ∈ L(X, Y). We say that T is a bounded operator from X to Y, denoted by
T ∈ B(X, Y), if and only if

∃k ∈ R, ∀x ∈ X, kT xkY ≤ kkxkX .

PROPOSITION 11.2. Let X and Y be normed linear spaces. Let T ∈ L(X, Y).
Then T is bounded if and only if T is continuous.

Proof. Forward Implication: Suppose that T is bounded. I will show that T is continuous.
Since T is bounded, ∃k ∈ R such that ∀x ∈ X, kT xkY ≤ kkxkX . Let x̂ ∈ X be arbitrary.
Let ε > 0 be arbitrary. Define δ := ε/k. Let x ∈ ballX (x̂, δ) be arbitrary. Then

kT x − T x̂kY = kT (x − x̂)kY , since T ∈ L(X, Y)

≤ kkx − x̂kX , by the choice of k
ε
< k · δ = k · = ε.
k

That is, T x ∈ ballY (T x̂, ε). So T is continuous.

Backward Implication: Suppose that T is continuous. I will show that T is bounded.
Since T is continuous, T is continuous at 0, in particular. So for ε = 1, ∃δ > 0 such that

59
60 CHAPTER 11. OPERATORS

∀x ∈ ballX (0, δ), T x ∈ ballY (0, 1). Let k := 2/δ. Let x ∈ X be arbitrary. Then

2kxkX δx
kT xkY = T , since T ∈ L(X, Y)
δ 2kxkX Y
2kxkX δx
= T , by definition of norm
δ 2kxkX Y
2kxkX
≤ · 1, by the choice of δ
δ
= kkxkX .

That is, kT xkY ≤ kkxkX . So T is bounded.

11.2 Operator Norm

DEFINITION 11.3 (Operator Norm). Let X and Y be normed linear spaces. Let
T ∈ B(X, Y). We define the operator norm of T , denoted by kT k, to be the number
given by  
kT k := inf k ∈ R : ∀x ∈ X, kT xkY ≤ kkxkX .

PROPOSITION 11.4 (Equivalent Definition of Operator Norms). Let X and Y be

normed linear spaces. Let T ∈ B(X, Y). Then
 
kT k = sup kT xkY ∈ R : x ∈ X, kxkX = 1 .

PROPOSITION 11.5. Let X and Y be normed linear spaces. Let T ∈ B(X, Y).
Then ∀x ∈ X, kT xkY ≤ kT kkxkX .

PROPOSITION 11.6. Let X, Y, and Z be normed linear spaces. Let T ∈ B(X, Y)

and S ∈ B(Y, Z). Then ST ∈ B(X, Z) and kST k ≤ kSkkT k.

11.3 Examples of Bounded Operators

11.3. EXAMPLES OF BOUNDED OPERATORS 61

EXAMPLE 11.7 (The Multiplication Operator). Let X = C([0, 1], C), k · k∞ . Let
f be a function in X. We define the multiplication operator on X, w.r.t. f , denoted
by Mf , as
Mf (g) = f g.

Then Mf is bounded and kMf k = kf k∞ .

Proof. Let g be an arbitrary function in X. Then

kMf gk∞ = kf gk∞

= sup |f (x)g(x)|
x∈[0,1]

= sup |f (x)||g(x)|
x∈[0,1]

≤ sup |f (x)| sup |g(x)|

x∈[0,1] x∈[0,1]

= kf k∞ kgk∞ .

That is, kMf gk∞ ≤ kf k∞ kgk∞ . So kf k∞ is an element of the set S = {k ∈ R : ∀g ∈

X, kMf gkY ≤ kkgkX }. So kMf k = inf(S) ≤ kf k∞ . Consider g0 given by g0 (x) = 1. Then
g0 in X. Then
kMf g0 k∞ = kf g0 k∞ = kf k∞ = kf k∞ kg0 k∞ .

Let k be an arbitrary element in S. Assume for the sake of contradiction that k < kf k∞ .
Then

kf k∞ kg0 k∞ = kMf g0 k∞
≤ kkg0 k∞
< kf k∞ kg0 k∞ .

This leads to a contradiction. So ∀k ∈ S, k ≥ kf k∞ . So kf k∞ is a lower bound for the

set S. So kMf k = inf(S) ≥ kf k∞ . Since kMf k ≤ kf k∞ and kMf k ≥ kf k∞ , we get
kMf k = kf k∞ .

EXAMPLE 11.8 (The Volterra Operator). Let X = C([0, 1], C), k · k∞ . Define
Z x
V f := x 7→ f (t)dt.
0
62 CHAPTER 11. OPERATORS

Then the Volterra Operator is bounded and kV k ≤ 1.

Proof. Let f be an arbitrary function in X with kf k∞ = 1. Then ∀x ∈ [0, 1],

Z x
|V f (x)| = f (t)dt
0
Z x
≤ |f (t)|dt
0
Z x
≤ sup |f (t)|dt
0 t∈[0,1]
Z x
= kf k∞ dt
Z0 x
= 1dt
0

= x.

That is, ∀x ∈ [0, 1], |V f (x)| ≤ 1. So kV f k∞ ≤ 1. Since ∀f ∈ X : kf k∞ = 1, kV f k∞ ≤ 1,

we get kV k ≤ 1.

EXAMPLE 11.9 (The Diagonal Operator). Let X = `2 (N). Let

 
d1
d2
 
 
D= .

 d3 

..
.

Then D is bounded if and only if (di )i∈N is bounded and kDk = k(di )i∈N k∞ .

Proof. Case 1.

X
kDxk22 = |di xi |2
i∈N
X
=≤ k(dj )j∈N k∞ |xi |2
i∈N
X
= k(dj )j∈N k∞ |xi |2
i∈N

= k(dj )j∈N k∞ kxk22 .

Case 2.
11.3. EXAMPLES OF BOUNDED OPERATORS 63

/ `∞ , ∃(dni )i∈N → ∞.
If (di )i∈N ∈

kDeni k2 = kdni eni k2

= |dni |keni k2
= |dni |.

So kDk ≥ kDeni k2 → ∞.

EXAMPLE 11.10 (Weighted Shifts).

• Let H = `2N . Let (wn )n∈N ∈ `∞

N . We define an unilateral forward weighted
shift W on H as
W (xn ) := (0, w1 x1 , w2 x2 , w3 x3 , ...).

i.e.,  
0
 
w1 0 
 
W =
 w2 0 .

 
 w3 0 
 
.. ..
. .

Then W is bounded and kW k = sup{|wn | : n ∈ N}.

• Let H = `2N . Let (vn )n∈N ∈ `∞

N . We define an unilateral backward weighted
shift V on H as
V (xn ) := (v1 x2 , v2 x3 , v3 x4 , ...).

Then V is bounded and kV k = sup{|vn | : n ∈ N}.

• Let H = `2Z . Let (un )n∈Z ∈ `∞

Z . We define a bilateral weighted shift U on H
as
U (xn ) := (un−1 xn−1 )n∈Z .

Then U is bounded and kU k = sup{|un | : n ∈ Z}.

EXAMPLE 11.11 (The Composition Operators). Let X = C([0, 1], C). Let ϕ ∈
C([0, 1], [0, 1]). We define the composition operator on X, denoted by Cϕ as

Cϕ (f ) := f ◦ ϕ.
64 CHAPTER 11. OPERATORS

Then Cϕ is contractive.

Proof.

kCϕ (f )k = sup |(f ◦ ϕ)(x)|

x∈[0,1]

≤ kf k∞ .

11.4 The Space of Bounded Linear Operators

PROPOSITION 11.12. Let X and Y be normed linear spaces. Then B(X, Y) is a

vector space and the operator norm is a norm on B(X, Y).

PROPOSITION 11.13. Let X and Y be normed linear spaces. Let B(X, Y) be the
space of bounded linear operators from X to Y. Then if Y is complete, B(X, Y) is
complete.

PROPOSITION 11.14. Let X and Y be normed linear spaces. Let k · k1 and

k · k2 be two equivalent norms on B(X, Y). Then T ∈ B(X, Y, k · k1 ) if and only if
T ∈ B(X, Y, k · k2 ).

11.5 Invertible Bounded Linear Operators

PROPOSITION 11.15. Let (X, k · k1 ) be a Banach space. Let S ∈ B(X) be a

bounded linear map that is invertible. Define a norm k · k2 on X as

kxk2 := kSxk1 .

Then k · k1 and k · k2 are equivalent.

Proof. On one hand, since S is bounded, ∃c1 such that ∀x ∈ X, kSxk1 ≤ c1 kxk1 . That is,
kxk2 ≤ c1 kxk1 .
11.5. INVERTIBLE BOUNDED LINEAR OPERATORS 65

On the other hand, since S is invertible, S −1 exists and is also bounded. Since S −1 is
bounded, ∃c2 such that ∀x ∈ X, kS −1 xk1 ≤ c2 kxk1 . Consider x = Sx, we get ∀x ∈ X,
kS −1 Sxk1 ≤ c2 kSxk1 . That is, kxk1 ≤ c2 kxk2 .
So k · k1 and k · k2 are equivalent.

PROPOSITION 11.16. Let (X, k · k) be a Banach space. Let S be a map in B(X)

that is invertible. Then

kS −1 k = (inf{kSxk : kxk = 1})−1 .

Proof.

kS −1 k = sup{kS −1 xk : kxk = 1}
 −1 
kS xk
= sup : kxk =
6 0
kxk
 −1 
kS Sxk
= sup : kxk =
6 0
kSxk
 
kxk
= sup : kxk =6 0
kSxk
 
1
= sup : kxk = 1
kSxk
= (inf{kSxk : kxk = 1})−1 .

That is,
kS −1 k = (inf{kSxk : kxk = 1})−1 .

