MATH 216—Introduction to Analysis

(Fall 2021)
Volker Runde

December 11, 2021

Contents
Introduction iii

Introduction (2020) iv

Introduction (2019) v

List of Symbols vi

0 Why prove things? 1

1 The Set N and Induction 6

2 The Set Q of Rational Numbers 17

3 The Set R of Real Numbers 19

4 The Difference between R and Q 24

5 Functions 30

6 Convergence of Sequences 33

7 lim sup, lim inf, and Cauchy Sequences 43

8 Subsequences 48

9 Infinite Series 53

10 Convergence Criteria 59

11 More on Absolute Convergence 64

12 The Exponential Function 69

13 Closed and Compact Subsets of R 74

14 Limits of Functions and Continuity 77

15 Properties of Continuous Functions and Uniform Continuity 83

16 Continuity of Inverse Functions 87

17 Differentiation 91

i
18 Local Extrema, the Mean Value Theorem, and Taylor’s Theorem 98

19 The Riemann Integral 104

20 Integration and Differentiation 116

21 Riemann Sums 120

22 The Complex Numbers C, and the Truth about exp, cos, and sin 125

Index 134

ii
 Anyone who cannot cope with mathematics is not fully human. At best he
is a tolerable subhuman who has learned to wear shoes, bathe, and not make
messes in the house.
Robert A. Heinlein, Time Enough for Love (1973).

Introduction
This is the third—and presumably, at least for the foreseeable future, last—incarnation
of my notes for MATH 216. Beyond the inevitable debugging and some nip-and-tuck, the
most significant difference relative to the previous version is the addition of exercises.

Volker Runde December 2021, Edmonton.

iii
 Mathematics rests on proof—and proof is eternal.
Saunders Mac Lane in: Responses to “Theoretical Mathematics: Toward a cultural
synthesis of mathematics and theoretical physics” by A. Jaffe and F. Quinn (1994).

Introduction (2020)
This is the second iteration of my notes for MATH 216. Several changes have been made
relative to the 2019 version:

• There are now a list of symbols and an index, which should make navigating the
notes easier.

• Some graphics have been added.

• A new section—numbered as zero—has been added: it is intended to be an appetizer

for the course, pointing out the necessity of proof in mathematics and giving some
examples for the method of indirect proof.

• The exponential function is now treated in greater detail and has its own section.

• Some material has been added to the section on Riemann sums.

Finally, and perhaps most importantly, the embarrassingly high number of typos has been
substantially reduced.

Volker Runde December 2020, Edmonton.

iv
 Ich behaupte aber, daß in jeder besonderen Naturlehre nur so viel eigentliche
Wissenschaft angetroffen werden könne, als darin Mathematik anzutreffen ist.
Immanuel Kant, Metaphysische Anfangsgründe der Naturwissenschaft (1786).

Introduction (2019)
These are the notes for MATH 216 as tought at the University of Alberta in the Fall Term
2019: it is a second year course aimed at students who have taken a full year of calculus,
but not yet been exposed to doing mathematics rigorously. The topics covered are the
same as in every first year cookie cutter course in calculus, but the level of treatment is
different. The focus is on a thorough development of the theory behind the calculations
of calculus. In particular, the aim was to introduce the students to techniques of proof
such as induction, indirect proof, and -δ arguments.
Up to Section 8, these notes follow

Kenneth A. Ross, Elementary Analysis, Second Edition. Springer Verlag

(2013)

fairly closely whereas from Section 14 onward, they are more or less patterned after

Otto Forster, Analysis 1. Vieweg (1984).

Volker Runde December 2019, Edmonton.

v
List of Symbols

∞, 27 lim supn→∞ xn = x, 43
−∞, 27 limn→∞ xn = x, 33
bxc, 48 limn→∞ xn = ∞, 37
√
x, 32 log x, 30
√n
x, 61
max S, 24
arctan x, 32 min S, 24

C, 125 N, 6
cos x, 30, 130 N0 , 11
n


df k , 12
dx (x0 ), 91 n!, 11
df
dx x=x , 91
d2 f
0 π, 131
dx2
, 96 Qn
dn f k=1 ak , 70
dxn , 96 P(S), 9
e, 69 Q, 17
exp x, 30, 69
exp z, 129 R, 19

f 0 (x0 ), 91 sin x, 30, 130

f 00 , 96 Sn , 32
f (n) , 96 (sn )∞ , 33
R b n=n0
x→x f (x) dx, 109
f (x) −→0 y0 , 77 Rab
f (x) → y0 , 77 f (x) dx, 107
∗a
x→x
f (x) −→0 ∞, 77 R∗ b
f (x) dx, 107
f (x) → ∞, 77 Pan
x→x ak , 6
f (x) −→0 −∞, 77 Pk=1
∞
f (x) → −∞, 77 k=1 ak , 53
sup S, 25
inf S, 25
tan x, 32
limx→x0 f (x) = y0 , 77 n→∞
xn −→ ∞, 37
limx→x0 f (x) = ∞, 77 n→∞
xn −→ x, 33
limx→x0 f (x) = −∞, 77
xn → ∞, 37
lim inf n→∞ xn = −∞, 43
xn → x, 33
lim inf n→∞ xn = x, 43
lim supn→∞ xn = ∞, 43 Z, 17

vi
0 Why prove things?
Once upon a time, there was a prince who was educated by private tutors. One day, the
math tutor set out to explain the Pythagorean Theorem to his royal student. The prince
wouldn’t believe it. So, the teacher proved the theorem, but the prince was not convinced.
The teacher presented another proof of the theorem, and then yet another, but the prince
would still shake his head in disbelief. Desperate, the teacher exclaimed: “Your royal
highness, I give you my word of honor that this theorem is true!” The prince’s face lit up:
“Why didn’t you say so right away?!”
Wouldn’t that be wonderful? A simple word of honor from the teacher, and the student
accepts the theorem as true. . .
Of course, it would be awful. Who makes sure that the teacher can be trusted? Where
did he get his knowledge from? Did he rely on another person’s word of honor? Was the
person from whom the teacher learned the theorem trustworthy? Where did that person
get their knowledge from? Did that person, too, trust someone else’s word of honor? The
longer the chain of words of honor gets, the shakier the theorem starts to look. It can’t
go on indefinitely: someone must have established the truth of the theorem some other
way—that someone must have proved it.
Why are mathematicians so obsessed with proofs? The simple answer is: because
they are obsessed with the truth. A proof is a procedure which, by applying certain rules,
establishes an assertion as true. Proofs do not only occur in mathematics. In a criminal
trial, for instance, the prosecution tries to prove that the defendant is guilty. Of course,
the rules according to which a proof is carried out depend very much on the context: they
are quite different in criminal law, experimental science, and mathematics.
Consider a chessboard consisting of 64 squares:

111111
000000
000000
111111
000000
111111
Figure 1: Chessboard and tiles

1
 Then take rectangular tiles as shown below the board: each of them covers precisely
two adjacent squares on the chessboard. It’s obvious that you can cover the entire board
with such tiles without any two of them overlapping. That’s straightforward, so why do
we need proof here?
To make things slightly more complicated, take a pair of scissors to the chessboard
and cut away the squares in the upper left and in the lower right corner. This is how it
will look like:

Figure 2: Chessboard with two squares removed

Now, try to cover this altered chessboard with the tiles without any two of them
overlapping. . .
If you really try this (preferably with a chessboard drawn on a piece of paper. . . ), you’ll
soon find out that—to say the least—it’s not easy, and maybe the nagging suspicion will
set in that it’s not even possible—but why?
There is, of course, the method of brute force to find out. There are only finitely many
ways to place the tiles on the chessboard, and if we try them all and see that in no case
the area covered by them is precisely the altered chessboard, then we are done. There
are two problems with this approach: firstly, we need to determine every possible way to
arrange the tiles on the chessboard, and secondly, even if we do, the number of possible
tile arrangements may be far too large for us to check them all. So, goodbye to brute
force. . .
So, if brute force fails us, what can we do? Remember, we are dealing with a chess-
board, and a chessboard not only consists of 64 squares in an eight by eight pattern—the
squares alter in color; 32 are white, and 32 are black:

2
 Figure 3: Colored chessboard

Each tile covers precisely two adjacent squares on the board, and two adjacent squares
on a chessboard are always different in color; so each tile covers one white square and one
black square. Consequently, any arrangement of tiles on the chessboard must cover the
same number of white and of black squares. But now, check the altered chessboard:

Figure 4: Colored chessboard with two squares removed

We removed two white squares, so the altered board has 30 white squares, but 32
black ones. Therefore, it is impossible to cover it with tiles—we proved it.
Does this smell a bit like black magic? Maybe, at the bottom of your heart, you
prefer the brute force approach: it’s the harder one, but it’s still doable, and maybe you
just don’t want to believe that the tiling problem is unsolvable unless you’ve tried every
possibility.
There are situations, however, where a brute force approach to truth is not only
inconvenient, but impossible. Have a look at the following mathematical theorem:

Theorem 0.1. Every integer greater than one is a product of prime numbers.

Is it true? And if so, how do we prove it?

3
 Let’s start with checking a few numbers: 2 is prime (and thus a product of prime
numbers), so is 3, 4 = 2 · 2, 5 is prime again, 6 = 2 · 3, 7 is prime, 8 = 2 · 2 · 2, 9 = 3 · 3,
and 10 = 2 · 5. So, the theorem is true for all integers greater than 2 and less than or
equal to 10. That’s comforting to know, but what about integers greater than 10? Well,
11 is prime, 12 = 2 · 2 · 3, 13 is prime, 14 = 2 · 7, 15 = 3 · 5, 16 = 2 · 2 · 2 · 2, . . . I stop
here because it’s useless to continue like this. There are infinitely many positive integers,
and no matter how many of them we can write as a product of prime numbers, there will
1010
always remain infinitely many left for which we haven’t shown it yet. Is 1010 +1 a
product of prime numbers? That number is awfully large. Even with the help of powerful
computers, it might literally take an eternity to find the prime numbers whose product it
is (if they exist. . . ). And if we have shown that the theorem holds true for every integer
10
1010 1010
up to 1010 + 1, we still don’t know about 1010 + 1.
Brute force leads nowhere here. Checking the theorem for certain examples might give
you a feeling for it—but it doesn’t help to establish its truth for all integers greater than
one.
Is the theorem possibly wrong? What would that mean? If not every integer greater
than one is a product of prime numbers, then there must be at least one integer a0 which
is not a product of prime numbers. Maybe, there is another integer a1 with 1 < a1 < a0
which is also not a product of prime numbers; if so replace a0 by a1 . If there is an integer
a2 with 1 < a2 < a1 which is not a product of prime numbers, replace a1 by a2 . And
so on. . . There are only finitely many numbers between 2 and a0 , and so, after a finite
number of steps, we hit rock bottom and wind up with an integer a > 1 with the following
properties: (a) a is not a product of prime numbers, and (b) it is the smallest integer with
that property, i.e., every integer greater than one and less than a is a product of prime
numbers.
Let’s think about this (hypothetical) number a. It exists if the theorem is false. What
can we say about it? It can’t be prime because then it would be a product (with just one
factor) of prime numbers. So, a isn’t prime, i.e., a = bc with neither b nor c being a or
1. This, in turn, means that 1 < b, c < a. By property (b) of a, the numbers b and c are
thus products of prime numbers, i.e., there are prime numbers p1 , . . . , pn , q1 , . . . , qm such
that b = p1 · · · pn and c = q1 · · · qm . But then

a = bc = p1 · · · pn q1 · · · qm

holds, and a is product of prime numbers, which contradicts (a).

We assumed that the theorem was wrong, and—based on that assumption—obtained
an integer a that is not a product of prime numbers only to see later that this was not
possible. The only way out of this dilemma is that our assumption was wrong: the theorem
1010
is true! (And we now know that 1010 + 1 is a product of prime numbers without having

4
to find them. . . )
The strategy we used to prove Theorem 0.1 is called indirect proof. We can’t show
something directly, so we assume it’s wrong and (hopefully) arrive at a contradiction.
Let’s try another (indirect) proof:

Theorem 0.2. There are infinitely many prime numbers.

Is this believable? There is no easy formula to calculate the n-th prime number, and
after putting down the first few prime numbers, it gets harder and harder to come up with
the next prime. So, is the theorem wrong and do we simple run out of prime numbers
after a while?
Assume this is so: there are only finitely many prime numbers, say p1 , . . . , pn . Set
a := p1 · · · pn + 1. By Theorem 0.1, a is a product of prime numbers. In particular, there
are a prime number q and a non-negative integer b with a = qb. Since p1 , . . . , pn are all
the prime numbers there are, q must be one of them. Let c be the product of all those pj
that aren’t q, so that a = qc + 1. We then obtain

0 = a − a = qc + 1 − qb = q(c − b) + 1,

and thus q(c − b) = −1. This, however, is impossible because c − b is a non-zero integer
and q ≥ 2.
We have thus again reached a contradiction, and Theorem 0.2 is proven.

5
1 The Set N and Induction
We denote the set {1, 2, 3, 4, . . .} of natural numbers by N.
Consider the sum of the first n odd natural numbers:
n = 1: 1 = 1;

n = 2: 1 + 3 = 4 = 22 ;

n = 3: 1 + 3 + 5 = 9 = 32 ;

n = 4: 1 + 3 + 5 + 7 = 16 = 42 .
This suggests:
Guess. For all n ∈ N, the sum of the first n odd integers is
n
X
1 + 3 + · · · + (2n − 1) = (2k − 1) = n2 .
k=1

How do we prove that this is true for all n ∈ N?

Properties of N.
(N 1) 1 ∈ N;

(N 2) if n ∈ N, then it has a successor n + 1 ∈ N;

(N 3) 1 is not a successor of any element in N;

(N 4) if n, m ∈ N have the same successor, then n = m;

(N 5) every ∅ 6= S ⊂ N has a smallest element.

Claim. Let S ⊂ N be such that:
(a) 1 ∈ S;

(b) if n ∈ S, then n + 1 ∈ S.
Then S = N.
Proof. Assume that S 6= N. Then

N \ S = {n ∈ N : n ∈ 6 ∅.
/ S} =

By (N 5), N \ S then must have a smallest element, say n0 . As 1 ∈ S, n0 6= 1 must hold.

Therefore n0 − 1 ∈ N exists and n0 − 1 ∈
/ N \ S, i.e., n0 − 1 ∈ S. But this means that

n0 = (n0 − 1) + 1 ∈ S,

which is a contradiction.
Therefore, S = N must hold.

6
Proof of our Guess. Set
n
( )
X
S := n∈N: (2k − 1) = n2 .
k=1

It is clear that 1 ∈ S. Suppose that n ∈ S. We will show that n + 1 ∈ S as well. Consider

n+1
X n
X
(2k − 1) = (2k − 1) + (2(n + 1) − 1)
| {z }
k=1 k=1 =2n+1
2
= n + 2n + 1
= (n + 1)2 .

This completes the proof.

We can distill something far more general out of this:

Principle of Mathematical Induction. For n ∈ N, let P (n) be a claim about n ∈ N

that may be true or not. Suppose that:

(a) P (1) is true;

(b) whenever P (n) is true, then P (n + 1) also true.

Then P (n) is true for all n ∈ N.

Proof. Set
S := {n ∈ N : P (n) is true}.

By (a), 1 ∈ S holds and by (b), n ∈ S implies n + 1 ∈ S. By the previous claim, this

means that S = N.

Example. We claim that

n
X n(n + 1)
k=
2
k=1

for all n ∈ N. According to the Principle of Induction, we proceed as follows.

n = 1 (induction anchor ): Obviously,
1
X 1·2
k=1=
2
k=1

holds.

7
 Pn n(n+1)
n n + 1 (induction step): Suppose that n ∈ N is such that k=1 = 2 . (This is
called the induction hypothesis). It follows that
n+1
X n
X
k= k + (n + 1)
k=1 k=1
n(n + 1)
= + (n + 1), by the induction hypothesis,
2
n(n + 1) 2(n + 1)
= +
2 2
(n + 1)(n + 2)
= ,
2
which proves that the claim is also true for n replaced by n + 1. By the Principle of
Induction, this proves the claim for all n ∈ N.
Example. We claim that 7n − 6n − 1 is divisible by 36 for all n ∈ N.
n = 1 (induction anchor ): Obviously,
71 − 6 − 1 = 0
is divisible by 36.
n n + 1 (induction step): Note that
7n+1 − 6(n + 1) − 1 = 7n+1 − 6n − 6 − 1
= 7n+1 − 6n − 7
= 7n+1 − 42n − 7 + 36n
= 7(7n − 6n − 1) + 36n.
As 7n −6n−1 is divisible by 36 by the induction hypothesis, it follows that 7n+1 −6(n+1)−1
is divisible by 36 as well. Induction thus proves that the claim is true for all n ∈ N.
Bernoulli’s Inequality. Let x ≥ −1. We claim that
(1 + x)n ≥ 1 + nx
for all n ∈ N. We use induction to prove it.
n = 1: Clearly,
(1 + x)1 = 1 + x = 1 + 1 · x
holds.
n n + 1: We have
(1 + x)n+1 = (1 + x)n (1 + x)
| {z }
≥0

≥ (1 + nx)(1 + x), by the induction hypothesis,

nx2
= 1 + nx + x + |{z}
≥0

≥ 1 + (n + 1)x.

8
Induction thus proves Bernoulli’s Inequality.
For any set S, its power set is defined as

P(S) := {A : A ⊂ S},

i.e., it is the set of all subsets of S.

Examples. 1. If S = ∅, then P(S) = {∅};

2. If S = {1}, then P(S) = {∅, {1}};

3. If S = {1, 2}, then P(S) = {∅, {1}, {2}, {1, 2}};

4. If S = {1, 2, 3}, then P(S) = {∅, {1}, {2}, {3}, {1, 2}, {2, 3}, {1, 3}, {1, 2, 3}}.
For the following n ∈ N, the power set P({1, . . . , n}) has therefore the following
number of elements:

n = 1: 2 = 21 elements;

n = 2: 4 = 22 elements;

n = 3: 8 = 23 elements.

This suggests that the following is true:

Claim. For all n ∈ N, the power set P({1, . . . , n}) has 2n elements.

Proof. This induction anchor, i.e., the case n = 1 is clear.

Suppose that the claim is true for n ∈ N. Let A ⊂ {1, . . . , n, n + 1}. Then there are
two—mutually exclusive—possibilities:

• n+1∈
/ A;

• n + 1 ∈ A.

It follows that

P({1, . . . , n, n + 1}) = {A ⊂ {1, . . . , n, n + 1} : n + 1 ∈

/ A}
˙
∪{A ⊂ {1, . . . , n, n + 1} : n + 1 ∈ A}

Here, ∪˙ stands for a disjoint union, i.e., there is no A ⊂ {1, . . . , n, n + 1} that lies is both
subsets of P({1, . . . , n, n + 1}). It follows that

#P({1, . . . , n, n + 1}) = #{A ⊂ {1, . . . , n, n + 1} : n + 1 ∈

/ A}
+ #{A ⊂ {1, . . . , n, n + 1} : n + 1 ∈ A},

9
where # stands for the number of elements in the set concerned.
If A ⊂ {1, . . . , n, n + 1} is such that n + 1 ∈
/ A, then it is obvious that A ⊂ {1, . . . , n}.
It follows that

/ A} = #P({1, . . . , n}) = 2n
#{A ⊂ {1, . . . , n, n + 1} : n + 1 ∈

by the induction hypothesis.

For A ⊂ {1, . . . , n, n + 1} such that n + 1 ∈ A, define

à := A \ {n + 1} ⊂ {1, . . . , n},

and for B ⊂ {1, . . . , n}, set

B 0 := B ∪ {n + 1}.

It follows that  0
à = A and f0 = B.
B

Hence, the subsets of {1, . . . , n, n + 1} containing n + 1 are in a one-to-one correspondence

with the subsets of {1, . . . , n}; in particular, there are 2n of them. It follows that

#P({1, . . . , n, n + 1})
= #{A ⊂ {1, . . . , n, n + 1} : n + 1 ∈
/ A} + #{A ⊂ {1, . . . , n, n + 1} : n + 1 ∈ A}
= 2n + 2n
= 2n+1 ,

which proves the claim.

The Principle of Induction allows for a straightforward and very useful generalization:

Generalization of the Principle of Induction. Let n0 ∈ Z, and let P (n) be a claim

about n ∈ Z with n ≥ n0 . Suppose that:

(a) P (n0 ) is true;

(b) whenever P (n) is true, then P (n + 1) also true.

Then P (n) is true for all n ∈ Z such that n ≥ n0 .

Example. Consider n2 and n + 1. We obtain for the first four values of n:

n = 1: 12 = 1 < 2 = 1 + 1;

n = 2: 22 = 4 > 3 = 2 + 1;

n = 3: 32 = 9 > 4 = 3 + 1;

10
n = 4: 42 = 16 > 5 = 4 + 1.

This suggests that n2 > n + 1 for n ≥ 2. We prove this by induction.

The induction anchor with n = 2 is clear.
n n + 1: Note that

(n + 1)2 = n2 + 2n + 1
> n + 1 + 2n + 1, by the induction hypothesis,
= 3n + 2
> n + 2,

which proves the claim.

We denote the set of non-negative integers {0, 1, 2, 3, . . .} = N ∪ {0} by N0 .
Example. For n ∈ N0 , its factorial n! is defined as follows:
(
1, n = 0,
n! =
1 · 2 · · · · · (n − 1) · n, n ∈ N.

We observe that

n = 1: 1! = 1 = 12 ;

n = 2: 2! = 2 < 4 = 22 ;

n = 3: 3! = 6 < 9 = 32 ;

n = 4: 4! = 24 > 16 = 42 ;

n = 5: 5! = 120 > 25 = 52 ;

n = 6: 6! = 720 > 36 = 62 .

This suggests that

n! > n2

for n ≥ 4.
The induction anchor is clear.
n n + 1: Observe that

(n + 1)! = (n + 1)n!
> (n + 1)n2 , by the induction hypothesis,
> (n + 1)2 , by the previous example,

which proves the claim.

11
 n


Definition 1.1. Let n, k ∈ N0 . We define the binomial coefficient k —pronounced: n
choose k—as 0 if k > n and  
n n!
:=
k k!(n − k)!
if k ∈ {0, . . . , n}.

Remark. It is obvious that    

n n
=
k n−k
for n ∈ N0 and k ∈ {0, . . . , n}.

Lemma 1.2 (Pascal’s Triangle). Let n ∈ N, and let k ∈ {1, . . . , n}. Then
     
n n−1 n−1
= +
k k−1 k

holds.

Remark. The reason why Lemma 1.2 is referred to as “Pascal’s Triangle” becomes apparent
in the following sketch:

1 1

1 2 1

1 3 3 1

1 4 6 4 1

1 5 10 10 5 1

Figure 5: Pascal’s Triangle

The first row contains 1 = 00 , and the second row contains 1 = 10 and 1 = 11 . For

the third row, we put 1 = 20 = 22 at the endpoints and obtain the midpoint by adding

1 and 1 from the row above to get the midpoint 2 = 21 . We proceed in this fashion.

Proof. We treat the case where k = n separately. It is then clear that

     
n n−1 n−1
= + .
n n−1 n
| {z } | {z } | {z }
=1 =1 =0

12
 We can thus suppose without loss of generality that 1 ≤ k ≤ n − 1. It follows that
   
n−1 n−1 (n − 1)! (n − 1)!
+ = +
k−1 k (k − 1)!(n − k)! k!(n − 1 − k)!
k(n − 1)! + (n − k)(n − 1)!
=
k!(n − k)!
n!
= .
k!(n − k)!

This proves the lemma.

n


The next lemma reveals why k is called n choose k.

Lemma 1.3. Let n ∈ N, and let k ∈ N0 . Then the number of subsets of {1, . . . , n} with
exactly k elements is nk .

Proof. Let cnk be the number of subsets of {1, . . . , n} with exactly k elements. It is clear
that  
n n
ck = 0 =
k
if k > n and that  
n
cn0 =1= .
0
We can therefore limit ourselves to the case where 1 ≤ k ≤ n.
n = 1: Then
1!
c11 = 1 =
1! 0!
holds, which establishes the induction anchor.
n + 1: Let n ∈ N, and suppose that cnk = nk for all k = 1, . . . , n (and, conse-


n
quently, for k = 0, 1, . . . , n). We shall deduce that
 
n+1 n+1
ck =
k

for all k = 1, . . . , n, n + 1. As
   
n+1 n+1
=1= = 1 = cn+1
0 = cn+1
n+1 ,
n+1 0

we can limit ourselves to the case where k = 1, . . . , n. Let A ⊂ {1, . . . , n, n + 1} have k

elements. Then one and only one of the following cases can occur:

• n+1∈
/ A;

• n + 1 ∈ A.

13
It follows that
 
n
#{A ⊂ {1, . . . , n, n + 1} : #A = k, n + 1 ∈
/ A} = cnk =
k

and  
n
#{A ⊂ {1, . . . , n, n + 1} : #A = k, n + 1 ∈ A} = cnk−1 = ,
k−1
by the induction hypothesis, and therefore
     
n n n+1
cn+1
k = #{A ⊂ {1, . . . , n, n + 1} : #A = k} = cnk + cnk−1 = + =
k k−1 k

by Pascal’s Triangle.

Binomial Theorem. Let n ∈ N0 . Then

n  
X n
(x + y)n = xk y n−k
k
k=0

holds for all x, y ∈ R.

Examples. 1. Let n = 2. It follows that

(x + y)2 = y 2 + 2xy + x2 = x2 + 2xy + y 2 ,

i.e., the usual (first) binomial formula.

2. Let x = y = 1. It follows that

n   n
n n
X n X
2 = (1 + 1) = = cnk = #P({1, . . . , n}).
k
k=0 k=0

3. Let n ∈ N, and let x = −1 and y = 1. It follows that

n  
k n
X
n
0 = (−1 + 1) = (−1) .
k
k=0

Proof. We use induction. If n = 0, then both sides of the equality are 1: this establishes
the induction anchor.
n n + 1: Suppose that
n  
n
X n
(x + y) = xk y n−k
k
k=0

14
holds for all x, y ∈ R. It follows that

(x + y)n+1 = (x + y)n (x + y)
n  
!
X n k n−k
= x y (x + y), by the induction hypothesis,
k
k=0
n   n  
X n k+1 n−k X n k n+1−k
= x y + x y
k k
k=0 k=0
n−1
X n n  
k+1 n−k n+1 n+1
X n k n+1−k
= x y +x +y + x y
k k
k=0 k=1
n   n  
n+1
X n k n+1−k
X n k n+1−k
=y + x y + x y + xn+1
k−1 k
k=1 k=1
n    
X n n
= y n+1 + + xk y n+1−k + xn+1
k−1 k
k=1
n  
X n + 1 k n+1−k
= y n+1 + x y + xn+1 , by Pascal’s Triangle,
k
k=1
n+1
X n + 1
= xk y n+1−k .
k
k=0

This completes the proof.

Exercises
1. Show that
n
X n(n + 1)(2n + 1)
k2 =
6
k=1
for all n ∈ N.

2. Show that
n
X n(n + 1)
(−1)k−1 k 2 = (−1)n−1
2
k=1
for all n ∈ N.

3. Show that
n
X n2 (n + 1)2
k3 =
4
k=1
for all n ∈ N.

4. Let a, b ∈ R. Show that

n
X
n+1 n+1
a −b = (a − b) ak bn−k
k=0

15
 for all n ∈ N0 .

5. Show that
n
X
k(k!) = (n + 1)! − 1
k=1

for all n ∈ N.
P
6. We use the symbol for finite sums. There is a similar notation for finite products.
Let m, n ∈ N be such that m ≤ n, and let am , am+1 , . . . , an ∈ R. Then we define
n
Y
ak := am am+1 · · · an .
k=m

Show that
n  
Y 1 n+1
1− 2 =
k 2n
k=2

for all n ≥ 2.

7. Show that
n−1 k
Y 1 nn
1+ =
k n!
k=1

for all n ∈ N such that n ≥ 2.

8. For each n ∈ N, let P (n) be the statement:

n2 + 5n + 1 is even.

(a) Show that P (n + 1) is true whenever P (n) is true.

(b) For which n ∈ N is P (n) actually true?

What’s the moral of this problem?

9. Guess for which n ∈ N the inequality 2n > n2 is true, and prove your guess using
induction.

16
2 The Set Q of Rational Numbers
We denote the set {0, 1, −1, 2, −2, . . .} of integers by Z (presumably because the German
word for “number” is “Zahl”), and the set of rational numbers
nm o
: m, n ∈ Z, n 6= 0
n
by Q (presumably because the word “quotient” starts with the letter “q”).
The following is (or at least ought to be) well known:
√
Example. 2 is not a rational number.

Definition 2.1. A real number x is called algebraic if there are cn , . . . , c1 , c0 ∈ Z with

cn 6= 0 such that
cn xn + cn−1 xn−1 + · · · + c1 x + c0 = 0.
m
Examples. 1. Let q ∈ Q, i.e., there are m, n ∈ Z with n 6= 0 such that q = n. It follows
that q solves the equation
nx − m = 0

and therefore is algebraic.

