A TASTE OF MATHEMATICS

AIME–T–ON LES MATHÉMATIQUES

Volume / Tome XIV

SEQUENCES AND SERIES

Margo Kondratieva
with Justin Rowsell

Memorial University
 The ATOM series
The booklets in the series, A Taste of Mathematics, are published by
the Canadian Mathematical Society (CMS). They are designed as enrichment
materials for high school students with an interest in and aptitude for
mathematics. Some booklets in the series will also cover the materials useful
for mathematical competitions at national and international levels.

La collection ATOM
Publiés par la Société mathématique du Canada (SMC), les livrets de la
collection Aime-t-on les mathématiques (ATOM) sont destinés au
perfectionnement des étudiants du cycle secondaire qui manifestent un intérêt
et des aptitudes pour les mathématiques. Certains livrets de la collection ATOM
servent également de matériel de préparation aux concours de
mathématiques sur l’échiquier national et international.

Editorial Board / Conseil de rédaction

Editor-in-Chief / Rédacteur-en-chef
Alistair Lachlan
Simon Fraser University / Université Simon Fraser

Associate Editors / Rédacteurs associés

Edward J. Barbeau
University of Toronto / Université de Toronto

Malgorzata Dubiel
Simon Fraser University / Université Simon Fraser

Frédéric Gourdeau
Laval University / Université Laval

Joseph Khoury
University of Ottawa / Université d’Ottawa

Antony Thompson
Dalhousie University / Université Dalhousie

Managing Editor / Rédacteur-gérant

Johan Rudnick
CMS / SMC
 A TASTE OF MATHEMATICS

AIME–T–ON LES MATHÉMATIQUES

Volume / Tome XIV

SEQUENCES AND SERIES

Margo Kondratieva
with Justin Rowsell

Memorial University
 Published by the Canadian Mathematical Society, Ottawa, Ontario
and produced by the CMS ATOM Office, Burnaby, BC, Canada

Publié par la Société mathématique du Canada, Ottawa (Ontario)

et produit par le Bureau ATOM de la SMC, Burnaby, BC, Canada

Printed in Canada by / imprimé au Canada par

Thistle Printing Limited

ISBN 978-0-919558-25-0

All rights reserved. No part of this publication may be reproduced or

transmitted in any form or by any means, electronic or mechanical, including
photocopying, recording, or any information or retrieval system now known or
to be invented, without permission in writing from the publisher: The Canadian
Mathematical Society, 209-1725 St. Laurent Blvd., Ottawa, ON, CANADA
K1G 3V4, except so far as may be allowed by law.

Tous droits réservés. Aucune partie de cet ouvrage ne peut être

reproduite ou utilisée par quelque procédé ou quelque façon que ce soit, y compris
les méthodes électroniques ou mécaniques, les enregistrements ou les systèmes
de mise en mémoire et d’information, sans l’accord préalable écrit de l’éditeur,
la Société mathématique du Canada, 209-1725 Boul. St. Laurent, Ottawa, ON,
CANADA K1G 3V4, sauf dans les limites prescrites par la loi.

c 2015
Canadian Mathematical Society / Société mathématique du Canada
 iii

Table of Contents
The authors v

FOREWORD v

INTRODUCTION 1

SEQUENCES 3

What is a sequence? 3

Exercises 4

Sequences which almost fit in a given interval 5

Exercises 7

What is the limit of a sequence? 8

Exercises 9

Properties of limits and some ways of finding them 10

Exercises 11

Recursive formulas 12

Exercises 13

A criterion for convergence 13

Monotonicity check using the derivative 15

Exercises 15

Converting recursive formulas to explicit formulas 16

Geometric sequences 18

Arithmetic sequences 18

Exercises 19

Some interesting sequences 19

The Fibonacci sequence 19

Euler sequence. The first remarkable limit 21

The second remarkable limit 23

The ”half or triple plus one” sequence 24

iv

Exercises 25

SERIES 26

Definitions and sigma notation 26

Finite arithmetic and geometric sums 27

Finite arithmetic sums. Triangular numbers. 27

Finite geometric sums 29

Geometric sums 29

General formula 30

Problems involving geometric series 31

Telescoping 41

Finding telescoping sums 41

Partial fractions 48

Harmonic series and paradox of shifted cards 51

Leibniz series and an unexpected effect of permutation 53

Hyperharmonic series and the Riemann zeta function 55

Power series and approximate evaluation of functions 57

L’Hospital’s Rule and power series 62

Fourier series 65

Exercises 68

APPENDIX 1: DERIVATIVES 69

Definition of the derivative 69

Important derivatives 70

Rules and properties 71

Exercises 72

APPENDIX 2: TRIANGLES 73
 v

The authors
Margo Kondratieva is an Associate Professor at the Memorial University (MUN).
She was born in Moscow (Russia) and received her Master of Science degree in the area of
Applied Mathematics before gaining her PhD in the area of Quantum Mechanics. After
working at the University of Manitoba and University of Minnesota – Duluth (USA), she
arrived at MUN, where she currently holds a joint appointment in the Department of
Mathematics and Statistics and the Faculty of Education.
Dr. Kondratieva is a recipient of the 2008 Motivation Teaching Award for her
teaching of undergraduate mathematics at MUN. Since 2006 she is a provincial coordinator
of various outreach mathematics activities for junior and senior high school students,
including Math Leagues games, the Blundon Seminar math camp, the Canadian Math
Kangaroo competition, and the Virtual Mathematical Marathon available on-line. She
is the author of five book chapters and more than twenty research papers related to
either mathematics or teaching of mathematics. This book is her first work directed to
secondary school students of mathematics.

Justin Rowsell was born in Newfoundland. He was a student of several calculus

and analysis courses taught by Dr. Kondratieva in 2004-2005 at MUN. This booklet has
resulted from a collaborative work during that period, aimed at collecting interesting
and non-standard problems related to summations of infinite series and explained using
mostly elementary techniques. Mr. Rowsell has completed a master’s degree in Pure
Mathematics from the University of Waterloo.

FOREWORD
Secondary school students are often familiar with finite arithmetic and geometric
series. Those who attempt a more advanced level of study become introduced to infinite
series and some formal techniques of their summation. However, many interesting,
non-standard, and important examples remain outside of students’ view and experience.
One reason is a cognitive difficulty related to dealing with infinite sequences of
numbers and the notion of convergence. The formal ‘epsilon-delta’ definition of limit often
presents a challenge even for university level students. Historically, there were different
approaches and interpretations of the concept of convergent series before it received its
modern interpretation. We found it important to refer to a few such historical facts,
dates, and names for the reader’ general information.
In this book, while maintaining a rigorous approach, we use a more intuitive
treatment of the topic. We refer to mostly elementary techniques involving solving
algebraic inequalities, linear and quadratic equations. We believe that the ideas we
explain and illustrate with many examples can be understood at the secondary school
level and help to develop a genuine understanding of the topic. An advanced familiarity
with the topic may foster a deeper study of mathematics at the university level.
Some of our problems are connected to Euclidean geometry or reveal other links
with topics studied at the secondary school level. We also illustrate how infinite sums
may appear while solving some word problems that do not explicitly refer to series and
convergence. We talk about some practical applications, such as calculations with an
approximation. As well, we introduce some notions and objects that are extremely
important in modern mathematics, for example, the Riemann zeta function and the
Dirichlet kernel. We hope that reading this book and solving the exercises will stimulate
students’ interest and fascination with this amazing area of mathematics.
 1

“It is always possible to think of a larger number...”

Aristotle.

INTRODUCTION

The word infinity comes from the Latin infinitus, “unbounded”, or unlimited
in quantity. Infinity is denoted by ∞, and is considered to be not a number but a
concept of increase beyond bounds. When we write n → ∞ we mean that variable
n eventually grows beyond any assigned value or exceeds any finite restriction.
It is the 5th century B.C., Greece. Zeno thinks about paradoxes related to
the notion of infinity, and asks “How can one get from point A to point B at all, if
to do that one needs to go from A to the mid-point C and then to the mid-point
between C and B, and so on, and thus to make an infinite number of iterations?”

A ................................................................................................................... ...................................................................................................................
1 B
. .
............................................................................................................................................................................................................................................
... . . . . . ...
....................................................... 1 ................................................................................. 1 ..................................... 1 .............. 1 ...... ..... ...............
. . . . . . ...
2 4 8 16

Figure 1. Zeno’s paradox.

In a response, Diogenes stands up and simply walks from A to B. In modern

mathematical language this story looks as follows:
1 1 1 1
+ + + + · · · = 1.
2 4 8 16
Here, the dots symbolize the addition of an infinite number of terms, where each
term is one half as large as the previous one.
The left hand side of the formula is an example of what is called an infinite
series. As we see, despite the fact that it has an infinite number of terms, the
result is a finite number. This formula shows that the distance between A and B
is finite, but it can be broken into an infinite number of parts. If an infinite series
has a finite sum, as in our example, it is called a convergent series.
Starting from the first term of the infinite series and adding the subsequent
terms one by one, we will obtain numbers —
1 1 1 1 1 1 1 1 1 1
S1 = , S2 = + , S3 = + + , S4 = + + + , ...
2 2 4 2 4 8 2 4 8 16
— none of which can be regarded as a final result, but which approach the final
result closer and closer as we sum more and more terms of the series. We can see
it better if we rewrite the previous line as
1 1 1 1
S1 = 1 − , S2 = 1 − , S3 = 1 − , S4 = 1 − , ....
2 4 8 16
The number which is approached in this infinite process is called the sum of the
infinite series.
2

There are some obvious situations where the process of accumulating more
and more summands does not lead to a number. For example, if you add the same
number infinitely many times
1 + 1 + 1 + 1 + ··· .
Adding terms one by one —
S1 = 1, S2 = 1 + 1, S3 = 1 + 1 + 1, S4 = 1 + 1 + 1 + 1, . . . ,
— Sn becomes arbitrarily large. In this case we say that the infinite series diverges
to infinity.
Another example is the infinite series
1 − 1 + 1 − 1 + 1 − 1 + ··· .
One can see that adding any odd number of terms we obtain 1, while the sum of
any even number of terms gives 0. That is, in the process of adding more terms
neither of the numbers is approached (in contrast with the first example). Such
an infinite series is also said to be divergent.
There are many examples, such as
1 1 1 1
1 + + + + + ··· ,
2 3 4 5
in which the answer to the question about convergence is not easy to find.
In this book we talk about infinite series; under what conditions they can
be represented by a finite number (are convergent), and how to find that number.
This subject requires careful treatment. Infinite series differ from finite series, and
thus you may not always trust your intuition based on your experience with finite
algebraic sums. Formal manipulations can easily lead to nonsense. Treating our
third example incorrectly, one can “prove” that 0 = 1 as follows
0 = (1 − 1) + (1 − 1) + (1 − 1) + · · · = 1 + (−1 + 1) + (−1 + 1) + · · · = 1.
This paradoxical regrouping is possible because the number of terms is infinite.
Notice that the expressions in parentheses are all zeros. Historically, this example
was given at the very beginning of 18th century by the Italian scholar Guido
Grandi to demonstrate the possibility of creating everything from nothing.
Another story from the past assigns the value 1/2 to the series
1
1 − 1 + 1 − 1 + 1 − 1··· = (false)
2
symbolizing that if two people share something indivisible (like a diamond ring),
to be fair they should take turns owning it; one year the first person owns the
ring, next year the other one, and so on, forever. Interestingly, Gottfried Leibniz
mentioned this incorrect result in a paper of 1674, in connection with geometric
series (which we will discuss in Section 2.3).
We will have a few more examples of infinite series which are contradictory
at first glance, and we will explain how to deal with them to avoid fallacies.
But, before starting our discussion of infinite series, we need to talk a little
about infinite sequences of numbers, the terms of which may be summed to form
series.
 3

CHAPTER 1: SEQUENCES

§1.1 What is a sequence?

Definition 1. A sequence is an ordered list of numbers. The numbers are
called terms (or elements) of the sequence.

The numbers could represent measurements of temperature made day by

day, the height of students in your class, or something else.
In mathematics, a sequence is often given by a formula for its n-th term,
this is called the explicit formula for the sequence.
Example 1. Let the n-th term be an = n2 , where n runs through the familiar
set of natural numbers
N = {1, 2, 3, 4, . . .}.
We say that the first term of the sequence (corresponding to n = 1) is a1 = 12 = 1,
the second term (corresponding to n = 2) being a2 = 22 = 4, the third term being
a3 = 32 = 9, and so on. Thus, the sequence given by the general term an = n2 is
{1, 4, 9, 16, 25, 36, . . .}.
We will employ the notation

n ∈ N,

which is a short way of saying that a number n is an element of the set of natural
numbers N. In general, x ∈ X means that x is an element of the set X. We will
use this notation regularly from now on.

Example 2. The sequence with n-th term an = n, where n ∈ N, is obviously

the sequence of natural numbers {1, 2, 3, 4, 5, ...}. 1

100
Example 3. The sequence with n-th term an = n , where n ∈ N, is

{100, 50, 100/3, 25, 20, ...}.

Example 4. The sequence with n-th term an = 1, n ∈ N, is an example of a

constant sequence. All its terms have the same value.

Example 5. The sequence with n-th term an = (−1)n , n ∈ N, belongs to the

class of alternating sequences since its terms alternate in sign: {−1, 1, −1, 1, −1, ...}.

We will talk mostly about sequences with an infinite number of terms, which
can be denoted by {an }∞
n=1 . We summarize the above examples and include some

1 We follow the tradition of using the same notation {1, 2, 3, 4, 5, ...} for both the set and the

sequence of natural numbers.

4

more in the table, and will use them in our future discussion.

Sequence n-th term 1st term 2nd term 3rd term 4th term 5th term
∞
{an }n=1 an a1 a2 a3 a4 a5

Example 1 n2 1 4 9 16 25

Example 2 n 1 2 3 4 5

100 100
Example 3 100 50 25 20
n 3

Example 4 1 1 1 1 1 1

Example 5 (−1)n −1 1 −1 1 −1

(−1)n 1 1 1 1
Example 6 −1 − −
n 2 3 4 5
n−1 1 2 1 3 4 2
Example 7 0 = =
n+1 3 4 2 5 6 3
1 1 1 1 1 1
Example 8
3n 3 9 27 81 243

Exercises
1. In each case list the first 10 terms of the sequence2 {an }∞
n=1 :

1
i) an = −n ii) an = n3 iii) an = iv) an = 2n v) an = n!
n2

2. Find n-th term of the sequence

1 1 1 1
i) {2, 5, 10, 17, 26, ...} ii) {−1, , − , , − , ...}
2 6 24 120
iii) {1, 3, 7, 15, 31, ...} iv) {3, 7, 11, 15, 19, ...}

1 1 1 1 1 1 1 3 9 27 81
v) {− , , − , , − , ...} vi) { , , , , , ...}
2 5 10 17 26 3 2 4 8 16 32

2 In (v) n-factorial denotes the product of all integers from 1 to n: n! = n(n − 1)(n − 2) · · · 1.

For example, 4! = 4 · 3 · 2 · 1 = 24.

 5

§1.2 Sequences which almost fit in a given interval

We use square bracket notation [A, B] for closed intervals. That is, the
interval from A to B including the endpoints: x ∈ [A, B] means A ≤ x ≤ B.
An open interval, one which does not include its endpoints, is denoted by (A, B).
Thus, x ∈ (A, B) means A < x < B.
In this section we become familiar with the situation in which a sequence
almost fits in a given open interval.

Definition 2. We say that almost all of the elements of a sequence {an }∞

n=1
belong to an interval (A, B) if all but a finite number of terms of the sequence
satisfy the inequality A < an < B.

Problem 1. In the following cases say whether almost all of the terms of
{an }∞
n=1 are in (A, B):

100
i) (A, B) = (0, 200), an = ;
n
100
Answer: Yes. All terms of the sequence are in the interval: 0 < < 200 for all
n
n ≥ 1.
100
ii) (A, B) = (0, 20), an = ;
n
Answer: Yes. Only the first five terms are outside the interval. To see that we
100
solve inequality < 20 and get n > 5. Thus, almost all of the terms of the
n
sequence are in the interval.
100
iii) (A, B) = (1, 200), an = ;
n
Answer: No. Only the first 99 terms belong to the interval, while the rest do not.
100
Solving 1 < we get n < 100. Thus, if n ≥ 100, then an ≤ 1, which places an
n
outside the given interval.
√
iv) (A, B) = (0, 2000), an = n;
√
Answer: No. The inequality n < 2000 is true only for n < 4, 000, 000. Thus only
the first 3,999,999 terms of this sequence belong to the interval, while the rest do
not.

v) (A, B) = (0, 10), an = (−1)n ;

Answer: No. All terms of this sequence with an even index a2 = a4 = a6 = · · · = 1,
and they do belong to the interval (0, 10). However, all terms of this sequence with
an odd index a1 = a3 = a5 = · · · = −1, and they do not belong to the interval
(0, 10).
6

Problem 2. Given a sequence, find an example (if possible) of an interval

to which almost all of the terms of the sequence belong.

Example Answer

no finite interval contains

E1: an = n2 , n ∈ N
almost all terms
no finite interval contains
E2: an = n, n ∈ N
almost all terms
100
E3: an = , n∈N (0, 0.2) but not (0.1, 0.2)
n

E4: an = 1, n ∈ N (0.5, 1.4)

E5: an = (−1)n , n ∈ N (−2, 2) but not (0, 2) or (−2, 1)

(−1)n
E6: an = , n∈N (−0.1, 0.22)
n
n−1
E7: an = , n∈N (0.5, 1.01) but not (0.5, 0.99)
n+1

Instead of giving just an example, we can attempt to give a complete description

of the intervals in Problem 2 as is required in the following problem.
Problem 3. Given a sequence, describe the intervals to which almost all of
the terms of the sequence belong.

Example Answer

100
E3: an = , n∈N (A, B) for any numbers A ≤ 0 and B > 0.
n

E4: an = 1, n ∈ N (A, B) with A < 1 and B > 1.

E5: an = (−1)n , n ∈ N (A, B) for any numbers A < −1 and B > 1.

(−1)n
E6: an = , n∈N (A, B) for any numbers A < 0 and B > 0.
n
n−1
E7: an = , n∈N (A, B) for any numbers A < 1 and B ≥ 1.
n+1
 7

Note that the general description given in the answers for Problem 3 provides
all possible particular examples which can serve as answers in Problem 2. The
general answer comes from careful analysis of all possible particular answers.
Let us consider Example 7 in more details. First note that all terms of the
n−1 2
sequence with n-th term an = = 1− , where n ≥ 1, are less than 1
n+1 n+1
and greater or equal to zero : 0 ≤ an < 1. Thus interval (0, 1) contains almost
all terms of the sequence. Obviously, any interval larger than that will also work,
so any interval (A, B) with A ≤ 0 and B ≥ 1 contains almost all terms of the
sequence. Now we note that even if A is greater than 0, the interval (A, 1) still
works as long as A is less than 1. That is, almost all terms of the sequence from
Example 7 belong to interval (A, B) for any A < 1 and B ≥ 1.
Indeed, given 0 < C < 1 we see that term an belongs to the interval (1−C, 1)
2−C
for any n > . To confirm that we simply solve the inequality
C
2
1−C <1− .
n+1
We can try some concrete numbers to make sense of this statement. For
example, take C = 0.1 and observe that 0.9 < an < 1 for all n ≥ 20; similarly,
take C = 0.01 and note that 0.99 < an < 1 for all n ≥ 200.

Exercises
1. Decide whether or not almost all the terms of {an }∞
n=1 belong to the given
interval:
1 1
i) an = , (0.05, 0.1) ii) an = π −n , (0, 1) iii) an = , (0, 5)
n 2−n
1
2. For the sequence an = , n ∈ N, (see Example 8) give an example of an
3n
interval to which almost all (but not all) of the terms belong (as in Problem 2).
Then describe all the intervals to which almost all of the terms of this sequence
belong (as in Problem 3).

3. For each sequence {an }∞n=1 either find an interval to which almost all (but not
all) of the terms belong or show that there is no finite interval containing almost
all terms.
1 √ √
i) an = ii) an = n5 iii) an = n iv) an = sin n
n

4. Justify the answers given in Problem 3 for sequences in Examples 3, 4, 5, 6.

8

§1.3 What is the limit of a sequence?

