An Essential

Mathematical Analysis

Handbook
Università degli studi di Roma ”La Sapienza”
Physics and Astrophysics BSc

Matteo Cheri

Theorems, proofs and some examples for the Physics undergrad

July 15, 2021

´
Version R δ(x) dx
 An Essential

Mathematical Analysis
Handbook

An essential handbook on Mathematical Analysis for BSc students majoring in Physics

Written by
Matteo Cheri
Università degli Studi di Roma ”La Sapienza”
Physics and Astrophysics BSc

LATEX 2ε inside, ViM powered.

July 15, 2021

´
Version R δ(x) dx
Contents

1 Complex Numbers and Functions 9

1.1 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Regions in C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 Extended Complex Plane Ĉ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Elementary Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.1 Complex Exponentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Properties of Trigonometric Functions . . . . . . . . . . . . . . . . . . . . . . . 12

2 Abstract Spaces 15
2.1 Metric Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Convergence and Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Vector Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 Hölder and Minkowski Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Differential Analysis 27
3.1 Digression on the Notation Used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1.1 Differential Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Curves in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Differentiability in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Differentiability in C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.1 Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.6 Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6.1 Critical Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6.2 Convexity and Implicit Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6.3 Lagrange Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4 Tensors and Differential Forms 51

4.1 Tensors and k-forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.1.1 Basic Definitions, Tensor Product and Wedge Product . . . . . . . . . . . . . . 51
4.1.2 Volume Elements and Orientation . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2 Tangent Space and Differential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2.1 External Differentiation, Closed and Exact Forms . . . . . . . . . . . . . . . . . 56

1
CONTENTS 2

4.3 Chain Complexes and Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.3.1 Singular n−cubes and Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.2 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5 Integral Analysis 61
5.1 Measure Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.1.1 Jordan Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.1.2 Lebesgue Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.2 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2.1 Lebesgue Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.2.2 Lebesgue Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Calculus of Integrals in R2 and R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.3.1 Double Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.3.2 Triple Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.3.3 Change of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.3.4 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3.5 Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.4 Integration in C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.4.1 Integration of Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . . . 81
5.4.2 Integral Representation of Holomorphic Functions . . . . . . . . . . . . . . . . 85
5.5 Integral Theorems in R2 and R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

6 Sequences, Series and Residues 91

6.1 Sequences of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Series of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.1 Power Series and Convergence Tests . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3 Series Representation of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.3.1 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.3.2 Laurent Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.3.3 Multiplication and Division of Power Series . . . . . . . . . . . . . . . . . . . . 102
6.3.4 Useful Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.4 Residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.4.1 Singularities and Residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.4.2 Classification of Singularities, Zeros and Poles . . . . . . . . . . . . . . . . . . . 105
6.5 Applications of Residue Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.5.1 Improper Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.5.2 General Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

7 Hilbert and Banach Spaces 115

7.1 Banach Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.1.1 Sequence Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.1.2 Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7.1.3 Function Spaces in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.2 Hilbert Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.3 Projections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.3.1 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.3.2 Projections and Orthogonal Projections . . . . . . . . . . . . . . . . . . . . . . 126
CONTENTS 3

7.3.3 Orthogonal Systems and Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

8 Distributions 131
8.1 Linear Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.1.1 Dual Spaces and Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.2 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.2.1 Local Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.2.2 Regular and Singular Distributions . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.2.3 Operations with Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
8.2.4 A Physicist’s Trick, The Dirac δ “Function” . . . . . . . . . . . . . . . . . . . . . 154
8.3 Integral Representation of Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.3.1 Dirac Delta and Heaviside Theta . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.3.2 Distributions in Rn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.4 Some Applications in Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.4.1 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

9 Ordinary Differential Equations 161

9.1 Existence of Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
9.2 Common Solving Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
9.2.1 First Order Linear ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
9.2.2 Second Order Linear ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

10 Fourier Calculus 175

10.1 Bessel Inequality and Fourier Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . 175
10.2 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
10.2.1 Fourier Series in L2 [−π, π] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
10.2.2 Fourier Series in L2 [a, b] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
10.2.3 Fourier Series in Symmetric Intervals, Expansion in Only Sines and Cosines . . . 181
10.2.4 Complex Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
10.2.5 Piecewise Derivability, Pointwise and Uniform Convergence of Fourier Series . . 185
10.2.6 Solving the Heat Equation with Fourier Series . . . . . . . . . . . . . . . . . . . 188
10.3 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
10.3.1 Fourier Integrals, Translations, Dilations . . . . . . . . . . . . . . . . . . . . . . 190
10.3.2 Behavior of Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
10.3.3 Inverse Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
10.3.4 Convolution Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
10.3.5 Solving the Heat Equation with Fourier Transforms . . . . . . . . . . . . . . . . 203

A Useful Concepts 207

A.1 Multi-Index Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
A.2 Properties of the Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

B Common Fourier Transforms 209

B.1 Common Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
CONTENTS 4
Introduction

The idea of writing this came from a mix of pure laziness and despair in preparing two exams, Analisi
Vettoriale (Vector Analysis) and Metodi e Modelli Matematici per la Fisica (Mathematical Models and
Methods for Physics, basically just pure Complex Analysis and Functional Analysis), that I have to sustain
in my BSc course in Physics here at Sapienza University of Rome.
These two courses are deeply intertwined and it’s really difficult to study them apart due to the sheer
volume of things that either are done for half in one course and half in the other one, or they simply
get generalized in the second, breaking up the logical flow one might get from studying these two
seemingly completely different subjects.
There will surely be errors in grammar, typing and probably some mathematical inaccuracies so for
any question, inquiry or you just want to say hi to me, don’t wait in sending me an email here
[email protected].
A huge thanks to anyone of you that’ll read this, I hope it will be useful for you as it is to me.

5
CONTENTS 6
Notation

• F Ordered scalar field, R or C when not specified

• an , (a)n Sequence
• ((a)k )n Sequence of sequences
• x, xµ Vector
• x ∧ y, µνγ xν y γ Cross product of x, y
∂f
• ∇f , ∂µ f , ∂xµ Gradient of f
∂f µ
• h∇, fi, div(f), ∂µ f µ , ∂xµ Divergence of f
• ∇ ∧ f, rot(f), µνγ ∂ ν f γ Rotor of f
∂f µ
• Jf(x), ∂µ f ν , ∂xν Jacobian matrix of f
2 ∂2f
• Hf (x), ∂µ ∂ν f , ∂µν f, ∂xµ ∂xν Hessian matrix of the function f

• ∀† For almost all

• Cc∞ (R) = K Space of test functions, (smooth with compact support)
• Zfm , Sf , Pfm Sets of zeros of order m, singularities or poles of order m of a function f
• an → a Simple convergence
• an ⇒ a Uniform convergence
A
• an −→ a Absolute convergence
T
• an −→ a Total convergence
• an *w a Weak convergence
• an *K a K−convergence
• an *? a K? −convergence

7
CONTENTS 8
1 Complex Numbers and Functions

§ 1.1 Complex Numbers

Definition 1.1.1 (Complex Numbers). Define with C the set of complex numbers, i.e. the set of
numbers z ∈ C : z = (x, y) and x, y ∈ R.
We define the real and imaginary parts of z as follows

Re(z) = x
(1.1)
Im(z) = y

Definition 1.1.2 (Operations in C). Take z1 , z2 ∈ C, then we define

z1 = z2 ⇐⇒ Re(z1 ) = Re(z2 ), Im(z1 ) = Im(z2 )
z1 + z2 = (Re(z1 ) + Re(z2 ), Im(z1 ) + Im(z2 ))
z1 z2 = (Re(z1 )Re(z2 ) − Im(z1 )Im(z2 ), Re(z1 )Im(z2 ) + Im(z1 )Re(z2 ))

Theorem 1.1. With the previous definitions the set C forms a field.
Definition 1.1.3 (Imaginary Unit). We define the imaginary unit i = (0, 1) ∈ C. From this definition
and the definition of product of two complex numbers, we have that i2 = −1
With this definition, we have
∀z ∈ C z = Re(z) + iIm(z) (1.2)

Definition 1.1.4 (Complex Conjugate). Taken z ∈ C, we call the complex conjugate of z the number
w such that
w = Re(z) − iIm(z) (1.3)
This number is denoted as z
Definition 1.1.5 (Complex Module). We define the module or norm of a complex number, the
following operator.
√ q
kzk = zz = Re2 (z) + Im2 (z) (1.4)

9
CHAPTER 1. COMPLEX NUMBERS AND FUNCTIONS 10

Definition 1.1.6 (Complex Inverse). The inverse of a complex number z ∈ C is defined as z −1 and it’s
calculated as follows
z
z −1 = 2 (1.5)
kzk

Definition 1.1.7 (Polar Form). Taken a complex number z ∈ C one can define it in polar form with its
modulus r and its argument θ. We have that, if z = x + iy
2
p
r = x2 + y 2 = kzk
y (1.6)
tan(θ) =
x

Definition 1.1.8 (Principal Argument). Taken arg(z) = θ we can define two different arguments, due
to the periodicity of the tan function.

1. Arg(z) ∈ (−π, π] called the principal argument

2. arg(z) = Arg(z) + 2kπ, k ∈ Z called the argument

As a rule of thumb, using the previous definition of argument of a complex number z = x + iy, we
have 
arctan(y/x) − π x < 0, y < 0

Arg(z) = arctan(y/x) x ≥ 0, z 6= 0 (1.7)

arctan(y/x) + π x < 0, y ≥ 0


Definition 1.1.9 (arg+ ). Given z ∈ C we define the arg+ (z) as the only value of arg(z) such that
0 ≤ θ < 2π.
In case we have a polydromic function, in order to specify we’re using this argument, there will be a +
as index. √ +
I.e. log+ (z), [z a ]+ , z , · · · and so on.

Theorem 1.2 (De Moivre Formula). A complex number z ∈ C in polar form can be written with
complex exponential and sine and cosine function as follows.
2 2
z = kzk ei arg z = kzk (cos(arg z) + i sin(arg z)) (1.8)

This formula easily generalizes the calculus of exponentials of complex

√ numbers. With this
definition, it’s obvious that the n−th root of a complex number n z has actually n − 1 results,
given the 2π−periodicity of the arg(z) function.

§ 1.2 Regions in C
Definition 1.2.1 (Line). A line λ in C, from z1 , z2 can be written as follows

λ(t) = z1 + t(z2 − z1 ) t ∈ [0, 1] (1.9)

1.3. ELEMENTARY FUNCTIONS 11

If t ∈ R this defines the line lying between z1 , z2 . Its non-parametric representation is the following
   
z − z1
{λ} := z ∈ C| Im =0 (1.10)
z2 − z1

Where z = λ(t).
Definition 1.2.2 (Circumference). A circumference γ centered in a point z0 ∈ C with radius R is defined
as follows
γ(θ) = z0 + Reiθ θ ∈ [0, 2π] (1.11)
Non parametrically, it can be defined as follows

{γ} := { z ∈ C| kz − z0 k = R} (1.12)

§§ 1.2.1 Extended Complex Plane Ĉ

Definition 1.2.3 (Extended Complex Plane). We define the extended complex plane Ĉ as follows

Ĉ = C ∪ {∞} (1.13)

This can be imagined by projecting C into the Riemann sphere centered in the origin.
Definition 1.2.4 (Points in Ĉ). Given a point z ∈ C, z = x + iy we can find its coordinates with the
following transformation
ẑ = (xt, yt, 1 − t) ∈ Ĉ (1.14)
Where the condition kẑk = 1 must hold, defining the value of t ∈ R Inversely, given ẑ = (x1 , x2 , x3 ) ∈ Ĉ
one finds
x1 + ix2
z= (1.15)
1 − x3

§ 1.3 Elementary Functions

Definition 1.3.1 (Exponential). The exponential function z 7→ ez with z ∈ C is defined as follows

ez = eRe(z)+iIm(z) = eRez (cos(Im(z)) + i sin(Im(z))) (1.16)

This gives
kez k = eRe(z)
(1.17)
arg(ez ) = Im(z) + 2πk k∈Z
We have therefore, for z, w ∈ C
ez ew = ez+w
ez (1.18)
= ez−w
ew
CHAPTER 1. COMPLEX NUMBERS AND FUNCTIONS 12

Definition 1.3.2 (Logarithm). We define the logarithm function z 7→ log z as follows

log(z) = log kzk + i arg(z) (1.19)

It’s evident how this function has multiple values for the same z value, and therefore is known as a
polydromic function, like the square root. We also define the principal branch of the logarithm as
Log(z)
Log(z) = log kzk + i Arg(z) (1.20)
Lastly we define the log+ (z) as follows

log+ (z) = log(kzk) + i arg+ (z) (1.21)

Definition 1.3.3 (Branch of the Logarithm). A general branch of the log function is defined as the
function f (z) : D ⊂ C −→ C such that
ef (z) = z (1.22)

§§ 1.3.1 Complex Exponentiation

Definition 1.3.4 (Complex Exponential). Taken s, z ∈ C, we define the complex exponential a follows,
taken z a variable
z s = es log(z) z 6= 0 (1.23)
Its derivative has the following value

d s
z = se(s−1) log(z) = sz s−1 (1.24)
dz
Alternatively, we define
sz = ez log(s) (1.25)

§§ 1.3.2 Properties of Trigonometric Functions

Definition 1.3.5 (Trigonometric Functions). Using De Moivre’s formula, we define
1 iz
e − e−iz


sin(z) =
2i (1.26)
1 iz
e + e−iz


cos(z) =
2

Definition 1.3.6 (Hyperbolic Functions). We define the hyperbolic functions as follows, given z = iy

sinh(y) = −i sin(iy)
(1.27)
cosh(y) = cos(iy)
1.3. ELEMENTARY FUNCTIONS 13

For a general value of z, we define

1 z
e − e−z


sinh(z) =
2 (1.28)
1 z
e + e−z


cosh(z) =
2

Theorem 1.3 (Trigonometric Identities). Given z, z1 , z2 ∈ C we have

sin2 (z) + cos2 (z) = 1

sin(z1 ± z2 ) = sin(z1 ) cos(z2 ) ± cos(z1 ) sin(z2 )
cos(z1 ± z2 ) = cos(z1 ) cos(z2 ) ∓ sin(z1 ) sin(z2 )
sin(z) = sin (Re(z)) cosh (Im(z)) + i cos (Re(z)) sinh (Im(y))
cos(z) = cos (Re(z)) cosh (Im(z)) − i sin (Re(z)) sinh (Im(y))
2 2
ksin(z)k = sin2 (Re(x)) + sinh (Im(y))
2 2
kcos(z)k = cos2 (Re(x)) + sinh (Im(y))
2 2
(1.29)
cosh (z) − sinh (z) = 1
sinh(z1 ± z2 ) = sinh(z1 ) cosh(z2 ) ± cosh(z1 ) sinh(z2 )
cos(z1 ± z2 ) = cosh(z1 ) cosh(z2 ) ± sinh(z1 ) sinh(z2 )
sinh(z) = sinh (Re(z)) cos (Im(z)) + i cosh (Re(z)) sin (Im(y))
cos(z) = cosh (Re(z)) cos (Im(z)) + i sinh (Re(z)) sin (Im(y))
2 2
ksin(z)k = sinh (Re(x)) + sin2 (Im(y))
2 2
kcos(z)k = cosh (Re(x)) + sin2 (Im(y))

Definition 1.3.7 (Inverse Trigonometric Functions). Given z ∈ C we define

 p 
arcsin(z) = −i log iz + 1 − z 2
 p 
arccos(z) = −i log z + i 1 − z 2 (1.30)
 
i i−z
arctan(z) = − log
2 i+z

Definition 1.3.8 (Inverse Hyperbolic Functions). Given z ∈ C we define

 p 
asinh(z) = log z + z 2 + 1
 p 
arccos(z) = log z + z 2 − 1 (1.31)
 
1 1+z
atanh(z) = log
2 1−z
CHAPTER 1. COMPLEX NUMBERS AND FUNCTIONS 14
2 Abstract Spaces

§ 2.1 Metric Spaces

§§ 2.1.1 Topology
Definition 2.1.1 (Metric Space). Let X be a non-empty set equipped with an application d, defined as
follows
d : X × X −→ F
(2.1)
(x, y) → d(x, y)
Where F is an ordered field.
The couple (X, d) is said to be a metric space, if and only if ∀x, y, z ∈ X the application d satisfies the
following properties

1. d(x, y) ≥ 0

2. d(x, x) = 0

3. d(x, y) = d(y, x)

4. d(x, y) ≤ d(x, z) + d(z, y)

Definition 2.1.2 (Ball). Let (X, d) be a metric space. We then define the open ball of radius r,
centered in x in X (BrX ), and the closed ball of radius r centered in x (BrX ) as follows

BrX (x) : = { u ∈ X| d(u, x) < r}

(2.2)
BrX (x) : = { u ∈ X| d(u, x) ≤ r}

When there won’t be doubts on on where the ball is defined, the superscript indicating the set of
reference will be omitted.

We’re now ready to define the topology on a metric space

Definition 2.1.3 (Open Set). Let (X, d) be a metric space, and A ⊆ X a subset. A is said to be an
open set if and only if
∀x ∈ X ∃BrX (x) ⊂ A (2.3)

15
CHAPTER 2. ABSTRACT SPACES 16

Definition 2.1.4 (Complementary Set). Let A be a generic set, then the set Ac is defined as follows

Ac := {a ∈
/ A} (2.4)

This set is said to be the complementary set of A.

It’s also obvious that A ∩ Ac = {}

Definition 2.1.5 (Closed Set). Alternatively to the notion of open set, we can say that E ⊆ X is a
closed set, if and only if
∀x ∈ E c ∩ X ∃BrX (x) ⊂ E c ∩ X (2.5)

Remark. A set isn’t necessarily open nor closed!

Proposition 1. 1. The set BrX (x) is open

2. The set BrX (x) is closed

Proof. Let A = BrX (x). If A is open, we have therefore, applying the definition of open set, that

∀x ∈ A ∃ > 0 : BX (x) ⊂ A

So
x0 ∈ A =⇒ d(x, x0 ) < r
∴  = r − d(x, x0 ) > 0
Then, by definition of open ball we have

y ∈ BX (x) =⇒ d(x, y) < 

Then, we can say that

d(y, x0 ) ≤ d(y, x) + d(x, x0 ) <  + d(x, x0 ) = r
∴ y ∈ BX (x) =⇒ y ∈ BrX (x0 ) ⊂ A
The demonstration of the second point is exactly the same, whereby we take E as our closed ball and
A = Ec

Proposition 2. Let (X, d)

1. The sets {}, X are open

2. The sets {}, X are closed

Tn
3. If {Ai }ni=1 is a collection of open sets, then A = i=1 Ai is open
Sn
4. If {Ci }ni=1 is a collection of closed sets, then C = i=1 Ci is closed

5. Let I ⊂ N be an index set, then

S
(a) If {Aα }α∈I is a collection of open sets, then B = α∈I Aα is open
T
(b) If {Cα }α∈I is a collection of closed sets, then D = α∈I Cα is closed
2.1. METRIC SPACES 17

Proof. The first two statements are of easy proof. Let BX ⊂ {}. This means that BX is empty and
therefore BX = {}, which makes it open by definition. Therefore we have that {}c = X, and X must
be closed, but if we reason a bit, we can say that ∀x ∈ X BX (x) ⊂ X, which means that X is open,
thus X c = {} must be closed.
Since we gave a proof for {} and X being open, we have that these two sets are both open and closed.
These two sets are said to be clopen.
For the other statements we use the De Morgan laws on set calculus, therefore we have
n
\
x∈ Ai =⇒ x ∈ Ai
i=1
∴ ∃i : BXi (x) ⊂ Ai

Taking  = mini∈I i we have

n
\
BX (x) ⊂ BX (x) =⇒ BX (x) ⊂ Ai = A
i=1

And A is open
If we let C = Ac we have that
n
!c n
\ [
c
C=A = Ai = Aci
i=1 i=1
∴ C is closed
For the last two we proceed as follows

x ∈ Aα =⇒ ∃α0 ∈ I : x ∈ Aα0
[
∴ ∃ > 0 : BX (x) ⊂ Aα0 ⊂ =B
α∈I

For the last one, we use the De Morgan laws and the proposition is demonstrated

Definition 2.1.6 (Internal Points, Closure, Border). Let (X, d) be a metric space and A ⊂ X a subset.
We define the following sets from A

1. A◦ = α∈I Gα is the set of internal points of A, where I is an index set and Gα ⊂ A are open
S

T
2. A = β∈J Fβ is the closure of A, where J is another index set and Fβ ⊂ A are closed
c
3. ∂A = A \ A◦ = A ∪ (A◦ ) is the border of A

Proposition 3. 1. A is an open set iff A = A◦

2. A is closed iff A = A
c
3. A◦ = (A◦ )
◦ c
4. A = (Ac )

CHAPTER 2. ABSTRACT SPACES 18

◦
5. (A ∩ B) = A◦ ∩ B ◦

6. A ∩ B = A ∪ B

Proof. Let O(A) be a collection of open sets, such that ∀G ∈ O(A) =⇒ G ⊂ A, then
[
A = A◦ =⇒ A = G
G∈O(A)

Therefore, being a union of a finite number of open sets, A is open.

For the same reason as before and the previous proposition, we have that A is closed
For the third proposition, we have
!c

c \ [ [
Ac = F = Fc = G = A◦
Ac ⊂F Ac ⊂F G∈O(A)

The others are easily demonstrated throw this process, iteratively

Proposition 4. Let (X, d) be a metric space, and A ⊂ X, x ∈ X

1. x ∈ A ⇐⇒ ∃ > 0 : B (x) ⊂ A

2. x ∈ A ⇐⇒ ∀ > 0 B (x) ∩ A 6= {}

3. x ∈ ∂A ⇐⇒ ∀ > 0 B (x) ∩ A 6= {} ∧ B (x) ∩ A 6= {}

Proof. 1 Let I(A) := { x ∈ X| ∃ > 0 : B (x) ⊂ A}

[
x ∈ I(A) =⇒ ∃ > 0 : B (x) ⊂ A, ∴ x ∈ G
G⊂A

But
x ∈ A◦ =⇒ ∃G ⊂ X open : x ∈ G =⇒ ∃ > 0 : B (x) ⊂ G ⊂ A
∴ A◦ ⊂ I(A) 3 x, I(A) ⊂ A by definition, ∴ I(A) = A◦

2 For the second proposition, we have

◦ c ◦
A = (Ac ) =⇒ x ∈ A ⇐⇒ x ∈ (Ac ) =⇒ ∀ > 0 B (x) 6⊂ Ac


∴ ∀ > 0 B (x) ∩ A 6= {}

3 For the last one, we have, taking into account the first two proofs

x ∈ ∂A ⇐⇒ x ∈ A \ A◦ =⇒ x ∈ A ∧ x ∈
/ A◦
1 ∧ 2 =⇒ x ∈ A ⇐⇒ ∀ > 0 B (x) ∩ A 6= {}
/ A◦ ⇐⇒ ∀ > 0 B (x) ∩ Ac 6= {}
∴x∈
2.2. CONVERGENCE AND COMPACTNESS 19

Definition 2.1.7 (Isometry). Let (X, d), (Y, ρ) be two metric spaces and f an application, defined as
follows
f : (X, d) → (Y, d)
f is said to be an isometry iff

∀x1 , x2 ∈ X, ρ(f (x1 ), f (x2 )) = d(x1 , x2 )

Remark. If f is an isometry, then f is injective, but it’s not necessarily surjective

Example 2.1.1. Let X = [0, 1] and Y = [0, 2], therefore

f :[0, 1] → [0, 2]
x → f (x) = x

f is obviously an isometry, since, for x, y ∈ [0, 1]

d(f (x), f (y)) = d(x, y)

But it’s obviously not surjective.

Definition 2.1.8 (Diameter of a Set). Let A be a set and the couple (A, d) be a metric space. We define
the diameter of A as follows
diam (A) = sup (d(x, y))
x,y∈A

§ 2.2 Convergence and Compactness

Definition 2.2.1 (Convergence). Let (X, d) be a metric space and x ∈ X. A sequence (xk )k≥0 in X is
said to converge in X and it’s indicated as xk → x ∈ X, iff

∀ > 0 ∃N > 0 : ∀k ≥ N, d(xk , x) <  ∴ lim xk = x

k→∞

Theorem 2.1 (Unicity of the Limit). Let (X, d) be a metric space and (xk )k≥0 a sequence in X. If
xk → x ∧ xk → y, then x = y

Definition 2.2.2 (Adherent point). Let (X, d) be a metric space and A ⊂ X. x ∈ X is said to be an
adherent point of A if ∃(xk )k≥0 ∈ A : xk → x ∈ X. The set of all adherent points of A is called
ad(A)

Definition 2.2.3 (Accumulation point). Let (X, d) be a metric space and A ⊂ X. x ∈ X is an

accumulation point of A, or also limit point of A if ∃(xk )k≥0 : xk 6= x ∧ xk → x ∈ ad(A)

Proposition 5. Let (X, d) be a metric space and A ⊂ X, then A = ad(A)

CHAPTER 2. ABSTRACT SPACES 20

Proof. Let Y = ad(A), then

x ∈ A =⇒ ∀ > 0 B (x) ∩ A 6= {}
∴ ∀n ∈ N B n1 (x) ∩ A 6= {} =⇒ ∀n ∈ N ∃xn ∈ B n1 (x)

But d(x, xn ) < n−1 , therefore x ∈ Y =⇒ x ∈ ad(A), and by definition

∃(xn )n≥0 : ∀ > 0 ∃N ∈ N : ∀k ≥ N d(xk , x) <  =⇒ xN ∈ B (x) ∴ xN ∈ A

∴ ∀ > 0 xN ∈ B (x) ∩ A 6= {} =⇒ x ∈ A =⇒ Y ⊂ A, ∴ Y = ad(A) = A

Proposition 6. Let (X, d) be a metric space and A ⊂ X. Then A is closed iff ∃(xk )k≥0 ∈ A : xk →
x ∈ A =⇒ ad(A) ⊂ A
Definition 2.2.4 (Dense Set). Let (X, d) be a metric space and A, B ⊂ X. A is said to be dense in
B iff B ⊂ A, therefore ∀ > 0 ∃y ∈ A : d(x, y) < . One example for this is Q ⊂ R, with the usual
euclidean distance defined through the modulus.
Definition 2.2.5. Let (X, d) be a metric space and (xk )k≥0 ∈ X. The sequence xk is said to be a
Cauchy sequence iff
∀ > 0 ∃N > 0 : ∀k, n ≥ N d(xk , xn ) < 

Proposition 7. Let (X, d) be a metric space and (xk )k≥0 ∈ X a sequence. Then, if xk → x, xk is a
Cauchy sequence
Definition 2.2.6 (Complete Space). Let (X, d) be a metric space. (X, d) is said to be complete iff
∀(xk )k≥0 ∈ X Cauchy sequences, we have xk → x ∈ X
Theorem 2.2 (Completeness). Let (X, d) be a metric space and Y ⊂ X. (Y, d) is complete iff
Y = Y in X
Proof. Let (Y, d) be a complete space, then

(xk ) ∈ Y Cauchy sequence =⇒ ∃y ∈ Y : xk → y

Let z ∈ ad(A) and ηk a subsequence of xk , then

∃(ηk ) ∈ Y : ηk → z =⇒ ∃y ∈ Y : ηk → y ∴ z = y =⇒ ad(Y ) ⊂ Y

Going the opposite way we have that ad(Y ) = Y and therefore Y = Y

Definition 2.2.7 (Compact Space). A metric space (X, d) is said to be compact or sequentially
compact if
∀(xk ) ∈ X xk → x ∈ X, ∃{(xkn )} : xkn → x ∈ X

Theorem 2.3 (Heine-Borel). Let K ⊂ Rn , then K is compact if and only if K is closed and
bounded
2.2. CONVERGENCE AND COMPACTNESS 21

<++>
Theorem 2.4. Let (X, d) be a compact space. Then (X, d) is also complete
Proof. (X, d) is compact, therefore

∀(xk ) ∈ X Cauchy sequence =⇒ xk → x ∈ X

Taken (xnk )k ∈ X a subsequence, we have

xk → x =⇒ xnk → x ∈ X

Definition 2.2.8 (Completely Bounded). Let (X, d) be a metric space. X is totally bounded iff
n
[
∃Y ⊂ X : ∀ > 0, ∀x ∈ Y X = B (x)
i=1

Definition 2.2.9 (Poligonal Chain). Let z, w ∈ C. We define a polygonal [z, w] as follows

[z, w] := { z, w ∈ C| z + t(w − z), t ∈ [0, 1] ⊂ R}

A polygonal chain will be indicated as follows Pz,w and it’s defined as follows
n−1
[
Pz,w = [zk , zk+1 ] = [z, z1 , · · · , zn−1 , w]
k=1

It can also be defined analoguously

√ for every metric space (X, d) 6= (C, k·k), where k·k : C → R is the
p
usual complex norm kzk = zz = Re(z)2 + Im(z)2
Definition 2.2.10 (Connected Space). Let (G, d) be a metric space, G is connected if

∀z, w ∈ G ∃Pz,w ⊂ G

Definition 2.2.11 (Contraction Mapping). Let (X, d) be a complete metric space. Let T : X −→ X. T
is said to be a contraction mapping or contractor if

∀x, y ∈ X ∃q ∈ [0, 1) : d(T (x), T (y)) ≤ qd(x, y) (2.6)

Note that a contractor is necessarily continuous.

Theorem 2.5 (Banach Fixed Point). Let (X, d) be a complete metric space, with X 6= {} and
equipped with a contractor T : X −→ X. Then

∃!x? ∈ X : T (x? ) = x? (2.7)

CHAPTER 2. ABSTRACT SPACES 22

Proof. Take x0 ∈ X and a sequence xn : N −→ X, where

xn = T (xn−1 ), ∀n ∈ N

It’s obvious that

d(xn+1 , xn ) = d(T (xn ), T (xn−1 )) ≤ qd(xn , xn−1 ) ≤ q n d(x1 , x0 )

We need to prove that xn is a Cauchy sequence. Let m, n ∈ N : m > n, then

d(xm , xn ) ≤ d(xm , xm−1 ) + · · · + d(xn+1 , xn ) ≤ q m−1 d(x1 , x0 ) + · · · + q n d(x1 , x0 )

Regrouping, we have
m−n−1 ∞  
n
X
k n
X
k n 1
d(xm , xn ) ≤ q d(x1 , x0 ) q ≤ q d(x1 , x0 ) q = q d(x1 , x0 )
1−q
k=0 k=0

By definition of convergence, we have then

∀ > 0, ∃N ∈ N : ∀n > N d(sn , s) < 

Then
q n d(x1 , x0 ) (1 − q)
<  =⇒ q n < , ∀n > N
1−q d(x1 , x0 )
Therefore, after taking m > n > N , we have

d(xm , xn ) < 

Therefore xn is a Cauchy sequence. Since (X, d) is a complete metric space, this sequence must have a
limit xn → x? ∈ X, but, by definition of convergence and limit, we have that by continuity
 
x? = lim xn = lim T (xn−1 ) = T lim xn−1 = T (x? )
n→∞ n→∞ n→∞

This point is unique. Take y ? ∈ X such that T (y ? ) = y ? 6= x? , then

0 < d(T (x? ), T (y ? )) = d(x? , y ? ) > qd(x? , y ? )

Therefore
∃!x? ∈ X : T (x? ) = x?
And x? is the fixed point of the contractor T

§ 2.3 Vector Spaces

Definition 2.3.1 (Vector Space). A vector space V over a field F is a set, where V 6= {} and it satisfies
the following properties, ∀u, v, w ∈ V and a, b ∈ F

1. u + v ∈ V sum closure
2.3. VECTOR SPACES 23

2. av ∈ V scalar closure

3. u + v = v + u

4. (u + v) + w = u + (v + w)

5. ∃!0 ∈ V : u + 0 = 0 + u = u

6. ∃!v ∈ V : u + v = 0 =⇒ v = −u

7. ∃!1 ∈ V : 1 · u = u

8. (ab)u = a(bu) = b(au) = abu

9. (a + b)u = au + bu

10. a(u + v) = au + av

Definition 2.3.2 (Norm). Let V be a vector space over a field F, then the norm is an application defined
as follows
k·k : V −→ F
Where it satisfies the following properties

1. kuk ≥ 0 ∀u ∈ V

2. kuk = 0 ⇐⇒ u = 0

3. kcuk = |c|kuk ∀u ∈ V c ∈ F

4. ku + vk ≤ kuk + kvk ∀u, v ∈ V

Definition 2.3.3 (Normed Vector Space). A normed vector space is defined as a couple (V, k·k),
where V is a vector space over a field F.

Proposition 8. A normed vector space (NVS), is also a metric vector space (MVS) if we define our
distance as follows
d(u, v) = ku − vk ∀u, v ∈ V

Definition 2.3.4 (Vector Subspace). Let V be a vector space and U ⊂ V. U is a vector subspace of V
iff

1. u, v ∈ U =⇒ u + v ∈ U

2. u ∈ U , a ∈ F =⇒ au ∈ U

Proposition 9. If (V, k·k) is an normed vector space and W ⊂ V is a subspace of V, then (W, k·k) is a
normed vector space
CHAPTER 2. ABSTRACT SPACES 24

Definition 2.3.5 (p-norm). Let (V, k·kp ) be a normed vector space. The norm k·kp is said to be a
p-norm if it’s defined as follows
  p1
dim(V)
X
kvkp :=  (vi )p  , ∀v ∈ V, ∀p ∈ N? := N ∪ {±∞} (2.8)
i=1

Setting p = ∞ we have that

kvk∞ = max |vi | (2.9)
i≤dim(V)

Definition 2.3.6 (Dual Space). Let V be a vector space over the field F, we define a linear functional
as an application ϕ : V −→ F such that ∀u, v ∈ V and c ∈ F
ϕ(u + v) = γ(u) + ϕ(v)
(2.10)
ϕ(λu) = λϕ(u)

Defining the sum of two linear functionals as (ϕ1 + ϕ2 )(v) = ϕ1 (v) + ϕ2 (v) we immediately see that
the set of all linear functionals forms a vector space over V, which will be called the dual space V ? .

§§ 2.3.1 Hölder and Minkowski Inequalities

Having defined p-norms, we can prove two inequalities that work with these norms, the Minkowski
inequality and the Hölder Inequality
Theorem 2.6 (Hölder Inequality). Let pq ∈ N? , where
1 1
+ =1
p q
Then
n
X
∀x, y ∈ Rn kxkp kykq ≥ |xk yk | (2.11)
k=1

Proof. Taking p = 1, we have q = ∞, and the demonstration is obvious

n
X n
X
kxkp kykq = kxk1 kpk∞ = max |yk | |xk | ≥ |xk yk |
k≤n
k=1 k=1

Else, if p > 1, we ave that

ap bq
ab ≤ + ∀a, b ≥ 0
p q
Let
x y
s= , t=
kxkp kykq
We have
n n n n
X p 1 X p
X q 1 X p
ksk = p |x k | = 1 = |t| = q |y|
kxkp kykq
k=1 k=1 k=1 k=1
2.3. VECTOR SPACES 25

Therefore
n n n
X 1X p 1X q
|sk tk | ≤ |sk | + |tk |
p q
k=1 k=1 k=1

Substituting again the definitions of s, t we have

n
X n
X
|yk xk | = kxkp kykq |sk tk | ≤ kxkp kykq
i=1 k=1

Theorem 2.7 (Minkowski Inequality). Let p ≥ 1, therefore ∀x, y ∈ Rn we have

kx + ykp ≤ kxkp + kykp (2.12)

Proof. We begin by writing explicitly the p-norm

n
X n
X
p p p−1
kx + ykp = (|xk | + |yk |) = (|xk | + |y|k ) (|xk | + |yk |)
k=1 k=1

p−1
Letting uk = (|xk | + |yk |) we have, after imposing the condition on q of the p-norm as q(p + 1) = p
and using that the sum is Abelian, we have
 ! q1
 Xn Xn

|xk |uk ≤ kxkp kukq = kxkp (|xk | + |yk |)p





k=1 k=1
X n n
! q1
 X
 p



 |yk |uk ≤ kykp kukq = kykp (|xk | + |yk |)
k=1 k=1

Therefore, summing and imposing that 1 − q −1 = p we have that

kx + ykp ≤ kxkp + kykq

CHAPTER 2. ABSTRACT SPACES 26
3 Differential Analysis

§ 3.1 Digression on the Notation Used

In this chapter (and from now on, mostly), we will use a notation which is called abstract index
notation with the Einstein summation convention. This is usually abbreviated in common literature
as the Einstein index notation. We will give here a brief explanation of how this notation actually
works, and why it’s so useful in shortening mathematical expressions. Let V be a vector space and V ?
be the dual space associated with V. Then we can write the elements v ∈ V, ϕ ∈ V ? with respect to
some basis as follows  
v1
 .. 
v = (v1 , v2 , · · · , vn ) =  . 
(3.1)
vn


ϕ = (ϕ1 , ϕ2 , · · · , ϕn ) = ϕ1 ϕ2 · · · ϕn

The first notation is the ordered tuple notation, meanwhile the second notation is the usual column/row
notation for vectors utilized in linear algebra. In Einstein notation we will have that

v −→ v i
(3.2)
ϕ −→ ϕi

Where the vector in the space will be indicated with a raised index (index, not power!) and the
covector with a lower index, where the index will span all the values i = 1, · · · , dim(V).
Let’s represent the scalar product in Einstein notation. Let’s say that we want to write the scalar product
hv, vi
dim(V)
X
hv, vi = vi vi −→ vi v i (3.3)
i=1

Note how we have omitted the sum over the repeated index. Now one might ask why it’s not written
as vi vi (or v i v i , since v ∈ V), and this is easily explained introducing the matrix gij , which is the matrix
of the scalar product.
Applying this matrix to v i we have gij v i . Note how the low index j is free and i is being summed over,
hence is a dummy index, this means that the result must have a lower index j for consistency. So we
can write vj = gij v i , and due to the lower index we already know that this is a covector, i.e. a linear

27
CHAPTER 3. DIFFERENTIAL ANALYSIS 28

functional V −→ F, hence it will “eat” a vector and “spew” a scalar (with no indices!). Feeding to this
covector the vector v j we have finally

hv, vi = vj v j = gij v i v j (3.4)

Where, algebraically we have “omitted” the definition of ι(·) = hv, ·i, which is the canonical isomor-
phism between V and V ? .
With this definition we have defined what mathematically are called musical isomorphisms, appli-
cations which raise and lower indexes. Ironically, this operation is called index gymnastics, since
we’re raising and lowering indices. Thanks to these conventions operations with matrices (and ten-
sors) become much much easier. Let aij and bij be two n × n matrices over the ordered field F. The
multiplication of these two matrices will simply be

cij = aik bkj (3.5)

Note how the k index gets “eaten”. This mathematical cannibalism is called contraction of the index
k. So, the trace for a matrix aij will be
tr(a) = aii (3.6)
And now comes the tricky part. In order to write determinants we need to define a symbol, the so
called Levi-Civita symbol, i1 ...in . In three dimensions it’s ijk , and follows the following rules

1
 even permutation of the indices
ijk = −1 uneven permutation of the indices (3.7)

0 i=j∨j =k∨k =i


In n dimensions, it becomes

1
 even permutation of indices
i1 ...ik = −1 uneven permutation of indices (3.8)

0 ii = ij for some i, j


It’s obvious by definition that this weird entity is completely antysimmetrical and unitary, and there-
fore it’s perfect for representing permutations (it’s also known as permutation symbol for a reason).
Therefore, remembering the definition of the determinant we can write, for an n × n matrix aij

det(a) = i1 ...in a1i1 a2i2 · · · anin = i1 ...in g ji1 a1j g ki2 a2k · · · g lin anin (3.9)

If dim(V) = 3, we can therefore immediately define the cross product of two vectors as follows

c = v × w −→ ci = g ij jkl v k wl (3.10)

(Note how we had to raise the index i).

From now on, we will start to use Greek letters for indices and Latin letters for labels, in order to avoid
confusions, simply look again at the formula for the determinant, it’s much clearer this way. In fact,
letting µ, ν, · · · be our indices and i, j, k, · · · our labels, we can write, for a matrix Aµν

det(A) = A = µ1 ...µn g νµ1 g σµ2 · · · g ζµn A1ν A2σ · · · Anζ (3.11)
3.1. DIGRESSION ON THE NOTATION USED 29

See? Much clearer, at least in my opinion.

Now we might want to understand how to write norms with this notation. For the usual Euclidean
norm it’s quite easy. So we can easily write
p p
kvk = hv, vi −→ vµ v µ (3.12)

In case we have a vector function f µ (xν ), the following notation will be used
q
kf(x)k −→ fµ f µ (xν ) (3.13)

Or, for the sum of two functions g µ (xν ) ± f µ (y ν )

q
kg(x) ± f(y)k −→ gµ g µ (xν ) + fµ f µ (y ν ) ± 2gµ (xν )f µ (y ν ) (3.14)

Or q
kg(x) ± f(y)k −→= gµν (g µ (xγ ) ± f µ (y γ ))(g ν (xγ ) ± f ν (y γ ))

A shorthand notation can be created by directly using the norm symbol, but with the contracted index
in the upper or lower position as follows

kg(x) ± f(y)k −→ kg µ (xν ) ± f µ (y ν )kµ (3.15)

For p-norms we have to watch out for a little detail. We have to add a square root in order to “fix” the
squaring of every element. So we get
q
p p
 p p
 p1
kvkp −→ p (vµ ) 2 (v µ ) 2 = (vµ ) 2 (v µ ) 2 (3.16)

1/p
Theorem 3.1. (vµ v µ )p/2 is wrong

Proof. It’s easy to see why it doesn’t work by expanding the sum on µ

p
(vµ v µ )p = v1 v 1 + v2 v 2 + · · · + vn v n
p p (3.17)
(vµ ) 2 (v µ ) 2 = (v1 v 1 )p + (v2 v 2 )p + · · · + (vn v n )p

Moreover, it’s time to bring down some formal rules for the usage of this notation

Theorem 3.2 (Rules for Index Calculus in Einstein Notation). 1. Free indices must be consistent
in both sides of the equation. I.e. aµν bµγδ = cνγδ . aµν bµγδ 6= cγνδ , aµν bµγδ 6= cνγσ

2. An index can be repeated only two times per factor and must be contracted diagonally. I.e.
aµ bµ fγδ = cδγ is defined correctly, aµ bµ , aµ bµ or aµ bµ fγµ are ill defined

3. Dummy indices can be replaced at will, since they don’t contribute to the “index
equation”
CHAPTER 3. DIFFERENTIAL ANALYSIS 30

§§ 3.1.1 Differential Operators

Differential operators will be defined formally in the next sections, but for now we will simply explain
how they actually work with this notation (and what are the advantages of such), alongside the usual
boldface notation.
We will begin by defining the derivative along the coordinate vectors (usually indicated with xµ ). We
will use the differential operator del (∂).
This operator will be used as follows
1. If there is no ambiguity for the coordinate system, the derivative alongside the coordinates xµ
will be indicated as ∂µ
2. In case of ambiguity, something will be added in order to distinguish the operators. I.e. let
(xµ , y ν ) be our coordinate system, then we will have ∂xµ or ∂yν
3. In every single case, even the differential operator must follow the index calculus rules
Now let f (xµ ) be some (scalar, there are no free indices) function of the coordinates. The derivative
(or gradient, it will soon be defined properly) can be written in various ways. In boldface notation it’s
usual to indicate this as ∇f , which can be translated as follows
∂f
∇f −→ ∂µ f = = f,µ (3.18)
∂xµ
Note how in the RHS it’s obvious that this quantity must be a vector due to the free index. The last one
is the comma notation for derivation, used for compacting (even more) the notation (Also check how
in the second notation, even if the index is raised, it behaves as a lower index, we will check deeply this
part in the section on differential forms).
Now comes the fun part. Higher order derivatives.
For the same function, we can define the Hessian matrix (the matrix of second derivatives) Hf, simply
applying two times the ∂ operator

2 ∂2f
Hf −→ ∂ν ∂µ f = ∂µν f = ∂µν f = = f,µν (3.19)
∂xµ ∂xν
Derivatives of order > 2 can then be defined recursively.
Now we might ask, what if we have a vector field F µ ? Nothing changes. We simply have to remember
to not repeat indices in order not to represent scalar products.
We have JF as the Jacobian matrix of F µ , basically the derivative matrix which in Einstein notation, as
before, has a quite obvious nature
∂F µ
JF −→ ∂ν F µ = = F,νµ (3.20)
∂xν
And so on, and so on…1
Let’s now define the divergence and curl operators. Take now a vector field g µ . We then have
∂g µ
∇ · g −→ gµν g µδ ∂δ g ν = ∂µ g µ = µ
= g,µ
∂xµ
(3.21)
∂g σ
∇ × g −→ µνσ g νδ ∂δ g σ = µνσ ∂ ν g σ = µνσ = µνσ g σ,ν
∂xµ
1 It’s quite fun to dive into the dumpster of Einstein notation, isn’t it?
3.1. DIGRESSION ON THE NOTATION USED 31

And therefore, defining the Laplacian as ∇2 = ∇ · ∇, we will simply have, for whatever function h

∂2h
∇2 h −→ g µν ∂ν ∂µ h = ∂ µ ∂µ h = = h,µ,µ (3.22)
∂xµ ∂xµ

Note how the operator ∂ µ appears. This can be seen as a derivation along the covector basis (xµ =
gµν xν ).
Now, we can go back to our mathematical rigor.
CHAPTER 3. DIFFERENTIAL ANALYSIS 32

§ 3.2 Curves in Rn
Definition 3.2.1 (Scalar Field). A scalar field is a function f : A ⊆ Rn −→ R where A is an open set
Definition 3.2.2 (Vector Field). A vector field is a function f µ : A ⊂ Rn −→ Rm where A is an open
set
Definition 3.2.3 (Continuity). A scalar field f : A −→ R is said to be continuous in a point pµ ∈ A if

∀ > 0 ∃δp : kxµ − pµ k < δ =⇒ |f (xµ ) − f (pµ )| <  (3.23)

A vector field f µ : A → Rm is said to be continuous instead if

∀ > 0 ∃δp : kxν − pν kν < δ =⇒ kf µ (xν ) − f µ (pν )kµ <  (3.24)

If this function is continuous ∀pµ ∈ A, then the vector field is said to be part of the space C(A), with
A ⊆ Rn
Definition 3.2.4 (Canonical Scalar Product). Let xµ , y µ ∈ Rn , the canonical scalar product is a
bilinear application h·, ·i : Rn × Rn −→ R where, if the components of the two vectors are defined as
xµ , y µ , is defined as
n
X
hx, yi = xi y i −→ xµ y µ (3.25)
i=1

It’s easy to see that the canonical scalar product induces the euclidean norm as follows
p p
kvk = kvk2 = hv, vi = vµ v µ (3.26)

Definition 3.2.5 (Curves in Rn ). A curve is an application ϕ : [a, b] ⊂ R → Rn .

The function ϕµ (t) = pµ , with t ∈ [a, b] is called the parametric representation of the curve.
Remembering how indexes work in this notation, we already know that this application can be
represented with an ordered n−tuple or a vector in Rn
Definition 3.2.6 (Regular Curves). A curve ϕµ (t) is said to be continuous if all its components are
continuous. A curve is said to be regular iff

ϕµ (t) ∈ C 1 ([a, b])

q (3.27)
ϕµ (t)ϕµ (t) 6= 0 t ∈ (a, b)

A curve is said to be piecewise regular if it’s not regular in [a, b] but it’s regular in a finite number of
subsets [an , bn ] ⊂ [a, b]
Definition 3.2.7 (Homotopy of Curves). Let γ µ , η µ be two curves from a set [a, b], [c, d] respectively.
These two curves are said to be homotopic to one another, and it’s indicated as γ µ ∼ η µ iff
∼
∃h : [c, d] −→ [a, b], h ∈ C([c, d]), h−1 ∈ C([a, b]), h(s) > h(t) for s > t : η µ = γ µ ◦ h (3.28)
3.2. CURVES IN RN 33

Definition 3.2.8 (Tangent Vector). The tangent vector of a regular curve is defined as the following
vector.
γ̇ µ
T µ (t) = p (3.29)
(γ˙µ γ̇ µ )
Where with γ̇ µ (t) we indicate the derivative of γ µ with respect to the only variable t.