√
2. 2 solves the equation
x2 − 2 = 0,

therefore is algebraic, but fails to be rational.

√
The following generalizes the statement that 2 is not rational:
c
Rational Zeroes Theorem. Let cn , . . . , c1 , c0 ∈ Z with n ≥ 1, cn 6= 0, and let d with
c, d ∈ Z and d 6= 0 having no prime factor in common solve the equation

cn xn + cn−1 xn−1 + · · · + c1 x + c0 = 0.

Then c divides c0 and d divides cn .

Proof. We have
 c n  c n−1 c
cn + cn−1 + · · · + c1 + c0 = 0.
d d d
Multiplying by dn yields

(1) cn cn + cn−1 cn−1 d + · · · + c1 cdn−1 + c0 dn = 0.

Solving for c0 dn , we obtain

c0 dn = −c cn cn−1 + cn−1 cn−2 d + · · · + c1 dn−1 .

17
This means that c divides c0 dn . As c and d have no prime factor in common, this is
possible only if c divides c0 .
Solving (1) for cn cn yields

cn cn = −d cn−1 cn−1 + · · · + c1 dn−2 + c0 dn−1 ,

so that d divides cn cn . Again, due to the fact that c and d have no prime factor in
common, this means that d divides cn .

Corollary 2.2. Consider the equation

xn + cn−1 xn−1 + · · · + c1 x + c0 = 0

with cn−1 , . . . , c1 , c0 ∈ Z such that c0 6= 0. Then any rational solution to this equation is
an integer dividing c0 .
Proof. Let c, d ∈ Z with d 6= 0 and having no prime factor in common be such that dc
solves the equation in question. Then, by the Rational Zeroes Theorem, d divides the
coefficient of xn , i.e., 1. It follows that d = ±1, so that dc = ±c. Again by the Rational
Zeroes Theorem, this means that c divides c0 .
√ √
Examples. 1. Let m ∈ N be such that m ∈ / Z. Assume that m is rational. Obvi-
√
ously, m solves the equation
x2 − m = 0.
√
By Corollary 2.2, this means that m must be an integer, which is a contradiction.
(It follows that the square root of any natural number that is not a perfect square
has to be irrational.)

2. Set
√
q
3
x = 2 + 5.
√
It follows that x2 = 2 + 3
5 and thus
√
x2 − 2 =
3
5.

Cubing this yields

5 = (x2 − 2)3 = x6 − 6x4 + 12x2 − 8,
i.e.,
x6 − 6x4 + 12x2 − 13 = 0.
If x were rational, then Corollary 2.2 would imply that x ∈ {±1, ±13}, which is
obviously nonsense. Therefore, x must be irrational.

Exercises
p
3
√
1. Is 5 − 3 a rational number?

18
3 The Set R of Real Numbers
For x, y ∈ R, the sum x + y and the product xy are defined in R; if x, y ∈ Q, then
x + y, xy ∈ Q as well. The following field axioms hold for x, y, z ∈ R:

(F 1) x + (y + z) = (x + y) + z (associativity of addition);

(F 2) x + y = y + x (commutativity of addition);

(F 3) x + 0 = x;

(F 4) there is a unique −x such that x + (−x) = 0 (existence and uniqueness of the

additive inverse);

(F 5) x(yz) = (xy)z (associativity of multiplication);

(F 6) xy = yx (commutativity of multiplication);

(F 7) x1 = x;

(F 8) if x 6= 0, the there is a unique x−1 such that xx−1 = 1 (existence and uniqueness of
the multiplicative inverse);

(F 9) x(y + z) = xy + xz (distributivity).

These axioms are satisfied by R and Q, but much more “exotic” examples are possible:
Example. Let Z2 := {0, 1} be equipped with an addition + and a multiplication · by the
following tables:

+ 0 1 · 0 1
0 0 1 and 0 0 0
1 1 0 1 0 1

It is routinely, albeit tediously verified that Z2 equipped with these operations satisfies (F
1) to (F 9).
Any set satisfying (F 1) to (F 9) is called a field.

Consequences. For x, y, z ∈ R, the following hold:

(i) x + z = y + z implies x = y;

(ii) x0 = 0;

(iii) (−x)y = −xy;

(iv) (−x)(−y) = xy;

19
 (v) if xz = yz with z 6= 0, then x = y;

(vi) if xy = 0, then x = 0 or y = 0.

Proof. (i) Suppose that x + z = y + z. This implies that

(x + z) + (−z) = (y + z) + (−z)

and so, by (F 1),

x + (z + (−z)) = y + (z + (−z)),

which yields x = y by (F 4) and (F 3).

(ii) By (F 3) and (F 9), we have

x0 = x(0 + 0) = x0 + x0,

so that

0 = x0 + (−x0) = (x0 + x0) + (−x0) = x0 + (x0 + (−x0)) = x0.

(iii) As x + (−x) = 0, (ii) yields that

0 = (x + (−x))y = xy + (−x)y,

so that (−x)y = −xy.

(iv) Note that, by (iii) and (ii),

(−x)(−y) + (−xy) = (−x)(−y) + (−x)y = (−x)((−y) + y) = 0.

(v) This is proven in the same way as (i).

(vi) Let xy = 0, and suppose that x 6= 0. It follows that

y = (x−1 x)y = x−1 (xy) = 0.

This completes the proof.

Remark. All these claims about R only require the field axioms (F 1) to (F 9). It is clear
that any field—such as Z2 with the addition and multiplication provided—also satisfies
the claims above.
The set R also has an order structure, i.e., it satisfies the following order axioms. For
x, y, z ∈ R, the following hold:

(O 1) we have x ≤ y or y ≤ x (note that this in an inclusive “or”, i.e., both can hold
simultaneously);

20
(O 2) if x ≤ y and y ≤ x, the x = y;

(O 3) if x ≤ y and y ≤ z, then x ≤ z;

(O 4) if x ≤ y, then x + z ≤ y + z;

(O 5) if x ≤ y and z ≥ 0, then xz ≤ yz.

Consequences. For x, y, z ∈ R, the following hold:

(i) if x ≤ y, then −y ≤ −x;

(ii) if x ≤ y and z ≤ 0, then xz ≥ yz;

(iii) if 0 ≤ x, y, then 0 ≤ xy;

(iv) 0 ≤ x2 ;

(v) 0 < 1;

(vi) if 0 < x, then 0 < x−1 ;

(vii) if 0 < x < y, then 0 < y −1 < x−1 .

Proof. (i) Set z := (−x) + (−y). It follows from (O 4) that

−y = x + z ≤ y + z = −x.

(ii) From (i), it follows that 0 ≤ −z, so that

−xz = x(−z) ≤ y(−z) = −yz

and therefore xz ≥ yz by (i) again.

(iii) Set x = 0 in (O 5). It follows that 0 ≤ yz. Relabeling the variables yields the claim.

(iv) If x ≥ 0, then x2 ≥ 0 holds. If x ≤ 0, then 0 ≤ −x, so that x2 = (−x)2 ≥ 0.

(v) It is clear that 1 = 12 ≥ 0. As x1 = x 6= 0 for all x ∈ R \ {0}, we have 1 6= 0 and

thus 1 > 0.

(vi) Assume that x > 0, but that 0 6< x−1 , i.e., x−1 ≤ 0 and therefore 0 ≤ −x−1 . It
follows that
0 ≤ x(−x−1 ) = −1,

which contradicts the fact that 1 > 0.

(vii) Similar.

21
Definition 3.1. For x ∈ R, define its absolute value or modulus |x| as

|x| := x

if x ≥ 0 and
|x| := −x

if x ≤ 0.

Remark. It is clear that |x| is well defined, i.e., if the two cases overlap, which means that
x ≥ 0 and x ≤ 0 and therefore x = 0, we have |x| = 0 in any case.

Theorem 3.2. The following are true for x, y ∈ R:

(i) |x| ≥ 0;

(ii) |xy| = |x||y|;

(iii) |x + y| ≤ |x| + |y| (triangle inequality).

Proof. (i) Clear.

(ii) We have four cases:

(a) x, y ≥ 0: then xy ≥ 0 as well, so that |x| = x, |y| = y, and |xy| = xy;

(b) x, y ≤ 0: them |x| = −x, |y| = −y, so that |x||y| = (−x)(−y) = xy = |xy|;
(c) x ≤ 0 and y ≥ 0: |x||y| = (−x)y = −xy = |xy|;
(d) x ≥ 0 and y ≤ 0: similar.

(iii) It is clear that

−|x| ≤ x ≤ |x| and − |y| ≤ y ≤ |y|.

Adding those inequalities yields

−(|x| + |y|) ≤ x + y ≤ |x| + |y|.

This means that

x + y ≤ |x| + |y| and − (x + y) ≤ |x| + |y|,

which yields the claim.

Corollary 3.3 (“Phony” Triangle Inequality). Let x, y ∈ R. Then

||x| − |y|| ≤ |x − y|

holds.

22
Proof. Note that
|x| = |y + (x − y)| ≤ |y| + |x − y|.

Interchanging the rôles of x and y yields

|y| ≤ |x| + |x − y|,

so that
±(|x| − |y|) ≤ |x − y|

and therefore
||x| − |y|| ≤ |x − y|.

This proves the claim.

Exercises
1. Let h√ i n √ o
2 := p + q 2 : p, q ∈ Q .
Q

Show that, if x ∈ Q 2 \ {0}, then x−1 ∈ Q 2 as well.

√  √ 

2. For n ∈ N, let x1 , . . . , xn ∈ R. Show that

n
X n
X
xk ≤ |xk |.
k=1 k=1

23
4 The Difference between R and Q
Definition 4.1. Let ∅ 6= S ⊂ R. Then:

(a) we call x0 ∈ R a maximum for S if x0 ∈ S and x ≤ x0 for all x ∈ S;

(b) we call x0 ∈ R a minimum for S if x0 ∈ S and x ≥ x0 for all x ∈ S.

We use the notation max S for the maximum and min S for the minimum of S.
Examples. 1. All non-empty finite subsets of R have a minimum and a maximum.

2. Q and Z have no maximum and no minimum.

3. N has no maximum, but min N = 1.

4. The set
S := q ∈ Q : q ≥ 0, q 2 ≤ 2

√
has 0 as its minimum, but no maximum because 2 ∈ / Q.

Definition 4.2. Let ∅ 6= S ⊂ R. Then:

(a) if M ∈ R satisfies x ≤ M for all x ∈ S, we call S bounded above by M and M an

upper bound for S;

(b) if m ∈ R satisfies x ≥ m for all x ∈ S, we call S bounded below by m and m a lower

bound for S;

(c) if S bounded both above and below, we call S bounded.

Examples. 1. Suppose that M := max S exists. Then M is an upper bound for S.

2. Suppose that m := min S exists. Then m is a lower bound for S.

3. Neither of Q, Z, or N is bounded above, and Q and Z aren’t bounded below either.

However, N is bounded below by its minimum 1.

4. Let a, b ∈ R be such that a < b, then we define:

(a, b) := {x ∈ R : a < x < b} (open interval from a to b),

[a, b] := {x ∈ R : a ≤ x ≤ b} (closed interval from a to b),
(a, b] := {x ∈ R : a < x ≤ b} (left open interval from a to b),

and

[a, b) := {x ∈ R : a ≤ x < b} (right open interval from a to b).

24
 Then all these sets are bounded below by a and above by b. However, a is a
minimum for [a, b) and [a, b], but not for (a, b) and (a, b], and b is a maximum for
(a, b] and [a, b], but not for (a, b) and [a, b).

5. If S is bounded with lower bound m and upper bound M , then

m≤x≤M

and, consequently,
−M ≤ −x ≤ −m

for all x ∈ S. Setting C := max{M, −m}, this means that

−C ≤ ±x ≤ C

and therefore
|x| ≤ C

for x ∈ S.

Definition 4.3. Let ∅ 6= S ⊂ R. Then:

(a) if S has a least upper bound, i.e., there is an upper bound M for S such that, whenever
M̃ < M , there is x ∈ S with x > M̃ , we call M the supremum of S, denoted by sup S;

(b) if S has a largest lower bound, we call it the infimum of S, denoted by inf S.

Examples. 1. If sup S and inf S exist, they are unique.

2. If max S exists, then sup S = max S; similarly, if min S exists, then inf S = min S.

3. Let a, b ∈ R be such that a < b. Then we have

sup(a, b) = sup(a, b] = sup[a, b) = sup[a, b] = b

and
inf(a, b) = inf(a, b] = inf[a, b) = inf[a, b] = a.

4. Consider
S = q ∈ Q : 0 ≤ q, q 2 ≤ 2 .


Then we have
√
inf S = min S = 0 and sup S = 2 ∈ R \ Q.

The last example shows that the following is false for Q instead of R:

25
Completeness Axiom. Let ∅ 6= S ⊂ R be bounded above. Then S has a supremum.

In fact, together with the field axioms and the order axioms the Completeness Axiom
characterizes R.

Corollary 4.4. Let ∅ 6= S ⊂ R be bounded below. Then S has an infimum.

Proof. Set
−S := {−x : x ∈ S}.

Let m be a lower bound for S, i.e., m ≤ x for all x ∈ S. It follows that −x ≤ −m for all
x ∈ S, so that −m is an upper bound for −S. From the Completeness Axiom, it follows
that sup(−S) exists. It is clear that −x ≤ sup(−S) and therefore x ≥ − sup(−S) for all
x ∈ S; hence, − sup(−S) is a lower bound for S. Assume that m̃ > − sup(−S) is a lower
bound for S. Then it follows that −m̃ < sup(−S) and thus cannot be an upper bound for
−S. This means that there is x ∈ S with −m̃ < −x, so that m̃ > x, which is impossible
if m̃ is a lower bound for S.

The following is crucial:

Archimedian Property. Let a, b > 0 be reals numbers. Then there is n ∈ N such that
na > b.

Proof. Assume that the claim is false, i.e., na ≤ b for all n ∈ N. This means that the set

S := {na : n ∈ N}

is bounded above by b and therefore has a supremum. As a > 0, we have sup S −a < sup S,
so that sup S − a cannot be an upper bound for S. It follows that there is n0 ∈ N such
that n0 a > sup S − a. This, however, entails that

(n0 + 1)a > sup S,

| {z }
∈S

which is a contradiction.

Corollary 4.5. Let a, b > 0 be real numbers. Then there are n, m ∈ N such that
1
b<n and < a.
m
Proof. Applying the Archimedian Property with a = 1 yields n ∈ N with n > b. Applying
the Archimedian Property again, with b = 1 this time, shows that there is m ∈ N such
1
that ma > 1, i.e., m < a.

Corollary 4.6 (Density of Q). Let a, b ∈ R be such that a < b. Then there is q ∈ Q such
that a < q < b.

26
Proof. We need to show that there are m ∈ Z and n ∈ N such that
m
a< < b.
n
1
As b − a > 0, there is n ∈ N such that b − a > n by the previous corollary. The subset

S := {ν ∈ Z : ν > na}

of N is non-empty by the Archimedian Property and therefore has a minimum m. It

follows that
m > na and m − 1 ≤ na,

so that
na < m ≤ na + 1 < na + n(b − a) = nb

and therefore
m
a< < b.
n
This completes the proof.

Corollary 4.7 (Density of R\Q). Let a, b ∈ R be such that a < b. Then there is r ∈ R\Q
such that a < r < b.

Proof. Choose q ∈ Q such √that a < q < b. As b − q > 0, there is n ∈ N such that
√
n(b − q) > 2, i.e., b − q > n2 and therefore
√
2
a<q+ < b,
n
| {z }
∈Q
/

which completes the proof.

The Symbols +∞ and −∞

We add the symbols +∞ and −∞ to R as objects that do not belong to R and equip
R ∪ {+∞, −∞} with and order structure with

−∞ < x < +∞

for all x ∈ R. For notational convenience, we will mostly write ∞ instead of +∞. We
also set
x+∞=∞ and x − ∞ := x + (−∞) := −∞

for all x ∈ R as well as

∞ + ∞ := ∞ and − ∞ − ∞ := −∞ + (−∞) := −∞.

27
Warning. Expressions like

∞−∞ or −∞+∞

are undefined and—even worse—cannot be defined in any meaningful way. They must be
avoided at all cost.
We use the following notation for a ∈ R:

(a, ∞) := {x ∈ R : x > a},

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

and

(−∞, ∞) := R.

Moreover, if ∅ 6= S ⊂ R is not bounded above, we set sup S := ∞, and if S is not bounded

below, we set inf S := −∞.
Example. Let S, T ⊂ R be non-empty, and define

S + T := {x + y : x ∈ S, y ∈ T }.

We claim that
sup(S + T ) = sup S + sup T.

Suppose first that S is not bounded above. Let M ∈ R, and let y0 ∈ T . As S is not
bounded above, there is x ∈ S such that x > M − y0 , so that x + y0 > M . Since
x + y0 ∈ S + T , this means that S + T is not bounded above. Similarly, if T is not
bounded above, then S + T is not bounded above either. It follows that, if S or T fails to
be bounded above, then

sup(S + T ) = ∞ = sup S + sup T.

We can therefore suppose without loss of generality that both S and T are bounded above.
Let x ∈ S and y ∈ T . Then
x + y ≤ sup S + sup T,

so that sup S + sup T is an upper bound for S + T , which entails

sup(S + T ) ≤ sup S + sup T.

28
Assume towards a contradiction that sup(S + T ) < sup S + sup T , i.e.,

 := sup S + sup T − sup(S + T ) > 0.

As sup S − 2 < sup S, there is x0 ∈ S such that x0 > sup S − 2 . Similarly, there is y0 ∈ T
such that y0 > sup T − 2 . It follows that
 
x0 + y0 > sup S − + sup S −
| {z } 2 2
∈S+T

= sup S + sup T − 
= sup S + sup T − sup S − sup T + sup(S + T )
= sup(S + T ),

which is a contradiction.

Exercises
1. Let ∅ 6= S ⊂ R be bounded such that inf S = sup S. What can you say about S?

2. Let a, b ∈ R be such that a < b. Show that there are infinitely many irrational
numbers r ∈ R \ Q such that a < r < b.

3. Let ∅ 6= S, T ⊂ R be bounded above. Show that

sup(S ∪ T ) = max{sup S, sup T }.

4. Determine inf S and sup S for

   
n 1
S := (−1) 1 − :n∈N .
n

5. Determine inf S and sup S for

 
1
S := : p is a prime number
p

(Hint: You can—and should —use the fact that there are infinitely many prime
numbers, so that, for each n ∈ N, there is a prime number p such that p ≥ n.)

6. Determine inf S and sup S for

 
n 1
S := n((−1) + 1) + : n ∈ N .
n

29
5 Functions
Definition 5.1. Let S and T be non-empty sets. Then a function f or map from S to
T —in symbols: f : S → T —is a rule that assigns to each s ∈ S a unique element f (s) ∈ T .

We use the notation

f : S → T, s 7→ f (s).

Examples. 1. Let
p(x) := cn xn + · · · + c1 x + c0

with cn , . . . , c1 , c0 ∈ R be polynomial. Then p defines a function

p : R → R, x 7→ p(x).

2. Let p and q 6= 0 be polynomials, and let Z := {x ∈ R : q(x) = 0}. Then

p(x)
r : R \ Z → R, x 7→
q(x)

defines a function from R \ Z into R.

3. sin, cos, and exp define functions from R to R.

4. The natural logarithm log defines a function

log : (0, ∞) → R, x 7→ log x

with log(exp x) = x for x ∈ R and exp(log x) = x for x ∈ (0, ∞).

5. Define f : R → R as follows:

 0,
 x∈/ Q,
1 p
f (x) := q, x = q 6= 0 with p ∈ Z and q ∈ N coprime,

1, x = 0.


Terminology. Let S and T be non-empty sets, and let f : S → T be a function. Then

we call

(a) S the domain of f ,

(b) T the codomain of f , and

(c) f (S) := {f (s) : s ∈ S} the range of f .

Definition 5.2. Let S and T be non-empty sets, and let f : S → T be a function. Then
we call f :

30
(a) injective if, for s1 , s2 ∈ S with s1 6= s2 , we have f (s1 ) 6= f (s2 );

(b) surjective if f (S) = T ;

(c) bijective if is both injective and surjective.

Remark. For any function f : S → T , the function

f : S → f (S), s 7→ f (s)

is surjective.
Examples. 1. Consider
f : R → R, x 7→ x2 .

Then f is neither injective nor surjective because f (−1) = 1 = f (1) and −1 ∈

/ f (R).

2. Consider
f : R → [0, ∞), x 7→ x2 .

The f is not injective, but surjective.

3. Consider
f : [0, ∞) → R, x 7→ x2 .

Then f is not surjective, but injective.

4. Finally,
f : [0, ∞) → [0, ∞), x 7→ x2

is bijective.

Definition 5.3. Let S and T be non-empty sets, and let f : S → T be a function. Then:

(a) for A ⊂ S, define

f (A) := {f (s) : s ∈ A},

the image of A under f ;

(b) for B ⊂ T , define

f −1 (B) := {s ∈ S : f (s) ∈ B},

the inverse image of B under f .

Observation. Suppose that f : S → T is injective, let t ∈ T , and set B := {t}. Then

f −1 (B) = {s ∈ S : f (s) = t}

is either empty—if t ∈
/ f (S)—or consists of one single element of S.

31
Definition 5.4. Let S and T be non-empty sets, and let f : S → T be injective. Then

f −1 : f (S) → S, f (s) 7→ s

is called the inverse function of f .

Examples. 1. If
f : [0, ∞) → [0, ∞), x 7→ x2 ,

then
√
f −1 : [0, ∞) → [0, ∞), y 7→ y.

2. If
f : R → (0, ∞), x 7→ ex ,

then
f −1 : (0, ∞) → R, y 7→ log y.

3. If  π π sin x
f: − , → R, x 7→ = tan x,
2 2 cos x
then  π π
f −1 : R → − , , y 7→ arctan y.
2 2

Exercises
1. Let S be a non-empty set. Show that there is no surjective map from S to P(S).
(Hint: Assume that there is a surjective map f : S → P(S) and consider the set
{x ∈ S : x ∈
/ f (x)}.)

2. For n ∈ N, let

Sn := {σ : {1, . . . , n} → {1, . . . , n} : σ is bijective}.

Use induction to show that #Sn = n!.

32
6 Convergence of Sequences
Definition 6.1. Let n0 ∈ Z and let T be a non empty set. A sequence in T is a function
s : {n ∈ Z : n ≥ n0 } → T .

Notation. We write:

• (sn )∞
n=n0 instead of s : {n ∈ Z : n ≥ n0 } → T ;

• sn instead of s(n) for n ∈ Z with n ≥ n0 .

We will mostly be concerned with sequences in R. In what follows, we will deal with
the case where n0 = 1 only, but it is clear how the statements extend to more general n0 .

Definition 6.2. Let (xn )∞ ∞

n=1 be a sequence in R. Then we say that (xn )n=1 converges to
x ∈ R or is convergent to x if, for each  > 0, there is n ∈ N such that |xn − x| <  for
all n ∈ N with n ≥ n . We call x the limit of (xn )∞
n=1 .

Notation. If x is the limit of (xn )∞

n=1 , we write
n→∞
x = lim xn or xn −→ x or xn → x.
n→∞

x3
x+ε
x4 x6 x7 x10
x x8
x2 x9 x 11
x5
x− ε
x1

1 2 3 4 5 6 7 8 9 10 11

Figure 6: limn→∞ xn = x

Examples. 1. Let c ∈ R, and let (xn )∞

n=1 be defined as xn := c for n ∈ N. Let  > 0,
and set n := 2021. Then
|xn − c| = 0 < 
holds for all n ≥ n , so that c = limn→∞ xn .

∞
2. Consider the sequence n1 n=1 . Let  > 0. By the Archimedian Property of R, there
is n ∈ N such that n1 < . Consequently, we have for n ≥ n that

1 1 1
= ≤ < .
n n n

33
 1
It follows that limn→∞ n = 0.

3. Let k ∈ N, and set xn := n1k . Let  > 0. By the previous example, there is n such
that n1 <  for all n ≥ n . It follows that

1 1 1
k
= k ≤ <
n n n
1
for n ≥ n , so that nk
→ 0.

4. Let xn := (−1)n for n ∈ N. What is the limit of (xn )∞

n=1 ?

Claim. (xn )∞
n=1 does not converge.

To see this, assume towards a contradiction that (xn )∞ n=1 has a limit x. Then there
is n1 ∈ N such that |xn − x| < 1 for n ≥ n1 . It follows that

2 = |xn − xn+1 | ≤ |xn − x| + |x − xn+1 | < 1 + 1 = 2

for n ≥ n1 , which is impossible.

Proposition 6.3. Let (xn )∞ ∞

n=1 be a convergent sequence in R. Then the limit of (xn )n=1
is unique.

Proof. Assume that there are x, x̃ ∈ R with x 6= x̃ that are both limits of (xn )∞
n=1 . Set

 := |x − x̃| > 0. As xn → x, there is n1 ∈ N such that |xn − x| < 2 for n ≥ n1 . As also
xn → x̃, there is n2 ∈ N such that |xn − x̃| < 2 for n ≥ n2 . For n ≥ max{n1 , n2 }, we
therefore have
 
 = |x − x̃| = |x − xn + xn − x̃| ≤ |x − xn | + |xn − x̃| < + = ,
2 2
which is a contradiction.

Theorem 6.4. Let (xn )∞ ∞

n=1 and (yn )n=1 be convergent sequences in R with limits x and
y, respectively. Then (xn + yn )∞
n=1 is convergent with

lim (xn + yn ) = x + y.
n→∞

Proof. Let  > 0. As xn → x, there is n1 ∈ N such that |xn − x| < 2 for n ≥ n1 , and as
yn → y, there is n2 ∈ N such that |yn − y| < 2 for n ≥ n2 . Set n := max{n1 , n2 }. Then
it follows for n ≥ n that
 
|(xn + yn ) − (x + y)| = |(xn − x) + (yn − y)| ≤ |xn − x| + |yn − y| < + = .
2 2
This means that xn + yn → x + y.

34
 We convene to call a sequence (xn )∞ n=1 bounded above, bounded below, or bounded if
the corresponding statement is true for the set {xn : n ∈ N}.
Example. Let θ > 1, and set xn := θn for n ∈ N. From Bernoulli’s Inequality, we obtain
that
xn = (1 + (θ − 1))n ≥ 1 + n(θ − 1)
| {z }
>0
for n ∈ N. By the Archimedian Property of R, this means that (xn )∞
n=1 cannot be bounded.

Proposition 6.5. Let (xn )∞ ∞

n=1 be a convergent sequence. Then (xn )n=1 is bounded.

Proof. Let x := limn→∞ xn . Then there is n1 ∈ N such that |xn − x| < 1 for n ≥ n1 and
therefore
|xn | = |xn − x + x| ≤ |xn − x| + |x| < 1 + |x|
for those n. Set
C := max{|x1 |, . . . , |xn1 −1 |, 1 + |x|},
so that |xn | ≤ C for n ∈ N. This proves the claim.

Theorem 6.6. Let (xn )∞ ∞

n=1 and (yn )n=1 be convergent sequences in R with limits x and
y, respectively. Then (xn yn )∞
n=1 is convergent with

lim xn yn = xy.
n→∞
Proof. As (yn )∞
n=1 is convergent, it is bounded, i.e., there is C ≥ 0 such that |yn | ≤ C for
all n ∈ N.
Let  > 0. Choose n1 ∈ N such that

|xn − x| <
2(C + 1)
for n ≥ n1 and n2 ∈ N such that

|yn − y| <
2(|x| + 1
for n ≥ n2 . Set n := max{n1 , n2 }. For n ≥ n , we then have
|xn yn − xy| = |xn yn − xyn + xyn − xy|
≤ |xn yn − xyn | + |xyn − xy|
= |xn − x||yn | + |x||yn − y|
≤ C|xn − x| + |x||yn − y|
C |x|
< +
2(C + 1) 2(|x| + 1)
 
≤ +
2 2
= .

35
Example.   
7 3 13
lim 2 + − 2 2 + 42 = 4.
n→∞ n n n
Theorem 6.7. Let (xn )∞ n=1 be a convergent sequence in R with
 limit x 6= 0. Then there
∞
1
is n0 ∈ N such that xn 6= 0 for all n ≥ n0 , and the sequence xn converges to x1 .
n=n0

|x|
Proof. Let n0 ∈ N be such that |xn − x| < 2 for n ≥ n0 . It follows that

|x|
|xn | = |(xn − x) + x| ≥ ||xn − x| − |x|| > >0
2
2
and therefore xn 6= 0 for n ≥ n0 . This also shows that |x1n | < |x| and therefore n ≥ n0 .
 ∞  ∞
Hence, the sequence x1n is bounded, as is the sequence xn1 x . Let C ≥ 0 be
n=n0 n=n0
such that |xn1 x| ≤ C for n ≥ n0 .

