Introdu

tion to Analysis

Irena Swanson
Reed College
Spring 2017
Table of contents

Preface 7
The briefest overview, motivation, notation 9
Chapter 1: How we will do mathematics 13
Section 1.1: Statements and proof methods 13
Section 1.2: Statements with quantifiers 25
Section 1.3: More proof methods, and negation 28
Section 1.4: Summation 34
Section 1.5: Proofs by (mathematical) induction 36
Section 1.6: Pascal’s triangle 45
Chapter 2: Concepts with which we will do mathematics 48
Section 2.1: Sets 48
Section 2.2: Cartesian product 56
Section 2.3: Relations, equivalence relations 58
Section 2.4: Functions 64
Section 2.5: Binary operations 72
Section 2.6: Fields 79
Section 2.7: Order on sets, ordered fields 82
Section 2.8: Increasing and decreasing functions 89
Section 2.9: The Least upper bound property of R 91
Section 2.10: Absolute values 93
Chapter 3: The field of complex numbers, and topology 96
Section 3.1: Complex numbers 96
Section 3.2: Functions related to complex numbers 99
Section 3.3: Absolute value in C 101
Section 3.4: Polar coordinates 104
Section 3.5: Topology on the fields of real and complex numbers 109
Section 3.6: The Heine-Borel theorem 113
4

Chapter 4: Limits of functions 116

Section 4.1: Limit of a function 116
Section 4.2: When something is not a limit 127
Section 4.3: Limit theorems 130
Section 4.4: Infinite limits (for real-valued functions) 138
Section 4.5: Limits at infinity 141
Chapter 5: Continuity 144
Section 5.1: Continuous functions 144
Section 5.2: Topology and the Extreme value theorem 149
Section 5.3: Intermediate value theorem 152
Section 5.4: Radical functions 156
Section 5.5: Uniform continuity 161
Chapter 6: Differentiation 164
Section 6.1: Definition of derivatives 164
Section 6.2: Basic properties of derivatives 168
Section 6.3: The Mean value theorem 176
Section 6.4: L’Hôpital’s rule 180
Section 6.5: Higher-order derivatives, Taylor polynomials 183
Chapter 7: Integration 188
Section 7.1: Approximating areas 188
Section 7.2: Computing integrals from upper and lower sums 197
Section 7.3: What functions are integrable? 200
Section 7.4: The Fundamental theorem of calculus 205
Section 7.5: Natural logarithm and the exponential functions 212
Section 7.6: Applications of integration 218
Chapter 8: Sequences 225
Section 8.1: Introduction to sequences 225
Section 8.2: Convergence of infinite sequences 230
Section 8.3: Divergence of infinite sequences and infinite limits 237
Section 8.4: Convergence theorems via epsilon-N proofs 241
Section 8.5: Convergence theorems via functions 247
Section 8.6: Bounded sequences, monotone sequences, ratio test 250
Section 8.7: Cauchy sequences, completeness of R, C 254
Section 8.8: Subsequences 258
Section 8.9: Liminf, limsup for real-valued sequences 261
 5

Chapter 9: Infinite series and power series 266

Section 9.1: Infinite series 266
Section 9.2: Convergence and divergence theorems for series 270
Section 9.3: Power series 277
Section 9.4: Differentiation of power series 282
Section 9.5: Numerical evaluations of some series 285
Section 9.6: Some technical aspects of power series 287
Section 9.7: Taylor series 291
Section 9.8: A special function 294
Section 9.9: A special function, continued 296
Section 9.10: Trigonometry 301
Section 9.11: Examples of L’Hôpital’s rule 304
Section 9.12: Further specialized uses of trigonometry 306
Appendix A: Advice on writing mathematics 311
Appendix B: What you should never forget 315
Index 319
Preface

These notes were written expressly for Mathematics 112 at Reed College, with first
usage in the spring of 2013. The title of the course is “Introduction to Analysis”. The
prerequisite is calculus. Recently used textbooks have been Steven R. Lay’s “Analysis,
With an Introduction to Proof ” (Prentice Hall, Inc., Englewood Cliffs, NJ, 1986, 4th
edition), and Ray Mayer’s in-house notes “Introduction to Analysis” (2006, available at
http://www.reed.edu/~mayer/math112.html/index.html).
In Math 112, students learn to write proofs while at the same time learning about bi-
nary operations, orders, fields, ordered fields, complete fields, complex numbers, sequences,
and series. We also review limits, continuity, differentiation, and integration. My aim for
these notes is to constitute a self-contained book that covers the standard topics of a course
in introductory analysis, that handles complex-valued functions, sequences, and series, that
has enough examples and exercises, that is rigorous, and is accessible to Reed College un-
dergraduates. I currently maintain two versions of these notes, one in which the natural,
rational and real numbers are constructed and the Least upper bound theorem is proved
for the ordered field of real numbers, and one version in which the Least upper bound
property is assumed for the ordered field of real numbers. You are currently reading the
shorter, latter version. See my Math 112 webpage www.reed.edu/~iswanson/112.html
for links to both versions.
Chapter 1 is about how we do mathematics: basic logic, proof methods, and Pascal’s
triangle for practicing proofs. Chapter 2 introduces foundational concepts: sets, Carte-
sian products, relations, functions, binary operations, fields, ordered fields, Archimedean
property for the set of real numbers. In particular, we assume that the set of familiar
real numbers forms an ordered field with the Least upper bound property. In Chapter 3
we construct the very useful field of complex numbers, and introduce topology which is
indispensable for the rigorous treatment of limits. I cover topology more lightly than what
is in the written notes. Subsequent chapters cover standard material for introduction to
analysis: limits, continuity, differentiation, integration, sequences, series, ending with the
P ∞ xk
development of the power series k=0 k! , the exponential and the trigonometric func-
tions. Since students have seen limits, continuity, differentiation and integration before, I
go through chapters 4 through 7 quickly. I slow down for sequences and series (chapters 8
and 9).
An effort is made throughout to use only what had been proved. When trigonometric
functions appear in earlier exercises they are accompanied with a careful listing of the
needed assumptions about them. For this reason, the chapters on differentiation and
integration do not have the usual palette of examples of other books. The final sections of
the last chapter make up for it and work out much trigonometry in great detail and depth.
8 Preface

I acknowledge and thank the support from the Dean of Faculty of Reed College to fund
exercise and proofreading support in the summer of 2012 for Maddie Brandt, Munyo Frey-
Edwards, and Kelsey Houston-Edwards. I also thank the following people for their valuable
feedback: Mark Angeles, Josie Baker, Marcus Bamberger, Anji Bodony, Zachary Campbell,
Nick Chaiyachakorn, Safia Chettih, Laura Dallago, Andrew Erlanger, Joel Franklin, Palak
Jain, Ya Jiang, Albyn Jones, Mason Kennedy, Christopher Keane, Michael Keppler, Ryan
Kobler, Oleks Lushchyk, Molly Maguire, Benjamin Morrison, Samuel Olson, Kyle Ormsby,
Angélica Osorno, Shannon Pearson, David Perkinson, Jeremy Rachels, Ezra Schwartz,
Marika Swanberg, Simon Swanson, Matyas Szabo, Ruth Valsquier, Xingyi Wang, Emerson
Webb, Livia Xu, Qiaoyu Yang, Dean Young, Eric Zhang, and Jialun Zhao. If you have
further comments or corrections, please send them to [email protected].
The briefest overview, motivation, notation

What are the meanings of the following:

5+6
7·9
8−4
4/5
√
2
4−8
1/3 = 0.333 . . . 1 = 3 · 1/3 = 0.999 . . .
(a · b)2 = a2 · b2
(a + b) · (c + d) = ac + ad + bc + bd
(a + b) · (a − b) = a2 − b2
(a + b)2 = a2 + 2ab + b2
√ √ √
a · b = ab (for which a, b?)
√ √ p √
What is going on: −4 · −9 = (−4)(−9) = 36 = 6,
√ √
−4 · −9 = 2i · 3i = −6
You know all of the above except possibly the complex numbers in the last two rows,
where obviously something went wrong. We will not resolve this last issue until later in
the semester, but the point for now is that we do need to reason carefully.
The main goal of this class is to learn to reason carefully, rigorously. Since one cannot
reason in a vacuum, we will (but of course) be learning a lot of mathematics as well:
sets, logic, various number systems, fields, the field of real numbers, the field of complex
numbers, sequences, series, some calculus, and that eix = cos x + i sin x.
We will make it all rigorous, i.e., we will be doing proofs. A proof is a sequence of
steps that logically follow from previously accepted knowledge.
But no matter what you do, never divide by 0. For further wise advice, turn to
Appendix A.

[Notational convention: Text between square brackets in this font should

be read as a possible reasoning going on in the background in your head,
and not as part of formal writing.]
† 1 Exercises with a dagger are invoked later in the text.
*2 Exercises with a star are more difficult.
Introdu tion to Analysis
Chapter 1: How we will do mathematics

1.1 Statements and proof methods

Definition 1.1.1 A statement is a reasonably grammatical and unambiguous sentence
that can be declared either true or false.

Why do we specify “reasonably grammatical”? We do not disqualify a statement just

because of poor grammar, nevertheless, we strive to use correct grammar and to express the
meaning clearly. And what do we mean by true or false? For our purposes, a statement
is false if there is at least one counterexample to it, and a statement is true if it has been
proved so, or if we assume it to be true.

Examples and non-examples 1.1.2

(i) The sum of 1 and 2 equals 3. (This is a statement; it is always true.)
(ii) Seventeen. (This is not a statement.)
(iii) Seventeen is the seventh prime number. (This is a true statement.)
(iv) Is x positive? (This is not a statement.)
(v) 1 = 2.* (This is a false statement.)
(vi) For every real number ǫ > 0 there exists a real number δ > 0 such that for all x,
if 0 < |x − a| < δ then x is in the domain of f and |f (x) − L| < ǫ. (This is a
statement, and it is (a part of) the definition of the limit of a (special) function f
at a being L. Out of context, this statement is neither true or false, but we can
prove it or assume it for various functions f .) )
(vii) Every even number greater than 4 can be written as a sum of two odd primes.
(This statement is known as Goldbach’s conjecture. No counterexample is
known, and no proof has been devised, so it is currently not known if it is true or
false.)
These examples show that not all statements have a definitive truth value. What makes
them statements is that after possibly arbitrarily assigning them truth values, different
consequences follow. For example, if we assume that (vi) above is true, then the graph

* This statement can also be written in plain English as “One equals two.” In mathematics it is acceptable
to use symbolic notation to some extent, but keep in mind that too many symbols can make a sentence hard to
read. In general we avoid starting sentences with a symbol. In particular, do not make the following sentence.
“=” is a verb. Instead make a sentence such as the following one. Note that “=” is a verb.
14 Chapter 1: How we will do mathematics

of f near a is close to the graph of the constant function L. If instead we assume that
(vi) above is false, then the graph of f near a has infinitely many values at some vertical
distance away from L no matter how much we zoom in at a. With this in mind, even “I
am good” is a statement: if I am good, then I get a cookie, but if I am not good, then you
get the cookie. On the other hand, if “Hello” were to be true or false, I would not be able
to make any further deductions about the world or my next action, so that “Hello” is not
a statement, but only a sentence.

A useful tool for manipulating statements is a truth table: it is a table in which the
first few columns may set up a situation, and the subsequent columns record truth values of
statements applying in those particular situations. Here are two examples of truth tables,
where “T ” of course stands for “true” and “F ” for “false”:

f (x) f is a constant function. f is continuous. f is differentiable everywhere.

x2 F T T
|x| F T F
7 T T T

x y xy > 0 xy ≤ 0 xy < 0
x>0 y >0 T F F
x>0 y ≤0 F T F
x<0 y >0 F T T
x<0 y ≤0 F F F

Note that in the second row of the last table, in the exceptional case y = 0, the
statement xy < 0 is false, but in “the majority” of the cases in that row xy < 0 is true.
The one counterexample is enough to declare xy < 0 not true, i.e., false.

Statements can be manipulated just like numbers and variables can be manipulated,
and rather than adding or multiplying statements, we connect them (by compounding the
sentences in grammatical ways) with connectors such as “not”, “and”, “or”, and so on.

Statement connecting:
(1) Negation of a statement P is a statement whose truth values are exactly opposite
from the truth values of P (under any specific circumstance). The negation of P
is denoted “ not P ” (or sometimes “~P ” or “not P ”).
Some simple examples: the negation of “A = B” is “A 6= B”; the negation
of “A ≤ B” is “A > B”; the negation of “I am here” is “I am not here” or
“It is not the case that I am here”.
 Section 1.1: Statements and proof methods 15

Now go back to the last truth table. Note that in the last line, the truth
values of “xy > 0” and “xy ≤ 0” are both false. But one should think that
“xy > 0” and “xy ≤ 0” are negations of each other! So what is going on,
why are the two truth values not opposites of each other? The problem is of
course that the circumstances x < 0 and y ≤ 0 are not specific enough. The
statement “xy > 0” is under these circumstances false precisely when y = 0,
but then “xy ≤ 0” is true. Similarly, the statement “xy ≤ 0” is under the
given circumstances false precisely when y < 0, but then “xy > 0” is true.
Thus, once we make the conditions specific enough, then the truth values of
“xy > 0” and “xy ≤ 0” are opposite, so that the two statements are indeed
negations of each other.
(2) Conjunction of statements P and Q is a statement that is true precisely when
both P and Q are true, and it is false otherwise. It is denoted “P and Q” or
“P ∧ Q”. We can record this in a truth table as follows:

P Q P and Q
T T T
T F F
F T F
F F F

(3) Disjunction of statements P and Q is a statement that is false precisely when

both P and Q are false, and it is true otherwise. We denote it as “P or Q” or as
“P ∨ Q”. In other words, as long as either P or Q is true, then P or Q is true. In
plain language, unfortunately, we use “or” in two different ways: “You may take
cream or sugar” says you may take cream or sugar or both, just like in the proper
logical way, but “Tonight we will go to the movies or to the baseball game” implies
that we will either go to the movies or to the baseball game but we will not do
both. The latter connection of two sentences is in logic called exclusive or, often
denoted xor. Even “either-or” does not disambiguate between “or” and “xor”,
but “either-or else” is the same as “xor”. The truth table for the two disjunctions
is:
P Q P or Q P xor Q
T T T F
T F T T
F T T T
F F F F
16 Chapter 1: How we will do mathematics

(4) Implication or a conditional statement is a statement of the form “P im-

plies Q,” or variants thereof, such as “If P then Q,” “Given P , Q follows,” “Q
whenever P ”. A symbolic abbreviation of implication is “P ⇒ Q.” An implication
is true when a true conclusion follows a true assumption, or whenever the assump-
tion is false. In other words, P ⇒ Q is false exactly when P is true and Q is false.
In an implication “P ⇒ Q”, P is called the antecedent and Q the consequent.
The truth table for implication is as follows:
P Q P ⇒Q
T T T
T F F
F T T
F F T

It may be counterintuitive that a false antecedent always makes the implication

true. Bertrand Russell once lectured on this and claimed that if 1 = 2 then he
(Bertrand Russell) was the pope. An audience member challenged him to prove
it. So Russell reasoned somewhat like this: “If I am the pope, then the consequent
is true. If the consequent is false, then I am not the pope. But if I am not the
pope, then the pope and I are two different people. By assumption 1 = 2, so we
two people are one, so I am the pope. Thus no matter what, I am the pope.”
Furthermore, if 1 = 2, then Bertrand Russell is similarly also not the pope.
Namely, if he is not the pope, the consequent is true, but if he is the pope, then
the pope and he are one, and since one equals two, then the pope and he are two
people, so Russell cannot be the pope.
A further discussion about why false antecedent makes the implication true is in
the next discussion (5).
We list here most possible rephrasings of “P implies Q”:

P implies Q.
If P then Q.
P is a sufficient condition for Q.
P only if Q.
Q if P .
Q provided P .
Q given P .
Q whenever P .
Q is a necessary condition for P .
 Section 1.1: Statements and proof methods 17

Unfortunately, the implication statement is not used consistently in informal spo-

ken language. For example, your grandmother may say: “You may have ice cream
if you eat your broccoli” when she means “You may have ice cream only if you eat
your broccoli.” Be nice to your grandmother and eat that broccoli even if she does
not express herself precisely because you know precisely what she means. But
in mathematics you do have to express yourself precisely! (Well, read the next
paragraph.)
Even in mathematics some shortcuts in precise expressions are acceptable. Here
is an example. The statements “An object x has property P if somethingorother
holds” and “An object x has property P if and only if somethingorother holds” (see
(5) below for “if and only if”) in general have different truth values and the proof
of the second is longer. However, the definition of what it means for an object
to have property P in terms of somethingorother is usually phrased with “if”, but
“if and only if” is meant. For example, the following is standard: “Definition:
A positive integer strictly bigger than 1 is prime if whenever it can be written as
a product of two positive integers, one of the two factors must be 1.” The given
definition, if read logically precisely, since it said nothing about numbers such as
4 = 2 · 2, would allow us to call 4 prime. However, it is an understood shortcut
that only the numbers with the stated property are called prime.
(5) Equivalence or the logical biconditional of P and Q stands for the compound
statement (P ⇒ Q) and (Q ⇒ P ). It is abbreviated “P ⇔ Q” or “P iff Q”, and is
true precisely when P and Q have the same truth values.
For example, for real numbers x and y, the statement “x ≤ y+1” is equivalent
to “x − 1 ≤ y.” Another example: “2x = 4x2 ” is equivalent to “x = 2x2 ,”
but it is not equivalent to “1 = 2x.” (Say why!)
We now backtrack on the truth values of P ⇒ Q. We can certainly fill in some
parts without qualms, leaving some unknown truth values x and y:

P Q P ⇒Q Q⇒P P ⇔Q
T T T T T
T F F x F
F T x F F
F F y y T
Since the last column above is the conjunction of the previous two, the last line
forces the value of y to be T . If x equals F , then the truth values of P ⇒ Q are the
same as the truth values of P ⇔ Q, which would say that the statements P ⇒ Q
and P ⇔ Q are logically the same. But this cannot be: “If r > 0 then r ≥ 0” is
true whereas “If r ≥ 0 then r > 0” is false. So this may convince you that the
18 Chapter 1: How we will do mathematics

truth values for the third and the fifth column have to be distinct, and this is only
possible if x is T .

Here is the truth table for all the connectives so far:

P Q not P P and Q P or Q P xor Q P ⇒Q P ⇔Q
T T F T T F T T
T F F F T T F F
F T T F T T T F
F F T F F F T T

One can form more elaborate truth tables if we start not with two statements P
and Q but with three or more. Examples of logically compounding P, Q, and R
are: P and Q and R, (P and Q) ⇒ Q, et cetera. For manipulating three statements,
we would fill a total of 8 rows of truth values, for four statements there would be
16 rows, and so on.

(6) Proof of P is a series of steps (in statement form) that establish beyond doubt that
P is true under all circumstances, weather conditions, political regimes, time of
day... The logical reasoning that goes into mathematical proofs is called deductive
reasoning. Whereas both guessing and intuition can help you find the next step
in your mathematical proof, only the logical parts are trusted and get written
down. Proofs are a mathematician’s most important tool; the book contains many
examples, and the next few pages give some examples and ideas of what proofs
are.
Ends of proofs are usually marked by , , //, or QED (for “quod erat demon-
strandum”, which is Latin for “that which was to be proved”). It is a good idea to
mark the completions of proofs especially when they are long or with many parts
and steps — that helps the readers know that nothing else is to be added.
The most trivial proofs simply invoke a definition or axiom, such as “An even
integer is of the form 2 times an integer,” “An odd integer is of the form 1 plus 2
times an integer,” or, “A positive integer is prime if whenever it can be written
as a product of two positive integers, one of the two factors is 1.”
Another type of proof consists of filling in a truth table. For example, P or ( not P )
is always true, no matter what the truth value of P is, and this can be easily verified
with the truth table:
P not P P or not P
T F T
F T T
 Section 1.1: Statements and proof methods 19

A formula using logical statements that is always true is called a tautology.

So P or not P is a tautology. Here is another example of tautology: ((P ⇒
Q) and P ) ⇒ Q, and it is proved below with the truth table:
P Q P ⇒Q (P ⇒ Q) and P ((P ⇒ Q) and P ) ⇒ Q
T T T T T
T F F F T
F T T F T
F F T F T

This particular tautology is called modus ponens, and its most famous example
is the following:
Every man is mortal. (If X is a man/human, then X is mortal.)
Socrates is human.
Therefore, Socrates is mortal.
Here is a more mathematical example of modus ponens:
Every differentiable function is continuous.
f is differentiable.
Therefore, f is continuous.
Another tautology is modus tollens: ((P ⇒ Q) and ( not Q)) ⇒ ( not P ). To
prove it, one constructs a truth table as before for modus ponens. — It is a
common proof technique to invoke the similarity principle with previous work
that allows one to not carry out all the steps, as I just did. However, whenever
you invoke the proof-similarity principle, you better be convinced in your mind
that the similar proof will indeed do the job; if you have any doubts, show all work
instead! In this case, I am sure that the truth table does the job, but if you are
seeing this for the first time, you may want to do the actual truth table explicitly
to get a better grasp on these concepts.
Here is a mathematical example of modus tollens:
Every differentiable function is continuous.
f is not continuous.
Therefore, f is not differentiable.
Here is another example on more familiar ground:
If you are in Oregon, then you are in the USA.
You are not in the USA.
Therefore, you are not in Oregon.

Some proofs can be pictorial/graphical. Here we prove with this method that for
any real numbers x and y, |x| < y if and only if −y < x < y. (We will see many
20 Chapter 1: How we will do mathematics

uses of absolute values.) Proof: [For a biconditional P ⇔ Q we need to

prove P ⇒ Q and Q ⇒ P .] The assumption |x| < y implies that y must be
positive, and the assumption −y < x < y implies that −y < y, which also says
that y must be positive. So, with either assumption, we can draw the following
part of the real number line:

−y 0 y
Now, by drawing, the real numbers x with |x| < y are precisely those real num-
bers x with −y < x < y. A fancier way of saying this is that |x| < y if and only if
−y < x < y.

Similarly, for all real numbers x and y, |x| ≤ y if and only if −y ≤ x ≤ y. (Here,
the word “similarly” is a clue that I am invoking the proof-similarity principle,
and a reader who wants to practice proofs or is not convinced should at this point
work through a proof by mimicking the steps in the previous one.)
Some (or actually most) proofs invoke previous results without re-doing the previ-
ous work. In this way we prove the triangle inequality, which asserts that for all
real numbers x and y, |x ± y| ≤ |x| + |y|. (By the way, we will use the triangle in-
equality intensely, so understand it well.) Proof: Note that always −|x| ≤ x ≤ |x|,
−|y| ≤ ±y ≤ |y|. Since the sum of smaller numbers is always less than or equal to
the sum of larger numbers, we then get that −|x| − |y| ≤ x ± y ≤ |x| + |y|. But
−|x| − |y| = −(|x| + |y|), so that −(|x| + |y|) ≤ x ± y ≤ |x| + |y|. But then by the
previous result, |x ± y| ≤ |x| + |y|.

Most proofs require a combination of methods. Here we prove that whenever x is

a real number with |x − 5| < 4, then |x3 − 3x| < 1030. Proof: The following is
standard formatting that you should adopt: first write down the left side of the
desired inequality (|x3 − 3x|), then start manipulating it algebraically, in interme-
diate steps add a clever 0 here and there, multiply by a clever 1 here and there,
rewrite, simplify, make it less than or equal to something else, and so on, every
step should be either obvious or justified on the right, until at the end you get the
quantity on the right (1030):
|x3 − 3x| ≤ |x3 | + |3x| (by the triangle inequality)
= |x|3 + 3|x|
= |x − 5 + 5|3 + 3|x − 5 + 5| (by adding a clever 0)
≤ (|x − 5| + 5)3 + 3(|x − 5| + 5) (by the triangle inequality
and since a ≤ b implies that a3 ≤ b3 )
 Section 1.1: Statements and proof methods 21

≤ (4 + 5)3 + 3(4 + 5) (since by assumption |x − 5| < 4)

= 93 + 3 · 9
< 103 + 3 · 10
= 1030.

Here is another pictorial proof, this one establishing the basis of trigonometry and
A
the definition of slope as rise over run: with the picture below, B = ab .

B
b

a
| {z }
A
Proof: The areas of the big and small triangles are 12 AB and 21 ab, and the area of
the difference is base times the average height, i.e., it is (A − a) b+B
2 . Thus
1 1 b+B
AB = ab + (A − a) .
2 2 2
By multiplying through by 2 we get that AB = ab + (A − a)(b + B) = ab + Ab +
AB − ab − aB, so that, after cancellations, Ab = aB. Then, after dividing through
A
by bB we get that B = ab .

Another common method of proof of a statement P is proof by contradiction:

you assume that not P is true and by using only correct logical and mathematical
steps you derive some nonsense (contradiction). This then says that not P is false,
so that not ( not P ) = P must be true. Beware: proofs by contradiction are in
general not considered elegant, nevertheless, they can be very powerful.
√
(Due to Pythagoras.) 2 is not a rational number. (A number is rational if
it is a ratio of two whole integers, where the denominator is not zero.) Proof
√ √
by contradiction: Suppose that 2 is rational. This means that 2 = ab for
some whole numbers a and b with b non-zero. Let d be the greatest common
divisor of a and b. Write a = a0 d and b = b0 d for some integers a0 , b0 6= 0.
√ √
Then 2 = ab = ab00 , so b0 2 = a0 , and so by squaring both sides we get
that 2b20 = a20 . So a20 = 2b20 is an even number, which by Exercise 1.1.18
means that a0 is an even number. Write a0 = 2a1 for some integer a1 . Then
4a21 = a20 = 2b20 , so that b20 = 2a21 is even, whence again b0 is even. But then
2 divides both a0 and b0 , so that 2d divides both a and b, which contradicts
22 Chapter 1: How we will do mathematics

the assumption that d was the greatest common divisor of a and b. Thus it
√
is not the case that 2 is rational, so it must be irrational.

Exercises for Section 1.1

1.1.1 Determine whether the following statements are true or false, and justify:
i) 3 is odd or 5 is even.
ii) If n is even, then 3n is prime.
iii) If 3n is even, then n is prime.
iv) If n is prime, then 3n is odd.
v) If 3n is prime, then n is odd.
vi) (P and Q) ⇒ P .
1.1.2 Sometimes statements are not written precisely enough. For example, “It is not the
case that 3 is prime and 5 is even” may be saying “ not (3 is prime and 5 is even),” or it
may be saying “( not (3 is prime)) and (5 is even).” The first option is true and the second
is false.
Similarly analyze several possible interpretations of the following ambiguous sentences:
i) If 6 is prime then 7 is even or 5 is odd.
ii) It is not the case that 3 is prime or if 6 is prime then 7 is even or 5 is odd.
General advice: Write precisely; aim to not be misunderstood.
1.1.3 Suppose that P ⇒ Q is true and that Q is false. Prove that P is false.
1.1.4 Prove that P ⇒ Q is equivalent to ( not Q) ⇒ ( not P ).
1.1.5 Fill in the following extra columns to the truth table in (5): P and not Q,
( not P ) and ( not Q), ( not P ) or ( not Q), P ⇒ not Q, ( not P ) ⇒ not Q. Are any of
the new columns negations of the columns in (5) or of each other?
1.1.6 Simplify the following statements:
i) (P and P ) or P .
ii) P ⇒ P .
iii) (P and Q) or (P or Q).
1.1.7 Prove, using truth tables, that the following statements are tautologies:
i) (P ⇔ Q) ⇔ [(P ⇒ Q) and (Q ⇒ P )].
ii) (P ⇒ Q) ⇔ (Q or not P ).
iii) (P and Q) ⇔ [P and (P ⇒ Q)].
iv) [P ⇒ (Q or R)] ⇔ [(P and not Q) ⇒ R].
1.1.8 Assume that P or Q is true and that R ⇒ Q is false. Determine with proof the truth
values of P, Q, R, or explain if there is not enough information.
 Section 1.1: Statements and proof methods 23

1.1.9 Assume that (P and Q) ⇒ R is false. Determine with proof the truth values of
P, Q, R, or explain if there is not enough information.
1.1.10 Suppose that x is any real number such that |x+2| < 3. Prove that |x3 −3x| < 200.
1.1.11 Suppose that x is any real number such that |x − 1| < 5. Find with proof a positive
constant B such that for all such x, |3x4 − x| < B.
1.1.12 (The quadratic formula) Let a, b, c be real numbers with a non-zero. Prove
(with algebra) that all solutions x of the quadratic equation ax2 + bx + c = 0 are of the
form
√
−b ± b2 − 4ac
x= .
2a

1.1.13 (Cf. Exercise 9.9.3.) Draw a unit circle and a line segment from the center to the
circle. Any real number x uniquely determines a point P on the circle at angle x radians
from the line. Draw the line from that point that is perpendicular to the first line. The
length of this perpendicular line is called sin(x), and the distance from the intersection of
the two perpendicular lines to the center of the circle is called cos(x). This is our definition
of cos and sin.
i) Consider the following picture inside the circle of radius 1:

y x

a) Label the line segments of lengths sin(x), sin(y), cos(y).

b) Use ratio geometry (from page 21) to assert that the smallest vertical line
in the bottom triangle has length sin(x) cos(y).
c) Use trigonometry and ratio geometry to assert that the vertical line in the
top triangle has length sin(y) cos(x).
d) Prove that sin(x + y) = sin(x) cos(y) + sin(y) cos(x).
1.1.14 In the spirit of the previous exercise, draw a relevant picture to prove cos(x + y) =
cos(x) cos(y) − sin(x) sin(y).
24 Chapter 1: How we will do mathematics

*1.1.15 (Logic circuits) Logic circuits are simple circuits which take as inputs logical
values of true and false (or 1 and 0) and give a single output. Logic circuits are composed
of logic gates. Each logic gate stands for a logical connective you are familiar with– it could
be and, or, or not (more complex logic circuits incorporate more). The shapes for logical
and, or, not are as follows:

Given inputs, each of these logic gates outputs values equal to the values in the associated
truth table. For instance, an “and” gate only outputs “on” if both of the wires leading
into it are “on”. From these three logic gates we can build many others. For example, the
following circuit is equivalent to xor.

input
output

input

A circuit that computes xor. The output on the right is on when

exactly one of the inputs on the left is on.

Make logic circuits that complete the following tasks. (It may be helpful to make logic
tables for each one.)
i) xor in a different way than the circuit above.
ii) “P implies Q” implication.
iii) xor for three inputs.
iv) Is a 3-digit binary number greater than 2?
v) Is a 4-digit binary string a palindrome?
1.1.16 Assuming that the area of the circle of radius r is πr 2 , convince yourself with
proportionality argument that the area of the region below, where x is measured in radians,
is 12 xr 2 .

r
x
 Section 1.2: Statements with quantifiers 25

†1.1.17 (Invoked in Theorem 9.9.4.) Let x be a small positive real number. Consider the
following picture with a circular segment of radius 1 and two right triangles:

i) Assert that the area of the small triangle is strictly smaller than the area of the
wedge, which in turn is strictly smaller than the area of the big triangle.
ii) Using the previous part and ratio geometry (from page 21), prove that
1
2
sin(x) cos(x) < 12 x < 12 tan x.
iii) Using the previous part, prove that 0 < cos(x) < sinx x < cos1 x .
1.1.18 (Odd-even integers)
i) Prove that the sum of two odd integers is an even integer.
ii) Prove that the product of two integers is odd if and only if the two integers are
both odd.
iii) Suppose that the product of two integers is odd. Prove that the sum of those two
integers is even.
iv) Suppose that the sum of the squares of two integers is odd. Prove that one of the
two integers is even and the other is odd.
v) Prove that the product of two consecutive integers is even. Prove that the product
of three consecutive integers is an integer multiple of 6.
vi) Prove that the sum of two consecutive integers is odd. Prove that the sum of three
consecutive integers is an integer multiple of 3.
√
1.1.19 Prove that 3 is not a rational number.

1.2 Statements with quantifiers

“The number x equals 1” is true for some x and false for some x. Thus “x = 1” is
not true or false universally. For determining a statement’s veracity we possibly need a
further qualification. We can use the universal quantifier “for all”, “for every”, or the
existential quantifier “there exists”, “for some”. Our example above could be modified
to one of the following:
(1) “There exists a real number x such that x = 1.”
(2) “For all real numbers x, x = 1,” which is logically the same as “For every real
number x, x = 1,” and even the same as “Every real number equals 1.”
Certainly the first statement is true and the second is false.
26 Chapter 1: How we will do mathematics

For shorthand we abbreviate “for all” with the symbol ∀, and “there exists” with
the symbol ∃. These abbreviations come in handy when we manipulate logical statements
further. The general forms of abbreviated statements with quantifiers are:
“ ∀x with a certain specification, P (x) holds” = “ ∀x P (x)”
“ ∃x with a certain specification, P (x) holds” = “ ∃x P (x)”

where P is some property that can be applied to objects x in question. The forms on the
left have an explicit specifications on the scope of the x, and in the forms on the right the
scope of the x is implicit.
Warning: For ease of readability it may be better to write out full words rather than
symbolic abbreviations.
We read the displayed statements above as “for all x, P of x [holds/is true]” and
“there exists x such that P of x [holds/is true]”, respectively. The part “such that” only
appears with the existential quantifier as a grammar filler but without any logical meaning;
it can be replaced with “for which”, and can sometimes be shortened further. For example:
“There exists a function f such that for all real numbers x, f (x) = f (−x)” can be rewritten
with equal meaning as “There exists a function f that is defined for all real numbers and
is even,” or even shorter as “There exists an even function.” (No “such that” appears in
the last two versions.)
Read the following symbolic statement (it defines the limit of the function f at a to
be L; see Definition 4.1.1):
∀ǫ > 0 ∃δ > 0 ∀x, 0 < |x − a| < δ ⇒ |f (x) − L| < ǫ.

When are the truth values and the negations of statements with quantifiers? We first
write a truth table with all the possible situations with regards to P in the first column,
and other columns give the truth values of the quantifier statements:
a possible situation ∀x P (x) ∀x not P (x) ∃x not P (x) ∃xP (x)
there are no x of specified type T vacuously T vacuously F F

there are x of specified type, T F F T

P true for all x

there are x of specified type, F T T F

P false for all x

there are x of specified type, P F F T T

true for some, false for some
 Section 1.2: Statements with quantifiers 27

A “for all” statement is true precisely when without exception all x with the given
description have the property P , and a “there exists” statement is true precisely when at
least one x with the given description satisfies property P . One proves a “for all” statement
by determining that each x with the given description has the property P , and one proves a
“there exists” statement by producing one specimen x with the given description and then
proving that that specimen has property P . If there are no x with the given specification,
then any property holds for those no-things x vacuously. For example, any positive real
number that is strictly smaller than −1 is also zero, equal to 15, greater than 20, product
of distinct prime integers, and any other fine property you can think of.
Notice that among the columns with truth values, one and three have opposite values,
and two and four have opposite values. This proves the following:

Proposition 1.2.1 The negation of “ ∀x P (x)” is “ ∃x not P (x)”. The negation of

“ ∃x P (x)” is “ ∀x not P (x).”

Thus “ ∀x P (x)” is false if there is even one tiny tiniest example to the contrary. “Every
prime number is odd” is false because 2 is an even prime number. “Every whole number
divisible by 3 is divisible by 2” is false because 3 is divisible by 3 and is not divisible by 2.

Remark 1.2.2 The statement “For all whole numbers x between 1/3 and 2/3, x2 is
irrational” is true vacuously. Another reason why “For all whole numbers x between 1/3
and 2/3, x2 is irrational” is true is that its negation, “There exists a whole number x
between 1/3 and 2/3 for which x2 is rational”, is false because there is no whole number
between 1/3 and 2/3: since the negation is false, we get yet more motivation to declare the
original statement true.

Exercises for Section 1.2

1.2.1 Show that the following statements are false by providing counterexamples.
i) No number is its own square.
ii) All numbers divisible by 7 are odd.
iii) The square root of all real numbers is greater than 0.
iv) For every real number x, x4 > 0.
1.2.2 Determine the truth value of the following statements, and justify your answers:
i) For all real numbers a, b, (a + b)2 = a2 + b2 .
ii) For all real numbers a, b, (a + b)2 = a2 + 2ab + b2 .
iii) For all real numbers x < 5, x2 > 16.
iv) There exists a real number x < 5 such that x2 < 25.
v) There exists a real number x such that x2 = −4.
28 Chapter 1: How we will do mathematics

vi) There exists a real number x such that x3 = −8.

vii) For every real number x there exists a positive integer n such that xn > 0.
viii) For every real number x and every integer n, |x| < xn .
ix) For every integer m there exists an integer n such that m + n is even.
x) There exists an integer m such that for all integers n, m + n is even.
xi) For every integer n, n2 − n is even.
xii) Every list of 5 consecutive integers has one element that is a multiple of 5.
xiii) Every odd number is a multiple of 3.
1.2.3 Explain why the following statements have the same truth values:
i) There exists x such that there exists y such that P holds for the pair (x, y).
ii) There exists y such that there exists x such that P holds for the pair (x, y).
iii) There exists a pair (x, y) such that P holds for the pair (x, y).
1.2.4 Explain why the following statements have the same truth values:
i) ∀x > 0 ∀y > 0, xy > 0.
ii) ∀y > 0 ∀x > 0, xy > 0.
iii) ∀(x, y), x, y > 0 implies xy > 0.
1.2.5 (Contrast with the switching of quantifiers in the previous two exercises.) Explain
why the following two statements do not have the same truth values:
i) For every x > 0 there exists y > 0 such that xy = 1.
ii) There exists y > 0 such that for every x > 0, xy = 1.
1.2.6 Rewrite the following statements using quantifiers:
i) 7 is prime.
ii) There are infinitely many prime numbers.
iii) Everybody loves Raymond.
iv) Spring break is always in March.
1.2.7 Let “xLy” represent the statement that x loves y. Rewrite the following statements
symbolically: “Everybody loves somebody,” “Somebody loves everybody,” “Somebody is
loved by everybody,” “Everybody is loved by somebody.” At least one statement should
be of the form “∀x ∃y, xLy”. Compare its truth value with that of “∃x ∀y, xLy”.
1.2.8 Find a property P of real numbers x, y, z such that “∀y ∃x ∀z, P (x, y, z)” and
“∀z ∃x ∀y, P (x, y, z)” have different truth values.
1.2.9 Suppose that it is true that there exists x of some kind with property P . Can we
conclude that all x of that kind have property P ? (A mathematician and a few other jokers
are on a train and see a cow through the window. One of them generalizes: “All cows in
this state are brown,” but the mathematician corrects: “This state has one cow one of
whose sides is brown.”)
 Section 1.3: More proof methods, and negation 29

1.3 More proof methods, and negation

When statements are compound, they can be harder to prove. Fortunately, proofs can
be broken down into simpler statements. Here is an essential chart of this breaking down.

Statement How to prove it

P (via contradiction). Suppose not P . Establish some nonsense
(that makes not P thus impossible so that P
must hold).

P and Q. Prove P . Prove Q.

P or Q. Suppose that P is false. Then prove Q.

Alternatively: Suppose that Q is false and
then prove P .
(It may even be the case that P is true always.
Then simply prove P . Or simply prove Q.)

If P then Q. Suppose that P is true. Then prove Q.

Contrapositively: Suppose that Q is false.
Prove that P is false.

P ⇔ Q. Prove P ⇒ Q. Prove Q ⇒ P .

For all x of a specified type, prop- Let x be arbitrary of the specified type. Prove
erty P holds for x. that property P holds for x.
(Possibly break up into a few subcases.)

There exists x of a specified type Find/construct an x of the specified type.

such that property P holds for x. Prove that property P holds for x.
Alternatively, invoke a theorem guaranteeing
that such x exists.

An element x of a specified type Suppose that x and x′ are both of specified

with property P is unique. type and satisfy property P . Prove that
x = x′ .
Alternatively, show that x is the only solution
to an equation, or the only element on a list,
or ....
30 Chapter 1: How we will do mathematics

Example 1.3.1 Prove that 2 and 3 are prime integers. (Prove that 2 is a prime integer
and that 3 is a prime integer.)

Proof. Let m and n be whole numbers strictly greater than 1. If m · n = 2, then 1 <
m, n ≤ 2, so m = n = 2, but 2 · 2 is not equal to 2. Thus 2 cannot be written as a product
of two positive numbers different from 1, so 2 is a prime number. If instead m · n = 3, then
1 < m, n ≤ 3. Then all combinations of products are 2 · 2, 2 · 3, 3 · 2, 3 · 3, none of which
is 3. Thus 3 is a prime number.

Example 1.3.2 Prove that a positive prime number is either odd or it equals 2. (One
may think of −2 as a prime number as well, that is why “positive” appears. But often the
term “prime” implicitly assumes positivity.)

Proof. Let p be a positive prime number. Suppose that p is not odd. Then p must be even.
Thus p = 2 · q for some positive whole number q. Since p is a prime, it follows that q = 1,
so that p = 2.

Example 1.3.3 Prove that if an integer is a multiple of 2 and a multiple of 3, then it is

a multiple of 6. (Implicit here is that the factors are integers.)

Proof. Let n be an integer that is a multiple of 2 and of 3. Write n = 2 · p and n = 3 · q for

some integers p and q. Then 2 · p = 3 · q is even, which forces that q must be even. Hence
q = 2 · r for some integer r, so that n = 3 · q = 3 · 2 · r = 6 · r. Thus n is a multiple of 6.

Example 1.3.4 Prove that for all real numbers x, x2 = (−x)2 .

Proof. Let x be an arbitrary real number. Then (−x)2 = (−x)(−x) = (−1)x(−1)x =

(−1)(−1)xx = 1 · x2 = x2 .

Example 1.3.5 Prove that there exists a real number x such that x3 − 3x = 2.

Proof. Observe that 2 is a real number and that 23 − 3 · 2 = 2. Thus x = 2 satisfies the
conditions.
 Section 1.3: More proof methods, and negation 31

Example 1.3.6 Prove that there exists a real number x such that x3 − x = 1.

Proof. Observe that f (x) = x3 − x is a continuous function. Since 1 is strictly between

f (0) = 0 and f (2) = 6, by the Intermediate value theorem (Theorem 5.3.1 in these notes)
[invoking a theorem rather than constructing x, as opposed to in the previ-
ous example]* there exists a real number x strictly between 0 and 2 such that f (x) = 1.

Example 1.3.7 (Mixture of methods) For every real number x strictly between 0 and 1
there exists a positive real number y such that x1 + y1 = xy
1
.

Proof. [We have to prove that for all x as specified some property holds.]
Let x be in (0, 1). [For this x we have to find y ...] Set y = 1 − x. [Was this
a lucky find? No matter how we got inspired to determine this y, we now
verify that the stated properties hold for x and y.] Since x is strictly smaller
than 1, it follows that y is positive. Thus also xy is positive, and y + x = 1. After dividing
the last equation by the positive number xy we get that x1 + y1 = xy 1
.
Furthermore, y in the previous example is unique: it has no choice but to be 1 − x.

Example 1.3.8 (The Fundamental theorem of arithmetic) Any positive integer

n > 1 can be written as pa1 1 pa2 2 · · · pakk for some positive prime integers p1 < · · · < pk and
some positive integers a1 , . . . , ak . (Another standard part of the Fundamental theorem of
arithmetic is that the pi and the ai are unique, but we do not prove that. Once you are
comfortable with proofs you can prove that part yourself.)

Proof. (You may want to skip this proof for now.) Suppose for contradiction that the
conclusion fails for some positive integer n. Then on the list 2, 3, 4, . . . , n let m be the
smallest integer for which the conclusion fails. If m is a prime, take k = 1 and p1 = m,
a1 = 1, and so the conclusion does not fail. Thus m cannot be a prime number, and so m =
m1 m2 for some positive integers m1 , m2 strictly bigger than 1. Necessarily 2 ≤ m1 , m2 < m.
By the choice of m, the conclusion is true for m1 and m2 . Write m1 = pa1 1 pa2 2 · · · pakk and
m2 = q1b1 q2b2 · · · qlbl for some positive prime integers p1 < · · · < pk , q1 < · · · < qk and some
positive integers a1 , . . . , ak , b1 , . . . , bl . Thus m = pa1 1 pa2 2 · · · pakk q1b1 q2b2 · · · qlbl is a product
of positive prime numbers, and after sorting and merging the pi and qj , the conclusion
follows also for m. But we assumed that the conclusion fails for m, which yields the
desired contradiction. Hence the conclusion does not fail for any positive integer.

* Recall that this font in brackets indicates the reasoning that should go on in the background in your head;
these statements are not part of a proof.
32 Chapter 1: How we will do mathematics

Example 1.3.9 Any positive rational number can be written in the form ab , where a and b
are positive whole numbers, and in any prime factorizations of a and b as in the previous
example, the prime factors for a are distinct from the prime factors for b.

Proof. [We have to prove that for all ...] Let x be a(n arbitrary) positive rational
number. [We rewrite the meaning of the assumption next in a more concrete
and usable form.] Thus x = ab for some whole numbers a, b. If a is negative, since x
is positive necessarily b has to be negative. But then −a, −b are positive numbers, and
x = −a
−b . Thus by possibly replacing a, b with −a, −b we may assume that a, b are positive.
[A rewriting trick.] There may be many different pairs of a, b, and we choose a pair for
which a is the smallest of all possibilities. [A choosing trick. But does the smallest
a exist?] Such a does exist because in a collection of given positive integers there is always
a smallest one. Suppose that a and b have a (positive) prime factor p in common. Write
a = a0 p and b = b0 p for some positive whole numbers a0 , b0 . Then x = ab00 , and since
0 < a0 < a, this contradicts the choice of the pair a, b. Thus a and b could not have had a
prime factor in common.

In order to be able to prove statements effectively, we often have to suppose the

negation of a part, say for proving statements with “or” and for proofs by contradiction.
Work and think through the following negations:
Statement Negation
not P P

P and Q not (P and Q) = ( not P ) or ( not Q)

P or Q not (P or Q) = ( not P ) and ( not Q)

P ⇒Q not (P ⇒ Q) = P and ( not Q)

P ⇔Q not (P ⇔ Q) = P ⇔ ( not Q) = ( not P ) ⇔ Q

For all x of a specified type, prop- There exists x of the specified type such that
erty P holds for x. P is false for x.

There exists x of a specified type For all x of the specified type, P is false for x.
such that property P holds for x.

Warning: The negation of a conditional statement is not another conditional statement!

Practice this one!
 Section 1.3: More proof methods, and negation 33

Example 1.3.10 There are infinitely many (positive) prime numbers. Proof by contra-
diction: (Due to Euclid.) Suppose that there are only finitely many prime numbers. Then
we can enumerate them all: p1 , p2 , . . . , pn . Let a = (p1 p2 · · · pn ) + 1. Since we know
that 2, 3, 5 are primes, necessarily n ≥ 3 and so a > 1. By the Fundamental theorem of
arithmetic (Example 1.3.8), a has a prime factor p. Since p1 , p2 , . . . , pn are all the primes,
necessarily p = pi for some i. But then p = pi divides a and p1 p2 · · · pn , whence it divides
1 = a − (p1 p2 · · · pn ), which is a contradiction. So it is not the case that there are only
finitely many prime numbers, so there must be infinitely many.
Another proof by contradiction: (Due to R. Meštrović, American Mathematical
Monthly 124 (2017), page 562.) Suppose that there are only finitely many prime numbers.
Then we can enumerate them all: p1 = 2, p2 = 3, . . . , pn . The positive integer p2 p3 · · · pn −2
has no odd prime factors, and since it is odd, it must be equal to 1. Hence p2 p3 · · · pn = 3,
which is false since p2 = 3, p3 = 5, and so on.

Exercises for Section 1.3

1.3.1 Prove that every whole number is either odd or even.
1.3.2 Prove that the successor of any odd integer is even.
1.3.3 Prove that if n is an even integer, then either n is a multiple of 4 or n/2 is odd.
√
1.3.4 Prove that if 0 < x < 1, then x2 < x < x.
√
1.3.5 Prove that if 1 < x, then x < x < x2 .
5
1.3.6 Prove that there exists a real number x such that x2 + 12 x = 16 .
1.3.7 Prove that the following pairs of statements are negations of each other:
i) P ⇔ Q.
(P and Q) or ( not P and not Q).
ii) P and (P ⇒ Q).
not P or not Q.
iii) (P ⇒ not Q) and (R ⇒ Q).
(P and Q) or (R and not Q).
iv) P and (Q or not R).
P ⇒ ( not Q and R).
1.3.8 Why are “f is continuous at all points” and “f is not continuous at 3” not negations
of each other?
1.3.9 Why are “Some continuous functions are differentiable” and “All differentiable func-
tions are continuous” not negations of each other?
34 Chapter 1: How we will do mathematics

1.3.10 Why is “P ⇒ not Q” not the negation of “P ⇒ Q”?

1.3.11 Negate the following statements:
i) The function f is continuous at 5.
ii) If x > y then x > z.
iii) For every ǫ > 0, there exists δ > 0 such that f (δ) = ǫ.
iv) For every ǫ > 0, there exists δ > 0 such that for all x, f (x · δ) = ǫx.
v) For every ǫ > 0, there exists δ > 0 such that for all x, 0 < |x − a| < δ implies
|f (x) − L| < ǫ.
1.3.12 Find at least three functions f such that for all real numbers x, f (x2 ) = x2 .
1.3.13 Prove that f (x) = x is the unique function that is defined for all real numbers and
that has the property that for all x, f (x3 ) = x3 .
1.3.14 Prove that the following are false:
i) For every real number x there exists a real number y such that xy = 1.
ii) 5n + 2 is prime for all non-negative integers n.
iii) For every real number x, if x2 > 4 then x > 2.
√
iv) For every real number x, x2 = x.
v) The cube of every real number is positive.
vi) The square of every real number is positive.
1.3.15 Prove that there exists a real number y such that for all real numbers x, xy = y.
1.3.16 Prove that there exists a real number y such that for all real numbers x, xy = x.
1.3.17 Prove that for every real number x there exists a real number y such that x+y = 0.
1.3.18 Prove that there exists no real number y such that for all real numbers x, x+y = 0.
1.3.19 Prove that for every real number y there exists a real number x such that x+y 6= 0.

1.4 Summation

There are many reasons for not writing out the sum of the first hundred numbers
in full length with 100 numbers and 99 plus signs: it would be too long, it would not
be any clearer, we would probably start doubting the intelligence of the writer, it would
waste paper and ink... The clear and short way of writing such a long sum is with the
summation sign Σ:
100
X 100
X
k or n.
k=1 n=1

The counters k and n above are dummy variables, they vary from 1 to 100. We could use any
other name in place of k or n. In general, if f is a function defined at m, m+1, m+2, . . . , n,
 Section 1.4: Summation 35

we use the summation sign for shortening as follows:

n
X
f (k) = f (m) + f (m + 1) + f (m + 2) + · · · + f (n).
k=m

This is one example where a good notation saves space (and it even sometimes clarifies
the concept). Typographically, when summation is displayed, the two bounds (m and n)
appear below and above the summation sign, but when the summation is in-line, the two
Pn
bounds appear to the right of the sign, like so: k=m f (k). (This prevents lines jamming
into each other.)
Now is a good time to discuss polynomials. A polynomial function is a function of
the form f (x) = a0 + a1 x + · · · + an xn for some non-negative integer n and some numbers
a0 , a1 , . . . , an . The largest d for which ad is non-zero is called the degree of the polynomial.
It is convenient to write this polynomial with the shorthand notation
n
X
n
f (x) = a0 + a1 x + · · · + an x = ak xk .
k=0

Here, of course, x0 stands for 1. When we evaluate f at 0, we get a0 = a0 +a1 ·0+· · ·+an ·0n
Pn
= k=0 ak 0k , and we deduce that notationally 00 stands for 1 here.

Remark 1.4.1 Warning: 00 could possibly be thought of also as lim+ 0x , which is surely
x→0
equal to 0. But then, is 00 equal to 0 or 1 or to something else entirely? Well, it turns
out that 00 is not equal to that zero limit – you surely know of other functions f for which
lim f (x) exists but the limit is not equal to f (c). (Consult also Exercise 7.5.7.)
x→c

Examples 1.4.2
5
X
(1) 2 = 10.
k=1
X5
(2) k = 15.
k=1
X4
(3) k 2 = 12 + 22 + 32 + 42 = 30.
k=1
X12
(4) cos(kπ) = cos(10π) + cos(11π) + cos(12π) = 1 − 1 + 1 = 1.
k=10
X2
(5) (4k 3 ) = −4 + 0 + 4 + 4 · 8 = 32.
k=−1
Xn
(6) 3 = 3n.
k=1
36 Chapter 1: How we will do mathematics
P0
We can even deal with empty sums such as k=1 ak : here the index starts at k = 1
and keeps increasing and we stop at k = 0, but there are no such indices k. What could
possibly be the meaning of such an empty sum? Note that
4
X 2
X 4
X 1
X 4
X 0
X 4
X
ak = ak + ak = ak + ak = ak + ak ,
k=1 k=1 k=3 k=1 k=2 k=1 k=1

or explicitly written out:

a1 + a2 + a3 + a4 = (a1 + a2 ) + (a3 + a4 ) = (a1 ) + (a2 + a3 + a4 ) = () + (a1 + a2 + a3 + a4 ),

from which we deduce that this empty sum must be 0. Similarly, every empty sum equals 0.

Similarly we can shorten products with the product sign Π:

n
Y
f (k) = f (m) · f (m + 1) · f (m + 2) · · · · · f (n).
k=m
Qn
In particular, for all non-negative integers n, the product k=1 k is used often and is
Qn
abbreviated as n! = k=1 k. See Exercise 1.4.6 for the fact that 0! = 1.

Exercises for Section 1.4

P4 P4
1.4.1 Compute k=0 (2k + 1), k=0 (k 2 + 2).
Pn Pn
1.4.2 Determine all non-negative integers n for which k=0 k= k=0 n.
1.4.3 Prove that
n
X n
X
i) c f (k) = cf (k).
k=m k=m
n
X n
X n
X
ii) f (k) + g(k) = (f (k) + g(k)).
k=m k=m k=m
P0 0·(−1)
1.4.4 Prove that k=1 k= 2
.
m−1
X n
X n
X
1.4.5 Prove that for all integers m ≤ n, f (k) + f (k) = f (k).
k=1 k=m k=1
1.4.6 Prove that the empty product equals 1. In particular, it follows that we can declare
0! = 1. This turns out to be very helpful notationally.
1.4.7 Prove:
5
Y
i) 2 = 32.
k=1
Y5
ii) k = 120.
k=1
 Section 1.5: Proofs by (mathematical) induction 37

1.5 Proofs by (mathematical) induction

So far we have learned a few proof methods. There is another type of proofs that
deserves special mention, and this is proof by (mathematical) induction, sometimes
referred to as the principle of mathematical induction. This method can be used when
one wants to prove that a property P holds for all integers n greater than or equal to an
integer n0 . Typically, n0 is either 0 or 1, but it can be any integer, even a negative one.
Induction is a two-step procedure:
(1) Base case: Prove that P holds for n0 .
(2) Inductive step: Let n > n0 . Assume that P holds for all integers n0 , n0 + 1, n0 +
2, . . . , n − 1. Prove that P holds for n.
Why does induction succeed in proving that P holds for all n ≥ n0 ? By the base case
we know that P holds for n0 . The inductive step then proves that P also holds for n0 + 1.
So then we know that the property holds for n0 and n0 + 1, whence the inductive step
implies that it also holds for n0 + 2. So then the property holds for n0 , n0 + 1 and n0 + 2,
whence the inductive step implies that it also holds for n0 + 3. This establishes that the
property holds for n0 , n0 + 1, n0 + 2, and n0 + 3, so that by inductive step it also holds for
n0 + 4. We keep going. For any integer n > n0 , in n − n0 step we similarly establish that
the inductive step holds for n0 , n0 + 1, n0 + 2, . . . , n0 + (n − n0 ) = n. Thus for any integer
n ≥ n0 , we eventually prove that P holds for it.
The same method can be phrased with a slightly different two-step process, with the
same result, and the same name:
(1) Base case: Prove that P holds for n0 .
(2) Inductive step: Let n > n0 . Assume that P holds for integer n − 1. Prove that
P holds for n.
Similar reasoning as in the previous case also shows that this induction principle
succeeds in proving that P holds for all n ≥ n0 .
Pn n(n+1)
Example 1.5.1 Prove the equality k=1 k= 2 for all n ≥ 1.
P1
Proof. Base case n = 1: The left side of the equation is k=1 k which equals 1. The right
side is 1(1+1)
2 which also equals 1. This verifies the base case.
Inductive step: Let n > 1 and we assume that the equality holds for n − 1. [We
want to prove the equality for n. We start with the expression on the
left (messier) side of the desired and not-yet-proved equation for n and
manipulate the expression until it resembles the desired right side.] Then
n n−1
!
X X
k= k +n
k=1 k=1
38 Chapter 1: How we will do mathematics

(n − 1)(n − 1 + 1)
= + n (by induction assumption for n − 1)
2
n2 − n 2n
= + (by algebra)
2 2
n2 + n
=
2
n(n + 1)
= ,
2
as was to be proved.
Pn
We can even prove the equality k=1 k = n(n+1) 2 for all n ≥ 0. Since we have already
P0
proved this equality for all n ≥ 1, it remains to prove it for n = 0. The left side k=1 k is
an empty sum and hence 0, and the right side is 0(0+1)2 , which is also 0.
Pn n(n+1)(n+2)(n+3)(n+4)
Example 1.5.2 Prove the equality k=1 k(k + 1)(k + 2)(k + 3) = 5
for all n ≥ 1.

Proof. Base case n = 1:

1
X
k(k + 1)(k + 2)(k + 3) = 1(1 + 1)(1 + 2)(1 + 3) = 1 · 2 · 3 · 4
k=1
1·2·3·4·5 1(1 + 1)(1 + 2)(1 + 3)(1 + 4)
= = ,
5 5
which verifies the base case.
Inductive step: Let n > 1 and we assume that the equality holds for n − 1. [We
want to prove the equality for n. We start with the expression on the
left side of the desired and not-yet-proved equation for n (the messier of
the two) and manipulate the expression until it resembles the desired right
side.] Then
n n−1
!
X X
k(k + 1)(k + 2)(k + 3) = k(k + 1)(k + 2)(k + 3) + n(n + 1)(n + 2)(n + 3)
k=1 k=1
(n − 1)(n − 1 + 1)(n − 1 + 2)(n − 1 + 3)(n − 1 + 4)
= + n(n + 1)(n + 2)(n + 3)
5
(by induction assumption)
(n − 1)n(n + 1)(n + 2)(n + 3) 5n(n + 1)(n + 2)(n + 3)
= + (by algebra)
5 5
n(n + 1)(n + 2)(n + 3)
= (n − 1 + 5) (by factoring)
5
n(n + 1)(n + 2)(n + 3)(n + 4)
= ,
5
as was to be proved.
 Section 1.5: Proofs by (mathematical) induction 39

Example 1.5.3 Assuming that the derivative of x is 1 and the product rule for deriva-
d
tives, prove that for all n ≥ 1, dx (xn ) = nxn−1 . (We introduce derivatives formally in
Section 6.1.)

Proof. We start the induction at n = 1. By calculus we know that the derivative of x1 is

1 = 1 · x0 = 1 · x1−1 , so equality holds in this case.
Inductive step: Suppose that equality holds for 1, 2, . . . , n − 1. Then
d n d
(x ) = (x · xn−1 )
dx dx
d d
= (x) · xn−1 + x (xn−1 ) (by the product rule of differentiation)
dx dx
n−1
=1·x + (n − 1)x · xn−2 (by induction assumption for 1 and n − 1)
= xn−1 + (n − 1)xn−1
= nxn−1 .

The following result will be needed many times, so remember it well.

Example 1.5.4 For any number x and all integers n ≥ 1,

(1 − x)(1 + x + x2 + x3 + · · · + xn ) = 1 − xn+1 .

Proof. When n = 1,
(1 − x)(1 + x + x2 + x3 + · · · + xn ) = (1 − x)(1 + x) = 1 − x2 = 1 − xn+1 ,
which proves the base case. Now suppose that equality holds for some integer n − 1 ≥ 1.
Then


(1 − x)(1 + x + x2 + · · · + xn−1 + xn ) = (1 − x) (1 + x + x2 + · · · + xn−1 ) + xn
= (1 − x)(1 + x + x2 + · · · + xn−1 ) + (1 − x)xn
= 1 − xn + xn − xn+1 (by induction assumption and algebra)
= 1 − xn+1 ,
which proves the inductive step.

Example 1.5.5 (Euclidean algorithm) Let f (x) and g(x) be non-constant polynomials.
Then there exist polynomials q(x) and r(x) such that f (x) = q(x) · g(x) + r(x) and such
that the degree of r(x) is strictly smaller than the degree of g(x).

Proof. We keep g(x) fixed and we prove that the claim holds for all polynomials f (x).
Write f (x) = a0 + a1 x + · · · + an xn for some numbers a0 , a1 , . . . , an and with an 6= 0,
and g(x) = b0 + b1 x + · · · + bm xm for some numbers b0 , b1 , . . . , bm and with bm 6= 0.
By assumption n, m ≥ 1. We proceed by induction on the degree n of f (x). If the
40 Chapter 1: How we will do mathematics

degree n of f (x) is strictly smaller than the degree m of g(x), then set q(x) = 0 and
r(x) = f (x). If n = m, then set q(x) = abnn and (necessarily) r(x) = f (x) − abnn g(x) =
a0 + a1 x + · · · + an xn − abnn (b0 + b1 x + · · · + bn xn ) = (a0 − abnn b0 ) + (a1 − abnn b1 )x + (a2 −
an 2 an n−1
bn b2 )x + · · · + +(an−1 − bn bn−1 )x . These are the base cases.
Now suppose that n > m. Set h(x) = a1 + a2 x + a3 x2 + · · · + an xn−1 . By induction on
n, there exist polynomials q1 (x) and r1 (x) such that h(x) = q1 (x) · g(x) + r1 (x) and such
that the degree of r1 (x) is strictly smaller than m. Then xh(x) = xq1 (x)·g(x)+xr1(x), and
by induction there exist polynomials q2 (x) and r2 (x) such that xr1 (x) = q2 (x)g(x) + r2 (x)
and such that the degree of r2 (x) is strictly smaller than m. Now set q(x) = xq1 (x) + q2 (x)
and r(x) = r2 (x) + a0 . Then the degree of r(x) is strictly smaller than m, and
q(x)g(x) + r(x) = xq1 (x)g(x) + q2 (x)g(x) + r2 (x) + a0
= xq1 (x)g(x) + xr1 (x) + a0
= xh(x) + a0
= f (x).
√
Example 1.5.6 For all positive integers n, n
n < 2.
√
Proof. Base case: n = 1, so n n = 1 < 2.
√
Inductive step: Suppose that n is an integer with n ≥ 2 and that n−1 n − 1 < 2. This
means that n − 1 < 2n−1 . Hence n < 2n−1 + 1 < 2n−1 + 2n−1 = 2 · 2n−1 = 2n , so that
√
n
n < 2.

Remark 1.5.7 There are two other equivalent formulations of mathematical induction for
proving a property P for all integers n ≥ n0 :
Mathematical induction, version III:
(1) Base case: Prove that P holds for n0 .
(2) Inductive step: Let n ≥ n0 . Assume that P holds for all integers n0 , n0 +
1, n0 + 2, . . . , n. Prove that P holds for n + 1.

Mathematical induction, version IV:

(1) Base case: Prove that P holds for n0 .
(2) Inductive step: Let n ≥ n0 . Assume that P holds for integer n. Prove that P
holds for n + 1.
Convince yourself that these two versions of the workings of mathematical induction
differ from the original two versions only in notation.
 Section 1.5: Proofs by (mathematical) induction 41

Exercises for Section 1.5: Prove the following properties for n ≥ 1 by induction.
Xn
n(n + 1)(2n + 1)
1.5.1 k2 = .
6
k=1
Xn  2
3 n(n + 1)
1.5.2 k = .
2
k=1
1.5.3 The sum of the first n odd integers is n2 .
Xn
1.5.4 (2k − 1) = n2 .
k=1
1.5.5 (Triangle inequality) For all real numbers a1 , . . . , an , |a1 + a2 + · · · + an | ≤ |a1 | +
|a2 | + · · · + |an |.
X n
1.5.6 (3k 2 − k) = n2 (n + 1).
k=1
1.5.7 1 · 2 + 2 · 3 + 3 · 4 + · · · + n(n + 1) = 31 n(n + 1)(n + 2).
1.5.8 7n + 2 is a multiple of 3.
1.5.9 3n−1 < (n + 1)!.
1 1 1 1 √
1.5.10 √ + √ + √ + · · · + √ ≥ n.
1 2 3 n
1 1 1 1 1
1.5.11 2 + 2 + 2 + · · · + 2 ≤ 2 − .
1 2 3 n n
1.5.12 Let a1 = 2, and for n ≥ 2, an = 3an−1 . Formulate and prove a theorem giving an
in terms of n (no dependence on other ai ).
1.5.13 8 divides 5n + 2 · 3n−1 + 1.
1.5.14 1(1!) + 2(2!) + 3(3!) + · · · + n(n!) = (n + 1)! − 1.
1.5.15 2n−1 ≤ n!.
Yn  
1 1
1.5.16 1− = .
k n
k=2
Yn  
1 n+1
1.5.17 1− 2 = .
k 2n
k=2
n
X
1.5.18 2k (k + 1) = 2n+1 n + 1.
k=0
P2n 1 n+2
1.5.19 k=1 k ≥ 2 .
P2n −1 1
Pn 1
† 1.5.20 (This is invoked in Example 9.1.9.) k=1 k2
≤ k=0 2k .
† 1.5.21 (This is invoked in the proof of Theorem 9.4.1.) For all numbers x, y, xn − y n =

P
(x − y) xn−1 + xn−2 y + xn−3 y 2 + · · · + y n−1 = (x − y) n−1
k=0 x
n−1−k k
y .
42 Chapter 1: How we will do mathematics

†1.5.22 (This is invoked in the Ratio tests Theorems 8.6.6 and 9.2.4.) Let r be a positive
real number. Suppose that for all positive integers n ≥ n0 , an+1 < ran . Prove that for all
positive integers n ≥ n0 , an+1 < r n−n0 an0 . (The strict inequality < can also be replaced
by ≤, >, ≥ throughout.)
1.5.23 Let An = 12 + 22 + 32 + · · · + (2n − 1)2 and Bn = 12 + 32 + 52 + · · · + (2n − 1)2 .
Find formulas for An and Bn , and prove them (by using algebra and previous problems,
and possibly not with induction).
1.5.24 (From the American Mathematical Monthly 123 (2016), page 87, by K. Gaitanas)
Pn−1 k 1
Prove that for every n ≥ 2, k=1 (k+1)! = 1 − n! .
1 1 1 1
1.5.25 Let An = 1·2 + 2·3 + 3·4 +···+ n(n+1) . Find a formula for An and prove it.
1.5.26 There are exactly n people at a gathering. Everybody shakes everybody else’s
hands exactly once. How many handshakes are there?
1.5.27 Pick a vertex V in a triangle. Draw n distinct lines from V to the opposite edge
of the triangle. If n = 1, you get the original triangle and two smaller triangles, for a total
of three triangles. Determine the number of distinct triangles obtained in this way with
arbitrary n.
1.5.28 Find with proof an integer n0 such that n2 < 2n for all integers n ≥ n0 .
1.5.29 Find with proof an integer n0 such that 2n < n! for all integers n ≥ n0 .
1.5.30 Let x be a real number. For any positive integer n define Sn = 1 +x +x2 +· · ·+xn .
i) Prove that for any n ≥ 2, xSn−1 + 1 = Sn , and that Sn (1 − x) = 1 − xn+1 .
ii) Prove that if x = 1, then Sn = n + 1.
n+1
iii) Prove that if x 6= 1, then Sn = 1−x
1−x . Compare with the proof by induction in
Example 1.5.4.
1.5.31 (Fibonacci numbers) Let s1 = 1, s2 = 1, and for all n ≥ 2, let sn+1 = sn + sn−1 .
This sequence starts with 1, 1, 2, 3, 5, 8, 13, 21, 34, . . .. (Many parts below are taken from the
book Fibonacci Numbers by N. N. Vorob’ev, published by Blaisdell Publishing Company,
1961, translated from the Russian by Halina Moss.)
i) Fibonacci numbers are sometimes “motivated” as follows. You get the rare gift of
a pair of newborn Fibonacci rabbits. Fibonacci rabbits are the type of rabbits who
never die and each month starting in their second month produce another pair of
rabbits. At the beginning of months one and two you have exactly that 1 pair of
rabbits. In the second month, that pair gives you another pair of rabbits, so at
the beginning of the third month you have 2 pairs of rabbits. In the third month,
the original pair produces another pair of rabbits, so that at the beginning of the
fourth month, you have 3 pairs of rabbits. Justify why the number of rabbits at
the beginning of the nth month is sn .
 Section 1.5: Proofs by (mathematical) induction 43
 √ n  √ n
ii) Prove that for all n ≥ 1, sn = √15 1+2 5 − √15 1−2 5 . (It may seem amazing
that these expressions with square roots of 5 always yield positive integers.) Note
that the base case requires proving this for n = 1 and n = 2, and that the inductive
step uses knowing the property for the previous two integers.
iii) Prove that s1 + s3 + s5 + · · · + s2n−1 = s2n .
iv) Prove that s2 + s4 + s6 + · · · + s2n = s2n+1 − 1.
v) Prove that s1 + s2 + s3 + · · · + sn = sn+2 − 1.
vi) Prove that s1 − s2 + s3 − s4 + · · · + s2n−1 − s2n = 1 − s2n−1 .
vii) Prove that s1 − s2 + s3 − s4 + · · · + s2n−1 − s2n + s2n+1 = s2n + 1.
viii) Prove that s1 − s2 + s3 − s4 + · · · √ + (−1)n+1 sn = (−1)n+1 sn−1 + 1.
ix) Prove that for all n ≥ 3, sn > ( 1+2 5 )n−2 .
x) Prove that for all n ≥ 1, s21 + s22 + · · · + s2n = sn sn+1 .
xi) Prove that sn+1 sn−1 − s2n = (−1)n .
xii) Prove that s1 s2 + s2 s3 + · · · + s2n−1 s2n = s22n .
xiii) Prove that s1 s2 + s2 s3 + · · · + s2n s2n+1 = s22n+1 − 1.
xiv) Prove that ns1 + (n − 1)s2 + (n − 2)s3 + · · · + 2sn−1 + sn = sn+4 − (n + 3).
xv) Prove that for all n ≥ 1 and all k ≥ 2, sn+k = sk sn+1 + sk−1 sn .
xvi) Prove that for all n, k ≥ 1, skn is a multiple of sn . (Use the previous part.)
xvii) Prove that s2n+1 = s2n+1 + s2n .
xviii) Prove that s2n = s2n+1 − s2n−1 .
xix) Prove that s3n = s3n+1 + s3n − s3n−1 .
 √   √ n
xx) Prove that sn+1 = 1+2 5 sn + 1−2 5 .
s3n+2 −1
xxi) Prove that s3 + s6 + s9 + · · · + s3n = 2 . (Use the previous part.)
s3n+2 +(−1)n+1 6sn−1 +5
xxii) Prove that s 31 + s32 + s33 + · · · + s3n = 10 .
 √
 (1 + 5)n  1
xxiii) Prove that sn − √ < .
 2n 5  2
(xxiv)* If you know a bit of number theory, prove that for all positive integers m, n, the
greatest common divisor of sm and sn is sgcd(m,n) .
(xxv)* Prove that sn is even if and only if n is a multiple of 3. Prove that sn is divisible
by 3 if and only if n is a multiple of 4. Prove that sn is divisible by 4 if and only
if n is a multiple of 6. Prove that sn is divisible by 5 if and only if n is a multiple
of 5. Prove that sn is divisible by 7 if and only if n is a multiple of 8. Prove that
there are no Fibonacci numbers that have the remainder of 4 when divided by 8.
Prove that there are no odd Fibonacci numbersthat are n divisible
 by 17. 
1 1 s sn
xxvi) If you know matrices, prove that for all n ≥ 2, = n+1 .
1 0 sn sn−1
44 Chapter 1: How we will do mathematics

1.5.32 (Via the grapevine, based on ideas of Art Benjamin, Harvey Mudd College, and
Dan Velleman, Amherst College) A tromino is a plane figure composed of three squares in
L-shape:

Prove that for every positive integer n, any 2n × 2n square grid with exactly one of the
squares removed can be tiled with trominoes.
1.5.33 (Spiral of Theodorus) Draw a triangle with vertices at (0, 0), (1, 0), (1, 1). The
√
hypotenuse has length 2.
i) One of the vertices of the hypotenuse is at (0, 0). At the other vertex of the
hypotenuse, draw an edge of length 1 at a right angle away from the first triangle.
Make a triangle from the old hypotenuse and this new edge. What is the length
of the hypotenuse of the new triangle?
ii) Repeat the previous step twice.
iii) Argue by induction that one can draw the square root of every positive integer.
1.5.34 (Tower of Hanoi) There are 3 pegs on a board. On one peg, there are n disks,
stacked from largest to smallest. The task is to move all of the disks from one peg to a
different peg, given the following constraints: you may only move one disk at a time, and
you may only place a smaller peg on a larger one (never a larger one on a smaller one).
Let Sn be the least number of moves to complete the task for n disks.
i) If n = 1, then what is the least number of moves it takes to complete the task?
What if there are 2 disks? Repeat for 3, 4, 5 disks.
ii) Make a recursive formula (defining Sn based on Sn−1 ) for this Sn . Then, make a
guess for a non-recursive formula for Sn (defining Sn based on n without invoking
Sn−1 ). Prove your guess using induction and the recursive formula that you wrote.
1.5.35 What is wrong with the following “proof by induction”? I will prove that 5n + 1
is a multiple of 4. Assume that this is true for n − 1. Then we can write 5n−1 + 1 = 4m
for some integer m. Multiply this equation through by 5 to get that
5n + 5 = 20m,
whence 5n + 1 = 4(5m − 1). As 5m − 1 is an integer, this proves that 5n + 1 is a multiple
of 4.

1.5.36 What is wrong with the following “proof by induction” besides the fact that the
conclusion is false for many n? I will prove that all horses are of the same color. This is the
same as saying that for any integer n ≥ 1 and any set of n horses, all the horses belonging
to the set have the same color. If n = 1, of course this only horse is the same color as
itself, so the base case is proved. Now let n > 1. If we remove one horse from this set, the
 Section 1.6: Pascal’s triangle 45

remaining n − 1 horses in the set are all of the same color by the induction assumption.
Now bring that one horse back into the set and remove another horse. Then again all of
these horses are of the same color, so the horse that was removed first is the same color as
all the rest of them.

1.6 Pascal’s triangle

Pascal’s triangle is very useful, so read this section with the exercises.
The following is rows 0 through 8 of Pascal’s triangle, and the pattern is obvious
for continuation into further rows:
row 0: . . . . . . . . . 1
row 1: . . . . . . . . 1 1
row 2: . . . . . . . 1 2 1
row 3: . . . . . . 1 3 3 1
row 4: . . . . . 1 4 6 4 1
row 5: . . . . 1 5 10 10 5 1
row 6: . . . 1 6 15 20 15 6 1
row 7: . . 1 7 21 35 35 21 7 1
row 8: . 1 8 28 56 70 56 28 8 1
Note that the leftmost and rightmost numbers in each row are all 1, and each of the other
numbers is the sum of the two numbers nearest to it in the row above it. We number the
slanted columns from left to right starting from 0: the 0th slanted column consists of all 1s,
the 1st slanted column consists of consecutive numbers 1, 2, 3, 4, . . ., the 2nd slanted column
consists of consecutive numbers 1, 3, 6, 10, . . ., and so on for the subsequent columns.


Let the entry in the nth row and kth column be denoted nk . We read this as “n
choose k”. These are loaded words, however, and we will eventually justify these words.
Pascal’s triangle is defined so that for all n ≥ 1 and all k = 0, 1, . . . , n − 1,
     
n n n+1
+ = .
k k+1 k+1


What would it take to compute 100 5
? It seems like we would need to write down rows
0 through 100 of Pascal’s triangle, or actually a little less, only slanted columns 0 through
5 of these 101 rows. That is too much drudgery! We will instead be smart mathematicians
and we will prove many properties of Pascal’s triangle in general, including shortcuts for


computing 100 5 . We will accomplish this through exercises, most of which can be proved
by mathematical induction.
46 Chapter 1: How we will do mathematics

Exercises for Section 1.6

1.6.1 Prove that the sum of the entries in the nth row is 2n .
1.6.2 Let k be an arbitrary non-negative integer. (This means that k is given to you not
as a specific number but as an unknown integer.) Prove by induction on n that the sum


of the entries in the slanted column k in rows j ≤ n is n+1
k+1 . In otherP
words, prove that
k

k+1
k+2


n


k + k + k +· · ·+ k = k+1 , which is the same as proving that j=k kj = n+1
n n+1
k+1 .

1.6.3 Prove that every integer n ≥ 0 has the property that for all k = 0, 1, 2, . . . , n,
n

n!
k = k!(n−k)! .


1.6.4 Prove that nk = n(n−1)(n−2)···(n−k+1)
k! .







100
100

1.6.5 Compute 42 , 52 , 62 , 72 , 82 , 100
2 , 100
3 , 4 , 5 .


1.6.6 Prove that nk is the number of possible k-member teams in a club with exactly n


members. For this reason nk is read n choose k.
† 1.6.7 (This is invoked in Theorem 2.8.2, Example 8.2.9, Theorem 6.2.3.) Prove that for
all non-negative integers n,
Xn  
n n k n−k
(a + b) = a b .
k
k=0

(Since a + b contains two summands, it is called a binomial, and the expansion of (a + b)n


is called the binomial expansion, with coefficients ni being called by yet another name
in this context: binomial coefficients.)
√
1.6.8 Express each of the following as a + b 2 for some integers a, b:
√ √ √ √ √
i) 2 − 1, ( 2 − 1)2 , ( 2 − 1)3 , ( 2 − 1)4 , ( 2 − 1)5 .
√ √
ii) Write each of the five expressions in the previous part in the form c − d for
some positive integers c, d.
(iii)* Do you see a relation between c and d for each expression in the previous part?
Is there a general rule? Can you prove it?
Pn
Pn

1.6.9 Prove that for all positive integers n, k=0 (−1)k nk = 0. Compute k=0 (−1)k nk
in case n = 0.
1.6.10 Prove that for any non-negative integer k,
n
X n(n + 1)(n + 2) · · · (n + k + 1)
j(j + 1)(j + 2) · · · (j + k) = .
j=1
k+1

(Hint: induction on n or instead use Exercises 1.6.2 and 1.6.4.)

 Section 1.6: Pascal’s triangle 47

1.6.11 Use Exercise 1.6.10 to get succinct simplifying formulas for

n
X n
X n
X n
X
j, j(j + 1), j(j + 1)(j + 2), j(j + 1)(j + 2)(j + 3).
j=1 j=1 j=1 j=1

i) Note that j 2 = j(j + 1) − j. Use the simplifications from above to prove that
Pn 2 n(n+1)(2n+1)
j=1 j = 6
.
ii) From j = j(j + 1)(j + 2) − 3j 2 − 2j = j(j + 1)(j + 2) − 3j(j + 1) + j develop the
3
Pn
formula for j=1 j 3 .
Pn
iii) Mimic the previous work to develop the formula for j=1 j 4 .

k
1.6.12 Prove that for all non-negative integers n and all k = 0, 1, . . . n, nk ≤ nk! .
1.6.13 Fix a positive integer k. Prove that there exists a positive number C such that for


all sufficiently large integers n, Cnk ≤ nk .