PROPOSITION 11.17. Let X be a Banach space. Then the set of invertible bounded
linear operators from X to X is an open set.
66 CHAPTER 11. OPERATORS
Chapter 12

Matrix Space

DEFINITION 12.1 (Symmetric Matrices). We denote by Sn the inner product space

with underlying set the set of n-by-n symmetric matrices over the real numbers
 
Sn := X ∈ Rn×n : X = X >

and inner product given by

hA, Bi := tr(AB).

PROPOSITION 12.2. Sn is a Hilbert space.

Proof. Rn×n is complete under any norm. Sn is a linear subspace of the finite dimensional
space Rn×n and hence closed. So Sn is a closed subset of a complete space and hence
complete.

DEFINITION 12.3. We denote by Sn+ the set of PSD matrices in Sn . We denote by

Sn++ the set of PD matrices in Sn .

67
68 CHAPTER 12. MATRIX SPACE
Chapter 13

Balanced Set

13.1 Definitions

DEFINITION 13.1 (Balanced Sets). Let X be a vector space over field F. Let S be
a subset of X. We say that S is balanced if

∀a ∈ F : |a| ≤ 1, aS ⊆ S.

13.2 Properties

PROPOSITION 13.2. Let X be a vector space over field F. Let B be a balanced

set in X. Then
∀a, b ∈ F : |a| ≤ |b|, aB ⊆ bB.

PROPOSITION 13.3. Balanced sets are path connected.

13.3 Stability of Balance

PROPOSITION 13.4 (Set Operations). • The union of balanced sets is bal-

anced.

69
70 CHAPTER 13. BALANCED SET

• The intersection of balanced sets is balanced.

PROPOSITION 13.5 (Convex Hull). The convex hull of a balanced set is balanced.

Proof. Let X be a vector space over field F. Let B be a balanced set in X. I will show
that conv(B) is balanced. Let k be an arbitrary element of F such that |k| ≤ 1. Let x be
an arbitrary element of k conv(B). Then ∃n ∈ N, x1 , ..., xn ∈ B, λ1 , ..., λn ∈ R such that
n
X n
X
λ1 , ..., λn ≥ 0, λi = 1, and x = k λi xi . Since |k| ≤ 1, B is balanced, and xi ∈ B, we
i=1 i=1
get kxi ∈ B. So
n
X n
X
x=k λ i xi = λi (kxi ) ∈ conv(B).
i=1 i=1

So k conv(B) ⊆ conv(B). So conv(B) is balanced.

PROPOSITION 13.6 (Closure). The closure of a balanced set is balanced.

PROPOSITION 13.7 (Linear Maps). • The scalar multiple of a balanced set is

also balanced.

• The (Minkowski) sum of two balanced sets is also balanced.

• The image of a balanced set under a linear operator is also balanced.

• The inverse image of a balanced set under a linear operator is also balanced.

13.4 The Balanced Hull Operator

DEFINITION 13.8 (Balanced Hull). Let X be a vector space over field F. Let S
be a subset of X. We define the balanced hull of S, denoted by balhull(S), to be the
smallest balanced set containing S.
13.5. THE BALANCED CORE OPERATOR 71

PROPOSITION 13.9 (Act on Other Properties).

• The balanced hull of a compact set is compact.

• The balanced hull of a totally bounded set is totally bounded.

• The balanced hull of a bounded set is bounded.

PROPOSITION 13.10. Let X be a vector space over field F. Let a be a scalar in

field F. Then
a balhull(S) = balhull(aS).

13.5 The Balanced Core Operator

DEFINITION 13.11 (Balanced Core). Let X be a vector space over field F. Let S
be a subset of X. We define the balanced core of S, denoted by balcore(S), to be the
largest balanced set contained in S.

PROPOSITION 13.12 (Act on Other Properties).

• The balanced core of a closed set is closed.

72 CHAPTER 13. BALANCED SET
Chapter 14

Topological Vector Space

14.1 Preliminaries: Absorbing Set

DEFINITION 14.1 (Absorbing Sets). Let X be a vector space over field F. Let S
be a subset of X. We say that S is absorbing if

∀x ∈ X, ∃r ∈ R : r > 0, ∀c ∈ F : |c| ≥ r, x ∈ cS.

i.e.,
[
nS = X.
n∈N

PROPOSITION 14.2. Every absorbing set contains the origin.

14.2 Definitions

DEFINITION 14.3 (Compatible). Let V be a vector space over field K. Let T be a

topology on V. We say that T is compatible with the vector space structure on V if
the addition and scalar multiplication operations on V are continuous.

73
74 CHAPTER 14. TOPOLOGICAL VECTOR SPACE

DEFINITION 14.4 (Topological Vector Space). We define a topological vector

space to be a vector space with a compatible Hausdorff topology.

14.3 Neighborhood Improvements

PROPOSITION 14.5. Let (V, τ ) be a topological vector space. Let U ∈ U0 be a

neighborhood of 0 in V. Then we have the followings.

1. ∃N ∈ U0 such that N + N ⊆ U .

2. ∃M ∈ U0 and ∃ε > 0 such that ∀0 < |k| < ε, we have kM ⊆ U .

[
3. V = nU .
n∈N

PROPOSITION 14.6. Any neighborhood of 0 in a topological vector space is ab-

sorbing.

PROPOSITION 14.7. Let V be a topological vector space over field K. Then every
neighborhood of 0 contains a open balanced neighborhood of 0.

Proof. Let U be an arbitrary element of U0 . Let µ denote the multiplication operation on

V. Then µ is continuous and hence µ−1 (U ) is a neighborhood of (0, 0) ∈ K × V. So there
exist an r > 0 and an open neighborhood N of 0 such that ball(0, r) × N ⊆ µ−1 (U ). Define
[
a set M by M := µ(ball(0, r), N ) = kN . Then M ⊆ U . Notice 0 ∈ N and hence
k∈K:|k|<r
0 ∈ kN for any k ∈ K. So 0 ∈ M.
Open: Since N is open and scalar multiplication is a homeomorphism, we get kN is
[
open for any k ∈ K. So kN is open. That is, M is open.
k∈K:|k|<r
Balanced: Let a be an arbitrary element of K such that |a| < 1. Then
[
aM = a kN , by definition of M
k∈K:|k|<r
[ [
= akN = kN
k∈K:|k|<r k∈K:|k|<|a|r
[
⊆ kN , since a < 1
k∈K:|k|<r
14.4. PROPERTIES OF TOPOLOGICAL VECTOR SPACES 75

= M, by definition of M.

That is, ∀a ∈ K : |a| < 1, we have aM ⊆ M. So M is balanced.

14.4 Properties of Topological Vector Spaces

PROPOSITION 14.8 (Normed Linear Spaces are Topological Vector Spaces). Let
X be a normed linear space over field K. Then X is a topological vector space with the
topology induced by the norm.

Proof. Part 1: Show that the norm topology is compatible with the vector space structure.
Let σ denote the addition operation in X. Let ((xα , yα ))α∈Λ be an arbitrary net in X × X
that converges to (x, y) ∈ X × X.

kσ(xα , yα ) − σ(x, y)k = k(xα + yα ) − (x + y)k = k(xα − x) + (yα − y)k

≤ kxα − xk + kyα − yk, by the triangle inequality
< ε/2 + ε/2 = ε.

That is, kσ(xα , yα ) − σ(x, y)k < ε. So σ is continuous.

Let µ denote the scalar multiplication operation in X. Let ((kα , xα ))α∈Λ be an arbitrary
net in K × X that converges to (k, x) ∈ K × X.

kµ(kα , xα ) − µ(k, x)k = kkα xα − kxk = kkα xα − kxα + kxα − kxk

≤ kkα xα − kxα k + kkxα − kxk, by the triangle inequality
= |kα − k|kxα k + |k|kxα − xk < ε/2 + ε/2 = ε.

That is, kµ(kα , xα ) − µ(k, x)k < ε. So µ is continuous.

Part 2: Show that the norm topology is Hausdorff.
Let x and y be arbitrary elements of X. Suppose that x 6= y. Define a number r by
r := kx − yk/2. Then ball(x, r) ∈ Ux and ball(y, r) ∈ Uy and ball(x, r) ∩ ball(y, r) = ∅. So
any two distinct points in X are separated by the norm topology. So X is Hausdorff.

PROPOSITION 14.9. Closure of a linear manifold of a topological vector space is

a closed linear subspace.
76 CHAPTER 14. TOPOLOGICAL VECTOR SPACE

Proof. Let V be a topological vector space. Let W be a linear manifold of V. We are to

prove that cl(W) is a linear subspace. Note that cl(W) is closed. So there remains only to
show that cl(W) is linear.
Let x and y be arbitrary elements of cl(W). Then there exists a net (xλ , yλ )λ∈Λ that
converges to (x, y). Since the addition operation σ is continuous, we have lim (xλ + yλ ) =
λ∈Λ
x + y. Since W is a linear subspace, xλ + yλ ∈ W. So x + y ∈ cl(W).
Let x be an arbitrary element of cl(W). Let k be an arbitrary element in K. Then there
exists a net (kλ, xλ )λ∈Λ that converges to (k, x). Since the scalar multiplication operation
µ is continuous, we have lim (kλ xλ ) = kx. Since W is a linear subspace, kλ xλ ∈ W. So
λ∈Λ
kx ∈ cl(W).

PROPOSITION 14.10. A quotient space of a topological vector space by a closed

subspace is a topological vector space with the quotient topology .

14.5 Operation on Sets in a Topological Vector Space

PROPOSITION 14.11 (Stability under Linear Combinations). Let X be a normed

vector space over F. Let K be a compact set in the space. Let C be a closed set in the
space. Then ∀α, β ∈ F, the set S given by S := αK + βC is closed.