Let  > 0. Choose n ≥ n0 such that |xn − x| < C+1 for n ≥ n . For those n, we
obtain
1 1 |xn − x| C
− = ≤ C|xn − x| < < .
xn x |xn x| C +1
This proves the claim.

Corollary 6.8. Let (xn )∞ ∞

n=1 and (yn )n=1 be convergent sequences in R with limits x and
y, respectively,
 ∞ such that y 6= 0. Then there is n0 ∈ N such that yn 6= 0 for all n ≥ n0 ,
xn
and yn converges such that
n=n0

xn x
lim = .
n→∞ yn y
Example.
4n3 − 3n2 + 17n − 666 4 − n3 + n172 − 666
n3 4
lim = lim = .
n→∞ 3n3 + n2 − n + 2021 n→∞ 3 + 1 − 12 + 2021 3
n n n3
Together, Theorems 6.4 and 6.6 and Corollary 6.8 are often refereed to as “the Limit
Laws”.

Divergence of Sequences
Definition 6.9. Let (xn )∞ ∞
n=1 be a sequence in R. We say that (xn )n=1 :

(a) diverges or is divergent if it does not converge;

(b) diverges to ∞ if, for each R ∈ R, there is nR ∈ N such that xn > R for n ≥ nR ;

(c) diverges to −∞ if, for each R ∈ R, there is nR ∈ N such that xn < R for n ≥ nR .

36
Notation. If (xn )∞
n=1 diverges to ∞, we write

n→∞
lim xn = ∞ or xn −→ ∞ or xn → ∞.
n→∞

Similar conventions apply to −∞.

Examples. 1. The sequence ((−1)n )∞
n=1 is divergent.

2. Every sequence diverging to ∞ or to −∞ is unbounded. However, (n(−1)n )∞

n=1 is
unbounded, but diverges neither to ∞ nor to −∞.
The following is clear:

Proposition 6.10. Let (xn )∞ ∞

n=1 be a sequence in R. Then (xn )n=1 diverges to ∞ if and
only if (−xn )∞
n=1 diverges to −∞.

Theorem 6.11. Let (xn )∞ ∞

n=1 and (yn )n=1 be sequences in R. Then:

(i) if limn→∞ xn = ∞ and if y = limn→∞ yn exists in R ∪ {∞}, then

lim (xn + yn ) = ∞;
n→∞

(ii) if limn→∞ xn = −∞ and if y = limn→∞ yn exists in R ∪ {−∞}, then

lim (xn + yn ) = −∞.

n→∞

Proof. We only prove (i). ((ii) can be proven analogously or be deduced from (i) using
Proposition 6.10.)
Case 1: y ∈ R.
In this case (yn )∞
n=1 is bounded, i.e., there is C ≥ 0 such that |yn | ≤ C for n ∈ N. Let
R ∈ R. As xn → ∞, there is nR ∈ N such that xn > R + C for all n ≥ nR . It follows for
those n that
xn + yn ≥ xn − C > R + C − C = R.

Consequently, limn→∞ (xn + yn ) = ∞ holds.

Case 2: y = ∞.
Let R ∈ R. Choose n1 , n2 ∈ N such that xn > R2 for n ≥ n1 and yn > R
2 for n ≥ n2 .
Set nR := max{n1 , n2 }. For n ≥ nR , it then follows that

R R
x n + yn > + = R.
2 2
This implies limn→∞ (xn + yn ) = ∞.

What happens with (xn + yn )∞

n=1 if xn → ∞ and yn → −∞? In this case, nothing can
be said.

37
Examples. 1. Let c ∈ R, and define

xn := n + c and yn := −n

for n ∈ N. Then we have xn → ∞, yn → −∞, and xn + yn = c → c.

2. Define
√
xn := n and yn := − n

for n ∈ N, so that xn → ∞ and yn → −∞. Let R ∈ R. We can suppose without

√ √ √
loss of generality that R ≥ 0. As n → ∞, there is nR ∈ N such that n > R + 1
for n ≥ nR , so that
√ √ √
n> n−1> R

for those n. We therefore obtain

√ √ √
√ √
x n + yn = n − n= n n−1 > R R=R

for n ≥ nR . It follows that xn + yn → ∞.

3. Let ck , . . . , c1 , c0 ∈ R with ck 6= 0, and let

p(x) := ck xk + · · · + c1 x + c0 .

We then have 
 c0 ,
 if k = 0,
lim p(n) = ∞, if k ≥ 1 and ck > 0,
n→∞ 
−∞, if k ≥ 1 and ck < 0.


Of course, if k = 0, the claim is clear.

Suppose that k ≥ 1 and that ck > 0. Set
ck−1 c1 c0
xn := ck + + · · · + k−1 + k .
n n n
The limit laws imply that xn → ck ; in particular, there is n1 ∈ N such that |xn −ck | <
ck
2 for all n ≥ n1 . It follows that
ck ck
xn = xn − ck + ck ≥ ck − |xn − ck | ≥ ck − =
2 2
2R
for those n. Let R > 0. As nk → ∞, there is n2 ∈ N such that nk > ck for all
n ≥ n2 . Set nR := max{n1 , n2 }. Then we obtain for all n ≥ nR that
2R ck
p(n) = nk xn > = R.
ck 2
It follows that p(n) → ∞.
The case with k ≥ 1 and ck < 0 is treated analogously.

38
 n
4. In the homework, it was shown that n2 ≤ 2n and therefore n ≤ 2n for all n ≥ 5. It
n n
follows that 2n → ∞. Let R ≥ 0. Then there is nR ∈ N such that 2n > R + 1 for
n
all n ≥ nR and therefore 2n − 1 > R for n ≥ nR . For those n, we obtain
 n 
2
2n − n = n − 1 > nR ≥ R.
n

All in all, 2n − n → ∞ holds.

Proposition 6.12. Let (xn )∞

n=1 be a sequence in (0, ∞). Then the following are equiva-
lent:

(i) limn→∞ xn = ∞;
1
(ii) limn→∞ xn = 0.
1
Proof. (i) =⇒ (ii): Let  > 0. Then there is n ∈ N such that xn >  for all n ≥ n , so
that
1 1
= <
xn xn
for those n. It follows that limn→∞ x1n = 0.
(ii) =⇒ (i): Let R ∈ R, and suppose without loss of generality that R > 0. As x1n → 0,
there is nR ∈ N such that x1n < R1 for all n ≥ nR , so that xn > R for all n ≥ nR . This
means that xn → ∞.

Example. Let θ ∈ (0, ∞) \ {1}.

If θ > 1, Bernoulli’s Inequality yields that

θn = (1 + (θ − 1))n ≥ 1 + n(θ − 1)

for n ∈ N. By the Archimedian Property of R, we have limn→∞ n(θ − 1) = ∞ and,

consequently, limn→∞ 1 + n(θ − 1) = ∞. It follows that limn→∞ θn = ∞.
If θ ∈ (0, 1), then 1θ > 1. It follows that
 n
1 1
lim = lim n = ∞,
n→∞ θ n→∞ θ

so that limn→∞ θn = 0.

Monotonic Sequences
Definition 6.13. Let (xn )∞ ∞
n=1 be a sequence in R. Then we call (xn )n=1 :

(a) increasing if xn+1 ≥ xn for n ∈ N;

(b) decreasing if xn+1 ≤ xn for n ∈ N;

39
(c) monotonic if it is increasing or decreasing.

Remark. If we have “>” and “<” in (a) and (b) instead of “≥” and “≤”, we call (xn )∞
n=1
strictly increasing or decreasing, respectively.

Theorem 6.14. Let (xn )∞ ∞

n=1 be a bounded, monotonic sequence. Then (xn )n=1 converges.

Proof. Without loss of generality, suppose that (xn )∞

n=1 is increasing. Set x := sup{xn :
n ∈ N}. We claim that x = limn→∞ xn .
Let  > 0. Then there is n ∈ N, such that xn > x − . For n ≥ n , this means that

x −  < xn ≤ xn +1 ≤ · · · ≤ xn−1 ≤ xn ≤ x < x + ,

i.e., |xn − x| < . This proves the claim.

Example. Define (xn )∞

n=1 recursively by letting

x2n−1 + 5
x1 := 5 and xn := for n ≥ 2.
2xn−1
We claim that
√
5 < xn+1 < xn ≤ 5
for n ∈ N and use induction to prove it.
n = 1: Clearly,
√
5 < |{z}
3 < |{z}
5 =5
=x2 =x1
holds.
√
n n + 1: As 5 < xn+1 , we have 5 < x2n+1 and therefore x2n+1 + 5 < 2x2n+1 . Division
by 2xn+1 yields
x2 + 5
xn+2 = n+1 < xn+1 ≤ 5.
2xn+1
As
√  √ 2
x2n+1 − 2 5xn+1 + 5 = xn+1 − 5 > 0,
we have
√
x2n+1 + 5 > 2 5xn+1 .
Division by 2xn+1 again yields
√ x2n+1 + 5
5< .
2xn+1
Consequently, (xn )∞
n=1 is bounded and decreasing, so that x = limn→∞ xn exists in R.
√
Clearly, x ≥ 5 most hold. On the other hand, we have
x2n−1 + 5 x2 + 5 x2 + 5
x = lim xn = lim = lim n = ,
n→∞ n→∞ 2xn−1 n→∞ 2xn 2x
√
so that x2 = 5. It follows that x = 5.

40
Theorem 6.15. Let (xn )∞
n=1 be an unbounded sequence in R. Then:

(i) if (xn )∞
n=1 is increasing, limn→∞ xn = ∞ holds;

(ii) if (xn )∞
n=1 is decreasing, limn→∞ xn = −∞ holds.

Proof. We only prove (i).

Let R ∈ R. As {xn : n ∈ N} is bounded below (by x1 ), it cannot be bounded above.
Therefore, there is nR ∈ N with xnR > R. As (xn )∞ n=1 is increasing, it follows that
xn ≥ xnR > R for all n ≥ nR . This means that limn→∞ xn = ∞.

Exercises
1. Let (xn )∞ ∞
n=1 and (yn )n=1 be convergent sequences in R with limits x and y, respec-
tively, such that xn ≤ yn for n ∈ N. Show that x ≤ y. (Hint: Assume towards a
contradiction that x > y, and set  := x−y
2 . What happens for large enough n?)

2. Let (xn )∞ ∞
n=1 and (yn )n=1 be convergent sequences in R such that xn < yn for n ∈ N.
Does this entail that limn→∞ xn < limn→∞ yn ?

3. Let (xn )∞ ∞
n=1 be a sequence in R converging to zero, and let (yn )n=1 be a bounded
sequence. Show that (xn yn )∞
n=1 is convergent with limn→∞ xn yn = 0.

4. Is the sequence (xn )∞

n=1 with

4n2 + sin n113 − 17n

xn :=
2n2 + cos(n)

for n ∈ N convergent? If so, determine its limit.

5. Let (xn )∞
n=1 be a a convergent sequence in Z. Show that there is n0 ∈ N such that
xn = xn0 for all n ≥ n0 . (Hint: Use the definition of convergence with  := 21 .)

6. Let (xn )∞ ∞
n=1 be a convergent sequence in R with limit x. Show that (|xn |)n=1 is also
convergent with
lim |xn | = |x|.
n→∞

7. Let (xn )∞
n=1 be a convergent sequence of non-negative reals with limit x. Show that
√
∞
xn n=1 is convergent with
√ √
lim xn = x.
n→∞

√ √ √ 2√ √ √
(Hint: x− ≤ x− y
y x + y = |x − y| for all x, y ≥ 0.)
√ ∞
8. Does the sequence n2 + n − n converge? If so, determine its limit.
n=1

41
 9. Let p be a non-zero polynomial. Show that limn→∞ p(n+1)
p(n) = 1
Pn
10. Let θ ∈ R \ {1}, and define sn := k=0 θk for n ∈ N0 . Show that
1 − θn+1
sn =
1−θ
for all n ∈ N0 and conclude that
1
lim sn =
n→∞ 1−θ
if θ ∈ (−1, 1).

11. Let (xn )∞ ∞

n=1 and (yn )n=1 be sequences in (0, ∞) such that limn→∞ xn = ∞ and that
limn→∞ yn exists in (0, ∞) ∪ {∞}. Show that

lim xn yn = ∞.
n→∞

What can you say if limn→∞ yn = 0?

12. (a) Show that

n
X 1 1
=1−
k(k + 1) n+1
k=1
for all n ∈ N.
(b) For n ∈ N, let sn := nk=1 k12 . Show that (sn )∞
P
n=1 converges. (Hint: Show that
∞
(sn )n=1 is bounded using (a).)

13. Define the sequence (xn )∞

n=1 recursively by letting
 
1 2
x1 := 2 and xn := xn−1 + for n ≥ 2.
2 xn−1
Show that
√
(∗) 2 ≤ xn+1 ≤ xn ≤ 2

for all n ∈ N. Conclude that (xn )∞

n=1 converges and compute its limit. (Hint: Note
that (∗) means that xn − xn+1 ≥ 0 and x2n − 2 ≥ 0 for all n ∈ N.)

14. Let the sequence (xn )∞

n=1 be defined recursively through
1 1
x1 := and xn := (x2n−1 + 1) for n ≥ 2.
2 3
Show that (xn )∞
n=1 converges and compute its limit.

15. Let θ > 0, and define the sequence (xn )∞

n=1 inductively through
√ p
x1 := θ and xn+1 = θ + xn for n ∈ N.
√
Show that (xn )∞
n=1 increases and is bounded above by 1 + θ (and therefore con-
verges), and compute its limit.

42
7 lim sup, lim inf, and Cauchy Sequences
Let (xn )∞
n=1 be a sequence in R that is bounded above. For n ∈ N, define

vn := sup{xk : k ≥ n}.

As
{xk : k ≥ n + 1} ⊂ {xk : k ≥ n},

we have vn+1 ≤ vn , i.e., the sequence (vn )∞

n=1 is decreasing, so that limn→∞ vn exists in
R ∪ {−∞}.

Definition 7.1. Let (xn )∞

n=1 be a sequence in R. Then:

(a) if (xn )∞ ∞
n=1 is bounded above, define the limit superior of (xn )n=1 as

lim sup xn := lim sup{xk : k ≥ n};

n→∞ n→∞

otherwise, define lim supn→∞ xn = ∞;

(b) if (xn )∞ ∞
n=1 is bounded below, define the limit inferior of (xn )n=1 as

lim inf xn := lim inf{xk : k ≥ n};

n→∞ n→∞

otherwise, define lim inf n→∞ xn = −∞.

Examples. 1. Clearly, we have

lim sup(−1)n = 1 and lim inf (−1)n = −1.

n→∞ n→∞

2. Equally clearly,

lim inf n = lim inf{k : k ≥ n} = lim n = ∞ = lim sup n

n→∞ n→∞ n→∞ n→∞

holds.
Remark. Note that

inf{xn : n ∈ N} ≤ lim inf xn ≤ lim sup xn ≤ sup{xn : n ∈ N}

n→∞ n→∞

and
lim sup(−xn ) = − lim inf xn .
n→∞ n→∞

Theorem 7.2. Let (xn )∞

n=1 be a sequence in R. Then:

(i) if limn→∞ xn exists in R ∪ {−∞, ∞}, then

lim inf xn = lim xn = lim sup xn ;

n→∞ n→∞ n→∞

43
 (ii) if lim inf n→∞ xn = lim supn→∞ xn , then limn→∞ xn exists in R ∪ {−∞, ∞}.

Proof. For n ∈ N, set

un := inf{xk : k ≥ n} and vn := sup{xk : k ≥ n},

so that
lim un = lim inf xn and lim vn = lim sup xn .
n→∞ n→∞ n→∞ n→∞

(i): We first treat the case where limn→∞ xn = ∞. Let R ∈ R. Then there is nR ∈ N
such that xn > R for all n ≥ nR . It follows that

un = inf{xk : k ≥ n} ≥ inf{xk : k ≥ nR } ≥ R

for such n. As R was arbitrary, this means that

∞ = lim un = lim inf xn ≤ lim sup xn ≤ ∞.

n→∞ n→∞ n→∞

The case where where limn→∞ xn = −∞ is dealt with similarly.

Suppose that x = limn→∞ xn exists in R. Let  > 0. Then there is n ∈ N such that
|xn − x| <  for n ≥ n and, in particular, xn < x +  for those n. It follows that

vn = sup{xk : k ≥ n} ≤ x + 

for n ≥ n , so that
lim sup xn = lim vn ≤ x + .
n→∞ n→∞

As  > 0 was arbitrary, this means that

lim sup xn ≤ x = lim xn .

n→∞ n→∞

Similarly, one sees that limn→∞ xn ≤ lim inf n→∞ xn . As lim inf n→∞ xn ≤ lim supn→∞ ,
this proves the claim.
(ii): Suppose first that lim inf n→∞ xn = lim supn→∞ xn = ∞. As xn ≥ un for all n ∈ N,
it follows that limn→∞ xn = ∞. The case where lim inf n→∞ xn = lim supn→∞ xn = −∞
is dealt with similarly.
Consider the case where

lim inf xn = lim sup xn = x ∈ R.

n→∞ n→∞

Let  > 0. As x = limn→∞ un , there is n1 ∈ N such that |un − x| <  for all n ≥ n1 ; in
particular, we have
x −  < un ≤ xn

44
for those n. Similarly, we find n2 ∈ N such that |vn − x| <  and therefore

xn ≤ vn < x + 

for n ≥ n2 . Letting n := max{n1 , n2 }, we obtain

x −  < xn < x + 

for n ≥ n , so that x = limn→∞ xn .

Cauchy Sequences
Question. For n ∈ N, define
n
X 1 1 1 1
sn := = 1 + + + ··· + .
k 2 3 n
k=1

Then (sn )∞n=1 is an increasing sequence, so that limn→∞ sn exists in R ∪ {∞}. Is it finite
or ∞, i.e., is (sn )∞
n=1 bounded or not?

Definition 7.3. A sequence (xn )∞ n=1 in R is called a Cauchy sequence if, for each  > 0,
there is n ∈ N such that |xn − xm | <  for all n, m ≥ n .

Proposition 7.4. Let (xn )∞ ∞

n=1 be a convergent sequence in R. Then (xn )n=1 is a Cauchy
sequence.

Proof. Set x := limn→∞ xn . Let  > 0. Then there is n ∈ N such that |xn − x| < 2 for
all n ≥ n . It follows that
 
|xn − xm | ≤ |xn − x| + |xm − x| < + =
2 2
for all n, m ≥ n . Therefore, (xn )∞
n=1 is a Cauchy sequence.

Lemma 7.5. Let (xn )∞ ∞

n=1 be a Cauchy sequence in R. Then (xn )n=1 is bounded.

Proof. Choose n0 ∈ N such that |xn − xm | < 1 for all n, m ≥ n0 , so that, in particular,
|xn − xn0 | < 1 for n ≥ n0 . It follows that

|xn | ≤ |xn − xn0 | + |xn0 | < 1 + |xn0 |

for n ≥ n0 . Set
C := max{|x1 |, . . . , |xn0 −1 |, 1 + |xn0 |},

so that |xn | ≤ C for all n ∈ N. This means that (xn )∞

n=1 is bounded.

Theorem 7.6. The following are equivalent for a sequence (xn )∞

n=1 in R:

45
 (i) (xn )∞
n=1 converges in R;

(ii) (xn )∞
n=1 is a Cauchy sequence.

Proof. (i) =⇒ (ii) is Proposition 7.4.

(ii) =⇒ (i): As (xn )∞
n=1 is bounded by Lemma 7.5, lim inf n→∞ xn and lim supn→∞ xn
exist in R. We shall invoke Theorem 7.2(ii) and show that

lim inf xn = lim sup xn .

n→∞ n→∞

Let  > 0, and choose n ∈ N such that |xn − xm | <  and, in particular, xn < xm + 
for n, m ≥ n . It follows that, for all n, m ≥ n , xm +  is an upper bound for {xk : k ≥ n},
so that
vn = sup{xk : k ≥ n} ≤ xm + .
Consequently,
lim sup xn −  = lim vn −  ≤ xm
n→∞ n→∞

holds for all m ≥ n . It follows that lim supn→∞ xn −  is a lower bound for {xk : k ≥ m}
for all m ≥ n so that

lim sup xn −  ≤ um = inf{xk : k ≥ m}

n→∞

for those m. This means that

lim sup xn −  ≤ lim un = lim inf xn .

n→∞ n→∞ n→∞

As  > 0 was arbitrary, this means that lim supn→∞ xn ≤ lim inf n→∞ xn and therefore
lim supn→∞ xn = lim inf n→∞ xn .
Pn 1
Example. For n ∈ N, consider sn := k=1 k . Observe that
2n n 2n 2n
X 1 X1 X 1 X 1 1 1
s2n − sn = − = ≥ =n =
k k k 2n 2n 2
k=1 k=1 k=n+1 k=n+1

for all n ∈ N. This means that (sn )∞n=1 cannot be a Cauchy sequence. It follows that
∞
(sn )n=1 does not converge and therefore has to be unbounded.

Exercises
1. Let ∅ 6= S be bounded above, let t ≥ 0, and set

tS := {tx : x ∈ S}.

Show that
sup(tS) = t sup S.

46
 Use this to conclude that, if (xn )∞
n=1 is any sequence that is bounded above in R
and t ≥ 0, then
lim sup txn = t lim sup xn .
n→∞ n→∞

2. Let (xn )∞ ∞
n=1 and (yn )n=1 be such that xn ≤ yn for all n ∈ N. Show that

lim inf xn ≤ lim inf yn and lim sup xn ≤ lim sup yn .

n→∞ n→∞ n→∞ n→∞

Use this to derive the Squeeze Theorem: If (xn )∞ ∞ ∞

n=1 , (yn )n=1 , and (zn )n=1 are se-
quences in R such that xn ≤ yn ≤ zn for all n ∈ N with (xn )∞ ∞
n=1 and (zn )n=1 con-
vergent and limn→∞ xn = limn→∞ zn , then (yn )∞
n=1 is convergent with limn→∞ yn =
limn→∞ xn = limn→∞ zn .

3. Let (xn )∞
n=1 be a sequence in R such that there is θ ∈ [0, 1) with

|xn+1 − xn | ≤ θn

for all n ∈ N. Show that (xn )∞

n=1 is a Cauchy sequence and therefore converges.
(Hint: Problem 10 in Section 6.)

47
8 Subsequences
Definition 8.1. Let (xn )∞ ∞ ∞
n=1 be a sequence. Then we call (yk )k=1 a subsequence of (xn )n=1
if there are n1 < n2 < · · · < nk < nk+1 < · · · in N such that yk = xnk .
1 ∞ 1 ∞



Examples. 1. ν
n n=1 is a subsequence of n n=1 for every ν ∈ N.

2. For n ∈ N, set xn := (−1)n . Then (x2n )∞ ∞

n=1 and (x2n−1 )n=1 are subsequences of
(xn )∞
n=1 . Note that
x2n = 1 and x2n−1 = −1

for n ∈ N.

3. For x ∈ R, the floor of x is defined as

bxc := max{m ∈ Z : m ≤ x}.

For n ∈ N, set jnk

xn := (−1)n .
2
Then (xn )∞
n=1 enumerates Z: x1 = 0, x2 = 1, x3 = −1, x4 = 2, x5 − 2, etc. Consider
the subsequences (x2n )∞ ∞
n=1 and (x2n−1 )n=1 . We have

x2n = n and x2n−1 = −(n − 1)

for n ∈ N.

Proposition 8.2. Let (xn )∞

n=1 be a sequence in R such that limn→∞ xn exists in R ∪
{−∞, ∞}. Then limk→∞ xnk = limn→∞ xn for every subsequence (xnk )∞ ∞
k=1 of (xn )n=1 .

Proof. Let (xnk )∞ ∞

k=1 be a subsequence of (xn )n=1 .
We first deal with the case where limn→∞ xn = ∞.
Let R ∈ R. Then there is nR ∈ N such that xn > R for all n ≥ nR . As n1 < n2 <
n3 < · · · , there is kR ∈ N such that nkR ≥ nR . It follows that nk ≥ nR for all k ≥ kR , so
that xnk > R for all k ≥ kR . This means that limn→∞ xnk = ∞.
The case where limn→∞ xn = −∞ is dealt with similarly.
We now turn to the case where x = limn→∞ xn ∈ R.
Let  > 0. Then there is n ∈ N such that |xn − x| <  for n ≥ n . As n1 < n2 < n3 <
· · · , there is k ∈ N such that nk ≥ n . It follows that nk ≥ n for all k ≥ k , so that
|xnk − x| <  for all k ≥ k . This means that limk→∞ xnk = x.

On the other hand, a sequence need not converge in order to have convergent (in
R ∪ {−∞, ∞}) subsequences:

48
 • ((−1)n )∞ 2n ∞
n=1 diverges whereas the subsequences ((−1) )n=1 and ((−1)
2n−1 )∞ con-
n=1
verge to 1 and −1, respectively.
 n 
∞
• (−1)n 2 n=1
does not converge in R ∪ {−∞, ∞}, but
 
2n 2n
lim (−1) = lim n = ∞
n→∞ 2 n→∞
 
2n−1 2n − 1
and lim (−1) = lim −(n − 1) = −∞.
n→∞ 2 n→∞

Proposition 8.3. Let (xn )∞

n=1 be a sequence in R. Then:

(i) if (xn )∞ ∞
n=1 is not bounded above, it has an increasing subsequence (xnk )k=1 such that
limk→∞ xnk = ∞;

(ii) if (xn )∞ ∞
n=1 is not bounded below, it has a decreasing subsequence (xnk )k=1 such that
limk→∞ xnk = −∞.

Proof. We only prove (i).

We will inductively construct a sequence n1 < n2 < n3 < · · · < nk < nk+1 < · · · in N
such that
xnk+1 ≥ max{xnk , k}

for k ∈ N. It is then clear that the corresponding subsequence (xnk )∞k=1 is increasing and
satisfies limk→∞ xnk = ∞.
Fix n1 ∈ N arbitrarily.
As (xn )∞n=n1 +1 is not bounded above there is n2 > n1 such that xn2 ≥ max{xn1 , 1}.
As (xn )∞n=n2 +1 is not bounded above, there is n3 > n2 such that xn3 ≥ max{xn2 , 2}.
Continue in this fashion. Suppose that n1 < n2 < · · · < nk have already been
constructed such that
xnj+1 ≥ max{xnj , j}

for j = 1, . . . , k − 1. As (xn )∞
n=nk +1 is not bounded above, there is nk+1 > nk such that
xnk+1 ≥ max{xnk , k}.

Definition 8.4. Let (xn )∞ n=1 be a sequence in R. We call x ∈ R an accumulation point

of (xn )n=1 if there is a subsequence (xnk )∞
∞ ∞
k=1 of (xn )n=1 with limk→∞ xnk = x.

Examples. 1. If (xn )∞
n=1 is convergent, then limn→∞ xn is its only accumulation point.

2. For n ∈ N, let xn := (−1)n . Then −1 and 1 are accumulation points of (xn )∞

n=1 ,
and it is easy to see that these are the only ones.

49
 3. For n ∈ N, set xn := n((−1)n + 1) + n1 . Then
1 1
lim x2n = lim 4n + =∞ and lim x2n−1 = lim = 0.
n→∞ n→∞ 2n n→∞ n→∞ 2n − 1
Therefore, 0 is an accumulation point of (xn )∞ n=1 . Let x ∈ R be any accumulation
point of (xn )n=1 , and let (xnk )k=1 be a subsequence of (xn )∞
∞ ∞
n=1 with limk→∞ xnk = x.
Assume that there are infinitely many even numbers among the nk ’s. The xnk ≥
2nk must hold for infinitely many k’s, so that (xnk )∞ k=1 cannot be bounded and
therefore not be convergent. Hence, there must be k0 ∈ N such that nk is odd for
all k ≥ k0 . This, however, means that—except for finitely many terms—(xnk )∞ k=1
∞
is a subsequence of (x2n−1 )n=1 , so that x = limn→∞ xnk = limn→∞ x2n−1 = 0. It
follows that 0 is the only accumulation point of (xn )∞ n=1 .

4. Every subsequence of (n(−1)n )∞n=1 is unbounded and therefore divergent. Conse-

quently, the sequence has no accumulation points.

Theorem 8.5. Let (xn )∞n=1 be a bounded sequence in R, and let S be the set of its accu-
mulation points. Then:

(i) S 6= ∅;

(ii) max S exists and equals lim supn→∞ xn ;

(iii) min S exists and equals lim inf n→∞ xn .

Proof. Set x := lim supn→∞ xn . We claim that x belongs to S and is an upper bound for
S. For n ∈ N, set again
vn := sup{xν : ν ≥ n}.
We will construct n1 < n2 < n3 < · · · in N such that

lim xnk = lim vn = lim sup xn

k→∞ n→∞ n→∞

by defining n1 < n2 < n3 < · · · inductively such that

1 1
x− < xnk < x +
k k
for k ∈ N.
By the definition of lim supn→∞ xn , there is ñ1 ∈ N such that

x − 1 < vn < x + 1

for all n ≥ ñ1 . As x − 1 cannot be an upper bound for {xν : ν ≥ ñ1 } by the definition of
lim supn→∞ xn , there is must be n1 ≥ ñ1 such that

x − 1 < xn1 ≤ vn1 < x + 1.