1.6.14 Give reasons why we should have nk = 0 for n < k or if either k or n is negative.
1.6.15 Let d be a positive integer. This is about summing entries in Pascal’s triangle
along the dth northwest-southeast slanted column: Prove by induction on n ≥ 0 that
Pn d+k

d+n+1


k=0 k = n .
*1.6.16 Prove that for all non-negative integers n,
Xn   !2 X  2n 
n−1
!2
2n k
2 −1=2 2k
2k 2k + 1
k=0 k=0

and !2
n 
X  X
n−1  !2
2n + 1 k 2n + 1 k
2 +1=2 2 .
2k 2k + 1
k=0 k=0
Pen n
k Pon n

k
For notation’s sake you may want to label En = k=0 2k 2 and On = k=0 2k+1 2 ,
where en , on are the largest integers such that 2en ≤ n and 2on + 1 ≤ n. The claim is then
2 2 2 2
that for all n ≥ 0, E2n − 1 = 2O2n and E2n+1 + 1 = 2O2n+1 . (Hint: use the definition
n


of k to rewrite En in terms of En−1 , On−1 . Proceed with induction.)
Chapter 2: Concepts with which we will do mathematics

2.1 Sets

What is a set? Don’t we already have an idea of what a set is? The following formal
definition is trying to make the intuitive idea precise:

Definition 2.1.1 A set is a collection of objects. These objects are called members or
elements of that set. If m is a member of a set A, we write m ∈ A, and also say that A
contains m. If m is not a member of a set A, we write m 6∈ A.

So, while perhaps the definition does not tell us much new about what we think of as
sets, it does tell us the notation and vocabulary with which we can write and talk about
sets. For example, the set of all polygons contains triangles, squares, rectangles, pentagons,
among other polygons. The set of all polygons does not contain circles or disks. The set of
all functions contains the trigonometric, logarithmic, exponential, constant functions, and
so on.

Examples and notation 2.1.2

(1) Intervals are sets:
(0, 1) is the interval from 0 to 1 that does not include 0, 1. (From the context you
should be able to distinguish between the interval (0, 1) and a point (0, 1) in the
plane.)
(0, 1] is the interval from 0 to 1 that does not include 0 but includes 1.
[0, 1) is the interval from 0 to 1 that does not include 1 but includes 0.
[0, 1] is the interval from 0 to 1 that includes 0, 1.
(5, ∞) is the interval of all real numbers strictly bigger than 5.
[5, ∞) is the interval of all real numbers bigger than or equal to 5.
(−∞, 5) is the interval of all real numbers strictly smaller than 5.
(−∞, 5] is the interval of all real numbers smaller than or equal to 5.
(−∞, ∞) is the set of all real numbers.
[a, a] is the “interval” consisting of precisely one number a.
(2) Descriptively we can say for example that a set A is the set of all prime numbers.
The description may be more loaded with symbols: A is the set of all real numbers x
 Section 2.1: Sets 49

with the property that x2 > 4x − 3. (Use the quadratic formula to prove that A
consists of the numbers that are strictly smaller than 1 or strictly larger than 3.)
(3) We can list the elements and surround them with curly braces to define a set:
(i) {1, 2, 3} is the set consisting of precisely 1, 2 and 3.
(i) {1, 2, 3, 2} is the set consisting of precisely 1, 2 and 3. Thus {1, 2, 3, 2} =
{1, 2, 3}.
(iii) {blue, “hello”, 5} is the set consisting precisely of the color blue, of the word
“hello”, and of number 5.
(iv) {1, 2, {1, 2}} is the set consisting of precisely of numbers 1 and 2 and of the
set {1, 2}. This set has exactly three distinct elements, it is not the same as {1, 2},
and it is not the same as {{1, 2}}.
(4) The set with no elements is called the empty set and is denoted ∅ or {}.
The set {∅} is not empty because it contains the empty set.
(5) When the list of elements is not small enough for reasonable explicit listing but
the pattern of elements is clear, we can start the list and then add “, . . .” when
the pattern is clear:
(i) {1, 2, 3, . . . , 10000} is the set of all positive integers that are at most 10000.
(ii) {1, 4, 9, . . . , 169} is the set of the first 13 squares of integers.
(iii) {0, 1, 2, 3, . . .} is the set consisting of all non-negative whole numbers. This
set is often denoted by N. WARNING: in some books, N stands for the set of
all positive whole numbers. To distinguish between the two, we will write N0 for
{0, 1, 2, 3, . . .} and N+ for {1, 2, 3, . . .}.
(6) Warning: {3, 5, 7, . . .} or {3, 5, 7, . . . , 101} could stand for the set of all odd primes
(up to 101), or possibly for the set of all odd whole numbers strictly greater
than 1 (up to 101). Avoid ambiguities: write more elements, or write an explicit
description of the elements instead.
(7) The set of all whole numbers is written Z, the set of all rational numbers is writ-
ten Q, the set of all real numbers is written R, and the set of all complex numbers is
written C (complex numbers are defined in Section 3.1, and until then do not worry
when “C” appears in the text). (I rely on your basic training to “understand” the
examples N, N0 , N+ , Z, Q, R, C.)
(8) We can also define sets propositionally: the set of all x (or x ∈ A) for which
property P holds can be we written as

{x : P (x)} or {x ∈ A : P (x)}.
Here are some explicit examples:
(i) {x ∈ R : x2 = x}, and this happens to be the set {0, 1}.
(ii) {x ∈ R : x > 0 and x < 1}, and this happens to be the interval (0, 1).
50 Chapter 2: Concepts with which we will do mathematics

(iii) Q = { ab : a, b ∈ Z, b 6= 0}.
(iv) {n ∈ N : n > 1 and if n = pq for some integers p, q then |p| = 1 or |q| = 1} is
the same as the set of all positive prime numbers.
(v) A = {x : x is a positive integer that equals the sum of its proper factors}.
Who are the elements of A? They have to be positive integers, and it is easy to
verify that 1, 2, 3, 4, 5 are not in A. But 6 has factors 1, 2, 3, 6, and the sum of
the factors 1, 2, 3 other than 6 equals 6. Thus 6 is an element of A. Verify that
28, 496, 8128 are also in A (and if you have a lot of time, program a computer to
verify that no other number smaller than 33 million is in A).
(9) Proving that a property P holds for all integers n ≥ n0 is the same as saying that
the set A = {n ∈ Z : P holds for n} contains {n0 , n0 + 1, n0 + 2, . . .}. By the
principle of mathematical induction, P holds for all integers n ≥ n0 is the same as
saying that n0 ∈ A and that n − 1 ∈ A implies that n ∈ A.

Summary of example sets, and their notation

∅ = {}: the set with no elements.
{a, b, c}, {a, b, . . . , z}.
{x : x can be written as a sum of three consecutive integers}.
N: the set of all Natural numbers. Depending on the book, this could be the set of all
positive integers or it could be the set of all non-negative integers. The symbols below are
unambiguous:
N0 : the set of all non-negative integers;
N+ : the set of all positive integers.
Z: the set of all integers (“Zahlen” in German).
Q: the set of all rational numbers (Quotients).
R: the set of all Real numbers.
C: the set of all Complex numbers (more about them starting in Section 3.1).

Just like numbers, functions, and logical statements, sets and their elements can also
be related and combined in meaningful ways. The list below introduces quite a few new
concepts that may be overwhelming at first, but in a few weeks you will be very comfortable
with them.

Subsets: A set A is a subset of a set B if every element of A is an element of B. In that

case we write A ⊆ B. For example, N+ ⊆ N0 ⊆ Z ⊆ Q ⊆ R. The non-subset relation
is expressed with the symbol 6⊆: R 6⊆ N.
Every set is a subset of itself, i.e., for every set A, A ⊆ A.
The empty set is a subset of every set, i.e., for every set A, ∅ ⊆ A.
 Section 2.1: Sets 51

If A is a subset of B and A is not equal to B (so B contains at least one element that
is not in A), then we say that A is a proper subset of B, and we write A ( B. For
example, N+ ( N0 ( Z ( Q ( R.

Equality: Two sets are equal if they consist of exactly the same elements. In other words,
A = B if and only if A ⊆ B and B ⊆ A.

Intersection: The intersection of sets A and B is the set of all objects that are in A
and in B:
A ∩ B = {x : x ∈ A and x ∈ B}.

When A ∩ B = ∅, we say that A and B are disjoint.

Union: The union of sets A and B is the set of all objects that are either in A or in B:

A ∪ B = {x : x ∈ A or x ∈ B}.

Intersections and unions of arbitrary families of sets: We have seen intersections

and unions of two sets at a time. We can also take intersections and unions of three,
four, five, and even infinitely many sets at a time. Verify the equalities below:
(A ∩ B) ∩ C = A ∩ (B ∩ C),
(A ∪ B) ∪ C = A ∪ (B ∪ C),
(A ∩ B) ∩ (C ∩ D) = A ∩ (B ∩ C ∩ D), etc
(A ∪ B) ∪ (C ∪ D) = (A ∪ (B ∪ C)) ∪ D, etc
(Verification of the first equality above: Let x ∈ (A ∩ B) ∩ C. This holds if and only
if x ∈ A ∩ B and x ∈ C, which holds if and only if x ∈ A, x ∈ B and x ∈ C, which in
turn holds if and only if x ∈ A and x ∈ (B ∩ C), i.e., if and only if x ∈ A ∩ (B ∩ C).)
Thus having established that parentheses above are irrelevant, we simply write the
four sets above as A ∩ B ∩ C, A ∪ B ∪ C, A ∩ B ∩ C ∩ D, A ∪ B ∪ C ∪ D, respectively.
More generally, given sets A1 , A2 , . . . , An , we write
n
\
Ak = A1 ∩ A2 ∩ · · · ∩ An = {a : a ∈ Ak for all k = 1, . . . , n},
k=1
[n
Ak = A1 ∪ A2 ∪ · · · ∪ An = {a : there exists k = 1, . . . , n such that a ∈ Ak },
k=1

or even more generally, when Ak are sets as k varies over a possibly infinite index set I,
then
\
Ak = {a : a ∈ Ak for all k ∈ I},
k∈I
52 Chapter 2: Concepts with which we will do mathematics
[
Ak = {a : there exists k ∈ I such that a ∈ Ak }.
k∈I

When I is the empty index set, one can argue as for empty sums and empty products
in Section 1.4 that
\
Ak = the universal set that contains all the Ak ,
k∈∅
[
Ak = ∅.
k∈∅

We return to this theme in Section 2.5. Note that the k in the subscripts are referred
to as indices of unions or intersections.
Complement: The complement of A in B is
B \ A = {b ∈ B : b 6∈ A}.

Some authors write B − A, but that has another meaning as well: B − A : {b − a : b ∈

B, a ∈ A}. Always try to use precise and unambiguous notation.
We often have an implicit or explicit universal set that contains all elements of our
current interest. Perhaps we are talking only about real numbers, or perhaps we are
talking about all functions defined on the interval [0, 1] with values being real numbers.
In that case, for any subset A of the universal set U , the complement of A is the
complement of A in U , thus U \ A, and this is denoted as Ac .

Summary notation and vocabulary

a ∈ A: a is an element of a set A.
A ⊆ B: A is a subset of set B; every element of A is an element of B.
A ( B: A ⊆ B and A 6= B; A is a proper subset of B.
A = B: A ⊆ B and B ⊆ A.
A ∩ B: the set of all elements that are in A and in B.
A and B are disjoint: A ∩ B = ∅.
A ∪ B: the set of all elements that are either in A or in B.
A \ B: the set of all elements of A that are not in B.
Ac : the set of all elements in the universal set that are not in A.

Example 2.1.3 We prove that Z = {3m + 4n : m, n ∈ Z}. Certainly for any integers m
and n, 3m + 4n is also an integer, so that {3m + 4n : m, n ∈ Z} ⊆ Z. Now let x ∈ Z. Then
x = 1 · x = (4 − 3) · x = 3(−x) + 4x,

so that x ∈ {3m + 4n : m, n ∈ Z}, whence Z ⊆ {3m + 4n : m, n ∈ Z}. Since we already

proved the other inclusion, the proof is done.
 Section 2.1: Sets 53

Example 2.1.4 We prove that A = {6m + 14n : m, n ∈ Z} equals the set B of all even
integers. Certainly for any integers m and n, 6m + 14n is an even integer, so that A ⊆ B.
Now let x ∈ B. Then x is even, so x = 2n for some integer n. Write
x = 2n = (14 − 2 · 6)n = 6(−2n) + 14n,

so that x ∈ {6m + 14n : m, n ∈ Z} = A. Thus B ⊆ A. Together with the first part this
implies that A = B.

Example 2.1.5 For each i ∈ N+ , let Ai = [i, ∞), Bi = {i, i + 1, i + 2, . . . , ∞}, and
Ci = (−i, i). Think through the following:
\ \ \
Ak = ∅, Bk = ∅, Ck = (−1, 1),
k∈N k∈N k∈N
[ [ [
Ak = [1, ∞), Bk = N+ , Ck = R.
k∈N k∈N k∈N

Example 2.1.6 For each real number r, let Ar = {r}, Br = [0, |r|]. Then
\ \ [ [
Ar = ∅, Br = {0}, Ar = R, Br = [0, ∞).
r∈R r∈R r∈R r∈R

Set operations can be represented with a Venn diagram, especially in the presence
of a universal set U . Here is an example:

A B U

On this Venn diagram, sets are represented by the geometric regions: A is the set
represented by the left circle, B is represented by the right circle, A ∩ B is the part of the
two circles that is both in A and in B, A ∪ B can be represented by the region that is
either in A or in B, A \ B is the left crescent after B is chopped out of A, etc. (There is
no reason why the regions for sets A and B are drawn as circles, but this is traditional.)
Sometimes we draw a few (or all) elements into the diagram. For example, in
54 Chapter 2: Concepts with which we will do mathematics

b
w U
A B

b
y b
t b
u
b
x b
z

b
v

we read that A = {x, y, t}, A ∩ B = {t}, et cetera.

Two disjoint sets A and B are represented by a Venn diagram as follows:

A B U

Proposition 2.1.7 For all sets A, B, C, we have A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C).

Proof. With the Venn diagram below, B ∪ C is the region filled with either horizontal
or vertical lines, A is the region filled with the Southeast-Northwest slanted lines, and so
A ∩ (B ∪ C) is the region that has simultaneously the Southeast-Northwest slanted lines
and either horizontal or vertical lines. Also, A ∩ C is the region that has horizontal and
Southeast-Northwest slanted lines, A ∩ B is the region that has vertical and Southeast-
Northwest slanted lines, so that their union (A ∩ B) ∪ (A ∩ C) represents the total region
of Southeast-Northwest slanted lines that either have horizontal or vertical cross lines as
well, which is the same as the region for A ∩ (B ∪ C).

C U

A B

We prove this also algebraically. We have to prove that A∩(B ∪C) ⊆ (A∩B)∪(A∩C)
and (A ∩ B) ∪ (A ∩ C) ⊆ A ∩ (B ∪ C).
Let x be arbitrary in A ∩ (B ∪ C). This says that x ∈ A and x ∈ B ∪ C, and the latter
says that either x ∈ B or x ∈ C. But then either x ∈ (A ∩ B) or x ∈ (A ∩ C), so that
 Section 2.1: Sets 55

x ∈ (A ∩ B) ∪ (A ∩ C). This proves one inclusion.

Now let x be arbitrary in (A ∩ B) ∪ (A ∩ C). This says that either x ∈ A ∩ B or
x ∈ A ∩ C, which in turn says that either x is in A and in B or else that x is in A and in
C. In any case, x ∈ A, and either x ∈ B or x ∈ C, so that x ∈ A and x ∈ B ∪ C, which
finally says that x ∈ A ∩ (B ∪ C). This proves the other inclusion.

Exercises for Section 2.1

2.1.1 Prove by induction on n that a set with n elements has exactly 2n distinct subsets.


2.1.2 Prove that a set with n elements has ni subsets with exactly i elements.
2.1.3 Prove the following:
√ √
i) {x ∈ R : x2 = 3} = { 3, − 3}.
ii) {x3 : x ∈ R} = R.
iii) {x2 : x ∈ R} = {x ∈ R : x ≥ 0} = [0, ∞).
iv) {2, 2, 5} = {2, 5} = {5, 2}.
v) {x ≥ 0 : x is an even prime number} = {2}.
vi) ∅ is a subset of every set. Elements of ∅ are green, smart, sticky, hairy, feathery,
prime, whole, negative, positive,...
vii) {x : x can be written as a sum of three consecutive integers} = {3n : n ∈ Z}.
viii) If A ⊆ B, then A ∩ B = A and A ∪ B = B.
2.1.4 Let U = {1, 2, 3, 4, 5, 6}, A = {1, 3, 5}, and B = {4, 5, 6}. Find the following sets:
i) (A \ B) ∪ (B \ A).
ii) U \ (B \ A).
iii) U ∪ (B \ A).
iv) U \ (A ∪ B).
v) (U ∩ A) ∪ (U ∩ B).
vi) A \ (A \ B).
vii) B \ (B \ A).
viii) {A} ∩ {B}.
2.1.5 Assume that A and B are disjoint subsets of U . For each part below, draw a Venn
diagram with A and B in U , and shade in the region described by the set: (i) U \ B,
(ii) A ∩ B, (iii) U \ (U \ A), (iv) (U \ A) ∩ (U \ B), (v) (U ∩ B) ∪ (U \ (B ∪ A)), (vi)
(B ∩ U ) ∪ (B \ U ), (vii) (U \ (U \ A)) ∪ B, (viii) (A ∪ B) ∪ (U \ A).
56 Chapter 2: Concepts with which we will do mathematics

2.1.6 Let A, B ⊆ U .
i) Prove that there exist at most 16 distinct subsets of U obtained from A, B, U by
intersections, unions, and complementation.
ii) If A and B are disjoint, prove that there exist at most 8 such distinct subsets.
iii) If A = B, prove that there exist at most 4 such distinct subsets.
iv) If A = B = U , prove that there exist at most 2 such distinct subsets.
v) If A = B = U = ∅, prove that there exists at most 1 such subset.
2.1.7 Let A, B, C ⊆ U . Prove the following statements:
i) (A ∩ C) \ B = (A \ B) ∩ (C \ B).
ii) (A \ B) ∪ (B \ A) = (A ∪ B) \ (A ∩ B).
iii) (A ∩ B) ∪ (U \ (A ∪ B)) = (U \ (A \ B)) \ (B \ A).
iv) U \ (A \ B) = (U \ A) ∪ B.
v) If U = A ∩ B, then A = B = U .
2.1.8 Let A, B ⊆ U .
i) Prove that (U \ A) ∩ (U \ B) = U \ (A ∪ B). (The intersection of the complements
is the complement of the union.)
ii) Prove that (U \ A) ∪ (U \ B) = U \ (A ∩ B). (The union of the complements is the
complement of the intersection.)
\ \ \
2.1.9 Compute (−1/k, 1/k), [−1/k, 1/k], {−1/k, 1/k}.
k∈N+ k∈N+ k∈N+
[ [ [
2.1.10 Compute (−1/k, 1/k), [−1/k, 1/k], {−1/k, 1/k}.
k∈N+ k∈N+ k∈N+

2.2 Cartesian product

The set {a, b} is the same as the set {b, a}, as any element of either set is also the
element of the other set. Thus, the order of the listing of elements does not matter. But
sometimes we want the order to matter. We can then simply make another new notation
for ordered pairs, but in general it is not a good idea to be inventing many new notations
and concepts; it is much better if we can reuse and recycle old ones. We do this next:

Definition 2.2.1 An ordered pair (a, b) is defined as the set {{a}, {a, b}}.

So here we defined (a, b) in terms of already known constructions: (a, b) is the set one
of whose elements is the set {a} with exactly one element a, and the remaining element of
(a, b) is the set {a, b} that has exactly two elements a, b if a 6= b and has exactly one element
otherwise. Thus for example the familiar ordered pair (2, 3) really stands for {{2}, {2, 3}},
(3, 2) stands for {{3}, {2, 3}}, and (2, 2) stands for {{2}, {2, 2}} = {{2}, {2}} = {{2}}.
 Section 2.2: Cartesian product 57

Proposition 2.2.2 (a, b) = (c, d) if and only if a = c and b = d.

Proof. [Recall that P ⇔ Q is the same as P ⇒ Q and P ⇐ Q. Thus the proof

will consist of two parts.]
Proof of ⇒: Suppose that (a, b) = (c, d). Then by the definition of ordered pairs,
{{a}, {a, b}} = {{c}, {c, d}}. If a = b, this says that {{a}} = {{c}, {c, d}}, so that
{{c}, {c, d}} has only one element, so that {c} = {c, d}, so that c = d. But then
{{a}, {a, b}} = {{c}, {c, d}} is saying that {{a}} = {{c}}, so that {a} = {c}, so that
a = c. Furthermore, b = a = c = d, which proves the consequent in case a = b. Now
suppose that a 6= b. Then {{a}, {a, b}} = {{c}, {c, d}} has two elements, and so c 6= d.
Note that {a} is an element of {{a}, {a, b}}, hence of {{c}, {c, d}}. Thus necessarily either
{a} = {c} or {a} = {c, d}. But {a} has only one element and {c, d} has two (since c 6= d),
it follows that {a} = {c}, so that a = c. But then {a, b} = {c, d}, and since a = c, it follows
that b = d. This proves the consequent in the remaining cases.
Proof of ⇐: If a = c and b = d, then {a} = {c} and {a, b} = {c, d}, so that
{{a}, {a, b}} = {{c}, {c, d}}.
Note that by our definition an ordered pair is a set of one or two sets.

Definition 2.2.3 For any sets A and B, the Cartesian product A × B of A and B is
the set {(a, b) : a ∈ A, b ∈ B} of all ordered pairs where the first component varies over all
elements of A and the second component varies over all elements of B.

In general, one can think of A × B as the “rectangle” with A on the horizontal side
and B on the perpendicular side.
Say, if A has 4 elements and B has 3 elements, then A × B can be represented by the
12 points in the rectangle with base consisting of elements of A and height consisting of
elements of B as follows:
b b b b b

B b b b b b

b b b b b

b b b b

A
If instead A and B are intervals as above, then A × B is the indicated rectangle.
When A and B extend infinitely far, then A × B is correspondingly a “large” rectangle:
The familiar real plane is the Cartesian product R × R.
(The three-dimensional space can be written as the Cartesian product R×(R×R) or the
Cartesian product (R × R) × R. In the former case we write elements in the form (a, (b, c)),
58 Chapter 2: Concepts with which we will do mathematics

and in the latter case we write them in the form ((a, b), c). Those extra parentheses are
there only for notation and to slow us down, they serve no better function, so by convention
we write elements simply in the form (a, b, c).)

Exercises for Section 2.2

2.2.1 How many elements are in ∅ × R?
2.2.2 Prove that A × (B ∪ C) = (A × B) ∪ (A × C).
2.2.3 Prove that A × (B ∩ C) = (A × B) ∩ (A × C).
2.2.4 Prove that (A ∩ C) × (B ∩ D) = (A × B) ∩ (C × D).
2.2.5 Give examples of sets A, B, C, D showing that (A∪C)×(B ∪D) 6= (A×B)∪(C ×D).
2.2.6 Let A have m elements and B have n elements. Prove that A × B has mn elements.
Prove that A × B has 2mn subsets.

2.3 Relations, equivalence relations

In this section we introduce relations in a formal way. Most relations that we eventually
analyze will be of familiar kind, such as ≤, <, is cousin, taller than, et cetera, but we can
get much more structure with a formal approach.

Definition 2.3.1 Any subset of A × B is called a relation on A and B. A relation

on A is a relation on A and A, i.e., a subset of A × A. We can give relation a name,
such as R, and in place of “(a, b) ∈ R” we alternatively write “aRb”. (We prefer to write
“ 3 ≤ 5” rather than “(3, 5) ∈ ≤”.)

Examples 2.3.2
(1) Some relations on the set A of all people are cousin, parent, older than, has same
birth date.
(2) Draw anything in R × R. That defines a relation on R (which most likely cannot
be expressed with a formula).
(3) Some relations on R are ≤, <, =, ≥, >. We can write (1, 2) ∈ ≤, or more familiarly,
1 ≤ 2. As a subset of R × R, ≤ consists of all points on or above the line y = x.
We can draw it (partially) as the shaded region below.
 Section 2.3: Relations, equivalence relations 59

(4) Another relation on R: R = {(a, b) : a, b ∈ R, a2 < b + 1}.

(5) If A = {1, 2} and B = {a, b}, then the following are all the possible relations on A
and B:

{(1, a), (1, b), (2, a), (2, b)},

{(1, b), (2, a), (2, b)},
{(1, a), (2, a), (2, b)},
{(1, a), (1, b), (2, b)},
{(1, a), (1, b), (2, a)},
{(1, a), (1, b)},
{(1, a), (2, a)},
{(1, a), (2, b)},
{(1, b), (2, a)},
{(1, b), (2, b)},
{(2, a), (2, b)},
{(1, a)},
{(1, b)},
{(2, a)},
{(2, b)},
{}.
60 Chapter 2: Concepts with which we will do mathematics

Remark 2.3.3 Relation names such as “≤” or “cousin” are familiar, but most relations
do not have such familiar or even good names; think of relation on A and B simply as a
possibly contrived subset of A × B.

Definition 2.3.4 Let R be a relation on A.

(1) R is reflexive if for all a ∈ A, aRa.
(2) R is symmetric if for all a, b ∈ A, aRb implies bRa.
(3) R is transitive if for all a, b, c ∈ A, if aRb, bRc, then aRc.
(4) R is an equivalence relation if it is reflexive, symmetric and transitive.

Examples 2.3.5
(1) ≤ on R is reflexive and transitive but not symmetric.
(2) < on R is transitive but not reflexive or symmetric.
(3) = on any set A is reflexive, symmetric, and transitive.
(4) Being a cousin is symmetric but not reflexive or transitive.
(5) Being taller than is ...

Definition 2.3.6 Let R be an equivalence relation on a set A. For each a ∈ A, the set
of all elements b of A such that aRb is called the equivalence class of a. We denote the
equivalence class of a with the shorthand [a].

For example, if R is the equality relation, then the equivalence class of a is {a}. If
R = A × A, then the equivalence class of any a is A. If A is the set of all students in
Math 112 this year, and aRb if students a and b are in the same section of Math 112, then
[a] is the set of all students that are in the same section as student a.

Theorem 2.3.7 Let R be an equivalence relation on a set A. Two equivalence classes are
either identical or they have no elements in common.

Proof. Let a, b ∈ A, and suppose that their equivalence classes have an element in common.
Call the element c.
We now prove that the equivalence class of a is a subset of the equivalence class of b.
Let d be any element in the equivalence class of a. Then aRd, aRc and bRc imply by
symmetry that dRa and cRb, so that by transitivity dRc. Then dRc, cRb and transitivity
give dRb, so that by symmetry bRd, which says that d is in the equivalence class of b. Thus
the equivalence class of a is a subset of the equivalence class of b.
A symmetric proof shows that the equivalence class of b is a subset of the equivalence
class of a, so that the two equivalence classes are identical.
 Section 2.3: Relations, equivalence relations 61

Remark 2.3.8 What this says is that whenever R is an equivalence relation on a set A,
then every element of A is in a unique equivalence class. Thus A is the disjoint union of
distinct equivalence classes. Conversely, if A = ∪i∈I Ai where the Ai are pairwise disjoint,
define R ⊆ A × A as (a, b) ∈ R precisely if a and b are elements of the same Ai . Then R is
an equivalence relation: reflexivity and symmetry are obvious, and for transitivity, suppose
that a and b are in the same Ai and b and c are in the same Aj . Since Ai and Aj have the
element b in common, by the pairwise disjoint assumption necessarily i = j, so that a and
c are both in Ai . Thus R is an equivalence relation.

Example 2.3.9 Let A = {1, 2, 3, 4, 5}. The writing of A as {1, 2} ∪ {3, 4} ∪ {5} makes the
following equivalence relation on A:
{(1, 1), (1, 2), (2, 1), (2, 2), (3, 3), (3, 4), (4, 3), (4, 4), (5, 5)}.

This means is that counting all the possible equivalence relations on A is the same as
counting all the possible writings of A as unions of pairwise disjoint subsets.

Important example 2.3.10 Let n be a positive integer. Let R be the relation on Z given
by aRb if a − b is a multiple of n. This relation is called congruence modulo n. It is
reflexive because for every a ∈ Z, a − a = 0 is an integer multiple of n. It is symmetric
because for all a, b ∈ Z, if aRb, then a − b = x · n for some integer x, and so b − a = (−x) · n,
and since −x is an integer, this proves that bRa. Finally, this relation R is transitive: let
a, b, c ∈ Z, and suppose that aRb and bRc. This means that a−b = x · n and b −c = y · n for
some integers x and y. Then a−c = a+(−b+b)−c = (a−b)+(b−c) = x·n+y·n = (x+y)·n,
and since x+y is an integer, this proves that aRc. Thus R is an equivalence relation. If aRb
for this relation R, we say that a is congruent to b modulo n, or that a is congruent
to b mod n. (Normally in the literature this is written as a ≡ b mod n.) We denote the
set of all equivalence classes with Z/nZ, and we read this as “Z mod n Z”. This set
consists of [0], [1], [2], . . . , [n − 1], [n] = [0], [n + 1] = [1], et cetera, so that Z/nZ has at
most n equivalence classes. Since any two numbers among 0, 1, . . . , n − 1 have difference
strictly between 0 and n, it follows that this difference is not an integer multiple of n, so
that [0], [1], [2], . . . , [n − 1] are distinct. Thus Z/nZ has exactly n equivalence classes. Two
natural lists of representatives of equivalence classes are 0, 1, 2, . . . , n − 1 and 1, 2, . . . , n.
(But there are infinitely many other representatives as well.)
For example, modulo 12, the equivalence class of 1 is the set {1, 13, 25, 37, . . .} ∪
{−11, −23, −35, . . .}, and the equivalence class of 12 is the set of all multiples of 12 (in-
cluding 0).
In everyday life we use congruence modulo 12 (or sometimes 24) for hours, modulo 12
for months, modulo 7 for days of the week, modulo 4 for seasons of the year, ...
There are exactly two equivalence classes for the congruences modulo 2: one consists of
62 Chapter 2: Concepts with which we will do mathematics

all the even integers and the other of all the odd integers. There is exactly one equivalence
class for the congruences modulo 1: all integers are congruent modulo 1 to each other. For
the congruences modulo 0, each equivalence class consists of precisely one element.

Example 2.3.11 (Construction of Z from N0 .) Consider the Cartesian product N0 × N0 .

Elements are pairs of the form (a, b), with a, b ∈ N0 . If a, b, c, d ∈ N0 , we will write
(a, b)R(c, d) if a + d = b + c. Thus R is a relation on N0 × N0 . (Certainly you are familiar
with Z, in which case you may want to think of this relation simply saying that (a, b)R(c, d)
if a − b = c − d. The problem is that in N0 we may not be able to subtract b from a and
still get an element from N0 .)
(1) R is reflexive: because for all (a, b) ∈ N0 × N0 , by commutativity of addition,
a + b = b + a, so that by definition of R, (a, b)R(a, b).
(2) R is symmetric: if (a, b)R(c, d), then by definition a + d = b + c, so that by
commutativity of addition, d + a = c + b, and by symmetry of the = relation,
c + b = d + a. But by definition this says that (c, d)R(a, b).
(3) R is transitive: if (a, b)R(c, d) and (c, d)R(e, f ), then by definition a + d = b + c
and c + f = d + e. It follows that (a + d) + (c + f ) = (b + c) + (d + e). By
associativity and commutativity of addition, (a + f ) + (c + d) = (b + e) + (c + d),
and by cancellation then a + f = b + e, which says that (a, b)R(e, f ).
Now we define a set Z to be the set of equivalence classes for this relation. Every element
(a, b) of N0 × N0 is in an equivalence class: if a ≥ b, then (a, b)R(a − b, 0), and if a < b, then
(a, b)R(0, b − a). Thus for each (a, b) ∈ N0 × N0 , there is an element in the equivalence class
of (a, b) of the form (0, e) or (e, 0) for some e ∈ N0 , and it is left for the reader to verify
that this e is unique. We can identify the set of equivalence classes with the usual integers
in Z as follows: the equivalence class of (e, 0) corresponds to the non-negative integer e
and the equivalence class of (0, e) corresponds to the non-positive integer −e.

Exercises for Section 2.3

2.3.1 Let A = {a, b} and B = {b, c, d}.
i) How many elements are in the set A×A?
ii) How many elements are in the set A×B?
iii) How many elements are in the set B×B?
iv) How many relations are there on B?
v) How many relations are there on A and B? On B and A?
vi) How many relations are there on A ∪ B and A ∩ B?
2.3.2 Let A have n elements and B have m elements. How many distinct relations on A
and B are there? Prove.
 Section 2.3: Relations, equivalence relations 63

2.3.3 In each part below, find a relation with the given properties. You may find such a
relation among number or human relations, or you may contrive a relation on a contrived
set A.
i) Reflexive, but not symmetric and not transitive.
ii) Reflexive and symmetric, but not transitive.
iii) Reflexive and transitive, but not symmetric.
iv) Symmetric, but not reflexive and not transitive.
v) Transitive, but not symmetric and not reflexive.
vi) Transitive and symmetric, but not reflexive.
2.3.4 Let A be a set with 2 elements. How many possible equivalence relations are there
on A? Repeat first for A with 3 elements, then for A with 4 elements.
2.3.5 Let A be a set with n elements. Let R be an equivalence relation on A with fewest
members. How many members are in R?
2.3.6 Let R be the relation on R given by aRb if a − b is a rational number. Prove that
R is an equivalence relation. Find at least three disjoint equivalence classes.
2.3.7 Let R be the relation on R given by aRb if a − b is an integer.
i) Prove that R is an equivalence relation.
ii) Prove that for any a ∈ R there exists b ∈ [0, 1) such that [a] = [b].
2.3.8 Let R be a relation on R × R given by (a, b)R(c, d) if a − c and b − d are integers.
i) Prove that R is an equivalence relation.
ii) Prove that for any (a, b) ∈ R × R there exists (c, d) ∈ [0, 1) × [0, 1) such that
[(a, b)] = [(c, d)].
iii) Prove that the set of equivalence classes can be identified with [0, 1) × [0, 1).
iv) For fun: check out the video game Asteroids online for a demonstration of this
equivalence relation. Do not get addicted to the game.
2.3.9 Let A be the set of all lines in the plane.
i) Prove that the relation “is parallel to” is an equivalence relation. Note that the
equivalence class of a non-vertical line can be identified by the (same) slope of the
lines in that class. Note that the vertical lines are in their own equivalence class.
ii) Prove that the relation “is perpendicular to” is not an equivalence relation.
2.3.10 For (a, b), (a′, b′ ) ∈ Z × Z \ {0}, define (a, b) ≀ (a′ , b′ ) if a · b′ = a′ · b.
i) Prove that ≀ is an equivalence relation. (Possibly mimic Example 2.3.11.)
ii) Describe the equivalence classes of (0, 1), (1, 1), (2, 3)?
iii) Find a natural identification between the equivalence classes and elements of Q.
64 Chapter 2: Concepts with which we will do mathematics

2.4 Functions

Here is the familiar definition of functions:

Definition 2.4.1 Let A and B be sets. A function from A to B is a rule that assigns to
each element of A a unique element of B. We express this with “f : A → B is a function.”
The set A is the domain of f and B is the codomain of f . The range or image of f is
Image(f ) = Range(f ) = {b ∈ B : b = f (a) for some a ∈ A}.

But in the spirit of introducing few new notions, let’s instead define functions with
the concepts we already know. Convince yourself that the two definitions are the same:

Definition 2.4.2 Let A and B be sets. A relation f on A and B is a function if for all
a ∈ A there exists b ∈ B such that (a, b) ∈ f and if for all (a, b), (a, c) ∈ f , b = c. In this
case we say that A is the domain of f , B is the codomain of f , and we write f : A → B.
The range of f is Range(f ) = {b ∈ B : there exists a ∈ A such that (a, b) ∈ f }.

Note that this second formulation is also familiar: it gives us all elements of the graph
of the function: b = f (a) if and only if (a, b) is on the graph of f . We will freely change
between notations f (a) = b and (a, b) ∈ f .
One should be aware that if f is a function, then f (x) is an element of the range and
is not by itself a function. (But often we speak loosely of f (x) being a function.)
To specify a function one needs to present its domain and its codomain, and to show
what the function does to each element of the domain.

Examples 2.4.3
(1) A function can be given with a formula. For example, let f : R → R be given by
f (x) = sin x. The range is [−1, 1].
√
(2) Here are formula definitions of two functions with domains [0, ∞): f (x) = x and
√ √
g(x) = − x. Note, however, that h(x) = ± x is NOT a function!
(3) There may be more than one formula for a function, each of which is applied to
distinct elements of the domain. For example, define f : N+ → Z by
 n−1
f (n) = 2 , if n is odd;
− n2 , if n is even.
(4) Let f : N+ → R be given by the description that f (n) equals the nth prime.
By Euclid’s theorem (proved on page 33) there are infinitely many primes so that
f is indeed defined for all positive integers. We know that f (1) = 2, f (2) = 3,
f (3) = 5, and with computer’s help I get that f (100) = 541, f (500) = 3571.
(5) For any set A, the identity function idA : A → A takes each x to itself.
 Section 2.4: Functions 65

(6) The function f : R → R given by f (x) = 1 for all x is not the identity function.
(7) Let b ∈ B. A function f : A → B given by f (a) = b for all a is called a constant
function.
(8) A function can be given by its graph:

From this particular graph we surmise that f (0) = 0, but with the precision of
the drawing and our eyesight it might be the case that f (0) = 0.000000000004.
Without any further labels on the axes we cannot estimate the numerical values
of f at other points. If numerical values are important, the graph should be filled
in with more information.
(9) A function may be presented by a table:

x f (x)
1 1
2 1
3 2
(10) A function may be presented with Venn diagrams and arrows:

×
r ×
r × ×
r ×

Remark 2.4.4 Two functions are the same if they have the same domains, the same
codomains, and if to each element of the domain they assign the same element of the
codomain.
For example, f : R → R and g : R → [0, ∞) given by f (x) = x2 and g(x) = x2 are
not the same! On the other hand, the functions f, g : R → R given by f (x) = |x| and
√
g(x) = x2 are the same.
66 Chapter 2: Concepts with which we will do mathematics

Notation 2.4.5 It is common to not specify the domain, in which case the domain is
implicitly the largest possible subset of R (after we introduce complex numbers we will
take the largest possible subset of C) on which the function is defined. For example, the
√
domain of f (x) = x1 is the set of all non-zero real numbers, and the domain of f (x) = x
is the set of all non-negative real numbers.

Definition 2.4.6 Let A, B, C, D be sets with B ⊆ C. If f : A → B and g : C → D,

then the composition g composed with f is a function g ◦ f : A → D such that for all
a ∈ A, (g ◦ f )(a) = g(f (a)).

If f (x) = (x + 1)2 and g(x) = x3 − 1, then

(g ◦ f )(x) = g(f (x)) = g((x + 1)2 ) = ((x + 1)2 )3 − 1 = (x + 1)6 − 1,

(f ◦ g)(x) = f (g(x)) = f (x3 − 1) = (x3 − 1 + 1)2 = x6 ,
g(x)f (x) = (x3 − 1)(x + 1)2 ,

and these three outcomes are distinct. For example, when plugging in x = 1, we get values
(g ◦ f )(1) = 63, (f ◦ g)(1) = 1, g(1)f (1) = 0.

Remark 2.4.7 It is common to write f 2 = f ◦ f , f 3 = f 2 ◦ f = f ◦ f ◦ f , etc. Some

exceptions to this notation are established in trigonometry: “sin2 (x)” stands for “(sin x)2 ”
and not for “sin(sin x)”.
√
Example 2.4.8 Let f (x) = x, and g(x) = x2 . The domain of f is R≥0 , and the domain
of g is R. It is possible to compose g ◦ f : R≥0 → R to obtain (g ◦ f )(x) = x, but when
composing f and g in the other order, (f ◦ g)(x) is not always equal to x. Namely, for any
negative number x, (f ◦ g)(x) = −x.
We just demonstrated that in composing two functions, the order matters.

Definition 2.4.9 A function f : A → B is injective (or one-to-one) if for all a, a′ ∈ A,

f (a) = f (a′ ) implies a = a′ . In other words, f is injective if whenever two things map via
f to one object, then those two things are actually the same.
A function f : A → B is surjective (or onto) if for all b ∈ B, there exists a ∈ A such
that f (a) = b. In other words, f is surjective if every element of the codomain is mapped
onto via f by some element of the domain.
A function f : A → B is bijective if it is injective and surjective.

For example, the identity function is bijective. The function f : R → R given by

f (x) = x2 is not injective because f (−1) = 1 = f (1). The function f : R≥0 → R given
√
by f (x) = x is injective because the square root function is strictly increasing (more
about that in Section 2.8), but it is not surjective because −1 is not the square root of
 Section 2.4: Functions 67
√
any non-negative real number. The function f : R≥0 → R≥0 given by f (x) = x is both
injective and surjective, thus bijective.
The following are all the possible functions f : {1, 2} → {1, 2} in tabular form:

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

1 1 1 1 1 2 1 2
2 1 2 2 2 1 2 2
The first and the last are neither injective nor surjective, but the middle two are bijective.
The following are all the eight possible functions f : {1, 2, 3} → {1, 2} in tabular form:

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

1 1 1 1 1 1 1 1 1 2 1 2 1 2 1 2
2 1 2 1 2 2 2 2 2 1 2 1 2 2 2 2
3 1 3 2 3 1 3 2 3 1 3 2 3 1 3 2

In this case no functions are injective, and all non-constant ones are surjective.

Theorem 2.4.10 The composition of two injective functions is injective.

Proof. Let f : A → B and g : C → D be injective functions, and suppose that B ⊆ C so

that we can compose the two.
If (g ◦ f )(a) = (g ◦ f )(a′ ), then g(f (a)) = g(f (a′ )). Since g is injective, it follows that
f (a) = f (a′ ), and since f is injective, it follows that a = a′ . Thus g ◦ f is injective.

Theorem 2.4.11 If f : A → B and g : B → C are surjective functions, so is g ◦f : A → C.

If f and g are both bijective, so is g ◦ f .

Proof. Let c ∈ C. Since g is surjective, there exists b ∈ B such that g(b) = c. Since f is
surjective, there exists a ∈ A such that f (a) = b. Thus (g ◦ f )(a) = g(f (a)) = g(b) = c, so
that g ◦ f is surjective.
The last statement follows from the first part and Theorem 2.4.10.

Definition 2.4.12 We have seen polynomial functions in Section 1.4: recall that for any
subset S ⊆ R, a function f : S → R is a polynomial function if there exist a non-negative
integer n and c0 , c1 , . . . , cn ∈ R such that for all x ∈ S, f (x) = c0 + c1 x + c2 x2 + · · · + cn xn .
If cn 6= 0, we say that the degree of f is n. A function f : S → R is a rational function
if there exist polynomial functions f1 , f2 : S → R such that for all x ∈ S, f2 (x) 6= 0 and
f (x) = f1 (x)/f2 (x).
Similarly there are polynomial and rational functions if all occurrences of “R” above
are replaced by “Q” or “C” .
68 Chapter 2: Concepts with which we will do mathematics

Polynomial and rational functions are very special, and are a workhorse of analysis.
(The reader has of course encountered trigonometric, exponential, and logarithmic func-
tions, which are not polynomial or rational.) The following are some special properties:

Theorem 2.4.13 If a polynomial function is not constant zero, it has only finitely many
roots. In fact, the number of roots is at most the degree of the polynomial.
The domain of a rational function is the complement of a finite subset of R (or C).

Proof. Suppose that f (x) = c0 + c1 x + c2 x2 + · · · + cn xn is not constant 0. Thus at least

one coefficient ci is non-zero. By possibly renaming we may assume that cn 6= 0. By the
non-constant assumption, n ≥ 1. If n = 1, then f has only one root, namely −c0 /c1 .
Suppose that n ≥ 2.* Let a be any root of f . By the Euclidean algorithm (Example 1.5.5),
there exist polynomial functions q and r such that

f (x) = q(x)(x − a) + r(x),

and the degree of r is strictly smaller than 1, i.e., r(x) is a constant. But if we plug x = a
into both sides, since a is a root of f , we get that r is the constant zero function, so that
f (x) = q(x)(x − a).

Necessarily the degree of q(x) is n − 1. By induction on the degree, q has at most n − 1

roots. If b is a root of f , then
0 = f (b) = q(b)(b − a),

so that either q(b) = 0 of b − a = 0. Thus b is either a root of q or b = a. Thus the roots of

f are a or any of the at most n − 1 roots of q, so that f has at most n roots. This proves
the first part.
A rational function is a quotient of two polynomial functions, and the rational function
is defined everywhere except where the denominator is 0. By the first part this excludes
only finitely many numbers.
This theorem is useful in that it assures us that the domain of a rational function
contains infinitely many points, or even better, that the domain is all except finitely many
real (or complex) numbers.
−c1 ±
qc2 −4c0 c2
* 2
If n = 2, then by the quadratic formula the roots of c0 + c1 x + c2 x are 1
2c2 . There are root
formulas for cases n = 3, 4, but executing them is very time-consuming, they require familiarity with complex
numbers, the solutions involve sums of cube roots with a few square roots thrown in for good measure, and
furthermore it can be hard to identify that the ensuing long expression simplifies to a nice number such as 2.
This, and the existence of computer capabilities, are the reasons that we do not teach such formulas. The formula
for solutions of cubic polynomials was discovered by Niccolò Fontana Tartaglia (1500–1557) and for quartic ones
by Lodovico Ferrari (1522–1565). Both formulas were popularized in a book by Gerolamo Cardano (1501–1576).
There are no formulas for general polynomials of degree n ≥ 5: not only do you and I do not know such a
formula, but Niels Henrik Abel (1802–1829) and Évariste Galois (1811–1832) proved that no formulas exist using
only radicals, sums, products, quotients.
 Section 2.4: Functions 69

(You are aware that the trigonometric functions sine and cosine have infinitely many
zeroes and that tangent and cotangent are not defined at infinitely many real numbers. This
fact, together with the previous theorem, establishes that these four trigonometric functions
are not polynomial or rational functions. For similar reasons the logarithmic functions are
not polynomial or rational. We have to work harder to prove that the exponential functions
are not polynomial or rational.)

Exercises for Section 2.4

2.4.1 Fix a positive integer n. Define f : Z → {0, 1, 2, 3, . . . , n−1} by f (a) is the remainder
after a is divided by n. In number theory instead of “f (a) = b” one says a mod n is b. We
can also define a similar (but not the same!) function g : Z → Z/nZ by g(a) = [a]. Graph
the function f for n = 2 and n = 3.
2.4.2 Define the floor function ⌊ ⌋ : R → R to be the function such that for all x ∈ R,
⌊x⌋ is the largest integer that is less than or equal to x. The range is Z. For example,
⌊π⌋ = 3, ⌊1⌋ = 1, ⌊−1.5⌋ = −2. Graph this function.
2.4.3 The ceiling function ⌈ ⌉ : R → R is the function for which ⌈x⌉ is the smallest
integer that is greater than or equal to x. The range is Z. For example, ⌈π⌉ = 4, ⌈1⌉ = 1,
⌈−1.5⌉ = −1. Graph this function.
2.4.4 Let A = [−a, a] or A = (−a, a), and let B be a subset of C. A function f : A → B is
an odd function (resp. even function) if for all x ∈ A, f (−x) = −f (x) (resp. f (−x) =
f (x)). Let n be a positive integer, c0 , c1 , . . . , cn complex numbers, and f the polynomial
function f (x) = c0 + c1 x + c2 x2 + · · · + cn xn .
i) Suppose that for all even k, ck = 0. Prove that f is an odd function.
ii) Suppose that for all odd k, ck = 0. Prove that f is an even function.
2.4.5 Find a set A and functions f, g : A → A such that f ◦ g 6= g ◦ f .
2.4.6 Let h : A → B, g : B → C, and f : C → D. Does f ◦ (g ◦ h) equal f ◦ g) ◦ h? Prove
or give a counterexample.
2.4.7 Let f : R × R → R and g : R → R × R be given by f (x, y) = x and g(x) = (x, 0).
i) Compute f ◦ g and g ◦ f . (This means: for each function find the domain and
codomain, and show what the function evaluates to at each point in the domain.)
ii) Which among f, g, f ◦ g, g ◦ f are injective, surjective, bijective?
70 Chapter 2: Concepts with which we will do mathematics

2.4.8 Let A = {1, 2}, B = {2, 4}, and C = {3, 6, 9}. Mark each of the following as a
function A → B, B → C, or C → A.
i) f = {(2, 6), (4, 3)}.
ii) g = {(9, 1), (6, 2), (3, 2)}.
iii) h = {(1, 4), (2, 2)}.
2.4.9 For each of the following functions, state the domains, codomains, and how they are
compositions of f, g, h from the previous exercise.
i) {(2, 2), (4, 2)}.
ii) {(9, 4), (3, 2), (6, 2)}.
iii) {(2, 6), (1, 3)}.
2.4.10 Let A be a set with 3 elements and B be a set with 2 elements.
i) How many possible functions are there from A to B?
ii) How many possible injective functions are there from A to B?
iii) How many possible surjective functions are there from A to B? From B to A?
iv) How many possible bijective functions are there from A to B?
v) How many possible functions from A to B are injective but not surjective?
vi) How many possible functions from A to B are surjective but not injective?
vii) How many functions from A to B are neither surjective nor injective?
2.4.11 The pigeonhole principle states that if n items (such as pigeons) are put into
m holes with n > m, then at least one hole has more than one item. Let A and B be sets
with only finitely many elements.
i) Use the pigeonhole principle to demonstrate that if A has more elements than B,
then f : A → B cannot be injective.
ii) Use the pigeonhole principle to demonstrate that if A has fewer elements than B,
then f : A → B cannot be surjective.
iii) Use the pigeonhole principle to demonstrate that if A and B do not have the same
number of elements, then f : A → B cannot be bijective.
2.4.12 Let A be a set with m elements and B a set with n elements.
i) How many possible functions are there from A to B?
ii) For which combinations of m, n are there injective functions from A to B?
iii) For m, n as in (ii), how many possible injective functions are there from A to B?
iv) For which combinations of m, n are there surjective functions from A to B?
v) For m, n as in (iv), how many possible surjective functions are there from A to B?
vi) For which combinations of m, n are there bijective functions from A to B?
vii) For m, n as in (vi), how many possible bijective functions are there from A to B?
 Section 2.4: Functions 71

2.4.13 Find an injective function f : N+ → N+ that is not surjective. Find a surjective

function g : N+ → N+ that is not injective. Compare with parts (ii) and (iv) of the previous
exercise. Does the pigeonhole principle apply?
2.4.14 In each part below, find f : R → R with the specified condition.
i) f is a bijective function.
ii) f is neither injective nor surjective.
iii) f is injective, but not surjective.
iv) f is surjective, but not injective.
2.4.15 Let f : A → B and g : B → C be functions.
i) Suppose that g ◦ f is injective. Prove that f is injective.
ii) Suppose that g ◦ f is surjective. Prove that g is surjective.
iii) Give an example of f, g such that g ◦ f is injective but g is not injective.
iv) Give an example of f, g such that g ◦ f is surjective but f is not surjective.
2.4.16 Find functions f, g : R → R such that f is surjective but not injective, g is injective
but not surjective, and f ◦ g is bijective.
2.4.17 Prove that if f and g are both injective functions, then f ◦ g is also injective.
2.4.18 Prove that if f and g are both surjective functions, then f ◦ g is also surjective.
2.4.19 Prove that if f and g are both bijective functions, then f ◦ g is also bijective.
2.4.20 Suppose f : A → B. We introduce the notation f (S) = {f (x) : x ∈ S} for any
subset S of A. Let C be an arbitrary subset of A.
i) Prove f (A) \ f (C) ⊆ f (A \ C).
ii) With proof, what condition will make f (A) \ f (C) = f (A \ C)?
2.4.21 For the following polynomial and rational functions, determine the domains (largest
sets on which the function is defined):
i) f (x) = x2 − x31−x .
4 3
ii) f (x) = x −2x
x6 +2
−1
.
1
(iii)* f (x) = x5 −17x2 +πx−4 . (Gottcha! We do not know the roots of this denominator.)
2.4.22 Prove that f : R → R given by f (x) = x3 − 1 is surjective and injective.
2.4.23 Let f : A → B be bijective. Prove that there exists g : B → A such that g ◦f = idA
and f ◦ g = idB . Such g is called the inverse of f .
2.4.24 A function f : A → B is called invertible if there exists g : B → A such that
g ◦ f = idA and f ◦ g = idB . Prove that every invertible function is bijective.
2.4.25 Let f : R → R be given by f (x) = x3 − 5. Prove that f is invertible and find its
inverse.
72 Chapter 2: Concepts with which we will do mathematics

2.4.26 (Lagrange interpolation) Let c1 , . . . , cn be distinct real (or complex) numbers.

(x−c1 )(x−c2 )···(x−cj−1 )(x−cj+1 )···(x−cn )
Set gj (x) = (cj −c1 )(cj −c2 )···(cj −cj−1 )(cj −cj+1 )···(cj −cn )
.

1, if j = i;
i) Prove that gj is a polynomial function and that gj (ci ) =
0, otherwise.
ii) Prove that for any real (or complex) numbers d1 , . . . , dn , there exists a polynomial g
of degree at most n − 1 such that for all j = 1, . . . , n, g(ci ) = di .
2.4.27 When defining a function, one has to pay attention that it is well-defined. Read
through the examples below showing how a definition can go wrong.
i) The codomain is not large enough: Let f : {1, 2} → {1, 2} be defined by f (x) = x2 .
ii) The function is not defined for all elements in the specified domain: Let f : R → R
be the inverse of the squaring function. Another example: let f : R → R take x
to 1/x.
iii) The function does not give the same answer: Let f : Z/12Z → Z be defined by
f ([a]) = a. Note that f ([0]) is 0, but since [0] = [12], then 0 = f ([0]) = f ([12]) =
12, but the integer 0 is not equal to the integer 12.
2.4.28 Let A be the set of all differentiable functions f : R → R. Define a relation R on
A by f Rg if f (0) = g(0).
i) Prove that R is an equivalence relation.
(ii)* Let S be the set of all equivalence classes. Define F : S → R by F ([f ]) = f (0).
Prove that F is a well-defined function, in the sense that if [f ] = [g], then F ([f ]) =
F ([g]).
(iii)* Prove that F is bijective.
2.4.29 Let f : N+ → Z and g : N+ → Q+ be defined by
 n−1
f (n) = 2 , if n is odd;
− n2 , if n is even,
and for any positive integer with prime factorization pe11 · · · pekk let g(pe11 · · · pekk ) =
f (e ) f (e )
p1 1 · · · pk k . Assuming/knowing that prime factorization of positive integers is unique
up to order, prove that g is bijective. (We just showed that N+ and Q+ have “the same
infinite number” of elements. This proof appeared in the article Counting the rationals
by Y. Sagher, in American Mathematical Monthly 96 (1989), page 823. More people are
familiar with the proof that appears on page 226.)
 Section 2.5: Binary operations 73

2.5 Binary operations

Definition 2.5.1 A binary operation on a set A is a function ◦ : A × A → A. An
arbitrary element of A × A is an ordered pair (a, b) for a, b ∈ A, and when we plug that
element into the function ◦, we should perhaps write ◦((a, b)), but normally one set of
parentheses is removed, so we write ◦(a, b). But depending on the exact form of ◦, we often
traditionally write a ◦ b rather than ◦(a, b) (see examples below).

Binary operations that we have seen are:

+, −, ·, /, ×, ◦, ∩, ∪, or , and , xor, =⇒, ⇐⇒ .

Examples and non-examples 2.5.2

(1) +, · on N, Z, Q, R (we of course write a + b rather than +(a, b) and a · b or even
ab rather than ·(a, b));
(2) Subtraction − on Z, Q, R (but not on N);
(3) Division / on Q \ {0}, R \ {0} (but not on Q, R,Q, R, N+ , Z \ {0});
(4) The additive inverse operation − is not a binary operation on R. (It is unary,
i.e., it acts on one number at a time.)
(5) The multiplicative inverse operations −1 is not a binary operation on R \ {0}. (It
is unary.)
(6) Let S be a set, and let A be the set of all functions f : S → S. The composition
(of functions) is a binary operation on A.
(7) ∩, ∪, \ are binary operations on the set of subsets of a set S.
(8) and and or are binary operations on the set of logical statements.

Definition 2.5.3 Let ◦ be a binary operation on a set A. An element e ∈ A is an identity

element for ◦ if for all a ∈ A, a ◦ e = a = e ◦ a.

(In this chapter we use “e” for the identity element; this is a symbol that is unrelated
to the base of the exponential function as in Definition 7.5.6.)

Examples 2.5.4 An identity (or is it the identity?) for + on Z, Q, N0 , R is 0. An identity

for multiplication on Z, Q, N0 , R is the number 1.
An identity for composition of functions is the identity function id (the function with
id(x) = x for all x).
Subtraction on Z does not have an identity element because if e were an identity
element, then 1 − e = 1 = e − 1, which says that e = 0 and e = 2, which is nonsense.
If S is a set and A is the collection of all subsets of S, then S is the identity element
for ∩ and ∅ is the identity element for ∪. The binary operation \ on A does not have an
identity unless the only element of A is the empty set.
The theorem below resolves the problem of “an identity” versus “the identity”.
74 Chapter 2: Concepts with which we will do mathematics

Theorem 2.5.5 Let ◦ be a binary operation on A. Suppose that e and f are both identities
for ◦. Then e = f . In other words, if an identity exists for a binary operation, it is unique.
Hence we talk about the identity for ◦.

Proof. Since for all a ∈ A, e ◦ a = a, we get in particular that e ◦ f = f . Also, for every
a ∈ A, a ◦ f = a, hence e ◦ f = e. Thus e = e ◦ f = f .
Note: we used symmetry and transitivity of the equality relation.

Definition 2.5.6 Let ◦ be a binary operation on A and suppose that e is its identity. Let
x be an element of A. An inverse of x is an element y ∈ A such that x ◦ y = e = y ◦ x.
To emphasize what the operation is, we may also say that y is a ◦-inverse of x (or see
specific terms below).

Examples and non-examples 2.5.7

(1) Let ◦ = + on Z. Then 0 is the identity element and every element has an additive
inverse.
(2) Let ◦ = + on N0 . Then 0 is the identity element and only 0 has an inverse.
(3) Let ◦ = · on Q. Then 1 is the identity element and every non-zero element has a
multiplicative inverse.
(4) Let ◦ = · on Z. Then 1 is the identity element and only ±1 have inverses.
(5) If S is a set and A is the collection of all subsets of S, then only S has an inverse
for ∩, the inverse being S itself, and only ∅ has an inverse for ∪, the inverse being ∅.
(6) Here is a new binary operation ◦ on the set S = {a, b, c, d} presented via its
multiplication table:
◦ a b c d
a a b c d
b b c d a
c c d a b
d d a b c

Note that a is the identity element, the inverse of a is a, the inverse of b is d, the
inverse of c is c, the inverse of d is b.

Definition 2.5.8 A binary operation ◦ on A is associative if for all a, b, c ∈ A, a◦(b◦c) =

(a ◦ b) ◦ c.
 Section 2.5: Binary operations 75

Examples and non-examples 2.5.9

(1) + and · are associative.
(2) −, / are not associative.
(3) Composition of functions is associative. (See Exercise 2.4.6.)
(4) ∩, ∪ are associative.

Theorem 2.5.10 Let ◦ be an associative binary operation on A with identity e. If x has

an inverse, that inverse is unique.

Proof. Let y and z be inverses of x. Then

y = y ◦ e (by property of identity)

= y ◦ (x ◦ z) (since z is an inverse of x)
= (y ◦ x) ◦ z (since ◦ is associative)
= e ◦ z (since y is an inverse of x)
= z (by property of identity).

Thus by the transitivity of equality, y = z.

Definition 2.5.11 We say that x is invertible if x has an inverse. The (abstract) inverse
is usually denoted x−1 .

Be careful! What is the number 5−1 if ◦ equals +?

If ◦ is associative, we will in the future omit parentheses in a ◦ b ◦ c (or in a ◦ b ◦ c ◦ d
et cetera), as the order of the computation does not matter.
If ◦ is not associative, we need to keep parentheses! For example, in Z, a − b − c − d can
have parentheses inserted in many different ways, and five different values can be obtained!
You may want to find specific four integers a, b, c, d for which you get 5 distinct values with
different placements of parentheses.

Notation 2.5.12 More generally, if ◦ is associative, we may and do omit parentheses

in expressions such as a1 ◦ a2 ◦ · · · ◦ an , as the meaning is unambiguous. We abbreviate
Qn
a1 ·a2 ·· · ··an (in this order!) to k=1 ak . When ◦ is addition, we abbreviate a1 ◦a2 ◦· · ·◦an
Pn
to k=1 ak . Examples were already worked out in Section 1.4.
76 Chapter 2: Concepts with which we will do mathematics

Notation 2.5.13 Just like for functions in Remark 2.4.7, also for arbitrary associative
binary operation ◦ we abbreviate a ◦ a with a2 , (a ◦ a) ◦ a with a3 , et cetera, and in general
for all positive integers n we write
an = an−1 ◦ a = a ◦ an−1 .

This notation is familiar also when ◦ equals multiplication: then 25 stands for 32. When
◦ is addition, then the abstract “25 ” stands for 10, but of course we prefer to not write 10
this way; instead we write it in the additive notation 2 + 2 + 2 + 2 + 2, or briefly, 5 · 2. The
empty product a0 makes sense if the set has the identity, and in that case a0 = e. (See
Exercise 1.4.6 for the first occurrence of an empty product.)
If a has a multiplicative inverse, then a−1 is that inverse, and in that case if ◦ is also
associative,
a−n = a−(n−1) ◦ a−1 .

To prove this by induction on n, we multiply

(a−(n−1) ◦ a−1 ) ◦ an = (a−(n−1) ◦ a−1 ) ◦ (a ◦ an−1 )

= a−(n−1) ◦ (a−1 ◦ a) ◦ an−1
= a−(n−1) ◦ an−1 = e,
and similarly an ◦ (a−(n−1) ◦ a−1 ) = e, which proves that a−n = a−(n−1) ◦ a−1 .

Theorem 2.5.14 Let ◦ be an associative binary operation on A. Let f, g have inverses.

Then g ◦ f also has an inverse, and the inverse is f −1 ◦ g −1 .

Proof. (g ◦ f ) ◦ f −1 ◦ g −1 = g ◦ (f ◦ f −1 ) ◦ g −1 = g ◦ e ◦ g −1 = e, and similarly

f −1 ◦ g −1 ◦ (g ◦ f ) = e, so that (g ◦ f )−1 = f −1 ◦ g −1 .
In particular, if ◦ equals +, then for all x, y ∈ A,
−(x + y) = (−x) + (−y),
and if ◦ equals ·, then
(x · y)−1 = y −1 · x−1 .

Theorem 2.5.15 If x is invertible, then its inverse is also invertible, and the inverse of
the inverse is x.

Proof. By definition of inverses of x, x−1 ◦ x = e = x ◦ x−1 , which also reads as “the inverse
of x−1 is x.”
 Section 2.5: Binary operations 77

Theorem 2.5.16 Cancellation. Let ◦ be an associative binary operation on a set A,

and z an invertible element in A. Then for all x, y ∈ A,

x ◦ z = y ◦ z ⇒ x = y,
z ◦ x = z ◦ y ⇒ x = y.

Proof. We prove only the first implication. If x ◦ z = y ◦ z, then (x ◦ z) ◦ z −1 = (y ◦ z) ◦ z −1 ,

hence by associativity, x ◦ (z ◦ z −1 ) = y ◦ (z ◦ z −1 ). Thus by the definition of inverses and
identities, x = x ◦ e = y ◦ e = y.
Another proof of the same fact is as follows:

x=x◦e
= x ◦ (z ◦ z −1 )
= (x ◦ z) ◦ z −1 (by associativity)
= (y ◦ z) ◦ z −1
= y ◦ (z ◦ z −1 ) (by associativity)
=y◦e
= y.

Definition 2.5.17 A binary operation ◦ on A is commutative if for all a, b ∈ A, a ◦ b =

b ◦ a.

Examples and non-examples 2.5.18

(1) +, · are commutative on all examples we have seen so far. (If you have seen
matrices, you know that matrix multiplication is not commutative.)
(2) ∩, ∪ are commutative.
(3) Function composition is not commutative (cf. Example 2.4.8).

We end this section with an important example. The reader is familiar with manip-
ulations below when n = 12 or n = 24 for hours of the day (we do not say “28 o’clock”),
when n = 7 for days of the week, when n = 3 for standard meals of the day, et cetera.

Important example 2.5.19 Let n be a positive integer. Recall the definition of Z/nZ
from Example 2.3.10: elements are equivalence classes [0], [1], [2], . . . , [n − 1]. Define + on
Z/nZ as follows: [a] + [b] = [a + b]. Well, first of all, is this even a function? Namely, we
need to verify that whenever [a] = [a′ ] and [b] = [b′ ], then [a + b] = [a′ + b′ ], which says
that any choice of representatives gives the same final answer. Well, a − a′ and b − b′ are
integer multiples of n, hence (a + b) − (a′ + b′ ) = (a − a′ ) + (b − b′) is a sum of two multiples
of n and hence also a multiple of n. Thus [a + b] = [a′ + b′ ], which says that + is indeed a
78 Chapter 2: Concepts with which we will do mathematics

binary operation on Z/nZ. It is straightforward to verify that + on Z/nZ is commutative

and associative, the identity elements is [0], and every element [a] has an additive inverse
[−a] = [n − a].
Similarly, we can define · on Z/nZ as follows: [a]·[b] = [a·b]. It is left to the reader that
this is a binary operation that is commutative and associative, and the identity elements
is [1]. The multiplication tables for n = 2, 3, 4 are below, where, for ease of notation, we
abbreviate “[a]” with “a”:
Z/2Z: · 0 1 Z/3Z: · 0 1 2 Z/4Z: · 0 1 2 3
0 0 0 0 0 0 0 0 0 0 0 0
1 0 1 1 0 1 2 1 0 1 2 3
2 0 2 1 2 0 2 0 2
3 0 3 2 1

Note that for multiplication in Z/4Z, [3] = [−1] is the multiplicative inverse of itself, and
[2] has no multiplicative inverse.

Exercises for Section 2.5

2.5.1 Let A be the set of all bijective functions f : {1, 2} → {1, 2}.
i) How many elements are in A?
ii) Prove that function composition ◦ is a binary operation on A.
iii) Write a multiplication table for ◦.
iv) What is the identity element?
v) Verify that every element of A is its own inverse.
vi) Verify that ◦ is commutative.
2.5.2 For any integer n ≥ 3, let A be the set of all bijective functions f : {1, 2, 3, . . . , n} →
{1, 2, 3, . . . , n}.
i) How many elements are in A?
ii) Prove that function composition ◦ is a binary operation on A.
iii) Find f, g ∈ A such that f ◦ g 6= g ◦ f .
2.5.3 Find a set A with a binary operation ◦ such that for some invertible f, g ∈ A,
(g ◦ f )−1 6= g −1 ◦ f −1 . (This is sometimes called the socks-and-shoes problem. Can you
think of why?)
2.5.4 Refer to Example 2.5.19: Write addition tables for Z/2Z, Z/3Z, Z/4Z.
2.5.5 Write multiplication tables for Z/5Z, Z/6Z, Z/7Z.
2.5.6 Determine all [a] ∈ Z/12Z that have a multiplicative inverse.
*2.5.7 Determine all [a] ∈ Z/pZ that have a multiplicative inverse if p is a prime number.
 Section 2.6: Fields 79

2.5.8 Here is an opportunity to practice induction. Let f, g : A → A be functions, with f

invertible. Prove that (f ◦ g ◦ f −1 )n = f ◦ g n ◦ f −1 . (Notation is from Remark 2.4.7.)
2.5.9 Consider the following binary operation ◦ on {a, b, c}: ◦ a b c
a a b c
b b a a
c c a a
i) Show that ◦ is commutative and has an identity element.
ii) Show that every element has an inverse, and that inverses need not be unique.
iii) Prove that ◦ is not associative. (Hint: Theorem 2.5.10.)
*2.5.10 Let S be the set of all logical statements. We define a relation ∼ on S as follows:
P ∼ Q if P and Q have the same truth values in all conditions. For example, “1 = 1” and
“2 = 2” are related via ∼.
i) Prove that ∼ is an equivalence relation on S. Let A be the set of all equivalance
classes.
ii) Verify that and , or and xor are naturally binary operations on A.
iii) Find the identity elements, if they exist, for each of the three binary operations.
iv) For each binary operation above with identity, which elements of A have inverses?
2.5.11 Define a binary operation ⊕ on R as a ⊕ b = a + b + 2, where + is the ordinary
addition on R. Prove that ⊕ is commutative and associative. Find the identity element of
⊕, and for each x ∈ R, find its inverse.
2.5.12 Define a binary operation ⊙ on R as a⊙b = a·+2·a+2·b+2, where + and · are the
ordinary addition and multiplication on R. Prove that ⊙ is commutative and associative.
Find the identity element of ⊙, and for each x ∈ R \ {−2}, find its inverse.

2.6 Fields

The motivation for the abstract definition of fields below comes from the familiar
properties of the set of all real numbers.

Definition 2.6.1 A set F is a field if it has two binary operations on it, typically denoted
+ and ·, and special elements 0, 1 ∈ F , such that the following identities hold for all
m, n, p ∈ F :
(1) (Additive identity) m + 0 = m = 0 + m.
(2) (Associativity of addition) m + (n + p) = (m + n) + p.
(3) (Commutativity of addition) m + n = n + m.
(4) (Multiplicative identity) m · 1 = m = 1 · m.
(5) (Distributivity) m · (n + p) = (m · n) + (m · p).
80 Chapter 2: Concepts with which we will do mathematics

(6) (Associativity of multiplication) m · (n · p) = (m · n) · p.

(7) (Commutativity of multiplication) m · n = n · m.
(8) (Existence of additive inverses) There exists r ∈ F such that m+r = r+m = 0.
(9) (Existence of multiplicative inverses) If m 6= 0, there exists r ∈ F such that
m · r = r · m = 1.
(10) 1 6= 0.
0 is called the additive identity and 1 is called the multiplicative identity.

It is standard to omit “·” when no confusion arises. Note that 2 · 222 + 4 is different
from 2 222 + 4, but 2 · x + 4 is the same as 2x + 4.
The familiar N+ , N0 , Z, Q, R all have the familiar binary operations + and · on them.
Among these, N+ lacks the additive identity, but all others have the additive identity 0. In
N0 , all non-zero elements lack additive inverses, and in Z, all non-zero elements other than
1 and −1 lack a multiplicative inverse. Thus N+ , N0 and Z are not fields.
We take it for granted that Q and R are fields. In Section 3.1 we construct a new
field, the field of complex numbers. There are many other fields out there, such as the set
of all real-valued rational functions with real coefficients. A few fields are developed in the
exercises to this section.

Notation 2.6.2 By Section 2.5, we know that the additive and multiplicative identities
and inverses are unique in a field. The additive inverse of m is denoted −m, and the
multiplicative inverse of a non-zero m is denoted m−1 , or also 1/m. The sum n + (−m) of
n and −m is also written as n − m, and the product n · m−1 of n and m−1 is also written as
n/m. The latter two operations are also called subtraction and division. The functions
− and −1 are unary, with domains F and F \ {0}, respectively. By Theorem 2.5.15,
−(−m) = m, and for any non-zero m, 11 = (m−1 )−1 = m.
m
Another bit of notation: · takes precedence over addition, so that “(a · b) + c” can
be written simply as “a · b + c”. Note the omitting of the multiplication symbol and
the parentheses on the right but not on the left in the theorem below; if we skipped the
parentheses on the left the meaning would be different and the equality false in the given
generality.

Theorem 2.6.3 (The other distributive property) If F is a field, then for all m, n, p ∈ F ,
(m + n)p = mp + np.

Proof. (m + n)p = p(m + n) (by commutativity of multiplication)

= pm + pn (by distributivity (5))
= mp + np (by commutativity of multiplication).
 Section 2.6: Fields 81

Theorem 2.6.4 If F is a field, then for all m ∈ F , m · 0 = 0 = 0 · m.

Proof. m · 0 = m · 0 + 0 (since 0 is the additive identity).

= m · 0 + (m · 0 + (−(m · 0))) (by the definition of additive inverses)
= (m · 0 + m · 0) + (−(m · 0)) (by associativity of addition)
= m · (0 + 0) + (−(m · 0)) (by the distributive property)
= m · 0 + (−(m · 0)) (since 0 is the additive identity)
= 0 (by the definition of additive inverses).

Since multiplication is commutative, it also follows that 0 · m = 0.

We cannot divide by 0. Never divide by 0. For one thing, in an abstract field,
dividing by 0 is simply not on the list of allowed operations, and for another, n/m always
stands for that unique element of the field, which, when multiplied by m, yields n. In other
words, n = (n/m) · m. If m somehow – horribly – happened to be 0, then we would have
n = (n/0)·0, and by Theorem 2.6.4, this product n would be 0. So, if we were to divide by 0,
the only number that could possibly be divided by it is 0 itself. But – continuing the horrible
detour – what should 0/0 be? How about the muddle in the following usage of other axioms
that seems to require also division of 1 by 0: 0/0 = 1·(0/0) = (1·0)/0 = (0·1)/0 = 0·(1/0).
In any case, “division” by 0 is inconsistent, and not allowed. In a mathematics paper, never
write “ x0 ” or “x/0”.
At this stage of your mathematical life, you will of course never write something like
“3/0” (my eyes hurt seeing this!), but a common college mistake that is essentially division
by 0 is canceling x when solving an equation such as x2 = 3x to obtain only one solution
x = 3. This cancellation was division by 0 when x was the other possible solution! Avoid
even hidden division by 0, so that you can find all the solutions.

Exercises for Section 2.6

2.6.1 Verify that the set {0} satisfies axioms (1)–(9) of fields, with 0 being the additive
and the multiplicative identity. Obviously {0} fails axiom (10).
2.6.2 Use the set-up in Example 2.3.10. Prove that Z/2Z is a field. Note that in this field
[2] = [0], so [2] does not have a multiplicative inverse. Also note that in this field, every
number has a square and cube root.
2.6.3 Use the set-up in Example 2.3.10. Prove that Z/3Z is a field. Note that in this field
[3] = [0], so [3] does not have a multiplicative inverse. Note that in this field, [2] is not the
square of any number.
82 Chapter 2: Concepts with which we will do mathematics

2.6.4 Use the set-up in Example 2.3.10, and let n be a positive integer strictly bigger than
1 that is not a prime integer. Prove that Z/nZ is not a field.
*2.6.5 Use the set-up in Example 2.3.10. Prove that Z/pZ is a field for any prime integer p.
Note that in Z/7Z, [2] is the square of [3] and of [4].
2.6.6 Let F be a field. Prove that (−1) · x is the additive inverse of x.
2.6.7 Let F be a field. Prove that for any x ∈ F , −(−x) = x. Prove that for any non-zero
x ∈ F , 1/(1/x) = x.
2.6.8 Let F be a field. Prove that for any x, y ∈ F , (−x) · y = −(x · y) = x · (−y). (Hint:
Use the definition of additive inverses.)
2.6.9 Let F be a field. Prove that for any x, y ∈ F , (−x) · (−y) = x · y.
2.6.10 Let x be a non-zero element of a field F . Then (−x)−1 = −x−1 .
2.6.11 (An unusual field.) Let ⊕ and ⊙ be binary operations on R as defined in Exer-
cises 2.5.11 and 2.5.12. Prove that R is a field with these two binary operations.

2.7 Order on sets, ordered fields

If < is a relation on a set S, we can define relations ≤, >, ≥ on S by

a ≤ b means that a < b or a = b.

a > b means that b < a.
a ≥ b means that a 6< b.
Conversely, if ≤ is a relation on S, then we can define < on S by

a < b if and only if a ≤ b and a 6= b,

which by before also defines >, ≥. Similarly, each of >, ≥ determines all four relations of
this form. Thus one of these relations on a set S implies that we have all four relations
naturally derived from the one. These relations impose the usual order on a set.
We use these relations in the usual sense on R, but we can also use < in other contexts:

Examples 2.7.1
(1) < can be the relation “is a proper subset of” on a set S of all subsets of some uni-
versal set U . In this case, ≤ means “is a subset of”, > means “properly contains”,
and ≥ means “contains”.
(2) < cannot be the relation “has strictly fewer elements” on the set S of all subsets
of the set {1, 2, 3, . . . , 100}. In this case, ≤ would mean “has fewer elements or is
the same”, rather than “has fewer or the same number of elements”.
 Section 2.7: Order on sets, ordered fields 83

Definition 2.7.2 Let ≤ be a relation on a set S.

An element b ∈ S is called an upper bound (resp. lower bound) of T (in S) if for
all t ∈ T , t ≤ b (resp. t ≥ b).*
A subset T of S is bounded above (resp. bounded below) (in S) if there exists an
upper bound (resp. lower bound) of T in S.
A set that is bounded above and below is called bounded.
An element c ∈ S is called a least upper bound, or supremum, of T , if it is an
upper bound of T , and if for all upper bounds b of T , c ≤ b. If c ∈ T , then c is also called
the maximum of T .
An element c ∈ S is called a greatest lower bound, or infimum, of T , if it is a
lower bound of T , and if for all lower bounds b of T , b ≤ c. If c ∈ T , then c is also called
the minimum of T .
The obvious standard abbreviations are: lub(T ) = sup(T ), glb(T ) = inf(T ), max(T ),
min(T ), possibly without parentheses around T .

Examples 2.7.3
(1) The set N0 has minimum 0. It is not bounded above, for any upper bound u
would be strictly smaller than the positive integer ⌈u⌉ + 1 (the ceiling function),
thus contradicting the assumption of upper bounds.
(2) The set T = {1/n : n ∈ N+ } has maximum 1, it is bounded below, the infimum is
0, and there is no minimum.
Proof: In long form, the set equals {1, 1/2, 1/3, 1/4, 1/5, . . .}. From this re-writing
it is clear that 1 is the maximum, that 0 is a lower bound and that 0 is not in the
set, so 0 cannot be the minimum. Why is 0 the largest lower bound, i.e., why is
0 the infimum of T ? Suppose that r is a positive real number. Set n = ⌈ r1 ⌉ + 1.
Then n is a positive integer, and n > r1 . By cross multiplying we get that r > n1 ,
which proves that r is not a lower bound on T . Since r was arbitrary, this proves
that no positive number is a lower bound on T , so that 0 is the greatest lower
bound on T .
(3) The set {1/p : p is a positive prime number} has maximum 1/2 and infimum 0.
(There are infinitely many prime numbers; see the proof on page 33.)
(4) The sets {(−1)n : n ∈ N+ } and {sin(x) : x ∈ R} both have maximum 1 and
minimum −1.
(5) The set of all positive rational numbers that are strictly smaller than π has mini-
mum 0 and supremum π, but it has no maximum.

* A sentence of the form “P is Q (resp. Q′ ) if R (resp. R′ )” is shorthand for two sentences: “P is Q if R”

and “P is Q′ if R′ ”.
84 Chapter 2: Concepts with which we will do mathematics

(6) The set {ex : x ∈ R} has no upper bound, it is bounded below with infimum 0,
but there is no minimum.
(7) The empty subset obviously has no minimum nor maximum. Every element of S
is vacuously an upper and a lower bound of the empty set.
(8) The set {x ∈ R : −3 < x − 5 < 3} has no minimum and maximum, but the
infimum is 2 and the supremum is 8. The set {x ∈ R : −3 ≤ x − 5 < 3} has
minimum 2, supremum 8, and no maximum. The set {x ∈ R : −3 < x − 5 ≤ 3}
has maximum 8, infimum 2, and no minimum. The set {x ∈ R : −3 ≤ x − 5 ≤ 3}
has minimum 2 and maximum 8.
(9) If T = {{}, {1}, {2}}, then the inclusion relation on T has minimum {}, and no
upper bounds in T . If we think of T as a subset of the set S of all subsets of {1, 2}
(or of the set S of all subsets of R), then T has supremum {1, 2}.
(10) If S is the set of all subsets of the set {1, 2, 3, . . . , 100} and < is the re-
lation “has strictly fewer elements”, then the empty set is the minimum
and {1, 2, 3, . . . , 100} is the maximum. If T is the subset of S consisting
only of sets with at most 2 elements, then the minimum of T is the empty


set, and there is no maximum or supremum in T . The 1002 elements
{1, 2}, {1, 3}, . . . , {1, 100}, {2, 3}, {2, 4}, . . . , {99, 100} are each greater than or
equal to all elements of T and they are not strictly smaller than any other element
of T .

