Calculus 1 – Spring 2019 Section 2

Jacob Shapiro
April 15, 2019

Contents
1 Logistics 2
1.1 How these lecture notes are meant . . . . . . . . . . . . . . . . . 3

2 Goal of this class and some motivation 4

2.1 Naive description of math . . . . . . . . . . . . . . . . . . . . . . 4
2.2 What is Calculus? . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Naive Naive Set Theory 6

4 Special sets of numbers 10

4.1 Intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5 Functions 13
5.1 Functions from R → R . . . . . . . . . . . . . . . . . . . . . . . . 17
5.1.1 Basic functions and their shapes . . . . . . . . . . . . . . 19
5.1.2 Trigonometric functions . . . . . . . . . . . . . . . . . . . 19
5.2 Special sets associated with a function . . . . . . . . . . . . . . . 21
5.3 Construction of new functions . . . . . . . . . . . . . . . . . . . . 23

6 Limits 25
6.1 The notion of a distance on R . . . . . . . . . . . . . . . . . . . . 26
6.2 Limits of sequences–functions from N → R . . . . . . . . . . . . . 28
6.3 Limits of functions from R → R . . . . . . . . . . . . . . . . . . . 37

7 Continuity of functions from R → R 46

8 Derivatives 50
8.1 Application: Minima and Maxima . . . . . . . . . . . . . . . . . 68
8.2 Convexity and concavity . . . . . . . . . . . . . . . . . . . . . . . 75
8.3 Application: Newton’s method . . . . . . . . . . . . . . . . . . . 78
8.4 Application: Linear Approximation . . . . . . . . . . . . . . . . . 79

1
9 Integrals 80
9.1 The supremum and infimum . . . . . . . . . . . . . . . . . . . . . 80
9.2 The Darboux integral . . . . . . . . . . . . . . . . . . . . . . . . 81
9.3 Properties of the integral . . . . . . . . . . . . . . . . . . . . . . 89

10 Important functions 101

10.1 The trigonometric functions . . . . . . . . . . . . . . . . . . . . . 101
10.2 The exponential and logarithmic functions . . . . . . . . . . . . . 102
10.2.1 The natural base for the logarithm . . . . . . . . . . . . . 103
10.3 The hyperbolic functions . . . . . . . . . . . . . . . . . . . . . . . 106
10.4 Special functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
10.4.1 The sinc function . . . . . . . . . . . . . . . . . . . . . . . 107
10.4.2 The square root . . . . . . . . . . . . . . . . . . . . . . . 107

11 Useful algebraic formulas to recall 107

11.1 Factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
11.2 Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

12 Dictionary of graphical symbols and acronyms 109

12.1 Graphical symbols . . . . . . . . . . . . . . . . . . . . . . . . . . 109
12.2 Acronyms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

1 Logistics
• Instructor: Jacob Shapiro [email protected]
• Course website: http://math.columbia.edu/~shapiro/teaching.html

• Location: 207 Mathematics Building

• Time: Mondays and Wednesdays 4:10pm-5:25pm (runs through January
23rd until May 6th 2019 for 28 sessions).
• Recitation sessions: Fridays 2pm-3pm in Hamilton 602.

• Office hours: Tuesdays 6pm-8pm and Wednesdays 5:30pm-7:30pm (or by

appointment), in 626 Mathematics (starting Jan 29th).
• Teaching Assistants:
– Donghan Kim [email protected]
– Mat Hillman [email protected]
– Ziad Saade [email protected]
• TAs office hours: Will be held in the help room (502 Milstein Center)
http://www.math.columbia.edu/general-information/help-rooms/502-milstein/
during the following times:

2
 – Mat: Fridays 10am-12pm.
– Donghan: Thursdays 4pm-6pm.
– Ziad: Tuesdays 4pm-6pm.
• Misc. information: Calculus @ Columbia http://www.math.columbia.
edu/programs-math/undergraduate-program/calculus-classes/
• Getting help: Your best bet is the office-hours, and then the help room.
If you prefer impersonal communication, you may use the Piazza website
to pose questions (even anonymously, if you’re worried). TAs will monitor
this forum regularly.

• Textbook: Lecture notes will be made available online. I will do my best

to follow the Columbia Calculus curriculum so as to make sure you can
go on to Calculus II smoothly. I will use material from various textbooks,
some of which include: Spivak [5], Apostol [1], Courant [2] and some-
times Stewart [6] just to set the timeline (since this is what Columbia’s
curriculum is based upon).

• Homework assignments: Assignments will appear on this website after

the Monday lecture every week and are meant to be solved before the
Monday lecture of the following week. Official solutions will be published
here one week before the relevant mid-term. You do not have to hand in
assignments every week, but you may present your solutions to the TAs
or me to gain extra credit (see below).
• Grading: There will be two midterms during lecture time (see below for
dates) and one final (after the last lecture). Your grade is (automatically)
the higher of the following two options: [Option A] The final carries 50%
weight, the midterms each carry 25% weight for a total of 100% of your
grade. [Option B] The final carries 40% weight, the midterms each carry
25% weight for a total of 90% of your grade. The remainder 10% is given to
you if you succeed in accomplishing the following task at least three times
during the semester: show up (no appointment necessary) to one of the
office hours of the TAs or me, successfully orally present a solution to one
of the homework assignments whose solution has not been published yet
(should take you not more than 10 minutes max), and in the same occasion
submit a neatly written down solution to the thing you just presented.

1.1 How these lecture notes are meant

These lecture notes are meant as a support for the lectures, by augmenting with
a few more details and possibly enriching with more complete explanations and
examples that I might not have had time to go over during the short lectures.
Since they are updated on a rather weekly basis, sometimes the earlier sections
which had already been covered in class, it is encouraged not to print out the
PDF but rather read it in digital form.

3
 Strictly speaking the material of calculus really starts in Section 4 onward
(Section 2 is a philosophical motivation and Section 3 sets up the language and
notation which is the basis of how we think about the various objects we deal
with).
Since calculus is not a proof-oriented class, most of the statements in this
text are not substantiated by a demonstration that explains why they are correct
(i.e., a proof), be it formal or not. Sometimes I chose to ignore this principle
and include the proof in the body of the text anyway, mostly because I felt
the demonstration was not very much beyond what would be required of an
average calculus student, and to cater to those readers who want to go a bit
deeper. The reader can recognize very easily these proofs because they start
with a Proof and are encased in a box. The contents of these proofs is not part
of the curriculum of the class and will not be required for the midterms or final.

2 Goal of this class and some motivation

This being one of the first classes you take in mathematics–even if you are not
a math major–has as its goal to expose you to mathematical thinking, which
could be thought of as a way to communicate and reason about abstract notions
in an efficient and precise (i.e. the opposite of vague) way. As such, it is first
and foremost a language that one has to study. Like your mother tongue, you
are exposed to math even before you know what a language is, or that you’re
undergoing the process of studying it. In this regard, one of our goals in this
class is to develop (still in a very naive level) the distinction between using the
language (when we use math to calculate the tip in a restaurant) and studying
that language to expand our horizons of thinking.
In order to do the latter (when we’re already adults–for children this is
easier), we must take a step back and talk about things that might look obvious
at first, or even unnecessary to discuss, simply in order to level the playing field
and make sure that we all (hopefully) understand each other and mean the same
thing when we use a certain ’phrase’. In this vein, the way I’m about to describe
our path might seem very opaque or pedestrian. However the promise is that if
you stick to this very sturdy banister you will be able to tread with confidence
to whichever new uncharted territories we reach.

2.1 Naive description of math

In math we deal with certain abstract objects, we give them names, or labels, we
consider various relations between them, and rules on how to manipulate them.
These rules are man-made, so to speak. There is very vast philosophy on what
extent this man-made universe corresponds to our physical reality in existence
(e.g. the connection between physics and math: does the physical reality build
or even constrain the kind of abstract mathematical structures that we can
come up with?) but the fact is that we don’t need an association with reality in
order to do math, and indeed, many branches of math have nothing to do with

4
reality at all. They are essentially studying the abstract structures that they
themselves have invented.
Imagine that you are playing the popular board game called monopoly. It
is loosely based on an economic system but at the end of the day it is a game
with rules who were invented by people and played by people. We could spend
time studying and exploring the various possibilities that arise as one plays the
game of monopoly. This would be one form of mathematical activity.
One of the main “mechanisms”, so to speak, of math making, is having an
arc to the story: we start with the given structure, and extract out of it certain
constraints that must hold given this structure. This is the basic mechanism of
logic, where for example if we say “this person is a student” and “students attend
lectures” we realize it must be the case that “this person attends lectures”. It is
this process that we will go through again and again, first describing the struc-
tures which we encounter and then “extracting” out of them new constraints.
Math can be done strictly with words (as I’ve been describing it so far) and
indeed this was mostly the approach taken in previous centuries. However, more
and more mathematicians realized that it is more efficient to use abbreviating
graphical symbols to lay down the abstract objects, structures and relations
of math. One writes down these graphical symbols on a piece of paper, on a
blackboard, or increasingly, into a computer, and this is a crucial way in which
we communicate about math nowadays, interlacing these graphical messages
within paragraphs of text which are supposed to introduce the rationale and
heuristics of what is really happening with the graphical symbols.
The graphical symbols are roughly organized as follows:
• A Latin or Greek single alphabet letter to denote the objects: a, b, c, · · · , x, y, z.
• Punctuation marks, “mathematical symbols” denote structures and rela-
tions: () , %, ∗, +, /, · · · , <, =.

• Of course we have the numbers themselves, which for our sake can be
thought of again as abstract objects but with honorary special labels:
1, 2, 3, . . . .

2.2 What is Calculus?

What has been described thus far could fit all of math in general. However, we
will venture into one particular area of math called “calculus”. Calculus means
in Latin “a pebble or stone used for counting”, and nowadays it is (mostly) the
word used to refer to a body of knowledge developed by Newton and Leibniz
around the mid 1600’s in order to study continuous rates of change of quantities
(for example instantaneous velocity of a physical object) or the accumulation
of quantities (e.g. the distance a physical object has traversed after a given
amount of time, given its acceleration).
While the main impetus to for the study of calculus comes from real life
questions, we will mostly fit it into our abstract framework so that we have a
“safe” way to deal with it without making conceptual mistakes.

5
 The main tool of calculus is the mathematical concept of a limit. The limit
has a stringent abstract definition using the abstract language, but intuitively it
is the end result of an imagined process (i.e. a series of steps) where we specify
the first few steps and imagine (as we cannot actually) to continue the process
forever and ask what would be the end result.

2.1 Example. Let us start with 1, then go to 12 , then 13 , 14 , and so on. Now
imagine that we continue taking more and more steps like this. What would be
the end result? The answer is zero, even though zero is never encountered after
any finite number of steps of this activity.
The concept of the limit is at the heart of anything we will do in this class,
and in particular, taking limits of sequences of numbers, where we have given
rules for generating the sequences of numbers (these are called functions).
Let us start now slightly more formally from the beginning.

3 Naive Naive Set Theory

Set theory is a (complex) branch of mathematics that stands at its modern
heart. Naive set theory is a way to present to mathematicians who are growing
up some of ideas of set theory in a way that obscures some of the hardest
questions of actual set theory and allow them to get going with math. What we
will do is naive “naive set theory”, which means we will very informally describe
what we need to setup a basic common language to allow us to study calculus
in an efficient way. The advantage will be that we will then have an entry point
to other branches of mathematics as well, such as actual naive set theory, but
also the beginning of analysis, topology or algebra. The best source for studying
more about set theory is Paul Halmos’ book of the same name [3].
Set theory is a corner of mathematics where the abstract objects are collec-
tion of yet other abstract objects (you will learn math often likes to be cute like
that). So a set is a collection of things, what they are–we don’t have to specify.
These things that a set contains could be sets themselves (this leads to nice
paradoxes). One graphical way to describe sets is using curly brackets, i.e. the
object (the set) which contains the objects a, b and c is graphically denoted by
{ a, b, c }. Note how one uses commas to separate distinct objects. While with
words it is obvious that the set that contains the objects a and b is the same
thing as the set that contains the objects b and a, in graphical symbols these
are a-priori two different things

{ a, b } versus { b, a }

but we declare that these two graphical symbols refer to one and the same
thing. Sometimes it is convenient to use three dots to let the reader know that
a certain number of steps should continue in an obvious fashion. For instance,
it is obvious that { a, b, . . . , f } really means { a, b, c, d, e, f }. Other times the
three dots mean a hypothetical continuation with no end, such as the case of

6
{ today, tomorrow, the day after tomorrow, . . . } where it is clear that there will
not be a final step to this process (setting aside fundamental questions about
compactness of spacetime and the universe), and that’s OK, since we actually
want to consider also hypothetical procedures. Such hypothetical procedures
are at the heart of limits, which lie at the heart of calculus. We say that such
sets, whose construction is hypothetical, have infinite size.
We often consider sets whose elements are numbers, as numbers for us are
currently just abstract mathematical objects, there should be no hindrance to
consider the set { 1 } if we also consider the set { a }, after all the graphical
symbols 1 or a are just labels. Using the bracket notation, we agree that there
is no “additional” meaning to the graphical symbol { a, a }, i.e., it merely means
the same thing as { a }.
Since sets themselves are abstract mathematical objects, we can just write
some letter, such as A or X or even a, to refer to one of them, rather than
enumerating its elements every time. Since what we mostly care about when
dealing with sets are their contents, i.e., the list of elements, it is convenient to
also have a graphical symbol to state whether an object (an element) resides in
a set or not. This is denoted via

a ∈ A means The object a lies in the set A.

a∈
/ A means The object a does not lie in the set A.

Note that using this graphical notation, it is clear that whatever appears to
the right of ∈ or ∈
/ must be a set.
Since all we know about sets is that they contain things, then we can “specify”
a set A by simply enumerating its contents, e.g. by using the curly-bracket
graphical notation. The special way to say that is using the equal = symbol:

A = { a, b, c } means A is the set { a, b, c }

means A is the set whose elements are a, b and c.

In many occasions, instead of describing a complicated situation in words, it

is many times easier to simply be able to refer to a set that contains absolutely
nothing at all (like an empty basket). This empty set is denoted by ∅ = { } (if
you think about it, there is only one such set, because since all we care about
sets is the list of things they contain, if both lists are equal (or empty) then
both sets are equal–so there is only one empty set).

a ∈ ∅ is always a false statement, regardless of what a is.

It is also convenient to have graphical notation that builds new sets out of
pre-given ones. For example:
• Union with the graphical symbol ∪: A ∪ B is the set containing all ele-
ments in either A or B. Example: A = { x, y }, B = { β,  }, A ∪ B =
{ x, β, y,  } (as we said, order doesn’t matter, and we are also not bound
to use Latin alphabet for labels of abstract objects.

7
 • Set difference with the symbol \: A \ B is the set containing elements in A
which are not in B. Example: A = { 1, 2 }, B = { 2 } gives A\B = { 1 }.
But B \ A = ∅.
• Intersection, with ∩: A ∩ B are all elements that are in both A and B.
Example: A = { 1, 2 }, B = { 2 } gives A ∩ B = { 2 } but A = { 1 } and
B = { 2 } gives A ∩ B = ∅.
Another important relation is not just between objects and sets, but also be-
tween sets and sets. Sometimes we would like to know whether all elements in
one set are also in another set. This is called a subset and is denoted as follows

A⊆B means a ∈ B whenever a ∈ A for any object a.

To verify that two sets A and B are the same (since what defines them is
the list of elements they contain), we must make sure of two things: A ⊆ B and
B ⊆ A. Hence

A=B means A⊆B∧B ⊆A

(note ∧ is the graphical symbol for the logic of ’and’, and we will also substitute
⇔ for ’means’).
3.1 Example. (A ∩ B) ∩ C = A ∩ (B ∩ C) (that is, the order of taking inter-
section does not change the end-result set). To really see why this is true, let
us proceed step by step. Suppose that x ∈ (A ∩ B) ∩ C. That means x ∈ A ∩ B
and x ∈ C. But x ∈ A ∩ B means x ∈ A and x ∈ B. Hence all together we
learn that the following are true: x ∈ A, x ∈ B and x ∈ C. This would be the
same end conclusion if we assumed that x ∈ A ∩ (B ∩ C). What we have learnt
is that x ∈ A ∩ (B ∩ C) whenever x ∈ (A ∩ B) ∩ C for any x. This is what we
said A ∩ (B ∩ C) ⊆ (A ∩ B) ∩ C means. This is half of the equality = statement.
The other half proceeds in the same way.

• An important structure is that of a product of sets. Given any two

sets A and B, we can form their product set A × B. This is a new
set whose elements consist of sets themselves. If A = { a, b, c, . . . } and
B = { x, y, z, . . . } then the set A × B is given by

{{ { a, 1 } , { x, 2 } } , { { b, 1 } , { x, 2 } } , { { c, 1 } , { x, 2 } } , . . .
. . . , { { a, 1 } , { y, 2 } } , { { b, 1 } , { y, 2 } } , { { c, 1 } , { y, 2 } } , . . .
. . . , { { a, 1 } , { z, 2 } } , { { b, 1 } , { z, 2 } } , { { c, 1 } , { z, 2 } } , . . . }

The point here is that in addition to form pairs, we also (arbitrarily) re-
fer to A as the first origin set, hence the 1 and B as the second origin
set, hence the 2, so that we are keeping track not only of the objects
in each pair, but from which origin set they’ve come from. In this way,
{ { a, 1 } , { x, 2 } } tells us immediately that a belongs to A and x belongs
to B.

8
 Because it is exhausting to write so many curly brackets, we agree on a
graphical notation that (,, ) means { { ,, 1 } , { , 2 } } for any , ∈ A
and  ∈ B. (,, ) is called an ordered pair. Example: { 1, 2 }×{ N, H } =
{ (1, N) , (2, N) , (1, H) , (2, H) }.
Clearly, (,, ) 6= (, ,), because { { ,, 1 } , { , 2 } } 6= { { ,, 2 } , { , 1 } }–
we can change orders within the curly brackets as we please, but we can’t
change move objects across curly brackets.

Another piece of notation that we want to discuss with sets involves their size.
The set { a } contains one object (no matter what a is) so we say its size (i.e.
the number of objects it contains is 1). Graphically we write two vertical lines
before and after the set in question to refer to its size, i.e.

|{ a }| = 1
|{ a, b }| = 2
...
|{ a, b, . . . , z }| = 26

Of course it is useful to agree that |∅| = 0 size ∅ contains no objects. When to

enumerate a set we must continue a hypothetical process to no end, for example,
the set { 1, 2, 3, . . . }, we say that the size of the set is infinite and graphically
we write the symbol ∞:

|{ 1, 2, 3, . . . }| = ∞

3.2 Definition. When a set is of size one, that is, when it has only one element,
we call it a singleton. Any set of the form { a } for any object a is a singleton.
In fact it is possible to turn the picture upside down, so to speak, and define
the numbers 1, 2, 3, . . . not as intrinsic abstract objects (how we used to think
about them so far) but as associated with a hierarchy of sets starting from the
empty one, with a natural association between the number we are naively used
to and the size of the constructed set:
1. Zero is associated with the set ∅, and we have |∅| = 0. We define zero to
be the empty set, making the empty set (rather than zero) the more basic
object and zero a derived object.
2. One is associated with the set { ∅ }. It is a singleton.
3. Two is associated with the set { ∅, { ∅ } }
4. etc.

You can find more about this in Halmos’ book.

3.3 Definition. A final piece of notation that we shall sometimes use is called
set-builder notation. This is a way to describe a subset B of a set A. Indeed,

9
our main way to describe sets (whether they are themselves subsets of other
sets or not doesn’t matter) is to enumerate their contents. For example:

A = The set of all people

and the subset

B = The set of all Americans

B is indeed a subset of A because all Americans are people (excluding pets

and so on). Indeed, B could be thought of as a sub-collection of the objects of
A, where all objects in the sub-collection obey a further constraint, namely, of
being American on top of being people. This is written graphically as

B = { a ∈ A | a is American }

In the above graphical notation, the symbol a is what is called a variable. It is

a generic label used to refer to any given element in A. To “build” B (or rather
to “imagine” all of its content) we must let a “run” through A.

4 Special sets of numbers

So far we have discussed sets as abstract collection of abstract objects. Let us
rely (in a sly way) on our pre-existing knowledge of objects which we refer to
as numbers, and imagine that we now start collecting them together into sets.
We can do whatever we want, so we can for example define a new set which
contains all of the following numbers

{ 1, 2, 3, 4, 5, . . . }

Note how now that we use the dots this means that the set has actually an
infinite number of elements. This is fine–in fact this is part of what makes
calculus interesting at all. This set above is called the natural numbers and is
denoted with the special graphical symbol N (blackboard N):

N = { 1, 2, 3, 4, 5, . . . } .

Using what we learnt, we know that 0 ∈

/ N yet 666 ∈ N.
4.1 Remark. In different geographical regions of the world, N may or may not
contain zero. In English-speaking environments, we mostly start N with 1, and
we shall do so with no exceptions until the end of the semester.
The next special set is the same thing but extended to the negative side:

{ 0, ±1, ±2, ±3, . . . } = { . . . , −5, −4, −3, −2, −1, 0, 1, 2, 3, 4, 5, . . . }

(recall the order of enumeration does not matter)

10
 This set is called the integers and is denoted by Z. Finally we would like to
include fractions as well
 
1 1 2 2 3 3 3
0, ±1, ± , ± , . . . , ±2, ± , ± , . . . , ±3, ± , ± , ± , . . .
2 3 3 4 2 4 5

i.e. any number that can be written in the form pq with p ∈ Z and q ∈ Z. In set
builder notation we would write
  
p 
p ∈ Z and q ∈ Z
q 

These are called the rationals and denoted by Q (for quotient). By the way,
rationals are not named such for being extra reasonable. The etymology is from
the word ’ratio’ which is also a quotient. There is also a decimal notation with
finitely many decimal digits after the point or periodic repeating, but let us skip
over that for now.
It turns out that there are certain numbers that exist (for example they
come from geometry or from physics) yet they are not in Q–they are irrational.
All of these numbers (which turn out to be the vast majority of all numbers,
where majority is meant in a certain sense, as we are trying to compare different
notions of infinity) are denoted by R and are called real numbers.
4.2 Remark. We already saw schematically (though not precisely) that there is
a way to build from the empty set all of N. There is also a concrete and precise
way (out of sets and manipulations of them) to construct Z out of N, Q out
of Z and R out of Q. We will not do so in this class as this material belongs
to a field of mathematics called analysis. If you are curious look at [4] under
Dedekind cuts.
4.3 Example. To give example for certain numbers in R \ Q, consider the
ratio between the circumference of a circle and its diameter. The ancient Greek
realized a while back that this number cannot be written in the form pq for some
p, q ∈ Z. To see this fact requires actually some work and preparation.