Proof. The case where β = 0 is trivial. I will assume β 6= 0. Let α, β ∈ F be arbitrary. Let
{si }i∈N be an arbitrary sequence in S that converges. Say the limit is s∞ . Since si ∈ S
for any i ∈ N and S = αK + βC, si = αki + βci for some ki ∈ K and some ci ∈ C, for
any i ∈ N. Since {ki }i∈N is a sequence in K and K is compact, there exists a convergent
subsequence {ki }i∈I of {ki }i∈N in K. Say {ki }i∈I converges to k∞ ∈ K. Since {si }i∈N
converges to s∞ , {si }i∈I also converges to s∞ . Since si = αki + βci , ci = β −1 (si − αki ).
Define c∞ := β −1 (s∞ − αk∞ ) Since {si }i∈I converges to s∞ and {ki }i∈I converges to k∞
and ci = β −1 (si − αki ), {ci }i∈I converges to c∞ . Since {ci }i∈I is a sequence in C and
converges to c∞ and C is closed, c∞ ∈ C. Since s∞ = αk∞ + βc∞ and k∞ ∈ K and
c∞ ∈ C, s∞ ∈ αK + βC. Since for any sequence in S that converges, the limit is also in S,
S is closed.

REMARK 14.12. The sum of two closed sets may not be closed.
14.5. OPERATION ON SETS IN A TOPOLOGICAL VECTOR SPACE 77

Proof. Counter-example 1
1
Consider A := {n : n ∈ N} and B := {n + : n ∈ N}.
n
(https://math.stackexchange.com/questions/124130/sum-of-two-closed-sets-in-mathbb-r-is-closed)
1
Their sum contains the sequence { }n∈N but does not contain 0.
n
Counter-example 2
Consider A := R × {0} and B := {(x, y) ∈ R2 : x, y > 0, xy ≥ 1}. Their sum is
R × R++ .

PROPOSITION 14.13. Let X be a normed vector space. Let S be a subset of X.

Let p be a vector in X. Then we have the followings.

1. p + int(S) = int(p + S),

2. p + cl(S) = cl(p + S).

Proof of (1). For one direction, let x be an arbitrary point in the set p + int(S). We are to
prove that x ∈ int(p + S). Since x ∈ (p + int(S)), (x − p) ∈ int(S). Since (x − p) ∈ int(S),
by definition of interior, there exists a radius r such that

B(x − p, r) ⊆ S.

It follows that B(x, r) ⊆ p + S. Since there exists a radius r such that B(x, r) ⊆ p + S, by
definition of interior,
x ∈ int(p + S).

For the reverse direction, let x be an arbitrary point in int(p + S). We are to prove that
x ∈ p + int(S). Since x ∈ int(p + S), by definition of interior, there exists a radius r such
that
B(x, r) ⊆ (p + S).

It follows that B(x − p, r) ⊆ S. Since there exists a radius r such that B(x − p, r) ⊆ S, by
definition of interior,
(x − p) ∈ int(S).

Since (x − p) ∈ int(S), we get x ∈ (p + int(S)).

Proof of (2). For one direction, let x be an arbitrary point in the set p + cl(S). We are to
prove that x ∈ cl(p + S). Since x ∈ (p + cl(S)), we get (x − p) ∈ cl(S). Since (x − p) ∈ cl(S),
by definition of closure, for any radius r, we have

B(x − p, r) ∩ S 6= ∅.
78 CHAPTER 14. TOPOLOGICAL VECTOR SPACE

It follows that B(x, r) ∩ (p + S) 6= ∅. Since for any radius r, B(x, r) ∩ (p + S) 6= ∅, by

definition of closure, we get
x ∈ cl(p + S).

For the reverse direction, let x be an arbitrary point in cl(p + S). We are to prove that
x ∈ (p + cl(S)). Since x ∈ cl(p + S), by definition of closure, for any radius r, we have

B(x, r) ∩ (p + S) 6= ∅.

It follows that B(x − p, r) ∩ S 6= ∅. Since for any radius r, B(x − p, r) ∩ S 6= ∅, by definition

of closure, we get
(x − p) ∈ cl(S).

Since (x − p) ∈ cl(S), we get x ∈ (p + cl(S)).

PROPOSITION 14.14. Let (V, k · k) be a normed vector space. Let S be a subset

of V . Let λ be a non-zero real number. Then

1. λ int(S) = int(λS).

2. λ cl(S) = cl(λS).

Proof of (1). For one direction, let x be an arbitrary point in λ int(S). We are to prove that
x ∈ int(λS). Since x ∈ λ int(S), we get x/λ ∈ int(S). Since x/λ ∈ int(S), by definition of
interior, there exists a radius r such that

B(x/λ, r) ⊆ S.

Let y be an arbitrary point in B(x, λr). Since y ∈ B(x, λr), we get ky − xk ≤ λr. Since
ky − xk ≤ λr, we get ky/λ − x/λk ≤ r. Since ky/λ − x/λk ≤ r, we get y/λ ∈ B(x/λ, r).
Since y/λ ∈ B(x/λ, r) and B(x/λ, r) ⊆ S, we get y/λ ∈ S. Since y/λ ∈ S, we get y ∈ λS.
Since any point in B(x, λr) is also in λS, we get B(x, λr) ⊆ λS. Since there exists a radius
r such that B(x, λr) ⊆ λS, by definition of interior, we get

x ∈ int(λS).

14.6 Finite-Dimensional Topological Vector Spaces

14.6. FINITE-DIMENSIONAL TOPOLOGICAL VECTOR SPACES 79

PROPOSITION 14.15. Let V be an n-dimensional topological vector space where

n ∈ N. Then V is homeomorphic to Kn via the map
n
X
ki ei 7→ (ki )ni=1 .
i=1

COROLLARY 14.16. Let V be a finite-dimensional vector space. Then there is a

unique topology T which makes V a topological vector space.

COROLLARY 14.17. Let (V, TV ) and (W, TW ) be topological vector spaces. Let T
be a linear map from V to W. Then if V is finite dimensional, T is continuous.

Proof. Assume that dim(V) < +∞. I will show that T is continuous. It suffices to show that
T is continuous at 0. Let {e1 , ..., en } be a basis for V. Let (xλ )λ∈Λ be a net of elements of V
Xn
that converges to 0. Say for each λ ∈ Λ, xλ = xλ,i ei . Since lim xλ = 0, ∀i ∈ {1, ..., n},
λ∈Λ
i=1
we have lim xλ,i = 0. Then
λ∈Λ

n
X
lim T xλ = lim T xλ,i ei
λ∈Λ λ∈Λ
i=1
n
X
= lim xλ,i T ei , since T is linear
λ∈Λ
i=1
n
X
= lim xλ,i T ei , since addition is continuous
λ∈Λ
i=1
Xn
= 0T ei = 0.
i=1

That is, lim T xλ = 0. It follows that T is continuous.

λ∈Λ
80 CHAPTER 14. TOPOLOGICAL VECTOR SPACE
Chapter 15

Continuous and Uniformly

Continuous Functions

15.1 Definition

DEFINITION 15.1 (Uniformly Continuous). Let V and W be topological vector

spaces. Let f be a function from V to W. We say that f is uniformly continuous if
∀U ∈ U0W , ∃N ∈ U0V such that ∀x, y ∈ V : x − y ∈ N , we have f (x) − f (y) ∈ U .

15.2 Extension of Continuous Linear Maps

PROPOSITION 15.2. Let V and W be topological vector spaces. Suppose that W

is complete. Let X be a linear manifold of V. Let T0 be a continuous linear map from
X to W. Then T0 extends to a continuous linear map T from cl(X ) to W.

PROPOSITION 15.3. Let X and Y be Banach spaces. Let M be a linear manifold

of X. Let T0 be a bounded linear operator from M to Y. Then T0 extends to a bounded
linear operator from cl(M) to Y and we have kT k = kT0 k.

15.3 Relation between the two Notions

81
82 CHAPTER 15. CONTINUOUS AND UNIFORMLY CONTINUOUS FUNCTIONS

PROPOSITION 15.4. Uniformly continuous functions are continuous.

PROPOSITION 15.5. Continuous linear maps are uniformly continuous.

Proof. Let V and W be topological vector spaces. Let T be a continuous linear map from
V to W. I will show that T is uniformly continuous. Fix a point x0 ∈ V. Let U0 be an
arbitrary element of U0W . Define U := T (x0 ) + U0 . Then U ∈ UTW(x0 ) . Since T is continuous
at x0 , ∃N ∈ UxV0 such that ∀x ∈ N , T (x) ∈ U . Define N0 := −x0 + N . Then N0 ∈ U0V . Let
x and y be arbitrary elements of V such that x − y ∈ N0 . Then

x − y ∈ N0
⇐⇒ x0 + x − y ∈ N, since N = x0 + N0
=⇒ T (x0 + x − y) ∈ U, by continuity of T
⇐⇒ T (x0 ) + T (x) − T (y) ∈ U, by linearity of T
⇐⇒ T (x) − T (y) ∈ U0 , since U = T (x0 ) + U0 .

So we have T (x) − T (y) ∈ U0 . So T is uniformly continuous.

PROPOSITION 15.6. Continuous conjugate linear maps are uniformly continuous

PROPOSITION 15.7. Continuous linear maps defined on a balanced and convex

subset are uniformly continuous.
Chapter 16

Complete Space

16.1 Cauchy Nets

DEFINITION 16.1 (Cauchy Net). Let (V, τ ) be a topological vector space. Let
(xλ )λ∈Λ be a net in V. We say that (xλ )λ∈Λ is a Cauchy net if ∀U ∈ U0 , ∃λ0 ∈ Λ
such that ∀λ1 , λ2 ≥ λ0 , we have xλ1 − xλ2 ∈ U .

PROPOSITION 16.2. Convergent nets are Cauchy.

Proof. Let V be a topological vector space. Let (xλ )λ∈Λ be a convergent net with limit
point x. Let U be an arbitrary element in U0 . Let N be an element in U0 that is balanced
and open and that N − N ⊆ U . Since lim xλ = x, ∃λ0 ∈ Λ such that ∀λ ≥ λ0 , xλ − x ∈ N .
λ∈Λ
Let λ1 and λ2 be arbitrary elements that are ≥ λ0 . Then

xλ1 − xλ2 = (xλ1 − x) − (xλ2 − x) ∈ N − N ⊆ U.