In the sequel we restrict < to relations that satisfy the trichotomy property:

Definition 2.7.4 A relation < on a set S satisfies the trichotomy if for all s, t ∈ S,
exactly one of the following relations holds:

s = t, s < t, t < s.

Examples 2.7.5
(1) The familiar < on R satisfies the trichotomy.
(2) If S is the set of all subsets of a universal set U , then the inclusion relation on S
satisfies the trichotomy.
(3) If S = {{}, {1}, {2}}, then the inclusion relation on S does not satisfy the tri-
chotomy.
 Section 2.7: Order on sets, ordered fields 85

Theorem 2.7.6 Let < on S satisfy the trichotomy. Then a supremum (resp. infimum) of
a non-empty subset T of S, if it exists, is unique.

Proof. Suppose that c, c′ are suprema of T in S. Both c and c′ are upper bounds on T , and
since c is a least upper bound, necessarily c′ ≤ c. Similarly c ≤ c′ , so that by trichotomy
c = c′ . This proves that suprema are unique, and a similar proof shows that infima are
unique.
Why did we assume that the subset T of S above be non-empty? By definition every
element of S is an upper bound for ∅, so in particular if S has no minimum, then ∅ has no
least upper bound.

Definition 2.7.7 Let F be a set with a binary operation +, with (additive) identity
0 ∈ F , and with a relation < satisfying the trichotomy. Define F + = {x ∈ F : 0 < x}, and
F − = {x ∈ F : x < 0}. Elements of F + are called positive, and element of F − are called
negative. Elements of F \ F + are called non-positive and element of F \ F − are called
non-negative.
We define intervals in F to be sets of the following form, where a, b ∈ F with a < b:

(a, b) = {x ∈ F : a < x < b},

(a, b] = {x ∈ F : a < x ≤ b},
[a, b) = {x ∈ F : a ≤ x < b},
[a, b] = {x ∈ F : a ≤ x ≤ b},
(a, ∞) = {x ∈ F : a < x},
[a, ∞) = {x ∈ F : a ≤ x},
(−∞, b) = {x ∈ F : x < b},
(−∞, b] = {x ∈ F : x ≤ b}.

Definition 2.7.8 We say that a field F is an ordered field if it has a relation < with the
following properties:
(1) < satisfies the trichotomy, i.e., for all x, y ∈ F , exactly one of the following is true:
x < y, x = y, y < x.

(2) (Transitivity of <) For all x, y, z ∈ F , if x < y and y < z then x < z.
(3) (Compatibility of < with addition) For all x, y, z ∈ F , if x < y then x + z <
y + z.
(4) (Compatibility of < with multiplication by positive elements) For all
x, y, z ∈ F , if x < y and 0 < z then xz < yz.
A subset of an ordered field is called an ordered set.
86 Chapter 2: Concepts with which we will do mathematics

Axiom 2.7.9 We take it as fact that R is an ordered field.

It follows that Q is an ordered field as well: it is a field, and the trichotomy, transitivity,
and compatibilities of < hold on Q as they hold on the larger set R. These properties of
< also hold on the subsets N+ , N0 and Z, even if the latter sets are not fields.

Theorem 2.7.10 Let F be an ordered set.

(1) For x ∈ F with the additive inverse −x ∈ F , x ∈ F + if and only if −x ∈ F − , and
x ∈ F − if and only if −x ∈ F + .
(2) 1 ∈ F + .
(3) For x ∈ F with the multiplicative inverse x−1 ∈ F , x ∈ F + if and only if x−1 ∈ F + ,
and x ∈ F − if and only if x−1 ∈ F − .

Proof. (1) x ∈ F + if and only if 0 < x, and by compatibility of < with addition this implies
that −x = 0 − x < x − x = 0, so that −x ∈ F − . The rest of (1) is equally easy.
(2) By assumption 1 6= 0. If 1 6∈ F + , then by trichotomy 1 < 0, and by (1), 0 < −1.
Thus by compatibility of < with multiplication by positive numbers, since −1 is supposedly
positive, 0 = 0·(−1) < (−1)·(−1). By Exercise 2.6.9, (−1)·(−1) = 1, which by transitivity
says that 0 < 1. Since we also assumed that 1 < 0, we get a contradiction to the trichotomy.
So necessarily 1 ∈ F + .
(3) Suppose that x ∈ F + . By trichotomy then exactly one of the following three
inequalities holds: x−1 < 0, x = 0, x−1 > 0. Let stand for the correct inequality (or
equality). By compatibility of < with multiplication by the positive number x, we then
have 1 = x · x−1 x · 0 = 0. By (2), the relation must equal >, so that x−1 > 0.
If instead x ∈ F − , then by (1), −x ∈ F + , and by what we have proved of (3),
(−x)−1 ∈ F + . By Exercise 2.6.10 then −x−1 = (−x)−1 ∈ F + , so that x−1 ∈ F − by (1).

Theorem 2.7.11 Let F be an ordered field.

(1) For x, y ∈ F + , x + y is also in F + .
(2) For x, y ∈ F + , x · y is also in F + .
(3) For x, y ∈ F − , x + y ∈ F − .
(4) For x, y ∈ F − , x · y ∈ F + .
(5) For x ∈ F + and y ∈ F − , x · y ∈ F − .

Proof. (1) By assumption, 0 < x and 0 < y. Then by compatibility of < with addition,
y = 0 + y < x + y, and since 0 < y, by transitivity of <, 0 < x + y, which proves that
x + y ∈ F +.
(2) By assumption, 0 < x and 0 < y. Then by compatibility of < with multiplication
by positive numbers, 0 = 0 · y < x · y, which proves that x · y ∈ F + .
The proofs of the rest are similar.
 Section 2.7: Order on sets, ordered fields 87

Theorem 2.7.12 (Archimedean property) Let x, y ∈ R. If y > 0, there exists p ∈ N+

such that x < py.

Proof. If x ≤ 0, set p = 1: this p is a positive integer, and x ≤ 0 < y = 1 · y.

Now suppose that x > 0. Set p = ⌈ xy ⌉ + 1. Then p is a positive integer, p > xy , and by
compatibility of > with multiplication by positive numbers, py > x.

Theorem 2.7.13 Between any two distinct real numbers there is a rational number
(strictly between them).

Proof. Let x, y ∈ R with x < y. Then y − x > 0, and by the Archimedean property, there
exists a positive integer p such that 2 < p(y − x). Let r = p1 (⌈px⌉ + 1) (recall that ⌈px⌉
is the ceiling function of px). So r is a rational number. By the definition of the ceiling
function, px ≤ ⌈px⌉ < ⌈px⌉+1. Since p is positive, so is p−1 , and by compatibility of < with
multiplication by positive numbers, x < p1 (⌈px⌉+1) = r. Furthermore, px+1 < 2+px < py
by the choice of p, so that ⌈px⌉ + 1 < py and finally r = p1 (⌈px⌉ + 1) < y.

Remark 2.7.14 It is also true that between any two real numbers there is an irrational
√
number. Namely, let x < y be real numbers. It is proved on page 21 that 2 is a
positive irrational number. By compatibility of < with multiplication by positive numbers
√ √
then x 2 < y 2. By the Archimedean property, there is a rational number r such that
√ √
x 2 < r < y 2. If r = 0, again by the Archimedean property there exists a rational
√ √
number s such that x 2 < 0 < s < y 2. So by possibly replacing r by this s we
may assume that r is a non-zero rational number. Then again by compatibility and by
√ √
Theorem 2.7.10, x < r/ 2 < y. But r is non-zero, so that r/ 2 is an irrational number
strictly between the given real numbers x and y.

Exercises for Section 2.7

2.7.1 Let F be an ordered field and let x ∈ F satisfy x > 1. Prove that for all positive
integers n, xn > 1 and xn+1 > x.
2.7.2 Let F be an ordered field and let x ∈ F satisfy 0 < x < 1. Prove that for all positive
integers n, 0 < xn < 1 and xn+1 < x.
2.7.3 (Bernoulli’s inequality) Prove that for all x ∈ R≥0 and all n ∈ N0 , (1 + x)n ≥
1 + nx.
2.7.4 Prove that if < is transitive then ≤ is transitive.
2.7.5 Prove that < (resp. ≤) is transitive if and only if > (resp. ≥) is transitive.
88 Chapter 2: Concepts with which we will do mathematics

2.7.6 (An exercise of this flavor is relevant in computing limits as in Section 4.1.) Under
what conditions is the minimum of {0.2, ǫ/7} equal to 0.2, and when is the minimum ǫ/7?
Similarly determine min{0.2, ǫ/7, ǫ2/4}.
2.7.7 Does the set {2n /n : n ∈ N+ } have a lower (resp. upper bound)? Justify. Repeat
with {n/2n : n ∈ N+ }.
2.7.8 For each of the subsets of R below, determine its minimum, maximum, infimum,
supremum, if applicable. Justify all answers.
i) {−1, 2, π, −7}.
ii) {(−1)n /n : n ∈ N+ }.
iii) The set of all positive prime numbers.
iv) {x ∈ R : −1 < x < 5}.
v) {x ∈ R : 2 ≤ x < 5}.
vi) {x ∈ Q : x2 < 2}.
vii) {x ∈ R : x2 + x − 1 = 0}.
viii) {x ∈ Q : x2 + x − 1 = 0}.
ix) {n/(n + 1) : n ∈ N0 }.
2.7.9 Suppose that a subset T of an ordered field has a minimum (resp., maximum, in-
fimum, supremum) b. Prove that the set −T = {−t : t ∈ T } has a maximum (resp.,
minimum, supremum, infimum) −b.
2.7.10 Suppose that a subset T of positive elements of an ordered field has a minimum
(resp., maximum, infimum, supremum) b. What can you say about the maximum (resp.,
minimum, supremum, infimum) of the set {1/t : t ∈ T }?
† 2.7.11 (Invoked in Theorem 7.3.4.) Let S and T be subsets of an ordered field F . Let
S + T = {s + t : s ∈ S, t ∈ T }.
i) If S and T are bounded above, prove that sup(S + T ) ≤ sup S + sup T .
ii) If S and T are bounded below, prove that inf(S + T ) ≥ inf S + inf T .
2.7.12 Let F be an ordered field. Prove that 2, 3 are positive (and so not zero).
2.7.13 Let F be a field and x ∈ F . Prove that x2 = 0 if and only if x = 0.
2.7.14 Let F be an ordered field and x ∈ F . Let x, y ∈ F be non-negative (resp. non-
positive) such that x + y = 0. Prove that x = y = 0.
2.7.15 Let F be an ordered field. Suppose that x ≤ y and p ≤ q. Prove that x + p ≤ y + q.
If in addition x < y or p < q, prove that x + p < y + q.
2.7.16 Let F be an ordered field and x, y ∈ F . Prove that x < y if and only if 0 < y − x.
Prove that x ≤ y if and only if 0 ≤ y − x.
2.7.17 Let F be an ordered field, and x, y ∈ F + with x < y. Prove that 1/y < 1/x.
 Section 2.8: Increasing and decreasing functions 89

2.7.18 Let F be an ordered field. Suppose that x < y and that x, y are non-zero. Does it
follow that 1/y < 1/x? Prove or give a counterexample.
†2.7.19 (In-betweenness in an ordered field) Let F be an ordered field. Let x, y ∈ F with
x < y. Prove that x < (x + y)/2 < y. (Why are we allowed to divide by 2?)
2.7.20 Find an ordered set without a minimum.
2.7.21 Let F be an ordered set. Prove that any non-empty finite subset S of F has a
maximum and a minimum. Prove that for all s ∈ S, min(S) ≤ s ≤ max(S).
2.7.22 Let m ∈ R be a positive number. Prove that there exists a positive integer N such
that 1/2N < m.
2.7.23 Let n > 1 be an integer and F = Z/nZ. (This was defined in Example 2.3.10.)
Prove that F is not an ordered set. In particular, using Exercise 2.6.5, for any prime
integer p, Z/pZ is a field that is not ordered.

2.8 Increasing and decreasing functions

Definition 2.8.1 Let F, G be ordered sets (as in the previous section), and let A ⊆ F . A
function f : F → G is increasing (resp. decreasing) on A if for all x, y ∈ A, x < y implies
that f (x) ≤ f (y) (resp. f (x) ≥ f (y)). If furthermore f (x) < f (y) (resp. f (x) > f (y) for
all x < y, then we say that f is strictly increasing (resp. strictly decreasing) on A. A
function is (strictly) monotone if it is (strictly) increasing or (strictly) decreasing.

Theorem 2.8.2 Let n be a positive integer and F an ordered field. Then the function
f : F → F defined by f (x) = xn is strictly increasing on F + ∪ {0}.

Proof. Let x, y ∈ F + with x < y. Then by Exercise 1.6.7,

f (y) − f (x) = y n − xn = (x + (y − x))n − xn
Xn   X n
n−1
n k n−k n
= x (y − x) −x = xk (y − x)n−k .
k k
k=0 k=0


But x, y − x are positive, and nk is a non-negative integer, so that f (y) − f (x) is the
sum of products of non-negative elements, so that f (y) − f (x) ≥ 0. Furthermore, one


of the non-negative summands is n0 x0 (y − x)n−0 = (y − x)n , which is positive, so that
f (y) − f (x) > 0.
90 Chapter 2: Concepts with which we will do mathematics

Corollary 2.8.3 Let n be a positive integer and F an ordered field. Suppose that x, y ∈
F + ∪ {0} have the property that xn < y n . Then x < y. (In other words, the nth radical
function is strictly increasing on F + ∪ {0}.)

Proof. If x = y, then xn = y n , which contradicts the assumption and trichotomy. If x > y,

then by Theorem 2.8.2, xn > y n , which also contradicts the assumption. So by trichotomy
x < y.

Theorem 2.8.4 If F, G are ordered sets and f : F → G is strictly monotone, then f is

injective, and there exists a strictly monotone function g : Range(f ) → F such that for all
x ∈ F , (g ◦ f )(x) = x and for all y ∈ G, (f ◦ g)(y) = y. In other words, g is the inverse of
the function f : F → Range(f ).

Proof. Let y ∈ Range(f ). Then y = f (x) for some x ∈ F . If also y = f (z) for some z ∈ F ,
since f is strictly monotone, x = z. So f is injective and x is unique. Thus we define
g : Range(f ) → F by g(y) = x. Then by definition for all x ∈ F , g(f (x)) = x and for all
y ∈ Range(f ), f (g(y)) = y. If f is increasing and y1 , y2 ∈ Range(f ) such that y1 < y2 , then
g(y1 ) < g(y2 ) for otherwise by the increasing property of f , y1 = f (g(y1)) ≥ f (g(y2 )) = y2 ,
which is a contradiction. Thus if f is increasing, so is g = f −1 . Thus if g = f −1 is
increasing, so is f = (f −1 )−1 . The same reasoning goes for the decreasing property.

Remark 2.8.5 Let F be an ordered set and G an ordered field. Let f, g : F → G be

functions. We define functions f + g, f · g : F → G by (f + g)(x) = f (x) + g(x) and
(f · g)(x) = f (x) · g(x). (Instead of f · g we simply write f g.)
(1) If f, g are both strictly increasing (resp. decreasing), then f +g is strictly increasing
(resp. decreasing).
(2) If f, g are both strictly increasing (resp. decreasing) and always take on positive
values, then f g is strictly increasing (resp. decreasing).
(3) If f, g are both strictly increasing (resp. decreasing) and always take on negative
values, then f g is strictly decreasing (resp. increasing).

Proof of (3): Let x, y ∈ F with x < y. Suppose that f and g are both increas-
ing functions, so that f (x) < f (y) < 0 and g(x) < g(y) < 0. Then −f (y), −g(x)
are positive numbers, so by compatibility of < with multiplication by positive numbers,
f (x)(−g(x)) < f (y)(−g(x)) and (−f (y))g(x) < (−f (y))g(y). By Exercise 2.6.8, this
says that −(f (x)g(x)) < −(f (y)g(x)) and −(f (y)g(x)) < −(f (y))g(y)). By transitiv-
ity of < then −(f (x)g(x)) < −(f (y))g(y)). By compatibility of < with addition, by
adding f (x)g(x) + f (y)g(y) we get that f (y)g(y) < f (x)g(x). With function notation,
(f g)(y) < (f g)(x), and since x and y were arbitrary, this says that f g is strictly decreas-
ing. The proof in the case where both f and g are strictly decreasing is similar.
 Section 2.9: The Least upper bound property of R 91

Exercises for Section 2.8

2.8.1 Let n be an odd positive integer and F an ordered field. Prove that the function
f : F → F defined by f (x) = xn is strictly increasing.
2.8.2 Let n be an even positive integer and F an ordered field. Prove that the function
f : F − ∪ {0} → F defined by f (x) = xn is strictly decreasing.
2.8.3 Let n be an odd positive integer and F an ordered field. Suppose that a, b ∈ F and
that an < bn . Prove that a < b.
2.8.4 Let F be an ordered field, a ∈ F + and f : F → F defined by f (x) = ax. Prove that
f is a strictly increasing function.
2.8.5 Let F be an ordered field, a ∈ F − and f : F → F defined by f (x) = ax. Prove that
f is a strictly decreasing function.
2.8.6 Prove that the composition of (strictly) increasing functions is (strictly) increasing.
Prove that the composition of (strictly) decreasing functions is (strictly) increasing.
2.8.7 Prove that the composition of a (strictly) increasing function with a (strictly) de-
creasing function, in any order, is (strictly) decreasing.
2.8.8 Suppose that f : F → G is strictly monotone.
i) Let B be a subset of F , and define g : B → G by g(x) = f (x). Prove that g is
strictly monotone.
ii) Define h : B → Range(g) by h(x) = f (x). Prove that h is bijective.
2.8.9 Give an example of a non-decreasing function f : R → R that is not injective.

2.9 The Least upper bound property of R

In this section we formalize and analyze the important Least upper bound property
of R that we accept without proof.

Axiom 2.9.1 (Least upper/greatest lower bound property) For any non-empty
subset T of R that is bounded above, sup(T ) exists in R.
Similarly, for any non-empty subset T of R that is bounded below, inf(T ) exists in R.

Moreover, by Theorem 2.7.6, sup(T ) and inf(T ) are unique.

√
For example, the non-empty bounded set {x ∈ Q : x2 < 2} in R has infimum − 2
√
and supremum 2 in R. In contrast, the same set has no infimum nor maximum in Q. So,
R satisfies the Least upper bound property but Q does not.
92 Chapter 2: Concepts with which we will do mathematics

We will use this Least upper bound property in the proof of the Intermediate value
theorem Theorem 5.3.1, in the proof of the Extreme value theorem Theorem 5.2.2, and in
many limit tricks for functions and sequences. Here is the first example of its usage.

Theorem 2.9.2 Every positive real number is the square of a positive real number.

Proof. Let y be a positive real number. Let T = {r ∈ R : 0 ≤ r and r 2 ≤ y}. Then

T contains 0, so it is non-empty. If for some r ∈ T , r > ⌈y⌉, then r > y ≥ r 2 , so that
by compatibility of multiplication/division by positive numbers, 1 ≥ r. But y is positive,
so ⌈y⌉ is bigger than or equal to 1, so that 1 ≥ r > ⌈y⌉ ≥ 1, which by transitivity of >
contradicts the trichotomy. So necessarily for all r ∈ T , r ≤ ⌈y⌉. This means that T is
bounded above. By the Least Upper Bound property of R, there exists a real number s
that is the least upper bound on T . Necessarily s ≥ 0 as 0 ∈ T .
We claim that s2 = y.
Suppose first that s2 < y. [We want to find a small positive number q such
that (s +q)2 < y, which would say that s +q ∈ T and thus give a contradiction
to s being an upper bound on T . The small number q might as well be of the
form p1 for some (large) positive integer p, so that we really want to find
some p ∈ N+ such that s2 + 2s 1
p + p2 < y, and for that it suffices to find
p ∈ N+ such that 2s + 1 < p(y − s2 ).] By the Archimedean property there exists a
positive integer p such that 2s + 1 < p(y − s2 ). We know that p1 > 0. Since p ≥ 1,
by compatibility of < (and ≤) with multiplication by the positive number p1 we have
that 1 ≥ p1 . Thus 2s + p1 ≤ 2s + 1 < p(y − s2 ). By compatibility of multiplication
again we get that (2s/p) + (1/p2 ) < y − s2 , so that by compatibility of < with addition,
(s + p1 )2 = s2 + 2s 1 1
p + p2 < y. This means that s + p ∈ T , so that s is not an upper bound
on T , which is a contradiction. By trichotomy we conclude that necessarily s2 ≥ y. This
in particular implies that s > 0.
Suppose next that s2 > y. [We want to find a small positive number q such
that (s − q)2 > y. This would prove that for all r ∈ T , r 2 ≤ y < (s − q)2 . We
need to guarantee that s − q is still positive, and then by the increasing
property of the squaring function on R+ , for all r ∈ T , r ≤ s − q, which
contradicts the fact that s is the supremum of T . This q might as well be
of the form p1 for some large positive integer p.] By the Archimedean property
there exists a positive integer p1 such that 2s < p1 (s2 − y), and there exists a positive
integer p2 such that 1 < p2 s. Let p = max{p1 , p2 }. Then 2s < p(s2 − y) and 1 < ps.
The assumption 1 < ps implies that s − p1 is positive. Similarly to before, we derive that
2s − p1 < p(s2 − y) and with compatibilities we derive that y < (s − p1 )2 . Then for any
r ∈ T , r 2 < y < (s − 1p )2 . But then by Corollary 2.8.3, r < s − 1p . Hence s − p1 is an
upper bound on T , contradicting the assumption that s is the least upper bound on T .
 Section 2.10: Absolute values 93

This proves that s2 ≤ y.

Then the trichotomy on R gives that s2 = y.

Exercises for Section 2.9

2.9.1 Let y be a positive real number. Prove that there exists a positive real number x
such that x3 = y.
*2.9.2 Let n be a positive integer and let y be a positive real number. Prove that there
exists a positive real number x such that xn = y.
*2.9.3 Let n be a positive odd integer and let y be a real number. Prove that there exists
a real number x such that xn = y.
Pn
2.9.4 Let S = { k=1 (−1)k k1 : n ∈ N+ }. Prove that S is bounded above and below.
Pn
2.9.5 Find the least upper bound of { k=0 101−k : n ∈ N+ }. (Hint: Example 1.5.4.)

2.10 Absolute values

Definition 2.10.1 Let F be an ordered field. The absolute value function | | : F → F
is a function defined as (
0; if x = 0;
|x| = x; if x ∈ F + ;
−x; if x ∈ F − .
This defines the absolute value function on the ordered fields Q and R.
The following theorem lists the familiar properties of absolute values, and the reader
may wish to prove them without reading the given proof.

Theorem 2.10.2 Let F be an ordered field.

(1) For all x ∈ F , |x| ≥ 0. Furthermore, x ≥ 0 if and only if x = |x|; and x ≤ 0 if and
only if x = −|x|. (In particular, |1| = 1.)
(2) For all x ∈ F , |x| = | − x|.
(3) For all x ∈ F , −|x| ≤ x ≤ |x|.
(4) For all x, a ∈ F , |x| ≤ a if and only if −a ≤ x ≤ a.
(5) For all x, a ∈ F , |x| < a if and only if −a < x < a.
(6) For all x, y ∈ F , |xy| = |x||y|.

Proof. (1) is from the definition.

(2) Certainly |0| = | − 0| = 0. Suppose that x ∈ F − . Then |x| = −x and −x ∈ F + so
that | − x| = −x. Thus |x| = −x = | − x|. If x ∈ F + , then −x ∈ F − , so that by what we
just proved, | − x| = | − (−x)| = |x|.
94 Chapter 2: Concepts with which we will do mathematics

(3) If x ≥ 0, then |x| = x, and if x < 0, then x < 0 < |x|. Thus by transitivity for
all x, x ≤ |x|. In particular, when applied to −x, this says that −x ≤ | − x| = |x|, and by
adding x − |x| to both sides we get that −|x| ≤ x.
(4) Suppose that |x| ≤ a. Then by (3) and transitivity, x ≤ a, and −x ≤ | − x| = |x| ≤
a, so that by transitivity and adding x − a to both sides, −a ≤ x.
(5) The proof of (5) is similar to that of (4).
(7) We may choose r and s ∈ {1, −1} such that rx ≥ 0 and sy ≥ 0. Then

|xy| = |(rx)(sy)| (by Exercises 2.6.8 and 2.6.9 and (2))

= (rx)(sy) (by Theorem 2.7.11)
= |rx| · |sy| (by the definition of r, s)
= |x||y| (by (2)).

In the next theorem we use the standard notation “±” when we mean that the result
should hold in place either for + or for −.

Theorem 2.10.3 The following inequalities hold for an ordered field F .

(1) Triangle inequality: For all x, y ∈ F , |x ± y| ≤ |x| + |y|.
(2) Reverse triangle inequality: For all x, y ∈ F , |x ± y| ≥ ||x| − |y|| = ||y| − |x||.

Proof. (1) By the first part of the previous theorem, −|x| ≤ x ≤ |x| and −|y| ≤ y ≤ |y|.
Thus

−(|x| + |y|) = (−|x|) + (−|y|) (by Theorem 2.5.14)

≤ x + (−|y|) (by compatibility of ≤ with addition)
≤ x + y (by compatibility of ≤ with addition)
≤ |x| + y (by compatibility of ≤ with addition)
≤ |x| + |y| (by compatibility of ≤ with addition),

so that by transitivity, −(|x| + |y|) ≤ x + y ≤ |x| + |y|. Thus by part (4) of the previous
theorem, |x + y| ≤ |x| + |y|. It follows that |x − y| = |x + (−y)| ≤ |x| + | − y|, and by the
second part again this is equal to |x| + |y|.
(2) By (1), |x| = |x ± y − ±y| ≤ |x ± y| + |y|, so that |x| − |y| ≤ |x ± y|. Similarly,
|y| − |x| ≤ |y ± x|. But by the second part of the previous theorem, |y − x| = | − (y − x)| =
|x − y| and |y + x| = |x + y|, so that |x ± y| ≥ |x| − |y| and |x ± y| ≥ |y| − |x| = −(|x| − |y|).
Since ||x| − |y|| is either |x| − |y| or |y| − |x|, (2) follows.
 Section 2.10: Absolute values 95

Theorem 2.10.4 Let F be an ordered field. Let r ∈ F .

(1) If r < ǫ for all ǫ ∈ F + , then r ≤ 0.
(2) If r > −ǫ for all ǫ ∈ F + , then r ≥ 0.
(3) If |r| < ǫ for all ǫ ∈ F + , then r = 0.

Proof. Proof of (1): By Theorem 2.7.10, 1 ∈ F + , so that 0 < 1, and by compatibility of

< with addition, 1 < 2. Thus by transitivity of <, 0 < 2, so that 2 is positive, and by
Theorem 2.7.10 (3), 2−1 ∈ F + . Since r > 0, by compatibility of < with multiplication by
positive numbers, ǫ = r/2 is a positive number. By assumption, r < ǫ = r/2. Again by
compatibility of < with multiplication, by multiplying through by 2r −1 , we get that 2 < 1,
which contradicts the trichotomy (since we already established that 1 < 2). Thus r 6∈ F + ,
so that r ≤ 0.
The proof of (2) is similar.
For (3), if |r| < ǫ for all ǫ ∈ F + , then −ǫ < r < ǫ. Then by (1) and (2), 0 ≤ r ≤ 0.
Since F + ∩ F − = ∅ by trichotomy, it follows that r = 0.

Exercises for Section 2.10

2.10.1 Let F be an ordered field. Prove that the absolute value function on F is increasing
on F + ∪ {0} and decreasing on F − ∪ {0}.
2.10.2 Let F be an ordered field.  
 
i) Prove that for all a, b ∈ F , |a| + |b| = |a| + |b|.
 
 
ii) Prove that for all a, b ∈ F + , |a| + |b| = a + b.
 
 
iii) Prove that for all a, b ∈ F − , |a| + |b| = −a − b.
2.10.3 Let F be an ordered field, a ∈ F and r ∈ F + . Express the sets {x ∈ F : |x−a| < r}
and {x ∈ F : |x − a| ≤ r} in interval notation.
2.10.4 (Triangle inequality) Let F be an ordered field and a1 , . . . , an ∈ F . Prove that

|a1 + a2 + · · · + an | ≤ |a1 | + |a2 | + · · · + |an |.

2.10.5 (Reverse triangle inequality) Let F be an ordered field and a1 , . . . , an ∈ F . Prove

that
|a1 + a2 + · · · + an | ≥ |a1 | − |a2 | − · · · − |an |.
 
 
Give an example of real numbers a1 , a2 , a3 with |a1 + a2 + a3 | 6≥ |a1 | − |a2 | − |a3 |.
2.10.6 Let F be an ordered field and S a subset of F .
i) Prove that S is bounded if and only if {|s| : s ∈ S} is bounded.
ii) Give an example of a set S in R that is bounded above but {|s| : s ∈ S} is not.
Chapter 3: The field of complex numbers, and topology

In this chapter we construct complex numbers from the real numbers. Even if we
want to measure only real numbers in the real world, the more general complex numbers
streamline many constructions, so they are an important tool. We prove the important
basic properties of complex numbers and their field C in the first two sections. As a starting
point we take it as fact that R is an ordered field with the Least upper bound property
and the Archimedean property.
The last two sections are about the Euclidean topology on R and C, which is crucial
in the subsequent chapters on limits. I cover those two sections lightly but invoke them as
needed.

3.1 Complex numbers

Note that there is no real number x such that x2 = −1. In this section we build the
smallest possible field containing R with an element whose square is −1.

Definition 3.1.1 Let C = R × R (the Cartesian product). Elements of C are called

complex numbers. Define binary operations + and · on C:
(a, b) + (c, d) = (a + c, b + d),
(a, b) · (c, d) = (ac − bd, ad + bc).

We represent complex numbers in the real plane like so:

b b
(0, 1)
(−3, 1) b
(2, 0.5)
(0, 0)
b b b

(−1, 0) (1, 0)
b
(0, −1)
 Section 3.1: Complex numbers 97

Definition 3.1.2 The horizontal axis, on which all complex numbers are of the form (r, 0),
is called the real axis, and the vertical axis, on which all complex numbers are of the form
(0, r), is called the imaginary axis.
The real part of (a, b) is Re(a, b) = a, and the imaginary part is Im(a, b) = b.

The illustration below shows a geometric interpretation of addition: to add (a, b) and
(c, d), draw the parallelogram using these two points and (0, 0) as three of the four vertices.
Think through why the sum is the fourth vertex.

(a + c, b + d)
b

(c, d) b

b
(a, b)
b

(A geometric interpretation of multiplication is in Theorem 3.4.1.)

Theorem 3.1.3 C is a field.

However, C is not an ordered field in the sense of Definition 2.7.8, and a rigorous proof
is left for Exercise 3.1.6.
Rather than giving a formal proof that C is a field, below is a list of the necessary
easy verifications. The reader is encouraged to verify all.
(1) ·, + are associative and commutative.
(2) · distributes over +.
(3) For all x ∈ C, (0, 0) + x = x. In other words, C has the additive identity 0 = (0, 0).
(The additive identity of a field is written as “0” even when it is an ordered pair
of real numbers.)
(4) For all (a, b) ∈ C, (−a, −b)+(a, b) = (0, 0). In other words, every element (a, b) has
an additive inverse −(a, b) = (−a, −b). By Theorem 2.5.10, the additive inverse is
unique.
(5) For all x ∈ C, (1, 0) · x = x. In other words, C has a multiplicative identity
1 = (1, 0). (The multiplicative identity of a field is written as “1” even when it is
an ordered pair of real numbers.)
(6) (1, 0) 6= (0, 0). i.e., 1 6= 0.
(7) Every non-zero element has a multiplicative inverse: for all (a, b) 6= (0, 0),
a −b a −b
(a, b)−1 = ( a2 +b 2 , a2 +b2 ) ∈ C and ( a2 +b2 , a2 +b2 ) · (a, b) = (1, 0). By Theo-
rem 2.5.10, the multiplicative inverse is unique. For example, the multiplicative
3 5i


inverse of (3, 5) is 34 , − 34 .
98 Chapter 3: The field of complex numbers, and topology

Notation 3.1.4 There is another, in some ways better, notation for elements of C which
you should adopt as well: (a, b) = a + bi, with (a, 0) = a and (0, b) = bi. This notational
convention does not lose any information, but it does save a few writing strokes. Addition
is easy: (a + bi) + (c + di) = (a + b) + (c + d)i, the additive inverse of a + bi is −a − bi,
the additive identity is 0. This notation justifies the possibly strange earlier definition of
multiplication in C:
(a, b) · (c, d) = (a + bi)(c + di) = ac + adi + bic + bidi = (ac − bd, ad + bc).

The multiplicative identity is 1, i = 0 + 1 · i is that number for which i2 = (−i)2 = −1, and
the multiplicative inverse of a non-zero a + bi is aa−bi
2 +b2 .

With this notation, the real part of a + bi is Re(a + bi) = a, and the imaginary part
is Im(a + bi) = b.
We next identify R as a subset of C: the operative word here is “identify”, as
R is not a subset of R × R, i.e., real numbers are not equal to ordered pairs of real
numbers. Nevertheless, with the natural identification of any real number r with the
complex number (r, 0) = r + i · 0 = r, we can think of R as a subset of C. The addition
and multiplication on R are compatible with those on C in the sense that for real numbers
r and s, the addition and multiplication of r and s as elements of R yield the same answers
as the corresponding addition and multiplication of r and s as elements of C.
In total we have the following natural inclusions that are compatible with addition
and multiplication:
N0 ⊆ Z ⊆ Q ⊆ R ⊆ C.

These number systems progressively contain more numbers and more solutions of more
equations. For example, the equation 1 + x = 0 does not have any solutions in N0 but it
does have one in Z; the equation 1 + 2x = 0 does not have any solutions in Z, but it does
have one in Q; the equation 2 − x2 = 0 does not have any solutions in Q, but it does have
√ √
two in R (namely 2 and − 2); the equation 2 + x2 = 0 does not have any solutions in R,
√ √
but it does have two in C (namely 2i and − 2i). Furthermore, the standard quadratic
formula always yields roots of quadratic equations in C.*

* One of the excellent properties of C is the Fundamental Theorem of Algebra: every polynomial with
coefficients in C or R has roots in C. The proof of this fact is proved in a junior-level class on complex analysis or
in a senior-level class on algebra. The theorem does not say how to find those roots, only that they exist. In fact,
there is another theorem in Galois theory that says that in general it is impossible to find roots of a polynomial
by using radicals, sums, differences, products, and quotients.
 Section 3.2: Functions related to complex numbers 99

Exercises for Section 3.1

3.1.1 Write the following elements in the form a + bi with a, b ∈ R:
1
i) 1+i .
ii) (4+i)(7−3i)
(1−i)(3+2i) .
iii) (2 − 3i)5 . (Hint: Exercise 1.6.7.)
√
3
3.1.2 Let x = 2
+ 2i . Draw x, x2 , x3 , x4 , x5 , x6 , x7 . What do you observe?
3.1.3 Draw the following sets in C:
i) {x : the real part of x is 3}.
ii) {x : the imaginary part of x is 3}.
iii) {x : the product of the real and imaginary parts of x is 3}.
iv) {x : the product of the real and imaginary parts of x is 0}.
3.1.4 Let x ∈ C such that x2 = −1. Prove that either x = i or x = −i.
3.1.5 Prove that the only x ∈ C with x2 = 0 is x = 0.
3.1.6 Prove that C is not ordered in the sense of Definition 2.7.8. Justify any facts (such
as that 0 < 1).

3.2 Functions related to complex numbers

We have established existence and properties of several functions of complex numbers:

(1) Inclusion of R into C: f (r) = r = r + 0 · i = (r, 0).
(2) Identity function on C: idC (x) = x. (Say why this function is different from the
function in part (1).)
(3) Real part Re : C → R.
(4) Imaginary part Im : C → R.
(5) Additive inverse − : C → C.
(6) Multiplicative inverse ( 1 ) : (C \ {0}) → C.
(7) Addition + : C × C → C.
(8) Multiplication · : C × C → C.
(9) Scalar multiplication functions: for any z ∈ C, multiplication by z is a function
with domain and codomain C: f (x) = zx.
There are further obvious functions:
(10) A function f : C → C is called polynomial if there exist a0 , a1 , . . . , an ∈ C such
that for all x ∈ C, f (x) = a0 + a1 x + a2 x2 + · · · + an xn .
(11) A function is called rational if it equals a polynomial function divided by a poly-
nomial function. Examples of rational functions are polynomial functions, as well
x2 +i
as such functions as x21+1 and x−3i .
100 Chapter 3: The field of complex numbers, and topology

From now on, when the domain of a function is not given explicitly, we take the
domain to be the largest possible subset of C on which the function makes sense. (Before
we took the largest possible subset of R; see Notation 2.4.5.) So, the domain of the last
two functions in the previous paragraph are C \ {i, −i} and C \ {3i}.
There is one more very important and powerful function on C, which at first may seem
unmotivated:

Definition 3.2.1 The complex conjugate of a + bi, is a + bi = a − bi ∈ C.

Geometrically, the complex conjugate of a number is the reflection of the number

across the real axis.

Theorem 3.2.2 Let x, y ∈ C. Then

(1) The complex conjugate of the complex conjugate of x equals x. In symbols: x = x.
(2) x + y = x + y.
(3) x · y = x · y.
(4) If y 6= 0, then y 6= 0 and (x/y) = x/y.
(5) x = x if and only if x ∈ R; and x = 0 if and only if x = 0.

Proof. Write x = a + bi and y = c + di for some a, b, c, d ∈ R. Certainly x = a − bi =

a + bi = x, proving (1). Addition in (2) is straightforward. The following proves (3):
x · y = (a + bi) · (c + di)
= ac − bd + (ad + bc)i
= ac − bd − (ad + bc)i
= (a − bi) · (c − di)
= x · y.
If y 6= 0, then by (3), x = (x/y)y = x/y · y, and so (4) follows.
Certainly x = x if and only if a + bi = a − bi, i.e., if and only if (a, b) = (a, −b), and
that holds if and only if b = 0. Thus x = x if and only if x = a ∈ R. Also, x = 0 if and only
if (a, b) = (0, 0), which is the same as saying that x = (a, −b) = (0, 0). This proves (5).

Remark 3.2.3 All functions above with domain and codomain equal to C were given
with some sort of algebraic formulation or description. How else can we represent such
a function? We certainly cannot give a tabular function formulation since the domain is
infinite. But we cannot draw such a function either: for the domain we would need to draw
the two-dimensional real plane, and the same for the codomain, so we would have to draw
the four-dimensional picture to see it all, and that is something we cannot do. So we need
to be satisfied with the algebraic descriptions of functions.
 Section 3.3: Absolute value in C 101

Exercises for Section 3.2

3.2.1 Let x, y ∈ C.
i) Prove that x + x = 2 Re x.
ii) Prove that x − x = 2i Im x.
iii) Prove that x · y + x · y = 2 Re(xy) = 2 Re x Re y + 2 Im x Im y.
3.2.2 Let x be a complex number. Prove that x · x is real.
3.2.3 Let x and y be non-zero complex numbers such that x · y is real. Prove that there
exists a real number r such that y = r · x.
3.2.4 Let x be non-zero complex number such that x2 is real. Prove that Re(x)·Im(x) = 0,
i.e., that either Re(x) or Im(x) is zero.

3.3 Absolute value in C

We have seen the absolute value function in ordered fields. The Pythagorean theorem
in the plane R × R motivates the natural definition of distance in C:

bi b
(a, b) = a + bi

a
Definition 3.3.1 The absolute value of a complex number (a, b) = a + bi is |a + bi| =
√
a2 + b2 ∈ R. The absolute value is also called the norm or the length. The distance
between complex numbers x and y is |x − y|.

Since the absolute value is a real number, this gives a way to partially compare complex
numbers, say by their lengths, or by their real components. But recall Exercise 3.1.6: C is
not an ordered field.
The absolute value of (a, 0) = a or (0, a) = ia is |a|; the absolute value of (1, 1) = 1 + i
√ √ √ √ √ √
is 2; the absolute value of (1, 2) = 1+i 2 is 3; the absolute value of (1, 3) = 1+i 3
√
is 4 = 2, et cetera.
102 Chapter 3: The field of complex numbers, and topology

Theorem 3.3.2 Let x, y ∈ C. Then

(1) |x| = |x|.
(2) x · x = |x|2 is a non-negative real number.
(3) x = 0 if and only if |x| = 0.
(4) | Im x|, | Re x| ≤ |x|.
(5) If x 6= 0, then x−1 = x/|x|2 .
(6) |xy| = |x| |y|.
(7) (Triangle inequality) |x ± y| ≤ |x| + |y|.
(8) (Reverse triangle inequality) |x ± y| ≥ ||x| − |y||.
√ p
Proof. Write x = a+bi for some a, b ∈ R. Then |x| = a2 + b2 = a2 + (−b)2 = |x|, which
proves (1). Also, x · x = (a + bi)(a − bi) = a2 − (bi)2 = a2 + b2 = |x|2 , and this is the sum
of two non-negative real numbers, and is thus non-negative. This proves (2). Furthermore,
since R is an ordered field, by Exercise 2.7.13, a2 , b2 ≥ 0, so by Exercise 2.7.14, a2 + b2 = 0
if and only if a = b = 0. This proves (3).
√
Since b2 ≥ 0, it follows that a2 ≤ a2 + b2 , so that by Corollary 2.8.3, |a| = a2 ≤
√
a2 + b2 . This proves that | Re x| ≤ |x|. Similarly | Im x| ≤ |x|. This proves (4).
If x 6= 0, then by (3), |x| is a non-zero (real, complex) number, and by (2), x/|x|2 is
the multiplicative inverse of x. This proves (5).
We could prove (6) with straightforward but laborious algebra by using that x = a+bi
and z = c + di for some real numbers a, b, c, d, and expanding the relevant sides, but the
following proof is better:
|xy|2 = (xy)(xy) (by (3))
= xyx y (by Theorem 3.2.2)
= xxyy (by associativity and commutativity of · in C)
= |x|2 |y|2 (by (3))
= (|x||y|)2 (by associativity and commutativity of · in R).
Now (6) follows by taking square roots of both sides.
To prove the triangle inequality, we also use the squares of the desired quantities to
avoid having to write the square root:
2
|x ± y| = (x ± y)(x ± y) (by (2))
= (x ± y)(x ± y) (by Theorem 3.2.2 (2))
= xx ± xy ± yx + yy (by algebra)
= |x|2 ± xy ± xy + |y|2 (by Theorem 3.2.2 (2))
= |x|2 ± 2 Re(xy) + |y|2 (by Exercise 3.2.1 iii))
 Section 3.3: Absolute value in C 103

≤ |x|2 + 2| Re(xy)| + |y|2 (comparison of real numbers)

≤ |x|2 + 2|xy| + |y|2 (by (4))
= |x|2 + 2|x||y| + |y|2 (by (6))
= |x|2 + 2|x||y| + |y|2 (by (1))
= (|x| + |y|)2,
and since the squaring function is strictly increasing on the set of non-negative real numbers,
it follows that |x ± y| ≤ |x| + |y|. This proves (7), and by Theorem 2.10.3 also (8).
A consequence of part (6) of this theorem is that if x ∈ C has absolute value greater
than 1, then positive integer powers of x have increasingly larger absolute values, if |x| < 1,
then positive integer powers of x have get increasingly smaller than 1, and if |x| = 1, then
all powers of x have absolute value equal to 1.
The absolute value allows the definition of bounded sets in C (despite not having an
order on C):

Definition 3.3.3 A subset A of C is bounded if there exists a positive real number M

such that for all x ∈ A, |x| ≤ M .

For example, any set with only finitely many elements is bounded: if A = {x1 , . . . , xn },
set M = max{|x1 |, . . . , |xn |} + 1, and then certainly for all x ∈ A, |x| < M .
The subset Z of C is not bounded. The infinite set {x ∈ C : |x| = 5} is bounded.
The set {in : n ∈ N+ } is bounded. The set {1/n : n ∈ N+ } is bounded. The set
{x ∈ C : the angle of x counterclockwise from the positive x-axis is π/4} is not bounded.
(Draw these sets.)

Exercises for Section 3.3

3.3.1 Draw the following sets in C:
i) {x : |x| = 3}.
ii) {x : |x − 2 + i| = 3}.
3.3.2 Let a ∈ C, let B be a positive real number, and let A = {x ∈ C : |x − a| ≤ B}.
Draw such a set in the complex plane assuming a 6= 0, and prove that A is a bounded set.
3.3.3 Let A and B be subsets of C. Define A + B = {a + b : a ∈ A, b ∈ B}, A · B = {a · b :
a ∈ A, b ∈ B}, and for any c ∈ C, let cA = {c · a : a ∈ A}. Compute A + B, A · B, and c · A
for the following A, B, c:
i) A = {1, 2, i}, B = {−1, i}, c = 4.
ii) A = R+ , B = R+ , c = −1.
104 Chapter 3: The field of complex numbers, and topology

3.3.4 Let A be a bounded subset of C.

i) Prove that for any complex number c, {ca : a ∈ A} is bounded.
ii) Prove that for any complex number c, {a + c : a ∈ A} is bounded.
iii) Prove that {a2 : a ∈ A} is bounded.
iv) Prove that for any positive integer n, {an : a ∈ A} is bounded.
v) Prove that for any polynomial function f (x), {f (a) : a ∈ A} is bounded.
3.3.5 Let a, b ∈ C. Suppose that for all real numbers ǫ > 0, |a − b| < ǫ. Prove that a = b.
(Hint: Theorem 2.10.4.)
*3.3.6 (Keep in mind that the square root function on C is yet to be discussed carefully;
see Exercise 5.4.6.) Discuss correctness/incorrectness issues with the following equalities:
√ √ √ √ p √
i) −6 = ( 3i)( 12i) = −3 −12 = (−3)(−12) = 36 = 6.
ii) (R. Bombelli, 1560, when solving the equation x3 = 15x + 4.)
q q
3 √ 3 √
4 = 2 + −121 + 2 − −121.
p √ p √ √
iii) (G. Leibniz, 1675) 1 + −3 + 1 − −3 = 6.

3.4 Polar coordinates

So far we have expressed complex numbers with pairs of real numbers either in ordered-
pair notation (x, y) or in the form x + yi. But a complex number can also be uniquely
determined from its absolute value and the angle measured counterclockwise from the
positive real axis to the line connecting (0, 0) and (x, y).

radius 1, angle π/2, 5π/2, . . .

b
b
radius 1, angle π/4, 9π/4, . . .
radius 2, angles π, −π, 3π, . . .
b b
radius 2, angles 0, 2π, . . .

radius 1, angles 3π/2, −π/2, . . .

b
radius 2, angle 5π/3, −π/3, . . .

Any choice of θ works for the complex number zero. The angles are measured in
radians. (While you may say degrees out loud, get into the habit of writing down radians;
later we will see how radians work better.) The angle is not unique; addition of any integer
multiple of 2π to it does not change the complex number.
 Section 3.4: Polar coordinates 105
√
For further examples, 1+i2 3 is on the unit circle centered
√
at the origin and is at angle
1−i 3
π/3 counterclockwise from the positive real axis, 2 is on the √same unit circle and at
angle −π/3 counterclockwise from the positive real axis, and −1+i 2
3
is on the same circle
and at angle 2π/3 counterclockwise from the positive real axis.
We refer to the entries in the ordered pair (x, y) ∈ R × R = C as Cartesian coordi-
nates. The coordinates (r, θ) consisting of the absolute value r of a complex number and
its angle θ (measured counterclockwise from the positive real axis) are referred to as polar
coordinates.
Numerical conversions between the two coordinate systems use trigonometry. If we
know r and θ, then x and y are given by:
x = r cos θ,
y = r sin θ,
and if we know x and y, then r and θ are given by:
p
r = x2 + y 2 ,

 anything, if x = 0 = y;


 π
 2,
 if x = 0 and y > 0;
θ = −π 2 , if x = 0 and y < 0;


 arctan(y/x) ∈ (−π/2, π/2),
 if x > 0;


arctan(y/x) ∈ (π/2, 3π/2), if x < 0.
Note that the angle is ±π/2 precisely when Re x = 0, that the angle is 0 when x is a
positive real number, that it is π when x is a negative real number, et cetera. Furthermore,
if the angle is not ±π/2, then the tangent of this angle is precisely Im x/ Re x.
We will see in Chapter 9 that the polar coordianates r, θ determine the complex number
as reiθ , but at this point we cannot yet make sense out of this exponentiation. Nevertheless,
they do hint at complex multiplication of reiθ and seiβ as being rsei(θ+β) , confirming that
the absolute value of the product is the product of the absolute values, and hinting that
the angle of the product is the sum of the two angles of the numbers.
We next prove this beautiful fact of how multiplication works geometrically without
reference to future chapters.

Theorem 3.4.1 (Fun fact) Let z be a complex number in polar coordinates r and θ.
Define functions M, S, R : C → C as follows:

M (x) = zx = Multiply x by z,
S(x) = rx = Stretch x by a factor of r,
R(x) = Rotate x by angle θ counterclockwise around (0, 0).
106 Chapter 3: The field of complex numbers, and topology

Then
M = S ◦ R = R ◦ S,

or in other words, multiplication by z is the same as stretching by r followed by or preceded

by rotating by the angle θ counterclockwise.

Proof. If z = 0 or x = 0, the conclusion is trivial, so we assume that x and z are non-zero.

By geometry of rotation, rotation and stretching by a positive real number can be done in
any order, i.e., R ◦ S(x) = R(rx) = rR(x) = S ◦ R(x).
So it suffices to prove that R◦ S(x) = M (x) for all x. We first prove this for the special
cases x = 1 and x = i, after which we prove it for general x.
The angle of M (1) = z is θ and its length is r. But (R ◦ S)(1) = R(r) also has length r
and angle θ, so that M (1) = (R ◦ S)(1).
Write z = (c, d) for some c, d ∈ R. Then M (i) = (−d, c), and we draw a few examples:

b
(c1 , d1 )
(c2 , d2 )
(−d1 , c1 ) b b b
(−d4 , c4 )

(−d2 , c2 ) b b b
(−d3 , c3 )
(c4 , d4 )
(c3 , d3 ) b

The complex number M (i) = (−d, c) has length equal to |(c, d)| = r. The angle between
(c, d) and (−d, c) is 90◦ , or π/2 radians, and more precisely, to get from z = (c, d) to
M (i) = (−d, c) we have to rotate counterclockwise by π/2. Thus the angle formed by M (i)
counterclockwise from the positive real axis is θ + π/2. But (R ◦ S)(i) = R(ri) also has the
same angle and length as M (i), so that M (i) = (R ◦ S)(i).
Now let x be general in C. Write x = a+bi for some a, b ∈ R. By geometry of rotation,
R(a + bi) = R(a) + R(bi) = aR(1) + bR(i). Then
R ◦ S(x) = S ◦ R(x) (as established from geometry)
= rR(x)
= rR(a + bi)
= r(aR(1) + bR(i))
= arR(1) + brR(i)
 Section 3.4: Polar coordinates 107

= aS(R(1)) + bS(R(i))
= aM (1) + bM (i) (by previously proved cases)
= az1 + bzi
= z(a + bi)
= M (x).

Theorem 3.4.2 For any non-zero complex number x and any integer n, the angle of xn
counterclockwise away from the positive x-axis is n times the angle of x. Also, |xn | = |x|n.

Proof. If n = 1, this is trivially true. Now suppose that the theorem is true for some
positive integer n. Then the angle of xn−1 counterclockwise away from the positive x-axis
is n − 1 times the angle of x, and by Theorem 3.4.1, the angle of xn = xxn−1 is the sum
of the angles of x and xn−1 , so that it is n times the angle of x. Similarly, by part (6) of
   
Theorem 3.3.2, |xn | = xxn−1  = |x| xn−1  = |x||x|n−1 = |x|n .
Thus by induction the theorem is proved for all positive n.
Still keep n positive. Since 1 = x−n xn has angle 0 and xn has angle n times the angle
of x, by Theorem 3.4.1, x−n must have angle −n times the angle of x. Also, by part (6)
of Theorem 3.3.2, 1 = |x−n | |xn | = |x−n | |x|n , so that |x−n | = |x|−n . Thus the theorem
holds for all non-zero n.
Finally, if n = 0, then then angle of xn = 1 is 0, which is 0 times the angle of x, and
|x0 | = |1| = 1 = |x|0 .
√
Thus for example, −1+2 3i is on the√unit circle at angle 2π/3 away from the positive
x-axis, and so the second power of −1+2 3i is on the unit circle at angle 4π/3 away from
the positive
√
x-axis, and the cube power is on the unit circle at angle 2π, i.e., at angle 0, so
−1+ 3i 3
that ( 2 ) = 1.

Corollary 3.4.3 Let n be a positive integer. Let A be the set of all complex numbers
on the unit circle at angles 0, 2π 2π 2π 2π
n , 2 n , 3 n , . . . , (n − 1) n . Then A equals the set of all the
complex number solutions to the equation xn = 1.

Proof. Let a ∈ A. By the previous theorem, an has length 1 and angle an integer multiple
of 2π, so that an = 1. If b ∈ C satisfies bn = 1, then |b|n = |bn | = 1, so that the non-
negative real number |b| equals 1. Thus b is on the unit circle. If θ is its angle, then the
angle of bn = 1 is by the previous theorem equal to nθ, so that nθ must be an integer
multiple of 2π. It follows that θ is an integer multiple of 2π
n , but all those angles appear for
the elements of A. Thus every element of A is a root, and every root is an element of A,
which proves the corollary.
Thus we can talk about nth roots of complex numbers. For non-negative real numbers
we choose its non-negative square root as the square root, but there is no natural choice
108 Chapter 3: The field of complex numbers, and topology

for the square root of a complex number; more on that is in the chapter on continuity in
Exercise 5.4.6.

Exercises for Section 3.4

3.4.1 Write the following complex numbers in the form a + bi with a, b ∈ R:
i) length 1 and at angle π/4 counterclockwise from the positive real axis.
ii) of length 1 and at angle −π/4 counterclockwise from the positive real axis.
iii) The product and the sum of the numbers from the previous two parts.
3.4.2 Draw the following points in the real plane, and think about Theorem 3.4.1:
i) 3 − 2i, i(3 − 2i),
ii) −2 − i, i(−2 − i).
iii) 2 − 3i, (1 + i)(2 − 3i).
iv) 1 − i, (1 + i)(1 − i).
3.4.3 Draw the set {x : the angle of x counterclockwise from the positive real axis is π/3}.
3.4.4 “Square” the pentagon drawn below. Namely, estimate the coordinates (real, imag-
inary or length, angle) of various points on the pentagon, square the point, and draw its
image on a different real plane. You need to plot the image not only of the five vertices, but
of several representative points from each side. (Hint: You may want to use Theorem 3.4.1.)

b
i

3.4.5 Prove that for any non-zero z ∈ C there exist exactly two elements in C whose
square equals z. (Hint: Theorem 2.9.2 and Theorem 3.4.1.)
 Section 3.5: Topology on the fields of real and complex numbers 109

3.4.6 Let z be non-zero in C and n ∈ N+ with polar coordinates r and θ. For and integer k,
let zk be the complex number whose absolute value equals r 1/n and whose angle measured
counterclockwise from the positive x axis is kθ/n.
i) Prove that zk is uniquely determined.
ii) Prove that for all k, zkn = z. (Hint: Theorem 3.4.1.)
iii) Prove that there the set {zk : k ∈ N} contains exactly n elements.

3.5 Topology on the fields of real and complex numbers

When reading this section, absorb the following main points of topology: open ball,
open set, limit point, closed set. The main object of this section is to introduce limit points
of sets so that we can in subsequent chapters talk about limits of functions, sequences, and
series.
By a topology on a set we mean that some sets are declared open, subject to the
conditions that the empty set and the whole set have to be open, that arbitrary unions of
open sets be open, and that finite intersections of open sets be open. In any topology, the
complement of an open set is called closed, but a set may be neither open nor closed. A
topology can be imposed on any set, not just R or C, but we focus on these two cases, and
in fact we work only with the “standard”, or “Euclidean” topology.

Definition 3.5.1 Let F be R or C. Let a ∈ F and let r be a positive real number. An

open ball with center a and radius r is a set of the form
B(a, r) = {x ∈ F : |x − a| < r}.

An open set in F is any set that can be written as a union of open balls.

The following are both B(0, 1), but the left one is a ball in R and the right one is a
ball in C. Note that they are different: by definition the left set is an open subset of R,
but if you think of it as a subset of C, it is not open (see Exercise 3.5.1).

−1 0 1 −1 0 1
110 Chapter 3: The field of complex numbers, and topology

Examples 3.5.2
(1) B(a, r) is open.
(2) F = ∪a∈F B(a, 1) is an open set.
(3) The empty set is an open set because it is vacuously a union of open sets (see
page 51).
(4) For real numbers a < b, the interval in R of the form (a, b) is an open set in R
because it is equal to B((a + b)/2, (b − a)/2). The interval (a, ∞) is open because
it equals ∪∞n=1 B(a + n, 1).
(5) The set A = {x ∈ C : 1 < Re x < 3 and 0.5 < Im x < 2} is open in C. Namely,
this set is the union ∪a∈A B(a, min{Re a − 1, 3 − Re a, Im a − 0.5, 2 − Im a}.

−2 −1 1 2 3

(6) The set A = {x ∈ C : Re x < 1, Im x < 2} is open in C. Namely, this set is the
union ∪a∈A B(a, min{1 − Re a, 2 − Im a}.

Theorem 3.5.3 B(a, r) has infinitely many points.

Proof. For each integer n ≥ 2, a + r/n ∈ B(a, r). Since r > 0, these numbers are all
distinct. Since N0 is infinite, so is the set of all integers that are at least 2.

Example 3.5.4 Thus if A is an open subset of F , then either A is empty or A has infinitely
many points. In particular, {a} is not open.

Theorem 3.5.5 Let A be an open set and let a ∈ A. Then there exists r > 0 such that
B(a, r) ⊆ A.

Proof. Since A is open, it is a union of open balls. Thus a is an element of one such ball
B(b, s), with B(b, s) ⊆ A.
Since a ∈ B(b, s), we have that |a − b| < s, so that r =
a
s − |a − b| is a positive real number. (On the illustration, this b

is the distance between a and the outside of the circle.) We b b

claim that B(a, r) ⊆ B(b, s). To prove this, let x ∈ B(a, r). Then
|x −b| = |x −a+a−b| ≤ |x −a| +|a−b| by the triangle inequality,
and since |x − a| < r = s − |a − b|, it means that |x − b| < s, so
that x ∈ B(b, s). This proves the claim, and hence it proves that B(a, r) ⊆ A.
 Section 3.5: Topology on the fields of real and complex numbers 111

Theorem 3.5.6 (Topology on F )

(1) ∅ and F are open.
(2) An arbitrary union of open sets is open.
(3) A finite intersection of open sets is open.

Proof. The empty set can be written as an empty union of open balls, so it is open
vacuously, and F = ∪a∈F B(a, 1), so that F is open. This proves (1).
Every open set is a union of open balls, and so the union of open sets is a union of
open balls, hence open. This proves (2).
Now let A1 , . . . , An be open sets. Let a ∈ A1 ∩ · · · ∩ An . By Theorem 3.5.5, for each
k = 1, . . . , n, there exists ri > 0 such that B(a, rk ) ⊆ Ak . Set r = min{r1 , . . . , rn }. Then
B(a, r) ⊆ ∩nk=1 B(a, rk ) ⊆ ∩nk=1 Ak . Thus for each a ∈ ∩nk=1 Ak there exists ra > 0 such
that B(a, ra ) ⊆ ∩nk=1 Ak . It follows that
∩nk=1 Ak = ∪a∈∩nk=1 Ak B(a, ra ).
This proves that ∩nk=1 Ak is a union of open balls, so it is open.
An arbitrary intersection of open sets need not be open: ∩∞
n=1 B(a, 1/n) = {a}, which
is not open.

Definition 3.5.7 Let A be an arbitrary subset of F and a ∈ F (not necessarily in A). We

say that a is a limit point of A if for all real numbers s > 0, B(a, s) contains elements of
A different from a.

Examples 3.5.8
(1) If A = {a}, then the set of limit points of A is the empty set.
(2) If A = Q, then the set of limit points of A is R.
(3) The set of limit points of ∅ is empty, the set of limit points of R is R, and the set
of limit points of C is C.
(4) The set of limit points of B(a, r) equals {x ∈ F : |x − a| ≤ r}.

Definition 3.5.9 A set A is a closed set if it contains all of its limit points.

Theorem 3.5.10 A is open if and only if F \ A is closed.

Proof. Suppose that A is open. Let x ∈ A. By Theorem 3.5.5, there exists r > 0 such that
B(x, r) ⊆ A. Thus B(x, r) ∩ (F \ A) = ∅, so that x is not a limit point of F \ A. Thus
no point of A is a limit point of F \ A, which proves that any limit points of F \ A are in
F \ A. Thus F \ A is closed.
Now suppose that F \ A is closed. Let x ∈ A. Since F \ A contains all of its limit
points, then x is not a limit point of F \ A. Thus by the definition of limit points, there
112 Chapter 3: The field of complex numbers, and topology

exists r > 0 such that B(x, r) ∩ (F \ A) is empty. This means that B(x, r) ⊆ A. Thus
A = ∪x∈A B(x, rx ) for appropriate rx > 0, so that A is open.
The following is now almost immediate from previous results:

Theorem 3.5.11 (Topology on F )

(1) ∅, F are closed sets.
(2) Arbitrary intersections of closed sets are closed.
(3) Finite unions of closed sets are closed.

Proof. Exercise 2.1.8 proves that the union of the complements of two sets equals the
complement of the intersection and that the it was proved that the intersection of the
complements of two sets equals the complement of the union, and an equally easily proved
mathematical truth is the following generalization to possibly many more sets:
[ \ \ [
F\ Ak = (F \ Ak ), F\ Ak = (F \ Ak ).
k∈I k∈I k∈I k∈I

With this, (2) and (3) follow from the last two theorems, and (1) is trivial.
Both ∅ and F are open and closed, and these turn out to be the only sets that are
both open and closed (see Exercise 3.5.2). Some sets are neither open nor closed (see
Exercise 3.5.1).

Exercises for Section 3.5

3.5.1 Let A be the open ball in R of radius 1 and centered at 0. Since R is a subset of C,
then A is also a subset of C. Prove that A is neither a closed nor an open subset of C.
3.5.2 Let F be either R or C, and let A be a subset of F that is both closed and open.
and let A be a closed and open subset of F . Prove that A = ∅ or A = F .
3.5.3 Prove that Q is neither an open nor a closed subset of R or C.
3.5.4 Sketch the following subsets of C. Determine their sets of limit points, and whether
the sets are open, closed, or neither: {x ∈ C : Im x = 0, 0 < Re x < 1}, {x ∈ C : Im x =
0, 0 ≤ Re x ≤ 1}, {x ∈ C : −2 ≤ Im x ≤ 2, 0 ≤ Re x ≤ 1}, {x ∈ C : 2 ≤ Im x ≤ −2, 0 ≤
Re x ≤ 1}, {1/n : n ∈ N+ }.
3.5.5 For each of the following intervals as subsets of R, determine the set of limit points,
and whether the set is open, closed, or neither in R:
(0, 1), [0, 1], [3, 5), [3, ∞).

3.5.6 Let a be a limit point of a set A. Suppose that a set B contains A. Prove that a is
a limit point of B.
 Section 3.6: The Heine-Borel theorem 113

3.5.7 Give examples of sets A ⊆ B ⊆ C and a ∈ C such that a is a limit point of B but
not of A.
3.5.8 Let A be a subset of C all of whose elements are real numbers. Prove that every
limit point of A is a real number.

3.6 The Heine-Borel theorem

Closed and bounded sets in C and R have many excellent properties – we will for
example see in Section 5.3 that when a good (say continuous) real-valued function has a
closed and bounded domain, then that function achieves a maximum and minimum value,
et cetera. The concept of uniform continuity (introduced in Section 5.5) needs the fairly
technical Heine-Borel theorems proved in this section.

Construction 3.6.1 (Halving closed and bounded subsets of R and quartering closed
and bounded subsets of C) Let A be a bounded subset of of R or of C, and let P be a
property that applies to subsets of A. Boundedness of A guarantees that A fits inside
a closed bounded rectangle R0 in C. If A is a subset of R, we take the “rectangle” R0
to have height zero. This rectangle can be halved lengthwise and crosswise to get four
equal closed subrectangles. In the next iteration we pick, if possible, one of these four
closed quarter subrectangles such that its intersection with A has property P . We call this
subrectangle R1 , and note that the side lengths are half the side lengths in R0 (and so the
area of R1 is one quarter of the area of R0 ). In general, once we have Rn , we pick its closed
quarter subrectangle Rn+1 such that Rn+1 ∩ A has property P and the sides of Rn+1 are
half the lengths of the sides in Rn . Write Rn = [an , bn] × [cn , dn ] for some real numbers
an ≤ bn and cn ≤ dn . By construction, for all n, bn − an = (b0 − a0 )/2n , and
a0 ≤ a1 ≤ a2 ≤ · · · ≤ an ≤ · · · ≤ bn · · · ≤ b2 ≤ b1 ≤ b0 .

This means that {a1 , a2 , a3 , . . .} is a non-empty subset of R that is bounded above, so that
by the Least upper bound theorem (Axiom 2.9.1), a = sup{a1 , a2 , a3 , . . .} is a real number.
By definition, a ≤ b1 , b2 , b3 , . . .. Similarly, b = inf{b1 , b2 , b3 , . . .} is a real number that is
greater than or equal to a. Suppose that a < b. Then by Exercise 2.7.22, there exists a
positive integer N such that 1/2N < (b − a)/(b0 − a0 ). But aN ≤ a ≤ b ≤ bN , so that
0 ≤ b − a ≤ bN − aN = (b0 − a0 )/2N < b − a, which contradicts trichotomy. Thus a = b,
i.e., we just proved that

sup{a1 , a2 , a3 , . . .} = inf{b1 , b2 , b3 , . . .}.

Similarly,
c = sup{c1 , c2 , c3 , . . .} = inf{d1 , d2 , d3 , . . .}.
114 Chapter 3: The field of complex numbers, and topology

This means that the intersection of all the Rn equals the set {a + ci}, consisting of exactly
one complex number. By the shrinking property of the subrectangles, for every δ > 0 there
exists a positive integer N such that RN ∩ A ⊆ B(a + ci, δ).
In particular, “quartering” of the closed and bounded region (interval) A =
[a0 , b0 ] ⊆ R, means halving the rectangle (interval), and the intersection of all the
chosen closed half-rectangles is a set with exactly one element. That element is in A, so a
real number.

Theorem 3.6.2 (The Heine-Borel theorem (in R, C)) Let A be a closed and bounded
subset of R or C. For each c ∈ A let δc be a positive number. Then there exists a finite
subset S of A such that A ⊆ ∪c∈S B(c, δc ).

Proof. We declare that a subset B of A satisfies (property) P if there exists a finite subset
S of B such that B ⊆ ∪c∈S B(c, δc ). We want to prove that A has P .
Suppose for contradiction that A does not have P . Since A is closed and bounded, it
fits inside a closed rectangle R0 . With Construction 3.6.1, we construct iteratively nested
subrectangles R0 ⊇ R1 ⊇ R2 ⊇ · · ·. The quarter subrectangles are chosen so that each
Rn ∩ A does not have P . This is true if n = 0 by assumption. Suppose that Rn has been
chosen so that Rn ∩A does not have P . If the intersection with A of each of the four quarter
subrectangles of Rn has P , i.e., if each of the four subrectangles (as in Construction 3.6.1)
intersected with A is contained in the union of finitely many balls B(c, δc ), then Rn ∩ A is
covered by finitely many such balls as well, which contradicts the assumption on Rn . Thus
it is possible to choose Rn+1 so that Rn+1 ∩ A does not have P . By construction, ∩∞ n=1 Rn
contains exactly one point. Let that point be x. Since each Rn ∩ A has infinitely many
points, x is a limit point of A, and since A is closed, necessarily x ∈ A.
By the shrinking sizes of the Rn , there exists a positive integer N such that Rn ⊆
B(x, δx ). But then Rn ∩ A has P , which contradicts the construction. Thus A has P .

Theorem 3.6.3 Let A be a closed and bounded subset of R or C, and for each a ∈ A
let δa be a positive number. Then there exist a finite subset S of A and a positive real
number δ such that A ⊆ ∪c∈S B(c, δc ) and such that for all x ∈ A there exists c ∈ S such
that B(x, δ) ⊆ B(c, δc ).

Proof. By Theorem 3.6.2, there exists a finite subset S of A such that A ⊆ ∪c∈S B(c, δc /2).
Let δ = 21 min{δc : c ∈ S}. Since S is a finite set, δ is a positive real number.
Let x ∈ A. By the choice of S there exists c ∈ S such that x ∈ B(c, δc /2). Let
y ∈ B(x, δ). Then

|y − c| = |y − x + x − c| ≤ |y − x| + |x − c| < δ + δc /2 ≤ δc ,
so that y ∈ B(c, δc ). It follows that B(x, δ) ⊆ B(c, δc ).
 Section 3.6: The Heine-Borel theorem 115

Exercises for Section 3.6

3.6.1 Let A = {x ∈ R : 2 ≤ x ≤ 3}. Prove that A ⊆ ∪a∈A B(a, 1/a). Does there exist a
finite set S of A such that A ⊆ ∪a∈S B(a, 1/a). If yes, find it, if no, explain why not.
3.6.2 Let A = {1/n : n ∈ N+ } ⊆ R. Prove that A ⊆ ∪a∈A B(a, a/3). Does there exist a
finite set S of A such that A ⊆ ∪a∈S B(a, a/3). Repeat with A = N+ .
3.6.3 Let A be a closed and bounded subset of R or C. Let I be a set and for each k ∈ I
S
let Uk be an open subset of F . Suppose that A ⊆ k∈I Uk . Prove that there exists a finite
S
subset K of I such that A ⊆ k∈K Uk . (Hint: Prove that for each a ∈ A there exists
δa > 0 such that B(a, δa ) is in some Uk . Apply Theorem 3.6.2.)
Chapter 4: Limits of functions

4.1 Limit of a function

Any calculus class teaches about limits, but the domains there are typically intervals
in R. Here we learn a more general definition for (more interesting) domains in C.

Definition 4.1.1 Let A be a subset of C and let f : A → C be a function. Suppose that

a complex number a is a limit point of A (see Definition 3.5.7). The limit of f (x) as x
approaches a is the complex number L if for every real number ǫ > 0 there exists a
real number δ > 0 such that for all x ∈ A, if 0 < |x − a| < δ, then |f (x) − L| < ǫ.
When this is the case, we write lim f (x) = L. Alternatively, in order to not make
x→a
lines crowded with subscripts, we can write limx→a f (x) = L.

It is important to note that we are not asking for f (a). For one thing, a may or may
not be in the domain of f , we only know that a is a limit point of the domain of f . We
are asking for the behavior of the function f at points near a.
We can give a simple geometric picture of this in case the domain and codomain are
subsets of R (refer to Remark 3.2.3 for why we cannot draw functions when domains and
codomains are subsets of C). Below are three graphs of real-valued functions defined on
a subset of R and with a being a limit point of the domain. In each, on the graph of
y = f (x) we cover the vertical line x = a, and with that information, we conclude that
limx→a f (x) = L.

L L L

a a a

The function f from the first graph above might be any of the following:
 Section 4.1: Limit of a function 117

L L L

a a a

In an intuitive sense we are hoping that f (x) for x near a can predict a value of f as we
get arbitrarily close to a. For example, we may not be able to bring x to 0 Kelvin, but if we
can take measurements f (x) for x getting colder and colder, perhaps we can theoretically
predict what may happen at 0 Kelvin. But how believable is our prediction? Perhaps for
our theory to be satisfactory, we need to run experiments at temperatures x that give us
f (x) within ǫ = 10 of the predicted value. Or when instruments get better, perhaps ǫ gets
smaller, say one thousandth. Or a new material is discovered which allows us to get ǫ
even smaller. But no matter what ǫ is determined ahead of time, for the prediction to be
believable, we need to determine a range x within a δ of a but not equal to a, for which
the f -values are within the given ǫ of the prediction.
A graphical way of representing the epsilon-delta definition of limits for real-valued
functions with domains in R is as follows: For every positive ǫ there exists a positive δ such
that for all x in the domain with 0 < |x − a| < δ (the x 6= a in the vertical gray band), the
value of f (x) is within ǫ of L (in the horizontal grey band):

L+ǫ
L
L−ǫ

a−δ a a+δ

Clearly if ǫ gets smaller, δ has to get smaller too; and if ǫ gets larger, we may keep the
old δ.
While these pictures can help our intuition, they do not constitute a proof: the def-
inition is an algebraic formulation, and as such it requires algebraic proofs. In the rest
of the section we examine many examples algebraically, with the goal of mastering the
epsilon-delta proofs. But epsilon-delta proofs are time-consuming, so in the future we will
want to replace them with some shortcuts. We will have to prove that those shortcuts
118 Chapter 4: Limits of functions

are logically correct, and the proofs will require mastering abstract epsilon-delta proofs.
Naturally, before we can master abstract epsilon-delta proofs, we need to be comfortable
with epsilon-delta proofs on concrete examples. In short, in order to be able to avoid
epsilon-delta proofs, we have to master them. (Ha!)

Example 4.1.2 lim (4x − 5) = 7.

x→3

Proof. The function that takes x to 4x − 5 is a polynomial function, so it is defined for

all complex numbers. Thus the domain of the function is C and 3 is a limit point of the
domain. Let ǫ > 0. [We are proving that for all real numbers ǫ > 0 something-
or-other holds. Recall that all proofs of this form start with “Let ǫ be
an arbitrary positive real number,” or abbreviated as we did. Now we have
to prove that the something-or-other holds. But this something-or-other
claims that there exists a real number δ > 0 with a certain property. Thus
we have to construct such a δ. In this first example I simply present a δ
that works, but in subsequent examples I show how to find a working δ. In
general, the proof should contain a specification of δ and a demonstration
that it works, but how one finds that δ is in general left to scratch work
or to inspiration. ] Set δ = ǫ/4. [Never mind how this magic ǫ/4 appears here;
wait until the next example where I present a process of finding δ.] Then
δ is a positive real number. [Sure! — Now we have to prove that for all x, if
0 < |x − 3| < δ, then |f (x) − L| < ǫ. The proof of “for all x . . .” starts with:]
Let x be an arbitrary complex number. [For this x we now have to prove that if
0 < |x − 3| < δ, then |f (x) − L| < ǫ. The proof of “If P then Q” starts with
“Assume P .”] Assume that 0 < |x − 3| < δ. [Now we have to prove Q, i.e., we
have to prove that |f (x) − L| = |(4x − 5) − 7| < ǫ. We do not simply write
“|(4x − 5) − 7| < ǫ” because we do not know that yet. We write the left side
of this inequality, and manipulate it – algebraically, often with triangle
inequalities and several steps, until we get < ǫ. ] Then
|(4x − 5) − 7| = |4x − 12|
= 4|x − 3|
< 4δ
ǫ
= 4 [Wasn’t ǫ/4 a clever choice of δ?]
4
= ǫ.

The commentary in the proof above is describing the thought process behind the proof
but need not and should not be written out in homework solutions. Below is a homework-
style solution:
 Section 4.1: Limit of a function 119

Proof of Example 4.1.2 without the commentary: The function that takes x to 4x − 5 is
a polynomial function, so it is defined for all complex numbers. Thus the domain of the
function is C and 3 is a limit point of the domain. Let ǫ > 0. Set δ = ǫ/4. Then δ is a
positive real number. Let x be an arbitrary complex number. Assume that 0 < |x − 3| < δ.
Then
|(4x − 5) − 7| = |4x − 12|
= 4|x − 3|
< 4δ
ǫ
=4
4
= ǫ.

In the previous example δ appeared magically as ǫ/4, and it happened to be a positive

number depending on ǫ that made the limit proof work. Other numbers would have worked
as well, such as ǫ/5, or ǫ/(10000 + ǫ), and so on. For good style choose simple formulations
over complicated ones. If some positive number works as δ, so does any smaller positive
number, so there is no smallest possible δ and there are many different correct choices of δ.
It is not necessary to find the largest possible δ for a given ǫ, and it is not even necessary
to show how you derived or chose your δ. However, it is necessary to show that your pick
of δ does satisfy the rest of the defining property of limit.
Think through the comments in the previous paragraph on limx→4 5 = 5: any positive
number works for δ.
In further examples we show all the necessary work for proofs of limits including how
to determine the δ. It is standard to use the proof-writing trick of partial filling-in: when
by the definition of limit it is time in the proof to declare what δ is, we typically write “Set
δ = ”, and in subsequent reasoning we fill in the underlined blank
with whatever restrictions seem necessary. Study the proofs below for how this is done,
but let me just say that each restriction puts an upper bound on δ, say bound δ above
by 1 and by ǫ2 /4, and this is accomplished by filling in “Set δ = min{1, ǫ2 /4 }”,
with room for possible further restrictions. (In Example 4.1.2 we only had one restriction,
namely δ ≤ ǫ/4, so that δ = min{ǫ/4} = ǫ/4.)
I find that the following two goals make these proofs more concrete and doable:
• Goal #1: Write |f (x) − L| as less than or equal to (something) times |x − a|,
If f is a rational function, then |x − a| better be a factor of |f (x) − L|!
• Goal #2: Find a positive constant upper bound B on the (something) found in
Goal #1. Then you make sure that δ ≤ ǫ/B.
You may want to keep these guidelines in mind as you read the proofs below.
120 Chapter 4: Limits of functions

Example 4.1.3 lim (4x2 − 5x + 2) = 28.

x→−2

Proof. The function that takes x to 4x2 − 5x + 2 is a polynomial function and it is

defined for all complex numbers. Thus −2 is a limit point of the domain. Let ǫ > 0.
Set δ = [δ to be determined still; the final write-up will
have this filled in, but we do not yet know δ.]. Then δ is a positive real number.
[Hoping, anyway.] Let x be any complex number. Suppose that 0 < |x + 2| < δ. Then
|(4x2 − 5x + 2) − 28| = |4x2 − 5x − 26| [Want Goal #1: x − a better be a factor.]
= |(x + 2)(4x − 13)|
= |4x − 13| · |x + 2| [Goal #1 accomplished]
[Goal #2: we want to bound above the coefficient |4x − 13| of |x − a| by a
constant. But obviously |4x − 13| is not bounded above by a constant for
all x. So we need to make a restriction on δ: let’s make sure that δ is at
most 1 (or 2, or 15, it does not matter what positive number you pick in this
example). So on the δ line write: δ = min{1, }. This guarantees that
δ will be 1 or smaller, depending on what gets filled in after the comma.]
= |4(x + 2 − 2) − 13| · |x + 2| (adding a clever zero)
= |4(x + 2) − 21| · |x + 2|
 
≤ |4(x + 2)| + 21 · |x + 2| (by the triangle inequality |a ± b| ≤ |a| + |b|)

< (4 · 1 + 21) · |x + 2| (since |x + 2| < δ ≤ 1)

= 25 · |x + 2| [Goal #2 accomplished.]
[Now go back to specifying δ at the beginning of the proof by filling in
with: δ = min{1, ǫ/B}. This means that δ is the smaller of 1 and ǫ/B, and in
particular δ ≤ 1 and δ ≤ ǫ/B.]
< 25 · δ
≤ 25 · ǫ/25
= ǫ.