4.4 Example. The square root of the a number is the answer to the question √
“what number do we multiply√ √by itself to get what we started
√ with?”. So 4 = 2
because √2 × 2 = 4, i.e. 4 × 4 = 4. √ Can we express 2 in a simple way too?
Clearly, 1 = 1 because 1 × 1 = 1, so 2 must be somewhere between 1 and 2.
If we take the middle 1.5 = 32 we get 1.5 × 1.5 = 94 = 2.25, so 1.5 is already too
much. What√ about 1.4? √ 1.4 × 1.4 = 1.96, so that’s already too little! It turns
out that 2 ∈ / Q, i.e. 2 ∈ R\Q. To see this, assume otherwise. Then we have
p p
q × q = 2 for some p, q ∈ Z. If both p and q are even, we can divide both by
2 and get the same number, so let us assume we have done that so that now
2
2 = pq2 with p, q integers not both even. This is the same as p2 = 2q 2 . That
means that p2 is even, i.e. it is of the form 2x for some x ∈ Z. This implies
that p is even, i.e., it is of the form 2y for some y ∈ Z (if p were odd, it would
2
be of the form 2y + 1, and then p2 = (2y + 1) = 4y 2 + 4y + 1, which is odd!)

11
That means that actually p2 = 4y 2 for some y ∈ Z, i.e. 4y 2 = 2q 2 . But then
2y 2 = q 2 , that is, q 2 is even, which implies q is even (as before). So both p and
q are even!
So Q has some “holes”, and the purpose of using R is to have a set that
contains everything. Indeed the whole point of calculus is limits, and the whole
point of limits is to continue procedures hypothetically with no end. These
hypothetical procedures are precisely where we may suddenly find ourselves out
of Q.
It is that everything set, R, that we geometrically interpret as a continuous
line, i.e. we associated the lack of holes with the concept of continuum. That
is why in physics when we think of a continuous time evolution, for instance,
we model the set of possible times as R. When we think of the set of possible
heights a ball could take as it is thrown up the air, we model that set as R, since
we imagine physical space to be a continuum with no holes, and N, Z and Q
cannot be appropriate to describe the set of all possible physical outcomes. So
you should have in your mind a picture of an infinite straight continuous line
when you think of R.
As we have seen, we also can consider products of sets, and so R × R could
be considered the set of all possible pairs of continuum values, that is, a plane
of continuum. For convenience we write R2 instead of R × R. This set of pairs
should be geometrically pictures as an infinite plane. Physical space, everything
around us, is R3 (at least in Newtonian mechanics).

4.1 Intervals
Sometimes it is convenient to specify subsets of R, which are intervals. Given
any two endpoints a ∈ R and b ∈ R such that a < b, we define the following sets

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

The first of which is called the closed interval between a and b, the second of
which the open interval between a and b. The last two don’t have special names.
Sometimes it is useful to have the restriction only on one side to obtain a
half-infinite interval, that is, to consider the set of all numbers larger than a for
some a ∈ R. This is achieved in an efficient way via the ∞ symbol as follows

(a, ∞) := { x ∈ R | x > a }
(−∞, a) := { x ∈ R | x < a }
[a, ∞) := { x ∈ R | x ≥ a }
(−∞, a] := { x ∈ R | x ≤ a }

12
5 Functions
Given two sets A and B, we may wish to construct a rule, or a way to map,
objects from A onto objects from B. For instance, if A = { a, b, c } and B =
{ x, y, z } then we may wish to “send” a to x, b to y and c to z. This rule defines
what is called a function, also referred to as a map. We can think of various
other functions from A to B, each one is distinct if it has a different way to map
the objects around. For example, consider the function which “sends” a to z, b
to y and c to x. It is yet another possible way to map the objects of A onto
those of B.
If we write a table of A and B laid out together in perpendicular directions
B ↓;A → a b c
x
y
z
then we may fill in the interior of the table with objects of the product,
A × B:
A×B a b c
x (a, x) (b, x) (c, x)
y (a, y) (b, y) (c, y)
z (a, z) (b, z) (c, z)
Then we can think of the first function we described, i.e. that sending a to
x, b to y and c to z as a way to pair elements of A and B, that is, as a subset of
A × B, namely, the subset { (a, x) , (b, y) , (c, z) }. Looking at the table above,
we can identify the function by coloring in red these pairs of the table
A×B a b c
x (a, x) (b, x) (c, x)
y (a, y) (b, y) (c, y)
z (a, z) (b, z) (c, z)
Similarly, the second function, that mapping a to z, b to y and c to x can be
associated with the list of pairs { (a, z) , (b, y) , (c, z) } and on the table of A × B
would this subset of pairs, associated to those pairs colored red looks like:
A×B a b c
x (a, x) (b, x) (c, x)
y (a, y) (b, y) (c, y)
z (a, z) (b, z) (c, z)
However, not all sets of pairs constitute function. The point is that by
considering the concept of functions, we are interesting in giving a rule, or a
guide to go from A to B. That means in particular that this rule should be
unambiguous, so that we don’t get stuck trying to decide. So the following list
of ordered pairs { (a, x) , (b, y) , (c, z) , (a, z) }, represented in the table as
A×B a b c
x (a, x) (b, x) (c, x)
y (a, y) (b, y) (c, y)
z (a, z) (b, z) (c, z)

13
 does not constitute an appropriate function, because it tells us simultane-
ously to send a to x as well as to send it to z. So we don’t know where to send
a. Hence a function should send objects of a to only one place, which means
that the set of pairs encoding the function shouldn’t have two pairs with the
same first component and different second component (e.g. (a, x) and (a, z)).
The converse, however, is perfectly fine. That is, the function { (a, x) , (b, x) , (c, x) },
which sends all of the elements of A to the same spot in B, is a perfectly fine
function.
Finally, we want to make sure that we know at all where to map any given
element, so if there is some element of the origin set that doesn’t exist as the
first fact of some pair in the set of pairs, we won’t know where to map it. We
agree to exclude such scenarios from “appropriate functions”.
Given these considerations, we make the
5.1 Definition. Given two sets A and B, a function f from A to B, written as
f : A → B, is a set of unambiguous rules to associate objects of A with objects
of B, i.e. it is a subset of pairs, i.e. of A × B, such that no two pairs have
the same first component and different second component, and such that all
elements of the origin set are covered as one enumerates all first components of
all pairs. The set A is called the domain of f , the set B is called the co-domain
of f . Sometimes one refers to the graph of f as the that subset of A × B which
specifies it.
Be ware the discrepancy between the intuitive meaning of the word graph
(we think of a geometric object) and the technical meaning given above (an
abstract subset of pairs). This will come up again and again in math, the
distinction between intuitive meanings of words from our daily lives and their
actual technical definition.
Let us introduce some graphical notation which will be used throughout the
course for functions:

1. Given a function f : A → B, suppose that the element a ∈ A gets mapped

to x ∈ B.
(a) Arrow notation: We write

a 7→ x

or

A 3 A 7→ x∈B

if we want to be slightly more explicit.

(b) Braces notation: We write

f (a) = x

(c) Subscript notation: We write

fa = x

14
 This is mainly used when A ∈ N, or when A is of the form A = T ×X
for two sets T and X, and instead of writing f ((t, x)) for some t ∈ T
and x ∈ X one writes ft (x).
5.2 Example. If the domain of a function, A, is empty, i.e. A = ∅, then
there are not many choices (since there is nothing to map) and so it suffices to
write f : ∅ → B (for any B), and there is just one unique function with this
domain. Similarly, if B contain only one element, then there is again nothing
to desscribe, because we have no choice. E.g. A = N, B = { / }, then we
know what f : A → B does. It merely converts any number into a /. This can
graphically be written as
f (n) = / for any n ∈ N
5.3 Example. If both the domain and co-domain are the same set, f : A → A,
then there is a special function which sends each element to itself. This is known
as the identity function, and is denoted as 1 : A → A for any set A. We have
1:A → A
a 7→ a
When the domain or codomain are rather large (think infinite), e.g. one
of our special sets of numbers, it sometimes becomes easier to give a formula
for what f does rather than specify one by one how it acts on each different
element, or enumerate a list of pairs (indeed, that would be literally impossible
for infinite sets). Consider the function
f :N → N
which adds 1 to any given number. So 1 7→ 2, 2 7→ 3, 3 7→ 4 and so on. An
easy way to encode that is to use variables, i.e. objects which are placeholders
for elements of a certain set1 . A variable is thus any object we could pick from
a certain set. For example, a variable n in the natural numbers is any choice
n ∈ N. We could have n = 5, n = 200 or n = 6000000 (but not n = −1, since
N contains only strictly positive integers). The point is, it is convenient not to
specify which element it is and work with a generic unspecified element. Once
we have a variable, we can easily write down the action of f as a succession of
algebraic operations, i.e. a formula in that variable:
f (n) = n + 1 for any n ∈ N
This specifies in a formula the same verbal description we gave earlier. The
variable is also called the argument of the function.
5.4 Remark. The most common and efficient way to describe a function is to
write two lines of text:
f :A → B
a 7→ some formula of a
1 We already encountered variables when we discussed set-builder notation

15
where the first line tells us the label of the function (in this case f ), the domain,
i.e. the origin set A, the codomain, i.e. the destination set B, and the second
line tells us how to map each object of A into B. In this case the second line is
in the form of a formula, but one could just as well list all possible mappings of
elements in A.
We can quickly run into problems with math, just as we would with natural
language. It doesn’t make sense to write “dry rain” even though we can easily
juxtapose the two words together. In the same way, if we try to write down

f :N → N
f (n) = n − 1 for any n ∈ N

we quickly realize this makes no sense! The reason is that for certain n ∈ N,
namely, for 1, if we apply the formula, we actually land outside of N, because 0 ∈
/
N! That means that the formula-way of describing functions can be dangerous,
that is, it can quickly lead us to write down nonsense. This is a manifestation of
the fact that just because we have a language with rules doesn’t mean that every
combination of any phrase will make sense. We still must be careful, especially
as we build shortcuts.
Here is another example:

f :N→ R
√
n →
7 n

Note that the same formula√ would not make sense with R replaced by N or even
Q, as we just learnt (e.g. 3 ∈
/ Q)!
5.5 Definition. When a function f : A → B has its domain A = N, i.e. the
natural numbers, one often calls that function a sequence and one writes its
argument in subscript notation, i.e.
√
fn = n

We can write many complicated formulas. For example, we can write what
is known as a piecewise formula:

a:R → R
(
x x≥0
x 7→
−x x<0

this means that before we apply the formula we must verify some conditions.
Sometimes it is helpful (though usually not unambiguous) to also sketch a
function. A sketch is the graphical arrangement of all possible values it could
take given all possible inputs. We have already seen how to do this in a rather
rudimentary way using the colorings of the graph of a function within the table
of A × B above. A sketch of a function is thus a way to geometrically draw the
graph of the function.

16
Figure
( 1: A plot of the graph of the function { 1, 2 } → { 1, 2 } given by
1 7→ 2
.
2 7→ 1

5.6 Example. f : { 1, 2 } → { 1, 2 } given by 1 7→ 2 and 2 7→ 1. Then

A × B = { (1, 1) , (1, 2) , (2, 1) , (2, 2) } and the graph of f may be described
as { (1, 2) , (2, 1) }. This can also be drawn graphically as in Figure 1.

5.1 Functions from R → R

First some general notions, which rely on the fact that for any two numbers
a, b ∈ R, we may compare them, i.e. we necessarily have exactly one of the
following: a < b, or b < a or a = b (this is not the case for any two objects
from any other set, it relies on many properties of R which we assume but leave
implicit).
5.7 Definition. A function f : R → R is called monotone increasing iff when-
ever a, b ∈ R and a ≤ b then f (a) ≤ f (b). It is monotonically decreasing iff
f (a) ≥ f (b). One can add the qualifier strictly to change ≤ into < and ≥ into
>.
5.8 Example. Take f : R → R given by f (x)
= x2 for any x ∈ R. The graph
of f is the subset of R×R given by pairs x, x2 for any x ∈ R, i.e.

x, x2 ∈ R2  x ∈ R



Since R may be pictured as an infinite line, R×R = R2 , the set of all pairs, should
be pictured as an infinite two-dimensional plane, in which case the graph of a
function f : R → R is a curve in that plane. The fact it is a curve, and nothing
else, is related to the fact that no two pairs have the same first component.
The shape of that curve is what we care about when drawing the graph of the

function, which is the sketch of that function. In the particular case of x, x2
for any x ∈ R, the shape of the curve is that of the familiar parabola as in
Figure 2.
5.9 Definition. The function

a:R → R
(
x x≥0
x 7→
−x x<0

17
 Figure 2: The parabola x 7→ x2 .

Figure 3: The sketch of the graph of the absolute value.

which we already described above to introduce the piecewise notation is called

the absolute value function. Its graph can be sketched as in Figure 3. One way
to figure out how to draw these functions is to draw a few pairs of points on
the plane and then extrapolate. For example, if we want to start plotting the
absolute value function, we start with a few points from the formula:
0 7→ 0
1 7→ 1
−1 7→ 1
2 7→ 2
−2 7→ 2
which correspond to the pairs (0, 0) , (1, 1) , (−1, 1) , (2, 2) , (−2, 2), and then we
mark points on a grid at these coordinates. In order to extrapolate it is useful
to know of a few building blocks of basic functions.
5.10 Remark. The absolute value function obeys the following properties: for
any two numbers x, y ∈ R:
|xy| = |x| |y| (1)
5.11 Definition. A function f : R → R is called bounded iff there is some
M ∈ R with M ≥ 0 such that
|f (x)| ≤ M
for all x ∈ R.
5.12 Example. The constant function f : R → R; x 7→ c (for some constant
c ∈ R) is bounded. Indeed, one can pick M := |c|.
5.13 Example. The parabolic function f : R → R; x 7→ x2 is not bounded.
5.14 Example. The parabolic function restricted to a finite interval is bounded.
Foe example, f |[−5,5] : [−5, 5] → R; x 7→ x2 . Indeed, one can pick M := 25.

18
Figure 4: The constant function R 3 x 7→ c ∈ R for all x ∈ R, for some constant
c ∈ R.

5.1.1 Basic functions and their shapes

5.15 Definition. Let c ∈ R be any given number. The constant function
f : R → R associated to c sends all elements of its domain to c:

f (x) = c (x ∈ R)

Graphically this function looks like a flat horizontal line at the height c as in
Figure 4.
5.16 Definition. Let a, b ∈ R be given (note this is short-cut notation for
a ∈ R and b ∈ R). Then the linear function f : R → R associated with a and b
is given by

f (x) = ax + b (x ∈ R)

Graphically this function looks like a straight line at an angle. The parameter
b sets the line’s height when it meets the vertical axis, and the number − ab is
where it meets the horizontal axis:

5.17 Definition. Let a, b, c ∈ R be given. Then the parabolic function f : R →

R associated with a, b, c is given by

f (x) = ax2 + bx + c (x ∈ R)

One can of course go on with these to any highest power of x, e.g. f (x) =
x100 which looks like this:

5.1.2 Trigonometric functions

We next want to introduce the trigonometric functions. These are special func-
tions because they do not have special algebraic formulas which define their
action. There are two possibilities: either we define them through a limiting
process (to be done later on) or we can define them pictorially through a geo-
metric picture. For now let us do the latter and just draw some pictures.
Let us draw a circle of radius 1 on the plane R2

19
 We know the entire circumference of the whole circle is 2π where π is some
special irrational number equal to about 3.14 which we cannot write out explic-
itly. Let us traverse, along the circle, starting from the point (1, 0) on the plane,
an arc of arc-length α, for some 0 ≤ α < 2π, and draw a right triangle whose
base is along the horizontal axis, has a point on the circle after arc-length α and
another vertex at the origin

The sinus function is defined as the height of this triangle (as a function of
α), and the cosine function is defined as the base length of this triangle (as a
function of α). Since we are on a circle, it makes sense to agree that after α > 2π
the sine and cosine functions assume the same values as if we were calculating
them with α − 2π, and similarly for α < 0, so that we get a definition for the
whole of R of a periodic function. Things to note:
1. cos (0) = 1, sin (0) = 0

2. cos π2 = 0, sin π2 = 1.

3. cos is decreasing on (0, π), sin is increasing on − π2 , π2 .

20
The cosine looks like this:

and the sine like this:

5.2 Special sets associated with a function

Given a function f : A → B, we already encountered the following sets associ-
ated with it:
1. The domain of f , which is just A.

2. The co-domain of f , which is just B.

3. The graph of f , encoding the same information as f itself, which is the
subset of A × B given by

graph (f ) = { (a, f (a)) ∈ A × B | a ∈ A }

We define a few more sets associated to a given function f :

5.18 Definition. The image of a function f : A → B is the subset of B given
by the following

im (f ) := { b ∈ B | There is some a ∈ A such that f (a) = b }

= { f (a) ∈ B | a ∈ A }

Note: we will not use the word “range” in this course, as it is ambiguous and
sometimes conflated with either co-domain or image.

21
5.19 Definition. Let f : A → B be a function between two sets A and B and
let S ⊆ A be a given subset. Then the image of S under f is the following
subset of B:
f (S) := { b ∈ B | There is some a ∈ S such that f (a) = b }
= { f (a) ∈ B | a ∈ S }
Note that in this graphical notation we use the braces notation on a whole set
rather than an object, and the result is then a set, rather than an object! This
notation can be confusing. Using this notion we can identify
f (A) = im (f )
5.20 Definition. Let f : A → B be a function between two sets A and B and
let S ⊆ B be a given subset. The pre-image of S under f is the following subset
of A:
f −1 (S) = { a ∈ A | f (a) ∈ S }
Note the introduction of a new notation: for the pre-image of a function f ,
we use the graphical symbol f −1 . Again this is a funny notation in the sense
that we plug in a set into f −1 and get back a set. Despite the notation, f −1 is
not a function. Of course for a function f : A → B we have f −1 (B) = A by
definition.
5.21 Definition. A function f : A → B is called surjective if im (f ) = B. That
means there are no elements of B left “uncovered” by f .
5.22 Definition. A function f : A → B is called injective if
 −1 
f ({ b }) ≤ 1 (b ∈ B)
which means that every point of B gets covered at most once (if not never) by
f . In other words, no two elements of A get sent to the same element of B, that
is, every destination point has a unique origin point, if it is in the image of f .
5.23 Definition. A function f : A → B is called bijective if it is surjective and
injective. Bijective functions should be thought of as reversible, because they
don’t lose information.
5.24 Example. The constant function f : R → R, x 7→ 5 for all x ∈ R
is not surjective, since im (f ) = { 5 } 6= R. It is not injective because while
f −1 ({ x }) = ∅ for all x 6= 5 and |∅| = 0, f −1 ({ 5 }) = R and |R| = ∞ > 1.
5.25 Example. The linear function f : R → R, x 7→ 5x for all x ∈ R is
bijective, because
im (f ) = { 5x | x ∈ R }
= R
−1
({ x }) = { y ∈ R | 5y = x } = 51 x which is of size one.

and f
What about the absolute value function?

22
5.3 Construction of new functions
5.26 Definition. Given two functions f : A → B and g : B → C, we define
their composition, denoted as g ◦ f , as a new function A → C given by the
formula

(g ◦ f ) (a) := g (f (a)) for any a ∈ A

which first applies f , and then g (considered as rules), all together passing
through B but ultimately producing a route (i.e. a function) from A → C. We
also can compose a function itself, if its co-domain is equal to its domain: if
f : A → A then

fn := f ◦ f ◦ · · · ◦ f
| {z }
n times

for any n ∈ N.
5.27 Example. If f : R → R is given by x 7→ sin (x) then f ◦ f : R → R is the
function given with the formula x 7→ sin (sin (x)) for any x ∈ R (we don’t ask
what that means geometrically).
5.28 Definition. A function f : A → B is called left-invertible iff there is some
other function g : B → A such that g ◦ f = 1 where 1 : A → A is the identity
function discussed above. Conversely, f is called right-invertible iff there is some
other function h : B → A such that f ◦ h = 1 where 1 : B → B is the identity
function. If f is both left and right invertible we call it invertible, and then the
left and right inverse are equal and unique g = h, in which case we denote that
inverse by f −1 = g = h (not to be confused with the pre-image notation, and
also not to be confused as an algebraic operation–we are not dividing anything
by anything else, this is merely graphical notation), so that by definition

f ◦ f −1 = 1B
−1
f ◦f = 1A

These last two equations are interesting, because they tell us that functions
themselves (rather than objects, numbers, or sets) are equal. But since we have
a precise way to think of functions as sets themselves, this is perfectly fine. Also
we use the short-hand notation of 1A to denote the (unique) identity function
A → A for any set A.
What kind of relationship is there between left or right invertibility and
injectivity surjectivity?
5.29 Definition. Given any function f : R → R, we can quickly define a new
function by an algebraic formula on f itself. For example, the function f + 3
has the formula

(f + 3) (x) = f (x) + 3 (x ∈ R)

23
Sometimes these shortcuts don’t always make sense and one has to be careful,
for example, with f1 . Other times the notation itself becomes ambiguous, for
2
example, f 2 could either mean f (f (x)) for any x ∈ R or it could mean (f (x))
for any x ∈ R. So in such cases one has to write out in words what one means.
Another possible confusion is with f −1 . Usually it means either the pre-image
or the (unique) inverse of a function, as defined above, if it exists. It usually
does not mean the function
1
x 7→ (x ∈ R)
f (x)

for which one usually uses the notation f1 instead.

We can also make formulas with two or more functions, whenever that makes
sense. So if f : R → R and g : R → R, then by f + g (or f − g, f g, fg , etc) we
mean a new function R → R whose formula is

x 7→ f (x) + g (x) (x ∈ R)

5.30 Definition. Given any function f : A → B and a subset X ⊆ A, we

define the restriction of f to X, denoted by f |X : X → B, as

f |X (a) := f (a) (a ∈ X)

So f |X and f have the same formula, but the former is restricted to act on a
smaller subset. This is sometimes a useful notion when considering the proper-
ties of functions, some of which may only hold on a subset but not on the whole
domain.
5.31 Example. Pick any number a ∈ R which is strictly positive, a > 1.
Consider the function expa : R → R which is given by

expa (x) := ax (x ∈ R)

If x = 0 the result is 1 (by convention). When x = n for some n ∈ N, we

know how to perform this operation. We merely raise a to the power n, i.e.
we compute a × a × · · · × a. When x = −n for some n ∈ N, we know that this
| {z }
n times
means to computer a × a × · · · × a, and then we must take the reciprocal of that
| {z }
n times
number, that√is, a1n . If x = n1 for some n ∈ N, then this should be the n-th root
of a, that is n a, and if x = − n1 for some n ∈ N, we get √ 1
n a . Hence all together
p
if x = q for some p, q ∈ Z we get
√
q
expa (x) = ap

which still doesn’t

√ tell us a lot, because as we already saw, we cannot write out
explicitly what 2 is, for instance,
√ but it at least gives us some constraint
√ √ on
what the answer should be (i.e. 2 should be that number such that 2 2 = 2).