Definition 3.2.9 (Tangent Line). A curve γ µ : [a, b] −→ Rn is said to have tangent line at a point
t0 ∈ [a, b] if it’s regular, therefore the line will have parametric equations

pµ (t) = γ µ (t0 ) + γ̇ µ (t0 )(t − t0 ) (3.30)

Definition 3.2.10 (Length of a Curve). The length of a curve γ : [a, b] −→ Rn is defined as follows
ˆ b q
Lγ := γ̇µ (t)γ̇ µ (t) dt (3.31)
a

Remark. In R2 , if a curve is defined in polar coordinates, it will appear as follows

ρ = ρ(θ), θ ∈ [θ0 , θ1 ] (3.32)

Its length will be given from the following integral

ˆ θ1 p
Lρ := (ρ0 (θ))2 + (ρ(θ))2 dθ (3.33)
θ0

The graph of a function f : [a, b] −→ R, f ∈ C 1 (a, b) can also be parametrized from a curve ϕµ (t),
where
ϕµ (t) → (t, f (t)) (3.34)
Its length will be then calculated with the following integral
ˆ b q
Lϕ := 1 + (ḟ (x))2 dx (3.35)
a

Theorem 3.3 (Length Invariance under Homotopy of Curves). Let {γ1 }, {γ2 } ⊂ Rn be two curves,
such that γ1 ∼ γ2 , then
Lγ1 = Lγ2

Proof. By definition we have that since the two curves are homotopic, the two domains of definition
of both are diffeomorphic to each other, and therefore

γ1µ (t) = γ2µ (ϕ(t)) (3.36)

CHAPTER 3. DIFFERENTIAL ANALYSIS 34

∼
Where ϕ : I −→ J is our C 1 diffeomorphism. Therefore
ˆ ˆ ˆ
Lγ1 = kγ̇1µ kµ dt = ˙
γ2µ (ϕ(t)) dt = k(γ2µ )0 kµ |ϕ̇(t)|dt
I I µ I

Then, changing variables t → ϕ(t) = s we get

ˆ ˆ
µ 0 −1
k(γ2µ )0 kµ ds = Lγ2

−1 0
Lγ1 = k(γ2 ) kµ ϕ̇ ϕ (s) (ϕ (s)) ds =
ϕ(I) J

Definition 3.2.11 (Curviline Coordinate). Let ϕ : [a, b] −→ Rn , we can define a function s(t) as follows
ˆ tq
s(t) = ϕ̇µ (τ )ϕ̇µ (τ ) dτ (3.37)
a

Then q
ds = ϕ̇µ (t)ϕ̇µ (t) dt (3.38)
And the length of a curve can also be indicated as follows
ˆ
Lϕ = ds (3.39)
ϕ

Definition 3.2.12 (Curvature, Normal Vector). The curvature of a curve is defined as follows
q q
κ(s) = Tµ (s)T µ (s) = ϕ̈µ (s)ϕ̈µ (s) (3.40)

(Note that kϕ0 (s)k = 1) The normal vector is similarly defined as

Ṫ µ (s)
N µ (s) = (3.41)
κ(s)

Definition 3.2.13 (Simple Curve, Closed Curve). A simple curve is an injective application γ : [a, b] −→
Rn . A curve is said to be closed iff γ µ (a) = γ µ (b)
Theorem 3.4 (Jordan Curve). Let γ µ be a simple and closed curve in R2 or C (note that C ' R2 ),
then the set {γ}c is defined as follows
{γ}c = {γ}◦ ∪ extr ({γ}) (3.42)
Note that {γ} ⊂ R2 is the image of the application γ and extr ({γ}) is the set of points that
lay outside of the closed curve.
In C everything that was said about curves holds, however one must watch out for the definition of
modulus, for a curve γ µ ∈ C we will have
q
|γ̇(t)| = (Re0 (γ))2 + (Im0 (γ))2 = γ(t)γ(t)
p
(3.43)
3.3. DIFFERENTIABILITY IN RN 35

§ 3.3 Differentiability in Rn
Definition 3.3.1 (Directional Derivative). Let A ⊆ Rn be an open set, and f : A −→ R. The function is
said to be derivable with respect to the direction v µ ∈ A at a point pµ ∈ A, if the following limit is
finite
f (pµ + hv µ ) − f (pµ )
∂vµ f (pν ) = lim (3.44)
h→0 h
If v µ = xµ then this is called a partial derivative, and it will be indicated in the following ways

∂f
= ∂µ f = ∂xµ f (3.45)
∂xµ

Definition 3.3.2 (Differentiability). A scalar field f : A ⊆ Rn −→ R, with A open, is said to be

differentiable in a point pµ ∈ A if and only if there exists a linear application aµ (pν ) = aµ , such that
the following limit is finite
R(pµ + hµ )
lim p =0 (3.46)
hµ hµ
p
hµ hµ →0

Where we define the function R as follows

R(pµ + hµ ) = f (pµ + hµ ) − (f (pµ ) + aµ hµ ) (3.47)

This means that p 

f (pµ + hµ ) = f (pµ ) + aµ hµ + O hµ hµ (3.48)

Theorem 3.5 (Consequences of Differentiability). Let f : A ⊆ Rn −→ R be a differentiable scalar

field in every point of A, then

1. f ∈ C(A)

2. f is differentiable in A, and aµ = ∂µ , where is the vector differential operator, composed

by the partial derivatives

3. f has directional derivatives in A and the following equation holds

∂vµ f (pν ) = ∂µ f (pν )v µ

4. ∂µ f indicates the maximum and minimum growth of the function f

5. There exist a tangent hyperplane to the graphic of the function at the point (pµ , f (pµ )) ∈
Rn+1 and has the following equation

xn+1 = f (pµ ) + ∂µ f (pν )(xµ − pµ )

CHAPTER 3. DIFFERENTIAL ANALYSIS 36

Proof. 1. f differentiable in A implies f ∈ C(A)

p 
p lim f (pµ + hµ ) = p lim (f (pµ ) + aµ hµ + O hµ hµ ) = f (pµ )
hµ hµ →0 hµ hµ →0

2. f differentiable in A implies f derivable in A

f (pµ + hei ) − f (pµ ) ai h + O(h)

∂i f (pµ ) = lim = lim = ai ∈ R
h→0 h h→0 h
Then p 
R(pµ + hµ ) = f (pµ + hµ ) − f (pµ ) + ∂µ f (pν )hµ = O hµ hµ

3. ∂vµ f = ∂µ f v µ

f (pµ + hwµ ) − f (pµ ) h∂µ f (pν )v µ + O(h)

∂vµ f (pν ) = lim = lim = ∂µ f (pν )v µ
h→0 h h→0 h

4. ∂µ f indicates the direction of maximum growth.

For Cauchy-Schwartz, we have
p p p √
∂vµ f ∂ vµ f = ∂µ f v µ ∂ν f v ν ≤ ∂µ f ∂ µ f vν v ν

Theorem 3.6 (Total Differential). Let f : A ⊂ Rn −→ R, with A open. If f ∈ C 1 (A) (i.e. the
derivatives of f are continuous), then f is differentiable in A, the vice versa is also true

Proof. We can write the following equation

f (pµ + hµ ) = f (p1 + h1 , · · · , pn ) − f (p1 , · · · , pn ) + · · · + f (p1 , · · · , pn + hn ) − f (p1 , · · · , pn ) (3.49)

For Lagrange, we will have

f (pi + hi ) = hi ∂i f (p1 , · · · , q i , · · · , pn ) = hi ∂i f (ci ) (3.50)

Therefore
n
|f (pµ + hµ ) − f (aµ ) − ∂µ f hµ | X
µ hi
lim p ≤ lim |∂i f (c i ) − ∂i f (p )| p =0 (3.51)
hµ hµ h→0 hµ hµ
p
hµ hµ →0
i=1

Therefore ∂i f (pµ ) is continuous and the function is differentiable.

Theorem 3.7. Given f : A → R with f ∈ C 1 (A). Then ∂µ f = 0 ∀xµ ∈ A implies that f is

constant in A
3.3. DIFFERENTIABILITY IN RN 37

Proof. Define A as A1 ∪ A2 where

A1 = {xµ ∈ A : f (xµ ) = f (xµ0 )} . A2 = {xµ ∈ A : f (xµ ) 6= f (xµ0 )}

Since A is open, both A1 , A2 must be open. Define a path γ from xµ to xµ0 and define the composite
function
ϕ(t) = f (xµ0 + t(xµ − xµ0 )) , t ∈ [0, 1]
Since f is differentiable and ∂µ f = 0 by definition ϕ is constant and therefore continuous, which
confirms that A1 is open.
Since A1 ∩ A2
Theorem 3.8 (Differentiability of Vector Fields, Jacobian Matrix). Let f µ : A ⊆ Rn −→ Rm be a vector
field and A an open set, then the function f µ is differentiable iff exists a matrix Jνµ ∈ Mnm (R)
such that
kf µ (pν + hν ) − f µ (pν ) − Jνµ hν kµ
p lim p (3.52)
hµ hµ →0 hµ hµ
Or, equivalently p 
fµ (pν + hν ) = f µ (pν ) + Jνµ hν + O hµ hµ (3.53)

The then Jνµ is the matrix of partial derivatives of the vector field, called the Jacobian matrix
of the vector field f µ , and can be calculated as follows

Jνµ (pσ ) = ∂ν f µ (pσ ) (3.54)

Theorem 3.9 (Composite Derivation). Let f µ : A ⊆ Rn −→ Rk and g ν : B ⊆ Rk −→ Rp be

two differentiable functions in pσ ∈ A, f µ (pσ ) ∈ B and A, B open sets, then hν = g ν ◦ f µ is
differentiable, and
∂σ hν = ∂µ g ν (f µ̃ )∂σ f µ (3.55)
Since µ = 1, · · · , k, ν = 1, · · · , p, σ = 1, · · · , n it’s obvious that ∂σ hν ∈ Mp,n (R), ∂µ g ν ∈
Mp,k (R), ∂σ f µ ∈ Mk,n (R).
Proof. We have that (g ν ◦ f µ ) = g ν (f µ (pσ )). Then g ν is differentiable at f µ (pσ ) if
 
g ν (f µ (pσ +sσ )) = g ν (f µ (pσ ))+∂σ g ν ◦ f µ̃ (f µ (pσ + sσ ) − f µ (pσ ))+O kf µ (pσ + sσ ) − f µ (pσ )kµ

Since f µ is differentiable, we have

g ν (f µ (pσ + sσ )) = g ν (f µ (pσ )) + ∂σ g ν ◦ f µ̃ ∂σ f µ +


 
+ ∂σ (g ν ◦ f µ ) O sµ sµ + O kf µ (pσ + sσ ) − f µ (pσ )kµ
p

Then, we must prove that

 
∂σ (g ν ◦ f µ ) O kf µ (pσ + sσ ) − f µ (pσ )kµ
√ lim √ µ =0
sµ sµ →0 sµ s
CHAPTER 3. DIFFERENTIAL ANALYSIS 38

But √
∂σ (g ν ◦ f µ ) O( sµ sµ )
√ µ →0
sµ s
And
kf µ (pσ + sσ ) − f µ (pσ )kµ ≤
p p p
p
∂ σ fµ ∂σ f µ sµ sµ + O sµ sµ ≤ C sµ sµ
Therefore
   
O kf µ (pσ + sσ ) − f µ (pσ )kµ O kf µ (pσ + sσ ) − f µ (pσ )kµ kf µ (pσ + sσ ) − f µ (pσ )k
µ
√ µ = √ µ →0
sµ s kf µ (pσ + sσ ) − f µ (pσ )kµ sµ s

§ 3.4 Differentiability in C
Definition 3.4.1 (Differentiability). A function f : G ⊂ C −→ C with G open, is said to be differen-
tiable or derivable at a point a ∈ G if exists finite the following limit

df f (z) − f (a)
= f 0 (a) = lim (3.56)
dz a
z→a z−a

As usual, if this holds ∀a ∈ G, the function is derivable in G

Theorem 3.10. If f : G ⊂ C −→ C is derivable in a ∈ G, then f is continuous in a

Proof.
f (z) − f (a)
lim (f (z) − f (a)) = lim (z − a) lim =0
z→a z→a z→a z−a

Theorem 3.11 (Some Simple Rules). Let f, g : G ⊂ C −→ C

1. (f ± g)0 (z) = f 0 (z) ± g 0 (z)

2. (f g)0 (z) = f 0 (z)g(z) + f (z)g 0 (z)

3. (f /g)0 (z) = f 0 (z)/g(z) − f (z)g 0 (z)/g 2 (z) ∀z ∈ G : g(z) 6= 0

4. f (z) = c =⇒ f 0 (z) = 0

5. f (z) = z n =⇒ f 0 (z) = nz n−1

Theorem 3.12 (Composite Function Derivation). Let f : G ⊂ C −→ C and g : F ⊂ C −→ C, where

f (G) ⊂ F . If f is derivable in a ∈ G and g is derivable in f (a) ∈ F , then g ◦ f is derivable, and
its derivative is calculated as follows
d dg df
(g ◦ f ) = = g 0 (f (a))f 0 (a) (3.57)
dz a dz f (a) dz a
3.4. DIFFERENTIABILITY IN C 39

Proof. Since G is open, ∃Br (a) ⊂ G. Therefore, taking a sequence (z)n ∈ Br (a) : limn→∞ (z)n = a.
Letting f (zn ) 6= a, we can directly write in the definition of derivative
(g ◦ f )(zn ) − (g ◦ f )(a)
lim = (g ◦ f )0 (a) = g 0 (f (a))f 0 (a)
n→∞ zn − a
Thus, rewriting the function inside the limit
(g ◦ f )(zn ) − (g ◦ f )(a) (g ◦ f )(zn ) − (g ◦ f )(a) f (zn ) − f (a)
= →0
zn − a f (zn ) − f (a) zn − a
Since f is continuous in a ∈ G
∼
Theorem 3.13 (Inverse Function Derivation). Let f : G ⊂ C −→ C be a bijective continuous map,
with f −1 (w) = z. If f (a) 6= 0 and it’s derivable at that same point, we have
df −1 1
= (3.58)
dw f (a) f 0 (a)

Proof. Since f is bijective and continuous we can write

df −1 f −1 (w) − f −1 (f (a)) z−a 1
= lim = lim = 0
dw f (a) w→f (a) w − f (a) z→a f (z) − f (a) f (a)

§§ 3.4.1 Holomorphic Functions

Definition 3.4.2 (Holomorphic Function). A function f : G ⊂ C −→ C is said to be holomorphic in its
domain G if G is open, and
df
∀z ∈ G ∃ = f 0 (z) (3.59)
dz
It is indicated as f ∈ H(G). It’s easy to demonstrate that this set is a vector space.
Theorem 3.14 (Cauchy-Riemann Equation). Let f : G ⊂ C −→ C, where G is open and f ∈ H(G).
Then, if we write z = x + iy (
Re(f (z)) = u(x, y)
(3.60)
Im(f (z)) = v(x, y)
We have that the function is holomorphic if and only if
∂u ∂v


 − =0
∂x ∂y

(3.61)
 ∂v ∂u

 + =0
∂x ∂y
Alternatively, it can be written as follows
∂f ∂f
+i =0 (3.62)
∂x ∂y
CHAPTER 3. DIFFERENTIAL ANALYSIS 40

Definition 3.4.3 (Wirtinger Derivatives). Before demonstrating the previous theorem, we define the
Wirtinger derivatives as follows.
Let z ∈ C, z = x + iy and f : G ⊂ C −→ C.
 
∂f 1 ∂ ∂



 ∂z = ∂f (z) = − i f (z)
2 ∂x ∂y
  (3.63)
 ∂f 1 ∂ ∂

 = ∂f (z) = +i f (z)
∂z 2 ∂x ∂y
Then, the Cauchy-Riemann equations will be equivalent to the following equation
∂f
= ∂f (z) = 0 (3.64)
∂z

Proof. Let f (z) = u(x, y) + iv(x, y) : G ⊂ C −→ C be a differentiable function in a point z0 , then as

we defined, we have that f ∈ H(B (z0 )), and therefore
df f (z0 + h) − f (z0 )
= lim
dz z0 h→0 h

And, therefore, along the imaginary axis and the real axis, we have
f (z0 + Re(h)) − f (z0 ) ∂f
lim =
Re(h)→0 Re(h) ∂x z0
f (z0 + iIm(h)) − f (z0 ) 1 ∂f ∂f
lim = = −i
Im(h)→0 iIm(h) i ∂y z0 ∂y z0

Due to the continuity of the derivative (f ∈ H(B (z0 ))) we must have an equality between these limits
∂f ∂f ∂f ∂f
+i =2 = 0, ∴ f ∈ H(B (z0 )) =⇒ =0
∂x ∂y ∂z ∂z
But, since f (z) = u(x, y) + iv(x, y), we will have that
 
∂f ∂ 1 ∂ ∂
= (u(x, y) + iv(x, y)) = +i (u(x, y) + iv(x, y))
∂z ∂z 2 ∂x ∂y
 
1 ∂u ∂u ∂v ∂v
= +i +i − =0
2 ∂x ∂y ∂x ∂y
∂u ∂v ∂v ∂u
∴ − =i +i
∂x ∂y ∂x ∂y
Rewriting the previous equation in a system, we immediately get back the Cauchy-Riemann equations
∂u ∂v
− =0
∂x ∂x
∂v ∂u
+ =0
∂x ∂y
3.5. SURFACES 41

Definition 3.4.4 (Whole Function). A function f : C −→ C is said to be whole iff f ∈ H(C)

Definition 3.4.5 (Singular Point). Let f : G ⊂ C −→ C be function such that if D = B (z0 ) \ {z0 } and
f ∈ H(D), then z0 is said to be a singular point of f
For functions f : [a, b] ⊂ R −→ C every theorem already stated for curves in Rn with n = 2 holds,
since C ' R2 . The only thing that should be checked thoroughly is that
 
Re(f (t))
f (t) = ∈ C|R
Im(f (t))

Is written as
f (t) = Re(f (t)) + iIm(f (t)) ∈ C

§ 3.5 Surfaces

Definition 3.5.1 (Regular Surface). Let K ⊂ R2 , K = E where E is an open and connected subset. A
regular surface in R3 is an application

rµ : K −→ R3

Such that
1. rµ ∈ C 1 (K), i.e. ∃∂ν rµ ∈ C(K)
2. rµ is injective in K
3. rank (∂ν rµ ) = 2
The image Im(rµ ) = Σ ⊂ R2 is then defined by the following parametric equations
 1
x(u, v) = r (u, v)

rµ (u, v) = y(u, v) = r2 (u, v) (3.65)

z(u, v) = r3 (u, v)


The third condition can be rewritten as follows

µνσ ∂1 rν ∂2 rσ = µνσ ∂1 rν ∂2 rσ 6= 0 ∀(u, v) ∈ K ◦ (3.66)

Remark. A function f ∈ C 1 (K) defines automatically a surface with parametric equations rµ (u, v) =
(u, v, f (u, v)). This surface is always regular since µνσ ∂1 rν ∂2 rσ = (−2u, −2v, 1) 6= 0 ∀(u, v) ∈ K
Definition 3.5.2 (Coordinate Lines). The curves obtained fixing one of the two variables are called
coordinate lines in the surface Σ. We have therefore, for a parametric surface rµ (u, v) and two fixed
values ũ, ṽ ∈ I ⊂ R
xµ1 (t) = rµ (t, ṽ)
(3.67)
xµ2 (t) = rµ (ũ, t)
CHAPTER 3. DIFFERENTIAL ANALYSIS 42

Example 3.5.1. The sphere centered in a point pµ0 ∈ R3 , pµ0 = (x0 , y0 , z0 ) with radius R ≥ 0 has the
following parametric equations
x = x0 + R sin(u) cos(v)
y = y0 + R sin(u) sin(v) (3.68)
z = z0 + R cos(v)
With (u, v) ∈ [0, π] × [0, 2π]. It’s a regular surface, since
kµνσ ∂1 rν ∂2 rσ kµ = R2 sin(u) > 0 ∀(u, v) ∈ [0, π] × [0, 2π]

Definition 3.5.3 (Curve on a Surface). Let γ : [a, b] ⊂ R −→ K ⊂ R3 be a regular curve, and

r : K −→ Σ, with the following parametric equations
(
µ u = u(t)
γ (t) = (3.69)
v = v(t)

The regular curve pµ (t) = rµ (u(t), v(t)) has Im pµ ⊂ Σ. If it passes for a point pµ0 = (u0 , v0 ) it has
tangent line
pµ (t) = pµ0 + ṙµ (t)(t − t0 ) = pµ0 + ∂u rµ (u(t), v(t))u̇(t) + ∂v rµ (u(t), v(t))v̇(t) (3.70)
The line is contained inside the following plane
 
(x − x0 ) (y − y0 ) (z − z0 )
det  ∂1 r1 ∂1 r2 ∂3 r 1  (3.71)
1 2
∂2 r ∂2 r ∂3 r 3
For a cartesian surface, i.e. the surface generated from the graph of a function f (x, y), the tangent
plane will be
z = f (xµ0 ) + ∂µ f (xν0 )(xµ − xµ0 ) (3.72)

Definition 3.5.4 (Normal Vector). The normal vector to a surface Σ, nµ (u, v) is the vector
1
nµ (u, v) = q µνσ ∂u rν ∂v rσ (3.73)
µνσ µδγ ∂1 rν ∂2 rσ ∂1 rδ ∂v rγ

For a cartesian surface we have

 µ
−∂1 f
1
nµ (x, y) = p −∂2 f  (3.74)
1 + ∂µ f ∂ µ f 1

Definition 3.5.5 (Implicit Surface). Let F : A ⊂ R3 −→ R be a function such that F ∈ C 1 (A), letting
Σ := { xµ ∈ R3 F (xµ ) = 0}. If xν0 ∈ Σ and ∂µ F (xν0 ) 6= 0, Σ coincides locally to a cartesian surface,
and the equation of the tangent plane at the point xν0 is the following
∂µ F (xν0 )(xµ − xµ0 ) = 0 (3.75)
3.6. OPTIMIZATION 43

Definition 3.5.6 (Metric Tensor). Let ds be the curviline coordinate of some curve γ µ inside a regular
surface Σ. Then we have that
sµ (t) = rµ (u(t), v(t)) (3.76)
And therefore

2
2
ds2 = drµ drµ = ∂1 rµ ∂1 rµ dx1 + 2∂1 rµ ∂2 rµ dx1 dx2 + ∂2 rµ ∂2 rµ dx2 (3.77)

In compact form, we can write

ds2 = gµν dxµ dxν (3.78)
And, the metric tensor gµν can defined as follows

gµν = ∂µ rσ ∂ν rσ (3.79)

Or, in matrix notation  µ

∂1 rµ ∂2 rµ

∂1 r ∂1 rµ
gµν = (3.80)
∂2 rµ ∂1 rµ ∂2 rµ ∂2 rµ µν

In usual mathematical notation we have

 
E F
gµν = (3.81)
F G µν

And it’s called the first fundamental quadratic form in the language of differential geometry. Then,
we can write

2
2
ds2 = E dx1 + 2F dx1 dx2 + G dx2 (3.82)

§ 3.6 Optimization

§§ 3.6.1 Critical Points

Theorem 3.15 (Fermat). Let pν ∈ A be a point of local minimal or maximal for the function
f : A ⊆ Rn −→ R with f ∈ C 1 (A). If f is differentiable in pν we have

∂µ f (pν ) = 0 (3.83)

The point pν satisfying this condition is then called a stationary point or a critical point for
the function f

Proof. Let v µ ∈ A be a direction. The function g(t) = f (pµ + tv µ ) has a point of local maximal or
minimal for t = 0. Then

F 0 (0) = ∂vµ f (pν ) = ∂µ f (pν )v µ = 0 =⇒ ∂µ f (pν ) = 0 (3.84)

CHAPTER 3. DIFFERENTIAL ANALYSIS 44

Definition 3.6.1 (Hessian Matrix). Let f : A ⊆ Rn −→ R, and let f ∈ C 1 (A), then we define the
Hessian matrix as the matrix of the second partial derivatives of the function f
 
∂11 f · · · ∂1n f
∂µ ∂ν f (xγ ) = ∂µν f (xγ )  ... .. ..  (xγ ) (3.85)

. . 
∂n1 f ··· ∂nn f µν

Theorem 3.16 (Schwarz). Let f : A ⊆ Rn −→ R, f ∈ C 2 (A), then

∂µν f = ∂νµ f (3.86)

Definition 3.6.2 (Nature of Critical Points). Let pγ be a critical point for a function f ∈ C 1 (A).
Then
1. ∂µν f (pγ ) is definite positive, then pγ is a local minimum
2. ∂µν f (pγ ) is definite negative, then pγ is a local maximum
3. ∂µν f (pγ ) is indefinite, then pγ is a saddle point
Theorem 3.17. Here is a list of some rules in order to determine the definition of the matrix
∂µν f .
Let v µ ∈ A be a direction, and pγ ∈ A a critial point of the function f : A ⊂ Rn −→ R then
1. If ∂µν f (pγ )v µ v ν > 0 ∀v µ ∈ A =⇒ ∂µν f (pγ ) positive definite
2. If ∂µν f (pγ )v µ v ν < 0 ∀v µ ∈ A =⇒ ∂µν f (pγ ) negative definite
3. If ∂µν f (pγ )v µ v ν ≥ 0 ∀v µ ∈ A =⇒ ∂µν f (pγ ) semi-positive definite
4. If ∂µν f (pγ )v µ v ν ≤ 0 ∀v µ ∈ A =⇒ ∂µν f (pγ ) semi-negative definite
5. If v µ 6= wµ are two directions, and ∂µν f (pγ )v µ v ν > 0 ∧ ∂µν f (pγ )wµ wν < 0 =⇒ ∂µν f (pγ )
indefinite
Theorem 3.18 (Sylvester’s Criteria). Let Aµν ∈ Mnn (R), and (Ak )µν be the reduced matrix with
order k ≤ n, then
1. detµν ((Ak )µν ) > 0 =⇒ Aµν positive definite
2. (−1)k detµν ((Ak )µν ) > 0 =⇒ Aµν negative definite
3. If detµν ((A2k )µν ) < 0 or if detµν ((A2k+1 )µν ) < 0 ∧ detµν ((A2n+1 )µν ) > 0 for k 6= n, then Aµν is
indefinite
Theorem 3.19 (Compact Weierstrass). Let f : K ⊆ Rn −→ R, f ∈ C(K), with K a compact set,
then
∃pµ , q µ ∈ K : min(f ) = f (pµ ) ≤ f (xµ ) ≤ max(f ) = f (q µ ) ∀xµ ∈ K (3.87)
K K
3.6. OPTIMIZATION 45

Proof. Being K a compact set, we have that every sequence (pµ )n converges inside the set, therefore,
letting (pµ )n being a minimizing sequence for f . Then there exist a converging subsequence (pµ )nk
such that
f (pµnk ) → f (pµ )
But, since (pµ )n is a minimizing sequence, we have

f (pµ ) = min(f )
K

By definition of minimizing sequence. Analogously, one can define a maximizing sequence and obtain
the same result for the maximum of the function in K
Theorem 3.20 (Heine-Cantor). Given f : K → R with K a compact set, then if f ∈ C(K), f is
uniformly continuous
<++>
Theorem 3.21 (Existence of Intermediate Values). Let f : A ⊂ Rn −→ R, f ∈ C(A), then, given
u ∈ R such that
min(F ) ≤ u ≤ max(F )
A A

Then
∃cµ ∈ A : f (cµ ) = u

Theorem 3.22 (Squeeze). Let g(xµ ), f (xµ ), h(xµ ) be three functions such that g ≤ f ≤ h for
xµ ∈ Bδ (xµ0 ), then

lim g(xµ ) = lim µ h(xµ ) = L =⇒ lim f (xµ ) = L

xµ →xµ
0 xµ →x0 xµ →xµ
0

Proof. By definition of limit we can say, supposing that the affirmation is true, that

∀ > 0, ∃δ > 0 : |f (xµ ) − L| < , kxµ − xµ0 k < δ

∀ > 0, ∃δ > 0 : |g(xµ ) − L| < , kxµ − xµ0 k < δ1
∀ > 0, ∃δ > 0 : |h(xµ ) − L| < , kxµ − xµ0 k < δ2

Since we’re working in Rn with the standard topology, the last condition on the right tells us that
we’re working in Bδ (xµ0 ), for which g ≤ f ≤ h for some δ3 > 0.
Taken δ = min {δ1 , δ2 , δ3 } then

∀ > 0 L −  ≤ g(xµ ) ≤ f (xµ ) ≤ h(xµ ) ≤ L + , xµ ∈ Bδ (xµ0 )

Therefore
|f (xµ ) − L| ≤ 
Which proves the theorem
<++>
CHAPTER 3. DIFFERENTIAL ANALYSIS 46

Theorem 3.23 (Closed Weierstrass). Let f : L ⊆ Rn −→ R. If L = L and f ∈ C(L) is a coercitive

function, i.e.
lim f (xµ ) = +∞ (3.88)
kxk→∞

Then
∃xµ ∈ L : min(f ) = f (xµ ) (3.89)
L

Proof. µ
p Let (p )n be a minimizing sequence for f in L. If this sequence wasn’t limited, we would have
that (pµ )n (pµ )n → ∞, and therefore

inf(f ) = lim f (pµn ) = +∞

L n→∞

Therefore (pµ )n must be limited, and the proof is the same as in the case of a compact set.
Theorem 3.24 (Topology and Functions). Let f : Rn −→ R, f ∈ C(Rn ). Then

{ xµ ∈ Rn | f (xµ ) < a ∈ R}
(3.90)
{ xµ ∈ Rn | f (xµ ) > b ∈ R}

Are open sets in Rn with the standard topology, and

{ xµ ∈ Rn | f (xµ ) ≤ a ∈ R}
(3.91)
{ xµ ∈ Rn | f (xµ ) ≥ b ∈ R}

Are closed sets

§§ 3.6.2 Convexity and Implicit Functions

Definition 3.6.3 (Convex Set). A set A ⊂ Rn is said to be convex if

λxµ + (1 − λ)y µ ∈ A ∀xµ , y µ ∈ A, ∀λ ∈ [0, 1] (3.92)

Analogously, a function f : A ⊂ Rn −→ R is said to be convex, if

f (λxµ + (1 − λ)y µ ) ≤ λf (xµ ) + (1 − λ)f (y µ ) ∀xµ , y µ ∈ A, ∀λ ∈ [0, 1] (3.93)

The function f is also known as a sublinear function

Also, the set
Ef = { (xµ , λ) ∈ A × R| f (xµ ) ≤ y} (3.94)
Is convex
Theorem 3.25 (Convexity). Let f : A ⊂ Rn −→ R.
1. f convex in A =⇒ f ∈ C(A)
2. f differentiable in A =⇒ f convex ⇐⇒ f (xµ ) ≥ f (pµ ) + h∇f (pµ ), xµ − pµ i
3. f ∈ C 2 (A) =⇒ f convex ⇐⇒ ∂µν f (xγ ) is positive semidefinite
3.6. OPTIMIZATION 47

Definition 3.6.4 (Matrix Infinite Norm). Let Aµν (xγ ) ∈ V −→ Mmn (F), where dim(V) = n. We can
define a norm for this space as follows
√
r
(µ)
kAµν k∞ = m max sup Aν Aν(µ) (xγ ) (3.95)
µ xγ ∈V

Theorem 3.26 (Average Value). Let f µ : A ⊆ Rn −→ Rm , with f ∈ C 1 (A), A an open set and
K ⊂ A a compact convex subset, then

kf µ (xν ) − f µ (y ν )kµ ≤ k∂ν f µ k∞ kxν − y ν kν (3.96)

Proof. Let rν (t) = (1 − t)y ν + txν be a smooth parametrization of a segment connecting the two
points xν , y ν , then
2 2
kf µ (rν (1) − f µ (rν (0)kµ ≤ ∂ ν fµ ∂ν f µ (rν (t)) ≤ sup(∂ ν fµ ∂ν f µ (rγ ))kxν − y ν kν
rγ
 
2
≤ m max sup ∂ ν f(µ) ∂ν f (µ) (xγ ) kxν − y ν kν
µ γ

Therefore
√ r
kf µ (xν ) − f µ (y ν )kµ ≤ m max sup ∂ ν f(µ) ∂ν f (µ) kxν − y ν kν =


µ γ
µ ν ν ν ν
= k∂ν f k∞ kx − y kν ∀x , y ∈ K

Theorem 3.27 (Implicit Functions, Dini). Let f µ : A ⊆ Rm × Rn −→ Rn , where f µ ∈ C 1 (A). Also

let (xν0 , y0γ ) ∈ A such that
f µ (xν0 , y0γ ) = 0
 µ
∂f
det 6= 0
µν ∂y γ
Then
∃B (xν0 ) = I ⊂ Rm , B (y0γ ) = J ⊂ Rn : f µ (xν , y γ ) = 0 ∀(xν , y γ ) ∈ I × J
Has a unique solution y γ = g γ (xν ) ∈ J, with g γ ∈ C 1 (I), and
−1
∂g γ ∂f µ ∂f µ

=− (3.97)
∂xν ∂y γ ∂xν

 −1
Proof. Let Bµγ = ∂y0γ f µ , then we know that

f µ (xν , y γ ) = 0 ⇐⇒ Bµγ f µ (xν , y σ ) = 0 ⇐⇒ Gγ (xν , y σ ) = y γ − Bµγ f µ (xν , y σ ) = 0

CHAPTER 3. DIFFERENTIAL ANALYSIS 48

We have therefore
Gγ (xν , g σ (xν )) = g γ (xν ) − Bµγ f µ (xν , g σ (xν )) = g γ (xν ) ∀xν ∈ B r (xν0 ) = I
∂Gγ ∂f µ
σ
= δσγ − Bµγ σ
∂y ∂y
∂Gγ γ γ ∂f µ
= δ σ − B µ = δσγ − δσγ = 0
∂y0σ ∂y σ
Now take (X, d) = (C(I, J), k·k∞ ), with J = B  (y0γ ), and define an application H : X −→ X such
that
H γ (wσ (xν )) = Gγ (xν , wσ (xν ))
We need to demonstrate that this application is a contraction, i.e. that ∃!g γ (xν ) : f µ (xν , g γ (xν )) =
0 ∀(xν , y γ ) ∈ I × J

kH γ (wσ (xν )) − y0γ kγ = kGγ (xν , wσ (xν )) − y0γ kγ ≤

≤ kGγ (xν , wσ (xν )) − Gγ (xν , y0σ )kγ + kGγ (xν , y0σ ) − Gγ (xν0 , y0ν )kγ ≤
∂Gγ
≤ kwσ (xν ) + y0σ kσ + kGγ (xν , y0σ ) − Gγ (xν0 , y0σ )kγ ≤ 
∂y σ ∞
Since kGγ (xν , y0γ ) − Gγ (xν0 , y0γ )kγ ≤ /2 and kwσ (xν ) − y0σ kσ ≤ , ∀(xν , y γ ) ∈ I × J, we have
∂Gγ 1
≤
∂y σ ∞ 2
Therefore
1 γ ν
kH γ (wγ (xν )) − H γ (v σ (xν ))kγ ≤ kw (x ) − v γ (xν )kγ
2
I.e. H is a contraction in C(I, J).
Due to the differentiability of f µ we can write
∀ > 0 ∃η : khν kν , kk γ kγ ≤ η =⇒ kf µ (xν , g γ (xν ) + k γ ) − f µ (xν , g γ (xν )) − ∂xν f µ hν − ∂yγ f µ k γ kµ ≤
≤ (khν kν + kk γ kγ )
Letting k γ = g γ (xν + hν ) − g γ (xν ) we have by definition
f µ (xν + hν , g γ (xν ) + k γ ) − f µ (xν , g γ (xν )) = 0
And therefore, putting ∂yγ f µ = δγµ
 
kg γ (xν + hν ) + ∂xν f µ hν kµ ≤  khν kν + kg γ (xν + hν ) − g γ (xν )kγ

Letting  = 1/2 we have kg γ (xν + hν ) − g γ (xν )k ≤ η1/2 , and we have

kg γ (xν + hν ) − g γ (xν )kγ ≤ kg γ (xν + hν ) − g γ (xν ) − ∂xν f γ hν kγ k∂xν f µ hν kγ
1 ν 
≤ kh kν + kg γ (xν + hν ) − g γ (xν )kγ + k∂xν f µ k∞ khν k
2
Which implies
kg γ (xν + hν ) − g γ (xν )kγ ≤ khν kν (1 + 2k∂xν f µ k∞ )
Which implies that g γ (xν ) is continuously differentiable in I. Whenever ∂yγ f µ 6= δγµ we can find a
transformed function f˜µ such that ∂yγ f˜µ = δγµ
3.6. OPTIMIZATION 49

§§ 3.6.3 Lagrange Multipliers

DefinitionS3.6.5 (Vinculated Critical Points). Let f : K ⊂ A ⊆ R2 −→ R with A open and f ∈ C 1 (A).
Let ∂K = k≤n γk with γk : [ak , bk ] −→ R2 . The critical points of f |∂K are found in Pik = γk (ti ) ∈ ∂K,
for which
d
f (Pik ) = 0
dt

Definition 3.6.6 (Argmax, Argmin). Let f µ : A −→ Rn a function which reach its maximum in
xνi ∈ A i = 1, · · · , m and its minimum at yjν ∈ A j = 1, · · · , k Then we can define

Arg max(f ) : = {xν1 , · · · , xνm }

A
(3.98)
Arg min(f ) : = {y1ν , · · · , ykν }
A

Theorem 3.28 (Lagrange Multipliers). Let f, g : A −→ R, f, g ∈ C 1 (A), A ⊆ Rn open, and

M = { xµ ∈ A| g(xµ ) = 0} and let xµ0 ∈ M : ∂m ug(xµ0 ) 6= 0, then xµ0 ∈ Arg maxM f ∨ Arg minM f
if it’s a free critical point of the Lagrangian

L(xµ , λ) = f (xµ ) − λg(xµ ) (xµ , λ) ∈ A × R (3.99)

I.e. ∃λ0 ∈ R : (xµ0 , λ0 ) solves (

∂µ f (xν ) = λ∂µ g(xν )
(3.100)
g(xµ ) = 0
Or, that
∂µ f (xν0 )
 
rank =1
∂µ g(xν0 )

Proof. Let ∂n g 6= 0, then we can see M as a graph of a regular implicit function of g, h : Rn−1 −→ R,
where
g(xµ , h(xµ )) = 0 ∀xµ ∈ Br (x˜µ 0 ) ⊂ Rn−1
Letting ϕ : (−, ) −→ Br (xµ0 ) a smooth curve, such that ϕµ (0) = xµ0 , we have that ψ ν (t) =
(ϕµ (t), h(t)) ∈ M is the parameterization of a smooth curve that passes through xµ0 ∈ M. We
have
d
f (ψ ν (0)) = ∂µ f φ̇µ (0) + ∂n f ḣ(φµ (0)) = ∂ν f (xµ0 )sν
dt
d
g(ψ ν (0)) = ∂ν g(xµ0 )sν
dt
With sν = ψ˙ν (0), therefore ∂ν f kψν (0) ∂ν g
Theorem 3.29 (Generalized Lagrange Multiplier Method). Let f, gi : A ⊆ Rn −→ R, 0 < i < n,
f, gi ∈ C 1 (A) with A an open set, let M := { xν ∈ A| g(xν ) = 0}. Take xν0 ∈ M such that

rank ∂ν g µ (xγ0 ) = k
CHAPTER 3. DIFFERENTIAL ANALYSIS 50

Then xν0 is a critical point for f |M , and it’s a free critical point for the Lagrangian L

L(xγ , λµ ) = f (xγ ) − λν g ν (xγ )

I.e. ∃(xγ0 , λν0 ) ∈ A × R solution of the system

(
∂ν f (xγ ) = ∂ν g µ (xγ )λν
g ν (xγ ) = 0

Alternatively, one can check that

∂µ f (xγ0 )
 
rank(A) = =k
∂ν g µ (xγ )
4 Tensors and Differential Forms

§ 4.1 Tensors and k -forms

§§ 4.1.1 Basic Definitions, Tensor Product and Wedge Product
Definition 4.1.1 (Multilinear Functions, Tensors). Let V be a real vector space, and take V k = V ×· · ·×V
k−times. A function T : V k −→ R is called multilinear if ∀i = 1, · · · , k, ∀a ∈ R, ∀v, w ∈ V
T (v1 , · · · , avi + wi , · · · , vk ) = aT (v1 , · · · , vi , · · · , vk ) + T (v1 , · · · , wi , · · · , vk ) (4.1)
A multilinear function of this kind is called k-tensor on V. The set of all k−tensors is denoted as T k (V)
and is a real vector space.
The tensor T is usually denoted as follows
Tµ1 ...µk (4.2)
Where each index indicates a slot of the multilinear application T (−, · · · , −)
Definition 4.1.2 (Tensor Product). Let S ∈ T k (V ), T ∈ T l (V), we define the tensor product S ⊗ T ∈
T k+l (V) as follows
(S ⊗ T )(v1 , · · · , vk , vk+1 , · · · , vk+l ) = S(v1 , · · · , vk )T (vk+1 , · · · , vk+l ) (4.3)
This product has the following properties
(S1 + S2 ) ⊗ T = S1 ⊗ T + S2 ⊗ T
S ⊗ (T1 + T2 ) = S ⊗ T1 + S ⊗ T2
(4.4)
(aS) ⊗ T = S ⊗ (aT ) = a(S ⊗ T )
(S ⊗ T ) ⊗ U = S ⊗ (T ⊗ U ) = S ⊗ T ⊗ U
If S = Sµ1 ...µk and T = Tµk+1 ...µk+l we have
(S ⊗ T )µ1 ...µk µk+1 ...µk+l = Sµ1 ...µk Tµk+1 ...νk+l (4.5)

Definition 4.1.3 (Dual Space). We define the dual space of a real vector space V as the space of all
linear functionals from the space to the field over it’s defined, and it’s indicated with V ? . I.e. let
ϕµ ∈ V ? , then ϕµ : V −→ R.
It’s easy to see how V ? = T 1 (V).

51
CHAPTER 4. TENSORS AND DIFFERENTIAL FORMS 52

Theorem 4.1. Let B = {vµ1 , · · · , vµn } be a basis for the space V, and let B ? := {ϕµ1 , · · · , ϕµn } be
the basis of the dual space, i.e. ϕµ vν = δνµ ∀ϕµ ∈ B ? , vµ ∈ B, then the set of all k-fold tensor
products has basis BT , where

BT := {ϕµ1 ⊗ · · · ⊗ ϕµk , ∀i = 1, · · · , n} (4.6)

Theorem 4.2 (Linear Transformations on Tensor Spaces). If fνµ : V −→ W is a linear transformation,

fµν ∈ L(V, W), one can define a linear transformation f ? : T k (W ) −→ T k (V ) as follows

f ? T (vµ1 , · · · , vµk ) = T (fνµ vµ1 , · · · , fνµ vµk )

Theorem 4.3. If g is an inner product on V (i.e. g : V × V −→ R, with the properties of an inner

product), there is a basis vµ1 , · · · , vµn of V such that g(vµ , vν ) = gµν = gνµ = g(vν , vµ ) = δµν .
This basis is called orthonormal with respect to T . Consequently there exists an isomorphism
∼
fνµ : Rn −→ V such that
g(fνµ xν , fνµ y ν ) = xµ y µ = gµν xµ y ν (4.7)
I.e.
f ? g(·, ·) = gµν (4.8)

Definition 4.1.4 (Alternating Tensor). Let V be a real vector space, and ω ∈ T k (V). ω is said to be
alternating if

ω(vµ1 , · · · , vµi , · · · , vµj , · · · , vµk ) = −ω(vµ1 , · · · , vµj , · · · , vµi , · · · , vµk )

(4.9)
ω(vµ1 , · · · , vµi , · · · , vµi , · · · , vµk ) = 0

Or, compactly
ωµ...ν...γ...σ = −ωµ...γ...ν...σ
(4.10)
ωµ...ν...ν...γ = 0
The space of all alternating k−tensors on V is indicated as Λk (V), and we obviously have that Λk (V) ⊂
T k (V).
We can define an application Alt : T k (V) −→ Λk (V) as follows
1 X
Alt(T )(v1µ , · · · , vkµ ) = µ
sgn(σ)T (vσ(1) µ
, · · · , vσ(k) ) (4.11)
k!
σ∈Σk

With σ = (i, j) a permutation and Σk the set of all permutations of natural numbers 1, · · · , k Compactly,
we define an operation on the indices, indicated in square brackets, called the antisymmetrization of
the indices inside the brackets.
This definition is much more general, since it lets us define a partially antisymmetric tensor, i.e. anti-
symmetric on only some indices.
1
Alt(Tµ1 ...µk ) = T[µ1 ...µk ] (4.12)
k!
4.1. TENSORS AND K-FORMS 53

As an example, for a 2−tensor aµν we can write

1
a[µν] = (aµν − aνµ ) = ãµν ∈ Λ2 (V) (4.13)
2
This is valid for general tensors. If we define a k−tensor over the product repeated k times for V and k
for its dual space V × · · · V × V ? × · · · × V ? , we can define the space T k (V × V ? ) = W. Let the basis
for this space be the following
BW := {vµ1 ⊗ · · · ⊗ vµk ⊗ ϕν1 ⊗ · · · ⊗ ϕνk }
Then an element Y of the space W can be written as follows
Y(vµ1 , · · · , vµk , ϕν1 , · · · , ϕνk ) = Yµν11 ...νk
...µk

We can define a new element Y ∈ Λk (V × V ? ) using the antisymmetrization brackets

[ν ...ν ]
Yµν11...µ
...νk
k
= Y[µ11 ...µkk ]

We can define also partially antisymmetric parts as follows

ν1 ...[νi νi+1 ]...νk 1  ν1 ...νi νi+1 ...νk 
Rµν11...ν
...µk = Yµ1 ...[µl µl+1 ]...µk =
k
Yµ1 ...µl µl+1 ...µk − Yµν11 ...νi+1 νi ...νk ν1 ...νi νi+1 ...νk ν1 ...νi νi+1 ...νk
...µl µl+1 ...µk + Yµ1 ...µl µl+1 ...µk − Yµ1 ...µl+1 µl ...µk
4!
Note how the indexes in the expressions with the label i and l simply got switched, and in the new
definition, the tensor R is antisymmetric in both the covariant (lower) indexes µl , µl+1 and in the
contravariant (upper) indexes νi , νi+1 , where obviously i, l ≤ k
Theorem 4.4. Let T ∈ T k (V) and ω ∈ Λk (V). Then

T[µ1 ...µk ] ∈ Λk (V)

ω[µ1 ...µk ] = ωµ1 ...µk (4.14)
T[[µ1 ...µk ]] = T[µ1 ...µk ]

Definition 4.1.5 (Wedge Product). Let ω ∈ Λk (V), η ∈ Λl (V). In general ω ⊗ η ∈

/ Λk+l (V), hence we
define a new product, called the wedge product, such that ω ∧ η ∈ Λ (V)
k+l

(k + l)!
ωµ1 ...µk ∧ ην1 ...νk = ω[µ1 ...µk ην1 ...νl ] (4.15)
k!l!
With the following properties
∀ω, ω1 , ω2 ∈ Λk (V), ∀η, η1 , η2 ∈ Λl (V), ∀a ∈ R, ∀f ? ∈ L : T k (V) −→ T l (V) ∀θ ∈ Λm (V)
(ω1 + ω2 ) ∧ η = ω1 ∧ η + ω2 ∧ η
ω ∧ (η1 + η2 ) = ω ∧ η1 + ω ∧ η2
(ω ∧ η) ∧ θ = ω ∧ (η ∧ θ)
(4.16)
aω ∧ η = ω ∧ aη = a(ω ∧ η)
ω ∧ η = (−1)kl η ∧ ω
f ? (ω ∧ η) = f ? (ω) ∧ f ? (η)
CHAPTER 4. TENSORS AND DIFFERENTIAL FORMS 54

Theorem 4.5. The set

{ϕµ1 ∧ · · · ∧ ϕµk , k < n} ⊂ Λk (V) (4.17)
Is a basis for the space Λk (V), and therefore
 
n n!
dim(Λk (V)) = =
k k!(n − k)!