The final version of the proof of lim (4x2 − 5x + 2) = 28 then looks like this:
x→−2

The function that takes x to 4x2 − 5x + 2 is a polynomial function and it is defined

for all complex numbers. Thus −2 is a limit point of the domain. Let ǫ > 0. Set δ =
min{1, ǫ/25}. Then δ is a positive real number. Let x be any complex number. Suppose
 Section 4.1: Limit of a function 121

that 0 < |x + 2| < δ. Then

|(4x2 − 5x + 2) − 28| = |4x2 − 5x − 26|

= |(x + 2)(4x − 13)|
= |4x − 13| · |x + 2|
= |4(x + 2 − 2) − 13| · |x + 2| (adding a clever zero)
= |4(x + 2) − 21| · |x + 2|
≤ (|4(x + 2)| + 21) · |x + 2| (by the triangle inequality |a ± b| ≤ |a| + |b|)
< (4 · 1 + 21) · |x + 2| (since |x + 2| < δ ≤ 1)
= 25 · |x + 2|
< 25 · δ
≤ 25 · ǫ/25
= ǫ.

We repeat the same type of discovery work on the next example, with fewer comments:

Example 4.1.4 lim (4/x2 ) = 4.

x→−1

Proof. The function that takes x to 4/x2 is defined for all non-zero complex numbers, so
−1 is a limit point of the domain. Let ǫ > 0. Set δ = . Then δ is a positive
real number. Let x be any complex number that satisfies 0 < |x + 1| < δ. Then
 
 4 − 4x2 
|4/x − 4| = 
2 
x2 
 
 (1 − x)(1 + x) 
= 4  

x2
 
1 − x
= 4  2  |1 + x|[Goal #1 accomplished: something times |x − a|.]
x


[Now we want the something 4 1−x x 2 to be at most some constant. Certainly
if we allow x to get close to 0, then (1 − x)/x2 will have a very large size,
so in order to find an upper bound, we need to make sure that x stays away
from 0. Since x is within δ of −1, in order to avoid 0 we need to make sure
that δ is strictly smaller than 1. For example, make sure that δ ≤ 0.4. Thus,
on the δ line write: δ = min{0.4, }.]
 
 2 − (x + 1) 
= 4   |1 + x| (by rewriting 1 − x = 2 − (x + 1))

|x|2
2 + |x + 1|
≤4 |1 + x| (by the triangle inequality)
x2
122 Chapter 4: Limits of functions

2.4
≤4 |1 + x| (since δ ≤ 0.4)
|x|2
9.6
= |1 + x| (by rewriting x = x + 1 − 1)
|x + 1 − 1|2
9.6
≤ |1 + x| (by the reverse triangle inequality because
0.62
|x + 1 − 1| ≥ 1 − |x + 1| > 1 − δ ≥ 1 − 0.4 = 0.6,
so that 1/|x + 1 − 1|2 < 1/0.62)
[On the δ line now write: δ = min{0.4, 0.62ǫ/9.6}.]
9.6
< δ
0.62
9.6
≤ 0.62 ǫ/9.6
0.62
= ǫ.
And here is another example for good measure, with δ already filled in:
4x3 +x
Example 4.1.5 lim = −2.
x→2i 8x−i
3
Proof. The domain of 4x +x
8x−i
consists of all complex numbers different from i/8, so 2i is a
limit point of the domain. Let ǫ > 0. Set δ = min{1, ǫ/9}. [It is so obvious that this
minimum of two positive numbers is positive that we skip the assertion “Thus
δ is a positive real number.” Do not omit the assertion or the checking of
its veracity for more complicated specifications of δ.] Let x be any complex
number different from i/8 such that 0 < |x − 2i| < δ. [Here we merged: “Let x be any
complex number different from i/8. Let x satisfy 0 < |x − 2i| < δ.” into one
shorter and logically equivalent statement “Let x be any complex number
different from i/8 such that 0 < |x − 2i| < δ.”] Then
 3   3 
 4x + x   4x + x 
   
 8x − i − (−2) =  8x − i + 2
 3 
 4x + x + 16x − 2i 
=  
8x − i 
 3 
 4x + 17x − 2i 
=  

8x − i
 
 (4x2 + 8ix + 1)(x − 2i) 
=  

8x − i
 
 (4x2 + 8ix + 1) 
=   |x − 2i| (Goal #1: x − a is a factor.)

8x − i
|4x2 | + |8ix| + 1
≤ |x − 2i| (by the triangle inequality)
|8x − i|
 Section 4.1: Limit of a function 123

4|(x − 2i + 2i)|2 + 8|x − 2i + 2i| + 1

≤ |x − 2i|
|8(x − 2i) + 15i|
4(|x − 2i| + 2)2 + 8(|x − 2i| + 2) + 1
≤ |x − 2i|
−8|x − 2i| + 15
(by the triangle and reverse triangle inequalities)
2
4(1 + 2) + 8(1 + 2) + 1
≤ |x − 2i| (since |x − 2i| < δ ≤ 1)
−8 + 15
61
= |x − 2i|
7
< 9δ
≤ 9ǫ/9
= ǫ.

The next example is of a limit is at a non-specific a.

Example 4.1.6 lim (x2 − 2x) = a2 − 2a.

x→a

Proof. Any a is a limit point of the domain of the given polynomial function. Let ǫ > 0.
Set δ = min{1, ǫ/(1 + |2a − 2|)}. Let x satisfy 0 < |x − a| < δ. Then

|(x2 − 2x) − (a2 − 2a)| = |(x2 − a2 ) − (2x − 2a)| (by algebra)

= |(x + a)(x − a) − 2(x − a)| (by algebra)
= |(x + a − 2)(x − a)| (by algebra)
= |x + a − 2| |x − a|
= |(x − a) + 2a − 2| |x − a| (by adding a clever 0)
≤ (|x − a| + |2a − 2|) |x − a| (by the triangle inequality)
≤ (1 + |2a − 2|) |x − a| (since |x − a| < δ ≤ 1)
< (1 + |2a − 2|)δ
≤ (1 + |2a − 2|)ǫ/(1 + |2a − 2|)
= ǫ.

Remark 4.1.7 Note that δ depends on ǫ and a, which are constants in the problem; δ is
not allowed to depend on x, as the definition goes:
“for all ǫ > 0 there exists δ > 0 such that for all x, etc”
so that x depends on δ, but δ does not depend on x.
By the definition of limits, δ is supposed to be a positive real number, not a function of x.
(See also Exercise 4.1.1.)
124 Chapter 4: Limits of functions

Remark 4.1.8 In all cases of rational functions, such as in examples above, Goal #1 is
to factor x − a from f (x) − L: if x − a is not a factor, check your limit or algebra for any
√
mistakes. In the next example, x − a is not a factor, but x − a is.
√
Example 4.1.9 lim 2x − 6 = 0.
x→3

Proof. The domain here is all x ≥ 3. So 3 is a limit point of the domain. Let ǫ > 0. Set
δ = ǫ2 /2. Let x > 3 satisfy 0 < |x − 3| < δ. Then
√ √ √
2x − 6 = 2 · x − 3
√ √ √
< 2 · δ (because is an increasing function)
√ p
= 2 · ǫ2 /2
= ǫ.

Often books consider the last example as a case of a one-sided limit (see definition
below) since we can only take the x from one side of 3. Our definition handles both-sided
and one-sided and all sorts of other limits with one simple notation, but we do have a use
for one-sided limits as well, so we define them next.

Definition 4.1.10 Let A ⊆ R, a ∈ R, L ∈ C, and f : A → C a function. Suppose that

a is a limit point of {x ∈ A : x > a} (resp. of {x ∈ A : x < a}). We say that the right-
sided (resp. left-sided) limit of f (x) as x approaches a is L if for every real number
ǫ > 0 there exists a real number δ > 0 such that for all x ∈ A, if 0 < x − a < δ (resp. if
0 < a − x < δ) then |f (x) − L| < ǫ. When this is the case, we write limx→a+ f (x) = L
(resp. limx→a− f (x) = L).
√
With this, Example 4.1.9 can be phrased as limx→3+ 2x − 6 = 0, and the proof goes
as follows: The domain A consists of all x ≥ 3, and 3 is a limit point of A∩{x : x > 3} = A.
Let ǫ > 0. Set δ = ǫ2 /2. Let x satisfy 0 < x − 3 < δ. Then
√ √ √
2x − 6 = 2 · x − 3
√ √
< 2· δ
√ p
= 2 · ǫ2 /2
= ǫ.
√
Thus, the two proofs are almost identical. Note that limx→3− 2x − 6 does not exist
because 3 is not the limit point of A ∩ {x ∈ R : x < 3} = ∅.
One-sided limits can also be used in contexts where limx→a f (x) does not exist. Below
is one example.
 Section 4.1: Limit of a function 125

x2 + 4, if x > 1;
Example 4.1.11 Let f : R → R be given by f (x) = Then
x − 2, if x ≤ 1.
limx→1+ f (x) = 5, limx→1− f (x) = −1.

Proof. Let ǫ > 0.

Set δ = min{1, ǫ/3}. Let x satisfy 0 < x − 1 < δ. Then
|f (x) − 5| = |x2 + 4 − 5| (since x > 1)
= |x2 − 1|
= |(x + 1)(x − 1)|
< |x + 1|δ (since x > 1, so x + 1 is positive)
= |x − 1 + 2|δ (by adding a clever 0)
≤ (|x − 1| + 2)δ (by the triangle inequality)
< (1 + 2)δ (since 0 < x − 1 < δ ≤ 1)
≤ 3ǫ/3
= ǫ.
This proves that limx→1+ f (x) = 5.
Set δ = ǫ. Let x satisfy 0 < 1 − x < δ. Then
|f (x) − (−1)| = |x − 2 + 1| (since x < 1)
= |x − 1|
<δ
= ǫ.

This proves that limx→1− f (x) = −3.

Exercises for Section 4.1

4.1.1 Below is an attempt at a “proof” that limx→3 x2 = 9. Explain how the two starred
steps below contain three mistakes.
Let ǫ > 0. Set δ =∗ ǫ/|x + 3|. Then
|x2 − 9| = |x − 3| · |x + 3|
<∗ δ|x + 3|
= ǫ.
126 Chapter 4: Limits of functions

4.1.2 Fill in the blanks of the following proof that limx→2 (x2 − 3x) = −2. Explain why
none of the inequalities can be changed into equalities.
so
that 2 is a . Let ǫ > 0. Set δ = . Let x satisfy
0 < |x − 2| < δ. Then
 
|(x2 − 3x) − (−2)| = x2 − 3x + 2
= |x − 1| |x − 2| (because )
= |x − 2 + 1| |x − 2| (by )
 
≤ |x − 2| + 1 |x − 2| (by )
< (3 + 1) |x − 2| (because )
< 4δ (because )
ǫ
≤4 (because )
4
= ǫ.

i) If the domain of this function is R as opposed to C, then |x − 2| < 4 can be shown

also with the following proof: Since |x − 2| < δ ≤ 3, then −3 < x − 2 < 3, so that
−2 < x − 1 < 4, which means that |x − 1| < 4. Say why this argument does not
work if the domain is C.
4.1.3 Determine the following limits and prove them with the epsilon-delta proofs.
i) lim (x3 − 4).
x→1
1
ii) lim .
x→2 x
iii) lim xx−4
2 +2 .
x→3 √
iv) lim x + 5.
x→4
x2 −9
v) lim .
x→3 x−3
4.1.4 Rework Example 4.1.3 with choosing δ to be at most 2 rather than at most 1.
4.1.5 Let b ∈ C and f, g : C → C with
 
x3 − 4x2 , if x 6= 5; x3 − 4x2 , if x = 5;
f (x) = g(x) =
b, if x = 5, b, 6 5.
if x =
Prove that the limit of f (x) as x approaches 5 is independent of b, but that the limit of
g(x) as x approaches 5 depends on b.
4.1.6 Suppose that a is a limit point of {x ∈ A : x > a} and of {x ∈ A : x < a}. Prove
that limx→a f (x) = L if and only if limx→a+ f (x) = L and limx→a− f (x) = L.
4.1.7 Suppose that a is a limit point of {x ∈ A : x > a} but not of {x ∈ A : x < a}. Prove
that limx→a f (x) = L if and only if limx→a+ f (x) = L.
 Section 4.2: When something is not a limit 127

4.1.8 Prove that limx→a (mx + l) = ma + l, where m and l are constants.

x
4.1.9 Let f : R → R be given by f (x) = |x|
. Prove that limx→0+ f (x) = 1 and that
limx→0− f (x) = −1.
4.1.10 Find a function f : R → R such that limx→0− f (x) = 2 and limx→0+ f (x) = −5.
4.1.11 Find a function f : R → R such that limx→0− f (x) = 2, limx→0+ f (x) = −5,
limx→1− f (x) = 3, and limx→1+ f (x) = 0. (Try to define such a function with fewest
possible words or symbols, but do use full grammatical sentences.)

4.2 When something is not a limit

Recall that limx→a f (x) = L means that a is a limit point of the domain of f , and that
for all real numbers ǫ > 0 there exists a real number δ > 0 such that for all x in the domain
of f , if 0 < |x − a| < δ then |f (x) − L| < ǫ. [Think of limx→a f (x) = L as statement
P , a being a limit point of the domain as statement Q, and the epsilon-delta
part as statement R. By definition, P is logically the same as the statement
Q and R.]
Thus if limx→a f (x) 6= L, then either a is not a limit point of the domain of f or else
it is not true that for all real numbers ǫ > 0 there exists a real number δ > 0 such that for
all x in the domain of f , if 0 < |x − a| < δ then |f (x) − L| < ǫ. [This simply says that
not P is the same as ( not Q) or ( not R).]
In particular,
lim f (x) 6= L and a is a limit point of the domain of f
x→a

means that it is not true that for all real numbers ǫ > 0 there exists a real number δ > 0
such that for all x in the domain of f , if 0 < |x − a| < δ then |f (x) − L| < ǫ. [This says
that ( not P ) and Q is the same as not R. You may want to write truth tables
for yourself.]
Negations of compound sentences, such as in the previous paragraph, are typically
hard to process and to work with in proofs. But by the usual negation rules of compound
statements (see chart on page 32), we successively rewrite this last negation into a form
that is easier to handle:

not For all real numbers ǫ > 0 there exists a real number δ > 0 such that


for all x in the domain of f , if 0 < |x−a| < δ then |f (x)−L| < ǫ.
[Negation of “For all z of some kind, property P holds” is “There is some z
of that kind for which P is false.” Hence the following rephrasing:]

= There exists a real number ǫ > 0 such that not there exists a real
128 Chapter 4: Limits of functions

number δ > 0 such that for all x in the domain of f , if 0 <

|x − a| < δ then |f (x) − L| < ǫ.
[Negation of “There exists z of some kind such that property P holds” is
“For all z of that kind, P is false.” Hence the following rephrasing:]
= There exists a real number ǫ > 0 such that for all real numbers
δ > 0, not for all x in the domain of f , if 0 < |x − a| < δ


then |f (x) − L| < ǫ.
[Negation of “For all z of some kind, property P holds” is “There is some z
of that kind for which P is false.” Hence the following rephrasing:]
= There exists a real number ǫ > 0 such that for all real numbers δ > 0,
there exists x in the domain of f such that not if 0 < |x − a| < δ


then |f (x) − L| < ǫ.
[Negation of “If P then Q” is “P and not Q.” Hence the following rephras-
ing:]
= There exists a real number ǫ > 0 such that for all real numbers δ > 0,
there exists x in the domain of f such that 0 < |x − a| < δ and


not |f (x) − L| < ǫ.
= There exists a real number ǫ > 0 such that for all real numbers δ > 0,
there exists x in the domain of f such that 0 < |x − a| < δ and
|f (x) − L| ≥ ǫ.
In summary, we just proved that:

Theorem 4.2.1 If a is a limit point of the domain of f , then limx→a f (x) 6= L means that
there exists a real number ǫ > 0 such that for all real numbers δ > 0, there exists x in the
domain of f such that 0 < |x − a| < δ and |f (x) − L| ≥ ǫ.
x
Example 4.2.2 The limit of |x| as x approaches 0 does not exist. In other words, for all
x
complex numbers L, limx→0 |x| 6= L.
x
The domain of the function that takes x to |x| is the set of all non-zero complex
x
numbers. For each non-zero x, |x| is a complex number of length 1 and with the same
angle as x. Thus the image of this function is the unit circle in C. Note that it is possible
to take two non-zero x very close to 0 but at different angles so that their images on the
unit circle are far apart. This is a geometric reasoning why the limit cannot exist. Next
we give an epsilon-delta proof.
x
Proof. The domain of the function that takes x to |x| is the set of all non-zero complex
numbers, so that 0 is a limit point of the domain. [Thus if the limit is not L, then
it must be the epsilon-delta condition that fails.] Set ǫ = 1. Let δ > 0 be an
 Section 4.2: When something is not a limit 129

arbitrary positive number. Let x = −δ/2 if Re(L) ≥ 0, and let x = δ/2 otherwise. Then
0 < |x| = |x − 0| < δ. If Re(L) ≥ 0, then
   
x −δ/2
Re − L = Re − L = −1 − Re(L) ≤ −1,
|x| | − δ/2|
x
so that | |x| − L| ≥ 1 = ǫ, and if Re(L) < 0, then
   
x δ/2
Re − L = Re − L = 1 − Re(L) > 1,
|x| |δ/2|
x
so that again | |x| − L| > 1 = ǫ. This proves the claim of the example.
i
Example 4.2.3 For all L ∈ C, limx→2 x−2 6= L.
A geometric reason for the non-existence of this limit is that as x gets closer to 2 (but
i
not equal to 2), the size of x−2 gets larger and larger.

Proof. Set ǫ = 1. Let δ > 0 be an arbitrary positive number. Set δ ′ = min{δ, 1/(|L| + 1)}.
Let x = 2 + δ ′ /2. Then 0 < |x − 2| < δ ′ ≤ δ, and
   
 i   2i 
   
 x − 2 − L =  δ ′ − L
 
 2i 
≥  ′  − |L| (by the reverse triangle inequality)
δ
≥ 2(|L| + 1) − |L| (since δ ′ ≤ 1/(|L| + 1))
≥1
= ǫ.

Example 4.2.4 For f : R → R given by the graph below, lim f (x) does not exist because
x→2
of the jump in the function at 2.

1 2 3

Here is an epsilon-delta proof. Say that the limit exists. Call it L. Set ǫ = 41 . Let δ

be an arbitrary positive number. If L ≥ 32 , set x = 2 + min 14 , 2δ , and if L < 32 , set
130 Chapter 4: Limits of functions

x = 2 − min{ 14 , 2δ }. In either case,

 
1 δ δ
0 < |x − 2| = min , ≤ < δ.
4 2 2
If L ≥ 23 , by our choice x = 2 + min{ 41 , δ2 }, so that f (x) = 1 + min{ 41 , δ2 } ≤ 1 + 41 ,
whence |f (x) − L| ≥ 41 = ǫ. Similarly, if L < 23 , by our choice x = 2 − min{ 41 , δ2 }, so that
f (x) = 2 − min{ 41 , 2δ } ≥ 2 − 14 = 1 + 34 , whence |f (x) − L| ≥ 41 = ǫ. Thus no L works, so
the limit of f (x) as x approaches 2 does not exist.

Exercises for Section 4.2

4.2.1 Prove that limx→3 (3x − 4) 6= −3.
4.2.2 Prove that limx→1 (x2 + 4) 6= −5.
x−3
4.2.3 Prove that limx→−3 x2 −9 does not exist.
4.2.4 Prove
 that for all a ∈ R, limx→a f (x) does not exist, where f : R → R is defined by
1, if x is rational;
f (x) =
0, if x is not rational.
√ √
4.2.5 Prove that limx→0 ( x − −x) does not exist. (Hint: The reason is different from
the reasons in all other examples.)

4.3 Limit theorems

While epsilon-delta proofs are a reliable method for proving limits, they do not help
in deciding what a limit may be. In this section we prove theorems that will efficiently
establish the limits for many functions. The proofs of these theorems require the epsilon-
delta machinery – as this is the definition of limits, but subsequent applications of these
theorems allow us to omit the time-consuming epsilon-delta proofs.

Theorem 4.3.1 If a limit exists, it is unique.

Proof. Suppose that both L1 and L2 are limits of f (x) as x approaches a.

First I give a FALSE proof: by assumption, L1 = limx→a f (x) = L2 , so that by
transitivity of equality, L1 = L2 . ×
What makes this proof false is that the equal sign in “limx→a f (x) = L” signifies, at this
point, not numerical equality, but that a is a limit point of the domain of f and that for
every ǫ > 0 there exists δ > 0 such that for all x in the domain of f , if 0 < |x − a| < δ then
|f (x) − L| < ǫ. Thus, we need a proof that will use these definitions instead.
 Section 4.3: Limit theorems 131

Real proof: Suppose that L1 and L2 are both limits. Let ǫ be an arbitrary positive
number. Then ǫ/2 is also positive, and by the definition of limits, for each i = 1, 2, there
exists δi > 0 such that for all x in the domain of f , if 0 < |x−a| < δi , then |f (x)−Li| < ǫ/2.
Set δ = min{δ1 , δ2 }. Then δ is a positive real number. Let x in the domain of f satisfy
0 < |x − a| < δ. Since a is a limit point of the domain, such x exists. Then
|L1 − L2 | = |L1 − f (x) + f (x) − L2 | (by adding a clever 0)
≤ |L1 − f (x)| + |f (x) − L2 | (by the triangle inequality)
< 2ǫ/2 (since δ ≤ δ1 , δ2 )
= ǫ,

which says that |L1 − L2 | < ǫ. Since ǫ was arbitrary, an application of Theorem 2.10.4
gives that |L1 − L2 | = 0, so that L1 − L2 = 0, and hence that L1 = L2 .
(If in the definition of limx→a f (x) we did not require that a be a limit point of the
domain of f , then x as in the proof above need not exist and we would not be able to
guarantee that limits are unique; in fact, any complex number L would be a limit.)

Theorem 4.3.2 Let a be a limit point of the domain of a function f with limx→a f (x) = L.
Suppose that L 6= 0. Then there exists δ > 0 such that for all x in the domain of f , if
0 < |x − a| < δ, then |f (x)| > |L|/2.
In particular, there exists δ > 0 such that for all x in the domain of f , if 0 < |x−a| < δ,
then f (x) 6= 0.

Proof. Set |L|/2 > 0, there exists δ > 0 such that for all x in the domain of f , if
0 < |x − a| < δ, then |f (x) − L| < |L|/2. Hence by the reverse triangle inequality, for the
same x, |L|/2 > |f (x) − L| ≥ |L| − |f (x)| = |L| − |f (x)|, which proves that |f (x)| > |L|/2.
In particular f (x) 6= 0.
The following theorem is very important, so study it carefully.

Theorem 4.3.3 Let A be the domain of f and g, and let a be a limit point of A, and let
c ∈ C. Suppose that limx→a f (x) and limx→a g(x) both exist. Then
(1) (Constant rule) lim c = c.
x→a
(2) (Linear rule) lim x = a.
x→a
(3) (Scalar rule) lim cf (x) = c lim f (x).
x→a x→a
(4) (Sum/difference rule) lim (f (x) ± g(x)) = lim f (x) ± lim g(x).
x→a x→a x→a
(5) (Product rule) lim (f (x) · g(x)) = lim f (x) · lim g(x).
x→a x→a x→a
f (x) lim f (x)
(6) (Quotient rule) If lim g(x) 6= 0, then lim = x→a .
x→a x→a g(x) lim g(x)
x→a
132 Chapter 4: Limits of functions

Proof. Both (1) and (2) were proved in Exercise 4.1.8.

Set L = limx→a f (x) and K = limx→a g(x).
(3) Let ǫ > 0. Then ǫ/(|c| + 1) is a positive number. (Note that we did not divide
by 0.) Since L = limx→a f (x), there exists δ > 0 such that for all x ∈ A, 0 < |x − a| < δ
implies that |f (x) − L| < ǫ/(|c| + 1). Hence for the same x,
|cf (x) − cL| = |c| · |f (x) − L| < |c| · ǫ/(|c| + 1) < ǫ,

which proves that limx→a cf (x) = cL = c limx→a f (x).

(4) Let ǫ > 0. Since L = limx→a f (x), there exists δ1 > 0 such that for all x ∈ A, if
0 < |x − a| < δ1 , then |f (x) − L| < ǫ/2. Similarly, since K = limx→a g(x), there exists
δ2 > 0 such that for all x ∈ A, if 0 < |x−a| < δ2 , then |g(x)−L| < ǫ/2. Set δ = min{δ1 , δ2 }.
Then δ is a positive number. Let x ∈ A satisfy 0 < |x − a| < δ. Then
|(f (x) ± g(x)) − (L ± K)| = |(f (x) − L) ± (g(x) − K)| (by algebra)
≤ |f (x) − L| + |g(x) − K| (by the triangle inequality)
< ǫ/2 + ǫ/2 (because 0 < |x − a| < δ ≤ δ1 , δ2 )
= ǫ,
which proves (4).
(5) Let ǫ > 0. Then min{ǫ/(2|L| + 2), 1}, min{ǫ/(2|K| + 1), 1} are positive numbers.
Since L = limx→a f (x), there exists δ1 > 0 such that for all x ∈ A, if 0 < |x − a| < δ1 ,
then |f (x) − L| < min{ǫ/(2|K| + 1), 1}. Similarly, since K = limx→a g(x), there exists
δ2 > 0 such that for all x ∈ A, if 0 < |x − a| < δ2 , then |g(x) − L| < ǫ/(2|L| + 2). Set
δ = min{δ1 , δ2 }. Then δ is a positive number. Let x ∈ A satisfy 0 < |x − a| < δ. Then by
the triangle inequality,

|f (x)| = |f (x) − L + L| ≤ |f (x) − L| + |L| < 1 + |L|,

and so

|f (x) · g(x) − L · K| = |f (x) · g(x) − f (x)K + f (x)K − L · K| (by adding a clever zero)
≤ |f (x) · g(x) − f (x)K| + |f (x)K − L · K| (by the triangle inequality)
= |f (x)| · |g(x) − K| + |f (x) − L| · |K| (by factoring)
ǫ ǫ
< (1 + |L|) · + |K| (since δ ≤ δ1 , δ2 )
2|L| + 2 2|K| + 1
< ǫ/2 + ǫ/2
= ǫ.

(6) Let ǫ > 0. Since K 6= 0, by Theorem 4.3.2, there exists δ0 > 0 such that for all
x ∈ A, if 0 < |x − a| < δ0 , then |g(x)| > |K|/2.
 Section 4.3: Limit theorems 133

The numbers |K|ǫ/4, |K|2 ǫ/(4|L| + 1) are positive numbers. Since L = limx→a f (x),
there exists δ1 > 0 such that for all x ∈ A, if 0 < |x − a| < δ1 , then |f (x) − L| < |K|ǫ/4.
Similarly, since K = limx→a g(x), there exists δ2 > 0 such that for all x ∈ A, if 0 < |x−a| <
δ2 , then |g(x) −L| < |K|2 ǫ/(4|L| +1). Set δ = min{δ0 , δ1 , δ2 }. Then δ is a positive number.
Let x ∈ A satisfy 0 < |x − a| < δ. Then
   
 f (x) L   f (x)K − Lg(x) 
    (by algebra)
 g(x) − K  =  Kg(x) 
 
 f (x)K − LK + LK − Lg(x) 
=  (by adding a clever zero)
Kg(x) 
   
 f (x)K − LK   LK − Lg(x) 
≤  + 
  Kg(x)  (by the triangle inequality)
Kg(x)
   
 f (x) − L  |L|  K − g(x) 
=  +   (by factoring)
g(x)  |K|  g(x) 
|K|ǫ 2 |L| |K|2 ǫ 2
< · + · (since δ ≤ δ0 , δ1 , δ2 )
4 |K| |K| 4|L| + 1 |K|
< ǫ/2 + ǫ/2
= ǫ.
This proves (6) and thus the theorem.

Theorem 4.3.4 (Power rule for limits) Let n be a positive integer. If limx→a f (x) = L,
then limx→a f (x)n = Ln .

Proof. The case n = 1 is the assumption. Suppose that we know the result for n − 1. Then


lim f (x)n = lim f (x)n−1 · f (x) (by algebra)
x→a x→a


= lim f (x)n−1 · lim (f (x)) (by the product rule)
x→a x→a
n−1
=L · L (by induction assumption)
= Ln .

So the result holds for n, and we are done by mathematical induction.

Theorem 4.3.5 (Polynomial function rule for limits) Let f be a polynomial function.
Then for all complex (or real) a, limx→a f (x) = f (a).

Proof. Because f is a polynomial function, it can be written as

f (x) = c0 + c1 x + c2 x2 + c3 x3 + · · · + cn xn
for some non-negative integer n and some constants c0 , c1 , . . . , cn . By the linear rule,
limx→a x = a. Hence by the power rule, for all i = 1, . . . , n, limx→a xi = ai . By the
134 Chapter 4: Limits of functions

constant rule, limx→a ci = ci , so that by the product rule limx→a ci xi = ci ai . Hence by

repeating the sum rule,


lim f (x) = lim c0 + c1 x + c2 x2 + · · · + cn xn = c0 + c1 a + c2 a2 + · · · + cn an = f (a).
x→a x→a

Theorem 4.3.6 (Rational function rule for limits) Let f be a rational function. Then
for all complex (or real) a in the domain of f , limx→a f (x) = f (a).

Proof. Let a be in the domain of f . Write f (x) = g(x)/h(x) for some polynomial functions
g, h such that h(a) 6= 0. By Theorem 2.4.13, the domain of f is the set of all except finitely
many numbers, so that in particular a is a limit point of the domain. By the polynomial
function rule for limits, limx→a g(x) = g(a) and limx→a h(x) = h(a) 6= 0. Thus by the
quotient rule, limx→a f (x) = g(a)/h(a) = f (a).

Theorem 4.3.7 (Absolute value rule for limits) For all a ∈ C, limx→a |x| = |a|.

Proof. This function is defined for all complex numbers, and so every a ∈ C is a limit point
of the domain. Let ǫ > 0. Set δ = ǫ. Then for all x ∈ C, by the reverse triangle inequality,
 
|x| − |a| ≤ |x − a| < δ = ǫ.

Theorem 4.3.8 (Real and imaginary parts of limits) Let f : A → C be a func-

tion, let a be a limit point of A, and let L ∈ C. Then limx→a f (x) = L if and only if
limx→a Re f (x) = Re L and limx→a Im f (x) = Im L.

Proof. First suppose that limx→a f (x) = L. Let ǫ > 0. By assumption there exists δ > 0
such that for all x ∈ A, 0 < |x − a| < δ implies that |f (x) − L| < ǫ. Then for the same x,
| Re f (x) − Re L| = | Re(f (x) − L)| ≤ |f (x) − L| < ǫ,

and similarly | Im f (x) − Im L| < ǫ which proves that limx→a Re f (x) = Re L and
limx→a Im f (x) = Im L.
Now suppose that limx→a Re f (x) = Re L and limx→a Im f (x) = Im L. By the scalar
and sum rules in Theorem 4.3.3, and by the definition of real and imaginary parts,

lim f (x) = lim (Re f (x) + i Im f (x)) = lim Re f (x) + i lim Im f (x) = Re L + i Im L = L.
x→a x→a x→a x→a
 Section 4.3: Limit theorems 135

Theorem 4.3.9 (The composite function theorem) Let h be the composition of

functions h = g ◦f . Suppose that a is a limit point of the domain of g ◦f , that lim f (x) = L
x→a
and that lim g(x) = g(L). Then lim h(x) = lim (g ◦ f )(x) = g(L).
x→L x→a x→a

Proof. Let ǫ > 0. Since lim g(x) = g(L), there exists δ1 > 0 such that for all x in the
x→L
domain of g, if 0 < |x−a| < δ1 then |g(x)−g(L)| < ǫ. Since lim f (x) = L, there exists δ > 0
x→a
such that for all x in the domain of f , if 0 < |x − a| < δ then |f (x) − L| < δ1 . Thus for the
same δ, if x is in the domain of h and 0 < |x−a| < δ, then |h(x)−g(L)| = |g(f (x))−g(L)| < ǫ
because |f (x) − L| < δ1 .
Perhaps the hypotheses on g in the theorem above seem overly restrictive, and you
think that the limit of g(x) as x approaches L need not be g(L) but an arbitrary K?
Consider the following example which shows that limx→a g(f (x)) then need not be K. Let
3, if x 6= 5;
f (x) = 5 and g(x) = . Then
7, otherwise.
lim f (x) = 5, lim g(x) = 3, and lim g(f (x)) = 7.
x→a x→5 x→a

Theorem 4.3.10 Suppose that f, g : A → R, that a is a limit point of A, that limx→a f (x)
and limx→a g(x) both exist, and that for all x ∈ A, f (x) ≤ g(x). Then
lim f (x) ≤ lim g(x).
x→a x→a

Proof. Let L = lim f (x), K = lim g(x). Let ǫ > 0. By assumptions there exists δ > 0
x→a x→a
such that for all x ∈ A, if 0 < |x − a| < δ, then |f (x) − L|, |g(x) − K| < ǫ/2. Then for the
same x,

K − L = K − g(x) + g(x) − f (x) + f (x) − L ≥ K − g(x) + f (x) − L

≥ −|K − g(x) + f (x) − L|
≥ −|K − g(x)| − |f (x) − L| (by the triangle inequality)
> −ǫ/2 − ǫ/2 = −ǫ.
Since this is true for all ǫ > 0, by Theorem 2.10.4, K ≥ L, as desired.

Theorem 4.3.11 (The squeeze theorem) Suppose that f, g, h : A → R, that a is a

limit point of A, and that for all x ∈ A \ {a}, f (x) ≤ g(x) ≤ h(x). If limx→a f (x) and
limx→a h(x) both exist and are equal, then limx→a g(x) exists as well and

lim f (x) = lim g(x) = lim h(x).

x→a x→a x→a

Proof. [If we knew that limx→a g(x) existed, then by the previous theorem,
limx→a f (x) ≤ limx→a g(x) ≤ limx→a h(x) = limx→a f (x) would give that the three
limits are equal. But we have yet to prove that limx→a g(x) exists.]
136 Chapter 4: Limits of functions

Let L = limx→a f (x) = limx→a h(x). Let ǫ > 0. Since limx→a f (x) = L, there
exists δ1 > 0 such that for all x, if 0 < |x − a| < δ1 then |f (x) − L| < ǫ. Similarly,
since limx→a h(x) = L, there exists δ2 > 0 such that for all x, if 0 < |x − a| < δ2 then
|h(x) − L| < ǫ. Now set δ = min{δ1 , δ2 }. Let x satisfy 0 < |x − a| < δ. Then
−ǫ < f (x) − L ≤ g(x) − L ≤ h(x) − L < ǫ,

where the first inequality holds because δ ≤ δ1 , and the last inequality holds because δ ≤ δ2 .
Hence −ǫ < g(x) − L < ǫ, which says that |g(x) − L| < ǫ, so that limx→a g(x) = L.

Exercises for Section 4.3

4.3.1 Determine the following limits by invoking appropriate results:
i) lim (x3 − 4x − 27), lim (x2 + 5).
x→2 x→2
3 3
(x − 4x − 27)
ii) lim .
x→2 (x2 + 5)2
|x3 − 4x − 27|
iii) lim .
x→2 (x2 + 5)3
4.3.2 If lim f (x) = L and lim g(x) = K, prove that lim (3f (x)2 − 4g(x)) = 3L2 − 4K.
x→a x→a x→a
4.3.3 Prove that lim xn = an for any non-zero complex number a and any integer n.
x→a
4.3.4 Prove that for any positive real number a and any whole number n, lim xn = an .
x→a
4.3.5 Prove that if limx→a f (x) = L, then limx→a |f (x)| = |L|. Give an example of a
function such that limx→a |f (x)| = |L| and limx→a f (x) does not exist.
4.3.6 Find functions f, g such that limx→a (f (x) + g(x)) exists but limx→a f (x) and
limx→a g(x) do not exist. Does this contradict the sum rule?
4.3.7 Find functions f, g such that limx→a (f (x)g(x)) exists but limx→a f (x) and
limx→a g(x) do not exist. Does this contradict the product rule?
x x

x x



4.3.8 Prove that lim |x| and lim 1 − |x| do not exist, but that lim |x| + 1 − |x|
x→0 x→0 x→0
does exist. Does this contradict the sum rule for limits? Justify.
|x|

4.3.9 Prove that lim x
and lim do not exist, but that lim x
· |x| does exist. Does
x→0 |x| x→0 x x→0 |x| x
this contradict the product rule for limits? Justify.
4.3.10 Let A ⊆ B ⊆ C, let a be a limit point of A, and let f : A → C, g : B → C. Suppose
that for all x ∈ A\ {x}, g(x) = f (x). Prove that if limx→a g(x) = L, then limx→a f (x) = L.
In particular, if a ∈ R is a limit point of A = B ∩ R and if limx→a g(x) = L, then the
restriction of g to A has the same limit point at a.
 Section 4.3: Limit theorems 137

4.3.11 The following information is known about functions f and g:

a f (a) lim f (x) g(a) lim g(x)

x→a x→a
0 1 2 6 4
1 −1 0 5 5
2 not defined 3 6 −6
3 4 4 not defined 5
4 5 1 4 7

For each part below provide the limit if there is enough information, and justify. (Careful,
the answers to the last two parts are different.)
i) lim (f (x) − g(x)).
x→0
ii) lim (f (x) + g(x)).
x→4
iii) lim (f (x) · g(x)).
x→3
f (x)
iv) lim .
x→2 g(x)
v) lim g(x) .
x→4 f (x)
vi) lim (g ◦ f )(x).
x→1
vii) lim (g ◦ f )(x).
x→4
4.3.12 Let A ⊆ C, let f, g : A → C, and let a be a limit point of A. Suppose that for all
x ∈ A \ {a}, f (x) = g(x). Prove that limx→a f (x) exists if and only if limx→a g(x) exists,
and if they both exist, then the two limits are equal.
4.3.13 Let f, g : R → R be defined by
 
x, if x ∈ Q; 1, if x 6= 0;
f (x) = g(x) =
0, otherwise, 0, otherwise.
i) Prove that lim f (x) = 0.
x→0
ii) Prove that lim g(x) = 1.
x→0
iii) Prove that lim (g ◦ f )(x) does not exist.
x→0
iv) Comment on Theorem 4.3.9.
*4.3.14 Let a, L, M be arbitrary real numbers. Construct functions F, G : R → R such
that limx→a F (x) = L, limx→L G(x) = M , but limx→a (G ◦ F )(x) 6= M . Comment on
Theorem 4.3.9. (Hint: modify f and g from the previous exercise.)
138 Chapter 4: Limits of functions

4.4 Infinite limits (for real-valued functions)

When the codomain of a function is a subset of an ordered field such as R, the values
of a function may grow larger and larger with no upper bound, or more and more negative
with no lower bound. In that case we may want to declare limit to be ∞ or −∞. Naturally
both the definition and how we operate with infinite limits requires different handling.

Definition 4.4.1 Let A ⊆ C, f : A → R a function, and a ∈ C. Suppose that a is a limit

point of A (and not necessarily in A).
We say that the limit of f (x) as x approaches a is ∞ if for every real number
M > 0 there exists a real number δ > 0 such that for all x ∈ A, if 0 < |x − a| < δ then
f (x) > M . We write this as lim f (x) = limx→a f (x) = ∞. Similarly we say that the
x→a
limit of f (x) as x approaches a is −∞ if for every real number M < 0 there exists a
real number δ > 0 such that for all x ∈ A, if 0 < |x − a| < δ then f (x) < M . We write this
as lim f (x) = limx→a f (x) = −∞.
x→a
The limit of f (x) as x approaches a from the right is −∞ if a is a limit point
of {x ∈ A : x > a}, and if for every real number M < 0 there exists a real number δ > 0
such that for all x ∈ A, if 0 < x − a < δ then f (x) < M . This is written as lim+ f (x)
x→a
= limx→a+ f (x) = −∞.
It is left to the reader to spell out the definitions of the following:

lim f (x) = ∞, lim− f (x) = ∞, lim− f (x) = −∞.

x→a+ x→a x→a

Note that we cannot use epsilon-delta proofs: no real numbers are within ǫ of infinity.
So instead we approximate infinity with huge numbers. In fact, infinity stands for that thing
which is larger than any real number. Thus for all M we can find x near a with f (x) > M
is simply saying that we can take f (x) arbitrarily large, which is more succinctly expressed
as saying that f (x) goes to ∞. (As far as many applications are concerned, a real number
larger than the number of atoms in the universe is as close to infinity as realistically possible,
but for proofs, the number of atoms in the universe is not large enough.)
1
Example 4.4.2 limx→0 |x|2 = ∞.

Proof. 0 is a limit point of the domain (in the field of real or complex numbers) of the
√
function that takes x to 1/|x|2 . Let M be a positive real number. Set δ = 1/ M . Let x
satisfy 0 < |x − 0| < δ, i.e., let x satisfy 0 < |x| < δ. Then
1 1
2
> 2 (because 0 < |x| < δ)
|x| δ
= M.
 Section 4.4: Infinite limits (for real-valued functions) 139

x+2
Example 4.4.3 limx→5+ x2 −25 = ∞.

Proof. Certainly 5 is the limit point of the domain of the given function. Let M > 0. Set
7
δ = min{1, 11M }. Let x satisfy 0 < x − 5 < δ. Then
x+2 5+2
2
> 2 (because x > 5 and x2 − 25 > 0)
x − 25 x − 25
7 1
= · (by algebra)
(x + 5) x − 5
7 1
> · (because 0 < x − 5 < δ ≤ 1, so 0 < x + 5 < 11, and x − 5 > 0)
11 x − 5
7 1
> · (because 0 < x − 5 < δ)
11 δ
7 1 7 7
≥ · 7 (because δ ≤ 11M , so 1/δ ≥ 1/( 11M ))
11 11M
= M.
1
Example 4.4.4 limx→0+ x = ∞.
1
Proof. Let M > 0. Set δ = 1/M . Then for all x with 0 < x − 0 < δ, x > 1/δ = M .
1
Example 4.4.5 limx→0− x = −∞.

Proof. Let M < 0. Set δ = −1/M . Then δ is a positive number. Then for all x with
0 < 0 − x < δ,
1
< −1/δ = M.
x
Example 4.4.6 We conclude that limx→0 x1 cannot be a real number, and it cannot be
either ∞ or −∞. Thus limx→0 x1 does not exist.
The following theorem is straightforward to prove, and so it is left for the exercises.

Theorem 4.4.7 Let f, g, h : A → R, and let a be a limit point of A. Suppose that

limx→a f (x) = L ∈ R, limx→a g(x) = ∞, limx→a h(x) = −∞. Then
(1) (Scalar rule) For any c ∈ R,
( (
∞, if c > 0; −∞, if c > 0;
lim (cg(x)) = −∞, if c < 0; lim (ch(x)) = ∞, if c < 0;
x→a x→a
0, if c = 0; 0, if c = 0.
(2) (Sum/difference rule)
lim (f (x) ± g(x)) = ±∞,
x→a
lim (f (x) ± h(x)) = ∓∞,
x→a
lim (g(x) − h(x)) = ∞,
x→a
lim (h(x) − g(x)) = −∞,
x→a
140 Chapter 4: Limits of functions

but we do not have enough information to determine limx→a (g(x) + h(x)) or even whether
it exists.
(3) (Product rule) 
∞, if L > 0;
lim (f (x) · g(x)) =
x→a −∞, if L < 0;

−∞, if L > 0;
lim (f (x) · h(x)) =
x→a ∞, if L < 0;
lim (g(x) · h(x)) = −∞.
x→a

We do not have enough information to determine the existence (or value) of limx→a (f (x) ·
g(x)) and of limx→a (f (x) · h(x)) in case L = 0.
(4) (Quotient rule)
f (x)
lim = 0,
x→a g(x)

f (x)
lim = 0,
x→a h(x)

g(x) ∞, if L > 0;
lim =
x→a f (x) −∞, if L < 0;

h(x) −∞, if L > 0;
lim =
x→a f (x) ∞, if L < 0.
limx→a (g(x)
We do not have enough information to determine the existence (or value) of f (x))
limx→a (h(x)
and of f (x)) in case L = 0. We also do not have enough information to determine
limx→a (g(x)
the existence (or value) of h(x)) .

Example 4.4.8 Define g, h1 , h2 , h3 , h4 : R \ {0} → R by g(x) = 1/x2 , h1 (x) = −1/x2 ,

h2 (x) = 17 − 1/x2 , h3 (x) = −1/x2 − 1/x4 , h4 (x) = −1/x2 − 1/x3 . We have seen that
limx→0 g(x) = ∞, and similarly that limx→0 h1 (x) = limx→0 h2 (x) = limx→0 h3 (x) =
limx→0 h4 (x) = −∞. However,
lim (g(x) + h1 (x)) = lim 0 = 0,
x→0 x→0
lim (g(x) + h2 (x)) = lim 17 = 17,
x→0 x→0
lim (g(x) + h3 (x)) = lim (−1/x4 ) = −∞,
x→0 x→0

but limx→0 (g(x) + h4 (x)) = limx→0 (−1/x3 ) does not exist. This justifies the “not enough
information” line in the sum/difference rule in Theorem 4.4.7.
Other “not enough information” lines are left for the exercises.
Distinguish between the scalar and the product rules: when limx→a f (x) = 0, if f is
a constant function, then limx→a (f (x)g(x)) = 0, but if f is not a constant function, then
we do not have enough information for limx→a (f (x)g(x)) = 0.
 Section 4.5: Limits at infinity 141

Exercises for Section 4.4

4.4.1 Give definitions for the following limits.
i) limx→a+ f (x) = ∞.
ii) limx→a− f (x) = ∞.
iii) limx→a− f (x) = −∞.
4.4.2 Prove that limx→0+ √1 = ∞.
x
1
4.4.3 Prove that limx→0− x3 = −∞.

4.4.4 This is about Theorem 4.4.7. Prove the scalar rule, the sum/difference rule, the
product rule, and the quotient rule.
4.4.5 Let g(x) = 1/x2 , f1 (x) = x3 , f2 (x) = x2 , f3 (x) = 17x2 , f4 (x) = x. It is easy to see
that limx→0 g(x) = ∞, limx→0 f1 (x) = limx→0 f2 (x) = limx→0 f3 (x) = limx→0 f4 (x) = 0.
i) Compute the limits limx→0 (f1 (x)g(x)), limx→0 (f2 (x)g(x)), limx→0 (f3 (x)g(x)),
limx→0 (f4 (x)g(x)).
ii) Justify the first “not enough information” in the product rule in Theorem 4.4.7.
iii) Justify the second “not enough information” in the product rule in Theorem 4.4.7.
4.4.6 Justify the three “not enough information” in the quotient rule in Theorem 4.4.7.
(Hint: Study the previous exercise.)

4.5 Limits at infinity

Definition 4.5.1 Let a ∈ R and L ∈ C.

For f : (a, ∞) → C, we say that limx→∞ = L if for every ǫ > 0 there exists a real
number N > 0 such that for all x > N , |f (x) − L| < ǫ.
For f : (−∞, a) → C, we say that limx→−∞ = L if for every ǫ > 0 there exists a real
number N < 0 such that for all x < N , |f (x) − L| < ǫ.
Below we switch to real-valued functions.
For f : (a, ∞) → R, we say that limx→∞ = ∞ if for every M > 0 there exists a real
number N > 0 such that for all x > N , f (x) > M . We say that limx→∞ = −∞ if for every
M < 0 there exists a real number N > 0 such that for all x > N , f (x) < M .
For f : (−∞, a) → R, we say that limx→−∞ = ∞ if for every M > 0 there exists a
real number N < 0 such that for all x < N , f (x) > M . We say that limx→−∞ = −∞ if
for every M < 0 there exists a real number N < 0 such that for all x < N , f (x) < M .
142 Chapter 4: Limits of functions

Example 4.5.2 limx→∞ (x5 − 16x4 ) = ∞.

Proof. Let M > 0. Set N = max{17, M 1/4}. Then for all x > N ,
x5 − 16x4 = x4 (x − 16)
≥ x4 (because x > N ≥ 17)
> N 4 (because x > N )
≥ (M 1/4 )4 (because N ≥ M 1/4 )
= M.
x5 −16x4
Example 4.5.3 limx→∞ x5 +4x2 = 1.

Proof. Let ǫ > 0. Set N = max{1, 20/ǫ}. Then for all x > N ,
 5   5 
 x − 16x4   x − 16x4 − x5 − 4x2 
   
 x5 + 4x2 − 1 =  x5 + 4x2 
 
 −16x4 − 4x2 
=  
x5 + 4x2 
 
 16x4 + 4x2 

= 5 
x + 4x2 
16x4 + 4x2
= 5 (because x > N ≥ 0)
x + 4x2
16x4 + 4x2
≤ (because x5 + 4x2 ≥ x5 > 0)
x5
16x4 + 4x4
≤ (because x > N ≥ 1 so that x2 < x4 )
x5
20x4
= 5
x
20
=
x
20
< (because x > N )
N
20
≤ (because N ≥ 20/ǫ)
20/ǫ
= ǫ.

Exercises for Section 4.5

4.5.1 Prove that limx→∞ (x2 − x) = ∞ and that limx→−∞ (x2 − x) = ∞.
4.5.2 Prove that limx→∞ (x3 − x) = ∞ and that limx→−∞ (x3 − x) = −∞.
 Section 4.5: Limits at infinity 143

4.5.3 Is there a limit rule of the following form: If limx→∞ f (x) = ∞ and limx→∞ g(x) =
∞, then limx→∞ (f (x) − g(x)) can be determined?
4.5.4 Prove the following limits:
2
i) limx→∞ 3x2x−4x+1
2 +2 = 3/2.
3x3 −4x+1
ii) limx→∞ 2x2 +2 = ∞.
3x2 −4x+1
iii) limx→∞ 2x3 +2 = 0.
√
2 +4
iv) limx→∞ 9xx+2 = 3.
√
2
v) limx→−∞ 9x x+2
+4
= −3.
4.5.5 For which rational functions f does limx→∞ f (x) = 0 hold? Justify.
4.5.6 For which rational functions f does limx→∞ f (x) exist and is not ±∞? Justify.
4.5.7 Let f : (a, ∞) → C and L ∈ C. Prove that limx→∞ f (x) = L if and only if
limx→∞ Re f (x) = Re L and limx→∞ Im f (x) = Im L.
4.5.8 Let f : (−∞, b) → C and L ∈ C. Prove that limx→−∞ f (x) = L if and only if
limx→−∞ Re f (x) = Re L and limx→−∞ Im f (x) = Im L.
Chapter 5: Continuity

Continuous functions from an interval in R to R are the ones that we can graph with-
out any holes or jumps, i.e., without lifting the pencil from the paper, so the range of such
functions is an interval in R as well. We make this more formal below, and not just for func-
tions with domains and codomains in R. The formal definition involves limits of functions.
All the hard work for that was already done in Chapter 4, so this chapter, after absorbing
the definition, is really straightforward. The big new results are the Intermediate value
theorem and the Extreme value theorem for real-valued functions (Section 5.3), existence
of radical functions (Section 5.4), and the new notion of uniform continuity (Section 5.5).

5.1 Continuous functions

Definition 5.1.1 A function f : A → B is continuous at a ∈ A if for all real numbers
ǫ > 0 there exists a real number δ > 0 such that for all x ∈ A, |x − a| < δ implies that
|f (x) − f (a)| < ǫ. We say that f is continuous if f is continuous at all points in its
domain.

Much of the time for us A is an interval in R, a rectangle in C, or B(a, r) in R or C,

et cetera. In the more general case, A may contain a point a that is not a limit point of A.

Theorem 5.1.2 Let f : A → B be a function and a ∈ A.

(1) If a is not a limit point of A, then f is continuous at a.
(2) If a is a limit point of A, then f is continuous at a if and only if limx→a f (x) = f (a).

Proof. In the first case, there exists δ > 0 such that B(a, δ) ∩ A = {a}. Thus by definition,
the only x ∈ A with |x − a| < δ is x = a, whence |f (x) − f (a)| = 0 is strictly smaller than
an arbitrary positive ǫ. Thus f : A → B is continuous at a.
Now assume that a is a limit point of A. Suppose that f is continuous at a. We need
to prove that limx→a f (x) = f (a). Let ǫ > 0. Since f is continuous at a, there exists δ > 0
such that for all x ∈ A, if |x − a| < δ, then |f (x) − f (a)| < ǫ. Hence for all 0 < |x − a| < δ,
|f (x) − f (a)| < ǫ, and since a is a limit point of A, this proves that limx→a f (x) = f (a).
Finally suppose that limx→a f (x) = f (a). We need to prove that f is continuous at a.
Let ǫ > 0. By assumption there exists δ > 0 such that for all x ∈ A, if 0 < |x − a| < δ then
|f (x) − f (a)| < ǫ. If x = a, then |f (x) − f (a)| = 0 < ǫ, so that for all x ∈ A, if |x − a| < δ,
then |f (x) − f (a)| < ǫ. Thus f is continuous at a.
 Section 5.1: Continuous functions 145

The next theorem is an easy application of Theorems 4.3.3, 4.3.4, 4.3.5, 4.3.6, 4.3.7,
4.3.8, and 4.3.9, at least when the points in question are limit points of the domain. When
the points in question are not limit points of the domain, the results below need somewhat
different but still easy proofs.

Theorem 5.1.3 We have:

(1) (Constant rule) Constant functions are continuous (at all real/complex numbers).
(2) (Linear rule) The function f (x) = x is continuous (at all real/complex numbers).
(3) (Absolute value rule) The function f (x) = |x| is continuous (at all real/complex
numbers).
(4) (Polynomial function rule) Polynomial functions are continuous (at all real/complex
numbers).
(5) (Rational function rule) Rational functions are continuous (at all points in the
domain).
(6) (Real and imaginary parts) f is continuous at a if and only if Re f and Im f are
continuous at a.
Suppose that f : A → C is continuous at a ∈ A. Then
(7) (Scalar rule) For any c ∈ C, cf is continuous at a.
(8) (Power rule) For any positive integer n, the function that takes x to (f (x))n is
continuous at a.
(9) (The composite function rule) If g is continuous at f (a), then g ◦ f is continuous
at a.
(10) (The composite power rule) If f is composable with itself i.e., if the range of f is
a subset of the domain of f , then f n = f ◦ f ◦ · · · ◦ f is continuous at a. (Example
of a function that is not composable with itself: f (x) = ln x.)
(11) (Restriction rule) For any subset B of A that contains a and in which a is a limit
point, the restriction of f to B is continuous at a. In other words, if g : B → C is
defined as g(x) = f (x) for all x ∈ B, then g is continuous at a. (Such g is called
the restriction of f to B. By this rule and by (1), the constant functions with
domain a subinterval in R are continuous.)
Now suppose that f, g : A → C, and that f and g are continuous at a ∈ A. Then
(12) (Sum/difference rule) f + g and f − g are continuous at a.
(13) (Product rule) f · g is continuous at a.
(14) (Quotient rule) If g(a) 6= 0, then f /g is continuous at a.

The theorem covers many continuous functions, but the following function for example
has to be verified differently.
146 Chapter 5: Continuity

Example 5.1.4 Let f : C → C be defined by


x2 − 4, if Re x > 1;
f (x) =
−3x3 , if Re x ≤ 1.
Then f is continuous at 1, because by the polynomial rules,

lim f (x) = lim (x2 − 4) = 12 − 4 = −3,

x→1,Re x>1 x→1,Re x>1

lim f (x) = lim −3x3 = −3 · 13 = −3,

x→1,Re x≤1 x→1,Re x≤1

so that limx→1 f (x) = −3 = −3 · 13 = f (1).

Note however that this function is not continuous at 1 + i:

lim f (x) = lim (x2 − 4) = (1 + i)2 − 4 = −4 + 2i,

x→1+i,Re x>1 x→1+i,Re x>1

lim f (x) = lim −3x3 = −3 · (1 + i)3 = 6 − 6i.

x→1+i,Re x≤1 x→1+i,Re x≤1

so that even limx→1+i f (x) does not exist.

Example 5.1.5 When we restrict the function in the previous examples to R, we get
g : R → R defined by 
x2 − 4, if x > 1;
g(x) =
−3x3 , if x ≤ 1.
As before, g is continuous at 1. Explicitly, by the polynomial rules,

lim+ g(x) = lim+ (x2 − 4) = 12 − 4 = −3,

x→1 x→1
lim− g(x) = lim− −3x3 = −3 · 13 = −3,
x→1 x→1
3
so that limx→1 g(x) = −3 = −3 · 1 = g(1). It is worth noting that precisely because of
continuity this function can be expressed in the following ways as well:
 
x2 − 4, if x ≥ 1; x2 − 4, if x ≥ 1;
g(x) = =
−3x3 , if x < 1 −3x3 , if x ≤ 1.
By the polynomial rules, this function is continuous at all other real numbers as well.

x + 1, if x < 0;
Example 5.1.6 Let f : (−1, 0) ∪ (0, 1] → R be defined by f (x) =
x − 1, if x > 0.
y

This function is continuous at all a < 0 because there f is the polynomial/linear function
 Section 5.1: Continuous functions 147

f (x) = x + 1, and f is function is continuous at all a > 0 because there f is the function
f (x) = x − 1. Thus f is continuous at all points in its domain, so that f is continuous.

Exercises for Section 5.1

5.1.1 Let c ∈ C.
i) Prove that the function f : C → C given by f (x) = c + x is continuous.
ii) Prove that the function g : C → C given by g(x) = cx is continuous.
5.1.2 Give an ǫ − δ proof to show that the functions Re, Im : C → R are continuous.
5.1.3 Give details of the proofs of Theorem 5.1.3.
5.1.4 The following information is known about functions f and g:
c f (c) lim f (x) g(c) lim g(x)
x→c x→c
0 1 2 3 4
1 −1 0 5 5
2 not defined −∞ 6 −6
3 4 4 not defined ∞
4 2 3 3 2
i) At which points is f continuous?
ii) At which points is |g| continuous?
iii) At which points is f g continuous?
5.1.5 Define a function f : C \ {0} → R such that for all non-zero x ∈ C, f (x) is the angle
of x counterclockwise from the positive real axis. Argue that f is not continuous.
5.1.6 Let f : A → C be a function, and let a ∈ A be a limit point of A. Let g : A → B
be a continuous invertible function such that its inverse is also continuous. Prove that
limx→a f (x) = L if and only if limz→g(a) f (g −1 (z)) = L.
5.1.7 Assume that the trigonometric functions sine and cosine are continuous everywhere.
(We will prove this in Theorem 9.9.4.) Define a function g : C \ {0} → R such that for all
non-zero x ∈ C, g(x) is the sine (respectively cosine) of the angle of x counterclockwise from
the positive real axis. Prove that f is continuous. Compare with the previous exercise.
5.1.8 Assume that the trigonometric function sine is continuous everywhere. (We will
prove this in Theorem 9.9.4.) Define a function f : R → R as

x sin(1/x), if x 6= 0;
f (x) =
0, if x = 0.
Prove that f is continuous at all points, including at 0. (You need a separate proof for
continuity at 0.)
148 Chapter 5: Continuity

5.1.9 Assume that the trigonometric function sine is continuous everywhere. (We will
prove this in Theorem 9.9.4.) Define a function f : R → R as

sin(1/x), if x 6= 0;
f (x) =
0, if x = 0.
i) Is f continuous at 0? Prove or disprove.
ii) Prove that f is continuous at all other points.

x, if x is rational;
5.1.10 Let f : R → R be given by f (x) =
0, otherwise.
i) Prove that f is continuous at 0.
ii) Prove that f is not continuous anywhere else.
*5.1.11 (The Thomae function, also called the popcorn function, the raindrop function,
and more) Let f : R → R be defined as
(
1
, if x = pq , where p ∈ Z and q ∈ N+ have only 1 as a common factor;
f (x) = q
0, if x is irrational.
i) Prove that f is not continuous at any rational number.
ii) Prove that f is continuous at all irrational numbers.

The exercises below modify the definition of continuity to get very different types of func-
tions. – The moral is that the order of quantifiers and implications is very important!
5.1.12 (This is from page 1177 of the Edward Nelson’s article “Internal set theory: a new
approach to nonstandard analysis.” Bull. Amer. Math. Soc. 83 (1977), no. 6, 1165–1198.)
A function f : A → B is suounitnoc at a ∈ A if for all real numbers ǫ > 0 there exists a
real number δ > 0 such that for all x ∈ A, |x − a| < ǫ implies that |f (x) − f (a)| < δ.
i) Prove that if f is suounitnoc at some a ∈ A, it is suounitnoc at every b ∈ A.
ii) Let f : R+ → R be given by f (x) = 1/x. Prove that at every a ∈ R+ , f is
continuous but not suounitnoc.
iii) Let f : R → R be given by f (x) = 1 if x is irrational and f (x) = 0 if x is rational.
Prove that at every a ∈ R, f is suounitnoc but not continuous.
5.1.13 A function f : A → B is ticonnuous at a ∈ A if there exists a real number
δ > 0 such that for all real numbers ǫ > 0 and for all x ∈ A, |x − a| < δ implies that
|f (x) − f (a)| < ǫ.
i) Suppose that f is ticonnuous at some a ∈ A. Prove that there exist a real number
δ > 0 and b ∈ B such that for all x ∈ A, |x − a| < δ implies that f (x) = b.
ii) Give an example of a continuous function that is not ticonnuous at every a in the
domain.
iii) Prove that every function that is ticonnuous at every point in the domain is
continuous.
 Section 5.2: Topology and the Extreme value theorem 149

5.1.14 A function f : A → B is connuousti at a ∈ A if for all real numbers ǫ > 0 there

exists a real number δ > 0 such that for all x ∈ A, |f (x) −f (a)| < ǫ implies that |x −a| < δ.
i) Let f : R → R be a constant function. Prove that f is not connuousti at any
a ∈ R.
ii) Let f : R → R be given by f (x) = 1 if x is irrational and f (x) = 0 if x is rational.
Prove that f is not connuousti at any a ∈ R.
iii) Let f : R → R be given by f (x) = x if x is irrational and f (x) = x + 1 if x is
rational. Prove that at every a ∈ R, f is connuousti.

5.2 Topology and the Extreme value theorem

Topology and continuity go hand in hand (see Exercise 5.2.5), but not in the obvious
way, as the next two examples show.
(1) If f : A → B is continuous and A is open, it need not be true that the range of f
is open in B, even if A is bounded. For example, let f : R → R or f : (−1, 1) → R
be the squaring function. Certainly f is continuous and R is an open set, but the
image of f is [0, ∞) or [0, 1), which is not open.
(2) If f : A → B is continuous and A is closed, it need not be true that the range of
1
f is closed in B. For example, let f : R → R be given by f (x) = 1+x 2 . This f is

continuous as it is a rational function defined on all of R. The domain A = R is

a closed set (and open), but the image of f is (0, 1], which is not closed (and not
open).
However, if f is continuous and its domain A is closed and in addition bounded, then
it is true that the image of f is closed. This fact is proved next.

Theorem 5.2.1 Let A be a closed and bounded subset of C or R, and let f : A → C be

continuous. Then the range of f is closed and bounded.

Proof. We first prove that the range is closed. Let b be a limit point of the range. We want
to prove that b is in the range. Since b is a limit point, by definition for every positive real
number r, B(b, r) contains an element of the range (even an element of the range that is
different from b). In particular, for every positive integer m there exists xm ∈ A such that
f (xm ) ∈ B(b, 1/m). If for some m we have f (xm ) = b, then we are done, so we may assume
that for all m, f (xm ) 6= b. Thus there are infinitely many xm . As in Construction 3.6.1,
we can construct nested quarter subrectangles Rn that contain infinitely many xm . There
is a unique complex number c that is contained in all the Rn . By construction, c is the
limit point of the set of the xm , hence of A. As A is closed, c ∈ A. But f is continuous
at c, so that for all ǫ > 0 there exists δ > 0 such that for all x ∈ A, if |x − c| < δ then
150 Chapter 5: Continuity

|f (x) − f (c)| < ǫ/2. In particular, for infinitely many large m, |xm − c| < δ, so that for
these same m, |f (xm ) − f (c)| < ǫ/2. But for all large m we also have |f (xm ) − b| < ǫ/2,
so that by the triangle inequality, for |f (c) − b| < ǫ. Since this is true for all positive ǫ, it
follows that f (c) = b. Thus any limit point of the range is in the range, so that the range
is closed.
Next we prove that the range is bounded. If not, then for every positive integer m there
exists xm ∈ A such that |f (xm )| > m. Again we use Construction 3.6.1, and this time
we construct nested quarter subrectangles Rn that contain infinitely many of these xm .
As before, there is a unique complex number c that is contained in all the Rn and in A,
and as before (with ǫ = 2), we get that for infinitely many m, |f (xm ) − f (c)| < 1. But
|f (xm)| > m, so that by the reverse triangle inequality,
|f (c)| ≥ |f (xm) − (f (xm ) − f (c))| ≥ |f (xm )| − |f (xm ) − f (c)| > m − 1

for infinitely many positive integers m. Since |f (c)| is a fixed real number, it cannot be
larger than all positive integers. Thus we get a contradiction to the assumption that the
range is not bounded, which means that the range must be bounded.

Theorem 5.2.2 (Extreme value theorem) Let A be a closed and bounded subset of C,
and let f : A → R be a continuous function. Then there exist l, u ∈ A such that for all
x ∈ A,
f (l) ≤ f (x) ≤ f (u).

In other words, f achieves its maximum value at u and its minimum value at l.

Proof. By Theorem 5.2.1, the range {f (x) : x ∈ A} of f is a closed and bounded subset
of R, so that its infimum L and supremum U are real numbers which are by closedness in
the range. Thus there exist u, l ∈ A such that L = f (l) and U = f (u).

Example 5.2.3 The function f : [−2, 2] → R given by f (x) = x2 − 6x + 5 achieves a

minimum and maximum. We can rewrite the function as f (x) = (x − 3)2 − 4, from which
it is obvious that the minimum of the function is achieved at 3 – but wait a minute, this
function is not defined at 3 and hence cannot achieve a minimum at 3. Here is a correction:
the quadratic function (x − 3)2 − 4 achieves its minimum at 3 and is decreasing on (−∞, 3),
so that f achieves its minimum on [−2, 2] at −2 and its maximum at 2.
Exercises 5.2.2 and 5.2.3 below give examples of continuous invertible functions whose
inverses are not continuous. The next theorem contains instead some positive results of
this flavor.
 Section 5.2: Topology and the Extreme value theorem 151

Theorem 5.2.4 Let F = R or F = C. Let A, B be subsets of F , let A be closed and

bounded, and let f : A → B be continuous and invertible. Then f −1 is continuous.

Proof. We need to prove that f −1 is continuous at every b ∈ B. Let ǫ > 0. Set a = f −1 (b).
The set B(a, ǫ) is open, so its complement is closed. Therefore C = A \ B(a, ǫ) = A ∩
(F \ B(a, ǫ)) is closed. As C ⊆ A, then C is also bounded. Thus by Theorem 5.2.1,
{f (x) : x ∈ C} is a closed and bounded subset of F . By injectivity of f , this set does not
contain b = f (a). The complement of this set contains b and is open, so that there exists
δ > 0 such that B(b, δ) ⊆ F \ {f (x) : x ∈ C). Now let y ∈ B with |y − b| < δ. Since f
is invertible, there exists x ∈ A such that y = f (x). By the choice of δ, x 6∈ C, so that
f −1 (y) = x ∈ B(a, ǫ) = B(f −1 (b), ǫ). In short, for every ǫ > 0 there exists δ > 0 such that
for all y ∈ B, if |y − b| < δ, then |f −1 (y) − f −1 (a)| < ǫ. This proves that f −1 is continuous
at b.

Exercises for Section 5.2

5.2.1 (Compare with Theorem 5.2.1.)
i) Show that f : (0, 1) → R defined by f (x) = x1 is a continuous function whose
domain is open and bounded but the range is not bounded.
ii) Show that f : (−1, 1) → R defined by f (x) = x2 is a continuous function whose
domain is open and bounded but the range is not open.
5.2.2 Consider the continuous function f in Example 5.1.6.
i) Prove that f has an inverse function f −1 : (−1, 1) → (−1, 0) ∪ (0, 1].
ii) Graph f −1 and prove that f −1 is not continuous.
iii) Compare with Theorem 5.2.4.
5.2.3 Define f : {x ∈ C : |x| < 1 or |x| ≥ 2} → C by

x, if |x| < 1;
f (x) = 1
2 x, otherwise.
Prove that f is continuous and invertible but that f −1 is not continuous. Compare with
Theorem 5.2.4.
5.2.4 Let A be a closed and bounded subset of C, and let f : A → C be continuous.
i) Why are we not allowed to talk about f achieving its maximum or minimum?
ii) Can we talk about the maximum and minimum absolute values of f ? Justify your
answer.
iii) Can we talk about the maximum and minimum of the real or of the imaginary
components of f ? Justify your answer.
152 Chapter 5: Continuity

*5.2.5 Let F be R or C, and let A and B be subsets of F . Prove that f : A → B is

continuous if and only if for every open subset U of F there exists an open subset V of F
such that the set {x ∈ A : f (x) ∈ U } = V ∩ A.
5.2.6 Let f : (0, 1) → R be defined by f (x) = x1 . Prove that f is continuous and that f
has neither a minimum nor a maximum. Explain why this does not contradict the Extreme
value theorem (Theorem 5.2.2).
5.2.7 (Outline of another proof of the Extreme value theorem for closed intervals in R, due
to Samuel J. Ferguson, “A one-sentence line-of-sight proof of the Extreme value theorem”,
American Mathematical Monthly 121 (2014), 331.) Let f : [a, b] → R be a continuous
function. The goal is to prove that f achieves its maximum at a point u ∈ [a, b]. Set
L = {x ∈ [a, b] : for all y ∈ [a, b], if y < x then f (y) ≤ f (x)}.
i) Prove that a ∈ L.
ii) Let c ∈ Bd (L) (boundary of L). Prove that c ∈ [a, b].
iii) Suppose that c 6∈ L.
− Prove that there exists y ∈ [a, c) such that f (y) > f (c).
− Prove that there exists δ > 0 such that for all x ∈ [a, b]∩B(c, δ), |f (x)−f (c)| <
(f (y) − f (c))/2. (Hint: use continuity at c.)
− Prove that there exists x ∈ L ∩ B(c, min{δ, c − y}).
− Prove that f (y) − f (x) > 0.
− Prove that y ≥ x.
− Prove that c − y > |x − c| ≥ c − x ≥ c − y, which is a contradiction.
iv) Conclude that c ∈ L and that L is a closed set.
v) Let u = sup(L). Prove that u exists and is an element of L.
vi) Let k > f (u). Prove that the set Sk = {x ∈ [c, b] : f (y) ≥ k} is closed, and in
particular that Sk , if non-empty, has a minimum.
vii) Suppose that for some x ∈ [a, b], f (x) > f (u).
− Prove that x > u and x 6∈ L.
− Prove that there exists x1 ∈ (u, x) such that f (x1 ) > f (x).
− Prove that there exist x1 ∈ (u, x), x2 ∈ (u, x1 ), x3 ∈ (u, x2 ), . . ., xn ∈
(u, xn−1 ), . . ., such that · · · > f (xn) > f (xn−1 ) > · · · > f (x3 ) > f (x2 ) >
f (x1 ) > f (x) > f (u).
− Prove that the set Sf (u) does not have a minimum.
viii) Conclude that f (u) is the maximum of f on [a, b].
5.2.8 Let A be a subset of R, and let f : A → R be a function with a maximum at
u ∈ A. What can you say about the slope of the line between (a, f (a)) and (u, f (u)) for
a ∈ A \ {u}?
 Section 5.3: Intermediate value theorem 153

5.3 Intermediate value theorem

In this section, all functions are real-valued. The reason is that we can only make
comparisons in an ordered field, and R is an ordered field (Axiom 2.7.9), and C is not
(Exercise 3.1.6).

Theorem 5.3.1 (Intermediate value theorem) Let a, b ∈ R with b > a, and let
f : [a, b] → R be continuous. Let k be a real number strictly between f (a) and f (b). Then
there exists c ∈ (a, b) such that f (c) = k.

Here is a picture that illustrates the Intermediate value theorem: for value k on the y-
axis between f (a) and f (b) there happen to be two c for which f (c) = k. The Intermediate
value theorem guarantees that one such c exists but does not say how many such c exist.

f (b) b

a c1 c2 b x

b
f (a)

Proof of Theorem 5.3.1: Set a0 = a and b0 = b. Apply Construction 3.6.1 of halving

intervals, with the property P of intervals being that k is strictly between the values of f
at the two endpoints. If for any n > 0, f ( an−1 +b
2
n−1
) = k, then the theorem is proved and
we can immediately stop the proof (and the construction).
When k is not equal to f ( an−1+b 2
n−1
), then since k is between f (an−1 ) and f (bn−1 ),
necessarily k is between f (an−1 ) and f ( an−1 +b 2
n−1
) or else between f ( an−1 +b
2
n−1
) and
f (bn−1 ). Choose that half [an , bn] of [an−1 , bn−1 ] which says that k is strictly between
f (an ) and f (bn ). Thus for all n, k is strictly between f (an ) and f (bn ). By construction
c = sup{a1 , a2 , a3 , . . .} = inf{b1 , b2 , b3 , . . .} is in [a, b]. So c is in the domain of f .
We will prove that for all ǫ > 0, |f (c) − k| < ǫ. Let ǫ > 0. Since f is continuous,
it is continuous at c, so there exists δ > 0 such that for all x ∈ [a, b], if |x − c| < δ,
then |f (x) − f (c)| < ǫ/3. By Exercise 2.7.22, there exists a positive integer n such that
1/2n < δ/(b − a). As an ≤ c ≤ bn , we have |an − c| ≤ |an − bn | = (b − a)/2n < δ, so
that |f (an ) − f (c)| < ǫ/3. Similarly, |f (bn) − f (c)| < ǫ/3. Hence by the triangle inequality,
|f (bn) − f (an )| < 2ǫ/3. But k is between f (an ) and f (bn), and thus both |f (an ) − k| and
154 Chapter 5: Continuity

|f (bn) − k| must be less than 2ǫ/3. Therefore

|f (c) − k| ≤ |f (c) − f (an )| + |f (an ) − k| < ǫ,

and since this true for all ǫ > 0, it follows by Theorem 2.10.4 that f (c) = k.
An important application of this theorem is in the next section, introducing the radical
functions. (So far we have sporadically used the square roots only, relying on some facts
from high school that we have not yet proved.)

c2 −2
Example 5.3.2 There exists a real number c such that c5 − 4 = c2 +2 .

2
Proof. Let f : R → R be defined by f (x) = x5 − 4 − xx2 −2 +2
. This function is a rational
function and defined for all real numbers, so that by Theorem 5.1.3, f is continuous. Note
that f (0) = −3 < 0 < f (2), so that by Theorem 5.3.1 there exists c in (0, 2) such that
2
f (c) = 0. In other words, c5 − 4 = cc2 −2
+2 .

Theorem 5.3.3 Let I be an interval in R, and let f : I → R be continuous. Then the

image of f is an interval in R.

Proof. For any c and d in the image of f , by the Intermediate value theorem (The-
orem 5.3.1), any real number between c and d is in the image of f , which proves the
theorem.
However, if f : A → B is continuous and injective and A is open, then the range of f
is open in B. We prove this first for A and B subsets of R, and then for subsets of C.
Compare the next theorem to Example 5.1.6.

Theorem 5.3.4 Let I be an interval in R, B a subset of R, and let f : I → B be

continuous and invertible. Then f −1 is continuous and f, f −1 are either both strictly
increasing or both strictly decreasing. Furthermore, if I is open, so is B.

Proof. Let a < b be in I. Since f has an inverse, f (a) 6= f (b), so that by trichotomy, either
f (a) < f (b) or f (a) > f (b).
For now we assume that f (a) < f (b). With that we prove that f is an increasing
function, i.e., that for any x, y ∈ I, if x < y then f (x) < f (y). First suppose that
x < y < z are in I. By invertibility, f (x), f (y), f (z) are distinct. If f (x) is between
f (y) and f (z), then an application of the Intermediate value theorem gives c ∈ (y, z) such
that f (c) = f (x). But x < y < c, so x and c are distinct, and f (c) = f (x) contradicts
invertibility of f . So f (x) is not between f (y) and f (z), and similarly f (z) is not between
f (x) and f (y). Thus necessarily f (x) < f (y) < f (z) or f (x) > f (y) > f (z). By setting
x = a and z = b we get that f is increasing on [a, b], by setting y = a and z = b we get that
 Section 5.3: Intermediate value theorem 155

f is increasing on I ∩ (−∞, a], and by setting x = a and y = b we get that f is increasing

on I ∩ [b, ∞). Thus f is increasing on I.
By Theorem 5.3.3 we know that B is an interval. By definition of inverses, f −1 is
strictly increasing.
If I is in addition open, let c ∈ B and a = f −1 (c). Since I is open and f is increasing,
there exist b1 , b2 ∈ A such that b1 < a < b2 . Thus f (b1 ) < f (a) = c < f (b2 ), and by
Theorem 5.3.3 we know that (f (b1), f (b2)) is an open subset of B that contains c. Thus B
is open.
Now we prove that f −1 is a continuous function at an arbitrary y be in the domain
of f −1 . Let ǫ > 0. Let c = f −1 (y). Let z ∈ (c, c + ǫ) ∩ I. (Of course, if c is the right
endpoint of I, there is no such z.) Since f is increasing, y = f (c) < f (z) and we set
δ1 = f (z) − y. Then for any x ∈ (y, y + δ1 ) = (f (c), f (z)), by the Intermediate value
theorem x is in the range of f , i.e., x is in the domain of f −1 , and since f −1 is increasing,
c = f −1 (y) < f −1 (x) < f −1 (f (z)) = z < c + ǫ. Thus |f −1 (x) − f −1 (y)| < ǫ. This
proves that for the given ǫ > 0 there exists δ1 > 0 such that for any x ∈ (y, y + δ1 ) ∩ I,
|f −1 (x) − f −1 (y)| < ǫ. Similarly, there exists δ2 > 0 such that for all x ∈ (y − δ2 , y) ∩ I,
|f −1 (y) − f −1 (x)| < ǫ. Now set δ = min{δ1 , δ2 }. By what we just proved, for all x in the
domain of f −1 , if |x − y| < δ, then |f −1 (x) − f −1 (y)| < ǫ. Thus f −1 is continuous at y,
and since y was arbitrary, f −1 is continuous.
Finally, suppose that f (a) > f (b) for some a < b in I. Set g = −f . Then g is
invertible and continuous, g(a) < g(b), and by the work so far we know that g is a strictly
increasing function with a continuous strictly increasing inverse g −1 . Thus f = −g is a
strictly decreasing function, and f −1 = −g −1 is a continuous strictly decreasing function.
In addition, if I is open, so is B.