24
It turns out that even if x ∈ R\Q one could proceed, via a limit procedure that
makes the sketch of expa look smooth when plotted on R (using the basic fact
that any element x ∈ R\Q has an element y ∈ Q arbitrarily close to it, so
intuitively, we define expa (x) as expa (y) (which we know how to compute)
where y ∈ Q is arbitrarily close to x ∈ R\Q).
As define, expa : R → R is not surjective, and hence not bijective. Indeed,
it is always larger than zero. That is, we have
im (expa ) = (0, ∞)
So we change the definition expa : R → R by modifying the co-domain to be
(0, ∞):
expa : R → (0, ∞)
to be defined by the same formula as before, and get a surjective function.
Actually expa is also injective. Indeed, we can verify this by verifying that if
expa (x) = expa (y) for some x, y ∈ R, then ax = ay (in HW1 you learn this is
one possible criterion for injectivity). Divide both sides of the equation by ay
to get ax a1y = 1. The basic rules of exponentiation imply now that ax−y = 1.
However, we know that only when exponentiating some number which is strictly
larger than 1 to power zero we get back 1, so that x − y = 0 necessarily. So
that means x = y and hence expa is indeed injective. Since expa : R → (0, ∞)
is injective and surjective, i.e. bijective, you learn in HW1 that means it has a
unique inverse exp−1a : (0, ∞) → R. This inverse is called the logarithm with
base a, and is denoted by loga : (0, ∞) → R.
5.32 Exercise. Both cos and sin functions when defined from R → R are not
injective nor surjective. However, one may modify both domain and co-domain
to make them bijective. How?

6 Limits
At the heart of calculus is the notion of a limit. The limit is a way to consider
a hypothetical process that cannot actually be carried out but whose result still
may have meaning. We have already encountered such hypothetical processes
when we first considered the set
N ≡ { 1, 2, 3, . . . }
where the dots mean the hypothetical process of continuing the list with no end.
Since this is not technically possible, this is merely a hypothetical notion. And
yet it is useful for us to collect together in one set all possible natural numbers,
which really just means that whatever large number one can think of, it is part
of N.
Yet another example that we already encountered was the hypothetical result
of a process of enlisting fractions with increasing denominators, i.e., the sequence
1 1 1 1
1, , , , , . . .
2 3 4 5

25
1.0

0.8

0.6

0.4

0.2

2 4 6 8 10

1
Figure 5: A plot of the graph of N 3 n 7→ n ∈ R.

which has no end. Since it has no end, the final result of this process is merely
hypothetical. And yet intuitively it is clear that the end result will be zero,
which really just means, whatever small number you can think of, one can find
a step in this process which will be smaller than that given number.

6.1 The notion of a distance on R

We say the distance between a pair of numbers is the magnitude of their differ-
ence, i.e. we only care about how far apart they are, but not in which direction.
Hence we define the distance function on pairs of numbers
d : R2 → [0, ∞)
(x, y) 7→ |x − y| for any (x, y) ∈ R2
The distance function has a number of important properties:
1. It is symmetric: d (x, y) = d (y, x) for any x, y ∈ R, because |a| = |−a| for
any a ∈ R.
2. Distance zero implies identity: d (x, y) = 0 for some x, y ∈ R implies
x = y. Indeed, the absolute value function, as we defined it, only takes
the value zero at zero. So |a| = 0 implies a = 0.
3. It obeys the so-called triangle inequality: for any three numbers x, y, z ∈ R
we have d (x, y) ≤ d (x, z) + d (z, y). The way to convince yourself that

26
 this is really true is to divide the analysis into cases. The easiest case is
that all three numbers are different and obey x < z < y. Then we have

d (x, y) ≡ |x − y|
(From the definition of the absolute value, because y > x)
= y−x
= y−z+z−x
(From the definition of the absolute value, because y > z and z > x)
= |y − z| + |z − x|

the other cases proceed similarly.

An important property of this distance is that it allows us to pinpoint exactly
6.1 Claim. Iff d (x, y) < α for some x, y ∈ R and some strictly positive number
α > 0 then we have

−α < x−y < α.

Indeed, from the definition of the absolute value, we know that d (x, y) ≡ |x − y|
is equal to x − y if x > y and y − x if y > x. Hence, either x > y, |x − y| = x − y,
and then x − y < α, or x < y, |x − y| = − (x − y), and then because α > 0 and
x − y < 0 we get x − y < 0 < α or just x − y < α. This shows you that the first
inequality holds. The second one proceeds similarly.
6.2 Claim. The distance function is translation invariant. That is: d (x, y) =
d (x − z, y − z) for any three numbers x, y, z ∈ R. To see this, we write out the
definition

d (x, y) = |x − y|
= |x − y + z − z|
= |x − z − (y − z)|
= d (x − z, y − z)

One can also show that

6.3 Claim. (Reverse triangle inequality) We have for any x, y ∈ R,

d (|x| , |y|) ≤ d (x, y) .

To see this, note that

d (|x| , |y|) ≡ ||x| − |y||

and

|x| = |x − y + y|
(Regular triangle inequality)
≤ |x − y| + |y|

27
so we have
|x| − |y| ≤ |x − y|
By symmetry (running the same argument after having exchange x with y) we
also have
|y| − |x| ≤ |x − y|
which is equivalent to (by multiplying the inequality by minus one):
|x| − |y| ≥ − |x − y|
so we conclude by Claim 6.1 that
||x| − |y|| ≤ |x − y|
which is what we wanted to show.

6.2 Limits of sequences–functions from N → R

More generally, let a : N → R be some function. Such functions whose do-
main is N have a special name: they are called sequences. The example above
corresponds to the sequence given by the formula
1
an = (n ∈ N)
n
but in principle this could be any formula. Here is another example
n
an = (−1) (n ∈ N)
if we list the values of this sequence we see the first few are equal to
−1, 1, −1, 1, −1, 1, −1, 1, . . .
and intuitively it is clear that this list does not “tend” to anything as we go
forward towards infinity, because it keeps jumping up and down between −1
and 1. Here is yet another example:
an = n2 (n ∈ N)
and again we list some of the first few elements
1, 4, 9, 16, . . .
the items on this list grow very quickly. So if “imagine” what would happen if
we continued to take more and more steps of this process, one possible way to
phrase the result would be to say that it tends to infinity. By that one means
that whatever (large) number one could come up with, there is a sufficient
number of steps of this process that may be taken so as to surpass this large
number.
So we have encountered so far three possible behaviors of a sequence a : N →
R:

28
 1. The sequence “converges” to some number c ∈ R as we plug in larger and
larger arguments, as was the case in the first example.
2. The sequence keeps jumping back and forth no matter how far we go.
3. The sequence keeps growing with no bound–it diverges.

We formalize these considerations in the following

6.4 Definition. A sequence a : N → R is said to have a limit L ∈ R iff for any
strictly positive number that one could pick, δ ∈ R with δ > 0, but as small as
one wants, there is some number N ∈ N (this number may depend on δ) such
that for all n ∈ N obeying n ≥ N , the following condition holds:

d (a (n) , L) < δ

That is, the distance between a (n) and the number L becomes as small
as one wants–one merely has to go far enough into the sequence, and how far
depends on how small the distance we ask for. Another way to say this is that
a (n) converges to L as n → ∞. The point about this concept is that the
distance between a (n) and L becomes smaller and smaller and smaller. If the
distance is “small”, but remains fixed as we enlarge n, the notion does not apply.
6.5 Remark. It is not possible that a (n) converges to L as n → ∞ and also
a (n) converges to L0 as n → ∞ if L 6= L0 .

Proof. We have for all n ≥ max (N, N 0 ), N being the threshold of distance
δ > 0 for the convergence of a to L and N 0 that to L0 ,

d (L, L0 ) ≤ d (L, a (n)) + d (a (n) , L0 )

≤ 2δ

That means that the distance between L and L0 can be made arbitrarily
small, that is, they are equal.

6.6 Definition. A sequence a : N → R is said to go to ∞ (respectively −∞), to

diverge to ±∞ iff for any strictly positive number that one could pick, M ∈ R,
there is some number N ∈ N (this number may depend on M ) such that for all
n ∈ N obeying n ≥ N , the following condition holds:

a (n) ≥ M

(respectively a (n) ≤ −M )
6.7 Definition. A sequence a : N → R is said to have no limit (one says the
limit does not exist), if there is no L ∈ R to which it converges, and it does not
go to either ∞ or −∞.
Different notations for this are:

29
 1. The limit notation: If L is a limit of a, then we write
lim a (n) = L
n→∞

2. or sometimes
lim a = L

3. or sometimes
a (n) → L (n → ∞)
4. and when it is not important what L is, but only that there is some L like
that, we write that lim a exists.
5. When a goes to infinity, we write
lim a (n) = ∞
n→∞

and say that the limit diverges.

6.8 Example. Going back to our initial example of N 3 n 7→ n1 ∈ R, let us see
why this converges to zero as n → ∞ according to the definition. We have
d (a (n) , 0) ≡ |a (n) − 0|
 
1
=  
n
(n > 0)
1
=
n
Let us pick some number δ > 0. If we want to arrange that n1 < δ, we equiv-
alently need n > 1δ . So if we take N to be the smallest integer larger than 1δ ,
then n ≥ N implies that n > 1δ !
6.9 Example. Take N 3 n 7→ n−1 n+1 ∈ R. Does this converge or diverge? Let us
list the first few elements
1 1 3 2 5 3 7 4 9
0, , , , , , , , , , . . .
3 2 5 3 7 4 9 5 11
this actually does converge to 1. The reason being that for n being very large,
i.e. much larger than 1, n − 1 and n + 1 are not very different, so their quotient
is very close to 1. To really see this, calculate
   
n−1 n − 1 
d ,1 ≡   − 1
n+1 n+1
 
n − 1 − n − 1
=   
n+1 
2
=
n+1
so if we want this distance to be arbitrarily small, we need to pick n such that
2 2
n+1 < δ or n > δ − 1.

30
 √ √
6.10 Example. Take N 3 n 7→ n + 1 − n ∈ R. Let us write out the first
few elements of this sequence (I used a computer to give approximate values of
the square roots):
0.41, 0.31, 0.26, 0.23, 0.21, 0.19, . . .
it seems to be going down, but does it converge to zero? The answer is yes,
again because as n is very large, the difference between n + 1 and n becomes
insignificant (essentially because n is much larger than 1!) So we try to calculate
√ √
√ √ 
d n + 1 − n, 0 =  n + 1 − n
= (The square root is monotone increasing, so this is positive)
√ √
= n+1− n
a2 − b2
 
Use the identity a − b =
a+b
n+1−n
= 2
(n + 1) + n2
1
= 2
2n + 2n + 1
Use 2n2 + 2n + 1 ≥ 2n2 for any n

1
≤
2n2
and we get the same story (we can make this arbitrarily small by taking n
arbitrarily large).
6.11 Example. Consider the sequence N 3 n 7→ n ∈ R. Can we show it goes
to infinity? Trivially, because for any big number that we can choose, M ∈ R,
there is some N ∈ N such that all n ≥ N will obey n ≥ M . In particular, take
N to be the first integer larger than M .
6.12 Claim. If a : N → R and b : N → R are two sequences which are equal
except for a finite number of elements, then their limit behavior is identical.
That is, if

a (n) = b (n)

for all n ≥ N , for some N ∈ N, then lim a = lim b if this exists (that is, either
both limits exist and converge to the same finite number, or both limits do not
exist, or both limits diverge to infinity or minus infinity).

Proof. Assume for simplicity that lim a exists and converges to a finite number
(the other cases being similar). Then we want to show lim b exists and equals
lim a. To show that, let us assume that Na (δ) ∈ N is that threshold of a such
that if n ≥ Na (δ) then

d (a (n) , lim a) < δ.

31
 Then if we pick n ≥ max ({ N, Na (δ) }) we have both d (a (n) , lim a) < δ and
a (n) = b (n), which implies

d (b (n) , lim a) < δ.

Hence b converges and lim b = lim a.

6.13 Claim. (Algebra of limits) If a, b are two sequences N → R which both have
finite limits, then lim (a

+ b) = lim a + lim b, (lim a) (lim b) = lim (ab). Also, if
lim b 6= 0, then lim ab = lim a a
lim b with the understanding of b being a sequence
that might be defined only after a finite number of terms.

Proof. Let us assume that both a and b have finite limits L1 and L2 . Let us
take the thresholds N1 (δ) , N2 (δ) ∈ N for each of these limits. That means
that given any δ > 0, if n ≥ N1 (δ) then

d (a (n) , L1 ) ≤ δ

and if n ≥ N2 (δ) then

d (b (n) , L2 ) ≤ δ

Let us define N (δ) := max ({ N1 (δ) , N2 (δ) }) (i.e. the largest of the two
thresholds, so that if n ≥ N (δ) then automatically both n ≥ N1 (δ) and
N2 (δ)). Then we have, using Claim 6.2

d (a (n) + b (n) , L1 + L2 ) = d (a (n) − L1 , L2 − b (n))

(Use triangle inequality with the third point being zero)
≤ d (a (n) − L1 , 0) + d (0, L2 − b (n))
(Use translation invariance again invariance, twice)
= d (a (n) , L1 ) + d (b (n) , L2 )
1


Hence, if n ≥ N 2δ , then

1 1
d (a (n) + b (n) , L1 + L2 ) ≤ δ+ δ=δ
2 2
which proves the first statement about the sum. To get the product, we use

32
Remark 5.10 together with the triangle inequality below to get

d (a (n) b (n) , L1 L2 ) = |a (n) b (n) − L1 L2 |

= |a (n) b (n) − L1 b (n) + L1 b (n) − L1 L2 |
= |(a (n) − L1 ) b (n) + L1 (b (n) − L2 )|
≤ |a (n) − L1 | |b (n)| + |L1 | |b (n) − L2 |
= |a (n) − L1 | |b (n) − L1 + L1 | + |L1 | |b (n) − L2 |
≤ |a (n) − L1 | (|b (n) − L1 | + |L1 |) + |L1 | |b (n) − L2 |
|L1 | (|a (n) − L1 | + |b (n) − L2 |) + |a (n) − L1 | |b (n) − L1 |
=
q 
2
so if we pick n ≥ N |L1 | + δ − |L1 | , then

q q  q 2
2 2 2
d (a (n) b (n) , L1 L2 ) ≤ |L1 | |L1 | + δ − |L1 | + |L1 | + δ − |L1 | + |L1 | + δ − |L1 |
 q  q 
2 2
= 2 |L1 | + |L1 | + δ − |L1 | |L1 | + δ − |L1 |
 q  q 
2 2
= |L1 | + |L1 | + δ |L1 | + δ − |L1 |
2 2
= |L1 | + δ − |L1 |
= δ

which proves the second statement about the product.

To prove the last statement, we assume lim b 6= 0. Consider first the case
1
that b (n) 6= 0 for all n ∈ N. Then we know what the sequence

b means, and1by
a


the multiplication law we just proved, we know that lim b = (lim a) lim b ,
so we only have to understand that
 
1 1
lim =
b lim b

To that end,
   
1 1  1 1 
d , =  b (n) − L2 

b (n) L2
 
 L2 − b (n) 
= 
 b (n) L2 


d (L2 , b (n))
=
|b (n)| |L2 |

33
 Now, we know that b (n) → L2 as n → ∞ and L2 6= 0 by hypothesis. That
means that

d (L2 , 0) ≤ d (L2 , b (n)) + d (b (n) , 0)

or

d (b (n) , 0) ≥ d (L2 , 0) − d (L2 , b (n))

Since L2 6= 0, d (L2 , 0) ≡ |L2 | is a nice strictly positive number. Since

b (n) → L2 as n → ∞, d (b (n) , L2 ) can be made arbitrarily small by taking
n above a certain threshold. For example, assume that n ≥ N2 21 d (L2 , 0) .

Then d (L2 , b (n)) ≤ 21 d (L2 , 0) or −d (L2 , b (n)) ≥ − 21 d (L2 , 0) so that we

can conclude all together d (b (n) , 0) ≥ 21 d (L2 , 0) or taking the reciprocal,
1 2 1


d(b(n),0) ≤ d(L2 ,0) . Hence we find for n ≥ N 2 2 d (L2 , 0) ,
 
1 1 d (L2 , b (n))
d , ≤ 2
b (n) L2 d (L2 , 0) |L2 |
2
= 2 d (L2 , b (n))
|L2 |

The final conclusion is that if we now take our new threshold to be

( ! )!
  2
1 |L2 |
max N2 d (L2 , 0) , N2 δ
2 2

for any δ > 0 and we take n to be larger than that threshold, we can conclude
that  
1 1
d , ≤δ
b (n) L2
This way of making the proof by assuming that b (n) 6= 0 for all n ∈ N also
tells us how to proceed in the other case. Indeed, we have just shown that
due to lim b 6= 0, there is a certain threshold, above which, b (n) 6= 0. So even
if that’s true in the beginning, Claim 6.12 shows it doesn’t matter.

6.14 Claim. (The Squeeze Theorem) If a : N → R, b : N → R and c : N → R are

sequences such that for each n ∈ N, a (n) ≤ b (n) ≤ c (n) and lim a = lim c then
lim a = lim b = lim c.

Proof. For convenience let l := lim a = lim c. Then for any δ > 0

|b (n) − l| < δ

and by Claim 6.1 this is equivalent to

l−δ < b (n) < l + δ

34
 However, we have a → l and c → l, so for any δ > 0 we can find N large
enough such that if n ≥ n then |a (n) − l| and |c (n) − l| are both smaller than
δ, that is, again by Claim 6.1, equivalent to

l−δ < a (n) < l + δ

l−δ < c (n) <l+δ

so we find, b (n) ≤ c (n) < l + δ and b (n) ≥ a (n) > l − δ which means that
|b (n) − l| < δ for all n ≥ N (the same threshold of both a and c). Since δ > 0
was arbitrary we are finished.

6.15 Remark. If we have two sequences a, b : N → R such that a (n) < b (n)
for any n ∈ N and such that both limits exist, we can “take the limit of the
inequality” and the inequality will still hold (though it stops being strict):
lim a ≤ lim b

Proof. If lim b = ∞ or lim a = −∞ then there is nothing to prove. Assume

first that both limits are finite. We already know then that the sequence
c := b − a converges to lim c = lim b − lim a. So our goal is to show that if
c (n) > 0 then lim c ≥ 0. Assume otherwise, that is, assume lim c < 0. Then
means that infinitely many n’s have c (n) < 0, as d (lim c, c (n)) is supposed
to be small, that is

c (n) < lim c + δ

for any δ > 0, for n large, and if we pick for example δ := − 21 lim c, we get
that c (n) is strictly negative, which cannot be.
The other cases follow easier reasoning.

6.16 Claim. We have the following special sequences, where α, p ∈ R and p > 0
1. If a : N → R is given by a (n) := n−p then
lim a = lim n−p
n→∞
= 0.
1
2. If a : N → R is given by a (n) := p n then
1
lim a = lim p n
n→∞
= 1.
1
3. If a : N → R is given by a (n) := n n then
1
lim a = lim n n
n→∞
= 1.

35
 nα
4. If a : N → R is given by a (n) := (1+p)n then

nα
lim a = lim n
n→∞ (1 + p)
= 0.

5. If a : N → R is given by a (n) := xn and x ∈ R with |x| < 1 then lim a = 0.

We will not include the proof for these (see [4], Theorem 3.20).
6.17 Claim. If a sequence a : N → R is monotone (as in Definition 5.7) then it
either converges to a finite number or it diverges to ∞ or to −∞. If it is both
monotone and bounded (as in Definition 5.11) then it necessarily converges to
a finite number.

Proof. Assume first that a is monotone increasing. This means that

a (n + 1) ≥ a (n) (n ∈ N)

If a is not bounded (as in Definition 5.11) then this fits the definition of a
sequence that diverges to infinity Definition 6.6. Then assume otherwise that
a is bounded by some constant M ≥ 0. Consider the set of numbers

im (a) ≡ { a (1) , a (2) , a (3) , . . . }

which is bounded by M from above and by a (1) from below due to the
monotonicity assumption. It is a fact that any bounded subset S ⊆ R has
what is called a least upper bound, denoted by sup (S), which is the smallest
possible upper bound on it. That is, it is an upper bound, and it is the
smallest in the set of all upper bounds. We will show that lim a exists by
showing that lim a = sup (im (a)) in this case.
First we need what is called the approximation property for the supremum.
It says the following: For any bounded set S ⊆ R, and for any ε > 0, there
is some element sε ∈ S such that sup (S) − ε < sε . Indeed assume otherwise.
Then there is some ε0 > 0 such that for all s ∈ S, sup (S) − ε ≥ s. But
then sup (S) − ε is an upper bound on S and since ε > 0, sup (S) − ε <
sup (S) so that sup (S) is not the least upper bound. Hence we have reached
a contradiction.
Using the approximation property for the supremum, let us return now
to the question of existence of lim a and its equality to sup (im (a)). Let
δ > 0 be given. Then we know by the approximation property that there is
some nδ ∈ N such that sup (im (a)) − δ < a (nδ ). Due to the monotonicity
assumption this implies that for all n ≥ nδ we have

sup (im (a)) − δ < a (n)

But also, from the fact that sup (im (a)) is an upper bound on im (a) it follows

36
 that for any n ∈ N,

a (n) ≤ sup (im (a)) < sup (im (a)) + δ

we conclude then that for all n ≥ nδ we have

|a (n) − sup (im (a))| < δ

which means that lim a = sup (im (a)).

6.3 Limits of functions from R → R

So far we have been dealing with limits sequence, which are functions N → R.
While there is a lot more to be said about such sequences (in particular the
Pn of infinite series, which are sequences a : N → R of the
whole development
form a (n) = m=1 b (m) for some other sequence b : N → R), let us turn our
attention to limits of other types of functions, namely, of functions R → R. The
limits of such functions are richer, since we can explore what happens as the
argument approaches more than just infinity. Indeed, if before, with sequences,
we had only one direction in which to probe the limit (namely, to keep going
forward in the direction of N), for functions whose domain is R, any point can
be a limit point, so to speak, which is intimately connected to the continuum
property of R referred to earlier (indeed limits of functions Q → R would make
less sense).
The idea is that since we are asking about the hypothetical process of what
would happen if we get nearer and nearer to a certain value in the domain (for
N that was the hypothetical value ∞, i.e. what happens to a (n) if n becomes
larger and larger), on R there is a whole continuum of values between any two
given points. Hence we could, for instance, get nearer and nearer to the value
zero without actually ever touching it. This brings us to our first
6.18 Example. Consider the function f : (0, ∞) → R given by f (x) := x1 .
There are two interesting directions for its limit that we can consider. The first
one is the one analogous to what we examined already in Example 6.8. In this
case we get the same result (we will define this formally shortly) that f (x) → 0
as x → ∞. However, now we may also approach x → 0 (which for sequences
was impossible since we were either at zero or we were always a fixed distance
away from it–at least distance 1). What happens to f (x) as x → 0? Apparently
it diverges to +∞. This is obvious from looking at the sketch of the graph of
the function as seen in Figure 6. We can also just plug in values smaller and
smaller:
1 7→ 1
1
7→ 2
2
1
7→ 3
3
...

