ETH Zürich

Analysis I: One Variable

Lecture Notes 2025

Alessio Figalli

December 30, 2025

Preface
These notes originate from the Analysis I course at ETH Zürich. Over the years, various
internal lecture notes have circulated; building on that tradition, I wrote a complete set of
notes during the academic year 2023/2024 and have since undertaken a thorough revision to
improve clarity, coherence, and self-containment. The present text focuses on analysis in one
real variable and is intended to reflect the first-semester course as it is taught to students in
Mathematics, Physics, and Interdisciplinary Natural Sciences.

The guiding philosophy is simple: definitions and theorems are introduced only when they
are truly needed, proofs are written with an emphasis on structure rather than length, and
examples are used to illuminate ideas rather than to replace them. While many topics will
look familiar from high school (limits, derivatives, integrals), the viewpoint is different: we
develop a rigorous theory that explains why the methods of calculus work and how they fit
together. The course requires little beyond basic algebra and an intuitive understanding of
functions; from there, we build the logical framework needed to support later studies, starting
from multivariable calculus in Analysis II.

The organization mirrors the flow of a first encounter with rigorous analysis. An introduc-
tory chapter places classical problems (such as the quadrature of the parabola) in a historical
and conceptual perspective, and it is followed by practical advice on studying effectively.
We then introduce the real numbers through their algebraic and order properties and isolate
completeness as the key axiom; consequences such as the Archimedean principle and dec-
imal expansions are discussed. Complex numbers are introduced early to streamline later
discussions of series and power series.
With this foundation, we turn to sequences, limits, and the first qualitative properties of
functions (boundedness, monotonicity, continuity). Compactness on closed intervals leads to
fundamental results such as the intermediate value property, existence of extrema, and uniform
continuity, and prepares the ground for a precise treatment of exponential and logarithmic
functions. The material on limits of functions and Landau notation formalizes the asymptotic
language used throughout analysis and applications. A short chapter on sequences of functions
(pointwise versus uniform convergence) anticipates the role of uniformity in interchange of
limits, differentiation, and integration.
Series are handled in parallel with sequences: we discuss non-negative and conditionally
convergent series, absolute convergence and its criteria, reordering, and products. Power se-
ries are studied both in the real and complex settings, with radius of convergence, termwise
operations, and the construction of the elementary functions (exponential, trigonometric, log-
arithmic) from their series expansions, including polar coordinates and the complex logarithm.
Differential calculus begins with the derivative as a limit and its geometric meaning, pro-
ceeds through the calculus rules, and culminates in the mean value theorem and its conse-
quences (monotonicity, convexity, l’Hôpital’s rule). Integration is developed via step functions
and Riemann sums, followed by integrability criteria and basic properties. The Fundamental
Theorem of Calculus binds differentiation and integration, after which we present standard
i
Chapter 0.0

techniques (parts, substitution, rational functions), improper integrals, and a first look at
special functions such as the Gamma function. A chapter on Taylor polynomials and analytic
functions closes the circle between local approximation and global information.
The final part provides a concise introduction to ordinary differential equations: linear and
autonomous first-order equations, linear second-order equations with constant coefficients, and
an existence–uniqueness theorem for first-order problems. The emphasis is on illustrating how
the tools developed earlier—continuity, differentiability, integration, and series—combine to
yield both qualitative and quantitative information about solutions.

Throughout, proofs are included for the core results used later in the text; when an argument
is instructive but not essential, it will be marked as “Extra material”. Numerous examples
and exercises are interspersed to consolidate understanding; serious engagement with them is
indispensable, as mastery in mathematics is achieved primarily through problem solving.

These notes will continue to evolve. I have aimed for a balance between brevity and com-
pleteness, avoiding unnecessary generality while keeping the pathway to further topics as
transparent as possible. I hope the text will serve both as a reliable companion during the
semester and as a reference to which students can return in later studies.

Version: December 30, 2025. ii

Contents

1 Introduction 2
1.1 Quadrature of the Parabola . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Naive Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Tips on Studying . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 The Real Numbers: Maximum, Supremum, and Sequences 10

2.1 The Axioms of the Real Numbers . . . . . . . . . . . . . . . . . . . . . . 10
2.1.1 Ordered Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Completeness Axiom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.1.3 Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.1 Definition of Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.2 The Absolute Value on the Complex Numbers . . . . . . . . . . . . . . . 38
2.3 Maximum and Supremum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3.1 Existence of the Supremum . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.3.2 Two-point Compactification . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4 Consequences of Completeness . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4.1 The Archimedean Principle . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.4.2 Decimal Fraction Expansion and Uncountability . . . . . . . . . . . . . 50
2.5 Sequences of Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.5.1 Convergence of Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.5.2 Convergent Subsequences and Accumulation Points . . . . . . . . . . . . 59
2.5.3 Addition, Multiplication, and Inequalities . . . . . . . . . . . . . . . . . 61
2.5.4 Bounded and Monotone Sequences . . . . . . . . . . . . . . . . . . . . . 64
2.5.5 Superior and Inferior Limits . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.5.6 Cauchy Sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.5.7 Improper Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.6 Sequences of Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . 74

3 Functions of one Real Variable 76

3.1 Real-valued Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.1.1 Boundedness and Monotonicity . . . . . . . . . . . . . . . . . . . . . . . 76
3.1.2 Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

iii
Chapter 0.0 CONTENTS

3.1.3 Sequential Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.2 Continuous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.2.1 The Intermediate Value Theorem . . . . . . . . . . . . . . . . . . . . . . 85
3.2.2 Inverse Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3 Continuous Functions on Compact Intervals . . . . . . . . . . . . . . . 90
3.3.1 Boundedness and Extrema . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.3.2 Uniform Continuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.4 Example: Exponential and Logarithmic Functions . . . . . . . . . . . 94
3.4.1 Definition of the Exponential Function . . . . . . . . . . . . . . . . . . . 94
3.4.2 Properties of the Exponential Function . . . . . . . . . . . . . . . . . . . 97
3.4.3 The Natural Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.5 Limits of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.5.1 Limit in the Vicinity of a Point . . . . . . . . . . . . . . . . . . . . . . . 103
3.5.2 One-sided Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.5.3 Landau Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.6 Sequences of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.6.1 Pointwise Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.6.2 Uniform Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

4 Series and Power Series 117

4.1 Series of Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.1.1 Series with Nonnegative Elements . . . . . . . . . . . . . . . . . . . . . . 119
4.1.2 Conditional Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.1.3 Leibniz Criterion for Alternating Series . . . . . . . . . . . . . . . . . . . 123
4.2 Absolute Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.2.1 Criteria for Absolute Convergence . . . . . . . . . . . . . . . . . . . . . 125
4.2.2 Reordering Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.2.3 Products of Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.3 Series of Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.4 Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.4.1 Radius of Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.4.2 Complex Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.5 Example: Exponential and Trigonometric Functions . . . . . . . . . 140
4.5.1 The Exponential Map as a Power Series . . . . . . . . . . . . . . . . . . 140
4.5.2 Sine and Cosine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.5.3 The Circle Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.5.4 Polar Coordinates and Multiplication of Complex Numbers . . . . . . . 148
4.5.5 The Complex Logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.5.6 Other Trigonometric and Hyperbolic Functions . . . . . . . . . . . . . . 150

5 Differential Calculus 153

5.1 The Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

Version: December 30, 2025. iv

Chapter 0.0 CONTENTS

5.1.1 Definition and Geometrical Interpretation . . . . . . . . . . . . . . . . . 153

5.1.2 Differentiation Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.2 Main Theorems of Differential Calculus . . . . . . . . . . . . . . . . . 164
5.2.1 Local Extrema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.2.2 The Mean Value Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.2.3 L’Hôpital’s Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.2.4 Monotonicity and Convexity via Differential Calculus . . . . . . . . . . . 173
5.3 Example: Differentiation of Trigonometric Functions . . . . . . . . 177
5.3.1 Sine and Arc Sine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.3.2 Cosine and Arc Cosine . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5.3.3 Tangent and Arc Tangent . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5.3.4 Hyperbolic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

6 The Riemann Integral 182

6.1 Step Functions and their Integral . . . . . . . . . . . . . . . . . . . . . 182
6.1.1 Decompositions and Step Functions . . . . . . . . . . . . . . . . . . . . 182
6.1.2 The Integral of a Step Function . . . . . . . . . . . . . . . . . . . . . . . 184
6.2 Definition and First Properties of the Riemann Integral . . . . . . 188
6.2.1 Integrability of Real-valued Functions . . . . . . . . . . . . . . . . . . . 188
6.2.2 Linearity and Monotonicity of the Riemann Integral . . . . . . . . . . . 190
6.3 Integrability Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
6.3.1 Integrability of Monotone Functions . . . . . . . . . . . . . . . . . . . . 195
6.3.2 Integrability of Continuous Functions . . . . . . . . . . . . . . . . . . . . 196
6.3.3 Integration and Sequences of Functions . . . . . . . . . . . . . . . . . . . 198

7 The Derivative and the Riemann Integral 200

7.1 The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . 200
7.1.1 The Fundamental Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.1.2 Integration by Parts and by Substitution . . . . . . . . . . . . . . . . . . 203
7.1.3 Improper Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
7.2 Integration and Differentiation of Power Series . . . . . . . . . . . . 209
7.3 Integration Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
7.3.1 Integration by Parts and by Substitution in Leibniz Notation . . . . . . 214
7.3.2 Integration by Parts: Examples . . . . . . . . . . . . . . . . . . . . . . . 215
7.3.3 Integration by Substitution: Examples . . . . . . . . . . . . . . . . . . . 216
7.3.4 Integration of Rational Functions . . . . . . . . . . . . . . . . . . . . . . 219
7.3.5 Definite Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.3.6 The Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
7.4 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.4.1 Taylor Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.4.2 Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231

8 Ordinary Differential Equations 234

Version: December 30, 2025. v

8.1 Ordinary Differential Equations (ODEs) . . . . . . . . . . . . . . . . . 234
8.1.1 Linear First Order ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . 237
8.1.2 Autonomous First Order ODEs . . . . . . . . . . . . . . . . . . . . . . . 242
8.1.3 Homogeneous Linear Second Order ODEs with Constant Coefficients . . 246
8.1.4 Non-Homogeneous Linear Second Order ODEs with Constant Coefficients253
8.2 Existence and Uniqueness for ODEs . . . . . . . . . . . . . . . . . . . . . 259
8.2.1 Existence and Uniqueness for First Order ODEs . . . . . . . . . . . . . 259
8.2.2 Extra material: Proof of Theorem 8.26 . . . . . . . . . . . . . . . . . . . 263
8.2.3 Higher Order ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

1
 Chapter 1

Introduction

1.1 Quadrature of the Parabola

Before embarking on our journey into the world of mathematical analysis, let us examine a
simple example that illustrates the way of thinking we aim to develop. At the same time, it
will serve as a first encounter with the ideas behind integral calculus.
Consider the set
P = (x, y) ∈ R2 | 0 ≤ x ≤ 1, 0 ≤ y ≤ x2 . (1.1)


Our goal is to determine its area.

y

1
1

x
0 1

This region, bounded above by a parabola, was already studied by Archimedes (ca. 287–212
BC), who computed it as the first example of a curvilinear area in the 3rd century BC.

2
 Chapter 1.2

For now, let us assume that we understand the symbols in (1.1) and that P corresponds
to the colored region in the figure. In particular, we take for granted the existence of the real
numbers R.
Of course, once we know integral calculus, the computation is straightforward. But at this
stage we do not want to rely on integrals. Instead, we must face the more basic question:

What is an area?

If we cannot answer this question precisely, then the task of “computing the area of P ” is
not well defined. To proceed, we reformulate our aim as follows.

Proposition 1.1. — Suppose there is a notion of area in R2 satisfying:

1. The area of the rectangle [a, b] × [c, d] = {(x, y) ∈ R2 | a ≤ x ≤ b, c ≤ y ≤ d} is equal to

(b − a)(d − c), where a, b, c, d are real numbers with a ≤ b, c ≤ d.

2. If F, G are domains in R2 with F contained in G, then the area of F is less than or

equal to the area of G.

3. For sets F, G in R2 without common points, the area of the union F ∪ G is the sum of
the areas of F and G.
1
Then the area of the set P defined in (1.1) (if it exists) must be equal to 13 .

In other words, without deciding in general how area is defined, we can already prove that
1
3 is the only consistent value for the area of P .
For the proof of Proposition 1.1 we need the following lemma:

Lemma 1.2. — For every natural number n ≥ 1,

n3 n2 n
12 + 22 + · · · + n2 = + + . (1.2)
3 2 6

The proof of this result uses a standard technique that will appear many times throughout
your studies: the principle of mathematical induction. The idea of induction is the following:
(i) First, you verify that the statement holds for the initial natural number where it is intended
to be true (usually n = 1). This is called the base case.
(ii) Then, you assume that the statement is true for some number n, and under this assump-
tion, you prove that it is also true for n + 1. This is called the induction step.
(iii) If both steps (i) and (ii) succeed, one concludes that the statement is valid for all natural
numbers because starting from n = 1, the induction step allows one to go to n = 2, then to
n = 3, then to n = 4, and so on.

Version: December 30, 2025. 3

 Chapter 1.2

Proof of Lemma 1.2. As mentioned above, we apply induction.

(i) Base case. For n = 1, the left-hand side of (1.2) is 1, while the right-hand side equals

1 1 1
+ + = 1.
3 2 6

Hence (1.2) holds for n = 1.

(ii) Induction step. Assume that for some n ≥ 1,

n3 n2 n
12 + 22 + · · · + n2 = + + . (1.3)
3 2 6

We want to show the validity of the same formula for n + 1, that is

(n + 1)3 (n + 1)2 n + 1
12 + 22 + · · · + n2 + (n + 1)2 = + + . (1.4)
3 2 6

To this aim, we can rely on the assumption that the formula already holds for n. So, using
(1.3) and expanding (n + 1)2 we get

n 3 n2 n
 
2 2 2
1 1 + · · · + n + (n + 1) = + + + (n2 + 2n + 1)
3 2 6
n3 3n2 13n
= + + + 1.
3 2 6

On the other hand, expanding the terms (n + 1)3 and (n + 1)2 , we get

(n + 1)3 (n + 1)2 n + 1 n3 + 3n2 + 3n + 1 n2 + 2n + 1 n + 1

+ + = + +
3 2 6 3 2 6
n3 3n2 13n
= + + + 1,
3 2 6

which matches the previous expression. Hence (1.4) holds.

(iii) Conclusion. Since the identity is true for n = 1, and the induction step shows it remains
true from n to n + 1, the principle of induction implies that (1.2) holds for all n ≥ 1.

Proof of Proposition 1.1. Assume that there is a notion of area satisfying the three properties
of the proposition, and that it is defined for P . Let A denote the area of P .
Fix n ≥ 1. We partition the interval [0, 1] into n subintervals of equal length n1 . Above each
subinterval we place a rectangle of width n1 . If we choose the right endpoint of each subinterval
to determine the height, we obtain the rectangles shown in Figure 1.1 (left images).

Version: December 30, 2025. 4

 Chapter 1.2

1 1

P P

0 k−1 k 1 0 k−1 k 1
n n n n

1 1

P P

0 k−1 k 1 0 k−1 k 1
n n n n

Figure 1.1: The cases n = 6 and n = 10. In the left
pictures the k-th rectangle has height
k 2 k−1 2
, while in the right pictures it has height .


n n

k 2
In this construction, the k-th rectangle has height n . Hence, by additivity of the area

and Lemma 1.2, their total area gives an upper sum:

1 12 22 n2
 
A≤ + + · · · +
n n2 n2 n2
1
= 3 (12 + 22 + · · · + n2 )
n 
1 n3 n2 n

= 3 + +
n 3 2 6
1 1 1
= + + .
3 2n 6n2

(The line segments where the rectangles meet have zero area and can be ignored.)
Similarly, if instead we choose the left endpoint of each subinterval, we obtain the rectangles

2
shown in Figure 1.1 (right images). In this case, the k-th rectangle has height k−1 n , and

Version: December 30, 2025. 5

 Chapter 1.2

their total area gives a lower sum:

12 (n − 1)2
 
1 0
A≥ + + · · · +
n n2 n2 n2
1 2 2 2


= 1 + 2 + · · · + (n − 1)
n3 
1 (n − 1)3 (n − 1)2 n − 1

= + +
n3 3 2 6
 3
n2 n

1 n
= + + − n2
n3 3 2 6
1 1 1
= − + .
3 2n 6n2

In summary, for every n ≥ 1,

1 1 1 1 1
− + ≤ A− ≤ + . (1.5)
2n 6n2 3 2n 6n2

The only real number A − 1

3 that satisfies these inequalities for all n is 0. Hence, A = 31 .

1 Remark 1.3. — In the language of Riemann integration (that we will discuss in Chapter 6),
we have shown that the lower sums converge to 13 from below, while the upper sums converge
to 31 from above; therefore, A = 31 .

To make the last step rigorous, one must prove that the only real number satisfying (1.5)
for all n ≥ 1 is 0. Intuitively, this is clear: as n grows, both bounds 2n
1
+ 6n1 2 and − 2n
1
+ 6n1 2
shrink toward 0, leaving no other possible value. At this stage, however, we cannot yet give
a fully rigorous proof, because we have not defined the real numbers precisely.
Moreover, we have not yet settled the question of which subsets of R2 admit an area, or
what exactly we mean by “area”. Here we have worked under the implicit assumption that such
a notion exists for sets like P . These foundational questions will be addressed in later classes,
with the introduction of the Riemann and Lebesgue integrals and the concept of measurable
sets.

1.2 Naive Set Theory

Note that in the previous examples we informally used the notion of a “set.” For completeness,
we now give a more precise (though still informal, or naive) description.

Version: December 30, 2025. 6

 Chapter 1.2

Naive Set Theory

The central assumptions of naive set theory are the following postulates:

(1) A set is a collection of distinct objects, called its elements.

(2) A set is completely determined by its elements (axiom of extensionality).

(3) If A(x) is any property of elements of a set X, then

{x ∈ X | A(x)}

denotes the set of all elements x in X for which A(x) holds (set-builder notation).

The empty set, written ∅ (or sometimes {}), is the set containing no elements.

In addition, Zermelo–Fraenkel set theory assumes the axiom of regularity, which implies
that no set is an element of itself.

1 We write x ∈ X if x is an element of the set X, and x ∈

/ X otherwise.
A set can sometimes be described by listing its elements explicitly, for example

X = {x1 , x2 , . . . , xn }.

More often, we describe sets using a property that characterizes their elements. According to
postulate (3), this is done using the so-called set-builder notation. For instance, both

{n ∈ Z | n is even} and {n ∈ Z | ∃ m ∈ Z : n = 2m}

denote the set of even integers.

Here, the symbol “∃” means “there exists”, while “∀” means “for all”. The symbols “|” and
“:” both mean “such that”.

For convenience, we summarize below some of the logical symbols that will be used through-
out these notes. With time and practice, this mathematical language will become familiar.

Version: December 30, 2025. 7

 Chapter 1.2

Logical Symbols
• ∀ : “for all” or “for every”.

• ∃ : “there exists”.

• ⇒ : “implies” or “if . . . then . . . ”.

• ⇐ : “is implied by”.

1 • ⇔ : “if and only if” (logical equivalence).

• ⇏ : “does not imply”.

For example,

x > 2 ⇒ x2 > 4, or equivalently x2 > 4 ⇐ x > 2;

x2 > 4 ⇏ x > 2, x ∈ Q ⇒ x ∈ R, x = 3 ⇔ 3x = 9.

Version: December 30, 2025. 8

Chapter 1.3 Tips on Studying

1.3 Tips on Studying

All beginnings are difficult. Starting university mathematics is not easy. The subject has
its own language and way of thinking, and you will need some time to become fluent in it.
The earlier you begin engaging actively with this new way of working, the more rewarding the
lectures will become.

Mathematics is learned by doing. You cannot learn mathematics by watching others, just
as you cannot learn to ski or play tennis by watching competitions on television. You have to
practice. Treat mathematics like a language: use it, speak it, and write it.

Exercises are the best teachers. The most effective way to learn is through solving problems.
Work on as many exercises as you can, try to explain the solutions to yourself and to others,
and attempt variations until you feel comfortable. Discussing the material with classmates is
also extremely helpful: explaining a proof or a solution often reveals whether you have really
understood it.

Collaboration vs. independence. Working in small groups is highly recommended. Discussion

makes abstract ideas more concrete and easier to grasp. But remember: in the end you must
be able to solve problems on your own. Always revisit exercises without help and check that
you can reproduce the reasoning independently.

Asking questions is a strength. Never hesitate to ask questions. Many of your classmates
will have the same doubts, and raising them helps everyone. It also provides valuable feedback
to lecturers and assistants about which points need further explanation. Learning to formulate
precise questions is itself an important skill, and the first year is the perfect time to practice
it.

Version: December 30, 2025. 9

 Chapter 2

The Real Numbers: Maximum,

Supremum, and Sequences

2.1 The Axioms of the Real Numbers

2.1.1 Ordered Fields

We all have an intuitive feeling for what real numbers are, but our goal is to introduce them
carefully and precisely. To get there, we first need a few building blocks.
2
Groups, Rings, and Fields

We begin with the notion of group. Loosely speaking, a group is a set equipped with an
“operation” that satisfies a list of properties. For our purposes, it is enough to know that a
operation is something that takes two elements of a set and gives back a third element of the
set. We now specify the properties that the operation must satisfy so that we can speak of a
group.

10
 Chapter 2.1

Definition 2.1: Groups

A group is a non-empty set G together with a rule (called an operation and usually
written as ⋆) that combines any two elements of G into another element of G. This
operation must satisfy three conditions:

• Associativity: No matter how you place parentheses, the result is the same: for
all a, b, c ∈ G,
(a ⋆ b) ⋆ c = a ⋆ (b ⋆ c).

• Neutral element: There is a special element e ∈ G such that combining it with

any a ∈ G leaves a unchanged: for any a ∈ G,

a ⋆ e = e ⋆ a = a.

• Inverse element: Every a ∈ G has a “partner” a−1 ∈ G that “cancels it out,”

giving the neutral element: ror any a ∈ G there exists a−1 ∈ G such that

a ⋆ a−1 = a−1 ⋆ a = e.

Note that, in general, one does not require that a⋆b = b⋆a. If the order of the operation
2 does not matter, i.e. a ⋆ b = b ⋆ a for all a, b ∈ G, the group is called commutative or
abelian.

Example 2.2. — Let us check some familiar sets and operations.

1. Natural numbers with addition. Take N = {0, 1, 2, 3, . . . } with the addition +.

• Addition is associative: (k + l) + m = k + (l + m) for all k, l, m ∈ N.

• The number 0 is neutral: 0 + n = n + 0 = n.
• However, no natural number except 0 has an inverse in N. For example, the inverse
of 3 is −3, which is not a natural number.

So (N, +) is not a group.

2. Integers with addition. Now take Z = {. . . , −2, −1, 0, 1, 2, . . . } with addition +.

• Addition is associative.
• 0 is the neutral element.
• Every integer has an inverse: the inverse of n is −n.

Thus (Z, +) is a group. It is also commutative since n + m = m + n.

Version: December 30, 2025. 11

 Chapter 2.1

3. Nonzero rational numbers with multiplication. Consider

n o
Q∗ = p
q p, q ∈ Z, p, q ̸= 0

with the multiplication ·. One can easily check that the multiplication is associative and
commutative, the neutral element is 1, and the inverse of pq is pq . Also, pq · rs = rs · pq .
Hence (Q∗ , ·) is a commutative group.

4. Permutations of three objects. Imagine we have three objects {1, 2, 3}. A permutation
is just a way of rearranging them. There are 6 possible rearrangements and together
they form the set S3 , whose elements are the following:

id (do nothing)
(1 2) (swap 1 and 2, leave 3)
(1 3) (swap 1 and 3, leave 2)
(2 3) (swap 2 and 3, leave 1)
(1 2 3) (send 1 7→ 2, 2 7→ 3, 3 7→ 1)
(1 3 2) (send 1 7→ 3, 3 7→ 2, 2 7→ 1).

2 The operation in this group is composition: do one rearrangement after another. More
explicitly, if σ and τ are two permutations, then σ ◦ τ means “first apply τ , then apply
σ”. With this operation, one can check all the properties:

• Associativity holds because if you do three rearrangements in a row, it does not

matter how you group them.
• The identity id is the neutral element.
• Every permutation can be undone, so each element has an inverse. For example,
the inverse of (1 2 3) is (1 3 2), since doing one after the other brings everything
back to the start.

Therefore (S3 , ◦) is a group, but in this case it is not commutative. For instance, let
σ = (1 2) and τ = (2 3).
- If we compute σ ◦ τ : τ sends 1 7→ 1, 2 7→ 3, 3 7→ 2. Then σ swaps 1 and 2. So overall
we get 1 7→ 2, 2 7→ 3, 3 7→ 1, which is the permutation (1 2 3).
- If we compute τ ◦ σ: σ swaps 1 and 2. Then τ swaps 2 and 3. So overall we get
1 7→ 3, 3 7→ 2, 2 7→ 1, which is the permutation (1 3 2).
Since (1 2 3) ̸= (1 3 2), we conclude that σ ◦ τ ̸= τ ◦ σ. Hence S3 is not commutative.

A couple of useful facts follow from the definition of group.

Version: December 30, 2025. 12

 Chapter 2.1

Lemma 2.3: Basic properties of groups

Let G be a group. Then:

1. The neutral element is unique.

2. The inverse of an element is unique.

3. The inverse of the inverse of an element is the element itself, namely (a−1 )−1 = a
for all a ∈ G.

Proof. (i) Assume that, in addition to e ∈ G, we have a second element e′ with the property
that e′ ⋆ a = a ⋆ e′ = a for all elements a ∈ G. Then, we can choose a = e to obtain

e ⋆ e′ = e.

Similarly, since e is a neutral element,

e ⋆ e′ = e′ .

Combining the two identities, we get

2 e = e ⋆ e′ = e′ .

This proves that e = e′ , so we can speak of the neutral element of a group.

(ii) Assume that for an element a ∈ G, there exist two elements b, c ∈ G that are both
inverse of a, namely
a ⋆ b = b ⋆ a = e, a ⋆ c = c ⋆ a = e.

Then, using associativity, we observe that

b = b ⋆ e = b ⋆ (a ⋆ c) = (b ⋆ a) ⋆ c = e ⋆ c = c.

This proves that the inverse element of a is unique, so we can speak of the inverse element,
and the notation a−1 makes sense.
(iii) Since a ⋆ a−1 = e, we deduce that a is the inverse of a−1 , thus

(a−1 )−1 = a. (2.1)

Groups capture the idea of combining elements with a single operation. But to describe
the arithmetic of numbers more faithfully, we also need a second operation (as we do with
addition and multiplication). This leads us to the notions of rings and fields.

Version: December 30, 2025. 13

 Chapter 2.1

Definition 2.4: Rings and Fields

A ring is a non-empty set R in which we can both “add” and “multiply” elements with
two operations “+” and “·”. Also, these two operations are compatible with each other.
More precisely:

• (R, +) is a commutative group, with neutral element denoted by 0.

• Multiplication · is associative, has a neutral element (usually written as 1),

and distributes over addition:

a · (b + c) = a · b + a · c, (b + c) · a = b · a + c · a for all a, b, c ∈ R.

If multiplication is also commutative, we call R a commutative ring.

Note that, unlike addition, we do not require that every element has an inverse for
multiplication.
A field is a special kind of commutative ring: every nonzero element has an inverse
for the multiplication. In other words, if R is a commutative ring, then R is a field if
R \ {0} forms a commutative group under multiplication.
Traditionally, we use the letter F to denote a field. We also write F ∗ = F \ {0} for the
set of all invertible elements of F .
2

Example 2.5. — Let us test these ideas on examples.

1. The integers Z. We already know that (Z, +) is a commutative group. To see if it is a

ring with the usual multiplication, we must check:

• Associativity of the multiplication: For all integers k, l, m ∈ Z, we have

(k · l) · m = k · (l · m).

• Neutral element for the multiplication: The neutral element for the multiplication
is 1 ∈ Z as, for all integers k ∈ Z, we have

1 · k = k · 1 = k.

• Distributivity: For all k, l, m ∈ Z we have

k · (l + m) = k · l + k · m

and
(k + l) · m = k · m + l · m.

Version: December 30, 2025. 14

 Chapter 2.1

Thus Z is a ring. Moreover, since the multiplication is commutative (namely, k ·l = l ·k),

it is a commutative ring.
However, Z is not a field: most integers do not have a multiplicative inverse inside Z.
For example, the inverse of 2 would be 12 , which is not an integer.

2. The rational numbers Q. With the usual addition and multiplication, Q is a commutative
ring. Moreover, every nonzero rational number pq has a multiplicative inverse pq , which
is also rational. Hence Q is a field.

So, in short: rings generalize the integers, while fields generalize the rationals.

Example 2.6. — A non-commutative ring. We consider the set M2 (Z) of all 2 × 2 tables
of integers (called “matrices”) usually written as
!
a b
.
c d

Addition and multiplication are defined as follows:

• Addition is done entry by entry:

2 ! ! !
a b a′ b′ a + a′ b + b′
+ = .
c d c′ d′ c + c′ d + d′

• Multiplication is slightly more involved:

! ! !
a b a′ b′ aa′ + bc′ ab′ + bd′
· = .
c d c′ d′ ca′ + dc′ cb′ + dd′

!
0 0
This structure satisfies the axioms of a ring with neutral elements for addition and
0 0
!
1 0
for multiplication, but multiplication is not commutative. Let us check explicitly
0 1
with ! !
0 1 0 0
A= , B= .
0 0 1 0
First compute AB:
! ! ! !
0 1 0 0 0·0+1·1 0·0+1·0 1 0
AB = · = = .
0 0 1 0 0·0+0·1 0·0+0·0 0 0

Version: December 30, 2025. 15

 Chapter 2.1

Now compute BA:

! ! ! !
0 0 0 1 0·0+0·0 0·1+0·0 0 0
BA = · = = .
1 0 0 0 1·0+0·0 1·1+0·0 0 1

We see that ! !
1 0 0 0
AB = ̸= = BA.
0 0 0 1

Thus M2 (Z) is a ring in which multiplication is not commutative.

Just as with groups, some useful facts follow immediately:

• The additive inverse of a is written −a, and the multiplicative inverse of a ̸= 0 is written
a−1 .

• In the current context, (2.1) applied to both addition and multiplication implies that

−(−a) = a, (a−1 )−1 = a whenever a ̸= 0. (2.2)

Also, we have the following

Lemma 2.7: Basic properties of fields

Let F be a field and let a, b ∈ F . Then:
2
1. 0 · a = a · 0 = 0.

2. a · (−b) = −(a · b) = (−a) · b. In particular, (−1) · a = −a.

3. (−a) · (−b) = a · b. In particular, (−a)−1 = −(a−1 ) whenever a ̸= 0.

Proof. (i) Since 0 is the neutral element for the addition, we have 0 = 0 + 0. Hence, using
distributivity, we get
0 · a = (0 + 0) · a = (0 · a) + (0 · a).

Adding −0 · a (i.e., the inverse of 0 · a for the addition), we deduce that 0 · a = 0. The case of
a · 0 is analogous.

(ii) By the distributive law,

a · b + a · (−b) = a · (b + (−b)) = a · 0 = 0.

So a·(−b) is the additive inverse of a·b, i.e., −(a·b) = a·(−b). Taking b = 1 gives −a = (−1)·a.
The validity of (−a) · b = −(a · b) follows exchanging a and b in the argument above.
(iii) By (ii) we know that −(a · b) = a · (−b). Hence, recalling (2.2),

a · b = −(a · (−b)).

Version: December 30, 2025. 16

 Chapter 2.1

On the other hand, applying (ii) with (−b) instead of b, we also have

−(a · (−b)) = (−a) · (−b).

Combining the two identities above, we conclude that (−a) · (−b) = a · b. Finally, taking
b = a−1 yields (−a) · (−(a−1 )) = a · a−1 = 1, which gives the second assertion.

Remark 2.8. — A natural question one may ask is the following: Can 0 (the additive
neutral element) and 1 (the multiplicative neutral element) be the same? If 0 = 1, then for
any a ∈ F , we would have
a = a · 1 = a · 0 = 0.

So the field would collapse to just one element, namely {0}. From now on, we will always
assume that a field has at least two distinct elements, therefore 0 ̸= 1.

Order relation

Next, we introduce the second ingredient of an ordered field: the order relation. To get there,
we first need a few basic notions.

Cartesian Product
2
Let X and Y be two sets. The cartesian product X × Y is the set of ordered pairs
of elements of X and Y :

X × Y = {(x, y) | x ∈ X, y ∈ Y }.

Example 2.9. — The cartesian product X × Y of X = {A, B, C, D, E, F, G, H} and Y =

{1, 2, 3, 4, 5, 6, 7, 8} is what we use to describe positions on a chessboard. Each pair corresponds
to a unique square. For instance, the black king starts the game on the square (E, 8) ∈ X ×Y .

Subsets
Let P and Q be sets.

• P is a subset of Q, written P ⊂ Q (or P ⊆ Q), if every element of P also belongs

to Q.

• P is a proper subset of Q, written P ⊊ Q, if P is a subset of Q but P ̸= Q.

• We write P ̸⊂ Q (or P ̸⊆ Q) if P is not a subset of Q.

Equivalent ways of saying “P is a subset of Q” are “P is contained in Q” or “Q is a superset

of P ,” written Q ⊃ P . Likewise, “Q is a proper superset of P ” is written Q ⊋ P .

Version: December 30, 2025. 17

 Chapter 2.1

Because of the axioms of naive set theory, two sets P and Q are equal exactly when both
P ⊂ Q and Q ⊂ P hold. For example, {x, y} = {z} holds if x = y = z. Notice that
multiplicities do not matter: {x, x, x} = {x}.

Definition 2.10: Relations

Let X be a set. A relation on X is a subset R ⊂ X ×X, that is, a collection of ordered
pairs of elements of X. If (x, y) ∈ R we write xRy. Common symbols for relations
include <, ≤, ∼, ≡, ∼
=.
If ∼ is a relation on X, we write x ̸∼ y if x ∼ y does not hold. A relation ∼ may have
the following properties:

1. Reflexive: x ∼ x for all x ∈ X.

2. Transitive: if x ∼ y and y ∼ z, then x ∼ z.

3. Symmetric: if x ∼ y, then y ∼ x.

4. Antisymmetric: if x ∼ y and y ∼ x, then x = y.

A relation is an equivalence relation if it is reflexive, transitive, and symmetric. It

is an order relation if it is reflexive, transitive, and antisymmetric.

2 Example 2.11. — On the set of integers Z, we consider three different relations.

• The usual relation ≤ (“less than or equal to”). This relation satisfies the following
properties:

1. Reflexive: n ≤ n for all n.

2. Transitive: if n ≤ m and m ≤ p, then n ≤ p.
3. Not symmetric: e.g. 7 ≤ 8 but 8 ̸≤ 7.
4. Antisymmetric: if n ≤ m and m ≤ n, then n = m.

So ≤ is an order relation.

• The relation < (“strictly smaller than”).

1. Not reflexive: no n satisfies n < n.

2. Transitive: if n < m and m < p, then n < p.
3. Not symmetric: e.g. 3 < 5 but 5 ̸< 3.
4. Antisymmetric (this is subtle): there are no integers n, m ∈ Z that satisfy both
n < m and m < n. Hence, the condition of antisymmetry holds because there is
nothing to check.

So < is neither an equivalence relation nor an order relation.

Version: December 30, 2025. 18

 Chapter 2.1

• On Z, define m ≡ n (mod 3) if m − n is divisible by 3.

1. Reflexive: for any m, m − m = 0 is divisible by 3, so m ≡ m (mod 3).

2. Symmetric: if m ≡ n (mod 3), then m − n is divisible by 3, hence also n − m =
−(m − n) is divisible by 3, so n ≡ m (mod 3).
2
3. Transitive: if m ≡ n (mod 3) and n ≡ p (mod 3), then both m − n and n − p are
divisible by 3, hence so is (m − n) + (n − p) = m − p, so m ≡ p (mod 3).

Therefore ≡ is an equivalence relation (this relation is called “congruence modulo 3” as

is usually written as “a ≡ b (mod 3)”).

Ordered Fields

We now introduce the concept of an ordered field. From now on, we use the arrow “ =⇒ ” to
denote implication: A =⇒ B means “A implies B”.

Definition 2.12: Ordered Field

Let F be a field, and let ≤ be an order relation on F . We call (F, ≤), or simply F , an
ordered field if the following hold:

1. Linearity of order: for all x, y ∈ F , at least one of x ≤ y or y ≤ x holds.

2. Compatibility with addition: for all x, y, z ∈ F ,

x ≤ y =⇒ x + z ≤ y + z.

3 3. Compatibility with multiplication: for all x, y ∈ F ,

0 ≤ x and 0 ≤ y =⇒ 0 ≤ x · y.

The following terminology is standard and will be used throughout:

• x ≤ y is read as “x is less than or equal to y.”

• y ≥ x means x ≤ y (“y is greater than or equal to x”).

• x < y means x ≤ y and x ̸= y (“x is strictly smaller than y”).

• x > y means y < x (“x is strictly greater than y”).

• An element x ∈ F is non-negative if x ≥ 0, and non-positive if x ≤ 0.

• An element x ∈ F is positive if x > 0, and negative if x < 0.

Version: December 30, 2025. 19

 Chapter 2.1

We often chain inequalities, e.g.

x≤y<z=a

stands for “x ≤ y, and y < z, and z = a.”

Example 2.13. — A standard example of an ordered field is the field of rational numbers
′
Q with the usual order. For two fractions pq and pq′ (with p, p′ ∈ Z and q, q ′ ∈ N), we define

3 p p′
≤ ′ if pq ′ ≤ p′ q,
q q

where the inequality on the right is the usual order on the integers. It is easy to check that
this order is compatible with addition and multiplication, so (Q, ≤) is indeed an ordered field.

Given (F, ≤) an ordered field, we want to prove a series of properties that follow from the
definitions. To simplify the notation, it is customary to write · for multiplication only if it
would otherwise be confusing. This is why, in proofs, · may disappear. For example, we may
write xy instead of x · y.

Version: December 30, 2025. 20

 Chapter 2.1

Lemma 2.14: Ordered field: basic consequences

Let (F, ≤) be an ordered field, and let x, y, z, w ∈ F . Then:

(a) (Trichotomy) Either x < y, or x = y, or x > y.

(b) If x < y and y ≤ z, then x < z. (Analogously, x ≤ y and y < z imply x < z.)

(c) (Addition of inequalities) If x ≤ y and z ≤ w, then x + z ≤ y + w. (Analogously,

x < y and z ≤ w imply x + z < y + w.)

(d) x ≤ y if and only if 0 ≤ y − x.

(e) x ≤ 0 if and only if 0 ≤ −x.

(f) x2 ≥ 0, and x2 > 0 if x ̸= 0.

(g) 0 < 1.

(h) If 0 ≤ x and y ≤ z, then xy ≤ xz.

(i) If x ≤ 0 and y ≤ z, then xy ≥ xz.

(j) If 0 < x ≤ y, then 0 < y −1 ≤ x−1 .

3 (k) If 0 ≤ x ≤ y and 0 ≤ z ≤ w, then 0 ≤ xz ≤ yw.

(l) If x + y ≤ x + z, then y ≤ z.

(m) If xy ≤ xz and x > 0, then y ≤ z.

Proof. (a) By linearity of the order, for any x, y at least one of x ≤ y or y ≤ x holds. If x ≤ y
and y ≤ x, antisymmetry gives x = y. Otherwise, exactly one strict inequality holds, giving
x < y or y < x.
(b) From x < y we have in particular x ≤ y. Since y ≤ z, transitivity yields x ≤ z. So, to
prove x < z, we need to exclude that x = z.
To see that, assume by contradiction that x = z. Then y ≤ z yields y ≤ x, which contradicts
x < y. Hence x < z.
The variant in parentheses is analogous.
(c) From x ≤ y, compatibility with addition (Definition 2.12) gives x + z ≤ y + z. From
z ≤ w we get y + z ≤ y + w. Transitivity yields x + z ≤ y + w.
For the strict variant, use (b).
(d) If x ≤ y, add −x to both sides to obtain 0 ≤ y − x. Conversely, if 0 ≤ y − x, add x to
get x ≤ y.
(e) Apply (d) with y = 0.

Version: December 30, 2025. 21

 Chapter 2.1

(f ) If x ≥ 0, then x2 ≥ 0 by multiplicative compatibility in Definition 2.12. If x ≤ 0, then

−x ≥ 0 by (e), hence x2 = (−x)2 ≥ 0; here x2 = (−x)2 follows from Lemma 2.7 (iii).
For strict positivity: if x ̸= 0 and x2 = 0, multiply by x−1 and use Lemma 2.7(i) to get
0 = 0 · x−1 = x2 x−1 = x · x · x−1 = x, a contradiction.
(g) By (f), 1 = 12 ≥ 0, and by Remark 2.8, 1 ̸= 0. Hence 0 < 1.
(h) From y ≤ z and (d), we have 0 ≤ z − y. If also 0 ≤ x, multiplicative compatibility
gives 0 ≤ x(z − y) = xz − xy. Apply again (d) to conclude that xy ≤ xz.
(i) From x ≤ 0 and (e) we get −x ≥ 0, and from y ≤ z and (d) we get 0 ≤ z − y. Then
0 ≤ (−x)(z − y) = xy − xz, so by (d) we have xy ≥ xz.
(j) We first assert that x−1 > 0: if by contradiction x−1 ≤ 0, then (h) with x ≥ 0 and
x−1 ≤ 0 would give 1 = xx−1 ≤ 0, contradicting (g).
Similarly, y −1 > 0.
3 Now, by (h), x−1 ≥ 0 and y −1 > 0 implies that x−1 y −1 ≥ 0. Combining this with x ≤ y and
(h), we get
y −1 = x x−1 y −1 ≤ y x−1 y −1 = x−1

as desired.
(k)–(m) These follow by combining (h)–(i) with (c) and (d). Detailed arguments are
assigned to Exercise 2.15.

Exercise 2.15. — Prove inferences (k)–(m) in Lemma 2.14.

What happens in (m) if the assumption x > 0 is replaced with x < 0? For each of these
inferences, also try to formulate and prove analogous statements for the strict relation “<”.

Now that we have a notion of an ordered field, we can construct the integers and rational
numbers.

Version: December 30, 2025. 22

 Chapter 2.1

Lemma 2.16: Integers and rationals inside an ordered field

Let (F, ≤) be an ordered field, and denote by 0 and 1 the neutral elements for addition
and multiplication, respectively. Then:

(i) The elements . . . , −2, −1, 0, 1, 2, . . . defined by

2 = 1 + 1, 3 = 2 + 1, . . . , −n = (−1) · n

are all distinct and satisfy

· · · < −2 < −1 < 0 < 1 < 2 < 3 < · · · .

We denote this set of elements by Z, and we call them “integers”.

(ii) Every fraction pq −1 with p, q ∈ Z, q ̸= 0, lies in F and the set of all such elements
is denoted by Q. Also,
3
Z ⊊ Q ⊆ F.

Proof. (i) By Lemma 2.14(g), we have 0 < 1. Then Lemma 2.14(c) yields 0 < 1 < 2 < 3 < · · · ,
and taking negatives gives · · · < −2 < −1 < 0. Hence all these elements are distinct.
(ii) For q ̸= 0, q is invertible in F ; define pq = pq −1 . The set of such fractions is a field
contained in F , which we denote by Q.
To show that Q strictly contains Z, consider 12 (the inverse of 2). Since 2 > 1, it follows
from Lemma 2.14(j) that 0 < 12 < 1, so 12 ̸∈ Z.

Exercise 2.17. — Show that

3
+ 4 ≥ 0 = {x ∈ R \ {0} | −3 ≤ x ≤ −1 or x > 0} .

x ∈ R \ {0} | x + x

(x+3)(x+1)
Hint: note that x + 3
x +4= x .

Version: December 30, 2025. 23

 Chapter 2.1

Functions
A function f from a set X to a set Y is an assignment of an element of Y to each
element of X. The element y ∈ Y to which x ∈ X is assigned is denoted f (x). We
write f : X → Y for a function from X to Y and sometimes also speak of a map,
mapping, or transformation.
The set X is the domain and Y the codomain.
We refer to the set X as domain, and the set Y as domain of values or codomain.
The set
{(x, f (x)) | x ∈ X} ⊂ X × Y

is called the graph of f .

In the context of a function f : X → Y , an element x of the domain of definition is
also called argument, and an element y = f (x) ∈ Y assumed by the function is also
called value of the function.
If f : X → Y is a function, one also writes

f :X → Y
x 7→ f (x),

3
where f (x) could be a concrete formula. We pronounce “7→” as “is mapped to”.
Two functions f1 : X1 → Y1 and f2 : X2 → Y2 are said to be equal if X1 = X2 , Y1 = Y2 ,
and f1 (x) = f2 (x) for all x ∈ X1 .

Definition 2.18: Absolute value and sign

Let (F, ≤) be an ordered field.

• The absolute value (or modulus) is the function | · | : F → F defined by


x, x ≥ 0,
|x| =
−x, x < 0.

• The sign is the function sgn : F → {−1, 0, 1} defined by




−1, x < 0,

sgn(x) = 0, x = 0,


1,

x > 0.

Version: December 30, 2025. 24

 Chapter 2.1

Lemma 2.19: Absolute value and sign: basic properties

Let (F, ≤) be an ordered field and let x, y ∈ F . Then:

(a) x = sgn(x) |x|, | − x| = |x|, sgn(−x) = − sgn(x).

(b) |x| ≥ 0, and |x| = 0 if and only if x = 0 (by trichotomy, Lemma 2.14 a).

(c) (Multiplicativity) sgn(xy) = sgn(x) sgn(y) and |xy| = |x| |y|.

(d) If x ̸= 0, then |x−1 | = |x|−1 .

(e) |x| ≤ y iff −y ≤ x ≤ y.

(f ) |x| < y iff −y < x < y.

(g) (Triangle inequality) |x + y| ≤ |x| + |y|.

(h) (Inverse triangle inequality) |x| − |y| ≤ |x − y|.

Proof. (a) This follows directly from the definition.

(b) This follows from the definition and Lemma 2.14(a).
(c) Check the four sign cases (depending on whether x, y are positive or negative).
(d) This follows from (c), because |x−1 ||x| = |1| = 1.
3 (e) We consider the case x ≥ 0, so that |x| = x (the case x ≤ 0 is similar).
If |x| ≤ y, since x = |x| ≥ 0 we get

−y ≤ 0 ≤ x ≤ y.

Conversely, if −y ≤ x ≤ y, since |x| = x we get |x| ≤ y.

(f ) This is proved similarly to (e), with strict inequalities.
(g) Thanks to (e) we have −|x| ≤ x ≤ |x| and −|y| ≤ y ≤ |y|. Adding these two inequalities,
we get
−(|x| + |y|) ≤ x + y ≤ |x| + |y|.

Applying (e) again, we conclude.

(h) From (g) we have |x| ≤ |x − y| + |y|, therefore

|x| − |y| ≤ |x − y|.

Exchanging the roles of x and y we also have |y| − |x| ≤ |y − x| = |x − y|, Combining these
two inequalities yields
−|x − y| ≤ |x| − |y| ≤ |x − y|,

and the result follows by applying (e).

Version: December 30, 2025. 25

 Chapter 2.1

Exercise 2.20. — For which x, y ∈ R does equality hold in the triangle inequality? And
3
in the inverse triangle inequality?

2.1.2 Completeness Axiom

An arbitrary ordered field is not enough for calculus: it may have “gaps”. The basic example
is the ordered field of rationals Q. Consider

X = {q ∈ Q : either q < 0, or q ≥ 0 and q 2 < 2}, Y = {q ∈ Q : q ≥ 0 and q 2 > 2}.

Then X and Y are non-empty, and every x ≤ y for every x ∈ X and y ∈ Y . However, there
is no rational number c with x ≤ c ≤ y for all such x, y (since there is no c ∈ Q with c2 = 2).
This “gap” shows that Q fails the completeness property below. To rule out such gaps we add
the completeness axiom. The need for such an axiom was already felt by the ancient Greeks
(Pythagoras, Euclid, Archimedes), but a precise formalisation came only in the 19th century
through the work of Weierstrass, Heine, Cantor, Dedekind, and others.

Definition 2.21: Completeness Axiom

Let (K, ≤) be an ordered field. We say that (K, ≤) is complete (or a completely
ordered field) if the following statement holds:

Let X, Y be non-empty subsets of K such that x ≤ y for all x ∈ X and y ∈ Y .

4 Then there exists c ∈ K lying between X and Y , in the sense that x ≤ c ≤ y for
all x ∈ X and y ∈ Y .

The statement above is called the completeness axiom.

Definition 2.22: Real Numbers

We call the field of real numbers any completely ordered field and denote it by R.

We will often visualise the real numbers as points on a straight line, also called the number
line.

−1 0 1 2 3 x y

Given x, y ∈ R, we interpret the relation x < y as “on the straight line, the point y lies to
the right of the point x”. With this representation of the real numbers, the following figure
represents the completeness axiom.

X c Y

Version: December 30, 2025. 26

 Chapter 2.1

Figure 2.1: Given X, Y ⊂ R with x ≤ y for all x ∈ X and y ∈ Y , there exists a number c in
between.

We note that while the number line is a helpful aid to build intuition, it should not replace
rigorous proofs.

Injective, surjective and bijective functions

Let f : X → Y be a function. We call f :

1. injective (or an injection) if x1 ̸= x2 =⇒ f (x1 ) ̸= f (x2 ) for all x1 , x2 ∈ X;

2. surjective (or a surjection) if for every y ∈ Y there exists x ∈ X with f (x) = y;

3. bijective (or a bijection) if it is both injective and surjective.

Thus, a function f : X → Y is not injective if there exist distinct x1 ̸= x2 ∈ X with

f (x1 ) = f (x2 ), and not surjective if there exists y ∈ Y such that f (x) ̸= y for all x ∈ X.

X Y X Y
f
f

Figure 2.2: Left: injective but not surjective. Right: surjective but not injective.

Image and preimage of a function

For f : X → Y and A ⊂ X, define the image of A under the function f as

f (A) = { y ∈ Y | ∃ x ∈ A with f (x) = y }.

For B ⊂ Y define the preimage of B under the function f as

f −1 (B) = { x ∈ X | f (x) ∈ B }.

Remark 2.23. — Saying that f : X → Y is surjective is equivalent to f (X) = Y . Equiva-

lently, f is surjective if f −1 ({y}) ̸= ∅ for every y ∈ Y .

Example 2.24. — Let f : R → R be the constant function x 7→ 0. Then f −1 ({0}) = R,

while f −1 ({y}) = ∅ for every y ̸= 0.

Version: December 30, 2025. 27

 Chapter 2.1

Example 2.25. — Let X, Y be finite sets with the same number of elements. Then for any
f : X → Y , injectivity and surjectivity are equivalent.
Proof. Suppose X and Y both have n elements, and list X = {x1 , . . . , xn }. If f is
injective, the n values f (xi ) are all distinct, so f (X) = {f (x1 ), . . . , f (xn )} has n elements.
Since f (X) ⊂ Y and |Y | = n, we must have f (X) = Y , i.e. f is surjective.
Conversely, to show that surjectivity implies injectivity, we prove that if f is not injective
then f is not surjective. So, assume there exist xi ̸= xj with f (xi ) = f (xj ). Then f (X) has
at most n − 1 elements, so it cannot coincide with Y and, therefore, f cannot be surjective.

Remark 2.26. — For infinite sets, injectivity and surjectivity need not be equivalent.
Consider f1 , f2 : N → N defined by

0, n = 0,
f1 (n) = n + 1, f2 (n) =
n − 1, n ≥ 1.

Then f1 is injective but not surjective, while f2 is surjective but not injective.

Exercise 2.27. — Reformulate the definitions of injectivity, surjectivity, and bijectivity

using the notions of image and preimage.
4
We conclude with the square root function on R≥0 = {x ∈ R : x ≥ 0} as an application
of completeness.

√
Exercise 2.28. — Show the existence and uniqueness of a bijective function · : R≥0 →
√
R≥0 with ( a)2 = a for all a ∈ R≥0 .

1. (Existence) For a ∈ R≥0 set

X = {x ∈ R≥0 | x2 ≤ a}, Y = {y ∈ R≥0 | y 2 ≥ a}.

By completeness, there exists c ∈ R with x ≤ c ≤ y for all x ∈ X, y ∈ Y . Show that

c ∈ X ∩ Y , hence c2 = a.
Hint: If by contradiction c ∈
/ X (i.e. c2 > a), choose small ε > 0 so that (c − ε)2 ≥ a.
2 −a
For instance, one can choose ε = c 2c and note that, with this choice,

c2 − a
(c − ε)2 = c2 − 2cε + ε2 > c2 − 2cε = c2 − 2c · = a.
2c

Hence c − ε ∈ Y , contradicting y ≥ c for all y ∈ Y .

The case c ∈
/ Y is analogous.

Version: December 30, 2025. 28

 Chapter 2.1

2. (Uniqueness) Prove that for every a ∈ R≥0 there is at most one c ∈ R≥0 with c2 = a.
Hint: First, prove that for all x, y ∈ R≥0 , the statements x < y and x2 < y 2 are
equivalent. Then use this fact to deduce the uniqueness of c.
√ √
Define · : R≥0 → R≥0 by a = c where c is as above. Show that:
√ √ √
3. · is increasing: if 0 ≤ x < y then x< y;
√
4. · is bijective;
√ √ √
5. xy = x y for all x, y ∈ R≥0 .

√
Exercise 2.29. — For all x ∈ R, show that x2 = |x|2 and x2 = |x|.

Working rules. In a field of real numbers as in Definition 2.22, the usual arithmetic rules
4
and equation manipulations hold (as always, division by 0 is undefined). The order relations
≤ and < satisfy the familiar laws for inequalities; in particular, multiplying an inequality by
a negative number reverses its direction. We will use these laws freely from now on. The new
ingredient is the completeness axiom, whose power will become clear as we apply it to prove
further results.
Do such numbers actually exist? Why do we say the real numbers? At this point
it is not obvious that a completely ordered field really exists. Moreover, we often speak of the
real numbers as if there were only one such object. In this course we will assume (in line with
your school experience) that a field of real numbers exists and is unique. In particular, you
may safely use R with its usual arithmetic, order rules, and completeness, and you can think
of R as the rational line with all gaps filled.
For those interested, the following section provides a brief and concrete explanation of why
this assumption is reasonable, with complete proofs reserved for later classes.

Extra material: Why the real numbers exist and are unique

In what follows, we sketch (without all details) the construction of R via Dedekind cuts and
the uniqueness argument for complete ordered fields.
Existence. Start from the rationals Q. Many real numbers are not rational, but we can
still “identify them” by stating which rationals lie to their left. This leads to the idea of
Dedekind cuts.
A Dedekind cut is a set C ⊂ Q with the following properties:

• C is nonempty and is not equal to Q;

• C is downward closed : if q ∈ C and r < q, then r ∈ C;

• C has no largest element: for every q ∈ C there exists r ∈ C with q < r.

Version: December 30, 2025. 29

Chapter 2.1

√
Intuitively, C collects all rationals “to the left” of a real number. For example, the cut of 2
is
{ q ∈ Q | either q < 0, or q ≥ 0 and q 2 < 2 }.

Define RDed to be the set of all Dedekind cuts with the following addition and multiplication:

C + D = {q + r | q ∈ C, r ∈ D}, C · D = {q · r | q ∈ C, r ∈ D}.

Also, define the order by inclusion: we say that C ≤ D if C ⊆ D.

With these definitions, one can show that RDed is an ordered field and, crucially, that it satisfies
the completeness axiom (V). In this construction, the rationals can be seen as a subset of RDed
via the “natural” cuts
Q ∋ r 7−→ { q ∈ Q | q < r } ∈ RDed .

Thus, at an intuitive level, real numbers exist: they can be realized as precise “cuts” of the
rational line filling all the “gaps”.
Uniqueness. Suppose K and L are two completely ordered fields, and let 0K , 1K and
0L , 1L denote their neutral elements for addition and multiplication. We identify 0K with 0L
and 1K with 1L . In this way, as a consequence of how the rationals are constructed inside any
ordered field (see Lemma 2.16), the rationals in K and in L are “the same” (i.e. the rational
q in K corresponds to the same q in L). In particular, we may regard a single set of rationals
Q as contained in both K and L.
Now, for x ∈ K, consider the sets

Xx = { q ∈ Q | q < x in K } ⊂ Q, Yx = { q ∈ Q | q > x in K } ⊂ Q.

These are nonempty, and we have q ≤ r for all q ∈ Xx and r ∈ Yx . Since Q ⊂ L, both Xx
and Yx are also contained in L. By completeness of L, there exists c ∈ L with

q≤c≤r for all q ∈ Xx , r ∈ Yx .

One then checks that

{ q ∈ Q | q < c in L } = Xx ,

so c is the unique element of L whose “cut of rationals to its left” is Xx .

This construction allows us to define a map from K to L that assigns to x ∈ K the element
c ∈ L constructed above:
IK,L : K → L, IK,L (x) = c.

One checks that this map preserves addition, multiplication, and order, and fixes every rational
(i.e. IK,L (q) = q for all q ∈ Q).
Repeating the construction with K and L swapped gives a map IL,K : L → K, and one
can check that

IL,K IK,L (x) = x for all x ∈ K, IK,L IL,K (y) = y for all y ∈ L.

Version: December 30, 2025. 30

 Chapter 2.1

Thus IK,L and IL,K are inverses, so IK,L : K → L is a bijection that “identifies” K and L.
In short, once the rationals are fixed inside a completely ordered field, every other element
is uniquely determined by how it “cuts” the rationals. This justifies speaking of the real
numbers and writing R.

2.1.3 Intervals

Definition 2.30: Intervals

Let a, b ∈ R. We define:

• the closed interval

[a, b] = { x ∈ R | a ≤ x ≤ b };

• the open interval

(a, b) = { x ∈ R | a < x < b };

• the half–open intervals

[a, b) = { x ∈ R | a ≤ x < b } and (a, b] = { x ∈ R | a < x ≤ b };

4
• the unbounded closed intervals

[a, ∞) = { x ∈ R | a ≤ x } and (−∞, b] = { x ∈ R | x ≤ b };

• the unbounded open intervals

(a, ∞) = { x ∈ R | a < x } and (−∞, b) = { x ∈ R | x < b }.

The intervals (a, b], [a, b), and (a, b) are non-empty exactly when a < b, while [a, b] is
non-empty exactly when a ≤ b (with [a, a] = {a} a degenerate interval). If an interval is
non-empty and bounded, we call a its left endpoint, b its right endpoint, and b − a its
length. Intervals of the forms [a, b], (a, b], [a, b), (a, b) are also called bounded intervals,
to distinguish them from the unbounded ones.
Some texts write open and half–open intervals using inverted square brackets, e.g. ]a, b[
instead of (a, b). In these notes we always use round brackets for open endpoints.

Version: December 30, 2025. 31

 Chapter 2.1

Set operations
Let P and Q be sets. The intersection P ∩ Q, the union P ∪ Q, the relative
complement P \ Q, and the symmetric difference P △Q are defined by

P ∩ Q = { x | x ∈ P and x ∈ Q },
P ∪ Q = { x | x ∈ P or x ∈ Q },
P \ Q = { x | x ∈ P and x ∈
/ Q },
P △Q = (P \ Q) ∪ (Q \ P ) = (P ∪ Q) \ (P ∩ Q).

These definitions are illustrated in the following pictures. Sketches of this kind are called
Venn diagrams.

P Q P Q

P ∩Q P \Q

5
P Q P Q

P ∪Q P △Q

If it is clear from the context that all sets under consideration are subsets of a given ambient
set X, then the complement of P in X is denoted by P c and defined as P c = X \ P.

Pc X
P

Figure 2.3: The complement P c = X \ P of P in X.

Version: December 30, 2025. 32

 Chapter 2.1

Union and intersection of several sets

Let A be a family of sets (i.e., a set whose elements are sets). We define the union
and the intersection of the sets in A as
[ \
A = { x | ∃ A ∈ A : x ∈ A }, A = { x | ∀ A ∈ A : x ∈ A }.
A∈A A∈A

If A = {A1 , A2 , . . .}, we also write

∞
[ ∞
\
An = { x | ∃ n ≥ 1 : x ∈ An }, An = { x | ∀ n ≥ 1 : x ∈ An }.
n=1 n=1

Example 2.31. — Let A = [x, ∞) : x ∈ R . Then A∈A A = R and A∈A A = ∅.

 S T

Alternatively, consider A = [x, ∞) : x ∈ [0, 1] . Then A∈A A = [0, ∞) and A∈A A =

 S T

[1, ∞).