Exercises for Section 5.3

5.3.1 Let f (x) = x3 + 4, g(x) = x4 + x2 . Prove that there exists c ∈ [−4, 4] such that
f (c) = g(c).
5.3.2 Assuming that cos is a continuous function, prove that there exists a real number x
such that cos x = x.
5.3.3 Let f : R → Q be a continuous function. Prove that f is a constant function.
i) Let g : C → Q be a continuous function. Prove that g is a constant function.
ii) Does a continuous function f : Q → R have to be constant?
5.3.4 (Fixed point theorem) Let f : [a, b] → [a, b] be a continuous function. Prove that
there exists c ∈ [a, b] such that f (c) = c.
156 Chapter 5: Continuity

5.3.5 Find real numbers a < b and a continuous function f : [a, b] → [a, b] such that
f (c) 6= c for all c ∈ (a, b).
5.3.6 Find real numbers a < b and a continuous function f : (a, b) → (a, b) such that
f (c) 6= c for all c ∈ (a, b).
5.3.7 The goal of this exercise is to prove that every polynomial of odd degree has a real
root. Write f (X) = a0 + a1 X + a2 X 2 + · · · + an X n for some real numbers a0 , a1 , a2 , . . . , an
such that n is odd and an 6= 0. Set b = |ann | max{|a0 |, |a1 |, . . . , |an−1 |, |an |}.
i) Prove that b is a positive real number. If f (b) = 0 or if f (−b) = 0, we have found
the root. So we may assume that f (b) and f (−b) are not zero.
ii) Justify all steps below:
|a0 + a1 (±b) + a2 (±b)2 + · · · + an−1 (±b)n−1 |
≤ |a0 | + |a1 b| + |a2 b2 | + · · · + |an−1 bn−1 | (by the triangle inequality)
|an |b |an |b2 |an |b3 |an |bn
≤ + + +···+ (by the choice of b)
n n n n
|an |b

≤ 1 + b + b2 + · · · + bn−1
n
|an |b n−1

≤ b + bn−1 + bn−1 + · · · + bn−1 (because b ≥ 1)
n
0 1 2 n−1 (place markers)
|an |b n−1
≤ nb
n
= |an |bn
= |an bn |.

iii) Justify all steps below:

|f (±b)| = |a0 + a1 (±b) + a2 (±b)2 + · · · + an−1 (±b)n−1 + an (±b)n |
≥ |an (±b)n | − |a0 + a1 (±b) + a2 (±b)2 + · · · + an−1 (±b)n−1 |
≥ 0.

iv) Prove that f (b) has the same sign (positive or negative) as an bn and that f (−b)
has the same sign as an (−b)n .
v) Prove that f (b) and f (−b) have opposite signs.
vi) Prove that f has a real root in (−b, b).
 Section 5.4: Radical functions 157

5.4 Radical functions

Let n be a positive integer. Define the function f (x) = xn with domain R if n is odd
and domain R≥0 otherwise.

Theorem 5.4.1 The range of f is R if n is odd and is R≥0 otherwise.

Proof. Certainly 0 = 0n is in the range. If a > 0, then

0n = 0 < a < a + 1 ≤ (a + 1)n .

The last inequality is by Exercise 2.7.1. Since exponentiation by n is a polynomial function,
it is continuous, and so by the Intermediate value theorem (Theorem 5.3.1), there exists
r ∈ (0, a + 1) such that a = r n . Thus every non-negative real number is in the range of f .
If a < 0 and n is odd, then similarly
(a − 1)n ≤ a − 1 < a < 0 = 0n ,

and the Intermediate value theorem guarantees that there exists r ∈ (a − 1, 0) such that
a = r n . So for odd n all real numbers are in the range.
By Theorem 2.8.2 and Exercise 2.8.1, f is strictly increasing. Thus by Theorem 2.8.4,
f has an inverse function f −1 . By the power rule, f is continuous, so that by Theorem 5.3.4,
f −1 is strictly increasing and continuous.

Definition 5.4.2 The function f (f (x) = xn ) is called exponentiation by n, where n is

called the exponent, or power, and x the base. The inverse of exponentiation by n is
√ √
the nth radical function, and is written as n . For each a in the domain, n a is called
the nth root of a.
√
We just established that for any positive integer n, n is defined on R≥0 , or even
√
on R if n is odd. We also just established that n is a strictly increasing continuous
function. (Compare to Theorem 2.9.2 in which we established with quite a bit of labor
√
that the square root function 2 is defined on R+ .) By continuity we immediately get the
following:

Theorem 5.4.3 (Radical rule for limits) For any positive integer n and any a in the
√
domain of n ,
√ √
lim n x = n a.
x→a

Raising a non-negative real number to an integer power mth power and then taking
the nth radical of the result yields the same things as first taking the nth radical of the
original non-negative real number and then raising the result to the mth power. We record
158 Chapter 5: Continuity
√
this function as n/m = ( )m/n , and call it exponentiation by m/n. Exponentiation by
1/n is the same as taking the nth radical.

Theorem 5.4.4 Let r ∈ Q. If r ≥ 0 let A = R≥0 and if r < 0 let A = R+ . Let f : A → A

be defined by f (x) = xr . Then f is continuous.
(1) If r > 0, then f is strictly increasing and has an increasing inverse.
(2) If r < 0, then f is strictly decreasing and has a decreasing inverse.
(3) If r = 0, then f is a constant function.

Proof. Write r = m/n, where n and m are integers and n 6= 0. Since m/n = (−m)/(−n),
by possibly multiplying by −1 we may assume that n > 0. Then f is a composition of
exponentiation by m with exponentiation by 1/n, in either order. Exponentiation by non-
negative m is continuous by the constant or power rule, exponentiation by negative m is
continuous by the quotient rule, and exponentiation by 1/n is continuous by Theorem 5.4.3.
Thus f is continuous by the composite rule.
If r > 0, then m, n > 0, and then f is the composition of two strictly increasing
functions, hence strictly increasing. If r < 0, then m < 0 and n > 0, so f is the compo-
sition of a strictly increasing and strictly decreasing function, hence strictly decreasing by
Exercise 2.8.7. Monotonicity of inverses follows from Theorem 5.3.4.
The last part is obvious.
We prove in Theorem 7.5.4 that exponentiation by arbitrary real numbers (not just
by rational numbers) is continuous.

Exercises for Section 5.4

5.4.1 Let n, m ∈ Q, and suppose that a and b are in the domain of exponentiation by n
and m. Prove:
i) an · bn = (ab)n .
ii) (an )m = amn .
iii) an · am = an+m .
iv) If a 6= 0, then a−n = 1/an .
√ √
5.4.2 Let f (x) = x − −x.
i) Prove that for all a in the domain of f , limx→a f (x) is inapplicable.
ii) Prove that f is continuous.
 Section 5.4: Radical functions 159

5.4.3 Determine the following limits, and justify all steps by invoking the relevant theo-
rems/rules: p
i) lim x2 − 3x + 4.
x→2 √
√
x− 2
ii) lim .
x→2 x2 + 4
x−2
iii) lim 2 .
x→2 x − 4
√ √
x− 2
iv) lim .
x→2 x2 − 4
√
5.4.4 Determine the domain of the function f given by f (x) = −x2 .
5.4.5 Here is an alternate proof of Theorem 5.4.3. Study the proof, and provide any
√
missing commentary. Let A be the domain of n .
i) Prove that an element of A is a limit point of A.
ii) Suppose that a = 0. Set δ = ǫn . Let x ∈ A satisfy 0 < |x − a| < δ. Since the nth
root function is an increasing function on A, it follows that
√ √  √  p √
 n x − n a =  n x = n |x| < n δ = ǫ.
√
iii) Suppose that a > 0. First let ǫ′ = min{ǫ, n a}. So ǫ′ is a positive number. Set
√ √ n
δ = min{(ǫ + n a)n − a, a − ( n a − ǫ′ )n }. Note that
√ √
(ǫ + n a)n − a > ( n a)n − a = 0,
√ √
0 ≤ n a − ǫ′ < n a,
√ n
a − ( n a − ǫ′ )n > 0,

so that δ is positive. Let x ∈ A satisfy 0 < |x − a| < δ. Then −δ < x − a < δ and
√ √ n
a − δ < x < δ + a. Since δ = min{(ǫ + n a)n − a, a − ( n a − ǫ′ )n }, it follows that
√ n
√

a − a − ( n a − ǫ′ )n ≤ a − δ < x < a + δ ≤ a + (ǫ + n a)n − a ,
√ √
or in other words, ( n a − ǫ′ )n < x < (ǫ + n a)n . Since the nth radical function is
increasing on R+ , it follows that
√ √ √ √
n
a − ǫ ≤ n a − ǫ′ < n x < ǫ + n a,
√ √ √ √
whence −ǫ < n x − n a < ǫ, so that | n x − n a| < ǫ, as desired.
iv) Assume that a is negative. Then necessarily n is an odd integer. By what we have
proved for positive numbers in the domain, such as for −a, there exists δ > 0 such
√ √ 
that for all x ∈ R, if 0 < |(−x) − (−a)| < δ, then  n −x − n −a < ǫ. But then
for x ∈ R with 0 < |x − a| < δ, since n is odd,
√ √   √ √
  √ √ 
 n x − n a = (−1) n x − n a  =  n −x − n −a < ǫ.
160 Chapter 5: Continuity

5.4.6 Recall from Exercise 3.4.5 that for every non-zero complex number a there exist
exactly two complex numbers whose squares are a. Let’s try to create a continuous square
root function f : C → C. (We will fail.)
i) Say that for all a in the first quadrant we choose f (a) in the first quadrant. Where
are then f (a) for a in the remaining quadrants?
ii) Is it possible to extend this square root function to a function f on all of C
(the positive and negative real and imaginary axes) in such a way as to make
limx→a f (x) = f (a) for all a ∈ C?
√ √ p √
iii) Explain away the problematic claims −4 · −9 = (−4)(−9) = 36 = 6,
√ √
−4 · −9 = 2i · 3i = −6 from page 9.
iv) Let D be the set of all complex numbers that are not on the negative real axis.
Prove that we can define a continuous square root function f : D → C.
(v)* Let θ be any real number, and let D be the set of all complex numbers whose
counterclockwise angle from the positive real axis is θ. Prove that we can define a
continuous square root function f : D → C.
5.4.7 (The goal of this exercise and the next one is to develop exponential functions
without derivatives and integrals. You will see in Section 7.4 that derivatives and integrals
give a more elegant approach.) Let c ∈ (1, ∞) and let f : Q → R+ be the function
f (x) = cx .
i) Why is f a function? (Is it well-defined, i.e., are we allowed to raise positive real
numbers to rational exponents?)
ii) Prove that f is strictly increasing. (Hint: Theorem 5.4.4.)
iii) Let ǫ > 0. Justify the following:
n+1
X n+1
X n+1
X
n+1 i i−1

i−1
(ǫ + 1) −1= (ǫ + 1) − (ǫ + 1) = ǫ(ǫ + 1) ≥ ǫ = (n + 1)ǫ.
i=1 i=1 i=1

Use the Archimedean property of R (Theorem 2.7.12) to prove that the set {(ǫ +
1)n : n ∈ N0 } is not bounded above.
iv) Prove that there exists a positive integer n such that c1/n < ǫ + 1.
v) Prove that there exists δ1 > 0 such that for all x ∈ (0, δ1 ) ∩ Q, |cx − 1| < ǫ.
vi) Prove that there exists δ2 > 0 such that for all x ∈ (−δ2 , 0) ∩ Q, |cx − 1| < ǫ.
vii) Prove that limx→0,x∈Q cx = 1.
viii) Prove that for any r ∈ R, limx→r,x∈Q cx exists and is a real number.
 Section 5.5: Uniform continuity 161

5.4.8 (Related to the previous exercise.) Let c ∈ R+ .

i) Prove that for any r ∈ R, limx→r,x∈Q cx exists and is a real number. (Hint: Case
c = 1 is special; case c > 1 done; relate the case c < 1 to the case c > 1 and the
quotient rule for limits.)
ii) We denote the limit in the previous part with cr . Prove that for all real numbers
c, c1 , c2 , r, r1 , r2 , with c, c1 , c2 > 0,
1
c−r = , (c1 c2 )r = cr1 cr2 , cr1 +r2 = cr1 cr2 , cr1 r2 = (cr1 )r2 .
cr
iii) Prove that the function g : R → R given by g(x) = cx is continuous. (This is easy.)

5.5 Uniform continuity

Definition 5.5.1 A function f is uniformly continuous if for all real numbers ǫ > 0
there exists a real number δ > 0 such that for all x and y in the domain, if |x − y| < δ,
then |f (x) − f (y)| < ǫ.

For example, constant functions are uniformly continuous.

In uniform continuity, given a real number ǫ > 0, there exists δ > 0 that depends only
on ǫ that makes some conclusion true, whereas in the definition of continuity at a, given a
real number ǫ > 0, there exists δ > 0 that depends on ǫ and on a for the same conclusion
to be true (with y = a). Thus the following is immediate:

Theorem 5.5.2 Every uniformly continuous function is continuous.

The converse is false in general:

Example 5.5.3 Let f (x) = x2 , with domain C. Since f is a polynomial function, it is

continuous. Suppose that f is also uniformly continuous. Then in particular for ǫ = 1
there exists δ > 0 such that for all x, y ∈ A, if |x − y| < δ then |f (x) − f (y)| < 1. Set x = δ1
and y = x + 2δ . Then |x − y| = δ/2 < δ and
 
 2   δ 2  δ2
 2 
|f (x) − f (y)| = x − y = δx +  = 1 + > 1 = ǫ.
4 4
This proves the negation of the definition of uniform continuity.
The converse of Theorem 5.5.2 does hold with some extra assumptions.
162 Chapter 5: Continuity

Theorem 5.5.4 Let A, B be closed and bounded subsets of R or C. Let f : A → B be

continuous. Then f is uniformly continuous.

Proof. Let ǫ > 0. Since f is continuous, for each a ∈ A there exists δa > 0 such that for
all x ∈ A, |x − a| < δa implies that |f (x) − f (a)| < ǫ/2. Note that A ⊆ ∪a∈A B(a, δa ).
By Theorem 3.6.3 there exists δ > 0 such that for all x ∈ A there exists a ∈ A such that
B(x, δ) ⊆ B(a, δa ). Let x, y ∈ A with |x − y| < δ. Since x, y ∈ B(x, δ) ⊆ B(a, δa ), it follows
that |x − a|, |y − a| < δa . Thus

|f (x) − f (y)| = |f (x) − f (a) + f (a) − f (y)|

≤ |f (x) − f (a)| + |f (a) − f (y)| (by the triangle inequality)
< ǫ/2 + ǫ/2 = ǫ.
√
Example 5.5.5 The continuous function is uniformly continuous.
√
Proof. We established in Section 5.4 that is continuous. Let ǫ > 0. We divide the
domain into two regions, one closed and bounded so we can invoke the theorem above, and
√
the other unbounded but where has a bounded derivative.
The first region is the closed and bounded interval [0, 2]. By the previous theorem
√ √
there exists δ1 > 0 such that for all a, x ∈ [0, 2], if |x − a| < δ1 then | x − a| < ǫ.
The second region is the unbounded interval [1, ∞). For a, x ∈ [1, ∞) with |x − a| < ǫ
we have
 √ √   
√ √ √ √ x + a   x−a 
| x − a| = ( x − a) √ √  = √ √ 
x + a  x + a
|x − a| √ √
≤ (because x, a ≥ 1)
2
< ǫ.
√
Finally, set δ = min{δ1 , ǫ, 1}. Let a and x be in the domain of such that |x−a| < δ.
Since |x − a| < δ ≤ 1, necessarily either x, a ∈ [0, 2] or x, a ≤ [1, ∞). We have analyzed
√ √
both cases, and we conclude that | x − a| < ǫ.

Theorem 5.5.6 Let f : A → B be uniformly continuous and C a subset of A. Let

g : C → B be defined as g(x) = f (x). Then g is uniformly continuous.

Proof. Let ǫ > 0. Since f is uniformly continuous, there exists δ > 0 such that for all
x, y ∈ A, if |x − y| < δ then |f (x) − f (y)| < ǫ. But then for any x, y ∈ C, if |x − y| < δ,
then |g(x) − g(y)| = |f (x) − f (y)| < ǫ.
 Section 5.5: Uniform continuity 163

Example 5.5.7 By Example 5.5.3 the squaring function is not uniformly continuous on
C or R, but when the domain is restricted to any bounded subset of C, that domain is a
subset of a closed and bounded subset of C, and so since squaring is continuous, it follows
by the previous theorem that squaring is uniformly continuous on the closed and bounded
set and hence on any subset of that.

Exercises for Section 5.5

5.5.1 Let f : C → C be given by f (x) = mx + l for some constants m, l. Prove that f is
uniformly continuous.
5.5.2 Which of the following functions are uniformly continuous? Justify your answers.
i) f : B(0, 1) → C, f (x) = x2 .
ii) f : (0, 1] → R, f (x) = 1/x.
iii) f : R → R, f (x) = x21+1 .
x
iv) f : R \ {0} → R, f (x) = |x| .
 x
, if x 6= 0;
v) f : R → R, f (x) = |x|
 0, otherwise.
1, if x is rational;
vi) f : R → R, f (x) =
0, otherwise.
vii) Re, Im : C → R.
viii) The absolute value function from C to R.
5.5.3 Suppose that f, g : A → C are uniformly continuous and that c ∈ C.
i) Prove that cf and f ± g are uniformly continuous.
ii) Is f · g uniformly continuous? Prove or give a counterexample.
5.5.4 Let f : (a, b) → C be uniformly continuous. Prove that there exists a continuous
function g : [a, b] → C such that f (x) = g(x) for all x ∈ (a, b).
5.5.5 Is the composition of two uniformly continuous functions uniformly continuous?
Prove or give a counterexample.
5.5.6 Let f : C → C be defined by f (x) = x3 .
i) Prove that f is continuous but not uniformly continuous.
ii) Find a uniformly continuous function g : C → C such that g ◦ f is uniformly
continuous.
iii) Find a uniformly continuous function g : C → C such that g ◦ f is not uniformly
continuous.
Chapter 6: Differentiation

The geometric motivation for differentiation comes from lines tangent to a graph of
a function f at a point (a, f (a)). For example, on the graph below are two secant lines
through (a, f (a)):

f (a)

a x

It appears that the line through (a, f (a)) and (x, f (x)) is closer to the tangent line to the
graph of f at (a, f (a)) if x is closer to a. Intuitively, the slope of the tangent line is the
limit of the slopes of the secant lines.

6.1 Definition of derivatives

Definition 6.1.1 Let A ⊆ C, and let a ∈ A be a limit point of A. A function f : A → C

is differentiable at a if
f (x) − f (a)
lim
x→a x−a
exists. In this case, we call this limit the derivative of f at a, and we use either Newton’s
 
notation f ′ (a) = (f (x))′x=a or Leibniz’s notation dx
df
(a) = dfdx
(x) 
x=a
.

A function is differentiable if it is differentiable at all points in its domain.

An alternative way of computing the derivatives is


df (x)  f (a + h) − f (a)
f ′ (a) =  = lim ,

dx x=a h→0 h
as this is simply a matter of writing x as a + h, and using that a + h = x → a if and only
if h → 0. (See Exercise 5.1.6.)
 Section 6.1: Definition of derivatives 165

Example 6.1.2 Let f (x) = mx + l, where m and l are complex numbers. Then for any
a ∈ C, Then
f (x) − f (a) (mx + l) − (ma + l) m(x − a)
f ′ (a) = lim = lim = lim = m.
x→a x−a x→a x−a x→a x−a
Alternatively,
f (a + h) − f (a) (m(a + h) + l) − (ma + l) mh
f ′ (a) = lim = lim = lim = m.
h→0 h h→0 h h→0 h

Example 6.1.3 Let f (x) = x2 . Then

f (x) − f (a)
f ′ (a) = lim
x→a x−a
x2 − a2
= lim
x→a x − a
(x − a)(x + a)
= lim
x→a x−a
= lim (x + a)
x→a
= 2a.
Alternatively,
f (a + h) − f (a)
f ′ (a) = lim
h→0 h
(a + h) − a2
2
= lim
h→0 h
a + 2ah + h2 − a2
2
= lim
h→0 h
2ah + h2
= lim
h→0 h
(2a + h)h
= lim
h→0 h
= lim (2a + h)
h→0
= 2a.

From now on, we will mostly use the alternative way of computing derivatives.

Example 6.1.4 Let f (x) = 1/x. Then

f (a + h) − f (a)
f ′ (a) = lim
h→0 h
1 1
−
= lim a+h a
h→0 h
166 Chapter 6: Differentiation
a a+h
a(a+h)
− a(a+h)
= lim (common denominator in the fractions)
h→0 h
a−a−h
a(a+h)
= lim
h→0 h
−h
= lim
h→0 a(a + h)h
−1
= lim
h→0 a(a + h)
−1
= 2
a
by the quotient rule for limits.
√
Example 6.1.5 Let f (x) = x. The domain of f is R≥0 , and for all a > 0,
f (a + h) − f (a)
f ′ (a) = lim
h→0 h
√ √
a+h− a
= lim
h→0 h
√ √ √ √
a+h− a a+h+ a
= lim ·√ √
h→0 h a+h+ a
(a + h) − a
= lim √ √ (since (x − y)(x + y) = x2 − y 2 )
h→0 h( a + h + a)
h
= lim √ √
h→0 h( a + h + a)
1
= lim √ √
h→0 a+h+ a
1
= √
2 a
by the linear, radical, composite, and quotient rules for limits. It is left to Exercise 6.1.4
to show that f is not differentiable at 0.

Example 6.1.6 Let f (x) = x3/2 . The domain of f is R≥0 , and for all a ≥ 0,
f (a + h) − f (a)
f ′ (a) = lim
h→0 h
(a + h) − a3/2
3/2
= lim
h→0 h
(a + h) − a3/2 (a + h)3/2 + a3/2
3/2
= lim ·
h→0 h (a + h)3/2 + a3/2
(a + h)3 − a3
= lim (since (x − y)(x + y) = x2 − y 2 )
h→0 h((a + h)3/2 + a3/2 )
 Section 6.1: Definition of derivatives 167

a3 + 3a2 h + 3ah2 + h3 − a3
= lim
h→0 h((a + h)3/2 + a3/2 )
(3a2 + 3ah + h2 )h
= lim
h→0 h((a + h)3/2 + a3/2 )

3a2 + 3ah + h2
= lim
h→0 (a + h)3/2 + a3/2
 2
 a3/23a+a3/2
, if a > 0;
= 2
 lim h = lim h1/2 = 0, if a = 0;
h→0 h3/2 h→0
3
= a1/2
2
by the linear, radical, composite, and quotient rules for limits. (Note that this f is differen-
tiable even at 0, whereas the square root function (previous example) is not differentiable
at 0.)
Note that in all these examples, f ′ is a function from some subset of the domain of f
to a subset of C, and we can compute f ′ at a number labeled x rather than a:
df (x) f (x + h) − f (x) f (z) − f (x)
f ′ (x) = = lim = lim .
dx h→0 h z→x z−x
The h-limit is perhaps preferable to the last limit, where it is z that varies and gets closer
and closer to x.

Example 6.1.7 The absolute value function is not differentiable at 0.

Proof. This function is not differentiable whether the domain is C or R. The reason is
that the limit of |h|−|0|
h
as h goes to 0 does not exist. Namely, if h varies over positive real
numbers, this limit is 1, and if h varies over negative real numbers, the limit is −1, so that
the limit indeed does not exist, and hence the derivative does not exist.
This gives an example of a continuous function that is not differentiable. (Any con-
tinuous function with a jagged graph is not differentiable.)

Exercises for Section 6.1

6.1.1 Prove that f : C → C given by f (x) = x3 is differentiable everywhere, and compute
the derivative function.
6.1.2 Prove that f : C \ {0} → C given by f (x) = 1/x2 is differentiable everywhere, and
compute the derivative function.
√
6.1.3 Prove that f : (0, ∞) → R given by f (x) = 1/ x is differentiable everywhere, and
compute the derivative function.
168 Chapter 6: Differentiation

6.1.4 Prove that the square root function is not differentiable at 0.

6.1.5 Prove that the function f : [0, ∞) → R given by f (x) = x3/2 is differentiable,
including at 0.
6.1.6 Let f : R → R be given by

x2 − 1, if x > 1;
f (x) =
x3 − x, if x ≤ 1.
Prove that f is differentiable at 1.
6.1.7 Let f : R → R be given by

x2 − 1, if x > 1;
f (x) = 4
x − 4x, if x ≤ 1.
Prove that f is continuous but not differentiable at 1.
6.1.8 Let f : R → R be given by

x2 , if x > 1;
f (x) =
x3 − x, if x ≤ 1.
Prove that f is not continuous and not differentiable at 1.

x2 , if x > 1;
6.1.9 Determine if the following function is differentiable at 1: f (x) =
2x, if x ≤ 1.
1 2 1 2
x , if x ≥ 0; x + 3, if x ≥ 0;
6.1.10 Let f, g : R → R be f (x) = 2 1 2 and g(x) = 2 1 2
− 2 x , if x < 0, −2x , if x < 0.
i) Prove that f is differentiable everywhere and that for all x ∈ R, f ′ (x) = |x|.
ii) Prove that g is not differentiable at 0.
iii) Prove that g is differentiable at any non-zero real number x with g ′ (x) = |x|.
6.1.11 Let f : (0, 1) → R be the square root function. Verify that f is differentiable,
bounded, even uniformly continuous, and that f ′ is not bounded. (Hint: Example 6.1.5.)

6.2 Basic properties of derivatives

Theorem 6.2.1 If f is differentiable at a, then f is continuous at a.

Proof. By definition of differentiability, a is a limit point of the domain of f and a is in the

domain of f . Furthermore, since limh→0 f (a+h)−f h
(a)
exists, by the product rule for limits
f (a+h)−f (a) ′
also limh→0 (h h ) exists and equals 0 · f (a) = 0. In other words, limh→0 (f (a +
h) − f (a)) = 0, so that by the sum rule for limits, limh→0 f (a + h) = limh→0 (f (a + h) −
f (a) + f (a)) = 0 + f (a) = f (a). Thus by Exercise 5.1.6, limx→a f (x) = f (a), so that f is
continuous at a.
 Section 6.2: Basic properties of derivatives 169

Theorem 6.2.2 (Basic properties of derivatives)

(1) (Constant rule) Constant functions are differentiable (at all real/complex numbers)
and the derivative is 0 everywhere.
(2) (Linear rule) The function f (x) = x is differentiable (at all real/complex numbers)
and the derivative is 1 everywhere.

Let A be a subset of C and let a ∈ A be a limit point of A. Suppose that f, g : A → C are

differentiable at a. Then
(3) (Scalar rule) For any c ∈ C, cf is differentiable at a and (cf )′ (a) = cf ′ (a).
(4) (Sum/difference rule) f ± g is differentiable at a and (f ± g)′ (a) = f ′ (a) ± g ′ (a).
(5) (Product rule) f · g is differentiable at a and (f · g)′ (a) = f ′ (a)g(a) + f (a)g ′ (a).
(6) (Quotient rule) If g(a) 6= 0, then f /g is differentiable at a and (f /g)′(a) =
f ′ (a)g(a)−f (a)g ′ (a)
(g(a))2
.

Proof. Parts (1) and (2) were already proved in part (1) of Example 6.1.2. (3) follows from
(cf )(a + h) − (cf )(h) cf (a + h) − cf (h) f (a + h) − f (h)
lim = lim = c lim = cf ′ (a),
h→0 h h→0 h h→0 h
and (4) follows from the sum rule for limits and from
(f + g)(a + h) − (f + g)(h) f (a + h) + g(a + h) − f (h) − g(h)
lim = lim
h→0 h h→0 h
f (a + h) − f (h) + g(a + h) − g(h)
= lim
h→0 h
 
f (a + h) − f (h) g(a + h) − g(h)
= lim +
h→0 h h
′ ′
= f (a) + g (a).

The following proves the product rule (5):

(f · g)(a + h) − (f · g)(a) f (a + h) · g(a + h) − f (a) · g(a)
lim = lim
h→0 h h→0 h
f (a + h) · g(a + h) − f (a) · g(a + h) + f (a) · g(a + h) − f (a) · g(a)
= lim (adding 0)
h→0 h
 
(f (a + h) − f (a))g(a + h) + f (a)(g(a + h) − g(a))
= lim
h→0 h
 
f (a + h) − f (a) g(a + h) − g(a))
= lim g(a + h) + f (a)
h→0 h h
′ ′
= f (a)g(a) + f (a)g (a),
where in the last step we used that f and g are differentiable at a and that g is continuous
at a (by Theorem 6.2.1). The proof of the quotient rule is similar, and the reader is invited
170 Chapter 6: Differentiation

to provide the commentary on the steps:

f f (a+h) f (a)
g (a + h) − fg (a) g(a+h) − g(a)
lim = lim
h→0 h h→0 h
f (a + h)g(a) − f (a)g(a + h)
= lim
h→0 hg(a + h)g(a)
f (a + h)g(a) − f (a)g(a) + f (a)g(a) − f (a)g(a + h)
= lim
h→0 hg(a + h)g(a)
f (a + h) · g(a + h) − f (a) · g(a + h) + f (a) · g(a) − f (a) · g(a + h)
= lim
h→0 hg(a + h)g(a)
 
f (a + h) · g(a + h) − f (a) · g(a + h) f (a) · g(a + h) − f (a) · g(a)
= lim −
h→0 hg(a + h)g(a) hg(a + h)g(a)
 
f (a + h) − f (a) g(a + h) f (a) g(a + h) − g(a)
= lim · − ·
h→0 h g(a + h)g(a) g(a + h)g(a) h
′ ′
f (a)g(a) − f (a)g (a)
= .
(g(a))2

Theorem 6.2.3 (Power rule) If n is a positive integer, then (xn )′ = nxn−1 .

Proof #1: Part (1) of Example 6.1.2 with m = 1 and l = 0 proves that (x1 )′ = 1, so that
the theorem is true when n = 1. Now suppose that the theorem holds for some positive
integer n. Then

(xn )′ = (xn−1 x)′

= (xn−1 )′ x + xn−1 x′ (by the product rule)
= (n − 1)xn−2 x + xn−1 (by induction assumption)
= (n − 1)xn−1 + xn−1
= nxn−1 ,

so that the theorem holds also for n, and so by induction also for all positive n.

Proof #2: The second proof uses binomial expansions as in Exercise 1.6.7:
(x + h)n − xn
(xn )′ = lim
h→0 h
Pn

n k n−k
k=0 k x h − xn
= lim
h→0 h
Pn−1 n
k n−k
x h
= lim k=0 k
h→0 h
Pn−1 n
k n−k−1
h k=0 k x h
= lim
h→0 h
 Section 6.2: Basic properties of derivatives 171

X
n−1 
n k n−k−1
= lim x h
h→0 k
k=0
n−1
X  
n k n−k−1
= x 0 (by the polynomial rule for limits)
k
k=0
 
n
= xn−1
n−1
= nxn−1 .

Theorem 6.2.4 (Polynomial, rational function rule for derivatives) Polynomial functions
are differentiable at all real/complex numbers and rational functions are differentiable at
all points in the domain.

Proof. The proof is an application of the sum, scalar, and power rules from Theorems 6.2.2
and 6.2.3.

Theorem 6.2.5 (The composite function rule for derivatives, aka the chain rule) Suppose
that f is differentiable at a, that g is differentiable at f (a), and that a is a limit point of the
domain of g ◦ f . (If f : A → B, g : B → C, and f is differentiable at a, then automatically
a is a limit point of A and hence of the domain of g ◦ f .) Then g ◦ f is differentiable at a,
and (g ◦ f )′ (a) = g ′ (f (a)) · f ′ (a).

Proof. Let ǫ > 0. Since f is differentiable at a, there exists δ1 > 0 such that for all a + h
in the domain of f , if 0 < |h| < δ1 then | f (a+h)−f h
(a)
− f ′ (a)| < min{1, ǫ/(2|g ′(f (a))| + 2)}.
For all such h, by the triangle inequality, | f (a+h)−f h
(a)
| < |f ′ (a)| + 1. By assumption g is
differentiable at f (a), so that there exists δ2 > 0 such that for all x in the domain of g, if
0 < |x − f (a)| < δ2 , then | g(x)−g(f
x−f (a)
(a))
− g ′ (f (a))| < ǫ/(2|f ′ (a)| + 2). Since f is differentiable
and hence continuous at a, there exists δ3 > 0 such that for all x in the domain of f , if
|x − a| < δ3 , then |f (x) − f (a)| < δ1 . Set δ = min{δ1 , δ2 , δ3 }. Let a + h be arbitrary in
the domain of g ◦ f such that 0 < |h| < δ. In particular a + h is in the domain of f . If
f (a + h) 6= f (a), then
   
 (g ◦ f )(a + h) − (g ◦ f )(a)   g(f (a + h)) − g(f (a)) 
 − g ′
(f (a)) · f ′
(a) = − g ′
(f (a)) · f ′
(a) 
 h   h 
 
 g(f (a + h)) − g(f (a)) f (a + h) − f (a) 
=  · − g ′ (f (a)) · f ′ (a)
f (a + h) − f (a) h
 
 g(f (a + h)) − g(f (a)) f (a + h) − f (a)
=  ′
− g (f (a)) ·
f (a + h) − f (a) h

f (a + h) − f (a) 
′
+ g (f (a)) · − g (f (a)) · f (a)
′ ′
h
172 Chapter 6: Differentiation
   
 g(f (a + h)) − g(f (a))   f (a + h) − f (a) 
≤ ′ 
− g (f (a)) ·  
f (a + h) − f (a) h 
 
 f (a + h) − f (a) 
+ |g (f (a))| 
′
− f (a)
′
h
ǫ ǫ
≤ · (|f ′ (a)| + 1) + |g ′ (f (a))| ′
2|f ′ (a)| + 2 2|g (f (a))| + 2
ǫ ǫ
< +
2 2
= ǫ.

Thus if there exists δ as above but possibly smaller such that f (a + h) 6= f (a) for all a + h
in the domain with 0 < |h| < δ, the above proves the theorem.
Now suppose that for all δ > 0 there exists h such that a + h is in the domain of f ,
0 < |h| < δ, and f (a + h) = f (a). Then in particular when h varies over those infinitely
many h, f ′ (a) = limh→0 f (a+h)−fh
(a)
= limh→0 h0 = 0. Also, for such h, (g ◦ f )(a + h) − (g ◦
f )(a) = g(f (a + h)) − g(f (a)) = 0, so that
 
 (g ◦ f )(a + h) − (g ◦ f )(a) 
 − g (f (a)) · f (a) = 0 < ǫ.
′ ′
 h
This analyzes all the cases and finishes the proof of the theorem.

Theorem 6.2.6 (Real and imaginary parts) Let A ⊆ R, and let a ∈ A be a limit
point of A. Let f : A → C. Then f is differentiable at a if and only if Re f and Im f are
differentiable at a, and in that case, f ′ = (Re f )′ + i(Im f )′ .

Proof. Since all h are necessarily real,

f (a + h) − f (a) Re(f (a + h) − f (a)) + i Im(f (a + h) − f (a))
=
h  h  
f (a + h) − f (a) f (a + h) − f (a)
= Re + i Im ,
h h
and by the definition of limits of complex functions, the limit of the function on the left
exists if and only if the limits of its real and imaginary parts exist.
Compare this last theorem with Exercises 6.2.4 and 6.2.5.
Exercises 5.2.2 and 5.2.3 each give an example of a continuous invertible function
whose inverse is not continuous. It is also true that the inverse of a differentiable invertible
function need not be differentiable. For example, the cubing function on R (f (x) = x3 )
is differentiable and invertible,
√ √
but its inverse, the cube root function, is not differentiable
3
0+h− 3 0
at 0, as the limit of h = h−1/3 does not exist as h goes to 0. Nevertheless, we can
connect differentiability of an invertible function to the derivative of the inverse, as shown
in the next theorem.
 Section 6.2: Basic properties of derivatives 173

Theorem 6.2.7 (Derivatives of inverses) Let A, B ⊆ C, let A be open, and let f :

A → B be an invertible differentiable function whose derivative is never 0. Then for all
b ∈ B, f −1 is differentiable at b, and
1
(f −1 )′ (b) = ,
f ′ (f −1 (b))
or in other words, for all a ∈ A, f −1 is differentiable at f (a) and
1
(f −1 )′ (f (a)) = .
f ′ (a)
Proof. Let b ∈ B. Then b = f (a) for some a ∈ A. Since A is open, there exists r > 0 such
that B(a, 2r) ⊆ A. Let C be any closed and bounded subset of A that contains B(a, r).
For example, C could be {x ∈ C : |x − a| ≤ r}. Define g : C → D = {f (x) : x ∈ C} as
g(x) = f (x). Then g is invertible and continuous. By Theorem 5.2.4, g −1 is continuous.
In particular, g −1 is continuous at b = f (a) = g(a).
Let ǫ > 0. Since g is differentiable at a, and g(a) = f (a) = b and g ′ (a) = f ′ (a), by
the limit
 definition of  derivatives, there exists δ > 0 such that for all x ∈ C, if |x − a| < δ
 g(x)−b 
then  x−a − f (a) < ǫ. By continuity of g −1 , there exists γ > 0 such that for all
′

−1
y ∈ D, if |y − b| < γ then |g (y) − a| = |g −1 (y) −1
 − g (b)| < δ. In particular, for all
   g(g (y))−b
−1 
such y,  g−1y−b ′ ′
(y)−a − f (a) =  g −1 (y)−a − f (a) < ǫ. This says that for all h ∈ C, if
   −1 
   
h
|h| < γ and b + h ∈ D, then  g−1 (b+h)−a − f ′ (a) =  g(g (b+h))−b
g −1 (b+h)−a − f ′
(a)  < ǫ. Thus
h
limh→0 g −1 (b+h)−a = f ′ (a), and by the quotient rule for limits,

g −1 (b + h) − g −1 (b) g −1 (b + h) − a 1 1
lim = lim = ′ = ′ −1 .
h→0 h h→0 h f (a) f (f (b))
But this holds for every C (and g which is f restricted to C), so that in particular,
f −1 (b + h) − f −1 (b) 1
(f −1 )′ (b) = lim = ′ −1 .
h→0 h f (f (b))
It should be noted that if we know that f −1 is differentiable, the proof of the last
part of the theorem above goes as follows. For all x ∈ B, (f ◦ f −1 )(x) = x, so that
(f ◦ f −1 )′ (x) = 1, and by the chain rule, f ′ (f −1 (x)) · (f −1 )′ (x) = 1. Then if f ′ is never 0,
1
we get that (f −1 )′ (x) = f ′ (f −1 (x)) .

Example 6.2.8 Let f : [0, ∞) → [0, ∞) be the function f (x) = x2 . We know that f
is differentiable at all points in the domain and that f ′ (x) = 2x. By Example 6.1.5 and
Exercise 6.1.4, the inverse of f , namely the square root function, is differentiable at all
positive x, but not at 0. The theorem above applies to positive x (but not to x = 0):
√ 1 1 1
( x)′ = (f −1 )′ (x) = = = √ .
f ′ (f −1 (x)) 2f −1 (x) 2 x
174 Chapter 6: Differentiation
√
Theorem 6.2.9 Let n be a positive integer. Then for all non-zero x in the domain of n
,
√ 1
( n x)′ = x1/n−1 .
n
Proof. Let A = R+ if n is even and let A = R \ {0} otherwise. Define f : A → A to
be f (x) = xn . We have proved that f is invertible and differentiable. The derivative is
√
f ′ (x) = nxn−1 , which is never 0. Thus by the previous theorem, f −1 = n is differentiable
with
√ 1 1 1 1 1
( n x)′ = ′ √ = √ n−1
= (n−1)/n
= x−(n−1)/n = x1/n−1 .
f ( x)
n
n( x)
n
nx n n

Theorem 6.2.10 (Generalized power rule) Let r be an arbitrary rational number and
let f : R+ → R+ be given by f (x) = xr . Then for all x, f ′ (x) = rxr−1 .

Proof. The power rule and quotient rules prove this in case r an integer, and the previous
theorem proves it in case r is one over a positive integer. Now suppose that r = m/n for
some integers m, n with n 6= 0. Since m/n = (−m)/(−n) is also a quotient of two integers,
we may write r = m/n so that m ∈ Z and n is a positive integer. Thus f is the composition
of exponentiation by m and by 1/n. By the chain rule,
1 1/n−1 m (m−1)/n+1/n−1
f ′ (x) = m(x1/n)m−1 ·
x = x = rxm/n−1/n+1/n−1 = rxr−1 .
n n
This proves the theorem for all rational r.
The theorem also holds for all real r. But to prove it for all real r one first needs to
define exponentiation by non-rational numbers. Such exponentiation was worked through
laboriously in Exercises 5.4.7 and 5.4.8, and if we were to continue that kind of labori-
ous treatment, the proof of the form of the derivative of such exponentiation would also
be laborious. So we postpone the definition of such exponentiation and the proof of its
derivative to Theorem 7.5.4, where with the help of integrals the definition and proofs write
themselves elegantly.

Exercises for Section 6.2

6.2.1 Provide the commentary for the proof of the quotient rule in Theorem 6.2.2.
6.2.2 Let f, g, h be differentiable. Prove that (f gh)′ = f ′ gh+f g ′ h+f gh′ . More generally,
prove a product rule formula for the derivative of the product f1 · f2 · · · · · fn .
6.2.3 Prove yet another form of the general power rule for derivatives: If f is dif-
ferentiable at a, then for every positive integer n, f n is differentiable at a, and (f n )′ (a) =
n(f (a))n−1f ′ (a).
6.2.4 Prove that the functions Re, Im : C → R are not differentiable at any a.
 Section 6.2: Basic properties of derivatives 175

6.2.5 (Compare with Theorem 6.2.6.) Prove that the absolute value function on R is
differentiable at all non-zero a ∈ R. Prove that the absolute value function on C is not
differentiable at any non-zero a ∈ C. (Hint: Let h = (r − 1)a for r near 1.)
6.2.6 The following information is known: c f (c) f ′ (c) g(c) g ′ (c)
0 1 2 6 4
1 −1 0 5 3
2 2 −3 6 −6
3 4 2 3 5
4 0 1 4 7
For each of the following, either provide the derivative or argue that there is not enough
information. In any case, justify every answer.
i) (f − g)′ (1) =
ii) (f · g)′ (2) =
 ′
iii) fg (3) =
′
iv) (g ◦ f ) (4) =
6.2.7 A function f is differentiable on (−2, 5) and f (3) = 4, f ′ (3) = −1. Let g(x) = 3x.
For each of the statements below determine whether it is true, false, or if there is not
enough information. Explain your reasoning.
i) f is constant.
ii) The slope of the tangent line to the graph of f at 3 is 4.
iii) f is continuous on (−2, 5).
iv) The derivative of (f ◦ g) at 1 is −3.
v) (f + g)′ (3) = 2.
x−1
6.2.8 Let f (x) = x−2 .
i) Find all a in the domain of f such that the tangent line to the graph of f at a has
slope −1.
ii) Find all a in the domain of f such that the tangent line to the graph of f at a has
slope 2. (Solutions need not be real.)
6.2.9 We assume the following familiar properties of the trigonometric functions: sin is
differentiable on R, sin′ = cos, and when restricted to [−π, π], sin has an inverse. Use
Theorem 6.2.7, the fact that (sin(x))2 + (cos(x))2 = 1, and that on [−π, π] cosine is non-
negative to prove that (sin−1 )′ (x) = √1−x
1
2
.
6.2.10 Make the requisite assumptions about the trigonometric functions to prove that
−1
(cos−1 )′ (x) = √1−x 2
.
6.2.11 Make the requisite assumptions about the trigonometric functions to prove that
1
(tan−1 )′ (x) = 1+x 2.
176 Chapter 6: Differentiation

6.3 The Mean value theorem

In this section the domains and codomains of all functions are subsets of R.

Theorem 6.3.1 Let f : [a, b] → R, and let c ∈ [a, b] such that f achieves an extreme
value at c (i.e., either for all x ∈ [a, b], f (c) ≤ f (x) or for all x ∈ [a, b], f (c) ≥ f (x)). Then
at least one of the following holds:
(1) c = a;
(2) c = b;
(3) f is not continuous at c;
(4) f is not differentiable at c;
(5) f is differentiable at c and f ′ (c) = 0.

Proof. It suffices to prove that if the first four conditions do not hold, then the fifth one
has to hold. So we assume that c 6= a, c 6= b, and that f is differentiable at c.
Suppose that f ′ (c) > 0. By the definition of derivative, f ′ (c) = limx→c f (x)−f
x−c
(c)
.
Thus for all x very near c but larger than c, f (x)−f x−c
(c)
> 0, so that f (x) − f (c) > 0, so
that f does not achieve its maximum at c. Also, for all x very near c but smaller than c,
f (x)−f (c)
x−c
> 0, so that f (x) − f (c) < 0, so that f does not achieve its minimum at c. This
is a contradiction, so that f ′ (c) cannot be positive. Similarly, f ′ (c) cannot be negative.
Thus f ′ (c) = 0.
Thus to find extreme values of a function, one only has to check if extreme values
occur at the endpoints of the domain, at points where the function is not continuous or
non-differentiable, or where it is differentiable and the derivative is 0. One should be aware
that just because any of the five conditions is satisfied, we need not have an extreme value
of the function. Here are some examples:
(1) The function f : [−1, 1] → R given by f (x) = x3 − x has neither the maximum
nor the minimum at the endpoints. 
x, if x > 0;
(2) Let f : [−1, 1] → R be given by f (x) = Then f is not continuous
1/2, if x ≤ 0.
at 0 but f does not have a minimum or maximum at 0.
x, if x > 0;
(3) Let f : [−1, 1] → R be given by f (x) = Then f is continuous
2x, if x ≤ 0.
and not differentiable at 0, yet f does not have a minimum or maximum at 0.
(4) Let f : [−1, 1] → R be given by f (x) = x3 . Then f is differentiable, f ′ (0) = 0, but
f does not have a minimum or maximum at 0.
 Section 6.3: The Mean value theorem 177

Theorem 6.3.2 (Darboux’s theorem) Let a < b be real numbers, and let f : [a, b] → R
be differentiable. Then f ′ has the intermediate value property, i.e., for all k between f ′ (a)
and f ′ (b) there exists c ∈ [a, b] such that f ′ (c) = k.

Proof. If f ′ (a) = k, we set c = a, and similarly if f ′ (b) = k, we set c = b. So we may

assume that k is strictly between f ′ (a) and f ′ (b).
The function g : [a, b] → R given by g(x) = f (x)−kx is differentiable, hence continuous.
Note that g ′ (x) = f ′ (x) − k, so that 0 is strictly between g ′ (a) and g ′ (b). If g ′ (a) > 0, let
c ∈ [a, b] such that g achieves a maximum at c, and if g ′ (a) < 0, let c ∈ [a, b] such that g
achieves a minimum at c. Such c exists by the Extreme value theorem (Theorem 5.2.2).
Note that g ′ (a) < 0 if and only if f ′ (a) < k, which holds if and only if f ′ (b) > k, which
in turn holds if and only if g ′ (b) > 0. Similarly g ′ (a) > 0 if and only if g ′ (b) < 0. Thus
for both choices of c, c cannot be a or b, so that c ∈ (a, b). By Theorem 6.3.1, g ′ (c) = 0.
Hence f ′ (c) = k.

Theorem 6.3.3 (Rolle’s theorem) Let a, b ∈ R with a < b, and let f : [a, b] → R be a
continuous function such that f is differentiable on (a, b). If f (a) = f (b), then there exists
c ∈ (a, b) such that f ′ (c) = 0.

Proof. By the Extreme value theorem (Theorem 5.2.2) there exist l, u ∈ [a, b] such that
f achieves its minimum at l and its maximum at u. If f (l) = f (u), then the minimum
value of f is the same as the maximum value of f , so that f is a constant function, and so
f ′ (c) = 0 for all c ∈ (a, b).
Thus we may assume that f (l) 6= f (u). It may be that f achieves its minimum at the
two endpoints, in which case u must be strictly between a and b. Similarly, it may be that
f achieves its maximum at the two endpoints, in which case l must be strictly between a
and b. In all cases of f (l) 6= f (u), either l or u is not the endpoint.
Say that l is not the endpoint. Then a < l < b. For all x ∈ [a, b], f (x) ≥ f (l),
so that in particular for all x ∈ (a, l), f (x)−f
x−l
(l)
≤ 0 and for all x ∈ (l, b), f (x)−f
x−l
(l)
≥ 0.
Since f ′ (l) = limx→l f (x)−f
x−l
(l)
exists, it must be both non-negative and non-positive, so
necessarily it has to be 0.
If instead u is not the endpoint, then a < u < b and for all x ∈ [a, b], f (x) ≤ f (u).
Thus in particular for all x ∈ (a, u), f (x)−f
x−u
(u)
≥ 0 and for all x ∈ (u, b), f (x)−f
x−u
(u)
≤ 0.
Since f ′ (u) = limx→u f (x)−f
x−u
(u)
exists, it must be both non-negative and non-positive, so
necessarily it has to be 0. Thus in all cases we found c ∈ (a, b) such that f ′ (c) = 0.
178 Chapter 6: Differentiation

Theorem 6.3.4 (Mean value theorem) Let a, b ∈ R with a < b, and let f : [a, b] → R
be a continuous function such that f is differentiable on (a, b). Then there exists c ∈ (a, b)
such that f ′ (c) = f (b)−f
b−a
(a)
.

Here is an illustration of this theorem: the slope f (b)−f

b−a
(a)
of the line from (a, f (a)) to
(b, f (b)) also equals the slope of the tangent line to the graph at some c between a and b:

a b x

Proof. Let g : [a, b] → R be defined by g(x) = f (x) − f (b)−f b−a

(a)
(x − a). By the sum and
scalar rules for continuity and differentiability, g is continuous on [a, b] and differentiable
on (a, b). Also, g(a) = f (a) and g(b) = f (b) − f (b)−fb−a
(a)
(b − a) = f (b) − (f (b) − f (a))
= f (a) = g(a). Thus by Rolle’s theorem, there exists c ∈ (a, b) such that g ′ (c) = 0. But
f (b) − f (a)
g ′ (x) = f ′ (x) − ,
b−a
f (b)−f (a) f (b)−f (a)
so that 0 = g ′ (c) = f ′ (c) − b−a , whence f ′ (c) = b−a .
The rest of this section consists of various applications of the Mean value theorem.
More concrete examples are left for the exercises.

Theorem 6.3.5 Let a, b ∈ R with a < b, and let f : [a, b] → R be a continuous function
such that f is differentiable on (a, b).
(1) If f ′ (c) ≥ 0 for all c ∈ (a, b), then f is non-decreasing on [a, b].
(2) If f ′ (c) > 0 for all c ∈ (a, b), then f is strictly increasing on [a, b].
(3) If f ′ (c) ≤ 0 for all c ∈ (a, b), then f is non-increasing on [a, b].
(4) If f ′ (c) < 0 for all c ∈ (a, b), then f is strictly decreasing on [a, b].
(5) If f ′ (c) = 0 for all c ∈ (a, b), then f is a constant function.

Proof of part (2): Let x, y ∈ [a, b] with x < y. By Theorem 6.3.4 there exists c ∈ (x, y)
such that f ′ (c) = f (x)−f
x−y
(y)
. Since f ′ (c) > 0 and x < y, necessarily f (x) < f (y). Since x
and y were arbitrary with x < y, then f is strictly increasing on [a, b].
 Section 6.3: The Mean value theorem 179

Example 6.3.6 Assuming that sin′ (x) ≤ 1 for all x ∈ R, we have that sin(x) ≤ x for all
x ≥ 0.

Proof. Let f (x) = x − sin(x). Then f is differentiable on R, and f ′ (x) = 1 − sin′ (x) ≥ 0.
By the previous theorem, f is non-decreasing, so that for all x ≥ 0, x − sin(x) = f (x) ≥
f (0) = 0, whence x ≥ sin(x).

Theorem 6.3.7 (Cauchy’s mean value theorem) Let a < b be real numbers and let
f, g : [a, b] → R be continuous functions that are differentiable on (a, b). Then there exists
c ∈ (a, b) such that
f ′ (c)(g(b) − g(a)) = g ′ (c)(f (b) − f (a)).
f ′ (c) f (b)−f (a)
In particular, if g ′ (c) 6= 0 and g(b) 6= g(a), this says that g ′ (c) = g(b)−g(a) .

Proof. Define h : [a, b] → R by h(x) = f (x)(g(b) − g(a)) − g(x)(f (b) − f (a)). Then h is con-
tinuous on [a, b] and differentiable on (a, b). Note that h(a) = f (a)(g(b)−g(a))−g(a)(f (b)−
f (a)) = f (a)g(b) − g(a)f (b) = h(b). Then by the Mean value theorem (Theorem 6.3.4)
there exists c ∈ (a, b) such that h′ (c) = 0, i.e., 0 = f ′ (c)(g(b) − g(a)) − g ′ (c)(f (b) − f (a)).

Theorem 6.3.8 (Cauchy’s mean value theorem, II) Let a < b be real numbers and
let f, g : [a, b] → R be continuous functions that are differentiable on (a, b) and such that
g ′ is non-zero on (a, b). Then g(b) 6= g(a), and there exists c ∈ (a, b) such that
f ′ (c) f (b) − f (a)
′
= .
g (c) g(b) − g(a)

Proof. By the Mean value theorem (Theorem 6.3.4) there exists c ∈ (a, b) such that
g ′ (c) = g(b)−g(a)
b−a . By assumption, g ′ (c) 6= 0, so that g(b) 6= g(a). The rest follows by
Cauchy’s mean value theorem (Theorem 6.3.7).

Exercises for Section 6.3

6.3.1 Let f : R → R be a polynomial function of degree n.
i) If f has m distinct roots, prove that f ′ has at least m − 1 distinct roots.
ii) Suppose that f has n distinct roots. Prove that f ′ has exactly n − 1 distinct roots.
(You may need Theorem 2.4.13.)
iii) Give an example of f with n = 2, where f has no roots but f ′ has 1 root. Why
does this not contradict the first part?
6.3.2 Let f : [0, 1] → C be given by f (x) = x3 + ix2 . Prove that there exists no c between
0 and 1 such that f ′ (c) = f (1)−f
1−0
(0)
. Does this contradict the Mean value theorem?
180 Chapter 6: Differentiation

6.3.3 Let f : [a, b] → C be continuous on [a, b] and differentiable on (a, b). Prove that
(a)
(a)

there exist c, d ∈ (a, b) such that Re(f ′ (c)) = Re f (b)−f
b−a and Im(f ′ (d)) = Im f (b)−f
b−a .
Give an example showing that c may not equal to d. Does this contradict the Mean value
theorem?
6.3.4 Prove the remaining parts of Theorem 6.3.5.
6.3.5 Let A be an open interval in R, and let f : A → R be differentiable with a bounded
derivative function. Prove that f is uniformly continuous.

6.4 L’Hôpital’s rule

The domains and codomains of all functions in this section are subsets of R.

Theorem 6.4.1 (L’Hôpital’s rule, easiest version) Let A ⊂ C, let a ∈ A be a limit

point of A, and f, g : A → C such that
(1) f (a) = g(a) = 0.
(2) f and g are differentiable at a.
(3) g ′ (a) 6= 0.
f (x) f ′ (a)
Then lim = ′ .
x→a g(x) g (a)
f (x) f (x) − f (a)
Proof. lim = lim (since f (a) = g(a) = 0)
x→a g(x) x→a g(x) − g(a)
f (x)−f (a)
x−a
= lim
x→a g(x)−g(a)
x−a
′
f (a)
= (by the quotient rule for limits since g ′ (a) 6= 0).
g ′ (a)
The rest of the versions of L’Hôpital’s rule in this section only work for real-valued
functions on domains that are subsets of R.

Theorem 6.4.2 (L’Hôpital’s rule) Let a ∈ R and let f, g be real-valued functions

defined on an interval containing a such that
(1) f (a) = g(a) = 0.
(2) f and g are differentiable near a but not necessarily at a.
(3) g ′ is non-zero near a. (This is a condition for x near a but not equal to a.)
f ′ (x)
(4) lim ′ = L.
x→a g (x)
f (x)
Then lim = L.
x→a g(x)
 Section 6.4: L’Hôpital’s rule 181
f ′ (x)
Proof. Let ǫ > 0. Since lim ′ = L, there exists δ > 0 such that for all x in the domain, if
x→a g (x)
′
0 < |x−a| < δ then | fg′ (x)
(x)
−L| < ǫ. Let x be one such number. By Theorem 6.3.8 there exists
′
a number c strictly between a and x such that fg′ (c) (c)
= fg(x)−g(a)
(x)−f (a)
. Since f (a) = g(a) = 0,
f ′ (c) f (x)
this says that g′ (c) = g(x) . Since 0 < |x − a| < δ and c is between a and x, it follows that
0 < |c − a| < δ. Hence    ′ 
 f (x)   f (c) 
   
 g(x) − L =  g ′ (c) − L < ǫ.
With our definition of derivatives, this last version includes the one-sided cases where
the domains for f and g are of the form [a, b), or of the form (b, a].
I should note that many books omit hypothesis (3). A counterexample if we omit that
hypothesis is given in Exercise 9.11.5.
The versions of L’Hôpital’s rule so far deal with limits of the form “zero over zero”.
There are similarly three versions for the form “infinity over infinity”. I write and prove
only the right-sided version:

Theorem 6.4.3 (L’Hôpital’s rule) Let a < b be real numbers. Let f, g : (a, b) → R be
differentiable with the following properties:
(1) lim f (x) = lim g(x) = ∞.
x→b− x→b−
(2) For all x ∈ (a, b), g ′ (x) 6= 0.
f ′ (x)
(3) lim ′ = L.
x→b− g (x)
f (x)
Then lim− = L.
x→b g(x)

Proof. Let ǫ > 0. By  ′ assumption there exists δ1 > 0 such that for all x ∈ (a, b), if
 f (x) 
0 < b − x < δ1 , then  g′ (x) − L < ǫ/4. By possibly replacing δ1 by min{δ1 , (b − a)/2} we
may assume that b − δ1 > a.
Set a0 = b − δ1 . Fix x such that 0 < b − x < δ1 . Then x ∈ (a0 , b) ⊆ (a, b). By
′
Theorem 6.3.8 there exists c ∈ (a0 , x) such that fg′ (c) (c)
= fg(x)−g(a
(x)−f (a0 )
0)
, and so
   ′ 
 f (x) − f (a0 )   f (c) 
 − L  =  − L  < ǫ/4.
 g(x) − g(a0 )   g ′ (c) 

Since lim f (x) = lim g(x) = ∞, there exists δ2 > 0 such that for all x with
x→b− x→b−
0 < b − x < δ2 , f (x) and g(x) are non-zero, and so we can define h : (b − δ2 , b) → R as
f (a0 )
1− f (x)
h(x) = g(a0 )
.
1− g(x)

By Theorem 4.4.7, limx→b− h(x) = 1. Thus there exists δ3 > 0 such that for all x, if
0 < b − x < δ3 , then |h(x) − 1| < min{ǫ/4(|L| + 1), 12 }. The ( 21 )-restriction in particular
182 Chapter 6: Differentiation

means that h(x) > 21 . Set δ = min{δ1 , δ2 , δ3 }. Then for all x with 0 < b − x < δ,
   
 f (x)   f (x)  1
   
 g(x) − L =  g(x) h(x) − Lh(x) |h(x)|
 
 f (x) − f (a0 )  1
=  − Lh(x)
g(x) − g(a0 ) |h(x)|
 
 f (x) − f (a0 ) 
≤ 2  − Lh(x) (since |h(x)| > 1/2)
g(x) − g(a0 )
  
 f (x) − f (a0 ) 
≤ 2  − L + |L − Lh(x)| (by the triangle inequality)
g(x) − g(a0 )
 ′  
 f (c) 
= 2  ′ − L + |L| |1 − h(x)|
g (c)
ǫ ǫ
<2 +
4 4
= ǫ.

x2 −1
Example 6.4.4 Compute lim 3 .
x→1 x −1

x2 −1 (x−1)(x+1)
Proof #1: By Example 1.5.4, x3 −1 = (x−1)(x2 +x+1) , so that by Exercise 4.3.10 and Theo-
2
x −1
rem 4.3.6, lim 3 = lim 2x+1 = 1+1 2
12 +1+1 = 3 .
x→1 x −1 x→1 x +x+1

Proof #2: By Theorem 4.3.5, lim (x2 − 1) = 0 = lim (x3 − 1), lim (x2 − 1)′ = lim 2x = 2,
x→1 x→1 x→1 x→1
x2 −1
and lim (x3 − 1)′ = lim 3x2 = 3. Thus by L’Hôpital’s rule (Theorem 6.4.3), lim 3
x→1 x→1 x→1 x −1
equals 23 .

More forms of L’Hôpital’s rule are in the exercises below, as well as in Section 9.11
after the exponential and trigonometric functions have been covered.

Exercises for Section 6.4

x√2 −64
6.4.1 Compute and justify lim 3
x+2
.
x→−8
2x
6.4.2 What is wrong with the following: Since limx→2 2 = 2, by L’Hôpital’s rule we have
x2 +3
that limx→2 2x−3 = 2.
6.4.3 Let n be a positive integer and a ∈ C. Use L’Hôpital’s rule to prove that
n
−an
limx→a x x−a = nan−1 . (We have proved the case of a = 1 previously, say how.)
6.4.4 Explicitly state the assumptions about properties of trigonometric functions that
are necessary to prove lim sin(x)
x = 1 and lim 1−cos(x)
x = 0.
x→0 x→0
 Section 6.5: Higher-order derivatives, Taylor polynomials 183

6.4.5 (Here the goal is to prove another version of L’Hôpital’s rule) Let f, g : (a, ∞) → R
be differentiable. Suppose that lim f (x) = lim g(x) = ∞, that g ′ (x) is non-zero for all x,
x→∞ x→∞
′
and that lim fg′ (x)
(x)
= L.
x→∞
′
i) Let ǫ > 0. Prove that there exists N > a such that for all x > N , | fg′ (x)
(x)
− L| < ǫ.
ii) Prove that there exists N ′ > N such that for all x ≥ N ′ , f (x), g(x) > 0.
iii) Prove that there exists N ′′ > N ′ such that for all x ≥ N ′′ , f (x) > f (N ′ ) and
g(x) > g(N ′ ).
′
iv) Prove that for all x > N ′′ there exists c ∈ (N ′ , x) such that fg′ (c) (c)
=
f (x) 1−f (N ′ )/f (x)
g(x) 1−g(N ′ )/g(x) .
f (x)
v) Prove that lim = L.
x→∞ g(x)
6.4.6 (Yet another version of L’Hôpital’s rule) Let f, g : (a, ∞) → R be differentiable.
Suppose that lim f (x) = lim g(x) = 0, that g ′ (x) is non-zero for all x, and that lim
x→∞ x→∞ x→∞
f ′ (x) f (x)
g ′ (x) = L. Prove that lim = L.
x→∞ g(x)

6.5 Higher-order derivatives, Taylor polynomials

Let f be a function from a subset of C to C.

If f is continuous at a, then near a, f is approximately the constant function f (a)
because limx→a f (x) = f (a). Among all constant functions, the function y = f (a) approx-
imates f at a best.
If f is differentiable near a and the derivative is continuous at a, then for all x near a but
not equal to a, f (x) = f (x) − f (a) + f (a) = f (x)−f
x−a
(a)
(x − a) + f (a) ∼
= f ′ (a)(x − a) + f (a),
so that for x near a (and possibly equal to a), f is approximately the linear function
f ′ (a)(x − a) + f (a), i.e., f is approximated by its tangent line.
This game keeps going, but for this we need higher order derivatives:

Definition 6.5.1 Let f be differentiable. If f ′ is differentiable, we write the derivative of

f ′ as f ′′ , or as f (2) . If f (n−1) is differentiable, we denote its derivative as f (n) . Using this
notation we also write f (1) = f ′ and f (0) = f .

For example, if f (x) = xm , then for all n ≤ m, f (n) (x) = m(m − 1)(m − 2) · · · (m −
n + 1)xm−n .
184 Chapter 6: Differentiation

Definition 6.5.2 Let f be a function with derivatives of orders 1, 2, . . . , n existing at a

point a in the domain. The Taylor polynomial of f (centered) at a of order n is

′ f ′′ (a) 2 f (n) (a)

Tn,f,a (x) = f (a) + f (a)(x − a) + (x − a) + · · · + (x − a)n
2! n!
Xn (k)
f (a)
= (x − a)k .
k!
k=0

Example 6.5.3 If f (x) = x4 − 3x3 + 4x2 + 7x − 10, then with a = 0,

T0,f,0(x) = −10,
T1,f,0(x) = −10 + 7x,
T2,f,0(x) = −10 + 7x + 4x2 ,
T3,f,0(x) = −10 + 7x + 4x2 − 3x3 ,
Tn,f,0 (x) = −10 + 7x + 4x2 − 3x3 + x4 for all n ≥ 4,

and for a = 1,
T0,f,1 (x) = −1,
T1,f,1 (x) = −1 + 10(x − 1),
T2,f,1 (x) = −1 + 10(x − 1) + (x − 1)2 ,
T3,f,1 (x) = −1 + 10(x − 1) + (x − 1)2 + (x − 1)3 ,
Tn,f,1 (x) = −1 + 10(x − 1) + (x − 1)2 + (x − 1)3 + (x − 1)4 for all n ≥ 4.
Note that for all n ≥ 4, Tn,f,0 (x) = Tn,f,1 (x) = f (x).
The following is a generalization of this observation:

Theorem 6.5.4 If f is a polynomial of degree at most d, then for any a ∈ C and any
integer n ≥ d, the nth-order Taylor polynomial of f centered at a equals f .

Proof. Write f (x) = c0 +c1 x+· · ·+cd xd for some c0 , c1 , . . . , cd ∈ C. By elementary algebra,
it is possible to rewrite f in the form f (x) = e0 + e1 (x − a) + e2 (x − a)2 + · · · + ed (x − a)d
for some e0 , e1 , . . . , ed ∈ C. Now observe that

f (a) = e0 ,
f ′ (a) = e1 ,
f ′′ (a) = 2e2 ,
f ′′′ (a) = 6e3 = 3!e3 ,
f (4) (a) = 24e4 = 4!e4 ,
..
.
 Section 6.5: Higher-order derivatives, Taylor polynomials 185

f (k) (a) = k!ek if k ≤ d,

f (k) (a) = 0 if k > d.
But then for n ≥ d,
n
X d
X d
X
f (k) (a) k k!ek k
Tn,f,a (x) = (x − a) = (x − a) = ek (x − a)k = f (x).
k! k!
k=0 k=0 k=0

Theorem 6.5.5 (Taylor’s remainder theorem over R) Let I be an interval in R,

and let (a − r, a + r) = B(a, r) ⊆ I. Suppose that f : I → R has derivatives of orders
1, 2, . . . , n + 1 on B(a, r). Then for all x ∈ B(a, r) there exist c, d between a and x such
that
f (n+1) (c)
f (x) = Tn,f,a (x) + (x − a)(x − c)n ,
n!
f (n+1) (d)
f (x) = Tn,f,a (x) + (x − a)n+1 .
(n + 1)!
Proof. Let g : B(a, r) → R be defined by
n
X f (k) (t)
g(t) = (x − t)k .
k!
k=0

Then g is differentiable on B(a, r), and

n
X n
X
′ f (k+1) (t) k kf (k) (t)
g (t) = (x − t) − (x − t)k−1
k! k!
k=0 k=0
n−1
X f (k+1) (t) X f (k) (t) n
f (n+1) (t)
= (x − t)n + (x − t)k − (x − t)k−1
n! k! (k − 1)!
k=0 k=1
(n+1)
f (t)
= (x − t)n .
n!
Note that g(x) = f (x), g(a) = Tn,f,a (x). By the Mean value theorem (Theorem 6.3.4),
there exists c strictly between a and x such that g ′ (c)(x − a) = g(x) − g(a). In other words,
f (n+1) (c)
n! − c)n (x − a) = f (x) − Tn,f,a (x), which proves the first formulation.
(x
By Cauchy’s mean value theorem (Theorem 6.3.7) applied to functions g(t) and h(t) =
(x −t)n+1 , there exists d between x and a such that h′ (d)(g(x) −g(a)) = g ′ (d)(h(x) −h(a)).
In other words,
f (n+1) (d)
−(n + 1)(x − d)n (f (x) − Tn,f,a (x)) = (x − d)n (0 − (x − a)n+1 ).
n!
f (n+1) (d)
Since d is not equal to x, we get that −(n + 1)(f (x) − Tn,f,a (x)) = n!
(−(x − a)n+1 ),
which proves the second formulation.
186 Chapter 6: Differentiation

Theorem 6.5.6 (Taylor’s remainder theorem over C) Let A be a subset of C or

of R, let a ∈ A, and let r ∈ R+ such that B(a, r) ⊆ A. Let f : A → C have higher order
derivatives of orders 1, 2, . . . , n on B(a, r) with n ≥ 1. Then for every ǫ > 0 there exists
δ > 0 such that if x ∈ B(a, δ), then
|f (x) − Tn,f,a (x)| < ǫ.

Proof. Let ǫ > 0. Let M = 1 + max{|f (a)|, |f ′(a)|, . . . , |f (n) (a)|}. Since f is differentiable
at a, it is continuous at a, so there exists δ1 > 0 such that for all qx ∈ B(a, r), ifq|x − a| < δ1 ,
ǫ ǫ
then |f (x)−f (a)| < ǫ/(n+1). Set δ = min{r, δ1 , ǫ/(M (n+1)), (M (n+1)) , . . . (M (n+1))
2 n }.
Let x satisfy |x − a| < δ. Then x is in the domain of f and Tn,f,a , and
 ′′

 f (a) f (n)
(a) 
|f (x) − Tn,f,a (x)| = f (x) − f (a) − f (a)(x − a) −
′ 2
(x − a) − · · · − (x − a) 
n
2! n!
 ′′   (n) 
 f (a)   
≤ |f (x) − f (a)| + |f ′ (a)| |x − a| +   |x − a|2 + · · · +  f (a)  |x − a|n
2!   n! 
ǫ ǫ ǫ ǫ
< +M +M +···+M
n+1 M (n + 1) M (n + 1) M (n + 1)
= ǫ.

More on Taylor polynomials and Taylor series is in the exercises in this section, in
Exercise 7.4.12, and in Section 9.3.

Exercises for Section 6.5

6.5.1 Prove that if f is a polynomial function, then for every a ∈ R, Tn,f,a = f for all n
greater than or equal to the degree of f .
6.5.2 (Generalized product rule.) Suppose that f and g have derivatives of orders 1, . . . , n
at a. Prove that
Xn  
(n) n (k)
(f · g) (a) = f (a)g (n−k) (a).
k
k=0

6.5.3 (Generalized quotient rule.) Prove the following generalization of the product rule:
Suppose that f and g have derivatives of orders 1, . . . , n at a and that g(a) 6= 0. Find and
prove a formula for the nth derivative of the function f /g.
√
6.5.4 Compute the Taylor polynomial of f (x) = 1 + x of degree 5 centered at a = 0.
Justify your work.
√
6.5.5 Compute the Taylor polynomial of f (x) = 1 − x of degree 10 centered at a = 0.
Justify your work.
 Section 6.5: Higher-order derivatives, Taylor polynomials 187

6.5.6 Let n ∈ N0 be given. Justify all work below.

1
i) Compute the Taylor polynomial of f (x) = 1−x of degree n centered at a = 0.
ii) Compute Tn,f,0 (0.5). (Hint: Use Example 1.5.4.)
iii) Compute Tn,f,0 (0.5) − f (0.5).
iv) Compute n such that |Tn,f,0(0.5) − f (0.5)| < 0.001.
v) Try to use Theorem 6.5.5 to determine n such that |Tn,f,0(0.5) − f (0.5)| < 0.001.
Note that this usage is not fruitful.
vi) Use Theorem 6.5.5 to determine n such that |Tn,f,0 (0.4) − f (0.4)| < 0.001.
6.5.7 Explicitly assume the necessary properties of trigonometric functions to compute
the Taylor polynomials of degree n of sin and cos centered at a = 0.
i) Estimate the error term with Theorem 6.5.5.
ii) In fact, calculators and computers use Taylor polynomials to compute values of
trigonometric and also exponential functions. Compute sin(1) and sin(100) to
within 0.01 of true value. What degree Taylor polynomial suffices for each? Use
Theorem 6.5.5.
6.5.8 Let f : C → R be the absolute value function.
i) Prove that f has derivatives of all orders at all non-zero numbers.
ii) Compute T2,f,i.
6.5.9 Let f : R → R be given by f (x) = |x|3 .
i) Prove that f is differentiable, and compute f ′ .
ii) Prove that f ′ is differentiable, and compute f ′′ .
iii) Prove that f ′′ is not differentiable.
6.5.10 Find a differentiable function f : R → R such that f ′ is not differentiable.
6.5.11 Find a function f : R → R that has derivatives of orders 1, 2, 3, 4, but such that
f (4) is not differentiable.
Chapter 7: Integration

The basic motivation for integration is computing areas of regions bounded by graphs
of functions. In this chapter we develop the theory of integration for functions whose
domains are subsets of R. The first two sections handle only codomains in R, and at the
end of Section 7.4 we extend integration to functions with codomains in C. We do not
extend to domains being subsets of C as that would require multi-variable methods and
complex analysis, which are not the subject of this course.

7.1 Approximating areas

In this section, domains and codomains of all functions are subsets of R. Thus we can
draw the regions and build the geometric intuition together with the formalism.
Let f : [a, b] → R. The basic aim is to compute the signed area of the region bounded
by the x-axis, the graph of y = f (x), and the lines x = a and x = b. By “signed” area we
mean that we add up the areas of the regions above the x-axis and subtract the areas of
the regions below the x-axis. Thus a signed area may be positive, negative or zero.

y = f (x)
y

a b x

In the plot above, there are many (eight) regions whose boundaries are some of the
listed curves, but only the shaded region (comprising of two of the eight regions in the
count) is bounded as a subset of the plane.
The simplest case of an area is when f is a constant function with constant value c.
Then the signed area is c · (b − a), which is positive if c > 0 and non-positive if c ≤ 0.
For a general f , we can try to approximate the area by rectangles, such as the following:
 Section 7.1: Approximating areas 189

y = f (x)
y

a b x

It may be hard to decide how close the approximation is to the true value. But we can
approximate the region more systematically, by having heights of the rectangles be either
the least possible height or the largest possible height, as below:

a b a b

Then clearly the true area is larger than the sum of the areas of the darker rectangles
on the left and smaller than the sum of the areas of the darker rectangles on the right.
We establish some notation for all this.

Definition 7.1.1 A function f : A → C is bounded if the range {f (x) : x ∈ A} of f is a

bounded subset of C.

Remark 7.1.2 If f is a bounded function with codomain R, then by the Least upper
bound theorem (Axiom 2.9.1), for any subset B of the domain, sup{f (x) : x ∈ B} and
inf{f (x) : x ∈ B} are real numbers.
190 Chapter 7: Integration

Definition 7.1.3 A partition of [a, b] is a finite subset of [a, b] that contains a and b. We
typically write a partition in the form P = {x0 , x1 , . . . , xn }, where x0 = a < x1 < x2 <
· · · < xn−1 < xn = b. (For example, in the figures above, n = 10.)
Let f : [a, b] → R be a bounded function.
The lower sum of f with respect to P is
n
X
L(f, P ) = inf{f (x) : x ∈ [xk−1 , xk ]}(xk − xk−1 ).
k=1

The upper sum of f with respect to P is

n
X
U (f, P ) = sup{f (x) : x ∈ [xk−1 , xk ]}(xk − xk−1 ).
k=1

By the previous remark, L(f, P ) and U (f, P ) are real numbers.

Clearly if f is a constant function f (x) = c for all x, then all lower and all upper sums
are c(b − a), so that every lower and every upper sum equals c(b − a). If instead f is a
non-constant function, then for every partition P there exists at least one subinterval of
P on which the supremum of the values of f is strictly bigger than the infimum of such
values, so that L(f, P ) < U (f, P ).
By the geometric set-up for all partitions P of [a, b],

L(f, P ) ≤ the signed area ≤ U (f, P ). (7.1.4)

In particular, if U (f, P ) − L(f, P ) < ǫ, then either U (f, P ) or L(f, P ) serves as an approx-
imation of the true signed area within ǫ of its true value. For most functions a numerical
approximation is the best we can hope for.

Example 7.1.5 Approximate the area under the curve y = f (x) = 30x4 + 2x between
x = 1 and x = 4. We first establish a partition Pn = {x0 , . . . , xn } of [1, 4] into n equal
subintervals. The length of each subinterval is (4 − 1)/n, and x0 = 1, so that x1 =
x0 + 3/n = 1 + 3/n, x2 = x1 + 3/n = 1 + 2 · 3/n, and in general, xk = 1 + k · 3/n.
Note that xn = 1 + n · 3/n = 4, as needed. Since f ′ (x) = 12x3 + 2 is positive on [1, 4],
it follows that f is increasing on [1, 4]. Thus necessarily for each i, the infimum of all
values of f on the ith subinterval is achieved at the left endpoint, and the supremum at
the right endpoint. In symbols, this says that inf{f (x) : x ∈ [xk−1 , xk ]} = f (xk−1 ) and
sup{f (x) : x ∈ [xk−1 , xk ]} = f (xk ). For example, with n = 1, L(f, P1 ) = f (1) · 3 = 96 and
U (f, P1) = f (4) · 3 = 3 · (30 · 44 + 2 · 4) = 23064. Thus the true area is some number between
96 and 23064. Admittedly, this is not much information. A computer program produced
the following better numerical approximations for lower and upper sums with respect to
 Section 7.1: Approximating areas 191

partitions Pn into n equal subintervals:

n L(f, Pn) U (f, Pn)

10 5061.2757 7358.0757
100 6038.727 6268.407
1000 6141.5217 6164.48967
10000 6151.8517 6154.148
100000 6152.885 6153.1148
1000000 6152.988 6153.011

Notice how the lower sums get larger and the upper sums get smaller as we take finer
partitions. We would like to conclude that the true area is between 6152.988 and 6153.011.
This is getting closer but may still be insufficient precision. For more precision, partitions
would have to get even finer, but the calculations slow down too.
The observed monotonicity is not a coincidence:

Theorem 7.1.6 Let P, R be partitions of [a, b] such that P ⊆ R. (Then R is called a

refinement of P , and R is said to be a finer partition than P .) Then
L(f, P ) ≤ L(f, R), U (f, P ) ≥ U (f, R).

Proof. Write P = {x0 , x1 , . . . , xn }. Let i ∈ {1, . . . , n}. Let R ∩ [xk−1 , xk ] = {y0 =

xk−1 , y1 , . . . , ym−1 , ym = xk }. By set inclusion, inf{f (x) : x ∈ [yj−1 , yj ]} ≥ inf{f (x) : x ∈
[xk−1 , xk ]}, so that
m
X
L(f, {y0, y1 , . . . , ym−1 , ym }) = inf{f (x) : x ∈ [yj−1 , yj ]}(yj − yj−1 )
j=1
Xm
≥ inf{f (x) : x ∈ [xk−1 , xk ]}(yj − yj−1 )
j=1
m
X
= inf{f (x) : x ∈ [xk−1 , xk ]} (yj − yj−1 )
j=1
= inf{f (x) : x ∈ [xk−1 , xk ]}(xk − xk−1 ),

and hence that

Xn n
X
L(f, R) = L(f, R ∩ [xk−1 , xk ]) ≥ inf{f (x) : x ∈ [xk−1 , xk ]}(xk − xk−1 ) = L(f, P ).
k=1 k=1

The proof for upper sums is similar.

192 Chapter 7: Integration

1, if x is rational;
Example 7.1.7 Let f (x) = Then for any partition P =
0, if x is irrational.
{x0 , . . . , xn } of [−2, 4], for all i, inf{f (x) : x ∈ [xk−1 , xk ]} = 0 and sup{f (x) : x ∈
Pn
[xk−1 , xk ]} = 1, so that L(f, P ) = 0 and U (f, P ) = k=1 1 · (xk − xk−1 ) = xn − x0 =
4 − (−2) = 6. Thus in this case, changing the partition does not produce better
approximations.

Theorem 7.1.8 For any partitions P and Q of [a, b], and for any bounded function f :
[a, b] → R,
L(f, P ) ≤ U (f, Q).

Proof. Let R = P ∪ Q. (Then R is a refinement of both P and Q.) By Theorem 7.1.6,

L(f, P ) ≤ L(f, R) and U (f, P ) ≥ U (f, R). Since always L(f, R) ≤ U (f, R), the conclusion
follows by transitivity of ≤.
It follows that if f is bounded on [a, b], then the set of all lower sums of f as P varies
over all the partitions of [a, b] is bounded above, so that the set of all lower sums has a real
least upper bound. Similarly, the set of all upper sums is bounded below and has a real
greatest lower bound.