37
 1.2

1.0

0.8

0.6

0.4

0.2

2 4 6 8 10

Figure 6: The graph of (0, ∞) 3 x 7→ x1 .

Hence it is clear that now for functions whose domain is R or a subset of it,
we need to measure the distance in the domain as well and not just let the
argument go to infinity (in the language of the previous section that mean all
n above a certain threshold N ). This gives us the following table of options for
the limit of a function f : R → R:

1. Probe the function at some point x ∈ R (which might not lie inside its
domain strictly speaking).
2. Probe the function at +∞ (this was the only thing which has an analogue
for sequences).
3. Probe the function at −∞.

• The result may converge to a finite number.

• The result may diverge to ±∞.
• The resulting limit may not exist.

Let us consider a few more examples:

6.19 Example. Consider the function f : R → R defined by x 7→ cos (x). This
function has no limit as x → ∞ because it keeps oscillating between ±1. Same
for sin.
6.20 Example. The function f : R → R defined by f (x) := expa (−x) for any
a > 1 (recall Example 5.31) has a limit of zero as x → ∞. (we won’t show this
now but we could relate it back to Claim 6.16).
6.21 Example. Sometimes the limit does not exist for a silly reason, for exam-
ple, that the function is different from the left or from the right
( of a given point.
1 x≥0
Indeed, consider the step function f : R → R given by x 7→ . Then
0 x<0
as we approach x → 0 from the right, we are always at 1, and as we approach
x → 0 from the left we are always at 0.

38
6.22 Definition. Let f : R → R. We say that limx→∞ f (x) exists and is equal
to some L < ∞ iff for any δ > 0 there is some Mδ > 0 such that if x > Mδ then
d (f (x) , L) < δ.
6.23 Definition. Let f : R → R. We say that limx→∞ f (x) diverges to ∞ iff
for all M > 0 there is some N > 0 such that if x > N then f (x) > M .
6.24 Remark. Similar definitions could be phrased concerning −∞, either in the
domain or in the co-domain of f .
6.25 Definition. (Limit point of a subset of R) Let A ⊆ R. The point l ∈ R
is called a limit point of A iff for any ε > 0 there is some a ∈ A\ { l } such that
d (a, l) < ε. We denote by A (called the closure of A) the union of A together
with the set of all its limit points.
6.26 Example. If A = { 1, 2, 3 } then A has no limit points, since we cannot
get arbitrarily close to any point from within A, as it is discrete.
6.27 Example. If A = (0, 1), then 1 is a limit point of A, even though 1 ∈ /A
itself. 0 is also a limit point, as well as any number in the interior of the interval
a ∈ (0, 1).
6.28 Example. If A = (0, 1) ∪ 2, the set of limit points of A is [0, 1]. In
particular, 2 is not a limit point of A. Then A = [0, 1] ∪ 2.
More often than not, when we talk about limit points, it will be applied
when we take a set A = (a, b) which is an interval and then we want to talk
about a or b as limit points of A.
6.29 Definition. Let f : A → R with A ⊆ R. Let x0 ∈ A be a limit point of
A. Then we say that limx→x0 f (x) exists and is equal to some L ∈ R iff for any
ε > 0 there is some δε > 0 such that for any x ∈ A such that d (x, x0 ) < δε we
have d (f (x) , L) < ε.
6.30 Definition. Let f : A → R with A ⊆ R. Let x0 ∈ A be a limit point
of A. Then we say that limx→x0 f (x) diverges to infinity iff for any M > 0
there is some δM > 0 such that for any x ∈ A such that d (x, x0 ) < δM we have
f (x) ≥ M .
This concept is extremely similar to the limit of a sequence. The only dif-
ference is that now we have a slightly different criterion of what “approaching”
means: we need to make the distance approached in the domain small as well.
6.31 Remark. The laws of limits of sequences we derived Claim 6.13, Claim 6.14,
Claim 6.17 also hold for limits of functions, and we don’t repeat them in this
context.
6.32 Example. Consider the limit limx→0 sin(x) x . This is related to a special
function called the sinc function (see its sketch in Figure 7). Strictly speaking
we define sinc : R → R via the following formula
(
sin(x)
x x 6= 0
sinc (x) :=
1 x=0

39
 1.0

0.8

0.6

0.4

0.2

-10 -5 5 10

-0.2

Figure 7: The sketch of the graph of the function sinc.

Note that the function x 7→ sin(x)

x strictly speaking does not make sense with
zero in its domain, this is the reason for the piecewise definition. Hence the limit
is asking if we can stitch the two parts of the piecewise definition together in a
sensible way, if we can show that
sin (x)
lim = 1
x→0 x
The first step is to show that for all x ∈ R\ { 0 } we have:
sin(x)
cos (x) ≤ x ≤ 1.

Once we have this inequality, we simply employ the squeeze theorem Claim 6.14
since limx→0 cos (x) = 1 (the proof of this fact is similar to related to Exam-
ple 6.33).
To prove the inequality we take the reciprocal:

x 1
1≤ sin(x) ≤ .
cos (x)
1
We multiply by 2 and sin (x) to get

1 x 1 sin (x)
sin (x) ≤ 2π π ≤ .
2 2 cos (x)
Now if we picture a right-triangle inscribed inside of the unit circle as follows:

40
 x
We note that the sector of the circle with vertices OCA has area 2π π. The
1 1
area of the triangle with vertices OCA is 2 sin (x) × 1 = 2 sin (x). The area of
the triangle with vertices OBC can be find using the sinuses theorem: Applying
it on the triangle AOD we get:
sin (x) cos (x) 1
= = π
x y 2

Applying it on the triangle BOC we get

BC 1 OB
= = π
x y 2

Comparing the two we hence find that the area of the triangle OBC is
1 1x 1 sin (x)
BC = =
2 2y 2 cos (x)
1 x 1 sin(x)
The picture then explains the inequality 2 sin (x) ≤ 2π π ≤ 2 cos(x) , and we’re
finished.
6.33 Example. limx→ π2 cos (x) = 0. We already do know that cos π2 = 0

from the geometric picture. The question is rather can we quantify that as we
approach x → π2 we really have cos (x) → 0 (later on we will see this is the
definition of continuity of cos at π2 ). The answer is yes: Given any ε > 0, we
want
!
d (cos (x) , 0) ≡ |cos (x)| < ε

to hold for all x ∈ R such that d x, π2 < δε (that is, δε > 0 is the ε-dependent

threshold we seek). The way we can prove this is by using the connection

41
between cos and sin. Indeed, cos (x) = − sin x − π2 for all x ∈ R. Hence we

need to study sin when we plug in small values of the argument. Looking at
the geometric picture though, sin (x) is always smaller than the arc, which is x.
Thus

d (cos (x) , 0) = |cos (x)|

  π 
= sin x −

2

 π
≤ x − 

2
so in this case we can just pick δε := ε!
xn −y n
6.34 Example. Consider limx→y x−y . After some algebraic manipulation
we have
n−1
X n−1
X
xn − y n = xn − y n + xn−k y k − xn−k y k
k=1 k=1
n−1
! n−1
!
X X
n n−k k n n−k k
= x + x y − y + x y
k=1 k=1
(Can expand the sums at no cost to simplify the expressions)
n−1
X n
X
n−k k
= x y − xn−k y k
k=0 k=1
(Change the index of the second sum from k 7→ k + 1)
n−1
X n−1
X
= xn−k y k − xn−k−1 y k+1
k=0 k=0
(Pull out a factor of x and y respectively (they always exist))
n−1
X n−1
X
= x xn−k−1 y k − y xn−k−1 y k
k=0 k=0
(Factorize again since both sums are the same)
n−1
X
= (x − y) xn−k−1 y k
k=0

Hence we learn that

n−1
xn − y n X
= xn−k−1 y k
x−y
k=0

so that
n−1
xn − y n X
lim = lim xn−k−1 y k
x→y x−y x→y
k=0

42
Now we can use Claim 6.13 to find
n−1
xn − y n X
lim = y n−k−1 y k
x→y x−y
k=0
n−1
X
= y n−1
k=0
= ny n−1
√ √
x+ε− x
6.35 Example. Consider the sequence limε→0 ε . Then as we already
saw in Example 6.10, we may factorize
√ √ (x + ε) − x
x+ε− x = √ √
x+ε+ x
ε
= √ √
x+ε+ x
and hence
√ √
x+ε− x 1
= √ √
ε x+ε+ x
√ √
which means,
√ √ 6.13, we only have to evaluate limε→0
using Claim x + ε+ x =
limε→0 x + ε + limε→0 x. Now
√ √
lim x + ε = x
ε→0

Indeed, from the equation above we get an estimate of

√ √ ε
x+ε− x = √ √
x+ε+ x
ε
≤ √
x
→ 0 as ε → 0 .

We conclude
√ √
x+ε− x 1
lim = √ .
ε→0 ε 2 x

6.36 Claim. If, for A, B ⊆ R, f : A → B is given such that lima→a0 f (a) =:

L1 exists and g : B → R such that limb→L1 g (b) =: L2 exists then for the
composition g ◦ f : A → R, the limit lima→a0 (g ◦ f ) (a) exists and equals

lim (g ◦ f ) (a) = lim g (b)

a→a0 b→L1

43
 Proof. This proof is extremely similar to Claim 6.43.
Since f (a) → L1 as a → a0 we have, for any ε > 0 some δ1 (ε) > 0 such
that if a ∈ A obeys d (a, a0 ) < δ1 (ε) then d (f (a) , L1 ) < ε.
Since g (b) → L2 as b → L1 , we have for any ε > 0 some δ2 (ε) > 0 such
that if b ∈ B obeys d (b, L1 ) < δ2 (ε) then d (g (b) , L2 ) < ε.
Hence for any ε > 0, if a ∈ A obeys d (a, a0 ) < δ1 (δ2 (ε)), we have
d (f (a) , L1 ) < δ2 (ε) so that d (g (f (a)) , L2 ) < ε. But this is precisely what
it means that (g ◦ f ) (a) → L2 as a → a0 .

6.37 Remark. This also gives us a reparametrization of limits. For example,

lim g (x0 + x) = lim g (x)

x→0 x→x0

Proof. We define the function f (x) := x0 + x. Then limx→0 f (x) = x0 easily

and we apply the claim to get

lim g (x0 + x) = lim g (f (x))

x→0 x→0
= lim g (x)
x→x0

sin(3x)
6.38 Example. Consider limx→0 x . Compare this to Example 6.32. We
have
sin (3x) sin (3x)
= 3
x 3x
If we define a new function f : x 7→ 3x for all x and sinc : x 7→ sin(x) x for all
sin(3x)
x 6= 0 then we have 3x = (sinc ◦ f ) (x). But we know that f (x) → 0 as
x → 0 (trivially) so that (sinc ◦ f ) (x) → 1 as x → 0 since sinc (x) → 1 as x → 0.
We conclude
sin (3x)
lim = 3
x→0 x
based on Example 6.32.
Example 6.21 pushes us to generalize our definition of limits to one sided
limits:
6.39 Definition. Let f : A → R and x0 ∈ A a limit point of A. Then the
left-sided limit of f at x0 exists and is equal to L, which is denoted by

lim f (x) = L
x→x−
0

iff for any ε > 0 there is some δ > 0 such that for all x ∈ A with x0 − x > δ we
have d (f (x) , L) < ε.

44
 The right-sided limit of f at x0 exists and is equal to L, which is denoted by
lim f (x) = L
x→x+
0

iff for any ε > 0 there is some δ > 0 such that for all x ∈ A with x − x0 > δ we
have d (f (x) , L) < ε.
6.40 Remark. Due to Claim 6.1, we can say that limx→x0 f (x) exists if and only
if both one-sided limits exist and are equal to each other.
(
1 x≥0
6.41 Example. Going back to Example 6.21, where f (x) := we
0 x<0
have
lim f (x) = 0
x→0−

and
lim f (x) = 1
x→0+

and indeed since the two limits are not equal we do not have that limx→0 f (x)
exists!
6.42 Example. If f : (0, 1) → R then there is no point to ask about the two-
sided limits at the end points 0 or 1, since the “other side” is not part of the
domain.
Actually there is a relationship between limits of sequences and limits of
functions!
6.43 Claim. Let f : A → R and x0 ∈ A be a limit point of A. Then limx→x0 f (x) =
L for some L ∈ R if and only if for any sequence a : N → A which converges to
x0 the new sequences f ◦ a : N → R converges to L.

Proof. First assume that for any sequence a : N → A which converges to x0

the new sequences f ◦ a : N → R converges to L. We then want to show that

lim f (x) = L.
x→x0

Hence let δ > 0 be given. We seek some ε > 0 such that if x ∈ A is such that
d (x, x0 ) < ε then d (f (x) , L) < δ. Assume the contrary. I.e. assume the limit
does not converge to L. Then that means that there is some δ0 > 0 such that
for each ε > 0 there is some xε ∈ A with d (xε , x0 ) < ε yet d (f (xε ) , L) > δ0 .
So pick ε to be a sequence such as n 7→ n1 . Hence there is some δ0 > 0
such that for each n ∈ N, there is some a (n) ∈ A with d (a (n) , x0 ) < n1 yet
d (f (a (n)) , L) > δ0 . But that means that a (n) → x0 yet (f ◦ a) (n) does not
converge to L as assumed. Thus we arrive at a contradiction.
Assume conversely that limx→x0 f (x) = L for some L ∈ R and let a : N →
A be any sequence converging to x0 . We want to show that f ◦ a : N → R

45
 converges to L. We know by assumption that for any ε > 0: (1) there is some
δε > 0 such that if x ∈ A is such that d (x, x0 ) < δε then d (f (x) , L) < ε; (2)
there is some Nε ∈ N such that if n ≥ Nε then d (a (n) , x0 ) < ε. Then for
n ≥ Nδε ,

d (a (n) , x0 ) < δε

so that

d (f (a (n)) , L) < ε

and so we have shown that for n ≥ Nδε , d (f (a (n)) , L) < ε, i.e. f ◦ a is a

sequence which converges to L.

7 Continuity of functions from R → R

7.1 Definition. The function f : A → R is called continuous at a limit point
x0 ∈ A iff limx→x0 f (x) = f (x0 ). If x0 ∈ A is not a limit point then f is defined
to be continuous at x0 .
7.2 Remark. Note that now for the limit x0 must really lie inside of A, the
domain of f . This means that if x0 is not a limit point of A (e.g. if A = { 1, 2, 3 }
then any f : A → R is continuous as A has no limit points, so there is nothing
to verify).
In principle continuity is a property of a function at a point of its domain.
But we can also think about continuity globally:
7.3 Definition. If f : A → R is continuous for all x ∈ A then f is called
continuous.
In general continuity is related to the sketch of the graph of a function having
no “jumps”, or not being too disjunct. Another way to think about continuity
is that it stipulates that the function at a point cannot be too wildly different
at nearby points. That is
7.4 Claim. Let f : A → R be given and pick some a ∈ A which is a limit
point and such that f is continuous at a. Then if we know the value of f at
a, i.e. if we know f (a), we can get an idea of what f (a + s) is for s ∈ R very
close to zero (i.e. a + s is very close to a). Indeed, continuity at a means that
limb→a f (b) = f (a). That means that we can make d (f (b) , f (a)) arbitrarily
small if we make d (b, a) arbitrarily small. More precisely, for any ε > 0 there
exists some δε > 0 such that if b ∈ A obeys d (b, a) < δε then d (f (b) , f (a)) < ε.
So if we pick s as any number within (−δε , δε ) we find that d (a + s, a) < δε
so that d (f (s + a) , f (a)) < ε. Unpacking what these distance estimates using
Claim 6.1 means that:
If we pick s ∈ (δε , δε ) then

f (a) − ε < f (a + s) < f (a) + ε

46
so once we know f (a) and continuity of f at a, we get a pretty good idea of
what f is nearby a. Coincidentally, this also tells us now something about the
meaning of δε : it is the size of the neighborhoud around a for which we get
estimates of size ε on f (a + s).
As a general rule of thumb, any function which can be written as a sequence
of algebraic manipulations (e.g. f (x) = 5x + 3 − 8 + x100 ) is continuous as long
as it is defined (e.g. x 7→ x1 is not continuous at zero as it is not defined at zero).
More complicated functions, such as cos : R → R, sin : R → R, expa : R →
(0, ∞), loga : (0, ∞) → R have to be examined and in principle their continuity
should not be taken for granted (though these ones listed turn out to be indeed
continuous).
Any function defined using the piecewise notation should be highly suspicious
in terms of its continuity.
7.5 Example. Going back to Example 6.21, it is clear that f there is continuous
on the whole of R except for the point zero, where it is not continuous.
7.6 Example. The function R 3 x 7→ x2 ∈ R is continuous. Indeed limy→x y 2 =
x2 for any x ∈ R. To see this, we calculate
d y 2 , x2 ≡ y 2 − x2 

 
 
2
= (y − x + x) − x2 
 
 
2
= (y − x) + 2x (y − x) + x2 − x2 
 

(Use triangle inequality)

2
≤ |y − x| + 2x |y − x|
2
≡ d (y, x) + 2xd (y, x)
√
So if we pick δε := x2 + ε − x and ask that d (y, x) < δε we find that that
p 2 p 
d y 2 , x2


≤ x2 + ε − x + 2x x2 + ε − x
p  p 
= x2 + ε − x x2 + ε − x + 2x
p  p 
= x2 + ε − x x2 + ε + x
= x2 + ε − x2
= ε
Note that δε > 0 for any x ∈ R. Indeed, if x < 0 this is obvious. If x ≥ 0, we
have
p ?
x2 + ε − x > 0
p ?
x2 + ε > x
?
x2 + ε > x2
ε > 0

47
indeed.
Actually we first started studying limits of functions R → R and then we
introduced the concept of continuity. But now that we have continuity and are
familiar with a few functions which are continuous, we may go back and use
this in order to calculate limits. Indeed, we
7.7 Claim. Let A, B ⊆ R be two given subsets. If f : A → R is continuous
at some limit point a0 ∈ A, g : B → A and has a limit at some limit point
b0 ∈ B which equals limb→b0 g (b) = a0 then we can “push” the limit through a
continuous function:
 
lim f (g (b)) = f lim g (b)
b→b0 b→b0

Proof. We know that f is continuous at a0 and it is a limit point. That means

that lima→a0 f (a) exists and equals f (a0 ). So we may applyClaim 6.36 (with
the roles of f and g actually reversed as in that claim) to obtain that

lim f (g (b)) = lim f (a)

b→b0 a→a0

= (Continuity of f at a0 )
= f (a0 )
(Hypothesis on g)
 
= f lim g (b) .
b→b0

7.8 Remark. Coincidentally this also shows us that the composition of contin-
uous functions is a continuous function:
 
lim f (g (x)) = f lim g (x)
x→x0 x→x0

= f (g (x0 )) .
7.9 Example. In one of the homework exercises we had to evaluate the limit
x
lim 22 = ?
x→0

Using the fact that y 7→ 2y is continuous, we can now “push” the limit inside
twice to get that this limit exists and equals 2.
One of the important consequences of continuity is the
7.10 Theorem. (Intermediate Value Theorem) Let f : [a, b] → R be continuous
and pick some c ∈ [f (a) , f (b)] (if f (b) < f (a) then reverse the order of the
interval). Then there is some x ∈ [a, b] such that f (x) = c, that is, f takes all
values in between the values at its end points.

48
 Proof. This ultimately related back to the fact that the image of an interval
under a continuous map is again an interval. Since this fact requires the
topological notion of connectedness, we shall not prove it here.

7.11 Example. Suppose we are looking for the solution x ∈ R of the equation
x21 −3x+1 = 0. Since there is this really large power of 21 we have no hope for a
closed form solution (such as the formula for the quadratic equation). However,
we know that the function R 3 x 7→ x21 − 3x + 1 ∈ R is continuous (it is just
some basic arithmetic operations). Furthermore, if we plug in x = −1 we get
21
(−1) − 3 (−1) + 1 = −1 + 3 + 1 = 3
and if we plug in x = +1 we get
21
(1) −3+1 = −1
Since 0 ∈ [−1, 3], somewhere between 3 and −1 the continuous function must
pass through zero, that is, there is a solution (one or more) to the equation
(though we still have no ideal what it is).
7.12 Claim. On any great circle around the world, for the temperature, pressure,
elevation, carbon dioxide concentration, if the simplification is taken that this
varies continuously, there will always exist two antipodal points that share the
same value for that variable.

Proof. Let f : [0, 2π] → R be a continuous function on the circle (parametrized

by angle), so that f (0) = f (2π). Let us pick any α ∈ [0, 2π] at arbi-
trary. Define d (α) := f (α) − f (α + π mod 2π). I.e., the difference be-
tween the value of f at α and the value at its anti-podal point. We know
that d (α + π) = −d (α) since we merely reverse the point with its anti-podal
point. Hence, d : [0, 2π] → R is a continuous function taking some value and
its negative, so that there must be some value α for which d (α) = 0, which
is precisely what we want.

7.13 Theorem. If f : [a, b] → R is continuous then there are two numbers

p, q ∈ [a, b] such that
f (p) ≥ f (x) ≥ f (q) (x ∈ [a, b])
i.e., f attains its maximum (at p) and minimum (at q).

Proof. The proof of this theorem is again related to notions of topology and
is thus outside the scope of our studies. It is related to “boundedness” and
“compactness” and the fact that continuity preserves these concepts.

7.14 Example. f : (0, 1) → R given by x 7→ x does not attain its maximum

within (0, 1). Indeed, for any t ∈ (0, 1), there is some s ∈ (0, 1) such that t < s,
because we are on the open interval, so we can get arbitrarily close to 1 without
actually touching it.

49
7.15 Corollary. Any continuous function f : [a, b] → R is bounded (as in
Definition 5.11).