Exercise 2.32. — Let X be a set and let A ⊆ P(X) be a collection of subsets of X. Show
the De Morgan laws for unions and intersections of sets:

5
!c !c
[ \ \ [
A = Ac , A = Ac
A∈A A∈A A∈A A∈A

Exercise 2.33. — 1. Show that a finite intersection of intervals is again an interval.

Describe the endpoints of a non-empty intersection in terms of the endpoints of the
given intervals.

2. When is the union of two intervals an interval again? In that case, what can you say
when you unite two intervals of the same type (open, closed, half–open)?

Definition 2.34: Neighbourhoods

Let x ∈ R. A neighbourhood of x is any set that contains an open interval I with
x ∈ I. For δ > 0, the open interval (x − δ, x + δ) is called the δ-neighbourhood of x.

Example 2.35. — Both [−1, 1] and Q ∪ [−1, 1] are neighbourhoods of 0 ∈ R (they contain,
say, (− 21 , 12 )), while [0, 1] is not a neighbourhood of 0.

Note that, for δ > 0 and x ∈ R,

(x − δ, x + δ) = { y ∈ R | |x − y| < δ },

Version: December 30, 2025. 33

 Chapter 2.1

so |x − y| can be interpreted as the distance from x to y. In particular, since |x − y| = |y − x|

for all x, y ∈ R (see Lemma 2.19(a)), the distance from x to y is equal to the one from y to
x.

Definition 2.36: Open and closed sets

A subset U ⊆ R is open if for every x ∈ U there exists an open interval I with
x ∈ I ⊆ U . A subset C ⊆ R is closed if its complement R \ C is open.

Open intervals are open, and closed intervals are closed. Intuitively, a set is open if every
point x in the set comes with a small interval around x still lying in the set. Note that “open”
is not the opposite of “closed”: the sets ∅ and R are both open and closed (their complements
5 are R and ∅, respectively). By contrast, Q ⊂ R and [a, b) ⊂ R are neither open nor closed.

Exercise 2.37. — Show that U ⊆ R is open if and only if for every x ∈ U there exists
δ > 0 such that (x − δ, x + δ) ⊆ U .

Exercise 2.38. — Let U be a family of open subsets of R, and let F be a family of closed
subsets of R. Show that
[ \
U is open, and C is closed.
U ∈U C∈F

(Optional.) Prove also: finite intersections of open sets are open, and finite unions of closed
sets are closed.

Version: December 30, 2025. 34

 Chapter 2.2

2.2 Complex Numbers

2.2.1 Definition of Complex Numbers

Starting from the field of real numbers R, we define the set of complex numbers as

C = R2 = {(x, y) | x, y ∈ R}.

We denote elements z = (x, y) ∈ C in the form z = x + iy, where i is the imaginary unit.
Here x ∈ R is the real part of z, written x = Re(z), and y ∈ R is the imaginary part,
written y = Im(z). Elements with Im(z) = 0 are called real, while those with Re(z) = 0 are
purely imaginary. Via the injective map R ∋ x 7→ x + i0 ∈ C, we identify R with the subset
of real numbers inside C.

iR
iy x + iy

−1 + i i 1+i

−1 0 1 x
5 i i R
−1 − −
2 2

Figure 2.4: The graphical representation of C is the complex plane. In this picture, the real
numbers lie on the real axis and the purely imaginary numbers on the imaginary axis.

As you may expect from previous knowledge, we want i to satisfy i2 = −1. To achieve this,
we define addition and multiplication on C so that it becomes a field. Additionally, we want
these operations to coincide with the usual addition and multiplication when considering real
numbers.
Since i2 = −1, using commutativity and distributivity we get

(x1 + iy1 )(x2 + iy2 ) = x1 x2 + ix1 y2 + iy1 x2 + i2 y1 y2 = (x1 x2 − y1 y2 ) + i(x1 y2 + y1 x2 ).

This motivates the following definition.

Definition 2.39: Addition and multiplication on C

On C = R × R we define addition and multiplication as follows:

(x1 , y1 ) + (x2 , y2 ) = (x1 + x2 , y1 + y2 ),

(x1 , y1 ) · (x2 , y2 ) = (x1 x2 − y1 y2 , x1 y2 + x2 y1 ).

Version: December 30, 2025. 35

 Chapter 2.2

Proposition 2.40: C is a Field

With the operations of Definition 2.39, together with the zero element (0, 0) and the
unit element (1, 0), the set C is a field.

Proof. Additive properties follow immediately from the corresponding properties in R: (0, 0)
is the additive identity and (−x, −y) is the additive inverse of (x, y).
We now check all the properties of multiplication by direct computation. We start from
the associativity.
Given (x1 , y1 ), (x2 , y2 ), (x3 , y3 ) ∈ C, we compute


(x1 , y1 ) · (x2 , y2 ) · (x3 , y3 ) = (x1 x2 − y1 y2 , x1 y2 + y1 x2 ) · (x3 , y3 )
= (x1 x2 x3 − y1 y2 x3 − x1 y2 y3 − y1 x2 y3 , x1 y2 x3 + y1 x2 x3 + x1 x2 y3 − y1 y2 y3 ).

Analogously, we have


(x1 , y1 ) · (x2 , y2 ) · (x3 , y3 ) = (x1 , y1 ) · (x2 x3 − y2 y3 , x2 y3 + y2 x3 )
= (x1 x2 x3 − y1 y2 x3 − x1 y2 y3 − y1 x2 y3 , x1 y2 x3 + y1 x2 x3 + x1 x2 y3 − y1 y2 y3 ).

Since the two expressions on the right-hand side coincide, the multiplication is associative.
5 Commutativity is also easy to check: indeed,

(x1 , y1 ) · (x2 , y2 ) = (x1 x2 − y1 y2 , x1 y2 + x2 y1 ) = (x2 , y2 ) · (x1 , y1 ).

The neutral element is (1, 0) since

(x, y) · (1, 0) = (x · 1 − y · 0, x · 0 + y · 1) = (x, y).

Distributivity is verified as follows:

(x1 , y1 ) · (x2 , y2 ) + (x3 , y3 ) = (x1 , y1 ) · (x2 + x3 , y2 + y3 )
= (x1 x2 + x1 x3 − y1 y2 − y1 y3 , y1 x2 + y1 x3 + x1 y2 + x1 y3 )
= (x1 x2 − y1 y2 , y1 x2 + x1 y2 ) + (x1 x3 − y1 y3 , y1 x3 + x1 y3 )
= (x1 , y1 ) · (x2 , y2 ) + (x1 , y1 ) · (x3 , y3 ),

Finally, it is immediate to check that the multiplication is commutative. Thus, C is a com-

mutative ring.
To conclude, given (x, y) ̸= (0, 0), one can directly verify that its multiplicative inverse is
 
x −y
x2 +y 2
, x2 +y 2
.

Version: December 30, 2025. 36

 Chapter 2.2

Indeed,
   
(x, y) · x
x2 +y 2
, x2−y
+y 2
= x · x2 +yx −y
2 − y · x2 +y 2 , y ·
x
x2 +y 2
+x· −y
x2 +y 2
 2 2 
= xx2 +y
+y 2 , yx−xy
2
x +y 2 = (1, 0).

This proves that C is a field.

From now on we write x + iy instead of (x, y), and abbreviate x + i0 as x, 0 + iy as iy, and
i1 simply as i. Then i2 = −1, and R naturally embeds into C. For z, w ∈ C we write zw for
their product. If z ̸= 0, its multiplicative inverse can be denoted by z −1 or z1 . For example,
i−1 = 1i = −i.

Definition 2.41: Complex Conjugation

For z = x + iy ∈ C we define its conjugate as z = x − iy. The mapping C ∋ z 7→ z ∈ C
is called complex conjugation.

iR
z = x + iy
y

5
0 x R

−y
z = x − iy

Figure 2.5: A complex number z and its conjugate z̄.

Lemma 2.42: Properties of Complex Conjugation

For all z, w ∈ C:

(i) zz = x2 + y 2 ∈ R≥0 . In particular, z z̄ = 0 if and only if z = 0.

(ii) z + w = z + w.

(iii) zw = z w.

Proof. Property (i) follows from the fact that, for z = x + iy, (x + iy)(x − iy) = x2 + y 2 . Also,
x2 + y 2 = 0 if and only if x + iy = 0.
Properties (ii) and (iii) follow from a direct computation, writing z = x1 +iy1 and w = x2 +iy2 :

z + w = (x1 + x2 ) + i(y1 + y2 ) = (x1 + x2 ) − i(y1 + y2 ) = (x1 − iy1 ) + (x2 − iy2 ) = z + w,

Version: December 30, 2025. 37

 Chapter 2.2

z · w = (x1 x2 − y1 y2 ) + i(x1 y2 + y1 x2 ) = (x1 x2 − y1 y2 ) − i(x1 y2 + y1 x2 )

= (x1 − iy1 ) · (x2 − iy2 ) = z · w.

Exercise 2.43. — Show that

z+z z−z
Re(z) = , Im(z) = ,
2 2i

and deduce that R = {z ∈ C | z = z}. Interpret these identities geometrically.

Since i2 = −1 < 0, property (f) in Lemma 2.19 implies that no order compatible with field
operations can exist on C. Nevertheless, calculus can be carried out on C, and this will be
studied in detail in the course on complex analysis. The reason is that C satisfies a suitable
extension of the completeness axiom, which you will meet in future courses.

2.2.2 The Absolute Value on the Complex Numbers

Since no ordering on C makes it an ordered field, we cannot use Definition 2.18 to define an
absolute value as we did on R. Still, we want a notion that extends the real absolute value
and preserves as many of its properties as possible. To this end, we use the root function from
5
Exercise 2.28.

Definition 2.44: Absolute Value

The absolute value (or norm) on C is the map | · | : C → R given by
√ p
|z| = zz = x2 + y 2 , z = x + iy ∈ C.

√ √
For x ∈ R, we have xx = x2 = |x|, so the complex absolute value extends the real one
from Definition 2.18 and the notation is consistent.

Note that |z| ≥ 0 for all z ∈ C, with equality if and only if z = 0 (see Lemma 2.42(i)).
Also, the absolute value is multiplicative:
√ √ √
|zw| = zw zw = zz ww = |z||w| for all z, w ∈ C.

In particular, for z ̸= 0,
z
z −1 = .
|z|2

These properties follow directly from Lemma 2.42. Also, geometrically, |z| = x2 + y 2 is the
p

Euclidean length of the segment going from 0 to z.

Next, we show that the absolute value satisfies the triangle inequality.

Version: December 30, 2025. 38

 Chapter 2.2

Proposition 2.45: Triangle Inequality

For all z, w ∈ C one has

|z + w| ≤ |z| + |w|.

iR

w
z
z+w

0 R
w

Figure 2.6: A visual representation of the triangle inequality: Placing w at the tip of z forms
a triangle with third side z + w. The length of this side is at most the sum of the other two,
namely, |z + w| ≤ |z| + |w|.

To prove the proposition, we first need the following result:

5
Lemma 2.46: Cauchy–Schwarz Inequality

If z = x1 + iy1 and w = x2 + iy2 , then

x1 x2 + y1 y2 ≤ |z||w|. (2.3)

Proof. We observe that

|z|2 |w|2 − (x1 x2 + y1 y2 )2 = (x21 + y12 )(x22 + y22 ) − (x1 x2 + y1 y2 )2

= x21 x22 + y12 y22 + y12 x22 + x21 y22 − (x21 x22 + y12 y22 + 2x1 x2 y1 y2 )
= y12 x22 + x21 y22 − 2x1 x2 y1 y2
= (y1 x2 − x1 y2 )2 ≥ 0.

This proves that (x1 x2 + y1 y2 )2 ≤ |z|2 |w|2 , so by Exercise 2.29,

|x1 x2 + y1 y2 | ≤ |z||w|.

Since x ≤ |x| for all x ∈ R, we obtain (2.3).

Proof of Proposition 2.45. For z = x1 + iy1 and w = x2 + iy2 , using Lemma 2.46 we have

|z + w|2 = (x1 + x2 )2 + (y1 + y2 )2

Version: December 30, 2025. 39

 Chapter 2.2

= |z|2 + |w|2 + 2(x1 x2 + y1 y2 )

5 ≤ |z|2 + |w|2 + 2|z||w| = (|z| + |w|)2 .

Taking square roots proves the result.

Because |z − w| represents the distance between z and w, this motivates the following defi-
nition:

Definition 2.47: Circular Disks

For z ∈ C and r > 0, we define the open disk with radius r > 0 around z as

B(z, r) = {w ∈ C | |z − w| < r},

and the closed disk with radius r > 0 around z as

B(z, r) = {w ∈ C | |z − w| ≤ r}.

In other words, the open disk B(z, r) is the set of points at distance strictly less than r
from z. We note that this definition is compatible with the one of neighborhood in R: if x ∈ R
and r > 0, then
B(x, r) ∩ R = (x − r, x + r).

iR
6

Exercise 2.48. — Let z1 , z2 ∈ C and r1 , r2 > 0. Show that, for every z ∈ B(z1 , r1 ) ∩
B(z2 , r2 ), there exists r > 0 such that

B(z, r) ⊆ B(z1 , r1 ) ∩ B(z2 , r2 ).

Illustrate your choice of r in a picture.

The following definition generalizes that in R from Exercise 2.37.

Version: December 30, 2025. 40

 Chapter 2.2

Definition 2.49: Open and Closed Sets

A set U ⊆ C is open if for every z ∈ U there exists r > 0 such that B(z, r) ⊆ U . A
set C ⊆ C is closed if its complement C \ C is open.
6
In particular, by Exercise 2.48, every open disk is open. More generally, arbitrary unions
of open sets are open. You will return to these notions in greater generality in your future
courses.

Version: December 30, 2025. 41

 Chapter 2.3

2.3 Maximum and Supremum

2.3.1 Existence of the Supremum

Definition 2.50: Bounded Sets, Maxima, and Minima

Let X ⊆ R be a subset of the real numbers.

• X is bounded from above if there exists s ∈ R such that x ≤ s for all x ∈ X.

Such a number s is called an upper bound of X. If s is an upper bound and
also an element of X, we say that s is the maximum of X and write

s = max(X).

• Analogously, X is bounded from below if there exists r ∈ R such that r ≤ x

for all x ∈ X. Such a number r is called a lower bound of X. If r is a lower
bound and also an element of X, we say that r is the minimum of X and write

r = min(X).

6
• X is called bounded if it is both bounded from above and bounded from below.

Remark 2.51. — If a set X ⊆ R has a maximum, then it is unique. Indeed, if x1 , x2 ∈ X

are both maxima, then x1 ≤ x2 (since x2 is a maximum) and x2 ≤ x1 (since x1 is a maximum),
so x1 = x2 .

A closed interval [a, b] with a < b has both a minimum and a maximum: a = min([a, b])
and b = max([a, b]). But not all sets have a maximum. For instance, the open interval (a, b)
does not have a maximum because the endpoint b, though an upper bound, is not contained
in the set. Similarly, R and unbounded intervals such as [a, ∞) or (a, ∞) have no maximum.

Definition 2.52: Supremum

Let X ⊆ R be a subset and let

A := {a ∈ R | x ≤ a for all x ∈ X}

be the set of all upper bounds of X. If A has a minimum, we call this minimum the
supremum of X and write
sup(X) = min(A).

Version: December 30, 2025. 42

 Chapter 2.3

In other words, the supremum of X is the smallest real number that is greater than or
equal to every element of X. Note that we can describe the supremum s = sup(X) as follows:

x ≤ s for all x ∈ X, and if t < s, then t is not an upper bound of X. (2.4)

This means that for every t < s, there exists some x ∈ X such that x > t. That is,

x ≤ s for all x ∈ X, and ∀ t < s ∃ x ∈ X such that x > t. (2.5)

The two characterizations (2.4) and (2.5) are equivalent.

Remark 2.53. — If a set X has a maximum, then this element is also the supremum.
Indeed, the maximum is an upper bound of X, and since it lies in X, no smaller upper bound
can exist.

Example 2.54. — Let

1
| n ≥ 1 = 0, 21 , 23 , 34 , . . . .
 
X := 1 − n

Then X is a bounded subset of R with sup(X) = 1.

• 1 is an upper bound of X, since 1 − 1

< 1 for all n ≥ 1.
6 n

• No number smaller than 1 is an upper bound, because for any t < 1 we can find n
large enough such that 1 − n1 > t. (We will justify this fact rigorously later using the
Archimedean Principle, see Corollary 2.60.)

• Therefore, sup(X) = 1, even though 1 ∈

/ X, so X has no maximum.

sup(X) = 1

Figure 2.7: The set X = 1 − 1

| n ≥ 1 . The points get arbitrarily close to 1, but never

n
reach it.

Note that not every set has a supremum. If X = ∅ or if X is unbounded from above,
then sup(X) does not exist. However, for any nonempty and bounded-above subset of R, the
supremum always exists:

Theorem 2.55: Existence of the Supremum

Let X ⊂ R be nonempty and bounded from above. Then sup(X) exists and is a real
number.

Proof. Since X is bounded from above, the set A := {a ∈ R | x ≤ a for all x ∈ X} of upper
bounds is nonempty. Since x ≤ a for any x ∈ X and a ∈ A, we can apply the completeness

Version: December 30, 2025. 43

 Chapter 2.3

axiom (Definition 2.21) to find c ∈ R such that

x ≤ c ≤ a for all x ∈ X and a ∈ A.

The first inequality implies that c is itself an upper bound (so c ∈ A), while the second
inequality tells us that c is smaller than or equal to every other upper bound. Hence, c =
min(A) = sup(X).

Proposition 2.56: Supremum and Set Operations

Let X and Y be nonempty subsets of R that are bounded from above. Define

X + Y := {x + y | x ∈ X, y ∈ Y } and X · Y := {x · y | x ∈ X, y ∈ Y }.

Then the sets X ∪ Y , X ∩ Y , and X + Y are also bounded from above. Moreover, if
X, Y ⊂ R≥0 (that is, x ≥ 0 and y ≥ 0 for all x ∈ X and y ∈ Y ), then X · Y is bounded
from above as well.
In these cases, the following formulas hold:

(1) sup(X ∪ Y ) = max{sup(X), sup(Y )},

(2) If X ∩ Y ̸= ∅, then sup(X ∩ Y ) ≤ min{sup(X), sup(Y )},

6 (3) sup(X + Y ) = sup(X) + sup(Y ),

(4) If X, Y ⊂ R≥0 , then sup(X · Y ) = sup(X) · sup(Y ).

Proof. We leave (1) and (2) to the reader.

(3) Let x0 = sup(X) and y0 = sup(Y ). For any z ∈ X + Y , there exist x ∈ X and y ∈ Y
such that z = x + y. Since x ≤ x0 and y ≤ y0 , we have

z = x + y ≤ x 0 + y0 ,

so x0 + y0 is an upper bound for X + Y . We now want to show that x0 + y0 = sup(X + Y ).

Let z0 = sup(X + Y ) and suppose, by contradiction, that

ε := x0 + y0 − z0 > 0.

Since x0 = sup(X), by the characterization (2.5) there exists x ∈ X such that x > x0 − ε/2.
Likewise, there exists y ∈ Y such that y > y0 − ε/2. Setting z = x + y, we obtain

ε ε
z > x0 − + y0 − = x0 + y0 − ε = z0 ,
2 2

contradicting the assumption that z0 is an upper bound for X + Y . Therefore, z0 = x0 + y0 .

Version: December 30, 2025. 44

 Chapter 2.3

(4) The proof is analogous. If all elements of X and Y are non-negative, and we set
x0 = sup(X) and y0 = sup(Y ), then for any z = x · y ∈ XY , we have

z = x · y ≤ x 0 · y0 ,

which shows that x0 · y0 is an upper bound for X · Y . Using a similar “ε-argument” as done
above when proving (3), one shows that this upper bound is sharp, i.e., x0 · y0 is the least
upper bound.

If X ⊆ R is nonempty and bounded from below, the largest lower bound of X is called
the infimum, denoted by inf(X). An existence result analogous to Theorem 2.55 holds for
infima as well. Moreover, the infimum can be expressed using the supremum:

inf(X) = − sup{−x | x ∈ X}.

This means that most results about infima can be deduced directly from those about suprema.

2.3.2 Two-point Compactification

6 In this section, we extend the notions of supremum and infimum to arbitrary subsets of R.
To do so, we introduce two formal symbols:

+∞ and − ∞,

which are not real numbers. We define the extended real number line (also called the
two-point compactification of R) by

R = R ∪ {−∞, +∞}.

Figure 2.8: The extended real number line R, obtained by adding −∞ on the far left and +∞
on the far right of the usual real line.

We extend the usual order relation ≤ on R to R by requiring that

−∞ < x < +∞ for all x ∈ R.

For simplicity, we often write ∞ instead of +∞.

Version: December 30, 2025. 45

 Chapter 2.3

We now introduce some standard (but informal) computation rules involving these symbols.
For all x ∈ R, we adopt the conventions:

∞ + x = ∞ + ∞ = ∞, −∞ + x = −∞ − ∞ = −∞.

If x > 0, then
x · ∞ = ∞ · ∞ = ∞, x · (−∞) = ∞ · (−∞) = −∞,

while for x < 0 we have

x · ∞ = −∞ · ∞ = −∞, x · (−∞) = −∞ · (−∞) = ∞.

These rules are widely used as notational shorthand, but one must handle them with care.
Expressions like
∞ − ∞, 0 · ∞, or similar

are undefined and should be avoided.

Definition 2.57: Supremum and Infimum in the Extended Line

6
Let X ⊆ R.

• If X is not bounded from above, we define sup(X) = ∞.

• If X = ∅, we define sup(∅) = −∞.

• If X is not bounded from below, we define inf(X) = −∞.

• If X = ∅, we define inf(∅) = ∞.

In this context, we refer to ∞ and −∞ as indefinite values.

In other words:

• Saying sup(X) = ∞ means that X is not bounded above; i.e.,

∀ x0 ∈ R ∃ x ∈ X such that x > x0 .

• Saying sup(X) = −∞ means that X is empty.

• Similarly, inf(X) = −∞ means that X is unbounded below, and inf(X) = ∞ means X

is empty.

Version: December 30, 2025. 46

 Chapter 2.4

2.4 Consequences of Completeness

We introduced the root function in Section 2.1 using the completeness axiom, and in Section
2.3 we used the same axiom to prove the existence of suprema. In this section, we discuss
further consequences of the completeness axiom.

2.4.1 The Archimedean Principle

The Archimedean principle states that for every real number x ∈ R there exists an integer n
greater than x. The following theorem, proved using the existence of suprema, gives a precise
formulation of this principle.

Theorem 2.58: Archimedean Principle

For every x ∈ R there exists exactly one n ∈ Z such that

n ≤ x < n + 1.

Proof. We first treat the case x ≥ 0. Define

E = {n ∈ Z | n ≤ x}.
7
Since 0 ∈ E and x is an upper bound, E is a non-empty subset of R bounded from above.
Hence, by Theorem 2.55, the supremum s0 = sup(E) exists. From the definition of supremum
we deduce:

(i) s0 ≤ x (because x is an upper bound);

(ii) there exists n0 ∈ E with s0 − 1 < n0 (otherwise s0 − 1 would also be an upper bound).

From (ii) we obtain s0 < n0 + 1, which implies

(iii) n0 + 1 ∈
/ E (otherwise s0 would not be an upper bound for E).

Moreover, since m ≤ s0 for every m ∈ E, we have m < n0 + 1 for all m ∈ E. As all elements
of E are integers,

m < n0 + 1 ⇐⇒ m − n0 < +1 ⇐⇒ m − n0 ≤ 0 ⇐⇒ m ≤ n0 .

Thus every m ∈ E is less than or equal to n0 , and since n0 ∈ E, we conclude that n0 is

the maximum of E. In particular, by Remark 2.53, the maximum is also the supremum, so
s0 = n0 .
Finally, recalling (iii) and the definition of E, we have n0 + 1 > x. Together with (i), this
shows
n0 = s0 ≤ x < n0 + 1,

Version: December 30, 2025. 47

 Chapter 2.4

establishing the claim for x ≥ 0.

Now, if x < 0, apply the previous argument to −x > 0. Then there exists m ∈ Z such that

m ≤ −x < m + 1,

which is equivalent to
−m − 1 < x ≤ −m.

If x = −m, then set n = −m. If x < −m, set n = −m − 1. In both cases, we obtain

n ≤ x < n + 1.

Finally, for uniqueness, assume that n1 , n2 ∈ Z both satisfy ni ≤ x < ni + 1. From

n1 ≤ x < n2 + 1 we deduce n1 < n2 + 1, and therefore n1 ≤ n2 . Reversing the roles of n1 and
n2 gives n2 ≤ n1 . Hence n1 = n2 .

Definition 2.59: Integer and Fractional Parts

The integer part ⌊x⌋ of x ∈ R is the integer n ∈ Z uniquely determined by Theorem
2.58 such that n ≤ x < n + 1. The map x 7→ ⌊x⌋ from R to Z is called the rounding
function. The fractional part of x is defined as

{x} = x − ⌊x⌋ ∈ [0, 1).

7
1
Corollary 2.60: n is arbitrarily small

For every ε > 0 there exists n ∈ N, with n ≥ 1, such that

1
< ε.
n

Proof. Applying Theorem 2.58 to x = 1

ε > 0, we find m ∈ Z such that

1
m≤ < m + 1.
ε

Set n := m + 1. In this way we have 0 < 1

ε < n, which is equivalent to n > 0 (therefore,
n ≥ 1) and n1 < ε.

Corollary 2.61: Density of Q

For every a, b ∈ R with a < b, there exists r ∈ Q such that a < r < b.

Proof. Set ε = b − a. By Corollary 2.60, there exists m ∈ N with 1

m < ε. Them, by Theorem
2.58 applies with x = ma, there exists n ∈ Z with

n ≤ ma < n + 1,

Version: December 30, 2025. 48

 Chapter 2.4

or equivalently,
n n+1
≤a< .
m m
Since 1
m < ε, by the two inequalities above we obtain

n+1 1
a< ≤a+ < a + ε = b.
m m

Thus r = n+1
m is a rational number between a and b.

Definition 2.62: Dense sets

A subset X ⊂ R is called dense in R if every open non-empty interval contains an
element of X.

By Corollary 2.61, Q is dense in R.

Archimedes’ principle can be generalized in several ways. The following generalization will
be useful when discussing decimal fractions.

Exercise 2.63. — Show the following analogue of Theorem 2.58: For every x ∈ R with
x ≥ 1, there exists exactly one n ∈ N such that

10n ≤ x < 10n+1 .

7
Exercise 2.64. — Show that Corollary 2.60 holds when we replace N by the powers of 10.
In other words: For every ε > 0 there exists n ∈ N such that

10−n < ε.

While the Archimedean principle looks “obvious”, it is actually more subtle than one may
think. Indeed, there exist ordered fields for which the Archimedean principle fails, as the
following exercise shows.

Exercise 2.65 (Advanced). — ex:non-archimedean An ordered field is called non-Archimedean

if the Archimedean property does not hold. Let
 
p(x)
F = : p, q ∈ R[x], q(x) ̸≡ 0 ,
q(x)

where R[x] denotes the set of polynomials with real coefficients (e.g. p(x) = 3x4 − 5x + 7,
q(x) = 2x2 − 1), and q(x) is different from the polynomial that is identically zero. Inside this
field, the set Z corresponds to the constant polynomial f (x) ≡ n ∈ Z.
We define the following order: f > g in F if f (x) − g(x) > 0 for all sufficiently large x ∈ R.

1. Show that F is an ordered field.

Version: December 30, 2025. 49

 Chapter 2.4

2. Show that F is non-Archimedean by proving that x is larger than every integer.

2.4.2 Decimal Fraction Expansion and Uncountability

A common way to understand real numbers is through their decimal expansions. Formally, a
decimal fraction is a sequence of integers

a0 , a1 , a2 , a3 , . . .

where a0 ∈ Z and 0 ≤ an ≤ 9 for all n ≥ 1. To such a sequence, we associate a real number

as follows.
Assume first that a0 ≥ 0 and define the following approximations:

x0 = a0 , y0 = a0 + 1,
a1 a1 1 a1 + 1
x1 = a0 + , y1 = a0 + + = a0 + ,
10 10 10 10
a1 a2 a1 a2 1 a1 a2 + 1
x2 = a0 + + , y2 = a0 + + + = a0 + + ,
10 100 10 100 100 10 100

and, more generally,

a1 an a1 an + 1
xn = a0 + + ... + n yn = a0 + + ... + . (2.6)
10 10 10 10n

7 Note that all the numbers xn and yn are rational. Also, since 0 ≤ ak ≤ 9 for k ≥ 1, one can
check that
x0 ≤ x1 ≤ · · · ≤ xn ≤ xn+1 ≤ · · · ≤ yn+1 ≤ yn ≤ · · · ≤ y1 ≤ y0 .

Thus, if we consider the sets X = {x0 , x1 , x2 , . . .} and Y = {y0 , y1 , y2 , . . .}, by the completeness
axiom there exists c ∈ R such that

xn ≤ c ≤ y n for all n ∈ N. (2.7)

Also, Exercise 2.64 shows that such a c is uniquely determined: since yn − xn = 10−n , if
c, d ∈ R both satisfy (2.7), then

|d − c| < 10−n for all n ∈ N,

which implies |d − c| = 0, i.e., c = d. We therefore define c to be the real number with decimal
expansion
c = a0 .a1 a2 a3 . . .
c
x1 x2 x3 y3 y2 y1

Figure 2.9: The rational numbers xn give better and better lower estimates for c, while the
rational numbers yn provide increasingly accurate upper estimates. The number c lies between
all of them.

Version: December 30, 2025. 50

 Chapter 2.4

Remark 2.66. — Alternatively, the number c may be defined either as the supremum of
the set X = {x0 , x1 , x2 , . . .} or as the infimum of the set Y = {y0 , y1 , y2 , . . .}.

If a0 < 0, we proceed as follows: consider the decimal expansion of the (−a0 , a1 , a2 , a3 , . . .),
apply the previous construction, and then define the desired number to be the negative of the
result.

Example 2.67. — Let

a0 = −33, a1 = 1, a2 = 6, an = 0 for n ≥ 3.

Then the construction above applied to (−a0 , a1 , a2 , . . .) gives 33.16, so the corresponding real
number is
c = −33.16.

Conversely, we ask: can every x ∈ R be written as a decimal expansion? The answer is

yes. For x ≥ 0, set a0 := ⌊x⌋ and define

an := ⌊10n x⌋ − 10⌊10n−1 x⌋, n ≥ 1. (2.8)

7
Then 0 ≤ an ≤ 9, and one checks that (2.7) holds, so that x has decimal expansion
a0 .a1 a2 a3 . . .. For x ≤ 0, apply the same procedure to −x and then change the sign of
a0 .
Note that different decimal expansions can represent the same number, e.g.

0.1999 . . . = 0.2000 . . . = 51 .

This phenomenon occurs exactly when the expansion eventually becomes all 9s. To avoid
ambiguity, we define a real decimal fraction as a sequence

a0 , a1 , a2 , a3 , . . .

with 0 ≤ an ≤ 9 for all n ≥ 1, such that for every n0 ≥ 1 there exists n ≥ n0 with an ̸= 9.

Exercise 2.68. — Let x ≥ 0 be a real number. Show that the sequence a0 , a1 , a2 , . . .

defined by (2.8) is a real decimal fraction. Conclude that this gives a bijection between R and
the set of all real decimal fractions.

Version: December 30, 2025. 51

 Chapter 2.4

Cardinality
Let X and Y be sets.

• We say X and Y have the same cardinality, written X ∼ Y , if there is a

bijection f : X → Y .

• We write X ≲ Y if there is an injection f : X → Y .

• The empty set has cardinality 0.

• A set X has finite cardinality |X| = n if there is a bijection with {1, . . . , n}.

• A set is infinite if it is not finite.

• A set is countable if it has a bijection to N. Its cardinality is denoted ℵ0 ,

pronounced Aleph-0.

• A set is uncountable if it is infinite but not countable.

Remark 2.69. — If X ≲ Y and Y ≲ X, then X ∼ Y . In other words, if there exist an

injective map f : X → Y and an injective map g : Y → X, then one can find a bijective map
h : X → Y . This nontrivial statement is the Schröder–Bernstein Theorem.
7
Example 2.70 (Some Countability Results). —

1. N and the even numbers have the same cardinality.

Let Neven := {0, 2, 4, 6, . . . }. The map

f1 : N → Neven , f1 (n) = 2n,

is a bijection, so N ∼ Neven .

2. N and Z have the same cardinality.

Define a bijection f2 : Z → N by

2n if n ≥ 0,
f2 (n) =
−2n − 1 if n < 0.

Explicitly:

0 7→ 0, 1 7→ 2, 2 7→ 4, 3 7→ 6, ..., −1 7→ 1, −2 7→ 3, −3 7→ 5, ...

This shows that Z ∼ N.

3. Q is countable.

Version: December 30, 2025. 52

 Chapter 2.4

First of all, since N ⊂ Q, the inclusion map n 7→ n is injective. So N ≲ Q.

To prove the converse, we first show that Q ≲ N × N. Indeed, every rational number
can be written as a reduced fraction pq with p ∈ Z and q ∈ N \ {0}. The map
 
p
f3 : Q → Z × N \ {0}, f3 q = (p, q)

is injective, so Q ≲ Z × N \ {0} ≲ Z × N. Using the map f2 from Example 2 we see that

that map


f4 : Z × N → N × N, f4 (n, m) = f2 (n), m ,

is a bijection, therefore Z × N ∼ N × N. Thus, we have proved that

Q ≲ Z × N ∼ N × N.

7 Next, we show that N × N ∼ N. Define the “Cantor pairing function”

(m + n)(m + n + 1)
f5 : N × N → N, f5 (m, n) = + n.
2

One can prove that f5 is a bijection (although only injectivity is needed here), as one
can also convince oneself by looking at Figure 2.10, hence

N × N ∼ N.

Putting everything together:

N ≲ Q ≲ Z × N ∼ N × N ∼ N,

so it follows from the Schröder–Bernstein Theorem, that Q ∼ N, i.e., the rational

numbers are countable.

Version: December 30, 2025. 53

 Chapter 2.4

54
9
(0,9)

44 53
8
(0,8) (1,8)

35 43 52
7
(0,7) (1,7) (2,7)

27 34 42 51
6
(0,6) (1,6) (2,6) (3,6)

20 26 33 41 50
5
(0,5) (1,5) (2,5) (3,5) (4,5)

14 19 25 32 40 49
7 4
(0,4) (1,4) (2,4) (3,4) (4,4) (5,4)

9 13 18 24 31 39 48
3
(0,3) (1,3) (2,3) (3,3) (4,3) (5,3) (6,3)

5 8 12 17 23 30 38 47
2
(0,2) (1,2) (2,2) (3,2) (4,2) (5,2) (6,2) (7,2)

2 4 7 11 16 22 29 37 46
1
(0,1) (1,1) (2,1) (3,1) (4,1) (5,1) (6,1) (7,1) (8,1)

0 1 3 6 10 15 21 28 36 45
0
(0,0) (1,0) (2,0) (3,0) (4,0) (5,0) (6,0) (7,0) (8,0) (9,0)

0 1 2 3 4 5 6 7 8 9

Figure 2.10: A representation of the function f5 . This map enumerates all elements of N × N.
The value of f5 (m, n) is in black; the coordinate (m, n) is in blue.

By Example 2.70 we know that Q is countable, namely, there exists a bijection f : N → Q.

Hence, if we define r0 = f (0), r1 = f (1), r2 = f (2), etc., we have proved that there exists a
(non-canonical) way of enumerating all the rational numbers. As we shall see now, this is not
possible for the real numbers.

Proposition 2.71: Uncountability of R

The set R is uncountable.
8
Proof. To show that R is uncountable, we shall actually prove that [0, 1) ⊂ R is uncountable.
Assume by contradiction [0, 1) is countable. Then there exists a bijective map f : N → [0, 1).
In other words, if we define xn = f (n), we have found a way to enumerate all the real numbers
in [0, 1):

x0 = 0.a0,1 a0,2 a0,3 a0,4 . . . , x1 = 0.a1,1 a1,2 a1,3 a1,4 . . . ,

Version: December 30, 2025. 54

 Chapter 2.4

x2 = 0.a2,1 a2,2 a2,3 a2,4 . . . , x3 = 0.a3,1 a3,2 a3,3 a3,4 . . . , ...

We now want to construct a new number x ∈ [0, 1) that is not in the list above. A possible
way to do this is the following: write x = 0.b1 b2 b3 b4 and define
  
5 if a0,1 ̸= 5,  5 if a1,2 ̸= 5, 5 if a ̸= 5,
2,3
b1 = , b2 = , b3 = ,
6 if a = 5,
0,1
 6 if a = 5, 1,2
6 if a = 5, 2,3
8
and, more in general, 
5 if a ̸ 5,
i,i+1 =
bi = for all i ≥ 1.
6 if a
i,i+1 = 5.

We now observe that the number x = 0.b1 b2 b2 . . . cannot be in the list, since for every i ≥ 0 it
differs from xi in the (i + 1)-th decimal place. This contradicts the assumption that we have
listed all real numbers. Therefore [0, 1) (and hence also R) is uncountable.

Extra material: An alternative proof of the uncountability of R via power sets

We have seen that sets like Z and Q are countable, whereas the set of real numbers R is
fundamentally larger. In this section, we present an alternative proof of this fact and discuss
in more detail the question: “How large is R?”

Power Set
Let X be a set. The power set P(X) of X is the set of all subsets of X:

P(X) := {A ⊆ X}.

Example 2.72. — If X = {0, 1, 2}, then

P(X) = {∅, {0}, {1}, {2}, {0, 1}, {0, 2}, {1, 2}, {0, 1, 2}}.

Theorem 2.73: Cantor’s Theorem

For any set X, the power set P(X) has strictly larger cardinality than X.

Proof. First of all, since there is an injective map i : X → P(X) given by i(x) = {x}, we see
that |X| ≤ |P(X)|.
Now, assume for contradiction that there exists a bijection f : X → P(X) and define the
set
A := {x ∈ X | x ∈/ f (x)}.

Since f is surjective, there exists a ∈ X such that f (a) = A. However we now see that this is
impossible. Indeed:

Version: December 30, 2025. 55

Chapter 2.4

• If a ∈ A, then by the definition of A it follows that a ∈

/ f (a) = A, contradiction.

• If a ∈
/ A, then by the definition of A it follows that a ∈ f (a) = A, again a contradiction.

Thus, no such bijection exists and P(X) is strictly larger than X.

We now show that the set of real numbers has the same cardinality as P(N).

Proposition 2.74: The reals have the same cardinality as P(N)

|R| = |P(N)|.

Proof. We first show that |P(N)| ≲ |[0, 1)| by constructing an injection

φ : P(N) ,→ [0, 1).

Given a subset A ⊂ N, define φ(A) as the real number with binary expansion

1 if n ∈ A,
φ(A) := 0.a0 a1 a2 a3 . . . where an =
0 if n ∈
/ A.

If A ̸= B, let n be the smallest element of A△B. Then φ(A) and φ(B) differ in the n-th digit,
so φ(A) ̸= φ(B). Hence φ is injective.
Conversely, we show that |[0, 1)| ≲ |P(N)|. Every real number x ∈ [0, 1) has a binary
expansion of the form
x = 0.b0 b1 b2 . . . with bi ∈ {0, 1}.

To such an expansion we can associate the subset

Ax := {n ∈ N | an = 1} ⊆ N.

This gives a map [0, 1) → P(N). There is only one subtlety: some numbers have two binary
expansions (for example, 0.011111 . . . = 0.100000 . . .). To avoid ambiguity, we agree to always
choose the expansion that ends with infinitely many zeros rather than the one ending with
infinitely many ones. With this convention, each real number in [0, 1) corresponds uniquely to a
subset of N. Therefore, we have defined an injection [0, 1) ,→ P(N), and hence |[0, 1)| ≲ |P(N)|.
By the Schröder–Bernstein Theorem, we obtain |[0, 1)| = |P(N)|.
Finally, it remains to see that [0, 1) and R have the same cardinality. An explicit injection
is given by
x 1
f : R → (0, 1), f (x) = + .
2(1 + |x|) 2
Thus |R| ≲ |(0, 1)| ≲ |[0, 1)|. Since trivially |[0, 1)| ≲ |R|, we conclude that

|R| ∼ |[0, 1)|.

Combining everything, we obtain |R| = |P(N)|.

Version: December 30, 2025. 56

Chapter 2.4

The cardinality of P(N) is denoted c and called the continuum. Hence, we have proved
that
|R| = c and ℵ0 < c.

A famous question, posed by Cantor, is whether there exists a set whose cardinality lies
strictly between ℵ0 and c. This is the Continuum Hypothesis. Remarkably, this question
can neither be proved nor disproved from the standard axioms of set theory (ZFC): it is
independent of them. This was shown by Gödel (1938) and Cohen (1963).

Version: December 30, 2025. 57

 Chapter 2.5 Sequences of Real Numbers

2.5 Sequences of Real Numbers

2.5.1 Convergence of Sequences

Let X be a set. Intuitively, a sequence in X is a list of elements x0 , x1 , x2 , . . . indexed by the
natural numbers. In this section we study sequences and their properties in R. We now give
a precise definition.

Definition 2.75: Sequences

A sequence in R is a function a : N → R. For n ∈ N we write a(n) = an and call an

the n-th term of the sequence. Instead of a : N → R one often writes (an )n∈N , (an )∞
n=0 ,
or (an )n≥0 .

Since we primarily use the letter x to denote a real number, for sequences of real numbers
we shall mostly write (xn )n∈N , (xn )∞
n=0 , or (xn )n≥0 .

Remark 2.76. — Throughout these notes we usually write sequences as (xn )∞ n=0 , but the
starting index is not essential. A sequence may equally well be defined from n = 1, or from
any other integer, without changing any of the notions or results below. In particular, shifting
the starting index does not affect the concepts of convergence, boundedness, or monotonicity,
which we will discuss later.
8

Example 2.77. — Examples of sequences are:

(i) (xn )∞
n=0 given by xn = (−1) ;
n

n=1 given by xn = n .
(ii) (xn )∞ 1

Definition 2.78: (Eventually) Constant Sequences

A sequence (xn )∞n=0 is constant if xn = xm for all m, n ∈ N. It is eventually constant

if there exist N ∈ N such that xn = xm for all n, m ≥ N .

Definition 2.79: Convergence of Sequences

Let (xn )∞
n=0 be a sequence in R. We say that (xn )n=0 converges (or is convergent)
∞

if there exists A ∈ R such that

∀ ε > 0 ∃ N ∈ N such that |xn − A| < ε for all n ≥ N .

In this case we write

lim xn = A (2.9)
n→∞

and call A the limit of (xn )∞

n=0 .

Version: December 30, 2025. 58

 Chapter 2.5 Sequences of Real Numbers

A priori it is not clear that a convergent sequence has only one limit. The following lemma
shows that the limit is indeed unique, so the notation (2.9) is justified.

Lemma 2.80: Uniqueness of the Limit

A convergent sequence (xn )∞

n=0 has exactly one limit.

Proof. Let A, B ∈ R be limits of (xn )∞

n=0 . Fix ε > 0. Then there exist NA , NB ∈ N such that
|xn − A| < ε for all n ≥ NA and |xn − B| < ε for all n ≥ NB . Setting N = max{NA , NB }, we
have
|A − B| ≤ |A − xN | + |xN − B| < ε + ε = 2ε.

Since ε > 0 is arbitrary, |A − B| = 0 and hence A = B.

Example 2.81. — A constant sequence (xn )∞ n=0 with xn = A ∈ R for all n converges to
A. Similarly, an eventually constant sequence converges to the value it eventually takes.

∞
Example 2.82. — The sequence n1 n=1 converges to 0, i.e., limn→∞ n1 = 0. Indeed, given
ε > 0, by Archimedes’ principle (Theorem 2.58) there exists N ∈ N with N1 < ε. Then, for all
n ≥ N , we have n1 − 0 = n1 ≤ N1 < ε.
8
Example 2.83. — The sequence (xn )∞ n=0 defined by xn = (−1) is not convergent, since
n

its terms alternate between 1 and −1 and do not approach any real number.

2.5.2 Convergent Subsequences and Accumulation Points

Let (xn )∞
n=0 be a sequence in R. A subsequence of (xn )n=0 is obtained by keeping only certain
∞

elements and discarding the others. For example,

x0 , x1 , x4 , x9 , x16 , x25 , . . .

is a subsequence. The formal definition is as follows.

Definition 2.84: Subsequences

Let (xn )∞
n=0 be a sequence. A subsequence is a sequence of the form (xnk )k=0 , where
∞

k=0 is a strictly increasing sequence of nonnegative integers, i.e. nk+1 > nk for all
(nk )∞
k ∈ N.

Remark 2.85. — Since nk+1 > nk for all k ∈ N, it follows by induction that nk ≥ k for
every k ∈ N.

Version: December 30, 2025. 59

 Chapter 2.5 Sequences of Real Numbers

Lemma 2.86: Subsequences of Convergent Sequences are Convergent

Let (xn )∞ ∞
n=0 be a sequence converging to A ∈ R. Then every subsequence (xnk )k=0 also
converges to A.

Proof. We leave the proof as an exercise.

8
A sequence can have convergent subsequences without being convergent itself. For example,
the sequence xn = (−1)n is not convergent, while the subsequences

(x2n )∞
n=0 and (x2n+1 )∞
n=0

are constant (equal to 1 and −1, respectively), hence convergent.

Definition 2.87: Accumulation Points of Sequences

Let (xn )∞n=0 be a sequence in R. A point A ∈ R is an accumulation point of (xn )n=0

∞

if for every ε > 0 and every N ∈ N there exists n ≥ N such that |xn − A| < ε.

Proposition 2.88: Subsequences and Accumulation Points

Let (xn )∞ ∞
n=0 be a sequence in R. A point A ∈ R is an accumulation point of (xn )n=0 if
and only if there exists a convergent subsequence of (xn )∞
n=0 with limit A.

Proof. Assume first A is an accumulation point. We construct (nk )k≥0 recursively:

- first, apply the definition of accumulation point with N = 1 and ε = 1 = 2−0 to find n0 ≥ 1
with |xn0 − A| ≤ 2−0 ;
9 - second, apply the definition of accumulation point with N = n0 + 1 and ε = 2−1 to find
n1 ≥ n0 + 1 with |xn1 − A| ≤ 2−1 ;
- more in general, given nk−1 , apply the definition of accumulation point with N = nk−1 + 1
and ε = 2−k to find nk > nk−1 with |xnk − A| ≤ 2−k .
Now, given ε > 0, pick N so that 2−N < ε. Then, for all k ≥ N we have

|xnk − A| ≤ 2−k ≤ 2−N < ε,

so limk→∞ xnk = A.
Conversely, assume that there exists a subsequence (xnk )∞
k=0 converging to A. Fix ε > 0
and N ∈ N. Since limk→∞ xnk = A, there exists N0 such that |xnk − A| < ε for all k ≥ N0 .
Hence, if we choose k = max{N0 , N }, because nk ≥ k (recall Remark 2.85) we have nk ≥ N
and |xnk − A| < ε. Thus A is an accumulation point.

Version: December 30, 2025. 60

 Chapter 2.5 Sequences of Real Numbers

Corollary 2.89: Infinitely Many Terms Near an Accumulation Point

If A ∈ R is an accumulation point of (xn )∞
n=0 , then for every ε > 0 there are infinitely
many n with xn ∈ (A − ε, A + ε).

k=0 with limk→∞ xnk = A. Hence

Proof. By Proposition 2.88, there exists a subsequence (xnk )∞
for any ε > 0 there exists K such that xnk ∈ (A − ε, A + ε) for all k ≥ K, providing infinitely
many elements of the sequence inside the interval (A − ε, A + ε).

Corollary 2.90: Accumulation Points of Convergent Sequences

A convergent sequence has exactly one accumulation point, namely its limit.

Proof. This follows from Lemma 2.86 and Proposition 2.88.

Example 2.91. — The sequence (xn )∞ n=0 , defined by xn = (−1) , has two accumulation
n

points: 1 and −1. Indeed, the subsequence (x2n )∞n=0 is constantly equal to 1, hence it converges
to 1, while the subsequence (x2n+1 )n=0 is constantly equal to −1, and therefore converges to
∞

−1.

Exercise 2.92. — Let (xn )∞ n=0 be a sequence in R, and let E ⊆ R be the set of its

9 accumulation points. Show that E is closed.

Exercise 2.93. — Construct a sequence (xn )∞ n=0 in R whose set of accumulation points is
the entire interval [0, 1].
Hint: Enumerate the rational numbers in [0, 1].

2.5.3 Addition, Multiplication, and Inequalities

Given two sequences (xn )∞
n=0 and (yn )n=0 , one can combine them through addition and mul-
∞

tiplication, and study how these operations affect their limits. Moreover, one can compare
sequences using inequalities and relate these inequalities to the corresponding limits.

Proposition 2.94: Limits and Operations

Let (xn )∞ ∞
n=0 and (yn )n=0 be sequences converging to A, B ∈ R, respectively. Then:

(1) The sequence (xn + yn )∞

n=0 converges to A + B.

(2) The sequence (xn yn )∞

n=0 converges to AB.

(3) Given α ∈ R, the sequence (αxn )∞

n=0 converges to αA.

(4) Suppose xn ̸= 0 for all n ∈ N and A ̸= 0. Then the sequence (x−1 ∞

n )n=0 converges
to A−1 .

Version: December 30, 2025. 61

 Chapter 2.5 Sequences of Real Numbers

Proof. We only prove (1) and (4), leaving (2) and (3) as exercises for the reader. To prove
(2), see Exercise 2.101 below.
(1) Fix ε > 0. Because limn→∞ xn = A and limn→∞ yn = B, there exist NA , NB ∈ N such
that
|xn − A| < 2ε ∀ n ≥ NA , |yn − B| < 2ε ∀ n ≥ NB .

Then, for all n ≥ N = max{NA , NB },

ε ε
|(xn + yn ) − (A + B)| ≤ |xn − A| + |yn − B| < 2 + 2 = ε.

|A|
(4) Since A ̸= 0, we can apply the definition of limit with ε = 2 to find N0 ∈ N such that
for all n ≥ N0 ,
|A|
|xn − A| < .
2
Using the reverse triangle inequality (Lemma 2.19(h)), this gives

|A|
|xn | ≥ |A| − |xn − A| > ∀ n ≥ N0 ,
2

so
1 2
< ∀ n ≥ N0 .
|xn | |A|
9 Hence, we have proved that

xn − A |xn − A| 2
x−1
n −A
−1
= = < |xn − A| ∀ n ≥ N0 .
xn A |xn | |A| |A|2

Now, given ε > 0, choose N1 ∈ N such that

|A|2
|xn − A| < ε ∀ n ≥ N1 .
2

Then, if we define N = max{N0 , N1 }, we have

2
x−1
n −A
−1
< |xn − A| < ε ∀ n ≥ N.
|A|2

Thus the sequence (x−1

n )n=0 converges to A .
∞ −1

Proposition 2.95: Limits and Inequalities

Let (xn )∞ ∞
n=0 and (yn )n=0 be sequences converging to A, B ∈ R, respectively.

1. If A < B, then there exists N ∈ N such that xn < yn for all n ≥ N .

2. If there exists N ∈ N such that xn ≤ yn for all n ≥ N , then A ≤ B.

Version: December 30, 2025. 62

 Chapter 2.5 Sequences of Real Numbers

Proof. We first prove (1). Let ε = 13 (B − A) > 0, so that A + ε < B − ε. We know that there
exist NA , NB ∈ N such that

n ≥ NA ⇒ |xn − A| < ε, n ≥ NB ⇒ |yn − B| < ε.

Thus, for N = max{NA , NB }, we have

xn < A + ε < B − ε < yn ∀n ≥ N,

proving (1).
For (2), suppose by contradiction that A > B. Then (1) implies the existence of N0 such
that xn > yn for all n ≥ N0 , contradicting the assumption xn ≤ yn for large n.

Remark 2.96. — In Proposition 2.95(2), even if xn < yn for all n, one cannot conclude
A < B. For instance, take
xn = − n1 , yn = n1 .
9 Then xn < yn for all n, but both sequences converge to 0.

Lemma 2.97: Sandwich Lemma

Let (xn )∞ ∞ ∞
n=0 , (yn )n=0 , (zn )n=0 be sequences such that, for some N ∈ N,

xn ≤ yn ≤ zn ∀ n ≥ N.

Suppose that (xn )∞ ∞ ∞

n=0 and (zn )n=0 both converge to the same limit. Then (yn )n=0 also
converges, and
lim xn = lim yn = lim zn .
n→∞ n→∞ n→∞

Proof. The proof is left as an exercise.

Exercise 2.98. — Compute the following limits, if they exist:

7n4 + 15 n2 + 5 n5 − 10
lim , lim , lim .
n→∞ 3n4 + n3 + n − 1 n→∞ n3 + n + 1 n→∞ n2 + 1

Version: December 30, 2025. 63

 Chapter 2.5 Sequences of Real Numbers

2.5.4 Bounded and Monotone Sequences

Definition 2.99: Bounded Sequences

A sequence (xn )∞
n=0 in R is called bounded if there exists a real number M ≥ 0 such
that
|xn | ≤ M ∀ n ∈ N.

Lemma 2.100: Convergent Sequences are Bounded

Every convergent sequence is bounded.

Proof. Let (xn )∞n=0 be a convergent sequence with limit A ∈ R. Choosing ε = 1 in the
definition of limit, there exists N ∈ N such that |xn − A| ≤ 1 for all n ≥ N . By the triangle
inequality (see Lemma 2.19(g)),

|xn | = |(xn − A) + A| ≤ |xn − A| + |A| ≤ 1 + |A| ∀ n ≥ N.

10
Set
M = max{1 + |A|, |x0 |, |x1 |, . . . , |xN −1 |}.

Then |xn | ≤ M for all n ∈ N, as required.

Exercise 2.101. — Prove statement (2) in Proposition 2.94.

Hint: Note that

|xn yn − AB| = |xn yn − xn B + xn B − AB| ≤ |yn − B| |xn | + |xn − A| |B|.

Since (xn )∞
n=0 converges, Lemma 2.100 guarantees that |xn | ≤ M for some M ∈ R. Use this
to conclude.

We will see later that every bounded sequence has at least one accumulation point, or
equivalently, a convergent subsequence. Before that, we introduce the notion of monotonicity,
which will allow us to identify a large class of automatically convergent sequences.

Version: December 30, 2025. 64

 Chapter 2.5 Sequences of Real Numbers

Definition 2.102: Monotone Sequences

A sequence (xn )∞
n=0 is called:

• (monotonically) increasing if m > n =⇒ xm ≥ xn ,

• strictly (monotonically) increasing if m > n =⇒ xm > xn ,

• (monotonically) decreasing if m > n =⇒ xm ≤ xn ,

• strictly (monotonically) decreasing if m > n =⇒ xm < xn .

If a sequence is increasing or decreasing, we call it monotone; if it is strictly increasing

or strictly decreasing, we call it strictly monotone.

Remark 2.103. — An equivalent formulation of motonone sequences can be given using

only successive terms:

• (xn )∞
n=0 is increasing if xn+1 ≥ xn for all n;

• (xn )∞
n=0 is strictly increasing if xn+1 > xn for all n;

• (xn )∞
n=0 is decreasing if xn+1 ≤ xn for all n;

10 • (xn )∞
n=0 is strictly decreasing if xn+1 < xn for all n.

Monotone bounded sequences are always convergent. This is proved in the next theorem
and illustrated in Figure 2.11.

xn → S
x0 x1 x2 x3 x4

S M

Figure 2.11: A monotonically increasing sequence (xn )∞n=0 represented along a single axis.
Each term xn is shown as a vertical segment; the sequence moves rightward, approaching the
supremum S without ever reaching the upper bound M . The limit of the sequence equals S.

Theorem 2.104: Convergence of Monotone Sequences

A monotone sequence (xn )∞n=0 converges if and only if it is bounded.

More precisely, let X = {x0 , x1 , x2 , x3 , . . .} ⊂ R denote the set of points in the sequence.

If (xn )∞
n=0 is increasing, then lim xn = sup(X);
n→∞

If (xn )∞
n=0 is decreasing, then lim xn = inf(X).
n→∞

Proof. If (xn )∞
n=0 converges, Lemma 2.100 shows that it is bounded.

Version: December 30, 2025. 65

 Chapter 2.5 Sequences of Real Numbers

Conversely, let (xn )∞ n=0 be a bounded monotone sequence. Without loss of generality,
assume it is increasing (otherwise consider (−xn )∞ n=0 ). Since (xn )n=0 is bounded from above,
∞

the set X = {x0 , x1 , x2 , x3 , . . .} has a supremum, say A = sup(X).

By definition of A:

(i) xn ≤ A for all n;

(ii) for every ε > 0 there exists N ∈ N such that xN > A − ε.

Then, for all n ≥ N , using (ii) and monotonicity we deduce that xn ≥ xN > A−ε. Combining
this with (i), we conclude that

A − ε < xn < A + ε for all n ≥ N ,

proving limn→∞ xn = A.

Remark 2.105. — If (xn )∞ n=0 is monotone and admits a bounded subsequence (xnk )k=0 ,
∞

then the whole sequence is bounded and, therefore, convergent by Theorem 2.104.
Indeed, assume for instance that (xn )∞
n=0 is increasing and (xnk )k=0 is bounded above by
∞

M . Then, by monotonicity and Remark 2.85,

10
x0 ≤ xk ≤ xnk ≤ M ∀ k ∈ N,

so (xn )∞
n=0 is bounded. The case of a decreasing sequence is analogous.

Exercise 2.106. — Let (xn )∞

n=0 be defined by x0 = 1 and
 
2 1
xn = xn−1 + for n ≥ 1.
3 xn−1

Show that (xn )∞ n=0 converges and determine its limit.

Hint: First, prove that the sequence converges to a nonnegative limit A. Then, show that the
limit satisfies the relation A = 23 A + A1 and use it to identify A.

Exercise 2.107. — Let (xn )∞ n=0 be monotonically increasing and (yn )n=0 monotonically
∞

decreasing, with xn ≤ yn for all n ∈ N. Show that both sequences converge and that

lim xn ≤ lim yn .
n→∞ n→∞

Illustrate your reasoning with a diagram similar to Figure 2.11.

Version: December 30, 2025. 66

 Chapter 2.5 Sequences of Real Numbers

2.5.5 Superior and Inferior Limits

Let (xn )∞ n=0 be a bounded sequence. To study its behaviour for large n, it is useful to consider
its tails:
X≥n = {xn , xn+1 , xn+2 , . . .} = {xk | k ≥ n} ⊂ R.

The concept of limit can be restated in terms of these tails: the sequence (xn )∞
n=0 converges
to A if and only if, for every ε > 0, there exists N ∈ N such that

X≥N ⊂ (A − ε, A + ε).

However, since not every sequence has a limit, we now introduce a related notion (the superior
and inferior limits), which always exist for bounded sequences.
For each n ∈ N, define

sn = sup(X≥n ) = sup xk , in = inf(X≥n ) = inf xk .

k≥n k≥n

Since X≥m ⊂ X≥n whenever m > n, we have

10 in ≤ im ≤ sm ≤ sn for all m > n.

Thus, (sn )∞
n=0 is a monotonically decreasing sequence, while (in )n=0 is monotonically increas-
∞

ing. Moreover, since (xn )∞ n=0 is bounded, both (sn )n=0 and (in )n=0 are bounded as well.
∞ ∞

Hence, by Theorem 2.104, both sequences converge. Their limits will be called the superior
limit and the inferior limit of (xn )∞
n=0 , respectively.

Remark 2.108. — In the sequel, we may use the shorthand

sup xk for sup{xk | k ≥ n}, inf xk for inf{xk | k ≥ n}.

k≥n k≥n

That is, the index range “k ≥ n” indicates that we are taking the supremum or infimum over
all terms of the sequence with index greater than or equal to n. With this notation,

sn = sup xk , in = inf xk .
k≥n k≥n

Note that, since xn ∈ X≥n ,

in ≤ xn ≤ sn ∀ n ∈ N. (2.10)

Version: December 30, 2025. 67

 Chapter 2.5 Sequences of Real Numbers

Definition 2.109: Superior and Inferior Limits

Let (xn )∞
n=0 be a bounded sequence in R. The numbers
   
lim sup xn = lim sup xk , lim inf xn = lim inf xk
n→∞ n→∞ k≥n n→∞ n→∞ k≥n

are called the superior limit and inferior limit of (xn )∞

n=0 , respectively.
From (2.10) and Proposition 2.95, we have

lim inf xn ≤ lim sup xn .

n→∞ n→∞

Example 2.110. — Let (xn )∞ n=1 be defined by xn = (−1) + n . We compute the corre-
n 1

sponding sequences (sn ) and (in ):

n 1 2 3 4 5 6 7 8 ...
3
xn 0 2 − 23 5
4... − 45 7
6 − 67 9
8
3 3 5 5 7 7 9 9
sn 2 2 4 4... 6 6 8 8
in −1 −1 −1 −1 −1 −1 −1 −1 . . .