That is, ∀U ∈ U0 , ∃λ0 such that ∀λ1 , λ2 ≥ λ0 , xλ1 − xλ2 ∈ U . So (xλ )λ∈Λ is Cauchy.

16.2 Complete Topological Vector Spaces

DEFINITION 16.3 (Cauchy Complete). Let (V, τ ) be a topological vector space.

We say that V is Cauchy complete if every Cauchy net in V converges in V.

83
84 CHAPTER 16. COMPLETE SPACE

PROPOSITION 16.4. Let V be a topological vector space. Let K be a complete set

in V. Then K is closed in V.
Chapter 17

Locally Convex Space

17.1 Preliminaries
17.1.1 Seminorms

PROPOSITION 17.1. Let V be a topological vector space. Let p be a seminorm on

V. Then p is continuous on V if and only if it is bounded above on some neighborhood
of 0.

Proof. Forward Direction: Assume that p is continuous. I will show that p is bounded
above on some neighborhood of 0. Define a set E by E := {x ∈ V : p(x) < 1}. Note that
range(p) = [0, +∞), [0, 1) is an open subset of [0, +∞). Since [0, 1) is open, p is continuous,
and E = p−1 ([0, 1)), E is open. Note that p(0) = 0 < 1. So 0 ∈ E. So E ∈ U0 . By definition
of E, p is bounded above by 1 on E.
Backward Direction: Assume that p is bounded above on some neighborhood of 0. I will
show that p is continuous. Say p is bounded above by M ∈ R+ on U ∈ U0 . Let ε > 0 be
ε
arbitrary. Define a set N ∈ U0 by N := U . Let x and y be arbitrary elements of V
M +1
ε
such that x − y ∈ N . Then x − y = u for some u ∈ U . So
M +1
ε ε ε
|p(x) − p(y)| ≤ p(x − y) = p( u) = p(u) ≤ M < ε.
M +1 M +1 M +1
That is, |p(x)−p(y)| < ε. So p is uniformly continuous on V and hence continuous on V.

17.1.2 Sublinear Functionals

85
86 CHAPTER 17. LOCALLY CONVEX SPACE

DEFINITION 17.2 (Sublinear Functional). Let V be a vector space over field K.

Let f be a function from V to R. We say that f is sublinear if it satisfies:

• Subadditivity:
∀x, y ∈ V, f (x + y) ≤ f (x) + f (y).

• Positive Homogeneity:

∀x ∈ V, ∀λ ≥ 0, f (λx) = λf (x).

PROPOSITION 17.3. Seminorms are sublinear functionals.

17.1.3 The Minkowski Functional

DEFINITION 17.4 (Minkowski Functional). Let V be a topological vector space.

Let E be a convex neighborhood of 0 in V. We define the Minkowski functional for
E, denoted by pE , to be a function from V to R given by

pE (x) := inf{r > 0 : x ∈ rE}.

PROPOSITION 17.5 (Convex). A Minkowski functional for a convex neighborhood

of 0 is a sublinear functional.

Proof. Let V be a topological vector space over field K. Let E be a convex neighborhood of
0 in V. Let p denote the Minkowski functional for E. I will show that p is sublinear.
Part 1: Show that ∀x, y ∈ V, we have p(x + y) ≤ p(x) + p(y).
Assume for the sake of contradiction that ∃x, y ∈ V such that p(x + y) > p(x) + p(y).
Define ε > 0 as ε := p(x + y) − (p(x) + p(y)). By definition, ∃rx > 0 such that x ∈ rx E and
rx < p(x) + ε/2, and ∃ry > 0 such that y ∈ ry E and ry < p(y) + ε/2. Since x ∈ rx E and
y ∈ ry E, we get x + y ∈ (rx + ry )E. So p(x + y) ≤ rx + ry . So

p(x + y) ≤ rx + ry < p(x) + ε/2 + p(y) + ε/2 = p(x + y).

That is, p(x + y) < p(x + y), a contradiction. So p(x + y) ≤ p(x) + p(y).
Part 2: Show that ∀x ∈ V, ∀k > 0, we have p(kx) = kp(x).
17.1. PRELIMINARIES 87

Let x be an arbitrary element of V. Let k be an arbitrary element of R such that k > 0.

Then

p(kx) = inf{r > 0 : kx ∈ rE} = inf{kr > 0 : kx ∈ krE}

= inf{kr > 0 : x ∈ rE} = k inf{r > 0 : x ∈ rE} = kp(x).

That is, p(kx) = kp(x).

PROPOSITION 17.6 (Balanced and Convex). A Minkowski functional for a balanced and convex
neighborhood of 0 is a seminorm.

Proof. Let V be a topological vector space over field K. Let E be a balanced convex neigh-
borhood of 0 in V. Let p denote the Minkowski functional for E. I will show that p is a
seminorm. I have showed that Minkowski functionals for convex sets are subadditive. It is
clear that ∀x ∈ V, p(x) ≥ 0. Let x be an arbitrary element of V. Let k be an arbitrary
element of K. If k = 0, then p(kx) = p(0x) = p(0) = 0 = 0p(x) = |k|p(x) and we are done.
Otherwise, k 6= 0. Then

p(kx) = inf{r > 0 : kx ∈ rE} = inf{|k|r > 0 : kx ∈ |k|rE}

= inf{|k|r > 0 : kx ∈ krE}, since E is balanced
= inf{|k|r > 0 : x ∈ rE} = |k| inf{r > 0 : x ∈ rE} = |k|p(x).

That is, p(kx) = |k|p(x).

PROPOSITION 17.7 (Open and Convex). Let V be a topological vector space.

Let E be an open and convex neighborhood of 0 in V. Let pE denote the Minkowski
functional for E. Then
E = {x ∈ V : pE (x) < 1}.

Proof. Let F denote the set {x ∈ V : pE (x) < 1}. I will show that E = F .
Forward Direction:
Let x be an arbitrary element of E. I will show that x ∈ F . Define a map f : R → V
by f (t) := tx. Then f is continuous. Since E is open in V and f : R → V is continuous,
we get f −1 (E) is open in R. Notice x = f (1) ∈ E. So 1 ∈ f −1 (E). Since f −1 (E) is open
and 1 ∈ f −1 (E), ∃δ > 0 such that 1 + δ ∈ f −1 (E). So f (1 + δ) ∈ E. So (1 + δ)x ∈ E. So
1 1
x∈ E. So pE (x) ≤ , which further, is < 1. So x ∈ F .
1+δ 1+δ
88 CHAPTER 17. LOCALLY CONVEX SPACE

Backward Direction:
Let x be an arbitrary element of F . I will show that x ∈ E. Since x ∈ F , by definition
of F , pE (x) < 1. So by definition of the Minkowski functional, ∃r0 > 0 such that r0 < 1
and x ∈ r0 E. Since r0 < 1, we have r0 E ⊆ E. So x ∈ E.

17.1.4 Separating Families

DEFINITION 17.8 (Separating Family of Seminorms). Let V be a vector space. Let

Γ be a family of seminorms on V. We say that Γ is separating if ∀x ∈ V such that
x 6= 0, ∃p ∈ Γ such that p(x) 6= 0.

DEFINITION 17.9 (Separating Family of Linear Functionals). Let V be a vector

space. Let L be a collection of linear functionals on V. Define for each ϕ ∈ L a
seminorm τϕ on V by τϕ (x) := |ϕ(x)|. We say that L is separating if the set Γ given
by Γ := {τϕ : ϕ ∈ L} is a separating family of seminorms.

17.2 Locally Convex Space

DEFINITION 17.10 (Locally Convex Space). Let (V, T ) be a topological vector

space. We say that T is locally convex if each point in V admits a neighborhood base
consisting of only convex sets.

PROPOSITION 17.11. Any convex neighborhood of 0 in a topological vector space

contains a convex, open, and balanced neighborhood of 0.

Proof. Let V be a topological vector space. Let U be an arbitrary element of U0V that is
convex. Since U ∈ U0V , U contains an open and balanced neighborhood N of 0. Then
conv(N ) ⊆ conv(U ) = U and conv(N ) is open and balanced. So conv(N ) is the set desired.
This completes the proof.
17.2. LOCALLY CONVEX SPACE 89

PROPOSITION 17.12. A quotient space of a locally convex space by a closed

subspace is a locally convex space with the quotient topology.

Proof. Let V be a locally convex space. Let W be a closed subspace of V. Let q denote the
canonical quotient map from V to V/W. Then q is continuous, open, and linear, and V/W
is a topological vector space with the quotient topology. So there remains only to show
V/W
that the quotient topology is locally convex. Let U be an arbitrary element of U0 . Since
V/W
U ∈ U0 and q is continuous, we get q −1 (U ) ∈ U0V . Since V is locally convex and U ∈ U0V ,
V/W
∃C ∈ U0V such that C ⊆ q −1 (U ). Since C ∈ U0V and q is open, we get q(C) ∈ U0 . Since
−1
C is convex and q is linear, q(C) is convex. Since C ⊆ q (U ), q(C) ⊆ U . So q(C) is the
set desired. This completes the proof.

DEFINITION 17.13 (Generated Topology). Let V be a vector space. Let Γ be a

separating family of seminorms on V. Define for each x ∈ V, F ⊆ Γ finite, and ε > 0 a
subset N (x, F, ε) of V by

N (x, F, ε) := {y ∈ V : ∀p ∈ F, p(x − y) < ε}.

Define a collection B of subsets of V by

B := {N (x, F, ε) : x ∈ V, F ⊆ F finite , ε > 0}.

We define the topology generated by Γ to be the topology generated by B.

THEOREM 17.14. Let V be a vector space. Let Γ be a separating family of

seminorms on V. Then Γ generates a locally convex topology on V.

THEOREM 17.15. Let (V, T ) be a locally convex space. Then there exists a sepa-
rating family Γ of seminorms on V that generates the original topology T .