As n increases, the terms an may approach a certain number. If this is
the case, we say that the sequence has a limit or, equivalently, that the sequence
converges. Intuitively, a limit is a target or a destination point for the sequence.
Some sequences reach their target; for example an = 1, n ∈ N, is always there.
Some other sequences never reach their limits; for example an = 100 n , n ∈ N, never
reaches the value zero – it gets extremely close to zero but no term of the sequence
actually takes that value. A sequence can also approach its target not only from
n
one side but from both sides: an = (−1) n , n ∈ N, approaches zero while the values
an are alternately positive and negative.

Following the approach of German mathematician Heinrich Eduard Heine,

who lived in the 19-th century, we will employ the following definition of the limit.

Definition 3. A number L is called the limit of a sequence {an }∞ n=1 as

n → ∞ if every open interval containing L also contains almost all terms of the
sequence. Then we write
lim an = L.
n→∞

Comments:
1. Note that at most one number L can be the limit of {an }∞ n=1 as n → ∞.
2. The word every in the definition is significant and refers to all without exception.
For example, suppose that someone tries to argue that an = 100 n has limit 1
because, for all B > 1, the interval (0, B) contains 1 as well as almost all terms of
the sequence. The inadequacy of this argument is clear when one points out that
(0.5, 1.5) contains 1 but only 133 terms of the sequence (for 67 ≤ n ≤ 199). Thus
1 is not the limit of the sequence.

100
Problem 4. Show that the limit as n → ∞ of the sequence an = n is
zero:
100
lim = 0.
n→∞ n
Solution: Note that every open interval containing zero has the form (A, B),
where A < 0 and B > 0. We need to show that, for any A < 0 and any B > 0,
almost all terms of the sequence under consideration satisfy the inequality:
A < an < B.
First we note that an = 100n is positive for any n ∈ N, thus A < 0 < an .
Second, solving the inequality 100
n < B for n, we find that given number B,
terms an < B for all n > 100
B . For instance, if B = 1 then an < 1 for all n > 100;
if B = 0.1 then an < 0.1 for all n > 1000, etc. Thus no matter how small the
length of the interval (A, B) is, only a finite number of the terms an are outside
(A, B).
According to the definition, this means that 0 is the limit of this sequence.
 9

Note that every open interval containing the number L has a form (A, B)
with A < L and B > L. Thus to show that lim an = L we take an open interval
n→∞
(A,B), with A < L and B > L, and demonstrate that almost all of the terms of the
sequence are in this interval no matter how small the length of the interval is. If
we fail to do so, i.e. we can justify that for a given sequence it is impossible to find
such a number L, we say that the sequence does not have a limit or, equivalently,
that the sequence diverges.

Problem 5. In each case use the definition to determine whether the given
sequence has a limit.
i) an = n2 , n ∈ N (Example 1).
Solution: In accordance with the definition, if a sequence has a limit then we must
be able to find an interval to which almost all of it terms belong. But the sequence
an = n2 runs away from any given interval as n becomes large. Thus, it does not
have a limit. In a case like this, when the terms of a sequence become arbitrarily
large, we also say that the sequence diverges to infinity and write

lim an = ∞.
n→∞

ii) an = (−1)n , n ∈ N (Example 5).

Solution: In accordance with the definition, if a sequence has a limit then we must
be able to find an arbitrarily small interval to which almost all of it terms belong.
The sequence an = (−1)n cannot fit inside any interval of length less then 2. Thus
it does not have a limit.
iii) an = n−1
n+1 , n ∈ N (Example 7).
Solution: From our previous discussion of this sequence in Problem 3 we know
that any interval (A, B) with A < 1 and B ≥ 1 contains almost all terms of the
sequence. The only problem is that there is no such a number L that belongs
to all of the intervals. For that reason we will consider a slightly different set of
intervals, namely (A, B) with A < 1 and B > 1. Note that we just changed the
condition for B from B ≥ 1 to B > 1, but it makes a big difference! Interval
(A, B) with A < 1 and B > 1 is a general description of an interval containing
1. Thus now every interval in the set contains almost all terms of the sequence as
well as the number L = 1. Therefore 1 is the limit of the sequence by definition.
We will consider another way of getting this result in Section 1.4.

Exercises
∞
1. Does the sequence {cos(πn)}n=1 converge?
∞
What about the sequence {sin(πn)}n=1 ?
Explain your answers.
n2 − 2 3n + 2
2. Justify the following formulas lim = 1 and lim = 3.
n→∞ n2 n→∞ n
10

§1.4 Properties of limits and some ways of finding them

It is often easier to find a limit using the properties which follow from the
definition, rather than using the definition itself. In this section we give some
examples of this.
From our previous discussion we have the following:
1
P1. lim = 0.
n→∞ n

P2. lim 1 = 1.
n→∞

P3. lim n = ∞.
n→∞

Provided lim an and lim bn exist, in the equalities P4, P5, P6, and P7 given
n→∞ n→∞
below, the limit on the left exists and equals the expression on the right:
P4. lim c · an = c · lim an for any constant c.
n→∞ n→∞

P5. lim (an + bn ) = lim an + lim bn .

n→∞ n→∞ n→∞

P6. lim (an · bn ) = ( lim an ) · ( lim bn ).

n→∞ n→∞ n→∞

an lim an
P7. lim = n→∞ , assuming that bn 6= 0, n ∈ N, and lim bn 6= 0.
n→∞ bn lim bn n→∞
n→∞

From the above properties one can also derive:

1
P8. lim k = 0, for any integer k > 0.
n→∞ n

P9. lim c = c, with c being a constant.

n→∞

P10. lim nk = ∞, for any integer k > 0.

n→∞

Let us now calculate the limit from the previous subsection (Problem 5 (iii)).
It could be done as follows:

n−1 1− 1 lim (1 − n1 ) lim 1 − lim 1

1−0
n n→∞ n→∞ n→∞ n
lim = lim 1 = = = = 1. (1)
n→∞ n + 1 n→∞ 1 +
n lim (1 + n1 ) lim 1 + lim 1 1+0
n→∞ n→∞ n→∞ n

Here we first divided the top and bottom of the fraction by n, (the highest
power of n in this case) then used property P7, then P5, and finally P1 and P2.
Note that the first step is necessary, because otherwise by properties P2, P3,
P4 and P5 both numerator and denominator of the fraction diverge to infinity, so
we cannot use property P7. But even if we could apply P7 formally, we would
have obtained
n−1 lim (n − 1) ∞
lim = n→∞ = .
n→∞ n + 1 lim (n + 1) ∞
n→∞
 11

The result is uncertain and the situation needs to be resolved, which we accomplish
by the division. This ∞∞ is what mathematicians call an indeterminate form. Some
other indeterminate forms are
0
, 00 , 1∞ .
0
We will have an example of the last of these cases when we talk about the Euler
sequence (see Section 1.9.2). If one arrives at an indeterminate form, one needs
to resolve it. A useful technique for doing this called L’Hospital’s rule is given in
Section 2.9.
Note that in some cases we reach expressions which look like indeterminate
forms but are not. For instance, for a constant c 6= 0 you can safely proceed as
follows:
0
= 0, c0 = 1, 0|c| = 0.
c
Formally you can also write 0c = ∞ if c > 0, and 0c = −∞ if c < 0. However in
contrast with previous formulas, here we do not have equality of two numbers as
both values are a short way of describing unboundedness.
The result in (1) may be generalized as follows. Consider the case in which
an is a rational function of n, i.e., the ratio of two polynomials of n of degrees m
and k
Bm nm + · · · + B1 n + B0
an = .
Ck nk + · · · + C1 n + C0
Then 
 0 
  if m < k
Bm
lim an = sign Ck ∞ if m > k
n→∞
 Bm

Ck if m = k.
Examples:
3n5 + 1
Example 9. an = . Here m = 5, k = 7. Thus lim an = 0.
4n7 + n + 1 n→∞
8
3n + 1
Example 10. an = . Here m = 8, k = 7 and B8 = 3, C7 = 4. Thus
4n7 + n + 1
lim an = ∞.
n→∞
3n5 + 1
Example 11. an = lim . Here m = k = 5 and B5 = 3, C5 = 4. Thus
n→∞ 4n5 + n + 1
3
lim an = .
n→∞ 4

Exercises
1. Provide detailed calculations for Examples 9, 10, and 11. Follow the method
given in (1) starting with the division of numerator and denominator by an
appropriate power of n. At each step show which property of limits you use.
2. In each case find the limit as n → ∞ of given sequence either by using the
definition of limit directly or the properties P1 - P10:
√
1 20 − n n
i) an = 3 ii) an = iii) an =
n − 3n2 − 3 2n + 12 n+1
12

§1.5 Recursive formulas

A sequence can be defined recursively as well as explicitly. A recursive
definition of a sequence is a definition which describes a term by using one or more
previous terms. The recursive definition of the natural numbers is an = an−1 + 1,
n > 1, with initial condition a1 = 1. The explicit formula is an = n, n ≥ 1.
When a sequence is defined recursively, there must always be initial conditions
given. The number of initial conditions needed to define a sequence uniquely is the
order of the recurrence relation. For example, the sequence an = 3an−1 , a1 = 1 is
of the first order, while an = 3an−1 + an−2 , a1 = 1, a2 = 3 is of the second order.
Finding the limit of a sequence from its recursive definition is straightforward
in some cases while in others it requires some thought. A straightforward example
is the following.

Problem 6. Find the limit of the sequence (asuming that the limit exists)
2 1
an = , a1 = .
1 + an−1 2
Solution: If we know that the sequence has a limit, we can derive the equation
that it must satisfy. From Definition 3 it is clear that, if a0 is assigned an arbitrary
value, and lim an is defined, equal L say, then lim an−1 = L also.
n→∞ n→∞
So, using properties P5, P7 and P9 of the limit, we can say that:
2
lim an =
n→∞ 1 + lim an−1
n→∞
2
L =
1+L
L2 + L = 2
2
L +L−2 = 0
L = −2 or L = 1

We know that a sequence can only have one limit so we want to find out which
one of these is the right value. By looking at our recursive formula, we know that
as long as a term is positive, the term after it must be positive. This is because
the term after it would be a ratio of two positive numbers. Since the first term is
positive, the term after it must be positive, as is the term after that one and so
on. Therefore, the limit must be non-negative. Thus, lim an = 1.
n→∞
Let us justify this result using the definition of the limit discussed earlier.
We need to show that any open interval that contains 1 will also contain an infinite
number of terms of the sequence {an }. For that we estimate the distance between
an and 1 as n increases. We find
(1 + an ) − 2 an − 1
1 − an+1 = = , so |1 − an+1 | = |1 − an | · |(an + 1)−1 |.
1 + an an + 1
 13

Now, if we could show that |(an + 1)−1 | ≤ r, for all n ≥ 1, then it would imply
that |1 − an+1 | ≤ |1 − a1 |rn < rn (since 1 − a1 = 0.5). Note that if 0 < r < 1 then
rn < 1/n for all n ≥ n0 starting from some positive n0 . In this case we would have
|1 − an+1 | < 1/n and conclude that every interval (1 − A, 1 + A), where A > 0
contains all terms an for n > 1/A.
Now let us find r < 1 such that |(an + 1)−1 | ≤ r. Calculate the few first
terms of the sequence: a1 = 1/2, a2 = 4/3, a3 = 6/7, a4 = 14/13, etc. Observe
that if an ≥ 1/2 then an + 1 ≥ 3/2 and thus (an + 1)−1 ≤ 2/3. Consequently,
an+1 ≤ 4/3. As well, if an ≤ 4/3 then an+1 ≥ 6/7 > 1/2. Thus if we start
with a term in the interval [1/2, 4/3], all terms of the sequence will remain in this
interval, and for all of them we have (an + 1)−1 ≤ r, where r = 2/3 < 1. Since
(2/3)n < 1/n for all n ≥ 1, we conclude that the limit of the sequence is 1.
This method of solving for the limit will not always work. We need to
know before using this method that the sequence does have a limit.
The following example illustrates that otherwise the method can lead to a wrong
conclusion.
Example 12. Let us study the sequence an+1 = 3an , a1 = 1. If we just consider
what this means, it is easy to give the answer. Every term is 3 times as large as
the term before it. If we take a positive number, and keep tripling it infinitely
many times, the value is going to go towards infinity, and we say that the sequence
diverges to infinity. If we had assumed that this sequence had a limit and tried to
solve for it using the method explained, we would have obtained
L = 3L.
Technically, infinity satisfies the above relation. But since we assumed that L is a
number (not infinity), we get L = 0, which is clearly false.
Note also that a sequence defined by recursive formula an+1 = 3an has limit
L = 0 if and only if the initial term a1 = 0.

Exercises
1. Find the limit of each recursively defined sequence given that it exists.
an 1
i) an+1 = , a1 = 3 ii) an+1 = a2n , a1 =
an − 2 2

§1.6 A criterion for convergence

Example 12 from previous the section shows that it is helpful to know
whether the limit of a sequence exists before attempting to evaluate it. The
theorem given in this section helps to determine when a sequence has a limit.

Definition 4. A sequence is called non-decreasing if an ≤ an+1 for all

n ∈ N. A sequence is called non-increasing if an ≥ an+1 for all n ∈ N. A sequence
is called monotone if it is either non-decreasing or non-increasing.
14

Definition 5. A sequence is called bounded if there are numbers K and M

such that K ≤ an ≤ M for all n ∈ N.

The following is a very important property of the real numbers whose proof
is beyond the scope of this work.

Theorem 1. If a sequence is monotone and bounded then it has a limit.

Neither hypothesis by itself is sufficient. A sequence which is monotone

and unbounded will diverge, like an = 3n . Also, a sequence which is bounded
but not monotone may not have a limit, like an = (−1)n . On the other hand,
a convergent sequence may not be monotone, an = (−1)n /n for example, but it
must be bounded.

Problem 7. Show that the sequence

1 + an−1
an = , a1 = 1
4
is monotone and bounded, and thus has a limit.
Solution: We can show both properties using the method of mathematical induction.
Looking at the first terms of the sequence, {1, 1/2, 3/8, ...}, we guess that the
sequence is non-increasing and bounded by K = 0 and M = 1.
First, we prove that the sequence is non-increasing. We see that a1 ≥ a2 .
Now assume that ak−1 ≥ ak . Then 1+ak−1 ≥ 1+ak , and 1+a4k−1 ≥ 1+a4k+1 . Using
the recursive formula which defines the sequence, we rewrite the last formula as
ak ≥ ak+1 . Thus, we have shown that the inequality ak−1 ≥ ak implies ak ≥ ak+1
for any k ≥ 2. Since this inequality is true for k = 1, we can conclude that
an ≥ an+1 for all n ∈ N, and by definition the sequence is non-increasing.
Second we prove that the sequence is bounded: 0 < an ≤ 1. Clearly,
0 < a1 ≤ 1. Now assume that 0 < ak ≤ 1. Since the terms are non-increasing,
ak+1 ≤ ak ≤ 1. Also, ak+1 = (1 + ak )/4 > 0 since ak > 0. Thus 0 < ak ≤ 1
implies 0 < ak+1 ≤ 1 for all k ≥ 1. We conclude that 0 < an ≤ 1 for all n ≥ 1.
Now, using Theorem 1, we conclude that the sequence has a limit. To find
the limit we use the method from the previous section and obtain L = 1/3.

Problem 8. Show that the sequence an = n−1

n+1 is monotone.
Solution: We write
(n + 1) − 1 n n−1
an+1 = = > = an .
(n + 1) + 1 n+2 n+1
The inequality follows from n(n+1) > (n+2)(n−1) which is equivalent to 0 > −2.
Recall that this sequence is bounded and has 1 as its limit.

The question of whether a sequence is monotone or not can also be addressed

by employing one rather useful tool which we now introduce. If you are not familiar
with derivatives, you should read Appendix A before continuing.
 15

§1.7 Monotonicity check using the derivative

Let us look at a sequence simply as a function whose domain is the positive
integers. So we will say that a sequence {an }∞ n=1 and a function f (x), x ≥ 1,
correspond to each other if and only if an = f (n) for all positive integers. For
example, the sequence an = n2 corresponds to the function f (x) = x2 and the
sequence an = sin n corresponds to the function f (x) = sin x.
This point of view is convenient because instead of checking the monotonicity
of a sequence, we now can ask this question about its corresponding function. In
Appendix A we tell you how to determine the monotonicity of a function by using
its derivative. This process is straightforward, and therefore applicable in a large
number of cases. Assuming that by now the reader is familiar with the Appendix,
let us consider the following example.

Problem 9. Is the sequence an = n2 − 50n, n ≥ 1 monotone?

Solution: The first few terms of the sequence are:
{−49, −96, −141, −184, −225, −264, −301}. From this, it seems that the sequence
is monotonically decreasing. But having written down any finite number of terms,
you still cannot be sure of what will happen to a sequence next, so the conclusion
cannot be made yet.
Let us consider the corresponding function. For that we just replace n with
x. This gives f (x) = x2 − 50x. The derivative of the function is f 0 (x) = 2x − 50.
Note that the derivative is negative for x < 25 and it is positive for x > 25. This
means that the function f (x) = x2 − 50x is decreasing on the interval x < 25 and
is increasing for x > 25. Consequently the sequence an = n2 − 50n is decreasing
for 1 ≤ n ≤ 24 and is increasing for n ≥ 26.
For the purpose of a convergency check, we are interested in the behavior of
the sequence as n → ∞. Therefore, when checking monotonicity, we can disregard
any finite number of terms. From our analysis we can conclude that the sequence
is monotonically increasing for n ≥ 26.

Problem 10. Apply the same approach for an = n−1 n+1 , n ∈ N, considered
previously.
Solution: The corresponding function is f (x) = x−1
x+1 . The derivative, calculated
using the quotient rule, is f 0 (x) = (x+1)−(x−1)
(x+1)2
2
= (x+1) 2 . This expression is always

positive for x ≥ 1 and thus the sequence is monotonically increasing for n ∈

{1, 2, 3, . . .}, which confirms the result obtained before by direct calculations.

Exercises
1. Give an example of a convergent and non-monotone sequence.
2. Give an example of a divergent and monotone sequence.
3. Give an example of a divergent and bounded sequence.
n2
4. Show that an = 2 is monotone and bounded. Find its limit.
n +1
16

§1.8 Converting recursive formulas to explicit formulas

It can often be difficult to find the limit of a sequence when we only know
its recursive definition. It is generally much easier to work with a sequence if we
have an explicit formula for an . For example:

Problem 11. Find the limits of the following sequences:

i) an = (1/2)n , n ≥ 0.
3 1
ii) bn = 2bn−1 − bn−2 , n ≥ 2 with b0 = 1, b1 = 2
4
Comment: In all previous examples the index n was a natural number, n ≥ 1.
However, sometimes we start counting from another number, for instance in this
problem from 0. This is a common practice, so in the future please pay attention
to this aspect of the problems as well.
Solution:
i) This sequence is decreasing since each term is half the previous one: an+1 =
(1/2)n+1 = (1/2)an < an . It is also bounded from below since all terms are
positive. Thus the limit exists. Since an = an−1 /2 for all n ≥ 1, the limit of the
sequence is 0 by the method of Section 1.5.
ii) If we knew that the sequence had a limit, it would be straightforward to show
that it is 0 by the method of Section 1.5. The problem is to justify that the limit
exists. To accomplish this we will show that the two formulas, (i) and (ii), actually
describe the same sequence, that is an = bn for all n ≥ 0. To support this we
begin by writing down the first six terms of each sequence,
1 1 1 1 1
{an } = {1, , , , , , . . .}
2 4 8 16 32
1 1 1 1 1
{bn } = {1, , , , , , . . .}
2 4 8 16 32
and observing that they are the same (note that indexes start with zero for both
sequences {an }∞ ∞
n=0 and {bn }n=0 in this example). Next we will show that an = bn
for all n ≥ 0. Thus, we will convert the definition of bn from recursive to explicit
form. We can do it by method of mathematical induction. First, observe that
b0 = 1 = (1/2)0 and b1 = 1/2 = (1/2)1 . Take n ≥ 1 and assume that bk = (1/2)k
for 0 ≤ k ≤ n. We need to show that these conditions imply bn+1 = (1/2)n+1 .
Indeed, using the recursive formula we have
 n  n−1    n−1  n−1  n+1
1 3 1 3 1 1 1 1
bn+1 = 2 − = 1− = = .
2 4 2 4 2 4 2 2
Thus, by principle of mathematical induction we have bk = (1/2)k = ak for k ≥ 0.
However, if we did not know ahead of time to which explicit formula we need
to compare a given recursive equation, we could use a different method. For doing
that we need a little bit of theory.
 17

Definition 6. A second order linear homogeneous recurrence relation is a

recurrence relation of the form

an = ran−1 + qan−2 , (2)

where r, q 6= 0 are constants. The quadratic equation

x2 − rx − q = 0 (3)

is called the characteristic equation of the recurrence relation (2).