Here sn = xn for even n and sn = xn+1 for odd n, hence

10    
lim sup xn = lim (−1)2n + 1
2n = lim 1 + 1
2n = 1.
n→∞ n→∞ n→∞

Instead, since xn ≥ −1 and limn→∞ x2n+1 = −1, we get in = −1 for all n ≥ 1, therefore
lim inf n→∞ xn = −1.

in s5 s3 s1

−1 0 1
x5 x3 x1 x6 x4 x2

Figure 2.12: Representation of the sequence xn = (−1)n + n1 . Green bars below the axis show
the first few values of xn : even indices approach 1, odd indices approach −1. Blue bars above
the axis illustrate the superior bounds sn = supk≥n xk : here s1 (= s2 ), s3 (= s4 ), and s5 (= s6 )
are labeled, while the unlabeled bars suggest the continued decrease toward 1. The red bar
at −1 marks in = inf k≥n xk = −1 for all n.

Lemma 2.111: Convergence and Superior/Inferior Limits

A bounded sequence (xn )∞

n=0 in R converges if and only if

lim sup xn = lim inf xn .

n→∞ n→∞

Version: December 30, 2025. 68

 Chapter 2.5 Sequences of Real Numbers

Proof. For every n ∈ N, define

in = inf xk , sn = sup xk ,
k≥n k≥n

and set
I = lim in = lim inf xn , S = lim sn = lim sup xn .
n→∞ n→∞ n→∞ n→∞

First, suppose that I = S. Since in ≤ xn ≤ sn (see (2.10)), the Sandwich Lemma 2.97
implies that the sequence (xn )∞
n=0 converges, and its limit equals I = S.

Conversely, assume that (xn )∞

n=0 converges to A ∈ R. Given ε > 0, there exists N ∈ N
such that
A − ε < xn < A + ε ∀ n ≥ N.

Then, for all n ≥ N , the same inequalities holds for sn and in :

A − ε ≤ in ≤ sn ≤ A + ε.

Taking limits and using Proposition 2.95, we obtain

A − ε ≤ I ≤ S ≤ A + ε.

10 Since ε > 0 is arbitrary, it follows that A = I = S, which proves the result.

Theorem 2.112: Superior and Inferior Limits are Accumulation Points

Let (xn )∞
n=0 be a bounded sequence and let

A = lim sup xn .
n→∞

Then A is an accumulation point of (xn )∞

n=0 , and for every ε > 0 the following hold:

(1) only finitely many elements satisfy xn ≥ A + ε;

(2) infinitely many elements satisfy A − ε < xn < A + ε.

An analogous statement holds for the inferior limit.

Proof. Recall the notation

sn = sup xk , n ∈ N.
k≥n

Since the sequence (sn )∞

n=0 is monotonically decreasing and converges to A, given ε > 0 there
exists N0 ∈ N such that
A ≤ sn < A + ε ∀ n ≥ N0 . (2.11)

We first prove that A is an accumulation point.

Version: December 30, 2025. 69

 Chapter 2.5 Sequences of Real Numbers

Fix N ∈ N and set N1 = max{N, N0 }. Since sN1 = supk≥N1 xk , there exists n1 ≥ N1 ≥ N0

such that
sN1 − ε < xn1 ≤ sN1 .

Thus, combining this bound with (2.11) we obtain

A − ε ≤ sN1 − ε < xn1 ≤ sN1 < A + ε.

This construction shows that for any ε > 0 and any N ∈ N, there exists n1 ≥ N such that
A − ε < xn1 < A + ε. Thus A is an accumulation point of (xn )∞
n=0 .

We now prove (1) and (2). From (2.11) we have xn < A + ε for all n ≥ N0 , so only finitely
many terms satisfy xn ≥ A + ε. This shows (1).
Also, since A is an accumulation point, it follows from Corollary 2.89 that infinitely many
terms of the sequence lie within any interval (A − ε, A + ε).

Corollary 2.113: Bounded Sequences have Convergent Subsequences

Every bounded sequence has at least one accumulation point and therefore possesses a
convergent subsequence.

Proof. By Theorem 2.112, the number

10 A = lim sup xn
n→∞

is always an accumulation point of (xn )∞ n=0 . Moreover, by Proposition 2.88, every accumula-
tion point is the limit of a convergent subsequence. Hence every bounded sequence admits at
least one convergent subsequence.

Exercise 2.114. — Let (xn )∞ n=0 be a bounded sequence in R, and let E ⊆ R be the set of
accumulation points of (xn )n=0 . Show that
∞

lim sup xn = max(E), lim inf xn = min(E).

n→∞ n→∞

Exercise 2.115. — Let (an )∞ n=0 , (bn )n=0 , (cn )n=0 be convergent sequences with limits
∞ ∞

A, B, C ∈ R, respectively. Define a new sequence (xn )∞ n=0 by




an , if n = 3k, k ∈ N,


xn = bn , if n = 3k + 1, k ∈ N,



cn , if n = 3k + 2, k ∈ N.



Compute lim sup xn , lim inf xn , and describe the set of accumulation points of (xn )∞
n=0 .
n→∞ n→∞

Version: December 30, 2025. 70

 Chapter 2.5 Sequences of Real Numbers

Exercise 2.116. — Let (xn )∞

n=0 be a bounded sequence such that (xn+1 −xn )n=0 converges
∞

to 0. Set
10 A = lim inf xn , B = lim sup xn .
n→∞ n→∞

Show that the set of accumulation points of (xn )∞

n=0 is the interval [A, B]. Construct an
example of such a sequence with A = 0 and B = 1.

2.5.6 Cauchy Sequences

Definition 2.117: Cauchy Sequences

A sequence (xn )∞
n=0 in R is called a Cauchy sequence if for every ε > 0 there exists
N ∈ N such that
|xn − xm | < ε ∀ m, n ≥ N.

Lemma 2.118: Cauchy Sequences are Bounded

Every Cauchy sequence is bounded.

Proof. By definition, there exists N ∈ N such that

|xn − xN | ≤ 1 ∀ n ≥ N.

11 Hence, for n ≥ N , we have |xn | ≤ |xN | + 1.

Now, define
M = max{ |x0 |, |x1 |, . . . , |xN −1 |, |xN | + 1 }.

Then |xn | ≤ M for all n ∈ N, so (xn )∞

n=0 is bounded.

n=0 converges if and only if it has

Exercise 2.119. — Show that a Cauchy sequence (xn )∞
a convergent subsequence.

Theorem 2.120: Convergence and Cauchy Sequences

A sequence (xn )∞
n=0 of real numbers converges if and only if it is a Cauchy sequence.

Proof. Suppose first that (xn )∞

n=0 converges to some A ∈ R, and let us prove that (xn )n=0 is
∞

a Cauchy sequence. Given ε > 0, choose N ∈ N such that

ε
|xn − A| < 2 ∀ n ≥ N.

Version: December 30, 2025. 71

 Chapter 2.5 Sequences of Real Numbers

Then for all m, n ≥ N ,

ε ε
|xn − xm | ≤ |xn − A| + |xm − A| < 2 + 2 = ε,

hence (xn )∞
n=0 is a Cauchy sequence.

Viceversa, let (xn )∞

n=0 be a Cauchy sequence. Since it is bounded (by Lemma 2.118),

k=0 converging to some limit

Corollary 2.113 implies that there exists a subsequence (xnk )∞
A ∈ R.
Given ε > 0, choose N0 ∈ N such that

ε
|xn − xm | < 2 ∀ m, n ≥ N0 ,

and choose N1 ∈ N such that

ε
|xnk − A| < 2 ∀ k ≥ N1 .

Let N = max{N0 , N1 }. Since nN ≥ N (see Remark 2.85), for all n ≥ N we have

ε ε
|xn − A| ≤ |xn − xnN | + |xnN − A| < 2 + 2 = ε.
11
Thus (xn )∞
n=0 converges to A.

Example 2.121. — Consider the condition

∀ ε > 0 ∃ N ∈ N such that |xn+1 − xn | < ε ∀ n ≥ N,

which is equivalent to saying that limn→∞ |xn+1 − xn | = 0. We note that this condition is not
equivalent to convergence. Indeed, consider the sequence

0, 1, 1 + 21 , 2, 2 + 31 , 2 + 32 , 3, 3 + 14 , 3 + 42 , 3 + 34 , 4, 4 + 51 , 4 + 25 , 4 + 35 , 4 + 45 , 5, 5 + 16 , . . .

which progresses between consecutive integers n − 1 and n in steps of size n1 . This sequence
is unbounded and hence not convergent, but the distances between successive elements tend
to zero. Therefore, the condition limn→∞ |xn+1 − xn | = 0 is insufficient for convergence.

2.5.7 Improper Limits

We now extend the notion of limit to allow the improper limit values +∞ (often abbreviated
as ∞) and −∞.

Version: December 30, 2025. 72

 Chapter 2.5 Sequences of Real Numbers

Definition 2.122: Improper Limits

Let (xn )∞
n=0 be a sequence in R.

We say that (xn )∞

n=0 diverges to +∞, and we write

lim xn = +∞,
n→∞

if for every M > 0 there exists N ∈ N such that xn > M for all n ≥ N .
Similarly, (xn )∞
n=0 diverges to −∞ if for every M > 0 there exists N ∈ N such that
xn < −M for all n ≥ N .
In both cases, we say that (xn )∞
n=0 has an improper limit.

An unbounded sequence need not diverge to +∞ or −∞. For instance, the sequence

0, −1, 2, −3, 4, −5, 6, −7, 8, −9, . . . ,

that is, xn = (−1)n n, is unbounded but neither diverges to +∞ nor to −∞.

Exercise 2.123. — Let (xn )∞ n=0 be an unbounded sequence of real numbers. Show that
there exists a subsequence which diverges either to +∞ or to −∞.
11 The notion of improper limit allows us to extend the definitions of superior and inferior
limits to unbounded sequences. If (xn )∞
n=0 is not bounded from above, then

sup xk = +∞ ∀ n ∈ N,
k≥n

and we write
lim sup xn = +∞.
n→∞

If (xn )∞
n=0 is bounded from above but not from below, then we define

lim sup xn = lim sup xk ,
n→∞ n→∞ k≥n

where the right-hand side is a real limit if the decreasing sequence supk≥n xk is bounded, and
the improper limit −∞ otherwise. The definition of lim inf extends analogously.

Exercise 2.124. — (Prove the sandwich principle for improper limits.) Let (xn )∞
n=0 and
(yn )n=0 be two sequences with xn ≤ yn for all n ∈ N. Show that:
∞

lim xn = +∞ =⇒ lim yn = +∞,

n→∞ n→∞

lim yn = −∞ =⇒ lim xn = −∞.

n→∞ n→∞

Version: December 30, 2025. 73

 Chapter 2.6 Sequences of Real Numbers

2.6 Sequences of Complex Numbers

Informally, a sequence of complex numbers is just like a sequence of real numbers, except
that each term is a complex number instead of a real one. Thus, we study ordered lists

z0 , z1 , z2 , . . .

where each element zn belongs to C. As in the real case, we are mainly interested in their
convergence, divergence, and limit behavior.
To analyze sequences in C, it is often sufficient to consider separately the corresponding
sequences of real and imaginary parts in R.

Definition 2.125: Sequences of Complex Numbers

A sequence of complex numbers (zn )∞

n=0 , where

zn = xn + iyn ,

is said to converge to a limit A + iB ∈ C if the two sequences of real numbers (xn )∞

n=0
and (yn )∞n=0 converge to A and B, respectively. In this case, we write

11 lim zn = A + iB.
n→∞

n=0 diverges to ∞ if the sequence of moduli (|zn |)n=0 diverges to

We say that (zn )∞ ∞

+∞, that is,

p
lim |zn | = lim x2n + yn2 = +∞.
n→∞ n→∞

Remark 2.126. — As for sequences of real numbers, one can consider subsequences of
sequences in C. Given a strictly increasing sequence of nonnegative integers (nk )∞
k=0 , the
corresponding subsequence is

(znk )∞ ∞
k=0 = (xnk + iynk )k=0 .

Exercise 2.127. — Let (zn )∞ n=0 be a convergent sequence in C. Show that the sequence
of moduli (|zn |)n=0 converges, and determine its limit. Conversely, does the convergence of
∞

n=0 imply the convergence of (zn )n=0 ?

(|zn |)∞ ∞

Exercise 2.128. — Given a complex number z ∈ C, consider the geometric sequence

n=0 defined by
(zn )∞
zn = z n = |z · .{z
. . · z} .
n times

Version: December 30, 2025. 74

 Chapter 2.6 Sequences of Real Numbers

Determine the set of all complex numbers z for which the sequence (z n )∞
n=0 converges.

Remark 2.129. — Most of the properties of convergent sequences in R remain valid for
11
complex sequences. In particular, limits in C are unique, and all standard limit rules (linearity,
product, quotient, etc.) hold. Each statement can be proved by applying the corresponding
property separately to the real and imaginary parts.

Version: December 30, 2025. 75

 Chapter 3

Functions of one Real Variable

In this chapter we study real-valued functions defined on subsets of R, typically intervals. The
central concept is continuity.

3.1 Real-valued Functions

3.1.1 Boundedness and Monotonicity

We begin with two elementary properties already encountered for sequences: boundedness
and monotonicity. A real-valued function is any function with values in R. We assume the
domain is a nonempty subset of R; informally, we speak of functions of one real variable.
For a nonempty set D ⊆ R, the set of real-valued functions on D is

F(D) = {f | f : D → R} .

For f1 , f2 ∈ F(D), α ∈ R, and x ∈ D we define

(f1 + f2 )(x) = f1 (x) + f2 (x), (αf1 )(x) = α f1 (x), (f1 f2 )(x) = f1 (x)f2 (x).
11
Given a ∈ R, we write f ≡ a for the constant function x 7→ a on D.

Remark 3.1. — With the operations above, F(D) is a commutative ring (the additive
identity is f ≡ 0 and the multiplicative identity is f ≡ 1).

A point x ∈ D is a zero of f ∈ F(D) if f (x) = 0. The zero set of f is {x ∈ D | f (x) = 0}.

We order F(D) pointwise: for f1 , f2 ∈ F(D),

f1 ≤ f2 ⇐⇒ f1 (x) ≤ f2 (x) ∀ x ∈ D,

f1 < f2 ⇐⇒ f1 (x) < f2 (x) ∀ x ∈ D.

We say that f ∈ F(D) is nonnegative if f ≥ 0, and positive if f > 0.

76
 Chapter 3.1 Real-valued Functions

Exercise 3.2. — Let N1 , N2 ⊆ D be the zero sets of f1 , f2 ∈ F(D), respectively. What is

the zero set of f1 f2 ?

Exercise 3.3. — Verify that the relation ≤ defined above on F(D) is an order relation.

Definition 3.4: Bounded Functions

Let D ̸= ∅ and f : D → R. We say that f is bounded from above if there exists
M > 0 such that
f (x) ≤ M ∀ x ∈ D.

We say that f is bounded from below if there exists M > 0 such that

f (x) ≥ −M ∀ x ∈ D.

We say that f is bounded if it is both bounded from above and from below. Equiva-
lently, f is bounded if there exists M > 0 such that

|f (x)| ≤ M ∀ x ∈ D.

Definition 3.5: Monotone Functions

11 Let D ⊆ R and f : D → R. The function f is:

1. increasing if x < y implies f (x) ≤ f (y) for all x, y ∈ D;

2. strictly increasing if x < y implies f (x) < f (y) for all x, y ∈ D;

3. decreasing if x < y implies f (x) ≥ f (y) for all x, y ∈ D;

4. strictly decreasing if x < y implies f (x) > f (y) for all x, y ∈ D.

We call f monotone if it is increasing or decreasing, and strictly monotone if it is

strictly increasing or strictly decreasing.

Example 3.6. — • Let D = [a, b] and f (x) = x2 . Then f is strictly increasing if a ≥ 0,

strictly decreasing if b ≤ 0, and not monotone if a < 0 < b.

• For any D ⊆ R and any odd integer n ≥ 1, the map x 7→ xn on D is strictly increasing.

• The rounding function ⌊·⌋ : R → R (see Definition 2.59) is increasing but not strictly
increasing.

• A constant function is both increasing and decreasing. Conversely, a function on D ⊆ R

that is both increasing and decreasing is constant.

Version: December 30, 2025. 77

 Chapter 3.1 Real-valued Functions

−1

11
Figure 3.1: A strictly monotone function is always injective but need not be surjective. For
example, f : R → R, f (x) = 18 x + sgn(x), is strictly increasing but not surjective (e.g. 12 is
not in the image).

Exercise 3.7. — Let D ⊆ R, and let f1 , f2 ∈ F(D) be strictly increasing. Show that:

(i) f1 + f2 is strictly increasing;

(ii) for a ∈ R, the function af1 is strictly increasing if a > 0, and strictly decreasing if a < 0;

(iii) if f1 > 0 and f2 > 0, then f1 f2 is strictly increasing.

3.1.2 Continuity

Definition 3.8: Continuous Functions

Let D ⊆ R and f : D → R. We say that f is continuous at x0 ∈ D if for all ε > 0
there exists δ > 0 such that

∀ x ∈ D, |x − x0 | < δ =⇒ |f (x) − f (x0 )| < ε.

We say that f is continuous on D if it is continuous at every point of D.

12

Remark 3.9. — It suffices to verify the implication above for small ε. Precisely:
Assume there exists ε0 > 0 such that for every ε ∈ (0, ε0 ] there is a δ > 0 with

|x − x0 | < δ =⇒ |f (x) − f (x0 )| < ε (∀ x ∈ D).

Then f is continuous at x0 .
Indeed, for ε > ε0 we can choose the number δ > 0 corresponding to ε0 to get

∀ x ∈ D, |x − x0 | < δ =⇒ |f (x) − f (x0 )| < ε0 < ε.

Version: December 30, 2025. 78

 Chapter 3.1 Real-valued Functions

In other words, if δ works for ε0 , then it works for all ε > ε0 .

The next figure shows a continuous function on D = [a, b) ∪ (c, d] ∪ {e}. For every x0 ∈ D
and every ε > 0, there exists δ > 0 such that all x ∈ D with |x−x0 | < δ satisfy |f (x)−f (x0 )| <
ε.

f
f (x0 ) + ε
f (x0 )
f (x0 ) − ε

a x0 − δ x0 x0 + δ b c d e

Figure 3.2: A continuous function with domain D = [a, b) ∪ (c, d] ∪ {e}.

Example 3.10. — • For a, b ∈ R, the affine function f (x) = ax + b is continuous.

Indeed, for x0 ∈ R and ε > 0, set δ = ε/|a| (if a = 0 then f is constant and we can
simply choose δ = 1). Then |x − x0 | < δ implies

|f (x) − f (x0 )| = |(ax + b) − (ax0 + b)| = |a(x − x0 )| = |a||x − x0 | < ε.

12

• The absolute value f (x) = |x| is continuous. Indeed, give ε > 0 simply choose δ = ε.
Then |x − x0 | < δ implies |f (x) − f (x0 )| = ||x| − |x0 || ≤ |x − x0 | < ε by inverse triangle
inequality.

• The rounding function f (x) = ⌊x⌋ is not continuous at integers. Indeed, if x0 ∈ Z, then
for any small δ > 0,
⌊x0 − 2δ ⌋ − ⌊x0 ⌋ = 1,

so the ε–δ condition fails for ε < 1.

f (x) = ⌊x⌋
f (x) = |x| 1

Figure 3.3: The absolute value function (left) and the floor function (right).

Version: December 30, 2025. 79

 Chapter 3.1 Real-valued Functions

√
Exercise 3.11. — Show that f : R → R, f (x) = x2 , and g : [0, ∞) → R, g(x) = x, are
both continuous.

Restriction
Let D ⊆ R and f : D → R. For any D′ ⊆ D the restriction of f to D′ is the function
f |D′ : D′ → R defined by

f |D′ (x) = f (x) ∀ x ∈ D′ .

We regard f |D′ and f as different functions unless D′ = D.

Exercise 3.12. — Let D ⊆ R and f : D → R be continuous. If D′ ⊆ D, show that f |D′

is continuous.

Proposition 3.13: Combination of Continuous Functions

Let D ⊆ R, and let f1 , f2 : D → R be continuous at x0 ∈ D. Then f1 + f2 , f1 f2 , and
af1 (for any a ∈ R) are continuous at x0 .

Proof. We first prove the result for the sum. Let ε > 0. Since f1 and f2 are continuous at x0 ,
there exist δ1 , δ2 > 0 such that
12
|x − x0 | < δ1 ⇒ |f1 (x) − f1 (x0 )| < 2ε , |x − x0 | < δ2 ⇒ |f2 (x) − f2 (x0 )| < 2ε .

So, choosing δ = min{δ1 , δ2 }, for |x − x0 | < δ we get

|(f1 + f2 )(x) − (f1 + f2 )(x0 )| ≤ |f1 (x) − f1 (x0 )| + |f2 (x) − f2 (x0 )| < ε,

which shows that f1 + f2 is continuous at x0 .

For the product, note that

|f1 (x)f2 (x) − f1 (x0 )f2 (x0 )| = |f1 (x)f2 (x) − f1 (x0 )f2 (x) + f1 (x0 )f2 (x) − f1 (x0 )f2 (x0 )|
≤ |f1 (x)f2 (x) − f1 (x0 )f2 (x)| + |f1 (x0 )f2 (x) − f1 (x0 )f2 (x0 )|
= |f2 (x)| |f1 (x) − f1 (x0 )| + |f1 (x0 )| |f2 (x) − f2 (x0 )|.

Now, first choose δ0 > 0 so that |x − x0 | < δ0 implies |f2 (x) − f2 (x0 )| < 1, so that

|x − x0 | < δ0 ⇒ |f2 (x)| < 1 + |f2 (x0 )|.

Then choose δ1 , δ2 > 0 so that

ε
|x − x0 | < δ1 ⇒ |f1 (x) − f1 (x0 )| < ,
2(1 + |f2 (x0 )|)

Version: December 30, 2025. 80

 Chapter 3.1 Real-valued Functions

ε
|x − x0 | < δ2 ⇒ |f2 (x) − f2 (x0 )| < .
2(1 + |f1 (x0 )|)
So, choosing δ = min{δ0 , δ1 , δ2 }, for |x − x0 | < δ we get

ε ε
|f1 (x)f2 (x) − f1 (x0 )f2 (x0 )| < |f2 (x)| + |f1 (x0 )|
2(1 + |f2 (x0 )|) 2(1 + |f1 (x0 )|)
ε ε
< (1 + |f2 (x0 )|) + |f1 (x0 )|
2(1 + |f2 (x0 )|) 2(1 + |f1 (x0 )|)
ε ε
< + = ε,
2 2

thus f1 f2 is continuous at x0 .
Finally, the statement about af1 follows by choosing f2 ≡ a (a constant function) and
using the product case proved above: since f1 and f2 are continuous at x0 , their product
f1 f2 = af1 is continuous at x0 .

Sum and Product Notation

Let n ∈ N and a0 , a1 , . . . , an ∈ R. We use the notation
n
X n
Y
aj = a0 + a1 + a2 + · · · + an , aj = a0 a1 a2 · · · an .
j=0 j=0
12
Here aj is a summand in the sum and a factor in the product; j is the index (or
running variable).
If J is a finite set and numbers (aj )j∈J are given, we write
X Y
aj and aj .
j∈J j∈J

By convention, for the empty index set ∅,

X Y
aj = 0, aj = 1.
j∈∅ j∈∅

Example 3.14. — A polynomial is a function of the form

n
X
p(x) = a0 + a1 x + · · · + an xn = aj xj , aj ∈ R.
j=0

Since x 7→ x is continuous and products/sums of continuous functions are continuous (Propo-

sition 3.13), every polynomial is continuous.

Version: December 30, 2025. 81

 Chapter 3.1 Real-valued Functions

Composition of Functions
Let f : X → Y and g : Y → Z. The composition is g ◦ f : X → Z, defined by
(g ◦ f )(x) = g(f (x)) for all x ∈ X.
Associativity. If f : W → X, g : X → Y , and h : Y → Z, then

h ◦ (g ◦ f ) = (h ◦ g) ◦ f.

Indeed, for all w ∈ W ,

h ◦ (g ◦ f )(w) = h (g ◦ f )(w) = h g(f (w)) = (h ◦ g) f (w) = (h ◦ g) ◦ f (w).

Therefore, we may omit parentheses and write h ◦ g ◦ f : W → Z.

Proposition 3.15: Composition of Continuous Functions

Let D1 , D2 ⊆ R, x0 ∈ D1 , and f : D1 → D2 be continuous at x0 . If g : D2 → R
is continuous at f (x0 ), then g ◦ f : D1 → R is continuous at x0 . In particular, the
composition of continuous functions is continuous.

Proof. Let ε > 0. By continuity of g at f (x0 ), there exists η > 0 such that
12
∀ y ∈ D2 , |y − f (x0 )| < η ⇒ |g(y) − g(f (x0 ))| < ε.

By continuity of f at x0 , there exists δ > 0 such that

∀ x ∈ D1 , |x − x0 | < δ ⇒ |f (x) − f (x0 )| < η.

Combining the implications gives, for x ∈ D1 ,

|x − x0 | < δ =⇒ |f (x) − f (x0 )| < η =⇒ |g(f (x)) − g(f (x0 ))| < ε.

Remark 3.16. — Applying Proposition 3.15 with g(y) = |y| (see Example 3.10), we see
that if f : D → R is continuous then x 7→ |f (x)| is continuous.

Exercise 3.17. — Show that f : R \ {0} → R, x 7→ x1 , is continuous. Deduce that if

g : D → R is continuous and has no zeros on D, then x 7→ g(x)
1
is continuous on D. Conclude
h(x)
that x 7→ g(x) is continuous on D whenever h, g : D → R are continuous and g has no zeros
on D.

Version: December 30, 2025. 82

 Chapter 3.1 Real-valued Functions

Exercise 3.18. — Let a < b < c and let f1 : [a, b] → R, f2 : [b, c] → R be continuous.
Define f : [a, c] → R by 
f (x), x ∈ [a, b),
1
f (x) =
f (x), x ∈ [b, c].
2

Show that f is continuous if and only if f1 (b) = f2 (b).

Exercise 3.19. — Let I ⊂ R be an open interval and let f : I → R be a function. Show

that f is continuous if and only if f −1 (U ) is open for every open set U ⊂ R.

Exercise 3.20. — Let D ⊆ R and f : D → R.

1. If f is continuous at x0 ∈ D, then there exist an open neighbourhood U of x0 and M > 0

such that |f (x)| ≤ M for all x ∈ D ∩ U .

2. If f is continuous at x0 ∈ D and f (x0 ) ̸= 0, then there exists an open neighbourhood U

of x0 such that f (x)f (x0 ) > 0 for all x ∈ D ∩ U (in other words, f (x) and f (x0 ) have
the same sign).

3.1.3 Sequential Continuity

12
Continuity admits a sequential characterization: a function f : D → R is continuous if and
only if it sends every convergent sequence in D to a convergent sequence with the corresponding
limit. This is called sequential continuity. We first fix a notation.

Notation for Limits of Sequences

Let (xn )∞
n=0 ⊆ R and x̄ ∈ R. We write

xn → x̄ or xn −−−→ x̄
n→∞

to mean
lim xn = x̄.
n→∞

Theorem 3.21: Continuity = Sequential Continuity

Let D ⊆ R, f : D → R, and x̄ ∈ D. Then f is continuous at x̄ if and only if for every

sequence (xn )∞
n=0 ⊂ D with xn → x̄ we have f (xn ) → f (x̄).

Proof. Assume that f is continuous at x̄. Then, given ε > 0, there exists δ > 0 such that

∀ x ∈ D, |x − x̄| < δ ⇒ |f (x) − f (x̄)| < ε.

Version: December 30, 2025. 83

 Chapter 3.1 Real-valued Functions

Also, since xn → x̄, there exists N ∈ N such that

n ≥ N =⇒ |xn − x̄| < δ.

Thus
n ≥ N =⇒ |f (xn ) − f (x̄)| < ε,

which implies that the sequence (f (xn ))∞

n=0 converges to f (x̄).

To prove the converse, assume that f is not continuous at x̄. This means that there exists
ε > 0 such that, for every δ > 0, there is x ∈ D with

|x − x̄| < δ and |f (x) − f (x̄)| ≥ ε.

12
Now, for every n ∈ N, we apply this property with δ = 2−n to find a point xn ∈ D such that

|xn − x̄| < 2−n and |f (xn ) − f (x̄)| ≥ ε.

Then the sequence constructed in this way satisfies xn → x̄ but f (xn ) ̸→ f (x̄).

Remark 3.22. — The proof above shows the following:

If f : D → R is not continuous at x̄, then there exists ε > 0 and a sequence (xn )∞
n=0 ⊂ D with
xn → x̄ such that |f (xn ) − f (x̄)| ≥ ε for all n ∈ N.

Exercise 3.23. — Let D ⊆ R and f : D → R be continuous. Suppose (xn )∞n=0 is a

sequence in D such that (f (xn ))n=0 converges. Must (xn )n=0 converge?
∞ ∞

Version: December 30, 2025. 84

 Chapter 3.2 Continuous Functions

3.2 Continuous Functions

3.2.1 The Intermediate Value Theorem

In this section we prove a fundamental theorem that formalises the idea that the graph of a
continuous function on an interval is a continuous curve, and thus cannot make any jumps.
We show that a continuous function f on an interval [a, b] contained in its domain takes all
intermediate values between f (a) and f (b). As we shall see, the proof relies on the existence
of the supremum, and therefore on the completeness axiom.

f (b)
(x̄, c)
y=c

f (a)
x
a x̄ b

Figure 3.4: A continuous function f : [a, b] → R must cross every horizontal line between f (a)
13 and f (b).

Theorem 3.24: Intermediate Value Theorem

Let f : [a, b] → R be a continuous function with f (a) ≤ f (b). Then, for every real
number c with f (a) ≤ c ≤ f (b), there exists x̄ ∈ [a, b] such that f (x̄) = c.

Proof. Fix c ∈ [f (a), f (b)], and define

X = {x ∈ [a, b] | f (x) ≤ c} .

Since a ∈ X and X ⊆ [a, b], the set X is nonempty and bounded from above. By Theorem 2.55,
its supremum
x̄ = sup(X) ∈ [a, b]

exists. We now use the continuity of f at x̄ to show that f (x̄) = c.

Since x̄ is the supremum of X, for each n ≥ 0 we can find a point xn ∈ X ∩ [x̄ − 2−n , x̄].
Then |xn − x̄| ≤ 2−n , hence xn → x̄. Also, by the definition of X, f (xn ) ≤ c. Thus, by
Theorem 3.21 (continuity of f along sequences),

f (x̄) = lim f (xn ),

n→∞

Version: December 30, 2025. 85

 Chapter 3.2 Continuous Functions

and Proposition 2.95 yields limn→∞ f (xn ) ≤ c. Therefore f (x̄) ≤ c.

Suppose, by contradiction, that f (x̄) < c and set ε := c − f (x̄) > 0. By continuity at x̄,
there exists δ > 0 such that for all x ∈ [a, b],

|x − x̄| < δ =⇒ |f (x) − f (x̄)| < ε,

hence f (x) < f (x̄) + ε = c. Therefore, by the definition of X,

(x̄ − δ, x̄ + δ) ∩ [a, b] ⊂ X.

Moreover, since f (x̄) < c ≤ f (b), we cannot have x̄ = b; hence x̄ < b.

Because x̄ < b, the interval (x̄, x̄ + δ) ∩ [a, b] ⊂ X is nonempty. Pick

y ∈ (x̄, x̄ + δ) ∩ [a, b] ⊂ X.

Then y ∈ X and y > x̄, which contradicts the defining property of the supremum: x̄ is an
upper bound of X, so X cannot contain elements larger than x̄. This contradiction shows
13 that f (x̄) ≥ c.
Together with f (x̄) ≤ c proved above, we conclude f (x̄) = c, as desired.

Remark 3.25. — If f : [a, b] → R is continuous with f (a) ≥ f (b), the theorem still holds
in the following form:

For every real number c with f (a) ≥ c ≥ f (b) there exists x̄ ∈ [a, b] such that f (x̄) = c.

To prove this, one can:

1. either repeat the previous proof with X = {x ∈ [a, b] | f (x) ≥ c};

2. or apply Theorem 3.24 to the function g = −f .

Exercise 3.26. — Let I be a non-empty interval and f : I → R a continuous injective

function. Show that f is strictly monotone.

Version: December 30, 2025. 86

 Chapter 3.2 Continuous Functions

3.2.2 Inverse Function Theorem

Identity and Inverse Function

Given a set X, the identity function idX : X → X is defined by

idX (x) = x for every x ∈ X.

If f : X → Y is bijective, there exists a unique function g : Y → X such that, for

each y ∈ Y , the value g(y) is the unique element x ∈ X satisfying f (x) = y. With this
definition,
g ◦ f = idX and f ◦ g = idY .

The function g is called the inverse function (or inverse mapping) of f , and is
denoted by f −1 .

Remark 3.27. — A function f : X → Y is bijective if and only if there exists a function

g : Y → X such that g ◦ f = idX and f ◦ g = idY .

In this subsection we show that every continuous strictly monotone function has an inverse
function that is also continuous.

13 y
y=x

(x, f (x))

(f (x), x)

f −1
x
f

Figure 3.5: The graphs of a strictly increasing function f and its inverse f −1 are reflections
across the line y = x. The points (x, f (x)) and (f (x), x) are mirror images.

Theorem 3.28: Inverse Function Theorem

Let I be an interval and f : I → R a continuous strictly monotone function. Then
f (I) is an interval, and the mapping f : I → f (I) has a continuous strictly monotone
inverse function f −1 : f (I) → I.

Proof. We may assume that I is non-empty and not a single point. Also, without loss of
generality, suppose f is strictly increasing (otherwise replace f with −f ).
Let J = f (I). Since f is strictly monotone, it is injective. Also, since by definition J =
f (I), it is surjective, hence bijective. Therefore there exists a unique inverse g = f −1 : J → I.

Version: December 30, 2025. 87

 Chapter 3.2 Continuous Functions

Because f is strictly increasing, we have

x1 < x2 ⇐⇒ f (x1 ) < f (x2 ) ∀ x1 , x2 ∈ I. (3.1)

Defining y1 = f (x1 ) and y2 = f (x2 ), this is equivalent to

y1 < y2 ⇐⇒ g(y1 ) < g(y2 ) ∀ y1 , y2 ∈ J.

Thus g is strictly increasing.

To show that J is an interval, y1 , y2 ∈ J, and assume without loss of generality that
y1 < y2 . Since J = f (I), (3.1) implies that y1 = f (x1 ) and y2 = f (x2 ) for some x1 , x2 ∈ I
with x1 < x2 . Now, by the Intermediate Value Theorem 3.24 applied to f : [x1 , x2 ] → R, we
have that all values c ∈ [y1 , y2 ] are in the image of f : [x1 , x2 ] → R, that is

[y1 , y2 ] ⊂ f ([x1 , x2 ]) ⊂ J.

Since y1 < y2 were two arbitrary points in J, this proves that J is an interval.
It remains to show that g = f −1 is continuous. Fix ȳ ∈ J, and suppose by contradiction
that g is not continuous at ȳ. Then, by Remark 3.22, there exist ε > 0 and a sequence
n=0 ⊂ J such that
(yn )∞

yn → ȳ but |g(yn ) − g(ȳ)| ≥ ε for all n ∈ N. (3.2)

13
Set xn = g(yn ) ∈ I and x̄ = g(ȳ). Then for every n ∈ N,

either xn ≤ x̄ − ε or xn ≥ x̄ + ε.

In particular, at least one of these cases must occur infinitely often. Without loss of generality,
assume xn ≤ x̄ − ε for infinitely many n, and extract a subsequence (xnk )∞ k=0 with xnk ≤ x̄ − ε
for all k. Since I is an interval, x̄ − ε ∈ I, and by strict monotonicity of f we obtain

ynk = f (xnk ) ≤ f (x̄ − ε) < f (x̄) = ȳ.

Taking the limit and using Proposition 2.95 gives (recall that yn → ȳ, see (3.2))

ȳ = lim ynk ≤ f (x̄ − ε) < f (x̄) = ȳ,

k→∞

a contradiction. Hence g is continuous.

Example 3.29. — Let n ∈ N, n ≥ 1. The function f : [0, ∞) → [0, ∞) defined by

f (x) = xn is continuous, strictly increasing, and surjective. By the Inverse Function Theorem,
there exists a continuous strictly increasing inverse [0, ∞) → [0, ∞), denoted by
√
n
x 7→ x,

Version: December 30, 2025. 88

 Chapter 3.2 Continuous Functions

called the n-th root. Furthermore, for m, n ∈ N with n ≥ 1, we define

m √ √
xn = n
x · ... · n x for x ∈ [0, ∞),
| {z }
m times

and, using Exercise 3.17,

m 1
x− n = m for x ∈ (0, ∞).
xn
y
y=x

13

(x, f (x))

(f (x), x)
√
f −1 (x) = x
x
f (x) = x2
√
Figure 3.6: The function f (x) = x2 and its inverse f −1 (x) = x are reflections of each other
across the line y = x.

Version: December 30, 2025. 89

 Chapter 3.3 Continuous Functions on Compact Intervals

3.3 Continuous Functions on Compact Intervals

In this section we show that continuous functions on bounded closed intervals — called
compact intervals — enjoy special properties.

3.3.1 Boundedness and Extrema

The key property of compact intervals, which will be used several times, is the following:

Lemma 3.30: Compactness

Let [a, b] be a compact interval, and let (xn )∞
n=0 be a sequence contained in [a, b]. Then
∞
there exists a subsequence (xnk )k=0 such that

lim xnk = x̄ for some x̄ ∈ [a, b].

k→∞

Proof. Since (xn )∞

n=0 is bounded (as it lies in [a, b]), Corollary 2.113 ensures the existence of
a convergent subsequence (xnk )∞ k=0 . Let x̄ denote its limit. Because a ≤ xnk ≤ b for all k,
Proposition 2.95 yields a ≤ x̄ ≤ b.

Theorem 3.31: Boundedness

14
Let [a, b] be a compact interval, and let f : [a, b] → R be continuous. Then f is bounded.

Proof. Assume by contradiction that f is unbounded. Then, for every n ∈ N, there exists
xn ∈ [a, b] such that |f (xn )| ≥ n. By Lemma 3.30, there is a subsequence (xnk )∞ k=0 converging
to some x̄ ∈ [a, b].
Since f is continuous, so is |f | (recall Remark 3.16), therefore |f (xnk )| → |f (x̄)| ∈ R. This
contradicts |f (xnk )| ≥ nk → ∞, so f must be bounded.

Exercise 3.32. — Find examples of:

1. a continuous but unbounded function on a bounded open interval;

2. a continuous but unbounded function on an unbounded closed interval;

3. a function unbounded on a compact interval but discontinuous at only one point.

Version: December 30, 2025. 90

 Chapter 3.3 Continuous Functions on Compact Intervals

Definition 3.33: Extreme Values

Let D ⊂ R and f : D → R.

• We say that f takes its maximum value at x0 ∈ D if f (x) ≤ f (x0 ) for all x ∈ D.
Then f (x0 ) is the maximum of f .

• We say that f takes its minimum value at x0 ∈ D if f (x) ≥ f (x0 ) for all x ∈ D.
Then f (x0 ) is the minimum of f .

Maxima and minima are called extreme values or extrema.

Theorem 3.34: Extreme Value Theorem

Let [a, b] be a compact interval, and let f : [a, b] → R be continuous. Then f attains
both its maximum and its minimum.

Proof. Theorem 3.31 guarantees that f is bounded, or equivalently, that f ([a, b]) ⊂ R is a
bounded subset of R. Thus, Theorem 2.55 implies that

S := sup f ([a, b])

exists. By definition of the supremum, for each n ∈ N there exists yn ∈ f ([a, b]) such that
S − 2−n ≤ yn ≤ S. Hence yn → S. Also, since yn ∈ f ([a, b]), there exists xn ∈ [a, b] such that
14 f (xn ) = yn .
Now, by Lemma 3.30, we can find subsequence (xnk )∞ k=0 such that xnk → x̄ ∈ [a, b]. By
continuity of f ,
f (x̄) = lim f (xnk ) = lim ynk = S,
k→∞ k→∞

so f attains its maximum at x̄.

Applying the same reasoning to −f shows that f also attains its minimum.

Exercise 3.35. — Does every continuous function f on the open interval (0, 1) attain its
maximum?

3.3.2 Uniform Continuity

Definition 3.36: Uniform Continuity

Let D ⊆ R. A function f : D → R is uniformly continuous if, for every ε > 0, there
exists δ > 0 such that

|x − y| < δ =⇒ |f (x) − f (y)| < ε for all x, y ∈ D.

Remark 3.37. — The difference between the usual definition of continuity and the one of
uniform continuity lies in how the choice of δ depends on the points considered.

Version: December 30, 2025. 91

 Chapter 3.3 Continuous Functions on Compact Intervals

For a function that is continuous at each x0 ∈ D, the δ in the definition may depend on
both ε and x0 : for every ε > 0 and each x0 , we can find a δ = δ(ε, x0 ) that works near x0 .
Uniform continuity is stronger: there exists a single δ = δ(ε) that works simultaneously
for all x, y ∈ D. In other words, the control on the variation of f does not deteriorate as
we move along the domain. This property is automatically satisfied on compact intervals for
continuous functions, as we will prove below.

Example 3.38. — We know that the function f (x) = x2 is continuous on R. We now

prove that it is not uniformly continuous. Indeed, fix ε = 1. We will show that for every δ > 0
there exist xδ , yδ ∈ R with

|xδ − yδ | < δ and |f (xδ ) − f (yδ )| ≥ 1.

Set
1 δ
yδ = , xδ = yδ + .
δ 2
Then |xδ − yδ | = δ
2 < δ, and
2
δ2

1 δ 1
|f (xδ ) − f (yδ )| = x2δ − yδ2 = + − = 1 + ≥ 1.
δ 2 δ2 4

This shows that no δ works for ε = 1, so f is not uniformly continuous on R.

14
Theorem 3.39: Uniform Continuity on Compact Intervals
Let [a, b] be a compact interval, and f : [a, b] → R continuous. Then f is uniformly
continuous.

Proof. Assume, by contradiction, that f is not uniformly continuous on [a, b]. Then there
exists ε > 0 such that for every δ > 0 one can find x, y ∈ [a, b] with

|x − y| < δ and |f (x) − f (y)| ≥ ε.

Taking δ = 2−n for each n ∈ N, we obtain sequences (xn )∞

n=0 and (yn )n=0 in [a, b] with
∞

|xn − yn | < 2−n and |f (xn ) − f (yn )| ≥ ε. (3.3)

By Lemma 3.30, the sequence (xn )∞

n=0 has a subsequence (xnk )k=0 converging to some x̄ ∈
∞

[a, b]. Then

|ynk − x̄| ≤ |ynk − xnk | + |xnk − x̄| < 2−nk + |xnk − x̄| −−−→ 0,
k→∞

so ynk → x̄ as well. Thus, by continuity of f and Theorem 3.21,

lim f (xnk ) = lim f (ynk ) = f (x̄),

k→∞ k→∞

Version: December 30, 2025. 92

 Chapter 3.3 Continuous Functions on Compact Intervals

therefore
|f (xnk ) − f (ynk )| ≤ |f (xnk ) − f (x̄)| + |f (x̄) − f (ynk )| −−−→ 0,
k→∞

which contradicts (3.3). Hence, f is uniformly continuous on [a, b].

Exercise 3.40. — Does Theorem 3.39 remain true for continuous functions on the open
interval (0, 1)?

Exercise 3.41. — In this exercise, we introduce another notion of continuity: Lipschitz

continuity.
14 1. A function f : D → R on a set D ⊂ R is Lipschitz continuous if there exists L ≥ 0
such that
|f (x) − f (y)| ≤ L|x − y| ∀ x, y ∈ D.

Give examples of Lipschitz continuous functions, and show that any such function is
uniformly continuous.
√
2. Let f (x) = x on R≥0 . Show that:

(i) f[0,1] : [0, 1] → R is not Lipschitz continuous;

(ii) f[1,∞) : [1, ∞) → R is Lipschitz continuous;
(iii) f : [0, ∞) → R is uniformly continuous.

Version: December 30, 2025. 93

 Chapter 3.4 Example: Exponential and Logarithmic Functions

3.4 Example: Exponential and Logarithmic Functions

In this section, we will use the notion of convergence and limits of sequences to define the
exponential function and prove some of its basic properties, such as continuity.

3.4.1 Definition of the Exponential Function

We start with a preliminary lemma that will be used repeatedly in this section.

Lemma 3.42: Bernoulli’s Inequality

For all a ∈ R with a ≥ −1 and all n ∈ N with n ≥ 1, it holds that

(1 + a)n ≥ 1 + na.

Proof. We proceed by induction. For n = 1 we have (1 + a)1 = 1 + a = 1 + 1 · a.

Now assume that the inequality holds for some n ≥ 1. Since 1 + a ≥ 0 by assumption, we
find

(1 + a)n+1 = (1 + a)n (1 + a) ≥ (1 + na)(1 + a) = 1 + na + a + na2 ≥ 1 + (n + 1)a,

which establishes the induction step and completes the proof.

14 Remark 3.43. — Using Bernoulli’s inequality and Archimedes’ principle one can prove the
following statement: For all x > 1 and all y ∈ R, there exists n ∈ N such that xn ≥ y.
Indeed, assume y > 0 (otherwise the inequality xn ≥ y is true for every n ∈ N) and apply
Bernoulli’s inequality with a = x − 1 > 0 to obtain (1 + a)n ≥ 1 + na ≥ na. Choosing n ≥ ay
(using Archimedes’ principle), it follows that xn ≥ y.

Proposition 3.44: Existence of the Exponential

Let x ∈ R. The sequence (an )∞
n=1 defined by
 x n
an = 1 +
n

is convergent, and its limit is a positive real number.

To prove this result, we first establish monotonicity.

Lemma 3.45: Monotonicity

Given x ∈ R, let n0 ∈ N satisfy n0 ≥ 1 and n0 > −x. Then the sequence (an )∞
n=n0
defined in Proposition 3.44 is increasing.

Version: December 30, 2025. 94

 Chapter 3.4 Example: Exponential and Logarithmic Functions

Proof. Note that

!n+1
x n+1 x


1 + n+1 1 + n+1
 
an+1 x
=
n = 1 +
an 1 + nx n 1 + nx
n(n + 1 + x) n+1 n + x
   n+1
n+x x
= = 1− .
n (n + 1)(n + x) n (n + 1)(n + x)

Also, for n ≥ n0 we have n + x > 0, hence

x x+n 1
≤ = ≤ 1,
(n + 1)(n + x) (n + 1)(n + x) n+1

or equivalently
x
− ≥ −1 ∀ n ≥ n0 .
(n + 1)(n + x)
Hence, by Bernoulli’s inequality (Lemma 3.42) applied with a = − (n+1)(n+x)
x
and n + 1 in
place of n, we get
 n+1  
an+1 n+x x n+x x
= 1− ≥ 1 − (n + 1)
an n (n + 1)(n + x) n (n + 1)(n + x)
 
n+x x
= 1− = 1,
n n+x

therefore an ≤ an+1 for all n ≥ n0 .

14
Proof of Proposition 3.44. Fix x ∈ R, and let n0 ∈ N satisfy n0 ≥ 1 and n0 > −x. By
Lemma 3.45, the sequence (an )∞n=n0 is increasing. If we can show that it is bounded, Theo-
rem 2.104 ensures convergence.
Case 1: x ≤ 0. In this case, −x = |x| and n0 > |x|. Then, for all n ≥ n0 > |x|,

x
0<1+ ≤ 1,
n

so  x n
0< 1+ ≤ 1.
n
Hence (an )∞n=n0 is an increasing sequence bounded above by 1, and therefore convergent. In
particular,  x n  x n
lim 1 + = sup 1 + > 0.
n→∞ n n≥n0 n

Case 2: x > 0. For n > x we have

 x n  x n  x2  n
1+ 1− = 1− 2 ≤ 1,
n n n

so  −n
x n 
 x −n (−x)
1≤ 1+ ≤ 1− = 1+ .
n n n

Version: December 30, 2025. 95

 Chapter 3.4 Example: Exponential and Logarithmic Functions

 n
(−x)
In other words, if we define bn = 1 + n , then

1
1 ≤ an ≤ ∀ n > x. (3.4)
bn
14
Since
  bn converges to a positive limit by Case 1, Proposition 2.94(4) implies that the sequence
1
bn also converges, and therefore it is bounded (Lemma 2.100). Recalling (3.4), we
n>x
conclude that the increasing sequence (an )∞
n=1 is also bounded, and therefore convergent.

Definition 3.46: Exponential Function

The exponential function exp : R → R>0 is defined by
 x n
exp(x) = lim 1+ ∀ x ∈ R.
n→∞ n

The Euler number is defined as

 1 n
e = exp(1) = lim 1+ . (3.5)
n→∞ n

Its numerical value is

e = 2.71828 18284 59045 23536 02874 71352 66249 77572 47093 69995 . . .
15
A useful consequence of Lemma 3.45 is the following bound.

Corollary 3.47: Growth of the Exponential

Given n ∈ N with n ≥ 1, the exponential function satisfies
 x n
exp(x) ≥ 1 + ∀ x > −n.
n

Proof. By Lemma 3.45 and Definition 3.46, for x > −n we have

an ≤ an+1 ≤ · · · ≤ exp(x).

Version: December 30, 2025. 96

 Chapter 3.4 Example: Exponential and Logarithmic Functions

3.4.2 Properties of the Exponential Function

Theorem 3.48: Properties of the Exponential Function

The exponential function exp : R → R>0 is bijective, strictly increasing, and continuous.
Moreover,

exp(0) = 1, (3.6)
exp(−x) = exp(x)−1 , (3.7)
exp(x + y) = exp(x) exp(y), (3.8)

for all x, y ∈ R.

Proof. We first verify the identities (3.6), (3.7), and (3.8).

1. Proof of (3.6). By definition,

 0 n
exp(0) = lim 1+ = lim 1n = 1.
n→∞ n n→∞

2. Proof of (3.7). Using Proposition 2.94(2) and the definition of exp,

 x n  x n  x 2 n
exp(x) exp(−x) = lim 1+ · lim 1 − = lim 1 − 2 .
15 n→∞ n n→∞ n n→∞ n
2
For n ≥ |x| we have − nx2 ≥ −1, so by Bernoulli’s inequality (Lemma 3.42),

x2  x 2 n
1− ≤ 1− 2 ≤1 ∀ n ≥ |x|.
n n

Since the left-hand side tends to 1, by Lemma 2.97 we get

 x 2 n
lim 1− = 1,
n→∞ n2

which implies exp(x) exp(−x) = 1 and therefore exp(−x) = exp(x)−1 .

3. Proof of (3.8). For n ≥ 1,

 x  y  x + y cn
1− 1− 1+ = 1 + 2,
n n n n

where
x+y
cn = −(x2 + y 2 ) − xy + xy .
n
x+y
For n ≥ |x| + |y| we have n ≤ 1, hence

x+y
−2|xy| ≤ −xy + xy ≤ 2|xy|.
n

Version: December 30, 2025. 97

 Chapter 3.4 Example: Exponential and Logarithmic Functions

Using (|x| − |y|)2 ≥ 0 we obtain 2|xy| ≤ x2 + y 2 , so for n ≥ |x| + |y|,

−2(x2 + y 2 ) ≤ cn ≤ 0,
2 2) 2 2)
and in particular ncn2 ≥ − 2(x n+y
2 ≥ − 2(x n+y ≥ −1 for all sufficiently large n. There-
fore, by Bernoulli’s inequality,

x2 + y 2 cn  cn n
1−2 ≤1+ ≤ 1+ 2 ≤1 for n large,
n n n

so Lemma 2.97 yields  cn n

lim 1+ = 1.
n→∞ n2
Now,

exp(x + y)  x n  y n  x + y n  cn n
= lim 1 − 1− 1+ = lim 1 + 2 = 1,
exp(x) exp(y) n→∞ n n n n→∞ n

which proves exp(x + y) = exp(x) exp(y).

We next establish the continuity, monotonicity, and bijectivity of exp.

We begin with two basic estimates that we will use repeatedly:

exp(x) ≥ 1 + x ∀ x ∈ R, (3.9)
15
which follows from Corollary 3.47 for x > −1, and is trivial for x ≤ −1 (since exp(x) > 0).
Using (3.7) together with (3.9) we obtain

1 1
exp(x) = ≤ ∀ x < 1. (3.10)
exp(−x) 1−x

A further consequence is that

 1
exp − < x < exp(x) ∀ x > 0. (3.11)
x

Indeed, the right inequality x < exp(x) follows from (3.9). For the left inequality, apply (3.9)
with y = x1 > 0:
1 1 1  1  1 −1  1 −1
exp ≥1+ > =⇒ exp − = exp < = x.
x x x x x x

1. Strict monotonicity. If x < y, then (3.9) yields exp(y − x) ≥ 1 + (y − x) > 1, hence

exp(y) = exp(x) exp(y − x) > exp(x).

2. Continuity. First, continuity at 0. Fix δ ∈ (0, 1) and consider x ∈ (−δ, δ).

(i) If x ∈ [0, δ), then by the monotonicity of exp we have exp(x) > exp(0) = 1, so it

Version: December 30, 2025. 98

 Chapter 3.4 Example: Exponential and Logarithmic Functions

follows from (3.10) that

1 1 δ
| exp(x) − exp(0)| = exp(x) − 1 ≤ −1< −1= .
1−x 1−δ 1−δ

(ii) If x ∈ (−δ, 0] we now have exp(x) < exp(0) = 1 (again by the monotonicity of exp),
so it follows from (3.9) that

δ
| exp(x) − exp(0)| = 1 − exp(x) ≤ 1 − (1 + x) = −x < δ ≤ .
1−δ

In other words, we proved that | exp(x) − exp(0)| < 1−δδ

for all x ∈ (−δ, δ).
Now, given ε > 0, choose δ = 1+ε ; with this choice it follows that 1−δ
ε δ
= ε, therefore

δ
x ∈ (−δ, δ) =⇒ | exp(x) − 1| < = ε,
1−δ

proving the continuity of exp at 0.

To show continuity at an arbitrary point x̄ ∈ R, write exp(x) = exp(x̄) exp(x − x̄), and
let (xn )∞
n=0 be a sequence with xn → x̄. Then, since xn − x̄ → 0 and exp is continuous
at 0, it follows from Theorem 3.21 that

lim exp(xn ) = exp(x̄) lim exp(xn − x̄) = exp(x̄) exp(0) = exp(x̄).

n→∞ n→∞
15
Since (xn )∞
n=0 is an arbitrary sequence converging to x̄, Theorem 3.21 implies that exp
is continuous at x0 .

3. Bijectivity. First of all, strict monotonicity implies injectivity. For surjectivity, fix a > 0
and set x0 = −a−1 and x1 = a. Then, thanks to (3.11),

exp(x0 ) < a < exp(x1 ).

Hence, by the continuity of exp and the Intermediate Value Theorem 3.24 applied on
[x0 , x1 ], there exists x ∈ [x0 , x1 ] with exp(x) = a. This shows surjectivity and concludes
the proof.

3.4.3 The Natural Logarithm

Definition 3.49: Logarithm

The unique inverse function
log : R>0 → R

of the bijective mapping exp : R → R>0 is called the logarithm.

Version: December 30, 2025. 99

 Chapter 3.4 Example: Exponential and Logarithmic Functions

Corollary 3.50: Properties of the Logarithm

The logarithm log : R>0 → R is strictly increasing, continuous, and bijective. Moreover,

log(1) = 0, (3.12)
log(a−1 ) = − log(a), (3.13)
log(ab) = log(a) + log(b), (3.14)

for all a, b > 0.

Proof. This follows directly from Theorem 3.48 and the Inverse Function Theorem 3.28. Equa-
tions (3.12), (3.13), and (3.14) follow from the corresponding properties of the exponential,
choosing x = log a and y = log b.

15

log(x)
exp(x) 1

Figure 3.7: Graphs of the exponential function and the logarithm. The auxiliary dashed lines
show that exp(x) ≥ x + 1 and log(x) ≤ x − 1.

The logarithm defined here is also called the natural logarithm to distinguish it from
logarithms with another base a > 1 (for instance a = 10 or a = 2). For any a > 1, we define

log x
loga (x) = ∀ x > 0.
log a

Version: December 30, 2025. 100

 Chapter 3.4 Example: Exponential and Logarithmic Functions

For example, log10 (10n ) = n for all n ∈ Z. Unless stated otherwise, log(x) always denotes the
natural logarithm, i.e., the logarithm to base e.
We can now define powers with arbitrary real exponents. For a > 0 and x ∈ R we set

ax = exp(x log a).

In particular, ex = exp(x log e) = exp(x) for all x ∈ R. Similarly, for x > 0 and a ∈ R,

xa = exp(a log x).

√
x 10 x1

√
x 2

√
x1/ 2

x0
1
√
x− 2
15
1

Figure 3.8: Graphs of a 7→ ax for various real exponents x.

Exercise 3.51. — Show that for x ∈ Q and a > 0, this definition agrees with the one for
rational powers from Example 3.29. Verify also the rules

log(ax ) = x log(a), ax ay = ax+y , (ax )y = axy

for all a > 0 and x, y ∈ R.

Exercise 3.52. — Let a > 0. Show that there exists a constant Ca > 0 such that
log(x) ≤ Ca xa for all x > 0.

√
Exercise 3.53. — Given a ∈ R, consider the sequence (xn )∞
n=1 given by xn =
n
na . Show
that this sequence converges, with
√
lim n na = 1.
n→∞

Exercise 3.54. — In this exercise, we introduce another notion of continuity (compare

with Exercise 3.41).

Version: December 30, 2025. 101

 Chapter 3.4 Example: Exponential and Logarithmic Functions

1. Let D ⊂ R and α ∈ (0, 1]. A real-valued function f on D is called α-Hölder continuous

if there exists L ≥ 0 such that

|f (x) − f (y)| ≤ L|x − y|α ∀ x, y ∈ D.

Show that any α-Hölder continuous function is uniformly continuous.

(Note: for α = 1, this reduces to the Lipschitz condition.)
15 2. Given α ∈ (0, 1], consider f : [0, ∞) → R, f (x) = xα . Show that f is α-Hölder
continuous.
y
First observe that tα ≥ t for all t ∈ [0, 1]. Applying this fact with t = x+y
x
and t = x+y
with x, y > 0, deduce that

(x + y)α ≤ xα + y α ∀ x, y ≥ 0, α ∈ (0, 1]. (3.15)

Finally, use (3.15) to prove that f is α-Hölder continuous.

Version: December 30, 2025. 102

 Chapter 3.5 Limits of Functions

3.5 Limits of Functions

We consider functions f : D → R defined on a subset D ⊂ R, and we wish to define the limit
of f (x) as x ∈ D approaches a point x0 ∈ R. Typical examples include D = R, D = [0, 1], or
D = (0, 1), with x0 = 0 in each case.

3.5.1 Limit in the Vicinity of a Point

Let D ⊂ R be non-empty, and let x0 ∈ R be such that

D ∩ (x0 − δ, x0 + δ) ̸= ∅ (3.16)

for all δ > 0. Whenever this holds, we say that x0 is an accumulation point of D. Note
that if x0 ∈ D, then (3.16) is automatically satisfied.
Condition (3.16) ensures that there exists a sequence of points in D converging to x0 .

Definition 3.55: Limit of a Function

Let f : D → R, and x0 be an accumulation point of D.
A number L ∈ R is called the limit of f (x) as x → x0 if, for every ε > 0, there exists
δ > 0 such that
∀ x ∈ D, |x − x0 | < δ =⇒ |f (x) − L| < ε.

15 In general, the limit of f (x) as x → x0 may not exist. However, if it exists, it is uniquely
determined. Hence, we speak of the limit and write

lim f (x) = L
x→x0

to indicate that the limit exists and is equal to L. Informally, this means that the function
values f (x) are arbitrarily close to L whenever x ∈ D is sufficiently close to x0 .
The limit of a function satisfies properties analogous to those of Proposition 2.94. More
precisely:
If f and g are functions on D such that

lim f (x) = L1 and lim g(x) = L2 ,

x→x0 x→x0

then
lim (f + g)(x) = L1 + L2 , lim (f g)(x) = L1 L2 .
x→x0 x→x0

Moreover, f ≤ g implies L1 ≤ L2 , and the sandwich lemma holds: if f ≤ h ≤ g and L1 = L2 ,

then limx→x0 h(x) = L1 = L2 .

