MA 2049

MA 2050

Introduction to Mathematical
Analysis

Chun-Hsiang Tsou
chtsou@[Link]

Department of Mathematics - National Central University

The 114th Scholar Year, Taoyuan, Taiwan (R.O.C.)
2
Contents

0 Preliminary 13
0.1 Vocabularies on Sets and Functions . . . . . . . . . . . . . . . . . . . . . . . 13
0.1.1 Mathematical Propositions . . . . . . . . . . . . . . . . . . . . . . . 13
0.1.2 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
0.1.3 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
0.1.4 Number Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
0.2 Countability of Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
0.3 Algebraic Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
0.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

1 System of Real Numbers 29

1.1 Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.2 Subsets in Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.3 Sequences in Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.3.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.3.2 Properties on Convergent Sequences . . . . . . . . . . . . . . . . . . 37
1.3.3 Extraction of Sequences . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.3.4 Cauchy Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.4 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.5 Limit Inferior and Limit Superior . . . . . . . . . . . . . . . . . . . . . . . . 47
1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.A Construction of Number Systems (Exercises) . . . . . . . . . . . . . . . . . 56

2 Numerical Series 69
2.1 General Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.1.1 Telescoping Series and Partial Sums . . . . . . . . . . . . . . . . . . 71
2.2 Series with Positive General Terms . . . . . . . . . . . . . . . . . . . . . . . 72
2.2.1 Generalities and Comparison Tests . . . . . . . . . . . . . . . . . . . 72
2.2.2 Other Test Methods and Asymptotic Behaviors . . . . . . . . . . . . 76
2.3 Series with Complex General Terms . . . . . . . . . . . . . . . . . . . . . . 79
2.3.1 Convergences and Absolute Convergences . . . . . . . . . . . . . . . 80
2.3.2 Permutation of Terms . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.3.3 Comparison with Integrals . . . . . . . . . . . . . . . . . . . . . . . . 84
2.3.4 Alternate Series, Abel Transforms and Cauchy Products . . . . . . . 87
2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3
4 CONTENTS

3 Topology - Elementary Introduction 99

3.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.2 Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.3 Basic Notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.4 Open Sets, Closed Sets and Neighborhoods . . . . . . . . . . . . . . . . . . 114
3.5 Interior and Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.6 Limits and Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.6.1 Limits and Continuity at a Point . . . . . . . . . . . . . . . . . . . . 123
3.6.2 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.6.3 Uniform and Lipschitz Continuity . . . . . . . . . . . . . . . . . . . 128
3.6.4 Continuous Linear Applications . . . . . . . . . . . . . . . . . . . . . 130
3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

4 Topology - Fundamental Theories 141

4.1 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.1.1 Complete Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.1.2 Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.1.3 Banach Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.1.4 Extension by Continuity . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.2 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.2.1 Sequence Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.2.2 Covering Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.2.3 Properties of Compacts Sets . . . . . . . . . . . . . . . . . . . . . . . 157
4.3 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.3.1 General Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.3.2 Arc Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.4 Finite Dimensional Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . 164
4.5 Space of Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

5 Inner Product Spaces 177

5.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.2 Norms on Inner Product Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.3 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.4 Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

6 Differential Calculus 195

6.1 Derivations of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
6.1.1 Directional Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . 195
6.1.2 Differentiation at a Point . . . . . . . . . . . . . . . . . . . . . . . . 196
6.1.3 Continuously Differentiable Functions . . . . . . . . . . . . . . . . . 200
6.2 Operations on Continuously Differential Functions . . . . . . . . . . . . . . 207
6.2.1 Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
6.2.2 Algebraic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.2.3 Curve Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
6.3 Higher Order Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
6.A Inverse and Implicit Function Theorems (Out of Program) . . . . . . . . . . 220
CONTENTS 5

6.A.1 Differential Calculus in Banach Spaces . . . . . . . . . . . . . . . . . 220

6.A.2 Mean Value Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.A.3 Statements of Inverse Function Theorems . . . . . . . . . . . . . . . 224
6.A.4 Proof of Local Inverse Function Theorem . . . . . . . . . . . . . . . 227
6.A.5 Implicit Function Theorem . . . . . . . . . . . . . . . . . . . . . . . 229

7 Sequences and Series of Functions 231

7.1 Sequence of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
7.1.1 Different Types of Convergence . . . . . . . . . . . . . . . . . . . . . 231
7.1.2 Properties of Uniform Convergence . . . . . . . . . . . . . . . . . . . 236
7.2 The Space of Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . 240
7.2.1 Partition on a segment . . . . . . . . . . . . . . . . . . . . . . . . . . 241
7.2.2 Piecewise Continuous Functions . . . . . . . . . . . . . . . . . . . . . 241
7.3 Series of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.3.1 Different Types of Convergence . . . . . . . . . . . . . . . . . . . . . 247
7.3.2 Properties on Convergent Series of Functions . . . . . . . . . . . . . 252
7.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
7.A Ascoli-Arzelà Theorem (Out of Program) . . . . . . . . . . . . . . . . . . . 266

8 Riemann Integration 269

8.1 Integrals on a Segment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
8.1.1 Riemann and Darboux Integrability . . . . . . . . . . . . . . . . . . 269
8.1.2 Integrals of Functions in Vector Values . . . . . . . . . . . . . . . . . 278
8.1.3 Properties of Integrals over a Segment . . . . . . . . . . . . . . . . . 285
8.2 Fundamental Theorems on Calculus . . . . . . . . . . . . . . . . . . . . . . 289
8.2.1 Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
8.2.2 Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . 291
8.3 Integrals on Any Interval . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
8.3.1 Integrability of Positive Real-Valued Functions . . . . . . . . . . . . 294
8.3.2 Integrability of Functions in Vector Values . . . . . . . . . . . . . . . 300
8.4 Convergence Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
8.4.1 Integrable and Square Integrable Spaces . . . . . . . . . . . . . . . . 308
8.4.2 Convergence Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 310
8.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314

9 Power Series 323

9.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
9.2 Convergences of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.3 Properties of Sum Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
9.4 Classical Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
9.4.1 Rational Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
9.4.2 Exponentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
9.4.3 Trigonometric and Hyperbolic Functions . . . . . . . . . . . . . . . . 335
9.5 Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
6 CONTENTS

10 Fourier Series 339

10.1 Periodic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
10.1.1 Definitions and General Properties . . . . . . . . . . . . . . . . . . . 339
10.1.2 Inner Product and Square Integrable Semi-Norm . . . . . . . . . . . 340
10.1.3 Trigonometric Polynomials and Series . . . . . . . . . . . . . . . . . 342
10.2 Fourier Coefficients and Series . . . . . . . . . . . . . . . . . . . . . . . . . . 344
10.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
10.2.2 Properties of Fourier Coefficients . . . . . . . . . . . . . . . . . . . . 345
10.3 Convergence Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
10.3.1 Pointwise Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 349
10.3.2 Normal Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
10.3.3 Convergence in Square Integrable Norm . . . . . . . . . . . . . . . . 353
10.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356

11 Fourier Transforms & Distributions 365

11.1 Fourier Transforms of Integrable Functions . . . . . . . . . . . . . . . . . . 367
11.2 Elementary Distribution Theory . . . . . . . . . . . . . . . . . . . . . . . . 373
11.2.1 Space of Test Functions . . . . . . . . . . . . . . . . . . . . . . . . . 373
11.2.2 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
11.2.3 Operations on Distributions . . . . . . . . . . . . . . . . . . . . . . . 378
11.2.4 Convolution Products . . . . . . . . . . . . . . . . . . . . . . . . . . 382
11.3 Fourier Transforms and Distributions . . . . . . . . . . . . . . . . . . . . . . 384
11.3.1 The Schwartz Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
11.3.2 Tempered Distributions . . . . . . . . . . . . . . . . . . . . . . . . . 388
11.3.3 Fourier Transforms on Tempered Distributions . . . . . . . . . . . . 389
11.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394

A Comparison Relations of Functions and Sequences 405

A.1 Domination and Negligence Relations of Functions . . . . . . . . . . . . . . 405
A.2 Equivalence between Functions . . . . . . . . . . . . . . . . . . . . . . . . . 408
A.3 Comparison of Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
A.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415

B Asymptotic Expansions 417

B.1 Definitions and Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
B.1.1 Expansion Near a Real Number . . . . . . . . . . . . . . . . . . . . . 417
B.1.2 Expansions of Usual Functions . . . . . . . . . . . . . . . . . . . . . 421
B.1.3 Expansions at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . 422
B.2 Operations on Asymptotic Expansions . . . . . . . . . . . . . . . . . . . . . 422
B.2.1 Sums and Products . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
B.2.2 Quotients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
B.2.3 Compositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
B.2.4 Integrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428

C Summation over Arbitrary Families 431

C.1 Summable Families of Positive Real Numbers . . . . . . . . . . . . . . . . . 431
C.2 Summable Families of Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . 441
C.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
C.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
CONTENTS 7

Index 457
8 CONTENTS
Notations

Number Systems
N natural numbers, (start with 0)
Z integers
Q rational numbers
R real numbers
C complex numbers
N∗ natural numbers excluded 0, (analogue for Q∗ , Z∗ , R∗ and C∗ )
R+ positive real numbers, (analogue for R− , R∗+ and R∗− )
K an arbitrary field, (the letter K stands for “corp” in french)

Sets and Logic

∅ empty set
{· | · } set definition
∈ and ∈ / belong to and not belong to
⊂ subset of
⊊ strict subset of
⊃ contain
∪ union
∩ intersection
∁ complement
℘ power set
\ relative complement
Card cardinal
∀ for all
∃ there exists
⇒ imply
⇐ sufficient condition
⇔ if, and only if
[[p, q]] integers between p and q

Constants and Usual Functions

9
10 CONTENTS

e Euler’s constant, base of natural logarithm

π ratio of a circle’s circumference to its diameter
i unit imaginary number
Re real part
Im imaginary part
z conjugation of z ∈ C
= equality
̸= not equal
:= definition
+ addition or plus sign
− subtraction or minus sign
× multiplication (usually omitted) or Cartesian product
÷ division
p
fraction p over q
q
! factorial
|·| magnitude, absolute value
arg(z) argument of the complex number z
√
square root
→ mapping or tends to
7 → function definition
f |A restriction of f on A
f ◦g f composited with g
exp(x) or ex exponential of x
ln natural logarithm
sin sinuous
cos cosinuous
tan tangent
cot cotangent
sinh hyperbolic sinuous
cosh hyperbolic cosinuous
tanh ! hyperbolic tangent
p
the number of combinations of q items among p elements
q

Orders and Limits

< strictly less then
≤ less then
> strictly larger then
≥ larger then
min minimum
CONTENTS 11

max maximum
sup supremum
inf infimum
(xn )n∈N sequence of general terms xn
AN the set of sequences with elements in the set A
∞ infinity
lim limit
lim sup limit superior
lim inf limit inferior
X
xn series of general terms xn
Y n
xn product of general terms xn
n
df
f ′ or derivation of the function f (respect to the variable x)
dx
∂α
or ∂ α partial derivation of order α
∂xα Z
integration

O domination relation
o negligence relation
∼ equivalence relation

Spaces and Topology

K[X] ring of polynomials with coefficients in the field K
Mm×n (K) space of matrices in m lines and n columns with coefficients in the field K
u ∗ adjoint of the linear application u
ker(u) kernel of the linear application u
im(u) image of the linear application u
⊕ direct sum
Span generated vector space
dim dimension
det determinant
GL(n) group of invertible matrices of order n
σ(M ) spectrum (the set of eigenvalues) of the matrix M
E vector space, (stands for “espace” in french)
d distance function
∥·∥ norm
(X, d) metric space of elements in X endowed with the distance function d
∞ 1
ℓ , ℓ and ℓ 2 space of bounded, summable and square summable sequences
∞
L and L p space of bounded and power-p integrable functions
C(X, Y ) space of continuous functions mapping from X into Y
12 CONTENTS

C k (E, F ) space of k-times differentiable functions mapping from E into F

C ∞ (E, F ) space of indefinitely differentiable functions mapping from E into F
B(a, r) and B(a, r) open and closed ball centered at a with the radius r
T topology
N (a) neighborhood of a
int(A) or Å interior of A
cl(A) or A closure of A
∂A boundary of A
L (E, F ) space of operators mapping from E into F
Isom(E, F ) set of isomorphisms mapping from E into F
⟨·, ·⟩ inner product or distribution evaluation
⊥ orthogonal, perpendicular
∨ common refinement
σ(f, P, t) Riemann sum
f or F(f )
b Fourier transform of f
D(Ω) space of test functions on Ω
D′ (Ω) space of distributions on Ω
S(Rn ) Schwartz space on Rn
S ′ (Rn ) space of tempered distributions on Rn

Abbrevations
s.t. such that
i.e. “id est” in latin, stands for “that is” or “in other words”
a priori (principally) stands for “in all generalities”
a.e. almost everywhere (used in Real Analysis)
 Chapter 0

Preliminary

In this chapter, we recall several basic definitions of the objects or concepts which will
frequently appear in the entire lecture note. Many of them have already introduced in
other courses such as Calculus and Basic Mathematics during the first year. There are
also some concepts which need to be redefined properly, such as real numbers R, but it is
however very helpful to have a image of the framework in the memory.

0.1 Vocabularies on Sets and Functions

In this section, we recall the notions on sets and functions in all generality.

0.1.1 Mathematical Propositions

• A proposition, also called as an assertion, is a mathematical sentence of which we
can assign a value of truth (True or False). For example, “1 + 1 = 2” is true; “π is an
integer ” is false; “x2 = 1” is true only if x = ±1.

• Usually, a mathematical proposition contain certain variables, which are called as

predicates, and the truth value of the proposition can be determinate once we know
precisely those predicates. For example, “x2 + y 2 = 1” contains the predicates x and
y, and this proposition is true only if (x, y) is situated on the unit circle.

• We say a proposition P implies a proposition Q (or we say Q is a consequence of

P ) and we denote P ⇒ Q if Q were true once P is true. For example, “ABC is a
2 2 2
rectangle triangle at A” implies “BC = AB + AC ”.

• We say two propositions P and Q are equivalent and we denote P ⇔ Q if P and

Q are simultaneously true. For example, Pythagoras theorem states that “ABC is a
2 2 2
rectangle triangle at A” and “BC = AB + AC ” are equivalent.

0.1.2 Sets
• We call a set any collection of objects, where the objects are called the elements or
members.

• We represent a set S by using one of the following manner,

13
14 PRELIMINARY

1.) indicate all of its elements, we write S = {x1 , y 7 , α, ∆, □, · · · };

2.) describe the condition satisfied for all elements, we write S = {x | x satisfies...}.

• If x is an element in the set S, we say that x belongs to S and we denote this relation
by x ∈ S.

• We say a set A is a subset of a set S or S include A if any element in A is also an

element in S. We denote this relation by A ⊂ S.

• We call the empty set and denote it by ∅. The empty set is a set which contains no
elements at all. In fact, for any set S, ∅ ⊂ S.

• Let S be a set, we call the power set of S and we denote it by ℘(S) the collection of
all subsets in S, i.e., ℘(S) := {A | A ⊂ S}.

• We call the union of two sets A and B the set of all elements which belong to A or
to B. We denote this set by A ∪ B, i.e., A ∪ B := {x | x ∈ A or x ∈ B}.

• We call the intersection of two sets A and B the set of all elements which belong
both to A and to B. We denote this set by A∩B, i.e., A∩B := {x | x ∈ A and x ∈ B}.

• We say two sets A and B are disjoint if A ∩ B = ∅. And we say a family of sets
{Aj | j ∈ J} is disjoint if they are pairwisely disjoint from one to each others.

• Let S be a given set and A ⊂ S be a subsets of S. We call the complement of A in

S the collection of all members in S which do not belong to A. We denote this set by
∁S A, A∁ or S \ A, i.e., S \ A := {x ∈ S | x ∈
/ A}.

• Following the previous point, we say the complement of A relative to an other

subset B ⊂ S the collection of all members in B which do not belong to A. We denote
this set by ∁B A or by B \ A. In fact, B \ A = B \ (A ∩ B).

• We use the following quantification notations,

– “∀x ∈ A...” means the statement is true for all elements in A.

– “∃x ∈ A...” means the statement is true for at least one element in A.

For example, the inclusion relation “A ⊂ S” can be formulated as “∀x ∈ A, x ∈ S”;

the non empty intersection “A ∩ B ̸= ∅” can be formulated as “∃x ∈ A, x ∈ B”.

• The Cartesiani,ii product of two sets A and B is the collection of the couples (a, b)
where a ∈ A and b ∈ B. That is, A × B := {(a, b) | a ∈ A, b ∈ B}.

• Following the previous point, we can define the Cartesian product of n sets A1 , A2 ,
· · · , An where n denotes a strictly positive integer (n = 1, 2, 3, · · · ). An element a ∈
A1 ×A2 ×· · · An is denoted by a = (a1 , a2 , · · · , an ) where aj ∈ Aj for all j ∈ {1, · · · , n}.
We call a such element a ∈ A as a n-tuple and for each j ∈ {1, · · · , n}, aj is called
the j-th component.
i
This adjective appears in many scientific contexts. Especially, the Cartesian coordinate system is
used everywhere. The name Cartesian refers to its inventor, René Descartes.
ii
René Descartes (1596-1650) French philosopher, scientist and mathematician. Symbolic person in the
Scientific Revolution, one of the founder of the rationalism thinkings. Famous by his statement “cogito, ergo
sum”(Je pense donc je suis, in french; I think therefore I am.).
0.1. VOCABULARIES ON SETS AND FUNCTIONS 15

• Following the previous point, in the cases where A1 = A2 = · · · = An = A, we denote

this Cartesian product by An .

We remark that the couples or n-tuples are ordered objects. That means, we say two
ordered objects are equal if each elements at the corresponding places are equal. If same
elements are placed in different positions, they are no longer considered to be equal.

0.1.3 Functions
Let A and B design two sets.

• We call a function, a mapping or an application from A into B any relationship

which links each elements x ∈ A and one particular member y ∈ B. We denote this
A −→ B
function by, f : , where x is called the variable and y is called the value
x 7−→ y
of f at x, we denote it by f (x).

• Let f : A → B, we call the domain of f the subset of A on where f is meaningful.

We call the image or the range of f on the subset U ⊂ A the collection of all values
of f (x) when the variable x describes the subset U . That is,

f (U ) := {f (x) | x ∈ U } ⊂ B.

Symmetrically, for all subset V ⊂ B, we call the pre-image of V by f the set of all
values of the variable x such that f (x) belongs to V . Also, we denote,

f −1 (V ) := {x ∈ A | f (x) ∈ V } ⊂ A.

• Let f : A → B, the graph of f is the following set, {(x, f (x)) ∈ A × B | x ∈ A}.

• Let f : A → B and S ⊂ A. We call the restriction of f on S, and we denote by f |S ,

S −→ B
the function f |S : . In the case where f is a restriction of an other
x 7−→ f (x)
function g, we say that g is an extension of f .

A −→ A
• The identity function is the following function, IdA : .
x 7−→ x

• Let f : A → B and g : B → C. We call the composite function defined by,

A −→ C
g◦f : .
x 7−→ g(f (x))

• Let f : A → B. We say f is

– injective or one-to-one if for all y ∈ B, there exists at most one x ∈ A such

that y = f (x);
– surjective or onto if for all y ∈ B, there exists at least one x ∈ A such that
y = f (x);
– bijective if f is injective and surjective. That is, for all y ∈ B, there exists
exactly one x ∈ A such that y = f (x).
16 PRELIMINARY

An injective function is also called an injection. Similarly, we call a surjection and

a bijection to refer respectively a surjective function and a bijective function.

• If a function f : A → B is bijective, we can define the inverse function, denoted by

f −1 , which links each element y ∈ B to its unique x ∈ A such that y = f (x). The
functions f and f −1 satisfies f ◦ f −1 = IdB and f −1 ◦ f = IdA .

0.1.4 Number Systems

In this section, we revise the number systems N, Z, Q, R and C in their usual approaches.
Remarking that we are about to introduce rigorously the system of real numbers R in
Chapter 1. And the natural numbers N, integers Z as well as rational numbers Q are
already introduced in the course of Basic Mathematics. Here we just recall their notations.

• N represents the set of natural numbers: N = {0, 1, 2, · · · }.

• Z represents the set of integers: Z = {· · · , −2, −1, 0, 1, 2, · · · }.

p
• Q represents the set of rational numbers: for any r ∈ Q, r = where p, q ∈ Z with
q
q ̸= 0.

• R represents the set of√real numbers: R contains all rational numbers and all irra-
tional numbers like 2, π, etc.

• C represents the set of complex numbers: for any z ∈ C, z = a + ib with a, b ∈ R

and i2 = −1. We call respectively the real numbers a and b by the real part of z and
the imaginary part of z. And we denote a = Re(z) as well as b = Im(z).

• Let n, m ∈ Z, n < m, we denote by [[n, m]] to represent the consecutive integers

between n and m, i.e., [[n, m]] = {n, n + 1, · · · , m}.

• We denote by N∗ the set N \ {0}. Analogically, the sets Z∗ , Q∗ , R∗ and C∗ are defined
by the same manner.

• We denote by R+ to represent positive real numbers, R+ := {x ∈ R | x ≥ 0}. Sym-

metrically we denote by R− the set of negative real numbers. Moreover, we use R∗+
to represent the strictly positive real numbers, namely R∗+ := {x ∈ R | x > 0}. And
the set R∗− is defined accordingly.

• For all complex number z ∈pC, we call the magnitude or the modulus of z the
positive real number |z| := Re(z)2 + Im(z)2 . Also, we call the argument or the
phase of z the following real number,
  
Im(z)


 arctan Re(z) if Re(z) ̸= 0,
arg(z) = π
 2 if z = iλ, λ > 0,
− π2

 if z = −iλ, λ > 0.
0.2. COUNTABILITY OF SETS 17

If z = 0, then its argument can be attributed with any real number. By using
Euleriii ’s formula

∀θ ∈ R, eiθ = cos θ + i sin θ,

we can deduce that any complex number z can be expressed into its polar form.

∀z ∈ C, z = reiθ ,
r = |z| ∈ R+ ,
θ = arg(z) + 2πn ∈ R. n ∈ Z.

0.2 Countability of Sets

In this section, we introduce the concept of countability of sets.

Definition 0.2.1. We say a non-empty set S is finite if there exists n ∈ N∗ and a bijection
mapping from [[1, n]] into S. The number n is thus unique and we call it the cardinal of
the finite set S, which is denoted by Card(S). We call a set is an infinite set if it is not
finite.

Remark 0.2.2. • ∅ is a finite set of cardinal 0.

• The unions or intersections of a finite number of finite sets are finite sets. The proof
of this property is left in Exercises.

Proposition 0.2.3 (Well-Ordering Principle). Any non-empty subset of N has a smallest

element.

Proof. This proposition comes directly from the definition of N. In fact, the well-ordering
principle is equivalent to the principle of mathematical induction. See in Exercise 1.A.19 in
Section 1.A.

Proposition 0.2.4. Let S be an infinite subset of N, then there exists a bijection from N
into S.

Proof. Since S is infinite, S is thus non-empty. From the well-ordering principle Proposi-
tion 0.2.3, there exists a smallest element s0 in S. We set then f (0) = s0 .
The other values of f are defined by recurrence. Let n ∈ N∗ , and assume that for all
j ∈ [[0, n − 1]], f (j) is already defined. We consider now the set

Sn := S \ {f (0), f (1), · · · , f (n − 1)}.

The set Sn is infinite, since if not, the set S is finite as a union of two finite sets. We then
apply again the well-ordering principle Proposition 0.2.3 and set that f (n) = min Sn .
iii
Leonhard Euler (1707-1783), German mathematician and physicist. One of the greatest mathematician
in the history. He contributed many fundamental works in number theory, complex analysis, graph theory
and topology. Several notations and constants in modern mathematics such as e, π and i were firstly
introduced by Euler.
18 PRELIMINARY

N −→ S
We now prove the injectivity and surjectivity of the function f : .
n 7−→ f (n)
From the construction, we deduce that for all n ∈ N, f (n + 1) ∈ Sn+1 ⊂ Sn . Since
f (n) = min Sn , we then have f (n) < f (n + 1), which implies that f is a strictly increasing
function over N. Thus, f is injective.
Assuming in the contrary that S\f (N) ̸= ∅. Let s ∈ S\f (N) ⊂ S, we then have f (0) < s
since f (0) is the smallest element in S and s ∈/ f (N). So, either f (m) < s < f (m + 1) for
certain m ∈ N or s > f (m) for all m ∈ N. In the case where f (m) < s < f (m+1) for certain
m ∈ N, we have that s ∈ Sm+1 = S \ {f (0), f (1), · · · , f (m)}. However, s < f (m + 1) =
min Sm+1 which contradicts to the construction. On the other hand, if s > f (m) for all
m ∈ N, we have s > f (m) ≥ m since f is a strictly increasing function on N. This inequality
holds for all m ∈ N, we have in particular that s > s, which is also contradictory.
In conclusion, f is injective and S = f (N). The proof is complete.

Definition 0.2.5. • We say two sets A and B are equipotent (also called alternatively
by similar, equivalent, equinumerous, or equipollent) or have the same cardinality
if there exists a bijection between A and B.

• We say a set S is countable or denumerable if there exists a bijection which maps

from a subset of N into S. In other words, if there exits an injection from S into N or
if S equipotent to a subset NS ⊂ N.

• We say a set S is uncountable if it is not countable.

Example 0.2.6. • Any finite set is countable.

• Any subset in N is countable.

• Any subset in a countable set is countable.

Solution. Let S be a set.

• Assuming that S is finite, then there exists n ∈ N∗ and a bijection f mapping from
[[1, n]] ⊂ N into S. Thus, S is countable.

• Assuming that S ⊂ N. S is either finite or infinite. The case where S is finite is

already studied in the previous point. If S is infinite, then S is an infinite subset of
N. It follows then from Proposition 0.2.4 that there exists a bijection mapping from
N into S. So, S is countable.

• Assuming that S ⊂ A where A is a countable set. Then, there exists an injection f

S −→ N
mapping from A into N. So, the restriction f |S : is also injective,
x 7−→ f (x)
which implies that S is countable.

Proposition 0.2.7. A set S is countable if, and only if either it is finite or there exists a
bijection mapping from N into S.
 0.2. COUNTABILITY OF SETS 19

Proof. If S is finite or if there exists a bijection mapping from N into S, the countability of
S is straightforward.
Reciprocally, we assume at the first time S is countable. By definition, there exists a
bijection f which maps from A ⊂ N into S. If A is a finite set, then there exists a bijection
g mapping from [[1, n]] into A with n ∈ N. So, the function f ◦ g is bijective from [[1, n]] into
S, which leads to that S is finite. In the case where A is an infinite subset in N, it follows
from Proposition 0.2.4, there exists g bijective from N into A and thus f ◦ g is bijective from
N into S.

Proposition 0.2.8. Let S be a non-empty set. Then the following assertions are equiva-
lents.
1.) S is countable.
2.) There exists a surjection from N into S.
3.) There exists a surjection from a countable set into S.
Proof. We will prove the implications 1.) ⇒ 3.) ⇒ 2.) ⇒ 1.).

S −→ S
1.) ⇒ 3.): Assuming S is countable, then the identity mapping IdS : satisfies the
x 7−→ x
condition 3.).
3.) ⇒ 2.): Let A be a countable set, and f be a surjective function mapping from A into S.
Since A is countable, and it follows from Proposition 0.2.4 that there exists a bijective
function g which maps from NA into A where NA = [[1, n]] with A being finite for
certain n ∈ N∗ and NA = N with A being infinite. In the case where NA = [[1, n]],
there exists a surjective function φ mapping from N into NA . We can take for example
N −→ [[1, n]]
φ : where p = qn + r with q ∈ N, r ∈ [[1, n]]. Thus, the function
p 7−→ r
f ◦ g ◦ φ is surjective from N into S. In the case where NA = N, it is sufficient to take
f ◦ g.
2.) ⇒ 1.): Assuming that f is surjectively mapping from N into S. For all s ∈ S, since f
is surjective, the set f −1 ({s}) ⊂ N is not empty. We then consider the function
S −→ N
g: −1
. The number min f −1 ({s}) is well-defined via the well-
s 7−→ min f ({s})
ordering principle for all s ∈ S. Let s1 , s2 ∈ S, such that g(s1 ) = g(s2 ) = n ∈ N. We
have then n ∈ f −1 ({s1 }) ⇒ f (n) = s1 . By the same reason, f (n) = s2 , and thus,
s1 = s2 . This implies that g is injective and in consequences, S is countable.

Proposition 0.2.9. The sets Z and N2 are countable.

Z −→ ( N
Proof. • The function f : 2m − 1 if m > 0 is bijective, then Z is
m 7−→
−2m if m ≤ 0
countable.
20 PRELIMINARY

N2 −→ N
• The function f : 1
is bijective, then N2 is
(n, m) 7−→ 2 (n + m)(n + m + 1) + m
countable.
The justification of the bijectivity of those two functions is left as exercises.

Corollary 0.2.10. The Cartesian product of two countable sets, even more generally a
finite number of countable sets, is countable.

Proof. Let A and B be countable sets. By definition, there exists injective functions
fA and fB mapping from respectively from A and B into N. Since the function φ :
A × B −→ N2
is an injective function from A × B into N2 . Others, since N2
(a, b) 7−→ (fA (a), fB (b))
is countable , there exists another injective function g mapping from N2 into N. Thus, the
function g ◦ φ is injective from A × B into N. So, A × B is countable. The generalized cases
with a product of finite number of countable sets can be deduced from the above result and
a simple mathematical induction.

Corollary 0.2.11. The set Q is countable.

p
Proof. For all r ∈ Q there exists p ∈ Z and q ∈ N∗ such that r = . This condition can be
q
Z × N∗ −→ Q
reformulate as the surjectivity of the function f : p . Since the sets Z and
(p, q) 7−→
q
N∗ are both countable, the set Z × N∗ is thus countable as the Cartesian product of two
countable sets. In consequences, Q is also countable as the image of a countable set by a
surjective function.

Proposition 0.2.12. The union of a countable family of countable sets is countable.

Proof. A countable family of countable sets can be expressed as a sequence of sets (An )n∈N
for which each term An , is a countable set. It follows then from Proposition 0.2.8, for all
n ∈ N, there exits a surjective function fn mapping from N into An . We[ consider now the
2
function F defined as follows. For all (n, m) ∈ N , F (n, m) = fn (m) ∈ An . Then, the
[ n∈N
function F is surjective from N2 into An . Applying again Proposition 0.2.8 and the fact
[ n∈N
that N2 is countable, the set An is countable.
n∈N

Corollary 0.2.13. Any finite union of countable sets is countable. And any intersection
of countable sets is countable.

Proof. Left in exercise.

Theorem 0.2.14. The set R is uncountable.

0.3. ALGEBRAIC STRUCTURES 21

Proof. Since the function x 7→ tan(π(x − 12 )) is bijectively mapping from (0, 1) into R. The
sets (0, 1) and R are thus equipotent. It is sufficient now to prove that (0, 1) is uncountable.
We assume in the contrary that (0, 1) is countable, and we will represent all real numbers
in (0, 1) in its decimal development. Then, we can write all the real numbers in (0, 1) as
follows.

r0 = 0.a0,0 a0,1 · · · a0,k · · · ;

r1 = 0.a1,0 a1,1 · · · a1,k · · · ;
..
.
rn = [Link],0 an,1 · · · an,k · · · ;
..
.

where an,k ∈ [[0, 9]] for all (n, k) ∈ N2 .

We then construct a new real number b = 0.b0 b1 · · · by setting bn = an,n + 1 for all n ∈ N
(in the case where an,n = 9, we set bn = 0). Then the number b ∈ (0, 1) can not be one of
the listed numbers (rn )n∈N above. Thus, the contradiction.
In conclusion, (0, 1) is not a countable set and therefore R is not either.

Remark 0.2.15. In a more general context, we can express the comparison of the cardinality
between sets in the following way. Let A and B be sets.
• If there exists an injection f mapping from A into B, then the function f forms a
bijection from A into its image f (A) ⊂ B. In the consequences, A and f (A) are
equipotent, which means A and f (A) have the “same quantity” of elements. Thus,
the range B contains “more elements” then A. We denote this relation by A ⪯ B.

• Following the previous point, if there exists only injections but no bijection between
A and B, we denote this relation by A ≺ B.

• If there exists a surjection mapping from A into B, we denote this relation by A ⪰ B.

This relation indicates that A contains “more elements” then B.

• Following the previous point, if there exists only surjections but no bijection between
A and B, we denote this relation by A ≻ B.

• Typical example from Theorem 0.2.14 sates that R ≻ N.

Remark 0.2.16. We can prove (in Exercises) that any infinite set contains at least one
countable subset in it. That is to say, the countable sets are “smallest” infinite sets. We
present this level of infinity by the symbol ℵ0 , and denote Card(N) = ℵ0 . We can show in
Exercise 0.4.12 that R and R2 have the same cardinality as well as the cardinality of them
equals to 2ℵ0 . Those who are interested in this direction can go further to explore different
levels of infinity like ℵ1 , ℵ2 and so on.

0.3 Algebraic Structures

In this section, we introduce basic algebraic structures, such as group, ring and field. Let
K be a non-empty set and we preserve this notation in the rest of this chapter.
22 PRELIMINARY

Definition 0.3.1. We call a binary operation or simply an operation on K any function

which maps from K × K into K.

Definition 0.3.2. Let ⋆ be an operation on K, we say the set K together with ⋆ is a group
if it satisfies the following axioms, and we denote it by (K, ⋆).

G-1.) For all x, y, z ∈ K, (x ⋆ y) ⋆ z = x ⋆ (y ⋆ z). Associativity

G-2.) There exists e ∈ K such that for all x ∈ K, x ⋆ e = e ⋆ x = x. Identity element

G-3.) For all x ∈ K there exists an element y ∈ K such that x ⋆ y = y ⋆ x = e. Inverse

element

Proposition 0.3.3. Let (K, ⋆) be a group. Then, the following assertions hold.

• The element e satisfying the second axiom of the definition of groups is unique. This
element is called the identity element of the group (K, ⋆).

• For all x ∈ K, the element y satisfying the third axiom of the definition of groups is
unique. This element is called the symmetric (or inverse) of x and it is usually
denoted by x−1 .

Proof. • Let e, e′ be identity elements in (K, ⋆), we have then e = e ⋆ e′ = e′ . So the

identity element is unique.

• Let x ∈ K and y, z ∈ K such that x ⋆ y = y ⋆ x = e as well as x ⋆ z = z ⋆ x = e, we

have then

y = y ⋆ e = y ⋆ (x ⋆ z) = (y ⋆ x) ⋆ z = e ⋆ z = z.

So the inverse element is unique.

Definition 0.3.4. We say that the operation ⋆ on the group (K, ⋆) is commutative if for
all x, y ∈ K, x ⋆ y = y ⋆ x. In this case, we say the group (K, ⋆) is a commutative group
or an abelianiv group.

Example 0.3.5. Let n ∈ N, n ≥ 2.

• (Z/nZ, +) is a group.

• The set of invertible matrices of order n ∈ N, equipped with the matrix multiplication
is a non commutative group (GL(n), ·).

Definition 0.3.6. We say a set K together with two operations + and · is a ring if it
satisfies the following axioms, and we denote it by (K, +, ·).
iv
Niels Henrik Abel (1802 - 1829), Norwegian mathematician. Famous for various pioneering works. He
was the first person who proved the impossibility to solve the general quinic equation in radicals. The Abel
Prize nowadays is considered as one of the most prestigious prize for mathematicians.
0.3. ALGEBRAIC STRUCTURES 23

R-1.) (K, +) is a commutative group.

R-2.) The operation · is associative and it possesses an identity element in K.

R-3.) For all x, y, z ∈ K, (x + y) · z = (x · z) + (y · z) as well as z · (x + y) = (z · x) + (z · y).

Distributivity

We denote this ring by (K, +, ·).

Example 0.3.7. (Z, +, ·) is a ring.

Remark 0.3.8. Let (K, +, ·) be a ring.

• We will call, in what follows, these two operations as addition and (internal) mul-
tiplication.

• We will denote by 0K the additive identity element (plausibly denoted by 0 if no

confusion will occurs) and by −x the additive inverse of x ∈ K. This element is also
called as the opposite of x.

• We define the subtraction operation “−” by ∀x, y ∈ K, x − y := x + (−y).

• We will denote by 1K the multiplicative identity element (plausibly denoted by 1 if

no confusion will occurs).

• We call an element x ∈ K is invertible for the multiplication · if there exists y ∈ K

such that x · y = y · x = 1K . We will denote by K∗ the set of invertible elements in K.

We can see that an element in a ring does not necessarily possesses an inverse respect
to the multiplication. While this final property is satisfied for every x ∈ K \ {0K }, the ring
becomes a field. In other words, a field is a ring such that K∗ = K \ {0K }.

Definition 0.3.9. We say a set K together with two operations + and · is a field if

K-1.) (K, +) is a commutative group with the identity element 0K ;

K-2.) Any non-zero element in K is invertible for multiplication · and (K∗ , ·) forms a com-
mutative group with the identity element 1K ;

K-3.) the multiplication · is distributive respect to the addition +.

We denote this field by (K, +, ·).

Remark 0.3.10. Concerning on notations, the multiplication symbol · is often omitted. For
example, we write xy at the place of x · y. Moreover, we use the powers indexes to represent
the multiplications by itself. For any x ∈ K, x ̸= 0K , we write by convention x0 = 1K and
for any n ∈ N, xn+1 = x · xn .
On the other hand, for all x ∈ K∗ , we denote by x−1 the inverse element for the
multiplication. Also, we will use the notations of fractions to represent the numbers in K.
p
We shall denote = q −1 · p for all p ∈ K and all q ∈ K∗ .
q
24 PRELIMINARY

In what follows, we consider an arbitrary field and denote it by (K, +, ·).

Proposition 0.3.11. • For any x ∈ K, x · 0K = 0K .

• For any x ∈ K, (−1K ) · x = −x.

• If 0K = 1K , then (K, +, ·) contains only one element. In this case, the field (K, +, ·)
is said to be trivial.

Proof. • Let x ∈ K. We have x · 0K = x · (0K + 0K ) = x · 0K + x · 0K . We then add, at

each sides of the equality, the opposite of x · 0K . Thus,

0K = x · 0K + (−x · 0K ) = x · 0K + x · 0K + (−x · 0K ) = x · 0K .

• Let x ∈ K,

(−1K ) · x + x = (−1K ) · x + 1K · x = (−1K + 1K ) · x = 0K · x = x · 0K = 0K .

This shows that (−1K ) · x is the opposite of x, then (−1K ) · x = −x.

• We assume that 0K = 1K in K. Let x ∈ K, we have x = x · 1K = x · 0K = 0K . So,

K = {0K }.

In what follows, we always assume a field is non trivial.

Remark 0.3.12. In the rest of this lecture note, we always assume a field is non trivial unless
we precisely mention that.

Example 0.3.13. Let K = {a, b, c} and the laws +, · are defined by the tables,

+ a b c · a b c
a a b c a a a a
b b c a b a b c
c c a b c a c b

Show that (K, +, ·) is a field.

Solution. Left in exercise

Example 0.3.14. We consider K := K × K and define

(a, b) ⊕ (c, d) := (a + c, b + d), (a, b) ⊙ (c, d) := (ac − bd, ad + bc).

Show that (K, ⊕, ⊙) is a field.

Solution. Left in exercise

Example 0.3.15. Let x, y ∈ K, then (x + y)(x − y) = x2 − y 2 .

0.3. ALGEBRAIC STRUCTURES 25

Solution.

(x + y)(x − y)
= (x + y) · (x + (−y)), (definition of subtraction)
= (x + y) · x + (x + y) · (−y), (distributivity)
= x · x + y · x + x · (−y) + y · (−y), (distributivity)
= x · x + y · x + x · (−1) · y + y · (−1) · y, (identity (−1) · y = −y)
= x · x + x · y + (−1) · x · y + (−1) · y · y, (commutativity)
= x · x + x · y + (−x · y) − y · y, (identity (−1) · y = −y)
= x2 − y 2 .

Definition 0.3.16. Let (K, +K , ·K ) be a field and (E, +) be a commutative group. We say
a function · : K × E → E is an external multiplication or a scalar multiplication if
the following relations hold for all λ, ν ∈ K and for all u, v ∈ E.

V-1.) (λ +K µ) · u = λ · u + µ · u;

V-2.) λ · (u + v) = λ · u + λ · v;

V-3.) λ · (µ · u) = (λ ·K µ) · u;

V-4.) 1K · u = u.

We call the set E together with the addition + and the external multiplication · as a K-
vector space and we denote it by (E, +, ·). Also, a vector is defined as any element in a
K-vector space (E, +, ·).

Definition 0.3.17. Let K be a field. We call an algebra over K any structure of type
(E, +, ·, ×) satisfying the following axioms.

A-1.) (E, +, ·) is a K-vector space.

A-2.) × is an operation on E.

A-3.) For all x, y, z ∈ E, (x + y) × z = x × z + y × z. Right distributivity

A-4.) For all x, y, z ∈ E, z × (x + y) = z × x + z × y. Left distributivity

A-5.) For all x, y ∈ E, α, β ∈ K, α · β · (x × y) = (α · x) × (β · y). Compatibility with

scalar multiplication

Example 0.3.18. The set of square matrices in real or complex coefficients Mn (K) (K = R
or C) equipped with matrix addition, scalar multiplication and matrix multiplication is an
algebra.
26 PRELIMINARY

0.4 Exercises
Exercise 0.4.1. Let (x, y) ∈ R2 . Write down the negations of the following propositions.

1.) “1 ≤ x < y”.

2.) “xy = 0”. (with the propositions with predicates respectively in x and in y)

3.) “x2 = 1 ⇒ x = 1”.

Exercise 0.4.2. Write down the negations of the following propositions.

1.) “∀x ∈ E, ∀x′ ∈ E, x ̸= x′ ⇒ f (x) ̸= f (x′ )”.

2.) “∀a ∈ Z, ∀b ∈ N∗ , ∃q ∈ Z, ∃r ∈ Z such that a = bq + r with 0 ≤ r < b”.

Exercise 0.4.3. Let E, F, G be three sets, f be a function mapping from E into F and g
be a function from F into G. Show that,

1.) if g ◦ f is injective, then f is injective;

2.) if g ◦ f is surjective, then g is surjective;

3.) if g ◦ f is injective and f is surjective, then g is injective;

4.) if g ◦ f is surjective and g is injective, then f is surjective.

Nested Increasing Sequences of Finite Sets

In this subsection, we want to establish a characterization of countable sets using the se-
quences of finite sets. Precisely speaking, we want to prove the following theorem.

Theorem 0.4.4. A set S is countable if, and only if there exists a sequence (Sn )n∈N of sets
which satisfies

• for all n ∈ N, Sn is a finite set;

• for all n ∈ N, Sn ⊂ Sn+1 ;

[
• S= Sn .
n∈N

At the same time, we define any sequence (Sn )n∈N which satisfies the three properties
in Theorem 0.4.4 as a nested increasing sequence of finite sets in S. For instance, the
sequence ([[0, n]])n∈N is a nested increasing sequence of finite sets in N.

Exercise 0.4.5. 1.) Give an increasing sequence of finite sets to Z.

2.) Give an increasing sequence of finite sets to N2 .

0.4. EXERCISES 27

Exercise 0.4.6. Let S be a countable set. Show that there exists an increasing sequence
of finite sets to S.
Hint: We can distinguish the cases where S is finite or infinite.

Exercise 0.4.7.[Let (Pn )n∈N be a sequence of sets such that for all n ∈ N, Pn is a finite
set. Show that Pn is countable.
n∈N

Exercise 0.4.8. Prove Theorem 0.4.4.

Exercise 0.4.9. We call an algebraic number any root of a polynomial with integer
coefficients. And we call a transcendent number any (complex) number which is not
algebraic. Show that the set of algebraic numbers is countable and there exists transcendent
numbers.

Exercise 0.4.10. Let A0 , A1 , B denote three arbitrary sets. We suppose that A0 and A1
are equipotent.

1.) Assume in addition that A1 and B are equipotent. Show that A0 and B are equipotent.

2.) We suppose in this question that A0 is finite and B is infinite. Also, we assume that
A0 ∩ B = ∅.

(a) Show that there exists a sequence of subsets B0 , B1 , · · · , Bn , · · · satisfying

• for all n ∈ N, Bn ⊂ B and Card(Bn ) = Card(A0 );
• for all n, m ∈ N, Bn ∩ Bm = ∅ once n ̸= m.
!
[ [
(b) Show that the Bn and A0 ∪ Bn are equipotent. Then deduce that
n∈N n∈N
A0 ∪ B and B are equipotent.

3.) Assume now A0 and A1 are countably infinite as well as that A0 ⊂ B and A1 ∩ B = ∅.
Show that B and A1 ∪ B are equipotent.

Exercise 0.4.11. In this Exercise, (a, b), [a, b] ⊂ R denotes respectively open and closed
intervals on real numbers.
π π
1.) Show that (− , ) and R are equipotent.
2 2
2.) Let a, b, α, β ∈ R such that a < b and α < β. Show that (a, b) and (α, β) are
equipotent.

3.) Show that [0, 1) and (0, 1) are equipotent.

4.) Show that [0, 1) and R are equipotent.

28 PRELIMINARY

Exercise 0.4.12. In this Exercise, we denote by {0, 1}N to represent the set of sequences
composed by 0 and 1. That is,

{0, 1}N = {(xn )n∈N | ∀n ∈ N, xn = 0 or 1} .

At the same time, we introduce the subsets A0 and A1 by,

n o
A0 := (xn )n∈N ∈ {0, 1}N ∃N ∈ N s.t. ∀n ≥ N, xn = 0 ;
n o
A1 := (xn )n∈N ∈ {0, 1}N ∃N ∈ N s.t. ∀n ≥ N, xn = 1 .

1.) Show that {0, 1}N is uncountable.

2.) Show that A0 and A1 are countable.

3.) Show that [0, 1) and {0, 1}N \ A1 are equipotent.

4.) Show that R and {0, 1}N are equipotent.

5.) Show that {0, 1}N × {0, 1}N and {0, 1}N are equipotent. Then deduce that R2 and R
are equipotent.
 Chapter 1

System of Real Numbers

In this chapter, we introduce the real number system. The main objective for studies is to
learn principal properties and how they can be derived from a minimum numbers of axioms.
In this entire chapter, we conserve all the notations which are previously defined.

1.1 Ordered Fields

Let S be a set.

Definition 1.1.1. We call a binary relation R on the set S any function which maps
from S × S into {True, False}. We denote by xRy the value of this function instead of
R(x, y) for all x, y ∈ S.

Definition 1.1.2. Let R be a binary relation on S, we say R be an order or precisely a

partial order on S if it satisfies the following axioms.

1.) For all x ∈ S, xRx. (Reflexivity)

2.) For all x, y ∈ S, if xRy and yRx then x = y. (Anti-symmetry)

3.) For all x, y, z ∈ S, if xRy and yRz then xRz. (Transitivity)

If, in addition, R satisfies that for all x, y ∈ S either xRy or yRx, we say the order R is a
total order over the set S.

Example 1.1.3. The relation < is not an order on N whereas ≥ is an order on N.

Definition 1.1.4. We say a field (K, +, ·) together with a total order R is a totally ordered
field or simply an ordered field if R is compatible with the field structure. That is, it
satisfies the following axioms.

1.) If xRy then (x + z)R(y + z) for all z ∈ K.

2.) If 0Rx and 0Ry then 0R(x · y).

The ordered field is denoted by (K, +, ·, R).

29
30 SYSTEM OF REAL NUMBERS

Example 1.1.5. ≥ is not compatible with the field (Q, +, ·).

From now on, (K, +, ·) denotes a field and ≤ denotes a total order on K such that
(K, +, ·, ≤) forms an ordered field.
Definition 1.1.6. Consider the ordered field (K, +, ·, ≤). we define the binary relations
≥, < and > as follows. Let x, y ∈ K,
1.) x ≥ y if y ≤ x.
2.) x < y if x ≤ y and x ̸= y.
3.) x > y if y < x.
Proposition 1.1.7 (Law of Trichotomy). Let x, y be elements in an ordered field (K, +, ·, ≤
), then exactly one of the relations x < y, x = y or x > y holds.
Proof. Let x, y ∈ K. Since the relation ≤ is total in K, we have either x ≤ y or y ≤ x.
• Case x ≤ y. From the definition of <, we have either x < y or x = y.
• Case y ≤ x. From the definition of <, we have either y < x or x = y. Others,
y < x ⇔ x > y from definition.

Given an ordered field (K, +, ·, ≤), we can “construct” the systems of natural numbers
N, of integers Z as well as the rational numbers Q inside the field K.
Definition 1.1.8. Let 0 and 1 be respectively the identity elements to the operations “+”
and “·” in the field (K, +, ·).
• The set of natural numbers N can be constructed by recurrence under the following
two rules, that 0 ∈ N and that if n ∈ N, then n + 1 ∈ N.
• The integers Z can be defined as Z = N ∪ {−n | n ∈ N}.
 
p ∗
• The rational numbers Q can be defined as Q = p ∈ Z, q ∈ Z . Here, the
q
p p p1 p2
fraction is defined by = q −1 p. Also, two fractions and are equal if, and
q q q1 q2
only if p1 q2 = p2 q1 .

Definition 1.1.9. The magnitude or the absolute value of x ∈ K, denoted by |x|, is

defined as
(
x if x ≥ 0,
|x| =
−x if x < 0.

Proposition 1.1.10 (Triangular Inequality). In an ordered field (K, +, ·, ≤), for all x, y ∈
K, it holds thati
||x| − |y|| ≤ |x ± y| ≤ |x| + |y|.
Proof. Left in Exercise 1.6.3.

Definition 1.1.11. An ordered field (K, +, ·, ≤) is said to be Archimedeanii if for all

i
The symbol ± stands for + or −.
ii
Archimedes of Syracuse (287 - 212 B.C.), Ancient Greek scientist.
1.2. SUBSETS IN ORDERED FIELDS 31

x ∈ K, there exists n ∈ Z such that x < n.

Example 1.1.12. The ordered field (Q, +, ·, ≤) is Archimedean.

Solution. Let x ∈ Q. If x < 0, then we choose n = 0. If x ≥ 0, there exists (p, q) ∈ N × N∗

p
such that x = ≤ p. We then choose n = p + 1. In both cases, we always have x < n and
q
the claim follows.

1.2 Subsets in Ordered Fields

Let (K, +, ·, ≤) be an ordered field.

Definition 1.2.1. We say the subset I of K to be an interval on K if I fits one of the

following cases.

1.) I = {x ∈ K | a < x < b} with a, b ∈ K such that a < b. In this case, I is denoted by
(a, b) and is called as an open interval.

2.) I = {x ∈ K | a < x} with a ∈ K. In this case, I is denoted by (a, +∞) and is called
as an unbounded open interval.

3.) I = {x ∈ K | x < b} with b ∈ K. In this case, I is denoted by (−∞, b) and is called as

an unbounded open interval.

4.) I = {x ∈ K | a ≤ x ≤ b} with a, b ∈ K such that a < b. In this case, I is denoted by

[a, b] and is called as a closed interval or a segment.

5.) I = {x ∈ K | a ≤ x} with a ∈ K. In this case, I is denoted by [a, +∞) and is called

as an unbounded closed interval.

6.) I = {x ∈ K | x ≤ b} with b ∈ K. In this case, I is denoted by (−∞, b] and is called as

an unbounded closed interval.

7.) I = {x ∈ K | a ≤ x < b} with a, b ∈ K such that a < b. In this case, I is denoted by

[a, b) and is called as a closed at left and open at right interval.

8.) I = {x ∈ K | a < x ≤ b} with a, b ∈ K such that a < b. In this case, I is denoted by

(a, b] and is called as an open at left and closed at right interval.

We define as well as the length of an interval the number |b − a| for bounded intervals, i.e.
cases 1.), 4.), 7.), 8.). For the unbounded intervals, their lengths are settled to be +∞.

Remark 1.2.2. We have also the following degenerate cases in the definitions of intervals.
In the most of cases, they are not considered as intervals.

• I = {a} with a ∈ K. In this case, I is called a singleton.

• I = K. In this case, I is sometimes denoted by (−∞, +∞).

Definition 1.2.3. Let S ⊂ K and a ∈ S.

32 SYSTEM OF REAL NUMBERS

• We say a is the largest element in S if for all x ∈ S, x ≤ a. When the largest element
exists, we call it the maximum and denote it by max(S).

• We say a is the smallest element in S if for all x ∈ S, a ≤ x. When the smallest

element exists, we call it the minimum and denote it by min(S).

Proposition 1.2.4. In the case of existence, the maximum and the minimum are unique.

Proof. Let S ⊂ K, we assume that S possess the maximums a, a′ ∈ S. By definition of

maximum, we have thus a ≤ a′ and a′ ≤ a. From the anti-symmetry of ≤, a = a′ . The case
of minimum follows exactly the same reasoning.

Definition 1.2.5. Let S ⊂ K and a ∈ K.

• We say a be an upper bound of S if for all x ∈ S, x ≤ a.

• We say a be a lower bound of S if for all x ∈ S, a ≤ x.

• If there exists an upper bound of S, we say that S is bounded from above.

• If there exists a lower bound of S, we say that S is bounded from below.

• We say the subset S is bounded if it is bounded from above and from below. This
condition can be also reformulated as follows,
∃M ∈ K, such that ∀x ∈ S, |x| ≤ M .

Definition 1.2.6. Let S be a subset of K.

• We call the supremum, or the least upper bound of S the smallest upper bound of
S, if it exists, and we denote it by sup(S).

• In the case where S is not bounded from above, we set sup(S) = +∞.

• We call the infimum, or the greatest lower bound of S the largest lower bound of S,
if it exists, and we denote it by inf(S).

• In the case where S is not bounded from below, we set inf(S) = −∞.

• We set by conventions, sup(∅) = −∞ and inf(∅) = +∞.

Remark 1.2.7. • In the cases of existence, it follows from Proposition 1.2.4 that the
supremum and infimum of a set S are respectively unique.

• In the cases where the maximum/minimum exists, we have immediately that sup(S) =
max(S) or inf(S) = min(S).

Example 1.2.8. In the ordered field (R, +, ·, ≤), show that

• sup(0, 3) = 3 and inf(0, 3) = 0.

• sup(N) does not exist, inf(N) = 0.

1.2. SUBSETS IN ORDERED FIELDS 33

• Let A = 2−k k ∈ N , then sup(A) = 1 and inf(A) = 0.



√ √
• Let B = x ∈ Q x2 < 2 , then sup(B) = 2 and inf(B) = − 2.


Answer the same questions in the ordered field (Q, +, ·, ≤).

Solution. • From the definition of the interval (0, 3), 3 is an upper bound of (0, 3). We
then prove that 3 is the least upper bound. Let x ∈ R, x < 3. If x < 1, we have
x < 1 ∈ (0, 3), and thus x is not an upper bound of (0, 3). If 1 ≤ x < 3, we have
x+3
x< < 3. This inequality shows that x is not an upper bound of (0, 3). So,
2
3 = sup(0, 3). The case of infimum is left in Exercises.

• Since R satisfies the Archimedean property, for all x ∈ R, there exists n ∈ Z such
that x < n. In consequence, x < n ≤ max{0, n} ∈ N, which implies x is not an upper
bound of N. Thus, sup(N) does not exist. The case of infimum is left in Exercises.

• For all k ∈ N, 2−k ≥ 0. This shows that 0 is a lower bound of A. Consider now ε > 0.
If ε > 1, we have ε > 1 = 2−0 ∈ A. So, ε can not be a lower bound of A. If 0 < ε ≤ 1,
we choose then k ∈ N satisfying k ≥ − log2 ε + 1. This choice is licit since R satisfies
the Archimedean property. We have thus 2−k < ε, which proves that ε can not be a
lower bound. In conclusion, 0 = inf(A). The case of sup(A) is left in Exercises.
√
• It follows
√ from Exercise 1.6.2 that for all x ∈ R +
√ , x 2 < 2 ⇔ x < 2. This implies
that 2 is an upper bound of B. To show that 2 = sup(B), we need an denseness
argument of Q in R. This property states that for all x ∈ R, we can always find a
number r ∈ Q such that r is as close to x as we want. This property will be introduced
in Proposition 1.2.11.
√
Since
√ Q is dense
√ in R, for all ε > 0, there exists always r ∈ Q such that
√ 2−ε < r <
2. Then, 2 − ε ∈ B is not√an upper bound. In other words, 2 is the smallest
upper bound, hence sup(B) = 2.

Proposition 1.2.9 (Characterization of the supremum/infimum). Let S ⊂ K and a ∈ K.

• a = sup(S) if, and only if

∀x ∈ S, x ≤ a and ∀ε > 0, ∃x ∈ S s.t. a − ε < x.

• a = inf(S) if, and only if

∀x ∈ S, x ≥ a and ∀ε > 0, ∃x ∈ S s.t. a + ε > x.

Proof. We will prove this proposition by double implications for the case of the supremum
and the case of the infimum follows exactly the same process.

⇒: We assume that a = sup(S). Since a is an upper bound of S, we have then ∀x ∈ S,

x ≤ a. Moreover, a is the smallest upper bound of S. That means, for all ε > 0,
a − ε < a is not an upper bound of S. There exists therefore an element x ∈ S such
that a − ε < x.
34 SYSTEM OF REAL NUMBERS

⇐: Let a ∈ K such that ∀x ∈ S, x ≤ a and ∀ε > 0, ∃x ∈ S such that a − ε < x. The first
condition shows that a is an upper bound of S and the second condition shows that
for any number y = a − ε which is strictly smaller then a can not be an upper bound.
In other words, a is the smallest upper bound. So, a = sup(S).

The following two propositions on the Archimedean property are useful in the future
applications.
Proposition 1.2.10 (Integer parts). Let (K, +, ·, ≤) be an Archimedean ordered field.
Then, for any x ∈ K, there exists a unique integer n ∈ Z such that n ≤ x < n + 1.
The integer n ∈ Z is called as the integer part of x and it is denoted by [x].
Proof. We begin with the uniqueness of the integer part. Let n, n′ ∈ Z such that n ≤ x <
n + 1 as well as n′ ≤ x < n′ + 1. We have immediately that n ≤ x < n′ + 1, which implies
that n ≤ n′ . Symmetrically, we have also n′ ≤ n. So, n = n′ by the anti-symmetry of the
order ≤.
Let x ∈ K, it follows from the Archimedean property that there exists an integer N ∈ Z
such that |x| < N . We consider the set S := {m ∈ N | N − m ≤ x}. Since |x| < N , it
follows that −N < −|x| ≤ x. And therefore, 2N ∈ S which also implies that S is non-
empty. Now, S is an non-empty subset in N. It follows from the well-ordering principle
that there exists Nm ∈ N such that Nm = min(S). We then set n = N − Nm ∈ Z. We
also remark that Nm ̸= 0, since if 0 ∈ S, then it holds N ≤ x ≤ |x| < N which leads to a
contradiction. So, we can deduce from the facts Nm ∈ S and Nm − 1 ∈ / S that n ≤ x as
well as n + 1 > x. Thus, the claim follows.

Proposition 1.2.11 (Denseness of Q in K). Let (K, +, ·, ≤) be an Archimedean ordered

field. Then, for any x, y ∈ K with x < y, there exists r ∈ Q such that x < r < y.
1
Proof. We begin with considering the number ∈ K∗+ . It follows from the Archimedean
y−x
1 1
property that there exists q ∈ N∗ such that < q, equivalently speaking, y > x + .
y−x q
Next, it follows from Proposition 1.2.10 that there exists p = [q · x] + 1 ∈ Z such that
p−1 p 1 p−1+1 p
p − 1 ≤ q · x < p, or that ≤ x < . Also, it holds that y > x + ≥ = .
q q q q q
p
Thus, the claim follows by setting r = ∈ Q.
q

Definition 1.2.12 (least upper bound property). We say that the ordered field (K, +, ·, ≤)
satisfies the least upper bound property if for any non empty subset S which is bounded
from above, S possesses the supremum in K.

Proposition 1.2.13. The least upper bound property implies Archimedean property.
Proof. Let (K, +, ·, ≤) be an ordered field satisfying the least upper bound property. Con-
sider x ∈ K.
• If x < 0, then we choose n = 0 ∈ Z.
1.3. SEQUENCES IN ORDERED FIELDS 35

• If x ≥ 0, we introduce the set S := {m ∈ Z | m ≤ x}. We have immediately 0 ∈ S.

Thus, S is non-empty and S is bounded from above by x. Applying the least upper
bound property, there exists t = sup(S) ∈ K. Since t is the smallest upper bound, t−1
is not an upper bound, which implies that there exists mt ∈ S such that mt > t − 1.
This condition implies also mt + 1 > t ⇒ mt + 1 ∈/ S ⇒ mt + 1 > x. We then choose
n = mt + 1 ∈ Z.

Hence, it always holds that x < n with the above choices of n.

1.3 Sequences in Ordered Fields

1.3.1 Definitions
Definition 1.3.1. We call a sequence of elements in S any function f which maps from
N into S. The value of f at n ∈ N is called the n-th term of the sequence. We use the
notations (f (n))n∈N or (fn )n∈N with fn = f (n) ∈ S to represent the sequences. In addition,
we denote by S N to represent the set of sequences of elements in S.

Definition 1.3.2. Let (xn )n∈N be a sequence of elements in K.

• We say the sequence (xn )n∈N is a constant sequence if there exists l ∈ K such that
xn = l for all n ∈ N.

• We say the sequence (xn )n∈N is bounded from above if there exists M ∈ K such
that for all n ∈ N, xn ≤ M . Also, any number M satisfying this condition is called
as an upper bound of the sequence (xn )n∈N .

• We say the sequence (xn )n∈N is bounded from below if there exists m ∈ K such
that for all n ∈ N, xn ≥ m. Also, any number m satisfying this condition is called as
a lower bound of the sequence (xn )n∈N .

• We say the sequence (xn )n∈N is bounded if it is bounded from above and from below.
That is, ∃M ∈ K∗+ such that ∀n ∈ N, |xn | ≤ M .

Remark 1.3.3. In the case of existence, we can also define the supremum or the infimum of
the sequence (xn )n∈N .

• If the set {xn | n ∈ N} possesses the supremum, then this supremum is called as the
supremum of (xn )n∈N and we denote it by sup xn .
n∈N

• If the set {xn | n ∈ N} possesses the infimum, then this infimum is called as the infi-
mum of (xn )n∈N and we denote it by inf xn .
n∈N

Proposition 1.3.4. The set of bounded sequences is stable by linear combinations and by
term-to-term multiplications. That is, if (xn )n∈N and (yn )n∈N are bounded sequences, then
for all λ ∈ K the sequences (xn + λyn )n∈N , (xn yn )n∈N are also bounded sequences.
36 SYSTEM OF REAL NUMBERS

Proof. Let (xn )n∈N and (yn )n∈N be bounded sequences. There exists Mx , My ∈ K∗+ such
that for all n ∈ N, |xn | ≤ Mx and |yn | ≤ My . Then, it follows from the triangular inequality
that for all λ ∈ K and n ∈ N, |xn + λyn | ≤ |xn | + |λ||yn | ≤ Mx + |λ|My ∈ K∗+ as well as
|xn yn | ≤ |xn ||yn | ≤ Mx My . Thus, (xn +λyn )n∈N and (xn yn )n∈N are bounded sequences.

Remark 1.3.5. This properties shows that the set of bounded sequences in K possesses a
structure of vector space. The complete justification is left as an exercise. We denote this
vector space by ℓ∞ (K).

Definition 1.3.6. Let (xn )n∈N be a sequence of elements in K.

• We say the sequence (xn )n∈N is increasing if for all n ∈ N, xn+1 ≥ xn .

• We say the sequence (xn )n∈N is decreasing if for all n ∈ N, xn+1 ≤ xn .

• We say the sequence (xn )n∈N is monotone if (xn )n∈N is increasing or decreasing.

• We say the sequence (xn )n∈N is strictly increasing, strictly decreasing or strictly
monotone if the corresponding inequalities become strict.

Remark 1.3.7. For an increasing sequence (xn )n∈N , we can deduce, by using a simple in-
duction, that for all n, m ∈ N, if n ≤ m then xn ≤ xm . The corresponding formulations for
other type of monotone sequences can be obtained similarly, which we left them as exercises.

Definition 1.3.8. Let (xn )n∈N be a sequence of elements in K.

• We say the sequence (xn )n∈N converges or tends to l ∈ K if for all ε ∈ K, ε > 0
there exists N ∈ N such that for all n ∈ N, n ≥ N we have |xn − l| ≤ ε. We also call
l as a limit of the sequence (xn )n∈N . Symbolically, we write

∀ε > 0, ∃N ∈ N s.t. ∀n ∈ N, n ≥ N ⇒ |xn − l| ≤ ε.

• We denote by xn −→ l or lim xn = l to express the convergence of (xn )n∈N to l.

n→+∞ n→+∞

• We say the sequence (xn )n∈N is convergent if there exits l ∈ K such that xn −→ l.
n→+∞

• We say the sequence (xn )n∈N diverges or it is divergent if (xn )n∈N is not convergent.

Remark 1.3.9. The natural number N in the above definition depends in general (sometimes
severely) on ε. That means, if we want that |xn − l| ≤ ε′ holds with ε′ < ε, we need a larger
N ′ > N . On the other hand, the inequality |xn − l| ≤ ε can be replaced by the strict one
|xn − l| < ε. The justification is left in Exercises.

Proposition 1.3.10. Let (xn )n∈N be a sequence of elements in K and let l ∈ K. Then,
lim xn = l if, and only if lim |xn − l| = 0.
n→+∞ n→+∞

Proof. It is a straightforward result from the definition of convergence by remarking that

|xn − l| = ||xn − l| − 0|.
1.3. SEQUENCES IN ORDERED FIELDS 37

Theorem 1.3.11 (Uniqueness of the limit). Let (xn )n∈N ∈ KN be a convergent sequence.
Then, there exists a unique l ∈ K such that lim xn = l.
n→+∞

Proof. We assume that l1 , l2 ∈ K be two limits of the convergent sequence (xn )n∈N . Let
ε > 0, by the definition of convergence, there exists N1 , N2 ∈ N such that for all n ∈ N,
ε ε
n ≥ N1 ⇒ |xn − l1 | ≤ as well as n ≥ N2 ⇒ |xn − l2 | ≤ . We then set N = max{N1 , N2 },
2 2
and it holds, for all n ∈ N, n ≥ N ,
ε ε
|l1 − l2 | ≤ |l1 − xn | + |xn − l2 | ≤ + ≤ ε.
2 2
The latter inequality shows that |l1 − l2 | is smaller then any number ε > 0. This equivalent
to say that |l1 −l2 | = 0. Thus, the sequence (xn )n∈N converges to the unique limit l ∈ K.

1.3.2 Properties on Convergent Sequences

Proposition 1.3.12. Let (xn )n∈N be a sequence of elements in K and l ∈ K. Then, it holds

1.) Any convergent sequence is bounded.

2.) If there exists a sequence (yn )n∈N which converges to 0 such that ∀n ∈ N, |xn −l| ≤ yn ,
then the sequence (xn )n∈N converges to l.

3.) If (xn )n∈N converges to l, then (|xn |)n∈N converges to |l|. And the reciprocal statement
is false.

4.) Let m ∈ K and m < l (resp. m > l), if (xn )n∈N converges to l, then there exists
n0 ∈ N such that xn > m (resp. xn < m) for all n ≥ n0 .

5.) If (xn )n∈N is a monotone sequence which converges to l, then the supremum (resp.
infimum) of (xn )n∈N equals to l in the case where (xn )n∈N is increasing (resp. de-
creasing).

Proof. 1.) Let (xn )n∈N be a sequence converging to l ∈ K. Then, there exists N ∈ N such
that for all n ∈ N, n ≥ N ⇒ |xn − l| ≤ 1. Hence, |xn | ≤ |l| + 1 for all n ≥ N . So, for
all m ∈ N, |xm | ≤ max{|x0 |, |x1 |, · · · , |xN −1 |, |l| + 1}. Thus, the sequence (xn )n∈N is
bounded.

2.) Let ε > 0, since the sequence (yn )n∈N converges to 0, there exists N ∈ N such that
|yn | ≤ ε once n ≥ N . Therefore, for n ≥ N , |xn − l| ≤ yn ≤ |yn | ≤ ε. This implies
that (xn )n∈N converges to l.

3.) It follows from the triangular inequality that for all n ∈ N, ||xn | − |l|| ≤ |xn − l|. Let
ε > 0, since xn −→ l, there exists N ∈ N such that |xn − l| ≤ ε once n ≥ N .
n→+∞
Combining with the inequality on ||xn | − |l||, we have then ||xn | − |l|| ≤ |xn − l| ≤ ε
once n ≥ N . That means, |xn | −→ |l|.
n→+∞

The reciprocal statement is false. Here is one counter-example. Consider (xn )n∈N
defined by, for all n ∈ N, xn = (−1)n . Then, (|xn |)n∈N converges to 1 as a constant
sequence whereas the sequence (xn )n∈N diverges.
38 SYSTEM OF REAL NUMBERS

l−m
4.) Since xn −→ l, we then choose ε = > 0. There exits then n0 ∈ N such
n→+∞ 2
l−m
that for all n ∈ N, n ≥ n0 implies that |xn − l| ≤ . We have in particular,
2
l+m
m< ≤ xn for all n ≥ n0 .
2
5.) We prove at the first step that l is an upper bound of the sequence (xn )n∈N . Assuming
on the contradictory that there exists N ∈ N such that xN > l. We then set δ =
xN − l
> 0. Since (xn )n∈N is an increasing sequence, we have, for all n ∈ N, n ≥ N ,
2
xn ≥ xN > l + δ. This implies that |xn − l| > δ once n ≥ N , which contradict with
the convergence. Thus, l ≥ xN for all N ∈ N.
Next, we prove that l = sup xn . Let ε > 0, from the convergence, there exists N ∈ N
n∈N
such that |xn − l| ≤ ε once n ≥ N . We have then, xn ≥ l − ε. Thus, l = sup xn
n∈N
according to the characterization of the supremum.

Lemma 1.3.13 (Sandwich). Let (xn )n∈N and (yn )n∈N be sequences of elements in K. We
assume that those two sequences converge to the same limit l ∈ K. If (zn )n∈N be a sequence
such that for all n ∈ N, xn ≤ zn ≤ yn , then (zn )n∈N also converge to this limit l.

Proof. For all n ∈ N, it follows from the triangular inequality, we have,

|zn − l| = |zn − xn + xn − l| ≤ |zn − xn | + |xn − l|

≤ |yn − xn | + |xn − l| = |yn − l + l − xn | + |xn − l|
≤ |yn − l| + 2|xn − l|.

Let ε > 0, since the sequences (xn )n∈N and (yn )n∈N converge simultaneously to l, there
ε ε
exists Nx , Ny ∈ N such that for all n ∈ N, |xn − l| ≤ if n ≥ Nx as well as |yn − l| ≤ if
3 3
n ≥ Ny . So, for all n ∈ N, n ≥ max{Nx , Ny } implies that

ε 2ε
|zn − l| ≤ |yn − l| + 2|xn − l| ≤ + ≤ ε.
3 3
Thus, the claim follows.

Proposition 1.3.14. Let (xn )n∈N and (yn )n∈N be two sequences of elements in K which
converge respectively to x and y. Then,

• if xn ≤ yn from certain rank then x ≤ y

• for all λ ∈ K, xn + λyn −→ x + λy;

n→+∞

• xn yn −→ xy.
n→+∞

xn xn x
• if y ̸= 0, then is well-defined from certain rank and lim = .
yn n→∞ yn y
1.3. SEQUENCES IN ORDERED FIELDS 39

x−y
Proof. • We assume in the contrary that x > y. We set ε = > 0. Since (xn )n∈N
3
converges to x and (yn )n∈N converges to y, there exits then Nx , Ny ∈ N such that
for all n ∈ N, |xn − x| ≤ ε if n ≥ Nx and |yn − y| ≤ ε if n ≥ Ny . So, for n ∈ N,
n ≥ max{Nx , Ny }, we have
x−y x + 2y 2x + y x−y
yn ≤ y + ε = y + = < =x− = x − ε ≤ xn .
3 3 3 3
This contradicts to the assumption of “xn ≤ yn from certain rank”. In conclusion,
x ≤ y.

• The case where λ = 0 corresponds exactly to the convergence of the sequence (xn )n∈N
and it is immediate. We assume now λ ̸= 0.
Let ε > 0, from the convergence of the sequences (xn )n∈N and (yn )n∈N , there exits
ε ε
then Nx , Ny ∈ N such that for all n ∈ N, |xn − x| ≤ if n ≥ Nx and |yn − y| ≤
2 2|λ|
if n ≥ Ny . Then, for n ∈ N and n ≥ max{Nx , Ny }, it follows from the triangular
inequality that,
ε ε
|xn + λyn − (x + λy)| ≤ |x − xn | + |λ||y − yn | ≤ + |λ| = ε.
2 2|λ|
Hence, xn + λyn −→ x + λy.
n→+∞

• We begin with the case where x = 0 or y = 0. Without of losing the generality, we

assume that x = 0. Since the sequence (yn )n∈N converges, it is a bounded sequence.
That is, there exists My,0 > 0 such that |yn | ≤ My,0 for all n ∈ N. Let ε > 0, from
ε
the assumption that xn −→ 0, there exists Nx,0 ∈ N such that |xn | ≤ for all
n→+∞ My,0
ε
n ∈ N, n ≥ Nx,0 . In the consequences, |xn yn | ≤ |xn ||yn | ≤ My,0 = ε. Thus,
My,0
xn yn −→ 0 = xy.
n→+∞
We consider now the cases x ̸= 0 and y ̸= 0. Applying the first result in Proposi-
tion 1.3.12, the sequence (yn )n∈N is bounded by certain positive number My > 0. Let
ε > 0, we shall use the convergences xn −→ x and yn −→ y to deduce that there
n→+∞ n→+∞
ε
exists Nx , Ny ∈ N and such that for all n ∈ N, |xn − x| ≤ if n ≥ Nx as well as
2My
ε
|yn − y| ≤ if n ≥ Ny . Using the triangular inequality, we can obtain that,
2|x|

|xn yn − xy| = |xn yn − xyn + xyn − xy|

ε ε
≤ |xn − x||yn | + |x||yn − y| ≤ My + |x| = ε.
2My 2|x|
In conclusion, xn yn −→ xy.
n→+∞

• Since y ̸= 0 and yn −→ y, there exists Ny,0 ∈ N such that for all n ∈ N, n ≥ Ny,0 ,
n→+∞
|y|
we have |yn − y| ≤ . Hence, it follows from the second triangular inequality,
2
|y|
|yn | = |yn − y + y| ≥ ||y| − |yn − y|| ≥ > 0. (1.1)
2
40 SYSTEM OF REAL NUMBERS
 
xn
In other words, |yn | > 0 from certain rank and thus the sequence is well-
yn n∈N
defined from certain rank.
xn x 1 1
To show that −→ , it is sufficient to show that −→ and the general
yn n→+∞ y yn n→+∞ y
case can be considered as the product of two convergent sequences. Let ε > 0, there
|y|2 ε
exists Ny ∈ N such that |yn − y| ≤ if n ≥ Ny . So, for n ∈ N, n ≥ max{Ny,0 , Ny },
2
we obtain following from (1.1),

1 1 |yn − y| 2|yn − y| 2 |y|2 ε

− = ≤ ≤ = ε.
yn y |yn ||y| |y|2 |y|2 2

1 1
Thus, −→ .
yn n→+∞ y

Remark 1.3.15. • Here, “xn ≤ yn from certain rank” stands that: there exists n0 ∈ N
such that for all n ≥ n0 , xn ≤ yn .

• The second point of this proposition shows that the set of convergent sequences is in
fact a subspace of bounded sequences and the mapping (xn )n∈N 7→ lim xn is linear.
n→+∞

Theorem
  1.3.16. An ordered field (K, +, ·, ≤) is Archimedean if, and only if the sequence
1
converges to 0.
n + 1 n∈N

Proof. We prove this theorem by double implication.

⇒: We assume now (K, +, ·, ≤) is an Archimedean ordered field. Let ε >0, there

 exists
1 1 1
m ∈ N∗ such that < m or that < ε. Since the sequence is
ε m n + 1 n∈N
decreasing, we deduce that for all n ∈ N, n ≥ m, it follows,

1 1 1
= ≤ < ε.
n+1 n+1 m
 
1
Thus, the sequence converges to 0.
n+1 n∈N


1
⇐: We assume that the sequence converges to 0. Let x ∈ K. In the case
n + 1 n∈N
where x ≤ 0, we then choose m = 1 ∈ Z. We assume now x > 0. From the convergence
1 1
of the sequence, there exists n ∈ N such that ≤ or that x ≤ n + 1. We then
n+1 x
choose m = n + 2 ∈ Z. So, the ordered field (K, +, ·, ≤) satisfies the Archimedean
property.
1.3. SEQUENCES IN ORDERED FIELDS 41

Definition 1.3.17 (Monotone convergence property). We say an ordered field (K, +, ·, ≤)

satisfies the monotone convergence property if any bounded and monotone sequence
of elements in K converges to a limit in K.

Example 1.3.18. Consider the sequence (yn )n∈N ∈ QN defined by

1 1
y0 = , and ∀n ∈ N, yn+1 = 1 .
2 2+ 2+yn

Show that (yn )n∈N is a bounded decreasing sequence in Q and it does not converge in Q.
1 1
Solution. If yn ∈ (0, ], we can deduce that yn+1 > 0 and that |yn+1 | ≤ . Therefore,
2 2
the boundedness of the sequence (yn )n∈N can be obtained from this implication and a
simple induction. On the other side, we can also prove that (yn )n∈N is decreasing using the
induction arguments. The detailed proof is left as an exercise.
If the sequence (yn )n∈N converges, we may thus assume that lim yn = l. Then, using
n→+∞
1
the algebraic operations on the convergent sequences, we obtain that l = 1 , which
2 + 2+l
√
implies that l = 2 − 1 ∈/ Q. So, the sequence (yn )n∈N does not converge in Q.

Proposition 1.3.19. The monotone convergence property implies Archimedean property.

Proof. We assume in the contrary that the Archimedean property does not hold. That
means, there exists certain x ∈ K such that for all n ∈ N, n ≤ x. We then consider the
sequence (sn )n∈N with ∀n ∈ N, sn = n. Then, the sequence (sn )n∈N is strictly increasing
and bounded from above by x. From the monotone convergence property, it has a limit
1
l ∈ K. Then, there exists N ∈ N such that for all n ∈ N, n ≥ N we have |sn − l| ≤ . And
3
it holds,
1 1 2
1 = |N − (N + 1)| = |sN − sN +1 | ≤ |sN − l| + |sN +1 − l| ≤ + = .
3 3 3
Thus, the contradiction. In conclusion, the Archimedean property is satisfied.

1.3.3 Extraction of Sequences

Definition 1.3.20. Let (xn )n∈N be a sequence of elements in K. We say a sequence (yn )n∈N
to be a subsequence or an extracted sequence of (xn )n∈N if there exists a strictly increasing
function φ : N → N, which is called as the extraction, such that

∀n ∈ N, yn = xφ(n) .

In this case, we often denote φ(j) = nj and thus yj = xφ(j) = xnj for j ∈ N.

Definition 1.3.21. Let (xn )n∈N be a sequence of elements in K and h ∈ K. We say h is

a cluster point of the sequence (xn )n∈N if there exists a subsequence (xnj )j∈N of (xn )n∈N
such that (xnj )j∈N converges to h.
42 SYSTEM OF REAL NUMBERS

Proposition 1.3.22. Any subsequences of a convergent sequence converges to the same

limit.

Proof. Let (xn )n∈N be a convergent sequence and let (yn )n∈N be a subsequence of (xn )n∈N
with ∀n ∈ N, yn = xφ(n) where φ designs the corresponding extraction. Since (xn )n∈N
converges, we design by l ∈ K its limit. Let ε > 0, there exists N ∈ N such that for
all n ∈ N, n ≥ N , |xn − l| ≤ ε. Others, since φ is strictly increasing function, we have
φ(m) ≥ m for all m ∈ N. So, if n ≥ N , |yn − l| = |xφ(n) − l| ≤ ε. Thus, the sequence
(yn )n∈N converges to l.

Example 1.3.23. Consider the sequence (xn )n∈N with xn = (−1)n for all n ∈ N. Then it
possesses two cluster points {1, −1}.

Solution. The subsequence (x2n )n∈N is a constant sequence (1)n∈N thus it converges to 1.
In addition, the subsequence (x2n+1 )n∈N is a constant sequence (−1)n∈N and it converges
to −1. So, 1 and −1 are cluster points of the sequence ((−1)n )n∈N .

Definition 1.3.24 (Bolzano-Weierstrass property). We say an ordered field (K, +, ·, ≤)

satisfies the Bolzanoiii -Weierstrassiv property if any bounded sequence of elements in
K possesses a convergent subsequence.

Proposition 1.3.25. Assume that (K, +, ·, ≤) satisfies the Bolzano-Weierstrass property,

then a sequence (xn )n∈N of elements in K is convergent if, and only if (xn )n∈N is bounded
and possesses a unique cluster point.

Proof. We prove this proposition by double implications.

⇒: We assume that the sequence (xn )n∈N ∈ KN converges to l ∈ K. Then, (xn )n∈N is
bounded since it is convergent. Also, any subsequence of (xn )n∈N converges to the
same limit l, which is equivalent to say that l is the unique cluster point of (xn )n∈N .

⇐: We assume now the sequence (xn )n∈N is bounded and possesses a unique cluster point
l ∈ K. We assume furthermore in the contrary that (xn )n∈N does not converge to
l. Then, there exists δ > 0 such that for all N ∈ N, there exists nN ∈ N such
that nN ≥ N and |xnN − l| > δ. Now, we construct an extraction φ by recurrence.
Firstly, φ(0) = n0 ∈ N. Then, for k ∈ N∗ if φ(0), φ(1), · · · , φ(k − 1) have already
constructed, we define φ(k) = nφ(k−1)+1 ∈ N. With this construction, ∀k ∈ N,
φ(k + 1) = nφ(k)+1 ≥ φ(k) + 1, which implies that the function φ : N → N is
strictly increasing hence it forms an extraction. Also, ∀k ∈ N, |xφ(k) − l| > δ. Since
(xn )n∈N is bounded, this subsequence (xφ(n) )n∈N is also bounded. Using the Bolzano-
Weierstrass property, the sequence (xφ(n) )n∈N possesses a convergent subsequence,
which is denoted by (yn )n∈N . We have thus yn −→ l′ ∈ K. We remark that l′ is
n→+∞
iii
Bernard Bolzano (1781 - 1848) Bohemian (Now, Czech Republic) mathematician and philosopher. He
concerned mostly on the logical statements in mathematics. It was him who firstly introduced the famous
ε − δ formulation, which gave the rigor and abstraction of mathematical analysis.
iv
Karl Theodor Wilhelm Weierstraß(1815 - 1897) German mathematician. Simply recognized as the
“father of modern analysis”, he gave the rigorous formulations on theorems in analysis, which are adapted
until nowadays.
1.3. SEQUENCES IN ORDERED FIELDS 43

also a cluster point of the sequence (xn )n∈N . On the other hands, since (yn )n∈N is a
subsequence of (xφ(n) )n∈N , we have then |yn − l| > δ for all n ∈ N. Then, for n ∈ N
being large enough, we have

δ δ
|l − l′ | ≥ ||l − yn | − |yn − l′ || > δ − = > 0.
2 2
We can thus deduce that |l′ − l| > 0 ⇔ l ̸= l′ . This is contradictory to the uniqueness
of the cluster point l. In conclusion, (xn )n∈N converges to l.

1.3.4 Cauchy Sequences

Definition 1.3.26. Let (xn )n∈N be a sequence of elements in K. We say the sequence
(xn )n∈N is a Cauchyv sequence if

∀ε > 0, ∃N ∈ N s.t. ∀n, m ∈ N, N ≤ n ≤ m ⇒ |xn − xm | ≤ ε.

Remark 1.3.27. Roughly speaking, a sequence is said to be Cauchy if the difference between
two arbitrary terms tends to zero. We can also reformulate the above definition as

∀ε > 0, ∃N ∈ N s.t. ∀n, p ∈ N, N ≤ n ⇒ |xn − xn+p | ≤ ε.

Proposition 1.3.28. Any convergent sequence is a Cauchy sequence.

Proof. Let (xn )n∈N ∈ KN such that xn −→ l ∈ K. Let ε > 0, by the convergence of
n→+∞
ε
(xn )n∈N , there exists N ∈ N, such that for all n ∈ N, n ≥ N implies |xn − l| ≤ . So, for
2
all n, m ∈ N, N ≤ n ≤ m, we have
ε ε
|xn − xm | ≤ |xn − l| + |xm − l| ≤ + ≤ ε.
2 2
Thus, the sequence (xn )n∈N is a Cauchy sequence.

Lemma 1.3.29. Any Cauchy sequence is bounded.

Proof. Let (xn )n∈N be a Cauchy sequence. There exists N ∈ N such that for all m ∈ N,
m ≥ N implies that |xm − xN | ≤ 1. In other words, xN − 1 ≤ xm ≤ xN + 1, we have
then |xm | ≤ max{|xN + 1|, |xN − 1|} ≤ |xN | + 1. So, for all n ∈ N, we have |xn | ≤
max{|x0 |, |x1 |, · · · , |xN −1 |, |xN | + 1}. It shows that the sequence (xn )n∈N is bounded.

Lemma 1.3.30. If a Cauchy sequence possesses a cluster point, then it converges.

v
Augustin-Louis Cauchy (1789 - 1857) French mathematician. Various branches in modern mathematics
can be seek to Cauchy’s fundamental works. He was believed to be one of the most influential mathematician
in the history.
44 SYSTEM OF REAL NUMBERS

Proof. Let (xn )n∈N ∈ KN be a Cauchy sequence and (xnj )j∈N be a subsequence of (xn )n∈N
such that xnj −→ l ∈ K where l is a cluster point of (xn )n∈N .
j→+∞
Let ε > 0, since the sequence (xn )n∈N is Cauchy and the subsequence (xnj )j∈N converges,
ε
there exists N, N ′ ∈ N such that for all n, m ∈ N, N ≤ n ≤ m implies that |xn − xm | ≤
2
ε
as well as for all j ∈ N, j ≥ N ′ implies that |xnj − l| ≤ . Since the extraction is strictly
2
increasing on N, we have nj ≥ j for all j ∈ N. We set now M = nmax{N,N ′ } ∈ N, we have
then M ≥ max{N, N ′ } also M = max{nN , nN ′ }. Let n ∈ N such that n ≥ M , we have
then,
ε ε
|xn − l| ≤ |xn − xM + xM − l| ≤ |xn − xM | + |xM − l| ≤ + ≤ ε.
2 2
Thus, the sequence (xn )n∈N converges to l.

1.4 Completeness
In this section, we firstly prove the equivalence between the properties on the ordered field
introduced previously. They state effectively one common thing, the so-called completeness.
Next, we “define” the system of real numbers R as a complete ordered field satisfying the
Archimedean property. On the other hand, one can proceed the classical way to construct
steps by steps from natural numbers N, then integers Z, next rational numbers Q, till
real numbers R and eventually complex numbers C. For those who are interested in this
approach, we recommend them to consult Section 1.A.
Theorem 1.4.1. In an ordered field (K, +, ·, ≤), the least upper bound property and the
monotone convergence property are equivalent.
Proof. We assume at first that (K, +, ·, ≤) satisfies the least upper bound property. Let
(xn )n∈N ∈ KN be a bounded sequence. Without of losing the generality, we suppose that
(xn )n∈N is an increasing sequence. Then, the set S := {xn | n ∈ N} is a non-empty and
bounded from above subset of K. From the least upper bound property, there exists l ∈ K
such that l = sup(S) = sup xn . We next show that lim xn = l.
n∈N n→+∞
Let ε > 0, since l = sup(S), it follows from the characterization of supremum that there
exists N ∈ N such that xN > l − ε. On the other hand, since (xn )n∈N is supposed to be
increasing and l is an upper bound of (xn )n∈N , we deduce that for all n ∈ N, if n ≥ N , we
have

l − ε < xN ≤ xn ≤ l.

It implies that lim xn = l. In the consequences, the monotone convergence property is

n→+∞
verified.
Reciprocally, we assume now that (K, +, ·, ≤) satisfies the monotone convergence prop-
erty. Consider a subset S ′ ⊂ K, which is non-empty and bounded from above. Since S ′ is
non-empty, there exists x0 ∈ S ′ . In parallel, since S ′ is bounded from above, there exists
y0 ∈ K being an upper bound of S ′ . We have immediately that x0 ≤ y0 .
We then construct two sequences (xn )n∈N and (yn )n∈N by recurrence as follows. The
xn + yn
initial terms x0 and y0 have previously defined. For n ∈ N, if is an upper bound of
2
xn + yn xn + yn
S ′ , we set xn+1 = xn and yn+1 = , otherwise, we set xn+1 = and yn+1 = yn .
2 2
1.4. COMPLETENESS 45

The construction shows that the sequence (xn )n∈N is increasing and that the sequence
(yn )n∈N is decreasing. We have at the same time that for all n ∈ N, x0 ≤ xn ≤ yn .
So, the sequence (yn )n∈N is bounded from below and thus be bounded. We then apply
the monotone convergence property to deduce that yn −→ l′ ∈ K. As an important
n→+∞
fact, we remark that for all n ∈ N, yn is an upper bound of S ′ whereas xn is always
not an upper bound of S ′ . In addition, the construction provides that for all n ∈ N,
|x0 − y0 |
|xn − yn | = −→ 0. We remark that this convergence holds from Exercise 1.6.15
2n n→+∞
and the Archimedean property can be derived from Proposition 1.3.19. In the consequences,
|xn − l′ | ≤ |xn − yn | + |yn − l′ | −→ 0, which is equivalent to say xn −→ l′ .
n→+∞ n→+∞
Now, we show that l′ = sup(S ′ ). Assuming in the contrary that l′ were not an upper
bound of S ′ . Then there exists a ∈ S ′ such that a > l′ . From the convergence yn −→ l′ ,
n→+∞
we have (from Proposition 1.3.12-4.) then yn < a from certain rank. This is contradictory
to the fact that yn remains always an upper bound of S ′ . In consequences, l′ is an upper
bound of S ′ . On the other hand, let ε > 0, since xn −→ l′ and (xn )n∈N is increasing,
n→+∞
there exist N ∈ N such that l′ − ε ≤ xN ≤ l′ . As xn is always not an upper bound of
S ′ , there exists aN ∈ S ′ such that aN > xN . Combining the previous results, we have
l′ − ε ≤ xN < aN ∈ S ′ . So, following from the characterization of supremum, l′ = sup(S ′ ).
In conclusion, the least upper bound property and monotone convergence property are
equivalent.

Theorem 1.4.2. In an ordered field (K, +, ·, ≤), the monotone convergence property implies
the Bolzano-Weierstrass property.
Proof. Let (xn )n∈N ∈ KN be a bounded sequence. There exists a0 , b0 ∈ K such that ∀n ∈ N,
xn ∈ [a0 , b0 ]. We then construct the sequences (an )n∈N , (bn )n∈N ∈ KN by recurrence as
follows.
Let n ∈ N, assuming that the constructed segment [an ,bn ] containsan infinity
 numbers 
an + bn an + bn
of terms in (xn )n∈N . Then, at least one of the segments an , and , bn
2 2
contains an infinity numbers of terms in (xn )n∈N . So, in the case where the segment

an + bn
an , contains an infinity numbers of terms in (xn )n∈N , then we set an+1 = an
2
an + bn an + bn
and bn+1 = . Otherwise, we set an+1 = and bn+1 = bn .
2 2
The construction shows that the sequences (an )n∈N and (bn )n∈N are respectively an
increasing and decreasing sequence. And it is straightforward that for all n ∈ N, a0 ≤
an ≤ bn ≤ b0 , which shows the two sequences are both bounded. Applying the monotone
convergence property, we have an −→ la ∈ K and bn −→ lb ∈ K.
n→+∞ n→+∞
On the other hand, let C = |a0 − b0 | ≥ 0. We deduce, using a simple induction from the
C
construction, that ∀n ∈ N, |an − bn | ≤ n −→ 0. So, for all n ∈ N, it holds,
2 n→+∞
|la − lb | ≤ |la − an | + |an − bn | + |bn − lb | −→ 0 + 0 + 0 = 0.
n→+∞

Therefore, the sequences (an )n∈N and (bn )n∈N converges to the same limit l := la = lb .
Now we consider the subsequence (xφ(n) )n∈N of (xn )n∈N defined by recurrence as follows.
We set φ(0) = 0. Let n ∈ N, since the segment [an+1 , bn+1 ] contains an infinity numbers of
46 SYSTEM OF REAL NUMBERS

terms in (xn )n∈N , it is always possible to choose φ(n + 1) ∈ N such that φ(n+ 1) > φ(n) and
xφ(n+1) ∈ [an+1 , bn+1 ]. Then, the subsequence (xφ(n) )n∈N satisfies ∀n ∈ N, an ≤ xφ(n) ≤ bn .
Since the sequences (an )n∈N and (bn )n∈N converge to the same limit l, it follows from
Sandwich Lemma 1.3.13 that xφ(n) −→ l.
n→+∞
In conclusion, the Bolzano-Weierstrass property is verified.

Theorem 1.4.3. In an ordered field (K, +, ·, ≤), if the Bolzano-Weierstrass property holds
then any Cauchy sequence converges.

Proof. Let (xn )n∈N ∈ KN be a Cauchy sequence. It follows from Lemma 1.3.29 that (xn )n∈N
is bounded. So, the Bolzano-Weierstrass property implies that there exists a convergent
subsequence (xnj )j∈N of (xn )n∈N . We then apply Lemma 1.3.30 to deduce that the Cauchy
sequence (xn )n∈N converges. Hence, the claim follows.

Theorem 1.4.4. In an Archimedean ordered field (K, +, ·, ≤), if any Cauchy sequence
converges then the monotone convergence property holds.

Proof. Let (xn )n∈N ∈ KN be a bounded monotone sequence. Without of losing the general-
ity, we can suppose that (xn )n∈N is increasing.
Assume now in the contrary that (xn )n∈N is not a Cauchy sequence. Then, there exists
δ > 0 such that for all N ∈ N, there exists n, m ∈ N, N ≤ n ≤ m which satisfy |xn −xm | > δ.
In other words, there exists two subsequences (xnj )j∈N and (xmj )j∈N of (xn )n∈N which satisfy
∀j ∈ N, |xnj − xmj | > δ. Moreover, it is sufficient to extract in the sequences (nj )j∈N and
(mj )j∈N to have the property that ∀j ∈ N, mj ≤ nj+1 .
On the other hand, since the field K is Archimedean, for all M ∈ K, there exists NM ∈ N
M
such that < NM , or that M < δNM . In the consequence,
δ
NM NM mNM
X X X
M < δNM = δ< |xnj − xmj | ≤ |xk − xk+1 | = xmNM − x0 .
j=1 j=1 k=1

This inequality shows that the sequence (xn )n∈N is in fact unbounded. Thus, the contra-
diction. So, the sequence (xn )n∈N is a Cauchy sequence.
Since any Cauchy sequences converge in K, then the monotone bounded sequence
(xn )n∈N converges in K. In conclusion, the monotone convergence property is verified.

In summary, we have the following equivalence result.

Corollary 1.4.5. In an Archimedean ordered field (K, +, ·, ≤), the following assertions are
equivalent.

1.) Any non empty and bounded from above (resp. below) subset possesses its supremum
(resp. infimum) in K.

2.) Any bounded monotone sequence converges.

3.) Any bounded sequence possesses a convergent subsequence.

4.) Any Cauchy sequence converges.

1.5. LIMIT INFERIOR AND LIMIT SUPERIOR 47

Definition 1.4.6. We say an Archimedean ordered field (K, +, ·, ≤) is complete if it

satisfies one of the above equivalent property.

Theorem 1.4.7. There exists a complete Archimedean ordered field.

Proof. Admitted theorem. For the construction, one can find it out in many classical
textbooks of Analysis, for example [5]. Also, we provide that in a long series of exercises in
Section 1.A.

Definition 1.4.8. The system of real numbers R is by definition a complete Archimedean

ordered field.

Theorem 1.4.9. The system of complex numbers C is complete in the sense of any Cauchy
sequence converges.
Proof. For all z ∈ C, we have |Re(z)| ≤ |z| and |Im(z)| ≤ |z|. Those inequality shows that
for any Cauchy sequence (zn )n∈N ∈ CN , its real part sequence (Re(zn ))n∈N ∈ RN as well
as its imaginary part sequence (Im(zn ))n∈N are both Cauchy sequences of real numbers. It
follows from the completeness of the field R, those two sequences converge in R. Therefore,
the sequence (zn )n∈N converges in C. In conclusion, the field C is complete.

1.5 Limit Inferior and Limit Superior

In this section, we consider a sequence (xn )n∈N of members in R.
Definition 1.5.1. • We say the sequence (xn )n∈N diverges to positive infinity or
tends to +∞ if

∀M ∈ R, ∃N ∈ N s.t. ∀n ∈ N, n ≥ N ⇒ xn ≥ M.

We denote this relation by lim xn = +∞ or by xn −→ +∞.

n→+∞ n→+∞

• We say the sequence (xn )n∈N diverges to negative infinity or tends to −∞ if

∀M ∈ R, ∃N ∈ N s.t. ∀n ∈ N, n ≥ N ⇒ xn ≤ M.

We denote this relation by lim xn = −∞ or by xn −→ −∞.

n→+∞ n→+∞

Definition 1.5.2. • The limit superior of (xn )n∈N , which we denoted by lim sup xn
n→+∞
or lim xn , is the infimum of the set {sup {xn | n ≥ k} | k ∈ N}.
n→+∞

• The limit inferior of (xn )n∈N , which we denoted by lim inf xn or lim xn , is the
n→+∞ n→+∞
supremum of the set {inf {xn | n ≥ k} | k ∈ N}.

Proposition 1.5.3. • The limit superior exists only if (xn )n∈N is bounded from above
and doesn’t diverge to negative infinity.
48 SYSTEM OF REAL NUMBERS

• The limit inferior exists only if (xn )n∈N is bounded from below and doesn’t diverge to
positive infinity.

• With the notations of supremum/infimum of a sequence, we have also,

lim sup xn = inf sup xn , and lim inf xn = sup inf xn .

n→+∞ k∈N n≥k n→+∞ k∈N n≥k

• In the cases of existence, it holds that

lim sup xn = lim sup xn , and lim inf xn = lim inf xn .

n→+∞ k→+∞ n≥k n→+∞ k→+∞ n≥k

Proof. • Assuming by contradiction that (xn )n∈N is not bounded from above. In other
words, for any k ∈ N, the set {xn | n ≥ k} is not bounded from above. Then the
numbers sup xn is not defined for all k ∈ N, as a direct consequence, the limit superior
n≥k
can not be defined in this case.
We assume now by contradiction that (xn )n∈N diverges to −∞. For all M ∈ R,
there exists N ∈ N such that ∀k ∈ N, k ≥ N , we (
have xk ≤ M . )
This implies that

sup xn ≤ M if k ≥ N . In the consequences, the set sup xn k ∈ N is not bounded

n≥k n≥k
from below. Thus, the limit superior does not exists.
In conclusion, the claim follows.

• It is sufficient to adapt the proof in the previous point.

• It is just a transcription of the definition using those notations.

• Consider for example the sequence (yk )k∈N defined by ∀k ∈ N, yk = sup xn , (yk )k∈N
n≥k
is clearly a decreasing sequence. The existence of lim sup xn shows that the sequence
n→+∞
(yk )k∈N is bounded from below. Using the monotone convergence property, the se-
quence (yk )k∈N converges in R. In addition, from the fact that any monotone sequence
converges to its supremum/infimum, we have inf yk = lim yk . Then, the claim fol-
k∈N k→+∞
lows. The claim on lim inf follows from the analogical arguments.

Proposition 1.5.4. The limit superior is the largest cluster point of (xn )n∈N . Symmetri-
cally, the limit inferior is the smallest cluster point of (xn )n∈N .

Proof. We deal with the case of limit superior, and the case of limit inferior can be deduced
from exactly the same arguments with a tiny modification.
We assume now lim sup xn = l ∈ R. From Proposition 1.5.3, the sequence (xn )n∈N
n→+∞
is bounded from above. We can thus define for all k ∈ N, yk := sup xn ∈ R via the
n≥k
least upper bound property. In addition, also from Proposition 1.5.3, the sequence (yk )k∈N
converges to l. For all k ∈ N, using the characterization of supremum, there exists mk ∈
1
N, mk > k such that yk+1 − < xmk ≤ yk+1 . We then extract a subsequence of
k+1
(xn )n∈N by recurrence. We set φ(0) = m0 ∈ N and for all n ∈ N, we choose φ(n +
1.6. EXERCISES 49

1) = mφ(n) . The condition mk > k implies that φ(n + 1) = mφ(n) > φ(n), which means
φ is strictly increasing,
 thus it is effectively
 an extraction. Nevertheless, the sequences
1
(yφ(n)+1 )n∈N and yφ(n)+1 − converge as the subsequences of convergences
φ(n) + 1 n∈N
sequences. Moreover, they converges to the same limit l. Consider now the subsequence
1
(xφ(n) )n∈N of (xn )n∈N , we have, for all n ∈ N, yφ(n)+1 − ≤ xφ(n+1) ≤ yφ(n)+1 . We
φ(n) + 1
then apply the Sandwich Lemma 1.3.13 to deduce that lim xφ(n) = l. This shows that l
n→+∞
is a cluster point of (xn )n∈N .
Let l′ ∈ R be an other cluster point of (xn )n∈N . Then, there exists another subsequence
(xψ(n) )n∈N of (xn )n∈N which converges to l′ . We assume now in the contrary that l′ > l.
l + l′ l + l′
From the simple result l < < l′ and Proposition 1.3.12-4.), we have that xψ(n) ≥
2 2
for n large enough. In the consequences, for n large enough, it holds,

l + l′
yψ(n) = sup xj ≥ xψ(n) ≥ .
j≥ψ(n) 2

This inequality contradicts with the fact that yk −→ l. Thus, l′ ≤ l.

k→+∞
In conclusion, the limit superior is the largest cluster point.

1.6 Exercises
In the following exercises, (K, +, ·, ≤) designs an ordered field.

Ordered Fields
Exercise 1.6.1. Let a, b, x, y, z ∈ K. Show that

1.) If a + x = a, then x = 0.
If a · x = a and a ̸= 0, then x = 1.

2.) If a + x = 0, then x = −a.

If a · x = 1 and a ̸= 0, then x = a−1 .

3.) If x · y = 0, then x = 0 or y = 0.

4.) If x ≤ y < z or x < y ≤ z then x < z. (transitivity of <)

5.) If a < b, then a + x < b + x. (compatibility of < and +)

If 0 < a and 0 < b, then 0 < a · b. (compatibility of < and ·)

6.) If a + x = b + x, then a = b.
If a + x ≤ (<)b + x, then a ≤ (<)b.
If a · x = b · x and x ̸= 0, then a = b.
If a · x ≤ (<)b · x and x > 0, then a ≤ (<)b.

7.) 0 · x = 0.

8.) −(−x) = x.

9.) −x = (−1) · x.
50 SYSTEM OF REAL NUMBERS

10.) If x ̸= 0, then x−1 ̸= 0 and (x−1 )−1 = x.

11.) If x ̸= 0 and y ̸= 0, then x · y ̸= 0 and (x · y)−1 = x−1 · y −1 .

12.) If x ≤ (<)y and 0 ≤ (<)z, then x · z ≤ (<)y · z.

If x ≤ (<)y and 0 ≥ (>)z, then x · z ≥ (>)y · z.

13.) If x ≤ (<)0 and y ≤ (<)0, then x · y ≥ (>)0.

If x ≤ (<)0 and y ≥ (>)0, then x · y ≤ (<)0.

14.) 0 < 1 and −1 < 0.

15.) x2 ≥ 0.

16.) If x > (<)0, then x−1 > (<)0.

Exercise 1.6.2. Let x, y ∈ K. Show that

1.) If 0 ≤ x < y, then x2 < y 2 .

2.) If 0 ≤ x, y and x2 < y 2 , then x < y.

And deduce that if x, y ≥ 0, then x ≤ (<)y ⇔ x2 ≤ (<)y 2 .

Exercise 1.6.3. Let | · | be the magnitude on K, and x, y ∈ K. Show that

1.) |x| ≥ 0.

2.) |x| = 0 if, and only if x = 0.

3.) −|x| ≤ x ≤ |x|.

4.) |x · y| = |x| · |y|.

5.) |x + y| ≤ |x| + |y|. (triangular inequality)

6.) ||x| − |y|| ≤ |x − y|. (second form of triangular inequality)

Exercise 1.6.4. Let x, y ∈ K and n ∈ N. Show that

1.) If 0 ≤ x < y, then xn < y n .

2.) If 0 ≤ x, y and xn < y n , then x < y.

Exercise 1.6.5. Let a, b ∈ K. Show that a ≤ b if, and only if for all ε > 0, a < b + ε.
1.6. EXERCISES 51

Subsets in Ordered Fields

Exercise 1.6.6. Show that any non empty subset of N possesses a minimum.

Exercise 1.6.7. Show that any non empty finite subset S of K possesses a maximum and
a minimum.

Exercise 1.6.8. Assuming the ordered field (K, +, ·, ≤) satisfies the least upper bound
property, and ∅ ̸= A ⊂ B ⊂ K with A bounded.
Show that inf(B) ≤ inf(A) ≤ sup(A) ≤ sup(B).

Exercise 1.6.9. Let A and B be bounded subsets in K and K satisfies the least upper
bound property. We introduce
• −A := {−x | x ∈ A};

• A + B := {x + y | x ∈ A, y ∈ B};

• AB := {xy | x ∈ A, y ∈ B}.
Are the following assertions true? Prove them in the case of truth or give a counter example.
1.) sup(−A) = − inf(A);

2.) sup(A + B) = sup(A) + sup(B);

3.) sup(A ∩ B) = min{sup(A), sup(B)};

4.) sup(A ∪ B) = max{sup(A), sup(B)};

5.) sup(AB) = sup(A) sup(B).

Exercise 1.6.10. Let A be an non-empty subset in K such that sup A > 0. Show that
there exists an element x > 0 and x ∈ A.

Sequences in Ordered Fields

Exercise 1.6.11. Let (xn )n∈N and (yn )n∈N be two bounded-from-below sequences. Is the
sequence (xn yn )n∈N bounded from below? Justify your answer.

Exercise 1.6.12. Let (xn )n∈N , (yn )n∈N be two convergent sequences. We assume that
xn −→ lx and yn −→ ly . Show that lim max{xn , yn } = max{lx , ly }.
n→+∞ n→+∞ n→+∞

Exercise 1.6.13. Let (xn )n∈N be a sequence of integers. Show that, if (xn )n∈N converges
in K, then (xn )n∈N is constant from certain rank.

Exercise 1.6.14. Let A be a non-empty subset in K. Show that the following properties
are equivalent.
52 SYSTEM OF REAL NUMBERS

1.) For all x, y ∈ K, x < y, there exists a ∈ A such that x < a < y.

2.) For all z ∈ K, there exists a sequence (an )n∈N ∈ AN such that z = lim an .
n→+∞

Exercise 1.6.15. Let (K, +, ·, ≤) be Archimedean, and 0 < α < 1. Show that lim αn =
n→+∞
0.

Exercise 1.6.16. Let (xn )n∈N be a sequence of elements in K and l ∈ K. Show that
(xn )n∈N converge to l if, and only if (|xn − l|)n∈N converge to 0.

Exercise 1.6.17. We recall here a mathematical proposition is a sentence which we can

determinate if it is true. There are logic rules that we can follow to simplify complicate
propositions. For example, ¬(P ∧ Q) ≡ (¬P ) ∨ (¬Q).
1.) Let x, ε ∈ K, write down the negation of the proposition “x ≤ ε”.

2.) Let x1 , x2 , ε ∈ K, write down the negations of the propositions “x1 ≤ ε and x2 ≤ ε”
as well as “x1 ≤ ε or x2 ≤ ε”.

3.) Let S ⊂ K and ε ∈ K, write down the negations of the propositions “∀x ∈ S, x ≤ ε”
as well as “∃x ∈ S, x ≤ ε”.
We assume from now on for all x ∈ K, there is a proposition that we denote it by
P (x).

4.) Write down the negations of the propositions “∀x ∈ S, P (x)” and “∃x ∈ S, P (x)”.

5.) Let (xn )n∈N be a sequence in K. We assume this sequence converge. Recall the
definition of the convergence using ε − N notations.

6.) We assume now the sequence (xn )n∈N diverge. Use ε − N notations to describe this
condition.

7.) Try to simplify the above ε − N argument using subsequences.

Exercise 1.6.18. Let (K, +, ·, ≤) be Archimedean, I ⊂ K be an interval and f : I → K be

a function. We define the continuity of f as follows.
• We say that f has a limit at c ∈ I if for any sequence (xn )n∈N of elements in I \ {c}
which converges to c, the sequence (f (xn ))n∈N converges in K.

• We say that f is continuous at c ∈ I if f has a limit at c and for any sequence (xn )n∈N
of elements in I \ {c} which converges to c, we have lim f (xn ) = f (c).
n→+∞

Show the following assertions.

1.) f has a limit at c if, and only if there exists L ∈ K satisfying for all ε > 0, there exists
δ > 0 such that for all x ∈ I ∩ [c − δ, c + δ], |f (x) − L| ≤ ε.

2.) f is continuous at c if, and only if for all ε > 0, there exists δ > 0 such that for all
x ∈ I ∩ [c − δ, c + δ], |f (x) − f (c)| ≤ ε.
1.6. EXERCISES 53

Exercise 1.6.19. Let (xn )n∈N ∈ KN . Show that (xn )n∈N converges if, and only if any
subsequence of (xn )n∈N converges.

Exercise 1.6.20. Let (K, +, ·, ≤) be Archimedean, and (xn )n∈N ∈ KN satisfying for all
1
n ∈ N, |xn − xn+1 | < .
n+1
1.) Is (xn )n∈N a Cauchy sequence? Prove it if so, or give a counter example if not.

2.) Assuming that there exists a subsequence of (xn )n∈N which is Cauchy, show that
(xn )n∈N is Cauchy.

3.) Does there exist a Cauchy subsequence of (xn )n∈N ? Prove it if so, or give a counter
example if not.

Exercise 1.6.21. Let (K, +, ·, ≤) be Archimedean, and f : K → K be a contraction, which

means there exists α ∈ K, 0 < α < 1 such that for all x, y ∈ K, |f (x) − f (y)| ≤ α|x − y|. Let
x0 ∈ K and consider the sequence (xn )n∈N defined by recurrence, ∀n ∈ N, xn+1 = f (xn ).
Show that (xn )n∈N is a Cauchy sequence.

Exercise 1.6.22. Let (xn )n∈N and (yn )n∈N be two Cauchy sequences in R, show that the
sequence (|xn − yn |)n∈N converges.

Exercise 1.6.23. Let a ∈ R, we consider the sequence (xn )n∈N ∈ RN defined by recurrence,
x0 = a and for all n ∈ N,

xn+1 = x2n − xn + 1.

For what values of a is the sequence (xn )n∈N (1) monotone? (2) bounded? (3) convergent?
Compute the limit in the cases of convergence.

Exercise 1.6.24. Let b, r ∈ R, b > 1 and r > 0. Show that

!r
sup xr = sup x = br .
x∈(1,b) x∈(1,b)

Exercise 1.6.25. In this exercise, we try to define the exponents and logarithms defined
on real numbers. Make sure you all understand properly the definitions of exponent and
logarithms functions for rational numbers. Let b ∈ R and b > 1.

1.) Show that for all r, s ∈ Q, we have br+s = br bs as well as brs = (br )s .

2.) Let x ∈ R, we set B(x) = bt ∈ R t ∈ Q, t ≤ x . Show that sup B(x) exists in R



and sup B(r) = br if r ∈ Q.

3.) We denote by bx the real number sup B(x). Show that for all x, y ∈ R, bx > 0 and
bx+y = bx by as well as bxy = (bx )y .
54 SYSTEM OF REAL NUMBERS

4.) Show that for all x, y ∈ R, x < y we have bx < by .

5.) Let y ∈ R, y > 0. Show that if there exists u, v ∈ R such that bu < y and bv > y, then
bu+1/n < y and bv−1/n > y for n sufficiently large.

6.) Let y ∈ R, y > 0. We set A = {w ∈ R | bw < y}. Show that sup A exists in R and
x = sup A satisfies bx = y. The real number x is defined as logb y.

Exercise 1.6.26. Consider the sequences (an )n∈N and (bn )n∈N given byvi ,

a0 = 1, b0 = 4;
n
X 1 1
∀n ∈ N∗ , an = , bn = an + .
k! n · n!
k=0

1.) Show that (an )n∈N is strictly increasing as well as (bn )n∈N is strictly decreasing.

2.) Show that the sequence (an )n∈N does not converge in (Q, +, ·, ≤).

Property of Nested Decreasing Segments

In the following exercises, we present an alternative way to define the completeness. It is
the use of adjacent sequences and nested decreasing segments.

Definition 1.6.27. Let (xn )n∈N and (yn )n∈N be sequences of elements in K. We call these
two sequences are adjacent if

• for all n ∈ N, xn ≤ yn ;

• (xn )n∈N is increasing and (yn )n∈N is decreasing;

• lim (xn − yn ) = 0.
n→∞

Definition 1.6.28. • We call a nested decreasing sequence of segments any se-

quence (Sn )n∈N of segments in K such that for all n ∈ N, Sn+1 ⊂ Sn .

• We say the ordered field (K, +, ·, ≤) satisfies the property of nested decreasing
segments \if all nested decreasing sequence of segments has a non empty intersection.
That is, Sn ̸= ∅.
n∈N

We want to show that the nested decreasing segments property is equivalent to the least
upper bound property.

Exercise 1.6.29. Let (Sn )n∈N = ([an , bn ])n∈N be a nested decreasing sequence of segments
in K. Show that, given k ∈ N, ak ≤ bn as well as bk ≥ an for all n ∈ N.

vi
By convention, 0! = 1.
1.6. EXERCISES 55

Exercise 1.6.30. Show that the least upper bound property property implies the property
of nested decreasing segments.

Exercise 1.6.31. Assume that the ordered field (K, +, ·, ≤) satisfies the least upper bound
property. Let (xn )n∈N and (yn )n∈N be adjacent sequences in K.
Show that (xn )n∈N and (yn )n∈N converge to the same limit.

Exercise 1.6.32. Assume that the ordered field (K, +, ·, ≤) satisfies the least upper bound
property. Let Sn be a sequence of nested decreasing segments such that |Sn | −→ 0. Here
n→+∞
|Sn | denotes the\length of the segment Sn .
Show that Sn = {l} with l ∈ K.
n∈N

Exercise 1.6.33. In this exercise, we don’t suppose any further assumption on the ordered
field (K, +, ·, ≤). Let A ⊂ K be a non-empty subset which is bounded from above.
Construct two sequences (xn )n∈N and (yn )n∈N satisfying,
• for all n ∈ N, xn ∈ A and yn is an upper bound of A;

• (xn )n∈N is increasing and (yn )n∈N is decreasing.

• lim (xn − yn ) = 0.
n→∞

Exercise 1.6.34. Assume that the ordered field (K, +, ·, ≤) satisfies the nested decreasing
segments property.
1.) Show that the sequences constructed in the previous exercise converge to the same
limit l ∈ K.

2.) Show that l = sup(A) and conclusion.

Exercise 1.6.35. Show that the nested decreasing segments property is equivalent to the
property that any adjacent sequences converge to the same limit.

Limit Inferior and Limit Superior

In all the following exercises, we consider the sequences (xn )n∈N , (yn )n∈N with strictly pos-
itive real coefficients.
Exercise 1.6.36. State the characterization of the limit superior and limit inferior using
cluster points of (xn )n∈N .

Exercise 1.6.37. We assume that lim inf yn exists, and let l = lim inf yn . We recall that a
n→+∞ n→+∞
necessary condition of the existence of limit inferior is that the sequence (yn )n∈N is bounded
from below. We denote by Ml this lower bound, that means, ∀n ∈ N, yn ≥ Ml .
56 SYSTEM OF REAL NUMBERS

1.) Let ε > 0, show that there exists only a finite numbers of terms in (yn )n∈N which are
smaller then l − ε.
xn+1 √
2.) Show that if lim = 0 then lim n xn = 0.
n→+∞ xn n→+∞
xn+1 √
We assume now lim inf > 0. We recall also that for all a > 0, lim n a = 1.
n→+∞ xn n→+∞

√ xn+1
3.) Let ε > 0, show that n
xn ≥ lim inf − ε from certain rank.
n→+∞ xn

xn+1 √
4.) Deduce that lim inf ≤ lim inf n xn .
n→+∞ xn n→+∞

Exercise 1.6.38 (Theorem of Structure of Subgroups in (R, +)). Let (K, +, ·, ≤) be an

ordered field satisfying the least upper bound property. In this exercise we will study the
subgroups contained in the group (K, +).
We recall here the definition of a subgroup. Let (S, ⋆) be a group with the identity
element e. We say (G, ⋆) is a subgroup in S if

• G ⊂ S;

• ∀x ∈ G, x−1 ∈ G;

• ∀x, y ∈ G, x ⋆ y ∈ G.

In what follows, we consider a subgroup (G, +) in (K, +).

1.) Let α ∈ K. Show that αZ := {αn | n ∈ Z} is a subgroup in (K, +).

2.) Deduce that if α ∈ G, then αZ ⊂ G.

3.) We suppose from now on that G ̸= {0}. Show that there exists α > 0 such that
αZ ⊂ G and the following infimum is well-defined,

a := inf(G ∩ K∗+ ).

4.) Let g, g ′ ∈ G with g < g ′ such that (g, g ′ ) ∩ G = ∅. Show that a = g ′ − g.

5.) We suppose that a > 0. Show that a ∈ G and deduce that G = aZ.

6.) We suppose that a = 0. Let x, y ∈ K such that x < y. Show that there exists g ∈ G
such that x ≤ g ≤ y.

7.) From now on, we take K = R. Show that Gπ := Z + 2πZ = {n + 2πm | n, m ∈ Z} is

a subgroup in (R, +).

8.) Show that for all x, y ∈ [−1, 1] with x < y, there exists n ∈ N such that x ≤ cos n ≤ y.

1.A Construction of Number Systems (Exercises)

This entire section should be treated independently.
 1.A. CONSTRUCTION OF NUMBER SYSTEMS (EXERCISES) 57

Natural Numbers
Definition 1.A.1 (Peano’s Axioms). We denote by N and call it the set of natural numbers,
any set satisfies the following axioms.
P-0.) 0 ∈ N.

P-1.) For any x ∈ N, there exists a unique successor in N. The successor of x is denoted as
x′ .

P-2.) The mapping x 7→ x′ is injective.

P-4.) 0 is not a successor of any natural number.

P-5.) If a property is possessed by 0 and is possessed by the successor of every natural num-
ber that possess it, then the property is possessed by all natural numbers. (Induction
Axiom)

Remark 1.A.2. In this subject, we recognize that the natural numbers N start at 0, this
is not consistent to the common definition in English word. However in France, this is
considered as a more appropriate definition of N.

Remark 1.A.3. Under this definition, we denote by N∗ the set N \ {0}.

Exercise 1.A.4. Let x, y ∈ N. Prove the following assertions

1.) if x ̸= y then x′ ̸= y ′ ;

2.) x ̸= x′ ;

3.) if x ̸= 0 then there exists a unique u ∈ N such that x = u′ .

We define now the addition on nature numbers. The addition of two natural numbers
is a function which maps from N × N into N. We denote this operation temporary denote
by +N . Let x, y ∈ N, the value of this function is called the sum of these two numbers, and
we temporary denote x +N y.
Definition 1.A.5. The addition on N is a operation satisfying the following axioms. For
all x, y ∈ N,
Add-1.) x +N 0 = x,

Add-2.) x +N y ′ = (x +N y)′ .

Exercise 1.A.6. We define the symbols 1 = 0′ , 2 = 1′ , 3 = 2′ and 4 = 3′ . Using the

definition of addition on N, show that 2 +N 2 = 4.

Exercise 1.A.7. Let x, y ∈ N.

Show that if x +N y = 0, then x = y = 0.

In the following exercises, we want to justify the Definition 1.A.5.

 58 SYSTEM OF REAL NUMBERS

Exercise 1.A.8. Let Pa and Pb two operations satisfying both the axioms in the definition
of addition. That is, for all x, y ∈ N,

xPa 0 = x, xPa y ′ = (xPa y)′ , and xPb 0 = x, xPb y ′ = (xPb y)′ .

Let x ∈ N be fixed, we design by à the set of y ∈ N such that xPa y = xPb y. That is,

à = {y ∈ N xPa y = xPb y}.

Show, by induction, that à = N.

Exercise 1.A.9. In the previous exercise, we have shown that the addition +N is unique.
Let x ∈ N, we say +N is well-defined at x if for all y ∈ N, x +N y ∈ N.

1.) Show, by induction, that for all y ∈ N, 0 +N y = y. In other words, +N is well defined
at 0.
Assuming now for x ∈ N, +N is well-defined at x.

2.) Show, by induction, that for all y ∈ N, x′ +N y = (x +N y)′ .

3.) Conclusion.

Now we have justified the addition on N. From now on, we will denote by x +N 1 the
successor of x ∈ N. Consider the volume of this subject, there are important properties we
won’t treat here. We just state them as theorems. In fact, they can be derived using the
same induction arguments. We left the readers who are interested to justify those results
by themselves.

Theorem 1.A.10 (Associativity of +N ). Let x, y, z ∈ N. It holds,

(x +N y) +N z = x +N (y +N z).

Theorem 1.A.11 (Commutativity of +N ). Let x, y ∈ N. It holds,

x +N y = y +N x.

Theorem 1.A.12. Let x ∈ N∗ and y ∈ N. It holds,

y ̸= x +N y.

Theorem 1.A.13. Let x, y, z ∈ N. If x ̸= y, then

x +N z ̸= y +N z.

Theorem 1.A.14 (Law of Trichotomy). Let x, y ∈ N, then exactly one of the following
condition holds,

Tri-1.) x = y,
 1.A. CONSTRUCTION OF NUMBER SYSTEMS (EXERCISES) 59

Tri-2.) there exists a unique u ∈ N∗ such that x = y +N u,

Tri-3.) there exists a unique u ∈ N∗ such that y = x +N u.

The previous theorem provides a way to define an order on N.

Definition 1.A.15. We define the relations <N and >N as follows. Let x, y ∈ N, we say

• x >N y if there exists u ∈ N∗ such that x = y +N u;

• x <N y if there exists u ∈ N∗ such that y = x +N u.

Then we can define the order ≤N on N. We say x ≤N y if x <N y or x = y.

Definition 1.A.16. For x, y ∈ N, we define the operation −N as follows,

• x −N y = 0 if x = y;

• x −N y = u if x >N y where u is the unique number in Theorem 1.A.14.

• if x <N y, x −N y is not defined.

Exercise 1.A.17. Justify ≤N is a total order on N. (reflexivity, antisymmetric, transitivity

and totality)

Theorem 1.A.18. The following assertions are equivalent.

Induc-1.) If S ⊂ N possesses the property that

0 ∈ S, and if n ∈ S, then n +N 1 ∈ S,

then S = N. (Principle of Mathematical Induction)

Induc-2.) If S ⊂ N possesses the property that

∀n ∈ N, if {0, 1, · · · , n} ⊂ S, then n +N 1 ∈ S,

then S = N. (Principle of Complete Induction)

Induc-3.) Every nonempty subset of N has a smallest element. (Well-Ordering Principle)

Exercise 1.A.19. Show that the well-ordering principle holds in N. Then prove Theorem
1.A.18.

Definition 1.A.20. The multiplication in N is an operation satisfying the following axioms.

For x, y ∈ N,

1.) x ×N 0 = 0,

2.) x ×N (y +N 1) = x ×N y +N x.
60 SYSTEM OF REAL NUMBERS

The justification of this definition follows exactly the same induction that we did for
addition. This is also left who are interested to proceed by themselves.
Exercise 1.A.21. Show that the multiplication ×N is
• commutative,

• distributive to +N ,

• associative.

Exercise 1.A.22. Let x, y ∈ N.

Show that if x ×N y = 0, then x = 0 or y = 0.

Integers
Definition 1.A.23. We define the integers Z as the Cartesian product {+, −} × N with
the following axiom,
⋆ (+, 0) = (−, 0).
Let p ∈ Z, p = (ps , pn ) with ps ∈ {+, −} and pn ∈ N.
• We call the sign of p and denote it by sign(p) the symbol ps ∈ {+, −}.

• We call the absolute value of p and denote it by |p|Z the natural number pn .
From the axiom ⋆, we can omit the signs of (+, 0) or (−, 0), and we simply denote it by 0.
As a convention, we always note the signs in front of the absolute values and omit the
parentheses. For example, −3 ∈ Z. We say an integer p ∈ Z is positive if sign(p) = +, and
p to be negative if sign(p) = −.
As an immediate consequence of the definition of Z, we have the following property,
Proposition 1.A.24. 0 is the only integer which is both positive and negative.
Like the notation of N∗ we denote by Z∗ the set Z \ {0}.

Definition 1.A.25. We define the addition +Z on Z by the following axioms. Let p, q ∈ Z,

1.) if sign(p) = sign(q), then |p +Z q|Z = |p|Z +N |q|Z and sign(p +Z q) = sign(p) = sign(q);

2.) if sign(p) ̸= sign(q) and |p|Z ≤N |q|Z , then |p +Z q|Z = |q|Z −N |p|Z and sign(p +Z q) =
sign(q);

3.) if sign(p) ̸= sign(q) and |q|Z ≤N |p|Z , then |p +Z q|Z = |p|Z −N |q|Z and sign(p +Z q) =
sign(p);

Exercise 1.A.26. Show that (Z, +Z ) is an abelian group.

We denote from now on by −p the inverse element of p ∈ Z for addition +Z . We can
thus define the subtraction operation −Z as follows. Let p, q ∈ Z, p −Z q := p +Z (−q).
 1.A. CONSTRUCTION OF NUMBER SYSTEMS (EXERCISES) 61

Definition 1.A.27. We define the multiplication on Z as follows. Let p, q ∈ Z,

(
+(|p|Z ×N |q|Z ) if sign(p) = sign(q),
p ×Z q =
−(|p|Z ×N |q|Z ) if sign(p) ̸= sign(q).

Exercise 1.A.28. Justify that ×Z is associative, commutative and distributive to +Z then

deduce that (Z, +Z , ×Z ) is a ring.

Exercise 1.A.29. Let a, b, p ∈ Z, show that

1.) p ∈ Z, (−1) ×Z p = −p,

2.) (a +Z b) ×Z (a −Z b) = a ×Z a −Z b ×Z b.

Definition 1.A.30. We define the binary relation ≤Z as follows. Let p, q ∈ Z, we say

p ≤Z q if there exists a positive integer u such that q = p +Z u.

Exercise 1.A.31. Justify that ≤Z is a total order on Z.

Remark 1.A.32. We can thus define accordingly the relations <Z , >Z and ≥Z .

Remark 1.A.33. Now we have constructed the integer system Z. It is easy to see that there
exists a bijection between the natural numbers and positive integers. Moreover, we can
also justify that the operations +Z , ×Z and the order ≤Z all coincide with +N , ×N and ≤N
respectively under this bijection. So, we identify the natural numbers and positive integers
and the inclusion N ⊂ Z have a sense.

Rational Numbers To define the rational numbers, we need to introduce the concept of
equivalence relation and equivalence classes.
Definition 1.A.34. An equivalence relation is a binary relation which is reflexive, sym-
metric and transitive.
Precisely speaking, let S be a set and R be a binary relation on S. R is an equivalence
relation if for all x, y, z ∈ S,
EqR-1.) xRx,

EqR-2.) if xRy then yRx,

EqR-3.) if xRy and yRz then xRz.

Exercise 1.A.35. We define a binary relation ∼ on the Cartesian product Z×Z∗ as follows.
Let (p1 , q1 ), (p2 , q2 ) ∈ Z × Z∗ ,

(p1 , q1 ) ∼ (p2 , q2 ) if, and only if p1 ×Z q2 = p2 ×Z q1 .

Show that ∼ is an equivalence relation.

62 SYSTEM OF REAL NUMBERS

Definition 1.A.36. Let R be an equivalence relation on S, and x ∈ S. We call the

equivalence class of x respect to the relation R, and we denote it as [x]R , the following
subset,

[x]R := {y ∈ S | xRy} .

And any elements in [x]R is called as a representative of the class.

Exercise 1.A.37. Let x, y ∈ S and [x]R , [y]R be the equivalence classes which represent
respectively x and y. We assume that [x]R ∩ [y]R ̸= ∅. Show that xRy and deduce that
every distinct equivalence classes are disjoint from one to each others.

p
Definition 1.A.38. We call a fraction any couple (p, q) ∈ Z × Z∗ , and we denote it by .
q

Remark 1.A.39. From Exercise 1.A.35, we know that ∼ is an equivalence relation of frac-
tions.

p
Exercise 1.A.40. Let be a fraction satisfying gcd(p, q) = 1.
   q 
p p ×Z x ∗
Show that = x∈Z .
q ∼ q ×Z x

Definition 1.A.41. We call a rational number any equivalence class of fractions respect
to the equivalence relation ∼. We denote the set of rational numbers by Q.

Let r ∈ Q, we know that r is an equivalence class of fractions. We

 can therefore find
p p
p ∈ Z and q ∈ Z∗ such that is a representative of r, that is, r = .
q q ∼
Definition 1.A.42. Let r1 , r2 ∈ Q, we define the addition +Q on Q as follows. Let
p1 p2
the fractions and respectively represent the equivalent classes r1 and r2 . That is,
  q1 q2 
p1 p2
r1 = and r2 = . We then define,
q1 ∼ q2 ∼
 
p1 ×Z q2 +Z p2 ×Z q1
r1 +Q r2 = .
q1 ×Z q2 ∼

Exercise 1.A.43. In Definition 1.A.42, show that r1 +Q r2 is independent of the choices

of representatives.