We will assume that the constants r, q satisfy the inequality r2 + 4q > 0, and
thus the quadratic equation (3) has two distinct real solutions.

Theorem 2. Let x1 and x2 be solutions of the quadratic equation (3). Let

A and B be arbitrary constants. Then, the formula

an = Axn1 + Bxn2 , n≥0 (4)

defines explicitly the same sequence as the one given by the recurrence relation
an = ran−1 + qan−2 with initial conditions a0 = A + B, a1 = Ax1 + Bx2 .

Let us apply Theorem 2 to sequence {bn } defined in Problem 11. We have

r = 2 and q = −3/4. The corresponding characteristic equation is x2 −2x+3/4 = 0
which has roots x1 = 1/2 and x2 = 3/2. From the initial conditions b0 = 1,
b1 = 1/2, we have A + B = 1, A/2 + 3B/4 = 1/2. Thus A = 1, B = 0, and
the explicit formula is bn = Axn1 + Bxn2 = (1/2)n . This is exactly what we have
claimed above.

Now we give the proof of the theorem.

Proof: First, we show that for any numbers A, B the n-th term given by
(4) will satisfy the recurrence relation (2). Suppose that an = Axn1 + Bxn2 then
an−1 = Axn−11 + Bxn−1
2 and an−2 = Axn−21 + Bx2n−2 . In order for (2) to hold,we
must have

Axn1 + Bxn2 = r(Axn−1

1 + Bxn−1
2 ) + q(Axn−2
1 + Bxn−2
2 ).

That is, we must have

A(xn1 − rxn−1
1 − qxn−2
1 ) + B(xn2 − rxn−1
2 − qxn−2
2 ) = 0,

which is equivalent to,

xn−2 A(x21 − rx1 − q) + B(x22 − rx2 − q) =

 
0.

This is clearly true for all n since both of the expressions in parentheses are equal
to zero because x1 and x2 are solutions to the characteristic equation (3).
The initial conditions for (2) follow from (4) with n = 0, 1.
18

We will need this theorem again to obtain the explicit formula for the
Fibonacci numbers (Section 1.9.1).

§1.8.1 Geometric sequences

Setting q = 0 in Theorem 2 we obtain the following:

Theorem 3. The recurrence relation

an = ran−1 , a0 = A,
where r, A are constants, defines the sequence given explicitly by
an = Arn , n ≥ 0.
Problem 12. Find the explicit formula for the sequence defined by the
recurrence relation an = 2an−1 with a0 = 3.
Solution: This is the first order linear homogeneous recurrence relation with r = 2
and A = 3. Thus, in the explicit form, an = 3 · 2n .

The sequences of the form an = Arn , A 6= 0 are called geometric. In other words,
a geometric sequence is a sequence where consecutive terms have a common ratio,
r. Some geometric sequences have a finite limit while some other ones do not
depending on the value of the common ratio r. There are four different cases and
they are described in the following table.

r lim Arn
n→∞

r ≤ −1 no limit
−1 < r < 1 0
r=1 A
r>1 ∞ or −∞ according
as A > 0 or A < 0

§1.8.2 Arithmetic sequences

Now we consider the non-homogeneous linear recurrence relations of the form
an = an−1 + d, a0 = A.
For the sequence defined by these equations the difference between any term
(after the first) and the one before is a constant d, called the common difference.
Such a sequence is called arithmetic. The equivalent explicit definition is
an = A + nd, n ≥ 0.
An example of an arithmetic sequence is {8, 5, 2, −1, −4, −7}. Its explicit definition
is an = 8 − 3n, n ≥ 0. The recursive definition is an = an−1 − 3, a0 = 8. The
common difference is −3.
 19

All arithmetic sequences (with d 6= 0) diverge. If d > 0 then lim an = ∞,

n→∞
and if d < 0 then lim an = −∞.
n→∞

Exercises
1. In each case convert the given recursive definition to an explicit definition
of the same sequence. Check your answer by verifying that the two definitions agree
for the first few terms. In each case decide whether the sequence is convergent or
divergent and explain your answer.
an−1
i) an = , a0 = 100 ii) an = 10 + an−1 , a0 = 100
10
3 1
iii) an = an−1 − an−2 , a0 = 1, a1 =
16 4

§1.9 Some interesting sequences

In this section we look at four well-known sequences. The first is notable
because of the properties of its members. The next two are interesting because of
their limits. The latter problem is attractive because it looks simple but is still
unsolved.

§1.9.1 The Fibonacci sequence

The Fibonacci sequence is usually defined by the recurrence relation
Fn = Fn−1 + Fn−2 , n > 2, with F1 = 1 and F2 = 1.
In other words, it is the sequence for which the first two terms are 1 and each term
after that is the sum of the two terms before it. The first few terms are
1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, ... .
The sequence is named after the Italian mathematician of the Middle Ages, who
used this example in his Book of Calculation (Liber Abaci) published in 1202. The
sequence represented the growth of the number of pairs of rabbits on an island if
each pair gives rise to a new pair every month starting from the second month of
their own life, and never die. Now, the sequence is referred to as the first model
of population dynamics.
If you were given the recursive formula for this sequence and were told to
find the 1000-th Fibonacci number (F1000 ), it would be a very tedious task. That
is why we need an explicit formula for the n-th term of the √
sequence. First we will
define the golden ratio. The golden ratio is equal to 1+2 5 ≈ 1.61 and is denoted
by φ. Taking q = r = 1, a0 = 0, and a1 = 1, in Theorem 2 we see that the
sequence referred to in the theorem is {0, F1 , F2 , F3 , ...}. In this case the roots of
the characteristic
√ equation
√ are φ and 1 − φ. We ensure that a1 = a2 = 1 by letting
A = 1/ 5 and B = −1/ 5. It follows that the n-th Fibonacci number is:
20

φn − (1 − φ)n
Fn = √ . (5)
5
Even though this equation contain radicals, every term of the sequence is an
integer.

There are many interesting properties of the Fibonacci numbers. Some of

them we list here.

1. The limit of the Fibonacci sequence is infinity, but the limit of the ratio
of consecutive terms is the golden ratio
Fn+1
lim = φ.
n→∞ Fn

2. The sum of the first N Fibonacci numbers is

F1 + F2 + · · · FN = FN +2 − 1.

For example, if N = 3 we have F1 + F2 + F3 = 1 + 1 + 2 = 4 = 5 − 1 = F5 − 1.

3. The absolute value of the difference between the square of a Fibonacci
number and the product of the previous and the next Fibonacci numbers is 1.

FN2 − FN −1 FN +1 = (−1)N +1 .
For example, if N = 4 then F42 − F3 F5 = 32 − 2 · 5 = −1.
4. The sum of the squares of two consecutive Fibonacci numbers is another
Fibonacci number
FN2 + FN2 +1 = F2N +1 .
For example, if N = 4 then F42 + F52 = 32 + 52 = 34 = F9 .
5. The difference of the squares of two consecutive Fibonacci numbers is the
product of two other Fibonacci numbers

FN2 +1 − FN2 = FN −1 FN +2 .

For example, if N = 4 then F52 − F42 = 52 − 32 = 16 = 2 · 8 = F3 F6 .

6. The only squares among Fibonacci numbers are F1 = 1, F2 = 1, F12 =
144.
7. Every 3rd Fibonacci number is even (i.e. divisible by F3 = 2). Every 4th
Fibonacci number is divisible by F4 = 3. Every 5th Fibonacci number is divisible
by F5 = 5. More generally, the ratio Fn·k /Fn is always an integer.
8. There are prime numbers amongst the Fibonacci numbers, like 1, 2, 3, 5, 13,
and so on. The question of whether there are an infinite or finite number of primes
amongst the Fibonacci numbers still remains open.

There are other interesting properties of the Fibonacci sequence. We will

provide one more, with its simple proof.
 21

Problem 13. Show that:

1 1 1
− =
Fn−1 · Fn Fn · Fn+1 Fn−1 · Fn+1
Solution: Start by finding a common denominator
1 1 Fn+1 − Fn−1
− =
Fn−1 · Fn Fn · Fn+1 Fn−1 · Fn · Fn+1

and then use the recursive definition of the Fibonacci sequence and substitute
Fn = Fn+1 − Fn−1 to allow cancellation:
Fn+1 − Fn−1 Fn
=
Fn−1 · Fn · Fn+1 Fn−1 · Fn · Fn+1
1
=
Fn−1 · Fn+1
which is our desired result.

§1.9.2 Euler sequence. The first remarkable limit

Some sequences have limits which are unexpected. In this section we give
one such example, which is due Leonard Euler, a Swiss mathematician of the 18th
century. Consider a sequence given by its n-th term
 n
1
an = 1 + , n ≥ 1.
n
We know that limn→∞ (1+(1/n)) = 1. But, because 1+(1/n) is raised to the power
n, none of the discussion above gives us a way of knowing whether limn→∞ an exists
or what its value is if it does exist.
We will have to find the limit of this sequence using another method.
First, note that the expression (1 + (1/n))n can be rewritten using the
binomial formula. Familiar examples of this formula are (a + b)2 = a2 + 2ab + b2
and (a + b)3 = a3 + 3a2 b + 3ab2 + b3 . In general for n ≥ 2 we have the following
sum of n + 1 terms:
     
n n−1 n n−2 2 n n−k k
(a + b)n = an + a b+ a b + ··· + a b + · · · + bn ,
1 2 k
 
n n!
where the binomial coefficients = are defined for 1 ≤ k < n. Using
k k!(n − k)!
the formula n! = n(n − 1)(n − 2) · · · 1 one can also write
 
n n(n − 1) · · · (n − k + 1)
= .
k k!
Setting a = 1 and b = 1/n we obtain
     
1 n n 1 n 1 n 1 1
(1 + ) = 1 + + 2
+ ··· + k
+ ··· + n.
n 1 n 2 n k n n
22

Now observe what happens to each term on the right hand side as n → ∞.
First, limn→∞ 1 = 1. Second,
 
n 1 n! 1
lim = lim = lim 1 = 1;
n→∞ 1 n n→∞ (n − 1)!1! n n→∞

Third,  
n 1 n! 1 n(n − 1) 1
lim 2
= lim 2
= lim 2
= ;
n→∞ 2 n n→∞ (n − 2)!2! n n→∞ n 2! 2!
Similarly, for any 1 ≤ k < n we have

 
n 1 n! 1 n(n − 1)(n − 2) · · · (n − k + 1) 1
lim · k = lim k
= lim k
= ;
n→∞ k n n→∞ (n − k)!k! n n→∞ n k! k!
Note that
       
1 n n 1 n 1 n 1 1
lim (1 + ) = lim 1 + + + · · · + + · · · + .
n→∞ n n→∞ 1 n 2 n2 k nk nn
The limits of individual terms found above suggest that the righthand side tends
to the infinite sum of inverse factorials
1 1
1+1+ + ··· + + ··· . (6)
2! k!
This can be shown rigorously by demonstrating that the sequence {(1 + 1/n)n }∞ n=1
is increasing and bounded and that it approaches the sum (6) as n → ∞.
As we will see in the next chapter, the infnite sum of inverse factorials (6)
is equal to the number e ≈ 2.71, the base of the natural logarithm. 3
1 1
1+1+ + ··· + + · · · = e.
2! k!
Thus, we claim:  n
1
lim 1+ = e. (7)
n→∞ n

The limit is known as the first remarkable limit. This limit can be also found
using logarithms, and L’Hospital’s rule (see Sec. 2.9 Example 3).
In fact, we have a more general formula. For any constant M ,
 n
M
lim 1 + = eM . (8)
n→∞ n
You will be able to prove (8) after reading section 2.9.

3 The natural logarithm log (x) of a number x > 0 is the power to which e would have to be
e
raised to equal x, and it also equals the area under the hyperbola y = 1/x on the segment from
1 to x. Since loge e = 1, the number e can be defined as the unique number greater than 1 such
that the area under the hyperbola y=1/x from 1 to e equals 1.
 23

§1.9.3 The second remarkable limit

 
1
Let us consider the sequence an = n sin , n ∈ N. It is not immediately
n
1
clear what the limit of this sequence is, since lim n = ∞, lim sin( ) = 0 and
n→∞ n→∞ n
0 · ∞ is an indeterminate form.
1
Let us introduce the notation xn = . Then
n
 
1 sin(xn )
lim n sin = lim .
n→∞ n n→∞ xn
Note that 0 < xn < π/2 for all n ≥ 1. Consider an acute positive angle of measure
xn radians, and consider a sector of a unit circle with inner angle xn . Let O be the
centre of the circle, and let A and B be the points where the angle meets the arc,
with A below, as shown in the Figure 2. Also, let us draw a line perpendicular to
OA which passes through B and let its intersection with OA be called C. Similarly,
extend OB, draw a line perpendicular to OA which runs through A and let its
intersection with OB be called D. We have,
D
.....
B.................... .......
. ...
...........
........ ............. ....
........ .... ...... ....
......
........ . . . .
....
........ . ....... .....
.
.
.
.
........... .. ... ... ...
.
.
.......... . .. .. ...
.
. . .
..... . . . ..
.
........ ..... .... .... .....
........ ... ... .... ...
...
............ .... .. ......
.
........ .... .. ......
.
............. ....
.. ......
.. ....
...........
. .
. .. ......
....
...
..... .
.
.
...
.
..
......... xn ............ . .................
.. . . . ..
O ............................................................................................................................................................................................
C A
Figure 2. A geometric proof of the second remarkable limit.

Now, |OA| = |OB| = 1, because they are both radii of the unit circle. So,
by the definition of sin and tan, we have |BC| = sin(xn ) and |DA| = tan(xn ). It
is easy to see from the figure that the area of the sector lies strictly between the
areas [OAB] and [OAD] of triangles 4OAB and 4OAD, respectively. Now, area
1 sin(xn ) 1 tan(xn )
[OAB] = |OA| · |BC| = , area [OAD] = |OA| · |AD| = .
2 2 2 2
xn xn
Since the area of the sector is π · = , we have
2π 2
sin(xn ) xn tan(xn )
< < .
2 2 2
2
Because > 0 for 0 < xn < π/2, we can multiply throughout the inequality
sin(xn )
2
by to obtain
sin(xn )
xn 1
1< <
sin(xn ) cos(xn )
which is equivalent to
sin(xn )
1> > cos(xn ).
xn
24

Note that cos(xn ) = |OC| in Figure 2 and |OA| = 1. As n increases C moves

to the right and becomes closer and closer to A. We claim that for every point E
in the segment OA, for all but a finite number of n, C lies to the right of E and
so lim cos(xn ) = |OA| = 1.
n→∞
Now we can apply the squeeze principle, which says that if an ≥ bn ≥ cn
for all n ≥ 1 and both sequences {an } and {cn } have the same limit L, then
1
lim bn = L. In our case both an = 1 and cn = cos( ) have the limit 1 as n → ∞,
n→∞ n
sin(xn ) 1
and bn = = n sin( ). Consequently,
xn n
 
1
lim n sin = 1. (9)
n→∞ n
This result is often referred to as the second remarkable limit.

§1.9.4 The “half or triple plus one” sequence

Let us consider the following recursive process. Take any odd number n.
Multiply it by 3 and add 1; divide the result, 3n + 1, by 2 as many times as
necessary to get an odd number. Repeat the process indefinitely.
For example, if a1 = 1 then 3 × 1 + 1 = 4 and after division by 2 twice we
get a2 = 1. Thus for the initial value 1, all terms of this sequence are 1. Other
examples are given in the table below:

1st term 2nd term 3rd term 4th term 5th term 6th term 7th term
a1 a2 a3 a4 a5 a6 a7

3 5 1 1 1 1 1

7 11 17 13 5 1 1

19 29 11 17 13 5 1

As you see, in all considered examples we eventually get a term equal to 1,

after which all following terms are 1’s. In other words, in all our examples the
sequence converges to 1. You are encouraged to try some other numbers as well. If
you start with an even number then first divide it by 2 as many times as necessary
to get an odd number and then continue as described above.
In 1937 Lothar Collatz proposed that no matter what number you start
with, you will always eventually reach 1. However, this statement has not yet
been either proved or disproved.
This situation demonstrates that even if some problems look simple, modern
mathematics may not be powerful enough to solve them.
 25

Exercises
1. i) Find the characteristic equation of the Fibonacci sequence and show that the
golden ratio φ is one of its roots. What is the other root?
ii) Derive the explicit formula (5) for the n-th term of the Fibonacci sequence.
iii) What do you know about the golden ration φ from geometry?
2. Prove property 5 of the Fibonnaci numbers.
3. Prove property 7 of the Fibonnaci numbers.
4. Calculate k5 for k = 1, 2, 3, 4 and use the binomial formula to expand (a + b)5

as a sum of six terms. Verify your result by direct multiplication.

5. Find the sum 41 + 42 + 43 . Prove that the sum of the binomial coefficients

     
n n n
+ + ··· + = 2n − 2
1 2 n−1
for all n ≥ 2.
6. Prove using the binomial formula that the sequence
 n ∞
1
1+
n n=1

is increasing.
1
7. i) Show that x sin is increasing for x > 1.
x
ii) Use (9) to show that for any number a
a
lim n sin = a.
n→∞ n
8. Justify the squeeze principle used in 1.9.3 from Definition 3.
9. Fix a positive integer k. Show that, for all but a finite number of n,
 n
1 1 1 1
1+ ≥ 1 + + + ... + .
n 1! 2! k!
26

CHAPTER 2: SERIES

§2.1 Definitions and sigma notation

Definition 7. A series is an expression of the form a1 + a2 + a3 + · · · which is
the sum of an infinite number of terms of the sequence {an }∞
n=1 .

The questions we ask about a series are: “Is it convergent?” and, if the
series is convergent, “What is its sum?” To make clear what we are talking about
we begin with a definition.

Definition 8. The n-th partial sum Sn of a sequence {an }∞

n=1 is the sum of the
first n terms of that sequence. That is,

Sn = a1 + a2 + a3 + · · · + an .

Note that the partial sums Sn form another sequence as n grows,

S1 = a1 , S2 = a1 + a2 , S3 = a1 + a2 + a3 , ... Sn = a1 + a2 + · · · + an .

The limit as n → ∞ of the sequence of partial sums, if it exists, gives the sum of
the series:
lim Sn = a1 + a2 + a3 + · · · .
n→∞
Definition 9. If the limit of the sequence of partial sums exists, the series is
convergent. Otherwise it is divergent.

This definition is due to Augustin-Louis Cauchy and appeared in his book,

Cours d’analyse, in 1821.
In 1820 Joseph Fourier introduced a shorthand way to write a series called
sigma notation. The partial sum Sn , represented by sigma notation, is:
n
X
Sn = a1 + a2 + a3 + · · · + an = ak .
k=1

The series and its sum, if it exists, will both be denoted by

∞
X
ak .
k=1

The following properties are used in algebraic manipulations

n
X n
X n
X n
X n
X n
X
(ak + bk ) = ak + bk , cak = c ak , c = cn. (10)
k=1 k=1 k=1 k=1 k=1 k=1

The first two of these properties are valid when n is replaced by ∞ provided the
sequence(s) on the right of the equation converge.
 27

§2.2 Finite arithmetic and geometric sums

Sometimes we might be asked questions like: What is the sum of the first
100 natural numbers? Or, what is the sum of the first 25 positive integer powers
of 2? The direct summation may seem tedious, but the questions can be quickly
answered
There are general formulae for the sum of a finite number of terms of an
arithmetic or geometric sequence, and we will talk about them in this section.

§2.2.1 Finite arithmetic sums. Triangular numbers.