Version: December 30, 2025. 103

 Chapter 3.5 Limits of Functions

Remark 3.56. — Let f : D → R be a function. If x0 ∈ D, then f is continuous at x0 if

and only if limx→x0 f (x) = f (x0 ).

Suppose that x0 ∈ D is an accumulation point of D \ {x0 }. Let f : D → R, and consider

its restriction f |D\{x0 } . It may happen that f is discontinuous at x0 , but that the limit

L = lim f |D\{x0 } (x) (3.17)

x→x0

nevertheless exists. In this case, the point x0 is called a removable discontinuity of f , and
one also writes
L = x→x
lim f (x). (3.18)
0
x̸=x0

If we now define 
f (x), x ∈ D \ {x0 },

fe(x) = (3.19)
L,

x = x0 ,

then fe is continuous at x0 . In other words, we can remove the discontinuity of f by redefining

its value at x0 to be L.
If instead x0 ∈
/ D but the limit in (3.18) exists, we call the function fe defined in (3.19) the
continuous extension of f to D ∪ {x0 }.
15 Arguing as in the proof of Theorem 3.21, we obtain the following result.

Lemma 3.57: Limit and Sequences

Let f : D → R. Then L = limx→x̄ f (x) if and only if, for every sequence (xn )∞
n=0 ⊂ D
converging to x̄, one has limn→∞ f (xn ) = L.

Exercise 3.58. — Prove Lemma 3.57.

We now state a result describing the behaviour of limits under composition with a contin-
uous function.

Proposition 3.59: Limit and Composition

Let E ⊂ R, and let f : D → E be such that the limit L = limx→x̄ f (x) exists and belongs
to E. If g : E → R is continuous at L, then

lim g(f (x)) = g(L).

x→x̄

Proof. Let (xn )∞

n=0 ⊂ D be a sequence converging to x̄. By Lemma 3.57 we have limn→∞ f (xn ) =
L. Since g is continuous at L, Theorem 3.21 gives limn→∞ g(f (xn )) = g(L). Because (xn )∞
n=0
was arbitrary, using Lemma 3.57 again, we conclude that limx→x̄ g(f (x)) = g(L).

Version: December 30, 2025. 104

 Chapter 3.5 Limits of Functions

We now introduce conventions for improper limits of functions, in analogy with improper
limits for sequences.

Definition 3.60: Improper Limits

Let f : D → R, and let x0 be an accumulation point of D.
We say that f diverges to +∞ as x → x0 , and write

lim f (x) = +∞,

x→x0

if for every M > 0 there exists δ > 0 such that

∀ x ∈ D, |x − x0 | < δ =⇒ f (x) ≥ M.

Analogously, f diverges to −∞ as x → x0 and we write lim f (x) = −∞, if for every

x→x0
M > 0 there exists δ > 0 such that
16
∀ x ∈ D, |x − x0 | < δ =⇒ f (x) ≤ −M.

3.5.2 One-sided Limits

It is often useful to consider limits taken from one side only and to allow x0 to be ±∞ as well.
To this end, let x0 ∈ R be such that

D ∩ (x0 , x0 + δ) ̸= ∅ (3.20)

for every δ > 0. In this case, we say that x0 is a right-hand accumulation point of D.
Analogously, if
D ∩ (x0 − δ, x0 ) ̸= ∅ (3.21)

for every δ > 0, we say that x0 is an left-hand accumulation point of D.

Version: December 30, 2025. 105

 Chapter 3.5 Limits of Functions

Definition 3.61: One-sided Limits

Let f : D → R, and let x0 ∈ R be a right-hand accumulation point of D.
A number L ∈ R is called the right-hand limit of f at x0 if, for every ε > 0, there
exists δ > 0 such that

x ∈ D ∩ (x0 , x0 + δ) =⇒ |f (x) − L| < ε.

In this case we write L = limx→x+ f (x).

0
We also allow improper one-sided limits. We say that

lim f (x) = +∞
x→x+
0

if for every M > 0 there exists δ > 0 such that

x ∈ D ∩ (x0 , x0 + δ) =⇒ f (x) ≥ M.

Similarly, limx→x+ f (x) = −∞ means: for every M > 0 there exists δ > 0 such that
0

x ∈ D ∩ (x0 , x0 + δ) =⇒ f (x) ≤ −M.

16 The left-hand limit is defined analogously, considering a left-hand accumulation point

of D and writing limx→x− f (x).
0

Next, we define the notion of limit at infinity.

Definition 3.62: Limits at Infinity

Let f : D → R, and assume that D ∩ (R, ∞) ̸= ∅ for every R > 0.
A number L ∈ R is called the limit of f as x → +∞ if, for every ε > 0, there exists
R > 0 such that
x ∈ D ∩ (R, ∞) =⇒ |f (x) − L| < ε.

We say that f diverges to +∞ as x → +∞ if, for every M > 0, there exists R > 0
such that
x ∈ D ∩ (R, ∞) =⇒ f (x) ≥ M.

The corresponding definitions for x → −∞ and divergence to −∞ are analogous.

Limits at +∞ can be converted into right-hand limits at 0 via inversion. Given f : D → R

as above, define

E = {x > 0 : x−1 ∈ D}, g : E → R, g(x) = f (x−1 ).

Version: December 30, 2025. 106

 Chapter 3.5 Limits of Functions

Then
lim f (x) = lim g(x),
x→+∞ x→0+

so one limit exists if and only if the other does.

Definition 3.63: One-sided Continuity and Jumps

Let f : D → R and x0 ∈ D. If limx→x+ f (x) exists and equals f (x0 ), then f is
0
continuous from the right at x0 . Continuity from the left is defined similarly.
We call x0 a jump point if both one-sided limits exist but are different:

L− := lim f (x) ∈ R, L+ := lim f (x) ∈ R, L− ̸= L+ .

x→x−
0 x→x+
0

The following graph represents a function with three points of discontinuity x1 , x2 , x3 .

16

x1 x2 x3

Figure 3.9: At x1 the discontinuity is removable (both one-sided limits exist and are equal,
but f (x1 ) differs). At x2 the function is continuous from the left but not from the right, and
x2 is a jump point. At x3 the function is continuous from the right; the left-hand limit does
not exist, hence x3 is not a jump point.

Example 3.64. — Let D = (0, ∞) and f (x) = xx = exp x log x . We want to compute

limx→0+ f (x). We split the argument into three steps.

y
2
(i) lim ye−y = 0. Indeed, by Corollary 3.47, for y > 0 one has ey ≥ 1 + 2 , hence
y→∞

y 4
0 ≤ ye−y ≤ y
2
≤ −−−→ 0
1+ y y→∞
2

by the sandwich lemma.

(ii) lim x log x = 0. For this, given ε > 0, choose R > 0 so that |ye−y | < ε for all y > R
x→0+
(this is possible thanks to point (i) above). Now, set δ = e−R and take x ∈ (0, δ). Then

Version: December 30, 2025. 107

 Chapter 3.5 Limits of Functions

y = − log x > R, which implies that

|x log x| = |e−y y| < ε ∀ x ∈ (0, δ).

(iii) lim xx = 1. Indeed, thanks to (ii) and the continuity of the exponential function, we
x→0+
can apply Proposition 3.59 to get

lim xx = lim exp x log x = exp(0) = 1.

x→0+ x→0+

Exercise 3.65. — Let a ∈ R. Compute, where defined,

x3 − x 2 − x − 2 3e2x + ex + 1 ex log x
lim , lim , lim , lim .
x→2 x−2 x→∞ 2e2x − 1 x→∞ xa x→∞ xa

Specify a suitable domain D (so that the functions are well defined) in each case.

3.5.3 Landau Notation

16 We introduce two standard notations that compare the asymptotic behaviour of a function to
that of another function (often called relative asymptotics). They are named after Edmund
Landau (1877–1938).

Definition 3.66: Big-O at a Point

Let f, g : D → R, and let x0 be an accumulation point of D. We write

f (x) = O g(x) as x → x0

if there exist M > 0 and δ > 0 such that

x ∈ D ∩ (x0 − δ, x0 + δ) =⇒ |f (x)| ≤ M |g(x)|.

We then say that f is a big-O of g as x → x0 .

If g(x) ̸= 0 for all x sufficiently close to x0 (with x ∈ D), then

f (x)
f (x) = O g(x) as x → x0 is bounded near x0 .


⇐⇒
g(x)

Version: December 30, 2025. 108

 Chapter 3.5 Limits of Functions

Definition 3.67: Big-O at Infinity

Let f, g : D → R, and assume that D ∩ (R, ∞) ̸= ∅ for every R > 0. We write

f (x) = O g(x) as x → +∞

if there exist M > 0 and R > 0 such that

x ∈ D ∩ (R, ∞) =⇒ |f (x)| ≤ M |g(x)|.

The definition for x → −∞ is analogous.

The big-O notation hides the precise bound by an implicit constant M , which is often
irrelevant for the argument that one is interested in.

Example 3.68. — • If f and g are bounded and continuous near x0 with g(x0 ) ̸= 0,
then f (x) = O g(x) as x → x0 .

• As x → 0, one has x2 = O(x), but x ̸= O(x2 ) (since x/x2 is unbounded near 0).
3x3 3x3
• As x → +∞, = O(1), but ̸= O(xα ) for α < 0.
x3 + 3 x3 + 3
16
As discussed above, the big-O means that f is bounded by a multiple of g. One may also
consider a stronger condition, namely that f is asymptotically negligible with respect to g.
This leads to the following definition.

Definition 3.69: Little-o at a Point

Let f, g : D → R, and let x0 be an accumulation point of D. We write

f (x) = o g(x) as x → x0

if, for every ε > 0, there exists δ > 0 such that

x ∈ D ∩ (x0 − δ, x0 + δ) =⇒ |f (x)| ≤ ε |g(x)|.

We then say that f is a little-o of g as x → x0 .

If g(x) ̸= 0 for all x near x0 (with x ∈ D), then

f (x)
f (x) = o g(x) as x → x0


⇐⇒ lim = 0.
x→x0 g(x)

Moreover, f (x) = o g(x) ⇒ f (x) = O g(x) .

Version: December 30, 2025. 109

 Chapter 3.5 Limits of Functions

Definition 3.70: Little-o at Infinity

Let f, g : D → R, and assume that D ∩ (R, ∞) ̸= ∅ for every R > 0. We write

f (x) = o g(x) as x → +∞

if, for every ε > 0, there exists R > 0 such that

x ∈ D ∩ (R, ∞) =⇒ |f (x)| ≤ ε |g(x)|.

The definition for x → −∞ is analogous.

Example 3.71. — • x = o(x2 ) as x → +∞, and x2 = o(x) as x → 0.

• For any α < 1,

3x3
= o |x|α as x → 0,


2x2 +x 10

but not for α ≥ 1. Indeed,

3x3 3
= |x|1−α −→ 0 as x → 0,
|x|α (2x2 + x10 ) 2 + x8
16 whenever α < 1.

In computations, one often uses Landau symbols as placeholders. Writing

f (x) + o g(x) as x → x0

means there is a function h : D → R with h(x) = o g(x) as x → x0 . Similarly for big-O.

Exercise 3.72. — Let p > 1, a ∈ R, and b > 0. Show:

1. xp = o(x) as x → 0;

2. x = o(xp ) as x → ∞;

3. xa = o(ex ) as x → ∞;

4. log x = o(xb ) as x → ∞.

Example 3.73. — Polynomial division gives, as x → ∞,

x3 − 7x2 + 6x + 2
 
1
=x−7+O = x − 7 + o(1) = x + O(1) = x + o(x).
x2 x

Version: December 30, 2025. 110

 Chapter 3.5 Limits of Functions

It may seem surprising that all four expressions are correct (and even useful) in different
contexts. Depending on the desired precision, one may use the more accurate form with error
term −7 + O x1 or the coarser estimate involving o(x).

Exercise 3.74. — Let f1 , f2 , g : D → R. If f1 (x) = o(g(x)) and f2 (x) = o(g(x)) as

x → x0 , prove that for all α1 , α2 ∈ R,

α1 f1 (x) + α2 f2 (x) = o g(x) as x → x0 .

Formulate and prove the analogous statement for big-O.

Exercise 3.75. — Let f1 , f2 , g1 , g2 : D → R. Show:

• If f1 (x) = o(g1 (x)) and f2 (x) = o(g2 (x)) as x → x0 , then f1 (x)f2 (x) = o g1 (x)g2 (x) .

• If f1 (x) = o(g1 (x)) and f2 (x) = O(g2 (x)) as x → x0 , then f1 (x)f2 (x) = o g1 (x)g2 (x) .

• If f1 (x) = O(g1 (x)) and f2 (x) = O(g2 (x)) as x → x0 , then f1 (x)f2 (x) = O g1 (x)g2 (x) .

16
3x2
Example 3.76. — Let f (x) = x + x3 + 4x4 + x7 and g(x) = x + . As x → 0 we have
1+x
3x2
f (x) = x+o(x2 ) (since x3 , x4 , x7 = o(x2 )), g(x) = x+O(x2 ) (since 1+x = 3x2 +O(x3 )).

Hence

f (x)g(x) = x + o(x2 ) x + O(x2 ) = x2 + x o(x2 ) + x O(x2 ) + o(x2 ) O(x2 ).

By the product for o(·) and O(·) (Exercise 3.75),

x o(x2 ) = o(x3 ), x O(x2 ) = O(x3 ), o(x2 ) O(x2 ) = o(x4 ).

Also, by the sum rule (Exercise 3.74),

o(x3 ) + O(x3 ) + o(x4 ) = O(x3 ).

Therefore,
f (x)g(x) = x2 + O(x3 ).

Version: December 30, 2025. 111

 Chapter 3.6 Sequences of Functions

3.6 Sequences of Functions

3.6.1 Pointwise Convergence

Definition 3.77: Sequences of Functions

A sequence of real-valued functions on a subset D ⊂ R is a family of functions fn :

D → R indexed by N. The function fn is called the n-th element of the sequence.
One often writes (fn )n∈N , (fn )∞
n=0 , or (fn )n≥0 for a sequence of functions.

As for sequences of real numbers, the starting index of a sequence of functions is not
essential and may differ from 0.

Definition 3.78: Pointwise Convergence

Let D ⊂ R, and let (fn )∞
n=0 be a sequence of functions fn : D → R. Let f : D → R be
another function. We say that (fn )∞
n=0 converges pointwise to f if for every x ∈ D,
the sequence of real numbers (fn (x))∞
n=0 converges to f (x). In this case, f is called the

17 pointwise limit of the sequence (fn )∞n=0 .

Exercise 3.79. — Show that the pointwise limit of a sequence of functions, if it exists, is
uniquely determined.

In the following example we show that, in general, continuity is not preserved under point-
wise convergence.

Example 3.80. — Let D = [0, 1] and, given n ≥ 1, define fn : D → R by fn (x) = xn . Then

the sequence of continuous functions (fn )∞
n=1 converges pointwise to the function f : D → R
given by 
0 if x < 1,

f (x) = lim fn (x) =
n→∞ 1 if x = 1.


The limit function f is not continuous on [0, 1].

Version: December 30, 2025. 112

 Chapter 3.6 Sequences of Functions

Figure 3.10: The sequence fn (x) = xn converges pointwise on [0, 1] to the function f (x) = 0
for x < 1 and f (1) = 1, which is not continuous at x = 1.

3.6.2 Uniform Convergence

As we have seen in the previous section, pointwise convergence of functions is not sufficient
to preserve continuity. We now introduce a stronger notion of convergence that ensures this
property. To motivate the definition, let us first rewrite the notion of pointwise convergence
17 using quantifiers.
Recall that (fn )∞
n=0 converges pointwise to f on D if, for every x ∈ D and every ε > 0,
there exists N ∈ N such that

|fn (x) − f (x)| < ε for all n ≥ N.

In this definition we first fix x ∈ D and ε > 0, and then find an index N that may depend
on both x and ε. In contrast, in the definition of uniform convergence below, we change the
order of quantifiers to ensure that N depends only on ε, and not on the particular choice of
x ∈ D.

Definition 3.81: Uniform Convergence

Let D ⊂ R, and let (fn )∞
n=0 be a sequence of functions fn : D → R. Let f : D → R be
another function. We say that (fn )∞
n=0 converges uniformly to f on D if, for every
ε > 0, there exists N ∈ N such that

|fn (x) − f (x)| < ε for all n ≥ N and all x ∈ D.

The condition |fn (x) − f (x)| < ε is equivalent to

f (x) − ε ≤ fn (x) ≤ f (x) + ε.

Version: December 30, 2025. 113

 Chapter 3.6 Sequences of Functions

This gives a useful geometric interpretation of uniform convergence: (fn ) converges uniformly
to f if, for every ε > 0, the graph of fn eventually lies entirely within the ε-tube around the
graph of f (see Figure 3.11).

Figure 3.11: Illustration of uniform convergence: for all sufficiently large n, the graph of fn
lies entirely within the ε-tube around the graph of f .

Exercise 3.82. — Let (fn )∞ n=0 be a sequence of functions fn : D → R. Show that if

17 (fn )n=0 converges uniformly to a function f , then (fn )∞
∞
n=0 also converges pointwise to f .

Theorem 3.83: Continuity under Uniform Convergence

Let D ⊂ R, and let (fn )∞
n=0 be a sequence of continuous functions fn : D → R converg-
ing uniformly to a function f : D → R. Then f is continuous.

Proof. o prove that f is continuous, we fix x̄ ∈ D and show that f is continuous at x̄.
Given ε > 0, the uniform convergence of fN to f provides N ∈ N such that

|fN (y) − f (y)| < ε

3 for all y ∈ D.

Also, since fN is continuous at x̄, there exists δ > 0 such that

|x − x̄| < δ =⇒ |fN (x) − fN (x̄)| < 3ε .

Then, for |x − x̄| < δ, we have

|f (x) − f (x̄)| ≤ |f (x) − fN (x)| + |fN (x) − fN (x̄)| + |fN (x̄) − f (x̄)|
ε ε ε
< 3 + 3 + 3 = ε,

which shows that f is continuous at x̄. Since x̄ is arbitrary, f is continuous on D.

Version: December 30, 2025. 114

 Chapter 3.6 Sequences of Functions

Remark 3.84. — Intuitively, uniform convergence allows us to exchange the order of taking
limits. More precisely, assume that (fn )∞
n=0 is a sequence of continuous functions converging
pointwise to f . Then, by the pointwise convergence and the continuity of the functions fn ,
we have

f (x̄) = lim fn (x̄), fn (x̄) = lim fn (x), f (x) = lim fn (x) for all x ∈ D.
n→∞ x→x̄ n→∞

Hence,
   
f (x̄) = lim fn (x̄) = lim lim fn (x) , lim f (x) = lim lim fn (x) .
n→∞ n→∞ x→x̄ x→x̄ x→x̄ n→∞

Note that the function f is continuous at x̄ if and only if f (x̄) = limx→x̄ f (x), which by the
identities above is equivalent to
   
lim lim fn (x) = lim lim fn (x) .
x→x̄ n→∞ n→∞ x→x̄

As we have seen, for pointwise convergence this interchange of limits may fail because f need
not be continuous. However, Theorem 3.83 ensures that this equality holds under uniform
convergence.

The next exercise reformulates this idea in terms of sequences.

17
Exercise 3.85. — Let (fn )∞n=0 be a sequence of uniformly continuous functions fn : D → R
converging uniformly to f : D → R. Let (xn )∞n=0 be a sequence in D converging to x̄ ∈ D.
Show that
lim fn (xn ) = f (x̄). (3.22)
n→∞

Find an example showing that pointwise convergence of (fn )∞

n=0 to f is not sufficient to
guarantee (3.22).

Exercise 3.86. — Let (fn )∞ n=0 be a sequence of functions fn : D → R, and let f : D → R.

Suppose D = D1 ∪ D2 for two subsets such that (fn |D1 ) converges uniformly to f |D1 and
(fn |D2 ) converges uniformly to f |D2 . Show that (fn )∞
n=0 converges uniformly to f on D.

Exercise 3.87. — Let (fn )∞ n=0 be a sequence of bounded functions fn : D → R. Show

that if (fn )n=0 converges uniformly to a function f : D → R, then f is bounded. Also, give
∞

an example of a sequence (fn ) of bounded functions converging pointwise to an unbounded

function.

Exercise 3.88. — Let (fn )∞n=0 be a sequence of uniformly continuous real-valued functions
on D converging uniformly to f : D → R. Show that f is uniformly continuous.

Version: December 30, 2025. 115

Chapter 3.6 Sequences of Functions

Extra material: Equicontinuity and the Arzelà–Ascoli Theorem

We have seen that any bounded sequence of points (xn )∞ n=0 admits a convergent subsequence.
Is there an analogous result for sequences of continuous functions? The Ascoli-Arzelà Theorem
addresses this question. We first need some definitions.

Definition 3.89: Equiboundedness and equicontinuity

Let (fn )∞
n=0 be a sequence of functions fn : D → R. We say that the sequence is
equibounded on D if there exists M > 0 such that

|fn (x)| ≤ M for all x ∈ D, n ∈ N.

Also, the sequence is equicontinuous on D if for every ε > 0 there exists δ > 0 such
that
|x − y| < δ =⇒ |fn (x) − fn (y)| < ε for all x, y ∈ D, n ∈ N.

Theorem 3.90: Ascoli-Arzelà

Let [a, b] be a compact interval, and let (fn )∞
n=1 be a sequence of continuous functions
fn : [a, b] → R that is equibounded and equicontinuous and pointwise bounded. Then
there exist a subsequence (fnk )∞
k=1 and a continuous function f : [a, b] → R such that
(fnk )k=0 converges uniformly to f on [a, b].
∞

Remark 3.91. — • By Theorem 3.83, the uniform limit in Arzelà–Ascoli is automati-

cally continuous; the theorem guarantees the existence of such a subsequence.

• The compactness of [a, b] is essential; without it, equiboundedness and equicontinuity

do not force any uniformly convergent subsequence on all of D (see Exercise 3.93).

Exercise 3.92. — Following Example 3.80, show that the family {xn | n ∈ N} is not
equicontinuous on [0, 1].
Hint: Fix ε = 12 . Then, for any δ > 0, choose x = 1, y = 1 − 2δ , and take n large enough so

n
that 1 − 2δ < 21 .

Exercise 3.93. — Let g(x) = max{(1 − |x|), 0} and define fn (x) = g(x − n) on R. Show
that (fn )∞
n=0 is equibounded and equicontinuous on R, fn (x) → 0 for each fixed x, but no
subsequence converges uniformly on R. Conclude that compactness of the domain cannot be
dropped from Theorem 3.90.

Version: December 30, 2025. 116

 Chapter 4

Series and Power Series

In this chapter we study series (infinite sums). They provide a framework to define many
classical functions; in particular, we will use series to define the trigonometric functions.

4.1 Series of Real Numbers

Definition 4.1: Convergent and Divergent Series

Let (an )∞
n=0 be a sequence of real numbers, and let A ∈ R. We say that the series
P∞
k=0 ak converges to A if
Xn
A = lim ak .
n→∞
k=0
P∞
In other words, computing the infinite sum k=0 ak means finding (if it exists) the
limit of the partial sums
Xn
sn = ak , n ∈ N.
k=0

We call an the n-th term (or n-th summand) of the series. If the limit exists, its
17
value A is the sum of the series.
If the limit does not exist, the series is said to be not convergent. In particular, if the
sequence of partial sums (sn )∞ n=0 diverges to +∞ (respectively, to −∞), we say that
the series diverges to +∞ (respectively, to −∞). This situation is therefore a specific
case of a series that does not converge.

Remark 4.2. — Unless otherwise specified, all series will consist of real numbers.

Proposition 4.3: Necessary Condition for Convergence

If the series ∞
P
k=0 ak converges, then an → 0 as n → ∞.

117
 Chapter 4.1 Series of Real Numbers

Pn
Proof. By assumption the partial sums sn = k=0 ak satisfy sn → A ∈ R. Then, for n ≥ 1,

an = sn − sn−1 −−−→ A − A = 0.
n→∞

P∞
Example 4.4 (Geometric Series). — For q ∈ R, the geometric series n=0 q
n converges if
and only if |q| < 1, and in this case
∞
X 1
qk = .
1−q
k=0

Indeed, if the series converges, then by Proposition 4.3 we must have q n → 0, hence |q| < 1.
Conversely, for |q| < 1 one proves by induction that
n
X 1 − q n+1
sn = qk = ∀ n ∈ N, q ̸= 1.
1−q
k=0

Also, since |q| < 1, q n+1 → 0. Thus,

1 − q n+1 1
sn = −−−→ .
1 − q n→∞ 1 − q

17
Example 4.5 (Harmonic Series). — The converse of Proposition 4.3 fails: the harmonic
series ∞k=1 k does not converge. To see this, consider n = 2 with ℓ ∈ N. Grouping terms
1 ℓ
P

gives

2ℓ
X 1 1 1 1 1 1  1 1
=1+ + + + + ··· + + · · · + ℓ−1 + ··· + ℓ
k 2 3 4 5 8 2 +1 2
k=1
1 1 1 1 1 1 1 1 1
≥1+ + + + + + + +··· + ℓ + ··· + ℓ
2 |4 {z 4} |8 8 {z 8 8} |2 {z 2}
= 12 = 12 = 12
1 1 ℓ
=1+ + ··· + = 1 + ,
2
| {z 2
} 2
ℓ times

which is unbounded as ℓ → ∞.

Exercise 4.6. — Let ∞

P∞
k=0 ak and k=0 bk be convergent series, and let α, β ∈ R. Show
P
P∞
that k=0 (αak + βbk ) converges and

∞
X ∞
X ∞
X
(αak + βbk ) = α ak + β bk .
k=0 k=0 k=0

Version: December 30, 2025. 118

 Chapter 4.1 Series of Real Numbers

Lemma 4.7: Convergence of the Tail

Let ∞
P P∞
k=0 ak be a series and fix N ∈ N. Then k=0 ak is convergent if and only if
P∞
k=N ak is convergent, and in that case

∞
X N
X −1 ∞
X
ak = ak + ak .
k=0 k=0 k=N

The same equivalence holds for divergence to +∞ or −∞.

Proof. For every n ≥ N ,

n
X N
X −1 n
X
ak = ak + ak .
k=0 k=0 k=N

Thus the partial sums of k=0 ak converge if and only if those of ∞

P∞
k=N ak do, and the identity
P

in the statement follows by letting n → ∞. The divergence case is analogous.

17

4.1.1 Series with Nonnegative Elements

Proposition 4.8: Nonnegative Series: Convergence vs. Divergence

Let ∞
P
k=0 ak be a series with nonnegative terms ak ≥ 0 for all k ∈ N. Then the partial
Pn ∞
sums sn = k=0 ak form an increasing sequence. If (sn )n=0 is bounded, the series
P∞
k=0 ak converges; otherwise it diverges to +∞.

Proof. Since an+1 ≥ 0, we have sn+1 = sn + an+1 ≥ sn for all n ∈ N, so (sn )∞

n=0 is increasing.
If the sequence (sn )n=0 is bounded, then it converges by Theorem 2.104. If the partial
∞

sums are not bounded, then they diverge to +∞.

Remark 4.9. — If ∞ k=0 ak has nonnegative terms, then (sn )n=0 is bounded if and only if
∞
P

it has a bounded subsequence (snk )∞

k=0 (see Remark 2.105).

Version: December 30, 2025. 119

 Chapter 4.1 Series of Real Numbers

Corollary 4.10: Comparison Test (Majorant/Minorant)

Let ∞
P P∞
k=0 ak and k=0 bk be series with 0 ≤ ak ≤ bk for all k ∈ N. Then

∞
X ∞
X
0≤ ak ≤ bk ,
k=0 k=0

and in particular
∞
X ∞
X
bk convergent =⇒ ak convergent,
k=0 k=0
∞
X ∞
X
ak divergent to + ∞ =⇒ bk divergent to + ∞.
k=0 k=0

These implications remain true if the inequalities 0 ≤ an ≤ bn hold only for all n ≥ N ,
for some N ∈ N.

Pn Pn
Proof. From ak ≤ bk we get k=0 ak ≤ k=0 bk for all n ∈ N. Therefore

∞
X n
X n
X ∞
X
ak = lim ak ≤ lim bk = bk .
n→∞ n→∞
k=0 k=0 k=0 k=0

18 The last part of the statement follows from Lemma 4.7.

P∞ P∞
Under the assumptions of the corollary, k=0 bk is called a majorant of k=0 ak , and
P∞ P∞
k=0 ak a minorant of k=0 bk . Hence the names majorant and minorant criterion.

P∞
Example 4.11. — The series 1
k=1 k2 is convergent. Indeed, for k ≥ 2,

1 1
ak = 2
≤ = bk ,
k k(k − 1)

and since 1
k(k−1) = 1
k−1 − k1 ,

n n  
X X 1 1 1
bk = − = 1 − −−−→ 1.
k−1 k n n→∞
k=2 k=2

P∞ P∞ P∞
By the comparison test, 1
k=2 k2 ≤ k=2 bk = 1, hence 1
k=1 k2 converges.

2n − 10
Example 4.12. — Consider an = . Since limn→∞ n2 an = 2, there exists
n3 − 10n + 100
N ∈ N such that for all n ≥ N , 0 ≤ n2 an ≤ 3, or equivalently

3
0 ≤ an ≤ ∀ n ≥ N.
n2

Version: December 30, 2025. 120

 Chapter 4.1 Series of Real Numbers

P∞
By Corollary 4.10 and Example 4.11, the series n=0 an converges.

Proposition 4.13: Cauchy Condensation Test

Let (ak )∞
k=0 be a decreasing sequence of nonnegative numbers. Then

∞
X ∞
X
ak converges ⇐⇒ 2k a2k converges.
k=0 k=0

Proof. Consider the partial sum of the series ∞ k=0 ak starting from k = 2 up to an index that
P

is a power of 2. Since the terms ak are decreasing, the following inequalities hold:
n+1
2X
ak = a2 + (a3 + a4 ) + (a5 + · · · + a8 ) + · · · + (a2n +1 + · · · + a2n+1 )
k=2

≤ a1 + (a2 + a2 ) + (a4 + · · · + a4 ) + · · · + (a2n + · · · + a2n )

|{z} | {z } | {z } | {z }
=1·a1 =2·a2 =4·a4 =2n a2n
n
X
= a1 + 2a2 + 4a4 + · · · + 2n a2n = 2k a2k ,
k=0

and similarly,
n+1
2X
18 ak = a2 + (a3 + a4 ) + (a5 + · · · + a8 ) + · · · + (a2n +1 + · · · + a2n+1 )
k=2

≥ a2 + (a4 + a4 ) + (a8 + · · · + a8 ) + · · · + (a2n+1 + · · · + a2n+1 )

|{z} | {z } | {z } | {z }
=1·a2 =2·a4 =4·a8 =2n a2n+1
n+1
1
1X k
= 2a2 + 4a4 + · · · + 2n+1 a2n+1 = 2 a2k .
2 2
k=1

In other words,
n n+1
2X n+1
X
k 1X k
2 a2k ≥ aj ≥ 2 a2k .
2
k=0 j=2 k=1

By Remark 4.9 and Corollary 4.10, the partial sums of one series are bounded if and only if
those of the other are. Hence, the two series converge or diverge together.

P∞
Example 4.14. — For p ∈ R, the series 1
n=1 np converges exactly when p > 1. Indeed:

• If p ≤ 0, then 1
np ≥ 1, so the series diverges (e.g., by Proposition 4.3).

• If p > 0, the sequence 1

is decreasing; so, it follows from Proposition 4.13,


np n≥1

∞ ∞ ∞
X 1 X 1 X
k
p
converges ⇐⇒ 2 k
k p
= 2 1−p converges.
n (2 )
n=1 k=0 k=0

Version: December 30, 2025. 121

 Chapter 4.1 Series of Real Numbers

Recalling Example 4.4, this holds exactly when 21−p < 1, i.e. p > 1.

Remark 4.15. — The argument in Example 4.14 provides another proof that the harmonic
series diverges (cf. Example 4.5).

P∞ 1
Exercise 4.16. — For p ∈ R, show that the series n=2 converges exactly when
n (log n)p
p > 1.
Hint: for p ≤ 0, compare with the harmonic series; for p > 0, use Proposition 4.13 and
Example 4.14.

P∞ 1
Exercise 4.17. — Decide whether the series n=3 converges or diverges.
n log n log log n

18 4.1.2 Conditional Convergence

Definition 4.18: Absolute and Conditional Convergence

A series ∞
P∞
k=0 ak is absolutely convergent if |ak | converges. It is condition-
P
P∞ P∞ k=0
ally convergent if k=0 ak converges but k=0 |ak | diverges.

A striking feature of conditionally convergent series is that their terms can be rearranged
to obtain any prescribed limit.

Theorem 4.19: Riemann Rearrangement Theorem

Let ∞
P
n=0 an be a conditionally convergent series and let A ∈ R. There exists a bijection
φ : N → N such that
X∞
A= aφ(n) .
n=0

Extra material: Proof of Theorem 4.19

Sketch of proof. Since ∞ an is conditionally convergent, we have an → 0 as n → ∞ (by

P
Pn=0
∞
Proposition 4.3), while n=0 |an | = ∞ by assumption.
Let
P = {n ∈ N : an ≥ 0}, N = {n ∈ N : an < 0},

and enumerate them in increasing order:

P = {p0 < p1 < . . .}, N = {n0 < n1 < . . .}.

Version: December 30, 2025. 122

 Chapter 4.1 Series of Real Numbers

Because the series is conditionally convergent, one can check that

∞
X ∞
X
apk = +∞, (−ank ) = +∞;
k=0 k=0

otherwise, the original series would either diverge or be absolutely convergent.

Now, the idea of the construction is as follows: given A ∈ R, we rearrange the terms by
alternately appending the least unused positive terms until the partial sum first exceeds A,
then the least unused negative terms until it drops below A, and so on. The divergence of the
two subseries ensures that this process never terminates and that every index is eventually
used.
More in detail, we define the bijection φ : N → N recursively. If A < 0, set φ(0) = n1 ; if
A ≥ 0, set φ(0) = p1 . Assume φ(0), . . . , φ(n) have been chosen, and let
n
X
sn = aφ(k) .
k=0

Then define 
min P \ {φ(0), . . . , φ(n)} , if sn < A,


φ(n + 1) =
min N \ {φ(0), . . . , φ(n)} , if sn ≥ A.


The map φ is injective by construction, and surjective because both subseries of positive and
negative terms diverge to +∞ (hence no index remains unused).
Since an → 0 as n → ∞, the successive corrections become arbitrarily small, and the
sequence of partial sums (sn )∞
n=0 converges to A. This establishes the theorem.

Exercise 4.20. — Fill in the details omitted in the proof of Theorem 4.19. Show also that
one can obtain A = ±∞ by a suitable rearrangement.

4.1.3 Leibniz Criterion for Alternating Series

Definition 4.21: Alternating Series

If (ak )∞
k=0 is a sequence of nonnegative numbers, the series
18
∞
X
(−1)k ak
k=0

is called the alternating series associated with the sequence (ak )∞

k=0 .

Version: December 30, 2025. 123

 Chapter 4.1 Series of Real Numbers

Proposition 4.22: Leibniz Criterion

Let (ak )∞
k=0 be a monotonically decreasing sequence of nonnegative numbers with ak →
0. Then the alternating series ∞ k
P
k=0 (−1) ak converges, and for all n ∈ N,

2n+1
X ∞
X 2n
X
(−1)k ak ≤ (−1)k ak ≤ (−1)k ak . (4.1)
k=0 k=0 k=0

Pn
Proof. Let sn = k=0 (−1)
ka
k. Since the sequence (an )∞
n=0 is decreasing and non-negative,
we have

s2n+2 = s2n −a2n+1 + a2n+2 ≤ s2n ,

| {z }
≤0

s2n+1 = s2n−1 +a2n − a2n+1 ≥ s2n−1 ,

| {z }
≥0

s2n+2 = s2n+1 +a2n+2 ≥ s2n+1 .

| {z }
≥0

for all n ∈ N. In other words,

18 s1 ≤ s3 ≤ . . . ≤ s2n−1 ≤ s2n+1 ≤ . . . ≤ s2n+2 ≤ s2n ≤ . . . ≤ s2 ≤ s0 .

This implies that the sequence (s2n )∞

n=0 is decreasing and bounded below, while the sequence
(s2n+1 )n=0 is increasing and bounded above. Thus, both limits A = limn→∞ s2n+1 and
∞

B = limn→∞ s2n exist and satisfy

s1 ≤ s3 ≤ . . . ≤ s2n−1 ≤ s2n+1 ≤ A ≤ B ≤ s2n+2 ≤ s2n ≤ . . . ≤ s2 ≤ s0 . (4.2)

In particular
0 ≤ B − A ≤ s2n+2 − s2n+1 = a2n+2 ∀ n ∈ N,

and because a2n+2 → 0, we deduce that A = B.

Also, (4.2) yields s2n+1 ≤ A = B ≤ s2n , which corresponds exactly to (4.1).

Example 4.23 (Alternating Harmonic Series). — The series

∞
X (−1)n+1 1 1 1
=1− + − + ···
n 2 3 4
n=1

P∞
converges by Proposition 4.22, whereas 1
n=1 n diverges (Example 4.5); hence it is only
conditionally convergent.

Version: December 30, 2025. 124

 Chapter 4.2 Absolute Convergence

4.2 Absolute Convergence

In this section we will look at absolutely convergent series and prove some convergence criteria.
As before, unless otherwise specified, all sequences consist of real numbers.

4.2.1 Criteria for Absolute Convergence

We begin by restating the concept of a Cauchy sequence in the context of convergent series.

Theorem 4.24: Cauchy Criterion for Series

The series ∞
P
k=0 ak converges if and only if, for every ε > 0, there exists N ∈ N such
that for all n > m ≥ N ,
Xn
ak < ε.
k=m+1

P∞
Proof. By definition, the series k=0 ak converges if and only if the sequence of partial sums

n
X
sn = ak
k=0

18 n=0 is a Cauchy sequence, that

converges. By Theorem 2.120, this occurs if and only if (sn )∞
is, |sn − sm | < ε for all n, m ≥ N . Since sn − sm = 0 when n = m, and the expression is
symmetric in n and m, it suffices to consider the case n > m. In this case,
n
X
sn − s m = ak ,
k=m+1

which proves the claim.

Example 4.25. — To see once more that the harmonic series diverges, note that for any
N ∈ N,
2N
X 1 1 1
≥ N· = .
k 2N 2
k=N +1

Thus, the Cauchy condition fails for ε = 12 , and the series cannot converge.

We can now prove that absolutely convergent series do indeed converge.

Version: December 30, 2025. 125

 Chapter 4.2 Absolute Convergence

Proposition 4.26: Absolute Convergence Implies Convergence

If a series ∞
P
n=0 an converges absolutely, then it converges and satisfies the generealized
triangle inequality
X∞ X∞
an ≤ |an |.
n=0 n=0

Proof. Since ∞ n=0 |an | converges, by the Cauchy criterion (Theorem 4.24) there exists N ∈ N
P

such that, for all n > m ≥ N ,

Xn
|ak | < ε.
18 k=m+1

By the triangle inequality,

n
X n
X
ak ≤ |ak | < ε,
k=m+1 k=m+1
P∞
so n=0 analso satisfies the Cauchy criterion and therefore converges.
Moreover, again by the triangle inequality,
n
X n
X ∞
X
ak ≤ |ak | ≤ |ak | ∀ n ∈ N,
k=0 k=0 k=0

and taking the limit as n → ∞ gives the desired inequality.

We now establish two classical criteria guaranteeing absolute convergence. In their proofs, we
repeatedly use the following fact:

Remark 4.27. — If a sequence (xn )∞ n=0 converges to α ∈ R, then Proposition 2.95 implies
the following facts:
(i) for any q > α there exists N ∈ N such that xn < q for all n ≥ N ;
(ii) for any r < α there exists N ∈ N such that xn > r for all n ≥ N .

19 Proposition 4.28: Cauchy Root Criterion

Given a sequence (an )∞
n=0 , define

p
n
α = lim sup |an | ∈ R ∪ {∞}.
n→∞

Then
∞
X ∞
X
α < 1 =⇒ an converges absolutely, α > 1 =⇒ an does not converge.
n=0 n=0

Version: December 30, 2025. 126

 Chapter 4.2 Absolute Convergence

Proof. Suppose α < 1 and set q = 2 ,

1+α
so that q ∈ (α, 1). By definition,
p
n
p
lim sup |an | = lim sup k |ak |.
n→∞ n→∞ k≥n

Thus xn = supk≥n |ak | → α. Since α < q, Remark 4.27(i) implies the existence of N ∈ N
p
k

such that
p
k
xn = sup |ak | < q ∀ n ≥ N,
k≥n

therefore
|ak | < q k ∀ k ≥ N,

Since q < 1, ∞ k=N |ak | converges by comparison with the geometric series (Example 2.128),
P
P∞
so n=0 an converges absolutely.
If α > 1, since the limsup is an accumulation point (Theorem 2.112), Proposition 2.88
implies the existence of a subsequence (ank )∞ such that limk→∞ nk |ank | = α. Hence, thanks
p
k=0
to Remark 4.27(ii) with r = 1, nk |ank | > 1 for all k large, or equivalently, |ank | > 1 for large
p

k. In particular, the sequence (an )∞

n=0 does not converge to 0. Recalling Proposition 4.3, this
P∞
implies that the series n=0 an does not converge.

Remark 4.29. — If α = 1, the root criterion is inconclusive:

• n 1/n → 1 (Exercise 3.53) but ∞n=1 1/n diverges (Example 4.5);

p P
19
• n 1/n2 → 1 (Exercise 3.53) but ∞n=1 1/n converges (Example 4.11).
p P 2

Proposition 4.30: D’Alembert’s Quotient Criterion

Given a sequence (an )∞

n=0 with an ̸= 0 for all n, assume that

|an+1 |
lim = α ∈ [0, ∞).
n→∞ |an |

Then
∞
X ∞
X
α < 1 =⇒ an converges absolutely, α > 1 =⇒ an does not converge.
n=0 n=0

Proof. The proof parallels that of the root criterion.

|an+1 |
If α < 1, set q = 1+α
2 ∈ (α, 1). Since |an | → α and α < q, by Remark 4.27(i) there exists
N ∈ N such that
|ak+1 |
<q ∀ k ≥ N.
|ak |
This gives

|ak | |ak−1 | |aN +1 | |aN |

|ak | = · · ... · · |aN | < q k−N |aN | = N q k ∀ k ≥ N.
|ak−1 | |ak−2 | |aN | q

Version: December 30, 2025. 127

 Chapter 4.2 Absolute Convergence

Since q < 1, the geometric comparison test shows that ∞ n=0 |an | converges absolutely.
P

If α > 1, then Remark 4.27(ii) with r = 1 implies the existence of N ∈ N such that

|ak+1 |
>1 ∀ k ≥ N.
|ak |

In particular,

|ak | |ak−1 | |aN +1 |

|ak | = · · ... · · |aN | > |aN | ∀ k ≥ N.
|ak−1 | |ak−2 | |aN |

Hence (an )∞
n=0 does not tend to 0, and by Proposition 4.3 the series does not converge.

Exercise 4.31 (Generalization of the Quotient Criterion). — Let (an )∞

n=0 be a sequence
of nonzero numbers, and define

|an+1 | |an+1 |
α+ = lim sup , α− = lim inf .
n→∞ |an | n→∞ |an |

Show that
∞
X ∞
X
α+ < 1 =⇒ an converges absolutely, α− > 1 =⇒ an does not converge.
n=0 n=0

19 Is the second implication still true if α− is replaced by α+ ?

4.2.2 Reordering Series

Theorem 4.32: Rearrangement of Absolutely Convergent Series

Let ∞
P
an be an absolutely convergent series, and let φ : N → N be a bijection. Then
P∞ n=0
n=0 aφ(n) is absolutely convergent, and

∞
X ∞
X
an = aφ(n) . (4.3)
n=0 n=0

P∞
Proof. Fix ε > 0. Since n=0 |an | converges, there exists N ∈ N such that

∞
X ε
|ak | < .
2
k=N +1

Let
M = max φ−1 (0), . . . , φ−1 (N ) .


Equivalently, M ∈ N is the smallest number such that

{a0 , . . . , aN } ⊂ {aφ(0) , . . . , aφ(M ) }.

Version: December 30, 2025. 128

 Chapter 4.2 Absolute Convergence

Then
{a0 , . . . , aN } ⊂ {aφ(0) , . . . , aφ(n) } ∀ n ≥ M,

therefore
n
X N
X X
aφ(ℓ) − ak = aφ(ℓ) .
ℓ=0 k=0 0≤ℓ≤n
φ(ℓ)>N

Moreover, since all indices φ(ℓ) > N with 0 ≤ ℓ ≤ n correspond to terms among {|ak | | k ≥
N + 1}, we have
X ∞
X
|aφ(ℓ) | ≤ |ak |.
0≤ℓ≤n k=N +1
φ(ℓ)>N

This implies that, for n ≥ M , we can estimate

n
X ∞
X n
X N
X ∞
X X ∞
X
aφ(ℓ) − ak = aφ(ℓ) − ak − ak = aφ(ℓ) − ak
ℓ=0 k=0 ℓ=0 k=0 k=N +1 0≤ℓ≤n k=N +1
φ(ℓ)>N
X ∞
X ∞
X
≤ |aφ(ℓ) | + |ak | ≤ 2 |ak | < ε.
0≤ℓ≤n k=N +1 k=N +1
φ(ℓ)>N

19 This shows that

n
X ∞
X
aφ(ℓ) → ak as n → ∞,
ℓ=0 k=0
P∞
which proves the identity (4.3). Applying the same reasoning to n=0 |an | shows that
P∞ P∞
n=0 |an | < ∞, hence aφ(n) is absolutely convergent.
P
n=0 |aφ(n) | =

4.2.3 Products of Series

P∞
Our next goal is to define the product of two absolutely convergent series n=0 an and
P∞
n=0 bn . The following result shows that, in this case, the product can be obtained by sum-
ming all possible products aj bk of their terms in any order (that is, by choosing an arbitrary
bijection between N and N × N).

Theorem 4.33: Product Theorem

Let ∞
P P∞
n=0 an and n=0 bn be absolutely convergent series, and let α : N → N × N be a
bijection. Writing α(n) = (α1 (n), α2 (n)), one has
∞
X ∞
 X  ∞
X
an bn = aα1 (n) bα2 (n) , (4.4)
n=0 n=0 n=0

and the series on the right converges absolutely.

Version: December 30, 2025. 129

 Chapter 4.2 Absolute Convergence

Proof. Consider first a bijection α : N → N × N, written α(n) = (α1 (n), α2 (n)), such that

{α(k) : 0 ≤ k < n2 } = {0, 1, . . . , n − 1}2 ∀ n ∈ N.

For example, (α(n))∞

n=0 could traverse the grid N × N as shown in Figure 4.1 below.

j∈N

α(15) = (0, 3)

α(4) = (0, 2)

α(3) = (0, 1)
α(k) path

α(0) = (0, 0) α(1) = (1, 0) i∈N

19 Figure 4.1: Illustration of the bijection α : N → N × N, which enumerates all lattice points
in N2 by following the blue path. For each n, the first n2 elements {α(k) | 0 ≤ k < n2 − 1}
exactly cover the square {0, 1, . . . , n − 1}2 .

Then, for every n ∈ N,

2 −1
nX n−1
! n−1
!
X X
|aα1 (k) ||bα2 (k) | = |aℓ | |bm | .
k=0 ℓ=0 m=0

Since the right-hand side is bounded by

∞ ∞
! !
X X
|aℓ | |bm | ,
ℓ=0 m=0

we have
2 −1
nX ∞
! ∞
!
X X
sup |aα1 (k) ||bα2 (k) | ≤ |aℓ | |bm | < ∞.
n∈N k=0 ℓ=0 m=0
P∞
This implies that the series k=0 aα1 (k) bα2 (k) converges absolutely. In particular, since it
converges, its value can be computed along every subsequence, therefore

∞ 2 −1
nX
X
aα1 (k) bα2 (k) = lim aα1 (k) bα2 (k)
n→∞
k=0 k=0

Version: December 30, 2025. 130

 Chapter 4.3 Absolute Convergence

Now, writing the identity

2 −1
nX n−1
! n−1
!
X X
aα1 (k) bα2 (k) = aℓ bm ,
k=0 ℓ=0 m=0

and taking the limit as n → ∞, Proposition 2.94(2) gives

∞ 2 −1
nX
X
aα1 (k) bα2 (k) = lim aα1 (k) bα2 (k)
n→∞
k=0 k=0
n−1 n−1 ∞ ∞
! ! ! !
X X X X
19 = lim aℓ lim bm = an bn ,
n→∞ n→∞
ℓ=0 m=0 n=0 n=0

which proves (4.4) for this specific bijection α.

For an arbitrary bijection β : N → N2 , define φ = α−1 ◦ β : N → N. Then β = α ◦ φ,
and writing β(n) = (β1 (n), β2 (n)) = (α1 (φ(n)), α2 (φ(n))), the Rearrangement Theorem 4.32
yields
∞ ∞ ∞ ∞ ∞
! !
X X X X X
aβ1 (n) bβ2 (n) = aα1 (φ(n)) bα2 (φ(n)) = aα1 (n) bα2 (n) = an bn .
n=0 n=0 n=0 n=0 n=0

Corollary 4.34: Cauchy Product

If ∞
P P∞
n=0 an and n=0 bn are absolutely convergent, then

∞ ∞ ∞ n
! ! !
X X X X
an bn = an−k bk ,
n=0 n=0 n=0 k=0

and the series on the right converges absolutely.

Proof. Consider the bijection α : N → N×N defined as the inverse of the function f5 : N×N →
20
N from Example 2.70(3), represented in Figure 2.10. Explicitly,

α(0) = (0, 0), α(1) = (1, 0), α(2) = (0, 1), α(3) = (2, 0), α(4) = (1, 1), ...
α(20) = (0, 5), . . . , α(31) = (4, 3), ..., α(49) = (5, 4), . . . , etc.

By Theorem 4.33,
∞ ∞ ∞
! !
X X X
an bn = aα1 (n) bα2 (n) .
n=0 n=0 n=0

Listing the terms explicitly and grouping them by diagonals as in Figure 4.2, we obtain

Version: December 30, 2025. 131

 Chapter 4.3 Absolute Convergence

(0,3)
n=3

(0,2) (1,2)
n=2

(0,1) (1,1) (2,1)

n=1

(0,0) (1,0) (2,0) (3,0)

n=0 i

Figure 4.2: Grouping by diagonals j + k = n in N × N. Each diagonal corresponds to the set

of terms an−k bk appearing in the n-th partial sum of the Cauchy product.

∞
X
aα1 (n) bα2 (n) = a0 b0 + (a0 b1 + a1 b0 ) + (a2 b0 + a1 b1 + a0 b2 )
n=0

+ (a3 b0 + a2 b1 + a1 b2 + a0 b3 ) + · · ·
20
∞ ∞ n
! !
X X X X
= aj bk = an−k bk .
n=0 j,k≥0 n=0 k=0
j+k=n

Finally, absolute convergence follows from the triangle inequality and Theorem 4.33:
∞ X
X n ∞ X
X n ∞
X
an−k bk ≤ |an−k bk | = |aα1 (n) ||bα2 (n) | < ∞.
n=0 k=0 n=0 k=0 n=0

P∞ n
Example 4.35. — Let q ∈ R with |q| < 1. Then n=0 q converges absolutely (see
Example 4.4). Applying the Cauchy product to this series with itself yields

∞
!2 ∞ X
n ∞
1 X X X
= qn = q n−k q k = (n + 1)q n .
(1 − q)2
n=0 n=0 k=0 n=0
P∞
This gives us an explicit formula for the value of the series n=0 nq
n:

∞ ∞ ∞
X X X 1 1 q
nq n = (n + 1)q n − qn = 2
− = .
(1 − q) 1−q (1 − q)2
n=0 n=0 n=0

Version: December 30, 2025. 132

 Chapter 4.3 Absolute Convergence

4.3 Series of Complex Numbers

To define the notion of a convergent series in C, it is sufficient to consider separately the
corresponding series of its real and imaginary parts in R.

Definition 4.36: Series of Complex Numbers

Let (zn )∞
n=0 = (xn +iyn )n=0 be a sequence of complex numbers, and let Z = A+iB ∈ C.
∞

The series n=0 zn is said to converge to Z if both real series ∞

P∞ P∞
n=0 xn and
P
n=0 yn
converge, with limits A and B, respectively:
∞
X ∞
X
xn = A, yn = B.
n=0 n=0
P∞ P∞
We say that n=0 zn converges absolutely if the series of moduli n=0 |zn | con-
verges.

Whenever the series ∞

P∞
n=0 zn and n=0 wn converge absolutely, their sum and product
P
20 are given (exactly as in the real case) by the following formulas:
∞
X ∞
X ∞
X
zn + wn = (zn + wn ),
n=0 n=0 n=0
∞
X ∞
 X  ∞ X
n  (4.5)
X
zn wn = zn−k wk .
n=0 n=0 n=0 k=0

Remark 4.37. — Let (zn )∞ n=0 be a sequence of complex numbers, and

= (xn + iyn )∞
P∞ n=0
assume that the series n=0 |zn | converges. Since

0 ≤ |xn | ≤ |zn |, 0 ≤ |yn | ≤ |zn | ∀ n ∈ N,

the Majorant Criterion (Corollary 4.10) implies that both ∞

P∞
n=0 |xn | and n=0 |yn | converge.
P

Hence, the series of real and imaginary parts are absolutely convergent.
Conversely, since |zn | ≤ |xn | + |yn |, the absolute convergence of ∞
P∞
n=0 xn and
P
n=0 yn
P∞
also implies the absolute convergence of n=0 zn . Therefore, absolute convergence in C is
equivalent to absolute convergence of the real and imaginary parts.

Version: December 30, 2025. 133

 Chapter 4.4 Power Series

4.4 Power Series

Our next goal is to investigate power series. These are series where the terms are powers of
the variable x ∈ R (or z ∈ C, if one considers complex power series) multiplied by coefficients.

4.4.1 Radius of Convergence

Definition 4.38: Power Series

A power series with real coefficients is a series of the form
∞
X
an xn ,
n=0

where (an )∞
n=0 is a sequence in R and x ∈ R. Here, x is the variable, and an ∈ R is
the coefficient of xn .
By convention, we set x0 = 1 for all x ∈ R, including x = 0. In other words, the first
term of the power series is always a0 .
20
Addition and multiplication of power series are given by
∞
X ∞
X ∞
X
an xn + bn xn = (an + bn )xn ,
n=0 n=0 n=0
∞
X ∞
 X  X∞ X
n 
an xn bn xn = an−k bk xn ,
n=0 n=0 n=0 k=0

where the product formula follows from Corollary 4.34:

∞
X ∞
 X  ∞ X
X n  ∞ X
X n 
n n n−k k
an x bn x = an−k x bk x = an−k bk xn .
n=0 n=0 n=0 k=0 n=0 k=0

A power series is a polynomial whenever only finitely many of its coefficients are nonzero.
The convergence of a power series depends on the coefficients (an )∞
n=0 and is characterized
in Theorem 4.41.

Version: December 30, 2025. 134

 Chapter 4.4 Power Series

Definition 4.39: Radius of Convergence

Let ∞n=0 an x be a power series, and define
n
P

p
n
ρ := lim sup |an |.
n→∞

The radius of convergence is defined as


0,

 ρ = ∞,

R = ρ−1 , 0 < ρ < ∞,


∞,

ρ = 0.

(Equivalently, R = 1
ρ with the conventions 1/0 = ∞ and 1/∞ = 0.)

In the following, when we write R ∈ [0, ∞], we mean that R is either a nonnegative real
number or R = ∞.

P∞
Exercise 4.40. — For each R ∈ [0, ∞], find a power series n=0 an x
n with radius of
convergence R.

20 Theorem 4.41: Convergence of Power Series

P∞
Let an xn be a power series with radius of convergence R ∈ (0, ∞]. Then
P∞ n=0n
n=0 an x converges absolutely for all x ∈ R with |x| < R, and does not converge
for all x ∈ R with |x| > R. In particular, for x ∈ (−R, R) we can define the function
f (x) = ∞ n
P
n=0 an x .

Proof. Let x ∈ R, and write ρ = lim supn→∞ |an | as in Definition 4.39. Then
p
n

p  p 
n n
lim sup |an xn | = lim sup |an | |x| = ρ|x|.
n→∞ n→∞
P∞
By the root criterion (see Proposition 4.28), the series n=0 an x converges absolutely if
n

ρ|x| < 1, and does not converge if ρ|x| > 1 (in particular, if ρ = 0, then it converges
absolutely for all x ∈ R). Since R = ρ1 , the result follows.

Theorem 4.42: Continuity of Power Series

Let ∞ n
P
n=0 an x be a power series with radius of convergence R ∈ (0, ∞], and define the
polynomials fn (x) = nk=0 ak xk . For any r ∈ (0, R), the sequence (fn )∞
P
n=0 converges
uniformly to f on [−r, r]. In particular, the power series defines a continuous function
f : (−R, R) → R.

Version: December 30, 2025. 135

 Chapter 4.4 Power Series

Proof. By Theorem 4.41 with x = r, the series ∞ n=0 |an |r converges. Hence, for every ε > 0
n
P
P∞
there exists N ∈ N such that k=N +1 |ak |rk < ε. Thus, for all x ∈ [−r, r] and all n ≥ N ,

∞
X ∞
X ∞
X
k k
|fn (x) − f (x)| = ak x ≤ |ak | |x| ≤ |ak |rk < ε.
k=n+1 k=n+1 k=N +1

This shows that (fn )∞ n=0 converges uniformly to f on [−r, r]. Since each fn is continuous
(being a polynomial), Theorem 3.83 implies that f is continuous on [−r, r]. As r < R is
arbitrary, f is continuous on (−R, R).

Example 4.43. — In general, the partial sums fn (x) = nk=0 ak xk do not converge uni-
P

formly to f (x) = ∞ k=0 ak x on the whole interval (−R, R).

k
P

To see this, consider the geometric series ∞ n=0 x . Its radius of convergence is R = 1, and
n
P
P∞ n
on (−1, 1) we have f (x) = n=0 x = 1−x 1
(see Example 4.4). If the convergence on (−1, 1)
were uniform, then applying the notion of uniform convergence with ε = 1 would give N ∈ N
such that, for all n ≥ N and x ∈ (−1, 1),

n
X 1
xk − < 1.
1−x
k=0

Taking n = N and using the triangle inequality, we would get

20
N N
1 X
k
X
<1+ x ≤1+ |x|k ≤ 1 + (N + 1) = N + 2 ∀ x ∈ (−1, 1),
1−x
k=0 k=0

a contradiction since limx→1− 1

1−x = ∞.

Exercise 4.44. — Find the radius of convergence R of the power series

∞ √ √
n
X n2 + n − n2 + 1
xn ,
n2
n=1

and study convergence at x = R and x = −R.

Exercise 4.45. — Let ∞ n=0 an x be a power series with an ̸= 0 for all n ∈ N, and assume
n
P
|an |
that the limit limn→∞ |an+1 | exists. Then

|an |
R = lim .
n→∞ |an+1 |

Hint: Define L = limn→∞ |a|an+1

n|
| . Combining Proposition 4.30 and Proposition 2.94(4), deduce
P∞
that n=0 an x converges absolutely for all x ∈ R with |x| < L, and does not converge for
n

all x ∈ R with |x| > L. Comparing this statement with Theorem 4.41, conclude that L = R.

Version: December 30, 2025. 136

 Chapter 4.4 Power Series

Proposition 4.46: Radius of Convergence of Sum and Product

P∞ n
P∞ n
Let R ≥ 0, and let n=0 an x and n=0 bn x be power series with radius of con-
vergence at least R. Then their sum and their Cauchy product also have radius of
convergence at least R.

Proof. By linearity and Corollary 4.34, the absolute convergence of an xn and bn xn for
P P

|x| < R implies that

∞
X ∞
X ∞
 X  ∞ X
X n 
(an + bn )x n
and an x n
bn x n
= an−k bk xn
n=0 n=0 n=0 n=0 k=0

both converge absolutely for |x| < R. Since a power series cannot converge for |x| larger than
its radius of convergence, each has radius of convergence at least R.

P∞
Example 4.47. — If n=0 an x
n has radius of convergence at least 1, then

∞ ∞
1 X X
an xn = (a0 + · · · + an ) xn ∀ x ∈ (−1, 1). (4.6)
1−x
n=0 n=0

P∞
20 Indeed, since n=0 x
n = 1
1−x for x ∈ (−1, 1) and has radius 1, (4.6) follows from Proposi-
tion 4.46.