Where dim(V) = n.
Therefore, dim(Λn (V)) = 1
Theorem 4.6. Let vµ1 , · · · , vµn be a basis for V, and take ω ∈ Λn (V), then, if wµ = aνµ vν

ω(wµ1 · · · wµn ) = det(aµν )ω(vµ1 , . . . , vµn ) (4.18)

µν

Or using the basis representation of a vector tµ = tµ wµ = tµ aνµ vν we have

ωµ1 ...µn tµ1 · · · tµn = det(aµν )ων1 ...νn tν1 · · · tνn (4.19)
µν

Proof. Define ηµ1 ...µn ∈ T n (Rn ) as

ηµ1 ...µn aµν11 aµν22 · · · aµvnn = ωµ1 ...µn aµν11 · · · aµνnn

Hence η ∈ Λn (Rn ) so η = λ det(·) for some λ, and

λ = ηµ1 ...µn eµ1 · · · eµn = ωµ1 ...µn v µ1 · · · v µn

§§ 4.1.2 Volume Elements and Orientation

Definition 4.1.6 (Orientation). The previous theorem shows that a ω ∈ Λn (V), ω 6= 0 splits the bases
of V in two disjoint sets.
Bases for which ω(Bv ) > 0 and for which ω(Bw ) < 0. Defining wµ = aµν v ν we have that the two bases
belong to the same group iff detµν (aµν ) > 0. We call this the orientation of the basis of the space.
The usual orientation of Rn is
[eµ ]
Given another two basis of Rn we can define (taking the first two examples)

[vµ ]
−[wµ ]

Definition 4.1.7 (Volume Element). Take a vector space V such that dim(V) = n and it’s equipped
with an inner product g, such that there are two bases (v µ1 , · · · , v µn ), (wµ1 , · · · , wµn ) that satisfy the
orthonormality condition with respect to this scalar product

gµν v µi v νj = gσγ wσi wγj = δij (4.20)

4.2. TANGENT SPACE AND DIFFERENTIAL FORMS 55

Then
ωµ1 ...µn v µ1 · · · v µn = ωµ1 ...µn wµ1 · · · wµn = det(aµν ) = ±1
µν

Where
wµ = aµν v ν
Therefore
∃!ω ∈ Λn (V) : ∃![wµ1 , · · · , wµn ] = O
Where O is the orientation of the vector space.
Definition 4.1.8 (Cross Product). Let v1µ , · · · , vnµ ∈ Rn+1 and define ϕν wν as follows
 µ 
v 1
 .. 
ϕν wν = det  . 
 
µ
v n

wν

Then ϕ ∈ Λ1 (Rn+1 ), and

∃!z µ ∈ Rn+1 : z µ wµ = ϕν wν
z µ is called the cross product, and it’s indicated as

z µ = v ν1 × · · · × v νn = µν1 ...νn v ν1 · · · v νn

§ 4.2 Tangent Space and Differential Forms

Definition 4.2.1 (Tangent Space). Let p ∈ Rn , then the set of all pairs { (p, v µ )| v µ ∈ Rn } is denoted
as Tp Rn and it’s called the tangent space of Rn (at the point. This is a vector space defining the
following operations

(p, av µ ) + (p, awµ ) = (p, a(v µ + wµ )) = a(p, v µ + wµ ) ∀v µ , wµ ∈ Rn , a ∈ R

Remark. If a vector v µ ∈ Rn can be seen as an arrow from 0 to the point v, a vector (p, v µ ) ∈ Tp Rn
can be seen as an arrow from the point p to the point p + v. In concordance with the usual notation
for vectors in physics, we will write (p, v µ ) = v µ directly, or vpµ when necessary to specify that we’re
referring to the vector v µ ∈ Tp Rn . The point p + v is called the end point of the vector vpµ .
Definition 4.2.2 (Inner Product in Tp Rn ). The usual inner product of two vectors vpµ , wpµ ∈ Tp Rn is
defined as follows
h·, ·ip :Tp Rn × Tp Rn −→ R
(4.21)
vpµ wµp = v µ wµ = k
Analogously, one can define the usual orientation of Tp Rn as follows

[(eµ1 )p , · · · , (eµn )p ]
CHAPTER 4. TENSORS AND DIFFERENTIAL FORMS 56

Definition 4.2.3 (Vector Fields, Again). Although we already stated a definition for a vector field, we’re
gonna now state the actual precise definition of vector field
Let p ∈ Rn be a point, then a function f µ (p) : Rn −→ Tp Rn is called a vector field, if ∀p ∈ A ⊆ Rn we
can define
f µ (p) = f µ (p)(eµ )p (4.22)
Where (eµ )p is the canonical basis of Tp Rn
All the previous (and already stated) considerations on vector fields hold with this definition.
Definition 4.2.4 (Differential Form). Analogously to vector fields, one can define k−forms on the
tangent space. These are called differential (k-)forms and “live” on the space Λk (Tp Rn ).
?
Let ϕµp 1 , · · · , ϕµp k ∈ (Tp Rn ) be a basis on such space, then the differential form ω ∈ Λk (Tp Rn ) is
defined as follows
X
ωµ1 ...µk (p) = ωµ1 ...µk ϕ[µ1 µk ]
p · · · ϕp → ωi1 ...ik (p)ϕi1 (p) ∧ · · · ∧ ϕik (p) (4.23)
i1 <...<ik

A function f : Tp Rn −→ R is defined as f ∈ Λ0 (Tp Rn ), or a 0−form. In general, so, we can write

without incurring in errors
f (p)ω = f (p) ∧ ω = f (p)ωµ1 ...µk (4.24)

§§ 4.2.1 External Differentiation, Closed and Exact Forms

Definition 4.2.5 (Differential). Now we will omit that we’re working on a point p ∈ Rn and we’ll use
the usual notation.
Let f : Tp Rn −→ R be a smooth (i.e. continuously differentiable) function, where f ∈ C ∞ , then,
using operatorial notation we have that ∂µ f (v) ∈ Λ1 (Rn ), therefore, with a small modification, we
can define
df (vpν ) = ∂µ f (v ν ) (4.25)
It’s obvious how dxµ (vpν ) = ∂ν xµ (v ν ) = v µ , therefore dxµ is a basis for Λ1 (Tp Rn ), which we will
indicate as dxµ , therefore ∀ω ∈ Λk (Tp Rn )
X
ωµ1 ...µk = ωµ1 ...µk dx[µ1 · · · dxµk ] → ωi1 ...ik (p) dxi1 ∧ · · · ∧ dxik (4.26)
i1 <...<ik

Basically, the vectors dxµ are the dual basis with respect to the canonical basis (eµ )p
Theorem 4.7. Since df (vpν ) = ∂ν f (v ν ) we have, expressing the differential of a function with
the basis vectors,
∂f
df = dxµ = ∂µ f dxµ (4.27)
∂xµ

Definition 4.2.6. Having defined a smooth linear transformation fνµ : Rn −→ Rm , it induces another
linear transformation ∂γ fνµ : Rn −→ Rm , which with some modifications becomes the application
(f? )µν : Tp Rn −→ Tf (p) Rm defined such that
 µ
(f? )µν (v ν ) = df |f (p) (v ν ) (4.28)
ν
4.2. TANGENT SPACE AND DIFFERENTIAL FORMS 57

Which, in turn, also induces a linear transformation f ? : Λk (Tf (p) Rm ) −→ Λk (Tp Rn ), defined as
follows. Let ωp ∈ Λk (Rm ), then we can define f ? ω ∈ Λk (Tf (p) Rn ) as follows

(f ? ωp )(vµ1 , . . . , vµk ) = ωf (p) (f? )µν11 vµ1 , · · · , (f? )µνkk vµk

(4.29)

(Just remember that in this way we are writing explicitly the chosen base, watch out for the indexes!)

Theorem 4.8. Let f : Rn −→ Rm be a smooth function, then

1. (f ? )µν (dxν ) = df = ∂ν f µ dxν

2. f ? (ω1 + ω2 ) = f ? ω1 + f ? ω2

3. f ? (gω) = (g ◦ f )f ? ω

4. f ? (ω ∧ η) = f ? ω ∧ f ? η

5. f ? h dx[µ1 · · · dxµn ] = h ◦ f detµν (∂ν f µ ) dx[µ1 · · · dxµn ]

d
Definition 4.2.7 (Exterior Derivative). We define the operator das an operator Λk (Tp V) −→ Λk+1 (Tp V)
for some vector space V. For a differential form ω it’s defined as follows

(dω)νµ1 ...µk = ∂[ν ωµ1 ...µk ] (4.30)

This, using the classical mathematical notation can be written as follows

X
dω = dωi1 ,...,ik ∧ dxi1 ∧ · · · ∧ dxik
i1 <...<ik
n (4.31)
X X ∂
dω = ω
j i1 ,...,ik
dxj ∧ dxi1 ∧ · · · ∧ dxik
i1 <...<ik j=1
∂x

Theorem 4.9 (Properties of d). 1. d(ω + η) = dω + dη

2. d(ω ∧ η) = dω ∧ η + (−1)k ω ∧ dη for ω ∈ Λk (V), η ∈ Λl (V)

2
3. ddω = d ω = 0

4. f ? (dω) = d(f ? ω)

Definition 4.2.8 (Closed and Exact Forms). A form ω is called closed iff

dω = 0 (4.32)

It’s called exact iff

ω = dη (4.33)

Theorem 4.10. Let ω be an exact differential form. Then it’s closed

CHAPTER 4. TENSORS AND DIFFERENTIAL FORMS 58

Proof. The proof is quite straightforward. Since ω is exact we can write ω = dρ for some differential
form ρ, therefore
2
dω = ddρ = d ρ = 0
Hence dω = 0 and ω is closed.

Example 4.2.1. Take ω ∈ Λ1 (R2 ), where it’s defined as follows

ωµ = p dx + q dy (4.34)

The external derivative will be of easy calculus by remembering the mnemonic rule d → ∂µ ∧ dxµ , or
also as ∂[ν then we have
dωµν = ∂[ν ωµ]
But  
∂1 ω1 ∂1 ω2
∂ν ωµ =
∂2 ω1 ∂2 ω2 µν

And
1 1
∂[ν ωµ] = (∂ν ωµ − ∂µ ων ) = (∂ω − ∂ω T )
2 2
Therefore  
1 0 ∂x q − ∂y p
dωµν =
2 ∂y p − ∂x q 0 µν
2 2
Which, expressed in terms of the basis vectors of Λ (R ), dx ∧ dy, we get

1 1
dω = (∂x q − ∂y p) dx ∧ dy + (∂y p − ∂x q) dy ∧ dx = (∂x q − ∂y p) dx ∧ dy (4.35)
2 2
Therefore
dω = 0 ⇐⇒ ∂x q − ∂y p = 0 (4.36)

Definition 4.2.9 (Star Shaped Set). A set A is said to be star shaped with respect to a point a iff
∀x ∈ A the line segment [a, x] ⊂ A

Lemma 4.2.1 (Poincaré’s). Let A ⊂ Rn be an open star shaped set, with respect to 0. Then every closed
form on A is exact

§ 4.3 Chain Complexes and Manifolds

§§ 4.3.1 Singular n−cubes and Chains
Definition 4.3.1 (Singular n−cube). A singular n-cube is an application c : [0, 1]n −→ A ⊂ Rn . In
general. A singular 0-cube is a function f : {0} −→ A and a singular 1−cube is a curve.

Definition 4.3.2 (Standard n−cube). We define a standard n-cube as a function I n : [0, 1]n −→ Rn
such that I n (xµ ) = xµ .
4.3. CHAIN COMPLEXES AND MANIFOLDS 59

Definition 4.3.3 (Face). Given a standard n−cube I n we define the (i, α)−face of the cube as
n
I(i,α) = (x1 , · · · , xi−1 , α, xi , · · · , xn−1 ) α = 0, 1 (4.37)

Definition 4.3.4 (Chain). Given n k−cubes ci , we define a n-chain s as follows

n
X
s= ai ci ai ∈ R (4.38)
i=1

Definition 4.3.5 (Boundary). Given an n−cube ci we define the boundary as ∂ci . For a standard
n−cube we have
n X
X
∂I n = (−1)i+α I(i,α)
n
(4.39)
i=1 α=0,1

For a k−chain s we define X X

∂s = ∂( ai ci ) = ai ∂ci (4.40)
i i

Where ∂s is a (k − 1)-chain
Theorem 4.11. For a chain c, we have that ∂∂c = ∂ 2 c = 0

§§ 4.3.2 Manifolds
Definition 4.3.6 (Manifold). Given a set M ⊂ Rn , it is said to be a k-dimensional manifold if
∀xµ ∈ M we have that
1. ∃U ⊂ Rk open set xµ ∈ U
and V ⊂ Rn and ϕ a diffeomorphism such that U ' V and
ϕ (U ∩ M ) = V ∩ Rk × {0} , i.e. U ∩ M ' Rk ∩ {0}
2. ∃U ⊂ Rk open and W ⊂ Rk open, xµ ∈ U and f : W −→ Rn a diffeomorphism
(a) f (W ) = M ∩ U
(b) rank (f ) = k ∀xµ ∈ W
(c) f −1 ∈ C(f (W ))
The function f is said to be a coordinate system in M
Definition 4.3.7 (Half Space). We define the k-dimensional half space Hk ⊂ Rk as

Hk := xµ ∈ Rk xi ≥ 0

(4.41)

Definition 4.3.8 (Manifold with Boundary). A manifold with boundary (MWB) is a manifold M such
that, given a diffeomorphism h, an open set U ⊃ M and an open set V ⊂ Rn

h (U ∩ V ) = V ∩ Hk ∩ {0}


(4.42)

The set of all points that satisfy this forms the set ∂M called the boundary of M
CHAPTER 4. TENSORS AND DIFFERENTIAL FORMS 60

Definition 4.3.9 (Tangent Space). Given a manifold M and a coordinate set f around xµ ∈ M , we
define the tangent space of M at xµ ∈ M as follows

f : W ⊂ Rk −→ Rn =⇒ f? Tx Rk = Tx M


(4.43)

Definition 4.3.10 (Vector Field on a Manifold). Given a vector field f µ we identify it as a vector field
on a manifold M if f µ (xν ) ∈ Tx M . Analogously we define a k−differential form
5 Integral Analysis

§ 5.1 Measure Theory

Definition 5.1.1 (Lower and Upper Sums). We define the upper and lower Riemann sums as follows.
Let f (x) be a function, then 
Xn
U(f, x) : = sup (f (t))




i=1 t∈[xk ,xk+1 ]

n
(5.1)
 X
 L(f, x) : = inf (f (t))



t∈[x ,x ]k k+1
i=1

A function is said to be Riemann integrable if limn→∞ (L(f, x) − U(f, x)) = 0

Definition 5.1.2 (Set Function). Let A be a set. We define the following function 1A (x) as follows
(
1 x∈A
1A (x) = (5.2)
0 x∈ /A

Theorem 5.1. The function 1Q is not integrable over the set [0, 1] with the usual definition of
the integral (Riemann sums)
Proof. Indicating the integral I as usual
ˆ 1
I= 1Q (x) dx
0

We see immediately that

U(1Q , x) = 1
L(1Q , x) = 0
Therefore 1Q (x) is not integrable in [0, 1] (with the Riemann integral)
Definition 5.1.3 (Measure). Let A ⊂ X be a subset of a metric space. We define the measure of the
set A, µ(A) as follows ˆ
µ(A) = 1A (x) dx (5.3)
X

61
CHAPTER 5. INTEGRAL ANALYSIS 62

Basically, what we did before, was demonstrating that the set Q ∩ [0, 1] is not measurable in the
Riemann integration theory. This is commonly indicated with saying that the set Q ∩ [0, 1] is not Jordan
measurable.
For clarity, let K be some measure theory. We will say that a set is K-measurable if the following
calculation exists ˆ
µK (A) = 1A (x) dx (5.4)
X

Definition 5.1.4 (Algebra). Let X 6= {} be a set. An algebra A over X is a collection of subsets of X

such that

1. {} ∈ A

2. X ∈ A

3. A ∈ A =⇒ Ac ∈ A
Sn Tn
4. A1 , · · · , An ∈ A =⇒ i=1 Ai , i1 Ai ∈ A

Example 5.1.1 (Simple Set Algebra). Let X = R2 and call R the set of all rectangles Ii ⊂ R? × R? ,
where R? = R ∪ {±∞}. It’s easy to see that this is not an algebra, since, by taking [0, 1] ∈ R, we have
that [0, 1]c ∈
/ R, hence it cannot be an algebra.
But, taken S as follows ( )
[n
2
S := A ⊂ R A = Ii Ii ∈ R
i=1

We can see easily, using De Morgan law, that S is an algebra.

§§ 5.1.1 Jordan Measure

Definition 5.1.5 (Disjoint Union). Taken two sets A, B, we define their disjoint union the binary
operation A t B as follows
A t B := A ∪ B \ A ∩ B (5.5)

Definition 5.1.6 (Simple Set). A set A is a simple set iff, for some Ri ∈ S, we have
n
G
A= Ri
i=1

Definition 5.1.7 (Measure of a Simple Set). Let A be a simple set, the Jordan measure of a simple set
is given by the sum of the measure of the rectangles, i.e. the “area” of A is given by the sum of the
area of each rectangle Ri
Xn
µJ (A) = µJ (Ri ) (5.6)
i=1
5.1. MEASURE THEORY 63

Definition 5.1.8 (External and Internal Measure). We define the external measure µJ and the internal
measure µJ as follows.
Taken a limited set B and a simple set A we have
µJ (B) = inf{ µJ (A)| B ⊂ A}
(5.7)
µJ (B) = sup{ µJ (A)| A ⊂ B}

A set is said to be Jordan measurable iff

µJ (B) = µJ (B) = µJ (B)

Remark (A Non Measurable Set). A good example for showing that the Jordan measure is the set we
were trying to measure, the set Q ∩ [0, 1]. We can easily see that
µJ (Q ∩ [0, 1]) = 1
µJ (Q ∩ [0, 1]) = 0

Therefore it’s not Jordan measurable.

From this we can jump to a new definition of measure, which is the Lebesgue measure where instead
of covering Q ∩ [0, 1] with a finite number of simple sets, we use sets which are formed from the
union of countable infinite simple sets.
We can define
Q ∩ [0, 1] := {q1 , q2 , · · · }
We then take  > 0 and define the following set
∞ h
[   i
A= qn − n
, qn + n
n=1
2 2

We have that
∞
X 
µ(A) ≤ = 2
n=1
2n−1
But µ(Q ∩ [0, 1]) ≤ µ(A) ≤ 2 → 0, therefore Q ∩ [0, 1] is measurable with µ(Q ∩ [0, 1]) = 0

§§ 5.1.2 Lebesgue Measure

Definition 5.1.9 (σ−Algebras and Borel Spaces). Given a non empty set X a σ−algebra on X is a
collection of subsets F such that
1. ∀A ∈ F , A ⊂ X
S∞
2. Let Ai ∈ F , i ∈ I : |I| = ℵ0 then i=1 Ai ∈ F
The couple (X, F) is called a Borel space or also a measurable space
Definition 5.1.10 (Measure). Given a Borel space (X, F), we can define an application
µ : F −→ [0, ∞] = R?+ (5.8)
Which satisfies the following properties
CHAPTER 5. INTEGRAL ANALYSIS 64

1. σ-additivity, given Ai ∈ F with i ∈ I ⊂ N, |I| ≤ ℵ0 , such that An ∩ Ak = {} for n 6= k

!
G X
µ Ai = µ(Ai )
i∈I i∈I

S∞
2. If Yj ⊂ X, with j ∈ J ⊆ N, µ(Yj ) < ∞ then X = j=1 Yj
Definition 5.1.11 (Measure Space). A measure space is a triplet (X, F, µ) with F a σ−algebra and µ
a measure.
Remark. The empty set has null measure.
Proof. Due to σ−additivity we have that
µ({}) = µ({} ∪ {}) = µ({}) + µ({})
Therefore, µ({}) = 0 necessarily.
Definition 5.1.12 (Lebesgue Measure). Consider again X = R2 and S the algebra of simple sets.
The external Lebesgue measure of a set B ⊂ R2 is then defined as follows
( ∞ ∞
)
X [
µL (B) := inf Area(Ri ) Ri ∈ S, B ⊂ Ri (5.9)
i=1 i=1

The set B is said to be Lebesgue measurable iff, ∀C ⊂ R 2

µL (C) = µL (C ∩ B) + µL (C \ B) (5.10)
If it’s measurable, then, µL (B) = µL (B) and it’s called the Lebesgue measure of the set.
In other words ∃ > 0 : ∃A, C ⊂ R2 , with A = A◦ , C = C such that
C ⊂ B ⊂ A ∨ µL (A \ C) <  (5.11)

Definition 5.1.13 (Borel Algebra). Let R be the set of all rectangles. The smallest σ−algebra containing
R is called the Borel algebra and it’s indicated as B
Definition 5.1.14 (Lebesgue Algebra). The set of (Lebesgue) measurable sets is a σ−algebra, which we
will indicate as L. In particular, we have that, if I is a rectangle, I ∈ L.
If we add the fact that in B there are null measure sets which have subsets which aren’t part of B, we
end up with the conclusion that B ⊂ L
Definition 5.1.15 (Null Measure Sets). A set with null measure is a set X ⊂ F such that
µ(X) = 0 (5.12)
Where µ is a measure function.
It’s obvious that sets formed by a single point have null measure.
I.e take a set A = {a}, then it can be seen as a rectangle with 0 area, and therefore
µ ({a}) = 0 (5.13)
5.2. INTEGRATION 65

Theorem 5.2. Every set such that |A| = ℵ0 has null measure

Corollary 5.1.1. Every line in R2 has null measure

Proof. Take the set A = {a1 , a2 , a3 , · · · }. Then, due to σ−additivity, we have

∞ ∞
!
G X
µ ({a1 , a2 , a3 , · · · }) = µ {ak } = µ ({ak }) = 0 (5.14)
k=1 k=1

For the corollary, it’s obvious if the line is thought as a rectangle in R2 with null area

§ 5.2 Integration
Definition 5.2.1 (Measurable Function). Given a Borel space (X, F) a measurable function is a
function f : X −→ F such that, ∀k ∈ F the following set is measurable

If := { k ∈ F| f (x) < k} (5.15)

Or, in other words If ∈ F , with F the given σ−algebra of the Borel space.
The space of all measurable functions on X will be identified as M(X)

Theorem 5.3. Given a set A ∈ F with F a σ−algebra, the function 1A (x) is measurable

Proof. We have that (

A k>1
I1A =
{} t ≤ 1

Therefore I1A ∈ F and 1A (x) is measurable

Definition 5.2.2 (Simple Measurable Function). Given a Borel space (X, F), a simple measurable
function is a function f : X −→ F which can be written as follows
n
X
f (x) = ck 1Ak (x) (5.16)
k=1

Where Ak ∈ F , ck ∈ F 0≤k≤n

Definition 5.2.3 (Integral). Given a measure space (X, F, µ) and a simple function f (x), we can define
the integral of the function f with respect to the measure µ over the set X as follows
ˆ n
X
f (x)µ (dx) = ck µ(Ak ) (5.17)
X k=1

For non negative functions we define the integral as follows

ˆ ˆ 
f (x)µ (dx) = sup g(x)µ (dx) (5.18)
X X
CHAPTER 5. INTEGRAL ANALYSIS 66

Where g(x) is a simple measurable function such that 0 ≤ g ≤ f .

If f assumes both negative and positive values we can write

f = f+ − f− (5.19)

Where (
f + = max {f, 0}
(5.20)
f − = max {−f, 0}

The integral, due to linearity, then will be

ˆ ˆ ˆ
f (x)µ (dx) = f + (x)µ (dx) − f − (x)µ (dx) (5.21)
X X X

With the only constraint that the function f (x) must be misurable in the σ-algebra F

§§ 5.2.1 Lebesgue Spaces

Definition 5.2.4 (Lp spaces). With the previous definitions, we can define an infinite dimensional
function space with the following properties
Given a measure space (X, F, µ) we have the following definition
 ˆ 
p p p
L (X, F, µ) = L (µ) := f : X −→ F| If ∈ F ∧ |f | µ (dx) < ∞ (5.22)
X

Defining the integral as an operator K̂µ [f ] we can see easily that this is a vector spaces due to the
properties of K̂µ .
It’s easy to note that if the chosen σ−algebra and measure are the Lebesgue ones, then this integral is
simply an extension of the usual Riemann integral.
It’s important to note that a norm in Lp (µ) can’t be defined as an usual integral p−norm, since there
are nonzero functions which have actually measure zero.

Definition 5.2.5 (Almost Everywhere Equality). Taken two functions f, g ∈ Lp (µ) we say that these two
function are almost everywhere equal if, given a set A := { x ∈ X| f (x) 6= g(x)} has null measure.
Therefore
f ∼ g ⇐⇒ µ(A) = 0 (5.23)
This equivalence relation creates equivalence classes of functions compatible with the vector space
properties of Lp (µ).

Definition 5.2.6 (Lp -Spaces). With the definition of the almost everywhere equality we can then define
a quotient space as follows
Lp (µ) = Lp (µ)\ ∼ (5.24)
This is a vector space, obviously, where the elements are the equivalence classes of functions f ∈ Lp (µ),
indicated as [f ].
If we define our σ−algebra and measure as the Lebesgue ones, this space is called the Lebesgue
space Lp (X), where an integral p−norm can be defined.
5.2. INTEGRATION 67

§§ 5.2.2 Lebesgue Integration

Note:
In this section the differential dx will actually indicate the Lebesgue measure µ (dx) used previously,
unless stated otherwise.

Theorem 5.4. Let f : E −→ F be a measurable function over E.

Given
F+∞ = x ∈ E| f (x) = +∞ ∧ F−∞ = x ∈ E| f (x) = −∞
Assuming E ⊂ X, with (X, L, µ) a Lebesgue measure space, we have that

µ (F+∞ ) = µ (F−∞ ) = 0

Proof. We can immediately say that

\
F+∞ = Fk ∈ L
k≥0

Letting r > 0 we will indicate with 1r (x) the set function of the set F+∞ ∩ Br (0), therefore we have
that
f + (x) ≥ k1r (x) ∀k ∈ N
Therefore ˆ ˆ
1
µ (F+∞ ∩ Br (0)) = 1r (x) dx ≤ f + (x) dx −→ 0
k E

The proof is analogous for F−∞

Theorem 5.5. Let (X, L, µ) be a measure space, where L is the Lebesgue σ−algebra and µ is
the Lebesgue measure. Given a function f ∈ L1 (X) we have that
ˆ
f (x) dx = 0 ⇐⇒ f ∼ 0 (5.25)
X

Proof. Let F0 = x ∈ X| f (x) > 0 = k≥0 F1/k .

T
Since f (x) > 1/k, ∀x ∈ F1/k , we have that, ∀k ∈ N
ˆ
µ(F1/k ) ≤ f (x) dx = 0
X

Through induction, we obtain that µ(F0 ) = 0

Theorem 5.6 (Monotone Convergence (B. Levi)). Let (f )k be a sequence of measurable functions
over a Borel space E, such that

0 ≤ f1 (x) ≤ · · · ≤ fk (x) ≤ · · · ∀x ∈ F ⊂ E, µ(F ) = 0

CHAPTER 5. INTEGRAL ANALYSIS 68

If fk (x) → f (x), we have that

ˆ ˆ
f (x) dx = lim fk (x) dx (5.26)
E k→∞ E

Or, in another notation ˆ ˆ

fk (x) dx → f (x) dx (5.27)
E E

Proof. Let F0k = 0 < y < fk (x) and F0 = 0 < y < f (x) be two sets defined as seen. They are all
measurable since fk (x), f (x) are measurable, and due to the monotony of fk (x) we have that
∞
G
F01 ⊂ F02 ⊂ · · · ⊂ F0k ⊂ · · · ∧ F0 = F0k
k=1

Due to σ−additivity of the measure function, we have that F0 is measurable, and that
∞
X
µ (F0 ) = µ (F0k ) ∴ µ (F0 ) = lim µ (F0k )
k−→∞
k=1

Notation (For Almost All). We now introduce a new (unconventional) symbol in order to avoid writing
too much, which would complicate the already difficult to understand theorems.
In order to indicate that we’re picking almost all elements of a set we will use a new quantifier, which
means that we’re picking all elements of a null measure subset of the set in question. The quantifier in
question will be the following
∀† (5.28)

Corollary 5.2.1. Let fk (x) be a sequence of non-negative measurable functions over a measurable set
E, then ∀† x ∈ E ˆ X Xˆ
fk (x) dx = fk (x) dx (5.29)
E k≥0 k≥0 E

Theorem 5.7 (Fatou). Let fk (x) be a sequence of measurable functions over a measurable set
E, such that ∀† x ∈ E ∃Φ(x) measurable : fk (x) > Φ(x), then
ˆ ˆ
lim inf fk (x) dx ≤ lim inf fk (x) dx
E k→∞ k→∞ E

Analogously happens with the lim sup of the sequence

Proof. Let hk (x) = fk (x) − Φ(x) ≥ 0 ∀† x ∈ E and gj (x) = infk≥k hk (x), then ∀k ≥ j we have
ˆ ˆ
gj (x) dx ≤ hk (x) dx
E E
5.2. INTEGRATION 69

It’s also (obviously) true taking the lim sup of the RHS, and for the theorem on the monotone conver-
gence, we have that
ˆ ˆ ˆ
lim gj (x) dx = lim gj (x) dx ≤ hk (x) dx
E j→∞ j→∞ E E
∴ lim gj (x) = sup gj (x) = sup inf hk (x) = lim inf hk (x)
j→∞ j j k≥j k→∞

Theorem 5.8 (Dominated Convergence (Lebesgue)). Let h(x) ≥ 0 be a measurable function on

the measurable set E such that for a sequence of measurable functions fk (x) we have that

|fk (x)| ≤ h(x) ∀† x ∈ E

And
f (x) = lim fk (x) ∀† x ∈ E
k→∞

Then ˆ ˆ
f (x) dx = lim fk (x) dx
E k→∞ E

Proof. By definition we have that −h(x) ≤ fk (x) ≤ h(x) ∀† x ∈ E, and we can apply Fatou’s theorem
ˆ ˆ ˆ ˆ
f (x) dx ≤ lim inf fk (x) dx ≤ lim sup fk (x) dx ≤ f (x) dx
E k→∞ E k→∞ E E

Corollary 5.2.2. Let E be a measurable set such that µ (E) < ∞ and let fk (x) be a sequence of
functions in E such that |fk (x)| ≤ M ∀† x ∈ E and fk (x) → f (x), ∀† x ∈ E. Then the theorem (5.8)
is valid.

Example 5.2.1. Take the sequence of functions fk (x) = kxe−kx over E = [0, 1]. We already know
that fk (x) −→ f (x) = 0 for x ∈ E, but fk (x) 6⇒ f (x) in E.
We have that
sup fk (x) = e−1 = h(x) 6= f (x)
E

We have that h(x) is measurable in E and we can apply the theorem (5.8)

Definition 5.2.7 (Carathéodory Function). Let (X, L, µ) be a measure space and A ⊂ Rn . f : X ×A −→

R is a Carathéodory function iff f (xµ , aν ) ∈ C(A) ∀aν ∈ A and f (xµ , aν ) ∈ M(X) ∀† xµ ∈ X

Definition 5.2.8 (Locally Uniformly Integrably Bounded). Let f : X × A −→ R be a Carathéodory

function. It’s said to be locally uniformly integrably bounded if ∀aν ∈ A ∃haν : X −→ R
measurable, and ∃B (aν ) ⊂ A, such that

∀y ν ∈ B (xµ ) |f (xµ , y ν )| ≤ haν (xµ )

Note that if µ is a finite measure, then f bounded =⇒ f locally uniformly integrably bounded or LUIB.
CHAPTER 5. INTEGRAL ANALYSIS 70

Theorem 5.9 (Leibniz’s Derivation Rule). Let (X, F, µ) be a measure space and A ⊂ Rn an open
set. If f : X × A −→ R is a LUIB Carathéodory function we can define
ˆ
µ
g(a ) = f (xν , aµ ) dµ (xσ ) ∈ C(A)
X

Then
∂xµ f (xν , aσ ) ∈ C(A)
Is LUIB, and therefore
g(aµ ) ∈ C 1 (A)
And ˆ
∂µ g = ∂aµ f (aν , xσ ) dµ (xγ )
X
In other terms ˆ ˆ
∂aµ f (aν , xσ ) dµ(xγ ) = ∂aµ f (aν , xσ ) dµ(xγ ) (5.30)
X X

Proof. Since f is a LUIB Carathéodory function we have that ∃haµ (xν ) : X −→ R and B (aµ ) ⊂ A :
∀y µ ∈ B (aν )
|f (y µ , xν )| ≤ haµ (xν )
Therefore ˆ
|g(aµ )| ≤ haµ (xν ) dµ(xσ ) < ∞
X

Now take a sequence (aµ )n : (aµ )n → aµ , then f ∈ C(A) =⇒ f (aµn , xν ) → f (aµ , xν ) ∀† xµ ∈

X, ∀aµn ∈ B (aµ )
∴ ∃N ∈ N : ∀n ≥ N |f (aµn , xν )| ≤ haµ (xν )
Then ˆ ˆ
g(aµn ) = f (aµn , xν ) dµ(xσ ) → f (aµ , xν ) dµ(xσ ) = g(aµ )
X X

Since f is differentiable and its derivative is measurable, we have for the mean value theorem

f (aµ + teµ , xν ) − f (aµ , xν ) = t∂µ f (ξ ν (t, xσ ), xγ )

If ξ µ (t, xν ) ∈ B (aµ ) we have that

|t∂µ f (ξ ν (t, xσ ), xγ )| ≤ haµ (xν )

And therefore ˆ
g(aµ + teµ ) − g(aµ ) 1
= t∂µ f (ξ ν (t, xσ ), xγ ) dµ(xδ )
t t X

For t → 0 ∂µ f (ξ ν , xσ ) → ∂µ f (aν , xσ ), and the LHS is simply the gradient of g. Therefore for theorem
(5.8) ˆ ˆ
ν ∂ ν σ γ
∂µ g(a ) = f (a , x ) dµ(x ) = ∂µ f (aν , xσ ) dµ(xγ )
∂aµ X X
5.3. CALCULUS OF INTEGRALS IN R2 AND R3 71

§ 5.3 Calculus of Integrals in R2 and R3

§§ 5.3.1 Double Integration
Theorem 5.10. Let E ⊂ R2 and F ⊂ R3 . Define Ex := { y ∈ R| (x, y) ∈ E} the sections of E
parallel to the y axis, then ˆ
µ(E) = µ1 (Ex ) dy (5.31)
R
Where with µi we indicate thei−dimensional measure on Rn .
Analogously, we define Fz := (x, y) ∈ R2 (x, y, z) ∈ F then
ˆ
µ(F ) = µ2 (Fz ) dz (5.32)
R

If we define Fxy := { z ∈ R| (x, y, z) ∈ F } we have

¨
µ(F ) = µ1 (Fxy ) dx dy (5.33)
R2

Proof. Let A ⊂ R2 open, and let Yk ⊂ R2 be rectangles such that

Y1 ⊂ Y2 ⊂ Y3 ⊂ · · ·
G∞
A= Yk
k=1

Then, due to σ−additivity, we have

ˆ
µ2 (A) = lim µ2 (Yk ) = lim µ1 (Ykx ) dx
k→∞ k→∞ R

But
Y1x ⊂ Y2x ⊂ · · ·
G∞
Ax = Ykx
k=1

Due to σ−additivity and the Beppo-Levi theorem we have that

ˆ ˆ
µ1 (Ax ) dx = lim µ1 (Ykx ) dx
R k→∞ R

Let E ⊂ R2 be a measurable set. Define a sequence of compact sets Ki and a sequence of open sets
Aj such that
K1 ⊂ · · · ⊂ Kj ⊂ E ⊂ Aj ⊂ · · · ⊂ A1
We have that limj→∞ µ2 (Aj ) = limj→∞ µ2 (Kj ) = µ2 (E) and that Kjx ⊂ E ⊂ Ajx .
From the previous derivation we can write that
ˆ
lim (µ1 (Ajx ) − µ1 (Kjx )) dx = 0
j→∞ R
CHAPTER 5. INTEGRAL ANALYSIS 72

Building a sequence of non-negative functions fj (x) = µ1 (Ajx )−µ1 (Kjx ) we have that fj (x) ≤ fj−1 (x)
and due to Beppo-Levi we have that
ˆ ˆ
lim fj (x) dx = lim fj (x) dx
j→∞ R R j→∞

And therefore µ1 (Kjx ) = µ1 (Ajx ), and

ˆ ˆ ˆ
∀† x ∈ R µ2 (Kj ) = µ1 (Kjx ) dx ≤ µ1 (Ex ) dx ≤ µ1 (Ajx ) = µ2 (Aj )
R R R

Theorem 5.11 (Fubini). Let f (x, y) be a measurable function in R2 , then

1. ∀† x ∈ R y 7→ f (x, y) is measurable in R
´
2. g(x) = R f (x, y) dy is measurable in R
˜ ´ ´
3. R2 f (x, y) dx dy = R R f (x, y) dx dy

Proof. Let f (x, y) ≥ 0. Defining F0 := { (x, y) ∈ E × R| 0 < z < f (x, y)} ⊂ R3 , we have that F0 is
measurable, and ¨
µ3 (F0 ) = f (x, y) dx dy
R2

But F0x is also measurable ∀† x ∈ R and therefore

ˆ ˆ ˆ
µ3 (F0 ) = µ2 (F0x ) dx = f (x, y) dx dy
R R R

Theorem 5.12 (Tonelli). Let f (x, y) be a measurable function and E ⊂ R2 be a measurable set.
If one of these integrals exists, the others also exist and have the same value
¨ ˆ ˆ  ˆ ˆ 
f (x, y) dx dy f (x, y) dx dy f (x, y) dy dx
R2 R R R R

Theorem 5.13 (Integration Over Rectangles). Let R = [a, b] × [c, d] ⊂ R2 be a rectangle, and f (x, y)
a measurable function over R. Then

´d
1. If ∀† x ∈ [a, b] ∃G(x) = c
f (x, y) dy, the function G(x) is measurable in [a, b] and
¨ ˆ b ˆ b ˆ d
f (x, y) dx dy = G(x) dx = f (x, y) dy dx
R a a c
5.3. CALCULUS OF INTEGRALS IN R2 AND R3 73

´b
2. If ∀† y ∈ [c, d] ∃F (y) = a
f (x, y) dx, the function F (y) is measurable in [c, d] and
ˆ ˆ d ˆ d ˆ b
f (x, y) dx dy = F (y) dy = f (x, y) dx dy
R2 c c a

If both are true, then

ˆ ˆ b ˆ d ˆ d ˆ b
f (x, y) dx dy = dx f (x, y) dy = dy f (x, y) dx (5.34)
R a c c a

Definition 5.3.1 (Normal Set). A set E ⊂ R2 is said to be normal with respect to the x axis if

E = (x, y) ∈ R2 a ≤ x ≤ b α(x) ≤ y ≤ β(x)



The definition is analogous for the other axes.

Theorem 5.14 (Integration over Normal Sets). Let E ⊂ R2 be a normal set with respect to the x
axis, and f (x, y) is a measurable function over E. Then
ˆ ˆ b ˆ β(x)
f (x, y) dx dy = dx f (x, y) dy (5.35)
E a α(x)

Theorem 5.15 (Dirichlet Inversion Formula). Take the triangle T := (x, y) ∈ R2 a ≤ y ≤ x ≤ b .



It can be considered normal with respect to both axes, and we can use the inversion formula
¨ ˆ b ˆ x ˆ b ˆ b
f (x, y) dx dy = dx f (x, y) dy = dy f (x, y) dx (5.36)
T a a a y

§§ 5.3.2 Triple Integration

Theorem 5.16 (Wire Integration). Let E ⊂ R3 be a normal set with respect to the z axis. If
f (x, y, z) is measurable in E we have
˚ ¨ ˆ g(x,y)
f (x, y, z) dx dy dz = dx dy f (x, y, z) dz (5.37)
E D h(x,y)

This is called the wire integration formula

Theorem 5.17 (Section Integration). Let F ⊂ R3 be a measurable set bounded by the planes
z = a and z = b with a < b. Taken z ∈ [a, b] we can define Fz and we have
˚ ˆ b ¨
f (x, y, z) dx dy dz = dz f (x, y, z) dx dy (5.38)
F a Fz

This is called the section integration formula

CHAPTER 5. INTEGRAL ANALYSIS 74

Theorem 5.18 (Center of Mass). Take a plane E ⊆ R2 with surface density ρ(x, y) > 0. We define
the total mass M as follows ¨
M= ρ(x, y) dx dy (5.39)
E
The coordinates of the center of mass will be the following
¨
1
xG = ρ(x, y)x dx dy
M E
¨ (5.40)
1
yG = ρ(x, y)y dx dy
M E

Theorem 5.19 (Moment of Inertia). Taken the same plane E, we define the moment of inertia
with respect to a line r as the following integral
¨
2
Ir = ρ(x, y) (d(pµ , r)) dx dy (5.41)
E

Where d(pµ , r) is the distance function between the point (x, y) and the rotation axis r.
Both formulas are easily generalizable in R3

§§ 5.3.3 Change of Variables

Definition 5.3.2 (Diffeomorphism). Let M, N ⊂ X be two subsets of a metric space X. The two sets are
∼
said to be diffeomorphic if ∃f : M −→ N an isomorphism such that f ∈ C 1 (M ) and f −1 ∈ C 1 (N ).
The application f is called a diffeomorphism.
Two diffeomorphic sets are indicated as follows

M 'N

∼
Theorem 5.20. Let A, B ⊂ Rn be two open sets and ϕµ : A −→ B a diffeomorphism, such that

ϕµ (E) = F

If f : E ⊂ B −→ R is measurable, we have that

ˆ ˆ ˆ
f (y µ ) dy µ = f (ϕµ (xν )) det ∂µ ϕν dxµ = f (ϕµ ) det ∂µ ϕν dxµ
E ϕ−1 (E) µν F µν

∼
Theorem 5.21 (Change of Variables). Let ϕµ : Rn −→ Rn be a diffeomorphism such that

ϕµ (xν ) = xµ ∀kxµ kµ > 1

And f : Rn −→ R a function such that supp f = K ⊂ Rn is a compact set. If f is measurable,

we have that ˆ ˆ
f (y µ ) dy µ = f (ϕµ (xν )) det ∂µ ϕν dxµ (5.42)
Rn Rn µν
5.3. CALCULUS OF INTEGRALS IN R2 AND R3 75

Proof. Take n = 2 without loss of generality. We can immediately write that

ˆ y1
g(y 1 , y 2 ) = f (η, y 2 ) dη
−∞

Then, for the fundamental theorem of integral calculus

∂1 g(y 1 , y 2 ) = f (y 1 , y 2 )

Taken c ∈ R, c > 1 : K ⊂ Q = [−c, c] × [−c, c], we have that ϕµ (xν ) = δνµ ∀kxµ kµ > 1 ∧ f (xµ ) =
0 ∀xµ ∈
/ Q.
Therefore f (ϕµ ) = 0 also and we have
ˆ ˆ ˆ
f (ϕµ ) det ∂µ ϕν dxγ = f (ϕµ ) det ∂µ ϕν dxγ = ∂1 g(ϕµ ) det ∂µ ϕν dxγ
Rn µν Q µν Q µν

But we have that

g(y µ ) = 0 ∀ y 1 ≥ c ∨ y 1 < −c
Define the following matrix Hµν
∂µ g(φγ )
 
Hµν =
∂µ ϕ2
Then we have that
det Hµν = ∂1 g(ϕµ ) det ∂µ ϕν
µν µν
µ µ
Writing g(ϕ ) = G(x ) we have

det Hµν = ∂1 G∂2 ϕ2 − ∂2 G∂1 ϕ2

µν

Thanks to the integration formula (5.34) we can then write

ˆ ˆ c ˆ c
γ 2
det Hµν dx = dx ∂1 G∂2 ϕ2 dxν
Q µν −c −c

Integrating by parts we get

ˆ ˆ c ˆ c
c c
det Hµν dxγ = G∂2 ϕ2 −c
− 2 2
G∂21 ϕ dx1 − G∂1 ϕ2 −c
− 2 2
G∂12 ϕ dx2
Q µν −c −c

But ∀xµ ∈ ∂Q ϕµ (xν ) = xµ =⇒ G(−c, x2 ) = g(−c, x2 ) = 0 ∧ G(c, x2 ) = g(c, x2 )

ˆ ˆ
γ
∴ det Hµν dx = f (xµ ) dxγ
Q µν Q

Theorem 5.22 (Common Coordinate Transformation in R2 and R3 ). 1. Polar Coordinates

(
x(ρ, θ) = ρ cos θ ρ ∈ R+
ϕµ (xν ) = (5.43a)
y(ρ, θ) = ρ sin θ θ ∈ [0, 2π)
CHAPTER 5. INTEGRAL ANALYSIS 76

 
ν cos θ −ρ sin θ
∂µ ϕ =
sin θ ρ cos θ (5.43b)
det ∂µ ϕν = ρ
µν

2. Spherical Coordinates

ρ ∈ R+

x(ρ, θ, φ) = ρ sin φ cos θ

µ ν
ϕ (x ) = y(ρ, θ, φ) = ρ sin φ sin θ θ ∈ [0, 2π) (5.44a)

z(ρ, θ, φ) = ρ cos φ φ ∈ [0, π]


 
sin φ cos θ −ρ sin φ sin θ ρ cos φ cos θ
∂µ ϕν =  sin φ sin θ ρ sin φ cos θ ρ cos φ sin θ 
cos φ 0 −ρ sin φ (5.44b)
det ∂µ ϕν = ρ2 sin φ
µν

3. Cylindrical Coordinates

ρ ∈ R+

x(ρ, θ, z) = ρ cos θ

µ ν
ϕ (x ) = y(ρ, θ, z) = ρ sin θ θ ∈ [0, 2π) (5.45a)

z(ρ, θ, z) = z z∈R


det ∂µ ϕν = ρ (5.45b)
µν

Definition 5.3.3 (Rotation Solids). Let D ⊂ R2 be a bounded measurable set contained in the half-plane
y = 0, x > 0. Suppose we let D “pop up” into R3 through a rotation by an angle θ0 around the z axis.
What has been obtained is a rotation solid E ⊂ R3 . We have that
˚ ¨ ˆ θ0 ¨ ¨
µ(E) = dx dy dz = ρ dρ dθ dz = θ0 ρ dρ dz = θ0 x dx dy (5.46)
E D 0 D D

Or
µ(E) = θ0 xG µ2 (D)

Theorem 5.23 (Guldino). The measure of a rotation solid is given by the measure of the rotated
figure times the circumference described by the center of mass of the solid.
This is exactly the previous formula.
5.3. CALCULUS OF INTEGRALS IN R2 AND R3 77

§§ 5.3.4 Line Integrals

Definition 5.3.4 (Line Integral of the First Kind). Given a scalar field f : A ⊂ R3 −→ R and a smooth
curve {γ} ⊂ R3 , we define the line integral of the first kind as follows
ˆ ˆ b
dγ µ
f ds = f (γ µ ) dt (5.47)
γ a dt µ

Theorem 5.24 (Center of Mass of a Curve). Given a curve γ µ : [a, b] −→ R3 with linear mass
density m : {γ} −→ R, we define the total mass of γ as follows
ˆ ˆ b
dγ µ
M= m ds = m(γ µ ) dt (5.48)
γ a dt µ

The center of mass is then defined as follows

ˆ
µ 1
xG = xµ m(xν ) ds (5.49)
M γ

Definition 5.3.5 (Line Integral of the Second Kind). Given a vector field f µ : A −→ R3 and a smooth
curve γ µ : [a, b] −→ A ⊂ R3 we define the line integral of the second kind as follows
ˆ ˆ b
µ dγµ
f Tµ ds = f µ (γ ν ) dt (5.50)
γ a dt

Defining a differential form ω = f µ dxµ we can also see this integral as follows
ˆ ˆ
ω= f µ Tµ ds (5.51)
γ γ

Where T µ is the tangent vector of the curve

Definition 5.3.6 (Conservative Field). Let f µ : A −→ R3 be a vector field such that f µ ∈ C 1 (A) and A
is open and connected. This field is said to be conservative, if ∀xµ ∈ A

∃U (xµ ) ∈ C 2 (A) : f µ = −∂ µ U (5.52)

The function U (xµ ) is called the potential of the field.

Theorem 5.25 (Line Integral of a Conservative Field). Given a conservative field f µ : A −→ R3 and
a smooth curve {γ} ⊂ A, γ µ : [a, b] −→ R3 with A open and connected, we have that
ˆ
f µ Tµ ds = U (γ(a)) − U (γ(b)) (5.53)
γ

Where U (xµ ) is the potential of the vector field.