Suppose that circles of equal diameter are packed tightly in n rows inside an
equilateral triangle (see Figure 3). How many circles are there? It is clear that we
have 1+2+3+· · ·+n circles. This sum has a special name, and appears frequently
in mathematics. It is called the n-th triangular number and can be calculated as

n(n + 1)
Tn = 1 + 2 + · · · + n = .
2

Carl Friedrich Gauss found this formula for triangular numbers when he was
around the age of ten. He used a unique method for this. He listed the n terms,
wrote this list backwards under the original list, and added vertically.

Tn = 1 + 2 + ··· + n
+ + + +
Tn = n + n−1 + ··· + 1
⇓ ⇓ ⇓ ⇓
2Tn = n+1 + n+1 + ··· + n+1

2Tn = n(n + 1)
n
Tn = (n + 1)
2
For a more general arithmetic sequence ak = a + (k − 1)d, k = 1, 2, ..., the
sum Sn of its first n terms is
n
X
Sn = (a + (k − 1)d)
k=1
n n n
!
X X X
= a+d k− 1
k=1 k=1 k=1
dn(n + 1) n
= na + − dn = (2a + (n − 1)d).
2 2
Here we have used the properties (10).
The natural numbers form an arithmetic sequence with a = 1 and d = 1.
Thus expression for Sn simplifies to Tn in this case.
28

Patterns show up in many triangular numbers. One example is the triangular

numbers whose indices are powers of 10. For those, we have
T10 = 55
T100 = 5050
T1000 = 500500.
Some more general formulas are
2
Tn+1 − Tn2 = (n + 1)3
Tn+1 + Tn = (n + 1)2 .
The reader can check these properties as a simple algebraic exercise.
Now let us present a problem which involves the triangular numbers.

Problem 14. Reconsider the construction stated earlier with the circles
inside the equilateral triangle (Figure 3). Let A be the area of the equilateral
triangle 4ABC and An be the total area occupied by the n rows of circles, show
that
An π
lim = √
n→∞ A 2 3
Solution: First, we know that the number of circles inside the triangle is the n-th
triangular number. Thus, the number of circles is n(n + 1)/2.
We will denote the side length of the triangle as L and we will denote the
radius of the circles as r. Now, the length of the side is the sum of the diameters
of n circles and two small portions on each end of the length. We will denote each
of these pieces as b and write L = 2nr + 2b. First, we find b in terms of r. For that
we draw a small right triangle 4BOE in the corner of the equilateral triangle. Its
vertices will be at the vertex B of the equilateral triangle, center O of the closest
circle, and point E of tangency of the closest circle with the side of the equilateral
triangle.

O
r

B C B b+r E

Figure 3. Circles inside an equilateral triangle.

Because the large triangle 4ABC is equilateral, its angles are π/3 radians,
and consequently the small angle ∠OBE in the right triangle is half of it, π/6
radians. From this we can see that
√ √
tan(π/6) = r/(b + r) ⇒ b + r = 3r ⇒ b = ( 3 − 1)r.
 29

√ √ L
Now, L = 2nr + 2( 3 − 1)r = 2(n + 3 − 1)r. Thus, r = √ .
2(n + 3 − 1)
The total area covered by n rows of circles is
n(n + 1) 2 πL2 n(n + 1)
An = πr = √ .
2 8(n + 3 − 1)2
√
Because the triangle is equilateral √ with side length
√ 2 L, its height is ( 3/2)L and
consequently its area is (1/2) · L · ( 3/2)L = 3 L /4. Now,
 
An 1
lim = lim An ·
n→∞ A n→∞ A
πL2 n(n + 1)
  
4
= lim √ · √
n→∞ 8(n + 3 − 1)2 3L2
n2 + n
 
π
= √ lim √ √
2 3 n→∞ n2 + 2( 3 − 1)n + ( 3 − 1)2
π π
= √ ·1= √ .
2 3 2 3
The limit is found using techniques learned in Section 1.4.

§2.2.2 Finite geometric sums

For a finite geometric series, there is a trick to find its sum as well. Let

S = 1 + r + r2 + · · · + rn ;

then, we have

r·S = r + r2 + r3 + · · · + rn + rn+1 .

Consequently

S − rS = 1 − rn+1 ,
(1 − r)S = 1 − rn+1 ,
1 − rn+1
S = .
1−r
In particular, the sum of the first 25 positive integer powers of 2 is 67,108,862.

§2.3 Infinite geometric sums

In this section we deal with an infinite geometric series of the form
∞
X
rn = 1 + r + r2 + r3 + · · · .
n=0
30

§2.3.1. The general formula

Geometric series are very nice due to the fact that there is a very simple way
to determine their convergence, and, if convergent, find their sum. A geometric
series diverges if |r| ≥ 1; otherwise it converges, and its sum is the following:
∞
X 1
rn = , for |r| < 1.
n=0
1−r

This formula has been known for a long time: it appeared in Euclid’s Elements,
written approximately 2300 years ago, and was used in the work of Archimedes.
There are several ways to prove this formula. One of them is motivated by
the following example.
Consider a geometric series with r = 12 :
1 1 1
1+ + + + ··· .
2 4 8
Now, let us assume that the sum is a finite number and denote it by x. Then
1 1 1
2x = 2 + 1 + + + + ... = 2 + x.
2 4 8
Thus, solving 2x = 2 + x, we get x = 2. We have
1 1 1
1+ + + + · · · = 2.
2 4 8
Remember, we already had similar formula in the introduction, where its motivation
was more geometrical than algebraic.
Now consider a geometric series with r = 1/q:
1 1 1
1+ + 2 + 3 + ··· .
q q q
Again, let us assume that the sum is a finite number and denote it by x. Then
1 1 1
qx = q + 1 + + 2 + 3 + · · · = q + x.
q q q
Thus x = q/(q−1). Substitute q = 1/r and do some algebra to obtain x = 1/(1−r).
Until now, it seems as if there is no restriction on the value of r. The
following example, where r = 2, shows that it is not so, and formal use of this
approach leads to a contradiction. Let x = 1 + 2 + 4 + 8 + 16 + · · · . The sum is
positive since each term is positive. On the other hand, 2x = 2+4+8+· · · = x−1.
Solving for x we have x = −1. Thus, we have just “proved” that the sum of an
infinite number of positive terms is equal to a negative number. This is clearly
false and thus, for r = 2 the series is divergent. Therefore, we conclude that this
“proof” works only under the assumption that the series converges, but does not
show when it does.

Another proof of the general formula, which gives a condition for convergence,
 31

follows directly from the formula for a finite geometric sum through the use of
Definition 9 and the properties of limits. Consider the n-th partial sum
1 − rn+1
Sn = 1 + r + · · · + rn = .
1−r
Then the limit
1 − lim rn+1
n→∞
lim Sn =
1−r n→∞

exists only if |r| < 1, when lim rn+1 = 0 and thus by definition
n→∞

1
1 + r + r2 + · · · = lim Sn = . (11)
n→∞ 1−r
This result is illustrated for 0 < r < 1 by Figure 4 below. It is a good exercise
to convince yourself that from the similarity of the triangles ACD and AP1 B it
follows that r = (s − 1)/s and thus s = (1 − r)−1 .
C
.......
.......... ..
............ ...
.............. .....
..
..
..
. ....................... .....
.... . .
....... ...... ...
....................... ....
...
.......... .... ....... ... ....
..... .
............................. ... ....
..... .... .. ... .
..
..
..
P
5................. ..... ........ ... ...
... .
. ....
....
.. .
.. .
......................................... . . .
.
. .
.
. ...
...
. ....
..
......... .... .
.... ... ..
. ..
. ....
...
..... .
. ... ...
. .
. .
. ....
..
P 3 . . . .
3............. .....
..... .. .....
.
.
.. .....
.
r .
.
.
.
.
.
.
.
.
.
.
.
....
...
...................................................... .
. .
. .
.
.....
. .
........ ..
.
..
.
.
2 ..... ... .
....... 4 .
.
. P .
.
. .
.
.
....
....
..........
.
.
......... ...
..
.
. . ... ..
. r .
.
.
.
.
.
.
.
.
.
.
.
.
.
. s−1
....
....
.
....... ..
. .... . . . .
...
.
..... 2 .... . . . . . .
. .
.....
...
.
. ..r..... .
.
.
.
.
.
.
.
.
.
.
. ....
P
1..................
.
..
.. .
.
. .
.... ......
... .
.
..
.
.
..
.
.
..
.
.
..
....
....
... .. ..
. r
..................................................................................
.
.
. ..
.
P .
. .
. .
. .
. ....
...
..
......... ... ..... ... 2 ..
. ..
. ..
. ..
. ....
...
....... ..
.
. ... .. ... .
.
. .
.
. .
.
. .
.
. ....
.....
..... ..
.
. ..... .
. .
. .
. .
. .
. ....
.........
. .
.
.
. ..... . . . . . ....
..........
. ..
.
. ... .. . . . . . ...
....
...
........
..... ..
.
.
.
.
.
r ..
.... ...
. ..
.
.
.
.
..
.
.
.
.
..
.
.
.
.
..
.
.
.
.
..
.
.
.
.
....
....
..
......... ..
.
. ..
. .. .
. .
. .
. .
. .
. ....
...
....... ..
.
. . .... .
.
.
.
.
.
.
.
.
. ....
.....
..... .. .....
.
. .. .. .. .. .. ....
A
..
....
......
.........
1 . ....
.
.
.
.
.
.
....
. . B r
.
.
.
.
.
.
2 .
.
.
.
.
. r 3 .
.
.
.
.
. r
.
.
.
.
.
.
.
.
.
.
.................................................................................................................... ..................................................................................................................................................................................................................................................................................................................................................................................................
.
.
D ..
.... .
.
....
. ..
.......
.
.......
.
...
0 ....
....
0 s 1 s 2 s 3 4 s s ··· s
....... ....
....... ....
...
......... ......
. .
..... .. .
...
.. ...
.......
.......
.......
Slope is r ....
...
... Slope is 1
....... ...
.
............ ... ....
... .
....... ....
....... ...

Figure 4. A sum of a geometric series.

§2.3.2 Problems involving geometric series

Let us consider the following situation. A fireman uses water from a tank
at a rate three times as big as the rate at which water is supplied to the tank. If
there were initially 600 litres of water in the tank, how many litres will be used
before the tank is empty?
We can approach this question as follows. Observe that by the time a given
amount of water is used by the fireman, a third of this amount will be supplied to
32

the tank. Thus after 600 litres are used, 600/3=200 litres will be supplied to the
tank. Similarly, by the time 200 litres are used, 200/3=600/9 litres will be supplied
to the tank, etc. If we continue in the same way, the total amount of water used
will be represented by the infinite geometric series 600 × (1 + 1/3 + 1/9 + ..). Using
(11) with r = 1/3 we find 1 + 1/3 + 1/9 + .. = 3/2. Thus, the answer is 900 litres.
The geometric sum (11) appears in many other problems involving an infinite
sequence of similar geometric objects (segments, squares, triangles, circles). Some
examples of this kind, including fractals, are collected in this section. Besides
that, using geometric series, one can see why any repeating decimal number can
be written as a fraction and thus is a rational number. Another example illustrates
how formula (11) helps deal with an infinite product.

Problem 15. The path of a ball. A ball falls from a height of 1m and
bounces to heights of 1/3 m, 1/9 m, 1/27 m, and so on. Find the total distance
that the ball covers after an infinite number of bounces.
Solution: The total distance is
   
1 1 1 2 1 1 2 1
1+2 + + + · · · = 1+ 1 + + + · · · = 1+ · = 1 + 1 = 2.
3 9 27 3 3 9 3 1 − 13

Here formula (11) with r = 1/3 was used. The answer is 2 m.

Problem 16. An infinite sequence of squares.

i) Consider an infinite sequence of squares such that vertices of each square coincide
with the midpoints of the sides of the previous square. Find the sum of the areas
of the squares, if the first square has an area of 1 m2 .

Figure 5. A sequence of squares.

ii) Let the first square be black, the second one be white, the third one be black
again, and so on. The colouring is shown on the right. Find the total area covered
by the black colour.

Solution: First note that the area of a square, whose vertices are the midpoints
of the sides of a given square, is exactly half that of the given square. Since this
pattern continues, we have an infinite sequence of squares in which each square
area is one-half as large as the area of the previous one. Thus we have a geometric
sequence of the areas of the squares: 1, 1/2, 1/4, 1/8, ....
 33

In question (i) we are asked about the total area of the squares, which
according to our previous discussion is the sum of the geometric series with r = 1/2:

1 1 1 1
1+ + + + ··· = 1 =2
2 4 8 1− 2

The answer for question (i) is 2 m2 .

To answer question (ii) we have to add every black square area and subtract
every white square area, which results in geometric series with r = −1/2:

1 1 1 1 2
1− + − + ··· = = .
2 4 8 1 − (− 12 ) 3

The answer for question (ii) is 2/3 m2 .

Problem 17. The length of a jagged line in an angle.

Consider an angle 0 < θ < π/2. Start with any point A0 on the side of the angle
and draw the perpendicular to the other side. Denote the point of intersection
by A1 . Then draw the perpendicular to the opposite side of the angle and denote
the intersection by A2 , and so on. By the jagged line we understand the resulting
piece-wise linear curve with turning points at A0 , A1 , A2 , ...Ak , ....
The problem is to find the length of the jagged line given the length of the
first segment |A0 A1 | = b.

.......
.........
A1 .........
..
..
..
. .........
....
.........
.............
A ........ .........
3................... ... ...
.
....... ..
. .... ....
......... .... ..... ....
......... ..... ....
..
... . .
.
.......... ... .. ... ...
. . . .
......... ......... ..... .... .....
...
.... . . .
..........
. .
.
. .. .
. .. ... ...
......... ...... ... ... .. ... . ...
.
..........
.. ... .. ... .. .
... ... .
... ...
........ .
... .. . .
.. .. .. . ... .
.. ...
..
..
. ...
....... ....... . . .
. .
..
.
. . .
. ...
..
... . .
....... ....... .... ... .... .... .... ... ...
.
... ...
. ...
.....
........ ........ .... .... .... .... .... ... .... ... ..
... .. . ... ...
. ...
.
........ ... ... ... .. ... ... .... ... .. . ... ... ...
...
........ . .
... .. .. .. .. . . ..... ... .. . ....... . ....... .
.
......... θ .
. . . .
. . .
. .. . . .
. .. . ...
.................................................................................................................................................................................................................................................................................................
A2 A0

Figure 6. A jagged line.

Solution: We will start by denoting the segments’ lengths by

|A0 A1 | = a1 , |A1 A2 | = a2 , |A2 A3 | = a3 , ...

Consider a sequence of similar right triangles A0 A1 A2 , A1 A2 A3 , ... They all

contain the angle θ with the side of length an+1 adjacent the angle θ, and side of
length an as the hypotenuse. Thus cos θ = an+1 /an where θ is a constant. That
implies the recurrence an+1 = an cos θ for all n ≥ 1. In particular, a2 = a1 cos θ,
and a3 = a2 cos θ = a1 cos2 θ. Similarly, we have an+1 = a1 cosn θ.
Recall that a1 = b, and so the segments’ lengths form a geometric sequence
an+1 = b cosn θ for all n ≥ 1. The common ratio an+1 /an = cos θ. Since 0 < θ <
π/2, 0 < cos θ < 1. This condition ensures us that the corresponding geometric
34

series, which gives the required length of the jagged line, converges, and we obtain:
|A0 A1 | + |A1 A2 | + |A2 A3 | + · · · = a1 + a2 + a3 + · · · =

b + b cos θ + b cos2 θ + · · · = b(1 + cos θ + cos2 θ + · · · ) =

∞
X b
b (cos θ)n =
n=0
1 − cos θ

Thus the length of the jagged line in an angle of size θ with first segment of
length b is b (1 − cos θ)−1 .
We invite the reader to check directly some special cases (e.g. θ = π/4 and
θ = π/3). What do you think will happen when θ is very close to 0 or π/2?

Problem 18. Two examples of a reduction to a geometric series.

18A. Find the infinite product
31/3 · 91/9 · 271/27 · 811/81 · · ·
Solution: Denote the product by N . Then N = 3M , where
1 2 3 4
M= + + + ··· .
3 9 27 81
Note that
M 1 2 3
= + + + ··· .
3 9 27 81
Thus
 
M 2 1 1 1 1 1 1 1 1
M− = M= + + + ··· = 1+ + + + ··· .
3 3 3 9 27 81 3 3 9 27
This geometric series in the parentheses with r = 1/3 has the sum 3/2. Therefore,
2M/3 = 1/2, and M = 3/4. Finally, the product is N = 33/4 .

18B. A number is written in the form of an “infinite” nested root with the
following periodic pattern v s
u r q
√
u
t
4 4 4 4 4...

Find the value of this number.

Solution: You may be familiar
√ with the following solution of this problem. Let
2
the
√ number be x. We have 4x = x. Squaring both sides we get 4x = x . Since
4 > 0 and the sequence of finite nested roots increases, we need x > 0. Thus
x = 4.
s r However, one may also use geometric series to solve this question. Note that
√
q p
4 4 4 4 4... = 4M , where M = 1/2 + (1/2)2 + (1/2)3 + ... + (1/2)k + ....
Using the formula for the sum of geometric series with first term 1/2 and common
 35

1/2
ratio r = 1/2, we obtain M = = 1. Thus, the value is 41 = 4.
1 − 1/2
Problem 19. Conversion of a repeating decimal to fractional form.
Convert 0.543 = 0.543543543 . . . into a fraction.
Solution:

0.543543543 . . . = 543 · (0.001001001 . . .)

= 543 · (0.001 + 0.000001 + 0.000000001 + · · · )
∞  n
X 1
= 543 ·
n=1
1000
∞
! !
X  1 n
= 543 · −1
n=0
1000
 
1
= 543 · 1 − 1
1 − 1000
   
1000 1 543
= 543 · − 1 = 543 · = .
999 999 999

Problem 20. Consider an isoceles 4A0 B0 C with angles A c0 = Bc0 , C

b = α,
and side |A0 B0 | = 1. Consider an infinite sequence of circles approaching angle C
and touching each other and the sides of the triangle (refer to the figure). Find
the total area occupied by the circles in terms of the angle α ∈ (0, π).

C
....
....
.......
.............
......................
. .
... .....
... ..
.... ....
.. ................. ..
......... ..........
.... ...
... ...
.... ....
.. .... .... ..
... ..................................... ....
......... .......
....
..
... ...
... ...
...... ..
...... .....
... .... . ... ...
.. ..
... ............. ............. ...
. .... . .
.. ........... ......................... ...
.. ....... ...... ...
.. ...... ......
.....
........ .....
....
.. ...
..
.. ...
..... .....
....... .....
... ... ... ..
... .... .. ....
.
... ... . .. ...
.
..
.... ... ..
...... ... ..
.. ........ ...... ..
.. .. ........ ..
A1......................................................................................................................................................................B1
... ............ ........
D1 ....... ....
... ........ ...... ..
. . .... ..
.. .... .... ..
.. ... ......
....
.... .....
...... ....
...
.
. ..
.... ...
..
.... ....
....... 0 ......
....... ... ..
... ... ........... .....
. ...
.... .... ... ....
... ... ..... ...
. ... ..
. ...
... ... .... . . ..
. .
. . ....... ..... .. ....
.. .... ...... ..... ...
..
.. ... ............ ...
..
.. ... ....... ... r0 ...
..
..
... .
. .
.
..
. . ... ..
..... ...... .... .
.. ..
... .
... .. . ...... .... ...... ..
... ............ .......
........ .. . ..... . .
..
. ..
..
............ β ...... .
.
. .. . . . ...
.... ...... ..
................................................................................................................................................................................................................
A0 D0 B0

Figure 7. Circles in an isosceles triangle.