∞
X
Exercise 4.48. — Compute n 2−n .
n=1

4.4.2 Complex Power Series

Analogously to the real case, we can consider series with complex coefficients and a complex
variable z ∈ C.

Definition 4.49: Complex Power Series

A complex power series with complex coefficients is a series of the form
∞
X
an z n ,
n=0

where (an )∞
n=0 is a sequence in C and z ∈ C.
Again, by convention, we set z 0 = 1 for all z ∈ C, including z = 0.

Version: December 30, 2025. 137

 Chapter 4.4 Power Series

Addition and multiplication are defined as in the real case:

∞
X ∞
X ∞
X
n n
an z + bn z = (an + bn )z n ,
n=0 n=0 n=0
∞
X ∞
 X  ∞ X
n  (4.7)
X
n n
an z bn z = an−k bk z n .
n=0 n=0 n=0 k=0

The radius of convergence is defined exactly as in Definition 4.39, replacing |x| by |z|.
With these definitions, the following theorem holds, and the proof is identical to the real case.
20
Theorem 4.50: Convergence of Complex Power Series
Let ∞ an z n be a power series with radius of convergence R ∈ (0, ∞]. Then the series
P
P∞ n=0 n
n=0 an z converges absolutely for all z ∈ C with |z| < R, and diverges for all z ∈ C
with |z| > R. In particular, for |z| < R one can define the (complex-valued) function
∞
X
f (z) = an z n .
n=0

Extra Material: Continuity of Complex Power Series

Before stating the analogue of Theorem 4.42, we recall the notions of continuity and uniform
convergence for functions defined on subsets of C.

Definition 4.51: Continuity for Complex Functions

Let D ⊂ C and f : D → C. We say that f is continuous at a point z0 ∈ D if for
every ε > 0 there exists δ > 0 such that

∀ z ∈ D, |z − z0 | < δ =⇒ |f (z) − f (z0 )| < ε.

If f is continuous at every z0 ∈ D, then f is said to be continuous on D.

Definition 4.52: Uniform Convergence for Complex Functions

Let D ⊂ C and let (fn )∞
n=0 be a sequence of functions fn : D → C and let f : D → C.
We say that (fn )n=0 converges uniformly to f on D if for every ε > 0 there exists
∞

N ∈ N such that for all n ≥ N and all z ∈ D,

|fn (z) − f (z)| < ε.

These definitions are identical in form to the real case, with the only difference that | · | now
denotes the modulus on C. With these definition, some proofs about continuous functions can
be repeated verbatim. In particular, polynomial functions are continuous; also, Theorem 3.83

Version: December 30, 2025. 138

Chapter 4.4 Power Series

remains valid: the uniform limit of continuous complex-valued functions is continuous. As a

consequence, we get the following:

Theorem 4.53: Continuity of Complex Power Series

Let ∞ n
P
n=0 an z be a complex power series with radius of convergence R ∈ (0, ∞], and
define the partial sums
Xn
fn (z) = ak z k .
k=0

For every r ∈ (0, R), the sequence (fn )∞

n=0 converges uniformly to f for |z| ≤ r. In
particular, the power series defines a continuous function for |z| < R.

Proof. The proof is identical to the real case. Since ∞ n=0 |an |r < ∞ for every r < R, the
n
P

sequence of partial sums (fn )∞

n=0 converges uniformly for |z| ≤ r. Since each fn is a polynomial
in z, it is continuous on C. Since the uniform limit of continuous functions is continuous, f is
continuous for |z| ≤ r. Finally, since r < R is arbitrary, f is continuous for |z| < R.

Version: December 30, 2025. 139

 Chapter 4.5 Example: Exponential and Trigonometric Functions

4.5 Example: Exponential and Trigonometric Functions

4.5.1 The Exponential Map as a Power Series

In Section 3.4 we introduced the real exponential function and established its main properties.
We now show that the exponential map can alternatively be defined via the exponential
series
∞
X xn
exp(x) = , (4.8)
n!
n=0

where
0! = 1, n! = 1 · 2 · . . . · n.

Since (n+1)!
n! 1
= n+1 → 0 as n → ∞, it follows directly from the quotient criterion (see
Exercise 4.45) that this series has infinite radius of convergence. Hence, Theorem 4.42 implies
that the right-hand side of (4.8) defines a continuous function on R.

Remark 4.54. — Alternatively, given N ∈ N, we note that

n! ≥ n · (n − 1) · . . . · (N + 1) · N ≥ N n−N +1 ;
21 | {z
n−N +1 terms
}

therefore, since n−N +1

n → 1 as n → ∞

p
n 1 1
ρ = lim sup |an | ≤ lim n−N +1 = .
n→∞ n→∞
N n N

Since N can be chosen arbitrarily large, we conclude that ρ = 0, therefore R = ∞.

The representation of the exponential function as a power series is, in many ways, more
flexible than its definition as a limit. Moreover, as we shall see, its complex version will
naturally connect with the sine and cosine functions.
Before discussing these connections, we first show that the two representations of the
exponential function (via series and via limit) coincide.

Proposition 4.55: Exponential Map as Power Series

For every x ∈ R,
∞
X xk  x n
= lim 1+ .
k! n→∞ n
k=0

Version: December 30, 2025. 140

Chapter 4.5 Example: Exponential and Trigonometric Functions

Extra material: Proof of Proposition 4.55

Proof. For every n ≥ 0, we have

n   n k−1 
x n X n x k xk Y

 X ℓ
1+ = = 1− .
n k nk k! n
k=0 k=0 ℓ=0

P∞ |x|k
Fix x ∈ R and ε > 0. Since k=0 k! < ∞, there exists N ∈ N such that

∞
X |x|k ε
< .
k! 2
k=N +1

In particular,
N ∞ ∞
X xk X xk X |x|k ε
− ≤ < . (4.9)
k! k! k! 2
k=0 k=0 k=N +1

Moreover, for n ≥ N ,

N n k−1  N k−1 n k−1 

xk xk Y |x|k |x|k Y
  Y  
X X ℓ X ℓ X ℓ
− 1− ≤ 1− 1− + 1−
k! k! n k! n k! n
k=0 k=0 ℓ=0 k=0 ℓ=0 k=N +1 ℓ=0
| {z }
≤1
N k−1 ∞
|x|k |x|k
 Y 
X ℓ X
≤ 1− 1− + .
k! n k!
k=0 ℓ=0 k=N +1

Hence, we proved that, for every n ≥ N ,

N N k−1 ∞
xk |x|k |x|k
 Y 
X x n X ℓ X
− 1+ ≤ 1− 1− + .
k! n k! n k!
k=0 k=0 ℓ=0 k=N +1

Since for each fixed k,

 k−1
Y 
ℓ
lim 1 − 1− = 0,
n→∞ n
ℓ=0

letting n → ∞ yields

N N
X xk  x n X xk  x n
− lim 1+ = lim − 1+
k! n→∞ n n→∞ k! n
k=0 k=0
N k−1 ∞
|x|k |x|k
 Y 
X ℓ X
≤ lim 1− 1− +
n→∞ k! n k!
k=0 ℓ=0 k=N +1
∞
X |x|k
= .
k!
k=N +1

Version: December 30, 2025. 141

 Chapter 4.5 Example: Exponential and Trigonometric Functions

By the triangle inequality and recalling (4.9), this gives

∞ N ∞
X xk  x n X xk  x n X xk
− lim 1+ ≤ − lim 1 + +
k! n→∞ n k! n→∞ n k!
k=0 k=0 k=N +1
∞ ∞ ∞
X |x|k X xk X |x|k
≤ + ≤2 < ε.
k! k! k!
k=N +1 k=N +1 k=N +1

Since ε > 0 is arbitrary, the identity (4.8) follows.

Figure 4.3: The exponential function and some of its partial sums.
21
Definition 4.56: The Complex Exponential Map
The complex exponential map is the function exp : C → C defined by
∞
X zn
exp(z) = , z ∈ C.
n!
n=0

For a positive real number a > 0 and z ∈ C, we define

az := exp(z log a), in particular ez = exp(z).

Before proving the main properties of the exponential function, recall the binomial formula:
for all z, w ∈ C and n ∈ N,
n    
X n n n!
(z + w) = n k
z w n−k
, = . (4.10)
k k k!(n − k)!
k=0

Version: December 30, 2025. 142

 Chapter 4.5 Example: Exponential and Trigonometric Functions

Theorem 4.57: Properties of the Complex Exponential

The complex exponential map exp : C → C is continuous, and for all z, w ∈ C,

ez+w = ez ew , |ez | = eRe(z) . (4.11)

In particular, |eix | = 1 for all x ∈ R.

zn
Proof. Since the series ∞n=0 n! has infinite radius of convergence, Theorem 4.53 implies that
P

exp : C → C is continuous.
For z, w ∈ C, using (4.5) and (4.10), we obtain

∞ ∞ ∞ X
n
! !
z w
X zn X wn X z k wn−k
e e = =
n! n! k!(n − k)!
n=0 n=0 n=0 k=0
∞ n ∞
n k n−k X (z + w)n
 
X 1 X
= z w = = ez+w .
n! k n!
n=0 k=0 n=0

To compute the modulus, note that complex conjugation is continuous and satisfies z k = z k
(see Lemma 2.42(3)), therefore

n n n
X zk X zk X zk
ez = lim = lim = lim = ez .
21 n→∞ k! n→∞ k! n→∞ k!
k=0 k=0 k=0

Since |w|2 = ww and w + w = 2 Re(w) for all w ∈ C, we have

|ez |2 = ez ez = ez ez = ez+z = e2 Re(z) ,

hence |ez | = eRe(z) . In particular, |eix | = e0 = 1 for all x ∈ R.

Remark 4.58. — Note that the proof of Proposition 4.55 can be repeated verbatim to

n
show that ez = lim 1 + nz for all z ∈ C.
n→∞

4.5.2 Sine and Cosine

Given x ∈ R, we split the power series of eix into its even and odd terms:
∞ n ∞ ∞
ix
X i n
X i2n 2n X i2n+1
e = x = x + x2n+1 .
n! (2n)! (2n + 1)!
n=0 n=0 n=0

Since i2n = (−1)n and i2n+1 = i(−1)n , we obtain

∞ ∞
ix
X (−1)n 2n
X (−1)n 2n+1
e = x +i x .
(2n)! (2n + 1)!
n=0 n=0

Version: December 30, 2025. 143

 Chapter 4.5 Example: Exponential and Trigonometric Functions

This motivates the following definitions of the sine and cosine functions:
∞ ∞
X (−1)n 2n+1 X (−1)n
sin(x) = x , cos(x) = x2n , (4.12)
(2n + 1)! (2n)!
n=0 n=0

so that the identity

eix = cos(x) + i sin(x)

holds for all x ∈ R.

As for the exponential series, the radius of convergence of the power series in (4.12) is
infinite. Therefore, by Theorems 4.41 and 4.42, sin and cos are continuous on R.

Definition 4.59: Even and Odd Functions

Let D ⊂ R be a set satisfying D = −D, that is,

x∈D =⇒ −x ∈ D.

A function f : D → R is called
21
• even if f (−x) = f (x) for all x ∈ D,

• odd if f (−x) = −f (x) for all x ∈ D.

Geometrically, an even function is symmetric with respect to the y-axis, while an odd
function is symmetric with respect to the origin.

Since (−x)2n+1 = −x2n+1 and (−x)2n = x2n for all n ∈ N, it follows directly from (4.12)
that

sin(−x) = − sin(x) (sin is an odd function),

cos(−x) = cos(x) (cos is an even function).

Exercise 4.60. — Prove that,for all x ∈ R,

| sin(x)| = | sin(|x|)| and | cos(x)| = | cos(|x|)|.

Version: December 30, 2025. 144

 Chapter 4.5 Example: Exponential and Trigonometric Functions

Theorem 4.61: From the Complex Exponential to Sine and Cosine

For all x ∈ R, the following relations hold:

eix − e−ix eix + e−ix

eix = cos(x) + i sin(x), sin(x) = , cos(x) = .
2i 2

For all x, y ∈ R, the trigonometric addition formulas are:

sin(x + y) = sin(x) cos(y) + cos(x) sin(y),

(4.13)
cos(x + y) = cos(x) cos(y) − sin(x) sin(y).

Proof. For x ∈ R we have

eix = cos(x) + i sin(x), e−ix = cos(−x) + i sin(−x) = cos(x) − i sin(x).

Adding and subtracting these two identities gives the formulas for cos(x) and sin(x) in terms
of eix and e−ix .
To prove the addition formulas, we recall that

ei(x+y) = eix eiy

(see (4.11)), therefore

21
cos(x + y) + i sin(x + y) = ei(x+y) = eix eiy
= (cos(x) + i sin(x))(cos(y) + i sin(y))



= cos(x) cos(y) − sin(x) sin(y) + i sin(x) cos(y) + cos(x) sin(y) .

Comparing real and imaginary parts yields the identities in (4.13).

In particular, setting y = x gives the angle-doubling formulas:

sin(2x) = 2 sin(x) cos(x), cos(2x) = cos2 (x) − sin2 (x). (4.14)

Recalling that |eix | = 1, we also obtain the circle identity:

cos2 (x) + sin2 (x) = 1 ∀ x ∈ R.

4.5.3 The Circle Number

Theorem 4.62: Existence of π as the First Positive Zero of Sine

There exists exactly one number π ∈ (0, 4) such that sin(π) = 0. For this number it
holds that
π
ei 2 = i, eiπ = −1, ei2π = 1.

Version: December 30, 2025. 145

 Chapter 4.5 Example: Exponential and Trigonometric Functions

x2n+1 ∞
Proof. For x ∈ (0, 2], the sequence (2n+1)! is monotonically decreasing. Hence, by the


n=0
Leibniz criterion for alternating series (Proposition 4.22), the following estimates hold:

x3 x3 x5
x− ≤ sin(x) ≤ x − + ∀ x ∈ [0, 2]. (4.15)
3! 3! 5!
x2n ∞
Analogously, the sequence (2n)! is monotonically decreasing for x ∈ [0, 1], therefore


n=0

x2 x2 x4
1− ≤ cos(x) ≤ 1 − + ∀ x ∈ [0, 1]. (4.16)
2 2 24

Note that sin(0) = 0. Also, thanks to (4.15), for x = 1 we have

1 1
sin(1) ≥ 1 − >√ .
6 2

Thus, because sin is continuous, the Intermediate Value Theorem (Theorem 3.24) implies the
existence of p ∈ (0, 1) such that sin(p) = √12 .
Because sin2 (p) + cos2 (p) = 1 and cos(x) ≥ 1 − 12 x2 > 0 for x ∈ [0, 1] (see (4.16)), this
implies that
1
q
cos(p) = 1 − sin2 (p) = √ .
2
In other words,
1+i
21 eip = cos(p) + i sin(p) = √ .
2
Now, if we define π = 4p ∈ (0, 4), we obtain

π (1 + i)2 π
ei 2 = ei2p = (eip )2 = = i, eiπ = (ei 2 )2 = i2 = −1, ei2π = (−1)2 = 1.
2

In particular, since cos(π) + i sin(π) = eiπ = −1, we deduce that

sin(π) = 0, cos(π) = −1.

This proves that we have found a number π ∈ (0, 4) that satisfies all the desired properties.
It remains to prove uniqueness.
Assume there exists s ∈ (0, 4), with s ̸= π, satisfying sin(s) = 0. From the estimate

x3  x2 
sin(x) ≥ x − =x 1− >0 for x ∈ (0, 2],
3! 6

we see that sin(x) has no zeros in (0, 2]. Hence, we deduce that π, s ∈ (2, 4).
Now, define r = |π − s|, so that r ∈ (0, 2). Then, by Exercise 4.60 and the addition
formula (4.13), we get

| sin(r)| = | sin(π − s)| = sin(π) cos(s) − cos(π) sin(s) = 0.

| {z } | {z }
=0 =0

Version: December 30, 2025. 146

 Chapter 4.5 Example: Exponential and Trigonometric Functions

However, this is impossible since sin has no zeros in (0, 2). So, such a number s cannot exist,
and π ∈ (0, 4) is uniquely determined by the condition sin(π) = 0.

Corollary 4.63: Periodicity of Sine and Cosine

sin(x + π2 ) = cos(x), cos(x + π2 ) = − sin(x),

sin(x + π) = − sin(x), cos(x + π) = − cos(x),

sin(x + 2π) = sin(x), cos(x + 2π) = cos(x).

Proof. From Theorem 4.62 we know that

sin( π2 ) = 1, cos( π2 ) = 0, sin(π) = 0, cos(π) = −1, sin(2π) = 0, cos(2π) = 1.

Using these values together with the addition formulas (4.13), the identities follow directly.
Alternatively, one can observe that
π π
ei(x+ 2 ) = ei 2 eix = ieix =⇒ cos(x + π2 ) + i sin(x + π2 ) = i cos(x) − sin(x),

which implies that sin(x + π2 ) = cos(x) and cos(x + π2 ) = − sin(x). The other identities can
be proved analogously.

From Corollary 4.63 it follows that both sin and cos are periodic functions with period 2π.
21 To determine sin(x) or cos(x) for any real x, it suffices to know their values on the interval
[0, π2 ].

Figure 4.4: Graphs of the sine and cosine functions.

Exercise 4.64. — Show that the zeros of sin : R → R are precisely the points in πZ ⊂ R,
and the zeros of cos : R → R are exactly the points in πZ + π2 . Also, show that cos(x) = 1 if
and only if x = 2nπ with n ∈ Z.

Exercise 4.65. — Show that

   
x+y x−y
sin(x) − sin(y) = 2 cos sin
2 2

for all x, y ∈ R. Use this to show that sin : [− π2 , π2 ] → [−1, 1] is strictly increasing and hence
bijective.

Version: December 30, 2025. 147

 Chapter 4.5 Example: Exponential and Trigonometric Functions

4.5.4 Polar Coordinates and Multiplication of Complex Numbers

Using the complex exponential function, we can express complex numbers in polar coordi-
nates, that is, in the form
z = reiθ = r cos(θ) + ir sin(θ),

where r = |z| is the distance of z from the origin, and θ is the angle between the positive real
axis R≥0 and the segment from 0 to z. In other words, if z = x + iy, then
p
x = r cos(θ), y = r sin(θ), r= x2 + y 2 .

21

Figure 4.5: Polar representation of a complex number.

If z ̸= 0, the angle θ is uniquely determined and is called the argument of z, denoted

θ = arg(z). The set of all complex numbers with absolute value one is

S1 = {z ∈ C | |z| = 1} = {eiθ | θ ∈ [0, 2π)},

and is called the unit circle in C.

Proposition 4.66: Existence of Polar Coordinates

For every z ∈ C\{0} there exist uniquely determined real numbers r > 0 and θ ∈ [0, 2π)
such that z = reiθ .

Extra material: Proof of Proposition 4.66

Proof. Let r = |z| and define w = zr . Then |w| = |z|

r = 1. We must show that there exists
a unique θ ∈ [0, 2π) such that w = e . We first prove the existence of θ, distinguishing two
iθ

cases according to the sign of Im(w), and then establish uniqueness.

Version: December 30, 2025. 148

 Chapter 4.5 Example: Exponential and Trigonometric Functions

Existence when Im(w) ≥ 0. Since Re(w)2 + Im(w)2 = 1, we have Re(w) ∈ [−1, 1]. Hence,
since cos(0) = 1 and cos(π) = −1, the Intermediate Value Theorem 3.24 ensures the existence
of θ ∈ [0, π] such that Re(w) = cos(θ). Noticing that, for such θ, both Im(w) and sin(θ) are
nonnegative, we get
p p
sin(θ) = 1 − cos2 (θ) = 1 − Re(w)2 = Im(w),

therefore w = eiθ .
Existence when Im(w) < 0. Since −w satisfies Im(−w) > 0, by the previous case there exists
ϑ ∈ (0, π) such that −w = eiϑ . Since eiπ = −1, we obtain

w = eiπ eiϑ = ei(π+ϑ) ,

where θ = π + ϑ ∈ (π, 2π).

′ ′
Uniqueness. Let θ, θ′ ∈ [0, 2π) satisfy eiθ = eiθ . Then ei(θ−θ ) = 1, that is,

sin(θ − θ′ ) = 0, cos(θ − θ′ ) = 1.

Since θ − θ′ ∈ (−2π, 2π), Theorem 4.62 and Corollary 4.63 imply θ − θ′ = 0. Hence θ = θ′ ,
proving uniqueness.

In polar coordinates, multiplication of complex numbers takes a simple geometric form: if

z = reiφ and w = seiψ , then
zw = rsei(φ+ψ) .

Thus, when multiplying two complex numbers, their magnitudes multiply and their arguments
add.
zw

21 w

ψ+φ

φ ψ
r s rs

Figure 4.6: Multiplication of complex numbers in polar coordinates.

Version: December 30, 2025. 149

 Chapter 4.5 Example: Exponential and Trigonometric Functions

Exercise 4.67. — Let w = reiθ ̸= 0. Show that the n-th roots of w (i.e. the solutions
z ∈ C of z n = w) are given by
n√ θ+2πk
o
n
r e i( n ) k = 0, 1, . . . , n − 1 .

In particular, for w = 1 the n-th roots are

 i2πk/n
e k = 0, 1, . . . , n − 1 ,

and are called the n-th roots of unity.

Exercise 4.68. — For every integer n ≥ 2, show that

n−1
X
ei2πk/n = 0.
k=0

21 4.5.5 The Complex Logarithm

We defined the real logarithm as the inverse of the bijective mapping exp : R → R>0 . For
complex numbers, however, the exponential map exp : C → C is not injective, since exp(ix) =
1 for all x = 2πn with n ∈ Z. To define a complex logarithm, one must therefore restrict the
exponential function to a suitable subset D ⊂ C such that the restricted map exp |D : D → C×
becomes bijective. There are many possible choices for such a domain D, but a detailed
discussion of this topic lies beyond the scope of this course.

4.5.6 Other Trigonometric and Hyperbolic Functions

In addition to the exponential, sine, and cosine functions, several related trigonometric
functions are defined.
The tangent and cotangent functions are defined by

sin(x) cos(x)
tan(x) = , cot(x) = ,
cos(x) sin(x)

for all x ∈ R such that the denominators are nonzero.

Version: December 30, 2025. 150

 Chapter 4.5 Example: Exponential and Trigonometric Functions

Figure 4.7: Graphs of the tangent and cotangent functions.

Exercise 4.69. — Show that, for x, y ∈ R where both sides are defined,

tan(x) + tan(y)
tan(x + y) = .
1 − tan(x) tan(y)
21
Find and prove an analogous addition formula for the cotangent function.

The hyperbolic sine and hyperbolic cosine are defined by the power series
∞ ∞
X x2k+1 X x2k
sinh(x) = , cosh(x) = .
(2k + 1)! (2k)!
k=0 k=0

Equivalently,
ex − e−x ex + e−x
sinh(x) = , cosh(x) = ,
2 2
and hence ex = cosh(x) + sinh(x) for all x ∈ R.
The hyperbolic tangent and hyperbolic cotangent are defined by

sinh(x) ex − e−x cosh(x) ex + e−x

tanh(x) = = x , coth(x) = = x ,
cosh(x) e + e−x sinh(x) e − e−x

where coth(x) is defined for all x ∈ R \ {0} (since sinh(x) ̸= 0 for x ̸= 0).
The functions sinh and tanh are odd, while cosh is even. Also, they satisfy the addition
formulas

sinh(x + y) = sinh(x) cosh(y) + cosh(x) sinh(y),

cosh(x + y) = cosh(x) cosh(y) + sinh(x) sinh(y),

Version: December 30, 2025. 151

 Chapter 4.5 Example: Exponential and Trigonometric Functions

and the hyperbolic identity

cosh2 (x) − sinh2 (x) = 1 ∀ x ∈ R.

21

Figure 4.8: Graphs of the hyperbolic sine and cosine.

Exercise 4.70. — Starting from the definitions of sinh and cosh, prove the above identities.

Version: December 30, 2025. 152

 Chapter 5

Differential Calculus

In this chapter we deal with differential calculus in one variable. This is of fundamental
importance for understanding functions on R.

5.1 The Derivative

5.1.1 Definition and Geometrical Interpretation

In this section, D ⊆ R denotes a nonempty set with no isolated points; that is, every x ∈ D is
an accumulation point of D \ {x}. A typical example is a nonempty interval containing more
than one point.

Definition 5.1: Derivative

Let f : D → R be a function and x0 ∈ D. We say that f is differentiable at x0 if the
limit
f (x) − f (x0 ) f (x0 + h) − f (x0 )
f ′ (x0 ) = x→x
lim = lim (5.1)
0 x − x0 h→0 h
x̸=x0 h̸=0
22
exists. In this case we call f ′ (x0 ) the derivative of f at x0 . If f is differentiable at
every point of D, then we also say that f is differentiable on D, and we call the
resulting function f ′ : D → R the derivative of f .

To simplify notation, we will often write

f (x) − f (x0 ) f (x0 + h) − f (x0 )

f ′ (x0 ) = lim = lim ,
x→x0 x − x0 h→0 h

without explicitly mentioning that x ̸= x0 or h ̸= 0. Note that the condition

f (x) − f (x0 )
f ′ (x0 ) = lim
x→x0 x − x0

153
 Chapter 5.1 The Derivative

can be rewritten as
f (x) − f (x0 ) − f ′ (x0 )(x − x0 )
lim = 0,
x→x0 x − x0
or equivalently, using the little-o notation from Definition 3.69,

f (x) − f (x0 ) − f ′ (x0 )(x − x0 ) = o(x − x0 ). (5.2)

Remark 5.2. — If f : D → R is differentiable at x0 , then it is also continuous at x0 .

Indeed, using (5.2),

lim f (x) = lim f (x0 ) + f ′ (x0 )(x − x0 ) + o(x − x0 ) = f (x0 ),

x→x0 x→x0

hence f is continuous at x0 .
df
An alternative notation for the derivative of f is dx . If x0 ∈ D is a right accumulation
22 point of D, then f is differentiable from the right at x0 if the right derivative

f (x) − f (x0 ) f (x0 + h) − f (x0 )

f+′ (x0 ) = lim = lim
x→x+
0
x − x0 h→0+ h

exists. Differentiability from the left and the left derivative f−′ (x0 ) are defined analo-
gously using the limit x → x−
0.

Affine Functions
An affine function is a function of the form x 7→ sx + r, for real numbers s and r.
The graph of an affine function is a nonvertical line in R2 . The parameter s in the
equation y = sx + r is called the slope of the line.

If f : D → R is differentiable at x0 ∈ D, the function x 7→ f (x0 ) + f ′ (x0 )(x − x0 ) is called

the affine approximation of f at x0 .

Version: December 30, 2025. 154

 Chapter 5.1 The Derivative

(x0 +h, f (x0 +h))

f (x0 +h) − L(x0 +h) = o(h)

(x0 , f (x0 ))

L(x0 +h) = f (x0 ) + f ′ (x0 )h

h
x
x0 x0 +h

Figure 5.1: The figure shows the graph of f (blue) and its tangent at x0 (gray dashed), which
corresponds to the graph of the affine approximation L(x) = f (x0 ) + f ′ (x0 )(x − x0 ). Note
that, when we move a distance h from x0 to x0 + h, the vertical error between f (x0 + h) and
L(x0 + h) is o(h).

22
Example 5.3. — • Constant functions are differentiable everywhere and have the zero
function as their derivative.

• The identity function f (x) = x is differentiable, and its derivative is the constant func-
tion 1. Indeed,
x − x0
f ′ (x0 ) = lim =1 ∀ x0 ∈ R.
x→x0 x − x0

Example 5.4. — The exponential function exp : R → R>0 is differentiable and its deriva-
tive is again the exponential function. Indeed, for x ∈ R, using that ex+h = ex eh we get
P∞ hk P∞ hk ∞ ∞
ex+h − ex eh − 1 k=0 k! −1 k=1 k!
X hk−1 X hn
= ex = ex x
=e = ex = ex .
h h h h k! (n + 1)!
k=1 n=0

xn
We now observe that the power series x 7→ ∞ n=0 (n+1)! has infinite radius of convergence
P
xn
(see, e.g., Exercise 4.45); in particular, the function g(x) = ∞ n=0 (n+1)! is continuous on R
P

and g(0) = 1!1 = 1. Hence

ex+h − ex
(ex )′ = lim = ex lim g(h) = ex g(0) = ex .
h→0 h h→0

Version: December 30, 2025. 155

 Chapter 5.1 The Derivative

More generally, let α ∈ C and define f : R → C by f (x) = eαx . Then, arguing as before, we
get
∞
eαx+αh − eαx eαh − 1 X αk hk−1
(eαx )′ = lim = eαx lim = eαx lim
h→0 h h→0 h h→0 k!
k=1
∞ ∞
X (αh)k−1 X (αh)n
= αeαx lim = αeαx lim = αeαx .
h→0 k! h→0 (n + 1)!
k=1 n=0

Example 5.5. — Let f : R \ {0} → R be given by f (x) = x1 . Then f is differentiable and

f ′ (x) = − x12 for all x ∈ R \ {0}:

1 1
x+h − x x − (x + h) 1 1
f ′ (x) = lim = lim = − lim = − 2.
h→0 h h→0 (x + h) x h h→0 (x + h) x x

Definition 5.6: Higher Derivatives

Let f : D → R be a function. We define the higher derivatives of f , if they exist, by

f (0) = f, f (1) = f ′ , f (2) = f ′′ , ..., f (n+1) = (f (n) )′

for all n ∈ N. If f (n) exists, we say that f is n-times differentiable. If the nth
22 derivative f (n) is also continuous, f is called n-times continuously differentiable.
We denote the set of n-times continuously differentiable functions on D by C n (D).

Equivalently, C 0 (D) is the set of real-valued continuous functions on D, and C 1 (D) is the
set of all differentiable functions whose derivative is continuous (these are called continuously
differentiable or of class C 1 ). Recursively, for n ≥ 1,

C n (D) = { f : D → R | f is differentiable and f ′ ∈ C n−1 (D) },

and we say that f ∈ C n (D) is of class C n .

Example 5.7. — Consider the function f : R → R defined by f (x) = sgn(x) x2 , or

equivalently 
x2 , x ≥ 0,
f (x) =
−x2 , x < 0.

For x > 0 and x < 0, we can compute the derivative directly:


2x, x > 0,
f ′ (x) =
−2x, x < 0.

Version: December 30, 2025. 156

 Chapter 5.1 The Derivative

For x = 0, we have

f (x) − f (0) sgn(x) x2

= = sgn(x) x −→ 0 as x → 0,
x−0 x

so f ′ (0) = 0. Altogether,

2x,

 x > 0,

′
f (x) = 0, x = 0, that is, f ′ (x) = 2|x|.


−2x, x < 0,


Hence f is continuously differentiable, i.e. f ∈ C 1 (R). However, since f ′ (x) = 2|x| is not
differentiable at x = 0, the function f is not of class C 2 .

Definition 5.8: Smooth Functions

We define
∞
\
C ∞ (D) = C n (D) = { f : D → R | f is differentiable infinitely many times },
n=0

and call functions f ∈ C ∞ (D) smooth or of class C ∞ .

22

Example 5.9. — The exponential function exp : R → R is smooth.

5.1.2 Differentiation Rules

As with continuous functions, we rarely reprove differentiability from first principles for each
new example; instead, we use general rules that reduce differentiability of compound expres-
sions to that of simpler ones.

Proposition 5.10: Derivative of Sum and Product

Let D ⊆ R and let x0 ∈ D be an accumulation point of D \ {x0 }. Let f, g : D → R be
differentiable at x0 . Then f + g and f · g are differentiable at x0 , and

(f + g)′ (x0 ) = f ′ (x0 ) + g ′ (x0 ), (5.3)

(f g)′ (x0 ) = f ′ (x0 ) g(x0 ) + f (x0 ) g ′ (x0 ). (5.4)

In particular, for any α ∈ R, the scalar multiple αf is differentiable at x0 and

(αf )′ (x0 ) = αf ′ (x0 ).

Version: December 30, 2025. 157

 Chapter 5.1 The Derivative

Proof. Using the properties of limit discussed in Section 3.5.1, we have

 
(f + g)(x) − (f + g)(x0 ) f (x) − f (x0 ) g(x) − g(x0 )
lim = lim +
x→x0 x − x0 x→x0 x − x0 x − x0
f (x) − f (x0 ) g(x) − g(x0 )
= lim + lim
x→x0 x − x0 x→x 0 x − x0
= f ′ (x0 ) + g ′ (x0 ),

and

(f g)(x) − (f g)(x0 ) (f (x) − f (x0 )) g(x) + f (x0 ) (g(x) − g(x0 ))

lim = lim
x→x0 x − x0 x→x0 x − x0
 
f (x) − f (x0 ) g(x) − g(x0 )
= lim g(x) + f (x0 ) lim
x→x0 x − x0 x→x 0 x − x0
 f (x) − f (x0 )    g(x) − g(x0 )
= lim · lim g(x) + f (x0 ) lim
x→x0 x − x0 x→x0 x→x0 x − x0
= f ′ (x0 )g(x0 ) + f (x0 )g ′ (x0 ),

where we used that g is continuous at x0 (see Remark 5.2) to conclude limx→x0 g(x) = g(x0 ).

Corollary 5.11: Higher Order Derivatives of the Product

Let f, g : D → R be n-times differentiable. Then f + g and f · g are also n-times
22
differentiable, with (f + g)(n) = f (n) + g (n) and
n  
(n)
X n
(f g) = f (k) g (n−k) .
k
k=0

In particular, for every α ∈ R, (αf )(n) = αf (n) .

Proof. For n = 1 this is Proposition 5.10. The general case follows by induction on n ≥ 1.

Corollary 5.12: Derivatives of Polynomials

Polynomial functions are differentiable on all of R. Moreover, (1)′ = 0 and (xn )′ =
nxn−1 for all n ≥ 1.

Proof. We argue by induction. The cases n = 0 and n = 1 were covered in Example 5.3. For
n > 1, assume (xn )′ = nxn−1 . Then, by (5.4), since xn+1 = x · xn ,

(xn+1 )′ = (x · xn )′ = 1 · xn + x · (nxn−1 ) = (n + 1)xn .

This proves the inductive step and establishes the result. Finally, the linearity of the derivative
(see (5.3)) yields the differentiability of any polynomial.

Version: December 30, 2025. 158

 Chapter 5.1 The Derivative

Example 5.13. — From Example 5.4 with α = ±1 and α = ±i,

(ex )′ = ex , (e−x )′ = −e−x , (eix )′ = ieix , (e−ix )′ = −ie−ix .

By Theorem 4.61,

(eix )′ − (e−ix )′ eix + e−ix

sin′ (x) = = = cos(x),
2i 2

and analogously cos′ (x) = − sin(x). Similarly, sinh′ (x) = cosh(x) and cosh′ (x) = sinh(x).

Theorem 5.14: Chain Rule

Let D, E ⊆ R, and let x0 ∈ D be an accumulation point of D \ {x0 }. Let f : D → E be
differentiable at x0 such that y0 = f (x0 ) is an accumulation point of E \ {y0 }, and let
g : E → R be differentiable at y0 . Then g ◦ f : D → R is differentiable at x0 and

(g ◦ f )′ (x0 ) = g ′ (f (x0 )) f ′ (x0 ).

Remark 5.15. — Heuristically, we would like to compute the derivative of the composition
by writing
g(f (x)) − g(f (x0 )) g(f (x)) − g(f (x0 )) f (x) − f (x0 )
= · ,
22 x − x0 f (x) − f (x0 ) x − x0
and then take the limit as x → x0 to get

(g ◦ f )′ (x0 ) = g ′ (f (x0 )) f ′ (x0 ).

This argument proves the result if f (x) ̸= f (x0 ) for x near x0 , but it is not a proof as stated:
if f (x) = f (x0 ) for infinitely many x close to x0 , the fraction g(ff(x))−g(f (x0 ))
(x)−f (x0 ) is undefined, and
one faces a 0/0 indeterminacy. In the proof below, we will provide a rigorous argument, but
it is helpful to keep the above heuristic in mind as the underlying intuition for the chain rule.

Proof. Observe that we can write

g(y) = g(y0 ) + [g(y) − g(y0 )]

= g(y0 ) + g ′ (y0 ) (y − y0 ) + [g(y) − g(y0 ) − g ′ (y0 ) (y − y0 )]
= g(y0 ) + g ′ (y0 ) (y − y0 ) + ω(y) (y − y0 ),

where the function ω : E → R is defined as


 g(y) − g(y0 ) − g ′ (y0 )

for y ∈ E \ {y0 },
ω(y) = y − y0
for y = y0 .

0

Version: December 30, 2025. 159

 Chapter 5.1 The Derivative

Since g is differentiable at y0 , it follows that ω(y) → 0 as y → y0 ; hence, the function ω is

continuous at y0 . Substituting y = f (x) and using y0 = f (x0 ),

g(f (x)) = g(f (x0 )) + g ′ (f (x0 )) [f (x) − f (x0 )] + ω(f (x)) [f (x) − f (x0 )],

therefore
 
g(f (x)) − g(f (x0 )) f (x) − f (x0 ) f (x) − f (x0 )
lim = lim g ′ (f (x0 )) + ω(f (x))
x→x0 x − x0 x→x0 x − x0 x − x0

22 = g ′ (f (x0 )) f ′ (x0 ) + ω(f (x0 )) f ′ (x0 ) = g ′ (f (x0 )) f ′ (x0 ),

| {z }
=0

where we used the continuity of ω at y0 = f (x0 ) to deduce that ω(f (x)) → ω(f (x0 )) = 0 as
x → x0 .

Remark 5.16. — By a nontrivial induction argument, if f : D → E and g : E → R are

n-times differentiable, then g ◦ f : D → R is n-times differentiable, and one can express the
nth derivative of g ◦ f in terms of sums and products of g (k) ◦ f and f (j) with 1 ≤ k, j ≤ n.
This is known as Faà di Bruno’s formula, but we will not explore it here.

Corollary 5.17: Quotient Rule

Let D ⊆ R, let x0 ∈ D be an accumulation point of D \ {x0 }, and let f, g : D → R be

differentiable at x0 . If g(x0 ) ̸= 0, then fg is differentiable at x0 and
 ′
f f ′ (x0 )g(x0 ) − f (x0 )g ′ (x0 )
(x0 ) = .
g g(x0 )2

Proof. Consider the function ψ : R\{0} → R given by ψ(y) = y1 . This function is differentiable
by Example 5.5, with ψ ′ (y) = − y12 . Then, by the chain rule (Theorem 5.14), g1 = ψ ◦ g is
differentiable at x0 , with
23
 ′
1 g ′ (x0 )
(x0 ) = ψ ′ (g(x0 ))g ′ (x0 ) = − .
g g(x0 )2

f
Applying now the product rule (Proposition 5.10), g =f· 1
g is differentiable at x0 , and
 ′
1 ′ g ′ (x0 ) f ′ (x0 )g(x0 ) − f (x0 )g ′ (x0 )
 
f 1
(x0 ) = f · (x0 ) = f ′ (x0 ) − f (x0 ) = .
g g g(x0 ) g(x0 )2 g(x0 )2

Example 5.18. — We want to compute the derivative of the function f : R → R given by

Version: December 30, 2025. 160

 Chapter 5.1 The Derivative

f (x) = exp sin sin(x2 ) . For this, let g(x) = sin sin(x2 ) , h(x) = sin(x2 ), and k(x) = x2 .

Then, since

f (x) = exp(g(x)), g(x) = sin(h(x)), h(x) = sin(k(x)),

applying the chain rule repeatedly and using that exp′ = exp and sin′ = cos, we get

f ′ (x) = exp(g(x))g ′ (x), g ′ (x) = cos(h(x)) h′ (x), h′ (x) = cos(k(x)) k ′ (x), k ′ (x) = 2x,

therefore
f ′ (x) = exp sin sin(x2 ) cos sin(x2 ) cos(x2 ) 2x




∀ x ∈ R.

Exercise 5.19. — Determine the derivative of the function x 7→ cos sin3 (exp(x)) .

In the next theorem, given f : D → E a continuous bijection, we investigate the derivative

of the inverse of a function. Notice that, for the derivative of f −1 to be well-defined at a
point ȳ = f (x̄), we must ensure that if x̄ ∈ D is an accumulation point of D \ {x̄}, then
f (x̄) is an accumulation point of E \ {f (x̄)}. This follows from continuity and injectivity: if
(xn )∞n=0 ⊂ D \ {x̄} satisfies xn → x̄, then (f (xn ))n=0 ⊂ E \ {f (x̄)} and f (xn ) → f (x̄).
∞

Theorem 5.20: Derivative of the Inverse

23
Let D, E ⊆ R, and let f : D → E be a continuous bijection whose inverse f −1 : E → D
is also continuous. Let x̄ ∈ D be an accumulation point of D \ {x̄}, and assume that f
is differentiable at x̄ with f ′ (x̄) ̸= 0. Then f −1 is differentiable at ȳ = f (x̄) and

1 1
(f −1 )′ (ȳ) = = .
f ′ (x̄) f ′ (f −1 (ȳ))

Proof. To compute (f −1 )′ (ȳ), take a sequence (yn )∞

n=0 ⊂ E \ {ȳ} with yn → ȳ and set
xn = f (yn ). Then
−1

−1
f −1 (yn ) − f −1 (ȳ)

xn − x̄ f (xn ) − f (x̄)
= = .
yn − ȳ f (xn ) − f (x̄) xn − x̄

By the continuity of f −1 , we have xn → x̄ = f −1 (ȳ). Thus, since f is differentiable at x̄ with

f ′ (x̄) ̸= 0, Proposition 2.94(4) implies
 −1
f (xn ) − f (x̄) 1
−→ ,
xn − x̄ f ′ (x̄)

proving that
f −1 (yn ) − f −1 (ȳ) 1
lim = ′ .
n→∞ yn − ȳ f (x̄)

Version: December 30, 2025. 161

 Chapter 5.1 The Derivative

Since the sequence (yn )∞

n=0 was arbitrary, Lemma 3.57 gives

f −1 (y) − f −1 (ȳ) 1
lim = ′ ,
y→ȳ y − ȳ f (x̄)

as desired.

f ′ (x0 ) x=y
Slope

1
f ′ (x0 )
Slope
(x0 , y0 )
f −1

(y0 , x0 )

23 Figure 5.2: An intuitive representation of Theorem 5.20. Let y0 = f (x0 ). Reflecting the graph
of f and the tangent line at (x0 , y0 ) across the line x = y in R2 yields the graph of f −1 and,
as asserted, the tangent line at (y0 , x0 ). A short calculation shows that the reflection of a line
with slope m across x = y has slope 1/m.

Example 5.21. — The function g : R \ {0} → R defined by g(y) = log |y| is differentiable,
with g ′ (y) = y1 for all y ∈ R \ {0}. Indeed, let f = exp : R → R>0 , so that log = f −1 . By
Theorem 5.20,
1
log′ (y) = ′ .
f (log(y))
Since exp′ = exp, for y > 0 we have

1 1
log′ (y) = = .
exp(log(y)) y

For y < 0, since g(y) = log(−y), by chain rule (Theorem 5.14) and applying the case above
to −y > 0, we get
1 1
g ′ (y) = − log′ (−y) = − = .
−y y

Example 5.22. — Given x > 0 and α ∈ R, we can compute the derivative of xα via

α
xα = exp α log x =⇒ (xα )′ = exp′ α log x α log′ (x) = exp α log x = αxα−1 .



x

Version: December 30, 2025. 162

 Chapter 5.1 The Derivative

This generalizes Corollary 5.12.

Example 5.23. — The logarithm f = log : R>0 → R is smooth. Indeed, by the example
above, f ′ (x) = x1 . By induction (using the Leibniz rule) one gets

1 2 (n − 1)!
f ′′ (x) = − , f (3) (x) = , and in general f (n) (x) = (−1) n−1 ∀ n ≥ 1.
x2 x3 xn

Exercise 5.24. — Consider the function ψ : R → R defined by

  
exp − 1 , x > 0,

ψ(x) = x
x ≤ 0.

0,

Show that ψ is smooth on R, and that all its derivatives at 0 vanish.

23

Hint: Show first by induction that, for all n ∈ N,

   
1 1
ψ (n)
(x) = exp − fn (x > 0), (5.5)
x x

where fn is a polynomial. Then, using that exp(−1/x) tends to 0 faster than any power of x
as x → 0+ (equivalently, exp(y) dominates every polynomial as y → +∞, by Corollary 3.47),
prove that
1


(n+1) ψ (n) (x) − ψ (n) (0) ψ(x)fn x 1

1
ψ (0) = lim = lim = lim ψ(x) fn x =0 ∀ n ∈ N.
x→0 x x→0+ x x→0+ x

Version: December 30, 2025. 163

 Chapter 5.2 Main Theorems of Differential Calculus

5.2 Main Theorems of Differential Calculus

5.2.1 Local Extrema

Definition 5.25: Local extrema

Let D ⊆ R and x0 ∈ D. We say that a function f : D → R has a local maximum at
x0 if there exists δ > 0 such that

f (x) ≤ f (x0 ) ∀ x ∈ D ∩ (x0 − δ, x0 + δ).

If the inequality is strict (i.e. f (x) < f (x0 ) for all x ∈ D ∩ (x0 − δ, x0 + δ) \ {x0 }), then
f has a strict local maximum at x0 .
A (strict) local minimum is defined analogously.
We call x0 a local extremum of f if f has either a local minimum or a local maximum
at x0 .

Proposition 5.26: Local Extrema vs. First Derivative

Let D ⊆ R and f : D → R. Suppose x0 ∈ D is a local extremum of f , that f is
differentiable at x0 , and that x0 is both a right-hand and a left-hand accumulation point
23 of D. Then
f ′ (x0 ) = 0.

Proof. Without loss of generality, assume f has a local maximum at x0 (otherwise replace f
by −f ). We first note that, for x close to x0 and to the right of it, we have f (x) − f (x0 ) ≤ 0
and x − x0 > 0. Hence,
f (x) − f (x0 )
f+′ (x0 ) = lim ≤ 0.
x→x+0
x − x0

Similarly, for x close to x0 and to the left of it, we have f (x) − f (x0 ) ≤ 0 and x − x0 < 0, so

f (x) − f (x0 )
f−′ (x0 ) = lim ≥ 0,
x→x−
0
x − x0

Since f is differentiable at x0 , the two one-sided derivatives coincide, that is,

f ′ (x0 ) = f+′ (x0 ) = f−′ (x0 ),

and therefore f ′ (x0 ) = 0.

An immediate consequence of the previous result is the following:

Version: December 30, 2025. 164

 Chapter 5.2 Main Theorems of Differential Calculus

Corollary 5.27: Local Extrema in an Interval

Let I ⊆ R be an interval and f : I → R. If x0 ∈ I is a local extremum of f , then at
least one of the following holds:

1. x0 is an endpoint of I;

2. f is not differentiable at x0 ;

3. f is differentiable at x0 and f ′ (x0 ) = 0.

In particular, all local extrema of a differentiable function on an open interval are zeros
of the derivative.

Exercise 5.28. — Let f : R → R be the polynomial f (x) = x3 − x.

(i) Find all local extrema of f .
(ii) Find all local extrema of the function |f | on [−3, 3].

5.2.2 The Mean Value Theorem

We now turn to general theorems of differential calculus and their consequences. Our first
23 question is whether the derivative of a differentiable function attains the slope of certain
secants; the next theorem is the starting point.

Theorem 5.29: Rolle’s Theorem

Let f : [a, b] → R be continuous on [a, b] and differentiable on (a, b). If f (a) = f (b),
then there exists ξ ∈ (a, b) with f ′ (ξ) = 0.

Proof. By Theorem 3.34, f attains its minimum and maximum on [a, b] at some points x0 , x1 ∈
[a, b]. By Proposition 5.26, any interior extremum has zero derivative. Thus, we consider two
cases:
(i) If either x0 or x1 lies in (a, b), we are done.
(ii) If both x0 and x1 are endpoints, since f (a) = f (b) then min f = max f = f (a) = f (b),
therefore f is constant. In particular, f ′ (x) = 0 for all x ∈ (a, b) and the result follows also in
this case.

An immediate consequence of this result is the following:

Corollary 5.30: Non-Vanishing Derivative implies Different Endpoints

Values
Let f : [a, b] → R be continuous on [a, b] and differentiable on (a, b). If f ′ (x) ̸= 0 for
all x ∈ (a, b), then f (a) ̸= f (b).

Version: December 30, 2025. 165

 Chapter 5.2 Main Theorems of Differential Calculus

Proof. If f (a) = f (b), then Rolle’s Theorem would imply the existence of ξ ∈ (a, b) with
f ′ (ξ) = 0, contradicting the assumption that f ′ never vanishes.

Theorem 5.31: Mean Value Theorem

Let f : [a, b] → R be continuous on [a, b] and differentiable on (a, b). Then there exists
ξ ∈ (a, b) such that
f (b) − f (a)
f ′ (ξ) = .
b−a

Proof. Define g : [a, b] → R by

f (b) − f (a)
g(x) = f (x) − (x − a).
b−a

Then g is continuous on [a, b], differentiable on (a, b), and satisfies

g(a) = f (a), g(b) = f (b) − (f (b) − f (a)) = f (a).

By Rolle’s Theorem (Theorem 5.29), there exists ξ ∈ (a, b) with

23 f (b) − f (a)
0 = g ′ (ξ) = f ′ (ξ) − ,
b−a

proving the result.

In words, Rolle’s Theorem states that if a differentiable function on an interval takes the
same value at both endpoints, then its slope must be zero somewhere in between (left image).
The Mean Value Theorem, on the other hand, asserts that for any differentiable function
on an interval, there exists a point where the slope equals the average slope (right image).
Moreover, the Mean Value Theorem can be reduced to Rolle’s Theorem by subtracting from
f a suitable linear function so that the endpoint values coincide.

f f (b)
f
f (a) = f (b)

f (a)

a b a b

Using the Mean Value Theorem, we can show that for differentiable functions on an interval,
24 the notion of Lipschitz continuity introduced in Exercise 3.41 is equivalent to having a bounded
derivative.

Version: December 30, 2025. 166

 Chapter 5.2 Main Theorems of Differential Calculus

Corollary 5.32: Lipschitz Continuity vs Bounded Derivative

Let f : [a, b] → R be continuous on [a, b] and differentiable on (a, b). Then f is Lipschitz
continuous on [a, b] if and only if f ′ is bounded on (a, b).

Proof. Suppose first that f is Lipschitz on [a, b] with constant L. This implies that, given
x, x0 ∈ (a, b) with x ̸= x0 ,
f (x) − f (x0 )
≤ L.
x − x0
Taking the limit as x → x0 gives |f ′ (x0 )| ≤ L, so f ′ is bounded on (a, b).
Conversely, suppose f ′ is bounded on (a, b), say |f ′ | ≤ M for all z ∈ (a, b). Then, given
x, y ∈ [a, b] with x < y, the Mean Value Theorem (Theorem 5.31) applied on the interval
[x, y] yields ξ ∈ (x, y) ⊂ (a, b) with

f (y) − f (x) = f ′ (ξ) (y − x),

therefore
|f (y) − f (x)| = |f ′ (ξ)| |y − x| ≤ M |y − x|.

Since x, y ∈ [a, b] are arbitrary, this shows that f is Lipschitz on [a, b] with Lipschitz constant
M.
24

Exercise 5.33. — Let [a, b] be a compact interval and f : [a, b] → R be continuously

differentiable. Show that f is Lipschitz continuous. What happens if compactness is dropped?

Example 5.34. — Let f : [0, 2π] → C be given by f (x) = eix = cos x + i sin x. Then
f (0) = f (2π) = 1, but
f ′ (x) = ieix ̸= 0 ∀ x ∈ [0, 2π].

Thus Rolle’s theorem and the Mean Value Theorem fail for complex-valued functions.

Theorem 5.35: Cauchy Mean Value Theorem

Let f, g : [a, b] → R be continuous on [a, b] and differentiable on (a, b). Then there exists
ξ ∈ (a, b) such that

g ′ (ξ) f (b) − f (a) = f ′ (ξ) g(b) − g(a) . (5.6)

If, in addition, g ′ (x) ̸= 0 for all x ∈ (a, b), then g(a) ̸= g(b) and

f ′ (ξ) f (b) − f (a)

′
= .
g (ξ) g(b) − g(a)

Version: December 30, 2025. 167

 Chapter 5.2 Main Theorems of Differential Calculus

Proof. Define the function F : [a, b] → R as

F (x) = g(x) f (b) − f (a) − f (x) g(b) − g(a) .

Then



F (a) = g(a) f (b) − f (a) − f (a) g(b) − g(a) = g(a)f (b) − f (a)g(b),



F (b) = g(b) f (b) − f (a) − f (b) g(b) − g(a) = g(a)f (b) − f (a)g(b).

Thus, by Rolle’s Theorem (Theorem 5.29), there exists ξ ∈ (a, b) with

0 = F ′ (ξ) = g ′ (ξ) f (b) − f (a) − f ′ (ξ) g(b) − g(a) ,

which is (5.6).
If g ′ (x) ̸= 0 for all x, then Corollary 5.30 yields g(a) ̸= g(b). Dividing (5.6) by g ′ (ξ) g(b) −
g(a) yields the second formula.

As with the Mean Value Theorem 5.31, Cauchy’s theorem has a geometric interpretation:
under the stated assumptions, the curve t 7→ f (t), g(t) has a tangent parallel to the line

through (f (a), g(a)) and (f (b), g(b)).

g(x)
24

g(b)

g(a)

f (x)
f (a) f (b)

The goal of the next exercises is to show that derivatives satisfy the intermediate value
property: although f ′ need not be continuous, it cannot have jump discontinuities. This result
is known as Darboux’s Theorem.

Exercise 5.36. — Follow the steps below to prove the following statement:
Let f : (a, b) → R be differentiable, and let x0 < x1 be points in (a, b) with f ′ (x0 ) ̸= f ′ (x1 ).
Show that, for every α between f ′ (x0 ) and f ′ (x1 ), there exists c ∈ (x0 , x1 ) such that f ′ (c) = α.

Version: December 30, 2025. 168

 Chapter 5.2 Main Theorems of Differential Calculus

1. Without loss of generality, assume f ′ (x0 ) < f ′ (x1 ) and fix α ∈ (f ′ (x0 ), f ′ (x1 )).

2. Consider the auxiliary function g(x) = f (x) − αx and show that g attains a minimum
at some c ∈ [x0 , x1 ].

3. Prove that c cannot be one of the endpoints:

(i) Since
g(x0 + h) − g(x0 )
−−−−→ g ′ (x0 ) = f ′ (x0 ) − α,
h h→0+

and f ′ (x0 ) < α, conclude that c ̸= x0 .

(ii) Similarly, using that

g(x1 ) − g(x1 − h)
−−−−→ g ′ (x1 ) = f ′ (x1 ) − α,
h h→0+

and f ′ (x1 ) > α, conclude that c ̸= x1 .

4. Deduce that c ∈ (x0 , x1 ), and use Proposition 5.26 to show that g ′ (c) = 0, hence
f ′ (c) = α.

24 5.2.3 L’Hôpital’s Rule

The results collectively known as L’Hôpital’s rule (also spelled L’Hospital) are named after
Guillaume François Antoine, Marquis de l’Hôpital (1661–1704). The rule likely goes back
to Johann Bernoulli, but was published by de l’Hôpital in his Analyse des Infiniment Petits
pour l’Intelligence des Lignes Courbes, the first systematic treatment of infinitesimal calculus.
However his approach was thoroughly geometric, as he had neither a rigorous notion of limit
nor of differentiability.

Theorem 5.37: L’Hôpital’s Rule

Let f, g : (a, b) → R be differentiable. Suppose:

1. g(x) ̸= 0 and g ′ (x) ̸= 0 for all x ∈ (a, b);

2. lim f (x) = lim g(x) = 0;

x→a+ x→a+

f ′ (x)
3. the limit L = lim exists.
x→a+ g ′ (x)
f (x)
Then lim exists and equals L.
x→a+ g(x)

Version: December 30, 2025. 169

 Chapter 5.2 Main Theorems of Differential Calculus

Proof. By (2), we can extend f and g continuously to [a, b) by setting f (a) = g(a) = 0. Fix
ε > 0. By (3), there exists δ > 0 such that

f ′ (ξ)
∈ (L − ε, L + ε) ∀ ξ ∈ (a, a + δ).
g ′ (ξ)

Now, for any x ∈ (a, a + δ), we can apply Cauchy’s Mean Value Theorem (Theorem 5.35) to
f and g on [a, x] to obtain some ξx ∈ (a, x) with

f (x) f (x) − f (a) f ′ (ξx )

= = ′ .
g(x) g(x) − g(a) g (ξx )

Since ξx ∈ (a, x) ⊂ (a, a + δ), it follows that

f (x) f ′ (ξx )
= ′ ∈ (L − ε, L + ε) for all x ∈ (a, a + δ).
g(x) g (ξx )

f (x)
24 Because ε > 0 is arbitrary, this proves that lim = L.
x→a+ g(x)
Theorem 5.37 is one of several versions of L’Hôpital’s rule. For instance, one can allow
both limits in (2) to be improper, i.e., limx→a+ g(x) = ±∞ and limx→a+ f (x) = ±∞ (with
arbitrary signs). More precisely:

Theorem 5.38: L’Hôpital’s Rule for Improper Limits

Let f, g : (a, b) → R be differentiable. Suppose:

1. g(x) ̸= 0 and g ′ (x) ̸= 0 for all x ∈ (a, b);

2. lim |f (x)| = lim |g(x)| = ∞;

x→a+ x→a+

f ′ (x)
3. the limit L = lim exists.
x→a+ g ′ (x)
f (x)
Then lim exists and equals L.
x→a+ g(x)

Extra material: Proof of Theorem 5.38

Proof. Fix ε > 0. By (3) there exists δ > 0 such that

f ′ (ξ)
∈ (L − ε, L + ε) ∀ ξ ∈ (a, a + δ).
g ′ (ξ)

For x ∈ (a, a + δ), apply Cauchy’s Mean Value Theorem (Theorem 5.35) on [x, a + δ] to obtain
ξx ∈ (x, a + δ) with
f (x) − f (a + δ) f ′ (ξx )
= ′ ∈ (L − ε, L + ε).
g(x) − g(a + δ) g (ξx )

Version: December 30, 2025. 170

 Chapter 5.2 Main Theorems of Differential Calculus

Hence
f (x) − f (a + δ)
∈ (L − ε, L + ε) ∀ x ∈ (a, a + δ). (5.7)
g(x) − g(a + δ)
We now observe that
g(a+δ)
f (x) f (x) − f (a + δ) 1 − g(x)
= · f (a+δ)
. (5.8)
g(x) g(x) − g(a + δ) 1 −
f (x)

Since |f (x)|, |g(x)| → ∞ as x → a+ , the second factor in (5.8) tends to 1. Thus, combining
(5.7) and (5.8), there exists η ∈ (0, δ) such that

f (x)
∈ (L − 2ε, L + 2ε) ∀ x ∈ (a, a + η).
g(x)

f (x)
Since ε > 0 is arbitrary, lim = L.
x→a+ g(x)

Example 5.39. — We illustrate the use of l’Hôpital’s rule (applied repeatedly) by com-
puting
sin x − x
lim .
x→0 x3
First, note that
lim (sin x − x) = 0, lim x3 = 0,
x→0 x→0

so we have an indeterminate form of type 0

0 and l’Hôpital’s rule applies. Therefore

sin x − x (sin x − x)′ cos x − 1

lim = lim = lim .
x→0 x3 x→0 (x3 )′ x→0 3x2

At this stage we still have

24
lim (cos x − 1) = 0, lim 3x2 = 0,
x→0 x→0

so the expression is again of type 0

0 and we may apply l’Hôpital’s rule a second time:

cos x − 1 (cos x − 1)′ − sin x

lim 2
= lim 2 ′
= lim .
x→0 3x x→0 (3x ) x→0 6x

Since
lim (− sin x) = 0, lim 6x = 0,
x→0 x→0

we still have a 0
0 form, and we apply l’Hôpital’s rule a third time:

− sin x − cos x 1
lim = lim =− .
x→0 6x x→0 6 6

Version: December 30, 2025. 171

 Chapter 5.2 Main Theorems of Differential Calculus

Thus
sin x − x 1
lim 3
=− .
x→0 x 6
This example shows how l’Hôpital’s rule can be iterated, provided that at each step the new
quotient is still of indeterminate form 00 (or ∞
∞ ) and the hypotheses of the theorem remain
satisfied.
Instead of the one-sided limits x → a+ or x → b− , one can also consider x → −∞ or
x → ∞:

Theorem 5.40: L’Hôpital’s Rule at Infinity

Let R > 0 and f, g : (R, ∞) → R be differentiable. Suppose:

1. g(x) ̸= 0 and g ′ (x) ̸= 0 for all x ∈ (R, ∞);

2. either lim f (x) = lim g(x) = 0 or lim |f (x)| = lim |g(x)| = ∞;

x→∞ x→∞ x→∞ x→∞

f ′ (x)
3. the limit L = lim exists.
x→∞ g ′ (x)

f (x)
Then lim exists and equals L.
x→∞ g(x)

Proof. If lim f (x) = lim g(x) = 0, apply Theorem 5.37 on 0, R1 to the functions x 7→ f x1



x→∞
x→∞
and x 7→ g x1 . If lim |f (x)| = lim |g(x)| = ∞, apply instead Theorem 5.38 on 0, R1 to the


24 x→∞ x→∞
functions x 7→ f x1 and x 7→ g x1 .