8 Derivatives
8.1 Definition. The derivative of f : A → R at x ∈ A, denoted by f 0 (x), is
defined as the limit
f (x + ε) − f (x)
f 0 (x) := lim
ε→0 ε
if it exists. If it does, then f is called differentiable at x. If f is differentiable on
the whole of A, then this defines now a new function f 0 : A → R whose formula
is A 3 x 7→ f 0 (x). This function is well-defined due to limits being unique, if
they exist. If f is differentiable only on a subset of A, say, B ⊆ A, then f 0 , as
a function, is only defined on B.
Sometimes different notations are used for the derivative, the most common
one is the Leibniz notation
d
f 0 (x) = f (x)
dx
The problem with this notation (and why we will not be using it) is that it
forces you to commit to give a name to the independent variable (x in this
case), further conflating the function f (a rule on all numbers) with the number
f (x) (f evaluated at the point x). This confusion between f and f (x), or
between f 0 and f 0 (x), we try to avoid. We prefer to think of the derivative as a
function itself regardless of the name of its argument, so that we prefer to write
f 0 with no mentioning of the name of the argument x.
There is, however, some benefit (and also danger) in the Leibniz notation,
since it helps one remember what the derivative actually is: it is the limit of a
quotient of the difference of the values of the function at near-by points divided
by the distance between the nearby points. One should think of d as “Delta” (the
Greek letter ∆) which stands for difference or change. Hence we are calculating
the quotient “difference in f ” by “difference in x”. Indeed oftentimes one sees
the notation
∆f
lim
∆x→0 ∆x

The danger with this notation is that it (sometimes) makes one take the quotient
too literally and forget that there is also a limit involved.
Another possible notation for the derivative is with the symbol ∂, which
stands usually in math for partial derivative if there are several variables on
which a function depends. However since for us most functions are of one
variable there is no distinction. The way one uses this notation is as

∂f = f0

50
or as

∂x f = f0

if one wishes to make explicit with respect to which variable the differentiation
is happening (which in this case is x), which is sometimes useful, especially if we
want to refer to a function via the formula defining it (i.e. when the domain and
codomain are implicitly obvious). Then one writes conveniently, for instance

∂xn = nxn−1

instead of the more cumbersome

0
(x 7→ xn ) = x 7→ nxn−1 .
d
To summarize, for us, there is no distinction between ∂x and dx .

8.2 Remark. Clearly this notion only makes sense if x ∈ A is a limit point of A.
8.3 Example. The derivative of any constant function is the constant zero
function.

Proof. The constant function f (no matter what the constant is) will always
have f (x+ε)−f
ε
(x)
= 0ε = 0, so that the limit is always zero, no matter which
x we plug in.

8.4 Example. The derivative of the parabola function f : R → R; x 7→ x2

exists and equals x 7→ 2x. Indeed,
f (x + ε) − f (x)
f 0 (x) = lim
ε→0 ε
2
(x + ε) − x2
= lim
ε→0 ε
2εx + ε2
= lim
ε→0 ε
= lim 2x + ε
ε→0
= lim 2x + lim ε
ε→0 ε→0
= 2x + 0
= 2x

8.5 Remark. Knowing the derivative of a function at a certain point gives us

additional information in order to approximate it away from that point (but still
near by), beyond the information already provided by continuity as in Claim 7.4.
Indeed, let us say that f : R → R is differentiable at some x ∈ R. That means
that
1
f 0 (x) = lim (f (x + ε) − f (x))
ε→0 ε

51
exists and is finite. Unpacking what the limit actually means, we get that for
any a > 0, there is some ba > 0 such that if |ε| < ba then
 
1
 (f (x + ε) − f (x)) − f 0 (x) < a

ε 

or (upon multiplying by |ε|)

|f (x + ε) − f (x) − εf 0 (x)| < |ε| a

≤ ba a
WLOG ba < 1
< a

that means that

f (x) + εf 0 (x) − a < f (x + ε) < f (x) + εf 0 (x) + a

so we get even more information about the function near by, namely, how and
in which direction it changes with ε; cf Claim 7.4.
8.6 Remark. Another, geometric interpretation of the derivative, as as the slope
of the function at a certain point. The slope is related to the angle of a straight
line which is tangent to the function at the given point. Recall a straight line
is a function of the form

R 3 x 7→ ax + b

where a, b ∈ R are the parameters that define the straight line. a is called its
slope and it is related to the angle that the straight line forms with the horizontal
axis. Indeed, it is the tangent of that angle α: a = tan (α). Hence the derivative
gives us the angle of the straight line which is tangent to the function at that
point, i.e., its slope at that point.
8.7 Remark. Yet another interpretation of the derivative is as instantaneous
rate of change of the function, at the given point. What that means is, given
any point, how quickly does the function increase (if its derivative is positive) or
decrease (if its derivative is negative) at a given point, which is of course related
to its slope at that point. In physics, if f : R → R denotes the function that gives
a particle’s position at a certain instance of time, then f 0 would corresponds to
its instantaneous velocity. We will see this in Claim 8.49.
f
8.8 Example. The derivative of the absolute value R 3 x 7→ |x| does not exist
at zero, but otherwise exists everywhere else.

|0 + ε| − |0|
f 0 (0) = lim
ε→0 ε
|ε|
= lim
ε→0 ε

52
Now if ε > 0 we get 1 and if ε < 0 we get −1, i.e. the left-sided limit is −1
and the right-sided limit is +1, so that the double-sided limit does not exist and
hence the function is not differentiable at zero. Anywhere else, e.g. if x > 0,

|x + ε| − |x|
f 0 (x) = lim
ε→0 ε
(x > 0 so |x| = x; For |ε| < x, |x + ε| = x + ε)
x+ε−x
= lim
ε→0 ε
= 1

and similar argument if x < 0.

8.9 Example. The derivative of sin is cos.

Proof. This was actually a problem on the midterm. Let us see how it works.
Pick any x ∈ R. Then we calculate
1
sin0 (x) ≡ lim (sin (x + ε) − sin (x))
ε→0 ε
We cannot evaluate this limit directly because it is
of the indeterminate form
0 a−b
cos a+b


0 . So we use sin (a) − sin (b) = 2 sin 2 2 to get
 ε  ε
sin (x + ε) − sin (x) = 2 cos x + sin
2 2
so that
1  ε ε
sin0 (x) = lim 2 cos x + sin
ε→0 ε 2 2
sin 2ε

 ε
= lim ε

cos x +
ε→0
2
2
(Use algebraic laws of limits)

! 
sin 2ε  ε 
= lim ε

lim cos x +
ε→0
2
ε→0 2
 
sin (a)
= Use the definition sinc (a) ≡
a
  ε    ε 
= lim sinc lim cos x +
ε→0 2 ε→0 2
Now both sinc and cos are continuous functions, so we may push the limit
through. Recall the limit of sinc from Example 6.32: lima→0 sinc (a) = 1.
Thus we find

sin0 (x) = cos (x)

53
 Since x ∈ R was arbitrary, we conclude sin0 = cos.

8.10 Example. The derivative of cos is − sin.

Proof. Pick any x ∈ R. Then we have

1
cos0 (x) ≡ lim (cos (x + ε) − cos (x))
ε→0 ε
Now again we have an indeterminate form 00 , so let us use the trigonometric
identity
   
a+b a−b
cos (a) − cos (b) = −2 sin sin
2 2

to get cos (x + ε) − cos (x) = −2 sin x + 2ε sin 2ε . Using a very similar

trick of identifying a sinc as in the example above we conclude that the limit
converges to − sin (x).

8.11 Claim. The derivative is linear. That means that if f and g are two
functions which are differentiable at some x ∈ R and α, β ∈ R then the new
function
αf + βg
is differentiable at x0 with derivative equal to αf 0 (x)+βg 0 (x), i.e., we can write
0
(αf + βg) = αf 0 + βg 0

Proof. We have

0 1
(αf + βg) (x) ≡ lim (αf (x + ε) + βg (x + ε) − αf (x) − βg (x))
ε→0 ε
1
= lim (αf (x + ε) − αf (x) + βg (x + ε) − βg (x))
ε→0 ε
 
1 1
= lim (αf (x + ε) − αf (x)) + (βg (x + ε) − βg (x))
ε→0 ε ε
(Use algebra of limits)
   
1 1
= lim (αf (x + ε) − αf (x)) + lim (βg (x + ε) − βg (x))
ε→0 ε ε→0 ε

(Use algebra of limits)

   
1 1
= α lim (f (x + ε) − f (x)) + β lim (g (x + ε) − g (x))
ε→0 ε ε→0 ε

(Use definition of derivatives of f and g)

≡ αf 0 (x) + βg 0 (x)

54
8.12 Theorem. If f is differentiable at some x ∈ R then f is continuous at
x.

Proof. Let us pick some point x ∈ R at which f is differentiable. That means

that the following limit exists and is finite
1
lim (f (x + ε) − f (x))
ε→0 ε
We want to show that limε→0 f (x + ε) exists and equals f (x) (this is the
definition of continuity according to Definition 7.1), that is, we want to show
that the following equation holds (and the limit in it exists)

lim f (x + ε) = f (x)
ε→0

ε
lim (f (x + ε) − f (x)) = (f (x + ε) − f (x))
lim
ε→0 εε→0
(Algebra of limits)
  1

= lim ε lim (f (x + ε) − f (x))
ε→0 ε→0 ε
0
= 0 · f (x)
= 0

8.13 Example. It is clear that the converse is false, namely, continuity does
not imply differentiability. The prime counter example is Example 8.8. The
absolute value is not differentiable at zero yet it is continuous at zero.
8.14 Claim. The derivative obeys the so-called Leibniz rule for products. That
means that if f and g are two functions differentiable at some x ∈ R then the
new function f g (the product) is also differentiable at x0 and its derivative is
equal to
0
(f g) (x) = f 0 (x) g (x) + f (x) g 0 (x)

i.e. one could write succinctly the Leibniz rule:

0
(f g) = f 0 g + f g0

55
 Proof. We have

0 1
(f g) (x) ≡ lim (f (x + ε) g (x + ε) − f (x) g (x))
ε→0 ε
1
= lim (f (x + ε) g (x + ε) − f (x + ε) g (x) + f (x + ε) g (x) − f (x) g (x))
ε→0 ε
1
= lim (f (x + ε) (g (x + ε) − g (x)) + (f (x + ε) − f (x)) g (x))
ε→0 ε
(Algebra of limits)
   
1 1
= lim f (x + ε) (g (x + ε) − g (x)) + lim (f (x + ε) − f (x)) g (x)
ε→0 ε ε→0 ε

Now we use again the algebra of limits, noting that because f is differen-
tiable at x, it is also continuous at x (as proven in Theorem 8.12) so that
limε→0 f (x + ε) = f (x). We find
   
0 1 1
(f g) (x) = f (x) lim (g (x + ε) − g (x)) + lim (f (x + ε) − f (x)) g (x)
ε→0 ε ε→0 ε

(Use differentiability of f and g)

= f (x) g 0 (x) + f 0 (x) g (x)

which is what we wanted to prove.

8.15 Example. The most important example of the product rule is when one
of the functions is a constant: Let f (x) = cg (x) for some c ∈ R, for all x ∈ R,
where g is a given function. Then
0
f0 = (cg)
= c0 g + cg 0
But c is just a constant, so c0 = 0 as we saw in Example 8.3, so we find f 0 = cg 0 .
8.16 Claim. Actually we already saw that if f (x) ≡ xn for some n ∈ N then
f 0 (x) = nxn−1 . Indeed, this was precisely Example 6.34! Actually this rule
works for any α ∈ R on [0, ∞) and not just n ∈ N: If f (x) = xα for all
x ∈ [0, ∞) then f 0 (x) = αxα−1 for all x ∈ [0, ∞).
8.17 Example. For instance, if f (x) := x12 for all x ∈ R \ { 0 }, then since
−2
1
x2 = x , we have f 0 (x) = −2x−3 = −2 x13 . Of course since the function f is
not defined at zero it is not differentiable there!
√
8.18 Example. Another example: if f (x) := x for all x ≥ 0 then since
√ 1 1 1
x = x 2 we have f 0 (x) = 21 x 2 −1 = 12 x− 2 = 12 11 = 2√
1
x
.
x2
We will see the proof for general n (i.e. not just n ∈ N) further below once
we understand the derivatives of log and exp.
8.19 Claim. For any a > 1, the logarithm function is differentiable and log0a (x) =
1
x loga (e) where e ≈ 2.718 is the natural base of the logarithm as in Defini-
tion 10.2. In particular, log0e (x) = x1 .

56
 Proof. We have
1
log0a (x) ≡ lim (loga (x + ε) − loga (x))
ε→0 ε
(Use logarithm laws)
  
1 x+ε
= lim loga
ε→0 ε x
1  ε 
= lim loga 1 +
ε→0 ε x
 x
Replace y :=
ε 
1 1
= lim y loga 1 +
y→∞ x y
(Use logarithm laws)
 y 
1 1
= lim loga 1+
y→∞ x y
 y 
1 1
= lim loga 1+
x y→∞ y
(Use continuity of log)
  y 
1 1
= loga lim 1 +
x y→∞ y
(Use definition of natural logarithm base *)
1
= loga (e)
x
where in the last step we used Definition 10.2.

8.20 Claim. For any a > 1, the exponential function is differentiable and its
derivative is equal to exp0a = loge (a) expa (recall e from Definition 10.2). In
particular, since loge (e) = 1 we find that

exp0e = expe .

Proof. We have, for any x ∈ R,

1
exp0a (x) ≡ lim(expa (x + ε) − expa (x))
ε
ε→0
(Use exponential laws)
1
= lim (expa (x + ε) expa (−x) − 1) expa (x)
ε→0 ε
1
= lim (expa (ε) − 1) expa (x)
ε→0 ε

57
 1
so we would be finished if we could show that limε→0 ε (expa (ε) − 1) =
loge (a). Let us rewrite

1 expa (ε) − 1
lim (expa (ε) − 1) = lim
ε→0 ε ε→0 loga (expa (ε))
1
= lim 1
ε→0
expa (ε)−1 loga (expa (ε) − 1 + 1)
(Use logarithm laws)
1
= lim  1

ε→0
loga (1 + expa (ε) − 1) expa (ε)−1
 
1
Use continuity of α 7→ to push the limit through
loga (α)
1
=  1

loga limε→0 (1 + expa (ε) − 1) expa (ε)−1

Now we use the fact that expa (ε) → 1 as ε → 0, since expa is continuity at
1
zero. Hence if g (ε) := expa (ε)−1 and f (α) := (1 + α) α , what we have is the
limit limε→0 (f ◦ g) (ε), and we have already learnt in Claim 6.36 that since
the limit of g at zero exists and equals zero, this limit equals limα→0 f (α), so
we find
1 1
lim (expa (ε) − 1) =  1

ε→0 ε loga limα→0 (1 + α) α

But this inner limit is precisely Definition 10.2, so that we find

1 1
lim (expa (ε) − 1) =
ε→0 ε loga (e)
(Use logarithm laws)
1
=
loge (e) / loge (a)
1
=
1/ loge (a)
= loge (a)

which is what we were looking for.

8.21 Claim. If f : A → B is differentiable at x ∈ A and g : B → R is differ-

entiable at f (x) then g ◦ f is differentiable x ∈ A and its derivative is equal
to
0
(g ◦ f ) (x) = g 0 (f (x)) f 0 (x)

58
or more succinctly
0
(g ◦ f ) = (g 0 ◦ f ) f 0

This is called the composition rule or the chain rule.

Proof. We have

0 1
(g ◦ f ) (x) ≡ lim (g (f (x + ε)) − g (f (x)))
ε→0 ε
(Rewrite the same thing)
g (f (x + ε)) − g (f (x)) 1
= lim (f (x + ε) − f (x))
ε→0 f (x + ε) − f (x) ε
(Limit of products equal to product of limits, if both exist, and use the fact that f is differ
 
g (f (x + ε)) − g (f (x))
= lim f 0 (x)
ε→0 f (x + ε) − f (x)

If we define
(
g(y)−g(f (x))
y−f (x) y ∈ R \ { f (x) }
q (y) := 0
g (f (x)) y = f (x)

Then we are interested in

g (f (x + ε)) − g (f (x))
lim = lim q (f (x + ε))
ε→0 f (x + ε) − f (x) ε→0

= lim (q ◦ f ) (x + ε)
ε→0

We note that q is continuous at f (x). Indeed, its continuity is equivalent to

the statement that the following limit exists and equals

g 0 (f (x)) = lim q (y)

y→f (x)

g (y) − g (f (x))
= lim
y→f (x) y − f (x)

which is the very definition of g being differentiable at f (x) (which we

assume). Now since q is continuous and f is continuous (as it is differ-
entiable, by Theorem 8.12), q ◦ f is continuous by Remark 7.8, so that
limε→0 (q ◦ f ) (x + ε) = (q ◦ f ) (x) = q (f (x)) = g 0 (f (x)) and we are fin-
ished.

8.22 Example. Let us try to evaluate the derivative of exp2 ◦ sin. This function
is given by the formula R 3 x 7→ 2sin(x) . We have by Claim 8.21 that
0 0
(exp2 ◦ sin) = (exp02 ◦ sin) sin0

59
Now we know from Example 8.9 that sin0 = cos and from Claim 8.20 we know
that exp02 = loge (2) exp2 . Hence
0
(exp2 ◦ sin) = loge (2) (exp2 ◦ sin) cos .

More explicitly, we get the formula

x 7→ loge (2) 2sin(x) cos (x)

8.23 Example. Recall that there are the so-called hyperbolic functions, which
are analogous to the trigonometric functions (with domain and codomain R):
the hyperbolic sinus is
ex − e−x
sinh (x) ≡
2
and the cosinus is
ex + e−x
cosh (x) ≡
2
and the tangent is
sinh (x)
tanh (x) ≡
cosh (x)
hence we can immediately calculate their derivatives and get (with m (x) ≡ −x
for all x ∈ R)
1 0

sinh0 = exp0 − (exp ◦m)
2
1
= (exp − (exp ◦m) m0 )
2
(Use m0 = −1)
1
= (exp + (exp ◦m))
2
≡ cosh

and
1 0

cosh0 = exp0 + (exp ◦m)
2
1
= (exp + (exp ◦m) m0 )
2
(Use m0 = −1)
1
= (exp − (exp ◦m))
2
≡ sinh

Compare this with Example 8.9 and Example 8.10 (i.e. the lack of minus sign
on cosh0 ).

60
8.24 Example. Now that we have the composition, logarithm and exponential
derivatives, let us prove Example 6.34 or Claim 8.16 for powers which are not
necessarily natural numbers. So let f (x) := xα for any α ∈ R and x ∈ (0, ∞).
We want to show that f 0 (x) = αxα−1 .
Since x > 0, it is valid to rewrite y = exp (log (y)) for any y > 0 since log is
the inverse of exp. So

f (x) = exp (log (xα ))

= exp (α log (x))

Now let us apply the chain rule on this to get

0
f 0 (x) = exp0 (α log (x)) (α log (x))
1
= exp (α log (x)) α
x
1
= xα α
x
= αxα−1

Now if x < 0 then xα does not necessarily make sense, because we have no
prescription to take a root of a negative number.
8.25 Example. Continuing the example above, if α = − 13 , then xα = √1
3 x and
√ p
we when x < 0, we could write 3 x = 3 − |x|. Now because taking roots is
multiplicative, we have
p √ p
3
− |x| = 3 −1 3 |x|
√ 3
p
and 3 −1 = −1 (since (−1) = −1) and 3 |x| we know how to do. More
formally, let f : R → R be defined by
√
 x x>0
3
√

3
f (x) := x= 0 x=0
 p

− |x| x < 0
3

For x > 0 we could apply Claim 8.16 and get

1 −2
f 0 (x) = x 3
3
whereas for x < 0 we need to apply the chain rule Claim 8.21 to get
1 −2
f 0 (x) = − |x| 3 (−1)
3
1 − 23
= |x|
3

61
At zero the situation is more delicate and the definition of f 0 from the limit has
to be employed:
1
f 0 (0) ≡ lim (f (ε) − f (0))
ε(
ε→0
√
1 3ε ε>0
= lim √
ε→0 ε − −ε ε < 0
3

( 2
ε− 3 ε>0
= lim 2
ε→0 |ε|− 3 ε<0
− 23
= lim |ε|
ε→0

this last limit does not exist–it diverges to +∞. Hence f is not differentiable
at zero. The geometric
meaning of this ∞ is that the slope of the tangent is
actually vertical! tan π2 = ∞.
1
8.26 Example. The derivative of f : R\ { 0 } → R whose formula is f (x) = x
is equal to f 0 (x) = − x12 .

Proof. We could either use the rule Claim 8.16 or we can appeal directly to
the definition:
 
1 1 1
f 0 (x) = lim −
ε→0 ε x+ε x
 
1 x−x−ε
= lim
ε→0 ε x (x + ε)
−1
= lim 2
ε→0 x + εx
 
1
Use continuity of x 7→
x
1
= − 2
x

8.27 Example. Let f : R → R be defined by the formula

(
x sin x1


x 6= 0
f (x) :=
0 x=0
When x 6= 0, we may use all of our theorems since all functions involved,
including x 7→ x1 , are differentiable. We have
    
0 1 1 1
f (x) = sin + x cos − 2
x x x
   
1 1 1
= sin − cos
x x x

62
At x = 0, we must revert to the definition:
1
f 0 (0) ≡ lim (f (ε) − f (0))
ε→0 ε  
1
= lim sin
ε→0 ε

This latter limit does not exist, so f is not differentiable at zero.

8.28 Example. Let f : R → R be given by

(
x2 sin x1


x 6= 0
f (x) :=
0 0

At x 6= 0 we get
    
0 1 2 1 1
f (x) = 2x sin + x cos − 2
x x x
   
1 1
= 2x sin − cos
x x

To understand what happens at zero we again must revert to the definition:

1
f 0 (0) = lim (f (ε) − f (0))
ε→0 ε  
1
= lim ε sin
ε→0 ε

This limit tends to zero by the squeeze theorem for example (since im (sin) =
[−1, 1]). Hence f 0 (0) exists and equals zero. However, f 0 as a function itself is
not continuous at zero, since we do not have

lim f 0 (ε) = 0
ε→0

Indeed, the left hand side does not even exist since cos 1ε does not converge to

anything as ε → 0. Hence f is an example of a function whose derivative exists,

but that the derivative function f 0 is not a continuous function. The fact it’s
not continuous means that of course it cannot be differentiable itself, that is,
f 00 does not exist:
If we try to calculate the second derivative we’ll find, for x 6= 0:
       
1 1 1 1 1
f 00 (x) = 2 sin + 2x cos − 2 + sin − 2
x x x x x
     
1 1 2 1
= 2 − 2 sin − cos
x x x x

63
whereas again at zero special care must be taken
1 0
f 00 (0) = lim (f (ε) − f 0 (0))
ε
ε→0
   
1 1 1
= lim 2ε sin − cos
ε→0 ε ε ε
    
1 1 1
= lim 2 sin − cos
ε→0 ε ε ε

this last limit indeed does not exist.

8.29 Claim. If f, g are two functions such that g does not take the value zero
(so that fg is a well-defined function–otherwise, restrict) and such that f, g are
both differentiable and g 0 also does not take the value zero, then
 0
f f 0 g − f g0
= .
g g2

Proof. Let us apply the various rules we have in order to figure this. Firstly,
let us define the function r (x) := x1 for all x 6= 0. This function takes a
number and gives its reciprocal. Then we get write g1 = r ◦ g, and so

f
= f (r ◦ g)
g

Hence to differentiate fg we need to apply both the Leibniz product rule

Claim 8.14 as well as the composition rule Claim 8.21:
 0
f 0
= (f (r ◦ g))
g
(Use Leibniz rule)
0

= f 0 (r ◦ g) + f (r ◦ g)
(Use composition rule on second term)
f0
= + f ((r0 ◦ g) g 0 )
g
(Rewriting the same thing...)
f 0g
= + (r0 ◦ g) f g 0
g2
2
Now in Example 8.26 we learn that r0 = −r2 . (i.e. r0 (x) = − x12 = −r (x) ).