We can proceed the similar calculus to show that (Q, +Q ) is an abelian group. The 
0
proof is also left who is interested to justify. The identity element in this group is ,
1 ∼
which we will denote simply as 0 (and we identify it with those “0” in Z and in N). Like
we introduced in N and in Z, we denote by Q∗ the set Q \ {0}.
 1.A. CONSTRUCTION OF NUMBER SYSTEMS (EXERCISES) 63

Exercise 1.A.44. Define the subtraction operation −Q on Q and show that (with the
notations in Definition 1.A.42)
 
p1 ×Z q2 −Z p2 ×Z q1
r1 −Q r2 = .
q1 ×Z q2 ∼

Definition 1.A.45. Let r1 , r2 ∈ Q, we define the multiplication ×Q on Q as follows.

  Let
p1 p2 p1
and respectively represent the equivalent classes r1 and r2 . That is, r1 = and
q1  q2 q1 ∼
p2
r2 = . We then define,
q2 ∼
 
p1 ×Z p2
r1 ×Q r2 = .
q1 ×Z q2 ∼

Exercise 1.A.46. Let r, r1 , r2 ∈ Q.

1.) Show that r1 ×Q r2 is independent to the choices of representatives.
 
−1
2.) Show that −r = ×Q r, where −r represents the inverse element of r for +Q .
1 ∼
3.) Show that (Q∗ , ×Q ) is an abelian group.

4.) Show that (Q, +Q , ×Q ) is a field.

Definition 1.A.47. Let r, r1 , r2 ∈ Q, we define the binary relation ≤Q on Q as follows.

p
• We say r ≤Q 0 if r possesses a representative fraction such that p ≤Z 0 and 0 <Z q.
q
• We say r1 ≤Q r2 if (r1 −Q r2 ) ≤Q 0.

Remark 1.A.48. We call a rational number r ∈ Q is negative if r ≤Q 0 and we call r

positive if −r ≤Q 0.

Exercise 1.A.49 (Law of Trichotomy in Q). Let r ∈ Q. Show that exactly one of the
following three assertions holds,
Tri-1.) r = 0,

Tri-2.) r ̸= 0 and r ≤Q 0,

Tri-3.) r ̸= 0 and 0 ≤Q r.

Exercise 1.A.50. Let r1 , r2 ∈ Q, we assume that r1 ≤Q 0 and r2 ≤Q 0.

Show that r1 +Q r2 ≤Q 0.

Exercise 1.A.51. Show that ≤Q is a total order on Q.

64 SYSTEM OF REAL NUMBERS

Remark 1.A.52. We can thus define accordingly the relations <Q , >Q , ≥Q as well as the
absolute value | · |Q ,
(
r if 0 ≤Q r,
|r|Q =
−r if r ≤Q 0.

Exercise 1.A.53. Show that (Q, +Q , ×Q , ≤Q ) is an ordered field.

Remark 1.A.54. Now, we have construct the rational number system Q. It is easy to seehthat
pi
there exists a bijection between the integers Z and the rational numbers of type r =
1 ∼
with p ∈ Z. Moreover, we can also justify that the operations +Q , ×Q and the order ≤Q all
coincide with +Z , ×Z and ≤Z respectively under this bijection. So, we identify the integers
and those rational umbers and the inclusion Z ⊂ Q have a sense.
Since we have already N ⊂ Z ⊂ Q, we will consider in the begging of next section
that all numbers are supposed to be in Q and the addition, subtraction, multiplication and
fractions are all considered in the Q-sense.

Real Numbers There are many equivalent ways to construct the real number systems.
The mainstream methods are Dedekind Cuts and Cauchy sequences. In this subject, we
choose the approach of using Cauchy sequences. This method of construction, originated
from Cantor, can also be adapted in the completion of any metric spaces.
Before we start, make sure you all understand the following notions, ordered field,
absolute values, sequences, convergence, Cauchy sequences.
Let (Q, +Q , ×Q , ≤Q ) be the ordered field of rational numbers we have constructed in the
previous section. We recall the following notations,

• For all r ∈ Q, we denote by |r|Q the absolute value of r, we remark that |r|Q ≥Q 0.
Q
• Let (xn )n∈N be a sequence converging in Q, we denote its limit by lim xn .
n→+∞

Definition 1.A.55. Let (xn )n∈N , (yn )n∈N be sequences of elements in Q, we define the
binary relation ≈ on QN as follows.
Q
(xn )n∈N ≈ (yn )n∈N if, and only if lim |xn −Q yn |Q = 0.
n→+∞

Exercise 1.A.56. Show that ≈ is an equivalence relation on QN .

Exercise 1.A.57. Let (xn )n∈N , (yn )n∈N be sequences of elements in Q, we assume that
(xn )n∈N is a Cauchy sequence in Q and (yn )n∈N ≈ (xn )n∈N .
Show that (yn )n∈N is also a Cauchy sequence in Q.
1.A. CONSTRUCTION OF NUMBER SYSTEMS (EXERCISES) 65

Definition 1.A.58. We define a real number x ∈ R any equivalence class of Cauchy

sequences in Q. That is,

R = {[(xn )n∈N ]≈ | (xn )n∈N is Cauchy in Q} .

Exercise 1.A.59. Let (xn )n∈N , (yn )n∈N be Cauchy sequences of elements in Q. Show that

1.) (xn +Q yn )n∈N is a Cauchy sequence in Q.

2.) (xn ×Q yn )n∈N is a Cauchy sequence in Q.

We define now the addition and multiplication on R.

Definition 1.A.60. Let x, y ∈ R and (xn )n∈N , (yn )n∈N be representative Cauchy sequences
in Q to x, y respectively. We define then,

• x + y = [(xn +Q yn )n∈N ]≈ ,

• x · y = [(xn ×Q yn )n∈N ]≈ .

Exercise 1.A.61. Justify the above definitions, i.e. they don’t depends on the choice of
representatives.

Exercise 1.A.62. Show that (R, +, ·) is a field. What types of sequences represent the
identity elements 0 and 1? Furthermore, what type of sequences represent the rational
numbers?

In the following exercises, we want to establish the law of trichotomy in R

Exercise 1.A.63. Let (xn )n∈N be a Cauchy sequence in Q. Show that exactly one of the
following three assertion holds,
Q
1.) lim xn = 0.
n→+∞

2.) there exists δ ∈ Q, δ >Q 0 and N ∈ N such that for all n ∈ N, n ≥N N , we have
xn ≥Q δ

3.) there exists δ ∈ Q, δ >Q 0 and N ∈ N such that for all n ∈ N, n ≥N N , we have
xn ≤Q −δ.

Exercise 1.A.64. Let (xn )n∈N and (yn )n∈N be Cauchy sequences in Q such that (xn )n∈N ≈
(yn )n∈N . Show that (xn )n∈N and (yn )n∈N have the same classification in the previous exer-
cise. Moreover, in the second and the third case, we can choose for (yn )n∈N the same δ as
for (xn )n∈N .

Definition 1.A.65. Let x, y ∈ R.

66 SYSTEM OF REAL NUMBERS

• We say x is strictly positive if there exists δ ∈ Q, δ >Q 0 such that for all repre-
sentative (xn )n∈N of x, there exists N ∈ N such that for all n ∈ N, n ≥N N , we have
xn ≥Q δ. We denote this relation by 0 < x.

• We say x is positive if 0 < x or 0 = x and we denote this relation by 0 ≤ x.

• We say x ≤ y if 0 ≤ (y − x).

Exercise 1.A.66. Justify ≤ is a total order on R.

Remark 1.A.67. We can now define all other comparison relations <, > and ≥.

Exercise 1.A.68. Justify (R, +, ·, ≤) is an ordered field.

Exercise 1.A.69. Show that (R, +, ·, ≤) is Archimedean.

Exercise 1.A.70. Let x, y ∈ R, 0 ≤ x < y.

1.) Show that there exists r ∈ Q such that 0 < r < y − x.

2.) Consider the successive numbers nr for n ∈ N. Show that there exists m ∈ N such
that x < mr and (m − 1)r ≤ x.

3.) Show that for this m obtained in the previous point, it satisfies x < mr < y.

4.) Deduce that Q is dense in R.

The following exercise is a reformulation of the denseness using the representative

Cauchy sequences.

Exercise 1.A.71. Let (xn )n∈N , (yn )n∈N be Cauchy sequences of elements in Q. We assume
Q
that lim (xn − yn ) ̸= 0.
n→+∞
Show that there exists r ∈ Q and N ∈ N such that for n ∈ N, n ≥N N we have either
xn <Q r <Q yn or yn <Q r <Q xn .

As we have defined the total order ≤, we can thus define the corresponding magnitude
on R. Let x ∈ R,
(
x if 0 ≤ x,
|x| =
−x if x ≤ 0.

Exercise 1.A.72. Let x ∈ R and (xn )n∈N be a representative Cauchy sequence in Q.

Show that (|xn |Q )n∈N is a representative of |x|.

We then start to consider the sequences and the convergence in R. In what follows, if
we didn’t precisely mention an object is in Q or mark the index like ≤Q , that means it is
considered in R.
1.A. CONSTRUCTION OF NUMBER SYSTEMS (EXERCISES) 67

Definition 1.A.73. Let (x(k) )k∈N be a sequence of elements in R, we say lim x(k) = y ∈ R
k→+∞
if, and only if

∀ε > 0, ∃K ∈ N s.t. ∀k ∈ N, k ≥N K ⇒ |x(k) − y| ≤ ε.

Exercise 1.A.74. Show that Definition 1.A.73 can be reformulate as follows.

For all εr ∈ Q, εr >Q 0, there exists K ∈ N such that for all k ∈ N, k ≥N K and for
(k)
any representatives (xn )n∈N and (yn )n∈N respectively of the real numbers x(k) and y, there
(k)
exists N ∈ N such that for all n ∈ N, n ≥N N , we have |xn − yn |Q ≤Q εr .

Let y ∈ R. We consider a representative (rn )n∈N ∈ QN of y. By definition, (rn )n∈N is a

Cauchy sequence in Q. For each n ∈ N, we consider rn as a real number and we introduce
(ren (k) )k∈N ∈ QN a representative of rn .

Exercise 1.A.75. Show that rn converge to y in R.

Hint: we can take for example (ren (k) )k∈N to be constant sequences (rn )k∈N .

Exercise 1.A.76. Let (xn )n∈N be a Cauchy sequence in R.

1.) Using the denseness result, show that there exists a sequence (rn )n∈N of elements in
Q such that |xn − rn | ≤ δn for all n ∈ N. (Here δn is a positive rational number which
we are able to set its value.)

2.) Show that (xn )n∈N converges in R.

3.) Final Conclusion.

Congratulations!
68 SYSTEM OF REAL NUMBERS
 Chapter 2

Numerical Series

In this chapter, we denote by K the field R or C. For study the nature of convergence or di-
vergence to a series, we need use some knowledge on the “speed of convergence/divergence”,
which should already introduced in the lectures of the first year. To describe rigorously this
concept, we introduce the comparison relations between sequences or functions by using
the symbols big O (dominance), small o (negligence) and ∼ (equivalence). Nevertheless,
we also introduce the asymptotic expansions of functions, which enlarges the notions of
Taylor’s expansions. In the propose of being self-content in this lecture note, we included
them in Appendices A and B.

2.1 General Definitions

Let (xn )n∈N be a sequence of elements in K.

Definition 2.1.1. • We call the sequence of partial sums of the sequence (xn )n∈N
the such defined sequence (Sn )n∈N ∈ KN :
n
X
∀n ∈ N, Sn = xk .
k=0

X
• We call the series of general terms xn and we denote by xn the sequence of
n
partial sums (Sn )n∈N .
X
• We say the series xn converges if (Sn )n∈N is convergent.
n
X
• If the series xn converges, we call the sum of the series the limit of (Sn )n∈N and
n
we denote
+∞
X n
X
xn = lim Sn = lim xk .
n→+∞ n→+∞
n=0 k=0

X
• In the case of (Sn )n∈N doesn’t converge, we say the series xn diverges.
n

69
70 NUMERICAL SERIES
X
Remark 2.1.2. • Any sequence (sn )n∈N ∈ RN can be seen as a series un , we can set
n
u0 = s0 and un = sn − sn−1 for n ∈ N∗ .
X
• In some cases of necessary, we denote xn for the series which start at the index
n≥n0
Xn
n0 , its partial sums becomes then, Sn = xk .
k=n0

• Many results in this chapter could be applicable in the further studies of the series of
elements in more abstract spaces, like Banach spaces.

X
Example 2.1.3. A geometric series an with a ∈ R or C is convergent if, and only if
n
|a| < 1. Moreover, in the case of convergence, the sum of the geometric series is given by
+∞
X 1
an = .
1−a
n=0

Solution. Left in Exercises.

X X
Definition 2.1.4. • Let α, β ∈ K, the linear combination α xn + β yn of two
X X X n n
series xn and yn is by definition the series (αxn + βyn ).
n n n
X
• In the case of complexes, the conjugate of a series zn is by definition the series
X n
zn .
n

X X X
Proposition 2.1.5. Let α, β ∈ K and let xn , yn , zn be convergent series. Then,
n n n

X X +∞
X +∞
X +∞
X
• the series α xn + β yn converges, and (αxn + βyn ) = α xn + β yn ;
n n n=0 n=0 n=0

X +∞
X +∞
X
• the series zn converges and zn = zn .
n n=0 n=0

Proof. Left in exercise.

Remark 2.1.6. The set of convergent series forms a K-vector space.

Proposition 2.1.7. For any convergent series, the general terms converge to 0.
2.1. GENERAL DEFINITIONS 71
X
Proof. Let xn be a convergent series. It follows from the definition, the sequence of
n
n
X
partial sums (Sn )n∈N converges to l ∈ K where Sn = xk . Then, the sequence (Sn−1 )n∈N∗
k=0
converges also to l as a subsequence of (Sn )n∈N . In the consequence, for all n ∈ N∗ ,
xn = Sn − Sn−1 . And it leads to lim xn = lim Sn − Sn−1 = l − l = 0.
n→+∞ n→+∞

Remark 2.1.8. The previous propositionXcan also be equivalently reformulate as follows. If

(xn )n∈N does not converge to 0, then xn diverges.
n

X +∞
X
Definition 2.1.9. Let xn be a convergent series and let S = xn . For all n ∈ N, the
n n=0
X +∞
X
series xk is also convergent. We call the sum Rn = xk to be the rest of order
k≥n+1 k=n+1
X
n of the series xn . And we have: Rn = S − Sn .
n

2.1.1 Telescoping Series and Partial Sums

X
Definition 2.1.10. We call xn as a telescoping series if the general terms xn can be
n
written as the difference Sn − Sn−1 .

In fact, the above definition is nothing else to a reformulation of the sequence of partial
sums. They provide the ideasXin analyzing the convergences of sequences or series. Precisely
speaking, to study a series xn , one can try to determinate its sequence of partial sums.
n

 
X 1 π
Example 2.1.11. The series arctan is convergent with the sum .
n
n2 + n + 1 2

Solution. Left in Exercise 2.4.7.

We can also proceed reciprocally. To study theXconvergence of a sequence (Sn )n∈N , we

transform this problem on the telescoping series xn with xn = Sn − Sn−1 .
n

n
!
X 1
Example 2.1.12. The sequence − ln n converges. The limit of this sequence
k
k=1 n∈N∗
is called as Euler’s constant and denoted by γ.

Solution. Left in Exercise 2.4.8.

72 NUMERICAL SERIES

2.2 Series with Positive General Terms

2.2.1 Generalities and Comparison Tests
X
Proposition 2.2.1. Let xn be a series with positive general terms. Then, the series
X n
xn converges if, and only if the sequence of partial sums (Sn )n∈N is bounded from above.
n
+∞
X
Moreover in this case, xn = lim Sn = sup Sn .
n→+∞ n∈N
n=0
Proof. Direct consequence of the monotone convergence property.

X X
Proposition 2.2.2 (Comparison test). Let xn and yn be two series with positive
n n
general terms such that for all n ∈ N, 0 ≤ xn ≤ yn .
X X +∞
X +∞
X
• If yn converges, then xn converges and 0 ≤ xn ≤ yn .
n n n=0 n=0
X X
• If xn diverges, then yn diverges.
n n
X
Proof. • We assume that yn converges. We design by (Sn )n∈N the sequence of partial
X n
sums to the series xn . Since the sequence (yn )n∈N is composed with positive
n
numbers, we have for all N ∈ N,
N
X N
X +∞
X
xn ≤ yn ≤ yn .
n=0 n=0 n=0
+∞
X
It shows that the sequence (Sn )n∈N is bounded from above and yn is an upper
n=0
X
bound of this sequence. From the previous proposition, the series xn converges
n
and
+∞
X +∞
X
0≤ xn = sup Sn ≤ yn .
n=0 n∈N n=0
X
• We assume that xn diverges. Then, for all M > 0, there exists N ∈ N such that
n
N
X
xn ≥ M , and it leads to
n=0
N
X N
X
M≤ xn ≤ yn .
n=0 n=0
X
Thus, the series yn diverges.
n
2.2. SERIES WITH POSITIVE GENERAL TERMS 73

Remark 2.2.3. This proposition remains true if we consider only the series started from
certain rank n0 ∈ N.
X X
Corollary 2.2.4 (Usual comparison test). Let xn and yn be two series with positive
n n
general terms.
X X
• If xn = O(yn ) when n → +∞ and yn converges, then xn converges.
n n
X X
• Assuming that xn ∼ yn when n → +∞, then xn converges if, and only if yn
n n
converges.
X 1
Example 2.2.5. 1.) The series converges.
n2
n≥1
X 1
2.) The series √ diverges.
n
n≥1
1 1 1 1
Solution. 1.) For all n ∈ N, n ≥ 2, we have 2 ≤ = − . Moreover,
 n n(n − 1) n−1 n
X 1 1
the telescoping series − converges since its sequence of partial sums
n−1 n
n≥2
is given by, ∀n ∈ N, n ≥ 2,
n  
X 1 1 1
Sn = − =1− −→ 1.
k−1 k n n→+∞
k=2
X 1
From the comparison test, the series converges.
n2
n≥1

1 2 √ √
2.) For all n ∈ N∗ , we have √ ≥ √ √ = 2( n + 1 − n). On the other hand,
X n√ n+ n+1
√
the telescoping series 2( n + 1 − n) diverge since its sequence of partial sums is
n
given by, ∀n ∈ N∗ ,
n √
X √ √
Sn = 2( k + 1 − k) = 2( n + 1 − 1) −→ +∞.
n→+∞
k=1
X 1
From the comparison test, the series √ diverges.
n
n≥1

Corollary 2.2.6 (Riemanni series). Let α ∈ R.

X 1
The series converges if, and only if α > 1.
nα
n≥1
i
Bernhard Riemann (1826 - 1866), German mathematician. He contributed fundamental pioneer re-
searches on various branches of mathematics. In analysis, “Riemann integral” provided a first rigorous
integral theory. Also, “Riemann hypothesis” is considered as the most famous unsolved conjenture in math-
ematics and plays a fundamental role in analytic number theory. In differential geometry, “Riemann surfaces”
provides a mathematical interpretation of general relativity.
74 NUMERICAL SERIES

Proof. Let α ∈ R, α ̸= 1. For all n ∈ N, n ≥ 2 , we have

!
1 α−1
 
1 1 1
− = 1− 1− .
(n − 1)α−1 nα−1 (n − 1)α−1 n

Using the usual equivalence relation (1 + un )p − 1 ∼ pun with un −→ 0, we have, by

n→+∞
1
taking un = − and p = α − 1,
n

1 α−1 α − 1
 
1− 1− ∼ .
n n

1 1 α−1 X 1
And hence, − ∼ . From the comparison test, the series
(n − 1)α−1 nα−1 nα nα
  n≥1
X 1 1
converges if, and only if the telescoping series α−1
− α−1 converges. It is
(n − 1) n
n≥2
immediate by investigating the sequences of partial sums to obtain that the latter series
converges if, and only if α > 1.
X1
The case α = 1 is corresponding to the harmonic series , which is a well-known
n
n≥1
divergent series. To prove this point we can use the equivalence relation while n → +∞,
 
1 1
∼ ln 1 + ∼ ln(n + 1) − ln n.
n n
X
Since the telescoping series (ln(n + 1) − ln n) diverges, it follows from the comparison
n≥1
X1
test, diverges.
n
n≥1

X
Corollary 2.2.7 (Riemann test). Let xn be a series with positive general terms.
n
 
1 X
• If there exists α > 1 such that xn = O , then xn converges.
nα n

β X
• If there exists α ≤ 1 and β > 0 such that xn ≥ α
from certain rank, then xn
n n
diverges.

X 1
Example 2.2.8 (Bertrandii series). The series converges if, and only if α > 1
nα (ln n)β
n≥2
or α = 1 < β.
ii
Joseph Bertrand (1822 - 1900), French mathematician and economist. He is known for Bertrand paradox
in probability.
2.2. SERIES WITH POSITIVE GENERAL TERMS 75

1+α
Solution. We assume at first α ̸= 1. We set γ = , and for all n ∈ N, n ≥ 2,
2
1 1 n(1−α)/2 1
α β
= γ β
:= γ un .
n (ln n) n (ln n) n
We discuss the following cases.
• Case α > 1.
– If β ≥ 0, we have either (ln n)β = 1, ∀n ∈ N, n ≥ 2 or (ln n)β −→ +∞. In
n→+∞
both cases, we have (ln n)β ≥ 1 from certain rank. Then, un ≤ n(1−α)/2 −→ 0.
  n→+∞
1 1
We have in particular un = O(1) and hence, α β
=O . From the
n (ln n) nγ
X 1
comparison test, the series converges.
n (ln n)β
α
n≥2
– If β < 0, then un = n (1−α)/2 (ln n)|β| .
We use the usual comparison relation
|β| ε
(ln n) = o(n ) for all ε > 0. Then it holds
un = n(1−α)/2 (ln n)|β| = o(n(1−α)/2+ε ).
The right-hand-side remains a bounded sequence if we choose ε small enough,
 for

α−1 1 1
example, we can choose ε = . In the consequences, α =O .
4 n (ln n)β nγ
X 1
From the Riemann test, the series converges.
n (ln n)β
α
n≥2

• Case α < 1. We have, for n ∈ N, n ≥ 3,

n(1−α)/2
un = ≥ n(1−α)/2 (ln n)−|β| .
(ln n)β
Let ε > 0, it follows from the usual comparison relation (ln n)|β| = o(nε ), we deduce
that (ln n)−|β| ≥ n−ε from certain rank. Hence, un ≥ n(1−α)/2−ε from certain rank.
1−α
It is sufficient to choose ε small enough, for example ε = , to obtain that
4
un −→ +∞.
n→+∞
1 1
In the consequence, ≥ γ from certain rank. We remark that γ < 1 in this
nα (ln n)β n
X 1
case. From the Riemann test Corollary 2.2.7, the series diverges.
n
n (ln n)β
α

1 1
• Case α = 1. If β ≤ 0, then β
≥ for all n ∈ N, n ≥ 3. Then the series
n(ln n) n
X 1
diverges from the Riemann test Corollary 2.2.7. On the other hand, if
n(ln n)β
n≥2
1
β > 0, we consider the function f : t 7→ , which is positive and decreasing on
t(ln t)β
[2, +∞). Moreover, the indeterminate integrals of f are given as follows
 1−β
Z t  t + C if β ̸= 1,
∀t > 2, f (s)ds = 1−β
2 
ln(ln t) + C if β = 1.
76 NUMERICAL SERIES

We can observe that if β > 0 then the indeterminate integral of f is bounded on

[2, ∞), which means that f is integrable on this interval, whereas f is not integrable
in the case β = 1.
X 1
We finally conclude with the integral test Theorem 2.3.12, that the series
nα (ln n)β
n≥2
converges if, and only if β > 1.

2.2.2 Other Test Methods and Asymptotic Behaviors

X
Proposition 2.2.9 (Root test). Let xn be a series with positive general terms.
n

√ X
• If there exists a subsequence (xσn )n∈N such that σn xσn ≥ 1 for all n ∈ N, then xn
n
diverges.
√ X
• If there exists α < 1 such that n xn ≤ α from certain rank, then xn converges.
n

Proof. • The divergence is straightforward since (xn )n∈N does not convergent to 0.
X
• We have in this case xn ≤ αn from certain rank. Then the convergence of xn
X n
follows from the convergence of the geometric series αn and the comparison test.
n

Remark 2.2.10. The conditions in the root test can also be formulated using the limit
superior.
√ X
• If lim sup n xn ≥ α > 1, then xn diverges.
n→+∞ n

√ X
• If lim sup n
xn ≤ α < 1, then xn converges.
n→+∞ n

X X
Proposition 2.2.11 (d’Alembertiii test; Ratio test). Let xn and yn be two series
n n
xn+1 yn+1
with strictly positive general terms. We assume that from certain rank, ≤ .
xn yn
X X
• If yn converges, then xn converges.
n n
X X
• If xn diverges, then yn diverges.
n n
iii
Jean Le Rond d’Alambert (1717 - 1783), French mathematician, physicist and philosopher. Co-editor
of l’Encrylopédie with Diderot.
2.2. SERIES WITH POSITIVE GENERAL TERMS 77

xn+1 yn+1
Proof. The condition ≤ from certain rank can be reformulate as follows. There
xn yn
xk+1 yk+1
exists N ∈ N such that for all k ∈ N, k ≥ N , ≤ . Then, for n ∈ N, n ≥ N , we
xk yk
have
n−1 n−1
Y xk+1 Y yk+1 xN
xn = xN ≤ xN = yn := Cyn ,
xk yk yN
k=N k=N

xN
where C = > 0.
yN
It follows therefore from the comparison test, the two claims on the convergence and
the divergence hold.

X
Corollary 2.2.12 (Usual d’Alembert test; Ratio test). Let xn be a series with strictly
n
positive general terms.
xn+1 X
• If ≥ 1 from certain rank, then xn diverges.
xn n
xn+1 X
• If there exists α < 1 such that ≤ α from certain rank, then xn converges.
xn n
xn+1
Proof. • The condition ≥ 1 from certain rank can be reformulate as follows. There
xn
xk+1
exists N ∈ N such that ∀k ∈ N, k ≥ N , ≥ 1. Hence, for all n ∈ N, n ≥ N ,
xk
n−1
Y xk+1
xn = xN ≥ xN > 0.
xk
k=N
X
This shows that the sequence (xn )n∈N does not converge to 0. Then the series xn
n
diverges.

• We can define the sequence (yn )n∈N := (αn )n∈N , and the claim follows via Proposi-
tion 2.2.11.

Remark 2.2.13. We can also reformulate the ratio test using the limit superior and inferior.
Let (xn )n∈N be a sequence of strictly positive general terms.
xn+1 X
• If lim inf > 1, then xn diverges.
n→+∞ xn
n
xn+1 X
• If lim sup < 1, then xn converges.
n→+∞ xn n
xn+1 X
• If lim = 1, then the convergence of xn remains undetermined.
n→+∞ xn
n
78 NUMERICAL SERIES
X X
Theorem 2.2.14. Let xn and yn be two series with positive general terms. We
X n n
assume that yn diverges.
n

1.) If xn = O(yn ), then

n n
!
X X
xk = O yk .
k=0 k=0

2.) If xn = o(yn ), then

n n
!
X X
xk = o yk .
k=0 k=0

3.) If xn ∼ yn , then
n
X n
X
xk ∼ yk .
k=0 k=0

Proof. Introducing the sequences of partial sums respectively to (xn )n∈N and (yn )n∈N . For
Xn Xn
all n ∈ N we set Xn = xk and Yn = yk .
k=0 k=0

1.) The relation xn = O(yn ) implies that thereX exists A > 0 and n0 ∈ N such that
xn ≤ Ayn holds for all n ∈ N, n ≥ n0 . Since yn diverges, there exists also n1 ∈ N
n
such that Xn0 ≤ AYn for all n ∈ N, n ≥ n1 . Then, for all n ∈ N, n ≥ max{n0 , n1 },
we have
n
X n
X
Xn = Xn0 + xk ≤ AYn + Ayk ≤ 2AYn .
k=n0 +1 k=n0 +1

Hence the claim follows.

) implies that there exists n0 ∈ N such that xn ≤ εyn

2.) Let ε > 0, the relation xn = o(ynX
for all n ∈ N, n ≥ n0 . Since yn diverges, there exists also n1 ∈ N such that
n
Xn0 ≤ εYn for all n ∈ N, n ≥ n1 . Then, for all n ∈ N, n ≥ max{n0 , n1 } + 1, we have
n
X n
X
Xn = Xn0 + xk ≤ εYn + εyk ≤ 2εYn .
k=n0 +1 k=n0 +1

Hence the claim follows.

3.) The relation xn ∼ yn holds if, and only if xn − yn = o(yn ). It is sufficient to apply the
previous result to obtain the claim on the equivalence relation.
2.3. SERIES WITH COMPLEX GENERAL TERMS 79
X X
Theorem 2.2.15. Let xn and yn be two series with positive general terms. We
X n n
assume that yn converges.
n

1.) If xn = O(yn ), then

+∞ +∞
!
X X
xk = O yk .
k=n+1 k=n+1

2.) If xn = o(yn ), then

+∞ +∞
!
X X
xk = o yk .
k=n+1 k=n+1

3.) If xn ∼ yn , then

+∞
X +∞
X
xk ∼ yk .
k=n+1 k=n+1

X
Proof. In each of the following cases, since the series yn converges, it follows from the
X n
comparison test, the series xn converges and the sequences of the resets converges to 0.
n
So it is licit to introduce the sequences of rests respectively to (xn )n∈N and (yn )n∈N . For all
+∞
X +∞
X
n ∈ N we set X̃n = xk and Ỹn = yk .
k=n+1 k=n+1

1.) If xn = O(yn ), there exists A > 0 and n0 ∈ N such that xn ≤ Ayn for all n ∈ N,
n ≥ n0 . Then, for all n ≥ n0 , we have X̃n ≤ AỸn . Hence the claim follows.

2.) Let ε > 0, the relation xn = o(yn ) implies that there exists n0 ∈ N such that xn ≤ εyn
for all n ∈ N, n ≥ n0 . Then, for all n ≥ n0 , we have X̃n ≤ εỸn . Hence the claim
follows.

3.) The relation xn ∼ yn holds if, and only if xn − yn = o(yn ). It is sufficient to apply the
previous result to obtain the claim on the equivalence relation.

2.3 Series with Complex General Terms

X
In this section, we consider a series xn with general terms xn ∈ C. We denote the
n
sequence of partial sums by (Sn )n∈N ∈ CN .
80 NUMERICAL SERIES

2.3.1 Convergences and Absolute Convergences

X
Theorem 2.3.1. The series xn converges if, and only if, the sequence of partial sums
n
(Sn )n∈N is a Cauchy sequence. That is,

n+p
X
∀ε > 0, ∃N ∈ N s.t. ∀n, p ∈ N, n ≥ N ⇒ |Sn+p − Sn | = xk ≤ ε.
k=n+1

Proof. Direct consequence of the completeness of C.

X X
Definition 2.3.2. We say xn is absolutely convergent if the series |xn | of positive
n n
general terms is convergent.

X
Theorem 2.3.3. The series xn converges if it is absolutely convergent. Moreover,
n

+∞
X +∞
X
xn ≤ |xn |.
n=0 n=0

X
Proof. We denote by S̃n the sequence of partial sums of the series |xn |. Since the series
X X n
xn is absolutely convergent, the series |xn | converges, which implies the sequence S̃n
n n
is a Cauchy sequence as a convergent sequence.
Let ε > 0, there exists N ∈ N such that for all n, p ∈ N, n ≥ N we have |S̃n+p − S̃n | ≤ ε.
Applying the triangular inequality, we have
n+p
X n+p
X
|Sn+p − Sn | = xk ≤ |xk | = S̃n+p − S̃n ≤ ε.
k=n+1 k=n+1

X
This shows that the sequence (Sn )n∈N is a Cauchy sequence, and hence the series xn
n
converges.
It follows also from the triangular inequality, we have, for all n ∈ N, |Sn | ≤ S̃n . Since
those two sequences are both convergent, we have, by letting n → +∞,
+∞
X +∞
X
xn ≤ |xn |.
n=0 n=0

X 1
Example 2.3.4. Let z ∈ C with Re(z) > 1. Then the series converges.
nz
n≥1
2.3. SERIES WITH COMPLEX GENERAL TERMS 81

Solution. Let n ∈ N∗ , it follows that

1 1
= n−z = e− ln n(Re(z)+iIm(z)) = e− ln nRe(z) = Re(z) .
nz n
X 1
Using Re(z) > 1 and Riemann series Corollary 2.2.7, the series converges. So,
n≥1
nRe(z)
X 1
the series is absolutely convergent hence it is convergent by Theorem 2.3.3.
nz
n≥1

Remark 2.3.5. We denote for all z ∈ C with Re(z) > 1,

+∞
X 1
ζ(z) = .
nz
n=1

This function ζ is called as Riemann’s Zeta Function. The famous Riemann’s hy-
pothesis is subject to study the behaviors of this function.

Lemma 2.3.6. The set of absolutely convergent series forms a subspace of the C-vector
space of convergent series.
X X
Proof. Take two absolutely convergent series xn and yn , and λ ∈ C, we remark that
n n
for all n ∈ N, |xn + λyn | ≤ |xn | + |λ||yn |. Then,
n
X n
X n
X n
X +∞
X +∞
X
|xk + λyk | ≤ (|xn | + |λ||yn |) = |xk | + |λ| |yk | ≤ |xn | + |λ| |yn |.
k=0 k=0 k=0 k=0 n=0 n=0

This inequality
X shows that the sequence of partial sums of the series with positive general
terms |xn + λyn | is bounded from above. Thus, it is convergent and therefore the series
X n
(xn + λyn ) is absolutely convergent.
n
Then, the set of absolutely convergent series is stable by linear combination. We also
knows that all absolutely convergent series are convergent series and the set of convergent
series forms a C-vector space via Remark 2.1.6. Hence, the claim follows.

X
Proposition 2.3.7. • Let
yn be a convergent series with positive general terms such
n X
that xn = O(yn ) while n → +∞, then the series xn absolutely converges.
n
X
• If yn is an absolutely convergent series with complex general terms such that xn ∼
n X
yn while n → +∞, then the series xn converges.
n

Proof. Left in Exercises.

82 NUMERICAL SERIES

2.3.2 Permutation of Terms

The following result states the possibility of adding the series in another order. Be careful,
you can do this only on absolutely convergent series.
Theorem 2.3.8 (Permutation of XTerms). Let σ be a bijection mapping from NX into itself
(i.e., σ is a permutation). If xn is absolutely convergent, then the series xσ(n) is
n n
+∞
X +∞
X
also absolutely convergent. Moreover, they have the same sum, xn = xσ(n) .
n=0 n=0

Proof. Let n ∈ N, we introduce M (n) := max{σ(0), σ(1), · · · , σ(n)}. It is immediate that

M (n) ≥ n and {σ(0), σ(1), · · · , σ(n)} ⊂ [[0, M (n)]]. Then, it holds that

n M (n) +∞
X X X
|xσ(k) | ≤ |xk | ≤ |xk |. (2.1)
k=0 k=0 k=0
X
It shows that the sequence of partial sums of the series |xσ(n) | is bounded from above.
X n
Then, it follows from the monotone convergence property that |xσ(n) | converges, which
X n
is equivalent to say xσ(n) is absolutely convergent.
n X
In addition, we remark that the series |xn | has positive general terms, then its sum
n
S equals to the supremum of the sequence of partial sums. That is,
+∞
( n )
X X
S= |xn | = sup |xk | n ∈ N .
n=0 k=0

Let ε > 0, it follows from the characterization of supremum, there exists m ∈ N such that
m
X
S−ε≤ |xk | ≤ S.
k=0

Since σ is bijective, we can set N (m) := max{σ −1 (0), σ −1 (1), · · · , σ −1 (m)} ∈ N. It holds
also that N (m) ≥ m and {σ −1 (0), σ −1 (1), · · · , σ −1 (m)} ⊂ [[0, N (m)]]. So, by using (2.1)
with n = N (m), we have,

m N (m)
X X
S−ε≤ |xσ(σ−1 (k)) | ≤ |xσ(k) | ≤ S.
k=0 k=0

Then, we apply again the characterization of supremum to deduce that

+∞
X +∞
X
|xσ(n) | = S = |xn |. (2.2)
n=0 n=0

Now, consider n ∈ N. We introduce the complementary set

An := [[0, M (n)]] \ {σ(0), σ(1), · · · , σ(n)}.

2.3. SERIES WITH COMPLEX GENERAL TERMS 83

And it holds that

M (n) n M (n) n
X X X X X X
xk − xσ(k) = xk ≤ |xk | = |xk | − |xσ(k) |.
k=0 k=0 k∈An k∈An k=0 k=0

Passing n → +∞, it follows from (2.2) that

M (n) n +∞ +∞
X X X X
|xk | − |xσ(k) | −→ |xk | − |xσ(k) | = S − S = 0.
n→+∞
k=0 k=0 k=0 k=0

In conclusion, we have also that

+∞
X +∞
X
xk = xσ(k) .
k=0 k=0

We should emphasis that the above result holds only for absolutely convergent series.
Without this assumption, we can not in general change the order of the summation.

1 1 1 1
Example 2.3.9. Consider the general terms xn = 1, −1, , − , · · · , , − , · · · . That is,
2 2 n n
1 1 ∗
x2n−1 = and x2n = − for all n ∈ N . Denoting by (Sn )n∈N the sequence of partial sums
n X n
to the series xn . We have then,
n≥1

1
∀n ∈ N∗ , S2n = 0, S2n−1 = .
n
X
We can deduce that Sn −→ 0. So, the series xn converges.
n→+∞
n≥1
Now, we change the general terms in the following order

1 1 1 1 1 1 1 1 1
xσ(n) =1, , −1, , , − , , , , , − , · · ·
2 3 4 2 5 6 7 8 3
1 1 1 1
· · · , p−1 , p−1 ,··· , p,− ,··· .
2 +1 2 +2 2 p

For p ∈ N∗ , we have in particular,

p−1
2X
1 1
≥ .
2p−1 +k 2
k=1

X
It show that sequence of partial sums of the series xσ(n) is not a Cauchy sequence, which
n
implies the divergence of the series.
84 NUMERICAL SERIES

2.3.3 Comparison with Integrals

In the rest part of this section, we present the relation between a series and an integral over
an interval. For this end, we introduce the concept of integrability of a function. In fact,
we will develop a complete theory on integrals next semester.

Definition 2.3.10. Let a ∈ R, and f : [a, +∞) → C be a piecewisely continuousiv function.

• Let A ∈ (a, +∞), we call the integral of f over the segment [a, A], which is denoted
Z Z A
by f= f (t)dt, the limit of the Riemann sumsv ,
[a,A] a

A n−1  
A−aX (A − a)
Z
f (t)dt = lim f a+ k .
a n→+∞ n n
k=0

• In the case where f is a positive function (i.e., ∀t ∈ [a, +∞), f (t) ≥ 0), we say the
Z A 
function f is integrable on the interval [a, +∞) if the set f (t)dt A > a ⊂ R+
a
is bounded from above. Z And we call theZ integral of f on [a, +∞) the supremum of
+∞
this set. We denote it by f or by f (t)dt. By using monotone convergence
[a,+∞) a
property we have,
Z +∞ Z A Z A
f (t)dt = sup f (t)dt = lim f (t)dt.
a A≥a a A→+∞ a

• In general cases (i.e., f is valued in C), the function f is called integrable if the
[a, +∞) −→ R+
positive valued function |f | : is integrable (in the sense of the
t 7−→ |f (t)|
previous case).

Proposition 2.3.11. Consider a function f which is the piecewisely continuous and positive
defined on [a, +∞), and a strictly increasing real sequence (tn )n∈N which tends to +∞. We
X Z tn+1
define a series xn with the general terms xn := f (t)dt.
n tn
X
Then, the series xn converges if, and only if the function f is integrable on [t0 , +∞).
n
In the case of convergence,
+∞
X Z tn Z +∞
xn = lim f (t)dt = f (t)dt.
n→+∞ t t0
n=0 0

Proof. We prove this proposition by double implications.

iv
That means, f is continuous on [a, +∞) except for an at most countable many points {a1 , a2 , · · · }.
Moreover, f possesses the left and right limits at each points of discontinuity.
v
Here, we choose the uniform partition over the segment [a, A]. In fact, the Riemann sums are defined
independently to the choice of partitions.
2.3. SERIES WITH COMPLEX GENERAL TERMS 85
X
⇒: We assume that the series xn converges. Since the function f is positive and
n X
(tn )n∈N is strictly increasing, we deduce that xn is a series with positive general
n
terms. So, its convergence shows the existence of a real number M > 0 such that for
Xn
all n ∈ N, xk ≤ M . Let A ∈ [t0 , +∞), the divergence tn −→ +∞ shows that
n→+∞
k=0
there exists nA ∈ N such that A ≤ tnA , and we have
Z A Z tnA A −1 Z tk+1
nX A −1
nX
f (t)dt ≤ f (t)dt = f (t)dt = xk ≤ M.
t0 t0 k=0 tk k=0

Hence, the function f is integrable on [t0 , +∞).

⇐: We assume now f is integrable on [t0 , +∞). Then there exists M > 0 such that for
Z A
all A > a, f (t)dt ≤ M . We have, by taking A = tn for all n ∈ N, that
a

n
X n Z
X tk+1 Z tn+1 Z tn+1
xk = f (t)dt = f (t)dt ≤ f (t)dt ≤ M.
k=0 k=0 tk t0 a

X
This implies the sequence of partial sums of the series xn is bounded from above.
X n X
Since the series xn has positive general terms, we conclude that the series xn
n n
converges. Moreover, the sum of this series can be obtained from the monotone
convergence property. Hence,
+∞
X Z tn Z +∞
xn = lim f (t)dt = f (t)dt.
n→+∞ t t0
n=0 0

Theorem 2.3.12 (Integral test). Let f be a function defined on [n0 , +∞) with n0 ∈ N. We
assume that f is piecewisely continuous, positive and decreasing function. Then,
X Z n 
• the series f (t)dt − f (n) converges;
n≥n0 +1 n−1

X
• the series f (n) converges if, and only if f is integrable on [n0 , +∞).
n≥n0 +1

Proof. Since f is decreasing, we have then f (n) ≤ f (t) ≤ f (n − 1) for all t ∈ [n − 1, n] and
all n ∈ N, n ≥ n0 . Integrating those inequalities on the interval [n − 1, n], we can obtain
that,
Z n
f (n) ≤ f (t)dt ≤ f (n − 1).
n−1
86 NUMERICAL SERIES

X Z n 
Then the series f (t)dt − f (n) has positive general terms and we have,
n≥n0 +1 n−1

n
X Z k  n
X
f (t)dt − f (k) ≤ (f (k − 1) − f (k)) = f (n0 ) − f (n) ≤ f (n0 ).
k=n0 +1 k−1 k=n0 +1

X Z n 
This shows the series f (t)dt − f (n) converges.
n≥n0 +1 n−1
Z n Z n 
For all n ∈ N, n ≥ n0 , we observe that f (t)dt = f (t)dt − f (n) + f (n),
n−1 n−1
X Z n X  Z n  X
hence f (t)dt = f (t)dt − f (n) + f (n). Since the series
n≥n n−1 n−1
X Z 0 n+1 n≥n0 +1 n≥n0 +1
X
f (t)dt − f (n) converges, the convergence of the series f (n) will have
n≥n0 +1 n−1 n≥n0 +1
X Z n
the same nature to the series f (t)dt. It follows from the previous proposition
n≥n0 +1 n−1
X Z n
that the series f (t)dt converges if, and only if the function f is integrable on
n≥n0 +1 n−1
X
[n0 , +∞). In the consequences, the series f (n) converges if, and only if f is integrable
n≥n0 +1
on [n0 , +∞).

Proposition 2.3.13 (Integral test, complex). Let n0 ∈ N and f be a complex-valued func-

tion of C 1 -class on [n0 , +∞) of whichthe derivative f ′ is integrable over [n0 , +∞). Then,
X Z n
the series f (t)dt − f (n) is absolutely convergent.
n≥n0 +1 n−1

Proof. Since f ′ is supposed to be integrable

Z n on [n
0 , +∞), it follows from the definition of
′
the integrability that the sequence |f (s)|ds is increasing and convergent in R.
n0 n≥n0
On the other hand, let n ≥ n0 + 1 and t ∈ [n − 1, n]. It follows from the fundamental
theorem of calculus that
Z n Z n Z n
′ ′
|f (n) − f (t)| = f (s)ds ≤ |f (s)|ds ≤ |f ′ (s)|ds.
t t n−1

In the consequences,
Z n Z n Z n Z n
f (t)dt − f (n) = (f (t) − f (n)) dt ≤ |f (n) − f (t)|dt ≤ |f ′ (s)|ds.
n−1 n−1 n−1 n−1
Z n 
′
We remark that the sequence |f (s)|ds is noting else but the sequence of
n0 n≥n0
X Z n
partial sums of the series |f ′ (s)|ds. Hence, the convergence of the sequence
n≥n0 +1 n−1
Z n  X Z n
|f ′ (s)|ds implies the convergence of the series |f ′ (s)|ds. Using the
n0 n≥n0 n≥n0 +1 n−1
2.3. SERIES WITH COMPLEX GENERAL TERMS 87

X Z n
comparison test, the series f (t)dt − f (n) converges. Thus, the claim follows.
n≥n0 +1 n−1

Corollary 2.3.14. Let n0 ∈ N and f be a complex-valued function of C 1X class on [n0 , +∞)

′
of which the derivative f is integrable over [n0 , +∞). Then, the series f (n) converges
n≥n0
Z n 
if, and only if the sequence f (t)dt has a limit.
n0 n∈N

Proof. Left in Exercises.

2.3.4 Alternate Series, Abel Transforms and Cauchy Products

Theorem 2.3.15 (Alternate series). Let (un )n∈N be X a decreasing sequence with positive
general terms which tends to 0. We consider the series (−1)n un , its sum S, the sequence
n
of partial sums (Sn )n∈N and the sequences of rests (Rn )n∈N . Then, it holds,
X
• the series (−1)n un converges;
n

• ∀n ∈ N, S2n+1 ≤ S ≤ S2n ;
+∞
X
• ∀n ∈ N, |Rn | = |S − Sn | = (−1)k uk ≤ un+1 .
k=n+1

Proof. For all n ∈ N, we have S2n = S2n−1 + u2n and S2n+1 = S2n − u2n+1 . Since the
sequence (un )n∈N is decreasing, we have u2n ≥ u2n+1 and thus,

S2n−1 = S2n − u2n ≤ S2n − u2n+1 = S2n+1 ≤ S2n .

Hence, S2n−1 ≤ S2n+1 ≤ S2n and similarly, S2n+1 ≤ S2n+2 ≤ S2n . Those relations imply
that the sequence (S2n+1 )n∈N is increasing, bounded from above by S0 and (S2n )n∈N is
decreasing, bounded from below by S1 . So, those two sequences converge. In addition, we
have S2n − S2n−1 = u2n −→ 0. Then the sequences (S2n+1 )n∈N and (S2n )n∈N converge to
n→+∞
the same limit S ∈ R, and it holds S = sup S2n+1 = inf S2n . We have thus, for all n ∈ N,
n∈N n∈N
S2n+1 ≤ S ≤ S2n . In the consequences, 0 ≤ S2n − S ≤ S2n − S2n+1 = u2n+1 as well as
0 ≤ S2n − S ≤ S2n − S2n−1 = u2n .

X (−1)n
Example 2.3.16. For all α > 0, the series converges. This series is absolutely
nα
n≥1
convergent only if α > 1.

As a generalization of the alternate series theorem, the Abel transform is often applied.

Theorem 2.3.17 (Abel transform). Let (an )n∈N and (bn )n∈N be sequences of complex num-
bers. We suppose that
88 NUMERICAL SERIES

n
!
X
• the sequence of partial sums ak is bounded;
k=0 n∈N

• (bn )n∈N is a decreasing sequence of positive real numbers which tends to 0.

X
Then, the series an bn converges.
n
Proof. See in Exercise 2.4.18

In the following theorem, we will study the possibility of “multiplier” two series and
look at the convergence.
X X
Definition 2.3.18. Let an and bn be series with complex general terms. We call
n n X
the Cauchy product of those two series the series cn given by,
n
n
X X
∀n ∈ N, cn = ak bn−k = ap bq .
k=0 p+q=n

X X
Theorem 2.3.19 (Cauchy Products). Let an and bn be absolutely convergent series.
X n n
Then, the Cauchy product cn of those series is also absolutely convergent. Moreover, it
n
holds that
  
+∞
X +∞
X +∞
X
cn =  ap   bq  .
n=0 p=0 q=0

Proof. We denote by X(An )n∈N

X, (Bn )n∈N
Xand (Cn )n∈N the sequences of partial sums respec-
tively to the series an , bn and cn .
n n n
Let n ∈ N, we define the subset,
Tn := (p, q) ∈ N2 p + q ≤ n .


As the immediate results, we remark that Tn ⊂ [[0, n]]2 ⊂ T2n . Equivalently speaking,
[[0, [n/2]]]2 ⊂ Tn ⊂ [[0, n]]2 .
Using binomial formula, we have that
X
An B n = ap bq ;
(p,q)∈[[0,n]]2
X
Cn = ap bq .
(p,q)∈Tn

From the definitions of ck , we have the estimations,

n
X n
X X n
X X X
|ck | = ap bq ≤ |ap |bq | = |ab ||bq |
k=0 k=0 p+q=k k=0 p+q=k (p,q)∈Tn
     
X n
X n
X X+∞ +∞
X
≤ |ap ||bq | =  |ap |  |bq | ≤  |ap |  |bq | .
(p,q)∈[[0,n]]2 p=0 q=0 p=0 q=0
2.3. SERIES WITH COMPLEX GENERAL TERMS 89

X X +∞
X +∞
X
Since the series an and bn are absolutely convergent, the sums |ap | and |bq |
n n p=0 q=0
are
X well-defined. So, this estimation shows that the sequence of partial sums of the Xseries
|cn | of positive real numbers is bounded from above. That is to say the series cn is
n n
absolutely convergent.
Now we estimate An Bn − Cn for n ∈ N,

X X X
|An Bn − Cn | = ap bq − ap bq = ap bq
(p,q)∈[[0,n]]2 (p,q)∈Tn (p,q)∈[[0,n]]2 \Tn
X X
≤ |ap ||bq | ≤ |ap ||bq |
(p,q)∈[[0,n]]2 \Tn (p,q)∈[[0,n]]2 \[[0,[n/2]]]2
X X
= |ap ||bq | − |ap ||bq |
(p,q)∈[[0,n]]2 (p,q)∈[[0,[n/2]]]2
     
n n [n/2] [n/2]
X X X X
= |ap |  |bq | −  |ap |  |bq | .
p=0 q=0 p=0 q=0
X X
Passing n → +∞, it follows from the absolute convergence of the series an and bn
n n
that
     
n n [n/2] [n/2]
X X X X
|An Bn − Cn | ≤  |ap |  |bq | −  |ap |  |bq |
p=0 q=0 p=0 q=0
     
X+∞ X+∞ X+∞ X+∞
−→  |ap |  |bq | −  |ap |  |bq | = 0
n→+∞
p=0 q=0 p=0 q=0

Hence, it holds that

  
+∞
X X+∞ +∞
X
cn =  ap   bq  .
n=0 p=0 q=0

X zn
Example 2.3.20. Let z ∈ C. We consider the series of complex numbers vi . We
n
n!
X 0n
denote r = |z| ≥ 0. If r = 0, it implies z = 0, then the convergence of is trivial.
n! n
We consider then the general cases r > 0. It follows from te usual comparison relations
rn
Proposition A.3.4 that rn = o(n!), which indicates that lim = 0. By using a simple
n→+∞ n!
change of indexes, we have,
rm rm rn−2 1 rn
lim = lim = lim = 2 lim = 0.
m→+∞ (m + 2)(m + 1)m! m→+∞ (m + 2)! n→+∞ n! r n→+∞ n!
vi
By convention, 0! = 1.
90 NUMERICAL SERIES

rm 1 rm 1
It shows that = o( ), in particular, = O( ). Also,
m! (m + 1)(m + 2) m! (m + 1)(m + 2)
X 1 1 1
the series is convergent since ∼ 2 and follows by
m
(m + 1)(m + 2) (m + 1)(m + 2) m
X 1
Riemann test. Or, can be simply seen the telescoping series of the
m
(m + 1)(m + 2)
 
1
convergent sequence − .
m + 2 m∈N
X rn
So, it follows from the comparison test that the series is convergent, in other
n
n!
X zn
words, the series is absolutely convergent. Then, we define the sum of this series as
n
n!
the complex exponentialvii of the number z and denote it by exp(z)viii . That is,
+∞ n
X z
exp(z) := .
n!
n=0

X an X bn
Consider now a, b ∈ C. From the above reasoning, we know the series and
n! n
n!
n
are absolutely convergent. Apply Theorem 2.3.19, the Cauchy product of those series is
absolutely convergent and
+∞ n
! +∞ ! +∞ !
X a X bn X X ap bq
exp(a) exp(b) = =
n! n! p! q!
n=0 n=0 n=0 p+q=n
+∞ n +∞
!
X 1 X n! X (a + b)n
= ak bn−k = = exp(a + b).
n! k!(n − k)! n!
n=0 k=0 n=0

Hence, the multiplication of exponentials exp(a + b) = exp(a) exp(b) holds.

2.4 Exercises
Exercise 2.4.1. Justify Example 2.1.3.

Exercise 2.4.2. Justify Proposition 2.1.5.

X
Exercise 2.4.3. Let xn be a convergent series with positive general terms. Show that
n
X
1.) x2n converges;
n

X √xn
2.) converges.
n
n≥1
vii
ATTENTION! You should distinguish this rigorous definition of exp and the intuitive understanding “e
power z”.
viii
It can be also denoted by ez .
2.4. EXERCISES 91

+∞
X 1
Exercise 2.4.4 (A Solution to Basel Problem). The determination of the value is
n2
n=1
well-known as the Basel problem, which was firstly resolved by Leonhard Euler in 1734.
X 1
1.) Justify the series converges.
n2
n≥1

2.) Let m ∈ N∗ . By using de Moivre’s formula, show that there exists a polynomial
Pm ∈ R[X] such that

∀θ ∈ R \ πZ, sin((2m + 1)θ) = (sin θ)2m+1 Pm (cot2 θ).

3.) Determinate the roots of Pm and factorize Pm .

m
X kπ
4.) Deduce the value of cot2 ( ).
2m + 1
k=1

1
5.) Let t ∈ (0, π2 ). Show that sin t ≤ t ≤ tan t then that cot2 t ≤ ≤ 1 + cot2 t.
t2
+∞
X 1
6.) Calculate the value of by using the previous questions.
n2
n=1

X (−1)n−1 X 1
7.) Show that the series and converges.
n2 (2n − 1)2
n≥1 n≥1

+∞ +∞
X (−1)n−1 X 1
8.) Determinate the values of and .
n2 (2n − 1)2
n=1 n=1

Summation by Regrouping Terms

X
Exercise 2.4.5. Let xm be a series of complex numbers. Let (pn )n∈N be a strictly
m
increasing sequence of natural numbers.

1.) Show that for all m ∈ N and m ≥ p0 , there exists a unique N (m) ∈ N such that
pN (m) < m ≤ pN (m)+1 .
pn
X m
X
2.) We define for all n, m ∈ N∗ , yn = xk and rm = xk . Show that the
k=pn−1 k=pN (m) +1
X X
series xm converges if, and only if yn converges and rm −→ 0. Furthermore,
m→+∞
m n
+∞
X +∞
X
xm = yn .
m=0 n=0

(−1)n−1 cos(2nπ/3)
Exercise 2.4.6. For all n ∈ N∗ , we denote xn = n−1
and yn = .
n + (−1) n
92 NUMERICAL SERIES
X
1.) Can we apply alternate series theorem on xn ? Justify your answer.
n
X
2.) Show that xn converges.
n
X
3.) Show that yn converges.
n

Telescoping Series and Integral Tests

 
X 1
Exercise 2.4.7. Consider the series arctan .
n
n2 + n + 1
 
1
1.) Let n ∈ N. Show that arctan = arctan(n + 1) − arctan n.
n2 + n + 1
 
X 1
2.) Show that the series arctan 2+n+1
converges and determinate its sum.
n
n

Exercise 2.4.8 (Euler’s constant γ). Let (Hn )n∈N be the sequence of partial sums of the
harmonic series.
n
X 1
∀n ∈ N∗ , Hn = .
k
k=1

X1
1.) Show that diverges.
n
n≥1

2.) Compare the series with integral, show that Hn ∼ ln n while n → +∞.
X
3.) Consider the sequence (Hn − ln n)n≥1 as partial sums of the series un . Show that
n≥1
Hn − ln n converges. This limit is called as Euler’s constant γ.

Exercise 2.4.9 (Comparison with integrals). Let f be a positive, continuous and decreasing
function on [n0 , +∞)

1.) Show, for all n ≥ n0 + 1, we have that

n
X Z n n−1
X
f (k) ≤ f (t)dt ≤ f (k).
k=n0 +1 n0 k=n0

as well as that
Z n+1 n
X Z n
f (t)dt ≤ f (k) ≤ f (n0 ) + f (t)dt.
n0 k=n0 n0
2.4. EXERCISES 93

X  Z n+1 
2.) Show that the series f (n) − f (t)dt is convergent. Also,
n≥n0 n

+∞ 
X Z n+1 
0≤ f (n) − f (t)dt ≤ f (n0 ).
n=n0 n

3.) Show that, the following asymptotic behavior holds while n → +∞,
X
• if the series f (n) diverges, then
n≥n0

n
X Z n
f (k) ∼ f (t)dt.
k=n0 n0

X
• if the series f (n) converges, then
n≥n0

+∞
X Z +∞
f (k) ∼ f (t)dt.
k=n+1 n

4.) We assume from now on that f is convex on [n0 , +∞). Show that for all n ≥ n0 ,
Z n+1
f (n) − f (n + 1)
≤ f (n) − f (t)dt.
2 n

X
5.) Following the previous question, show that, if the series f (n) converges, then
n≥n0

+∞  Z n+1 
f (n0 ) X
≤ f (n) − f (t)dt ≤ f (n0 ).
2 n=n n
0

Exercise 2.4.10. Justify the convergence of Riemann series by using integral tests.

Exercise 2.4.11. We use the notations introduced in Exercise 2.4.8. Recall that
n
X
∀n ≥ 1, Hn − ln n = uk .
k=1

1.) Let α > 1. Show that it holds while n → +∞,

Z n
1 1
α
∼ α
dt,
n n−1 t

2.) Deduce that

+∞
X 1 1
α
∼ .
k (α − 1)nα−1
k=n
94 NUMERICAL SERIES

1
3.) Show that the Euler’s constantix satisfies ≤ γ ≤ 1.
2
1 1
4.) Justify that un ∼ − . Then show that
2 n2
1 1
Hn = ln n + γ − + O( 2 ).
2n n
holds while n → +∞.

Ratio Test: Delicate Cases

X
Exercise 2.4.12. Consider a series xn with strictly positive general terms. We suppose
n
that the asymptotic behavior holds while n → +∞,
xn+1 α 1
= 1 − + O( β ),
xn n n
where α ∈ R and β > 1.
X
1.) Is the ratio test sufficient to determinate the convergence of the series xn ? Justify
n
your answer.

2.) Let (yn )n∈N be a sequence of strictly positive real numbers.

 Show that the sequence
X yn+1
(yn )n∈N converges if, and only if the series ln converges.
n
yn

(n + 1)α xn+1
 
3.) For all n ∈ N∗ , we set wn = ln . Give the asymptotic expansion of
nα xn
wn while n → +∞.

4.) Show that the sequence (nα xn )n∈N converges to a non-zero limit in R.
X
5.) Deduce that if α > 1 then the series xn converges and it diverges if α ≤ 1.
n

n!
Exercise 2.4.13. For all n ∈ N∗ , we introduce xn = .
nn e−n
 
xn+1
1.) Give the asymptotic expansion of ln while n → +∞.
xn
√  n n
2.) Deduce that there exists K > 0 such that n! ∼ K n .
e
3.) Do Exercises A.4.5 to A.4.7 to determinate the value of K.

4.) Conclude with Stirling’s Formula.

ix
The approximate value of γ is given as γ = 0.577215664901533....
2.4. EXERCISES 95

Study the Convergences of Series

Exercise 2.4.14. Study the convergences of the following series.
X (−1)n X (−1)n
1.) √ and .
n
n + (−1)n ln n n
n + (−1)n ln n

(−1)n
  X  
X 1
2.) ln cos α and ln 1 + α
with α ∈ R∗+ .
n
n n
n
  
X
2 1 X 1
3.) cos πn ln 1 − and 2
. It is necessary to justify the definition of
n
n n
n sin n
the second series.

Exercise 2.4.15. Let α, β ∈ R. Study the convergences of the following series.

X (−1)n
1.) by assuming α ̸= β.
n
nα + (−1)n−1 nβ

X esin n
2.) .
n
nα
X  p 
3.) sin π n2 + αn + β .
n

√ √
Exercise 2.4.16. 1.) Let n ∈ N. Show that (3 + 5)n + (3 − 5)n is an even integer.
X  √ 
2.) Study the convergence of the series sin π(3 + 5)n .
n

Exercise
X 2.4.17 (Difficult, may skip at the first look). Determinate whether the series
xn is convergent or not where xn is given in the following cases,
n

√ √
1 n+1− n−1
1.) n
!α with α ∈ R; 5.) (−1) n
with α ∈ R;
X√ k nα
k Z n
α α
k=1 6.) e−n et dt with α ∈ R;
1
n p
Y 3
2.) (2 − e1/k ); 7.) sin(π n3 + λnα ) with λ ̸= 0 and α ≤
k=1 2;
√
n−1
sin(π n)
1 8.) with α ∈ R;
nα
X
n
3.) (−1) ;
p(n − p) Z +∞ n−t
p=1 e
9.) dt;
!ln(n) n n+t
+∞
X (−1)k−n Z 1
4.) ; 10.) (−1) n
cos(nt2 )dt;
k
k=n 0
96 NUMERICAL SERIES

+∞
12.) sin(e−n ).
Z
n
11.) e−(ln t) dt with a ≥ 1;
a

Abel Transform
Exercise 2.4.18. Consider two sequences (an )n∈N and (bn )n∈N X of complex numbers. We
denote by (An )n∈N the sequence of partial sums of the series an . That is, for all n ∈ N,
n
n
X
An = ak . We make the following assumptions
k=0

• The sequence (An )n∈N is bounded.

• (bn )n∈N is a decreasing sequence of positive real numbers which tends to 0.

Define for all n ∈ N, un = an bn .

1.) Let n, p ∈ N∗ . Prove the Abel transform,

n+p
X n+p−1
X
uk = An+p bn+p + Ak (bk − bk+1 ) − An bn+1 .
k=n+1 k=n+1

X
2.) Show that the series un converges by using Cauchy sequences.
n

Exercise 2.4.19. Let θ ∈ R and α > 0.

n
X
1.) For all n ∈ N∗ , calculate cos(kθ).
k=1

X cos(nθ)
2.) Determinate the convergence of the series .
nα
n≥1

3.) Is this series absolutely convergent? Justify your answer.

Exercise 2.4.20. Consider (an )n∈N be a sequence of positive real numbers. We define the
following function f and sequence (tn )n∈N .

• For all x ∈ R+ , f (x) := Card {n ∈ N | an ≥ x}.

• x0 = sup ak ; ∀n ∈ N∗ , xn = sup {ak | k ∈ N, ak < xn−1 }.

k∈N

We suppose that an −→ 0.
n→+∞

1.) Show that for all x ∈ R+ , f (x) is well-defined also that f vanishes on [M, +∞) for
certain M > 0.

2.) Justify that (xn )n∈N is positive, decreasing and tending to 0.

2.4. EXERCISES 97
X
3.) Consider from now on the series bn defined by ∀n ∈ N, bn = f (xn )(xn − xn+1 ).
n
Show that
n φ(n)
X X
∀n ∈ N, ∃φ(n) ∈ N s.t. bk ≤ ak .
k=0 k=0

4.) Let n ∈ N. Show that there exists ψ(n) ∈ N such that

n ψ(n)
X 1 X
ak − n ≤ bk .
2
k=0 k=0

X X
5.) Deduce that an converges if, and only if bn converges. Furthermore, show that
X n n
if an converges, then it holds
n

+∞
X +∞
X
an = bn .
n=0 n=0

X
A.) Show that the series an converges if, and only if f (x) is integrable on (0, +∞).
n
X
B.) Show that if an converges, then it holds
n

+∞
X Z +∞
an = f (x)dx.
n=0 0
98 NUMERICAL SERIES
 Chapter 3

Topology - Elementary
Introduction

In this chapter, K = R or C.

3.1 Metric Spaces

In this section, S denotes a non-empty set.
Definition 3.1.1. We call a distance function on the set S any function d which maps
from S 2 into R+ and satisfies the following axioms.
D-1.) ∀x, y ∈ S, d(x, y) = d(y, x). Symmetry

D-2.) ∀x, y, z ∈ S, d(x, z) ≤ d(x, y) + d(y, z). Triangular inequality

D-3.) ∀x, y ∈ S, d(x, y) = 0 if, and only if x = y. Separation property

Definition 3.1.2. We call a metric space any non empty set S together with a distance
d on S. We denote it by (S, d).

Proposition 3.1.3 (first and second triangular inequality). Let (S, d) be a metric space.
For all x, y, z ∈ S, it holds that

d(x, z) ≤ d(x, y) + d(y, z),

and that

|d(x, z) − d(y, z)| ≤ d(x, y).

Proof. Straightforward from the definition of metric spaces.

Definition 3.1.4. Let X and Y be two non-empty subsets in the metric space (S, d).
• The distance from a point x ∈ S to X is defined by

d(x, X) := inf d(x, y).

y∈X

99
100 TOPOLOGY - ELEMENTARY INTRODUCTION

• The distance between the subsets X and Y is defined by

d(X, Y ) := inf d(x, y).

x∈X, y∈Y

• The Hausdorffi distance between X and Y is defined by

( )
dH (X, Y ) := max sup d(x, Y ), sup d(y, X) .
x∈X y∈Y

• The diameter of the subset X is defined by

diam(X) := sup d(x, y).

x,y∈X

Proposition 3.1.5. Let X be a non-empty subset in the metric space (S, d). Then, for all
x, y ∈ S,

|d(x, X) − d(y, X)| ≤ d(x, y).

Proof. Let z ∈ X. From the definition of the distance d(x, X) and the triangular inequality,
we have

d(x, X) ≤ d(x, z) ≤ d(x, y) + d(y, z).

This inequality implies that

d(x, X) − d(x, y) ≤ d(y, z)

holds for all z ∈ X. We remark that the left-hand-side d(x, X) − d(x, y) ∈ R is independent
to z. In other words, the previous inequality implies that d(x, X) − d(x, y) is a lower bound
to the set {d(y, z) | z ∈ X}. Since the infimum d(y, X) = inf {d(y, z) | z ∈ X} is the largest
lower bound, we obtain thus, d(x, X) − d(x, y) ≤ d(y, X). Hence,

d(x, X) − d(y, X) ≤ d(x, y).

By symmetry, it is sufficient to exchange the symbols x and y to deduce

d(y, X) − d(x, X) ≤ d(x, y).

Thus, the claim follows.

Remark 3.1.6. This proposition states that for any non-empty set X in a metric space (S, d),
the function which measuring the distance from one point to this set X is 1-Lipschitz
continuous. The classification of continuities will be introduced in Section 3.6.3.
i
Felix Hausdorff (1868 - 1942), German mathematician. Famous for his contributive works on topology,
set theory, measure theory and functional analysis.
3.2. NORMED SPACES 101

3.2 Normed Spaces

In this section, E denotes a K-vector space.
Definition 3.2.1. We call a norm on the space E any function ∥ · ∥ maps from E into R+
and satisfying the following axioms.
N-1.) ∀(α, x) ∈ K × E, ∥αx∥ = |α| · ∥x∥. Homogeneity
N-2.) ∀x, y ∈ E, ∥x + y∥ ≤ ∥x∥ + ∥y∥. Triangular inequality
N-3.) ∀x ∈ E, ∥x∥ = 0 if, and only if x = 0E . Separation property

Definition 3.2.2. We call a normed vector space or simply a normed space any
K-vector space together with a norm ∥ · ∥ on E. We denote it by (E, ∥ · ∥).

Remark 3.2.3. • Any normed space can be considered as a metric space with the in-
duced distance d(x, y) = ∥x − y∥ defined on E.
• We call a unit vector any vector x ∈ E with ∥x∥ = 1. In fact, for all x ∈ E \ {0E },
x
x possesses the unique decomposition x = ∥x∥x̂ where ∥x∥ ∈ R∗+ and x̂ = is a
∥x∥
unit vector. Furthermore, we call the unit sphere the set of unit vectors in E. In
particular, we denote by Sn−1 the unit sphere in Rn for all n ∈ N∗ .
• Let (E, ∥ · ∥) be a normed space. For any subspace F in E, we can associate F with
the norm ∥ · ∥F obtained by the restriction of ∥ · ∥ on F . This norm ∥ · ∥F is called as
the induced norm of ∥ · ∥ on F and the space (F, ∥ · ∥F ) (we can simply denote it as
(F, ∥ · ∥)) is called as the induced normed subspace of (E, ∥ · ∥) on F .

Proposition 3.2.4. Let (E, ∥ · ∥) be a normed space and let n ∈ N∗ .

1.) For all (α1 , · · · , αn ) ∈ Kn and all (x1 , · · · , xn ) ∈ E n , we have the inequality of
sublinearity
n
X n
X
αj xj ≤ |αj |∥xj ∥.
j=1 j=1

2.) For all x, y ∈ E we have the second triangular inequality

∥x∥ − ∥y∥ ≤ ∥x − y∥.

Proof. 1.) Direct consequence of the homogeneity and the triangular inequality.
2.) Let x, y ∈ E, we write x = (x − y) + y and apply the triangular inequality to obtain,
∥x∥ = ∥(x − y) + y∥ ≤ ∥x − y∥ + ∥y∥.
It implies that ∥x∥ − ∥y∥ ≤ ∥x − y∥.
From the symmetry, we have also
∥y∥ − ∥x∥ ≤ ∥y − x∥ = ∥x − y∥.
Then the claim follows.
102 TOPOLOGY - ELEMENTARY INTRODUCTION

Definition 3.2.5. Let ∥ · ∥N1 and ∥ · ∥N2 be norms on the vector space E. We say the norm
∥ · ∥N1 is equivalent to ∥ · ∥N2 if there exists c, C > 0 such that,

∀x ∈ E, c∥x∥N1 ≤ ∥x∥N2 ≤ C∥x∥N1 .

Proposition 3.2.6 (Reflexivity and Symmetry of Equivalence of Norms). Let E be a vector

space and ∥ · ∥, ∥ · ∥N1 , ∥ · ∥N2 be norms defined on E. Then, it holds that

• ∥ · ∥ is equivalent to itself;

• ∥ · ∥N1 is equivalent to ∥ · ∥N2 if, and only if ∥ · ∥N2 is equivalent to ∥ · ∥N1 .

Proof. Left in Exercises.

Using the previous proposition, we will say the norms ∥ · ∥N1 and ∥ · ∥N2 are equivalent
if the one is equivalent to the other.

Proposition 3.2.7 (Transitivity of Equivalence of Norms). Let ∥ · ∥N1 , ∥ · ∥N2 and ∥ · ∥N3
be norms defined on E. We assume that ∥ · ∥N1 , ∥ · ∥N2 are equivalent and ∥ · ∥N2 , ∥ · ∥N3
are equivalent. Then, ∥ · ∥N1 and ∥ · ∥N3 are equivalent.

Proof. The two equivalence relations between ∥ · ∥N1 , ∥ · ∥N2 and between ∥ · ∥N2 , ∥ · ∥N3
imply the existence of the four strictly positive constants c1,2 , c2,3 , C1,2 and C2,3 satisfying
that ∀x ∈ E,

c1,2 ∥x∥N1 ≤ ∥x∥N2 ≤ C1,2 ∥x∥N1 ,

as well as

c2,3 ∥x∥N2 ≤ ∥x∥N3 ≤ C2,3 ∥x∥N2 .

Combining those inequalities, we obtain that

c1,2 c2,3 ∥x∥N1 ≤ ∥x∥N3 ≤ C1,2 C2,3 ∥x∥N1 .

Thus the claim follows.

Remark 3.2.8. We have shown that the equivalence between norms is reflexive, symmetric
and transitive. Thus, it is an equivalence relation.

Example 3.2.9 (p-norms on Kn ). Let n ∈ N∗ and consider the vector space Kn , we define
for all x = t (x1 , · · · , xn ) ∈ Kn ,

n
!1
 X p

|xi |p

if 1 ≤ p < ∞,

∥x∥p :=

 i=1
max{|x1 |, · · · , |xn |} if p = ∞.


Then ∥ · ∥p is a norm on Kn .
3.2. NORMED SPACES 103

Solution. 1.) Homogeneity. Let α ∈ K, x ∈ Kn , we have

n
X n
X n
X
∥αx∥pp = |αxi |p = |α|p |xi |p = |α|p |xi |p ,
i=1 i=1 i=1

which leads to ∥αx∥p = |α|∥x∥p in the cases where 1 ≤ p < ∞. Also,

∥αx∥∞ = max{|αx1 |, · · · , |αxn |} = |α| max{|x1 |, · · · , |xn |} = |α|∥x∥∞ .

2.) Triangular inequality. Let x, y ∈ Kn , using the triangular inequality in R, we have

∥x + y∥∞ = max{|x1 + y1 |, · · · , |xn + yn |} ≤ max{|x1 | + |y1 |, · · · , |xn | + |yn |}

≤ max{|x1 |, · · · , |xn |} + max{|y1 |, · · · , |yn |} = ∥x∥∞ + ∥y∥∞ .

The cases where 1 ≤ p < ∞ is left in Exercise 3.7.1.

n
!1
X p

3.) Separation property. If x = 0Kn , then ∥x∥p = 0p = 0 or max{0, · · · , 0} = 0.

i=1
n
X
Reciprocally, consider a vector x ∈ Kn such that ∥x∥p = 0. Then, |xi |p = 0 or
i=1
max{|x1 |, · · · , |xn |} = 0 and it leads to xi = 0 for all 1 ≤ i ≤ n. Thus, x = 0Kn .

Remark 3.2.10. The triangular inequality ∥x + y∥p ≤ ∥x∥p + ∥y∥p is called the Minkowskiii
inequality.

Proposition 3.2.11. Let ∥ · ∥p denote the p-norms on Kn with 1 ≤ p < ∞. For all x ∈ Kn ,
the following inequalities hold

1.) ∥x∥∞ ≤ ∥x∥1 ≤ n∥x∥∞ ,

√
2.) ∥x∥∞ ≤ ∥x∥2 ≤ n∥x∥∞ ,
√
3.) ∥x∥2 ≤ ∥x∥1 ≤ n∥x∥2 ,

4.) ∥x∥p ≤ ∥x∥1 ≤ n1−1/p ∥x∥p .

Proof. Let x = t (x1 , · · · , xn ) ∈ Kn .

1.)
n
X n
X
∥x∥∞ = max |xi | ≤ |xi | = ∥x∥1 ≤ max |xi | = n∥x∥∞ .
1≤i≤n 1≤i≤n
i=1 j=1
ii
Hermann Minkowski (1864 - 1909), Russian/German mathematician and physicist. Several works on
geometry and number theory. He introduced the “Minkowski spacetime” to interpret Einstein’s special
relativity.
104 TOPOLOGY - ELEMENTARY INTRODUCTION

2.) We may assume i0 ∈ [[1, n]] be the index such that |xi0 | = max{|x1 |, · · · , |xn |} = ∥x∥∞ .
Since the function t 7→ t2 is increasing on R+ , we have
n
X
∥x∥2∞ = |xi0 |2 ≤ |xi |2 = ∥x∥22 ;
i=1
n
X Xn
∥x∥22 = |xi |2 ≤ |xi0 |2 = n∥x∥2∞ .
i=1 j=1

3.) Applying the Cauchy-Schwarziii inequalityiv on Kn ,

n n
!2
X X X
∥x∥22 = 2
|xi | ≤ |xi ||xj | = |xi | = ∥x∥21
i=1 (i,j)∈[[1,n]] i=1
n
!2 n
! n
!
X X X
= 1|xi | ≤ 12 |xi |2 = n∥x∥22 .
i=1 i=1 i=1

4.) For the case where x = 0Kn , this inequality is immediate. We assume now x ̸= 0Kn . It
follows from the separation property of the norm ∥ · ∥1 that ∥x∥1 > 0. Then it is licit
|xi |
to introduce the numbers αi = ∈ [0, 1] for all i ∈ [[1, n]]. In parallel, it follows
∥x∥1
from the convexity of the function t 7→ tp , the inequality tp ≤ t holds on [0, 1] since
p ≥ 1. Thus,
n n n n
1 X p
X p
X 1 X
|x i | = α ≤ αi = |xi | = 1.
∥x∥p1 i
∥x∥1
i=1 i=1 i=1 i=1

In the consequence, we obtain that

n
|xi |p ≤ ∥x∥p1 .
X
∥x∥pp =
i=1

On the other hand, it follows from Hölderv inequalityvi in Kn ,

n
X
∥x∥1 = 1|xi | ≤ ∥1Kn ∥p′ ∥x∥p = n1−1/p ∥x∥p .
i=1

Here, we denote by 1Kn to represent the vector t (1, 1, · · · , 1) and p′ be the conjugate
1 1
exponent of p, i.e., the number which satisfies + ′ = 1.
p p
iii
Karl Hermann Amadeus Schwarz (1843 - 1921), German mathematician. He was known for his work in
complex analysis.
iv
It is the well-known inequality on inner product spaces, its general forms are introduced in Proposi-
tion 5.2.7 and Proposition 5.2.6. In Kn , where the usual Euclidean or Hermitian structure is endowed, it
is stated as follows. Let u, v ∈ Kn , u = t (u1 , · · · , un ) and v = t (v1 , · · · , vn ). Then ⟨u, v⟩ ≤ ∥u∥2 ∥v∥2 ,
v v
Xn u n u n
uX uX
precisely speaking, ui vi ≤ t |ui | t
2 |vi |2 .
i=1 i=1 i=1
v
Ludwig Otto Hölder (1859 - 1937), German mathematician.
vi
It is a very useful inequality and having various forms in different contexts. We provide the suitable one
in Exercise 3.7.1
3.2. NORMED SPACES 105

Corollary 3.2.12. For any 1 ≤ p, q ≤ ∞, the norms ∥ · ∥p and ∥ · ∥q on Kn are equivalent.

Proof. From the first and the least equivalence relations between ∥ · ∥p and ∥ · ∥1 in Propo-
sition 3.2.11, we norms ∥ · ∥p are equivalent to ∥ · ∥1 for all 1 ≤ p ≤ ∞. Moreover, from the
transitivity of the equivalence relation in Proposition 3.2.7, all norms ∥ · ∥p are equivalent
from one to each others.

Remark 3.2.13. In fact, we will see in Theorem 4.4.10 that on a finite dimensional vector
space, all norms are equivalent from one to each others.

Example 3.2.14 (Matrix norms). We denote by Mm×n (K) the K-vector space of matrices
with m lines and n columns. We set E = Kn and F = Km . Then all linear applications
from E into F can be represented by a matrix in Mm×n (K). We denote for the moment
by ∥ · ∥p,E and by ∥ · ∥p,F the p-norms on E and on F .
We define, for all A ∈ Mm×n (K),

∥Ax∥p,F
∥A∥p,M = sup = sup ∥Ax∥p,F . (3.1)
x∈E ∥x∥p,E x∈E
x̸=0E ∥x∥p,E =1

• Show that for all x ∈ E, ∥Ax∥p,F ≤ ∥A∥p,M ∥x∥p,E .

• Justify ∥ · ∥p,M is a norm on Mm×n (K).

• Let A = (ai,j )1≤i≤m , then

1≤j≤n
(m )
X
– if p = 1, ∥A∥1,M = max |ai,j | ;
1≤j≤n
i=1
 
Xn 
– if p = ∞, ∥A∥∞,M = max |ai,j | ;
1≤i≤m  
j=1
q q 
– if p = 2, ∥A∥2,M = ρ( AA) = max λ λ ∈ σ(t AA) .
t

Solution. Let 1 ≤ p ≤ ∞, we consider the function ∥ · ∥p,M defined on Mm×n (K) by (3.1).

• At first, we prove the inequality ∀x ∈ E, ∥Ax∥p,F ≤ ∥A∥p,M ∥x∥p,E . If x = 0E , then

Ax = 0F by the linearity of A. If x ̸= 0E , it follows from the definition that
 
∥Ax∥p,F ∥Ay∥p,F 
∥Ax∥p,F = ∥x∥p,E ≤  sup  ∥x∥p,E = ∥A∥p,M ∥x∥p,E . (3.2)

∥x∥p,E y∈E ∥y∥p,E
y̸=0E

Hence, the claim follows.

• We verify that ∥ · ∥p,M is indeed a norm on Mm×n (K)

106 TOPOLOGY - ELEMENTARY INTRODUCTION

– Homogeneity. Let A ∈ Mm×n (K) and α ∈ K. It follows from the linearity of the
scalar multiplication and the homogeneity of the norm ∥ · ∥p,F , we have, for all
x ∈ E, ∥αAx∥p,F = |α|∥Ax∥p,F . And hence,

∥αAx∥p,F ∥Ax∥p,F
∥αA∥p,M = sup = |α| sup = |α|∥A∥p,M .
x∈E ∥x∥p,E x∈E ∥x∥p,E
x̸=0E x̸=0E

– Triangular inequality. Let A, B ∈ Mm×n (K). It follows from the linearity of

multiplication with a vector and the triangular inequality of the norm ∥ · ∥p,F as
well as (3.2) that for all x ∈ E,

∥(A + B)x∥p,F ≤ ∥Ax + Bx∥p,F ≤ ∥Ax∥p,F + ∥Bx∥p,F

≤ (∥A∥p,M + ∥B∥p,M ) ∥x∥p,E .

In particular if x ̸= 0E , we have

∥(A + B)x∥p,F
≤ ∥A∥p,M + ∥B∥p,M .
∥x∥p,E

It implies that the positive real number ∥A∥p,M + ∥B∥p,M is an upper bound
∥(A + B)x∥p,F
of the quotient for all x ∈ E \ {0E }. From the definition of the
∥x∥p,E
supremum, we deuce that

∥(A + B)x∥p,F
∥A + B∥p,M = sup ≤ ∥A∥p,M + ∥B∥p,M .
x∈E ∥x∥p,E
x̸=0E

Hence, the triangular inequality holds.

– Separation property. If A = 0Mm×n (K) , then for all x ∈ E,

∥Ax∥p,F = ∥0Mm×n (K) x∥p,F = ∥0F ∥p,F = 0.

∥Ax∥p,F 0
And hence, ∥A∥p,M = sup = sup = 0. On the other hand,
x∈E ∥x∥p,E x∈E ∥x∥p,E
x̸=0E x̸=0E
let A ∈ Mm×n (K) such that ∥A∥p,M = 0. Then, using the inequality (3.2), we
deduce that for all x ∈ E, ∥Ax∥p,F = 0 and hence Ax = 0F from the separation
property of ∥ · ∥p,F . In the consequences, A = 0Mm×n (K) . Then the separation
property of ∥ · ∥p,M is verified.

• Let A = (ai,j )1≤i≤m , then for x = t (x1 , · · · , xn ) ∈ E, we have

1≤j≤n

   P 
n
a1,1 · · · a1,n x1 j=1 a1,j xj
 . .. ..   .   .. 
Ax =  ..
 . .   .  
 .  =  . .

Pn
am,1 · · · am,n xn j=1 am,j xj
3.2. NORMED SPACES 107

n
X
– If p = 1, and let x ∈ E such that ∥x∥1,E = |xj | = 1,
j=1

m X
X n m X
X n n X
X m
∥Ax∥1,F = ai,j xj ≤ |ai,j ||xj | = |ai,j ||xj |
i=1 j=1 i=1 j=1 j=1 i=1
n m
(m )
X X X
= |xj | |ai,j | ≤ max |ai,j | .
1≤j≤n
j=1 i=1 i=1

The equality is achieved vii

m
( m for)x = (δj,j0 )1≤j≤n where j0 is the index such that
X X
|ai,j0 | = max |ai,j | . We can thus obtain,
1≤j≤n
i=1 i=1
(m )
X
∥A∥1,M = max |ai,j | .
1≤j≤n
i=1

– If p = ∞, and let x ∈ E such that ∥x∥∞,E = max |xj | = 1,

1≤j≤n
   
n
X Xn  Xn 
∥Ax∥∞,F = max ai,j xj ≤ max |ai,j ||xj | ≤ max |ai,j | .
1≤i≤m 1≤i≤m   1≤i≤m  
j=1 j=1 j=1

The equality is achieved for x = t (e−iθ1 , · · · , e−iθn ). Here, for j ∈[[1, n]], θj =
Xn Xn 
arg(ai0 ,j ) and i0 refers to the index such that |ai0 ,j | = max |ai,j | .
1≤i≤m  
j=1 j=1
We can thus obtain,
 
Xn 
∥A∥∞,M = max |ai,j | .
1≤i≤m  
j=1

– We consider now the case where p = 2. We endow the spaces E = Kn and

F = Km with their Euclidean or Hermitian structuresviii . That means, the inner
n
X
product ⟨u, v⟩E := ui vi for all u, v ∈ E as well as the analogue definition for
i=1
those in F . Then, let x ∈ E, ∥Ax∥22,F = ⟨Ax, Ax⟩F . In a matrix representation,
we can write ∥Ax∥22,F = x∗ A∗ Ax, where x∗ and A∗ signify the adjoints respec-
tively to the matrices x and A. In the case of K = R, A∗ = t A whereas A∗ = t A
if K = C.
We can observe that A∗ A ∈ Mn×n (K) is a Hermitian matrix if K = C (this
matrix is symmetric if K = R). From the theories in Linear Algebra, the matrix
A∗ A is ortho-diagonalizable. There exists therefore an unitary (resp. orthogonal)
(
vii 1 if j = j0
Here, δi,j represents the Kronecker symbol, which is defined by δj,j0 = .
0 if j ̸= j0
viii
For rigorous definitions, see Chapter 5.
108 TOPOLOGY - ELEMENTARY INTRODUCTION

matrixix P ∈ U (n) (resp. P ∈ O(n) if K = R) such that P ∗ A∗ AP = D with the

matrix D is diagonal composed with the eigenvalues of the matrix A∗ A. That
means,
 
λ1 0 · · · 0
 0 λ2 (0) 0 
 
D=
 .. . . 
 . (0) . . .. 


0 · · · 0 λn

where ∀1 ≤ i ≤ n, λi ∈ σ(A∗ A) counted with the multiplicity.

With this spectral decomposition, we have that

∥Ax∥22,F ≤ max {λi }∥P x∥22,E = max {λi }(x∗ P ∗ P x) = max {λi }∥x∥22,E .
1≤i≤n 1≤i≤n 1≤i≤n

Also, the equality is achieved when P x coincides with an eigenvector of the

eigenvalue max {λi }x to the matrix A∗ A. We can thus obtain,
1≤i≤n

p p
∥A∥2,M = ρ(A∗ A) = max {λ | λ ∈ σ(A∗ A)}.

Example 3.2.15 (Bounded, Summable and Square-Summable Sequences). We consider

the sequences of elements in Cxi . Let u = (un )n∈N ∈ CN be a sequence of complex numbers.

• We denote by ℓ∞ (C) the set of bounded sequences. We define ∥u∥ℓ∞ (C) = sup |un | ∈
n∈N
R+ , then (ℓ∞ (C), ∥ · ∥ℓ∞ (C) ) is a normed space.
X
• We say u is summable if the series un is absolutely convergentxii , and denote by
n
ℓ1 (C) the set of summable sequences. We define

+∞
X
∥u∥ℓ1 (C) = |un | ∈ R+ .
n=0

Then (ℓ1 (C), ∥ · ∥ℓ1 (C) ) is a normed space.

ix
An orthogonal matrix (resp. unitary) matrix of order n, n ∈ N∗ , is by definition a n × n real (resp.
complex) matrix M such that t M M = In (resp. t M M = In ). The set of orthogonal (resp. unitary) matrices
of order n is denoted by O(n) (resp. U (n)).
x
We denote it by ρ(A∗ A) and call it the spectral radius of the matrix A∗ A.
xi
Of course, those results hold for sequences of real numbers.
+∞
X
xii
We might write “ |un | < +∞” to reformulate this absolute convergence property.
n=0
ATTENTION!! DON’T ABUSE TO USE THE SYMBOL ∞.
3.2. NORMED SPACES 109
X
• We say u is square-summable if the series |un |2 is convergentxiii and denote by
n
ℓ2 (C) the set of square-summable sequences. We define
v
u +∞
uX
∥u∥ℓ2 (C) = t |un |2 ∈ R+ .
n=0

Then (ℓ2 (C), ∥ · ∥ℓ2 (C) ) is a normed space.

• We have the following relation of strict inclusions,
ℓ1 (C) ⊊ ℓ2 (C) ⊊ ℓ∞ (C).

Solution. • We have seen in Remark 1.3.5 that the set of bounded sequences ℓ∞ (C)
forms a C-vector space. We then verify ∥ · ∥ℓ∞ (C) defined a norm on ℓ∞ (C). Let
u = (un )n∈N and v = (vn )n∈N be sequences in ℓ∞ (C).
– Let α ∈ C. If α = 0, it is immediate αu = 0CN and hence, ∥αu∥ℓ∞ (C) = sup 0 = 0.
n∈N
If α ̸= 0, for all n ∈ N, it follows that |αun | = |α||un | ≤ |α| sup |un | = |α|∥u∥ℓ∞ (C) .
n∈N
Also, since |α| > 0, for all ε > 0, there exists n ∈ N, such that ∥u∥ℓ∞ (C) −
ε
≤ |un | ≤ ∥u∥ℓ∞ (C) . So, |α|∥u∥ℓ∞ (C) − ε ≤ |α||un | ≤ |α|∥u∥ℓ∞ (C) . Using the
|α|
characterization of supremum, we conclude with sup |αun | = |α|∥u∥ℓ∞ (C) . Thus,
n∈N
the homogeneity follows.
– For all n ∈ N, it follows from the triangular inequality in the normed space
(C, |·|), we have |un +vn | ≤ |un |+|vn | ≤ sup |un |+sup |vn | = ∥u∥ℓ∞ (C) +∥v∥ℓ∞ (C) .
n∈N n∈N
It implies that the number ∥u∥ℓ∞ (C) + ∥v∥ℓ∞ (C) ≥ 0 is an upper bound of the
sequence (un + vn )n∈N = u + v. Hence, ∥u + v∥ℓ∞ (C) ≤ ∥u∥ℓ∞ (C) + ∥v∥ℓ∞ (C) and
the triangular inequality follows.
– If u = 0CN , then ∥u∥ℓ∞ (C) = sup{0, 0, · · · } = 0. Reciprocally, if ∥u∥ℓ∞ (C) = 0,
n∈N
then sup |un | = 0, it implies that for all n ∈ N, un = 0. Thus the separation
n∈N
property follows.
• We have seen in Lemma 2.3.6 that the set of absolutely convergent series form a C-
vector space. We then verify ∥ · ∥ℓ1 (C) defined a norm on ℓ1 (C). Let u = (un )n∈N and
v = (vn )n∈N be sequences in ℓ1 (C).
– Let α ∈ C, it follows from the linearity of the summation that ∥αu∥ℓ1 (C) =
+∞
X +∞
X
|αun | = |α| |un | = |α|∥u∥ℓ1 (C) . Thus the homogeneity follows.
n=0 n=0
– By using the triangular inequality in (C, | · |), we have
+∞
X +∞
X +∞
X +∞
X
∥u + v∥ℓ1 (C) = |un + vn | ≤ (|un | + |vn |) ≤ |un | + |vn |
n=0 n=0 n=0 n=0
= ∥u∥ℓ1 (C) + ∥v∥ℓ1 (C) .
+∞
X
xiii
Same remark, we might write “ |un |2 < +∞”.
n=0
110 TOPOLOGY - ELEMENTARY INTRODUCTION

These inequalities hold since each involved series are convergent. Then, the
triangular inequality follows.
+∞
X
– If u = 0CN , then ∥u∥ℓ1 (C) = 0 = 0. Reciprocally, if ∥u∥ℓ1 (C) = 0, then for
n=0
+∞
X
all n ∈ N, |un | ≤ |un | = 0, it implies that for all n ∈ N, un = 0. Thus the
n=0
separation property follows.

• We first show that the set ℓ2 (C) forms a C-vector space. Let u = (un )n∈N and
v = (vn )n∈N be sequences in ℓ2 (C) as well as let λ ∈ C. For all n ∈ N, we have

|un + λvn |2 = |un |2 + |λ|2 |vn |2 + 2Re(λvn un ).

X 3.2.16 and remarking that |Re(λvn un )| ≤ |λ||uX

From the following Lemma n vn |, we
deduce that the series Re(λvn un ) is absolutely convergent. So, the series |un |2 ,
X X n n
|vn |2 and Re(λvn un ) are all convergent, and hence the sequence (un + λvn )n∈N
n n
is square-summable. This shows that ℓ2 (C) is stable by linear combination. In the
consequence, ℓ2 (C) forms a C-vector space as a vector subspace of the space of bounded
sequences ℓ∞ (C).
Next, ∥ · ∥ℓ2 (C) defines a norm on ℓ2 (C) since for all u ∈ ℓ2 (C), we have ∥u∥ℓ2 (C) =
q
∥(|un |2 )n∈N ∥ℓ1 (C) and the three properties follows from those of ∥ · ∥ℓ1 (C) .
X
• Let u = (un )n∈N ∈ ℓ1 (C). Since the series un is absolutely convergent, then
n
|un | −→ 0 and (un )n∈N is bounded. Then, the sequence (|un |2 )n∈N can be considered
n→+∞
as a product of (|un |)n∈N with a bounded X sequence. In other words, we have |un |2 =
O(|un |). By the convergence of the series |un | and the comparison test, the series
X n
|un |2 converges. Thus, u ∈ ℓ2 (C), in the consequences, ℓ1 (C) ⊂ ℓ2 (C). Moreover,
n  
1 X 1
1 2
we have also ℓ (C) ̸= ℓ (C). The sequence ∈ ℓ2 (C) since the series
n n∈N∗ n2
  n≥1
1 X1
converges. But / ℓ1 (C) since the series
∈ diverges.
n n∈N∗ n
n≥1
2 2
X v = (vn )n∈N ∈ ℓ (C), then v is bounded since (|vn | )n∈N is bounded because
Let
|vn |2 converges. So, ℓ2 (C) ⊂ ℓ∞ (C). Moreover, we have also ℓ2 (C) ̸= ℓ∞ (C). The
n
constant sequence (1)n∈N is clearly an example of bounded and not square-summable
sequence.

Lemma 3.2.16. Let ℓ2 (C) denote the set of square-summable sequences of elements in C.
Let u, v ∈ ℓ2 (C), u = (un )n∈N and v = (vn )n∈N . Then, the sequence (un vn )n∈N is summable.
3.2. NORMED SPACES 111

Proof. We begin with recalling the following basic inequality. For all a, b ∈ R+ , 2ab ≤ a2 +b2 .
|un |2 |vn |2
Let u, v ∈ ℓ2 (C), for all n ∈ N, we have |un vn | = |un ||vn | ≤ + . Summing them
2 2
up in the series, for all n ∈ N,
n n n n +∞ +∞
! !
X X 1 X X 1 X X
|uk vk | = |uk ||vk | ≤ |uk |2 + |vk |2 ≤ |un |2 + |vn |2 .
2 2
k=0 k=0 k=0 k=0 n=0 n=0
X
It shows that the sequence of partial sums of the series with positive general terms |un vn |
n
is bounded from above, and hence this series is convergent. In other words, the sequence
(un vn )n∈N is summable, i.e. (un vn )n∈N ∈ ℓ1 (C).

Remark 3.2.17. We can endow the vector space ℓ2 (C) with an inner product ⟨·, ·⟩ℓ2 (C) defined
+∞
X
2
by ∀u, v ∈ ℓ (C), ⟨u, v⟩ℓ2 (C) = un vn . Then, (ℓ2 (C), ⟨·, ·⟩ℓ2 (C) ) is an inner product space.
n=0

Example 3.2.18. Let I ⊂ R be a segment and consider the set C(I, C) if continuous
functions mapping from I into C. We define for all 1 ≤ p < ∞,
Z 1
p
p
∥f ∥Lp (I,C) := |f (x)| dx ,
I

and

∥f ∥L∞ (I,C) = sup |f (x)|.

x∈I

Then, (C(I, C), ∥ · ∥Lp (I,C) ) is a normed space for all 1 ≤ p ≤ ∞.

Solution. Left in Exercise 3.7.2

Proposition 3.2.19. The norms ∥ · ∥Lp (I,C) and ∥ · ∥Lq (I,C) on C(I, C) are not equivalent if
p ̸= q.
Proof. We consider the segment I = [0, 1] and the functions fn : x 7→ xn for all n ∈ N. Let
1 ≤ p < ∞ and n ∈ N, we have,
Z 1 1/p  1/p
np 1
∥fn ∥Lp (I,C) = x dx = ;
0 np + 1
∥fn ∥L∞ (I,C) = 1.

We have then ∥fn ∥Lp (I,C) −→ 0. So, there does not exist a constant C > 0 such that
n→+∞
∥f ∥L∞ (I,C) ≤ C∥f ∥Lp (I,C) holds for all f ∈ C(I, C), the sequences (fn )n∈N provides a counter
example. Hence, ∥ · ∥Lp (I,C) and ∥ · ∥L∞ (I,C) are not equivalent.
Let 1 ≤ p < q < ∞, we have then
 1/q  1/p
1 1
∥fn ∥Lq (I,C) nq+1 (np + 1) 1/p p 1/p 1 + np
= 1/p
= 1/q
= 1/q n1/p−1/q  1/q = An n
1/p−1/q
,
∥fn ∥Lp (I,C) 1

(nq + 1) q 1
np+1 1 + nq
112 TOPOLOGY - ELEMENTARY INTRODUCTION

 1/p
1
p1/p 1+ np p1/p
where An = −→ .
q 1/q
1/q n→+∞
q 1/q

1
1+ nq
∥fn ∥Lq (I,C)
Since p < q ⇒ 1/p − 1/q > 0 ⇒ n1/p−1/q −→ +∞, we have then −→
n→+∞ ∥fn ∥Lp (I,C) n→+∞
+∞. Then, there does not exist any constant C > 0 such that ∥f ∥Lq (I,C) ≤ C∥f ∥Lp (I,C)
holds for all f ∈ C(I, K), the sequences (fn )n∈N provides a counter example. Hence, the
norms ∥ · ∥Lp (I,C) and ∥ · ∥Lq (I,C) are not equivalent.

3.3 Basic Notions

From this section and until the next Chapter, we will introduce the notions about Topology,
such as open, closed, compact, etc. We will denote by (X, dX ) or (Y, dY ) the metric spaces,
and by (E, ∥ · ∥E ) or (F, ∥ · ∥F ) the normed spaces. We want to point out at all beginning
that the notions defined for metric spaces are automatically verified by normed spaces via
Remark 3.2.3.

Definition 3.3.1. Let a ∈ X and r ∈ R∗+ . We call the open ball centered at a and of
radius r the following set,

B(a, r) := {x ∈ X | d(a, x) < r} .

We call the closed ball centered at a and of radius r the following set,

B(a, r) := {x ∈ X | d(a, x) ≤ r} .

We call the sphere centered at a and of radius r the following set,

S(a, r) := {x ∈ X | d(a, x) = r} .

In fact, S(a, r) = B(a, r) \ B(a, r). Moreover, we call the unit sphere the sphere S(0E , 1)
in a normed space (E, ∥ · ∥E ).

Example 3.3.2. Describe the open balls in R2 respect to the norms ∥ · ∥∞ , ∥ · ∥1 and ∥ · ∥2 .

Solution. See in the oral course.

Definition 3.3.3. We say a set A ⊂ X is bounded if A is contained in an open ball.

Remark 3.3.4. The following results are direct consequences of the definition of bounded
sets.

• A set A is bounded if, and only if for all a ∈ X, there exists r > 0 such that
A ⊂ B(a, r).

• A set A in a normed space E is bounded if, and only if there exists M > 0 such that
for all x ∈ A, ∥x∥ ≤ M .
3.3. BASIC NOTIONS 113

• Finite sets, all subsets in a bounded set and the union of a finite number of bounded
sets are bounded.

Definition 3.3.5. We say a sequence (xn )n∈N of elements in the metric space (X, d) con-
verges to l ∈ X if the sequence of real numbers (d(xn , l))n∈N converges to 0.
We can also define the convergence using the ε − N notations. xn −→ l if, and only if
n→+∞

∀ε > 0, ∃N ∈ N s.t. ∀n ∈ N, n ≥ N ⇒ d(xn , l) ≤ ε.

We say a sequence diverges if it does not converge. In the case of convergence, the
element l ∈ X such that (xn )n∈N converges to it is called a limit of (xn )n∈N .
Proposition 3.3.6. Any convergent sequence is bounded.
Proof. Left in Exercises.

Theorem 3.3.7 (Uniqueness of the limit). Let (xn )n∈N be a convergent sequence of elements
in the metric space (X, d), then (xn )n∈N converges to a unique limit l ∈ X.
Proof. Left in Exercises

Xa normed space (E, ∥ · ∥). We

Definition 3.3.8. Let (xn )n∈N be a sequence of elements in
call the series of general terms xn , and we denote it by xn , the sequence of partial
n
sums (sn )n∈N defined by,
n
X
∀n ∈ N, sn = xk ∈ E.
k=0

X
Definition 3.3.9. We say the series xn is convergent if the sequence of partial sums
n
(sn )n∈N is convergent. In the case of convergence, the limit of (sn )n∈N is called the sum of
X +∞
X
the series xn , and we denote it by xn .
n n=0
Let p ∈ N∗
and for all i ∈ [[1, p]], we denote by (Xi , dXi ) a metric space. We consider now
Yp
the Cartesian product X = Xi . For any element x ∈ X, we denote x = (x1 , x2 , · · · , xp )
i=1
where for all i ∈ [[1, p]], xi ∈ Xi is called as the i-th component of x.
Definition 3.3.10. Let dX be the function defined as follows,
∀x, y ∈ X, dX (x, y) = max dXi (xi , yi ).
1≤i≤p

Then, dX is a distance on X.
We call X as the product metric space of the spaces (Xi , dXi )1≤i≤p .

Proposition 3.3.11. A sequence (x(k) )k∈N in the product metric space X is convergent if,
(k)
and only if each component sequence (xi )k∈N converges for all i ∈ [[1, p]].
Proof. Left in Exercises.
114 TOPOLOGY - ELEMENTARY INTRODUCTION

3.4 Open Sets, Closed Sets and Neighborhoods

Definition 3.4.1. Let U be a subset in the metric space (X, d), we say that U is an open
set, or that U is open, if for all x ∈ U , there exists r > 0 such that B(x, r) ⊂ U .

Remark 3.4.2. • By definition, ∅ and X are open sets.

• In the above definition of open sets, the open ball B(x, r) can be replaced by a closed
ball B(x, r). The justification of this point is left in Exercises.

Proposition 3.4.3. • Any open ball is an open set.

• Any open interval (a, b) ⊂ R is an open set in the normed space of real numbers
endowed with the absolute values function (R, | · |).
Proof. Left in Exercises.

Proposition 3.4.4. The open sets in X satisfy the following assertions.

1.) ∅ and X are open sets.
2.) The union of any family of open sets is open.
3.) The intersection of a finite number of open sets is open.
Proof. 1.) Straightforward from definition.
2.) Let (U
[i )i∈I be a family of open sets in (X, d) where I denotes the set of indexes. Let
x∈ Ui then there exists i0 ∈ I such that x ∈ Ui0 . Since Ui0 is open, there exists
i∈ I [ [
r > 0 such that B(x, r) ⊂ Ui0 ⊂ Ui . Thus, is an open set.
i∈I i∈I
\
3.) Let (Ui )i∈[[1,N ]] be N open sets in (X, d) where N ∈ N∗ . Let x ∈ Ui , then
1≤i≤N
x ∈ Ui for all i ∈ [[1, N ]]. Since each Ui is an open set, there exists ri > 0 such that
B(x, ri ) ⊂ Ui . Let i0 ∈ [[1, N ]] be the index such that ri0 = min ri , we have then
1≤i≤N
\
B(x, ri0 ) ⊂ B(x, ri ) ⊂ Ui for all i ∈ [[1, N ]]. In other words, B(x, ri0 ) ⊂ Ui .
1≤i≤N
\
Thus, Ui is an open set.
1≤i≤N

Remark 3.4.5. No matter the family of open sets is finite, infinite, even uncountable, the
union is always open.

Remark 3.4.6. Let S be a set. We call a topology on the set S any collection T ⊂ ℘(S) of
all subsets satisfying the three analogue assertions in Proposition 3.4.4. Precisely speaking,
T-1.) ∅ ∈ T and S ∈ T ;
[
T-2.) T is stable by any union, i.e., for any subsetxiv J ⊂ T , U ∈T;
U ∈J
[
xiv
In the case where J = ∅, it follows by the convention that = ∅.
∅
3.4. OPEN SETS, CLOSED SETS AND NEIGHBORHOODS 115

T-3.) T\is stable by any finite intersection, i.e., let N ∈ N∗ , if ∀i ∈ [[1, N ]], Ui ∈ T then
Ui ∈ T .
1≤i≤N

We then define the topological space (S, T ) composed with the set S together with
its topology T . In the context of metric spaces, the topology is defined via the distance
function d and then by the open balls B(x, r).

Proposition 3.4.7. Let ∥ · ∥1 and ∥ · ∥2 be two equivalent norms on the vector space E,
then the open sets in the normed space (E, ∥ · ∥2 ) are also open in (E, ∥ · ∥1 ), and vice-versa.
In other words, the equivalent norms define the same topology.
Proof. Let x ∈ E and r ∈ R∗+ , we shall denote respectively by B∥·∥1 (x, r) an by B∥·∥2 (x, r)
the open balls in (E, ∥·∥1 ) and in (E, ∥·∥2 ) centered at x with a radius r. Since the norms ∥·∥1
and ∥ · ∥2 are equivalent, there exist c, C > 0 such that for all z ∈ E, c∥z∥1 ≤ ∥z∥2 ≤ C∥z∥1 .
1
For all y ∈ B∥·∥1 (x, r), we have ∥y − x∥1 < r and hence, ∥y − x∥2 ≤ ∥y − x∥1 < r. This
C
implies that y ∈ B∥·∥2 (x, Cr). In other words, B∥·∥1 (x, r) ⊂ B∥·∥2 (x, Cr). Symmetrically,
r
we have also that B∥·∥2 (x, r) ⊂ B∥·∥1 (x, ).
c
Let U ⊂ E be an open set for ∥ · ∥2 , there exists r2 > 0 such that B∥·∥2 (x, r2 ) ⊂ U and
r2
hence, B∥·∥1 (x, ) ⊂ B∥·∥2 (x, r2 ) ⊂ U . Then U is an open set for ∥ · ∥1 . Symmetrically,
C
if U be an open set for ∥ · ∥1 , there exists r1 > 0 such that B∥·∥2 (x, r1 ) ⊂ U and hence,
B∥·∥2 (x, cr1 ) ⊂ B∥·∥1 (x, r1 ) ⊂ U . Then U is an open set for ∥ · ∥2 .

Definition 3.4.8. Let U be a subset in the metric space (X, d), we say U is a closed set,
or we say U is closed, if the complement U ∁ = X \ U of U is an open set.

Remark 3.4.9. By definition, ∅ and X are closed sets.

Proposition 3.4.10. • Any singleton is closed.

• Any closed ball is closed.
Proof. • Let a ∈ X, we consider the set {a} ⊂ X. For all x ∈ {a}∁ , x ∈ X and x ̸= a.
From the separation property of the distance function d, we have that d(x, a) > 0.
1
Then, for all y ∈ B(x, d(x, a)), it follows from the second triangular inequality, we
2
1 1
have d(y, a) ≥ |d(x, a) − d(x, y)| ≥ d(x, a) − d(x, a) = d(x, a) > 0. This inequality
2 2
1 ∁
implies that y ̸= a and hence, B(x, d(x, a)) ⊂ {a} . So, {a}∁ is an open set and thus
2
{a} is a closed set.
• Let x0 ∈ X and let r ∈ R∗+ , we consider the closed ball B(x0 , r) ⊂ X. From the
definition of closed balls, the complement of B(x0 , r) can be expressed as, B(x0 , r)∁ =
{y ∈ X | d(y, x0 ) > r}. Let y ∈ B(x0 , r)∁ , we can write d(y, x0 ) = r + t with t > 0.
t
Let z ∈ B(y, ), it follows from the second triangular inequality that d(z, x0 ) ≥
2
t t t
|d(y, x0 ) − d(z, y)| = r + t − = r + > r. Then, B(y, ) ⊂ B(x0 , r)∁ and hence,
2 2 2
B(x0 , r)∁ is open. So, the closed ball B(x0 , r) is closed.
116 TOPOLOGY - ELEMENTARY INTRODUCTION

Proposition 3.4.11. The closed sets in X satisfy the following assertions.

1.) ∅ and X are closed sets.
2.) The union of a finite number of closed sets is closed.
3.) The intersection of any family of closed sets is closed.
Proof. 1.) Straightforward from definition.
2.) Let (Ui )i∈[[1,N ]] be N closed subsets in X where N ∈ N∗ . Then for all i ∈ [[1, N ]],
 ∁
[ \
Ui∁ is an open set in X. Using relationxv  Ui  = Ui∁ , we have that
1≤i≤N 1≤i≤N
 ∁
[
 Ui  is an open set as the intersection of a finite numbers of open sets
1≤i≤N
[
(Ui∁ )i∈[[1,N ]] . Thus, Ui is a closed set.
1≤i≤N

3.) Let (Ui )i∈I be a family of closed subsets in X where I denotes the set of indexes.
!∁
\ [
Then for all i ∈ I, Ui∁ is an open set in X. Using relation Ui = Ui∁ , we
i∈I i∈I
!∁
\ \
have that Ui is an open set as the union of open sets (Ui∁ )i∈I . Thus, Ui is
i∈I i∈I
a closed set.

Proposition 3.4.12 (Sequence Characterization of Closed Sets). A subset U in X is closed

if, and only if any convergent sequence of elements in U possesses its limit in U .
Proof. We prove this proposition by double implications.
⇒: Let U be a closed set. We consider a convergent sequence (xn )n∈N of elements in U .
Let lim xn = l ∈ X and we assume in the contrary that l ∈ U ∁ . Since U ∁ is an
n→+∞
open set, there exists ε0 > 0 such that B(l, ε0 ) ⊂ U ∁ . From the convergence of the
sequence (xn )n∈N , there exists N ∈ N such that for all n ∈ N, n ≥ N ⇒ d(xn , l) ≤ ε0 .
In other words, xn ∈ B(l, ε0 ) ⊂ U ∁ from certain rank. This is contradictory to the
fact that (xn )n∈N is composed by elements in U . Thus, l ∈ U .
xv
Here we use De Morgan’s Law, which states the following theorem. Let B be a set and let (Ai )i∈I
be a family of sets where I denotes the set of index. Then it holds,
\ [ [ \
B\ Ai = (B \ Ai ), and B \ Ai = (B \ Ai ).
i∈I i∈I i∈I i∈I

With the notations of complements, we have

!∁ !∁
\ [ ∁ [ \
Ai = Ai , and Ai = A∁i .
i∈I i∈I i∈I i∈I

The proof of this theorem can be obtained directly from the definitions of unions and intersections. We will
use systematically this theorem and call it “passing to the complements”.
3.4. OPEN SETS, CLOSED SETS AND NEIGHBORHOODS 117

⇐: We suppose that any convergent sequence (xn )n∈N of elements in U possesses its limit
1
in U . Let y ∈ U ∁ , we consider the sets B(y, ) ∩ U for all n ∈ N∗ . We assume
n
1
in the contrary that for all n ∈ N∗ , B(y, ) ∩ U ̸= ∅. Then we can construct a
n
∗ 1
sequence (xn )n∈N such that for all n ∈ N , xn ∈ B(y, ) ∩ U . In other words, (xn )n∈N
n
is a sequence of elements in U and it converges to y. From the property that any
convergent sequence of elements in U possesses its limit in U , we obtain that y ∈ U .
1
Thus, the contradiction. We then deduce that B(y, ) ∩ U = ∅ from certain rank.
n
1
That means, there exists N ∈ N∗ such that B(y, ) ⊂ U ∁ . So, U ∁ is an open set,
N
which is equivalent to say that U is a closed set.

Example 3.4.13. • Let (xn )n∈N be a sequence of elements in X which converges to

l ∈ X. Then the set A := {xn | n ∈ N} ∪ {l} is closed.

• The set B := (x, y) ∈ R2 xy = 1 is a closed set in (R2 , ∥ · ∥∞ ).



Solution. • Let y ∈ A∁ . Since y = ̸ l, we set ε0 = d(y, l) > 0. From the convergence

ε0
of (xn )n∈N , there exists N ∈ N such that xn ∈ B(l, ) for all n ≥ N . Then, for all
2
ε0 ε0
n ∈ N, n ≥ N , we have d(xn , y) ≥ |d(xn , l) − d(y, l)| ≥ ε0 − = > 0. We then
2 2
1
set r = min{d(x0 , y), · · · , d(xN −1 , y), ε0 } > 0 since y ∈ A∁ . We have then for all
3
1
n ∈ N, if n ∈ [[0, N − 1]], we have d(xn , y) > d(xn , y) ≥ r, if n ≥ N , we have also
3
ε0 ε0
d(xn , y) ≥ > = r. They imply the fact that A ∩ B(y, r) = ∅. And hence,
2 3
B(y, r) ⊂ A∁ . It shows that A∁ is an open set and therefore A is a closed set.

• Consider a sequence (xn , yn )n∈N of elements in B. We assume that this sequence

converges to (lx , ly ) ∈ R2 in (R2 , ∥ · ∥∞ ). Then the sequences of each component
(xn )n∈N and (yn )n∈N converge in R respectively to lx and ly . On the other hand, we
have also for all n ∈ N, xn yn = 1 since (xn , yn ) ∈ B. Thus,

lx ly = ( lim xn )( lim yn ) = lim xn yn = lim 1 = 1.

n→+∞ n→+∞ n→+∞ n→+∞

It shows that (lx , ly ) ∈ B. From the sequence characterization of closed sets, B is a

closed set.

Definition 3.4.14. Let a ∈ X, we say a subset V in X is a neighborhood of a if there

exists an open set U in X such that a ∈ U ⊂ V . We denote by N (a) the collection of all
neighborhoods of a.

Remark 3.4.15. Let a ∈ X, we consider the neighborhoods of a, N (a) ⊂ ℘(X).

N-1.) For all U ∈ N (a), a ∈ U .
118 TOPOLOGY - ELEMENTARY INTRODUCTION

N-2.) If U ⊂ V and U ∈ N (a), then V ∈ N (a).

N-3.) For all U, V ∈ N (a), U ∩ V ∈ N (a).

N-4.) For all U ∈ N (a), there exists V ∈ N (a) such that for all b ∈ V , U ∈ N (b).

Proposition 3.4.16. Let (xn )n∈N be a sequence of elements in X and l ∈ X. Then,

(xn )n∈N converges to l if, and only if

∀V ∈ N (l), ∃N ∈ N s.t. ∀n ∈ N, n ≥ N ⇒ xn ∈ V.

Proof. We prove this proposition by double implications.

⇒: We assume that xn −→ l. Let V be a neighborhood of l, there exists an open set U

n→+∞
such that a ∈ U ⊂ V . And then there exists ε > 0 such that B(l, ε) ⊂ U ⊂ V . From
the convergence of (xn )n∈N , there exists N ∈ N such that d(xn , l) ≤ ε once n ≥ N . In
other words, for all n ≥ N , xn ∈ B(l, ε) ⊂ V .

⇐: It is sufficient to choose V = B(l, ε) for ε > 0.

3.5 Interior and Closure

Let A be a subset of (X, d).

Definition 3.5.1. We say x ∈ X be an interior point of A if A is a neighborhood of x.

We call the interior of A, and we denote it by Å or int(A) , the collection of all interior
points of A. Namely,

Å = int(A) = {x ∈ A | ∃U ⊂ A, U is open s.t. x ∈ U } .

Proposition 3.5.2. The interior of A is the largest open set contained in A.

Proof. Let x ∈ int(A), from the definition of the interior, A is a neighborhood of x. Then
there exists U being an open set such that x ∈ U ⊂ A. Since U is an open set, there exists
r > 0 such that B(x, r) ⊂ U . Therefore, for all y ∈ B(x, r), y ∈ B(x, r) ⊂ U ⊂ A. It implies
that A is also a neighborhood of y, then y ∈ int(A). In the consequences, B(x, r) ⊂ int(A).
Thus, int(A) is an open set.
Let U ⊂ A be an open set containing in A. For all x ∈ U , there exists r > 0 such that
B(x, r) ⊂ U ⊂ A, which means that A is a neighborhood of x. So, x ∈ int(A) and hence,
U ⊂ int(A).
In conclusion, int(A) is the largest open set contained in A.

Remark 3.5.3. We[can also deduce that int(A) is the union of all open sets contained in
A, i.e. int(A) = U where O(A) = {U ⊂ A | U is open} ⊂ ℘(A). The proof is left in
U ∈O(A)
Exercises.
3.5. INTERIOR AND CLOSURE 119

Corollary 3.5.4. Let A and B be subsets in X. Then the following assertions hold.

1.) A is an open set if, and only if A = int(A).

2.) int(∅) = ∅.

3.) A ⊂ B ⇒ int(A) ⊂ int(B).

4.) int(A) ∪ int(B) ⊂ int(A ∪ B).

5.) int(A ∩ B) = int(A) ∩ int(B).

Proof. 1.) We prove this point by double implications.

⇒: We assume now A is an open set. It is obvious that A is an open set contained in

A and if U is an open set contained in A, then U ⊂ A. In other words, A is the
largest open set contained in A. From Proposition 3.5.2, we have int(A) = A.
⇐: We assume that A = int(A). It is shown in Proposition 3.5.2 that int(A) is an
open set, and hence A is an open set.

2.) Since ∅ is an open set and the fact that if U ⊂ ∅, then U = ∅. Then the claim
follows.

3.) From int(A) ⊂ A ⊂ B, we have that int(A) is an open set contained in B. Using
Proposition 3.5.2 it follows that, int(A) ⊂ int(B).
We remark that the reciprocal statement is false. Taking for example A = [0, 1) ∪ {3}
and B = (−1, 2]. Then we have int(A) = (0, 1) and int(B) = (−1, 2), which imply
that int(A) ⊂ int(B). But we don’t have A ⊂ B.

4.) The relations int(A) ⊂ A and int(B) ⊂ B imply that int(A) ∪ int(B) ⊂ A ∪ B. Then,
the set int(A) ∪ int(B) is an open set (as the union of two open sets) contained in
A ∪ B. From Proposition 3.5.2, we have int(A) ∪ int(B) ⊂ int(A ∪ B).
We remark that the inclusion here could be strict. Taking for example A = (0, 1] and
B = [1, 2). We have thus int(A) = (0, 1) and int(B) = (1, 2) as well as A ∪ B = (0, 2).
So, int(A) ∪ int(B) = (0, 1) ∪ (1, 2) = (0, 2) \ {1} =
̸ (0, 2) = int(A ∪ B).

5.) Let x ∈ int(A ∩ B), then there exists an open set U such that x ∈ U ⊂ A ∩ B.
We have thus, U ⊂ A and U ⊂ B and then, x ∈ int(A) and x ∈ int(B). Hence,
x ∈ int(A) ∩ int(B). It implies that int(A ∩ B) ⊂ int(A) ∩ int(B).
Reciprocally, the relations int(A) ⊂ A and int(B) ⊂ B imply that int(A) ∩ int(B) ⊂
A ∩ B. Then, the set int(A) ∩ int(B) is an open set (as the intersection of two open
sets) contained in A∩B. From Proposition 3.5.2, we have int(A)∩int(B) ⊂ int(A∩B).
In conclusion, int(A ∩ B) = int(A) ∩ int(B).

Example 3.5.5. Let (E, ∥ · ∥) be a normed space and F be a vector subspace of E. If

F ̸= E, then int(F ) = ∅.
120 TOPOLOGY - ELEMENTARY INTRODUCTION

Solution. We assume in the contrary that int(F ) ̸= ∅. Let a ∈ int(F ), there exists then
ε > 0 such that B(a, ε) ⊂ int(F ) since int(F ) being an open set. It is obvious that for all
x ∈ B(0E , ε), x = x+a−a where x+a ∈ B(a, ε). Since F is a vector space and B(a, ε) ⊂ F ,
∥y∥ εy
we have that B(0E , ε) ⊂ F . Then, for all y ∈ E, y ̸= 0E , we can write y = where
ε ∥y∥
εy
∈ B(0E , ε) ⊂ F . So, y ∈ F and it implies that F = E. Thus the contradiction, we
∥y∥
conclude that int(F ) = ∅.

Definition 3.5.6. We say x ∈ X be an adherent point or a closure point of A if for

any neighborhood V of x, the intersection V ∩ A is not empty. Furthermore, we call the
closure of A, and we denote it by A or cl(A) the collection of adherent points of A.

Proposition 3.5.7. Let A be a subset in X. Then, the following assertions hold.

1.) cl(A) is a closed set;

2.) cl(A)∁ = int(A∁ );

3.) cl(A) is the smallest closed set containing A.

Proof. 1.) Let x ∈ cl(A)∁ , from the definition, there exists a neighborhood V of x such
that V ∩ A = ∅. We deduce then that there exists U being open and r > 0 such that
B(x, r) ⊂ U ⊂ V , which implies that any point in B(x, r) possesses a neighborhood
V which has an empty intersection with A. So, B(x, r) ⊂ cl(A)∁ and hence cl(A)∁ is
an open set. Passing to the complements, cl(A) is closed.

2.) From the definition, we have immediately that any point in A is a closure point, i.e.,
A ⊂ cl(A). Passing to the complements, it holds that cl(A)∁ ⊂ A∁ . The previous
point indicates that cl(A) is closed, which is equivalently to say cl(A)∁ is open. Then,
cl(A)∁ is an open set contained in A∁ .
Consider U being an open set contained in A∁ . For all y ∈ U , there exists r′ > 0 such
that B(y, r′ ) ⊂ U ⊂ A∁ . It implies also that B(y, r′ ) ∩ A = ∅. Then, y ∈ cl(A)∁ and
hence U ⊂ cl(A)∁ . In conclusion, cl(A)∁ is the largest open set contained in A∁ . From
Proposition 3.5.2, cl(A)∁ = int(A∁ ).

3.) We have shown in the first point that cl(A) is a closed set. Let U be a closed set
containing A, that is, A ⊂ U . Passing to the complements, we have that U ∁ ⊂ A∁ and
hence U ∁ is an open set contained in A∁ . So, U ∁ ⊂ int(A∁ ) = cl(A)∁ . Passing again
to the complements, we obtain cl(A) ⊂ U .

Corollary 3.5.8. Let A and B be subsets in X. Then, the following assertions hold.

1.) A is a closed set if, and only if A = cl(A).

2.) cl(∅) = ∅.

3.) A ⊂ B ⇒ cl(A) ⊂ cl(B).

3.5. INTERIOR AND CLOSURE 121

4.) cl(A) ∪ cl(B) = cl(A ∪ B).

5.) cl(A ∩ B) ⊂ cl(A) ∩ cl(B).

Proof. They can be obtained by passing to the complements in each statements of Corol-
lary 3.5.4.

Proposition 3.5.9. A point x is a closure point of A if, and only if there exists a sequence
(an )n∈N of elements in A which converges to x.

Proof. We prove this proposition by double implications.

1
⇒: Let x ∈ cl(A), we take successively V = B(x, ) as a neighborhood of x for all
n
1
n ∈ N . From the definition of cl(A), B(x, ) ∩ A ̸= ∅ for all n ∈ N∗ . We can thus
∗
n
1
chose an ∈ B(x, ) ∩ A and the sequence (an )n∈N forms a sequence of elements in A
n
which converges to x.

⇐: Let x ∈ X such that x = lim an where (an )n∈N denotes a sequence of elements in A.
n→+∞
Let V be a neighborhood of x, there exists then r > 0 such that B(x, r) ⊂ V . From
the convergence of the sequence (an )n∈N , there exists N ∈ N such that an ∈ B(x, r)
once n ≥ N . Hence, aN ∈ B(x, r) ∩ A ⊂ V ∩ A, which implies that V ∩ A ̸= ∅. So,
x ∈ cl(A).

Proposition 3.5.10. Let x ∈ X, then x ∈ cl(A) if, and only if d(x, A) = 0.

Proof. Recalling Proposition 3.5.9, x ∈ cl(A) if, and only if there exists a sequence (an )n∈N of
elements in A such that an −→ x. We then prove this proposition by double implications.
n→+∞

⇒: Let x ∈ cl(A), we consider the sequence (an )n∈N described above. Then 0 ≤ d(x, A) =
inf d(x, a) ≤ d(x, an ) for all n ∈ N. From the convergence an −→ x, then
a∈A n→+∞
d(x, an ) −→ 0. Thus, 0 ≤ d(x, A) ≤ inf d(x, an ) = 0, implies that d(x, A) = 0.
n→+∞ n∈N

⇐: Let x ∈ X such that d(x, A) = 0. Recalling the definition of the distance d(x, A) =
inf d(x, y) and the characterization of the infimum, we have that for all ε > 0, there
y∈A
1
exists a ∈ A such that 0 = d(x, A) ≤ d(x, a) < 0+ε. Taking for example ε = −→ 0,
n n→∞
1
we deduce that for all n ∈ N∗ , there exists an ∈ A such that d(x, an ) ≤ . We have
n
thus constructed a sequence (an )n∈N of elements in A which converges to x. Thus,
x ∈ cl(A).

Definition 3.5.11. We say a subset A ⊂ X is dense in (X, d) if cl(A) = X.

 122 TOPOLOGY - ELEMENTARY INTRODUCTION

Proposition 3.5.12. The following assertions are equivalent.

1.) A is dense in (X, d).
2.) For all x ∈ X and δ > 0, there exits a ∈ A such that d(x, a) ≤ δ.
3.) For all x ∈ X, there exists a sequence (an )n∈N ∈ AN which converges to x.
4.) X is the only closed set containing A.
5.) Any non empty open set in X has a non empty intersection with A.
Proof. We will proceed the proof as follows: 1.) ⇒ 2.) ⇒ 3.) ⇒ 4.) ⇒ 5.) ⇒ 1.).
1.) ⇒ 2.): It is sufficient to apply the definition of the closures and take B(x, δ) as the considered
neighborhood.
1
2.) ⇒ 3.): It is sufficient to take for example δ = for all n ∈ N∗ .
n
3.) ⇒ 4.): Let U be a closed set containing A. Let x ∈ X, it follows from 3.) that there exists
(an )n∈N of elements in A such that an −→ x. We observe also that for all n ∈ N,
n→+∞
an ∈ A ⊂ U , so, (an )n∈N is also a convergent sequence of elements in U . Applying
the sequence characterization of closed sets, we have x ∈ U . Hence, X ⊂ U and then
U = X.
4.) ⇒ 5.): Let U be an open set such that U ∩ A = ∅, which means also U ⊂ A∁ . Passing to
the complements, we have that A ⊂ U ∁ and U ∁ being a closed set. From 4.), we have
that U ∁ = X and hence U = ∅. And the assertion 5.) follows by contraposition.
5.) ⇒ 1.): Follows directly from the definition of the closures.

Example 3.5.13. The set Q of rational numbers is dense in R, the set R \ Q of irrational
numbers is also dense in R.

Definition 3.5.14. • We call the boundary of A the subset ∂A := cl(A) \ int(A).

• We call the exterior of A the subset int(A∁ ), and we denote it by ext(A).

Proposition 3.5.15. The following three sets int(A), ∂A and ext(A) are respectively open,
closed and open. In addition, they form a partitionxvi of X.
Proof. Left in Exercises.

Example 3.5.16. • In a non-empty normed space, the open balls B(x, r) and closed
balls B(x, r) have their boundary:

∂B(x, r) = ∂B(x, r) = S(x, r) = B(x, r) \ B(x, r).

In a general metric space, this result is in general false.

xvi
We say a partition of X any family of disjoint subsets in X of which the union equals to X.
3.6. LIMITS AND CONTINUITY 123

• In (R, | · |), ∂Q = R since Q = R and Q̊ = ∅.

Let A be a non-empty subset in X. In the following proposition, we want to study the

metric space (A, d|A ) where d|A is the restriction of the distance d on A. We call d|A as
the induced distance on A and (A, d|A ) is called as the induced metric space on A of
(X, d).
Proposition 3.5.17. Let V ⊂ A. Then V is an open (resp. closed) set in the space (A, d|A )
if, and only if there exists an open (resp. closed) set U ⊂ X such that V = U ∩ A.
Proof. We prove this proposition only for the case where V is open. The case where V is
closed can be obtained from the cases of open sets and passing them to the complements.
We will denote by BA (x, r) to represent the open balls in the metric space (A, d|A ). Let
V is an open set in (A, d|A ) and x ∈ V . Then there exists rx > 0 such that BA (x, rx ) ⊂ V .
Consider the set U defined by,
[
U= B(x, rx ).
x∈V

Then the set U is an open set in (X, d) as an union of open balls. And we have V ⊂ U ∩ A.
In parallel, we have also,
[
U ∩A⊂ (B(x, rx ) ∩ A) ⊂ V.
x∈V

We obtain thus V = U ∩ A.
Reciprocally, Let V = U ∩ A with U being an open set in (X, d). Let x ∈ V , we have in
particular x ∈ U . Then there exists r > 0 such that B(x, r) ⊂ U . In fact,

B(x, r) ∩ A = {y ∈ X | d(x, y) < r} ∩ A = {y ∈ A | d|A (x, y) < r} = BA (x, r).

Thus, BA (x, r) ⊂ V and hence V is an open set in (A, d|A ).