50
By the definition of lim supn→∞ xn again, there is ñ2 ∈ N such that
1 1
x− < vn < x +
2 2
1
for all n ≥ ñ2 . As x − 2 cannot be an upper bound for {xν : ν ≥ ñ2 }, there is n2 ≥
max{n1 + 1, ñ2 } with
1 1
x−
< xn2 < x + .
2 2
Suppose now that n1 < n2 < · · · < nk have already been constructed with
1 1
x− < xnj < x +
j j
for j = 1, . . . , k. Choose ñk+1 ∈ N such that
1 1
x− < vn < x +
k+1 k+1
for all n ≥ ñk+1 . Then choose nk+1 ≥ max{nk + 1, ñk+1 } such that
1 1
x− < xnk+1 < x + .
k+1 k+1
1
Let  > 0. Choose k ∈ N such that k <  for all k ≥ k ; it follows that

x −  < xnk < x + 

for those k. All in all, x = limk→∞ xnk . This means that x ∈ S, so that, in particular,
S 6= ∅, which proves (i).
Let x̃ ∈ S be arbitrary. Then there is a subsequence (xnν )∞ ∞
k=1 of (xn )n=1 such that
xνk → x̃. As
xνk ≤ vνk

for k ∈ N it follows that

x̃ = lim xνk ≤ lim vνk = x.
k→∞ k→∞

Hence, x is an upper bound for S and therefore its maximum. This proves (ii).
(iii) is proven analogously.

Corollary 8.6 (Bolzano–Weierstraß Theorem). Every bounded sequence in R has a con-

vergent subsequence.

Corollary 8.7. The following are equivalent for a bounded sequence (xn )∞
n=1 in R:

(i) (xn )∞
n=1 is convergent;

(ii) (xn )∞
n=1 has exactly one accumulation point.

51
Proof. (i) =⇒ (ii) is clear by Proposition 8.2.
(ii) =⇒ (i): Let S be set of accumulation points of (xn )∞
n=1 and suppose that S is a
singleton set. By Theorem 8.5(i) and (ii), this means that

lim sup xn = max S = min S = lim inf xn .

n→∞ n→∞

By Theorem 7.2, this means that (xn )∞

n=1 converges.

Exercises
1. Let (xn )∞
n=1 be a sequence in R. Show that x ∈ R is an accumulation point of
∞
(xn )n=1 if and only if, for each  > 0, there are infinitely many n ∈ N such that
|xn − x| < . (Hint: For the “only if ” part, inductively construct a subsequence
(xnk )∞ ∞ 1
k=1 of (xn )n=1 with |xnk − x| < k for k ∈ N.)

2. Let ∅ 6= S ⊂ R be bounded above. Show that there is an increasing sequence

(xn )∞
n=1 in S with limn→∞ xn = sup S.

3. Let (xn )∞ ∞
n=1 be a sequence in R, let x be an accumulation point of (xn )n=1 , and let
(k )∞
k=1 be a sequence of strictly positive reals such that limk→∞ k = 0. Show that
there is a subsequence (xnk )∞ ∞
k=1 of (xn )n=1 such that |xnk − x| < k for all k ∈ N.

52
9 Infinite Series
Let a1 , a2 , a3 , . . . ∈ R. What is
∞
X
ak = a1 + a2 + a3 + · · ·
k=1

supposed to mean? A naive approach leads easily to nonsensical “results”. For instance,
let ak := (−1)k for k ∈ N. Then—depending on how we bracket the summands—we
obtain
∞
X
ak = −1 + 1 − 1 + 1 − 1 + 1 − 1 + 1 − . . .
k=1
(
(−1 + 1) + (−1 + 1) + (−1 + 1) + (−1 + 1) − · · · = 0,
=
−1 + (1 − 1) + (1 − 1) + (1 − 1) + (1 − 1) + · · · = −1,
so that 0 = −1 and thus 1 = 0. . .
A more rigorous approach is therefore required.
P∞
Definition 9.1. Let (ak )∞
k=1 be a sequence in R. Then the infinite series k=1 ak is the
∞
sequence (sn )n=1 where
Xn
sn := ak
k=1
for n ∈ N. The terms sn are called the partial sums of ∞
P P∞
k=1 ak . We say that ak
Pk=1
∞
converges if there is s ∈ R such that s = limn→∞ sn ; otherwise, we say that k=1 ak
diverges.

Notation. If ∞
P
k=1 ak converges and s = limn→∞ sn , we write
∞
X
ak = s.
k=1
P∞ ∞
The symbol k=1 ak can therefore mean two objects: the sequence (sn )n=1 of partial
sums and—in the case of convergence—its limit. The reasons for this are historical. If
(sn )∞
n=1 diverges to −∞ or ∞, we write
∞
X ∞
X
ak = −∞ or ak = ∞,
k=1 k=1

respectively.
P∞ k
Examples. 1. Consider k=1 (−1) . As
n
(
X 0, for even n ∈ N,
sn = (−1)k =
k=1
−1, for odd n ∈ N,
P∞ k
The series k=1 (−1) diverges.

53
 2. Geometric Series: As shown in the exercises,
∞
X 1
θk =
1−θ
k=0

holds for θ ∈ (−1, 1).

3. Harmonic Series: As seen before,

∞
X 1
= ∞.
k
k=1

P∞ 1 P∞ 1
4. On the other hand, k=1 k2 converges as shown in the exercises; in fact, k=1 k2 =
π2
6 .

The following is immediate:

P∞
Lemma 9.2. Let a1 , a2 , a3 , . . . ≥ 0. Then k=1 ak converges if and only if (sn )∞
n=1 is
bounded.

In this situation, we write

∞
X
ak < ∞.
k=1

Example. Let p > 0. We have

∞
(
X 1 = ∞, for p = 1,
kp < ∞, for p = 2.
k=1
P∞ 1
Which precisely are the p > 0 for which k=1 kp < ∞? As
n n
X 1 X 1
p
≥
k k
k=1 k=1

for p ≤ 1 and
n n
X 1 X 1
p
≤
k k2
k=1 k=1
P∞ 1 P∞ 1
for p ≥ 2, it is clear that k=1 kp = ∞ for p ∈ (0, 1] and k=1 kp < ∞ for p ∈ [2, ∞).
But what if p ∈ (1, 2)?

Theorem 9.3 (Cauchy’s Compression Theorem). Let (ak )∞

k=1 be a decreasing sequence
of non-negative reals. Then we have
∞
X ∞
X
ak < ∞ ⇐⇒ 2k a2k < ∞.
k=1 k=1

54
Proof. For n ∈ N, set
n
X n
X
sn := ak and Sn := 2k a2k .
k=1 k=1
P∞
“=⇒”: Set s := k=1 ak , and note that

s ≥ s2n
= a1 + a2 + a3 + · · · + a2n
= a1 + a2 + (a3 + a4 ) + (a5 + · · · + a8 ) + · · · + (a2n−1 +1 + · · · + a2n )
| {z } | {z } | {z }
≥2a4 ≥4a8 ≥2n−1 a2n

≥ a1 + a2 + 2a4 + 4a8 + · · · + 2n−1 a2n

1
= a1 + Sn
2
and therefore
Sn ≤ 2(s − a1 ).
P∞ k
for n ∈ N. It follows that (Sn )∞
n=1 is bounded, so that k=1 2 a2k < ∞.
“⇐=”: As (sn )n=1 is increasing, it is enough to show that (s2n )∞
∞
n=1 is bounded. Set
P∞ k
S := k=1 2 a2k . For n ∈ N, we have

s2n = a1 + a2 + a3 + · · · + a2n
= a1 + (a2 + a3 ) + (a4 + · · · + a7 ) + · · · + (a2n−1 + · · · + a2n −1 ) + a2n
| {z } | {z } | {z }
≤2a2 ≤4a4 ≤2n−1 a2n−1

≤ a1 + Sn−1 + a2n
≤ 2a1 + S,
which proves the boundedness of (s2n )∞
n=1 .

∞
Example. Let p > 0. As k1p k=1 is decreasing, Cauchy’s Compression Theorem applies,
so that
∞ ∞ ∞
X 1 X 2k X
< ∞ ⇐⇒ = (21−p )k < ∞
kp (2k )p
k=1 k=1 k=1
1−p
⇐⇒ 2 <1
⇐⇒ p > 1.
P∞ 1
Hence, for instance, k=1 3 < ∞ holds.
k2
Theorem 9.4 (Cauchy Criterion for Infinite Series). Let (ak )∞ k=1 be a sequence in R.
P∞
Then k=1 ak converges if and only if, for each  > 0, there is n ∈ N such that
n
X
ak < 
k=m+1

55
for all n > m ≥ n .
P∞
Proof. Let (sn )∞
n=1 be the sequence of partial sums of k=1 ak and note that
P∞
k=1 ak converges ⇐⇒ (sn )∞
n=1 converges
⇐⇒ ∞
(sn )n=1 is a Cauchy sequence
⇐⇒ for every  > 0, there is n ∈ N such that
|sn − sm | <  for all n, m ≥ n
⇐⇒ for every  > 0, there is n ∈ N such that
|sn − sm | <  for all n > m ≥ n .
As
n
X m
X n
X
sn − sm = ak − ak = ak
k=1 k=1 k=m+1
for n > m, this yields the claim.
P∞
Corollary 9.5. Let (ak )∞
k=1 be a sequence in R such that k=1 ak converges. Then
limk→∞ ak = 0.

Proof. Let  > 0, and let n ∈ N be as in the Cauchy Criterion. Then

n
X
|an | = ak < 
k=(n−1)+1

for all n ≥ n + 1.

Absolute Convergence
P∞
Definition 9.6. Let (ak )∞
k=1 ak be a sequence in R. We call k=1 ak absolutely convergent
P∞
if k=1 |ak | < ∞.

Example. Let θ ∈ (−1, 1). Then the geometric series ∞ k

P
k=0 θ is absolutely convergent.
P∞
Proposition 9.7. Let (ak )∞k=1 be a sequence in R such that k=1 ak converges absolutely.
P∞
Then k=1 ak converges.

Proof. Let  > 0. Applying the Cauchy Criterion to ∞

P
k=1 |ak |, we obtain n ∈ N such
that
Xn
|ak | < 
k=m+1
for all n > m ≥ n . It follows that
n
X n
X
ak ≤ |ak | < 
k=m+1 k=m+1
P∞
for n > m ≥ n . Applying the Cauchy Criterion again—this time to k=1 ak —we obtain
that ∞
P
k=1 ak converges.

56
 Does the converse hold?

Theorem 9.8 (Alternating Series Test). Let (ak )∞ be a decreasing sequence of non-
P∞ k=1 k−1
negative reals such that limk→∞ ak = 0. Then k=1 (−1) ak converges.

Proof. For n ∈ N, set

n
X
sn := (−1)k−1 ak .
k=1
As
s2n+2 − s2n = −a2n+2 + a2n+1 ≥ 0
for n ∈ N, the subsequence (s2n )∞ ∞
n=1 of (sn )n=1 is increasing. Similarly, we have

s2n+1 − s2n−1 = a2n+1 − a2n ≤ 0

for n ∈ N, so that (s2n−1 )∞

n=1 is decreasing. Moreover,

s2 ≤ s2n = s2n−1 − a2n ≤ s2n−1 ≤ s1

holds for n ∈ N, so that both (s2n )∞ ∞

n=1 and (s2n−1 )n=1 are bounded and therefore conver-
gent.
Set s := limn→∞ s2n−1 . We claim that s = ∞ k−1 a .
P
k=1 (−1) k
Let  > 0. Then there is n1 ∈ N such that |s2n−1 − s| < 2 for all n ≥ n1 . As
limk→∞ ak = 0, there is n2 ∈ N such that |an | < 2 for all n ≥ n2 . Set n := max{2n1 , n2 },
and let n ≥ n .
Case 1: n is odd, i.e., n = 2m − 1 for some m ∈ N. As n ≥ 2n1 , we have 2m ≥ 2n1 + 1,
so that m ≥ n1 . It follows that

|sn − s| = |s2m−1 − s| < < .
2
Case 2: n is even, i.e., n = 2m for some m ∈ N. It follows that
 
|sn − s| = |s2m − s| = |s2m−1 − an − s| ≤ |s2m−1 − s| + |an | ≤ + = .
2 2
This completes the proof.

∞
Example. As k1 k=1 is decreasing and converges to zero, the Alternating Series Test yields
(−1)k−1 (−1)k−1
the convergence of the alternating harmonic series ∞ ; in fact, ∞
P P
k=1 k k=1 k =
P∞ 1
log 2. But as k=1 k = ∞, it does not converge absolutely.

Proposition 9.9. Let (ak )∞ ∞

k=1 and (bk )k=1 be sequences in R, and let α, β ∈ R. Then:

(i) if ∞
P P∞ P∞
k=1 ak and k=1 bk converge, k=1 (αak + βbk ) converges as well, and
∞
X ∞
X ∞
X
(αak + βbk ) = α ak + β bk
k=1 k=1 k=1

holds;

57
 (ii) if ∞
P P∞ P∞
k=1 ak and k=1 bk converge absolutely, k=1 (αak + βbk ) converges absolutely
as well.

Proof. (i) follows immediately from the limit laws.

For (ii), let n ∈ N, and note that
n
X n
X
|αak + βbk | ≤ (|αak | + |βbk |)
k=1 k=1
n
X n
X
= |α| |ak | + |β| |bk |
k=1 k=1
∞
X ∞
X
≤ |α| |ak | + |β| |bk |.
k=1 k=1
P∞
It follows that k=1 |αak + βbk | < ∞.

Exercises
1. For k ∈ N, let
1 (−1)k
ak := − √ ,
k k
P∞ k−1 a diverges. Why doesn’t this
so that limk→∞ ak = 0. Show that k=1 (−1) k
contradict the Alternating Series Test?
P∞
2. Let (ak )∞ ∞
k=1 and (bk )k=1 be sequences of non-negative reals such that k=1 ak < ∞
P∞ P∞ √
and k=1 bk < ∞. Show that k=1 ak bk < ∞. (Hint: First, prove the inequality
√
between the arithmetic and the geometric mean: ab ≤ a+b 2 for all a, b ≥ 0.)
P∞
3. Let (n )∞
n=1 be a sequence of non-negative reals such that k=1 k < ∞, and let
∞
(xn )n=1 be a sequence in R such that

|xn+1 − xn | ≤ n

for all n ∈ N. Show that (xn )∞

n=1 converges. Does this conclusion remain valid if we
only require that limn→∞ n = 0?

58
10 Convergence Criteria
Theorem 10.1 (Comparison Test). Let (ak )∞ ∞
k=1 and (bk )k=1 with bk ≥ 0 for k ∈ N.

(i) Suppose that ∞

P
k=1 bk < ∞ and that there is n0 ∈ N such that |ak | ≤ bk for k ≥ n0 .
P∞
Then k=1 ak converges absolutely.

(ii) Suppose that ∞

P
k=1 bk = ∞ and that there is n0 ∈ N such that ak ≥ bk for k ≥ n0 .
P∞
Then k=1 ak diverges.

Proof. (i): For n ≥ n0 , we have

n
X 0 −1
nX n
X 0 −1
nX n
X 0 −1
nX ∞
X
|ak | = |ak | + |ak | ≤ |ak | + bk ≤ |ak | + bk .
k=1 k=1 k=n0 k=1 k=n0 k=1 k=1
P∞ P∞
Hence, the sequence of partial sums of k=1 |ak | is bounded. This means that k=1 ak
converges absolutely.
(ii): For n ≥ n0 , we have
n
X 0 −1
nX n
X 0 −1
nX n
X
ak = ak + ak ≥ ak + bk ,
k=1 k=1 k=n0 k=1 k=n0
P∞ P∞
so that the sequence of partial sums of k=1 ak is unbounded. Therefore, k=1 ak di-
verges.

Example. For k ∈ N, let

sin(43k 7 − 4k + 13)
ak := .
4k 2 + cos(ek4 − 7k)
As
1
|ak | ≤ 2
3k
P∞ 1 P∞
for k ∈ N and k=1 3k2 < ∞, the series k=1 ak converges absolutely.

Corollary 10.2 (Limit Comparison Test). Let (ak )∞ ∞

k=1 and (bk )k=1 with bk ≥ 0 for k ∈ N.
|ak |
(i) Suppose that ∞
P P∞
k=1 bk < ∞ and that limk→∞ bk exists in R. Then k=1 ak con-
verges absolutely.
P∞ ak
(ii) Suppose that k=1 bk = ∞ and that limk→∞ bk exists in (0, ∞) ∪ {∞}. Then
P∞
k=1 ak diverges.
 ∞
Proof. (i): Let n0 ∈ N be such that bk 6= 0 for all k ≥ n0 . As |abkk | converges, it
k=n0
|ak |
is bounded, which means that there is C ≥ 0 with ≤ C, i.e., |ak | ≤ Cbk . Apply the
bk
Comparison Test.
(ii): Let n0 ∈ N and δ > 0 be such that bk 6= 0 and abkk > δ, i.e., ak > δbk for k ≥ n0 .
Again, apply the Comparison Test.

59
Example. For k ∈ N, let
4k 2 − 3k + 3
ak :=
3k 3 + 2k − 4
and
1
bk := .
k
It follows that
ak 4k 3 − 3k 2 + 3k 4 − k3 + k32 4
bk
=
3k 3 + 2k − 4
= 2 4 → 3 > 0.
3 + k2 − k3
P∞
Therefore, k=1 ak diverges.

Theorem 10.3 (Ratio Test). Let (ak )∞

k=1 be a sequence in R.

|ak+1 |
(i) Suppose that there are n0 ∈ N and θ ∈ (0, 1) such that ak 6= 0 and |ak | ≤ θ for all
k ≥ n0 . Then ∞
P
k=1 ak converges absolutely.

|ak+1 |
(ii) Suppose that there are n0 ∈ N and θ ≥ 1 such that ak 6= 0 and |ak | ≥ θ for all
k ≥ n0 . Then ∞
P
k=1 ak diverges.

Proof. (i): Clearly, |ak+1 | ≤ θ|ak | holds for all k ≥ n0 .

We claim that |ak | ≤ θk−n0 |an0 | for all k ≥ n0 .
This is clear for k = n0 . Suppose the claim is true for k ≥ n0 . Then we obtain

|ak+1 | ≤ θ|ak | ≤ θ θk−n0 |an0 | = θ(k+1)−n0 |an0 |.

Induction yields the claim.

As ∞
P k−n0 |a | = θ −n0 |a |
P∞ k
k=1 θ n0 n0 k=1 θ < ∞, it follows from the Comparison Test
P∞
that k=1 ak converges absolutely.
(ii): Clearly, |ak+1 | ≥ θ|ak | holds for all k ≥ n0 . Inductively, we obtain that

|ak | ≥ θk−n0 |an0 | ≥ |an0 | > 0

P∞
for k ≥ n0 . This means that (ak )∞
k=1 cannot converge to zero, so that k=1 ak must
diverge.

Corollary 10.4 (Limit Ratio Test). Let (ak )∞

k=1 be a sequence in R such that there is
n0 ∈ N with ak 6= 0 for all k ≥ n0 . Then:
P∞ |ak+1 |
(i) k=1 ak converges absolutely if limk→∞ |ak | < 1;
P∞ |ak+1 |
(ii) k=1 ak diverges if limk→∞ |ak | > 1 in R ∪ {∞}.

60
 |ak+1 |
Remark. Nothing can be said if limk→∞ |ak | = 1. For instance, let

1 1
ak := and bk :=
k2 k
|ak+1 | |bk+1 |
for k ∈ N. Then limk→∞ |ak | = limk→∞ |bk | = 1 holds whereas

∞
X ∞
X
ak < ∞ and bk = ∞.
k=1 k=1

xk P∞
Example. Fix x ∈ R. For k ∈ N, let ak := k! . Clearly, k=0 ak converges if x = 0. If
x 6= 0, we have ak 6= 0 for all k ∈ N0 and

|ak+1 | |x|k+1 k! |x|

= k
= → 0.
|ak | (k + 1)! |x| k
P∞
Therefore, k=0 ak converges absolutely.

Theorem 10.5 (Root Test). Let (ak )∞

k=1 be a sequence in R.
p
(i) Suppose that there are n0 ∈ N and θ ∈ (0, 1) such that k |ak | ≤ θ for all k ≥ n0 .
Then ∞
P
k=1 ak converges absolutely.
p
(ii) Suppose that there is θ ≥ 1 such that k |ak | ≥ θ for infinitely many k ∈ N. Then
P∞
k=1 ak diverges.

Proof. (i): We have |ak | ≤ θk for k ≥ n0 . Applying the Comparison Test yields the claim.
(ii): We have |ak | ≥ θk ≥ 1 for infinitely many k ∈ N. This rules out that (ak )∞ k=1
converges to zero, so that ∞
P
k=1 ak must diverge.

Example. For k ∈ N, let

2 + (−1)k
ak := ,
2k−1
so that
(
1
ak+1 2 + (−1)k+1 2k−1 1 2 + (−1)k+1 6, for even k,
= = =
ak 2k 2 + (−1)k 2 2 + (−1)k 3
2, for odd k.

Hence, we cannot apply the Ratio Text to the series ∞

P
k=1 ak . However, we have
r √
k
√ k
k 2(2 + (−1) ) 6 1
k
ak = ≤ →
2k 2 2
√
Therefore, there is n0 ∈ N such that k ak < 23 for k ≥ n0 , so that ∞
P
k=1 ak converges
absolutely.

Corollary 10.6 (Limit Root Test). Let (ak )∞

k=1 be a sequence in R. Then:

61
 P∞ p
(i) k=1 ak converges absolutely if lim supk→∞ k |ak | < 1;
P∞ pk
(ii) k=1 a k diverges if lim supk→∞ |ak | > 1 in R ∪ {∞}.
p
Proof. (i): Let θ ∈ R be such that lim supk→∞ k |ak | < θ < 1. Then there is n0 ∈ N such
p
that k |ak | < θ. The claim then follows from the Root Test. q p
(ii): Let k1 < k2 < k3 < · · · be such that limj→∞ kj |akj | = lim supk→∞ k |ak | > 1.
q p
Then there must be j0 ∈ N such that kj |akj | ≥ 1 for all j ≥ j0 , i.e., k |ak | ≥ 1 for
infinitely many k ∈ N. The Root Test then yields the divergence of ∞
P
k=1 ak .
p
As for the Limit Ratio Test, no conclusion is possible if limk→∞ k |ak | = 1.

Exercises
1. Determine whether or not the following series converge or converge absolutely:
∞ ∞ ∞
X cos(νπ) X k! X 1
(a) ; (b) ; (c) .
ν+1 kk m log m
ν=0 k=1 m=2

(Hint for (c): Cauchy’s Comppression Theorem.)

2. Determine whether or not the following series diverge, converge, or converge abso-
lutely:
∞
X 3 2 sin k
(a) (−1)k +3k −7k+13 ;
cos k + k 2
k=1
∞  
X 1 2ν
(b) ;
νν ν
ν=1
∞
Xnn
(c) ;
n!
n=1
∞    
X 2n . 3n
(d) ;
n n
n=1
∞
X (−1)ν + 2
(e) ;
(−1)ν−1 ν
ν=1
P∞ sin( π2 +mπ )
(f) m=1
√
m
.
P∞ k
3. Let p be a polynomial, and let θ ∈ (−1, 1). Show that the series k=1 p(k)θ
converges absolutely.
√
4. (a) Show that lim supn→∞ n
n ≤ 1.

62
 √
n
√
(b) Conclude that that limn→∞ n = 1 and limn→∞ n r = 1 for all r > 0.
√
(Hint for (a): Assume that lim supn→∞ n n > 1 and arrive a contradiction to the
fact that ∞ n
P
n=1 n θ < ∞ for all θ ∈ (0, 1).)

5. Let p and q be polynomials, let ν be the degree of p, and let µ be the degree of q.
Suppose that n0 is such that q(k) 6= 0 for all k ≥ n0 . Show that the series ∞ p(k)
P
k=n0 q(k)
converges if and only if µ − ν ≥ 2. (Hint: Limit Comparison Test.)

6. Let (an )∞ ∞
n=1 be a sequence in R such that lim inf n→∞ |an | = 0, and let (Rk )k=1 be
a sequence of non-zero reals. Show that (an )∞ ∞
n=1 has a subsequence (ank )k=1 such
P∞
that k=1 Rk ank converges absolutely.

7. Show that that

n
X nxn+1 − (n + 1)xn + 1
kxk = x
(1 − x)2
k=1

for all n ∈ N, and conclude that

∞
X x
kxk =
(1 − x)2
k=1

if |x| < 1.

63
11 More on Absolute Convergence
P∞
Theorem 11.1. Let (ak )∞k=1 be a sequence in R such that k=1 ak converges absolutely,
P∞
and let σ : N → N be bijective. Then k=1 aσ(k) is also absolutely convergent with
P∞ P∞
k=1 aσ(k) = k=1 ak .

Proof. Set x := ∞
P
k=1 ak . Let  > 0. By the Cauchy Criterion for infinite series, there is
n0 ∈ N such that nk=n0 |ak | < 3 for all n ≥ n0 , so that
P

∞
X  
|ak | ≤ < .
3 2
k=n0

It follows that
0 −1
nX ∞ ∞
X X 
x− ak = ak ≤ |ak | < .
2
k=1 k=n0 k=n0

Choose n ∈ N so large that

{σ −1 (1), . . . , σ −1 (n0 − 1)} ⊂ {1, . . . , n },

i.e.,
{1, . . . , n0 − 1} ⊂ {σ(1), . . . , σ(n )}.

Let n ≥ n . We obtain
n
X 0 −1
nX 0 −1
nX n
X
x− aσ(k) ≤ x − ak + ak − aσ(k)
k=1 k=1 k=1 k=1
 X
< + |ak |
2
k∈{σ(1),...,σ(n)}\{1,...,n0 −1}
∞
 X
≤ + |ak |
2
k=n0
 
< +
2 2
= .

This proves that ∞

P
k=1 aσ(k) = x.
The same argument applied to the series ∞
P P∞
k=1 |ak | shows that k=1 |aσ(k) | converges,
P∞
so that k=1 ak is absolutely convergent.

Theorem 11.2 (Riemann’s Rearrangement Theorem). Let (ak )∞ k=1 be a sequence in R

P∞
such that k=1 ak converges, but is not absolutely convergent, and let x ∈ R ∪ {−∞, ∞}.
Then there is a bijection σ : N → N such that ∞
P
k=1 aσ(k) = x.

64
Proof. We need to show to find a rearrangement of a1 , a2 , a3 , . . . such that the rearrange
series converges to x.
Let b1 , b2 , b3 , . . . denote the non-negative terms of (ak )∞
k=1 , and let c1 , c2 , c3 , . . . the
P∞
strictly negative ones. We claim that k=1 bk = ∞. Otherwise,
∞
X ∞
X ∞
X
ck = ak − bk
k=1 k=1 k=1

would converge as would consequently

∞
X ∞
X ∞
X
|ak | = bk − ck .
k=1 k=1 k=1
P∞
This would contradict the hypothesis that k=1 ak does not converge absolutely. Simi-
larly, one sees that ∞
P
k=1 ck = −∞.
Case 1: x = ∞.
As ∞
P
k=1 bk = ∞, there is n1 ∈ N such that
n1
X
bk > 1 − c1 .
k=1
P∞
As k=n1 +1 bk = ∞, there is n2 ∈ N, n2 > n1 such that
n+1
X n2
X
bk + c1 + bk > 2 − c2 .
k=1 k=n1 +1
P∞
As k=n2 +1 bk = ∞, there is n3 ∈ N, n3 > n2 such that
n+1
X n2
X n3
X
bk + c1 + bk + c2 + > 3 − c3 .
k=1 k=n1 +1 k=n2 +1

Continuing in this fashion, we obtain a rearrangement

b1 , . . . , bn1 , c1 , bn1 +1 , . . . bn2 , c2 , bn2 +1 , . . . , bn3 , c3 , bn+3+1 , . . .

of a1 , a2 , a3 , . . ..
For N ∈ N, let sN denote the N th partial sum of the rearranged series. If ν ∈ N is
such that N ≥ nν , then sN ≥ ν. It follows that the rearranged series diverges to ∞.
Case 2: x = −∞.
This is dealt with similarly.
Case 3: x ∈ R.
Choose n1 ∈ N minimal such that
n1
X
bk > x.
k=1

65
Then choose m1 ∈ N minimal such that
n1
X m1
X
bk + ck < x.
k=1 k=1

Now choose n2 ∈ N, n2 > n1 minimal such that

n1
X m1
X n2
X
bk + ck + bk > x.
k=1 k=1 k=n1 +1

Then choose m2 ∈ N, m2 > m1 such that

n1
X m1
X n2
X m2
X
bk + ck + bk + ck < x.
k=1 k=1 k=n1 +1 k=m1 +1

Continuing in this fashion, we obtain a rearrangement

b1 , . . . , bn1 , c1 , . . . , cm1 , bn1 +1 , . . . bn2 , cm1 +1 , . . . cm2 , bn2 +1 , . . . , bn3 , cm2 +1 , . . .

of a1 , a2 , a3 , . . ..
For N ∈ N, let sN denote the N th partial sum of the rearranged series. Then sN is of
one of the following forms for some ν ∈ N:
n1
X m1
X n
X
(2) bk + ck + · · · + bk
k=1 k=1 k=nν +1

with n ≤ nν+1 or
n1
X m1
X n
X m
X
(3) bk + ck + · · · + bk + ck
k=1 k=1 k=nν +1 k=mν +1

with m ≤ mν+1 .
Suppose that SN is of the form (2). Then the minimality of nν+1 yields that

|x − SN | ≤ bnν+1 ,

whereas if n = nn+1 and

|x − SN | ≤ −cmν

if n < nν+1 . Similar estimates hold if SN is of the form (3). All in all

|x − SN | ≤ max{bnν+1 , cmν , −cmν+1 , bnν }

holds.
As ∞
P
k=1 ak converges, limk→∞ ak = 0 holds and therefore limk→∞ bk = limk→∞ ck = 0
as well. It follows that x = limN →∞ sN .