64
 Hence we have r0 ◦ g = − g12 . We learn that
 0
f f 0 g f g0
= − 2
g g2 g
f 0 g − f g0
=
g2
which is the result we were looking for.

8.30 Example. Let us try to evaluate tan0 . We know that tan ≡ sin
cos . Hence
using Claim 8.29
 0
sin
tan0 =
cos
sin0 cos − sin cos0
=
cos2
Now we use the rules Example 8.9 and Example 8.10 to find
cos2 + sin2
tan0 =
cos2
However, cos2 + sin2 ≡ 1 (make a drawing if you need to, but this is the Pythago-
rian theorem). Hence
1
tan0 =
cos2
1
This last expression, cos , is sometimes called the secant.
Similarly, we have
 0
0 sinh
tanh =
cosh
sinh0 cosh − sinh cosh0
=
cosh2
cosh − sinh2
2
=
cosh2
Now there is a similar identity for the hyperbolic functions that says that
cosh2 − sinh2 = 1 (you can verify this directly) so that
1
tanh0 =
cosh2
8.31 Theorem. ( The Hospital rule; L’Hôpital’s rule) If f : A → R and g :
A → R are both differentiable, and if for some limit point of A, a (which is
not necessarily inside of A!) we have limx→a f (x) = limx→a g (x) = 0 or (both)
0
±∞, and g 0 (x) 6= 0 for all x ∈ A and limx→a fg0 (x)
(x)
exists, then
f (x) f 0 (x)
lim = lim
x→a g (x) x→a g 0 (x)

65
 0
8.32 Example. The requirement that limx→a fg0 (x)(x)
exists is crucial. Consider
f (x) = x + sin (x) and g (x) = x at a = ∞. Then

f 0 (x) = 1 + cos (x)

and

g 0 (x) = 1
0
so that fg0 (x)
(x)
= 1+cos(x)
1 = 1 + cos (x). The limit here does not exist as x → ∞.
But we can work with the original quotient to get
x + sin (x) sin (x)
lim = lim 1 +
x→∞ x x→∞ x
= 1 + lim sinc (x)
x→∞
| {z }
=0
= 1

which exists!
8.33 Example. Consider the limit limx→0 exp(x)−1
x2 +x . Since exp (0) = 1, we have
the indeterminate form 00 . But proceeding with the hospital’s rule, we get

expa (x) − 1 expa (x)

lim = lim
x→0 x2 + x x→0 2x + 1
limx→0 expa (x)
=
limx→0 (2x + 1)
1
=
1
= 1

8.34 Claim. The derivative of the inverse function: Let f : A → B be an

invertible function in the sense of Definition 5.28. That means that there is
some f −1 : B → A such that f ◦ f −1 = 1B and f −1 ◦ f = 1A . Assume that A
and B are subsets of R and let us further assume that f is differentiable and
also that its derivative does not take the value zero. Then

0 1
f −1 = ◦ f −1
f0

Proof. We know that f −1 ◦ f = 1A by definition of the inverse, where

(1A ) (x) = x for all x ∈ A. Let us differentiate the RHS of the equation:

10A = 1

i.e. the constant function which always equals 1. On the other hand if we

66
 differentiate the LHS of the equation, using Claim 8.21 we find

0 
0 
f −1 ◦ f = f −1 ◦ f f 0

So we find

0 
1 = f −1 ◦ f f0

or
1
0
= f −1 ◦f.
f0

If we now apply f −1 to both sides of the equation from the right (using
f ◦ f −1 = 1B ), we get
1
0
◦ f −1 = f −1
f0
which is the result we were looking for.
8.35 Example. Let us use this rule in order to find the derivative of arcsin.
Recall that arcsin is defined as the inverse of sin. Since sin is in general not
invertible, we must restrict its domain and codomain in order to really get an
honest inverse. Then if we re-define sin as
h π πi
sin : − , → [−1, 1]
2 2
then it is both injective and surjective, and so it is indeed invertible, and we
denote its inverse by arcsin (whatever it is, we do not really know how to get a
formula for it...) We only know that
(sin ◦ arcsin) (x) = x (x ∈ [−1, 1])
 h π π i
(arcsin ◦ sin) (x) = x x∈ − ,
2 2
Using these two relations it is enough to find arcsin0 , even though we still have
no idea what the formula for arcsin is!
1
arcsin0 = ◦ arcsin
sin0
1
= ◦ arcsin
cos
You might say this is useless, since we still don’t have a formula for arcsin.
However, cos acting on arcsin is something we can figure out, since we can re-
write
 π π(from the Pythagorian theorem sin2 + cos2 = 1 and using the fact that on
− 2 , 2 the sin is always increasing, so there sin0 is positive and cos = sin0 , so


we pick the positive version of the square root to get:

p
cos = 1 − sin2

67
so that
q
2
cos (arcsin (x)) = 1 − (sin (arcsin (x)))
p
= 1 − x2

We find that
1
arcsin0 (x) = √
1 − x2
which is quite remarkable since we still have no formula for what arcsin is! The
sign may be found by working out when arcsin is increasing vs. decreasing.

8.36 Example. Consider the function f : [0, ∞) → [0, ∞) given by f (x) := x2

for all x ∈ [0, ∞). Since we restrict the domain of f (from the naive choice of
R), f is injective. Since we restrict the codomain of f (from the naive choice √ of
R), f is surjective. Hence it is bijective and its inverse is given by f −1 (x) = x.
with the same domain and co-domain. Using Claim 8.34 we can then get make
sure that our expectation of the derivative of f −1 goes through. Indeed, using
Example 6.34 or Example 8.18 we know that

0 1
f −1 (x) = 1
2x 2
And from Claim 8.34 we get (Using f (x) = x2 so f 0 (x) = 2x = 21 (x) and so
1
f 0 (x) ≡ r (x))

0 1
f −1 = ◦ f −1
f0
1
= r ◦ f −1
2
1 1
= √
2 ·

8.1 Application: Minima and Maxima

8.37 Definition. Let f : A → R be given. Then a ∈ A is a maximum point
of f iff f (x) ≤ f (a) for all x ∈ A. Similarly, a is a minimum point of f iff
f (x) ≥ f (a) for all x ∈ A.

8.38 Definition. A point is an extremum point iff it is either a maximum point

or a minimum point.
8.39 Definition. Let f : A → R be given. Then a ∈ A is a local maximum
point of f iff there is some ε > 0 such that f (x) ≤ f (a) for all x ∈ A which
obeys d (x, a) < ε. Similarly, a is a local minimum point of f iff there is some
ε > 0 such that f (x) ≥ f (a) for all x ∈ A which obeys d (x, a) < ε.

68
 Figure 8: Local and global extrema (source: Wikipedia).

8.40 Definition. A point is a local extremum point iff it is either a local

maximum or a local minimum point.
The distinction is that sometimes a function gets to a valley which is not
absolutely the lowest point of a function, so we call that point a local minimum.
8.41 Definition. Let f : A → R be a given differentiable function. Then a ∈ A
is a stationary point iff f 0 (a) = 0.

8.42 Theorem. If f : [a, b] → R has a local extremum point at x ∈ (a, b)

and if f is differentiable at x, then f 0 (x) = 0, i.e. x is a stationary point for
f.

Proof. Assume first that the extremum a maximum. We know that

1
f 0 (x) ≡ lim (f (x + ε) − f (x))
ε→0 ε

 means that for any η > 0 there is some

By the definition of the limit, that
δη > 0 such that if |ε| < δη then  1ε (f (x + ε) − f (x)) − f 0 (x) < η. Without
loss of generality assume that δη is chosen such that x + δη < b and x − δη > a
(otherwise just make δη smaller, and the same conclusion would hold, since
this should work any ε such that |ε| < δη ). Then if −δη < ε < 0, x − δη <
x + ε < x, so we have (by assumption that x is a maximum point) that
f (x) ≥ f (x + ε) and also ε < 0, that is,
1
(f (x + ε) − f (x)) ≥ 0
ε

69
 since both numerator and denominator are negative. Hence
1
f 0 (x) ≥ (f (x + ε) − f (x)) − η
ε
≥ 0−η
= −η

Since η was arbitrary with the constraint, this means that f 0 (x) ≥ 0. Simi-
larly, if 0 < ε < δη , then x < x + ε < x + δη so that again by the assumption
of the maximum, f (x + ε) ≤ f (x) and so now
1
(f (x + ε) − f (x)) ≤ 0
ε
from which we learn that
1
f 0 (x) ≤ (f (x + ε) − f (x)) + η
ε
≤ 0+η
= η

and again since η > 0 was arbitrary this implies f 0 (x) ≤ 0.

Because f 0 (x) ≥ 0 and f 0 (x) ≤ 0, it can only be that f 0 (x) = 0.
The case where x is a local minimum proceeds similarly.
8.43 Remark. Note the converse is false, that is, finding a stationary point does
not mean that we found a local extremum!
8.44 Example. Consider the function f : [−1, 1] → R given by f (x) = x3 for
all x ∈ [−1, 1]. The derivative is f 0 (x) = x2 . If we want to find stationary
points we must solve the equation f 0 (x) = 0 for x, that is, x2 = 0 in this case.
The solution to this equation is just the point x = 0. Looking at Figure 9, it is
immediately clear that the point x = 0 is not a local extremum point, so what
we have is a converse of Theorem 8.42. Such stationary points which are not
local extrema are called inflection points.
8.45 Theorem. (Rolle’s theorem) Let f : [a, b] → R be continuous such that f
is differentiable on (a, b). If f (a) = f (b), then there must be some x ∈ (a, b)
such that f 0 (x) = 0.

Proof. If f is constant then we are finished, because then f 0 = 0. Otherwise,

there must be some t ∈ (a, b) for which f (t) > f (a) or f (t) < f (a).
Assume the former. Then by Theorem 7.13, there is some y ∈ [a, b] at
which f attains its maximum. Since f (t) > f (a), it must be that y ∈ (a, b).
By Theorem 8.42, f 0 (y) = 0 (the point here is that y is not one of the end
points a or b).
If on the other hand the latter applies, i.e., f (t) < f (a), then again by
Theorem 7.13 we find some y ∈ [a, b] at which f attains its minimum. Now

70
 1.0

0.5

-1.0 -0.5 0.5 1.0

-0.5

-1.0

Figure 9: The sketch of the graph of [−1, 1] 3 x 7→ x3 .

since f (t) < f (a), this means y ∈ (a, b) necessarily, and again, f 0 (y) = 0 by
Theorem 8.42.
In either case, f 0 (y) = 0 for some y ∈ (a, b) (and not y = a or y = b).

8.46 Example. If differentiability fails somewhere in the middle then the con-
clusion of Rolle’s theorem fails. For instance, take f (x) = |x|. We know this
isn’t differentiable at x = 0 and indeed if f ’s domain is [−1, 1] then there is no
point in (−1, 1) at which the derivativeis zero, even though f (−1) = f (1) = 1.
1
 x>0
We actually already saw that f 0 (x) = −1 x < 0.

undefined x = 0


8.47 Example. Differentiability √ is only required at the end points to apply

Rolle’s theorem. Take f (x) := 1 − x2 on [−1, 1]. Then the function is contin-
uous on [−1, 1] and differentiable on (−1, 1). Indeed, we have
1
f 0 (x) = √ (−2x)
2 1 − x2
x
= −√
1 − x2
for all x 6= ±1. But for x = ±1, the denominator is zero and indeed the function
is not differentiable there. None the less, Rolle’s theorem applies (since it only
requires differentiability on the open interval) and so there is indeed a point at

71
which f 0 is zero in the interior of (−1, 1), and that point is zero, as can be seen
from the formula.
8.48 Theorem. (The mean value theorem) If f : [a, b] → R is continuous, and
differentiable at least on (a, b), then there is some point x ∈ (a, b) such that
f (b) − f (a)
f 0 (x) =
b−a

Proof. Let g : [a, b] → R be continuous, and differentiable on (a, b). Define

h (x) := (f (b) − f (a)) g (x) − (g (b) − g (a)) f (x) (x ∈ [a, b])

Then h is continuous on [a, b] and differentiable at least on (a, b) and

h (a) = f (b) g (a) − f (a) g (b) = h (b)

by direct calculation. But now by Theorem 8.45 there must be some c ∈ (a, b)
for which h0 (c) = 0. Note that

h0 (x) = (f (b) − f (a)) g 0 (x) − (g (b) − g (a)) f 0 (x)

and so at such a point c ∈ (a, b) we get

(f (b) − f (a)) g 0 (c) = (g (b) − g (a)) f 0 (c)

Now if we pick g := 1, then g 0 = 1 and g (b) − g (a) = b − a and we find

f (b) − f (a)
f 0 (c) =
b−a
as desired.

8.49 Claim. Let f : R → R be differentiable. Let a, b ∈ R such that a < b. If

f 0 (x) ≥ 0 for all x ∈ [a, b] then f is increasing on [a, b]. If f 0 (x) ≤ 0 for all
x ∈ [a, b] then f is decreasing on [a, b].

Proof. Assume first that f 0 (x) ≥ 0 for all x ∈ [a, b]. Let t, s ∈ [a, b] such that
t < s. To show that f is increasing would mean to show that

f (t) ≤ f (s) .

Using Theorem 8.48 on [t, s] we learn that there must be some ξ ∈ (t, s) such
that
f (t) − f (s)
f 0 (ξ) = .
t−s

72
Figure 10: The mean value theorem.

73
 However, since by assumption f 0 (x) ≥ 0 for all x ∈ [a, b], and (t, s) ⊆ [a, b],
we must have f 0 (ξ) ≥ 0. Also, we know t − s > 0, so that

f (t) − f (s) = f 0 (ξ) (t − s)

| {z } | {z }
≥0 >0
≥ 0.

The proof follows very similar path if f 0 (x) ≤ 0 is assumed.

Claim 8.49 allows us to organize stationary points as follows:

8.50 Corollary. If f : R → R is a differentiable function such that for some
x ∈ R, f 0 (x) = 0 (i.e., x is a stationary point) then:
1. If f 0 is negative left to x and positive right to x, i.e., it goes from decreasing
to increasing, then x is a local minimum for f .
2. If f 0 is positive left to x and negative right to x, i.e., it goes from increasing
to decreasing, then x is a local maximum for f .
3. If f 0 is positive left to x and positive right to x, or alternatively it is
negative left to x and negative right to x, then x is an inflection point for
f.
8.51 Corollary. If f : [a, b] → R is a given function, then the global extrema
points of f will be attained at one of the following points:
1. Stationary points.
2. Boundary points (i.e. either a or b).
3. Points of non-differentiability.
8.52 Example. Consider the absolute value function f : [−1, 1] → R given by
f (x) := |x| for all x ∈ [−1, 1]. As we know, f is differentiable everywhere except
at zero, and
(
0 −1 x < 0
f (x) = .
1 x>0

Figure 3 shows that zero is a global minimum for the function, but it is clearly
not a stationary point since f is not differentiable there. The global maxima
are the boundaries −1 and 1, again, not stationary points.
8.53 Claim. If f : R → R is twice differentiable at some x ∈ R and f 0 (x) = 0
then:
1. If f 00 (x) > 0 then x is a point of local minimum.
2. If f 00 (x) < 0 then x is a point of local maximum.
If f 00 (x) = 0 the test is not informative.

74
Figure 11: (Wikipedia) Convex function, because the purple line is above the
black line.

Proof. Using Claim 8.49, we see that if f 00 (x) > 0, then f 0 is strictly increas-
ing near x, which means that it must change from negative to positive, so
that using Corollary 8.50 we learn that x is a local minimum for f .
Conversely, if f 00 (x) < 0, then by Claim 8.49 f 0 is strictly decreasing near
x, so it changes from positive to negative, so that again by Corollary 8.50 it
follows that x is a local maximum for f .
If f 00 (x) = 0, then we found a stationary point for f 0 , which does not give
further information about how it changes from being increasing to decreasing
or vice versa near x, i.e., nothing about the nature of the stationary point for
f.

8.2 Convexity and concavity

8.54 Definition. Let a, b ∈ R such that a < b and let f : [a, b] → R be given.
f is called convex iff for any x, y ∈ [a, b] and for any t ∈ [0, 1], f we have

f ((1 − t) x + ty) ≤ (1 − t) f (x) + tf (y) .

f is called strictly convex iff for any x, y ∈ [a, b] and for any t ∈ [0, 1], f we have

f ((1 − t) x + ty) < (1 − t) f (x) + tf (y) .

f is called (strictly) concave iff −f is (strictly) convex.

8.55 Remark. If we think the function

[0, 1] 3 t 7→ f ((1 − t) x + ty) ∈ R

75
we see that it is merely the restriction of f onto the interval [x, y], re-parametrized
so as to elapse that interval at length one (i.e. with the variable t ∈ [0, 1] instead
of [x, y]) so it goes between (0, f (x)) and (1, f (y)). On the other hand,

[0, 1] 3 t 7→ (1 − t) f (x) + tf (y)

= f (x) + t (f (y) − f (x))

corresponds to the straight line between the points (0, f (x)) ∈ R2 and (1, f (y)) ∈
R2 with slope f (y) − f (x). Hence, the requirement of convexity is that the
graph of the function between any two points always lies below the straight line
between the two points, as in Figure 11.
8.56 Claim. f : [a, b] → R is convex if and only if for any s, t, u ∈ [a, b] such
that
a<s<t<u<b
we have
f (t) − f (s) f (u) − f (t)
≤ .
t−s u−t

t−s
Proof. Assume f is convex. Define λ := u−s . Since s < t < u holds, λ ∈
u−s−t+s u−t
(0, 1). Also, 1 − λ = u−s = u−s . Finally, note that

u−t
1−λ =
u−s
l
(1 − λ) (u − s) = u−t
l
−s − λ (u − s) = −t
l
t = (1 − λ) s + λu .

Hence we may apply convexity of f to obtain

f (t) = f ((1 − λ) s + λu)

≤ (1 − λ) f (s) + λf (u)
u−t t−s
= f (s) + f (u)
u−s u−s

76
 which is in turn equivalent to

(u − s) f (t) ≤ (u − t) f (s) + (t − s) f (u)

l
(u − s) f (t) − (u − s) f (s) ≤ (u − t) f (s) + (t − s) f (u) − (u − s) f (s)
l
(u − s) (f (t) − f (s)) ≤ (t − s) (f (u) − f (s))
l
f (t) − f (s) f (u) − f (s)
≤ .
t−s u−s
Now if we take

(u − s) f (t) ≤ (u − t) f (s) + (t − s) f (u)

and multiply it by −1 we get

(s − u) f (t) ≥ (t − u) f (s) + (s − t) f (u)

l
(s − u) f (t) + (u − s) f (u) ≥ (t − u) f (s) + (s − t) f (u) + (u − s) f (u)
l
(u − s) (f (u) − f (t)) ≥ (u − t) (f (u) − f (s))
l
f (u) − f (s) f (u) − f (t)
≤ .
u−s u−t
Plugging these two together we find the condition in the claim. Since all
these steps have been equivalences, we can go backwards to show that if this
condition is true then f is convex.

8.57 Claim. Let f : [a, b] → R be differentiable. Then f is convex iff f 0 is

monotone non-decreasing.

Proof. Let x, y ∈ [a, b] such that x < y. Pick now also some ε > 0 and δ > 0.
Assume f is convex. Then we have by (two applications of) Claim 8.56 on
x, x + ε, y and on x + ε, y, y + δ, we get:

f (x + ε) − f (x) f (y) − f (x + ε) f (y + δ) − f (y)

≤ ≤ .
ε y−x−ε δ
Ignoring the middle part and taking the two limits ε → 0, δ → 0 we find that
f 0 (x) ≤ f 0 (y). To prove the converse we can use the mean value theorem
Theorem 8.48.

8.58 Corollary. Let f : [a, b] → R be twice differentiable. Then f is convex iff

77
f 00 is non-negative.

Proof. If f 00 is non-negative, then by Claim 8.49 f 0 is monotone non-decreasing,

so we can apply Claim 8.57.

Connecting this with Claim 8.53, we can interpret now that if a function is
strictly convex and has a stationary point, then that stationary point is neces-
sarily a minimum!
8.59 Claim. Let f : [a, b] → R be differentiable and convex. Then f 0 is contin-
uous.

Proof. Since f is convex, we learn that f 00 is non-negative by Corollary 8.58.

8.60 Example. If f 00 is strictly positive then f is strictly convex, but the

converse is not true. Consider f : [−1, 1] → R given by f (x) = x4 for all
x ∈ R2 . Then f 00 (x) = 12x2 , which is zero at x = 0. However, f is strictly
convex.

8.3 Application: Newton’s method

Let f : [a, b] → R be given such that f is differentiable and f 0 is differentiable
as well (i.e. f is twice differentiable), and such that f (a) < 0, f (b) > 0,
f 0 (x) ≥ δ > 0, and 0 ≤ f 00 (x) ≤ M for all x ∈ [a, b]. Let ξ be the unique point
in (a, b) at which f (ξ) = 0 (which is guaranteed to be unique by Theorem 8.45;
see HW7Q2.2).
Pick any x1 ∈ (ξ, b), and define a sequence N → R, n 7→ xn by the formula

f (xn )
xn+1 := xn − (n ∈ N) .
f 0 (xn )

Geometrically, the point xn+1 is chosen so that it is where the straight line with
slope f 0 (xn ) passing through (xn , f (xn )) intercepts the horizontal axis:

f (xn ) − 0
= f 0 (xn ) .
xn − xn+1

8.61 Claim. xn+1 < xn and

lim xn = ξ.
n→∞

Proof. First let us prove that xn ≥ ξ. We know already that x1 > ξ. Assume
that for all n ≤ m for some m ∈ N, xn > ξ. Check xm+1 . Apply the MVT

78
 Theorem 8.48 on f 0 between (ξ, xm ) to get some c ∈ (ξ, xm ) such that

f (xm ) − f (ξ)
| {z }
=0
= f 0 (c) .
xm − ξ

But since f 0 is increasing and c < xm , we have

f (xm )
≤ f 0 (xm )
xm − ξ
l
f (xm ) ≤ f 0 (xm ) (xm − ξ)
l
f (xm )
− xm ≤ −ξ
f 0 (xm )
l
xm+1 ≥ ξ.

This in turn implies (by knowledge that ξ is the unique zero point of f ,
and that f is increasing) that f (xn ) > 0 for all n. We also know f 0 (xn ) ≥ δ
for all n. Hence
−f (xn )
xn+1 − xn =
f 0 (xn )
< 0.

But now Claim 6.17 implies that limn→∞ xn exists and equals ξ.