3.6 Limits and Continuity

In this section, (X, dX ) and (Y, dY ) denote the metric spaces. We consider A ⊂ X be a
non-empty subset in X and f is a function mapping from A into Y .

3.6.1 Limits and Continuity at a Point

Definition 3.6.1. Let a ∈ cl(A), we say f tends to an element l ∈ Y at the point a, if

∀ε > 0, ∃δ > 0 s.t. ∀x ∈ A, dX (x, a) ≤ δ ⇒ dY (f (x), l) ≤ ε.

This definition can be reformulate using the neighborhoods.

Proposition 3.6.2. The function f tends to l at a if, and only if

∀V ∈ N (l), ∃U ∈ N (a) s.t. U ∩ A ⊂ f −1 (V ).

Proof. Left in Exercises

124 TOPOLOGY - ELEMENTARY INTRODUCTION

Theorem 3.6.3 (Uniqueness of the limit). If f tends to both l1 and l2 at a, then l1 = l2 .

Proof. Let ε > 0, since f tends to l1 at a, there exists δ1 > 0 such that for all x ∈ A,
ε ε
dX (x, a) ≤ δ1 ⇒ dY (f (x), l1 ) ≤ . Similarly, we have dX (x, a) ≤ δ2 ⇒ dY (f (x), l2 ) ≤ for
2 2
certain δ2 > 0. Then, for x ∈ A ∩ BX (x, min{δ1 , δ2 }), we have,
ε ε
dY (l1 , l2 ) ≤ dY (f (x), l1 ) + dY (f (x), l2 ) ≤ + = ε.
2 2
The above inequality holds for all ε > 0, we can thus conclude that dY (l1 , l2 ) = 0. From
the separation property, l1 = l2 .

Remark 3.6.4. • If f tens to a certain element l at a we say that f possesses a limit

lim f (x) as well as f (x) −→ l.
at a and the limit l is denoted by lim f or x→a
a x→a
x∈A

• In the cases where A ⊂ R we say that f possesses a limit from left at a if f |A∩(−∞,a)
possesses a limit on a. And we denote the limit by lim f or lim f (x) or x→a
lim f (x).
a− x→a−
x∈A x∈A,x<a

• Similarly, we can define the limit from right at a, they are denoted by lim f or
a+
lim f (x) or lim f (x).
x→a
x→a+ x∈A,x>a
x∈A

• If A ⊂ R is an unbounded set from upper (resp. from below), we say that f possesses
a limit l at +∞ (res. at −∞), if

∀ε > 0, ∃u ∈ R s.t. ∀x ∈ A, x ≥ u (resp. x ≤ u) ⇒ dY (f (x), l) ≤ ε.

Definition 3.6.5. We say that the function f is continuous at a ∈ A if f tends to f (a)

at a. Using the ε − δ formulation, that means

∀ε > 0, ∃δ > 0 s.t. ∀x ∈ A, dX (x, a) ≤ δ ⇒ dY (f (x), f (a)) ≤ ε.

Proposition 3.6.6. Using the formulation with neighborhoods, the function f is continuous
at a if, and only if

∀V ∈ N (f (a)), f −1 (V ) ∈ N (a).

Proof. Left in Exercises

Theorem 3.6.7 (Sequence characterization of limits). A function f mapping from A ⊂

(X, dX ) into (Y, dY ) tends to l ∈ Y at a ∈ cl(A) if, and only if for any sequence (xn )n∈N of
elements in A which converges to a, the sequence (f (xn ))n∈N converges to l.
Proof. We prove this theorem by double implications.
⇒: We assume that f tends to l at a. Let (xn )n∈N be a sequence of elements in A which
converges to a. Let ε > 0, from the convergence of f at a, there exists δ > 0 such
that dY (f (x), l) ≤ ε once dX (x, a) ≤ δ. From the convergence of (xn )n∈N , there exists
N ∈ N such that dx (xn , a) ≤ δ once n ≥ N . So, if n ≥ N , we have dX (xn , a) ≤ δ and
thus dY (f (xn ), l) ≤ ε. That means exactly (f (xn ))n∈N converges to l.
3.6. LIMITS AND CONTINUITY 125

⇐: We assume in the contrary that f does not tend to l at a. From the negation of the
convergence definition,

∃ε0 > 0 s.t. ∀δ > 0 ∃xδ ∈ A s.t. dX (xδ , a) ≤ δ and dY (f (xδ ), l) > ε0 .

1
We then take for example δ = in the above statement, then there exists a sequence
n
(xn )n∈N such that xn −→ a and dY (f (xn ), l) > ε0 for all n ∈ N. This is contra-
n→+∞
dictory to the assertion that for any sequence (xn )n∈N converging to a, the sequence
(f (xn ))n∈N converges to l. Hence, the function f tends to l at a.

The following two corollaries are immediate consequences to Theorem 3.6.7. Their
justifications are left in Exercises.

Corollary 3.6.8. If f tends to l at a, then l ∈ cl(f (A)).

Corollary 3.6.9. A function f is continuous at a ∈ A if, and only if for any sequence
converging to a, the sequence (f (xn ))n∈N converges to f (a).

Theorem 3.6.10 (Function compositions). Let f be a function mapping from A ⊂ (X, dX )

into (Y, dY ) and g be a function mapping from B ⊂ (Y, dY ) into (Z, dZ ) such that f (A) ⊂ B.
If f tends to b at a ∈ cl(A) and if g tends to l at b ∈ cl(B), then g ◦ f tends to l at a.

Proof. Let ε > 0, since g(y) −→ l, there exists δ > 0 such that if dY (y, b) ≤ δ, we have
y→b
dZ (g(y), l) ≤ ε. Similarly, since f (x) −→ b, there exist η > 0 such that if dX (x, a) ≤ η,
x→a
we have dY (f (x), b) ≤ δ. Combining those results, for all x ∈ A, if dX (x, a) ≤ η, we have
dY (f (x), b) ≤ δ and then dZ (g(f (x)), l) ≤ ε, which implies that g(f (x)) −→ l.
x→a

Corollary 3.6.11. If f is continuous at a and g is continuous at f (a), then g ◦ f is

continuous at a.

Proof. Direct consequence of Theorem 3.6.10.

3.6.2 Continuity
Definition 3.6.12. Let f be a function mapping from A ⊂ (X, dX ) into (Y, dY ). We say
that f is continuous on A, or f is a continuous function on A, if f is continuous at each
point of A.

Theorem 3.6.13 (Set characterization of continuous functions). A function f mapping

from (X, dX ) into (Y, dY ) is continuous if, and only if the pre-image by f of any open set
U ⊂ (Y, dY ) is an open set in (X, dX ).

Proof. We prove this theorem by double implications.

126 TOPOLOGY - ELEMENTARY INTRODUCTION

⇒: Let f be a continuous function mapping from (X, dX ) into (Y, dY ) and let U ⊂ Y
be an open set. Consider a point x ∈ f −1 (U ), we have immediately that f (x) ∈ U
and there exists ε > 0 such that BY (f (x), ε) ⊂ U . Since f is continuous on X,
f is thus in particular continuous at x. Then there exists δ > 0 such that for all
y ∈ X, dX (y, x) ≤ δ ⇒ dY (f (y), f (x)) ≤ ε. In other words, if y ∈ B X (x, δ) then
f (y) ∈ B Y (f (x), ε) ⊂ U . We obtain thus for all y ∈ B X (x, δ), f (y) ∈ U . Equivalently
speaking, B X (x, δ) ⊂ f −1 (U ). Hence f −1 (U ) is an open set.

⇐: We assume that f −1 (U ) ⊂ X is an open set for any open set U ⊂ Y . Let x ∈ X

and ε > 0. We consider the open ball BY (f (x), ε) ⊂ Y . From the assumption,
f −1 (BY (f (x), ε)) is an open set in X. It is obvious that x ∈ f −1 (BY (f (x), ε)). So,
there exists δ > 0 such that BX (x, δ) ⊂ f −1 (BY (f (x), ε)). This inclusion shows that
dY (f (y), f (x)) < ε once d(y, x) < δ. Thus, f is continuous at x for all x ∈ X, and
therefore f is continuous on X.

Corollary 3.6.14. A function f mapping from (X, dX ) into (Y, dY ) is continuous if, and
only if the pre-image by f of any closed set U ⊂ (Y, dY ) is a closed set in (X, dX ).

Proof. We remark that for all subset U ⊂ Y , then f −1 (Y \ U ) = X \ f −1 (U ). And the

corollary follows directly from Theorem 3.6.13 by passing to the complements.

Proposition 3.6.15. The set of invertible matrices is an open subset in the set of square
matrices.

Mn (K) −→ K
Proof. The function det : defined on the space of square matrices
M 7−→ det(M )
Mn (K) is continuous since it is polynomialxvii respect to the coefficients. Then the invertible
matrices are characterized by the matrices with non zero determinants. In other words,
GL(n) = det−1 (K \ {0}). In addition, the set K \ {0} is an open set. So, it follows from the
set characterization of continuous function that GL(n) is an open set as the pre-image of
the open set K \ {0} by the continuous function det.

Proposition 3.6.16. Let f be a continuous function on A ⊂ (X, dX ) into (Y, dY ) and g

be a continuous function on B ⊂ (Y, dY ) into (Z, dZ ) such that f (A) ⊂ B. Then, g ◦ f is
continuous on A.

Proof. Direct consequence of Theorem 3.6.10.

xvii
To justify this statement, it is sufficient to apply a simple induction on developing the determinant. Or
it can be seen via the formula
X n
Y
det M = ε(σ) mi,σ(i) .
σ∈Sn i=1
3.6. LIMITS AND CONTINUITY 127

Definition 3.6.17. We call a homeomorphism between the metric spaces (X, dX ) and
(Y, dY ) any function f mapping from X into Y such that f is bijective and both f and
f −1 are continuous. If such function f exists, we say that the metric space (X, dX ) is
homeomorphic to (Y, dY ) and vice versa.

Remark 3.6.18. We assume that f is a homeomorphism between X and Y .

• In this case, we call the metric spaces (X, dX ) and (Y, dY ) are homeomorphic or
topologically isomorphic.

• For any open set U ⊂ X, f (U ) is an open set in Y . The same conclusion holds for
closed sets.

• For any open set V ⊂ Y , f −1 (V ) is an open set in X. The same conclusion holds for
closed sets.

To sum up, if the spaces (X, dX ) and (Y, dY ) are homeomorphic between them, then they
have the same topology.

Theorem 3.6.19 (Extension of identities). Let f and g be continuous functions mapping

from (X, dX ) into (Y, dY ). If f and g coincide on a dense subset A of X, then f = g.

Proof. Let x ∈ X, since A is dense in X, there exists a sequence (an )n∈N of elements in A
which converges to x. From the continuity of the functions f and g, it follows f (an ) −→
n→+∞
f (x) as well as g(an ) −→ g(x). Moreover, since f and g coincide on A, we have then
n→+∞
f (an ) = g(an ) for all n ∈ N. Passing n to +∞, we obtain thus f (x) = g(x). Hence, f = g
on X.

We consider now a special case when Y = Y1 ×Y2 ×· · ·×Yp where each (Yi , dYi ), i ∈ [[1, p]]
is a metric space. Let f be a function mapping from A ⊂ (X, dX ) into the product metric
space (Y, dY ) where dY is defined in Definition 3.3.10.

Definition 3.6.20. Let i ∈ [[1, p]], we call the i-th component of the function f each
function fi : A → Yi satisfying ∀x ∈ A, f (x) = (f1 (x), f2 (x), · · · , fp (x)).

Proposition 3.6.21. • A function f : A → Y tends to a limit l = (l1 , l2 , · · · , lp ) ∈ Y

at a ∈ cl(A) if, and only if for any i ∈ [[1, p]], the component fi of the function f tends
to li at a.

• Following the previous point, the function f is continuous on A if, and only if all
components fi are continuous on A.

Proof. • We prove this point by double implications.

⇒: Let i ∈ [[1, p]], we consider the projection function

Y −→ Yi
Pi : .
y = (y1 , · · · , yp ) 7−→ yi
128 TOPOLOGY - ELEMENTARY INTRODUCTION

We now justify that the projection function is continuous on Y . Let ε > 0 and
ỹ ∈ Y . Consider y ∈ Y such that dY (y, ỹ) ≤ ε. From the definition of product
metric space Definition 3.3.10, we have thus in particular

dYi (Pi (y), Pi (ỹ)) = dYi (yi , ỹi ) ≤ max {dYj (yj , ỹj )} = dY (y, ỹ) ≤ ε
1≤j≤p

This inequality shows that Pi (y) −→ Pi (ỹ) holds for all ỹ ∈ Y . Hence, Pi is
y→ỹ
continuous on Y .
Let f be a function A → Y which tends to l at a. It follows from the function
composition Theorem 3.6.10 that for all i ∈ [[1, p]], fi = Pi ◦ f tends to li at a.
⇐: We assume now for all i ∈ [[1, p]], fi (x) −→ li . Let ε > 0, then there exists
x→a
δ1 , · · · , δp ∈ R∗+ such that for all i ∈ [[1, p]], if x ∈ A, dX (x, a) ≤ δi , we have
dYi (fi (x), li ) ≤ ε. We then set δ = min{δ1 , · · · , δp } > 0 and it holds,

∀x ∈ A, dX (x, a) ≤ δ ⇒ dY (f (x), l) = max dYi (fi (x), li ) ≤ ε.

1≤i≤p

Hence, f (x) −→ l.
x→a

• Direct consequence of the previous point.

3.6.3 Uniform and Lipschitz Continuity

Definition 3.6.22. We say a function f is uniformly continuous on A if,

∀ε > 0, ∃δ > 0 s.t. ∀x, y ∈ A, dX (x, y) ≤ δ ⇒ dY (f (x), f (y)) ≤ ε.

As an immediate result, all uniformly continuous functions are continuous.

Definition 3.6.23. Let f be a function which maps from (X, dX ) into (Y, dY ) is a K-
Lipschitz continuous functionxviii with K ∈ R+ if

∀x, y ∈ X, dY (f (x), f (y)) ≤ KdX (x, y).

And we say f is a Lipschitz continuous function if it is K-Lipschitz continuous for some

K ≥ 0.

Example 3.6.24. • From the second triangular inequality, the norm ∥ · ∥ in a normed
space (E, ∥ · ∥) is a 1-Lipschitz continuous function.

• The function x 7→ dX (x, A) is a 1-Lipschitz continuous function on X. It is a direct

consequence of Proposition 3.1.5.

• The function x 7→ x2 defined on R is continuous on but uniformly continuous.

• The function x 7→ x2 defined on the segment [a, b] is uniformly continuous, it is in

fact Lipschitz continuous.
xviii
Rudolf Otto Sigismund Lipschitz (1832 - 1903), German mathematician. He worked on analysis, differ-
ential geometry also on number theory.
3.6. LIMITS AND CONTINUITY 129

√
• The function x 7→ x defined on R+ is uniformly continuous but not Lipschitz con-
tinuous.

Proposition 3.6.25. Any Lipschitz continuous function is uniformly continuous.

Proof. Let f be a K-Lipschitz continuous function mapping from (X, dX ) into (Y, dY )
where K ≥ 0 and (X, dX ), (Y, dY ) denote the metric spaces. Then, for all x, y ∈ X,
dY (f (x), f (y)) ≤ KdX (x, y). If K = 0, we have immediately that for all x, y ∈ X,
dY (f (x), f (y)) = 0 ⇒ f (x) = f (y) from the separation property of dY . Hence f is a
constant function, which is uniformly continuous on X.
ε
We assume now K > 0. Let ε > 0, we then choose δ = . For all x, y ∈ X such
K
that dX (x, y) ≤ δ, we have dY (f (x), f (y)) ≤ KdX (x, y) ≤ Kδ = ε. Thus, f is uniformly
continuous.

Remark 3.6.26. From Proposition 3.6.25, we can see that Lipschitz continuous functions
are particular cases of uniformly continuous functions. In fact, we can describe the concept
of uniform continuity by using the modulus of continuity.

• Precisely speaking, let ω be a continuous positive-valued function on R+ such that

lim ω(ε) = 0. We say the function f : X → Y admits ω as modulus of continuity if
ε→0+

∀x, y ∈ X, dY (f (x), f (y)) ≤ ω(dX (x, y)).

• The K-Lipschitz continuous functions are exactly the functions which admit x 7→ Kx
as the modulus of continuity.

• If ω : x 7→ xα with α ∈ (0, 1), we say such functions are Hölder continuous. That
is,

∀x, y ∈ X, dY (f (x), f (y)) ≤ dX (x, y)α .

Proposition 3.6.27. The composition of two Lipschitz continuous functions is also a Lip-
schitz continuous function.

Proof. Let f be a Lipschitz continuous function mapping from (X, dX ) into (Y, dY ) and g
be a Lipschitz continuous function mapping from (Y, dY ) into (Z, dZ ) where (X, dX ), (Y, dY )
and (Z, dZ ) denote three metric spaces. Then, there exists Kf , Kg ≥ 0 such that for all
x1 , x2 ∈ X, all y1 , y2 ∈ Y we have

dY (f (x1 ), f (x2 )) ≤ Kf dX (x1 , x2 ), and dZ (g(y1 ), g(y2 )) ≤ Kg dY (y1 , y2 ).

So, for all x1 , x2 ∈ X, it holds,

dZ (g ◦ f (x1 ), g ◦ f (x2 )) = dZ (g(f (x1 )), g(f (x2 )))

≤ Kg dY (f (x1 ), f (x2 )) ≤ Kg Kf dX (x1 , x2 ).

It shows that g ◦ f : X → Z is a Kg Kf -Lipschitz continuous function.

 130 TOPOLOGY - ELEMENTARY INTRODUCTION

Proposition 3.6.28. Let u be a linear application from (E, ∥ · ∥E ) into (F, ∥ · ∥F ), then the
following assertions are equivalent.

1.) u is a Lipschitz continuous function.

2.) ∃K ≥ 0 s.t. ∀x ∈ E, ∥u(x)∥F ≤ K∥x∥E .

3.) ∃r > 0 and K ≥ 0 s.t. ∀x ∈ E, ∥x∥E = r ⇒ ∥u(x)∥F ≤ K

Proof. We will prove the implications 1.) ⇒ 2.) ⇒ 3.) ⇒ 1.).

1.) ⇒ 2.): Assuming that u is Lipschitz continuous, then there exists K ≥ 0 such that for all
x, y ∈ E, we have ∥u(x) − u(y)∥F ≤ K∥x − y∥E . Since u is supposed to be linear,
u(0E ) = 0F . Then the assertion 2.) follows by taking y = 0E .

2.) ⇒ 3.): It is straightforward by taking r = 1.

3.) ⇒ 1.): Let x, y ∈ E, we may assume, without of losing the generality, that x ̸= y. Consider
r
the vector z := (x − y) ∈ E satisfying ∥z∥E = r. We then have that
∥x − y∥E
∥u(z)∥F ≤ K. Using the linearity of u and the homogeneity of ∥ · ∥F , we obtain that
 
∥x − y∥E
∥u(x) − u(y)∥F = ∥u(x − y)∥F = u z
r F
∥x − y∥E ∥x − y∥E K
= u(z) = ∥u(z)∥F ≤ ∥x − y∥E .
r F r r

K
Hence, u is -Lipschitz continuous function.
r

3.6.4 Continuous Linear Applications

Let (E, ∥ · ∥E ) and (F, ∥ · ∥F ) be normed spaces.

Theorem 3.6.29. Let u be a linear application mapping from E into F . Then, the following
assertions are equivalent.

1.) u is bounded on the unit sphere of (E, ∥ · ∥E ).

2.) u is Lipschitz continuous.

3.) u is continuous.

4.) u is continuous at a point.

5.) u is continuous at the origin point 0E .

Proof. We will prove the implications 1.) ⇒ 2.) ⇒ 3.) ⇒ 4.) ⇒ 5.) ⇒ 1.).

1.) ⇒ 2.): The boundedness of u on the unit sphere stats that there exists K ≥ 0 such that for
all x ∈ E, ∥x∥E = 1, we have ∥u(x)∥F ≤ K. Then it follows from Proposition 3.6.28
that u is Lipschitz continuous.
 3.6. LIMITS AND CONTINUITY 131

2.) ⇒ 3.): Since u is Lipschitz continuous, it follows from Proposition 3.6.25 that u is uniformly
continuous and hence u is continuous on E.

3.) ⇒ 4.): From the definition of continuity, u is continuous on E means that u is continuous at
any particular point.

4.) ⇒ 5.): Assuming that u is continuous at a ∈ E. Let ε > 0, there exists δ > 0 such that for
all x ∈ E, ∥x − a∥E ≤ δ, we have ∥u(x) − u(a)∥F ≤ ε. Using the linearity of u, for all
y ∈ E, ∥y∥E ≤ δ, we have

∥u(y) − u(0E )∥F = ∥u(y) − 0F ∥F = ∥u(y)∥F = ∥u((y + a) − a)∥F

= ∥u(y + a) − u(a)∥F ≤ ε.

Hence, u is continuous at 0E .

5.) ⇒ 1.): It follows the definition of the continuity of u at 0E with ε = 1 that there exists δ0 > 0
such that for all y ∈ E, ∥y∥E ≤ δ0 , we have ∥u(y)∥F ≤ 1. Let x ∈ E with ∥x∥E = 1,
we have then ∥δ0 x∥E = δ0 . Using the linearity of u, we obtain
 
1 1 1
∥u(x)∥F = u δ0 x = ∥u(δ0 x)∥F ≤ .
δ0 F δ0 δ0
Thus, u is bounded on the unit sphere.

Example 3.6.30. We consider here E = C([0, 1], R) the set of continuous functions on the
segment [0, 1] equipped with the Lp norms introduced in Example 3.2.18.
Z 1
• The application φ : f 7→ f (t) sin(t)dt is linear and continuous on the space (E, ∥ ·
0
∥L∞ ([0,1],R) ).

• The application ψ : f 7→ f (1) is linear but not continuous on the space (E, ∥·∥L1 ([0,1],R) )
Solution. • The linearity of φ comes from the linearity of integration. We now prove φ
is continuous. Let f ∈ E with ∥f ∥L∞ ([0,1],R) = 1, we have
Z 1 Z 1
|φ(f )| = f (t) sin(t)dt ≤ |f (t)|| sin t|dt
0 0
Z 1
≤ ∥f ∥L ([0,1],R)
∞ | sin t|dt ≤ 1 − cos 1.
0

Hence, φ is bounded on the unit sphere in (E, ∥ · ∥L∞ ([0,1],R) ). From Theorem 3.6.29,
φ is continuous.

• The linearity of ψ is trivial. We consider the sequence of functions (fn )n∈N ∈ E N

[0, 1] −→ R
given by for all n ∈ N, fn : . We have then for all n ∈ N,
t 7−→ (n + 1)tn
Z 1 Z 1
∥fn ∥L1 ([0,1],R) = |f (t)|dt = (n + 1) tn dt = 1.
0 0
132 TOPOLOGY - ELEMENTARY INTRODUCTION

Also, ψ(fn ) = fn (1) = n + 1 −→ +∞. Hence, ψ is not bounded on the unit sphere
n→+∞
of (E, ∥ · ∥L1 ([0,1],R) ). In conclusion, ψ is not continuous.

Next, we introduce some results on the continuity of linear applications on normed

vector spaces. To do so, we need recall some elements in Linear Algebra.

• We call a linear applicationxix (or a linear map) any function φ which maps from
a K-vector space E into another K-vector space F and satisfies the following linearity
condition

∀x, y ∈ E, ∀α, β ∈ K, φ(αx + βy) = αφ(x) + βφ(y).

• The kernel of a linear application φ is the set of points in E where φ vanishes, that is,
ker(φ) := {x ∈ E | φ(x) = 0F }. From the linearity of φ, ker(φ) is a vector subspace of
E. The image of φ refers to the set im(φ) := {φ(x) | x ∈ E} ⊂ F . From the linearity
of φ, im(φ) is a vector subspace of F .

• We call a linear form defined on the K-vector space E any linear application mapping
from E into K.

• We say the subset H in E is a hyperplane if H is a vector subspace in E with

codimension 1. Precisely speaking, if there exists a non null vector v ∈ E \ {0E }
such that E = H ⊕ Kv. Here Kv denotes the one-dimensional vector space generated
from v, i.e., Kv := {λv | λ ∈ K}.

• In the cases where E is of finite dimensions n ∈ N∗ , then the hyperplanes in E are

the subspaces of dimension n − 1.

• The Fundamental Homomorphism Theorem states that for any linear application φ :
E → F , there exists a vector subspace S ⊂ E such that E = ker(φ) ⊕ S and that
φ|S : S → im(φ) ⊂ F is an isomorphism.

• A direct consequence of the fundamental homomorphism theorem is the Rank-nullity

Theorem, which states that if E is of finite dimensional vector space, then

dim E = dim ker(φ) + rank(φ),

where rank(φ) = dim im(φ).

• It also follows from the fundamental homomorphism theorem that for any hyperplane
H in E, there exists a non-null linear form φ such that H = ker(φ).

Proposition 3.6.31. Let (E, ∥ · ∥) be a normed space and H be a vector hyperplane in E.

Then, H is either closed or dense in E.
xix
In this lecture note, we will principally use the term “linear application” instead of “linear map” to
avoid confusing repetitions of the words “map” and “mapping”.
3.6. LIMITS AND CONTINUITY 133

Proof. We prove at the first time that H is a closed vector subspace in E. It is clear that
H ⊂ E and H is a closed set. Let x, y ∈ H, then there exists sequences (xn )n∈N and
(yn )n∈N of elements in H and they converge respectively to x and to y. Let λ ∈ K, the
sequence (xn + λyn )n∈N is composed with elements in H and it converges to x + λy. Thus,
x + λy ∈ H, which justifies that H is a closed vector subspace in E.
Now we distinguish the cases where H = H or H ⊊ H. If H = H, then H is closed. In
the other case H ⊊ H, there exists a vector v ∈ H \ H. Since H is a hyperplane, we can
write thus E = H ⊕ Ku with u ∈ E \ {0E }. Then there exists λ ∈ K such that v = h + λu
with h ∈ H. We deduce that λ ̸= 0 since if so, then v ∈ H which is contradictory to the
1
assumption. We then write u = (v − h) and remarking that v, h ∈ H. Then, u ∈ H, we
λ
can obtain that E = H ⊕ Ku ⊂ H. Hence, E = H which means H is dense in E.

Example 3.6.32. We consider the space of summable sequences ℓ1 (C) introduced in Ex-
ample 3.2.15 and we endow this space with the norm ∥ · ∥∞ . We have already seen that
ℓ1 (C) ⊊ ℓ∞ (C). So, it is licit to endow ℓ1 (C) with this norm.
• The subset H0 := (un )n∈N ∈ ℓ1 (C) u0 = 0 is a closed hyperplane in (ℓ1 (C), ∥ · ∥∞ ).


+∞
( )
X
• The subset H = (un )n∈N ∈ ℓ (C) 1
un = 0 is a dense hyperplane in the same
n=0
normed space.

ℓ1 (C) −→ C
Proof. • We consider the mapping φ0 : . The linearity of φ0 is
(un )n∈N 7−→ u0
straightforward. Also, for all u = (un )n∈N ∈ ℓ1 (C), we have that

|φ0 (u)| = |u0 | ≤ sup |un | = ∥u∥ℓ∞ (C) .

n∈N

Combined with the linearity of φ0 , this inequality shows that φ0 is 1-Lipschitz con-
tinuous on ℓ1 (C). So, H0 = ker(φ0 ) = φ−1
0 ({0CN }) is an hyperplane and it is closed as
the pre-image of the closed set {0CN } in C by the continuous function φ0 via the set
characterization of continuous functions.
ℓ1 (C) −→ C
+∞
• Consider the mapping φ : X . The linearity of φ corresponds
(un )n∈N 7 →
− un
n=0
exactly the linearity of the sum of convergent series. So, H = ker(φ) is an hyperplane
in ℓ1 (C). We next to show H is dense in ℓ1 (C).
Let u = (un )n∈N ∈ ℓ1 (C). Without of losing the generality, we may assume that
+∞
X
S := un ̸= 0. Let ε > 0, it follows from the Archimedean property in R that there
n=0
|S|
exists N ∈ N such that < N . We then construct the sequence v = (vn )n∈N as
ε
follows
(
S
un − N if n ∈ [[0, N − 1]],
∀n ∈ N, vn =
un if n ≥ N.
134 TOPOLOGY - ELEMENTARY INTRODUCTION
X X
The convergence of the series |vn | follows from those of |un | since the sequences
n n
u and v are equal from the rank N . So, v ∈ ℓ1 (C). Also, we have that
+∞ +∞
X X S
φ(v) = vn = un − N = 0.
N
n=0 n=0

This implies that v ∈ H. Moreover, it holds also that

|S|
∥u − v∥ℓ∞ (C) = sup |un − vn | = ≤ ε.
n∈N N
Hence, H is dense in ℓ1 (C).

Proposition 3.6.33. Let (E, ∥ · ∥) be a normed space and φ be a linear form on E.

• The kernel of φ is either closed or dense in E.
• φ is continuous if, and only if ker(φ) is closed.
Proof. • Since the kernel of a linear form is an hyperplane in E, it follows from Propo-
sition 3.6.31 that any hyperplane in E is either closed or dense. Thus the claim
follows.
• We prove this point by double implications.
⇒: We assume that φ is continuous, then ker(φ) can be considered the pre-image of
the closed set {0E } by the continuous function φ. From the set characterization
of continuous function, ker(φ) is closed.
⇐: We suppose now ker(φ) is closed in E. Assuming in the contrary that φ is not
continuous on E. It follows from Theorem 3.6.29 that φ is not continuous on the
origin point 0E . So, there exists a sequence (xn )n∈N ∈ E N converging to 0E such
that for all n ∈ N, |φ(xn )| ≥ ε0 with a constant ε0 > 0.
Introduce now the sequence (yn )n∈N given by,
xn
∀n ∈ N, yn = ∈ E.
φ(xn )
We remark that since |φ(xn )| ≥ ε0 > 0, then φ(xn ) ̸= 0 and it is licit to calculate
xn
. From this definition and the convergence of (xn )n∈N , we have in one side,
φ(xn )
∥xn ∥E ∥xn ∥
∀n ∈ N, ∥yn ∥E = ≤ −→ 0.
|φ(xn )| ε0 n→+∞
On the other hand, it follows from the linearity of φ that for all n ∈ N,
1
φ(yn ) = φ(xn ) = 1 ⇒ φ(y0 − yn ) = φ(y0 ) − φ(yn ) = 1 − 1 = 0.
φ(xn )
Passing n → +∞, we can conclude that the sequence (y0 −yn )n∈N has all elements
in ker(φ) and it converges to y0 with φ(y0 ) = 1 ̸= 0. It implies that lim (y0 −
n→+∞
yn ) = y0 ∈
/ ker(φ).
Thus, the contradiction with the sequence characterization of closed sets. By
contraposition, φ is continuous.
3.7. EXERCISES 135

⇐(var.): We suppose that ker(φ) is a closed set. From the previous point, ker(φ) is
therefore not dense in E, we have then ker(φ) = ker(φ) ⊊ E. Then, there
exists x̃ ∈ E such that x̃ ∈/ ker(φ), equivalently speaking, φ(x̃) ̸= 0. We set
x̃
now x = and obtain immediately that φ(x) = 1 ̸= 0. This shows also
φ(x̃)
x ∈ ker(φ)∁ and ker(φ)∁ is an open set. Then, there exists r > 0 such that
B(x, r) ⊂ ker(φ)∁ . We assume now in the contrary that there exists z̃ ∈ S(0, 1)
1 −z̃ ∥z̃∥E
such that |φ(z̃)| ≥ . We set z = and it comes that ∥z∥E = ≤ r. In
r φ(z̃) |φ(z̃)|
the consequences, let y = x + z ∈ B(x, r) and we have
 
−z̃ φ(z̃)
φ(y) = φ(x) + φ(z) = φ(x) + φ =1− = 0.
φ(z̃) φ(z̃)

Thus, the contradiction with the fact B(x, r) ⊂ ker(φ)∁ . In conclusion, we have
1
|φ(z)| < for all z ∈ S(0, 1). That means also φ is bounded on the unit sphere,
r
taking in account of the linearity of φ, it follows from Theorem 3.6.29 that φ is
continuous.

3.7 Exercises
Exercise 3.7.1. In this exercise, we want to establish the Minkowski inequality in (Kn , ∥·∥p )
introduced in Example 3.2.9.

1.) Show that for all x, y ∈ Kn , ∥x + y∥1 ≤ ∥x∥1 + ∥y∥1 .

We assume from now on 1 < p < ∞, we introduce the conjugate exponent of p the
1 1
number 1 < p′ < ∞ satisfying + ′ = 1.
p p
2.) Show that the function t 7→ ln t is a concave function on R∗+ .

3.) Let α, β ∈ R∗+ and let 0 < t < 1. Show that αt β (1−t) ≤ tα + (1 − t)β.
′
ap bp ′
4.) Let a, b ∈ R+ . Show that ab ≤ + ′ and the equality holds if, and only if ap = bp .
p p
This inequality is called the Young’s inequality of productxx .
n
X
5.) Let x, y ∈ Kn satisfying that ∥x∥p = ∥y∥p′ = 1. Show that |xj yj | ≤ 1.
j=1

n
X
6.) Deduce that for all x, y ∈ Kn , |xj yj | ≤ ∥x∥p ∥y∥p′ . This inequality is called the
j=1
Hölder inequality.
n
X
7.) Let u, v ∈ Kn . Show that |uj ||vj |p−1 ≤ ∥u∥p ∥v∥pp−1 .
j=1
xx
William Henry Young (1863 - 1942), English mathematician. Professor at Cambridge University. He
worked on measure theory, Fourier series and differential calculus.
136 TOPOLOGY - ELEMENTARY INTRODUCTION

8.) Let x, y ∈ Kn . Show the Minkowski inequality ∥x + y∥p ≤ ∥x∥p + ∥y∥p .

Exercise 3.7.2. Justify Example 3.2.18.

Exercise 3.7.3. Justify Proposition 3.3.6.

Exercise 3.7.4. Justify Proposition 3.4.3.

Exercise 3.7.5. Let (X, dX ), (Y, dY ) be metric spaces and f be a function mapping from
A ⊂ X into B ⊂ Y . Let a ∈ cl(A) and l ∈ Y .
Show that x→a
lim f (x) = l if, and only if for all sequence (xn )n∈N of elements in A which
x∈A
converges to a, there exits a subsequence (xσ(n) )n∈N of (xn )n∈N such that f (xσ(n) ) −→ l.
n→+∞

Exercise 3.7.6. Let (E, ∥ · ∥) be a normed vector space, and F ⊂ E be a vector subspace
in E. Show that if F̊ ̸= ∅, then F = E.

Exercise 3.7.7. Let (X, d) be a metric space.

1.) Let F be a closed subset in X. Show that

\ there exists a nested intersecting sequence
(Un )n∈N of open sets in X such that Un = F (See Definition 4.1.8 for nested
n∈N
intersecting sequences).

2.) Let F and G be two disjoint closed sets in X. Show that there exists disjoint open
sets U and V such that F ⊂ U and G ⊂ V .

3.) Show that there exists a continuous function f mapping from X into R such that
f −1 ({1}) = F and f −1 ({0}) = G.

Hint: consider the distance function x 7→ d(x, F ).

Exercise 3.7.8. Let (X, d) be a metric space and A ⊂ X. We recall that ∂A = A \ Å.

1.) Give an example of subsets A and B in R such that

∂(A ∪ B) ∪ ∂(A ∩ B) ̸= (∂A) ∪ (∂B).

2.) Show that, for all non-empty subsets A, B ⊂ X,

∂(A ∪ B) ∪ ∂(A ∩ B) ⊂ (∂A) ∪ (∂B).

3.) Show that, if A, B ⊂ X are closed non-empty subsets,

∂(A ∪ B) ∪ ∂(A ∩ B) = (∂A) ∪ (∂B).

3.7. EXERCISES 137

Exercise 3.7.9. We consider in this Exercise the space ℓ∞ (R) of bounded sequences of
elements in R, and we endow this space with the infinity norm ∥ · ∥ℓ∞ (R) which is defined
by, for all x = (xn )n∈N ∈ ℓ∞ (R), ∥x∥ℓ∞ (R) = sup |xn |. We say a sequence (xn )n∈N ∈ RN is
n∈N
almost zero if there exists only a finite number of terms which are not zero. We denote by
V the set of almost zero sequences.
Show that V is the set of sequences which converge to 0.

Exercise 3.7.10. Following the previous exercise, let λ = (λn )n∈N ∈ RN . We consider the
+∞
X
map Tλ : (xn )n∈N 7→ λ n xn .
n=0

1.) Show that Tλ is well-defined and is linear on V .

2.) Which condition does λ need to satisfy such that Tλ is continuous on V ?

Exercise 3.7.11. Let f and g two continuous functions mapping from [a, b] ⊂ R into it self.
We suppose that f ◦ g = g ◦ f . We denote by f n and g n their n times iterations f ◦ · · · ◦ f
and g ◦ · · · ◦ g.

1.) Assume now f > g on [a, b]. Show that there exists K > 0 such that for all n ∈ N∗ ,
and x ∈ [a, b], f n (x) ≥ Kn + g n (x).

2.) Deduce that there exists x ∈ [a, b] such that f (x) = g(x).

|x + ty|
Exercise 3.7.12. We define a function L on R2 . For all (x, y) ∈ R2 , L(x, y) := sup .
t∈R 1 + t2

1.) Show that L is a norm on R2 .

2.) Show that L is equivalent to the Euclidean norm ∥ · ∥2 on R2 . Give the constants c
and C of the equivalence.

3.) Give an explicit expression of L in functions of x and y. Represent the unit ball
respect to the norm L.

Exercise 3.7.13. Let E be the spaces of C 1 -class functions mapping from [0, 1] into R, we
define for all f ∈ E,

N∞ (f ) := sup |f (t)|;
t∈[0,1]
Z 1
N1 (f ) := |f (t)|dt;
0
Nd (f ) := |f (0)| + sup |f ′ (t)|.
t∈[0,1]

Show that

1.) N∞ , N1 and Nd define norms on E;

138 TOPOLOGY - ELEMENTARY INTRODUCTION

2.) for all f ∈ E, N1 (f ) ≤ N∞ (f ) ≤ Nd (f );

3.) the norms N∞ and Nd are not equivalent;

4.) the norms N1 and N∞ are not equivalent.

5.) Let Ω = {f ∈ E | ∀t ∈ [0, 1], f (t) > 0}. For each of the above norms, is Ω an open
set?

Hint: we can try to construct a sequence of functions in E such that it converges to 0 for
one norm and doesn’t for another.

Exercise 3.7.14. We consider the space of square matrices (Mn (C), ∥ · ∥M ) where n ∈ N∗
and ∥ · ∥M denotes a matrix norm.

1.) We call a matrix M ∈ Mn (C) is nilpotent if there exists p ∈ [[1, n]] such that
M p = 0Mn (C) . Show that the set of nilpotent matrices is a closed set in Mn (C).

2.) We call the spectral radius of a matrix M ∈ Mn (C) and denote it by ρ(M ) the
maximum of the absolute values of the eigenvalues of M . That is,

ρ(M ) := max {|λ| | λ ∈ σ(M )} .

Show that the function ρ : M 7→ ρ(M ) is not a norm on Mn (C).

3.) Show that for all M ∈ Mn (C), ρ(M ) ≤ ∥M ∥M .

4.) Let M ∈ Mn (C), we call the similitude class of the matrix M the following set

Sim(M ) := T −1 M T

T ∈ GL(n) .

Show that for any matrix A ∈ Sim(M ), A has the same eigenvalues as M .

5.) Following from the previous question, show that if 0Mn (C) ∈ Sim(M ), then M is
nilpotent.
Hint: We may consider the functions A 7→ det(A − λIn ) to show that 0 is the unique
eigenvalue of M .

6.) Let M be a nilpotent matrix. Show that there exists Mε ∈ Sim(M ) such that
∥Mε ∥M ≤ ε. Then deduce that 0Mn (C) ∈ Sim(M ).
Hint: we may think on the trigonalization
 or Jordan form of M and the transition
1 (0)
ε
 
 
matrices Tε =  ..
.
.
 
 
(0) εn−1

7.) Let M ∈ Mn (C) and ε > 0. Show that there exists Mε ∈ Sim(M ) such that ∥Mε ∥M ≤
ρ(M ) + ε.
3.7. EXERCISES 139

8.) Let M ∈ Mn (C). Prove the following Gels’fand’s formulaxxi ,

1
lim ∥M k ∥M
k
= ρ(M ).
k→+∞

Exercise 3.7.15. In this exercises we will study the subgroups contained in (R, +). In
what follows, we denote by G a subgroup in (R, +).
1.) Let α ∈ R∗ . Show that αZ := {αn | n ∈ Z} is a subgroup in (R, +).

2.) Deduce that if α ∈ G, then αZ ⊂ G.

3.) We suppose from now on that G ̸= {0}. Show that there exists α > 0 such that
αZ ⊂ G and the following infimum is well-defined,

a := inf(G ∩ R∗+ ).

4.) Let g, g ′ ∈ G with g < g ′ such that (g, g ′ ) ∩ G = ∅. Show that a = g ′ − g.

5.) We suppose that a > 0. Show that a ∈ G.

6.) Following the previous question, show that G = aZ.

7.) We suppose that a = 0. Let x ∈ R and n ∈ N∗ . Show that there exists gn ∈ G such
1
that x ≤ gn ≤ x + . Deduce that G is dense in R.
n
8.) Show that Gπ := Z + 2πZ = {n + 2πm | n, m ∈ Z} is a subgroup in (R, +).

9.) Following the previous question, show that Gπ is dense in R.

10.) Show that {cos n | n ∈ N} = [−1, 1].

Exercise 3.7.16. Show that, in a normed space (E, ∥ · ∥), the only sets simultaneously
open and closed are ∅ and E.

Exercise 3.7.17. Consider the space C([0, 1], R) of continuous functions defined on [0, 1]
into R. We recall that the infinity norm ∥ · ∥L∞ ([0,1],R) is defined by, ∀f ∈ C([0, 1], R),
∥f ∥L∞ ([0,1],R) = sup |f (x)|.
x∈[0,1]

1.) Let A := {f ∈ C([0, 1], R) | f (0) = 0}. Show that A is a closed subset in (C([0, 1], R), ∥·
∥L∞ ([0,1],R) ).
 Z 1 
2.) Let B := f ∈ C([0, 1], R) f (t)dt ≥ 1 . Show that B is a closed subset in
0
(C([0, 1], R), ∥ · ∥L∞ ([0,1],R) ).

3.) Show that for all f ∈ A ∩ B, ∥f ∥L∞ ([0,1],R) > 1.

xxi
Israı̈l Moyseyovich Gel’fand (1913-2009), Soviet and American mathematician and biologist. He is
considered as one of the greatest mathematician in the 20th century. Known for his works on group theory,
representation theory and functional analysis.
140 TOPOLOGY - ELEMENTARY INTRODUCTION

4.) Construct a sequence of functions (fn )n∈N of elements in A ∩ B satisfying in addition

that ∥fn ∥L∞ ([0,1],R) −→ 1.
n→+∞

5.) Is the distance d(x, U ) between a point x to a closed set U achieved in general? Justify
your answer.
 Chapter 4

Topology - Fundamental Theories

In this chapter, we denote for instance by (X, d) a metric space and by (E, ∥ · ∥) a normed
space on the field K = R or C.

4.1 Completeness
4.1.1 Complete Metric Spaces
Definition 4.1.1. We say a sequence (xn )n∈N of elements in a metric space (X, d) is a
Cauchy sequence if it satisfies,

∀ε > 0, ∃N ∈ N s.t. ∀n, m ∈ N, n, m ≥ N ⇒ d(xn , xm ) ≤ ε.

Remark 4.1.2. There are also equivalent formulations of Cauchy sequences, we present here
three of them.
• ∀ε > 0, ∃N ∈ N s.t. ∀n, p ∈ N, n ≥ N ⇒ d(xn , xn+p ) ≤ ε.

• There exists (αn )n∈N ∈ RN

+ with αn −→ 0 such that ∀n, p ∈ N, d(xn , xn+p ) ≤ αn .
n→+∞

• (in a normed spaces (E, ∥ · ∥)) For all V ∈ N (0), there exists N ∈ N such that for all
n, p ∈ N, n ≥ N ⇒ xn − xn+p ∈ V .

Proposition 4.1.3. The following properties satisfied by Cauchy sequences of real numbers
also hold in any metric space.
• Any Cauchy sequence is bounded.

• Any convergent sequence is a Cauchy sequence.

• Any Cauchy sequence possessing a cluster point converges to this point.

Proof. It is sufficient to replace the absolute values by the distance functions in the proofs
of Proposition 1.3.28 and Lemmas 1.3.29 and 1.3.30.

Proposition 4.1.4. Let f be an uniformly continuous function mapping from (X, dX ) into
(Y, dY ). Then the image by f of any Cauchy sequence in X is a Cauchy sequence in Y .

141
142 TOPOLOGY - FUNDAMENTAL THEORIES

Proof. Let (xn )n∈N be a Cauchy sequence in X. We consider the sequence (f (xn ))n∈N ∈ Y N .
Let ε > 0, it follows from the uniform continuity of f , there exists δ > 0 such that for all x, y,
dX (x, y) ≤ δ ⇒ dY (f (x), f (y)) ≤ ε. Also, since (xn )n∈N is a Cauchy sequence, there exists
N ∈ N such that for all n, m ∈ N, n, m ≥ N ⇒ dX (xn , xm ) ≤ δ. Then, for n, m ≥ N we
have dX (xn , xm ) ≤ δ ⇒ dY (f (xn ), f (xm )) ≤ ε, which means that the sequence (f (xn ))n∈N
is a Cauchy sequence.

Definition 4.1.5. We say a metric space (X, d) is complete if any Cauchy sequence of
elements in X converges.

Definition 4.1.6. Let A be a non empty subset in (X, d), we say the subset A is complete
if for any Cauchy sequence of elements in A converges to a limit in A.

Theorem 4.1.7. The normed spaces (R, | · |) and (C, | · |) are complete.

Proof. They are consequences of Definition 1.4.8 and Theorem 1.4.9.

Definition 4.1.8 (Nested decreasing sequences). We say the sequence (Fn )n∈N of subsets
in (X, d) forms a nested decreasing sequencei if for all n ∈ N, Fn+1 ⊂ Fn .

Proposition 4.1.9. A metric space (X, d) is complete if, and only if for any nested de-
creasing sequence of bounded and closed subsets (Fn )n∈N such that diam(Fn ) −→ 0 has a
n→+∞
non empty intersection.

Proof. We prove this proposition by double implications.

⇒: We suppose that (X, d) is complete. Let (Fn )n∈N be a nested decreasing sequence of
closed and bounded sets satisfying diam(Fn ) −→ 0. Since each set Fn is supposed
n→+∞
to be non-empty, we can thus choose an element xn ∈ Fn for all n ∈ N. The nested
relation ∀n ∈ N, Fn+1 ⊂ Fn implies that xm ∈ Fn once m ≥ n. For N ∈ N, we
have then xn , xm ∈ FN once n, m ≥ N . So, d(xn , xm ) ≤ diam(FN ) −→ 0. This
N →+∞
proves that (xn )n∈N is a Cauchy sequence, hence it converges by the completeness of
(X, d). Let l = lim xn ∈ X. For N ∈ N, we know that xn ∈ FN once n ≥ N . So,
n∈N
the sequence (xn )n≥N is a convergent sequence of elements in FN . From the sequence
characterization of closed sets and the fact that FN is closed, we\have that l ∈ FN .
Since the latter result holds for all N ∈ N we have that l ∈ FN . Thus, the
N ∈N
sequence (Fn )n∈N has a non-empty intersection.

⇐: We assume that any nested decreasing sequence (Fn )n∈N of closed and bounded sets
satisfying diam(Fn ) −→ 0 has a non-empty intersection. Let (xn )n∈N be a Cauchy
n→+∞
sequence in (X, d). For all k ∈ N∗ , there exists Nk ∈ N such that for all n ∈ N,
1 1
n ≥ Nk ⇒ dX (xNk , xn ) ≤ k . In other words, xn ∈ B(xNk , k ). We then consider the
2 2
i
This kind of sequences are also called as “decreasing sequences of nested boxes”.
4.1. COMPLETENESS 143

1
sequences (Fk )k∈N where Fk := B(xNk , ). It is clear that each Fk is closed and
2k−1
1
bounded as a closed ball. It is also immediate that diam(Fk ) = k−1 −→ 0. For all
2 k→+∞
1
k ∈ N, we can choose Nk+1 > Nk such that xn ∈ B(xNk+1 , k+1 ) once n ≥ Nk+1 . We
2
1 1
have in particular that d(xNk , xNk+1 ) ≤ k . So, for all y ∈ Fk+1 = B(xNk+1 , k ), we
2 2
have
1 1 1
d(y, xNk ) ≤ d(xNk , xNk+1 ) + d(xNk+1 , y) ≤ k
+ k = k−1 .
2 2 2
Hence, y ∈ Fk and also Fk+1 ⊂ Fk . This justifies that (Fk )k∈N is a nested de-
creasing sequence of closed and bounded
\ sets with the diameters converging to 0.
From the assumption, there exists l ∈ Fk . Since l ∈ Fk for all k ∈ N, we have
k∈N
1
then d(xNk , l) ≤ −→ 0. We can thus deduce that the Cauchy sequence
2k−1 k→+∞
(xn )n∈N possesses a subsequence (xNk )k∈N which converges to l. We then conclude
that xn −→ l. Hence, (X, d) is complete.
n→+∞

Proposition 4.1.10. Any complete subset in a metric space is closed.

Proof. Since all convergent sequences are Cauchy sequences, then all convergent sequences
have their limit in the complete subset and the claim follows from the sequence characteri-
zation of closed sets.

Proposition 4.1.11. Any closed subset in a complete metric space is complete.

Proof. Let A ⊂ X be a closed subset set in the complete metric space (X, d). Consider a
Cauchy sequence (an )n∈N of elements in A. From the completeness of (X, d), the sequence
(an )n∈N converges. Also, from the sequence characterization of closed sets, the limit of
(an )n∈N belongs to A. Hence, the Cauchy sequence (an )n∈N converges in A, which means
thus A is a complete subset.

Theorem 4.1.12. The product space of two complete metric spaces is complete.
(1) (2)
Proof. Consider the product space (X1 × X2 , dX1 ×X2 ). Let (xn )n∈N = (xn , xn )n∈N be a
Cauchy sequence in (X1 × X2 , dX1 ×X2 ). It follows from Definition 3.3.10 and the defini-
(1) (2)
tion of Cauchy sequences, both (xn )n∈N and (xn )n∈N are Cauchy sequences respectively
in (X1 , dX1 ) and in (X2 , dX2 ). Using the completeness of (X1 , dX1 ) and (X2 , dX2 ), the se-
(1) (2)
quences (xn )n∈N and (xn )n∈N are both convergent. Hence, the sequence (xn )n∈N converges
in (X1 × X2 , dX1 ×X2 ), and thus the space (X1 × X2 , dX1 ×X2 ) is complete.

Definition 4.1.13. We say a function f mapping between two metric spaces is a contrac-
tion if f is K-Lipschitz continuous function with 0 < K < 1.
144 TOPOLOGY - FUNDAMENTAL THEORIES

Theorem 4.1.14 (Banach Fixed Point, Contraction Mapping Principle). Let f be a con-
traction which maps from a complete metric space (X, d) into itself.
Then, f possesses a unique fixed point l ∈ X, i.e. f (l) = l.
In fact, for all x ∈ X the sequence (f n (x))n∈N converges to l and it satisfies,

∀n ∈ N, d(f n (x), l) ≤ K n d(x, l).

Proof. Let x ∈ X, we consider the sequence (xn )n∈N defined by recurrence. We set x0 = x
and for all n ∈ N, xn+1 = f (xn ). In other words, xn = f n (x). From the K-Lipschitz
continuity of the function f , we can obtain by using an immediate induction that for all
m ∈ N,

d(xm , xm+1 ) = d(f (xm−1 ), f (xm )) ≤ Kd(xm−1 , xm ) ≤ K m d(x0 , x1 ).

Let n ∈ N and p ∈ N∗ , we can deduce from the fact K < 1 that,

p−1 p−1
1 − Kp
X X  
n+k n
d(xn , xn+p ) ≤ d(xn+k , xn+k+1 ) ≤ K d(x0 , x1 ) = K d(x0 , x1 )
1−K
k=0 k=0
d(x0 , x1 ) n
≤ K −→ 0.
1−K n→+∞

Hence, the sequence (xn )n∈N is a Cauchy sequence in (X, d). From the completeness of
(X, d), the sequence (xn )n∈N converges. We then set l := lim xn ∈ X. Since f is Lipschitz
n→+∞
continuous, f is therefore continuous, we have then f (xn ) −→ f (l). In addition, we have
n→+∞
also f (xn ) = xn+1 −→ l. From the uniqueness of the limit, it follows that f (l) = l. Thus,
n→+∞
l is a fixed point of f in (X, d).
We now prove the uniqueness of the fixed point of f . Let l, l′ be two fixed points of f ,
it follows from the K-Lipschitz continuity of f ,

d(l, l′ ) = d(f (l), f (l′ )) ≤ Kd(l, l′ ) ⇒ (1 − K)d(l, l′ ) ≤ 0

Knowing that K < 1, the latter inequality is possible only if d(l, l′ ) ≤ 0. We then deduce
from the positivity and the separation property of the distance function d that l = l′ . Hence,
the fixed point of the function f is unique.
Let l be the fixed point of f in (X, d) and x ∈ X. Using the relation f (l) = l and an
immediate induction, we have for all n ∈ N,

d(f n (x), l) = d(f (f n−1 (x)), f (l)) ≤ Kd(f n−1 (x), l) ≤ K n d(x, l).

4.1.2 Banach Spaces

Definition 4.1.15. We call a Banachii space any complete normed space.

Example 4.1.16. The normed spaces (Kn , ∥ · ∥p ) are complete for 1 ≤ p ≤ ∞.

ii
Stephan Banach (1892 - 1945), Polish mathematician. He was considered as the founder of modern
functional analysis.
4.1. COMPLETENESS 145

Solution. We observe at first that the normed space (Kn , ∥ · ∥∞ ) is exactly the n-times
product space (K, | · |) × · · · × (K, | · |). From the completeness of the field K = R or C and
from Theorem 4.1.12, the normed space (Kn , ∥ · ∥∞ ) is complete.
On the other hand, it follows from the results in Proposition 3.2.11 and Corollary 3.2.12
that any norms ∥·∥p , 1 ≤ p ≤ ∞, on Kn are equivalent. From Proposition 3.4.7, they defines
the same open sets and hence they define the same neighborhoods. Since the convergence
of sequences could be defined via neighborhoods, any sequence is convergent in (Kn , ∥ · ∥p )
if, and only if it is convergent in (Kn , ∥ · ∥q ) for any 1 ≤ p, q ≤ ∞. The same equivalence
holds also for Cauchy sequences. Hence, (Kn , ∥ · ∥p ) is complete if, and only if (Kn , ∥ · ∥∞ )
is complete. In conclusion, (Kn , ∥ · ∥p ) is a Banach space for all 1 ≤ p ≤ ∞.

Example 4.1.17. The spaces of sequences on C, (ℓ∞ (C), ∥ · ∥ℓ∞ (C) ), (ℓ1 (C), ∥ · ∥ℓ1 (C) ) and
(ℓ2 (C), ∥ · ∥ℓ2 (C) ) are complete. (For the definition of those spaces, see Example 3.2.15.)
Solution. We will prove the case of (ℓ∞ (C), ∥·∥ℓ∞ (C) ) and the other cases are left in Exercises.
Consider (x(k) )k∈N be a Cauchy sequence in (ℓ∞ (C), ∥ · ∥ℓ∞ (C) ). For all k, n ∈ N, we
(k)
denote by xn ∈ C the element indexed by n in the sequence x(k) ∈ ℓ∞ (C).
Let ε > 0 and consider n ∈ N being fixed. Using the fact that (x(k) )k∈N is a Cauchy
sequence in ℓ∞ (C), there exists k0 ∈ N such that
∀k, p ∈ N, k ≥ k0 ⇒ ∥x(k) − x(k+p) ∥ℓ∞ (C) ≤ ε. (4.1)

From the definition of the ℓ∞ -norm, we have |x(k) (k+p)

n − xn | ≤ ∥x(k) − x(k+p) ∥ℓ∞ (C) . In the
(k)
consequences, the sequence (xn )k∈N is a Cauchy sequence in C. From the completeness of
C, this sequence converges, and we denote its limit by ln ∈ C.
Furthermore, since the sequence (x(k) )k∈N is a Cauchy sequence, it is therefore bounded.
That is to say, there exists M > 0 such that ∥x(k) ∥ℓ∞ (C) ≤ M for all k ∈ N. It follows from
(k )
the convergence lim x(k)
n = ln that there exists k1 ∈ N such that |xn
1
− ln | ≤ 1. And we
k→+∞
have the estimation,
|ln | = |ln − x(k
n
1)
+ x(k 1) (k1 )
n | ≤ |ln − xn | + ∥x
(k1 )
∥ℓ∞ (C) ≤ 1 + M.
We remark that the constant M is independent of the index n ∈ N. Thus, this estimation
shows that the sequence (ln )n∈N ∈ CN is bounded, i.e., l := (ln )n∈N ∈ ℓ∞ (C).
Next, we use again (4.1) to duce that there exists k2 ∈ N such that
ε
∀k, p ∈ N, k ≥ k2 ⇒ ∥x(k) − x(k+p) ∥ℓ∞ (C) ≤ .
2
For k ≥ k2 , we use again the convergence lim xn(k) = ln , by choosing pk,n ∈ N large enough
k→+∞
(k+pk,n ) ε
such that |xn − ln | ≤ . We can now deduce the estimation,
2
(k+p ) (k+p ) (k+p )
|x(k) (k)
n − ln | ≤ |xn − xn
k,n
| + |xn k,n − ln | ≤ ∥x(k) − x(k+pk,n ) ∥ℓ∞ (C) + |xn k,n − ln |
ε ε
≤ + = ε.
2 2
Remarking this estimation holds for all n ∈ N and all k ≥ k2 , we can thus conclude with
∥x(k) − l∥ℓ∞ (C) −→ 0.
k→+∞

Hence, the normed space (ℓ∞ (C), ∥ · ∥ℓ∞ (C) ) is complete.

146 TOPOLOGY - FUNDAMENTAL THEORIES

Example 4.1.18. The space C([0, 2], R) endowed with the L1 -norm is not complete.

Solution. We consider the sequence of functions (fn )n∈N defined by

[0, 2] −→ ( R
∗
∀n ∈ N , fn : xn if 0 ≤ x < 1,
x 7−→
1 if 1 ≤ x ≤ 2

It is clear that fn ∈ C([0, 2], R) for all n ∈ N∗ . Moreover, for all n, p ∈ N∗ , we have the
estimation
Z 1 Z 2
∥fn − fn+p ∥L1 ([0,2]) = |xn − xn+p |dx + |1 − 1|dx
0 1
1 1 1
= − ≤ −→ 0.
n+1 n+p+1 n + 1 n→+∞

Hence, the sequence (fn )n∈N is a Cauchy sequence in (C([0, 2], R), ∥ · ∥L1 ([0,2],R) ).
On the other hand, consider x ∈ [0, 2].

• If x ∈ [0, 1), then fn (x) = xn −→ 0.

n→+∞

• If x ∈ [1, 2], then fn (x) = 1 −→ 1.

n→+∞

Thus, the sequence of functions (fn )n∈N converges pointwiselyiii to the following function

[0, 2] −→ ( R
f: 0 if x ∈ [0, 1)
x 7−→
1 if x ∈ [1, 2].

In the consequences,
Z 1 Z 2
n 1
∥fn − f ∥L1 ([0,2]) = |x − 0|dx + |1 − 1|dx = −→ 0. (4.2)
0 1 n + 1 n→+∞

This shows that (fn )n∈N converges to f in the L1 -norm.

However, the function f is not continuous on [0, 2]. Thus, the sequence of functions
(fn )n∈N does not converge in the normed space (C([0, 2], R), ∥ · ∥L1 ([0,2],R) ). In the conse-
quence, the space (C([0, 2], R), ∥ · ∥L1 ([0,2],R) ) is not complete.

X
Proposition 4.1.19. Let (E, ∥ · ∥) be a Banach space. A series xn of elements in E is
n
convergent if, and only if
n+p
X
∀ε > 0, ∃N ∈ N s.t. ∀n, p ∈ N, n ≥ N ⇒ xk ≤ ε.
k=n+1
iii
There are various modes of convergences for sequences of functions. We will go further on this subject
in the next semester.
4.1. COMPLETENESS 147

n
X
Proof. We consider the sequence of partial sums (Sn )n∈N defined by Sn = xk . Since
k=0
the space is (E, ∥ · ∥) is complete,
X a sequence is convergent if ,and only if it is a Cauchy
sequence. So, the series xn is convergent if, and only if the sequence of partial sums
n
(Sn )n∈N is a Cauchy sequence. Thus, the claim follows.

X
Definition 4.1.20. We say a series xn of elements in a Banach space (E, ∥ · ∥) is
X n
absolutely convergent if the numerical series ∥xn ∥ is convergent
n

X
Theorem 4.1.21. Any absolutely convergent series xn in a Banach space (E, ∥ · ∥) is
n
convergent and it satisfies
+∞
X +∞
X
xn ≤ ∥xn ∥.
n=0 n=0
X
Proof. Let xn be an absolutely convergent series in E. Then the numerical series
X n
∥xn ∥ is convergent, which implies that the sequence of partial sums of this series is
n
also a Cauchy sequence. It follows from the triangular inequality that for all n, p ∈ N, we
have
n+p
X n+p
X
xk ≤ ∥xk ∥.
k=n+1 k=n+1
X
This implies that the sequence of partial sums of xn is also a Cauchy sequence. From
X n
the completeness of E, the series xn converges.
n
For all N ∈ N, we can obtain from the triangular inequality that
N
X N
X +∞
X
xn ≤ ∥xn ∥ ≤ ∥xn ∥.
n=0 n=0 n=0

Since the considered series are convergent, we can thus pass N → +∞ to obtain that
+∞
X +∞
X
xn ≤ ∥xn ∥.
n=0 n=0

Remark 4.1.22. The reciprocal statement of this theorem is true. That is, if any absolutely
convergent series converges in a normed space, then this normed space is complete. See in
Exercise 4.6.27.
148 TOPOLOGY - FUNDAMENTAL THEORIES

4.1.3 Banach Algebras

Let (E, +, ·, ×) be an algebra over K and ∥ · ∥ be a norm on the vector space (E, +, ·).

Definition 4.1.23. We say the norm ∥ · ∥ is an algebraic norm on E if it satisfies,

• x, y ∈ E, ∥x × y∥ ≤ ∥x∥∥y∥

• ∥1E ∥ = 1.

In the case, we call the normed space (E, ∥ · ∥) as a normed algebra.

Definition 4.1.24. Let (E, ∥ · ∥) be a normed algebra, we say E to be a Banach algebra

if the normed space (E, ∥ · ∥) is complete. In other words, if a Banach space (F, ∥ · ∥F )
possesses an algebraic norm ∥ · ∥F , then F is a Banach algebra.

In the rest of this subsection, we suppose that (E, ∥ · ∥) is a Banach algebra.

Proposition 4.1.25 (Neumanniv Series). Let a ∈ E such that ∥a∥ < 1. Then,

• the element 1E − a ∈ E is invertible;

X
• the series an is absolutely convergent;
n

+∞
X
• an = (1E − a)−1 .
n=0
X
Proof. Let a ∈ E, we consider the series an . Since ∥ · ∥ is an algebraic norm, we have
n X
then for all n ∈ N, ∥an ∥ ≤ ∥a∥n . Since ∥a∥ < 1, the numerical series ∥a∥n converges.
X n
n
Then the series a is absolutely convergent.
n
Let N ∈ N, we have
N N
!
X X
n
a (1E − a) = (1E − a) an = 1E − aN +1 −→ 1E .
N →+∞
n=0 n=0

From the continuity of the multiplication at left, as well as at right, we obtain therefore
+∞ +∞
!
X X
n
a (1E − a) = (1E − a) an = 1E
n=0 n=0

+∞
X +∞
X
This shows that both 1E − a and an are invertible and an = (1E − a)−1 .
n=0 n=0

Example 4.1.26. In the space of square matrices Mn×n (K). We have the following results.
iv
Karl Gottfried Neumann (1832-1925), German mathematician. Professor at Leipzig University. He
worked in integral equations and potential theory.
4.1. COMPLETENESS 149

• For any matrix norm ∥·∥p,M given in Example 3.2.14, the normed space (Mn×n (K), ∥·
∥p,M ) is a Banach algebra. Moreover, those norms are equivalent from one to each
others.

• For any matrix M ∈ Mn×n (K), if the eigenvalues of the matrix t M M all have an
absolute value strictly less then 1, then Id − M is invertible and its inverse can be
calculated via the Neumann series.

• The set of invertible matrices GL(n) is an open and dense set.

GL(n) −→ GL(n)
• The function Inv : is continuous.
M 7−→ M −1

Solution. • We can see the matrix norm ∥ · ∥p,M as the operator norm on the space
of operators L (Kn , Kn ) where Kn is endowed with the norm ∥ · ∥p . From the result
in Example 4.1.16, (Kn , ∥ · ∥p ) is complete. Then, applying Theorem 4.5.8 the space
(Mn×n (K), ∥ · ∥p,M ) is complete. Next, we prove the norm ∥ · ∥p,M is an algebraic
norm.
Let A, B ∈ Mn×n (K), it follows from the first result in Example 3.2.14, that for all
x ∈ Kn , ∥Ax∥p ≤ ∥A∥p,M ∥x∥p . We then deduce that

∀x ∈ Kn , ∥ABx∥p ≤ ∥A∥p,M ∥Bx∥p ≤ ∥A∥p,M ∥B∥p,M ∥x∥p .

∥ABx∥p
It implies that for all x ∈ Kn , x ̸= 0Kn , we have ≤ ∥A∥p,M ∥B∥p,M . Then,
∥x∥p
∥ABx∥p
∥A∥p,M ∥B∥p,M can be considered as an upper bound of the function x 7→
∥x∥p
for all x ̸= 0Kn . In conclusion, the following relation holds

∥ABx∥p
∥AB∥p,M = sup ≤ ∥A∥p,M ∥B∥p,M .
x∈Kn ∥x∥p
x̸=0

In conclusion, ∥ · ∥p,M is an algebraic norm and hence (Mn×n (K), ∥ · ∥p,M ) is a Banach
algebra.
Finally, since the space Mn×n (K) is finite dimensional, it follows from Theorem 4.4.10
that those norms are equivalent.

• It follows from the previous point that the matrix norms defines the algebraic norms
on Mn×n (K). Also, from Example 3.2.14, let M ∈ Mn×n (K), ∥M ∥2,M is given by
the largest absolute value of the eigenvalues of the Hermitian matrix t M M . Thus the
claim follows from Proposition 4.1.25.

• We have seen in Proposition 3.6.15 that GL(n) is an open set.

To prove the denseness, we will use a spectral argument. Let M ∈ Mn×n (K), since
the field C is algebraically closedv , the matrix M is C-trigonalizablevi . There exists
then the transition matrix P ∈ GL(n) such that M = P −1 T P where T is an upper
v
It is the consequence of Fundamental Theorem of Algebra, which states that any complex polynomial
has at least one root in C.
vi
In an equivalent manner, we can also use the arguments of Jordan forms.
150 TOPOLOGY - FUNDAMENTAL THEORIES

triangular matrix with the diagonal coefficients composed by its eigenvalues. Let
σ(M ) = {λ1 , · · · , λn } with λi ∈ C, 1 ≤ i ≤ n, counted with the multiplicity. Let
ε > 0, we consider then the matrices defined by Mε := M +εId = P −1 Tε P . Here Tε =
T + εId is an upper triangular matrix with the diagonal coefficients λ1 + ε, · · · , λn + ε.
We will denote by m the smallest absolute value of non zero eigenvalues. That is,
m := min {|λ| | λ ∈ σ(M ), λ ̸= 0}. Thus, it is sufficient to choose ε < m to obtain
that λ + ε ̸= 0 for all λ ∈ σ(M ). In other words, all eigenvalues of the matrix Tε are
non zero. Since the matrices Mε and Tε are similar, all eigenvalues of Mε are non
zero. Hence, the matrix Mε is invertible.
In conclusion, for ε sufficiently small, there exists an invertible matrix Mε ∈ GL(n)
such that ∥Mε − M ∥2,M ≤ ε. Thus the set GL(n) is dense in Mn×n (K).
• We can also use Neumann series to prove that GL(n) is open. Let ∥ · ∥p,M be a
matrix norm given in Example 3.2.14. Consider M ∈ GL(n). It is clear that M and
M −1 are not zero matrices and from the separation property of the norm, we may set
r := ∥M −1 ∥−1
p,M > 0.
Let Mh ∈ Mn×n (K) be a matrix in the open ball Bp,M (M, r). Since M is invertible,
we may write Mh = (In − h)M where h := (M − Mh )M −1 . Since ∥ · ∥p,M is an
algebraic norm, we have
∥h∥p,M ≤ ∥M − Mh ∥p,M ∥M −1 ∥p,M < r · r−1 = 1.
Using Proposition 4.1.25, the matrix In − h is invertible and its inverse is given by the
sum of the Neumann series,
+∞
X
(In − h)−1 = hk .
k=0
In the consequences, Mh = (In − h)M is also invertible, i.e., Mh ∈ GL(n). We have
thus proved that Bp,M (M, r) ⊂ GL(n) and hence, GL(n) is open.
• Following the previous point, we have
+∞
X
Mh−1 −M −1
=M −1 −1
(In − h) −M −1
=M −1
hk .
k=1
By applying the norm ∥ · ∥p,M , we deduce that
+∞
X ∥h∥p,M
∥Mh−1 − M −1 ∥p,M ≤ ∥M −1 ∥p,M ∥h∥p,M = ∥M −1 ∥p,M −→ 0.
1 − ∥h∥p,M ∥h∥p,M →0
k=1
It shows that the function Inv is continuous.

X an
Proposition 4.1.27 (Exponential). Let a ∈ E, the exponential series is absolutely
n! n
convergent. The sum of this series is called the exponential of a and is denoted by exp(a).
X ∥a∥n
Proof. Let a ∈ E, the numerical series converges, and the claim follows.
n
n!

Remark 4.1.28. We can thus define the exponential of a square matrix, and use it in the
resolution of linear systems of differential equations.
4.2. COMPACTNESS 151

4.1.4 Extension by Continuity

For a continuous function f defined on X, it is obvious that any restriction f |U on any subset
U ⊂ X is also continuous. Furthermore, if f possesses stranger properties of continuity, like
uniform or Lipschitz continuity, those properties hold automatically on f |U .
The things will become less obvious if we concern on the reciprocal statements. This
question consist to ask weather a continuous function U can be extended into a continuous
function on X? We can except that some additional conditions are necessary for the function
f or for the spaces where f is defined.
In this section, we present the answer of this problem. The key argument are complete-
ness and uniform continuity. The following theorems are important, they will be used in
theoretical construction of integrals.

Theorem 4.1.29. Let U be a dense subset in the metric space (X, dX ) and f be a uniformly
continuous function mapping from U into a complete metric space (Y, dY ). Then, there
exists a unique continuous function g mapping from X into Y such that g|U = f . Moreover,
g is respectively uniformly continuous or K-Lipschitz continuous or isometric if f possesses
the corresponding property.

Proof. Left in Exercises 4.6.21 to 4.6.23.

Theorem 4.1.30. Let U be a dense subspace in the normed space (E, ∥ · ∥E ) and T be a
continuous linear application which maps from U into a Banach space (F, ∥ · ∥F ). Then,
there exists a unique continuous linear application Te which maps from E into F such that
Te|E = T . Moreover, Te is an isometry if T is too.

Proof. Left in Exercises 4.6.21 to 4.6.23.

Remark 4.1.31. Theorem 4.1.30 is a particular case of the famous Hahn-Banach Theo-
rem, which is fundamental in Functional Analysis.

4.2 Compactness
4.2.1 Sequence Definition
Definition 4.2.1. We say a metric space (X, d) is sequentially compact if any sequence
of elements in X possesses a convergent subsequence. And we say a subset A ⊂ X is sequen-
tially compact if any sequence of elements in A possesses a subsequence which converges to
a limit in A.

Proposition 4.2.2. Any sequentially compact subset in a metric space is complete.

Proof. Let A ⊂ X be a sequentially compact subset. Consider a Cauchy sequence (xn )n∈N
of elements in A. From the sequential compactness, (xn )n∈N possesses a subsequence which
converges to a limit in A. Thus, the sequence (xn )n∈N is convergent as a Cauchy sequence
possessing a convergent subsequence. Hence, the claim follows.
152 TOPOLOGY - FUNDAMENTAL THEORIES

Proposition 4.2.3. Let A be a closed subset in the metric space (X, d). Then, A is sequen-
tially compact if, and only if any nested decreasing sequence (Fn )n∈N of non-empty closed
subsets in A has a non-empty intersection.

Proof. We prove this proposition by double implications.

⇒: We suppose that the subset A ⊂ (X, d) is sequentially compact. Consider a nested

decreasing sequence (Fn )n∈N of non-empty closed subsets in A. Since Fn is supposed
to be non-empty for all n ∈ N, we then choose xn ∈ Fn ⊂ A. Then the sequence
(xn )n∈N is a sequence of elements in A, then it possesses a convergent subsequence
(xσ(n) )n∈N with l := lim xσ(n) ∈ A because A is sequentially compact. Since the
n→+∞
family of sets (Fn )n∈N is nested decreasing, we have then xσ(n) ∈ FN once σ(n) ≥ N .
Then, for all N ∈ N, the sequence (xσ(n) )σ(n)≥N is a convergent sequence of elements
in the closed set FN . From the sequence
\ characterization of closed sets, we have
l ∈ FN for all N ∈ N. Hence, l ∈ Fn and the family (Fn )n∈N has a non-empty
n∈N
intersection.

⇐: We suppose that for any nested decreasing sequence of non-empty closed subsets in
A has a non-empty intersection. Consider a sequence (xn )n∈N of elements in A. For
all n ∈ N, we set Fn = cl {xk | k ≥ n} ⊂ A since A is closed. It is clear that Fn is
non-empty,
\ closed and satisfying Fn+1 ⊂ Fn for all
\n ∈ N. From the assumption,
Fn ̸= ∅, we then consider the element l ∈ Fn ⊂ A. We now construct
n∈N n∈N
a subsequence (xq(n) )n∈N in (xn )n∈N by recurrence. We start with q(0) = 0. For
all j ∈ N, since l ∈ Fq(j)+1 = cl {xk | k ≥ q(j) + 1}, there exists then xq(j+1) with
q(j + 1) ≥ q(j) + 1 satisfying that d(xq(j+1) , l) ≤ 2−(j+1) . Hence, there exists a
subsequence (xq(n) )n∈N of (xn )n∈N such that xq(n) −→ l. Thus, the subset A is
n→+∞
sequentially compact.

Proposition 4.2.4. Let (xn )n∈N be a convergent sequence in (X, d), and let l be its limit.
Then, the set A := {xn | n ∈ N} ∪ {l} is sequentially compact.

Proof. Since (xn )n∈N is convergent, any subsequence of (xn )n∈N converges to the same limit.
In other words, l is the unique cluster point of the set {xn | n ∈ N}. So, A = cl {xn | n ∈ N}
is closed.
Let (Fn )n∈N be a nested decreasing sequence of non-empty closed subsets in A. We
consider the following cases

• l ∈ Fn for all n ∈ N;

• l∈
/ FN for certain N ∈ N.
\
In the first case, we have immediately that l ∈ Fn and hence (Fn )n∈N has a non-empty
n∈N
intersection.
In the second case, since FN is a closed subset and xn −→ l ∈ FN∁ , we deduce that
n→+∞
there exists p ∈ N such that xn ∈ FN∁ once n ≥ p . By the nested decreasing relation
4.2. COMPACTNESS 153

Fk ⊂ FN for all k ≥ N , we have therefore that if k ≥ N , Fk contains only a finite number

of terms of (xn )n∈N . In other words, Fk ⊂ A is a finite set once k ≥ N . Since Fk is always
supposed to be non-empty, then the sequence (F \n )n∈N is stationary from certain rank. That
means, Fk+1 = Fk for all k ≥ N ′ ≥ N . Thus, Fn = FN ′ ̸= ∅.
n∈N
In conclusion, the sequence (Fn )n∈N has a non-empty intersection. From Proposi-
tion 4.2.3, the set A is sequentially compact.

4.2.2 Covering Definition

Definition 4.2.5 (Covers). Let (X, d) be a metric space and F := {Ui ⊂ X | i ∈ I} ⊂ ℘(X)
denote a family of subsets in X indexed by an arbitrary non-empty set I. Consider A ⊂ X.
[
• We say the family F is a cover of A if A ⊂ Ui . In particular, if I is a finite set,
i∈I
we say F is a finite cover of A.

• If F is a cover of A, we say F is an open cover of A if, in addition, for all i ∈ I, Ui

is an open subset in X.

• We call a subcover of the cover F any sub-family F ′ ⊂ F where {Ui | i ∈ J} with

J ⊂ I which remains a cover of A.
[
• We say A is not finitely covered by F if for any finite subset J ⊂ I, A ⊈ Ui .
i∈J

In general topology, the compact sets are defined using the open covers.

Definition 4.2.6. Let A be a subset in a topological space (X, T ), we say A is (topolog-

ical) compact if for any open cover (Ui )i∈I of A, it possesses a finite subcover. That is, if
[ N
[
A⊂ Ui , then there exists σ(1), σ(2), · · · , σ(N ) ∈ I with N ∈ N∗ such that A ⊂ Uσ(k) .
i∈I k=1

Remark 4.2.7. We say the metric space (X, d) is compact (resp. sequentially compact) if
X is compact (resp. sequentially compact) in (X, d).

In the rest part of this subsection, we will prove the definitions of compactness using
sequences and covers are effectively equivalent when we work in metric spaces.

Proposition 4.2.8. Any compact set in a metric space is bounded and closed.

Proof.
[ Let A ⊂ (X, d) be a compact set. We consider the following open cover A ⊂ X ⊂
B(a, r) where a ∈ X be an arbitrary point. Since A is supposed to be compact, there
r>0
exists a finite subcover of this open cover. That means, there exists r1 < r2 < · · · < rp with
[p
∗
p ∈ N such that A ⊂ B(a, ri ) = B(a, rp ). Thus, A is bounded.
i=1
We then prove that A is closed. Let y ∈ A∁ , we consider the open cover
[
A ⊂ {y}∁ = B(y, r)∁ .
r>0
154 TOPOLOGY - FUNDAMENTAL THEORIES

N
[
From the compactness of A, there exists 0 < r1 < · · · < rN such that A ⊂ B(y, ri )∁ . By
i=1
the relation B(y, r)∁
⊂ B(y, r′ )∁ r′
for all 0 < < r, we have that A ⊂ B(y, r1 )∁ . Passing to
the complementary, we have thus B(y, r1 ) ⊂ A∁ . This implies that A∁ is an open set and
hence A is a closed set.

Lemma 4.2.9. Any topological compact set is sequentially compact.

Proof. Let A be a compact set and (xn )n∈N be a sequence of elements in A.

For all n ∈ N, consider the following sets, Fn = cl({xk | k ≥ n}). As the immediate
consequences of this construction, we obtain the properties, for all n ∈ N,

• Fn is closed since it is the closure of a subset;

• Fn ⊂ A = A since A is closed by its compactness and Proposition 4.2.8;

• Fn+1 ⊂ Fn .
\
Define now F as the intersection of all Fn , i.e., F = Fn ⊂ A.
n∈N

1.) We assume at first that F is not empty, then for any l ∈ F , l ∈ Fn for all n ∈ N. We
can thus construct a subsequence (xq(n) )n∈N by induction.

• q(0) = 0.
• For all n ∈ N, since l ∈ Fq(n)+1 = cl({xk | k ≥ q(n) + 1}), there exists q(n + 1) ∈
N, q(n + 1) ≥ q(n) + 1 such that d(xq(n+1) , l) ≤ 2−n .

This construction shows that the subsequence xq(n) converges to l ∈ A.

2.) We assume now F = ∅. For all n ∈ N, we consider the set Un = Fn∁ . Then, Un is
open as the complementary of a closed set. We have also,
!∁
[ [ \
U= Un = Fn∁ = Fn = F ∁ = ∅∁ = X.
n∈N n∈N n∈N
[
This shows that Un is an open cover of A. From the definition of compact sets,
n∈N
p
[
there exists a finite open subcover of this cover. That means, A ⊂ Uσ(i) where
i=1
p ∈ N and ∀ i ∈ [[1, p]], σ(i) ∈ N.
The relation Fn+1 ⊂ Fn implies that Un ⊂ Un+1 . Let q = max {σ(i) | i ∈ [[1, p]]} ∈ N,
we obtain in fact, A ⊂ Uq . We can thus deduce that Fq ⊂ A∁ which means that the
sequence (xn )n∈N has at most q terms and this is absurd.

In conclusion, A is sequentially compact.

4.2. COMPACTNESS 155

Definition
[ 4.2.10. We say a subset A ⊂ X is precompact if for all ε > 0, the open cover
A⊂ B(x, ε) possesses a finite open subcover. That means, there exists x1 , x2 , · · · , xp ∈
x∈A
p
[
A, p ∈ N such that A ⊂ B(xi , ε).
i=1

Notice that the points xi and the number p depend on ε a priori.

Proposition 4.2.11. Any subset in a precompact set is precompact.

Proof. Let A ⊂ X be a precompact set and let S ⊂ A. For all ε > 0, we define the open
cover FS = {B(x, ε) | x ∈ S} of S. We will show n that εFS possesses
o a finite open subcover.
Consider now the open cover of A: FA = B(x, ) x ∈ A . Since A is precompact,
2
p
[ ε
there exists finite points a1 , a2 , · · · , ap such that A ⊂ B(ai , ). Without of losing the
2
i=1
ε
generality, we may assume that m ≤ p and for all i ∈ [[1, m]], B(ai , ) ∩ S ̸= ∅ also
2
m
[ ε
S⊂ B(ai , ).
2
i=1
ε
We then pick for i ∈ [[1, m]], bi ∈ B(ai , ) ∩ S. Since
2
ε ε
diam(B(ai , ) ∩ S) ≤ diam(B(ai , )) = ε,
2 2
ε
we have then B(ai , ) ∩ S ⊂ B(bi , ε). Thus,
2
m m
[ ε [
S⊂ B(ai , ) ⊂ B(bi , ε),
2
i=1 i=1

where for all i ∈ [[1, m]], bi ∈ S.

So, {B(bi , ε) | i ∈ [[1, m]]} is a finite open subcover of FS .

Lemma 4.2.12. Any sequentially compact subset is precompact and complete.

Proof. Let A be a sequentially compact subset in X. The completeness of A follows from

Proposition 4.2.2. It remains to prove A is precompact.
We assume in the contrary that A is not precompact. Then, there exists ε > 0 such
that the cover {B(x, ε) | x ∈ A} does not possess any finite open subcover. We construct a
sequence (xn )n∈N by induction as follows.

• x0 ∈ A.
n
[
• For all n ∈ N, the union of balls B(xj , ε) could not cover A. Since if so, the
j=0
set {B(x, ε) | x ∈ A} has a finite open subcovers. That means, there exists xn ∈
[n
A\ B(xj , ε).
j=0
156 TOPOLOGY - FUNDAMENTAL THEORIES

The such constructed sequence (xn )n∈N satisfies d(xn , xj ) ≥ ε for all j ≤ n.
On the other hand, A is supposed to be sequentially compact, then there exists a con-
vergent subsequence (xq(n) )n∈N of (xn )n∈N . However, we have also d(xq(n) , xq(n−1) ) ≥ ε.
This shows the subsequence is even not a Cauchy sequence. This is a contradiction with
the convergence of (xq(n) )n∈N . In conclusion, A is precompact.

Lemma 4.2.13. If a subset is precompact and complete, then it is topological compact.

Proof. Let A ⊂ X. We suppose that A is precompact and complete. Consider F =
{Ui | i ∈ I} being an open cover of A. We assume in the contrary that A is not finitely
covered by F .
We construct a sequences (Fn )n∈N of subsets in A as well as a sequence (xn )n∈N ∈ AN
which satisfy the following induction property.
P(n): Fn ⊂ A, diam(Fn ) ≤ 2−n , Fn is not finitely covered by F . And xn ∈ Fn with
d(xn , xn+1 ) ≤ 2−n .
The construction is given as follows.
p0
[
• Since A is precompact, there exists a1 , a2 , · · · , ap0 ∈ A such that A ⊂ B(ai , 1).
i=1
Therefore, there exists q0 ∈ N, 1 ≤ q0 ≤ p0 such that B(aq0 , 1) is not finitely recovered
by F . Because if so, F possesses a finite subcover of A. We then set F0 = B(aq0 , 1)∩A
and x0 = aq0 .
• Assuming that Fn ⊂ A is already constructed, we know that Fn is precompact since A
is precompact and followed by Proposition 4.2.11. So, there exists a1 , a2 , · · · , apn ∈ Fn
[pn
such that Fn ⊂ B(ai , 2−(n+1) ). From the induction property, Fn is not finitely
i=1
covered by F . Using the same argument in the first step, we can find qn+1 ∈ [[1, pn ]]
such that the ball B(aqn+1 , 2−(n+1) ) is not finitely covered by F . We can thus set
Fn+1 = B(aqn+1 , 2−(n+1) ) ∩ A and xn+1 = aqn+1 . Then we can obtain that xn+1 =
aqn+1 ∈ Fn = B(aqn , 2−n ). Hence, d(xn , xn+1 ) ≤ 2−n .
By induction, P(n) is true for all n ∈ N, we have thus constructed such sequences.
X
It follows by the triangular inequality and the convergence of the series 2−n that the
n
sequence (xn )n∈N is a Cauchy sequence. Since A is complete, xn −→ a ∈ A.
n→+∞
Since F is a cover of A, there exists k ∈ I such that a ∈ Uk . Moreover, since Uk is
open, there exists r > 0 such that B(a, r) ⊂ Uk .
The convergences xn −→ a and 2−n −→ 0 imply that there exists N ∈ N such
n→+∞ n→+∞
r r
that d(xN , a) ≤ and that 2−N ≤ . So, FN ⊂ B(xN , 2−N ) ⊂ B(a, r) ⊂ Uk . This is a
4 4
contradiction with the fact that FN is not finitely recovered by F .

Theorem 4.2.14 (Borelvii -Lebesgueviii ). In a metric space, a subset is topologically compact

if, and only if it is sequentially compact.
vii
Émile Borel (1871 - 1956), French mathematician. He contributed fundamental works in measure theory
and probalility.
viii
Henri Lebesgue (1875 - 1941), French mathematician. He was famous in integration theory. The
“Lebesgue integral” is considered the most suitable integral theory nowadays.
4.2. COMPACTNESS 157

Proof. This is the summary of Lemmas 4.2.9, 4.2.12 and 4.2.13.

In what follows in this lecture note, since we will only considered metric spaces or normed
spaces, we say simply a compact set which means both it is compact via finite open covers
or via convergent subsequences.

4.2.3 Properties of Compacts Sets

Proposition 4.2.15. Any closed subset in a compact set is compact.

Proof. Let A ⊂ A′ ⊂ X with A being a closed set and A′ being a compact set. We consider
a sequence (an )n∈N ∈ AN of elements in A. We have immediately that (an )n∈N is also a
sequence of elements in A′ . By the compactness of A′ , there exists a subsequence (aσ(n) )n∈N
of (an )n∈N which is convergent in A′ . Then, the sequence (aσ(n) )n∈N is a convergent sequence
of elements in A. From the sequence characterization of closed sets, its limit belongs to A.
Thus, the sequence (an )n∈N possesses a convergent subsequence in A. In consequences, A
is compact.

Theorem 4.2.16. Let (X, dX ) and (Y, dY ) denote the metric spaces. And A ⊂ X, B ⊂ Y
denote the compact subsets in the corresponding metric spaces. Then, the Cartesian product
A × B is compact in the product space (X × Y, dX×Y ).

Proof. Let (xn , yn )n∈N ∈ (A × B)N be a sequence of elements in the set A × B, then
(xn )n∈N ∈ AN and (yn )n∈N ∈ B N . From the compactness of the set A, there exists a
subsequence (xφ(n) )n∈N of (xn )n∈N which converges in A. We remark that (yφ(n) )n∈N is
also a sequence of elements in B. From the compactness of B, there exists a subsequence
(yφ(ψ(n)) )n∈N of (yφ(n) )n∈N which converges in B. Since the sequence (xφ(ψ(n)) )n∈N is a
subsequence of the convergent sequence (xφ(n) )n∈N , so it converges in A.
In conclusion, the sequence (xφ(ψ(n)) , yφ(ψ(n)) )n∈N converges in A × B and hence A × B
is compact.

Theorem 4.2.17 (Bolzano-Weierstrass in R). A set of real numbers is compact if, and only
if it is bounded and closed.

Proof. From Proposition 4.2.8, any compact set is bounded and closed. Inversely, we con-
sider A ⊂ R being a bounded and closed subset in R. Let (an )n∈N ∈ AN , then (an )n∈N
is a bounded sequence in R. From the Bolzano-Weierstrass property in R, the bounded
sequence (an )n∈N has a subsequence (aφ(n) )n∈N which converges in R. Then applying the
sequence characterization of closed sets, this subsequence (aφ(n) )n∈N converges in A. Thus,
A is compact.

Theorem 4.2.18 (Bolzano-Weierstrass in (Rn , ∥ · ∥∞ )). In the normed space (Rn , ∥ · ∥∞ ),

a subset is compact if, and only if it is bounded and closed.

Proof. From Proposition 4.2.8, any compact set is bounded and closed.
Reciprocally, let A ⊂ Rn be a bounded and closed subset. From the boundedness of A,
there exists M > 0 such that A ⊂ B(0Rn , M ) = [−M, M ]n . The segment [−M, M ] ⊂ R is
bounded and closed in (R, | · |). From Theorem 4.2.17, [−M, M ] is a compact set in (R, | · |).
158 TOPOLOGY - FUNDAMENTAL THEORIES

We remark that the normed space (R2 , ∥ · ∥∞ ) is the product space of (R, | · |) and (R, | · |)
since the definition of ∥ · ∥∞ in R2 defines exactly the product distance of (R, | · |) × (R, | · |).
Using a simple induction, we can deduce that the normed space (R2 , ∥ · ∥∞ ) is exactly the
n-times product space of (R, | · |). We then use Theorem 4.2.16 to deduce that [−M, M ]n is
a compact space in (Rn , ∥ · ∥∞ ). Then, A can be considered as a closed subset in a compact
set. Hence, it is compact via Proposition 4.2.15.

Corollary 4.2.19 (Bolzano-Weierstrass in Kn ). In the normed space (Kn , ∥ · ∥p ) where

1 ≤ p ≤ ∞ and K = R or C, a subset is compact if, and only if it is bounded and closed.
Moreover, any bounded sequence in those normed spaces possesses at least one cluster point.
Proof. We consider at the first time the spaces (Rn , ∥ · ∥p ) for 1 ≤ p ≤ ∞. It follows from
the equivalence of p-norms in Rn Corollary 3.2.12 that ∥ · ∥p and ∥ · ∥∞ are equivalent so
they define the same topology. Since those norms define the same open sets, they define
the same open covers as well as the open subcovers. So, they define also the same compact
sets. Thus, the claim for the spaces (Rn , ∥ · ∥p ) follows from Theorem 4.2.18.
C −→ R2
We next consider the space (C, | · |). Introducing f : . It is
z 7−→ (Re(z), Im(z))
R2 −→ C
immediate that f is bijective with the inverse given by, f −1 : . Also,
(x, y) 7−→ x + iy
p
for all z ∈ C, |z| = Re(z)2 + Im(z)2 = ∥f (z)∥2 . This equality implies that if zk −→ z in
k→+∞
(C, | · |), then f (zk ) −→ f (z) in (R2 , ∥ · ∥2 ). Hence f is continuous mapping from (C, | · |)
k→+∞
into (R2 , ∥·∥2 ). The symmetric continuity result for f −1 holds also from the same argument.
Thus, f is an homeomorphism between (C, | · |) and (R2 , ∥ · ∥2 ). In the consequence, (C, | · |)
and (R2 , ∥ · ∥2 ) define the same topology. Using the result on (Rn , ∥ · ∥p ), the claim for
(C, | · |) follows.
The general result for (Cn , ∥ · ∥p ) follows then the same instructions. We consider firstly
the product space (C, ∥ · ∥∞ ) then we use the equivalent norms ∥ · ∥p on Cn . We can thus
obtain that the claim follows for (Kn , ∥ · ∥p ) for K = R or C and for all 1 ≤ p ≤ ∞.
Let (xk )k∈N be a bounded sequence in (Kn , ∥ · ∥p ). Then there exists M > 0 such that
xk ∈ B ∥·∥p (0Kn , M ) for all k ∈ N. The closed ball B ∥·∥p (0Kn , M ) is bounded and closed in
(Kn , ∥·∥p ). Then it is compact. So, the sequence (xk )k∈N possesses a convergent subsequence
in B ∥·∥p (0Kn , M ), and therefore (xk )k∈N has at least one cluster point.

Theorem 4.2.20. Let f : X → Y be a continuous function. Then, for any compact subset
A ⊂ X, its image by f , f (A) ⊂ Y , is compact.
Proof. Let (yn )n∈N be a sequence of elements in f (A). Then, there exists a sequence
(xn )n∈N ∈ AN such that yn = f (xn ) for all n ∈ N. From the compactness of A, there
exists a subsequence (xσ(n) )n∈N such that xσ(n) −→ x ∈ A. Thus, we remark that the
n→+∞
sequence (yσ(n) )n∈N is a subsequence of (yn )n∈N . From the continuity of f , we have that
yσ(n) = f (xσ(n) ) −→ f (x) ∈ f (A). Hence the set f (A) is compact.
n→+∞

Corollary 4.2.21. Let f be a continuous function which maps from X into Y and A ⊂ X
be compact.
4.2. COMPACTNESS 159

• If Y = R, then f (A) is bounded and there exists xm , xM ∈ A such that f (xm ) =

min f (x) = inf f (x) and f (xM ) = max f (x) = sup f (x).
x∈A x∈A x∈A x∈A

• If Y is a normed space, then f (A) is bounded and the supremum/infimum are achieved.

Proof. • From Theorem 4.2.20 and Proposition 4.2.8, f (A) is bounded and closed in R
and hence the claim follows.

• We consider the function x 7→ ∥f (x)∥Y . This function is continuous, as the composite

functions of two continuous functions f and ∥ · ∥Y , mapping from X into R. Then it
is sufficient to apply the first point.

Theorem 4.2.22 (Heineix ). Any continuous function defined on a compact set is uniformly
continuous on it.

Proof. Let f be a continuous function mapping from (X, dX ) into (Y, dY ) and let A ⊂ X
be a compact set.
We assume in the contrary that f is not uniformly continuous on A. Then, there exists
ε0 > 0 such that for all δ > 0, there exists xδ , yδ ∈ A such that dX (xδ , yδ ) ≤ δ and
1
dY (f (xδ ), f (yδ )) > ε0 . We then choose for example δ = for all n ∈ N∗ to deduce that
n
1
there exists the sequences (xn )n∈N and (yn )n∈N of elements in A satisfying dX (xn , yn ) ≤
n
and dY (f (xn ), f (yn )) > ε0 for all n ∈ N∗ .
Since (xn )n∈N ∈ AN and A is compact, the sequence (xn )n∈N possesses a subsequence
(xσ(n) )n∈N such that xσ(n) −→ a ∈ A. From the convergence of (xσ(n) )n∈N and the
n→+∞
1
relation dX (xn , yn ) ≤ −→ 0. We deduce that the sequence (yσ(n) )n∈N converges also
n n→+∞
to a. From the continuity of the function f at a ∈ A, we have that f (xσ(n) ) −→ f (a) as
n→+∞
well as that f (yσ(n) ) −→ f (a). In the consequences, it holds
n→+∞

dY (f (xσ(n) ), f (yσ(n) )) ≤ dY (f (xσ(n) ), f (a)) + dY (f (yσ(n) ), f (a)) −→ 0.

n→+∞

This is contradictory to the condition dY (f (xn ), f (yn )) > ε0 > 0 for all n ∈ N∗ . Hence, f
is uniformly continuous on A.

Theorem 4.2.23. Let f be a bijective function mapping from (X, dX ) into (Y, dY ), and let
A ⊂ X be a compact set. If f is continuous on A, then f −1 is continuous on f (A).

Proof. We suppose that f is continuous on A. Let U be a closed set in A. It follows from

Proposition 4.2.15 that U is a compact set. Applying now Theorem 4.2.20, we deduce that
f (U ) ⊂ Y is a compact set and thus f (U ) is a closed set from Proposition 4.2.8. Since
f is bijective, f (U ) is the pre-image of U by the function f −1 , we obtain therefore that
for any closed set U ⊂ f (A), U has a closed pre-image by f −1 . It follows from the set
characterization of continuous function Theorem 3.6.13 that f −1 is continuous.

ix
Eduard Heine (1821 - 1881), German mathematician. He worked in real analysis.
 160 TOPOLOGY - FUNDAMENTAL THEORIES

Corollary 4.2.24. Any bijective continuous function which maps from a compact metric
space (X, dX ) into a metric space (Y, dY ) is a homeomorphism.
Proof. Direct consequence of Theorem 4.2.23.

4.3 Connectedness
Let (X, d) be a metric space and A ⊂ X be a subset.

4.3.1 General Definition

Proposition 4.3.1. The following assertions are equivalent.
1.) If X = O1 ∪O2 with O1 , O2 to be open sets satisfying O1 ∩O2 = ∅, then either O1 = ∅
or O2 = ∅.
2.) If X = F1 ∪ F2 with F1 , F2 to be closed sets satisfying F1 ∩ F2 = ∅, then either F1 = ∅
or F2 = ∅.
3.) If A ⊂ X is simultaneously open and closed, then either A = ∅ or A = X.
4.) Any continuous function f : X → Z is constant.
Proof. We prove the following implications: 1.) ⇒ 2.) ⇒ 3.) ⇒ 4.) ⇒ 1.).
1.) ⇒ 2.): We assume that X = F1 ∪ F2 with F1 , F2 being closed sets and F1 ∩ F2 = ∅. Since
F1 ∩ F2 = ∅, we have then F1 ⊂ F2∁ as well as F2 ⊂ F1∁ . Thus, X = F1 ∪ F2 ⊂ F2∁ ∪ F1∁ .
Since X is the entire space, we have then X = F1∁ ∪ F2∁ . On the other hand, we pass
to the complementary on the condition X = F1 ∪ F2 to obtain that F1∁ ∩ F2∁ = ∅. We
remark also that the sets F1∁ and F2∁ are both open sets. Thus, from 1.) we have either
F1∁ = ∅ or F2∁ = ∅. And it is immediate to obtain that either F1 = ∅ or F2 = ∅.
2.) ⇒ 3.): Let A ⊂ X be open and closed, then A∁ is also open and closed. Since X = A ∪ A∁
and A ∩ A∁ = ∅, it follows from 1.) or 2.) we have either A = ∅ or A = X.
3.) ⇒ 4.): Let f be a continuous function mapping from X into Z. Let x0 ∈ X, and let n0 =
Z2 −→ R+
f (x0 ) ∈ Z. We endow the set Z with the metric dZ : where
(n, m) 7−→ |n − m|
| · | denote the absolute value in R. Then (Z, dZ ) is a metric space. We consider now
the set {n0 } ⊂ Z. The set {n0 } is a closed set since it is a singleton. Moreover,
we have also BZ (n0 , 1/2) = {n0 }. Thus, {n0 } is at the same time open and closed.
We then use the continuity of the function f to deduce that the set f −1 ({n0 }) is
open and closed in (X, d). From 3.), f −1 ({n0 }) is either empty or equal to X. Since
x0 ∈ f −1 ({n0 }), f −1 ({n0 }) is not empty, we then conclude that f −1 ({n0 }) = X and
hence f is a constant function.
4.) ⇒ 1.): We suppose that X = O1 ∪ O2 with O1 , O2 being open and O1 ∩ O2 = ∅. We consider
X −→ ( Z
the function characteristic 1O1 : 1 if x ∈ O1 . The function 1O1 is
x 7−→
0 if x ∈ O2
continuous since all open sets in (Z, dZ ) have the open pre-images in X. From 4.), the
function 1O1 is a constant function, we have then either 1O1 = 0 or 1O1 = 1. Those
two cases correspond exactly either O1 = ∅ or O2 = ∅.
4.3. CONNECTEDNESS 161

Definition 4.3.2. We say (X, d) to be a connected space if it satisfies one of the above
assertions.

The connectedness of the subset A can be in fact defined via induced topology (A, d|A ).
An equivalent definition is presented as follows.

Definition 4.3.3. We say that A is connected if for any open sets O1 , O2 in (X, d)
satisfying

A ⊂ O1 ∪ O2 , A ∩ O1 ∩ O2 = ∅,

then O1 ∩ A = ∅ or O2 ∩ A = ∅.
On the other side, we say A is disconnected or separated if A is not connected.

Lemma 4.3.4. If a connected subset C in X such that C ∩ int(A) ̸= ∅ and C ∩ ext(A) ̸= ∅,

then C ∩ ∂A ̸= ∅.

Proof. We recall that int(A), ext(A) and ∂A form a partition in X (Proposition 3.5.15).
That means, X = int(A) ∪ ext(A) ∪ ∂A and those three sets are pairwisely disjoint. Then,
(∂A)∁ = int(A) ∪ ext(A) with int(A) and ext(A) being both open. So, if C ∩ ∂A = ∅, then
C ⊂ (∂A)∁ = int(A) ∪ ext(A). And C ∩ int(A) ∩ ext(A) ⊂ int(A) ∩ ext(A) = ∅. From the
connectedness of C, we have either C ∩ int(A) = ∅ or C ∩ ext(A) = ∅. This is contradictory
to the assumption. By contraposition, the claim follows.

Theorem 4.3.5. The segment [0, 1] on R is connected.

Proof. Let [0, 1] = O1 ∪ O2 with O1 , O2 being open and O1 ∩ O2 = ∅. Without of losing

the generality, we assume that 0 ∈ O1 .
Consider now the set V defined by, V := {x ∈ [0, 1] | [0, x] ⊂ O1 }. And we set also
m := sup V ∈ [0, 1]. Let l ∈ [0, m), it follows from the characterization of supremum that
there exists x ∈ V such that l < x < m. Thus, [0, l] ⊂ [0, x] ⊂ O1 . We obtain thus,
[0, m) ⊂ O1 . We now consider the point m ∈ [0, 1]. If m ∈ O2 , since O2 is open, there exists
ε > 0, ε < m such that (m − ε, m + ε) ⊂ O2 . We have then (m − ε, m) ⊂ O1 ∩ O2 and
this is contradictory to the assumption O1 ∩ O2 = ∅. Hence, m ∈ O1 . In the consequences,
[0, m] = [0, m) ∪ {m} ⊂ V . Finally, if m < 1, there exists δ > 0, δ < 1 − m such that
(m − δ, m + δ) ⊂ O1 . We have also that [0, m + δ/2] = [0, m] ∪ [m, m + δ/2] ⊂ O1 . It implies
that m + δ/2 ∈ V and it is contradictory to the definition of m.
In conclusion, m = 1 and thus, [0, 1] ⊂ O1 and hence [0, 1] ∩ O2 = ∅. In other words,
[0, 1] is connected.

Theorem 4.3.6. Let (Ai )i∈I be a family of connected subsets in (X, d).
\ [
• If Ai ̸= ∅, then Ai is connected.
i∈I i∈I
162 TOPOLOGY - FUNDAMENTAL THEORIES

n
[
• If I = {1, 2, · · · , n}, n ∈ N, and Ai ∩ Ai+1 ̸= ∅ for all i ∈ [[1, n − 1]], then Ai is
i=1
connected.
[
Proof. • We consider a continuous function f : Ai → Z. Since each Ai is connected,
i∈I \
we have then f is constant on each Ai . Take now x0 ∈ Ai , we obtain thus for all
i∈I [ [
i ∈ I, all x ∈ Ai , f (x) = f (x0 ). This shows that f is constant on Ai . Thus, Ai
i∈I i∈I
is connected.

• We prove this point by recurrence. The case n = 1 is trivial. Assuming that the
property holds for n ∈ N∗ , we consider the connected sets A1 , · · · , An , An+1 satisfying
n
[
Ai ∩ Ai+1 ̸= ∅ for all i ∈ [[1, n]]. the, the set B := Ai is connected by the recurrence
i=1
hypothesis. It follows from the fact ∅ ̸= An ∩ An+1 ⊂ B ∩ An+1 and the previous
n+1
[
result that B ∪ An+1 = Ai is connected.
i=1

Remark 4.3.7. The intersection of two connected sets is not in general connected.

Theorem 4.3.8. Let A ⊂ X be a connected subset and f : X → Y be a continuous function.

Then, f (A) is connected.

Proof. Consider a continuous function g : f (A) → Z, then the function g ◦ f : A → Z is

also continuous. From the connectedness of A, the function g ◦ f is constant on A. Thus,
g is constant on A, this implies the connectedness of f (A).

4.3.2 Arc Connectedness

Definition 4.3.9. Let a, b ∈ X, we call an arc or a path joining a and b any continuous
function f : [0, 1] → X such that f (0) = a and f (1) = b in the case of existence.

Definition 4.3.10. We say the subset A is arc-connected or path-connected if for any

pair of points (a, b) ∈ A2 , there exists an arc joining a and b.

Theorem 4.3.11. • If the subset A is arc-connected, then A is connected.

• The reciprocal statement is true if A is an open set in a normed space (E, ∥ · ∥).

Proof. • Let f be a continuous function mapping from A into Z and let x0 ∈ A. For
all x ∈ A, since A is arc-connected, there exists a continuous function φ defined [0, 1]
such that φ(0) = x0 and φ(1) = x. Thus, the function f ◦ φ is a continuous function
mapping from [0, 1] into Z. From the connectedness of [0, 1] ⊂ R, the function f ◦ φ
is a constant function, which implies that f (x) = f ◦ φ(1) = f ◦ φ(0) = f (x0 ). This
relation holds for all x ∈ A. Hence f is a constant on A so A is connect.
4.3. CONNECTEDNESS 163

• Let A ⊂ E be an open subset. Let x0 ∈ A, we denote by A0 the set of all the points
which are arc-connected with x0 . That is

A0 := {x ∈ A | ∃φ ∈ C([0, 1], A) s.t. φ(0) = x0 , φ(1) = x} .

We prove at first that A0 is an open set. Let x ∈ A0 ⊂ A, there exists then a

continuous function φ defined on [0, 1] such that φ(0) = x0 and φ(1) = x. Since A is
an open set, there exists then r > 0 such that B(x, r) ⊂ A. Then, let y ∈ B(x, r), we
consider the function ψ defined on [0, 1] as follows,
(
φ( 2t ) if 0 ≤ t ≤ 12
ψ(t) =
2tx + (2t − 1)(y − 2x) if 12 ≤ t ≤ 1

Thus, the function ψ is continuous on both segments [0, 21 ] and [ 21 , 1]. And ψ has a
common limit at 12 from left and from right, so it is continuous on [0, 1]. Moreover, for
all t ∈ [0, 21 ], ψ(t) = φ( 2t ) ∈ A as well as for all t ∈ [ 21 , 1], ψ(t) = 2tx+(2t−1)(y −2x) ∈
B(x, y) ⊂ A. We obtain thus ψ ∈ C([0, 1], A). From the definition of A0 , we obtain
that B(x, r) ⊂ A0 and hence A0 is an open set.
On the other hand, we consider the set A0 ∩ A. Let y ∈ A0 ∩ A ⊂ A, there exists
r > 0 such that B(y, r) ⊂ A. Also, since y ∈ A0 , B(y, r) ∩ A0 ̸= ∅. There exists
then x ∈ A0 ∩ B(y, r). We can thus construct exactly by the same way to conclude
that y ∈ A0 . It implies that A0 = A0 ∩ A. This shows that A0 is open and closed in
the induced normed space (A, ∥ · ∥A ). From the connectedness of A, we have wither
A0 = A or A0 = ∅. It is obvious that x0 ∈ A0 . We conclude thus A0 = A.

Example 4.3.12. Let (E, ∥ · ∥) be a normed space.

• Convex sets. We say a subset A ⊂ E to be convex if for all pairs of points (a, b) ∈ A2 ,
the segment {at + b(1 − t) | 0 ≤ t ≤ 1} ⊂ A. Convex sets are arc-connected.

• Star-shaped sets. We say a subset A ⊂ E to be star shaped if there exists a ∈ A

such that for all b ∈ A, the segment {at + b(1 − t) | 0 ≤ t ≤ 1} ⊂ A. Star-shaped sets
are arc-connected.

• The set cl( (x, sin x1 ) 0 < x ≤ 1 ) ⊂ R2 is connected but is not arc-connected.


Theorem 4.3.13 (Connected subsets in R). Let A be a subset in R. Then, A is connected

if, and only if A is an interval.

Proof. We prove this theorem by double implications.

⇒: We assume in the contrary that A is not an interval. Then, there exists a, b, c ∈ R

such that a < c < b with a, b ∈ A and c ∈/ A. We consider then a function φ defined
by φ(x) = 1 if x > c and φ(x) = 0 if x < c. Then, φ is a continuous function mapping
from A into Z. Thus the contradiction with the connectedness of A. In conclusion, A
is an interval.

⇐: Any interval in R is connected since it is convex.

164 TOPOLOGY - FUNDAMENTAL THEORIES

Theorem 4.3.14 (Intermediate Values). Let f be a continuous function which maps from
X into R and A be a connected subset in X. If α, β ∈ f (A), then for all γ ∈ [α, β], there
exists t ∈ A such that f (t) = γ.

Proof. From Theorem 4.3.8, the set f (A) is a connected set in R. Then it is an interval by
Theorem 4.3.13. Since α, β ∈ f (A), from the connectedness of an interval in R, [α, β] ⊂
f (A). Hence the claim follows.

4.4 Finite Dimensional Normed Spaces

In this section, (E, ∥ · ∥) denotes a normed space.

Theorem 4.4.1. Any finite dimensional normed space is complete.

Proof. We prove this theorem by induction on the dimension of the vector space E. We set
the induction hypothesis P(n) for n ∈ N as follows. P(n): For any normed space (E, ∥ · ∥)
of dimension n, E is complete.
If n = 0, then E is reduced to the origin point E = {0E }. It is trivial that {0E } is
complete.
We assume now n > 0 and that P(n − 1) holds. Let (E, ∥ · ∥) be a normed space of
dimension n. We consider a non null linear form φ defined on E, then the kernel F := ker(φ)
is a subspace of dimension n−1, i.e. a hyperplane. From the induction hypothesis P(n−1),
it follows that (F, ∥ · ∥F ) is complete where ∥ · ∥F designs the induced norm of ∥ · ∥ on F . In
particular, F is a complete subset in the normed space E, and thus it is closed byuUsing
Proposition 4.1.10. In addition, we can deduce from Proposition 3.6.33 that φ is continuous
on E.
Let ê ∈ E such that φ(ê) = 1, we remark that it is licit since φ is supposed to be
K × F −→ E
non null. We consider now the function u : . This function u is
(λ, y) 7−→ λê + y
E −→ K×F
bijective and its inverse is given by u−1 : . Since the linear
x 7−→ (φ(x), x − φ(x)ê)
form φ is continuous, we deduce that u and u−1 are both continuous. Hence, u defines a
homeomorphism between E and K × F . Using Theorem 4.1.12, the space K × F is complete
and therefore E is complete. Thus P(n) holds.

Corollary 4.4.2. Any finite dimensional subspace of a normed space is closed.

Proof. Let F ⊂ E be a finite dimensional subspace. From Theorem 4.4.1, (F, ∥ · ∥F ) is

complete. We remark that for any Cauchy sequence in the induced normed space (F, ∥ · ∥F )
is also a Cauchy sequence in (E, ∥ · ∥). Thus, F is a complete subset in E. It follows then
from Proposition 4.1.10 that F is closed.

Proposition 4.4.3. Any linear form on a finite dimensional normed space is continuous.
4.4. FINITE DIMENSIONAL NORMED SPACES 165

Proof. Let φ be a linear form on a finite dimensional space. It is shown in Corollary 4.4.2
that the kernel ker(φ) is a closed subspace. Then the continuity of φ follows from Proposi-
tion 3.6.33.

Theorem 4.4.4. Let (E, ∥ · ∥E ) and (F, ∥ · ∥F ) be normed spaces. We assume that E has
finite dimensions. Then, any linear application which maps from E into F is continuous.

Proof. Let dim E = n ∈ N∗ and (e1 , · · · , en ) be a basis in E. We denote by (e∗1 , · · · , e∗n ) the
dual basis to (e1 , · · · , en ) in the dual spacex E ∗ . Then,
n
X
∀x ∈ E, x = e∗i (x)ei .
i=1

From Proposition 4.4.3, the linear forms e∗1 , · · · , e∗n are continuous. Let u be a linear appli-
cation defined on E, we have the following decomposition
n
X
∀x ∈ E, u(x) = e∗i (x)u(ei ).
i=1

Hence, the continuity of u follows from the continuities of e∗1 , · · · , e∗n .

Theorem 4.4.5 (Heine-Borel). In a finite dimensional normed space, the compact sets are
exactly the bounded and closed sets.

Proof. Let (E, ∥·∥) be a normed space of finite dimensions dim E = n ∈ N∗ and (e1 , · · · , en )
be a basis in E. We consider the isomorphismxi u from E into Kn of the representation in
the basis (e1 , · · · , en ). It follows from Theorem 4.4.4 that u and u−1 are both continuous
since E and Kn are finite dimensional. Hence, u defines also a homeomorphism between
(E, ∥ · ∥) and (Kn , ∥ · ∥∞ ). On the other hand, it follows from Corollary 4.2.19 that the
compact sets in (Kn , ∥ · ∥∞ ) are exactly the bounded and closed subsets. Since the space
(E, ∥ · ∥) is homeomorphic to (Kn , ∥ · ∥∞ ), then the claim follows.

Remark 4.4.6. In General Topology, we say a topological space satisfies the Heine-Borel
property if every bounded and closed set is compact.

Corollary 4.4.7 (General Bolzano-Weierstrass). Any bounded sequences of elements in a

finite dimensional normed space possesses a convergent subsequence.
x
We call the dual space of E and denote by E ∗ the space of all linear forms defined on E. If E has
finite dimensions, then E ∗ has the same dimensions as E, dim E = dim E ∗ . Moreover, let dim E = n ∈ N∗
n
X
and (e1 , · · · en ) be a basis in E. Then, for all x ∈ E, there exists a unique linear combination x = αi (x)ei
i=1
where ∀i ∈ [[1, n]], αi (x) ∈ K. We can prove (left in Exercises) that for all i ∈ [[1, n]], the application
e∗i : x 7→ αi (x) is linear, so e∗1 , · · · , e∗n ∈ E ∗ are called the dual basis to e1 , · · · , en .
xi
We call an isomorphism any bijective linear application. Following the previous footnote, we can show
E −→ Kn
(also left in Exercises) that the application u : is linear and bijective. We
x 7−→ (α1 (x), · · · , αn (x))
call the function u as the isomorphism of the representation in the basis (e1 , · · · , en )
166 TOPOLOGY - FUNDAMENTAL THEORIES

Proof. Direct consequences of Theorem 4.4.5.

Example 4.4.8. The unit closed ball B(0RN , 1) in the Banach space (ℓ1 (R), ∥ · ∥ℓ1 (R) ) (see
Example 3.2.15) is not compact.

Solution. We consider the sequence (s(k) )k∈N of elements in ℓ1 (R) defined by, ∀k ∈ N,
+∞
X
s(k) = (δk,n )n∈N xii . It is clear that for all k ∈ N, ∥s(k) ∥ℓ1 (R) = δk,n = 1, and thus
n=0
s(k) ∈ B(0RN , 1).
We assume in the contrary that there exists a subsequence (s(σ(k)) )k∈N which converges
in (ℓ1 (R), ∥ · ∥ℓ1 (R) ). We denote by s = (sn )n∈N ∈ ℓ1 (R) the limit of this subsequence. Then,
for all N ∈ N, we have
+∞
(σ(k))
X
|δσ(k),N − sN | = |sN − sN | ≤ |s(σ(k))
n − sn | = ∥s(σ(k)) − s∥ℓ1 (R) −→ 0.
k→+∞
n=0

On the other hand, we know that δσ(k),N = 0 for all σ(k) ̸= N and hence the sequence
(δσ(k),N )k∈N is stationary at 0 from the rank N + 1. We then conclude that sN = 0 for all
N ∈ N. Thus, s = 0RN .
However, we have always ∥s(k) − 0RN ∥ℓ1 (R) = 1. This is contradictory to the conver-
gence. So the sequence (s(k) )k∈N does not posses any convergent subsequence. Thus, the
set B(0RN , 1) in (ℓ1 (R), ∥ · ∥ℓ1 (R) ) is not compact.

Theorem 4.4.9 (Rieszxiii ). In a normed space E, the unit closed ball B(0E , 1) is compact,
if, and only if E has finite dimensions.

Proof. ⇐: If E is a finite dimensional normed space, it follows from Heine-Borel Theo-

rem 4.4.5 that the compact sets are exactly closed and bounded subsets. Then, the
unit closed ball is compact.

⇒: We assume that the closed ball B(0E , 1) in (E, ∥ · ∥) is compact, then the unit sphere
S(0E , 1) ⊂ B(0E , 1) is a closed subset in a compact set. From Proposition 4.2.15,
the unit sphere S(0E , 1) is also a compact set. We consider now the open cover
[ 1
S(0E , 1) ⊂ B(x, ). Since S(0E , 1) is compact, then the open cover possesses
2
x∈S(0E ,1)
a finite subcover. Thus, there exists x1 , · · · , xp ∈ S(0E , 1) with p ∈ N∗ such that
[ 1
S(0E , 1) ⊂ B(xi , ).
2
i∈[[1,p]]

Let F = Span(x1 , · · · , xp ) ⊂ E be the subspace of E generated by the unit vectors

x1 , · · · , xp . It is immediate that F has finite dimensions dim F ≤ p. Using Theo-
rem 4.4.1 and Corollary 4.4.2, (F, ∥ · ∥F ) is complete where ∥ · ∥F designs the induced
norm of ∥ · ∥ in F and hence F is closed in (E, ∥ · ∥). If F = E, then the proof is
finished. We assume therefore from now on F ⊊ E. There exists then x ∈ E \ F .
xii
Here, we used again the Kronecker symbols.
xiii
Federic Riesz (1880 - 1956), Hungarian mathematician. He and his brother Marcel Riesez made several
fundamental contributions in functional analysis.
4.4. FINITE DIMENSIONAL NORMED SPACES 167

It is immediate that x ̸= 0E since 0E ∈ F because F is a vector space. We deduce

x
then x̂ := ∈ S(0E , 1) \ F . Since F is closed, then r := d(x̂, F ) > 0 since if not,
∥x∥
d(x̂, F ) = 0 ⇔ x̂ ∈ F = F because F is closed. Using the characterization of infimum,
there exists y ∈ F such that
3
r ≤ s := ∥x̂ − y∥ ≤ r.
2
[ 1
Since S(0E , 1) ⊂ B(xi , ), there exist k ∈ [[1, p]] such that the unitary vector
2
i∈[[1,p]]
1 1
s−1 (x̂ − y) ∈ B(xk , ). That means s−1 (x̂ − y) = xk + h with ∥h∥ < . We have thus,
2 2
x̂ = y + sxk + sh.

We observe that y ∈ F and xk ∈ F . It implies that y + sxk ∈ F . In the consequences,

s 3
d(x̂, F ) ≤ ∥x̂ − (y + sxk )∥ = ∥sh∥ < ≤ r.
2 4
This is impossible since we have settled r = d(x̂, F ).
In conclusion, we have E = F and thus E has finite dimensions.

Theorem 4.4.10. All norms defined on a finite dimensional vector space are equivalent
from one to each others.

Proof. Let E be vector space of dimension n ∈ N∗ . We denote by (e1 , · · · , en ) ∈ E n a basis

n
X
of E. Then, for each x ∈ E, x can be written as x = xj ej where (x1 , · · · , xn ) ∈ Kn . We
j=1
introduce a norm ∥ · ∥1 defined as follows
n
X
∀x ∈ E, ∥x∥1 = |xj |.
j=1

We can show that ∥ · ∥1 defines a norm on E, the justification is left in Exercises.

Let ∥ · ∥ be a norm defined on E. It follows from the positivity and the separation
property that for all j ∈ [[1, n]], ∥ej ∥ > 0. And for all x ∈ E,

n
X n
X
∥x∥ = xj e j ≤ ∥ej ∥|xj | ≤ C∥x∥1 ,
j=1 j=1

where the constant C = max{∥e1 ∥, · · · , ∥en ∥} is independent of x. This inequality shows

also that the norm ∥ · ∥ is continuous on the normed space (E, ∥ · ∥1 ).
On the other hand, we assume in the contrary that there does not exist a constant c > 0
such that for all x ∈ E, c∥x∥1 ≤ ∥x∥. In other words, for all ε > 0, there exists xε ∈ E such
xε
that ∥xε ∥ < ε∥xε ∥1 . In the latter inequality, we can replace xε by , this is licit since
∥xε ∥1
168 TOPOLOGY - FUNDAMENTAL THEORIES

xε ̸= 0 if the inequality holds. And the condition becomes that there exists xε ∈ E with
∥xε ∥1 = 1 such that ∥xε ∥ < ε.
1
Moreover, we can take for example ε = for k ∈ N∗ . We have thus obtained a sequence
k
1
(x )k∈N such that for all k ∈ N , ∥x ∥1 = 1 and ∥x(k) ∥ ≤ . We remark that the unit
(k) ∗
∗ (k)
k
sphere in (E, ∥ · ∥) is a compact set since it is bounded and closed in a finite dimensional
vector space via Heine-Borel Theorem 4.4.5.
From the compactness of the unit sphere in (E, ∥ · ∥1 ), there exists a convergent subse-
quence of (x(k) )k∈N∗ . Let (y (k) )k∈N∗ be this subsequence and let z = lim y (k) . We have
k→+∞
then ∥z∥1 = 1. However, we have also ∥x(k) ∥ −→ 0. By the continuity of ∥·∥ on (E, ∥·∥1 ),
k→+∞
it follows that ∥z∥ = 0 and this leads to z = 0. It is contradictory with ∥z∥1 = 1.
In conclusion, there exists c > 0 such that for all x ∈ E, c∥x∥1 ≤ ∥x∥. Combining the
other inequality ∥x∥ ≤ C∥x∥1 , the norm ∥ · ∥ is equivalent to ∥ · ∥1 . By the transitivity of
the equivalence relations between norms, the claim follows.

4.5 Space of Operators

In this section, (E, ∥ · ∥E ) and (F, ∥ · ∥F ) denote two normed spaces. We consider the linear
applications from E into F .
Definition 4.5.1. We denote by L (E, F ) the set of linear and continuous applications
mapping from E into F . An element in L (E, F ) is also called as an operator. In the case
E = F , we denote it in short by L (E).

Proposition 4.5.2. L (E, F ) is a vector space.

Proof. Left in Exercise 4.6.1.

Let u be a linear application from E into F , we have seen in Theorem 3.6.29 that u is
continuous if, and only if u is bounded on the unit sphere in E. Using the linearity of u
and the homogeneity of the norms ∥ · ∥E , ∥ · ∥F , we can define a norm on L (E, F ) in the
following way.
Definition 4.5.3. We call the operator norm of a continuous linear application u map-
ping from E into F the following real number,
∥u(x)∥F
∥u∥L (E,F ) := sup ∥u(x)∥F = sup ∥u(x)∥F = sup
x∈E x∈E x∈E ∥x∥E
∥x∥E =1 ∥x∥E ≤1 x̸=0E

= inf {M > 0 | ∀x ∈ E, ∥u(x)∥F ≤ M ∥x∥E } .

Remark 4.5.4. The operator norm on L (E, F ) corresponds to the uniform convergence
topology for functions defined on the unit sphere of E. We will investigate more about this
concept in nest semester.

Proposition 4.5.5. Let u ∈ L (E, F ), then the operator norm ∥u∥L (E,F ) satisfies,
• ∀ x ∈ E, ∥u(x)∥F ≤ ∥u∥L (E,F ) ∥x∥E ,
4.5. SPACE OF OPERATORS 169

• if there exists K > 0 such that for all x ∈ E, ∥u(x)∥F ≤ K∥x∥E , then ∥u∥L (E,F ) ≤ K.
Proof. Left in Exercise 4.6.2.

L (E, F ) −→ R+
Proposition 4.5.6. The function ∥·∥L (E,F ) : defines a norm.
u 7−→ ∥u∥L (E,F )
Proof. Left in Exercise 4.6.3.

Proposition 4.5.7. Let (G, ∥ · ∥G ) be a normed space and f ∈ L (E, F ), g ∈ L (F, G).
Then, g ◦ f ∈ L (E, G) and we have,
∥g ◦ f ∥L (E,G) ≤ ∥g∥L (F,G) ∥f ∥L (E,F ) .
Proof. Left in Exercise 4.6.4.

Theorem 4.5.8. If (F, ∥ · ∥F ) is a Banach space, then (L (E, F ), ∥ · ∥L (E,F ) ) is also a

Banach space.
Proof. Let (un )n∈N be a Cauchy sequence in (L (E, F ), ∥ · ∥L (E,F ) ).
Let x ∈ E, the application u 7→ u(x) is linear. Also, it is ∥x∥-Lipschitz continuous
mapping from (L (E, F ), ∥ · ∥L (E,F ) ) into (F, ∥ · ∥F ). Thus, the sequence (un (x))n∈N is a
Cauchy sequence in the complete space (F, ∥ · ∥F ). By the completeness of F , the sequence
(un (x))n∈N converges. We consider then the function u defined by,
∀x ∈ E, u(x) = lim un (x).
n→+∞

It is immediate that u is linear on E from the linearity of passing to the limit.

We next prove that u is continuous. Since (un )n∈N is a Cauchy sequence, then it is
bounded. That is, there exists M > 0 such that for all n ∈ N, ∥un ∥L (E,F ) ≤ M . From the
definition of operator norm, we have that
∀n ∈ N, ∀x ∈ S(0E , 1), ∥un (x)∥F ≤ ∥un ∥L (E,F ) ≤ M,
where SE (0, 1) designs the unit sphere in E. We then pass n → +∞ in the above inequality
to obtain that
∀x ∈ S(0E , 1), ∥u(x)∥F ≤ M.
It proves that u is bounded on the unit sphere S(0E , 1). Using Theorem 3.6.29, u is
continuous and hence, u ∈ L (E, F ).
We then prove that ∥un − u∥L (E,F ) −→ 0. Let ε > 0, there exists N ∈ N such that
n→+∞
for all n, p ∈ N, n ≥ N , we have ∥un − un+p ∥L (E,F ) ≤ ε. Then, for all x ∈ S(0E , 1),
∥un (x) − un+p (x)∥F ≤ ε. We then pass p → +∞ to obtain that ∥un (x) − u(x)∥F ≤ ε. From
the definition of operator norm, we have that, for all n ≥ N
∥un − u∥L (E,F ) = sup ∥un (x) − u(x)∥F ≤ ε.
x∈S(0E ,1)

It shows that ∥un − u∥L (E,F ) −→ 0.

n→+∞
In conclusion, there exists u ∈ L (E, F ) such that ∥un − u∥L (E,F ) −→ 0. Thus,
n→+∞
(L (E, F ), ∥ · ∥L (E,F ) ) is complete.
170 TOPOLOGY - FUNDAMENTAL THEORIES

In the case where E and F are finite dimensional spaces, we have seen in Theorem
4.4.4 that all linear applications from E into F are continuous. Thus, the studies on the
operators mapping between finite dimensional normed spaces can be reduced into the studies
on matrix norms, which are already introduced in Example 3.2.14.
We assume now E = (Kn , ∥·∥) where K = R or C and since all norms in finite dimensional
normed spaces are equivalent, it is not necessary for us to precise the norm ∥ · ∥ here. Then,
L (E) = Mn×n (K) (square matrices of order n with coefficients in K).

4.6 Exercises
Exercise 4.6.1. Justify Proposition 4.5.2.

Exercise 4.6.2. Justify Proposition 4.5.5.

Exercise 4.6.3. Justify Proposition 4.5.6.

Exercise 4.6.4. Justify Proposition 4.5.7.

Exercise 4.6.5. Let (E, ∥·∥) be a Banach space. We consider a nested decreasing sequence
of closed balls (Bn )n∈N . That is, for all n ∈ N, Bn = B(xn , rn ) and Bn+1 ⊂ Bn .

1.) Show that for all n, m ∈ N, n ≤ m, we have ∥xn − xm ∥ ≤ rn − rm .

2.) Show that the sequence of real numbers (rn )n∈N converges.

3.) Show that the sequence (xn )n∈N ∈ E N converges.

\
4.) Show that there exists x ∈ E and r ∈ R+ such that Bn = B(x, r).
n∈N

Exercise 4.6.6. Let (E, ∥ · ∥) be a Banach space and T be a contraction mapping from E
into itself.

1.) Let y ∈ E, show that the equation x = T (x) + y of unknown x has a unique solution
in E.

2.) We define the mapping S := IdE − T . Show that S is bijective from E into E.

3.) Show that S −1 is a Lipschitz continuous function on E.

Exercise 4.6.7. Let (E, ∥ · ∥) be a normed space of finite dimension. Let F be a non-empty
closed subset in E.

1.) Let x ∈ E, show that the distance d(x, F ) is achieved. (i.e., there exists y ∈ F such
that ∥x − y∥ = d(x, F ).)

2.) Let K be a compact set in E, show that the distance d(K, F ) is achieved.
4.6. EXERCISES 171

3.) Is the distance between two non-empty closed sets achieved?

4.) Let f be a continuous function mapping from R into R. Show that for any point
P ∈ R2 , the distance between P and the graph Γ of f is achieved.

Exercise 4.6.8. Let (E, ∥ · ∥) be a normed space, A ⊂ E be a closed subset and B ⊂ E be

a compact subset.