Proof. Since (V, T ) is locally convex, it admits a neighborhood base B0 at 0 consisting of

only convex, balanced, and open sets. Denote the Minkowski functional for each E ∈ B0
by pE . Define Γ := {pE : E ∈ B0 }. Since each E is balanced and convex, each pE is a
seminorm. Now I claim that Γ is separating. Let x be an arbitrary element of V such that
x 6= 0. Since x 6= 0 and T is Hausdorff, ∃U ∈ U0T such that x ∈
/ U . Since V is locally convex,
I can assume without loss of generality that U is convex, balanced, and open. Since U is
90 CHAPTER 17. LOCALLY CONVEX SPACE

open, U = {x ∈ V : pU (x) < 1}. Since x ∈

/ U , we get pU (x) ≥ 1. So pU (x) 6= 0. So pU is
the function desired. So Γ is separating. Let T 0 denote the topology generated by Γ. Now
I claim that T 0 = T . Let B 0 denote the base generated by Γ.
Forward Direction: Show that T ⊆ T 0 .
Let E be an arbitrary element of B0 . Then E is convex and open and hence E =
N (0, pE , 1) ∈ B 0 ⊆ T 0 . That is, E ∈ T 0 . So T 0 contains the neighborhood base B0 at 0 for
T . So T ⊆ T 0 .
Backward Direction: Show that T 0 ⊆ T .
Since B 0 is a base for T 0 , N0 := {N (0, F, ε) : F ⊆ Γ finite , ε > 0} is a neighborhood
base at 0 for T 0 . Let N be an arbitrary element of N0 . Then N = N (0, F, ε) for some
F ⊆ Γ finite and some ε > 0. Let pE be an arbitrary element of F . Then E is convex
and open and hence pE is bounded above by 1 on E. So pE is continuous under T . So
\
N (0, pE , ε) = p−1
E ([0, ε)) is open under T . So N (0, F, ε) = N (0, pE , ε) is open under
pE ∈F
T . So T contains the neighborhood base N0 at 0 for T 0 . So T 0 ⊆ T .

REMARK 17.16. The above two theorems say that separating families of seminorms
on vector spaces give rise to locally convex topologies, and that all locally convex
topologies arise in this manner.

EXAMPLE 17.17. The norm topology is exactly the locally convex topology gener-
ated by Γ = {k · k}.

17.3 Relation to Other Topologies

PROPOSITION 17.18. A locally convex topology is equivalent to a metric topology

if and only if it can be generated by a countable family of seminorms.

PROPOSITION 17.19. A locally convex topology is equivalent to a norm topology

if and only if it can be generated by a finite family of seminorms.
17.4. CONTINUITY IN LOCALLY CONVEX SPACES 91

17.4 Continuity in Locally Convex Spaces

PROPOSITION 17.20. Let (V, T ) be a locally convex space. Let Γ be a separating

family of seminorms on V that generate T . Let p be a seminorm on V. Then p is
continuous if and only if ∃κ > 0 and p1 , ..., pm ∈ Γ where m ∈ N such that

∀x ∈ V, p(x) ≤ κ max{pi (x)}m

i=1 .

Proof. Forward Direction: Assume that p is continuous. I will show that ∃κ > 0 and
p1 ..pm ∈ Γ where m ∈ N such that ∀x ∈ V, p(x) ≤ κ max{pi (x)}m
i=1 . Notice [0, 1) is open in
[0, +∞). Since p is a continuous function from V to [0, +∞) and [0, 1) is open in [0, +∞), we
get M := p−1 ([0, 1)) is open in V. Note that p(0) = 0 ∈ [0, 1). So 0 ∈ M. So M is an open
neighborhood of 0 in V. Then M contains some basic neighborhood N := N (0, {pi }m
i=1 , ε)
for some m ∈ N, p1 ..pm ∈ Γ, and ε > 0. Consider an arbitrary element x of V. Let rx
denote the number max{pi (x)}m
i=1 .

• Case 1: rx = 0. Then for any k > 0, we have

∀i = 1..m, pi (kx) = kpi (x) ≤ krx = k · 0 = 0 < ε.

So kx ∈ N and hence kx ∈ M. So p(kx) < 1. So

1 1 1
p(x) = p(kx) < · 1 = .
k k k
Since k > 0 was chosen arbitrarily, we get p(x) = 0. So p(x) = rx ≤ 1 · rx .

• Case 2: rx > 0. Then we have

ε ε ε ε
∀i = 1..m, pi ( x) = pi (x) ≤ rx = < ε.
2rx 2rx 2rx 2
ε ε ε
So x ∈ N and hence x ∈ M. So p( x) < 1. So
2rx 2rx 2rx
2rx ε 2rx 2
p(x) = p( x) < · 1 = · rx .
ε 2rx ε ε

2
Take κ := max{1, }. Then p(x) ≤ κrx . That is,
ε
∀x ∈ V, p(x) ≤ κ max{pi (x)}m
i=1 .

Backward Direction: Assume that ∃κ > 0 and p1 ..pm ∈ Γ where m ∈ N such that
∀x ∈ V, p(x) ≤ κ max{pi (x)}m
i=1 . I will show that p is continuous. Note that N :=
92 CHAPTER 17. LOCALLY CONVEX SPACE

N (0, {pi }m
i=1 , 1) is an open neighborhood of 0 in V. Consider an arbitrary element x of N .
Then ∀i = 1..m, pi (x) < 1. So

p(x) ≤ κ max{pi (x)}m

i=1 < κ · 1 = κ.

So p is bounded above by κ on an open neighborhood of 0. So p is continuous.

PROPOSITION 17.21. Let (V, TV ) and (W, TW ) be two locally convex spaces. Let
ΓV be a separating family of seminorms on V that generate TV . Let ΓW be a separating
family of seminorms on W that generate TW . Let T be a linear map from V to W.
Then the followings are equivalent.

1. T is continuous.

2. ∀q ∈ ΓW , ∃κ > 0, ∃p1 , ..., pm ∈ ΓV such that

∀x ∈ V, q(T x) ≤ κ max{pi (x)}m

i=1 .

Proof. Forward Direction Assume that T is continuous. I will show that (2) holds. Let
q be an arbitrary element of ΓW . Then q is a continuous seminorm. Note that q ◦ T is also
a continuous seminorm. By the previous proposition, (2) holds.
Backward Direction: Assume that (2) holds. I will show that T is continuous. Let
U ∈ U0W . Let q1 , ..., qn ∈ ΓW such that NW := N (0, {q1 , ..., qn }, ε) ⊆ U . For each i ∈
{1, ..., n}, since qi ∈ ΓW , qi is continuous. So each qi ◦ T is continuous on V. So NV :=
N (0, {q1 ◦ T, ..., qn ◦ T }, ε) ∈ U0V . Let x be an arbitrary element of NV . Then T x ∈ NW ⊆ U .
So T is continuous at 0 and hence continuous on V.

COROLLARY 17.22. Let (V, T ) be a locally convex space. Let f be a linear

functional on V. Then f is continuous if and only if there is a continuous seminorm p
on V such that
∀x ∈ V, |f (x)| ≤ p(x).

17.5 Convergence in Locally Convex Spaces

PROPOSITION 17.23. Let (V, T ) be a locally convex space. Let Γ be a separating

family of seminorms on V that generates the original topology T . Let (xλ )λ∈Λ be a net
17.6. STRONG OPERATOR TOPOLOGY 93

in V. Then (xλ )λ∈Λ converges to a point x ∈ V if and only if

∀p ∈ Γ, lim p(xλ − x) = 0.
λ∈Λ

17.6 Strong Operator Topology

17.7 Weak Operator Topology

94 CHAPTER 17. LOCALLY CONVEX SPACE
Chapter 18

The Hahn-Banach Theorem

18.1 The Extension Theorems

THEOREM 18.1 (The Hahn-Banach Theorem - 1). Let V be a vector space over
field R. Let M be a linear manifold in V. Let p be a sublinear functional on V. Let
f be a linear functional on M. Suppose that ∀m ∈ M, we have f (m) ≤ p(m). Then
there exists a linear functional g on V such that g|M = f and ∀x ∈ V, g(x) ≤ p(x).

THEOREM 18.2 (The Hahn-Banach Theorem - 2). Let V be a vector space over
field K. Let M be a linear manifold in V. Let p be a seminorm on V. Let f be a linear
functional on M. Suppose that ∀m ∈ M, we have |f (m)| ≤ p(m). Then there exists a
linear functional g on V such that g|M = f and ∀x ∈ V, |g(x)| ≤ p(x).

COROLLARY 18.3. Let (V, T ) be a locally convex space. Let M be a linear

manifold in V. Let f ∈ M∗ . Then ∃g ∈ V ∗ such that g|M = f .

Proof. Since (V, T ) is a locally convex space, there exists a separating family ΓV of semi-
norms that generate T . Since (V, T ) is a locally convex space and M is a linear manifold
in V, M is also a locally convex space. Note that ΓM := {p|M : p ∈ ΓV } is a separating
family of seminorms on M that generates TM . Since f ∈ M∗ , ∃κ > 0 and p1 , ..., pn ∈ ΓV
such that
∀m ∈ M, |f (m)| ≤ κ max{p1 (m), ..., pn (m)}.

95
96 CHAPTER 18. THE HAHN-BANACH THEOREM

Define a function q : V → R by

q(x) := κ max{p1 (x), ..., pn (x)}.

Then q is a continuous seminorm on V and we have ∀m ∈ M, |f (m)| ≤ q(m). By the

Hahn-Banach Theorem - 2, there exists a linear functional on V such that g|M = f and
∀x ∈ V, |g(x)| ≤ q(x). Since g is a linear functional on V and q is a continuous seminorm
on V such that ∀x ∈ V, |g(x)| ≤ q(x), we get that g is continuous.