CHAPTER 5. INTEGRAL ANALYSIS 78

Definition 5.3.7 (Rotor). Given a vector field f µ : A −→ R3 with f µ ∈ C 1 (A), we define the rotor of
the vector field as follows
rot(f µ ) = µνγ ∂ ν f γ (5.54)

Theorem 5.26. Given f µ a conservative vector field on an open connected set A, we have that

µνγ ∂ ν f γ = 0 (5.55)
Alternatively, if γ µ : [a, b] −→ R3 is the parameterization of a smooth closed curve, we have
that ˛
f µ Tµ ds = 0 (5.56)
γ

§§ 5.3.5 Surface Integrals

Definition 5.3.8 (Area of a Surface). Given rµ : K ⊂ R2 −→ Σ ⊂ R3 a smooth surface, we have that
given its metric tensor gµν (u, v) we have that
ˆ ¨ q ¨ p
µ (Σ) = dσ = det gµν du dv = EG − F 2 du dv (5.57)
Σ K µν K

For a cartesian surface S we have that

ˆ ¨ r  2
µ (S) = ds = 1 + k∂µ f kµ dx dy (5.58)
S K

Definition 5.3.9 (Rotation Surface). Given a smooth curve γ µ : [a, b] −→ R3 , the rotation of this curve
around the z−axis generates a smooth surface Σ with the following parameterization
 1
γ (t) cos θ

µ
r (t, θ) = γ 2 (t) sin θ (t, θ) ∈ [a, b] × [0, θ0 ] (5.59)

 3
γ (t)
The area of a rotation surface is calculated as follows
ˆ b
s 2  2 2
1 dγ 1 dγ
µ (Σ) = θ0 γ (t) + dt (5.60)
a dt dt

Theorem 5.27 (Guldino II). Given Σ a smooth rotation surface defined as before, we have that
its area will be ˆ
µ (Σ) = θ0 x1 ds = θ0 x1G Lγ (5.61)
γ

Where x1G is the first coordinate of the center of mass of the curve, calculated as follows
ˆ
1 1
xG = x1 ds
Lγ γ
5.3. CALCULUS OF INTEGRALS IN R2 AND R3 79

Definition 5.3.10 (Surface Integral). Given a smooth surface Σ ⊂ R3 with parameterization rµ : K −→

Σ and a scalar field h : R3 −→ R, we define the surface integral of h as follows
ˆ ¨ q
h(xµ ) dσ = h(rµ ) det gµν du dv (5.62)
Σ K µν

If Σ is a cartesian surface, we have

ˆ ¨ q
µ 2
h(xµ ) dσ = h(x1 , x2 , f ) 1 + (k∂µ f k ) dx dy (5.63)
Σ K

Definition 5.3.11 (Center of Mass of a Surface). Given a smooth surface Σ with parameterization
rµ (u, v) and mass density δ, we define its total mass as follows
ˆ
M= δ dσ (5.64)
Σ

Its center of mass xµG will be calculated as follows

ˆ
µ 1
xG = xµ δ(xν ) dσ (5.65)
M Σ

Definition 5.3.12 (Moment of Inertia of a Surface). Given a smooth surface Σ with parameterization
rµ (u, v) and mass density δ we define its moment of inertia around an axis r, I, as the following
integral ˆ
2
I= δ(xµ ) (d(pµ , r)) dσ pµ ∈ Σ (5.66)
Σ

Definition 5.3.13 (Orientable Surface). A smooth surface with parameterization rµ : K ⊂ R2 −→ Σ ⊂

R3 is said to be orientable if ∀γ : [a, b] −→ Σ smooth closed curve, we have, given nµ the normal
vector of the surface
nµ (γ ν (a)) = nµ (γ ν (b)) (5.67)
Another way of formulating it is
nµ (xν ) ∈ C(K) (5.68)

Definition 5.3.14 (Boundary of a Surface). Given a smooth surface as before, we define the boundary
∂Σ as follows
∂Σ = Σ \ Σ (5.69)
Note how, given the parameterization rµ , we have rµ (∂K) = ∂Σ
Definition 5.3.15 (Closed Surface). A surface Σ ⊂ R3 is said to be closed iff ∂Σ = {}
Definition 5.3.16 (Flux). Given a vector field f µ : A ⊂ R3 −→ R3 and a smooth orientable surface
Σ ⊂ A, we define the flux of the vector field f µ on the surface as follows
ˆ ¨
µ µ
ΦΣ (f ) = f nµ dσ = f µ (rν )µγσ ∂1 rγ ∂2 rσ du dv (5.70)
Σ K
CHAPTER 5. INTEGRAL ANALYSIS 80

§ 5.4 Integration in C
Definition 5.4.1 (Piecewise Continuous Function). Let γ : [a, b] −→ C be a piecewise continuous curve
such that {γ} ⊂ D ⊂ C, and f : D −→ C, f ∈ C(D). Then the function (f ◦ γ) γ 0 (t) : [a, b] −→ C
is a piecewise continuous function
Definition 5.4.2 (Line Integral in C). Let γ : [a, b] −→ D ⊂ C be a piecewise continuous curve and
f : D −→ C a measurable function f ∈ C(D).
We define the line integral over γ the result of the application of the integral operator K̂γ [f ], where
ˆ ˆ b
K̂γ [f ] = f (z) dz = (f ◦ γ) γ 0 (t) dt (5.71)
γ a

Where ∀† z ∈ {γ} f (z) is defined

Theorem 5.28 (Properties of the Line Integral). Let z, w, t ∈ C, f, g ∈ M(C) and {γ}, {η}, {κ} three
smooth curves, then
1. K̂γ [zf + wg] = z K̂γ [f ] + wK̂γ [g]

2. γ ∼ η =⇒ K̂γ [f ] = K̂η [f ]

3. γ = η + κ =⇒ K̂γ [f ] = K̂η+κ [f ] = K̂η [f ] + K̂κ [f ]

4. K̂γ+w [f (z)] = K̂γ [f (z + w)]

Notation. If a measurable function f (z) has the same value of the integral for different curves between
two points z1 , z2 ∈ C, we will write directly
ˆ ˆ z2
f (z) dz = f (z) dz
γ z1

Theorem 5.29 (Darboux Inequality). Let f : D −→ C be a measurable function and γ : [a, b] −→

{γ} ⊂ D ⊆ C piecewise smooth. Then
ˆ
f (z) dz ≤ Lγ sup kf (z)k
γ z∈{γ}

Proof. The proof is quite straightforward using the definition given for the line integral
ˆ ˆ b ˆ b
0
f (z) dz = (f ◦ γ) γ (t) dt ≤ k(f ◦ γ) γ 0 (t)k dt ≤
γ a a
ˆ b
≤ sup kf (z)k kγ 0 (t)k dt = Lγ sup kf (z)k
z∈{γ} a z∈{γ}
5.4. INTEGRATION IN C 81

§§ 5.4.1 Integration of Holomorphic Functions

Definition 5.4.3 (Primitive). Let f : D −→ C and F : D −→ C be two functions and D ⊂ C an open
and connected set. F (z) is said to be the primitive function or antiderivative of f in D if

dF
= f (z) ∀z ∈ D (5.72)
dz

Notation. Given a closed curve γ and a measurable function f (z) we define the following notation
ˆ ˛
f (z) dz = f (z) dz
γ γ

Theorem 5.30 (Existence of the Primitive Function). Let f : D −→ C f ∈ C(D) with D ⊂ C open
and connected. Then these statements are equivalent
1. ∃F : D −→ C : F 0 (z) = f (z)
´z ´
2. ∀z1 , z2 ∈ D, ∀{γ} ⊂ D piecewise smooth f (z) dz = z12 f (z) dz
γ
¸
3. ∀γ : [a, b] −→ {γ} ⊂ D closed piecewise smooth γ f (z) dz = 0

Proof. 1 =⇒ 2. As with the hypothesis we have that ∃F : D −→ C : F 0 (z) = f (z) ∀z ∈ D. Given

two points z1 , z2 ∈ D and taken a smooth curve γ : [a, b] −→ D : γ(a) = z1 ∧ γ(b) = z2 . Therefore
ˆ ˆ b ˆ b
f (z) dz = (f ◦ γ) γ 0 (t) dt = (F 0 ◦ γ) γ 0 (t) dt
γ a a

The result of the integral is obviously F (z2 ) − F (z1 ), therefore we can immediately write that, if
ˆ ˆ z2
∃F : D −→ C : F 0 (z) = f (z) =⇒ f (z) dz = f (z) dz
γ z1

2 =⇒ 1 Taken a point z0 ∈ D, any point z ∈ D can be connected with a polygonal to z0 since D is

connected. The integral of f over this polygonal is obviously path-independent, hence we can define
the following function ˆ z
F (z) = f (w) dw
z0

Since D is open we can define δz ∈ R, δz > 0 ∧ ∃Bδ1 (z) ⊂ D. Taken ∆z ∈ C : k∆zk < δ1 we have
that ˆ z+∆z
F (z + ∆z) − F (z) = f (w) dw
z

Dividing by ∆z and taking the limit as ∆z → 0 we have that using the Darboux inequality we get that
ˆ z+∆z
F (z + ∆z) − F (z) 1
− f (z) = f (w) dw ≤ 
∆z k∆zk z
CHAPTER 5. INTEGRAL ANALYSIS 82

2 =⇒ 3. Taken an arbitrary piecewise smooth curve γ and z1 6= z2 ∈ {γ}. We can now find two
curves such that γ(t) = γ1 (t) − γ2 (t). Since the integral of f is path independent, we get
ˆ ˆ ˆ
f (z) dz = f (z) dz − f (z) dz = 0
γ γ1 γ2

3 =⇒ 2 is exactly as before but with the opposite reasoning.

Example 5.4.1. Let’s calculate the integral of functions fn (x) = z −n n ∈ N for a closed simple
piecewise smooth curve γ such that 0 ∈
/ {γ}.
For n > 1 we have that f ∈ C(D) where D = C \ {0}, and we have that
ˆ
1 z −(n−1)
dz = − +w w ∈C
zn n−1

Therefore, for every closed simple piecewise smooth curve γ : 0 ∈

/ {γ} we have
˛
1
n
dz = 0
γ z

For n = 1 we still have that f ∈ C(D) but @F (z) : D −→ C primitive of f1 (z), but there exists one in
the domain G of holomorphy of the logarithm.
Although we have that G ⊂ D, and we can take a curve γ : 0 ∈ extr γ, and therefore {γ} ⊂ G and
we have that ˛
1
dz = 0
γ z
If we otherwise have 0 ∈ γ ◦ the integral is non-zero.
Take a branch of the logarithm σ and a curve η has only one point of intersection with such branch
at zi = u0 eiα . Taken η(a) = η(b) = u0 eiα , we define η : [a + , b + ] −→ C with  > 0 : η (t) =
η(t) ∀t ∈ [a + , b + ], then ˛ ˛
1 1
dz = lim dz
η z →0 η z

Therefore, ∀z ∈ C \ {σ} we have that

d log z 1
=
dz z
And therefore ˛
1
dz = log (η(b − )) − log (η (a + ))
η z
For  → 0 we have
ˆ
1
dz = (log(u0 ) + i(α + 2π)) − (log(u0 ) + iα) = 2πi
η z

√
Example 5.4.2. Let’s calculate the integral of f (z) = z along a closed simple piecewise smooth
curve γ : [a, b] −→ C : 0 ∈ γ ◦ and it intersects the line σα = u0 eiα , where
√ √ θ
z= rei 2 r ∈ R+ , θ ∈ (α, α + 2π], α ∈ R
5.4. INTEGRATION IN C 83

Taken a parametrization γ(t) : γ(a) = γ(b) = u0 eiα we have that f (z) ∈ H(D) where D = C \ {σα }.
Proceding as before, we have ˛ ˛
√ √
z dz = lim z dz
γ →0 γ

Since it has a primitive in D we can write

˛
√ 2 √ γ (b−) 2 √ 3 2 √ 3 4 √ 3
lim z dz = lim z z = u0 u0 e 2 i(α+2π) − u0 u0 e 2 iα = − u0 u0 e 2 iα
→0 γ

3 →0 γ (a+) 3 3 3

Lemma 5.4.1. Taken a closed simple pointwise smooth curve γ : [a, b] −→ C and taken D = {γ}◦ ∪γ =
{γ}◦ and a function f ∈ H(D), for a finite cover of D, Q composed by squares Qj ∈ Q ∀j ∈ [1, N ] ⊂ N,
we have that

f (z) − f (zj ) df
∃zj ∈ Qj ∩ {γ}◦ : − <  ∀z ∈ Qj ∩ {γ}◦ \ {zj }
z − zj dz zj

Proof. Going by contradiction, let’s say that

∃ > 0 : @zj ∈ Qj ∩ {γ}◦

d
Taken a finite subcover Qn where diam(Qnj ) = 2n ∀Qj ∈ Q we can define for some k ∈ K ⊂ N
[
An = Qnk ∩ {γ}◦ ∀n ∈ N
k∈K

We have that An+1 ⊂ An , and taking a sequence (w)n ∈ {γ}◦ we have due to the compactness of
{γ}◦ that ∃(w)nj → w ∈ {γ}◦ . Since f ∈ H({γ}◦ ) we have that f is holomorphic in w, therefore

f (z) − f (w) df
∀ > 0 ∃δ > 0 : − <  ∀z ∈ Bδ (w) \ {w}
z−w dz w
√
2
Taken an ñ such that diam(Qñj ) = 2ñ d < δ we have that still w ∈ An ∀n ∈ N, and due to its closedness
we can also say
∃Nj ∈ N : ∀nj > Nj (w)nj ∈ An
Therefore
∃k0 ∈ N : w ∈ Qñk0 ∩ {γ}◦ ⊂ Añ

Theorem 5.31 (Cauchy-Goursat). Taken γ : [a, b] −→ C a closed simple piecewise smooth curve
and D = {γ} ∪ {γ}◦ and a function f ∈ H(D), we have
˛
f (z) dz = 0 (5.73)
γ
CHAPTER 5. INTEGRAL ANALYSIS 84

Proof. Using the previous lemma we can say that for a finite cover {γ}, Qj ∈ Q ∃zj ∈ Qj ∩ {γ}◦ and
a function 
 f (z) − f (zj ) − f 0 (zj ) z 6= zj

δj (z) = z − zj

0 z=z j

Which is countinuous and ◦

 δj (z) < ∀z ∈ Qj ∩ {γ} .
Taken a curve {ηj } = ∂ Qj ∩ {γ}◦ , and the expansion of f (z) in the region, we have that

f (z) = f (zj ) + f 0 (zj )(z − zj ) + δj (z)(z − zj )

˛ ˛ ˛ ˛
f (z) dz = (f (zj ) − zj f 0 (zj )) dz + f 0 (zj ) z dz + δj (z)(z − zj ) dz
ηj ηj ηj ηj

The first two integrals on the second line are null, and we have therefore
˛ ˛
f (z) dz = δj (z)(z − zj ) dz
ηj ηj

SN
By definition {γ} = j=1 {ηj } and therefore

˛ N ˛
X
f (z) dz = δj (z)(z − zj ) dz
γ j=1 ηj

Using the Darboux inequality we have immediately that

˛ N
X ˛ N
X √
f (z) dz ≤ δj (z)(z − zj ) dz ≤  2d(4d + Lj )
γ j=1 ηj j=1

Using the theorem on the Jordan curve, we have that ∃Qn ∈ Q such that {γ} ⊂ Qn . Taken diam(Qn ) =
D
˛ XN
√
f (z) dz ≤  2D(4D + L) → 0
γ j=1

Definition 5.4.4 (Simple Connected Set). An open set G ⊂ X with X some metric space, is said to be
simply connected iff ∀{γj } ⊂ G simple curves we have that γj ∼ 0.
γ ∼ 0 implies that the curve is homotopic to a point

Theorem 5.32 (Cauchy-Goursat II). Let G ⊂ C open and simply connected. Then, ∀f ∈ H(G), {γ} ⊂
G with γ simple closed and smooth
˛
f (z) dz = 0
γ
5.4. INTEGRATION IN C 85

Proof. 1. The curve γ doesn’t intersect itself.

˛ ˛
f (z) dz = f (z) dz = 0
γ 0

2. The curve γ intersects

Sn itself n − 1 times.
Then {γ} = k=1 {γk } with γk simple smooth non intersecting curves. Since {γk } ⊂ G ∀k =
1, · · · , n, {γk } ∼ 0, we have
˛ n ˛
X
f (z) dz = f (z) dz = 0
γ k=1 γk

Theorem 5.33. Let G ⊂ C be a simply connected open set. If f ∈ H(G), then there exists a
primitive for f (z)

§§ 5.4.2 Integral Representation of Holomorphic Functions

Definition 5.4.5 (Positively Oriented Curve). The parametrization of a curve in C is said to be positively
oriented if its parametrization is taken such the path taken results counterclockwise.
Notation. The integral over a closed positively oriented parametrization of a curve γ is indicated as
follows ffi

Theorem 5.34 (Cauchy Integral Representation). Taken a positively oriented closed simple piece-
wise smooth curve γ : [a, b] −→ C and a function f : G ⊂ C −→ C such that if D = {γ} ∪ {γ}◦ ⊂
G, f ∈ H(D), we have that
ffi
1 f (w)
f (z) = dw ∀w ∈ {γ}◦ (5.74)
2πi γ w − z

Proof. Taken γρ (θ) = z + ρeiθ such that γρ ∼ γ, {γρ } ⊂ {γ}◦ is a simple curve, we have
ffi ffi
f (w) f (w)
dw = dw
γ w−z γρ w − z

Then, using that ffi

1
dw = 2πi
γ w−z
We get ffi ffi
f (z) f (w) − f (z)
dw − 2πif (z) = dw
γ w−z γρ w−z
CHAPTER 5. INTEGRAL ANALYSIS 86

Since f ∈ H({γ}◦ ) we have that

∀ > 0 ∃δ > 0 : kz − wk < δ =⇒ kf (z) − f (w)k < 

Taken ρ < δ we get, using the Darboux inequality

ffi ffi
f (w) − f (z) f (w) − f (z)
dw ≤ 2π =⇒ dw = 0
γρ w−z γ w−z

Theorem 5.35 (Derivatives of a Holomorphic Function). Let D ⊂ C be an open set and f : D −→ C

a function f ∈ H(D), then f ∈ C ∞ (D) and
ffi
dn f n! f (w)
n
= dw (5.75)
dz 2πi γ (w − z)n+1

Where γ is a closed simple piecewise smooth curve such that z ∈ {γ}◦ and {γ} ⊂ D
Corollary 5.4.1. Let f ∈ H(D), then

dn f
∀n ∈ N ∈ H(D)
dz n

Theorem 5.36 (Morera). Let D ⊂ C be an open and connected set. Take f : D −→ C : f ∈ C(D).
Then, if ∀{γ} ⊂ D closed piecewise smooth
˛
f (z) dz = 0 =⇒ f ∈ H(D) (5.76)
γ

Proof. Since f ∈ C(D) ∃F (z) ∈ C 1 (D) : f (z) = F 0 (z). Since C 1 (C) ' H(C) we have that, due to
the previous corollary
dF
= f (z) ∈ H(D)
dz

Theorem 5.37 (Cauchy Inequality). Let f ∈ H(BR (z0 )) with z0 ∈ C. If kf (z)k ≤ M ∀z ∈ BR (z0 )

df n!M
≤ (5.77)
dz z0 Rn

dn f
Proof. Take γr (θ) = z0 + reiθ with θ ∈ [0, 2π], r > R, then the derivative dz n can be written using
z0
the Cauchy integral representation, since f ∈ H(Br (z0 ))
ffi
dn f n! f (w)
= dw
dz n z0 2πi γr (w − z0 )n+1
5.5. INTEGRAL THEOREMS IN R2 AND R3 87

Using the Darboux inequality we have then

dn f n! n!M
≤ sup kf (z)k ≤ n
dz n z0
n
r z∈{γr } r

Since r < R we therefore have

dn f n!M
≤
dz n z0 Rn

Theorem 5.38 (Liouville). Let f : C −→ C a function such that f ∈ H(C), i.e. whole. If
∃M > 0 : kf (z)k ≤ M ∀z ∈ C the function f (z) is constant

Proof. f ∈ H(C), kf (z)k ≤ M and we can write, taken γR (θ) = z + Reiθ with θ ∈ [0, 2π]
ffi
0 1 f (w)
f (z) = dz
2πi γR (w − z)2

For Darboux ffi

0 1 f (w) supz∈{γR } kf (z)k M
kf (z)k ≤ dz ≤ ≤
2π γR (w − z)2 R R
Since R > 0 is arbitrary, we can say directly that kf 0 (z)k = 0 and therefore f (z) is constant ∀z ∈ C.

Theorem 5.39 (Fundamental Theorem of Algebra). Take a polynomial Pn (z) ∈ Cn [z], where Cn [z]
is the space of complex polynomials with variable z and degree n. If we have
n
X
Pn (z) = ak z k , z, ak ∈ C, an 6= 0
k=0

We can say that ∃z0 ∈ C : Pn (z0 ) = 0

Proof. As an absurd, say that ∀z ∈ C, Pn (z) 6= 0. Then f (z) = 1/Pn (z) ∈ H(C).
Since limz→∞ Pn (z) = ∞, we have that kf (z)k ≤ M ∀z ∈ C, and limz→∞ f (z) = 0.
Therefore ∃R > 0 : ∀kzk > R, kf (z)k < 1. Since f ∈ H(C), we have that f ∈ C(B R (z)). Due to the
Liouville theorem we have that f (z) is constant

§ 5.5 Integral Theorems in R2 and R3

Theorem 5.40 (Gauss-Green). Given D ⊂ R2 a set with a piecewise smooth parameterization of

∂D and two functions α, β : A ⊆ R2 −→ R and D ⊂ A
¨ ˆ ¨ ˆ
∂x β dx dy = β(x, y) dy, ∂y α dx dy = − α(x, y) dx dy (5.78)
D ∂+D D ∂D
CHAPTER 5. INTEGRAL ANALYSIS 88

Theorem 5.41 (Stokes). Given D ⊂ R2 an open set with ∂D piecewise smooth and a vector
field f µ : A −→ R2 with D ⊂ A
ˆ ˆ
3µν ∂ µ f ν dx dy = f µ tµ ds (5.79)
D ∂+D

Where tµ is the vector tangent to ∂ + D

Theorem 5.42 (Gauss 1). Given D ⊂ Rn open set with ∂D piecewise smooth and a vector field
f µ : A −→ Rn with D ⊂ A ¨ ˆ
∂µ f µ dx dy = f µ nµ ds (5.80)
D ∂+D

Where nµ is the normal vector to ∂ + D

Theorem 5.43 (Stokes for Surfaces). Given a smooth surface Σ ⊂ R3 with parameterization rµ
and a vector field f µ : A −→ R3 with Σ ⊆ A
ˆ ˆ
nµ µνγ ∂ ν f γ dσ = f µ tµ ds (5.81)
Σ ∂+Σ

Where tµ is the tangent vector to the border of the surface

Theorem 5.44 (Useful Identities). Given u, v ∈ C 2 (Ω) and a vector field f µ ∈ C 2 (Ω, R3 )
ˆ ˆ
∂µ ∂ µ v dx dy dz = nµ ∂µ v dσ
ˆ Ω
ˆ
∂Ω
ˆ
µ µ
u∂µ f dx dy dz = − f ∂µ w dx dy dz + uf µ nµ dσ
ˆ Ω
ˆΩ
ˆ
∂Ω
(5.82)
u∂µ ∂ µ v dx dy dz = − ∂µ u∂ µ v dx dy dz + unµ ∂µ v dσ
ˆ Ω
ˆ Ω ∂Ω
µ µ µ µ
(u∂µ ∂ v − w∂µ ∂ u) dx dy dz = (un ∂µ v − wn ∂µ u) dσ
Ω ∂Ω

We can analogously write these theorems in the language of differential forms and manifolds, after
giving a couple of definitions
Definition 5.5.1 (Volume Element). Given a k−dimensional compact oriented manifold M with bound-
ary and ω ∈ Λk (M ) a k−differential form on M , we define the volume of M as follows
ˆ ˆ
V (M ) = dV = ω (5.83)
M M

Where dV is the volume element of the manifold, given by the unique ω ∈ Λk (M ), defined as follows

ω = f dxµ1 ∧ · · · ∧ dxµk (5.84)

With f an unique function.
For M ⊂ R3 with nµ as outer normal and ω ∈ Λ2 (M ) we can write immediately, by definition

ωµν v µ wν = nµ µνγ v ν wγ = dA
5.5. INTEGRAL THEOREMS IN R2 AND R3 89

Therefore
µ
dA = kµνγ v ν wγ k (5.85)
Which is the already known formula.
For a 2−manifold we can write immediately the following formulas

dA = n1 dy ∧ dz + n2 dz ∧ dx + n3 dx ∧ dy (5.86)

And, on M  1
n dA = dy ∧ dz

n2 dA = dz ∧ dx (5.87)

 3
n dA = dx ∧ dy

Theorem 5.45 (Gauss-Green-Stokes-Ostogradskij). Given M a smooth manifold with boundary, c

a p−cube in M and ω ∈ Λ(M ) we have
ˆ ˆ ˆ
dω = c? dω = ω (5.88)
c [0,1]p ∂c

In general, we can write ˆ ˆ

dω = ω (5.89)
M ∂M

Definition 5.5.2 (Gauss-Green, Differential Forms). Given M ⊂ R2 a compact 2−manifold with

boundary and two functions α, β : M −→ R with α, β ∈ C 1 (M ) defining

ω = α dx + β dy (5.90)

We have ˆ ˆ ˆ ¨  
∂β ∂α
α dx + β dy = ω= dω = − dx ∧ dy (5.91)
∂M ∂M M M ∂x ∂y

Proof. Take ω = α dx + β dy, then

 
∂β ∂α
dω = d(α dx + β dy) = − dx ∧ dy
∂x ∂y

Theorem 5.46 (Gauss, Differential Forms). Given M a 3−manifold smooth with boundary and
compact with outer normal nµ and a vector field f µ ∈ C 1 (M ), we have
ˆ ˆ
µ
∂µ f dV = f µ nµ dA (5.92)
M ∂M
CHAPTER 5. INTEGRAL ANALYSIS 90

Proof. Taken the following differential form

ω = f 1 dy ∧ dz + f 2 dz ∧ dx + f 3 dx ∧ dy

We have, using the formulas (5.87)

ω = f µ nµ dA
And
dω = ∂µ f µ dV
Therefore ˆ ˆ ˆ ˆ
µ
∂µ f dV = dω = ω= f µ nµ dA
M M ∂M ∂M

Theorem 5.47 (Stokes, Differential Forms). Given M ⊂ R3 a compact oriented smooth 2−manifold
with boundary with nµ as outer normal and tµ as tangent vector in ∂M , given a vector field
f µ ∈ C 1 (A) where M ⊂ A, we have
ˆ ˆ
nµ µνγ ∂ ν f γ dA = f µ tµ ds (5.93)
M ∂M

Proof. Taking the following differential form

ω = f µ dxµ

We have that

dω = (∂2 f 3 − ∂3 f 2 ) dy ∧ dz + (∂3 f 1 − ∂1 f 3 ) dz ∧ dx + (∂1 f 2 − ∂2 f 1 ) dx ∧ dy

Using the formulas (5.87) we have

dω = nµ µνγ ∂ ν f γ dA
Since in R2 we have tµ ds = dxµ we therefore have

f µ tµ ds = f µ dxµ = ω

And therefore ˆ ˆ ˆ ˆ
µ ν γ
n µνγ ∂ f dA = dω = ω= f µ tµ ds
M M ∂M ∂M

These last formulas are a good example on how they can be generalized through the use of
differential forms, bringing an easy way of calculus in Rn of the various integral theorems, all condensed
in one formula, the Gauss-Green-Stokes-Ostogradskij theorem
6 Sequences, Series and Residues

§ 6.1 Sequences of Functions

Definition 6.1.1 (Sequence of Functions). Let S be a set and (X, d) a metric space, a sequence of
functions is defined as follows
fn :S −→ (X, d)
(6.1)
s → fn (s)
Where, ∀n ∈ N a function f(n) : S −→ (X, d) is defined
Definition 6.1.2 (Pointwise Convergence). A sequence of functions (fn )n≥0 is said to converge point-
wise to a function f : S −→ (X, d), and it’s indicated as fn → f , if
∀ > 0, ∀x ∈ S ∃N (x) ∈ N : d(fn (x), f (x)) <  ∀n ≥ N (x) (6.2)
It can be indicated also as follows
lim (fn (x)) = f (x) (6.3)
n→∞

Definition 6.1.3 (Uniform Convergence). Defining an k·k∞ = supi≤n |·| we have that the convergence
of a sequence of functions is uniform, and it’s indicated as fn ⇒ f , iff
∀ > 0 ∃N ∈ N : d(fn (x), f (x)) <  ∀n ≥ N ∀x ∈ S (6.4)
Or, using the norm k·k∞
∀ > 0 ∃N ∈ N : kfn − f k∞ <  (6.5)

Theorem 6.1 (Continuity of Uniformly Convergent Sequences). Let (fn )n≥0 : (S, dS ) −→ (X, d) be a
sequence of continuous functions. Then if fn ⇒ f , we have that f ∈ C(S), where C(S) is the
space of continuous functions
Proof.

∀x ∈ S, ∃ > 0 : fn ⇒ f, ∴ ∀n ≥ N ∈ N : d(fn (x), f (x)) <
3
 (6.6)
fn ∈ C(S) =⇒ ∃δ > 0 : d(fn (x), fn (y)) < , ∀x, y ∈ S : dS (x, y) < δ
3
∴ d(f (x), f (y)) ≤ d(f (x), fn (x)) + d(fn (x), fn (y)) + d(fn (y), f (y)) <  ⇐⇒ dS (x, y) < δ

91
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 92

Theorem 6.2 (Integration of Sequences of Functions). Let (fn )n≥0 be a sequence of functions such
that fn ⇒ f Then we can define the following equality
ˆ b ˆ b ˆ b
lim fn (x) dx = lim fn (x) dx = f (x) dx (6.7)
n→∞ a a n→∞ a

Proof. We already know that in the closed set [a, b] we can say, since fn ⇒ f , that

∀ > 0 ∃N ∈ N : ∀n ≥ N kfn − f k∞ < (6.8)
b−a

Then, we have that

ˆ b ˆ b
∀n ≥ N fn (x) dx − f (x) dx ≤ kfn − f k∞ (b − a) <  (6.9)
a a

Theorem 6.3 (Differentiation of a Sequence of Functions). Define a sequence of functions as

fn : I −→ R, with fn (x) ∈ C 1 (I). If

1. ∃x0 ∈ I : fn (x0 ) → l

2. fn0 ⇒ g ∀x ∈ I

Then
fn (x) ⇒ f =⇒ ∀x ∈ I, f 0 (x) = lim fn0 (x) = g(x) (6.10)
n→∞

Proof. For the fundamental theorem of integral calculus, we can write, using the regularity of the
fn (x) that
ˆ x
fn (x) = fn (x0 ) + fn (t) dt
x0

Taking the limit we have

ˆ x
lim fn (x) = l + g(t) dt = f (x)
n→∞ x0
∴ f 0 (t) = g(t)

But, we also have that

∀ > 0 kfn0 − f 0 k∞ ≤ |fn (x0 ) − l| + kfn0 − gk∞ (b − a) < 

∴ fn ⇒ f, fn0 ⇒ f 0
6.2. SERIES OF FUNCTIONS 93

§ 6.2 Series of Functions

Let now, for the rest of the section, (X, d) = C.

Definition 6.2.1 (Series of Functions). Let (fn )n≥0 ∈ C be a sequence of functions, such that fn : S → C.
We can define the series of functions as follows
n
X
sn (x) = fk (x) (6.11)
k=1

Definition 6.2.2 (Convergent Series). A series of functions sn (x) : S → C is said to be convergent or

pointwise convergent if
Xn
sn (x) = fk (x) −→ s(x) (6.12)
k=0

Where s(x) : S → C is the sum of the series.

This means that
∞
X
∀x ∈ S, lim sk (x) = fk (x) = s(x) (6.13)
k→∞
k=0

Theorem 6.4. Necessary Condition for the convergence of a series of functions:

Let (fn ) ∈ C be a succession, then the series sn (x) defined as follows, converges to the
function s(x)
n
X ∞
X
sn (x) = fk (x) = s(x) = fk (x)
k=0 k=0

Proof.
∀x ∈ S lim fk (x) = lim (sn (x) − sn+1 (x)) = 0
k→∞ n→∞

Definition 6.2.3 (Uniform Convergence). A series of functions is said to be uniformly convergent if

and only if
X∞ Xn
fk (x) ⇒ s(x) ⇐⇒ sn (x) = fk (x) ⇒ s(x) (6.14)
k=0 k=0

Definition 6.2.4 (Absolute Convergence). A series of functions is said to be absolutely convergent if

and only if
∞ ∞
A
X X
fk (x) −→ s(x) =⇒ |fk (x)| → s(x) (6.15)
k=0 k=0
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 94

P∞ A
Theorem 6.5. Let k=0 fk (x) −→ s(x), then
∞ ∞
A
X X
fk (x) −→ s(x) =⇒ fk (x) → s(x) (6.16)
k=0 k=0

Proof. Let
n
X ∞
X
sn (x) = fk (x) ∴ ∃g(x) : (S, d) −→ C, ∃N (x) ∈ N : g(x) − fk (x) =
k=0 k=0
X∞
= |fk (x)| <  ∀n ≥ N (x)
k=n+1

∴ ∀n, m ∈ N, m > n
m
X ∞
X
|sm (x) − sn (x)| = fk (x) ≤ |fk (x)| <  ∀x ∈ S
k=n+1 k=n+1

∴ (sn (x)) is a Cauchy series in C =⇒ sk (x) → s(x)

Definition 6.2.5 (Total Convergence). A series of functions sk (x) is said to be totally convergent if

1. ∃Mk : supS |fk (x)| ≤ Mk ∀k ≥ 1

P∞
2. k=0 Mk → M

T
The total convergence is then indicated as sk (x) −→ s(x)

Proposition 10. Let

n
X
sn (x) = fn (x)
k=0

Then

1. fn (x) ∈ C(S) ∧ sk (x) ⇒ s(x) =⇒ s(x) ∈ C(S)

´ ´
2. fn (x) ∈ C(S), sk (x) ⇒ s(x) =⇒ s(x) dx = limk→∞ sk (x) dx

A
3. sk (x) −→ s(x) =⇒ sk (x) → s(x)

A
4. sk (x) ⇒ s(x) =⇒ sk (x) −→ s(x)

T
5. sk (x) −→ s(x) =⇒ sk (x) ⇒ s(x)
6.2. SERIES OF FUNCTIONS 95

§§ 6.2.1 Power Series and Convergence Tests

Theorem 6.6 (Weierstrass Test). Let (fn ) : (S, d) → C a sequence of functions.
If we have that
∀n > N ∈ N ∃Mn > 0 : |fn (x)| ≤ Mn
n
X ∞
X X
∴ ∀x ∈ S fk (x) ≤ Mk → M ∴ fk (x)n ⇒ s(x)
k=0 k=1 k=0

Definition 6.2.6 (Power Series). Let z, z0 , (an ) ∈ C. A power series centered in z0 is defined as
follows
X∞
ak (z − z0 )k (6.17)
k=0

Example 6.2.1. Take the geometric series. This is the best example of a power series centered in
z0 = 0, and it has the following form
∞
X
zk (6.18)
k=0

We can expand it as follows

m
X 1 + z n+1
z k = (1 − z) 1 + z + z 2 + · · · + z m = 1 − z n+1 =


∀|z| 6= 1 (6.19)
1−z
k=0

Taking the limit, we have, therefore

∞
X 1
zk = ∀|z| < 1 (6.20)
1−z
k=0

P∞
Theorem 6.7 (Cauchy-Hadamard Criteria). Let k=0 ak (z−z0 )k be a power series, with an , z, z0 ∈ C.
We define the Radius of convergence R ∈ R? = R ∪ {±∞}, with the Cauchy-Hadamard criteria
1


 +∞ =0


 R
1 1

1
= lim sup |an | n = l 0< =l<∞ (6.21)
R n→∞ 
 R
1


0

= +∞
R
Then sk (z) ⇒ s(z) ∀|z| ∈ (−R, R)
Theorem 6.8 (D’Alambert Criteria). From the power series we have defined before, we can write
the D’Alambert criteria for convergence as follows
1 ak+1 ak
= lim =⇒ R = lim (6.22)
R k→∞ ak k→∞ ak+1
Where R is the previously defined radius of convergence
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 96

Theorem 6.9 (Abel). Let R > 0, then if a power series converges for |z| = R, it converges
uniformly ∀|z| ∈ [r, R] ⊂ (−R, R]. It is valid analogously for x = −R

Remark (Power Series Integration). If the series has R > 0 and it converges in |z| = R, calling s(x) the
sum of the series, with x = |z| we can say that
ˆ R ∞ ˆ R ˆ ∞
RX ∞
X X Rk+1
s(x) dx = ak xk dx = ak xk dz = ak (6.23)
0 0 0 k+1
k=0 k=1 k=0

Remark (Power Series Derivation). If Abel’s theorem holds, we have also that, if we have s(x) our
power series sum, we can define the n−th derivative of this series as follows
∞
dn s X
= k(k − 1) · · · (k − n + 1)ak xk−n (6.24)
dxn
k=n

§ 6.3 Series Representation of Functions

§§ 6.3.1 Taylor Series
Theorem 6.10 (Taylor Series Expansion). Let f : D −→ C be a function such that f ∈ H(BR (z0 )),
with Br (z0 ) ⊆ D. Then
n
X 1 dn f
f (z) = (z − z0 )n kz − z0 k < r (6.25)
n=0
n! dz n z0

Proof. Taken z ∈ Br (z0 ) and γ(t) = z0 + reit t ∈ [0, 2π] and kz − z0 k < r < R we can write, using
the integral representation of f
ffi ffi
1 f (w) 1 f (w)
f (z) = dw = dw
2πi γ (w − z) 2πi γ (w − z0 ) − (z − z0 )

From basic calculus we know already that if z 6= w

1 1 − (z/w)n
  z n 
1 1
= + =
w−z w 1 − z/w 1 − z/w w
n−1
1  z n X 1  z n
= +
w−z w w w
k=0

Therefore, inserting it back into the integral representation, we have

X (z − z0 )k ffi f (w) (z − z0 )n
ffi
f (w)
f (z) = k+1
dw + dw
2πi γ (w − z0 ) 2πi γ [(w − z0 ) − (z − z0 )] (w − z0 )n
k=0
6.3. SERIES REPRESENTATION OF FUNCTIONS 97

On the RHS as first term we have the k−th derivative of f and on the right there is the so called
remainder Rn (z). Therefore
n
X 1 dk f
f (z) = (z − z0 )k + Rn (z)
k! dz k z0
k=0

n→∞
It’s easy to demonstrate that Rn (z) −→ 0, and therefore
∞
X 1 dk f
f (z) = (z − z0 )k
k! dz k z0
k=0

Definition 6.3.1 (Taylor Series for Scalar Fields). Given a function f : A ⊂ Rn −→ R f ∈ C m (A), given
a multi-index α one can define the Taylor series of the scalar field as follows
X 1
f (x) = ∂ α f (x0 )(x − x0 )α + Rm (x)
α!
|α|≤m

Where, the remainder is defined in integral form as follows

X (x − x0 )α ˆ 1
Rm (x) = (m + 1) (1 − t)m ∂ α f (x0 + tx − tx0 ) dt
α! 0
|α|=m+1

Definition 6.3.2 (MacLaurin Series). Taken a Taylor series, such that z0 = 0, we obtain a MacLaurin
series.
∞
X 1 dk f
f (z) = zk (6.26)
k! dz k z=0
k=0

Definition 6.3.3 (Remainders). We can have two kinds of remainder functions while calculating series:
n
1. Peano Remainders, Rn (z) = O(kz − z0 k )
2. Lagrange Remainders, Rn (x) = (n + 1)!−1 f (n+1) (ξ)(x − x0 )n+1 , x, x0 ∈ R ξ ∈ (x, x0 )
What we saw before as Rn (z) is the remainder function for functions f : D ⊂ C −→ C.
A particularity of remainder function is that Rn (z) → 0 always, if f is holomorphic
Theorem 6.11 (Integration of Power Series II). Let f, g : BR (z0 ) −→ C and {γ} ⊂ BR (z0 ) a piecewise
smooth path. Taken
X∞
f (z) = an (z − z0 )n g ∈ C({γ})
n=0

We have that ˆ ˆ
∞
X
n
an g(z)(z − z0 ) dz = g(z)f (z) dz (6.27)
n=0 γ γ
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 98

Proof. Since f, g ∈ C({γ}) by definition, and f ∈ H(Br (z0 )) with r < R, we have that ∃K̂γ [f g].
Firstly we can write that ∀z ∈ BR (z0 )
n−1
X n−1
X ∞
X
g(z)f (z) = ak g(z)(z − z0 )k + g(z)Rn (z) = ak g(z)(z − z0 )k + g(z) ak (z − z0 )k
k=0 k=0 k=n

Then we can write

ˆ n−1
X ffi ˆ
k
g(z)f (z) dz = ak g(z)(z − z0 ) dz + g(z)Rn (z) dz
γ k=0 γ γ

Letting M = supz∈{γ} kg(z)k, and noting that kRn (z)k <  for ∀ > 0 and for some n ≥ N ∈ N, z ∈
{γ} we have, using the Darboux inequality
ˆ
g(z)Rn (z) dz ≤ M Lγ  → 0
γ

Theorem 6.12 (Holomorphy of Power Series). If a function f (z) is expressable as a power series
P∞
f (z) = k=0 ak (z − z0 )k , kz − z0 k < R we have that f ∈ H(BR (z0 ))

Proof. Take the previous theorem on the integration of power series, and choose g(z) = 1. Since
g(z) ∈ H(C) we also have that it’ll be continuous on all paths {γ} ⊂ C piecewise smooth.
Take now a closed piecewise smooth path {γ}, then we can write
ffi ffi X
∞ ∞
X ffi
f (z)g(z) dz = ak (z − z0 )k = ak (z − z0 )k dz
γ γ k=0 k=0 γ

Since the function h(z) = (z − z0 )k ∈ H(C) ∀k = 6 1, we have, thanks to the Morera and Cauchy
theorems ffi
f (z) dz = 0 =⇒ f (z) ∈ H(BR ({γ}))
γ

P∞
Corollary 6.3.1 (Derivative of a Power Series II). Take f (z) = k=0 ak (z − z0 )k kz − z0 k < R. Then,
∀z ∈ BR (z0 ) we have that
∞
df X
= ak k(z − z0 )k−1 (6.28)
dz
k=1

Proof. Taken z ∈ BR (z0 ) and a continuous function g(w) ∈ C({γ}), with {γ} ⊂ BR (z0 ) a closed
simple piecewise smooth path. If z ∈ {γ}◦ and
 
1 1
g(w) =
2πi (w − z)2
6.3. SERIES REPRESENTATION OF FUNCTIONS 99

We have, using the integral representation for holomorphic functions

ffi ∞ ffi
1 f (w) X ak (w − z0 )k
dw = dw
2πi γ (w − z)2 2πi γ (w − z0 )2
k=0

Since h(w) = (w − z0 )k ∈ H(C) ∀k 6= 1 we have, using again the integral representation for
holomorphic functions ffi
1 f (w) df
dw =
2πi γ (w − z)2 dz
ffi k
1 (w − z0 )
dw = k(z − z0 )k−1
2πi γ (w − z)2
Therefore
ffi ∞
1 f (w) X df
2
dw = ak k(z − z0 )k =
2πi γ (w − z) dz
k=0

Corollary 6.3.2 (Uniqueness of the Taylor Series). Taken an holomorphic function f ∈ H(D) with
D ⊂ C some connected open set, we have that
∞
X 1 dk f
f (z) = ak (z − z0 )k ak = ∀kz − z0 k < R
k! dz k z0
k=0

Proof. Taken g(z) a continuous function over a closed piecewise simple path {γ} ⊂ C, where
 
1 1
g(z) =
2πi (z − z0 )k+1

We have that
ffi ∞ ffi
1 f (z) X ak
n+1
dz = (z − z0 )k−n−1 dz
2πi γ (z − z0 ) 2πi γ
k=1

The integral on the RHS evaluates to δnk , and thanks to the integral representation of f (z) we can write
ffi
1 f (z) 1 dn f
dz = = n!an
2πi γ (z − z0 )n+1 n! dz n z0

§§ 6.3.2 Laurent Series

Definition 6.3.4 (Annulus Domain). Let 0 ≤ r < R ≤ ∞ and z0 ∈ C, we define the annulus set as
follows
ArR (z0 ) := { z ∈ C| r < kz − z0 k < R} (6.29)
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 100

Special cases of this are the ones where r = 0, R = ∞ and r = 0, R = ∞

A0,R (z0 ) = BR (z0 ) \ {z0 }
Ar,∞ (z0 ) = C \ B r (z0 )
A0,∞ (z0 ) = C \ {z0 }

Theorem 6.13 (Laurent Series Expansion). Let f : AR1 R2 (z0 ) −→ C be a function such that
f ∈ H (AR1 R2 (z0 )), and {γ} ⊂ AR1 R2 (z0 ) a closed simple piecewise smooth curve.
Then f is expandable in a generalized power series or a Laurent series as follows
∞ ∞ ∞
X X c−
n
X
f (z) = c+ n
n (z − z0 ) + n
= ck (z − z0 )k (6.30)
n=0 n=1
(z − z 0 )
k=−∞

Where the coefficients are the following

ffi
− 1
cn = f (z)(z − z0 )n−1 dz n ≥ 0
2πi γ
ffi
+ 1 f (z)
cn = dz n > 0 (6.31)
2πi γ (z − z0 )n+1
ffi
1 f (z)
ck = dz k ∈ Z
2πi γ (z − z0 )k+1

Proof. Taken a random point z ∈ AR1 R2 (z0 ), a closed simple piecewise smooth curve {γ} ⊂ AR1 R2 (z0 )
and two circular smooth paths {γ2 }, {γ3 } : {γ2 } ∪ {γ3 } = ∂Ar1 r2 (z0 ) ⊂ AR1 R2 (z0 ) ∧ {γ} ⊂ Ar1 r2 (z0 )
and a third circular path {γ3 } ⊂ Ar1 r2 (z0 ), we can write immediately, using the omotopy between all
the paths ffi ffi ffi
f (w) f (w) f (w)
dw = dw + dw
γ2 w − z γ1 w − z γ3 w −z
Using the Cauchy integral representation we have that the integral on γ3 yields immediately 2πif (z),
hence we can write
ffi ffi
1 f (w) 1 f (w)
f (z) = dw + dw
2πi γ2 (w − z0 ) − (z − z0 ) 2πi γ1 (z0 − z) − (w − z0 )
Using the two following identities for z 6= w
n−1
1 1  z n X 1  z k
= +
w−z w−z w w w
k=0
 w n n
1 1 X 1  w k
= +
z−w z−w z w z
k=1

We obtain that
n−1 ffi n ffi
X (z − z0 )k f (w) X 1
f (z) = k+1
dw + ρn (z) + f (w)(w − z0 )k−1 dw + σn (z)
2πi γ2 (w − z0 ) 2πi(z − z0 )k γ1
k=0 k=1
6.3. SERIES REPRESENTATION OF FUNCTIONS 101

−
Where, after choosing appropriate substitutions with some coefficients c+
k , ck we have
n−1 n
X X c−
f (z) = c+ k
k (z − z0 ) + ρn (z) +
k
+ σn (z)
(z − z0 )k
k=0 k=1

Where ρn , σn are the two remainders of the series expansion, and are
ffi
(z − z0 )n f (w)
ρn (z) = dw
2πi γ2 [(w − z 0 ) − (z − z0 )] (w − z0 )n
ffi
1 f (w)
σn (z) = dw
2πi(z − z0 )n γ1 (w − z0 ) − (z − z0 )
n→∞
In order to prove the theorem we now need to demonstrate that ρn , σn −→ 0. Taken M1 =
supz∈{γ1 } kf (z)k, M2 = supz∈{γ2 } kf (z)k, we have, using the fact that both γ1 , γ2 are circular
 n
M2 kz − z0 k n→∞
kρn (z)k ≤ −→ 0 kz − z0 k < r2
1 − r12 kz − z0 k r2
 n
M1 r1 n→∞
kσn (z)k ≤ 1 −→ 0 r1 < kz − z0 k
r1 kz − z0 k − 1
kz − z 0 k

And the theorem is proved.