66
Theorem 11.3 (Cauchy Product Formula). Let (ak )∞ ∞
k=0 and (bk )k=0 be sequences in R
P∞ P∞
such that k=0 ak and k=0 bk converge absolutely. For k ∈ N0 , set
k
X
ck := aj bk−j .
j=0
P∞
Then the series k=0 ck converges absolutely such that
∞ ∞
! ∞ !
X X X
ck = ak bk .
k=0 k=0 k=0

Proof. For n ∈ N, set

n
X n
X n
X
An := ak , Bn := bk , Cn := ck
k=0 k=1 k=1

and
n n
! !
X X
Pn := |ak | |bk | .
k=0 k=0

We claim that
lim (An Bn − Cn ) = 0.
n→∞

Note first that, for n ∈ N, we have

n X
X k X X
Cn = aj bk−j = aj bl and An Bn = aj bl ,
k=0 j=0 0≤j,l 0≤j,l≤n
j+l≤n

so that
X
An B n − C n = aj bl .
0≤j,l≤n
j+l>n

Let  > 0. By the absolute convergence of ∞

P P∞ ∞
k=0 ak and k=0 bk , the sequence (Pn )n=1
converges and therefore is a Cauchy sequence. This means that there is n ∈ N such that

Pn − Pn = |Pn − Pn | < 

67
for all n ≥ n . For n ≥ 2n , we obtain
X
|An Bn − Cn | ≤ |aj bl |
0≤j,l≤n
j+l>n
X
≤ |aj bl |
0≤j,l≤n
j+l>2n
X
≤ |aj bl |
0≤j,l≤n
j>n or l>n
X X
= |aj ||bl | − |aj ||bl |
0≤j,l≤n 0≤j,l≤n

= Pn − Pn
< .

This proves our claim, so that

∞ ∞ ∞
! !
X X X
ck = ak bk .
k=0 k=0 k=0

holds.
To see that ∞
P
k=0 ck actually converges absolutely, let n ∈ N, and observe that

n n X
k n n ∞ ∞
! ! ! !
X X X X X X
|ck | ≤ |aj ||bk−j | ≤ |ak | |bk | ≤ |ak | |bk | ,
k=0 k=0 j=0 k=0 k=0 k=0 k=0

which completes the proof.

Exercises
1. (Failure of the Cauchy Product Formula without absolute convergence.) For k ∈ N0 ,
set
k
(−1)k X
ak := bk := √ and ck := aj bk−j .
k+1 j=0

Show that ∞
P P∞ ∞
k=0 ak and k=0 bk converge, but that (ck )k=0 does not converge to
P∞
zero (so that, in particular, k=0 ck does not converge).

68
12 The Exponential Function
The exponential function exp : R → R is defined for x ∈ R as
∞
X xk
exp(x) := .
k!
k=0

As we saw, the series converges absolutely for each x ∈ R. For x, y ∈ R, we obtain from
the Cauchy Product Formula
∞
! ∞ !
X xk X yk
exp(x) exp(y) =
k! k!
k=0 k=0
∞ X
k
X xj y k−j
= , by Theorem 11.3,
j! (k − j)!
k=0 j=0
∞ k
X 1 X k!
= xj y k−j
k! j!(k − j)!
k=0 j=0 | {z }
=(kj)
∞
X (x + y)k
= , by the Binomial Theorem,
k!
k=0

= exp(x + y).

By induction, we see that

· · + x}) = exp(x)n
| + ·{z
exp(nx) = exp(x
n times

for any x ∈ R and n ∈ N. We define Euler’s constant as e := exp(1), so that exp(n) = en

for all n ∈ N. Clearly, exp(0) = 1 = e0 holds. For any x ∈ R, we have

1 = exp(0) = exp(x − x) = exp(x) exp(−x),

1
so that exp(x) 6= 0 and exp(−x) = exp(x) . In particular,

1 1
exp(−n) = = n = e−n
exp(n) e

holds for all n ∈ N. All in all, we have exp(m) = em for all m ∈ Z. Next, note that
x x  x 2
exp(x) = exp + = exp >0
2 2 2
m
for any x ∈ R. Let q ∈ Q, and choose m ∈ Z and n ∈ N such that q = n. We have

exp(q)n = exp(nq) = exp(m) = em ,

69
so that
√
exp(q) = n
em = eq .

In view of this, we define

ex := exp(x)

for all x ∈ R.
We defined Euler’s constant e to be exp(1). We claim that

1 n
 
e = lim 1 +
n→∞ n

We proceed by proving several inequalities.

Claim 1.
n
1 n X 1
 
2≤ 1+ ≤ .
n k!
k=0

for all n ∈ N.

Proof. By Bernoulli’s Inequality, we have

1 n
 
n
1+ ≥1+ =2
n n

for all n ∈ N, which proves the first inequality.

For n ∈ N, note that
 n n  
1 X n 1
1+ =
n k nk
k=0
 
n k
X Y n − k + j  1
=1+ 
j nk
k=1 j=1
n k
X 1 Y n−k+j
=1+
k! n }
k=1 j=1 | {z
≤1
n
X 1
≤1+
k!
k=1
n
X 1
= ,
k!
k=0

which proves the second inequality.

Claim 2.
∞
X 1
e= ≤ 3.
k!
k=0

70
Proof. First note that, by induction, it is clear that

k! ≥ 2k−1

for all k ∈ N.
Let n ∈ N, and note that
n n n n−1  
X 1 X 1 X 1 X 1 1
=1+ ≤1+ =1+ =1+2 1− n .
k! k! 2k−1 2k 2
k=0 k=1 k=1 k=0

Letting n → ∞ yields the equality.

1 n ∞



In particular, Claims 1 and 2 guarantee that the sequence 1+ n n=1
is bounded.

Claim 3.
1 n 1 m
   
1+ > 1+
n m
for all n, m ∈ N with n > m.

Proof. Let n, m ∈ N with n > m. Then the Binomial Theorem yields

n  
1 n X n 1
 
1+ =
n k nk
k=0
m  
X n 1
>
k nk
k=0
m k
X 1 Y n−j+1
=1+
k! n
k=1 j=1
m k  
X 1 Y j−1
=1+ 1−
k! n
k=1 j=1
m k  
X 1 Y j−1
≥1+ 1−
k! m
k=1 j=1
m k
X 1 Y m−j+1
=1+
k! m
k=1 j=1
m  
X m 1
=
k mk
k=0
1 m
 
= 1+ ,
m
which proves the claim.

n
∞
By Claims 1 and 2, the sequence 1 + n1 n=1
is bounded below by 2 and above by 3.

n
Claim 3 asserts that it is also increasing. Therefore, we now know that limn→∞ 1 + n1
exists. It remains to be shown that it equals e.

71
Claim 4.
1 n
        
1 1 1 2 m−1 1
1+ ≥1+1+ 1− + ··· + 1 − 1− ··· 1 −
n n 2! n n n m!
for all n, m ∈ N with n > m.

Proof. For n > m, the Binomial Theorem yields

n  
1 n X n 1
 
1+ =
n k nk
k=0
m  
X n 1
≥
k nk
k=0
m k
X 1 Y n−j+1
=1+ ,
k! n
k=1 j=1
m k  
1 Y
X j−1
=1+ 1−
k! n
k=1 j=1
      
1 1 1 2 m−1 1
=1+1+ 1− + ··· + 1 − 1− ··· 1 − .
n 2! n n n m!
which proves the claim.

We can now put it all together.

Fix m ∈ N. Letting n → ∞, we obtain that
m
1 n
 
1 1 X 1
lim 1 + ≥ 1 + 1 + + ··· + = .
n→∞ n 2! m! k!
k=0

1 n


Letting m → ∞ yields limn→∞ 1 + n ≥ e. The reversed inequality follows from Claim
1.
More generally, one can show that
 x n
ex = exp(x) = lim 1+
n→∞ n
for all x ∈ R (see Exercise 2 below).

Exercises
1. Let f : R → R be a non-zero, continuous function such that f (x + y) = f (x)f (y) for
all x, y ∈ R. Show that there is C ∈ R such that f (x) = exp(Cx) for x ∈ R.

2. Show that show that  x n

exp(x) = lim 1+
n→∞ n
for all x ∈ R. Proceed as follows:

72
• for x ≥ 0, modify the argument in the case where x = 1;

n  2 n

n 
• for x < 0, note that 1 + nx 1 − nx = 1 − nx2 , and observe what happens
as n → ∞.

73
13 Closed and Compact Subsets of R
Definition 13.1. Let F ⊂ R. We call F closed if, for every convergent sequence (xn )∞
n=1
in F , i.e., xn ∈ F for all n ∈ N, its limit also lies in F .

Examples. 1. ∅ and R are closed for trivial reasons.

2. All singleton subsets of R, i.e., sets of the form {x} with x ∈ R are closed.

3. Let a < b. Then (−∞, a], [a, b], and [b, ∞) are closed.
1
4. The set (0, 1] is not closed. To see this, let xn := n ∈ (0, 1] for n ∈ N. Then
0 = limn→∞ xn does not lie in (0, 1].

5. Let x ∈ R. For each n ∈ N, there is qn ∈ Q with x < qn < x + n1 , so that

x = limn→∞ qn . As x may be irrational, this means that Q is not closed.

6. Let (xn )∞
n=1 be a sequence in R, and let

S := {x ∈ R : x is an accumulation point of (xn )∞

n=1 }.

We claim that S is closed. To see this, let (sν )∞

ν=1 be a convergent sequence in S,
and let s := limν→∞ sν . We must find a subsequence (xnk )∞ ∞
k=1 of (xn )n=1 such that
s = limk→∞ xnk . We can find ν1 < ν2 < ν3 < · · · such that

1
|sνk − s| <
k
for all k ∈ N. For any k ∈ N, set
 
1
Nk := n ∈ N : |sνk − xn | < .
k

As sνk is an accumulation point for (xn )∞ n=1 for each k ∈ N, the set Nk is infinite
for each k ∈ N. Pick n1 ∈ N1 . As N2 is infinite, there is n2 ∈ N2 with n2 > n1 .
Continue in this fashion. Suppose that n1 < n2 < · · · < nk have already been chosen
such that nj ∈ Nj for j = 1, . . . , k. As Nk+1 is infinite, there is nk+1 ∈ Nk+1 with
nk+1 > nk . For n1 < n2 < n3 < · · · chosen this way, we have
1 1 2
|s − xnk | ≤ |s − sνk | + |sνk − xnk | < + =
k k k
for all k ∈ N. This means that s = limk→∞ xnk , so that s ∈ S.

Definition 13.2. Let S ⊂ R. The closure of S is defined as

n o
S := x ∈ R : there is a sequence (xn )∞
n=1 in S with x = lim xn . n→∞

74
Remarks. 1. S ⊂ S for every S ⊂ R.

2. S = S if and only if S is closed.

3. Q = R.

Proposition 13.3. Let S ⊂ R. Then S is closed. Moreover, S ⊂ F for any closed F ⊂ R

with S ⊂ F .

Proof. We first prove the “moreover” part.

Let F ⊂ R be closed such that S ⊂ F , and let x ∈ S. Then there is a sequence (xn )∞
n=1
∞
in S such that x = limn→∞ xn . As S ⊂ F , the sequence (xn )n=1 also lies in F , and as F
is closed, x = limn→∞ xn ∈ F holds.
To see that S is indeed closed, let (sn )∞
n=1 be a convergent sequence in S, and set
s := limn→∞ sn . We need to find a sequence (xk )∞ k=1 in S such that s = limk→∞ xk .
Choose n1 < n2 < n3 < · · · such that
1
|snk − s| <
k
for k ∈ N. For each k ∈ N, the term snk is the limit of a convergent sequence in S; in
particular, there is xk ∈ S such that |snk − xk | < k1 . It follows that
1 1 2
|xk − s| ≤ |xk − snk | + |snk − s| < + =
k k k
for k ∈ N, so that s = limk→∞ xk .

Definition 13.4. A set K ⊂ R is called compact if it is both closed and bounded.

Proposition 13.5. The following are equivalent for K ⊂ R:

(i) K is compact;

(ii) every sequence in K has a convergent subsequence with limit in K.

Proof. (i) =⇒ (ii): Let (xn )∞ ∞

n=1 be a sequence in K. As K is bounded, so is (xn )n=1 . The
Bolzano–Weierstraß Theorem then yields a convergent subsequence (xnk )∞ ∞
k=1 of (xn )n=1 .
As K is closed, limk→∞ xnk ∈ K holds.
(ii) =⇒ (i): Assume that (ii) holds, but that K is not compact. That leaves two
(possibly overlapping) possibilities: K is not closed or K is not bounded.
Case 1: K is not closed. Then there is a convergent sequence (xn )∞ n=1 in K such
that limn→∞ xn ∈ ∞ ∞
/ K. If (xnk )k=1 is any subsequence of (xn )n=1 , then limk→∞ xnk =
limn→∞ xn ∈ / K, so that (ii) is violated.
Case 2: K is not bounded. In this case there is, for each n ∈ N, some xn ∈ K with
|xn | ≥ n. The sequence (xn )∞
n=1 has no bounded and therefore no convergent subsequence.
This contradicts (ii).

75
Exercises
1. Let S ⊂ R. Show that

S = {x0 ∈ R : for each  > 0, there is x ∈ S with |x − x0 | < }.

2. Show the following:

(a) if F1 , . . . , Fm ⊂ R are closed, then F1 ∪ · · · ∪ Fm is closed;

T
(b) if F ⊂ P(R) is such that each F ∈ F is closed, then {F : F ∈ F} is closed.

3. A set U ⊂ R is called open if its complement R \ U is closed. Show that U ⊂ R

is open if and only if, for each x ∈ U , there is  > 0 such that (x − , x + ) ⊂ U .
Conclude that (a, b) is open for a < b.

76
14 Limits of Functions and Continuity
Definition 14.1. Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D.

(a) Suppose that y0 ∈ R is such that limn→∞ f (xn ) = y0 for every sequence (xn )∞
n=1 in
D with limn→∞ xn = x0 . Then we say that f (x) converges to y0 as x tends to x0 and
call y0 the limit of f (x) as x tends to x0 ; we write
x→x
lim f (x) = y0 or f (x) −→0 y0 or f (x) → y0 .
x→x0

(b) Suppose that limn→∞ f (xn ) = ∞ for every sequence (xn )∞n=1 in D with limn→∞ xn =
x0 . Then we say that f (x) diverges to ∞ as x tends to x0 ; we write
x→x
lim f (x) = ∞ or f (x) −→0 ∞ or f (x) → ∞.
x→x0

(c) Suppose that limn→∞ f (xn ) = −∞ for every sequence (xn )∞

n=1 in D with limn→∞ xn =
x0 . Then we say that f (x) diverges to −∞ as x tends to x0 ; we write
x→x
lim f (x) = −∞ or f (x) −→0 −∞ or f (x) → −∞.
x→x0

Example. Define (
0, x ∈ Q,
f : R → R, x 7→
1, x ∈ R \ Q,
and let x0 ∈ R be arbitrary. For every n ∈ N, there are qn ∈ Q and rn ∈ R \ Q such that
1
x0 < qn , rn < x0 + ,
n
so that x0 = limn→∞ qn = limn→∞ rn . As

lim f (qn ) = 0 6= 1 = lim f (rn ),

n→∞ n→∞

the limit limx→x0 f (x) does not exist.

Theorem 14.2. Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D. Then the following are
equivalent for y0 ∈ R:

(i) limx→x0 f (x) = y0 ;

(ii) for every  > 0, there is δ > 0 such that |f (x) − y0 | <  for every x ∈ D with
|x − x0 | < δ.

77
Proof. (i) =⇒ (ii): We proceed indirectly. Assume that (ii) is false. This means that there
is 0 > 0 such that for every δ > 0, there is xδ ∈ D with |xδ −x0 | < δ, but |f (xδ )−y0 | ≥ 0 .
In particular, for each n ∈ N, there is xn ∈ D with |xn − x0 | < n1 and |f (xn ) − y0 | ≥ 0 .
It follows that xn → x0 , but that f (xn ) 6→ y0 . This is a contradiction.
(ii) =⇒ (i): Let (xn )∞
n=1 be a sequence in D such that xn → x0 . Let  > 0, and let
δ > 0 be as specified by (ii). Then there is n ∈ N such that |xn − x0 | < δ for all n ≥ n
and consequently |f (xn ) − y0 | <  for n ≥ n . This means that limn→∞ f (xn ) = y0 .

Proposition 14.3. Let ∅ 6= D ⊂ R, let f, g : D → R, let x0 ∈ D, and suppose that

limx→x0 f (x) and limx→x0 g(x) exist in R. Then the limits limx→x0 (f (x) + g(x)) and
limx→x0 f (x)g(x) exist in R, namely

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

x→x0 x→x0 x→x0

and   
lim f (x)g(x) = lim f (x) lim g(x) .
x→x0 x→x0 x→x0

Moreover, if limx→x0 g(x) 6= 0, then there is δ > 0 such that g(x) 6= 0 for all x ∈ D with
|x − x0 | < δ, and we have
f (x) limx→x0 f (x)
lim = .
x→x0 g(x) limx→x0 g(x)
Proof. The first part is clear by the Limit Laws for sequences.
For the “moreover” part, set y0 := limx→x0 g(x), so that |y0 | > 0. Choose δ > 0 such
that |g(x) − y0 | < |y20 | for all x ∈ D with |x − x0 | < δ. It then follows that

|y0 |
|g(x)| = |g(x) − y0 + y0 | ≥ ||g(x) − y0 | − |y0 || > >0
2
for all x ∈ D with |x − x0 | < δ. The formula then follows again from the limit laws.

Just for the record, we define:

Definition 14.4. Let D ⊂ R not be bounded above, and let f : D → R.

(a) Suppose that y0 ∈ R is such that limn→∞ f (xn ) = y0 for every sequence (xn )∞
n=1 in
D with limn→∞ xn = ∞. Then we say that f (x) converges to y0 as x tends to ∞ and
call y0 the limit of f (x) as x tends to ∞; we write
x→∞
lim f (x) = y0 or f (x) −→ y0 or f (x) → y0 .
x→∞

(b) Suppose that limn→∞ f (xn ) = ∞ for every sequence (xn )∞

n=1 in D with limn→∞ xn =
∞. Then we say that f (x) diverges to ∞ as x tends to ∞; we write
x→∞
lim f (x) = ∞ or f (x) −→ ∞ or f (x) → ∞.
x→∞

78
(c) Suppose that limn→∞ f (xn ) = −∞ for every sequence (xn )∞
n=1 in D with limn→∞ xn =
∞. Then we say that f (x) diverges to −∞ as x tends to ∞; we write
x→∞
lim f (x) = −∞ or f (x) −→ −∞ or f (x) → −∞.
x→∞

It is obvious how the analogous definition with D not bounded below and x → −∞
will look like.
Example. Let c0 , c1 , . . . , cn ∈ R with cn 6= 0, and let

p(x) = cn xn + · · · + c1 x + c0 .

We the have 

c0 , if n = 0,
lim p(x) = ∞, if n ≥ 1 and cn > 0,
x→∞ 
−∞, if n ≥ 1 and cn < 0,


and 

 c0 , if n = 0,

∞, if n ≥ 1 is even and cn > 0,




lim p(x) = −∞, if n ≥ 1 is even and cn < 0,
x→−∞ 
∞, if n ≥ 1 is odd and cn < 0,





−∞, if n ≥ 1 is odd and cn > 0.


Definition 14.5. Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D. We say that f is

continuous at x0 if limx→x0 f (x) = f (x0 ).

The following are immediate from Theorem 14.2 and Proposition 14.3:

Corollary 14.6. Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D. Then the following are
equivalent:

(i) f is continuous at x0 ;

(ii) for every  > 0, there is δ > 0 such that |f (x) − f (x0 )| <  for every x ∈ D with
|x − x0 | < δ.

79
 y

f (x )

f (x 0 ) + ε

f (x 0 )

f (x 0 ) − ε

x0 −δ x0 x0+ δ x

Figure 7: Continuity of f at x0 in terms of  and δ

Corollary 14.7. Let ∅ 6= D ⊂ R, let f, g : D → R, let x0 ∈ D, and suppose f and g are

continuous at x0 . Then f + g and f g are continuous at x0 . Moreover, if g(x0 ) 6= 0, then
there is δ > 0 such that g(x) 6= 0 for all x ∈ D with |x − x0 | < δ, and the restriction of fg
to (x0 − δ, x0 + δ) ∩ D is continuous at x0 .

Definition 14.8. Let ∅ 6= D ⊂ R, and let f : D → R. We say that f is continuous if it

is continuous at each x0 ∈ D.

Examples. 1. Polynomials are continuous on all of R.

2. Rational functions are continuous on their natural domain, i.e., there where the
denominator is non-zero.

Continuity of the Exponential Function

Recall that the exponential function exp : R → R is defined as
∞
X xk
exp(x) :=
k!
k=0

for x ∈ R. Is it continuous?

80
 Let x0 ∈ R, and let (xn )∞
n=1 be such that xn → x0 . It follows from the Limit Laws
that
∞ ∞ ∞
X xkn X xk X xk0
lim exp(xn ) = lim = lim n = = exp(x0 ),
n→∞ n→∞ k! n→∞ k! k!
k=0 k=0 k=0
so that exp is continuous at x0 . What is problematic with this argument?
The problem is that the summation symbol in the definition of exp is not a sum, but
a limit of sums, so that the above chain of equalities becomes, in fact,
m
X xk n
lim exp(xn ) = lim lim
n→∞ n→∞ m→∞ k!
k=0
m ∞ m
X xk n
X xk X xk
= lim lim = lim lim n = lim 0
= exp(x0 ).
m→∞ n→∞ k! m→∞ n→∞ k! m→∞ k!
k=0 k=0 k=0

The problem lies with the second equality: we are interchanging the order of limn→∞ and
limm→∞ . Interchanging the order of limits, however, can be treacherous:
1 m 1 m
   
lim lim 1 − = 0 6= 1 = lim lim 1 − .
n→∞ m→∞ n m→∞ n→∞ n
To prove that exp is indeed continuous, we therefore need to do more work.
Example. We will show that exp is continuous. First, we will see that exp is continuous
at 0. Let (xn )∞
n=1 be a sequence in R such that limn→∞ xn = 0. We need do show that
limn→∞ exp(xn ) = 1. Let  > 0, and let C ≥ 0 be such that |xn | ≤ C for n ∈ N. As
P∞ C k Pm Ck 
k=0 k! converges, there is m ∈ N such that k=m +1 k! < 2 for all m > m and,
P∞ k
consequently, k=m +1 Ck! ≤ 2 . Define
m
X xk
p(x) :=
k!
k=0

for x ∈ R. Then p is a polynomial—and therefore continuous—with p(0) = 1. It follows

that there is n ∈ N such that |p(xn ) − 1| < 2 for all n ≥ n . We therefore obtain for
n ≥ n that
∞
X xk n
| exp(xn ) − 1| = −1
k!
k=0
∞
X xkn
≤ |p(xn ) − 1| +
k!
k=m +1
∞
 X Ck
< +
2 k!
k=m +1
 
≤ +
2 2
= .

81
This proves the continuity of exp at 0. Let x0 ∈ R be arbitrary, and let (xn )∞ n=1 be a
sequence in R such that xn → x0 . It follows that limn→∞ (xn − x0 ) = 0, so that

lim exp(xn ) = lim exp((xn − x0 ) + x0 ) = lim exp(xn − x0 ) exp(x0 ) = exp(x0 ).

n→∞ n→∞ n→∞

Hence, exp is continuous.

Exercises
1. Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D. Show that the following are
equivalent:

(i) limx→x0 f (x) = ∞;

(ii) for every R ∈ R, there is δ > 0 such that f (x) > R for all x ∈ D with
|x − x0 | < δ.

2. Let ∅ 6= D ⊂ R be not bounded above, and let f : D → R. Show that the following
are equivalent for y0 ∈ R:

(i) limx→∞ f (x) = y0 ;

(ii) for every  > 0, there is R > 0 such that |f (x) − y0 | <  for all x ∈ D with
x > R.

3. Define f : R → R by letting

 0,
 x∈/ Q,
1 p
f (x) := q, x = q 6= 0 with p ∈ Z and q ∈ N coprime,

1, x = 0.


Show that f is discontinuous at every x ∈ Q, but continuous at every x ∈ R \ Q.

82
15 Properties of Continuous Functions and Uniform Con-
tinuity
Let ∅ 6= S be any set, and let f : S → R be any function. We call f bounded if its range
f (S) is bounded.

Theorem 15.1. Let ∅ 6= K ⊂ R be a compact set, and let f : K → R be continuous.

Then f is bounded, and there are xmin , xmax ∈ K such that

f (xmin ) = inf f (K) and f (xmax ) = sup f (K).

Proof. Assume towards a contradiction that f is not bounded, i.e., for each n ∈ N, there
is xn ∈ K such that |f (xn )| ≥ n. As K is compact, there is a subsequence (xnk )∞ k=1 of
∞
(xn )n=1 converging to some x0 ∈ K. As f is continuous, this means that limk→∞ f (xnk ) =
f (x0 ), which means that (f (xnk ))∞
k=1 is bounded. This contradicts the fact that

|f (xnk )| ≥ nk ≥ k

for all k ∈ N.
For any n ∈ N, there is xn ∈ K such that
1
f (xn ) > sup f (K) − .
n
The sequence (xn )∞ ∞
n=1 has a convergent subsequence (xnk )k=1 with limk→∞ xnk =: xmax ∈
K. The continuity of f yields that
 
1
f (xmax ) = lim f (xnk ) ≥ lim sup f (K) − = sup f (K) ≥ f (xmax ).
k→∞ k→∞ nk

In a similar vein, the existence of xmin ∈ K is proven.

Theorem 15.2 (Intermediate Value Theorem). Let a < b, and let f : [a, b] → R be
continuous. Then, for any c between f (a) and f (b), there is xc ∈ [a, b] such that f (xc ) = c.

83
 y

f (x)

f (a)
c
f (b)

a xc b x

Figure 8: Intermediate Value Theorem

Proof. Without loss of generality, we can focus on the case where f (a) < c < f (b). (If
f (a) = c or if f (b) = c the case is clear, and if f (a) > c > f (b), we just replace f by −f
and c by −c.)
Let
S := {x ∈ [a, b] : f (x) ≤ c}.

Then S is a non-empty, bounded subset of R and therefore has a supremum xc . We claim

that f (xc ) = c. Let (xn )∞
n=1 be a sequence in S such that limn→∞ xn = xc . The continuity
of f yields that f (xc ) = limn→∞ f (xn ) ≤ c. As f (b) > c, it is clear that xc < b. Therefore
there is n0 ∈ N such that xc + n1 < b for n ≥ n0 . By the definition of xc , it is clear that
 
1
f xc + >c
n
1
for all n ≥ n0 . As xc + n → xc , the continuity of f again yields that
 
1
f (xc ) = lim f xc + ≥ c.
n→∞ n

This completes the proof.

Example. Let p be a polynomial of odd degree, i.e., there are c0 , c1 , . . . , cν with cν 6= 0

and ν odd such that
p(x) = cν xν + · · · + c1 x + c0 .