THEOREM 18.4 (The Hahn-Banach Theorem - 3). Let X be a normed linear space.
Let M be a linear manifold in X. Let f ∈ M∗ . Then ∃g ∈ X∗ such that g|M = f and
that kgk = kf k.

COROLLARY 18.5. Let V be a locally convex space. Let {xi }m

i=1 be a linearly
independent set of vectors in V where m ∈ N. Let k1 , ..., km be arbitrary elements of
K. Then ∃g ∈ V ∗ such that ∀i = 1..m, g(xi ) = ki .

COROLLARY 18.6. Let V be a locally convex space. Let M be a finite-dimensional

linear manifold of V. Then M is topologically complemented.

Proof. Let {mi }ni=1 be a basis for M where n = dim(M). Then {mi }ni=1 is a linearly
independent set of vectors in V. By Corollary 18.5, for each i = 1..n, ∃ρi ∈ V ∗ such that
m
\
ρi (mj ) = δi,j . Define Y := ker(ρi ). Since the ρi ’s are continuous, the ker(ρi )’s are closed.
i=1
So Y is closed. Since dim(M) < ∞, M is closed.
Now I will show that V = M + Y. Let v be an arbitrary element of V. Define for i = 1..n
n
X
a scalar ki as ki := ρi (v). Define a point m as m := ki mi . Then m ∈ M. Define a point
i=1
y as y := v − m. Then ∀i = 1..n, we have
n
X n
X
ρi (y) = ρi (v − m) = ρi (v − kj mj ) = ρi (v) − kj ρi (mj )
j=1 j=1
n
X
= ki − kj δi,j = ki − ki = 0.
j=1
18.1. THE EXTENSION THEOREMS 97

n
\
That is, ρi (y) = 0. So ∀i = 1..n, y ∈ ker(ρi ). So y ∈ ker(ρi ) = Y. So ∀v ∈ V, v = m + y
i=1
where m ∈ M and y ∈ Y. So V = M + Y.
Now I will show that M ∩ Y = {0}. Note that 0 ∈ M ∩ Y. Let z be an arbitrary element
n
X
n
of M ∩ Y. Since z ∈ M, there exist scalars {rj }j=1 such that z = rj mj . On one hand,
j=1
n
X
since z = rj mj , ∀i = 1..n, we have
j=1

Xn n
X n
X
ρi (z) = ρi ( rj mj ) = rj ρi (mj ) = rj δi,j = ri .
j=1 j=1 j=1

n
\
That is, ρi (z) = ri . On the other hand, since z ∈ Y = ker(ρi ), ∀i = 1..n, we have
i=1
n
X
ρi (z) = 0. So ∀i = 1..n, ri = 0. So z = rj mj = 0. So M ∩ Y = {0}.
j=1
So M is topologically complemented by Y.

COROLLARY 18.7. Let X be a normed linear space. Let x ∈ X. Then

kxk = max{|x∗ (x)| : x∗ ∈ X∗ , kx∗ k ≤ 1}.

i.e., ∃x∗ ∈ X∗ with kx∗ k = 1 such that kxk = |x∗ (x)|.

COROLLARY 18.8. Let X be a normed linear space. Then the canonical embedding
J : X → X∗∗ is an isometry.

Proof. Let x be an arbitrary element of X. We are to prove that kxkX = kJxkX∗∗ . Let x̂
denote Jx. On one hand, for any y ∗ ∈ X∗ , we have

|x̂(y ∗ )| = |y ∗ (x)| ≤ ky ∗ kkxk.

So kx̂k ≤ kxk. On the other hand, by Corollary 18.7, there exists x∗ ∈ X∗ with kx∗ k ≤ 1
such that |x∗ (x)| = kxk. So

kx̂k ≥ |x̂(x∗ )| = |x∗ (x)| = kxk.

That is, kx̂k ≥ kxk. Since ∀x ∈ X, kxk = kJxk, we have that J is an isometry.
98 CHAPTER 18. THE HAHN-BANACH THEOREM

COROLLARY 18.9. Let X be a normed linear space. Let Y be a closed subspace

of X. Let z ∈ X \ Y. Then ∃x∗ ∈ X∗ with kx∗ k = 1 such that x∗ |Y = 0 and
x∗ (z) = d(z, Y).

/ Y, Y 6= z + Y. By Corollary 18.7, ∃ξ ∗ ∈ (X/Y)∗ with kξ ∗ k = 1 such that

Proof. Since z ∈
|ξ ∗ (z + Y)| = kz + Yk = d(z, Y). Let q be the canonical quotient map from X to X/Y.
Define a map from X to K as x∗ := ξ ∗ ◦ q.
Show that x∗ ∈ X∗ :
Clearly x∗ is linear. Recall that kξ ∗ k = 1 and that q is a contraction map and hence
kqk ≤ 1. So kx∗ k ≤ kξ ∗ kkqk ≤ 1. So x∗ ∈ X∗ .
Show that kx∗ k = 1:
Since kξ ∗ k = 1, we can find a sequence (tn )n∈N in X/Y such that ∀n ∈ N, we have
1
ktn k ≤ 1 and 1 − < |ξ ∗ (tn )| ≤ 1. So lim |ξ ∗ (tn )| = 1. Define for each n ∈ N a point
n n∈N
n
xn ∈ X to be such that q(xn ) = tn . Then ∀n ∈ N, we have
n+1
n n
kxn + Yk = kq(xn )k = k tn k = ktn k < ktn k ≤ 1.
n+1 n+1
That is, kxn + Yk < 1. So ∀n ∈ N, ∃yn ∈ Y such that kxn + yn k < 1. On the other hand,
we have

lim |x∗ (xn + yn )| = lim |ξ ∗ (q(xn + yn ))| = lim |ξ ∗ (xn + yn + Y)| = lim kxn + yn + Yk
n∈N n∈N n∈N n∈N
∗ ∗
= lim kxn + Yk = lim |ξ (xn + Y)| = lim |ξ (q(xn ))|
n∈N n∈N n∈N
∗ n n ∗
= lim |ξ ( tn )| = lim | ξ (tn )|, by linearity of ξ ∗
n∈N n+1 n∈N n + 1
n
= lim · lim |ξ ∗ (tn )| = 1 · 1 = 1.
n∈N n + 1 n∈N

That is, lim |x∗ (xn + yn )| = 1. Since ∀n ∈ N, kxn + yn k < 1 and lim |x∗ (xn + yn )| = 1, we
n∈N n∈N
get kx∗ k ≥ 1. Recall that we have proved kkx∗ k ≤ 1. So kkx∗ k = 1.
Show that x∗ |Y = 0:
Let y be an arbitrary element of Y. Then we have

x∗ (y) = ξ ∗ (q(y)) = ξ ∗ (y + Y) = d(y, Y) = 0.

So x∗ |Y = 0.
Show that x∗ (z) = d(z, Y):
Note that
x∗ (z) = |ξ ∗ (q(z))| = |ξ ∗ (z + Y)| = d(z, Y).
That is, x∗ (z) = d(z, Y).
18.2. SEPARATION RESULTS 99

18.2 Separation Results

(bug)

PROPOSITION 18.10. Let V be a locally convex space over field K. Let G be a

non-empty, open, and convex set in V. Suppose that 0 ∈
/ G. Then there exists a closed
hyperplane M in V such that G ∩ M = ∅.

Proof.
Case 1: K = R.
Since G 6= ∅, take x0 ∈ G. Define a set H as H := x0 − G. Then H is non-empty, open,
convex, and 0 ∈ H. Let pH denote the Minkowski functional on H. Since H is an open
convex neighborhood of 0, H = {x ∈ V : pH (x) < 1}. Define a set W by W := Rx0 . Then
W is a linear manifold in V. Define a map f : W → R by f (kx0 ) := kpH (x0 ). Then f is a
linear functional on W. Note that

f (kx0 ) = kpH (x0 ) = pH (kx0 ), for k ≥ 0, and

f (kx0 ) = kpH (x0 ) < 0 ≤ pH (kx0 ), for k < 0.

not finished

THEOREM 18.11 (The Hahn-Banach Theorem - 4). Let V be a locally convex

space. Let A and B be two non-empty, open, convex, and disjoint sets in V. Then
∃f ∈ V ∗ , ∃κ ∈ R such that

∀a ∈ A, b ∈ B, <f (a) > κ > <f (b).

THEOREM 18.12 (The Hahn-Banach Theorem - 5). Let V be a locally convex

space. Let A and B be two non-empty, closed, convex, and disjoint sets in V. Suppose
B is compact. Then ∃f ∈ V ∗ , ∃α, β ∈ R such that

∀a ∈ A, b ∈ B, <f (a) ≥ α > β ≥ <f (b).

100 CHAPTER 18. THE HAHN-BANACH THEOREM

COROLLARY 18.13. Let V be a locally convex space. Let A be a non-empty set in

V. Then the closed convex hull conv(A) equals the intersection of all closed half-spaces
that contain A.

Proof. Let Ω denote the set of all closed half-spaces that contain A.
Forward Direction:
\
Note that ∀S ∈ Ω, S is closed and convex. So S is closed and convex. Note also
S∈Ω
\ \
that A ⊆ S. So conv(A) ⊆ S.
S∈Ω S∈Ω
Backward Direction:
Let z be an arbitrary element outside conv(A). Then conv(A) and {z} are two non-
empty, closed, convex, and disjoint sets and we have that {z} is compact. By the Hahn-
Banach theorem, version 5, ∃f ∈ V ∗ , ∃α, β ∈ R such that

∀a ∈ conv(A), <f (a) ≥ α > β ≥ <f (z).

Define a set S0 by S0 := {x ∈ V : <f (x) ≥ α}. Then S0 is a closed half-space of V and

\ \
z∈
/ S0 . So z ∈
/ S. So S ⊆ conv(A).
S∈Ω S∈Ω
Chapter 19

Reflexive Banach Space

19.1 Definitions

DEFINITION 19.1 (Reflexive). Let X be a Banach space. Let J denote the canonical
embedding of X into X∗∗ . We say that X is reflexive if J is an isometric isomorphism
between X and X∗∗ .