Remark 5.41. — As discussed in the following exercise, the proofs of Theorems 5.37, 5.38,
′ (x)
and 5.40 also apply (with small modifications) when fg′ (x) diverges to +∞ or −∞. In this
f (x) f ′ (x)
case, one concludes that g(x) diverges to the same infinite limit as g ′ (x) .

f ′ (x)
Exercise 5.42. — Prove Theorem 5.37 in the case where limx→a+ g ′ (x) = +∞.
f ′ (ξ)
Hint: Repeat the argument in the proof of Theorem 5.37, replacing the condition g ′ (ξ) ∈
f ′ (ξ)
(L − ε, L + ε) with inequalities of the form g ′ (ξ) > M for arbitrary M > 0.

Exercise 5.43. — Compute the following limits using L’Hôpital’s rule:

sin x − x ex − x − 1 x4 − 4x
(a) lim , (b) lim , (c) lim , (d) lim x3 ex .
x→0+ x2 sin x x→0 cos x − 1 x→2 sin(πx) x→−∞

Exercise 5.44. — Let a < b and let f : [a, b] → R be continuous. Suppose x0 ∈ [a, b], that
f is differentiable on [a, b] \ {x0 }, and that limx→x0 f ′ (x) exists. Show that f is differentiable
at x0 and that f ′ is continuous at x0 .

Version: December 30, 2025. 172

 Chapter 5.2 Main Theorems of Differential Calculus

Exercise 5.45. — Let I = (a, b) ⊆ R and let f : I → R be twice differentiable. Applying

L’Hôpital’s rule twice, show that

f (x + h) − 2f (x) + f (x − h)
f ′′ (x) = lim ∀ x ∈ I.
h→0 h2

5.2.4 Monotonicity and Convexity via Differential Calculus

The mean value theorem allows us to characterize some properties of functions via the deriva-
tive. In what follows, f ′ ≥ 0 means f ′ (x) ≥ 0 for all x ∈ I. Here I always denotes a nontrivial
interval (nonempty and not a single point).

Proposition 5.46: Monotonicity vs. First Derivative

Let I ⊆ R be an interval, and let f : I → R be differentiable. Then

f ′ ≥ 0 ⇐⇒ f is increasing.

Proof. If f is increasing, then f (x + h) − f (x) ≥ 0 for h > 0, and f (x + h) − f (x) ≤ 0 for

h < 0. Hence, in both cases, f (x+h)−f
h
(x)
≥ 0, therefore

f (x + h) − f (x)
f ′ (x) = lim ≥ 0.
h→0 h
24 Conversely, suppose f is not increasing. Then there exist x1 < x2 with f (x2 ) < f (x1 ). By
the Mean Value Theorem 5.31, there exists ξ ∈ (x1 , x2 ) such that

f (x2 ) − f (x1 )
f ′ (ξ) = < 0,
x2 − x1

so f ′ ̸≥ 0 on I.

Remark 5.47. — If f ′ > 0, the same argument shows that f is strictly increasing. However
the converse fails: the function f (x) = x3 is strictly increasing, but f ′ (0) = 0.

Corollary 5.48: Constant Functions vs. First Derivative

Let I ⊆ R be an interval and f : I → R. Then f is constant if and only if f is
differentiable and f ′ (x) = 0 for all x ∈ I.

Proof. The derivative of a constant is 0.

Conversely, if f ′ = 0, then f ′ ≥ 0 and −f ′ ≥ 0, so by Proposition 5.46 both f and −f are
increasing, hence f is constant.

Exercise 5.49. — Let I ⊆ R and f : I → R. Show that f is a polynomial if and only if f

is smooth and there exists n ∈ N such that f (n) ≡ 0.

Version: December 30, 2025. 173

 Chapter 5.2 Main Theorems of Differential Calculus

Definition 5.50: Convex Functions

Let I ⊆ R and f : I → R. We call f convex if, for all a, b ∈ I with a < b and all
t ∈ (0, 1),
(5.9)


f (1 − t)a + tb ≤ (1 − t)f (a) + tf (b).

We call f strictly convex if the inequality in (5.9) is strict. A function g : I → R is

(strictly) concave if −g is (strictly) convex.

Geometrically, (5.9) says that on every interval [a, b] the graph of f lies below the secant
through (a, f (a)) and (b, f (b)).
An equivalent definition of a convex function is the following:
f : I → R is convex if for all a, b ∈ I with a < b and all x ∈ (a, b),

f (x) − f (a) f (b) − f (x)

≤ , (5.10)
x−a b−x

and strictly convex if the inequality is strict.

Geometrically, this definition corresponds to saying that the “right” secant is steeper than the
24
“left” one.
f

xt 7→ (1 − t)f (a) + tf (b)

a xt = (1 − t)a + tb b

Figure 5.3: In blue, a convex function. In red, the secant through (a, f (a)) and (b, f (b)),
which lies above the function f . In green, the secants between (a, f (a)), (xt , f (xt )) and
(xt , f (xt )), (b, f (b)), the latter being steeper than the former.

Exercise 5.51. — Show that (5.9) for all t ∈ (0, 1) is equivalent to (5.10) for all x ∈ (a, b).

Version: December 30, 2025. 174

 Chapter 5.2 Main Theorems of Differential Calculus

Proposition 5.52: Convexity vs. Monotonicity of the First Derivative

Let I ⊆ R and let f : I → R be differentiable. Then f is convex if and only if f ′ is
increasing.

Proof. Assume f ′ is increasing. Then, for a < b and x ∈ (a, b), the Mean Value Theorem 5.31
applied on the intervals [a, x] and [x, b] yields ξ ∈ (a, x) and ζ ∈ (x, b) such that

f (x) − f (a) f (b) − f (x)

f ′ (ξ) = , f ′ (ζ) = .
x−a b−x

Since f ′ is increasing we have f ′ (ξ) ≤ f ′ (ζ), so (5.10) follows. Since a < b ∈ I and x ∈ (a, b)
are arbitrary, f is convex.
Conversely, assume f is convex. Given a < b, consider h > 0 small enough so that
a + h < b − h and apply (5.10) twice: first, applying it on the interval (a, b − h) with x = a + h
we get
f (a + h) − f (a) f (b − h) − f (a + h)
≤ ;
h (b − h) − (a + h)
then, applying it on the interval (a + h, b) with x = b − h we obtain

f (b − h) − f (a + h) f (b) − f (b − h)
≤ .
25 (b − h) − (a + h) h

Combining these two inequalities we deduce that, for all sufficiently small h > 0,

f (a + h) − f (a) f (b) − f (b − h)
≤ . (5.11)
h h

Letting h → 0+ gives f ′ (a) ≤ f ′ (b). Since a < b are arbitrary, f ′ is increasing.

Exercise 5.53. — Under the assumptions of Proposition 5.52, prove that f is strictly
convex if and only if f ′ is strictly increasing.

Corollary 5.54: Convexity vs. Second Derivative

Let I ⊆ R and let f : I → R be twice differentiable. Then f is convex if and only if
f ′′ ≥ 0.

Proof. By Proposition 5.52, f is convex if and only if f ′ is increasing. Applying Proposi-

tion 5.46 to f ′ , we have that f ′ increasing if and only if f ′′ ≥ 0.

Exercise 5.55. — Under the assumptions of Corollary 5.54, show that if f ′′ (x) > 0 for all
x ∈ I, then f is strictly convex. Is the converse true?

Version: December 30, 2025. 175

 Chapter 5.2 Main Theorems of Differential Calculus

Exercise 5.56. — The function f : (0, ∞) → R, x 7→ x log x, is strictly convex. Indeed, f

is smooth and
1
f ′ (x) = log x + 1, f ′′ (x) = > 0 ∀ x > 0,
x
so Exercise 5.55 applies. Moreover, limx→0+ x log x = 0 (see Example 3.64) and limx→0+ f ′ (x) =
−∞, as reflected in the graph below.

25

Exercise 5.57 (Minima of Convex Functions). — Let I ⊆ R and f : I → R be convex.

Show that every local minimum of f is a global minimum.

Exercise 5.58. — Show that for all real numbers x ≥ −1 and p ≥ 1, the continuous
Bernoulli inequality holds:
(1 + x)p ≥ 1 + px.

Hint: Define f : [−1, ∞) → R as f (x) = (1 + x)p − 1 + px. Prove that f is convex and satisfies
f (0) = f ′ (0) = 0. Use these facts to deduce that f ≥ 0 on [−1, ∞).

Exercise 5.59. — Given α ∈ (0, 1], show that the function x 7→ xα is concave on (0, ∞),
and use this fact to provide an alternative proof of (3.15).

Version: December 30, 2025. 176

 Chapter 5.3 Example: Differentiation of Trigonometric Functions

5.3 Example: Differentiation of Trigonometric Functions

In this section we study the derivative and monotonicity properties of the trigonometric func-
tions.

5.3.1 Sine and Arc Sine

Recalling Exercise 5.13, the functions sin : R → R and cos : R → R are smooth and satisfy

sin′ (x) = cos(x), cos′ (x) = − sin(x).

By Theorem 4.62 and Exercise 4.64, the zeros of cos : R → R are the points { π2 + kπ | k ∈ Z},
and cos(0) = 1. By the Intermediate Value Theorem (Theorem 3.24), it follows that sin′ (x) =
cos(x) > 0 for all x ∈ (− π2 , π2 ). Hence, by Remark 5.47, the function

sin : [− π2 , π2 ] → [−1, 1] (5.12)

is strictly increasing and bijective (recall that sin(− π2 ) = −1 and sin( π2 ) = 1). Consequently,
the restriction of the sine function to [− π2 , π2 ] has an inverse, denoted

arcsin : [−1, 1] → [− π2 , π2 ],

called the arcsine.

The following figure shows the graphs of the sine function on [− π2 , π2 ] and of its inverse.
25
π
2 arcsin(x)
1 sin(x)

− π2 π
2 −1 1

−1
− π2

Remark 5.60. — Since sin′′ = − sin, it follows that sin is convex on − π2 , 0 and concave

on 0, π2 .

By Theorem 5.20, the arcsine is differentiable at s whenever the derivative of the sine at
x = arcsin(s) is nonzero. Since sin′ = cos vanishes only at the endpoints of − π2 , π2 , for
 

x ∈ − π2 , π2 and s = sin(x) we obtain

1 1 1
arcsin′ (s) = =p 2
=√ ,
cos(x) 1 − sin (x) 1 − s2

Version: December 30, 2025. 177

 Chapter 5.3 Example: Differentiation of Trigonometric Functions

where we used that cos(x) > 0 for x ∈ (− π2 , π2 ), thus cos(x) =

p
1 − sin2 (x).

5.3.2 Cosine and Arc Cosine

A similar discussion applies to the cosine function. The cosine is strictly decreasing on the
interval [0, π], and satisfies cos(0) = 1 and cos(π) = −1. Hence,

cos : [0, π] → [−1, 1]

is bijective.

π
1

π
π 2
0 2 π

−1 cos(x)
arccos(x)
−1 0 1

The inverse function is called the arccosine and is denoted by

25 arccos : [−1, 1] → [0, π].

Applying the differentiation rule for the inverse function, for x ∈ (0, π) and s = cos(x) we
obtain
1 1 1
arccos′ (s) = = −p = −√ .
− sin(x) 1 − cos2 (x) 1 − s2

Remark 5.61. — Since cos′′ = − cos, it follows that cos is concave on 0, π2 and convex

on π2 , π .

5.3.3 Tangent and Arc Tangent

Consider the restriction tan : (− π2 , π2 ) → R of the tangent function. Using the quotient rule
(Corollary 5.17) we find
 ′
′ sin(x) cos(x) cos(x) − sin(x)(− sin(x)) 1
tan (x) = = 2
=
cos(x) cos (x) cos2 (x)

for all x ∈ (− π2 , π2 ). Hence tan is strictly increasing on (− π2 , π2 ). Moreover,

sin(x) sin(x)
lim tan(x) = lim = +∞, lim tan(x) = lim = −∞.
π − π − cos(x) π + π + cos(x)
x→( 2 ) x→( 2 ) x→(− 2 ) x→(− 2 )

Version: December 30, 2025. 178

 Chapter 5.3 Example: Differentiation of Trigonometric Functions

Hence, by the Intermediate Value Theorem, the tangent function is bijective:

tan : (− π2 , π2 ) → R.

tan(x)

π
2
arctan(x)

− π2 0 π
2 0

− π2

25
Its inverse
arctan : R → − π2 , π2

is called the arctangent. By Theorem 5.20, for x ∈ (− π2 , π2 ) and s = tan(x) we obtain

1
arctan′ (s) = 1 = cos2 (x).
cos2 (x)

Since
sin2 (x) 1 − cos2 (x) 1
s2 = tan2 (x) = 2
= 2
= − 1,
cos (x) cos (x) cos2 (x)
it follows that 1 + s2 = 1
cos2 (x)
, therefore

1
arctan′ (s) = ∀ s ∈ R.
1 + s2

The cotangent behaves similarly. The restriction cot |(0,π) : (0, π) → R is strictly decreasing
and bijective. Its inverse
arccot : R → (0, π)

is called the arccotangent and satisfies

1
arccot′ (s) = − ∀ s ∈ R.
1 + s2

Version: December 30, 2025. 179

 Chapter 5.3 Example: Differentiation of Trigonometric Functions

5.3.4 Hyperbolic Functions

We now perform the analogous analysis for the hyperbolic trigonometric functions:

ex − e−x ex + e−x sinh(x) ex − e−x

sinh(x) = , cosh(x) = , tanh(x) = = x .
2 2 cosh(x) e + e−x

We have sinh′ (x) = cosh(x) > 0 for all x ∈ R, so by Proposition 5.46 the hyperbolic sine
is strictly increasing. Since limx→∞ sinh(x) = +∞ and limx→−∞ sinh(x) = −∞, the function

sinh : R → R

is bijective. Its inverse

arsinh : R → R

is called the inverse hyperbolic sine. By Theorem 5.20, for x ∈ R and s = sinh(x) we
obtain
1 1 1
arsinh′ (s) = =q =√ .
cosh(x) 1 + s2
1 + sinh2 (x)

Moreover, the inverse hyperbolic sine has a closed formula. Indeed, starting from sinh(x) = s,
we get
ex − e−x
25 =s =⇒ e2x − 2sex − 1 = 0.
2
Setting y = ex , this becomes y 2 − 2sy − 1 = 0, yielding
p
y =s± 1 + s2 .
√
Since y = ex > 0, the admissible root is y = s + 1 + s2 , hence
p p
ex = y = s +


1 + s2 =⇒ x = log s + 1 + s2 .

In other words, p

arsinh(s) = log s + 1 + s2 , s ∈ R.

The hyperbolic cosine satisfies cosh′ (x) = sinh(x) and cosh′′ (x) = cosh(x) > 0 for all
x ∈ R. Thus, cosh is strictly convex (Corollary 5.54) and has a global minimum at x = 0,
since cosh′ (0) = sinh(0) = 0. For x > 0, cosh′ (x) > 0, so cosh is strictly increasing on [0, ∞).
As cosh(0) = 1 and limx→∞ cosh(x) = +∞, we deduce

cosh : [0, ∞) → [1, ∞)

is bijective. Its inverse

arcosh : [1, ∞) → [0, ∞)

Version: December 30, 2025. 180

 Chapter 5.3 Example: Differentiation of Trigonometric Functions

is called the inverse hyperbolic cosine. It is differentiable on [1, ∞) and satisfies, for
s = cosh(x) with x > 0 (thus s > 1),

1 1 1
arcosh′ (s) = =p =√ .
sinh(x) cosh(x)2 − 1 s2 − 1

Furthermore, arguing as done before for the inverse hyperbolic sine,

p

arcosh(s) = log s + s2 − 1 ∀ s ≥ 1.

25 The inverse hyperbolic tangent,

 
1 1+s
artanh : (−1, 1) → R, artanh(s) = 2 log
1−s

is the inverse function of the strictly increasing bijection tanh : R → (−1, 1), and it satisfies

1
artanh′ (s) = ∀ s ∈ (−1, 1).
1 − s2

Exercise 5.62. — Verify all assertions made in Paragraphs 5.3.4, 5.3.4, and 5.3.4.

Version: December 30, 2025. 181

 Chapter 6

The Riemann Integral

In this chapter, we take the idea from Section 1.1 and extend it to the notion of the Riemann
integral, using the concepts of supremum and infimum.

6.1 Step Functions and their Integral

6.1.1 Decompositions and Step Functions

Partitions
Two sets A, B are called disjoint if A ∩ B = ∅. For a collection A of sets, we say that
the sets in A are pairwise disjoint if for all A1 , A2 ∈ A with A1 ̸= A2 it holds that
A1 ∩ A2 = ∅.
Let X be a set. A partition of X is a family P of non-empty pairwise disjoint subsets
of X such that
[
X= P.
P ∈P

In other words, each P ∈ P is non-empty, and every element of X belongs to exactly one
26
P ∈ P.

P2 P3
P1
P4

Figure 6.1: Schematic representation of a partition P = {P1 , . . . , P4 } of a set X.

For the following discussion, we fix two real numbers a < b and work with the compact
interval [a, b] ⊂ R.

182
 Chapter 6.1 Step Functions and their Integral

Definition 6.1: Decomposition of an Interval

A decomposition of [a, b] is a finite sequence of points

a = x0 < x1 < . . . < xn−1 < xn = b,

with n ∈ N. The points x0 , . . . , xn are called the division points of the decomposition.

Formally, a decomposition of [a, b] is a finite subset of [a, b] containing a and b, together

with the ordering of its elements. Each decomposition induces a natural partition of [a, b]:

[a, b] = {x0 } ∪ (x0 , x1 ) ∪ {x1 } ∪ . . . ∪ (xn−1 , xn ) ∪ {xn },

which we will use implicitly from now on.

A decomposition
a = y0 < y1 < . . . < ym = b

is called a refinement of the decomposition

a = x0 < x1 < . . . < xn = b

if
{x0 , x1 , . . . , xn } ⊆ {y0 , y1 , . . . , ym }.
26
The notion of refinement defines a partial order on the set of all decompositions of [a, b]. Note
that any two decompositions of [a, b] admit a common refinement given by the union of all
division points.

Definition 6.2: Step Functions

A function f : [a, b] → R is called a step function if there exists a decomposition

a = x0 < x1 < . . . < xn = b

such that, for each k = 1, . . . , n, the restriction of f to the open interval (xk−1 , xk ) is
constant. In this case, we say that f is a step function with respect to the decomposition
a = x0 < x1 < . . . < xn = b.

x0 = a x1 x2 x3 x4 x5 x6 = b

Figure 6.2: The graph of a step function on the interval [a, b]. Note that the values of f at
the division points of the decomposition are irrelevant for the definition of a step function.

Version: December 30, 2025. 183

 Chapter 6.1 Step Functions and their Integral

Proposition 6.3: Linearity of the Space of Step Functions

Let f, g : [a, b] → R be step functions, and α, β ∈ R. Then αf + βg is also a step
function.

Proof. Let f be a step function with respect to the decomposition a = x0 < x1 < . . . < xn = b,
and let g be a step function with respect to the decomposition a = y0 < y1 < . . . < ym = b.
The union of all division points {x0 , . . . , xn } ∪ {y0 , . . . , ym } defines a new decomposition

a = z0 < z1 < . . . < zN = b

that is a common refinement of the two. Since both f and g are constant on each open interval
(zk−1 , zk ), so is the function αf + βg. Thus αf + βg is a step function with respect to this
decomposition.

Example 6.4. — Constant functions are step functions.

Remark 6.5. — As in the proof of Proposition 6.3, one can show that the product of two
step functions is again a step function. Moreover, step functions are bounded, since they take
only finitely many values.
26

6.1.2 The Integral of a Step Function

Definition 6.6: Integral of a Step Function

Let f : [a, b] → R be a step function with respect to a decomposition

a = x0 < x1 < . . . < xn = b.

We define the integral of f on [a, b] as the real number

Z b n
X
f (x) dx = ck (xk − xk−1 ), (6.1)
a k=1

where ck denotes the constant value of f on the interval (xk−1 , xk ).

Remark 6.7. — For non-negative step functions f ≥ 0, the value of the integral (6.1) can
be interpreted geometrically as the total area of the rectangles of height ck and base length
xk − xk−1 . In general, the integral represents the signed net area enclosed between the graph
of f and the x-axis.

Version: December 30, 2025. 184

 Chapter 6.1 Step Functions and their Integral

At this stage, in (6.1), the symbols and dx should be regarded as purely formal. Histor-
R

ically, the symbol is an elongated S for “sum,” and dx indicates an “infinitesimal length,”
R

i.e. xk − xk−1 in the limit of an infinitely fine decomposition. This notation was introduced
by Leibniz (1646–1716).

+ +
+
x0 = a x1 x2 x3 x4 = b
–

Figure 6.3: For a non-negative step function f ≥ 0, the integral in (6.1) represents the area
of the set {(x, y) ∈ R2 | a ≤ x ≤ b, 0 ≤ y ≤ f (x)}, and in general the signed area.

Remark 6.8. — Definition 6.6 raises a potential issue: a priori, the right-hand side of (6.1)
might depend on the chosen decomposition. To show that this is not the case, let

26 a = y0 < · · · < ym = b

be another decomposition of [a, b] with respect to which f is a step function. We must prove
that
Xn Xm
ck (xk − xk−1 ) = dk (yk − yk−1 ), (6.2)
k=1 k=1

where dk denotes the constant value of f on (yk−1 , yk ). We argue in three steps:

1. If the decomposition a = y0 < · · · < ym = b is a refinement of a = x0 < · · · < xn = b

differing by a single additional division point yℓ ∈ (xℓ−1 , xℓ ), then the two sums in (6.2)
coincide, since cℓ = dℓ = dℓ+1 and

cℓ (xℓ − xℓ−1 ) = dℓ (yℓ − yℓ−1 ) + dℓ+1 (yℓ+1 − yℓ ).

2. By induction on the number of additional division points, (6.2) holds for any refinement.

3. As shown in the proof of Proposition 6.3, any two decompositions admit a common re-
finement. Comparing both sums with the sum corresponding to this common refinement
proves (6.2) in full generality.

Therefore, the value of the integral of a step function is independent of the chosen decom-
position.

Version: December 30, 2025. 185

 Chapter 6.1 Step Functions and their Integral

Proposition 6.9: Linearity of the Integral of Step Functions

Let f, g : [a, b] → R be step functions, and let α, β ∈ R. Then
Z b Z b Z b
(αf + βg)(x) dx = α f (x) dx + β g(x) dx.
a a a

Proof. As in the proof of Proposition 6.3, we can find a decomposition a = x0 < . . . < xn = b
such that both f and g (and hence αf + βg) are constant on each interval (xk−1 , xk ). If f
takes the value ck and g the value dk on (xk−1 , xk ), then αf + βg takes the value αck + βdk .
Thus
Z b n
X
(αf + βg)(x) dx = (αck + βdk )(xk − xk−1 )
a k=1
Xn n
X
=α ck (xk − xk−1 ) + β dk (xk − xk−1 )
k=1 k=1
Z b Z b
=α f (x) dx + β g(x) dx,
a a

as claimed.

Proposition 6.10: Monotonicity of the Integral of Step Functions

26 Let f, g : [a, b] → R be step functions such that f ≤ g. Then
Z b Z b
f (x) dx ≤ g(x) dx.
a a

Proof. As in the proofs of Proposition 6.3, we can find a decomposition a = x0 < . . . < xn = b
such that both f and g are constant on each interval (xk−1 , xk ). Writing ck and dk for their
respective values, the assumption f ≤ g implies ck ≤ dk for all k = 1, . . . , n. Hence
Z b n
X n
X Z b
f (x) dx = ck (xk − xk−1 ) ≤ dk (xk − xk−1 ) = g(x) dx.
a k=1 k=1 a

Applying Proposition 6.10 with g ≡ 0, we deduce the following:

Corollary 6.11: Positivity of the Integral of Step Functions

If f : [a, b] → R is a step function such that f (x) ≥ 0 for all x ∈ [a, b], then
Z b
f (x) dx ≥ 0.
a

Version: December 30, 2025. 186

 Chapter 6.1 Step Functions and their Integral

Exercise 6.12. — Let [a, b] and [b, c] be two compact intervals, and let f1 : [a, b] → R and
f2 : [b, c] → R be step functions. Define

f1 (x), x ∈ [a, b),

f : [a, c] → R, f (x) =
f2 (x), x ∈ [b, c].


26
1. Show that f is a step function on [a, c].

2. Prove that Z c Z b Z c
f (x) dx = f1 (x) dx + f2 (x) dx.
a a b

3. Show that every step function on [a, c] is of the form described above.

Version: December 30, 2025. 187

 Chapter 6.2 Definition and First Properties of the Riemann Integral

6.2 Definition and First Properties of the Riemann Integral

As in the last section, we consider functions on a compact interval [a, b] ⊂ R. To alleviate
Rb
notation, we write SF for the set of step functions on [a, b]. Also, we often write a f dx in
Rb
place of a f (x) dx.

6.2.1 Integrability of Real-valued Functions

Before defining lower and upper sums, we recall a simple but useful property of the supremum
and infimum of two related sets. We will use this fact several times in what follows.

Relation between Supremum and Infimum

Let A, B ⊂ R be nonempty sets such that s ≤ t for all s ∈ A and t ∈ B. Then

sup A ≤ inf B. (6.3)

Moreover,

sup A = inf B ⇐⇒ for every ε > 0 there exist s ∈ A and t ∈ B s.t. t − s < ε. (6.4)

The following definition of integrability is a variant of Riemann’s definition, which goes

back to the French mathematician Jean-Gaston Darboux (1842–1917).
26
Definition 6.13: Lower and Upper Sums
Let f : [a, b] → R be a function. Define the sets of lower sums L(f ) ⊂ R and upper
sums U(f ) ⊂ R by
Z b  Z b 
L(f ) = ℓ dx ℓ ∈ SF and ℓ ≤ f , U (f ) = u dx u ∈ SF and f ≤ u .
a a

If f is bounded, then these sets are non-empty. Indeed, if |f | ≤ M , then the constant step
functions
ℓ(x) = −M ∀ x ∈ [a, b], u(x) = M ∀ x ∈ [a, b],

satisfy ℓ ∈ L(f ) and u ∈ U(f ).

For ℓ, u ∈ SF with ℓ ≤ f ≤ u, Proposition 6.10 gives
Z b Z b
ℓ dx ≤ u dx.
a a

This implies that s ≤ t for all s ∈ L(f ) and t ∈ U(f ), so (6.3) yields

sup L(f ) ≤ inf U (f ).

Version: December 30, 2025. 188

 Chapter 6.2 Definition and First Properties of the Riemann Integral

Definition 6.14: Riemann Integral

A bounded function f : [a, b] → R is Riemann integrable if sup L(f ) = inf U(f ). In
this case, this common value is called the Riemann integral of f , and we write
Z b
f dx = sup L(f ) = inf U(f ).
a

We call a the lower (integration) limit and b the upper (integration) limit, and the
Rb
function f the integrand of the integral a f dx. If f ≥ 0 is Riemann integrable, we interpret
Rb
the number a f dx as the area of the set

{(x, y) ∈ R2 | a ≤ x ≤ b, 0 ≤ y ≤ f (x)}.

Remark 6.15. — For now we only discuss Riemann integrability and the Riemann inte-
gral, so we will simply say “integrable” and “integral”. Note, however, that there is another
fundamental theory, the Lebesgue integral, which we will not cover in this course.

Proposition 6.16: Riemann Integrability Condition

Let f : [a, b] → R be bounded. Then f is Riemann integrable if and only if for every
ε > 0 there exist step functions ℓ, u ∈ SF such that
26
Z b
ℓ≤f ≤u and (u − ℓ) dx < ε.
a

In this case,
Z b Z b Z b Z b
f dx − ℓ dx < ε, u dx − f dx < ε.
a a a a

Proof. By (6.4) applied with A = L(f ) and B = U(f ) we obtain

f is Riemann integrable ⇐⇒ sup L(f ) = inf U(f )

⇐⇒ ∀ ε > 0 ∃ s ∈ L(f ), t ∈ U (f ) : t − s < ε
Z b Z b
⇐⇒ ∀ ε > 0 ∃ ℓ, u ∈ SF : ℓ ≤ f ≤ u and u dx − ℓ dx < ε
a a
Z b
⇐⇒ ∀ ε > 0 ∃ ℓ, u ∈ SF : ℓ ≤ f ≤ u and (u − ℓ) dx < ε,
a

where we used Proposition 6.9 to deduce that

Z b Z b Z b
u dx − ℓ dx = (u − ℓ) dx.
a a a

Version: December 30, 2025. 189

 Chapter 6.2 Definition and First Properties of the Riemann Integral

Finally, the concluding inequalities follow from

Z b Z b Z b Z b
ℓ dx ≤ f dx ≤ u dx and (u − ℓ) dx < ε.
a a a a

It is useful to note that the Riemann integral extends the integral of step functions; in
particular, we can speak of the Riemann integral of a step function, see Exercise 6.17.

Exercise 6.17. — Let f : [a, b] → R be a step function. Show that f is Riemann integrable
and that its Riemann integral equals its step-function integral.

Exercise 6.18. — Repeat the proof of Proposition 1.1 and show, in the language of this
R1
section, that f : [0, 1] → R, x 7→ x2 , is Riemann integrable with 0 x2 dx = 31 . Also,

L(f ) = − ∞, 31 and 1



26 U(f ) = 3, ∞ .

Example 6.19. — Not all functions are Riemann integrable. Indeed, consider the function
f : [0, 1] → R defined by 
1, x ∈ Q,
f (x) =
0, x ∈
/ Q.

We claim that f is not Riemann integrable.

Let u ∈ SF with f ≤ u, and let 0 = x0 < · · · < xn = 1 be a decomposition such that
u is constant ck on (xk−1 , xk ). Since Q is dense in R, there exists x ∈ (xk−1 , xk ) ∩ Q, hence
1 = f (x) ≤ u(x) = ck , so ck ≥ 1. Therefore
Z 1 n
X n
X
u(x) dx = ck (xk − xk−1 ) ≥ (xk − xk−1 ) = xn − x0 = 1.
0 k=1 k=1

Thus inf U(f ) ≥ 1, and taking u ≡ 1 gives inf U (f ) = 1. A similar argument with lower sums
shows that sup L(f ) = 0. Hence f is not Riemann integrable.

6.2.2 Linearity and Monotonicity of the Riemann Integral

Theorem 6.20: Linearity of the Riemann Integral

27 If f, g : [a, b] → R are integrable and α, β ∈ R, then αf + βg is integrable and
Z b Z b Z b
(αf + βg) dx = α f dx + β g dx.
a a a

Version: December 30, 2025. 190

 Chapter 6.2 Definition and First Properties of the Riemann Integral

Proof. Given ε > 0, Proposition 6.16 yields step functions ℓ1 , ℓ2 , u1 , u2 with

Z b Z b
ℓ1 ≤ f ≤ u1 , ℓ2 ≤ g ≤ u2 , (u1 − ℓ1 ) dx < ε, (u2 − ℓ2 ) dx < ε,
a a

and Z b Z b Z b Z b
f dx − ℓ1 dx < ε, g dx − ℓ2 dx < ε.
a a a a

Assume first α, β ≥ 0. Then

αℓ1 + βℓ2 ≤ αf + βg ≤ αu1 + βu2 ,

and
Z b 
Z b Z b
(αu1 + βu2 ) − (αℓ1 + βℓ2 ) dx = α (u1 − ℓ1 ) dx + β (u2 − ℓ2 ) dx < (α + β)ε.
a a a

Since ε > 0 is arbitrary, this proves that αf + βg is integrable. Moreover, by the triangle
inequality and Proposition 6.9,
Z b Z b Z b Z b Z b
(αf + βg) dx − α f dx − β g dx ≤ (αf + βg) dx − (αℓ1 + βℓ2 ) dx
a a a a a
Z b Z b Z b
+ (αℓ1 + βℓ2 ) dx − α ℓ1 dx − β ℓ2 dx
27 |a {z a a
}
=0
Z b Z b Z b Z b
+α ℓ1 dx − f dx + β ℓ2 dx − g dx
a a a a
≤ (α + β)ε + αε + βε = 2(α + β)ε.

Since ε > 0 is arbitrary, the linearity identity follows.

The case when one of α, β is negative is analogous, but one needs to reverse the corre-
sponding inequalities. For instance, if α ≥ 0 and β < 0, then

αℓ1 + βu2 ≤ αf + βg ≤ αu1 + βℓ2 ,

and
Z b 
Z b Z b
(αu1 + βℓ2 ) − (αℓ1 + βu2 ) dx = α (u1 − ℓ1 ) dx + |β| (u2 − ℓ2 ) dx < (α + |β|)ε.
a a a

This implies again that αf + βg is integrable, and the linearity identity holds similarly.

Exercise 6.21. — Let f : [a, b] → R be Riemann integrable, and let f ∗ : [a, b] → R

be obtained by changing the value of f at finitely many points. Show that f ∗ is Riemann
integrable and has the same Riemann integral as f .

Version: December 30, 2025. 191

 Chapter 6.2 Definition and First Properties of the Riemann Integral

Proposition 6.22: Monotonicity of the Riemann Integral

Z b Z b
Let f, g : [a, b] → R be integrable. If f ≤ g, then f dx ≤ g dx.
a a

Proof. Since f ≤ g, for any step function ℓ with ℓ ≤ f we have ℓ ≤ g. This implies that
L(f ) ⊆ L(g), therefore
Z b Z b
f dx = sup L(f ) ≤ sup L(g) = g dx.
a a

Positive and Negative Parts

Given a function f : D → R, we define its positive part f + : D → R and negative
part f − : D → R by

f + (x) = max{0, f (x)}, f − (x) = − min{0, f (x)}.

These satisfy

|f | + f |f | − f
f = f + − f −, |f | = f + + f − , f+ = , f− = .
27 2 2

Moreover, for any functions f, g : D → R,

f ≤ g =⇒ f + ≤ g + , f ≤ g =⇒ f − ≥ g − .

|f |
f

a b a b

f+
f−

a b a b

Figure 6.4: Top left: graph of f . Top right: graph of |f |. Bottom left: graph of f + .
Rb
Bottom right: graph of f − . The integral a f dx is the signed area (positive minus negative),
Rb Rb Rb Rb Rb
whereas a |f | dx is the total area. Moreover, a f dx = a f + dx − a f − dx and a |f | dx =
Rb + Rb −
a f dx + a f dx.

Version: December 30, 2025. 192

 Chapter 6.2 Definition and First Properties of the Riemann Integral

Remark 6.23. — For any real numbers z1 , z2 ∈ R, one has

(z1 − z2 )+ ≥ z1+ − z2+ . (6.5)

Indeed, since z + ≥ z and z + ≥ 0 for all z ∈ R, by applying these inequalities with z = z1 − z2

and z = z2 we obtain

z1 = (z1 − z2 ) + z2 ≤ (z1 − z2 )+ + z2+ and 0 ≤ (z1 − z2 )+ + z2+ .

Hence (z1 − z2 )+ + z2+ is greater than or equal to both z1 and 0, and therefore

z1+ = max{z1 , 0} ≤ (z1 − z2 )+ + z2+ ,

which yields (6.5) after rearranging.

Theorem 6.24: Triangle Inequality for the Riemann Integral

Let f : [a, b] → R be integrable. Then f + , f − , and |f | are integrable, and

Z b Z b
f dx ≤ |f | dx.
a a

Rb
Proof. Fix ε > 0. Since f is integrable, there exist step functions ℓ ≤ f ≤ u with a (u−ℓ) dx <
27 ε. Then ℓ+ and u+ are step functions with ℓ+ ≤ f + ≤ u+ .
Since u − ℓ ≥ 0, we have (u − ℓ) = (u − ℓ)+ . Moreover, applying (6.5) with z1 = u(x) and
z2 = ℓ(x), we obtain

(u(x) − ℓ(x))+ ≥ u(x)+ − ℓ(x)+ ∀ x ∈ [a, b].

Hence Z b Z b Z b
+ + +
(u − ℓ ) dx ≤ (u − ℓ) dx = (u − ℓ) dx < ε,
a a a

so f + is integrable. By Theorem 6.20, also f − = f + − f and |f | = 2f + − f are integrable.

Finally,
Z b Z b Z b Z b Z b Z b
+ − + −
f dx = f dx − f dx ≤ f dx + f dx = |f | dx.
a a a a a a

Exercise 6.25. — Let a < b < c. Show that f : [a, c] → R is integrable if and only if f |[a,b]
and f |[b,c] are integrable, and in that case
Z c Z b Z c
f dx = f |[a,b] dx + f |[b,c] dx.
a a b

Version: December 30, 2025. 193

 Chapter 6.2 Definition and First Properties of the Riemann Integral

Exercise 6.26. — Let f : [a, b] → R be integrable and λ > 0. Define g : [λa, λb] → R by
g(x) = f (λ−1 x). Show that g is integrable and
Z b Z λb
λ f dx = g dx.
a λa

Exercise 6.27. — Let f : [a, b] → R be integrable. Show that the function F : [a, b] → R
given by Z x
27 F (x) = f (t) dt
a

is continuous.

Exercise 6.28. — Let f : [0, 1] → R be integrable and ε > 0. Show that there exists a
continuous function g : [0, 1] → R such that
Z 1
|f (x) − g(x)| dx < ε.
0

Version: December 30, 2025. 194

 Chapter 6.3 Integrability Theorems

6.3 Integrability Theorems

6.3.1 Integrability of Monotone Functions

As before we work on a compact interval [a, b] ⊂ R. Note that every monotone function
f : [a, b] → R is bounded; for instance, if f is increasing then f (a) is a lower bound and f (b)
is an upper bound.

Theorem 6.29: Monotone Functions are Integrable

Every monotone function f : [a, b] → R is Riemann integrable.

Proof. Without loss of generality, f is increasing (otherwise replace f by −f and use Propo-
sition 6.20). We want to apply Proposition 6.16: given ε > 0, we need to construct step
Rb
functions ℓ, u ∈ SF such that ℓ ≤ f ≤ u and a (u − ℓ) dx < ε.
Fix n ∈ N (to be chosen later) and the uniform partition

k
a = x0 < x1 < . . . < xn = b, xk = a + (b − a).
n

Define the step functions ℓ, u : [a, b] → R as

ℓ(x) = f (xk−1 ) and u(x) = f (xk ) for x ∈ (xk−1 , xk ), k = 1, . . . , n,

27 ℓ(x) = u(x) = f (x) for x ∈ {x0 , . . . , xn }.

Note that, since f is increasing, ℓ ≤ f ≤ u. Moreover, for each k we have u − ℓ = f (xk ) −

f (xk−1 ) on (xk−1 , xk ). Recalling that xk − xk−1 = b−a
n , this yields

b n n
b−aX
Z X


(u − ℓ) dx = f (xk ) − f (xk−1 ) (xk − xk−1 ) = f (xk ) − f (xk−1 )
a n
k=1 k=1
b−a
b−a

= f (xn ) − f (x0 ) = f (b) − f (a) .
n n

Choosing n ∈ N so large that b−a

n f (b)−f (a) < ε, Proposition 6.16 implies that f is Riemann

integrable.

Figure 6.5: A monotone function, with upper and lower sums arbitrarily close.

Version: December 30, 2025. 195

 Chapter 6.3 Integrability Theorems

Using the additivity property in Exercise 6.25, Theorem 6.29 extends to functions that are
only piecewise monotone.

Definition 6.30: Piecewise Monotone Functions

A function f : [a, b] → R is piecewise monotone if there exists a decomposition

a = x0 < x1 < . . . < xn = b

such that f |(xk−1 ,xk ) is monotone for every k = 1, . . . , n.

Corollary 6.31: Piecewise Monotone Functions are Integrable

Every bounded piecewise monotone function f : [a, b] → R is Riemann integrable.

Proof. Combine Theorem 6.29 with Exercises 6.21 and 6.25.

6.3.2 Integrability of Continuous Functions

Using boundedness and uniform continuity on compact intervals (Theorems 3.31 and 3.39),
27 we can prove that continuous functions are integrable.

Theorem 6.32: Continuous Functions are Integrable

Every continuous function f : [a, b] → R is Riemann integrable.

Proof. Let f : [a, b] → R be continuous and fix ε > 0. By uniform continuity (Theorem 3.39),
there exists δ > 0 such that

|x − y| < δ =⇒ |f (x) − f (y)| < ε ∀ x, y ∈ [a, b]. (6.6)

Choose a partition a = x0 < . . . < xn = b with xk − xk−1 < δ. For each k set

ck = min{f (x) | xk−1 ≤ x ≤ xk }, dk = max{f (x) | xk−1 ≤ x ≤ xk },

which exist by Theorem 3.34, and let yk , zk ∈ [xk−1 , xk ] satisfy f (yk ) = ck and f (zk ) = dk .
Then, since |yk − zk | ≤ xk − xk−1 < δ, (6.6) yields dk − ck < ε.
Define now the step functions ℓ, u : [a, b] → R as

ℓ(x) = ck and u(x) = dk for x ∈ (xk−1 , xk ), k = 1, . . . , n,

ℓ(x) = u(x) = f (x) for x ∈ {x0 , . . . , xn }.

Version: December 30, 2025. 196

 Chapter 6.3 Integrability Theorems

Then ℓ ≤ f ≤ u and u − ℓ = dk − ck on (xk−1 , xk ), hence

Z b n
X n
X
(u − ℓ) dx = (dk − ck )(xk − xk−1 ) < ε (xk − xk−1 ) = ε(b − a).
a k=1 k=1

Since ε > 0 is arbitrary, f is integrable.

By Exercise 6.25, the previous theorem extends to piecewise continuous functions.

Definition 6.33: Piecewise Continuous Functions

A function f : [a, b] → R is piecewise continuous if there exists a decomposition

a = x0 < x1 < . . . < xn = b

such that f |(xk−1 ,xk ) is continuous for all k and both one-sided limits limx→x+ f (x)
k−1
and limx→x− f (x) exist. Equivalently, each f |(xk−1 ,xk ) extends to a continuous function
k
on [xk−1 , xk ].

Corollary 6.34: Piecewise Continuous Functions are Integrable

Every piecewise continuous function f : [a, b] → R is Riemann integrable.

Proof. Apply Theorem 6.32 to the continuous extensions on each subinterval, and use Exer-
27
cises 6.21 and 6.25.

Most “common” functions are piecewise continuous or piecewise monotone, hence integrable
by Theorems 6.29 and 6.32. Note also that there are continuous functions that are not
monotone on any open subinterval.

Figure 6.6: Approximate representation of a continuous function which is not monotone on

any open subinterval.

Exercise 6.35. — Let f : [a, b] → R be continuous. Prove that

Z b
f ≡0 ⇐⇒ |f (x)| dx = 0.
a

Version: December 30, 2025. 197

 Chapter 6.3 Integrability Theorems

Exercise 6.36. — Let C be the space of continuous functions on [a, b] and define I : C → R
by
Z b
I(f ) = f dx.
a

Show that I is continuous with respect to the uniform norm; that is, for every ε > 0 there
27
exists δ > 0 such that

sup |f (x) − g(x)| < δ =⇒ |I(f ) − I(g)| ≤ ε.

x∈[a,b]

Hint: Take δ = b−a .

ε

6.3.3 Integration and Sequences of Functions

Let (fn )∞
n=0 , with fn : [a, b] → R, be a sequence of integrable functions. Assume that fn
converges pointwise or uniformly to f : [a, b] → R. Is f integrable? And if so, does
Z b Z b
lim fn dx = f dx
n→∞ a a

hold?
In general, the pointwise limit of integrable functions need not be integrable. Also, as the
following example shows, even when the pointwise limit f is integrable, one may have that
limn fn ̸= f .
R R

Example 6.37. — Let D = [0, 1] and define fn : D → R by

28 


n2 x if 0 ≤ x ≤ 2n
1
,


fn (x) = n2 n1 − x if 2n1
≤ x ≤ n1 ,





if n1 ≤ x ≤ 1.

0


Each fn is continuous (hence integrable). Also, its graph is a triangle of base 1

n and height n2 ,
so Z 1
1 1 n 1
fn (x) dx = · · = .
0 2 n 2 4
Moreover, fn (0) = 0 for all n, and for every x > 0 we have fn (x) = 0 for all n > 1/x, hence
fn (x) → 0. Thus fn converges pointwise to the constant function f = 0, but
Z 1 Z 1
1
fn dx = ̸= 0 = f dx.
0 4 0

Version: December 30, 2025. 198

 Chapter 6.3 Integrability Theorems

n
2

fn

1 1
n

On the other hand, as the next result shows, uniform convergence is sufficient for both
integrability of the limit and interchange of limit and integral.

Theorem 6.38: Uniform Convergence and Riemann Integrals Commute

Let (fn )∞
n=0 , with fn : [a, b] → R, be a sequence of integrable functions converging
uniformly to f : [a, b] → R. Then f is integrable and
Z b Z b
f dx = lim fn dx. (6.7)
a n→∞ a
28
Proof. Fix ε > 0. By uniform convergence, there exists N such that |fn − f | ≤ ε on [a, b] for
all n ≥ N .
Rb
Since fN is integrable, there exist step functions ℓ, u with ℓ ≤ fN ≤ u and a (u − ℓ) dx < ε.
Set ℓ̂ = ℓ − ε and û = u + ε. Then ℓ̂, û ∈ SF. Also, since |fN − f | < ε,

ℓ̂ = ℓ − ε ≤ fN − ε ≤ f ≤ fN + ε ≤ u + ε = û

and (because û − ℓ̂ = u − ℓ + 2ε)

Z b Z b
(û − ℓ̂) dx = (u − ℓ) dx + 2ε(b − a) < ε + 2ε(b − a).
a a

As ε > 0 is arbitrary, Proposition 6.16 yields that f is integrable.

Moreover, using monotonicity (Proposition 6.22) and the triangle inequality for the Rie-
mann integral (Theorem 6.24),
Z b Z b Z b Z b
f dx − fn dx = (f − fn ) dx ≤ |f − fn | dx ≤ ε (b − a) ∀ n ≥ N,
a a a a

proving (6.7).

Version: December 30, 2025. 199

 Chapter 7

The Derivative and the Riemann

Integral

In this chapter we study the interplay between the Riemann integral (Chapter 6) and differ-
entiation (Chapter 5). These connections are fundamental for the developments that follow.

7.1 The Fundamental Theorem of Calculus

Throughout this section we fix a compact interval I ⊆ R that is nonempty and contains more
than one point. For brevity, we write integrable for Riemann integrable.

7.1.1 The Fundamental Theorem

Definition 7.1: Primitive Function

Let I ⊂ R be an interval and f : I → R a function. Any differentiable function
F : I → R such that F ′ = f is called a primitive (or antiderivative) of f .

28 Remark 7.2. — As the following exercise shows, a primitive may not always exist.

Exercise 7.3. — Show that there is no differentiable function F : R → R with F ′ (x) =

sgn(x) for all x ∈ R.
Hint: use Darboux’s Theorem (see Exercise 5.36).

The next result is known as the Fundamental Theorem of (Integral and Differential)
Calculus, going back to Leibniz, Newton, and Barrow

200
 Chapter 7.1 The Fundamental Theorem of Calculus

Theorem 7.4: Fundamental Theorem of Calculus

Let f : [a, b] → R be continuous. Then:
(i) For every C ∈ R, the function F : [a, b] → R defined by
Z x
F (x) = f (t) dt + C (7.1)
a

is a primitive of f .
(ii) Every primitive F : [a, b] → R of f has the form (7.1) for some constant C.

Proof. By Theorem 6.32, f is integrable.

Let F be defined as in (7.1). To prove (i), we fix x0 ∈ [a, b] and we want to show that
F (x0 ) = f (x0 ). To this aim, fix ε > 0. By continuity, there exists δ > 0 such that
′

z ∈ [a, b], |z − x0 | < δ =⇒ |f (z) − f (x0 )| < ε (7.2)

Now, given x ∈ (x0 , x0 + δ) ∩ [a, b], it follows from Exercise 6.25 that
Z x Z x0 
F (x) − F (x0 ) 1
− f (x0 ) = f (t) dt − f (t) dt − f (x0 )
x − x0 x − x0 a a
Z x
1
= f (t) dt − f (x0 ) .
x − x 0 x0
28
Also, Z x Z x
1 1
f (x0 ) = f (x0 ) dt = f (x0 ) dt.
x − x0 x0 x − x0 x0

Combining these two equations and using Theorem 6.24, we get

Z x Z x
F (x) − F (x0 ) 1 1
− f (x0 ) = f (t) dt − f (x0 ) dt
x − x0 x − x 0 x0 x − x 0 x0
Z x
1

= f (t) − f (x0 ) dt
x − x 0 x0
Z x
1
≤ |f (t) − f (x0 )| dt.
x − x 0 x0

Note now that, in the last integral, t ∈ [x0 , x] ⊂ [x0 , x0 + δ) ∩ [a, b]. Hence, it follows from
(7.2) that |f (t) − f (x0 )| < ε, therefore
x
F (x) − F (x0 )
Z
1
− f (x0 ) < ε dt = ε.
x − x0 x − x0 x0

Similarly, if x ∈ (x0 − δ, x0 ) ∩ [a, b], then

x0 x0
F (x) − F (x0 )
Z Z
1
1
− f (x0 ) = f (t) − f (x0 ) dt ≤ |f (t) − f (x0 )| dt < ε.
x − x0 x0 − x x x0 − x x

Version: December 30, 2025. 201

 Chapter 7.1 The Fundamental Theorem of Calculus

In summary, we proved that

F (x) − F (x0 )
x ∈ [a, b], |x − x0 | < δ =⇒ − f (x0 ) < ε,
x − x0

therefore
F (x) − F (x0 )
F ′ (x0 ) = lim = f (x0 ),
x→x0 x − x0
as desired.
Rx
′
We now prove (ii). Let F be a primitive of f . Then, since a f (t) dt = f (x) (by (i)),
 Z x ′
F (x) − f (t) dt = F ′ (x) − f (x) = f (x) − f (x) = 0 ∀ x ∈ (a, b).
a
Rx
By Corollary 5.48, this implies that F (x)− a f (t) dt is constant on [a, b], concluding the proof
of (ii).

f (x0 )
<ε
f (x)
f

28 x0 x

Figure 7.1: This figure illustrates the key estimate in the proof: F (x) − F (x0 ) equals the
rectangle f (x0 )(x − x0 ) plus a red area with absolute value < ε|x − x0 |, hence F (x)−F
x−x0
(x0 )
→
f (x0 ) as x → x0 .

Corollary 7.5: Integral vs. Derivative

If F : [a, b] → R is continuously differentiable, then for all x ∈ [a, b],
Z x
F (x) = F (a) + F ′ (t) dt.
a

Rx
Proof. Since F is a primitive of F ′ , Theorem 7.4 yields F (x) = a F ′ (t) dt + C. Evaluating
at x = a gives C = F (a).

Corollary 7.6: Riemann Integral and Primitives

If f : [a, b] → R is continuous and F is a primitive of f , then
Z b
f (t) dt = F (b) − F (a).
a

Proof. Apply Corollary 7.5 with F ′ = f and x = b.

Version: December 30, 2025. 202

 Chapter 7.1 The Fundamental Theorem of Calculus

Example 7.7. — For all a < b:

Z b
1. ex dx = eb − ea ,
a
Z b
2. sin x dx = − cos b + cos a,
a
Z b
3. cos x dx = sin b − sin a,
a
Z b
4. sinh x dx = cosh b − cosh a,
a
Z b
5. cosh x dx = sinh b − sinh a,
a
b
b1+α − a1+α
Z
6. xα dx = for 1 + α ̸= 0 and 0 < a < b,
a 1+α
Z b
dx
7. = log b − log a for 0 < a < b.
a x

Exercise 7.8. — Let f : [a, b] → R be discontinuous at at most finitely many points. Show
Rx
that F (x) = a f (t) dt is continuous on [a, b], differentiable at every continuity point of f , and
satisfies F ′ (x) = f (x) at those points.
28

Exercise 7.9. — Let f : [a, b] → R be continuous. Show that there exists ξ ∈ (a, b) with
Z b
f (x) dx = f (ξ) (b − a).
a

7.1.2 Integration by Parts and by Substitution

b
Given a function h : [a, b] → R, we use the notation h(x) a := h(b) − h(a).


Theorem 7.10: Integration by Parts

If f, g : [a, b] → R are continuously differentiable, then
Z b b
Z b
′
f ′ (x) g(x) dx.

f (x) g (x) dx = f (x)g(x) a −
a a

Proof. By Proposition 5.10, (f g)′ = f ′ g + f g ′ . Rearranging and integrating, thanks to Corol-

lary 7.6 we get
Z b Z b Z b  b
Z b
f g ′ dx = (f g)′ dx − f ′ g dx = f g a − f ′ g dx.
a a a a

Version: December 30, 2025. 203

 Chapter 7.1 The Fundamental Theorem of Calculus

As a convention, for any h : [a, b] → R,

Z a Z b
h(x) dx = − h(x) dx. (7.3)
b a

Theorem 7.11: Integration by Substitution, 1st Form

Let I, J ⊂ R be intervals, f : I → J be continuously differentiable, and g : J → R be

continuous. For any [a, b] ⊂ I,
Z b Z f (b)
′
g(f (x)) f (x) dx = g(y) dy.
a f (a)

Ry
Proof. Fix y0 ∈ J and set G(y) = y0 g(t) dt. Since G′ = g, by the chain rule (see Theo-
rem 5.14) we get (G◦f )′ = G′ (f )f ′ = g(f )f ′ . Integrating this identity and using Corollary 7.6
yields
Z b Z b
′
g(f (x)) f (x) dx = (G ◦ f )′ (x) dx = G(f (b)) − G(f (a))
a a
Z f (b) Z f (a) Z f (b)
= g(t) dt − g(t) dt = g(t) dt.
y0 y0 f (a)

28
Before stating the next result, we note the following: If h : [a, b] → R is continuously
differentiable with h′ ̸= 0, then h′ has constant sign on [a, b], so h is strictly monotone and in-
vertible; h−1 is continuous by Theorem 3.28 and differentiable on (h(a), h(b)) by Theorem 5.20.
Furthermore, since (h−1 )′ = h′ ◦h1
−1 , also (h
−1 )′ is continuous.

Theorem 7.12: Integration by Substitution, 2nd Form

Let I, J ⊂ R be intervals, f : I → J be C 1 , and g : J → R be continuous. Let [a, b] ⊂ I

and assume f ′ (x) ̸= 0 on [a, b]. If f −1 : [f (a), f (b)] → R denotes the inverse of f |[a,b] ,
then Z b Z f (b)
g(f (x)) dx = g(y) (f −1 )′ (y) dy.
a f (a)

Proof. In order to apply Theorem 7.11, we first observe that

Z b Z b Z b
g(f (x)) ′ g(f (x))
g(f (x)) dx = f (x) dx = f ′ (x) dx.
a a f ′ (x) a f ′ ◦ f −1 (f (x))
g
So we can apply Theorem 7.11 with f ′ ◦f −1
in place of g to get

Z b Z f (b)
g(y)
g(f (x)) dx = dy.
a f (a) f ′ (f −1 (y))

Version: December 30, 2025. 204

 Chapter 7.2 The Fundamental Theorem of Calculus

28 Since 1
f ′ ◦f −1
= (f −1 )′ (recall Theorem 5.20), the result follows.

7.1.3 Improper Integrals

A function f : I → R is locally integrable if f |[a,b] is integrable for every compact [a, b] ⊂ I.

Definition 7.13: Improper Integrals

Let I ⊆ R be a nonempty interval and f : I → R be locally integrable. Set c = inf I ∈
R ∪ {−∞} and d = sup I ∈ R ∪ {∞}, and fix x0 ∈ I. We define the improper integral
of f on I by
Z d Z x0 Z b
f (x) dx := lim f (x) dx + lim f (x) dx,
c a→c+ a b→d− x0

whenever both limits exist and the sum is well-defined (we do not allow the indetermi-
nate form ∞−∞). Here the first limit is taken over a ∈ I with c < a < x0 (interpreting
a → −∞ if c = −∞) and the second over b ∈ I with x0 < b < d (interpreting b → +∞
if d = ∞). If the value is finite we say the integral converges; if it is ±∞ we say it di-
verges to ±∞; otherwise, it does not converge. When defined, the value is independent
of the choice of x0 .
29

Example 7.14. — Consider the integral of f (x) = x1 over the interval I = (0, 1). The
function f is continuous, and hence locally integrable, on I. However, f is unbounded near 0,
so it is not integrable on [0, 1] in the usual Riemann sense, and the integral over (0, 1) must
be understood as an improper integral.
Since the only problem is at 0 and the integrand is bounded and continuous up to 1, we
can write Z 1 Z 1
1 1
dx = lim dx.
0 x a x
a→0+

Using the primitive log x, we compute

Z 1
1
dx = [log x]1a = log 1 − log a = − log a,
a x

therefore Z 1
1
dx = lim (− log a) = +∞.
0 x a→0+

Thus the improper integral over (0, 1) diverges to +∞.

Version: December 30, 2025. 205

 Chapter 7.2 The Fundamental Theorem of Calculus

Example 7.15. — Recalling that arctan′ (x) = 1

1+x2
, we have
Z ∞
dx  b π
2
= lim arctan x 0 = lim arctan b = .
0 1+x b→∞ b→∞ 2

R∞
Example 7.16. — For α ∈ R, we want to compute 1 x−α dx. Note that

b b b
x1−α b1−α − 1
Z  Z
−α dx
x dx = = for α ̸= 1, = log b for α = 1,
1 1−α 1 1−α 1 x

so letting b → ∞ we obtain

∞
1 
Z
−α
, α > 1,
x dx = α − 1
1 
∞, α ≤ 1.

In particular, the integral converges if and only if α > 1.

As we now show, when f ≥ 0 the improper integral over an interval always exists in the
extended sense: it either converges to a finite value or diverges to +∞.

Lemma 7.17: Improper Integral of Nonnegative Functions

Let I ⊆ R be a nonempty interval with c = inf I and d = sup I, and let f : I → [0, ∞)
29 be locally integrable. Fix any x0 ∈ I. Then the one-sided limits
Z x0 Z b
L− := lim f (x) dx, L+ := lim f (x) dx
a→c+ a b→d− x0

exist in [0, ∞], and

Z d Z β
f (x) dx = L− + L+ = sup f (x) dx ∈ [0, ∞].
c c<α<β<d α

In particular, the improper integral over I always exists in the extended sense and equals
+∞ whenever the supremum is +∞.

Rx Rb
Proof. Since f ≥ 0, the maps a 7→ a 0 f (for a < x0 ) and b 7→ x0 f (for b > x0 ) are monotone,
hence the limits L− , L+ exist in [0, ∞]. Also, for any c < a < x0 b < d,
Z b Z x0 Z b
af = f+ f.
a a x0

Taking suprema gives

Z β Z x0 Z b
sup f = sup f + sup f = L− + L+ ,
c<α<β<d α a<x0 a b>x0 x0

Version: December 30, 2025. 206

 Chapter 7.2 The Fundamental Theorem of Calculus

Rd
which equals the definition of c f above.

Example 7.18. — Consider

Z ∞ Z −1 Z 1 Z ∞
2 2 2 2
e−x dx = e−x dx + e−x dx + e−x dx.
−∞ −∞ −1 1

2
Since x2 ≥ x for x ≥ 1, we have e−x ≤ e−x on [1, ∞), hence
Z ∞ Z ∞ b
2
e−x dx ≤ e−x dx = lim − e−x = e−1 < ∞.

b→∞ 1
1 1

R −1 2 R∞ 2 R1 2 R1
By symmetry, −∞ e−x dx = 1 e−x dx < ∞. Finally, −1 e−x dx ≤ −1 1 dx = 2. Thus the
improper integral converges.