1.) Show that A + B := {x + y | x ∈ A, y ∈ B} is closed.

2.) If B is only a closed set, does A + B remain closed?

Exercise 4.6.9. Let (E, ∥ · ∥) be a normed space and K be a compact subset which does
not contain 0E .

1.) Show that the set F := {λx | λ ∈ R+ , x ∈ K} is closed.

2.) Does the set F remain closed if K is just a closed subset?

3.) Same question if K contains 0E .

Exercise 4.6.10. Consider C([a, b], R) the space of continuous functions mapping from the
segment [a, b] into R.
Z b
1.) Is C([a, b], R) complete for the norm N1 : f 7→ |f (t)|dt?
a

2.) Same question for the norm N∞ : f 7→ sup |f (t)|.

t∈[a,b]

Exercise 4.6.11. Let n ∈ N, we denote by O(n) the set of orthogonal matrices of order n.
We recall that M ∈ O(n) ⇔ t M M = Id.
Give a suitable norm on the space of matrices Mn (R) and show that O(n) is a compact set
in Mn (R).

Exercise 4.6.12. Let K be a compact subset in a metric space (X, d). Let f be a function
mapping from K into itself and satisfying

∀ x, y ∈ K, x ̸= y ⇒ d(f (x), f (y)) < d(x, y).

1.) Consider the function x 7→ d(x, f (x)), show that f possesses a unique fixed point
α ∈ K.

2.) Let x0 ∈ K, we consider the sequence (xn )n∈N defined by recurrence, ∀n ∈ N, xn+1 =
f (xn ) and vn = d(xn , α). Show that the sequence of real numbers (vn )n∈N converges
to vl ≥ 0.

3.) Show that (xn )n∈N has a convergent subsequence (xφ(n) )n∈N with the limit xl ∈ K.
Deduce also that vl = d(xl , α).
172 TOPOLOGY - FUNDAMENTAL THEORIES

4.) Show that the vφ(n) −→ d(f (xl ), α) and deduce that xl = α.
n→+∞

5.) Show that the sequence (xn )n∈N converges to α.

6.) Show that for all ε > 0, there exists N ∈ N such that for all x0 ∈ K and n ∈ N,
n ≥ N ⇒ d(f n (x0 ), α) ≤ ε.

Exercise 4.6.13. Let K be a compact subset in a metric space (X, d). Consider a function
f : K → K which satisfies

∀x, y ∈ K, d(f (x), f (y)) ≥ d(x, y).

The objective of this exercise is to show that f is a bijective isometry.

1.) Show that f is injective.

Let x0 ∈ K, we construct a sequence (xn )n∈N given by, ∀n ∈ N, xn = f n (x0 ).

2.) Show that (xn )n∈N possesses a Cauchy subsequence (xφ(n) )n∈N where φ : N → N
denotes an extraction.

3.) Show that we can extract again the subsequence (xφ(n) )n∈N into (xφ(n)
e )n∈N such that
e + 1) − φ(n))
the sequence of integers (φ(n e n∈N is still strictly increasing.

4.) Denote now ∀n ∈ N, p(n) := φ(n

e + 1) − φ(n).
e Show that xp(n) −→ x0 .
n→+∞

5.) Deduce that f is an isometry, i.e., ∀x, y ∈ K, d(f (x), f (y)) = d(x, y).

6.) Show that f is surjective. Conclusion.

Exercise 4.6.14. Let E = ℓ2 (N, R) be the space of square summable real sequences. For
all k ∈ N, we (k)
 denote by e the sequence (δk,n )n∈N .
1 (k) [
Show that e k ∈ N∗ {0E } is compact.
k

Exercise 4.6.15. Let B be a bilinear application mapping from E1 × E2 into F where

E1 , E2 , F are supposed to be finite dimensional normed spaces. We denote respectively
their norms by ∥ · ∥E1 , ∥ · ∥E2 and ∥ · ∥F .

1.) Assuming that there does not exist a constant K > 0 such that for all (x1 , x2 ) ∈
E1 × E2 , ∥B(x1 , x2 )∥F ≤ K∥x1 ∥E1 ∥x2 ∥E2 . Show that for all n ∈ N, there exists
(x1,n , x2,n ) ∈ S(0E1 , 1) × S(0E2 , 1) such that ∥B(x1,n , x2,n )∥F > n where S(0E1 , 1) and
S(0E2 , 1) denote the unit spheres respectively in E1 and E2 .

2.) Show that the sequences (x1,n )n∈N and (x2,n )n∈N possess the convergent subsequences
(x1,q(n) )n∈N and (x2,q(n) )n∈N .

3.) Show that there exists (x˜1 , x˜2 ) ∈ S(0E1 , 1) × S(0E2 , 1) such that the linear application
B(·, x˜2 ) : x1 7→ B(x1 , x˜2 ) is not bounded in a neighborhood of x˜1 .

4.) Deduce the contradiction by discussing the continuity of the linear application B(·, x˜2 ).
4.6. EXERCISES 173

5.) Show that there exists a constant K > 0 such that for all (x1 , x2 ) ∈ E1 × E2 ,
∥B(x1 , x2 )∥F ≤ K∥x1 ∥E1 ∥x2 ∥E2 .

Exercise 4.6.16. Let (E, ∥ · ∥) be a normed space and F ⊂ E be a finite dimensional

subspace in E.

1.) Show that F is closed.

2.) Let x ∈ E, show that the distance d(x, F ) is achieved and d(x, F ) = d(−x, F ).

3.) Show that if x ∈

/ F , then d(x, F ) > 0.

Exercise 4.6.17. Let (E, ∥ · ∥) be a real Banach space and (an )n∈N be a family of linearly
independent vectors in E. We set Vn := Span(a0 , · · · , an ).

1.) Show that there exists a sequence (µn )n∈N of strictly positive real numbers such that

1
∀n ∈ N, ∥µn+2 an+2 ∥ ≤ d(µn+1 an+1 , Vn ).
3
X
2.) Show that the series µn an converges.
n

+∞
X [
3.) Show that s := µn an ∈
/ Vn .
n=0 n∈N

4.) Show that a real Banach space has its dimension either finite or uncountably infinite.

Cantorxiv Set
Exercise 4.6.18. Let I be an interval in R. We assume that I is covered by the open
intervals (In )n∈N . In this exercise, we want to establish that
+∞
X
|I| ≤ |In |.
n=0

Here, |I| ∈ R+ represents the length of the interval I. Precisely speaking, we set
(
|b − a| if I is bounded with endpoints a and b,
|I| =
+∞ if I is unbounded.

NS
[
1.) Let S = [a, b] ⊂ I, show that there exists NS ∈ N such that S ⊂ Ij .
j=0

xiv
Georg Cantor (1845-1918), German mathematician. His set theory is considered as the foundation of
modern mathematics. The notions of countability of sets was introduced by Cantor.
174 TOPOLOGY - FUNDAMENTAL THEORIES

NS
X
2.) Using mathematical induction on the number NS , show that |S| ≤ |Ij |.
j=0
[
3.) Show that I = Sp where for each p ∈ N, Sp is a segment and Sp ⊂ Sp+1 .
p∈N

4.) Conclusion.

Exercise 4.6.19. Let (rn )n∈N ∈ RN . Consider the following set

+∞
[ 
1 1
U := rn − n+1 , rn + n+1 .
2 2
n=0

1.) Show that U is an open set containing all elements of (rn )n∈N and strictly included
in R.
2.) Using the previous result, show that there does not exits any real sequences which
contain all real numbers.
3.) Show that there exists a sequence (qn )n∈N containing all rational numbers. Then
deduce that there exists an open set U such that Q ⊂ U ⊊ R.

Exercise 4.6.20. Consider the Cantor set,

+∞
( )
N∗
X cn
K := x ∈ R ∃(cn )n∈N ∈ {0, 2} , x = .
3n
n=1

1.) Justify K ⊂ R is well-defined.

Let (Uj )j∈J be a open covering of K in R. We assume in the contrary that K does
not have a finite subcover of (Uj )j∈J .
Let p ∈ N∗ , a = (a1 , · · · , ap ) ∈ {0, 2}p , we denote by Ka the subset of K,
 
p +∞
 ∗
X an X cn 
Ka := x ∈ K ∃(cn )n∈N ∈ {0, 2}N , x = + .
 3n 3n 
n=1 n=p+1

2.) Show that K = K(0) ∪ K(2) and deduce that for all a = (a1 , · · · , ap ) ∈ {0, 2}p ,

Ka = K(a1 ,··· ,ap ,0) ∪ K(a1 ,··· ,ap ,2) .

∗
3.) Show that there exits a sequences (an )n∈N ∈ {0, 2}N such that for all p ∈ N∗ ,
K(a1 ,··· ,ap ) does not have a finite subcover of (Uj )j∈J .
+∞
X an
4.) Consider now x0 := ∈ K. Show that there exists j0 ∈ J and p ∈ N∗ such that
3n
n=1

1
K(a1 ,··· ,ap ) ⊂ B(x0 , ) ⊂ Uj0 .
3p
5.) Show that K is a compact set.
4.6. EXERCISES 175

Extension by Continuity
In what follows, (X, dX ) designs a metric space and (Y, dY ) refers to a complete metric
space.

Exercise 4.6.21. Let f be a function mapping from U ⊂ X into Y and a ∈ U . Show that
if f possesses a limit at a, then f satisfies

∀ε > 0, ∃δ > 0 s.t. ∀x, y ∈ BX (a, δ) ∩ U, dY (f (x), f (y)) ≤ ε.

This condition is called the Cauchy criteria of continuity.

Exercise 4.6.22. We assume that f satisfies the Cauchy criteria of continuity at a.

1.) Let (xn )n∈N be a sequence of elements in U which converges to a. Show that the
sequence (f (xn ))n∈N converges to a limit l ∈ Y .

2.) Following the previous question, let (yn )n∈N be another sequence of elements in U
which converges also to a. Show that (f (yn ))n∈N converges to the same limit l ∈ Y .

3.) Conclusion.

Exercise 4.6.23. We assume here that U is dense in X and f is uniformly continuous on

U.

1.) Show that f satisfies the Cauchy criteria of continuity at any point a ∈ X.

2.) Define the extension of f on X, which means, a function g : X → Y such that g|U = f .

3.) Show that g is uniformly continuous on X.

Hint: we can use the sequences xn −→ x and yn −→ y.
n→+∞ n→+∞

4.) We assume in addition that f is K-Lipschitz continuous on U for K > 0. Show that
g is also K-Lipschitz continuous on X.

5.) Now, f is a continuous linear application mapping from U ⊂ E into F . We suppose

that U is a dense vector subspace in the normed space E and F is a Banach space.
Show that f can be uniquely extended into a continuous linear application mapping
from E into F .

Exercise 4.6.24. Let f be a derivable function defined on (0, 1]. We assume that f ′ is
bounded on (0, 1].

1.) Show that f can be extended into a Lipschitz continuous function on [0, 1].

2.) We assume furthermore f ′ possesses a right-limit at 0. Show that f can be extended

into a derivable function on [0, 1].

Exercise 4.6.25. 1.) State the Banach Fixed Point Theorem.

176 TOPOLOGY - FUNDAMENTAL THEORIES

2.) Prove the Banach Fixed Point Theorem.

X −→ X
3.) Let f be a contraction in (X, d). Show that the function g : is
x 7−→ x + f (x)
continuous and bijective.
Hint: For the surjectivity, you may consider the function x 7→ y − f (x).

Exercise 4.6.26. Let (X, d) be a complete metric space and f be a function mapping from
X into X. We assume that there exists p ∈ N∗ such that f p is a contraction. Show that f
has a unique fixed point.
Hint: Here, f p designs the p-times iteration, f p = f ◦ f ◦ · · · ◦ f . Think on applying Banach
fixed point theorem.

Exercise 4.6.27. Let (E, ∥ · ∥) be a normed space. We suppose that any absolutely con-
vergent series of elements in E converges. Show that (E, ∥ · ∥) is a Banach space.
Hint: Let (xn )n∈N be a Cauchy sequence of elements in E, construct a subsequence
(xφ(n) )n∈N such that ∀n ∈ N, ∥xφ(n) − xφ(n+1) ∥ ≤ 2−n .

Exercise 4.6.28. Let Ω be an open set in (X, d), Ω ̸= X, and K ⊂ Ω be a compact

subset. Show that there exists δ > 0 such that Kδ ⊂ Ω. Here Kδ is defined by, Kδ :=
{x ∈ X | d(x, K) ≤ δ}.
Hint: We recall that the distance function x 7→ d(x, A) is continuous for any non-empty
subset A ⊂ X.

Wandering Points
In the study of dynamic systems, there is a concept called wandering points. Consider
f : Rn → Rn being a continuous function, n ∈ N∗ and K ⊂ Rn being a compact subset. We
assume that f (K) ⊂ K. A point p ∈ K is said to be a wandering point for f if there exists
a neighborhood U of p such that (f −m (U ))m∈N is a pairwisely disjoint family of sets.
Exercise 4.6.29. Let p ∈ K and M ∈ N∗ , we assume that p ∈
/ f M (K). Show that
(f −m ({p}))m∈N is a pairwisely disjoint family of sets.

Exercise 4.6.30. Following the previous Exercise, we want to show that in fact, p is a
wandering point.
1.) Denoting by Br the open ball B(p, r) centered at p with the radius r > 0, show that
there exists δ > 0 such that f M (K) ∩ Bδ = ∅.
2.) Assuming that p is not a wandering point for f , show that for all ε ∈ (0, δ), there exists
pε ∈ K and m1 (ε), m2 (ε) ∈ N, 0 ≤ m1 (ε) < m2 (ε) < M such that f m1 (ε) (pε ) ∈ Bε
also f m2 (ε) (pε ) ∈ Bε .
3.) Show that there exists m1 , m2 ∈ N with 0 ≤ m1 < m2 < M and a sequence (qn )n∈N ∈
K N such that f m1 (qn ) −→ p and f m2 (qn ) −→ p.
n→+∞ n→+∞

4.) Conclusion.
 Chapter 5

Inner Product Spaces

In this chapter, we consider the K-vector spaces denoted by E and F where K = R or C.

We start with a recall on linear applications.

5.1 Definitions
Definition 5.1.1. We say a function φ : E → F is linear if

∀α, β ∈ K, ∀u, v ∈ E, φ(αu + βv) = αφ(u) + βφ(v).

Definition 5.1.2. We call a bilinear form on E any application φ which maps from E ×E
into K satisfying

• for all y ∈ K, the application φ(·, y) : x 7→ φ(x, y) is linear,

• for all x ∈ K, the application φ(x, ·) : y 7→ φ(x, y) is linear.

Definition 5.1.3. We say a bilinear form φ on E is symmetric if

∀x, y ∈ E, φ(x, y) = φ(y, x).

Now we distinguish the cases weather K = R or K = C. We assume for the moment

K = R.

Definition 5.1.4. If E be a real vector space, we say a bilinear symmetric form φ is

positive if it satisfies,

∀x ∈ E, φ(x, x) ≥ 0.

Moreover, we say φ is positive definite if φ is positive and satisfies the separation property,

φ(x, x) = 0 if, and only if x = 0E .

Definition 5.1.5. We call an inner product or a scalar product on a real vector space
E any bilinear symmetric positive definite form.

177
178 INNER PRODUCT SPACES

In the case where K = C, the bilinear symmetric positive forms can not be simply
adapted. As an example, let φ be a complex bilinear symmetric form, then φ cannot be
positive. Taking x ∈ E such that φ(x, x) ̸= 0, we have then φ(eiθ x, eiθ x) = e2iθ φ(x, x). We
can thus choose θ such that φ(eiθ x, eiθ x) < 0.
So, we need to introduce the corresponding structures for complex vector spaces. We
assume that E is a complex vector space.

Definition 5.1.6. We call an application φ on E is semi-linear if

∀α, β ∈ C, ∀u, v ∈ E, φ(αu + βv) = αφ(u) + βφ(v).

Definition 5.1.7. We call a sesquilineari form on E any application φ which maps from
E × E into C satisfying

• for all y ∈ C, the application φ(·, y) : x 7→ φ(x, y) is semi-linear,

• for all x ∈ C, the application φ(x, ·) : y 7→ φ(x, y) is linear.

Remark 5.1.8. In different contexts, there are someone who define sesquilinear forms as an
application to be linear on the first variable and to be semi-linear on the second variable.

Definition 5.1.9. We say a sesquilinear form φ on E is Hermitianii if

∀x, y ∈ E, φ(x, y) = φ(y, x).

Definition 5.1.10. We say a sesquilinear Hermitian form φ is positive if it satisfies,

∀x ∈ E, φ(x, x) ≥ 0.

Moreover, we say φ is positive definite if φ is positive and satisfies the separation property,

φ(x, x) = 0 if, and only if x = 0E .

Definition 5.1.11. We call a Hermitian product on a complex vector space E any

sesquilinear Hermitian positive definite form.

Definition 5.1.12. We call a real inner product space (resp. complex) or a real pre-
Hilbertiii space (resp. complex) any real ([Link]) vector space E equipped with an
inner product (resp. Hermitian product) ⟨·, ·⟩. We denote this space by (E, ⟨·, ·⟩). In the
cases when we want emphasis or avoid confusions, we denote the inner product by ⟨·, ·⟩E .
i
The root word “sesqui-” from Latin stands for “one time and half”
ii
Charles Hermite (1822 - 1901), French mathematician. He worked in number theory, quadrature forms
and differential equations. He was one of the first persons to use matrices. He also proved firstly the
transcendency of the constant e.
iii
David Hilbert (1862 - 1943), German mathematician. One of the most influential mathematician in 19th-
20th century. He contributed in various branches of mathematics such as analysis, algebra and foundation
of mathematics. Famous for his 23 problems proposed in 1900.
5.1. DEFINITIONS 179

Definition 5.1.13. We call a Euclideaniv space or a Hermitian space respectively any

real or complex inner product space of finite dimensions.

Example 5.1.14. We present the often-occurred inner product spaces.

n
X
• Rn equipped with ⟨x, y⟩Rn = t xy = xi yi .
i=1
n
X
• Cn t
equipped with ⟨x, y⟩Cn = xy = xi yi .
i=1

• The square summable sequences ℓ2 (C) equipped with the inner product
+∞
X
⟨u, v⟩ℓ2 (C) := un vn .
n=1

We will see a more detailed analysis about this space in the chapter of Fourier series.

• Let I be a segment and let C(I, C) be the set consists of continuous functions on I.
We define the inner product associated to the square-integrable norm, given by,
Z
⟨f, g⟩L2 (I,C) := f (x)g(x)dx.
I

Solution. We only verify the definition of inner products in the case where K = C since the
case of real numbers are covered by those of complex numbers.

• – The right-linearity and the left-semi-linearity follows from the linearity of matrix
product.
– Following the previous point, the sesquilinear form ⟨·, ·⟩ is Hermitian from the
usual multiplication of complex numbers.
Xn
n
– Let x ∈ C such that ⟨x, x⟩ = 0. We have then |xi |2 = 0, which implies that
i=1
|xi | = 0 for all i ∈ [[1, n]]. In other words, x = 0Cn . So, the sesquilinear form ⟨·, ·⟩
is positive definite. In conclusion, ⟨·, ·⟩ defines an inner product on Cn .
X
• – We have seen in Lemma 3.2.16 that for all u, v ∈ ℓ2 (C), the series un vn
n
converges. Then the right-linearity and the left-semi-linearity follows from the
linearity of the sum of a series.
– Let u, v ∈ ℓ2 (C), it follows then,

+∞
X +∞
X
⟨u, v⟩ℓ2 (C) = un vn = vn un = ⟨v, u⟩ℓ2 (C) .
n=0 n=0

It is licit since all series mentioned above are convergent. Thus, ⟨·, ·⟩ℓ2 (C) is
Hermitian.
iv
Euclid (ca. 300 B.C.), Ancient Greek mathematician.
180 INNER PRODUCT SPACES

+∞
X
– Let u ∈ ℓ2 (C) such that ⟨u, u⟩ℓ2 (C) = 0, which implies that |un |2 = 0. We
n=0
deduce that un = 0 for all n ∈ N. Thus, u = 0CN and ⟨·, ·⟩ℓ2 (C) is positive-definite.
In conclusion, ⟨·, ·⟩ℓ2 (C) defines an inner product on ℓ2 (C).
• – The right-linearity and the left-semi-linearity follow from the linearity of inte-
gration.
– Let f, g ∈ C(I, C),
Z Z
⟨f, g⟩L2 (I) = f (x)g(xdx) = g(x)f (x)dx = ⟨g, f ⟩L2 (I) .
I I
Z
– Let f ∈ C(I, C) such that ⟨f, f ⟩L2 (I,C) = 0. We have then |f (x)|2 dx = 0. From
I
the continuity of f , we can deducev that f (x) = 0 for all x ∈ I, which implies
that ⟨·, ·⟩L2 (I,C) is positive-definite. In conclusion, ⟨·, ·⟩L2 (I,C) defines an inner
product on C(I, C).

5.2 Norms on Inner Product Spaces

Let (E, ⟨·, ·⟩) be a real or complex inner product space.
q
Definition 5.2.1. For all x ∈ E, we set ∥x∥ = ⟨x, x⟩.

Proposition 5.2.2 (Polarization Identity, real case). Let E be a real inner product space,
then for all x, y ∈ E,
1
∥x + y∥2 − ∥x∥2 − ∥y∥2


⟨x, y⟩ =
2
1
∥x∥2 + ∥y∥2 − ∥x − y∥2


=
2
1
∥x + y∥2 − ∥x − y∥2 .


=
4
Proof. Left in Exercise 5.5.1.

Proposition 5.2.3 (Polarization Identity, complex case). Let E be a complex inner product
space, then for all x, y ∈ E,
1
∥x + y∥2 − ∥x∥2 − ∥y∥2


Re ⟨x, y⟩ =
2
1
∥x∥2 + ∥y∥2 − ∥x − y∥2


=
2
1
∥x + y∥2 − ∥x − y∥2 ,


=
4
as well as
1
∥x + y∥2 − ∥x − y∥2 − i∥x + iy∥2 + i∥x − iy∥2 .


⟨x, y⟩ =
4
v
It is very recommend for all of you to prove this point in Exercises.
5.2. NORMS ON INNER PRODUCT SPACES 181

Proof. Left in Exercise 5.5.1.

Corollary 5.2.4 (Parallelogram Identity). It holds in real or complex inner product spaces
that for all x, y ∈ E,

∥x + y∥2 + ∥x − y∥2 = 2 ∥x∥2 + ∥y∥2

Proof. Left in Exercise 5.5.1.

Remark 5.2.5. We can see from Propositions 5.2.2 and 5.2.3 that the inner product can be
expressed in functions of norms. Given a normed space (E, ∥ · ∥), we can show that if the
parallelogram identity Corollary 5.2.4 holds on E, then it possesses an inner product which
is given by polarization identities Propositions 5.2.2 and 5.2.3.

Proposition 5.2.6 (Cauchy-Schwarz Inequality, complex case). Let E be a complex inner

product space, then for all x, y ∈ E,

|Re ⟨x, y⟩ | ≤ ∥x∥∥y∥,

while the equality occurs if, and only if x and y are linearly dependent on R.

| ⟨x, y⟩ | ≤ ∥x∥∥y∥,

while the equality occurs if, and only if x and y are linearly dependent on C.

Proof. • The case where y = 0E is trivial. We assume from now on y ̸= 0E and consider
R −→ R
the function f : . We have then from the positivity of
λ 7−→ |⟨x + λy, x + λy⟩|
⟨·, ·⟩ that for all λ ∈ R, f (λ) = ∥y∥2 λ2 + 2Re ⟨x, y⟩ λ + ∥x∥2 ≥ 0. We remark that f
is a polynomial function with second degree which is always positive on R. We know
then the discriminant to f is negative, and we obtain that |Re ⟨x, y⟩ | ≤ ∥x∥∥y∥. Also,
the equality occurs if, and only if f has a unique root on R. It can be easily seen that
this root is not 0. So, the equality occurs if, and only if there exists λ ∈ R∗ such that
x + λy = 0E . Equivalently speaking that x and y are linearly dependent on R.

• It is immediate that
D
iθ
E exists θ ∈ R such that ⟨x, y⟩ = | ⟨x, y⟩ |e . We have
there
thus, | ⟨x, y⟩ | = eiθ x, y ∈ R. From the previous result, we have then | ⟨x, y⟩ | =
D E
Re eiθ x, y ≤ ∥eiθ x∥∥y∥ = ∥x∥∥y∥. And the equality occurs if, and only if eiθ x and
y are linearly dependent on R. Equivalently speaking, x and y are linearly dependent
on C.

As a particular case of the above proposition, we have the following Cauchy-Schwarz

inequality for real inner product spaces.
182 INNER PRODUCT SPACES

Proposition 5.2.7 (Cauchy-Schwarz Inequality, real case). Let E be a real inner product
space, then for all x, y ∈ E,

| ⟨x, y⟩ | ≤ ∥x∥∥y∥,

while the equality occurs if, and only if (x, y) is linearly dependent.

Proposition 5.2.8. The function x 7→ ∥x∥ defines a norm on E, it is called as the Eu-
clidean norm or the Hermitian norm induced by the inner product ⟨·, ·⟩.

Proof. We verify if ∥ · ∥ satisfies the definition of a norm.

• The homogeneity comes from the bilinearity or sesquilinearity of the inner product.

• Let x, y ∈ E, it follows from Cauchy-Schwarz inequality Proposition 5.2.6 or Proposi-

tion 5.2.7 that

∥x + y∥2 = ∥x∥2 + ∥y∥2 + 2Re ⟨x, y⟩ ≤ ∥x∥2 + ∥y∥2 + 2∥x∥∥y∥ = (∥x∥ + ∥y∥)2 .

Thus, the triangular inequality follows by taking the square root.

• The separation property follows from the definite-positivity of ⟨·, ·⟩.

Proposition 5.2.9. For a fixed x ∈ E, the function y 7→ ⟨x, y⟩ is linear and continuous on
E.

Proof. The linearity comes from the definition. For the continuity, it follows from Cauchy-
Schwarz inequality and the linearity that the function y 7→ ⟨x, y⟩ is ∥x∥-Lipschitz continu-
ous, so continuous on E.

Proposition 5.2.10. Let (E, ⟨·, ·⟩) be a Euclidean or Hermitian space, then

• E is complete;

• the compact subsets in E are exactly the bounded and closed subsets.

Proof. It follows from the completeness of the field K = R or C and from Theorem 4.4.1
that any finite dimensional normed space is complete, so does E. Also, it follows from
Heine-Borel Theorem 4.4.5 that the compact sets in E are exactly bounded and closed
subsets.
5.3. ORTHOGONALITY 183

5.3 Orthogonality
Let (E, ⟨·, ·⟩) be a real or complex inner product space.

Definition 5.3.1. We say two vectors x, y ∈ E are orthogonal, and we denote this relation
by x ⊥ y, if ⟨x, y⟩ = 0.

Remark 5.3.2. • If x ⊥ y then y ⊥ x.

• x ⊥ x if, and only if x = 0.

• We say two subsets X and Y of E are orthogonal if for all (x, y) ∈ X × Y , x ⊥ y.

Definition 5.3.3. We call the orthogonal of a subset X of E the following subset,

X ⊥ := {y ∈ E | ∀x ∈ X, x ⊥ y} .

Remark 5.3.4. • We call the orthogonal of x ∈ E, and we denote simply by x⊥ , the

subset {x}⊥ . Moreover, we can consider x⊥ as the kernel of the linear form y → ⟨x, y⟩,
which is continuous. So, x⊥ is a closed hyperplane if x ̸= 0 and x⊥ = E if x = 0.

• X ⊥ Y if, and only if Y ⊂ X ⊥ or X ⊂ Y ⊥ .

• X ⊥ = (SpanX)⊥ .

• {0E }⊥ = E and E ⊥ = {0E }.

Proposition 5.3.5. Let X ⊂ E, then,

• X ⊥ is a closed vector subspace in E;

• if X ⊂ Y ⊂ E, then Y ⊥ ⊂ X ⊥ ;

• X ⊥ = cl(X)⊥ .

Proof. • Let λ ∈ K and x, y, z ∈ E. If x ⊥ y and x ⊥ z, it follows from the linearity of

y 7→ ⟨x, y⟩, that x ⊥ (y + λz). This shows that X ⊥ is stable by linear combination
and thus X ⊥ is a vector subspace. Let x ∈ X, the set x⊥ := {y ∈ E | ⟨x, y⟩ = 0} is a
closed set as\the pre-image of the closed set {0} by the continuous function y 7→ ⟨x, y⟩.
So, X ⊥ = x⊥ is closed as the intersection of closed sets.
x∈X

• Let y ∈ Y ⊥ . For all x ∈ X, we have immediately that x ∈ Y since X ⊂ Y . Then,

⟨x, y⟩ = 0. Hence, y ∈ X ⊥ .

• From X ⊂ cl(X), we have then cl(X)⊥ ⊂ X ⊥ . Reciprocally, let x ∈ X ⊥ , we have

X ⊂ x⊥ . From the first point, x⊥ is a closed set. Then, cl(X) ⊂ x⊥ . In other words,
x ∈ cl(X)⊥ .
184 INNER PRODUCT SPACES

Theorem 5.3.6 (Pythagorasvi ). Let x1 , x2 , · · · , xp ∈ E be vectors which are orthogonal

from one to each others, then
p 2 p
X X
xi = ∥xi ∥2 .
i=1 i=1

p 2 * p p
+
X X X
Proof. It is sufficient to calculate xi = xi , xi , then to apply the orthogo-
i=1 i=1 i=1
nality of the vectors x1 , x2 , · · · , xp .

Definition 5.3.7. Let (vj )j∈J be a family (the indexing set J could be finite, countable
even uncountable) of vectors in the inner product space E.
• We say the family is orthogonal if vi ⊥ vj for all i, j ∈ J and i ̸= j.

• We say the family is orthonormal if it is orthogonal and ∥vj ∥ = 1 for all j ∈ J.

Proposition 5.3.8. An orthogonal family of non null vectors is linearly independent.

Proof. Let (vj )j∈J be an orthogonal family of non null vectors. Consider the vectors
n
X
v1 , · · · , vn in the family and the linear equation 0E = αj vj of unknowns α1 , · · · , αn ∈ K.
j=1
We have then, for all i ∈ [[1, n]],
* n + n
X X
0 = ⟨0E , vi ⟩ = αj vj , vi = αj δj,i = αi .
j=1 j=0

Thus, the family (vj )j∈J is linearly independent.

Theorem 5.3.9 (Gramvii -Schmidtviii Orthonormalization Procedure). Let (xn )n∈N be a

family of linearly independent vectors in the inner product space E. Then, there exists a
unique orthonormal family (ên )n∈N such that,

∀n ∈ N, Span{ê0 , ê1 , · · · , ên } = Span{x0 , x1 , · · · , xn }, and ⟨ên , xn ⟩ > 0.

Proof. We construct the orthonormmal family (ên )n∈N by recurrence. Since the family
(xn )n∈N is linearly independent, it is then composed by non null vectors. We set
x0
ê0 = .
∥x0 ∥
Assuming that the vectors ê0 , · · · , ên has already constructed. We set
n
X
ê′n+1 = xn+1 − ⟨êk , xn+1 ⟩ ek .
k=0
vi
Pythagoras of Samos (ca. 570 - ca. 495 B.C.), Ancient Greek mathematician and philosopher.
vii
Jorgen Pedersen Gram (1850 - 1916), Denish mathematician.
viii
Erhard Schmidt (1876 - 1959), Baltic German mathematician.
5.3. ORTHOGONALITY 185

The vector ê′n+1 is non null since xn+1 ∈

/ Span{x0 , · · · , xn } = Span{ê0 , · · · , ên }. Then, for
all j ∈ [[0, n]], we have
n
X n
X
êj , ê′n+1 = ⟨êj , xn+1 ⟩ − ⟨êk , xn+1 ⟩ ⟨êj , êk ⟩ = ⟨êj , xn+1 ⟩ − ⟨êk , xn+1 ⟩ δj,k = 0.
k=0 k=0

It shows that ê′n+1 is orthogonal to all the vectors in ê0 , · · · , ên . We set then

ê′n+1
ên+1 = .
∥ê′n+1 ∥

Thus, the vectors ê0 , · · · , ên+1 are orthonormal. At the same time, we have
n
* !+
X 1
⟨ên+1 , xn+1 ⟩ = ên+1 , ê′n+1 + ⟨êk , xn+1 ⟩ êk = ′ > 0.
∥ên+1 ∥
k=0

We next prove the uniqueness of the orthonormal family (ên )n∈N . Let (fn )n∈N be an
orthonormal family satisfying the same conditions. We then show by recurrence that fn = ên
for all n ∈ N. The relation Span{ê0 } = Span{x0 } = Span{f0 } shows that f0 = α0 ê0 with
α0 ∈ K. The normalization condition ∥ê0 ∥ = ∥f0 ∥ = 1 shows that |α0 | = 1. Also, the
condition ⟨f0 , x0 ⟩ > 0 implies that α0 = 1 and hence, f0 = ê0 . We assume that for
all j ∈ [[0, n]], fj = êj . From the condition Span{ê0 , · · · , ên+1 } = Span{x0 , · · · , xn+1 } =
n+1
X
Span{f0 , · · · , fn+1 }, we have fn+1 = αj êj with α0 , · · · , αn+1 ∈ K. Using the orthogonal
j=0
relations, we have for all k ∈ [[1, n]]

n+1
* +
X
0 = ⟨fk , fn+1 ⟩ = êk , αj êj = αk .
j=0

Also,

1 = ∥fn+1 ∥2 = ⟨fn+1 , fn+1 ⟩ = ⟨αn+1 ên+1 , αn+1 ên+1 ⟩ = |αn+1 |2 .

Using now the condition ⟨fn+1 , xn+1 ⟩ > 0, we obtain that αn+1 ⟨ên+1 , xn+1 ⟩ > 0, which
implies that αn+1 = 1. Thus, fn+1 = ên+1 . In conclusion, we have (fn )n∈N = (ên )n∈N .

We recall here the projections on the subspaces of E.

Definition 5.3.10. Let F and G be subspaces of E such that they are in direct sum,
F ⊕ G = E. We call the projector on F and paralleling to G the linear application
p : E → E such that p ◦ p = p and ker(p) = G, im(p) = F .

Definition 5.3.11. We call an orthogonal projector any projector p on E such that

ker(p) and im(p) are orthogonal between them.

Proposition 5.3.12. Any orthogonal projector is continuous.

186 INNER PRODUCT SPACES

Proof. Let p be an orthogonal projector on F and paralleling to G with E = F ⊕ G. Let

x ∈ E, there exists y ∈ G such that x = y + p(x) with y ⊥ p(x). Using Pythagoras
Theorem 5.3.6, we have ∥x∥E = ∥y∥E + ∥p(x)∥E . This inequality implies that ∥p(x)∥E ≤
∥x∥E . We can then deduce that p is bounded on the unit sphere in E and this implies that
the linear application p is continuous via Theorem 3.6.29.

Remark 5.3.13. Let F be a subspace of finite dimensions in E, then there exists a unique
orthogonal projector on F , we denote it by pF . The complete statement of this result is
the following one.

Proposition 5.3.14. Let F be a finite dimensional vector subspace in the inner product
space E. Then, the following assertions hold.

• F possesses an orthogonal supplementary in E.

• If e1 , · · · , en is an orthonormal basis of F , then the orthogonal projector pF from E

on F is given by,
n
X
∀x ∈ E, pF (x) = ⟨ej , x⟩ej
j=1

Proof. Let F ⊂ E be a finite dimensional subspace with dim F = n ∈ N∗ . There exists

then a linearly independent family of n vectors in F . Then, we can construct an orthonor-
mal family e1 , · · · , en by the Gram-Schmidt procedure Theorem 5.3.9. It is therefore an
orthonormal basis of F . We consider then the projector defined by,
n
X
∀x ∈ E, pF (x) = ⟨ej , x⟩ej .
j=1

Let x ∈ E, we have immediately that pF (x) ∈ F . Also, for all k ∈ [[1, n]]
n n
* +
X X
⟨ek , x − pF (x)⟩ = ⟨ek , x⟩ − ek , ⟨ej , x⟩ ej = ⟨ek , x⟩ − ⟨ej , x⟩ δj,k = 0.
j=1 j=1

It implies that x − pF (x) ∈ F ⊥ . By writing x = pF (x) + (x − pF (x)), we obtain thus

E = F ⊕ F ⊥.

Remark 5.3.15. In the case where E is a Euclidean or Hermitian space, we have immediately
that, let F be a vector subspace in E, then

• F ⊕ F ⊥ = E;

• (F ⊥ )⊥ = F .

Proposition 5.3.16. Let F be a subspace of finite dimensions in E, and let x ∈ E. Then,

the orthogonal projection pF (x) of x on F achieves the distance d(x, F ) from x to F .
5.3. ORTHOGONALITY 187

Proof. Let y ∈ F , we write x − y = x − pF (x) + pF (x) − y by remarking that x − pF (x) ∈ F ⊥

and pF (x) − y ∈ F . Those two vectors are orthogonal. From Pythagoras Theorem 5.3.6,
we have

∥x − y∥2 = ∥x − pF (x)∥2 + ∥pF (x) − y∥2 .

We obtain thus, for all y ∈ F ,

∥x − pF (x)∥ ≤ ∥x − y∥.

The equality is achieved at pF (x) ∈ F . Hence, the claim follows.

Theorem 5.3.17 (Bessel Inequality). Let (vn )n∈N∗ be an orthonormal family of vectors in
E. Then, for all x ∈ E, the sequence (⟨vn , x⟩)n∈N∗ is square-summable. Moreover, it holds
that
+∞
X
|⟨vn , x⟩|2 ≤ ∥x∥2 . (5.1)
n=1

Proof. Let n ∈ N, we consider F := Span{v1 , v2 , · · · , vn } be the finite-dimensional vector

subspace in E generated from the vectors v1 , v2 , · · · , vn . Since those vectors are orthonor-
mal, then they consists an orthonormal basis in F . Applying Proposition 5.3.14, we have
the decomposition

x = pF (x) + (x − pF (x)).
n
X
Here, pF (x) = ⟨vj , x⟩ vj and x − pF (x) ∈ F ⊥ . From the orthogonal relations and
j=1
Pythagoras Theorem 5.3.6, we have,
n
2
X
∥x∥2 = ∥pF (x)∥2 + ∥x − pF (x)∥2 = ⟨vj , x⟩ + ∥x − pF (x)∥2 .
j=1

Using a simple fact ∥x − pF (x)∥2 ≥ 0, we obtain,

n
2
X
⟨vj , x⟩ ≤ ∥x∥2 .
j=1
X
This inequality shows that the n-th partial sum of the series of positive numbers |⟨vn , x⟩|2
n≥1
X
is bounded from above by ∥x∥2 . Hence, the series |⟨vn , x⟩|2 converges and it is equivalent
n≥1
to say the sequence (⟨vn , x⟩)n∈N∗ is square-summable. Remarking that
   
+∞ n n
2 2
X X X
|⟨vn , x⟩|2 = sup  ⟨vj , x⟩  = lim  ⟨vj , x⟩  ,
n∈N∗ n→+∞
n=1 j=1 j=1

hence, inequality (5.1) holds by passing n → +∞.

188 INNER PRODUCT SPACES

Remark 5.3.18. The Bessel inequality Theorem 5.3.17 holds also for an uncountable family
of vectors. For the precise meaning of the sum over an arbitrary family of numbers or
vectors, we recommend to consult Appendix C.

Theorem 5.3.19 (Parseval’s Identity). Let (vn )n∈N∗ be an orthonormal family of vectors
in E and let x ∈ E. Then, the equality holds
+∞
X
|⟨vn , x⟩|2 = ∥x∥2 (5.2)
n=1

if, and only if x ∈ Span {vn | n ∈ N∗ }.

Proof. For all n ∈ N∗ , we define Fn := Span{v1 , v2 , · · · , vn } and F := Span {vn | n ∈ N∗ }.
It is clear that Fn ⊂ F . Let x ∈ E.
⇒: We suppose (5.2) holds. Let ε > 0, it follows from the convergence of the series that
there exists N ∈ N∗ such that
N
X
∥x∥2 − ε2 < |⟨vn , x⟩|2 ≤ ∥x∥2 .
n=1

Notice that FN is a finite dimensional subspace in E, the results in Proposition 5.3.14

sate that
N
X
2 2
d(x, FN ) = ∥x∥ − ∥pFN (x)∥ = ∥x∥ − 2 2
|⟨vn , x⟩|2 .
n=1

Then, we obtain that

v
u
u N
X
d(x, F ) ≤ d(x, FN ) ≤ t∥x∥2 − |⟨vn , x⟩|2 ≤ ε.
n=1

Hence, we deduce with d(x, F ) = 0, which is equivalent to say x ∈ F .

⇐: We assume that x ∈ F . Let ε > 0, there exists y ∈ F such that ∥x − y∥2 ≤ ε2 . Since
y ∈ F , y can be written as an linear combination of vectors in (vn )n∈N∗ . Thus, there
XN
exists N ∈ N∗ such that y = ⟨vn , y⟩ vn ∈ FN . Then, using Proposition 5.3.16, we
n=1
have,
∥x − pFN (x)∥2 = d(x, FN )2 ≤ ∥x − y∥2 ≤ ε2 .
N
X
2 2
Notice that ∥x − pFN (x)∥ = ∥x∥ − ∥pFN (x)∥ = ∥x∥ − 2 2
|⟨vn , x⟩|2 . We can thus
n=1
obtain,
N
X
∥x∥2 − ε2 ≤ |⟨vn , x⟩|2 ≤ ∥x∥2 .
n=1

Passing ε → 0, the equality (5.2) holds.

5.4. HILBERT SPACES 189

5.4 Hilbert Spaces

Definition 5.4.1. We call a Hilbert space any complete inner product space. That
means, a Hilbert space H satisfies

• H is a vector space.

• H is endowed with a (real or complex) inner product ⟨·, ·⟩H .

q
• The normed space (H, ∥ · ∥H ) is complete, where ∥x∥H = ⟨x, x⟩H for all x ∈ H.

Proposition 5.4.2. • Any finite dimensional inner product space is a Hilbert space.

• Any Hilbert space has its dimension either to be finite or to be uncountable.

Proof. The first point follows directly from Proposition 5.2.10. Consider now a Hilbert
space H with a dimension being countable infinite. Let (en )n∈N be an orthonormal basis
(the existence of such basis is guaranteed by XGram-Schmidt
X orthonormalization procedure
en
Theorem 5.3.9) of H, we consider the series vn := of elements in H. We
n n
(n + 1)2
1
have then, from the triangular inequality , for all n ∈ N, ∥vn ∥H ≤ . Since the series
(n + 1)2
X 1 X
2
converges, the series vn is absolutely convergent. From the completeness
(k + 1) n
k X
of H, the series vn converges. However, it is impossible to decompose into a linear
n
+∞
X en
combination of the limit . Thus, the contradiction. In conclusion, the E has
(n + 1)2
n=0
its dimensions either being finite or being uncountable.

The following theorem states that the result in Proposition 5.3.16 holds in Hilbert spaces
where the assumption of finite dimensions is released.

Theorem 5.4.3. Let H be a Hilbert space and F ⊂ H be a closed vector subspace. Let
x ∈ F , the following results hold.

1.) There exists xF ∈ F such that this point achieves the distance d(x, F ).

2.) xF is characterized by ∀y ∈ F , ⟨x − xF , y⟩H = 0. And xF is unique.

3.) H = F ⊕ F ⊥ and (F ⊥ )⊥ = F .

4.) The mapping pF : x 7→ xF is linear and satisfies p2F = pF , i.e., it is an orthogonal

projector.

Proof. 1.) If x ∈ F , it is sufficient to take xF = x and the claim follows.

Now, we assume that x ∈ / F . From the characterization d(x, F ) = 0 ⇔ x ∈ F = F , we
can deduce that d(x, F ) > 0. Next, by using the characterization of infimum, for all
1
n ∈ N, there exists a vector yn ∈ F such that d(x, F )2 ≤ ∥x − yn ∥2H ≤ d(x, F )2 + n .
2
190 INNER PRODUCT SPACES

Then, we construct the subspaces Vn as well as the vectors vn by

∀n ∈ N, Vn = Span{y0 , y1 , · · · , yn } ⊂ F, vn = pVn (x) ∈ Vn .
Here, pVn refers to the orthogonal projector on the finite dimensional space Vn . For
all n ∈ N, it follows from Proposition 5.3.16 that ∥x − vn ∥H = d(x, Vn ). In the
consequences we have the estimation
1
d(x, F )2 ≤ ∥x − vn ∥2H ≤ ∥x − yn ∥2H ≤ d(x, F )2 +. (5.3)
2n
Also, by using the relation Vn ⊂ Vn+1 , we obtain that vn − vn+1 ∈ Vn+1 . Then, x −
⊥ , which implies (x−v
vn+1 ∈ Vn+1 n+1 ) ⊥ (vn −vn+1 ). Using Pythagoras Theorem 5.3.6
and the previous estimation, we have
1
∥vn − vn+1 ∥2H = ∥x − vn ∥2H − ∥x − vn+1 ∥2H ≤ .
2n
So, for all n, p ∈ N, it holds that
p−1 p−1
X X 1
∥vn − vn+p ∥2H ≤ ∥vn+k − vn+k+1 ∥2H ≤
2n+k
k=0 k=0
1 1 − 21p 1
= n 1 ≤ n−1 −→ 0.
2 1− 2 2 n→+∞

This shows that the sequence (vn )n∈N is a Cauchy sequence in H. Notice that H is
complete, that (vn )n∈N ∈ F N and that F is closed, we can deduce that the sequence
(vn )n∈N converges to a limit xF ∈ F . And it is sufficient to pass n → +∞ in (5.3) to
obtain that ∥x − xF ∥H = d(x, F ).
2.) Consider y ∈ F and t ∈ C. Since xF achieves the distance, we have then
∥x − xF ∥2H ≤ ∥x − xF + ty∥2 = ∥x − xF ∥2H + |t|2 ∥y∥2H + 2Re(t ⟨x − xF , y⟩H )

We may set θ = arg(⟨x − xF , y⟩H ) ∈ [0, 2π) and t = |t|e−iθ . Then it becomes,
2 ⟨xF − x, y⟩H ≤ |t|∥y∥2H .
Letting |t| → 0, we can obtain that ⟨xF − x, y⟩H = 0. In conclusion, x − xF ⊥ y and
hence, x − xF ∈ F ⊥ .
Furthermore, the uniqueness of xF is derived by the same arguments in the proof of
Proposition 5.3.16.
3.) In the previous point, we have shown the decomposition x = xF + (x − xF ) with
xF ∈ F and x − xF ∈ F ⊥ . Thus, the claims follow.
4.) Consider x, y ∈ H and α, β ∈ C. Then, αpF (x) + βpF (y) ∈ F and for all z ∈ F , we
have,
⟨(αx + βy) − (αpF (x) + βpF (y)), z⟩H = α ⟨x − pF (x), z⟩H + b ⟨y − pF (y), z⟩H
= α0 + β0 = 0.
Thus, we obtain pF (αx + βy) = αpF (x) + βpF (y), which means pF is linear.
Moreover, for all x ∈ F , pF (x) = xF = x. It indicates that p2F = pF .
5.4. HILBERT SPACES 191

The results in Theorem 5.4.3 can be also generalized into convex subsets, which has
important application in Optimization.

Proposition 5.4.4. If F is a closed and convex subset in H, for all x ∈ H, there exists a
unique orthogonal projection pF (x) on F , which is characterized by,

• ∥x − pF (x)∥H = d(x, F ),

• for all y ∈ F , Re(⟨x − pF (x), y − pF (x)⟩H ) ≤ 0.

Proof. See in Functional Analysis.

Theorem 5.4.5 (Riesz-Fréchetix Representation). Let H be a Hilbert space. Let φ : H → C

be a continuous linear formx on H. Then, there exists a unique vφ ∈ H such that

∀x ∈ H, φ(x) = ⟨vφ , x⟩H .

Proof. We prove at first the uniqueness. If vφ,1 and vφ,2 both satisfy the condition, it
comes directly from the linearity that ∀x ∈ H, ⟨vφ,1 − vφ,2 , x⟩H = φ(x) − φ(x) = 0. Then,
vφ,1 − vφ,2 ∈ H⊥ = {0H }. Hence, vφ,1 = vφ,2 .
Introduce now the kernel F := ker(φ). It follows from the continuity and the linearity
of φ that F is a closed vector subspace in H via set characterization of continuous functions
Theorem 3.6.13.
If F = H, we have immediately that ∀x ∈ H, φ(x) = 0. Then, it is sufficient to set
vφ = 0H .
We assume now F ⊊ H. Then, we can find v0 ∈ H \ F . Then, let v1 be the orthogonal
v0 − v1
projection of v0 on F obtained by Theorem 5.4.3 and let v2 = . We have then
∥v0 − v1 ∥H

• v2 ∈ F ⊥ ;

• ∥v2 ∥H = 1.

Let x ∈ H. It follows simply from the linearity that

 
v2 φ(v2 )
φ x − φ(x) = φ(x) − φ(x) =0
φ(v2 ) φ(v2 )
v2
In other words, x − φ(x) ∈ ker(φ) = F . In the consequence,
φ(v2 )
 
v2 φ(x)
0 = v2 , x − φ(x) = ⟨v2 , x⟩H − ∥v2 ∥2H .
φ(v2 ) H φ(v2 )

Hence, φ(x) = ⟨φ(v2 )v2 , x⟩H and the claim follows by setting vφ = φ(v2 )v2 .

Corollary 5.4.6. Let H be a Hilbert space, and F ⊂ H be a subspace in H. Then, it holds,

ix
René Maurice Fréchet (1878-1973), French mathematician. He contributed in functional fnalysis, prob-
ability, and topology.
x
In Functional Analysis, a continuous linear form is describe as an element in the dual space of H. We
denote φ ∈ H′ .
192 INNER PRODUCT SPACES

• (F ⊥ )⊥ = cl(F );

• F is dense in H if, and only if F ⊥ = {0H }.

Proof. • Recall Proposition 5.3.5, (F ⊥ )⊥ is a closed subset in H. It is also immediate

from the orthogonal relations that F ⊂ (F ⊥ )⊥ . So, it follows that cl(F ) ⊂ (F ⊥ )⊥ .
On the other hand, since cl(F ) is a closed vector subspace in H, we have by applying
Theorem 5.4.3 that cl(F ) = (cl(F )⊥ )⊥ . Using the relation F ⊂ cl(F ) and Proposi-
tion 5.3.5, we have cl(F )⊥ ⊂ F ⊥ . Again, it follows (F ⊥ )⊥ ⊂ (cl(F )⊥ )⊥ . In conclusion,
we have

(F ⊥ )⊥ ⊂ (cl(F )⊥ )⊥ = cl(F ) ⊂ (F ⊥ )⊥ .

Hence, the claim follows.

• If F is dense in H, it follows from Proposition 5.3.5 that

{0H } = H⊥ = (cl(F ))⊥ = F ⊥ .

Reciprocally, if F ⊥ = {0H }, it follows from the previous point that

cl(F ) = (F ⊥ )⊥ = {0H }⊥ = H.

Hence, the claim follows.

Remark 5.4.7. The results in Corollary 5.4.6 remain true in any arbitrarily Banach space.
The precise meaning of the orthogonality in a normed space (without introducing inner
products) and their proofs are consequences of Hahn-Banach Theorem. See in the course
Functional Analysis.

5.5 Exercises
Exercise 5.5.1. Justify Propositions 5.2.2 and 5.2.3 and Corollary 5.2.4.

Exercise 5.5.2. Show that the unit closed ball in a complex inner product space is strictly
convex. That means, let B be the unit closed ball, for all x, y ∈ B, and all t ∈ (0, 1).
∥tx + (1 − t)y∥ < 1.

Exercise 5.5.3. Let x, y be non-zero vectors in a real inner product space, show that
1 1 1
2
x− 2
y = ∥x − y∥.
∥x∥ ∥y∥ ∥x∥∥y∥

Exercise 5.5.4. Let E be a real inner product space. Determine all numbers λ such that
for all x, y ∈ E the following inequality holds,

∥x∥ + ∥y∥ ≤ λ max{∥x + y∥, ∥x − y∥}.

5.5. EXERCISES 193

Exercise 5.5.5. Let E be a complex inner product space. Determine all numbers λ such
that for all x, y ∈ E the following inequality holds,

∥x∥ + ∥y∥ ≤ λ max{∥x + y∥, ∥x − y∥}.

Exercise 5.5.6. Let x1 , · · · , xn be n vectors in a real inner product space such that for all
i, j ∈ [[1, n]], i ̸= j, ⟨xi , xj ⟩ ≤ 0. In addition, there exists a vector x such that ⟨x, xi ⟩ > 0 for
all i ∈ [[1, n]]. Show that the vectors x1 , · · · , xn are linearly independent.

The objective of the following exercise is to introduce a numerical method to calculate

integrals, which is called as weighted orthogonal polynomials.

Exercise 5.5.7. Let I be an interval on R. We denote by ˚ I its interior. Let θ be a

continuous function defined on ˚
I in theZ values of strictly positive real numbers such that for
all polynomial functions P , J(P ) := P (t)θ(t)dt < +∞. We remark that such function
I
2
exists, for example when I = R, we can choose θ(t) = e−t .

1.) Show that the application (P, Q) 7→ J(P Q) defines an inner product on the space of
polynomial functions.

2.) Show that there exists (Pn )n∈N of polynomial functions such that

• for all n ∈ N, Pn is polynomial of degree n with a strictly positive dominate

coefficient, that means Pn (t) = an tn + an−1 tn−1 + · · · + a0 with an > 0;
• for all n, m ∈ N, J(Pn Pm ) = 1 if n = m and J(Pn Pm ) = 0 if not.
1
3.) Let I = (−1, 1) and θ : t 7→ √ . Determinate the sequence (Pn )n∈N in this case.
1 − t2
4.) Show that for each n ∈ N, Pn has n distinguish roots z1 , · · · , zn in ˚
I.

5.) We assume from now n ∈ N∗ to be a fixed degree. Let i ∈ [[1, n]]. Show that there
exists a unique polynomial function Li of degree less then n−1 and satisfies Li (zj ) = 1
if i = j and Li (zj ) = 0 if not. We denote then λi = J(Li ).

6.) Show that, for all polynomial function P with a degree strictly less then 2n, we have
n
X
J(P ) = λi P (zi ).
i=1

7.) Show that in fact, λi = J(L2i ) > 0 for all i ∈ [[1, n]].

8.) Show that the following recurrence relation holds,

Pn+2 = an (X − bn )Pn+1 + cn Pn .

Exercise 5.5.8. Let E be an inner product space and F ⊂ E be a vector subspace. We

assume that E = F ⊕ F ⊥ and pF be the orthogonal projector on F . Show that for all
x ∈ E, the distance dist(x, F ) is achieved at pF (x) and deduce dist(x, F ) = ∥x − pF (x)∥.
194 INNER PRODUCT SPACES
 Chapter 6

Differential Calculus

In this chapter, we denote by (E, ∥ · ∥E ) and by (F, ∥ · ∥F ) the non empty real normed
spaces of finite dimensions. We shall choose arbitrarily the bases BE = (u1 , u2 , · · · , um )
and BF = (v1 , v2 , · · · , vn ) respectively to E and to F . In a point view of practice, we may
assume that E = Rm , F = Rn and choose the canonical basis of them.
Xn
Also, for any function f which maps from E into F , we will write f (x) = fi (x)vi
i=1
where for all i ∈ [[1, n]], fi : E → R denotes the i-th component function.

6.1 Derivations of Functions

Let U be an open set in E and f be a function which maps from U into F .

6.1.1 Directional Derivatives

Let a ∈ U and h ∈ E. Since U is open and a ∈ U , there exists therefore an open interval
(−εa,h , εa,h ) with εa,h > 0 such that {a + th | −εa,h < t < εa,h } ⊂ U . Then, the function
t 7→ f (a + th) is defined for |t| small enough.

Definition 6.1.1. We say the function f is derivable at a in the direction h if the

following limit exists,

f (a + th) − f (a)
lim .
t→0 t
In the case, this limit is called as the directional derivative of f at a in the direction
h, which is denoted by Dh f (a).

Remark 6.1.2. f is derivable at a in the direction h if, and only if, all components fi ,
i ∈ [[1, n]] are derivable at a in the direction h. And we have,
n
X
Dh f (a) = Dh fi (a)vi .
i=1

We recall that BE = (u1 , u2 , · · · , um ) forms a basis of E.

195
196 DIFFERENTIAL CALCULUS

Definition 6.1.3. Let j ∈ [[1, m]], we say f possesses the j-th partial derivative at a if
f is derivable at a in the direction uj . And we denote it Dj f (a).

Remark 6.1.4. In the case E = Rm and BE refers to the canonical basis (x̂1 , · · · , x̂m ), we
denote the partial derivatives as follows. Let 1 ≤ j ≤ m,
∂f f (a1 , · · · , xj , · · · , am ) − f (a1 , · · · , aj , · · · , am )
Dj f (a) = ∂j f (a) = (a) = lim
∂xj xj →aj xj − aj

Definition 6.1.5. If f possesses all partial derivatives at a, we call the Jacobian matrix
at a in the bases BE and BF , and we denote by Jf (a), the following matrix,
 
Jf (a) = Dj fi (a) 1≤i≤n ∈ Mn×m (R).
1≤j≤m

Remark 6.1.6. In the cases E = Rm , F = Rn and BE , BF are canonical basis, the Jacobian
matrix is given as follows.
 
∂fi
Jf (a) = ∂x j
(a) 1≤i≤n .
1≤j≤m

Example 6.1.7. Consider the function f : R2 → R,

( 2
y
x if x ̸= 0,
f (x, y) =
0 if x = 0.

Then, f is derivable at (0, 0) in any direction (a, b) ∈ R2 whereas f is not continuous at

(0, 0).

6.1.2 Differentiation at a Point

Let a ∈ U . Since U is open, the function h 7→ f (a + h) is defined on a neighborhood of 0E .

Definition 6.1.8. We say f is differentiable at a if there exists a linear (and continuous)

application u such that the asymptotic behavior holds,

f (a + h) = f (a) + u(h) + o(∥h∥E ).

∥h∥E →0

We say also u is tangent to f at a.

Remark 6.1.9. • The asymptotic expansion is equivalently to say there exists a function
ε mapping from U − {a} into F such that

∀h ∈ U − {a}, f (a + h) = f (a) + u(h) + ∥h∥E ε(h),

and satisfying

lim ε(h) = 0F .
∥h∥E →0
6.1. DERIVATIONS OF FUNCTIONS 197

• In the previous point, the set U − {a} is defined as

U − {a} = {x − a | x ∈ U } .

In general manner, let A, B be subspaces in a vector space and let λ ∈ C, we define,

A + B = {x + y | x ∈ A, y ∈ B} ,
λA = {λx | x ∈ A} .

• Since E and F are finite dimensional normed spaces, it follows from Theorem 4.4.4
that any linear application u mapping from E into F is continuous. So, the linear
application u appeared previously is in fact an operator in L (E, F ).

Proposition 6.1.10. If u ∈ L (E, F ) is tangent to f at a, then for all h ∈ E, Dh f (a) =

u(h).

Proof. The differentiability of f at a provides the existence of a function ε which tends to

0F at 0E and satisfying

f (a + h) = f (a) + u(h) + ∥h∥E ε(h),

for all h ∈ U − {a}. Then the directional derivation of f at a in the direction h are given
by

f (a + th) − f (a) [f (a) + u(th) + |t|∥h∥E ε(th)] − f (a)

Dh f (a) = lim = lim
t→0 t t→0 t
= u(h).

Definition 6.1.11. If f is differentiable at a, we call the derivative of f at a and we

denote by df (a) the unique linear application u which is tangent to f at a.

Remark 6.1.12. The differentiability of f at a can also be rewritten as

f (x) = f (a) + df (a)(x − a) + o(∥x − a∥E ).

x→a

Example 6.1.13. 1.) Let E = Mn (R), the function exp (see Proposition 4.1.27 for
definition) is differentiable at 0Mn (R) (zero matrix) and its derivative is IdMn (R) .

2.) We say a function f : U ⊂ C → C is derivable at z0 ∈ U in the sense of complex

variables if the following limit exists in C,

f (z) − f (z0 )
lim .
z→z0 z − z0
This limit is called the derivative of f at z0 in the sense of complex variables, we
denote it by f ′ (z0 ).
198 DIFFERENTIAL CALCULUS

In this case, we consider a function f˜ defined on R2 by f˜(x, y) = f (x + iy). Then,

f˜ is differentiable at (x0 , y0 ) where z0 = x0 + iy0 . Moreover, the partial derivatives
satisfies the Cauchy-Riemann equation,

∂ f˜ ∂ f˜
(x0 , y0 ) = i (x0 , y0 ).
∂y ∂x
This type of functions are called as the holomorphic functions, which will be introduced
in the course of Complex Analysis.
Solution. 1.) From the definition of the function exp, for all h ∈ Mn (R), we have
+∞ n +∞ n
X h X h
exp(h) = = Id + h + = exp(0) + IdMn (R) (h) + r(h),
n! n!
n=0 n=2

+∞ n
X h
where r(h) = . We have also the estimation,
n!
n=2

+∞ ∥h∥n +∞ ∥h∥n−2
Mn (R) Mn (R)
X X
∥r(h)∥Mn (R) ≤ = ∥h∥2Mn (R) = o(∥h∥Mn (R) ).
n! n! h→0
n=2 n=2

X ∥h∥n−2
Mn (R)
We remark that the series converges. Hence, d exp(0Mn (R) ) = IdMn (R) .
n!
n≥2

2.) We suppose that f is derivable at z0 in the sense of complex variables. We have then

f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + o(|z − z0 |),

z→z0

where f ′ (z0 ) ∈ C. By making a change of variable z = x + iy, the function f˜ defined

on R2 satisfies then,

f˜(x, y) = f˜(x0 , y0 ) + f ′ (z0 )(x − x0 + i(y − y0 )) + o(∥(x − x0 , y − y0 )∥).

(x,y)→(x0 ,y0 )

The mapping (hx , hy ) 7→ f ′ (z0 )(hx + ihy ) is linear on R2 . This shows that f˜ is
∂ f˜
differentiable at (x0 , y0 ). Finally, we remark that (x0 , y0 ) = f ′ (z0 ) as well as that
∂x
∂ f˜
(x0 , y0 ) = if ′ (z0 ). Thus, the Cauchy-Riemann equation is verified.
∂y

Proposition 6.1.14. If f is differentiable at a, then f is continuous at a.

Proof. Left in Exercise.

Proposition 6.1.15. The function f is differentiable at a if, and only if all of its compo-
nents fi , i ∈ [[1, n]], are differentiable at a. We have also,
n
X
df (a)(h) = dfi (a)(h)vi .
i=1
6.1. DERIVATIONS OF FUNCTIONS 199

Proof. We prove this proposition by double implications.

⇒: We suppose that the function f is differentiable at a. Then the development holds,

f (a + h) = f (a) + df (a)(h) + ∥h∥E ε(h),

∥h∥E →0

where ε denotes a function which tends to 0F at 0E . Then, the i-th component of the
function f satisfies

fi (a + h) = fi (a) + df (a)i (h) + ∥h∥E εi (h).

∥h∥E →0

This relation shows that fi is differentiable at a and that dfi (a) = df (a)i . It follows
thus, for all h ∈ U − {a},
n
X n
X
df (a)(h) = df (a)i (h)vi = dfi (a)(h)vi .
i=1 i=1

⇐: It is straightforward by summation.

Corollary 6.1.16. Assume that f is differentiable at a. Then,

m m
X X ∂f
• for h = λj uj , Dh f (a) = λj (a);
∂xj
j=1 j=1

• the
 matrix  representing df (a) in the basis BE and BF is the Jacobian matrix Jf (a) =
Dj fi (a) 1≤i≤n
1≤j≤m

Proof. • Using the linearity of the differential df (a) and Proposition 6.1.10, we obtain,

Xm
Dh f (a) = df (a)(h) = df (a)( λj uj )
j=1
m m m
X X X ∂f
= λj df (a)(uj ) = λj Dj f (a) = λj (a).
∂xj
j=1 j=1 j=1

• Direct consequence of the previous point and Proposition 6.1.15.

Definition 6.1.17. We say f is differentiable on U if f is differentiable at each point in

U −→ L (E, F )
U . We call then the differential of f the application df : .
x 7−→ df (x)
200 DIFFERENTIAL CALCULUS

6.1.3 Continuously Differentiable Functions

Definition 6.1.18. We say f is continuously differentiable or is of C 1 -class, if f is
differentiable and for each h ∈ E, the mapping x →
7 Dh f (x) is continuous.

GL(n) −→ GL(n)
Example 6.1.19. • The inversion map Inv : is continuously
M 7−→ M −1
differentiable. And the differential of Inv is given by,

GL(n) −→ L (Mn (R))

dInv : .
M 7−→ h 7→ −M −1 hM −1

• The function exp is continuously differentiable on Mn (R).

Solution. • Let M ∈ GL(n), it is shown in Proposition 3.6.15 that GL(n) is an open

set and thus, for h ∈ Mn (R), ∥h∥Mn (R) being small enough, M + h ∈ GL(n). From
the identity (M + h)−1 (M + h) = In , we have then

(M + h)−1 M + (M + h)−1 h = In ⇒ (M + h)−1 = M −1 − (M + h)−1 hM −1 .

Then it holds,

(M + h)−1 = M −1 − [M −1 − (M + h)−1 hM −1 ]hM −1

= M −1 − M −1 hM −1 + (M + h)−1 hM −1 hM −1 .

It is cleat that the application h 7→ −M −1 hM −1 is linear on Mn (R). Also, the

3
continuityi of the function Inv shows that ∥(M + h)−1 ∥Mn (R) ≤ ∥M −1 ∥Mn (R) for
2
∥h∥Mn (R) being small enough. Hence,

3
∥(M + h)−1 hM −1 hM −1 ∥Mn (R) ≤ ∥M −1 ∥3Mn (R) ∥h∥2Mn (R) = o(∥h∥Mn (R) ).
2
So, the function Inv is differentiable at each points in GL(n) and the differential is
given as above. On the other hand, let h ∈ Mn (R) be a fixed matrix, it is also
immediate that the function M 7→ −M −1 hM −1 is continuous using the continuity of
the function Inv. In conclusion, the function Inv is continuously differentiable.

• Let M ∈ Mn (R), we have for all h ∈ Mn (R),

+∞
X (M + h)k − M k
exp(M + h) − exp(M ) = .
k!
k=0

We then need to calculate the development of (M + h)k for all k ∈ N. Using

a simple induction, we have that (M + h)k = M k + Lk (M, h) + Rk (M, h) where
i
We can show this point by using te comatrices. Let M = (mi,j )i,j∈[[1,n]] ∈ GL(n), then M −1 =
1 t 
Cof(mi,j ) 1≤i≤n . The continuity comes from the continuity of the functions det and M 7→
det M 1≤j≤m
Cof(mi,j ).
6.1. DERIVATIONS OF FUNCTIONS 201

k−1
X
Lk (M, h) = M j hM k−1−j and Rk (M, h) is the sum of 2k − (k + 1) terms of the
j=0
products in which each product the factor h appears at least two times. Then, the
mapping Lk (M, ·) : h 7→ Lk (M, h) is linear respect to h and the estimation holds,
∥Lk (M, h)∥Mn (R) ≤ k∥M ∥k−1
Mn (R) ∥h∥Mn (R) . This estimation shows that the series
X Lk (M, h)
is absolutely convergent. Thus, the mapping
k!
k

+∞
X Lk (M, h)
L(M, ·) : h 7→
k!
k=0

is a linear application on Mn (R).

On the other hand, we develop (∥M ∥Mn (R) + ∥h∥Mn (R) )k to obtain that

(∥M ∥Mn (R) + ∥h∥Mn (R) )k

k
!
k
∥M ∥k−p p
X
= ∥M ∥kMn (R) + k∥M ∥k−1
Mn (R) ∥h∥Mn (R) + Mn (R) ∥h∥Mn (R) .
p=2
2

From the definition of Rk (M, h), we have that

k
!
k
∥M ∥k−p p
X
∥Rk (M, h)∥Mn (R) ≤ Mn (R) ∥h∥Mn (R)
p=2
2
= (∥M ∥Mn (R) + ∥h∥Mn (R) )k − ∥M ∥kMn (R) − k∥M ∥k−1
Mn (R) ∥h∥Mn (R) .

X Rk (M, h)
This estimation shows that the series is absolutely convergent and sat-
k!
k
isfies the estimation
+∞
X ∥Rk (M, h)∥
≤
k!
k=0
+∞ +∞ ∥M ∥k +∞ ∥M ∥k−1
X (∥M ∥Mn (R) + ∥h∥Mn (R) )k X Mn (R)
X Mn (R)
− − ∥h∥Mn (R)
k! k! (k − 1)!
k=0 k=0 k=1
= e∥M ∥Mn (R) +∥h∥Mn (R) − e∥M ∥Mn (R) − ∥h∥Mn (R) e∥M ∥Mn (R)
= e∥M ∥Mn (R) (e∥h∥Mn (R) − 1 − ∥h∥Mn (R) ).

Using the Taylor expansion of the real-valued function t 7→ et , we obtain that while
+∞
X Rk (M, h)
∥h∥Mn (R) → 0, = o(∥h∥Mn (R) ). So, the function exp defined on Mn (R)
k!
k=0
is differentiable at M and its differential is given by the mapping L(M, ·).
We next show that exp is a C 1 function. Let h ∈ Mn (R), we consider the function
L(·, h) : M 7→ d(exp)(M )(h). It follows from the continuity of the matrix multiplica-
k−1
X
tions that for any k ∈ N, the function Lk (·, h) : M 7→ M j hM k−1−j is continuous.
j=0
 202 DIFFERENTIAL CALCULUS

+∞
X Lk (M, h)
To show that the sum function L(·, h) : M 7→ is continuous, we will show
k!
k=0
X Lk (·, h)
that the series of functions is normally convergent in all closed ballsii . Let
k!
k
R > 0, for all M ∈ Mn (R), ∥M ∥Mn (R) ≤ R, we have

∥Lk (M, h)∥Mn (R) ≤ kRk−1 ∥h∥Mn (R) .

X ∥Lk (M, h)∥Mn (R)

From the convergence of the series exponential, the series con-
k!
k
X Lk (·, h)
verges and hence the series of functions is normally convergent on the
k!
k
X Lk (·, h)
closed ball B Mn (R) (0, R). So, the sequence of partial sums to converges
k!
k
uniformally to its limit and the limit is continuous on B Mn (R) (0, R). We conclude
+∞
X Lk (M, h)
thus the function L(·, h) : M 7→ is continuous on B Mn (R) (0, R). Since
k!
k=0
it is true for all R > 0, the function L(·, h) is continuous everywhere. Finally, the
function exp is of class C 1 .

Proposition 6.1.20. The following assertions are equivalent.

1.) f is continuously differentiable.

2.) f is differentiable and for each j ∈ [[1, m]], Dj f is continuous.

3.) f is differentiable and df : U → L (E, F ) is continuous.

Proof. We prove the implications 1.) ⇒ 2.) ⇒ 3.) ⇒ 1.).

1.) ⇒ 2.): It is sufficient to take h = uj for each j ∈ [[1, m]].

2.) ⇒ 3.): Following from Proposition 6.1.15, f is differentiable if, and only if all its components
fi , i ∈ [[1, n]] are differentiable. Let j ∈ [[1, m]] and x, h ∈ E, we have the decomposition
Xn
Dj f (x)(h) = Dj fi (x)(h)vi . Then, the mapping x 7→ Dj f (x)(h) is continuous if,
i=1
and only if all components
 x 7→ Dj fi (x)(h) are continuous. Using Corollary 6.1.16,
the mapping x 7→ Dj fi (x) 1≤i≤n is continuous and hence df : U → L (E, F ) is
1≤j≤m
continuous.

3.) ⇒ 1.): We assume that df : U → L (E, F ) is continuous. Then, for all convergent sequence
(xn )n∈N , xn −→ x ∈ U , we have df (xn ) −→ df (x) in L (E, F ). Let h ∈ U − a,
n→+∞ n→+∞

ii
For the arguments of series of functions and their different types of convergence, they will be introduced
in the course of the second semester.
6.1. DERIVATIONS OF FUNCTIONS 203

we have then
∥Dh f (xn ) − Dh f (x)∥F = ∥(df (xn ) − df (x))(h)∥F
≤ ∥df (xn ) − df (x)∥L (E,F ) ∥h∥E −→ 0.
n→+∞

And the claim 1.) follows.

Theorem 6.1.21. f is continuously differentiable if, and only if all partial derivatives of
f are continuous.
Proof. We prove this theorem by double implications.
⇒: It follows directly from Proposition 6.1.20.
⇐: We suppose now f possesses the continuous partial derivatives Dj f for all j ∈ [[1, m]].
We are going to show that f is differentiable on U . Let a ∈ U and h ∈ B(0, R)
Xm
where R > 0 such that B(a, R) ⊂ U . We write h in the basis BE : h = λj uj .
j=1
Since E is a finite dimensional normed space, it follows from Theorem 4.4.10 that
any norms defined on E are equivalent, we will adapt the ∞−norm in the following
Xm
reasoning. That means, ∥h∥E = λj uj = max |λj |. Then we define the
j∈[[1,m]]
j=1
∞
k
X
vectors hk := λj uj and it is immediate that ∥hk ∥E ≤ ∥h∥E for all k ∈ [[1, m]]. We
j=1
have also,
m
X
f (a + h) − f (a) = (f (a + hk ) − f (a + hk−1 )). (6.1)
k=1

Since f possesses the partial derivatives, we have then, for all k ∈ [[1, m]],
f (a + hk ) = f (a + hk−1 ) + λk Dk f (a + hk−1 ) + rk (h),
where the function rk satisfies rk (h) = o(|λk |). From the continuity of the mapping
h→0
x 7→ Dk f (x), we have then Dk f (a + hk−1 ) −→ Dk f (a). Summing up the previous
∥h∥E →0
equality for all k ∈ [[1, m]], we obtain then,
m
X
f (a + h) = f (a) + λk Dk f (a) + r(h)
k=1
= f (a) + u(h) + r(h). (6.2)
m
X
Here, the linear application u is defined by ∀h = λk uk ∈ B(0E , R), u(h) =
k=1
m
X
λk Dk f (a). We use again (6.1) to obtain,
k=1
m
X
r(h) = (f (a + hk ) − f (a + hk−1 ) − λk Dk f (a)).
k=1
204 DIFFERENTIAL CALCULUS

For all k ∈ [[1, m]], we consider the function vk defined by

∀t ∈ [0, λk ], vk (t) = f (a + hk−1 + tuk ) − tDk f (a).

Since f possesses the continuous partial derivatives, the function vk is therefore con-
tinuously differentiable on [0, λk ]. We have then,

∀t ∈ [0, λk ], vk′ (t) = Dk f (a + hk−1 + tuk ) − Dk f (a).

We apply now the mean-value theorem on vk . There exists sk ∈ [0, λk ] such that
vk (λk ) − vk (0)
vk′ (sk ) = . It follows then,
λk − 0
1
[f (a + hk ) − f (a + hk−1 ) − λk Dk f (a)] = Dk f (a + hk−1 + sk uk ) − Dk f (a).
λk
Hence,

∥f (a + hk ) − f (a + hk−1 ) − λk Dk f (a)∥F
≤ |λk | sup ∥Dk f (a + hk−1 + suk ) − Dk f (a)∥F .
s∈[0,λk ]

We remark that the vector hk−1 + shk satisfies that for s ∈ [0, λk ],

∥hk−1 + suk ∥E = max{|λ1 |, · · · , |λk−1 |, |s|} ≤ max |λj | = ∥h∥E .

j∈[[1,m]]

Summing up these inequalities for all k ∈ [[1, m]], we obtain that

m
X
∥r(h)∥F ≤ ∥h∥E sup ∥Dk f (a + l) − Dk f (a)∥F .
∥l∥E ≤∥h∥E k=1

Let ε > 0, it follows from the continuity of the partial derivatives, that there exists
δ ∈ (0, R) such that for all l ∈ E, ∥l∥E ≤ δ implies that
m
X
∥Dk f (a + l) − Dk f (a)∥F ≤ ε.
k=1

So, for all h ∈ B(0E , R) satisfying ∥h∥E ≤ δ, ∥r(h)∥F ≤ ε∥h∥E . In other words,
r(h) = o(h). Combined with (6.2), we have shown that f is differentiable at
∥h∥E →0
a ∈ U . Since it is applicable on every points in U , we have then f is differentiable
on U . In conclusion, f is continuously differentiable on U via Proposition 6.1.20-item
2.).

Proposition 6.1.22. Any linear application u ∈ L (E, F ) is continuously differentiable.

The differential of u at each point a ∈ E equals to u.
Proof. The claim follows immediately from the relation

∀a, h ∈ E, u(a + h) = u(a) + u(h).

The continuity of the differential u comes from the continuity of linear applications defined
on finite dimensional spaces.
6.1. DERIVATIONS OF FUNCTIONS 205

Proposition 6.1.23. Let E1 , E2 be normed spaces of finite dimensions. Then, any bilinear
application B : E1 ×E2 → F is continuously differentiable. At each point (a1 , a2 ) ∈ E1 ×E2 ,
the differential of B equals to,

E1 × E2 −→ F
dB(a1 , a2 ) : .
(h1 , h2 ) 7−→ B(h1 , a2 ) + B(a1 , h2 )

Proof. It follows from the result in Exercise 4.6.15 that there exists K > 0 such that for all
(x1 , x2 ) ∈ E1 × E2 , ∥B(x1 , x2 )∥F ≤ K∥x1 ∥E1 ∥x2 ∥E2 . So, it follows from the bilinearity of
B that for all (a1 , a2 ), (h1 , h2 ) ∈ E1 × E2 ,

B(a1 + h1 , a2 + h2 ) = B(a1 , a2 ) + B(h1 , a2 ) + B(a1 , h2 ) + B(h1 , h2 ). (6.3)

Since the space E1 ×E2 is of finite dimensions, we then endow this space the norm ∥·∥E1 ×E2
defined by,

∀(x1 , x2 ) ∈ E1 × E2 , ∥(x1 , x2 )∥E1 ×E2 = max{∥x1 ∥E1 , ∥x2 ∥E2 }.

We have then,
• the mapping (h1 , h2 ) 7→ B(h1 , a2 ) + B(a1 , h2 ) is linear on E1 × E2 ;
• ∥B(h1 , h2 )∥F ≤ K∥h1 ∥E1 ∥h2 ∥E2 ≤ K∥(h1 , h2 )∥2E1 ×E2 , and hence, B(h1 , h2 ) =
(h1 ,h2 )→0
o(∥(h1 , h2 )∥E1 ×E2 ).
Thus, B is differentiable at each point (a1 , a2 ) ∈ E1 × E2 . Moreover, with a fixed (h1 , h2 ) ∈
E1 × E2 , the function (a1 , a2 ) 7→ dB(a1 , a2 )(h1 , h2 ) is continuous as a linear application
defined on a finite dimensional normed space. In conclusion, B is continuously differentiable.

Proposition 6.1.24. If E = R and U ⊂ R be an open set. f is continuously differentiable

if, and only if f is continuously derivable on U and its differential at a ∈ U equals to
U −→ F
df : .
h 7−→ hf ′ (a)

L (R, F ) −→ F
Proof. We begin with the remark that the mapping ψ : carries out
u 7−→ u(1)
an isomorphism between the spaces L (R, F ) and F . The inverse of this mapping is given
F −→ L (R, F )
by ψ −1 : . This remark shows that the function f is differentiable at a
v 7−→ h 7→ hv
if, and only if there exists v ∈ F such that while |h| → 0,

f (a + h) = f (a) + hv + o(|h|).

Equivalently speaking,
f (a + h) − f (a)
lim = v = f ′ (a) ∈ F.
h→0 h
So, f is differentiable at a if, and only if it is derivable at a. In addition, f is continuously
differentiable if, and only if f is continuously derivable, i.e. the mapping x 7→ f ′ (x) is
continuous.
206 DIFFERENTIAL CALCULUS

We assume now F = R. The linear applications mapping from E in to F are now the
linear forms on E. That is, L (E, F ) = L (E, R) = E ∗ . Where E ∗ denote the dual space of
E, of which the dimension is also m. Since BE = (u1 , · · · , um ) is a basis on E, we denote
by (dx1 , dx2 , · · · , dxm ) the corresponding basis in E ∗ . That means, for j ∈ [[1, m]], dxj is
the unique linear form on E such that dxj (uj ) = 1.
Proposition 6.1.25. If f is differentiable at a, then
m
X ∂f
df (a) = (a)dxj
∂xj
j=1

m
X
Proof. For all h ∈ U − a, we decompose h in the basis BE : h = λj uj . Using the
j=1
definition of the linear forms dx1 , · · · , dxm , we have that ∀j ∈ [[1, m]], dxj (h) = λj . From
Corollary 6.1.16, we have
 
m m m
X ∂f X ∂f X ∂f
df (a)(h) = λj (a) = (a)dxj (h) =  (a)dxj  (h).
∂xj ∂xj ∂xj
j=1 j=1 j=1

Remark 6.1.26. If f is continuously differentiable, we have

m
X ∂f
df = dxj ,
∂xj
j=1

is a continuous function mapping from U into E ∗ .

We assume in addition that E is a Euclidean space equipped with an inner product ⟨·, ·⟩
and that BE is an orthonormal basis, e.g. E = Rm , and BE is the canonical basis. In this
E −→ R
case, for all j ∈ [[1, m]], dxj : .
y 7−→ ⟨uj , y⟩
Definition 6.1.27. If f is differentiable at a, we call the gradient of f at a, which is
−−→
denoted by grad
D −−→ E f (a) or by ∇f (a), the unique vector in E such that for all h ∈ E, df (a)(h) =
grad f (a), h . In fact,

m
−−→ X ∂f
grad f (a) = (a)uj .
∂xj
j=1

Remark 6.1.28. • The above definition is based from Riesz-Fréchet Representation The-
orem 5.4.5. In fact, we are in finite dimensional spaces, the proof of this theorem can
be founded in Linear Algebra.
 
−−→ ∂f ∂f
• In the case E = R , grad f (a) =
m (a), · · · , (a) ∈ Rm .
∂x1 ∂xm
−−→
• If f is continuously differentiable on U , then grad f is a continuous function on U into
E, which is called a vector field.
6.2. OPERATIONS ON CONTINUOUSLY DIFFERENTIAL FUNCTIONS 207

6.2 Operations on Continuously Differential Functions

We conserve the notations in the precious section.

6.2.1 Chain Rule

Let (G, ∥ · ∥G ) be a real normed space of finite dimensions and g be a function which
maps from V ⊂ F into G such that V is open and f (U ) ⊂ V . We can thus consider the
composition g ◦ f .

Theorem 6.2.1 (Chain Rule). We assume that f and g are continuously differentiable,
then g ◦ f is continuously differentiable and moreover,

∀x ∈ U, d(g ◦ f )(x) = dg(f (x)) ◦ df (x).

Proof. Let a ∈ U and b = f (a) ∈ V . The conditions of differentiability of f and g respec-

tively at a and b give that for all h ∈ U − {a} and all l ∈ V − {b},

f (a + h) = f (a) + df (a)(h) + ∥h∥E ε1 (h),

g(b + l) = g(b) + dg(b)(l) + ∥l∥F ε2 (l),

where ε1 (h) −→ 0F and ε2 (l) −→ 0G . We denote also by l the function ∀h ∈ U − {a},

∥h∥E →0 ∥l∥E →0
l(h) := f (a + h) − f (a) ∈ V − {b}. Then, it holds,

(g ◦ f )(a + h) = g(f (a + h)) = g(b + l(h)) = g(b) + dg(b)(l(h)) + ∥l(h)∥F ε2 (l(h))

= (g ◦ f )(a) + [dg(f (a)) ◦ df (a)](h) + r(h),

where

r(h) = ∥h∥E dg(b)(ε1 (h)) + ∥l(h)∥F ε2 (l(h)).

We have also the estimations,

∥l(h)∥F ≤ ∥df (a)∥L (E,F ) ∥h∥E + ∥h∥E ∥ε1 (h)∥F ,

∥r(h)∥G ≤ ∥h∥E ∥dg(b)∥L (F,G) ∥ε1 (h)∥F + ∥l(h)∥F ∥ε2 (l(h))∥G


≤ ∥h∥E ∥dg(b)∥L (F,G) ∥ε1 (h)∥F + (∥df (a)∥L (E,F ) + ∥ε1 (h)∥F )∥ε2 (l(h))∥G .

Then, when h → 0E , we have r(h) = o(∥h∥E ). Hence, g ◦ f is differentiable at a with the

differential dg(f (a)) ◦ df (a) ∈ L (E, G).
If, in addition, the functions df and dg are continuous respectively on U and on V . It
follows from the continuity of composite functions that the function x 7→ dg(f (x)) ◦ df (x)
is continuous on U . Thus, g ◦ f is continuously differentiable.

Let BG = (w1 , · · · , wp ) be a basis in G, we can thus define the partial derivatives of g

as well as g ◦ f .

Proposition 6.2.2 (Chain Rule). • The Jacobian matrix of g ◦ f is given as follows,

J(g ◦ f )(x) = Jg(f (x)) · Jf (x).

208 DIFFERENTIAL CALCULUS

• The partial derivatives of g ◦ f are given as follows, for k ∈ [[1, p]], j ∈ [[1, m]],
n
X
Dj (g ◦ f )k (x) = Di gk (f (x))Dj fi (x).
i=1

Proof. They are direct consequences of Theorem 6.2.1.

Remark 6.2.3. When E = Rm , F = Rn , G = Rp and their basis are canonical, the above
chain rule can be written in the following useful formula,
n
∂(g ◦ f )k X ∂gk ∂fi
(x) = (f (x)) (x).
∂xj ∂yi ∂xj
i=1

Proposition 6.2.4. If E = R and f is continuously derivable on an open interval I and g

is continuously differentiable, then g ◦ f is continuously derivable and

∀x ∈ I, (g ◦ f )′ (x) = dg(f (x))(f ′ (x)).

Proof. From Proposition 6.1.24, the function f is continuously differentiable, then using
Theorem 6.2.1, we obtain that the function g◦f is continuously differentiable on I. Applying
again Proposition 6.1.24, the function g◦f is continuously derivable and we have the formula,

∀x ∈ I, (g ◦ f )′ (x) = d(g ◦ f )(x)(1) = [dg(f (x)) ◦ df (x)](1)

= dg(f (x))(df (x)(1)) = dg(f (x))(f ′ (x)).

We assume now G = R, it follows from the chain rule, we have, for x ∈ U ,

m n
!
X X ∂g ∂fi
d(g ◦ f )(x) = (f (x)) (x) dxj
∂yi ∂xj
j=1 i=1
 
n m
X ∂g X ∂fi
= (f (x))  (x)dxj  .
∂yi ∂xj
i=1 j=1

m
X ∂fi
We remark that dfi (x) = (x)dxj for all i ∈ [[1, n]]. And it becomes
∂xj
j=1

n
X ∂g
d(g ◦ f )(x) = (f (x))dfi (x).
∂yi
i=1

If we substitute yi = fi , then dyi = dfi also,

n
X ∂g
dg = dyi .
∂yi
i=1

This is called the invariant property of the differentials.

6.2. OPERATIONS ON CONTINUOUSLY DIFFERENTIAL FUNCTIONS 209

6.2.2 Algebraic Properties

Let f , g be continuous differentiable functions which map from U ⊂ E into F .

Proposition 6.2.5. Let α, β ∈ R, then αf + βg is continuously differentiable and

d(αf + βg) = αdf + βdg.

Proof. Left in Exercises.

This proposition induces directly the following corollary.

Corollary 6.2.6. The set C 1 (U, F ) of continuously differentiable functions mapping from
U into F forms a vector space.

Proof. Left in Exercises.

Proposition 6.2.7. Let v ∈ L (F, G), then v ◦ f is continuously differentiable and

d(v ◦ f ) = v ◦ df.

Proof. Left in Exercises.

Let F1 , F2 be normed spaces of finite dimensions. Consider a bilinear application B :

F1 × F2 → G and the functions f1 : U → F1 as well as f2 : U → F2 . We denote by B(f1 , f2 )
the function x 7→ B(f1 (x), f2 (x)) which maps from U into G.

Proposition 6.2.8. If f1 , f2 are continuously differentiable, then B(f1 , f2 ) is continuously

differentiable and we have for x ∈ U ,

∀h ∈ E, d(B(f1 , f2 ))(x)(h) = B(df1 (x)(h), f2 (x)) + B(f1 (x), df2 (x)(h)).

Proof. The function (f1 , f2 ) : x 7→ (f1 (x), f2 (x)) mapping from U into F1 × F2 is clearly
continuously differentiable with the differential d(f1 , f2 )(x) : h 7→ (df1 (x)(h), df2 (x)(h)) ∈
L (E, F1 × F2 ) for all x ∈ U . Then, the bilinear function B(f1 , f2 ) is the composition of
B and (f1 , f2 ). It follows from then Proposition 6.1.23 and Theorem 6.2.1 that B(f1 , f2 ) is
continuously differentiable and its differential is given by,

∀x ∈ U, h ∈ E, dB(f1 , f2 )(x)(h) = dB(f1 (x), f2 (x))(df1 (x)(h), df2 (x)(h))

= B(df1 (x)(h), f2 (x)) + B(f1 (x), df2 (x)(h)).

Example 6.2.9. Let (E, ⟨·, ·⟩) be a Euclidean space.

• The scalar product (x, y) 7→ ⟨x, y⟩ is continuously differentiable with the differential
at (x, y) ∈ E given by: (h, l) 7→ ⟨h, y⟩ + ⟨x, l⟩.

• The quadrature form Q : x 7→ ⟨x, x⟩ = ∥x∥2 is continuously differentiable and

−−→
dQ(x)h = 2⟨x, h⟩ for all x, h ∈ E. Thus, grad Q(x) = 2x.
210 DIFFERENTIAL CALCULUS

• The norm N : x 7→ ∥x∥ =

p
Q(x) is continuously differentiable on E \ {0E }. The
⟨x, h⟩ −−→ x
differential at x ∈ E \ {0E } is given by, dN (x) : h 7→ . Thus, grad N (x) = .
∥x∥ ∥x∥

Corollary 6.2.10. If F = K, K = R or C, then

• the product of the functions f and g is continuously differentiable and we have, for
x ∈ E,

d(f g)(x) = g(x)df (x) + f (x)dg(x);

• the set C 1 (U, K) is a K-algebra in which the invertible elements are non-vanishing
functions.

Proof. We remark that the product of two functions (f, g) 7→ f g is bilinear, and the claim
follows from Proposition 6.2.8.

6.2.3 Curve Integrals

Let f be a continuously differentiable function which maps from U into F . We assume
that U is arc-connected and consider a, b ∈ U . Let φ be a C 1 function on [0, 1] such that
φ(0) = a, φ(1) = b and γ := φ([0, 1]) ⊂ U .

Proposition 6.2.11. Under the above assumption, we have

Z 1
f (b) − f (a) = df (φ(t))(φ′ (t))dt.
0

Proof. Since f is continuously differentiable and φ is continuously derivable, it follows from

Proposition 6.2.4 that f ◦ φ is continuously derivable with the derivative

∀t ∈ (0, 1), (f ◦ φ)′ (t) = df (φ(t))(φ′ (t)).

Then the claim follows from the fundamental theorem of calculus.

Corollary 6.2.12. In the case E is a Euclidean space and F = R, we have

1D
−−→ −−→
Z E Z
f (b) − f (a) = grad f (φ(t)), φ′ (t) dt = grad f (s) · ds.
0 γ

Proof. Direct consequence of Proposition 6.2.11.

6.3. HIGHER ORDER DERIVATIVES 211

6.3 Higher Order Derivatives

We conserve all notations in the previous sections. We assume in what follows that E = Rm ,
F = Rn and BE , BF refers to the canonical basis. Let k ∈ N ∪ {∞} and f be a function
mapping from U ⊂ Rm into Rn where U is open.
Remark 6.3.1. We remark that the results in this section still hold in general finite dimen-
sional normed spaces E and F . In fact, E and F are isomorphic respectively to Rm and
Rn under the representations in the basis BE and BF . And all norms in E or in F are
equivalent. we can thus restrict our studies under those assumptions.

Definition 6.3.2. • We say f is of C 0 -class if f is continuous.

• We say f is k times continuously differentiable or of C k -class for k ∈ N∗ if f is

differentiable and Dh f is of C k−1 -class for any h ∈ Rm .

• We say f is indefinitely differentiable or of C ∞ -class even or smooth if f is of

C k -class for all k ∈ N.

Proposition 6.3.3. • The set C k (U, Rn ) of C k -class functions mapping from U into Rn
is a vector space.

• For all h ∈ Rm , the mapping Dh : f 7→ Dh f is a linear application from C k (U, Rn )

into C k−1 (U, Rn ).

Proof. • We will justify all linear combinations of C k -class functions are also of C k -class
by induction on k ∈ N. If k = 0, then the claim follows from the continuity of the
linear combination of continuous functions. We assume now the claim holds for k ∈ N.
Let f, g design C k+1 -class functions and α, β ∈ R. We have then in particular that
f, g are continuously differentiable functions. It follows then from Proposition 6.2.5
that αf + βg is continuously differentiable with d(αf + βg) = αdf + βdg. Using the
induction assumption, αdf + βdg is of C k -class since df and dg are. Hence d(αf + βg)
is of C k -class and thus αf + βg is of C k+1 -class. By induction, the claim holds for all
k ∈ N.

• Left in Exercises.

Proposition 6.3.4. The function f is of C k -class if ,and only if all of its components are
of C k -class.

Proof. The claim can be obtained by iterative applications of Proposition 6.1.15 and Propo-
sition 6.3.6.

Proposition 6.3.5. • If F = R or C, then the product of two C k -class functions is a

C k -class function.

• The set C k (U, K) is a K-algebra in which the invertible elements are non vanishing
functions. (K = R or C)
212 DIFFERENTIAL CALCULUS

Proof. The claims can be obtained by iterative applications of Corollary 6.2.10.

Proposition 6.3.6. For all k ∈ N∗ , the following assertions are equivalent.

1.) f is of C k -class.

∂f
2.) f is differentiable and for all j ∈ [[1, m]], : U → Rn is of C k−1 -class.
∂xj

3.) f is differentiable and df : U → L (Rm , Rn ) is of C k−1 -class.

Proof. The claims can be obtained by iterative applications of Proposition 6.1.20.

Proposition 6.3.7. Let f be a C k -class function and g be a C k -class function satisfying the
assumptions in Theorem 6.2.1. Then, the composition g ◦ f is also a C k -class function.

Proof. The claim can be obtained by iterative applications of Theorem 6.2.1.

Definition 6.3.8. Let f be a C k -class function and p ∈ [[1, k]], we define the partial differ-
entials of order p by iterations. For j1 , · · · , jp ∈ [[1, m]],

∂pf ∂ p−1 f
   
∂ ∂ ∂
= ··· f= .
∂xj1 · · · ∂xjp ∂xj1 ∂xjp ∂xj1 ∂xj2 · · · ∂xjp

∂p
As a direct consequence from Proposition 6.3.3, is a linear application from
∂xj1 · · · ∂xjp
C k (U, Rn ) into C k−p (U, Rn ).

Definition 6.3.9. • Let q ∈ N, we call a linear differential operator of order ≤ q any

linear application of the form,
X X ∂p
D= α(j1 ,··· ,jp ) (x) ,
∂xj1 · · · ∂xjp
p≤q (j1 ,··· ,jp )∈[[1,m]]p

where a(j1 ,··· ,jp ) ∈ C 0 (U, Rn ). This application maps from C q (U, Rn ) into C 0 (U, Rn ).

• We call a linear differential operator of order q any linear differential operator ≤ q

but not ≤ q − 1.

• We say a linear differential operator with constant coefficients if the functions α(j1 ,··· ,jp )
are constants.

Theorem 6.3.10 (Schwarz). If f is a C 2 -class function, then for all i, j ∈ [[1, m]],

∂2f ∂2f
= .
∂xi ∂xj ∂xj ∂xi
6.3. HIGHER ORDER DERIVATIVES 213

Proof. Let a ∈ U and r > 0 such that for all t, s ∈ [−r, r], a + tx̂i + sx̂j ∈ U . We begin with
∂f
applying Proposition 6.2.11 on the C 1 -class function .
∂xi
Let 0 < λ ≤ r and for all t ∈ [−r, r],
Z λ
∂f ∂f ∂2f
(a + tx̂i + λx̂j ) − (a + tx̂i ) = (a + tx̂i + sx̂j )ds.
∂xi ∂xi 0 ∂xj ∂xi

We apply again the same argument on f , and we obtain,

Z λ
∂f
f (a + λx̂i + λx̂j ) − f (a + λx̂j ) = (a + tx̂i + λx̂j )dt,
0 ∂xi

as well as
Z λ
∂f
f (a + λx̂i ) − f (a) = (a + tx̂i )dt.
0 ∂xi
Then,
∆(λ) := f (a + λx̂i + λx̂j ) − f (a + λx̂j ) − f (a + λx̂i ) + f (a)
Z λ Z λ
∂2f

= (a + tx̂i + sx̂j )ds dt.
0 0 ∂xj ∂xi

We remark that the expression of ∆(λ) is symmetric respect to i and j, which implies that
Z λ Z λ
∂2f

∆(λ) = (a + tx̂i + sx̂j )dt ds.
0 0 ∂xi ∂xj

We consider now the function φ : [0, r]2 → R defined by

 2
∂2f

∂ f
φ(t, s) = − (a + tx̂i + sx̂j ).
∂xi ∂xj ∂xj ∂xi
From the above calculations, we have immediately, for all λ ∈ (0, r],
Z λ Z λ 
φ(t, s)dt ds = 0.
0 0

∂2f ∂2f
On the other hand, from the continuity of the functions and on U , the
∂xi ∂xj ∂xj ∂xi
function φ is continuous on [0, r]2 .
Let ε > 0, using the continuity of φ at (0, 0), there exists δ > 0 such that |φ(t, s) −
φ(0, 0)| ≤ ε for all 0 ≤ t, s ≤ δ. We have then the estimation,
Z δ Z δ  Z δ Z δ 
δ 2 |φ(0, 0)| = φ(0, 0)dt ds = φ(t, s) − φ(0, 0)dt ds
0 0 0 0
Z δ Z δ 
≤ |φ(t, s) − φ(0, 0)|dt ds ≤ εδ 2 .
0 0

This shows |φ(0, 0)| ≤ ε for all ε > 0. Thus,

∂2f ∂2f
(a) − (a) = φ(0, 0) = 0.
∂xi ∂xj ∂xj ∂xi
214 DIFFERENTIAL CALCULUS

Remark 6.3.11. • For notation, if i = j, we denote the second order partial derivative
∂2f
by and adapt this notation in what follows for higher order derivatives.
∂x2i
• We can generalize this theorem on all directional derivatives. If f is a C 2 -class function,
then for all u, v ∈ Rm ,

Du (Dv f ) = Dv (Du f ).

The Schwarz Theorem shows that the composition of differential operators is commuta-
tive in the algebra L (C ∞ (U, Rm )). By this commutativity, the partial differential operator
of order p in Definition 6.3.8 can be written as
∂p ∂p
= .
∂xj1 · · · ∂xjp ∂x1 · · · ∂xpmm
p1

Here, the m-tuples (p1 , · · · , pm ) ∈ Nm is called a multi-index. In addition, (p1 , · · · , pm )

m
X
satisfies pj = p and for each j ∈ [[1, m]], pj refers to the occurrence number of the index
j=1
j in the set {j1 , · · · , jp }.
We present here the notation for multi-indexes.
Definition 6.3.12. Let β = (β1 , · · · , βm ) ∈ Nm be a multi-index of order m.
m
X
• We call the length of β, and denote it by |β|, the natural number |β| := βj .
j=1

m
Y
• We call the factorial of β, and denote if by β!, the number β! := βj !.
j=1

• Let h = (h1 , · · · , hm ) ∈ Rm , we call the exponent of h with power β, and denote

m
β
Y
it by hβ , the real number hβ := hj j .
j=1

• Likewise, we call the partial derivative of the multi-index β, and we denote it

∂β β ∂β ∂ |β|
by ∂ β or , the differential operator ∂ = := .
∂xβ ∂xβ ∂xβ1 1 · · · ∂xβmm
Therefore, all differential operators can be rewritten using the multi-indexes. For in-
stance, the differential operator of order ≤ q introduced in Definition 6.3.9 can be written
in a unique way by,
X ∂β
D= αβ (x) .
∂xβ
|β|≤q

m
X
Let h = hj x̂j , the derivation operator in the direction h is
j=1

m
X ∂
Dh = hj .
∂xj
j=1
6.3. HIGHER ORDER DERIVATIVES 215

We can calculate the power p of Dh in the algebra L (C ∞ (U, Rm )) just like the polynomial
functions,
 
m pj
hj  ∂p X hβ ∂ β
Dhp = p!
X Y
= p! .
pj ! ∂xp11 · · · ∂xpmm

m
β! ∂xβ
(p1 ,··· ,pm )∈N j=1 |β|=p
p1 +···+pm =p

∂ ∂
Example 6.3.13. • Riemann operator. R = +i .
∂x ∂y
m
X ∂2
• Laplacianiii ∆ = .
j=1
∂x2j

m
∂2 1 X ∂2
• Wave operator or d’Alambertian □ = − .
∂t2 c2
j=1
∂x2j

m
X ∂2
∂
• Heat or Diffusion operator H = −k .
∂t ∂x2j
j=1

Theorem 6.3.14 (Tayloriv ). Let f ∈ C k+1 (U, Rn ) with k ∈ N∗ . If [a, a + h] ⊂ U , we have

k Z 1
X 1 p (1 − s)k
f (a + h) = Dh f (a) + Dhk+1 f (a + sh) ds
p! 0 k!
p=0
X hβ ∂ β f X hβ Z 1 ∂ β f
= (a) + (k + 1) (a + sh)(1 − s)k ds.
β! ∂xβ β! 0 ∂xβ
|β|≤k |β|=k+1

[0, 1] −→ Rn
Proof. It is sufficient to consider the function φ : . Since f is of
s 7−→ f (a + sh)
C k+1 -class, the function φ is then continuously k + 1-times derivable. And the formula is
obtained from the Taylor expansion of φ.

Here, we consider only f ∈ C 2 (U, R).

Definition 6.3.15. • We say a point a ∈ U is a critical point of f if df (a) = 0L (Rm ,R) .

• We call the Hessianv matrix of f at x ∈ U the following matrix

 2 
Hess(f )(x) = ∂x∂i ∂x
f
j
(x)
i,j∈[[1,m]]

Theorem 6.3.16. Let a ∈ U be a critical point of f . If in addition, the Hessian matrix of

f at a is definite positive, then a is a local minimum of f .
iii
Pierre Simon de Laplace (1749 - 1827), French mathematician and astrologist. One of the most influential
scientist in his era. He has several important contribution in analysis, in probability and in astrology. He
was the representative person of the “determinist” in the philosophy of science.
iv
Brook Taylor (1685 - 1731), English mathematician. Known by Taylor expansions.
v
Otto Hesse (1811 - 1874), German mathematician. He worked in algebra and geometry.
216 DIFFERENTIAL CALCULUS

Proof. Let r > 0 such that B(a, r) ⊂ U , we consider h ∈ B(0, r). It follows from Taylor’s
formula Theorem 6.3.14 that
1
f (a + h) = f (a) + ⟨h, Hess(f )(a)h⟩ + ∥h∥2Rm ε(h).
2
Here ⟨·, ·⟩ designs the scalar product in Rm and ε(h) −→ 0. From the definition of the
h→0
Hessian matrix and Schwarz Theorem 6.3.10, the Hessian matrix Hess(f )(a) is symmetric.
Combined with the hypothesis that Hess(f )(a) is definite positive, we deduce that the
m
p form (h, l) 7→ ⟨h, Hess(f )(a)l⟩ defines a scalar product in R . Hence, the mapping
bilinear
m m
h 7→ ⟨h, Hess(f )(a)h⟩ defines a norm on R . Since R is a finite dimensional normed
space, all norms are equivalent. There exists therefore c, C > 0 such that for all h ∈ Rm ,
we have

c∥h∥2Rm ≤ ⟨h, Hess(f )(a)h⟩ ≤ C∥h∥2Rm .

c
From the convergence ε(h) −→ 0, there exists δ > 0 such that ∥h∥Rm ≤ δ ⇒ |ε(h)| ≤ .
h→0 2
Then, for ∥h∥ ≤ min{r, δ}, we have
1 c
f (a + h) = f (a) + ⟨h, Hess(f )(a)h⟩ + ∥h∥2Rm ε(h) ≥ f (a) + ∥h∥2Rm .
2 2
Thus, a is a local minimum of f .

Remark 6.3.17. In the case where m = 2, let a be a critical point of f and M = Hess(f )(a)
be the Hessian matrix at this point.

• If M is definite positive, then a is a local minimum.

• If M is definite negative, then a is a local maximum.

• If M has the eigenvalues λ1 , λ2 ∈ R such that λ1 λ2 < 0, then a is a saddle point.

• If M has 0 as an eigenvalue, then we can not conclude the behavior of f near a.

6.4 Exercises
Exercise 6.4.1. Justify Proposition 6.1.14.

Exercise 6.4.2. We endow the space Mn (K) with its matrix norm ∥ · ∥M , K = R or C.
GL(n) −→ Mn (K)
We recall that the function inverse Inv : is continuous.
M 7−→ M −1

1.) Let M ∈ GL(n). Show that for h ∈ Mn (K), ∥h∥M small enough, M + h is invertible.

2.) Show that, (M + h)−1 − M −1 = L(h) + o(∥h∥M ), where L : Mn (K) → Mn (K) be

h→0
a linear application. Give the explicit formula of L.

3.) Show that the function Inv is continuously differentiable.

6.4. EXERCISES 217

Exercise 6.4.3. Let A : t 7→ A(t) be a continuously differentiable function from an interval

on R into GL(n). Show that the function A−1 : t 7→ A(t)−1 is continuously differentiable
d(A−1 )
and give its differential .
dt

Exercise 6.4.4. Does there exists a norm on Rn which is differentiable?

x sin y − y sin x
Exercise 6.4.5. Let f be the function defined on R2 by f (x, y) = if (x, y) ̸=
x2 + y 2
(0, 0) and f (0, 0) = 0. Show that f is a C 1 -class function.

Exercise 6.4.6. Let f be a continuous function on R2 . We assume that the partial deriva-
∂f ∂f
tives and are continuous on R2 \ {(0, 0)}. We assume moreover those partial deriva-
∂x ∂y
tives possess the limits at (0, 0). Show that f is in fact a C 1 -class function.

Exercise 6.4.7. Let f be a continuous function on U = R2 \ {(0, 0)} satisfying that for all
x ∈ U and all t > 0, f (tx) = tα f (x) with α > 0.

1.) Show that f can extend by 0 at (0, 0) into a continuous function on R2 .

We denote from now on by f the extended function on R2 .

2.) We assume that α ∈ (0, 1). Show that f is differentiable at (0, 0) if, and only if f = 0
on R2 .

3.) We assume that α = 1. Show that f is differentiable at (0, 0) if, and only if f is an
affine function.
x3 + y 3
4.) Let g be the function defined on R2 by g(x, y) = and g(0, 0) = 0. Is g a
x2 + y 2
C 1 -class function?

5.) Show that if α > 1, f is differentiable at (0, 0).

6.) We assume that α > 1 and f is a C 1 -class function on U . Show that f is of C 1 -class
function on R2 .
p √
7.) Let α > 2 and h be the function defined on R2 by h(x, y) = ( |x| + y)α . Show that
h is a C 1 -class function.

 
1
Exercise 6.4.8. 1.) Let f be the function defined on Rn by f (x) = exp − 2 if x ̸= 0
|x|
and f (0) = 0. Show that f is a C ∞ -class function.

2.) Same question for the function φ given by,

  
 exp −1 if |x| < 1
φ(x) = 1 − |x|2
0 if |x| ≥ 1.

218 DIFFERENTIAL CALCULUS

Differentiation of Determinant
Let n ∈ N∗ , we recall here some elements in Linear Algebra.

• ⟨·, ·⟩ and ∥ · ∥ refer the usual scalar product and Euclidean norm on Rn .

• Mn (R) designs the space of square matrices of order n in real coefficients.

• In designs the unit matrix. That is, the matrix in which the diagonal coefficients all
equal to 1 and other coefficients all equal to 0.
 
• A matrix T = ti,j is said to be an upper triangular matrix if for all 1 ≤ i, j ≤
1≤i,j≤n
n such that i > j, then ti,j = 0.

• GL(n) designs the set of invertible matrices of order n.

• Let A ∈ Mn (R), we denote by Cof(A) the cofactor matrix of A. If A ∈ GL(n), we

1 t
have A−1 = Cof(A).
det(A)
• Let A ∈ Mn (R), the trace of A, tr(A) designs the sum of all diagonal coefficients of
  Xn
A. That is, if A = ai,j , then tr(A) = ai,i .
1≤i,j≤n
i=1

• Let
 A
t
 ∈ Mn (R), A designs
 the
 transposition of the matrix A. That is, if A =
t
ai,j , then A = aj,i . Furthermore, for all x ∈ Rn , ⟨x, Ax⟩ =
1≤i,j≤n 1≤i,j≤n
⟨t Ax, x⟩.

• An eigenvalue of the matrix A is a root in C of the characteristic polynomial x 7→

det(xIn − A). We denote it by λ and σ(A) refers to the set of eigenvalues (counted
with the multiplicity).

• The matrix A is said to be similar to B if there exists P ∈ GL(n) such that A =

P −1 BP .

• Any matrix A ∈ Mn (R) is C-triangularizable. That is, A is similar to an upper

triangular matrix (this triangular matrix is in Mn (C)).

• If a matrix A is symmetric, i.e. t A = A, then A is R-diagonalizable, which means A

is similar to a diagonal matrix.

Exercise 6.4.9. Show the following results in Linear Algebra. Let A ∈ Mn (R).

1.) For all B ∈ Mn (R), tr(AB) = tr(BA) and t (AB) = t B t A.

2.) If the matrix A is similar to B, then tr(A) = tr(B).

X
3.) tr(A) = λ.
λ∈σ(A)
Y
4.) det(A) = λ.
λ∈σ(A)
6.4. EXERCISES 219

Exercise 6.4.10. Let A ∈ Mn (R).

1.) Give the definition of the matrix norm ∥ · ∥M on Mn (R) associated to ∥ · ∥.

r
2.) Show that ∥A∥M = max |λ|.
λ∈σ(t AA)

3.) Deduce that for if λ ∈ σ(A), then |λ| ≤ ∥A∥M .

Exercise 6.4.11. Show that when ∥H∥M → 0,

det(In + H) = 1 + tr(H) + O(∥H∥2M ).

Exercise 6.4.12. Show that det is a C ∞ function on Mn (R) and and calculate its differ-
ential at each point A ∈ GL(n).

Exercise 6.4.13. We define the function ⟨·, ·⟩m by

∀A, B ∈ Mn (R), ⟨A, B⟩m = tr(t AB).

Show that ⟨·, ·⟩m defines an inner product on Mn (R).

Exercise 6.4.14. Show that for all A ∈ GL(n), ∇ det(A) = Cof(A).

Exercise 6.4.15. Show that GL(n) is an open and dense set in Mn (R).

Exercise 6.4.16. Show that that for all M ∈ Mn (R), ∇ det(M ) = Cof(M ).

Exercise 6.4.17. Show that the function f : x 7→ ∥x∥ is differentiable on E \ {0} and
calculate its differential at each point x ∈ E \ {0}.

Mn (R) −→ M
Exercise 6.4.18. Let p ∈ N∗ , we consider the function f : . Show
M 7−→ M p
that f is differentiable and give its differential at each point M ∈ Mn (R).

Exercise 6.4.19. Consider the function f mapping from R2 into R defined by,
xy 2
(
x 2 +y 2 if (x, y) ̸= (0, 0)
f (x, y) =
0 if (x, y) = (0, 0).

1.) Show that f is of C 1 -class in R2 \ {(0, 0)} and calculate the gradient at each point.

2.) Is the function f continuous at (0, 0)? Justify your answer.

220 DIFFERENTIAL CALCULUS

3.) Is the function f differentiable on R2 ? Justify your answer.

Exercise 6.4.20. Consider the function f mapping from R2 into R defined by,

 √ xy if (x, y) ̸= (0, 0)
f (x, y) = x2 +y 2
 0 if (x, y) = (0, 0).

1.) Show that f is of C 1 -class in R2 \ {(0, 0)} and calculate the gradient at each point.

2.) Is the function f continuous at (0, 0)? Justify your answer.

3.) Is the function f differentiable on R2 ? Justify your answer.

Exercise 6.4.21. Let Ω be an open set in Rn (n ∈ N∗ ) and f be a function mapping from

Ω into R. We suppose that f is differentiable on Ω and f (x0 ) ̸= 0 for certain x0 ∈ Ω.
1
Show that g := is defined on a neighborhood of x0 ; g is differentiable at x0 and calculate
f
∇g(x0 ).

6.A Inverse and Implicit Function Theorems (Out of Pro-

gram)
In this section, we will present two important theorems in Differential Geometry. The
content is therefore less related to applied mathematics. For those who are still interested
in such topics, we recommend to consult specialized books e.g. [1].
Let k ∈ N∗ ∪ {∞}. Throughout this section, V and W denote respectively the open sets
in the Banach spaces (not necessary to be finite-dimensional) E and F .

6.A.1 Differential Calculus in Banach Spaces

Here, we suppose that the space of variables as well as that of values are Banach spaces,
which would be infinitely dimensional spaces. So, it is indispensable to use intrinsic for-
mulations, i.e., independent to the chosen coordinates, to express the differentiability.

Definition 6.A.1. Let f : V → W and x0 ∈ V .

• We say f is (Fréchet)-differentiablevi at x0 if there exists a continuous linear

application u ∈ L (E, F ) such that the following asymptotic behavior holds while
∥h∥E → 0.

f (x0 + h) = f (x0 ) + u(h) + o(∥h∥E ).

• We call the operator u as the differential of f at x0 and denote by df (x0 ).

• We say f is differentiable on V if it is differentiable at each point in V .

vi
Or simply differentiable.
6.A. INVERSE AND IMPLICIT FUNCTION THEOREMS (OUT OF PROGRAM) 221

As an important concept, we need to introduce the notion of isomorphisms.

Definition 6.A.2. Let E and F be vector spaces. We call an isomorphism any linear
and bijective mappings from E into F . We also say that u is invertible.
As the immediate consequences, for a linear and bijective mapping u, its inverse map
u−1 is also linear and satisfies,
IdE = u−1 ◦ u, IdF = u ◦ u−1 .

We remark that a linear application is not necessary to be continuous (the continuity

is automatic only if we are in finite dimensions via Theorem 4.4.4). Also, the linear and
continuous applications are called as operators, and denoted as an element in L (E, F ).
The following theorem gives a characterization of continuous isomorphisms.
Theorem 6.A.3. Let E and F be Banach spaces. We assume a linear application u : E →
F is invertible and continuous. Then, its inverse u−1 : F → E is also continuous.
Proof. Consequence of the Open Mapping Theorem. See in Functional Analysis.
In other words, for a continuous isomorphism, its inverse map is automatically continuos.

Definition 6.A.4. We denote by Isom(E, F ) the set of continuous isomorphisms mapping

from E into F .

The following results have somewhere appeared in previous chapters. They will useful
in the future.
Proposition 6.A.5. Let E and F be Banach spaces. Then,
1.) The space of operators L (E, F ) is a Banach space. And L (E, E) is a Banach algebra
for the composition law.
2.) The set Isom(E, F ) is open in L (E, F ).

Isom(E, F ) −→ Isom(F, E)
3.) The inversion map Inv : is differentiable.
u 7−→ u−1
Proof. 1.) They correspond exactly to Proposition 4.5.7 and Theorem 4.5.8.
2.) Consider u ∈ Isom(E, F ) and let h ∈ L (E, F ). Since u is linear and invertible, we
have
u + h = u ◦ (IdE + u−1 ◦ h).
Notice also that the inequality
∥u−1 ◦ h∥L (E,E) ≤ ∥u−1 ∥L (F,E) ∥h∥L (E,F ) < 1
holds once we have ∥h∥L (E,F ) < ∥u−1 ∥−1 L (F,E) . In the consequence, by using the
Banach algebra structure of L (E, E), the operator IdE + u−1 ◦ h is invertible and its
inversion can be expressed by Neumann series Proposition 4.1.25. Then, u + h can be
considered as the composition of two invertible operators, it is therefore invertible.
1
In conclusion, for all h ∈ L (E, F ) with ∥h∥L (E,F ) < −1 , the operator u + h
∥u ∥L (F,E)
is an isomorphism from E into F . This shows that Isom(E, F ) is an open set in
L (E, F ).
222 DIFFERENTIAL CALCULUS

3.) Consider u ∈ Isom(E, F ) and h ∈ L (E, F ). From the previous point, if ∥h∥L (E,F ) is
small enough, u + h is also invertible. Hence,

(u + h)−1 = (IdE + u−1 ◦ h)−1 ◦ u−1 .

Here, (IdE + u−1 ◦ h)−1 can be developed in Neumann series.

+∞
X +∞
X
(IdE + u−1 ◦ h)−1 = (−u−1 ◦ h)n = IdE − u−1 ◦ h + (−u−1 ◦ h)n .
n=0 n=2

Then, when ∥h∥L (E,F ) being small enough, we have,

+∞
!
X
(u + h)−1 = u−1 − u−1 ◦ h ◦ u−1 + (−u−1 ◦ h)n ◦ u−1 .
n=2

We have also the estimation,

+∞ +∞
!
X X
(−u−1 ◦ h)n ◦ u−1 ≤ (−u−1 ◦ h)n ∥u−1 ∥L (F,E)
n=2 L (F,E) n=2 L (E,E)
+∞
X
≤ ∥u−1 ∥L (F,E) ∥u−1 ∥nL (F,E) ∥h∥nL (E,F )
n=2
1
= ∥u−1 ∥3L (F,E) ∥h∥2L (E,F ) = o(∥h∥L (E,F ) ).
1− ∥u−1 ∥ L (F,E) ∥h∥L (E,F )

Thus, the inversion map Inv : u 7→ u−1 is differentiable with the differential

∀h ∈ L (E, F ), dInv(u)(h) = −u−1 ◦ h ◦ u−1 .

It is not difficult to verify dInv(u) ∈ L (L (E, F ), L (E, F )).

6.A.2 Mean Value Inequalities

Now, we present the generalization of the mean value theorem or inequality, which is already
introduced in Calculus. We start with recalling the C k -class functions.

Definition 6.A.6. Let f : V → W be a differentiable function.

• We say f is continuously differentiable or of C 1 -class if the differential

V −→ L (E, F )
df :
x 7−→ df (x)

is continuous on V .

• Let k ∈ N∗ , we say f is of C k -class if df is of C k−1 -class.

• We say f is of C ∞ -class or smooth if f is of C k -class for all k ∈ N∗ .

6.A. INVERSE AND IMPLICIT FUNCTION THEOREMS (OUT OF PROGRAM) 223

Theorem 6.A.7 (Mean Value Ineqality). Let V ⊂ E be an open set and let f : V → F be
a differentiable function on V .

• Cases E = R. Consider a segment [a, b] ⊂ U . We suppose there exists a continuous

function g : [a, b] → R which is differentiablevii on (a, b) and satisfies

∀t ∈ (a, b), ∥df (t)∥L (R,F ) ≤ g ′ (t).

Then, it holds that ∥f (b) − f (a)∥F ≤ g(b) − g(a).

• General cases. We assume that V is convex and that the differential df is bounded on
V . Then, for all x, y ∈ V , it holds that

∥f (y) − f (x)∥F ≤ ∥y − x∥E sup ∥df (z)∥L (E,F ) .

z∈V

Proof. • Let ε > 0. We will establish the inequality holds for all t ∈ [a, b],

∥f (t) − f (a)∥F ≤ g(t) − g(a) + ε(t − a) + ε. (6.4)

We observe at first that (6.4) holds for t = a. Introduce the set

A := {t ∈ [a, b] | (6.4) holds on [a, t]} .

We notice that the set A is a connected component of the pre-image of (−∞, ε] by

the function t 7→ ∥f (t) − f (a)∥F − (g(t) − g(a)) − ε(t − a). Using then the continuities
of f and g, we deduce that A is a closed set.
Let c := sup A. It is immediate from the closedness that c ∈ A. We assume in the
contrary that c < b.
Using the differentiability of f and g at the point c, there exists δ > 0 such that for
all s ∈ (0, δ),
εs
∥f (c + s) − f (c)∥F ≤ s∥df (c)∥L (R.F ) + ,
2
′ εs
|g(c + s) − g(c) − sg (c)| ≤ .
2
Apply the condition ∥df (c)∥L (R,F ) ≤ g ′ (c), we obtain,

∥f (c + s) − f (c)∥F ≤ g(c + s) − g(c) + εs.

In the consequence,

∥f (c + s) − f (a)∥F ≤ ∥f (c + s) − f (c)∥F + ∥f (c) − f (a)∥F

≤ g(c + s) − g(c) + sε + g(c) − g(a) + ε(c − a) + ε
= g(c + s) − g(a) + ε(c + s − a) + ε.

It shows that c + s ∈ A for all s ∈ (0, δ). And it is contradictory to the definition of c.
In conclusion, c = b. In other words, (6.4) holds for all t ∈ [a, b]. The claim of this
point follows by passing ε → 0.
vii
in the sense of a function with one real variable
224 DIFFERENTIAL CALCULUS

• The convexity assumption states that

{(1 − t)x + ty | t ∈ [0, 1]} ⊂ V.

Then, the claim follows from the previous point by considering the functions

fe : t 7→ f ((1 − t)x + ty);

ge : t 7→ t∥y − x∥E sup ∥df (z)∥L (E,F ) .
z∈V

Corollary 6.A.8. Let f : V → W be a differentiable function and let x0 ∈ V . We assume

that df is continuous at x0 . Then, for all ε > 0, there exists r > 0 such that the function
g : x 7→ f (x) − f (x0 ) − df (x0 )(x − x0 ) is ε-Lipschitz continuous in the ball B(x0 , r).

Proof. From the differentiability of f implies the function g is also differentiable on V . And
dg is continuous on at x0 with

dg(x0 ) = df (x0 ) − df (x0 ) = 0L (E,F ) .

Let ε > 0. It follows from the continuity of dg, there exists r > 0 such that ∥dg(x)∥L (E,F ) ≤
ε for all x ∈ B(x0 , r). Then, the claim follows from the convexity of B(x0 , r) and Theo-
rem 6.A.7.

6.A.3 Statements of Inverse Function Theorems

Definition 6.A.9. We call the function f as a C k -diffeomorphism from V into W if

• f is a bijection from V into W ;

• f is of C k -class;

• the inverse function f −1 : W → V is also a C k -class function.

We begin with a remark. Notice that a bijective and C 1 -class function can be a homeo-
morphism without being a diffeomorphism. A typical counter example is the function

R −→ R
f:
x 7−→ x3

It is obvious that f is C 1 -class and is a homeomorphism mapping R into itself. However,

the inverse function

R −→ R
f −1 : √
x 7−→ 3 x

is not differentiable at 0. We can also observe that f ′ (x) = 3x2 with f ′ (0) = 0. In fact, we
the following lemma is a direct consequence from chain rule Theorem 6.2.1.
6.A. INVERSE AND IMPLICIT FUNCTION THEOREMS (OUT OF PROGRAM) 225

Lemma 6.A.10. Let f be a homeomorphismviii from V into W . We assume that f is

differentiable at a point x ∈ V . Then, the inverse function f −1 is differentiable at y =
f (x) ∈ W if, and only if df (x) ∈ Isom(E, F ). Moreover, we have,

d(f −1 )(y) = df (x)−1 .

Proof. ⇒: We assume that f −1 is differentiable at y = f (x). Using Chain Rule Theo-

rem 6.2.1 on the identities f −1 ◦ f = IdE and f ◦ f −1 = IdF , we obtain,

IdE = d(f −1 )(y) ◦ df (x), and IdF = df (x) ◦ d(f −1 )(y).

They show that d(f −1 )(y) is invertible and d(f −1 )(y) = df (x)−1 .
⇐: Since f is differentiable at x, we have the asymptotic expansion,

f (x + hx ) = f (x) + df (x)(hx ) + ∥hx ∥E εx (hx ).

Here, εx : V − {x} → F satisfying ∥εx (hx )∥F −→ 0. Since we suppose that

∥hx ∥E →0
df (x) ∈ Isom(E, F ), we have also df (x)−1 ∈ L (F, E) by Theorem 6.A.3 and,

hx = df (x)−1 [f (x + hx ) − f (x) − ∥hx ∥E εx (hx )]. (6.5)

On the other had, using the bijectivity of f , for all hy ∈ W − {y}, we can write
f (x + hx ) = y + hy with hx = f −1 (y + hy ) − x. Moreover, (6.5) becomes,

hx = df (x)−1 (hy − ∥hx ∥E εx (hx )).

Taking the norms, we have,

∥hx ∥E ≤ ∥df (x)−1 ∥L (F,E) (∥hy ∥F + ∥hx ∥E ∥εx (hx )∥F ).

In the consequences, the inequality

∥df (x)−1 ∥L (F,E)
∥hx ∥E ≤ ∥hy ∥F (6.6)
1 − ∥df (x)−1 ∥L (F,E) ∥εx (hx )∥F

holds for ∥hx ∥E small enough such that 1 > ∥df (x)−1 ∥L (F,E) ∥εx (hx )∥F . And this
scenario is guaranteed by the convergence ∥εx (hx )∥F −→ 0.
∥hx ∥E →0

Now, we rewrite (6.5) as

f −1 (y + hy ) − f −1 (y) = df (x)−1 hy + ∥hy ∥F εy (hy ),

∥hx ∥E
where εy (hy ) = εx (hx ). By using the estimation (6.6), we have
∥hy ∥F

∥df (x)−1 ∥L (F,E)

∥εy (hy )∥F ≤ ∥εx (hx )∥E .
1 − ∥df (x)−1 ∥L (F,E) ∥εx (hx )∥F

The continuity of f −1 implies that hx → 0E in E while hy → 0F in F . Letting

∥hy ∥F → 0, it follows that ∥εx (hx )∥E → 0, and hence, ∥εy (hy )∥F → 0.
In conclusion, f −1 is differentiable at y = f (x) with the differential df (x)−1 .

viii
That means f is continuous, bijective and f −1 is also continuous. See in Definition 3.6.17.
226 DIFFERENTIAL CALCULUS

Proposition 6.A.11. Let f be a homeomorphism from V into W . We assume that f is of

C 1 -class on V . Then, f is a C 1 -diffeomorphism from V into W if, and only if for all x ∈ V ,
df (x) ∈ Isom(E, F ).
Proof. ⇒: If f is a C 1 -diffeomorphism, it is in particular differentiable at each point x in
V . It is sufficient to apply Lemma 6.A.10 to obtain that df (x) ∈ Isom(E, F ).
⇐: Applying Lemma 6.A.10, the inverse function f −1 is differentiable at each point x ∈ V
and d(f −1 )(x) = df (x)−1 . Precisely speaking, the differential of f −1 is given by,

W −→ L (F, E)
d(f −1 ) : .
y 7−→ df (f −1 (y))−1

We can see d(f −1 ) as the composition of the following three functions

1.) f −1 : W → V is continuous since f is a homeomorphism;
2.) df : V → L (E, F ) is continuous since f is supposed to be C 1 -class on V ;
3.) Inv : Isom(E, F ) → Isom(F, E) is continuous since it is differentiable from
Proposition 6.A.5.
In conclusion, d(f −1 ) is of C 1 -class on W .

The main objective in this section is to characterize the C k -diffeomorphisms, which are
formulated into the following two fundamental theorems.

Theorem 6.A.12 (Local Inverse Function). Let U ⊂ E be an open set. Let f : U → F be

a C k -class function. We suppose furthermore that there exists a point x0 ∈ U such that

df (x0 ) ∈ Isom(E, F ).

Then, there exists an open neighborhood V ∈ N (x0 ), V ⊂ U and an open set W ⊂ F such
that f is a C k -diffeomorphism from V into W .

The proof of Local Inverse Function Theorem 6.A.12 is long and difficult, we will show
it in the next subsection.

Corollary 6.A.13 (Global Inverse Function). Let U ⊂ E be an open set. The C k -class
function f is a C k -diffeomorphism from U into its image f (U ) ⊂ F if, and only if
• f is injective on U ;
• for all x ∈ U , df (x) ∈ Isom(E, F ).
Proof. ⇒: It is sufficient to apply Proposition 6.A.11.
⇐: We apply the Local Inverse Function Theorem 6.A.12 at each point x ∈ U since we
have df (x) ∈ Isom(E, F ). Then, for all x ∈ U , there exists Vx ⊂ U such that f is
a C k -diffeomorphism from an open set Vx into another open set Wx ⊂ f (U ). In the
consequences,
[ [
f (U ) ⊂ Wx ⊂ f (U ) = f (U ).
x∈f (U ) x∈f (U )
6.A. INVERSE AND IMPLICIT FUNCTION THEOREMS (OUT OF PROGRAM) 227
[
It shows that f (U ) = Wx is an open set since it is the union of open sets Wx .
x∈f (U )

On the other hand, the injectivity of f implies that f is bijective from U into f (U ).
In addition, for all y ∈ f (U ), denoting x = f −1 (y), the restriction of f −1 on Wx is
simply the inverse of the restriction of f on Vx , i.e., f −1 |Wx = f |−1
Vx since f forms a
−1
C -diffeomorphism from Vx into Wx . In conclusion, f is also of C k -class.
k

Remark 6.A.14. The main differences between Proposition 6.A.11 and Corollary 6.A.13
consist that
• in Proposition 6.A.11, f −1 is supposed to be continuous whereas in Corollary 6.A.13,
this condition is dropped out.
• in Proposition 6.A.11, the assumption “image f (V ) = W is open” is indispensable
whereas in Corollary 6.A.13, this condition can be derived from Local Inverse Function
Theorem 6.A.12.