36

Solution: Note that the first circle of the sequence is inscribed in 4A0 B0 C, and
the k th circle of the sequence is inscribed in a triangle which is similar to 4A0 B0 C.
We claim that the coefficient of similarity (see Appendix B.4) is (tan β)2k , where
β = (π − α)/4, and will show that now.
Let CD0 be the altitude, and let O be the incentre of 4A0 B0 C. Angle
CA0 B0 = (π − α)/2 = 2β and from 4A0 CD0 which is right angled, we obtain
\
the length |CD0 | = (1/2) tan(2β). From 4A0 OD0 which is also right angled, we
get the inradius r0 = |OD0 | = (1/2) tan β. Thus, the diameter of the first circle is
tan β.
Denote the point of tangency of the first and the second circles by D1 . Draw
a line parallel to the side A0 B0 through D1 . Denote the points of intersection of
the line with the sides A0 C and B0 C by A1 and B1 respectively. The second
circle is inscribed in 4A1 B1 C. 4A1 B1 C is similar to 4A0 B0 C with coefficient
(tan β)2 which follows from calculating the ratio of their heights (see Appendix B
for double angle formulas):
|CD1 | |CD0 | − 2r0 2 tan β
= =1− = (tan β)2 .
|CD0 | |CD0 | tan(2β)
By the same process it can be shown that the third circle of the sequence
is inscribed in 4A2 B2 C which is similar to 4A1 B1 C with coefficient (tan β)2 ,
and thus similar to 4A0 B0 C with coefficient (tan β)4 . In fact, for each circle,
the corresponding circumscribed triangle Ak Bk C is similar to the previous one
Ak−1 Bk−1 C with coefficient of similarity (tan β)2 .
From the coefficients of similarity, we know the ratio of the radii of inscribed
circles. Let rk be the inradius of Ak Bk C. Then
r1 r2 rk
= (tan β)2 , = (tan β)4 , = (tan β)2k , for all k ≥ 1.
r0 r0 r0
From this we immediately obtain the total area of the circles
∞ ∞ ∞ ∞
!
2
X
2 2
X rk2 2
X rk2 2
X πr02
πr0 + πrk = πr0 1 + 2 = πr0 2 = πr0 (tan β)4k = .
r0 r0 1 − (tan β)4
k=1 k=1 k=0 k=0

tan β π tan2 β
Recall that r0 = , so the total area occupied by the circles is ,
2 4 (1 − tan4 β)
where β = (π − α)/4. Note that, if the triangle is equilateral, then
π π 3π
α = , β = , and the total area of the circles is .
3 6 32
Problem 21a. The Cantor Set. The Cantor set, named after Georg
Cantor is constructed as follows. We start with the closed interval [0,1] which
clearly has length 1. We divide it into 3 equal intervals, with the middle interval
being an open
interval (it does not contain its endpoints). We remove the middle
interval 13 , 32 , and this leaves the two intervals 0, 13 , and 32 , 1 . We take each of
  

these intervals, divide them into 3 identical sized intervals, and remove the middle
open interval again. We repeat this process forever. The Cantor set is the set of
all real numbers which remain after all the intervals have been removed.
 37

The problem is to show that the total length of all the intervals removed
is 1, which is equal to the length of the original interval. (Despite this fact, the
Cantor set contains infinitely many numbers!)
1
Solution: The first interval removed has length . There are now 2 other intervals
3
1
remaining, each with length . We remove an interval from each of the remaining
3
1 1 1
ones, each interval removed has length · = . There are now 4 intervals
3 3 9
1
remaining. Each of them has length . We remove an interval from each of them.
9
1 1 1
Each interval removed has length · = . Therefore, if we repeat this process,
3 9 27
the total length of intervals removed is:
1 1 1 1 1 1
+2· +4· + ··· = 20 · + 2 1 · 2 + 22 · 3 + · · ·
3 9 27 31 3 3
∞ ∞  n
X 2n 1X 2 1 1
= n+1
= = · 2 = 1.
n=0
3 3 n=0
3 3 1 − 3

Thus, as desired, the total length of the intervals removed is 1.

What are the numbers that are left in the Cantor set? Some examples are:
1 2 1 2 7 8 1 1 1
0, , , , , , , , , .
3 3 9 9 9 9 27 81 243
If we look carefully at the procedure which constructs the set we find that a number
is in the Cantor set if and only if it is the sum of a series
∞
a1 a2 a3 X ak
1
+ 2 + 3 + ··· = , where ak is either 0 or 2.
3 3 3 3k
k=1

8 2 2
The series is infinite in some cases and in others not. For example, = + 2
9 3 3
1 ∞ 2
P
corresponds to a1 = a2 = 2, ak = 0 for k ≥ 3. However, = is represented
3 k=2 3k
by an infinite geometric series and corresponds to a1 = 0, ak = 2 for k ≥ 2. The
removal of the middle third of each interval corresponds to the condition ak 6= 1 in
the representation above. The Cantor set includes the endpoints of each removed
open interval, but not only these points. Even if we had removed the middle
closed interval in each step instead of removing the open interval, the set of points
remaining would have been infinite.

Problem 21b. The Sierpinski Carpet. The Sierpinski carpet is very

similar to the Cantor set. It can be viewed as being a 2-dimensional counterpart
of the Cantor set. It is constructed as follows. We start with a square of side length
38

1. We partition this square into 9 smaller squares of equal area, and remove the
interior of the middle square. Now, we take each of the remaining 8 squares and
partition them into 9 smaller squares. We again remove the interior of the middle
square in each one. We take each of the 8 · 8 = 64 remaining squares and partition
them, and again remove the interior of the middle square. We continue the process
forever. Show that the total area of the removed squares is 1, which implies that
the Sierpinski carpet has an area 0.

Figure 8. The Sierpinski Carpet.

1
Solution: The first square removed has area 2 . Then we remove 8 more squares,
 2 3
1 1
each with area = 4 . Now we remove 64 more squares, each with area
32 3
 2
1 1
= 6 . We continue this process forever. The total area of the removed
33 3
squares is:
1 1 1 1 1 1
2
+ 8 · 4 + 64 · 6 + · · · = 80 · + 81 · 2 + 82 · 3 + · · ·
3 3 3 9 9 9
∞ ∞  n
X 8n 1X 8 1 1
= n+1
= = · 8 = 1.
n=0
9 9 n=0 9 9 1− 9

Thus, as desired, the total area of the removed squares is 1. Just as in the Cantor
set, even though it may appear that there is nothing remaining, there are infinitely
many points remaining. In particular, if C is the Cantor set, then the sets4 C ×[0, 1]
and [0, 1] × C are both subsets of the Sierpinski carpet.

Problem 22. The Koch Snowflake. The Koch snowflake is constructed

as follows. Start with an equilateral triangle of side length 1. Then, partition
each side into 3 equal portions. On the middle portion of each side, place another
equilateral triangle, with side length equal to the length of that portion. Now, do
the same thing on each side of the resulting figure and continue indefinitely. The
figure that appears as the limit of these constructions is the Koch Snowflake.
4 Here A × B denotes the Cartesian product of two sets A and B, that is, the set of all pairs

(a, b), where a ∈ A and b ∈ B.

 39

.. .. .
...... ...... ......
... ... ... ... ............... ...............
... ...... ... ...... ... .
... ... ... ... .... .
. ..... .
... ... ... ... ... . . ... ..........
... ... ................................... ................................. ............... ................ ............ ............
... ... ... ... ... ..
..... ...
... ... ... . . .....
... ... ...
... ... ..
...............
...
.... ...
... ... ....
. ... .............
... ... ... .
.... . .
....
.. ... ... ... ... ...
.... ... ... ... ............... .............
.... ... ... ... ... .
.
... ...
... ..... ...
... ... .....
. . . ...
.................................................................................................. .................................... ..
................................. ............... ............... ................. ..............
... ... ...... ... .. ....
... . .. . ....
... ... .
. ... .
... ..
... ... .............. ..............
..... ......
...

Figure 9. The Koch Snowflake.

If you were to calculate the area and perimeter for the first few steps, you will
notice that they are both increasing with each step. There is one major difference
though. The sequence of areas is bounded because each figure can actually be
inscribed inside the same circle as the original triangle. Thus, the sequence of areas
is bounded from above √ and monotonically increasing and is therefore convergent,
2 3
the resulting area is . The same is not true for the sequence representing
5
the perimeter. That sequence increases without bound, and thus, the perimeter is
infinite. You should try to show both of these facts using geometric series.

Problem 23. The Four Mice Problem.

Four mice are placed initially at the corners A, B, C, and D of a square
room. Each mouse runs directly towards the mouse on its right, continuously
changing direction as the mouse on the right moves on its path. All four mice run
at the same constant speed, and they continue running, spiralling inwards, until
they all meet at the centre of the square. The side length of the room is 1 unit.
What is the length of the path run by each mouse?
This problem can be found under various names such as the turtle, dog,
beetle, or mice problem. Some Internet sites also provide animations and variations
(see e.g. http://mathworld.wolfram.com/MiceProblem.html). There are many
approaches to solving it. Here we give a solution which relies on the geometric
sum.

B........................................................................C 1
......................................................................
C
... .... ....
... ... .......
... .... .... ....
... .... .... ....
... ..... .... ...
...
....
. ....
. ....
... ... ....
... .. .
C2................................................................................D ....
... .
. 2 ....
... .... ....
. . ......
.
. ....
... ... .... .. .....
.
... .. .
. .... ...
... ...... ...
... .
.
.
. ...... .....
... .... . .
. .
... .... ... .
.
. .... ...
B1 ......... . . .
... .
. ..... D1
... ..... ......
.... ...... ... .... .... ...
... .... ... .
. ...... ....
.... .... . . . .
... .... ..... .... ....... .....
.. .... .. ... ... ....
.... ..
... ................................................................................... ....
..... ....
. . .
. A.. ...
...
...
B 2 ....
..
... ... 2 ....
.
..... .. ....
.
.... .... ..... ....
.... .... ..
... ...
. .
...........................................................................................................................................................
A A1 D

Figure 10. An auxiliary Four Mice Problem.

First, we consider an auxiliary problem. In this problem the difference from

the previous one is that the mice change direction at discrete intervals instead of
40

continuously. The mouse starting from A initially runs half of the length of side
AD ending at A1 , and similarly for the other three mice. So at the end of the first
interval the four mice have reached A1 , B1 , C1 , D1 respectively. Now the mice
change direction. The mouse at A1 runs half the length of the side A1 D1 ending
at A2 , and similarly for the other three mice. At the end of the second interval
the four mice have reached A2 , B2 , C2 , D2 respectively (see Figure 10). The mice
continue moving in this fashion. After infinitely many iterations they meet at the
centre of the square. The question is the same as before: what is the total length
of the path run by each mouse?
Solution: Observe that√if the side length of ABCD is 1 unit then the side
length√of A1 , B1 , C1 , D1 is 1/ 2, the side length of the next square A2 , B2 , C√2 , D2
is (1/ 2)2 , and the side length of the n-th square An , Bn , Cn , Dn is (1/ 2)n .
Since each mouse runs a half of the side of each square in this infinite sequence of
squares, the path length is
  n   
1 1 1 1 1 2
1 + √ + ··· + √ + ··· = √ = √ .
2 2 2 2 1 − 1/ 2 2− 2
Next, we modify the auxilary problem as follows. Let k ≥ 2. As before each
of the four mice is pursuing the mouse on its right, each starting from one of the
four corners A, B, C, D of the room. Again, the mice change direction at discrete
intervals. This time, at the first stage, the mouse which starts from A runs 1/k
along the side towards D stopping at A1 which is 1/k from A. Similarly, for the
other three mice. At the second stage the mouse at A1 runs 1/k of the side A1 D1
towards D1 stopping at A2 which is (1/k) · A1 D1 from A1 . Similarly, for the other
three mice. The question is the same as before: what is the total length of the
path run by each mouse?
Solution: Note that for k = 2 we have the problem that we already solved.
In general, by the Pythagorean theorem each square An+1 Bn+1 Cn+1 Dn+1 in the
sequence is √
k 2 − 2k + 2
q
2 2
(1/k) + [1 − (1/k)] =
k
times the size of the previous square An Bn Cn Dn . Since each mouse runs 1/k of
the side of each square in the sequence, the total path length is:
∞ √ !n
1X k 2 − 2k + 2 1
= √
k n=0 k 2
k − k − 2k + 2
In the limit as k → ∞, the paths run by the mice become the paths run
when each mouse is always running directly towards the mouse on its right — as
in the original Four Mice Problem. Thus the solution to the Four Mice Problem
is
1
L = lim √ .
k→∞ k − 2
k − 2k + 2
√
To evaluate the limit we multiply numerator and denominator by k+ k 2 − 2k + 2:
√ √
1 k + k 2 − 2k + 2 k + k 2 − 2k + 2
√ = 2 =
k − k 2 − 2k + 2 k − (k 2 − 2k + 2) 2k − 2
 41

Now we see that

√ p
k + k 2 − 2k + 2 1 + 1 − (2/k) + (2/k 2 ) dividing top and
lim = lim bottom by k
k→∞ 2k − 2 k→∞ 2 − (2/k)
p √
limk→∞ 1 + 1 − (2/k) + (2/k 2 ) 1+ 1
= = = 1.
limk→∞ 2 − (2/k) 2
Therefore L = 1 and the path of each mouse has length 1.

§2.4 Telescoping
In this section we will introduce a method which can help to find the sum
of a given series. This method only works for a certain class of problems, and it
requires more creativity than the use of the geometric formula. Nevertheless, it is
a very important method because it is one of the very few ways to find the sum
of infinite as well as finite summations.

§2.4.1. Finding telescoping sums

The main idea of the method is to change the appearance of the sum and
look for similar terms to cancel with each other. We will first illustrate the idea
with a pair of finite summations.
Problem 24. Give an explicit formula for the following sum.

1 · 1! + 2 · 2! + 3 · 3! + · · · + n · n!

Solution. Use k · k! = (k + 1) · k! − k! = (k + 1)! − k!. Then

1 · 1! + 2 · 2! + 3 · 3! + · · · + n · n! = (2! − 1!) + (3! − 2!) + (4! − 3!) + · · · + ((n + 1)! − n!) .

After cancellations we obtain 1 · 1! + 2 · 2! + 3 · 3! + · · · + n · n! = (n + 1)! − 1 .

Problem 25. Find the sum

1 1 1 1
+ + + ··· + .
2·3 3·4 4·5 1000 · 1001
Solution: Of course the sum could be found by adding all the terms manually, but
it can be done faster and easier by telescoping.
Note that
1 1 1 1 1 1 1 1 1
= − , = − , ... , = − .
2·3 2 3 3·4 3 4 1000 · 1001 1000 1001
Then we rewrite the sum as
1 1 1 1 1 1 1 1
− + − + − + ··· + − . (12)
2 3 3 4 4 5 1000 1001
Many terms cancel each other. In fact, all terms other than the first and the last
42

1 1 999
cancel. Thus, the sum is − = .
2 1001 2002
Problem 26. Consider a sequence of triangles the n-th having vertices:
   
1 1
(1, 2), ,0 , ,0 , n = 2, 3, 4, . . . , 1000.
n n+1
Find their total area.
2 (1,2)

0 1
1001
1/5 1/4 1/3 1/2 1

Figure 11. A sequence of triangles.

Solution: Each of the triangles has (1, 2) as one vertex and two other vertices on
the x-axis. Thus, each of the triangles has height 2. The length of base of the n-th
triangle on the x-axis is n−1 − (n + 1)−1 . Since the area of a triangle is a half of
the height times base, the area of the n-th triangle is the same as the length of its
base. To find the total area we need to sum the areas of all 999 triangles. This,
in fact, leads to the same telescoping procedure (12) as in the previous problem.
After cancellations we again obtain 1/2 − 1/1001 = 999/2002.
However, the
problem has a geometric solution. The segment joining (1, 2)
and (n + 1)−1 , 0 is a common side of the n-th and (n + 1)-th triangles, for
2 ≤ n ≤ 999. Thus the total area is the area of the big triangle with vertices (1, 2),
(1/2, 0), (1/1001, 0) which has height 2 and base length 1/2 − 1/1001 = 999/2002.
So the area of the big triangle is 999/2002.
This problem is useful for us in two ways. First it provides a geometric
interpretation of the telescoping procedure, and second, we can generalize this
result to the case of arbitrary N > 2 (not only N = 1000 as in the two problems
above):
1 1 1 1 1 1
+ + + ··· + = − . (13)
2·3 3·4 4·5 N · (N + 1) 2 N +1
Both algebraic and geometric (using triangles’ area) solutions will work.
The advantage of having a general formula is that we can now ask ourselves:
What will happen if we let N go to infinity?
Following the geometric interpretation, the areas of the N − 1 triangles add
up to the area of the triangle with vertices (1, 2), (1/2, 0), (1/(N + 1), 0). In the
limit N → ∞, the vertices of the resulting triangle become (1, 2), (1/2, 0), (0, 0);
the area of the triangle is 1/2, and this is the sum of the infinite series:
∞
X 1 1
= . (14)
n=2
n(n + 1) 2
 43

When we find the area (13) of the triangle with vertices (1, 2), (1/2, 0),
(1/(N + 1), 0), we, in fact, find a partial sum of the series. Thus, finding the total
sum this way agrees with our previous concept of convergence of a series in terms
of the limit of the sequence of its partial sums (see Definition 3):
 
1 1 1
S = lim − = .
N →∞ 2 N +1 2

There is one more geometrical interpretation of this formula which is provided

by the following example.

Problem 27. Figure 11 shows two circles α and β of radius 1 that touch
at X. τ is a common tangent line and γ1 is the unique circle touching α, β, and
τ ; γ2 is the unique circle touching α, β, and γ1 . For each n ≥ 1, let γn+1 be the
unique circle touching α, β, and γn . Show that the diameter dn of γn is equal to
1
dn = , n ≥ 1.
n(n + 1)

A X B
1 1

γ3
1 1
γ2

r1
α O1
β
γ1
τ

Figure 12. A sequence of circles.

Solution: We will start by finding the radius r1 of γ1 . Let A and O1 be centres of

cercles α and γ1 respectively.
Draw a right triangle with vertices O1 , A, and X. The legs of this right
triangle have lengths XO1 = 1 − r1 and AX = 1. The hypotenuse of this triangle
has length AO1 = 1 + r1 . We can use the Pythagorean theorem to obtain:

(1 − r1 )2 + 1 = (1 + r1 )2
1 − 2r1 + r12 + 1 = 1 + 2r1 + r12
1 = 4r1
1
r1 =
4
Thus, the diameted of γ1 is d1 = 2r1 = 1/2.
44

The following statement establishes a relationship between diameters of two

touching circles in a slightly more general situation.

Lemma. Let α, β be circles of radius 1, with centres A, B respectively,

which touch at X. Let γ be a circle which touches both α and β externally and δ
be the unique circle which touches α, β, and γ externally as in Figure 12. If the
diameter of γ is 1/[c(c+1)], where c > 0, then the diameter of δ is 1/[(c+1)(c+2)].

X
A 1 1 B

1 P
δ

r
r
α O β
γ

Figure 13. Touching circles

Note. Since every d > 0 can be written uniquely in the form 1/[c(c + 1)]
with c > 0 the lemma applies to every circle γ touching α and β externally. It is
convenient to write the diameter of γ in the form 1/[c(c + 1)] because it gives us
a neat way of expressing the relationship between the two diameters.

Proof: Let O, P be the centres of γ, δ respectively, and r, s their radii.

Applying the Pythagorean theorem to 4AXP gives
(r + 1)2 = XO2 + 12 .
p p
Therefore XO = (r + 1)2 − 1 and similarly XP = (s + 1)2 − 1. Since OP =
XO − XP and γ and δ touch, we have
p p
(r + 1)2 − 1 − (s + 1)2 − 1 = r + s
which is the same as
p p
(r + 1)2 − 1 − r = (s + 1)2 − 1 + s (15)
Substituting r = 1/[2c(c + 1)], the left hand side of (15) becomes:
s 2
1 1
+1 −1 −
2c(c + 1) 2c(c + 1)
1 p  1
= 1 + 4c(c + 1) − 1 =
2c(c + 1) c+1
 45

p
Substituting in (15) gives the equation 1/(c + 1) = (s + 1)2 − 1 + s for s.
Subtracting s from both sides and then squaring both sides we get
1 2s
− + s2 = s2 + 2s
(c + 1)2 c+1
The unique solution is s = 1/[2(c + 1)(c + 2)] and so the diameter of δ is
1/[(c + 1)(c + 2)] as claimed.