Theorem 6.14 (Convergence of a Laurent Series). Being defined on an annulus set, the Laurent
series of a function must have two radii of convergence. Given a function f holomorphic on
a set AR1 R2 (z0 ) we have
1 p
= lim sup n kcn k
R2 n→∞
p (6.32)
R1 = lim sup n kc−n k
n→∞
It’s equivalent of showing the convergence of the two series
∞ ∞
X X c−
f (z) = c+ k
k (z − z0 ) +
k
(z − z0 )k
k=0 k=1

Theorem 6.15 (Integral of a Laurent Series). Let f (z) ∈ H (AR1 R2 (z0 )) and take {γ} ⊂ AR1 R2 (z0 ) a
piecewise smooth curve, and g ∈ C({γ}), then we have
∞
X ˛ ˛
cn g(z)(z − z0 )n dz = g(z)f (z) dz
n=−∞ γ γ

Proof. We begin by separating the sum in two parts, ending up with the following
X∞ ˛ ˛
c+n g(z)(z − z 0 )n
dz = g(z)f+ (z) dz
n=0 γ γ
∞ ˛ ˛
X g(z)
c−
n dz = g(z)f− (z) dz
n=1 γ (z − z0 )n γ
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 102

Which is analogous to the integration of Taylor series. The same could be obtained keeping the bounds
of the sum in all Z
As for Taylor series, in a completely analogous fashion, a Laurent series is holomorphic and unique.
The derivative of a Laurent series, is then obviously the following
∞
df X
= cn n(z − z0 )n−1
dz n=−∞

§§ 6.3.3 Multiplication and Division of Power Series

P∞
Theorem
P∞ 6.16 (Product of Power Series). Take f (z) = n=0 an (z − z0 )n , z ∈ BR1 (z0 ) and g(z) =
n=0 b n (z − z0 )n , z ∈ BR2 (z0 ). Then
∞
X n
X
f (z)g(z) = cn (z − z0 )n , cn = ak bn−k kz − z0 k < min (R1 , R2 ) = R
n=0 k=0

Proof. Due to the holomorphy of both f and g, we have that the function f g has a Taylor series
expansion
∞
X
f (z)g(z) = cn (z − z0 ) kz − z0 k < R
n=0

We have then, using Leibniz’s derivation rule

n  
1 dn 1 X n dk f dn−k g
cn = f (z)g(z) = =
n! dz n n! k dz k z0 dz n−k z0
k=0
n
X 1 dk f 1 dn−k g
= =
k! dz k z0 (n − k)! dz n−k z0
k=0
Xn
= ak bn−k
k=0

Theorem 6.17 (Division of Power Series). Taken the two functions as before, with the added
necessity that g(z) 6= 0, we have that
∞ n−1
!
f (z) X n 1 X
= dn (z − z0 ) dn = an − dk bn−k
g(z) n=0 b0
k=0

Proof. Everything hold as in the previous proof. Remembering that (f /g)g = f and using the previous
theorem, we obtain
Xn
an = dk bk−n
k=0
6.3. SERIES REPRESENTATION OF FUNCTIONS 103

And therefore, inverting

n−1
an 1 X
dn = − dk bn−k
b0 b0
k=0

§§ 6.3.4 Useful Expansions

∞
X zn
ez = kzk < ∞ (6.33)
n=0
n!

∞
1 iz −iz

X (−1)n 2n+1
sin(z) = e −e = z kzk < ∞ (6.34)
2i n=0
(2n + 1)!

∞
d X (−1)n 2n
cos(z) = sin(z) = z kzk < ∞ (6.35)
dz n=0
(2n)!

∞
X z 2n
cosh(z) = cos(iz) = kzk < ∞ (6.36)
n=0
(2n)!

∞
d X z 2n+1
sinh(z) = cosh(z) = kzk < ∞ (6.37)
dz n=0
(2n + 1)!

∞
1 X
= zn kzk < 1 (6.38)
1 − z n=0

∞
1 X
= (−1)n z n kzk < 1 (6.39)
1 + z n=0

∞
1 X
= (−1)n (z − 1)n kz − 1k < 1 (6.40)
z n=0

∞  
X s
(1 + z)s = zn s ∈ C, kzk < 1 (6.41)
n
n=0

∞
1
X 1
ez = n
0 < kzk < ∞ (6.42)
n=0
n!z
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 104

§ 6.4 Residues

§§ 6.4.1 Singularities and Residues

Definition 6.4.1 (Singularity). Given a function f : G −→ C we define a singularity a point z0 ∈ G
such that
∀ > 0 ∃z ∈ B (z0 ) : f (z) is holomorphic (6.43)

Definition 6.4.2 (Isolated Singularity). Given a function f : G −→ C we define an isolated singularity

a point z0 ∈ G such that
∃r > 0 : f ∈ H (A0r (z0 )) (6.44)

Definition 6.4.3 (Residue). Let z0 ∈ G be an isolated singularity of f : G −→ C, then ∃r > 0 : ∀z ∈

A0r (z0 ) the following Laurent series expansion holds
∞ ∞ ∞
X X bn X
f (z) = an (z − z0 )n + n
= cn (z − z0 )n
n=0 n=1
(z − z 0 ) n=−∞

The residue of the function f in z0 is defined as follows

Res f (z) = b1 = c−1 (6.45)

z=z0

A second definitiion is given by the following contour integral

ffi
1
Res f (z) = f (z) dz
z=z0 2πi γ

Where γ is a simple closed path around z0

Definition 6.4.4 (Winding Number). Given a closed curve {γ} we define the winding number or
index of the curve around a point z0 the following integral
˛
1 dz
n(γ, z0 ) = (6.46)
2πi γ z − z0

Theorem 6.18 (Residue Theorem). Given a function f : G −→ C such that f ∈ H(D) where
D = G \ {z1 , · · · , zn } and zk are isolated singularities, we have, taken a closed piecewise
smooth curve {γ}, such that {z1 , · · · , zn } ⊂ {γ}◦
˛ ∞
X
f (z) dz = 2πi n(γ, zk ) Res f (z) (6.47)
γ z=zk
k=0
6.4. RESIDUES 105

Proof. Firstly we can say that γ ∼ k γk where γk are simple curves around each zk , then since the
P
function is holomorphic in the regions A0r (zk ) with k = 1, · · · , n we can write
∞
X
f (z) = cn (z − zk )n
n=−∞

Therefore, we have
˛ n ˛
X n
X ∞
X ˛
f (z) dz = f (z) dz = cj (z − zk )j dz
γ k=0 γk k=0 j=−∞ γk

We can then use the linearity of the integral operator and write
˛ n −2 ˛ ˛ ∞ ˛
X X dz X
f (z) dz = cj (z − zk )j dz + c−1 + cj (z − zk )j dz
γ j=−∞ γ k γk
z − z k j=0 γk
k=0

Thanks to the Cauchy theorem we already know that the first and last integrals on the RHS must be
null, therefore
˛ n ˛
X dz
f (z) dz = c−1
γ γk z − zk
k=0
Recognizing the definition of residue and the winding number of the curve, we have the assert
˛ X
f (z) dz = 2πi n(γ, zk ) Res f (z)
γ z=zk
k=0

Definition 6.4.5 (Residue at Infinity). Given a function f : G −→ C and a piecewise smooth closed
curve γ. If f ∈ H({γ} ∪ extr {γ}) we have
ffi  
1 1
f (z) dz = −2πi Res f (z) = 2πi Res 2 f (6.48)
γ z=∞ z=0 z z

Theorem 6.19. Given a function f : G −→ C as before, if the function has zk singularities with
k = 1, · · · , n
n
X
Res f (z) = Res f (z) (6.49)
z=∞ z=zk
k=1

§§ 6.4.2 Classification of Singularities, Zeros and Poles

Definition 6.4.6 (Pole). Given a function f (z) with an isolated singular point in z0 ∈ C, we have that
in A0r (z0 ) the function can be expanded with a Laurent series
∞ ∞
X X bk
f (z) = ak (z − z0 )k +
(z − z0 )k
k=0 k=1

The point z0 is called a pole of order m if bk = 0 ∀k > m

CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 106

Definition 6.4.7 (Removable Singularity). Given f (z), z0 as before, we have that z0 is a removable
singularity if bk = 0 ∀k ≥ 1
Definition 6.4.8 (Essential Singularity). Given f (z), z0 as before, we have that z0 is an essential
singularity if bk 6= 0 for infinite values of k
Definition 6.4.9 (Meromorphic Function). Let f : G ⊂ C −→ C be a function. f is said to be
meromorphic if f ∈ H(G̃) where G̃ = G \ {z1 , · · · , zn } where zk ∈ G are poles of the function
Theorem 6.20. Let z0 be an isolated singularity of a function f (z). z0 is a pole of order m if
and only if
1 dm−1 g
f (z) = g ∈ H(B (z0 ))  > 0 (6.50)
(m − 1)! dz m−1 z0

Proof. Let f : G −→ C be a meromorphic function and g : G −→ C, g ∈ H(G) where f (z) has a pole
in z0 ∈ G and g(z0 ) 6= 0
g(z)
f (z) =
(z − z0 )m
Since g(z) is holomorphic in z0 we have that, for some r
∞
X 1 dk g
g(z) = (z − z0 )k z ∈ Br (z0 )
k! dz0k
k=0

And therefore, ∀z ∈ A0r (z0 )

∞
X 1 dk g
1
f (z) = g(z) = (z − z0 )k−m
(z − z0 )m k! dz0k
k=0

Since g(z0 ) 6= 0 we have the assert.

Alternatively we start by hypothesizing that z0 is already a pole of order m for f , and therefore we can
write the following Laurent expansion for some r > 0
∞
X
f (z) = ck (z − z0 )k ∀z ∈ A0r (z0 )
k=−m

Where c−m 6= 0. Therefore, we write

(
(z − z0 )m f (z) z ∈ A0r (z0 )
g(z) =
c−m z = z0
And, expanding g(z) for z ∈ Br (z0 ) we obtain
∞
X
g(z) = c−m + c−m+1 (z − z0 ) + · · · + c−1 (z − z0 )m−1 + ck (z − z0 )k+m
k=0

g(z) is holomorphic in the previous domain of expansion, and therefore we have, since the Taylor
expansion is unique
1 dm−1 g
c−1 = = Res f (z)
(m − 1)! dz0m−1 z=z0
6.4. RESIDUES 107

Definition 6.4.10 (Zero). Let f : G −→ C be a holomorphic function. Taken z0 ∈ G, it’s said to be a

zero of order m if
 k
d f
 k = 0 k = 1, · · · , m − 1

dz0

m

 d
 f 6= 0
dz0m

Theorem 6.21. The point z0 ∈ G is a zero of order m for f if and only if

g(z)
f (z) = g(z0 ) 6= 0, g ∈ H(G)
(z − z0 )m

Proof. Taken f (z) = (z − z0 )m g(z) such that g(z0 ) 6= 0 we can expand g(z) with Taylor and at the
end obtain
∞
X 1 dk g
f (z) = (z − z0 )k+m
k! dz0k
k=0

Since this is a Taylor expansion also for f (z) we have that, for j = 1, · · · , m − 1

dj f dm f
=0 = m!g(z0 ) 6= 0
dz0j dz0m

The same is obtainable with the vice versa demonstrating the theorem

Notation. Let f be a meromorphic function. We will define the following sets of points accordingly

1. Zfm as the set of zeros of order m

2. Sf as the set of isolated singularities of f

3. Pfm as the set of poles of order m

We immediately see some special cases

1. Pf∞ is the set of essential singularities of f

2. Pf1 is the set of removable singularities of f

Theorem 6.22. Let f : D −→ C be a function such that f ∈ H(D), with D an open set, then

1. f (z) = 0 ∀z ∈ D

2. ∃z0 : f (k) (z0 ) = 0 ∀k ≥ 0

3. Zf ⊂ D has a limit point

CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 108

Proof. 3) =⇒ 2)
Take z0 ∈ D as the limit point of Zf . Since f ∈ C(D) we have that z0 ∈ Zfm . therefore

f (z) = (z − z0 )m g(z) g(z0 ) 6= 0, g ∈ H(D) =⇒ ∃δ > 0 : g(z) 6= 0 ∀z ∈ Bδ (z0 )

Therefore
f (z) 6= 0 ∀z ∈ A0δ (z0 )
2) =⇒ 1) 
Suppose that Zf (k) := z ∈ D| f (k) (z) = 0 6= {}. We have to demonstrate that this set is clopen in
D.
Take z ∈ Zf (k) and a sequence (z)k ∈ Zf (k) such that zk → z. We have then

f (k) (z) = lim f (k) (zk ) = 0

k→∞

Therefore Zf (k) = Zf (k) and the set is closed.

Take then z ∈ Zf (k) ⊂ D, since D is open we have that ∃r > 0 : Br (z) ⊂ D, therefore

∞
(
X
k z=w
∀w ∈ Br (z), z 6= w f (w) = ak (w − z) = 0 =⇒
k=0
ak = 0 ∀k ≥ 0

Since w 6= z we have that Br (z) ⊂ Zf (k) and the set is open. Taking both results we have that the set
is clopen and D = Zf (k)

Corollary 6.4.1. Let f, g : D −→ C and f, g ∈ H(D). We have that f = g iff the set {f (z) = g(z)}
has a limit point in D

Corollary 6.4.2 (Zeros of Holomorphic Functions). Let f : D −→ C be a non-constant function

f ∈ H(D) with D an open connected set. Then

∀z ∈ Zfm m<∞

Proof. Take z0 ∈ Zf , then since f is non-constant we have that Zf has no limit points in D, therefore

dk f dm f
∃δ > 0 : f (z) 6= 0 ∀z ∈ A0δ (z0 ) ∧ ∃m ≥ 1 : = 0 k ∈ [0, m), 6= 0
dz0k dz0m

Therefore z0 ∈ Zfm

Theorem 6.23. Let f : D −→ C be a meromorphic function, such that

p(z)
f (z) = p, q ∈ H(D)
q(z)

If z0 ∈ Zqm such that p(z0 ) 6= 0, then z0 ∈ Pfm

6.4. RESIDUES 109

Proof. z0 ∈ Zqm is an isolated singularity of q, therefore

∃δ > 0 : q(z) 6= 0 ∀z ∈ A0δ (z0 ) ∴ z0 ∈ Sp/q
m
We therefore can take q(z) = (z − z0 ) g(z) and we have
p(z) h(z)
f (z) = m
=
g(z)(z − z0 ) (z − z0 )m
Where h(z) is a holomorphic function such that h(z0 ) 6= 0. By definition of pole we have z0 ∈ Pfm
Theorem 6.24 (Quick Calculus of Residues for Rational Functions). If f (z) = p(z)/q(z) as before,
there is a quick rule of thumb for calculating the residue in z0 . We can write
1 dm−1 h
Res f (z) =
z=z0 (m − 1)! dz0m−1
If the pole is a removable singularity, we have z0 ∈ Pf1 and
p(z0 )
Res f (z) =
z=z0 q 0 (z0 )

Theorem 6.25. Let f be a meromorphic function. If z0 ∈ Pfm we have

lim f (z) = ∞
z→z0

Proof.
g(z)
z0 ∈ Pfm =⇒ f (z) = , z0 ∈
/ Zg
(z − z0 )m
Then
1 (z − z0 )
lim = lim =0
z→z0 f (z) z→z0 g(z)

Theorem 6.26. If z0 ∈ Pf1 , ∃ > 0 such that f ∈ A0 (z0 ) and kf (z)k ≤ M , ∀z ∈ A0 (z0 )
Proof. By definition we have that
∃r > 0 : f ∈ H(A0 (z0 ))
And therefore the function is Laurent representable in this set as follows
∞
X
f (z) = ck (z − z0 )k 0 < kz − z0 k < 
k=0

Taken the following holomorphic function

z ∈ A0 (z0 )

 f (z)

∞
g(z) = X

 ck (z − z0 ) z = z0
z=0


We have that g ∈ H B (z0 ) and therefore kf (z)k ≤ M ∀z ∈ A0 (z0 )
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 110

Lemma 6.4.1 (Riemann). Take a function f ∈ H (A0 (z0 )) for some  > 0, then if kf (z)k ≤ M ∀z ∈
A0 (z0 )
The point z0 is a removable singularity for f

Proof. In the set of holomorphy the function is representable with Laurent, therefore

∞ ∞
X X c−
f (z) = c+ k
k (z − z0 ) +
k
(z − z0 )k
k=0 k=1

We have that the coefficients c−

k are the following, where we integrate over a curve {γ} := { z ∈ C| kz − z0 k = ρ < }

ffi
1
c−
k = f (z)(z − z0 )k−1 dz
2πi γ

The function is limited, and therefore for Darboux

c− k
k ≤ρ M →0 ∀k ≥ 1

Therefore z0 ∈ Pf1

Theorem 6.27 (Quick Calculus Methods for Residues). Let f be a meromorphic function, then

1. z0 ∈ Pfn then
1 dn−1 n
Res f (z) = lim (z − z0 ) f (z) (6.51)
z=z0 (n − 1)! z→z0 dz n−1

2. z0 ∈ Pfm and f (z) = p(z)/(z − z0 )m , where p ∈ Ck [z] with k ≤ m − 2 and p(z0 ) 6= 0, then

p(z)
Res f (z) = Res =0
z=z0 z=z0 (z − z0 )m

§ 6.5 Applications of Residue Calculus

§§ 6.5.1 Improper Integrals

Definition 6.5.1 (Improper Integral). An improper integral is defined as the integral of a function in
a domain where such function has a divergence, or where the interval is infinite. Some examples of
6.5. APPLICATIONS OF RESIDUE CALCULUS 111

such integrals, given a function f (x) with divergences at a, b ∈ R are the following
ˆ ∞ ˆ R
f (x) dx = lim f (x) dx
c R→∞ c
ˆ d ˆ d
f (x) dx = lim f (x) dx
−∞ R→∞ −R
ˆ ∞ ˆ ˆ R
f (x) dx = f (x) dx = lim f (x) dx
−∞ R R→∞ −R
ˆ b ˆ b−
f (x) dx = lim+ f (x) dx
a →0 a+
ˆ h ˆ a− ˆ h
!
f (x) dx = lim f (x) dx + f (x) dx a ∈ (e, h)
e →0+ e a+

Definition 6.5.2 (Cauchy Principal Value). The previous definitions give rise to the following definition,
the Cauchy principal value. Given an improper integral we define the Cauchy principal value as
follows
Let f (x) be a function with a singularity c ∈ (a, b), and g(x) another function then
ˆ ∞ ˆ ˆ R
PV g(x) dx = PV g(x) dx = lim g(x) dx
−∞ R R→∞ −R
ˆ b ˆ ˆ !
c− b
PV f (x) dx = lim f (x) dx + f (x) dx
a →0+ a c+

In the first case. PV is usually omitted.

For a complex integral, if γR (t) = Reit is a circumference, we have
ˆ ˆ
PV f (z) dz = lim f (z) dz
γR R→∞ γR

Notation (Circumferences and Parts of Circumference). For a quick writing of the integrals in this
section, we will use this notation for the following circumferences
CR (t) = Reit t ∈ [0, 2π]
it
CRαβ = Re t ∈ [α, β]
+ it
CR (t) = Re t ∈ [0, π]
− −it
CR (t) = Re t ∈ [0, π]
C˜± R = ±
CR × [−R, R]

Hypothesis 1. Let R0 > 0 and f ∈ C(D), where D := { z ∈ C| kzk ≥ R0 } ∪ R and

lim zf (z) = 0
z→∞
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 112

Hypothesis 2. Let R0 > 0 and f ∈ C(D), where D := { z ∈ C| kzk ≥ R0 } ∪ R and

lim f (z) = 0
z→∞

Theorem 6.28. If (1) holds true, then

ˆ
+ −
PV f (z) dz = 0 γR = CR , CR , CR (6.52)
γR

Also, if f (x) is a real function

ˆ ˆ ˆ
f (x) dx = PV f (z) dz = PV f (z) dz (6.53)
+ −
R C̃R C̃R

Theorem 6.29. Let f (z) be an even function, if (1) holds we have

ˆ ∞ ˆ ˆ
1 1
f (x) dx = PV f (z) dz = PV f (z) dz (6.54)
0 2 +
C̃R 2 −
C̃R

Theorem 6.30. Let f (z) = g(z k ), k ≥ 2. If (1) holds

ˆ ∞ ˆ
1
f (x) dx = 2iπ PV f (z) dz (6.55)
0 1−e k C̃R0,2π/k

Theorem 6.31. If (2) holds

ˆ ˆ
iλx
f (x)e dx = PV f (z)eiλz dz λ>0
+
R C̃R
ˆ ˆ (6.56)
f (x)eiλx dx = PV f (z)eiλz dz λ>0
−
R C̃R

From this, we can write then, for λ > 0

ˆ ˆ !
iλz
f (x) cos(iλx) dx = Re PV f (z)e dz λ>0
+
R C̃R
ˆ ˆ ! (6.57)
iλz
f (x) sin(iλx) dx = Im PV f (z)e dz λ>0
+
R C̃R

Hypothesis 3. Let f (z) = g(z)h(z) with g(z) a meromorphic function such that Sg 6⊂ R+ and
1. h ∈ H(C \ R+ )
2. limz→∞ zf (z) = 0
6.5. APPLICATIONS OF RESIDUE CALCULUS 113

3. limz→0 zf (z) = 0

Definition 6.5.3 (Pacman Path). Let ΓRr be what we will call as the pacman path, this path is formed
by 4 different paths

γ1 (t) = reit t ∈ [, 2π − ]

γ2 = [−R, R]
(6.58)
γ3 (t) = Reit t ∈ [, 2π − ]
γ4 = [−R, R]
We will abbreviate this as Γ
Theorem 6.32. Given f (x) a function such that (3) holds, we have that
ˆ ∞ ˆ
g(x)∆h(x) dx = PV g(z)h(z) dz (6.59)
0 Γ

Where
∆h(x) = lim+ (h(x + i) − h(x − i)) (6.60)
→0

In general, we have the following conversion table

h(z) ∆h(x)

1
− 2πi log+ (z) 1

log+ (z) −2πi

2
log+ (z) −2πi log(x) + 4π 2
(6.61)
log+ (z) − 2πi log+ (z) −4πi log(x)

i 2 1
4π log+ (z) + 2 log+ (z) log(x)


[z α ]+ xα 1 − e2πiα

All the previous integrals are solved through a direct application of the residue theorem.

§§ 6.5.2 General Rules

Theorem 6.33 (Integrals of Trigonometric Functions). Let f (cos θ, sin θ) be some rational function
of cosines and sines. Then we have that
ˆ 2π ˆ
z + z −1 z − z −1 dz
 
f (cos θ, sin θ) dθ = f , (6.62)
0 kzk=1 2 2i iz
CHAPTER 6. SEQUENCES, SERIES AND RESIDUES 114

Theorem 6.34 (Integrals of Rational Functions). Let f (x) = pn (x)/qm (x) with m ≥ n + 2 and
qm (x) 6= 0 ∀x ∈ R, then ˆ
pn (x) X pn (z)
dx = 2πi Res (6.63)
q
R m (x) z=zk qm (z)
k

Lemma 6.5.1 (Jordan’s Lemma). Let f (z) be a holomorphic function in A := { z ∈ C| kzk > R0 , Im(z) ≥ 0}.
Taken γ(t) = Reit 0 ≤ t ≤ π with R > R0 .
If ∃MR > 0 : kf (z)k ≤ MR ∀z ∈ {γ} and MR → 0, we have that
ˆ
PV f (z)eiaz dz = 0 a > 0 (6.64)
γ

Theorem 6.35. Let f (x) = pn (x)/qm (x) and m ≥ n + 1 with qm (x) 6= 0 ∀x ∈ R, then ∀a > 0 we
have that ˆ
pn (x) iax X pn (z) iaz
e dx = 2πi Res e (6.65)
R mq (x) z=zk q m (z)
k

Lemma 6.5.2. Let f (z) be a meromorphic function such that z0 ∈ Pf1 and γr± are semi circumferences
parametrized as follows
γr± (t) = z0 + re±iθ θ ∈ [−π, 0]
Then ˆ
PV f (z) dz = ±πi Res f (z) (6.66)
γr± z=z0

Theorem 6.36. Let f (x) = pn (x)/qm (x) with m ≥ n + 2 and qm (x) has xj ∈ Zg1 R
then
ˆ
pn (x) X pn (z) X pn (z)
dx = 2πi Res + πi Res (6.67)
R qm (x) z=zk qm (z)
j
z=xj qm (z)
k

If g(x) = rα (x)/sβ (x)eiax and β ≥ α + 1 with xj ∈ Zg1 R , then ∀a > 0

ˆ
rα (x) iax X rα (z) iaz X rα (z) iaz
e dx = 2πi Res e + πi Res e (6.68)
R sβ (x) z=zk sβ (z)
j
z=xj sβ (z)
k

zk are all the zeros of q, s contained in the plane {Im(z) > 0}

7 Hilbert and Banach Spaces

§ 7.1 Banach Spaces

§§ 7.1.1 Sequence Spaces
Definition 7.1.1 (Banach Space). Given a space and a norm (X, k·k), the space is said to be a Banach
space if it’s complete with respect to the norm k·k.
I.e. remembering the definition of completeness, we have that ∀(x)k ∈ X Cauchy sequence, xk →
x∈X
Notation (The Field F). Here in this section, the field F should be intended as either the field of real
numbers R or the field of complex numbers C
Definition 7.1.2 (Sequence Space). As a n-tuple in the field Fn can be seen as a sequence, as follows

x ∈ Fn , x = (x1 , x2 , · · · , xn ) = (xk )nk=1

We can imagine a sequence as a point in a space. We will call this space FN , and an element of this
space will be indicated as follows

x ∈ FN , x = (x)n = (x1 , x2 , · · · , xn , · · · ) = (xk )∞

k=1

Therefore, every point in FN is a sequence. Note that the infinite sequence of 0s and 1s will be indicated
as 0 = (0)n , 1 = (1)n
Definition 7.1.3 (Sequence of Sequences). We can see a sequence of sequences as a mapping from N
to the space FN , as follows
x :N −→ FN
n → ((x)k )n
It’s important to note how there are two indexes, since every element of the sequence is a sequence in
itself (i.e. ((x)k )n ∈ FN for any fixed n ∈ N)
Definition 7.1.4 (Convergence of a Sequence of Sequences). A sequence of sequences is said to
converge to a sequence in FN if and only if

lim k(x)k − ((x)k )n k = 0 (7.1)

n→∞

For some norm k·k

115
CHAPTER 7. HILBERT AND BANACH SPACES 116

Definition 7.1.5 (Pointwise Convergence). A sequence of sequence is said to converge pointwise to a

sequence in FN if and only if
∀k ∈ N lim ((x)k )n = (x)k (7.2)
n→∞

And it’s indicated as ((x)k )n → (x)k

Example 7.1.1. Take the following sequence of sequences in FN
k
((x)k )n =
n
This sequence converges pointwise to the null sequence, since
k
lim ((x)k )n = lim = (0)k
n→∞ n→∞ n

Space of Bounded Sequences

Definition 7.1.6 (Limited Sequence Space). Let (x)k ∈ FN . Calling the space of bounded sequences as
`∞ (F), we have that (x)k ∈ `∞ (F) if and only if

sup |(x)n | = M ∈ F (7.3)

n∈N

Therefore, this space is defined as follows

`∞ (F) := { (x)n ∈ FN sup |(x)n | < M, M ∈ F} (7.4)

n∈N

Theorem 7.1. The application k·k∞ = supn∈N |·| is a norm in `∞ (F)

Proof. 1) k(x)n k ≥ 0 ∀(x)n ∈ FN , k(x)n k∞ = 0 ⇐⇒ (x)n = (0)n , by definition of sup the first
statement is obvious, meanwhile for the second

0 ≤ |(x)n | ≤ sup |(x)n | = 0 =⇒ |(x)n | = 0 ∴ (x)n = (0)n

n∈N

2) kc(x)n k∞ = |c|k(x)n k∞

kc(x)n k∞ = sup |c(x)n | = sup |c||(x)n | = |c| sup |(x)n | = |c|k(x)n k∞

n∈N n∈N n∈N

3) k(x)n + (y)n k∞ ≤ k(x)n k∞ + k(y)n k∞

sup |(x)n + (y)n | ≤ sup (|(x)n | + |(y)n |) = sup |(x)n | + sup |(y)n | = k(x)n k∞ + k(y)n k∞
n∈N n∈N n∈N n∈N

Since `∞ (F) is a vector space, the couple (`∞ (F), k·k∞ ) is a normed vector space
Remark. Let V be a vector space over some field F. If dim(V) = ∞, a closed and bounded subset
W ⊂ V isn’t necessarily compact, whereas, a compact subset Z ⊂ V is necessarily closed and bounded.
7.1. BANACH SPACES 117

Example 7.1.2. Take V = `∞ (F) and W = B1 ((0)n ), where

B1 ((0)n ) := { (x)n ∈ F∞ | k(x)n k∞ ≤ 1}

We have that diam(B1 ) = 2, therefore this set is bounded and closed by definition.
Take the canonical sequence of sequences ((e)k )n , defined as follows:

((e)k )n = ((0)k , (0)k , · · · , (0)k , (1)k , (0)k , · · · ), for some k ∈ N

Therefore, ∀n 6= m
((e)k )n − ((e)k )m ∞
= k(1)k k∞ = 1
Therefore there aren’t converging subsequences, and therefore B1 can’t be compact.

Space of Sequences Converging to 0

Definition 7.1.7 (Space of Sequences Converging to 0). The space of sequences converging to 0 is
indicated as `0 (F) and is defined as follows

`0 (F) := { (x)n ∈ FN (x)n → 0} (7.5)

Proposition 11. `0 (F) ⊂ `∞ (F), and the couple (`0 (F), k·k∞ ) is a normed vector space, where the
norm k·k∞ gets induced from the space `∞ (F)

Proof.
lim (x)k = 0 =⇒ ∀ > 0 ∃N ∈ N : |(x)n | <  ∀n ≥ N
k→∞
∴ sup |(x)n | =  ≤ M ∈ F =⇒ (x)n ∈ `∞ (F), ∴ `0 (F) ⊂ `∞ (F)
n∈N

`p (F) Spaces
Definition 7.1.8. The sequence space `p (F) is defined as follows
p
`p (F) := { (x)n ∈ FN k(x)n kp = M ∈ F} (7.6)

Where k·kp is the usual p−norm

Proposition 12. The application k·kp : `p (F) −→ F is a norm in `p (F), and the couple (`p (F), k·kp ) is a
normed vector space

Proof. We begin by proving that `p (F) is actually a vector space, therefore 1) ∀(x)n , (y)n ∈ `p (F), (x)n +
(y)n = (x + y)n ∈ `p (F)
∞
X p p
(x + y)n ∈ `p (F) =⇒ |(x)n + (y)n | = k(x)n + (y)n kp < M ∈ F
n=0
p p p
k(x)n + (y)n kp ≤ k(x)n kp + k(y)n kp < M ∈ F
CHAPTER 7. HILBERT AND BANACH SPACES 118

2) ∀(x)n ∈ `p (F), c ∈ F, c(x)n ∈ `p (F)

p
c(x)n ∈ `p (F) =⇒ kc(x)n kp < M ∈ F
∞
X ∞
X
p p p p p p
kc(x)n kp = |c(x)n | = |c| |(x)n | = |c| k(x)n kp < M ∈ F
n=0 n=0

Remark. (x)n ∈ `p (F) =⇒ (x)n ∈ `0 (F).

p
Proof. The proof is simple, taking (y)n = |(x)n | , we can see that (y)n → 0, therefore (x)n → 0 and
(x)n ∈ `0 (F)

Space of Finite Sequences

Definition 7.1.9 (Space of Finite Sequences). The space of finite sequences is indicated as `f (F) and
it’s defined as follows
`f (F) := { (x)n ∈ FN (x)n = 0 ∀n > N ∈ N} (7.7)
It’s already obvious that `f (F) ⊂ `p (F) ⊂ `q (F) ⊂ `0 (F) ⊂ `∞ (F), where p < q ∈ R+ \ {0} where
p < q ∈ R+ \ {0}

§§ 7.1.2 Function Spaces

Notation. In this case, when there will be written the field F, we might either mean R only, i.e.
functions R −→ R, or R; C, i.e. functions R −→ C.

Definition 7.1.10 (Some Function Spaces). We are already familiar from the basic courses in one
dimensional real analysis, about the space of continuous functions C(A), where A ⊂ R. We can define
three other spaces directly, adding some restrictions.

1. Cb (F) := { f ∈ C(F)| supx∈F (f (x)) ≤ M ∈ F}

2. C0 (F) := { f ∈ C(F)| limx→∞ (f (x)) = 0}

3. Cc (F) := { f ∈ C(F)| f (x) = 0 ∀x ∈ Ac ⊂ F} i.e. Cc (F) := { f ∈ C(F)| supp (f ) is compact },

where with supp we indicate the following set suppF (f ) := { x ∈ F| f (x) 6= 0}

Due to the properties of continuous functions, these spaces are obviously vector spaces.

Proposition 13. We have Cc (F) ⊂ C0 (F) ⊂ Cb (F) ⊂ C(F), the application

kf ku = kf k∞ = sup |f (x)| (7.8)

x∈A

Is a norm in C(A), whereas

kf ku = kf k∞ = sup |f (x)| (7.9)
x∈F

Is a norm in the other three spaces

7.1. BANACH SPACES 119

Proof. The inclusion of these spaces is obvious, due to the definition of these. For the proof that the
application k·ku is a norm, it’s immediately given from the proof that the application k·k∞ is a norm in
`∞ (F), and that k·ku = k·k∞
Remark. Take fn ∈ Cb (F) a sequence of functions. The uniform convergence of this sequence means
that fn → f in the norm k·ku = k·k∞
Proposition 14. If f ∈ C0 (F), then f is uniformly continuous
Proof. Let f ∈ C0 (F), then

∀ > 0 ∃l : |x| ≥ l =⇒ |f (x)| <
2
Since every continuous function is uniformly continuous in a closed set, then

∀ > 0 ∃δ : ∀x, y ∈ [−L − 1, L + 1] ∧ |x − y| < δ =⇒ |f (x) − f (y)| < 

Hence we can have two cases. We either have |x − y| < δ or x, y ∈ [−L − 1, L + 1]. Hence we have,
in the first case
|f (x) − f (y)| ≤ |f (x)| + |f (y)| < 
Or, in the second case

∀ > 0 ∃δ > 0 : |x − y| < δ =⇒ |f (x) − f (y)| < 

Demonstrating our assumption

Cp (F) spaces
Definition 7.1.11. We can define a set of function spaces analogous to the `p (F) spaces. These spaces
are the Cp (F) spaces. We define analogously the p−norm for functions as follows
sˆ
p p
kf kp := |f (x)| dx (7.10)
F

Thanks to what said about `p (F) spaces and p−norms, it’s already obvious that these spaces are normed
vector spaces
Remark. Watch out! Cp (F) 6⊂ C0 (F), and Cp (F) 6⊂ Cq (F) for 1 ≤ p ≤ q. It’s easy to find counterex-
amples
Proposition 15. If 1 ≤ p ≤ q, then

Cp (F) ∩ Cb (F) ⊂ Cq (F)

Proof. Let f ∈ Cp (F) ∩ Cb (F). Therefore supx∈F |f (x)| < M ∈ F, then

ˆ ˆ ˆ
q p q−p q−p p
|f (x)| dx = |f (x)| |f (x)| dx ≤ M |f (x)| dx ≤ ∞
F F F

Therefore f ∈ Cp (F) ∩ Cb (F) =⇒ f ∈ Cq (F)

CHAPTER 7. HILBERT AND BANACH SPACES 120

§§ 7.1.3 Function Spaces in Rn

Definition 7.1.12 (Seminorm). A seminorm is an application k·kα,β : A −→ F with A a function space
and α, β multiindices, where

kf kα,β := xα ∂ β f ∞
= sup xα ∂ β f (x) (7.11)
x∈F

Definition 7.1.13 (Schwartz Space). The space S(Rn ) is called the Schwartz space, and it’s defined as
follows n o
S(Rn ) := f ∈ C ∞ (Rn )| kf kα,β < ∞, α, β multiindices (7.12)

Example 7.1.3. Taken p(x) ∈ R[x] a polynomial, a common example of functions f (x) ∈ S(R) is the
following.
2n
f (x) = p(x)e−a|x| (7.13)
With a > 0, n ∈ N
Theorem 7.2. A function f ∈ C ∞ (Rn ) is in S(Rn ) if ∀β multiindex, ∀a > 0 ∃Cα,β such that
Cα,β
∂ β f (x) ≤   a2 ∀x ∈ Rn (7.14)
2
1 + kxk

Proof. Taken n = 1 and f ∈ C ∞ (R), then

j
j Cj,k |x|
xj ∂ k f (x) = |x| ∂ k f (x) ≤ j ≤ Cj,k ∀x ∈ R
(1 + x2 ) 2

Therefore
kf kj,k ≤ Cj,k < ∞ =⇒ f ∈ S(R)
Taken f ∈ S(R) we have that, if |x| ≥ 1
a a a
(1 + x2 ) 2 ≤ 2 2 |x|

Taken j = dae
a a

k xj ∂ k f (x) kf kj,k 2 2 kf kj,k 2 2 kf k0,k

∂ f (x) = j
≤ a ≤ a ≤ a |x| < 1
|x| |x| (1 + x2 ) 2 (1 + x2 ) 2
a
Taken Cj,k = 2 2 max kf kdae,k , kf k0,k the assertion is demonstrated

Definition 7.1.14 (Space of Compact Support Function). Given a function with compact support f , we
define the space of compact functions Cc∞ (Rn ) as the space of all such functions.
We have obviously that Cc∞ (Rn ) ⊂ S(Rn )
Theorem 7.3. Both Cc∞ (Rn ) and S(Rn ) are dense in (Cp (Rn ), k·kp )
7.2. HILBERT SPACES 121

Theorem 7.4 (Other Function Spaces). 1. C(R) Space of continuous functions

2. R[x] Space of real polynomials
3. C k (R) Space of continuous k−derivable functions
4. Cck (R) Space of functions f ∈ C k (R) with compact support
5. C ∞ (R) Space of infinitely differentiable (smooth) functions
6. C0 (R) Space of smooth functions with limx→±∞ f (x) = 0
7. Cc∞ (R) Space of smooth functions with compact support
We have the obvious inclusions
`f ⊂ `p ⊂ · · · ⊂ `0 ⊂ RN
Cc (R) ⊂ C0 (R) ⊂ · · · ⊂ Cp (R) ⊂ C(R)
R[x] ⊂ C ∞ (R) ⊂ · · · ⊂ C(R)
Cc∞ (R) ⊂ C0∞ (R) ⊂ C ∞ (R)

§ 7.2 Hilbert Spaces

Definition 7.2.1 (Hermitian Product). Given V a complex vector space, and an application h·, ·i : V −→ C
such that ∀u, v, z ∈ V, c, d ∈ C
1. hv, vi ≥ 0
2. hv, vi = 0 ⇐⇒ v = 0
3. hu, vi = hv, ui
4. hu + v, zi = hu, zi + hv, zi
5. hcu, dvi = cdhu, vi
The application h·, ·i is called an Hermitian product in V, and the couple (V, h·, ·i) is called an
Euclidean space
Remark. It’s usual in physics that for a Hermitian product, we have that

hcu, vi = chu, vi (7.15)

Definition 7.2.2 (Hilbert Space). Given (V, h·, ·i) an euclidean space. It’s said to be a Hilbert space if
it’s complete
Theorem 7.5 (Cauchy-Schwartz Inequality). Given (V, h·, ·i) a complex euclidean space, then
∀u, v ∈ V
2
khu, vik ≤ hu, uihv, vi (7.16)
CHAPTER 7. HILBERT AND BANACH SPACES 122

Proof. Taken t ∈ C, we define p(t) = htu + v, tu + vi. Then by definition of the Hermitian product,
we have
2
p(t) = ktk hu, ui + thu, vi + thv, ui + hv, vi
Writing hu, vi = ρeiθ , t = se−iθ we have

p(se−iθ ) = s2 hu, ui + 2sρ + hv, vi ≥ 0 ∀s ∈ R

Then, by definition, we have

2
ρ2 = khu, vik ≤ hu, uihv, vi

Theorem 7.6 (Induced Norm). Given a Hermitian product h·, ·i we can define an induced norm
k·k by the definition p
k·k = h·, ·i (7.17)

Theorem 7.7. Addition, multiplication by a scalar and the scalar product are continuous in an
euclidean space V, then, given two sequences un −→ u ∈ V, vn → v ∈ V and

un + vn −→ u + v
c ∈ C =⇒ cun −→ u
hun , vn i −→ hu, vi

Proof. Thanks to Cauchy-Schwartz we have

|hu, vi − hun , vn i| = |hu, vi − hu, vn i + hu, vn i − hun , vn i| ≤ |hu, vi − hu, vn i| + |hu, vn i − hun , vn i| =
= kukkv − vn k + kvn kku − un k

Since the successions are convergent, we have that ∃M > 0 : kvn k ≤ M ∀n ∈ N, therefore

|hu, vi − hun , vn i| ≤ max{kuk, M } (kv − vn k + ku − un k) → 0

Example 7.2.1 (Some Euclidean Spaces). 1) `2 (C)

Given x, y ∈ `2 (C) we define the scalar product as
∞
X
hx, yi = xi yi
i=1

2) `2 (µ), a weighted sequence space, where

∞
( )
X 2
2 N
` (µ) := x ∈ C µi |xi | < ∞, µi ∈ R, µi > 0 ∀i
i=1
7.3. PROJECTIONS 123

Given x, y ∈ `2 (µ) we define

∞
X
hx, yi = µi xi yi
i=1

3) C2 (C) Given f, g ∈ C2 (C) we define

ˆ
hf, gi = f (x)g(x) dx
R

4) C2 (C, p(x) dx), weighted function spaces, where

 ˆ 
+
C2 (C, p(x) dx) := f ∈ C(C)| f (x)f (x)p(x) dx < ∞, p(x) ∈ C(R; R )
R

Given f, g ∈ C2 (C, p(x) dx) we define

ˆ
hf, gi = f (x)g(x)p(x) dx
R

The spaces C2 aren’t complete therefore they’re not Hilbert spaces. The spaces L2 (C) and the
weighted alternative are the completion of such spaces and are therefore Hilbert spaces

Theorem 7.8 (Polarization Identity). Given a complex euclidean space (V, h·, ·i) we have, ∀u, v ∈ V
1 2 2

2 2

hu, vi = ku + vk − ku − vk + i ku + ivk ku − ivk (7.18)
4

Theorem 7.9 (Parallelogram Rule). Let (V, k·k) be a normed vector space. A necessary and
sufficient condition that the norm is induced by a scalar product is that
 
2 2 2 2
ku + vk + ku − vk = 2 kuk + kvk ∀u, v ∈ V (7.19)

§ 7.3 Projections
§§ 7.3.1 Orthogonality
Definition 7.3.1 (Angle). Given a real euclidean space (V, h·, ·i) we define the angle θ = u∠v as follows
 
hu, vi
θ = arccos (7.20)
kukkvk

Definition 7.3.2 (Orthogonal Complement). Given an euclidean vector space V and two vectors u, v,
we say that the two vectors are orthogonal u ⊥ v if

hu, vi = 0 (7.21)
CHAPTER 7. HILBERT AND BANACH SPACES 124

If X ⊂ V and ∀x ∈ X we have that

hu, xi = 0
We say that u ∈ X ⊥ where this space is called the Orthogonal Complement of X, i.e.

X ⊥ := { v ∈ V| hv, wi = 0 ∀w ∈ X} (7.22)

Theorem 7.10. Given X ⊂ V with V an euclidean space, the set X ⊥ is a closed subspace of V
Proof. X ⊥ is a subspace, hence ∀v1 , v2 ∈ X ⊥ and c1 , c2 ∈ C we have

hc1 v1 + c2 v2 , wi = c1 hv1 , wi + c2 hv2 , wi ∀w ∈ X

Hence c1 v1 + c2 v2 ∈ X ⊥ .
Given a sequence (v)n ∈ X ⊥ : (v)n −→ v ∈ V we have, given w ∈ X

hvn , wi = 0 ∀n ∈ N

Due to the continuity of the scalar product we have that

lim hvn , wi = 0 = hv, wi

n→∞

Therefore v ∈ X ⊥ and the subspace X ⊥ is closed in V

Theorem 7.11. Given X, Y ⊂ V with V an euclidean space, we have

X ⊂ Y =⇒ Y ⊥ ⊂ X ⊥
 ⊥
X ⊥ = (X)⊥ = span(X)

Proof. Taken v ∈ Y ⊥ we have by definition

hv, yi = 0 ∀y ∈ Y

Since X ⊂ Y we have then

hv, xi = 0 ∀x ∈ X
Therefore Y ⊥ ⊂ X ⊥ .
By definition we have that X ⊂ X ⊂ span(X), and thanks to the previous proof

X ⊥ ⊃ (X)⊥ ⊃ (span(X))⊥

Taken w ∈ span(X) we have X

w= ci wi wi ∈ X
i

And given v ∈ X ⊥ , we get X

hv, wi = ci hv, wi i = 0
i
7.3. PROJECTIONS 125

Now take w ∈ span(X). Take a sequence (w)n ∈ span(X) such that (w)n −→ w. Thanks to the
continuity of the scalar product we have
hv, wi = hv, lim wn i = lim hv, wn i = 0
n→∞ n→∞

Demonstrating that X ⊥ = (span(X))⊥

Lemma 7.3.1. Let V be a Hilbert space. Given W ⊂ V a closed subspace. Given v ∈ V
∃!w0 ∈ W : ∀w ∈ W d = kv − w0 k ≤ kv − wk

Proof. Take d = infw∈W kv − wk. By definition of infimum we have that ∃(w)n ∈ W such that
lim kv − wn k = d
n→∞

Using the parallelogram rule, we have that

2
2 2 2 2 1
kwn − wk k = k(wn − v) + (v − wk )k = 2kv − wn k + 2kv − wk k − 4 (wn + wk ) − v
2
Since 1/2(wn + wk ) ∈ W we have by definition of d
1
(wn + wk ) − v ≥ d
2
Therefore, we can rewrite
2 2 2
kwn − wk k ≤ 2kv − wn k + 2kv − wk k − 4d2
Therefore, we have
2
∀ > 0 ∃N ∈ N : ∀n, k ≥ N kwn − wk k ≤ 4(d + )2 − 4d2
Hence (w)n is a Cauchy sequence. Since by definition V is complete and W ⊂ V is closed, we have
that W is also complete, therefore (w)n → w0 ∈ W and we have
kv − w0 k = d
Now suppose that ∃w1 , w2 ∈ W such that the previous is true, i.e.
kv − w1 k = kv − w2 k ≤ kv − wk ∀w ∈ W
Taken w3 = 1/2(w1 + w2 ) we have
2 2 1 2
kv − w3 k = kv − w1 k − kw2 − w1 k
4
Taken d = kv − w1 k = kv − w2 k, z1 = v − w3 and z2 = 1/2(w1 − w2 ) we get
 
2 2 2 2
kz1 + z2 k + kz1 − z2 k = 2 kz1 k + kz2 k

And therefore
1 2 2

2 1 2
d2 = kv − w1 k + kv − w2 k = kv − w3 k + kw1 − w2 k
2 4
I.e. if w1 6= w2 , w3 is the infimum between v ∈ V and W
CHAPTER 7. HILBERT AND BANACH SPACES 126

§§ 7.3.2 Projections and Orthogonal Projections

Theorem 7.12 (Projection). Given W ⊂ V closed subspace of a Hilbert space, we have
v =w+z ∀v ∈ V, w ∈ W, z ∈ W ⊥

Proof. Given v ∈ V, due to the previous lemma we have that ∃!w ∈ W such that
d = kv − wk ≤ kv − w0 k ∀w0 ∈ W
Taken z = v − w, and an element x ∈ W, define the vector w + tx with t ∈ C. Since W is a subspace
w + tx ∈ W and ∀t ∈ C we have
2 2 2 2
d2 ≤ kv − (w + tx)k = kv − wk − thv − w, x, −ithx, v − wi + ktk kxk
Writing hx, v − wi = khx, v − wikeiθ and t = se−iθ with s ∈ R we have
2
−2skhv − w, xik + s2 kxk ≥ 0 ∀s ∈ R
Which implies
hv − w, xi = 0 =⇒ z = v − w ∈ W ⊥
Therefore there exists a representation v = w + z with w ∈ W, z ∈ W ⊥ Now, we suppose that
v = w0 + z 0 , then
0 = (w − w0 ) + (z − z 0 )
Therefore
2 2
0 = k(w − w0 ) + (z − z 0 )k = kw − w0 k + kz − z 0 k
Therefore the representation is unique.
Theorem 7.13. If W ⊂ V with V a Hilbert space, we have

⊥
W⊥ = W
If W is closed
⊥
W⊥ =W

Proof. Taken w ∈ W we have that w ⊥ v with v ∈ W ⊥ , therefore w ∈ (W ⊥ )⊥ . Therefore

⊥
W ⊂ W⊥
Since the space on the right is closed, we have
⊥
⊥
W = (W ⊥ ) = W ⊥

⊥
Now taken w ∈ W ⊥ , since W is a closed subspace by definition, we can write
⊥
w =v+z v ∈ W, z ∈ W = W⊥
We have w ⊥ z, and therefore
2
kzk = hz, w − vi = 0 =⇒ w = v ∈ W
7.3. PROJECTIONS 127

Definition 7.3.3 (Orthogonal Projection). Given a closed subspace W ⊂ V we can define an operator
π̂W : V −→ W such that
π̂W v = w ⇐⇒ w ∈ W, v − w ∈ W ⊥ (7.23)
π̂W is linear and called a orthogonal projection

Theorem 7.14. Given W ⊂ V a closed subspace of the Hilbert space V, then given an orthogonal
projection π̂W : V −→ W we have, ∀v, z ∈ V and another closed subspace Z ⊂ V

1. π̂W
2
= π̂W

2. hπ̂W v, zi = hv, π̂W zi ≥ 0

3. If Z ⊆ W ⊥ π̂W ◦ π̂Z = π̂Z ◦ π̂W = 0

4. If Z ⊂ W π̂W ◦ π̂Z = π̂Z ◦ π̂W = π̂W

Definition 7.3.4 (Direct Sum). An euclidean space V is called the direct sum of closed subspaces
Vi ⊂ V and it’s indicated as follows
∞
M
V = V1 ⊕ V2 ⊕ · · · = Vk (7.24)
k=1

If

1. The spaces Vk are orthogonal in couples

P∞
2. ∀v ∈ V v = k=1 vk with vk ∈ Vk

Corollary 7.3.1. Given a Hilbert space V and a closed subspace W, then

V = W ⊕ W⊥ (7.25)

§§ 7.3.3 Orthogonal Systems and Bases

Definition 7.3.5 (Orthogonal System). A set of vectors X ⊂ V X 6= {} is said to be an orthogonal
system if ∀u, v ∈ X, u 6= v u ⊥ v.