6.A.4 Proof of Local Inverse Function Theorem

Lemma 6.A.15. Let B(x0 , r) ⊂ E denote the open ball centered at x0 with the radius r > 0
contained in the Banach space E. Consider a continuous function f : B(x0 , r) → E. We
suppose that the function φ : x 7→ x − f (x) is a contraction, i.e., K-Lipschitz continuous
with 0 < K < 1. Then, by setting y0 := f (x0 ) ∈ E, it holds that
• there exists an open set V satisfying x0 ∈ V ⊂ B(x0 , r) such that f is an homeomor-
phism from V into B(y0 , (1 − K)r);
1
• the inverse function f −1 : B(y0 , (1 − K)r) → V is -Lipschitz continuos.
1−K
Proof. We will show in order that 1.) f is injective on B(x0 , r), 2.) f is surjective into
B(y0 , (1 − K)r), 3.) the inverse function f −1 is Lipschitz continuous on V ⊂ B(x0 , r).
1.) Consider x1 , x2 ∈ B(x0 , r). Since φ is a contraction, we have,
∥(x1 − x2 ) − (f (x1 ) − f (x2 ))∥E = ∥φ(x1 ) − φ(x2 )∥E ≤ K∥x1 − x2 ∥E .
Using the triangular inequality in E, we deduce that
∥x1 − x2 ∥E − ∥f (x1 ) − f (x2 )∥E ≤ ∥(x1 − x2 ) − (f (x1 ) − f (x2 ))∥E ≤ K∥x1 − x2 ∥E .
Hence,
1
∥x1 − x2 ∥E ≤ ∥f (x1 ) − f (x2 )∥E . (6.7)
1−K
This estimation shows that the condition f (x1 ) = f (x2 ) implies x1 = x2 . Thus, f is
injective.
2.) Let y ∈ B(y0 , (1 − K)r). We will construct the pre-image of y by f using Picard’s
iterations. Consider now the sequence (xn )n∈N ∈ E N given by,
∀n ∈ N, xn+1 = y + φ(xn ).
At first, we need to prove the sequence (xn )n∈N is well-defined, i.e., for all n ∈ N,
xn ∈ B(x0 , r).
228 DIFFERENTIAL CALCULUS

• x0 ∈ B(x0 , r) is obvious.
• For n = 1, we have x1 − x0 = y + φ(x0 ) − x0 = y − y0 . Then, ∥x1 − x0 ∥E =
∥y − y0 ∥E < (1 − K)r < r, and hence, x1 ∈ B(x0 , r).
• Assume now n ∈ N∗ and xj ∈ B(x0 , r) for all j ∈ [[0, n]]. Since φ is a contraction,
we deduce that

∥xj+1 − xj ∥E = ∥φ(xj ) − φ(xj−1 )∥E ≤ K∥xj − xj−1 ∥E .

It is sufficient to repeat the above inequality to obtain,

∥xj+1 − xj ∥E ≤ K j ∥x1 − x0 ∥ = K j ∥y − y0 ∥E . (6.8)

Also, we have,
n
X n
X
∥xn+1 − x0 ∥E ≤ ∥xj+1 − xj ∥E ≤ ∥y − y0 ∥E Kj
j=0 j=0
+∞
X 1 (1 − K)r
≤ ∥y − y0 ∥E Kj = ∥y − y0 ∥E < = r. (6.9)
1−K 1−K
j=0

It shows xn ∈ B(x0 , r).

Following the mathematical induction, the sequence (xn )n∈N is well-defined.

On the other hand, let n, p ∈ N , it follows from (6.8), that
n+p−1 n+p−1
X X ∥y − y0 ∥E n
∥xn+p − xn ∥E ≤ ∥xj+1 − xj ∥E ≤ ∥y − y0 ∥E Kj ≤ K −→ 0.
1−K n→+∞
j=n j=n

And the convergence is independent to the parameter p. It shows that (xn )n∈N is a
Cauchy sequence in E, and hence it converges to x ∈ E. Passing n → +∞ in (6.9),
we obtain x ∈ B(x0 , r). Also, using the continuity of φ, we have,

x = lim xn+1 = lim (y + φ(xn )) = y + φ( lim xn ) = y + φ(x).

n→+∞ n→+∞ n→+∞

It shows that y = f (x). Hence, f is surjective.

3.) It is sufficient to set V as the pre-image of B(y0 , (1 − K)r) by f . Then we have

a ∈ V ⊂ B(x0 , r) by using the previous points. Then, f is bijective from V into
B(y0 , (1−K)r) and the Lipschitz continuity of f −1 follows directly from the inequality
(6.7).

Proof of Local Inverse Function Theorem 6.A.12. We begin with assuming k = 1.

Under the assumptions in Theorem 6.A.12, df (x0 ) ∈ Isom(E, F ), we can thus consider
the function

U −→ E
f1 : −1
.
x 7−→ df (x0 ) (f (x))
6.A. INVERSE AND IMPLICIT FUNCTION THEOREMS (OUT OF PROGRAM) 229

Then, f1 is of C 1 -class on U since it can be seen as the composition of three C 1 -class functions
f , df (x0 ) and the inversion map. In fact, for h ∈ U − {x0 } such that ∥h∥E being small
enough, we have,

f1 (x0 + h) = df (x0 )−1 [f (x0 ) + df (x0 )(h) + ∥h∥E ε0 (h)]

= f1 (x0 ) + IdE (h) + ∥h∥E df (x0 )−1 (ε0 (h)).

Using the continuity of df (x0 )−1 and the convergence ε0 (h) −→ 0, we obtain the differ-
∥h∥E →0
entiability of f1 at x0 as well as df1 (x0 ) = IdE .
Let K ∈ (0, 1). Since f1 is of C 1 -class on U , the differential x 7→ df1 (x) is continuous
on U . It follows from the consequence of Mean Value Inequality Corollary 6.A.8 that there
exists r > 0 such that g : x 7→ f1 (x) − f1 (x0 ) − df1 (x0 )(x − x0 ) is K-Lipschitz continuous on
B(x0 , r). Introducing also the function φ : x 7→ x − f1 (x), we have, for all x, y ∈ B(x0 , r),

φ(x) − φ(y) = (x − y) − (f1 (x) − f1 (y)) = (x − y) − [(g(x) − g(y)) + df1 (x0 )(x − y)]
= g(y) − g(x).

It shows that φ is a contraction on B(x0 , r). We are at the position to apply Lemma 6.A.15.
From the lemma, there exists an open neighborhood Ve ⊂ B(x0 , r) of x0 such that f1
is an homeomorphism from Ve into B(f1 (x0 ), (1 − K)r). In the consequences, since df (x0 )
is invertible, the original function f = df (x0 ) ◦ f1 is an homeomorphism mapping from Ve
into Wf := df (x0 )(B(f1 (x0 ), (1 − K)r)) ⊂ F . We remark that W f is also an open set in F
by using Theorem 6.A.3.
It remains to prove the inverse function f −1 is also of C 1 -class. Now, f is an homeo-
morphism from Ve into W f and it is differentiable at x0 ∈ Ve . Moreover, f is of class C 1 ,
which states that the mapping x 7→ df (x) is continuous. Notice df (x0 ) ∈ Isom(E, F ) and
Isom(E, F ) is an open set in L (E, F ) via Proposition 6.A.5. We can deduce that there
exists an open set V ⊂ Ve such that for all x ∈ V , df (x) ∈ Isom(E, F ) as the pre-image of
an open set B(df (x0 ), δ) by the continuous mapping x 7→ df (x). Here δ > 0 is small enough
such that B(df (x0 ), δ) ⊂ Isom(E, F ). Now, we can apply Proposition 6.A.11 to deduce that
f is a C 1 -diffeomorphism mapping from V into W = f (V ), which is also an open set since
f is an homeomorphism.
The cases where k ≥ 2 can be obtained by proceeding mathematical inductions. The
proof is finished.

6.A.5 Implicit Function Theorem

Theorem 6.A.16 (Implicit Funciton). Let E, F and G be Banach spaces. Consider f :
U → G where U ⊂ E × F being an open set. We assume that

• f is of C k -class on U ;

• (x0 , y0 ) ∈ U such that f (x0 , y0 ) = 0G ;

• the partial derivative ∂y f (x0 , y0 ) ∈ L (F, G) is an isomorphism.

Then, there exists

1.) an open neighborhood V of (x0 , y0 ) in E × F ;

2.) an open neighborhood W of x0 in E;

230 DIFFERENTIAL CALCULUS

3.) a C k -class function g : W → F ,

which satisfy

(x, y) ∈ V, and f (x, y) = 0G ⇔ x ∈ W, and y = g(x).

Proof. We will use Local Inverse Function Theorem 6.A.12. To do this, we introduce

U −→ E×G
f1 : .
(x, y) 7−→ (x, f (x, y))

Calculating the differential of f1 , it gives,

E×F −→ E×G
df1 (x0 , y0 ) : .
(h, k) 7−→ (h, ∂x f (x0 , y0 )h + ∂y f (x0 , y0 )k)

Since ∂y f (x0 , y0 ) is an isomorphism, the differential df1 (x0 , y0 ) is also an isomorphism, of

which the inverse map is given by,

E × G −→ E×F
df1 (x0 , y0 )−1 : −1
.
(u, v) 7−→ (u, ∂y f (x0 , y0 ) [v − ∂x f (x0 , y0 )(u)])

Apply now the Local Inverse Function Theorem 6.A.12, there exists an open neighborhood
V of (x0 , y0 ) in U ⊂ E × F and an open neighborhood W1 of f1 (x0 , y0 ) = (x0 , 0G ) such that
f1 is a C 1 -diffeomorphism from V into W1 .
Now, it is licit to introduce g1 = f1−1 mapping from W1 into V . In addition, we can
write

∀(x, z) ∈ W1 , g1 (x, z) = (x, g0 (x, z)).

We have thus defined the function g0 : W1 → F , which is of C 1 -class since g1 is. Moreover,
we have the equivalence,

(x, y) ∈ V, f (x, y) = z ⇔ (x, y) ∈ V, f1 (x, y) = (x, z) ∈ W1

⇔ (x, z) ∈ W1 , g1 (x, z) = (x, y) ∈ V
⇔ (x, z) ∈ W1 , g0 (x, z) = y.

In particular, taking z = 0G , we obtain that

(x, y) ∈ V, f (x, y) = 0G ⇔ (x, 0G ) ∈ W1 , g0 (x, 0G ) = y.

W −→ F
Define now W := {x ∈ E | (x, 0G ) ∈ W1 } and g : . Since W1 is open
x 7−→ g0 (x, 0G )
in E × G, the hyperplane W is thus open in E. Also, since g0 is of C 1 -class, the function g
is also of C 1 -class. Hence, the proof is finished.
 Chapter 7

Sequences and Series of Functions

In this chapter, (E, ∥ · ∥E ) and (F, ∥ · ∥F ) denote the normed spaces over the field K := R
or C. We will consider in all generality the functions mapping from A into F where A ⊂ E
refers a non-empty subset in E.

7.1 Sequence of Functions

In this section, we denote respectively by (fn )n∈N and f a sequence of functions and a
function mapping from A into F .

7.1.1 Different Types of Convergence

Definition 7.1.1. We say the sequence of functions (fn )n∈N converges pointwisely to
the function f if for all x ∈ A, the sequence of vectors (fn (x))n∈N converges to f (x) in the
normed space (F, ∥ · ∥F ). In the case of pointwise convergence, we say f is the pointwise
limit of the sequence of functions (fn )n∈N .

Remark 7.1.2. • We say the sequence of functions (fn )n∈N converges pointwisely if for
all x ∈ A, the sequence (fn (x))n∈N ∈ F N converges.
• The pointwise limit f of a sequence of functions (fn )n∈N is unique since for all x ∈ A,
the limit of the sequence (fn (x))n∈N is unique.

Example 7.1.3. Let E = F = R.

1.) We consider the sequence of functions (fn )n∈N defined on [0, 1] by, for all n ∈ N,
x ∈ [0, 1], fn (x) = xn . We have immediately that (fn )n∈N converges pointwisely to
the function f given by,
(
0 if x ∈ [0, 1),
f (x) =
1 if x = 1.

2.) Consider the sequence of functions (fn )n∈N defined on R by, for all n ∈ N,
(
n2 x if |x| ≤ n1 ,
fn (x) = 1
x if |x| > n1 .

231
232 SEQUENCES AND SERIES OF FUNCTIONS

Then, the sequence of functions (fn )n∈N converges pointwisely and its pointwise limit
function f is given by,
(
0 if x = 0,
f (x) = 1
x if x ̸= 0.

Definition 7.1.4. We say the sequence of functions (fn )n∈N converges uniformly on A
if for all ε > 0, there exists Nε ∈ N such that

∀n ∈ N, n ≥ Nε ⇒ ∀x ∈ A, ∥fn (x) − f (x)∥F ≤ ε.

In this case, we say f is the uniform limit on A of (fn )n∈N .

Remark 7.1.5. • We say the sequence of functions (fn )n∈N converges uniformly on A if
there exists a function f defined on A such that (fn )n∈N converges uniformly on A to
f.

• The sequence of functions (fn )n∈N converges uniformly on A to f is equivalent to say

the following convergence,

sup ∥fn (x) − f (x)∥F −→ 0. (7.1)

x∈A n→+∞

• Following from the previous point, the uniform convergence on A can be equivalently
+ satisfying αn −→ 0 such that
reformulate as: there exists a sequence (αn )n∈N ∈ RN
n→+∞
sup ∥fn (x) − f (x)∥F ≤ αn .
x∈A

• Reciprocally, if the sequence of functions (fn )n∈N does not converge uniformly on A
to f , then there exists ε0 > 0 and a sequence (xn )n∈N ∈ AN as well as a subsequence
(fq(n) )n∈N of functions such that for all n ∈ N, ∥fq(n) (xn ) − f (xn )∥F > ε0 .

Remark 7.1.6. Using the ε − N formulations, the pointwise convergence and the uniform
convergence can be written as follows.

• The sequence of functions (fn )n∈N converges pointwisely to f if, and only if

∀x ∈ A, ∀ε > 0, ∃Nx,ε ∈ N s.t. ∀n ∈ N, n ≥ Nx,ε ⇒ ∥fn (x) − f (x)∥F ≤ ε.

• The sequence of functions (fn )n∈N converges uniformly to f if, and only if

∀ε > 0, ∃Nε ∈ N s.t. ∀n ∈ N, n ≥ Nε ⇒ ∀x ∈ A, ∥fn (x) − f (x)∥F ≤ ε.

Proposition 7.1.7. If the sequence of functions (fn )n∈N converges uniformly to f , then it
converges pointwisely to f .

Proof. Left in Exercise 7.4.1.

7.1. SEQUENCE OF FUNCTIONS 233

Example 7.1.8. 1.) In the first case in Example 7.1.3, the sequence of functions (fn )n∈N
does not converge uniformly on [0, 1] to f . In fact, we have that for all n ∈ N,
αn := sup |xn − f (x)| = 1 which does not converge to 0. We then deduce that this
x∈[0,1]
sequence of functions (fn )n∈N does not converge uniformly on [0, 1] since if so, the
uniform limit must coincide with the pointwise limit.

2.) Let γ ∈ R+ and (fn )n∈N∗ be a sequence of functions defined by, for all n ∈ N,
R+ −→ R
fn : . Then, let x ∈ R+ be given, we have immediately that
x 7−→ n xe−nx
γ

fn (x) −→ 0. So, the sequence of functions (fn )n∈N∗ converges pointwisely to the
n→+∞
null function. We then define for all n ∈ N, αn := sup |fn (x) − 0| = sup |nγ xe−nx |.
x∈R+ x∈R+
1 nγ−1
We can obtain that αn = fn ( ) = . We finally deduce that if γ < 1, then the
n e
sequence of functions (fn )n∈N∗ converges uniformly on R+ to the null function and if
γ ≥ 1, the convergence is only pointwise.

3.) Consider the sequence of functions (fn )n∈N defined by, for all n ∈ N, ∀x ∈ E, fn (x) =
nx
. We can obtain immediately that (fn )n∈N converges pointwisely to the
1 + n∥x∥E
x
function f given by, ∀x ∈ E \ {0E }, f (x) = and f (0E ) = 0. Let r > 0 and
∥x∥E
x ∈ B(0, r)∁ , we have the estimation

nx x 1 1
∥fn (x) − f (x)∥E = − = ≤ .
1 + n∥x∥E ∥x∥E E 1 + n∥x∥E 1 + nr

Hence, the sequence of functions (fn )n∈N converges uniformly on B(0E , r)∁ to the
function f .

Proposition 7.1.9. Let (fn )n∈N and (gn )n∈N be sequences of functions which converges
uniformly on A respectively to f and g. Let α, β ∈ R or C. Then, the sequence of functions
(αfn + βgn )n∈N converges uniformly on A to αf + βg.
Proof. For all x ∈ A, we have the estimation

∥(αfn + βgn )(x) − (αf + βg)(x)∥F ≤ |α|∥fn (x) − f (x)∥F + |β|∥gn (x) − g(x)∥F
≤ |α| sup ∥fn (x) − f (x)∥F + |β| sup ∥gn (x) − g(x)∥F .
x∈A x∈A

And it leads to

sup ∥(αfn + βgn )(x) − (αf + βg)(x)∥F

x∈A
≤ |α| sup ∥fn (x) − f (x)∥F + |β| sup ∥gn (x) − g(x)∥F −→ 0.
x∈A x∈A n→+∞

Hence, the claim follows.

Remark 7.1.10. The statement of Proposition 7.1.9 holds for pointwise convergence.
234 SEQUENCES AND SERIES OF FUNCTIONS

Definition 7.1.11. We say the sequence of functions (fn )n∈N converges uniformly on any
compact in A to f if for any compact subset K ⊂ A, the sequence of functions (fn |K )n∈N
converges uniformly on K to f |K .

Remark 7.1.12. If (fn )n∈N converges uniformly on A, then it converges uniformly on any
compact in A. The reciprocal statement is false. Consider the case where fn : x 7→ xn on
[0, 1). This sequence of functions converges uniformly on any compact in [0, 1) to the null
function but it does not converges uniformly on [0, 1).

Example 7.1.13. Consider the second case in Example 7.1.8, the sequences of functions
(fn )n∈N∗ does not converge uniformly on R+ if γ ≥ 1. However, this sequences of functions
converges on any compact in R∗+ . In fact, let [a, b] ⊂ R∗+ be a compact set, we have
1
sup |fn (x) − 0| = fn (a) = nγ ae−na −→ 0 for all n ∈ N such that n ≥ .
x∈[a,b] n→+∞ a

Definition 7.1.14. We say the sequence of functions (fn )n∈N is a uniform Cauchy se-
quence on A if for all ε > 0, there exists N ∈ N such that for all n, m ∈ N, N ≤ n ≤ m ⇒
∀x ∈ A, ∥fn (x) − fm (x)∥F ≤ ε.

Remark 7.1.15. The above definition can be reformulate as follows.

∀ε > 0, ∃N ∈ N, ∀n, p ∈ N, N ≤ n ⇒ ∀x ∈ A, ∥fn (x) − fn+p (x)∥F ≤ ε.

Theorem 7.1.16. Let (fn )n∈N be a sequence of functions mapping from the subset A ⊂ E
into the Banach space F . Then, (fn )n∈N is uniformly convergent on A if, and only if
(fn )n∈N is a uniform Cauchy sequence on A.
Proof. We prove this theorem by double implications.
⇒: We assume that the sequence of functions (fn )n∈N converges uniformly on A to f .
Let ε > 0, there exists N ∈ N such that for all n ∈ N, n ≥ N , we have ∀x ∈ A,
ε
∥fn (x) − f (x)∥F ≤ . Then, for all n, p ∈ N, n ≥ N , we have that ∀x ∈ A,
2
ε ε
∥fn (x) − fn+p (x)∥F ≤ ∥fn (x) − f (x)∥F + ∥f (x) − fn+p (x)∥F ≤ + = ε.
2 2
Hence, (fn )n∈N is a uniformly Cauchy sequence on A.

⇐: We assume that the sequence of functions is a uniform Cauchy sequence on A. We

prove firstly that the sequence of functions (fn )n∈N converges pointwisely then we
prove in fact, this convergence is uniform.
Let ε > 0, from the uniform convergence of (fn )n∈N on A, there exists N ∈ N such
that for all n, p ∈ N, if n ≥ N , we have,

∀y ∈ A, ∥fn (y) − fn+p (y)∥F ≤ sup ∥fn (x) − fn+p (x)∥F ≤ ε. (7.2)
x∈A

This shows that for y ∈ A being given, the sequence (fn (y))y∈N is a Cauchy sequence
in F . Since F is supposed to be a Banach space, the sequence (fn (y))y∈N converges
7.1. SEQUENCE OF FUNCTIONS 235

A −→ F
to a limit ly ∈ F . We then define a function f : . Then,
y 7−→ ly = lim fn (y)
n→+∞
we can pass p → +∞ in (7.2) to obtain that for all n ∈ N, n ≥ N ,

∀y ∈ A, ∥fn (y) − f (y)∥F ≤ ε.

Thus, the sequence of functions (fn )n∈N converges uniformly on A to f .

Definition 7.1.17. We denote by L∞ (A, F ) the vector space of bounded functions mapping
from A into F . And we define a function ∥ · ∥L∞ (A,F ) on L∞ (A, F ) by,

∀f ∈ L∞ (A, F ), ∥f ∥L∞ (A,F ) := sup ∥f (x)∥F .

x∈A

Proposition 7.1.18. 1.) ∥ · ∥L∞ (A,F ) defines a norm on L∞ (A, F ).

2.) Let (fn )n∈N be a sequence of functions of elements in L∞ (A, F ). Then, (fn )n∈N con-
verges uniformly on A if, and only if (fn )n∈N converges in the normed vector space
(L∞ (A, F ), ∥ · ∥L∞ (A,F ) ).

Proof. 1.) Left in Exercises.

2.) ⇒: We assume that (fn )n∈N converges uniformly on A to a function f . Then, there
exists N1 ∈ N such that for all x ∈ A, ∥fN1 (x) − f (x)∥F ≤ 1. We have then
∀x ∈ A,

∥f (x)∥F = ∥f (x) − fN1 (x) + fN1 (x)∥F ≤ ∥f (x) − fN1 (x)∥F + ∥fN1 (x)∥F
≤ 1 + ∥fN1 ∥L∞ (A,F ) .

It proves that f is bounded, i.e., f ∈ L∞ (A, F ). Then, it follows from (7.1) that

∥fn − f ∥L∞ (A,F ) = sup ∥fn (x) − f (x)∥F −→ 0. (7.3)

x∈A n→+∞

Hence, (fn )n∈N converges uniformly in L∞ (A, F ) under the norm ∥ · ∥L∞ (A,F ) .
⇐: We assume now there exists f ∈ L∞ (A, F ) such that (7.3) holds. Let ε > 0,
there exists N ∈ N such that

∀n ∈ N, n ≥ N ⇒ ∥fn − f ∥L∞ (A,F ) ≤ ε,

We deduce then that for all n ≥ N , ∀x ∈ A, ∥fn (x) − f (x)∥F ≤ ε. That means
(fn )n∈N converges uniformly on A to f .

Proposition 7.1.19. Let (fn )n∈N and (gn )n∈N be sequences of functions in L∞ (A, K). We
assume that (fn )n∈N converges uniformly on A to f and (gn )n∈N converges uniformly on A
to g. Then, (fn gn )n∈N converges uniformly on A to f g.
236 SEQUENCES AND SERIES OF FUNCTIONS

Proof. We have seen in Proposition 7.1.18 that f, g ∈ L∞ (A, K), as a direct consequence,
f g ∈ L∞ (A, K). Also, there the sequences (fn )n∈N and (gn )n∈N are bounded sequences for
the norm ∥ · ∥L∞ (A,K) as convergent sequences. Then, there exists M > 0 such that for all
n ∈ N, ∥fn ∥L∞ (A,K) ≤ M as well as ∥gn ∥L∞ (A,K) ≤ M . In addition, for all φ, ψ ∈ L∞ (A, K),
it is also immediate that ∥φψ∥L∞ (A,K) ≤ ∥φ∥L∞ (A,K) ∥ψ∥L∞ (A,K) . Let n ∈ N, it follows,
∥fn gn − f g∥L∞ (A,K) = ∥fn gn − f gn + f gn − f g∥L∞ (A,K)
≤ ∥(fn − f )gn ∥L∞ (A,K) + ∥f (gn − g)∥L∞ (A,K)
≤ ∥fn − f ∥L∞ (A,K) ∥gn ∥L∞ (A,K) + ∥f ∥L∞ (A,K) ∥gn − g∥L∞ (A,K)
≤ M ∥fn − f ∥L∞ (A,K) + ∥f ∥L∞ (A,K) ∥gn − g∥L∞ (A,K) −→ 0.
n→+∞

Thus, the claim follows by applying Proposition 7.1.18.

Remark 7.1.20. The statement of Proposition 7.1.19 is false if the functions (fn )n∈N and
(gn )n∈N are no longer bounded. As an example, we may consider the sequence of functions
1
defined on R. Let fn = and gn : x 7→ x for n ∈ N∗ . We have that (fn )n∈N converges
n
uniformly on R to the function zero and (gn )n∈N converges uniformly to the function x 7→ x
x
on R. But the product fn gn : x 7→ does not converge uniformly on R to the product of
n
limit functions, i.e. the zero function.

Theorem 7.1.21. If F is a Banach space, then the space (L∞ (A, F ), ∥ · ∥L∞ (A,F ) ) is also
complete.
Proof. Let (fn )n∈N be a Cauchy sequence in (L∞ (A, F ), ∥·∥L∞ (A,F ) ). From the definition of
∥ · ∥L∞ (A,F ) , (fn )n∈N is a uniform Cauchy sequence of on A. Then applying Theorem 7.1.16,
(fn )n∈N converges uniformly on A. From Proposition 7.1.18, the sequence (fn )n∈N converges
in the normed space (L∞ (A, F ), ∥ · ∥L∞ (A,F ) ).
Remark 7.1.22. Combining the results in Proposition 7.1.19 and Theorem 7.1.21, the space
of bounded functions with values in K = R or C, (L∞ (A, K), ∥ · ∥L∞ (A,K) ), forms a Banach
algebra.

7.1.2 Properties of Uniform Convergence

Theorem 7.1.23. Let a ∈ A, we suppose that
• for all n ∈ N, fn is continuous at a;
• (fn )n∈N converges uniformly on A to f .
Then, f is continuous at a.
Proof. Let ε > 0. Since (fn )n∈N converges uniformly on A to f , there exists N ∈ N such
ε
that for all x ∈ A, ∥fN (x) − f (x)∥F ≤ . Also, since fN is continuous at a, there exists
3
ε
δ > 0 such that for all x ∈ A, ∥x − a∥E ≤ δ implies that ∥fN (x) − fN (a)∥F ≤ . Then, for
3
all x ∈ A, ∥x − a∥E ≤ δ we have,
∥f (x) − f (a)∥F ≤ ∥f (x) − fN (x)∥F + ∥fN (x) − fN (a)∥F + ∥fN (a) − f (a)∥F
ε ε ε
≤ + + = ε.
3 3 3
Thus, f is continuous at a.
7.1. SEQUENCE OF FUNCTIONS 237

As the direct consequence of Theorem 7.1.23, we have the following theorem on the
continuity of the limit function of a uniformly convergent sequence of functions.
Corollary 7.1.24. Let (fn )n∈N be a sequence of continuous functions which converges uni-
formly on A. Then the uniform limit function is continuous on A.

Theorem 7.1.25. We suppose that

• for all n ∈ N, fn is continuous on A;

• the sequence of functions (fn )n∈N converges uniformly on any compact in A to f .

Then, f is continuous on A.
Proof. Let a ∈ A and (xn )n∈N be a sequence of elements in A which converges to a. We
have seen in the chapter of Topology that the set S := {xn | n ∈ N}∪{a} is a compact set in
A. Applying Theorem 7.1.23, the function f is continuous on S. Hence, f (xn ) −→ f (a)
n→+∞
which implies that f is continuous at a. Since a is chosen arbitrary in A, we deduced
therefore that f is continuous on A.

Proposition 7.1.26. Let A be a compact subset in E and let F be a Banach space. Then,
the space (C(A, F ), ∥·∥L∞ (A,F ) ) of continuous functions mapping from A into F is a Banach
space.
Proof. Since A is compact, for all continuous function f ∈ C(A, F ), the image f (A) is a
compact subset in F , it is therefore closed and bounded. In other words, the function f is
bounded on A. We have then C(A, F ) ⊂ L∞ (A, F ). Also, it follows from Theorem 7.1.23
as well as its corollary, any convergent sequence in (C(A, F ), ∥ · ∥L∞ (A,F ) ) has its limit being
a continuous function. In terms of topology, C(A, F ) is a closed subset in L∞ (A, F ).
On the other hand, it follows from the completeness of F and Theorem 7.1.21, the space
(L∞ (A, F ), ∥ · ∥L∞ (A,F ) ) is complete. Thus, (C(A, F ), ∥ · ∥L∞ (A,F ) ) is complete as a closed
subset in a complete space.

Theorem 7.1.27 (Interchange of Limits). Let a ∈ A. We suppose that

• F is complete;

• for all n ∈ N, the function fn possesses a limit ln ∈ F at a;

• (fn )n∈N converges uniformly on A to f .

Then,
1.) the sequence (ln )n∈N converges to l ∈ F ;

2.) the function f has a limit at a ∈ A;

3.) lim f (x) = l.

x→a

Equivalently speaking, we have

lim lim fn (x) = lim ln = l = lim f (x) = lim lim fn (x).

n→+∞ x→a n→+∞ x→a x→a n→+∞
238 SEQUENCES AND SERIES OF FUNCTIONS

Proof. Since F is complete and (fn )n∈N converges uniformly on A, it follows from Theo-
rem 7.1.16 that (fn )n∈N is a uniform Cauchy sequence on A. Let ε > 0, there exists N ∈ N
such that for all n, p ∈ N, n ≥ N implies that ∀x ∈ A, ∥fn (x) − fn+p (x)∥F ≤ ε. From the
convergences lim fn (x) = ln and lim fn+p (x) = ln+p , we have thus ∥ln − ln+p ∥F ≤ ε, which
x→a x→a
implies that the sequence (ln )n∈N is a Cauchy sequence in F . From the completeness of F ,
the sequence (ln )n∈N converges in F . We denote the limit of this sequence by l ∈ F from
now on.
Let ε > 0, it follows from the convergence ln −→ l, there exists N1 ∈ N such that
n→+∞
ε
for all n1 ∈ N, n1 ≥ N1 implies that ∥ln1 − l∥F ≤ . Also, using the uniform convergence,
3
ε
there exists N2 ∈ N such that for all n2 ∈ N, n2 ≥ N2 implies that ∥fn2 (x) − f (x)∥F ≤
3
ε
for all x ∈ A. We chose now q ∈ N, q ≥ max{N1 , N2 } then it follows ∥lq − l∥F ≤ as well
3
ε
as ∀x ∈ A, ∥fq (x) − f (x)∥F ≤ . Moreover, from the convergence lim fq (x) = lq , there
3 x→a
ε
exists δ > 0 such that for all x ∈ A, ∥x − a∥E ≤ δ implies that ∥fq (x) − lq ∥F ≤ . Then,
3
for x ∈ A, ∥x − a∥E ≤ δ, it holds,
ε ε ε
∥f (x) − l∥F ≤ ∥f (x) − fq (x)∥F + ∥fq (x) − lq ∥F + ∥lq − l∥F ≤ + + = ε.
3 3 3
Thus, lim f (x) = l.
x→a

Remark 7.1.28. This theorem still holds at the infinity endpoints on R, i.e., if A = [c, +∞)
with a = +∞ or A = (−∞, c] with a = −∞. The corresponding proof is left in exercises.

Example
 x nx 7.1.29. Consider the sequence of functions (fn )n∈N defined by, ∀x ∈ R∗+ , fn (x) =
. Then, the sequence of functions (fn )n∈N converges pointwisely to the function
n
zero. If the sequence of functions (fn )n∈N converges uniformly on R∗+ , then the uniform
limit coincides with the pointwise limit. It follows from Theorem 7.1.27 that the sequence
(fn (0+ ))n∈N∗ converges to 0. However, for all n ∈ N∗ , lim fn (x) = 1. This contradiction
x→0+
shows that the sequence (fn )n∈N does not converges uniformly on R∗+ .

Theorem 7.1.30. Let I ⊂ R be an interior-non-empty interval. Consider the sequence of

functions (fn )n∈N mapping from I into a Banach space F . We suppose that

• for all n ∈ N, fn is continuously differentiable on I (i.e. fn ∈ C 1 (I, F ));

• (fn )n∈N converges pointwisely on I to f .

• the sequence of the derivatives (fn′ )n∈N converges uniformly on any compact in I.

Then,

1.) in fact, (fn )n∈N converges uniformly on any compact in I to f ;

2.) f is continuously differentiable on I and f ′ = lim fn′ .

n→+∞
7.1. SEQUENCE OF FUNCTIONS 239

Proof. 1.) Let [a, b] ⊂ I denotes a compact set. For all n, p ∈ N and x ∈ [a, b], since fn and
fn+p are continuously differentiable on I, it follows from the Mean Value Inequality
Theorem 8.2.14,

∥fn (x) − fn+p (x)∥F ≤ ∥fn (a) − fn+p (a)∥F + |b − a| sup ∥fn′ (y) − fn+p
′
(y)∥F
y∈[a,x]

≤ ∥fn (a) − fn+p (a)∥F + |b − a| sup ∥fn′ (y) − fn+p

′
(y)∥F .
y∈[a,b]

We remark that the least right-hand-side is independent of the point x ∈ [a, b]. We
then deduce that

sup ∥fn (x) − fn+p (x)∥F ≤ ∥fn (a) − fn+p (a)∥F + |b − a| sup ∥fn′ (y) − fn+p
′
(y)∥F
x∈[a,b] y∈[a,b]

Since (fn )n∈N converges pointwisely on I, the sequence (fn (a))n∈N is a Cauchy se-
quence. Combining the uniform convergence on any compact in I of the sequence of
functions (fn′ )n∈N , the right-hand-side above tends to 0 while n → +∞ independently
to p ∈ N.
Thus, the sequence of functions (fn )n∈N is a uniform Cauchy sequence on [a, b]. From
the completeness of F , (fn )n∈N converges uniformly on [a, b].

2.) Let x0 ∈ ˚
I and let δ > 0 such that [x0 − δ, x0 + δ] ⊂ I. We consider the sequence of
functions (gn )n∈N defined by,

fn (x0 + h) − fn (x0 )
∀h ∈ [−δ, δ] \ {0}, gn (h) = .
h
Let n, p ∈ N, we apply again the Mean Value Inequality Theorem 8.2.14 on the
function fn − fn+p and on [x0 , x0 + h] to obtain the following estimation. For all
h ∈ [−δ, δ] \ {0},
1
∥gn (h) − gn+p (h)∥F = ∥(fn − fn+p )(x0 + h) − (fn − fn+p )(x0 )∥F
h
≤ sup ∥fn′ (y) − fn+p
′
(y)∥F .
y∈[x0 −δ,x0 +δ]

In the consequences,

sup ∥gn (h) − gn+p (h)∥F ≤ sup ∥fn′ (y) − fn+p

′
(y)∥F .
h∈[−δ,δ]\{0} y∈[x0 −δ,x0 +δ]

From the uniform convergence of the sequence (fn′ )n∈N , the right-hand-side above
tends to 0 while n → +∞ independently to p ∈ N. Hence, the sequence of functions
(gn )n∈N converges uniformly on [−δ, δ]\{0}. Furthermore, it follows from the pointwise
convergence of (fn )n∈N , that the uniform limit function of (gn )n∈N is given by,

f (x0 + h) − f (x0 )
∀h ∈ [−δ, δ] \ {0}, g(h) := lim gn (h) = .
n→+∞ h
On the other hand, we have also,

∀n ∈ N, lim gn (h) = fn′ (x0 ).

h→0
240 SEQUENCES AND SERIES OF FUNCTIONS

We then apply Theorem 7.1.27 with A = [−δ, δ] \ {0} and a = 0 ∈ A to conclude that
the function g has a limit at 0. This shows that the function f is differentiable at x0 .
So,

f (x0 + h) − f (x0 )
f ′ (x0 ) = lim = lim g(h) = lim fn′ (x0 ).
h→0 h h→0 n→+∞

This equality shows that the derivative f ′ is the pointwise limit of the sequence of
functions (fn′ )n∈N . We remark also that (fn′ )n∈N is supposed to be convergent uni-
formly on any compact in I and fn′ is continuous for each n ∈ N. We finally use
Theorem 7.1.25 to obtain that f ′ is continuous on I. That is, f ∈ C 1 (I, F ).

Corollary 7.1.31. Let I ⊂ R be an interior-non-empty interval and let k ∈ N∗ ∪ {∞}.

Consider the sequence of functions (fn )n∈N mapping from I into a Banach space F . We
suppose that

• for all n ∈ N, fn is of C k -class on I;

• (fn )n∈N converges pointwisely on I to f .

(j)
• for all j ∈ [[1, k]]i the sequence of the j-th derivatives (fn )n∈N converges uniformly on
any compact in I to a function gj .

Then,

1.) in fact, (fn )n∈N converges uniformly on any compact in I to f ;

2.) f is of C k -class on I and f (j) = gj = lim fn(j) for all j ∈ [[1, k]].
n→+∞

Proof. Left in Exercise 7.4.2.

Remark 7.1.32. • In the case k ̸= ∞, the statement of Corollary 7.1.31 remains true if
(k)
we only assume that the sequence of the k-th derivatives (fn )n∈N converges uniformly
on any compact in I.

• We can also derive the versions of Theorem 7.1.30 and Corollary 7.1.31 for the func-
tions possessing the partial derivatives. This generalization is also left in Exercises for
readers.

7.2 The Space of Continuous Functions

In this section, we recall at first some essential notions about Riemann integrals on a segment
of real numbers, which will be applied in Chapter 8. Then, we introduce two important
results, namely Theorems 7.2.11 and 7.2.16, on the functions with a finite numbers of
discontinuity.
Let [a, b] ⊂ R denotes a segment of real numbers with a, b ∈ R satisfying a < b.
i
If k = ∞, [[1, k]] = N∗ .
7.2. THE SPACE OF CONTINUOUS FUNCTIONS 241

7.2.1 Partition on a segment

Definition 7.2.1. We call a partitionii on [a, b] any finite sequence P = (x0 , x1 , · · · , xn )
with n ∈ N∗ such that

a = x0 < x1 < · · · < xn = b.

We also call the norm (or mesh) of the partition P the largest length of each subintervals,

|P | = max (xj − xj−1 ).

j∈[[1,n]]

Example 7.2.2. Let n ∈ N∗ , the uniform partition of the segment [a, b] is given by,
b−a
∀j ∈ [[0, n]], xj = a + j .
n
b−a
The norm of this partition is therefore .
n

Definition 7.2.3. Let P = (x0 , x1 , · · · , xn ) and P ′ = (y0 , y1 , · · · , ym ) are partitions of the

segment [a, b]. We say P is finer or being a refinement of the partition P ′ if any term in
P ′ is also a term in P . i.e.,

∀j ∈ [[0, m]], yj ∈ {x0 , x1 , · · · , xn }.

Proposition 7.2.4. For any partitions P and P ′ on [a, b], there exists a common refine-
ment denoted by P ∨ P ′ which is simultaneously finer then P and then P ′ .

Proof. We can consider any partition on [a, b] as a finite subset in [a, b] containing a and
b. The relations of refinement defines an order (partial, not total) on the sets of partitions
in [a, b], which is corresponding to the inclusions on the finite subsets. Since the reunion
P ∪ P ′ is also a finite set containing a and b, we deduce then the existence of the common
refinement P ∨ P ′ .

7.2.2 Piecewise Continuous Functions

Definition 7.2.5. A function f : [a, b] → F is said to be piecewise continuous on
[a, b] if there exists a partition (x0 , x1 , · · · , xn ) on [a, b] such that for all j ∈ [[1, n]], the
restriction f |(xj−1 ,xj ) can be extended into a continuous function on [xj−1 , xj ]. We denote
by PC([a, b], F ) the set of piecewise continuous functions mapping form [a, b] into F .

Remark 7.2.6. We recall here some fundamental properties of the piecewise continuous
functions. Let f ∈ PC([a, b], F ).

1.) A partition (x0 , x1 , · · · , xn ) satisfying Definition 7.2.5 is called an adapted partition

to f . In the following points, we discuss the adapted partition (x0 , x1 , · · · , xn ).

2.) For all j ∈ [[1, n]], f is continuous at each points in (xj−1 , xj ).

ii
It should be distinguished with the “partitions” of a set
242 SEQUENCES AND SERIES OF FUNCTIONS

3.) For all j ∈ [[0, n − 1]], f possesses a right limit at xj .

4.) For all j ∈ [[1, n]], f possesses a left limit at xj .

5.) If f is continuous on [a, b], then f is piecewise continuous on [a, b].

6.) Any piecewise continuous function is bounded. Since f |(xj−1 ,xj ) can be extended into a
continuous function on [xj−1 , xj ] and [xj−1 , xj ] is compact in R, this extended function
has thus a compact image in F , which is in particular bounded. And the partition is
finite, we deduce therefore that f is bounded on [a, b].

7.) Any refinement of the partition (x0 , x1 , · · · , xn ) is also an adapted partition to f .

Proposition 7.2.7. The set of piecewise continuous functions PC([a, b], F ) forms a vector
space over the field R or C.

Proof. Let f, g ∈ PC([a, b], F ) and α, β ∈ R or C. Introduce the partitions (x0 , x1 , · · · , xp )

and (y0 , y1 , · · · , yq ) which are adapted respectively to f and to g where p, q ∈ N∗ . From
the properties in Section 7.2.1 about partitions, we may denote by (z0 , z1 , · · · , zr ) their
common partition with r ∈ N∗ . Then, we deduce that f and g are both continuous in
each subinterval (zj , zj+1 ), i ∈ [[0, r − 1]]. In the consequence, the function αf + βg is
continuous on the subinterval (zj , zj+1 ) and it possesses the limits from left and from right
at zi . Thus, αf + βg ∈ PC([a, b], F ). It is therefore a vector subspace in the vector space
L∞ ([a, b], F ).

Definition 7.2.8. • We call a step function on [a, b] any piecewise constant func-
tion on [a, b]. That is, if φ is a step function, there exists a partition (x0 , x1 , · · · , xn )
such that φ is a constant function on each subinterval (xj−1 , xj ) for all j ∈ [[1, n]].

• We denote by ST ([a, b], F ) the set of step functions on [a, b].

• We call the characteristic function or indication function of the subset A ⊂ E

the function 1A defined by,
(
1 if x ∈ A,
1A (x) =
0 if x ∈/ A.

• Using the characteristic functions, a step function φ on [a, b] can be expressed as

follows,
n
cj 1(xj−1 ,xj ) ,
X
φ|[a,b]\{x0 ,x1 ,··· ,xn } =
j=1

where ∀j ∈ [[1, n]], cj ∈ F represents the value of φ on the subinterval (xj−1 , xj ). The
values of φ at the points of discontinuity {x0 , x1 , · · · , xn } can be attributed arbitrarily,
which will play less importantly in the future.

Proposition 7.2.9. The set of step functions forms a vector space over the field R or C.
7.2. THE SPACE OF CONTINUOUS FUNCTIONS 243

Proof. Left in Exercise 7.4.3.

Proposition 7.2.10. Let f ∈ PC([a, b], F ). Then there exists a function g which is con-
tinuous on [a, b] and a step function φ such that f = g + φ.

Proof. We prove this proposition by using the mathematical induction on the number of
discontinuity of the function f .

• If f has 0 points of discontinuity on [a, b]. That is, f is continuous, then the statement
is trivial by setting g = f and φ = 0.

• Assuming that the statement is true for any piecewise continuous function possessing
n points of discontinuity with n ∈ N. Consider a function f ∈ PC([a, b], F ) possessing
n + 1 points of discontinuity. Let c ∈ (a, b) be a point of discontinuity of f . We define
the functions h and ψ as follows,

h(x) = f (x) if x ∈ [a, c),

h(c) = lim f (t),
t→c−
h(x) = f (x) − lim f (t) + lim f (t) if x ∈ (c, b];
t→c+ t→c−
ψ(x) = 0 if x ∈ [a, c],
ψ(x) = lim f (t) − lim f (t) if x ∈ (c, b].
t→c+ t→c−

Then, ψ is a step function and if follows f = h + ψ. Moreover, h is continuous at

c, this implies that h possesses n points of discontinuity on [a, b]. Applying now the
recurrence assumption, h = g + φ with g being continuous on [a, b] and φ being a step
function. In conclusion, f = g + φ + ψ and the claim follows.

Theorem 7.2.11. Any piecewise continuous function on a segment is the uniform limit of
a sequence of step functions. Equivalently speaking, the set of step functions is dense in the
normed space (PC([a, b], F ), ∥ · ∥L∞ ([a,b],F ) ).

Proof. Using Proposition 7.2.10, it is sufficient to prove the statement for continuous func-
tions.
Consider f ∈ C([a, b], F ). It follows from Hein’s Theoremiii that f is uniformly continu-
ous on [a, b] since f is continuous on the compact set [a, b]. Let ε > 0, there exists δ > 0 such
that for all x, y ∈ [a, b], |x − y| ≤ δ implies
 that ∥f (x) − f (y)∥F ≤ ε. Consider the partition
b−a
{x0 , x1 , · · · , xn+1 } given by, n = + 1 and for all j ∈ [[0, n]], xj = x + jδ. We have
δ
then xj+1 − xj ≤ δ and hence, ∥f (x) − f (aj )∥F ≤ ε for all x ∈ [xj , xj+1 ). We define then
the step function g by g(x) = f (xj ) for all x ∈ [xj , xj+1 ) and for all j ∈ [[0, n]]. Thus, we
have ∥f (x) − g(x)∥F ≤ ε for all x ∈ [a, b], which is equivalent to say ∥f − g∥L∞ ([a,b],F ) ≤ ε.
Then the claim follows.

iii
See the course on Topology in the first semester
244 SEQUENCES AND SERIES OF FUNCTIONS

Theorem 7.2.12. Let f be a continuous function on [a, b] and ε > 0. Then, there exists a
piecewise affine functioniv g defined on [a, b] such that ∥f − g∥L∞ ([a,b],F ) ≤ ε.

Proof. From Hein’s Theorem, f is uniformly continuous on the compact [a, b]. Let ε > 0,
there exits δ > 0 such that for all x, y ∈ [a, b], |x − y| ≤ δ implies that ∥f (x) − f (y)∥F ≤ ε.
We consider a partition P = (x0 , x1 , · · · , xn ) on [a, b] such that the length |P | ≤ δ. Then,
we construct a piecewise affine function g as follows

• For all j ∈ [[0, n]], g(xj ) = f (xj ).

xj − t t − xj−1
• For all j ∈ [[1, n]], ∀t ∈ (xj−1 , xj ), g(t) = f (xj−1 ) + f (xj ).
xj − xj−1 xj − xj−1
Hence, for all t ∈ (xj−1 , xj ),
xj − t t − xj−1
∥f (t) − g(t)∥F ≤ ∥f (xj−1 ) − f (t)∥F + ∥f (xj ) − f (t)∥F ≤ ε.
xj − xj−1 xj − xj−1

It is obvious that on the points x0 , x1 , · · · , xn , ∥g(xj ) − f (xj )∥F = 0. We conclude thus on

[a, b], ∥f − g∥L∞ ([a,b],F ) ≤ ε.

Corollary 7.2.13. Any piecewise continuous function on a segment is the uniform limit of
a sequence of piecewise affine functions. Equivalently speaking, the set of piecewise affine
functions is dense in PC([a, b], F ) for the uniform convergence topology.

Definition 7.2.14. Let f be a continuous function mapping from [0, 1] into R or C. Let
n ∈ N, we call the Bernsteinv polynomial associated to f of n-th degree the following
polynomial Bn defined by,
n
!
X n k
Bn (X) = X k (1 − X)n−k f ( ). (7.4)
k n
k=0

Proposition 7.2.15. Let f be a continuous function mapping from [0, 1] into R or C.

Then, the sequence (Bn )n∈N of Bernstein polynomial functions associated to f converges
uniformly on [0, 1] to f .

Proof. We begin with some technical results. Then we prove the statement using the uni-
form continuity of f .

• Let n ∈ N, it follows from the binomial expansion, for all x, y ∈ R,

n
!
X n
(x + y)n = xk y n−k .
k=0
k
iv
An affine function is by definition the composition of a linear application and a translation. That is,
let f : E → F be affine then ∀x ∈ E, f (x) = Lx + y0 where L ∈ L (E, F ) and y0 ∈ F . A function φ is said
to be piecewise affine on [a, b] if there exists a partition (x0 , x1 , · · · , xn ) on [a, b] such that for all j ∈ [[1, n]],
the restriction φ|(xj−1 ,xj ) is an affine function, i.e., ∀t ∈ (xj−1 , xj ), φ(t) = aj t + bj with aj , bj ∈ F
v
Sergei Natanovich Bernstein (1880-1968), Soviet (now Ukraine) mathematician. He worked on partial
differential equations, differential geometry, probability theory and approximation theory.
7.2. THE SPACE OF CONTINUOUS FUNCTIONS 245

We can consider the above formula as an identity of two polynomial functions in x

with a fixed y. We differentiate both sides respect to x and it give,
n
!
X n
n(x + y)n−1 = kxk−1 y n−k .
k=0
k

By differentiating the second time, we will obtain

n
!
X n
n(n − 1)(x + y)n−2 = k(k − 1)xk−2 y n−k .
k=0
k

Taking y = 1 − x and multiplying respectively by x and by x2 , we have the identities

n
!
X n
1= xk (1 − x)n−k , (7.5)
k=0
k
n
!
X n
nx = kxk (1 − x)n−k , (7.6)
k=0
k
n
!
X n
n(n − 1)x2 = k(k − 1)xk (1 − x)n−k . (7.7)
k=0
k

• Following the previous point and let k ∈ [[0, n]]. The following identity holds
(k − nx)2 = k(k − 1) + (1 − 2nx)k + n2 x2 .
Using the relations (7.5), (7.6) and (7.7), we have
n
!
X n
(k − nx)2 xk (1 − x)n−k
k=0
k
n n
! !
X n k n−k
X n
= k(k − 1)x (1 − x) + (1 − 2nx) kxk (1 − x)n−k + n2 x2
k=0
k k=0
k
= n(n − 1)x2 + (1 − 2nx)nx + n2 x2 = nx(1 − x). (7.8)
n
We remark that for all x ∈ [0, 1], nx(1 − x) ≤ .
4
• Since the function f is continuous on the compact set [0, 1], f is uniformly continuous
on [0, 1] and the maximum of |f | is achieved. We introduce then M := max |f (x)| ∈
x∈[0,1]
R+ . It is immediate that for all x, y ∈ [0, 1], |f (x) − f (y)| ≤ 2M .
Let ε > 0, from the uniform continuity of f , there exists δ > 0 such that for all
ε
x, y ∈ [0, 1], |x − y| ≤ δ implies that |f (x) − f (y)| ≤ . We fix now a point x ∈ [0, 1]
2
and n ∈ N. Let k ∈ [[0, n]], we distinguish the following cases.
k k ε
– If x − ≤ δ, then f (x) − f ( ) ≤ .
n n 2
k
– If x − > δ, equivalently speaking, (k − nx)2 > n2 δ 2 , we have then
n
k (k − nx)2
f (x) − f ( ) ≤ 2M ≤ 2M .
n n2 δ 2
246 SEQUENCES AND SERIES OF FUNCTIONS

We then conclude that for all k ∈ [[0, n]],

k ε (k − nx)2
f (x) − f ( ) ≤ + 2M . (7.9)
n 2 n2 δ 2

• We consider now the Bernstein polynomial functions (Bn )n∈N given by (7.4). Using
(7.5), (7.8) and (7.9), we obtain that for all x ∈ [0, 1],
n
!  
X n k n−k k
|f (x) − Bn (x)| = x (1 − x) f (x) − f ( )
k n
k=0
n
!
(k − nx)2
 
X n k n−k ε
≤ x (1 − x) + 2M
k 2 n2 δ 2
k=0
ε M
≤ +
2 2nδ 2
Since the quantities ε, δ and M are independent of n, we can thus choose for n large
enough such that |f (x) − Bn (x)| ≤ ε holds for all x ∈ [0, 1]. This means exactly that
(Bn )n∈N converges uniformly on [0, 1] to f .

Theorem 7.2.16 (Stonevi -Weierstrass). Any continuous function mapping from [a, b] into
R or C is the uniform limit of a sequence of polynomial functions. Equivalently speaking, the
space of polynomial functions is dense in the space of continuous functions over a segment.

Proof. Let f be a continuous function defined on [a, b]. We consider a change of variables
[0, 1] −→ [a, b]
φ : . So, the function f ◦ φ is a continuous function defined
x 7−→ a + (b − a)x
on [0, 1]. From Proposition 7.2.15, f ◦ φ is the uniform limit of the Bernstein polynomial
functions (Bn )n∈N associated to f ◦ φ. Thus, the function f is the uniform limit of the
x−a
sequence of polynomial functions (Bn ◦ φ−1 ) where ∀x ∈ [a, b], Bn ◦ φ−1 (x) = Bn ( )
b−a
which define effectively polynomial functions.

Remark 7.2.17. 1.) The statement of Stone-Weierstrass Theorem 7.2.16 will be false on
1
an arbitrary interval. For instance, the function x 7→ defined on (0, 1] is not
x
bounded. It can not be a limit uniform of a sequence of polynomial functions. Since
all polynomial functions are bounded on the interval (0, 1]. From Proposition 7.1.18,
the uniform limit of a sequence of bounded functions is a bounded function. This is
1
contradictory to the fact that x 7→ is unbounded.
x
2.) We will see in the chapter on Fourier series the Trigonometric Weierstrass Theorem
which states that the space of functions spanned by x 7→ sin nx and x 7→ cos nx is
dense in the space of 2π-periodic continuous functions.
vi
Marshall Harvey Stone (1903-1989), American mathematician. He works in real analysis, functional
analysis and topology. He generalized the origianl result of Weierstrass’ approximation theorem and simplied
the proof.
7.3. SERIES OF FUNCTIONS 247

7.3 Series of Functions

We proceed the analysis about the series of functions just like what we have done in the
chapter on numerical series. For the sequence of functions (fn )n∈N , we can construct the
Xn
sequence of partial sums (Sn )n∈N given by, for all n ∈ N, Sn = fk . We also denote by
X k=0
fn the series of functions. In this section, we consider such series of functions mapping
n
from A into F .

7.3.1 Different Types of Convergence

X
Definition 7.3.1. Let fn be a series of functions mapping from A into F .
n
X
• We say the series of functions fn converges pointwisely on A if the sequence
n
of partial sums (Sn )n∈N convergesX
pointwisely on A. Equivalently speaking, for each
fixed x ∈ A, the numerical series fn (x) converges in F .
n

• In the case of convergence, the pointwise limit of the sequence (Sn )n∈N is denote by
+∞
X X
S= fn and is called as the sum function of the series fn
n=0 n
X
• Following the previous point, let x ∈ A, the numerical series fn (x) converges in
n
+∞ +∞
!
X X
F . It is also immediate that fn (x) = fn (x).
n=0 n=0
X
• We call the rest function of order n of the convergent series of functions fn , the
n
function Rn defined by,
+∞
X
∀x ∈ A, Rn (x) = fk (x).
k=n+1

For all x ∈ A and all n ∈ N, we have S(x) = Sn (x) + Rn (x).

X
Definition 7.3.2. We say the series of functions fn converges uniformly on A if the
n
sequence of partial sums (Sn )n∈N converges uniformly on A.

X
Proposition 7.3.3. If the series of functions fn converges uniformly on A. Then,
n
X
1.) the series fn converges pointwisely on A;
n

2.) the sequence of functions (fn )n∈N converges uniformly on A to the zero function.
248 SEQUENCES AND SERIES OF FUNCTIONS

Proof. Left in Exercise 7.4.4.

X
Proposition 7.3.4. If the series of functions fn converges pointwisely on A. Then,
X n
fn converges uniformly on A if, and only if the sequence of rests (Rn )n∈N converges
n
uniformly to the null function.
X
Proof. Using the definition of uniform convergence, the series fn converges uniformly
n
on A, if and only if the sequence of partial sums (Sn )n∈N converges uniformly on A to the
sum function S. Equivalently speaking,

∥Rn ∥L∞ (A,F ) = ∥S − Sn ∥L∞ (A,F ) −→ 0.

n→+∞

Thus, the claim follows.

Example 7.3.5. Consider the sequence of functions (fn )n∈N defined by, ∀n ∈ N, fn :
R+ −→ R X
−nx
. Fixing at first x ∈ R∗+ , the numerical series fn (x) converges since
x 7−→ xe n
1 X
fn (x) = xe−nx = O( 2 ) and via the comparison test. In addition, the series fn (0) =
n n
X X
0 is convergent. So, the series of functions fn converges pointwisely on R+ . Moreover,
n n
the sequence of partial sums can be calculated as follows,
n n
X X x(1 − e−(n+1)x ) x
∀x ∈ R∗+ , Sn (x) = fk (x) = xe−kx = −x
−→ .
1−e n→+∞ 1 − e−x
k=0 k=0

Also,

xe−(n+1)x
∀x ∈ R∗+ , Rn (x) = S(x) − Sn (x) = .
1 − e−x
We have then in particular,

1 e−1 1
Rn ( )= −1/(n+1)
−→ .
n+1 (n + 1)(1 − e ) n→+∞ e
X
This relation shows that the series fn does not converge uniformly on R∗+ .
n

X
Definition 7.3.6. We say the series of functions fn converges uniformly on any
X n
compact in A if for any compact subset K ⊂ A, the series fn converges uniformly on
n
K.
7.3. SERIES OF FUNCTIONS 249

Example 7.3.7. We consider again the previous example. Let [a, b] ⊂ R∗+ be a compact
subset. We can obtain, for n large enough,

ae−(n+1)a
∀x ∈ [a, b], |Rn (x)| ≤ −→ 0.
1 − e−a n→+∞

X
Thus, the series fn converges uniformly on any compact in R∗+ .
n

Proposition 7.3.8. We assume that F is a Banach space. Then, the series of functions
X
fn converges uniformly on A if, and only if for all ε > 0, there exists N ∈ N such that
n
for all n, p ∈ N, n ≥ N implies that
n+p
X
∀x ∈ A, fk (x) ≤ ε.
k=n+1 F
X
Proof. From the definition of uniform convergence of series of functions, fn converges
n
uniformly on A if, and only if the sequence of partial sums (Sn )n∈N converges uniformly on
A. Since F is complete, it follows from Theorem 7.1.16 that (Sn )n∈N converges uniformly on
A if, and only if (Sn )n∈N is a uniform Cauchy sequence on A. Hence, the claim follows.

X
Definition 7.3.9. We say the series of functions fn converges absolutely on A if for
X n
all x ∈ A, the numerical series ∥fn (x)∥F converges.
n

X (−1)n
Example 7.3.10. The series of functions converges pointwisely on R∗+ and
nx
n≥1
absolutely on (1, +∞).

X
Remark 7.3.11. The absolute convergence of a series of functions fn is equivalent to
X n
the pointwise convergence of the series ∥fn ∥F where for all n ∈ N, ∥fn ∥F denotes the
n
function x 7→ ∥fn (x)∥F .

Proposition 7.3.12. If F is a Banach space, then the absolute convergence implies the
pointwise convergence of series of functions.
X
Proof. Let x ∈ A, from the completeness of F , the convergence of the series ∥fn (x)∥F
X X n
implies the convergence of the series fn (x). Thus, the series of functions fn converges
n n
pointwisely on A.
250 SEQUENCES AND SERIES OF FUNCTIONS
X
Definition 7.3.13. We say the series of functions fn converges normally on A if
n

• for all n ∈ N, fn is a bounded function, i.e., f ∈ L∞ (A, F );

X
• the numerical series ∥fn ∥L∞ (A,F ) converges.
n

X
Proposition 7.3.14. The series of functions fn converges normally on A if, and only
X n
if there exists a series of positive numbers αn such that
n

• for all n ∈ N and all x ∈ A, ∥fn (x)∥F ≤ αn ;

X
• αn converges.
n

Proof. Left in Exercise 7.4.5.

X
Theorem 7.3.15. If F is a Banach space and the series of functions fn converges
X n
normally on A, then fn converges absolutely and uniformly on A. Moreover, it holds,
n

+∞
X +∞
X
fn ≤ ∥fn ∥L∞ (A,F ) . (7.10)
n=0 L∞ (A,F ) n=0

X
Proof. Let fn be a series of functions which converges normally. Then, for all x ∈ A,
n X
∥fn (x)∥F ≤ ∥fn ∥L∞ (A,F ) and the numerical series ∥fn ∥L∞ (A,F ) converges. Then, from
X n X
the comparison test, the series ∥fn (x)∥F converges. Hence, fn converges absolutely
n n
on A. X
On the other hand, since ∥f ∥L∞ (A,F ) converges, its sequence of partial sums is there-
n
n+p
X
fore a Cauchy sequence. That is, for all p ∈ N∗ , ∥fk ∥L∞ (A,F ) −→ 0. We have thus,
n→+∞
k=n+1
for all x ∈ A,
n+p
X n+p
X n+p
X
∥Sn+p (x) − Sn (x)∥F = fk (x) ≤ ∥fk (x)∥F ≤ ∥fk ∥L∞ (A,F ) −→ 0.
n→+∞
k=n+1 F k=n+1 k=n+1

So, the sequence of partial sums (Sn )n∈N is a uniformly Cauchy sequence on A. From
Xthe
completeness of F and Theorem 7.1.16, (Sn )n∈N converges uniformly on A. Thus, fn
n
converges uniformly on A.
7.3. SERIES OF FUNCTIONS 251

We have, for all N ∈ N and x ∈ A,

N
X N
X N
X +∞
X
fn (x) ≤ ∥fn (x)∥F ≤ ∥fn ∥L∞ (A,F ) ≤ ∥fn ∥L∞ (A,F ) .
n=0 F n=0 n=0 n=0
X
Since fn converges uniformly, we have thus,
n

+∞
X +∞
X
∀x ∈ A, fn (x) ≤ ∥fn ∥L∞ (A,F ) .
n=0 F n=0

Hence, (7.10) follows.

Remark 7.3.16. Theorem 7.3.15 is equivalent to say any absolutely convergent series in the
space (L∞ (A, F ), ∥ · ∥L∞ (A,F ) ) converges. This is a consequence of the completeness result
Theorem 7.1.21.
X
Example 7.3.17. 1.) The series xe−nx converges normally on any interval [a, +∞)
n
where a > 0. Since for n ∈ N being large enough, the function x 7→ xe−nx is decreasing
+∞), which implies that ∀x ∈ [a, +∞), |xe−nx | ≤ ae−na . And it is immediate
on [a,X
that ae−na converges.
n
X
2.) The series an cos(bn x) with 0 < a < 1 and b ∈ R is normally convergent on R.
n
X
3.) Consider the series of functions fn defined on [0, 1] by, ∀n ∈ N∗ ,
n
(
1
0 if x ∈ [0, n+1 ] ∪ [ n1 , 1],
fn (x) = 1 1
n if x ∈ ( n+1 , n1 ).

Let x ∈ [0, 1]. If x = 0, then fn (x) = 0 for all n ∈ N∗ . If x ∈ (0, 1], there exists Nx ∈ N
1
such that < x, we deduce then fn (x) = 0 for all n ≥ Nx . Hence, the series of
Nx
functions converges pointwisely
X on [0, 1]. Also, since
Xfn (x) are positive numbers, we
have then the series |fn (x)| converges. Thus, fn is absolutely convergent on
n n
[0, 1].
n+p
X
Next, for all n, p ∈ N∗ , we can calculate fk to deduce that for all x ∈ [0, 1],
k=n+1

n+p
X 1
fn (x) ≤ −→ 0.
n + 1 n→+∞
k=n+1
X
From the completeness of R and Proposition 7.3.8, the series of functions fn con-
n
verges uniformly on [0, 1].
252 SEQUENCES AND SERIES OF FUNCTIONS

1 X
Finally, since for all n ∈ N∗ , ∥fn ∥L∞ ([0,1],R) = , the series ∥fn ∥L∞ ([0,1],R) is
n n
X
divergent. fn is not normally convergent on [0, 1].
n
X
In conclusion, the series of functions fn converges on [0, 1] absolutely and uniformly
n
but not normally.
X
Proposition 7.3.18. Let (−1)n gn be a series of functions. We suppose that
n
• for all n ∈ N, gn : A → R is a positive function;
• for all x ∈ A, the sequence (gn (x))n∈N is decreasing.
• the sequence of functions (gn )n∈N converges uniformly to zero function.
X
Then, the series (−1)n gn converges uniformly on A.
n

Proof. From the assumption, we remark that for all x ∈ A, the sequence (gn (x))n∈N is
aXpositive, decreasing and tends to 0. From X the theorem on alternate series, the series
(−1)n gn (x) converges. Hence, the series (−1)n gn converges pointwisely. Denoting by
n n
S the sum function, it follows also from the theorem of alternate series that for all x ∈ A,
n
X
S(x) − (−1)k gk (x) ≤ |gn+1 (x)| ≤ ∥gn+1 ∥L∞ (A,F ) −→ 0.
n→+∞
k=0
X
Thus, the series (−1)n gn converges uniformly on A.
n

7.3.2 Properties on Convergent Series of Functions

Theorem 7.3.19. Let a ∈ A, we suppose that
• for all n ∈ N, the function fn is continuous at a;
X
• the series of functions fn converges uniformly on any compact in A (or simply
n
uniform on A).
Then, the sum function S is continuous at a. Moreover, if the functions fn are continuous
on A, then S is continuous on A.
Proof. It is sufficient to consider the sequence of partial sums and apply the results on
uniform convergence of sequences of functions.

X
Example 7.3.20. 1.) Consider the series of functions fn defined by, ∀n ∈ N, fn :
n
einx X
x 7→ 2 . Then, fn converges normally on R. Since ∀n ∈ N, fn is continuous,
n n
+∞ inx
X e
then the sum function S : x 7→ is continuous on R.
n2
n=0
7.3. SERIES OF FUNCTIONS 253

X 1
2.) Consider the series of functions defined on the complex open half-plane D :=
nz
n≥1
1 1
{z ∈ C | Re(z) > 1}. Then, for all z ∈ D and all n ∈ N∗ , z = Re(z) and the series
n n
X 1 X 1
is convergent. So, the series of functions converges absolutely and
n Re(z) nz
n≥1 n≥1
hence pointwisely on D. Let K ⊂ D be a compact subset, then there exists a > 1
1 1 X 1
such that ∀z ∈ K, Re(z) ≥ a. From the estimation z ≤ a , we obtain that
n n nz
n≥1
X 1
is uniformly convergent on K. Thus, converges uniformly on any compact in
nz
n≥1
+∞
1 X 1
D. Since the functions z 7→ z is continuous on D, then the sum function z 7→
n nz
n=1
is continuous on D via Theorem 7.3.19.
We call the Riemann’s Zeta Function the sum function of this series and we denote
+∞
X 1
ζ(z) := .
nz
n=1

X (−1)n
3.) Applying Proposition 7.3.18, the series of functions converges uniformly on
nx
n≥1
(−1)n
any compact in R∗+ . From the continuity of each functions x 7→ and Theo-
nx
+∞
X (−1)n
rem 7.3.19, the function x 7→ is continuous on R∗+ .
nx
n=1

Corollary 7.3.21. If E be a Banach algebra, of which the identity element is denoted by

Id, then the function x 7→ (Id − x)−1 is continuous on the unit open ball in E. Also, the
function x 7→ exp(x) is continuous on E.

Proof. We have seen the the chapter of Topology that for all x ∈ B(0E , 1), (Id − x) is
+∞
X
invertible and its inverse can be expressed as the sum of Neumann series. That is, xn =
X n=0
(Id − x)−1 . We then consider the series of functions xn . For any 0 < r < 1, we have,
n X
from the algebraic norm, x ∈ B(0E , r) ⇒ ∥xn ∥E ≤ ∥x∥nE ≤ rn . Thus, the series xn
n
converges normally, and hence uniformly, on the closed ball B(0E , r). By the continuity
+∞
X
of each functions x 7→ xn , we deduce then the function sum x 7→ xn = (Id − x)−1 is
[ n=0
continuous on B(0E , r). From the simple relation B E (0, r) = B(0E , 1), we obtain thus
0<r<1
the continuity of the function x 7→ (Id − x)−1 on B(0E , 1).
X xn
Consider the series of functions . Let x ∈ B(0E , r) for any r > 0, we have the
n
n!
254 SEQUENCES AND SERIES OF FUNCTIONS

following estimation

xn ∥x∥nE rn 1
∀n ∈ N, ≤ ≤ = O( ).
n! E n! n! n→+∞ n2
X xn
Hence, the series of functions converges absolutely on E and normally on B(0E , r).
n
n!
X
Since r > 0 can be chosen arbitrary, the series of functions fn converges normally, hence
n
xn
uniformly on any compact in E. Notice the continuities of each functions x 7→ , the sum
n!
+∞ n
X x
function exp : x 7→ is continuous on E.
n!
n=0

Theorem 7.3.22. Let a ∈ A, we suppose that

• for all n ∈ N, fn possesses a limit ln ∈ F at a;

X
• the series of functions fn converges uniformly on A.
n

Then,
X
1.) the series ln converges in F ;
n

+∞
X
2.) the sum function S = fn possesses a limit at a;
n=0

+∞
X
3.) lim S(x) = ln .
x→a
n=0

Equivalently speaking, we have

+∞  +∞ +∞
!
X  X X
lim fn (x) = ln = lim S(x) = lim fn (x) .
x→a x→a x→a
n=0 n=0 n=0

Proof. It is straightforward by using the sequence of partial sums and applying Theo-
rem 7.1.27.

X 1
Example 7.3.23. Consider the series of functions defined on (1, +∞). We have,
nx
n≥1
∗ 1 1 X1
for all n ∈ N , lim x = . And the series diverges. So, the contrapositive of
x→1+ n n n
n≥1
Theorem 7.3.22 shows that the series does not converge uniformly on (1, +∞).

Theorem 7.3.24. Let I ⊂ R be an interior-non-empty interval. Consider the sequence of

functions (fn )n∈N mapping from I into a Banach space F . We suppose that
7.4. EXERCISES 255

• for all n ∈ N, fn is continuously differentiable on I (i.e. fn ∈ C 1 (I, F ));

X
• fn converges pointwisely on I to S.
n
X
• the series of functions fn′ converges uniformly on any compact in I.
n

Then,
X
1.) in fact, fn converges uniformly on any compact in I to f ;
n

+∞
X
2.) S is continuously differentiable on I and S ′ = fn′ .
n=0

Proof. It is straightforward by using the sequence of partial sums and applying Theo-
rem 7.1.30.

Corollary 7.3.25. Let I ⊂ R be an interior-non-empty interval and let k ∈ N∗ ∪ {∞}.

Consider the sequence of functions (fn )n∈N mapping from I into a Banach space F . We
suppose that

• for all n ∈ N, fn is of C k -class on I;

X
• fn converges pointwisely on I to S.
n
X
• for all j ∈ [[1, k]] the series of the j-th derivatives fn(j) converges uniformly on any
n
compact in I to a function Sj .

Then,
X
1.) in fact, fn converges uniformly on any compact in I to f ;
n

+∞
X
(j)
2.) S is of C k -class on I and S = Sj = fn(j) for all j ∈ [[1, k]].
n=0

Proof. Left in Exercise 7.4.6.

Remark 7.3.26. We can also derive the version of Theorem 7.3.24 for the functions possessing
the partial derivatives. This generalization is also left to the readers.

7.4 Exercises
Exercise 7.4.1. Justify Proposition 7.1.7.

Exercise 7.4.2. Justify Corollary 7.1.31.

256 SEQUENCES AND SERIES OF FUNCTIONS

Exercise 7.4.3. Justify Proposition 7.2.9.

Exercise 7.4.4. Justify Proposition 7.3.3.

Exercise 7.4.5. Justify Proposition 7.3.14.

Exercise 7.4.6. Justify Corollary 7.3.25.

Exercise 7.4.7. Consider the sequence of functions (fn )n∈N given by,

1 2
 
∗
∀n ∈ N , fn (x) = x +
n

1.) Does the sequence of functions (fn )n∈N converge pointwisely on R? Justify your
answer and give its limit function.

2.) Same question for uniform convergence.

3.) Same question for uniform convergence on any compact.

1
Solution. 1.) Let x ∈ R be fixed. We consider the sequence (fn (x))n∈N . Since x+ −→
n n→+∞
7 y 2 , we have fn (x) −→ x2 .
x, it follows from the continuity of the function y →
n→+∞
Hence, the sequence of functions (fn )n∈N converges pointwisely to the function f :
x 7→ x2 .

2.) If the sequence of functions (fn )n∈N converges uniformly on R, then its uniform limit
function is the pointwise limit function. Let x ∈ R, n ∈ N∗ , we calculate |fn (x) −
1 2
   
2 1 1
f (x)| = x + −x = 2x + . Let n ∈ N∗ be fixed, the function x 7→
n n n
1 1
2x + is unbounded on R. Thus, the sequence of functions (fn )n∈N does not
n n
converges uniformly on R.

3.) Let K be a compact in R, that means K is closed and bounded. Then, there exists a
segment [a, b] such that K ⊂ [a, b]. For all n ∈ N∗ , we calculate sup |fn (x) − f (x)|.
x∈[x,b]
It follows that,

1 2
 
1 1
sup |fn (x) − f (x)| ≤ sup x+ − x2 = sup 2x +
x∈K x∈[a,b] n x∈[a,b] n n
 
1 1
≤ 2 max{|a|, |b|} + −→ 0
n n n→+∞

Thus, the sequence of functions (fn )n∈N converges uniformly on K. Since K is chosen
arbitrary on R, we conclude that (fn )n∈N converges uniformly on any compact in R.
7.4. EXERCISES 257

Exercise 7.4.8. Consider the sequence of functions (fn )n∈N defined on R by

∀n ∈ N, fn (x) = sin x cosn x.

1.) Show that (fn )n∈N converges pointwisely on R and give its limit function f .

2.) Does the sequence (fn )n∈N converges uniformly on R? Justify your answer.

Exercise 7.4.9. Let (fn )n∈N be a sequence of functions defined from [a, b] ⊂ R into C where
[a, b] is a segment. We assume that (fn )n∈N converges pointwisely to f on [a, b]. Moreover,
we assume that for all n ∈ N, fn is K-Lipschitz continuous on [a, b] where K > 0.

1.) Show that f is K-Lipschitz continuous on [a, b].

2.) State the negation of uniform convergence using subsequences.

3.) Show that (fn )n∈N converges uniformly to f on [a, b].

Exercise 7.4.10. Let (fn )n∈N be a sequence of polynomial functions defined from R into
R. We assume that (fn )n∈N converges uniformly to a function f on R. Show that f is a
polynomial function.

Exercise 7.4.11. Consider the sequence of functions (fn )n∈N defined by,

x2n
∀n ∈ N, x ∈ R, fn (x) = .
1 + x2n

1.) Show that (fn )n∈N converges pointwisely on R and give its limit function f .

2.) Does (fn )n∈N converge uniformly on R? Justify your answer.

3.) Determinate the intervals where (fn )n∈N converges uniformly on them.

Solution. 1.) We distinguish the cases where |x| < 1, |x| = 1 and |x| > 1.

x2n
• If |x| < 1, we have then x2n −→ 0. Hence, fn (x) = −→ 0.
n→+∞ 1 + x2n n→+∞
x2n 1 1
• If |x| = 1, we have fn (x) = 2n
= . Hence, fn (x) −→ .
1+x 2 n→+∞ 2

x2n 1
• If |x| > 1, then it follows that fn (x) = = −→ 1.
1+x 2n 1 + x−2n n→+∞
So, the sequence of functions (fn )n∈N converges pointwisely to the function f given
by,

 0 if |x| < 1,

1
2 if |x| = 1,
f (x) =

1 if |x| > 1.

258 SEQUENCES AND SERIES OF FUNCTIONS

2.) We observe that the limit function f is not continuous on R. Also, the function fn
is continuous on R for each n ∈ N. If the sequence of functions (fn )n∈N converges
uniformly on R, it follows from the Theorem of continuity of uniformly convergent
sequence of functions, the limit function f will be continuous on R. Thus, by contra-
diction, the sequence of functions (fn )n∈N does not converge uniformly on R.
We can also estimate for example that
1 1
sup |fn (x) − f (x)| ≥ sup |fn (x) − f (x)| = sup 2n
= ↛ 0.
x∈R x∈(1,+∞) x∈(1,+∞) 1 + x 2 n→+∞

3.) Let a, b ∈ (−1, 1) with a < b, we have, for all n ∈ N and all x ∈ [a, b],

x2n max{|a|, |b|}2n

|fn (x) − f (x)| = |fn (x) − 0| = ≤ −→ 0.
1 + x2n 1 n→+∞

Thus, (fn )n∈N converges uniformly on [a, b].

Also, let c ∈ R, c > 1. We have, for all n ∈ N and all x ∈ [c, +∞),
1 1
|fn (x) − f (x)| = |fn (x) − 1| = 2n ≤ −→ 0.
1+x 1 + c2n n→+∞
Hence, (fn )n∈N converges uniformly on [c, +∞). From the symmetric arguments, the
sequence (fn )n∈N converges also uniformly on (−∞, −c].

Exercise 7.4.12 (Dinivii ’s Theorem). Consider a sequence of functions (fn )n∈N . We sup-
pose that

• the domain A is compact in the space (E, ∥ · ∥E );

• each function fn is continuous and real-valued, i.e., fn ∈ C(A, R);

• (fn )n∈N converges pointwisely to a continuous function f on A;

• the sequence (fn )n∈N is increasing, i.e.,

∀n ∈ N, ∀x ∈ A, fn (x) ≤ fn+1 (x);

Let ε > 0 be fixed. For all n ∈ N, we define a set Un as follows.

Un := {x ∈ A | |fn (x) − f (x)| < ε} .

1.) Show that each Un is an open set.

2.) Show that the sequence (Un )n∈N is increasing, i.e., ∀n ∈ N, Un ⊂ Un+1 .
[
3.) Show that A = Un .
n∈N

4.) Show that there exists N (ε) ∈ N such that A = UN (ε) .

vii
Uliss Dini (1845-1918), Italian mathematician and politician. He worked on real analysis.
7.4. EXERCISES 259

5.) Show that (fn )n∈N converges uniformly on A to f .

X
Exercise 7.4.13. Consider the series of functions gn defined by,
n

2x
∀n ∈ N∗ , x ∈ R, gn (x) = .
x2 + n2
X
1.) Show that gn converges pointwisely.
n
X
2.) Show that gn does not converge normally on R.
n

+∞
X
3.) Is the sum function g := gn continuous on R? Justify your answer.
n=1

Solution. 1.) Let x ∈ R and n ∈ N∗ , we have the estimation,

2x 2|x|
|gn (x)| = ≤ 2 .
x2 +n 2 n
X 1
Since the series 2
converges, it follows from the comparison test that the numer-
n
X n X
ical series gn (x) converges for each x ∈ R. Thus, the series of functions gn
n n
converges pointwisely on R.

2.) We determinate the extreme values of each function gn on R. Let n ∈ N∗ ,

2(x2 + n2 ) − (2x)2 2n2 − 2x2

gn′ (x) = = .
(x2 + n2 )2 (x2 + n2 )2

The resolution of gn′ (xm ) = 0 gives us xm = ±n. Hence, the function gn achieve its
maximum at xm = n and we have,
1
∥gn ∥L∞ (R,R) = sup |gn (x)| = |gn (±n)| = .
x∈R n
X1 X
Since the series does not converges, the series gn does not converges normally
n
n n
on R.

3.) Let [a, b] ⊂ R be a segment, we have the following estimation. For all n ∈ N,

2x 2 max{|a|, |b|}
sup |gn (x)| = sup ≤ .
x∈[a,b] x∈[a,b] x2 +n 2 n2
X 1
Since the series 2
converges, it follows from the comparison test that the numer-
n
X n X
ical series sup |gn (x)| converges. Thus, the series of functions gn converges
n x∈[a,b] n
260 SEQUENCES AND SERIES OF FUNCTIONS

normally
X on the segment [a, b] and hence, it converges uniformly. In other words,
gn converges uniformly on any compact in R. Applying the theorem of continuity
n
+∞
X
of uniformly convergent series of functions, the sum function g := gn is continuous
n=1
on R.

X
Exercise 7.4.14. Consider the series of functions fn defined by,
n

xn
∀n ∈ N∗ , x ∈ R, fn (x) = .
n
X
1.) Determinate the intervals on R where fn is normally convergent.
n

+∞
X
2.) Show that the sum function f := fn is of class C ∞ on I where I is an interval to
n=1
be determinate.

3.) Give the differential equation satisfied by f and resolve it.

Exercise 7.4.15. Let f : R 7→ F such that its second derivative f ′′ is bounded on R.

Consider the sequence of functions (gn )n∈N defined by,
   
∗ 1
∀n ∈ N , x ∈ R, gn (x) = n f x + − f (x) .
n

Show that (gn )n∈N converges uniformly to f ′ on R.

Exercise 7.4.16. Let [a, b] ⊂ R be a segment.

Z b
1.) For all n, m ∈ N, calculate xm cos(nx)dx.
a
Z b
2.) Let P be a polynomial function. Show that P (x) cos(nx)dx −→ 0.
a n→+∞

3.) Show that for any continuous function f defined on [a, b], we have
Z b
f (x) cos(nx)dx −→ 0.
a n→+∞

This property is called as the Riemann-Lebesgue Lemma.

Z b
4.) Let g be a continuous function on [a, b] and satisfying for all m ∈ N, xm g(x)dx = 0.
a
Show that g = 0 on [a, b].
7.4. EXERCISES 261

Exercise 7.4.17. Let φ be a function defined on R∗+ . We suppose that

• φ is not identically zero on R∗+ ;

• lim φ(x) = lim φ(x) = 0.

x→0+ x→+∞

1.) Consider the sequence of functions (fn )n∈N defined by

∀n ∈ N∗ , x ∈ R∗+ , fn (x) = φ(nx).

Show that the sequence of functions (fn )n∈N converges pointwisely and give the limit
function.

2.) Does (fn )n∈N converges uniformly on R∗+ ? Justify your answer.

3.) Consider the sequence of functions (gn )n∈N defined by

x
∀n ∈ N∗ , x ∈ R∗+ , gn (x) = φ( ).
n
Study the pointwise and uniform convergence of the sequence of functions (gn )n∈N .

4.) Show that the sequences of functions (fn )n∈N and (gn )n∈N converge uniformly on any
compact in R∗+ .

X
Exercise 7.4.18. Consider a series of functions fn given by,
n

∀n ∈ N, fn = 1[n,n+1) .
X
1.) Show that fn converges pointwisely on R+ and give its sum function.
n

2.) Study the normal convergence. Justify your answer.

3.) Study the uniform convergence on R+ . Justify your answer.

4.) Study the uniform convergence on any compact in R+ . Justify your answer.
+∞
X +∞
X
5.) Does the relation lim fn (x) = lim fn (x) holds? Justify your answer.
x→+∞ x→+∞
n=0 n=0

Exercise 7.4.19. Consider the following function f given by,

+∞
X 1
∀x ∈ R∗+ , f (x) = .
1 + n2 x
n=1

1.) Justify that f is well-defined.

2.) Show that the corresponding series of functions converges normally on [a, +∞) for all
a > 0.

3.) Do we have the uniform convergence on R∗+ ? Justify your answer.

262 SEQUENCES AND SERIES OF FUNCTIONS

4.) Show that f is indefinitely differentiable on R∗+ and give f (k) with k ∈ N.

5.) Existence and the value of lim f (x).

x→+∞

6.) Answer the questions 1.) and 2.) for the function g given by,
+∞  
X 1 1
∀x ∈ R∗+ , g(x) = − .
n2 x 1 + n2 x
n=1

π2 1
7.) Show that f (x) ∼ .
x→+∞ 6 x
π 1
8.) Using Riemann sums (rectangle method), show that f (x) ∼ √ .
x→0+ 2 x

X
Exercise 7.4.20. Consider the series of functions fn defined by,
n

e−nx
∀n ∈ N, x ∈ R+ , fn (x) = .
1 + n2
X
1.) Justify the pointwise and uniform convergence of the series of functions fn .
n

+∞
X
2.) Show that the sum function S := fn is of class C 2 on R∗+ .
n=0

3.) Deduce that S satisfies the differential equation

1
∀x ∈ R∗+ , S ′′ (x) + S(x) = .
ex −1

Cauchy-Lipschitz Theorem
In this subsection, we are about to study the existence and uniqueness of the following first
order differential equation
dx
= f (t, x),
dt
combined with the initial condition

x(t0 ) = x0 .

Let I ⊂ R be an interior non empty interval and t0 ∈ I, we say a function φ : I → R is a

solution to the above initial value problem if it satisfies

 dφ = f (t, φ(t)) ∀t ∈ I,
dt (IVP)
 φ(t ) = x .
0 0

The main objective is to prove the following result.

7.4. EXERCISES 263

Theorem 7.4.21 (Local Cauchy-Lipschitzviii ). Let f be a function satisfying

• f is a continuous function defined on an open subset D ⊂ R × R;

• f is (uniformly) Lipschitz continuous respect to the variable x, i.e., there exists L > 0
such that

∀(t, x), (t, x′ ) ∈ D, |f (t, x) − f (t, x′ )| ≤ L|x − x′ |.

Then, for any initial condition (t0 , x0 ) ∈ D, there exists α > 0 and a unique solution to the
initial value problem (IVP) on the interval [t0 − α, t0 + α].

Exercise 7.4.22. Let (t0 , x0 ) ∈ D.

1.) Show that there exists strictly positive numbers rt and rx such that

S := [t0 − rt , t0 + rt ] × [x0 − rx , x0 + rx ] ⊂ D.

2.) Deduce that f is bounded over S.

Let I be an interval. We define the integral operator T : ψ 7→ T (ψ) for any continuous
function ψ : I → R as follows,

I −→ R
T (ψ) :
Z t .
t 7−→ x0 + f (s, ψ(s))ds
t0

We remark that the above integral might not be well-defined since the data point (s, ψ(s))
could lay outside of D.

Exercise 7.4.23. Let h > 0, we denote Ih := [t0 − h, t0 + h] and introduce the space of
functions

Fh := {ψ ∈ C(Ih , R) | ∀s ∈ Ih , (s, ψ(s)) ∈ S} .

Determinate a value of h in terms of rx , rt or ∥f ∥L∞ (S,R) such that for all ψ ∈ Fh , we have
T (ψ) ∈ Fh .

Exercise 7.4.24. Now, we consider a function φ0 ∈ Fα with 0 < α < h. We define the
Picardix ’s iterations (φn )n∈N given by the recursive relation
Z t
∀n ∈ N, φn+1 = T (φn ) : t 7→ x0 + f (s, φn (s))ds.
t0

1.) Give an example of φ0 .

viii
This theorem has various names. It is also called as Picard-Lindelöf Theroem, Picard’s existence Theo-
rem or Existence and uniqueness Theorem.
ix
Charles Émile Picard (1856-1941), French mathematician. One of the most influential mathematician
(after H. Poincaré) in his era. He contributed on differential equations, complex analysis, algebraic curves,
algebraic topology, etc.
264 SEQUENCES AND SERIES OF FUNCTIONS

2.) Justify that (φn )n∈N is well-defined.

3.) Show that

∀n ∈ N∗ , ∥φn+1 − φn ∥L∞ (Iα ,R) ≤ αL∥φn − φn−1 ∥L∞ (Iα ,R) .

4.) Deduce a value of α such that (φn )n∈N converges uniformly on Iα .

Exercise 7.4.25. Let (gn )n∈N : I → R be a sequence of continuous functions which con-
verges uniformly to a function g. Show that, for all a, b ∈ I, we have
Z b Z b
g(s)ds = lim gn (s)ds.
a n→+∞ a

Exercise 7.4.26. We denote by φ the uniform limit of the Picard iteration (φn )n∈N .

1.) Show that φ satisfies the integral equation,

Z t
∀t ∈ Iα , φ(t) = x0 + f (s, φ(s))ds. (intEq)
t0

2.) Show that the integral equation (intEq) possesses at most one solution in Fα .

3.) Prove Theorem 7.4.21.

Exponential and Logarithmic of Matrices

Let n ∈ N∗ , we denote by Mn (C) the space of square matrices of order n with complex
coefficients. We denote by ∥ · ∥ the usual norm on the space Cn .

Exercise 7.4.27. It is known that the following function ∥ · ∥M defines a norm on Mn (C).

∥Ax∥
∀A ∈ Mn (C), ∥A∥M := sup .
x∈Cn \{0} ∥x∥

1.) Show that for all A ∈ Mn (C),

∥A∥M = sup ∥Ax∥ = inf {T > 0 | ∀x ∈ Cn , ∥Ax∥ ≤ T ∥x∥} .

x∈Cn , ∥x∥=1

2.) Show that the norm ∥ · ∥M is algebraic.

Exercise 7.4.28. Let A ∈ Mn (C), we introduce the following function.

+∞ k k
X A t
∀t ∈ R, exp(At) = In + .
k!
k=1

1.) Justify this function is well-defined on R.

7.4. EXERCISES 265

2.) Show that this function is indefinitely differentiable on R and give its first derivative.

3.) Let x0 ∈ Cn . Resolve the following initial value problem


 dx = Ax,
dt
 x(0) = x .
0

Exercise 7.4.29. Let A, B ∈ Mn (C). We assume that A and B commute between them,
i.e., AB = BA.
1.) Show that

∀t ∈ R, exp(At)B = B exp(At).

2.) Let x0 ∈ Cn . Show that the function t 7→ exp(At) exp(Bt)x0 satisfies the initial value
problem

 dx = (A + B)x,
dt
 x(0) = x .
0

3.) Deduce that exp(A + B) = exp(A) exp(B) = exp(B) exp(A).

4.) Does the previous identity hold for A and B being not commuting?

5.) Show that exp(A) ∈ GL(n) and exp(A)−1 = exp(−A).

6.) Consider now a function g : I → Mn (C) where I ⊂ R is an open interval. We suppose

that g is continuously differentiable and for all t ∈ I, g(t)g ′ (t) = g ′ (t)g(t). Show that
d
∀t ∈ I, exp(g(t)) = exp(g(t))g ′ (t) = g ′ (t) exp(g(t)).
dt

Exercise 7.4.30. Let A ∈ Mn (C). We define

+∞
X (−1)k−1
∀t ∈ I ⊂ R, log(In + At) = Ak tk .
k
k=1

1.) Justify that this function is well-defined on I = (−∥A∥−1 −1

M , ∥A∥M ).

2.) Justify that for all t ∈ I, (In + At) is invertible and (In + At)−1 commute with
log(In + At).

3.) Justify and calculate the first derivative of t 7→ log(In + At).

4.) Define now a function J : t 7→ exp(− log(In + At))(In + At). Show that J is a constant
function on I.

5.) Show that if ∥A∥M < 1, then exp(log(In + A)) = In + A.

266 SEQUENCES AND SERIES OF FUNCTIONS

Exercise 7.4.31. A matrix N ∈ Mn (C) is called nilpotent if there exists p ∈ N such that
N p = 0.

1.) Assuming A is a nilpotent matrix, show that the results in Exercise 7.4.30 hold with
I = R.

2.) Assuming A = λIn + N with λ ∈ C \ {0} and N nilpotent, show that

 
1
A = exp µIn + log(In + N ) ,
λ

where µ ∈ C such that eµ = λ.

3.) Using the arguments on Jordan forms, prove that for any invertible matrix A ∈ GL(n),
there exists B ∈ Mn (C) such that exp(B) = A.

7.A Ascoli-Arzelà Theorem (Out of Program)

Theorem 7.A.1 (Ascolix -Arzelàxi ). Let K be a compact subset in a metric space (X, d)
and S be a bounded set in the space C(K, C) endowed with the uniform convergence norm,
i.e., ∥ · ∥L∞ (K) . We suppose that the functions in S are equicontinuous, that means

∀ε > 0, ∃δ > 0 s.t. ∀f ∈ S, d(x, y) ≤ δ ⇒ |f (x) − f (y)| ≤ ε.

Then, S is relatively compactxii in C(K).

The simplest version of the Ascoli-Arzelà Theorem 7.A.1 can be formulated as follows
in the case where K = [a, b] being a segment.

Corollary 7.A.2. Let (fn )n∈N be a sequence of continuous functions mapping from the
segment [a, b] ⊂ R into C. We suppose that

• (fn )n∈N is uniformly bounded;

• (fn )n∈N is equicontinuous.

Then, (fn )n∈N has a subsequence which converges uniformly on [a, b].

To prove the famous Ascoli-Arzelà Theorem 7.A.1, we begin with the following lemma
on compact sets.

Lemma 7.A.3. Let K be a compact set in a metric space (X, d). Then, there exists a
countable and dense subset A = {ak | k ∈ N} contained in K. Moreover, for all ε > 0,
there exists Mε ∈ N such that for all x ∈ K, there exists k ∈ N with k ≤ Mε such that
d(x, ak ) ≤ ε.
x
Giulio Ascoli (1843-1896), Italian mathematician. He worked on thoery of functions and Fourier series.
xi
Cesare Arzelà (1847-1912), Italian mathematician. Professor at University of Bolongna. He worked on
functional analysis and theory of functions.
xii
We say a set A is relatively compact if its closure A is compact. In finite dimensional spaces, the
relative compactness is equivalent to the boundedness.
7.A. ASCOLI-ARZELÀ THEOREM (OUT OF PROGRAM) 267

Proof. Let (rn )n∈N be a sequence of strictly positive real numbers which converges to 0.
Typically, we may assume that rn = 2−n for all n ∈ N. Let n ∈ N be fixed. We consider
the following open cover of the set K.
[
K⊂ B(x, rn ),
x∈K

where B(x, r) denotes the open ball centered at x with the radius r in the metric space K.
From the compactness of K, this open cover has a finite subcover. So, there exists mn ∈ N
m
[n
and the points x(1, n), · · · , x(mn , n) ∈ K such that K ⊂ B(x(j, n), rn ).
j=1
We then define A := {x(j, n) | j ∈ [[1, mn ]], n ∈ N}. Since mn is a finite number for all
n ∈ N, we can easily deduce that the points x(j, n) in A can be indexed by N2 , which is also
a countable set. So, we shall denote from now on the set A as a sequence A = {ak | k ∈ N}.
On the other hand, let ε > 0. It follows from the convergence rn −→ 0 that there
n→+∞
m
[ N

exists N ∈ N such that 0 < rn < ε for all n ≥ N . Using the relation K ⊂ B(x(j, N ), rN ),
j=1
for any point x ∈ K, we can find a point ak := x(j, N ) ∈ A such that x ∈ B(ak , rN ). This
is equivalently to say that the distance between x and ak is smaller then ε. We remark also
XN
that the index k can be chosen such that k ≤ mn := Mε and the number Mε depends
n=0
only on ε. In conclusion, A is dense in K and the additional property is satisfied.

Now, we prove the main theorem.

Proof of Ascoli-Arzelà Theorem 7.A.1. We will prove the equivalent statement of the the-
orem. For any sequence of functions (fn )n∈N in S, there exists a subsequence (fφ(n) )n∈N
which converges uniformly in C(K). Here φ denotes an extraction, that means a strictly
increasing function mapping from N into N.
Since K is compact, we apply Lemma 7.A.3 to introduce a countable and dense subset
A := {ak | k ∈ N} where (ak )k∈N denotes a sequence of elements in K. Next, we proceed the
famous Cantor’s diagonal process. We start with considering the sequence (fn (a0 ))n∈N
of complex numbers. Since S is bounded in (C(K), ∥ · ∥L∞ (K,C) ), there exists M > 0 such
that for all f ∈ S, ∥f ∥L∞ (K,C) ≤ M . We have thus immediately that

∀n ∈ N, |fn (a0 )| ≤ ∥fn ∥L∞ (K,C) ≤ M.

In other words, the sequence (fn (a0 ))n∈N is bounded in C. Using the completeness of C
(or the Bolzano-Weierstrass Theorem), (fn (a0 ))n∈N has a convergent subsequence. This is
equivalently to say that there exists an extraction φ0 : N → N being strictly increasing such
that fφ0 (n) (a0 ) −→ y0 ∈ C.
n→+∞
The next step of the process consists to repeat the similar extraction on the sequence
(fφ0 (n) (a1 ))n∈N . Using again the uniform boundedness of S and the completeness of C, we
can extract another convergent subsequence fφ0 (φ1 (n)) (a1 ) in C. Then, by repeating such
extractions, we can obtain a sequence of functions (φk )k∈N of which each maps from N into
N and is strictly increasing. Furthermore, we have also

∀k ∈ N, ∃yk ∈ C s.t. fφ0 ◦φ1 ◦···◦φk (n) (ak ) −→ yk .

n→+∞
268 SEQUENCES AND SERIES OF FUNCTIONS

Consider now the function φ : N → N defined by, ∀n ∈ N, φ(n) = φ0 ◦ φ1 ◦ · · · ◦ φn (n). Since

each φk is strictly increasing function, we have that ∀n, k ∈ N, φk (n) > n. So, it holds that

∀n ∈ N, φ(n) = φ0 ◦ φ1 ◦ · · · ◦ φn (n) < φ0 ◦ φ1 ◦ · · · ◦ φn (n + 1)

< φ0 ◦ φ1 ◦ · · · ◦ φn (φn+1 (n + 1)) = φ0 ◦ φ1 ◦ · · · ◦ φn+1 (n + 1)
= φ(n + 1).

Hence, (fφ(n) )n∈N is a subsequence of the sequence (fn )n∈N . From the construction, for all
k ∈ N, the sequence (fφ(n) )n≥k is a subsequence of (fφ0 ◦φ1 ◦···◦φk (n) )n≥k , which is convergent
in C. Thus, we have,

∀k ∈ N, fφ(n) (ak ) −→ yk ∈ C.
n→+∞

This is exactly to say the sequence of functions (fφ(n) )n∈N converges pointwisely on the
countable set A.
Now we will prove the sequence (fφ(n) )n∈N is a Cauchy sequence in (C(K, C), ∥·∥L∞ (K,C) ).
Let ε > 0. From the equicontinuity of S, there exists δ(ε) > 0 such that for all k ∈ N and
ε
all x, y ∈ K, d(x, y) ≤ δ(ε) ⇒ |fφ(k) (x) − fφ(k) (y)| ≤ . Consider now x ∈ K, and we use
3
the denseness result Lemma 7.A.3 of A in K. There exists ak ∈ A such that d(ak , x) ≤ δ(ε).
In parallel, the index k can be chosen k ≤ Mδ(ε) ∈ N which depends only on ε. In addition,
we have seen that the sequence of functions (fφ(n) )n∈N converges pointwisely on A. It leads
to that the sequence (fφ(n) (ak ))n∈N is a Cauchy sequence in C. So, there exists Nε ∈ N and
ε
it holds that |fφ(n) (a) − fφ(n+p) (a)| ≤ for all n, p ∈ N with n ≥ Nε . We remark that the
3
number Nε depends on the point ak ∈ A, which depends a priori on the considered point x.
However, since we can always choose k ≤ Mδ(ε) ∈ N, the convergence shown in the diagonal
process depends in fact only on ε. Hence, for all n, p ∈ N with n ≥ Nε ,

|fφ(n) (x) − fφ(n+p) (x)|

≤ |fφ(n) (x) − fφ(n) (a)| + |fφ(n) (a) − fφ(n+p) (a)| + |fφ(n+p) (x) − fφ(n+p) (a)|
ε ε ε
≤ + + = ε.
3 3 3
Since x ∈ K is chosen arbitrary, we can conclude that ∥fφ(n) −fφ(n+p) ∥L∞ (K,C) ≤ ε if n ≥ Nε .
And this is exactly to say that (fφ(n) )n∈N is a Cauchy sequence in (C(K, C), ∥ · ∥L∞ (K,C) ).
We have seen in Proposition 7.1.26 that the space (C(K, C), ∥ · ∥L∞ (K,C) ) is complete. Thus,
the sequence (fφ(n) )n∈N converges in this space. Hence, the claim follows.
 Chapter 8

Riemann Integration

Let I ⊂ R be an interval of real numbers and F be a Banach space. In this chapter, we will
consider the functions mapping from I into F .

8.1 Integrals on a Segment

In this section, [a, b] ⊂ R denotes a segment in R with a < b. We will only discuss in this
section the scenarios where I = [a, b].

8.1.1 Riemann and Darboux Integrability

To define rigorously the integrals, we begin with the simplest case. In this subsection, we
consider only the real-valued functions, i.e., F = R.

Definition 8.1.1. Let φ be a step function defined on [a, b] and (x0 , x1 , · · · , xn ) be a

partition adapted to φ. For all i ∈ [[1, n]], we denote by ci ∈ R the value of φ on the
subinterval (xi−1 , xi ). We call the integral of φ on [a, b] the following value,
Z n
X
φ= (xi − xi−1 )ci ∈ R. (8.1)
[a,b] i=1

Proposition 8.1.2. The integrals of step functions satisfy the following properties.

1.) Let φ be a step function, its integral over [a, b] is independent to the adapted partition.
Z
2.) The integral φ 7→ φ is a linear application on the space of step functions.
[a,b]

Z Z
3.) φ ≤ |φ|.
[a,b] [a,b]

Proof. 1.) We consider at first the case with another partition which is finer and contains
exactly one element then the adapted partition P := (x0 , x1 , · · · , xn ). That means, we
consider the partition of type P ′ := (x0 , x1 , · · · , xp , x′ , xp+1 , · · · , xn ) where p ∈ [[0, n]].
We also denote by Int(φ, P ) the value calculated from the right-hand side of (8.1)
under the partition P . And Int(φ, P ′ ) is defined accordingly.

269
270 RIEMANN INTEGRATION

Since φ is a step function, it is constant on all subintervals (xi−1 , xi ) for all i ∈ [[1, n]].
We have in particular that for all x ∈ (xp , xp+1 ), φ(x) = cp . Hence,

p−1
X n
X
′ ′ ′
Int(φ, P ) = (xi − xi−1 )ci + (x − xp )cp + (xp+1 − x )cp + (xi − xi−1 )ci
i=1 i=p+1
p−1
X n
X
= (xi − xi−1 )ci + (xp+1 − xp )cp + (xi − xi−1 )ci
i=1 i=p+1
Xn
= (xi − xi−1 )ci = Int(φ, P ).
i=1

Hence, the claim follows in this case. We can thus proceed a mathematical induction
on the numbers of extra points in the refinement partition to deduce that the claim
follows for any refinement partition.
Next, let P and P ′ be adapted partitions to the step function φ. It follows from
Proposition 7.2.4 there exists a common refinement P ∨P ′ on [a, b]. From the previous
result, we have thus

Int(φ, P ) = Int(φ, P ∨ P ′ ) = Int(φ, P ′ ).

Hence, the claim follows.

2.) Let φ and ψ be step functions on [a, b] and α, β ∈ R. Let (x0 , x1 , · · · , xn ) be an

adapted partition of [a, b] common to f and to g (from the previous reasoning, it is
licit). Let i ∈ [[1, n]], we denote respectively by ci and di the values of the functions φ
and ψ on the subinterval (xi−1 , xi ). Then, it follows,
Z n
X
αφ + βψ = (xi − xi−1 )(αci + βdi )
[a,b] i=1
Xn n
X Z Z
=α (xi − xi−1 )ci + β (xi − xi−1 )di = α φ+β ψ.
i=1 i=1 [a,b] [a,b]

Thus, the integral is a linear application.

3.) From the definition and triangular inequality, we have

Z n
X n
X Z
φ = (xi − xi−1 )ci ≤ (xi − xi−1 )|ci | = |φ|. (8.2)
[a,b] i=1 i=1 [a,b]

Proposition 8.1.3. Let φ and ψ beZ step functions

Z on [a, b]. We assume that φ = ψ on
i
[a, b] except for finite points . Then, φ= ψ.
[a,b] [a,b]
i
That means for all x ∈ [a, b] \ S, φ(x) = ψ(x) where S denotes a finite set.
8.1. INTEGRALS ON A SEGMENT 271

Proof. We denote by (y1 , · · · , ym ) the ordered points in [a, b] where φ ̸= ψ. From Proposi-
tion 7.2.9, the function φ − ψ is also a step function on [a, b]. In addition, φ − ψ equal to
zero on [a, b] \ (y1 , · · · , ym ). Hence,
Z m−1
X
(φ − ψ) = 0(y1 − a) + 0(yi+1 − yi ) + 0(b − ym ) = 0.
[a,b] i=1

Ans the claim follows using the linearity of the integral map Proposition 8.1.2.

Proposition 8.1.4. Let φ, ψ be step functions.

Z
1.) We assume that φ ≥ 0 except for finite points, then φ ≥ 0.
[a,b]
Z Z
2.) We assume that φ ≤ ψ except for finite points, then φ≤ ψ.
[a,b] [a,b]

Proof. 1.) Let φ be a positive step function on [a, b] except for finite points. Consider
each subinterval (xi−1 , xi ), since it is an open non empty interval in real numbers, it
contains therefore an infinity numbers of points. According to the assumption, the
coefficient ci of φ on (xi−1 , xi ) is positive. We have thus,
Z n
X
φ= (xi − xi−1 )ci ≥ 0.
[a,b] i=1

2.) Let φ and ψ be step function satisfying φ ≤ ψ except for finite points. Following
Proposition 7.2.9, the function ψ − φ is a positive step function except for finite
points. Applying the linearity of the integral map Proposition 8.1.2 and the previous
result, we have
Z Z Z
(ψ − φ) ≥ 0 ⇔ φ≤ ψ.
[a,b] [a,b] [a,b]

Proposition 8.1.5. Let c ∈ (a, b) and φ be a function defined on [a, b]. Then, φ is a step
function on [a, b] if, and only if the restrictions φ|[a,c] and φ|[c,b] are both step functions.
Furthermore, we have,
Z Z Z
φ= φ|[a,c] + φ|[c,b] .
[a,b] [a,c] [c,b]

Proof. ⇒: We assume that φ is a step function on [a, b]. Consider an adapted par-
tition P = (x0 , x1 , · · · , xn ) of the function φ. We may assume, without of los-
ing the generality that c ∈ (xj−1 , xj ) for certain j ∈ [[1, n]] (the cases where c ∈
{x1 , x2 , · · · , xn−1 } are trivial). We can thus construct a refinement of P given by,
P ′ = (x0 , x1 , · · · , xj−1 , c, xj , · · · , xn ). Then, P ′ is also an adapted partition of φ.
Since φ is constant on each intervals (xi−1 , xi ) for all i ∈ [[1, n]]. We have in particular
that φ is constant on each intervals (xi−1 , xi ) for i ∈ [[1, j − 1]] as well as on (xj−1 , c).
272 RIEMANN INTEGRATION

Thus, φ is constant on each subintervals of the adapted partition (x0 , x1 , · · · , xj−1 , c)

of [a, c]. Hence, φ is a step function on [a, c].
From the similar reason on [c, b], we deduce that φ is also a step function on [c, b]. It
follows from (8.1) that
Z Xn
φ= (xi − xi−1 )ci
[a,b] i=1
j−1
X n−1
X
= (xi − xi−1 )ci + (c − xj−1 )cj + (xj − c)cj + (xi − xi−1 )ci
i=1 i=j+1
Z Z
= φ|[a,c] + φ|[c,b]
[a,c] [c,b]

⇐: We assume now φ|[a,c] and φ|[c,b] are both step functions. Let (x0 , x1 , · · · , xn ) and
(y0 , y1 , · · · , ym ) be adapted partitions respectively to φ|[a,c] and to φ|[c,b] . Then, the
finite sequence P = (x0 , x1 , · · · , xn = c = y0 , y1 , · · · , ym ) is therefore a partition of
[a, b]. We have also, the function φ is constant on each subintervals of P . Hence, φ is
a step function on [a, b].

Definition 8.1.6. We call a tagged partition (P, t) on the segment [a, b] any partition
P = (x0 , x1 , · · · , xn ) together with a finite sequence t = (t1 , t1 , · · · , tn ) satisfying

∀i ∈ [[1, n]], xi−1 ≤ ti ≤ xi .

Let f be a bounded function defined on [a, b]. We call the Riemann sum of the function
f on the tagged partition (P, t) the following real number
n
X
σ(f, P, t) := (xi − xi−1 )f (ti ).
i=1

Remark 8.1.7. As an immediate result, we can consider the following step function φ defined
by,
n
f (ti )1(xi−1 ,xi ) .
X
φ=
i=1
Z
And it follows that σ(f, P, t) = φ.
[a,b]

Definition 8.1.8. We say a bounded function f ∈ L∞ ([a, b], R) is Riemann integrable

on [a, b] if there exists a number If ∈ R such that for all ε > 0, there exists δ > 0 such that
for any tagged partition (P, t) with |P | ≤ δ, we have

|If − σ(f, P, t)| ≤ ε.

In the case,
Z we call the number If as the Riemann integral of f on [a, b] and denote it
by (R) f.
[a,b]
8.1. INTEGRALS ON A SEGMENT 273

We introduce as well as an equivalent definition of Riemann integrability, which is easier

to manipulate the calculus.

Definition 8.1.9. Let f ∈ L∞ ([a, b], R).

• We call and denote the lower Darbouxii sum the following supremum
Z (Z )
f := sup φ φ ∈ ST ([a, b], R), φ ≤ f .
[a,b] [a,b]

• We call and denote the upper Darboux sum the following infimum
Z (Z )
f := inf φ φ ∈ ST ([a, b], R), φ ≥ f .
[a,b] [a,b]

• We say f is Darboux integrable if the two bounds are equals. In the case, the equal
bounds are called as the Darboux integral of f on [a, b] and we denote it by
Z Z Z
(D) f= f= f.
[a,b] [a,b] [a,b]

Remark 8.1.10. Since f is a bounded function, it is immediate that ∀x ∈ [a, b], inf f ≤
[a,b]
f (x) ≤ sup f . From the positivity of integral Proposition 8.1.4, for any step function φ such
[a,b]
that φ ≤ f ≤ sup f , we have
[a,b]

Z Z
φ≤ sup f = (b − a) sup f.
[a,b] [a,b] [a,b] [a,b]

(Z )
This shows that the set φ φ ∈ ST ([a, b], R), φ ≤ f ⊂ R is bounded from above.
[a,b]
So it is licit to consider its supremum, i.e. the lower Darboux sum. We can follow the
analogue arguments to deduce the existence of the upper Darboux sum.
Moreover, we always have that
Z Z
f≤ f.
[a,b] [a,b]

Theorem 8.1.11. A bounded function f ∈ L∞ ([a,Zb], R) is Riemann

Z integrable if, and only
if it is Darboux integrable. Moreover, we have (R) f = (D) f.
[a,b] [a,b]
ii
Jean-Gaston Darboux (1842-1917), French mathematician. He worked on geometry and analysis. Many
famous mathematician were Phd students of Darboux, like Émile Borel, Élie Cartan, Émile Picard, Thomas
Stieljes.
274 RIEMANN INTEGRATION

Proof. ⇒: We assume now f is Riemann integrable. We will prove that

(Z )
If = sup φ φ ∈ ST ([a, b], R), φ ≤ f . (8.3)
[a,b]

Z
Let φ ∈ ST ([a, b], R) and φ ≤ f . We assume by contradictory that If < φ.
Z [a,b]

We denote then ε0 := φ − If > 0. Let Pφ = (x0 , x1 , · · · , xn ) be an adapted

[a,b]
partition of φ. We consider a tagged partition (P, t) such that P = (y0 , y1 , · · · , ym )
is a refinement of Pφ and that for all tagged points tj , tj ∈ [a, b] \ {x0 , x1 , · · · , xn }.
From the condition φ ≤ f , we deduce that for all tj , φ(tj ) ≤ f (tj ). Introducing a step
m
f (tj )1(yj−1 ,yj ) , we can obtain that φ ≤ ψ on [a, b]. Hence, it follows
X
function ψ :=
j=1
from Proposition 8.1.4 that
Z Z
σ(f, P, t) = ψ≥ φ = I f + ε0 .
[a,b] [a,b]

We remark that the above inequality holds no matter how fine the mesh |P | is. Thus,
this is contradictory toZ the definition of Riemann integrability of f . In the conse-
quences, we have If ≥ φ for all φ ∈ ST ([a, b], R) such that φ ≤ f .
[a,b]
On the other hand, let ε > 0, from the Riemann integrability of f , there exists δ > 0
ε
such that for any tagged partition (P, t) with |P | ≤ δ, we have |If −σ(f, P, t)| ≤ . Let
3
P = (x0 , x1 , · · · , xn ) be a partition with |P | ≤ δ. We consider two tags t = (t1 , · · · , tn )
and t′ = (t′1 , · · · , t′n ) on the partition P such that ∀i ∈ [[1, n]], f (t′i ) ≤ f (ti ). We have
then
n
X
(f (ti ) − f (t′i ))(xi − xi−1 ) = |σ(f, P, t) − σ(f, P, t′ )|
i=1
2ε
≤ |σ(f, P, t) − If | + |If − σ(f, P, t′ )| ≤ .
3
Consider now a step function φ defined by,

∀i ∈ [[1, n]], φ|(xi−1 ,xi ) = inf f (t); φ(xi ) = f (xi ).

t∈(xi−1 ,xi )

We have hence, φ ≤ f on [a, b] and

Z
σ(f, P, t) − φ
[a,b]
n
( n )
X X
= f (ti )(xi − xi−1 ) − inf f (t′i )(xi − xi−1 ) ∀i ∈ [[1, n]], t′i ∈ (xi−1 , xi )
i=1 i=1
n
( )
X 2ε
= sup (f (ti ) − f (t′i ))(xi − xi−1 ) ∀i ∈ [[1, n]], t′i ∈ (xi−1 , xi ) ≤
3
i=1
8.1. INTEGRALS ON A SEGMENT 275
Z
2ε ε 2ε
In the consequence, φ ≥ σ(f, P, t) −
≥ If − − = If − ε. From the char-
[a,b] 3 3 3
acterization of supremum, the relation (8.3) holds. We can use analogue arguments
to establish that
(Z )
If = inf φ φ ∈ ST ([a, b], R), φ ≥ f .
[a,b]

Z Z
In conclusion, f is Darboux integrable and (R) f = If = (D) f.
[a,b] [a,b]

⇐: We assume now f is Darboux integrable. We set

Z (Z )
If := (D) f = inf φ φ ∈ ST ([a, b], R), φ ≥ f
[a,b] [a,b]
(Z )
= sup φ φ ∈ ST ([a, b], R), φ ≤ f .
[a,b]

Since f is bounded, we set also

H := sup f (x) − inf f (x) ≥ 0.

x∈[a,b] x∈[a,b]

Let ε > 0, it follows from the Darboux integrability

Z of f , there exists stepZfunctions φ+
ε ε
and φ− such that φ+ ≥ f , φ− ≤ f and φ+ − If ≤ as well as If − φ− ≤ .
[a,b] 2 [a,b] 2
We denote by P+ and P− the adapted partitions respectively to φ+ and φ− . Consider
their common refinement P+ ∨ P− = (x0 , x1 , · · · , xN ) with N ∈ N∗ . We set

1 n ε o
δ= min x1 − x0 , · · · , xN − xN −1 , > 0.
2 HN
Let (P, t) be a tagged partition on [a, b] such that |P | ≤ δ. From the definition of
δ, for each subinterval [yj−1 , yj ] in the partition P = (y0 , y1 , · · · , ym ), we have two
possible cases,

1.) [yj−1 , yj ] ⊂ (xi−1 , xi ) for certain i ∈ [[1, N ]];

2.) there exists a unique i ∈ [[1, N ]] such that yj−1 ≤ xi ≤ yj .

We remark furthermore that the second case might appear at most N times for all
subintervals in P . We denote respectively by S1 and S2 the set of indexes j ∈ [[1, m]]
such that [yj−1 , yi ] satisfies the first case and the second case. As an immediate result,
S1 ∪ S2 = [[1, m]].
We next consider the tagged values f (tj ) for tj ∈ [yj−1 , yj ]. According to the above
two possible cases, we have the following results.

1.) If j ∈ S1 , we have ∀s, s′ ∈ [yj−1 , yi ], φ− (s) = φ− (s′ ) ≤ f (s′ ) ≤ φ+ (s′ ) = φ+ (s).

2.) If j ∈ S2 , the things become more complicated. Since inf f ≤ f ≤ φ+ , it
[a,b]
leads immediately to that sup f = inf f + H ≤ φ+ + H. We have, for all
[a,b] [a,b]
276 RIEMANN INTEGRATION

s, s′ ∈ [yj−1 , yj ], f (s′ ) ≤ sup f ≤ φ+ (s) + H. Also, the similar result holds for
[a,b]
φ− , we could deduce that,

∀s, s′ ∈ [yj−1 , yi ], φ− (s) − H ≤ f (s′ ) ≤ φ+ (s) + H.

1(yj−1 ,yj ) . Except for a finite points, we obtain

X
Introduce the step function ψ := H
j∈S2
the following relation,
m
f (tj )1(yj−1 ,yj ) ≤ φ+ + ψ.
X
φ− − ψ ≤
j=1

At the same time, we have also,

Z X ε
ψ=H (yj − yj−1 ) ≤ H|P |Card(S2 ) ≤ HN δ ≤ .
[a,b] 2
j∈S2

Applying Proposition 8.1.4, we have thus,

Z Z m Z
f (tj )1(yj−1 ,yj ) ≤
X
(φ− − ψ) ≤ (φ+ + ψ).
[a,b] [a,b] j=1 [a,b]

We can thus obtain using the linearity of the integral mapping,

Z Z Z
ε ε
(φ+ + ψ) = φ+ + ψ ≤ If + + = If + ε.
[a,b] [a,b] [a,b] 2 2
Z
Symmetrically, we have also (φ− − ψ) ≥ If − ε.
[a,b]
We finally conclude that
Z m
f (tj )1(yj−1 ,yj ) ≤ If + ε.
X
If − ε ≤ σ(f, P, t) =
[a,b] j=1

Hence, f is Riemann integrable and its Riemann integral equals to the Darboux
integral.

From now on, since the two definitions are equivalent, we have no necessity to distinguish
them. We say simply a function
Z f is integrable on [a, b] if it is Riemann or Darboux
integrable and we denote by f its Riemann or Darboux integral.
[a,b]

Proposition 8.1.12. Any piecewise continuous function is integrable.

Proof. We prove at first the statement for continuous functions. Let g be a continuous
function defined on [a, b]. Since [a, b] is a compact subset in R, it follows from Hein’s
Theoremiii that g is uniformly continuous on [a, b]. Let ε > 0, there exists δ > 0 such that
iii
See in the course of first semester.
8.1. INTEGRALS ON A SEGMENT 277

ε
for all x, y ∈ [a, b], |x − y| ≤ δ implies that |g(x) − g(y)| ≤ . We consider then the
b−a  
b−a
uniform partition P = (a, a + s, a + 2s, · · · , a + ns = b) where n = + 1 ∈ N∗ and
δ
b−a
s= . Introduce now the step functions φ+ and φ− given by,
n

 φ+ (x) =
 max g, if x ∈ [a + (i − 1)s, a + is) ∀i ∈ [[1, n]];
[a+(i−1)s,a+is]

 φ− (x) =
 min g, if x ∈ [a + (i − 1)s, a + is) ∀i ∈ [[1, n]].
[a+(i−1)s,a+is]

ε
We obtain thus φ− ≤ g ≤ φ+ , also that ∀x ∈ [a, b], φ+ (x) − φ− (x) ≤ . In the
b−a
consequences,
Z Z Z Z Z
ε
g− g≤ φ+ − φ− = (φ+ − φ− ) ≤ (b − a) = ε.
[a,b] [a,b] [a,b] [a,b] [a,b] b−a
Z Z
We conclude thus g= g, so g is integrable over [a, b].
[a,b] [a,b]
Consider now a function f ∈ PC([a, b], R). Using Proposition 7.2.10, there exists a
continuous function g and a step function φ such that f = g + φ. We remark that from
Proposition 7.2.9, the set of step functions forms a vector space. If ψ+ is a step function
such that g ≤ ψ+ , then ψ+ +φ is a step function such that f ≤ ψ+ +φ. We deduce therefore
Z Z Z Z Z Z
that f≤ g+ φ. Similarly, we have also that f≥ g+ φ. So,
[a,b] [a,b] [a,b] [a,b] [a,b] [a,b]

Z Z Z Z Z Z
g+ φ≤ f≤ f≤ g+ φ.
[a,b] [a,b] [a,b] [a,b] [a,b] [a,b]

Z Z
From the result on continuous functions, we have g = g. The above inequality
[a,b] [a,b]
Z Z
implies hence that f= f , so f is integrable over [a, b].
[a,b] [a,b]

Proposition 8.1.13. The function 1Q ∈ L∞ ([0, 1], R) is not Riemann integrable on [0, 1].
Proof. Consider a tagged partition (P, t) on [0, 1] where t = (t1 , · · · , tn ) ∈ [0, 1]n . Since the
set Q is dense in R, we can always choose the tagged values t1 , t2 , · · · , tn ∈ Q ∩ [0, 1]. In this
case, σ(1Q , P, t) = 1. On the other side, the set R \ Q is also dense in R. Following the same
processes, we can obtain another tagged values t′ ∈ ([0, 1] \ Q)n such that σ(1Q , P, t′ ) = 0.
These examples shows that 1Q is not Riemann integrable.
In fact, we can calculate
Z the integral of this function using Borel-Lebesgue integra-
tion, which gives that 1Q dµ = 0. This theory will be introduced in the course of Real
[0,1]
Analysis.

The concept of Riemann integrability can be extended to the functions with values in
C or Cn . The generalization to functions valued in an arbitrarily Banach space will be
introduced in the next subsection.
278 RIEMANN INTEGRATION

Definition 8.1.14. • Let f ∈ L∞ ([a, b], C), we write f = Re(f ) + iIm(f ) where Re(f )
and Im(f ) denote the real part ant imaginary part of f , which are both bounded
functions valued in R. We say f is integrable on [a, b] if Re(f ) and Im(f ) are integrable.
In the case, the integral of f is given by,
Z Z Z
f= Re(f ) + i Im(f ) ∈ C.
[a,b] [a,b] [a,b]

• Let f ∈ L∞ ([a, b], Cn ), we denote by fi ∈ L∞ ([a, b], C) the i-th component of the
function f for i ∈ [[1, n]]. We say f in integrable on [a, b] if for all i ∈ [[1, n]], fi is
integrable on [a, b]. In the case, the integral of f is given by,
R 
Z [a,b] f1
 .  n
. 
 . ∈C .
f =
[a,b] R
[a,b] fn

8.1.2 Integrals of Functions in Vector Values

In this subsection, we discuss the functions valued in the Banach space F . Similarly to the
cases of functions in R, we define the Riemann integrable functions as follows.
Definition 8.1.15. Let f ∈ L∞ ([a, b], F ) and (P, t) be a tagged partition on [a, b].
• We call the Riemann sum of f on (P, t) the following number,
n
X
σ(f, P, t) := f (ti )(xi − xi−1 ) ∈ F,
i=1

where P = (x0 , x1 , · · · , xn ) and t = (t1 , · · · , tn ).

• We say f is Riemann integrable on [a, b] if there exists a number If ∈ F such that for
all ε > 0, there exists δ > 0 such that for any tagged partition (P, t) with |P | ≤ δ, we
have
∥If − σ(f, P, t)∥F ≤ ε.
In the case,
Z we call the number If as the Riemann integral of f on [a, b] and denote
it by (R) f.
[a,b]

Remark 8.1.16. As we have mentioned in the subsection about Riemann integration for
real valued functions, the above definition is not easy to manipulate. However, we can not
introduce generally the “Darboux integrals” for vector-valued functions since we don’t have
in general an order in an arbitrary Banach space. Nevertheless, the idea to approach a
function by using step functions can be always adapted.

Definition 8.1.17. Let φ be a step function defined on [a, b] and (a0 , a1 , · · · , an ) be a

partition adapted to φ. For all i ∈ [[1, n]], we denote by ci ∈ F the value of φ on the
subinterval (ai−1 , ai ). The integral of φ is given by (8.1), i.e.
Z X n
φ= (ai − ai−1 )ci ∈ F.
[a,b] i=1
8.1. INTEGRALS ON A SEGMENT 279

Remark 8.1.18. As the immediate consequences, Propositions 8.1.2, 8.1.3 and 8.1.5 hold
also. And (8.2) becomes furthermore,
Z Z Z
φ ≤ ∥φ∥F ≤ ∥φ∥L∞ ([a,b],F ) = (b − a)∥φ∥L∞ ([a,b],F ) . (8.4)
[a,b] [a,b] [a,b]
F

Here, ∥φ∥F denotes the function x 7→ ∥φ(x)∥F , which is a step function in ST ([a, b], R).

Z
Proposition 8.1.19. The integral map over [a, b], φ 7→ φ, is a continuous func-
[a,b]
tion on the normed space of step functions endowed with the uniform convergence norm,
(ST ([a, b], F ), ∥ · ∥L∞ ([a,b],F ) ).

Proof. We conserve the notations in the proof of the previous proposition. We have, from
the linearity of integral, for all x, y ∈ [a, b],
Z Z Z n
X
φ− ψ = (φ − ψ) = (ai − ai−1 )(ci − di ) (8.5)
[a,b] [a,b] [a,b] i=1
F F F
n
X
≤ (ai − ai−1 )∥ci − di ∥F ≤ (b − a)∥φ − ψ∥L∞ ([a,b],F ) . (8.6)
i=1

Hence the integral map is (b − a)-Lipschitz continuous on the space of step functions for
∥ · ∥L∞ ([a,b],F ) .

Proposition 8.1.20. Let (φn )n∈N be a !

sequence of step functions which converges uniformly
Z
on [a, b]. Then, the sequence φn ∈ F N converges.
[a,b]
n∈N

Proof. Since the sequence (φn )n∈N converges uniformly on [a, b], it is therefore a Cauchy
sequence in the normed space (L∞ ([a, b], F ), ∥ · ∥L∞ ([a,b],F ) ). From the Lipschitz continuity
shown in Proposition 8.1.19, we have, for all n, p ∈ N,
Z Z
φn − φn+p ≤ (b − a)∥φn − φn+p ∥L∞ ([a,b],F ) −→ 0
[a,b] [a,b] n→+∞
F

Z !
Hence, φn is a Cauchy sequence in F . From the completeness of F , it converges.
[a,b]
n∈N

Definition 8.1.21. Let f ∈ PC([a, b], F ) and (φn )n∈N be a sequence of step functions which
converges uniformly on [a, b] to f . We define the integral of f the following limit
Z Z
f = lim φn . (8.7)
[a,b] n→+∞ [a,b]
280 RIEMANN INTEGRATION

Remark 8.1.22. From the denseness result Theorem 7.2.11, for any f ∈ PC([a, b], F ), there
exists a sequence (φn )n∈N of step functions which! converges uniformly on [a, b] to f . Also,
Z
from Proposition 8.1.20 the sequence φn ∈ F N converges. Thus, the above
[a,b]
n∈N
definition is licit.
Z
Proposition 8.1.23. Let f ∈ PC([a, b], F ), the integral f is independent of the choice
[a,b]
of the uniform convergent sequence of step functions (φn )n∈N .
Proof. Let (φn )n∈N and (ψn )n∈N be sequences of steps functions which converge uniformly to
the same function f ∈ PC([a, b], F ). Then, the sequence of functions (φn −ψn )n∈N converges
uniformly to the zero function on [a,! b]. It follows from the continuity of the integral
Z
Proposition 8.1.19 that φn − ψn converges to 0. Hence, the claim follows.
[a,b]
n∈N

Z
Proposition 8.1.24. The integral map φ 7→ φ is a linear continuous map on the
[a,b]
normed space (PC([a, b], F ), ∥ · ∥L∞ ([a,b],F ) ).
Proof. The linearity of the integral map of piecewise continuous functions comes from that
of step functions and the limit. We now prove the continuity. Let f, g ∈ PC([a, b], F ), we
assume the general case f ̸= g. It follows from the denseness of step functions in PC([a, b], F )
Theorem 7.2.11, there exists the sequence of step functions (φn )n∈N and (ψn )n∈N which
converge uniformly on [a, b] respectively to f and to g. Let ε > 0, there exists N ∈ N such
ε ε
that for all n ≥ N , we have ∥f − φn ∥L∞ ([a,b],F ) ≤ as well as ∥g − ψn ∥L∞ ([a,b],F ) ≤ . Then
2 2
it follows from (8.5),
Z Z
φn − ψn ≤ (b − a)∥φn − ψn ∥L∞ ([a,b],F )
[a,b] [a,b]
F
≤ (b − a)(∥φn − f ∥L∞ ([a,b],F ) + ∥f − g∥L∞ ([a,b],F ) + ∥g − ψn ∥L∞ ([a,b],F ) )
≤ (b − a)(ε + ∥f − g∥L∞ ([a,b],F ) ).
Z Z
Since the left-hand-side in the above inequality converges to f− g , we then
[a,b] [a,b]
F
pass n → +∞ to obtain that
Z Z
f− g ≤ (b − a)(ε + ∥f − g∥L∞ ([a,b],F ) ).
[a,b] [a,b]
F

We remark that this relation holds for all ε > 0, we conclude thus,
Z Z
f− g ≤ (b − a)∥f − g∥L∞ ([a,b],F ) .
[a,b] [a,b]
F

This inequality shows that the integral map on PC([a, b], F ) is (b − a)-Lipschitz continuous.
8.1. INTEGRALS ON A SEGMENT 281

PropositionZ 8.1.25. ZIf two functions f, g ∈ PC([a, b], F ) coincide on [a, b] except for finite
points, then f= g.
[a,b] [a,b]

Proof. Since f − g is a step function which equals to the null function except for a finite
set, the integral of f − g is thus zero. From the linearity of integral mapping, the claim
follows.

Remark 8.1.26. From Propositions 8.1.20, 8.1.23 and 8.1.24, the integral map of piecewise
continuous functions are in fact the unique extension by denseness of the integral map of
step functions. In a general manner, we have the following two theorems which can be
proved using the similar arguments.