84
If cν > 0, then
lim p(x) = ∞ and lim p(x) = −∞.
x→∞ x→−∞
Hence, there are a < 0 such that p(a) < 0 and b > 0 such that p(b) > 0. By the
Intermediate Value Theorem, there is x ∈ (a, b) such that p(x) = 0. If cν < 0, replace p
by −p. In any case, a polynomial of odd degree has a zero in R.
Definition 15.3. Let ∅ 6= D ⊂ R. Then f : D → R is called uniformly continuous if, for
each  > 0, there is δ > 0 such that |f (x) − f (y)| <  for all x, y ∈ D such that |x − y| < δ.
By Corollary 14.6, it is clear that every uniformly continuous function is continuous.
But does the converse hold?
Examples. 1. The function
1
f : (0, ∞) → R, x 7→
x
is clearly continuous. Let  := 1, and assume that f is uniformly continuous. Let
δ > 0 be as in the definition of uniform continuity. As limn→∞ n1 = 0, there is nδ ∈ N
1 1
such that n+1 − n < δ for n ≥ nδ . However, we have
   
1 1
f −f = |(n + 1) − n| = 1 ≥ 
n+1 n
for all n ≥ nδ . Therefore, f is not uniformly continuous.

2. The function
f : [0, ∞) → R, x 7→ x2
is continuous. Assume that f is uniformly continuous, and let  := 1. Then there is
δ > 0 such that |f (x) − f (y)| < 1 for all x, y ≥ 0 with |x − y| < δ. Choose, x := 2δ
and y := 2δ + 2δ . It follows that |x − y| = 2δ < δ. However, we have

|f (x) − f (y)| = |x − y|(x + y)

 
δ 2 2 δ
= + +
2 δ δ 2
δ4
≥
2δ
= 2.
Therefore, f is not uniformly continuous.

3. The function
f : [0, 1] → R, x 7→ x2
is uniformly continuous continuous. Let  > 0, and note that

|f (x) − f (y)| = |x − y|(x + y) ≤ 2|x − y|

for x, y ∈ [0, 1]. Set δ := 2 .

85
 That last example is no coincidence:

Theorem 15.4. Let ∅ 6= K ⊂ R be compact, and let f : K → R be continuous. Then f

is uniformly continuous.

Proof. Assume that f is not uniformly continuous. This means that there is 0 > 0 such
that, for each δ > 0, there are xδ , yδ ∈ K with |xδ − yδ | < δ such that |f (xδ ) − f (yδ )| ≥
0 . In particular, for each n ∈ N, there are xn , yn ∈ K such that |xn − yn | < n1 and
|f (xn ) − f (yn )| ≥ 0 . As K is compact, (xn )∞ ∞
n=1 has a subsequence (xnk )k=1 converging to
some x0 ∈ K. As clearly limn→∞ (xn − yn ) = 0, it follows that x0 = limk→∞ ynk as well.
The continuity of f yields

lim f (xnk ) = f (x0 ) = lim f (ynk ),

k→∞ k→∞

which contradicts that |f (xn ) − f (yn )| ≥ 0 for all n ∈ N.

Exercises
1. Let ∅ 6= K ⊂ R be compact, and let f : K → R be continuous. Show that f (K) is
compact.

2. Show that the following are equivalent for a set ∅ 6= K ⊂ R:

(i) K is compact;
(ii) every continuous function f : K → R is bounded.

3. Let f : [0, 1] → R be continuous such that f ([0, 1]) ⊂ [0, 1]. Show that f has a fixed
point, i.e., there is x0 ∈ [0, 1] such that f (x0 ) = x0 . (Hint: Apply the Intermediate
Value Theorem to the function [0, 1] 3 x 7→ x − f (x).)

4. Let a < b, and let f, g : [a, b] → R be continuous such that f (a) ≤ g(a) and
f (b) ≥ g(b). Show that there is x0 ∈ [a, b] such that f (x0 ) = g(x0 ).

5. Let f : [0, 2] → R be continuous such that f (0) = f (2). Show that there are
x, y ∈ [0, 2] with |x − y| = 1 and f (x) = f (y). (Hint: Apply the Intermediate Value
Theorem to the auxiliary function [0, 1] 3 x 7→ f (x + 1) − f (x).)

6. Show that there is no continuous function f : [0, 1] → R that attains each of its
values exactly twice. (Hint: Assume that there is such a function, and play around
with the Intermediate Value Theorem.)

7. Let ∅ 6= D ⊂ R, let f : D → R be uniformly continuous, and let (xn )∞

n=1 be a
∞
Cauchy sequence in D. Show that (f (xn ))n=1 is a Cauchy sequence.

86
16 Continuity of Inverse Functions
Definition 16.1. Let ∅ 6= D ⊂ R, and let f : D → R. We call f :

(a) increasing if f (x1 ) ≤ f (x2 ) for all x1 , x2 ∈ D with x1 < x2 ;

(b) strictly increasing if f (x1 ) < f (x2 ) for all x1 , x2 ∈ D with x1 < x2 ;

(c) decreasing if f (x1 ) ≥ f (x2 ) for all x1 , x2 ∈ D with x1 < x2 ;

(d) strictly decreasing if f (x1 ) > f (x2 ) for all x1 , x2 ∈ D with x1 < x2 .

If f is increasing or decreasing, we call it monotonic; if it is strictly increasing or strictly

decreasing, we call it strictly monotonic.

Clearly, every strictly monotonic function is injective.

Proposition 16.2. Let a < b, and let f : [a, b] → R be continuous and injective. Then f
is strictly monotonic.

Proof. As f is injective, f (a) 6= f (b) must hold, i.e., f (a) < f (b) or f (a) > f (b). We can
suppose without loss of generality that f (a) < f (b).
We first claim that

f (a) = min f ([a, b]) and f (b) = max f ([a, b]).

To see that f (a) = min f ([a, b]), assume that there is x0 ∈ (a, b] such that f (x0 ) < f (a).
As f (x0 ) < f (a) < f (b), the Intermediate Value Theorem yields c ∈ (x0 , b) such that
f (c) = f (a). This violates the injectivity of f . Similarly, the claim for f (b) is proven.
Assume towards a contradiction that f is not increasing, i.e., there are x1 , x2 ∈ [a, b]
with x1 < x2 such that f (x1 ) ≥ f (x2 ). By the injectivity of f , the case f (x1 ) = f (x2 )
cannot occur, so that f (x1 ) > f (x2 ). It is clear that both x1 = a and x2 = b cannot
happen. This means that x1 6= a or x2 6= b.
Suppose that x1 6= a, i.e., a < x1 . Choose c such that f (x2 ) < c < f (x1 ). By the
Intermediate Value Theorem, there is x3 ∈ (x1 , x2 ) such that f (x3 ) = c. As

f (a) < f (x2 ) < c < f (x1 ),

the Intermediate Value Theorem also yields x4 ∈ (a, x1 ) with f (x4 ) = c. This contradicts
the injectivity of f .
Similarly, we deal with the case where x2 6= b.

Theorem 16.3. Let a < b, and let f : [a, b] → R be continuous and injective. Then f
maps [a, b] onto [α, β]—with α = f (a) and β = f (b) if f is strictly increasing and α = f (b)
and β = f (a) if f is strictly decreasing—, and the inverse function f −1 : [α, β] → [a, b] is
also continuous.

87
Proof. The part about f mapping [a, b] onto [α, β] with α and β as stated is clear by
Proposition 16.2. What remains to be shown is the continuity of f −1 .
Let y0 ∈ [α, β], and let (yn )∞n=1 be a sequence in [α, β] with y0 = limn→∞ yn . We
need to show that limn→∞ f (yn ) = f −1 (y0 ). We assume that this is not the case. This
−1

means that there are 0 > 0 and n1 < n2 < n3 < · · · such that |f −1 (ynk ) − f −1 (y0 )| ≥ 0
for all k ∈ N. As the sequence (f −1 (ynk ))∞
k=1 is bounded, it has a convergent subsequence
by the Bolzano–Weierstraß Theorem. We may replace (f −1 (ynk ))∞ k=1 by this subsequence
−1 ∞
and suppose that (f (ynk ))k=1 already converges. (This spares us atrocious notation like
ynkl .) Set x0 := limk→∞ f −1 (ynk ). As f is continuous, we have

y0 = lim ynk = lim f (f −1 (ynk )) = f (x0 )

k→∞ k→∞

and therefore f −1 (y0 ) = x0 , so that |f −1 (ynk ) − x0 | ≥ 0 for all k ∈ N. This is a

contradiction.

Examples. 1. Let n ∈ N. The function

f : [0, ∞) → [0, ∞), x 7→ xn

is strictly increasing and maps [0, ∞) onto [0, ∞). For any 0 ≤ a < b, the restriction
of f to [a, b] has a continuous inverse by Theorem 16.3, namely
√
[an , bn ] → [a, b], x 7→ n
x.

This means that the restriction of

√
f −1 : [0, ∞) → [0, ∞), x 7→ n
x

onto [α, β] is continuous for all 0 ≤ α < β. As any convergent sequence in [0, ∞) is
contained in an interval [0, R] with R > 0, this means that f −1 is continuous.

2. We claim that the exponential function exp : R → R is strictly increasing. To see

this, let x > 0, and note that
∞ ∞
X xk X xk
exp(x) = =1+ > 1.
k! k!
k=0 k=1
| {z }
>0

For x1 < x2 , we therefore obtain

exp(x2 ) = exp(x1 + x2 − x1 ) = exp(x1 ) exp(x2 − x1 ) > exp(x1 ).

| {z }
>0
| {z }
>1

88
 Next, we claim that

lim exp(x) = ∞ and lim exp(x) = 0.

x→∞ x→−∞

As
1
exp(−x) =
exp(x)
for all x ∈ R, it is enough to prove that limx→∞ exp(x) = ∞. Recall that

exp(n) = en

for n ∈ N where
∞ ∞
X 1 X 1
e = exp(1) = =2+ > 2.
k! k!
k=0 k=2

Let (xn )∞be a sequence in R with limn→∞ xn = ∞, and note that limn→∞ bxn c =
n=1
∞ as well. We obtain

exp(xn ) ≥ exp(bxn c) = ebxn c > 2bxn c

for n ∈ N, so that limn→∞ exp(xn ) = ∞.

Let c ∈ (0, ∞). As limx→−∞ exp(x) = 0, there is a < 0 with exp(a) < c, and since
limx→∞ exp(x) = ∞, there is b > 0 with exp(b) > c. The continuity of exp and the
Intermediate Value Theorem yield xc ∈ (a, b) with exp(xc ) = c. This means that
exp(R) = (0, ∞).
By Theorem 16.3, the restriction of exp to [a, b] has a continuous inverse for every
a < b, namely
[exp(a), exp(b)] → [a, b], x 7→ log x.

As in the previous example, the continuity of

log : (0, ∞) → R, x 7→ log x

follows.
Just for the record, we define now ax for a > 0 and any x ∈ R.

Definition 16.4. Let a > 0. Define

expa : R → R, x 7→ exp(x log a).

Properties. For any a > 0:

1. expa is continuous;

2. expa (x + y) = expa (x) expa (y) for all x, y ∈ R;

89
 3. expa (q) = aq for all q ∈ Q.

The third property motivates to define

ax = expa (x)

for all a > 0 and x ∈ R.

90
17 Differentiation
Definition 17.1. Let ∅ 6= D ⊂ R, and let f : D → R. Then f is called differentiable at
x0 ∈ D if
f (x) − f (x0 )
lim
x→x0 x − x0
x∈D\{x0 }

exists. This limit is called the (first) derivative of f at x0 . If f is differentiable at each

point of D, we call it differentiable on D.

Notation. Symbols for the first derivative of f at x0 are

df df
f 0 (x0 ) or (x0 ) or
dx dx x=x0

Remarks. 1. It is tacitly understood throughout that, whenever we speak of differen-

tiability at a point x0 ∈ D that x0 ∈ D \ {x0 }, i.e., there is at least one sequence in
D \ {x0 } converging to x0 .

2. Clearly, f is differentiable at x0 if and only if

f (x0 + h) − f (x0 )
lim
h→0 h
h6=0
x0 +h∈D

exists.
Examples. 1. Let n ∈ N, and let

f : R → R, x 7→ xn .

Let x0 , h ∈ R be such that h 6= 0. The Binomial Theorem yields

n  
X n k n−k
n
f (x0 + h) = (x0 + h) = x h ,
k 0
k=0

so that
n−1
X 
n k n−k
f (x0 + h) − f (x0 ) = x h
k 0
k=0

and therefore
n−1   n−2  
f (x0 + h) − f (x0 ) X n k n−1−k X n k n−1−k
= x h = x h + nxn−1 .
h k 0 k 0 0
k=0 k=0
Pn−2 n
xk0 hn−1−k = 0, this means that f is differentiable at x0 with


As limh→0 k=0 k
f 0 (x0 ) = nxn−1
0 .

91
 2. We claim that the exponential function is differentiable at 0 and that

d exp
= 1.
dx x=0

For h 6= 0, note that

∞ ∞
!
exp(h) − 1 1 X hk X hk−1
= −1 = .
h h k! k!
k=0 k=1

Let (hn )∞
n=1 be a sequence in R \ {0} converging to zero. Let C ≥ 0 be such that
|hn | ≤ C for all n ∈ N. Let  > 0. It follows from the Limit Ratio Test that
P∞ C k−1
< ∞; by the Cauchy Criterion, there is therefore m ∈ N such that
Pk=1
∞
k!
C k−1 
k=m +1 k! ≤ 2 . Set
m
X xk−1
p(x) := .
k!
k=1

Then p is a polynomial—and thus continuous—and p(0) = 1. Choose n ∈ N such

that |p(hn ) − 1| < 2 for all n ≥ n . For those n, we have
∞
exp(hn ) − 1 X hk−1
n
−1 = −1
hn k!
k=1
m ∞
X hk−1
n
X |hn |k−1
≤ −1 +
k! k!
k=1 k=m +1
∞
X C k−1
≤ |p(hn ) − 1| +
k!
k=m +1
 
< +
2 2
= .

exp(h)−1 d exp
This proves that limh→0 h = 1, i.e., dx x=0 = 1 as claimed. Let x0 , h ∈ R
h6=0
be such that h 6= 0. As

exp(x0 + h) − exp(x0 ) exp(x0 ) exp(h) − exp(x0 ) exp(h) − 1

= = exp(x0 ) ,
h h h
it follows that exp is differentiable at x0 such that

d exp
= exp(x0 ).
dx x=x0

Proposition 17.2. Let ∅ 6= D ⊂ R, let x0 ∈ D, and let f : D → R be differentiable at

x0 . Then f is continuous at x0 .

92
Proof. Let (xn )∞
n=1 be a sequence in D with x0 = limn→∞ xn . We distinguish two cases.
Case 1: There is n0 ∈ N such that xn = x0 for all n ≥ n0 . Then it is obvious that
limn→∞ f (xn ) = f (x0 ).
Case 2: There are n1 < n2 < n3 < · · · with xnk 6= x0 for all k ∈ N. We can suppose
that
{n1 , n2 , n3 , . . .} = {n ∈ N : xn 6= x0 }.
We have
f (xnk ) − f (x0 )
lim (f (xnk ) − f (x0 )) = lim (xnk − x0 )
k→∞ k→∞ xnk − x0
f (xnk ) − f (x0 )
= lim lim (xnk − x0 ) = f 0 (x0 )0 = 0.
k→∞ xnk − x0 k→∞

Let  > 0, and choose k ∈ N such that |f (xnk ) − f (x0 )| <  for all k ≥ k . Set n := nk ,
and let n ≥ n . Then there are two possibilities: either xn = x0 or xn 6= x0 , i.e.,
n ∈ {n1 , n2 , n3 , . . .}, so that n is of the form nk , necessarily with k ≥ k . In either case,
|f (xn ) − f (x0 ) <  holds. All in all, we have limx→x0 f (x) = f (x0 ).

Proposition 17.3 (Differentiation Laws). Let ∅ 6= D ⊂ R, let x0 ∈ D, and let f, g : D →

R be differentiable at x0 . Then:

(i) f + g is differentiable at x0 with

(f + g)0 (x0 ) = f 0 (x0 ) + g 0 (x0 );

(ii) f g is differentiable at x0 with

(f g)0 (x0 ) = f (x0 )g 0 (x0 ) + f 0 (x0 )g(x0 );

f
(iii) if g(x0 ) 6= 0 there is δ > 0 such that g(x) 6= 0 for all x ∈ D with |x − x0 | < δ and g
is differentiable at x0 with
 0
f f 0 (x0 )g(x0 ) − f (x0 )g 0 (x0 )
(x0 ) = .
g g(x0 )2

Proof. (i) is immediate from the Limit Laws.

(ii): Let h 6= 0 be such that x0 + h ∈ D, and note that

f (x0 + h)g(x0 + h) − f (x0 )g(x0 )

h
f (x0 + h)g(x0 + h) − f (x0 + h)g(x0 ) + f (x0 + h)g(x0 ) − f (x0 )g(x0 )
=
h
g(x0 + h) − g(x0 ) f (x0 + h) − f (x0 )
= f (x0 + h) + g(x0 )
| {z } | h
{z } | h
{z }
h→0
→ f (x0 ) h→0 0 h→0 0
→ g (x 0) → f (x0 )

93
(iii): The existence of δ follows from the continuity of g at x0 . We first treat the case
where f ≡ 1. Let h 6= 0 be such that |h| < δ and x0 + h ∈ D. We then have
1 1
g(x0 +h) − g(x0 ) 1

1 1

= −
h h g(x0 + h) g(x0 )
 
1 g(x0 ) − g(x0 + h)
=
h g(x0 + h)g(x0 )
g(x0 + h) − g(x0 ) 1
=−
h } g(x + h)g(x0 )
| {z | 0 {z }
h→0 0 h→0
→ g (x 0) → g(x0 )2

h→0 g 0 (x0 )
→ − ,
g(x0 )2

which proves the claim in this particular case. For general f , apply (ii) and the special
case:
 0    0
f 0 1 1
(x0 ) = f (x0 ) (x0 ) + f (x0 ) (x0 )
g g g
f 0 (x0 ) f (x0 )g 0 (x0 )
= −
g(x0 ) g(x0 )2
f (x0 )g(x0 ) − f (x0 )g 0 (x0 )
0
= .
g(x0 )2

This completes the proof.

Theorem 17.4 (Chain Rule). Let ∅ 6= Dg ⊂ R and ∅ 6= Df ⊂ R, let g : Dg → R and

f : Df → R be such that g(Dg ) ⊂ Df , and let x0 ∈ Dg be such that g is differentiable at
x0 and f is differentiable at g(x0 ). Then f ◦ g is differentiable at x0 with

(f ◦ g)0 (x0 ) = f 0 (g(x0 ))g 0 (x0 ).

Proof. Define f˜: Df → R by letting

( f (y)−f (g(x ))
0
, if y 6= g(x0 ),
f˜(y) := y−g(x0 )
f 0 (g(x0 )), otherwise.

As f is differentiable in g(x0 ), we have

lim f˜(y) = f 0 (g(x0 )) = f˜(g(x0 ));

y→g(x0 )

moreover,
f (y) − f (g(x0 )) = f˜(y)(y − g(x0 ))

94
holds for all y ∈ Df . It follows that

f (g(x)) − f (g(x0 )) f˜(g(x))(g(x) − g(x0 ))

lim = lim
x→x 0 x − x0 x→x 0 x − x0
x∈Dg \{x0 } x∈Dg \{x0 }
g(x) − g(x0 )
= lim f˜(g(x)) lim
x→x0 x→x0 x − x0
x∈Dg \{x0 } x∈Dg \{x0 }

= f 0 (g(x0 ))g 0 (x0 ),

which proves the claim.

Theorem 17.5 (Differentiability of the Inverse Function). Let a < b, and let f : [a, b] → R
be continuous and strictly monotonic, and let [α, β] := f ([a, b]). Then, if f is differentiable
at x0 ∈ [a, b] with f 0 (x0 ) 6= 0, the inverse function f −1 : [α, β] → [a, b] is differentiable at
f (x0 ) with
df −1 1
=
dy y=f (x0 ) f 0 (x0 )

Proof. Set y0 := f (x0 ), and let y ∈ [α, β] \ {y0 }; set x := f −1 (y). It follows that

f −1 (y) − f −1 (y0 ) x − x0
= .
y − y0 f (x) − f (x0 )

As f −1 is continuous by Theorem 16.3, x tends to x0 as y tends to y0 . This means that

f −1 (y) − f −1 (y0 ) x − x0 1 1
lim = x→x
lim = lim = ,
y→y 0 y − y0 0 f (x) − f (x0 ) x→x0 f (x)−f (x0 ) f 0 (x 0)
y6=y0 x6=x0 x6=x0 x−x0

which proves the theorem.

Remark. With y = f (x), so that x = f −1 (y), Theorem 17.5 takes on the catchy form

dx 1
= dy ,
dy
dx

i.e., the derivative of the inverse is the inverse of the derivative.

Examples. 1. As exp : R → (0, ∞) is differentiable exp−1 = log : (0, ∞) → R is
differentiable at every y0 ∈ (0, ∞) and

d log 1 1 1
= = = .
dy y=y0
d exp exp(log y0 ) y0
dx x=log y
0

2. For r ∈ R, define

fr : (0, ∞) → R, x 7→ xr := expx (r) := exp(r log x).

95
 Then fr is differentiable, and the rules of differentiation and the previous example
yield
r r
fr0 (x) = exp(r log x) = exp(r log x)
x exp(log x)
= r exp(− log x) exp(r log x) = r exp((r − 1) log x) = rfr−1 (x) = rxr−1

for x > 0.

Higher Derivatives
If ∅ 6= D ⊂ R, and let f : D → R be differentiable on D. Suppose that the derivative
f 0 : D → R is again differentiable. Then the derivative of f 0 is called the second derivative
2
of f and denoted by f 00 or ddxf2 . Inductively, one can go on an define, for each n, the n-th
n
derivative f (n) or ddxnf . It is customary, to identify the zeroth derivative f (0) with f . If f 0 ,
f 00 , . . . , and f (n) exist, we call f n-times differentiable; if, furthermore, f (n) is continuous,
we call f n-times continuously differentiable (in the case where n = 1, we simply call f
continuously differentiable).

Exercises
1. Let ∅ 6= D ⊂ R, let n ∈ N0 , and let f, g : D → R be n-times differentiable. Show
that f g is n-times differentiable with
n  
(n)
X n (k)
(f g) (x) = f (x)g (n−k) (x).
k
k=0

for all x ∈ D. (Hint: Review the proof of the Binomial Theorem.)

2. Show that the function

(
x sin x1 , x 6= 0,


f : [0, 1] → R, x 7→
0, x = 0,
is continuous on [0, 1], differentiable on (0, 1], but not differentiable at 0.

3. Show that the function

(
1
x2 sin


x , x 6= 0,
f : [0, 1] → R, x 7→
0, x = 0,
is differentiable on [0, 1] whereas f 0 : [0, 1] → R is not continuous at 0.

4. Let ∅ 6= D ⊂ R. Then f : D → R is called Lipschitz continuous if there is C ≥ 0

such that
|f (x) − f (y)| ≤ C|x − y|
for all x, y ∈ D.

96
(a) Show that every Lipschitz continuous function is uniformly continuous.
(b) Suppose that D = [a, b] with a < b and that f is continuously differentiable on
[a, b]. Show that f is Lipschitz continuous. (Hint: Mean Value Theorem)
(c) Show that
√
f : [0, 1] → R, x 7→ x

is uniformly continuous, but not Lipschitz continuous. (Hint: What do you

1
observe if you assume that f is Lipschitz continuous and choose x = (C+1)2

and y = 0?)

97
18 Local Extrema, the Mean Value Theorem, and Taylor’s
Theorem
Definition 18.1. Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D. Then we say that f
has:

(a) a local maximum at x0 if there is  > 0 with (x0 − , x0 + ) ⊂ D and f (x) ≤ f (x0 )
for all x ∈ (x0 − , x0 + );

(b) a local minimum at x0 if there is  > 0 with (x0 − , x0 + ) ⊂ D and f (x) ≥ f (x0 )
for all x ∈ (x0 − , x0 + ).

If f has a local maximum or a local minimum at x0 , we say that it has a local extremum
at x0 .

Theorem 18.2 (First Derivative Test). Let ∅ 6= D ⊂ R, let f : D → R, and let x0 ∈ D

be such that f has a local extremum at x0 and is differentiable at x0 . Then f 0 (x0 ) = 0.

Proof. We can suppose without loss of generality that f has a local maximum at x0 .
Let  > 0 be as in the definition of a local maximum. Let h ∈ (−, 0). Then it follows
that f (x0 + h) ≤ f (x0 ) and h < 0 and therefore
f (x0 + h) − f (x0 )
≥ 0,
h
so that
f (x0 + h) − f (x0 ) f (x0 + h) − f (x0 )
f 0 (x0 ) = lim = lim ≥ 0.
h→0 h h→0 h
h6=0 h∈(−,0)
x0 +h∈D

On the other hand, we have

f (x0 + h) − f (x0 )
≤0
h
for h ∈ (0, ) and therefore
f (x0 + h) − f (x0 ) f (x0 + h) − f (x0 )
f 0 (x0 ) = lim = lim ≤ 0.
h→0 h h→0 h
h6=0 h∈(0,)
x0 +h∈D

This means that f 0 (x0 ) = 0 as claimed.

Remarks. 1. The existence of  > 0 with (x0 − , x0 + ) ⊂ D in Definition 18.1 is

crucial for Theorem 18.2 to hold: the function

f : [0, 1] → R, x 7→ x

attains its minimum at 0 and its maximum at 1, but f 0 (0) = f 0 (1) = 1.

98
 2. If f : D → R is differentiable at x0 such that f 0 (x0 ) = 0, then f need not have a
local extremum there: consider

f : R → R, x 7→ x3

at x0 = 0.

Lemma 18.3 (Rolle’s Theorem). Let a < b, let f : [a, b] → R be continuous and dif-
ferentiable on (a, b), and suppose that f (a) = f (b). Then there is ξ ∈ (a, b) such that
f 0 (ξ) = 0.

f (x )

a b x

Figure 9: Rolle’s Theorem

Proof. If f is constant, then the claim is trivial.

We can therefore suppose that f is not constant, i.e., there is x0 ∈ (a, b) with f (x0 ) >
f (a) = f (b) or f (x0 ) < f (a) = f (b). In any case, f attains its maximum or minimum
on [a, b]—which it must attain because f is continuous and [a, b] is compact—at point
ξ ∈ (a, b). Then f has a local extremum at ξ, so that f 0 (ξ) = 0 by Theorem 18.2.

Theorem 18.4 (Mean Value Theorem). Let a < b, and let f : [a, b] → R be continuous
and differentiable on (a, b). Then there is ξ ∈ (a, b) such that

f (b) − f (a)
f 0 (ξ) = .
b−a

99
 y

f (x )

a b x

Figure 10: Mean Value Theorem

Proof. Define
f (b) − f (a)
g : [a, b] → R, x 7→ f (x) − (x − a).
b−a
Then g is continuous, differentiable on (a, b), and satisfies g(a) = f (a) = g(b). By Rolle’s
Theorem, there is ξ ∈ (a, b) with

f (b) − f (a)
0 = g 0 (ξ) = f 0 (ξ) − ,
b−a
which proves the claim.

Corollary 18.5. Let a < b, and let f : [a, b] → R be continuous and differentiable on
(a, b) such that f 0 (x) = 0 for all x ∈ (a, b). Then f is constant.

Proof. Assume otherwise, i.e., there are x, y ∈ [a, b] with f (x) 6= f (y), i.e., f (x)−f
x−y
(y)
6= 0.
By the Mean Value Theorem, however, there is ξ ∈ (min{x, y}, max{x, y}) with

f (x) − f (y)
0 = f 0 (ξ) = .
x−y
This is a contradiction.

Theorem 18.6 (Taylor’s Theorem). Let a < b, let n ∈ N0 , and let f : [a, b] → R be
(n + 1)-times differentiable on [a, b]. Then, for any x, x0 ∈ [a, b] with x 6= x0 , there is

100
ξ ∈ (min{x, x0 }, max{x, x0 }) such that
n
X f (k) (x0 ) f (n+1) (ξ)
f (x) = (x − x0 )k + (x − x0 )n+1 .
k! (n + 1)!
k=0

Proof. Let x, x0 ∈ [a, b] be such that x 6= x0 ; without loss of generality, suppose that
x < x0 . Set
n
!
(n + 1)! X f (k) (x0 )
y := f (x) − (x − x0 )k ,
(x − x0 )n+1 k!
k=0

so that
n
X f (k) (x0 ) y
f (x) = (x − x0 )k + (x − x0 )n+1 .
k! (n + 1)!
k=0

We will show that there is ξ ∈ (x, x0 ) such that f (n+1) (ξ) = y.

Define
n
X f (k) (t) y
g : [a, b] → R, t 7→ f (x) − (x − t)k − (x − t)n+1 ,
k! (n + 1)!
k=0

so that g(x0 ) = g(x) = 0. By Rolle’s Theorem, there is ξ ∈ (x, x0 ) such that g 0 (ξ) = 0.
Note that
n
!
X f (k+1) (t) f (k) (t) y
g 0 (t) = −f 0 (t) − (x − t)k − (x − t)k−1 + (x − t)n
k! (k − 1)! n!
k=1
f (n+1) (t) y
=− (x − t)n + (x − t)n ,
n! n!
so that
f (n+1) (ξ) y
0=− (x − ξ)n + (x − ξ)n
n! n!
and thus y = f (n+1) (ξ).