Definition 7.3.6 (Orthonormal System). Given an orthogonal system X ⊂ V such that ∀u ∈ X we

have kuk = 1, the system X is called an orthonormal system

Theorem 7.15. Given an orthogonal system X ⊂ V with (V, h·, ·i) an euclidean space, then we
have that X is a system of linearly independent vectors

Definition 7.3.7 (Basis). Given an orthogonal and complete set of vectors (v)α in an euclidean space V
it’s said to be an orthogonal basis of V. If it’s an orthonormal and complete set of vectors it’s said to
be an orthonormal basis of V
CHAPTER 7. HILBERT AND BANACH SPACES 128

Lemma 7.3.2. Given an orthogonal system (v)nk=1 ∈ V and let u ∈ V an arbitrary vector. Then given
z ∈ V as follows
n
X hu, vk i
z := u − 2 vk
k=1
kvk k
We have that z ⊥ vi ∀1 ≤ i ≤ n and therefore z ⊥ span {v1 , · · · , vn }
Proof. ∀i = 1, · · · , n it’s obvious that hz, vi i = 0, therefore
⊥
z ∈ {v1 , · · · , vn }
Therefore
⊥ ⊥ ⊥
z ∈ {v1 , · · · , vn } = span {v1 , · · · , vn } = span {v1 , · · · , vn }

Theorem 7.16 (Gram-Schmidt Orthonormalization). Given V an euclidean space and (v)n∈N ∈ V a

set of linearly independent vectors. Then ∃(u)n∈N ∈ V orthonormal system such that
1. un is a linear combination of vi ∀0 ≤ i ≤ n, i.e.
n
X
un = ank vk ann 6= 0
k=1

2. vn can be written as follows

n
X
vn = bnk uk bnn 6= 0
k=1

Therefore, ∀n ∈ N we have that

span {v1 , · · · , vn } = span {u1 , · · · , un }

Proof. Defining
n−1
X hvn , vk i wn
wn = vn − 2 wk un =
kwk k kwn k
k=1
⊥
We can say immediately that ∀n ≥ 1 wn ∈ {w1 , · · · , wn−1 } . By induction we can say that it holds
∀(w)n∈N , therefore (w)n is an orthogonal system and (u)n∈N is an orthonormal system
We can also say that
k−1
X
vn = wn + βnj wj
j=1

I.e. ∀n ≥ 1 wn is a linear combination of {v1 , · · · , vn }, therefore, by definition

span {v1 , · · · , vn } = span {w1 , · · · , wn } = span {u1 , · · · , un }
7.3. PROJECTIONS 129

Example 7.3.1. 1) Legendre Polynomials

Using the Gram-Schmidt orthonormalization procedure, we can find an orthonormal system {p0 , · · · , pn } ⊂
C2 [−1, 1] starting from the following system
(v)n := 1, x, x2 , x3 , x4 , · · · , xn


The final result will be called the Legendre Polynomials We begin by taking the canonical scalar
product in C2 [−1, 1] as follows
ˆ 1
hf, gi = f (x)g(x) dx
−1
And using that
ˆ 2

1  n = 2k ∈ N
xn dx = n+1
−1 
0 n = 2k ∈ N
Therefore, we have that
ˆ 1
2
w0 = 1 kw0 k = dx = 2
−1
ˆ 1
1 2 −1 2 2
w1 = x − x dx = x kw1 k = x dx =
2 −1 1 3
ˆ 1  2
1 2 1 8
w2 = x2 − kw2 k = x2 dx =
3 −1 3 45
..
.
Normalizing, we find
1
p0 = √
2
r
3
p1 (x) = x
2
r   r
45 2 1 1 5
3x2 − 1


p2 (x) = x − =
8 3 2 2
..
.
And so on. The set (p)n is called the set of Legendre polynomials 2) Hermite Polynomials
2
Using the same procedure, we can find the Hermite polynomials Hn (x) in the space C2 (R, e−x dx).
The first 5 are the following
H1 = 1
H2 (x) = x
1
H3 (x) = x2 −
2
3 3
H4 (x) = x − x
2
3
H5 (x) = x4 − 3x2 +
4
CHAPTER 7. HILBERT AND BANACH SPACES 130

Theorem 7.17 (Existence of an Orthonormal Basis). Given a separable or complete euclidean

space V, there always exists an orthonormal basis
Proof. Taken V a separable euclidean space and (v)n∈N a dense subset of V.
Removing the linearly independent elements of this subset, we can call the new linearly independent
set (w)n∈N . We have obviously

span {(w)∞ ∞
n=1 } = span {(v)n=1 }

span {(w)∞ ∞
n=1 } = span {(v)n=1 } = V

Orthonormalizing the system (w)n∈N with the Gram-Schmidt procedure I obtain then a new set (u)n∈N
such that
span {(u)n∈N } = span {(w)n∈N } = V
(u)n∈N is complete and therefore a basis.
Remark. If V is a complete euclidean space but not separable, we can find thanks to Zorn’s lemma a
maximal orthonormal basis, but
1. There isn’t a standard procedure for finding this basis
2. Taken the basis (u)α∈I it can’t be numerable. If it was then V = span {(u)α }. Taken X =
spanQ {(u)k } as the set of finite linear combinations with rational coefficients, we have that
X = span {(u)k } and therefore X = V contradicting the fact that V is not separable
8 Distributions

§ 8.1 Linear Functionals

§§ 8.1.1 Dual Spaces and Functionals
Definition 8.1.1 (Dual Space II). Given (V, k·k) a normed vector space over a field F, we define the
following set
V ? := { f : V −→ F| v 7→ f (v)}
An element f ∈ V ? is called a continuous linear functional, with the following properties, ∀f, g ∈
V ? , u, v ∈ V, λ, λ1 , λ ∈ F
f (λ1 u + λ2 v) = λ1 f (u) + λ2 f (v)
(f + g)(u) = f (u) + g(v) (8.1)
(λf )(u) = f (λu) = λf (u)
With these properties the set V has a vector space structure, and it’s called the dual space
?

Definition 8.1.2 (Bounded Linear Functional). Given f ∈ V ? , we call f a bounded linear functional
if, ∀x ∈ V
sup |f (x)| < ∞ (8.2)
kxk≤1

Definition 8.1.3 (Dual Norm). Given a normed vector space (V, k·k) we can give a structure of normed
vector space to its dual with the couple (V ? , k·k? ), where the application k·k? is called the dual norm

k·k? : V ? −→ F

The dual norm is defined in two ways. ∀f ∈ V ? , v ∈ V

|f (v)|
kf k?1 = sup (8.3a)
v6=0 kvk

kf k?2 = sup |f (x)| (8.3b)

kvk≤1

Where f ∈ V ? is a bounded linear functional.

131
CHAPTER 8. DISTRIBUTIONS 132

Theorem 8.1. For a bounded linear functional f ∈ V ? the two definitions of the dual norm
coincide, i.e.
kf k?1 = kf k?2 = kf k?
And we can define the following inequality

|f (v)| ≤ kf k? kvk ∀v ∈ V (8.4)

Proof. Since f ∈ V ? is bounded we have that both norms exist and must be finite. We can therefore
write, thanks to the homogeneity of f , taken v ∈ V
 
v
kf k?1 = sup f = sup |f (v)| ≤ sup |f (v)| = kf k?2
v6=0 kvk kvk=1 kvk≤1

I.e. kf k?1 ≤ kf k?2 , analogously

|f (v)| |f (v)|
kf k?2 = sup |f (v)| ≤ sup ≤ sup = kf k?2
kvk≤1 0<kvk≤1 kvk v6=0 kvk

I.e. kf k?1 ≥ kf k?2 , therefore we have

kf k?1 = kf k?2 = kf k?

The inequality is obvious taken the definition of supremum

Theorem 8.2. Given f ∈ V ? with V normed vector space, we have that the following assump-
tions are equivalent

1. f is continuous

2. f is continuous at the origin

3. f is bounded

Proof. 1) =⇒ 2)
Since f ∈ V ? is linear by definition, we have that it’s also continuous and injective, therefore

lim f (v) = f (0) = 0

v→0

2) =⇒ 3)
Since f is continuous at the origin, we have by definition of continuity and limit

∀ > 0 ∃δ > 0 : kxk < δ =⇒ |f (x)| <  x∈V

Taken u = δx ∈ V, we have kuk = |δ|kxk and

u 1
|f (x)| = f = |f (u)| ≤ 
δ δ
8.1. LINEAR FUNCTIONALS 133

Therefore, if kxk ≤ 1 we have that |f (u)| ≤ δ, therefore

|f (u)| ≤ kf k? kuk =⇒ kf k? kxk ≤ kf k? ≤ 

3) =⇒ 1) By definition of continuity and boundedness we have

∀v, w ∈ V, ∀ > 0 ∃δ > 0 : ku − wk < δ =⇒ |f (u) − f (w)| < 

Through the linearity of f we have

|f (u) − f (w)| = |f (u − w)| ≤ kf k? ku − wk ≤ δkf k?

−1
Taken δ = kf k? , we have
|f (u) − f (v)| ≤ δkf k? = 

Corollary 8.1.1. Given f ∈ V ? , we have that, given C ∈ F a constant

(
∀x ∈ V |f (x)| ≤ Ckxk
kf k? = C ⇐⇒
∀ > 0 ∃x ∈ V : |f (x)| ≥ (C − )kxk

Proof. We have, by definition of kf k?

kf k? = sup |f (x)| ≤ sup |ckxk| = C

kxk≤1 kxk≤1

In the second case, supposing kf k? < C and taken  = 1/2(C − kf k? )

∃x ∈ V : |f (x)| ≥ (C − kf k? )kxk

Definition 8.1.4 (Kernel). Given a linear functional f ∈ V ? we define the kernel of f as the set of
zeros of f , i.e.
ker f = { v ∈ V| f (v) = 0} (8.5)

Theorem 8.3 (Riesz Representation Theorem). Given (V, h·, ·i) a Hilbert space, we can uniquely
∼
define its dual V ? through the isomorphism φv : V −→ F defined as follows

∀u, v ∈ V φv (u) = hu, vi

I.e. v 7→ h·, vi.

This isomorphism has the following properties

1) ∀v ∈ V φv ∈ V ? , kφv k? = kvk
2) ∀c1 , c2 ∈ F, v1 , v2 ∈ V φc1 v1 +c2 v2 = c1 φv1 + c2 φv2 (8.6)
?
3) f ∈ V =⇒ ∃v ∈ V : f = φv
CHAPTER 8. DISTRIBUTIONS 134

Proof. Taken w, v, z ∈ V and c, s ∈ F we have by definition

φv (w + z) = hw + z, vi = hw, vi + hz, vi = φv (w) + φv (z)

φv (cw) = hcw, vi = chw, vi = cφv (w)
φcw+sz (v) = hv, cw + szi = chv, wi + shv, zi = cφw (v) + sφz (v)

We also have, thanks to Cauchy-Schwartz

|φv (z)| = |hz, vi| ≤ kzkkvk

2
|φv (v)| = |hv, vi| ≤ kvk ∴ kφv k? = kvk

For the last one we take f ∈ V ? . If ker f = V we have by definition that

f (v) = 0 ∀v ∈ V ∴f =0

Taken v = 0 we then have

φ0 (v) = h0, vi = 0 = f (v) ∀v ∈ V

Supposing now ker f 6= V there must be w ∈ ker f ⊥ : w 6= 0. Taken u ∈ V we can write

f (u) f (u) f (u)

u = u1 + w =u− w+ w
f (w) f (w) f (w)

Therefore, through the linearity of f

 
f (u)
f (u1 ) = f u− w = f (u) − f (w) = f (u − v) = 0 ∴ u1 ∈ ker f
f (w)

Since w ∈ ker f ⊥ , we have that

hu1 , wi = 0

By definition of φv , then

f (u) f (w) f (w) f (u)

φv (u) = hv, u1 i + hv, wi = h 2 w, u1 i + h 2 w, f (w) wi
f (w) kwk kwk

Then
f (w) f (w) f (u)
φv (u) = 2 hw, u1 i + 2 hw, wi = f (u)
kwk kwk f (w)

Therefore
φv (u) = f (u) ∀f ∈ V ? , ∀u ∈ V
8.2. DISTRIBUTIONS 135

§ 8.2 Distributions
§§ 8.2.1 Local Integrability
Definition 8.2.1 (Weak Convergence). Given a normed vector space (V, k·k), a sequence (v)k∈N ∈ V
is said to be weakly convergent to v ∈ V and it’s indicated as vk *w v if

∀f ∈ V ? lim f (vk ) = f (v)

k→∞

Theorem 8.4. Given a normed vector space V and a sequence (v)k∈N ∈ V such that vk → v ∈ V
(i.e. converges strongly), we have that

vk → v =⇒ vk *w v

Proof. Given vk → v, we have by definition that

vk → v ⇐⇒ lim kvk − vk = 0
k→∞

Now, writing the definition of weak convergence and applying the linearity of the limit

vk *w v ⇐⇒ ∀f ∈ V ? lim (f (vk ) − f (v)) = 0

k→∞

Therefore
∀f ∈ V ? lim (f (vk − v)) ≤ kf k? lim kvk − vk
k→∞ k→∞

Therefore we have that, since f is bounded

∃C ∈ F : lim (f (vk − v)) ≤ C lim kvk − vk ∀f ∈ V ?

k→∞ k→∞

And therefore
vk → v =⇒ vk *w v

Remark. The opposite isn’t necessarily true.

Proof. Take (V, k·k) = (`2 , k·k2 ) and take the standard sequence (ei )j where

(ei )j = (0, 0, · · · , 0, (1)j , 0, · · · )

|{z}
i−th

And (1)j is the identity sequence.

We have that (ei )j 6→ (a)i since
√
k(ei )j − (ei )k k2 = 2 ∀j 6= k
CHAPTER 8. DISTRIBUTIONS 136

Now, considering that (`2 , k·k2 ) is a Hilbert space, we have that, due to Riesz representation theorem

∀f ∈ `2? f = h·, ai = φa a ∈ `2

Therefore
∞
X
f (eij ) = heij , ai i = ai eij = aj → 0
i=1

Therefore
∀(a)j ∈ `2 ∀f ∈ `2? lim f (eij ) = 0
j→∞

Hence, by definition eij *w 0

Definition 8.2.2 (Isolated Singularities II). Given a function f : R −→ C, it’s said to have isolated
singularities if  
Sf := x ∈ R| lim f (y) 6= f (x)
y→x

Doesn’t have accumulation points, i.e. if |Sf | < ∞

Definition 8.2.3 (Piecewise Continuity). A function f : [a, b] ⊂ R −→ C is said to be piecewise
continuous if
1. Sf 6= {} and |Sf | < ∞
2. ∀u ∈ Sf
∃ lim+ f (x) = f (u+ )
x→u
(8.7)
∃ lim− f (x) = f (u− )
x→u

3.
∃ lim+ f (x) = f (a+ )
x→a
(8.8)
∃ lim− f (x) = f (b− )
x→b

Definition 8.2.4 (Jump). Given a piecewise continuous function f we define the jump of the function
at a discontinuity x ∈ Sf as follows

∆f (x) = f (x+ ) − f (x− )

Definition 8.2.5 (Piecewise Differentiability). A piecewise continuous function f : [a, b] −→ C is said to

be piecewise differentiable if and only if
1. f is piecewise continuous in [a, b]
2. f 0 is piecewise continuous in [a, b]
Definition 8.2.6 (Local Integrability). A function f : R −→ C is said to be locally integrable if
8.2. DISTRIBUTIONS 137

1. Sf 6= {} and |Sf | < ∞

2. ∀a, b ∈ R, a < b we have that
ˆ b
|f (x)| dx < ∞
a

The set of locally integrable functions forms a subspace of L1 (R) and it’s indicated as L1loc (R)
Theorem 8.5. Let f : R −→ C be a piecewise continuous function, then f ∈ L1loc (R)
Proof. Take f (x) a piecewise continuous function, then ∀a, b ∈ R ∃(u)0≤k≤n ∈ R : u0 = a, un = b
for which we have f ∈ C((uk−1 , uk )) ∀0 ≤ k ≤ n. I can therefore define f̃k : [uk−1 , uk ] −→ C, where

x ∈ (uk−1 , uk )

f (x)

+
f̃k (x) = f (uk−1 ) x = uk−1

f (u−

k) x = uk

We have that
∞
X
f (x) = f̃k (x)
k=0

And therefore ˆ b n ˆ
X uk n ˆ
X uk
|f (x)| dx = |f (x)| dx = f̃k (x) dx
a k=0 uk−1 k=0 uk−1

Therefore f (x) ∈ L1loc (R) and the theorem is proved

Theorem 8.6 (Integration by Parts in L1loc (R)). Let f, g ∈ L1loc ([a, b]), then indicating the evaluation
of a function at two points as [f (x)]w z we have
ˆ b
−
[f g]ba+ = (f 0 (x)g(x) + f (x)g 0 (x)) dx+
a
X (8.9)
f (x− )∆g(x) + g(x− )∆f (x) + ∆f (x)∆g(x)


+
x∈Sf ∪Sg

Proof. Take H : [a, b] −→ C, then for the fundamental theorem of calculus

ˆ b
− + dH X
H(b ) − H(a ) = dx + ∆H(x)
a dx x∈SH

Taken H(x) = f (x)g(x) we have

∆(f g)(x) = f (x+ )g(x+ ) − f (x− )g(x− )

With some manipulation we have

∆(f g)(x) = f (x+ )g(x+ ) − f (x+ )g(x− ) + f (x+ )g(x− ) − f (x− )g(x− ) = f (x+ )∆g(x) + g(x− )∆f (x) =
= f (x− ) + ∆f (x) ∆g(x) + g(x− )∆f (x) = f (x− )∆g(x) + g(x− )∆f (x) + ∆f (x)∆g(x)


CHAPTER 8. DISTRIBUTIONS 138

Therefore
ˆ b X
(f g)(b− ) − (f g)(a+ ) = (f g)0 (x) dx + f (x− )∆g(x) + g(x− )∆f (x) + ∆f (x)∆g(x)


a x∈Sf ∪Sg

§§ 8.2.2 Regular and Singular Distributions

Definition 8.2.7 (Test Functions). A function f : R −→ C is said to be a test function if supp{f } is

compact and f ∈ C ∞ (R), i.e. f ∈ Cc∞ (R)
This space is usually denoted as follows Cc∞ (R) = K

Definition 8.2.8 (K-convergence). Given a sequence (f )n∈N ∈ K, it’s said to be K−convergent if

1. ∃I ⊂ R : ∀x ∈ I c fn (x) = 0
(k)
2. ∀k ∈ N fn (x) ⇒ f (n) (x)

Then, it’s indicated as fn *K f

Definition 8.2.9 (Distribution). A distribution is a continuous linear functional ϕ : K −→ C, i.e.

∀(f )n ∈ K, fn *K f =⇒ ϕ(fn ) = ϕ(f )

By definition of dual space, we have that ϕ ∈ K?

Theorem 8.7. Given g ∈ L1loc (R) and ϕg : K −→ C, if we have

ˆ
ϕg (f ) = g(x)f (x) dx ∀f ∈ K
R

Then ϕg ∈ K?

Proof. Using the fact that f ∈ K we can immediately say that |f (x)| ≤ M ∀x ∈ [−a, a] ⊂ R, therefore
ˆ ˆ a ˆ a
|g(x)f (x|) dx = |g(x)f (x)| dx ≤ M |g(x)| dx < ∞
R −a −a

Alternatively, using the defintion of integral we can say that

ϕg (αf + βh) = αϕg (f ) + βϕg (h) ∀α, β ∈ C, ∀f, h ∈ K

We only need to show that this application is K−continuous. I take fn *K f , with fn ∈ K.

It’s obvious that fn ⇒ f , and using the linearity and that g ∈ L1loc (R), calling A = kgk, we have

|ϕg (f ) − ϕg (fn )| ≤ Akf − fn ku → 0

Therefore ϕg (fn ) → ϕg (f ) ∀fn ∈ K, fn *K f

8.2. DISTRIBUTIONS 139

Definition 8.2.10 (Regular Distribution). A distribution f ∈ K? is said to be regular if ∃g ∈ L1loc (R)

such that
f = h·, gi = ϕg
I.e. ˆ
f (h) = ϕg (h) = hh, gi = h(x)g(x) dx
R

Definition 8.2.11 (Ceiling and Floor Functions). We define the ceiling function d·e : R −→ Z and the
floor functions b·c : R −→ Z as follows

bxc = max { m ∈ Z| m ≤ x}
(8.10)
dxe = min { n ∈ Z| n ≥ x}

Definition 8.2.12 (Heaviside Function). We define the Heaviside function as a function H : R −→ R

as follows (
1 x≥0
H(x) =
0 x<0

It’s obviously piecewise continuous and therefore H ∈ L1loc (R)

A secondary definition is the one that follows H̃ : R −→ R
(
1 x>0
H̃(x) =
0 x≤0

Example 8.2.1 (Floor and Ceiling Distributions). Since both the floor and ceiling distributions are locally
integrable, we can build a regular distribution ϕ ∈ K? as follows
ˆ Xˆ k+1
ϕbxc (f ) = f (x)bxc dx = kf (x) dx
R k∈Z k
ˆ Xˆ k+1
ϕdxe (f ) = f (x)dxe dx = (k + 1)f (x) dx
R k∈Z k

Example 8.2.2 (Theta Distribution). Given H ∈ L1loc (R) we can define the associated theta distribu-
tion ϕH = ϑ ∈ K, as follows
ˆ ˆ
ϑ(f ) = f (x)H(x) dx = f (x) dx
R R+

It’s already obvious that ϕH̃ = ϕH , but it’s better to formalize it in the following theorem

Theorem 8.8. Let f, g ∈ L1loc (R) such that { x ∈ R| f (x) 6= g(x)} = {u1 , · · · , un }. Then we have
ϕg = ϕf ∈ K?
CHAPTER 8. DISTRIBUTIONS 140

Proof. We have by Riesz theorem that

ˆ ˆ u1 X ˆ uk+1
n−1 ˆ ∞
ϕg (f ) = g(x)f (x) dx = f (x)g(x) dx + f (x)g(x) + f (x) dx
R −∞ k=2 uk un

But, since f (x) = g(x), ∀x ∈ (uk , uk+1 ) we have

ˆ ˆ u1 X ˆ uk+1
n−1 ˆ ∞
f (x)g(x) dx = f (x)g(x) dx + f (x)g(x) dx + f (x)g(x) dx =
R −∞ k=2 uk un
ˆ u1 X ˆ uk+1
n−1 ˆ ∞ ˆ
= f (x)h(x) dx + f (x)h(x) dx + f (x)h(x) dx = f (x)h(x) dx
−∞ k=2 uk un R

Therefore
ϕg = ϕh

Definition 8.2.13 (Singular Distribution). A singular distribution is a distribution f ∈ K? for which,

given g ∈ L1loc (R), f 6= ϕg , where ˆ
ϕg (h) = h(x)g(x) dx
R

Definition 8.2.14 (Dirac Delta Distribution). An example of singular distribution is the Dirac delta
distribution δa ∈ K? . This distribution is defined as follows

δa (f ) = f (a) ∀f ∈ K, a ∈ R

Theorem 8.9. Given the Dirac delta distribution δa , @δ(x) ∈ L1loc (R) such that (taken a = 0
without loss of generality) ˆ
δ0 (f ) = f (x)δ(x) dx
R

Proof. Let’s say that ∃δ(x) ∈ L1loc (R), therefore, we could define δ0 = h·, δ(x)i. This function therefore
must have these properties
δ(x) 6= 0 ⇐⇒ x = 0, x ∈ Sδ
But, since for b ∈ R, b 6= 0, we have that δ(x) → δ(b) continuously, i.e. δ(b) = A > 0, therefore

A
∃ > 0 : δ(x) ≥ ∀x ∈ [b − , b + ]
2
But, then ∃f ∈ K such that f (x) → f (b) continuously, f > 0, f (b) = 1. Taken  < b we can say

1
∃˜ ∈ (0, ) : f (x) ≥ ∀x ∈ [b − 0 , b + 0 ]
2
8.2. DISTRIBUTIONS 141

Therefore, we have that

ˆ ˆ  ˆ 
A1 0 A
f (x)δ(x) dx = f (x)δ(x) dx ≥ dx =
R − − 22 2

But, by definition f (0) = 0, therefore

ˆ
f (x)δ(x) dx = 0 =⇒ ϕδ (f ) = ϕ0 (f )
R

Therefore, the distribution δ0 6= h·, δ(x)i

Notation (Common Abuse of Notation). A common abuse of notation in the usage of the Dirac delta
distribution, is supposing that it’s writable like a regular distribution, i.e. supposing that ∃δ(x) ∈ L1loc (R),
the “Dirac delta function” such that
ˆ
δa (f ) = f (x)δ(x − a) dx = f (a) ∀f ∈ K
R

This is obtainable only if the “delta function” is defined as follows

(
+∞ x = 0
δ(x) = (8.11)
0 x 6= 0

Note that ˆ
δ0 (1) = δ(x) dx = 1
R

Theorem 8.10. Given gn ∈ L1loc (R) such that gn ∈ C[−an , an ] with an → 0, an > 0 ∀n ∈ N, we
have that, if ˆ
gn (x) dx = 1
R
Then
lim ϕgn (f ) = δ0 (f )
n→∞

Proof. We use the definition of limit, therefore we have

ˆ ˆ an
f (x)gn (x) dx − δ0 (f ) ≤ |f (x) − f (0)|gn (x) dx
R −an

Using that the integral over the real axis of gn is unitary

ˆ
f (x)gn (x) dx − δ0 (f ) ≤ sup |f (x) − f (0)| → 0
R |x|≤an

Therefore
lim ϕgn (f ) = δ0 (f )
n→∞
CHAPTER 8. DISTRIBUTIONS 142

Corollary 8.2.1. Given g ∈ C1 (R) a non-negative function g ≥ 0 such that

ˆ
g(x) dx = 1
R

Then, if we put gn (x) = ng(nx) we have that

lim ϕgn (f ) = δ0 (f )
n→∞

Proof. As before, we use the definition, and therefore we have

ˆ ˆ ˆ
f (x)gn (x) dx − δ0 (f ) ≤ |f (x) − f (0)|gn (x) dx = n |f (x) − f (0)|g(nx) dx
R R R

Using the substitution u = nx we therefore get

ˆ ˆ u
|f (x) − f (0)|g(nx)n dx = f − f (0) g(u) du
R R n
But, by definition of g(x), we have that ∃L > 0 : ∃ > 0 for which
ˆ L ˆ
1−≤ g(u) du ≤ 1 ∧ g(u) du ≤ 
−L |u|≥L

And therefore
ˆ u ˆ L   ˆ
u u
f − f (0) g(u) du = f − f (0) g(u) du + f − f (0) g(u) du
R n −L n |u|≥L n
By using the properties of the integral, we know that
ˆ u  u 
f − f (0) g(u) du ≤ sup |f (x) − f (0)| +  sup f − |f (0)|
R n x≤ L u∈R n
n

Since f (x) → f (0) continuously, due to the fact that f ∈ K, we have that
ˆ
|f (x) − f (0)|g(nx)n dx ≤ 2kf ku
R

Therefore ˆ
lim f (x)gn (x) dx − δ0 (f ) = 0
n→∞ R

Definition 8.2.15 (PV (x−n ) −Distribution). Taken n > 0, n ∈ N, we have that x−n ∈ / L1loc (x) and
?
therefore there is no associated distribution ϕx−n ∈ K .
A useful thing we could do is utilizing the definition of the Cauchy principal value of the function.
Therefore, we define the following singular distribution
  ˆ a n−1
X f (k) (0)
!
1 1 k
PV [f ] = lim f (x) − x dx (8.12)
xn a→∞ −a xn k!
k=0
8.2. DISTRIBUTIONS 143

Theorem 8.11. The application PV(x−n ) : K −→ R defined as before is a distribution, hence

PV(x−n ) ∈ K?
Proof. We already know that this distribution is linear, since ∀f, g ∈ K, ∀c, d ∈ R
  ˆ a n−1
!
1 1 X 1 
(k) (k)

PV [cf + dg] = lim cf (x) + dg(x) + cf (0) + dg (0) xk dx =
xn a→∞ −a xn k!
k=0
   
1 1
= c PV [f ] + d PV [g]
xn xn
Secondly we must show that the integral is well defined.
Let f ∈ K, then
ˆ R n−1
!
n 1 X 1 (k)
IR (f ) = f (x) − f (0)xk dx, n ∈ N, R > 0
−R xn k!
k=0

Using the Lagrange formulation for the remainder of the McLaurin expansion, we have
n−1
X f (k) (0) k
Rn−1 (x) = f (x) − x = xn rn−1 (x)
k!
k=0

And therefore we have ˆ R

n
IR (x) = rn−1 (x) dx
−R

Since rn−1 (x) ∈ C ∞ (R) (it’s a polynomial) and supp f ⊂ [−b, b] for some b ∈ R, b > 0, we have that
for R > b
X f (k) (0) ˆ
n−1
1
n n
IR (f ) = Ib (f ) − n−k
dx
k! b≤|x|≤R x
k=0

The last integral evaluates to


ˆ
 
2 1 1
− n−k mod 2 = 0, n − k > 2

1 
n−k
dx = n − k − 1 bn−k−1 Rn−k−1
b≤|x|≤R x 
0 else

Therefore, introducing the following dummy index j = n − k where j mod 2 = 0, we have

n−1
f (j) (0)
 
1 X 2
PV (f ) = Ibn (f ) −
xn j=0
j! (j − 1)bj−1

Therefore the integral defining the distribution is well defined.

Lastly we need to prove that PV(x−n )(fn ) *K PV(x−n )(f ), ∀(f )n ∈ K : fn → f , i.e. that it’s a
continuous functional.
Going back to the previous definition we have that
Rn−1 (x) 1 dn−1 f
rn−1 (x) = = (ξ(x)) ξ(x) ∈ (0, x)
xn xn dxn−1
CHAPTER 8. DISTRIBUTIONS 144

Therefore
n 2R
|IR (f )| ≤ 2 sup |rn−1 (x)| ≤ kf ku
x∈[−R,R] n!
Which means
n−1
f (j) (0)
 
1 2b (n) X 2
PV (f ) ≤ f +
xn n! u
j=0
j! (j − 1)bj−1

Taken b1 = max {b, 1} we have that supp f ⊂ [−b1 , b1 ] and whenever j − 1 > 1, b1 ≥ 1

2
≤2
(j − 1)bj−1
1

Therefore, finally
n−1
f (j) (0)
 
1 2 X
PV (f ) ≤ max {b, 1} f (n) +2
xn n! u
j=0
j!

Now, defined fj ∈ K and hj (x) = f (x) − fj (x) with fj → f we have that supp hj ⊂ [−b, b], ∀j ∈ N.
(n)
By definition we have that hj ⇒ 0 and hj ⇒ 0 ∀n ∈ N

n−1 (j)

1

2 hj (0)
(n)
X
PV (hj ) ≤ max {b, 1} hj +2
xn n! u
j=0
j!

This means
(k) 
∀k ∈ N, ∀ > 0, ∃NR ∈ N : ∀k ≥ Nj hj (x) ≤
2(b + 1 + )
Chosen N = max0≤k≤n−1 Nk we have that ∀j ≥ N
 
  n−1
1 2 X 1
PV (hj ) ≤ max {b, 1} + ≤→0
xn 2(b + 1 + ) j=0
j!

Therefore, PV(x−n ) ∈ K?

§§ 8.2.3 Operations with Distributions

Definition 8.2.16 (Weak Derivative). Given u ∈ L1 ([a, b]), we define the weak derivative v ∈ L1 ([a, b])
if, ∀h ∈ Cc∞ ([a, b]) we have
ˆ b ˆ b
u(x)h0 (x) dx = − v(x)h(x) dx
a a

The function v(x) will then be identified as follows

v(x) = D u(x)
8.2. DISTRIBUTIONS 145

Theorem 8.12 (Operations with Distributions). Given ϕ, γ ∈ K? , f ∈ K, h ∈ C ∞ (R) and c ∈ C we

define the following operations in K?

+ : K? × K? −→ K?
· : C × K? −→ K?
(8.13)
◦ : C ∞ × K? −→ K?
D : K? −→ K?

Where, they act as follows

+(ϕ(f ), γ(f )) = (ϕ + γ)(f ) = ϕ(f ) + γ(f )

·(c, ϕ(f )) = (cϕ)(f ) = cϕ(f )
◦(h, ϕ(f )) = (hϕ)(f ) = ϕ(hf )
D ϕ(f ) = −ϕ(f 0 )

The last operation is the distributional derivative.

Theorem 8.13. Given g(x) ∈ L1loc (R) and ϕg ∈ K? its associated distribution. Then, ∀h ∈
C ∞, f ∈ K
(hϕg )(f ) = ϕg (hf ) = ϕhg (f ) (8.14)

Proof. The proof is quite straightforward

ˆ ˆ
(hϕg )(f ) = ϕg (hf ) = g(x) (h(x)f (x)) dx = f (x) (g(x)h(x)) dx = ϕhg (f )
R R

Theorem 8.14. Taken g ∈ L1loc (R) and g 7→ ϕg ∈ K? its associated distribution, we have that

D ϕg = ϕD g

Where D g is the weak derivative of g

Proof. ∀f ∈ K we have that

ˆ ˆ
0 0
D ϕg (f ) = −ϕg (f ) = − f (x)g(x) dx = f (x) D g(x) dx = ϕD g (f )
R R

Theorem 8.15. ∀g, f ∈ L1loc (R) and given ϕg , γf ∈ K? , c ∈ C their associated distributions, we
have that
ϕg + ϕf = ϕg+f
ϕcf = cϕf
CHAPTER 8. DISTRIBUTIONS 146

Proof. The proof is obvious using the linearity of the integral operator
Theorem 8.16 (Useful Identities). Here is a list of some useful identities
xδ0 = 0
 
1
xPV = ϕ1 (8.15)
x
x D δ0 = −δ0

Proof. 1) xδ0 = 0.
Taken f ∈ K
xδ0 (f ) = δ0 (xf ) = 0f (0) = 0
1


2) xPV xn = ϕ1
Again, taken g ∈ K
    ˆ
1 1
x PV (g) = PV (xg) = lim f (x) dx = ϕ1
x x →0 |x|≥0

For the last one, taken h ∈ K

x D δ0 (f ) = D δ0 (xf ) = −δ0 (f + xf 0 ) = −f (0) − 0f (0) = −f (0) = −δ0 (f )

Notation (Abuse of Notation). Given g ∈ L1loc (R) and its associated distribution g 7→ ϕg , is quite
common to use the original function to indicate actually the distribution. Together with this, the
distributional derivative D gets indicated as an usual derivative, therefore
ϕg → g(x)
dg
D ϕg → g 0 (x) =
dx
Therefore it isn’t uncommon to see identities like this
 
1
xPV =1
x
Where actually 1 → ϕ1 is the identity distribution.
Or
δ00 = −δ0
This makes an easy notation for calculating distributional derivatives and have some calculations, but
one should watch out to this common abuse of notation
Definition 8.2.17 (K? Convergence). Given (σ)n∈N ∈ K? a sequence of distributions, it’s say to
K? −converge if
(σ)n (f ) → σ(f ) ∈ K?
It’s indicated as follows
(σ)n *? σ
8.2. DISTRIBUTIONS 147

Theorem 8.17. Given κ ∈ K? a distribution, we have that f ∈ C ∞ (K) in the sense of distribu-
tional derivatives, where the n−th derivative is defined as follows
Dn κ(f ) = (−1)n κ(f (n) ) ∀f ∈ K

Proof. Taken f ∈ K we have that f ∈ Cc∞ (R) ⊂ C ∞ (R) hence f is smooth, therefore we have, for
κ ∈ K?
D κ(f ) = −κ(f 0 )
Iterating, we get for n = 2
D2 κ(f ) = − D κ(f 0 ) = κ(f 00 )
Iterating till n we get finally
Dn κ(f ) = (−1)n κ(f (n) )
Hence the derivability depends only on the test function f , and since it’s smooth, the distributional
derivatives can be defined ∀n ∈ N
Example 8.2.3. Take for example the delta distribution δ0 ∈ K? , given f ∈ K we want to calculate
the following value
Dn δ0 (f )
Using the previous identity, we have that
Dn δ0 (f ) = (−1)n δ0 [f (n) ] = (−1)n f (n) (0)

Theorem 8.18 (Chain Rule). Given h ∈ C ∞ (R) and η ∈ K? , we have that

D(hη) = h0 η + h D η
Where the distributional derivative is identical to the usual derivative for h and it’s the usual
distributional derivative for η
Proof. Taken h ∈ C ∞ , η ∈ K? and f ∈ K we have
(hη)(f ) = η(f h)
Therefore, using the identity for derivating a distribution we have that
D(hη)[f ] = −(hη)[f 0 ] =
= −η[hf 0 ] = −η[(hf )0 − h0 f ] = −η[(hf )0 ] + η[h0 f ] =
= (h D η)[f ] + (h0 η)[f ]

Theorem 8.19. Given gn (x) : R −→ C a continuous sequence of functions such that gn ⇒ g ∈

C(R). Then, taken gn 7→ ϕgn ∈ K? the associated distribution, we have that
ϕgn *? ϕg ∈ K?
CHAPTER 8. DISTRIBUTIONS 148

Proof. By definition of convergence, we have that, given f ∈ K, where supp f ⊂ [−a, a]

|ϕgn (f ) − ϕg (f )| ≤ ϕ|gn −g| (|f |) ≤ 2akgn − gku kf ku → 0
Therefore, ϕgn *? ϕg
Theorem 8.20. Given (η)n ∈ K? a sequence of distribution, then if (η)n *? η we have that
D ηn *? D η

Proof. By definition, we have

lim D ηn (f ) = − lim ηn (f 0 ) = −η(f 0 ) = D η(f )
n→∞ n→∞

Example 8.2.4 (The Absolute Value Distribution). Taken g(x) = |x| we have that (obviously) |x| ∈
L1loc (R), therefore there exists a distribution ϕ|x| ∈ K? defined as follows
ˆ
ϕ|x| (f ) = |x|f (x) dx
R

We have that the distributional derivative it’s actually the function sgn (x), (in this case, since it’s a locally
integrable function it coincides with its weak derivative), therefore it’s not unusual to see expressions
like this
d
|x| = sgn(x) weakly/distributionally
dx
The proof of this is quite easy. Taken f ∈ K
ˆ
D ϕ|x| (f ) = −ϕ|x| (f 0 ) = − |x|f 0 (x) dx =
R
ˆ ˆ
= [xf (x)]R− − [xf (x)]R+ + f (x) dx − f (x) dx =
R−
ˆ R+

= sgn(x)f (x) dx = ϕsgn(x) (f )

R

Since we have that

D ϕg = ϕD g
Where D g is intended as the weak derivative of g, we have that
D |x| = sgn(x)
Where in order to emphasize tha this is a weak derivative, one could use the notation
Dw |x| = sgn(x)
Note that in literature it’s common to use the abuse of notation written beforehand in general and in
this particular case.
Also note that if the function we had used would have been an ordinarily derivable function, the weak
derivative would have coincided with the ordinary derivative.
8.2. DISTRIBUTIONS 149

Theorem 8.21. In general, given g ∈ C 1 (R), we can say that g(|x|) has the following weak
derivative
D g(|x|) = D g(|x|) sgn(x)

Proof. Since g(|x|) ∈ L1loc (R) we can say that exists g 7→ ϕg ∈ K? such that
D ϕg(|x|) = ϕD g(|x|)
Where D g is the weak derivative of g(|x|). Since it’s a derivative (and it’s also demonstrable) we can
use the rules for composite derivation, and knowing that D |x| = sgn(x) we have
D ϕg(|x|) = ϕD g(|x|) = ϕD g D |x| = ϕsgn D g

Theorem 8.22 (Derivative of the ϑ Distribution). Given ϑ ∈ K? the theta distribution, we define
ˆ ˆ
ϑ(f ) = f (x)H(x) dx = f (x) dx
R R+

The derivative of this distribution is

D ϑ = δ0

Proof. We have ∀f ∈ K
ˆ ˆ
0 0
D ϑ(f ) = −ϑ(f ) = − f (x)H(x) dx = − f 0 (x) dx = f (0) = δ0 (f )
R R+

Therefore
D ϑ = δ0
Or, using a common abuse of notation
ϑ = ϑ(x) = H(x), D H(x) = δ(x)

Notation (Piecewise Derivative). Given g : R −→ C a piecewise differentiable function, we define the

differential operator D◦ as follows
(
g 0 (x) ∃g 0 (x)
D◦ g(x) = (8.16)
0 @g 0 (x)

Theorem 8.23. Given f : R −→ C a piecewise differentiable function and Sf := {u1 , · · · , uk }

isolated singularities, then
∞
X
D f (x) = D◦ f (x) + ∆f (ui )δ(ui )
i=1
CHAPTER 8. DISTRIBUTIONS 150

Proof. Taken Sf := {u} without loss of generality, we have that

lim f (x) = f (u+ )

x→u+
lim f (x) = f (u− )
x→u−
∆f (u) = f (u+ ) − f (u− )

And, ∀g ∈ K, taken f 7→ ϕf ∈ K?
ˆ u ˆ ∞
0 0
D ϕf (g) = −ϕf (g ) = − g (x)f (x) dx − g(x)f (x) dx
−∞ u

Integrating by parts and rebuilding the definition of ∆f (u) we have

ˆ
D ϕf (g) = g(u)∆f (u) + g(x) D◦ f (x) dx = ϕD◦ g (f ) + ∆f (u)δu (f )
R

And therefore
D f (x) = D◦ f + ∆f (u)δ(u)

Example 8.2.5 (Deriving the Sign Function). Take a ∈ R and the function sgn(x − a), using the
previous formula we have that Ssgn(x−a) = {a} and ∆ sgn(a) = 2, therefore

D sgn(x − a) = D◦ sgn(x − a) + 2δ(x − a)

Since D◦ sgn(x − a) = 0 we have finally

D sgn(x − a) = 2δ(x − a)

And equivalently
D sgn(a − x) = −2δ(x − a)

Example 8.2.6 (A General Piecewise Differentiable Function). Take the function g : R −→ R defined
as follows (
1 x<0
g(x) =
x−2 x≥0
We have that (
0 x≤0
D◦ g(x) = H(x) =
1 x>0
Since the discontinuity is in the origin and we have ∆g(0) = −3 we have

D g(x) = H(x) − 3δ(x)

8.2. DISTRIBUTIONS 151

Example 8.2.7 (Derivative of the Floor Function). Given the floor function bxc : R −→ Z we have that
Sbxc = Z, ∆bxc = 1 and D◦ bxc = 0 therefore

X ∞
X
Dbxc = δ(x − k) = δ(x − k)
k∈Z k=−∞

n √ o
Given instead a different function, bx2 c we have that D◦ bx2 c = 0 and Sbx2 c = k ∈ Z| ± k ⊂R
√
with ∆b± kc = ±1, therefore

√ ∞ √
X √ X √
Dbx2 c = δ(x − k) − δ(x − −k) = δ(x − k) − δ(x − −k)
k≥1 k=1

Theorem 8.24. Given h ∈ C ∞ (R) a smooth function and a ∈ R, we have that

n  
n
X n (k)
h(x) D δ(x − a) = (−1) k
h (a) Dn−k δ(x − a)
k
k=0

Proof. Taken f ∈ K we have that

 
(h(x) Dn δa )(f ) = Dn δa (hf ) = δa (hf )(n)

Using now Leibnitz’s composite derivation rule, we have

n   k
dn X n d h dn−k f
h(x)f (x) =
dxn k dxk dxn−k
k=0

We have that " #

n  
n
X n
(h(x) D δa )(f ) = δa (−1)n h(k) (x)f (n−k) (x) =
k
k=0
n  
X n
= (−1)k h(k) (x) Dn−k δa (f )
k
k=0

Example 8.2.8. Take as an example the following distributional derivative

D7 xex D2 δ0

We have from that formula that

d2 d x
xex D2 δ0 = (xex )δ0 = (e + xex )δ0 = (2 + x)ex δ0
dx2 dx
CHAPTER 8. DISTRIBUTIONS 152

And therefore h0 (0) = 1, h00 (0) = 2

Which means that
     
2 2 0 2 00
xex D2 δ0 = h(0) D2 δ0 − h (0) D δ0 + h (0)δ0
2 1 0

Inserting the values, we get

xex D2 δ0 = 2 D δ0 − 2δ0
The derivative is now trivial, and we get

D7 xex D2 δ0 = 2 D7 (δ0 − D δ0 ) = 2 D7 δ0 − D8 δ0

Theorem 8.25. Given k, m ∈ N and a function f ∈ C k (R) we have that


0
 k<m
Dk (xm f ) (0) =
 
k (
 m m!f k − m)(0) k ≥ m


And
Di (xm )(0) = m!δm
i

Proof. We apply immediately the Leibnitz chain rule and we have

m    
X k k
Dk (xm f ) (0) = f (k−j) (0)m!δjm = m! f (k−m) (0) ∀k ≥ m
j m
j=0

Theorem 8.26 (Properties of the PV(x−n ) Distribution). Here there will be listed some properties
of the PV(x−n ) distribution
1. D log |x| = PV(x−n )
2. Given n, m ∈ N, then 
m−n
  x
 m≥n
1
xm PV =
 
1
xn PV
 m<n
xn−m

3. Dm PV(x−n ) = (−m)!PV(x−n−m )
Proof. 1) Taken log |x| ∈ L1loc (R) we know that ∃ϕlog |x| ∈ K? such that log |x| 7→ ϕlog |x| and therefore
we can write as follows the previous derivative, that ∀f ∈ K
ˆ  ˆ ∞
D ϕlog |x| (f ) = −ϕlog |x| (f 0 ) = − lim f 0 (x) log(−x) dx − lim f (x) log(x) dx
→0− −∞ →0+ 
8.2. DISTRIBUTIONS 153

Integrating by parts we get

ˆ  
f (x) 1
D ϕlog |x| (f ) = lim log  (f () − f (−)) + lim dx = PV (f )
→0+ →0+ |x|≥ x xn

Or  
1
D log |x| = PV
xn
2) Taken f ∈ K
ˆ R n−1
!
Dk (xm f )(0) k
    
1 1 1 X
xm PV (f ) = PV (x m
f ) = lim m
x f (x) − x dx
xn xn R→∞ −R xn k!
k=0

Taken m ≥ n we have that the sum is null ∀k ≤ m, therefore since k ≤ n − 1 < m it sums to 0
Therefore   ˆ
n m 1 m
lim IR (x f ) = PV (x f ) = xm−n f (x) dx = ϕxm−n (f )
R→∞ xn R
Taken m < n we have that
n−1 n−1 n−m−1
X xk k m X  k  m! X f (k) (0)
D (x f )(0) = f (k−m) (0) = xk
k! m k! k!
k=0 k=m k=0

Therefore
ˆ R n−m−1
!
f k (0) k
 
n m 1 X 1
lim IR (x f ) = lim lim f (x) − x dx = PV (f )
R→∞ R→∞ →0 −R xn−m k! xn−m
k=0

3) Taken n ∈ N, n > 0, m = 1 we have

   
1 1
D PV n
(f ) = −PV (f 0 ) = − lim IR
n
(f 0 )
x xn R→∞

Integrating by parts we have

" n−1
!#R
1 X f (k+1) (0)
n 0 k+1
IR (f ) = f (x) − f (0) − x +
xn (k + 1)!
k=0 −R
ˆ R n−1 (k+1)
!
n X f (0) k
+ n+1
f (x) − f (0) − x dx =
−R x (k + 1)!
k=0
n
! n
!
1 X f (k) (0) k 1 X f (k) (0) k n+1
= n f (R) − R − f (−R) − (−R) + nIR (f )
R k! (−R)n k!
k=0 k=0

Since supp f ⊂ [−b, b] we have two cases, R > b and R < b.