Theorem 7.19: Integral Test for Series

Let f : [0, ∞) → [0, ∞) be monotone decreasing. Then, for every N ∈ N,

N +1 Z N +1 N
29
X X
f (n) ≤ f (x) dx ≤ f (n).
n=1 0 n=0

In particular,
∞
X Z ∞
f (n) converges ⇐⇒ f (x) dx converges.
n=1 0

Proof. By monotonicity, f is locally integrable. Define step functions on [0, ∞) by

u(x) = f (⌊x⌋), ℓ(x) = f (⌈x⌉),

where ⌊x⌋ is the rounding function (i.e. the largest integer ≤ x), while ⌈x⌉ denotes the smallest
integer ≥ x. Then ℓ ≤ f ≤ u, and for N ≥ 1,

N
X +1 Z N +1 Z N +1 Z N +1 N
X
f (n) = ℓ(x) dx ≤ f (x) dx ≤ u(x) dx = f (n).
n=1 0 0 0 n=0

The result follows by taking the limit as N → ∞.

Version: December 30, 2025. 207

 Chapter 7.2 The Fundamental Theorem of Calculus

f (2)
f (3)
f

0 1 2 3 4 5 6 7 8 9 10

Example 7.20. — The harmonic series can be written as {f (n)}∞ n=0 with f (x) = 1+x ,
1

hence it diverges since Z ∞

dx
= ∞.
0 1+x
Moreover, Theorem 7.19 allows us to estimate its partial sums. Let HN = N n=1 n be the
1
P

N -th partial sum. Applying the integral test with f (x) = 1+x
1
, we obtain

N +1 Z N +1 N
1 dx 1
29
X X
≤ ≤ .
1+n 0 1+x 1+n
n=1 n=0

The left-hand side is

N +1 N +2
X 1 X 1
= = HN +2 − 1,
1+n n
n=1 n=2

the integral is Z N +1
dx  N +1
= log(1 + x) 0 = log(N + 2),
0 1+x
and the right-hand side is
N N +1
X 1 X 1
= = HN +1 .
1+n n
n=0 n=1

Thus, for every N ≥ 1,

HN +2 − 1 ≤ log(N + 2) ≤ HN +1 .

In particular, HN → ∞ as N → ∞, and these inequalities show that the harmonic series

grows like log N .
In contrast, ∞ n=1 n2 converges because
1
P

Z ∞
dx
< ∞.
0 (1 + x)2

Version: December 30, 2025. 208

 Chapter 7.2 The Fundamental Theorem of Calculus

7.2 Integration and Differentiation of Power Series

Rx n+1 n+1
From Example 7.7(6), for n ≥ 0 we have 0 tn dt = xn+1 , i.e. xn+1 is a primitive of xn . Also,
by Corollary 5.12, (xn )′ = nxn−1 . These formulas allow us to integrate and differentiate
polynomials. We now address integrating and differentiating power series.
We recall the limit r
n 1
lim =1 (7.4)
n→∞ n
see Exercise 3.53. Also, we shall use the following fact.

Remark 7.21. — Let (an )∞

n=0 be a sequence of nonnegative numbers with

L = lim sup an < ∞,

n→∞

and let (bn )∞

n=0 and (γn )n=0 satisfy bn → 1 and γn → 1. Then
∞

lim sup aγnn bn = L.

n→∞
29
In other words, multiplying by a factor that tends to 1, or raising to an exponent that tends
to 1, does not change the value of the lim sup.

Exercise 7.22. — Prove Remark 7.21.

Hint: Fix ε ∈ (0, 1). Since bn , γn → 1 as n → ∞, there exists N such that

1 − ε ≤ bn , γn ≤ 1 + ε for all n ≥ N.

Observe now that, given a ≥ 0, for every s ∈ [1 − ε, 1 + ε] it holds

min a1−ε , a1+ε ≤ as ≤ max a1−ε , a1+ε ,

 

where the minimum/maximum depends on whether a ≤ 1 or a ≥ 1. Apply this with a = an

and s = γn , and combine with 1 − ε ≤ bn ≤ 1 + ε, to deduce

(1 − ε) min a1−ε 1+ε

≤ aγnn bn ≤ (1 + ε) max a1−ε 1+ε
for all n ≥ N.
 
n , an n , an

Then take the lim sup in these inequalities and finally let ε → 0.

Version: December 30, 2025. 209

 Chapter 7.2 The Fundamental Theorem of Calculus

Theorem 7.23: Integration of Power Series

Let f (x) = ∞ n
P
n=0 an x have radius of convergence R > 0. Then

∞
X an
F (x) = xn+1
n+1
n=0

has the same radius of convergence R and is a primitive of f on (−R, R).

an−1
Proof. Set ρ = lim supn→∞ n |an |, so that R = ρ−1 . Define c0 = 0 and cn = for n ≥ 1,
p
n
so F (x) = ∞ n=0 cn x . Noticing that
n
P

r   n−1
n 1 n
p
n
p
n−1
|cn | = |an−1 | ,
n
q
it follows from (7.4) and Remark 7.21 (applied with bn = n n1 and γn = n )
n−1
that
p
n n−1
p
lim sup |cn | = lim sup |an−1 | = ρ,
n→∞ n→∞

hence also F has radius of convergence R.

We now want to prove that F ′ = f . Fix [−r, r] ⊂ (−R, R) and define the polynomials
fn (x) = nk=0 ak xk . Then
P
Z x n
29
X ak k+1
fn (t) dt = x .
0 k+1
k=0

By Theorem 4.42, the sequence of functions (fn )∞

n=0 converge uniformly to f on [−r, r], so
Theorem 6.38 yields
Z x Z x
f (t) dt = lim fn (t) dt ∀ x ∈ [−r, r].
0 n→∞ 0

On the other hand, again by Theorem 4.42,

n
X ak k+1
lim x = F (x) ∀ x ∈ [−r, r].
n→∞ k+1
k=0
Rx
This proves that F (x) = 0 f (t) dt on [−r, r], so Theorem 7.4 implies that F ′ (x) = f (x) on
[−r, r]. Since [−r, r] ⊂ (−R, R) is arbitrary, we proved that F ′ = f on (−R, R).

Corollary 7.24: Differentiation of Power Series

Let f (x) = ∞ n
P
n=0 an x have radius of convergence R > 0. Then f is differentiable on
(−R, R) with
∞
X
f ′ (x) = nan x n−1 ∀ x ∈ (−R, R),
n=1

and the series on the right has radius of convergence R.

Version: December 30, 2025. 210

 Chapter 7.2 The Fundamental Theorem of Calculus

Proof. Let cn = (n + 1)an+1 and g(x) = ∞

P∞
n=0 cn x . Let R̄ be the radius
n−1 = n
P
n=1 nan x
of g. Then Theorem 7.23 implies that the power series
∞
X cn n+1
G(x) = x
n+1
n=0

has radius R̄ and is a primitive of g. But

∞ ∞ ∞
X (n + 1)an+1 n+1
X
n+1
X
G(x) = x = an+1 x = an xn = f (x) − a0 .
n+1
n=0 n=0 n=1

This implies that G and f have the same radius of convergence (so R̄ = R) and that g = G′ =
(f − a0 )′ = f ′ .

Exercise 7.25. — Let f (x) = ∞ n=0 an x have radius of convergence R > 0. Show that f
n
P

is C ∞ on (−R, R) and express f (n) as a power series for each n ∈ N.

Exercise 7.26. — Let f (x) = ∞

P∞
n=0 an x and g(x) = n=0 bn x with radii Rf , Rg > 0,
n n
P

and set R = min{Rf , Rg }. Prove that if f (x) = g(x) for all x ∈ (−R, R), then an = bn for all
n. In particular, Rf = Rg .
29

Exercise 7.27. — Let α ∈ R. The goal of this exercise is to show that

∞  
X α n
α
(1 + x) = x ∀ x ∈ (−1, 1), (7.5)
n
n=0
 
α α(α − 1) · · · (α − n + 1)
where α
is defined as .


n =
n n!

(a) Show that, for α ∈/ N, the series f (x) = ∞ α n

n=0 n x has radius of convergence 1.
P

(b) Compute f ′ and show that

f (x)
f ′ (x) = α ∀ x ∈ (−1, 1). (7.6)
1+x

f
′
(c) Let g(x) = (1 + x)α and use (7.6) to prove g = 0 on (−1, 1). Conclude the validity of
(7.5) from the fact that f (0) = g(0) = 1.

Example 7.28. — We have already seen in Example 4.23 that, as a consequence of the
Leibniz criterion (Proposition 4.22), the alternating harmonic series converges. However, we

Version: December 30, 2025. 211

 Chapter 7.2 The Fundamental Theorem of Calculus

were unable to determine the value of the series. We now show that
∞
X (−1)n+1
= log(2).
n
n=1

Using the geometric series and the fundamental theorem of calculus, for x ∈ (−1, 1) we
have
∞
1 1 X
(log(1 + x))′ = = = (−1)n xn ,
1+x 1 − (−x)
n=0

hence, since log(1) = 0, Theorem 7.23 yields

Z x X (−1)n ∞ X (−1)k+1 ∞
1
log(1 + x) = dt = xn+1 = xk .
0 1+t n+1 k
n=0 k=1

xk
For x ∈ [0, 1], the terms ak = k are nonnegative, decreasing, and converge to 0, so by
Proposition 4.22,

2n 2n+1
X (−1)k+1 k
X (−1)k+1 k
x ≤ log(1 + x) ≤ x ∀ n ∈ N.
k k
k=1 k=1

Letting x → 1− yields that, for each n ∈ N,

2n 2n+1
29 X (−1)k+1 X (−1)k+1
≤ log(2) ≤ .
k k
k=1 k=1

Now, using again Proposition 4.22, we can let n → ∞ to get the desired result.

Example 7.29. — Similarly to what we did in the previous exercise, we note that
∞
′ 1 X
arctan (x) = = (−1)k x2k ∀ x ∈ (−1, 1),
1 + x2
k=0

hence Z x ∞
X (−1)k
1
arctan x = dt = x2k+1 .
0 1 + t2 2k + 1
k=0

Using the Leibniz criterion as before yields

2n+1 2n
X (−1)k 2k+1 X (−1)k
x ≤ arctan(x) ≤ x2k+1 ∀ n ∈ N.
2k + 1 2k + 1
k=0 k=0

Letting first x → 1− gives

2n+1 2n
X (−1)k π X (−1)k
≤ arctan(1) = ≤ ∀ n ∈ N,
2k + 1 4 2k + 1
k=0 k=0

Version: December 30, 2025. 212

 Chapter 7.2 The Fundamental Theorem of Calculus

and letting n → ∞ gives

∞
X (−1)n π
= .
2n + 1 4
n=0

Sometimes the primitive of a function cannot be expressed in terms of the “standard”

elementary functions.

Example 7.30 (Integral Sine). — The integral sine is the primitive function Si : R → R
of the continuous function
29 (
sin(x)
x if x ̸= 0
x ∈ R 7→
1 if x = 0
Rx sin(t)
with the normalisation Si(0) = 0, that is Si(x) = 0 t dt. Thanks to Theorem 7.23, the
function Si can be expressed as a power series:

x ∞
xX ∞
(−1)n 2n (−1)n
Z Z
sin(t) X
Si(x) = dt = t dt = x2n+1
0 t 0 n=0 (2n + 1)! (2n + 1)!(2n + 1)
n=0

for all x ∈ R.

Version: December 30, 2025. 213

 Chapter 7.3 Integration Methods

7.3 Integration Methods

Let I ⊆ R be an interval, and f : I → R a function. The notation
Z
f (x) dx = F (x) + C

means that F is a primitive function (antiderivative) of f . In the expression F (x) + C, C is

read as an indefinite constant, usually called integration constant. Since the domain I of
f is an interval, two primitive functions of f differ by a constant, which makes the notation
meaningful. One calls F (x) + C the indefinite integral of f . Indefinite integrals of special
functions can be found in tables or by means of computer algebra systems. In this section,
we show general methods to determine indefinite integrals.
Throughout this section, I ⊆ R denotes a non-empty interval that is not a single point.
Also, all functions in this section are real-valued functions with domain I that are integrable
on any compact interval [a, b] ⊆ I.

7.3.1 Integration by Parts and by Substitution in Leibniz Notation

In the computation of indefinite integrals, it is convenient to use Leibniz notation. This
notation allows us to reformulate, in a natural formalism, both integration by parts and by
substitution (see Section 7.1.2). We recall that the derivative of a function h is denoted by h′
30 or by dh
dx . In this section, the second notation (called Leibniz notation) will be useful.
Integration by Parts. Let f and g be functions with primitives F and G, respectively.
Recall that, from the product rule for the derivative in Proposition 5.10, it follows that
(F G)′ = f G + F g. This implies the integration by parts formula
Z Z
F (x) g(x) dx = F (x)G(x) − f (x) G(x) dx + C. (7.7)

In Leibniz notation, f = dF
dx and g = dx . This leads to the notation f dx = dF and g dx = dG,
dG

and integration by parts is sometimes written as

Z Z
F dG = F G − G dF + C,

which should be understood as a short form of formula (7.7).

Integration by Substitution. Let J be an interval and let f : I → J be a differentiable
function. If G : J → R is a primitive of g then, by the chain rule in Theorem 5.14, [G(f (x))]′ =
g(f (x)) f ′ (x) for all x ∈ I. From this it follows that
Z
g(f (x)) f ′ (x) dx = G(f (x)) + C.

Version: December 30, 2025. 214

 Chapter 7.3 Integration Methods

Since G(u) = g(u) du + C, we obtain

R

Z Z
g(f (x)) f ′ (x) dx = g(u) du + C (7.8)

where we used the change of variables u = f (x). The substitution rule is also called change
of variable, as one has replaced the variable u in g(u) du by u = f (x). In Leibniz notation
R

this is very natural: if u = f (x) then du = f ′ (x) dx, and (7.8) follows.
We also recall the second form of the substitution rule: if f ′ ̸= 0 we can set x = f −1 (u) so
that du
dx
= (f −1 )′ (u), and obtain
Z Z
dx
g(f (x)) dx = g(u) du + C, (7.9)
du

see Section 7.1.2.

7.3.2 Integration by Parts: Examples

Example 7.31. — We want to calculate the indefinite integral xex dx. Since ex = (ex )′ ,
R

using (7.7) we get

Z Z Z Z
x x ′ x ′ x x
xe dx = x(e ) dx = xe − x · e dx + C = xe − ex dx + C.

30
Since ex dx = ex + C, we conclude that
R

Z
xex dx = xex − ex + C.

We note that it is sufficient to use only one integration constant C in such calculations, since
several such constants can be combined into one.

Example 7.32. — We calculate the integral

R
log(x) dx:
Z Z Z
log(x) dx = log(x) · 1 dx = log(x) · x′ dx
Z
= log(x) · x − log′ (x) x dx + C
Z
1
= log(x) · x − x dx + C
x
Z
= log(x) · x − 1 dx + C = x log(x) − x + C.

Suggestion: To ensure that the final result is correct, differentiate the result and check if you
get the original function. For instance, in this case, one can easily check that

(x log(x) − x + C)′ = log(x).

Version: December 30, 2025. 215

 Chapter 7.3 Integration Methods

Exercise 7.33. — Give a recursive formula for calculating the indefinite integrals
Z Z Z
n x n
x e dx , x sin(x) dx , xn cos(x) dx

for n ∈ N.

Exercise 7.34. — Calculate

Z Z
xs log(x)dx , eax sin(bx) dx

for all s, a, b ∈ R. Note that the case s = −1 needs to be treated separately, in analogy with
Example 7.7(6)-(7).

7.3.3 Integration by Substitution: Examples

Example 7.35. — We want to compute x
dx. Let u = f (x) = 1 + x2 , so that
R
1+x2
du = f ′ (x) dx = 2x dx. Then we find
Z Z Z
x 1 1 1 1 1 1
2
dx = 2
(2x dx) = du = log |u| = log(1 + x2 ) + C.
1+x 2 1+x 2 u 2 2
30
R√
Example 7.36. — Given r > 0, we want to compute the indefinite integral r2 − x2 dx.
Due to the trigonometric identity r2 − r2 sin(θ)2 = r cos(θ) it is convenient to use the
p

change of variable x = r sin(θ), θ ∈ (− π2 , π2 ). With this choice we have dx = r cos(θ) dθ,

therefore
Z p Z p Z
2
2 2
r − x dx = r 2 2 2
r − r sin(θ) cos(θ) dθ = r cos2 (θ) dθ.

To compute cos2 (θ) dθ we use integration by parts as follows:

R

Z Z
2
cos (θ) dθ = cos(θ) sin′ (θ) dθ
Z
= cos(θ) sin(θ) − cos′ (θ) sin(θ) dθ + C
Z
= cos(θ) sin(θ) + sin2 (θ) dθ + C.

Since sin2 (θ) = 1 − cos2 (θ), we get

Z Z Z
2
cos (θ) dθ = cos(θ) sin(θ) + 1 dθ − cos2 (θ) dθ + C
Z
= cos(θ) sin(θ) + θ − cos2 (θ) dθ + C,

Version: December 30, 2025. 216

 Chapter 7.3 Integration Methods

therefore
Z Z
1
2 cos2 (θ) dθ = cos(θ) sin(θ) + θ + C cos2 (θ) dθ =


=⇒ cos(θ) sin(θ) + θ + C.
2

(Note that, since C ∈ R is arbitrary, in the last formula we still write C in place of 2 .)
C
This
proves that

r2
Z p Z
2
cos2 (θ) dθ =


2 2
r − x dx = r sin(θ) cos(θ) + θ + C.
2

(Again, we write C in place of Cr2 .) Recalling that x = r sin(θ) with θ ∈ (− π2 , π2 ), it follows

q
2
that θ = arcsin xr and cos(θ) = 1 − xr2 , therefore

r2
Z p
1 p x


r2 − x2 dx = x r2 − x2 + arcsin r + C.
2 2

Substitutions like in Example 7.39 are called trigonometric substitutions. We will not
always argue carefully in these calculations and will rather trust the Leibniz notation, but
recall that, to apply (7.9), there must be invertibility of the function when we express the old
variable by the new variable.
For the following list of trigonometric substitutions, let n ∈ Z.
n
30 • In expressions of the form (a2 − x2 ) 2 for a > 0, as already seen in the example above,
one considers the substitution x = a sin(θ) with θ ∈ (− π2 , π2 ), giving dx = a cos(θ) dθ
1
and (a2 − x2 ) 2 = a cos(θ).
n
• In expressions of the form (a2 + x2 ) 2 for a > 0, the substitution
n x = a tan(θ) with θ ∈
n
(− 2 , 2 ) yields dx = cos2 (θ) dθ and (a + x ) 2 = cos(θ) .
π π a 2 2 a

• Although this is not a trigonometric substitution, we still note the following: For the
n n
expression x(a2 − x2 ) 2 or the expression x(a2 + x2 ) 2 , the substitutions u = a2 − x2 and
u = a2 + x2 , respectively, allow us to compute the indefinite integrals.

Example 7.37. — (i) Given a > 0, using the substitution x = a tan(θ), recalling that
1
a
(a2 + x2 ) 2 = cos(θ) and dx = cosa2 (θ) dθ (recall that tan′ (θ) = cos12 (θ) ), we get

cos3 (θ) a
Z Z Z
1 1 1
3 dx = dθ = 2 cos(θ) dθ = sin(θ) + C
(a2 + x2 ) 2 a3 cos2 (θ) a a2
1 x
= 2
tan(θ) cos(θ) + C = √ + C.
a a2 a2 + x2

(ii) Choosing u = 1 − x2 (so that du = −2x dx), we have

Z Z
p 1 1 1
2
x 1 − x dx = − u1/2 du = − u3/2 + C = − (1 − x2 )3/2 + C.
2 3 3

Version: December 30, 2025. 217

 Chapter 7.3 Integration Methods

Certain indefinite integrals can be computed with hyperbolic substitutions. For instance,
n
for expressions of the form (x2 − a2 ) 2 with a ∈ R, the substitution x = a cosh(u) yields
1
dx = a sinh(u) du and (x2 − a2 ) 2 = a sinh(u).

Example 7.38. — Using the substitution x = cosh(u) (so dx = sinh(u) du), we compute
Z p Z q Z
x2 − 1 dx = cosh2 (u) − 1 sinh(u) du = sinh2 (u) du.

In analogy to the argument used in Example 7.36, we compute sinh2 (u) du as follows:
R

Z Z
2
sinh (u) du = cosh(u) sinh(u) − cosh2 (u) du + C
Z
1 + sinh2 (u) du + C


= cosh(u) sinh(u) −
Z
= cosh(u) sinh(u) − u − sinh2 (u) du + C,

This yields

cosh(u) sinh(u) − u
Z Z
2
2 sinh (u) du = cosh(u) sinh(u)−u+C =⇒ sinh2 (u) du = +C,
2

hence √
cosh(u) sinh(u) − u x x2 − 1 − arcosh(x)
Z p
30 2
x − 1 dx = +C = + C.
2 2
Another method that we would like to mention briefly here is the so-called half-angle
method (or Weierstrass substitution). This is useful for the integral of expressions like
cos2 (x)+cos(x)+sin(x)
sin(x) or , see also Remark 7.42 below. We show this method in detail in
1
1+sin(x)
the next example.

Example 7.39. — We want to compute sin(x) dx, and we consider the change of variable
R 1

u = tan 2 . We can note that, by the doubling angle formulas for sine and cosine (see (4.14)),
x

it follows that
sin( x )
x x 2 cos 2x 2 tan x2


(2) 2u
sin(x) = 2 sin cos = = 2 x

=
2 2 sin2 ( x2 ) 1 + tan 1 + u2
1 + cos2 x 2
(2)
2
and analogously cos(x) = 1−u1+u2
.
Furthermore, the relation u = tan x
implies that x = 2 arctan(u), therefore dx = 2


2 1+u2
du
(recall that arctan′ (s) = 1+s
1
2 ).

Using these formulas, we get

1 + u2 2
Z Z Z
1 1 x


dx = du = du = log |u| + C = log tan 2 + C.
sin(x) 2u 1 + u2 u

Version: December 30, 2025. 218

 Chapter 7.3 Integration Methods

7.3.4 Integration of Rational Functions

A function of the form f (x) = p(x)

q(x) for polynomials p and q ̸= 0 is called a rational function.
In this section, we show a procedure for computing the indefinite integral of a rational function
f = pq on an interval I on which q has no zeros. By polynomial division with remainder, one
can always write f = pq in the form f = g+ qr , where g and r are polynomials with deg r < deg q.
The polynomial function g is easy to integrate. Therefore, we always assume that the degree
of p is smaller than the degree of q.
We start by integrating some elementary rational functions. Let a ∈ R and n ∈ N with
n ≥ 2. Then:
Z
1
dx = log |x − a| + C (7.10)
x−a
Z
1 1
n
dx = (x − a)1−n + C (7.11)
(x − a) 1−n
Z
1 1
arctan xa + C (7.12)


2 2
dx =
a +x a
Z
x 1
2 2
dx = log(a2 + x2 ) + C (7.13)
a +x 2
Z
x 1
2 2 n
dx = (a2 + x2 )1−n + C (7.14)
(a + x ) 2(1 − n)

The integrals (7.10) and (7.11) are calculated with substitution u = x−a, for (7.12) substitute
30 u = xa , for (7.13) and (7.14) substitute u = a2 + x2 .
To integrate a general rational function, we use what is called the partial fraction decom-
position of rational functions. Let p, q be polynomials without nontrivial common divisors
such that q ̸= 0 and deg p < deg q.
First, factorize the polynomial q into linear and quadratic factors

q(x) = (x − a1 )k1 · · · (x − an )kn (x2 + b1 x + c1 )ℓ1 · · · (x2 + bm x + cm )ℓm

Then, the rational function p(x)

q(x) can be rewritten as a linear combination of rational functions
of the form
1 1 x
, , ,
(x − ai )k (x2 + bj x + cj )ℓ (x2 + bj x + cj )ℓ
for some k ≤ ki and ℓ ≤ ℓj , and then one needs to integrate each of these individual terms.

x4 + 1
Z
Example 7.40. — We want to calculate the indefinite integral dx. We first
x2 (x + 1)
rewrite the denominator:
x4 + 1 x4 + 1
= .
x2 (x + 1) x3 + x 2
We now perform polynomial division with remainder. We want to write

x4 + 1 = (x3 + x2 ) q(x) + r(x),

Version: December 30, 2025. 219

 Chapter 7.3 Integration Methods

where deg r < 3. The leading term of x4 + 1 is x4 , and the leading term of x3 + x2 is x3 , so
4
the first term of the quotient must be xx3 = x:

x · (x3 + x2 ) = x4 + x3 .

Subtracting this from x4 + 1 gives

(x4 + 1) − (x4 + x3 ) = −x3 + 1.

−x3
Now the leading term of the remainder is −x3 , so the next term in the quotient is x3
= −1:

−1 · (x3 + x2 ) = −x3 − x2 .

Subtracting this from −x3 + 1 gives

(−x3 + 1) − (−x3 − x2 ) = x2 + 1.

The remainder is now x2 + 1, which has degree 2 < 3, so the division stops. Thus

x4 + 1 = (x3 + x2 )(x − 1) + (x2 + 1),

and therefore
x4 + 1 x2 + 1 x2 + 1
30 = x − 1 + = x − 1 + .
x3 + x 2 x3 + x 2 x2 (x + 1)
We now look for a partial fraction decomposition of

x2 + 1
.
x2 (x + 1)

The denominator factors as x2 (x + 1), where:

• x2 is a repeated linear factor,

• x + 1 is a simple linear factor.

For a factor x2 , the general rule is that the numerator should be a polynomial of degree at
most 1 (one less than the power of the factor), so we take a numerator of the form ax + b.
For a simple linear factor x + 1, the numerator is a polynomial of degree at most 0, that is, a
constant c. Thus the most general form of the decomposition is

x2 + 1 ax + b c
2
= 2
+
x (x + 1) x x+1

for some real numbers a, b, c to be determined. To determine a, b, c we multiply both sides by

x2 (x + 1), which gives

x2 + 1 = ax(x + 1) + b(x + 1) + cx2 = ax2 + ax + bx + b + cx2 .

Version: December 30, 2025. 220

 Chapter 7.3 Integration Methods

Comparing the coefficients of x2 , x, and the constant term, we obtain the linear system

a + c = 1, a + b = 0, b = 1,

so a = −1, b = 1, and c = 2. In summary,

x4 + 1 x2 + 1 1 1 2
2
= x − 1 + 2
=x−1− + 2 + ,
x (x + 1) x (x + 1) x x x+1

and therefore

x4 + 1
Z Z Z Z Z Z
1 1 1
dx = x dx − 1 dx − dx + dx + 2 dx
x2 (x + 1) x x2 x+1
x2 1
= − x − log |x| − + 2 log |x + 1| + C.
2 x

Example 7.41. — We calculate the indefinite integral x(x2 +2x+2) 1

dx. Note that the
R

polynomial x + 2x + 2 has no real zeros. For the partial fraction decomposition, we look for
2

a, b, c ∈ R such that
1 a bx + c
2
= + 2 .
x(x + 2x + 2) x x + 2x + 2
To find a, b, c we multiply both sides by x(x2 + 2x + 2) and get

30 1 = a(x2 + 2x + 2) + (bx + c)x = ax2 + 2ax + 2a + bx2 + cx,

thus
a + b = 0, 2a + c = 0, 2a = 1,

which gives a = 12 , b = − 12 , and c = −1. It follows that

Z Z Z
1 1 1 1 x+2
2
dx = dx − 2
dx
x(x + 2x + 2) 2 x 2 x + 2x + 2
Z
1 1 x+2
= log |x| − dx + C
2 2 (x + 1)2 + 1
Z
1 1 u+1
= log |x| − du + C (u = x + 1)
2 2 u2 + 1
Z Z
1 1 u 1 1
= log |x| − 2
du − − 2
du + C
2 2 u +1 2 u +1
1 1 1
= log |x| − log(u2 + 1) − arctan(u) + C
2 4 2
1 1 2 1
= log |x| − log((x + 1) + 1) − arctan(x + 1) + C,
2 4 2

where we used (7.13) and (7.14).

In some cases, the above procedure may also lead to compute integrals of the form
1
dx for an a ∈ R and n ≥ 2, which (as explained previously) we can handle with the
R
(a2 +x2 )n

Version: December 30, 2025. 221

 Chapter 7.3 Integration Methods

trigonometric substitution tan(u) = xa .

Remark 7.42. — Now that we know how to integrate rational functions, we can rediscuss
the half-angle method introduced before. This allows one to compute the integral of rational
expressions in sine and cosine. In fact, with the substitution u = tan x2 , using that

2u 1 − u2 2
sin(x) = , cos(x) = , dx = du,
1 + u2 1 + u2 1 + u2

(see Example 7.39), one ends up with the integral of a rational function in u.
30
cos(x)
Exercise 7.43. — Calculate the indefinite integral dx using the substitution
R
2+sin(x)
u = tan x2 .

Remark 7.44. — Sometimes, one substitution or the other is carried out because there
is a nested function in the function to be integrated, and there is simply no other method
√
available. For example, in the integral sin( x) dx, none of the mentioned methods are
R
√
available, but one is tempted to set u = x, and this indeed leads to an integral that one can
solve. Similarly, in an integral of the form 1+e x dx, one sets u = e .
R 1 x

7.3.5 Definite Integrals

Using the methods developed in the previous sections, we can now calculate both integrals on
compact intervals [a, b] and improper integrals (recall Definition 7.13). We now discuss some
examples of the latter.

Z 1
Example 7.45. — We compute the improper integral log(x) dx (this is improper since
0
x 7→ log(x) is unbounded as x → 0+ ). Recalling (7.32) and Example 3.64, we have
Z 1 Z 1 1
31

log(x) dx = lim log(x) dx = lim x log(x) − x a
0 a→0+ a a→0+


= lim log(1) − 1 − a log(a) + a = −1.
a→0+

Z π
2
Exercise 7.46. — Calculate tan(x) dx.
0

Z ∞
Exercise 7.47. — Decide for which p ∈ R≥0 the improper integral x sin(xp ) dx con-
0
verges.

Version: December 30, 2025. 222

 Chapter 7.3 Integration Methods

7.3.6 The Gamma Function

The Gamma-function Γ is defined, for s ∈ (0, ∞), by the improper integral
Z ∞
Γ(s) = xs−1 e−x dx. (7.15)
0

To verify that this improper integral indeed converges, we examine the integration limits 0
and ∞ separately. For 0 < a < b we find, using integration by parts,
Z b Z b Z b
s−1 −x 1 s ′ −x 1  s −x b 1
x e dx = (x ) e dx = x e a
+ xs e−x dx. (7.16)
a s a s s a

We obtain
b
1  s −x b 1 b s −x
Z  Z 
s−1 −x
x e dx = lim x e a
+ x e dx
0 a→0 s s a
1 b s −x
Z
1
= bs exp(−b) + x e dx,
s s 0

where the integral on the right is an actual Riemann integral since the function xs e−x is
continuous on [0, b]. To investigate the upper limit of integration, we note that there exists
31 R > 0 such that ex > xs+2 holds for all x > R. Thus
Z ∞ Z R Z ∞
s −x s −x
x e dx ≤ x e dx + x−2 dx < ∞
0 0 R

which shows that (7.16) converges as b → ∞. Specifically, we obtain

Z ∞  Z b  Z ∞
s−1 −x 1 s −x 1
x e dx = lim 1 s
s b exp(−b) + x e dx = xs e−x dx.
0 b→∞ s 0 s 0

This shows that the gamma function satisfies the relation

Γ(s + 1) = s Γ(s) (7.17)

for all s ∈ (0, ∞), from which one can deduce that the Gamma function extends the factorial
function from N to (0, ∞). In fact
Z ∞ Z ∞ ∞
0 −x
= e−x dx = e−x 0 = e0 = 1,

Γ(1) = x e dx =
0 0

therefore (7.17) implies that

Γ(n + 1) = nΓ(n) = n(n − 1)Γ(n − 1) = . . . = n! Γ(1) = n! ∀ n ∈ N.

Version: December 30, 2025. 223

 Chapter 7.4 Taylor Series

7.4 Taylor Series

7.4.1 Taylor Approximation

As we have seen in Chapter 5, given a differentiable function f : D → R, the derivative f ′ (x0 )
gives the slope of the tangent to the graph of f at x0 . Moreover, the corresponding affine
function
x 7→ f (x0 ) + f ′ (x0 )(x − x0 )

approximates the function f within an error o(|x − x0 |) as x → x0 , see (5.2). The idea behind
Taylor’s theorem is that the “quality” of the approximation can be increased by considering
higher-order polynomials instead of affine approximations.
In this section, it will be convenient to use the following abuse of notation: given a, b ∈ R,
irrespective of the order between a and b, [a, b] denotes the interval between them. In other
words, for all a, b ∈ R, [a, b] and [b, a] denote the same interval.
We also recall that, if a < b, then
Z b Z b
f (x) dx ≤ |f (x)| dx,
a a
31
see Theorem 6.24. If instead b < a, then a minus sign appears (recall (7.3)) and we get
Z b Z a Z a Z b Z b
f (x) dx = f (x) dx ≤ |f (x)| dx = − |f (x)| dx = |f (x)| dx .
a b b a a

In conclusion, independently of the order of a and b, we always have

Z b Z b
f (x) dx ≤ |f (x)| dx .
a a

Let I ⊆ R be an open interval, and f : I → R be an n times differentiable function. The

n-th Taylor approximation of f around a point x0 ∈ I is the polynomial function
n
X f (k) (x0 )
Pn (x) = (x − x0 )k . (7.18)
k!
k=0

Note that, with this choice of the coefficients, P (k) (x0 ) = f (k) (x0 ) for k ∈ {0, . . . , n}.
We will state and prove different versions of Taylor’s Theorem. We begin with this first
version:

Version: December 30, 2025. 224

 Chapter 7.4 Taylor Series

Theorem 7.48: Taylor Expansion to Order n with Integral Remainder

Let n ≥ 1, f : [a, b] → R an n-times continuously differentiable function, and fix
x0 ∈ [a, b]. Then, for all x ∈ [a, b],
x
(x − t)n−1
Z
f (x) = Pn−1 (x) + f (n) (t) dt, (7.19)
x0 (n − 1)!

where Pn−1 is the (n − 1)-th Taylor approximation of f defined in (7.18).

Remark 7.49. — In the above theorem, the assumption that f is an n-times contin-
uously differentiable function guarantees that the integral of the continuous function t 7→
n−1
f (n) (t) (x−t)
(n−1)! exists.

Proof. The proof follows by induction on n and integration by parts.

If n = 1 then f : [a, b] → R is continuously differentiable and, by Corollary 7.5, we get
x x
(x − t)0
Z Z
f (x) = f (x0 ) + f ′ (t) dt = P0 (x) + f (1) (t) dt,
x0 x0 0!

31 as desired.
To explain the idea behind the inductive step, assume first n = 2 (so f is twice continuously
differentiable). Then, in the integral above, we can apply integration by parts to the functions
f ′ (t) and g(t) = t − x. Indeed, since g ′ = 1 and g(x) = 0, we get
Z x
f (x) = f (x0 ) + f ′ (t)g ′ (t) dt
x0
x
Z x
= f (x0 ) + f ′ (t)g(t) x0 − f ′′ (t)g(t) dt

x0
Z x
′
= f (x0 ) + f (x0 )(x − x0 ) + f ′′ (t)(x − t) dt
x0
x
(x − t)1
Z
= P1 (x) + f (2) (t) dt.
x0 1!

This proves the case n = 2.

More generally, assume that the statement of the theorem is true for some n ≥ 1 and that
f : [a, b] → R is an (n + 1)-times continuously differentiable function. Then, by the induction
hypothesis,
n−1
X f (k) (x0 ) Z x
k (x − t)n−1
f (x) = (x − x0 ) + f (n) (t) dt
k! x0 (n − 1)!
k=0

Version: December 30, 2025. 225

 Chapter 7.4 Taylor Series

n (x−t)n−1
for all x ∈ [a, b]. If we set g(t) = − (x−t)
n! , then g (t) =
′
(n−1)! and it follows from integration
by parts that

n−1 Z x
X f (k) (x0 ) k
h
(n)
ix
f (x) = (x − x0 ) + f (t)g(t) − f (n+1) (t)g(t) dt
k! x0 x 0
k=0
n−1
X f (k) (x0 ) Z x
(x − x0 )n (x − t)n
= (x − x0 )k + f (n) (x0 ) + f (n+1) (t) dt
k! n! x 0
n!
k=0
n Z x
X f (k) (x0 ) k (x − t)n
= (x − x0 ) + f (n+1) (t) dt.
k! x0 n!
k=0

This proves the induction step, and hence the result.

Next, we prove the following alternative version of Taylor’s Theorem. Here, we do not
require f (n) to be continuous, but only to exist. Note that in the case n = 1, this result
corresponds to the Mean Value Theorem 5.31.

Theorem 7.50: Taylor Expansion to Order n with Lagrange Remainder

Let n ≥ 1, f : [a, b] → R an n-times differentiable function, and fix x0 ∈ [a, b]. Then,
for all x ∈ [a, b] with x ̸= x0 there exists ξL ∈ (x0 , x) such that

f (n) (ξL )
f (x) = Pn−1 (x) + (x − x0 )n . (7.20)
31 n!

Remark 7.51. — If x = x0 , the formula holds trivially: one may take ξL = x0 , since

f (n) (ξL )
f (x0 ) = Pn−1 (x0 ) and (x0 − x0 )n = 0.
n!

Proof. Fix x ∈ (a, b). Without loss of generality, assume x > x0 (the case x < x0 is analogous)
and consider the function g : (a, b) → R defined as

n−1
f (n−1) (t) X f (k) (t)
g(t) = f (t) + f (1) (t)(x − t) + . . . + (x − t)n−1 = (x − t)k . (7.21)
(n − 1)! k!
k=0

Then g(x) = f (x) and g(x0 ) = Pn−1 (x). Also, its derivative is given by

n−1 n−1
′
X f (k+1) (t) X f (k) (t)
g (t) = (x − t)k − k(x − t)k−1
k! k!
k=0 k=0
n−1 n−1
X f (k+1) (t) X f (k) (t)
= (x − t)k − (x − t)k−1
k! (k − 1)!
k=0 k=1
n−1 n−2
X f (k+1) (t) X f (k+1) (t) f (n) (t)
= (x − t)k − (x − t)k = (x − t)n−1 .
k! k! (n − 1)!
k=0 k=0

Version: December 30, 2025. 226

 Chapter 7.4 Taylor Series

Hence, applying the Cauchy Mean Value Theorem 5.35 in the interval [x0 , x] to the functions
g(t) and h(t) = −(x − t)n , we deduce the existence of a point ξL ∈ (x0 , x) such that

31 f (n) (ξL )
f (x) − Pn−1 (x) g(x) − g(x0 ) g ′ (ξL ) (n−1)! (x − ξL )n−1 f (n) (ξL )
= = = = .
(x − x0 )n h(x) − h(x0 ) h′ (ξL ) n(x − ξL )n−1 n!

This implies (7.20) and concludes the proof.

We can now state our two versions of Taylor’s approximation, using the little-o and the big-O
notation.

Corollary 7.52: Taylor Approximation with Little-o

Let n ≥ 1, f : [a, b] → R an n-times continuously differentiable function, and fix
x0 ∈ [a, b]. Then, for all x ∈ [a, b],

f (x) = Pn (x) + o(|x − x0 |n ) as x → x0 . (7.22)

Proof. Thanks to Theorem 7.48,

x
(x − t)n−1
Z
f (x) = Pn−1 (x) + f (n) (t) dt
x0 (n − 1)!
x x
(x − t)n−1
(x − t)n−1
Z Z
(n)
= Pn−1 (x) + f (x0 ) dt + f (n) (t) − f (n) (x0 ) dt.
x0 (n − 1)! x0 (n − 1)!

Also, using the change of variable s = x − t, we note that

x x−x0
(x − t)n−1 (x − x0 )n (x − x0 )n
Z Z
32 1 1
dt = sn−1 ds = = , (7.23)
x0 (n − 1)! (n − 1)! 0 (n − 1)! n n!

therefore
x
(x − t)n−1 (x − x0 )n
Z
f (n) (x0 ) dt = f (n) (x0 ) .
x0 (n − 1)! n!
Hence, we can write
Z x
(x − x0 )n
(n)

(x − t)n−1
f (x) = Pn−1 (x) + f (x0 ) + f (n) (t) − f (n) (x0 ) dt
n! x0 (n − 1)!
Z x (7.24)

(x − t)n−1
= Pn (x) + f (n) (t) − f (n) (x0 ) dt.
x0 (n − 1)!

Now, given ε > 0, it follows from the continuity of f (n) at x0 that there exists δ > 0 such that
|f (n) (x)−f (n) (x0 )| < ε for all x ∈ (x0 −δ, x0 +δ)∩[a, b]. Therefore, if x ∈ (x0 −δ, x0 +δ)∩[a, b],
we can bound the integrand in the last integral by

(x − t)n−1 |x − t|n−1
f (n) (t) − f (n) (x0 ) ≤ |f (n) (t) − f (n) (x0 )|
(n − 1)! (n − 1)!

Version: December 30, 2025. 227

 Chapter 7.4 Taylor Series

|x − t|n−1
<ε ∀ t ∈ [x0 , x].
(n − 1)!

Hence, using (7.23) again, we get

Z x
(x − t)n−1
|f (x) − Pn (x)| ≤ f (n) (t) − f (n) (x0 ) dt
x0 (n − 1)!
x
|x − t|n−1 |x − x0 |n
Z
<ε dt = ε ,
x0 (n − 1)! n!

which shows that f (x) − Pn (x) = o(|x − x0 |n ) as x → x0 .

Corollary 7.53: Taylor Approximation with Big-O

Let n ≥ 1, f : [a, b] → R an n-times differentiable function, and fix x0 ∈ [a, b]. Assume
that there exists M > 0 such that |f (n) (x)| ≤ M for all x ∈ [a, b]. Then

f (x) = Pn−1 (x) + O(|x − x0 |n ) as x → x0 . (7.25)

Proof. Given x ∈ [a, b], we apply (7.20) to find a point ξL ∈ [x0 , x] such that

f (n) (ξL )
f (x) − Pn−1 (x) = (x − x0 )n .
n!
32 Since |f (n) (ξL )| ≤ M , this implies

M
|f (x) − Pn−1 (x)| ≤ |x − x0 |n ,
n!

therefore f (x) − Pn−1 (x) = O(|x − x0 |n ) as x → x0 , as desired.

Example 7.54. — As mentioned earlier, if f is differentiable at x0 , then

f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + o(|x − x0 |) as x → x0 .

Taylor’s Theorem allows us to obtain more accurate approximations when f has higher regu-
larity.

1. If f is twice differentiable and f ′′ is bounded in a neighborhood of x0 , then Corollary 7.53

(with n = 2) gives

f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + O(|x − x0 |2 ) as x → x0 .

If in addition f ′′ is continuous (equivalently, f ∈ C 2 ), then Corollary 7.52 yields

1
f (x) = f (x0 ) + f ′ (x0 )(x − x0 ) + f ′′ (x0 )(x − x0 )2 + o(|x − x0 |2 ) as x → x0 .
2

Version: December 30, 2025. 228

 Chapter 7.4 Taylor Series

2. If f is smooth, then Corollary 7.52 yields

f (x) = Pn (x) + o(|x − x0 |n ) as x → x0 ,

while Corollary 7.53 applied with n + 1 in place of n gives

f (x) = Pn (x) + O(|x − x0 |n+1 ) as x → x0 .

Hence, in the case when f is smooth, the bound on f − Pn provided by Corollary 7.53
is often more convenient.

Example 7.55. — We illustrate how Taylor expansions can be used to compute limits and
to obtain expansions of more complicated functions.

sin(x) − x
1. Compute the limit lim .
x→0 x3
We consider the Taylor polynomial of sin at 0 of degree 3. Since

sin(0) = 0, sin′ (0) = 1, sin′′ (0) = 0, sin(3) (0) = −1,

the Taylor polynomial of degree 3 at 0 is

0 2 −1 3 x3
32 P3 (x) = 0 + 1 · x + ·x + ·x =x− .
2 6 6

Taylor’s Theorem gives

x3
sin(x) = P3 (x) + O(x4 ) = x − + O(x4 ) as x → 0.
6

We stop the expansion at order 3 because the limit involves x3 in the denominator; any
term of order ≥ 4 will disappear in the limit.
Then  x3  x3
sin(x) − x = x − + O(x4 ) − x = − + O(x4 ),
6 6
and thus 3
sin(x) − x − x6 + O(x4 ) 1
= = − + O(x) as x → 0.
x3 x3 6
Hence
sin(x) − x 1
lim =− .
x→0 x3 6

2. Compute the Taylor expansion of sin2 (x) near 0.

Using the expansion above,

x3
sin(x) = x − + O(x4 ) as x → 0,
6

Version: December 30, 2025. 229

 Chapter 7.4 Taylor Series

we obtain sin2 (x) by squaring:

 x3 2
sin2 (x) = x − + O(x4 )
6
x4 x6 x3
= x2 − + + 2x · O(x4 ) − · O(x4 ) + O(x4 ) · O(x4 ).
3 36 2

We now observe that

2x · O(x4 ) = O(x5 ), x3 · O(x4 ) = O(x7 ), O(x4 ) · O(x4 ) = O(x8 ).

All these terms are of order at least x5 , so they may be merged into a single remainder
6
term O(x5 ). The term x36 can also be absorbed into the remainder for the same reason.
This proves that
x4
sin2 (x) = x2 − + O(x5 ),
3
and the Taylor polynomial of sin2 at 0 of degree 4 is

x4
P4 (x) = x2 − .
3

This illustrates a general strategy: once we know the expansion of a function (here
sin), we can obtain expansions of combinations of that function (here sin2 ) by algebraic
operations on the corresponding Taylor polynomials.
32

Remark 7.56. — In the previous example we showed that

x3
sin(x) = x − + O(x4 ) as x → 0.
6

These coefficients agree with those appearing in the power series expansion of sin. In Sec-
tion 7.4.2 below, we will see that sin is an analytic function, meaning that its Taylor series
around 0 actually converges to sin(x). For such functions, the Taylor polynomial of degree n
is simply the truncation of an infinite series that represents the function itself.

Example 7.57. — We can use the Taylor approximation to refine the discussion in Section
5.2.1. Let f : (a, b) → R be an n-times continuously differentiable function. Suppose x0 ∈
(a, b) satisfies
f ′ (x0 ) = . . . = f (n−1) (x0 ) = 0.

Then the following implications hold:

• If f (n) (x0 ) < 0 and n is even, then f has an isolated local maximum in x0 .

• If f (n) (x0 ) > 0 and n is even, then f has an isolated local minimum in x0 .

• If f (n) (x0 ) ̸= 0 and n is odd, then x0 is not a local extremum of f .

Version: December 30, 2025. 230

 Chapter 7.4 Taylor Series

All three statements follow from (7.20), which, in this case, takes the form

f (n) (ξL )
f (x) = f (x0 ) + (x − x0 )n , ξL ∈ (x0 , x).
n!

Indeed, if f (n) (x0 ) > 0, by continuity there exists δ > 0 such that f (n) (ξL ) > 0 for ξL ∈
(x0 , x) ⊂ (x0 − δ, x0 + δ). If n is even, then (x − x0 )n > 0 for x ̸= x0 and we deduce that
f (x) > f (x0 ) for x ∈ (x0 − δ, x0 + δ) with x ̸= x0 . If n is odd, then (x − x0 )n changes sign
when considering x > x0 and x < x0 , so x0 is not a local extremum of f .
On the other hand, if f (n) (x0 ) < 0 and n is even, the same argument as above shows that
f (x) < f (x0 ) for x ∈ (x0 − δ, x0 + δ) with x ̸= x0 , while in the case n odd x0 is not a local
extremum of f .

7.4.2 Analytic Functions

Motivated by Taylor’s Theorem, one might expect that if f is smooth and we replace the
finite Taylor polynomial Pn by the full Taylor series
∞
X f (k) (x0 )
(x − x0 )k ,
k!
k=0

then this infinite series should converge to f (x). Unfortunately, this is not true in general:
32
only rather special smooth functions enjoy this property.
Note that the Taylor series is centered at x0 instead of 0 (i.e., xn is replaced with (x−x0 )n ).
Hence, all theorems about power series from Section 4.4 still hold, but taking into account
that now x0 plays the role of the center. In particular, if the series has radius of convergence
R > 0, then it converges for all x ∈ (x0 − R, x0 + R), while it diverges for |x − x0 | > R.

Definition 7.58: Analytic Functions

Let I ⊆ R be an interval and x0 ∈ I. A smooth function f : I → R is called analytic at
x0 if the Taylor series of f around x0 has radius of convergence R > 0 and there exists
δ ∈ (0, R) such that
∞
X f (n) (x0 )
f (x) = (x − x0 )n ∀ x ∈ (x0 − δ, x0 + δ) ∩ I.
n!
n=0

We say f is analytic in I if f is analytic at all points in I.

In other words, analytic functions f : I → R are characterized by the fact that, for every
point x0 ∈ I, there exists a power series that converges to f in a neighborhood of x0 .
As the next example shows, there are smooth functions f whose Taylor series converges to
a function different from f .

Version: December 30, 2025. 231

 Chapter 7.4 Taylor Series

Example 7.59. — Consider the function ψ : R → R defined by

(
exp − x1 if x > 0


ψ : x ∈ R 7→ .
0 if x ≤ 0

As shown in Exercise 5.24, ψ is smooth on R and satisfies ψ (n) (0) = 0 for all n ∈ N. Hence,
the Taylor series of the function ψ at the point x0 = 0 is the zero series:
∞ ∞
X ψ (n) (0) X 0 n
xn = x = 0.
n! n!
n=0 n=0

This series has an infinite radius of convergence and converges to the function 0. Since
ψ(x) > 0 holds for all x > 0, the Taylor series does not converge to ψ, and so ψ is not analytic
at the point x0 = 0.

The next result provides a criterion that guarantees that the Taylor series of f converges
to f in a neighborhood of x0 .

Theorem 7.60: A Criterion for Analyticity at x0

Let I ⊆ R be an interval and f : I → R a smooth function. Given x0 ∈ I, assume that

there exist constants r, C0 , A > 0 such that
32
|f (n) (x)| ≤ C0 An n! for all x ∈ (x0 − r, x0 + r) ∩ I, n ∈ N.

Then f is analytic at x0 .

Proof. We first estimate the radius of convergence R of the Taylor series. If we define an =
f (n) (x0 )
, then the Taylor series is equal to ∞
n=0 an (x − x0 ) . Thus, thanks to our assumption
n
P
n!
on the size of |f (n) |, it follows that

C0 An n!
|an | ≤ = C0 An .
n!

This implies that

p p p
lim sup n
|an | ≤ lim sup n C0 An = lim sup n C0 A = A,
n→∞ n→∞ n→∞

therefore, by the definition of radius of convergence (see Definition 4.39), R ≥ A1 .

Now, fix δ < min r, A1 . Given x ∈ (x0 −δ, x0 +δ)∩I, we apply (7.20) and our assumption


on the size of |f (n) | to deduce that

|f (n) (ξL )|
|f (x) − Pn−1 (x)| ≤ |x − x0 |n ≤ C0 An |x − x0 |n ≤ C0 (Aδ)n .
n!

Version: December 30, 2025. 232

 Chapter 7.4 Taylor Series

Since Aδ < 1 (because δ < A ),

1
letting n → ∞ we conclude that

∞
X f (n) (x0 )
f (x) = lim Pn−1 (x) = (x − x0 )n ∀ x ∈ (x0 − δ, x0 + δ) ∩ I,
n→∞ n!
n=0

as desired.

As a direct consequence of Theorem 7.60, we immediately deduce the following:

Corollary 7.61: A Criterion for Analyticity

Let f : [a, b] → R be a smooth function, and assume there exist constants C0 , A > 0
such that
32
|f (n) (x)| ≤ C0 An n! for all x ∈ [a, b], n ∈ N. (7.26)

Then f is analytic on [a, b].

Exercise 7.62. — 1. Show that the functions exp, sin, sinh satisfy the property (7.26)
on any interval [a, b] ⊂ R.

2. Show that the function log satisfies (7.26) on any interval [a, b] ⊂ (0, ∞).

3. Let f, g : [a, b] → R be functions satisfying (7.26). Show that f + g and f · g also satisfy
this property (possibly with different constants C0 and A).

Version: December 30, 2025. 233

 Chapter 8

Ordinary Differential Equations

Setting up and solving differential equations is one of the main practical uses of calculus. Such
equations are used in physics, chemistry, biology, and many other sciences. They are also
central in areas like structural engineering, modern economics, and information technology.

8.1 Ordinary Differential Equations (ODEs)

In this section we study ordinary differential equations (ODEs). They describe how an un-
known quantity depends on one real variable (often denoted by x or t) and how this dependence
is constrained by relations involving its derivatives.
It is convenient to fix some notation from the beginning:

• we usually denote the unknown function by u;

• the independent variable is denoted by x (or by t when it represents time);

• letters like f , g, a0 , a1 will typically denote given functions or constants appearing in

33 the equation (data of the problem).

Although derivatives are usually denoted using ′ (so u′ , u′′ , etc.), it is common to use a
dot to denote derivatives with respect to time (so u̇, ü, etc.).

Definition 8.1: ODEs

An ordinary differential equation (ODE) is a relation involving a function u : R →
R of a real variable x ∈ R and its derivatives. The general form of an n-th order ODE
is
G x, u(x), u′ (x), u′′ (x), . . . , u(n) (x) = 0, (8.1)

where G : Rn+2 → R is a given function.

234
 Chapter 8.1

In many examples the independent variable is time t and the equation describes the evo-
lution of a system, but we keep the generic notation x unless we want to stress the time
interpretation.
ODEs can be classified according to several criteria:

1. Order: An ODE is of order n if u(n) is the highest derivative appearing in the equation.
For instance:

(a) u′′ + u = 0 ⇝ second order.

(b) u(3) = x2 u + x ⇝ third order.
(c) (u′ )2 + u − x3 = 0 ⇝ first order.

2. Linearity: An ODE is linear if it is linear in u and its derivatives. Otherwise, it is

nonlinear. Here “linear” means that u, u′ , u′′ , . . . appear only to the first power and are
not multiplied with each other.

(a) u′′ + u = 0 ⇝ linear.

(b) u′′ + u2 = 0 ⇝ nonlinear (because of u2 ).
(c) u′′ + u′ u = 0 ⇝ nonlinear (because of the product u′ u).
(d) u(3) = x2 u + x ⇝ linear.
33 (e) (u′ )2 + u − x3 = 0 ⇝ nonlinear (because of (u′ )2 ).

3. Homogeneity (for linear ODEs): For a linear ODE, we say it is homogeneous if all terms
involve the function or its derivatives. Equivalently, if u is a solution then Au is a
solution for all A ∈ R. If there is an additional term that does not depend on u (a
“forcing term”), the equation is non-homogeneous.

(a) u′′ + u = 0 ⇝ homogeneous.

(b) u(3) = x2 u + x ⇝ non-homogeneous (because of the term +x).
(c) u(3) = x2 u ⇝ homogeneous.

Example 8.2. — We now present some classic examples of ODEs and their applications.
In each case we indicate what plays the role of the unknown u.

1. Newton’s Law of Cooling: This law states that

“The rate of heat loss of a body is proportional to the difference between its
temperature and the temperature of the surrounding environment.”

If T (t) denotes the temperature of the object at time t, the law gives


Ṫ (t) = −k T (t) − Tenv ,

Version: December 30, 2025. 235

 Chapter 8.1

where:

• Tenv is the ambient temperature (assumed constant);

• k > 0 is a constant describing the heat transfer.

Here the unknown function is u(t) = T (t), and the given data are k and Tenv . This is a
linear, non-homogeneous, first-order ODE.

2. Harmonic Oscillator: Consider a mass attached to a spring. Let x(t) denote its dis-
placement from equilibrium. The restoring force is proportional to x:

Fspring = −kx,

with k > 0 (Hooke’s law). Newton’s law mẍ = F gives

mẍ(t) = −kx(t),

or equivalently
ẍ(t) + ω 2 x(t) = 0,
q
where ω = m k
> 0 is the angular frequency. Here u(t) = x(t) is the unknown, and ω
33 is a given parameter. This is linear, homogeneous, and second order.
If we also take into account friction proportional to the velocity ẋ, we obtain the damped
harmonic oscillator:
ẍ(t) + 2ζω ẋ(t) + ω 2 x(t) = 0,

where ζ ≥ 0 is the damping ratio. This equation models many vibrating systems in
physics and engineering.

3. Logistic Population Growth: In population dynamics, a common model for a population

with limited resources is the logistic equation
 
P (t)
Ṗ (t) = rP (t) 1 − ,
K

where:

• P (t) is the population at time t;

• r > 0 is the intrinsic growth rate;
• K > 0 is the carrying capacity (maximum sustainable population).

Here the role of u is played by P , while r and K are given parameters. This is a nonlinear
first-order ODE.

Version: December 30, 2025. 236

 Chapter 8.1

4. Bessel Equation: The Bessel equation with parameter α ∈ R is

x2 u′′ (x) + xu′ (x) + (x2 − α2 )u(x) = 0.

This is linear, homogeneous, and second order. The unknown is u(x), and α is a given
parameter. Its solutions are the Bessel functions, which appear, for instance, in problems
of heat conduction and wave propagation in cylindrical geometries, and in quantum
mechanics.

5. Airy Equation: The Airy equation is

u′′ (x) − α2 xu(x) = 0,

where α > 0 is a given constant. The corresponding solutions are called Airy functions.
They arise, for example, in quantum mechanics when studying a particle in a linear or
triangular potential.

So far, we have only considered single equations, but one can also study systems of ODEs
with several unknown functions u1 , . . . , un . We will not go into this now, but many ideas are
similar.
33
In addition, solutions are often required to satisfy extra conditions such as u(0) = 0
(prescribed position at time 0) and/or u′ (0) = 1 (prescribed velocity at time 0). When these
conditions are imposed at a single time (typically t = 0), they are called initial conditions.
More general conditions (for instance at two different points, such as u(0) = 0 and u(1) = 1)
are called boundary conditions.
Later we shall see that, under suitable assumptions on the data (for example on a function
f appearing in the equation), prescribing initial conditions often leads to a unique solution.
This is the content of the Cauchy–Lipschitz (or Picard–Lindelöf) theorem.

8.1.1 Linear First Order ODEs

We now consider linear first-order ODEs. Throughout this subsection we fix a non-empty
interval I ⊆ R that is not a single point, and we study equations of the form

u′ (x) + f (x)u(x) = g(x),

where f and g are given continuous functions on I, and u is the unknown.

We start with the homogeneous case g ≡ 0.

Version: December 30, 2025. 237

 Chapter 8.1

Theorem 8.3: Homogeneous Linear 1st Order ODEs

Let f : I → R be continuous and consider the homogeneous first-order linear ODE

u′ (x) + f (x)u(x) = 0 ∀ x ∈ I. (8.2)

Let F : I → R be a primitive of f . Then all C 1 solutions u : I → R of (8.2) are of the

form
u(x) = Ae−F (x) , A ∈ R.

In other words, the set of solutions of (8.2) forms a one-dimensional linear subspace of
C 1 (I).

Proof. Given A ∈ R, define u(x) = Ae−F (x) . Then

u′ (x) = −F ′ (x)Ae−F (x) = −f (x)Ae−F (x) = −f (x)u(x) ∀ x ∈ I,

so u solves the ODE.

Conversely, let u ∈ C 1 (I) solve (8.2) and set v(x) = eF (x) u(x). Then

′
v ′ (x) = eF (x) u(x) + eF (x) u′ (x) = eF (x) f (x)u(x) − f (x)u(x) = 0


∀ x ∈ I.
33
By Corollary 5.48, we deduce that v(x) = A for some A ∈ R, hence u(x) = Ae−F (x) .

Remark 8.4. — In the previous result, solutions are written using a primitive F of f .
Since primitives are defined up to an additive constant, we can replace F by F + C for any
C ∈ R. This amounts to replacing Ae−F (x) by Ae−C e−F (x) , and since A ∈ R is arbitrary, this
does not change the set of solutions.

Example 8.5. — 1. Constant–coefficient homogeneous equation. Consider the differential

equation
u′ (x) + 3u(x) = 0.

This is of the form u′ (x) + f (x)u(x) = 0 with f (x) = 3, and a primitive of f is F (x) = 3x.
By Theorem 8.3, all solutions are of the form

u(x) = Ae−F (x) = Ae−3x , A ∈ R.

2. Variable–coefficient homogeneous equation. Consider the differential equation

2
u′ (x) − u(x) = 0, x ∈ (0, ∞).
x

Version: December 30, 2025. 238

 Chapter 8.1

Here f (x) = − x2 , so a primitive is F (x) = −2 log x = − log(x2 ). Therefore

2)
u(x) = Aelog(x = Ax2 , A ∈ R.