Theorem 8.1.27. Let U be a dense subset in the metric space (X, dX ) and f be a uniformly
continuous function mapping from U into a complete metric space (Y, dY ). Then, there
exists a unique continuous function g mapping from X into Y such that g|U = f . Moreover,
g is respectively uniformly continuous or K-Lipschitz continuous or isometric if f possesses
the corresponding property.

Proof. Left in Exercises.

Theorem 8.1.28. Let U be a dense subspace in the normed space (E, ∥·∥E ) and f be a linear
continuous function mapping from U into a Banach space (F, ∥ · ∥F ). Then, there exists a
unique linear continuous function g mapping from E into F such that g|E = f . Moreover,
g is K-Lipschitz continuous or isometric if f possesses the corresponding property.

Proof. Left in Exercises.

Z
Summing up the results in this subsection, the integral mapping on [a, b], φ 7→ φ,
[a,b]
can be extended uniquely by continuity from the set of step functions ST ([a, b], F ) into
the set of piecewise continuous functions PC([a, b], F ) thanks to the denseness result The-
orem 7.2.11. In fact, we can define the integral for any uniform limit of step functions via
this approach. Precisely speaking, we introduce the following space of functions.

Definition 8.1.29. We call the set of regulated functionsiv the closure of step functions
ST ([a, b], F ) in the normed space (L∞ ([a, b], F ), ∥ · ∥L∞ ([a,b],F ) ). Equivalently speaking, f is
said to be a regulated function if there exists a sequence of step functions (φn )n∈N such that
(φn )n∈N converges uniformly to f on [a, b]. We simply denote the set of regulated functions
by ST ([a, b], F ).

Remark 8.1.30. • From Theorem 7.2.11, ST ([a, b], F ) is dense in PC([a, b], F ). We have
immediately that PC([a, b], F ) ⊂ ST ([a, b], F ).
iv
This concept is firstly introduced by Nicolas Bourbaki in 1949. Nicolas Bourbaki, is an imaginary
person created by a group of french mathematician from École Normale Supérieure (ENS). They used this
pseudonym in the propose to edit textbooks in pure mathematics. Representative persons in Bourbaki group
contain Henri Cartan, André Weil, Jean Dieudonné, Laurent Schwartz, etc.
282 RIEMANN INTEGRATION

• We can easily observe that ST ([a, b], F ) is a closed vector subspace in the complete
normed space (L∞ ([a, b], F ), ∥ · ∥L∞ ([a,b],F ) ). Thus, it is also a complete normed space.
However, the space (PC([a, b], F ), ∥·∥L∞ ([a,b],F ) ) is not complete although ST ([a, b], F )
is dense in PC([a, b], F ).
Z
• From Theorem 8.1.27, the integral mapping on [a, b], φ 7→ φ can be extended
[a,b]
in to a linear continuous function in the space of regulated functions. Equivalently
speaking, the integral of a regulated function f can be defined by (8.7).
• Following the previous point, the operator norm of the integral mapping is equal to
(b − a).

Example 8.1.31. Consider the function f defined by,

(
1 1 1
f (x) = n if x ∈ ( n+1 , n ]
0 if x = 0.

As well as the sequence of functions (fn )n∈N defined by, for all n ∈ N∗ ,
(
1 1 1
p if x ∈ ( p+1 , p ], ∀p < n
fn (x) = 1 1
n if x ∈ [0, n ].

We can prove that f is the uniform limit of the sequence of step functions (fn )n∈N , so f
is a regulated function, f ∈ ST ([0, 1], R). On the other hand, since f possesses an infinity
numbers of points of discontinuity, f is not piecewise continuous on [a, b].
Furthermore, we can also prove that the sequence of step functions (fn )n∈N is a Cauchy
sequence in (PC([0, 1], R), ∥ · ∥L∞ ([0,1],R) ). And this sequence is not convergent.

Theorem 8.1.32. Any regulated functions in ST ([a, b], F ) is Riemann integrable. And the
Riemann integral of f equals to the integral given by (8.7).
Proof. Let f ∈ ST ([a, b], F ), from the above descriptions, we can define the integral of f
on [a, b] by (8.7) and set the number
Z
If := f ∈ F.
[a,b]

Let ε > 0, it follows from the denseness of step functions in the set of regulated functions
ε
that there exists φ ∈ ST ([a, b], F ) such that ∥φ − f ∥L∞ ([a,b],F ) ≤ . From the
3(b − a)
continuity of integral mapping, we have
Z Z
ε
φ− f ≤ (b − a)∥φ − f ∥L∞ ([a,b],F ) ≤ .
[a,b] [a,b] 3
F

Let Pφ = (x0 , x1 , · · · , xN ) be an adapted partition to φ with N ∈ N∗ . We set

 
1 ε
δ = min x1 − x0 , · · · , xN − xN −1 , > 0.
3 N ∥f ∥L∞ ([a,b],F )
Let (P, t) be a tagged partition on [a, b] such that |P | ≤ δ. From the definition of δ, for
each subinterval [yj−1 , yj ] in the partition P = (y0 , y1 , · · · , ym ), we have two possible cases,
8.1. INTEGRALS ON A SEGMENT 283

1.) [yj−1 , yj ] ⊂ (xi−1 , xi ) for certain i ∈ [[1, N ]];

2.) there exists a unique i ∈ [[1, N ]] such that yj−1 ≤ xi ≤ yj .

We remark furthermore that the second case might appear at most N times for all subin-
tervals in P . We denote respectively by S1 and S2 the set of indexes j ∈ [[1, m]] such that
[yj−1 , yi ] satisfies the first case and the second case. As an immediate result, S1 ∪S2 = [[1, m]].
According to the above two possible cases, we have the following results.

1.) If j ∈ S1 , we have, for all s, s′ ∈ [yj−1 , yi ],

ε
∥f (s) − φ(s′ )∥F = ∥f (s) − φ(s)∥F ≤ ∥φ − f ∥L∞ ([a,b],F ) ≤ . (8.8)
3(b − a)

2.) If j ∈ S2 , we simply remark that for all s ∈ [yj−1 , yj ],

∥φ(s)∥F ≤ ∥φ∥L∞ ([a,b],F ) ≤ ∥f ∥L∞ ([a,b],F ) + ∥φ − f ∥L∞ ([a,b],F )

ε
≤ ∥f ∥L∞ ([a,b],F ) + . (8.9)
3(b − a)

We next consider the tagged values f (tj ) for tj ∈ [yj−1 , yj ]. Introduce now the step
m
f (tj )1(yj−1 ,yj ) as well as ψ := f (tj )1(yj−1 ,yj ) . We remark that the
X X
˜
functions f :=
j=1 j∈S2
Riemann sum of f can be written as the integral of f˜, i.e.,
m m Z
f (tj )1(yj−1 ,yj )
X X
σ(f, P, t) = f (tj )(yj − yj−1 ) = (8.10)
j=1 j=1 [yj−1 ,yj ]
Z m Z
f (tj )1(yj−1 ,yj ) =
X
= f˜. (8.11)
[a,b] j=1 [a,b]

From (8.8) and (8.9), we have the following estimations.

1.) If j ∈ S1 , for all s ∈ (yj−1 , yj ),

ε
∥f (tj ) − (φ + ψ)(s)∥F = ∥f (tj ) − φ(s)∥F ≤ .
3(b − a)
In the consequences,
Z   ε
f˜ − (φ + ψ) ≤ (yj − yj−1 ) . (8.12)
[yj−1 ,yj ] 3(b − a)
F

2.) If j ∈ S2 , for all s ∈ (yj−1 , yj ),

ε
∥f (tj ) − (φ + ψ)(s)∥F = ∥φ(s)∥F ≤ ∥f ∥L∞ ([a,b],F ) + .
3(b − a)
In the consequences,
Z  
  ε
f˜ − (φ + ψ) ≤ (yj − yj−1 ) ∥f ∥L∞ ([a,b],F ) + . (8.13)
[yj−1 ,yj ] 3(b − a)
F
284 RIEMANN INTEGRATION

Since S1 ∪ S2 = [[1, m]], we can deuce from Proposition 8.1.5 that

Z   XZ   XZ  
˜
f − (φ + ψ) = ˜
f − (φ + ψ) + f˜ − (φ + ψ) .
[a,b] j∈S1 [yj−1 ,yj ] j∈S2 [yj−1 ,yj ]

In the consequences,
Z   X Z  
f˜ − (φ + ψ) ≤ f˜ − (φ + ψ)
[a,b] j∈S1 [yj−1 ,yj ]
F F
X Z  
+ ˜
f − (φ + ψ) .
j∈S2 [yj−1 ,yj ]
F

Using (8.12) and (8.13), we have,

Z m

˜
 X ε X
f − (φ + ψ) ≤ (yj − yj−1 ) + ∥f ∥L∞ ([a,b],F ) (yj − yj−1 ).
[a,b] 3(b − a)
F j=1 j∈S2

ε
We recall that Card(S2 ) ≤ N , (yj − yj−1 ) ≤ |P | ≤ δ and δ ≤ . So, we
3N ∥f ∥L∞ ([a,b],F )
have,
Z   ε ε 2ε
f˜ − (φ + ψ) ≤ + = .
[a,b] 3 3 3
F

Nevertheless, we have also,

Z X ε
ψ ≤ ∥f ∥L∞ ([a,b],F ) (yj − yj−1 ) ≤ ∥f ∥L∞ ([a,b],F ) N δ ≤ .
[a,b] 3
F j∈S2

Thus,
Z Z Z
    2ε ε
f˜ − φ ≤ f˜ − (φ + ψ) + ψ ≤ + = ε.
[a,b] [a,b] [a,b] 3 3
F F F
Z
Recalling (8.10), the above inequality shows exactly σ(f, P, t) − φ ≤ ε. In
[a,b]
F
conclusion, f is Riemann integrable and the Riemann integral is given by (8.7).

Corollary 8.1.33. Any piecewise continuous function is Riemann integrable and the Rie-
mann integral is given by (8.7).
Proof. Direct consequence of Theorem 8.1.32 and the relation PC([a, b], F ) ⊂ ST ([a, b], F ).

(
sin( x1 ) if x ∈ (0, 1]
Example 8.1.34. The function x 7→ is Riemann integrable on
0 if x = 0.
[0, 1] but it is not a regulated function.
8.1. INTEGRALS ON A SEGMENT 285

8.1.3 Properties of Integrals over a Segment

Proposition 8.1.35. The inequality (8.4) holds for any function in PC([a, b], F ).
Proof. Let f ∈ PC([a, b], F ), consider a sequence of step functions (φn )n∈N which converges
uniformly to f . Then, for all n ∈ N, (8.4) implies that
Z Z
φn ≤ ∥φn ∥F ≤ (b − a)∥φn ∥L∞ ([a,b],F ) .
[a,b] [a,b]
F

Using the continuity of the integral mapping, we can pass n → +∞ in the above inequality
and it leads to
Z Z
f ≤ ∥f ∥F ≤ (b − a)∥f ∥L∞ ([a,b],F ) .
[a,b] [a,b]
F

Thus, the claim follows.

Proposition 8.1.36. Let f ∈ PC([a, b], F ) and α ∈ R. Define a function fα : x 7→ f (x − α)

on the segment [a − α, b − α]. Then, the function fα is piecewise continuous on [a − α, b − α]
and it holds,
Z Z
f= fα . (8.14)
[a,b] [a−α,b−α]

Proof. The piecewise continuity of fα is straightforward from the piecewise continuity of f

on the corresponding subintervals.
Let φ be a step function on [a, b]. We can write in a general manner that ∀t ∈ [xi−1 , xi ],
φ(t) = ci where (x0 , x1 , · · · , xn ) denotes a partition of [a, b]. Then, (x0 −α, x1 −α, · · · , xn −α)
is an adapted partition of φα on [a − α, b − α], and it follows,
Z X n Xn Z
φ= (xi − xi−1 )ci = ((xi − α) − (xi−1 − α))ci = φα .
[a,b] i=1 i=1 [a−α,b−α]

So, (8.14) holds for step functions. The case of piecewise continuous functions follows by
considering a sequence of step functions which converges uniformly.

Using the same arguments, we can prove the following results.

Corollary 8.1.37. Let f be a piecewise continuous function on [a, b].
x
1.) If λ > 0, we consider the function g defined by, ∀x ∈ [a, b], g(x) = f ( ). Then, g is
λ
piecewise continuous on [λa, λb] and it holds,
Z Z
g=λ f.
[λa,λb] [a,b]

2.) If we consider the function g defined by, ∀x ∈ [a, b], g(x) = f (−x). Then, g is
piecewise continuous on [−b, −a] and it holds,
Z Z
g= f.
[−b,−a] [a,b]
286 RIEMANN INTEGRATION

3.) Let T be a continuous linear application mapping from F into a complete normed
space G. Then, the function T ◦ f is also piecewise continuous on [a, b] and it holds,
Z ! Z
T f = T ◦ f.
[a,b] [a,b]

4.) If f ∈ PC([a, b], C), then

Z Z Z ! Z Z ! Z
f= f, Re f = Re(f ), Im f = Im(f ).
[a,b] [a,b] [a,b] [a,b] [a,b] [a,b]

Proposition 8.1.38. Let [c, d] be a segment included in [a, b] and f ∈ PC([a, b], F ). Then,
f |[c,d] ∈ PC([c, d], F ) and f 1[c,d] ∈ PC([a, b], F ). Moreover,
Z Z Z
f= f |[c,d] = f 1[c,d] . (8.15)
[c,d] [c,d] [a,b]

Proof. The results f |[c,d] ∈ PC([c, d], F ) and f 1[c,d] ∈ PC([a, b], F ) can be obtained by
considering a common refinement of P and (a, c, d, b) where P is an adapted partition of f .
The relation (8.15) is straightforward if f is a step function. Then the general cases can
be deduced using a uniformly convergent sequence of step functions.

Corollary 8.1.39. Let (x0 , x1 , · · · , xn ) be a partition on [a, b] and f ∈ PC([a, b], F ). Then,
Z n Z
X
f= f. (8.16)
[a,b] i=1 [xi−1 ,xi ]

n
f 1[xi−1 ,xi ] . Then
X
Proof. Since (x0 , x1 , · · · , xn ) is a partition on [a, b], we have then f =
i=1
applying the linearity of the integral mapping and Proposition 8.1.38, the relation (8.16)
holds.

Definition 8.1.40. Let I be an interval on R. We say f ∈ PC(I, F ) if f is piecewise

continuous on any segment included in I. Let f be a such function and a, b ∈ I, we denote
 Z


 f if a < b,
Z b  Z[a,b]

f (t)dt = − f if a > b,
a 


 [b,a]
0 if a = b.


Using the above notations, we can rewrite the properties introduced in this subsection
as follows. We find back the usual writing styles on integrals.
Let f ∈ PC(I, F ) and a, b ∈ I.
Z a Z b
1.) f (t)dt = − f (t)dt.
b a
8.1. INTEGRALS ON A SEGMENT 287

Z b
2.) The map f 7→ f (t)dt is a linear continuous mapping from PC(I, F ) into F .
a
Z b
3.) f (t)dt ≤ (b − a) sup ∥f (t)∥F .
a F t∈[a,b]

Z b−α Z b
4.) If α ∈ R such that a − α, b − α ∈ I, then f (t)dt = f (t − α)dt.
a−α a
Z λb Z b
5.) If λ ∈ R∗ such that λa, λb ∈ I, then f (t)dt = λ f (λt)dt.
λa a

6.) Let c ∈ I, it holds

Z b Z c Z b
f (t)dt = f (t)dt + f (t)dt (8.17)
a a c

7.) Let f, g ∈ PC([a, b], R) such that f ≤ g on [a, b]. Then,

Z Z
f≤ g.
[a,b] [a,b]

8.) Let f ∈ PC([a, b], R) with values in positive real numbers.

Z We assume that f (x0 ) > 0
where x0 is a point where f is continuous. Then, f > 0.
[a,b]

Proposition 8.1.41. We define the application ∥ · ∥L1 ([a,b],F ) : PC([a, b], F ) → R+ by,
Z b
∀f ∈ PC([a, b], F ), ∥f ∥L1 ([a,b],F ) = ∥f (t)∥F dt.
a

Then, ∥ · ∥L1 ([a,b],F ) defines a semi-normv on PC([a, b], F ).

Proof. Left in Exercises.

Corollary 8.1.42. The application ∥ · ∥L1 ([a,b],F ) defines a norm on the space of continuous
functions C([a, b], F ).

Proof. Left in Exercises.

Remark 8.1.43. In the space of continuous functions on [a, b], the norms ∥ · ∥L∞ ([a,b],F ) and
∥ · ∥L1 ([a,b],F ) are not equivalent.

Theorem 8.1.44. Let (fn )n∈N be a sequence of functions.

v
We recall that a semi-norm refers to any positive function which satisfies the homogeneity and the
triangular inequality. If, in addition, a semi norm satisfies the separation property, then it is a norm.
288 RIEMANN INTEGRATION

• We suppose that for all n ∈ N, fn ∈ PC([a, b], F ) and the sequence of functions
converges uniformly on [a, b] to a function f ∈ PC([a, b], F ). Then,
Z Z
lim fn = f. (8.18)
n→+∞ [a,b] [a,b]

• We suppose furthermore that for all n ∈ N, fn is continuous on [a, b] and (fn )n∈N
converges uniformly on [a, b] to f . Then, (8.18) holds.
X
• We suppose that the series of continuous functions fn converges uniformly on
n
[a, b]. Then, the series of integrals of fn converges in F and we have,
Z +∞
X +∞ Z
X
fn = fn . (8.19)
[a,b] n=0 n=0 [a,b]

Proof. We remark that the second point is just a particular case of the first point, and the
limit function f in the second point is continuous via Theorem 7.1.23.
Concerning on the first point, for all n ∈ N, it follows from the continuity and the
linearity of the integral mapping, we have
Z Z Z
fn − f = (fn − f )
[a,b] [a,b] [a,b]
F F
Z b
≤ ∥fn (t) − f (t)∥F dt ≤ (b − a)∥fn − f ∥L∞ ([a,b],F ) −→ 0.
a n→+∞

Hence, the claim follows.

Finally,
X it is sufficient to apply the second point on the sequence of partial sums of the
series fn to deduce the convergence and (8.19).
n

Remark 8.1.45. The statements in Theorem 8.1.44 also hold for regulated functions.
X
Example 8.1.46. Let cn be a series of complex numbers which is absolutely conver-
n
R −→ C
gent. For all n ∈ N, we define the function fn : . We have then, ∀n ∈ N,
x 7−→ cn einx
X
∥fn ∥L∞ (R,C) ≤ |cn |. From the comparison test, the series ∥fn ∥L∞ (R,C) converges, which
X n
implies that the series of functions fn is normally convergent, so it is uniformly conver-
n X
gent on R. Introduce now the sum function of the series fn given by,
n

+∞
X
∀x ∈ R, f (x) = cn einx .
n=0

Using the continuities of the functions fn and Theorem 7.1.23, the sum function f is con-
tinuous on R.
8.2. FUNDAMENTAL THEOREMS ON CALCULUS 289

Let p ∈ N, we can proceed the same processes to deduce that

+∞
X
∀x ∈ R, f (x)e−ipx = cn ei(n−p)x .
n=0

Moreover, it follows from Theorem 8.1.44 on the segment [0, 2π] that
Z 2π +∞
X Z 2π +∞
X
−ipt i(n−p)t
f (t)e dt = cn e dt = cn 2πδn,p = 2πcp .
0 n=0 0 n=0

8.2 Fundamental Theorems on Calculus

In this section, we generalize the theorems about derivations and integrations which are
already presented in Calculus. We will consider the functions mapping from an interior
non-emptyvi interval I into a Banach space (F, ∥ · ∥F ).

8.2.1 Differentiation
Let f : I → F be a function and a ∈ I.

Definition 8.2.1. The function f is called differentiable (or derivable) at a if the fol-
lowing difference quotient has a limit limit when x → a.

f (x) − f (a)
f ′ (a) = lim ∈ F. (8.20)
x→a x−a
df
We call the quantity f ′ (a) the derivative of f at a, we denote it also by (a).
dx
Equivalently speaking, the following asymptotic behavior of f holds in a neighborhood
of a.

f (x) = f (a) + (x − a)f ′ (a) + o(|x − a|).

This relation deduce directly the following result.

Proposition 8.2.2. If f is differentiable at a then f is continuous at a.

Definition 8.2.3. We say the function f is left-differentiable, if the quotient of difference

in (8.20) has only the left-limit . We call this left-limit the left-derivative of f at a.
We define also the right-derivative of f at a accordingly.

Remark 8.2.4. If a is one of the endpoint of I, then the differentiablility of f at a means

the left-differentiablility (or right-differentiablility, depends the location of the point a).
In the case where a is an interior point of I, f is differentiable at a if, and only if f is
left-differentiable and right-differentiable at a also the derivatives at both two sides are
equal.

vi
That means, I is not reduced into ∅ or a singleton.
290 RIEMANN INTEGRATION

Theorem 8.2.5 (Characterization of Constant Functions). A function f equals to a con-

stant on the interval I if, and only if f is differentiable with the zero derivative on I.

Proof. ⇒: It is straightforward from the definition.

⇐: We assume that f is differentiable with the zero derivative on I. Let ε > 0 and a ∈ I.
We introduce the set

A := {t ∈ I | ∥f (t) − f (a)∥F ≤ ε|t − a|} .

It is clear that a ∈ A. Since f is continuous, the function g : t 7→ ∥f (t)−f (a)∥F −ε|t−a|

is also continuous on I. Thus, A equals to the pre-image of the closed set R− by the
continuous function g. Using the set characterization of continuous functionsvii , the
set A is a closed subset in I.
We assume in the contrary that A ̸= I. There exists thus b ∈ I such that b ∈ / A. We
may assume without losing the generality that b > a. We set then ã := sup(A ∩ [a, b]).
Since A and [a, b] are closed, it follows that ã ∈ A. From the condition f ′ (ã) = 0,
there exists a neighborhood V of ã such that

f (t) − f (ã)
∀t ∈ V, ≤ ε.
t − ã F

Taking now t > ã sufficiently closed to ã, we have,

∥f (t) − f (a)∥F ≤ ∥f (t) − f (ã)∥F + ∥f (ã) − f (a)∥F ≤ ε(t − b) + ε(ã − a) = ε(t − a).

It shows that t ∈ A ∩ [a, b], which is contradictory to the definition of ã.

Hence, A = I, that means for all ε > 0 and t ∈ I, it holds that ∥f (t)−f (a)∥F ≤ ε|t−a|.
It implies that f (t) = f (a), thus the claim follows.

Definition 8.2.6. • We say the function f is differentiable on the interval I if it is

differentiable at each point in I.

• Let f be a differentiable function on I. We call the derivative of f the function

df
f ′ : x 7→ f ′ (x), this function is also denoted by .
dx
• We call a differentiable function f is continuously differentiable or of C 1 -class on
I if its derivative f ′ is continuous on I.

• We denote the set of continuously differentiable functions on I by C 1 (I, F ).

• By convention, the set C 0 (I, F ) refers to the set of continuous functions on I (i.e.,
C 0 (I, F ) = C(I, F )).

• Let k ∈ N∗ , we denote by C k (I, F ) the set of k-times continuously differentiable

functions on I. That means, f ∈ C k (I, F ) if, and only if f is differentiable on I and
its derivative f ′ is of class C k−1 .
vii
See in the chapter of Topology
8.2. FUNDAMENTAL THEOREMS ON CALCULUS 291

• We denote by C ∞ (I, F ) the set of indefinitely differentiable or smooth functions

on I. That \ means, f ∈ C ∞ (I, F ) if, and only if f ∈ C k (I, F ) for all k ∈ N (i.e.,
C ∞ (I, F ) = C k (I, F )).
k∈N

dk f
• We denote by f (k) or the k-times derivative of f if it exists.
dxk

Proposition 8.2.7. Let k ∈ N ∪ {∞}, K = R or C.

• The set C k (I, F ) is a vector space.

• The derivation f 7→ f ′ is a linear application mapping from C k (I, F ) into C k−1 (I, F ).

• Let J ⊂ R be another interval on real numbers. Let f ∈ C 1 (I, F ) and φ ∈ C 1 (J, R)

such that φ(J) ⊂ I. Then, f ◦ φ ∈ C 1 (J, F ) and the following chain rule holds.

(f ◦ φ)′ = (f ′ ◦ φ)φ′ .

• Let f, g ∈ C k (I, K), then f g ∈ C k (I, K) and the following Leibniz formula holds.
n
!
(n)
X n
∀n ∈ N, n ≤ k, (f g) = f (p) g (n−p) .
p=0
p

Proof. Left in Exercises.

8.2.2 Fundamental Theorem of Calculus

Definition 8.2.8. A function g is called as an indefinite integral (or an antiderivative,
or a primitive function) of f ∈ C(I, F ) if g is differentiable on I and g ′ = f .

Proposition 8.2.9. Two indefinite integrals of the same continuous function f are equal
up to a constant.

Proof. Let g1 and g2 are indefinite integrals to f on I. It follows from the linearity of
derivation that

(g1 − g2 )′ = g1′ − g2′ = f − f = 0.

Using the Characterization of Constant Functions Theorem 8.2.5, we deduce that g1 − g2

is a constant on I.

Theorem 8.2.10 (Fundamental Thoerem of Calculus). Let f ∈ C(I, F ) and a ∈ I.

Z x
1.) The function ga : x 7→ f (t)dt is the unique indefinite integral of f which vanishes
a
at a.
292 RIEMANN INTEGRATION

2.) For all indefinite integral g of f and all a, b ∈ I, it holds,

Z b
f (t)dt = g(b) − g(a).
a

Proof. 1.) The uniqueness of the function ga comes from Proposition 8.2.9. We next
justify ga is differentiable and its derivative equals to f .
Let x0 ∈ I and ε > 0. Using the continuity of f at x0 , there exists δ > 0 such that
∀t ∈ I, |t − x0 | ≤ δ ⇒ ∥f (t) − f (x0 )∥F ≤ ε. So, for all x ∈ I, |x − x0 | ≤ δ, we have,
Z x
∥ga (x) − ga (x0 ) − (x − x0 )f (x0 )∥F = f (t)dt − (x − x0 )f (x0 )
x0 F
Z x Z
= (f (t) − f (x0 ))dt ≤ ∥f − f (x0 )∥F ≤ ε|x − x0 |.
x0 F [x0 ,x]

ga (x) − ga (x0 )
Hence, for all x ∈ B F (x0 , δ), − f (x0 ) ≤ ε. This is equivalent to
x − x0 F
say

ga (x) − ga (x0 )
lim = f (x0 ).
x→x0 x − x0

2.) This result is obviously true for g = ga . Using Proposition 8.2.9, it follows that
g = ga + c with c ∈ F being a constant. And the claim follows straightforwardly.

As a direct consequences of the Fundamental Theorem of Calculus, we have the following

corollaries.

Corollary 8.2.11. Let f be a C 1 -class function on I, then for all a, b ∈ I, it holds,

Z b
f (b) − f (a) = f ′ (t)dt.
a

Corollary 8.2.12 (Integration by Parts). Let f, g ∈ C 1 (I, K) and a, b ∈ I. It holds that

Z b Z b
f ′ (t)g(t)dt = f (b)g(b) − f (a)g(a) − f (t)g ′ (t)dt.
a a

Proof. It is sufficient to apply Leibniz formula and the Fundamental Theorem of Calculus.

Corollary 8.2.13 (Change of Variables). Let φ be a C 1 -class function on J where J is an

interval contained in I. Let f be a continuous function on I. Then, for all a, b ∈ J, it
holds,
Z φ(b) Z b
f (t)dt = φ′ (s)f (φ(s))ds.
φ(a) a
8.2. FUNDAMENTAL THEOREMS ON CALCULUS 293

Proof. It is sufficient to apply the chain rule and the Fundamental Theorem of Calculus.

Theorem 8.2.14 (Mean Value Inequality). Let f ∈ C(I, F ) be a C 1 -class function in the
interior ˚
I. Let a, b ∈ ˚
I. It holds that
∥f (b) − f (a)∥F ≤ (b − a) sup ∥f ′ (x)∥F .
x∈[a,b]

Proof. It is consequence from the Fundamental Theorem of Calculus Theorem 8.2.10 and
Proposition 8.1.35. We have,
Z b Z
′
∥f (b) − f (a)∥F = f (t)dt ≤ ∥f ′ ∥F ≤ (b − a) sup ∥f ′ (x)∥F .
a F [a,b] x∈[a,b]

Proposition 8.2.15. Let f be a continuous function on I and g is an indefinite integral of

f on I. Let [a, b] ⊂ I be a segment and x0 ∈ [a, b]. Then, it holds,
∥g∥L∞ ([a,b],F ) ≤ ∥g(x0 )∥F + ∥f ∥L1 ([a,b],F ) ≤ ∥g(x0 )∥F + |b − a|∥f ∥L∞ ([a,b],F ) .
Proof. Let x ∈ [a, b], it follows from the Fundamental Theorem of Calculus that
Z x
g(x) = g(x0 ) + f (t)dt.
x0
In the consequences,
Z Z
∥g(x)∥F ≤ ∥g(x0 )∥F + ∥f ∥F ≤ ∥g(x0 )∥F + ∥f ∥L∞ ([a,b],F )
[x0 ,x] [x0 ,x]
= ∥g(x0 )∥F + |b − a|∥f ∥L∞ ([a,b],F ) .
Hence the claim follows.

Corollary 8.2.16. Let (fn )n∈N be a sequence of continuous functions which converges uni-
formly on any compact in I to a function f . Let x0 ∈ ˚
I. Consider the sequence of functions
(gn )n∈N defined by
Z x
∀n ∈ N, x ∈ I, gn (x) = fn (t)dt.
x0
Then, the sequence of functions (gn )n∈N converges uniformly on any compact in I to an
indefinite integral of f .
Proof. We define the function g by
Z x
∀x ∈ I, g(x) = f (t)dt.
x0
It is obvious that g is an indefinite integral of f . Using the linearity of integral, we deduce
that gn − g is an indefinite integral of fn − f for all n ∈ N.
Let [a, b] ⊂ I be a segment containing x0 . Applying Proposition 8.2.15 on the segment
[a, b], we obtain that
∥gn − g∥L∞ ([a,b],F ) ≤ |b − a|∥fn − f ∥L∞ ([a,b],F ) .
Here ∥ · ∥L∞ ([a,b],F ) denotes the infinity norm on the segment [a, b]. Passing n → +∞ in the
above inequality, we conclude that (gn )n∈N converges uniformly to g on the segment [a, b].
Since any segment in I is concluded in a segment containing x0 , thus the claim follows.
294 RIEMANN INTEGRATION

8.3 Integrals on Any Interval

In this section, we consider the integrals over an interior non empty interval I ⊂ R. We
shall denote by a, b ∈ Rviii the two extreme values of the interval I. As an important fact,
we essentially work on the space of piecewise continuous functions on I.

Definition 8.3.1. We say f is piecewise continuous mapping from the interval I into a
normed space F if for any segment J ⊂ I, the restriction f |J is piecewise continuous on J.
The set of such functions is denoted by PC(I, F ).

Remark 8.3.2. Let f ∈ PC(I, F ).

• f might contain an infinity numbers of discontinuous points on I.

• Let S be the set of points where f is discontinuous. Then for any segment J ⊂ I,
S ∩ J is a finite set. We call such subsets as negligible sets in I ix .

• Assuming for all x ∈ I, there is a property P(x). We say P(x) is true almost
everywherex on I (or for almost every x ∈ I) if P(x) is true for x ∈ I \ S where S
is a negligible set on I.

8.3.1 Integrability of Positive Real-Valued Functions

Definition 8.3.3. Let f be a piecewise
Z continuous function inpositive real values. We say
f is integrable on I if the set f J is a segment, J ⊂ I is bounded from above. In
J
the case, we define the integral of f on the interval I by,
Z Z b Z 
f= f (t)dt := sup f J is a segment, J ⊂ I .
I a J

Remark 8.3.4. Let f be a piecewise function in positive real values on the segment [a, b].

• This definition is coherent with the Riemann integration for the piecewise continuous
functions defined on a segment. In fact, for all c, d ∈ R such that a ≤ c < d ≤ b, we
have
Z Z Z Z Z
f= f+ f+ f≥ f.
[a,b] [a,c] [c,d] [d,b] [c,d]

Since f is positive, each integrals in the previous inequality are positive. And the
equality holds for a = c and b = d.

• This function f is integrable on the intervals (a, b], [a, b) and (a, b). And we have,
Z Z Z Z
f= f= f= f.
[a,b] (a,b] [a,b) (a,b)
viii
Here, the plausible cases where a = +∞ or b = −∞ are taken into considerations.
ix
The concept of negligible sets comes from Real Analysis. In fact, this terms corresponds to the subsets
with zero measure.
x
We also denote it in abbreviation a.e.. For a full investigation of this concept, please refer to Real
Analysis.
8.3. INTEGRALS ON ANY INTERVAL 295

Let Jn = [an , bn ] with

Z a < anZ < bn < b. It follows from the same argument as the
previous point that f ≤ f . Then we can take the sequences (an )n∈N and
Jn [a,b]
(bn )n∈N which are respectively decreasing and increasing such that an −→ a and
Z  n→+∞Z
bn −→ b. We have then the sequence f is increasing and lim f =
n→+∞ Jn n→+∞ J
Z n∈N n

f . Hence, the claim follows.

[a,b]

Proposition 8.3.5. Let I be an interior non empty interval on R. There exists a sequence
of segments (Jn )n∈N which satisfies
• for all n ∈ N, Jn ⊂ Jn+1 ;
[
• Jn = I.
n∈N

The sequence (Jn )n∈N is called as a nested increasing sequence of segments in I.

Proof. Let a, b ∈ R be the two extreme values of the interval I. Consider the sequences
(an )n∈N and (bn )n∈N constructed as follows.
• If a ∈ R and a ∈ I, we set an = a for all n ∈ N.

• If b ∈ R and b ∈ I, we set bn = b for all n ∈ N.

• In the other cases, we set a < a0 < b0 < b. And (an )n∈N , (bn )n∈N are respectively
decreasing and increasing sequences such that an −→ a and bn −→ b (available
n→+∞ n→+∞
even if a = −∞ or b = +∞).
Then, the sequence of segments (Jn )n∈N defined by Jn = [an , bn ] satisfies the desired prop-
erty.

Remark 8.3.6. Let (Jn )n∈N be a nested increasing sequence of segments in I. Then, any
compact subset J in I is contained in Jn for certain n ∈ N. In fact, since a compact set
is[closed and bounded, we can set α = min J ∈ I and β = max J ∈ I. From the relation
Jn , there exists p, q ∈ N such that α ∈ Jp and β ∈ Jq . Then we set n = max{p, q} and
n∈N
deduce that J ⊂ [α, β] ⊂ Jn .

Proposition 8.3.7. Let f be a piecewise continuous function in positive real values defined
on I. Let (Jn )n∈N be a nested increasing sequence of segments in I. Then the following
assertions are equivalent.
1.) f is integrable on I.
Z 
2.) The sequence f is bounded from above.
Jn n∈N
Z 
3.) The sequence f is convergent.
Jn n∈N
296 RIEMANN INTEGRATION

Furthermore, we have
Z Z Z
f = sup f = lim f.
I n∈N Jn n→+∞ J
n

Proof. We assume at first

Z f is integrable on I. Then there exists M > 0 such that for all
compact subset J ⊂ I, f ≤ M . Since any segment in real numbers is compact, we have
Z J Z 
therefore, f ≤ M for all n ∈ N. Reciprocally, we assume that the sequence f
Jn Jn n∈N
is bounded from above. From the previous
Z Z any compact set J ⊂ I is contained in
remark,
certain term J ⊂ Jn . We have thus, f ≤ sup f . We have established 1.) ⇔ 2.).
Z J  n∈N J n

We remark that the sequence f is increasing. Thus, the equivalence 2.) ⇔ 3.)
Jn n∈N
can be obtained from the monotone convergence property. Furthermore, the equality follows
from the same argument.

Example 8.3.8. 1.) The function exponential x 7→ ex is integrable on R− = (−∞, 0].

Since ([−n, 0])n∈N is a nested increasing sequence of segments in R− and for all n ∈ N,
Z 0
et dt = 1 − e−n ≤ 1. Hence,
−n

Z 0 Z 0
et dt = lim et dt = lim (1 − e−n ) = 1.
−∞ n→+∞ −n n→+∞

2.) Let α ∈ R. We consider the function x 7→ xα defined on [1, +∞). The sequence
([1, n])n∈N∗ is a nested increasing sequence of segments in [1, +∞). And we have,
(
n nα+1 −1
if α ̸= −1,
Z
α 1+α
t dt =
1 ln n if α = −1.

α
Z +∞ x 7→ x is integrable on [1, +∞) if, and only if α < −1. In the
Thus, the function
−1
case, we have tα dt = .
1 1+α

3.) From the similar reason, the function x →7 xα is integrable on (0, 1] if, and only if
Z 1
1
α > −1. In the case, we have tα dt = .
0 1+α

Proposition 8.3.9. Let f and g be piecewise continuous functions in positive real values
on I.

1.) We suppose that f ≤ g.

• If g is integrable on I, then f is integrable on I.

• If f is not integrable on I, then g is not integrable on I.
8.3. INTEGRALS ON ANY INTERVAL 297

2.) We suppose that f and g are integrable on I. Then, for all λ, µ ∈ R+ , the function
λf + µg is also integrable on I and it holds,
Z Z Z
(λf + µg) = λ f + µ g. (8.21)
I I I

Proof. 1.) Consider a nested increasing sequence (Jn )n∈N of segments in I. Since f and
g are piecewise continuous functions and f ≤ g, it follows that for all n ∈ N,
Z Z
f≤ g.
Jn Jn
Z 
So, if g is integrable on I, the sequence g is bounded from above. Thus the
Z  J n n∈N

sequence f is also bounded from above. Hence f is integrable on I. The

Jn n∈N
second point can be deduced by contraposition.

2.) It is immediate that λf + µg is piecewise continuous and positive on I. Let (Jn )n∈N
be the sequence considered in the previous point. For all n ∈ N, we have
Z Z Z
(λf + µg) = λ f +µ g.
Jn Jn Jn

Since the functions f and g are integrable on I, the integrals in the right hand side
are convergent when n → +∞. Hence, the left hand side converges, which implies the
integrability of λf + µg and the equality (8.21) holds.

2 2
Example 8.3.10. 1.) The function x 7→ e−x is integrable on R− since ∀x ∈ R− , e−x ≤
e1−x = ee−x and the function x 7→ e−x is integrable on R− .
| sin x|
2.) The function x 7→ is integrable on [1, +∞) if α > 1.
xα

Proposition 8.3.11. Let f be a piecewise continuous and positive function on I. The

following assertions are equivalent.
1.) f is integrable on I.

2.) The indefinite integrals of f are bounded on I.

3.) The indefinite integrals of f converges at a also at b.

Proof. Let F be an indefinite integral of f . Since f is positive on I, the indefinite integral
F is therefore an increasing function. Then, the equivalence 2.) ⇔ 3.) can be deduced from
the monotone convergence property.
For any segment J = [α, β] ⊂ I, it follows from the Fundamental Theorem of Calculus
that
Z β
f (t)dt = F (β) − F (α).
α
298 RIEMANN INTEGRATION
Z 
If F is bounded on I, then the set f J is a segment, J ⊂ I is bounded. Hence,
J
2.) ⇒ 1.).
We assume now f is integrable on I. If F does not converge at a, since F is increasing
Z β
on I, it implies that F diverges to −∞ at a. Let β ∈ I be fixed, we obtain that f (t)dt =
Z  α

F (β) − F (α) −→ +∞. Thus the set f J is a segment, J ⊂ I is not bounded. By

α→a J
contraposition, F converges at a. Using similar arguments, F converges at b. Hence,
1.) ⇒ 3.).

Example 8.3.12. The function − ln is integrable on (0, 1] since x 7→ x − x ln x is an

indefinite integral of ln and lim (x − x ln x) = 0.
x→0

Proposition 8.3.13. Let f be a piecewise continuous and positive function on I and c ∈ ˚ I.

Then, f is integrable on I if, and only of f is integrable on I ∩ (−∞, c] and on I ∩ [c, +∞).
Furthermore, we have
Z b Z c Z b
f (t)dt = f (t)dt + f (t)dt. (8.22)
a a c

Proof. Consider a nested increasing sequence (Jn )n∈N of segments in I. For all n ∈ N, we
denote Jn = [an , bn ]. Then, from certain rank n ≥ n0 ∈ N, we have Jn ∩ (−∞, c] = [an , c]
and Jn ∩ [c, +∞) = [c, bn ]. We remark also that ([an , c])n≥n0 forms a nested increasing
sequence of segments in I ∩ (−∞, c] and the symmetric result holds on I ∩ [c, +∞).
We assume at first that f is integrable on I. Since f is positive, the following inequality
holds for n large enough
Z Z Z
f≤ f ≤ f.
[an ,c] [an ,bn ] I

Z !
In other words, the sequence f is bounded from above. Thus, f is integrable
[an ,c]
n≥n0
on I ∩ (−∞, c]. The integrability on I ∩ [c, +∞) follows from the similar arguments.
We assume now f is integrable on I ∩ (−∞, c] and on I ∩ [c, +∞). Then, for n ≥ n0 ,
we have
Z Z Z Z Z
f= f+ f≤ f+ f.
Jn [an ,c] [c,bn ] I∩(−∞,c] I∩[c,+∞)

Hence, f is integrable on I. By passing n → +∞, the identity (8.22) holds.

Remark 8.3.14. From Proposition 8.3.13, to answer weather the function f is integrable on
the interval I, it is sufficient to study the integrability near the endpoints a and b because
the cases of integrals on segments are already studied in the previous section. We will focus
on the intervals of type I = [a, b) with a ∈ R, b ∈ R or b = +∞.
8.3. INTEGRALS ON ANY INTERVAL 299

Proposition 8.3.15. Let f and g be a piecewise continuous and positive function on I =

[a, b).
• If f = O(g) and g is integrable on I, then f is integrable on I.
b

• If g = o(f ) and g is not integrable on I, then f is not integrable on I.

b

• If f ∼ g, then f and g have the same integrability.

b

Proof. • Since f = O(g), there exist M > 0 and c ∈ I such that

b

∀x ∈ [c, b), f (x) ≤ M g(x).

Applying Proposition 8.3.9, the function x 7→ M g(x) is integrable on [c, b) and hence
f is integrable on [c, b). Also, from Proposition 8.3.13, f is integrable on I.

• The negligence condition implies that there exists c ∈ I such that ∀x ∈ [c, b), g(x) <
f (x). Applying Propositions 8.3.9 and 8.3.13, the function f is not integrable on I.

• If f and g are equivalent in a neighborhood of b, we have at the same time that

f = O(g) and g = O(f ). Thus the claim follows from the first point.
b b

Example 8.3.16. 1.) The function f : x 7→ (1 + x3 )α is integrable on R+ if, and only if

−1
α< , since f (x) ∼ x3α when x → +∞.
3
ln x
2.) The function f : x 7→ is integrable on (0, 1). The integrability of f can be
x−1
established from analyzing its behavior near the 0 and 1.

• We have lim f (x) = 1. So the function f can be extended into a continuous

x→1−
function on (0, 1]. Thus, f is integrable in a neighborhood of 1.
• From f (x) ∼ − ln x ans the function − ln is integrable on (0, 1], so f is integrable
in a neighborhood of 0.

Proposition 8.3.17. Let (bn )n∈N be a sequence of elements in I = [a, b) which tends to b.
Let f be a piecewise continuous
Z bn and  positive function on I. Then, f is integrable on I if,
and only if the sequence f (t)dt is convergent. In that case, we have,
a n∈N
Z b Z bn
f (t)dt = lim f (t)dt.
a n→+∞ a
Z x
Proof. The function F : x 7→ f (t)dt is an indefinite integral of f . Since f is positive, F
a
is increasing on I. So, the integrability of f is equivalent to the boundedness of the function
F on I via the monotone convergence property. Furthermore, the boundedness of F on I is
equivalent to the convergence of F at b which is then equivalent to the convergence of the
sequence (F (bn ))n∈N since bn −→ b.
n→+∞
300 RIEMANN INTEGRATION

We remark that if (bn )n∈N is increasing, then ([a, bn ]) forms a sequence of nested in-
creasing sequence of segments in I. And the Proposition 8.3.17 can be deduced directly by
applying Proposition 8.3.7.

Corollary 8.3.18. Let (bn )n∈N be a sequence of elements in I = [a, b) which tends to b.
Let f be a piecewise continuous and positive function on I. Then, f is integrable on I if,
X Z bn
and only if the series f (t)dt is convergent. In that case, we have,
n bn−1

Z b ∞ Z
X bn
f (t)dt = f (t)dt.
b0 n=1 bn−1

Proof. It is sufficient to consider the sequence of partial sums and apply Proposition 8.3.17.

Recall now the theorem which is already appeared in the chapter on numerical series.

Theorem 8.3.19 (Integral test). Let f be a function defined on [n0 , +∞) with n0 ∈ N. We
assume that f is piecewisely continuous, positive and decreasing function. Then,
X Z n 
• the series f (t)dt − f (n) converges;
n≥n0 +1 n−1

X
• the series f (n) converges if, and only if f is integrable on [n0 , +∞).
n≥n0 +1

Proof. See in the course of the first semester.

Remark 8.3.20. The condition f (x) −→ 0 is a neither necessary nor sufficient condition
x→+∞
for the integrability of f on the interval [a, +∞).
1
• The function x 7→ converges to 0 when x → +∞ and this function is not integrable
x
on [1, +∞).

• Consider the function f defined by,

(
1
n if x ∈ [n, n + n3
]
f (x) =
0 otherwise.
Z n+1
1 X 1
We have, for all n ∈ N, . f (t)dt =
And the series is convergent. Thus,
n n2 n
n2
f is integrable on [1, +∞). However, f is not convergent when x → +∞, it is even
unbounded on [1, +∞).

8.3.2 Integrability of Functions in Vector Values

In this subsection, we consider the piecewise continuous functions defined on I with values
in the Banach space F .
8.3. INTEGRALS ON ANY INTERVAL 301

Definitions

Definition 8.3.21. We say a function f ∈ PC(I, F ) is integrable on I if the function

∥f ∥F : x 7→ ∥f (x)∥F is integrable. We denote
Z
∥f ∥L1 (I,F ) := ∥f ∥F ∈ R+ .
I

The following proposition is a direct consequence of Proposition 8.3.9

Proposition 8.3.22. Let f : I → F and g : I → R+ be piecewise continuous functions

such that for all x ∈ I, ∥f (x)∥F ≤ g(x). If g is integrable on I and, then f is integrable on
I. Moreover,
Z
∥f ∥L1 (I,F ) ≤ g.
I

Example 8.3.23. Let α ∈ R.

eix
1.) The function x 7→ is integrable on [1, +∞) if, and only if α > 1, since ∀x ∈ [0, +∞),
xα
eix 1
α
= α.
x x

sin x sin x 1
2.) Consider the function f : x 7→ on [1, +∞). For all x ∈ [0, +∞), ≤ α,
xα xα x
and the function x 7→ x−α is integrable on [0, +∞) if α > 1. Applying Proposi- X
tion 8.3.22, f is integrable on [1, +∞) if α > 1. If α ≤ 1, we consider the series un
n
defined by,
nπ
| sin t|
Z
∗
∀n ∈ N , un = dt.
(n−1)π tα
Z nπ
1 2 X
Thus, un ≥ | sin t|dt = . We have then un diverges in the
(nπ)α (n−1)π (nπ)α n
cases where α ≤ 1. Hence, f is not integrable on [1, +∞) if α ≤ 1.

As what we have done in the previous subsection, the studies of integrability can be
restricted in the neighborhoods of the extreme points. The following proposition is a direct
consequence of Proposition 8.3.15.

Proposition 8.3.24. Let f : I → F and g : I → R+ be piecewise continuous functions on

I = [a, b) with a ∈ R, b ∈ R. We suppose that g is integrable on I and f = O(g). Then, f
b
is integrable on I.

Proposition 8.3.25. The set L1 (I, F ) of piecewise continuous and integrable functions on
I forms a vector space.
302 RIEMANN INTEGRATION

Proof. It is sufficient to apply Propositions 8.3.9 and 8.3.22 as well as the following trian-
gular inequality. Let f, g ∈ L1 (I, F ) and λ, µ ∈ R or C,

∀x ∈ I, ∥(λf + µg)(x)∥F ≤ |λ|∥f (x)∥F + |µ|∥g(x)∥F .

Lemma 8.3.26. Let J and K be segments in I such that J ⊂ K ⊂ I. Then, for all
f ∈ PC(I, F ), we have
Z Z Z Z
f− f ≤ ∥f ∥F − ∥f ∥F .
K J F K J

Proof. Denoting J = [α, β] and K = [α′ , β ′ ], we have, from (8.17), that

Z Z Z α Z β′
f− f= f (t)dt + f (t)dt.
K J α′ β

Therefore,
Z Z Z α Z β′ Z Z
f− f ≤ ∥f (t)∥F dt + ∥f (t)∥F dt = ∥f ∥F − ∥f ∥F .
K J F α′ β K J

Theorem 8.3.27. Let f ∈ PC(I, F ) be an integrable function  on

Z I. Let (Jn )n∈N be a
sequence of nested increasing segments in I. Then, the sequence f converges in
Jn n∈N
F . Moreover, this limit does not depend on the choice of the sequence (Jn )n∈N .

Proof. Since f is integrable, from definition, the function ∥f∥ZF : x 7→ ∥f (x)∥F is positive
and integrable on I. Using Proposition 8.3.7, the sequence ∥f ∥F is convergent,
Jn n∈N
hence it is a Cauchy sequence. Let n, p ∈ N. It follows from Lemma 8.3.26 and Jn ⊂ Jn+p
that
Z Z Z Z
f− f ≤ ∥f ∥F − ∥f ∥F −→ 0.
Jn+p Jn Jn+p Jn n→+∞
F
Z 
Thus, the sequence f is a Cauchy sequence in F . From the completeness of F ,
Jn n∈N
it converges in F .
We consider now two sequences (Jn )n∈N and (Kn )n∈N of nested increasing segments in
I such that ∀n ∈ N, Jn ⊂ Kn . It follows also from Lemma 8.3.26 that
Z Z Z Z
f− f ≤ ∥f ∥F − ∥f ∥F .
Kn Jn F Kn Jn
8.3. INTEGRALS ON ANY INTERVAL 303
Z Z Z
Using again Proposition 8.3.7, we have lim ∥f ∥F = ∥f ∥F = lim ∥f ∥F .
n→+∞ J I n→+∞ K
Z Z n Z n 
Hence, it follows that f− f −→ 0. Knowing that the sequences f
Kn Jn n→+∞ Jn
Z  F n∈N

and f are convergent, we deduce then

Kn n∈N
Z Z
lim f = lim f.
n→+∞ J n→+∞ K
n n

In the general cases, consider (Jn )n∈N = ([αn , βn ])n∈N and (J ′ n )n∈N = ([αn′ , βn′ ])n∈N
being two sequences of nested increasing segments in I. We then construct for all n ∈ N,
Kn = [min(αn , αn′ ), max(βn , βn′ )]. Then, (Kn )n∈N is also a sequence of nested increasing
segments in I and we have Jn ⊂ Kn , Jn′ ⊂ Kn for all n ∈ N. Applying the previous result,
we have then
Z Z Z
lim f = lim f = lim f.
n→+∞ J n→+∞ K n→+∞ J ′
n n n

Definition 8.3.28. Let f ∈ PC(I,

Z F ) be an integrable function on I. We call the integral
of f the limit of the sequence f where (Jn )n∈N is a sequence of nested increasing
Jn n∈N Z Z b
segments in I. We denote the integral of f by f or f (t)dt where a, b ∈ R refers to the
I a
endpoints of I.

Properties
Z
Proposition 8.3.29. The integral mapping on I, f 7→ f is a linear application on
I
L1 (I, F ).

Proof. Let f, g ∈ L1 (I, F ) and λ, µ ∈ R or C. Consider (Jn )n∈N being a sequence of nested
increasing segments in I. It follows from the linearity of the integral map on a segment
Proposition 8.1.24 that for all n ∈ N,
Z Z Z
(λf + µg) = λ f +µ g.
Jn Jn Jn

Form Theorem 8.3.27, the sequences of integrals over Jn appeared above are all convergent.
We then pass n → +∞ to obtain that
Z Z Z
(λf + µg) = λ f + µ g.
I I I
304 RIEMANN INTEGRATION

Proposition 8.3.30 (Triangular Inequality). Let f ∈ L1 (I, F ), we have,

Z Z
f ≤ ∥f ∥F .
I F I

Proof. Let (Jn )n∈N be a sequence of nested increasing segments in I. It follows from Propo-
sition 8.1.35 that for all n ∈ N,
Z Z
f ≤ ∥f ∥F .
Jn F Jn

Since f is integrable, the two sides in the above inequality are both convergent when n →
+∞. By passing n → +∞, the claim follows.

Proposition 8.3.31. Let T ∈ L (F, G) be a linear continuous application mapping from F

into a Banach space G. Let f ∈ L1 (I, F ). Then, T ◦ f ∈ L1 (I, G) and we have,
Z Z 
T ◦f =T f ∈ G.
I I

Proof. Since T is linear and continuous, there exists M > 0 such that for all y ∈ F ,
∥T (y)∥G ≤ M ∥y∥F . In the consequence, for all x ∈ I, ∥T ◦ f (x)∥G = ∥T (f (x))∥G ≤
M ∥f (x)∥F . This proves the integrability of T ◦ f on I.
Let (Jn )n∈N be a sequence of nested increasing segments in I. Using Corollary 8.1.37,
we obtain that
Z Z 
T ◦f =T f .
Jn Jn

Then, the claim follows by passing n → +∞ and the linearity of T .

We remark next that the properties satisfied by piecewise continuous and positive func-
tions hold also for functions in vector values. Precisely speaking, we have the following
results.
Proposition 8.3.32. Let f ∈ PC(I, F ) be a piecewise continuous function on I.
• Let c ∈ ˚
I. Then, f is integrable on I if, and only of f is integrable on I ∩ (−∞, c] and
on I ∩ [c, +∞). Furthermore, the equation (8.22) holds.

• Let I = [a, b) with

Z xa ∈ R and b ∈ R. We assume that f is integrable. Then, the
function g : x 7→ f (t)dt has a limit when x → b. Also,
a
Z b Z x
f (t)dt = lim g(x) = lim f (t)dt.
a x→b x→+∞ a

• Let a, b ∈ R be the endpoints of the interval I. Let (αn )n∈N and (βn )n∈N be sequences
of elements in I such that αn −→ a and βn −→ b. We assume that f is integrable
n→+∞ n→+∞
on I. Then,
Z b Z βn
f (t)dt = lim f (t)dt.
a n→+∞ α
n
8.3. INTEGRALS ON ANY INTERVAL 305

• We assume that f is integrable on the segment [a, b]. Then, f is integrable on (a, b],
on [a, b) and on (a, b). Also,
Z Z Z Z
f= f= f= f.
[a,b] (a,b] [a,b) (a,b)

• We assume that f is integrable on I. Then, for any interval I ′ ⊂ I, f is integrable on

I ′ and it follows,
Z Z
f = f 1I ′ .
I′ I

Proof. Left in Exercises.

Nevertheless, we present some results about the integrability of functions with values in
finite dimensional spaces.

Proposition 8.3.33. Let f be a piecewise continuous function defined on I.

• Case f ∈ PC(I, R): We define the positive part f + and negative part f − of f by,
∀x ∈ I, f + (x) = max{0, f (x)} and f − (x) = max{0, −f (x)}. The functions f + and
f − are therefore piecewise continuous and positive on I. As an immediate result, we
have f = f + − f − and |f | = f + + f − . Then, f is integrable on I if, and only if f +
and f − are integrable on I. In the case, we have
Z Z Z Z Z Z
−
f= f − f+
and |f | = f + f − .
+
I I I I I I

• Case f ∈ PC(I, C): The function f is integrable on I if, and only if the real part
Re(f ) ∈ PC(I, R) and the imaginary part Im(f ) ∈ PC(I, R) are integrable on I. And
we have,
Z Z Z
f = Re(f ) + i Im(f ).
I I I

In parallel, the function f is integrable on I if, and only if its conjugate f is integrable
on I. Also, it follows
Z Z
f = f.
I I

• Case f ∈ PC(I, Cn ): We denote by fi ∈ PC(I, C) the i-th component of the function

f . Then, the function f is integrable on I if, and only if for all i ∈ [[1, n]], fi is
integrable on I. And we have,
R 
Z I f1
 . 
f = . 
 . .
I R
I fn
306 RIEMANN INTEGRATION

Improper Integrals
Definition 8.3.34. Let f be a piecewise continuous function on I = [a, b) with aZ ∈ R and
x
b ∈ R. We say that f has an improper integral on I if the function g : x 7→ f (t)dt
a
defined on I has a limit when x → b.

Remark 8.3.35. From Proposition 8.3.32, any integrable function has its improper integral
on I. However, there are functions possessing the improper integrals without being inte-
grable on I. We remark also that the improper integrals can be defined on the intervals of
type I = (a, b] or I = (a, b) with a, b ∈ R.

Example 8.3.36. Let I = [2π, +∞).

eix 1
• We consider the function f defined by, ∀x ∈ I, f (x) = . Since |f (x)| = and the
x x
1
function x 7→ is not integrable on I. The function f is not integrable on I.
x
On the other hand, using the integration by parts, we have, for all x > 2π,
Z x it Z x it
e i ieix e
g(x) := dt = − −i 2
dt.
2π t 2π x 2π t
Z x it
1 e
Since the function x 7→ 2 is integrable on I, the function x 7→ 2
dt has a limit
x 2π t
when x → +∞. Hence, the function g has a limit when x → +∞. Thus, although f
is not integrable on I, f has an improper integral on I.
• Following the previous point. We consider the real part and the imaginary part of
sin x cos x
the function f . We have thus, the functions x 7→ and x 7→ are not
x
Z +∞ Zx +∞
sin t cos t
integrable on I = [2π, +∞). However, the integrals dt and dt
2π t 2π t
sin x
exist. Moreover, since the function x 7→ possesses a finite limit when x → 0, the
Z +∞ x
sin t
integral dt exists .
0 t

Comparison Relations
Proposition 8.3.37. Let I = [a, b) with a ∈ R and b ∈ R. Let f ∈ PC(I, F ) and g ∈
PC(I, R+ ). We suppose that g is integrable.
1.) If f = O(g), then f is integrable on I. Also, when x → b,
b
Z b Z b 
f (t)dt = O g(t)dt .
x x

2.) If f = o(g), then f is integrable on I. Also, when x → b,

b
Z b Z b 
f (t)dt = o g(t)dt .
x x
8.3. INTEGRALS ON ANY INTERVAL 307

Proof. From Proposition 8.3.24, the function f is integrable on I.

1.) The condition f = O(g) implies that there exists A ∈ I and M > 0 such that for all
b
x ∈ [A, b), ∥f (x)∥F ≤ M g(x). Let A ≤ x < y < b, we have,
Z y Z y Z y
f (t)dt ≤ ∥f (t)∥F dt ≤ M g(t)dt.
x F x x

Since the functions f and g are integrable on I, we can thus pass y → b and it follows,
Z b Z b
f (t)dt ≤ M g(t)dt.
x F x

2.) We proceed the same method. Let ε > 0, there exists A ∈ I such that for all x ∈ [A, b),
we have ∥f (x)∥F ≤ εg(x). Thus, for all x ∈ [A, b), we have
Z b Z b
f (t)dt ≤ ε g(t)dt.
x F x

Proposition 8.3.38. Let I = [a, b) with a ∈ R and b ∈ R. Let f ∈ PC(I, F ) and g ∈

PC(I, R+ ). We suppose that g is not integrable.
1.) If f = O(g), then it follows, when x → b,
b
Z x Z x 
f (t)dt = O g(t)dt .
a a

2.) If f = o(g), then it follows, when x → b,

b
Z x Z x 
f (t)dt = o g(t)dt .
a a
Z x
Proof. Since g is positive and not integrable on I, we have that g(t)dt −→ +∞.
a x→b

1.) The condition f = O(g) implies that there exists A ∈ I and M > 0 such that for all
b
x ∈ [A, b), ∥f (x)∥F ≤ M g(x). Let A ≤ x < b, we have,
Z x Z x Z x Z x
f (t)dt ≤ ∥f (t)∥F dt ≤ M g(t)dt ≤ M g(t)dt.
A F A A a
Z x
From the relation g(t)dt −→ +∞, there exists B > 0 such that for all x ≥ B, we
a x→b
have
Z A Z x
f (t)dt ≤ g(t)dt.
a F a

Thus, for x ∈ I, x ≥ max{A, B}, it holds that

Z x Z A Z x Z x
f (t)dt ≤ f (t)dt + f (t)dt ≤ (1 + M ) g(t)dt.
a F a F A F a
308 RIEMANN INTEGRATION

2.) Let ε > 0, the condition f = o(g) implies that there exists A ∈ I and such that for all
b
x ∈ [A, b), ∥f (x)∥F ≤ εg(x). Let A ≤ x < b, we have,
Z x Z x Z x Z x
f (t)dt ≤ ∥f (t)∥F dt ≤ ε g(t)dt ≤ ε g(t)dt.
A F A A a
Z x
From the relation g(t)dt −→ +∞, there exists B > 0 such that for all x ≥ B, we
a x→b
have
Z A Z x
f (t)dt ≤ε g(t)dt.
a F a

Thus, for x ∈ I, x ≥ max{A, B}, it holds that

Z x Z A Z x Z x
f (t)dt ≤ f (t)dt + f (t)dt ≤ 2ε g(t)dt.
a F a F A F a

Corollary 8.3.39. Let f and g be piecewise continuous and positive functions on I = [a, b).
We assume that f ∼ g.
b

1.) If g is integrable on I, then f is also integrable on I. And we have, when x → b,

Z b Z b
f (t)dt ∼ g(t)dt.
x x

2.) If g is not integrable on I, then f is neither integrable on I. And we have, when

x → b,
Z x Z x
f (t)dt ∼ g(t)dt.
a a

Proof. It is sufficient to use Propositions 8.3.37 and 8.3.38 by remarking that g − f = o(g)
in a neighborhood of b.

8.4 Convergence Theorems

8.4.1 Integrable and Square Integrable Spaces
Z
Proposition 8.4.1. The mapping ∥ · ∥L1 (I,F ) : f 7→ ∥f ∥F defines a semi-norm on the
I
vector space L1 (I, F ).

Proof. Left in Exercises.

Remark 8.4.2. • ∥ · ∥L1 (I,F ) is not a norm in the space of piecewise continuous functions
since ∥f ∥L1 (I,F ) = 0 implies that f = 0 on I almost everywhere (that means, except for
a set of discontinuous points which remains finite in any intersection with a segment).
8.4. CONVERGENCE THEOREMS 309

• ∥ · ∥L1 (I,F ) defines a norm on the space L1c (I, F ) = L1 (I, F ) ∩ C(I, F ) of continuous
and integrable functions on I.

• The normed space (L1c (I, F ), ∥ · ∥L1 (I,F ) ) is not complete.

Z
• The integral mapping f 7→ f is linear and continuous on (L1c (I, F ), ∥ · ∥L1 (I,F ) ) with
I
an operator norm which equals to 1.

Definition 8.4.3. A sequence of integrable functions (fn )n∈N is said to be convergent in

L1 (I, F ) if there exists f ∈ L1 (I, F ) such that lim ∥fn − f ∥L1 (I,F ) = 0. In the case, we
n→+∞
have also
Z Z
lim fn = f.
n→+∞ I I

Proof. It follows from the triangular inequality Proposition 8.3.30,

Z Z Z Z
fn − f = (fn − f ) ≤ ∥fn − f ∥F −→ 0.
I I I I n→+∞
F F

Definition 8.4.4. We call a piecewise continuous function f with values in R or C is

square integrable if the functions x 7→ |f (x)|2 is integrable on I. We denote by L2 (I, C)
the set of square integrable functions. Let f ∈ L2 (I, C), we define
sZ
∥f ∥L2 (I,C) := |f |2 .
I

Also, we denote by L2c (I, C) the set of continuous and square-integrable functions on I.

Proposition 8.4.5. Let f, g be square integrable functions on I, then their product f g is

integrable on I.

Proof. Since the function (|f | − |g|)2 is positive, we have the inequality 2|f ||g| ≤ |f |2 + |g|2 .
Also, since f and g are both square integrable, the right-hand-side is therefore integrable.
Thus, the claim follows.

Proposition 8.4.6. The set L2 (I, C) of continuous square integrable functions forms a
vector space.

Proof. We only prove here the set L2 (I, C) is stable under linear combinations. Let f, g ∈
L2 (I, C) and λ, µ ∈ C, we have,

|λf + µg|2 = (λf + µg)(λf + µg) = |λ|2 |f |2 + |µ|2 |g|2 + λµf g + λµf g.

Using the square integrability of f and g as well as Proposition 8.4.5, the claim follows.
310 RIEMANN INTEGRATION

Proposition 8.4.7. Let f, g ∈ L2 (I, C), we define,

Z
⟨f, g⟩L2 (I,C) := f g.
I

Then, ⟨·, ·⟩L2 (I,C) defines a sesquilinear, Hermitian and positive form on L2 (I, C). We have
also, for all f ∈ L2 (I, C),

∥f ∥2L2 (I,C) = ⟨f, f ⟩L2 (I,C) .

Proof. Left in Exercises.

Proposition 8.4.8 (Cauchy-Schwarz Inequality). Let f, g ∈ L2 (I, C), it holds,

Z 2 Z Z
fg ≤ |f |2 |g|2 .
I I I

The equality holds if, and only if there exists λ ∈ C such that g = λf except for finite points.

Proof. It is sufficient to adapt the proof of Cauchy-Schwarz inequality in an inner product

space.

Definition 8.4.9. A sequence of integrable functions (fn )n∈N is said to be convergent in

L2 (I, C) if there exists f ∈ L2 (I, C) such that lim ∥fn − f ∥L2 (I,C) = 0.
n→+∞

8.4.2 Convergence Theorems

Here, we present some important theorems stating weather it is licit to interchange the
symbols of limit and integration when it comes to calculate the integral of the limit function
of a sequence of functions. However, this intention is generally false for an arbitrary sequence
of functions, even if this sequence of functions converges uniformly.

Example 8.4.10. Consider the sequence of functions (fn )n∈N defined by, for all n ∈ N∗ ,
(
1 2
fn (x) = n if x ∈ [0, n ]
0 if x > n2 .

Then, (fn )n∈N converges uniformly to the null function on R+ . But we have,
Z
1
fn = n2 = n −→ +∞.
R+ n n→+∞

Proposition 8.4.11. Let I be a bounded interval and (fn )n∈N be a sequence of piecewise
continuous and integrable on I. We suppose that (fn )n∈N converges uniformly to a piecewise
continuous function f on I. Then, f is integrable on I. And we have,
Z Z
f = lim fn .
I n→+∞ I
8.4. CONVERGENCE THEOREMS 311

Proof. Since (fn )n∈N converges uniformly to f , there exists N ∈ N such that for all ∥fN −
f ∥L∞ (I,F ) ≤ 1. Then, for all x ∈ I, ∥f (x)∥F ≤ ∥fN (x)∥F +∥f (x)−fN (x)∥F ≤ 1+∥fN (x)∥F .
Since I is bounded, the constant function x 7→ 1 is integrable on I. From the previous
inequality, the function f is integrable on I.
Denoting by |I| the length of the bounded interval I, we have
Z Z Z Z
f − fn ≤ ∥f − fn ∥F ≤ ∥f − fn ∥L∞ (I,F ) ≤ |I|∥f − fn ∥L∞ (I,F ) −→ 0.
I I I I n→+∞
F

Hence, the claim follows.

Theorem 8.4.12 (Lebesgue’s Dominated Convergence Theorem). Let (fn )n∈N be a se-
quence of piecewise continuous functions on I. We suppose that
• the sequence (fn )n∈N converges pointwisely to a piecewise continuous function f ;

• there exists an integrable function ϕ on I such that

∀n ∈ N, ∀x ∈ I, ∥fn (x)∥F ≤ ϕ(x).

Then, f is integrable on I and we have,

Z Z
f = lim fn .
I n→+∞ I

Proof. Difficult theorem, we admit this result. In fact, the Riemann integration theory is
insufficient to establish this theorem. To prove this theory, it is needed to apply Measure
Theory and Lebesgue integration theory, which will be introduced in the course of Real
Analysis.

Example 8.4.13. • Consider the sequence of functions (fn )n∈N defined by,

∀x ∈ [0, 1], fn (x) = n2 xe−nx .

Then, (fn )n∈N converges pointwisely to null function on [0, 1]. However,
Z 1 Z n Z 1
−t −n
fn (t)dt = te dt = 1 − (n + 1)e −→ 1 ̸= 0 = 0dt.
0 0 n→+∞ 0

This example shows that only pointwise convergence is insufficient to interchange limit
and integral. To do so, either the uniform convergence or the domination condition
are essential.

• Consider the sequence of functions (fn )n∈N defined on R+ by,

  n2
 x2
1− 2 if x ∈ [0, n],

fn (x) = n

 0 if x > n.
2
Then, the sequence of functions (fn )n∈N converges pointwisely to f : x 7→ e−x .
Also, from the convexity of the exponential function, we have, for all x ∈ [0, n],
312 RIEMANN INTEGRATION

x2 x2
≤ e− n2 . Hence, for all x ∈ R+ , we have 0 ≤ fn (x) ≤ e−x . Remarking that the
2
1− 2
n 2
function x 7→ e−x is integrable on R+ , we then apply Theorem 8.4.12 to show that
n n2 +∞ √
t2
Z Z Z Z
2 π
lim 1− 2 dt = lim fn = f= e−t dt = .
n→+∞ 0 n n→+∞ R
+ R+ 0 2

Theorem 8.4.14 (Theorem of Continuity under Integral Signs). Let f be a function map-
ping from A × I into F where A ⊂ E and E is a normed space. We suppose that
• for all x ∈ A, the function f (x, ·) : t 7→ f (x, t) is piecewise continuous on I;

• for all t ∈ I, the function f (·, t) : x 7→ f (x, t) is continuous on A;

• there exists an integrable function ϕ on I such that

∀x ∈ A, ∀t ∈ I, ∥f (x, t)∥F ≤ ϕ(t).

Z
Then, the function g : x 7→ f (x, t)dt is continuous on A.
I

Z ∈ A, from the domination

Proof. At first, we justify the function g is well-defined. For all x
condition, the function f (x, ·) is integrable on I. Thus, g(x) = f (x, t)dt is well-defined.
I
Let a ∈ A and (xn )n∈N ∈ AN be a sequence of elements in A such that xn −→ a in E.
n→+∞
We define then the sequence of functions (fn )n∈N by, ∀n ∈ N, fn = f (xn , ·) : t 7→ f (xn , t).
From the second assumption, we have, for all t ∈ I, lim fn (t) = f (a, t). In other words,
n→+∞
the sequence of functions (fn )n∈N converges pointwisely to the function f (a, ·).
Next, using the domination condition and applying Theorem 8.4.12, we deduce that
Z Z Z
lim g(xn ) = lim f (xn , t)dt = lim fn = f (a, t)dt = g(a).
n→+∞ n→+∞ I n→+∞ I I

Hence, g is continuous at any arbitrary point in A, equivalently speaking, g is continuous

on A.

Theorem 8.4.15 (Theorem of Derivation under Integral Signs). Let f be a function map-
ping from J × I into F where J is also a non empty interior interval in R. We suppose
that
• for all t ∈ I, the function f (·, t) : x 7→ f (x, t) is differentiable on J;

• for all x ∈ J, the function f (x, ·) : t 7→ f (x, t) is integrable on I;

∂f ∂f
• for all x ∈ J, the partial derivative (x, ·) : t 7→ (x, t) is piecewise continuous on
∂x ∂x
I;

• there exists an integrable function ϕ on I such that

∂f
∀x ∈ J, ∀t ∈ I, (x, t) ≤ ϕ(t).
∂x F
8.4. CONVERGENCE THEOREMS 313
Z
Then, the function g : x 7→ f (x, t)dt is differentiable on J and we have,
I
Z
′ ∂f
∀x ∈ J, g (x) = (x, t)dt.
I ∂x

Proof. Let a ∈ J, we define, for all x ∈ J and all t ∈ I,


 f (x, t) − f (a, t) if x ̸= a

h(x, t) = x−a
∂f
(a, t) if x = a.


∂x
Let t ∈ I be fixed, since the function f (·, t) is differentiable on J, the function x 7→ h(x, t)
is continuous on J. Consider now x ∈ J, x ̸= a. It follows from the Mean Value Inequality
Theorem 8.2.14 that
Z x Z x
∂f ∂f
∥f (x, t) − f (a, t)∥F = (s, t)ds ≤ (s, t) ds ≤ ϕ(t)|x − a|.
a ∂x F a ∂x F

Thus,

∥f (x, t) − f (a, t)∥F

∥h(x, t)∥F = ≤ ϕ(t).
|x − a|

When x = a, the above inequality follows directly fromZthe domination condition.

Thus, applying Theorem 8.4.14, the function x 7→ h(x, t)dt is continuous on J. We
I
then apply the linearity of the integral mapping to obtain that, for all x ∈ J, x ̸= a,
Z  Z
g(x) − g(a) f (x, t) − f (a, t)
Z Z
1
= f (x, t)dt − f (a, t)dt = dt = h(x, t)dt.
x−a x−a I I I x−a I

g(x) − g(a)
This shows that the function x 7→ can be extended into a continuous function
x−a
on J, which implies the existence of the limit of this function at a. Hence, g is differentiable
at a and the claim follows.

Corollary 8.4.16. Let f be a function mapping from J ×I into F and k ∈ N∗ . We suppose

that

• for all t ∈ I, the function f (·, t) : x 7→ f (x, t) is of C k -class on J;

• there exists the integrable functions ϕ0 , ϕ1 , · · · , ϕk on I such that

∂if
∀i ∈ [[0, k]], ∀x ∈ J, ∀t ∈ I, (x, t) ≤ ϕi (t).
∂xi F

Z
Then, the function g : x 7→ f (x, t)dt is of C k -class on J and we have,
I

∂if
Z
(i)
∀i ∈ [[0, k]], ∀x ∈ J, g (x) = (x, t)dt.
I ∂xi
314 RIEMANN INTEGRATION

Proof. Left in Exercises.

Theorem 8.4.17 (Fubini). Let f be a function mapping from J × I into F . We suppose

that f is integrable on J × I. Then,
Z
• the function g1 : x 7→ f (x, y)dy is integrable on J;
I
Z
• the function g2 : y 7→ f (x, y)dx is integrable on I;
J
Z Z Z
• g1 = g2 = f , equivalently speaking,
J I J×I
Z Z  Z Z 
f (x, y)dx dy = f (x, y)dy dx.
I J J I

Proof. Difficult theorem, we admit this result.

8.5 Exercises
Exercise 8.5.1. Let f be a continuous function mapping the segment [a, b] ⊂ R into C.
For any p ∈ [1, +∞), we define
Z !1/p
p
∥f ∥Lp ([a,b].C) = |f | .
[a,b]

1.) Give the definition of the infinity norm ∥f ∥L∞ ([a,b],C) of f on [a, b].

2.) Show that there exists c ∈ [a, b] such that |f (c)| = ∥f ∥L∞ ([a,b],C) .

3.) Show that ∥f ∥Lp ≤ (b − a)1/p ∥f ∥L∞ ([a,b],C) .

4.) Let ε > 0. Show that

∃δ > 0 s.t. ∥f ∥Lp ([a,b].C) ≥ (2δ)1/p (∥f ∥L∞ ([a,b],C) − ε).

5.) Show that lim ∥f ∥Lp ([a,b].C) = ∥f ∥L∞ ([a,b],C) .

p→+∞

Exercise 8.5.2. In the following cases, is the function f integrable on the interval R∗+ =
(0, +∞)? Justify your answers.

1.) f : x 7→ (ln x)e−x .

sin x
2.) f : x 7→ .
x3/2
 
1
3.) f : x 7→ sin .
x
8.5. EXERCISES 315

Exercise 8.5.3. Study the existence of the following integrals.

Z +∞
2
1.) e−(ln x) dx.
0
Z +∞  
1
2.) sin dx.
0 x2
1
ln x ln(1 − x)
Z
3.) dx, where α ∈ R.
0 xα

Exercise 8.5.4. Let f be a differentiable function mapping from R into C. We assume

that f and f ′ are integrable on R.

1.) Show that f has the limits when x → ±∞.

2.) Show that f tends to 0 when x → ±∞.

3.) Let g : R → C be a differentiable function such that g and g ′ are bounded on R. Show
that the functions f ′ g and g ′ f are integrable on R and deduce that
Z Z
f ′ g = − g ′ f.
R R

Exercise 8.5.5. Let f be a uniformly continuous and positive function on R+ . We suppose

that f is integrable on R+ . Show that f tends to 0 when x → +∞.

Exercise 8.5.6. Let f ∈ PC(R, C). We suppose that f is integrable on R. Consider the
sequence (un )n∈N given by,
Z
∀n ∈ N, un = f (t) cos(nt)dt.
R

1.) Justify that the sequence (un )n∈N is well-defined.

2.) Show that lim un = 0.

n→+∞

Application of Lebesgue’s dominated convergence theorem

t
Exercise 8.5.7. Let f be the function defined by, ∀t ∈ R∗+ , f (t) = .
et −1
1.) Show that f is integrable on R∗+ .

2.) For all n ∈ N∗ , we define the function fn on R∗+ by, fn (t) = te−nt . Calculate In :=
Z +∞
fn (t)dt.
0
X
3.) Show that the series of functions fn is pointwisely convergent to f on R∗+ .
n≥1
316 RIEMANN INTEGRATION
Z +∞
4.) Calculate f (t)dt.
0

Exercise 8.5.8. Consider a sequence (an )n∈N given by,

Z π
sinn t
∀n ∈ N, an = √ dt.
0 t

1.) Justify the definition of (an )n∈N .

2.) Show that the sequence of functions in the integrals is convergent on (0, π). Give its
limit function and the type of convergence.

3.) Show that the sequence (an )n∈N converges and determinate its limit.

X
Exercise 8.5.9. Consider the series of functions fn given by,
n

∀n ∈ N, fn (t) = (−1)n t2n .

Z 1
1.) Calculate fn (t)dt for all n ∈ N.
0

XZ 1
2.) Show that the series fn (t)dt converges.
n 0

X
3.) Show that the series of functions fn converges on (0, 1). Give the sum function
n
and the type of convergence.
+∞
X (−1)n π
4.) Show that = .
2n + 1 4
n=0

Exercise 8.5.10. Let f be a smooth function on R and a ∈ R. Consider the function g

given by,

 f (x) − f (a) if x =

̸ a
∀x ∈ R, g(x) = x−a
 f ′ (a) if x = a

1.) By using the function t 7→ f (tx + (1 − t)a), give an integral representation of g.

2.) Show that g is also smooth on R.

3.) Deduce the Hadamard’s Lemma: For any smooth function φ on R such that φ(0) =
0, there exists another smooth function ψ such that ∀x ∈ R, φ(x) = xψ(x).
8.5. EXERCISES 317

Original Proof of Stone-Weierstrass Theorem

In this subject, the main objective is to provide the original proof by Weierstrass of his
famous theorem on approximation by polynomials of any continuous function, i.e., Stone-
Weierstrass Theorem 7.2.16.

Definition 8.5.11. We call the heat kernel (also known as Gaussianxi kernel or Gauss-
Weierstrass kernel) the family of functions (Kt )t>0 given by,
1 − x2
∀t > 0, ∀x ∈ R, Kt (x) = √ e 4t .
4πt

Exercise 8.5.12. Let t > 0.

1.) Show that Kt is integrable on R.

Z +∞
2.) Calculate Kt (x)dx.
−∞

3.) Consider now δ > 0. We define

Z Z −δ Z +∞
IK (δ, t) = Kt = Kt (x)dx + Kt (x)dx.
R\[−δ,δ] −∞ δ

Show that IK (δ, t) −→ 0 when δ is fixed.

t→0

Definition 8.5.13. Let f and g be piecewisely continuous functions on R. We define by

the following formula the convolution product, denoted by f ∗ g, of those two functions.
Z +∞
∀x ∈ R, f ∗ g(x) = f (x − y)g(y)dy.
−∞

We remark that the integral in the definition is, by default, not well-defined for an
arbitrary pair of functions.

Exercise 8.5.14. Let f ∈ L∞ (R, R) and g ∈ L1 (R, R).

1.) Show that f ∗ g is well-defined.

2.) Deduce that f ∗ g ∈ L∞ (R, R) and that

∥f ∗ g∥L∞ (R,R) ≤ ∥f ∥L∞ (R,R) ∥g∥L1 (R,R) .

3.) Show that f ∗ g = g ∗ f .

4.) Consider x0 ∈ R and δ > 0. Show that

! Z !
|f ∗ g(x0 )| ≤ sup |f (y)| ∥g∥L1 (R,R) + ∥f ∥L∞ (R,R) |g| .
y∈[x0 −δ,x0 +δ] R\[−δ,δ]
xi
Karl Friedrich Gauß(1777-1855), German mathematician, astronomer and physicist. One of the greatest
mathematician in the history.
318 RIEMANN INTEGRATION

5.) We assume in addition that f is continuous on R. Let x0 ∈ R. Show that

|f ∗ Kt (x0 ) − f (x0 )| −→ 0.
t→0

6.) We assume that f is uniformly continuos on R. Show that

∥f ∗ Kt − f ∥L∞ (R,R) −→ 0.
t→0

Definition 8.5.15. Let f ∈ PC(R, R).

• We say a function f vanishes on a set S if for all x ∈ S, f (x) = 0.

• We call the support of f the closure of the complementary of the set where f vanishes.
That is,

supp(f ) := {x ∈ R | f (x) ̸= 0}.

Exercise 8.5.16. Let f ∈ PC(R, R). We assume that f has a compact support.

1.) Show that f ∗ g is well-defined for any g ∈ PC(R, R).

2.) Show that f ∗ ge is a polynomial function if ge is a polynomial function.

3.) Let h > 0 such that supp(f ) ⊂ [−h, h]. Show that for any ϕ ∈ PC(R, R), it holds that

∥f ∗ ϕ∥L∞ ([−h,h],R) ≤ 2h∥f ∥L∞ ([−h,h],R) ∥ϕ∥L∞ ([−2h,2h],R) .

Exercise 8.5.17. Let [a, b] ⊂ R denote a segment, we consider now a continuous function
f ∈ C([a, b], R).

1.) Justify that f is uniformly continuous and bounded on [a, b].

2.) Deduce that f can be extended into a function fe which is also uniformly continuous,
bounded and compactly supported on R. We assume from now on that

h > 0 s.t. [a, b] ⊂ supp(fe) ⊂ [−h, h].

3.) From this question, let ε > 0 be given. Show that, for all t > 0, there exists a
polynomial function ϕt such that ∥Kt − ϕt ∥L∞ ([−2h,2h],R) ≤ ε.

4.) Justify that fe ∗ ϕt is a polynomial function. Then give an estimation of ∥fe ∗ ϕt − fe ∗

Kt ∥L∞ ([−h,h],R) in terms of ε.

5.) Prove Stone-Weierstrass Theorem: there exists a polynomial function p such that
∥f − p∥L∞ ([a,b],R) ≤ ε.
8.5. EXERCISES 319

Euler’s Gamma Function

Exercise 8.5.18. Let x ∈ R∗+ .
1.) Show that the function t 7→ tx−1 e−t is integrable on R∗+ .
We define the Euler’s Gamma function given by
Z +∞
∗
∀x ∈ R+ , Γ(x) = tx−1 e−t dt.
0

2.) Let [a, b] ⊂ R∗+ be a segment. Show that Γ is continuous on [a, b] then deduce that Γ
is continuous on R∗+ .

3.) Show that Γ is indefinitely differentiable on R∗+ .

4.) Show that for all x ∈ R∗+ , Γ(x + 1) = xΓ(x) then deduce that for all n ∈ N∗ ,
Γ(n + 1) = n!.
1 1
5.) Calculate Γ( ) then Γ(n + ) for all n ∈ N.
2 2
1
6.) Using the continuity of Γ at x = 1, show that Γ(x) ∼ when x → 0.
x

Exercise 8.5.19. Let x ∈ R∗+ . We define the following sequence (un )n∈N ,

t n
Z n  
∗ x−1
∀n ∈ N , un = t 1− dt.
0 n
1.) Justify that (un )n∈N is well-defined.
n
x Y
x
nx n! 1 + k1

1 n
2.) Show that for all n ∈ N∗ , un = = .
x(x + 1) · · · (x + n) x n+1 1 + xk
k=1

3.) Show that (un )n∈N converges and deduce thatxii

+∞
x
1 Y 1 + k1
Γ(x) = .
x 1 + xk
k=1

Exercise 8.5.20. Generally speaking, when it comes to study the asymptotic behavior of
the functions given under the form
Z
g(x) = f (x, t)dt,
I

we might apply the following instructions.

The integral function f presents often a peak at tx , when the parameter x is fixed,
also the distribution of f is concentrated in the neighborhoods of tx . In this scenario, we
could replace f by an approximation near tx such that the integral on a corresponding
neighborhood can be simply obtained.
xii
This formula is considered as the original definition of the Gamma function, it is dated to 1729 in a
letter from Euler to Goldbach.
320 RIEMANN INTEGRATION

1.) Let x > 0 be fixed. Determinate the extrema of the function t 7→ tx e−t on R∗+ . Study
the variations of this function.
Z +∞  x x Z +∞
x+1 −t x
2.) Show that Γ(x + 1) = x (te ) dt = x e(ln(1+t)−t)x dt.
0 e −1

3.) Now, consider δ ∈ (0, 1) fixed.

ln(1 + δ)
(a) Using the convexity of ln, show that ln(1 + t) ≤ t for all t ≥ δ.
δ
(b) Assuming x ≥ 1, show that
Z +∞
e(ln(1+t)−t)x dt ≤ C1 Ax1 ,
δ

where C1 > 0 and A1 ∈ (0, 1), they depend only on δ.

Z −δ
(c) Show that e(ln(1+t)−t)x dt ≤ C2 Ax2 .
−1

4.) Let ε > 0. Show that there exists δ ∈ (0, 1) such that for all t ∈ [−δ, δ], it holds,
 
1 2 1
− t ≤ ln(1 + t) − t ≤ − − ε t2 .
2 2

√
Z δ r
2 π
5.) Let p > 0. Show that lim x e−pxt dt = .
x→+∞ −δ p
Z δ
6.) Give an asymptotic equivalence of e(ln(1+t)−t)x dt while x → +∞.
−δ
Z +∞
7.) Give an equivalence of e(ln(1+t)−t)x dt while x → +∞. Next prove Stirling’s
−1
√  x x
formula: Γ(x + 1) ∼ 2πx .
x→+∞ e

Laplace Transforms and Applications

Exercise 8.5.21. Let f : R∗+ → R be a piecewise continuous function. Let ξ ∈ R∗+ , we call
the Laplace Transform of f at ξ the following integral in the case of existence
Z +∞
L(f )(ξ) := f (t)e−ξt dt.
0

1.) Let ξ0 ∈ R+ such that the function t 7→ f (t)e−ξ0 t is integrable on R∗+ .

Show that the Laplace transform of f is well-defined for all ξ ∈ [ξ0 , +∞).

2.) We assume that the Laplace transform of f is defined on the open interval (a, +∞)
where a ∈ R∗+ .
d
Show that L(f ) is differentiable on (a, +∞) and give the integral formula for L(f ).
dξ
8.5. EXERCISES 321

1
3.) We consider from now on the function g : t 7→ .
1 + t2 Z
Show that the function g is integrable on R+ and calculate g.
R+

4.) Show that the Laplace transform L(g) is two times differentiable on R∗+ and satisfies
the differential equation
1
∀t ∈ R∗+ , y ′′ (t) + y(t) = . (DE)
t

5.) Let (ξn )n∈N be a sequence of strictly positive real numbers such that ξn −→ 0. Cal-
n→+∞
culate the limit lim L(g)(ξn ) by using Lebesgue’s dominated convergence theorem.
n→+∞

6.) Following the previous question, determinate lim L(g)(ξ).

ξ→+∞

Exercise 8.5.22. We consider the following function h defined for all ξ ∈ R∗+ ,
Z +∞
sin t
h(ξ) := dt.
0 ξ +t
In the following questions, we begin by considering a particular ξ0 ∈ R∗+ .
eit
1.) Is the function t 7→ integrable on R∗+ ? Justify your answer.
ξ0 + t
1 − cos t
2.) Show that the function t 7→ is integrable on R∗+ .
(ξ0 + t)2
3.) Let M > 0. By using integration by parts on the segment [0, M ], show that the
sin t
function t 7→ has an improper integral on R∗+ and that
ξ0 + t
Z +∞
1 − cos t
h(ξ0 ) = dt.
0 (ξ0 + t)2

4.) Let a ∈ R∗+ . Show that h is differentiable on the interval [a, +∞), then deduce that
h is differentiable on R∗+ .
5.) By using integration by parts, show that
Z +∞
sin t
∀ξ ∈ R∗+ , h′ (ξ) = − dt.
0 (ξ + t)2

6.) Show that h satisfies (DE) on R∗+ .

7.) From the theory of differential equations, there exists A, B ∈ R such that

∀ξ ∈ R∗+ , L(g)(ξ) − h(ξ) = A cos ξ + B sin ξ.

By letting ξ → +∞, show that A = B = 0.

Z +∞
sin t π
8.) Show that dt = .
0 t 2
322 RIEMANN INTEGRATION
 Chapter 9

Power Series

9.1 Definitions
Definition 9.1.1. Let (an )n∈N ∈ CN be a sequence of complex numbers. We call a power
X C −→ C
series any series of functions fn where ∀n ∈ N, fn : . And we denote
n z 7−→ an z n
X
this series of functions by an z n . The sequence (an )n∈N is called as the coefficients of
n
this power series.

Remark 9.1.2. • You should be aware of the notations of those for power series and
those for numerical series when the complex number z is fixed.
X
• We also denote by an tn and call it the power series on real variables when the
n
functions fn take the real variable t for all n ∈ N.
• PowerXseries can be defined in a more general way. In fact, any series of functions of
type an (z − z0 )n is a power series where z0 ∈ C. It is sufficient to proceed a change
n
of variables ze = z − z0 to bring those power series into the forms in Definition 9.1.1.
X X
Definition 9.1.3. Let an z n and bn z n be power series.
n n
X
• We call the sum of those two power series the power series (an + bn )z n .
n
X
• We call the Cauchy product of those two power series the power series cn z n
n
given by
n
X
∀n ∈ N, cn = ak bn−k .
k=0
X
• We call the product of the power series an z n with the scalar λ ∈ C the power
X n
series (λan )z n .
n

323
324 POWER SERIES

The set of power series, endowed with the above algebraic structure forms a commutative
algebra.

9.2 Convergences of Power Series

X
In this section, we consider the power series an z n .
n

Proposition 9.2.1. Let z0 X∈ C. We assume that the sequence (an z0n )n∈N ∈ CN is bounded.
Then, the numerical series an z n converges absolutely for all z ∈ C such that |z| < |z0 |.
n

Proof. For z0 = 0, the result is immediate. We suppose from now on that |z0 | > 0
Since the sequence (an z0n )n∈N is bounded. Then there exists M > 0 such that ∀n ∈ N,
|a0 ||z0 |n ≤ M . Also, for all z ∈ C, we have,
n
|z|n z
∀n ∈ N, |an z n | = |an z0n | n
≤M .
|z0 | z0
X z n
The relation |z| < |z0 | implies the geometric series is convergent. Thus, it follows
z
nX 0
from the comparison test, we deduce that the series an z n is absolutely convergent.
n

X
Definition 9.2.2. We call the radius of convergence of the power series an z n the
n
number R ∈ R+ given as the supremum of the set composed by all real numbers r ≥ 0 such
that the sequence (an rn )n∈N is bounded.

Remark 9.2.3. If for all r > 0, the sequence (an rn )n∈N is always bounded, then we say
R = +∞.

X
Theorem 9.2.4. Let R ∈ R+ be the radius of convergence of the power series an z n .
n
Let z ∈ C.
X
• If |z| < R, then the numerical series an z n converges absolutely.
n
X
• If |z| > R, then the numerical series an z n diverges.
n

Proof. If R = 0, then for all z ∈ C∗ , |z| n

X> 0. It implies that the sequence (an z ) is
unbounded, which implies that the series an z n diverges. We assume from now on that
n
R > 0. Consider z ∈ C.

• If |z| < R, there exists then r ∈ R such that |z| < r < R. Then the claim follows from
Proposition 9.2.1 by setting z0 = r.
9.2. CONVERGENCES OF POWER SERIES 325
X
• If |z| > R, then the series an z n diverges since the sequence (an z n )n∈N is un-
n
bounded.

X X X
Example 9.2.5. • The power series an z n , |an |z n , λan z n for λ ∈ C have the
n n n
same radius of convergence.
X zn
• The radius of convergence of the power series is infinity.
n
n!
X
• The radius of convergence of the power series nn z n is 0.
n

X zn
• The radius of convergence of the power series equals to 1.
n
n≥1

Solution. Left in Exercises.

X
Definition 9.2.6. Let R be the radius of convergence of the power series an z n . We
n
define
X
• the disc of convergence of the power series an z n the open ball in C, D :=
n
{z ∈ C | |z| < R};

• the circle of convergence the boundary C := {z ∈ C | |z| = R} of the disc of con-

vergence.
+∞
X
Also, we call the sum function of the power series the function S : z 7→ an z n which is
X n=0
n
defined whenever the numerical series an z converges.
n

As the direct consequences of Theorem 9.2.4, the sum function of a power series is
defined on the disc of convergence D and it is not defined on the exterior of D ∪ C = D.
In general, the behavior of the power series on the circle of convergence is uncertain. For
instance, the following power series have all a radius of convergence which equals to 1.
X
Example 9.2.7. • z n diverges for all z ∈ C, |z| = 1;
n

X zn
• converges for all z ∈ C, |z| = 1;
n2
n≥1

X zn
• converges for z = −1 but diverges for z = 1.
n
n≥1
326 POWER SERIES

In the following proposition, for a number L ∈ R+ , we set by convention,

(
1 0 if L = +∞
=
L +∞ if L = 0.
X
Proposition 9.2.8. Let R ∈ R+ be the radius of convergence of the power series an z n .
n

1
• If the sequence ( n |an |)n∈N∗ converges to L ∈ R+ , then R = .
p
L
 
|an+1 | 1
• If the sequence converges to L ∈ R+ , then R = .
|an | n∈N L

Proof. Direct consequences of root test and ratio test. Left in Exercises.

X X
Now we consider two power series an z n and bn z n . And we denote by Ra and Rb
n n
respectively their radii of convergence.

Proposition 9.2.9. • If |an | ≤ |bn | for all n ∈ N then Ra ≥ Rb .

• If an = O(bn ) then Ra ≥ Rb .

• If an ∼ bn then Ra = Rb .

Proof. • Let r > 0 such that (bn rn )n∈N is bounded. The relation |an | ≤ |bn | implies
that the sequence (an rn )n∈N is also bounded. Hence, Ra ≥ Rb .

• The domination relation an = O(bn ) means that there exists

X M > 0 such that |an | ≤
M |bn | from certain rank. In parallel, the power series (M bn )z n has the radius of
n
convergence equals to Rb . Then, the claim follows from the previous point.

• The equivalence relation an ∼ bn means that an = O(bn ) and bn = O(an ). And the
previous point implies that Ra ≥ Rb and Rb ≥ Ra .

Theorem 9.2.10.
X We denote by Ra+b and Rab respectively
X the radii of convergence of the
n n
power series (an + bn )z and the Cauchy product cn z where
n n

n
X
∀n ∈ N, cn = ak bn−k .
k=0

• If Ra ̸= Rb , then Ra+b = min{Ra , Rb }.

• If Ra = Rb , then Ra+b ≥ Ra = Rb .

• Rab ≥ min{Ra , Rb }.
9.2. CONVERGENCES OF POWER SERIES 327

Moreover, for all z ∈ C such that |z| < min{Ra , Rb }, we have

+∞
X +∞
X +∞
X
(an + bn )z n = an z n + bn z n ;
n=0 n=0 n=0
+∞ +∞ +∞
! !
X X X
n n n
cn z = an z bn z .
n=0 n=0 n=0
X
Proof. • Let z ∈ C such that |z| < min{Ra , Rb }. It is immediate that the series an z n
X n
and bn z n are absolutely convergent. By using a simple estimation |(an + bn )z n | ≤
n X
|an z | + |bn z n |, the term-to-term sum series
n
(an + bn )z n is also absolutely conver-
n
gent. This implies that Ra+b ≥ min{Ra , Rb } and it holds
+∞
X +∞
X +∞
X
n n
(an + bn )z = an z + bn z n .
n=0 n=0 n=0

• We assume without of losing the generality that Ra < Rb . Consider now z ∈ C with
Ra < |z| < Rb . Then, the sequence (bn z n )n∈N is bounded and (an z n )n∈N is unbounded.
As the sum of those sequences, the sequence ((an + bn )z n )n∈N is unbounded. In the
consequences, Ra+b ≤ Ra and therefore, Ra+b = min{Ra , Rb }.
X X
• Let z ∈ C such that |z| < min{Ra , Rb }. The numerical series an z n and bn z n
n n
are absolutely convergent. It follows from the
X result of the Cauchy products of two
absolutely convergent series that the series cn z n is also absolutely convergent. It
n
implies that Rab ≥ min{Ra , Rb } and the identity,
+∞ +∞
! +∞ !
X X X
n n n
cn z = an z bn z .
n=0 n=0 n=0

X
Theorem 9.2.11. The power series (n + 1)an+1 z n has the same radius of convergence
X n
as an z n .
n
X X
Proof. We denote respectively by R and R′ the radii of convergence of an z n and (n+
n n
1)an+1 z n . Let r ∈ R+ . If r < R′ , then the sequence ((n + 1)an+1 rn )n∈N is bounded, which
implies there exists M > 0 such that for all n ∈ N,
Mr
(n + 1)|an+1 |rn ≤ M ⇔ |an+1 |rn+1 ≤ .
n+1
Sine the above right-hand-side is also bounded, we deduce that the sequence (an rn )n∈N is
also bounded. Thus, R ≥ R′ .
328 POWER SERIES

On the other hand, if r < R, then we can choose ε > 0 such that r < r + ε < R.
Then there exists M > 0 such that sequence (an (r + ε)n )n∈N is bounded by M . By setting
r+ε n+1
δ := > 1, it follows from the comparison relation (n + 1) = o(δ n ) that ≤ 1 from
r δn
certain rank. So, we have for n ∈ N being large enough,

(n + 1)rn
 
n 1 n+1 M
(n + 1)|an+1 |r = n+1
|an+1 |(r + ε)n+1 = n
|an+1 |(r + ε)n+1 ≤ .
(r + ε) r+ε δ r

It shows that the sequence ((n + 1)an+1 rn ) is also bounded. Hence, R′ ≥ R. In conclusion,
R′ = R.

Theorem 9.2.12. A power series converges normally (thus uniformally) on any compact
subset contained in its disc of convergence. Moreover, it diverges on any point at the exterior
of the disc of convergence. However, there are no general results on the convergence of a
power series on its circle of convergence.
X
Proof. Let R ∈ R+ be the radius of convergence of the power series an z n . The special
n
case R = 0 is immediate, we assume now R > 0. Let A ⊂ D be a compact set contained
in the disc of convergence D := {z ∈ C | |z| < R}. Since the function z 7→ |z| is continuous
on C, it is also continuous on A. From the compactness of A and the continuity of this
function, there exists z0 ∈ A such that for all z ∈ S, |z| ≤ |z0 | < X
R. For all n ∈ N, we have
n n
∀z ∈ S, |an z | ≤ |an ||z0 | . By using Theorem 9.2.4, the series |an ||z0 |n converges. It
X n
implies that the series of functions an z n converges normally on A. Also, Theorem 9.2.4
X n
n
states also that an z diverges on C \ D. Furthermore, Example 9.2.7 shows that the
n
convergence on the circle of convergence depends on the considered power series.

Remark 9.2.13. 1.) In general, a power series does not converge uniformly on the disc of
convergence D.

2.) If a power series converges uniformally on the disc of convergence D, it converges also
uniformally on D = D ∪ C. (Left in Exercises)
X
3.) If there exists z0 ∈ C such that an z0n diverges, then the power series does not
n
converge uniformally on D.
X
4.) If the series |an |Rn (with R ∈ R+ ) converges, then the power series converges
n
normally on D.

9.3 Properties of Sum Functions

X
In this section, we consider a power seriesan z n with a radius of convergence R ∈ R+ and
n X
we denote by S the sum function defined on where the numerical series an z n converges.
n
9.3. PROPERTIES OF SUM FUNCTIONS 329

Theorem 9.3.1. The sum function S is continuous in the disc of convergence D :=

{z ∈ C | |z| < R}.

Proof. It is the consequence of Theorem 9.2.12 and Theorem X 7.1.25 because for all n ∈ N
the function z 7→ an z n is continuous on C and the power series an z n converges uniformly
n
on any compact in D.

Theorem 9.3.2. The function t 7→ S(t) is of class C ∞ on (−R, R) and the successive
derivations are obtained as the sum functions of the successive term-to-term derivations of
the power series. Precisely speaking, for all k ∈ N,

+∞ +∞
(k)
X dk X (n + k)! n
∀t ∈ (−R, R), S (t) = an k tn = an+k t .
dt n!
n=0 n=0

On the other hand, this function t 7→ S(t) possesses the indefinite integrals under the
following form

+∞
X an n+1
t 7→ z0 + t ,
n+1
n=0

where z0 ∈ C.

Proof. We deal only the case where k = 1, the general cases can be obtained by induction.
(−R, R) −→ C
We denote for all n ∈ N, fn : . It is clear that fn is continuously
t 7−→ an tn
′
differentiable and fn+1 (t) = (n + 1)an+1 tn for t ∈ (−R, R). By using Theorem 9.2.11, the
X
power series (n+1)an+1 z n has the same radius of convergence. So, the series of functions
X n
fn′ converges uniformally on any compact in (−R, R). Applying now Theorem 7.1.30,
n
the function t 7→ S(t) is of C 1 -class on (−R, R) also,

+∞
X +∞
X
∀t ∈ (−R, R), S ′ (t) = fn′ (t) = an (n + 1)tn .
n=0 n=0

On the other hand, it is sufficient to apply Proposition 8.4.11 on the segment [0, t] with
t ∈ (−R, R) to obtain the indefinite integrals of t 7→ S(t).

This result can be generalized into the following one by a simple composition of C 1
functions.

Corollary 9.3.3. Let I ⊂ R be an interior-non-empty interval and φ be a C 1 -class function

mapping from I into D. Then, the function t 7→ S(φ(t)) is a C 1 -function and its differential
is given by t 7→ S ′ (φ(t))φ′ (t). Here, S ′ refers to the sum function of the power series
X
(n + 1)an+1 z n .
n
330 POWER SERIES

X I −→ C
Proof. We consider the series of functions fn where fn : for all
n t 7−→ an (φ(t))n
n ∈ N. Let J ⊂ I be a compact set. It follows from the continuity of φ that φ(J) ⊂ D
is a compact set. Then, there exists t0 ∈ J such that r := sup |φ(t)| = |φ(t0 )| < R. In
t∈J
the consequences, for all n ∈ N, sup |fn (t)| ≤ |an |rn . From Theorem 9.2.12, the series of
t∈J
functions converges normally, thus uniformly, on J. Also,
′
∀n ∈ N, sup |fn+1 (t)| ≤ (sup |φ′ (t)|)(n + 1)|an+1 |rn .
t∈J t∈J
X X
From the convergence of the series (n + 1)an+1 rn , the series of functions fn′ converges
n n
normally, thus uniformally, on J. So, it follows from Theorem 7.1.30 that the sum function
t 7→ S(φ(t)) is of C 1 -class function. And its differential is given by
+∞ +∞
d X X
∀t ∈ I, S(φ(t)) = fn′ (t) = φ′ (t) (n + 1)an+1 φ(t)n = S ′ (φ(t))φ′ (t).
dt
n=0 n=0

The following important result will be introduced in a more general context, the Complex
Analysis.
Proposition
X 9.3.4 (Cauchy’s Integral Formula). Let S be the sum function of the power
n
series an z with a radius of convergence R ∈ R+ . Let 0 < r < R and k ∈ N it holds
n
that
Z 2π
1
S(reiθ )e−ikθ dθ = ak rk . (9.1)
2π 0
X
Proof. Using Theorem 9.2.12, the power series an z n converges uniformally to the sum
n
function S on the compact set B(0, r). Consider now the functions gn : θ 7→ an rn einθ e−ikθ
defined on the segment [0, 2π] for all n ∈ N. Those functions are clearly continuous Xand
hence integrable on the segment [0, 2π]. We can deduce thus the series of functions gn
n
converges uniformally to the function θ 7→ S(reiθ )e−ikθ . By using Theorem 7.3.19, this
function θ 7→ S(reiθ )e−ikθ is also continuous on [0, 2π]. Also, applying Proposition 8.4.11,
we obtain that
Z 2π +∞ Z 2π
X +∞
X Z 2π
iθ −ikθ n
S(re )e dθ = gn (θ)dθ = an r ei(n−k)θ dθ = 2πak rk .
0 n=0 0 n=0 0

Hence, the claim follows.

Corollary 9.3.5 (Liouvillei ’s Theorem). Let S be the sum function of a power series with
its radius of convergence equals to +∞. We assume furthermore that S is bounded on C.
Then, S is a constant function.
i
Joseph Liouville (1809-1882), French mathematician. Several contributions on number theory, complex
analysis, differntial geometry, and partial differntial equations. He proved the existence of transcendent num-
bers. The Strum-Liouville Theorem can be applied in many physic model to resolve problems in Hamiltonian
mecanics.
9.4. CLASSICAL POWER SERIES 331

Proof. Since the radius of convergence equals to infinity, it follows from the Cauchy’s Inte-
gral formula Proposition 9.3.4 that for all k ∈ N and r > 0,
Z 2π
1
k
|ak |r = S(reiθ )e−ikθ dθ ≤ ∥S∥L∞ (C,C) .
2π 0

Here, ∥S∥L∞ (C,C) = sup |S(z)| is well-defined since S is supposed to be bounded on C. Thus,
z∈C
by passing r → +∞, we obtain that ak = 0 for all k ̸= 0. In conclusion, S is a constant
function.

9.4 Classical Power Series

9.4.1 Rational Fractions
P (z)
We recall here a rational fraction F ∈ C(X) refers to any function F := z 7→ where
Q(z)
P, Q ∈ C[X] are polynomials with coefficients in C such that P and Q have no common
factors. We call z0 ∈ C is a pole of the rational fraction F if z0 is a root of Q.

Proposition 9.4.1. The following power series have the radii of convergences equal to 1.
Moreover, let z ∈ C with |z| < 1, it holds that
+∞
1 X
• = zn;
1−z
n=0

+∞
!
1 X n+k−1 n
• for all k ∈ N∗ , = z .
(1 − z)k k − 1
n=0

Proof. • Classical geometric series, already seen previously.

!
n + k − 1
• Let k ∈ N∗ . For all n ∈ N, we define an = . It is immediate that the
k−1
sequence (an )n∈N is polynomial on n. So, it holds that |an z n | −→ 0 for |z| < 1 and
n→+∞
that |an z n | −→ +∞ for |z| > 1. We deduce that the radius of convergence of the
n→+∞
X
power series an z n equals to 1. Now, we prove the following identity,
n

+∞
!
1 X n+k−1 n
∀z ∈ C, |z| < 1, = z . (9.2)
(1 − z)k k − 1
n=0

The case k = 1 corresponds to classical geometric series. Next, we assume for certain
k ∈ N∗ , the!formula (9.2) holds. Denoting by S the sum function of the power series
X n+k
z n , we have thus, for all |z| < 1,
n k

+∞ +∞
! ! !!
X n+k n X n+k n−1+k
(1 − z)S(z) = (1 − z) z = − zn.
n=0
k n=0
k k
332 POWER SERIES
! ! !
n+k n−1+k n+k−1
By using the identity − = and the induction
k k k−1
1
assumption, we can thus deduce that (1 − z)S(z) = for all |z| < 1. Hence,
(1 − z)k
(9.2) holds for k + 1 and the claim follows by induction.

Theorem 9.4.2. Any rational fraction F ∈ C(X) of which 0 is not a pole can be written as
the sum function of a power series of which the radius of convergence equals to the minimum
of the magnitudes of its poles.
Proof. Let z1 , z2 , · · · , zp ∈ C \ {0} with p ∈ N∗ be the poles of the fraction F . Then we can
decompose F into simple elements,
α1 α2 αp
∀z ∈ C \ {z1 , z2 , · · · , zp }, F (z) = k
+ k
+ ··· + .
(z − z1 ) 1 (z − z2 ) 2 (z − zp )kp
Here, α1 , α2 , · · · , αp ∈ C and k1 , k2 , · · · , kp ∈ N∗ . Then, for all j ∈ [[1, p]] and all |z| < |zj |,
it follows from Proposition 9.4.1 that
+∞
! 
αj αj 1 αj X n + kj − 1 z n
= = .
(z − zj )kj (−zj )kj kj (−zj )kj n=0 kj − 1 zj
 
1 − zzj

Summing up the above equalities for all j ∈ [[1, p]], we can deduce that F can be written as
the sum function of a power series for all |z| < ρ := min{|z1 |, |z2 |, · · · , |zp |}. On the other
hand, since F diverges when z → zj∗ where zj∗ refers to the pole such that |zj∗ | = ρ, this
power series can not have a radius of convergence larger then ρ. Hence the claim follows.

9.4.2 Exponentials
From the comparison relation Rn = o(n!) for all R > 0, we deduce that the radius of
X zn
convergence of the power series is infinity. So, we can define the exponential function
n
n!
on C.
Definition 9.4.3. We call the complex exponential function, which is denoted by
X zn
exp : z 7→ ez , the sum function of the power series . That is,
n
n!
+∞ n
X z
∀z ∈ C, ez = . (9.3)
n!
n=0

Remark 9.4.4. • By applying Theorem 9.3.2, the exponential function exp satisfies the
differential equation y ′ = y on R with the initial condition y(0) = 1. This shows that
the definition using power series (9.3) provides a generalization into complex variables
of real exponentials, that we have learnt in Calculus.
• We can show that exp possesses an inverse function which equals to an indefinite
1
integral of x 7→ . This corresponds to what we have learnt in Calculus of the
x
definition of the natural logarithmic function ln.
9.4. CLASSICAL POWER SERIES 333

Proposition 9.4.5. For all z1 , z2 ∈ C, it holds that ez1 +z2 = ez1 ez2 .
Proof. Using Theorem 9.2.10 and the fact that exponentials has a radius of convergence
infinity, the product of two complex exponentials ez1 and ez2 can be obtained by the Cauchy
product of power series. Hence,
+∞ n
! +∞ ! +∞ n !
X z X zn X X z k z n−k
ez1 ez2 = 1 2
= 1 2
n! n! k! (n − k)!
n=0 n=0 n=0 k=0
+∞ n
! ! +∞
X 1 X n X (z1 + z2 )n
k n−k
= z1 z2 = = ez1 +z2 .
n! k n!
n=0 k=0 n=0

Proposition 9.4.6. For all z ∈ C, the following relations hold,

• ez ̸= 0; • |ez | = eRe(z) ;
1
• e−z = ; • |ez | = 1 ⇔ z ∈ iR.
ez
• ez = ez ;

Proof. Left in Exercises.

Theorem 9.4.7. The complex exponential z 7→ ez is a continuous surjective morphism

from the additive group (C, +) into the multiplicative group (C∗ , ×).
Proof. Proposition 9.4.5 shows that exp is a morphism mapping from (C, +) into (C∗ , ×).
Also, the continuity of exp on C follows from Theorem 9.3.1. In the next, we prove the
surjectivity of exp.
Let z0 ∈ C∗ .
• Case z0 ∈
/ R− . Introduce at first the parametric function
[0, 1] −→ C
f: .
t 7−→ (1 − t) + tz0

It is clear that f is a C 1 -class function. A simple calculation shows that f (t) = 0 ⇔

1
t= . Since z0 ∈
/ R− , we deduce that f does not vanish on [0, 1]. So, we can
1 − z0
consider the function ψ given by,
Z t ′
f (s)
∀t ∈ [0, 1], g(t) = ds.
0 f (s)

Clearly, this function g is also of class C 1 . In the consequence, the composed function
h : t 7→ f (t)e−g(t) is also C 1 class function. Applying Corollary 9.3.3, we have
f ′ (t) −g(t)
 
′ ′
∀t ∈ [0, 1], h (t) = f (t) − f (t) e = 0.
f (t)
It shows that h is a constant function on [0, 1]. Hence, h(1) = h(0) = 1. In the
consequences, z0 = f (1) = h(1)eg(1) . It proves thus the existence of g(1) ∈ C such
that z0 = eg(1) .
334 POWER SERIES

• Case z0 ∈ R∗− . Then we may set t = ln |z0 | ∈ R. Also, from the previous point, there
exists θ ∈ C such that i = eθ . Hence, z0 = et+2θ .

Theorem 9.4.8. The function Φ : t 7→ eit is a continuous surjective morphism from

the additive group (R, +) into the multiplicative group (U, ×) where U := {z ∈ C | |z| = 1}
denotes the unitary complex numbers. Moreover, ker(Φ) = aZ with a > 0.
Proof. Theorem 9.4.7 shows that Φ is a continuous morphism mapping from (R, +) into
(C∗ , ×). In addition, Proposition 9.4.6 shows that Φ(R) ⊂ U . For the surjectivity, we
consider z0 ∈ U . Theorem 9.4.7 shows that there exists z ∈ C such that ez = z0 . Also,
since |z0 | = 1, Proposition 9.4.6 shows that z ∈ iR. Hence, Φ is a continuous surjective
morphism mapping from (R, +) into (U, ×).
In the consequences, ker(Φ) := t ∈ R eit = 1 is a subgroup in (R, +). From the
Theorem of structure of subgroups in the additive group of real numbers (the exercise we
have done in the Chapter of Real Numbers), that ker(Φ) is either discret or dense in R. If
ker(Φ) is dense in R, then Φ(t) = 0 for all t ∈ R from the continuity of Φ. We deduce then
there exists a > 0 such that ker(Φ) = aZ.

Now, we can define in a purely analytical way the constant π.

Definition 9.4.9. We denote by π > 0 the strictly positive constant
a
π :=
2
where a refers to the constant such that ker(Φ) = aZ in Theorem 9.4.8. Equivalently
speaking, we have
1
min t > 0 eit = 1 .

π :=
2

Corollary 9.4.10. Let t ∈ R. Then eit = 1 ⇔ t ∈ 2πZ.

Proof. Direct consequence of Theorem 9.4.8.

Proposition 9.4.11. There exists a unique continuous function φ defined on C \ R− such

that φ(1) = 0 and
∀z ∈ C \ R− , eφ(z) = z.
Proof. Taking the proof of Theorem 9.4.7, we consider the function φ given by,
Z 1
z−1
∀z ∈ C \ R− , φ(z) = ds.
0 (1 − s) + sz

Hence, it holds that eφ(z) = z for all z ∈ C \ R− . Now we prove the continuity of φ.
z−1
Let R > 0. The function (z, s) 7→ is continuous on {z ∈ C | |z| < R} × [0, 1].
(1 − s) + sz
Moreover,
z−1
∀(z, s) ∈ {z ∈ C | |z| < R} × [0, 1], ≤ R + 1.
(1 − s) + sz
9.4. CLASSICAL POWER SERIES 335

The integrability of the constant function s 7→ R + 1 and the Theorem of continuity under
integral signs Theorem 8.4.14 imply the continuity of the function φ on {z | |z| < R}. Since
R is chosen arbitrary, we conclude that φ is continuous on its definition domain C \ R− .
Let φ1 , φ2 be continuous functions such that for all z ∈ C \ R− , z = eφ1 (z) = eφ2 (z) .
Applying Proposition 9.4.5 and Corollary 9.4.10, we have φ1 (z) − φ2 (z) ∈ 2πiZ. Due to the
set C \ R− is connected, there exists m ∈ Z such that ∀z ∈ C \ R− , φ1 (z) − φ2 (z) = 2πim.
Then, the condition φ(1) = 0 implies the uniqueness of such function.

Definition 9.4.12. We call the function φ in Proposition 9.4.11 as the principal deter-
mination of logarithmic, which coincides with ln on R∗+ . We denote also,

C \ R− −→ C
log : .
z 7−→ log(z) := φ(z)

9.4.3 Trigonometric and Hyperbolic Functions

Definition 9.4.13. We call the complex sinus and cosinus function the following func-
tions mapping from C into C.
+∞
eiz − e−iz X z 2n+1
∀z ∈ C, sin z = = (−1)n ,
2i (2n + 1)!
n=0
+∞
eiz + e−iz X z 2n
cos z = = (−1)n .
2 (2n)!
n=0

Definition 9.4.14. We call the complex hyperbolic sinus and cosinus function the
following functions mapping from C into C.
+∞
ez − e−z X z 2n+1
∀z ∈ C, sinh z = = ,
2 (2n + 1)!
n=0
+∞
ez + e−z X z 2n
cosh z = = .
2 (2n)!
n=0

The following results can be obtained by manipulating power series.

Proposition 9.4.15. For all z ∈ C,

1.) cos(iz) = cosh(z); 6.) e−z = cosh(z) − sinh(z);

2.) cosh(iz) = cos(z); 7.) eiz = cos(z) + i sin(z);

3.) sin(iz) = i sinh(z); 8.) e−iz = cos(z) − i sin(z);

4.) sinh(iz) = i sin(z); 9.) cos2 (z) + sin2 (z) = 1;

5.) ez = cosh(z) + sinh(z); 10.) cosh2 (z) − sinh2 (z) = 1.

Proof. Left in Exercises.

336 POWER SERIES

Proposition 9.4.16. For all a, b ∈ C.

1.) cos(a + b) = cos(a) cos(b) − sin(a) sin(b);
2.) sin(a + b) = cos(a) sin(b) + sin(a) cos(b);
3.) cosh(a + b) = cosh(a) cosh(b) + sinh(a) sinh(b);
4.) sinh(a + b) = cosh(a) sinh(b) + sinh(a) cosh(b).
Proof. Left in Exercises.

9.5 Analytic Functions

Definition 9.5.1.
X We say a function f : A ⊂ R → C is analytic at 0 if there exists a
power series an tn with the radius of convergence R > 0 as well as a number r ∈ (0, R)
n
such that (−r, r) ⊂ A and
+∞
X
∀t ∈ (−r, r), f (t) = an tn .
n=0

Definition 9.5.2. We say the function f is analytic at t0 ∈ R if the function t 7→ f (t − t0 )

is analytic at 0.

Remark 9.5.3. • The property “analytic” describes the local behaviors of functions.
So, if g coincides with an analytic function f in a neighborhood of t0 , then g is also
analytic at t0 .
• If the function f is analytic at each points in R, we say f is an entire analytic
function.
• From what we have introduced previously, the functions such as exponentials, trigono-
metric, hyperbolic or logarithmic functions (here, we mean t 7→ ln(1 + t)) are analytic
functions at 0.

Proposition 9.5.4. Let f be an analytic function at t0 ∈ R. Then, there exists r > 0 such
that f is C ∞ -class on (t0 − r, t0 + r). Moreover, the power series which expands f is given
by the Taylor series,
+∞ (n)
X f (t0 )
∀t ∈ (t0 − r, t0 + r), f (t) = (t − t0 )n .
n!
n=0
X
Proof. Let an (t − t0 )n be the power series which expands f . Denoting by R > 0 its
n
radius of convergence, then, there exists 0 < r < R such that
+∞
X
∀t ∈ (t0 − r, t0 + r), f (t) = an (t − t0 )n .
n=0

f (n) (t0 )
Using Theorem 9.3.2, f is C ∞ on (t0 − r, t0 + r) and for all n ∈ N, an = . Hence,
n!
the claim follows.
9.5. ANALYTIC FUNCTIONS 337

Remark 9.5.5. 1.) The previous proposition states also that the power series which ex-
pands a analytic function is uniquely given by Taylor series.

2.) There exists the C ∞ functions which is not analytic. Here is an example,
1
(
e− t2 if t ̸= 0,
f (t) =
0 if t = 0.

We can show that f is C ∞ on R and each derivatives are all zero at 0. If f is analytic
at 0, then f equals to null function in a neighborhood of 0, which is contradictory to
its definition.
X
Proposition 9.5.6. Let f be an analytic function at 0 and an tn be its expansion power
n
series. Then,
• if f is even, then ∀n ∈ N, a2n+1 = 0;

• if f is odd, then ∀n ∈ N, a2n = 0.

Proof. Left in Exercises.

Proposition 9.5.7. Let f be a C ∞ function on the interval (−a, a) with a > 0. We assume
that there exists ρ, M > 0 such that
M n!
∀t ∈ (−a, a), ∀n ∈ N, |f (n) (t)| ≤ .
ρn
Then, f is analytic at 0 with a radius of convergence R ≥ min{a, ρ}.
Proof. It follows from Taylor’s expansion of f at 0 that
n t
f (k) (0) f (n+1) (t − s) n
X Z
k
∀t ∈ (−a, a), ∀n ∈ N, f (t) = t + s ds
k! 0 n!
k=0

By using the given estimation, let t ∈ [−η, η] with 0 < η < min{a, ρ}, we have
t  n+1
f (n+1) (t − s) n η n+1
Z
η
s ds ≤ sup |f (n+1) (s)| ≤M −→ 0.
0 n! s∈[−η,η] (n + 1)! ρ n→+∞

X f (n) (0)
This estimation shows that the rests to the power series tn converges unifor-
n
n!
mally on any compact in (− min{a, ρ}, min{a, ρ}). From Proposition 7.3.4, the power series
X f (n) (0)
tn has a radius of convergence larger then min{a, ρ}. Hence, the claim follows
n
n!
via Proposition 9.5.4.

Corollary 9.5.8 (Bernstein’s Theorem). Let f be a C ∞ function on the interval (−a, a)

with a > 0. We assume that for all n ∈ N, f (n) is positive on (−a, a). Then, f is analytic
at 0.
338 POWER SERIES

a
Proof. We choose ρ > 0 with ρ < . Let t ∈ [−ρ, ρ]. It follows also from Taylor’s expansion
2
of f at t ∈ (−a, a) that for all n ∈ N,
n t+ρ
f (k) (t) f (n+1) (2t + ρ − s) n
X Z
k
f (t + ρ) = ρ + s ds.
k! t n!
k=0

Since f and all its derivatives are all positive on (−a, a), the right-hand-side above is thus
positive. We have then,
ρn (n)
0≤ f (t) ≤ f (t + ρ) ≤ sup |f (x)|.
n! x∈[−2ρ,2ρ]

Applying Proposition 9.5.7, the claim follows.

 Chapter 10

Fourier Series

In this chapter, we consider only the functions mapping from R into C.

10.1 Periodic Functions

10.1.1 Definitions and General Properties
Definition 10.1.1. Let f : R → C be a function.
• We call a period of f any p ∈ R such that ∀t ∈ R, f (t + p) = f (t).
• We say f is periodic if f possesses a period p ∈ R∗+ . Precisely speaking, we say in
this case f is p-periodic.

Proposition 10.1.2. Let f be a periodic function.

• If p is a period of f then for all n ∈ Z, np is also a period of f .
• The set of periods of f forms a subgroup in (R, +).
• Let G be a subgroup in (R, +), then either G = αZ, α ∈ R∗+ , or G is dense in R.
(Theorem of Structure of Subgroups in (R, +))
• f is completely determined by f |[a,a+p) for any a ∈ R.
• Let p ∈ R∗+ , the set of p-periodic and continuous functions, which is denote by
Cp (R, C), forms a sub-algebra in the algebra of continuous functions C(R, C).
• The previous statement holds also for p-periodic and piecewisely continuous functions.
We denote by PC p (R, C) the algebra of such functions.
• Any periodic piecewise continuous function is bounded on R.
• Any periodic continuous function is uniformly continuous on R.
• Let p ∈ R∗+ , a ∈ R and g ∈ PC([a, a + p], C). If g(a + p) = g(a), then there exists a
unique p-periodic piecewise continuous function f ∈ PC p (R, C) such that f |[a,a+p] = g.
In fact, f is given by,
  
t−a
∀t ∈ R, f (t) = g p+a ,
p
where {x} = x − [x] denotes the fraction part of the real number x ∈ R.

339
340 FOURIER SERIES

Proof. Left in Exercises.

In the rest of this chapter, we consider only the 2π-periodic functions. In the case where
it is necessary to consider anarbitrary
 period p, it is sufficient to introduce the function g
2π
defined by, ∀t ∈ R, g(t) = f t .
p

Definition 10.1.3. Let f ∈ PC 2π (R, C). We say f satisfies the Dirichleti condition at
t ∈ R if
1 1
f (t) = (f (t+ ) + f (t− )) = ( lim f (s) + lim f (s)).
2 2 s→t+ s→t−

We denote by D2π (R, C) the set of 2π-periodic piecewise continuous functions satisfying the
Dirichlet condition on R.

We remark that this condition is automatically verified on all continuous points. Also,
the set D2π (R, C) forms a sub-algebra in PC 2π (R, C). For any function f ∈ PC 2π (R, C),
there exists a unique function f˜ ∈ D2π (R, C) which coincides with f on all points of conti-
nuity.

Proposition 10.1.4. Let f ∈ PC 2π (R, C) and a ∈ R, we have

Z Z
f= f.
[a,a+2π] [0,2π]

Proof. From the periodicity of f , we have, for all x ∈ [0, a], f (x) = f (x + 2π). So,
Z a Z a Z a+2π
f (t)dt = f (t + 2π)dt = f (t)dt.
0 0 2π

And it follows,
Z a+2π Z 0 Z 2π Z a+2π
f (t)dt = f (t)dt + f (t)dt + f (t)dt
a a 0 2π
Z 0 Z 2π Z a
= f (t)dt + f (t)dt + f (t)dt
a 0 0
Z 2π
= f (t)dt.
0

10.1.2 Inner Product and Square Integrable Semi-Norm

Proposition 10.1.5. Let f, g ∈ PC 2π (R, C), we define,
Z
1 1
⟨f, g⟩ := fg = ⟨f, g⟩L2 ([0,2π],C) .
2π [0,2π] 2π

Then, ⟨·, ·⟩ is a sesquilinear, Hermitian and positive form on the vector space PC 2π (R, C).
i
Peter Gustav Lejeune Dirichlet (1805-1859), German mathematician. He contributed in analytical
number theory, Fourier series, potential theory and boundary value problems.
10.1. PERIODIC FUNCTIONS 341

Proof. Left in Exercises.

Proposition 10.1.6. A function f ∈ PC 2π (R, C) satisfies ⟨f, f ⟩ = 0 if, and only if f = 0

on R almost everywhere.
Proof. By 2π-periodicity of f , it is sufficient to prove the statement on the segment [0, 2π].
Since f is piecewise continuous on [0, 2π], we denote by (x0 , x1 , · · · , xn ) an adapted partition
of f on [0, 2π]. Also, we denote by fi = f |(xi−1 ,xi ) ∈ C((xi−1 , xi ), C) for all i ∈ [[1, n]]. We
have then,
Z n Z
1 2 1 X
⟨f, f ⟩ = |f | = |fi |2 .
2π [0,2π] 2π (xi−1 ,xi )
i=1

Assuming now ⟨f, f ⟩ = 0, it follows from the previous inequality that, for all j ∈ [[1, n]],
Z n Z
X
2
0≤ |fj | ≤ |fi |2 = ⟨f, f ⟩ = 0.
(xj−1 ,xj ) i=1 (xi−1 ,xi )

Since fi is continuous on (xi−1 , xi ), we deduce that fi = 0 on (xi−1 , xi ). Hence, f = 0

on [0, 2π] except for the points x0 , x1 , · · · , xn . Then using the 2π-periodicity, f = 0 on R
almost everywhere.
The reciprocal implication is straightforward from calculation.

Definition 10.1.7. Let f ∈ PC 2π (R, C), we call the square integrable semi-norm of f
the positive real number,
p 1
∥f ∥2 := ⟨f, f ⟩ = √ ∥f ∥L2 ([0,2π],C) .
2π

Remark 10.1.8. Although ⟨·, ·⟩ and ∥·∥2 don’t satisfy the separation property in PC 2π (R, C),
we will call them in what follows respectively “inner product” and “semi-norm”.
• We can show, the complete proofs are left in Exercises, that the Cauchy-Schwarz
inequality holds in PC 2π (R, C) endowed with ⟨·, ·⟩ and ∥ · ∥2 .
• We can also define the orthogonal relations in PC 2π (R, C). Let f, g ∈ PC 2π (R, C), we
say f and g are orthogonal if ⟨f, g⟩ = 0. We denote f ⊥ g.
• The Pythagoras Theorem holds on PC 2π (R, C). Let f, g ∈ PC 2π (R, C) such that f ⊥ g
then,
∥f + g∥22 = ∥f ∥22 + ∥g∥22 . (10.1)

• If, there are sufficiently conditions to determinate the values on the discontinuous
points, ⟨·, ·⟩ and ∥ · ∥2 could define effectively an inner product and a norm.

Corollary 10.1.9. (C2π (R, C), ⟨·, ·⟩) and (D2π (R, C), ⟨·, ·⟩) are inner product spaces.
Proof. It is sufficient to prove the separation property. It is also left in Exercises.

Remark 10.1.10. Using Cauchy-Schwarz inequality, we can show that,

1 1
∀f ∈ PC 2π (R, C), ∥f ∥L1 ([0,2π],C) ≤ ∥f ∥2 = √ ∥f ∥L2 ([0,2π],C) ≤ ∥f ∥L∞ (R,C) .
2π 2π
342 FOURIER SERIES

10.1.3 Trigonometric Polynomials and Series

Definition 10.1.11. Let n ∈ Z, we denote by en the function defined by,
∀x ∈ R, en (x) = einx = cos(nx) + i sin(nx).
∞ (R, C).
We have immediately that en ∈ C2π

Proposition 10.1.12. The family (en )n∈Z is orthonormal in (C2π (R, C), ⟨·, ·⟩).
Proof. Let n ∈ Z, we have,
Z Z 2π
1 1
⟨en , en ⟩ = 2
|en | = e−int eint dt = 1.
2π [0,2π] 2π 0

Also, for all n, m ∈ Z, n ̸= m, we have,

Z 2π
1 ei(m−n)2π − e0
Z
1 1
⟨en , em ⟩ = en em = e−nt emt dt = = 0.
2π [0,2π] 2π 0 2π m−n

In conclusion, ⟨en , em ⟩ = δn,m ii and hence the claim follows.