8.62 Claim. [[4] Exercise 5.25 (d)] One can prove (though this is beyond the
scope of this class since it uses Taylor’s theorem) that
 2(n−1)
2δ M
0 ≤ xn − ξ ≤ (x1 − ξ) (n ∈ N) .
M 2δ
so that we get an upper bound on how far we are from the true value ξ and our
approximation xn at any given step n ∈ N.

8.4 Application: Linear Approximation

Another way to approximate a function with the derivative is via the linear
approximation, i.e. going back to Remark 8.5. There, we found that if f : R → R
is differentiable, and we know, for some x ∈ R, both f (x) and f 0 (x), then if ε
is a small number, it is not a bad approximation to write down

f (x + ε) ≈ f (x) + εf 0 (x) .

The precise meaning of the symbol ≈ was given in Remark 8.5, see also HW7Q2.1.

79
9 Integrals
9.1 The supremum and infimum
The following notions of supremum and infimum have been discussed a bit in the
proof of Claim 6.17 (where also an approximation property has been presented)
but we repeat them here since they are essential for the definition of the integral.
9.1 Definition. Let S ⊆ R. Then S is called bounded from above iff there is
some M ∈ R such that for all s ∈ S, s ≤ M . M is then called an upper bound
on S (note that many upper bounds exist once one exists).
9.2 Example. S = R is not bounded from above. S = (0, 1) is bounded above,
1 is an upper bound, but also 2 etc.
9.3 Remark. Sets bounded from below, and lower bounds, are defined similarly.

9.4 Definition. Let S ⊆ R be a set bounded from above. Then a supremum

on S is a a number α ∈ R such that the following two conditions hold:
1. α is an upper bound on S, in the sense above.
2. If β ∈ R is any other upper bound on S, then α ≤ β.

Another name for the supremum is least upper bound. One then writes α =
sup (S).
9.5 Definition. Let S ⊆ R be a set bounded from below. Then an infimum on
S is a number α ∈ R such that the following two conditions hold:
1. α is a lower bound on S, in the sense above.

2. If β ∈ R is any other lower bound, then α ≥ β.

Another name for the infimum is greatest lower bound. One then writes α =
inf (S).
9.6 Theorem. R has a completeness property that any subset of it S bounded
from above has a supremum in R, and every set T bounded from below has an
infimum in R.
√
√
9.7 Example. Consider the set 0, 2 ∩ Q. This set is bounded above by 2.
√
But the supremum, which is 2, is not in Q. We have to embed Q in R in order
to talk about the supremum of this set.
With the supremum and infimum of subsets of R, we are finally ready to
start discussing the integral.

80
 Figure 12: The integral’s geometric interpretation.

9.2 The Darboux integral

The notion of the integral has many different interpretations, perhaps the most
immediate one is the geometric interpretation which says that for a function
f : [a, b] → R, its integral is the area between:
 
1. The vertical line defined by the set of points (x, y) ∈ R2  x = a .
 
2. The vertical line defined by the set of points (x, y) ∈ R2  x = b .
 
3. The horizontal line defined by the set of points (x, y) ∈ R2  y = 0 .
 
4. The curve for the function defined by the set of points (x, y) ∈ R2  y = f (x) .

Since the curve of the function may be very complicated, we want to devise a way
to understand very general functions instead of restricting ourselves to simple
shapes (like triangles and squares). The way we do it is via approximation by
many small rectangles, and this is rigorously defined as follows.
9.8 Definition. (The Darboux integral ) Let f : [a, b] → R be a bounded func-
tion (as in Definition 5.11). We define its upper Darboux sum as the limit
b
lim S a (f, N )
N →∞

with
N −1     
b b−a X  b−a b−a
Sa (f, N ) := sup f (x) ∈ R  x ∈ a + n , (n + 1)
N n=0 N N

(if the limit exists at all) and its lower Darboux sum as

lim S ba (f, N )
N →∞

81
with
N −1     
b−a X b−a b−a
S ba (f, N )

:= inf f (x) ∈ R  x ∈ a + n , (n + 1)
N n=0 N N

(again, if the limit exists). If these two limits exist and are equal, i.e., if
b
lim S a (f, N ) = lim S ba (f, N )
N →∞ N →∞

then f is called integrable on [a, b] and the result of these limits is called its
integral on [a, b], and denoted by
Z b
f
a

or sometimes by
Z b
f (x) dx .
a
which is much more common and natural than the cumbersome
Z b Z b
(x 7→ f (x)) ≡ f (x) dx .
a a

Note that since f is bounded, the set

   
 b−a b−a
f (x) ∈ R  x ∈ a + n , (n + 1)
N N

is bounded (from above and below) for each n, and so it necessarily has a
supremum and an infimum.

9.9 Remark. Another common name for this construction is the Riemann in-
tegral, which is defined in a slightly different way, but the end result can be
proven to be equivalent to our Darboux integral.
9.10 Example. The simplest example is the integral of the constant function.
Indeed let

f : [a, b] → R

be given by

f (x) = c

for some c ∈ R. Then we always get

   
 b−a b−a
f (x) ∈ R  x ∈ a + n
 , (n + 1) = {c}
N N

82
 Figure 13: The two Darboux sum limits approximating the integral.

regardless of n, so that the upper Darboux sums are both equal to

N −1 N −1
1 X b−a X
lim sup ({ c }) = lim c
N →∞ N N →∞ N
n=0 n=0
b−a
= c lim N
N →∞ N
= c lim (b − a)
N →∞
= c (b − a) .
The result is the same for the lower Darboux sum, and so the constant function
is integrable and equal to the constant times the length of the interval, i.e. the
area of the rectangle trapped between the constant, the horizontal axis, and the
limits a and b.
9.11 Example. Let us now take the linear function, f : [a, b] → R defined by
f (x) = αx + β
for some α, β ∈ R. Let us assume that α > 0, so that f is monotonically
increasing (otherwise one has to flip all the logic). Note that while this function
is not bounded if the domain were R, it is bounded on [a, b]. On each set
   
 b−a b−a
f (x) ∈ R  x ∈ a + n
 , (n + 1)
N N
we have due to the monotonically increasing property, that the supremum is
equal to
   
b−a b−a
f a + (n + 1) = α a + (n + 1) +β
N N

83
and the infimum to
   
b−a b−a
f a+n = α a+n +β
N N

Let us attempt to calculate the upper Darboux sum:

N −1     
b b−a X  b−a b−a
lim S a (f, N ) = lim sup f (x) ∈ R  x ∈ a + n , (n + 1)
N →∞ N →∞ N N N
n=0
N −1    
b−a X b−a
= lim α a + (n + 1) +β
N →∞ N N
n=0
−1
 2 NX ! !
b−a
= lim α (b − a) a + (n + 1) + β (b − a)
N →∞ N n=0
PN ! !
n=1 n
= (b − a) α a + (b − a) lim +β
N →∞ N2
PN
Now we use the formula n=1 n = 21 N (N + 1) (see Claim 11.4) to get that this
equals
  1  
2 N (N + 1)
(b − a) α a + (b − a) lim +β
N →∞ N2
    
1 1
= (b − a) α a + (b − a) lim 1 + +β
2 N →∞ N
   
1
= (b − a) α a + (b − a) + β
2
α 
= (b − a) (b + a) + β
2
α 2
b − a2 + β (b − a) .


=
2
Similarly, let us verify that the lower Darboux sum is equal to this same thing:
N −1     
b−a X b−a b−a
S ba

lim (f, N ) = lim inf f (x) ∈ R  x ∈ a + n , (n + 1)
N →∞ N →∞ N N N
n=0
N −1    
b−a X b−a
= lim α a+n +β
N →∞ N N
n=0
PN −1 !
2 n=0 n
= α (b − a) a + α (b − a) lim + β (b − a)
N →∞ N2

84
now
PN −1 PN
n=0 n (n − 1)
n=1
lim = lim
N →∞ N2 N →∞ N2
PN PN
n=1 n − n=1
= lim
N →∞ N2
1
N (N + 1) − N
= lim 2
N →∞ N2
1
=
2
so we indeed get the same result, as
1 2
α (b − a) a + α (b − a)
 2 
1
= α (b − a) a + (b − a)
2
1
= α (b − a) (b + a)
2
α 2
b − a2 .


=
2
In conclusion, the two Darboux sum limits equal, so that f is integrable and we
conclude
Z b Z b
f ≡ (αx + β) dx
a a
1 2
b − a2 + β (b − a) .


= α
2
This makes perfect sense thinking about the meaning of the integral geomet-
rically, as it is precisely the area of the trapezoid defined by the straight line
f.
9.12 Example. Here is an example of when a function is not integrable. Define

f : [a, b] → R

by the formula
(
1 x∈Q
f (x) :=
0 x∈
/Q

Then f keeps jumping between zero and 1. The lower Darboux sum will be
zero, but the upper Darboux sum will be 1 and so we’ll get that

b−a 6= 0

hence the two limits both do exist, but are not equal, and so the function is not
integrable.

85
9.13 Claim. For any N ∈ N, we have
b
S ba (f, N ) ≤ S a (f, N ) .

Proof. The infimum of a set is always smaller than the supremum of that
same set.

9.14 Claim. The upper Darboux sums define a monotone decreasing subsequence
in N and the lower Darboux sums define a monotone increasing subsequence in
N:

S ba (f, N ) ≤ S ba (f, 2N ) (N ∈ N)
b b
Sa (f, N ) ≥ Sa (f, 2N ) (N ∈ N) .

Proof. Let us consider just the upper sums (the lower sums follow a similar
argument). To make the notation a bit shorter and the argument clearer,
let us (without loss of generality assume that a = 0 and b = 1; otherwise
one may rescale the function afterwards). Then we have (with f [ n , n+1 ] :=
 N N

sup f (x)  x ∈ n , n+1

  

N N for the sake of brevity)

N −1
1 1 X
S 0 (f, N ) ≡ f n n+1 .
N n=0 [ N , N ]

This calculation represents a division of the interval [0, 1] into N subintervals,

each of length N1 . I.e. the boundary points for the sub-intervals are

1 2 3 N −1
0, , , ,..., ,1.
N N N N
1 1
For S 0 (f, 2N ), we divide [0, 1] into 2N intervals, each of length 2N , so this
is actually a subdivision of the previous one, since now the boundary points
of the sub-intervals are
1 1 3 2
0 , , , ,...,1.
2N N 2N N

86
 1
Thus we can re-write S 0 (f, N )
N −1
1 1 X
S 0 (f, N ) = f n n+1
N n=0 [ N , N ]
N −1 N −1
1 X 1 X
= f [ n , n + 1 ]∪[ n + 1 , n+1 ] + f n n 1 n+1
2N n=0 [ N , N + 2N ]∪[ N + 2N , N ]
n 1
2N n=0 N N 2N N 2N N
(sup (A ∪ B) ≥ sup (A))
N −1 N −1
1 X 1 X
≥ f[ n , n + 1 ] + f n 1 n+1
2N n=0 N N 2N 2N n=0 [ N + 2N , N ]
N −1 N −1
1 X 1 X
= f [ 2n , 2n+1 ] + f 2n+1 2n+2
2N n=0 2N 2N 2N n=0 [ 2N , 2N ]
2N −1
1 X
= f n n+1
2N n=0 [ 2N , 2N ]
1
≡ S 0 (f, 2N )

9.15 Theorem. A function f : [a, b] → R is integrable if and only if for any

ε > 0 there is some Nε ∈ N such that the upper and lower Darboux sums at Nε
are at most ε far away from each other, that is, that
b
S a (f, Nε ) − S ba (f, Nε ) < ε.

Proof. In order to prove this, we need the notions of lim inf and lim sup which
we have not yet introduced.

9.16 Theorem. A function f : [a, b] → R is integrable if it is continuous.

9.17 Remark. Note that since f : [a, b] → R is continuous and defined on a
closed interval, it is automatically bounded via Corollary 7.15, so that it was
valid to ask if it is integrable at all.
9.18 Example. The converse is false. Consider the function
f : [0, 1] → R
(
1
5 x= 2
x 7→ 1
.
6
0 x= 2
1


This function is clearly not continuous, because limx→ 12 f (x) = 0 yet f 2 = 5.
However, it is integrable, and
Z b
f = 0.
a

87
9.19 Theorem. A function f : [a, b] → R is integrable if it is monotone (in-
creasing or decreasing).

Proof. Assume that f is monotone increasing without loss of generality. Then

      
 b−a b−a b−a
sup f (x) ∈ R  x ∈ a + n
 , (n + 1) = f a + (n + 1)
N N N

and
      
 b−a b−a b−a
inf f (x) ∈ R  x ∈ a + n , (n + 1) = f a+n
N N N

Hence we have
N −1    
b b−a X b−a b−a
S a (f, N ) − S ba (f, N ) = f a + (n + 1) −f a+n
N n=0 N N
"N −1   N −1  #
b−a X b−a X b−a
= f a + (n + 1) − f a+n
N n=0
N n=0
N
−1
"N   N   #
b−a X b−a X b−a
= f a+n − f a+n
N n=1
N n=0
N
−1  −1 
"   NX  N #
b−a b−a b−a X b−a
= f a+N + f a+n − f (a) − f a+n
N N n=1
N n=1
N
b−a
= (f (b) − f (a))
N
But this can be made arbitrarily small, so that by Theorem 9.15 we conclude
that f is integrable.

9.20 Remark. Again the assumption of monotonicity for a function on a closed

interval implies boundedness immediately. To see this, let us assume (without
loss of generality) that f is monotone increasing. That means that f attains its
maximum at b and its minimum at a so that

f (a) ≤ f (x) ≤ f (b) (x ∈ [a, b])

and so max ({ |f (b)| , |f (a)| }) is a bound on f . Hence it was legitimate to ask

whether f was integrable at all.
In light of Example 9.18, it is not surprising that the following is true:
9.21 Theorem. A function f : [a, b] → R is integrable if it is bounded and has
only finitely many points of discontinuity.
Note this last theorem stands in no contrast to Example 9.12, since in that
example, f had infinitely many points of discontinuity.

88
 1.0

0.5

0.2 0.4 0.6 0.8 1.0

-0.5

-1.0

1


Figure 14: The function x 7→ sin x is integrable.

9.22 Example. Let f : [0, 1] → R be given by

(
sin x1


x ∈ (0, 1]
f (x) := .
0 x=0

This function is certainly bounded, and in fact, it is continuous on (0, 1) and

has one point of discontinuity, namely, at zero. Hence by Theorem 9.21 it is
integrable! By the way, Z 1  
1
sin dx
0 x
has no explicit formula.

9.3 Properties of the integral

9.23 Theorem. Let f : [a, b] → R and g : [a, b] → R be two integrable functions.
Then f + g is integrable, and
Z b Z b Z b
(f + b) = f+ g.
a a a

Furthermore, if α ∈ R then αf is integrable and

Z b Z b
αf = α f.
a a

89
9.24 Theorem. (Monotonicity) Let f : [a, b] → R and g : [a, b] → R be two
integrable functions. If f (x) ≤ g (x) for all x ∈ [a, b] then
Z b Z b
f ≤ g.
a a

9.25 Theorem. Let f : [a, b] → R be integrable and c ∈ (a, b). Then f |[a,c] :
[a, c] → R and f |[c,b] : [c, b] → R are integrable, and
Z c Z b Z b
f+ f = f.
a c a

A good way to remember this result is that if A, B ⊆ R are two sets with no
intersection, then
Z Z Z
f = f+ f,
A∪B A B

i.e. the integral on the union is the sum of the integrals.

9.26 Theorem. Let f : [a, b] → R be integrable and bounded by M ∈ R, i.e.,
|f (x)| ≤ M for all x ∈ R. Then
Z 
 b 
f  ≤ M (b − a) .
 

 a 

9.27 Theorem. Let f : [a, b] → R and g : [a, b] → R be two integrable functions.

Then f g : [a, b] → R is integrable.
9.28 Remark. Note that even though this result says that f g is integrable, there
is nothing like the Leibniz rule (Claim 8.14) for integrals! It is in principle very
hard to calculate the integral of f g given the integrals of f and g separately.
9.29 Theorem. Let f : [a, b] → R be integrable. Then |f | : [a, b] → R is
integrable, and
Z 
 b  Z b
f ≤ |f | .
 

 a  a

9.30 Theorem. (Leibniz integral rule) Let f : [a, b] → R be integrable. Define

F : [a, b] → R by the formula
Z x
F (x) := f (x ∈ [a, b]) .
a

Then F is continuous, and if f is continuous at some x ∈ [a, b], then F is

differentiable at x, and
F 0 (x) = f (x) .

90
Proof. For continuity, let us check that
?
F (x) = lim F (x + ε)
ε→0
Z x+ε
= lim f.
ε→0 a

Now using Theorem 9.25 we have

Z x+ε Z x Z x+ε
f = f+ f
a a x
Z x+ε
= F (x) + f.
x

so that we have to prove

Z x+ε
lim f = 0.
ε→0 x

Since f is bounded (as it is integrable!), say, by M ≥ 0, we have by Theo-

rem 9.26
Z x+ε 
 

 f  ≤ εM
x

and so
R x+ε
−εM ≤ x
f ≤ εM

and so by the squeeze theorem Claim 6.14 the limit is zero. We learn that F
is indeed continuous.
Let us now further assume that f is continuous at some x ∈ [a, b], and
verify that F is differentiable at that same x. We have

? 1
F 0 (x) = lim (F (x + ε) − F (x))
ε
ε→0
Z x+ε Z x 
1
= lim f− f .
ε→0 ε a a

Now using Theorem 9.25 again we have

Z x+ε Z x Z x+ε
f = f+ f
a a x

91
 so that
Z x+ε
0 ? 1
F (x) = lim f.
ε→0 ε x

Since f is continuous at x, for any e > 0 there is some de > 0 such that if
y ∈ [a, b] is such that |y − x| < de then |f (y) − f (x)| < e. So when |ε| < de ,
we have for any y ∈ [x, x + ε] that |x − y| < de and so

f (x) − e ≤ f (y) ≤ f (x) + e

Hence we find for such ε using Theorem 9.24 that

Z x+ε Z x+ε
R x+ε
(f (x) − e) dy ≤ x f (y) dy ≤ (f (x) + e) dy
x x

but the outer integrals are of constant functions (in y) so that (upon dividing
by 1ε we learn)
R x+ε
f (x) − e ≤ 1ε x f (y) dy ≤ f (x) + e
 R 
 x+ε
which is equivalent to  1ε x f (y) dy − f (x) < e. Since e > 0 was arbitrary,

x+ε
we learn that limε→0 1ε x f exists and equals f (x), i.e.
R

F 0 (x) = f (x) .

9.31 Remark. If one thinks of integration as aR map from functions to functions,

x
taking a function f into a new function x 7→ a f , i.e.,
 Z x 
f 7→ x 7→ f
a
and as differentiation as a map from functions to functions f to their derivatives
(using the notation in the end of Definition 8.1 for the derivative)
f 7→ ∂f
then we learn
R that in a certain sense, Theorem 9.30 says that, ∂ is the left
inverse to :
Z
∂◦ = 1functions .

9.32 Theorem. (The fundamental theorem of calculus) If f : [a, b] → R is

differentiable and if f 0 : [a, b] → R is integrable, then
Z b
f 0 = f (b) − f (a)
a
b
=: f (x)|x=a .

92
 Rb
Proof. We know by Definition 9.8 that a
f 0 is going to be approximated from
b
below by S ba (f 0 , N ) and from above by S a (f 0 , N ) for some finite N , and that
these approximations become better as N grows larger. Consider the upper
approximation,
N −1     
b 0 b−a X 0
 b−a b−a
Sa (f , N ) ≡ sup f (x) ∈ R  x ∈ a + n
 , (n + 1) .
N n=0 N N

Since f is differentiable on each subinterval, we may apply the mean value

theorem Theorem 8.48 to  get that (forb−a 
N ), for each n = 0, . . . , N − 1,
fixed
there is some xn ∈ a + n b−a N , (n + 1) N such that

f a + (n + 1) b−a − f a + n b−a



0 N N
f (xn ) =
a + (n + 1) b−a
N −a−n N
b−a

f a + (n + 1) b−a − f a + n b−a



N N
= b−a
N

Now when we calculate

N −1 N −1
b − a X f a + (n + 1) b−a − f a + n b−a



b−a X 0 N N
f (xn ) = b−a
N n=0 N n=0 N

the sum telescopes and only the first and last terms survive:
N −1
b − a X f a + (n + 1) b−a − f a + n b−a



N N
= b−a
N n=0 N
N −1    
X b−a b−a
= f a + (n + 1) −f a+n
n=0
N N
        
b−a b−a b−a
= f a+ − f (a) + f a + 2 −f a+ + ··· +
N N N
    
b−a b−a
+ f a+N − f a + (N − 1)
N N
= f (b) − f (a) .

However, trivially, since xn ∈ a + n b−a b−a

 

N , (n + 1) N ,
       
b−a b−a b−a
f 0 (x) ∈ R  x ∈ a + n f 0 (x) ∈ R  x ∈ a + n
 
inf , (n + 1) ≤ f 0 (xn ) ≤ sup ,
N N N

93
 and hence when we sum up,
PN −1 b
S ba (f 0 , N ) ≤ b−a
N n=0 f 0 (xn ) ≤ S a (f 0 , N )

But we just learnt above that the inner term is independent of N (due to the
telescoping) and simply equals f (b) − f (a), hence,
b
S ba (f 0 , N ) ≤ f (b) − f (a) ≤ S a (f 0 , N ) .

Taking the limit N → ∞ of this inequality (Remark 6.15) we learn that

Z b Z b
0
f ≤ f (b) − f (a) ≤ f0
a a

which is equivalent to
Z b
f0 = f (b) − f (a) .
a

9.33 Remark. Theorem 9.32, which is a culmination of our entire effort in this
course, says that differentiation is the right inverse of integration, i.e.
Z
◦∂ = 1functions .

9.34 Remark. Theorem 9.32 will be our main hammer or work horse to “solve”
integrals, rather than the explicit definition Definition 9.8 which is explicitly
worked out in Example 9.11 (i.e. one almost never follows the procedure in Ex-
ample 9.11, which is really presented here more for illustration than an actual
computational tool). Using Theorem 9.32 we learn that if we can re-write a
function as the derivative of another function, then we can immediately inte-
grate. This is easier said than done, and many many functions for which one
can prove (by brute-force) integrability, one still cannot write down an explicit
formula for the result of the integral.
At any rate, this tells us immediately the following rules, by essentially
undoing Section 8:

1. Derivative of power law: for any α ∈ R \ { −1 } from Claim 8.16,

Z b Z b
1
xα dx = ∂ xa+1 dx
a a a + 1
b
1 α+1 

= x .
α+1 
x=a

(Note indeed this formula would not make sense for α = −1 due to α + 1
in the denominator).

94
2. Derivative of reciprocal from Claim 8.19:
Z b Z b
1
dx = log0
a x a
b
= log|a .