3. Trigonometric coefficient. Consider the differential equation

u′ (x) + sin x u(x) = 0.

Here f (x) = sin x, so a primitive is F (x) = − cos x. Hence

y(x) = Ae−F (x) = Aecos x , A ∈ R.

Next, we consider the non-homogeneous linear first-order ODE

u′ (x) + f (x)u(x) = g(x) ∀ x ∈ I, (8.3)

where f, g : I → R are given continuous functions.

To motivate the solution formula, we use the method of variation of constants. For the
homogeneous equation we know that every solution has the form Ae−F (x) with A ∈ R. The
idea is to replace this constant by a function and look for a solution of the non-homogeneous
equation of the form
33 u(x) = H(x)e−F (x) ,

for some function H ∈ C 1 (I). With this choice,

u′ (x) = H ′ (x)e−F (x) − F ′ (x)H(x)e−F (x) = H ′ (x)e−F (x) − f (x)u(x).

Thus u solves (8.3) if and only if

H ′ (x)e−F (x) = g(x),

which means that H must be a primitive of the function g(x)eF (x) .

This motivates the general solution formula:

Theorem 8.6: Non-Homogeneous Linear 1st Order ODEs

Let f, g : I → R be continuous, and consider the non-homogeneous first-order linear
ODE (8.3). Let F : I → R be a primitive of f , and let H : I → R be a primitive of
geF . Then every C 1 solution u : I → R of (8.3) is of the form

u(x) = H(x)e−F (x) + Ae−F (x) , A ∈ R.

In particular, the set of solutions of (8.3) is a one-dimensional affine subspace of C 1 (I).

Version: December 30, 2025. 239

 Chapter 8.1

Proof. If H is a primitive of geF , then H + A is also a primitive for any constant A. Hence,
by the same computation as the one performed above, it follows that

u(x) = (H(x) + A)e−F (x)

solves (8.3). Indeed

u′ (x) = (H(x) + A)′ e−F (x) − F ′ (x)(H(x) + A)e−F (x)

= H ′ (x)e−F (x) − f (x)u(x) = g(x) − f (x)u(x).

Conversely, let u be any solution of (8.3), and set v(x) = u(x) − H(x)e−F (x) . Then

v ′ (x) = u′ (x) − H ′ (x)e−F (x) − F ′ (x)H(x)e−F (x)

= −f (x)u(x) + g(x) − g(x)eF (x) e−F (x) + f (x)H(x)e−F (x)
= −f (x)u(x) + g(x) − g(x) + f (x)(u(x) − v(x)) = − f (x) v(x).

Thus v solves the homogeneous equation (8.2). By Proposition 8.3, we have v(x) = Ae−F (x)
for some constant A. Therefore,

u(x) = v(x) + H(x)e−F (x) = Ae−F (x) + H(x)e−F (x) ,

which proves the result.

33
The previous results give explicit formulas to solve every linear first-order ODE of the form
u′+ f u = g. In concrete situations, the difficulty lies in computing a primitive F of f and
then a primitive of g(x)eF (x) .
As we have seen, solutions depend on a free parameter A ∈ R. This allows us to impose
an initial condition of the form u(x0 ) = u0 , which uniquely determines A.

Example 8.7. — We solve the ODE

2
u′ (x) − 2xu(x) = ex , u(0) = 1, (8.4)

2
on R. Here u is the unknown, and f (x) = −2x, g(x) = ex are given. According to Theo-
rem 8.6, we first find a primitive of f :

F (x) = −x2 .

Then we consider
2 2
g(x)eF (x) = ex e−x = 1,

whose primitive is H(x) = x. Thus u must be of the form

2
u(x) = (x + A)ex .

Version: December 30, 2025. 240

 Chapter 8.1

Imposing the condition u(0) = 1 gives A = 1, hence

2
u(x) = (x + 1)ex . (8.5)

Remark 8.8. — If one forgets the formula from Theorem 8.6, it is enough to remember
the following procedure for solving (8.3). We start from

u′ (x) + f (x)u(x) = g(x),

and multiply both sides by a function ew(x) :

u′ (x)ew(x) + f (x)u(x)ew(x) = g(x)ew(x) .

We look for w such that the left-hand side is the derivative of u(x)ew(x) , that is

′
u(x)ew(x) = u′ (x)ew(x) + w′ (x)u(x)ew(x) .

So we require
w′ (x) = f (x).

If we choose w = F to be any primitive of f (the additive constant does not matter), then

′
33 u(x)eF (x) = g(x)eF (x) ,

and therefore Z
F (x)
u(x)e = geF + A,

for some A ∈ R. Thus, if H is a primitive of geF , this reads

u(x)eF (x) = H(x) + A =⇒ u(x) = H(x)e−F (x) + Ae−F (x) .

Example 8.9. — We revisit Example 8.7: we want to solve the ODE

2
u′ (x) − 2xu(x) = ex , u(0) = 1,

using only the integrating-factor method described in Remark 8.8.

We multiply both sides by ew(x) and choose w such that

′
u(x)ew(x) = u′ (x)ew(x) + w′ (x)u(x)ew(x)

reproduces the left-hand side of the equation. Thus we require w′ (x) = f (x) = −2x, and
2
therefore we may take w(x) = −x2 . Multiplying the ODE by e−x gives

2
′ 2 2 2
u(x)e−x = e−x u′ (x) − 2xu(x) = e−x ex = 1.

Version: December 30, 2025. 241

 Chapter 8.1

Integrating, we get

2 2
u(x)e−x = x + A =⇒ u(x) = (x + A)ex .

Imposing u(0) = 1 gives A = 1, so

2
33 u(x) = (x + 1)ex .

Exercise 8.10. — Find a solution of

u′ (x) − 4
+ 1 u(x) = x4 ,


x u(1) = 1,

in the interval (0, ∞).

8.1.2 Autonomous First Order ODEs

We next study autonomous first-order ODEs, where the rate of change of the unknown depends
only on its current value, and not explicitly on x:

u′ (x) = f u(x) , (8.6)

where f : R → R is a given continuous function, and u is the unknown. The function f tells
us how u should change depending on its current value.
A standard way to solve such equations is the method of separation of variables. If
u(x) = C ∈ R for some C such that f (C) = 0, then the constant function u = C is a solution
of the ODE. Otherwise, if f (u(x)) ̸= 0, we can divide both sides by f (u(x)):

34 u′ (x)
= 1.
f (u(x))

Integrating and using the substitution formula (7.8), we obtain

Z Z Z
1 1
du = u′ (x) dx = 1 dx = x + A, (8.7)
f (u) f (u(x))

where A is a constant of integration.

If H is a primitive of 1/f , this reads

H(u(x)) = x + A =⇒ u(x) = H −1 (x + A).

Since we assumed f (u(x)) ̸= 0 on the interval under consideration, we have H ′ = 1

f ̸= 0 there,
so H is strictly monotone and therefore invertible on that interval.

Version: December 30, 2025. 242

 Chapter 8.1

Example 8.11. — We study the logistic growth model, treating it as an autonomous

first-order ODE and solving it explicitly.
Consider  
u(x)
u′ (x) = r u(x) 1 − , (8.8)
K
where r > 0 is the growth rate and K > 0 is the carrying capacity. Here u(x) denotes the
population at “time” x. From the modelling point of view one is usually interested in solutions
with
0 ≤ u(x) ≤ K,

since u < 0 has no meaning as a population, and K represents an upper bound (the maximal
sustainable population).
Mathematically, the right-hand side vanishes at u = 0 and u = K, so the constant functions
u(x) = 0 and u(x) = K are solutions. To solve the ODE by separation of variables we assume
0 < u(x) < K on the interval under consideration (so that we may safely divide by u and
K − u).
We rearrange and integrate:

K u′ (x)
Z Z
K

=r =⇒ du = r dx = rx + A.
u(x) K − u(x) u(K − u)

34 We decompose the rational function

K 1 1
= + ,
u(K − u) u K −u

so that
Z Z Z  
K 1 1 u
du = du + du = log u − log(K − u) = log .
u(K − u) u K −u K −u

Thus
 
u(x) u(x)
log = rx + A =⇒ = erx+A
K − u(x) K − u(x)
=⇒ u(x) = Kerx+A − u(x)erx+A
1 + erx+A u(x) = Kerx+A


=⇒
erx+A
=⇒ u(x) = K .
1 + erx+A

If u0 ∈ (0, K) denotes the initial population at x = 0, then

eA u0
u0 = K =⇒ eA = .
1 + eA K − u0

Version: December 30, 2025. 243

 Chapter 8.1

Substituting back gives the explicit solution

Ku0
u(x) =
.
u0 + K − u0 e−rx

It is useful to briefly discuss the different initial conditions:

• If u(0) = 0, then u(x) = 0 is a (constant) solution. In other words, if the population is

zero at the initial time, then it remains zero for all times.

• If u(0) = K, then u(x) = K is a (constant) solution. In this case the population stays
exactly at the carrying capacity.

• If 0 < u(0) < K, the above formula applies and one checks that 0 < u(x) < K for all x,
with u(x) → K as x → ∞. The population grows and approaches the carrying capacity
from below.

• If u(0) > K, then u′ (0) < 0, so u decreases. One can check that the same explicit
formula holds also in this case (with u0 > K). In particular, u(x) > K for all x, and
u(x) → K as x → ∞: the population decreases towards the carrying capacity from
34
above.

The method of separation of variables applies also to equations of the form

u′ (x) = f (u(x)) g(x),

where f, g : R → R are continuous. Assuming f (u(x)) ̸= 0 on a suitable interval, we divide:

u′ (x)
= g(x).
f (u(x))

Integrating, Z Z
1
du = g(x) dx + A.
f (u)
If H is a primitive of 1/f and G a primitive of g, then

u(x) = H −1 G(x) + A .



H u(x) = G(x) + A =⇒

We summarize this discussion in the next:

Version: December 30, 2025. 244

 Chapter 8.1

Theorem 8.12: Separable Equations

Let I, J ⊆ R be intervals, let f : J → R and g : I → R be continuous, and assume

f (y) ̸= 0 for all y ∈ J. Let H be a primitive of 1/f on J, and let G be a primitive of
g on I. Then every C 1 solutions u : I → J of

u′ (x) = f u(x) g(x) (8.9)

is of the form
u(x) = H −1 G(x) + A ,


A ∈ R.

Proof. Since f never vanishes, H ′ = 1/f has constant sign. Hence H is strictly monotone and
therefore invertible.
If u(x) = H −1 (G(x) + A), then by the chain rule (Theorem 5.14) and the formula for the
derivative of the inverse function (Theorem 5.20), we get

1 1
u′ (x) = (H −1 )′ (G(x)+A) G′ (x) = g(x) = g(x) = f (u(x)) g(x),
H′◦ H −1 (G(x) + A) H ′ (u(x))

so u solves (8.9)
Conversely, suppose u solves (8.9). Then

′ 1
H ◦ u(x) = H ′ (u(x)) u′ (x) = f (u(x))g(x) = g(x) = G′ (x).
34 f (u(x))

Thus H(u(x)) − G(x) has derivative zero, and is therefore constant:

H(u(x)) = G(x) + A for some A ∈ R.

Applying H −1 gives
u(x) = H −1 G(x) + A ,

which completes the proof.

Example 8.13. — We solve the separable equation

u′ (x) = x u(x)2

on an interval where u does not vanish. We rewrite the equation by separating variables:

u′
= x.
u2

Integrating both sides gives

x2
Z Z
du 1
= x dx, =⇒ − = + A, A ∈ R.
u2 u 2

Version: December 30, 2025. 245

 Chapter 8.1

Solving for u yields the family of solutions

1
u(x) = − x2 , A ∈ R.
2 +A

If an initial condition u(x0 ) = u0 (with u0 =

̸ 0) is prescribed, we determine the constant A
by inserting x = x0 :
1 x2 1 x2
− = 0 + A =⇒ A = − − 0 .
u0 2 u0 2
Thus the unique solution with u(x0 ) = u0 is

1
u(x) = − x20
.
x2 1
2 − u0 − 2

8.1.3 Homogeneous Linear Second Order ODEs with Constant Coefficients

We now move to second-order linear ODEs. These are considerably more difficult to solve
than first-order ones in general, so we start with the simplest case: homogeneous equations
with constant coefficients,
u′′ (x) + a1 u′ (x) + a0 u(x) = 0, (8.10)

where a0 , a1 ∈ R are given constants and u is the unknown. Such equations already cover
34
many important applications (for instance, oscillations and damped vibrations).

Example 8.14. — • For u′′ = 0, affine functions are solutions: u(x) = Ax + B, with
A, B ∈ R.

• For u′′ − u = 0, ex and e−x are solutions, therefore

u(x) = Aex + Be−x , A, B ∈ R,

is a solution.

• For u′′ + u = 0, sin and cos are solutions, therefore

u(x) = A sin(x) + B cos(x), A, B ∈ R.

is a solution. Since sine and cosine can be written in terms of e±ix , one can also rewrite
the solutions above as

u(x) = Ceix + De−ix , C, D ∈ C,

and then re-express them in terms of real-valued sine and cosine (recall that we are
interested in real-valued functions).

Version: December 30, 2025. 246

 Chapter 8.1

Exercise 8.15. — Check the assertions in Example 8.14.

The last two examples suggest looking for solutions of (8.10) of the form

u(x) = eαx , α ∈ C.

With this choice,

u′′ (x) + a1 u′ (x) + a0 u(x) = α2 + a1 α + a0 u(x),

so u is a solution if and only if

α2 + a1 α + a0 = 0.

The quadratic polynomial

p(t) = t2 + a1 t + a0

is called the characteristic polynomial. Its roots determine the shape of the solutions. We
distinguish three cases according to the discriminant ∆ = a21 − 4a0 .
• Case 1: ∆ > 0. The polynomial p(t) has two distinct real roots
√ √
−a1 + ∆ −a1 − ∆
α= , β= . (8.11)
2 2

Then x 7→ eαx and x 7→ eβx are two linearly independent solutions, and therefore
34
u(x) = Aeαx + Beβx , A, B ∈ R,

is a solution of (8.10).
• Case 2: ∆ < 0. Then p(t) has two complex-conjugate roots
√ √
a1 −∆ a1 −∆
α + iβ = − + i , α − iβ = − − i , (8.12)
2 2 2 2

with β > 0. The complex-valued functions x 7→ e(α±iβ)x solve (8.10), and hence their real and
imaginary parts are real solutions. This gives

u(x) = Aeαx sin(βx) + Beαx cos(βx), A, B ∈ R.

• Case 3: ∆ = 0. Then p(t) has a double real root

a1
α=− , (8.13)
2

so x 7→ eαx is a solution of (8.10). To find another independent solution, recall the special
case u′′ = 0, where two linearly independent solutions are 1 and x, which can be written as
eγx and xeγx with γ = 0.

Version: December 30, 2025. 247

 Chapter 8.1

This suggests that x 7→ xeαx might be a solution. Indeed,

′′
′
xeαx + a1 xeαx + a0 xeαx = α2 + a1 α + a0 xeαx + (2α + a1 )eαx = 0,


| {z } | {z }
=0 =0

where the first term vanishes because α is a root of p, and the second vanishes by (8.13).
34 Hence
u(x) = Aeαx + Bxeαx , A, B ∈ R,

solves (8.10).
For second-order ODEs it is customary to prescribe both the value of u and the value of
its derivative at some point (for instance, u(0) = 1 and u′ (0) = 0). The two constants A, B
in the formulas above are precisely what we need in order to satisfy two such conditions.

We now want to prove that the families of solutions described in Paragraph 8.1.3 indeed give
all solutions of (8.10).

Theorem 8.16: Existence and Uniqueness: the Homogeneous Case

Given a0 , a1 ∈ R, let ∆ = a21 − 4a0 and consider the following solutions of (8.10):

∆>0: u1 (x) = eαx , u2 (x) = eβx , α, β as in (8.11),

∆<0: u1 (x) = eαx sin(βx), u2 (x) = eαx cos(βx), α, β as in (8.12),
∆=0: u1 (x) = eαx , u2 (x) = xeαx , α as in (8.13).

If u ∈ C 2 (I) solves (8.10), then there exist A, B ∈ R such that

u = Au1 + Bu2 .

35 In other words, the set of solutions of (8.10) forms a two-dimensional linear subspace
of C 2 (I).

Proof. We treat the case ∆ > 0 (the other cases are similar). Assume for simplicity that 0 ∈ I
(otherwise, fix x0 ∈ I and repeat the argument with x0 in place of 0).
Observing that
u1 (0) = u2 (0) = 1, u′1 (0) = α > β = u′2 (0),

by defining
αu2 (x) − βu1 (x) u1 (x) − u2 (x)
v1 (x) = , v2 (x) = ,
α−β α−β
it follows that v1 and v2 are solutions of (8.10) that satisfy

v1 (0) = 1, v1′ (0) = 0, v2 (0) = 0, v2′ (0) = 1.

Version: December 30, 2025. 248

 Chapter 8.1

Now, given u ∈ C 2 (I) solving (8.10), set

w(x) = u(x) − u(0)v1 (x).

Then, w is a solution satisfying w(0) = 0. Our goal is to show that w is a multiple of u2 . For
this, consider the function

W (x) = w(x)v2′ (x) − w′ (x)v2 (x),

called the Wronskian of w and v2 , and compute W ′ (x). Differentiating gives

W ′ (x) = w′ (x)v2′ (x) + w(x)v2′′ (x) − w′′ (x)v2 (x) − w′ (x)v2′ (x) = w(x)v2′′ (x) − w′′ (x)v2 (x).

Since both w and v2 solve (8.10), we have

w′′ = −a1 w′ − a0 w, v2′′ = −a1 v2′ − a0 v2 .

Substituting into the expression for W ′ gives

W ′ (x) = w(−a1 v2′ − a0 v2 ) − (−a1 w′ − a0 w) v2

= −a1 wv2′ − a0 wv2 + a1 w′ v2 + a0 wv2

35 = a1 (w′ v2 − wv2′ ) = −a1 W (x).

Thus W satisfies the first-order linear ODE

W ′ (x) + a W (x) = 0, W (0) = 0.

By Theorem 8.3 W (x) = Ae−a1 x , and the condition W (0) = 0 implies that W = 0 on I. In
other words,
w(x)v2′ (x) − w′ (x)v2 (x) = 0 ∀ x ∈ I.

Now, on any interval J where v2 does not vanish, we can rewrite this as

v2′ (x)
w′ (x) = w(x) ∀ x ∈ I ∩ J.
v2 (x)

v′
Thus, since v22 = (log |v2 |)′ , we can apply Theorem 8.3 to deduce the existence of a constant
A ∈ R such that
w(x) = A elog |v2 (x)| = A |v2 (x)| ∀ x ∈ I ∩ J.

Since |v2 | is either equal to v2 or to −v2 on J (because v2 does not vanish), we proved that
on each interval J where v2 does not vanish, there exists a constant AJ ∈ R such that

w(x) = AJ v2 (x).

Version: December 30, 2025. 249

 Chapter 8.1

In our situation, one can check that v2 vanishes only at 0, so there exist A− , A+ ∈ R such
that
w = A− v2 on (−∞, 0) ∩ I, w = A+ v2 on (0, ∞) ∩ I.

If either (−∞, 0) ∩ I or (0, ∞) ∩ I are empty, we have proved that w is a multiple of v2 on I.

Otherwise we observe that, by the continuity of w and v2 , recalling that v2 (0) = 0 we get

w(0) = lim w(x) = lim A− v2 (x) = A− v2 (0) = 0,

x→0− x→0−

therefore,
w = A − v2 on (−∞, 0] ∩ I, w = A+ v2 on [0, ∞) ∩ I.

Thus, recalling that v2′ (0) = 1,

′ w(−h) − w(0) A− v2 (−h) − A− v2 (0)

w− (0) = lim = lim = A− (v2 )′− (0) = A− ,
h→0+ h h→0 + h

and analogously w+
′ (0) = A v ′ (0) = A . Since w ∈ C 2 (I) we have that w ′ (0) = w ′ (0), so
+ 2 + − +
A− = A+ , and therefore w(x) = A+ v2 (x) on all of I.
In conclusion

αu2 (x) − βu1 (x) u1 (x) − u2 (x)

u(x) = u(0)v1 (x) + A+ v2 (x) = u(0) + A+ ,
35 α−β α−β

which is a linear combination of u1 and u2 , as claimed.

Remark 8.17 (Zero initial data). — From the explicit formulas in Theorem 8.16 one can
easily check that the only solution u ∈ C 2 (I) of the homogeneous equation u′′ +a1 u′ +a0 u = 0
satisfying u(x0 ) = u′ (x0 ) = 0 for some x0 ∈ I is the zero solution.
For instance, in the case ∆ > 0 we have

u(x) = Aeαx + Beβx , α ̸= β,

so 
Aeαx0 + Beβx0 = 0,
u(x0 ) = 0, u′ (x0 ) = 0 =⇒
Aαeαx0 + Bβeβx0 = 0.

Multiplying the first equation by α and subtracting from the second gives (β − α) B eβx0 = 0,
therefore B = 0. Going back to the first equation, we also get A = 0.
The cases ∆ = 0 and ∆ < 0 are handled similarly. In conclusion, no nontrivial solution
can vanish together with its derivative at a point.

As a consequence of the previous remark, we obtain the following.

Version: December 30, 2025. 250

 Chapter 8.1

Corollary 8.18: Wronskian and linear dependence

Let v1 , v2 ∈ C 2 (I) be solutions of (8.10) on an interval I ⊆ R, and let

W (x) = v1 (x)v2′ (x) − v1′ (x)v2 (x)

be their Wronskian. If W (x0 ) = 0 for some x0 ∈ I, then v1 and v2 are linearly dependent
on I.

Proof. Consider the 2 × 2 matrix

!
v1 (x0 ) v2 (x0 )
M (x0 ) = .
v1′ (x0 ) v2′ (x0 )

The condition W (x0 ) = 0 means precisely that det M (x0 ) = 0, so there exists a nonzero
vector (A, B) ∈ R2 such that

A v1 (x0 ) + B v2 (x0 ) = 0, A v1′ (x0 ) + B v2′ (x0 ) = 0.

35 Define
w(x) = A v1 (x) + B v2 (x).

Then w solves (8.10) and satisfies

w(x0 ) = 0, w′ (x0 ) = 0.

By Remark 8.17, the only solution with these initial data is the trivial one, so w = 0 on I.
This proves that A v1 + B v2 = 0, thus v1 and v2 are linearly dependent on I.

Example 8.19. — We consider a vertical spring whose upper end is attached to a fixed
ceiling, a weight of mass m attached to its lower end, and the mass can move only in the
vertical direction. There is a unique position where the spring force balances gravity; we call
this the equilibrium position and denote it by u = 0. If x denotes time, we write u(x) for
the vertical displacement of the mass from this equilibrium position at time x (with upward
displacements taken as positive).
By Newton’s law, the mass m multiplied by the acceleration u′′ (x) equals the total force
acting on the mass.

Version: December 30, 2025. 251

 Chapter 8.1

u(x) x

A first contribution to the total force comes from the spring: according to Hooke’s law, the
restoring force is −ku, where k > 0 is the spring constant. In addition, friction acts against
the motion; we assume this friction force is proportional to the velocity and given by −du′ ,
where d ≥ 0 is the damping constant.
The equation of motion is therefore

mu′′ = −du′ − ku,

or equivalently
d ′ k
u′′ + u + u = 0.
m m
This is a homogeneous linear ODE of second order with constant coefficients, with unknown
35
u and given constants m, d, k. Setting
r
k d
ω= , ζ= ,
m 2mω

we can rewrite it as
u′′ + 2ζωu′ + ω 2 u = 0

(see also Example 8.2(2)). The characteristic polynomial is

p(t) = t2 + 2ζωt + ω 2

with discriminant ∆ = 4(ζ 2 − 1)ω 2 .

If ∆ < 0 (equivalently ζ < 1), the friction is small compared to the spring strength and
the system oscillates:
p
u(x) = e−ζωx A sin(γx) + B cos(γx) ,


γ= 1 − ζ 2 ω.

The constants A, B are determined by the initial position u(0) and the initial velocity u′ (0).
In the special case ζ = 0 there is no damping, and u is periodic.
If friction is large compared to the spring strength (ζ ≥ 1, so ∆ ≥ 0), the oscillations
disappear, and the weight returns monotonically to equilibrium:

Version: December 30, 2025. 252

 Chapter 8.1

• If ζ > 1, then
 p 
u(x) = Ae−λ1 x + Be−λ2 x , λ1,2 = ζ ± ζ 2 − 1 ω.

• If ζ = 1, then
u(x) = Ae−ωx + Bxe−ωx .

One can check that ζ − ζ 2 − 1 < 1 for all ζ > 1, so the fastest exponential convergence
p

is achieved when ζ = 1. This type of behavior is desirable, for example, in door-closing

mechanisms. Here again the constants A, B are fixed by prescribing suitable initial conditions.

8.1.4 Non-Homogeneous Linear Second Order ODEs with Constant Coef-

ficients
We now add a forcing term and consider the non-homogeneous linear second-order ODE with
constant coefficients
u′′ (x) + a1 u′ (x) + a0 u(x) = g(x), (8.14)

where a0 , a1 ∈ R are constants and g ∈ C 0 (I) is a given function. The unknown is again u.
With the notation of Paragraph 8.1.3, let u1 , u2 be two linearly independent solutions of
the homogeneous equation (8.10):
35
∆>0: u1 (x) = eαx , u2 (x) = eβx , α, β as in (8.11),
∆<0: u1 (x) = eαx sin(βx), u2 (x) = eαx cos(βx), α, β as in (8.12),
∆=0: u1 (x) = eαx , u2 (x) = xeαx , α as in (8.13).

We look for a solution u of (8.14) of the form

u(x) = H1 (x)u1 (x) + H2 (x)u2 (x),

where H1 , H2 are unknown functions. This is the method of variation of constants in the
second-order setting: we know the solutions of the homogeneous equation, and we allow the
coefficients in the linear combination to depend on x.
We compute
u′ = (H1′ u1 + H2′ u2 ) + (H1 u′1 + H2 u′2 ),

and
u′′ = (H1′ u1 + H2′ u2 )′ + (H1′ u′1 + H2′ u′2 ) + (H1 u′′1 + H2 u′′2 ).

Hence

u′′ + a1 u′ + a0 u = (H1′ u1 + H2′ u2 )′ + (H1′ u′1 + H2′ u′2 ) + (H1 u′′1 + H2 u′′2 )
+ a1 (H1′ u1 + H2′ u2 ) + a1 (H1 u′1 + H2 u′2 ) + a0 (H1 u1 + H2 u2 ).

Version: December 30, 2025. 253

 Chapter 8.1

Since u1 and u2 solve the homogeneous equation, we have

H1 (u′′1 + a1 u′1 + a0 u1 ) = 0, H2 (u′′2 + a1 u′2 + a0 u2 ) = 0,

therefore

u′′ + a1 u′ + a0 u = (H1′ u1 + H2′ u2 )′ + (H1′ u′1 + H2′ u′2 ) + a1 (H1′ u1 + H2′ u2 ).

For u to solve the non-homogeneous ODE (8.14), we can choose H1′ and H2′ so that

H1′ (x)u1 (x) + H2′ (x)u2 (x) = 0, H1′ (x)u′1 (x) + H2′ (x)u′2 (x) = g(x). (8.15)

From the first equation, we can write

u1 ′
H2′ = − H . (8.16)
u2 1

Substituting into the second equation gives

u1 ′ ′ u2 g
H1′ u′1 − u H =g =⇒ H1′ = .
u2 2 1 u′1 u2 − u′2 u1

Inserting this into (8.16) yields

u1 u2 g u1 g
35 H2′ = − = ′ .
u2 u′1 u2 − u′2 u1 u2 u1 − u′1 u2

Note that the denominator

u′1 u2 − u′2 u1 = − u1 u′2 − u′1 u2

is (up to sign) the Wronskian of u1 and u2 . Since u1 and u2 are linearly independent solutions
of the homogeneous equation, Corollary 8.18 implies that this Wronskian never vanishes on
I. In particular, the above formulas for H1′ and H2′ are well-defined on all of I. Therefore,
Z Z
u2 g u1 g
H1 = dx, H2 = dx.
u1 u2 − u′2 u1
′ u′2 u1 − u′1 u2

This shows that if H1 and H2 are primitives of the functions above, then

u = H1 u1 + H2 u2

is a particular solution of (8.14). Finally, the general solution is obtained by adding any
solution of the homogeneous equation:

u = Au1 + Bu2 + H1 u1 + H2 u2 , A, B ∈ R.

We summarize this in the following proposition.

Version: December 30, 2025. 254

 Chapter 8.1

Theorem 8.20: Existence and Uniqueness: the Non-Homogeneous Case

Given a0 , a1 ∈ R, let ∆ = a21 − 4a0 and consider the following solutions of (8.10):

∆>0: u1 (x) = eαx , u2 (x) = eβx , α, β as in (8.11),

∆<0: u1 (x) = eαx sin(βx), u2 (x) = eαx cos(βx), α, β as in (8.12),
∆=0: u1 (x) = eαx , u2 (x) = xeαx , α as in (8.13).

u2 g u1 g
Let H1 and H2 be primitives of and ′ , respectively. If u ∈
u′1 u2 ′
− u2 u1 u2 u1 − u′1 u2
C 2 (I) solves (8.14), then there exist A, B ∈ R such that

u = Au1 + Bu2 + H1 u1 + H2 u2 .

In other words, the set of solutions of (8.14) forms a two-dimensional affine subspace
of C 2 (I).

Proof. First we show existence. By the computation above, if H1 and H2 are primitives of

u2 g u1 g
and ,
u′1 u2 − u′2 u1 u′2 u1 − u′1 u2

then the function

35 up = H1 u1 + H2 u2

satisfies
u′′p + a1 u′p + a0 up = g,

so up is a particular solution of (8.14). Since the equation is linear, for any A, B ∈ R the
function
u = Au1 + Bu2 + up

also solves (8.14). This proves existence of solutions of the stated form.
For uniqueness, let u ∈ C 2 (I) be any solution of (8.14), and define

v = u − up = u − (H1 u1 + H2 u2 ).

Then v solves the homogeneous equation (8.10) and therefore, by Theorem 8.16, there exist
A, B ∈ R such that
v = Au1 + Bu2 .

Hence u = v + up = Au1 + Bu2 + H1 u1 + H2 u2 , as desired.

Although this method is very general, in practice the integrals defining H1 and H2 can be
quite complicated. In some (very special) situations it is easier to guess a particular solution

Version: December 30, 2025. 255

 Chapter 8.1

by trying functions of the form

p(x)eγx , p(x)eγx cos(ηx), p(x)eγx sin(ηx),

where p is a polynomial and γ, η > 0 are chosen to reflect the structure of g.

Example 8.21. — Solve

u′′ (x) + u(x) = 1, u(0) = 0, u′ (0) = 1.

The homogeneous equation u′′ + u = 0 has solutions A cos(x) + B sin(x), so we look for a
solution of the form
u(x) = H1 (x) cos(x) + H2 (x) sin(x),

where H1 , H2 are unknown. This is exactly the method of variation of constants specialized
to this case.
The computation leads to the system

H1′ (x) cos(x) + H2′ (x) sin(x) = 0, −H1′ (x) sin(x) + H2′ (x) cos(x) = 1.

Solving for H1′ and H2′ as in the general method gives

35
Z Z
sin(x)
H1 = − 2 dx = − sin(x) dx,
sin (x) + cos2 (x)
Z Z
cos(x)
H2 = dx = cos(x) dx.
cos (x) + sin2 (x)
2

Hence we may take H1 = cos(x) and H2 = sin(x), which gives a particular solution

u(x) = cos(x) cos(x) + sin(x) sin(x) = cos2 (x) + sin2 (x) = 1.

(In this case, one could also have tried to guess this particular solution!) The general solution
is therefore
u(x) = 1 + A cos(x) + B sin(x).

Imposing u(0) = 0 and u′ (0) = 1 (two initial conditions for the unknown u) yields

u(x) = 1 − cos(x) + sin(x).

Example 8.22. — Solve

u′′ (x) + u(x) = sin(x), u(0) = 0, u′ (0) = 1.

Version: December 30, 2025. 256

 Chapter 8.1

Again, the homogeneous solutions are A cos(x) + B sin(x). The method of variation of con-
stants gives Z Z
sin(x) sin(x)
H1 = − dx = − sin2 (x) dx,
sin2 (x) + cos2 (x)
Z Z
cos(x) sin(x)
H2 = dx = cos(x) sin(x) dx.
cos2 (x) + sin2 (x)
One possible choice is

1 1
H2 = − cos2 (x).


H1 = cos(x) sin(x) − x ,
2 2

This gives the particular solution

1
u(x) = H1 (x) cos(x) + H2 (x) sin(x) = − x cos(x).
2

Hence the general solution is

x cos(x)
u(x) = − + A cos(x) + B sin(x).
2

Imposing u(0) = 0 and u′ (0) = 1, we find

1 3
u(x) = − x cos(x) + sin(x).
35 2 2

Remark 8.23. — In Example 8.22, notice the appearance of the factor x in front of cos(x) in
the particular solution. This occurs because sin(x) and cos(x) already solve the homogeneous
equation, so a particular solution cannot be obtained by taking a simple linear combination
of them.

The situation from Example 8.22 illustrates a general principle behind the method of un-
determined coefficients, which provides a systematic way to guess a particular solution of a
linear ODE with constant coefficients. The procedure can be summarised as follows:

• Begin by making a guess that has the same functional form as the forcing term g(x).

– If g(x) is a polynomial of degree n, guess a general polynomial of degree n.

– If g(x) = eαx , guess Aeαx .
– If g(x) = sin(βx) or cos(βx), guess A cos(βx) + B sin(βx).
– If g(x) is a sum of such terms, combine the corresponding guesses.

• Check whether this first guess lies in the space of solutions of the homogeneous equation.
If it does, then substituting it into the differential equation produces zero on the left-
hand side, so it cannot match the non-zero forcing term g(x).

Version: December 30, 2025. 257

 Chapter 8.1

• If the guess lies in the homogeneous solution space, multiply it by x. For a second-order
equation, if this new guess still belongs to the homogeneous space, multiply by x2 . At
this point the guess will not solve the homogeneous equation, so no further powers are
needed.

This explains why Example 8.22 requires a particular solution of the form x cos(x): the
functions sin(x) and cos(x) solve the homogeneous equation, so we should look for a particular
solution of the form Ax cos(x) + Bx sin(x). A similar phenomenon occurs when g(x) = eαx
and α is a root of the characteristic equation. In that case one should try Axeαx . However,
when ∆ = 0, then both eαx and xeαx solve the homogeneous equation, so the correct guess
for a particular solution is Ax2 eαx .
35
Exercise 8.24. — Solve the following ODEs (the unknown function is always u):

1. u′′ (x) + u′ (x) + u(x) = sin(2x), u(0) = 0, u′ (0) = 1.

Hint: Look for a particular solution of the form a sin(2x) + b cos(2x).

2. u′′ (x) + 4u(x) = cos(2x), u(0) = 1, u′ (0) = 0.

Hint: Look for a particular solution of the form ax cos(2x) + bx sin(2x).

3. u′′ (x) + u′ (x) − 2u(x) = x2 , u(0) = 2, u′ (0) = 1.

Hint: Look for a particular solution of the form ax2 + bx + c.

4. u′′ (x) + 2u′ (x) − 3u(x) = cos(x) + x, u(0) = 1, u′ (0) = 1.

Hint: Look for a particular solution of the form a sin(x) + b cos(x) + cx + d.

Version: December 30, 2025. 258

 Chapter 8.2

8.2 Existence and Uniqueness for ODEs

8.2.1 Existence and Uniqueness for First Order ODEs

Our goal now is to present the general theory of first-order ODEs for real-valued functions on
the real line. In general, a first-order ODE is an equation relating x, u(x), and u′ (x),

G x, u(x), u′ (x) = 0.

In this section we restrict to equations for which one can “isolate” u′ , so that one can write
the ODE in normal form:
u′ (x) = f (x, u(x)).

Definition 8.25: First-order ODEs in normal form

36
A first-order ODE in normal form is an equation of the type

u′ (x) = f (x, u(x)),

where f : R2 → R is a given function and u : R → R is the unknown.

The Cauchy–Lipschitz Theorem, also known as the Picard–Lindelöf Theorem, is a funda-

mental result in the theory of ODEs. It ensures the existence and uniqueness of solutions
under suitable conditions on f .
In the next theorem we need to assume that f is continuous as a function of the two
variables x and y. This means that, for any point (x0 , y0 ) in the domain and for any ε > 0,
there exists δ > 0 such that

|x − x0 | < δ and |y − y0 | < δ =⇒ |f (x, y) − f (x0 , y0 )| < ε.

Version: December 30, 2025. 259

 Chapter 8.2

Theorem 8.26: Cauchy-Lipschitz: Global Version

Let f : R × R → R satisfy the following conditions:

1. f is continuous in R × R;

2. f is Lipschitz with respect to the second variable; that is, there exists a constant
L > 0 such that

|f (x, y1 ) − f (x, y2 )| ≤ L|y1 − y2 | ∀ x ∈ R, y1 , y2 ∈ R.

Then, for any point (x0 , y0 ) ∈ R × R there exists a unique C 1 function u : R → R such
that (
u′ (x) = f (x, u(x)) for all x ∈ R,
(8.17)
u(x0 ) = y0 .

As we shall see in Section 8.2.2 below, the proof is based on the method of successive
approximations, also known as Picard iterations. It involves constructing a sequence of con-
tinuous functions that converge to the solution of the differential equation. Before diving into
the proof of this important theorem, we first discuss some examples and generalizations.
Theorem 8.26 guarantees that solutions to the first order ODE (8.17) are unique when f
is Lipschitz in the second variable. This assumption is crucial, as the next example shows.
36

Example 8.27. — Consider the ODE

u′ (x) = |u(x)|α , u(0) = 0, (8.18)

with α ∈ (0, 1].

• For α = 1 the function f (y) = |y| is Lipschitz, since

|f (y1 ) − f (y2 )| = ||y1 | − |y2 || ≤ |y1 − y2 |.

Hence, Theorem 8.26 guarantees that the solution is unique. Since the constant function
u = 0 is a solution, this is the unique solution.
• For α < 1 the function f (y) = |y|α is not Lipschitz. Indeed, if this function were Lipschitz,
then there would exist a constant L > 0 such that

|f (y) − f (0)| = |y|α ≤ L|y| ∀ y ∈ R.

This would imply

1 ≤ L|y|1−α ∀ y ∈ R \ {0},

but this is false since limy→0 |y|1−α = 0 (recall that α < 1).

Version: December 30, 2025. 260

 Chapter 8.2

Note that, also in this case, the constant function u = 0 is a solution. We now try to use
the method of separation of variables (recall Section 8.1.2) to find a second solution that is
not zero, say with u(x) > 0 somewhere:

u′ (x)
u′ (x) = u(x)α =⇒ = 1,
u(x)α

so, by integration, we obtain

u(x)1−α
Z Z
du
= dx = x + A =⇒ = x + A.
uα 1−α

Choosing A = 0 (which is compatible with u(0) = 0) we get

1
u(x) = (1 − α)x 1−α
,

which is positive for x > 0. Hence, the function


 0 for x ≤ 0,
u(x) = 1
 (1 − α)x
1−α for x > 0,

is a second solution of (8.18). Actually, given any value x0 ≥ 0, all the functions
36

 0 for x ≤ x0 ,
ux0 (x) = 1
 (1 − α)(x − x )
1−α
0 for x > x0 ,

solve (8.18), so there are infinitely many solutions.

Motivated by the previous example, one may wonder if the solution of (8.18) is unique for
α > 1. We begin with the following observation, stated as an exercise.

Exercise 8.28. — Let α > 1. Prove that the function f : R → R given by f (y) = |y|α is
locally Lipschitz (i.e., it is Lipschitz in every compact interval [a, b]), but it is not Lipschitz
on the whole R.
Hint: Use Corollary 5.32.

By the previous exercise, we see that Theorem 8.26 does not apply to (8.18) when α > 1.
Still, since this function is locally Lipschitz, one may hope that some existence and uniqueness
theorem still holds. This is indeed the case, as implied by the local version of the Cauchy–
Lipschitz Theorem stated below. As we shall discuss later, since now the function f is only
assumed to be locally Lipschitz, in general we cannot find a solution u defined on the whole
R.

Version: December 30, 2025. 261

 Chapter 8.2

Theorem 8.29: Cauchy-Lipschitz: Local Version

Let I ⊂ R be an interval, and let f : I × R → R satisfy the following conditions:

1. f is continuous in I × R;

2. f is locally Lipschitz with respect to the second variable; that is, for every pair of
compact intervals [a, b] ⊂ I and [c, d] ⊂ R there exists a constant L > 0 such that

|f (x, y1 ) − f (x, y2 )| ≤ L|y1 − y2 | ∀ x ∈ [a, b], y1 , y2 ∈ [c, d].

Then, for any point (x0 , y0 ) ∈ I × R there exist an interval I ′ ⊂ I containing x0 and a
unique C 1 function u : I ′ → R such that
(
u′ (x) = f (x, u(x)) for all x ∈ I ′ ,
(8.19)
u(x0 ) = y0 .

In other words, under a local Lipschitz assumption, one can only guarantee the existence
and uniqueness of a solution on some interval around x0 . Moreover, as long as the solution u(x)
remains bounded within I ′ , one can continue applying Theorem 8.29 to extend the interval I ′
as much as possible.
36 To better understand why solutions are defined only on some interval I ′ ⊂ I, we consider
the following example.

Example 8.30. — Consider the ODE

u′ (x) = u(x)2 , u(0) = 1.

The function f (y) = y 2 is locally Lipschitz, so Theorem 8.29 applies.

To find the solution, we use separation of variables. More precisely, since u(0) > 0, by
continuity u will be positive in a neighborhood of 0 and we get

u′ (x)
u′ (x) = u(x)2 =⇒ = 1,
u(x)2

so, by integration, we obtain

Z Z
du 1
= dx = x + A =⇒ − = x + A.
u2 u(x)

Choosing x = 0 this implies A = −1 and we get

1
u(x) = .
1−x

Version: December 30, 2025. 262

 Chapter 8.2

Note that this function solves the ODE on (−∞, 1), but limx→1− u(x) = ∞, so we cannot
extend this solution beyond x = 1.

Exercise 8.31. — Consider the ODE

u′ (x) = u(x)α , u(0) = 1,

36
with α > 1. Show that Theorem 8.29 applies and find the unique solution (which again blows
up in finite time).

Although Theorem 8.29 guarantees that most nonlinear ODEs have a unique solution (at
least locally in x), nonlinear ODEs are very difficult to solve and there are no general tech-
niques to tackle such problems, neither in practice nor in theory. Therefore, in applications,
one often resorts to numerical methods.

8.2.2 Extra material: Proof of Theorem 8.26

Remark 8.32. — In the proof, we shall use the following fact: if v : R → R is a contin-
uous function, then also the function s 7→ f (s, v(s)) is continuous. This is a consequence of
the continuity of f and the fact that the composition of continuous functions is continuous.
Although we did not prove this fact in this specific setting where f depends on two variables,
this can be proved in the same way as in Proposition 3.15.

Proof of Theorem 8.26. Let L > 0 be a Lipschitz constant for f with respect to the second
variable, as in Theorem 8.26. We first prove local existence on a short interval around x0 ,
then uniqueness on that interval, and finally extend the solution to the whole R.
• Step 1: An equivalent integral equation. We first show that u : R → R is a C 1 solution
to (8.17) if and only if u is a continuous function satisfying
Z x
u(x) = y0 + f (s, u(s)) ds ∀ x ∈ R. (8.20)
x0

Indeed, if u solves the ODE, then by integration (see Corollary 7.5) we deduce the validity of
(8.20).
Conversely, if u is a continuous function satisfying (8.20), then Theorem 7.4 and Re-
mark 8.32 imply that Z x
x 7→ f (s, u(s)) ds
x0

is a primitive of the continuous function f (x, u(x)). Hence u′ (x) = f (x, u(x)) for all x, so u is
C 1 . Finally, choosing x = x0 in (8.20) we deduce that u(x0 ) = y0 .
Therefore, to prove existence, it suffices to construct a solution to (8.20). This will be
accomplished by constructing what are known as Picard approximations, that is, a sequence
of functions that converge to a solution of (8.20).

Version: December 30, 2025. 263

Chapter 8.2

• Step 2: Construction of Picard approximations. First, we define the continuous

function u0 : R → R as
u0 (x) = y0 ∀ x ∈ R.

Then we define u1 : R → R as
Z x
u1 (x) = y0 + f (s, u0 (s)) ds.
x0

The integral is well-defined since u0 is continuous and therefore s 7→ f (s, u0 (s)) is continuous
(see Remark 8.32). We also observe that u1 is the primitive of a continuous function, so it is
C 1 (and, in particular, continuous).
More generally, given n ∈ N, once the continuous function un : R → R is constructed, we
define Z x
un+1 (x) = y0 + f (s, un (s)) ds.
x0

Again, since un is continuous, also s 7→ f (s, un (s)) is continuous, and therefore un+1 is C 1
(and in particular continuous).
• Step 3: Convergence of Picard approximations on a short interval. Set

1
τ= (so that Lτ = 21 ).
2L

We now prove the uniform convergence of the sequence of Picard approximations un on the
interval [x0 − τ, x0 + τ ] by showing that this sequence corresponds to the partial sums of a
uniformly convergent series of functions.
Define vk = uk − uk−1 for k ≥ 1, so that
n
X
un (x) = y0 + vk (x).
k=1

We want to prove that the series

∞
X
u∞ (x) = y0 + vk (x)
k=1

converges absolutely for every x ∈ [x0 −τ, x0 +τ ], so that the function u∞ : [x0 −τ, x0 +τ ] → R
is well-defined, and that the sequence of functions {un }∞ n=0 converges uniformly to u∞ on
[x0 − τ, x0 + τ ].
To this end, we observe that
Z x

vn+1 (x) = un+1 (x) − un (x) = f (s, un (s)) − f (s, un−1 (s)) ds ∀ x ∈ R,
x0

Version: December 30, 2025. 264

Chapter 8.2

therefore, by the Lipschitz regularity of f in the second variable,

Z x
|vn+1 (x)| ≤ f (s, un (s)) − f (s, un−1 (s)) ds
x0
Z x Z x (8.21)
≤L un (s) − un−1 (s) ds = L vn (s) ds
x0 x0

for every x ∈ R and n ≥ 1.

For n ≥ 1, define
an = max |vn (x)|.
x∈[x0 −τ,x0 +τ ]

Given x ∈ [x0 − τ, x0 + τ ], we have [x0 , x] ⊂ [x0 − τ, x0 + τ ], hence

Z x
|vn (s)| ds ≤ an |x − x0 | ≤ an τ,
x0

and combining this with (8.21) yields

an
|vn+1 (x)| ≤ Lan τ = .
2

Since x is arbitrary in [x0 − τ, x0 + τ ], this proves that

an
an+1 ≤ Lan τ = ∀ n ≥ 1.
2

By induction we deduce
an+1 ≤ 2−n a1 ∀ n ≥ 1.

Thus, for every x ∈ [x0 − τ, x0 + τ ] and every k ≥ 1,

|vk (x)| ≤ 2−(k−1) a1 .

By the majorant criterion, the series ∞ k=1 vk (x) converges absolutely for every x ∈ [x0 −
P

τ, x0 + τ ].
Moreover, for x ∈ [x0 − τ, x0 + τ ] and n ≥ 0,
∞
X ∞
X ∞
X
|u∞ (x) − un (x)| = vk (x) ≤ |vk (x)| ≤ a1 2−(k−1)
k=n+1 k=n+1 k=n+1 (8.22)
−n
= a1 2 −−−→ 0,
n→∞

so the sequence of functions {un }∞

n=0 converges uniformly to u∞ on [x0 − τ, x0 + τ ].
• Step 4: The limit function u∞ solves (8.20). We now take the limit as n → ∞ in
Z x
un+1 (x) = y0 + f (s, un (s)) ds, ∀ x ∈ [x0 − τ, x0 + τ ].
x0

The left-hand side converges to u∞ (x) uniformly on [x0 − τ, x0 + τ ] by Step 3.

Version: December 30, 2025. 265

Chapter 8.2

For the right-hand side, recalling (8.22) and that Lτ = 1/2, we estimate
Z x Z x
|u∞ (s) − un (s)| ds ≤ L|x − x0 |a1 2−n ≤ a1 2−(n+1) ,


f (s, u∞ (s)) − f (s, un (s)) ds ≤ L
x0 x0

which proves that

Z x Z x
f (s, un (s)) ds → f (s, u∞ (s)) ds as n → ∞.
x0 x0

So we conclude that u∞ solves (8.20) on [x0 −τ, x0 +τ ], which shows the existence of a solution
u∞ on [x0 − τ, x0 + τ ].
• Step 5: Local uniqueness on [x0 − τ, x0 + τ ]. Let u1 , u2 : [x0 − τ, x0 + τ ] → R be two
solutions of (8.17), and therefore of (8.20). Then, for each x ∈ [x0 − τ, x0 + τ ],
Z x

u1 (x) − u2 (x) = f (s, u1 (s)) − f (s, u2 (s)) ds,
x0

so Z x Z x
|u1 (x) − u2 (x)| ≤ f (s, u1 (s)) − f (s, u2 (s)) ds ≤ L |u1 (s) − u2 (s)| ds.
x0 x0

Define
a= max |u1 (x) − u2 (x)|.
x∈[x0 −τ,x0 +τ ]

For x ∈ [x0 − τ, x0 + τ ] we then have

a
|u1 (x) − u2 (x)| ≤ La|x − x0 | ≤ Lτ a = ,
2

so taking the maximum over x ∈ [x0 − τ, x0 + τ ] yields

a
a≤ .
2

This implies that a = 0, that is, u1 = u2 on [x0 − τ, x0 + τ ].

• Step 6: Global existence and uniqueness. We now extend the solution uniquely to the
whole R by iterating the local existence and uniqueness argument.
Set x1 = x0 + τ and y1 = u∞ (x1 ). Applying the same Picard construction with initial data
(x1 , y1 ) gives a unique solution u(1) on [x1 − τ, x1 + τ ] = [x0 , x0 + 2τ ]. On the overlap [x0 , x1 ]
both u∞ and u(1) solve (8.17) with the same value at x1 , so by Step 5 they coincide there.
Hence we can glue them and obtain a single solution on [x0 − τ, x1 + τ ] = [x0 − τ, x0 + 2τ ].
Iterating this construction to the right, we obtain a unique solution on [x0 − τ, ∞). A
completely analogous construction to the left (starting from x0 − τ and moving to the left
by steps of length τ ) yields a unique solution on (−∞, x0 + τ ]. Gluing the left and right
pieces together (again using local uniqueness on overlaps) gives a unique solution u : R → R
satisfying (8.17). This completes the proof of Theorem 8.26.

Version: December 30, 2025. 266

 Chapter 8.2

8.2.3 Higher Order ODEs

Before considering general higher-order ODEs, we begin with a familiar second-order linear
equation and show how it can be rewritten as a system of first-order ODEs.

Example 8.33 (Linear second-order equation as a first-order system). — Let a0 , a1 : I → R

be continuous functions and consider the linear homogeneous second-order ODE

u′′ (x) + a1 (x)u′ (x) + a0 (x)u(x) = 0 ∀ x ∈ I. (8.23)

Define
U1 (x) = u(x), U2 (x) = u′ (x).

Then
U1′ (x) = U2 (x), U2′ (x) = u′′ (x) = −a1 (x)U2 (x) − a0 (x)U1 (x),

so (8.23) is equivalent to the first-order system

(
U1′ (x) = U2 (x),
U2′ (x) = −a1 (x)U2 (x) − a0 (x)U1 (x).
36
Thus, prescribing the values of u(x0 ) and u′ (x0 ) at some x0 ∈ I is equivalent to prescribing
the initial condition (U1 (x0 ), U2 (x0 )) for the associated first-order system.

The same idea extends to arbitrary higher-order ODEs. More precisely, suppose we are
given an n-th order ODE of the form

G x, u(x), u′ (x), u′′ (x), . . . , u(n) (x) = 0,

and assume that the highest derivative can be isolated and written as

u(n) (x) = f x, u(x), u′ (x), . . . , u(n−1) (x) , (8.24)

where f : R × Rn → R is a given function.

We introduce the variables

U1 = u, U2 = u′ , U3 = u′′ , . . . , Un = u(n−1) .

By definition, we have

U1′ = U2 , U2′ = U3 , ..., ′

Un−1 = Un ,

Version: December 30, 2025. 267

 Chapter 8.2

and equation (8.24) becomes

Un′ = u(n) = f x, U1 , U2 , . . . , Un .

Therefore, the n-th order equation (8.24) is equivalent to the first-order system



 U1′ = U2 ,
U2′ = U3 ,




..

 .
′




 Un−1 = Un ,
 U ′ = f x, U , U , . . . , U
.

n 1 2 n

The Cauchy–Lipschitz Theorem (both its global and local versions) extends to systems
of first-order ODEs and ensures existence and uniqueness of solutions whenever the right-
hand side is continuous and (locally) Lipschitz with respect to the variables (U1 , . . . , Un ). In
particular, once the initial conditions

U1 (x0 ) = u(x0 ), U2 (x0 ) = u′ (x0 ), ..., Un (x0 ) = u(n−1) (x0 )

are prescribed at some x0 ∈ I, there exists a unique (local) solution to the system, and hence
36 a unique solution to the original n-th order ODE.
As an application, we obtain a classical structure result for linear second-order equations.

Theorem 8.34: Existence and Uniqueness for Linear Second-Order ODEs

Let a0 , a1 : I → R be continuous and bounded functions. Then the set of solutions of

the linear homogeneous equation

u′′ (x) + a1 (x)u′ (x) + a0 (x)u(x) = 0 ∀x ∈ I

is a two-dimensional linear subspace of C 2 (I).

Proof. Fix x0 ∈ I and define U1 = u, U2 = u′ . As shown in Example 8.33, the equation can
be rewritten as the first-order system
(
U1′ = U2 ,
U2′ = −a1 (x)U2 − a0 (x)U1 .

The right-hand side is linear in (U1 , U2 ) and continuous and bounded in x, so it is Lipschitz
in both U1 and U2 . Thus, the Cauchy–Lipschitz Theorem for systems ensures the existence
and uniqueness of solutions once (U1 (x0 ), U2 (x0 )) is prescribed. Equivalently, the original
second-order equation has a unique solution once u(x0 ) and u′ (x0 ) are given.

Version: December 30, 2025. 268

 Chapter 8.2

Let u1 be the unique solution satisfying u1 (x0 ) = 1 and u′1 (x0 ) = 0, and let u2 be the
unique solution satisfying u2 (x0 ) = 0 and u′2 (x0 ) = 1. By linearity, Au1 + Bu2 is a solution
for every A, B ∈ R.
Conversely, if u is any solution and we set A = u(x0 ) and B = u′ (x0 ), then the function
36 v = u − Au1 − Bu2 solves the equation and satisfies v(x0 ) = v ′ (x0 ) = 0. By uniqueness, v
must be identically zero, and therefore u = Au1 + Bu2 .

The higher-dimensional version of the Cauchy–Lipschitz Theorem for systems will be stud-
ied in Analysis 2. Understanding the proof of the one-dimensional case presented in Sec-
tion 8.2.2 provides a solid foundation for that more general theory.

Version: December 30, 2025. 269

Index

absolute value, 38 decreasing, 77

accumulation point, 60 dense, 49
affine distance, 34
function, 154, 224
equibounded, 116
Airy equation, 237
equicontinuity, 116
arc tangent, 179
Euler’s number, 96
arccotangent, 179
eventually constant, 58
area, 189
exponential function, 96
argument, 148
exponential series, 140
Arzelà–Ascoli theorem, 116
associativity, 82 factor, 81
fractional part, 48
base
logarithm, 100 geometric sequence, 74
Bessel equation, 237
bijection, 27 higher derivative, 156
bijective, 27 Hölder continuous, 102
boundary conditions, 237
identity function, 87
bounded, 42, 64
image, 27
bounded function, 77
imaginary part, 35
Cauchy sequence, 71 imaginary unit, 35
Cauchy–Schwarz inequality, 39 improper limit, 73
change of variable, 215 increasing, 77
closed, 34 indefinite values, 46
closed disk, 40 index, 81
coefficient, 134 inferior limit, 68
completeness axiom, 26 Infimum, 45
complex conjugation, 37 injection, 27
complex numbers, 35 injective, 27
continuous integrand, 189
piecewise, 197 integration constant, 214
continuous extension, 104 integration limit, 189
continuum, 57 inverse function, 87
Inverse Hyperbolic Cosine, 181
decimal fraction, 50 Inverse Hyperbolic Sine, 180
decomposition, 183
270
Chapter 8.2 INDEX

inverse mapping, 87 square root function, 28

strictly monotone, 77
Lebesgue integral, 189
subsequence, 59
limit, 58
summand, 81
line, 154
superior limit, 68
Lipschitz continuous, 93
supremum, 42
lower bound, 42
surjection, 27
lower sums, 188
surjective, 27
maximum, 42
two-point compactification, 45
minimum, 42
monotone, 77 uniform convergence, 113, 138
piecewise, 196 uniformly continuous, 91
unit circle, 148
natural logarithm, 100
upper bound, 42
neighbourhood, 33
upper sums, 188
norm, 38
number line, 26 variable, 76, 134

open, 34 zero, 76
open disk, 40

partial fraction decomposition, 219

partition, 182, 183
piecewise continuous, 197
piecewise monotone, 196
pointwise convergence, 112
polar coordinates, 148
power series, 134, 137
preimage, 27

radius of convergence, 135

rational function, 219
real decimal fraction, 51
real part, 35
real-valued, 76
removable discontinuity, 104
restriction, 80
root, 89
rounding function, 48
running variable, 81

sequence, 58, 112

sequential continuity, 83
smooth function, 157
Version: December 30, 2025. 271
Bibliography

[ACa2003] N. A’Campo, A natural construction for the real numbers arXiv preprint 0301015,
(2003)

[Apo1983] T. Apostol, A proof that Euler missed: Evaluating ζ(2) the easy way The Mathe-
matical Intelligencer 5 no.3, p. 59–60 (1983)

[Aig2014] M. Aigner and G. M. Ziegler, Das BUCH der Beweise Springer, (2014)

[Amm2006] H. Amann und J. Escher, Analysis I, 3. Auflage, Grundstudium Mathematik,

Birkhäuser Basel, (2006)

[Bla2003] C. Blatter, Analysis I ETH Skript, [Link] blatter/[Link]

(2003)

[Bol1817] B. Bolzano, Rein analytischer Beweis des Lehrsatzes, daß zwischen je zwei Werthen,
die ein entgegengesetztes Resultat gewähren, wenigstens eine reelle Wurzel der Gleichung
liege, Haase Verl. Prag (1817)

[Boo1847] G. Boole, The mathematical analysis of logic Philosophical library, (1847)

[Can1895] G. Cantor, Beiträge zur Begründung der transfiniten Mengenlehre Mathematische

Annalen 46 no.4, 481–512 (1895)

[Cau1821] A.L. Cauchy, Cours d’analyse de l’école royale polytechnique L’Imprimerie Royale,
Debure frères, Libraires du Roi et de la Bibliothèque du Roi. Paris, (1821)

[Ded1872] R. Dedekind, Stetigkeit und irrationale Zahlen Friedrich Vieweg und Sohn, Braun-
schweig (1872)

[Die1990] J. Dieudonné, Elements d’analyse Editions Jacques Gabay (1990)

[Hat02] A. Hatcher, Algebraic Topology Cambridge University Press (2002)

[Hil1893] D. Hilbert, Über die Transzendenz der Zahlen e und π Mathematische Annalen 43,
216-219 (1893)

[Hos1715] G.F.A. Marquis de l’Hôpital, Analyse des Infiniment Petits pour l’Intelligence des
Lignes Courbes 2nde Edition, F. Montalant, Paris (1715)

272
Chapter 8.2 BIBLIOGRAPHY

[Lin1894] E. Lindelöf, Sur l’application des méthodes d’approximations successives à l’étude

des intégrales réelles des équations différentielles ordinaires Journal de mathématiques pures
et appliquées 10 no.4, 117–128 (1894)

[Rus1903] B. Russell, The principles of mathematics WW Norton & Company, (1903)

[Rot88] J. J. Rotman, An introduction to Algebraic Topology Graduate Texts in Mathematics

119 Springer 1988

[Smu1978] R. Smullyan, What is the name of this book? Prentice-Hall, (1978)

[Zag1990] D. Zagier, A one-sentence proof that every prime p ≡ 1 mod 4 is a sum of two
squares. Amer. Math. Monthly 97, no.2, p. 144 (1990)

Version: December 30, 2025. 273