Definition 7.1.9 The lower integral of f over [a, b] is

L(f ) = sup{L(f, P ) : as P varies over partitions of [a, b]},

and the upper integral of f over [a, b] is

U (f ) = inf{U (f, P ) : P varies over partitions of [a, b]}.

We say that f is integrable over [a, b] when L(f ) = U (f ). We call this common
value the integral of f over [a, b], and we write it as
Z b Z b Z b
f= f (x) dx = f (t) dt.
a a a

Theorem 7.1.8 shows that L(f ) ≤ U (f ). Equality holds sometimes (for integrable
functions), but not always.
Ra
For example, every function f : [a, a] → R is integrable over [a, a], and a f = 0.
If f is a constant function f (x) = c for all x, then all lower and all upper sums are
Rb
c(b − a), so that a f = c(b − a), and all constant functions are integrable. We have seen
before that if f is non-constant function, then for every partition P , L(f, P ) < U (f, P ),
yet for good functions it can happen that L(f ) ≤ U (f ).
Example 7.1.7 shows that sometimes strict inequality holds. Note that we have not
yet proved that the function in Example 7.1.5 is integrable, but Section 7.3 we will prove
that every continuous function is integrable over a closed bounded interval.
 Section 7.1: Approximating areas 193

Theorem 7.1.10 Let f : [a, b] → R be bounded. Then f is integrable over [a, b] if and
only if for all ǫ > 0 there exists a partition Pǫ of [a, b] such that U (f, Pǫ) − L(f, Pǫ) < ǫ.

Proof. Suppose that f is integrable over [a, b]. Then L(f ) = U (f ). Let ǫ > 0. Since L(f )
is the supremum of all lower sums L(f, P ) as P varies over partitions of [a, b], there exists
a partition P1 of [a, b] such that L(f ) − L(f, P1) < ǫ/2. Similarly, there exists a partition
P2 of [a, b] such that U (f, P2) − U (f ) < ǫ/2. Let P = P1 ∪ P2 . Then P is a partition of
[a, b], and by Equation (7.1.4) and by Theorem 7.1.6,
L(f, P1) ≤ L(f, P ) ≤ U (f, P ) ≤ U (f, P2).

Thus since L(f ) = U (f ), we have that

U (f, P )−L(f, P ) ≤ U (f, P2)−L(f, P1) = U (f, P2 )−U (f )+L(f )−L(f, P1) < ǫ/2+ǫ/2 = ǫ.

Now suppose that for every ǫ > 0 there exists a partition Pǫ of [a, b] such that U (f, Pǫ)−
L(f, Pǫ) < ǫ. By the supremum/infimum definitions of lower and upper integrals,
0 ≤ U (f ) − L(f ) ≤ U (f, Pǫ) − L(f, Pǫ) < ǫ.

Since the non-negative constant U (f ) − L(f ) is strictly smaller than every positive number
ǫ, it means, by Theorem 2.10.4, that U (f ) − L(f ) = 0, so that f is integrable.

Example 7.1.11 Let


1, if x = n1 for some n ∈ N+ ;
f (x) =
0, otherwise.
R1
Then f is integrable over [0, 1] and 0 f = 0. Namely, let ǫ > 0. By Theorem 2.7.12,
there exists a positive integer p such that 1 < pǫ. Thus p1 < ǫ. Set d = 2p2 (p+1) 1
, x0 =
1 1
0, for k = 1, . . . , p, set x2k−1 = p−k+2 − d, x2k = p−k+2 + d, and x2p+1 = 1. Then
P = {x0 , x1 , . . . , x2p+1 } is a partition of [0, 1]. On each subinterval of the partition, the
infimum of all possible values of f on that subinterval is 0, so that L(f, P ) = 0. However,
sup{f (x) : x ∈ [x0 , x1 ]} = sup{f (x) : x ∈ [x2k−1 , x2k ]} = 1, and for k ≥ 1, sup{f (x) : x ∈
[x2k , x2k+1]} = 0. Thus

U (f, P ) − L(f, P ) = U (f, P )

p p
X X
= 1 · (x1 − x0 ) + 1 · (x2k − x2k−1 ) = x1 − x0 + 2d
k=1 k=1
1 1
= − d + 2dp < + 2pd
p+1 p+1
1 2p 1 1 1
= + 2 = + =
p + 1 2p (p + 1) p + 1 p(p + 1) p
< ǫ.
194 Chapter 7: Integration

Thus by the previous theorem, f is integrable over [0, 1].

The following theorem says that to compute (lower, upper) integrals of integrable
functions, we need not use all the possible partitions. It suffices to use, for example, only
partitions into equal-length subintervals. The second part of the theorem below gives a
formulation of integrals that looks very technical but is fundamental for applications (see
Section 7.6).

Theorem 7.1.12 Let f : [a, b] → R be bounded and integrable. For each real number
(r) (r) (r)
r > 0 let Pr = {x0 , x1 , . . . , xnr } be a partition of [a, b] such that each subinterval
(r) (r)
[xk−1 , xk ] has length at most r. Then
Z b
lim L(f, Pr ) = lim U (f, Pr ) = f.
r→0+ r→0+ a
(r) (r) (r)
Furthermore, if for each r > 0 and each k = 1, 2, . . . , nr , ck is arbitrary in [xk−1 , xk ],
then
nr
X Z b
(r) (r) (r)
lim f (ck )(xk − xk−1 ) = f.
r→0+ a
k=1

Proof. Let ǫ > 0. By Theorem 7.1.10, there exists a partition P of [a, b] such that
U (f, P ) − L(f, P ) < ǫ/2. Let P = {y0 , y1 , . . . , yn1 }. Since f is bounded, there exists a
positive real number M such that for all x ∈ [a, b], |f (x)| < M . Let r be a positive real
number that is at most 8Mǫn1 . The elements of P other than a and b may be contained in
two subintervals, so that at most 2n1 − 1 subintervals in Pr contain the n1 + 1 elements
of P . I will refer to these subintervals of Pr special. Let Pr = {x0 , x1 , . . . , xmr }. Then

L(f, P ∪ Pr ) ≥ L(f, Pr ) (by Theorem 7.1.6)

mr
X
= mk (f )(xk − xk+1 )
k=1
mr
X mr
X
= mk (f )(xk − xk+1 ) + mk (f )(xk − xk+1 )
k=1,special k=1,non-special
mr
X
≥ (−M )(xk − xk+1 )
k=1,special
mr
X
+ (inf of f on the corresp. subinterval of P ) (xk − xk+1 )
k=1,non-special
(since the non-special [xk−1 , xk ] is a subset of a subinterval in P )
mr
X
≥ (−M − (inf of f on the corresp. subinterval of P )) (xk − xk+1 )
k=1,special
 Section 7.1: Approximating areas 195
mr
X
+ (inf of f on the corresp. subinterval of P ) (xk − xk+1 )
k=1
(by adding a 0)
mr
X
≥ (−2M )(xk − xk+1 ) + L(f, P )
k=1,special

But the total length of the non-special subintervals is at most r(2n1 − 1) < 2rn1 < ǫ/4M,
so that for all r < 8Mǫn1 ,
ǫ ǫ
L(f, P ∪ Pr ) − L(f, P ) ≥ L(f, Pr ) − L(f, P ) ≥ (−2M ) =− .
4M 2
Furthermore, both L(f, P ) and the lower sum of its refinement L(f, P ∪ Pr ) are within ǫ/2
of L(f ), and since they are both less than or equal to L(f ), it follows that L(f, P ) and
L(f, P ∪ Pr ) are within ǫ/2 of each other. Thus L(f, Pr ) is within ǫ/2 of L(f ) as well.
Similarly, for all r sufficiently small, U (f, Pr ) is within ǫ/2 of U (f ) = L(f ). Hence
U (f, Pr ) − L(f, Pr ) < ǫ, which proves the first paragraph of the theorem.
Furthemore, by compatibility of order with multiplication by positive numbers and
addition, we have that for each r,
nr
X (r) (r) (r)
L(f, Pr ) ≤ f (ck )(xk − xk−1 ) ≤ U (f, Pr ),
k=1
Pnr (r) (r) (r) Rb
So that for all r sufficiently close to 0, k=1 f (ck )(xk − xk−1 ) is within ǫ of a f . Thus
Pnr (r) (r) (r) Rb
by the definition of limits, limr→0+ k=1 f (ck )(xk − xk−1 ) = a f .

Notation 7.1.13 In the definition of integral we sometimes write “dx” or “dt”, and some-
times we do not. There is no need to write “dx” when we are simply integrating a function f ,
Rb
as in “ a f ”: we seek the signed area determined by f over the domain from a to b. For
this it does not matter if we like to plug x or t or anything else into f . But when we
write “f (x)” rather than “f ”, then we add “dx”, and the reason is that f (x) is an element
of the codomain and is not a function. Why do we have to be pedantic? If x and t are
non-dependent variables, by the constant rule established above we then have
Z b
f (x) dt = f (x)(b − a),
a
and specifically, by geometric reasoning,
Z 4 Z 4
x dx = 8, x dt = 4x.
0 0
Thus writing “dx” versus “dt” is important, and omitting it can lead to confusion: is
the answer the constant 8, or is it 4x depending on x? Furthermore, if x and t are not
196 Chapter 7: Integration

independent, we can get further values too. Say if x = 3t, then

Z 4 Z 4
x dt = 3t dt = 24.
0 0
In short, use notation precisely.

Exercises for Section 7.1

7.1.1 Prove that if P = {x0 , . . . , xn } is a partition of [a, b] into n equal parts, then xk =
a + k b−a
n .

1, if x < 2;
7.1.2 Let f (x) = Let P be the partition of [0, 3] into two equal intervals
0, if x ≥ 2.
and Q the partition of [0, 3] into three equal intervals.
i) Compute L(f, P ) and U (f, P ).
ii) Compute L(f, Q) and U (f, Q).
iii) Compare L(f, P ) and L(f, Q). Why does this not contradict Theorem 7.1.6?
7.1.3 Let P be a partition of [a, b], and suppose that L(f, P ) = U (f, P ). Prove that f
must be constant on [a, b].
7.1.4 Let Pn be a partition of [−1, 1] into n equal parts. Let f (x) = 1 if x = 0 and
f (x) = 0 otherwise.
i) Graph f and conclude that f is not continuous at 0.
ii) Compute L(f, Pn) and U (f, Pn).
7.1.5 Use geometry to compute the following integrals:
Z a
i) f=
Zar p
ii) r 2 − x2 dx =
Z0 5
iii) (4x − 10) dx =
3 
Z 5  5, if x < −7;

2, if −7 ≤ x < 1;
iv) f , where f (x) =
−1  3x,
 if 1 ≤ x < 3;
9 − x, if 3 ≤ x.
7.1.6 Use geometry to compute the following integrals (t and x do not depend on each
other). Justify all work.
Z rp
i) r 2 − x2 dt =
Z0 5
ii) (4x − 10) dt =
3
 Section 7.2: Computing integrals from upper and lower sums 197

Z  5, if t < −7;
5 
2, if −7 ≤ t < 1;
iii) f , where f (t) =
−1 
 3t, if 1 ≤ t < 3;
9 − x, if 3 ≤ t.
7.1.7 In EasyLand they define Eantegrals as follows: first they divide the interval [a, b]
into n equal subintervals, so each has length b−a n . By Exercise 7.1.1, the kth subinterval
 b−a b−a
given in this way is a + (k − 1) n , a + k n . Rather than finding the minimum and
maximum of f on this subinterval, they simply Eapproximate f on the subinterval by
plugging in the right endpoint a + k b−a n , so that the signed area over the kth subinterval


b−a b−a
is Eapproximately f a + k n n . Thus the Eapproximate signed area of f over [a, b]
Pn

b−a b−a
via this partition is k=1 f a + k n n . If this sum has a limit as n goes to infinity,
then EasyLanders declare the Eantegral of f over [a, b] to be
b Xn  
 f = lim b − a b−a
f a+k .
a n→∞ n n
k=1
b
Suppose that f : [a, b] → R is integrable. Prove that the Eantegral  f exists and equals
Rb a
the integral a f .

1, if x is rational;
7.1.8 Use Eantegrability from the previous exercise. Let f (x) =
0, if x is irrational.
By Example 7.1.7 we know that f is not integrable over [0, 2].
2
i) Prove that f is Eantegrable over [0, 2] and find its Eantegral  f .
0
√2 2 √2 2 2
ii) Compute  f, √ f , and prove that  f + √ f 6=  f .
0 2 0 2 0

7.2 Computing integrals from upper and lower sums

The definition of integrals appears daunting: we seem to need to compute all the
possible lower sums to get at the lower integral, all the possible upper sums to get at the
upper integral, and only if the lower and upper integrals are the same do we have the precise
integral. In Example 7.1.5 in the previous section we have already seen that numerically
we can often compute the integral to within desired precision by taking finer and finer
partitions. In this section we compute some precise numerical values of integrals, and
without computing all the possible upper and lower sums. Admittedly, the computations
are time-consuming, but the reader is encouraged to read through them to get an idea of
what calculations are needed to follow the definition of integrals. In Section 7.4 we will see
very efficient shortcuts for computing integrals, but only for easy/good functions.
198 Chapter 7: Integration

Example 7.2.1 Let f (x) = x on [2, 6]. We know that the area under the curve between
Z 6
x = 2 and x = 6 is 16. Here we compute that indeed f = 16. For any positive integer n
2
let Pn = {x0 , . . . , xn } be the partition of [2, 6] into n equal parts. By Exercise 7.1.1,
xk = 2 + k n4 . Since f is increasing, on each subinterval [xk−1 , xk ] the minimum is xk−1
and the maximum is xk . Thus
n
X
U (f, Pn) = xk (xk − xk−1 )
k=1
Xn  
4 4
= 2+k
n n
k=1
Xn n  2
4 X 4
= 2 + k
n n
k=1 k=1
 2
4 n(n + 1) 4
=2 n+
n 2 n
8(n + 1)
=8+ .
n
It follows that for all n,
8(n + 1)
U (f ) = inf{U (f, P ) : P varies over partitions of [2, 6]} ≤ U (f, Pn) = 8 + ,
n
so that U (f ) ≤ 8 + 8 = 16. Similarly,
n
X
L(f, Pn) = xk−1 (xk − xk−1 )
k=1
Xn  
44
= 2 + (k − 1)
nn
k=1
Xn n  2
4 X 4
= 2 + (k − 1)
n n
k=1 k=1
 2
4 (n − 1)n 4
=2 n+
n 2 n
8(n − 1)
=8+ ,
n
and
L(f ) = sup{L(f, P ) : P varies over partitions of [2, 6]} ≥ sup{L(f, Pn) : n ∈ N>0 } = 16.
All together this says that

16 ≤ L(f ) ≤ U (f ) ≤ 16,
 Section 7.2: Computing integrals from upper and lower sums 199
R6
so that L(f ) = U (f ) = 16, and finally that 2
f = 16.
Note that we did not compute all the possible lower and upper sums, but we computed
enough of them. We knew that we computed enough of them because as n goes large,
sup{L(f, Pn) : n} = inf{U (f, Pn) : n}.

Example 7.2.2 We compute the integral for f (x) = x2 over [1, 7]. For any positive inte-
ger n let Pn = {x0 , . . . , xn } be the partition of [1, 7] into n equal parts. By Exercise 7.1.1,
xk = 1 + k n6 . Since f is increasing on [1, 7], on each subinterval [xk−1 , xk ] the minimum is
x2k−1 and the maximum is x2k . Thus, by using Exercise 1.5.1 in one of the steps below:
n
X
U (f, Pn) = x2k (xk − xk−1 )
k=1
Xn  2
6 6
= 1+k
n n
k=1
Xn  
12 2 36 6
= 1+k +k 2
n n n
k=1
Xn n n  3
6 X 12 6 X 2 6
= + k + k
n n n n
k=1 k=1 k=1
 3
6 n(n + 1) 12 6 n(n + 1)(2n + 1) 6
= n+ +
n 2 n n 6 n
36(n + 1) 36(n + 1)(2n + 1)
=6+ + ,
n n2
so that U (f ) ≤ inf{U (f, Pn) : n} = 6 + 36 + 72 = 114. Similarly,
L(f ) ≥ L(f, Pn) = 114,
R7
whence 114 ≤ L(f ) ≤ U (f ) ≤ 114 and 1 f = 114.
R2√
Example 7.2.3 The goal of this exercise is to compute 0 x dx. For the first attempt,
let Pn = {x0 , . . . , xn } be the partition of [0, 2] into n equal intervals. By Exercise 7.1.1,
xk = 2k
n
. The square root function is increasing, so that
n
X √
U (f, Pn) = xk (xk − xk−1 )
k=1
Xn r
2k 2
=
n n
k=1
 3/2 X
n
2 √
= k.
n
k=1
200 Chapter 7: Integration
Pn √
But now we are stuck: we have no simplification of k=1 k, and we have no other

Pn √
2 3/2
immediate tricks to compute inf{ n k=1 k : n}.
But it is possible to compute enough upper and lower sums for this function to get the
2·(n−1)2 2·n2
integral. Namely, for each positive integer n let Qn = {0, n22 , 2·4 2·9
n2 , n2 , . . . , n2 , n2 =
2k2 √
2}. This is a partition of [0, 2] into n (unequal) parts, with xk = n2 . Since is an
increasing function, on each subinterval [xk−1 , xk ] the minimum is achieved at xk−1 and
the maximum at xk , so that
X n √  
2 k 2k 2 2(k − 1)2
U (f, Qn) = −
n n2 n2
k=1
√ n
2 2X

= 3 k k 2 − (k − 1)2
n
k=1
√ n
2 2X
= 3 k (2k − 1)
n
k=1
√ n √ n
4 2X 2 2 2X
= 3 k − 3 k
n n
k=1 k=1
√ √
4 2 n(n + 1)(2n + 1) 2 2 n(n + 1)
= 3 − 3 (by Exercise 1.5.1)
n√ 6 n 2
2 2
= (n + 1)(2(2n + 1) − 3) (by factoring)
6n2
√
2
= (n + 1)(4n − 1).
3n2
√ √
Thus U (f ) ≤ inf{U (f, Pn)√: n} = inf{ 3n22 (n + 1)(4n − 1) : n} = 4 2
3 . Similarly, L(f ) ≥
√
4 2
R2√
3 , so that 0 x dx = 4 3 2 .

Exercises for Section 7.2

R2
7.2.1 Mimic examples in this section to compute x3 dx. (You may need Exercise 1.5.2.)
0
R2 √
7.2.2 Mimic examples in this section to compute 0 3 x dx.

7.3 What functions are integrable?

In this section, we continue to assume that the domains and codomains of all functions
are subsets of R.
 Section 7.3: What functions are integrable? 201

Theorem 7.3.1 Every continuous real-valued function on [a, b] is integrable over [a, b],
where a, b ∈ R with a < b.

Proof. Let f : [a, b] → R be continuous. We need to prove that L(f ) = U (f ). By

Theorem 2.10.4 it suffices to prove that for all ǫ > 0, U (f ) − L(f ) ≤ ǫ.
So let ǫ > 0. By Theorem 5.5.4, f is uniformly continuous, so there exists δ > 0 such
that for all x, c ∈ [a, b], if |x−c| < δ then |f (x)−f (c)| < ǫ/(b−a). Let P = {x0 , x1 , . . . , xn}
be a partition of [a, b] such that for all k = 1, . . . , n, xk − xk−1 < δ. (For example, this can
be accomplished as follows: by Theorem 2.7.12, there exists n ∈ N+ such that (b − a) < nδ,
and then P can be taken to be the partition of [a, b] into n equal parts.) Then
U (f, P ) − L(f, P )
Xn
= (sup{f (x) : x ∈ [xk−1 , xk ]} − inf{f (x) : x ∈ [xk−1 , xk ]}) (xk − xk−1 )
k=1
Xn
ǫ
≤ (xk − xk−1 ) (by uniform continuity since xk − xk−1 < δ)
(b − a)
k=1
X n
ǫ
= (xk − xk−1 )
(b − a)
k=1
= ǫ.
But U (f ) ≤ U (f, P ) and L(f ) ≥ L(f, P ), so that 0 ≤ U (f ) −L(f ) ≤ U (f, P ) −L(f, P ) ≤ ǫ.
Thus U (f ) = L(f ) by Theorem 2.10.4, so that f is integrable over [a, b].
Both the proof above as well as Theorem 7.1.12 prove that for a continuous function f
on [a, b],
Z b Xnr
(r) (r) (r)
f = lim f (ck )(xk − xk−1 ),
a r→0+
k=1
(r) (r) (r)
where for each r > 0, {x0 , x1 , . . . , xnr } is a partition of [a, b] into subintervals of length
at most r, and for each k = 1, . . . , nr , ck ∈ [xk−1 , xk ].

Theorem 7.3.2 Let f : [a, b] → R be bounded. Let S = {s0 , s1 , . . . , sm } be a finite subset

of [a, b] with s0 = a < s1 < · · · < sm = b. Suppose that for all d > 0 and all k f is
integrable over [sk−1 + d, sk − d]. (By the previous theorem, Theorem 7.3.1, any f that is
continuous at all x ∈ [a, b] \ S satisfies this hypothesis.) Then f is integrable over [a, b],
and Z b Xm Z sk
f= f.
a k=1 sk−1

Proof. Let ǫ > 0. Since f is bounded, there exists a positive real number M such that
for all x ∈ [a, b], |f (x)| < M . Let e = 31 min{s1 − s0 , s2 − s1 , . . . , sm − sm−1 } and d =
202 Chapter 7: Integration

min{e, ǫ/(4M (m + 1)(2m + 1))}. By assumption, for each k = 1, . . . , m − 1, f is integrable

on the interval [sk−1 + d, sk − d]. Thus by Theorem 7.1.10, there exists a partition Pk of
[sk−1 + d, sk − d] such that U (f, Pk ) − L(f, Pk ) < ǫ/(2m + 1). Now let P = {a} ∪ P1 ∪ P2 ∪
· · · ∪ Pm−1 ∪ {b}. Then P is a partition of [a, b], and
m
X
U (f, P ) − L(f, P ) = (U (f, P ∩ [sk−1 + d, sk − d]) − L(f, P ∩ [sk−1 + d, sk − d]))
k=1
+ (U (f, P ∩ [a, a + d]) − L(f, P ∩ [a, a + d]))
m−1
X
+ (U (f, P ∩ [sk − d, sk + d]) − L(f, P ∩ [sk − d, sk + d]))
k=1
+ (U (f, P ∩ [b − d, b]) − L(f, P ∩ [b − d, b]))
m
X m−1
X
≤ (U (f, Pk ) − L(f, Pk )) + 2M · 2d + 2M · 2d + 2M · 2d
k=1 k=1
Xm
ǫ
< + 4M (m + 1)d
2m + 1
k=1
< ǫ.
Rb Pm R sk
Thus by Theorem 7.3.1, f is integrable. Furthermore, a f − k=1 sk−1 f is bounded
Pm Pm
above by U (f, P ) − k=1 L(f, Pk ) and below by L(f, P ) − k=1 U (f, Pk ), and the upper
Rb P R sk
and lower bounds are within 2ǫ of 0, so that a f − m k=1 sk−1 f is within 2ǫ of 0. Since ǫ
Rb Pm R sk
is arbitrary, it follows by Theorem 2.10.4 that a f = k=1 sk−1 f.
Ra
Notation 7.3.3 What could possibly be the meaning of f if a < b? In our definition
b Rb
of integrals, all partitions started from a smaller a to a larger b to get a f . If we did
reverse b and a, then the widths of the subintervals in each partition would be negative
(xk − xk−1 = −(xk−1 − xk )), so that all the partial sums and both the lower and the upper
integrals would get the negative value. Thus it seems reasonable to declare
Z a Z b
f =− f.
b a
In fact, this is exactly what makes Theorem 7.3.2 work without any order assumptions
Rb Rc Rb
on the sk . For example, if a < c < b, by Theorem 7.3.2, a f = a f + c f , whence
Rc Rb Rb Rb Rc
a
f = a f − c f = a f + b f.

Theorem 7.3.4 Suppose that f and g are integrable over [a, b], and that c ∈ R. Then
Rb Rb Rb
f + cg is integrable over [a, b] and a (f + cg) = a f + c a g.

Proof. We first prove that L(cg) = U (cg) = cL(g) = cU (g). If c ≥ 0, then

L(cg) = sup{L(cg, P ) : P a partition of [a, b]}
 Section 7.3: What functions are integrable? 203

= sup{cL(g, P ) : P a partition of [a, b]}

= c sup{L(g, P ) : P a partition of [a, b]}
= cL(g)
= cU (g)
= c inf{U (g, P ) : P a partition of [a, b]}
= inf{cU (g, P ) : P a partition of [a, b]}
= inf{U (cg, P ) : P a partition of [a, b]}
= U (cg),
and if c < 0, then

L(cg) = sup{L(cg, P ) : P a partition of [a, b]}

= sup{cU (g, P ) : P a partition of [a, b]}
= c inf{U (g, P ) : P a partition of [a, b]}
= cU (g)
= cL(g)
= c inf{L(g, P ) : P a partition of [a, b]}
= sup{cL(g, P ) : P a partition of [a, b]}
= inf{U (cg, P ) : P a partition of [a, b]}
= U (cg).
Rb Rb
This proves that cg is integrable with a
(cg) = c a
g.

Let ǫ > 0. By integrability of f and cg there exist partitions P, Q of [a, b] such that
U (f, P ) − L(f, P ) < ǫ/2 and U (cg, Q) − L(cg, Q) < ǫ/2. Let R = P ∪ Q. Then R is a
partition of [a, b], and by Theorem 7.1.6, U (f, R)−L(f, R) < ǫ/2 and U (cg, R)−L(cg, R) <
ǫ/2. By Exercise 2.7.11, for every partition of [a, b], and in particular for the partition R,
L(f + cg, R) ≥ L(f, R) + L(cg, R), and U (f + cg, R) ≤ U (f, R) + U (cg, R). Then

0 ≤ U (f + cg) − L(f + cg)

≤ U (f + cg, R) − L(f + cg, R)
≤ U (f, R) + U (cg, R) − L(f, R) − L(cg, R)
< ǫ.
Thus U (f + cg) − L(f + cg) = 0 by Theorem 2.10.4, and so f + cg is integrable. The
inequalities L(f + cg, R) ≥ L(f, R) + L(cg, R), and U (f + cg, R) ≤ U (f, R) + U (cg, R)
furthermore prove that L(f )+L(cg) ≤ L(f +cg) ≤ U (f +cg) ≤ U (f )+U (cg) = L(f )+L(cg),
204 Chapter 7: Integration

so that finally
Z b Z b Z b Z b Z b
(f + cg) = f+ (cg) = f +c g.
a a a a a

Theorem 7.3.5 Let a, b ∈ R with a < b. Let f, g : [a, b] → R be integrable functions such
that f (x) ≤ g(x) for all x ∈ [a, b]. Then
Z b Z b
f≤ g.
a a

Here is a picture that illustrates this theorem: the values of g are at each point in the
domain greater than or equal to the values of f , and the area under the graph of g is larger
than the area under the graph of f :
y = g(x)
y = f (x)

a b x

Proof. By assumption on every subinterval I of [a, b], inf{f (x) : x ∈ I} ≤ inf{g(x) : x ∈ I}.
Thus for all partitions P of [a, b], L(f, P ) ≤ L(g, P ). Hence L(f ) ≤ L(g), and since f and
Rb Rb
g are integrable, this says that a f ≤ a g.

Exercises for Section 7.3

R1 R3 R3 R2
7.3.1 The following is known: 0 f = 5, 2 f = 6, 0 f = 15, 1 g = −3. Compute and
R2 R3 R1
justify 1 (3f − 4g), 1 5f , 2 3g.

x, if x is rational;
7.3.2 Let f (x) =
0, if x is irrational.
i) Prove that f is not integrable over [0, 1].
ii) Does this contradict Theorem 7.3.2? Justify.
Ra
7.3.3 Let f : [−a, a] → R be an integrable odd function. Prove that −a f = 0.
Ra Ra
7.3.4 Let f : [−a, a] → R be an integrable even function. Prove that −a f = 2 0 f .
7.3.5 Let f : [a, b] → R be piecewise continuous as in Theorem 7.3.2.
i) Prove that |f | isintegrable over [a, b].
R b  R b
ii) Prove that  a f  ≤ a |f |.
 Section 7.4: The Fundamental theorem of calculus 205

7.3.6 Find a function f : [0, 1] → R that is not integrable over [0, 1] but such that |f | is
integrable over [0, 1].
7.3.7 So far we have seen that every differentiable function is continuous and that every
continuous function is integrable.
i) Give an example of a continuous function that is not differentiable.
ii) Give an example of an integrable function that is not continuous.
*7.3.8 Let f : [a, b] → R be monotone. Prove that f is integrable over [a, b].

7.4 The Fundamental theorem of calculus

Despite first appearances, it turns out that integration and differentiation are related.
For this we have two versions of the Fundamental theorem of calculus.

Theorem 7.4.1 (The Fundamental theorem of calculus, I) Let f, g : [a, b] → R such

that f is integrable over [a, b] and g is differentiable with g ′ = f . Then
Z b
f = g(b) − g(a).
a

Proof. Let P = {x0 , x1 , . . . , xn } be a partition of [a, b]. Since g is differentiable on [a, b], it
is continuous on each [xk−1 , xk ] and differentiable on each (xk−1 , xk ). Thus by the Mean
value theorem (Theorem 6.3.4), there exists ck ∈ (xk−1 , xk ) such that f (ck ) = g ′ (ck ) =
g(xk )−g(xk−1 )
xk −xk−1 . By the definition of lower and upper sums,
n
X
L(f, P ) ≤ f (ck )(xk − xk−1 ) ≤ U (f, P ).
k=1

But
n
X n
X g(xk ) − g(xk−1 )
f (ck )(xk − xk−1 ) = (xk − xk−1 )
xk − xk−1
k=1 k=1
Xn
= (g(xk ) − g(xk−1 ))
k=1
= g(xn ) − g(x0 )
= g(b) − g(a),

so that L(f, P ) ≤ g(b) − g(a) ≤ U (f, P ), whence

L(f ) = sup{L(f, P ) : P a partition of [a, b]}
≤ g(b) − g(a)
206 Chapter 7: Integration

≤ inf{U (f, P ) : P a partition of [a, b]}

= U (f ).
Since f is integrable over [a, b], by definition L(f ) = U (f ), and so all inequalities above
Rb
have to be equalities, so that necessarily a f = g(b) − g(a).
The general notation for applying Theorem 7.4.1 is as follows: if g ′ = f , then
Z b b

f = g(x) = g(b) − g(a).
a a
′
For example, assuming that cos (x) = − sin(x), we have that
Z π π

sin x dx = − cos(x) = − cos(π) − (− cos(0)) = −(−1) + 1 = 2.
0 0
If we instead had to compute this integral with upper and lower sums, it would take us a
lot longer and a lot more effort to come up with the answer 2.
In general, upper and lower sums and integrals are time-consuming and we want to
avoid them if possible. The fundamental theorem of calculus that we just proved enables
us to do that for many functions: to integrate f over [a, b] one needs to find g with g ′ = f .
Such g is called an antiderivative of f . For example, if r is a rational number different
r+1
from −1, then by the power rule (Theorem 6.2.10), an antiderivative of xr is xr+1 . By the
r+1
scalar rule for derivatives, for any constant C, xr+1 + C is also an antiderivative. It does
not matter which antiderivative we choose to compute the integral:
Z b  r+1   r+1 
r b a br+1 ar+1
x dx = +C − +C = − ,
a r+1 r+1 r+1 r+1
so that the choice of the antiderivative is irrelevant.

Definition 7.4.2 (Indefinite integral) If g is an antiderivative of f , we write also

Z
f (x) dx = g(x) + C,

where C stands for an arbitrary constant.

R R R R
For example, 3x2 dx = x3 + C, 3 dx = 3x + C, t dx = tx + C, x dx = 12 x2 + C,
and so on. (Study the differences and similarities of the last three.)
So far we have seen 2x for rational exponents x. Exercise 5.4.7 also allows real ex-
ponents, and proves that this function of x is continuous. Thus by Theorem 7.3.1 this
R
function is integrable. We do not yet know 2x dx, but in Theorem 7.5.5 we will see that
R x R 2
2 dx = ln12 2x + C. For 2(x ) dx instead, you and I do not know an antiderivative,
we will not know one by the end of the course, and there actually is no “closed-form”
antiderivative. This fact is due to a theory of Joseph Liouville (1809–1882). What is the
 Section 7.4: The Fundamental theorem of calculus 207

meaning of “closed-form”? Here is an oblique answer: Exercise 9.8.5 claims that there ex-
2
ists an infinite power series (sum of infinitely many terms) that is an antiderivative of 2(x ) .
2
Precisely because of this infinite sum nature, the values of any antiderivative of 2(x ) cannot
be computed precisely, only approximately. Furthermore, according to Liouville’s theory,
that infinite sum cannot be expressed in terms of the more familiar standard functions,
and neither can any other expression for an antiderivative. It is in this sense that we say
2
that 2(x ) does not have a “closed-form” antiderivative.
(It is a fact that in the ocean of all functions, those for which there is a “closed-form”
antiderivative form only a tiny droplet.)
At this point we know very few methods for computing antiderivatives. We will in
time build up rigorously a bigger stash of functions: see the next section (Section 7.5) and
the chapter on power series (Section 9.4).
The simplest method for finding more antiderivatives is to first find a differentiable
function and compute its derivative, and voilá, the original function is an antiderivative of
its derivative. For example, by the chain and power rules, (x2 + 3x)100 is an antiderivative
of 100(x2 + 3x)99 (2x + 3).

Theorem 7.4.3 (The Fundamental theorem of calculus, II) Let f : [a, b] → R be

continuous. Then for all x ∈ [a, b], f is integrable over [a, x], and the function g : [a, b] → R
Rx
given by g(x) = a f is differentiable on (a, b) with
Z x
d
f = f (x).
dx a

Proof. Since f is continuous over [a, b], it is continuous over [a, x], so that by Theorem 7.3.1,
f is integrable over [a, x]. Thus g is a well-defined function. Let c ∈ (a, b). We will prove
that g is differentiable at c.
Let ǫ > 0. By continuity of f at c, there exists δ > 0 such that for all x ∈ [a, b], if
|x − c| < δ then |f (x) − f (c)| < ǫ. Thus on [c − δ, c + δ] ∩ [a, b], f (c) − ǫ < f (x) < f (c) + ǫ,
so that by Theorem 7.3.5,
Z max{x,c} Z max{x,c} Z max{x,c}
(f (c) − ǫ) ≤ f≤ (f (c) + ǫ).
min{x,c} min{x,c} min{x,c}

But by integrals of constant functions,

Z max{x,c}
(f (c) ± ǫ) = (max{x, c} − min{x, c})(f (c) ± ǫ) = |x − c|(f (c) ± ǫ).
min{x,c}

Thus
Z max{x,c}
|x − c|(f (c) − ǫ) ≤ f ≤ |x − c|(f (c) + ǫ).
min{x,c}
208 Chapter 7: Integration

If x 6= c, dividing by |x − c| and rewriting the middle term says that

Rx
f
f (c) − ǫ ≤ c ≤ f (c) + ǫ,
x−c
Rx R x 
f  f 
whence −ǫ ≤ x−c −f (c) ≤ ǫ, and  x−c − f (c) < ǫ. Then for all x ∈ [a, b], if 0 < |x−c| < δ,
c  c

then
   R x R 

 g(x) − g(c)   a f − ac f 
 
− f (c)  =  − f (c) 
 x−c  x−c 
 Rx 
 
 c f 
= − f (c)  (by Theorem 7.3.2 and Notation 7.3.3)
 x−c 
< ǫ.
g(x) − g(c)
Thus lim exists and equals f (c), i.e., g ′ (c) = f (c).
x→c x−c
Rx
It is probably a good idea to review the notation again. The integral a f can also be
written as Z x Z x Z x
f= f (t) dt = f (z) dz.
a a a
This is a function of x because x appears in the bound of the domain of integration. Note
Rx
similarly that a f (t) dz are functions of t and x but not of z. Thus by the Fundamental
theorem of calculus, II,
Z x Z x
d d
f = f (x), and f (t) dz = f (t).
dx a dx a
Rx
Do not write “ a f (x) dx”: this is trying to say that x varies from a to x, so one
occurrence of the letter x is constant and the other occurrence varies from a to that
constant, which mixes up the symbols too much!

So far we have defined integrals of real-valued functions. By Theorem 7.4.1, if g ′ = f ,

Rb Rb
then a f = g(b) − g(a), i.e., a g ′ = g(b) − g(a). If g is complex-valued, we know that
Rb
g ′ = (Re g)′ + i(Im g)′ , so that it would make sense to define a g ′ as the integral of (Re g)′
plus i times the integral of (Im g)′ . Indeed, this is the definition:

Definition 7.4.4 Let f : [a, b] → C be a function such that Re f and Im f are integrable
over [a, b]. Define the integral of f over [a, b] to be
Z b Z b Z b
f= Re f + i Im f.
a a a

The following are then immediate generalizations of the two versions of the fundamen-
tal theorem of calculus:
 Section 7.4: The Fundamental theorem of calculus 209

Theorem 7.4.5 (The Fundamental theorem of calculus, I, for complex-valued

functions) Let f, g : [a, b] → C such that f is continuous and g is differentiable with
g ′ = f . Then
Z b
f = g(b) − g(a).
a

Theorem 7.4.6 (The Fundamental theorem of calculus, II, for complex-valued

functions) Let f : [a, b] → C be continuous. Then for all x ∈ [a, b], f is integrable over
Rx
[a, x], and the function g : [a, b] → R given by g(x) = a f is differentiable on (a, x) with
Z x
d
f = f (x).
dx a

Exercises for Section 7.4

7.4.1 Compute and justify or verify the integrals below. Clever guessing is fine.
Z 1
i) x3 dx =
0
Z 1
√
ii) x dx =
0
Z 1
iii) 16x(x2 + 4)7 dx =
Z0 1
√
iv) (3 x + 4)(x3/2 + 2x + 3)7 dx =
0
7.4.2 Compute the integrals below, assuming that t and x do not depend on each other.
Z 1
i) x3 dx =
Z0 1
ii) x3 dt =
Z0 x
iii) x3 dt =
0
7.4.3 Below t and x do not depend on each other. Compute the following derivatives,
possibly using
Z xTheorem 7.4.3.
d
i) t3 dt =
dx Z0
x
d
ii) x3 dt =
dx 0
Z t
d
iii) x3 dx =
dx 0
Z t
d
iv) t3 dx =
dx 0
210 Chapter 7: Integration

7.4.4 Suppose that f : [a, b] → R+ is continuous and that f (c) > 0 for some c ∈ [a, b].
Rb
Prove that a f > 0.
7.4.5 (Integration by substitution) By the chain rule for differentiation, (f ◦ g)′ (x) =
f ′ (g(x))g ′ (x).
Rb
i) Prove that a f ′ (g(x))g ′ (x) dx = f (g(b)) − f (g(a)).
R
ii) Prove that f ′ (g(x))g ′ (x) dx = f (g(x)) + C.
iii) Compute the following integrals: explicitly state f, g in applying this rule:
Z 3
(2x − 4)10 dx =
Z2 3
4x + 3
dx =
1 (2x2 + 3x)10
Z 3
4x + 3
p dx =
1 2x2 + 3x)10
Z 3 p
3
(8x + 6) 2x2 + 3x dx =
1
7.4.6 (Integration by parts) By the product rule for differentiation, (f · g)′ (x) =
f ′ (x)g(x) + f (x)g ′ (x).
Rb Rb
i) Prove that a f ′ (x)g(x) dx = f (b)g(b) − f (a)g(a) − a f (x)g ′ (x) dx.
R R
ii) Prove that f ′ (x)g(x) dx = f (x)g(x) − f (x)g ′ (x) dx.
iii) Compute the following integrals: explicitly state f, g in applying this rule:
Z 1
(4x + 3)(5x + 1)10 dx =
Z−1
1
4x + 3
√ dx =
−1 2x + 4

7.4.7 Prove by integration by parts the following improper integral value for a non-negative
integer n:
Z 1
1
(−x2n ln x)dx = .
0 (2n + 1)2

7.4.8 Compute the following derivatives. (Hint: the Fundamental theorem of calculus and
the chain rule.)
Z 3x q
d √
i) t4 + 5 t dt.
dx 2 √
Z x √
d t+5 t
ii) dt.
dx −x2 t1 00 − 2t5 0 + t7 − 2
Z h(x)
d
iii) f (t) dt.
dx g(x)
Z h(x)
d
iv) f.
dx g(x)
 Section 7.4: The Fundamental theorem of calculus 211

7.4.9 Consult Exercise 7.4.13.

R∞
i) Compute 1 x12 dx.
R∞
ii) For any rational number p < −1, compute 1 xp dx. (The same is true for real
p < −1, but we have not developed enough properties for such functions.)
7.4.10 Consult Exercise 7.4.18.
R1
i) Compute 0 √1x dx.
R1
ii) For any rational number p > −1, compute 0 xp dx. (The same is true for real
p > −1, but we have not developed enough properties for such functions.)
7.4.11 (Mean value theorem for integrals) Let f : [a, b] → R be continuous. Prove
that there exists c ∈ (a, b) such that
Z b
1
f (c) = f.
b−a a
7.4.12 Let f have continuous derivatives of order up to n + 1 on the interval [a, b].
i) Justify how for any x in [a, b],
Z x
f (x) = f (a) + f ′ (t) dt.
a

ii) Integrate the integral above by parts, and rewrite, to get that
Z x
′
f (x) = f (a) + (x − a)f (a) + (x − t)f ′′ (t) dt.
a

iii) Use induction, integration by parts, and rewritings, to get that

Z x
f ′ (a) f (n) (a) n (x − t)n (n+1)
f (x) = f (a) + (x − a) + · · · + (x − a) + f (t) dt.
1! n! a n!
iv) Say why you cannot apply the Fundamental theorem of calculus, II, to compute
d
Rx n+1 (n+1)
dx a (x − t) f (t) dt.
v) (Taylor’ remainder formula in integral form) Consult Section 6.5 for Taylor poly-
Rx n
nomials to prove that f (x) = Tn,f,a (x) + a (x−t)
n!
f (n+1) (t) dt.
7.4.13 (Improper integral, unbounded domain) Let f : [a, ∞) → R be continuous.
i) Discuss how our construction/definition of integrals fails when the domain is not
bounded.
RN
ii) Prove that for all N ∈ [a, ∞), a f exists.
RN
iii) If limN →∞ a f exists, we call the limit the (improper) integral of f over [a, ∞).
R∞
We denote it a f . Observe that this is a limit of limits. Similarly formulate the
Rb
definition of −∞ g for a continuous function g : (−∞, b] → R.

3x + 1, if x < 10; R∞
iv) Let f (x) = . Compute and justify 0 f .
0, otherwise.
212 Chapter 7: Integration

7.4.14 Work out Exercises 7.3.3, 7.3.4, and 7.3.5 for complex-valued functions.
R∞
7.4.15 Let p > 1. Prove that 1 x1p dx exists, and compute it. Comment on what happens
to the integral if p ≤ 1.
7.4.16 (Improper integral, unbounded domain) Let f : R → R be bounded such
Rc R∞
that for some c ∈ R, −∞ f and c f exist in the sense of Exercise 7.4.13.
Re R∞
i) Prove that for all e ∈ R, −∞ f and e f exist and that
Z c Z ∞ Z e Z ∞
f+ f= f+ f.
−∞ c −∞ e
R∞
We denote this common value as −∞ f .
R∞ RN
ii) Prove that −∞ f = limN →∞ −N f .
7.4.17 Let f : R → R over R. This exercise is meant to show that integrability of f
RN
over R cannot be simply defined as the existence of the limit limN →∞ −N f . Namely, let

1, if ⌊x⌋ (the floor of x) is even;
f (x) =
−1, otherwise.
i) Sketch the graph of this function.
RN
ii) Prove that for every positive real number N , −N f = 0.
iii) Prove that f is not integrable over [0, ∞) or over (−∞, 0) in the sense of Exer-
cise 7.4.13.
7.4.18 (Improper integral, unbounded function) Let f : (a, b] → R be continuous.
i) Discuss how our construction/definition of integrals fails when the domain does
not include the boundaries of the domain.
Rb
ii) Prove that for all N ∈ (a, b), N f exists.
Rb
iii) If limN →a+ N f exists, we call this limit the (improper) integral of f over [a, b],
Rb Rb
and we denote it a f . Similarly formulate a f if the domain of f is [a, b) or (a, b).
Rb
iv) Let f : (a, b) → R be given by f (x) = 1. Prove that a f = b − a.

7.5 Natural logarithm and the exponential functions

The function that takes a non-zero x to 1/x is continuous everywhere on its domain
since it is a rational function. Thus by Theorem 7.3.1 and Notation 7.3.3, for all x > 0,
Rx 1
1 x
dx is well-defined. This function has a familiar name:
 Section 7.5: Natural logarithm and the exponential functions 213

Definition 7.5.1 The natural logarithm is the function

Z x
1
ln x = dt
1 t
for all x > 0.

We prove below all the familiar properties of this familiar function.

Remark 7.5.2
R1
(1) ln 1 = 1 1t dt = 0.
Rx Rx
(2) By geometry, for x > 1, ln x = 1 1t dt > 0, and for x ∈ (0, 1), ln x = 1 1t dt =
R1
− x 1t dt < 0.
(3) By the Fundamental theorem of calculus (Theorem 7.4.3), for all b ∈ R+ , ln is
differentiable on (0, b), so that ln is differentiable on R+ . Furthermore, ln′ (x) = x1 .
(4) ln is continuous (since it is differentiable) on R+ .
(5) The derivative of ln is always positive. Thus by Theorem 6.3.5, ln is everywhere
increasing.
(6) Let c ∈ R+ , and set g(x) = ln(cx). By the chain rule, g is differentiable, and
1
g ′ (x) = cx c = x1 = ln′ (x). Thus the function g − ln has constant derivative 0. It
follows by Theorem 6.3.5 that g − ln is a constant function. Hence for all x ∈ R+ ,

ln(cx) − ln(x) = g(x) − ln(x) = g(1) − ln(1) = ln(c) − 0 = ln(c).

This proves that for all c, x ∈ R+ ,

ln(cx) = ln(c) + ln(x).

(7) By the previous part, for all c, x ∈ R+ ,

c
ln = ln(c) − ln(x).
x
(8) For all non-negative integers n and all c ∈ R+ , ln(cn ) = n ln(c). We prove this
by mathematical induction. If n = 0, then ln(cn ) = ln(1) = 0 = 0 ln c = n ln c.
Now suppose that equality holds for some n − 1. Then ln(cn ) = ln(cn−1 c) =
ln(cn−1 ) + ln(c) by what we have already established, so that by the induction
assumption ln(cn ) = (n − 1) ln(c) + ln(c) = n ln(c).
(9) For all rational numbers r and all c ∈ R+ , ln(cr ) = r ln(c). Here is a proof. We
have proved this result if r is a non-negative integer. If r is a negative integer,
then −r is a positive integer, so that by the previous case, ln(cr ) = ln(1/c−r ) =
ln(1) − ln(c−r ) = 0 − (−r) ln(c) = r ln(c), which proves the claim for all integers.
Now write r = m r
n for some integers m, n with n 6= 0. Then n ln(c ) = n ln(c
m/n
)=
m r m
ln(c ) = m ln(c), so that ln(c ) = n ln(c) = r ln(c).
214 Chapter 7: Integration

(10) The range of ln is R = (−∞, ∞). Here is a proof. By geometry, ln(0.5) < 0 < ln(2).
Let y ∈ R+ . By Theorem 2.7.12, there exists n ∈ N+ such that y < n ln(2). Hence
ln 1 = 0 < y < n ln(2) = ln(2n ), so that since ln is continuous, by the Intermediate
value theorem (Theorem 5.3.1), there exists x ∈ (1, 2n) such that ln(x) = y. If
y ∈ R− , then by the just proved we have that −y = ln(x) for some x ∈ R+ , so
that y = − ln(x) = ln(x−1 ). Finally, 0 = ln(1). Thus every real number is in the
range of ln.
(11) Thus ln : R+ → R is a strictly increasing continuous and surjective function. Thus
by Theorem 2.8.4, ln has an inverse ln−1 : R → R+ . By Theorem 5.3.4, ln−1 is
increasing and continuous.
(12) By Theorem 6.2.7, the derivative of ln−1 is
1
(ln−1 )′ (x) = ′ −1 = ln−1 (x).
ln (ln (x))
(13) For all x, y ∈ R,
  −1 
ln−1 (x) −1 ln (x) −1 −1

−1



−1 = ln ln −1 = ln ln ln (x) − ln ln (y) = ln−1(x − y) .
ln (y) ln (y)
We have proved that for all c ∈ R+ and r ∈ Q, ln−1 (r ln(c)) = ln−1 (ln(cr )) = cr , and
we have proved that for all r ∈ R, ln−1 (r ln(c)) is well-defined. This allows us to define
exponentiation with real (not just rational) exponents:

Definition 7.5.3 Let c ∈ R+ and r ∈ R. Set

cr = ln−1 (r ln(c)).

This definition gives rise to two functions:

(1) The generalized power function with exponent r when c varies and r is con-
stant;
(2) The exponential function with base c when r varies and c is constant.

Theorem 7.5.4 Let r ∈ R. The function f : R+ → R+ given by f (x) = xr is differen-

tiable, with f ′ (x) = rxr−1 . This function is increasing if r > 0 and decreasing if r < 0.

Proof. By definition, f (x) = ln−1 (r ln(x)), which is differentiable by the chain and scalar
rules and the fact that ln and its inverse are differentiable. Furthermore, the derivative is
−1
f ′ (x) = ln−1 (r ln(x)) · xr = r lnln−1(r(ln(x))
ln(x))
= r ln−1 (r ln(x) − ln(x)) = r ln−1 ((r − 1) ln(x)) =
rxr−1 . The monotone properties then follow from Theorem 6.3.5.
 Section 7.5: Natural logarithm and the exponential functions 215

Theorem 7.5.5 Let c ∈ R+ . The function f : R → R+ given by f (x) = cx is differentiable,

and f ′ (x) = (ln(c))cx . This function is increasing if c > 1 and decreasing if c ∈ (0, 1).

Proof. By definition, f (x) = ln−1 (x ln(c)), which is differentiable by the chain and scalar
rules and the fact that ln−1 is differentiable. Furthermore, the derivative is f ′ (x) =
ln−1 (x ln(c)) · ln(c) = f (x) · ln(c) = (ln(c))cx . The monotone properties then follow from
Theorem 6.3.5.
We next give a more concrete form to ln−1 .

Definition 7.5.6 Let e = ln−1 (1) (so that ln(e) = 1). The constant e is called Euler’s
constant.

Since ln(e) = 1 > 0 = ln(1), by the increasing property of ln it follows that e > 1.
Now let f (x) = 1/x. This function is non-negative on [1, ∞). If P is a partition of
[1, 3] into 5 equal parts, then L(f, P ) ∼ = 0.976934, if P is a partition of [1, 3] into 6 equal
parts, then L(f, P ) ∼= 0.995635, and if P is a partition of [1, 3] into 7 equal parts, then
∼
L(f, P ) = 1.00937. This proves that L(f ) > 1 over the interval [1, 3]. On [1, 3], the function
R3
f is continuous and thus integrable, so that ln 3 = 1 f > 1 = ln e. Since ln is an increasing
function, this means that e < 3. By geometry ln(2) < U (f, {1, 2}) = 1 = ln e, so that
similarly e > 2. We conclude that e is a number strictly between 2 and 3.
1 1 1
Note that U (f, {1, 1.25, 1.5, 1.75, 2, 2.25, 2.5}) = 0.25(1 + 1.25 + 1.5 + 1.75 + 21 + 2.25
1
)=
2509
R 2.5
2520 < 1, so that ln 2.5 = 1 f is strictly smaller than this upper sum. It follows that
ln 2.5 < 1 = ln e and 2.5 < e. If P is a partition of [1, 2.71828] into a million pieces of equal
length, a computer gives that U (f, P ) is just barely smaller than 1, so that 2.71828 < e.
If P is a partition of [1, 2.718285] into a million pieces of equal length, then L(f, P ) is just
barely bigger than 1, so that e < 2.718285. Thus e ∼ = 2.71828.
A reader may want to run further computer calculations for greater precision. A dif-
ferent and perhaps easier computation is in Exercise 7.5.13.

Theorem 7.5.7 ln−1 (x) = ex .

Proof. By definitions,
ex = ln−1 (x · ln(e)) = ln−1 (x · ln(ln−1 (1))) = ln−1 (x · 1) = ln−1 (x).

We have already proved in page 214 that the derivative of ln−1 is ln−1 :

Theorem 7.5.8 (ex )′ = ex .

216 Chapter 7: Integration

Exercises for Section 7.5

7.5.1 Prove that for all x > 0, 1/(1 + x) < ln(1 + x) − ln x < 1/x. (Hint: Use the geometry
of the definition of ln.)
7.5.2 Let c ∈ R+ .
x
i) Prove that for all x ∈ R, cx = eln(c ) = ex ln(c) .
Rb 1
6 1, then a cx dx = ln(c)
ii) Prove that if c = (cb − ca ).
R x
7.5.3 Use integration by substitution (Exercise 7.4.5) to compute x2 +4 dx.
R 2
7.5.4 Let c ∈ R+ . Use integration by substitution (Exercise 7.4.5) to compute xc(x ) dx.
R
7.5.5 Let c ∈ R+ . Use integration by parts (Exercise 7.4.6) to compute xcx dx.
R
7.5.6 Use integration by parts (Exercise 7.4.6) to compute ln(x) dx.
7.5.7 This exercise is about applying various versions of the L’Hôpital’s rule.
i) Prove that lim lnxx = 0.
x→∞
x ln x
ii) Prove that lim x ln x = 0. (Hint: Use one of x ln x = 1/ ln x or x ln x = 1/x , and
x→0+
perhaps not both work.)
iii) Compute and justify lim ln(xx ).
x→0+
iv) Compute and justify lim xx .
x→0+
7.5.8 (Logarithmic differentiation) Sometimes it is hard or even impossible to compute
3
the derivative√of a function. Try for example f (x) = xx , f (x) = (x2 + 2)x +4 , or f (x) =
x2 (x−1)3 (x+4)3 x2 +1
√
3
x+2(x+7)3 (x−2)4
. There is another way if the range of the function consists of positive
real numbers: apply ln of both sides, take derivatives of both sides, and solve for f ′ (x).
For example, if f (x) = xx , then ln(f (x)) = ln(xx ) = x ln(x), so that
f ′ (x) ′ ′ x
= (ln(f (x))) = (x ln(x)) = ln(x) + = ln(x) + 1,
f (x) x
so that f ′ (x) = xx (ln(x) + 1).
3
i) Compute and justify the derivative of f (x) = (x2 + 2)x +4 . √
2
(x−1)3 (x+4)3 x2 +1
ii) Compute and justify the derivative of f (x) = x √
3
x+2(x+7)3 (x−2)4
.
1
7.5.9 Prove that lim+ ln(1 + x) = 1. (Hint: L’Hôpital’s rule.)
x→0 x  c
i) Prove that for any c ∈ R+ , lim x ln 1 + = c. (Hint: Change variables.)
x→∞ x
 c x
+
ii) Prove that for any c ∈ R , lim 1 + = ec .
x→∞ x
7.5.10 Prove that f (x) = (1 + x1 )x is a strictly increasing function. (Hint: use logarithmic
differentiation and Exercise 7.5.1.)
7.5.11 Prove that for all x ∈ R, (1 + x1 )x < e. (Hint: previous two exercises.)
 Section 7.5: Natural logarithm and the exponential functions 217

ex −1
7.5.12 Prove that limx→0 x = 1. You may want to apply L’Hôpital’s rule (Theo-
rem 6.4.3).
7.5.13 Let f (x) = ex .
i) For any positive integer n, compute the Taylor polynomial Tn,f,0 for ex of degree n
centered at 0 (Definition 6.5.2).
ii) Use Taylor’s remainder theorem (Theorem 6.5.5) to show that for any x ∈ R and
any ǫ > 0 there exists n ∈ N such that |f (x) − Tn,f,0 (x)| < ǫ.
iii) Use the Taylor’s remainder theorem (Theorem 6.5.5) to prove that |e−T8,f,0 (1)| =
|e1 − T8,f,0(1)| < 0.00001.
iv) Compute T8,f,0 (1) to 7 significant digits.
v) (Unusual) We computed some digits of e on page 215. On a computer, try to get
more digits of e with those methods, and with methods in this exercise. Which
method is faster? Can you streamline either method?
7.5.14 Let f (x) = ex . Use L’Hôpital’s rule (Theorem 6.4.3) and induction on n to prove
that
ex − Tn,f,0 (x) 1
lim n+1
= .
x→0 x (n + 1)!
(Hint: Rewrite for L’Hôpital’s rule.)
 2
−1/x
7.5.15 Let f : R → R be given by f (x) = e , if x 6= 0;
0, if x = 0.
i) Prove by induction on n ≥ 0 that for each n there exists a polynomial function
hn : R → R such that

1 −1/x2
f (x) = hn ( x ) · e
(n) , if x 6= 0;
0, if x = 0.
(At non-zero x you can use the chain rule, the derivative of the exponential func-
tion, and the power rule for derivatives. However, f (n+1) (0) is – but of course
(n) (n)
– computed as limh→0 f (h)−f h
(0)
, and then you have to use L’Hôpital’s rule.
You do not have to be explicit about the polynomial function.)
ii) Compute the nth Taylor polynomial for f centered at 0.
7.5.16 (A friendly competition between e and π) Use calculus (not a calculator!) to
determine which number is bigger, eπ or π e . You may assume that 1 < e < π. (Hint:
compute the derivative of some function.)
218 Chapter 7: Integration

7.6 Applications of integration

The Fundamental theorems of calculus relate integration with differentiation. In par-

Rb
ticular, to compute a f , if we know an antiderivative g of f , then the integral is easy.
However, as already mentioned after the Fundamental theorem of calculus I, many func-
tions do not have a “closed-form” antiderivative. One can still compute definite integrals
up to a desired precision, however: we take finer and finer partitions of [a, b], and when
U (f, P ) and L(f, P ) are within a specified distance from each other, we know that the true
integral is somewhere in between, and hence up to the specified precision either L(f, P ) or
Rb
U (f, P ) stands for a f . In applications, such as in science and engineering, many integrals
have to be and are computed in this way because of the lack of closed-form antiderivatives.
In this section we look at many applications that exploit the original definition of
integrals via sums over finer and finer partitions. For many concrete examples we can then
solve the integral via antiderivatives, but for many we have to make do with numerical
approximation.

7.6.1 Length of a curve

Let f : [a, b] → R be a continuous function. If the graph of f is a line, then by the

p
Pythagorean theorem the length of the curve is (b − a)2 + (f (b) − f (a))2 . For a general
curve it is harder to determine its length from (a, f (a)) to (b, f (b)). But we can do the
standard calculus trick: let P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on each subinterval
[xk−1 , xk ] we “approximate” the curve with the line (xk−1 , f (xk−1)) to (xk , f (xk )), compute
p
the length of that line as (xk − xk−1 )2 + (f (xk ) − f (xk−1 ))2 , and sum up all the lengths:
n
X p
(xk − xk−1 )2 + (f (xk ) − f (xk−1 ))2 .
k=1

Whether this is an approximation of the true length depends on the partition, but geomet-
rically it makes sense that the true length of the curve equals
n
X p
lim (xk − xk−1 )2 + (f (xk ) − f (xk−1 ))2 ,
k=1

as the partitions {x0 , x1 , . . . , xn } get finer and finer. But this is not yet in form of The-
orem 7.1.12. For that we need to furthermore assume that f is differentiable on (a, b).
Then by the Mean value theorem (Theorem 6.3.4) for each k = 1, . . . , n there exists
ck ∈ (xk−1 , xk ) such that f (xk ) − f (xk−1 ) = f ′ (ck )(xk − xk−1 ). If in addition we as-
sume that f ′ is continuous, then it is integrable by Theorem 7.3.1, and by Theorem 7.1.12
 Section 7.6: Applications of integration 219

the true length of the curve equals

n p
X
lim (xk − xk−1 )2 + (f ′ (ck )(xk − xk−1 ))2
P ={x0 ,x1 ,...,xn }finer and finer,ck ∈[xk−1 ,xk ]
k=1
n p
X
= lim 1 + (f ′ (ck ))2 (xk − xk−1 )
P ={x0 ,x1 ,...,xn }finer and finer,ck ∈[xk−1 ,xk ]
k=1
Z b p
= 1 + (f ′ (x))2 dx.
a

We just proved:

Theorem 7.6.1 If f : [a, b] → R is continuous and differentiable such that f ′ is continuous

on [a, b], then the length of the curve from (a, f (a)) to (b, f (b)) is
Z bp
1 + (f ′ (x))2 dx.
a

7.6.2 Volume of a surface area of revolution, disk method

Let f : [a, b] → R be continuous. We rotate the region between x = a and x = b and
bounded by the x-axis and the graph of f around the x-axis. If the graph is a horizontal
line, then the rotated region is a disk of height b − a and radius f (a), so that its volume
is π(f (a))2(b − a). For a general f we let P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on
each subinterval [xk−1 , xk ] we “approximate” the curve with the horizontal line y = f (ck )
for some ck ∈ [xk−1 , xk ], we compute the volume of the solid of revolution obtained by
rotating that approximated line over the interval [xk−1 , xk ] around the x-axis, and sum up
all the volumes:
Xn
π(f (ck ))2 (xk − xk−1 ).
k=1
Geometrically it makes sense that the true volume equals
n
X
lim π(f (ck ))2 (xk − xk−1 ),
P ={x0 ,x1 ,...,xn }finer and finer,ck ∈[xk−1 ,xk ]
k=1
Rb
and by Theorems 7.1.12 and 7.3.1, this equals π a
(f (x))2 dx. This proves:

Theorem 7.6.2 If f : [a, b] → R is continuous, then the volume of the solid of revolution
obtained by rotating around the x-axis the region between x = a and x = b and bounded
by the x-axis and the graph of f is
Z b
π (f (x))2 dx.
a
220 Chapter 7: Integration

7.6.3 Volume of a surface area of revolution, shell method

Let 0 ≤ a ≤ b, and let f, g : [a, b] → R be continuous such that for all x ∈ [a, b]
f (x) ≤ g(x). We rotate the region between y = a and y = b and bounded by the graphs
of x = f (y) and x = g(y) around the x-axis. If f (x) = c and g(x) = d are constant
functions, then the solid of revolution is a hollowed disk, with the outer border of height
d − c and radius b and the hole inside it has the same height but radius a. Thus the volume
is π(d − c)(b2 − a2 ). For general f and g we let P = {y0 , y1 , . . . , yn} be a partition of
[a, b]; on each subinterval [yk−1 , yk ] we “approximate” the curve f, g with the horizontal
line f (ck ), g(ck ) for some ck ∈ [yk−1 , yk ], we compute the volume of the solid of revolution
obtained by rotating that approximated region over the interval [yk−1 , yk ] around the x-
axis, and sum up all the volumes:
n
X
π(g(ck ) − f (ck ))(yk2 − yk−1
2
).
k=1

Geometrically it makes sense that the true volume equals

n
X
lim π(g(ck ) − f (ck ))(yk2 − yk−1
2
)
P ={y0 ,y1 ,...,yn }finer and finer,ck ∈[yk−1 ,yk ]
k=1
n
X
= lim π(g(ck ) − f (ck ))(yk + yk−1 )(yk − yk−1 ),
P ={y0 ,y1 ,...,yn }finer and finer,ck ∈[yk−1 ,yk ]
k=1
Rb
and by Theorems 7.1.12 and 7.3.1, this equals π a
(g(y) − f (y))2y dy. This proves:

Theorem 7.6.3 If 0 ≤ a ≤ b and f, g : [a, b] → R are continuous, then the volume of the
solid of revolution obtained by rotating around the x-axis the region between y = a and
y = b and bounded by the graphs of x = f (y) and x = g(y), is
Z b
2π y(g(y) − f (y)) dy.
a

Example 7.6.4 The volume of the sphere of radius r is 43 πr 3 .

Proof. We rotate the upper-half circle of radius r centered at the origin around the x-
axis. The circle of radius r centered at the origin consists of all points (x, y) such that
p p
x2 + y 2 = r 2 , so we have g(y) = r 2 − y 2 and f (y) = − r 2 − y 2 . Thus the volume is
Z r p   r
4 
4π 2
y r 2 − y 2 dy = − π(r − y ) 2 3/2  = 4 π(r 2)3/2 = 4 πr 3 .
3  3 3
0 0
 Section 7.6: Applications of integration 221

7.6.4 Surface area of the surface area of revolution

It will take quite a few steps to analyze the simple case before we get to an integral.
We first rotate the line segment y = mx from x = 0 to x = b > 0 around the x-axis.
We assume for now that m 6= 0.

In this way we obtain a right circular cone of height b and base radius |m|b. The perimeter
of that base circle is of course 2π|m|b. If we cut the cone in a straight line from a side to
p
the vertex, we cut along an edge of length b2 + (mb)2 , and we get the wedge as follows:

perimeter 2π|m|b

radius
p
b2 + (mb)2
p
Without the clip in the disc, the perimeter would be 2π b2 + (mb)2 , but our perimeter
is only 2π|m|b. Thus the angle subtended by the wedge is by proportionality equal to
√2π|m|b 2π = √2π|m| . The area of the full circle is radius squared times one half of
2π 2 2
b +(mb) 1+m2
the full angle, and so proportionally the area of our wedge is radius squared times one half
p 2π|m|
of our angle, i.e., the surface area of this surface of revolution is ( b2 + (mb)2 )2 2√1+m2
√
= π|m|b2 1 + m2 . Note that even if b < 0, the surface area is the absolute value of
√
πmb2 1 + m2 .
Thus if m is not zero and 0 ≤ a < b or a < b ≤ 0, then the surface area of revolution
obtained by rotating the line y = mx from x = a to x = b equals the absolute value of
√
πm(b2 − a2 ) 1 + m2 . Note the geometric requirement that at a and b the line is on the
same side of the x-axis.
222 Chapter 7: Integration

Now suppose that we rotate the line y = mx + l around the x-axis, with m 6= 0 and
a < b and both on the same side of the intersection of the line with the x-axis. This
intersection is at x = −l/m. By shifting the graph by l/m to the right, this is the same as
rotating the line y = mx from x = a + l/m to x = b + l/m, and by the previous case the
surface area of this is the absolute value of
p
πm((b + l/m)2 − (a + l/m)2 ) 1 + m2

p
= πm b2 + 2bl/m + l2 /m2 − (a2 + 2al/m + l2 /m2 ) 1 + m2

p
= πm b2 − a2 + 2(b − a)l/m 1 + m2
p
= πm(b − a)(b + a + 2l/m) 1 + m2
p
= π(b − a)(m(b + a) + 2l) 1 + m2 .

If instead we rotate the line y = l (with m = 0) around the x-axis, we get a ring whose
√
surface area is (b − a)2π|l|, which is the absolute value of π(b − a)(m(b + a) + 2l) 1 + m2 .
Thus for all m, the surface area of the surface of revolution obtained by rotating the line
y = mx + l from x = a to x = b around the x-axis is the absolute value of
p
π(b − a)(m(b + a) + 2l) 1 + m2 ,
with the further restriction in case m 6= 0 that a and b are both on the same side of the
x-intercept.
Now let f : [a, b] → R≥0 be a differentiable (and not necessarily linear) function. Let
P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on each subinterval [xk−1 , xk ] we approximate
the curve with the line from (xk−1 , f (xk−1)) to (xk , f (xk )). By the assumption that f
takes on only non-negative values we have that both xk−1 , xk are both on the same side of
f (x )−f (x )
the x-intercept of that line. The equation of the line is y = xkk −xk−1k−1 (x − xk ) + f (xk ),
so that m = f (xxkk)−f (xk−1 )
−xk−1
and l = − f (xxkk)−f (xk−1 )
−xk−1
xk + f (xk ). Since f is differentiable, by
the Mean value theorem (Theorem 6.3.4) there exists ck ∈ (xk−1 , xk ) such that f ′ (ck ) =
f (xk )−f (xk−1 )
xk −xk−1 . Thus m = f ′ (ck ) and l = f (xk ) − f ′ (ck )xk .
We rotate that line segment around the x-axis, compute its volume of that solid of
revolution, and add up the volumes for all the subparts:
X n  p 
 
π(xk − xk−1 ) (f ′ (ck )(xk + xk−1 ) + 2 (f (xk ) − f ′ (ck )xk )) 1 + (f ′ (ck ))2 
k=1
n
X p
= π |(f ′ (ck )(−xk + xk−1 ) + 2f (xk ))| 1 + (f ′ (ck ))2 (xk − xk−1 ).
k=1

By Theorems 7.1.12 and 7.3.1 we get the following:

 Section 7.6: Applications of integration 223

Theorem 7.6.5 If f : [a, b] → R≥0 is differentiable with continuous derivative, then the
surface area of the solid of revolution obtained by rotating around the x-axis the curve
y = f (x) between x = a and x = b is
Z b p Z b p
′ ′
π 2
|f (x)(−x + x) + 2f (x)| 1 + (f (x)) dx = 2π f (x) 1 + (f ′ (x))2 dx.
a a

Exercises for Section 7.6

7.6.1 Compute the perimeter of the circle with methods from this section.
7.6.2 Compute the volume of the sphere of radius r. Justify your work.
2 2 2
7.6.3 Compute the volume of the ellipsoid whose boundary satisfies xa2 + yb2 + zb2 = 1.
(We need multi-variable calculus to be able to compute the volume of the ellipsoid whose
2 2 2
boundary satisfies xa2 + yb2 + zc2 = 1.)
7.6.4 Compute the volume of a conical pyramid with base radius r and height h.
7.6.5 Compute the volume of a doughnut (you specify its dimensions).
7.6.6 Compute the surface area of the sphere of radius r.
7.6.7 If a tiny particle of mass m is rotating around a circle of radius r, its moment of
inertia of the particle is I = mr 2 . Likewise, if a (circular) hoop of radius r and of mass m
is rotating around its center, its moment of inertia of the particle is I = mr 2 . Now suppose
that we have a thin circular plate of radius 5a meters and mass 4 kilograms. (If you wish,
let b be the thickness of the plate and ρ the mass divided by the volume.) The goal of this
exercise is to find its moment of inertia about its diameter.
i) Draw the circular plate with center at origin. Let the axis of rotation be the y-axis.
ii) Let P be a partition of [0, 5]. Prove that the moment
q of inertia of the sliver of
4
the plate between xk−1 and xk is approximately 52 x2k − x2k−1 c2k , where ck ∈
[xk−1 , xk ].
iii) Prove that the moment of inertia of the rotating circular plate is 50 kilogram
meters squared.
7.6.8 If a constant force F moves an object by d units, the work done is W = F d. Suppose
that F depends on the position as F (x) = kx for some constant k. Find the total work
done between x = a and x = b.
224 Chapter 7: Integration

7.6.9 In hydrostatics, (constant) force equals (constant) pressure times (constant) area,
and (constant) pressure equals the weight density of the water w times (constant) depth
h below the surface. But most of the time we do not have tiny particles but large objects
where depth, pressure, and surface areas vary. For example, an object is completely sub-
merged under water from depth a to depth b. At depth h, the cross section area of the
object is A(h). Compute the total force exerted on the object by the water.
Chapter 8: Sequences

In this chapter, Sections 8.5 and 8.4 contain identical results in identical order, but
the proofs are different. You may want to learn both perspectives, or you may choose to
omit one of the two sections.

8.1 Introduction to sequences

Definition 8.1.1 An infinite sequence is a function with domain N+ and codomain C.

If s is such a function, instead of writing s(n), it is common to write sn , and the sequence
s is commonly expressed also in all of the following notations (and obviously many more):
s = {s1 , s2 , s3 , . . .} = {sn }∞ ∞
n=1 = {sn }n≥1 = {sn }n = {sn+3 }n=−2 = {sn−4 }n>4 = {sn }.

The nth element sn = s(n) in the ordered list is called the nth term of the sequence.

The notation {s1 , s2 , s3 , . . .} usually stands for the set consisting of the elements
s1 , s2 , s3 , . . ., and the order of a listing of elements in a set is irrelevant. Here, however,
{s1 , s2 , s3 , . . .} stands for the sequence, and the order matters. When the usage is not
clear from the context, we add the word “sequence” or “set” as appropriate.
The first term of the sequence {2n − 1}n≥4 is 7, the second term is 9, et cetera. The
point is that even though the notating of a sequence can start with an arbitrary integer,
the counting of the terms always starts with 1.
Note that sn is the nth term of the sequence s, whereas {sn } = {sn }n≥1 is the sequence
in which n plays a dummy variable. Thus

s = {sn } =
6 sn .

Examples and notation 8.1.2

(1) The terms of a sequence need not be distinct. For any complex number c, {c} =
{c, c, c, . . .} is called a constant sequence.
(2) For any complex numbers c and d, the sequence {c, d, c, d, c, d, . . .} can be written
more concisely as { d−c n c+d
2 (−1) + 2 }n .
(3) The sequence {(−1)n } = {−1, 1, −1, 1, . . .} has an infinite number of terms, and
its range is the finite set {−1, 1}. The range of the sequence {in } is the set
{i, −1, −i, 1}.
226 Chapter 8: Sequences

(4) The sequence s with s(n) = n + 4 for all n ≥ 1 can be written as {n + 4} =

{n + 4}n≥1 = {5, 6, 7, . . .} = {n}n≥5 . This sequence is different from the sequence
{n} = {1, 2, 3, . . .}.
(5) The sequence {2n} is an abbreviation for the sequence {2, 4, 6, 8, . . .}. This is not
the same as the sequence {4, 2, 8, 6, 12, 10, . . .}, namely, if we scramble the order
of the terms, we change the sequence.
(6) The sequence of all odd positive integers (in the natural order) can be written as
{2n − 1} = {2n − 1}n≥1 = {2n + 1}n≥0 (and in many other ways).
(7) Terms of a sequence can be defined recursively. For example, let s1 = 1, and for
each n ≥ 1 let sn+1 = 2sn . Then the sequence of these sn has a non-recursive form
√
sn = 2n−1 . If however, s1 = 1 and for each n ≥ 1, sn+1 = 2s3n + sn + 2+ln(sn )+1,
then we can certainly compute any one sn by invoking sk for k < n, but we will
not get a closed-form for sn as in the previous example.
(8) (Fibonacci numbers) Let s1 = 1, s2 = 1, and for all n ≥ 2, let sn+1 = sn +sn−1 .
This sequence starts with 1, 1, 2, 3, 5, 8, 13, 21, 34, . . ., and obviously keeps growing.
(For more on these numbers, see  Exercise
 1.5.31,  where  in particular it is proved
√ n √ n
that for all integers n, sn = √1 1+ 5
− √1 1− 5
.)
5 2 5 2
(9) Some sequences, just like functions, do not have an algebraic expression for terms.
For example, let s be the sequence whose nth term is the nth prime number.
This sequence starts with 2, 3, 5, 7, 11, 13, 17, 19, 23, and we could write many more
terms out explicitly, but we do not have a formula for them. (This s is indeed an
infinite sequence since there are infinitely many primes, as proved on page 33.)
(10) Note that {n}n∈Z is NOT a sequence because the list has no first term.
(11) However, one can scramble the set Z of all integers into a sequence. One way
of doing this as as follows: {0, 1, −1, 2, −2, 3, −3, 4, −4, . . .}, which algebraically
equals 
n/2, if n is even;
sn =
−(n − 1)/2, otherwise.
(“Otherwise” applies to odd (positive) integers.)
(12) One can scramble the set Q+ of all positive rational numbers into a sequence via
a diagonal construction as follows. First of all, each positive rational number can
be written in the form a/b for some positive integers a, b. Rather than plotting the
fraction a/b, we will plot the point (a, b) in the plane. Refer to Plot 8.1.2: the bold
points are elements of N+ × N+ , and each such (a, b) is identified with the fraction
a/b. Every positive rational number appears in this way somewhere as a bold point,
and all appear multiple times because ab = ac bc . Now we want to systematically
enumerate these bold points/rational numbers. If we first enumerate all of them
in the first row, and then proceed to the second row, well, actually, we never get
 Section 8.1: Introduction to sequences 227

to the second row as we never finish the first row. So we need a cleverer way of
counting, and that is done as follows. We start counting at (1, 1), which stands
for 1/1 = 1. We then proceed through all the other integer points in the positive
quadrant of the plane via diagonals as in Plot 8.1.2. The given instructions would
enumerate positive rational numbers as 1/1, 2/1, 1/2, 1/3, 2/2. Ah, but 2/2 has
already been counted as 1/1, so we do not count 2/2. Thus, the proper counting
of positive rational numbers in this scheme starts with:
1/1, 2/1, 1/2, 1/3, 2/2, 3/1, 4/1, 3/2, 2/3, 1/4, 1/5, 2/4, 3/3, 4/2, 5/1, 6/1, 5/2, 4/3,
et cetera, where the crossed out numbers are not part of the sequence because they
had been counted earlier. For example, the fifth term is 3.

b b b b b b

b b b b b b

b b b b b b

b b b b b b

b b b b b b

1 b b b b b b

1 2 3 4 5 6

Plot 8.1.2 Beginning of the enumeration of the positive rational numbers.

It is important to note that every positive rational number appears on this list,
and because we are skipping any repetitions, it follows that every positive rational
number appears on this list exactly once. Thus this gives an enumeration of
positive rational numbers. *
A different enumeration of Q+ is given with an algebraic formulation in Exer-
cise 2.4.29.
(13) If {sn } is a sequence of positive rational numbers in which every positive rational
number appears exactly (or at least) once, we can construct from it a sequence in

* Here is a fun exercise: look at the ordered list of positive rational numbers above, including the crossed-out
(n+m−2)2
fractions. Verify for a few of them that n/m is in position 2 + 3n+m−4
2 + 1 on the list. Namely, it is a
(x+y−2)2 3x+y−4 + 2 +
fact that f (x, y) = 2 + 2 + 1 gives a bijection of (N ) with N . This was first proved by Rudolf
Fueter and George Pólya, but the proof is surprisingly hard, using transcendence of er for algebraic numbers r,
so do not attempt to prove this without much more number theory background.
228 Chapter 8: Sequences

which every rational number appears exactly (or at least) once as follows:

 0, if n = 1;
qn = sn/2 , if n is even;
s , if n ≥ 3 is odd.
(n−1)/2

This sequence starts with 0, s1 , −s1 , s2 , −s2 , s3 , −s3 , . . .. Since every positive ratio-
nal number is one of the sn (exactly once/at least once), so every rational number
is on this new list (exactly once/at least once).
Incidentally, it is impossible to scramble R or C into a sequence. This can be proved
with a so-called Cantor’s diagonal argument, which we are not presenting here,
but an interested reader can consult other sources.
(14) Sequences are functions, and if all terms of the sequence are real numbers, we can
plot sequences in the usual manner for plotting functions. The following is part of
a plot of the sequence {1/n}.
sn
1 b

b
b
b b

n
b b b b b b b b b b

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Plot 8.1.2 sn = 1/n

(15) Another way to plot a sequence is to simply plot and label each sn in the complex
plane or on the real line. We plot three examples below.

s1 , s3 , s5 , . . . s2 , s4 , s6 , . . . s3 s2 s1
−1 1 0 1

s1
b
i/2
s3 s4 s2 b

b
b
b b
b b
b b bb b b
b bb b b b

−1 1

Plot 8.1.2 Plots of the image sets of {(−1)n}, {1/n} and {(−1)n/n + 0.5i/n}.