The result stated in Problem 27 follows by induction on n. We already know

that d1 = 1/2. The induction step is provided by the lemma because, for each n,
γn+1 is related to γn in exactly the same way as δ to γ in the lemma. Consequently
we have
1 1 1 1
d1 = , d2 = , d3 = , · · · , dn = ,···
1·2 2·3 3·4 n(n + 1)
As one can see from the Figure 13, the infinite sum of the diameters dn is
equal to the radius of the two largest circles, which is 1. Thus we obtain
∞ ∞
X X 1
dn = = 1.
n=1 n=1
n(n + 1)

Note that this formula easily follows from (14) by adding 1/2 to each side.

Problem 28. Find the sum

∞
X k 2 + 3k + 1
.
(k + 2)!
k=0

Solution: Note that k 2 + 3k + 1 = (k 2 + 3k + 2) − 1 = (k + 1)(k + 2) − 1. Also,

recall that k! = 1 · 2 · 3 · · · (k − 1) · k. Thus, (k + 2)! = k!(k + 1)(k + 2). With this
in hand we can rewrite the general term as
k 2 + 3k + 1 (k 2 + 3k + 2) − 1 (k + 1)(k + 2) 1 1 1
= = − = − .
(k + 2)! (k + 2)! (k + 2)! (k + 2)! k! (k + 2)!
Using the convention that 0! = 1 the N -th partial sum is now
1 1 1 1 1 1 1 1
SN = − + − + − ··· + − .
0! 2! 1! 3! 2! 4! N ! (N + 2)!
After cancellations we obtain
1 1
SN = 1 + 1 − − .
(N + 1)! (N + 2)!
Finally, we take the limit as N → ∞ and see that SN → 2.
Thus, the answer is
∞
X k 2 + 3k + 1
= 2.
(k + 2)!
k=0
46

The next two examples refer to the Fibonacci numbers Fn . Telescoping

rearrangements of the series are possible due to the identities which we derived in
Section 1.9.1.

Problem 29. Show that:

∞
X 1
= 1
n=2
F n−1 · Fn+1

1 1 1
Solution: Use the identity − = from Problem 13
Fn−1 · Fn Fn · Fn+1 Fn−1 · Fn+1
to write the general term as a difference of two others
∞ ∞  
X 1 X 1 1
= − .
F
n=2 n−1
· Fn+1 n=2
Fn−1 · Fn Fn · Fn+1

1
Denoting by an for n ≥ 2 observe that the series has the form
P∞ Fn−1 · Fn
n=2 (an − an+1 ).
Therefore the partial sum is
N
X 1 1
SN = = 1 − aN +1 = 1 − .
n=2
F n−1 · Fn+1 FN · FN +1

We now can take the limit of the partial sum to get the desired result:
∞  
X 1 1
= lim 1 − = 1.
F
n=2 n−1
· Fn+1 N →∞ FN · FN +1

Problem 30. Show that:

∞
X Fn
= 2.
F
n=2 n−1
· Fn+1

Solution: We start by using the definition of the Fibonacci sequence to change the
form of Fn in the numerator, and then write the expression as a difference of two
terms:
Fn Fn+1 − Fn−1 1 1
an = = = −
Fn−1 · Fn+1 Fn−1 · Fn+1 Fn−1 Fn+1
Writing out the sum of the first N terms, we get:
         
1 1 1 1 1 1 1 1
SN = 1 − + 1− + − + − + ··· + − .
2 3 2 5 3 8 FN −1 FN +1
 47

For each n ≥ 1, the negative portion of an will always cancel with the positive
portion of an+2 . Thus, the N -th partial sum is:

1 1 1 1
SN = 1 + 1 − − =2− −
FN FN +1 FN FN +1

Taking the limit, we get the desired result:

∞  
X Fn 1 1
= lim 2 − − = 2.
F
n=2 n−1
· Fn+1 N →∞ FN FN +1
Note that even if you manage to find a rearrangement which allows you to
cancel terms, an infinite series may still be divergent. Remember, to prove that
a series converges you need to show that the sequence of partial sums converges.
Otherwise, the series is divergent. A few examples where the telescoping technique
does not lead to a convergent result are presented below.

Problem 31. Explain why the series

∞
X
(−1)n = −1 + 1 − 1 + 1 − 1 + 1 − · · ·
n=1

is divergent.
Solution: It looks as though the cancellations take place again and the sum must
be 0. We have to work out this example by the definition, and thus find the limit
of the sequence of partial sums. We obtain
S1 = −1, S2 = 0, S3 = −1, S4 = 0, etc.
We observe that all the terms of the sequence of partial sums with even
indices are equal to 0 while those with odd indices are equal to −1. Thus, the
sequence alternates between two numbers and therefore does not have a limit. The
series is therefore divergent.

Problem 32. Is the following series convergent or divergent?

∞  
X n
ln
n=2
n+1
 
n
Solution: Since ln = ln n − ln(n + 1), the N -th partial sum is
n+1
SN = (ln 2 − ln 3) + (ln 3 − ln 4) + (ln 4 − ln 5) + · · · + (ln N − ln(N + 1)).
After cancellations we have SN = ln 2 − ln(N + 1).
Notice that so far everything looks similar to Example 2, but that does not
mean that we will have a similar result. The calculation of the limit gives the final
answer. We have
lim SN = ln 2 − lim ln(N + 1) = −∞.
N →∞ N →∞
48

Thus, the series is divergent.

Problem 33. Is the following series convergent or divergent?

∞
X 1
√ √
n=0
n+1+ n
√ √ √ √
Solution: First we use ( n + 1 + n)( n + 1 − n) = n + 1 − n = 1 to write
1 √ √
√ √ = n + 1 − n.
n+1+ n
Now the N -th partial sum is
√ √ √ √ √ √ √ √
SN = 1 − 0 + 2 − 1 + 3 − 2 + · · · + N + 1 − N .
√
After cancellations we have SN = N + 1. Here again,
√ the calculation of the limit
gives the final answer. We have lim SN = lim N + 1 = ∞. Thus, the series
N →∞ N →∞
is divergent.

From these examples we see that there are many ways to make the telescoping
method work. In the next subsection we show that the method of partial fractions
is a very valuable tool when finding the sum of a telescoping series whose n-th
term is a rational function of n.

§2.4.2 Partial fractions

In this subsection we deal with sequences whose n-th term is a rational
expression of n. That is, an = P (n)/Q(n), where P and Q are polynomials with
the degree of P less than that of Q. We know that any polynomial in n can
be factored as a product of linear terms n − d and irreducible quadratic terms
an2 + bn + c with b2 < 4ac (recall that if b2 ≥ 4ac then such a quadratic term can
be factored into two linear terms, and thus it is called reducible). The method of
partial fractions aims to rewrite the ratio an = P (n)/Q(n) as a sum of fractions
whose denominators are factors of Q(n). The trick from Problems 25 and 26
(section 2.4.1) uses such a representation:
1 1 1
= − .
(n + 1)(n + 2) (n + 1) (n + 2)
The goal of this subsection is to convert this “trick” into a rigorous method. The
following example shows how the method works.

Problem 34. Find the partial fraction representation for the rational
1
expression 2 .
n + 3n + 2
Solution: We know that n2 + 3n + 2 = (n + 1)(n + 2); let us asume that there exist
real numbers A and B such that
1 1 A B
2
= = + .
n + 3n + 2 (n + 1)(n + 2) n+1 n+2
 49

Now we need to find the values of A and B. This task is called equating coefficients.
Multiplying both sides of the last equation by (n + 1)(n + 2) we obtain

1 = A(n + 2) + B(n + 1) = (A + B)n + (2A + B).

As you can see, the coefficient of n on the left side of this equation is zero, so
A + B = 0. Also, the constant on the left is 1 while the constant on the right is
2A + B. So 2A + B = 1. We now have two equations and two unknowns so we
can solve the equations for A and B. The first equation yields A = −B. Then the
second equation gives A = 1 and B = −1. Thus,
1 1 1
= − .
n2 + 3n + 2 n+1 n+2

The numerators in the partial fractions will not always be constants. They
were only constants in the previous example because the denominators were linear
functions of n.
However, for a denominator of degree d > 1 the corresponding numerator is
a polynomial in n of degree d − 1. For example, if the denominator has factors
(n + 1)(n2 + 1) then the corresponding numerators are A and Bn + C.

Problem 35. Find the partial fraction decomposition for

n2 + 2n + 2
.
(n + 2)(n2 + 3n + 3)

Solution: Since n2 + 3n + 3 is irreducible (because b2 − 4ac = 9 − 12 = −3 < 0),

it will be one of the denominators of the partial fractions, while n + 2 is the other
one. Thus, we have

n2 + 2n + 2 A Bn + C
= + .
(n + 2)(n2 + 3n + 3) n + 2 n2 + 3n + 3

Multiplying both sides by (n + 2)(n2 + 3n + 3), we get

n2 + 2n + 2 = A(n2 + 3n + 3) + (Bn + C)(n + 2)

= An2 + 3An + 3A + Bn2 + 2Bn + Cn + 2C
= (A + B)n2 + (3A + 2B + C)n + (3A + 2C).

Now, by equating coefficients, we just have to solve the system

A + B = 1, 3A + 2B + C = 2, 3A + 2C = 2.

This can be easily solved to find A = 2, B = −1, and C = −2. Therefore

n2 + 2n + 2 2 n+2
= − 2 .
(n + 2)(n2 + 3n + 3) n + 2 n + 3n + 3
50

Sometimes partial fractions and telescoping work well together. Here are
two such examples.

Problem 36. Find

∞
X k
4 + k2 + 1
.
n=1
k

Solution: It is easy to see that the denominator can be factored using a quick trick

k 4 + k 2 + 1 = (k 4 + 2k 2 + 1) − k 2 = (k 2 + 1)2 − k 2

which is a difference of squares, so k 4 + k 2 + 1 = (k 2 + k + 1)(k 2 − k + 1). Now we

can find the partial fractions for the general term. Let for some A, B, C, D
k k Ak + B Ck + D
= 2 = 2 + .
k4 + k2 + 1 (k + k + 1)(k 2 − k + 1) k + k + 1 k2 − k + 1

Multiplying both sides by k 4 + k 2 + 1 we obtain

k = (Ak + B)(k 2 − k + 1) + (Ck + D)(k 2 + k + 1)

= (A + C)k 3 + (−A + B + C + D)k 2 + (A − B + C + D)k + (B + D).

Solving corresponding linear system yields: A = 0, B = −1/2, C = 0, and

D = 1/2. Thus, the general term is now equal to:
 
k 1 1 1
ak = 4 = −
k + k2 + 1 2 k2 − k + 1 k2 + k + 1
 
1 1 1
= −
2 k 2 − k + 1 (k + 1)2 − (k + 1) + 1
When we write each of the first n terms in this fashion and add them together the
negative portions of ak cancels with the positive portion of ak+1 for all 1 ≤ k ≤ n
leaving
 
1 1 1 1
Sn = − = 1− 2 .
2(11 − 1 + 1) 2((n + 1)2 − (n + 1) + 1) 2 n +n+1
Therefore
∞
X 1 1
= lim Sn = .
k 4 + k 2 + 1 n→∞ 2
k=1

Comment: The critical point in a telescoping sum is to express ak in the form

f (k) − f (k + 1).
 51

Problem 37. Find the sum:

1 1 1
+ + + ···
1·3 3·5 5·7
Solution: It is easy to see that for k ≥ 1 the k-th term
   
1 1 1 1 1 1 1
ak = = − = − .
(2k − 1)(2k + 1) 2 2k − 1 2k + 1 2 2k − 1 2(k + 1) − 1
When we write each of the first n terms in this fashion and add them together the
negative portions of ak cancels with the positive portion of ak+1 for all 1 ≤ k ≤ n
leaving
1 1 1 1
Sn = − = − .
2(2 · 1 − 1) 2(2(n + 1) − 1) 2 4n + 2
Taking the limit as n → ∞ the desired sum is
 
1 1 1 1 1 1
+ + + · · · = lim − = .
1·3 3·5 5·7 n→∞ 2 4n + 2 2

§2.5 Harmonic series and paradox of shifted cards

2
Recall that the harmonic mean of two numbers a and b is given by .
a−1 + b−1
The series for which each term is equal to the harmonic mean of its neighbors is
called the harmonic series, and it has the very simple form
∞
X 1 1 1 1 1
= 1 + + + + ··· .
n=1
n 2 3 4 5
In 1673, Gottfried Wilhelm Leibniz knew that this series was divergent. This
is not an obvious observation, because it diverges very slowly, and if one calculates
its partial sums it is hard to say whether they are bounded or not. Leonhard
Euler’s book Differential Calculus written in 1755, gives the following examples:
the sum of the first 1000 terms of the Harmonic series is approximately 7.485,
and the sum of first 1, 000, 000 terms of the series is approximately 14.393. Euler
himself found the numbers by a certain trick, not by direct summation. To have
the sum close to 100, we would have to take approximately 2143 terms.
Another interesting fact is that none of the partial sums for the harmonic
series is an integer.
To prove that the harmonic series is divergent we group the terms of the
series as follows
       
1 1 1 1 1 1 1 1 1 1 1
1+ + + + + + + + + ··· + + + ··· + +· · ·
2 3 4 5 6 7 8 9 16 17 32
Note that in the n-th set of parentheses, the last (and smallest) term is 1/2n , and
the number of terms is 2n−1 . Thus, the sum in each set of parentheses is greater
than 2n−1 /2n = 1/2. For instance,
1 1 1 1 1 1 1 1 1 1
+ >2· = , + + + >4· = ,
3 4 4 2 5 6 7 8 8 2
52

and so on. The total sum therefore is greater than an infinite sum of halves and
thus, the series is divergent.
This divergence helps explain an interesting effect — an experiment which
you can perform.
Place cards, say, from a standard deck, at the left edge of a desk so that
the card on top extends half its length beyond the card underneath it; the second
top card extends a quarter of its length beyond the card underneath it; the third,
1
fourth,..., n-th from the top, respectively, extends one-sixth, one-eighth,..., 2n of
its length beyond the card immediately underneath it. The bottom card rests on
the desk with its left edge lying along the left edge of the desk. This pile of cards
will not fall to the floor even though the top card extends far beyond the edge of
the desk.

..................................................................................................................................................
... .
.. .....
... ....
............................ ....
.... ........................................................................ ...
...
...
................................................ ....
................................................ ....
. ................................................
.... ............................................................................. ....
.... ... ... ....
.... ... ........................................................... .... ...
... .. ..
. .
... .... . ....
... . ...
.
. .. .. ....
. . .
.... ..... ..... .... .... ....
.... .... .... .... .... ....
.... ... ... .... .... ...
... .. ...
. ... ... ....
... . ...
.
. ... ... ....
. . . . .
.... ..... ..... .... .... ....
.... .... .... .... .... ...
....................................................................................................................................................

Figure 14. Paradox of the shifted cards.

Now we will explain this effect. Let N be the number of cards. We suppose
that the length of each card is 1, and that the origin of measurement is the left
edge of the top card.

Figure 15. Shifted cards in a coordinate plane.

Let xn be the coordinate of the left edge of the n-th card from the top. Thus
x1 = 0, x2 = 12 , x3 = 12 + 14 = 34 , and generally for n ≥ 2,
  n−1
1 1 1 1 1X1
xn = xn−1 + = 1 + + ··· + = .
2(n − 1) 2 2 n−1 2 k
k=1

The position of the centre of mass of the n-th card is at x̄n = xn + 21 .

 53

Since all the cards are assumed to be identical, the centre of mass ḡn of the
top n cards is located at the mean of the centres of mass of the individual cards,
so that
x̄1 + x̄2 + · · · + x̄n
ḡn = .
n
In the table below we give the values of xn , x̄n , and ḡn for small values of n.
1 1
n xn = xn−1 + 2(n−1) x̄n = xn + 2 ḡn = (x̄1 + · · · + x̄n )/n
1 1
1 0 2 2
1 3
2 2 1 4
1 1 3 5 11
3 2 + 4 = 4 4 12
3 1 11 17 25
4 4 + 6 = 12 12 24

We make some observations. Since the edge of the desk is located at xN and
since the harmonic series diverges, the top card in the pile can be arbitrarily far
to the left of the edge of the desk when sufficiently many cards are laid.
The second observation is that from the table ḡn = xn+1 for 1 ≤ n ≤ 4. We
prove by induction that it holds for all positive integers n. Suppose that it holds
up to the index k, that is ḡn = xn+1 for 1 ≤ n ≤ k. Then
(k + 1)ḡk+1 = x̄1 + · · · + x̄k+1 = (x̄1 + · · · x̄k ) + x̄k+1 = kḡk + x̄k+1 .
1
Now, use the assumption of the induction step and the relation x̄n = xn + 2 to
get  
1 1
(k + 1)ḡk+1 = kxk+1 + xk+1 + = (k + 1)xk+1 + .
2 2
1
Finally, use xk+2 = xk+1 + 2(k+1) to obtain
 
1 1
(k + 1)ḡk+1 = (k + 1) xk+2 − + = (k + 1)xk+2 .
2(k + 1) 2
This implies that ḡk+1 = xk+2 , and thus by induction ḡn = xn+1 for 1 ≤ n < N .
What is the significance of this result? It says that the centre of mass of the
top n cards lie directly on top of the left edge of (n + 1)-th card from the top, and
that the centre of mass of all the cards above the bottom card lying on the desk
is directly above the edge of the desk. Thus, even though the upper cards in the
stack extend far to the left of the desk, the stack will not fall to the floor.
The harmonic series is a member of a large class of series which are called
hyperharmonic. We will talk about them in a later section.