Remark. For n = 0, Taylor’s Theorem is just the Mean Value Theorem (up to a slightly
stronger differentiability hypothesis).
Taylor’s Theorem can be used to derive the Second Derivative Test for local extrema:

Corollary 18.7 (Second Derivative Test). Let a < b, let f : (a, b) → R be twice continu-
ously differentiable, and let x0 ∈ (a, b) be such that f 0 (x0 ) = 0. Then:

(i) if f 00 (x0 ) < 0, then f has a local maximum at x0 ;

(ii) if f 00 (x0 ) > 0, then f has a local minimum at x0 .

101
Proof. We only deal with the case where f 00 (x0 ) < 0.
Since f 00 is continuous and (a, b) is an open interval, there is  > 0 with (x0 −, x0 +) ⊂
(a, b) such that f 00 (x) < 0 for all x ∈ (x0 − , x0 + ). Fix x ∈ (x0 − , x0 + ) \ {x0 }. By
Taylor’s Theorem, there is ξ between x and x0 such that
<0 ≥0
z }| { z }| {
0 f 00 (ξ) (x − x0 )2
f (x) = f (x0 ) + f (x0 )(x − x0 ) + ≤ f (x0 ),
| {z } | 2!
{z }
=0
≤0

which proves the claim.

Exercises
1. Let f : R → R be a differentiable function such that f 0 = f and f (0) = 1. Show
f
that f = exp. (Hint: Differentiate exp .)

2. Let r ∈ R, and let f : (0, ∞) → R be a differentiable function such that f (1) = 1

and
x f 0 (x) = r f (x)

all x > 0. Show that

f (x) = xr
f (x)
for x > 0. (Hint: Consider the function (0, ∞) 3 x 7→ xr and differentiate it.)

3. Let f, g : R → R be differentiable functions such that

f (0) = 0, g(0) = 1, f 0 = g, and g 0 = −f.

Show that
f (x) = sin x and g(x) = cos x

for x ∈ R. (Hint: Consider the function

R → R, x 7→ (f (x) − sin x)2 + (g(x) − cos x)2 ,

and differentiate it.)

4. Let a < b, and let f, g : [a, b] → R be continuous and differentiable on (a, b). Show
that there is ξ ∈ (a, b) such that

f 0 (ξ)(g(b) − g(a)) = g 0 (ξ)(f (b) − f (a)).

(Hint: Apply Rolle’s Theorem to a suitable auxiliary function.)

102
5. Let f : R → R be such that there is C ≥ 0 with

|f (x) − f (y)| ≤ C(x − y)2

for all x, y ∈ R. Show that f is constant.

6. Let a < b, and let f : [a, b] → R be continuous and differentiable on (a, b). Show
that:

(a) if f 0 (x) ≥ 0 for all x ∈ (a, b), then f is increasing, and if f 0 (x) > 0 for all
x ∈ (a, b), then f is strictly increasing;
(b) if f is increasing, then f 0 (x) ≥ 0 for all x ∈ (a, b).

Give an example of a strictly increasing function f that is differentiable on (a, b)

such that there is ξ ∈ (a, b) with f 0 (ξ) = 0. (Hint for (a): Mean Value Theorem.)

7. Let n ∈ N, and let f : R → R be n-times differentiable such that f (n) ≡ 0. Show

that f is a polynomial of degree at most n − 1. (Hint: Taylor’s Theorem.)

8. Let f : [a, b] → R be differentiable, and let c ∈ R be such that f 0 (a) ≤ c ≤ f 0 (b).

Show that there is ξ ∈ [a, b] such that f 0 (ξ) = c.
Hint: First, consider the case where c = 0.
Warning: You must not suppose that f 0 is continuous!

103
19 The Riemann Integral
Intuitively, the Riemann integral of a (non-negative) function over an interval can be
thought of as the area between the graph of the function and the x-axis. We want to
make this notion precise.
For constant functions, it is straightforward how one would define a Riemann integral:

Definition 19.1. Let a < b, and let f : [a, b] → R be constant, i.e., f (x) = c for all
x ∈ [a, b] and fixed c ∈ R. Then the Riemann Integral of f from a to b is defined as
Z b
f (x) dx := c(b − a).
a

Of course, we want to integrate more functions.

Given a < b, we call a finite number of points with

a = x0 < x1 < · · · < xn = b

a partition of [a, b], and we call a function f : [a, b] → R a step function if there is a partition
a = x0 < x1 < · · · < xn = b such that f is constant on (xj−1 , xj ) for j = 1, . . . , n.

a x1 x2 x3 b x

Figure 11: Graph of a step function

104
Definition 19.2 (Riemann Integral of Step Functions). Let a < b, and let f : [a, b] → R
be a step function defined with respect to the partition a = x0 < x1 < · · · < xn = b. Then
the Riemann Integral of f from a to b is defined as
Z b n
X
f (x) dx := f (ξj )(xj − xj−1 )
a j=1

with ξj ∈ (xj−1 , xj ) for j = 1, . . . , n.

a x1 x2 x3 b x

Figure 12: Integral of a step function

Question. Let a = y0 < y1 < · · · < ym = b be another partition such that f is constant
on (yj−1 , yj ) for j = 1, . . . , m. This would then yield
Z b m
X
f (x) dx = f (ηj )(yj − yj−1 )
a j=1

with ηj ∈ (yj−1 , yj ) for j = 1, . . . , n. Does this give the same value?

The answer is “yes” (fortunately. . . ): Let a = z0 < z1 < · · · < zp = b be a common

refinement of a = x0 < x1 < · · · < xn = b and a = y0 < y1 < · · · < ym = b, i.e.,

{z0 , z1 , . . . , zp } ⊃ {x0 , x1 , · · · , xn } ∪ {y0 , y1 , . . . , ym }.

105
 a x 1 y1 x2 y3 x3 x4 x5 b

= = =

=
=

=
= =

= =
x0 z1 z2 y2 z4 z5 z6 y4 x6

=
y0 z3 z7 y5

=
z0 z8

Figure 13: Common refinement {z0 , . . . , z8 } of {x0 , . . . , x6 } and {y0 , . . . , y5 }

It follows that f is constant on (zj−1 , zj ) for j = 1, . . . , p. For each j = 1, . . . , p, choose

ζj ∈ (zj−1 , zj ). We then obtain
n
X n
X X
f (ξj )(xj − xj−1 ) = f (ξj ) (zν − zν−1 )
j=1 j=1 (zν−1 ,zν )⊂(xj−1 ,xj )
n
X X
= f (ζν )(zν − zν−1 )
j=1 (zν−1 ,zν )⊂(xj−1 ,xj )
p
X
= f (ζν )(zν − zν−1 )
ν=1
m
X X
= f (ζν )(zν − zν−1 )
j=1 (zν−1 ,zν )⊂(yj−1 ,yj )
m
X X
= f (ηj ) (zν − zν−1 )
j=1 (zν−1 ,zν )⊂(yj−1 ,yj )
m
X
= f (ηj )(yj − yj−1 ).
j=1

Remark. If we define a step function f : [a, b] → R with respect to a partition a = x0 <

x1 < · · · < xn = b, we do not make any requirements for the values of f at the partition
points x0 , x1 , . . . , xn . This means, in particular, that if we change a step function’s values
at only finitely many points, the value of its Riemann integral will not be affected.
We collect the basic properties of the Riemann integral of step functions:

Theorem 19.3. Let a < b, let f, g : [a, b] → R be step functions, and let α, β ∈ R. Then:

(i) αf + βg is a step function such that

Z b Z b Z b
αf (x) + βg(x) dx = α f (x) dx + β g(x) dx;
a a a

(ii) if f ≤ g, i.e., f (x) ≤ g(x) for all x ∈ [a, b], then

Z b Z b
f (x) dx ≤ g(x) dx
a a
holds.

106
Proof. (i): Let a = x0 < x1 < · · · < xn = b be a partition of [a, b] such that f is a step
function with respect to it, and let a = y0 < y1 < · · · < ym = b be a partition of [a, b] such
that g is a step function with respect to it. Replacing both a = x0 < x1 < · · · < xn = b
and a = y0 < y1 < · · · < ym = b with a common refinement if necessary, we can
suppose without loss of generality that f and g are both step functions with respect
to a = x0 < x1 < · · · < xn = b, i.e., both functions are constant on (xj−1 , xj ) as is,
consequently, αf + βg for j = 1, . . . , n, so that αf + βg is a step function as well. Choose
ξj ∈ (xj−1 , xj ) for j = 1, . . . , n, and note that
Z b n
X
αf (x) + βg(x) dx = (αf (ξj ) + βg(ξj ))(xj − xj−1 )
a j=1
Xn n
X
=α f (ξj )(xj − xj−1 ) + β g(ξj )(xj − xj−1 )
j=1 j=1
Z b Z b
=α f (x) dx + β g(x) dx.
a a

(ii) is straightforward.

Definition 19.4. Let a < b, and let f : [a, b] → R be bounded. Then:

(a) the lower integral of f from a to b is defined as

Z b Z b 
f (x) dx := sup φ(x) dx : φ : [a, b] → R is a step function with φ ≤ f ;
a
∗ a

(b) the upper integral of f from a to b is defined as

Z∗ b Z b 
f (x) dx := inf ψ(x) dx : ψ : [a, b] → R is a step function with ψ ≥ f .
a a

Examples. 1. Let a < b, and let f : [a, b] → R be a step function. It follows that the
supremum and the infimum the the definitions of the lower and the upper integral
of f are in fact attained—at φ = ψ = f —, so that

Z b Z b Z∗ b
f (x) dx = f (x) dx = f (x) dx.
a a
∗ a

2. Consider (
1, x ∈
/ Q,
f : [0, 1] → R, x 7→
0, x ∈ Q.

107
 Let φ, ψ : [0, 1] → ∞ be step functions with φ ≤ f ≤ ψ, and let a = x0 < x1 < · · · <
xn = b be such that φ and ψ are constant on (xj−1 , xj ) for j = 1, . . . , n. Choose
ξj ∈ (xj−1 , xj ) ∩ Q and ηj ∈ (xj−1 , xj ) \ Q for j = 1, . . . , n. It follows that
Z b n
X n
X
φ(x) dx = φ(ξj )(xj − xj−1 ) ≤ f (ξj )(xj − xj−1 ) = 0
a j=1 j=1

and Z b n
X n
X
ψ(x) dx = ψ(ηj )(xj − xj−1 ) ≥ f (ηj )(xj − xj−1 ) = 1
a j=1 j=1

and therefore
Z b Z∗ b
f (x) dx = 0 < 1 = f (x) dx.
a
∗ a

(This function is known as the Dirichlet Function.)

Proposition 19.5. Let a < b, let f, g : [a, b] → R be bounded, and let t ∈ R be non-
negative. Then we have

Z∗ b Z∗ b Z∗ b Z∗ b Z∗ b
f (x) + g(x) dx ≤ f (x) dx + g(x) dx and tf (x) dx = t f (x) dx.
a a a a a

Proof. For the first claim, let  > 0. By the definition of the upper integral, there are step
functions ψ1 , ψ2 : [a, b] → R with ψ1 ≥ f and ψ2 ≥ g such that

Z b Z∗ b Z b Z∗ b
 
ψ1 (x) dx < f (x) dx + and ψ2 (x) dx < g(x) dx + .
a a 2 a a 2

It follows that
Z b Z b Z b Z∗ b Z∗ b
ψ1 (x) + ψ2 (x) dx = ψ1 (x) dx + ψ2 (x) dx < f (x) dx + g(x) dx + .
a a a a a

As ψ1 + ψ2 is a step function with ψ1 + ψ2 ≥ f + g, it follows that

Z∗ b Z b Z∗ b Z∗ b
f (x) + g(x) dx ≤ ψ1 (x) + ψ2 (x) dx < f (x) dx + g(x) dx + .
a a a a

Since  > 0 is arbitrary, this proves the claim.

For the second claim, we can suppose without loss of generality that t > 0. Let
ψ : [a, b] → R be a step function with ψ ≥ f . Then tψ is a step function with tψ ≥ tf . It
follows that
Z∗ b Z b Z b
tf (x) dx ≤ tψ(x) dx = t ψ(x) dx
a a a

108
and thus
Z∗ b Z∗ b
tf (x) dx ≤ t f (x) dx.
a a
On the other hand, we have

Z∗ b Z∗ b Z∗ b
1 1
f (x) dx = t f (x) dx ≤ tf (x) dx,
a a t t a

so that multiplication with t yields the reversed inequality.

Corollary 19.6. Let a < b, let f, g : [a, b] → R be bounded, and let t ∈ R be non-negative.
Then we have
Z b Z b Z b Z b Z b
f (x) + g(x) dx ≥ f (x) dx + g(x) dx and tf (x) dx = t f (x) dx.
∗ a ∗ a ∗ a ∗ a ∗ a

Proof. Observe that

Z b Z∗ b
f (x) dx = − − f (x) dx
a
∗ a

(and same for g), and apply the previous proposition.

Definition 19.7 (Riemann Integral). Let a < b. We call f : [a, b] → R Riemann integrable
if it is bounded and satisfies
Z∗ b Z b
f (x) dx = f (x) dx.
a
∗ a

In this case, we define

Z b Z∗ b
f (x) dx := f (x) dx
a a
as the Riemann integral of f from a to b.

Examples. 1. Step functions are Riemann integrable.

2. The Dirichlet Function is not Riemann integrable.

Proposition 19.8. Let a < b. Then f : [a, b] → R is Riemann integrable if and only if,
for each  > 0, there are step functions φ, ψ : [a, b] → R with φ ≤ f ≤ ψ such that
Z b Z b
ψ(x) dx − φ(x) dx < .
a a

109
Proof. Suppose that f is Riemann integrable, and let  > 0. Then there are φ, ψ : [a, b] →
R with φ ≤ f ≤ ψ such that
Z b Z b Z b Z b
 
φ(x) dx > f (x) dx − and ψ(x) dx < f (x) dx + .
a a 2 a a 2
Multiplying the first inequality with −1 and adding it to the second yields the claim.
Conversely, let  > 0, and let φ, ψ : [a, b] → R be step functions with φ ≤ f ≤ ψ and
Z b Z b
ψ(x) dx − φ(x) dx < .
a a

As φ and ψ are bounded, so is f , and we have

Z b Z b
f (x) dx ≥ φ(x) dx
a
∗ a
Z b Z b Z b Z∗ b
= ψ(x) dx − ψ(x) − φ(x) dx ≥ ψ(x) dx −  ≥ f (x) dx − .
a a a a

As  > 0 is arbitrary, this means that

Z b Z∗ b
f (x) dx ≥ f (x) dx.
a
∗ a

The reversed inequality holds trivially. This completes the proof.

Examples. 1. Let a < b, and let f : [a, b] → R be continuous. Then f is Riemann

integrable. To see this, let  > 0. As f is uniformly continuous, there is δ > 0 such

that |f (x) − f (y)| < b−a for all x, y ∈ [a, b] with |x − y| < δ. Choose a partition
a = x0 < x1 < · · · < xn = b such that xj − xj−1 < δ for j = 1, . . . , n. Set

mj := min f ([xj−1 , xj ]) and Mj := max f ([xj−1 , xj ])

for j = 1, . . . , n, and define

(
f (xj ), if x = xj with j ∈ {0, 1, . . . , n},
φ : [a, b] → R, x 7→
mj , if x ∈ (xj−1 , xj ) with j ∈ {1, . . . , n},
and
(
f (xj ), if x = xj with j ∈ {0, 1, . . . , n},
ψ : [a, b] → R, x 7→
Mj , if x ∈ (xj−1 , xj ) with j ∈ {1, . . . , n}.

It is then clear that φ, ψ : [a, b] → R are step functions with φ ≤ f ≤ ψ. For

j = 1, . . . , n, let ξj , ηj ∈ [xj−1 , xj ] be such that f (ξj ) = mj and f (ηj ) = Mj . As
xj − xj−1 < δ, it follows that |ξj − ηj | < δ and, consequently,

Mj − mj = |f (ξj ) − f (ηj )| < .
b−a

110
 We conclude that
Z b Z b n n
X X 
ψ(x) dx − φ(x) dx = (Mj − mj )(xj − xj−1 ) < (xj − xj−1 ) = .
a a b−a
j=1 j=1

2. Let a < b, and let f : [a, b] → R be monotonic. Then f is Riemann integrable. To

see this, suppose first without loss of generality that f is increasing. Let  > 0, and
let n ∈ N. For j = 0, 1, . . . , n, set
b−a
xj := a + j.
n
Define
(
f (xj−1 ), if x ∈ [xj−1 , xj ) with j ∈ {1, . . . , n},
φ : [a, b] → R, x 7→
f (b), x = b,

and
(
f (xj ), if x ∈ [xj−1 , xj ) with j ∈ {1, . . . , n},
ψ : [a, b] → R, x 7→
f (b), x = b.

Then φ, ψ : [a, b] → R are step functions with φ ≤ f ≤ ψ such that

Z b Z b Xn n
X
ψ(x) dx − φ(x) dx = f (xj )(xj − xj−1 ) − f (xj−1 )(xj − xj−1 )
a a j=1 j=1
n
b−a X
= (f (xj ) − f (xj−1 ))
n
j=1
b−a
(f (b) − f (a)).
=
n
Rb Rb
For sufficiently large n, we therefore obtain a ψ(x) dx − a φ(x) dx < .

Theorem 19.9 (Properties of the Riemann Integral). Let a < b, let f, g : [a, b] → R be
Riemann integrable, and let α ∈ R. Then:

(i) f + g is Riemann integrable with

Z b Z b Z b
f (x) + g(x) dx = f (x) dx + g(x) dx;
a a a

(ii) αf is Riemann integrable with

Z b Z b
αf (x) dx = α f (x) dx;
a a

(iii) if f ≤ g, then
Z b Z b
f (x) dx ≤ g(x) dx;
a a

111
(iv) |f | is Riemann integrable with
Z b Z b
f (x) dx ≤ |f (x)| dx;
a a

(v) f g is Riemann integrable.

Proof. (i): From Proposition 19.5 and Corollary 19.6, we conclude that

Z∗ b Z∗ b Z∗ b
f (x) + g(x) dx ≤ f (x) dx + g(x) dx
a a a
Z b Z b Z b
= f (x) dx + g(x) dx ≤ f (x) + g(x) dx,
∗ a ∗ a ∗ a

which yields the claim.

(ii): Suppose first that α ≥ 0. By Proposition 19.5 and Corollary 19.6, we then have
Z∗ b Z b Z b
αf (x) dx = α f (x) dx = α f (x) dx,
a a
∗ a

which proves the claim for non-negative α. Suppose now that α < 0, and observe that
Z b Z∗ b Z∗ b Z b
αf (x) dx = − − αf (x) dx = −(−α) f (x) dx = α f (x) dx.
a a a
∗ a

and, similarly,
Z∗ b Z b
αf (x) dx = α f (x) dx.
a a
This proves the claim for negative α as well.
(iii) is obvious.
(iv): Define f + , f − : [a, b] → R as follows:
( (
+ f (x), f (x) ≥ 0, − −f (x), f (x) ≤ 0,
f (x) := and f (x) :=
0, otherwise, 0, otherwise.
It is clear that
f = f+ − f− and |f | = f + + f − .
We claim that both f + and f − are Riemann integrable. Let  > 0, and let φ, ψ : [a, b] → R
Rb Rb
be step functions such that φ ≤ f ≤ ψ and a ψ(x) dx − a φ(x) dx < . It is then clear
that φ+ and ψ + are step functions with φ+ ≤ f + ≤ ψ + such that
Z b Z b Z b
+ +
ψ (x) dx − φ (x) dx = ψ + (x) − φ+ (x) dx
a a a
Z b Z b
+ + − −
≤ ψ (x) − φ (x) − (ψ (x) − φ (x)) dx ψ(x) − φ(x) dx < .
a | {z } a
≤0

112
It follows that f + is Riemann integrable. Similarly, one sees that f − is Riemann integrable.
It follows that |f | = f + + f − is Riemann integrable and that
Z b
f (x) dx
a
Z b Z b Z b Z b Z b
+ − + −
= f (x) dx − f (x) dx ≤ f (x) dx + f (x) dx = |f (x)| dx.
a a a a a

(v): We first treat the case where g = f , i.e., we show that f 2 is integrable. As
f 2 = |f |2 , we my replace f by |f | and therefore suppose that f ≥ 0. As f is bounded,
we can multiply it with a suitable scalar and also suppose that f ≤ 1. Let  > 0, and
Rb Rb
let φ, ψ : [a, b] → R be step functions with φ ≤ f ≤ ψ and a ψ(x) dx − a φ(x) dx < 2 .
Replacing φ by φ+ and ψ by
(
ψ(x), ψ(x) ≤ 1,
[a, b] → R, x 7→
1, ψ(x) > 1,

we can suppose without loss of generality that 0 ≤ φ ≤ f ≤ ψ ≤ 1. Then φ2 and ψ 2 are

also step functions with φ2 ≤ f 2 ≤ ψ 2 . It follows that
Z b Z b Z b
2 2
ψ(x) dx − φ(x) dx = ψ(x)2 − φ(x)2 dx
a a a
Z b
= (ψ(x) − φ(x))(ψ(x) + φ(x)) dx
a
Z b
≤2 ψ(x) − φ(x) dx
a

<2
2
= ,

so that f 2 is Riemann integrable.

For the general case, just note that
1
f g = ((f + g)2 − (f − g)2 ),
4
which completes the proof.

Corollary 19.10 (Mean Value Theorem of Integration). Let a < b, let f : [a, b] → R be
continuous, and let g : [a, b] → [0, ∞) be Riemann integrable. Then there is ξ ∈ [a, b] such
that Z b
Z b
f (x)g(x) dx = f (ξ) g(x) dx.
a a

113
Proof. Let
m := inf f ([a, b]) and M := sup f ([a, b]),

so that mg ≤ f g ≤ M g and, consequently,

Z b Z b Z b
m g(x) dx ≤ f (x)g(x) dx ≤ M g(x) dx.
a a a

Hence, there is c ∈ [m, M ] with

Z b Z b
f (x)g(x) dx = c g(x) dx.
a a

By the Intermediate Value Theorem, there is ξ ∈ [a, b] with f (ξ) = c.

Exercises
1. Is the function (
x, x ∈ Q,
f : [0, 1] → R, x 7→
0, x ∈
/ Q,
Riemann integrable? If so, evaluate its integral.

2. For n ∈ N, define step functions φn , ψn : [0, 1] → R by letting

(
k−1
, if x ∈ k−1 , nk for some k ∈ {1, . . . , n},


φn (x) = n n
1, x = 1,

and (
k
 k−1
, nk for some k ∈ {1, . . . , n},


n, if x ∈ n
ψn (x) =
1, x = 1.
R1 R1 R1
Compute 0 φn (x) dx and 0 ψn (x) dx and use this to evaluate 0 x dx.

3. Let (
1, x ∈ n1 : n ∈ N ,

f : [0, 1] → R, x 7→
0, otherwise.
and, for n ∈ N, let φn , ψn : [0, 1] → R be defined through
(
0, 0 ≤ x < n1 ,
φn (x) =
f (x), n1 ≤ x ≤ 1,

and (
1, 0 ≤ x < n1 ,
ψn (x) =
f (x), n1 ≤ x ≤ 1.
R1
Show that f is Riemann integrable and that 0 f (x) dx = 0.

114
4. Let a < b, and let f be the function defined Exercise 3 in Section 14. Show that f
Rb
is Riemann integrable on [a, b] with a f (x) dx = 0.

5. Let a < b, and let f : [a, b] → R be a function such that |f | : [a, b] → R is Riemann
integrable. Does this necessarily mean that f is Riemann integrable?
Rb
6. Let a < b, and let f : [a, b] → [0, ∞) be continuous such that a f (x) dx = 0. Show
that f (x) = 0 for all x ∈ [a, b]. (Hint: A sketch might help.)

7. Let a < b, let f : [a, b] → R be Riemann integrable, and let g : [a, b] → R be such
that {x ∈ [a, b] : f (x) 6= g(x)} is finite. Show that g is Riemann integrable and
Rb Rb
a g(x) dx = a f (x) dx. Proceed as follows:

• first suppose, that f is a step function and show that g is also a step function
Rb Rb
and that a g(x) dx = a f (x) dx;
• then use the definition of the Riemann integral to prove the general case.

115
20 Integration and Differentiation
Lemma 20.1. Let a < b < c. Then f : [a, c] → R is Riemann integrable if and only if the
restrictions of f to both [a, b] and [b, c] are Riemann integrable, in which case
Z c Z b Z c
f (x) dx = f (x) dx + f (x) dx
a a b

holds.

Proof. Boring.

For convenience, we agree to define

Z a Z b Z a
f (x) dx = 0 and f (x) dx = − f (x) dx
a a b

if a > b. With these conventions, the formula of Lemma 20.1 remains valid for any choice
of a, b, c ∈ [min{a, b, c}, max{a, b, c}].
In what follows, we use the term interval for any set of one of the following forms:

• (a, b), [a, b], (a, b], or [a, b) with a < b;

• (a, ∞), [a, ∞), (−∞, a), or (−∞, a] with a ∈ R;

• R.

Theorem 20.2. Let I ⊂ R be an interval, let f : I → R be continuous, and let x0 ∈ I.

Then Z x
F : I → R, x 7→ f (t) dt
x0
is an antiderivative of f , i.e., F is differentiable such that F 0 = f . Moreover, if G : I → R
is any antiderivative of f , then F − G is constant.

Proof. Let x ∈ I, and let (hn )∞

n=1 be a sequence in R\{0} such that hn → 0 and x+hn ∈ I
for all n ∈ N. We obtain
Z x+hn Z x 
F (x + hn ) − F (x) 1
= f (t) dt − f (t) dt
hn hn x0 x0
Z x+hn Z x0 
1
= f (t) dt + f (t) dt
hn x0 x
Z x+hn
1
= f (t) dt
hn x
for all n ∈ N by Lemma 20.1. By the Mean Value Theorem of Integration (with g ≡ 1),
there is, for each n ∈ N, a number ξn between x and x + hn such that
Z x+hn Z x+hn
f (t) dt = f (ξn ) 1 dt = f (ξn )hn ,
x x

116
so that
F (x + hn ) − F (x)
= f (ξn ).
hn
As hn → 0, it follows that ξn → x, and since f is continuous, we conclude that

F (x + hn ) − F (x)
lim = lim f (ξn ) = f (x).
n→∞ hn n→∞

Hence, F is an antiderivative of f .
Let G : I → R be any antiderivative of f . As

(F − G)0 = F 0 − G0 = f − f = 0,

it is clear that F − G must be constant.

Corollary 20.3 (Fundamental Theorem of Calculus). Let I ⊂ R be an interval, let

f : I → R be continuous, and let F : I → R be an antiderivative of f . Then
Z b
f (x) dx = F (b) − F (a)
a

for all a, b ∈ I.

Proof. Let a, b ∈ I. Define

Z x
F̃ : I → R, x 7→ f (t) dt.
a

By Theorem 20.2, F̃ is an antiderivative of f with

Z b
F̃ (a) = 0 and F̃ (b) = f (x) dx,
a

so that Z b
f (x) dx = F̃ (b) − F̃ (a).
a

As F − F̃ is constant, this yields the claim.

Notation. For the right hand side of the formula in Corollary 20.3, we often write
b x=b
F (x) or F (x) .
a x=a

If F is an antiderivative of f , we use the symbols

Z Z
F (x) = f (x) dx or F (x) = f (x) dx + C,

the indefinite integral of f , where the last expression emphasizes that an antiderivative is
unique only up to an additive constant.

117
Examples. 1.
xr+1
Z
xr dx =
r+1
for r ∈ R \ {−1}.

2. Z
1
dx = log x.
x
3. Z
exp(x) dx = exp(x).

4. Z Z
sin x dx = − cos x and cos x dx = sin x.

Corollary 20.4 (Change of Variables). Let I ⊂ R be an interval, let f : I → R be

continuous, let a < b, and let φ : [a, b] → R be continuously differentiable such that
φ([a, b]) ⊂ I. Then
Z b Z φ(b)
0
f (φ(t))φ (t) dt = f (x) dx
a φ(a)

holds.

Proof. Let F : I → R be an antiderivative of f . The Chain Rule then yields

(F ◦ φ)0 = (F 0 ◦ φ)φ0 = (f ◦ φ)φ0 ,

i.e., F ◦ φ is an antiderivative of (f ◦ φ)φ0 . It follows from Corollary 20.3 that

Z b Z φ(b)
0
f (φ(t))φ (t) dt = F (φ(b)) − F (φ(a)) = f (x) dx
a φ(a)

as claimed.

Corollary 20.5 (Integration by Parts). Let a < b, and let f, g : [a, b] → R be continuously
differentiable. Then
Z b Z b
b
0
f (x)g (x) dx = f (x)g(x) − f 0 (x)g(x) dx
a a a

holds.