There is an obvious arithmetic on sequences (just like there is on functions):

{sn } ± {tn } = {sn ± tn },

{sn } · {tn } = {sn · tn },
c{sn } = {csn },
 Section 8.1: Introduction to sequences 229

{sn }/{tn} = {sn /tn } (if tn 6= 0 for all n).

One has to make sure to add/multiply/divide equally numbered terms of the two
sequences, such as in the following:
{n}n≥3 + {n}n≥2 = {n + 1}n≥2 + {n}n≥2 = {2n + 1}n≥2 .
Here are a few further examples of arithmetic on sequences, with + and · binary
operations on the set of sequences:

{2n } + {−2n } = {0},

{2n } + {(−2)n } = {0, 8, 0, 32, 0, 128, 0, 512, 0, . . .},
2{2n } = {2n+1 },
{2n } · {2−n } = {1},
{2n − 1} + {1} = {2n},
{(−1)n n} + {2/n} = {(−1)nn + 2/n},
{in }/{(−i)n} = {(−1)n }.

Exercises for Section 8.1

8.1.1 Express algebraically the ordered sequence of all positive integer multiples of 3.
8.1.2 Think of a sequence whose terms are all between 2 and 3.
8.1.3 Plot the sequences {n2 } and {1/n2 }. Compare the plots.
8.1.4 Plot the sequences {(−1)n } and {cos(nπ)}.
8.1.5 Plot the image set of the sequence s = {in }: draw the real and imaginary axes and
the unit circle centered at the origin; on this circle, plot i1 , i2 , i3 , i4 , i5 , i6 , and label each
correspondingly with “s1 ”, “s2 ”, “s3 ”, “s5 ”, “s6 ”. Label also s20 , s100, s101, s345.
8.1.6 Prove that the difference of the ordered sequence of all positive odd integers and the
constant sequence {1} is the ordered sequence of all non-negative even integers.
8.1.7 We have seen that + and · are binary operations on the set of infinite sequences.
i) Prove that {0} is the identity for + and that {1} is the identity for ·.
ii) Prove that every infinite sequence has an inverse for +.
iii) Show that not every infinite sequence has an inverse for this operation. What
infinite sequences have an inverse for this operation?
8.1.8 Sequences can also be finite. Two examples of finite sequences are: (i) last exam
scores in a class arranged in alphabetic order by the student; and (ii) last exam scores in
a class arranged in ascending order by score. Give two more examples of finite sequences.
230 Chapter 8: Sequences

8.2 Convergence of infinite sequences

Definition 8.2.1 A sequence s = {sn } converges to L ∈ C if for every real number ǫ > 0
there exists a positive real number N such that for all integers n > N , |sn − L| < ǫ.
If s converges to L, we also say that L is the limit of {sn }. We use the following
notations for this:
sn → L, {sn } → L,
lim s = L, lim sn = L, lim{sn } = L, lim(sn ) = L,
lim sn = L, lim {sn } = L, lim{sn } = L, lim (sn ) = L,
n→∞ n→∞ n n→∞

and, to save vertical space, just like for limits of functions, we also use a variation on the
last three: limn→∞ sn = L, limn→∞ {sn } = L, limn→∞ (sn ) = L.
We say that a sequence is convergent it it has a limit.

For example, the constant sequence s = {c} converges to L = c because for all n,
|sn − L| = |c − c| = 0 is strictly smaller than any positive real number ǫ.
The sequence s = {300, −5, π, 4, 0.5, 106, 2, 2, 2, 2, 2, . . .} converges to L = 2 because
for all n ≥ 7, |sn − L| = |2 − 2| = 0 is strictly smaller than any positive real number ǫ.
In conceptual terms, a sequence {sn } converges if the tail end of the sequence gets
closer and closer to L; you can make all sn with n > N get arbitrarily close to L by simply
increasing N a sufficient amount.
We work out examples of epsilon-N proofs; they are similar to the epsilon-delta proofs,
and we go through them slowly at first. Depending on the point of view of your class, the
reader may wish to skip the rest of this section for an alternative treatment in Section 8.5
in terms of limits of functions. More epsilon-N proofs are in Section 8.4. Be aware that
this section is more concrete; the next section assumes greater ease with abstraction.

Example 8.2.2 Consider the sequence s = {1/n}. Plot 8.1.2 gives a hunch that lim sn = 0,
and now we prove it. [(Recall that any text between square brackets in this
font is what should approximately be going through your thoughts, but it
is not something to write down in a final solution.) By the definition of
convergence, we have to show that for all ǫ > 0 some property holds. All
proofs of this form start with:] Let ǫ be an arbitrary positive number. [Now we
have to show that there exists an N for which some other property holds.
Thus we have to construct an N . Usually this is done in retrospect, one
cannot simply guess an N , but in the final write-up, readers see simply that
educated guess – more about how to guess educatedly later:] Set N = 1ǫ . Then
N is a positive real number. [Now we have to show that for all integers n > N ,
|sn − 0| < ǫ. All proofs of statements of the form “for all integers n > N ”
 Section 8.2: Convergence of infinite sequences 231

start with:] Let n be an (arbitrary) integer with n > N . [Finally, we have to prove
the inequality |sn − 0| < ǫ. We do that by algebraically manipulating the
left side until we get the desired final < ǫ:]
|sn − 0| = |1/n − 0|
= |1/n|
= 1/n (because n is positive)
< 1/N (because n > N > 0)
1
= (because N = 1/ǫ) [Ah, what a clever guess that was!]
(1/ǫ)
= ǫ.
So we conclude that |sn − 0| < ǫ, which proves that lim sn = 0.
Just as in the epsilon-delta proofs where one has to find a δ, similarly how does one
divine an N ? In the following two examples we indicate this step-by-step, not as a book
or your final homework solution would have it recorded.

Example 8.2.3 Let sn = { n1 ((−1)n + i(−1)n+1 )}. If we write out the first few terms, we
find that {sn } = {−1 + i, 1/2 − i/2, −1/3 + i/3, . . .}, and we may speculate that lim sn = 0.
Here is plot of the image set of this sequence in the complex plane:

s1
b
i/2
s3
b
b
b b
b b bb
b
bb b b
b b
b
s4
−1
b
b
s2 1
b

We prove that lim sn = 0. Let ǫ > 0. Set N = . [We will eventually fill
in what the positive real number N should be, but at this point of the proof
simply leave some blank space. Assuming that N is in place, we next need to
prove that for every integer n > N , the relevant property as in the ǫ − N
definition of limits holds. The proof of “for every integer n > N ” always
starts with:] Let n be an integer strictly bigger than N . [We want to make sure
eventually that n is positive, which is guaranteed if N is positive, but with
blank N , we will simply assume in the algebra below that N is positive.]
Then
 
1 
|sn − 0| =  ((−1) + i(−1)
n n+1
) − 0
n
232 Chapter 8: Sequences
 
1
=   · |(−1)n + i(−1)n+1 | (because |ab| = |a||b|)
n
1
= · |(−1)n + i(−1)n+1 | (because n is positive)
n
1 p
= · ((−1)n )2 + ((−1)n+1 )2
n
1 √
= · 2
n
1 √
< · 2 (because n > N )
N
√ √ √
[Aside: we want/need 2/N ≤ ǫ, and 2/N = ǫ is a possibility, so set N = 2/ǫ.
Now go ahead, write that missing information on N in line 1 of this proof!]
√
2 √
= √ (because N = 2/ǫ)
2/ǫ
= ǫ,
which proves that for all n > 1/ǫ, |sn − 0| < ǫ. Since ǫ is arbitrary, this proves that
lim sn = 0.

Thus a polished version of the example just worked out looks like this:
 √
We prove that lim n1 ((−1)n + i(−1)n+1 ) − 0 = 0. Let ǫ > 0. Set N = 2/ǫ. Then
N is a positive real number. Let n be an integer strictly bigger than N . Then
 
1 
 n
|sn − 0| =  ((−1) + i(−1) n+1
) − 0
n
 
1
=   · |(−1)n + i(−1)n+1 | (because |ab| = |a||b|)
n
1
= · |(−1)n + i(−1)n+1 | (because n is positive)
n
1 p
= · ((−1)n )2 + ((−1)n+1 )2
n
1 √
= · 2
n
1 √
< · 2 (because n > N )
N√
2 √
= √ (because N = 2/ǫ)
2/ǫ
= ǫ,
√
which proves that for all n > 2/ǫ, |sn − 0| < ǫ. Since ǫ is arbitrary, this proves that
lim sn = 0.
 Section 8.2: Convergence of infinite sequences 233
2
2n+3n
Example 8.2.4 Claim: lim 3+4n+n 2 = 3. Proof: Let ǫ > 0. Set N = . Let n be
an integer strictly bigger than N . Then
   
 2n + 3n2   2n + 3n2 3(3 + 4n + n2 ) 
  
 3 + 4n + n2 − 3 =  3 + 4n + n2 − 3 + 4n + n2 
 
 −9 − 10n 
=  
3 + 4n + n2 
9 + 10n
= (because n > 0) [Assuming that N > 0.]
3 + 4n + n2
n + 10n
≤ (because n ≥ 9) [Assuming that N ≥ 8.]
3 + 4n + n2
11n
=
3 + 4n + n2
11n
≤ 2 (because 3 + 4n + n2 > n2 , so 1/(3 + 4n + n2 ) < 1/n2 )
n
11
=
n
11
< (because n > N )
N
11
≤ (because N ≥ 11/ǫ so 1/N ≤ 1/(11/ǫ)) [Assuming this.]
11/ǫ
= ǫ,

which was desired. Now (on scratch paper) we gather all the information we used about N :
N > 0, N ≥ 8, N ≥ 11/ǫ, and that is it. Thus on the first line we fill in the blank part:
Set N = max{8, 11/ǫ}, which says that N is either 8 or 11/ǫ, whichever is greater, so that
N ≥ 8 and N ≥ 11/ǫ.
The polished version of this proof would go as follows:

2
2n+3n
Example 8.2.5 Claim: lim 3+4n+n 2 = 3. Proof: Let ǫ > 0. Set N = max{8, 11/ǫ}. Then

N is a positive real number. Let n be an integer strictly bigger than N . Then

   
 2n + 3n2   2n + 3n2 3(3 + 4n + n2 ) 
  
 3 + 4n + n2 − 3 =  3 + 4n + n2 − 3 + 4n + n2 
 
 −9 − 10n 
=  
3 + 4n + n2 
9 + 10n
= (because n > 0)
3 + 4n + n2
n + 10n
≤ (because n ≥ 9)
3 + 4n + n2
11n
=
3 + 4n + n2
234 Chapter 8: Sequences

11n
≤ (because 3 + 4n + n2 > n2 , so 1/(3 + 4n + n2 ) < 1/n2 )
n2
11
=
n
11
< (because n > N )
N
11
≤ (because N ≥ 11/ǫ so 1/N ≤ 1/(11/ǫ))
11/ǫ
= ǫ,

which proves that for all n > N , |sn − 3| < ǫ. Since ǫ is arbitrary, this proves that the limit
of this sequence is 3.
Below is a polished proof of a very similar problem.
2
−2n+3n
Example 8.2.6 Claim: lim 9−4n+n 2 = 3. Proof: Let ǫ > 0. Set N = max{2, 20/ǫ}. Then

N is a positive real number. Let n be an integer strictly bigger than N . Then

   
 −2n + 3n2   −2n + 3n2 3(9 − 4n + n2 ) 
  
 9 − 4n + n2 − 3 =  9 − 4n + n2 − 9 − 4n + n2 
 
 −27 + 10n 
=  
9 − 4n + n2 
−27 + 10n
= (because n > N ≥ 2, so n ≥ 3, so 10n − 27 > 0,
9 − 4n + n2
and n2 − 4n + 9 = (n − 3)2 + 2n > 0)
10n
< (because 10n − 27 < 10n)
9 − 4n + n2
10n
= √ √
9 − 4n + (1/ 2)n2 + (1 − 1/ 2)n2
10n √
< √ (because 0 < (n − 4)2 + 2 = 2(9 − 4n + (1/ 2)n2 ))
(1 − 1/ 2)n2
10 √ 1√ 1
< (because 1 − 1/ 2 > 0.5, so 1−1/ 2
< 0.5 )
0.5n
20
=
n
20
< (because n > N )
N
20
≤ (because N ≥ 20/ǫ so 1/N ≤ 1/(20/ǫ))
20/ǫ
= ǫ,
which proves that for all n > N , |sn − 3| < ǫ. Since ǫ is arbitrary, this proves that the limit
of this sequence is 3.
 Section 8.2: Convergence of infinite sequences 235

Example 8.2.7 Let r ∈ C such that |r| < 1. Then limn→∞ r n = 0.

Proof. If r = 0, the sequence {r n } is the constant zero sequence so certainly the limit is 0.
So we may assume without loss of generality that r 6= 0. Let ǫ > 0. Set N =
ln(min{ǫ,0.5})
ln |r| .
Since ln of numbers between 0 and 1 is negative, we have that N is a positive
number. Let n be an integer with n > N . Then
|r n − 0| = |r|n
= |r n−N | |r|N
< 1n−N |r|N (by Theorem 7.5.5)
= min{ǫ, 0.5}
≤ ǫ.


1 n
Example 8.2.8 lim 1+ n = e.

Proof. All the hard work for this has been done already in Exercise 7.5.9. Let ǫ > 0. By

x
Exercise 7.5.9, limx→∞ 1 + x1 = e. Thus there exists N > 0 such that for all x > N ,

 

 1 + 1 x − e < ǫ. In particular, for any integer n > N ,  1 + 1 n − e < ǫ.
x n

Example 8.2.9 lim n1/n = 1.

Proof. For all n ≥ 2, by Exercise 1.6.7,

n  
X
1/n n 1/n n n
n = (n ) = (n − 1 + 1) = (n1/n − 1)k .
k
k=0

Each of the summands is non-negative, and if we only use the summands with k = 0 and
k = 2, we then get that
1
n ≥ 1 + n(n − 1)(n1/n − 1)2 .
2

By subtracting 1√we get that n−1 ≥ 21 n(n−1)(n1/n −1)2 , so that for n ≥ 2, n2 ≥ (n1/n −1)2 ,
and hence that √n2 ≥ n1/n − 1. Certainly n1/n − 1 ≥ 0 for all n ≥ 1. It follows that for all
√
n ≥ 2, and even for all n ≥ 1, 0 ≤ n1/n − 1 ≤ √2 . Now let ǫ > 0. Set N = max{2, 2/ǫ2}.
n √
Then N is a positive real number. Let n > N be an integer. Then 0 ≤ n1/n − 1 ≤ √2 <
√ n
√2 = ǫ, which proves that |n1/n − 1| < ǫ, and hence proves this limit.
N
236 Chapter 8: Sequences

Example 8.2.10 Let M be a positive real number. Then lim M 1/n = 1.

Proof. First suppose that M ≥ 1. Then certainly for all integers n ≥ M , we have that
1 ≤ M 1/n ≤ n1/n . Let ǫ > 0. By the previous example, there exists N > 0 such
that for all integers n > N , 0 < n1/n − 1 < ǫ. Then for all integers n > max{M, N },
0 ≤ M 1/n − 1 ≤ n1/n − 1 < ǫ, which implies that |M 1/n − 1| < ǫ. This proves the example
in case M ≥ 1.
Now suppose that M < 1. By assumption, 1/M > 1, so by the previous case (and
by using that (1/M )1/n = 1/M 1/n ), there exists N > 0 such that for all integers n > N ,
0 ≤ 1/M 1/n − 1 < ǫ. By adding 1 to all three parts in this inequality we get that
1 ≤ 1/M 1/n < ǫ+1, so that by compatibility of < with multiplication by positive numbers,
1
ǫ+1
< M 1/n ≤ 1. Hence by compatibility of < with addition,
1 ǫ
0 ≤ 1 − M 1/n < 1 − = < ǫ,
ǫ+1 ǫ+1
since ǫ + 1 > 1. Thus |1 − M 1/n | < ǫ, which proves that lim M 1/n = 1.

Exercises for Section 8.2

8.2.1 Let sn = 1/n2 . Fill in the blanks of the following proof that lim{sn } = 0.
Let ǫ > 0. Set N = . Then if n > N ,
 
1 

|sn − 0| =  2 − 0
n
1
= (because )
n2
1
< 2 (because )
N
= (because )
= ǫ.

8.2.2 Prove that for any real number L, lim{ n1 + L} = L.

8.2.3 Prove that for any positive real number k, lim{1/nk } = 0.
8.2.4 Prove that lim sin(n)
n
= 0.
8.2.5 Prove that lim 2n+1
n2 −4
= 0.
8.2.6 Prove that lim 3n+4 3
2n = 2 .
√ √
8.2.7 Prove that lim n + 1 − n = 0.
√ √
8.2.8 Prove that for every positive integer k, lim n+k− n = 0.
 Section 8.3: Divergence of infinite sequences and infinite limits 237
Pn 1
8.2.9 Suppose that the terms of a sequence are given by sn = k=1 k(k+1) .
Pn 1 1
i) Using induction on n, prove that k=1 k(k+1) = 1 − n+1 .
ii) Use this to find and prove the limit of {sn }.
8.2.10 Prove the following limits:
ln(n)
i) lim ln(n+1) = 1. (Hint: epsilon-N proof, continuity of ln.)
ln(ln(n))
ii) lim ln(ln(n+1)) = 1.
8.2.11 Let r ∈ R. By Theorem 2.7.13, for every positive integer n, there exists a rational
number sn ∈ (r − n1 , r + n1 ). Prove that {sn } converges to r.
8.2.12 Let r be a real number with a known decimal expansion. Let sn be the rational
number whose digits n + 1, n + 2, n + 3, et cetera, beyond the decimal point are all 0, and
all other digits agree with the digits of r. (For example, if r = π, then s1 = 3.1, s2 = 3.14,
s7 = 3.1415926, et cetera.) Prove that lim{sn } = r. (Repeat with binary expansions if you
know what a binary expansion is.)
√
n!
8.2.13 Prove that limn { n! } = 1.
n
8.2.14 What is wrong with the following “proof” that limn→∞ 2n+1
= 12 .
1
−1
“Proof.” Let ǫ > 0. Set N = 2ǫ2 . Let n > N . Then
   
 n 1   2n − (2n + 1) 

 2n + 1 − 2  =  2(2n + 1) 
 
 −1 
=  
2(2n + 1) 
1
=
2(2n + 1)
1
< (because all terms are positive)
2(2N + 1)
= ǫ.

8.3 Divergence of infinite sequences and infinite limits

The sequence {(−1)n} alternates in value between −1 and 1, and does not seem to
converge to a single number. The following definition addresses this situation.
238 Chapter 8: Sequences

Definition 8.3.1 A sequence diverges if it does not converge. In other words, {sn }
diverges if for all complex numbers L, lim{sn } =
6 L.

By the usual negation of statements (see chart on page 32), lim{sn } =

6 L means:

not For all real numbers ǫ > 0 there exists a positive real number N


such that for all integers n > N , |sn − L| < ǫ.

= There exists a real number ǫ > 0 such that not there exists a positive


real number N such that for all integers n > N , |sn − L| < ǫ.

= There exists a real number ǫ > 0 such that for all positive real numbers


N , not for all integers n > N , |sn − L| < ǫ.

= There exists a real number ǫ > 0 such that for all positive real numbers


N there exists an integer n > N such that not |sn − L| < ǫ.

= There exists a real number ǫ > 0 such that for all positive real numbers
N there exists an integer n > N such that |sn − L| ≥ ǫ.

Example 8.3.2 {(−1)n} is divergent. Namely, for all complex numbers L, lim sn 6= L.

Proof. Set ǫ = 1 (half the distance between the two values of the sequence). Let N be
an arbitrary positive number. If Re(L) > 0, let n be an odd integer greater than N ,
and if Re(L) ≤ 0, let n be an even integer greater than N . In either case, |sn − L| ≥
| Re(sn ) − Re(L)| ≥ 1 = ǫ.
The sequence in the previous example has no limit, whereas the sequence in the next
example has no finite limit:

6 L.
Example 8.3.3 For all complex numbers L, lim{n} =

Proof. Set ǫ = 53 (any positive number will work). Let N be a positive real number. Let n
be any integer that is strictly bigger than N and strictly bigger than |L| + 53 (say strictly
bigger than N + |L| + 53). Such an integer exists. Then by the reverse triangle inequality,
|n − L| ≥ |n| − |L| ≥ 53 = ǫ.
The last two examples are different: the first one has no limit at all since the terms
oscillate wildly, but for the second example we have a sense that its limit is infinity. We
formalize this:
 Section 8.3: Divergence of infinite sequences and infinite limits 239

Definition 8.3.4 A real-valued sequence {sn } diverges to ∞ if for every positive real
number M there exists a positive number N such that for all integers n > N , sn > M . We
write this as lim sn = ∞.
A real-valued sequence {sn } diverges to −∞ if for every negative real number M
there exists a positive real number N such that for all integers n > N , sn < M . We write
this as lim sn = −∞.

Example 8.3.5 lim n = ∞.

Proof. Let M > 0. Set N = M . (As in epsilon-delta or epsilon-N proofs, we must figure
out what to set N to. In this case, N = M works). Let n ∈ N+ with n > N . Since N = M ,
we conclude that n > M , and the proof is complete.
√
n
Example 8.3.6 lim n! = ∞.

Proof. Let M > 0. Let M0 be an integer that is strictly greater than M . By Example 8.2.10
there exists N1 > 0 such that for all integers n > N1 ,
s 
 M0 !  M +1−M
n  0
 M
− 1 < .
 (M0 + 1) 0  M0 + 1

Since M0 ! = M0 (M0 − 1)(M0 − 2) · · · 3 · 2 · 1 < (M0 + 1)M0 , this means that

s
M0 ! M0 + 1 − M M
0≤ 1− n < = 1 − ,
(M0 + 1)M0 M0 + 1 M0 + 1
q
so that n (M0M 0!
+1)M0
> MM0 +1
. Set N = max{N1 , M0 }. Then for all integers n > N ,

M0 !
n! = n(n − 1) · · · (M0 + 1) · M0 ! ≥ (M0 + 1)n−M0 · M0 ! = (M0 + 1)n ,
(M0 + 1)M0
so that s
√
n M0 ! M
n! ≥ (M0 + 1) n M
> (M0 + 1) = M.
(M0 + 1) 0 M0 + 1

Theorem 8.3.7 (Comparison theorem (for sequences with infinite limits)) Let
{sn }, {tn} be real-valued sequences such that for all sufficiently large n (say for n ≥ N for
some fixed N ), sn ≤ tn .
(1) If lim sn = ∞, then lim tn = ∞.
(2) If lim tn = −∞, then lim sn = −∞.

Proof. (1) By assumption lim sn = ∞ for every positive M there exists a positive N ′ such
that for all integers n > N ′ , sn > M . Hence by assumption tn ≥ sn for all n ≥ N we get
240 Chapter 8: Sequences

that for every positive M and for all integers n > max{N, N ′ }, tn > M . Thus by definition
lim tn = ∞.
Part (2) has an analogous proof.
2
Example 8.3.8 lim n n+1 = ∞.
2
Proof. Note that for all n ∈ N+ , n n+1 = n + n1 ≥ n, and since we already know that
2
lim n = ∞, it follows by the comparison theorem above that lim n n+1 = ∞.
2
Example 8.3.9 lim n n−1 = ∞.
2
Proof. Note that for all integers n > 2, n n−1 = n − n1 ≥ n2 . We already know that
lim n = ∞, and it is straightforward to prove that lim n2 = ∞. Hence by the comparison
2
theorem, lim n n−1 = ∞.
Or, we can give an M − N proof. Let M > 0. Set N = max{2, 2M }. Let n be an
integer strictly bigger than N . Then
n2 − 1 1 n N
=n− ≥ > ≥ M.
n n 2 2
Theorem 8.3.10 Let {sn } be a sequence of positive numbers. Then lim sn = ∞ if and
only if lim s1n = 0.

Proof. Suppose that lim sn = ∞. Let ǫ > 0. By the definition of infinite limits, there
exists a positive number N such that for all integers n > N , sn > 1/ǫ. Then for the same
n, 0 < s1n < ǫ, so that | s1n | < ǫ. This proves that lim s1n = 0.
Now suppose that lim s1n = 0. Let M be a positive number. By assumption lim s1n = 0
there exists a positive number N such that for all integers n > N , | s1n − 0| < 1/M . Since
each sn is positive, it follows that for the same n, s1n < 1/M , so that sn > M . This proves
that lim sn = ∞.

Exercises for Section 8.3

8.3.1 Prove that the following sequences diverge:
√
i) { n}.
√
ii) {2 n }.
iii) {(n + 1)3 − n3 }.
iv) {(−1)n + 3/n}.
8.3.2 Suppose that {sn } diverges and {tn } converges. Prove that {sn ± tn } diverges.
8.3.3 Suppose that {sn } diverges and {tn } converges to a non-zero number. Prove that
{sn · tn } diverges.
 Section 8.4: Convergence theorems via epsilon-N proofs 241

8.3.4 Give an example of two divergent sequences {sn }, {tn } such that {sn +tn } converges.
8.3.5 Let {sn } be a sequence of negative numbers. Prove that lim sn = −∞ if and only if
lim s1n = 0.
8.3.6 Given the following sequences, find and prove the limits, finite or infinite, if they
exist. Otherwise, prove divergence:
i) { nn+5
3 −5 }.
2
−n
ii) { 2n
3n2 −5 }.
n
iii) { (−1)
n−3
}.
n
iv) { (−1)
n+1
n
}.
3
−8n
v) { nn2 +8n }.
2
vi) { 1−nn }.
n
vii) { 2n! }.
n!
viii) { (n+1)! }.
8.3.7 Prove or give a counterexample:
i) If {sn } and {tn } both diverge, then {sn + tn } diverges.
ii) If {sn } converges and {tn } diverges, then {sn + tn } diverges.
iii) If {sn } and {tn } both diverge, then {sn · tn } diverges.
iv) If {sn } converges and {tn } diverges, then {sn · tn } diverges.
8.3.8 Find examples of the following:
sn
i) A sequence {sn } of non-zero terms such that lim{ sn+1 } = 0.
sn+1
ii) A sequence {sn } of non-zero terms such that lim{ sn } = 1.
iii) A sequence {sn } of non-zero terms such that lim{ sn+1
sn
} = ∞.
iv) A sequence {sn } such that lim{sn+1 − sn } = 0.
v) A sequence {sn } such that lim{sn+1 − sn } = ∞.
8.3.9 Suppose that the sequence {sn }n diverges to ∞ (or to −∞). Prove that {sn }n
diverges. (The point of this exercise is to parse the definitions correctly.)

8.4 Convergence theorems via epsilon-N proofs

All of the theorems in this section are also proved in Section 8.5 with a different
method; here we use the epsilon-N formulation for proofs without explicitly resorting to
functions whose domains have a limit point. I recommend reading this section and omitting
Section 8.5.
242 Chapter 8: Sequences

Theorem 8.4.1 If a sequence converges, then its limit is unique.

Proof. Let {sn } be a convergent sequence. Suppose that {sn } converges to both L and L′ .
Then for any ǫ > 0, there exists an N such that |sn − L| < ǫ/2 for all n > N . Likewise, for
any ǫ > 0, there exists an N ′ such that |sn − L′ | < ǫ/2 for all n > N ′ . Then by the triangle
inequality, |L − L′ | = |L − sn + sn − L′ | ≤ |L − sn | + |sn − L′ | < ǫ/2 + ǫ/2 = ǫ. Since ǫ is
arbitrary, by Theorem 2.10.4 it must be the case that |L − L′ | = 0, i.e., that L = L′ .

Theorem 8.4.2 Suppose that lim{sn } = L and that L 6= 0. Then there exists a positive
number N such that for all integers n > N , |sn | > |L|/2. In particular, there exists a
positive number N such that for all integers n > N , sn 6= 0.

Proof. Note that p = |L|/2 is a positive real number. Since lim{sn } = L, it follows that
there exists a real number N such that for all integers n > N , |sn − L| < |L|/2. Then by
the reverse triangle inequality (proved in Theorem 2.10.3),
|sn | = |sn − L + L| = |(sn − L) + L| ≥ |L| − |sn − L| > |L| − |L|/2 = |L|/2.

Theorem 8.4.3 Suppose that lim sn = L and lim tn = K. Then

(1) (Constant rule) For any complex number c, lim{c} = c.
(2) (Linear rule) lim{1/n} = 0.
(3) (Sum/difference rule) lim{sn ± tn } = L ± K.
(4) (Scalar rule) For any complex number c, lim{csn } = cL.
(5) (Product rule) lim{sn tn } = LK.
(6) (Quotient rule) If tn 6= 0 for all n and K 6= 0, then lim{sn /tn } = L/K.
(7) (Power rule) For all positive integers m, lim{sm n}= L .
m

Proof. Part (1) was proved immediately after Definition 8.2.1. Part (2) was Example 8.2.2.
Part (3): Let ǫ > 0. Since lim sn = L, there exists a positive real number N1 such
that for all integers n > N1 , |sn − L| < ǫ/2. Since lim tn = K, there exists a positive real
number N2 such that for all integers n > N2 , |tn − K| < ǫ/2. Let N = max{N1 , N2 }. Then
for all integers n > N ,
|(sn ± tn ) − (L ± K)| = |(sn − L) ± (tn − K)|
≤ |sn − L| + |tn − K| (by the triangle inequality)
< ǫ/2 + ǫ/2 (since n > N ≥ N1 , N2 )
= ǫ.
This proves (3).
Part (4): Let ǫ > 0. Note that ǫ/(|c| + 1) is a positive number. Since lim sn = L, there
exists N such that for all integers n > N , |sn − L| < ǫ/(|c| + 1). Then for the same n,
 Section 8.4: Convergence theorems via epsilon-N proofs 243
|c|
|csn − cL| = |c| · |sn − L| ≤ |c|ǫ/(|c| + 1) = |c|+1 ǫ < ǫ. Since ǫ is any positive number, we
conclude that csn converges to cL.
Part (5): Let ǫ > 0. Since lim{sn } = L, there exists N1 such that for all integers
n > N1 , |sn −L| < 1. Thus for all such n, |sn | = |sn −L+L| ≤ |sn −L|+|L| < 1+|L|. There
also exists N2 such that for all integers n > N2 , |sn − L| < ǫ/(2|K| + 1). By assumption
lim{tn } = K there exists N3 such that for all integers n > N3 , |tn − K| < ǫ/(2|L| + 2). Let
N = max{N1 , N2 , N3 }. Then for all integers n > N ,
|sn tn − LK| = |sn tn − sn K + sn K − LK| (by adding a clever 0)
= |(sn tn − sn K) + (sn K − LK)|
≤ |sn tn − sn K| + |sn K − LK| (by the triangle inequality)
= |sn (tn − K)| + |(sn − L)K|
= |sn | · |tn − K| + |sn − L| · |K|
ǫ ǫ
< (|L| + 1) · + · |K|
2|L| + 2 2|K| + 1
ǫ ǫ
< +
2 2
= ǫ,
which proves (5).
Part (6): Let ǫ > 0. Since lim{sn } = L, there exists N1 such that for all integers
n > N1 , |sn − L| < 1. Thus for all such n, |sn | = |sn − L + L| ≤ |sn − L| + |L| < 1 + |L|.
There also exists N2 such that for all integers n > N2 , |sn − L| < |K|ǫ2 . By assumption
2
|K| ǫ
lim{tn } = L there exists N3 such that for all integers n > N3 , |tn − K| < 4(1+|L|) . Since
K 6= 0, by Theorem 8.4.2 there exists a positive number N4 such that for all integers
n > N4 , |tn | > |K|/2. Let N = max{N1 , N2 , N3 , N4 }. Then for all integers n > N ,
   
 sn L   K · s n − L · tn 
 − = 
 tn K  K · tn 
 
 K · s n − tn s n + tn s n − L · tn 
=   (by adding a clever 0)
K · tn 
   
 K · s n − tn s n   tn s n − L · tn 
≤  +
 
 (by the triangle inequality)

K · tn K · tn
1 1 1
= |K − tn ||sn | + |L − sn |
|K| |tn | K
|K|2 ǫ 1 2 |K|ǫ 1
< (1 + |L|) +
4(1 + |L|) |K| |K| 2 |K|
ǫ ǫ
= +
2 2
= ǫ,
which proves (6).
244 Chapter 8: Sequences

Part (7): The proof is by induction on m, the base case being the assumption. If
limn→∞ {1/nm−1 } = 0, then by the product rule, limn→∞ {1/nm } = limn→∞ {1/nm−1 ·
1/n} = limn→∞ {1/nm−1 } · limn→∞ {1/n} = 0. This proves (7).
5n−2
Example 8.4.4 Suppose sn = 3n+4 . To prove that lim sn = 5/3, we note that sn =
5n−2 1/n 5−2/n
3n+4 1/n = 3+4/n .By the linear rule, lim(1/n) = 0, so by the scalar rule, lim(2/n) =
lim(4/n) = 0. Thus by the constant, sum, and difference rules, lim{5 − 2/n} = 5 and
lim{3 + 4/n} = 3, so that by the quotient rule, lim sn = 5/3.
2 2
3n+2 1/n 3/n+2/n
Example 8.4.5 Let sn = n3n+2 2 −3 . Note that sn = n2 −3 · 1/n2 = 1−3/n2 . By the linear
rule, lim 1/n = 0, so that by the scalar rule, lim 3/n = 0, and by the product and scalar
rules, lim 2/n2 = lim 3/n2 = 0. Thus by the sum and difference rules, lim{3/n + 2/n2 } = 0
and lim{1 − 3/n2 } = 1. Finally, by the quotient rule, lim sn = 0/1 = 0.

Theorem 8.4.6 (Power, polynomial, rational rules for sequences) For any positive
integer m, limn→∞ {1/nm } = 0. If f is a polynomial function, then lim{f (1/n)} = f (0). If
f is a rational function that is defined at 0, then lim{f (1/n)} = f (0).

Proof. By the linear rule (in Theorem 8.4.3), limn→∞ {1/n} = 0, and by the power rule,
for all positive integers m, limn→∞ {1/nm } = 0.
Now write f (x) = a0 + a1 x + · · · + ak xk for some non-negative integer k and some
complex numbers a0 , a1 , . . . , ak . By the constant, power and repeated sum rules,

lim f (1/n) = lim a0 + a1 (1/n) + a2 (1/n)2 + · · · + ak (1/n)k
n→∞ n→∞

= a0 + lim a1 (1/n) + lim a2 (1/n)2 + · · · + lim ak (1/n)k

n→∞ n→∞ n→∞
2 k
= a0 + a1 · 0 + a2 · 0 + · · · + ak · 0
= a0 = f (0).
This proves the polynomial rule.
Finally, let f be a rational function. Write f (x) = g(x)/h(x), where g, h are polynomial
functions with h(0) 6= 0. By the just-proved polynomial rule, limn→∞ g(1/n) = g(0),
limn→∞ h(1/n) = h(0) 6= 0, so that by the quotient rule, limn→∞ f (1/n) = g(0)/h(0) =
f (0).

Theorem 8.4.7 (The composite rule for sequences) Suppose that lim sn = L. Let
g be a function whose domain contains L and all terms sn . Suppose that g is continuous
at L. Then lim g(sn ) = g(L).

Proof. Let ǫ > 0. Since g is continuous at L, there exists a positive number δ > 0 such
that for all x in the domain of g, if |x − L| < δ then |g(x) − g(L)| < ǫ. Since lim sn = L,
 Section 8.4: Convergence theorems via epsilon-N proofs 245

there exists a positive number N such that for all integers n > N , |sn − L| < δ. Hence for
the same n, |g(sn) − g(L)| < ǫ.
In particular, since the absolute value function, the real part, and the imaginary part
functions are continuous everywhere, we immediately conclude the following:

Theorem 8.4.8 Suppose that lim sn = L. Then

(1) lim |sn | = |L|.
(2) lim Re sn = Re L.
(3) lim Im sn = Im L.

Furthermore, since the real and imaginary parts determine a complex number, we
moreover get:

Theorem 8.4.9 A sequence {sn } of complex numbers converges if and only if the se-
quences {Re sn } and {Im sn } of real numbers converge.

Proof. By Theorem 8.4.8 it suffices to prove that if lim{Re sn } = a and lim{Im sn } = b,

then lim{sn } = a + bi. Let ǫ > 0. By assumptions there exist positive real numbers N1 , N2
such that for all integers n > N1 | Re sn − a| < ǫ/2 and such that for all integers n > N2
| Im sn − b| < ǫ/2. Set N = max{N1 , N2 }. Then for all integers n > N ,
|sn − (a + bi)| = | Re sn + i Im sn − a − bi| ≤ | Re sn − a||i|| Im sn − b| < ǫ/2 + ǫ/2 = ǫ.

Theorem 8.4.10 (Comparison of sequences) Let s and t be convergent sequences of

complex numbers. Suppose that |sn | ≤ |tn | for all except finitely many n. Then | lim sn | ≤
| lim tn |.
If in addition for all except finitely many n, sn , tn are real numbers with sn ≤ tn , then
lim sn ≤ lim tn .

Proof. Let L = lim sn , K = lim tn . By Theorem 8.4.8, lim |sn | = |L|, lim |tn | = |K|.
Suppose that |L| > |K|. Set ǫ = (|L| − |K|)/2|. By the definition of convergence, there
exist N1 , N2 > 0 such that if n is an integer, n > N1 implies that −ǫ < |sn | − |L| < ǫ, and
n > N2 implies that −ǫ < |tn | − |K| < ǫ. Let N3 be a positive number such that for all
integers n > N3 , |sn | ≤ |tn |. If we let N = max{N1 , N2 , N3 }, then for integers n > N we
have that |tn | < |K| + ǫ = (|L| + |K|)/2 = |L| − ǫ < |sn |. This contradicts the assumption
|sn | ≤ |tn |, so that necessarily |L| ≤ |K|.
The proof of the second part is very similar and is left to the exercises.
246 Chapter 8: Sequences

Theorem 8.4.11 (The squeeze theorem for sequences) Suppose that s, t, u are se-
quences of real numbers and that for all n ∈ N+ , sn ≤ tn ≤ un . If lim s and lim u both
exist and are equal, then lim t exists as well and
lim s = lim t = lim u.

Proof. Set L = lim s = lim u. Let ǫ > 0. Since lim s = L, there exists a positive N1 such
that for all integers n > N1 , |sn − L| < ǫ. Since lim u = L, there exists a positive N2 such
that for all integers n > N2 , |un − L| < ǫ. Set N − max{N1 , N2 }. Let n be an integer
strictly greater than N . Then −ǫ < sn − L ≤ tn − L ≤ un − L < ǫ, so that |tn − L| < ǫ.
Since ǫ was arbitrary, this proves that lim t = L.

Example 8.4.12 lim(n + 1)1/n = 1.

n+1 1/n

1/n
Proof. For n ≥ 2, 1 ≤ n ≤ 32 ≤ n1/n. Thus by the previous theorem and
n+1 1/n

by Example 8.2.9, lim{ n ) = 1. Hence by Theorem 8.4.3 and by Example 8.2.9,
n
n o n
o
1/n n+1 1/n 1/n n+1 1/n
lim{(n + 1) } = lim n
n = lim n
lim{n1/n} = 1.

Exercises for Sections 8.4 and 8.5

8.4.1 Compute the following limits, and justify:
i) lim n21−2 .
2
3n +20
ii) lim 2n3 −20 .
−2
iii) lim −5n2n
−2 −n−1 .
2
iv) lim n 4n
+4n−2
2 +2 .
3
v) lim −3n +2
2n3 +n .
−4
2n +2
vi) lim 3n−2 +n−1 +2 .

8.4.2 Give an example of a sequence {sn } and a number L such that lim |sn | = |L| but
{sn } does not converge. Justify your example.
n o
(−1)n
8.4.3 Prove that lim n = 0.
†8.4.4 Let {sn } be a convergent sequence of positive real numbers. Prove that lim sn ≥ 0.

m
8.4.5 Prove that for all integers m, limn→∞ n+1n = 1.
8.4.6 Prove that limn→∞ {3 − n2 } = 3.
√ √
8.4.7 Prove that lim 3 n + 1 − 3 n = 0.
 Section 8.5: Convergence theorems via functions 247


1 n
8.4.8 By Example 8.2.8, lim 1+ n = e. Determine the following limits:

n
i) n n+1
n . o


1 2n
ii) 1 + 2n .
n
o
2n
iii) 1 + n1 .


1 n

iv) n 1 + 2n .o

n+1
v) 1 + n1 .
√
8.4.9 Let s1 be a positive real number. For each n ≥ 1, let sn+1 = sn . Prove that
lim sn = 1. (Hint: Prove that lim sn = lim s2n .)

8.5 Convergence theorems via functions

The results in this section are the same as those in Section 8.4, but here they are proved
with theorems about limits of functions that were proved in Chapter 4. So, a connection is
made between limits of functions and limits of sequences. The reader may omit this section
(or the previous one). This section is more abstract; one has to keep in mind connections
with functions as well as theorems about limits of functions to get at theorems about limits
of sequences. Exercises for this section appear at the end of Section 8.4.
For any sequence s we can define a function f : {1/n : n ∈ N+ } → C with f (1/n) = sn .
Conversely, for every function f : {1/n : n ∈ N+ } → C we can define a sequence s with
sn = f (1/n).
The domain of f has exactly one limit point, namely 0. With this we have the usual
notion of limx→0 f (x) with standard theorems from Section 4.3,

Theorem 8.5.1 Let s, f be as above. Then lim sn = L if and only if limx→0 f (x) = L.

Proof. (⇒) Suppose that lim sn = L. We have to prove that limx→0 f (x) = L. Let ǫ > 0.
By assumption lim sn = L, there exists a positive real number N such that for all integers
n > N , |sn − L| < ǫ. Let δ = 1/N . Then δ is a positive real number. Let x be in the
domain of f such that 0 < |x − 0| < δ. Necessarily x = 1/n for some positive integer
n. Thus 0 < |x − 0| < δ simply says that 1/n < δ = 1/N , so that N < n. But then by
assumption |f (1/n) − L| = |sn − L| < ǫ, which proves that limx→0 f (x) = L.
(⇐) Now suppose that limx→0 f (x) = L. We have to prove that lim sn = L. Let
ǫ > 0. By assumption limx→0 f (x) = L there exists a positive real number δ such that for
all x in the domain of f , if 0 < |x − 0| < δ then |f (x) − L| < ǫ. Set N = 1/δ. Then N
is a positive real number. Let n be an integer greater than N . Then 0 < 1/n < 1/N = δ,
so that by assumption |f (1/n) − L| < ǫ. Hence |sn − L| = |f (1/n) − L| < ǫ, which proves
that lim sn = L.
248 Chapter 8: Sequences

Example 8.5.2 (Compare the reasoning in this example with the epsilon-N proofs of
5n−2 1/n 5−2/n
Section 8.2.) Let sn = 5n−2
3n+4 . We note that sn = 3n+4 · 1/n = 3+4/n . The corresponding
function f : {1/n : n ∈ N+ } is f (x) = 5−2x 3+4x , and by the scalar, sum, difference, and
quotient rules for limits of functions, limx→0 f (x) = 5−0
3+0 = 5/3, so that by Theorem 8.5.1,
lim sn = 5/3.

3n+2 3n+2 1/n2 3/n+2/n2

Example 8.5.3 Suppose sn = n2 −3 . Note that sn = 2
n −3 · 1/n2 = 1−3/n2 . The
+ 3x+2x2
corresponding function f : {1/n : n ∈ N } is f (x) = 1−3x2
, and by the scalar,
sum,
difference, product, and quotient rules for limits of functions, lim sn = limx→0 f (x) =
0+0
1−0 = 0.

Theorem 8.5.4 The limit of a converging sequence is unique.

Proof. Let {sn } be a convergent sequence. Suppose that {sn } converges to both L
and L′ . Let f : {1/n : n ∈ N+ } be the function corresponding to s. By Theorem 8.5.1,
limx→0 f (x) = L and limx→0 f (x) = L′ . By Theorem 4.3.1, L = L′ . This proves uniqueness
of limits for sequences.

Theorem 8.5.5 Suppose that lim{sn } = L and that L 6= 0. Then there exists a positive
number N such that for all integers n > N , |sn | > |L|/2. In particular, there exists a
positive number N such that for all integers n > N , sn 6= 0.

Proof. Let f be the function corresponding to s. Then limx→0 f (x) = L, by Theorem 4.3.2
there exists δ > 0 such that for all x in the domain of f , if x < δ then |f (x)| > |L|/2.
Set N = 1/δ. Let n be an integer strictly greater than N . Then 1/n < 1/N = δ, so
|sn | = |f (1/n)| > |L|/2.

Theorem 8.5.6 Suppose that lim sn = L and lim tn = K. Then

(1) (Constant rule) For any complex number c, lim{c} = c.
(2) (Linear rule) lim{1/n} = 0.
(3) (Sum/difference rule) lim{sn ± tn } = L ± K.
(4) (Scalar rule) For any complex number c, lim{csn } = cL.
(5) (Product rule) lim{sn tn } = LK.
(6) (Quotient rule) If tn 6= 0 for all n and K 6= 0, then lim{sn /tn } = L/K.
(7) (Power rule) For all positive integers m, lim{sm n}= L .
m

Proof. Let f, g : {1/n : n ∈ N+ } → C be given by f (1/n) = sn , g(1/n) = tn . By

Theorem 8.5.1, limx→0 f (x) = L and limx→0 g(x) = K. Theorem 4.3.3 proves parts (3),
(4), (5), (6) for f, g, hence via Theorem 8.5.1 also for s, t. Theorem 4.3.4 finishes the proof
of (7).
 Section 8.5: Convergence theorems via functions 249

For part (1), declare f (1/n) = c and for part (2), declare f (1/n) = 1/n. Again
Theorems 4.3.3 and 8.5.1 easily finish the proofs of (1), (2).
The following theorem for sequences follows immediately from the corresponding
power, polynomial, and rational rules for functions:

Theorem 8.5.7 (Power, polynomial, rational rules for sequences) Let f be a poly-
nomial function. Then lim{f (1/n)} = f (0). In particular, for any positive integer m,
limn→∞ {1/nm } = 0. If f is a rational function that is defined at 0, then lim{f (1/n)} =
f (0).

Theorem 8.5.8 (The composite rule for sequences) Suppose that lim sn = L. Let g
be a function whose domain contains L and all terms sn . Suppose that g is continuous at
L. Then lim g(sn ) = g(L).

Proof. Let f (1/n) = sn . By Theorem 8.5.1, limx→0 f (x) = L, and by assump-

tion limx→L g(x) = g(L). Thus by the composite function theorem (Theorem 4.3.9),
limx→0 g(f (x)) = g(L), so that by Theorem 8.5.1, lim g(sn ) = g(L).
In particular, since the absolute value function, the real part, and the imaginary part
functions are continuous everywhere, we immediately conclude the following:

Theorem 8.5.9 Suppose that lim sn = L. Then

(1) lim |sn | = |L|.
(2) lim Re sn = Re L.
(3) lim Im sn = Im L.

Since the real and imaginary parts determine a complex number, we also have:

Theorem 8.5.10 A sequence {sn } of complex numbers converges if and only if the se-
quences {Re sn } and {Im sn } of real numbers converge.

Proof. Let f (1/n) = sn . By Theorem 4.3.8, limx→0 f (x) = L if and only if

limx→0 Re f (x) = Re L and limx→0 Im f (x) = Im L, which is simply a restatement of
the theorem.
250 Chapter 8: Sequences

Theorem 8.5.11 (Comparison of sequences) Let s and t be convergent sequences of

complex numbers. Suppose that |sn | ≤ |tn | for all except finitely many n. Then | lim sn | ≤
| lim tn |.
If in addition for all except finitely many n, sn , tn are real numbers with sn ≤ tn , then
lim sn ≤ lim tn .

Proof. Let A be the set of those 1/n for which |sn | ≤ |tn | in the first case and for which
sn ≤ tn in the second case. Let f, g : A → R be the functions f (1/n) = |sn |, g(1/n) = |tn |
in the first case, and f (1/n) = sn , g(1/n) = tn in the second case. By assumption
for all x in the domain, f (x) ≤ g(x). Since 0 is a limit point of the domain (despite
omitting finitely many 1/n) and since by Theorem 8.5.1, limx→0 f (x) and limx→0 g(x)
both exist, by Theorem 4.3.10, limx→0 f (x) < limx→0 g(x). In the first case, this translates
to | lim sn | ≤ | lim tn |, and in the second case it translates to lim sn ≤ lim tn .

Theorem 8.5.12 (The squeeze theorem for sequences) Suppose that s, t, u are se-
quences of real numbers and that for all n ∈ N+ , sn ≤ tn ≤ un . If lim s and lim u both
exist and are equal, then lim t exists as well and
lim s = lim t = lim u.

Proof. Let f, g, h : {1/n : n ∈ N+ } → C be functions defined by f (1/n) = sn , g(1/n) = tn ,

h(1/n) = un . The assumption is that for all x in the domain of f, g, h, f (x) ≤ g(x) ≤ h(x),
and by Theorem 8.5.1 that limx→0 f (x) = limx→0 h(x). Then by the squeeze theorem
for functions (Theorem 4.3.11), limx→0 f (x) = limx→0 g(x) = limx→0 h(x). Hence by
Theorem 8.5.1, lim s = lim t = lim u.

Example 8.5.13 lim(n + 1)1/n = 1.

n+1 1/n


3 1/n
Proof. For n ≥ 2, 1 ≤n n o≤ 2 ≤ n1/n. Thus by the previous theorem and


n+1 1/n
by Example 8.2.9, lim n = 1. Hence by Theorem 8.5.6 and by Example 8.2.9,
n
o n
o
1/n n+1 1/n 1/n n+1 1/n
lim{(n + 1) } = lim n n = lim n lim{n1/n} = 1.

Exercises for this section appear at the end of Section 8.4.

 Section 8.6: Bounded sequences, monotone sequences, ratio test 251

8.6 Bounded sequences, monotone sequences, ratio test

Definition 8.6.1 A sequence {sn } is bounded if there exists a positive real number B
such that for all integers n, |sn | ≤ B.
If all sn are real numbers, we say that {sn } is bounded above (resp. below) if there
exists a real number M such that for all positive integers n, sn ≤ M (resp. sn ≥ M ). In
other words, {sn } is bounded if the set {sn : n ∈ N+ } is a subset of B(0, M ) for some real
number M .
 n √
4+i
Sequences {4+5i}, {1/n}, { 5+2i }, {(−1)n }, {3−4in}, { 3n4n
2 −4 }, {
n
n} are bounded
2 √ √
(the latter by Example 1.5.6), but {n}, {(−1)nn}, { n n+1 }, { n n!}, {2 n } are not bounded.
The next theorem provides many examples of bounded sequences.

Theorem 8.6.2 Every convergent sequence is bounded.

Proof. Let {sn } be a convergent sequence with limit L. Thus there exists a positive integer
N such that for all integers n > N , |sn −L| < 1. Set B = max{|s1 |, |s2|, |s3|, . . . , |sN +1 |, |L|+
1}. Then for all positive integers n ≤ N , |sn | ≤ B by definition of B, and for n > N ,
|sn | = |sn − L + L| ≤ |sn − L| + |L| < 1 + |L| ≤ B.
Another proof of the theorem above is given in the next section via Cauchy sequences.

Definition 8.6.3 A sequence {sn } of real numbers is called non-decreasing (resp. non-
increasing, strictly increasing, strictly decreasing) if for all n, sn ≤ sn+1 (resp.
sn ≥ sn+1 , sn < sn+1 , sn > sn+1 ). Any such sequence is called monotone.
2
Sequences {1/n}, {−n} are strictly decreasing, { n n+1 } is strictly increasing, {(−1)n n}
2
is neither increasing nor decreasing, { n n+5 }n≥1 is neither increasing nor decreasing, but
2
{ n n+5 }n≥2 is strictly increasing.

Theorem 8.6.4 (Bounded monotone sequences) Let {sn } be a bounded sequence

of real numbers such that for some integer N , {sn }n≥N is non-decreasing (resp. non-
increasing). Then lim sn exists, and equals the least upper bound (resp. greatest lower
bound) of the set {sN , sN +1 , sN +2 , . . .}.

Proof. Suppose that for all n ≥ N , sn ≤ sn+1 . By the Least upper theorem (Axiom 2.9.1),
the least upper bound of the set {sN , sN +1 , sN +2 , . . .} exists. Call it L.
Let ǫ > 0. Since L is the least upper bound, there exists a positive integer N ′ ≥ N
such that 0 ≤ L − sN ′ < ǫ. Hence for all integers n > N ′ , sN ′ ≤ sn , so that

0 ≤ L − sn ≤ L − sN ′ < ǫ,
which proves that for all n > N ′ , |sn − L| < ǫ. Thus lim sn = L.
252 Chapter 8: Sequences

The proof of the case of sn ≥ sn+1 for all n ≥ N is similar.

The theorem below was already proved in Example 8.2.7.

Theorem 8.6.5 (Ratio test for sequences) Let r ∈ C with |r| < 1. Then lim r n = 0.

Proof. If r = 0, the sequence is the constant zero sequence, so of course its limit is 0. Now
suppose that r 6= 0. By Exercise 2.7.2, for all positive integers n, 0 < |r|n+1 < |r|n . Thus
the sequence {|r|n} is a non-increasing sequence that is bounded below by 0 and above
by 1. By Theorem 8.6.4, L = lim |r|n = inf{|r|n : n ∈ N+ }. Since 0 is a lower bound and
L is the greatest of lower bounds of {|r|n : n ∈ N+ }, necessarily 0 ≤ L.
Suppose that L > 0. Then L(1 − |r|)/(2|r|) is a positive number. Since L is the
infimum of the set {|r|, |r|2, |r|3, . . .}, there exists a positive integer N > N1 such that
0 ≤ |r|N − L < L 1−|r|
2|r| . By multiplying by |r| we get that
 
N +1 1 − |r| L(1 − |r|) (1 + |r|) (1 + 1)
|r| < |r|L + L = L |r| + =L <L = L,
2 2 2 2
and since L is the infimum of all powers of |r|, we get that L ≤ |r|N +1 < L, which is a
contradiction. So necessarily L = 0. Hence for every ǫ > 0 there exists N ∈ R+ such that
for all integers n > N , ||r n | − 0| < ǫ. But this says that |r n − 0| < ǫ, so that lim r n = 0 as
well.

Theorem 8.6.6 (Ratio test for sequences) Let {sn } be a sequence of non-zero  complex
 sn+1 
numbers and let L be a real number in the interval [0, 1). Assume that lim  sn  = L or
 
 sn+1 
that there exists a positive integer K such that for any integer n ≥ K,  sn  ≤ L.
Then lim sn = 0.

Proof. Let r be a real number strictly between L and 1. Then r and r − L are positive
numbers.
Under the first
 (limit)
 condition, there exists a positive number K such that for all
 sn+1  
integers n > K,  sn  − L < r − L. Then for all n > K,
      
 sn+1   sn+1   sn+1  
 =  − L + L ≤   − L + L ≤ r − L + L = r.
 sn   sn   sn  

Thus both  conditions say that there exists a positive K such that for all integers
 sn+1 
n > K,  sn  ≤ r. We may replace this K by any larger number, and the conclusion still
holds. So from now on we assume that K is a positive integer.
Let ǫ > 0. By Theorem 8.6.5, lim r n = 0. Thus there exists M ≥ 0 such that for all
ǫr K
integers n > M , |r n − 0| < |sK|
. Set N = max{M, K}. Let n be an integer strictly greater
 Section 8.6: Bounded sequences, monotone sequences, ratio test 253

than N . Then
 
 sn s n−1 s n−2 s K+1 
|sn − 0| =  · · ···· · sK 
sn−1 sn−2 sn−3 sK
       
 sn   sn−1   sn−2   sK+1 
=  · · ···· 
 sK  · |sK |
sn−1   sn−2   sn−3 
≤ r n−K |sK |
= r n |sK |r −K
< ǫ.

Exercises for Section 8.6

8.6.1 Compute the limits. (Use the comparison test, ratio test, decreasing property, or
other tricks.)
i) limn→∞ 21n = 0.
(−1)n
ii) limn→∞ 3n
= 0.
n
(−4)
iii) limn→∞ 7n = 0.
n
k
iv) limn→∞ n! = 0 for all k ∈ C.
n
v) limn→∞ kn = 0 for all non-zero k ∈ C with |k| > 1.
nm
vi) limn→∞ kn = 0 for all non-zero k ∈ C with |k| > 1 and all integers m.
8.6.2 Give an example of a sequence {sn } of non-zero complex numbers such that
lim{sn+1 /sn } exists and has absolute value strictly smaller than 1.
8.6.3 Give an example of a sequence {sn } of non-zero complex numbers such that
lim{sn+1 /sn } = i/2.
√
8.6.4 Let sn be the nth Fibonacci number. Prove that lim sn+1
sn =
1+ 5
2 . (Hint: Exer-
cise 1.5.31.)
8.6.5 Let r ∈ C satisfy |r| > 1. Prove that the sequence {r n } is not bounded.
8.6.6 (Ratio test again, and compare to Theorem 8.6.6) Let {sn } be a sequence of non-zero
complex numbers such that lim sn+1
sn = L. Suppose that |L| < 1. Prove that lim sn = 0.
8.6.7 Consider sequences {(−1)n } and {1}. For each, determine whether it converges, and
the limit of ratios | sn+1
sn
|. Comment why the Ratio Test does not apply.
8.6.8 Let {sn } be a sequence of real numbers such that for all n ≥ 1, sn ≤ sn+1 (resp.
sn ≥ sn+1 ). Prove that {sn } is convergent if and only if it is bounded.

n
8.6.9 Prove that the sequence 1 + n1 n
is strictly increasing. (Hint: Exercise 7.5.10.)
√n
8.6.10 Prove that the sequence { n! } is strictly increasing.
254 Chapter 8: Sequences

8.6.11 Let {sn } be a non-decreasing sequence. Prove that the set {s1 , s2 , s3 , . . .} is
bounded below.
8.6.12 (Monotone sequences) Let {sn } be a monotone sequence of real numbers.
i) Suppose that {sn } is not bounded above. Prove that {sn } is non-decreasing and
that lim sn = ∞.
ii) Suppose that {sn } is not bounded below. Prove that {sn } is non-increasing and
that lim sn = −∞.
iii) Prove that lim sn is a real number if and only if {sn } is bounded.

8.7 Cauchy sequences, completeness of R, C

So far, to determine convergence of a sequence required knowing the limit. In this

section we prove an alternate machinery for deciding that a sequence converges without
knowing what its limit is. This will be used for subsequential limits and in comparison
theorems for convergence of series in the next chapter.

Definition 8.7.1 A sequence {sn } is Cauchy if for all ǫ > 0 there exists a positive real
number N such that for all integers m, n > N , |sn − sm | < ǫ.

Theorem 8.7.2 Every Cauchy sequence is bounded.

Proof. Let {sn } be a Cauchy sequence. Thus for ǫ = 1 there exists a positive integer N
such that for all integers m, n > N , |sn − sm | < 1. Then the set {|s1 |, |s2|, . . . , |sN |, |sN +1 |}
is a finite and hence a bounded subset of R. Let M ′ an upper bound of this set, and let
M = M ′ + 1. It follows that for all n = 1, . . . , N , |sn | < M , and for n > N , |sn | =
|sn − sN +1 + sN +1 | ≤ |sn − sN +1 | + |sN +1 | < 1 + M ′ = M . Thus {sn } is bounded by M .

Theorem 8.7.3 Every convergent sequence is Cauchy.

Proof. Let {sn } be a convergent sequence. Let L be the limit. Let ǫ > 0. Since lim sn = L,
there exists a positive real number N such that for all n > N , |sn − L| < ǫ/2. Thus for all
integers m, n > N ,

|sn − sm | = |sn − L + L − sm | ≤ |sn − L| + |L − sm | < ǫ/2 + ǫ/2 = ǫ.

 Section 8.7: Cauchy sequences, completeness of R, C 255

Remark 8.7.4 It follows that every convergent sequence is bounded (this was already
proved in Theorem 8.6.2).
The converse of Theorem 8.7.3 is not true if the field in which we are working is Q.
√
For example, let sn be the decimal approximation of 2 to n digits after the decimal point.
Then {sn } is a Cauchy sequence of rational numbers: for every ǫ > 0, let N be a positive
integer such that 1/10N < ǫ. Then for all integers n, m > N , sn and sm differ at most in
digits N + 1, N + 2, . . . beyond the decimal point, so that |sn − sm | ≤ 1/10N < ǫ. But {sn }
does not have a limit in Q, so that {sn } is a Cauchy but not a convergent sequence.
But over R and C, all Cauchy sequences are convergent, as we prove next.

Theorem 8.7.5 (Completeness of R, C) Every Cauchy sequence in R or C is conver-

gent.

Proof. First let {sn } be a Cauchy sequence in R (as opposed to in C). By Theorem 8.7.2,
{sn } is bounded. It follows that all subsets {s1 , s2 , s3 , . . .} are bounded too. In particular,
by the Least upper bound theorem (Axiom 2.9.1), un = sup{sn , sn+1 , sn+2 , . . .} is a real
number. For all n, un ≥ un+1 because un is the supremum of a larger set. Any lower
bound on {s1 , s2 , s3 , . . .} is also a lower bound on {u1 , u2 , u3 , . . .}. Thus by Theorem 8.6.4,
the monotone sequence {un } has a limit L = inf{u1 , u2 , u3 , . . .}.
We claim that L = lim{sn }. Let ǫ > 0. Since {sn } is Cauchy, there exists N1 > 0
such that for all integers m ≥ n > N1 , |sn − sm | < ǫ/2. Thus if we fix n > N1 , then for
all m ≥ n, we have that sm < sn + ǫ/2. But un is the least upper bound on all sm for
m ≥ n, so that sm ≤ un < sn + ǫ/2, and in particular, sn ≤ un < sn + ǫ/2. It follows
that |sn − un | < ǫ/2 for all integers n > N1 . Since L = inf{u1 , u2 , u3 , . . .}, there exists an
integer N2 such that 0 ≤ uN2 − L < ǫ/2. Set N = max{N1 , N2 }. Let n > N be an integer.
By the definition of the un , L ≤ un ≤ uN ≤ uN2 , so that 0 ≤ un − L ≤ uN2 − L < ǫ/2.
Hence

|sn − L| = |sn − un + un − L| ≤ |sn − un | + |un − L| < ǫ/2 + ǫ/2 = ǫ.

This proves that every real Cauchy sequence converges.
Now let {sn } be a Cauchy sequence in C. We leave it to Exercise 8.7.1 to prove that
then {Re sn } and {Im sn } are Cauchy. By what we have proved for real sequences, there
exist a, b ∈ R such that lim Re sn = a and lim Im sn = b. Then by Theorem 8.5.10 or
Theorem 8.4.9, {sn } converges to a + bi.
256 Chapter 8: Sequences

Example 8.7.6 (Harmonic sequence) The sequence {1 + 12 + 31 + · · · + n1 }n is not Cauchy,

does not converge, is monotone, and is not bounded.

Proof. Let sn be the nth term of the sequence. If m > n, then

1 1 1 1 m−n n
sm − sn = + + + ···+ ≥ =1− .
m m−1 m−2 n+1 m m
Thus if m = 1000n, then sm − sn > 1 − 1/1000 = 0.999. In particular, if ǫ = 1/2, then
|sm − sn | <
6 ǫ. Thus the sequence is not Cauchy. By Theorem 8.7.3 it is not convergent,
and by Theorem 8.7.2 it is not bounded. It is certainly strictly increasing.
So necessarily the terms of {1 + 21 + 31 + · · · + n1 } get larger and larger beyond bound.
But this sequence gets large notoriously slowly – the 1000th term of the sequence is smaller
than 8, the 10000th term is smaller than 10, the 100000th term is barely larger than 12.
(One could lose patience in trying to see how this sequence grows without bound.)
Here is a thought experiment related to this: we can stack finitely many books to form
a bridge to the Moon and beyond! All books weigh one unit of weight and are one unit
along the longest side. The books are stacked so that each one protrudes out from the
heap below as much as possible and with keeping the center of mass stable. The topmost
book, by the uniform assumption, protrudes out 12 of its length. Actually, for the book
to be stable this should actually be 21 minus a tad, but that tad can be taken to be say
1
, say, or much much less. We will ignore this tiny tad, but an interested reader
googolplex
may wish to work through the stacking below with (varied!) corrections being incorporated
in all the centers of mass and all the protrusions.
This diagram shows the top two books.

The center of mass of the top two books is clearly at 43 from the right-hand edge of the
bottom book, so that the third book from the top down should protrude from underneath
the second one 41 .

2·1+1· 12
The center of mass of this system, measured from the rightmost edge, is 3 = 56 ,
so that the fourth book has to protrude out 16 .

3·1+1· 12
The center of mass of this system, measured from the rightmost edge, is 4 = 78 ,
so that the fifth book has to protrude out 18 units.
(n−1)·1+1· 1
In general, the center of mass of the top n books is at n
2
= 2n−1
2n measured
1
from the rightmost edge, so that the (n + 1)st book should protrude out by 2n units.
 Section 8.7: Cauchy sequences, completeness of R, C 257

Thus the total protrusion of the top n books equals

 
1 1 1 1
1+ + +···+
2 2 3 n−1
units. We just proved that this sum grows beyond limit. Thus we can reach the Moon
with enough books (and a platform to stand on).

Exercises for Section 8.7

† 8.7.1 (Invoked in Theorems 8.8.4 and 8.7.5.) Prove that {sn } is Cauchy if and only if
{Re sn } and {Im sn } are Cauchy.
8.7.2 Suppose that {sn } and {tn } are Cauchy sequences.
i) Prove that {sn ± tn } is a Cauchy sequence.
ii) Prove that {sn · tn } is a Cauchy sequence.
iii) Suppose that {sn } is Cauchy, and that for some positive number B, |tn | ≥ B for
all n ∈ N+ . Prove that {sn /tn } is Cauchy.
iv) Prove that for all c ∈ C, {csn } is a Cauchy sequence.
8.7.3 Give examples of non-Cauchy sequences {sn }, {tn } such that {sn + tn } is a Cauchy
sequence. Why does this not contradict the previous exercise? Repeat for {sn · tn }.
8.7.4 Let r ∈ C satisfy |r| > 1. Prove that lim r n diverges. (Hint: Check the Cauchy
property.)
8.7.5 Review the proof in Example 8.3.2 that {(−1)n} is divergent. Give another proof
using the contrapositive of Theorem 8.7.3.
8.7.6 Give examples of sequences with the listed properties.
i) A Cauchy sequence in Q that is not convergent.
ii) A bounded sequence that is not convergent.
iii) A bounded sequence that is not Cauchy.
iv) A real increasing sequence that is not Cauchy.
v) A real Cauchy sequence that is increasing.
8.7.7 Say why the sequences below do not exist.
i) A convergent sequence that is not Cauchy.
ii) A Cauchy sequence in R that is not convergent.
iii) A bounded real increasing sequence that is not Cauchy.
258 Chapter 8: Sequences

8.8 Subsequences
Definition 8.8.1 A subsequence of an infinite sequence {sn } is an infinite sequence
{sk1 , sk2 , sk3 , . . .} where 1 ≤ k1 < k2 < k3 < · · · are integers. Notations for such a
subsequence are:
{skn }, {skn }n , {skn }∞
n≥1 , {skn }n≥1 , {skn }n∈N+ .

Note that for all n, kn ≥ n.

Examples 8.8.2 Every sequence is a subsequence of itself. Sequences {1/2n}, {1/3n},

{1/(2n + 1)} are subsequences of {1/n}, and {1}, {−1}, {(−1)n+1 } are subsequences of
{(−1)n }. The constant sequence {1} is not a subsequence of {1/n}, because the lat-
ter sequence does not have infinitely many terms equal to 1. If {sn } is the sequential
enumeration of Q+ on page 227, then {1/n}, {n}, { n+1n
}, { n12 } are subsequences, but
n
{ n+2 } = {1/3, 1/2, . . .} is not.

Theorem 8.8.3 A subsequence of a convergent sequence is convergent, with the same

limit. A subsequence of a Cauchy sequence is Cauchy.

Proof. Let {sn } be a convergent sequence, with limit L, and let {skn } be a subsequence.
Let ǫ > 0. By assumption there exists a positive number N such that for all integers
n > N , |sn − L| < ǫ. Since n ≤ kn , it follows that |skn − L| < ǫ. Thus {skn } converges.
The proof of the second part is similar.

Theorem 8.8.4 Every bounded sequence has a Cauchy subsequence.

Proof. This proof uses the halving construction already encountered in Construction 3.6.1,
and here the property P is that the subset contains infinitely many elements of the sequence.
Let {sn } be a bounded sequence (of real or complex numbers). Let M be a positive
real number such that for all n, |sn | ≤ M . Let a0 = c0 = −M and b0 = d0 = M . The
sequence {sn } has infinitely many (all) terms in the rectangle R0 = [a0 , b0 ] × [c0 , d0 ]. Set
l0 = 0. (If all sn are real, we may take c0 = d0 = 0, or perhaps better, ignore the second
coordinates.)
We prove below that for all m ∈ N+ there exists a subsequence {skn } all of whose
terms are in the rectangle Rm = [am , bm ] × [cm , dm ], where bm − am = 2−m (b0 − a0 ),
[am , bm ] ⊆ [am−1 , bm−1 ]. dm − cm = 2−m (d0 − c0 ), [cm , dm ] ⊆ [cm−1 , dm−1 ]. Furthermore,
we prove that there exists lm > lm−1 such that slm ∈ Rm .
Namely, given the (m − 1)st rectangle Rm−1 , integer lm−1 such that slm−1 ∈ Rm−1 ,
and a subsequence {skn } all of whose terms are in Rm−1 , divide Rm−1 into four equal-sized
subrectangles. Necessarily at least one of these four subrectangles contains infinitely many
 Section 8.8: Subsequences 259

elements of {skn }, so pick one such subrectangle, and call it Rm . Therefore there exists
a subsequence of {skn } that is contained in Rm , and that subsequence of {skn } is also a
subsequence of {sn }. We call it {skn′ }. Since we have infinitely many kn′ , in particular
there exists kn′ > lm−1 , and we set lm = kn′ . Thus slm ∈ Rm .
By construction, {sln }n is a subsequence of {sn }n . We next prove that {sln }n is a
Cauchy sequence. Let ǫ > 0. Since the either side length of the mth subrectangle Rm
equals the corresponding side length of R0 divided by 2m , by Exercise 2.7.22 there exists
a positive integer N such that any side length of RN is strictly smaller than the constant
ǫ/2. Let m, n > N be integers. Then slm , sln are in RN , so that
p
|slm − sln | ≤ (one side length of Rn )2 + (other side length of Rn )2 < ǫ.
The following is now an immediate consequence of Theorem 8.7.5:

Theorem 8.8.5 Every bounded sequence in C has a convergent subsequence.

Example 8.8.6 We work out the construction of a subsequence as in the proof on the
bounded sequence {(−1)n − 1}. For example, all terms lie on the interval [a0 , b0 ] = [−4, 4].
Infinitely many terms lie on [a1 , b1 ] = [−4, 0], and on this subinterval I arbitrarily choose the
second term, which equals 0. Infinitely many terms lie on [a2 , b2 ] = [−4, −2], in particular,
I choose the third term −2. After this all terms of the sequence in [a2 , b2 ] are −2, so that
we have built the Cauchy subsequence {0, −2, −2, −2, . . .} (and subsequent [an , bn ] all have
bn = −2). We could have built the Cauchy subsequence {−2, −2, . . .}, or, if we started
with the interval [−8, 8], we could have built the Cauchy subsequences {0, 0, −2, −2, . . .}
or {−2, 0, 0, −2, −2, . . .}, and so on.

Definition 8.8.7 A subsequential limit of a sequence {sn } is a limit of any subsequence

of {sn }. Thus a subsequential limit can be a complex number, or as in Definition 8.3.4, it
can be ±∞.

Here are a few examples in tabular form:

sequence {sn } the set of subsequential limits of {sn }
convergent {sn } {lim sn }
{(−1)n} {−1, 1}
{(−1)n − 1} {0, −2}
{(−1)n + 1/n} {−1, 1}
{(−1)nn + n + 3} {3, ∞}
{in } {i, −1, −i, 1}
{n} {∞}
{(−1)nn} {−∞, ∞}
260 Chapter 8: Sequences

Theorem 8.8.8 Every unbounded sequence of real numbers has a subsequence that has
limit −∞ or ∞.

Proof. If {sn } is not bounded, choose k1 ∈ N+ such that |sk1 | ≥ 1, and once kn−1 has been
chosen, choose an integer kn > kn−1 such that |skn | ≥ n. Now {skn }n is a subsequence
of {sn }. Either infinitely many among the skn are positive or else infinitely many among
the skn are negative. Choose a subsequence {sln }n of {skn }n such that all terms in {sln }
have the same sign. If they are all positive, then since sln ≥ n for all n, it follows that
limn→∞ sln = ∞, and if they are all negative, then since sln ≤ −n for all n, it follows that
limn→∞ sln = −∞.

Exercises for Section 8.8

8.8.1 Find the subsequential limit sets of the following sequences, and justify:
√ √
i) {(1/2 + i 3/2)n }, {(1/2 − i 3/2)n }.
ii) {1/2n }.
√
iii) {(1/2 + i 3/2)n + 1/2n }.
√
iv) {(1/2 + i 3/2)n · 1/2n}.
√ √
v) {(1/2 + i 3/2)n + (1/2 − i 3/2)n }.
√ √
vi) {(1/2 + i 3/2)n · (1/2 − i 3/2)n}.
8.8.2 Consider the sequence enumerating Q+ as on page 227. Prove that the set of its
subsequential limits equals R≥0 .
n 2 o
n +(−1)n n(n+1)
8.8.3 Prove that the real-valued sequence n+1
is unbounded. Find a subse-
quence that diverges to ∞ and a subsequence that converges to −1.
8.8.4 Let m be a positive integer, and let c be the complex number of absolute value 1
at angle 2π/m counterclockwise from the positive real axis. Find the set of subsequential
limits of {cn }n .
8.8.5 Give examples of sequences with the listed properties, if they exist. If they do not
exist, justify.
i) A sequence with no convergent subsequences.
ii) A sequence whose set of subsequential limits equals {1}.
iii) A sequence whose set of subsequential limits equals {1, 3}.
iv) A sequence whose set of subsequential limits equals {1, 2, 12}.
8.8.6 Suppose that a Cauchy sequence has a convergent subsequence. Prove that the
original sequence is convergent as well.
8.8.7 Prove that a sequence of real numbers contains a monotone subsequence.
 Section 8.9: Liminf, limsup for real-valued sequences 261

8.9 Liminf, limsup for real-valued sequences

Recall that by Axiom 2.9.1 every subset T of R bounded above has a least upper
bound sup T = lub T in R, and that every subset T bounded below has a greatest lower
bound inf T = inf T in R. We extend this definition by declaring
sup T = ∞ if T is not bounded above,
inf T = −∞ if T is not bounded below.
The same definitions apply to sequences thought of as sets: sup{1/n}n = 1, inf{1/n}n = 0,
sup{n}n = ∞, inf{n}n = 1, sup{(−1)n}n = 1, inf{(−1)n }n = −1, et cetera.
Much analysis of sequences has to do with their long-term behavior rather than with
their first three, first hundred, or first million terms — think convergence or the Cauchy
property of sequences. We will see further usage of such tail-end analysis in the next
chapter (for convergence criteria for series). Partly with this goal in mind, we apply infima
and suprema to sequences of tail ends of sequences:

Definition 8.9.1 Let {sn } be a real-valued sequence. The limit superior lim sup and
limit inferior lim inf of {sn } are:

lim sup sn = inf{sup{sn : n ≥ m} : m ≥ 1},

lim inf sn = sup{inf{sn : n ≥ m} : m ≥ 1}.

For any positive integers m1 < m2 ,

inf{sn : n ≥ m2 } ≤ sup{sn : n ≥ m2 } ≤ sup{sn : n ≥ m1 },

inf{sn : n ≥ m1 } ≤ inf{sn : n ≥ m2 } ≤ sup{sn : n ≥ m2 }.
Hence for any positive integers m1 , m2 , inf{sn : n ≥ m2 } ≤ sup{sn : n ≥ m1 }, so that
lim inf sn = sup{inf{sn : n ≥ m} : m ≥ 1} ≤ inf{sup{sn : n ≥ m} : m ≥ 1} = lim sup sn .
If {sn } is bounded by A below and B above, then also A ≤ lim inf sn ≤ lim sup sn ≤ B,
so that in particular by the Least upper bound theorem (Axiom 2.9.1), lim inf sn and
lim sup sn are real numbers.
In other words, lim sup sn is the infimum of the set of all the suprema of all the
tail-end subsequences of {sn }, and analogously, lim inf sn is the supremum of the set of
all the infima of all the tail-end subsequences of {sn }. In the plot below, the sequence
{sn }, drawn with dots, oscillates between positive and negative values but with peaks
and valleys getting smaller and smaller. The connected top decreasing dots denote the
sequence {sup{sm : m ≥ n}}n and the connected bottom increasing dots represent the
sequence {inf{sm : m ≥ n}}n :
262 Chapter 8: Sequences

b
b b
b
b
b
b b b
b b
b b b b b
b b b b b b
b b b b b b b b b
b b b b b b
b b b
b b b b b b b b b b b b
b b b b b b
b b b b b b b b
b b b b b b b
b b b b b b
b b b b
b b
b b
b
b
b

b b

b b

For lim inf sn to be a real number it is not enough for the sequence to be bounded
below: for example, {n} is bounded below but lim inf{n} = sup{inf{n : n ≥ m} : m ≥
1} = sup{m : m ≥ 1} = ∞. Similarly, for lim sup sn to be a real number it is not enough
for the sequence to be bounded above.

Theorem 8.9.2 If {sn } converges to L, then lim inf sn = lim sup sn = L.

Proof. Let ǫ > 0. Then there exists N > 0 such that for all integers n > N , |sn − L| < ǫ.
Thus for all m ≥ N , sup{sn : n ≥ m} ≥ L + ǫ, so that L − ǫ ≤ lim sup sn ≤ L + ǫ. Since
this is true for all ǫ > 0, it follows by Theorem 2.10.4 that L = lim sup sn . The other part
is left to the reader.

Remark 8.9.3 (Ratio test for sequences) With the new language, Theorem 8.6.6 can
be rephrased as follows: If {sn } is a sequence of non-zero complex numbers such that
lim sup{|sn+1 /sn |} < 1, then lim sn = 0. The proof there already accomplishes this. On
the other hand, the ratio test for divergence in Exercise 8.6.6 is not phrased in the most
general form. One generalization is that if lim sup{|sn+1 /sn |} > 1, then {sn } diverges. The
proof is simple. Let r ∈ (1, lim inf{|sn+1 /sn |}). By definition of liminf as supremum of some
infima, this means that there exists an integer m such that inf{|sn+1 /sn | : n ≥ m} > r.
Thus by an easy induction, for all n > m, |sn | > r n−m |sm |, and then by the Comparison
test (Theorem 8.3.7), {|sn |} diverges to infinity, hence {sn } does not converge to a complex
number.
It turns out that there is an important connection between limsup, liminf, and subse-
quential limits:
 Section 8.9: Liminf, limsup for real-valued sequences 263

Theorem 8.9.4 Let {sn } be a bounded sequence of real numbers. Then the supremum
of the set of all subsequential limits equals lim sup sn , and the infimum of the set of all
subsequential limits equals lim inf sn .

Proof. Proof of the limsup part only: Let A = lim sup{sn }, let S be the set of all subse-
quential limits of {sn }, and let U = sup(S). Since the sequence is bounded, A and U are
real numbers.
Let ǫ > 0. Since A = inf{sup{sn : n ≥ m} : m ≥ 1}, there exists m0 ≥ 1 such that
sup{sn : n ≥ m0 } − A < ǫ. Thus for all n ≥ m0 , sn − A < ǫ. But then any subsequential
limit of {sn } is a subsequential limit of {sn }n≥m0 , so that this limit must be at most A + ǫ.
Thus A + ǫ is an upper bound on all subsequential limits of {sn }, so that U ≤ A + ǫ. Since
ǫ is an arbitrary positive number, by Theorem 2.10.4 this means that U ≤ A.
By definition of U , there exists a convergent subsequence {skn } such that U −
lim{skn }n < ǫ/2. Let L = lim{skn }n . So U − L < ǫ/2, and there exists a positive real
number N such that for all integers n > N , |skn − L| < ǫ/2. Thus for all n > N ,

skn > L − ǫ/2 > (U − ǫ/2) − ǫ/2 = U − ǫ.