§2.6 Leibniz series and an unexpected effect of permutation

The series named after Gottfried Wilhelm Leibniz is a convergent one:
∞
X 1 1 1 1 1
(−1)n+1 = 1 − + − + − · · · = ln 2. (16)
n=1
n 2 3 4 5
54

Note that it differs from the divergent harmonic series only by the alternation
of the sign of its terms, and hence it is classified as an alternating series.
It is not hard to show that the series is convergent. In fact, any alternating
series
∞
X
(−1)n an , an > 0
n=1
whose general term an is both decreasing and tends to zero (i.e. an+1 ≤ an and
an → 0) is convergent. The partial sums of such a series jump above and below
its limit, getting closer as n increases.
It is a little more tricky to show that the series converges to ln 2. To see
this you will need to read a later section about power series representations of
functions and use formula (23) for the logarithmic function ln(x + 1) at x = 1.
The series is convergent, but it has an amazing property: by changing the
order of the terms, one can get different results for its sum! Series with this
property were called conditionally convergent by Georg Bernhard Riemann around
1854.
Let us show for this example that after a certain permutation of the terms,
the sum is equal to (3/2) ln 2.
First write our original statement
1 1 1 1 1 1 1
ln 2 = 1 −+ − + − + − + ···
2 3 4 5 6 7 8
Now divide both sides by 2 and fill in zeros for further convenience
ln 2 1 1 1 1 1
=0+ +0− +0+ +0− +0+ + ···
2 2 4 6 8 10
Now add the two formulas to obtain
3 ln 2 1 1 1 1 1
= 1 + 0 + − + + 0 + − + ···
2 3 2 5 7 4
Note that on the right side, you have the same terms as in the Leibniz series,
but written in a different order. On the left side, you have a different result!
An even more surprising result, known as Riemann’s Theorem, is as follows:
one can get any number by rearranging the terms of a conditionally convergent
series. Niels Henrik Abel was so upset by this statement that he once said that
conditionally convergent series are the devil’s production. Some more examples of
the devil’s production are series of the form
∞
X 1
(−1)n+1 , 0 < s < 1.
n=1
ns

In particular, for s = 1/2,

∞
X 1
(−1)n+1 √
n=1
n
is conditionally convergent.
 55

§2.7 Hyperharmonic series and the Riemann zeta function

The Riemann zeta function is a well known object in analysis, in particular
due to a problem that has remained open for more than a century and which is
known as the 8th Hilbert problem or the Riemann hypothesis about the zeroes of
the zeta function. Who knows, maybe you will witness or even participate in its
solving!
The most characteristic property of the zeta function is its representation
by either the infinite sum with respect to all natural numbers n, also known as
hyperharmonic series,
∞
X 1
ζ(s) = ,
n=1
ns
or an infinite product with respect to all prime numbers p,
Y  −1
1
ζ(s) = 1− s .
p
p prime

Either of these formulas may serve as a definition of the zeta function when the
corresponding expression converges.
Recall that a prime number is a natural number which has only two divisors:
one and itself. The fundamental significance of prime numbers is manifested in
the uniqueness (up to the order of the factors) of the representation of any natural
number as a product of powers of primes, like 72 = 32 · 23 . This fact is the basis of
the equivalence of the two representations of the zeta function given above. It also
explains why the study of the zeta function can have applications in number theory,
in particular for questions about the distribution of the prime numbers. The
equivalence of the two representations for the zeta function was first established
by Euler, who considered only real values of the argument s. Later, Riemann
introduced the function ζ(s) by means of another equation, valid for any value of
the argument s, including complex numbers. He showed that his definition agrees
with the one provided by the hyperharmonic series for all s, both real and complex,
for which the series converges.
Our goal in this section is to investigate the convergence of hyperharmonic
series for real values of s. It is clear thatPfor s = 0 and any s < 0, e.g. s =
∞
−1/2, s = −2, the hyperharmonic series n=1 n1s is divergent. The answer is
less obvious for positive s, but we know that for s = 1 it is the harmonic series
which diverges. In fact, the value s = 1 is the P delimiter between the convergent
∞
and divergent hyperharmonic series. That is, n=1 n1s is convergent for s > 1,
and divergent otherwise. For instance, it converges for s = 1.1 and diverges for
s = 1/2. (However, a hyperharmonic series with 0 < s ≤ 1 becomes conditionally
convergent with the insertion of an alternating sign (−1)n because an = n−s is
decreasing and tends to 0 as n → ∞.)
Let us first demonstrate the convergence of the series for ζ(2) given by the
hyperharmonic series for s = 2
∞
X 1
ζ(2) = .
n=1
n2
56

Consider the sequence of partial sums

1 1 1
Sn = 1 + 2
+ 2 + ··· + 2.
2 3 n
It is obvious that the sequence grows; that is, Sn < Sn+1 .
Let us show that the sequence is bounded above by 2.
Using the technique of telescoping sum from Section 4.2, for every n we have
1 1 1
Sn = 1+ + 2 + ··· + 2
22 3 n
1 1 1
≤ 1+ + + ··· +
1·2 2·3 (n − 1) · n
     
1 1 1 1 1 1
= 1+ − + − + ··· + −
1 2 2 3 n−1 n
1
= 2 − < 2.
n
This confirms that the partial sums Sn are indeed bounded above by 2. The
series representing ζ(2) does converge.
P∞
To prove convergence of n=1 n1s for all s > 1 one may observe that such
series have positive terms and are bounded from above by a convergent geometric
series, and therefore converge. In detail, we write
     
1 1 1 1 1 1 1 1
ζ(s) = 1 + + s + + s+ s+ s + + ··· + s + ···
2s 3 4s 5 6 7 8s 15
Here we group together all terms from the 2n -th to the (2n+1 − 1)-th position, for
n = 1, 2, 3... (just like we did to to show that the harmonic series diverges). For
example, in the first group we have the 2nd and 3rd terms, then the 4th to 7th
terms in the second group, the 8th to 15th terms in the third group, and so on.
Note that there are 2n terms in nth group. For instance, two in the first group,
four in the second, eight in the third and so on. Obviously, since s > 1 the larget
term in each group in the first one, 1/(2n )s . Thus, the nth group’s sum is less
than  n
n 1 1 1
2 × n s = n(s−1) = .
(2 ) 2 2s−1
∞
rn , where r = 1/2s−1 .
P
Thus the whole series is less then the geometric series
n=1
Now for s > 1 we have 0 < r < 1, and so the geometric series converges. But the
sum of our hyperharmonic
P∞ series is even less than the the sum of the geometric
series, therefore n=1 n1s for s > 1 converges as well.
Note that for s = 1 we get r = 1, and the above approach no longer works.
This by itself does not show the divergence for s = 1. But we have proved the
divergence earlier, showing that the harmonic series can be compared to an infinite
sum of constant terms (1/2), and thus its sum is infinite.
To find the actual value of the sum of a convergent hyperharmonic series is
a harder problem. For s = 2, it was solved by Euler in 1734. We show its value
here without a proof:
 57

∞
X 1 π2
ζ(2) = = . (17)
n=1
n2 6
It turns out that when the argument is an even number, the value of the zeta
function can be expressed using π. For example
∞ ∞
X 1 π4 X 1 π6
ζ(4) = = , ζ(6) = = .
n=1
n4 90 n=1
n6 945

The same is not true for odd values of the argument though. For instance, it
∞
1
P
took many researchers quite an effort to prove that ζ(3) = n3 is an irrational
n=1
number. This problem was considered hopeless until Roger Apéry finally solved
it. This number is now called the Apéry constant. An approximate value is

ζ(3) ≈ 1.202056903159594285399738161511450

It was first obtained with such precision by Andrey Markov in 1889. This improved
a result of Thomas Stieltjes from 2 years earlier when he published a table of the
values of the Riemann zeta function with 32 decimals places for integral values of
p from 2 to 70.
To appreciate these results you may try to obtain ζ(3) by definition using
your pocket calculator. See how many terms you must add to get just a few correct
digits. Again, mathematicians devised various tricks to get the result.

§2.8 Power series and approximate evaluation of functions

Not so long ago people didn’t have calculators, but mathematicians
√ knew
how to calculate an approximate value for some expressions like 3 or sin(π/12).
Even now in the computer era, have you ever wondered how your calculator or
computer performs such operations? Here is a hint.
Everyone can evaluate a polynomial at any number.
For example, if f (x) = x2 + 2x + 1 and x = 3 then f (3) = 16.
It is interesting that many other functions can be approximated arbitrarily
closely by polynomials and, if we extend these polynomials into infinite sums (such
sums are called power series), they represent the corresponding function. A good
example of how this works is given by the geometric series. Rewriting the formula
for the geometric series backwards, and replacing r with x we obtain
∞
1 X
= 1 + x + x2 + · · · = xn , |x| < 1. (18)
1−x n=0

This means that for any |x| < 1, the function f (x) = (1 − x)−1 is the same as
the infinite sum on the right hand side. This example gives you an idea of such
representations, but it looks like the function itself is much simpler than the power
series. Thus, in this case there is no profit in the replacement of the former by
the latter. Nevertheless, one can use the right hand side to evaluate the function
58

approximately by taking just the first few terms of the infinite series. For example,
the formula
1
≈ 1 + x, |x| < 1 (19)
1−x
allows you to find such results as
1 1
≈ 1.1 or ≈ 0.998.
0.9 1.002
quickly in your head.
In the first case we took x = 0.1 and in the second case we took x = −0.002.
When we say approximately equal we normally need to estimate the accuracy of
our approximated result, that is, the difference between the exact and approximate
values. Formula (19) gives better accuracy for smaller x. For larger x, like x = 0.9,
using just two terms is not as accurate. To improve the accuracy, one needs to
take more terms from the right hand side of (18), like (1 − x)−1 ≈ 1 + x + x2 .

Before discussing the approximation of functions by power series we state

some important properties of power series whose justification lies beyond the scope
of this work.
The first point is that for every power series
S(x) : a0 + a1 x + a2 x2 + . . .
there exists r ≥ 0, which is either a real number or ∞, called the radius of
convergence of S such that S(x) converges for every x such that |x| < r, and
S(x) diverges for every x such that |x| > r. When 0 < r < ∞, S(x) may converge
when x takes one of the values r, −r, both, or neither. The open interval (−r, r)
is called the interval of convergence.
If r = 0, the series is not interesting to us. So below assume that r > 0.
With S(x) we associate a function f by:
f (x) = a0 + a1 x + a2 x2 + . . . ( for all x such that S(x) converges) (20)
A crucial property of power series is that, in the interval of convergence of
S(x), f (x) is differentiable and

f 0 (x) = a1 + 2a2 x + 3a3 x2 + 4a4 x3 + . . . (−r < x < r).

0
That is, f (x) exists and is represented by the power series obtained by
differentiating S(x) term by term. Moreover, the resulting series has the same
interval of convergence as S(x).
Thus differentiating (18) we get the formula
1
= 1 + 2x + 3x2 + 4x3 + . . . (−1 < x < 1) (21)
(1 − x)2
Since the interval of convergence does not change we can differentiate (20)
term by term as many times as we like to obtain series representing the successive
derivatives of f (x). Differentiating n times we get
f (n) (x) = (n · (n − 1) · . . . 3 · 2 · 1)an + (a series of powers of x) (−r < x < r).
 59

Substituting x = 0, we see that f (n) (0) = n! an , which implies that an = f (n) (0)/n!.
This is important. It shows that, just as we obtain a function from a power
series by (21), so from a function f we obtain a power series:

f 0 (0) f 00 (0) 2 f (3) (0) 3

f (0) + x+ x + x + ... (22)
1! 2! 3!
This is called the Maclaurin series of f . From the properties stated above, if
there is any series which represents f in a non-empty interval (−r, r), then it is
the Maclaurin series of f .
Approximating functions by power series
To approximate f by a power series in some interval (−r, r) we can proceed
as follows:
Step 1. By computing the derivatives of f at 0 compute the Maclaurin series of f :
∞
X f (n) (0)
a 0 + a 1 x + a 2 x2 + . . . = an xn , an = , n≥0
n=0
n!

Notice that (18) is the Maclaurin series of 1/(1 − x) and (21) is the Maclaurin
series of 1/(1 − x)2 .
Step 2. Find r > 0 such that
∞
X
ak xk = f (x) (−r < x < r).
k=0

Of course, the larger r is the better.

Step 3. Let Sn (x) denote the partial sum a0 +a1 x+a2 x2 +. . .+an xn of the Maclarin
series of f . Find an expression B(x, n) in x and n which bounds |f (x) − Sn (x)|
for x ∈ (−r, r) such that

lim B(x, n) = 0 (−r < x < r).

n→∞

In the case of (18),

n ∞
1 X X |x|n+1
− xk = xk = (−1 < x < 1).
1−x 1−x
k=0 k=n+1

So we can take B(n, x) = |x|n+1 /(1 − x) in this case.

In the case of (21), one can use the Cauchy remainder formula to see that
n
1 X
k−1 |x|n
− kx < (n + 1)n (−1 < x < 1).
(1 − x)2 (1 − x)3
k=1

Step 4. Given  > 0 and a value of x in the interval found in step 2 for which we
want the value of f (x), use the expression B(x, n) in step 3 to find out how many
terms of the series are needed to compute f (x) with an error of less than .
60

In example (17), letting x = 1/4 and  = 0.01, the bound for the error is
B(1/4, n) = 1/(3 · 4n ). Since for n = 3, (1/3 · 43 ) < 0.01, the first four terms
1 + x + x2 + x3 are sufficient in this case.
In example (21), letting x = 1/4 and  = 0.01, the bound for the error from step
3 is
(n + 1)n
B(1/4, n) = 3 n−3
3 ·4
Since for n = 7, (8 · 7)/(33 · 44 ) < 0.01, the first seven terms 1 + 2x + 3x2 + 4x3 +
5x4 + 6x5 + 7x6 are sufficient in this case.
The technique for approximating functions by series described above is most
effective for approximating f (x) when x is close to 0. The same technique can be
used to approximate f (x) for x near a. Let g(x) = f (x + a). Assuming that g(x)
is represented by its Maclaurin series in some interval, we have
∞
X g (k) (0)
f (x + a) = g(x) = xk (−r < x < r)
k!
k=0

for some r > 0. Note that g (k) (0) = f (k) (a). Thus, substituting x − a for x in the
line above we have
∞
X f (k) (a)
f (x) = (x − a)k (−r < x − a < r).
k!
k=0

This is called the Taylor series for f (x) at x = a.

The following Maclaurin series for some other well-known functions are very
easily computed:
∞
X xn
ln(1 + x) = (−1)n+1 (−1 < x ≤ 1) (23)
n=1
n

x3 x5 x7 x9
sin x = x − + − + − ... (|x| < ∞) (24)
3! 5! 7! 9!

x2 x4 x6 x8
cos x = 1 − + − + − ... (|x| < ∞) (25)
2! 4! 6! 8!

In general, the tasks mentioned in steps 2 and 3 above, finding the interval
on which the Maclaurin series represents the function and finding an error bound
B(n, x) are beyond what we can explain in this work. However, there are a lot of
cases in which suitable error bounds can be found in a simple way. For example,
in (23), for 0 ≤ x ≤ 1, the terms are decreasing in absolute value and alternating
in sign. So we can take B(n, x) = xn+1 /(n + 1) for 0 ≤ x ≤ 1.
Note that power series yield interesting examples for summations. Fixing x,
we obtain a number series and its value. For example, let us find the value of
π2 π4 π6 π8
1− + − + − ···
2! 4! 6! 8!
 61

It is easy to see that the value is −1, by noticing that the sum is of the form
∞
X x2k
(−1)k with x = π. Thus, by (25) we immediately see that the sum is
(2k)!
k=0
cos π = −1.
1
Another example is the use of (21) with x = to show that
3
∞
X n 9
n−1
= . (26)
n=1
3 4

Finally, using formula (23) with x = 1 gives (16) from our discussion of Leibniz
series.

Caution. We can only use this if we know that the power series is truly representative
of the function at a given point x. We already saw that otherwise, this may
lead to nonsense like the sum of positive terms being equal to a negative number
∞
2k = −1) if we use x = 2 in formula (18) while it is only valid for |x| < 1.
P
(
n=0
Let us conclude this section with a problem that incorporates one of the examples.

Problem 38. The double sum. (Putnam, 1999).

Evaluate
∞ X ∞
X m2 n
m=1 n=1
3m (n3m + m3n )
Solution: First note that if we interchange the summation variables, m and n, the
result will be the same. That is
∞ X ∞ ∞ X ∞
X m2 n X n2 m
= .
m=1 n=1
3 (n3 + m3 ) n=1 m=1 3 (m3n + n3m )
m m n n

With this, we can make the expression look more symmetric:

∞ X
∞ ∞ X ∞
!
1 X m2 n X n2 m
+ .
2 m=1 n=1
3m (n3m + m3n ) n=1 m=1 3n (m3n + n3m )

Rewriting the summands with a common denominator, we have

∞ X
∞ ∞ ∞
X m2 n 1 X X 3n m2 n + 3m n2 m
=
m=1 n=1
3m (n3m + m3n ) 2 m=1 n=1 3m 3n (n3m + m3n )
∞ ∞
1 X X mn(n3m + m3n )
=
2 m=1 n=1 3m 3n (n3m + m3n )
∞ ∞
1 X X mn
=
2 m=1 n=1 3m 3n
∞ ∞
!!
1 X m 1X n
=
2 m=1 3m 3 n=1 3n−1
62

Now we use (26) and continue to get

∞  
1 X m 1 9
= ·
2 m=1 3m 3 4
∞
1 9 X m
= ·
2 12 m=1 3m
∞
9 1 X m
= ·
24 3 m=1 3m−1
9 9 9
= · = .
72 4 32

Please note that a double sum is simply a sum whose summands are themselves
sums. Also, we found the common denominator by switching the order of the sums.
We were able to do this because each term is positive and the series is convergent.
In general, we have to be careful with switching the order of summation, as seen
in a past example with a conditionally convergent series. Finally, we were able to
bring m/(3m ) outside of the sum whose index variable is n because, with respect
to n, m/(3m ) is a constant.

§2.9 L’Hospital’s Rule and power series

Let
∞
X ∞
X
f (x) = a n xn , g(x) = bn x n
n=0 n=0

be two functions represented by the power series shown in an interval (−r, r),
where r > 0. Then we have:
An instance of L’Hospital’s rule:
If f (0) = g(0) = 0 and one of the limits

f (x) f 0 (x)
lim , lim
x→0 g(x) x→0 g 0 (x)

exists, then so does the other and the limits are equal.
Here is the explanation. We start with an example. Suppose that

g(x) = b5 x5 + b6 x6 + · · · , f (x) = a5+k x5+k + a6+k x6+k + · · · ,

where k ≥ −4 is a fixed integer number, and b5 6= 0, a5+k 6= 0. Then

f (x) a5+k x5+k + a6+k x6+k + · · · a5+k xk + a6+k x1+k + · · ·

= =
g(x) b5 x5 + b6 x6 + · · · b5 + b6 x + · · ·
 63

and
f 0 (x) a5+k (5 + k)x4+k + a6+k (6 + k)x5+k + · · ·
=
g 0 (x) 5b5 x4 + 6b6 x5 + · · ·

a5+k (5 + k)xk + a6+k (6 + k)x1+k + · · ·

=
5b5 + 6b6 x + · · ·
Now there are three cases to consider. First, for k = 0

f (x) a5 + a6 x + · · · f 0 (x) 5a5 + 6a6 x + · · ·

= , =
g(x) b5 + b6 x + · · · g 0 (x) 5b5 + 6b6 x + · · ·
so
f (x) f 0 (x) a5
lim = lim 0 = .
x→0 g(x) x→0 g (x) b5
Second, for k > 0 (e.g. k = 1)

f (x) f 0 (x)
lim = lim 0 = 0.
x→0 g(x) x→0 g (x)

Third, for −4 ≤ k < 0 (e.g. k = −2) in both ratios f /g and f 0 /g 0 in the reduced
form the limit of the denominator is a constant (b5 and 5b5 respectively), but the
numerator is unbounded as x → 0. Thus, limits of both ratios do not exist.

We can proceed similarly in the general case. Let m(f ) be the least n such that
an 6= 0, and m(g) be the least n such that bn 6= 0. Since f (0) = g(0) = 0,
a0 = b0 = 0 and so m(f ), m(g) ≥ 1. It follows that m(f 0 ) = m(f ) − 1, and
m(g 0 ) = m(g) − 1. In the above example m(f ) = 5 + k ≥ 1 (since k ≥ −4),
m(g) = 5, m(f 0 ) = 4 + k, and m(g 0 ) = 4.
Case 1. m(f ) < m(g). This corresponds to −4 ≤ k < 0 in the above example. For
ease of notation let p denote m(f ). Then

f (x) ap xp + ap+1 xp+1 + ap+2 xp+2 + . . .

=
g(x) bp+1 xp+1 + bp+2 xp+2 + . . .

ap + ap+1 x + ap+2 x2 + . . .
=
bp+1 x + bp+2 x2 + . . .

Note that here we require that the leading coefficient in the numerator ap 6= 0,
but there is no assumption about the leading coefficient of the denominator.
Since power series functions are continuous in the interval of convergence and,
in particular at 0, as x → 0 the limit of the numerator is ap 6= 0 and the the
limit of the denominator is 0. Therefore the last fraction is unbounded as x → 0.
So limx→0 f (x)/g(x) does not exist. By the same token, since m(f 0 ) < m(g 0 ),
limx→0 f 0 (x)/g 0 (x) also does not exist.
Case 2. m(f ) ≥ m(g). This corresponds to k ≥ 0 in the above example. For ease
64

of notation let q denote m(g). Then

f (x) aq xq + aq+1 xq+1 + aq+2 xq+2 + . . .
=
g(x) bq xq + bq+1 xq+1 + bq+2 xq+2 + . . .

aq + aq+1 x + aq+2 x2 + . . .
=
bq + bq+1 x + bq+2 x2 + . . .