Supposing R > b we have
n
f (k) (0)
 
X 1 1
n
IR (f 0 ) = nIR
n+1
(f ) − −
k! (−R)n−k Rn−k
k=0
CHAPTER 8. DISTRIBUTIONS 154

Taken a new index j = n − k such that j mod 2 = 0 we have

n
f (j) (0)
 
X 1 1
n
IR (f 0 ) = n+1
nIR −2 −→ −nPV (f )
j=0
j! Rn−k xn+1

In case that m > 1, by iteration we get

   
m 1 1
D PV = (−m)!PV
xn xn+m

§§ 8.2.4 A Physicist’s Trick, The Dirac δ “Function”

Definition 8.2.18 (Composite Delta). Supposing that a Dirac δ exists also as a function, we can imagine
describing it as a composite function (δ ◦ f )(x) = δ(f (x)).
Watch out, since δ(f (x)) 6= δ0 (f )

Definition 8.2.19 (Giving it Some Meaning). Taken gn ∈ L1loc (R) a sequence of functions such that
gn ∈ C[−an , an ] with an > 0, an → 0, we have that ∃!gn 7→ ϕgn such that ϕgn *? δ0
A good choice for this sequence would be the Gaussian function γn defined as follows
n 2 2
γn (x) = √ e−n x (8.17)
π

Then, we have that ϕγn *? δ0 , or written in a clear abuse of notation with non-existing functions

lim γn (x) = δ(x)

n→∞

Which, actually means that

ˆ
lim γn (x)f (x) dx = f (0) = δ0 (f ) ∀f ∈ K
n→∞ R

Then, letting b(x) ∈ C 1 (R) we can define the following quantity

ˆ ˆ
f (x)δ(b(x)) dx = lim γn (b(x))f (x) dx ∀f ∈ K
R n→∞ R

Theorem 8.27. Let g ∈ C 1 (R) be a function with isolated and simple zeros Zg := {x1 , x2 , x3 , · · · },
then ∀f ∈ K we have
ˆ ˆ X δx (f )
k
lim γn (g(x))f (x) dx = f (x)δ(g(x)) dx = 0 (x )|
(8.18)
n→∞ R R |g k
xk ∈Zg
8.2. DISTRIBUTIONS 155

Proof. Taken g(x) ∈ C 1 (R) such that g(x) ≥ g(y) ∀x ≥ y, supposing that Zg = {x0 } we have, since
it’s a simple zero, that g 0 (x0 ) 6= 0
Taking in the previous integral the substitution y = g(x), we have that x = g −1 (y) and dx =
D g −1 (y) dy, therefore
ˆ ˆ sup(g)
f (g −1 (y))
I = lim γn (g(x))f (x) dx = lim γn (y) dy
n→∞ R n→∞ inf(g) g 0 (g(y))

Evaluating the integral we have

f (g −1 (0)) f (x0 ) 1
I= = 0 = 0 δx
g 0 (g −1 (0)) g (x0 ) |g (x0 )| 0

This is easily generalizable if there is more than one zero

Example 8.2.9 (Composition with a Constant). Take δ(ax) with g(x) = ax and a 6= 0. Since Zg = {0}
using the previous formula we have g 0 (0) = a and therefore
1
δ(ax) = δ(x)
|a|

Example 8.2.10. Taken g(x) = atan(x) we have that Zg := {0} and

1
g 0 (x) = g 0 (0) = 1
1 + x2
Therefore
δ(atan(x)) = δ(x)

Example 8.2.11 (Composition with a Polynomial). Take g(x) = (x2 − a2 ), then Zg := {−a, a} and
g 0 (±a) = ±2a, therefore, if a 6= 0
1
δ(x2 − a2 ) = (δ(x − a) + δ(x + a))
2|a|

Example 8.2.12 (Composition with an Exponential). Taken g(x) = ex we have that Zg = {} therefore
δ(ex ) = 0
Example 8.2.13 (Composition with a Trigonometric Function). Taken now g(x) = cos(x) we have
that Zg = { k ∈ Z| kπ/2} and g 0 (x) = − sin(x), therefore
 
0 kπ
g (xk ) = − sin = (−1)k+1
2

Therefore
X   ∞  
kπ X kπ
δ(cos(x)) = δ x− = δ x−
2 2
k∈Z k=−∞
CHAPTER 8. DISTRIBUTIONS 156

Example 8.2.14 (Calculating an Integral with the Composite Delta). Take the following integral
ˆ
e−|x| δ(sin(2x)) dx
R

In order for solving it we expand the composite delta. We have that Zg = { k ∈ Z| kπ/2} and
g 0 (xk ) = (−1)k 2, therefore
ˆ ∞   ∞
−|x| 1 X −|x| kπ 1 X − kπ
e δ(sin(2x)) dx = e δ x− = e 2

R 2 2 2
k=−∞ k=−∞

Fixing the sum and summing, we have

ˆ X kπ 1 1 1
e−|x| δ(sin(2x)) dx = e− 2 − = π −
R k∈N
2 1 − e 2 2

§ 8.3 Integral Representation of Distributions

§§ 8.3.1 Dirac Delta and Heaviside Theta
Theorem 8.28 (Integral Representation of the Delta Distribution). Taken the non-existent δ(x)
function, we can represent it as follows
¨
1
δ(x − y) = eik(x−y) f (x) dk dx (8.19)
2π R2

Which means ¨
1
δy (f ) = eik(x−y) f (x) dk dx
2π R2

Theorem 8.29 (Integral Representation of the Theta Distribution). Taken ϑ(x) the Heaviside regular
distribution we can write ∀x ∈ R \ {0}
ˆ
1 eikx
ϑ(x) = PV dk (8.20)
2πi R k − i

Proof. This equation has two cases, one with x > 0 and one with x < 0. Therefore taking x > 0 we
have ˆ
eikx
I(x) = PV dk
R k − i

This integral can be evaluated using the residue theorem. We have defining the function f (k) as follows

1
f (k) =
k − i
8.3. INTEGRAL REPRESENTATION OF DISTRIBUTIONS 157

The integral can be seen as follows

ˆ
I(x) = PV eikx f (k) dk
R

This can be written as a complex integral, writing the transformation x → z with z ∈ C

˛
eizx
I(x) → J(x) = lim dz
R→∞ γR z − i

Where the path is the following

+
{γR } = CR ∪ [−R, R]
Writing it all out explicitly we have that
ˆ ˆ R
eizx eizx
J(x) = lim dz + lim dz = JR + I(Rez)
R→∞ CR z − i R→∞ −R z − i

Due to the Jordan lemma we know for sure that JR = 0 therefore we have that, writing Re(z) = x,
thanks to the residue theorem
X eizx
J(x) = I(x) = 2πi Res
z=zk z − i
zk ∈Sf

The set of singularities of f is formed by a single point Sf = {i} which is a simple pole, and therefore,
since i ∈ {γR }◦ we have

eizx eizx
J(x) = 2πi Res = 2πi lim (z − i) = e−x
z=i z − i z→i z − i

Therefore, we get
J(x) = 2πie−x
For getting back to the first integral, we have

1
I(x) = lim+ e−x = 1 = ϑ(x) x > 0
2πi →0

In the second case, for x < 0 we have that the new curve will be
− −
{γR } = CR ∪ [−R, R]
− ◦
Since Sf 6⊂ {γR } we have thanks to Cauchy-Goursat that

1 1 −
I(x) = J (z) = 0 = ϑ(x) x < 0
2πi 2πi
Proving our assumption
CHAPTER 8. DISTRIBUTIONS 158

§§ 8.3.2 Distributions in Rn
Definition 8.3.1 (Scalar Test Fields). Given a scalar field f : Rn −→ C one can define the space of
scalar test fields K(Rn ) as follows
K(Rn ) = Cc∞ (Rn ) (8.21)
A function f ∈ Cc∞ (Rn ) is a function such that supp f ⊂ A ⊂ Rn with A a compact set and
∃∂ α f ∀|α| ∈ N
Definition 8.3.2 (K(Rn )−Convergence). Given fn ∈ K(Rn ) a sequence of scalar fields. fn converges
to f ∈ K(Rn ) if
1. ∃A ⊂ Rn compact set such that fn (Ac ) = {0}
2. ∀α ∈ Nn multi-index ∂ α fn ⇒ ∂ α f , ∀x ∈ A
It’s indicated as
fn *Kn f

Definition 8.3.3 (Multidimensional Distribution). A n−dimensional distribution is a continuous

functional ϕ : K(Rn ) −→ C such that

∀gn *Kn g ϕ(gn ) → ϕ(g) (8.22)

Then ϕ ∈ K? (Rn )
Theorem 8.30 (Operations in K? (Rn )). We can define the usual distributional operations in
K(Rn ). The usual operations defined for K? are defined in the same way, the only difference
is given by the definition of the derivative Given α ∈ Nn , f ∈ K(Rn ), ϕ ∈ K? (Rn )

∂ α ϕ(f ) = (−1)|α| ϕ(∂ α f ) (8.23)

Example 8.3.1 (Laplacian of a Distribution). Taken ∂ α = ∂ 2 = ∂µ ∂ µ

∂µ ∂ µ ϕ(f ) = ϕ(∂µ ∂ µ f )

Remark (Local Integrability). The local integrability of some functions is different in Rn . In fact taken
the spherical (n − 1)-dimensional coordinate transformation we have

dn x = rn−1 dr dSn−1
−a
And taken the function g(xν ) = kxµ kµ we get
ˆ ˆ ˆ
f (x) n rn−1
a d x= f (x) dSn−1 dr
Rn kxµ kµ R Sn−1 ra
−a
Which means that kxµ kµ ∈ L1loc (Rn ) ∀a < n
8.4. SOME APPLICATIONS IN PHYSICS 159

§ 8.4 Some Applications in Physics

§§ 8.4.1 Electrostatics
Example 8.4.1 (Point Charge). Taken q a point charge in the origin (0, 0, 0) we define the potential
generated by this charge as follows (in cgs)

∂µ ∂ µ V (xν ) = −4πρ(x)

Where ρ(x) is the charge density, defined as

ρ(x) = qδ 3 (xν )

With δ 3 being the 3−d Dirac delta

Therefore !
1
∂µ ∂ µ = −4πδ 3 (xν )
kxµ kµ

Which means actually, that ∀f ∈ K(R3 )

˚
∂µ ∂ µ f 3
d x = −4πf (0)
R3 kxµ kµ

Proof. Since f ∈ K(R3 ) we have that ∃R > 0 : f (xµ ) = 0∀x ∈

/ B R (0)
2
Therefore we get that ˚ ˆ
∂µ ∂ µ f 3 ∂µ ∂ µ f 3
µ
d x = lim d x
R3 kx kµ →0 ≤kxµ k ≤R kxµ k
µ µ
n o
Taken AR (0) = xµ ∈ R3  ≤ kxµ kµ ≤ R We have that the second integral becomes, using Stokes’
theorem
˚ ! ¨ " !#
1 3 1 1
f (x)∂µ ∂ µ f (x) d x + ∂µ f − f ∂µ nµ dσ
AR kxµ kµ ∂AR kxµ kµ kxµ kµ

Changing to spherical coordinates we have that

!   
µ 1 1 ∂ 2 ∂ 1
∂µ ∂ = r =0
kxµ kµ r2 ∂r ∂r r

Therefore we get
˚ ¨   
1 3 1 1
∂µ ∂ µ f d x = − ∂µ f − f ∂µ rµ dS2
AR kxµ kµ kxµ kµ = r r

Which, taken ∂µ r−1 = −r̂µ r−2 we have

˚ ¨ ¨
1 µ 3 µ
µ
∂µ ∂ f d x = − r̂ ∂µ f r dS2 − f (x) dS2
AR kx kµ kxµ kµ = kxµ kµ =
CHAPTER 8. DISTRIBUTIONS 160

We have for the first integral that

¨
r̂µ ∂µ f r dS2 ≤ 4π sup k∂µ f k
kxµ kµ = R3

And for the second

¨
f (x) dS2 − 4πf (0) ≤ 4π sup |f (xν ) − f (0)|
kxµ kµ = kxµ kµ =

Since  → 0+ and since f ∈ K(R3 ) we have that the first and second integral converge to 0, and
therefore ˚ ¨
1 µ 3
µ
∂µ ∂ f d x = − lim f (x) dS2 = −4πf (0)
R3 kx kµ →0 kxµ kµ =
9 Ordinary Differential Equations

§ 9.1 Existence of Solutions

1
You might test that assumption at your convenience

Definition 9.1.1 (Differential Equation). Given y ∈ C m (I), I ⊆ R we define a differential equation

of order m the following
F (x, y, y 0 , · · · , y (m) ) = 0
If the equation can be rewritten as follows

dm y
= f (x, y, · · · , y (m−1) )
dxm
It’s called in normal form.
If only total derivatives appear, the differential equation is called ordinary, whereas if also partial
derivatives appear, the differential equation is called partial.

Theorem 9.1 (Reduction of Order). Given an ODE (ordinary differential equation) of order m,
one can reduce the order of the equation through a mapping to Rm , where
 
y(x)
 y 0 (x) 
(y, y 0 , · · · , y (m) ) 7→ y µ (x) =  ..
 
.

 
y (m) (x)

And f (x, y, · · · , y (m) ) 7→ f µ (x, y µ ).

The equation becomes a first order differential equation

dy µ
= f µ (x, y µ )
dx
1 Captain Picard, Star Trek: The Next Generation

161
CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 162

If f is linear, this system can be expressed in matrix form, where

dy µ
= Aµν y µ
dx

Definition 9.1.2 (Cauchy Problem). A Cauchy or initial value problem is defined as the following
system (
y 0 (x) = f (x, y(x))
(9.1)
y(x0 ) = y0
Where y0 , t0 ∈ R are known values
Theorem 9.2. Given the Cauchy problem (9.1) we say that y(x) ∈ C 1 is a solution of the system
if and only if is a continuous solution of the following integral equation
ˆ x
y(x) = y0 + f (s, y(s)) ds (9.2)
x0

Proof. Supposing y ∈ C 1 we have that for the fundamental theorem of integral calculus
ˆ x ˆ x
y(x) = y(x0 ) + y 0 (s) ds = y0 + f (s, y(s)) ds
x0 x0

Instead, considering f ◦ y we have that the new composed function must be continuous, therefore we
have for the fundamental theorem of integral calculus that y(x) ∈ C 1 and that
ˆ x0
y(x0 ) = y0 + f (s, y(s)) ds = y0
x0

And therefore ˆ x
dy d
= f (s, y(s)) ds = f (x, y(x))
dx dx x0

Which gives back the thesis

Theorem 9.3 (Picard-Lindelöf-Cauchy Existence). Let A ⊂ R2 with f : A −→ R a continuous
function which defines an ODE (9.1) where
ˆ x
y(x) = y0 + f (s, y(s)) ds
x0

Let a, b > 0 and take R = [x0 − a, x0 + a] × [y0 − b, y0 + b] = Ia × Jb ⊂ A. Supposing f Lipschitz

continuous on the second variable, i.e.
∃L > 0 : |f (x, u) − f (x, w)| ≤ L|u − w| ∀(x, u), (x, w) ∈ R
Then, if M = maxR |f (x, y)|
∃ > 0 : ∃!y(x) ∈ C 1 (B (y0 ))
Where  = min {a, b/M, 1/L}
Therefore, there is an unique local solution y(x) to the ODE
9.1. EXISTENCE OF SOLUTIONS 163

Proof. We firstly define a Banach space (X, k·ku ) where X := { y ∈ C (I )| ky(x) − y0 ku ≤ b}.
Define the Picard operator as follows
T̂ : X −→ X
Where ˆ x
T̂ (w) = y0 + f (s, w(s)) ds = z(x)
x0

We have
ˆ x
T̂ (y) − T̂ (z) = T̂ (y − z) = f (s, u(s)) − f (s, v(s)) ds ≤ Lku − vku ∀u, v, y, z ∈ X
u u x0 u

Since x ∈ I is completely arbitrary we have therefore, putting d(x, y) = kx − yku

d(y, z) = d(T̂ (y), T̂ (z)) ≤ Ld(u, v)

Since L ∈ (0, 1), T̂ is a contractor and it has a single fixed point in X.

Taking this fixed point as y(x) we have
ˆ x
T̂ (y) = y0 + f (s, y(s)) ds = y(x)
x0

Which is our searched solution, and it’s unique

Definition 9.1.3 (Maximal and Global Solutions). Given y(x) a solution to the Cauchy problem (9.1)
we define a maximal solution ym (x) a solution for which ym (x) = y(x) ∀x ∈ I and which still
solves the problem in a set (a, b) ⊃ I , and (a, b) is the biggest set for which ym (x) solves the ODE.
If ym is defined ∀x ∈ R, the solution is called global

Theorem 9.4 (Prolungability of Solutions). Given the ODE (9.1) and let f : (a, b) × R −→ R. Let
c1 , c2 ∈ K ⊂ (a, b) with K a compact set, such that

|f (x, y)| ≤ c1 + c2 |y(x)| x ∈ K, ∀y ∈ R

Then the solution can be extended in the whole set (a, b). The previous statement means
that f is sublinear in the second variable

Theorem 9.5. Let ym (x) be a maximal solution to (9.1) defined on (a, b). Then ∀K ⊂ A ∃δ >
0 : ∀x 6∈ (a + δ, b − δ), (x, ym (x)) 6∈ K

Theorem 9.6. Taken y(x) a solution to the ODE (9.1). If ∃c > 0 such that

|y(x)| ≤ c ∀t ∈ A

Then ∃ym (x) : (a, b) −→ R

Lemma 9.1.1 (Peano-Gronwall Inequality). Let ϕ : I ⊂ R −→ R be a function ϕ ∈ C(I), if

ˆ t
|ϕ(t)| ≤ c + L |ϕ(s)| ds
t0
CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 164

Then the following inequality holds

|ϕ(t)| ≤ ceL|t−t0 |
Therefore, if y 0 (t) = f (t, y) is an ODE, taken f (t, y) Lipschitz continuous on the second variable and
chosen ϕ(t) = u(t) − v(t) where u, v are two solutions, we have that
|f (t, u) − f (t, v)| ≤ ceL|t−t0 |

§ 9.2 Common Solving Methods

Definition 9.2.1 (Separable ODE). Given y 0 (t) = f (t, y(t)) an ODE we define a separable ODE a
differential equation such that
y 0 (t) = f (t, y(t)) = g(t)h(y)

Method 1 (Separation of Variables). Suppose that we have the previous ODE, then, thanks to the
Picard-Lindelöf-Cauchy theorem we can immediately say that
1. There is at least one local solution in B (t0 ) if f ∈ C, i.e. if g, h ∈ C
2. There is a unique local solution in B (t0 ) if ∂y f ∈ C, i.e. if h ∈ C 1
The first thing to do is finding all constant solution to the ODE, called the equilibrium solutions.
Then, a general solution can be found by direct integration, where we’re integrating the following
differential form
1
dy = g(t) dt
h(y)
Integrating, and putting H(t), G(t) as the two primitives, we can say that
H(y) = G(t) + H(y0 ) − G(t0 )
Where c = H(y0 ) − G(t0 ) is the integration constant in terms of the initial values, if given
Example 9.2.1. Take the following initial value problem
(
w0 (x) = w2 (x)
w(0) = 1

We have f (x, w) = w2 (x) i.e. the equation is separable, therefore

ˆ ˆ
w0 (x)
dx = dx − 1
w2 (x)
Therefore, integrating the LHS
1
− =x−1
w(x)
And
1
w(x) =
1−x
9.2. COMMON SOLVING METHODS 165

Definition 9.2.2 (Autonomous Differential Equation). Given a differential equation in normal form
y 0 (x) = f (x, y(x)), it’s said to be an autonomous differential equation if

f (x, y(x)) = g(y(x))

Therefore
y 0 (x) = g(y(x)) (9.3)
Note that it’s a separable equation, and therefore, writing the primitive of 1/g(y) = G(y) we have that
the general solution of this problem would be

G(y) = x + c

Where c ∈ R is a constant

Example 9.2.2. Take the following initial value problem

(
y 0 (x) = t|y|
y(0) = y0

This is obviously a separable ODE, and we have f (t, y(t)) = t|y|.

We have that f ∈ C(R2 ) therefore there exists at least one solution for this equation.
Now, applying Picard-Lindelöf-Cauchy we can see that, ∀[α, β] ⊂ R compact set

|f (t, y) − f (t, z)| = |t|y| − t|z|| = |t|||y| − |z|| ≤ max {|α|, |β|} |y − z|

Taken L = max {|α|, |β|} we have that f is Lipschitz continuous in the second variable in every compact
set K ⊂ R and therefore there exists a unique solution y : K −→ R.
Successively we move on to integrate directly the differential equation. We have
ˆ y ˆ t
dy
= t dt
y0 |y| t0

I.e.
y 1 2
t − t20


log =
y0 2
y 1 2 1 2
= e 2 t0 e 2 t
y0
Since t0 = 0 we get
1 2
|y|(t) = |y0 |e 2 t
The solutions depend on the sign of y0 and we finally get
( 1 2
y0 e 2 t y0 > 0
y(t) =
− 12 t2
y0 e y0 < 0
CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 166

§§ 9.2.1 First Order Linear ODEs

Definition 9.2.3 (Integrating Factor). An example of linear ODE is the following kind of equation

y 0 (t) = p(t)y(t) + q(t) p, q ∈ C(I), I ⊆ R (9.4)

This equation has a unique solution in all of R, since f (t, y) is Lipschitz continuous and sublinear in R.
We immediately know that the solution y ∈ C 1 exists since f ∈ C(R2 ) and by Cauchy-Picard-Lindelöf
we have
|f (t, y) − f (t, z)| = |p(t)y(t) − p(t)z(t)| = |p(t)||y − z| ≤ kpku |y − z|

Taken L = kpku we see immediately that f satisfies the theorem and the solution is uniquely defined
in every compact set K ⊂ R.
We also see that f is sublinear, since

|f (t, y)| = |p(t)y(t) + q(t)| ≤ kpku |y(t)| + kqku

Which means that the solution can be extended in R, and ∃!y : R −→ R solution to the ODE.
We now proceed to integrate the ODE, for which we will proceed with a different method than usual.
Let µ(t) be what we will call the integrating factor, defined as follows
´
µ(t) = e− p(t) dt

Which has the property that p(t)µ(t) = µ0 (t)

Method 2 (Integrating Factor). Let y 0 (t) = p(t)y(t) + q(t) be a linear ODE, we solve by finding the
integrating factor µ(t) and then multiplying both sides of the equayion by the integrating factor and
moving the p(t)y(t) term from the RHS to the LHS

µ(t)y 0 (t) − p(t)µ(t)y(t) = q(t)µ(t)

By definition of the integrating factor we have that p(t)µ(t) = −µ0 (t) and therefore

µ(t)y 0 (t) + µ0 (t)y(t) = q(t)µ(t)

Using the chain rule we get

d
(µ(t)y(t)) = µ(t)q(t)
dt
Integrating
ˆ
µ(t)y(t) = µ(t)q(t) dt + c

And therefore, the final general solution will be the following

ˆ
1 c
y(t) = µ(t)q(t) dt +
µ(t) µ(t)
9.2. COMMON SOLVING METHODS 167

Example 9.2.3. Take the following initial value problem

 y 0 (t) = 1 y(t) + 3t3



t
y(−1) = 2


We immediately see the previous definition, therefore we can immediately say that since f is Lipschitz
continuous and sublinear the solution to the problem will be uniquely defined on all of R.
For integrating this ODE we use the integrating factor method, taking p(t) = t−1 and q(t) = 3t3 .
By definition, we have µ(t) = |t|, and therefore, since t < 0, µ(t) = −t−1 , which gives
ˆ
1
− y(t) = −3 t2 dt + c
t
Therefore
y(t) = t4 − ct
Applying the initial condition
y(−1) = 1 + c = 2 =⇒ c = 1
And therefore the final, unique solution of the initial value problem is the following

y(t) = t4 − t

Theorem 9.7 (General Solutions). Taken y 0 (x) + p(x)y(x) = q(x) a linear non homogeneous ODE,
we define the associated homogeneous ODE as the equation

z 0 (x) + p(x)z(x) = 0

The solution of the complete ODE will be given by the sum of the particular solution y(x)
and the homogeneous solution z(x)
Proof. Take y(x) a solution of the equation and y(x) the particular solution of the ODE, we then have,
since both are solutions (
y 0 (x) + p(x)y(x) = q(x)
y 0 (x) + p(x)y(x) = q(x)
Subtracting term a term we have

(y 0 (x) − y(x)) + p(x)(y(x) − y(x)) = 0

Chosen y 0 (x) − y(x) = z(x) we have that

z 0 (x) + p(x)z(x) = 0

Therefore z(x) is a solution, but it’s also the solution to the associated homogeneous ODE, and therefore,
inverting for the general solution y(x), we have

y(x) = y(x) + z(x)

CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 168

Example 9.2.4. Taken the following initial value problem

y 0 (x) + 2 y(x) = 1


x x2
y(−1) = 2


We want to find the general integral of the equation, y(x). We begin by solving the associated
homogeneous equation
2
z 0 (x) + z(x) = 0
x
In order to integrate this we use the integrating factor µ(x), where
 ˆ x 
1
dx = exp (2 log(|x|)) = exp log x2 = x2



µ(x) = exp 2
−1 x

We multiply by the integrating factor, and get

d 2

x z(x) = 0
dx
Therefore
c
x2 z(x) = c =⇒ z(x) =
x2
The particular solution y(x) will be
ˆ x
1 1
y(x) = µ(x) dx
µ(x) x0 x2

Therefore ˆ x
1 1
y(x) = dx =
x2 −1 x
The general solution will be y(x) = y(x) + z(x), and we finally get

c 1
y(x) = 2
+
x x
Imposing the initial condition we have

y(−1) = c − 1 = 2 =⇒ c = 3

Therefore
3 1
y(x) = +
x2 x
Note how we could have found the general integral directly using the formula
ˆ x ˆ x
c 1 3 1
y(x) = 2 + 2 dx = 2 + 2 dx
x x −1 x x −1
9.2. COMMON SOLVING METHODS 169

Method 3 (Variation of Constants). Given a linear ODE y 0 (t) + a(t)y = f (t), with a, f ∈ C(I), I ⊆ R
suppose that z(t) is the solution of the homogeneous equation, we have therefore
ˆ 
c
z(t) = = c exp a(t) dt
µ(t)

We can find a particular solution through variation of constants, i.e. we suppose c = c(t), and we
have ˆ 
y(t) = c(t) exp a(t) dt

Imposing this on the differential equation, we have

 ˆ   ˆ 
y 0 (t) = c0 (t) exp − a(t) dt − a(t)c(t) exp − a(t) dt

And therefore
1 a(t)c(t)
(c0 (t) − a(t)c(t)) + = f (t)
µ(t) µ(t)
Therefore ˆ
c0 (t)
= f (t) =⇒ c(t) = f (t)µ(t) dt
µ(t)
Which implies ˆ
1
y(t) = f (t)µ(t) dt
µ(t)
We then recover the previous formula by adding the particular and homogeneous solutions
ˆ
c 1
y(t) = + f (t)µ(t) dt
µ(t) µ(t)

§§ 9.2.2 Second Order Linear ODEs

Definition 9.2.4 (Initial Value Problem for 2nd Order Linear ODEs). Given a second order linear
differential equation y 00 (x) + a(x)y 0 (x) + b(x)y(x) = f (x) we define the initial value problem for
such differential equation as follows
 00 0
y (x) + a(x)y (x) + b(x)y(x) = f (x)

y 0 (x0 ) = y00

y(x0 ) = y0


This kind of problem, if a, b, f ∈ C(I) I ⊆ R is said to be well defined

Theorem 9.8 (Space of Solutions). Given a second order linear ODE, the solutions of the ho-
mogeneous equation span a vector space of dimension 2 S, and the general solution of the
homogeneous equation has the following form

zg (x) = c1 z1 (x) + c2 z2 (x) ∈ S

CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 170

I.e., if we define a matrix Wµν (t) as follows

 
z1 (t) z2 (t)
Wµν (t) =
z10 (t) z20 (t)

If z1 , z2 solve the differential equation, the matrix Wµν (t) is non-singular

Proof. Suppose z1 , z2 ∈ C 2 (I) are two solutions. If they’re linearly inddependent we have that they
solve the following system with c1 , c2 ∈ R \ {0}
(
c1 z1 (x) + c2 z2 (x) = 0
c1 z10 (x) + c2 z20 (x) = 0

This system is solvable only if the determinant of the matrix Wµν is non zero, and therefore the two
functions must be linearly independent and span the vector space of solutions S
Method 4 (Characteristic Polynomial). Take the following linear homogeneous ODE of order 2

z 00 (x) + az 0 (x) + bz(x) = 0 a, b ∈ R

We begin by “guessing” a solution in exponential form z(x) = eλx , where λ ∈ C. Therefore, reinserting
into the ODE, we have
eλx λ2 + aλ + b = 0

Since the exponential is never zero ∀x ∈ R, ∀λ ∈ C we have that the polynomial must be zero, and
therefore we get two roots
a 1p 2
λ1,2 = − ± a − 4b
2 2
We can now discern 3 cases.
Case 1), a2 − 4b > 0
There are two real roots λ1 , λ2 , and therefore we have

z(x) = c1 eλ1 x + c2 eλ2 x

Case 2), a2 − 4b < 0

There are two complex conjugate roots λ, λ, and the solution will be
 
z(x) = Re c1 eλx + c2 eλx = c1 eRe(λ)x cos(Im(λ)x) + c2 eRe(λ)x sin(Im(λ)x)

Or
z(x) = AeRe(λ)x cos(Im(λ) + ϕ)
With A, ϕ ∈ R Case 3) a2 − 4b = 0
There is only one root λ = a/2 with multiplicity 2. In order to find the general integral z(x) we impose
the variation of constants, and we have

z(x) = c(x)eλx

And
z 0 (x) = c0 (x)eλx + λc(x)eλx = eλx (c0 (x) + λc(x))
z 00 (x) = c00 (x)eλx + λc0 (x)eλx + λ2 c(x)eλx = eλx (c00 (x) + λc0 (x) + λ2 c(x))
9.2. COMMON SOLVING METHODS 171

Since λ is a root of the characteristic polynomial, we end up with

c00 (x) = 0 =⇒ c(x) = c2 + c1 t
Therefore
a
z(x) = eλt (c1 + c2 t) = e− 2 t (c1 + c2 t)

Method 5 (Similarity Method). Another method for finding a particular solution is the similarity method.
Given the following linear ODE
y 00 (x) + ay 0 (x) + by(x) = f (x)
The particular solution of this problem can be found using the shape of f (x).
Case 1) f (x) ∈ Rn [x]
In this case we will suppose that the solution will be a polynomial, and we can discern 3 cases

q(x)
 b 6= 0
y(x) = xq(x) b = 0, a 6= 0

 2
x q(x) b = 0, a = 0
Where q(x) ∈ Rn [x]
Case 2) f (x) = Aeλx , λ ∈ C
In this case we will suppose the particular solution as the real part of a complex exponential, where
y(x) = Re (ỹ(x)) = γ(x)Re eλx

Where γ(x) ∈ C 2 (I), I ⊂ R is some unknown function. From this we will have, deriving twice
ỹ 0 (x) = eλx (γ 0 (x) + λγ(x))
ỹ 00 (x) = eλx λ2 γ(x) + 2λγ 0 (x) + γ 00 (x)

Substituting into the differential equation we obtain

γ 00 (x) + (2λ + a)γ 0 (x) + (λ2 + bλ + a)γ(x) = A
We get again 3 different cases
a) λ2 + aλ + b = 0
In this case we have
A A
γ(x) = =⇒ ỹ(x) = 2 eλx
λ2 + aλ + b λ + aλ + b
b) λ2 + aλ + b = 0, 2λ + a 6= 0
In this gase instead we have
A A
γ 0 (x) = =⇒ ỹ(x) = xeλx
2λ + a 2λ + a
c) λ2 + aλ + b = 0, 2λ + a = 0
Lastly we have
A 2 λx
γ 00 (x) = A =⇒ ỹ(x) = x e
2
The solution for the differential equation will lastly be either the real (or imaginary, if there is a sine in
the RHS) part of the function we found.
CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 172

Example 9.2.5. We will use the similarity method to solve the following differential equation

y 00 (x) + 2y 0 (x) + 3y(x) = 2e3x

We see immediately that f (x) = 2e3x so we choose a particular solution y(x) with the following form

y(x) = ke3x

We therefore have
y 0 (x) = 3ke3x
y 00 (x) = 9ke3x

Substituting
1
18k = 2 =⇒ k =
9
Therefore our searched solution will be the following

1 3x
y(x) = e
9

Example 9.2.6. Let’s now solve the following differential equation

y 00 (x) + 2y 0 (x) − 3y(x) = 2e−3x

Following the same approach, we see that the characteristic polynomial is null, and we end up with a
contradiction, therefore we find a solution of the following kind

y(x) = kxe−3x

Deriving the particular solution, we have

y 0 (x) = ke−3x (1 − 3x)

y 00 (x) = 3ke−3x (3x − 2)

Substituting
3k(3x − 2) + 2k(1 − 3x) − 3kx = 2
−4k = 2
1
k=−
2
Therefore the particular solution we’re searching for is, finally
x
y(x) = − e−3x
2
9.2. COMMON SOLVING METHODS 173

Method 6 (Variation of Constants). Another method that might be used to solve a second order linear
differential equation is the method of variation of constants.
Given the following differential equation and the associated homogeneous equation

y 00 (x) + ay 0 (x) + by(x) = f (x)

z 00 (x) + az 0 (x) + bz(x) = 0

Suppose that z1 (x), z2 (x) are two linearly independent homogeneous solutions, and

z(x) = c1 z1 (x) + c2 z2 (x)

We search for a particular solution supposing that the two constants are actually functions, and we
have
y(x) = c1 (x)z1 (x) + c2 (x)z2 (x)
Due to the necessity of a non-singular Wronskian determinant, we impose the following condition

c01 (x)z1 (x) + c02 (x)z2 (x) = 0

Deriving two times the particular solution and then substituting in the original equation we end up
with the following system (
c01 (x)z1 (x) + c02 (x)z2 (x) = 0
c01 (x)z10 (x) + c02 (x)z20 (x) = f (x)
The necessity for the linear independence of the solutions is clear here, since the system is solvable if
and only if the Wronskian determinant is zero.
We solve the system by substitution, and we have
z (x)

c01 (x) = −c02 (x) 2

 z1 (x)
c0 (x) =
 f (x) − c01 (x)z10 (x)
2
z20 (x)


Therefore
z1 (x)

0
c1 (x) = − f (x)

z1 (x)z20 (x) − z2 (x)z10 (x)


z2 (x)
c02 (x) =

 f (x)
z1 (x)z20 (x) − z2 (x)z10 (x)
Written in terms of the Wronskian determinant detµν Wµν (x) we finally have as particular solution
ˆ ˆ
z2 (x) z1 (x)
y(x) = z2 (x) f (x) dx − z1 (x) f (x) dx + c
detµν Wµν (x) detµν Wµν (x)

And the general solution of the differential equation will be the following
ˆ x ˆ x
z2 (x) z1 (x)
y(x) = z2 (x) f (x) dx − z1 (x) f (x) dx + c1 z1 (x) + c2 z2 (x)
x0 detµν Wµν (x) x0 detµν Wµν (x)
CHAPTER 9. ORDINARY DIFFERENTIAL EQUATIONS 174
10 Fourier Calculus

§ 10.1 Bessel Inequality and Fourier Coefficients

Definition 10.1.1 (Fourier Coefficients). Suppose (uk )k∈N = U ⊂ V, with (V, h·, ·i) an euclidean space,
taken v ∈ V we can define an operator F̂U : V −→ CN such that
∀v ∈ V F̂U (v) = c ∈ CN
Where
c = (hv, u1 i, hv, u2 i, · · · ) ∈ CN
The coefficients hv, uk i ∈ C are called the Fourier coefficients of the vector v ∈ V
Theorem 10.1 (Bessel Inequality & Parseval’s Theorem). Given (uk )k∈N = U ⊂ V an orthonormal
system and V an euclidean space with v ∈ V. Taken α1 , · · · , αk ∈ C some coefficients and
defined the two following sums
n
X n
X
Sn = hv, uk iuk = ck uk
k=1 k=1
Xn
Snα = αk uk
k=1

Then
kv − Sn k ≤ kv − Snα k
∞
X 2 2
kck k ≤ kvk
k=1
The last inequality is known as Bessel’s inequality
Lastly we also have Parseval’s equality or Parseval’s theorem, which states
∞
X ∞
X ∞
X 2 2
v= hv, uk iuk = ck uk ⇐⇒ kck k = kvk
k=1 k=1 k=1

Due to this the operator F̂U actually acts into ` (C), i.e. 2

F̂U : V −→ `2 (C)

175
CHAPTER 10. FOURIER CALCULUS 176

Proof. By definition of euclidean norm and using the bilinearity of the scalar product we have
2 2
0 ≤ kv − Snα k = kvk − 2Re (hv, Snα i) + kSnα k =
n
! n
2
X X 2
= kvk − 2Re hv, uk iαk + kαk k
k=1 k=1

Therefore
n
X n
X
2 2 2
0 ≤ kvk − kck k + kαk − ck k
k=1 k=1

The minimum on the left is given for αk = ck and therefore, since Snc = Sn we have

kv − Sn k ≤ kv − Snα k

And, using the non-negativity of the norm operator, putting n → ∞ we have

n
X n
X
2 2 2 2
0 ≤ kv − Sn k = kvk − kck k =⇒ kck k ≤ kvk
k=1 k=1

Therefore
∞
X 2 2
khv, uk ik ≤ kvk
k=1

Which means that the sum on the left converges uniformly, and therefore ck ∈ `2 (C). This demonstrates
that F̂U : V −→ `2 (C) and Bessel’s inequality.
This also gives Parseval’s equality, since, for n → ∞

∞ 2 ∞
X 2
X 2
v− hv, uk iuk = 0 ⇐⇒ kvk = khv, uk ik
k=1 k=1

Due to the uniform convergence in V we have therefore

∞
X
F̂U (v) = hv, uk iuk
k=1

Definition 10.1.2 (Closed System). An system (uk )k∈N ∈ V is said to be closed iff ∀v ∈ V
∞
X
2 2
kvk = khv, uk ik
k=1
X∞
v= hv, uk iuk
k=1
10.1. BESSEL INEQUALITY AND FOURIER COEFFICIENTS 177

Theorem 10.2 (Closeness and Completeness). Given an orthonormal system U = (uk )k∈N ∈ V
with V an euclidean space, we have that U is a complete set if and only if U is a closed system.
If U is complete or closed, V is separable
Proof. Defined Sn the partial sums of the Fourier representation of v (ndr the series that represents v
with respect to the system (uk )), we have that for the theorem to be true the following two things
must hold
lim Sn = v span(uk ) = V
n→∞

I.e. ∀ > 0 ∃N ∈ N, α1 , · · · , αN ∈ C such that kv − Snα k < . Using Bessel-Parseval we have

α
0 ≤ kv − SN k ≤ kv − SN k<

Proving the closure of the system if the space V is complete.

Taken (uk )k∈N a closed system, we have that Sn → v, therefore v ∈ ad (span(U)), which implies

v ∈ span(U) =⇒ V = span(U)

The last implication is given by the fact that v ∈ V is arbitrary, and it implies the completeness of U and
the separability of V
Theorem 10.3 (Riesz-Fisher). Given V a hilbert space and U = (uk )k∈N ∈ V an orthonormal
system, therefore ∀c ∈ `2 ∃v ∈ V : F̂U [v] = c and

ck = hv, uk i
∞
X
2 2 2
kvk = kck2 = kck k
k=1
∞
X
v= hv, uk iuk
k=1

Proof. Taken a sequence (vk ) ∈ V defined as follows

m
X
vn = ck uk
k=1

This sequence is a Cauchy sequence, therefore it converges to v ∈ V, since

m 2 m m
2
X X X
kvn − vm k = ck u k =h ck uk , ck uk i =
k=n+1 k=n+1 k=n+1
Xm m
X m
X 2
= ck ci hui , uk i = kck k
k=n+1 i=n+1 k=n+1

By definition, since c ∈ `2 , the sum on the right converges, therefore

∞
X
2 2
kvn − vm k ≤ kck k < ∞
k=n+1
CHAPTER 10. FOURIER CALCULUS 178

Which means, that ∀ > 0 ∃N ∈ N such that

∞
X
2 2
kvn − vm k ≤ kck k <  ∀n ≥ N
k=n+1

Which implies that vn → v and

∞
X
v= ck uk
k=1
We can now write hv, uk i = hvn , uk i + hv − vn , uk i.
We have
X n
∀n ≥ k hvn , uk i = ci hui , uk i = ck
i=1
For Cauchy-Schwartz we also have that
khv − vn , uk ik ≤ kv − vn k → 0
Which implies that ck = hv, uk i and therefore
∞
X ∞
X
v= hv, uk iuk = ck uk
k=1 k=1

§ 10.2 Fourier Series

§§ 10.2.1 Fourier Series in L2 [−π, π]
Definition 10.2.1 (Fourier Series). Given a function f ∈ L2 [−π, π] we define the Fourier series
expansion of this function the following expression
∞
a0 X
f (x) ∼ + ak cos(kx) + bk sin(kx) (10.1)
2
k=1

Where ˆ
1 π


 ak = f (x) cos(kx) dx
 π −π
ˆ (10.2)
 1 π
 bk =
 f (x) sin(kx) dx
π −π
The notation ∼ indicates that the Fourier series of the function converges to the function f (x). Usually
an abuse of notation is used, where the function is actually set as equal to the Fourier expansion.
Definition 10.2.2 (Trigonometric Polynomial). A function p ∈ L2 [−π, π] is said to be a trigonometric
polynomial if, for some coefficients αk , βk we have
∞
α0 X
p(x) = + αk cos(kx) + βk sin(kx) (10.3)
2
k=1
10.2. FOURIER SERIES 179

Theorem 10.4 (Completeness of Trigonometric Functions). Given (uk ), (vk ) ∈ L2 [−π, π] two se-
quences of functions, where (
uk (x) = cos(kx)
vk (x) = sin(kx)

The set {uk , vk } is orthogonal and complete, i.e. a basis in L2 [−π, π]

Remark. These trigonometric identities always hold, ∀n, k ∈ N, n 6= k

1
cos(nx) cos(kx) = (cos[(n + k)x] + cos[(n − k)x])
2
1
sin(nx) sin(kx) = (cos[(n − k)x] − cos[(n + k)x]) (10.4)
2
1
cos(nx) sin(kx) = (sin[(n + k)x] − sin[(n − k)x])
2

Proof. We begin by demonstrating that the two function sequences uk , vk are orthogonal in L2 [−π, π].
Therefore, by explicitly writing the scalar product, we have, for k 6= n
ˆ π ˆ
1 π
hun , uk i = cos(nx) sin(kx) dx = cos[(n + k)x] + cos[(n − k)x] dx
−π 2 −π

Therefore  π
1 sin[(n + k)x] sin[(n − k)x]
hun , uk i = + =0
2 n+k n−k −π

Analogously  π
1 sin[(n − k)x] sin[(n + k)x]
hvn , vk i = − =0
2 n−k n+k −π

And, finally  π
1 cos[(n + k)x] cos[(n − k)x] 1 1
hun , vk i = − = − =0
2 n+k n−k −π 2 2
Which demonstrates that, for k 6= n uk ⊥ un , vk ⊥ vn , uk ⊥ vk .
Now, taken a trigonometric polynomial p(x) ∈ L2 [−π, π] we need to prove that span {uk , vk } =
L2 [−π, π], i.e.
∀ > 0 ∀f ∈ L2 [−π, π] kp − f k2 < 

We have already that C[−π, π] = L2 [−π, π] and that for a Weierstrass theorem (without proof), every
periodic function with period 2π is the uniform limit of a trigonometric polynomial.
Using these two results, given f ∈ L2 [−π, π], ∃g ∈ C[−π, π] : kf − gk2 < /3. Taken ĝ(x) as the
periodic extension of g(x), for Weierstrass we have
  
kg − ĝk2 < kp − ĝk2 < =⇒ kp − ĝku < √
3 3 3 2π

Therefore, finally kf − pk2 < 

CHAPTER 10. FOURIER CALCULUS 180

Theorem 10.5 (Parseval Identity). Given f ∈ L2 [−π, π] we have that

ˆ π 2 ∞
2 |a0 | X 2 2
|f (x)| dx = + |ak | + |bk |
−π 2
k=1

Proof. The proof is quite straightforward, since trigonometric polynomials form a basis for L2 [−π, π]
we have that this is simply the already known Parseval identity, since
∞
X
2 2
kf k2 = kck k
k=0

Writing ck = ak + bk we have
2 ∞
2 ka0 k X 2 2
kf k2 = + kak k + kbk k
2
k=1

§§ 10.2.2 Fourier Series in L2 [a, b]

Definition 10.2.3 (Basis of the Space). In order to define a trigonometric basis in L2 [a, b] with a 6= b,
we can use a simple coordinate transformation onto the {(uk ), (vk )} basis of the space L2 [−π, π].
Therefore, taken
π
y(x) = (2x − a − b)
b−a
The new basis on L2 [a, b] will be
 
kπ

(uk (y(x))) = cos(ky(x)) = cos
 (2x − a − b)
 b−a
 
 kπ
 (vk (y(x))) = sin(ky(x)) = sin
 (2x − a − b)
b−a
The completeness of this basis is given by the fact that, this change of coordinates is a smooth
diffeomorphism between L2 [−π, π], L2 [a, b].
Definition 10.2.4 (General Fourier Series). With the previous definition, the Fourier series of a function
f ∈ L2 [a, b] is given as follows
∞    
ã0 X kπ kπ
f (x) ∼ + ãk cos (2x − a − b) + b̃k sin (2x − b − a) (10.5)
2 b−a b−a
k=1

Where  ˆ b  
1 kπ
ã = f (x) cos (2x − b − a) dx


 k
 b−a a b−a
ˆ b   (10.6)
 1 kπ
 b̃k = f (x) sin (2x − b − a) dx


b−a a b−a
10.2. FOURIER SERIES 181

§§ 10.2.3 Fourier Series in Symmetric Intervals, Expansion in Only Sines and Cosines
Definition 10.2.5. We firstly begin finding the Fourier series of a function in L2 [−l, l]. Using the
previous general case in L2 [a, b] and setting a = −l, b = l we have ∀f ∈ L2 [−l, l]
∞    
a0 X kπx kπx
f (x) ∼ + ak cos + bk sin (10.7)
2 l l
k=1

With coefficients ˆ
1 l
  
kπx
ak = l f (x) cos dx



−l l
ˆ (10.8)
1 l
 
 kπx
 bk = f (x) sin dx


l −l l

Theorem 10.6. Taken the space L2 [0, π] we have that both trigonometric sequences (uk (x))
and (vk (x)) are orthogonal bases in this space, and the following equalities hold.
∀f ∈ L2 [0, π]
∞
a0 X
f (x) ∼ 0 + a0k cos(kx)
2
k=1
∞
X
f (x) ∼ b0k sin(kx)
k=1
Where ˆ
2 π
a0k = f (x) cos(kx) dx
π 0
ˆ π
2
b0k = f (x) sin(kx) dx
π 0

Proof. The proof of this theorem is straightforward, we firstly define the even and uneven extensions
of the function f (x) in L2 [−π, π] as follows
(
f (x) x ∈ [0, π]
f e (x) =
f (−x) x ∈ [−π, 0)

And (
u f (x) x ∈ [0, π]
f (x) =
−f (−x) x ∈ [−π, 0)
Expanding both these functions in [−π, π] we have that, indicating the coefficients of each as aek , bek , auk , buk
ˆ ˆ
1 π e 2 π
aek = f (x) cos(kx) dx = f (x) cos(kx) dx = a0k
π −π π 0
bek = 0
auk = 0
ˆ ˆ
1 π u 2 π
buk = f (x) sin(kx) dx = f (x) sin(kx) dx = b0k
π −π π 0
CHAPTER 10. FOURIER CALCULUS 182

Therefore
∞
a0 X
f (x) ∼ 0 +
e
a0k cos(kx)
2
k=1
∞
X
f u (x) ∼ b0k sin(kx)
k=1

Which implies that

2 2
kf e − Sn k2 = 2kf − Sn k[0,π] → 0
And
2 2
kf u − Sn k2 = 2kf − Sn k[0,π] → 0
Proving the theorem.
Example 10.2.1. Taken the function f (x) = x2 x ∈ [−l, l] we want to find the Fourier expansion of
this function.
Since x2 is even, thanks to the previous theorem we know that the coefficients bk = 0 in all the set of
definition, therefore
∞  
2 a0 X kπx
x ∼ + ak cos
2 l
k=1

We firstly calculate the coefficient a0 of the expansion

ˆ
1 l 2 2l2
a0 = x dx =
l −l 3

The coefficients ak can be calculated using the fact that x2 is even, and therefore we have
ˆ l ˆ l
1 l 2
      
kπx 1 2 kπx l 4 kπx
ak = x cos dx = x sin − x sin dx =
l −l l l l kπ −l kπ 0 l
  l ˆ l  
4l kπx 4l kπx
= x cos − sin dx
(kπ)2 l −l (kπ)2 0 l

Since the last integral is 0 we have

l
(−1)k l2
 
4l kπx
ak = x cos =
(kπ)2 l −l (kπ)2

The searched Fourier expansion is therefore

∞
l2 4l2 X (−1)k
 
kπx
x2 ∼ + 2 cos
3 π k2 l
k=1

Example 10.2.2 (Parseval’s Equality). Having now found the Fourier expansion for x2 , we can use
Parseval’s equality in order to calculate the sum
∞
X 1
k2
k=1
10.2. FOURIER SERIES 183

Thanks to Parseval, we therefore have

ˆ l ∞
2 1 2l4 16l2 X 1
x2 2 = 4
x dx = + 4
l −l 9 π k4
k=1

The integral on the left is obvious, and moving the terms around we finally have
∞ ˆ l
π4 π4 π4 2 2
 
X 1 4
= x dx − = −
k4 16l5 −l 36 16 5 9
k=1

Therefore
∞
X 1 π4
=
k4 90
k=1

§§ 10.2.4 Complex Fourier Series

Theorem 10.7 (Complex Exponential Basis). Taken the space L2 [−π, π], and defining a system
(ek )k∈Z = eikx , this system is an orthogonal basis for the space.