19.2 Properties
(bug * 2)

PROPOSITION 19.2 (Closed Subspace). Let X be a reflexive Banach space. Let

Y be a closed subspace of X. Then Y is a reflexive Banach space.

Proof. Since X is a Banach space and Y is a closed subspace of X, Y is a Banach space. Let
J denote the canonical embedding of Y into Y∗∗ . By the Hahn-Banach Theorem (Corollary
18.8), J is an isometry and hence automatically injective.
To see that J is surjective, consider an arbitrary element y ∗∗ ∈ Y∗∗ . Define a map
x∗∗ : X∗ → K by x∗∗ (z ∗ ) := y ∗∗ (z ∗ |Y ). Then it is easy to check that x∗∗ is linear and
continuous. So x∗∗ ∈ X∗∗ . Since X is reflexive, ∃x ∈ X such that x̂ = x∗∗ .
Assume for the sake of contradiction that x ∈
/ Y. Since Y is a closed subspace of X and
x ∈ X \ Y, By the Hahn-Banach Theorem, ∃g ∈ X∗ such that g|Y = 0 and g(x) 6= 0. Now
we have x̂(g) = g(x) 6= 0 and x∗∗ (g) = y ∗∗ (g|Y ) = y ∗∗ (0) = 0. So x̂(g) 6= x∗∗ (g). However,
this contradicts to the previous conclusion that x̂ = x∗∗ . So x ∈ Y.

101
102 CHAPTER 19. REFLEXIVE BANACH SPACE

Now I will show that J(x) = y ∗∗ . Consider an arbitrary element w∗ ∈ Y∗ . Let v ∗ ∈ X∗

be a Hahn-Banach extension of w∗ . Then

y ∗∗ (w∗ ) = y ∗∗ (v ∗ |Y ), since v ∗ is an extension of w∗

= x∗∗ (v ∗ ), by definition of x∗∗
= x̂(v ∗ ), by the choice of x
= v ∗ (x), by definition of x̂
= w∗ (x), since v ∗ is an extension of w∗
= J(x)(w∗ ), by definition of J(x).

That is, y ∗∗ (w∗ ) = J(x)(w∗ ). So J(x) = y ∗∗ . So J is surjective. This completes the proof.

PROPOSITION 19.3 (Dual Space). Let X be a Banach space. Then X is reflexive

if and only if X∗ is reflexive.

Proof. Forward Direction: Assume that X is reflexive. I will show that X∗ is reflexive.

not finished

PROPOSITION 19.4 (Image under Isometric Isomorphism). Let X and Y be

normed linear spaces. Suppose that X is reflexive and that there is an isometric iso-
morphism between X and Y.

PROPOSITION 19.5. Let X be a Banach space. Then X is reflexive if and only if

X1 is weakly compact.

Proof. Forward Direction: Assume that X is reflexive. I will show that X1 is weakly
compact. Since X∗ is a Banach space, by the Banach-Alaoglu Theorem, X∗∗
1 is weak*-
compact. i.e., X∗∗ ∗∗ ∗∗ ∗ ∗∗
1 is compact in the space (X , σ(X , X )). Since X is reflexive, X̂ = X .
So X is isometrically isomorphic to X∗∗ . So X1 is compact in the space (X, σ(X, X∗ )). i.e.,
X1 is weakly compact.
Backward Direction: Assume that X1 is weakly compact. I will show that X is
c1 is compact in the space (X∗∗ , σ(X∗∗ , X∗ )). i.e., X
reflexive. X c1 is weak*-compact. Since
19.2. PROPERTIES 103

the weak* topology is Hausdorff and Xc1 is weak*-compact, Xc1 is weak*-closed. Since X
is a Banach space, by the Goldstine’s Theorem, Xc1 is weak*-dense in X∗∗ . Since Xc1 is
1
∗∗ c ∗∗ ∗∗
weak*-closed and weak*-dense in X , X1 = X . So X̂ = X . So X is reflexive.
1 1

don’t understand
104 CHAPTER 19. REFLEXIVE BANACH SPACE
Chapter 20

Weak Topology

20.1 Definitions

DEFINITION 20.1 (Dual Pair). Let V be a vector space over field K. Let L be a
linear manifold of V # and a separating family of linear functionals on V. We define a
dual pair to be the pair (V, L).

DEFINITION 20.2 (Topology Generated by Linear Functionals). Let (V, L) be a

dual pair. Define for each ϕ ∈ L a function pϕ : V → R by pϕ (x) := |ϕ(x)|. Then each
pϕ is a seminorm on V. Define a family ΓL of seminorms on V by ΓL := {pϕ : ϕ ∈ L}.
Then ΓL is a separating family of seminorms on V. We define the topology generated
by L, denoted by σ(V, L), to be the locally convex topology generated by ΓL .

DEFINITION 20.3 (Weak Topology). Let (V, T ) be a locally convex space. Then
V ∗ is a linear manifold of V # . By the Hahn-Banach theorem, we get V ∗ is separating.
So (V, V ∗ ) is a dual pair. We define the weak topology on V to be the topology
σ(V, V ∗ ) induced by the family V ∗ .

DEFINITION 20.4 (Weak* Topology). Let (V, T ) be a locally convex space. Then
V̂ is a linear manifold of (V ∗ )# and a separating family of linear functionals on V ∗
So (V ∗ , V̂) is a dual pair. We define the week* topology on V ∗ to be the topology

105
106 CHAPTER 20. WEAK TOPOLOGY

σ(V ∗ , V̂) induced by the family V̂.

20.2 Properties

PROPOSITION 20.5. Let (V, L) be a dual pair where V is a vector space over field
K and L is a linear manifold of L# . Then L = (V, σ(V, L))∗ .

Proof. Forward Direction: Clearly if f ∈ L, then f is σ(V, L)-continuous and hence

f ∈ (V, σ(V, L))∗ .
Backward Direction: Let f ∈ (V, σ(V, L))∗ . I will show that f ∈ L. Define a seminorm
p : V → R by p(x) := |f (x)|. Since f is σ(V, L)-continuous, p is also σ(V, L)-continuous. So
∃n ∈ N, ∃ρ1 , ..., ρn ∈ L, ∃κ ∈ R such that κ > 0 and that ∀x ∈ V, p(x) ≤ κ max{|ρi (x)|}ni=1 .
\n
Then if x ∈ ker(ρi ), |f (x)| = p(x) ≤ κ max{0}ni=1 = 0 and hence x ∈ ker(f ). So
i=1
n
\
ker(ρi ) ⊆ ker(f ). So f ∈ span{ρi }ni=1 and hence f ∈ L. This completes the proof.
i=1

PROPOSITION 20.6. Let (V, T ) be a topological space. Let (xλ )λ∈Λ be a net
in (V, T ) that converges to some point x in V. Then (xλ )λ∈Λ also converges to x in
(V, σ(V, V ∗ )).

PROPOSITION 20.7. Let (V, TV ) and (W, TW ) be locally convex spaces. Let T be
a continuous linear map from (V, TV ) to (W, TW ). Then T is also a continuous linear
map from (V, σ(V, V ∗ )) to (W, σ(W, TW )).

PROPOSITION 20.8. The weak* topology is a weaker topology than the weak
topology.

Proof Idea. The week* topology is a topology induced by the pre-dual and the weak topology
is a topology induced by the dual. But the pre-dual sits inside the dual. Therefore it is
harder to converge in the weak topology because there are more functionals that have to be
continuous. You’ll need more open sets to make those extra functionals continuous.
20.3. THEORY ON BANACH SPACES 107

20.3 Theory on Banach Spaces

PROPOSITION 20.9. Let X be a finite-dimensional Banach space. Then the norm,

weak, and weak* topologies on X all coincide.

PROPOSITION 20.10. Let X be a Banach space. Let X∗ denote the dual space of
X. Let τ∗ denote the weak topology on X∗ induced by elements of X as

lim x∗α = x∗ ⇐⇒ ∀x ∈ X, lim x∗α (x) = x∗ (x).

α α

Then (X∗ , τ∗ ) is a topological vector space.

20.3.1 The Uniform Boundedness Principle

THEOREM 20.11 (The Uniform Boundedness Principle). Let X and Y be Banach

spaces. Let A be a family of bounded linear maps from X to Y. Suppose that ∀x ∈ X,
we have Mx := sup{kT xkY : T ∈ A} < ∞. Then sup{kT k : T ∈ A} < ∞.

COROLLARY 20.12. Let X be a Banach space. Let S be a set in X. Then S is

bounded if and only if

∀x∗ ∈ X∗ , sup{|x∗ (s)| : s ∈ S} < ∞.

COROLLARY 20.13. Let X be a Banach space. Let S be a subset of X∗ . Then S

is bounded if and only if

∀x ∈ X, sup{|s∗ (x) : s∗ ∈ S} < ∞.