3. Derivative of trigonometric functions Example 8.10:

Z b Z b
sin = − cos0
a a
Z b
= − cos0
a
b
= − cos|a

and Example 8.9 yields

Z b Z b
cos = sin0
a a
b
= sin|a .

4. Derivative of exponential from Claim 8.20:

Z b Z b
exp = exp0
a a
b
= exp|a .

5. Derivative of hyperbolic trigonometric functions from Example 8.23:

Z b Z b
sinh = cosh0
a a
b
= cosh|a

and
Z b Z b
cosh = sinh0
a a
b
= sinh|a .

6. More complicated functions, e.g., from Example 8.30

Z b Z b
1
= tan0
a cos2 a
b
= tan|a

95
 and
Z b Z b
1
= tanh0
a cosh2 a
b
= tanh|a .

From Example 8.35 we get

Z b Z b
1
√ dx = arcsin0
a 1 − x2 a
b
= arcsin|a .

9.35 Theorem. (Change of variables) Let f : [a, b] → R be continuous (and

hence integrable by Theorem 9.16) and ϕ : [A, B] → [a, b] be continuous such
that ϕ0 : [A, B] → R is continuous. Then
Z ϕ(B) Z B
f = (f ◦ ϕ) ϕ0 .
ϕ(A) A

Rx
Proof. Define F : [a, b] → R by F (x) := a f . Then by Theorem 9.30, since
f is continuous, F 0 = f . Let us calculate, using Theorem 9.32
Z B Z B
0
(f ◦ ϕ) ϕ = (F 0 ◦ ϕ) ϕ0
A A
Z B
0
= (F ◦ ϕ)
A
(Use the fundamental theorem of calc.)
= F (ϕ (B)) − F (ϕ (A))
Z ϕ(B)
= F0
ϕ(A)
Z ϕ(B)
= f.
ϕ(A)

9.36 Example. Consider the function

g : [0, 2] → R
p
x 7→ 2x x2 + 1 .

We are interested in Z 2 Z 2 p
g≡ 2x x2 + 1dx .
0 0

96
To this end, let us try to use the change of variables theorem above. Define
ϕ : [0, 2] → [1, 5] by
ϕ (x) := x2 + 1
Then ϕ is continuously differentiable, and ϕ0 (x) = 2x, which is a linear function,
which we already saw was integrable (Example 9.11). Also note that
p
g (x) = ϕ0 (x) ϕ (x)
and so applying
√ the change of variables theorem with f : [1, 5] → R defined via
f (x) := x for all x ∈ [1, 5] we find that
Z 2 p Z 5
2
√
2x x + 1dx = xdx .
0 1
√ √
The point being, it is much easier to integrable x 7→ x than x 7→ 2x x2 + 1.
√
The key here was to define ϕ and to observe that g = ϕ0 ϕ.
Now using Remark 9.34 we find that
Z 5 5
√ 2 3 
xdx = x2 
1 3 1
2 3 
= 52 − 1 .
3
9.37 Example. Let us consider the integral
Z 4
x
dx.
2 2
0 (1 + x )

Let us define ϕ (x) := 1 + x2 , in which case ϕ0 (x) = 2x so that if f (x) := 1

2x2
then
x 1 ϕ0 (x)
=
(1 + x2 ) 2 ϕ (x)2
= f (ϕ (x)) ϕ0 (x)
then the change of variables Theorem 9.35 says
Z 4 Z 4
x
dx = ϕ0 (f ◦ ϕ)
2 2
0 (1 + x ) 0
Z ϕ(4)
= f
ϕ(0)
Z 17
1
= dx
1 2x2
x=17
1
− x−1 

=
2 x=1
 
1 1
= 1− .
2 17

97
9.38 Example. Consider Z 4
1
dx .
0 1 + x2
Define ϕ (x) := tan (x). Then ϕ0 (x) = cos(x)
1
2 and so if f (x) :=
1
1+x2 then
change of variables says
Z 4 Z 4
1
2
dx = f
0 1+x 0
Z ϕ(ϕ−1 (4))
= f
ϕ(ϕ−1 (0))
Z ϕ−1 (4)
= (f ◦ ϕ) ϕ0
ϕ−1 (0)
Z ϕ−1 (4)
1 1
= .
ϕ−1 (0) 1 + tan2 cos2

1 1 cos2
But now, 1+tan2 = sin2
= cos2 + sin2
= cos2 and hence we find
1+ cos2

1 1
= 1.
1 + tan2 cos2
Rd
Since c
= d − c we have
Z 4
1
dx = ϕ−1 (4) − ϕ−1 (0)
0 1 + x2

But ϕ−1 ≡ tan−1 ≡ arctan−1 . So the integral equals

Z 4
1
dx = arctan (4) − arctan (0) .
0 1 + x2

9.39 Remark. By the way, Example 9.38 raises an interesting point: since

sin (x)
limπ tan (x) = limπ
x→ 2 x→ 2 cos (x)
= ∞

we must have (via the definition arctan ≡ tan−1 )

π
lim arctan (x) =
x→∞ 2
and similarly
π
lim arctan (x) = − .
x→−∞ 2

98
Hence it makes sense to think of
Z b
1
dx = arctan (b) − arctan (a)
a 1 + x2
with a → −∞ and b → +∞, and the result is called an improper integral. I.e.,
Z b
1
lim lim dx = lim arctan (b) − lim arctan (a)
a→−∞ b→∞ a 1 + x2 b→∞ a→−∞
π  π
= − −
2 2
= π.

One writes this in compact form as

Z ∞
1
dx = π .
−∞ 1 + x2
Since this result involves a limit of the end points a and b, it is called an improper
integral.
9.40 Definition. An improper integral is the result of the limit (if it exists)
after integrating
Z b
lim lim f,
a→a0 b→b0 a
where a0 ∈ R or a0 = −∞ and b0 ∈ R or b0 = ∞.
9.41 Theorem. (Integration by parts) If f, g : [a, b] → R are differentiable such
that f 0 , g 0 are integrable, then
Z b Z b
f g 0 = f (b) g (b) − f (a) g (a) − f 0g .
a a

Proof. Using Claim 8.14, we have

0
(f g) = f 0 g + f g0 .

But Theorem 9.32 says that

Z b
0 b
(f g) = f g|a .
a

We learn that
Z b Z b
b
f 0g + f g0 = f g|a .
a a

which is essentially the result.

99
9.42 Example. Consider the function R 3 x 7→ x sin (x) ∈ R. Let us define
f (x) := x (with the same domains and co-domains). Then
Z b Z b
x sin (x) dx = f sin
a a
Z b
= − f cos0
a
(integration by parts)
Z b
b
= −f cos|a + f 0 cos
a
(Use f 0 = 1)
Z b
b
= −f cos|a + cos
a
Z b
b
= −f cos|a + sin0
a
(Use fundamental thm. of calc.)
b
= −f cos + sin|a .

The point is we know how to integrate trigonometric functions, but we don’t

know how to integrate x 7→ x sin (x), and that’s when integration by parts could
help.
9.43 Example. [Courant] We have
Z b Z b
0
xex dx = x (x 7→ ex ) dx
a a
Z b
b
= xex |a − ex dx
a
b
= ex (x − 1)|a .

9.44 Example. More generally,

Z b Z b
b
xf 0 (x) dx = xf (x)|a − f.
a a

9.45 Example. [Courant] We have

Z b Z b
0
arcsin = arcsin (x 7→ x)
a a
Z b
b
= x arcsin (x)|a − x arcsin0
a
Z b
b x
= x arcsin (x)|a − √ dx
a 1 − x2

100
for this last integral we do a change of variables with ϕ (x) := 1 − x2 so that
ϕ0 (x) = −2x and hence with f (x) := √1x we find
Z b Z b
x 1 0
√ dx =
ϕ (x) f (ϕ (x)) dx
a a −2
1 − x2
1 ϕ(b)
Z
= − f
2 ϕ(a)
1 √ ϕ(b)
= − 2 xϕ(a) .
2
Rb
9.46 Example. [Courant] Consider a eαx sin (βx) dx. In this example re-
peated integration by parts will result in an algebraic equation:
Z b Z b  0
αx αx 1
e sin (βx) dx = − e x 7→ cos (βx) dx
a a β
b Z b
1 1
= − eαx cos (βx) + αeαx cos (βx) dx

β a a β
b Z b  0
1 α 1
= − eαx cos (βx) + eαx x 7→ sin (βx) dx

β a β a β
b b !
α b αx
Z
αx 1 α αx 1
 
= −e cos (βx) +
 e sin (βx) −
 e sin (βx) dx
β a β β a β a
b
α2 b αx
 Z
1 α
− eαx cos (βx) + 2 eαx sin (βx)  − 2

= e sin (βx) dx .
β β a β a
Rb
We solve this for a
eαx sin (βx) dx to get
Z b  b
αx 1 1 αx α αx 
e sin (βx) dx = 2 − e cos (βx) + 2 e sin (βx) 
a 1+ αβ2
β β a
1 b
= eαx [α sin (βx) − β cos (βx)]|a .
β 2 + α2
9.47 Remark. We summarize that our main tools to evaluate integrals is combine
Remark 9.34 together with Theorem 9.23, Theorem 9.35 and Theorem 9.41.
This is not a lot, and indeed most functions which can be integrated don’t
admit an explicit formula for their integral.

10 Important functions
10.1 The trigonometric functions
Recall the definitions and properties of sin, cos, tan, cot etcetera discussed in
Section 5.1.2.

101
 sin
• The tangent function is defined as the quotient tan ≡ cos whenever cosine
is non-zero (so one must restrict its domain of definition).
• The cotangent function is defined as the quotient cot ≡ cos
sin whenever sine
is non-zero (so one must restrict its domain of definition).

• These two definitions mean that one must mainly keep in mine sin and
cos whereas properties of tan and cot may be inferred by the quotient
definition.
• There is a special (irrational) number, denoted by π, equal to approxi-
mately 3.1415. Geometrically it is the ratio of a circle’s circumference to
its diameter. It is also a convenient way to measure arc lengths on the
circle of radius 1 for that reason: an arc-length of 2π is the entire circle,
π is half the circle, π2 is one-quarter of it, etc. Naturally the trigonomet-
ric functions, which related to arc-lengths of the unit circle, have special
values corresponding to special multiples of π:
– sin (nπ) = 0 for all n ∈ Z.
n
– cos (nπ) = (−1) for all n ∈ Z.
n
– sin n π2 = − (−1) for all n ∈ Z.

– cos n π2 = 0 for all n ∈ Z.

• The trigonometric functions are periodic of period 2π:

cos (x + 2π) = cos (x) (x ∈ R)

sin (x + 2π) = sin (x) (x ∈ R)

π
• The sine and cosine are related by a shift of the angle in 2:
 π
sin (x) = cos x − (x ∈ R)
2

• They obey the pythagoras law:

2 2
cos (x) + sin (x) = 1 (x ∈ R)

• Their image is in [−1, 1] and they are continuous throughout their domain
R. They do not have limits at ±∞ as they keep oscillating.

10.2 The exponential and logarithmic functions

Recall the definitions and properties of expa and loga for a > 1 discussed in
Example 5.31:
1. loga is defined on (0, ∞) and has as its image the whole of R. It is strictly
monotone increasing. It is a continuous function.

102
 2. loga (1) = 0.
3. loga (a) = 1.
4. limx→0 loga (x) = −∞.

5. limx→∞ loga (x) = +∞.

6. loga (xy) = loga (x) + loga (y).
 
7. loga xy = loga (x) − loga (y).

8. loga (xp ) = p loga (x) for all p ∈ R.

9. loga (x) ≤ Ca (x − 1) for some strictly positive constant (independent of
x) Ca .

10. expa is defined on R, and has as its image (0, ∞) (so it is always strictly
positive). It is strictly monotone increasing. It is a continuous function.
11. expa (0) = 1.
12. limx→−∞ expa (x) = 0.

13. limx→∞ expa (x) = ∞.

14. expa (1) = a.
15. expa (x + y) = expa (x) expa (y).
expa (x)
16. expa (x − y) = expa (y) .
p
17. expa (px) = (expa (x)) for all p ∈ R.

• In the context of evaluating limits, it is useful to know that

x = expa (loga (x)) (x > 0)

x = loga (expa (x)) (x ∈ R)

10.2.1 The natural base for the logarithm

n
10.1 Claim. The limit limn→∞ 1 + n1 exists and equals some number between
2 and 3.

103
Proof. We have by the binomial theorem (see below)
 n n
1 X n! 1
1+ =
n j=0
j! (n − j)! nj
n
X n (n − 1) · · · (n − j + 1) (n − j)! 1
=
j=0
j! (n − j)! nj
n
X 1 n (n − 1) · · · (n − j + 1)
=
j=0
j! nj
n    
X 1 1 j−1
= 1− ··· 1 −
j=0
j! n n
        
1 1 1 1 2 1 1 n−1
= 1+1+ 1− + 1− 1− + ··· + 1− ··· 1 −
2! n 3! n n n! n n

So that
 n  n+1
1 1
1+ − 1+
n n+1
n     n+1
X 1    
X 1 1 j−1 1 j−1
= 1− ··· 1 − − 1− ··· 1 −
j=0
j! n n j=0
j! n+1 n+1
n     X n    
X 1 1 j−1 1 1 j−1
= 1− ··· 1 − − 1− ··· 1 − −
j=0
j! n n j=0
j! n+1 n+1
   
1 1 n
− 1− ··· 1 −
(n + 1)! n+1 n+1
n        
X 1 1 j−1 1 j−1
= 1− ··· 1 − − 1− ··· 1 − −
j=0
j! n n n+1 n+1
   
1 1 n
− 1− ··· 1 −
(n + 1)! n+1 n+1
 
1 1
Using ≤
n+1 n
n        
X 1 1 j−1 1 j−1
≤ 1− ··· 1 − − 1− ··· 1 − −
j=0
j! n n n n
   
1 1 n
− 1− ··· 1 −
(n + 1)! n+1 n+1
   
1 1 n
= − 1− ··· 1 −
(n + 1)! n+1 n+1
≤ 0

104
 1 n


This shows that the sequence N 3 n 7→ 1 + n is increasing.
It is also bounded, since
 n n    
1 X 1 1 j−1
1+ = 1− ··· 1 −
n j=0
j! n n
 
j−1 1
≥ for all j
n 2
n
X 1
≤ 1+
j=0
2j
Pn
The latter sequence, N 3 n 7→ 1 + j=0 21j actually converges to 3. Be-

n
ing monotone increasing and bounded, N 3 n 7→ 1 + n1 converges by
Claim 6.17. Since it is increasing and the first term is larger than 2, we
find that  n
1
lim 1 + exists and ∈ (2, 3)
n→∞ n

n
10.2 Definition. We define e := limn→∞ 1 + n1 and call it the natural base
for the logarithm. It turns out that e ≈ 2.718 and is irrational. When we don’t
write the subscript a for expa or loga , we mean a = e. The reason why it is
called the natural base for the logarithm is because its derivative is given by
1
log0 (x) =
x
and also that

log (x) ≤ x−1

(i.e. the constant Ce = 1).

10.3 Claim. We have log (x) ≤ x − 1 for any x ∈ (0, ∞).

Proof. Let us first assume that x > 1. Then we may apply the mean value
theorem Theorem 8.48 to get that there must be some y ∈ (1, x) such that

log (x) − log (1)

log0 (y) =
x−1
Now
1
log0 (y) =
y

105
 and log (1) = 0, so we find

1 log (x)
=
y x−1
1
Since y ∈ (1, x), y > 1 so y < 1, and so we find

log (x)
1 <
x−1
moving the denominator to the other side of the inequality we get the result.
When x < 1, we have again by an application of Theorem 8.48 that there
is some y ∈ (x, 1) such that

1 log (1) − log (x)

=
y 1−x
1
However now y > 1 so that

− log (x)
1 >
1−x
and we find the same result.
If x = 1, then log (x) = 0 and x − 1 = 0 so the inequality is actually an
equality.

10.3 The hyperbolic functions

Recall from Example 8.23 where we introduced the hyperbolic trigonometric
functions, which are analogous to the trigonometric functions. This follows
from the Euler identities
1 ix
e − e−ix


sin (x) =
2i
and
1 ix
e + e−ix


cos (x) =
2
(If that means nothing to you because of the i, then ignore this short introduc-
tion). By analogy we define, with R being the domain and codomain,
1 x
e − e−x


sinh (x) :=
2
and
1 x
e + e−x


cosh (x) :=
2

106
10.4 Special functions
10.4.1 The sinc function
The sinc function, which is discussed in Example 6.32, is defined as
(
1 x=0
R 3 x 7→ sinc (x) := sin(x) ∈R
x x 6= 0

The example cited shows that this function is continuous at the origin. It
converges to zero at ±∞.

10.4.2 The square root

√
The square-root function · : [0, ∞) → [0, ∞) is continuous, positive, and
monotone increasing. Furthermore it obeys
√ √ √
x+y ≤ x+ y (x, y ≥ 0)

11 Useful algebraic formulas to recall

11.1 Factorizations
11.1 Claim. Newton’s binomial theorem: If a, b ∈ R and n ∈ N then
n
n
X n!
(a + b) = aj bn−j
j=0
j! (n − j)!

11.2 Claim. If a, b ∈ R then

a2 − b2 = (a − b) (a + b)

In fact, we have the more general formula for n ∈ N:

n−1
X
an − bn = (a − b) an−k−1 bk
k=0

which is proven in Example 6.34.

These formulas are especially useful when one is dealing with square roots,
e.g.
√ √ a−b
a− b = √ √ .
a+ b

11.3 Claim. Recall if ax2 + bx + c = 0 for some a, b, c ∈ R then the two solutions
x1 , x2 ∈ R can be written as
1  p 
x1,2 = −b ± b2 − 4ac
2a

107
given that b2 − 4ac > 0. If b2 − 4ac = 0 then there is only one solution. If
b2 − 4ac < 0 then there are no solutions. If there are two solutions, then we
may factorize
   
2 1  p
2
1  p
2
ax + bx + c = x− −b − b − 4ac x− −b + b − 4ac
2a 2a

and if there is only one solution then we factorize

 2
1
ax2 + bx + c = x− (−b)
2a
PN
11.4 Claim. We have n=1 n = 21 N (N + 1) for all N ∈ N.

11.2 Inequalities
1
• If a < b and a, b > 0 then a > 1b .

• If a < b then −a > −b.

First of all it is important to note that, as in Claim 6.1

11.5 Claim. For any ε > 0, the inequality (for the unknown x ∈ R)

|x| < ε

is equivalent to the two simultaneous inequalities (for the unknown x ∈ R)

−ε < x <ε

which is sometimes useful in algebraic manipulations of limits.

11.6 Claim. The inequality ax2 + bx + c < α for some a, b, c, α ∈ R has to be
solved as follows: If a > 0, this is an upwards parabola. Hence to be smaller
than some value (α), the unknown x must be constrained to the interior between
two points x1 , x2 . These two points are given by the solution to the equation
ax2 + bx + c − α = 0, assuming b2 − 4a (c − α) > 0, so that there are two unique
solutions. If on the other hand a < 0, this is a downwards parabola, and so the
unknown x must be constrained to the left or right of x1 and x2 respectively,
assuming x1 < x2 .
If x1 = x2 more care has to be taken: If a > 0, then any x solves the
inequality and if a < 0 then no x solves the inequality.
11.7 Claim. If f is monotone increasing and a < b then f (a) ≤ f (b). If f is
strictly monotone increasing then f (a) < f (b). If f is monotone decreasing
then f (a) ≥ f (b) and if f is strictly monotone decreasing then f (a) > f (b).
This can be useful, for example, the function x 7→ x2 is monotone increasing
for x > 0, so that if 0 < a < b then a2 < b2 .

108
12 Dictionary of graphical symbols and acronyms
12.1 Graphical symbols
1. The equivalent symbol, ≡, means the following: if a and b are any two
objects, then a ≡ b means we have agreed, at some earlier point, that
a and b are two different labels for one and the same thing. Example:
{ a, a } ≡ { a }.
2. The definition symbol, :=, means the following: if a and b are any two
objects, then a := b means that right now, through this very equation, we
are agreeing that a is a new label for the pre-existing object b. The main
difference to ≡ is about when this agreement happens: The ≡ symbol is
a reminder about our previous conventions, whereas the := symbol is an
event of establishing a new convention. These both should be contrasted
with the equal sign =, which merely says that two things (turn out) to be
equal, whether by convention or not is not specified.
3. ∞ means the size of a set whose number of elements is unbounded.
(
x x≥0
4. x 7→ |x| means the absolute value function, i.e. |x| ≡ .
−x x < 0

5. The maximum or minimum of a set of numbers is the largest element of

that set. Note this only works when we take the maximum or minimum
or elements in R (or its subsets)!

max ({ 1, 2, 100 }) = 100

min ({ 1, 2, 100 }) = 1

6. Sometimes it is useful to write out the sum of numbers in a compact way.

ThePgraphical symbol for that is the capital Greek letter Sigma, written
as . We use it as follows: Let us assume we have a sequence a : N → R
and S ⊆ N is given. For example S = { n1 , n2 , . . . , nN }. Then
X
a (n) ≡ a (n1 ) + a (n2 ) + · · · + a (nN )
n∈S

For instance if S = { 1, 2, . . . , N } then we usually write

N
X
a (n) ≡ a (1) + a (2) + · · · + a (N )
n=1

7. The factorial of a number n! is a short-hand notation for the following

arithmetic operation:

n! ≡ n (n − 1) (n − 2) . . . 2

109
 so that

2! = 2
3! = 3·2=6
4! = 4 · 3 · 2 = 24
5! = 5 · 4 · 3 · 2 = 120
···

From this we see immediately that n! = n (n − 1)! for any n ∈ N.

12.2 Acronyms
1. s.t. means “such that”.
2. w.r.t. means “with respect to”.
3. l.h.s. means “left hand side”, usually of an equation. r.h.s. means “right
hand side”.
4. iff means “if and only if”, which is a relationship between two statements,
meaning the first implies the second and the second implies the first.
5. “The origin” means either 0 ∈ R or (0, 0) ∈ R2 .

6. WLOG means “without loss of generality”.

References
[1] Tom M. Apostol. Calculus, Vol. 1: One-Variable Calculus, with an Intro-
duction to Linear Algebra. Wiley, 1991.
[2] Richard Courant and Fritz John. Introduction to Calculus and Analysis,
Vol. 1 (Classics in Mathematics). Springer, 1998.
[3] Paul R. Halmos. Naive Set Theory. Martino Fine Books, 2011.

[4] Walter Rudin. Principles of Mathematical Analysis (International Series in

Pure and Applied Mathematics). McGraw-Hill Education, 1976.
[5] Michael Spivak. Calculus, 4th edition. Publish or Perish, 2008.
[6] James Stewart. Calculus: Early Transcendentals 8th Edition. Cengage
Learning, 2015.

110