Proof. Using Euler’s formula for complex exponentials we have

(ek )k∈Z = eikx = cos(kx) + i sin(kx) = uk (x) + ivk (x)

Therefore, due to the linearity of the scalar product, these functions are orthogonal to each other, and
due to linearity we also have

span{eikx } = span{cos(kx), sin(kx)}

Which, implies
span{eikx } = L2 [−π, π]
Note that
2
eikx 2
= 2π

Definition 10.2.6 (Complex Fourier Series). Given f ∈ L2 [−π, π] we can now define a Fourier expansion
in complex exponentials as follows
∞
X hf (x), eikx i X
f (x) ∼ 2 eikx = ck eikx (10.9)
k=−∞ keikx k2 k∈Z

Where, the coefficients will be

ˆ π
hf (x), eikx i 1
ck = 2 = f (x)eikx dx (10.10)
keikx k2 2π −π
CHAPTER 10. FOURIER CALCULUS 184

Note that if f (x) : R −→ R we have

ˆ π ˆ π ˆ π
1 1 1
ck = f (x)e ikx
dx = f (x)eikx dx = f (x)e−ikx dx = c−k
2π −π 2π −π 2π −π

Therefore, for a real valued function

∞
X ∞
X
ck eikx + c−k e−ikx = c0 + 2 Re ck eikx

f (x) ∼ c0 + (10.11)
k=1 k=1

Example 10.2.3. Taken the function f (x) = ex , x ∈ [−π, π], we want to find the Fourier series in
terms of complex exponentials. Since f (x) is a real valued function, we have
∞
X
e x ∼ c0 + 2 Re ck eikx


k=1

The coefficients will be

ˆ π
1 1
sinh(π)
c0 = ex dx = eπ − e−π =
2π −π 2π π
ˆ π  
1 1 1 
ck = e(1−ik)x dx = eπ(1−ik) − e−π(1−ik)
2π −π 2π 1 − ik

The second expression can be seen as follows

1 (−1)k
eikπ eπ − e−ikπ e−π =


ck = sinh(π)
2π(1 − ik) π(1 − ik)

The final expansion will then be given from finding the real part of this coefficient times the basis vector,
i.e.
(−1)k sinh(π) ikx (−1)k sinh(π)
 
Re e = Re ((1 + ik)(cos(kx) + i sin(kx)))
π(1 − ik) 1 + k2

The last calculation is obvious, and we therefore have

(−1)k sinh(π) ikx (−1)k sinh(π)

 
Re e = (cos(kx) − k sin(kx))
π(1 − ik) 1 + k2

And the final solution will be

∞
sinh(π) 2 sinh(π) X (−1)k
ex ∼ + (cos(kx) − k sin(kx))
π π 1 + k2
k=1
10.2. FOURIER SERIES 185

§§ 10.2.5 Piecewise Derivability, Pointwise and Uniform Convergence of Fourier Series

Definition 10.2.7 (One Sided Derivatives). Let f : [a, b] −→ C be a piecewise continuous function and
let x ∈ [a, b). f (x) is said to be right (or left) derivable, if the following limits exist

0 f (x + ) − f (x+ )
f+ (x) = lim
→0+ 
0 f (x + ) − f (x+ )
f− (x) = lim− −
→0 
Where
f (x± ) = lim± f (y)
y→x

Example 10.2.4. Take the following function

0 x<0



1

f (x) = x=0
2


1−x x>0

We have
0 f () − f (0+ ) 1−−1
f+ (0) = lim+ = = −1
→0  
0 f () − f (0+ ) 0
f− (0) = lim+ = =0
→0 − 
It’s important to see how the right and left derivatives might not coincide with the right and left limits
of the derivative, as explained in the following theorem
Theorem 10.8. Let f (x) : [a, b] −→ C be a piecewise differentiable function, then given x ∈ [a, b)
we have
0
f+ (x) = f 0 (x+ )
0
f− (x) = f 0 (x− )

Proof. Since f (x) is piecewise differentiable, we have that ∃γ > 0 such that f (x) is differentiable
∀x ∈ (x, x + γ) and we can define f 0 (x+ )
Therefore ∀α > 0, ∃1 > 0 such that

∀y ∈ (x, x + 1 ) f 0 (y) − f 0 (x+ ) < α

Thanks to the Lagrange theorem, and the definition of one sided limit, we have, given 0 < δ <  ≤ 1
f (x + ) − f (x + δ)
− f 0 (x+ ) < α
−δ
Which implies therefore
f (x + ) − f (x+ )
lim lim − f 0 (x+ ) = 0
→0+ δ→0+ 
CHAPTER 10. FOURIER CALCULUS 186

Which also implies, by definition

0 f (x + ) − f (x+ )
f+ (x) = lim = f 0 (x+ )
→0+ 

Definition 10.2.8 (Periodic Extension). Given a function f : [−π, π] −→ C, non periodic, we define the
periodic extension f̃ : R −→ C as

f̃ (x) = f (x + 2kπ) k ∈ Z, x + 2kπ ∈ (−π, π]

Note that it coincides with the same function, given that x ∈ (−π, π] and therefore the periodic
extension has discontinuities of the first kind at the points xk = (2k + 1)π, k ∈ Z

Lemma 10.2.1 (Riemann-Lebesgue). Let f ∈ C[a, b] be a function such that f 0 is piecewise continuous
(also holds ∀f ∈ L1 [a, b]), then
ˆ b
lim f (x) sin(sx) dx = 0
s→∞ a

Proof. The proof comes directly from the calculus of the integral
ˆ b ˆ b
1 b 1
f (x) sin(sx) dx = − [f (x) cos(sx)]a + f 0 (x) sin(sx) dx
a s s a

Since kf ku = M and kf 0 ku = M 0 , we have

ˆ b
1
f (x) sin(sx) dx ≤ (2M + |b − a|M 0 ) → 0
a |s|

Definition 10.2.9 (Dirichlet Kernel). We define the Dirichlet kernel as the following function

! n
1 sin 2n+1
2
z 1 1X
Dn (z) = = + cos(kz)
2π sin z2 2π π
k=1

0
Lemma 10.2.2. Given f : R −→ C a piecewise continuous function, and x ∈ R such that exists f+ (x)
we have that ˆ π
 1

 lim f (x + z)Dn (z) dz = f (x+ )
 n→∞ 0 2
ˆ 0
1
f (x + z)Dn (z) dz = f (x− )

 lim

n→∞ −π 2
10.2. FOURIER SERIES 187

Theorem 10.9. Given f : R −→ C a 2π−periodic piecewise continuous function and letting Sn

be the n-th partial sum of the Fourier series expansion of f and letting x ∈ R such that exist
both the left and right derivative in the point, we have that
f is continuous in x

f (x)
lim Sn (x) = 1
 f (x+ ) + f (x− ) f is not continuous inx


n→∞
2
Or also
1
f (x+ ) + f (x− )


lim Sn (x) =
n→∞ 2

Proof. By definition of the Fourier expansion, we have that, in the trigonometric basis
n
a0 X
Sn (x) = + ak cos(kx) + bk sin(kx)
2
k=1

Inserting the usual definitions of the Fourier coefficients and using the fact that the sum is finite, hence
it converges, we have
"ˆ n
! #
π
1 1 X
Sn (x) = f (s) + cos(kx) cos(ks) + sin(kx) sin(ks) ds
π −π 2
k=1

Simplifying
ˆ π n
!
1 1 X
Sn (x) = f (s) + cos (k(s − x)) ds
π −π 2
k=1

Rearranging the second term we see that it’s the Dirichlet kernel, and using a transformation z = s − x,
we have ˆ π
Sn (x) = f (x + z)Dn (z) dz
−π

Note that the extremes of integration are the same since both these functions are 2π−periodic.
Using the definition of the integral between f and the Dirichlet kernel, we have finally the statement
of the theorem, in the case that the function has a discontinuity at the point x
1
f (x+ ) + f (x− )


lim Sn (x) =
n→∞ 2

Theorem 10.10 (Pointwise Convergence of the Fourier Series). Given a piecewise continous 2π−pe-
riodic function f : R −→ C, we have that the Fourier series converges pointwise to the
following two cases, in case the function is continous or not in the point x ∈ R
Theorem 10.11 (Uniform Convergence of the Fourier Series). Given a 2π−periodic function f ∈
C(R), such that its derivative is piecewise continuous, we have that
Sn ⇒ f
CHAPTER 10. FOURIER CALCULUS 188

Proof. Since f (x) ∈ C(R) for the previous theorem we have that Sn (x) → f (x) ∀x ∈ R, therefore
∞
a0 X
f (x) = + ak cos(kx) + bk sin(kx) x ∈ R
2
k=1

Using the Weierstrass M-test we have that, taken the following sequence

ck = ak cos(kx) + bk sin(kx)

The sequence is limited as follows

|ck | ≤ |ak ||cos(kx)| + |bk ||sin(kx)| ≤ |ak | + |bk | = Mk

In order to check that the sum of this extremum is convergent we find the Fourier expansion of the
derivative of f
∞
a0 X
f 0 (x) ∼ 0 + a0k cos(kx) + b0k sin(kx)
2
k=1

It’s not hard to prove that 

b
a0k = − k

k
 b0k = ak

k
Therefore
∞ ∞ ∞
X X 1 1X 0 2 2 2
|ak | + |bk | = (|a0k | + |b0k |) ≤ |ak | + |b0k | + 2 < ∞
k 2 k
k=1 k=1 k=1

Since ak , a0k , bk , b0k 2

∈ ` (R) Therefore, the sum converges and for Weierstrass’ M-test it converges
uniformly

§§ 10.2.6 Solving the Heat Equation with Fourier Series

Definition 10.2.10 (Heat Equation). In physics, the equation governing heat transfer is the heat
equation a partial differential equation of order 2 in space and of order 1 in time.
The equation is the following
∂u ∂2u
=λ 2 (10.12)
∂t ∂x

Example 10.2.5 (Heat Equation with Von Neumann Boundary Conditions). We firstly write the heat
equation with Von Neumann boundary conditions

∂t u = λ∂x2 u



∂x u(0, t) = ∂x u(l, t) = 0 (10.13)

u(x, 0) = f (x)


Where x ∈ [0, l], t > 0

We suppose that u(x, t) is expressible as a uniformly convergent Fourier series of only sines or cosines.
10.2. FOURIER SERIES 189

Since we want the derivative on the x to vanish in order to satisfy immediately the boundary conditions,
we suppose the following expansion
∞  
a0 (t) X kπx
u(x, t) = + ak (t) cos
2 l
k=1

Deriving, we have
∞
a00 (t) X 0
 
kπx
∂t u = + ak (t) cos
2 l
k=1
∞  
πX kπx
∂x u = − kak (t) sin
l l
k=1
∞
π2 X 2
 
2 kπx
∂x u = − 2 k ak (t) cos
l l
k=1

It’s immediate to see that the boundary conditions are immediately satisfied, and therefore, reinserting
it back into the differential equation, we get
∞ ∞
π2 λ X 2 a0 (t) X 0
   
kπx kπx
− 2 k ak (t) cos = 0 + ak (t) cos
l l 2 l
k=1 k=1

Therefore, we end up with the following infinite system of ODEs of order 1

 0
a0 (t) = 0

 2
kπ
a0k (t) = −λ
 ak (t)
l

With k > 1, k ∈ N. Therefore, integrating we get

kπ

2
a0 = a0 ak (t) = ak e−λ l t

Reinserting into the second boundary condition u(x, 0) = f (x) we end up determining the coefficients
as the cosine-Fourier coefficients of the function f (x)
ˆ l  
2 kπx
ak = f (x) cos dx
l 0 l
ˆ l
1
a0 = f (x) dx
l 0

Therefore, the complete solution to the PDE is

ˆ l ∞  ˆ l  
1 2X − kπ

2
λt kπx kπs
u(x, t) = f (x) dx + e l cos f (s) cos ds
2l 0 l l 0 l
k=1
CHAPTER 10. FOURIER CALCULUS 190

Example 10.2.6 (Dirichlet Boundary Conditions). Let’s now take the heat equation with different
boundary conditions, namely
2

 ∂t u = λ∂x u

u(x, 0) = f (x)

u(0, t) = u(l, t) = 0

+
Where (x, t) ∈ [0, l] × R \ {0} Since the first derivative doesn’t appear in the boundary conditions
we choose a particular form of u(x, t) in terms of an only sine Fourier expansion (assuming uniform
convergence). We have therefore
∞  
X kπx
u(x, t) = bk (t) sin
l
k=1

Deriving, we get therefore

∞  
X kπx
∂t u = b0k (t) sin
l
k=1
∞
π2 X 2
 
kπx
∂x2 u = − k b k (t) sin
l2 l
k=1

Reinserting into the differential equation, we have

∞
λπ 2 k 2
   
X kπx kπx
b0k (t) sin + bk (t) sin =0
l l2 l
k=1

Therefore, equating the coefficients for the infinite ODEs we get

 2
kπ
b0k (t) = bk (t)
l
kπ

2
bk (t) = bk e−λ l t

And our particular solution will be, therefore

∞  
X 2 2
− λkl2π t kπx
u(x, t) = bk e sin
l
k=1

Imposing the last condition we get

∞  ˆ l  
X 2 2
− λkl2π t kπx kπs
u(x, t) = e sin f (x) sin ds
l 0 l
k=1

§ 10.3 Fourier Transform

§§ 10.3.1 Fourier Integrals, Translations, Dilations
Proposition 16 (Extending the Fourier Series). Let f : R −→ C be a non periodic function and
fl : [−l, l] ⊂ R −→ C be a function with periodic extension that converges to f (x) for l → ∞.
10.3. FOURIER TRANSFORM 191

We have, using the complex exponential basis that

ˆ
1 X −ikπx l ikπs
fl (x) ∼ e l fl (s)e l ds
2l −l
k∈Z

Sending l → ∞ we have that the sum of coefficients behaves like a Riemann sum and converges to the
following integral ˆ
g(λ)eikx dx = f (x)
R

Where the last equality is given by the fact that fl (x) → f (x).
We define the integral used for finding these “coefficients” the Fourier Integral Transform of f
ˆ
1
g(λ) = F̂[f ](λ) = f̂ (λ) f (x)e−iλx dx
2π R

Definition 10.3.1 (Parity, Translation and Dilation Operators). Let f : R −→ C be some function. We
define the following operators

P̂ [f ](x) = f (−x) Parity

T̂a [f ](x) = f (x − a) Translation
Φ̂a [f ](x) = f (ax) Dilation

Definition 10.3.2 (Fourier Operator). Given a function f ∈ L1 (R) we define the Fourier operator
F̂[f ] as follows ˆ
F̂[f ] = f (x)e−iλx dx λ ∈ R
R

Which is basically the Fourier transform f̂ of the function f .

Note that F̂ : L1 (R) → L1 (R), since
ˆ ˆ
F̂[f ] = f (x)e−iλx dx = |f (x)| dx
R R

Theorem 10.12 (Properties of the Fourier Transform). Given f, g ∈ L1 (R) and a, b ∈ C we hafe that
1. F̂[af + bg] = aF̂[f ] + bF̂[g]

2. F̂[f ](λ) = F̂[f ](−λ)

3. Im(f ) = 0 =⇒ F̂[f ](λ) = F̂[f ](−λ)

 
4. Im(f ) = 0, f even =⇒ Im F̂[f ] = 0, F̂[f ] even
 
5. Im(f ) = 0, f uneven =⇒ Re F̂[f ] = 0, F̂[f ] uneven
CHAPTER 10. FOURIER CALCULUS 192

Theorem 10.13 (Action of the Dilation, Parity and Translation Operators on the Fourier Operator).
Given f ∈ L1 (R) and a ∈ R \ {0} we have
1. F̂ P̂ = P̂ F̂
2. F̂ T̂a = e−iλa F̂
3. T̂a F̂ = F̂[eiax f (x)]
−1
4. F̂ Φ̂a = |a| Φ̂a−1 F̂
−1 −iλb
5. F̂Φa T̂b = |a| e Φ̂a−1 F̂
−1 −iλb/a
6. F̂ T̂b Φ̂a = |a| e Φ̂a−1 F̂
7. F̂ D̂ = iλF̂
Example 10.3.1 (Fourier Transform of the Set Index Function). Let f (x) = 1[−a,a] (x) be the index
function of [−a, a], we have
ˆ ˆ a
i 2
F̂[1[−a,a] ](λ) = 1[−a,a] (x)e−iλx dx = e−iλx dx = [e−iλx ]a−a = sin(λa)
R −a λ λ

For a = 1/2 we have 1−1/2,1/2 (x) = rect(x) and therefore

 
sin(λ/2) λ
F̂[rect(x)](λ) = = sinc
λ/2 2π

Example 10.3.2 (Fourier Transform of the Triangle Function). We define the triangle function tri(x) =
max {1 − |x|, 0} = rect(x/2)(1 − |x|). We then have
ˆ ˆ 1
F̂[max {1 − |x|, 0}] = max {1 − |x|, 0} e−iλx dx = (1 − |x|)e−iλx dx
R −1

Using the properties of the absolute value and using a change in coordinayes we have
ˆ 1 ˆ 1
(1 − x) eiλx + e−iλx dx = 2


F̂[max {1 − |x|, 0}] = (1 − x) cos(λx) dx
0 0

By direct integration, we therefore get

ˆ
2 1
 
2 4 λ
F̂[max {1 − |x|, 0}] = sin(λx) dx = 2 (1 − cos(λ)) = 2 sin2 (λ/2) = sinc2
λ 0 λ λ 2π

Example 10.3.3 (A Couple Fourier Transforms More). 1) f (x) = H(x)e−ax where a ∈ R, a > 0 This
one is quite straightforward. We have
ˆ ˆ +∞
e−(a+iλ)x

1
F̂[H(x)e−ax ] = H(x)e−(a+iλ)x dx = e−(a+iλ)x = − =
R R+ a + iλ 0 a + iλ
10.3. FOURIER TRANSFORM 193

2) f (x) = 1/(1 + x2 ) For this we immediately choose to use the residue theorem, so, we firstly suppose
that λ < 0 and we choose a path enclosing the region Im(z) < 0, for which the only singularity is
given by z̃ = −i ˆ
e−iλz
   −iλz 
1 e
F̂ = lim dz = −2πi Res
1 + x2 R→∞ γ − 1 + z 2
R
z=−i 1 + z 2

Therefore
e−iλz
   
1
F̂ = −2πi lim (z + i) = πeλ
1 + x2 z→−i (z + i)(z − i)
Analogously, for λ > 0 we choose a curve enclosing the region Im(z) > 0, and therefore, noting that
the only encompassed singularity is z̃ = i
ˆ
e−iλz
   −iλz 
1 e
F̂ = lim dz = 2πi Res
1 + x2 R→∞ γ + 1 + z 2
R
z=i 1 + z2

Therefore
e−iλz
   
1
F̂ = 2πi lim (z − i) = πe−λ
1 + x2 z→i (z + i)(z − i)
Uniting both cases, i.e. for λ ∈ R, we have finally
 
1
F̂ = πe−|λ|
1 + x2

3) f (x) = e−a|x| Last but not least, we can calculate this Fourier transform using the properties of the
Fourier operator. Firstly
e−a|x| = H(x)e−ax + H(−x)eax
We can also write this as follows

e−a|x| = H(x)e−ax + P̂ [H(x)e−ax ]

Therefore, using the linearity of the Fourier transform and the behavior of it under parity transformations,
we have  
1 1 1 1 2a
F̂[e−a|x| ] = + P̂ = + = 2
a + iλ a + iλ a + iλ a − iλ a + λ2

2
Example 10.3.4 (A Particular Way of Solving a Fourier Integral). Take now the function f (x) = e−x ,
using the properties of this function under derivation we can build easily a differential equation

df
= −2xf (x)
dx
Applying the Fourier operator on both sides we get
 
df
F̂ = −2F̂ [xf (x)]
dx
d
iλF̂[f (x)] = −2i F̂[f (x)]
dλ
CHAPTER 10. FOURIER CALCULUS 194

Therefore, we’ve built a differential equation in the Fourier domain, where the searched function is
actually the Fourier transform of the initial equation. Solving, we get

d log λ
F̂[f (x)] = −
dλ 2
´ 2 √
Therefore, imposing the condition of integration on all R and remembering that R
e−x dx = π, we
have
√ λ2 2
F̂[f (x)] = πe− 4 = F̂[e−x ]

§§ 10.3.2 Behavior of Fourier Transforms

Theorem 10.14. Let f ∈ L1 (R) be some function. Taken f̂ (λ) = F̂[f ](λ), we have that f̂ ∈ C0 (R)

Proof. Firstly, we need to demonstrate that f̂ ∈ C(R). Therefore, by definition of continuity we have
ˆ
f̂ (λ + ) − f̂ (λ) = f (x)e−iλx (e−ix − 1) dx
R

Using the properties of the modulus operator, we have that

ˆ
f̂ (λ + ) − f̂ (λ) ≤ |f (x)| e−ix − 1 dx
R

For some a ∈ R, we also have that

ˆ ˆ ˆ
|f (x)| e−ix − 1 dx ≤ |f (x)| e−ix − 1 dx + 2 |f (x)| e−ix − 1 dx
R |x|≤a |x|>a

And for x ≤ a we can also say

 x 
2
e−ix − 1 = (1 − cos(x)) − sin2 (x) = 4 sin2 ≤ (x)2 ≤ (a)2
2
Letting kf k1 be the usual p integral norm on L1 (R), we have therefore
ˆ ˆ ˆ
|f (x)| e−ix − 1 dx + 2 |f (x)| e−ix − 1 dx ≤ |a|kf k1 + 2 |f (x)| e−ix − 1 dx
|x|≤a |x|>a |x|>a

The last integral goes to 0 for a = −1/2 , therefore

f̂ (λ + ) − f̂ (λ) ≤ |a|kf k1

Proving that f̂ ∈ C(R).

Instead, for proving that f̂ (λ) → 0 for λ → ∞, we have
ˆ ˆ
f̂ (λ) ≤ f (x)e−iλx dx + 2 f (x)e−iλx dx
|x|≤a |x|>a
10.3. FOURIER TRANSFORM 195

We have then
ˆ
f̂ (λ) ≤ f (x)e−iλx dx + 
|x|≤a

For the Riemann-Lebesgue lemma we therefore have

ˆ
lim f̂ (λ) = lim f (x)e−iλx dx = 0
λ→∞ λ→∞ |x|≤a

Theorem 10.15. Given f ∈ C p−1 (R) a function, such that ∂ p f (x) is piecewise continuous, and
∂ k f (x) ∈ L1 (R) for k = 1, · · · , p. If this holds, we have that
1. F̂[∂ k f ](λ) = (iλ)k F̂[f ], for k = 1, · · · , p
2. limλ→±∞ λp F̂[f ](λ) = 0
Proof. Through integration by parts of the definition of the Fourier transform we have

F̂[f 0 ](λ) = [f (x)e−iλx ]R + iλF̂[f ](λ)

If the evaluation of f (x)e−iλx on all R gives back 0 we have that the first part of the theorem is
demonstrated through iteration.
Using that f 0 ∈ L1 (R) tho, we can define using the fundamental theorem of calculus
ˆ x
f (x) = f (0) + f 0 (s) ds
0

Also, since f 0 ∈ L1 (R) we must have that the limits at ±∞ of f (x) must be finite, therefore

lim f (x)e−iλx = 0
x→±∞

And
F̂[f 0 ](λ) = iλF̂[f ](λ)
Through this, we therefore have by iteration that
1
λp F̂[f ](λ) = F̂[f (p) ](λ)
ip
Which, thanks to Riemann-Lebesgue, gives
1
lim λp F̂[f ](λ) = lim F̂[f ](λ) = 0
λ→∞ ip λ→∞

Theorem 10.16. Given f ∈ L1 (R) such that xk f ∈ L1 (R) for k = 1, · · · , p, we have that F̂[f ] ∈
C p (R), and
∂ k F̂[f ](λ) = F̂[(−ix)k f (x)](λ)
CHAPTER 10. FOURIER CALCULUS 196

Proof. In order to see if it’s true, we start for the first derivative and apply the definition. Therefore,
given f̂ (λ) = F̂[f ](λ) ˆ
1
(f̂ (λ + ) − f̂ (λ)) − (−ix)f (x)e−iλx dx
 R

Using the triangle inequality and the definition of f̂ (λ) we have

ˆ ˆ
e−ix − 1 e−ix − 1
|f (x)| + ix dx = |xf (x)| + i dx
R  R x

Dividing the improper integral around a point a ∈ R we have that everything is lesser or equal to the
following quantity
ˆ ˆ
e−ix − 1 e−ix − 1
|xf (x)| + i dx + 2 |xf (x)| + i dx
|x|≤a x |x|>a x

We also have that

e−ix − 1 cos(x) − 1 x sin(x) |cos(x) − 1| |x − sin(x)|
+i ≤ +i ≤ +
x x x |x| |x|

Using the Taylor expansions for the cosine and the sine we get, approximating, that
1 2
|cos θ − 1| ≤ θ
2
1 3
|sin(θ) − θ| = |θ|
3!
Therefore
e−ix − 1 1 2 1 3
+ i ≤ |x| + |x|
x 2 6
Putting that back into the integral, we have that it must be smaller or equal to the following quantity
ˆ 2
! ˆ 2
!
|x| |x| x |x|
|xf (x)| + dx + 2 |xf (x)| + dx
|x|≤a 2 6 |x|>a 2 6

Using in the first integral that |x| ≤ |a| we get, that all the quantity will be surely less than the
supremum of such, and therefore
  ˆ
|a| |a|
+ kxf (x)k1 + 2 |xf (x)| dx
2 6 |x|>a

Therefore, imposing as before a = −1/2 , we get

ˆ
f̂ (λ + ) − f̂ (λ)
lim − (−ix)f (x)e−iλx dx = 0
→0  R

Proving the thesis of the theorem

10.3. FOURIER TRANSFORM 197

Theorem 10.17 (Invariance of the Schwartz Space under Fourier Transforms). Given f ∈ S(R) ⊂ L1 (R),
then F̂[f ] ∈ S(R).
 
Proof. We will prove a weaker assumption. Since S(R) = C ∞ (R), k·kj,k where the seminorm
k·kj,k is defined as follows
kf kj,k = xj ∂ k f u
<∞ j, k ∈ N
We have, ∀f ∈ S(R) that for a given constant Ca,k ∈ R

Ca,k
∂kf ≤ a a, k ∈ N, x ∈ R
(x2 + 1) 2

Therefore, taken a = j + 2
ˆ ˆ j ˆ
j k |x| 1
x ∂ f (x) dx ≤ Cj+2,k j dx ≤ Cj+2,k dx < ∞
R R (x2 + 1) 2 +1 R x2 + 1

Therefore, we have that xj ∂ k f ∈ L1 (R) ∀j, k ∈ N. But we can also write the following result using
Leibniz’s rule
j  
j k

X j
∂ x f (x) = ∂ m xj ∂ j−m f (x) ∈ L1 (R)
m=0
m

Therefore, applying the Fourier transform and using the previous property, we have

(iλ)j (∂ k f̂ (λ)) = (iλ)j (−i)k F̂[xk f ](λ) = (−i)k F̂[∂ j (xk f )](λ) ∈ C0 (R)

Which finally gives

f̂ = F̂[∂ j (xk f )](λ) <∞ ∀j, k ∈ N
j,k j,k

Which finally gives

f ∈ S(R) =⇒ F̂[f ] ∈ S(R)

§§ 10.3.3 Inverse Fourier Transform

Definition 10.3.3 (Fourier Antitransform). Let f ∈ L1 (R), we define the Fourier Antitransform as
F̂ a [f ](x) the following computation
ˆ
1
F̂ a [f ](x) = f (λ)eiλx dλ (10.14)
2π R

Note that, taken the transformation λ = −u we get

ˆ
1 1
F̂ a [f ](x) = f (−u)e−iux du = F̂ ◦ P̂ [f ]
2π R 2π

This brings to the definition of the following theorem

CHAPTER 10. FOURIER CALCULUS 198

Theorem 10.18 (Inversion Formula). Given f ∈ S(R), we have that F̂ a F̂[f ] = F̂ F̂ a [f ] = 1̂[f ] = f ,
therefore F̂ a = F̂ −1 and the Fourier transform is a bijective map in S(R)
∼
F̂ : S(R) −→ S(R)

Proof. Weakening the statement, we can say that taken f ∈ K such that f (x) = 0 for |x| > a ∈ R,
and taken  ∈ (0, 1/a), we have that, Fourier expanding the function in [−−1 , ], we get
ˆ 1
X 2  
f (x) = ck e ikπx
, ck = f (x)e−ikπx dx → F̂[f ](kπ)
 − 1 2
k∈Z

Therefore, we can immediately write

X
f (x) = F̂[f ](kπ)eikπx
k∈Z

Letting λk, = kπ we have

ˆ
1 X 1
f (x) = F̂[f ](λk, )eiλk, ∆λk, → F̂[f ](λ)eiλx dx = F̂ a F̂[f ]
2π 2π R
k∈Z

Where we let  → 0+ in the Fourier series, which written in that way gives a Riemann-Lebesgue sum
converging to the integral of the antitransform, therefore proving that in S(R) F̂ a = F̂ −1

Theorem 10.19 (Plancherel). Given f, g ∈ S(R), and let h·, ·i be the usual scalar product defined
as follows ˆ
hf, gi = f (x)g(x) dx
R

Then, we have
√
F̂[f ] = 2πkf k2
2 (10.15)
hF̂[f ], F̂[g]i = 2πhf, gi

Proof. For Parseval we have

ˆ
2 X 2 1 X
kf (x)k dx = |ck | = F̂[f ](λk, ) ∆λk,
R 2 2π
k∈Z k∈Z

Taking the limit  → 0+ the sum on the right converges to the following value
ˆ ˆ 2 √
2 1
|f (x)| dx = F̂[f ] dλ =⇒ kf k2 = 2πkf k2
R 2π R

Then for the polarization identity, and this result, we get the final proof of the theorem
10.3. FOURIER TRANSFORM 199

Theorem 10.20 (Continuity Expansion). Let p, q ∈ C(R), then if p(x) = q(x) ∀x ∈ Q we have that

p(x) = q(x) ∀x ∈ R

This holds for any dense subset of R

Proof. Since Q is dense in R we can take a sequence (xn )n∈N ∈ Q such that xn → x ∈ R. Therefore,
using the continuity of p, q we have
   
p(x) = p lim xn = lim p(xn ) = lim q(xn ) = q lim xn = q(x)
n→∞ n→∞ n→∞ n→∞

Theorem 10.21 (Extension of the Inversion Formula). Let f, g ∈ S(R). Taken a metric dS (·, ·) :
S(R) × S(R) → R defined as follows
∞ X
∞
X 1 kf − gkj,k
dS (f, g) = j, k ∈ N
j=0 k=0
2j+k 1 + kf − gkj,k

Where k·kj,k is the Schwartz seminorm.

Since K ⊂ S(R) is dense with this norm, we have that we can extend continuously the Fourier
inversion formula ∀f ∈ S(R), we have

F̂ a F̂ = F̂ F̂ a = 1̂ =⇒ F̂ a = F̂ −1 ∀f ∈ S(R)

Theorem 10.22. Given f ∈ L1 (R), then, since we might have that F̂[f ] ∈ / L1 (R), using the
Cauchy principal value
ˆ
1 1
PV F̂[f ](λ)eiλx dλ = f (x− ) + f (x+ )


∀x ∈ R (10.16)
2π R 2

And, if f is continuous in x, we have that

ˆ
1
PV F̂[f ](λ)eiλx dλ = F̂ −1 F̂[f ] = 1̂[f (x)] = f (x)
2π R

Theorem 10.23 (New Calculation Rules). With what we added so far, in operatorial form, we can
write down two new calculation rules. Supposing the inversion formula holds, and therefore
F̂ a = F̂ −1
1  
F̂ −1 = F̂ P̂
2π (10.17)
F̂ −1 F̂ = 2π 1̂
CHAPTER 10. FOURIER CALCULUS 200

Example 10.3.5. Let’s calculate the Fourier transform of the following function f : R −→ C
1
f (x) =
(x + i)3

Acting symbolically on this we know already that F̂ [H(x)e−x ] = (1 + iλ)−1 , therefore

 
1
F̂ −1 (λ) = 2πH(−λ)eλ
1 + ix
Applying the parity operator and multiplying the inverse-transformed function by −i we obtain
 
−1 1
F̂ = −2iπH(λ)e−λ
x+i
Lastly, we can derive it twice and divide the result by two, obtaining
 
−1 1
F̂ = iπλ2 H(λ)e−λ
(x + i)3

§§ 10.3.4 Convolution Product

Definition 10.3.4 (Convolution Product). Given f, g ∈ L1 two bounded functions, we define the
convolution of these two functions as follows
ˆ
(f ∗ g)(x) = f (y)g(x − y) dy (10.18)
R

Theorem 10.24. Defined the convolution product of two bounded functions, we have that
∗ : L1 (R) × L1 (R) −→ L1 (R)
And
kf ∗ gku ≤ kf ku kgk1
kf ∗ gk1 ≤ kf k1 kgk1

Proof. The proof of the first result is direct

ˆ ˆ
f (y)g(x − y) dx ≤ kf ku |g(x − y)| dy = kf ku kgk1
R R

For the second proof, we begin taking a compact set [a, b] ⊂ R and we move forward sending a → −∞
and b → ∞. Therefore
ˆ b ˆ ˆ b−y ¨
|(f ∗ g)(x)| dx ≤ |f (y)| |g(u)| du ≤ |f (y)||g(u)| dy du = kf k1 kgk1
a R a−y R2

Therefore we have that

∗ : L1 (R) × L1 (R) −→ L1 (R)
And that the convolution product is bounded
10.3. FOURIER TRANSFORM 201

Theorem 10.25 (Properties of the Convolution Product). Given f, g, h ∈ L1 (R) three bounded
functions, then
f ∗g =g∗f
(10.19)
(f ∗ g) ∗ h = f ∗ (g ∗ h) = f ∗ g ∗ h
These properties are easily demonstrable using the properties of the integral

Theorem 10.26 (Derivation of a Convolution). Given f ∈ L1 (R) a bounded function, and g ∈

L1 (R) ∩ C 1 (R) a bounded function, we have that

df ∗ g dg
=f∗ = f ∗ g0 (10.20)
dx dx

Proof. Written A(x, t) = f (t)g(x − t) we have that

ˆ
(f ∗ g)(x) = A(x, t) dt
R

And therefore ˆ
df ∗ g ∂
= A(x, t) dt
dx ∂x R

Since we also have that, due to the boundedness of g 0 , that

∂A
= |f (t)g 0 (x − t)| ≤ M |f (t)|
∂x

Due to the fact that f (t) ∈ L1 (R) the integral is well defined, and using Leibniz’s derivation rule, we
have ˆ ˆ
df ∗ g ∂A
= dt = f (t)g 0 (x − t) dt = (f ∗ g 0 )(x)
dx R ∂x R

Theorem 10.27 (Fourier Transform of a Convolution). Given f, g ∈ L1 (R) two bounded functions,
we have that
F̂[f ∗ g] = F̂[f ]F̂[g] (10.21)

Proof. The proof of this is quite direct, we have that

ˆ ˆ ˆ
−iλx −iλx
F̂[f ∗ g](λ) = (f ∗ g)(x)e dx = e f (y)g(x − y) dy dx =
ˆR ˆ R R

= f (y)e−iλy g(x − y)e−iλ(x−y) dx dy = F̂[f ]F̂[g]

R R

Where on the last equality we used the Fubini-Tonelli theorem

CHAPTER 10. FOURIER CALCULUS 202

Example 10.3.6 (Convolution of Two Set Functions). Given a set A = [−a, a] and a set B = [−b, b],
we know that
1A (x)1B (x) = 1A∩B (x)
1[−a,a] (x + y) = 1[−a+y,a+y] (x)
Suppose we need to calculate the convolution product 1A ∗ 1A . The calculation is quite easy with
those properties ˆ
(1A ∗ 1A )(x) = 1A (y)1A (x − y) dy
R
Taken C = [−a + x, a + x], we have
ˆ
(1A ∗ 1A )(x) = 1A∩C (y) dy = µ (A ∩ C)
R

Where, we know already that


0
 x < 2a, x > 2a
µ (A ∩ C) = 2a − x x ∈ [0, 2a]

2a + x x ∈ [2a, 0]


Summarized
µ (A ∩ C) = max {2a − |x|, 0} = (1A ∗ 1A )(x)
Remembering the definition of the rect(x), tri(x) functions, we get that

(rect ∗ rect)(x) = max {1 − |x|, 0} = tri(x)

Example 10.3.7 (Two Gaussians). Taken α, β ∈ R we might want to calculate the following convolution
product ˆ
−αx2 −βx2 2 2
e ∗e = e−αy e−β(x−y) dx
R
The direct calculation of the integral is immediate, but we might want to test here the power of the
last theorem that was stated. Hence we have
h 2 2
i h 2 2
i
F̂ −1 F̂[e−αx ∗ e−βx ] = F̂ −1 F̂[e−αx ]F̂[e−βx ]

Using that r
2 π − λ2
F̂[eax ](λ) = e 4a
a
We have immediately that the searched convolution will be
 4

π

α+β
2 2 −λ
e−αx ∗ e−βx = √ F̂ −1 e 4 αβ (x)
αβ
Which gives s
 
π λ2 π 1 αβ − α+β
  
α+β αβ
−1 − x2
√ F̂ e 4 αβ
(x) = √ √ e
αβ αβ π α + β
10.3. FOURIER TRANSFORM 203

With some rearrangement of the constant terms, we get finally the expected result
r
π
 
2 2 − αβ x2
e−αx ∗ e−βx = e α+β
α+β

§§ 10.3.5 Solving the Heat Equation with Fourier Transforms

The power of Fourier Calculus, as we seen in the Fourier series section, also comes when dealing with
differential equations. In this section, we will consider once again the heat equation, but instead of
considering a finite rod, we will suppose that the rod is actually infinite. We therefore get the following
partial differential equation
∂2u

 ∂u
=k 2 x ∈ R, t ∈ R+ \ {0}
∂t ∂x
u(x, 0) = u0 (x) x ∈ R


We begin by applying the Fourier transform on the solution as follows. The transformed variable will
be indicated as an index
ˆ
v(λ, t) = F̂x [u(x, t)](λ) = u(x, t)e−iλx dx (10.22)
R

Using the properties of the Fourier transform, we have

F̂x [∂x2 u](λ) = (iλ)2 F̂x [u(x, t)](λ) = −λ2 v(λ, t)
Supposing also that v(x, t) is derivable with respect to t (i.e. Leibniz’s rule holds), we have
ˆ
∂v ∂u −iλx
= F̂[∂t u(x, t)](λ) = e dx
∂t R ∂t

And the heat equation, becomes after the Fourier transformation

 ∂v = −kλ2 v(λ, t)

λ ∈ R, t ∈ R+ \ {0}
∂t (10.23)
v(λ, 0) = F̂[u0 (x)](λ) λ ∈ R


The solution is almost immediate, and we therefore get

2 2
v(λ, t) = v(λ, 0)e−kλ t
= F̂[u0 (x)](λ)e−kλ t

Note that
2 1 h x2 i
e−kλ = √ t
F̂x e− 4kt (λ)
2 πkt
Therefore, remembering that the product of transforms gives the transform of the convolution
1 h x2
i
v(λ, t) = F̂x [u(x, t)] = √ F̂x u0 ∗ e− 4kt (λ)
2 πkt
And, the searched solution will therefore be the following
1 x2
u(x, t) = √ u0 ∗ e− 4kt
2 πkt
CHAPTER 10. FOURIER CALCULUS 204
Appendices

205
A Useful Concepts

§ A.1 Multi-Index Notation

In order to ease various calculations one can utilize more abstract index constructions. One of these is
the multi-index notation, where instead of having an index i ∈ N or j ∈ Z, one constructs a “vector”
of indices, like α = (a1 , · · · , an ) ∈ Nn or β = (b1 , · · · , bn ) ∈ Zn .
This notation includes a set of operations on such multi-indexes, defined as follows

Theorem A.1 (Operations on Multi-indexes). Given a multi-index α ∈ Nn , one can define the
following operations on them

n
X
|α| = ai
i=1
n
(A.1)
Y
α! = ai !
i=1

Given x ∈ Rn and the del operator ∂ one can also write

n
Y
xα = xai i
i=1
n
(A.2)
Y ∂ |α| ∂ |α|
α
∂ = ∂iai = a1 an =
i=1
∂x1 · · · ∂xn ∂xα

207
APPENDIX A. USEFUL CONCEPTS 208

§ A.2 Properties of the Fourier Transform

Here’s a list of the properties of the Fourier transform, useful for dealing with calculations in operatorial
form
F̂ P̂ = P̂ F̂
F̂ T̂a = e−iλa F̂
T̂a F̂ = F̂ e−iax f (x) a ∈ R, f ∈ L1 (R)
 

1
F̂ Φ̂a = Φ̂ 1 F̂ a ∈ R \ {0}
|a| a
e−iλb
F̂ Φ̂a T̂b = Φ̂ a1 F̂ a ∈ R \ {0}, b ∈ R
|a|
b
(A.3)
e−iλ a
F̂ T̂b Φ̂a = Φ̂ a1 F̂ a ∈ R \ {0}, b ∈ R
|a|
F̂ ∂ˆ = iλF̂
∂ˆ F̂ = F̂ [−ixf (x)] f ∈ L1 (R)
1 1
F̂ −1 = F̂ P̂ = P̂ F̂
2π 2π
F̂ −1 F̂ = F̂ F̂ −1 = 2π 1̂
B Common Fourier Transforms

§ B.1 Common Fourier Transforms

In this appendix, you’ll find a table of some particular Fourier transforms that might pop up in Fourier
calculus, and might also be helpful for calculating more complex transforms.
Note that
ˆ
F̂[f ](λ) = f (x)e−iλx dx (B.1)
R

f (x) Conditions F̂[f ](λ)

2 sin(λa)
1[−a,a] (x)
λ

rect(x) sinc(λ/2π)

tri(x) sinc2 (λ/2π)

sinc(x) rect(λ/2π)

sin[(λ + 1)a] sin[(λ − 1)a]

cos(x)1[−a,a] +
λ+1 λ−1

209
APPENDIX B. COMMON FOURIER TRANSFORMS 210

f (x) Conditions F̂[f ](λ)

2a
e−|x|
λ2 + a2

1
H(x)e−ax a>0
a + iλ

sgn(a)
H(ax)e−ax
a + iλ

1 π −a|λ|
e
a2 + x2 a

√
1 4ac − b2 π −β|λ|+i λ
β= >0 e 2a
a2 x2 + bx + c 2|a| βa

1 π −|λ|
e (|λ| + 1)
(x2 + 1)2 2

1 |λ|
− √2 λ π


πe sin 2 + 4
1 + x4

r
2 π − λ2
e−ax e 4a
a

r
2 π iλ2
eix (1 + i)e− 4
2

r
−ix2 π iλ2
e (1 − i)e 4
2

√ h
λ2 −π
i
cos x2 π cos 4

√ h 2 i
− π sin λ 4−π


sin x2

 
πλ
sech(x) π sech
2

x2 √ λ2
e− 2 (−i)n 2πe 2
p√ Hn (x) p√ Hn (λ)
π2n n! π2n n!
B.1. COMMON FOURIER TRANSFORMS 211

Remember that:  x 2 x


 sinc = sin



 2π x 2
 rect(x) = 1[− 1 , 1 ] (x)

2 2
(B.2)

 tri(x) = max {1 − |x|, 0}

1


 sech(x) =


cosh(x)
APPENDIX B. COMMON FOURIER TRANSFORMS 212
Bibliography

[Ces18] F. Cesi. Rudimenti di Analisi Infinito Dimensionale, Versione 0.3 (Armadillo). 2018.
[Fla18] Eugenio Montefusco Flavia Lanzara. Esercizi Svolti di Analisi Vettoriale con Elementi di
Teoria. Edizioni LaDotta, 2018. isbn: 9788898648429.
[Pre13] C. Presilla. Elementi di Analisi Complessa, Funzioni di una Variabile. Springer, 2013.
isbn: 9788847055001.
[Spi65] Michael Spivak. Calculus on Manifolds, a Modern Approach to Classical Theorems of
Advanced Calculus. Addison-Wesley Publishing Company, 1965. isbn: 0805390219.

213