Note that here we require that the leading coefficient in the denominator bq 6= 0,
but there is no assumption about the leading coefficient of the numerator.
In the last fraction the limit as x → 0 of the numerator is aq and the limit of the
denominator is bq 6= 0. So limx→0 f (x)/g(x) = aq /bq exists. Also,
m(f 0 ) ≥ m(g 0 ). For f 0 and g 0 the corresponding coefficients are qaq , qbq respectively.
Thus limx→0 f 0 (x)/g 0 (x) = (qaq )/(qbq ) = aq /bq also exists.
We can prove a corresponding version of L’Hospital’s rule at x = a for
functions f, g represented by power series in x − a in an interval (a − r, a + r).
Let us give two examples of the instance of L’Hospital’s rule shown above.
sin x
Example 12. lim = 1.
x→0 x
In this example we verify the result from 2.6.3 using two new methods. First,
substituting x = 0 gives 00 . So the hypothesis of L’Hospital’s rule is satisfied and
we have
d
sin x (sin x) cos x cos 0
lim = lim dx d = lim = =1
x→0 x x→0 (x) x→0 1 1
dx
Alternatively, using (23), the Maclaurin series for sin x, we get
3 5 7
sin x x − x3! + x5! − x7! + . . .
lim = lim
x→0 x x→0 x
x2 x5 x7
 
= lim 1 − + − + ... = 1
x→0 3! 5! 7!
2 5 7
because the function represented by 1 − x3! + x5! − x7! + . . . is continuous at x = 0.
ln(1 + x)
Example 13. lim = 1.
x→0 x
Applying the rule, the given limit is equal to
d/dx [ln(1 + x)] (1 + x)−1 limx→0 (1 + x)−1 1
lim = lim = = = 1.
x→0 d/dx[x] x→0 1 limx→0 1 1

This result can be used to calculate the limit of the Euler sequence, see 2.6.2.
The first observation we need is that
 n  x
1 1
lim 1 + = lim 1 +
n→∞ n x→∞ x
provided that the limit on the right exists. This is clear from the definitions of
the limits on the two sides. So it is sufficient to evaluate the limit on the right.
 65

Let exp(x) = ex denote the exponential function which is the inverse function of
ln(x). The rest of the calculation is:
 x
1 1/y
lim 1+ = lim (1 + y) by substituting x = 1/y
x→∞ x y→0+
1/y
= lim (1 + y) provided this limit exists
y→0
  
1/y
= lim exp ln (1 + y) since exp is the inverse of ln
y→0
 
ln(1 + y)
= lim exp by a property of ln
y→0 y
 
ln(1 + y) since exp is continuous
= exp lim
y→0 y assuming inner limit exists
= exp(1) = e from Example 13

One should be careful to apply L’Hospital’s rule only when the hypotheses
are satisfied. For example, the application

cos x d/dx [cos x] − sin x

lim = lim = lim = 0
x→0 1+x x→0 d/dx [1 + x] x→0 1

gives nonsense. The function (cos x)/(1 + x) is continuous at x = 0 because it is

a quotient of continuous function with denominator nonzero at x = 0. Thus the
true value of the limit is (cos 0)/(1 + 0) = 1.

§2.10 Fourier series

In this section we touch on the subject of Fourier series. We begin with a
problem about a finite trigonometric sum called the Dirichlet kernel.

Problem 39. Prove that for N ≥ 0 and x 6= 2πk, k ∈ Z,

sin[(N + 1/2)x]
1 + 2 cos x + 2 cos(2x) + 2 cos(3x) + · · · + 2 cos(N x) = .
sin[x/2]

Solution: Multiply both sided by sin[x/2] to get rid of the fraction on the righthand
side. To simplify the new lefthand side use the trigonometric identity

2 cos A sin B = sin(A + B) − sin(A − B),

with A = nx, n = 1, 2, . . . N , and B = x/2. Then we have

 x    3x   x    5x   
3x
sin + sin − sin + sin − sin + ···
2 2 2 2 2
     
1 1
+ sin N + x − sin N − x .
2 2
66

After cancelling sin(x/2) with − sin(x/2), sin(3x/2) with − sin(3x/2), and

so on, we are left with only one term sin (N + 21 )x as required. Notice that we
 

have used the method of telescoping which we studied in §2.4.

The Dirichlet kernel is related to a very important kind of series called a
Fourier series which takes the form:
1
a0 + a1 cos x + b1 sin x + a2 cos(2x) + b1 sin(2x) + a3 cos(3x) + b3 sin(3x) + . . .
2
and may also be written
∞
1 X
a0 + (an sin(nx) + bn cos(nx)) . (27)
2 n=1

Here the coefficients an and bn are fixed real numbers and x is a free variable.
As with a power series in x, when x is given a particular value, we get a series
of numbers which may or may not converge. Thus, just as a power series in x
determines a function f (x) defined for those x for which the series converges, the
same is true of a Fourier series.
For each n ≥ 1, the functions cos(nx) and sin(nx) are periodic with period
(2π)/n in the sense that, for all x,
     
2π 2π
cos n x + = cos(nx), sin n x + = sin(nx).
n n

Since 2π is the least common multiple of 2π, (2π)/2, (2π)/3, (2π)/4, . . . , if f (x) is
represented by the Fourier series shown above, then f (x) is periodic with period
2π. This is the big difference between representing a function by a power series
and representing a function by a Fourier series.
P∞
We saw in §2.8 that there is a way of choosing the coefficients in n=0 an xn
to obtain a series which represents a given function f . We take an = f (n) (0)/n!.
As we saw above this “recipe” works in many important cases.
The situation for Fourier series is analogous. Let f be a function with domain
(−π, π). If an and bn are given by the formulas

1 π 1 π
Z Z
an = f (x) cos(nx) dx (n ≥ 0), bn = f (x) sin(nx) dx (n ≥ 1),
π −π π −π

then (27) is called the Fourier series of f (x).

In many cases the Fourier series of f (x) represents f (x) on (−π, π). We give
two examples.
∞
X 2
Example 1. x = (−1)n+1 sin (nx) (−π < x < π).
n=1
n
Notice that all the coefficients an are 0 in this case because x is an odd function.
Similarly, in the power series expansion of an odd function such as sin x only odd
 67

powers of x appear.
∞
π2 X 4
Example 2. x2 = + (−1)n 2 cos (nx) (−π ≤ x ≤ π).
3 n=1
n
Notice that all the coefficients bn are 0 in this case because x2 is an even function.
Similarly, in the power series expansion of an even function such as cos x only even
powers of x appear.
Proving that the Fourier series of f (x) converges to the function f (x) at a
given point x requires mathematical techniques which can not be discussed in this
text. However, once such a correspondence is established, Fourier series expansions
give rise to some interesting results regarding sums of series. For example, letting
x = π/2 in the above Example 1 we obtain
1 1 1 1 π
1− + − + + ··· = .
3 5 7 9 4
As well, letting x = π in Example 2 we obtain
∞
π2 X 4
π2 = + ,
3 n=1
n2

which immediately gives the value of ζ(2) shown in (17).

The Dirichlet kernel with which we began this section plays a key role in the
proof that, under certain conditions, the Fourier series of f (x) converges to f (x)
for all x in (−π, π).
Fourier series were invented by Joseph Fourier in his study of the diffusion of
heat in a metal plate. He published his work on the problem in Théorie Analytique
de la Chaleur in 1822. Fourier series have found many applications in physics and
engineering.
68

Exercises
1. Find an appropriate formula or method from this chapter to show that
∞ 1000
X 1 1 X
(i) n
= (ii) 2n + 3 = 1, 004, 000
n=1
4 3 n=1
∞
175 X 1 1
(iii) 1.76 = (iv) =
99 n=1
n(n + 1)(n + 2) 4
∞ ∞
X n 4 X 3n n
(v) n
= (vi) = 48
n=1
4 9 n=1
4n−1
∞ ∞
X 1 π2 X 1 π2
(vii) 2
= (viii) 2
=
n=1
(2n) 24 n=1
(2n − 1) 8
∞ ∞
X xn X πn
(ix) = ex (x) = eπ
n=0
n! n=0
n!
∞ ∞
X 6n ln 13 − ln 7 X 1
(xi) (−1)n+1 n+1
= (xii) = ln 2
n=1
n7 7 n=1
2n(2n − 1)
 n
ln(1 + kx) k
(xiii) lim =k (xiv) lim 1 + = ek
x→0 x n→∞ n

cos x − 1 1
(xv) lim nsin(k/n) = k (xvi) lim =−
n→∞ x→0 x2 2

N
X sin(N x) + sin[(N + 1)x]
(xvii) 1+2 cos(nx) = , N ≥ 1, x 6= πk, k ∈ Z
n=1
sin x

2. Recall where in the text the each of the following names were mentioned and
what problem or theory is related to them.
Abel, Apéry, Archimedes, Cantor, Cauchy, Euclid, Euler, Fibonacci, Fourier,
Gauss, L’Hospital, Koch, Leibniz, Markov, Maclaurin, Riemann, Sierpinsky,
Stieltjes, Taylor, Zeno.
 69

APPENDIX 1: DERIVATIVES

In this book we use derivatives as an auxiliary tool in several places. First,

they are useful to check whether a function or sequence is monotonic. Secondly,
they are needed for finding Maclaurin series representations of functions. And
finally, they are required for L’Hopital’s Rule for finding limits. In this appendix
we will introduce the definition of the derivative, along with the derivatives of
some useful functions, and also a few rules which can be used to simplify our task
of finding a derivative.
A 1.1 Definition of the derivative

The derivative of a function y = f (x) at a point x0 is a number. It can be

dy
denoted in many ways including y 0 (x0 ), f 0 (x0 ), and (x0 ). Geometrically, this
dx
number represents the slope of the line tangent to the curve y = f (x) at the point
(x0 , f (x0 )). Algebraically, it is the instantaneous rate of change of the function at
the point, so it shows how the function changes if the argument rises from x to
x + h as h → 0

f (x + h) − f (x)
f 0 (x) = lim
h→0 (x + h) − x
f (x + h) − f (x)
= lim .
h→0 h

Let us use the definition of the derivative to find the derivative of an

elementary function.
Problem 40. Let f (x) = ax, where a is a constant. Then

f (x + h) − f (x)
f 0 (x) = lim
h→0 h
a(x + h) − ax
= lim
h→0 h
ah
= lim
h→0 h
= lim a
h→0
= a

which makes sense because f (x) = ax is a straight line with slope a, and the rate
of change of this function is the slope. Thus, the derivative is also equal to a at
any point x0 .

Here’s another example.

70

Problem 41. Let f (x) = ax2 , where a is a constant. Then,

f (x + h) − f (x)
f 0 (x) = lim
h→0 h
a(x + h)2 − ax2
= lim
h→0 h
ax2 + 2ahx + ah2 − ax2
= lim
h→0 h
2
2ahx + ah
= lim
h→0 h
= lim (2ax + ah)
h→0
= 2ax.
This tells you that the derivative, and the slope of the tangent line, of the curve
f (x) = ax2 at any point (x0 , f (x0 )), is equal to 2ax0 . For instance, the curve has
a horizontal tangent line at the origin since the derivative at x0 = 0 is 2 · 0 = 0.
The curve has a positive slope for x > 0, meaning that the rate of change of the
function is positive and therefore, the function is increasing for x > 0. Likewise,
for all x < 0 the slope of the tangent line is negative and the function is decreasing.
The relationship between the slope of the tangent line and the rate of change
of a function, expressed numerically using the derivative, becomes a valuable tool
for the monotonicity check.
i) If we have a positive derivative for all x in an interval, then the function
is increasing on that interval.
ii) If we have a negative derivative for all x in an interval, then the function
is decreasing on that interval.
iii) If we have a derivative equal to zero for all x in an interval, then the
function is constant on that interval.
Now, we will present several rules and techniques for finding derivatives.
Each of them can be derived from the definition of the derivative.

A 1.2. Important derivatives

It would be inappropriate for someone to be expected to use limits whenever
they need to find a derivative because this can sometimes be a long and tedious
process. The following are important derivatives that are presented here without
proof:

f (x) f 0 (x) f (x) f 0 (x)

c 0 ln x 1/x
n n−1
x nx sin x cos x
ex ex cos x − sin x
x x
a , a>0 (ln a)a tan x sec2 x

In these tables a, c, and n are real constants with a > 0.

 71

These derivatives can be combined with the following rules to easily find the
derivatives of many functions. We will refer to a function which has a derivative
as differentiable.
A 1.3. Rules and properties
1. The Constant Multiple Rule
The constant multiple rule states that if f is a differentiable function and c
is a real number, then cf is also differentiable and:
d
[cf (x)] = cf 0 (x),
dx
d
where dx means “the derivative with respect to x”.

d
Problem 42. What is dx [3x2 ]?
Solution: With c = 3 and f (x) = x2 we can use the constant multiple rule. We
have, f 0 (x) = 2x, and therefore dx
d
[3x2 ] = 3(2x) = 6x.

2. The Sum and Difference Rule

The sum (and difference) rule states that the sum (or difference) of two
differentiable functions is differentiable and is equal to the sum (or difference) of
their derivatives.
d
[f (x) ± g(x)] = f 0 (x) ± g 0 (x)
dx
d
Problem 43. What is dx [ex + 3x2 ]?
Solution: We will use the sum rule, with f (x) = ex and g(x) = 3x2 . f 0 (x) = ex
and g 0 (x) = 6x which we found in the last example. Thus, dx
d
[ex + 3x2 ] = ex + 6x.

3. The Product Rule

The product rule states that the product of two differentiable functions is
also differentiable, and
d
[f (x)g(x)] = f 0 (x)g(x) + f (x)g 0 (x)
dx
We can see from this that the constant multiple rule is just a special case of
the product rule, with g(x) = c.

d
Problem 44. What is dx [3x sin x]?
Solution: We will use the product rule with f (x) = x and g(x) = sin x. f 0 (x) = 1
and g 0 (x) = cos x, so dx
d
[x sin x] = sin x + x cos x and using the constant multiple
d
rule we get dx [3x sin x] = 3 sin x + 3x cos x.

4. The Quotient Rule

The quotient rule states that the quotient f /g of two differentiable functions
(with g(x) 6= 0) is itself differentiable and
f 0 (x)g(x) − f (x)g 0 (x)
 
d f (x)
=
dx g(x) [g(x)]2
72

d
x
Problem 45. What is dx ex ?
Solution: We will use the quotient
 x  rule with f (x) = x and g(x) = ex . f 0 (x) = 1
0 ex −xex
and g (x) = e . Therefore dx ex = e2x = 1−x
x d
ex .

5. The Chain Rule

The chain rule involves the composition of functions. The composition read
“f of g of x” can be written as (f ◦ g)(x) or f (g(x)). Finding the composition of
the two functions involves substituting g(x) for x in the expression for f (x).

Problem 46. Find the composition f (g(x)) for f (x) = sin x and g(x) = ex .
Solution: Replacing x by g(x) in the expression for f (x) we quickly obtain
f (g(x)) = f (ex ) = sin(ex ).
The chain rule states, if f and g are differentiable, then their composition
of f and g is also differentiable, and
d
[f (g(x))] = f 0 (g(x)) · g 0 (x).
dx

Thus for the composite function sin(ex ) we have

d
[sin(ex )] = cos(ex ) · ex .
dx
d
Problem 47. What is dx [ln ex ]?
First Solution: This is f (g(x)) with f (x) = ln x and g(x) = ex . f 0 (x) = 1/x and
g 0 (x) = ex so dxd
[f (g(x))] = f 0 (g(x)) · g 0 (x) = e1x · ex = 1
Second Solution: ln x and ex are inverse functions of each other, so ln ex = x.
d d
Therefore, dx [ln ex ] = dx [x] = 1.

As you can see, both of the previous solutions had the exact same answer even
though they were approached using totally different methods.

Exercises
Find the derivatives of the following functions:

1. i) 17x100 ii) 3 cos x iii) −4 ln x

iv) 5x v) 4x−1 vi) π sin x

2. i) ex + e2x ii) x3 + x2 iii) cos x + sin x

iv) 5 − x v) 3x − x3 vi) ln x − x

3. i) sin x · cos x ii) 1/2 sin 2x iii) ex · ex

iv) x ln x − x v) 3x · ex vi) ex sin x
 73

4. i) (sin x)/(cos x) ii) (ex )/(x2 ) iii) (3x )/(4x )

iv) (x3 )/(x2 ) v) (sin x)/(ex ) vi) (ex )/(ln x)

5. i) ln(ln x) ii) esin x iii) eln x

iv) (sin x)4 v) ln(sin x) vi) (ln x)12

APPENDIX B: TRIANGLES
Here we have collected a few facts which are useful for problems dealing with
triangles.
I. Similar triangles
Similar figures are re-scaled images of each other. If either of the following
is true, then the two triangles ABC and A0 B 0 C 0 are similar:

1. Â = Â0 , B̂ = B̂ 0 .
AB AC BC
2. = 0 0 = 0 0 = k. The number k is called coefficient of
A0 B 0 AC BC
similarity.
AB AC
3. Â = Â0 , = 0 0.
A0 B 0 AC
II. Right triangle
A
...........
.... ...............
... α ............
... .........
... .........
... .........
......... c
b ..... .........
.........
..... .........
.........
... .........
... .........
........... .........
...
..............................................................................................................................................
C a B

Figure 16. Right triangle.

1. a2 + b2 = c2
b a a
2. cos α = , sin α = , tan α = .
c c b

α sin α cos α tan α

0 0 1 0
√ √
π/6 1/2 3/2 1/ 3
√ √
π/4 1/ 2 1/ 2 1
√ √
π/3 3/2 1/2 3
π/2 1 0 undefined
74

2 tan α
3. cos(2α) = cos2 α−sin2 α, sin(2α) = 2 sin α cos α, tan(2α) = .
1 − tan2 α
III. Equilateral triangle
B
...
.......
... ...
..... ......
..
. ...
.
. ...
... ...
... ..........................................
.............
. .......
.. ....
.. ...
.... .....
........ ......
. ..
... ....
. ... ... ...
... ... .... .. ...
..
.
.. ... ...
.... ......
.. .... .
.. r .. ...
.
. ...... .
... ... . ...
... .... ....... ...
..........................................................................................................................
A C

Figure 17. Equilateral triangle.

1. Â = B̂ =√Ĉ = π/3, |AB| = |BC| = |CA| = a.

3a2
2. Area is .
4
3. Any altitude is an angle bisector and a median as well. Altitudes intersect
at a point, which is also the incenter
√ and circumcenter.
3a
4. Radius of the incircle is .
6
IV. Isoceles triangle

C
...
....
.. ..
... ....
.. ..
.. ..
.. ..
.. ..
.. ..
.. ..
... ..
..
... ..
.. ..
. ..
... ..
..
... ..
.
. ..
. ..
... ..
.. ..
.
... .......................................... ....
............ ....... ..
..... ......
......
... .....
....
.. ..
.. .............
...
........ .....
.. ..
. . . ............ ........
. ... . r ....
.
.. . ... ..
.. .. .. ..
.. .. .. ..
.. ..... .. .. ....
.. ..... .
. ..
.. ....... ...... ..
.. ... ........ ..
........................................................................................................................
A a B

Figure 18. Isoceles triangle.

1. Â = B̂ = (π − Ĉ)/2, |BC| = |CA|, |BA| = a.

a
2. Let CD be the altitude. Then |CD| = tan Â.
2 !
a Â
3. The center of incircle lies on CD. The radius r of incircle is tan .
2 2
 ATOM
A Taste Of Mathematics / Aime-T-On les Mathématiques

1. Edward J. Barbeau Mathematical Olympiads’ Correspondence Program (1995-1996 )

2. Bruce L.R. Shawyer Algebra — Intermediate Methods
3. Peter I. Booth, John Grant McLoughlin and Bruce L.R. Shawyer Problems
for Mathematics Leagues
4. Edward J. Barbeau and Bruce L.R. Shawyer Inequalities
5. Richard Hoshino and John Grant McLoughlin Combinatorial Explorations
6. Peter I. Booth, John Grant McLoughlin and Bruce L.R. Shawyer Problems
for Mathematics Leagues — II
7. Jim Totten Problems of the Week
8. Peter I. Booth, John Grant McLoughlin and Bruce L.R. Shawyer Problems
for Mathematics Leagues — III
9. Edward J. Barbeau The CAUT Problems
10. Naoki Sato Modular Arithmetic
11. Bruce L.R. Shawyer and Bruce B. Watson Problems for Junior Mathematics
Leagues
12. Edward J. Barbeau Transformational Geometry
13. Edward J. Barbeau Quadratics and Complex Numbers
Cost per volume (including shipping and handling charges):
Regular Rate $15.00 — CMS Member’s Rate $12.00 (excluding taxes).
For non-Canadian shipping addresses the rates quoted must be paid in US funds.
Coût par volume (frais y compris d’expédition et de manutention) :
Taux régulier 15,00 $ — Tarif de membre de la SMC 12,00 $ (avant taxes).
Pour une adresse à l’étranger, paiement requis en devises américaines.
For more information and to order:
Pour plus d’informations et pour compléter une commande :
cms.math.ca/Publications/Books/atom
or contact:
ou contacter :
CMS/SMC Publications
209-1725 boul. St. Laurent Blvd, Ottawa, Ontario, CANADA K1G 3V4

Email: publications@cms.math.ca Courriel : publications@smc.math.ca

 Canadian Mathematical Society / Société mathématique du Canada
209-1725 boul. St. Laurent Blvd, Ottawa, Ontario, CANADA K1G 3V4
cms.math.ca