Definition 10.1.13. We call a trigonometric polynomial of degree N ∈ N any lin-

ear combination of the family (en )n∈[[−N ,N ]] . We denote by PN the set of trigonometric
polynomial functions of degree N .
As an immediate result, for all P ∈ PN , there exists a unique sequence (cn )n∈[[−N ,N ]] ∈
C2N +1 such that
N
X
P = cn en .
n=−N

It is also straightforward form Proposition 10.1.12 that

Z 2π
1
∀n ∈ [[−N , N ]], cn = ⟨en , P ⟩ = e−int P (t)dt.
2π 0
Moreover, we can use trigonometric functions to express the function P . There exists also
unique sequences (an )n∈[[0,N ]] and (bn )n∈[[1,N ]] such thatiii ,
N
a0 X
∀t ∈ R, P (t) = + an cos(nt) + bn sin(nt).
2
n=1

The coefficients (an )n∈[[0,N ]] and (bn )n∈[[0,N ]] can be calculated directly from the inner prod-
ucts. Let n ∈ [[0, N ]],
1 2π
Z
an = 2⟨cos(nt), P ⟩ = cos(nt)P (t)dt;
π 0
1 2π
Z
bn = 2⟨sin(nt), P ⟩ = sin(nt)P (t)dt.
π 0
ii
We recall here that δn,m denotes the Kronecker symbol, which is defined by, for all n, m ∈ Z δn,m = 1
if n = m and δn,m = 0 if n ̸= m.
iii
In this chapter, we will abusively use the notations cos(nt) and sin(nt) to express the functions t 7→
cos(nt) and t 7→ sin(nt).
10.1. PERIODIC FUNCTIONS 343

We impose systematically that b0 = 0. Also, by identifying the real and imaginary parts of
the coefficients (cn )n∈[[−N ,N ]] , we have, for all n ∈ [[0, N ]],

1 1
cn = (an − ibn ), c−n = (an + ibn );
2 2
an = cn + c−n , bn = i(cn − c−n ).

Proposition 10.1.14. Let P ∈ PN , it follows,

N N
X 1 1X
∥P ∥22 = |cn | = |a0 |2 +
2
(|an |2 + |bn |2 ).
4 2
n=−N n=1

Proof. Left in Exercises.

Definition 10.1.15. We call a trigonometric series any series of functions given by

X X
cn en = c0 + (cn en + c−n e−n ),
n n≥1

where (cn )n∈Z is a sequence of complex numbers indexed by Z.

Here, we introduce the series indexed by Z. In aX

general manner, let (x
Xn )n∈Z be a family
of elements in a normed space. We denote also by xn the series x0 + (x−n + xn ). Its
n n≥1
sequence of partial sums (SN )N ∈N is therefore given by,
N
X
∀N ∈ N, SN = x0 + (x−n + xn ).
n=1
X
We say the series xn converges if the sequence of partial sums converges, which means
n
N
!
X
the sequence xn converges.
n=−N N ∈N

X
Theorem 10.1.16. The trigonometric series cn en is normally convergent on R if, and
X n
only if the series |cn | of complex numbers indexed by Z converges.
n
X
Proof. Recalling that the normal convergence of a series of function fn means the con-
X n
vergence of the numerical series ∥fn ∥L∞ (R,C) where ∥ · ∥L∞ (R,C) denotes the norm of
n
uniform convergence.
Let n ∈ N∗ , it is straightforward that for all x ∈ R,

|c−n e−inx + cn einx | ≤ |c−n | + |cn |.

344 FOURIER SERIES

1
By writing cn = |cn |eiθn with θn ∈ [0, 2π), we can choose x0 = (θ−n − θn ) ∈ R to obtain
2n
that

|c−n e−inx0 + cn einx0 | = ||c−n |eiθ−n e−inx0 + |cn |eiθn einx0 |

i
= |(|c−n | + |cn |)e 2 (θ−n +θn ) | = |c−n | + |cn |.

This implies that ∥c−n e−n + cn en ∥L∞ (R,C) = |c−n | + |cn |. Hence, the normal convergence of
X X
the series cn en is equivalent to the convergence of the numerical series |cn |.
n n

Remark 10.1.17. Using the functions sin and cos, the previous theorem can be also formulate
as follows.
a0 X
The trigonometric series + (an cos(nt) + bn sin(nt)) is normally convergent on R if,
2
n≥1
X
and only if the series (|an | + |bn |) converges.
n

10.2 Fourier Coefficients and Series

In this section, we consider a function f ∈ PC 2π (R, C).

10.2.1 Definitions
Definition 10.2.1. We call the exponential Fourieriv coefficients of a function f ∈
PC 2π (R, C) the sequence (cn (f ))n∈Z indexed by Z given by,
Z 2π
1
∀n ∈ Z, cn (f ) = ⟨en , f ⟩ = f (t)e−int dt.
2π 0
Also, we call the trigonometric Fourier coefficients of f the sequences (an (f ))n∈N and
(bn (f ))n∈N given by, for all n ∈ N,

1 2π
Z
an (f ) = 2⟨cos(nt), f ⟩ = f (t) cos(nt)dt;
π 0
1 2π
Z
bn (f ) = 2⟨sin(nt), f ⟩ = f (t) sin(nt)dt.
π 0

Remark 10.2.2. • It always holds that b0 (f ) = 0. And for all n ∈ N,

1 1
cn (f ) = (an (f ) − ibn (f )), c−n (f ) = (an (f ) + ibn (f ));
2 2
an (f ) = cn (f ) + c−n (f ), bn (f ) = i(cn (f ) − c−n (f )).

• It follows from Proposition 10.1.4 that the Fourier coefficients cn (f ), an (f ) or bn (f )

can be calculated by the integrals on any segment with a length equals to 2π.

• For all n ∈ Z the mapping cn : f 7→ cn (f ) defines a linear form on PC 2π (R, C).

iv
Jean Baptiste Joseph Fourier (1768-1830), French mathematician and physicist. His Fourier Series and
Fourier Transforms are fundamental tools in modern engineering.
10.2. FOURIER COEFFICIENTS AND SERIES 345

• We denote by fb the mapping n 7→ cn (f ). In the consequences, the application f 7→ fb

is linear which maps from PC 2π (R, C) into the space of sequences of complex numbers
indexed by Z.

Definition 10.2.3. We call the Fourier series of the function f ∈ PC 2π (R, C), the fol-
lowing trigonometric series
X a0 (f ) X
S(f ) := cn (f )en = + (an (f ) cos(nt) + bn (f ) sin(nt)) .
n
2 ∗
n∈N

Let N ∈ N, we call the Fourier sum of order N of f , and denote it by SN (f ), the N -th
term in the sequence of partial sums in the Fourier series of f . That is,
N N
X
int a0 (f ) X
SN (f )(t) = cn (f )e = + (an (f ) cos(nt) + bn (f ) sin(nt)) .
2
n=−N n=1

10.2.2 Properties of Fourier Coefficients

Proposition 10.2.4. Let n ∈ Z. The following assertions hold.
1
• |cn (f )| ≤ ∥f ∥2 , |cn (f )| ≤ ∥f ∥L1 ([0,2π],C) and |cn (f )| ≤ ∥f ∥L∞ (R,C) .
2π
• cn (f ) = c−n (f ).

• If f is valued in R, then c−n (f ) = cn (f ). Also, an (f ) and bn (f ) are real numbers.

b b b
• The flip of
b f , denoted by f , is defined by, ∀t ∈ R, f (t) = f (−t). Then, f ∈ PC 2π (R, C)
and c−n (f ) = cn (f ).

• If f is an odd function, then c−n (f ) = −cn (f ) as well as an (f ) = 0.

• If f is an even function, then c−n (f ) = cn (f ) as well as bn (f ) = 0.

• Let a ∈ R we denote by τa the translation operation on functions defined on R. That

means, τa f is given by, ∀x ∈ R, τa f (x) = f (x + a). Then, τa f ∈ PC 2π (R, C) and
cn (τa f ) = eina cn (f ).

Proof. Left in Exercises.

In the next Proposition 10.2.6, we present the relation between the Fourier coefficients
of a periodic function and those of its “derivative”. To do so, we introduce the following
concept of piecewisely C 1 functions and pseudo-derivatives.

Definition 10.2.5. Let f be a function defined on a segment [a, b] ⊂ R and valued in a

normed space F .

1.) We say f is a piecewisely C 1 -class function on [a, b] if there exists a partition

(x0 , x1 , · · · , xn ) on [a, b] such that for all j ∈ [[1, n]], the restriction f |(xj−1 ,xj ) is a
d
C 1 -class function and its derivative f |(xj−1 ,xj ) possesses a left-limit at xj−1 and a
dt
right-limit at xj respectively.
346 FOURIER SERIES

2.) The space of piecewisely C 1 -class is denoted by PC 1 ([a, b], F ).

3.) We call the pseudo-derivative of f ∈ PC 1 ([a, b], F ), and denoted it also by f ′ , the
piecewisely continuous function given by,
d
∀j ∈ [[1, n]], ∀x ∈ (xj−1 , xj ), f ′ (x) = f| (x).
dt (xj−1 ,xj )
And the values on x0 , x1 , · · · , xn are attributed arbitrary.v

Proposition 10.2.6. If f ∈ PC 12π (R, C) being continuous on R, then for all n ∈ Z,

cn (f ′ ) = incn (f ).

Proof. Since f ∈ PC 12π (R, C), we denote by (0 = x0 , x1 , · · · , xm = 2π), m ∈ N∗ , an adapted

partition of [0, 2π] such that f is continuously differentiable on each subintervals (xj−1 , xj )
for all j ∈ [[1, m]]. Let n ∈ Z, we calculate,
m Z
′ 1 X xj ′
cn (f ) = f (t)e−int dt.
2π xj−1
j=1

Also, using the integration by parts, it follows that

Z xj Z xj
′ −int − −inxj −inxj−1
f (t)e dt = f (xj )e +
− f (xj−1 )e + in f (t)e−int dt.
xj−1 xj−1

−
It follows from the continuity of f that for all j ∈ [[1, m]], f (x+
j ) = f (xj ) = f (xj ). Summing
all of them up, we have,
m
in xj
Z
′ 1 X −inxj −inxj−1
cn (f ) = (f (xj )e − f (xj−1 )e )+ f (t)e−int dt
2π 2π xj−1
j=1

in 2π
Z
1 −2πin −0in
= (f (2π)e − f (0)e )+ f (t)e−int dt.
2π 2π 0

Using the 2π-periodicity, f (2π) = f (0), the claim follows.

Corollary 10.2.7. If f is piecewisely C k -class and 2π-periodic function on R with k ∈ N∗ ,

we have, for all n ∈ Z,

cn (f (k) ) = (in)k cn (f ).

Theorem 10.2.8. Let (cn )n∈Z ∈ CZ X be a family of complex numbers indexed by Z. We

suppose that the trigonometric series cn en converges uniformly on R and we denote by
n
f the sum function of this series of functions. Then, we have

∀n ∈ Z, cn (f ) = cn .
v
We will see in the proof that in fact, those values on the discontinuous points do not effect the conclusion.
10.2. FOURIER COEFFICIENTS AND SERIES 347
X
Proof. Let m ∈ Z. Since the series of functions cn en converges uniformly on R to the
n X
function f and the function e−m is bounded on R, the series of functions cn en e−m
n
converges uniformly to the function f e−m on R. We remark also that for all n ∈ Z the
function cn en em is continuous on R. We then apply Theorem 8.1.44 to obtain that
Z 2π Z Z +∞
1 1 1 X
cm (f ) = f (t)e−imt dt = f e−m = cn en e−m
2π 0 2π [0,2π] 2π [0,2π] n=−∞
+∞ Z +∞ Z 2π
1 X 1 X
= cn en e−m = cn ei(n−m)t dt = cm
2π n=−∞ [0,2π] 2π n=−∞ 0

To obtain the least equality, we have used Proposition 10.1.12.

Proposition 10.2.9. Let f ∈ PC 2π (R, C) and N ∈ N. Then, the Fourier sum of order N ,
SN (f ) is the unique trigonometric polynomial p ∈ PN such that f − p is orthogonal to the
space PN .

Proof. For all n ∈ [[−N , N ]], we have, ⟨en , SN (f )⟩ = cn (f ). This implies that

⟨en , f − SN (f )⟩ = ⟨en , f ⟩ − ⟨en , SN (f )⟩ = cn (f ) − cn (f ) = 0.

⊥.
In other words, we have the relation f − SN (f ) ∈ PN
⊥ . From the
We now prove the uniqueness result. Let p ∈ PN such that f − p ∈ PN
previous result, we have that
⊥
SN (f ) − p = (f − p) − (f − SN (f )) ∈ PN .

⊥ = {0}. Hence, p = S (f ).
Thus, SN (f ) − p ∈ PN ∩ PN N

Corollary 10.2.10. Let f ∈ C2π (R, C) or D2π (R, C) and N ∈ N. Then, the Fourier sum
of order N , SN (f ) is the orthogonal projection of f on the vector subspace PN .

Theorem 10.2.11 (Besselvi ’s Inequality). Let f ∈ PC 2π (R, C) and N ∈ N, it holds,

N
X
|cn (f )|2 ≤ ∥f ∥22 .
n=−N

⊥ and h ∈ P ⊥ .
Proof. It follows from Proposition 10.2.9 that f = SN (f )+h with SN (f ) ∈ PN N
Applying Pythagoras theorem (10.1), we have,

∥f ∥22 = ∥SN (f )∥22 + ∥h∥22 .

vi
Friedrich Wilhelm Bessel (1784-1846), German mathematician and astronomer.
348 FOURIER SERIES

On the other hand, using Proposition 10.1.12, we obtain that

* N N
+
X X
2
∥SN (f )∥2 = ⟨SN (f ), SN (f )⟩ = cn (f )en , cm (f )em
n=−N m=−N
N
X N
X N
X N
X
= cn (f )cm (f )⟨en , em ⟩ = cn (f )cm (f )δn,m
n=−N m=−N n=−N m=−N
N
X
= |cn (f )|2 .
n=−N

Thus,
N
X
|cn (f )|2 = ∥SN (f )∥22 = ∥f ∥22 − ∥h∥22 ≤ ∥f ∥22 .
n=−N

Corollary 10.2.12. Let f ∈ PC 2π (R, C), then the series

X 1 1X
|cn (f )|2 and |a0 (f )|2 + |an (f )|2 + |bn (f )|2


n
4 2 n
are convergent.
Proof.
X From Bessel’s inequality Theorem 10.2.11, the sequence of partial sums of the series
|cn (f )|2 is increasing and bounded by ∥f ∥22 . Hence, it is convergent. The trigonometric
n
series follows exactly the same argument.
Equivalently speaking, Corollary 10.2.12 shows that the Fourier coefficients of f is square
summable. Also, the application f 7→ fb is a linear which maps from PC 2π (R, C) into the
space of square summable sequences ℓ2 (Z, C).

Lemma 10.2.13 (Riemann-Lebesgue). Let f ∈ PC 2π (R, C), then cn (f ) −→ 0. Equiva-

n→±∞
lently speaking, let [α, β] be a segment such that f is continuous on it, we have,
Z β Z β
f (t) cos(nt)dt −→ 0, and f (t) sin(nt)dt −→ 0.
α n→+∞ α n→+∞
X
Proof. It is a direct consequence of the convergence of the series |cn (f )|2 . An alternative
n
proof relies on Stone-Weierstrass Theorem 7.2.16. The complete proof by this method is
left in Exercise 7.4.16.

Corollary 10.2.14. If f is piecewise C k -class, then cn (f ) = o(|n|−k ) as well as an (f ) =

o(n−k ) and bn (f ) = o(n−k ).
Proof. From Corollary 10.2.7, we have that, for all n ∈ Z∗ ,
1
|cn (f )| = |cn (f (k) )|.
|n|k
Since f (k) ∈ PC 2π (R, C), it follows from Riemann-Lebesgue lemma that cn (f (k) ) −→ 0.
n→+∞
So, the claim follows.
10.3. CONVERGENCE THEOREMS 349

10.3 Convergence Theorems

10.3.1 Pointwise Convergence
Proposition 10.3.1. Let N ∈ N. We call the Dirichlet’s kernel of degree N the following
trigonometric polynomial
N
X
DN := en .
n=−N

Then, DN satisfies,
Z
1
• DN = 1;
2π [0,2π]

sin(N + 12 )t
• ∀t ∈ R \ 2πZ, DN (t) = . And DN (t) = 2N + 1 for all t ∈ 2πZ.
sin 2t
Proof. Left in Exercise.

Proposition 10.3.2. Let f ∈ PC 2π (R, C) and N ∈ N, we have, for all t ∈ R,

Z 2π
1
SN (f )(t) = f (t − u)DN (u)du.
2π 0
Proof. It is straightforward from the calculation and a change of variable u := t − s,
N N  Z 2π 
X
int
X 1 −ins
SN (f )(t) = cn (f )e = f (s)e ds eint
2π 0
n=−N n=−N
2π N t−2π
−1
Z Z
1 X
in(t−s)
= f (s) e ds = f (t − u)DN (u)du
2π 0 2π t
n=−N
Z 2π
1
= f (t − u)DN (u)du.
2π 0

Theorem 10.3.3 (Dirichlet). Let f be a piecewise continuous, 2π-periodic function and

t0 ∈ R. We suppose that the quantities

f (t0 − t) − f (t−
0) f (t0 + t) − f (t+
0)
and
t t
have the limits when t → 0+ . Then, the convergence holds,
1
f (t− +


SN (f )(t0 ) −→ 0 ) + f (t0 ) .
N →+∞ 2
Proof. We introduce the following notations,
1
• l0 := f (t− +


0 ) + f (t0 ) ;
2
350 FOURIER SERIES

• g is the function defined by ∀t ∈ R, g(t) = f (t + t0 ), i.e. g = τt0 f ;

1 1
b
• u is the function defined by ∀t ∈ R, u(t) = (g(t) + g(−t)), i.e. u = (g + g); we
2 2
remark that u is an even function;
1 1
b
• v is the function defined by ∀t ∈ R, v(t) = (g(t) − g(−t)), i.e. v = (g − g); we
2 2
remark that v is an odd function.
Applying Proposition 10.2.4, we have, for all N ∈ N,
N b !
1 1
b
X
SN (u)(0) = SN (g) + SN (g) (0) = (cn (g) + cn (g))
2 2
n=−N
N N N
!
1 X X X
= (cn (g) + c−n (g)) = cn (g) = cn (τt0 f )
2
n=−N n=−N n=−N
N
X
= cn (f )eint0 = SN (f )(t0 ).
n=−N

We define moreover the function w by,

(
u(t) if t ∈ R \ 2πZ
w(t) =
l0 if t ∈ 2πZ.

Then, the function w is piecewise continuous, 2π-periodic functions on R and it holds,

1 1
lim w(t) = lim (g(t) + g(−t)) = lim (f (t + t0 ) + f (−t + t0 ))
t→0+ t→0+ 2 t→0+ 2
!
1 1
f (t− ) + f (t+


= lim f (t) + lim f (t) = 0 0 ) = l0 .
2 t→t+0 t→t−
0
2

We have the same limit when t → 0− . By 2π-periodicity, we deduce that w is continuous

in the neighborhoods of 2πZ. Similarly, we have also
−
w(t) − w(0) 1 f (t + t0 ) − f (t+
0 ) + f (−t + t0 ) − f (t0 )
lim = lim
t→0+ t 2 t→0+ t
f (t + t0 ) − f (t+
0) f (−t + t0 ) − f (t−
0)
From the assumption, the limits lim and lim both
t→0+ t t→0+ t
exist. Then the previous equality shows that w has the right-derivative at any point of
2πZ. And the left-derivative can be obtained from the fact that w is an even function. We
consider now the following function

R \ 2πZ −→ C
h: w(t)−w(0) .
t 7−→ eit −1

Since eit − 1 ∼ it when t → 0, the function h can be extended by continuity into a piece-
wise continuous and 2π-periodic function on R, which we always denote it by h. In the
consequences, we can write

∀t ∈ R, w(t) = l0 + h(t)(eit − 1).

10.3. CONVERGENCE THEOREMS 351

Calculating now the Fourier coefficients of w, we have, for all n ∈ Z,

cn (w) = l0 δ0,n + cn (he1 ) − cn (h) = l0 δ0,n + cn−1 (h) − cn (h).

On the other hand, the function u has the same Fourier coefficients as w since they are
equal except on 2πZ. We obtain thus,
N
X
SN (u)(0) = l0 + (cn−1 (h) − cn (h)) = l0 + c−N −1 (h) − cN (h).
n=−N

Finally, we apply Riemann-Lebesgue Lemma 10.2.13, cN (h) −→ 0 and it holds,

N →±∞

1
f (t− +


SN (f )(t0 ) = SN (u)(0) −→ l0 = 0 ) + f (t0 ) .
N →+∞ 2

Corollary 10.3.4. Let f be a piecewise C 1 -class and 2π periodic function. Then,

1
f (t− ) + f (t+ ) .


∀t ∈ R, lim SN (f )(t) =
N →+∞ 2

Corollary 10.3.5 (Dirichlet’s Theorem). Let f ∈ D2π (R, C)∩PC 12π (R, C), then the Fourier
series converges pointwisely to f on R. That is,
+∞ +∞
X a0 X
∀t ∈ R, f (t) = cn (f )eint = + (an (f ) cos(nt) + bn (f ) sin(nt)) .
n=−∞
2
n=1

Example 10.3.6. Consider the 2π-periodic function f of which its restriction on (−π, π)
is given by,

 1 if t ∈ (0, π)

f (t) = 0 if t=0

−1 if t ∈ (−π, 0)


Then, f is piecewise C 1 -class and satisfies the Dirichlet condition on R. Calculate now its
Fourier coefficients, let n ∈ N,

• an (f ) = 0 since f is an odd function;

(
Z π 0 if n = 2p
1 2
• bn (f ) = f (t) sin(nt)dt = (1 − (−1)n ) = 4 .
π −π nπ π(2p+1) if n = 2p + 1

Applying Dirichlet’s Theorem 10.3.3 as well as Corollary 10.3.5, we have,

+∞
4X 1
∀t ∈ R, f (t) = sin(2p + 1)t.
π 2p + 1
p=0
352 FOURIER SERIES

π
In particular, we take t = ∈ (0, π) and it holds,
2
+∞ +∞
(−1)p
 
4X 1 (2p + 1)π X π
1= sin ⇔ = .
π 2p + 1 2 2p + 1 4
p=0 p=0

We remark at first that for all p ∈ N, the function t 7→ sin(2p + 1)t is continuous on
[0, 2π]. If the Fourier series converges uniformly on [0, 2π], it follows from Theorem 7.1.23
that its sum function f is continuous on [0, 2π]. It is clear that f is not continuous. By
contraposition, the convergence of Fourier series is not uniform in this case. Nevertheless,
we can use Theorem 7.1.25 to show that the Fourier series converges uniformly on any
segments in (0, π).
Let P ∈ N∗ , we consider S2P −1 (f ) the Fourier sum of degree 2P − 1 of f , that is,

P −1
4 X 1
∀t ∈ R, S2P −1 (f )(t) = sin(2p + 1)t.
π 2p + 1
p=0

By calculating its derivative,

P −1
′ 4 X 2 sin(2P t)
∀t ∈ R \ πZ, S2P −1 (f )(t) = cos(2p + 1)t = ,
π π sin t
p=0

kπ
we obtain that the extreme values of S2P −1 (f ) are achieved at for k ∈ [[1, 2P − 1]].
2P
π
Hence, the maximum of S2P −1 (f ) is achieved at and equals to
2P
P −1  
π 4 X 1 (2p + 1)π
max S2P −1 (f ) = S2P −1 (f )( )= sin .
R 2P π 2p + 1 2P
p=0

2 sin t
We can observe that this quantity is the Riemann sum of the function t 7→ on the
  π t
π 2π Pπ (2p + 1)π
partition (0, , ··· , ) with the tagged values on . Thus,
P P P 2P p∈[[0,P −1]]

Z π
2 sin t
lim (max S2P −1 (f )) = dt ≈ 1.179 > 1.
P →+∞ R π 0 t

We conclude that the maximum values of the Fourier sums of f does not converges to the
maximum value of f . This kind of results is called as Gibbsvii phenomena.

10.3.2 Normal Convergence

1
PropositionX 10.3.7. Let f be a piecewise C -class, 2π-periodic and continuous on R. Then,
the series |cn (f )| converges.
n
vii
Josiah Willard Gibbs (1839-1903), American mechanical engineer and scientists. His contributions are
principally on thermodynamics and physical chemistry.
10.3. CONVERGENCE THEOREMS 353

Proof. We denote also by f ′ the pseudo-derivative of f and it is immediate that f ′ ∈

1
PC 2π (R, C). It follows from Proposition 10.2.6 that cn (f ) = cn (f ′ ) for all n ∈ Z∗ . Using
X 1 in X
Corollary 10.2.12 and the fact that is convergent, we obtain the series |cn (f )| is
n
n2 n
convergent as the term-to-term product of two square summable series.

As the immediate consequences, we remark that (cn (f ))n∈Z ∈ ℓ1 (Z, C), (an (f ))n∈N ∈
ℓ1 (N, C) and (bn (f ))n∈N ∈ ℓ1 (N, C).

Theorem 10.3.8.X Let f be a piecewise C 1 -class, 2π-periodic and continuous on R. Then,

the Fourier series cn (f )en converges normally to f on R.
n
X
Proof. From Proposition 10.3.7, the numerical series |cn (f )| converges. We then apply
X n
Theorem 10.1.16 to obtain that the Fourier series cn (f )en is normally convergent on R.
n
Also, since f is piecewise C 1 -class, 2π-periodic and continuous on R, it satisfies the Dirichlet
condition. From Corollary 10.3.5, the Fourier series converges pointwisely to f on R. In
conclusion, the Fourier series converges normally to f on R.

Theorem 10.3.9 (Trigonometric Weierstrass). Let f ∈ C2π (R, C). For all ε > 0, there
exists a trigonometric polynomial p such that ∥f − p∥L∞ (R,C) ≤ ε. In other words, the set of
trigonometric polynomial is dense in the space of continuous, 2π-periodic functions endowed
with the uniform convergence topology.

Proof. Let ε > 0, we remark that f is continuous on the segment [0, 2π]. It follows from
Theorem 7.2.12 that there exists a piecewise affine function g on [0, 2π] such that
ε
∥f − g∥L∞ (R,C) ≤ .
2
Moreover, it is shown in the proof of Theorem 7.2.12 that we have in fact, g(0) = f (0) =
f (2π) = g(2π). We can thus construct a piecewise affine, 2π-periodic function, which we
ε
always denote by g, such that ∥f − g∥L∞ (R,C) ≤ .
2
Now, the function g is piecewise C 1 -class, 2π-periodic and continuous on R. Applying
Theorem 10.3.8, the Fourier series of g converges normally, hence uniformly, to g on R. So,
ε
there exists N ∈ N such that ∥g − SN (g)∥L∞ (R,C) ≤ . In conclusion, we have
2
∥f − SN (g)∥L∞ (R,C) ≤ ∥f − g∥L∞ (R,C) + ∥g − SN (g)∥L∞ (R,C) ≤ ε.

10.3.3 Convergence in Square Integrable Norm

In this subsection, we consider f ∈ PC 2π (R, C).

Proposition 10.3.10. Let f ∈ PC 2π (R, C) and N ∈ N, the mapping p 7→ ∥p − f ∥2 from

PN into R archives its minimum at the unique point SN (f ) ∈ PN .
354 FOURIER SERIES

⊥ . Using
Proof. For all p ∈ PN , it follows from Proposition 10.2.9 that f − SN (f ) ∈ PN
Pythagoras theorem, we have,

∥f − p∥22 = ∥f − SN (f )∥22 + ∥SN (f ) − p∥22 ≥ ∥f − SN (f )∥22 .

The equality holds if, and only if ∥SN (f ) − p∥2 = 0. Since p and SN (f ) belong both in the
subspace PN , we deduce that the equality holds if, and only if p = SN (f ).

Lemma 10.3.11. Let f ∈ PC 2π (R, C) and ε > 0. There exists a function g ∈ C2π (R, C)
such that ∥f − g∥2 ≤ ε.

Proof. By the invariance of integrals of 2π-periodic functions over a segments of length 2π

Proposition 10.1.4, we can assume that f is continuous at 0. Let (x0 , x1 , · · · , xm , xm+1 ) be
an adapted partition of [2, π] with m ∈ N∗ , x0 = 0 and xm+1 = 2π. The discontinuous
points of f are therefore x1 , x2 , · · · , xm . Let ε > 0, we define
( )
1 πε2
δ = min x1 − x0 , x2 − x1 , · · · , xm+1 − xm , >0
2 2m∥f ∥2L∞ (R,C)

and a 2π-periodic function g given as follows. For all j ∈ [[1, m]],

• if t ∈ (xj−1 + δ, xj − δ), then g(t) = f (t);

xj + δ − t t − (xj − δ)
• if t ∈ [xj − δ, xj + δ], then g(t) = f (xj − δ) + f (xj + δ).
2δ 2δ
From this construction, g is 2π-periodic and continuous on R. Also, for all t ∈ [xj −δ, xj +δ],
we have,

xj + δ − t t − (xj − δ)
|g(t)| ≤ |f (xj − δ)| + |f (xj + δ)| ≤ ∥f ∥L∞ (R,C) .
2δ 2δ
Therefore, ∥g∥L∞ (R,C) ≤ ∥f ∥L∞ (R,C) .
We calculate now ∥f − g∥22 .
Z m Z
1 1 X
∥f − g∥22 = |f − g| = 2
|f − g|2
2π [0,2π] 2π [x j −δ,xj +δ]
j=1
m Z m Z
1 X 2 1 X
≤ (|f | + |g|) ≤ (2∥f ∥L∞ (R,C) )2
2π [xj −δ,xj +δ] 2π [xj −δ,xj +δ]
j=1 j=1
4m∥f ∥2L∞ (R,C)
= δ ≤ ε2 .
π
Hence, the claim follows.

Theorem 10.3.12. Let f ∈ PC 2π (R, C) and ε > 0. There exists a trigonometric polynomial
[
p such that ∥f − p∥2 ≤ ε. In other words, the space of trigonometric polynomials PN
N ∈N
is dense in the space of piecewise continuous and 2π-periodic functions for the topology of
square integrable semi-norm.
10.3. CONVERGENCE THEOREMS 355

Proof. Let ε > 0, it follows from Lemma 10.3.11, there exists g ∈ C2π (R, C) such that
ε
∥f − g∥2 ≤ . In addition, it follows from the Trigonometric Weierstrass Theorem 10.3.9,
2
ε
there exists a trigonometric polynomial p such that ∥g − p∥2 ≤ ∥g − p∥L∞ (R,C) ≤ . In the
2
consequence,
ε ε
∥f − p∥2 ≤ ∥f − g∥2 + ∥g − p∥2 ≤ + = ε.
2 2

Theorem 10.3.13. Let f ∈ PC 2π (R, C). Then the sequence (∥f − SN (f )∥2 )N ∈N converges
to 0.
Proof. Let ε > 0, it follows from Theorem 10.3.12 that there exists a trigonometric polyno-
mial p such that ∥f − p∥2 ≤ ε. We denote by N the degree of the polynomial p, i.e. p ∈ PN .
Using now Proposition 10.3.10, we have ∥f − SN (f )∥2 ≤ ∥f − p∥2 .
For all N ′ ∈ N, N ′ ≥ N , it is immediate that PN ⊂ PN ′ . In the consequence, SN (f ) ∈
PN ′ . Applying again Proposition 10.3.10, we obtain that
∥f − SN ′ (f )∥2 ≤ ∥f − SN (f )∥2 ≤ ∥f − p∥2 ≤ ε.
Thus, the claim follows.

Corollary 10.3.14 (Parsevalviii -Bessel Formula or Plancherelix Identity). For all piece-
wisely continuous and 2π-periodic function f , it holds that
+∞ +∞
X |a0 (f )|2 1 X
∥f ∥22 2
|an (f )|2 + |bn (f )|2 = ∥fb∥2ℓ2 (Z,C) .


= |cn (f )| = +
n=−∞
4 2
n=1

Proof. We apply the triangular inequality |∥f ∥2 − ∥SN (f )∥2 | ≤ ∥f − SN (f )∥2 and Theo-
rem 10.3.13 to show that the sequence (∥SN (f )∥2 )N ∈N converges to ∥f ∥2 .

Remark 10.3.15. • We can reformulate Parseval’s Formula Corollary 10.3.14 in the fol-
lowing equivalent way. The application f 7→ fb is an isometryx between (C2π (R, C), ∥ ·
∥2 ) and (ℓ2 (Z, C), ∥ · ∥ℓ2 (Z,C) ). That means, f 7→ fb is linear and satisfies,
1
∀f ∈ C2π (R, C), √ ∥f ∥L2 ([0,2π],C) = ∥f ∥2 = ∥fb∥ℓ2 (Z,C) .
2π
• We can replace C2π (R, C) by D2π (R, C) in the previous point. Since the vector space
PC 2π (R, C) endowed with ∥·∥L2 ([0,2π],C) is not a normed space (there are always a finite
numbers of points on any segment of which the separation property is not verified),
we can not use the term “isometry” on the space PC 2π (R, C).
• Let f, g ∈ PC 2π (R, C), it follows from the polarization identities that
+∞ +∞
X a0 (f )a0 (g) 1 X  
⟨f, g⟩ = cn (f )cn (g) = + an (f )an (g) + bn (f )bn (g) .
n=−∞
4 2
n=1
viii
Marc-Antoine Parseval (1755-1836), French mathematician. Known for his result on Fourier series and
transforms.
ix
Michel Plancherel (1885-1967), Swiss mathematician. He worked on analysis, mathematical physis and
algebra.
x
By definition, an isometry between two normed spaces is a linear application which conserves the length.
356 FOURIER SERIES

10.4 Exercises
Exercise 10.4.1. For all n ∈ N, we define

sin(n + 21 )t
∀t ∈ R \ (2πZ), fn (t) = .
sin 2t

1.) Determinate the limits of fn on 2πZ. We still call fn its extension by continuity.

2.) Let m, n ∈ N. Justify the existence of

Z 2π
1
Im,n := fm (t)fn (t)dt.
2π 0

n
X
3.) Show that fn (t) = eikt for all t ∈ R.
k=−n

4.) Show that Im,n = 2 min(m, n) + 1.

Convolution with Fejér’s Kernel

Exercise 10.4.2. For any f, g ∈ PC 2π (R, C), we define the convolution product, denoted
by f ∗ g as follows
Z π
1
∀x ∈ R, f ∗ g(x) = f (x − t)g(t)dt.
2π −π

We recall also the Dirichlet’s kernel of degree n ∈ N the following trigonometric polyno-
mial
n
X
Dn : t 7→ eikt .
k=−n

We define the Fejérxi ’s kernel of degree N ∈ N the following function

N
1 X
FN := Dn .
N +1
n=0

1.) Show that f ∗ g is 2π-periodic and f ∗ g = g ∗ f .

2.) Show that if f is continuous on R then f ∗ g is continuous on R.

Z π Z π
1 1
3.) Show that for all N ∈ N, DN (t)dt = FN (t)dt = 1.
2π −π 2π −π
xi
Lipót Fejér (1880-1959), Hungarian mathematician. He worked on harmonic analysis. He deeply effected
the development of mathematics in Hungry. Paul Erdös, George Pólya and John von Neumann were his
Phd students.
10.4. EXERCISES 357

4.) Show that for all n ∈ N,


sin(n+ 12 )t

sin 2t
if t ∈ R \ (2πZ),
Dn (t) =
 2n + 1 if t ∈ 2πZ.

Then deduce that for all N ∈ N,

  
(N +1)t
 1 sin2 2
N +1 sin2 2t
if t ∈ R \ (2πZ),
FN (t) =
N +1 if t ∈ 2πZ.


Z π Z −δ
5.) Let δ ∈ (0, π). Show that the integrals FN (t)dt and FN (t)dt tend to 0 as
δ −π
N → +∞.

6.) From now on, we consider a function f ∈ C2π (R, C). Show that f is uniformly contin-
uous on R.

7.) Let (fN )N ∈N be the sequence of functions defined by,

∀N ∈ N, fN = f ∗ FN .

Show that fN is a trigonometric polynomial, and give its coefficients.

8.) Let N ∈ N. Show that

Z π
1
∀x ∈ R, fN (x) − f (x) = (f (x − t) − f (x))FN (t)dt.
2π −π

Then deduce that (fn )n∈N converges to f uniformly on [Link]

Exercise 10.4.3. Consider the function f defined by, f (t) = t2 on (−π, π) and extended
by 2π-periodicity on R.

1.) Calculate the trigonometric Fourier coefficients of f .

+∞
π2 X (−1)n
2.) Show that for all t ∈ [−π, π], t2 = +4 cos(nt).
3 n2
n=1

+∞ +∞
X 1 π2 X (−1)n+1 π2
3.) Show that = also that = .
n2 6 n2 12
n=1 n=1

Exercise 10.4.4. Show that

+∞
π 2 X cos(2nt)
∀t ∈ [0, π], t(π − t) = − .
6 n2
n=1
xii
Now, we have proved the Trigonometric Weierstrass Theorem, which states that the set of trigono-
metric polynomials is dense in the normed space (C2π (R, C), ∥ · ∥L∞ (R,C) ).
 358 FOURIER SERIES

Exercise 10.4.5. We say a function f ∈ PC 2π (R, C) is Hölder continuous with exponent

α > 0 if it holds

∀x, y ∈ R, |f (x) − f (y)| ≤ C|x − y|α ,

with C > 0 being a constant. We consider a such function f in the following questions.
1.) Justify that f is uniformly continuous on R.

2.) What can we say if α > 1? We assume now 0 < α ≤ 1.

M
3.) Show that, there exists M > 0 such that for all n ∈ Z∗ , |cn (f )| ≤ .
     |n|α 
2kπ 2kπ 2kπ 2(k + 1)π
Hint: f (t) = f + f (t) − f , and integral over , .
n n n n

Exercise 10.4.6. Let f ∈ PC 2π (R, C). We use (cn (f ))n∈Z to represent the Fourier coeffi-
cients of f . Show that the following assertions are equivalent.
(A1) f is of C ∞ class on R.

(A2) ∀k ∈ N, ∃Mk > 0 s.t. ∀n ∈ Z, |nk cn (f )| ≤ Mk .

Solution. We prove the implications in two directions.
A1 ⇒ A2: Since f is smooth and 2π-periodic on R, it is immediate that any order derivative of
f is also continuous and 2π-periodic on R. It follows from the relation between the
Fourier coefficients of f and those of its derivatives, we obtain,

∀n ∈ Z, cn (f ′ ) = incn (f ).

Using mathematical induction, we have,

∀k ∈ N, ∀n ∈ Z, cn (f (k) ) = (in)k cn (f ).

Let k ∈ N, since the k-th order derivative f (k) is 2π-periodic and continuous on R, it
follows from the convergence theorem of Fourier series in square integrable norm that
its Fourier coefficients (cn (f (k) ))n∈Z is square summable. Then, from the necessary
condition of convergent series, we have |cn (f (k) )|2 −→ 0. So, those coefficients are
n→±∞
bounded as a convergent sequence. That is,

∃Mk > 0 s.t. ∀n ∈ Z, |nk cn (f )| = |cn (f (k) )| ≤ Mk .

And the claim (A2) follows.

A2 ⇒ A1: Taking at first k = 2, we have,

M2
∀n ∈ Z∗ , |cn (f )| ≤ .
n2
1
In other words, cn (f ) = O( 2 ) as n → ±∞. From the comparison test and the
X 1 n X
convergence of series , the numerical series cn (f ) is absolutely convergent.
n
n2 n
10.4. EXERCISES 359
X
Using the result (Theorem 10.1.16) in the course, the Fourier series of f , cn (f )en ,
n
is normally (hence uniformly) convergent on R. Combining with the convergence
theorem in square integrable norm, the Fourier series of f converges to f in ∥ · ∥2 .
Since the uniform convergence implies the convergence in square integrable norm, we
deduce that the Fourier series of f converges to f uniformly on R. We have thus the
development in Fourier series,
+∞
X
∀t ∈ R, f (t) = cn (f )eint .
n=−∞
X
So, f can be considered as the sum function of the series of functions cn (f )en .
n
Notice that for all n ∈ Z, the function en is smooth on R. Let k ∈ N, we have,
dk
∀n ∈ Z, ∥ cn (f )en ∥L∞ (R) = ∥(in)k cn (f )en ∥L∞ (R)
dtk
|nk+2 cn (f )| Mk+2
= |nk cn (f )| = 2
≤
n n2
X 1
Using again the comparison test and the convergence of series , we deduce that
n
n2
X dk
the series of functions cn (f )en is normally (hence uniformly) convergent on
n
dtk
R. Apply the result (Corollary 7.3.25) in the course, we conclude that f is k-times
differentiable on R. Since k ∈ N can be chosen arbitrary, the function f is thus smooth
on R.

Poisson Summation Formula

Exercise 10.4.7. We denote by S(R) the set of C ∞ functions on R and rapid decreasing.
Precisely speaking, we definexiii
n o
S(R) : f ∈ C ∞ (R, C) ∀p, q ∈ N, ∃Cp,q > 0 s.t. ∀t ∈ R, |tp f (q) (t)| ≤ Cp,q .
2
1.) Show that the function φ : t 7→ e−t belongs to S(R).
2
Hint: You can prove, by induction, that φ(q) (t) = Pq (t)e−t with Pq being a polynomial
function.
In what follows, we consider a function f ∈ S(R).
2.) Show that f is integrable on R.
3.) Let n ∈ Z, we consider the function fn : R → C defined by, fn (t) =Xf (t + 2πn). Show
that, for any q ∈ N and segment [a, b] ⊂ R, the series of functions fn(q) is normally
n∈Z
convergent on [a, b].
xiii
For more detail, see Definition 11.3.1.
360 FOURIER SERIES

+∞
X
4.) We define the function F : R → C by F (t) = fn (t). Show that F is of C 1 -class
n=−∞
and 2π-periodic on R.
XZ 2π
5.) Let m ∈ Z. Show that the numerical series fn (t)e−imt dt is convergent. Then
n∈Z 0
+∞ Z 2π
1 X
deduce that cm (F ) = fn (t)e−imt dt.
2π n=−∞ 0

6.) By justifying the function t 7→ f (t)e−imt is integrable on R, show that

Z +∞
1
cm (F ) = f (t)e−imt dt.
2π −∞
We recall that the Fourier Transform of f is defined by,
Z +∞
∀ξ ∈ R, f (ξ) =
b f (t)e−iξt dt.
−∞

7.) Prove the following Poissonxiv Summation Formula,

+∞ +∞
X 1 X b
f (2πn) = f (m).
n=−∞
2π m=−∞

Integral Inequalities
Exercise 10.4.8. Let f be a 2π-periodic and C 1 -class function mapping from R into C.
1.) Give the relation between the Fourier coefficients cn (f ) and cn (f ′ ).
Z π
2.) We assume that f (t)dt = 0. Prove the following inequality
−π
Z π Z π
2
|f (t)| dt ≤ |f ′ (t)|2 dt.
−π −π

3.) Following the previous question, determinate the functions such that the equality
occurs.
Solution. 1.) (See Proposition 10.2.6) Let n ∈ Z. Since f is of C 1 -class and 2π-periodic,
it is immediate that f ′ is continuous and 2π-periodic. Calculating now the Fourier
coefficient cn (f ′ ), it following from the integration by parts,
Z 2π
′ 1
cn (f ) = f ′ (t)e−int dt
2π 0
 Z 2π 
1 −int 2π −int
= f (t)e 0
− f (t)(−in)e dt .
2π 0
xiv
Siméon Denis Poisson (1781-1840), French mathematician and physicist. He contributed on statistics,
complex analysis, partial differential equations and their applications in physics.
10.4. EXERCISES 361

From 2π-periodicity of f , we have f (2π)e−in(2π) = f (0)e−in0 . Then, it follows that

Z 2π
′ 1
cn (f ) = (in) f (t)e−int dt = incn (f ).
2π 0

Remark: Since f is 2π-periodic, the above integrals can be replaced by any integral on
an interval of length 2π.

2.) We observe at first that

Z 2π Z π
1 −i0t 1
c0 (f ) = f (t)e dt = f (t)dt = 0,
2π 0 2π −π

also that c0 (f ′ ) = i0c0 (f ) = 0.

Next, we apply Parseval’s formula Corollary 10.3.14 to obtain that
Z π +∞
X
|f (t)|2 dt = 2π∥f ∥22 = 2π∥fb∥2ℓ2 (Z,C) = 2π |cn (f )|2 ,
−π n=−∞

as well as that
Z π +∞
X
|f ′ (t)|2 dt = 2π∥f ′ ∥22 = 2π∥fb′ ∥2ℓ2 (Z,C) = 2π |cn (f ′ )|2
−π n=−∞

In parallel, using the result in question 1.), we have, for all n ∈ Z∗ ,

|cn (f )|2 ≤ n2 |cn (f )|2 = |incn (f )|2 = |cn (f ′ )|2 .

Thus, combining the above results, we can deduce that

Z π +∞
X X
2
|f (t)| dt = 2π |cn (f )|2 = 2π |cn (f )|2
−π n=−∞ n∈Z∗
X +∞
X Z π
≤ 2π |cn (f ′ )|2 = 2π |cn (f ′ )|2 = |f ′ (t)|2 dt.
n∈Z∗ n=−∞ −π

Hence, the claim follows.

3.) From the analysis in 2.), the equality occurs if, and only if,
Z π Z π +∞
X
0= |f ′ (t)|2 dt − |f ′ (t)|2 dt = (|cn (f ′ )|2 − |cn (f )|2 )
−π −π n=−∞
+∞
X
= (n2 − 1)|cn (f )|2 .
n=−∞

Since n2 − 1 and |cn (f )|2 are both positive numbers, we can deduce that the equality
occurs if, and only if ∀n ∈ Z∗ , n2 − 1 = 0 or cn (f ) = 0. We obtain thus, cn (f ) = 0 for
all n ∈ Z with |n| ≥ 2.
362 FOURIER SERIES

Finally, since f is supposed to be continuous over R, it follows from the normal

convergence of Fourier series that
+∞
X
∀t ∈ R, f (t) = cn (f )eint = c1 (f )eit + c−1 (f )e−it .
n=−∞

In conclusion, the equality occurs if, and only if f ∈ Span(eit , e−it ).

Exercise 10.4.9.
Z π Let f be a 2π-periodic and C 1 -class function mapping from R into C.
We assume f (t)dt = 0. Prove that
−π
Z π
π
∥f ∥2L∞ (R,C) ≤ |f ′ (t)|2 dt.
6 −π

Exercise 10.4.10. Let g be a C 1 -class function defined on [0, π] such that g(0) = g(π) = 0.
Prove the following Wirtinger inequality
Z π Z π
2
|g(t)| dt ≤ |g ′ (t)|2 dt.
0 0

Euler’s Infinite Product of Sinus

Exercise 10.4.11. Let x ∈ (0, 1).
+∞
tx−1
Z
1.) Justify the existence of the integral I(x) := dt.
0 1+t
X (−1)n
2.) Justify the convergence of the series .
n
n+x

1 +∞
tx−1 X (−1)n
Z
3.) Show that dt = .
0 1+t n+x
n=0

X (−1)n +∞
1
4.) Deduce that that I(x) = − 2x .
x n2 − x2
n=1

Exercise 10.4.12. Let x ∈ (0, 1), we consider the 2π-periodic function f given by

∀t ∈ [−π, π], f (t) = cos(xt).

1.) Calculate the trigonometric Fourier coefficients an (f ) and bn (f ) for n ∈ N.

2.) Show that the Fourier series of f converges. Give the type of convergence.
Z +∞ x−1 Z +∞ −x
t π t
3.) Show that dt = . Then calculate dt.
0 1+t sin(πx) 0 1+t
10.4. EXERCISES 363
Z +∞
4.) Show that for all t > 0, Γ(1 − x) = t (ts)−x e−ts ds.
0

5.) Prove Euler’s duplication formula,

π
Γ(x)Γ(1 − x) = .
sin(πx)

Exercise 10.4.13. Let y ∈ (0, π).

1.) Justify the following series converges and show that

+∞
1 X 2y
cot y = + .
y y 2 − n2 π 2
n=1

Y y2

2.) Show that the infinite product 1− converges.
n2 π 2
n≥1

X 2y
3.) Show that the series of functions converges uniformly on any compact
y 2 − (nπ)2
n≥1
in (0, π).
+∞ +∞
! !
sin y 2 Y y2 Y 1
4.) Show that ln( ) = ln( ) + ln (1 − 2 2 ) − ln (1 − 2 ) .
y π n π 4n
n=1 n=1
n
Y 1 ((2n)!)2
5.) Let n ∈ N∗ . Show that (1 − ) = (2n + 1) .
4n2 16n (n!)4
k=1

+∞
Y 1 2
6.) Using Stirling’s formula, show that (1 − 2
)= .
4n π
n=1

7.) Prove Euler’s infinite product for sinus,

+∞
Y y2

sin y = y 1− 2 2 .
n π
n=1
364 FOURIER SERIES
 Chapter 11

Fourier Transforms & Distributions

In this chapter, the main objective is to present the famous Fourier transforms, which
plays an extremely important role in mathematics for engineering. At the first look, the
Fourier transforms can be considered as the “continuous version” of Fourier series, i,e.,
the limit scenarios while the considered functions have a period tending to infinity. In the
consequences, the sums of series would be interpreted as integrals.
We will begin with introducing the Fourier transforms of integrable functions, i.e., the
cases where the integrals are rigorously well-defined. Then we introduce some fundamental
properties of Fourier transforms, such as it transforms derivatives into products and vice
versa.
The second part of this chapter is reserved to give a slight view of the theory of distri-
butions, which are also known as the “generalized functions”. In modern developments of
physics and engineering, to restrict the Fourier transforms in integrable functions is largely
insufficient to satisfy the acquirements for applications. In the consequences, mathemati-
cians enlarged the concept of functions to give a vast space where the Fourier transforms
have its mathematical meaning. Due to the limited volume of this course, most of proofs and
auxiliary results are omitted. For those who are interested, we encourage them to explore
more advanced subjects such as Real Analysis, Functional Analysis and Partial Differential
Equations.
After having the concept of distributions, we present the generalized version of Fourier
transforms. In this context, we can derive lots of properties, including the most important
Fourier inversion formula as well as Parseval & Plancherel theorems. The main objective to
include such an advanced topic in a lecture note for undergraduate students is to provide a
solid mathematical support for the day they might seek rigorous meanings of the subjects
they are manipulating.

Notations
As an important matter, we shall adapt the following notations of differential calculus,
which will be used in this entire chapter.

1.) n ∈ N∗ and Rn endows with its usual Euclidean structure. A vector x ∈ Rn is denoted
by x = (x1 , · · · , xn ).
n
X
2.) The scalar product between x, y ∈ Rn is denoted and defined by, x · y = xj yj . And
j=1

365
366 FOURIER TRANSFORMS & DISTRIBUTIONS
v
√ u n 2
uX
n
the magnitude of a vector x ∈ R is given by, |x| := x · x = t xj .
j=1

n
X
3.) For any multi-index α ∈ Nn , we denote and define its length |α| by, |α| := αj as
j=1
n
Y
well as its factorial, α! := αj !.
j=1

n
α
Y
4.) For all h ∈ Rn and α ∈ Nn , we define, hα := hj j .
j=1

5.) The partial derivative of order α ∈ Nn are denoted by,

∂α ∂ α1 ∂ α2 ∂ αn ∂ |α|
∂α = = · · · = .
∂xα ∂xα1 1 ∂xα2 2 ∂xαnn ∂xα1 1 · · · ∂xαnn

6.) In what follows, Ω refers in all generality to a non-empty open subset in Rn . And the
term “function” always indicates the mappings defined on Ω and valued in C.

7.) Let k ∈ N∗ , we call a function φ mapping from Ω into C is k-times continuously

differentiable or of C k -class if all partial derivatives ∂ α φ are continuous over Ω for all
multi-index α ∈ Nn such that |α| ≤ k.

8.) The set of C k -class functions (which is also a vector space) is denoted by C k (Ω). We
denote as well as by C k (Ω) the space of C k -class functions φ such that ∂ α φ can be
extended continuously over Ω for all |α| ≤ k.

9.) We say the function φ is smooth, or indefinitely differentiable, if φ is of C k -class for

all k ∈ N.

10.) The space of smooth functions on Ω is denoted by C ∞ (Ω). Equivalently speaking,

∞
\ ∞
\
C ∞ (Ω) = C k (Ω) and C ∞ (Ω) = C k (Ω).
k=0 k=0

11.) Assuming that f and g are functions of C k -class on Ω, then f g is also a C k -class
function on Ω and moreover for all multi-index α ∈ Nn , |α| ≤ k, the following Leibniz
formula holds,
!
X α
∂ α (f g) = ∂ β f ∂ α−β g, (11.1)
N
β
β∈N
β≤α
!
α α!
where = . The proof of this formula is left in Exercises.
β β!(α − β)!

12.) The space L1 (Ω) denotes the set of functions which are measurable and integrable (in
the sense of Borel-Lebesgue integrals) over Ω. We note
Z
∥f ∥L1 (Ω) := |f (x)|dx.
Ω
11.1. FOURIER TRANSFORMS OF INTEGRABLE FUNCTIONS 367

13.) Let 1 ≤ p < +∞, the Lp −space is defined by

Lp (Ω) := f : Ω → C f is measurable and |f |p ∈ L1 (Ω) .



This space endows naturally with the following Lp −norm,

Z 1/p
∥f ∥Lp (Ω) := |f (x)|p dx .
Ω

14.) We define also the L∞ -space for uniformly bounded functions,

L∞ (Ω) := f : Ω → C f is measurable and ∃M > 0 such that |f | ≤ M a.e.i ,



and the L∞ -norm is given by,

∥f ∥L∞ (Ω) := inf {M | |f (x)| ≤ M a.e.} .

11.1 Fourier Transforms of Integrable Functions

Definition 11.1.1. Let f ∈ L1 (Rn ), we define the Fourier transform of f the following
function denoted by fb or by F(f ) even or by Fx→ξ (f (x)).
Z
∀ξ ∈ Rn , fb(ξ) := f (x)e−iξ·x dx.
Rn

Remark 11.1.2. There are other equivalent definitions of Fourier transforms. Different
definitions could lead to different versions of the same formulas. In the case when you need
to consult other documents on Fourier transforms, it is important to check the definition.
We present two other definitions in the follows.
Z
1
1.) fb(ξ) := f (x)e−iξ·x dx.
(2π)n/2 Rn
Z
2.) fb(ξ) := f (x)e−2πiξ·x dx.
Rn

Proposition 11.1.3. Let f ∈ L1 (Rn ), fb is well-defined and being continuous on Rn .

Proof. Let f ∈ L1 (Rn ) and ξ ∈ Rn . The function x 7→ f (x)e−iξ·x is integrable on Rn

since ∀x ∈ Rn , |f (x)e−iξ·x | ≤ |f (x)| and f ∈ L1 (Rn ). Thus, fb is well-defined. Using the
same domination condition and applying the Theorem of Continuity under Integral Signs
Theorem 8.4.14, the function fb is continuous on Rn .

L1 (Rn ) −→ L∞ (Rn )
Proposition 11.1.4. The Fourier transform F : is a continuous
f 7−→ fb
linear operator.
i
The abbreviation a.e. stands for almost everywhere. In the Measure Theory, we say a property P is
satisfied almost everywhere on a set S if P is true on S except for a subset of zero measure.
368 FOURIER TRANSFORMS & DISTRIBUTIONS

Proof. The linearity of F comes from the linearity of integration. Let f ∈ L1 (Rn ), we have
immediately the following estimation
Z Z
n −iξ·x
∀ξ ∈ R , |f (ξ)| ≤
b |f (x)e |dx = |f (x)|dx ≤ ∥f ∥L1 (Rn ) . (11.2)
Rn Rn

This implies that fb ∈ L∞ (Rn ) and ∥fb∥L∞ (Rn ) ≤ ∥f ∥L1 (Rn ) . Then, the application F is
linear and bounded on the unit sphere of L1 (Rn ). Hence F is continuous.

Lemma 11.1.5 (Riemann-Lebesgue). Let f ∈ L1 (Rn ), then lim fb(ξ) = 0.

|ξ|→+∞

Proof. We shall use the denseness result Theorem 11.2.9. Let φ ∈ D(Rn ). Since supp(φ)
is a compact set, it is closed and bounded. There exists then R > 0 such that supp(φ) ⊂
[−R, R]n . And it follows that for all ξ ∈ Rn ,
Z Z Pn
φ(ξ)
b = φ(x)e −iξ·x
dx = φ(x1 , · · · , xn )e−i j=1 ξj xj dx1 · · · dxn
Rn [−R,R]n
Z R Z R Yn
= ··· φ(x1 , · · · , xn ) e−iξj xj dx1 · · · dxn .
−R −R j=1

We consider now the values ξ ∈ Rn such that |ξ| → +∞. Then, there exists certain
component j0 ∈ [[1, n]] such that |ξj0 | → +∞. Using Fubini’s Theorem 8.4.17, we can
integrate the j0 -th variable at first. Hence,
Z R Z R Y Z R 
−iξj xj −iξj0 xj0
φ(ξ)
b = ··· e φ(x1 , · · · , xn )e dxj0 dx1 · · · dxn
−R −R 1≤j≤n −R
j̸=j0

From integration by parts, we have,

Z R
φ(x1 , · · · , xn )e−iξj0 xj0 dxj0
−R
R
e−iξj0 xj0 e−iξj0 xj0
Z R
∂φ
= φ(x1 , · · · , xn ) − (x1 , · · · , xn ) dxj0
−iξj0 −R −R ∂xj0 −iξj0
Z R
1 ∂φ
= (x1 , · · · , xn )e−iξj0 xj0 dxj0 .
iξj0 −R ∂xj0
Then it holds,
Z R Z R n
1 ∂φ Y
φ(ξ)
b = ··· (x1 , · · · , xn ) e−iξj xj dx1 · · · dxn
iξj0 −R −R ∂xj0
j=1
Z
1 ∂φ 1 d ∂φ
= (x)e−iξ·x dx = (ξ).
iξj0 [−R,R]n ∂xj0 iξj0 ∂xj0

Applying the above equality and the estimation (11.2), we have,

1 d∂φ 1 ∂φ
|φ(ξ)| ≤ ≤ −→ 0. (11.3)
|ξj0 | ∂xj0 |ξj0 | ∂xj0
b
L∞ (Rn ) L1 (Rn ) |ξ|→+∞
11.1. FOURIER TRANSFORMS OF INTEGRABLE FUNCTIONS 369

Let f ∈ L1 (Rn ) and ε > 0, from Theorem 11.2.9, there exists φ ∈ D(Rn ) such that
ε
∥f − φ∥L1 (Rn ) ≤ . Using the convergence (11.3), there exists M > 0 such that for all
2
ε
ξ ∈ Rn , |ξ| ≥ M ⇒ |φ(ξ)|
b ≤ . By applying those results and (11.2), for |ξ| ≥ M , we have
2
|fb(ξ)| ≤ |fb(ξ) − φ(ξ)|
b + |φ(ξ)|
b ≤ ∥fb − φ∥
b L∞ (Rn ) + |φ(ξ)|
b
ε ε
≤ ∥f − φ∥L1 (Rn ) + |φ(ξ)|
b ≤ + = ε.
2 2
Hence, lim fb(ξ) = 0.
|ξ|→+∞

Example 11.1.6. Calculation of F(1(−1,1) ).

Solution. We recall that for all x ∈ R,
(
1 if −1 < x < 1
1(−1,1) (x) =
0 if x ∈
/ (−1, 1).

For all ξ ∈ R,
1
e−iξ − eiξ
Z Z
sin(ξ)
F(1(−1,1) )(ξ) = 1(−1,1) (x)e −iξx
dx = e−iξx dx = =2 = 2sinc(ξ).
R −1 −iξ ξ

Let a ∈ Rn , we denote by τa to represent the translation operation by a. That means,

if f denotes a function defined on Rn , then ∀x ∈ Rn , τa f (x) = f (x + a).
Proposition 11.1.7. Let f ∈ L1 (Rn ), a ∈ Rn and j ∈ [[1, n]], then the following assertions
hold.
1.) If the function fej : x 7→ xj f (x) belongs also to L1 (Rn ), then fb has its continuous
∂ b
partial derivative given by i f (ξ) = Fx→ξ (xj f (x))(ξ).
∂ξj
∂f
2.) If f is differentiable on the direction xj and the j-th partial derivative ∈ L1 (Rn ),
∂xj
∂f
then F( )(ξ) = iξj F(f )(ξ).
∂xj

3.) Fx→ξ (f (x − a))(ξ) = e−ia·ξ F(f )(ξ); equivalently, τd

a f (ξ) = e
ia·ξ b
f (ξ).

4.) Fx→ξ (eia·x f (x))(ξ) = fb(ξ − a) = τ−a fb(ξ).

b b
1 bξ
5.) For h ∈ R∗ , Fx→ξ (f (hx))(ξ) = f ( ). In particular, f = fb.
b
|h|n h
Proof. 1.) We consider the function F : (x, ξ) 7→ f (x)eiξ·x . A simple calculation leads
∂F
to, i = xj f (x)eiξ·x . We have then for all ξ ∈ Rn , |F (·, ξ)| ≤ |f | ∈ L1 (Rn ) and
∂ξj
∂F
(·, ξ) ≤ |fej | ∈ L1 (Rn ). Then, the claim follows from the Theorem of Derivation
∂ξj
under Integral Signs Theorem 8.4.15.
370 FOURIER TRANSFORMS & DISTRIBUTIONS

∂f
2.) Under the assumptions, the Fourier transform F( ) is well-defined. We next cal-
∂xj
culate the integral over Rn . We shall denote a point x ∈ Rn by x = (xj , x′ ) with
∂f
x′ ∈ Rn−1 . Then, since ∈ L1 (Rn ), it follows from Fubini Theorem 8.4.17 that we
∂xj
can integrate at first on the variable xj . That is,
Z Z Z +∞
∂f ∂f ∂f ′ ′
F( )= (x)e−iξ·x dx = (xj , x′ )e−iξj xj −iξ ·x dxj dx′
∂xj Rn ∂xj Rn−1 −∞ ∂xj
Z Z +∞ 
−iξ ′ ·x′ ∂f ′ −iξj xj
= e (xj , x )e dxj dx′ .
Rn−1 −∞ ∂x j

Z +∞
∂f
We are about to calculate the integral (xj , x′ )e−iξj xj dxj . Let R > 0 and
−∞ ∂x j
x′ ∈ Rn−1 , it follows from the integration by parts that
Z R Z R
∂f R
(xj , x′ )e−iξj xj dxj = f (xj , x′ )e−iξj xj + iξj f (xj , x′ )e−iξj xj dxj .
−R ∂xj −R −R

∂f ∂f
Since ∈ L1 (Rn ), it is immediate that xj 7→ (xj , x′ ) ∈ L1 (R). It follows
∂xj ∂xj
Z R
′ ′ ∂f
from the identity f (R, x ) = f (0, x ) + (xj , x′ )dxj that the right-hand side
0 ∂xj
converges while R → ±∞. We have then f (xj , x′ ) −→ l± ∈ R. We use now the
xj →±∞
fact f ∈ L1 (Rn ) ⇒ xj 7→ f (xj , x′ ) ∈ L1 (R). This condition implies that l± = 0 since
if not, the function xj 7→ f (xj , x′ ) is no longer integrable over R. In consequences, we
have
R
f (xj , x′ )e−iξj xj −→ 0.
−R R→+∞

Z +∞ Z +∞
∂f
Thus, (xj , x′ )e−iξj xj dxj = iξj f (xj , x′ )e−iξj xj dxj . Importing it in the
−∞ ∂x j −∞
∂f
calculus of F( ), the claim follows.
∂xj

3.) It is sufficient to apply the change of variable y = x − a.

4.) Straightforward from calculations.

5.) By definition of Fourier transforms, we calculate ∀ξ ∈ Rn ,

Z
Fx→ξ (f (hx))(ξ) = f (hx)e−iξ·x dx
Rn
Z +∞ Z +∞ Pn
= ··· f (hx1 , · · · , hxn )e−i j=1 ξj xj dx1 · · · dxn .
−∞ −∞

We then apply the change of variables y = hx, that is, y1 = hx1 , · · · , yn = hxn . If
11.1. FOURIER TRANSFORMS OF INTEGRABLE FUNCTIONS 371

h > 0, we have thus

Z +∞ Z +∞ Pn
··· f (hx1 , · · · , hxn )e−i j=1 ξj xj dx1 · · · dxn
−∞ −∞
Z +∞ Z +∞ Pn yj
dy1 dyn
= ··· f (y1 , · · · , yn )e−i j=1 ξj h ···
−∞ −∞ h h
Z
1 ξ 1 bξ
= n f (y)e−i h ·y dy = f( )
h Rn hn h
Similarly, if h < 0, we have,
Z +∞ Z +∞ Pn
··· f (hx1 , · · · , hxn )e−i j=1 ξj xj dx1 · · · dxn
−∞ −∞
Z −∞ Z −∞ Pn yj
dy1 dyn
= ··· f (y1 , · · · , yn )e−i j=1 ξj h ···
+∞ +∞ h h
+∞ +∞
(−1)n
Z Z Pn ξj
= ··· f (y1 , · · · , yn )e−i j=1 h yj
dy1 · · · dyn
hn −∞ −∞
Z
1 ξ 1 bξ
= f (y)e−i h ·y dy = f( )
|h|n Rn |h|n h

Lemma 11.1.8. The Fourier transforms of Gaussian functions are given as follows.
x2
1.) Let f : x 7→ e− 2 defined on R. Then, ∀ξ ∈ R,
√ ξ2
fb(ξ) = 2πe− 2 . (11.4)
Pn
2 x2j
2.) Let ε > 0 and fε : x 7→ e−ε|x| = e−ε defined on Rn . Then, ∀ξ ∈ Rn ,
j=1

r n
π |ξ|2
fε (ξ) =
b e− 4ε . (11.5)
ε

Proof. 1.) We calculate at first the derivatives of the Gaussian function f .

x2
∀x ∈ R, f ′ (x) = −xe− 2 .

In terms of equations, the function f satisfies the following differential equation

∀x ∈ R, y ′ (x) + xy(x) = 0 (11.6)

Applying now the Fourier transform on this equation, we obtain,

Fx→ξ (y ′ (x)) + Fx→ξ (xy(x)) = 0

d
⇔∀ξ ∈ R, iξb y (ξ) + i yb(ξ) = 0
dξ

Hence, yb satisfies the same equation (11.6). Denoting λ = fb(0) ∈ R, it follows then the
functions λf and fb satisfy at the same time the equation (11.6) as well as the initial
372 FOURIER TRANSFORMS & DISTRIBUTIONS

condition y(0) = λ. Applying Cauchy-Lipschitz Theoremii , we have that fb = λf .

Finally, we calculate
Z
x2 √
λ = f (0) =
b e− 2 dx = 2π.
R

Thus, the formula (11.4) follows.

2.) Let ξ ∈ Rn , standard calculations lead to

Z Z Pn Z n
2 2 2
e− e−iξj xj −εxj dx.
Y
fbε (ξ) = e−ε|x| e−iξ·x dx = j=1 (iξj xj +εxj ) dx =
Rn Rn Rn j=1

2
Since for each j, the function x 7→ e−iξj xj −εxj depends only on the variable xj , we can
write the above integral as follows,
n Z n
−iξj xj −εx2j
Y Y
fbε (ξ) = e dxj = Ij .
j=1 R j=1
Z
2
Then, it is sufficient to calculate for each j ∈ [[1, n]] the integral Ij = e−iξj xj −εxj dxj .
√ R
Using a change of variables y = 2εxj , we have,
Z Z ξ 2 Z ξ
−iξj xj −εx2j 1 −i √ j y− y2 1 −i √ j y − y
2
Ij = e dxj = √ e 2ε dy = √ e 2ε e 2 dy
R 2ε R 2ε R
1 ξj
= √ fb( √ ).
2ε 2ε

Applying now (11.4) we obtain,

r  ξ2
π j
Ij = e− 4ε .
ε

And hence,
n r n Pn ξ2 r n
π π |ξ|2
− j=1 j
Y
fbε (ξ) = Ij = e 4ε = e− 4ε .
ε ε
j=1

Thus, (11.5) follows.

x2
Remark 11.1.9. The relation (11.4) stands that the Gaussian function x 7→ e−√2 is an
eigenfunction to the continuous linear operator F associated to the eigenvalue 2π.
ii
This theorem sates the existence and uniqueness of the solution to the differential equations. It is one
of the fundamental results
( in the studies in ordinary differential equations. Precisely speaking, consider the
dy
dt
= f (t, y)
initial value problem of the unknown function y defined on an open interval interval I
y(t0 ) = y0
with t0 ∈ I and y0 ∈ R. The Cauchy-Lipschitz Theorem states that if the function f is Lipschitz continuous
on the second variable y, then this initial value problem has a unique solution in a neighborhood of t0 .
11.2. ELEMENTARY DISTRIBUTION THEORY 373

11.2 Elementary Distribution Theory

The theory of distributions had introduced for the objectives to expand the notions of
functions so that it could be reasonable to describe the derivations of discontinuous func-
tions. So we also call them as generalized functions. The theory of distributions has
already various applications in the studies of physics, signal processing, electronics and
other domains.
The original motivation to develop such theory could seek back to Paul Dirac’s works on
modeling the impulsion in 1920s. At that time, a function δ, which is called Dirac mass
nowadays, had introduced as follows.
(
+∞ if x = 0
δ(x) =
0 if x ̸= 0,

and, for any continuous function φ,

Z +∞
δ(x)φ(x)dx = φ(0).
−∞

It is clear that δ is not, in a point of view of rigorous mathematics meanings, a well-defined

function. It was dated to 1940s that a complete mathematical interpretation of such concept
was developed by Laurent Schwartz. The main idea was to interpret a function f , instead of
knowing each value f (x) Z on the domain of definition, we can also determinate all the values
+∞
of the scalar product f (x)φ(x)dx for φ belongs to a set of functions that should be
−∞
carefully chosen. An other important application of the theory of distribution is the studies
in partial differential equations.
For a more complete exploration of the theory, we recommend to consult the following
masterpieces:

• Laurent Schwartz Théorie des Distributions Hermann Paris 1966

• Lars Hörmander The Analysis of Linear Partial Differential Operators I - Distribution

Theory and Fourier Analysis Springer-Verlag Berlin 1990

11.2.1 Space of Test Functions

Definition 11.2.1. Let φ be a continuous function defined on Ω. We call the support of
φ and we denote it by supp(φ) the closure of the set where φ does not vanishes.

Remark 11.2.2. • We say φ vanishes at x ∈ Rn if φ(x) = 0.

• We obtain from the definition of the support that supp(φ)∁ is the largest open set in
Ω where φ vanishes.

Definition 11.2.3. We denote by D(Ω) or by C0∞ (Ω) and call it the space of test functions
the set of functions mapping from Ω into C, indefinitely differentiable with a compact
support in Ω.
374 FOURIER TRANSFORMS & DISTRIBUTIONS

Proposition 11.2.4. D(Ω) forms a vector space.

Proof. Left in Exercise 11.4.1.

Remark 11.2.5. Let φ ∈ D(Ω), we can extend φ into a function φ̃ ∈ D(Rn ).

Proposition 11.2.6.
(  
−1
exp 1−|x|2
if |x| < 1,
φ(x) = (11.7)
0 if |x| ≥ 1.

Then φ ∈ D(Rn ) and supp(φ) = B(0, 1).

Proof. Left in Exercise 11.4.2.

Remark 11.2.7. Let φ be the function defined by (11.7), from what is preceding, φ ∈ D(Rn ).
We set for ε > 0,
Z −1
1 x
φε (x) = n φ(x)dx φ( ). (11.8)
ε Rn ε

Then, the functions φε are smooth and their supports are B(0, ε). Moreover,
Z
φε (x)dx = 1.
Rn

The sequence defined by (11.8) is called a regularizing sequence (in some references, it
is also called as mollifiers or an approximation to the identity). That means, a sequence
of test functions (φε )ε>0 satisfying while ε → 0+ ,

• φε (x) ≥ 0, ∀x ∈ Rn ,
\
• supp(φε ) → {0Rn }, i.e., supp(φε′ ) ⊂ supp(φε ) if ε′ ≤ ε and supp(φε ) = {0Rn },
ε>0
Z
• φε (x)dx = 1.
Rn

Theorem 11.2.8 (Cut-off functions). Let K ⊂ Ω be a compact subset of Ω and let δ > 0.
Then, there exists χ ∈ D(Ω) such that,

• 0 ≤ χ(x) ≤ 1 for all x ∈ Ω,

• χ(x) = 1 for x ∈ K,

• χ(x) = 0 if x ∈ Ω\ K̃ where K̃ is a compact neighborhoodiii of K such that dist(y, K) ≤

δ for all y ∈ K̃.

Such function χ is called a cut-off function over K.

iii
That means, there exists a neighborhood V of K and a compact set K̃ such that K ⊂ V ⊂ K̃.
 11.2. ELEMENTARY DISTRIBUTION THEORY 375

Proof. See in the course Partial Differential Equations

Let 1 ≤ p < ∞, the following theorem shows that any functions in Lp (Ω) can be
approximated by compact supported smooth functions.

Theorem 11.2.9. D(Ω) is dense in Lp (Ω). That is, D(Ω) ,→ Lp (Ω).

dense

Proof. Difficult theorem, see in the course Partial Differential Equations.

Definition 11.2.10. Let (φk )k∈N be a sequence of functions in D(Ω) and let φ ∈ D(Ω), we
say that (φk )k∈N convergesiv to φ in D(Ω) if

• there exists K ⊂ Ω, K compact such that supp(φk ) ⊂ K for all k ∈ N,

• for all multi-index α ∈ Nn , ∂ α φk converge to ∂ α φ uniformly on Ω.

11.2.2 Distributions
Definition 11.2.11. We call a distribution or a generalized function any continuous
linear formv defined on D(Ω). Precisely speaking, a mapping T : D(Ω) → C is said to be a
distribution if T satisfies

Linearity: for all f, g ∈ D(Ω) and λ, µ ∈ C, T (λf + µg) = λT (f ) + µT (g);

Continuity: for any sequence (φk )k∈N which converges in D(Ω), φk −→ φ, then (T (φk ))k∈N
k→∞
converges to T (φ) in C.

We denote by D′ (Ω)vi the set of distributions on Ω and we note ⟨T, φ⟩ at the place of T (φ).
In the case of possible confusion, we might denote it by ⟨T, φ⟩D′ ×D .

The following proposition give an alternative definition of distributions.

Proposition 11.2.12. Let T be a mapping defined on D(Ω). Then, T is a distribution if,

and only if T satisfies

1.) T is a complex valued linear form defined on D(Ω),

in Ω, there exists C > 0 and m ∈ N such that for any

2.) for any compact subset K X
∞ vii
φ ∈ CK (Ω) , |T (φ)| ≤ C ∥∂ α φ∥L∞ (K) .
|α|≤m

Proof. See in the course Partial Differential Equations.

We denote by L1loc (Ω) the set of functions which are integrable on any compact in Ω.
iv
Here, we define the convergence in the space D(Ω) without introducing any norm or distance function.
In fact, the spaces of test functions D(Ω) as well as the space of distributions D′ (Ω), which will be introduced
later, are non-metrizable spaces. That means, it is impossible to find out any distance function or norm
on those spaces. The topology of those spaces are considered in the frameworks of topological vector
spaces (TVS).
v
It is also called a functional
vi
In fact, the space D′ (Ω) is exactly the topological dual space of D(Ω).
vii
This space stands for the set of smooth functions with a support contained in K
376 FOURIER TRANSFORMS & DISTRIBUTIONS

Proposition 11.2.13. Let f ∈ L1loc (Ω), we define for all φ ∈ D(Ω),

Z
⟨Tf , φ⟩ = f (x)φ(x)dx. (11.9)
Ω

Then, Tf ∈ D′ (Ω).

Proof. The linearity comes from the linearity of the integration.

In parallel, let φ ∈ D(Ω) such that supp(φ) ⊂ K ⊂ Ω with K compact, we have the
estimation,
Z
|⟨Tf , φ⟩| ≤ |f (x)||φ(x)|dx ≤ ∥f ∥L1 (K) ∥φ∥L∞ (K) .
K

Combined with the linearity of Tf , the continuity of Tf comes directly from this estimation.

Definition 11.2.14. Let T ∈ D′ (Ω), we say that T is a regular distribution if there

exists f ∈ L1loc (Ω) such that ⟨T, φ⟩ = ⟨Tf , φ⟩ for all φ ∈ D(Ω), where Tf is defined by
(11.9).

Remark 11.2.15. The mapping f 7→ Tf which associate f ∈ L1loc (Ω) and Tf ∈ D′ (Ω) is
linear and injective but it is not surjective.

Example 11.2.16. Let a ∈ Rn , we define for all φ ∈ D(Rn ),

⟨δa , φ⟩ = φ(a). (11.10)

Then δa ∈ D′ (Rn ).

Remark 11.2.17. The distributions defined by (11.10) are called the Dirac mass at a ∈ Rn ,
we denote also by δ to refer the Dirac mass at origin, i.e. δ = δ0 .

Proposition 11.2.18. The Dirac mass δa is not a regular distributions, which shows that
the mapping f 7→ Tf is not surjective over L1loc (Ω).

Proof. See in the course Partial Differential Equations.

Definition 11.2.19. We say a sequence of distributions (Tk )k∈N converge to a distribution

T ∈ D′ (Ω), if for all φ ∈ D(Ω), ⟨Tk , φ⟩ −→ ⟨T, φ⟩ in C.
k→+∞
X
Analogically, we say a series Tk of distributions converges to a “sum distribution” T if
k
k
X
the sequence of partial sums Sk = Tj converge to T in D′ (Ω).
j=0

Theorem 11.2.20. Let Tk ∈ D′ (Ω), ∀k ∈ N. If for all φ ∈ D(Ω), the sequence (⟨Tk , φ⟩)k∈N
converges in C, then (Tk )k∈N converge in D′ (Ω).
11.2. ELEMENTARY DISTRIBUTION THEORY 377

Proof. Consequences of Banach-Steinhaus Theorem or Uniform Boundedness Principle. See

in the course Partial Differential Equations.

Example 11.2.21 (The principal value of 1/x). For ε > 0, we define for φ ∈ D(R),
Z
φ(x)
⟨Tε , φ⟩ = dx.
|x|≥ε x

Then we have,

• Tε ∈ D′ (R),

• (Tε )ε>0 converge in D′ (R) while ε → 0. The limit is called the principal value of 1/x,
which is denoted by p.v.( x1 ), that is,
  Z
1 φ(x)
p.v.( ), φ = lim dx.
x ε→0 |x|≥ε x

Solution. • Let ε > 0, we have for all φ ∈ D(R),

Z Z
φ(x) 1 D E
⟨Tε , φ⟩ = dx = 1R\(−ε,ε) (x) φ(x)dx = T1R\(−ε,ε) x1 , φ .
|x|≥ε x R x

1
We remark here the function x 7→ 1R\(−ε,ε) (x) ∈ L1loc (R). Hence, Tε can be consid-
x
ered as a regular distribution and thus Tε ∈ D′ (R).

• Let 0 < ε′ < ε, we calculate

Z Z
φ(x) φ(x)
⟨Tε′ , φ⟩ − ⟨Tε , φ⟩ = dx − dx
|x|≥ε′ x |x|≥ε x
Z −ε′ Z ε
φ(x) φ(x)
= dx + dx,
−ε x ε′ x
= (φ(ε) − φ(−ε)) ln(ε) − (φ(ε′ ) − φ(−ε′ )) ln(ε′ )
Z ε
− ln(x)(φ′ (x) + φ′ (−x))dx.
ε′

Then we have the estimation

Z ε
|⟨Tε′ , φ⟩ − ⟨Tε , φ⟩| ≤ |f (ε)| + |f (ε′ )| + | ln(x)||φ′ (x) + φ′ (−x)|dx,
ε′

where we set f (x) = (φ(x) − φ(−x)) ln(x).

Z x
Using the mean value inequality, |f (x)| = φ′ (t)dt | ln(x)| ≤ 2∥φ′ ∥L∞ (R) |x ln(x)|.
−x
Further, the function x → |x ln(x)| is an increasing function for x positive and close
enough to 0. We have thus,

|f (ε)| + |f (ε′ )| ≤ 4∥φ′ ∥L∞ (R) |ε ln(ε)|.

 378 FOURIER TRANSFORMS & DISTRIBUTIONS

In parallel, it holds,
Z ε Z ε
′ ′ ′
| ln(x)||φ (x) + φ (−x)|dx ≤ 2∥φ ∥L∞ (R) | ln(x)|dx,
ε′ 0
≤ 2∥φ′ ∥L∞ (R) |ε − ε ln(ε)|.

Combining those estimations, we obtain,

|⟨Tε′ , φ⟩ − ⟨Tε , φ⟩| ≤ 4∥φ′ ∥L∞ (R) |ε ln(ε)| + 2∥φ′ ∥L∞ (R) |ε − ε ln(ε)| −→ 0
ε→0

This estimation shows that, for any sequence (εn )n∈N such that εn > 0 and εn → 0,
the sequence (⟨Tεn , φ⟩)n∈N is in fact a Cauchy sequence in C, which is thus convergent.
Then we apply Theorem 11.2.20 and the claim follows.

11.2.3 Operations on Distributions

Definition 11.2.22. Let T ∈ D′ (Ω), 1 ≤ j ≤ n. We define the partial derivative of T
∂T
respect to the variable xj , which is denoted as , by the following formula,
∂xj
   
∂T ∂φ
∀φ ∈ D(Ω), , φ = − T, . (11.11)
∂xj ∂xj

Proposition 11.2.23. Let T ∈ D′ (Ω), then

∂T
• ∈ D′ (Ω) for all 1 ≤ j ≤ n,
∂xj
∂2T ∂2T
• = for all 1 ≤ j, k ≤ n,
∂xj ∂xk ∂xk ∂xj

• ⟨∂ α T, φ⟩ = (−1)|α| ⟨T, ∂ α φ⟩ for all φ ∈ D(Ω) and all α ∈ Nn .

Proof. Let α ∈ Nn be a multi-index and define the mapping T1 : φ 7→ ⟨T, ∂ α φ⟩ on the space
D(Ω). We verify T1 is a continuous linear form on D(Ω).
Well-defined: Let φ ∈ D(Ω). The function φ is smooth and supported in a compact K ⊂ Ω.
So, the partial derivative ∂ α φ is also a smooth function on Ω. Also, φ vanishes on
Ω \ K, which implies that ∂ α φ vanishes on the same subset. In the consequences,
∂ α φ ∈ D(Ω). Thus, the mapping T1 : D(Ω) → C is well-defined.

Linearity: The linearity of T1 comes from those of T and partial derivation.

Continuity: Consider a sequence (φk )k∈N which converges to φ in D(Ω). From the definition
of convergence in D(Ω), the supports of φk are contained in a common compact
K ⊂ Ω. In the consequence, the supports of ∂ α φk lies also in K for each k ∈ N.
Also, for any multi-index, β ∈ Nn the sequence of functions ∂ α (∂ β φk ) = ∂ α+β φk
converges uniformly to ∂ α+β φ = ∂ α (∂ β φ) on Ω while k → +∞. In the consequence,
the sequence of functions (∂ α φk )k∈N converges to ∂ α φ in D(Ω). Finally, using the
continuity of T , we deduce that T1 (φk ) = ⟨T, ∂ α φk ⟩ converges to ⟨T, ∂ α φ⟩ = T1 (φ) in
C while k → +∞. In conclusion, T1 is continuous on D(Ω).
11.2. ELEMENTARY DISTRIBUTION THEORY 379

Proposition 11.2.24. The derivation is continuous on D′ (Ω). If Tk −→ T in D′ (Ω),

k→+∞
then for all α ∈ Nn , ∂ α Tk → ∂ α T in D′ (Ω).

Proof. Left in Exercise 11.4.3.

(
1 if x > 0
Example 11.2.25. Consider the Heavisideviii function H(x) = . It
0 if x ≤ 0
is immediate that H ∈ L1loc (R), we can thus define a regular distribution TH ∈ D′ (R).
Moreover, let φ ∈ D(R),
Z +∞
′ ′
⟨TH , φ⟩ = −⟨TH , φ ⟩ = − φ′ (x)dx = φ(0) = ⟨δ, φ⟩.
0

So, TH′ = δ.

Example 11.2.26. Let f (x) = ln |x| for all x ∈ R \ {0}. Since f ∈ L1loc (R), we can define
a regular distribution Tf ∈ D′ (R).
Let φ ∈ D(R),
Z Z
⟨Tf′ , φ⟩ = −⟨Tf , φ′ ⟩ = − f (x)φ′ (x)dx = − lim f (x)φ′ (x)dx.
R ε→0+ |x|≥ε

Since φ is compactly supported, there exists M > ε such that supp(φ) ⊂ [−M, M ].
Z Z M Z −ε
′ ′
f (x)φ (x)dx = f (x)φ (x)dx + f (x)φ′ (x)dx
|x|≥ε ε −M
M Z M −ε Z −ε
φ(x) φ(x)
= f (x)φ(x) − dx + f (x)φ(x) − dx
ε ε x −M −M x
Z
φ(x)
= −φ(ε) ln(ε) + φ(−ε) ln(ε) − dx.
|x|≥ε x
Z ε
Using the mean value inequality, |φ(ε) − φ(−ε)| = φ′ (t)dt ≤ 2ε∥φ′ ∥L∞ (R) . We have
−ε
thus, |(φ(ε) − φ(−ε)) ln(ε)| ≤ 2∥φ′ ∥L∞ (R) |ε ln(ε)| −→ 0.
ε→0
In conclusion, for any φ ∈ D(R), ⟨f ′ , φ⟩ = ⟨p.v.( x1 ), φ⟩. That is, (ln |x|)′ = p.v.( x1 ).

Definition 11.2.27. Let I be an open interval in R and {a1 , · · · , am } denote the interior
points in I which satisfy a1 < a2 < · · · < am .
We say a function f is of piecewisely C k -class on I if f is of C k -class on any open intervals
of I \ {a1 , · · · , am } and has the limits at left and at right on each points of a1 , · · · , am .
We denote respectively by f (a∓ j ) the left and right limits of f at aj , for 1 ≤ j ≤ m.
viii
Oliver Heaviside (1850-1925), English mathematician and physicist. Known for his contributions on
solving differential equations and understandings on Maxwell’s equations.
380 FOURIER TRANSFORMS & DISTRIBUTIONS

Theorem 11.2.28 (Jumps’ formula). Let f be a function of piecewisely C 1 -class on I, then,

m
dTf X
−
= Tf ′ + (f (a+
j ) − f (aj ))δaj , (11.12)
dx
j=1

where Tf denotes the regular distribution defined by f ∈ L1loc (I) and f ′ denotes the function
defined by the derivative of f on I \ {a1 , · · · , am }.
Proof. Let φ ∈ D(I), supp(φ) is a compact set contained in I, which implies that there
exists p, q ∈ I such that supp(φ) ⊂ [p, q] ⊂ I. We are going to verify
  m
dTf X
−
, φ = ⟨Tf ′ , φ⟩ + (f (a+
j ) − f (aj ))⟨δaj , φ⟩.
dx
j=1

In the case where the points of discontinuity {a1 , · · · , am } ⊂ I \ [p, q]. We have immediately
that f is continuously differentiable on [p, q]. Then, it follows from the integration by parts,
    Z Z q
dTf dφ dφ dφ
, φ = − Tf , = − f (x) (x)dx = − f (x) (x)dx
dx dx I dx p dx
Z q Z
= −f (x)φ(x)|qp + f ′ (x)φ(x)dx = f ′ (x)φ(x)dx = ⟨Tf ′ , φ⟩.
p I

Hence (11.12) follows since ∀j ∈ [[1, m]], ⟨δaj , φ⟩ = φ(aj ) = 0.

Next, we may assume that the points of discontinuity {ar , ar+1 , · · · , as } ⊂ [p, q] for
certain 1 ≤ r ≤ s ≤ m. This time, f is continuously differentiable on each open intervals
(aj , aj+1 ) for all r −1 ≤ j ≤ s where we rename the points by setting ar−1 = p and as+1 = q.
It also follows from the integration by parts,
    Z Z q
dTf dφ dφ dφ
, φ = − Tf , = − f (x) (x)dx = − f (x) (x)dx
dx dx I dx p dx
s s
!
Z a Z aj+1
X j+1
dφ X aj+1 ′
=− f (x) (x)dx = − f (x)φ(x)|aj − f (x)φ(x)dx .
aj dx aj
j=r−1 j=r−1

We remark also for all j ∈ [[r − 1, s]],

f (x)φ(x)|aaj+1
j
= f (a− +
j+1 )φ(aj+1 ) − f (aj )φ(aj ).

Summing up for all index j, we can obtain the following results since φ(aj ) = 0 for all
j∈/ [[r, s]],
s
X s
X s
X
f (x)φ(x)|aaj+1
j
= (f (a−
j ) − f (a+
j ))φ(aj ) = (f (a− +
j ) − f (aj ))⟨δaj , φ⟩
j=r−1 j=r j=r
*m +
X
− +
= (f (aj ) − f (aj ))δaj , φ .
j=1

At the same time, we have as well as

s
X Z aj+1 Z q Z
′ ′
f (x)φ(x)dx = f (x)φ(x)dx = f ′ (x)φ(x)dx = ⟨Tf ′ , φ⟩.
j=r−1 aj p I

Regrouping the above results in the integration by parts, the formula (11.12) follows.
 11.2. ELEMENTARY DISTRIBUTION THEORY 381

Proposition 11.2.29. Let T ∈ D′ (Ω) and f ∈ C ∞ (Ω), then the mapping

φ ∈ D(Ω) 7→ ⟨T, f φ⟩ (11.13)

defines a distribution.

Proof. We verify T2 : φ 7→ ⟨T, f φ⟩ is a continuous linear form on D(Ω).

Well-defined: Let φ ∈ D(Ω). The function f φ is also a smooth function as the product of two
smooth functions and the derivatives can be obtained from Leibniz formula. Also,
since φ is supported in a compact set K ⊂ Ω, φ(x) = 0 for all x ∈ Ω \ K, which leads
to f φ vanishes on the same subset. Hence, f φ is a smooth and compactly supported
function on Ω, that is, f φ ∈ D(Ω). Thus, T2 (φ) = ⟨T, f φ⟩ is well-defined on D(Ω).

Linearity: The linearity of T2 comes from those of T and multiplication with a fixed function.

Continuity: Consider now a sequence of functions (φk )k∈N which converges to φ in D(Ω). There
exists a common compact set K ⊂ Ω such that for all k ∈ N, supp(φk ) ⊂ K. As an
immediate result, we have supp(f φk ) ⊂ K for all k ∈ N. In parallel, let α ∈ Nn be a
multi-index, it follows from Leibniz formula,
! !
X α X α
α α−β β
∂ (f φk ) = ∂ f ∂ φk −→ ∂ α−β f ∂ β φ = ∂ α (f φ).
β≤α
β k→+∞
β≤α
β

The above convergence while k → +∞ is uniform on Ω since φk → φ in D(Ω). In the

consequence, the sequence of functions (f φk )k∈N converges to f φ in D(Ω). So, using
the continuity of T , we have,

T2 (φk ) = ⟨T, f φk ⟩ −→ ⟨T, f φ⟩ = T2 (φ).

k→+∞

It shows that T2 is continuous on D(Ω).

Definition 11.2.30. Let T ∈ D′ (Ω) and f ∈ C ∞ (Ω), the product of T by a smooth function
f is defined as the distribution given by (11.13), and is denoted by f T ∈ D′ (Ω).

Proposition 11.2.31. Let (Tk )k∈N , T ∈ D′ (Ω) and (fk )k∈N , f ∈ C ∞ (Ω).

• If Tk → T in D′ (Ω), then f Tk → f T in D′ (Ω).

• If ∀α ∈ Nn , ∂ α fk → ∂ α f uniformly on any compact in Ω, then fk T → f T in D′ (Ω).

Proof. Left in Exercise 11.4.4.

Remark 11.2.32. It is impossible to define in a general way the multiplication between two
distributions. The proof of this proposition is contained in the course Partial Differential
Equations.
382 FOURIER TRANSFORMS & DISTRIBUTIONS

11.2.4 Convolution Products

We first introduce the convolution product between two functions.
Definition 11.2.33. Let f, g be functions, we call the convolution product between f
and g and denote it by f ∗ g once the following integral has a sense,
Z
n
∀x ∈ R , f ∗ g(x) = f (x − y)g(y)dy.
Rn

As an immediate result by using a change of variables, it holds that f ∗ g = g ∗ f . We

remark furthermore that the convolution between functions is in fact an essential tool in
many branches of Analysis. For instance in the proof of Theorem 11.2.9 and the construction
of cur-off functions Theorem 11.2.8. For those who are interested, we recommend to consult
the course Partial Differential Equations or the references therein.
Definition 11.2.34. We define the convolution product of a distribution T ∈ D′ (Rn ) and
a function φ ∈ D(Rn ). Let x ∈ Rn , we set
b
T ∗ φ(x) := ⟨T, φ(x − ·)⟩ = T, τx φ ,

where φ(x − ·) refers to the function y 7→ φ(x − y) ∈ D(Rn ).

Proposition 11.2.35. Let T ∈ D′ (Rn ) and φ ∈ D(Rn ). Then,

1.) T ∗ φ is continuous.

2.) T ∗ φ is a smooth function on Rn . Moreover, for all α ∈ Nn , we have ∂ α (T ∗ φ) =

(∂ α T ) ∗ φ = T ∗ (∂ α φ).
Proof. 1.) Let x ∈ Rn and (xk )k∈N be a convergent sequence in Rn such that xk −→ x.
k→+∞
We define then for all k ∈ N, φk : y 7→ φ(xk − y) ∈ D(Rn ). We are about to prove the
sequence (φk )k∈N converges to φ(x − ·) in D(Rn ).
Since the sequence (xk )k∈N converges to x, it is bounded. Then there exists r >
0 such that ∀k ∈ N, xk ∈ B(x, r). Moreover, from the definition of φk , we have
supp(φk ) = {xk − y | y ∈ supp(φ)} ⊂ B(x, r) − supp(φ). Here, B(x, r) − supp(φ) :=

x − y x ∈ B(x, r), y ∈ supp(φ) . From the compactness of B(x, r) and supp(φ),
we deduce that the set K := B(x, r) − supp(φ) is also compact. Thus, for all k ∈ N,
supp(φk ) ⊂ K with K compact.
xk − x
On the other hand, let α ∈ Nn and y ∈ Rn . By setting the unit vector û := ,
|xk − x|
we have

∂ α (φk − φ(x − ·))(y) = (−1)|α| (∂ α φ(xk − y) − ∂ α φ(x − y))

Z |xk −x|
|α| d α
= (−1) ∂ φ(x − y + tû)dt.
0 dt
X
It implies that ∥∂ α (φk − φ(x − ·))∥L∞ (Rn ) ≤ |xk − x| ∥∂ β φ∥L∞ (Rn ) −→ 0.
k→+∞
β∈Nn ,
|β|=|α|+1
In other words, (∂ α φk )k∈N converges uniformly to ∂ α φ. Thus, (φk )k∈N converges to
φ(x − ·) in D(Rn ).
11.2. ELEMENTARY DISTRIBUTION THEORY 383

Now we have (φk )k∈N converges to φ(x − ·) in D(Rn ). From the continuity of T , we
have ⟨T, φk ⟩ −→ ⟨T, φ⟩ in C. In terms of the convolution product, it comes out
k→+∞
T ∗ φ(xk ) −→ T ∗ φ(x). In conclusion, T ∗ φ is continuous on Rn .
k→+∞

2.) For all j ∈ [[1, n]] and t ∈ R∗ , it follows from the straightforward calculation that,

T ∗ φ(x + txbj ) − T ∗ φ(x) 1 1

= ⟨T, φ(x + txbj − ·)⟩ − ⟨T, φ(x − ·)⟩
t t
 t 
φ(x + txbj − ·) − φ(x − ·)
= T, .
t

We can also adapt the method used in the previous point to prove that

φ(x + txbj − ·) − φ(x − ·) ∂φ

−→ (x − ·) in D(Rn ).
t t→0 ∂xj

Hence, T ∗ φ possesses the j-th partial differentiation and it is given by

∂(T ∗ φ) ∂φ
=T ∗ .
∂xj ∂xj

∂φ ∂φ
From the previous point, since ∈ D(Rn ), the function T ∗ is continuous and
∂xj ∂xj
∂(T ∗ φ)
thus, is continuous. So, the function T ∗ φ possesses all continuous partial
∂xj
differentiation, it is therefore continuously differentiable. We use then an immediate
induction to deduce that in fact, T ∗ φ is a smooth function and for any multi-index
α ∈ Nn ,

∂ α (T ∗ φ) = T ∗ (∂ α φ).

Finally, we proceed the calculation, for all y ∈ Rn ,

∂α
 
α α |α|
(∂ T ) ∗ φ(y) = ⟨∂ T, φ(y − ·)⟩ = (−1) T, α φ(y − ·) = ⟨T, ∂ α φ(y − ·)⟩.
∂x

Thus, the claim follows.

Example 11.2.36. Let φ ∈ D(Rn ) and x ∈ Rn .

1.) Let a ∈ Rn , δa ∗ φ(x) = ⟨δa , φ(x − ·)⟩ = φ(x − a). In particular, δ ∗ φ(x) = φ(x).
Z
1 n
2.) Let f ∈ Lloc (R ), Tf ∗ φ(x) = ⟨Tf , φ(x − ·)⟩ = f (y)φ(x − y)dy = f ∗ φ(x). We
Rn
find thus the convolution product between functions.

Remark 11.2.37. We can see in Proposition 11.2.35 that f ∗ φ is a smooth function even
though f ∈ L1loc (Rn ) might be discontinuous. So the convolution product is a regulariz-
ing operator on functions. This property could serve to prove the denseness result Theo-
rem 11.2.9.
384 FOURIER TRANSFORMS & DISTRIBUTIONS

11.3 Fourier Transforms and Distributions

In order to extend the concept of Fourier transforms to the functions not only in L1 (Rn ), we
use the similar approaches as the distributions. In the following subsections, we will define
the space of “test functions” for Fourier transforms, which is called the Schwartz space.
Next, we introduce the generalized Fourier transforms as linear continuous forms defined
on the space of test functions, that is the so-called tempered distributions.

11.3.1 The Schwartz Space

Definition 11.3.1. We call a rapid decreasing function any function φ ∈ C ∞ (Rn )
satisfying

∀ α, β ∈ Nn , ∃ Cα,β > 0 s.t. sup |xα ∂ β φ(x)| ≤ Cα,β .

x∈Rn

We denote by S(Rn ) the set of rapid decreasing functions on Rn . It is immediate that

S(Rn ) is a vector space, and we call it the Schwartz space.

Proposition 11.3.2. • D(Rn ) ⊂ S(Rn ).

2
• For any z ∈ C such that Re(z) < 0, then the function f : x 7→ ez|x| belongs to S(Rn ).

• All rational functionsix are not in S(Rn ).

• For any φ1 , φ2 ∈ S(Rn ), φ1 φ2 ∈ S(Rn ).

• For any φ ∈ S(Rn ) and any polynomial function P , then P φ ∈ S(Rn ).

• For any φ ∈ S(Rn ), and any multi-index α ∈ Nn , then ∂ α φ ∈ S(Rn ).

Proof. In Exercise 11.4.5.

Definition 11.3.3. Let q ∈ N, we denote by Nq the semi-norm on S(Rn ) defined by,

X
∀φ ∈ S(Rn ), Nq (φ) := sup |xα ∂ β φ(x)|.
n
α,β∈Nn x∈R
|α|,|β|≤q

Remark 11.3.4. Using this semi-norm, the definition of S(Rn ) can be reformulate as,

φ ∈ S(Rn ) ⇔ ∀q ∈ N, Nq (φ) < +∞.

Definition 11.3.5. Let (φk )k∈N be a sequence of elements in S(Rn ), and ψ ∈ S(Rn ). We
say the sequence (φk )k∈N converge to ψ in S(Rn ) if for all q ∈ N, Nq (φk − ψ) −→ 0.
k→+∞

ix P (x)
A rational function is defined as the quotient of two polynomial functions, e.g., f (x) = such that
Q(x)
P and Q have no common factors.
11.3. FOURIER TRANSFORMS AND DISTRIBUTIONS 385

Remark 11.3.6. As an immediate result, the multiplications by polynomial functions and

all partial derivations are continuous operations on S(Rn ).

Proposition 11.3.7. For all 1 ≤ p ≤ ∞, it follows that S(Rn ) ⊂ Lp (Rn ). In fact, the space
S(Rn ) is continuously embedded in Lp (Rn ), which means for any sequence (φk )k∈N of
functions, if (φk )k∈N converges in S(Rn ), then it converges in Lp (Rn ).

Proof. See in the course Partial Differential Equations.

Corollary 11.3.8. Let 1 ≤ p < ∞, S(Rn ) is dense in Lp (Rn ).

Proof. Direct consequence of Theorem 11.2.9 and the fact D(Rn ) ⊂ S(Rn ) ⊂ Lp (Rn ).

Theorem 11.3.9. D(Rn ) is dense in S(Rn ).

Proof. See in the course Partial Differential Equations.

With all above properties on the Schwartz space S(Rn ), the Fourier transform F can be
applies on all functions φ ∈ S(Rn ) ⊂ L1 (Rn ) and the corresponding properties in Proposi-
tion 11.1.7 are also available on S(Rn ). We recall them in the below.

Proposition 11.3.10. Let φ ∈ S(Rn ), a ∈ Rn and j ∈ [[1, n]], then the following assertions
hold.
∂
• φ
b is a C 1 -class function and φ(ξ)
b = Fx→ξ (−ixj φ(x))(ξ).
∂ξj
∂φ
• F( )(ξ) = iξj F(φ)(ξ).
∂xj

• Fx→ξ (φ(x − a))(ξ) = e−ia·ξ F(φ)(ξ).

• Fx→ξ (eia·x φ(x))(ξ) = φ(ξ

b − a).

Theorem 11.3.11 (Fourier inversion formula). The Fourier transform F : φ 7→ φ

b is a
bicontinuous automorphism on S(Rn ). Precisely speaking, it satisfies,

• for all φ ∈ S(Rn ), φ

b ∈ S(Rn );

• F is linear and bijective from S(Rn ) into itself;

• the inverse F −1 is given by,

Z b
−1 1 ix·ξ 1 1
F (φ)(x) = e φ(ξ)dξ = φ(−x) = φ(x); (11.14)
(2π)n Rn (2π)n (2π)n
b b

• both F and F −1 are continuous from S(Rn ) into itself.

Proof. Let φ ∈ S(Rn ).

386 FOURIER TRANSFORMS & DISTRIBUTIONS

• We first justify that φ

b ∈ S(Rn ). Let q ∈ N, and α, β ∈ Nn , |α|, |β| ≤ q, it follows from
Proposition 11.3.10 that
ξ α ∂ β φ̂(ξ) = i|α|+|β| Fx→ξ (∂ α (xβ φ(x)))(ξ).
Using Leibniz formula (11.1) and Proposition 11.3.2, we have
!
X α
α β
∂ (x φ) = ∂ γ xβ ∂ α−γ φ ∈ S(Rn ) ⊂ L1 (Rn ).
γ∈Nn
γ
γ≤α

It implies that Fx→ξ (∂ α (xβ φ(x))) ∈ L∞ (Rn ), in other words, Nq (φ)

b < ∞. Since this
condition holds for any q, we have then φ b ∈ S(Rn ).
• Now we are about to justify the inversion formula (11.14). We start with
Z Z Z 
eix·ξ φ(ξ)dξ
b = eix·ξ e−iξ·y φ(y)dy dξ.
Rn Rn Rn

Let x ∈ Rn , we consider the function S(y, ξ) := eix·ξ e−iξ·y φ(y). However, S ∈

/ L1 (Rnξ ×
n
Ry ) since the estimation |S(y, ξ)| = |φ(y)| ∈ 1 n
/ L (Rξ ). Thus, we can not directly apply
Fubini Theorem 8.4.17 to interchange the integrals.
On the other hand, we can apply Lebesgue’s Dominated Convergence Theorem 8.4.12
to prove that
Z Z
2
ix·ξ
e φ(ξ)dξ
b = lim eix·ξ e−ε|ξ| φ(ξ)dξ
b
Rn ε→0+ Rn

Let ε > 0, we have now, |S(y, ξ)e −ε|ξ|2 2

| ≤ e−ε|ξ| |φ(y)| ∈ L1 (Rnξ ×Rny ). Then, we apply
Fubini Theorem 8.4.17 and it holds,
Z Z Z 
ix·ξ −ε|ξ|2 ix·ξ −ε|ξ|2 −iξ·y
e e φ(ξ)dξ
b = e e e φ(y)dy dξ
Rn Rn Rn
Z Z 
−i(y−x)·ξ −ε|ξ|2
= φ(y) e e dξ dy
n Rn
ZR  
2
= φ(y) Fξ→ζ (e−ε|ξ| )(y − x) dy
Rn

We then apply Lemma 11.1.8, it leads to

Z r n Z
π |x−y|2
ix·ξ
e φ(ξ)dξ
b = lim e− 4ε φ(y)dy
Rn ε→0+ ε Rn

(x − y)
We affect a change of variables u = √ to obtain,
2 ε
√ n √
Z Z
2
ix·ξ
e φ(ξ)dξ
b = (2 π) lim e−|u| φ(x − 2 εu)du.
Rn ε→0+ Rn

We then use again the Lebesgue’s Dominated Convergence Theorem 8.4.12 to derive
that
Z
eix·ξ φ(ξ)dξ
b = (2π)n φ(x).
Rn

Thus, (11.14) holds.

11.3. FOURIER TRANSFORMS AND DISTRIBUTIONS 387

• See in the course Partial Differential Equations.

Proposition 11.3.12 (Parseval formula). Let φ, ψ ∈ S(Rn ). Then,

Z Z
1.) φ(x)ψ(x)dx
b = φ(x)ψ(x)dx;
b
Rn Rn
Z Z
1
2.) φ(x)ψ(x)dx = n
φ(x)
b ψ(x)dx.
b
Rn (2π) Rn

Proof. 1.) From the definition of Fourier transform,

Z Z Z 
−ix·y
φ(x)ψ(x)dx
b = e φ(y)dy ψ(x)dx.
Rn Rn Rn

Since S(Rn ) ⊂ L1 (Rn ), the functions considered in the integrals are both integrable
respect to the variables x and y. Then, using Fubini Theorem 8.4.17,
Z Z Z
φ(x)ψ(x)dx
b = e−ix·y φ(y)ψ(x)dydx
Rn Rn Rn
Z Z  Z
−ix·y
= e ψ(x)dx φ(y)dy = φ(x)ψ(x)dx.
b
Rn Rn Rn

1
2.) Let ω = φ,
b we calculate
(2π)n
Z Z
1 1
ω
b (ξ) = e−ix·ξ φ(x)dx = eix·ξ φ(x)dx
(2π)n (2π)n
b b
Rn Rn
= F −1 (φ)(ξ)
b = φ(ξ).

We then apply the first point with ω and ψ, it follows,

Z Z Z Z
1
φ(x)ψ(x)dx = ω
b (x)ψ(x)dx = ω(x)ψ(x)dx
b = φ(x)
b ψ(x)dx.
b
Rn Rn Rn (2π)n Rn

Corollary 11.3.13 (Plancherel). For any φ ∈ S(Rn ), it holds,

1
∥φ∥L2 (Rn ) = ∥φ∥
b L2 (Rn ) .
(2π)n/2

Proof. We endow the space L2 (Rn ) with the Hermitian product, let φ, ψ ∈ L2 (Rn ) with the
values in C,
Z
⟨φ, ψ⟩L2 (Rn ) = φ(x)ψ(x)dx.
Rn

Then, the claim is straightforward using Parseval formula Proposition 11.3.12.

388 FOURIER TRANSFORMS & DISTRIBUTIONS

Using the above propositions, we are able to prove the following results in the Fourier
transforms of convolution products.

Proposition 11.3.14. Let φ, ψ ∈ S(Rn ), then

1.) φ ∗ ψ ∈ S(Rn );

∗ψ =φ
2.) φ[ bψ;
b

1
3.) φψ
c = b ∗ ψ.
φ b
(2π)n

Proof. Left in Exercise 11.4.6.

11.3.2 Tempered Distributions

Definition 11.3.15. We call a tempered distribution any continuous linear form on
S(Rn ). The set of tempered distributions is denoted by S ′ (Rn ).

Remark 11.3.16. Let T ∈ S ′ (Rn ) and φ ∈ S(Rn ).

• We will also denote by ⟨T, φ⟩ at the place of T (φ). In the case of possible confusion,
we might use ⟨T, φ⟩S ′ ×S .

• The continuity of T is understood in the sense that if φk −→ φ in S(Rn ), then

k→+∞
⟨T, φk ⟩ −→ ⟨T, φ⟩ in C.
k→+∞

Example 11.3.17. For all a ∈ Rn , δa ∈ S ′ (Rn ).

Solution. Left in Exercise 11.4.7.

The following proposition shows the relation between tempered distributions S ′ (Rn ) and
distributions D′ (Rn ).

Proposition 11.3.18. The following assertions hold.

1.) Let T ∈ S ′ (Rn ) then there exists C > 0 and q ∈ N such that

∀φ ∈ D(Rn ), ⟨T, φ⟩D′ ×D ≤ CNq (φ). (11.15)

2.) Using the denseness result Theorem 11.3.9, (11.15) holds for φ ∈ S(Rn ) and for
⟨·, ·⟩S ′ ×S

3.) Let T ∈ D′ (Rn ). We assume that there exists C > 0 and q ∈ N such that (11.15)
holds. Then, there exists a unique tempered distribution T̃ ∈ S ′ (Rn ) such that T̃ = T
on D′ (Rn ).

Proof. See in the course Partial Differential Equations.

11.3. FOURIER TRANSFORMS AND DISTRIBUTIONS 389

It is also expected that the regular distributions Tf could be defined similarly by Defi-
nition 11.2.14 in the sens of tempered distributions for all f ∈ L1loc (Rn ). However, not all
functions in L1loc (Rn ) could be defined in the senseZof tempered distributions. For instance,
2 2
the function x 7→ e|x| ∈ L1loc (Rn ) but the integral e|x| φ(x)dx is not well-defined for any
R
φ ∈ S(R). For tempered distributions, we need more additional conditions to define a such
distribution via a given function.

Proposition 11.3.19. The following assertions hold.

1.) Let f ∈ L1loc (Rn ) such that |f (x)| ≤ C(1 + |x|)r for all x ∈ Rn with C > 0 and r ∈ N.
Then, Tf ∈ S ′ (Rn ).

2.) Let 1 ≤ p ≤ ∞ and f ∈ Lp (Rn ). Then Tf ∈ S ′ (Rn ).

3.) For all α, β ∈ Nn , xα ∂ β T ∈ S ′ (Rn ). We remark that the multiplication with smooth
functions and partial derivatives are defined in the same way as the distributions
D′ (Rn ).

Proof. See in the course Partial Differential Equations

Definition 11.3.20. A function f ∈ C ∞ (Rn ) is said to be moderate increasing if for all

β ∈ Nn , there exists Cβ > 0 and mβ ∈ N such that

∀x ∈ Rn , |∂ β f (x)| ≤ Cβ (1 + |x|)mβ .

Proposition 11.3.21. Let T ∈ S ′ (Rn ) and f be a moderate increasing function. Then,

f T ∈ S ′ (Rn ).

Proof. See in the course Partial Differential Equations.

Definition 11.3.22. We say a sequence of tempered distribution (Tk )k∈N converge to T ∈

S ′ (Rn ) if for all φ ∈ S(Rn ), ⟨Tk , φ⟩ −→ ⟨T, φ⟩.
k→+∞

Proposition 11.3.23. Let (Tk )k∈N be a sequence of tempered distributions such that for
any φ ∈ S(Rn ), the complex sequence (⟨Tk , φ⟩)k∈N converges. Then there exits T ∈ S ′ (Rn )
such that (Tk )k∈N converges to T in S ′ (Rn ).

Proof. See in the course Partial Differential Equations.

11.3.3 Fourier Transforms on Tempered Distributions

Let T ∈ S ′ (Rn ). It follows from the linearity and the continuity of the Fourier transform
on S(Rn ), the mapping φ 7→ ⟨T, φ⟩b defines thus a tempered distribution.

Definition 11.3.24. Let T ∈ S ′ (Rn ), we call the Fourier transform of T , and denote by
Tb or F(T ), the tempered distribution,

∀φ ∈ S(Rn ), ⟨Tb, φ⟩ = ⟨T, φ⟩.

b
390 FOURIER TRANSFORMS & DISTRIBUTIONS

Example 11.3.25. We have the Fourier transforms of tempered distributions in S ′ (Rn ).

1.) δb = 1.

2.) δba = ξ 7→ e−ia·ξ

3.) Fx→ξ (eiax ) = (2π)n δa

Let a ∈ Rn and T ∈ S ′ (Rn ), we also denote by τa T the tempered distribution defined

by, ∀φ ∈ S(Rn ), ⟨τa T, φ⟩ = ⟨T, τ−a φ⟩. Such definition is deduced from the following result
(the justification is left in exercises). For any regular distribution Tf with f ∈ L1loc (Rn ), we
have τa Tf = Tτa f . Similarly to Proposition 11.1.7, we have the following properties on the
Fourier transforms of tempered distributions.

Proposition 11.3.26. Let T ∈ S ′ (Rn ) and j ∈ [[1, n]], the following assertions hold.
∂ b \
1.) T = (−ix j )T
∂ξj

\∂
2.) T = iξj Tb
∂xj

−ia·x T
3.) τa Tb = e\
ia·ξ b
4.) τd
aT = e T

Proof. Left in Exercise 11.4.8.

Theorem 11.3.27. The Fourier inversion formula for tempered distributions is given by,

S(Rn ) −→ C
∀T ∈ S ′ (Rn ), F −1 (T ) : .
φ 7−→ ⟨T, F −1 (φ)⟩

Proof. Straightforward from the definition of Fourier transforms of tempered distributions

and the Fourier inversion formula Theorem 11.3.11.

Proposition 11.3.28. The Fourier transform induces an isometry on L2 (Rn ).

Precisely speaking, let f ∈ L2 (Rn ), we consider the tempered distribution
Z
Tf ∈ S ′ (Rn ) : φ 7→ f (x)φ(x)dx.
Rn

Then, F(Tf ) ∈ L2 (Rn ), in the sense of there exists a unique function g ∈ L2 (Rn ) such that
Tg = F(Tf ) in S ′ (Rn ). Furthermore, we have

1
∥f ∥L2 (Rn ) = ∥g∥L2 (Rn ) . (11.16)
(2π)n/2

Proof. See in the course Partial Differential Equations.

11.3. FOURIER TRANSFORMS AND DISTRIBUTIONS 391

Definition 11.3.29. We define the convolution product of a tempered distribution T ∈

S ′ (Rn ) and a rapid decreasing function φ ∈ S(Rn ) exactly by the same way as those in
D′ (Rn ) × D(Rn ). Let x ∈ Rn ,
b
T ∗ φ(x) = ⟨T, φ(x − ·)⟩ = T, τx φ .

We have the analogue results as Proposition 11.2.35.

Theorem 11.3.30. Let T ∈ S ′ (Rn ) and φ ∈ S(Rn ). Then, T ∗ φ ∈ C ∞ (Rn ) and is a

moderate increasing function. Moreover, ∂ α (T ∗ φ) = (∂ α T ) ∗ φ = T ∗ (∂ α φ).

Proof. We adapt the similar method used in Proposition 11.2.35.

Let x ∈ Rn and (xk )k∈N be a convergent sequence in Rn such that xk −→ x. We
k→+∞
define then for all k ∈ N, φk : y 7→ φ(xk − y) ∈ S(Rn ). We are about to prove the sequence
(φk )k∈N converges to φ(x − ·) in S(Rn ).
Let q ∈ N and α, β ∈ Nn be multi-indexes such that |α|, |β| ≤ q. Then the straightfor-
ward calculation gives, for all y ∈ Rn ,

∂ β (φk − φ(x − ·))(y) = (−1)|β| [∂ β φ(xk − y) − ∂ β φ(x − y)]

Z |xk −x|
|β| d β
= (−1) ∂ φ(x − y + tû)dt,
0 dt
xk − x
where û := deigns a unit vector. Hence, for all y ∈ Rn ,
|xk − x|

y α ∂ β (φk − φ(x − ·))(y) ≤ Nq+1 (φ)|xk − x| −→ 0.

k→+∞

This implies that φk −→ φ(x − ·) in S(Rn ). Next, it follows from the continuity of T , we
k→+∞
obtain that

T ∗ φ(xk ) = ⟨T, φ(xk − ·)⟩ = ⟨T, φk ⟩ −→ ⟨T, φ(x − ·)⟩ = T ∗ φ(x).

k→+∞

Thus, T ∗ φ is continuous.
Next, we can follow the same arguments in the proof of Proposition 11.2.35 to show
that T ∗ φ possesses the continuous partial derivatives. So, T ∗ φ is a C ∞ -class function and
the partial derivatives are given by ∂ α (T ∗ φ) = (∂ α T ) ∗ φ = T ∗ (∂ α φ) for all α ∈ Nn .
Now, it remains to show that T ∗ φ is a moderate increasing function. It follows from
Proposition 11.3.18 there exists CT > 0 and q ∈ N such that |⟨T, ψ⟩| ≤ CT Nq (ψ) for all
ψ ∈ S(Rn ). Hence, let x ∈ Rn ,

|∂ β (T ∗ φ)(x)| = |T ∗ (∂ β φ)(x)| = ⟨T, ∂ β φ(x − ·)⟩

≤ CT Nq (∂ β φ(x − ·)) ≤ CT Nq+|β| (φ(x − ·)).

Recalling the semi-norm Nq̃ , it comes out

X X
Nq̃ (φ(x − ·)) := sup |y α̃ ∂ β̃ φ(x − y)| = sup |(x − z)α̃ ∂ β̃ φ(z)|.
y∈Rn z∈Rn
α̃,β̃∈Nn α̃,β̃∈Nn
|α̃|,|β̃|≤q |α̃|,|β̃|≤q
392 FOURIER TRANSFORMS & DISTRIBUTIONS

By developing the polynomial (x − z)α̃ , we obtain that

X
|(x − z)α̃ | ≤ Cα̃ (1 + |x|)|α̃| |z γ |,
γ∈Nn ,|γ|≤|α̃|

where Cα̃ > 0 which comes from the combination coefficients and it depends only on α̃.
Summing all up, we can deduce thus,

Nq̃ (φ(x − ·)) ≤ Cq̃ (1 + |x|)q̃ Nq̃ (φ),

with the constant Cq̃ depends only on q̃. We finally obtain that,

|∂ β (T ∗ φ)(x)| ≤ CT Cq+|β| Nq+|β| (φ)(1 + |x|)q+|β| = Cβ (1 + |x|)mβ .

Thus, T ∗ φ is a moderate increasing function.

Theorem 11.3.31. The convolution product T ∗ φ can be also considered as a tempered

distribution. In addition, for all ψ ∈ S(Rn ),
b
⟨T ∗ φ, ψ⟩ = ⟨T, φ ∗ ψ⟩ (11.17)

Proof. From Theorem 11.3.30, the convolution T ∗ φ is a smooth and moderate increasing
function. We can apply Proposition 11.3.19 to consider T ∗ φ as a tempered distribution.
It is then given by, for all ψ ∈ S(Rn ),
Z Z Z
⟨T ∗ φ, ψ⟩ = T ∗ φ(x)ψ(x)dx = ψ(x)⟨T, φ(x − ·)⟩dx = ⟨T, ψ(x)φ(x − ·)⟩dx.
Rn Rn Rn

Next, we are about to justify (11.17).

At first, we suppose that ψ ∈ D(Rn ) and supp(ψ) ⊂ K with K being compact in Rn .
Rn × Rn −→ C
We define a function f : . Let y ∈ Rn , it follows from the
(x, y) 7−→ ψ(x)φ(x − y)
Z b
convolution product of functions that f (x, y)dx = φ ∗ ψ(y). Since K is compact, there
Rn
exists then A > 0 such that K ⊂ [−A, A]n . We define a function g on Rn × Rn as follows.
For all x = (x1 , · · · , xn ) ∈ Rn and y ∈ Rn ,
Z xn Z x1
g(x, y) = ··· f ((t1 , · · · , tn ), y)dt1 · · · dtn .
−A −A

Using the Theorem of Derivation under Integral Signs Theorem 8.4.15, we can obtain that
for all x ∈ Rn and q ∈ N, Nq (g(x, ·)) ≤ ∥ψ∥L∞ (Rn ) (2A)n Nq (φ). Hence, g(x, ·) ∈ S(Rn ).
Consider now the function G defined by, for all x ∈ Rn ,

G(x) = ⟨T, g(x, ·)⟩.

We can thus follow the proof of Theorem 11.3.30 to show that G is a smooth function and
for any multi-index α ∈ Nn ,

∂ α G(x) = ⟨T, ∂xα g(x, ·)⟩ (11.18)

11.3. FOURIER TRANSFORMS AND DISTRIBUTIONS 393

∂n
Taking now α = (1, · · · , 1), we have ∂xα g(x, y) = g(x, y) = f (x, y). We then
∂x1 · · · ∂xn
integrate the relation (11.18) to obtain that
Z xn Z x1 Z xn Z x1
··· ⟨T, f (t, ·)⟩dt1 · · · dtn = ··· ⟨T, ∂xα g(t, ·)⟩dt1 · · · dtn
−A −A −A −A
Z xn Z x1
= ··· ∂ α G(t)dt1 · · · dtn = G(x) = ⟨T, g(x, ·)⟩
−A Z xn −A Z x1 
= T, ··· f (t, ·)dt1 · · · dtn
−A −A
We can thus choose x1 = · · · = xn = A and use the fact that the function f is supported in
K × Rn to deduce that
Z Z  Z  b
⟨T ∗ φ, ψ⟩ = ⟨T, ψ(t)φ(t − ·)⟩dt = ⟨T, f (t, ·)⟩dt = T, f (t, ·)dt = ⟨T, φ ∗ ψ⟩.
Rn Rn Rn

Hence, (11.17) follows for ψ ∈ D(Rn ).

Finally, since T ∈ S ′ (Rn ) and following from Proposition 11.3.18, there exists C > 0
and q ∈ N such that for all ψ ∈ D(Rn ) ⊂ S(Rn ),
b b
T, φ ∗ ψ D′ ×D ≤ CNq (φ ∗ ψ).
b b
Also, by calculating the semi-norm Nq (φ ∗ ψ), we can obtain the estimation Nq (φ ∗ ψ) ≤
∥φ∥L∞ (Rn ) Nq (ψ). Hence,
b
T, φ ∗ ψ D′ ×D ≤ C∥φ∥L∞ (Rn ) Nq (ψ).
b
Apply again Proposition 11.3.18 the distribution ψ 7→ ⟨T, φ ∗ ψ⟩ defines a tempered distri-
bution. Thus, the claim follows.

The following theorem states the Fourier transforms of a convolution product in S ′ (Rn )×
S(Rn ). In short, it is a generalization of Proposition 11.3.14.
Corollary 11.3.32. Let T ∈ S ′ (Rn ) and φ ∈ S(Rn ). Then it follows that
1.) F(T ∗ φ) = F(φ)F(T ).
1
2.) F(φT ) = F(T ) ∗ F(φ).
(2π)n
Proof. 1.) We consider here T ∗ φ as a tempered distribution. Let ψ ∈ S(Rn ), it follows
from Theorem 11.3.31
b
⟨F(T ∗ φ), ψ⟩ = ⟨T ∗ φ, ψ⟩ = ⟨T, φ ∗ ψ⟩.
b b

For all y ∈ Rn , we deduce from Parseval’s formula Proposition 11.3.12 that

b Z b Z Z
φ ∗ ψ(y) =
b φ(y − x)ψ(x)dx =
b φ(x − y)ψ(x)dx =
b \
φ(x − y)ψ(x)dx
R n Rn Rn
Z
= e−iy·x φ(x)ψ(x)dx
b = F(φψ)(y).
b
Rn
Hence,
b
⟨F(T ∗ φ), ψ⟩ = ⟨T, φ ∗ ψ⟩
b = ⟨T, F(φψ)⟩
b = ⟨F(T ), φψ⟩
b = ⟨F(φ)F(T ), ψ⟩.
Thus, the claim follows.
394 FOURIER TRANSFORMS & DISTRIBUTIONS

2.) We proceed similarly, let ψ ∈ S(Rn ), it follows from the definition of Fourier trans-
forms of tempered distributions and the multiplication by smooth functions, we have
that

⟨F(φT ), ψ⟩ = ⟨φT, ψ⟩
b = ⟨T, φψ⟩.
b

Also, it follows from Proposition 11.3.14 that

b b b
−1 1 1 1 [
φψb = F (φ) b ∗ ψ.
b
b ψb = φ
bbψb = φ
bψb = φ
(2π)n (2π)n (2π)n

Applying again Theorem 11.3.31, we have

 b 
D E 1 [
⟨F(φT ), ψ⟩ = T, φψb = n
b∗ψ
T, φ
(2π)
b
1 1
= n
⟨F(T ), φ
b ∗ ψ⟩ = ⟨F(T ) ∗ φ,
b ψ⟩.
(2π) (2π)n

Thus, the claim follows.

11.4 Exercises
Exercise 11.4.1. Justify Proposition 11.2.4.

Exercise 11.4.2. Justify Proposition 11.2.6.

Exercise 11.4.3. Justify Proposition 11.2.24.

Exercise 11.4.4. Justify Proposition 11.2.31.

Exercise 11.4.5. Justify Proposition 11.3.2.

Exercise 11.4.6. Justify Proposition 11.3.14.

Exercise 11.4.7. Justify Example 11.3.17.

Exercise 11.4.8. Justify Proposition 11.3.26.

Exercise 11.4.9. For k ∈ N∗ , we introduce the function fk defined on R given by

(
1
fk (x) = k if |x| < k,
0 if |x| ≥ k.
11.4. EXERCISES 395

1.) Show that the sequence of functions (fk )k∈N converges uniformly on R.

2.) Does (fk )k∈N converge in L1 (R)?

3.) Calculate, for each k ∈ N∗ , the Fourier transform fbk .

4.) Is the sequence of functions (fbk )k∈N convergent? Write down your ideas.

Solution. 1.) Let x ∈ R. There exists certain k0 ∈ N∗ such that |x| < k0 and hence,
1
∀k ∈ N∗ , k ≥ k0 > |x| we have, fk (x) = −→ 0. Thus, (fk )k∈N converges
k k→+∞
pointwisely to the function 0. Moreover, we have also,

1
∥fk − 0∥L∞ (R) = sup |fk (x)| = −→ 0.
x∈R k k→+∞

Hence, (fk )k∈N converges uniformly to 0 on R.

2.) From 1.), we have already determinate the limit function 0. If the sequence (fk )k∈N
converges in L1 (R), then the limit function must be 0 (almost everywhere on R). We
then calculate for each k ∈ N∗ ,
Z +∞ Z k
1
∥fk − 0∥L1 (R) = |fk (x) − 0|dx = dx = 2 ̸−→ 0.
−∞ −k k k→+∞

So, (fk )k∈N does not converges in L1 (R).

3.) Let k ∈ N∗ and ξ ∈ R, we calculate,

Z +∞ Z k
1 −iξx
fbk (ξ) = fk (t)e−iξx dx = e dx
−∞ −k k
1 e−iξk − eiξk sin(kξ)
= =2 = 2sinc(kξ).
k −iξ kξ

sin(kξ)
4.) We have, for all k ∈ N, fbk (0) = 2sinc(0) = lim 2 = 2. Also, let ξ ∈ R∗ , fbk (ξ) =
ξ→0 kξ
sin(kξ)
2 −→ 0. Thus, the sequence of functions (fbk )k∈N converges pointwisely to
kξ k→+∞
the function f˜ given by,
(
2 if ξ = 0
f˜(ξ) =
0 if ξ ̸= 0.

This convergence can not be uniform. Since for any k ∈ N∗ the function fbk is continu-
ous over R, if this convergence is uniform, it follows from the theorem of continuity of
uniform convergent sequence of functions that the limit function f˜ must be continuous
over R. It is clear that f˜ is not continuous on R. By contraposition, this sequence of
functions (fbk )k∈N does not converge uniformly.
396 FOURIER TRANSFORMS & DISTRIBUTIONS

Exercise 11.4.10. Consider the function 1(−1,1) defined by,

(
1 if −1 < x < 1
1(−1,1) (x) =
0 if not.

1.) Calculate the Fourier transform of 1(−1,1) .

2.) Justify that 1(−1,1) ∈ L2 (R) and also 1\ 2
(−1,1) ∈ L (R).

+∞
sin2 x
Z
3.) Calculate the integral dx.
−∞ x2

Exercise 11.4.11. Let a ∈ R. Calculate the Fourier transforms of the following functions.
x2
You may consider the Gaussian function G : x 7→ e− 2 on R.
(x−a)2
1.) f : x 7→ e− 2 .
x2
2.) f : x 7→ sin(ax)e− 2 .
x2
3.) f : x 7→ e−|a| 2 .
x2
4.) f : x 7→ xe− 2 .

Exercise 11.4.12. Let a > 0, and b, c ∈ R. Justify and calculate the Fourier transform of
R −→ R
the function f : −(ax 2 +bx+c) .
x 7−→ e

Z +∞
2
Exercise 11.4.13. Calculate the integral e−x cos(7x)dx.
−∞

Exercise 11.4.14. Determinate weather the following functional T defines a distribution

on R or not.
Z 1
1.) ⟨T, φ⟩ = φ(t)dt.
0
Z 1
2.) ⟨T, φ⟩ = |φ(t)|dt.
0

+∞
X
3.) ⟨T, φ⟩ = φ(j).
j=1

+∞
X 1
4.) ⟨T, φ⟩ = φ( ).
j
j=1
11.4. EXERCISES 397

Exercise 11.4.15. Let φ ∈ D(R). For all n ∈ Z, we define

Z
cen (φ) = einx φ(x)dx.
R

1.) Justify the definitions of cen (φ) for all n ∈ Z.