20.3.2 The Banach-Steinhaus Theorem

(bug)
108 CHAPTER 20. WEAK TOPOLOGY

THEOREM 20.14 (The Banach-Steinhaus Theorem). Let X and Y be Banach

spaces. Let (Tn )n∈N be a sequence in B(X, Y). Suppose that ∀x ∈ X, ∃yx ∈ Y such
that lim Tn x = yx . Define a map T : X → Y by T x := yx . Then we have the followings:
n∈N

1. sup kTn k < +∞.

n∈N

2. T is a bounded linear map and kT k ≤ lim inf kTn k.

n∈N

Proof. Part 1: Since ∀x ∈ X, lim Tn x = yx , we have Mx := sup kTn xk < ∞. By the

n∈N n∈N
Uniform Boundedness Principle, we get M := sup kTn k < ∞.
n∈N
Part 2: Clearly T is linear. Let x be an arbitrary element of X. Then

kT xk = kyx k = k lim Tn xk ≤ lim inf kTn kkxk = (lim inf kTn k) · kxk.
n→∞ n→∞ n∈N

So kT k ≤ lim inf kTn k. So T is a bounded linear map.

n∈N

not finished
don’t know why

COROLLARY 20.15. Let X be a Banach space. Let (xn )n∈N be a sequence that
converges to x in (X, σ(X, X∗ )). Then sup kxn k < ∞ and kxk ≤ lim inf kxn k.
n∈N n∈N

COROLLARY 20.16. Let X be a Banach space. Let (x∗n )n∈N be a sequence that
converges to x∗ in (X∗ , σ(X∗ , X)). Then sup kx∗n k < ∞ and kx∗ k ≤ lim inf kx∗n k.
n∈N n∈N

20.3.3 The Banach-Alaoglu Theorem

THEOREM 20.17 (The Banach-Alaoglu Theorem). Let X be a Banach space. Then

the closed unit ball X∗1 of X∗ is weak*-compact.
20.3. THEORY ON BANACH SPACES 109

COROLLARY 20.18. Let X be a Banach space. Then X is isometrically isomorphic

to a subspace of (C(L, K), k · k∞ ) where L is some compact Hausdorff space.

20.3.4 Goldstine’s Theorem

THEOREM 20.19 (Goldstine’s Theorem). Let X be a Banach space. Let J denote

the canonical embedding of X into X∗∗ . Then J(X1 ) is weak*-dense in X∗∗
1 .

20.3.5 Metrizability

PROPOSITION 20.20. Let X be a Banach space. Then X1 is weakly metrizable if

and only if X∗ is separable.

PROPOSITION 20.21. Let X be a Banach space. Then X∗1 is weak*-metrizable if

and only if X is separable.
110 CHAPTER 20. WEAK TOPOLOGY
Chapter 21

Locally Compact Space

21.1 The F. Riesz’s Theorem

THEOREM 21.1 (The F. Riesz’s Theorem). Let X be a topological vector space

over field K. Then X is finite-dimensional if and only if it is locally compact.

111
112 CHAPTER 21. LOCALLY COMPACT SPACE
Chapter 22

Adjoint Operator

22.1 Banach Space Adjoint

DEFINITION 22.1 (Banach Space Adjoint). Let X and Y be Banach spaces. Let
T be a bounded linear operator from X to Y. We define the Banach space adjoint
of T , denoted by T ∗ , to be a bounded linear operator from Y∗ to X∗ given by

T ∗ y ∗ (x) := y ∗ T x.

PROPOSITION 22.2. kT ∗ k = kT k.

PROPOSITION 22.3. T ∗ is invertible if and only if T is invertible.

22.2 Hilbert Space Adjoint

DEFINITION 22.4 (Adjoint Matrix). Let A be an m × n matrix. We define the

adjoint of A, denoted by A∗ , to be an n × m matrix given by

(A∗ )ij := (A)ji .

113
114 CHAPTER 22. ADJOINT OPERATOR

DEFINITION 22.5 (Adjoint Operator). Let V and W be inner product spaces. Let
T be a linear map from V to W . We define the adjoint of T , denoted by T ∗ , to be a
map from W to V such that

∀x ∈ V, ∀y ∈ W, hT (x), yiW = hx, T ∗ (y)iV .

PROPOSITION 22.6 (Existence). Let V be a finite-dimensional inner product

space and T be a linear operator on V . Then the adjoint of T exists.

PROPOSITION 22.7 (Uniqueness). Let V be an inner product space and T be a

linear operator on V . Then the adjoint of T is unique, provided that it exists.

22.3 Properties of the Adjoint Operator

PROPOSITION 22.8. Let V be an inner product space. Then

1. (IV )∗ = IV where IV is the identity operator on V .

2. T ∗∗ = T for any linear operator T on V .

PROPOSITION 22.9. Let V be an inner product space and T be a linear operator

on V . Then T ∗ is also linear.

PROPOSITION 22.10. Let V be an inner product space. Then

1. For any linear operators T and U ,

(T + U )∗ = T ∗ + U ∗ .

2. For any linear operator T ,

(cT )∗ = c · T ∗ .
22.4. NORMAL OPERATORS 115

3. For any linear operator T and U ,

(T U )∗ = U ∗ T ∗ .

PROPOSITION 22.11. Let X be a finite-dimensional Hilbert space. Let T be an

invertible linear operator on X. Let T ∗ denote the Hilbert space adjoint of T . Then
T ∗ is also invertible and we have

(T −1 )∗ = (T ∗ )−1 .

PROPOSITION 22.12. Let X and Y be finite-dimensional Hilbert spaces. Let T

be a linear operator from X to Y. Let T ∗ denote the Hilbert space adjoint of T . Then

ran(T ∗ ) = (ker(T ))⊥ .

Proof. Forward Direction: I will show that ran(T ∗ ) ⊆ (ker(T ))⊥ . Let x be an arbitrary
element of ran(T ∗ ) ⊆ X. Then ∃y ∈ Y such that x = T ∗ y. Let k be an arbitrary element of
ker(T ). Then T k = 0. So

hk, xi = hk, T ∗ yi = hT k, yi = h0, yi = 0.

That is, ∀k ∈ ker(T ), we have hx, ki = 0. So x ∈ (ker(T ))⊥ . So ran(T ∗ ) ⊆ (ker(T ))⊥ .
Backward Direction: I will show that (ker(T ))⊥ ⊆ ran(T ∗ ). Let x be an arbitrary
element of (ker(T ))⊥ ⊆ X. Assume for the sake of contradiction that x ∈
/ ran(T ∗ ). Then
∃x0 ∈ (ran(T ∗ ))⊥ such that hx, x0 i =
6 0. Note that T ∗ T x0 ∈ ran(T ∗ ). Since x0 ∈ (ran(T ∗ ))⊥
and T ∗ T x0 ∈ ran(T ∗ ), we have hx0 , T ∗ T x0 i = 0. So

hT x0 , T x0 i = hx0 , T ∗ T x0 i = 0.

So T x0 = 0. So x0 ∈ ker(T ). Since x0 ∈ ker(T ) and hx, x0 i 6= 0, we get x ∈

/ (ker(T ))⊥ . This
contradicts to the assumption that x ∈ (ker(T ))⊥ . So x ∈ ran(T ∗ ). So (ker(T ))⊥ ⊆ ran(T ∗ ).
This completes the proof.

22.4 Normal Operators

DEFINITION 22.13 (Normal). Let V be an inner product space and T be a linear

116 CHAPTER 22. ADJOINT OPERATOR

operator on V . We say that T is normal if T T ∗ = T ∗ T .

Chapter 23

Convolution

DEFINITION 23.1 (Convolution). Let f and g be functions from R to R. We define

the convolution of f and g, denoted by f ∗ g, to be a function on R given by
Z +∞
(f ∗ g)(t) := f (τ )g(t − τ )dt.
−∞

117
118 CHAPTER 23. CONVOLUTION
Chapter 24

Coercive Functions

24.1 Definitions

DEFINITION 24.1 (Coercive). Let f be a function from Rd to R∗ . We say that f

is coercive if lim f (x) = +∞.
kxk→∞

24.2 Properties

PROPOSITION 24.2. Let f be a proper lower semi-continuous function from Rd

to R∗ . Let K be a compact set in Rd . Assume K ∩ dom(f ) 6= ∅. Then f attains its
minimum over K.

Proof.
Define m := inf f (x).
x∈K
Since m = inf f (x), there exists a sequence {xi }i∈N in K such that lim f (xi ) = m.
x∈K i→∞
Since K is compact and {xi }i∈N ⊆ K, there exists a convergent subsequence {xi }i∈I in
K where I is an infinite subset of N.
Say the limit is x∞ where x∞ ∈ K.
Since lim f (xi ) = m, we get lim f (xi ) = m.
i→∞ i∈I,i→∞
Since lim f (xi ) = m, we get lim inf f (xi ) = m.
i∈I,i→∞ i∈I,i→∞
Since f is lower semi-continuous and lim xi = x∞ , we get f (x∞ ) ≤ lim inf xi .
i∈I,i→∞ i∈I,i→∞
That is, f (x∞ ) ≤ m.

119
120 CHAPTER 24. COERCIVE FUNCTIONS

Since m = inf f (x), we have ∀x ∈ K, f (x) ≥ m.

x∈K
In particular, f (x∞ ) ≥ m.
Since f (x∞ ) ≥ m and f (x∞ ) ≤ m, f (x∞ ) = m.
Since f is proper, f (x∞ ) = m 6= −∞.
So f attains its minimum at point x∞ .

PROPOSITION 24.3. Let f be a proper, lower semi-continuous, and coercive func-

tion from Rd to R∗ . Let C be a closed subset of Rd . Assume C ∩ dom(f ) 6= ∅. Then f
attains its minimum over C.

Proof.
Since C ∩ dom(f ) 6= ∅, take x ∈ C ∩ dom(f ).
Since f is coercive, ∃R such that ∀y, kyk > R, we have f (y) ≥ f (x).
Since x ∈ C ∩ dom(f ) and ∀y, kyk > R, we have f (y) ≥ f (x), the set of minimizers of f
over C is the same as the set of minimizers of f over C ∩ ball[0, R].
Since C and ball[0, R] are both closed, C ∩ ball[0, R] is closed.
Since ball[0, R] is bounded, C ∩ ball[0, R] is bounded.
Since C ∩ ball[0, R] is closed and bounded, by the Heine-Borel Theorem, C ∩ ball[0, R]
is compact.
Since f is proper and lower semi-continuous and C ∩ ball[0, R] is compact, f attains its
minimum over C ∩ ball[0, R].
So f attains its minimum over C.
Chapter 25

Unclassified Results

PROPOSITION 25.1. Let (X, d) be a compact metric space. Let L(X) be the set
of all Lipschitz functions from X to R. Let C(X) be the set of all continuous functions
from X to R. Then L(X) is dense in C(X).

121