Proof. Set F := f g, so that F 0 = f 0 g + f g 0 and therefore

Z b Z b Z b
b
0 0
f (x)g(x) dx + f (x)g (x) dx = F 0 (x) dx = F (b) − F (a) = f (x)g(x)
a a a a

as claimed.

118
Exercises
1. Prove Lemma 20.1.

2. Let a < b, and let f, g : [a, b] → R be continuously differentiable such that f (a) ≤
g(a) and f 0 ≤ g 0 . Show that f ≤ g.

3. Let a < b, let f : [a, b] → R be continuously differentiable, and let

Z b
F : R → R, x 7→ f (t) sin(xt) dt.
a

Show that limx→∞ F (x) = limx→−∞ F (x) = 0. (Hint: Integration by Parts.)

4. Let a < b, and let f, g : [a, b] → R be Riemann integrable. By Theorem 19.9(v), f g

is also Riemann integrable. Does then
Z b Z b  Z b 
f (x)g(x) dx = f (x) dx g(x) dx
a a a

necessarily hold?

119
21 Riemann Sums
Our approach to Riemann integration—via order theoretic arguments—is not the original
one. It has the advantage that it is fast and almost effortlessly yields the Riemann
integrability of monotonic functions. Usually, Riemann integrability—and the Riemann
integral—are defined in terms of Riemann sums.

Definition 21.1. Let a < b, let f : [a, b] → R, let a = x0 < x1 < · · · < xn = b, and let
ξk ∈ [xk−1 , xk ] for k = 1, . . . , n. Then the expression
n
X
f (ξk )(xk − xk−1 )
k=1

is called the Riemann sum of f with respect to the partition a = x0 < x1 < · · · < xn = b
with support points ξ1 , . . . , ξn .

f (x )

a x1 x2 x3 ξ b x
ξ1 ξ2 ξ3 4

Figure 14: A Riemann sum

If a function is Riemann integrable, then its Riemann integral can be arbitrarily well
approximated by Riemann sums:

Theorem 21.2. Let a < b, and let f : [a, b] → R be Riemann integrable. Then, for each
 > 0, there is δ > 0 such that, for every partition a = x0 < x1 < · · · < xn = b with

120
maxk=1,...,n xk − xk−1 < δ and any choice of support points ξ1 , . . . , ξn with ξk ∈ [xk−1 , xk ]
for k = 1, . . . , n, we have
Z b n
X
f (x) dx − f (ξk )(xk − xk−1 ) < .
a k=1

Proof. Let  > 0. Then there are step functions φ, ψ : [a, b] → R with φ ≤ f ≤ ψ such that
Rb 
a ψ(x) − φ(x) dx < 2 . Choose a = t0 < t1 < · · · < tm = b such that φ and ψ restricted
to (tj−1 , tj ) are constant for j = 1, . . . , m. Choose C > 0 such that C ≥ |f (x)| for all

x ∈ [a, b], and set δ := 8Cm .
Let a = x0 < x1 < · · · < xn = b be a partition of [a, b] with maxk=1,...,n xk − xk−1 < δ
and pick ξk ∈ [xk−1 , xk ] for k = 1, . . . , n. Define
(
0, if x ∈ {x0 , x1 , . . . , xn },
f˜: [a, b] → R, x 7→
f (ξk ), if x ∈ (xk−1 , xk ) with k ∈ {1, . . . , n}.

Clearly, f˜ is a step function and its Riemann integral from a to b is

Z b n
X
f˜(x) dx = f (ξk )(xk − xk−1 ),
a k=1

i.e., the Riemann sum of f with respect to the partition a = x0 < x1 < · · · < xn = b with
support points ξ1 , . . . , ξn .
The following are clear:

• φ − 2C ≤ f˜ ≤ ψ + 2C;

• if k ∈ {1, . . . , n} and j ∈ {1, . . . , m} are such that [xk−1 , xk ] ⊂ (tj−1 , tj ) then

φ(x) ≤ f˜(x) ≤ ψ(x)

holds for all x ∈ (xk−1 , xk ).

If k ∈ {1, . . . , n} is such that there is no j ∈ {1, . . . , m} with [xk−1 , xk ] ⊂ (tj−1 , tj ),

then there is jk ∈ {0, 1, . . . , n} such that tjk ∈ [xk−1 , xk ]. As any two of the intervals
[x0 , x1 ], . . . , [xn−1 , xn ] have at most one point in common, this means that the number of
all such k is at most 2m. Set
n
[
I= (xk−1 , xk ).
k=1
there is j ∈ {1, . . . , m} such that [xk−1 , xk ] ⊂ (tj−1 , tj )

Define (
0, x ∈ I,
Φ : [a, b] → R, x 7→
2C, otherwise.

121
Obviously, Φ is a step function and satisfies

φ − Φ ≤ f˜ ≤ ψ + Φ.

There are at most 2m intervals (xk−1 , xk ) on which Φ is non-zero. This means that
Z b

Φ(x) dx < 2C(2mδ) = .
a 2

This implies
Z b Z b Z b
 
φ(x) dx − ≤ f˜(x) dx ≤ ψ(x) dx + .
a 2 a a 2
As Z b Z b Z b Z b
 
f (x) dx − ≤ φ(x) dx and ψ(x) dx ≤ f (x) dx + ,
a 2 a a a 2
it follows that Z b Z b
f (x) dx − f˜(x) dx < .
a a
This completes the proof.

In fact, more is true:

Theorem 21.3. Let a < b. Then the following are equivalent for f : [a, b] → R:

(i) f is Riemann integrable;

(ii) there is I ∈ R such that, for each  > 0, there is δ > 0 such that, for every partition
a = x0 < x1 < · · · < xn = b with maxk=1,...,n xk −xk−1 < δ and any choice of support
points ξ1 , . . . , ξn with ξk ∈ [xk−1 , xk ] for k = 1, . . . , n, we have
n
X
I− f (ξk )(xk − xk−1 ) < .
k=1

In this case, we have

Z b
I= f (x) dx.
a

Proof. (i) =⇒ (ii) is the content of Theorem 21.2

(ii) =⇒ (ii): We first claim that f must be bounded. Assume towards a contradiction
that f is unbounded. Choose δ > 0 and a partition a = x0 < x1 < · · · < xn = b with
maxk=1,...,n xk − xk−1 < δ such that
n
X
I− f (ξk )(xk − xk−1 ) < 1.
k=1

122
for all ξk ∈ [xk−1 , xk ] for k = 1, . . . , n. As f is unbounded, there must be k0 ∈ {1, . . . , n}
such that f is unbounded on [xk0 −1 , xk0 ]. For k ∈ {1, . . . , n} \ {k0 }, fix ξk ∈ [xk−1 , xk ].
Choose ξk0 ∈ [xk0 −1 , xk0 ] such that

n
1 X 1
|f (ξk0 )| ≥ I− f (ξk )(xk − xk−1 ) + .
xk0 − xk0 −1 xk0 − xk0 −1
k=1
k6=k0

It follows that

n
X n
X
I− f (ξk )(xk − xk−1 ) ≥ |f (ξk0 )|(xk0 − xk0 −1 ) − I − f (ξk )(xk − xk−1 ) ≥ 1,
k=1 k=1
k6=k0

which is a contradiction. Therefore, f is bounded.

Let  > 0. Choose δ > 0 such that such that, for every partition a = x0 < x1 < · · · <
xn = b with maxk=1,...,n xk − xk−1 < δ and any choice of support points ξ1 , . . . , ξn with
ξk ∈ [xk−1 , xk ] for k = 1, . . . , n, we have
n
X 
I− f (ξk )(xk − xk−1 ) < .
4
k=1

Let a = x0 < x1 < · · · < xn = b be such a partition. For k = 1, . . . , n, set

mk := inf f ([xk−1 , xk ]) and Mk := sup f ([xk−1 , xk ]).

Define
(
f (x), if x ∈ {x0 , x1 , . . . , xn },
φ : [a, b] → R, x 7→
mk , if x ∈ (xk−1 , xk ) with k ∈ {1, . . . , n},

and (
f (x), if x ∈ {x0 , x1 , . . . , xn },
ψ : [a, b] → R, x 7→
Mk , if x ∈ (xk−1 , xk ) with k ∈ {1, . . . , n}.
Then φ and ψ are step functions such that φ ≤ f ≤ ψ. For k = 1, . . . , n choose ξk , ηk ∈
[xk−1 , xk ] such that
 
f (ξk ) − mk < and Mk − f (ηk ) < .
4(b − a) 4(b − a)

123
It follows that
Z b n
X n
X n
X
φ(x) dx − f (ξk )(xk − xk−1 ) = mk (xk − xk−1 ) − f (ξk )(xk − xk−1 )
a k=1 k=1 k=1
n
X
= (f (ξk ) − mk )(xk − xk−1 )
k=1
n
 X
< xk − xk−1
4(b − a)
k=1

<
4
and, analogously,
Z b n
X 
ψ(x) dx − f (ηk )(xk − xk−1 ) < .
a 4
k=1

We conclude that
Z b
I− φ(x) dx
a
n n Z b
X X   
≤ I− f (ξk )(xk − xk−1 ) + f (ξk )(xk − xk−1 ) − φ(x) dx < + =
a 4 4 2
k=1 k=1

as well as Z b

I− ψ(x) dx < ,
a 2
so that
Z b Z b Z b Z b
 
ψ(x) dx − φ(x) dx ≤ ψ(x) dx − I + I − φ(x) dx < + = .
a a a a 2 2

In view of Proposition 19.8, this means that f is Riemann integrable. Finally, Theorem
Rb
21.2 yields that I = a f (x) dx because I has to be unique (see Exercise 1 below).

In fact, the original definition of the Riemann integral uses Theorem 21.3(ii) and
defines the Riemann integral to be I.

Exercises
1. Show that I in Theorem 21.3(ii) is unique.

124
22 The Complex Numbers C, and the Truth about exp, cos,
and sin
Definition 22.1. The complex numbers C are the set

R2 := {(x, y) : x, y ∈ R}

equipped with addition + and multiplication · defined by

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

and

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

for (x1 , y1 ), (x2 , y2 ) ∈ C.

Proposition 22.2. C is a field (with zero element (0, 0) and identity element (1, 0)).

Proof. Except for the existence of the multiplicative inverse, verifying the field axioms is
routine.
Let (x, y) ∈ C \ {(0, 0)}, so that x2 + y 2 > 0. It follows that
   
x −y x −y −y x
(x, y) · , = x − y , x + y
x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2 x2 + y 2
 2
x + y 2 −xy + yx

= ,
x2 + y 2 x2 + y 2
= (1, 0),
 
i.e., x
x2 +y 2
, x2−y
+y 2
is the multiplicative inverse of (x, y).

We identify R with the set

{(x, 0) : x ∈ R} ⊂ C.

Addition and multiplication in C restricted to this set just “are” addition and multipli-
cation of R. For x ∈ R, we therefore simply write x to denote the complex number
(x, 0).
Set i := (0, 1), the imaginary unit. We have

i2 = (0, 1) · (0, 1) = (−1, 0) = −1,

i.e., the equation z 2 = −1 has the solution i in C (−i also solves the equation).

125
 √
Warning. One often sees the expression −1 for i, but this is problematic because one
can easily arrive at nonsense this way:
√ p √ √
1= 1= (−1)(−1) = −1 −1 = −1.
√
The notation −1 should therefore be avoided.
The reason for the apparent paradox is the following:

Proposition 22.3. C cannot be ordered.

Proof. Assume that we can define an order relation “≤” on C such that the order axioms
are satisfied.
Consider 0 and i.
By (O 1), 0 ≤ i or i ≤ 0 must hold.
If 0 ≤ i, then 0 ≤ −1 by (O 5) and therefore 1 ≤ 0, which cannot be.
If i ≤ 0, then 0 ≤ −i, and—again by (O 5)—

0 ≤ (−i)2 = i2 = −1.

This proves the claim.

√
If y > 0, the equation x2 = y has two solutions, of which y is defined to be the
positive one. If y < 0, there is no such way to tell the two solutions of x2 = y apart.
Given z = (x, y) ∈ C, we have

z = (x, 0) + (0, y) = (x, 0)(1, 0) + (y, 0)(0, 1) = x + iy,

which is the usual way to write complex numbers.

Definition 22.4. Let z = x + iy ∈ C. Then:

(a) Re z := x is the real part of z;

(b) Im z := y is the imaginary part of z;

(c) z̄ := x − iy is the complex conjugate of z;

p
(d) |z| := x2 + y 2 is the absolute value or modulus of z.

126
 In terms of these definitions, we have for z, w ∈ C:

z = Re z + i Im z,
z̄ = Re z − i Im z,
z̄¯ = z,
1
Re z = (z + z̄),
2
1
Im z = (z − z̄),
2i
z + w = z̄ + w̄,
zw = z̄ w̄,
√
|z| = z z̄,

and
z̄
z −1 =
|z|2
6 0.
if z =
As for the absolute value on R, we have:
Theorem 22.5. The following are true for z, w ∈ C:
(i) |z| ≥ 0 with |z| = 0 if and only if z = 0;

(ii) |zw| = |z||w|;

(iii) |z + w| ≤ |z| + |w|.

Proof. (i) is clear.
For (ii), note that

|zw|2 = zwzw = (z z̄)(ww̄) = |z|2 |w|2 ,

and take roots.

(iii): First note that

Re z w̄ ≤ |z w̄| = |z w̄| = |z||w̄| = |z||w|,

so that
|z + w|2 = (z + w) (z + w)
= z z̄ + z w̄ + wz̄ + ww̄
= |z|2 + z w̄ + z w̄ + |w|2
= |z|2 + 2 Re z w̄ + |w|2
≤ |z|2 + 2|z||w| + |w|2
= (|z| + |w|)2 .

127
Taking roots again yields the claim.

Given the absolute value on C, we can define convergence of sequences in C:

Definition 22.6. Let (zn )∞ ∞
n=1 be a sequence in C. Then we say that (zn )n=1 converges
to z ∈ C or is convergent to z if, for each  > 0, there is n ∈ N such that |zn − z| <  for
all n ∈ N with n ≥ n . We call z the limit of (zn )∞
n=1 .

Remarks. 1. We use the same notation for limits of complex sequences as for real ones.

2. The Limit Laws hold for complex sequences as for real ones.
Theorem 22.7. Let (zn )∞
n=1 be a sequence in C. Then the following are equivalent:

(i) (zn )∞
n=1 converges in C;

(ii) (Re zn )∞ ∞
n=1 and (Im zn )n=1 converge in R.

In this case, we have

lim zn = lim Re zn + i lim Im zn .
n→∞ n→∞ n→∞
Proof. (i) =⇒ (ii): Let z ∈ C be the limit of ∞
(zn )n=1 . As

|zn − z̄| = |zn − z| = |zn − z|

for n ∈ N, we see that limn→∞ zn = z̄. The Limit Laws yield

1 1
lim Re zn = lim (zn + zn ) = (z + z̄) = Re z.
n→∞ n→∞ 2 2
A similar argument yields limn→∞ Im zn = Im z.
(ii) =⇒ (i): Let x := limn→∞ Re zn and y := limn→∞ Im zn . The Limit Laws then
yield that
lim zn = lim Re zn + i Im zn = x + iy.
n→∞ n→∞
This completes the proof.

Definition 22.8. A sequence (zn )∞ ∞

n=1 in C is called bounded if (|zn |)n=1 is bounded in R.

Remark. It is easy to see that (zn )∞ ∞

n=1 is bounded if and only if both (Re zn )n=1 and
(Im zn )∞
n=1 are bounded.

Corollary 22.9. Every bounded sequence in C has a convergent subsequence.

Proof. We cannot just copy the proof of the real case because it involves order arguments.
Theorem 22.7, however, allows us to steer around this difficulty.
Let (zn )∞ ∞
n=1 be a bounded sequence in C. Then (Re zn )n=1 is a bounded sequence
in R and therefore has a convergent subsequence (Re znk )∞ ∞
k=1 . As (Im zn )n=1 is bounded
in R, so is its subsequence (Im znk )∞k=1 , which therefore has a convergent subsequence
(Im znkν )ν=1 . Then (Re znkν )ν=1 and (Im znkν )∞
∞ ∞
ν=1 are convergent sequences in R, so that
∞
(znkν )ν=1 converges in C.

128
 Of course, we can also define complex Cauchy sequences:

Definition 22.10. A sequence (zn )∞ n=1 in C is called a Cauchy sequence if, for each  > 0,
there is n ∈ N such that |zn − zm | <  for all n, m ≥ n .

As for real sequences, we have:

Theorem 22.11. The following are equivalent for a sequence (zn )∞

n=1 in C:

(i) (zn )∞
n=1 converges in C;

(ii) (zn )∞
n=1 is a Cauchy sequence.

Proof. (i) =⇒ (ii): The proof for real sequences carries over verbatim.
(ii) =⇒ (i): As in the proof of Theorem 22.7, we see that (zn )∞ n=1 is also a Cauchy
sequence. It is easy to see that the sum of two Cauchy sequences and the product of a
Cauchy sequence with a constant are again Cauchy sequences. As in the proof of Theorem
22.7, it then follows that (Re zn )∞ ∞
n=1 and (Im zn )n=1 are Cauchy sequences and therefore
convergent in R. By Theorem 22.7, this means that (zn )∞ n=1 converges in C.

One can now go on and define:

• convergence and absolute convergence of infinite series in C;

• limits of functions from subsets of C into C;

• continuity of functions from subsets of C into C.

In most cases, it is straightforward how the definitions from the real case can be adapted
to the complex situation. Most of the results from Sections 6 to 16—as long as they do
not involve the fact that R can be ordered—carry over. In particular, this is true for
everything we proved for absolutely convergent series.

exp(z) for z ∈ C
We defined
∞
X xk
exp(x) =
k!
k=0
for x ∈ R: this was possible because the series converges absolutely by the Limit Ratio
Test. The same argument shows that the series also converges if x ∈ R is replaced by
some z ∈ C.

Definition 22.12. The exponential function exp : C → C is defined by letting

∞
X zk
exp(z) :=
k!
k=0

for z ∈ C.

129
 It is straightforward that
∞ ∞
X z̄ k X zk
exp(z̄) = = = exp(z)
k! k!
k=0 k=0

for all z ∈ N. The Cauchy Product Formula holds for absolutely convergent complex series
as it holds over R: the proof carries over verbatim. As a consequence, we obtain—as over
R—that
exp(z + w) = exp(z) exp(w)
for z, w ∈ C. As over R, it follows that exp(C) ⊂ C \ {0}. Furthermore, the proof of the
continuity of exp carries over from R to C. As over R, we also write ez instead of exp(z)
for z ∈ C.
Definition 22.13. Define cos, sin : R → R by letting

cos x := Re eix and sin x := Im eix

for x ∈ R.
For the remainder of this section, we shall only work with this definition of cos and
sin. It is immediate that
eix = cos x + i sin x
for x ∈ R.
Properties. 1. cos and sin are continuous, even differentiable.
This follows from Theorem 22.7.

2. For x ∈ R, we obtain

(cos x)2 + (sin x)2 = |eix |2 = eix eix = eix eix = eix e−ix = e0 = 1.

3. Let x, y ∈ R. By definition,

ei(x+y) = cos(x + y) + i sin(x + y)

holds. On the other hand, the exponential law implies

ei(x+y) = eix eiy
= (cos x + i sin x)(cos y + i sin y)
= (cos x)(cos y) − (sin x)(sin y) + i((sin x)(cos y) + (cos x)(sin y)).
Comparing the two expressions, we obtain

cos(x + y) = (cos x)(cos y) − (sin x)(sin y)

and sin(x + y) = (sin x)(cos y) + (cos x)(sin y),

i.e., the addition formulae for cos and sin.

130
 4. It is obvious that 


 1, k ≡ 0 mod 4,

 i, k ≡ 1 mod 4,
ik =


 −1, k ≡ 2 mod 4,

 −i, k ≡ 3 mod 4,
for k ∈ N0 . It follows that
∞
ix
X (ix)k
e =
k!
k=0
∞ ∞
X (ix)k X (ix)k
= +
k! k!
k=0 k=0
k even k odd
∞ ∞
X (ix)2k X (ix)2k+1
= +
(2k)! (2k + 1)!
k=0 k=0
∞ 2k ∞
X x X x2k+1
= (−1)k +i (−1)k
(2k)! (2k + 1)!
|k=0 {z } |k=0 {z }
=cos x =sin x

for x ∈ R, i.e.,
∞ ∞
X x2k X x2k+1
cos x = (−1)k and sin x = (−1)k .
(2k)! (2k + 1)!
k=0 k=0

d x
5. We saw that dx e = ex for x ∈ R; in a similar way, one sees that

d ix
e = ieix .
dx
It follows that
d d d ix
cos x + i sin x = e = ieix = − sin x + i cos x,
dx dx dx
so that
cos0 x = − sin x and sin0 x = cos x

for x ∈ R.

What is π?
It is obvious that cos 0 = 1.

Claim.
1
cos 2 ≤ − .
3

131
Proof. Apply Taylor’s Theorem with f = cos, x0 = 0, x = 2, and n = 3: there is ξ ∈ (0, 2)
such that
cos ξ
cos 2 = 1 − 2 + 16.
24
It follows that
16 2 1
cos 2 ≤ 1 − 2 + = −1 + = − ,
24 3 3
which proves the claim.

The Intermediate Value Theorem immediately yields that cos has a zero in (0, 2).

Definition 22.14. Define π ∈ R as

π := 2 inf{x ≥ 0 : cos x = 0}.

π


Claim. π is well defined, strictly positive and less than 4, and cos 2 = 0. In particular,
π
2 is the least non-negative zero of cos.

Proof. Let
S := {x ≥ 0 : cos x = 0}.
Clearly, S is bounded below and not empty. So, π is well defined. Let (xn )∞
n=1 be a
sequence in S such that xn → inf S. As cos is continuous, it follows that
π 
cos = cos(inf S) = lim cos xn = 0.
2 n→∞
π
As cos 0 = 1, and since cos has a zero in (0, 2), it follows that 2 ∈ (0, 2), i.e., π ∈ (0, 4).

Claim (Values of eix ). The following are true:

π 3π
ei 2 = i, eiπ = −1, ei 2 = −i, and e2πi = 1.

Proof. As cos π2 = 0, we have

  π 2   π 2
sin = 1 − cos = 1,
2 2
i.e., sin π2 = ±1. Assume towards a contradiction that sin π2 = −1. Then the Mean

Value Theorem yields ξ ∈ 0, π2 such that

sin π2 − sin 0


2
cos ξ = π = − < 0,
2 π

so that cos has a zero in 0, π2 by the Intermediate Value Theorem. This contradicts the

definition of π. It follows that sin π2 = 1 and therefore

π
π  π 
ei 2 = cos + i sin = i.
2 2
nπ
As ei 2 = in for n ∈ N, the other claims follow.

132
 In particular, Euler’s Identity holds:

eiπ + 1 = 0.

We obtain the following table of values for cos and sin:

π 3π
x 0 2 π 2 2π
cos x 1 0 −1 0 1
sin x 0 1 0 −1 0

Together with the addition formulae for cos and sin, this yields:

Claim. The following are true for all x ∈ R:

cos (x + 2π) = cos x and sin (x + 2π) = sin x,

cos(x + π) = − cos x and sin(x + π) = − sin x,

and
π  π 
cos x = sin −x and sin x = cos −x
2 2
Claim.
{x ∈ R : sin x = 0} = {nπ : n ∈ Z}.

Proof. Clearly, cos(−x) = cos x for x ∈ R, so that cos x > 0 for x ∈ − π2 , π2 . As

sin x = cos π2 − x , it follows that sin x > 0 for x ∈ (0, π), and since sin(x + π) = − sin x,

we conclude that sin x < 0 for all x ∈ (π, 2π). Consequently, sin only has the zeroes 0 and
π in [0, 2π).
x
Let x ∈ R be such that sin x = 0, and set m := 2π , so that x = 2mπ + ξ with
ξ ∈ [0, 2π). It follows that

sin ξ = sin(x − 2mπ) = sin x = 0,

so that ξ ∈ {0, π}, i.e., x = 2mπ or x = (2m + 1)π.

Claim. n π o
{x ∈ R : cos x = 0} = nπ + : n ∈ Z .
2
π


Proof. This follows from the previous claim because cos x = − sin x − 2 for all x ∈
R.

133
Index

absolute function, 83
convergence, 56 sequence, 35
value, 22 set, 24
accumulation point, 49
Cauchy
addition
criterion for infinite series, 55
associativity of, 19
Product Formula, 67
commutativity of, 19
sequence, 45
existence and uniqueness of inverse,
Cauchy’s Compression Theorem, 54
19
Chain Rule, 94
alternating harmonic series, 57
Change of Variables, 118
Alternating Series Test, 57
closed
Archimedian Property, 26
interval, 24
associativity
set, 74
of addition, 19
closure, 74
of multiplication, 19
codomain, 30
axioms
commutativity
field, 19
of addition, 19
order, 20
of multiplication, 19
Bernoulli’s Inequality, 9 compact set, 75
bijectivity, 31 Comparison Test, 59
binomial coefficient, 12 Limit, 59
Binomial Theorem, 14 Completeness Axiom, 26
Bolzano–Weierstraß Theorem, 51 continuous function, 80
bound at x0 , 79
lower, 24 convergence
largest, 25 absolute, 56
upper, 24 of a function, 77
least, 25 of a sequence, 33
bounded
decreasing function, 87
above
density
sequence, 35
of Q, 26
set, 24
of R \ Q, 27
below
derivative
sequence, 35
first, 91
set, 24

134
Derivative Test strictly decreasing, 87
First, 98 strictly increasing, 87
Second, 101 strictly monotonic, 87
Differentiability of the Inverse Function, surjective, 31
95 uniformly continuous, 85
differentiable function, 91 Fundamental Theorem of Calculus, 117
Differentiation Laws, 93
geometric series, 54
distributivity, 19
domain, 30 harmonic series, 54
alternating, 57
Euler’s
higher derivatives, 96
constant, 69
identity, 133 image of a set under a function, 31
existence and uniqueness of increasing function, 87
additive inverse, 19 indirect proof, 5
multiplicative inverse, 19 induction
exponential function, 69, 80 anchor, 7
hypothesis, 8
factorial, 11
principle of, 7
field, 19
step, 8
axioms, 19
infimum, 25
first derivative, 91
infinite series, 53
First Derivative Test, 98
absolutely convergent, 56
floor, 48
Cauchy criterion, 55
function, 30
partial sum of, 53
bijective, 31
injectivity, 31
bounded, 83
integers, 17
codomain of, 30
integral
continuous, 80
lower, 107
at x0 , 79
upper, 107
decreasing, 87
Integration by Parts, 118
differentiable, 91
Intermediate Value Theorem, 83
domain of, 30
interval
exponentioal, 69
closed, 25
increasing, 87
left open, 25
injective, 31
open, 25
monotonic, 87
right open, 25
range of, 30
inverse
Riemann integrable, 109
function, 32

135
 image of a set under a function, 31 numbers
natural, 6
largest lower bound, 25
properties of, 6
least upper bound, 25
rational, 17
left open interval, 24
Limit open interval, 24
Comparison Test, 59 order axioms, 20
Laws, 36
partial sum, 53
Ratio Test, 60
Pascal’s Triangle, 12
Root Test, 61
phony triangle inequality, 22
limit
power set, 9
inferior, 43
Principle of Induction, 7
of a function, 77
proof, 1
of a sequence, 33
indirect, 5
superior, 43
local range, 30
extremum, 98 Ratio Test, 60
maximum, 98 Limit, 60
minimum, 98 rational numbers, 17
lower Rational Zeroes Theorem, 17
bound, 24 Riemann
largest, 25 Integral, 109
integral, 107 of a constant function, 104
map, see function of a step function, 104
maximum, 24 properties of, 111
Mean Value Theorem, 99 sum, 120
of Integration, 113 Riemann’s Rearrangement Theorem, 64
minimum, 24 right open interval, 24
modulus, 22 Rolles’s Theorem, 99
monotonic function, 87 Root Test, 61
multiplication Limit, 61
associativity of, 19 Second Derivative Test, 101
commutativity of, 19 sequence, 33
existence and uniqueness of inverse, bounded, 35
19 above, 35
n choose k, 12 below, 35
natural numbers, 6 Cauchy, 45
properties of, 6 convergent, 33

136
 decreasing, 39 bound, 24
strictly, 40 least, 25
divergent, 36 integral, 107
to −∞, 36
to ∞, 36
increasing, 39
strictly, 40
limit of, 33
monotonic, 40
recursive definition of, 40
series
geometric, 54
harmonic, 54
alternating, 57
infinite, 53
absolutely convergent, 56
Cauchy criterion, 55
set
bounded, 24
above, 24
below, 24
closed, 74
closure of, 74
compact, 75
singleton, 74
strictly
decreasing function, 87
increasing function, 87
monotonic function, 87
subsequence, 48
supremum, 25
surjectivity, 31

Taylor’s Theorem, 100

triangle inequality, 22
phony, 22

uniformly continuous function, 85

upper

137