2.) Show that for all k ∈ N, there exists Mφ,k > 0 such that

∀n ∈ Z, |nk cen (φ)| ≤ Mk,φ .

A
n )n∈Z ∈ C such that an = O(|n| ) while n → ±∞ with A ∈ R. Show that the
3.) Let (aX Z

series an cen (φ) converges.

n

+∞
X
4.) Deduce that the mapping T : φ 7→ an cen (φ) is a distribution.
n=−∞

5.) Show that if (an )n∈Z ∈ ℓ1 (Z, C), then T is a regular distribution.

Solution. 1.) It is sufficient to prove the integrability of the function x 7→ einx φ(x). This
function is a continuous function as a product of two continuos ones and it has compact
support on R since φ ∈ D(R). So, this function is integrable over R and the number
cen (φ) is well-defined.

2.) Since φ ∈ D(R), we may assume that supp(φ) ⊂ [−Lφ , Lφ ] with Lφ ∈ R+ . Since the
case n = 0 is trivial, we take n ∈ Z∗ and apply the integration by parts to obtain
that,
Lφ Lφ Lφ
einx einx ′
Z Z
cen (φ) = einx φ(x)dx = φ(x) − φ (x)dx
−Lφ in −Lφ −Lφ in
Z Lφ
i
= einx φ′ (x)dx.
n −Lφ

Here, since supp(φ) ⊂ [−Lφ , Lφ ], we have that φ(−Lφ ) = φ(Lφ ) = 0. Repeat this
processes k times, we obtain that,

ik Lφ inx (k)
Z
cen (φ) = k e φ (x)dx.
n −Lφ

In the consequences,
Z Lφ Z Lφ
k inx (k)
|n cen (φ)| = e φ (x)dx ≤ |φ(k) (x)|dx ≤ 2Lφ ∥φ(k) ∥L∞ (R) .
−Lφ −Lφ

Hence, it is sufficient to choose Mk,φ = 2Lφ ∥φ(k) ∥L∞ (R) to obtain the desired estima-
tion.

3.) The condition an = O(|n|A ) can be reformulate as

∃C > 0 s.t. ∀n ∈ Z, |an | ≤ C|n|A .

 398 FOURIER TRANSFORMS & DISTRIBUTIONS

Apply the result in the previous question, we have,

Mk,φ
∀n ∈ Z∗ , |an cen (φ)| ≤ C|n|A = CMk,φ |n|A−k .
|n|k

Therefore, it is sufficient to choose k ∈ N such that A − k ≤ −2 (for instance,

1
k ≥ [A] + 3) to obtain the asymptotic behavior an cen (φ) = O( 2 ). Combining with
X 1 X n
the convergence of the series , the series an cen (φ) converges.
n
n2 n

4.) We verify that T is a continuous linear form on D(R).

Well-defined: Straightforward consequence of the previous question.

Linearity: It follows from the linearity of integration and those of sum.
Continuity: Consider a sequence of functions (φj )j∈N which converges to the null function in
D(R). Since the supports of each φj are contained in a common compact set,
which can be assumed to be [−L, L] with L ∈ R+ , we can thus set Lφj = L in
the estimation in question 2.) for all j ∈ N. Next, we choose k ∈ N, k ≥ [A] + 3.
It follows from the questions 2.) and 3.), we have,

(k) 1
∀j ∈ N, ∀n ∈ Z∗ , |an cen (φj )| ≤ CMk,φj |n|A−k ≤ 2CL∥φj ∥L∞ (R) .
n2
In the consequence, for all j ∈ N,
+∞ +∞
X (k)
X 1
|T (φj )| ≤ |an cen (φj )| ≤ |a0 |L∥φj ∥L∞ (R) + 4CL∥φj ∥L∞ (R)
n=−∞
n2
n=1
2π 2 (k)
≤ |a0 |L∥φj ∥L∞ (R) + CL∥φj ∥L∞ (R) .
3
Now, we pass j → +∞ and it follows from the uniform convergence of φj as well
(k)
as that of φj , we conclude that

T (φj ) −→ 0.
j→+∞

Combining with the linearity of T , the mapping T is continuous over D(R).

X
5.) Since the series an is absolutely convergent, it follows from the result (Theo-
n X
rem 10.1.16) in the course that the trigonometric series an en is normally (hence
n
uniformly) convergent on R. We denote then the sum function of this series by f .
That is ,
+∞
X
∀x ∈ R, f (x) = an einx .
n=−∞

Also, since for all n ∈ Z, the function an en : x 7→ an einx is continuous on R, it

follows from the result (Theorem 7.3.19) in the course that the sum function f is also
continuous on R. We can thus deduce, f ∈ L1loc (R).
11.4. EXERCISES 399
X
Let φ ∈ D(R). Consider now the series of functions gn indexed by Z and given by,
n

∀n ∈ Z, ∀x ∈ R, gn (x) = an einx φ(x).

Taking the L∞ norm on those functions, we have,

∀n ∈ Z, ∥gn ∥L∞ (R) ≤ ∥φ∥L∞ (R) |an |.

X X
The convergence of |an | implies that the series of functions gn is normally
n n
(hence uniformly) convergent on R. It is also immediate from the pointwise conver-
gence that
+∞
X
∀x ∈ R, gn (x) = f (x)φ(x).
n=−∞

We remark also that for all n ∈ Z supp(gn ) ⊂ supp(φ), and hence,

Z Z
gn (x)dx = an einx φ(x)dx.
R supp(φ)

Apply the result (Proposition 8.4.11 or you can apply dominated convergence theorem)
in the course, we have
+∞
X +∞
X Z
⟨T, φ⟩ = an cen (φ) = an einx φ(x)dx
n=−∞ n=−∞ R
+∞ Z
X +∞
X Z Z +∞
X
= gn (x)dx = gn (x)dx = gn (x)dx
n=−∞ R n=−∞ supp(φ) supp(φ) n=−∞
Z Z
= f (x)φ(x)dx = f (x)φ(x)dx = ⟨Tf , φ⟩ .
supp(φ) R

+∞
X
Hence, T is the regular distribution associate to f (x) = an einx .
n=−∞

Exercise 11.4.16. Let α ∈ D(R) such that α(0) = 1.

1.) Show that

φ(x) − φ(0)α(x)
∀φ ∈ D(R), x 7→ ∈ D(R).
x
Hint: Use Hadamard’s Lemma, see Exercise 8.5.10.

2.) Let T ∈ D′ (R) such that xT = 0 in D′ (R). Show that T = Cδ with C ∈ C being a
constant.

3.) Let a ∈ R. Resolve the equation (x − a)T = 0 in D′ (R).

400 FOURIER TRANSFORMS & DISTRIBUTIONS

Exercise 11.4.17. The main objective of this exercise is to calculate the Fourier transform
of the Heaviside function. We recall the following functions.
• the Heaviside function H,
(
1 if x > 0,
H(x) =
0 if x ≤ 0.

1
• the principle value of ,
x
Z
1 φ(x)
∀φ ∈ S(R), ⟨p.v.( ), φ⟩ = lim dx.
x ε→0 |x|≥ε x

x2
• the Gaussian g, ∀x ∈ R, g(x) = e− 2 .
In the following questions, we shall consider H as a tempered distribution in S ′ (R).
1
1.) Justify that xp.v.( ) = 1 in S ′ (R).
x
dH
2.) Calculate in S ′ (R).
dx
3.) Calculate δ.
b

b − p.v.( 1 ) = Cδ for certain C ∈ C.

4.) Show that iH
x
5.) Give, without proof, the Fourier transform gb.

6.) Calculate ⟨H,

b g⟩.

1
7.) Calculate ⟨p.v.( ), g⟩.
x
b = −ip.v.( 1 ) + πδ.
8.) Show that H
x

Exercise 11.4.18 (Limit of Dirichlet’s Kernels). Now, we set I = (−π, π) ⊂ R. Recall that,
m
R −→ C X
for all k ∈ Z, ek : and the Dirichlet’s kernel of degree m ∈ N, Dm = ek .
t 7−→ eikt k=−m
We will consider the functions ek and Dm as regular distributions in D′ (I).
1.) Let φ ∈ D(I). Show that there exists Cφ > 0 such that
Cφ
∀k ∈ Z, |⟨ek , φ⟩| ≤ .
1 + k2

2.) Deduce that (Dm )m∈N converges in D′ (I).

3.) Show that

sin(m + 12 )t
∀m ∈ N, ∀t ∈ I, Dm (t) = .
sin 2t
11.4. EXERCISES 401

4.) Show that the following result holds in D′ (I)

lim Dm = 2πδ.
m→+∞

Hint: We can use Hadamard’s Lemma, see Exercise 8.5.10, and Riemann-Lebesgue’s
Lemma.

Exercise 11.4.19. Determinate the following limits in D′ (R) when h → 0.

1
1.) (δh − δ−h );
2h
1
2.) (δ2h + δ−2h − 2δ0 );
4h2
1 x2
3.) √ e− h2 .
πh

Exercise 11.4.20. Show that, the solutions in D′ (R) to the equation T ′ = 0 are constant
distributions.

Exercise 11.4.21. Calculate the following derivatives in D′ (R).

1
1.) T = arctan( );
x
2.) T = ae−a|x| with a > 0;

3.) T = ln |x|.

Exercise 11.4.22 (Uncertainty Principle). In this problem, ℏ > 0 denotes a positive con-
stant. Let φ ∈ S(R), we consider a function ψ defined by,
Z +∞
1
∀x ∈ R, ψ(x) = √ φ(p)eipx/ℏ dp.
2πℏ −∞
1 b p
1.) Show that ∀p ∈ R, φ(p) = √ ψ( ).
2πℏ ℏ
r
ℏ b′ p
2.) Deduce that ∀p ∈ R, pφ(p) = −i ψ ( ).
2π ℏ
3.) We define the following quantities,
Z +∞
σx2 := x2 |ψ(x)|2 dx,
−∞
Z+∞
σp2 := p2 |φ(p)|2 dp.
−∞

Justify that σx and σp are well-defined.

402 FOURIER TRANSFORMS & DISTRIBUTIONS

4.) Show that

ℏ2 b′ 2
σp2 = ∥ψ ∥L2 (R) = ℏ2 ∥ψ ′ ∥2L2 (R) .
2π

5.) Prove the following inequality

ℏ
σx σp ≥ ∥ψ∥2L2 (R) .
2

Exercise 11.4.23. Let φ ∈ S(R) and p > 0.

X
1.) Show that the series φ(pk) converges in C.
k∈Z
X
2.) Deduce that that the series of tempered distributions δpk converges in S ′ (R).
k∈Z

+∞
X
3.) We introduce the sampling functionx S given by, S := δk . Show that
k=−∞

+∞
X
Sb = 2π δ2πk .
k=−∞

Hint: Here, it is essential to use Poisson summation formulaxi , which states that
+∞ +∞
X 1 X
φ(2πk) = φ(k).
b
2π
k=−∞ k=−∞

4.) Prove that in S ′ (R) it holds,

+∞ +∞
X 2π X
F( δpk ) = δ( 2π )k . (⋆)
p p
k=−∞ k=−∞

Hint: Here, it is also essential to use the scaling formula for Fourier transforms,
1 ξ
Fx→ξ (f (ax))(ξ) = n fb( ).
|a| a

Exercise 11.4.24. In this exercise, we consider a function f ∈ PC 2π (R, C) and a sequence

(ck )k∈Z of complex numbers indexed by Z. We introduce and suppose furthermore that
• The function f0 is defined by
(
f (x) if x ∈ [0, 2π]
f0 (x) =
0 if x ∈
/ [0, 2π]

• (ck )k∈Z is moderate increasing, i.e., there exists C > 0 and m ∈ N such that ∀k ∈ Z,
|ck | ≤ C(1 + |k|)m .
x
This function is also called as Dirac comb, impulse train or sha function
xi
See in Exercise 10.4.7.
11.4. EXERCISES 403

N
X
1.) Let p > 0 and N ∈ N, we introduce the tempered distribution TN := ck δpk .
k=−N
Show that the sequence (TN )N ∈N converges in S ′ (R).

2.) For all k ∈ Z, we ipkx as tempered distributions. Prove

Xconsider the functions x 7→ e
that the series ck eipkx converges in S ′ (R).
k∈Z

3.) For k ∈ Z, calculate the Fourier transform in S ′ (R) of the tempered distribution δpk .

4.) Show that fb0 is a smooth and moderate increasing function on R.

5.) Justify that the mapping given by

Z
∀φ ∈ S(R), ⟨Tf , φ⟩ = f (x)φ(x)dx
R

defines a tempered distribution.

6.) Let φ ∈ S(R). Show that

+∞ Z
X 2π
⟨T
cf , φ⟩ = f0 (t)φ(t
b + 2πk)dt.
k=−∞ 0

7.) Following the previous question and letting k ∈ Z, show that

Z 2π Z +∞
f0 (t)φ(t
b + 2πk)dt = fb0 (x)e−i2πkx φ(x)dx = ⟨fb0 δd
2πk , φ⟩.
0 −∞

8.) By using Poisson Summation formula (⋆) in Exercise 11.4.23, show that
+∞
X
T
cf = fb0 (k)δk . (∗)
k=−∞

9.) Show that, in S ′ (R), the following relation holds,

+∞ +∞
1 X b ikx
X
f (x) = f0 (k)e = ck (f )eikx .
2π
k=−∞ k=−∞

Exercise 11.4.25. Consider a function f ∈ PC 2π (R, C).

1.) Show that the mapping Tf given by
Z 2π
∀φ ∈ S(R), ⟨Tf , φ⟩ = f (t)φ(t)dt
0

defines a tempered distribution.

2.) Let φ ∈ S(R). Show that there exists Cφ > 0 such that

|⟨Tf , φ⟩| ≤ Cφ ∥f ∥L2 ((0,2π),C) .

404 FOURIER TRANSFORMS & DISTRIBUTIONS

3.) Deduce that if a sequence (fN )N ∈N of piecewise continuous 2π-periodic functions

converges in L2 ((0, 2π), C), then (TfN )N ∈N converges in S ′ (R).

4.) Prove the relation (∗) by using the theorem of convergence of Fourier series (you
should give which type).

5.) Let p > 0 and let f be a piecewise continuous and p-periodic function mapping from
R into C. Prove that
+∞
cf = 2π 2πk
X
T fb0 ( )δ 2πk .
p p p
k=−∞
 Appendix A

Comparison Relations of Functions

and Sequences

In this lecture note, we present the appendices A and B concerning to introduce the asymp-
totic behaviors of functions defined on R as well as those of sequences of real numbers. In
Appendix A, we will present the concept of using the symbols big O, small o as well as the
equivalence -. The ideas of those notations were originated from Bachmann and Landaui .
Those notations provide a useful tool in mathematical analysis to describe precisely the
behavior of the considered function in a particular point or at infinity by using knowledge
of the usual functions such like polynomials, fractionals, exponentials or logarithms. An-
other important scenario to use those notations appears when it comes to describe the time
complexity of algorithms in Computer Sciences.
In this entire chapter, except for where is explicitly mentioned, we consider the functions
defined in a neighborhood of a ∈ R := R∪{±∞}. Precisely speaking, any function appeared
in this chapter is real-valued and defined on an interval I ⊂ R, which corresponds to one of
the following type:

• I = (a − h, a + h) with h > 0 while a ∈ R,

• I = (M, +∞) with M ∈ R while a = +∞,

• I = (−∞, M ) with M ∈ R while a = −∞.

The interval I is called a neighborhood interval of a and we always assume that the
considered functions are continuous on I.

A.1 Domination and Negligence Relations of Functions

Definition A.1.1. Let a ∈ R and f , g be functions defined on a neighborhood I of a.

1.) We say that f is dominated by g in a neighborhood of a if there exists a bounded

function u defined on I such that f = ug on I. We denote this relation by f = O(g)
a
or by f (x) = O(g(x)).
x→a

i
Paul Bachmann (1837 - 1920) and Edmund Landau (1877 - 1938) were German mathematicians.

405
406 COMPARISON RELATIONS OF FUNCTIONS AND SEQUENCES

2.) We say that f is negligible in front of g in a neighborhood of a if there exists a

function ε defined on I such that f = εg and lim ε(x) = 0. We denote this relation
x→a
by f = o(g) or by f (x) = o(g(x)).
a x→a

   
1 2 2 1
Example A.1.2. 1.) x sin = O(x ). Since x 7→ sin is bounded.
x x→0 x
   
2 1 3/2 1/2 1
2.) x sin = o(|x| ). Since lim |x| sin = 0.
x x→0 x→0 x

x2
3.) x2 = o(sin x). Since lim = 0.
x→0 x→0 sin x

4.) A function f is bounded in a neighborhood of a if, and only if f = O(1).

a

5.) A function f converges to 0 at a if, and only if f = o(1).

a

The following results can be immediately deduced from the properties of bounded func-
tions or functions which tend to 0. Let a ∈ R and f , f1 , f2 , g, g1 , g2 design the functions
defined on a neighborhood I of a.

• If f = o(g), then f = O(g).

a a

• If f1 = O(g) and f2 = O(g), then f1 + f2 = O(g).

a a a

• If f1 = O(g1 ) and f2 = O(g2 ), then f1 f2 = O(g1 g2 ).

a a a

• If f1 = o(g) and f2 = o(g), then f1 + f2 = o(g).

a a a

• If f1 = o(g1 ) and f2 = O(g2 ), then f1 f2 = o(g1 g2 ).

a a a

• If f1 = o(g1 ) and f2 = o(g2 ), then f1 f2 = o(g1 g2 ).

a a a

• If f = O(g1 ) and g1 = O(g2 ), then f = O(g2 ).

a a a

• If f = o(g1 ) and g1 = O(g2 ), then f = o(g2 ).

a a a

• If f = O(g1 ) and g1 = o(g2 ), then f = o(g2 ).

a a a

• If f = o(g1 ) and g1 = o(g2 ), then f = o(g2 ).

a a a

Proposition A.1.3. The following results holds.

1.) If f = O(g) and g is bounded on I, then f is bounded on I.

a

2.) If f = O(g) and lim g(x) = 0, then lim f (x) = 0.

a x→a x→a

3.) If f = o(g) and g is bounded on I, then lim f (x) = 0.

a x→a
A.1. DOMINATION AND NEGLIGENCE RELATIONS OF FUNCTIONS 407

Proof. 1.) In this case, the function f can be written as the product of two bounded
functions on I, then f is bounded on I.

2.) In this case, the function f can be written as the product of a bounded function and
a function tending to 0 at a, then f tends to 0 at a.

3.) Same arguments as the previous point.

Proposition A.1.4. We assume that the function g does not vanish on I \ {a}. Then,
f
1.) f is dominated by g if, and only if is bounded on I \ {a},
g
f (x)
2.) f is negligible in front of g if, and only if lim = 0.
x→a g(x)
Proof. 1.) We prove this point by double implications.

⇒: We assume that f is dominated by g, then there exists a bounded function u

f
such that f = ug on I. Then = u on I and hence it is bounded on I \ {a}.
g
f f (x)
⇐: We assume that is bounded on I \ {a}. We set u(x) = for all x ∈
g g(x)
I \ {a}. We need then to determine the value of u at a. If g(a) ̸= 0, we then set
f (a) f
u(a) = . If g(a) = 0, it follows from the boundedness of the function , we
g(a) g
deduce that f (a) = 0. Hence, we can set u(a) as any fixed number, for instance,
u(a) = 1. With those settings, we have u is bounded on I and thus f = O(g).
a

2.) We prove this point by double implications.

⇒: If f = g, there exists a function ε tending to 0 at a such that f = εg on I. So,

a
f (x) f (x)
= ε(x) for all x ∈ I \ {a} and therefore, lim = 0.
g(x) x→α g(x)

f (x) f (x)
⇐: We assume that lim = 0. We then set ε(x) = for all x ∈ I \ {a}.
x→a g(x) g(x)
Since the function g is supposed continuous on I, thus it is bounded. And the
f f
function f can be written as the product of and g where g is bounded and
g g
tends to 0 at a. This shows that lim f (x) = 0. We can thus set ε(a) = 0 and
x→a
the claim f = o(g) holds.
a

Example A.1.5. The following relations hold.

• ln x = o(x).
x→+∞
 
1
• exp − = o(x2 ).
x x→0+
408 COMPARISON RELATIONS OF FUNCTIONS AND SEQUENCES

Generally speaking, let α > 0 and β > 0, we have, when x → +∞,

1.) (ln x)α = o(xβ );

2.) xα = o(eβx );

3.) e−αx = o(x−β ).

A.2 Equivalence between Functions

Definition A.2.1. Let f and g defined on I, we say that f is equivalent to g or f and g
are equivalent on a neighborhood of a if there exists a function u defined on I such that
f = ug and lim u(x) = 1. We denote this relation by f ∼ g or by f (x) ∼ g(x).
x→a a x→a

Remark A.2.2. We can obtain immediately that f ∼ g if, and only if there exists a function
a
ε defined on I and tend to 0 at a such that f = (1 + ε)g. In other words, f − g = o(g).
a

Proposition A.2.3. If f is equivalent to g in a neighborhood of a, then g is equivalent to

f in a neighborhood of a.

Proof. Since f is equivalent to g in a neighborhood I of a, there exists a function u which

tends to 1 at a. Then the function u does not vanish in a neighborhood I ′ of a, which we
1
can suppose that I ′ ⊂ I. So, the function x 7→ is defined on I ′ and tends to 1 at a,
u(x)
1
then the relation g = f shows that g ∼ f .
u a

Example A.2.4. Let f be a polynomial function defined by

n
X
f (x) = ck xk ,
k=p

where cp ̸= 0 and cn ̸= 0.

1.) a = 0. For x ∈ R, we have

 
n
X ck k−p 
f (x) = cp xp 1 + x .
cp
k=p+1

Since lim xr = 0 once r > 0, we have then f (x) ∼ cp xp

x→0 x→0

2.) a = ±∞. For x ̸= 0, we have

 
n−1
X ck
f (x) = cn xn  + 1 ∼ cn xn .
cn xn−k x→±∞
k=p
A.2. EQUIVALENCE BETWEEN FUNCTIONS 409

Proposition A.2.5. If g does not vanish on I \ {a}, then the function f is equivalent to g
f (x)
if, and only if lim = 1.
x→a g(x)

Proof. From Remark A.2.2, f is equivalent to g if, and only if f − g = o(g). Then using
a
Proposition A.1.4, this relation holds if, and only if

f (x) − g(x) f (x)

lim = 0 ⇔ lim = 1.
x→a g(x) x→a g(x)

Example A.2.6. 1.) Let f and g are two functions defined on I\{a}
√ with strictly positive
√
values. If f is equivalent to g in a neighborhood of a, then f and g are equivalent
in a neighborhood of a.

2.) When x → 1, we have

x3 − 1 = (x − 1)(x2 + x + 1) ∼ 3(x − 1).

In the consequences,
√
• 3 x3 − 1 ∼ 3 3(x − 1).
p

• (x3 − 1)α ∼ 3α (x − 1)α for α ∈ R.

3.) When x → +∞, we have

p
x3 − 2x2 + x + 3 ∼ x3/2 .

Proposition A.2.7. Let f and g be equivalent functions in a neighborhood of a. If g has

a limit at a, then f has also a limit at a. Moreover,

lim f (x) = lim g(x).

x→a x→a

Proof. Immediate consequence of f = ug with lim u(x) = 1.

x→a

Remark A.2.8. • The statement of the previous proposition still holds if g diverges to
±∞ when x tends to a. In this cases, the functions f and g diverges simultaneously
to ±∞ with the same sign while x → a.

• The reciprocal statement remains true only if the functions f and g have a common
non-zero finite limit while x → a.

Proposition A.2.9. Let f and g be equivalent functions in a neighborhood of a. Then the

following assertions hold.

1.) If g is positive on I, then f is positive on a neighborhood of a.

2.) If g does not vanish on I, then neither does f in a neighborhood of a.

410 COMPARISON RELATIONS OF FUNCTIONS AND SEQUENCES

3.) If g does not vanish on I \ {a}, then neither does f |I\{a} in a neighborhood of a.
Proof. Immediate consequence of f = ug with lim u(x) = 1.
x→a

Proposition A.2.10. Let f be a differentiable function at a ∈ R and if f ′ (a) ̸= 0, then

while x → a,

f (x) − f (a) ∼ f ′ (a)(x − a).

Proof. Immediate consequence of the definition of f ′ and the fact f ′ (a) ̸= 0.

Example A.2.11. When x → 0,

• ex − 1 ∼ x • tan x ∼ x

• ln(1 + x) ∼ x • arctan x ∼ x

• sin x ∼ x • sinh x ∼ x

• arcsin x ∼ x • tanh x ∼ x

Proposition A.2.12. Let f and g be equivalent functions in a neighborhood of a. Let φ be

a function define on J ⊂ R and mapping into I such that lim φ(t) = a. Then, f ◦ φ and
t→α
g ◦ φ are equivalent in a neighborhood of α.
Proof. Let u be a function defined on I such that u(x) −→ 1 and f = ug on I. Then, in a
x→a
neighborhood of α, we have

f (φ(t)) = u(φ(t))g(φ(t)),

with lim u(φ(t)) = 1. Thus the claim follows.

t→α

Example A.2.13. 1.) When x → 0, we have esin x − 1 ∼ sin x.

2.) When x → 0, we have (1 + x)α − 1 = eα ln(1+x) − 1 ∼ α ln(1 + x) where α ∈ R∗ .

3.) When x → 0, we have ln(cos x) ∼ cos x − 1.

Proposition A.2.14. In a neighborhood of a, the following assertions hold.

1.) If f ∼ g and g ∼ h, then f ∼ h.

2.) If f1 ∼ g1 and f2 ∼ g2 , then f1 f2 ∼ g1 g2 .

f1 g1
3.) If f1 ∼ g1 and f2 ∼ g2 and none of those functions vanish on I \ {a}, then ∼ .
f2 g2
Proof. 1.) We can find functions u and v defined on a neighborhood of a such that f = ug,
g = vh and they both tend to 1 at a. Therefore, f = uvh in a neighborhood of a and
lim u(x)v(x) = 1. Thus, the transitivity holds.
x→a
A.2. EQUIVALENCE BETWEEN FUNCTIONS 411

2.) We can find functions u1 and u2 defined on a neighborhood of a such that f1 = u1 g1 ,

f2 = u2 g2 and they both tend to 1 at a. Therefore, f1 f2 = u1 u2 g1 g2 in a neighborhood
of a and lim u1 (x)u2 (x) = 1. Thus, claim on the product function holds.
x→a

3.) We can find functions u1 and u2 defined on a neighborhood of a such that f1 = u1 g1 ,

f1 u1 g1
f2 = u2 g2 and they both tend to 1 at a. Therefore, = in a neighborhood of
f2 u2 g2
u1 (x)
a and lim = 1. Thus, claim on the quotient function holds.
x→a u2 (x)

Example A.2.15. 1.) The transitivity of the equivalence relation implies that, in a
neighborhood of 0,

esin x − 1 ∼ sin x ∼ x,

as well as for α ∈ R∗ ,

(1 + x)α − 1 ∼ α ln(1 + x) ∼ αx.

2.) Let f be a rational function (i.e. can be written as a quotient of polynomial functions)
which is defined by,
n
X
ak xk
k=p
f (x) = m with ap , an , bq , bm ̸= 0.
X
bk xk
k=q

ap xp ap
• In a neighborhood of 0, we have f (x) ∼ ∼ xp−q .
bq xq bq
an x n an n−m
• In a neighborhood of ±∞, we have f (x) ∼ m
∼ x .
bm x bm
x x  x
3.) In a neighborhood of 0, we have sin ∼ and cos x − 1 = cos 2 −1 =
x 2 2 2
−2 sin2 . They implies that,
2
x2
cos x − 1 ∼ − ,
2
also that,

x2
ln(cos x) = ln(1 + (cos x − 1)) ∼ cos x − 1 ∼ − .
2

x2
4.) In a neighborhood of 0, cosh x − 1 ∼ .
2
√
5.) In a neighborhood of 0, arccos(1 − x) ∼ 2x.
412 COMPARISON RELATIONS OF FUNCTIONS AND SEQUENCES

6.) Let f : R∗+ → R be a function defined by

   
1 1
∀x > 0, f (x) = exp − exp .
x2 (x + 1)2
Then, for x → +∞ we have,
   
1 1 1
f (x) = exp 1 − exp −
x2 (x + 1)2 x2
   
1 2x + 1
= exp 1 − exp − 2
x2 x (x + 1)2
   
2x + 1 1
∼ 1 − exp − 2 2
since lim exp =1
x (x + 1) x→+∞ x2
2x + 1 2x + 1
∼ 2 since lim 2 =0
x (x + 1)2 x→+∞ x (x + 1)2
2
∼ 3.
x

Remark A.2.16 (important). There are not general results on the sum or the difference
of two equivalent functions. In general, if f1 ∼ g1 and f2 ∼ g2 , we can’t conclude that
f1 + f2 ∼ g1 + g2 .

Example A.2.17. 1.) We consider the function x 7→ sin 2x + tan 3x when x → 0. We

have already that sin 2x ∼ 2x and tan 3x ∼ 3x. We can write it as follows
 
sin 2x tan 3x
sin 2x + tan 3x = x + .
x x
 
sin 2x tan 3x
Since lim + = 5, we can conclude that
x→0 x x

sin 2x + tan 3x ∼ 5x.

2.) We consider now the function x 7→ sin 2x − tan 2x when x → 0. We can write it as
follows
 
sin 2x tan 2x
sin 2x − tan 2x = x − .
x x
 
sin 2x tan 2x
Since lim − = 0, we can not conclude in this case about the sum of
x→0 x x
the equivalence relations. We can only obtain that

sin 2x − tan 2x = o(x).

3.) We consider the function x 7→ sin 2x + cos x − 1 when x → 0. We already have

x2 x2
sin 2x ∼ 2x and cos x − 1 ∼ − . Here we remark that = o(x) in a neighborhood
2 2
of 0. We can write it as follows
 
sin 2x x cos x − 1
sin 2x + cos x − 1 = x + .
x 2 x2 /2
A.3. COMPARISON OF SEQUENCES 413
 
sin 2x x cos x − 1
Since lim + = 2, we can obtain that
x→0 x 2 x2 /2

sin 2x + cos x − 1 ∼ 2x.

4.) We consider that function x 7→ x + ln x when x → +∞. We already know that

ln x
ln x = o(x) when x → +∞, this relation can be also formulate as lim = 0.
x→+∞ x
Then we write
 
ln x
x + ln x = x 1 + .
x

Applying the previous limit, we have then x + ln x ∼ x.

Those previously presented examples can be summarized into the following proposition.

Proposition A.2.18. If f and g be functions such that g = o(f ) in a neighborhood of a,

then it follows f + g ∼ f .

A.3 Comparison of Sequences

In this section, we introduce some useful notions to describe the asymptotic behaviors
between real sequences. Let (xn )n∈N and (yn )n∈N be real sequences.

Definition A.3.1. • We say (xn )n∈N is dominated by (yn )n∈N if there exits a bounded
sequence (wn )n∈N such that xn = wn yn from certain rank. We denote this relation by
xn = O(yn ).

• We say (xn )n∈N is negligible in front of (yn )n∈N if there exits a sequence (wn )n∈N
converging to 0 such that xn = wn yn from certain rank. We denote this relation by
xn = o(yn ).

• We say the sequence (xn )n∈N is equivalent to (yn )n∈N if there exits a sequence
(wn )n∈N converging to 1 such that xn = wn yn from certain rank. We denote this
relation by xn ∼ yn .
We have immediately that if xn ∼ yn then yn ∼ xn . We then say those two sequences
are equivalent.

Remark A.3.2. • The relation of dominance can be also reformulated as follows, there
exists M > 0 such that |xn | ≤ M |yn | from certain rank.

• The relation of negligence can be reformulated as follows, for all ε > 0, there exists
N ∈ N such that for all n ≥ N , |xn | ≤ ε|yn |.

Proposition A.3.3. If yn ̸= 0 from certain rank, then

xn
• xn = O(yn ) if, and only if is bounded from certain rank.
yn
xn
• xn = o(yn ) if, and only if lim = 0.
n→+∞ yn
414 COMPARISON RELATIONS OF FUNCTIONS AND SEQUENCES

xn
• xn ∼ yn if, and only if lim = 1.
n→+∞ yn

Proof. Straightforward from the definitions, and left in Exercises.

We can use the results in the comparison of functions to obtain the corresponding
comparison of sequences. Let (un )n∈N be a sequence of numbers in I and un −→ a.
n→+∞

1.) If f = O(g), then f (un ) = O(g(un )).

a

2.) If f = o(g), then f (un ) = o(g(un )).

a

3.) If f ∼ g, then f (un ) ∼ (g(un )).

a

Proposition A.3.4. The following useful comparison relations of negligence hold.

1.) n = o(n2 ); 6.) nα = o(nβ ) if α < β;

√
2.) n = o(n); 7.) nα = o(an ) for all α > 0, a > 1;
 
1 1
3.) =o √ ; 8.) an = o(nα ) for all α > 0, 0 < a < 1;
n n
1 9.) (ln n)β = o(nα ) for all α > 0, β > 0;
4.) = o(1);
n
5.) 1 = o(n); 10.) an = o(n!) for all a > 1.

Proof. Left in Exercises.

Proposition A.3.5. Let (xn )n∈N , (x′ n )n∈N , (yn )n∈N , (y ′ n )n∈N and (wn )n∈N be sequences
of real numbers.

• If xn ∼ yn and yn ∼ wn , then xn ∼ wn .

• If xn ∼ yn and x′n ∼ yn′ , then xn x′n ∼ yn yn′ .

• If xn ∼ yn and x′n ∼ yn′ and those sequences don’t vanish from certain rank, then
xn yn
′
∼ ′.
xn yn
Proof. Left in Exercises

Example A.3.6. Let (un )n∈N be a sequence of real numbers which converges to 0. The
following comparison relations of equivalence hold.

1.) eun − 1 ∼ un ; 5.) sinh un ∼ un ;

2.) ln(1 + un ) ∼ un ; 6.) tanh un ∼ un ;
3.) sin un ∼ un ; 7.) arcsin un ∼ un ;
4.) tan un ∼ un ; 8.) arctan un ∼ un ;
A.4. EXERCISES 415

u2n u2n
9.) cos un − 1 ∼ − ; 10.) cosh un − 1 ∼ ;
2 2
11.) (1 + un )p − 1 ∼ pun with p ∈ R.

Theorem A.3.7 (Stirlingii ’s Formula).

√  n n
n! ∼ 2πn .
e
Proof. See in Exercises A.4.1 to A.4.8.

A.4 Exercises
Stirling’s Formula
Exercise A.4.1. For all n ∈ N∗ , calculate the integral,
Z n
In := ln(t)dt.
1

Exercise A.4.2. Let k ∈ N∗

1.) Show that ln is a concave function on R+ .

2.) Show that for all t ∈ [0, 1], ln(k + t) ≥ (1 − t) ln(k) + t ln(k + 1).
t−k
3.) Show that for all t ∈ [k, k + 1], ln(t) ≤ ln(k) + .
k
k+1−t
4.) Show that for all t ∈ [k, k + 1], ln(t) ≤ ln(k + 1) − .
k+1
5.) Show that
Z k+1  
1 1 1 1 1
(ln(k) + ln(k + 1)) ≤ ln(t)dt ≤ (ln(k) + ln(k + 1)) + − .
2 k 2 4 k k+1

6.) Show that for all n ∈ N∗ ,

 
1 1 1 1
ln(n!) − ln(n) ≤ In ≤ ln(n!) − ln(n) + 1− .
2 2 4 n

Exercise A.4.3. We consider the sequence (un )n∈N defined by, for all n ∈ N∗ ,
1
un = ln(n!) − ln(n).
2
Show that the sequence (In − un )n∈N∗ is increasing and convergent.

ii
James Stirling, Scottish mathematician 1692-1770.
416 COMPARISON RELATIONS OF FUNCTIONS AND SEQUENCES

Exercise A.4.4. We set L := lim (In − un ). Show that while n → +∞,

n→+∞

nn √ 1−L
n! ∼ ne .
en

Exercise A.4.5. Let n ∈ N, we defineiii ,

Z π/2
Wn := sinn (t)dt.
0

Show that, for all n ∈ N,

(n + 2)Wn+2 = (n + 1)Wn .

Exercise A.4.6. Show that for all n ∈ N,

(2n)! π
W2n = .
22n (n!)2
2

n+1
Exercise A.4.7. 1.) Show that for all n ∈ N, Wn ≤ Wn+1 ≤ Wn .
n+2
2.) Show that while n → +∞, Wn+1 ∼ Wn .

3.) Show that (n + 1)Wn+1 Wn = C with C is a constant independent of n. And give the
value of C.
r
π
4.) Show that while n → +∞, Wn ∼ .
2n

Exercise A.4.8. Prove Stirling’s formula, while n → +∞,

√  n n
n! ∼ 2πn .
e

iii
Those integrals are called the Wallis integrals.
 Appendix B

Asymptotic Expansions

Here, we will introduce an elementary theory about the asymptotic analysis. It can be
considered as a complementary of Taylor Expansions introduced in Calculus. The general
theory was developed independently by Poincaréi and Stieljesii .
In this chapter, we preserve the notations in the previous one. Let n be a natural
number and a ∈ R. The functions considered in this chapter are real-valued and defined in
a neighborhood I of a.

B.1 Definitions and Examples

We begin with the general definition.

Definition B.1.1. We say the function f possesses an asymptotic expansion of order n at

a (a could take ±∞) if there exists n + 1 functions φ0 , φ1 , · · · , φn such that

• for all k ∈ [[1, n]], φk (x) = o(φk−1 (x)) while x → a;

n
X
• f (x) − φk (x) = o(φn (x)) while x → a.
k=0

From the next subsection, we will no longer consider the general functions φ0 · · · φn .
Instead, we shall consider only the cases where φk can be taken as the polynomial basis xk
or fractionals x−k .

B.1.1 Expansion Near a Real Number

In this subsection, we suppose always that a ∈ R.

Definition B.1.2. We say the function f possesses an asymptotic expansion of order

n in a neighborhood of a if there exists the real numbers α0 , α1 · · · , αn and a function
ε defined on I \ {a} such that
i
Henri Poincaré (1854 - 1912), French mathematician, physicist, engineer and philosopher. Several fun-
damental theories in modern science, such chaos theory and relativity, can be traced up to his original works.
He was considered as the genius after Issac Newton.
ii
Thomas Joannes Stieljes (1856 - 1894), Dutch mathematician. He was famous by his pioneer works in
the theoretical analysis, especially the Riemann-Stieljes integral.

417
418 ASYMPTOTIC EXPANSIONS

n
X
• for all x ∈ I, f (x) = αk (x − a)k + (x − a)n ε(x);
k=0

• lim ε(x) = 0.
x→a

Remark B.1.3. In the above definition, we suppose always that f is defined in a neighbor-
hood of a. This definition can be also formulated using the small o symbol. Namely,
n
X
f (x) − αk (x − a)k = o(|x − a|n ).
x→a
k=0

Also,
n
X
f (x) = αk (x − a)k + o(|x − a|n ).
x→a
k=0

Example B.1.4. 1.) Consider the function f on (−1, 1) given by,

f (x) = x − x2 + 2x3 + x3 ln(1 + x).

Then, f possesses an asymptotic expansion of order 3 at 0 as follows

f (x) = x − x2 + 2x3 + x3 ε(x).

x→0

Here, ε(x) = ln(1 + x) −→ 0.

x→0

R \ {1} −→ R
2.) Consider the function f : 1
. Then f possesses an asymptotic
x 7−→ 1−x
expansion of any order n ∈ N. In fact, we have
n
1 − xn+1
 
X
k 1 n x
∀x ∈ R \ {1}, x = = −x .
1−x 1−x 1−x
k=0

This implies that

n
1 X
∀x ∈ R \ {1}, = xk + xn ε(x),
1−x
k=0

x
with ε(x) = −→ 0.
1 − x x→0
3.) It is sufficient to replace −x by x2 in the previous example to obtain that
n
1 X
∀x ∈ R, 2
= (−1)k x2k + (−1)n x2n ε(−x2 ).
1+x
k=0

1
And this implies that the function x 7→ possesses the asymptotic expansions
1 + x2
of order 2n at 0.
B.1. DEFINITIONS AND EXAMPLES 419

Theorem B.1.5 (Taylor-Young or Taylor with Peano Remainder). Let f be a C n -class

function on the neighborhood I of a ∈ R. Then, it holds that
n
X f (k) (a)
∀x ∈ I, f (x) = (x − a)k + ε(x)(x − a)n . (B.1)
k!
k=0

Here, ε is a function defined on I such that lim ε(x) = 0.

x→a
In terms of asymptotic expansions, it is equivalent to say while x → a,
n
X f (k) (0)
f (x) = (x − a)k + o(|x − a|n ).
k!
k=0

Proof. If n = 0, it is sufficient to take ε(x) = f (x) − f (a) and (B.1) follows from the
continuity of f . We suppose from now on that n ∈ N∗ .
We define the function ε as follows. ε(a) = 0 and
n
X f (k) (a)
f (x) − (x − a)k
k!
k=0
∀x ∈ I \ {a}, ε(x) = .
(x − a)n
And we can immediately deduce the formula (B.1). It remains to prove lim ε(x) = 0.
x→a
f (n) (a)
Consider now the function g given by, ∀x ∈ I, g(x) = f (x) − (x − a)n . Then g
n!
is of class C n on I and for all k ∈ [[0, n − 1]], g (k) (a) = f (k) (a). Also, g (n) (a) = 0. We apply
now Taylor Expansion formula with integral remainder to the function g and it holds that
for all x ∈ I,
n−1
X f (k) (a) Z x
(x − t)n−1 (n)
g(x) = (x − a)k + g (t)dt
k! a (n − 1)!
k=0
n−1
X f (k) (a) Z 1
k n (1 − u)n−1 (n)
= (x − a) + (x − a) g (a + (x − a)u)du.
k! 0 (n − 1)!
k=0
On the other hand, using the definition of ε, we have that
n−1
X f (k) (a)
∀x ∈ I \ {a}, g(x) = (x − a)n ε(x) + (x − a)k .
k!
k=0
It becomes then,
1
(1 − u)n−1 (n)
Z
∀x ∈ I \ {a}, ε(x) = g (a + (x − a)u)du.
0 (n − 1)!
We remark that g (n) is a continuous function on I and g (n) (a) = 0. Let ε′ > 0, it follows
from the continuity of this function at a, there exists δ > 0 such that for all x ∈ I,
|x − a| ≤ δ ⇒ |g (n) (x)| ≤ ε′ . Hence,
Z 1
(1 − u)n−1 (n)
∀x ∈ I, |x − a| ≤ δ, |ε(x)| = g (a + (x − a)u)du
0 (n − 1)!
Z 1
(1 − u)n−1 (n)
≤ |g (a + (x − a)u)|du
0 (n − 1)!
ε′ ε′
Z 1
≤ (1 − u)n−1 du = 2 ≤ ε′ .
(n − 1)! 0 n
420 ASYMPTOTIC EXPANSIONS

This implies that lim ε(x) = 0. Thus, the claim follows.

x→0

Proposition B.1.6 (Uniqueness of Asymptotic Expansions). We suppose that the func-

tion f possesses two asymptotic expansions at a ∈ R. Namely, there exists two n-tuples
(α0 , α1 , · · · , αn ) and (β0 , β1 , · · · , βn ) such that when x → a,
n
X n
X
k n
f (x) = αk (x − a) + o(|x − a| ), and f (x) = βk (x − a)k + o(|x − a|n ).
k=0 k=0

Then, for all j ∈ [[0, n]], αj = βj .

Proof. From the assumption, there exists two functions ε1 and ε2 which tend to 0 when
x → a such that
n
X n
X
∀x ∈ I, f (x) = αk (x − a)k + (x − a)n ε1 (x) = βk (x − a)k + (x − a)n ε2 (x).
k=0 k=0

We assume now by the contradictory that (α0 , α1 , · · · , αn ) ̸= (β0 , β1 , · · · , βn ). We denote

by p the smallest index such that αp ̸= βp . We have thus,
n
X
∀x ∈ I, αp (x − a)p + αk (x − a)k + (x − a)n ε1 (x)
k=p+1
Xn
= βp (x − a)p + βk (x − a)k + (x − a)n ε2 (x).
k=p+1

By dividing (x − a)p in the above equality, it holds that,

n
X
∀x ∈ I \ {a}, αp + αk (x − a)k−p + (x − a)n−p ε1 (x)
k=p+1
Xn
= βp + βk (x − a)k−p + (x − a)n−p ε2 (x).
k=p+1

Next, we pass x → a to obtain that αp = βp . This leads to the contradiction. Hence,

(α0 , α1 , · · · , αn ) = (β0 , β1 , · · · , βn ).

Remark B.1.7. In fact, the study of the asymptotic expansion of a function f at a can be
obtained those of the function h 7→ f (a + h) at 0.

Remark B.1.8. The following results can be deduced from the definitions. The proofs are
left in Exercises.

• The function f possesses an asymptotic expansion of order 0 at a if, and only if f has
a limit when x → a.

• The function f possesses an asymptotic expansion of order 1 at a if, and only if f is

differentiable at a.
B.1. DEFINITIONS AND EXAMPLES 421

However, those properties will no longer hold when n ≥ 2. Here is an example of a function
possessing an asymptotic expansion
 but not 2-times differentiable.
1
Consider f : x 7→ x + x3 sin on R∗ and f (0) = 0. We can thus write ∀x ∈ R,
x2   
2 1 1
f (x) = x + x ε(x) with ε(x) = x sin 2
and ε(0) = 0. Since x 7→ sin is bounded
x x2
on R∗ , we have thus lim ε(x) = 0. This shows that f possesses an asymptotic expansion of
x→0
order 2 at 0.
On the other hand, we calculate the derivative of f .
   
∗ ′ 2 1 1
∀x ∈ R , f (x) = 1 + 3x sin − 2 cos .
x2 x2
This shows that f ′ has no limit when x → 0. In the consequence, f is not 2-times differen-
tiable at 0.

B.1.2 Expansions of Usual Functions

By applying Taylor-Young Theorem B.1.5, the following functions possess the asymptotic
expansions at 0 of any order n.
1
• Inverse function x 7→ .
1−x
n
1 X
= xk + o(xn ).
1−x
k=0

• Exponential function and hyperbolic functions.

n
X xk
ex = + o(xn ),
k!
k=0
n
X x2k+1
sinh(x) = + o(x2n+2 ),
(2k + 1)!
k=0
n
X x2n
cosh(x) = + o(x2n+1 .)
(2n)!
k=0

• Trigonometric functions.
n
X x2n+1
sin(x) = (−1)k + o(x2n+2 ),
(2n + 1)!
k=0
n
X x2n
cos(x) = (−1)k + o(x2n+1 ).
(2n)!
k=0

• Logarithmic function.
n
X xk
ln(1 + x) = (−1)k+1 + o(xn ).
k
k=1

• Power function x 7→ (1 + x)α with α ∈ R.

α(α − 1) 2 α(α − 1) · · · (α − n + 1) n
(1 + x)α = 1 + αx + x + ··· + x + o(xn ).
2! n!
422 ASYMPTOTIC EXPANSIONS

B.1.3 Expansions at Infinity

Definition B.1.9. We say the function f possesses an asymptotic expansion of order n at
±∞ if there exists a n + 1-tuple (α0 , α1 , · · · , αn ) such that when x → ±∞,
n  
X αk 1
f (x) = +o .
xk xn
k=0

x
Example B.1.10. Consider the function f : x 7→ defined on R \ {1}. We search its
x−1
1
asymptotic expansion at +∞. Applying a change of variables u = , we have,
x
1
∀x ∈ (1, +∞), f (x) = .
1−u
1
We know that the function u 7→ possesses an asymptotic expansion at 0 of any order
1−u
n, which is given by,
n
1 X
= uk + o(un ).
1 − u u→0
k=0

Hence, it holds that

n  
X 1 1
f (x) = + o .
x→+∞ xk xn
k=0

B.2 Operations on Asymptotic Expansions

It is also essential to know how to manipulate the asymptotic expansions. As we mentioned
in Remark B.1.7, the results of asymptotic expansions at a real number a can be obtained
from the study at 0. So, the results in this chapter will only concern on the asymptotic
expansions at 0 or at infinity.

B.2.1 Sums and Products

Proposition B.2.1. Let f and g be functions possessing the asymptotic expansions at 0 of
order n, which can be expressed as follows.

f (x) = P (x) + o(xn ), and g(x) = Q(x) + o(xn ).

Here, P and Q refer to the polynomial functions of degree n.

Then, the functions f + g and f g possess the asymptotic expansions at 0 of order n given
by

f (x) + g(x) = P (x) + Q(x) + o(xn ),

f (x)g(x) = R(x) + o(xn ),

where R is the polynomial function obtained from the product P Q by conserving all the
terms of degrees less then n.
B.2. OPERATIONS ON ASYMPTOTIC EXPANSIONS 423

Proof. Since f and g possess the asymptotic expansions at 0, there exists the functions ε1
and ε2 which tend to 0 when x → 0 and satisfy,

∀x ∈ I, f (x) = P (x) + xn ε1 (x),

g(x) = Q(x) + xn ε2 (x).

By setting ε(x) = ε1 (x) + ε2 (x), we have

∀x ∈ I, f (x) + g(x) = P (x) + Q(x) + xn ε(x).

Also, it holds that lim ε(x) = lim ε1 (x) + lim ε2 (x) = 0. Thus, the statement for the sum
x→0 x→0 x→0
function f + g holds.
On the other hand, by calculating the product, we have,

∀x ∈ I, f (x)g(x) = P (x)Q(x) + P (x)xn ε2 (x) + Q(x)xn ε1 (x).

Introducing the polynomial function R in the statement, so we can write P (x)Q(x) =

R(x) + xn+1 S(x) with S being a polynomial function. Thus, it becomes,

∀x ∈ I, f (x)g(x) = R(x) + xn+1 S(x) + P (x)xn ε2 (x) + Q(x)xn ε1 (x)

= R(x) + xn (xS(x) + P (x)ε2 (x) + Q(x)ε1 (x)).

We set now ε(x) = xS(x) + P (x)ε2 (x) + Q(x)ε1 (x) and it holds that lim ε(x) = 0. Thus,
x→0
the claim follows.

Corollary B.2.2. Let f and g be functions possessing the asymptotic expansions of order
n at a ∈ R. Then their sum f + g and product f g posses also the asymptotic expansions of
order n at a.

Proof. It is sufficient to consider the functions h 7→ f (a+h) and h 7→ g(a+h) then apply the
previous proposition to deduce the results in the cases where a ∈ R. For the cases a = ±∞,
1 1
we consider the functions u 7→ f ( ) and u 7→ g( ) then apply the previous proposition.
u u

Example B.2.3. 1.) The asymptotic expansion of order 3 at 0 to the function f defined
1
on R \ {1} by f (x) = − ex .
1−x
1
We have seen the asymptotic expansions of the functions x 7→ and exp of order
1−x
3 at 0. They are given as follows.
1
= 1 + x + x2 + x3 + o(x3 )
1−x
x2 x3
ex = 1 + x + + + o(x3 ).
2 6
Since f is the difference of those functions, the asymptotic expansion of f is thus
obtained by the difference of those to the above functions. So,

x2 5x3
f (x) = + + o(x3 ).
2 6
424 ASYMPTOTIC EXPANSIONS

2.) The asymptotic expansion of order 3 at 0 to the function g defined on (−1, +∞) by
cos x
g(x) = √ .
1+x
1
The functions cos and x 7→ √ = (1 + x)−1/2 posses the asymptotic expansions
1+x
of order 3 at 0 given as follows
1
cos x = 1 − x2 + o(x3 ),
2
1 1 3 5
√ = 1 − x + x2 − x3 + o(x3 ).
1+x 2 8 16
Hence, by multiplying the above expansions, we obtain that
1 1 1
g(x) = 1 − x − x2 − x3 + o(x3 ).
2 8 16

B.2.2 Quotients
Proposition B.2.4. Let u be a function such that lim u(x) = 0. We assume furthermore
x→0
1
that u possesses the asymptotic expansions of order n at 0. Then, the function x 7→
1 − u(x)
possesses the asymptotic expansion of order n at 0.
1
Proof. The condition lim u(x) = 0 implies that the function x 7→ is defined in a
x→0 1 − u(x)
neighborhood of 0.

• Case n = 0. We have seen that a function possesses an asymptotic expansion of order

0 is equivalent to say this function converges at 0. It is straightforward by applying the
1
limit of function composition to deduce that lim = 1. Hence, the function
x→0 1 − u(x)
1
x 7→ possesses an asymptotic expansion of order 0 at 0.
1 − u(x)
1
• We assume now n ≥ 1. We have seen that the asymptotic expansion of h 7→ at
1−h
0 of order n is given by,
n
1 X
∀h ∈ R \ {1}, = hk + hn ε1 (h),
1−h
k=0

with lim ε1 (h) = 0. Now we replace h by u(x) in the above relation and it becomes
h→0

n
1 X
= u(x)k + u(x)n ε1 (u(x)),
1 − u(x)
k=0

for x belongs to a neighborhood of 0.

Using the asymptotic expansion of u and the fact lim u(x) = 0, we can write in a
x→0
neighborhood of 0 that

u(x) = α1 x + xε2 (x),

B.2. OPERATIONS ON ASYMPTOTIC EXPANSIONS 425

where α1 ∈ R and lim ε2 (x) = 0. In the consequence, it holds that

x→0

n
1 X
= u(x)k + xn (α1 + ε2 (x))n ε1 (u(x)).
1 − u(x)
k=0

From lim u(x) = 0 and lim ε1 (h) = 0, we deduce that lim ε1 (u(x)) = 0. Also,
x→0 h→0 x→0
lim (α1 + ε2 (x))n = α1n . Then, we have that
x→0

xn (α1 + ε2 (x))n ε1 (u(x)) = o(xn ).

x→0

Hence, it holds that

n
1 X
= u(x)k + o(xn )
1 − u(x)
k=0

n
X
We remark that the function x 7→ u(x)k possesses the asymptotic expansion at 0
k=0
of order n since it is composed from the products and sums of the functions possessing
the asymptotic expansions. Thus, the claim follows.

Example B.2.5. Determinate the following asymptotic expansions at 0.

1
1.) The function f : x 7→ of order 2.
1 + x + 12 x2
1 1
We set u(x) = −x − x2 . It is immediate that lim u(x) = 0. Then f (x) =
2 x→0 1 − u(x)
in a neighborhood of 0. We then apply the proof of the previous proposition to obtain
that
1
f (x) = = 1 + u(x) + u(x)2 + o(u(x)2 ).
1 − u(x)
So,

1 2 2
   
1 1 2
1 2 = 1 + −x − x + −x − x + o(x2 )
1 + x + 2x 2 2
1
= 1 − x + x2 + o(x2 ).
2

1
2.) The function g : x 7→ of order 4.
cos x
 π π 1 1
For all x ∈ − , , we have that g(x) = = . From the
2 2 1 − (1 − cos(x)) 1 − u(x)
asymptotic expansion of cos at 0, we have also
1 1
u(x) = 1 − cos x = x2 − x4 + o(x4 ) −→ 0.
2 24 x→0
426 ASYMPTOTIC EXPANSIONS

Applying the proof of the previous proposition, we can obtain that in a neighborhood
of 0,

1 4 2
   
2 2 1 2 1 4 1 2
g(x) = 1 + u(x) + u(x) + o(u(x) ) = 1 + x − x + x − x + o(x4 ).
2 24 2 24
1 1 4
Here, since u(x) ∼ x2 , we have that u(x)2 ∼ x when x → 0. This is why we only
2 4
need to calculate up to u(x)2 not to u(x)4 . In conclusion, we have that
1 1 5
= 1 + x2 + x4 + o(x4 ).
cos x 2 24
3.) The function tan of order 5.
We apply the previous result and the asymptotic expansion of sin at 0 to obtain that
when x → 0,
  
sin x 1 3 1 5 5 1 2 5 4 4
tan(x) = = x− x + x + o(x ) 1 + x + x + o(x )
cos x 6 120 2 24
1 2
= x + x3 + x5 + o(x5 ).
3 15

Proposition B.2.6. Let f and g be functions defined on I possessing the asymptotic ex-
f
pansions at a of order n. If g has a non-zero limit at a, then the function possesses the
g
asymptotic expansion of order n at a.
Proof. If a ∈ R, we will consider the functions h 7→ f (a + h) and h 7→ g(a + h). If a = ±∞,
1 1
we will consider the functions u 7→ f ( ) and u 7→ g( ). So, we can always consider only
u u
the asymptotic expansions at 0. In the following part of this proof, we assume a = 0.
Since lim g(x) = l ̸= 0, it holds that in a neighborhood of 0,
x→0

f (x) f (x) f (x) 1 f (x) 1

= =  = .
g(x) l − (l − g(x)) l g(x) l 1 − u(x)
1− 1−
l
g(x)
We set here u(x) = 1 − which tends to 0 at 0 and possesses an asymptotic expansion
l
1
at 0. So, we apply Proposition B.2.4 to deduce the asymptotic expansion of x 7→ .
1 − u(x)
f
Hence, the function possesses the asymptotic expansion at 0 as the product of such
g
functions.

1
Example B.2.7. The asymptotic expansion of f : x 7→ of order 3 at 0.
1 + ex
Since 1 + ex −→ 2, we need to transform
x→0

1 1
∀x ∈ R, f (x) = .
1 − ex

2
1−
2
B.2. OPERATIONS ON ASYMPTOTIC EXPANSIONS 427

1 − ex
By setting u(x) = , we have u(x) −→ 0. Then we use the asymptotic expansion of
2 x→0
1
exp and u 7→ , which are given as follows.
1−u
1 − ex 1 1 1
= − x − x2 − x3 + o(3),
2 2 4 12
1
= 1 + u + u + u + o(u3 ).
2 3
1−u
Now we apply the proof in the previous proposition, we have
   2
1 1 2 1 3 3 1 1 2 1 3 3
2f (x) =1 − x + x + x + o(x ) + x + x + x + o(x )
2 4 12 2 4 12
 3
1 1 1
− x + x2 + x3 + o(x3 ) + o(x3 )
2 4 12

Developing the above formula, we obtain thus,

1 1 1
f (x) = − x + x3 + o(x3 ).
2 4 24

B.2.3 Compositions
It is also possible to determinate the asymptotic expansions of the compositions of func-
tions. The statements of general theorems are complicated, we will introduce this part by
examples.

Example B.2.8. Asymptotic expansion of order 3 at 0 to the function f : x 7→ esin(x) .

Let u(x) = sin x on R, we have u(x) −→ 0. Moreover, since sin x ∼ x, it follows that
x→0 0
u(x)3 ∼ x3 . So, applying the following asymptotic expansions of order 3
0

1 1
eu = 1 + u + u2 + u3 + o(u3 ),
u→0 2 6
1 3
sin x = x − x + o(x3 ),
x→0 6
we can obtain that in a neighborhood of 0,
   2  3
1 1 1 1 1
esin x = 1 + x − x3 + o(x3 ) + x − x3 + o(x3 ) + x − x3 + o(x3 ) + o(x3 )
6 2 6 6 6
1
= 1 + x + x2 + o(x3 ).
2

 
sin x
Example B.2.9. Asymptotic expansion of order 4 at 0 to the function g : x 7→ ln .
x
We begin with the asymptotic expansion to sin at 0.

1 1 5
sin x = x − x3 + x + o(x5 ).
6 120
428 ASYMPTOTIC EXPANSIONS

sin x
Then, we set u(x) = − 1 and it holds that
x
sin x 1 1 4
u(x) = − 1 = − x2 + x + o(x4 ).
x 6 120
1
We obtain thus lim u(x) = 0 and u(x) ∼ − x2 . It is therefore sufficient to calculate the
x→0 6
asymptotic expansion to ln(1 + u) of order 2. We then use the asymptotic expansion
1
ln(1 + u) = u − u2 + o(u2 )
2
to deduce that
    
sin x sin x
ln = ln 1 + −1 = ln(1 + u(x))
x x
1
= u(x) − u(x)2 + o(u(x)2 )
2
   2
1 2 1 4 4 1 1 2 1 4
= − x + x + o(x ) − − x + 4
x + o(x ) + o(x4 )
6 120 2 6 120
1 1 4
= − x2 − x + o(x4 ).
6 180

B.2.4 Integrations
Proposition B.2.10. We suppose that f is continuous and possesses an asymptotic ex-
n
X
pansion in a neighborhood of 0 of order n, which equals to αk xk . We denote by F an
k=0
indeterminate integral of f . Then, F possesses an asymptotic expansion at 0 or order n + 1
given by,
n
X αk k+1
F (x) = F (0) + x + o(xn+1 ).
k+1
k=0

Proof. From the assumptions, there exists a function ε defined on I such that
n
X
∀t ∈ I, f (t) = αk tk + tn ε(t).
k=0

Integrating this equality from 0 to x, we have that

Z x Z xXn n Z x
k n
X αk k+1
F (x) − F (0) = f (t)dt = αk t + t ε(t)dt = x + tn ε(t)dt.
0 0 k+1 0
k=0 k=0

On the other hand, we have also that lim ε(x) = 0. Let ε′ > 0, this convergence implies
x→0
that there exists δ > 0 such that for all x ∈ I, |x| ≤ δ ⇒ |ε(x)| ≤ ε′ . Thus, for all
x ∈ [−δ, δ] ∩ I,
Z x Z x
|x|n+1
tn ε(t)dt ≤ |tn ε(t)|dt ≤ ε′ .
0 0 n+1
B.2. OPERATIONS ON ASYMPTOTIC EXPANSIONS 429

This relation shows that

Z x
1
lim tn ε(t)dt −→ 0.
x→0 xn+1 0 x→0

Using the small o formulation

Z x
tn ε(t)dt = o(xn+1 ).
0 x→0

Hence, the claim follows.

Example B.2.11. Determinate the asymptotic expansions at 0 to the following functions.

1.) x 7→ ln(1 + x).

2.) x 7→ arctan x.

3.) x 7→ arcsin x.
1
Solution. 1.) We know the asymptotic expansion to x 7→ at 0,
1+x
n
1 X
= (−1)k xk + o(xn ).
1+x
k=0

1
Since x 7→ ln(1 + x) is an indeterminate integral of x 7→ , we then apply the
1+x
previous proposition to obtain that
n
X xk+1
ln(1 + x) = (−1)k + o(xn+1 )
k+1
k=0
x2 x3 xn+1
=x− + + · · · + (−1)n + o(xn+1 ).
2 3 n+1

1
2.) We know the asymptotic expansion to x 7→ at 0,
1 + x2
n
1 X
2
= (−1)k x2k + o(x2n ).
1+x
k=0

1
Since x 7→ arctan x is an indeterminate integral of x 7→ , we then apply the
1 + x2
previous proposition to obtain that
n
X x2k+1
arctan x = (−1)k + o(x2n+1 )
2k + 1
k=0
x3 x5 x2n+1
=x− + + · · · + (−1)n + o(x2n+1 ).
3 5 2n + 1
430 ASYMPTOTIC EXPANSIONS

1
3.) We know the asymptotic expansion to x 7→ √ at 0,
1 − x2
1+ 1 3 1 × 3 × · · · × (2n − 1) 2n
√ = 1 + x2 + x4 + · · · + x + o(x2n+1 ).
1 − x2 2 8 2 × 4 × · · · × 2n
1
Since x 7→ arcsin x is an indeterminate integral of x 7→ √ , we then apply the
1 − x2
previous proposition to obtain that

1 3 1 × 3 × · · · × (2n − 1) x2n+1
arcsin x = x + x3 + x5 + · · · + + o(x2n+2 ).
6 40 2 × 4 × · · · × 2n 2n + 1
 Appendix C

Summation over Arbitrary Families

Now, the main objective is to give a rigorous mathematical sense of the sums like
X
xj .
j∈J

We have seen in Chapter 2 the concept of series, in other words, the summation of a
set of numbers indexed by N (or by N∗ depending the contexts). It is a natural intension
to generalize this concept of summation to any set indexed by an arbitrary countable set,
because we have seen in Section 0.2 that a countable set is bijectively related to N.
On the other hand, the results in Section 2.3.2 on permutations of terms show that the
convergence of a series depends in fact how the general terms are ordered. It indicates that
the “sum” of a convergent series does not, in all generality, give the result of adding up all
of the terms in the series. This is why we need to introduce the mathematical meaning of
“adding them all up” of a set of numbers or vectors.
In this section, we denote by J to present an arbitrary set of indexes, which might be
even uncountable a priori. We consider as well as the family {xj | k ∈ J} indexed by J with
the general terms xj being numbers or vectors. In fact, we will see in Proposition C.3.4
that if a family is summable, then at most countable terms could be no zero. In practice,
J is often indexed by N, by Z or by Np .
In this entire section, we use the following notations.

• We consider the family {xj | j ∈ J} ⊂ E indexed by J and valued in E.

• The elements xj are called as general terms.

• E denotes the vector space where the general terms xj belong to.

• The family {xj | j ∈ J} is also denoted by (xj )j∈J like sequences.

• the set of all families indexed by J and valued in E is denoted by E J ;

• J ′ ⊂ J refers in general to a finite subset in J.

C.1 Summable Families of Positive Real Numbers

In this section, we consider the scenario ∀j ∈ J, xj ∈ R+ .

431
432 SUMMATION OVER ARBITRARY FAMILIES

Definition C.1.1. We say the family (xj )j∈J is summable if the following set is bounded
from above,
 
X 
xj J ′ ⊂ J, J ′ finite ⊂ R+ .
 ′ 
j∈J

In that case, we call this supremum of this set as the sum of (xj )j∈J . And we denote
 
X X 
xj := sup xj J ′ ⊂ J, J ′ finite ∈ R+ .
 ′ 
j∈J j∈J

Example C.1.2. The family 2−n n ∈ N is summable.



Solution. In this case, J = N. For all finite subset J ′ ⊂ N, we may denote NJ ′ = max(J ′ ) ∈
N. Then,
NJ ′
X X 1 − 2−NJ ′ −1
2−n ≤ 2−n = = 2 − 2−NJ ′ ≤ 2.
1 − 2−1
n∈J ′ n=0
( )
X
Hence, the set 2−n J ′ ⊂ J, J ′ finite is bounded from above, which means the family
n∈J ′
2−n n ∈ N is summable.

′
XFor calculate the sum, it is sufficient to consider J = [[0, N ]] for all N ∈ N. It gives
2 = 2 − 2 . Using the monotony and convergence of the sequence (2−N )N ∈N , we
−n −N

n∈J ′
deduce that
 
 X 
sup 2−n N ∈ N = 2.
 
n∈[[0,N ]]
X
In conclusion, 2−n = 2.
n∈N

Proposition C.1.3. Let (xj )j∈J beXa summable family of positive real numbers. Then, its
sum is a positive real number and xj = 0 if, and only if xj = 0 for all j ∈ J.
j∈J

Proof. The sum of a summable family of positive real numbers is the supremum of a bounded
subset in R+ . Then,
X it is a positive real number.
Assume that xj = 0. For all j0 ∈ J it follows that
j∈J
 
X X  X
0 ≤ x j0 = xj ≤ sup xj J ′ ⊂ J, J ′ finite = xj = 0.

j∈J ′

j∈{j0 } j∈J

It gives then xj0 = 0 holds for all j0 ∈ J.

′
XReciprocally, if xj = 0 forXall j ∈ J, it is immediate that for any finite subset J ⊂ J,
xj = 0. And it implies xj = 0.
j∈J ′ j∈J
C.1. SUMMABLE FAMILIES OF POSITIVE REAL NUMBERS 433

Definition C.1.4. We call a nested increasing sequence of finite sets in J any sequence
(J ′ n )n∈N which satisfies
• for all n ∈ N, Jn′ is finite and Jn′ ⊂ Jn+1
′ ;
[
• Jn′ = J.
n∈N

We recall Theorem 0.4.4, which states that the family J is countable if, and only if it
possesses a nested increasing sequence of finite sets.

Proposition C.1.5. Let (xj )j∈J be a family of positive real numbers. Assume that J is
countable and let (J ′ n )n∈N be a nested increasing sequence of finite sets 
in J. Then,
 the
X
family (xj )j∈J is summable if, and only if the sequence of real numbers  xj  is
j∈Jn′
n∈N
bounded from above. Moreover, we have,
   
X X  X
xj = sup xj n ∈ N = lim  xj  .
 ′  n→+∞ ′
j∈J j∈Jn j∈Jn

Proof. We prove the double implications.

⇒: Since for all n ∈ N, Jn′ is finite, it follows from the summability of (xj )j∈J , we have
X X
xj ≤ xj .
j∈Jn′ j∈J

And the direct implication holds.

 
X 
⇐: Reciprocally, we may set M := sup xj n ∈ N . Consider J ′ ⊂ J be a finite

j∈Jn′

[
subset. Since Jn′ = J, there exists n ∈ N such that J ′ ⊂ Jn′ . And we have
n∈N
X X
xj ≤ xj ≤ M.
j∈J ′ j∈Jn′

It shows the summability of (xj )j∈J .

 
X
Moreover, the sum of (xj )j∈J can be calculated as the limit of the sequence  xj 
j∈Jn′
n∈N
following from the monotone convergence property.

Corollary C.1.6. XThe family {xn | n ∈ N} of positive real numbers if summable if, and
only if the series xn is convergent. In that case of convergence, we have
n

X +∞
X
xn = xn .
n∈N n=0
 434 SUMMATION OVER ARBITRARY FAMILIES

Proof. It is sufficient to consider ([[0, n]])n∈N as a nested increasing sequence of finite sets in
N.

Proposition C.1.7. Let (xn )n∈Z be a family of positive real numbers indexed by Z. Then,
the following assertions are equivalent.

1.) The family (xn )n∈Z is summable.

X X
2.) The series xn and x−n are convergent.
n n
X
3.) The series (xn + x−n ) is convergent.
n

n
!
X
4.) The sequence xk is convergent.
k=−n n∈N

If one of the above property is satisfied, we have moreover,

+∞ +∞ +∞ n
!
X X X X X
xn = xn + x−n = x0 + (xn + x−n ) = lim xk .
n→+∞
n∈Z n=0 n=1 n=1 k=−n

Proof. We prove the successive implications.

1.)⇒2.): Since [[0, n]] is a finite subset in Z for all n ∈ N, it follows

! from the summability of
Xn X
the family (xn )n∈N , the sequence of partial sums xk of the series xn is
k=0 n∈N n
X above. Combined with the fact xn ≥ 0 for all n ∈ N, we deduce that
bounded from
the series xn is convergent via the monotone convergence property in R. And the
X n
series x−n follows from the same arguments.
n

2.)⇒3.): It is simply the sum of two convergent series.

X
3.)⇒4.): It is nothing else but the sequence of partial sums of the series (xn + x−n ).
n

4.)⇒1.): It is sufficient to consider ([[−n, n]])n∈N as a nested increasing sequence of finite sets
in Z.

Hence, the claims follow from Proposition C.1.5.

Proposition C.1.8. Let (aj )j∈J and (bj )j∈J be families of positive real numbers indexed
by J.

1.) We suppose that all j ∈ J, aj ≤ bj . Then, (bj )j∈J is summable implies that (aj )j∈J
is summable. In that case, we have moreover,
X X
aj ≤ bj .
j∈J j∈J
C.1. SUMMABLE FAMILIES OF POSITIVE REAL NUMBERS 435

2.) We suppose that (aj )j∈J and (bj )j∈J are summable. Then, for all λ, µ ∈ R+ , the
family (λaj + µbj )j∈J is also summable and we have,
X X X
(λaj + µbj ) = λ aj + µ bj .
j∈J j∈J j∈J

Proof. 1.) Let J ′ ⊂ J be a finite set, it follows from the relations aj ≤ bj for all j ∈ J ′ ⊂ J
that
X X X
aj ≤ bj ≤ bj
j∈J ′ j∈J ′ j∈J
X
Notice that the sum bj is independent of the choice of J ′ . It shows the summability
j∈J
of (aj )j∈J . Taking the supremum over all finite subsets J ′ ⊂ J, we obtain
X X
aj ≤ bj .
j∈J j∈J

2.) Consider a finite subset J ′ ⊂ J, it follows from the linearity of finite sums, we have,
X X X X X
(λaj + µbj ) = λ aj + µ bj ≤ λ aj + µ bj .
j∈J ′ j∈J ′ j∈J ′ j∈J j∈J

It shows the summability of the family (λaj + µbj )j∈J and its sum equals to the
right-hand-side just above by taking the supremum.

Proposition C.1.9. Let (xj )j∈J be a summable family of positive real numbers. Then for
any subset J˜ ⊂ J, the family (xj )j∈J˜ is also summable. Moreover,
X X
xj ≤ xj .
j∈J˜ j∈J

Proof. We consider a family (yj )j∈J of positive real numbers which is given by,
(
xj , if j ∈ J˜
yj =
0, if not.

Then, we have yj ≤ xj holds for all j ∈ J and the X

claim follows
X from Proposition C.1.8.
˜
Moreover, since yj and xj coincides on J, we have xj = yj . Hence, the inequality
j∈J˜ j∈J
follows also.

Proposition C.1.10. Let (xj )j∈J be a summable family of positive real numbers. Let
σ : J → J be a bijective, i.e., a permutation. Then, the family (xσ(j) )j∈J is also summable.
Moreover,
X X
xj = xσ(j) .
j∈J j∈J
436 SUMMATION OVER ARBITRARY FAMILIES

Proof. For any finite set J ′ ⊂ J, the image σ(J ′ ) ⊂ J is also finite. It shows the summability
of the family (xσ(j) )j∈J and we have,
X X
xσ(j) ≤ xj .
j∈J j∈J

Next, it follows from the bijectivity of σ, xj = xσ−1 (σ(j)) for all j ∈ J. Using the summability
of (xσ(j) )j∈J and the previous result, we have,
X X X
xj = xσ−1 (σ(j)) ≤ xσ(j) .
j∈J j∈J j∈J

Combining those inequalities, the claim follows.

Theorem
[ C.1.11 (Associativity of the Sum). Let {Jω | ω ∈ Ω} be a partition of J, i.e.,
J= Jω and Jω ∩ Jω′ = ∅ if ω ̸= ω ′ . Then, the family (xj )j∈J is summable if, and only
ω∈Ω
if
X
• For all ω ∈ Ω, if (xj )j∈Jω is summable we denote sω = xj .
j∈Jω

• The family (sω )ω∈Ω is summable.

In that case, we have moreover,
 
X X X X
xj = sω =  xj  .
j∈J ω∈Ω ω∈Ω j∈Jω

Proof. ⇒: We assume the family (xj )j∈J is summable. For each ω ∈ Ω, it follows from
Jω ⊂ J and Proposition C.1.9 that (xj )j∈Jω is summable.
[
Consider a finite set Ω′ ⊂ Ω, we have always Jω ⊂ J. Since the subsets Jω are
ω∈Ω′
disjoints from one to each others, then, it follows also from Proposition C.1.9
 
X X X X X
sω =  xj  = xj ≤ xj .
ω∈Ω′ ω∈Ω′
S
j∈Jω j∈ ω∈Ω′ Jω j∈J

It shows the summability of the family (sω )ω∈Ω .

⇐: We assume that (xj )j∈J satisfies the two properties. Consider J ′ ⊂ J be a finite
subset. For all k ∈ J ′ , since the family {Jω | ω ∈ Ω} forms a partition of J,
[there
′
exists a subset ωk ∈ Ω such that k ∈ Jωk . In the consequence, we have J ⊂ Jωk .
k∈J ′
Furthermore,
 
X X X X X X
xj ≤ xj =  xj  = sωk ≤ sω .
j∈J ′ k∈J ′ k∈J ′
S
j∈ k∈J ′ J ωk j∈Jωk ω∈Ω

Then, (xj )j∈J is summable.

X X
Finally, the relation sω = xj comes from the above inequalities.
ω∈Ω j∈J
C.1. SUMMABLE FAMILIES OF POSITIVE REAL NUMBERS 437
X
Example C.1.12. 1.) Let an be a convergent series with positive real general terms.
n≥1
Define
+∞
X ak
∀n ∈ N∗ , xn = n .
k(k + 1)
k=n

• Justify the definition of xn .

X
• Study the convergence of xn and calculate its sum in the case of convergence.
n≥1

2.) Let x1 , · · · , xk be real numbers in (0, 1). Show that

k k
X Y p
Y 1
xj j = .
1 − xj
(p1 ,··· ,pk )∈Nk j=1 j=1

3.) Let n ∈ N, n ≥ 2, we denote by π(n) the largest prime divisor of n (e.g. π(29) = 29
and π(256) = 2). Study the convergence of the series
X 1
.
nπ(n)
n≥2

Solution. 1.) • We notice at first that the sequence (an )n∈N is bounded since it con-
verges to 0 as the general terms of a convergent series. For all n ∈ N, we have
the asymptotic behavior as k → +∞,
ak 1
= O( 2 ).
k(k + 1) k
X 1
Since the series converges by Riemann criteria, it follows from the com-
k2
k≥n
X ak
parison test that the series converges. That is, the number xn is
k(k + 1)
k≥n
well-defined.
X
• From the definitions of xn , the series xn is composed with positive real num-
n≥1
bers. Introduce now the set of indexes J := {(n, k) | n, k ∈ N∗ , k ≥ n} as well
as the subsets Jn = {(n, k) | k ∈ N∗ , k ≥ n}. Then, the family (Jn )n∈NX
forms a
partition of J. Apply Theorem C.1.11, the convergence of the series xn is
n≥1
X
equivalent to the summability of the family y(n,k) , where the general terms
(n,k)∈J
y(n,k) is given by,
nak
∀n, k ∈ N∗ , k ≥ n, y(n,k) = .
k(k + 1)

To prove the summability and calculate the sum, we consider another partition
of the family J, which is given by, ∀k ∈ N∗ , Jek = {(n, k) | n ∈ N∗ , n ≤ k}. Then,
438 SUMMATION OVER ARBITRARY FAMILIES

for all k ∈ N∗ the terms y(n,k) is summable on Jek since Jek is a finite set. Also,
we have that
k
X ak X ak
y(n,k) = n= .
k(k + 1) 2
(n,k)∈Jek n=1

X
In parallel, since the series an is convergent, we deduce by applying Theo-
n≥1
rem C.1.11 that the family (y(n,k) )(n,k)∈J is summable and its sum is equal to
+∞
1X
ak .
2
k=1
X
In conclusion, the series xn is convergent and we have,
n≥1

+∞ +∞
X X nak 1X
xn = = ak .
k(k + 1) 2
n=1 n,k∈N∗ , n≥k k=1

2.) We prove the claim by induction on k ∈ N∗ . For the case k = 1, the geometric series
X p 1
x11 is convergent with the sum . So, the claim holds in this case.
p
1 − x 1
1

Assuming k ∈ N∗ and  holds, we consider the numbers x1 , x2 , · · · , xk+1 ∈

 the claim
k+1
Y pj
(0, 1) and the family  xj  . We divide the set of indexes into the
j=1
(p1 ,··· ,pk+1 )∈Nk+1
partition,

Nk+1 = N × Nk
= (0, p2 , · · · , pk+1 )(p2 ,··· ,pk+1 )∈Nk ∪ (1, p2 , · · · , pk+1 )(p2 ,··· ,pk+1 )∈Nk ∪ · · ·
[
= (p, p2 , · · · , pk+1 )(p2 ,··· ,pk+1 )∈Nk .
p∈N

Now, we 
apply the induction
 assumption for k. For all p ∈ N, the family of positive
k+1
p
numbers xp1
Y
xj j  is summable with the sum
j=2
(p2 ,··· ,pk+1 )∈Nk
 
k+1 k+1
xp1 p 1
xj j  = xp1
X Y Y
sp :=
1 − xj
(p2 ,··· ,pk+1 )∈Nk j=2 j=2

X
And the series sp is also convergent as a geometric series, we have thus,
p
 
k+1 k+1
X X
xp1
Y p
Y 1
xj j  = .
1 − xj
p∈N (p2 ,··· ,pk+1 )∈Nk j=2 j=1

Then, the claim follows for k + 1 from Theorem C.1.11.

C.1. SUMMABLE FAMILIES OF POSITIVE REAL NUMBERS 439

3.) We denote by pk the k-th smallest prime number, e.g., p1 = 2, p4 = 7, · · · . For all
k ∈ N∗ we introduce the set

Ak := {n ∈ N | π(n) = pk } .

It is not difficult to see that the sets Ak are disjoint from one to each others. Thus,
we obtain a partition of natural numbers except 0 and 1.
[
N \ {0, 1} = Ak .
k∈N∗
 
1
Then, the claim is to study the summability of the family of numbers .
nπ(n) n≥2
In view of Theorem C.1.11, we concern on the summability of this family on each Ak .
Let k ∈ N∗ and n ∈ Ak , we can decompose n into prime factors as follows,
k
α
Y
n = pk pj j .
j=1

k 
1 Y 1 αj

1
Here, αj ∈ N for all 1 ≤ j ≤ k. So, = 2 . In the consequences,
nπ(n) pk j=1 pj

k 
1 αj

X 1 1 X Y
= 2 .
nπ(n) pk pj
n∈Ak α1 ,··· ,αk ∈N j=1

Applying the previous question, the right-hand-side above is summable then, the
1
family of numbers is summable over the set Ak . Moreover, we can calculate
nπ(n)
its sum,
k  k
1 αj

X 1 1 X Y 1 Y 1
sk := = 2 = 2 .
nπ(n) pk pj pk j=1 1 − 1/pj
n∈Ak α1 ,··· ,αk ∈N j=1
X
Next, we are about to study the convergence of the series sk . The convexity of
k≥1
the function x 7→ − ln(1 − x) on the interval [0, 1/2] shows that

∀x ∈ [0, 1/2], − ln(1 − x) ≤ λx,

3
where λ = 2 ln 2 < . Then, we have,
2
k k
X X 1
− ln(1 − 1/pj ) ≤ λ .
pj
j=1 j=1

Remarking that since pj is prime, so 2pj is not, we can thus deduce with the integral
test,
k k k 2pk Z 2pk +1
3X 1 X 1 X 1 X 1 1
= + ≤ ≤1+ dt = 1 + ln(2pk ).
2 pj pj 2pj m 2 t−1
j=1 j=1 j=1 m=1
440 SUMMATION OVER ARBITRARY FAMILIES

2λ
In the consequences, by denoting µ = < 1,
3
k
X
− ln(1 − 1/pj ) ≤ µ(1 + ln 2 + ln pk ).
j=1

(2e)µ X
Thus, we obtain that sk ≤ . Since 2 − µ < 1, the series sk converges via
pk2−µ k
Riemann test.
X 1
In conclusion, the series converges.
nπ(n)
n≥2

Theorem C.1.13 (Double Sums). Let (x(p,q) )(p,q)∈N2 be a family of positive real numbers
indexed by N2 . Then the following assertions are equivalent.

1.) The family (x(p,q) )(p,q)∈N2 is summable.

 
X X X+∞
2.) For all p ∈ N, the series x(p,q) converges and the series  x(p,q)  converges.
q p q=0

 
X X X+∞
3.) For all q ∈ N, the series xp,q converges and the series  x(p,q)  converges.
p q p=0

!
X X
4.) The series x(p,q) converges.
n p+q=n

In that case, we have moreover,

   
+∞ X+∞ +∞ X+∞ +∞
!
X X X X X
xp,q =  x(p,q) =
  x(p,q) =
 x(p,q) .
(p,q)∈N2 p=0 q=0 q=0 p=0 n=0 p+q=n

Proof. It is sufficient to apply Theorem C.1.11 with the following partitions of N2 respec-
tively.
[
• N2 = {p} × N.
p∈N
[
• N2 = N × {q}.
q∈N
[
• N2 = (p, q) ∈ N2 p + q = n .
n∈N
C.2. SUMMABLE FAMILIES OF VECTORS 441

C.2 Summable Families of Vectors

From now on, (E, ∥ · ∥) denotes a complete normed space (for the definition of such space,
please refer to Chapter 4) over the field K = R or C. In common scenarios, we choose E
either R or C.
Later, we will meet the concept of series valued in an arbitrary vector space. For the
convergence of such series, please refer to Definitions 3.3.8, 3.3.9 and 4.1.20.

Definition C.2.1. We say the family (xj )j∈J ∈ E J is summable if the family (∥xj ∥)j∈J
of positive real numbers is summable. In that case, we denote
X
∥(xj )j∈J ∥ℓ1 (J,E) = ∥xj ∥.
j∈J

Also, we denote the set of summable families (indexed by J and valued in E) by ℓ1 (J, E).

Proposition C.2.2. Let (xj )J∈J and (yj )j∈J be families of vectors in E. Then, the fol-
lowing assertions hold.

• If for all j ∈ J, ∥xj ∥ ≤ ∥yj ∥, then the summability of (yj )j∈J implies that of (xj )j∈J .

• If (xj )J∈J is summable and J ′ ⊂ J, then the family (xj )J∈J ′ is summable.

Proof. It is sufficient to consider the families (∥xj ∥)j∈J and (∥yj ∥)j∈J of positive real num-
bers, then apply Propositions C.1.5 and C.2.5.

Proposition C.2.3. The set of summable families ℓ1 (J, E) endowed with ∥ · ∥ℓ1 (J,E) forms
a normed space.

Proof. At first, we prove ℓ1 (J, E) has the vector space structure. It is clear that ℓ1 (J, E)
is a subset in the vector space E J endowed with the term-to-term addition and external
multiplication. So, it is sufficient to verify that ℓ1 (J, E) is stable under linear combination.
Let x = (xj )j∈J and y = (yj )j∈J be summable families over J. Let λ ∈ K. Then,
x+λy = (xj +λyj )j∈J ∈ E J . Consider a finite subset J ′ ⊂ J, it follows from the homogeneity
and the triangular inequality of ∥ · ∥, we have,
X X X
∥xj + λyj ∥ ≤ ∥xj ∥ + |λ| ∥yj ∥ ≤ ∥x∥ℓ1 (J,E) + |λ|∥y∥ℓ1 (J,E) . (C.1)
j∈J ′ j∈J ′ j∈J ′

We remark that the right-hand-side is independent to the choice of J ′ . Hence, the family
(∥xj + λyj ∥)j∈J is summable, equivalently speaking, x + λy ∈ ℓ1 (J, E).
Furthermore, it is sufficient by taking the supremum of the left-hand-side of (C.1) with
λ = 1 to obtain the triangular inequality

∥x + y∥ℓ1 (J,E) ≤ ∥x∥ℓ1 (J,E) + ∥y∥ℓ1 (J,E) .

For the homogeneity, we choose x = (0E )j∈J and (C.1) gives

∥λy∥ℓ1 (J,E) ≤ |λ|∥y∥ℓ1 (J,E) .

442 SUMMATION OVER ARBITRARY FAMILIES

1
By the same reason and remarking that y = (λy), we obtain,
λ
1
∥y∥ℓ1 (J,E) ≤ ∥λy∥ℓ1 (J,E) .
|λ|

Hence, it holds that ∥λy∥ℓ1 (J,E) = |λ|∥y∥ℓ1 (J,E) . Here, we work on the general case λ ̸= 0.
For the case λ = 0, the result is immediate.
Finally, for the separation property, it is immediate that ∥(0E )∥ℓ1 (J,E) = 0. We assume
now x = (xj )j∈J ∈ ℓ1 (J, E) satisfying that ∥x∥ℓ1 (J,E) = 0. Then, it holds that for all k ∈ J,
X
0 ≤ ∥xk ∥ = ∥xj ∥ ≤ ∥x∥ℓ1 (J,E) = 0.
j∈{k}

And the separation property of ∥ · ∥ gives xk = 0E . Hence, the separation property for
∥ · ∥ℓ1 (J,E) follows.

Definition C.2.4. • We call the support of the family (xj )j∈J the indexes j ′ ∈ J such
that xj ′ ̸= 0E . We denote those indexes by supp((xj )j∈J ). That is,

supp((xj )j∈J ) := j ′ ∈ J xj ′ ̸= 0E .


• We say a family (xj )j∈J has finite support or is finitely supported if its support
supp((xj )j∈J ) ⊂ J is a finite set.

• We denote by E0J the set of families indexed by J and valued in E which have finite
supports.

Proposition C.2.5. Any finitely supported family is summable. Precisely speaking, E0J is
a vector subspace in ℓ1 (J, E).

Proof. Let (xj )j∈J be a family with its finite support K := supp((xj )j∈J ) ⊂ J. Then, for
all finite subset J ′ ⊂ J, we have
X X X
∥xj ∥ = ∥xj ∥ ≤ ∥xj ∥.
j∈J ′ j∈J ′ ∩K j∈K

Remarking that the right-hand-side is independent to the choice of J ′ , we deduce that

(xj )j∈J ∈ ℓ1 (J, E). In fact, the equality occurs once J ′ ⊃ K and it follows that
X
∥(xj )j∈J ∥ℓ1 (J,E) = ∥xj ∥. (C.2)
j∈K

Furthermore, it is not difficult ti verify that the support of the sum of two families is
contained in the union of each supports, and since the two families has finite supports we
can deduce that the support of the sum family is also finite. Hence, E0J is stable under
linear combinations.
In conclusion, E0J is a vector subspace in ℓ1 (J, E).

Proposition C.2.6. E0J is dense in the normed space (ℓ1 (J, E), ∥ · ∥ℓ1 (J,E) ).
C.2. SUMMABLE FAMILIES OF VECTORS 443

Proof. Let x = (xj )j∈J ∈ ℓ1 (J, E). Let ε > 0. From the definition of ∥ · ∥ℓ1 (J,E) and the
characterization of supremum, there exists a finite subset J ′ ⊂ J such that
X
∥x∥ℓ1 (J,E) − ε < ∥xj ∥ ≤ ∥x∥ℓ1 (J,E) ,
j∈J ′

equivalently speaking,
X
0≤ ∥xj ∥ < ε
j∈J\J ′

Then, we consider the family (x̃j )j∈J given by,

(
xj if j ∈ J′
x̃j =
0E otherwise

It is clear that x̃ = (x̃j )j∈J ∈ E0J . Apply Proposition C.2.5, the family x̃ is also summable,
so does for x − x̃. Thus, we obtain from (C.2) that
X
0 ≤ ∥x − x̃∥ℓ1 (J,E) = ∥xj ∥ < ε.
j∈J\J ′

Hence, the claim follows.

Definition C.2.7. Let (xj )j∈J be a family with the finite support K ⊂ J. We define and
denote the sum of this family over J the following vector,
X X
xj := xj ∈ E.
j∈J j∈K

(E0J , ∥ · ∥ℓ1 (J,E) ) −→ (E, ∥ · ∥)

X
Proposition C.2.8. The summation mapping s : (xj )j∈J 7−→ xj is
j∈J
linear and continuous.

Proof. The linearity comes directly from that in E since any linear combination of finitely
supported family possesses also a finite support. For the continuity, consider a finitely
supported family (xj )j∈J , it follows from the triangular inequality in E that

X X X
xj = xj ≤ ∥xj ∥ = ∥(xj )j∈J ∥ℓ1 (J,E) .
j∈J j∈K j∈K

Here, K is the support of (xj )j∈J , which is finite. Combining with the linearity, the sum-
mation mapping s is 1-Lipschitz continuous. Hence, the claim follows.

We next define the sum of a summable family valued in E by using Theorem 4.1.30.
444 SUMMATION OVER ARBITRARY FAMILIES

Definition C.2.9. The summation map can be uniquely extended into a continuous linear
application se on the space ℓ1 (J, E). The sum of a summable family is defined as the value
of this extended map. We denote
X
∀(xj )j∈J ∈ ℓ1 (J, E), xj := se((xj )j∈J ).
j∈J

Proposition C.2.10. Let T be a continuous linear application which maps from E into
another Banach space F , i.e., T ∈ L (E, F ). Let (xj )j∈J ∈ ℓ1 (J, E). Then, (T (xj ))j∈J ∈
ℓ1 (J, F ) and we have,
 
X X
T xj  = T (xj ).
j∈J j∈J

Proof. Since T is continuous and linear, it follows from Theorem 3.6.29 that there exists
M > 0 such that for all x ∈ E, ∥T (x)∥F ≤ M ∥x∥E . Consider a finite subset J ′ ⊂ J, we
have
X X
∥T (xj )∥F ≤ M ∥xj ∥E ≤ M ∥(xj )j∈J ∥ℓ1 (J,E) .
j∈J ′ j∈J ′

Thus, the family (T (xj ))J∈J is summable. Moreover, it follows from the linearity of T
that the claimed equality holds for any finitely supported family. So, the equality holds on
ℓ1 (J, E) from the continuity of T and Theorem 3.6.19. Hence, the claims follow.

Corollary C.2.11. For a real number x ∈ R, we define x+ := max(x, 0) and x− :=

max(−x, 0).
−
• Let (xj )j∈J ∈ RJ . Then, (xj )j∈J ∈ ℓ1 (J, R) if, and only if both (x+
j )j∈J and (xj )j∈J
are summable. In that case, we have
X X X
xj = x+
j − x−
j .
j∈J j∈J j∈J

• Let (xj )j∈J ∈ CJ . Then, (xj )j∈J ∈ ℓ1 (J, C) if, and only if both (Re(xj ))j∈J and
(Im(xj ))j∈J are summable. In that case, we have
X X X
xj = Re(xj ) + i Im(xj ).
j∈J j∈J j∈J

• Let (xj )j∈J ∈ CJ . Then, (xj )j∈J ∈ ℓ1 (J, C) if, and only if both (xj )j∈J is summable.
In that case, we have
X X
xj = xj .
j∈J j∈J
C.2. SUMMABLE FAMILIES OF VECTORS 445

• Let E be a finite dimensional space and let (ê1 , ê2 , · · · , ên ) be a basis in E. We denote
the decomposition in this basis as follows
n
X
∀x ∈ E, x= x(k) êk .
k=1

Consider a family (xj )j∈J indexed by J and valued in E. Then, (xj )j∈J is summable
(k)
if, and only if each family (xj )j∈J indexed by J and valued in R or C is summable.
In that case, we have,
 
X Xn X (k)
xj =  xj  êk .
j∈J k=1 j∈J

Proof. It is sufficient to apply Proposition C.2.10 by remarking the corresponding operations

are linear and continuous.

Proposition
X C.2.12. A family (xn )n∈N ∈ E N indexed by N is summable if, and only if the
series xn is absolutely convergent. Moreover, we have,
n

X +∞
X
xn = xn .
n∈N n=0

Proof. From the definition of summability of (xn )n∈N , the set

( )
X
∥xk ∥ J ′ ⊂ N, J ′ finite
k∈J ′

is bounded from above. In particular, by choosing J ′ = [[0, n]] Xfor all n ∈ N, we obtain
that the sequence of partial sums of the series of positive terms ∥xn ∥ is bounded from
n
[Link] implies the convergence of this series and hence the absolute convergence of the
series xn . Reciprocally, since each finite subset J ′ ⊂ N is contained in an integer interval
n
[[0, ′
Xmax J ]], we can follow the same arguments to deduce that the absolute convergence of
xn implies the summability of the family (xn )n∈N .
n
X +∞
X
It is clear that the equality xn = xn holds for the families with finite supports.
n∈N n=0
Then, it holds also for summable families from the denseness result Proposition C.2.6.

Proposition C.2.13. The sum of a summable family is invariant under any permutation.
Proof. Let σ : J → J be a bijection, i.e., a permutation. Consider a summable family
(xj )j∈J ∈ ℓ1 (J, E) and its permuted family (xσ(j) )j∈J . For any finite subset J ′ ⊂ J, we
have,

X X X
xσ(j) ≤ ∥xσ(j) ∥ = ∥xj ′ ∥ ≤ ∥(xj )j∈J ∥ℓ1 (J,E) .
j∈J ′ j∈J ′ j ′ ∈σ(J ′ )
446 SUMMATION OVER ARBITRARY FAMILIES

It shows the family (xσ(j) )j∈J is also summable. Moreover, by taking the supremum over
all finite subsets J ′ ⊂ J, we obtain that

X
xσ(j) ≤ ∥(xj )j∈J ∥ℓ1 (J,E) . (C.3)
j∈J

X
This equality shows also that the linear application (xj )j∈J 7→ xσ(j) is 1-Lipschitz con-
j∈J
tinuous. Remarking that this application coincides with the sum mapping s on finitely
supported families E0J . Using the denseness argument Proposition C.2.6, we obtain the
equality
X X
xj = xσ(j) .
j∈J j∈J

Remark C.2.14. By taking σ = IdJ , the inequality (C.3) becomes,

X X
xj ≤ ∥xj ∥. (C.4)
j∈J j∈J

Remark C.2.15. The result in Proposition C.2.13 holds automatically for any absolutely
convergent series. For a non-absolutely convergent series, even it is convergent, we can
construct the permutations such that,

• it diverges but bounded;

• it diverges to infinity;

• it converges to any other limit we want;

• it possesses an infinity number of cluster points.

Theorem C.2.16 (Associativity of the Sum). Let {Jω | ω ∈ Ω} be a partition of J. Let

(xj )j∈J be a summable family valued in E. Then, it holds that
X
• For all ω ∈ Ω, the family (xj )j∈Jω is summable. We denote sω = xj .
j∈Jω

• The family (sω )ω∈Ω is summable.

•
 
X X X X
xj = sω =  xj  .
j∈J ω∈Ω ω∈Ω j∈Jω
C.2. SUMMABLE FAMILIES OF VECTORS 447

Proof. • The summability of (xj )j∈Jω comes directly from Proposition C.2.2. Also, by
X X
using (C.4), we have ∥sω ∥ = xj ≤ ∥xj ∥.
j∈Jω j∈Jω

•   family (∥xj ∥)j∈J and apply Theorem C.1.11, we deduce that the family
Consider the
X
 ∥xj ∥ is summable. Combining the inequality obtained in the previous
j∈Jω
ω∈Ω
point and Proposition C.2.2, the family (sω )ω∈Ω is summable. Also, we have the
estimation,
   
X X X X X X X
 xj  = sω ≤ ∥sω ∥ ≤  ∥xj ∥ = ∥xj ∥.
ω∈Ω j∈Jω ω∈Ω ω∈Ω ω∈Ω j∈Jω j∈J

ℓ1 (J, E) −→ E 
It shows the linear application sΩ : X X is 1-Lipschitz
(xj )j∈J 7−→  xj 
ω∈Ω j∈Jω
continuous.

• To show this equality, we are about to show the two summation mapping s and sΩ
are equal. Assuming that the family (xj )j∈J has a finite support J ′ ⊂ J, we have
immediately a partition {J ′ ∩ Jω | ω ∈ Ω} of J ′ . Since J ′ is finite, the intersections
Jω ∩ J ′ is either finite set or empty and only a finitely many intersections are non-
empty. Then, it holds that
X X X
s((xj )j∈J ) = xj = xj = sΩ ((xj )j∈J ).
j∈J ′ ω∈Ω j∈Jω ∩J ′

It shows that s and sΩ coincides on E0J and the claim follows by the denseness argu-
ment Proposition C.2.6.

X (−1)n−1 X 1
Example C.2.17. Consider the series . From the convergence of , we
n2 n2
n≥1 n≥1
know that this alternate series is absolutely convergent. Or, we can say the family of real
(−1)n−1

numbers is summable, thus, we can exchange the order of the summation.
n2 n≥1
Splitting this sum into the odd indexes and even indexes, it holds

+∞ +∞ +∞
X (−1)n−1 X 1 X 1
= −
n2 (2n − 1)2 4n2
n=1 n=1 n=1
+∞ +∞ +∞
X 1 X 1 X 1
= +
n2 (2n − 1)2 4n2
n=1 n=1 n=1
448 SUMMATION OVER ARBITRARY FAMILIES

+∞
X 1 π2
Applying the result of Basel’s Problem, = , we can deduce that
n2 6
n=1

+∞
X 1 π2
= ,
(2n − 1)2 8
n=1
+∞
X (−1)n−1 π2
= .
n2 12
n=1

Proposition C.2.18. Let (xj )j∈J be a summable family of elements in E. We assume that
J is countable and (J ′ n )n∈N is a nested increasing sequence of finite sets in J. Then,
 
X
1.) the sequence  xj  ∈ E N converges;
j∈Jn′
n∈N

2.) the limit in the previous point is independent to the nested increasing sequence of finite
sets (J ′ n )n∈N . That is, if (J ′′ n )n∈N be another nested increasing sequence of finite sets
in J. Moreover, it holds
   
X X X
xj = lim  xj  = lim  xj  .
n→+∞ n→+∞
j∈J j∈Jn′ j∈Jn′′

Proof. Introduce the partition,

!
[
J= J0′ ∪ ′
Jn+1 \ Jn′ .
n∈N
 
X
Then, we apply Theorem C.2.16 to deduce that the family  xj  is summable.
′
j∈Jn+1 \Jn′
X X n∈N
Equivalently speaking, the series xj converges in E. Furthermore, this series is
′
n j∈Jn+1 \Jn′
 
X
the telescoping series corresponding to the sequence  xj  . Thus, we conclude with
j∈Jn′
  n∈N
X X X
the convergence of the sequence  xj  . And we have, xj = lim xj , also
n→+∞
j∈Jn′ j∈J j∈Jn
X n∈N
the sum xj is independent to the choice of the nested increasing sequence (Jn′ )n∈N .
j∈J

Corollary C.2.19. Let (xp,q )p,q∈N be a summable family indexed by N2 and valued in E.
Then, the following assertions hold
C.2. SUMMABLE FAMILIES OF VECTORS 449
X
• For all p ∈ N, the series xp,q is absolutely convergent. Moreover, the series
q
 
X X+∞
 xp,q  is absolutely convergent.
p q=0
X
• For all q ∈ N, the series xp,q is absolutely convergent. Moreover, the series
p
 
X X+∞
 xp,q  is absolutely convergent.
q p=0
!
X X
• The series xp,q is absolutely convergent.
n p+q=n

Moreover, we have,

X +∞ X
X +∞ +∞ X
X +∞ +∞ X
X
xp,q = xp,q = xp,q = xp,q .
p,q∈N2 p=0 q=0 q=0 p=0 n=0 p+q=n

Proof. It is sufficient to apply Theorem C.2.16 with the following partitions of N2 respec-
tively.
[
• N2 = {p} × N.
p∈N
[
• N2 = N × {q}.
q∈N
[
• N2 = (p, q) ∈ N2 p + q = n .
n∈N

X zn
Example C.2.20. 1.) Let z ∈ C with |z| < 1. We consider the series and
1 − z 2n
n≥1
X z 2n−1
.
1 − z 2n−1
n≥1

zn
Since |z| < 1, we have the equivalence relations while n → +∞, ∼ |z|n and
1 − z 2n
z 2n−1
∼ |z|2n−1 . Those relations as well as the comparison tests show that the
1 − z 2n−1
considered series are absolutely convergent, hence, convergent.
For n ∈ N∗ be fixed, we have,
+∞ +∞ +∞
zn X X X
2n
= zn z 2nk = z n(2k+1) = z n(2k−1) .
1−z
k=0 k=0 k=1
450 SUMMATION OVER ARBITRARY FAMILIES

We consider then the family (z n(2k−1) )(n,k)∈N∗ 2 of complex numbers indexed by N∗ 2 .

X
For n ∈ N∗ be fixed, the series |z|n(2k−1) is convergent since |z|2n < 1 then we
k≥1
denote,
+∞
X |z|n
sn := |z|n(2k−1) = .
1 − |z|2n
k=1
X
By using the same equivalence relation above, the series sn converges. Thus, the
n≥1
family (z n(2k−1) )(n,k)∈N∗ 2 is summable and we can exchange the order of summations.
Hence, it holds,
+∞ +∞ X+∞
!
X zn X
n(2k−1)
X
= z = z n(2k−1)
1 − z 2n
n=1 n=1 k=1 (n,k)∈N∗ 2
+∞ X+∞ +∞
!
X X z 2k−1
= z n(2k−1) = .
1 − z 2k−1
k=1 n=1 k=1

X zn X z 2n−1
In conclusion, the two series and are convergent and their
1 − z 2n 1 − z 2n−1
n≥1 n≥1
sums are equal.

2.) We define

2(p − q)
∀p, q ∈ N, up,q = .
(p + q + 1)(p + q + 2)(p + q + 3)

An elementary calculation shows that up,q can be decomposed with

2q + 1 2q + 2 2q + 3
up,q = − +2 −
p+q+1 p+q+2 p+q+3
   
1 1 1 1
= (2q + 3) − − (2q + 1) − .
p+q+2 p+q+3 p+q+1 p+q+2
X
• For q ∈ N be fixed, the above decomposition shows that the series up,q con-
p
verges as a telescoping series of a sequence tending to 0. Moreover, by calculating
its sum, we have,
+∞
X 2q + 3 2q + 1 1 1
sq := up,q = − = − .
q+2 q+1 q+1 q+2
p=0
X
Hence, the series sq converges and we have,
q
 
+∞ +∞ +∞ +∞  
X X X X 1 1
 up,q  = sq = − = 1.
q+1 q+2
q=0 p=0 q=0 q=0
C.2. SUMMABLE FAMILIES OF VECTORS 451

• From the definition of up,q , we have uq,p = −up,q . In the consequences,

   
+∞ X
X +∞ +∞ X
X +∞
 up,q  =  −uq,p  = −1.
p=0 q=0 q=0 p=0

• Let
Xn ∈ N. It follows from the symmetry up,q = −uq,p , we can deduce that
up,q = 0. Hence,
p+q=n

+∞
!
X X
up,q =0
n=0 p+q=n

Those cases shows that the family (up,q )(p,q)∈N2 is not summable and the order of
summation could give different results.

Proposition C.2.21. The normed space (ℓ1 (J, E), ∥ · ∥ℓ1 (J,E) ) is complete once the space
E is complete.

Proof. We assume (E, ∥ · ∥) is a complete normed space. Consider a Cauchy sequence

(n)
(x(n) )n∈N in ℓ1 (J, E). Here, for all n ∈ N, x(n) = (xj )j∈J denotes a summable family of
vectors in E indexed by J.
Let ε > 0. Since (x(n) )n∈N is Cauchy, there exists N ∈ N such that for all n, p ∈ N,
n ≥ N implies ∥x(n+p) − x(n) ∥ℓ1 (J,E) ≤ ε. Then for all j ∈ J, we have

(n+p) (n)
∥xj − xj ∥ ≤ ∥x(n+p) − x(n) ∥ℓ1 (J,E) ≤ ε,

(n)
once n ≥ N . It shows that (xj )n∈N is a Cauchy sequence in E. From the completeness of
(n)
E, this sequence converges, so, there exists xj ∈ E such that lim xj = xj . Now, we are
n→+∞
about to prove that the family x := (xj )j∈E is summable and it is the limit of (x(n) )n∈N in
ℓ1 (J, E).
Consider a non-empty finite subset J ′ ⊂ J, and choose j ∈ J ′ . It follows from the
(n) (n) 1
convergence lim xj = xj , there exists Nj ∈ N such that ∥xj − xj ∥ ≤ once
n→+∞ Card(J ′ )
n ≥ Nj . Then, for n ∈ N, n ≥ max′ Nj the following estimation holds,
j∈J

X X (n) (n)
 X
(n)
X (n)
∥xj ∥ ≤ ∥xj − xj ∥ + ∥xj ∥ = ∥xj − xj ∥ + ∥xj ∥
j∈J ′ j∈J ′ j∈J ′ j∈J ′
(n)
≤ 1 + ∥x ∥ℓ1 (J,E) ≤ 1 + M.

Here, M > 0 be an upper bound of the sequence (x(n) )n∈N in ℓ1 (J, E), it is licit since any
Cauchy sequence is bounded. So, it shows the summability of the family (xj )j∈J .
Next, let ε > 0 and J ′ ⊂ J be a non-empty finite subset in J. Since (x(n) )n∈N is Cauchy,
ε
there exists N ∈ N such that for all n, p ∈ N, n ≥ N implies ∥x(n+p) − x(n) ∥ℓ1 (J,E) ≤ . For
2
(n)
each j ∈ J ′ , we apply the convergence lim xj = xj . There exists Pj ∈ N such that for
n→+∞
452 SUMMATION OVER ARBITRARY FAMILIES

(n+p) ε
all n ∈ N and p ≥ Pj , we have ∥xj − xj ∥ ≤ . So, for n ≥ N and p ≥ max′ Pj ,
2Card(J ′ ) j∈J
we have the estimation,
X (n) X (n) (n+p)
X (n+p)
∥xj − xj ∥ ≤ ∥xj − xj ∥+ ∥xj − xj ∥
j∈J ′ j∈J ′ j∈J ′
(n+p)
X
≤ ∥x(n) − x(n+p) ∥ℓ1 (J,E) + ∥xj − xj ∥
j∈J ′
ε ε
≤ + Card(J ′ ) = ε.
2 2Card(J ′ )

Note that the choice of N is independent to J ′ , we can thus take the supremum of the
left-hand-side over J ′ ⊂ J to deduce that for n ≥ N , it holds

∥x(n) − x∥ℓ1 (J,E) ≤ ε.

Hence, lim ∥x(n) − x∥ℓ1 (J,E) = 0.

n→+∞

In the following definition, E = C. We remark also that this case can be generalized
into any Banach algebra.

Definition C.2.22. We say a family (xj )j∈J indexed by J of complex numbersi is square-
summable if the family (|xj |2 )j∈J is summable. In that case, we denote
sX
∥(xj )j∈J ∥ℓ2 (J,C) = |xj |2 .
j∈J

Also, we denote by ℓ2 (J, C) to presents the set of summable families indexed by J and
valued in C.

Proposition C.2.23. If two families (xj )j∈J and (yj )x∈J are square-summable, then the
family (xj yj )j∈J is summable.

|x|2 + |y|2
Proof. It is the consequence of the inequality |xy| ≤ and Proposition C.2.2.
2

Corollary C.2.24. The set of square-summable families ℓ2 (J, C) forms a vector space.

Proof. Direct consequence of the previous proposition.

Proposition C.2.25. The vector space ℓ2 (J, C) possesses the inner produce space structure
of which the inner product is given by,
X
⟨(xj )j∈J , (yj )j∈J ⟩ℓ2 (J,C) = xj yj .
j∈J
i
We consider here E = C for the reason of well-defined internal multiplications. We can totally replace
C by any Banach algebra.
C.3. APPLICATIONS 453

Proof. The justification of the linearity at left and at right, the symmetry or hermitian, and
separation property is left in Exercises.

Corollary C.2.26 (Cauchy-Schwarz Inequality). Let (xj )j∈J and (yj )x∈J be two square-
summable families. Then, it holds that
2   
X X X
xj yj ≤  |x2j |  |yj |2  .
j∈J j∈J j∈J

And the equality occurs if, and only if there exists λ ∈ C such that ∀j ∈ J, xj = λyj .

C.3 Applications
Example C.3.1 (Cauchy Products). The Cauchy product of two absolutely convergent
series is also absolutely convergent and its sum equals to the product of the sums of those
two series.
Solution.
[ It is the consequence of Theorem C.2.16 by considering the partition N2 =
{(p, q) | p + q = n}.
n∈N

X
Example C.3.2. Let z ∈ C with |z| < 1. Then, the geometric series z n is absolutely
n
1
convergent with the sum . Applying the Cauchy product of this series with itself, we
1−z
have,
+∞
! +∞ +∞
1 X X
p q
XX
2
= z z = (n + 1)z n .
(1 − z) p+q=n
n=0 n=0 n=0

X X
Example C.3.3 (Dirichlet Product of Two Series). Let xn and yn be absolutely
n≥1 n≥1
convergent series.
X X
• We consider the family (xp , yq )(p,q)∈N∗ 2 of complex numbers. Since xn and yn
n≥1 n≥1
are absolutely convergent, we have, the series of general terms
 
X X+∞
|xp yq | = |xp |  |yq |
q∈N∗ q=1

is also absolutely convergent. Thus, the family (xp , yq )(p,q)∈N∗ 2 is summable. From
Theorem C.2.16, we have,
      
X +∞ X
X +∞ +∞
X X+∞ +∞
X +∞
X
xp yq =  x p yq  = xp  yq  =  xp   yq  .
(p,q)∈N∗ 2 p=1 q=1 p=1 q=1 p=1 q=1
454 SUMMATION OVER ARBITRARY FAMILIES
[
Furthermore, by choosing the partition N∗ 2 = {(p, q) | p, q ∈ N∗ , pq = n}, we
n∈N∗
obtain also from Theorem C.2.16 that,

+∞
! +∞ !  +∞   +∞  +∞
!
X X X X X X X
xn yn =  xp   yq  = xp yq = xp yq
n=1 n=1 p=1 q=1 (p,q)∈N ∗2 n=1 pq=n
 
+∞
X X
=  xd y np  .
n=1 p∈N∗ , p|n

• For all n ∈ N∗ , we denote by d(n) to represent the number of positive divisors of

n. We remark that d(n) = Card {p ∈ N∗ | p|n}. Let α > 1. By applying the previous
point, we have
 
+∞ +∞ +∞
!
X d(n) X X 1 X X 1 1
=  =
nα n α pα q α
n=1 n=1 p∈N∗ , p|n n=1 pq=n
+∞
! +∞ !
X 1 X 1
= = ζ(α)2 .
nα nα
n=1 n=1

+∞
X 1
Here, ζ means Riemann’s zeta function, given by ζ(z) = for Re(z) > 1.
nz
n=1

• For all n ∈ N∗ , we introduce Euler’s totient function, denote by φ(n), to represent

the number of positive integers which are relative prime to n. There is a
result in Number Theory that
X
φ(p) = n.
p∈N∗ , p|n

Let α > 2. We have, by combining previous results,

+∞ +∞ +∞
! !
X φ(n) X 1 X φ(q) X φ(q)
ζ(α) = =
nα pα qα pα q α
n=1 n=1 n=1 (p,q)∈N∗ 2
+∞ +∞
X X φ(p) X 1
= = = ζ(α − 1).
nα nα−1
n=1 p∈N∗ , p|n n=1

In conclusion,
+∞
X φ(n) ζ(α − 1)
= .
nα ζ(α)
n=1

Proposition C.3.4. Any summable family has an at most countable support.

C.4. EXERCISES 455

Proof. Let x = (xj )j∈J be a summable family. For the case ∥x∥ℓ1 (J,E)=0 , it follows from the
separation property of the norm ℓ1 (J, E), we have that supp(x) = ∅. We assume then from
now on that ∥x∥ℓ1 (J,E) > 0.
∥x∥ℓ1 (J,E)
 
∗
For all n ∈ N , we introduce the set Jn := j ∈ J ∥xj ∥ ≥ . Then we have,
n
Card(Jn ) ≤ n, which indicates that Jn is a finite set. Also, it is immediate that Jn ⊂ Jn+1
and hence,
[
supp(x) = Jn .
n∈N∗

In other words, the support supp(x) possesses a nested increasing sequence of finite subsets.
Applying Theorem 0.4.4, supp(x) is countable.

C.4 Exercises
Exercise C.4.1. In this Exercise, we are about to reproduce the proof of Associativity of
Sum Theorem C.2.16, which is originally given and known by Fubini and has its analogue
in the theory of integration.
Let (E, ∥ · ∥) be a Banach space and J be a countable set. We consider a family
u := (uj )j∈J indexed by J of vectors in E.

1.) We suppose that u is summable. Show that,  forany nested increasing sequence
X
(Jn′ )n∈N of finite subsets in J, the sequence  uj  of elements in E converges.
j∈Jn′
n∈N
Furthermore, this limit is independent to the choice of the nested increasing sequence
(Jn′ )n∈N .

2.) We adapt the notations

X in the previous question. Show that u is summable if, and
only if the series αn is convergent, where
n
X X
α0 = ∥uj ∥; ∀n ∈ N∗ , αn = ∥uj ∥.
j∈J0 j∈Jn \Jn−1

3.) Now, u is indexed by J = N × N. Show that u = (up,q )p,q∈N is summable if, and only
if it satisfies one of the following condition.
X
• For all p ∈ N being fixed, the series ∥up,q ∥ converges. Also, the series
q
 
X X+∞
 ∥up,q ∥ converges.
p q=0
X
• For all q ∈ N being fixed, the series ∥up,q ∥ converges. Also, the series
p
 
X X+∞
 ∥up,q ∥ converges.
q p=0
456 SUMMATION OVER ARBITRARY FAMILIES

Furthermore, if the condition is fulfilled, we have

     
+∞ X
X +∞ +∞ X
X +∞ +∞ X
X n
 up,q  =  up,q  =  up,n−p  .
p=0 q=0 q=0 p=0 n=0 p=0

Exercise C.4.2. In this Exercise, (E, ∥ · ∥) denotes a Banach space over the filed C and
x = (xj )j∈J is a family of vectors in E indexed by an arbitrary set J.

1.) Show that if x is summable over J then, for any continuous linear form φ : E → C,
the family (φ(xj ))j∈J ∈ CJ of complex numbers is summable and it holds that
 
X X
φ(xj ) = φ  xj  .
j∈J j∈J

2.) Show that, if E is finite dimensional, then x is summable if, and only if every families
of each components are summable.

3.) Show that, if the family x is summable then the following set is bounded
 
X 
′ ′
x J ⊂ J, J finite .
 ′ j 
j∈J

4.) Show that, in the cases where E = R or be finite dimensional spaces, the family x is
summable if, and only if the following set is bounded
 
X 
xj J ′ ⊂ J, J ′ finite .
 ′ 
j∈J

5.) Show that, if the family x is summable then there exists a vector S ∈ E such that for
all ε > 0, there exists a finite subset K ⊂ J such that for all finite subset J ′ satisfying
K ⊂ J ′ ⊂ J, we have

X
xj − S ≤ ε.
j∈J ′

And show that the reciprocal statement holds if E is finite dimensional.

Index

Abel, 22 Bourbaki, 281

adjoint, 107
algebra, 25 Cantor, 173
Banach, 148 ’s diagonal process, 267
normed, 148 cardinal, 17
algebraically closed, 149 Cartesian product, 14
almost everywhere, 294 Cauchy, 43
anti-aymmetry, 29 criteria of continuity, 175
arc, 162 product, 88, 323
Archimedes, 30 sequence, 43, 141
argument, 16 chain rule, 207, 291
Arzelà, 266 change of variables, 292
Ascoli, 266 Characterization
assertion, 13 Constant Functions, 290
associativity, 22, 58 sequence, of closed sets, 116
asymptotic expansion, 417 sequence, of limits, 124
set, of continuous functions, 125
ball supremum/infimum, 33
closed, 112 closure, 120
open, 112 codimension, 132
Banach, 144 coefficients
Basel problem, 91 exponential Fourier, 344
Bernstein, 244 trigonometric Fourier, 344
polynomial, 244 commutativity, 22, 58
Bertrand, 74 compact
Bessel, 347 pre-, 155
bijective, 15 relatively, 266
binary relation, 29 sequentially, 151
Bolzano, 42 topological, 153
Borel, 156 compatibility, 25, 29
bound complement, 14
lower, 32 complete, 47, 142
upper, 32 condition
boundary, 122 Dirichlet, 340
bounded, 32, 112 conjugate exponent, 104
from above, 32 continuity
from below, 32 modulus of, 129
sequences, 35 continuous, 124

457
458 INDEX

Hölder, 129, 358 Dirichlet, 340

Lipschitz, 128, 263 disjoint, 14
piecewise, 241 distance, 99
piecewisely, 84 Hausdorff, 100
uniformly, 128 induced, 101, 123
contraction, 143 distribution, 375
converge, 36, 69, 113 regular, 376
absolute, 80, 147 tempered, 388
absolutely, 249 distributivity, 23
normally, 250 left/right, 25
pointwise, 146 diverge, 36, 69, 113
pointwisely, 231, 247 to infinity, 47
uniformly, 232, 247 domination, 405, 413
uniformly on any compact, 234, 248
eigenfunction, 372
convergence
eigenvalue, 108
circle of, 325
element, 13
disc of, 325
identity, 22
radius of, 324
inverse, 22
cover, 153
embedding
open, 153
continuous, 385
sub-, 153
equicontinuity, 266
equipotent, 18
d’Alembert, 76
equivalence, 102, 408, 413
Darboux, 273
class, 62
upper/lower sum, 273
relation, 61
denseness, 34, 121
Euclid, 179
derivable
Euler, 17
directional, 195
’s constant, 71, 92
derivative, 197, 289
’s formula, 17
directional, 195
duplication formula, 363
left/right-, 289
infinite product for sinus, 363
partial, 196, 214
exponential, 150
pseudo-, 346
complex, 90
Descartes, 14
extension, 15
diagonalizable
ortho-, 107 Fejér, 356
diameter, 100 field, 23
diffeomorphism ordered, 29
C k -, 224 trivial, 24
differentiable, 196, 289 finer, 241
k times continuously, 211 flip, 345
continuously, 200 form
indefinitely, 211 bilinear, 177
left/right-, 289 linear, 132
differential equation sesquilinear, 178
first order, 262 formula
Dini, 258 Cauchy’s Integral, 330
Dirac mass, 376 Fourier inversion, 385
INDEX 459

Gel’fand’s, 139 phenomena, 352

jump’s, 380 gradient, 206
Leibniz, 291, 366 Gram, 184
Parseval, 387 graph, 15
Parseval-Bessel, 355 group, 22
Poisson Summation, 360 abelian/commutative, 22
Stirling, 320, 415 sub-, 56
Fourier, 344
Fréchet, 191 Hölder, 104
differentiable, 220 Hausdorff, 100
fraction, 23, 62 Heaviside, 379
rational, 331 Heine, 159
function, 15 Hermite, 178
Hermitian, 178
C ∞ -class, 211
Hesse, 215
C k -class, 211
Hilbert, 178
affine, 244
homeomorphic, 127
analytic, 336
homeomorphism, 127
characteristic, 242
homogeneity, 101
complex exponential, 332
hyperplane, 132
complex hyperbolic sinus/cosinus,
335 identity
complex sinus/cosinus, 335 parallelogram, 181
composite, 15 Plancherel, 355
cut-off, 374 polarization, 180
distance, 99 image, 15, 132
entire analytic, 336 pre-, 15
Euler’s Gamma, 319 inequality
generalized, 375 Bessel, 187, 347
Heaviside, 379 Cauchy-Schwarz, 104, 181, 310
identity, 15 Hölder, 104, 135
indication, 242 mean value, 223, 293
inverse, 16 Minkowski, 103, 136
moderate increasing, 389 triangular, 30, 99, 101, 304
piecewise constant, 242 Young’s, 135
piecewisely C 1 -class, 345 infimum, 32
primitive, 291 initial condition, 262
rapid decreasing, 359, 384 initial value problem, 262, 372
regulated, 281 injective/one-to-one, 15
rest, 247 integrable, 84, 294
smooth, 211, 366 Darboux, 273
step, 242 Riemann, 272
sum, 247, 325 square-, 309, 341
functional, 375 integral, 84, 269
Darboux, 273
Gauss, 317 improper, 306
Gel’fand, 139 indefinite, 291
general terms, 69, 113 Riemann, 272
Gibbs, 352 integral equation, 264
460 INDEX

integration symmetric, 107

Borel-Lebesgue, 277 unitary, 107
interior, 118 maximum, 32
intersection, 14 member, 13
interval, 31 mesh, 241
closed, 31 minimum, 32
open, 31 Minkowski, 103
invertible, 23 modulus, 16
isomorphism, 221 mollifiers, 374
multi-index, 214
kernel, 132
Dirichlet’s, 349 negligence, 406, 413
Fejér, 356 negligible, 294
heat/Gaussian/Gauss-Weierstrass, neighborhood, 117
317 Neumann, 148
norm, 101
Laplace, 215 p-, 102
Law algebraic, 148
De Morgan’s, 116 Euclidean, 182
of Trichotomy, 30, 58 Hermitian, 182
Lebesgue, 156 matrix, 105
Lemma operator, 168
Hadamard’s, 316 square-integrable, 179
Riemann-Lebesgue, 368 numbers
sandwich, 38 algebraic, 27
limit, 36 complex, 16
from left/right, 124 integers, 16, 60
pointwise, 231 natural, 16, 57
superior/inferior, 47 rational, 16, 61
uniform, 232 real, 16, 64
uniqueness, 37, 113, 124 transcendent, 27
linear
algebra, 107, 132 operation
application, 132, 177 addition, 23, 57
map, 132 binary, 22
Liouville, 330 multilpication(internal), 23
Lipschitz, 128 multiplication(external/scalar), 25
logarithmic operator, 168
principal determination of, 335 differential, 212, 214
integral, 263
magnitude, 16 order, 29
magnitude/absolute value, 30 total, 29
mapping, 15 orthogonal, 183
matrix orthonormal, 184
Hermitian, 107
Hessian, 215 Parseval, 355
Jacobian, 196 part
nilpotent, 138 imaginary, 16
orthogonal, 107 integer, 34
INDEX 461

real, 16 proposition, 13
partition, 122, 241 Pythagoras, 184
adapted, 241
tagged, 272 range, 15
Peano refinement, 241
’s axioms, 57 common, 241
period, 339 reflexivity, 29, 102
periodic, 339 rest, 71
permutation, 82 restriction, 15
phase, 16 Riemann, 73
Picard, 263 sum, 272, 278
’s iterations, 227, 263 sums, 84
Plancherel, 355 Zeta Function, 81, 253
Poincaré, 417 Riesz, 166
point ring, 22
adherent, 120 Schmidt, 184
closure, 120 Schwarz, 104
cluster, 41 segment, 31
critical, 215 semi-linear, 178
interior, 118 sequence, 35
wandering, 176 adjacent, 54
Poisson, 360 bounded, 108
polar decreasing, 36
form, 17 increasing, 36
pole, 331 monotone, 36
positive, 177 nested decreasing, 54, 142
definite, 177 nested increasing, 26, 433
predicates, 13 partial sums, 69, 113, 247
principle square-summable, 109
complete induction, 59 sub-, 41
contraction mapping, 144 summable, 108
mathematical induction, 59 uniform Cauchy, 234
well-ordering, 17, 34, 59 series, 69, 113
product alternate, 87
convolution, 317, 382, 392 Bertrand, 74
Hermitian, 178 Fourier, 345
inner, 111, 177 geometric, 70
scalar, 177 harmonic, 74
projector, 185 Neumann, 148
orthogonal, 185 power, 323
property Reimann, 73
Archimedean, 30 telescoping, 71
Bolzano-Weierstrass, 42 trigonometric, 343
Heine-Borel, 165 sesquilinear, 178
least upper bound, 34 set, 13
monotone convergence, 41 arc-connected, 162
nested decreasing segments, 54 Cantor, 173
separation, 99, 101 closed, 115
462 INDEX

connected, 161 support, 318, 373, 442

convex, 163 finite, 442
countable, 18 supremum, 32
disconnected, 161 surjective/onto, 15
empty, 14 symbol
finite, 17 Kronecker, 107
infinite, 17 symmetric, 177
open, 114 symmetry, 102
power, 14
star-shaped, 163 Taylor, 215
uncountable, 18 tend to, 123
similitude class, 138 test
singleton, 31 comparison, 72
space integral, 85, 300
L1 , 308 ratio, 76
L1loc , 375 Riemann, 74
L2 , 309 root, 76
L∞ , 235 Theorem
Lp , 367 Ascoli-Arzelà, 266
ℓ1 , 108, 441 Banach Fixed Point, 144
ℓ2 , 109, 452 Bernstein, 337
ℓ∞ , 108 Bolzano-Weierstrass in Rn , 157
Banach, 144 Borel-Lebesgue, 156
connected, 161 Cauchy-Lipschitz, 372
dual, 165, 191 Continuity under Integral Signs, 312
Euclidean, 179 Derivation under Integral Signs, 312
Hermitian, 179 Dini’s, 258
Hilbert, 189 Dirichlet, 349, 351
induced metric, 123 Fubini, 314
inner product, 111, 178 Fundamental Homomorphism, 132
metric, 99 Fundamental, of Calculus, 291
normed, 101 General Bolzano-Weierstrass, 165
pre-Hilbert, 178 Gram-Schmidt Orthonormalization
product, 113, 143 Procedure, 184
Schwartz, 384 Hahn-Banach, 151, 192
topological, 115 Heine, 159
topological vector (TVS), 375 Heine-Borel, 165
vector, 25 Implicit Function, 229
spectral radius, 108, 138 Interchange of Limits, 237
sphere, 112 Intermediate Values, 164
Stieljes, 417 Inverse Functoin, 226
Stirling, 415 Lebesgue’s Dominated Convergence,
Stone, 246 311
subset, 14 Liouville, 330
sum, 69, 323, 432 local Cauchy-Lipschitz, 263
Fourier, 345 Plancherel, 387
summable, 432, 441 Pythagoras, 184
square-, 452 Rank-nullity, 132
INDEX 463

Riesz, 166 trigonalizable, 149

Riesz-Fréchet Representation, 191, trigonometric polynomial, 342
206
Schwarz, 212 union, 14
Stone-Weierstrass, 246 unit
structure of subgroups in (R, +), 56 sphere, 101, 112
Taylor, 215 vector, 101
Taylor-Young, 419
value, 15
Trigonometric Weierstrass, 353
vanish, 318
topology, 114
variable, 15
transform
vector, 25
Abel, 87, 96
field, 206
Fourier, 360, 367, 389
Laplace, 320 Weierstraß, 42
transitivity, 29, 102
translation, 345, 369 Young, 135
464 INDEX
Bibliography

[1] Henri Cartan Calcul Différentiel, Hermann, Paris, 1967.

[2] Ching-Hsiao Arthur Cheng An Introduction to Mathematical Analysis, Lecture note in

National Central University, 2022.

[3] Claude Deschamps and André Warusfel Mathématiques TOUT EN UN 1er année
MPSI-PCSI, Dunod, Paris, 2003.

[4] Claude Deschamps and André Warusfel Mathématiques TOUT EN UN 2e année MP,
PC, PSI, Dunod, Paris, 2001.

[5] E. Landau Foundation of Analysis, Chelsea Pub., New York, 1951.

[6] S. Meignen Analyse fonctionelle: du continu au discret lecture notes in ENSIMAG

second year, 2015/2016.

465