Thus for any integer m ≥ N , the supremum of {sn : n > m} must be at least U − ǫ, so
that A, the limsup of {sn } must be at least U − ǫ. Hence by Theorem 2.10.4 this means
that A ≤ U .
It follows that A = U .

Theorem 8.9.5 Let {sn }, {tn} be bounded sequences in R. Then lim sup sn +lim sup tn ≥
lim sup(sn + tn ) and lim inf sn + lim inf tn ≤ lim inf(sn + tn ).

Proof. Let a = lim sup sn , b = lim sup tn , c = lim sup(sn + tn ). By boundedness, a, b, c ∈ R.

Let ǫ > 0. Recall that a = inf{sup{sn : n ≥ m} : m ≥ 1}. Then there exists a positive
integer m such that

sup{sn : n ≥ m} − ǫ < a ≤ sup{sn : n ≥ m}.

In particular for all n > m, sn − ǫ < a. By possibly increasing m, we similarly get that
for all n > m in addition tn − ǫ < b. Thus for all n > m, sn + tn − 2ǫ < a + b. Thus
c = inf{sup{sn + tn : n ≥ m} : m ≥ 1} ≤ a + b + 2ǫ. Since ǫ is arbitrary, it follows that
c ≤ a + b, which proves the first part. The rest is left as an exercise.
264 Chapter 8: Sequences

Theorem 8.9.6 Let {sn } and {tn } be sequences of non-negative real numbers such that
lim sn is a positive real number L. Then lim sup(sn tn ) = L lim sup tn and lim inf(sn tn ) =
L lim inf tn .

Proof. Let ǫ > 0. Set ǫ′ = min{L/2, ǫ}. By assumption there exists N > 0 such that for all
integers n > N , |sn − L| < ǫ′ . Then L − ǫ′ < sn < L + ǫ′ . Thus each sn is positive, and in
fact |sn | > L/2. It follows that (L − ǫ′ )tn ≤ sn tn ≤ (L + ǫ′ )tn . But then since L ± ǫ′ ≥ 0,
(L−ǫ′ ) lim sup tn = lim sup(L−ǫ′ )tn ≤ lim sup sn tn ≤ lim sup(L+ǫ′ )tn = (L+ǫ′ ) lim sup tn .

The proof of the liminf part is similar.

Exercises for Section 8.9

8.9.1 Let {sn } and {tn } be sequences of real numbers such that lim sn is a positive real
number L. Prove that lim sup(sn +tn ) = L+lim sup tn and lim inf(sn +tn ) = L+lim inf tn .
8.9.2 Compute lim inf and lim sup for the following sequences. Justify your work.
i) { sinn n }.
ii) {(−1)n n!}.
iii) {2−n }.
iv) {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}.
v) {(4 + n1 )(−1)n }.
vi) The sequence of all positive prime numbers.
vii) The sequence of all multiplicative inverses of positive prime numbers.
8.9.3 Prove the following for any sequence {sn }:
i) inf(−sn ) = − sup(sn ).
ii) lim inf(−sn ) = − lim sup(sn ).
8.9.4 Suppose that {sn } converges to L. Finish the proof of Theorem 8.9.2, namely prove
that lim inf sn = L.
8.9.5 For every positive integer n, let s2n−1 = 1 and s2n = 1/2n .
i) Compute lim inf sn , lim inf sn .
ii) Prove that lim sup sn+1
sn > 1.
iii) Does this contradict the ratio test as in Remark 8.9.3.
8.9.6 Find bounded real-valued sequences {sn }, {tn} such that lim sup sn + lim sup tn >
lim sup(sn + tn ). (Compare with Theorem 8.9.5.)
 Section 8.9: Liminf, limsup for real-valued sequences 265

8.9.7 Let {sn }, {tn } be bounded sequences in R.

i) Finish the proof of Theorem 8.9.5, namely prove that lim inf sn + lim inf tn ≤
lim inf(sn + tn ).
ii) Find such {sn }, {tn} so that lim inf sn + lim inf tn < lim inf(sn + tn ).
8.9.8 Suppose that lim sn = ∞. Prove that the set of subsequential limits of {sn } is
empty.
8.9.9 Finish the proof of Theorem 8.9.4, namely prove that the infimum of the set of all
subsequential limits of a bounded sequence equals the liminf of the sequence.
8.9.10 Let {sn } be a sequence of positive real numbers. Prove that lim sup s1n = 1
lim inf sn .
Chapter 9: Infinite series and power series

In this section we will handle (some) infinite sums, and in particular functions that
arise as infinite sums of higher and higher powers of a variable x. The culmination of
the chapter and the course are the exponential and trigonometric function, with their
properties.
Warning: Finite sums are possible by the field axioms, but infinite sums need not
make any sense at all. For example,

1 + (−1) + 1 + (−1) + 1 + (−1) + 1 + (−1) + 1 + (−1) + 1 + (−1) + 1 + (−1) · · ·

may be taken to be 0 or 1 depending on which consecutive pairs are grouped together in
a sum, or it could even be taken to be 3 by taking the first three positive 1s, and then
matching each successive −1 in the sum with the next not-yet-used +1. In this way each
±1 in the expression is used exactly once, so that the sum can indeed be taken to be 3.
Similarly, we can make the limit be 4, −17, et cetera.
This should convince you in addition that in infinite sums, the order of addition
matters! For more on the order of addition, see ‘exorderseries2’ and Exercise 9.2.16.
Infinite sums require special handling, but limits of sequences prepared the ground for
that.

9.1 Infinite series

Definition 9.1.1 For an infinite sequence {an } of complex numbers, define the corre-
sponding sequence of partial sums
{a1 , a1 + a2 , a1 + a2 + a3 , a1 + a2 + a3 + a4 , . . .}.
Pn
We denote the nth term of this sequence sn = k=1 ak . The (infinite) series corre-
P∞
sponding to the sequence {an } is k=1 ak (whether this “infinite sum” makes sense or
not).
P P
When the range of indices is clear, we write simply k ak or ak .
 Section 9.1: Infinite series 267

Example 9.1.2 For the sequence {1}, the sequence of partial sums is {n}. If a 6= 1, by
Pn n+1
Example 1.5.4 the sequence of partial sums of {an } is { k=1 ak }n = { a a−1−a }n . In par-
n
ticular, the sequence of partial sums of {(−1)n} is {{ (−1)2 −1 }n = {−1, 0, −1, 0, −1, 0, . . .}.
We have encountered shifted sequences, such as {an }n≥m , and similarly there are
P∞
shifted series: k=m ak stands for the limit of the sequence of partials sums, but in this
case, the nth partial sum is sn = am + am+1 + am+2 + · · · + am+n−1 .
P∞
Definition 9.1.3 (Most of the time and by default we take m = 1.) The series k=m ak
Pn−m+1
converges to L ∈ C if the sequence { k=m ak }n converges to L. We say then that L
P∞
is the sum of the series and we write k=m ak = L.
If the series does not converge, it diverges.

Just like for sequences, when a series diverges, it may diverge to ∞ or to −∞, or it
may simply have no limit.
Since a sequence {sn } converges if and only if {sn + c} converges (where c is any
P∞ P∞
constant), it follows that k=1 ak converges if and only if k=m ak converges, and then
∞
X ∞
X
ak = a1 + a2 + · · · + am−1 + ak .
k=1 k=m

The following follows immediately from the corresponding results for sequences:
P∞ P∞
Theorem 9.1.4 Let A = k=1 ak , B = k=1 bk , and c ∈ C.
(1) If A, B ∈ C, then
X∞
(ak + cbk ) = A + cB.
k=1

(2) If all ak , bk are real numbers and A, B ∈ R ∪ {∞, −∞}, then


 ∞, if A = ∞, B ∈ R ∪ {∞} and c ≥ 0;



 ∞, if A = ∞, B ∈ R ∪ {−∞} and c ≤ 0;



 −∞, if A = −∞, B ∈ R ∪ {−∞} and c ≥ 0;


X ∞  −∞, if A = −∞, B ∈ R ∪ {∞} and c ≤ 0;
(ak + cbk ) = ∞, if A ∈ R, B = ∞ and c > 0;


k=1 
 −∞, if A ∈ R, B = ∞ and c < 0;



 ∞, if A ∈ R, B = −∞ and c < 0;


 −∞,
 if A ∈ R, B = −∞ and c > 0;
A, if c = 0.
Pn Pn
Proof. (1) By assumption, the sequences { k=1 ak } and { k=1 bk } converge to A and B
in C, respectively. By the theorem on the convergence of sums of sequences (Theorem 8.4.3)
Pn Pn Pn
then k=1 (ak + cbk ) = k=1 ak + c k=1 bk converges to A + cB.
Other parts are proved similarly.
268 Chapter 9: Infinite series and power series

Remark 9.1.5 This theorem justifies the binary operation of addition on the set of of
convergent infinite series:
∞
! ∞
! ∞
X X X
ak + bk = (ak + bk ).
k=1 k=1 k=1

It is hard to immediately present examples of this because we know so few limits of

infinite series. There are examples in the exercises.
P∞
Theorem 9.1.6 If r ∈ C satisfies |r| < 1, then the geometric series k=1 r k−1 converges
1
P∞ r
to 1−r , so k=1 r k = 1−r .
Pn n
Proof. By Example 1.5.4, k=1 r k−1 = 1−r n
1−r . By Theorem 8.6.5, lim r = 0. Thus by the
scalar and sum rules for limits of sequences (Theorem 8.5.6 or Theorem 8.4.3),
Xn  
k−1 1 − rn 1 rn 1
lim r = lim = lim − = .
n→∞ n→∞ 1 − r n→∞ 1 − r 1−r 1−r
k=1
Pn Pn r
Thus limn→∞ k=1 r k = r limn→∞ k=1 r k−1 = 1−r .
In particular, the familiar decimal expansion 0.33333 · · · of 1/3 can be thought of as
P∞
the infinite sum k=1 103k . The sequence of its partial sums is {0.3, 0.33, 0.333, 0.3333, . . .},
and by the theorem above,
X∞ ∞  k−1
3 3 X 1 3 1 3 10 1
k
= = · 1 = · = .
10 10 10 10 1 − 10 10 9 3
k=1 k=1

A little more work is expressing 5.523 = 5.523523523523 · · · (repeating 523) as a fraction:

X∞  
523 523 523 1
5.523 = 5 + + + + · · · = 5 + 523
1000 10002 10003 1000k
k=1
1
1000 1 523 4995 + 523
= 5 + 523 1 = 5 + 523 =5+ =
1 − 1000 1000 − 1 999 999
5518
= .
999
P∞ 1
Example 9.1.7 The harmonic series k=1 k diverges to ∞ by Example 8.7.6.
P∞ 1
Example 9.1.8 k=1 k·(k+1) = 1.
Pn 1 1
Proof. By Exercise 8.2.9, k=1 k(k+1) =1− n+1 . Thus
∞
X n
X  
1 1 1
= lim = lim 1 − = 1.
k(k + 1) n→∞ k(k + 1) n→∞ n+1
k=1 k=1
 Section 9.1: Infinite series 269
P∞ 1
Example 9.1.9 The series k=1 k2 converges.

Proof #1: By Exercise 1.5.20 and Theorem 9.1.6,

n
2X −1 n+1
X 1 ∞
X
1 1 1
0≤ 2
≤ k−1
≤ k−1
= = 2,
k 2 2 1 − 1/2
k=1 k=1 k=1
P2n −1 Pn
so that { k=1 k12 }n and { k=1 k12 }n are bounded increasing sequences of real numbers.
P∞
By Theorem 8.6.4, the sequence has a limit that is at most 2, so that k=1 k12 converges.
Pn
Proof #2: Since k 2 > k(k − 1), we have that for k > 2, k12 < k(k−1) 1
. Thus k=1 k12 =
Pn Pn P∞ P∞
1 + k=2 k12 < 1 + k=2 k(k−1) 1 1
. By Example 9.1.8, the series k=2 k(k−1) 1
= k=1 k(k+1)
Pn
converges, so that the increasing sequence { k=1 k12 } of partial sums is bounded above, so
P∞
that by Theorem 8.6.4, k=1 k12 converges.
P∞ 2
It turns out that k=1 k12 = π6 , but this is much harder to prove. (An indication of
the proof is in Exercise 9.12.5.)
P∞
Theorem 9.1.10 If k=1 ak converges, then lim an = 0, and the sequence {an } is
bounded.
P∞
Proof. By assumption, k=1 ak converges, so that by definition, the sequence {sn } of
partial sums converges, and is thus Cauchy. In particular, for every ǫ > 0 there exists
N > 0 such that for all integers n > N + 1, |an | = |sn − sn−1 | < ǫ. Thus lim an = 0, and
by Theorem 8.6.2, {an } is bounded.
The converse of this theorem is of course false; see Example 9.1.7.

Theorem 9.1.11 Let {an } be a sequence of non-negative real numbers. If the sequence
P
{a1 + a2 + · · · + an } of partial sums is bounded above, then an converges.

Proof. The sequence {a1 + a2 + · · · + an } of partial sums is monotone and bounded above,
so it converges by Theorem 8.6.4.

Theorem 9.1.12 Let {an } be a sequence of complex numbers, and let m be a positive
P∞ P∞
integer. Then k=1 an converges if and only if k=m an converges. Furthermore in this
P∞ P∞
case, k=1 an = (a1 + a2 + · · · + am−1 ) + k=m an .

Proof. Let sn = a1 + a2 + · · · + an , and tn = am + am+1 + · · · + an . By the constant

and sum rules for sequences (Theorem 8.5.6 or Theorem 8.4.3), the sequence {sn }n =
{a1 + a2 + · · · + am−1 }n + {tn }n converges if the sequence {tn } converges, and similarly,
{tn }n = {sn }n − {a1 + a2 + · · · + am−1 }n converges if {sn } converges.
270 Chapter 9: Infinite series and power series

Exercises for Section 9.1

P∞
9.1.1 Let r ∈ C satisfy |r| ≥ 1. Prove that r k diverges.
k=1
P∞ P∞ P∞
9.1.2 Compute and justify the following sums: k=1 21k , k=6 31k , k=8 52k .
P∞ 2k+1
9.1.3 Prove that k=1 k2 ·(k+1)2 converges, and find the sum. (Hint: Do some initial
experimentation with partial sums, find a pattern for partial sums, and prove the pattern
with mathematical induction.)
9.1.4 Let an = (−1)n . Prove that the sequence of partial sums {a1 + a2 + · · · + an } is
bounded but does not converge. How does this not contradict Theorem 9.1.11?
9.1.5 For each k ∈ N+ let xk be an integer between 0 and 9.
P ∞ xk
i) Prove that k=1 10 k converges.

ii) What does this say about decimals?

iii) Find the sum if xk = 4 for all k. Express the sum with its decimal expansion and
also as a ratio of two positive integers.
iv) Find the sum if {xn } = {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}. Express the sum with its
decimal expansion and also as a ratio of two positive integers.
P∞ xk
(v)* Prove that whenever the sequence {xn } is eventually periodic, then k=1 10 k is

a rational number.
P∞ P∞ P∞
9.1.6 Prove that k=1 ak converges if and only if k=1 Re ak and k=1 Im ak converge.
P∞ P∞ P∞
Furthermore, k=1 ak = k=1 Re ak + i k=1 Im ak .
P∞
9.1.7 Suppose that lim an 6= 0. Prove that k=1 ak diverges.
9.1.8 Determine with proof which series converge.
P∞
i) k=1 k1k .
P∞

ii) k=1 k13 + ik .
P
iii) ∞ 1
k=1 k! .
9.1.9 Let {an } and {bn } be complex sequences, and let m ∈ N+ such that for all n ≥ 1,
P P∞
an = bn+m . Prove that ∞ k=1 ak converges if and only if k=1 bk converges.

9.2 Convergence and divergence theorems for series

P
Theorem 9.2.1 (Cauchy’s criterion for series) The infinite series ak converges if
and only if for all real numbers ǫ > 0 there exists a real number N > 0 such that for all
integers n ≥ m > N ,
|am+1 + am+2 + · · · + an | < ǫ.
P
Proof. Suppose that ak converges. This means that the sequence {sn } of partial sums
converges, and by Theorem 8.7.3 this means that {sn } is a Cauchy sequence. Thus for all
 Section 9.2: Convergence and divergence theorems for series 271

ǫ > 0 there exists N > 0 such that for all integers m, n > N , |sn − sm | < ǫ. In particular
for n ≥ m > N , |am+1 + am+2 + · · · + an | = |sn − sm | < ǫ.
Now suppose that for all real numbers ǫ > 0 there exists a real number N > 0 such
that for all integers n ≥ m > N , |am+1 + am+2 + · · · + an | < ǫ. This means that the
sequence {sn } of partial sums is Cauchy. By Theorem 8.7.5, {sn } is convergent. Then by
P
the definition of series, k ak converges.

Theorem 9.2.2 (Comparison test (for series)) Let {an } be a real and {bn } a complex
P P
sequence. If ak converges and if for all n, an ≥ |bn |, then bk converges.

Proof. Note that all an are non-negative, and for all integers n ≥ m,
|bm+1 + bm+2 + · · · + bn | ≤ |bm+1 | + |bm+2 | + · · · + |bn |
≤ am+1 + am+2 + · · · + an
= |am+1 + am+2 + · · · + an |.
P
Since ak converges, by Theorem 9.2.1, for every ǫ > 0 there exists N > 0 such that for all
integers n > m > N , |am+1 + am+2 + · · · + an | < ǫ, and hence |bm+1 + bm+2 + · · · + bn | < ǫ.
P
Thus again by Theorem 9.2.1, bk converges.

Theorem 9.2.3 (Comparison test (for series)) Let {an } be a complex and {bn } a real
P P
sequence. If ak diverges and if for all n, bn ≥ |an |, then bk diverges.

Proof. For all integers n ≥ m, |bm+1 + bm+2 + · · · + bn | ≥ |am+1 | + |am+2 | + · · · + |an | ≥

P
|am+1 +am+2 +· · ·+an |. Since k ak is divergent, by Theorem 9.2.1, there exists ǫ > 0 such
that for all N > 0 there exist integers n > m > N , such that |am+1 + am+2 + · · · + an | ≥ ǫ,
P
and hence |bm+1 + bm+2 + · · · + bn | ≥ ǫ. Thus again by Theorem 9.2.1, bk is divergent.

Theorem 9.2.4 (Ratio test) (Compare to Remark 8.9.3.) Let {an } be a sequence of
non-zero complex numbers. 
 an+1  P P
(1) If lim sup  an  < 1, then |ak | and ak converge.
  P P
 
(2) If lim inf  aan+1
n
 > 1, then |ak | and ak diverge.

Proof. Let L = lim sup | aan+1

n
|. Suppose that L < 1. Let r be a real number in the
open interval (L, 1). Since L = inf{sup{ |a|an+1
n|
|
: n ≥ m} : m ≥ 1} and r > L, it fol-
lows that there exists m ≥ 1 such that r > sup{ |a|an+1
n|
|
: n ≥ m}. Thus for all n ≥ m,
P
n
|an+1 | < r|an |. Thus by Exercise 1.5.22, |am+n | < r |am |. The geometric series k r k
P
converges by Theorem 9.1.6, and by Theorem 9.1.4, k am r k converges. Thus by Theo-
P∞ P∞ P
rem 9.2.2, k=1 |am+k | and k=1 am+k converge. Hence by Theorem 9.1.12, |ak | and
P
ak converge. This proves (1).
272 Chapter 9: Infinite series and power series

Now let L = lim inf | aan+1

n
|, and suppose that L > 1. Let r be a real number in the
open interval (1, L). Since L = sup{inf{ |a|an+1
n|
|
: n ≥ m} : m ≥ 1} and r < L, it follows that
there exists m ≥ 1 such that r < inf{ |a|an+1
n|
|
: n ≥ m}. Thus for all n ≥ m, |an+1 | > r|an |.
Thus by a straightforward modification of Exercise 1.5.22, |am+n | > r n |am |. The geometric
P P
series k r k diverges by Exercise 9.1.1, and by Theorem 9.1.4, k am r k diverges. Thus by
P∞ P∞ P
Theorem 9.2.2, k=1 |am+k | and k=1 am+k diverge. Hence by Theorem 9.1.12, |ak |
P
and ak diverge. This proves (2).
This ratio test for convergence of series does not apply when lim sup | aan+1 | = 1 or
an+1 Pn
lim inf | an | = 1. The reason is that under these assumptions the series k |ak | and
P
k ak sometimes converge and sometimes diverge. For example, if an = 1/n for all n,
Pn
lim sup | aan+1 | = lim inf | aan+1 | = 1, and k=1 k1 diverges; whereas if an = 1/n2 for all n,
n n Pn
then lim sup | aan+1
n
| = lim inf | aan+1
n
| = 1, and k=1 k12 converges.

Theorem 9.2.5 (Root test for series) Let {an } be a sequence of complex numbers.
Let L = lim sup |an |1/n .
P P
(1) If L < 1, then k |ak |, k ak converge.
P P
(2) If L > 1, then k |ak |, k ak diverge.

Proof. If L < 1, choose r ∈ (L, 1). Since L = inf{sup{|an |1/n : n ≥ m} : m ≥ 1} and r > L,
there exists m ≥ 1 such that r > sup{|an |1/n : n ≥ m}. Thus for all n ≥ m, r n ≥ |an |.
P k
Thus by the Comparison test (Theorem 9.2.2), since the geometric series r converges,
P P
we have that ak and |ak | converge. The proof of (2) is similar, and is omitted here.

Theorem 9.2.6 (Alternating series test) If {an } is a sequence of positive real numbers
P∞
such that lim an = 0 and a1 ≥ a2 ≥ a3 ≥ · · ·, then k=1 (−1)k ak converges.

Proof. Let m, n be positive integers. Then

0 ≤ (an − an+1 ) + (an+2 − an+3 ) + · · · + (an+2m − an+2m+1 )

= an − an+1 + an+2 − an+3 + · · · + an+2m − an+2m+1
= an − (an+1 − an+2 ) − (an+3 ) − an+4 ) − · · · − (an+2m−1 − an+2m ) − an+2m+1
≤ an ,
and similarly
0 ≤ an − an+1 + an+2 − an+3 + · · · + an+2m ≤ an .
P k
Thus by Cauchy’s criterion Theorem 9.2.1, k (−1) ak converges.
 Section 9.2: Convergence and divergence theorems for series 273
P
Example 9.2.7 Recall from Example 9.1.7 that the harmonic series k 1/k diverges.
P P
k
But the alternating series k (−1) /k converges by this theorem. (In fact, k (−1)k /k
converges to − ln 2, but proving the limit is harder – see the proof after Example 9.7.4.
We examine this infinite series more carefully:
1 1 1 1 1 1 1 1 1
−1 + − + − + − + − + −···
2 3 4 5 6 7 8 9 10


We cannot rearrange the terms in this series as 21 + 41 + 16 + 81 + 10 1
+ · · · minus


1 + 31 + 51 + 17 + 19 + · · · because both of these series diverge to infinity. More on
changing the order of summation is in Exercise 9.2.16 and in Theorem 9.2.9.
P∞ P∞
Definition 9.2.8 The series k=1 ak is called absolutely convergent if k=1 |ak | con-
verges.
P∞
Theorem 9.2.9 Let k=1 ak be absolutely convergent. Let r : N+ → N+ be a bijective
P∞
function. Then k=1 ar(k) converges.
P∞
Proof. By comparison test (Theorem 9.2.2), k=1 ak converges to some number L ∈ C.
P∞
We will prove that k=1 ar(k) converges to L.
P∞
Let ǫ > 0. Since k=1 ak = L, there exists N0 > 0 such that for all integers
Pn P∞
n > N0 , | k=1 ak − L| < ǫ/2. Also, since k=1 |ak | converges, by Cauchy’s criterion for
Pn
sequences, there exists N1 > 0 such that for all integers n ≥ m > N1 , k=m+1 |ak | <
ǫ/2. Pick an integer N > max{N0 , N1 }. Since r is invertible, we can define M =
max{r −1 (1), r −1(2), . . . , r −1 (N )}. Let n be an integer strictly bigger than M . Then
by definition the set {r(1), r(2), . . . , r(n)} contains 1, 2, . . . , N . Let K = {r(k) : k ≤
n} \ {1, 2, . . . , N }. Then
     
X n  X N X  X N  X
     
 ar(k) − L =  ak + ak − L ≤  ak − L + |ak | < ǫ
     
k=1 k=1 k∈K k=1 k∈K

since all the finitely many indices in K are strictly bigger than N0 .

Theorem 9.2.10 (Integral test for series convergence) Let f : [1, ∞) → [0, ∞) be a
Rn P∞
decreasing function. Suppose that for all n ∈ N+ , 1 f exists. Then k=1 f (k) converges
Rn
if and only if limn→∞ 1 f exists and is a real number.
P∞ Rn
(It is not necessarily the case that k=1 f (k) equals limn→∞ 1 f .)

Proof. Since f is decreasing, for all x ∈ [n, n + 1], f (n) ≥ f (x) ≥ f (n + 1). Thus
Z n+1 Z n+1 Z n+1
f (n + 1) = f (n + 1)dx ≤ f (x)dx ≤ f (n)dx = f (n).
n n n
P
Suppose that k f (k) converges. Then by the definition limn→∞ (f (1) + f (2) + · · · +
R n+1 R2 R3 R n+1
f (n)) exists. By the displayed inequalities, 1 f = 1 f + 2 f + · · · + n f ≤ f (1) +
274 Chapter 9: Infinite series and power series
R n+1
f (2) + · · · + f (n), so that { 1 f }n is a bounded increasing sequence of real numbers, so
R n+1 Rn
that limn→∞ 1 f exists, and hence that limn→∞ 1 f exists.
Rn
Conversely, suppose that limn→∞ 1 f exists. Let L ∈ R be this limit. By the dis-
R3 R n+1 R n+1
played inequalities, f (2) + f (3) + · · · + f (n + 1) ≤ 2 f + · · · + n f = 2 f . Since f
takes on only non-negative values, this says that f (2) + f (3) + · · · + f (n + 1) ≤ L. Thus
{f (2) + · · · + f (n + 1)}n is a non-decreasing sequence that is bounded above by L. Thus
by Theorem 8.6.4, this sequence converges. By adding the constant f (1), the sequence
P
{f (1) + · · · + f (n + 1)}n converges, so that by the definition of series, k f (k) converges.

Theorem 9.2.11 (The p-series convergence test) Let p be a real number. The series
P p
k k converges if p < −1 and diverges if p ≥ −1.

Proof. If p = −1, then the series is the harmonic series and hence diverges. If p ≥ −1,
then np ≥ n−1 for all n by Theorem 7.5.5. Thus by the comparison test (Theorem 9.2.3),
P p
k k diverges.
Now suppose that p < −1. The function f : [1, ∞) → R given by f (x) = xp is
differentiable, continuous, and decreasing. Since f is continuous, for all positive integers n,
Rn Rn Rn p p+1

1
f exists. By the Fundamental theorem of calculus, 1
f = 1
x dx = n p+1−1 . By the
composite rule for sequences (either Theorem 8.5.8 or Theorem 8.4.7), since the function
that exponentiates by the positive −(p +1) is continuous at all real numbers and lim n1 = 0,

−(p+1) Rn
it follows that lim np+1 = lim n1 = 0−(p+1) = 0, so that lim 1 f exists and equals
−1
P p
p+1 . Thus by the Integral test (Theorem 9.2.10), k k converges.

Exercises for Section 9.2

P P∞ (−1)k P∞ P∞ (−1)k
9.2.1 Prove that ∞ 1
k=0 (2k+1)2 , k=0 (2k+1)2
, 1
k=1 k2 and k=1 k2
all converge.
9.2.2 Prove that the following statements are equivalent. (They are also all true, but we
do not yet have enough methods to prove them.)
P∞ 1 π2
(1) k=0 (2k+1)2 = 8 .
P∞ 1 π2
(2) k=1 k2 = 6 .
P∞ (−1)k π2
(3) k=0 (2k+1)2 = 16 .
P∞ (−1)k 2
(4) k=1 k2 = π12 .
P∞ (−1)k P∞
(Hint for a part: Write out the first few summands of k=1 k2 + 12 k=1 k12
P∞ k P∞
= k=1 (−1)k2 + 2
1
k=1 (2k)2 .)
 Section 9.2: Convergence and divergence theorems for series 275

9.2.3 For each of the following series, determine with proof whether they converge or
diverge. You may need to use Examples 8.2.9 and 8.3.6.
X∞
3i
i) √ .
k 2 + k4
k=1
X∞
1
ii) √ .
k=1
k
X∞
1
iii) .
k3
k=1
X∞
2k
iv) .
k!
k=1
X∞
2k
v) .
k3
k=1
X∞
1
vi) √ .
k=1
k 2
P P
9.2.4 Find a convergent series k ak and a divergent series k bk with lim sup |an |1/n =
lim sup |bn |1/n = 1.
9.2.5 Make a list of all encountered criteria of convergence for series.
9.2.6 The goal of this exercise is to show that if the ratio test (Theorem 9.2.4) determines
the convergence/divergence of a series, then the root test (Theorem 9.2.5) determines it as
well. Let {an } be a sequence of non-zero complex numbers.
i) Suppose that lim sup | aan+1
n
| < 1. Prove that lim sup |an |1/n < 1.
ii) Suppose that lim inf | aan+1
n
| > 1. Prove that lim sup |an |1/n > 1.
P 5k
9.2.7 Apply the ratio test (Theorem 9.2.4) and the root test (Theorem 9.2.5) to ∞ k=1 k! .
P∞
Was one test easier? Repeat for k=1 k1k .
P∞
9.2.8 Let {an } be a complex sequence, and let c ∈ C. Is it true that k=1 ak converges
P∞
if and only if k=1 cak converges? If true, prove; if false, give a counterexample.
P∞
9.2.9 Let an = (−1)n /n, bn = 2(−1)n . Prove that k=1 ak converges, that for all n,
P
|bn | = 2, and that ∞ ak
k=1 bk diverges.
P∞
9.2.10 (Compare with Exercise 9.2.9.) Suppose that k=1 |ak | converges. Let {bn } be a
P∞
sequence of complex numbers such that for all n, |bn | > 1. Prove that k=1 abkk converges.
276 Chapter 9: Infinite series and power series

9.2.11 (Summation by parts) Let {an }, {bn} be complex sequences. Prove that
n
X n
X n
X k
X
ak bk = an bk − (ak+1 − ak ) bj .
k=1 k=1 k=1 j=1

(Hint: Set a0 = b0 = 0. Let f, g : [1, ∞) → C be defined as follows: for each n ∈ N+ , on the

interval [n, n+1), f is the the constant function an −an−1 and g is the constant function bn .
R n+1 R n+1 Pn
Both f and g are piecewise continuous, F (n) = 1 f = an , G(n) = 1 g = k=1 bk .
R R
The problem should remind you of integration by parts (F g) = F G − (f G).)
P ∞ xk
9.2.12 Let x ∈ C with |x| = 1 and x 6= 1. Prove that k=1 k converges. (Hint:
Exercise 9.2.11, Example 1.5.4.)
P∞ P∞
9.2.13 Prove that if k=1 ak converges then k=1 akk converges. (Hint: Exercise 9.2.11.)
9.2.14 (The Dirichlet test) Let {an } be a decreasing real sequnce such that lim an = 0,
and let {bn } be a sequence of complex numbers whose sequence of partial sums is bounded.
P∞
i) Prove that the series k=1 ak bk converges. (Hint: Exercise 9.2.11.)
ii) Prove the alternating series test using part i).
9.2.15 The following guides through another proof of the p-series convergence test for
p < −1. (Confer Theorem 9.2.11 for the first proof).
i) Prove that for each positive integer n there exists c ∈ (n, n+1) such that (p+1)cp =
(n + 1)p+1 − np+1 . (Hint: Mean value theorem (Theorem 6.3.4).)
ii) Prove that for all positive integers n,
(n + 1)p+1 − np+1 np+1 − (n − 1)p+1
(n + 1)p < and np < .
p+1 p+1
iii) Prove by induction on n ≥ 1 that
n
X 1 p
kp ≤ np+1 + .
p+1 p+1
k=1
P
iv) Prove that the positive sequence { nk=1 k p }n of partial sums is an increasing
sequence bounded above.
Pn P∞
v) Prove that the sequence { k=1 k p }n converges, and so that k=1 k p converges.
9.2.16 (Order of summation in infinite sums is important.)
P P∞
i) Prove that ∞ 1
k=1 2k and
1
k=1 2k+1 diverge. Refer to Example 9.2.7.
Pn Pn P2n+1 k
ii) Observe that k=1 2k 1 1
− k=1 2k+1 = k=1 (−1) k .
P∞ (−1)k P∞ 1 P∞ 1
iii) Argue that k=1 k 6= k=1 2k − k=1 2k+1 . Why does this not contradict
the “expected” summation and difference rules?
 Section 9.3: Power series 277

9.2.17 (Raabe’s test) Let a1 , a2 , . . . be positive real numbers such that for some α > 1
P
and for some N ∈ N, aan+1 n
≤ 1− α n for all n ≥ N . Prove that n an converges. (Hint:
Let f (x) = 1 − x . Use the Mean value theorem to get c ∈ (x, 1) such that f ′ (c)(1 − x) =
α
P −α
f (1) − f (x). Conclude that 1 − xα ≤ (1 − x)α. Apply this to x = 1 − n1 . Use that n
converges.)

9.3 Power series

In this section we deal with sums where the index varies through N0 , and furthermore,
the terms of the sequence are special functions rather than constants:

Definition 9.3.1 A power series is an infinite series of the form

∞
X X
ak xk = ak xk = a0 + a1 x + a2 x2 + a3 x3 + · · · ,
k=0 k≥0

where a0 , a1 , a2 , . . . are fixed complex numbers, and x is a variable that can be replaced by
any complex number. (By convention as on page 35, 00 = 1.)

Think through the following table of reading off the coefficients an of xn :

power series an
X∞
xk an = 1
k=0
X∞ 
2k+1 3 5 7 1, if n is odd;
x = x+x +x + x +··· an =
0, if n is even.
k=0
(
X∞ 0, if n is odd;
x2k+12 = x12 + x14 + x16 + x18 + · · · an = 0, if n is even and n < 12;
k=0 1, if n ≥ 12 is even.
X∞
kxk an = n
k=0
X∞
xk 1
an = n!
k!
k=0
X∞
(kx)k an = nn
k=0

The partial sums of power series are polynomials, so a power series is a limit of poly-
nomials.
278 Chapter 9: Infinite series and power series

A power series is a function of x, and the domain is to be determined. Clearly, 0 is in

P∞
the domain of every power series: plugging in x = 0 returns k=0 ak 0k = a0 . If all except
finitely many an are 0, then the power series is actually a polynomial, and is thus defined
P
on all of C. When 1 is in the domain, evaluation of the power series k ak xk at x = 1 is
P
the (ordinary) series k ak .
The most important question that we address in this section is: which x are in the
domain of the power series, i.e., for which x does such an infinite series converge. We prove
that for every power series whose domain is not all of C there exists a non-negative real
number R such that the series converges for all x ∈ C with |x| < R and the series diverges
for all x ∈ C with |x| > R. What happens at x with |x| = R depends on the series.

P∞
Example 9.3.2 Let f (x) = k=0 xk . By Theorem 9.1.6, the domain of f contains all
complex numbers with absolute value strictly smaller than 1, and by Theorem 9.1.10, the
domain of f contains no other numbers, so that the domain equals {x ∈ C : |x| < 1}.
P∞ 1
Moreover, for all x in the domain of f , by Theorem 9.1.6, k=0 xk = 1−x . Note that the
1
domain of 1−x is strictly larger than the domain of f .
Whereas for general power series it is impossible to get a true numerical infinite sum,
P P
for geometric series this is easy: f ( 21 ) = k≥0 21k = 1−1 1 = 2, f ( 31 ) = k≥0 31k = 1−1 1 = 32 ,
P 1
2 3
f (0.6) = k≥0 0.6k = 1−0.6 = 2.5.

P
Theorem 9.3.3 (Root test for the convergence of power series) Let ak xk be a
power series, and let α = lim sup |an |1/n. Define R by

 1/α, if 0 < α < ∞;
R = 0, if α = ∞;

∞, if α = 0.
P P
Then for all x ∈ C with |x| < R, |ak ||x|k and ak xk converge in C, and for all x ∈ C
P P
with |x| > R, |ak ||x|k and ak xk diverge.

Proof. By the definition of limits, α is either a non-negative real number or ∞. We apply

the Root test for series (Theorem 9.2.5): lim sup |an xn |1/n = |x| lim sup |an |1/n = |x|α. If
|x|α < 1, then both of the series converge, and if |x|α > 1, then the two series diverge. If
α = 0, then |x|α < 1 is true for all x ∈ C, so R = ∞ has the stated property. If α = ∞,
then |x|α < 1 is true only for x = 0, so R = 0 has the stated property. If 0 < α < ∞, then
|x|α < 1 is true only for all x ∈ C with |x| < 1/α = R.
 Section 9.3: Power series 279

Definition 9.3.4 The R from Theorem 9.3.3 is called the radius of convergence of the
P
series ak xk .

This is really a radius of convergence because inside the circle B(0, R) the series con-
verges and outside of the circle the series diverges. Whether the power series converges at
points on the circle depends on the series; see Example 9.3.6.
P P P P P
Theorem 9.3.5 The series ak xk , |ak |xk , |ak ||x|k , kak |x|k , kak xk−1 ,
P 2 P
k ak xk , k(k − 1)ak xk , all have the same radius of convergence.
p √ √
Proof. By Example 8.2.9, lim n |n| = 1. For any integer n ≥ 2, 1 ≤ n n − 1 ≤ n n, so that
√ p
by the squeeze theorem, lim n n − 1 = 1. Thus by Theorem 8.9.6, lim sup n |n(n − 1)an | =
p p p
lim n n(n − 1)·lim sup n |an | = lim sup n |an |. This proves that the α as in Theorem 9.3.3
P P
for ak xk is the same as the α for k(k − 1)ak xk , which proves that these two power
series have the same radius of convergence. The proofs of the other parts are similar.
P
Example 9.3.6 We have seen that xk has radius of convergence 1. By Theorem 9.1.10,
this series does not converge at any point on the unit circle. By the previous theorem,
P∞ P∞
the radius of convergence of k=1 k1 xk and of k=1 k12 xk is also 1. By the p-series test
P∞
(Theorem 9.2.11) or by the harmonic series fact, k=1 k1 xk diverges at x = 1, and by the
P∞
alternating series test (Theorem 9.2.6), k=1 k1 xk converges at x = −1. By the p-series
P∞
test (Theorem 9.2.11), k=1 k12 converges, so that by the comparison test Theorem 9.2.2,
P∞ 1 k
k=1 k2 x converges on the unit circle.
P
Theorem 9.3.7 Let the radius of convergence for ak xk be R. Then the radius of
P √
convergence for ak x2k is ∞ if R = ∞ and it is R otherwise.
p
Proof. Let α = lim sup n |an |. In the second power series, the nth coefficient is 0 if n is
odd, and it is an/2 xn if n is even. Then applying the root test to this power series gives
n√ p √ p √ p √ p o
1 2 3 4 5 6 7 8
lim sup 0, |a1 |, 0, |a2 |, 0, |a3 |, 0, |a4 |, . . .
np p p p o
= lim sup 2 |a1 |, 4 |a2 |, 6 |a3 |, 8 |a4 |, . . .
(0s do not contribute to limsup of non-negative numbers)
q qp qp qp 
p
1 2 3 4
= lim sup |a1 |, |a2 |, |a3 |, |a4 |, . . .
r np o
1
p
2
p3
p4
= lim sup |a1 |, |a2 |, |a3 |, |a4 |, . . .
√
= α,

and the conclusion follows.

280 Chapter 9: Infinite series and power series
P P
Example 9.3.8 Similarly to the last example and by Theorem 9.3.5, xk , kxk ,
P 2 k P P 2k+1
k x , k(k − 1)xk , x all have radius of convergence 1.

Theorem 9.3.9 (Ratio test for the convergence of power series) Suppose that all
an are non-zero complex numbers.
  P P
n 
(1) If |x| < lim inf  aan+1 , then |ak ||x|k and ak xk converge.
 a  P P
(2) If |x| > lim sup an+1 n 
, then |ak ||x|k and ak xk diverge.
 a  P
n 
Thus if lim an+1 exists, it equals the radius of convergence of ak xk .

Warning: Compare with the Ratio test for convergence of series (Theorem 9.2.4) where
fractions are different. Explain to yourself why that is necessarily so, possibly after going
through the proof below.

Proof. The two series converge in case x = 0, so that we may assume that x 6= 0. We may
then apply the Ratio test for convergence of series (Theorem 9.2.4):
   
 an+1 xn+1   an+1 
lim sup   = |x| lim sup  
an xn   an 
  
 an+1 
= |x| inf{sup    : n ≥ m : m ≥ 1}
an 
( )
1
= |x| inf   :m≥1
n 
inf{ aan+1 : n ≥ m}
|x|
= n  a  o
n 
sup inf{ an+1 : n ≥ m} : m ≥ 1
|x|
=  .
n 
lim inf  aan+1
If this is strictly smaller than 1, then the two series converge, which proves (1). Similarly,
 
 an+1 xn+1  |x|

lim inf  =  ,
an x n  lim sup aan 
n+1

and if this is strictly larger than 1, then the two series diverge. The last part is then
immediate by the definition of radius of convergence.

Examples 9.3.10
P 2k+1
(1) In Example 9.3.8 we established via the root test that x has radius of
convergence 1. The ratio test is inapplicable for this power series. However, note
P 2k+1 P
that x = x (x2 )k , and by the ratio test for series (not power series), this
series converges for non-zero x if lim sup |(x2 )k+1 /(x2 )k | < 1, i.e., if |x2 | < 1, i.e.,
if |x| < 1, and it diverges if |x| > 1.
 Section 9.3: Power series 281
P∞

x k
(2) The radius of convergence of k=1 k is ∞. For this we apply the root test:
 1/n
α = lim sup  1n 
n = lim sup 1
n = 0.
P∞ xk
 1 
(3) By the ratio test, the radius of convergence of is lim n!
1
 = lim (n+1)! =
k=1 k! n!
(n+1)!

lim(n + 1) = ∞. The root test gives α = lim sup |1/n!|1/n| = lim sup(1/n!)1/n, and
by Example 8.3.6 this is 0. Thus the radius of convergence is ∞ also by the root
test.

Exercises for Section 9.3

P∞
9.3.1 Let the radius of convergence of k=0 ak xk be R. Let c ∈ C.
P∞ P∞
i) Prove that c· k=0 ak xk = k=0 (c·ak )xk is convergent with radius of convergence
equal to R if c 6= 0, and with radius of convergence ∞ otherwise.
P
ii) Prove that ∞ k
k=0 ak (cx) is convergent with radius of convergence equal to ∞ if
c = 0 and radius R/|c| otherwise.
P
9.3.2 Suppose that k ak converges for some ak ∈ C.
P
i) Prove that the function defined as the power series k ak xk has radius of conver-
gence at least 1.
P∞
ii) Give an example of ak ∈ C for which the radius of convergence of k=0 ak xk is
strictly greater than 1. (Bonus points for easiest example.)
P∞
iii) Give an example of ak ∈ C for which the radius of convergence of k=0 ak xk is
equal to 1.
9.3.3 Compare with the previous exercise: find the radii of convergence of the power series
P∞ k P∞ k
P∞ k
P∞ k P∞ k
P∞ k P∞ k
k=0 x , k=0 (−1)x , k=0 (2x) , k=0 x + k=0 (−1)x , k=0 x + k=0 (2x) .

9.3.4 Compute and justify the radius of convergence for the following series:
P k
i) 3x .
P
ii) (3x)k .
P
iii) 3kxk .
P
iv) k(3x)k .
P xk
v) k3
.
P 3x k
vi) k3 .
P (3x)k
vii) k3 .
9.3.5 Compute and justify the radius of convergence for the following series:
P xk
i) k.
P kxk
ii) 2k .
P k xk
iii) (2k)k
.
282 Chapter 9: Infinite series and power series
P∞ P∞
9.3.6 Let k=0 ak xk and k=0 bk xk be convergent power series with radii of convergence
P∞ P∞
R1 and R2 , respectively. Let R = min{R1 , R2 }. Prove that k=0 ak xk ± k=0 bk xk =
P∞ k
k=0 (ak ±bk )x is convergent with radius of convergence at least R. (Hint: Theorem 9.1.4.)
P∞
9.3.7 Let R be the radius of convergence of k=0 ak xk . Let p be a positive integer.
P
i) Determine the radius of convergence of ∞ k
k=p ak x .
P∞
ii) Determine the radius of convergence of k=0 ak xpk .
P∞
9.3.8 What would be a sensible definition for generalized power series k=0 ak (x − a)k ?
P∞
What would be a sensible definition of the radius of convergence of k=0 ak (x −a)k ? Draw
a relevant picture in C.

9.4 Differentiation of power series

Power series are functions. In this section we prove that they are differentiable at all
x inside the circle of convergence. Since a differentiable function is continuous, it follows
that a power series is continuous inside the circle of convergence.
Recall that for any differentiable function f , f ′ (x) = limh→0 f (x+h)−f (x)
h P , and any
P∞ P ∞ n
power series k=0 ak xk is actually the limit of a sequence: k=0 ak xk = lim{ k=0 ak xk }n .
Thus Pn Pn
X∞
k ′ lim{ k=0 ak (x + h)k }n − lim{ k=0 ak xk }n
( ak x ) = lim .
h→0 h
k=0
Pn Pn
Certainly by the sum rule for convergent series, lim{ k=0 ak (x+h)k }n −lim{ k=0 ak xk }n =
Pn
lim{ k=0 ak ((x + h)k − xk )}n , and by the constant rule we get that
∞
X n
X
k ′ (x + h)k − xk
( ak x ) = lim lim ak .
h→0 n→∞ h
k=0 k=0

If we could change the order of limits, then we would get by the polynomial rule for
derivatives that ∞ n
X X
k ′
( ak x ) = lim lim kak xk−1 .
n→∞ h→0
k=0 k=0

In fact, it turns out that this is the correct derivative, but our reasoning above was based
on an unproven (and generally false) switch of the two limits.
We give a correct proof of derivatives in the rest of the section. By Theorem 9.3.5
P∞ P∞ P∞
we already know that the series k=0 ak xk and k=0 kak xk−1 = k=1 kak xk−1 have the
same radius of convergence.
The following theorem is not necessarily interesting in its own right, but it is a stepping
stone in the proof of derivatives of power series.
 Section 9.4: Differentiation of power series 283
P
Theorem 9.4.1 Let k ak xk have radius convergence R. Let c ∈ C satisfy |c| < R. Then
P∞
the function g(x) = k=1 ak (xk−1 + cxk−2 + c2 xk−3 + · · · + ck−1 ) is defined on B(0, R)
and is continuous at c.

Proof. There is nothing to prove if R = 0, so we may assume that R > 0.

P∞ k−1
First of all, k=1 kak x converges on B(0, R) by Theorem 9.3.5. In particular,
P∞ k−1
P ∞ k−1
k=1 kak c = k=1 ak (c + cck−2 + c2 ck−3 + · · · + ck−1 ) is well-defined, but this is
simply g(c), so that c is in the domain of g.
Let ǫ > 0 and let d ∈ R satisfy |c| < d < R. Then again by Theorem 9.3.5, d is in the
P P P P
domain of k ak xk , k kak xk−1 , and also k k(k − 1)ak xk−2 and k k(k − 1)|ak ||x|k−2.
P
Set D = 21 k k(k − 1)|ak |dk−2 , and δ = min{d − |c|, D+1 ǫ
}. Then δ is positive. Let c ∈ C
with 0 < |x − c| < δ. Then by the triangle inequality, |x| = |x − c + c| ≤ |x − c| + |c| <
P
δ + |c| ≤ d − |c| + |c| = d < R, so that x is in the domain of the power series k ak xk .
P
But is x in the domain of g? Since the radius of convergence of k ak xk is R, by
P P
Theorem 9.3.5 also the radius of convergence of k kak xk−1 is R, so that k kak dk−1
converges. Then from the Comparison theorem (Theorem 9.2.2) and from
|ak (xk−1 + cxk−2 + c2 xk−3 + · · · + ck−1 )|
≤ |ak |(|x|k−1 + |c||x|k−2 + |c|2 |x|k−3 + · · · + |c|k−1 ) (by the triangle inequality)
≤ |ak |(dk−1 + ddk−2 + d2 dk−3 + · · · + dk−1 )
= k|ak |dk−1

we deduce that x is in the domain of g. Furthermore,

X X
|g(x) − g(c)| = | ak (xk−1 + cxk−2 + c2 xk−3 + · · · + ck−1 ) − kak ck−1 |
X
=| ak (xk−1 + cxk−2 + c2 xk−3 + · · · + ck−1 − kck−1 )|
X
=| ak ((xk−1 − ck−1 ) + (cxk−2 − ck−1 ) + (c2 xk−3 − ck−1 ) + · · · + (ck−1 − ck−1 )|
X
=| ak ((xk−1 − ck−1 ) + c(xk−2 − ck−2 ) + c2 (xk−3 − ck−3 ) + · · · + ck−2 (x − c))|
X
=| ak (x − c)((xk−2 + xk−3 c + xk−4 c2 + · · · + ck−2 )
+ c(xk−3 + xk−4 c + xk−5 c2 + · · · + ck−3 )
+ c2 (xk−4 + xk−5 c + xk−6 c2 + · · · + ck−4 ) + · · · + ck−2 )| (by Exercise 1.5.21)
X
≤ |ak ||x − c|((|x|k−2 + |x|k−3 |c| + |x|k−4 |c|2 + · · · + |c|k−2 )
+ |c|(|x|k−3 + |x|k−4 |c| + |x|k−5 |c|2 + · · · + |c|k−3 )
+ |c|2 (|x|k−4 + |x|k−5 |c| + |x|k−6 |c|2 + · · · + |c|k−4 ) + · · · + |c|k−2 )
(by the triangle inequality)
X
≤ |x − c| |ak |((dk−2 + dk−3 d + dk−4 d2 + · · · + dk−2 )
284 Chapter 9: Infinite series and power series

+ d(dk−3 + dk−4 d + dk−5 d2 + · · · + dk−3 )

+ d2 (dk−4 + dk−5 d + dk−6 d2 + · · · + dk−4 ) + · · · + dk−2 )
X
= |x − c| |ak |((k − 1)dk−2 + (k − 2)dk−2 + (k − 3)dk−2 + · · · + dk−2 )
X k(k − 1) k−2
= |x − c| |ak | d (by Example 1.5.1)
2
= |x − c|D
< δD
< ǫ.
This proves that g is continuous at c.
P∞
Theorem 9.4.2 Let f (x) = k=0 ak xk have radius of convergence R. Then f is differen-
P∞
tiable on B(0, R), and for all c ∈ B(0, R), f ′ (c) = k=1 kak ck−1 . The derivative function
P∞ P∞
f ′ (x) = k=0 kak xk−1 = k=1 kak xk−1 also has radius of convergence R.

Proof. Let c, x ∈ B(0, R). Then

P k
P k
f (x) − f (c) k ak x − k ak c
=
x−c x−c
P
ak (xk − ck )
= k
x−c
P
ak (x − c)(xk−1 + xk−2 c + xk−3 c2 + · · · + ck−1 )
= k (by Exercise 1.5.21)
x−c
X
= ak (xk−1 + xk−2 c + xk−3 c2 + · · · + ck−1 ),
k

which is the function g from the previous theorem. In that theorem we proved that g is
continuous at c, so that
X∞
′ f (x) − f (c)
f (c) = lim = lim g(x) = g(c) = kak ck−1 .
x→c x−c x→c
k=1

Then the theorem follows from Theorem 9.3.5.

P∞
Theorem 9.4.3 Suppose that the power series k=0 ak xk has radius of convergence R
P∞
and that the power series k=0 bk xk has radius of convergence S. Suppose that for some
positive r ≤ R, S the two power series take on the same values at all z ∈ Br (0). Then for
all n ≥ 0, an = bn .
P∞ P∞
In other words, if for all x ∈ Br (0), k=0 ak xk = k=0 bk xk , then the two series are
identical.
P∞ P∞
Proof. Let f (x) = k=0 ak xk and g(x) = k=0 bk xk . By assumption a0 = f (0) = g(0) =
b0 . Let n ∈ N+ . By Theorem 9.4.2, f and g are differentiable on Br (0), and since they
 Section 9.5: Numerical evaluations of some series 285

are identical, their nth derivatives agree on Br (0) as well. By an n-fold application of the
derivative,
X∞
(n)
f (x) = ak k(k − 1)(k − 2) · · · (k − n + 1)xk−n,
k=n

and similarly for g (n) . But then

an n(n − 1)(n − 2) · · · (n − n + 1) = f (n) (0) = g (n) (0)bnn(n − 1)(n − 2) · · · (n − n + 1),

so that an = bn .

Exercises for Section 9.4

1
9.4.1 Consider the function f (x) = 1−x and its corresponding geometric series on B(0, 1).
i) Compute a derivative of f as a rational function and as a power series.
ii) Compute an antiderivative of f as a rational function and as a power series.
P∞ x2k+1
9.4.2 Let f (x) = k=0 (2k+1)k! .
i) Find the radius of convergence of the series.
ii) Compute the derivative of f .
P∞ xk
iii) By Example 9.7.3 we know that for every real x, ex = k=0 k! . Find the series
2
for e(x ) .
2 c
iv) Find a power series whose derivative is e(x ) . (Comment: ab might stand for (ab )c
c c
or for a(b ) . But the first form is simply abc , so we would not write ab to stand for
c c
that. So, it is standard notation to write ab when we mean a(b ) , In particular,
2 2
e(x ) can be written more plainly as ex .)
v) Address the no-closed-form discussion on page 206.

9.5 Numerical evaluations of some series

Differentiation of power series is a powerful tool. For all complex numbers x ∈ B(0, 1)
P∞ 1
the geometric series k=0 xk converges to 1−x . Certainly it is much easier to compute
1
1−x than the infinite sum. We can exploit geometric series and derivatives of power series
to compute many other infinite sums. Below we provide a few illustrations of the method.
286 Chapter 9: Infinite series and power series
P∞ k
Example 9.5.1 k=1 2k−1 = 4.
P∞
Proof. Let f (x) = k=0 xk . This is the geometric series with radius of convergence 1 that
1
P∞ P∞
converges to 1−x (Example 9.3.2). By Theorem 9.4.2, f ′ (x) = k=0 kxk−1 = k=1 kxk−1 ,
and by Theorem 9.3.5, the radius of convergence of f ′ is also 1. Thus 12 is in the domain
of f ′ . Since we have two ways of expressing f (as power series and as a rational function),
there is similarly also a second form for f ′ :
 ′
′ 1 1
f (x) = = .
1−x (1 − x)2
P∞ k
P∞
k−1
From the two forms we deduce that k=0 2k−1 = k=0 k 21 1
= f ′ (1/2) = (1−1/2) 2 = 4.

P∞ k2
P∞ k2
Example 9.5.2 k=0 2k = 6, and k=0 2k−1 = 12.
P∞ k
Proof. As in the previous example we start with the geometric series f (x) = k=0 x
′
P∞ k−1 1
that converges on B(0, 1). Its derivative f (x) = k=0 kx = (1−x)2 also converges
P ∞ P∞ 2 k−1
on B(0, 1). Then xf ′ (x) = k=0 kx k
and its derivative (xf ′
(x)) ′
= k=0 k x also
converge on B(0, 1). From
X∞  ′
2 k−1 ′ ′ x 1+x
k x = (xf (x)) = 2
=
(1 − x) (1 − x)3
k=0
P∞ k2   P∞ k 2
1+ 21
we deduce that k=0 2k−1 = (1−1/2) 3 = 3/2
1/8 = 12, and k=0 2k = 6.

P∞ 1+x
Example 9.5.3 From the previous example we know that k=0 k 2 xk−1 = (1−x)3 . Multi-
P∞ x+x2
plying both sides by x gives k=0 k 2 xk = (1−x) 3 , and differentiation gives

X∞  
3 k−1 d x + x2 1 + 4x + x2
k x = = .
dx (1 − x)3 (1 − x)4
k=0
P∞ k3 1+4·0.5+0.52
P∞ 3
It follows that k=0 2k−1 = (1−0.5)4 = 52 and k=0 2kk = 26.
P∞ 1
P∞ k
P∞ k 2 P∞ k3
Summary: k=0 2 k−1 = 2, k=0 2 k−1 = 4, k=0 2k−1 = 12, k=0 2k−1 = 26. Is it
P∞ k 4
possible to predict k=0 2k−1 ?
P∞ 1
Example 9.5.4 k=1 k2k = ln 2.
P∞ k
Proof. Let f (x) = k=1 xk . The radius of convergence of f is 1, so 1/2 is in its domain.
P∞ 1
Also, f ′ (x) = k=1 xk−1 = 1−x , so that f (x) = − ln(1 − x) + C for some constant C. In
P∞ P∞
k
particular, C = −0 + C = − ln(1 − 0) + C = f (0) = 0, so that k=1 k21k = k=1 k1 12 =
f ( 21 ) = − ln(1 − 21 ) = − ln( 21 ) = ln(2).
 Section 9.6: Some technical aspects of power series 287

Exercises for Section 9.5

P∞
9.5.1 Compute and justify k=1 k31k .
P∞ k
9.5.2 Consider k=1 k+2·3 k . According to my computer the partial sum of the first 1000

terms is a rational number whose numerator and denominator take several screen pages,
so this sum as a rational number is hard to comprehend. So instead I computed curtailed
decimal expansions for this and for a few other sums: according to my computer, the
partial sum of the first 10 terms is about 0.719474635555091, the partial sum of the first
100 terms is about 0.719487336054311, the partial sum of the first 1000 terms is about
0.719487336054311. What can you suspect? How would you go about proving it?

9.6 Some technical aspects of power series

This section is meant as a reference, and should probably be skipped in a first class
on power series.
The result below seems obvious, but the proof is nevertheless a bit involved.
P∞ k
P∞ k
Theorem 9.6.1 Let k=0 ak x and k=0 bk x be convergent power series with radii
of convergence R1 and R2 , respectively. Let R = min{R1 , R2 }. Then on B(0, R)
Pn Pn
the product sequence {( k=0 ak xk ) · ( k=0 bk xk )}n converges to the power series
P∞ Pk k
k=0 ( j=0 aj bk−j )x .
P∞ P∞ P∞ Pk
We write this as ( k=0 ak xk ) · ( k=0 bk xk ) = k=0 ( j=0 aj bk−j )xk on B(0, R).

Proof. If R = 0, there is nothing to prove. So we may assume that R is a positive real

number or it is ∞. Let x ∈ B(0, R).
P∞ P∞
Set L = k=0 ak xk and K = k=0 bk xk (these limits exist as |x| < R1 , R2 ), sn =
Pn k
Pn k
Pn
k=0 ak x , tn = k=0 bk x , and cn = j=0 aj bn−j . By the theorem on the convergence
of products of sequences (Theorem 8.4.3), lim(sn tn ) = LK. Thus we need to prove that
P∞ k
k=0 ck x = LK.
Let ǫ > 0. By convergence of products, there exists N1 > 0 such that for all integers
n > N1 , |sn tn − LK| < ǫ/2. By expanding (repeated “foiling”) and by grouping,
n
! n 
X X
s n tn = ak xk  bj xj 
k=0 j=0

X k
n X n X
X n
k
= am bk−m x + am bn+k−m xn+k
k=0 m=0 k=1 m=k
Xn n
X n
X
= ck x k + am bn+k−m xn+k .
k=0 k=1 m=k
288 Chapter 9: Infinite series and power series
P∞ P∞ e K,
e respec-
Since |x| < R ≤ R1 , R2 , the series k=0 |ak xk | and k=0 |bk xk | converge, to L,
tively. Thus there exists N2 > 0 such that for all integers n > N2 ,
∞ 
X∞ X Xn  ǫ
k  k k 
|ak x | =  |ak x | − |ak x | < ,
  4K e +1
k=n+1 k=0 k=0
P∞ ǫ
and similarly there exists N3 > 0 such that for all integers n > N3 , k=n+1 |bk xk | < .
e+1
4L
Now let n > max{2N1 , N2 , N3 }. Then, with possibly adding more non-negative terms in
the third line below,
n X
X n
 
am bn+k−m xn+k 
k=1 m=k
n ⌊n/2⌋
X X  n n
 X X  
=  m
am x bn+k−m x n+k−m 
+ am xm bn+k−m xn+k−m 
k=1 m=k k=1 m=max{k,⌊n/2⌋+1}
     
⌊n/2⌋ n n ⌊n/2⌋
X X   X X  
≤  m 
|am x | bk x k 
+  |am xm |  bk xk 
m=1 k=⌊n/2⌋ m=⌊n/2⌋ k=1

e ǫ ǫ e
≤L + K
e+1
4L e +1
4K
ǫ
< .
2
Thus
   
Xn  X n 
   
 ck xk − LK  =  ck xk − sn tn + sn tn − LK 
   
k=0 k=0
 n 
X 
 k 
≤ ck x − sn tn  + |sn tn − LK|
 
k=0
 
Xn X n 
 n+k 
< am bn+k−m x  + ǫ/2
 
k=1 m=k
< ǫ/2 + ǫ/2 = ǫ.
P∞
Theorem 9.6.2 Let f (x) = k=0 ck xk have radius of convergence a positive real num-
P∞
ber R. Let a ∈ C with |a| = R such that k=0 ck ak converges. Let B be an open ball
centered at a, and let g : B → C be continuous. If f (x) = g(x) for all x ∈ B(0, R) ∩ B,
then f (a) = g(a).

Proof. Let ǫ > 0. We want to show that |f (a) − g(a)| < ǫ, which will via Theorem 2.10.4
then prove that f (a) = g(a). It suffices to prove the inequality |f (a) − g(a)| < ǫ under the
additional assumption that ǫ < 1.
P∞
Since k=0 ck ak converges, by Cauchy’s criterion Theorem 9.2.1, there exists a positive
 Section 9.6: Some technical aspects of power series 289
Pn PN +m
integer N such that for all integers n ≥ N , | k=N ck ak | < ǫ/4. Let sm = k=N ck ak .
By assumption, for all m ≥ 1, |sm | < ǫ/4 < 1. Furthermore, cN = s0 /aN , and for n > N ,
cN +n = (sn − sn−1 )/aN +n .
Let r be a positive real number strictly smaller than 1. First we do a simple expansion:

r N s0 +(s1 − s0 )r N +1 + (s2 − s1 )r N +2 + · · · + (sn − sn−1 )r N +n

= r N cN aN + cN +1 aN +1 r N +1 + cN +2 aN +2 r N +2 + · · · + cN +n aN +n r N +n
N
X +n
= ck (ra)k .
k=N
P 
 N +n k
Thus by more rewriting and by the triangle inequality,  k=N ck (ra)  equals
 
= r N s0 + (s1 − s0 )r N +1 + (s2 − s1 )r N +2 + · · · + (sn − sn−1 )r N +n 
 
= s0 (r N − r N +1 ) + s1 (r N +1 − r N +2 ) + · · · + sn−1 (r N +n−1 − r N +n ) + sn r N +n 


= r N (1 − r) s0 + s1 r + · · · + sn−1 r n−1 + sn r n 



≤ r N (1 − r) |s0 | + |s1 |r + · · · + |sn−1 |r n−1 + |sn |r n
ǫ


≤ r N (1 − r) 1 + r + · · · + r n−1 + r n (since |sm | < ǫ/4 for sufficiently large m)
4
ǫ
= r N (1 − r n + r n ) (by Example 1.5.4)
4
ǫ
= rN
4
ǫ
< .
4
P∞ 
In particular,  k=N ck (ra)k  ≤ ǫ . 4
Since polynomial functions are continuous, there exists δ1 > 0 such that for all x ∈
PN −1 PN −1 
B(a, δ1 ),  k=0 ck xk − k=0 ck ak  < ǫ/4.
Since g is continuous at a, there exists δ2 > 0 such that for all x ∈ B(a, δ2 ), |g(x) −
g(a)| < ǫ/4. Let r ∈ (0, 1) such that 1 − r < R1 min{δ1 , δ2 }. Then |a − ra| < δ1 , δ2 , so that
 
 NX −1 NX−1 
 
|g(a) − f (a)| = g(a) − g(ra) + g(ra) − f (ra) + f (ra) − ck ak + ck ak − f (a)
 
k=0 k=0
   
 N
X −1  NX −1 
 k  k 
≤ |g(a) − g(ra)| + |g(ra) − f (ra)| + f (ra) − ck a  +  ck a − f (a)
   
k=0 k=0
   
X ∞ N
X −1 N
X −1  X ∞ 
   
= |g(a) − g(ra)| + 0 +  ck (ra)k + ck (ra)k − ck ak  +  ck ak 
   
k=N k=0 k=0 k=N
 ∞  N −1   
ǫ  X  X  X 
NX−1 ∞
k  k k  k
< + ck (ra)  +  ck (ra) − ck a  +  ck a 
4      
k=N k=0 k=0 k=N
290 Chapter 9: Infinite series and power series
ǫ ǫ ǫ ǫ
< + + + = ǫ.
4 4 4 4
Since ǫ is arbitrary, by Theorem 2.10.4, f (a) = g(a).

Exercises for Section 9.6

P∞ P∞
9.6.1 Let k=0 ak xk , k=0 bk xk be power series and R a positive real number such that
P∞ k

P∞ k


for all x ∈ B(0, R), k=0 ak x · k=0 b k x = 1. In other words, at each x in B(0, R),
the two infinite series are multiplicative inverses of each other.
P∞
i) Prove that for all x ∈ B(0, R), k=0 ak xk 6= 0.
ii) Prove that a0 6= 0.
Pk
iii) Prove that b0 = a10 , and that for all k > 0, bk = − a10 j=1 aj bk−j . (Hint:
Theorem 9.6.1.) Suppose that there exists M ∈ R such that for all non-negative
integers n, |an xn | < M .
M +|a0 | k−1
iv) Prove that for all integers k ≥ 1, |bk xk | ≤ |aM 0|
2( |a0 | ) . (Hint: induction.)
P
9.6.2 (Abel’s lemma) Suppose that a0 , a1 , a2 , . . . are complex numbers and that k ak
P
converges. Let f (x) = k ak xk . The goal of this exercise is to prove that f is defined on
B(0, 1) ∪ {1}, that it is continuous on B(0, 1), and that as a function on the real interval
[0, 1], it is continuous.
i) Prove that the domain of f includes B(0, 1) ∪ {1}.
ii) Prove that f is continuous on B(0, 1). (Hint: Invoke a theorem.)
P
Set L = k ak , and for n ≥ 0, set sn = a0 + a1 + · · · + an − L.
Pn Pn
iii) Prove that k=0 sk (1−x)xk = sn (1−xn+1 )+ k=0 ak (xk −1). (Hint: Summation
by parts, see Exercise 9.2.11.)
iv) Prove that for every ǫ > 0 there exists N ∈ N such that for all integers n ≥ N ,
|sn | < ǫ.
Pn
v) Prove that the sequence { k=0 sk (1 − x)xk }n converges for all x ∈ B(0, 1) ∪ {1}.
P∞
vi) Prove that f (x) − L = k=0 sk (1 − x)xk for all x ∈ B(0, 1) ∪ {1}.
vii) Let ǫ > 0. Prove that there exists N ∈ R such that for all integers m ≥ N and for
P∞ 
all real x in the interval [0, 1],  k=m sk (1 − x)xk  < ǫ.
viii) Let ǫ > 0. Prove that for every positive integer n there exists δ > 0 such that for
Pn 
all complex x ∈ (B(0, 1) ∪ {1}) ∩ B(1, δ),  k=0 sk (1 − x)xk  < ǫ.
ix) Let ǫ > 0. Prove that there exists δ > 0 such that for every real x in the interval
P∞ 
[max{0, 1 − δ}, 1],  k=0 sk (1 − x)xk  < ǫ.
x) Let g : [0, 1] → C be defined by g(x) = f (x). Prove that g is continuous.
 Section 9.7: Taylor series 291
P∞ P∞
*9.6.3 Let f (x) = k=0 ak xk and g(x) = k=0 bk xk .
i) Express (f ◦ g)(x) as a power series in terms of the ai and bj .
ii) What is special for the power series of composition if a0 = 0?
iii) Assuming that a0 = 0, and given the radii of convergence for f and g, what would
it take to find the radius of convergence of the composition series?

9.7 Taylor series

A common way of generating power series comes from approximations of functions.

Definition 9.7.1 Let a be in the domain of a function f , and assume that f has derivatives
P∞ (k)
of all orders at a. The Taylor series of f (centered) at a is the series k=0 f k!(a) (x−a)k .

If a = 0, a Taylor series is a power series (as defined in this chapter), and for other a
this is also a power series, but of a more general kind.

Remark 9.7.2 Let f : A → C have derivatives of all orders at a. We shift the domain
and the function as follows: Let B = {x − a : x ∈ A}, and let g : B → C be defined by
P
g(x) = f (x + a). By straightforward calculus, k ak (x − a)k is a Taylor series for f at a
P
if and only if k ak xk is a Taylor series for g at 0. Thus any analysis of Taylor series can
by a shift be transformed into a Taylor series that is a power series.
Theorems 6.5.5 and 6.5.6 say that some Taylor series are convergent power series at
each x near 0 and that they converge to the value of the original function. We analyze one
such example below, and a few examples are left for the exercises. One of the Taylor series
in the exercises does not converge to its function at any point other than at a.

Example 9.7.3 Let f (x) = ex , where the domain of f is R. It is easy to compute the
Taylor series for f :
X∞
xk
.
k!
k=0

By the ratio test or the root test, this series converges for all x ∈ C, not only for x ∈ R.
(More on this series is in Sections 9.8 and 9.9, where we learn more about exponentiation
by complex numbers.) Let x ∈ R. By Theorem 6.5.5, for every positive integer n there
exists dn between 0 and x such that
n
X
x xk edn
e − = xn+1 .
k! (n + 1)!
k=0
292 Chapter 9: Infinite series and power series

Thus      
 n
X k   e|x|
 x x   edn n+1 

e − = x  ≤  x n+1 .

 k!  (n + 1)! (n + 1)!
k=0
n o
e|x|
The sequence (n+1)! xn+1 converges to 0 by the ratio test for sequences. This proves
that for each x ∈ R, the Taylor polynomials for ex approximate ex arbitrarily closely, and
that the Taylor series at each x equals ex :
∞
X
x xk
e = .
k!
k=0

In practice, the justification above is precisely how one (a human or a computer) com-
putes values ex to arbitrary precision. For example, to compute e0.1 to within 0.0001,
– following the machinery above – we take M = 1 > 0.1, and the smallest integer N
P16 0.1k
for which 21N ≤ 0.0001 22M −1M(3M
!
)M
is 16. Thus to within 0.0001, e0.1 is k=0 k! =
21021144462931669584965107651 ∼ −0.1
19020718080000000000000000000 = 1.10517091807565. To compute e , the same N works
P 16 (−0.1) k
7011749326061779176954720883 ∼
for desired precision, giving e−0.1 to be k=0 k! = 7749181440000000000000000000 =
π
0.90483741803596. To compute e to within 0.0001, we need bigger M , say M = 4,
P31 k
and then N = 31, giving eπ to be k=0 πk! = 23.1406926327793 ± 0.0001.

Example 9.7.4 Let f (x) = ln(x + 1), where the domain of f is (−1, ∞). It is straightfor-
ward to compute the Taylor series for f centered at 0:
∞
X (−1)k−1 xk
.
k
k=1

By the Ratio test for series, Theorem 9.2.4, the radius of convergence for this series is 1.
It is worth noting that the domain of the function f is all real numbers strictly bigger
than −1, whereas the computed Taylor series converges at all complex numbers in B(0, 1)
and diverges at all complex (including real) numbers whose absolute value is strictly bigger
than 1. By Example 9.1.7, the series diverges at x = −1, and by Theorem 9.2.6, it converges
at x = 1. You should test and invoke Theorem 6.5.5 to show that for all x ∈ (−1, 1),
P∞ k−1 k
ln(x + 1) = k=1 (−1) k x .
Incidentally, since ln(x + 1) is continuous on its domain and since its Taylor series
converges at x = 1 by the Alternating series test (Theorem 9.2.6), it follows from Theo-
rem 9.6.2 that
X∞
(−1)k−1
= ln 2.
k
k=1

With the help of Taylor series we can similarly get finite-term expressions for other
infinite sums. Below is another, harder example, and the reader may wish to skip it.
 Section 9.7: Taylor series 293
√
Example 9.7.5 Let f (x) = 1 − x. The domain of f is the interval (−∞, 1]. On the
subinterval (−∞, 1) the function has derivatives of all orders:
1 1 1 1 1 3
f ′ (x) = − (1 − x)−1/2 , f ′′ (x) = − · (1 − x)−3/2 , f ′′′ (x) = − · · (1 − x)−5/2 , . . . ,
2 2 2 2 2 2
1 1 3 2n − 3
f (n) (x) = − · · · · · · · (1 − x)−(2n−1)/2.
2 2 2 2
Thus the Taylor series for f centered at 0 is
∞
X 1 · 3 · 5 · · · (2k − 3)
1
1− x− k
xk .
2 k!2
k=2

For large n, the quotient of the (n + 1)st coefficient divided by the nth coefficient equals
2n−1
2(n+1) , whose limsup equals 1. Thus by the Ratio test for power series (Theorem 9.3.9),
the Taylor series converges absolutely on B(0, 1), and in particular it converges absolutely
on (−1, 1). Furthermore, the quotient 2(n+1)2n−1
= 2(n+1)−3 4/3
2(n+1) above is at most 1 − n for all
n ≥ 4. Thus by Raabe’s test (Exercise 9.2.17), this Taylor series converges at x = 1, and so
it converges absolutely on [−1, 1]. But what does it converge to? Consider x ∈ (−1, 1). We
use the integral form of the Taylor’s remainder theorem (Exercise 7.4.12): the difference
between Tn,f,0 (x) − f (x) is
Z x Z x n
(x − t)n 1 1 3 2n − 1 −(2n+1)/2 x−t 1 · 3 · 5 · · · (2n − 1)
· · · ··· (1 − t) dt = √ dt.
0 n! 2 2 2 2 0 1−t n!2n+1 1 − t
As the integrand goes to 0 with n, and as |x| < 1, the integral goes to 0 with n, so that
the Taylor series converges to f on (−1, 1). (Incidentally, an application of Exercise 9.6.2
shows that the Taylor series is continuous on [−1, 1], and as f is also continuous there,
necessarily the Taylor series converges to f on [−1, 1].)

Exercises for Section 9.7

1
9.7.1 Let f (x) = 1−x .
(n)
i) Compute f (x) for all integers n ≥ 0.
ii) Compute and justify the Taylor series of f centered at 0.
iii) Determine the radius of convergence.
iv) What is the domain of f and what is the domain of its Taylor series?
1
v) Compute the Taylor series for f (x) = 1−x centered at 2. Determine its radius of
convergence (for this you have to take coefficients of powers of x − 2 rather than
of powers of x).
1
vi) Compute the Taylor series for f (x) = 1−x centered at 3. Determine its radius of
convergence. Compare with the previous parts.
294 Chapter 9: Infinite series and power series

9.7.2 Let f (x) = ln(x+1), where the domain of f is (−1, ∞) as in Example 9.7.4. Compute
f (0.001) from one of its Taylor polynomials to four digits of precision. Show all work, and
in particular invoke Theorem 6.5.5.
9.7.3 Let f : R → R be given by
 2

f (x) = e−1/x , if x 6= 0;
0, if x = 0
as in Exercise 7.5.15.
i) Compute and justify the Taylor series for f centered at 0.
ii) Compute the radius of convergence for the series.
iii) Is it possible to compute f (0.001) from this series? Discuss.
*9.7.4 In introductory analysis courses we typically handle long, even infinite, sums but
not so much long products. In Example 9.7.5 we proved that the Taylor series for f (x) =
√
1 − x centered at 0 converges on [−1, 1]. This in particular means that the sequence
1·3·5···(2n−3)
{ n!2n }n converges to 0. Can you prove this directly without involving Raabe’s
test?

9.8 A special function

Define the power series

∞
X xk
E(x) = .
k!
k=0

By Examples 9.3.10 (3), the domain of E is C. Thus by Section 9.4, E is differentiable

everywhere.

Remarks 9.8.1
P∞ k−1 P∞ xk−1 P∞ k
(1) E ′ (x) = k=1 kxk! = k=1 (k−1)! = k=0 xk! = E(x).
(2) E(0) = 1.
(3) Let a ∈ C. Define g : C → C by g(x) = E(x + a) · E(−x). Then g is a product of
two differentiable functions, hence differentiable, and
g ′ (x) = E ′ (x + a) · E(−x) + E(x + a) · E ′ (−x) · (−1)
= E(x + a) · E(−x) − E(x + a) · E(−x)
= 0,

so that g is a constant function. This constant has to equal E(a) · E(−0) = E(a).
We conclude that for all a and x in C, E(x + a)E(−x) = E(a).
(4) Thus for all c, d ∈ C, E(c + d) = E(c)E(d) (set c = x + a, d = −x).
 Section 9.8: A special function 295

(5) By induction on n and by the previous part, it follows that for all positive integers
n and all a ∈ C, E(na) = (E(a))n .
(6) Part (3) applied to a = 0 gives that 1 = E(0) = E(x)E(−x). We conclude that
1
E(x) is never 0, and that E(−x) = E(x) .
n
(7) By parts (5) and (6), E(na) = (E(a)) for all integers n and all a ∈ C.
(8) Let a, b ∈ R. By part (4), E(a + bi) = E(a) · E(bi). Thus to understand the
function E : C → C, it suffices to understand E restricted to real numbers and E
restricted to i times real numbers.

Theorem 9.8.2 For all x ∈ R, E(x) = ln−1 (x) = ex , the exponential function from Sec-
tion 7.5. In particular, E(1) equals the Euler’s constant e.
E(x)
Proof. Define f : R → R by f (x) = ex . Then f is differentiable and
E ′ (x)ex − E(x)(ex )′ E(x)ex − E(x)ex
f ′ (x) = = = 0.
(ex )2 (ex )2
Thus f is a constant function, so that for all x ∈ R, f (x) = f (0) = E(0) 1
e0 = 1 = 1. Thus
E(x) = ex . By Theorem 7.5.7, this is the same as ln−1 (x). In particular, E(1) = e1 = e.
The following is now immediate:

Theorem 9.8.3 The function E restricted to R has the following properties:

(1) The range is all of R+ .
(2) E restricted to R is invertible with E −1 (x) = ln(x).
Pn 1
The sequence { k=0 k! }n of partial sums for E(1) = e1 = e is convergent, and
starts with 1, 2, 2.5, 83 ∼
= 2.66667, 65 ∼ 163 ∼ 1957 ∼ 685 ∼
24 = 2.70833, 60 = 2.71667, 720 = 2.71806, 252 =
2.71825, 1−9601 ∼ 98641 ∼
40320 = 2.71828, 36288 = 2.71828. The last decimal approximation turns out
to be correct to 5 decimal places.
In the next section we analyze the case of E restricted to i times real numbers. By
(8) in Remarks 9.8.1, we will then understand E completely.

Exercises for Section 9.8

9.8.1 Use the definition of ln in this section to prove that for all x, y ∈ R+ , ln(xy) =
ln x + ln y, and that for all integers n, ln(xn ) = n ln x.
9.8.2 Use the definition of ln in this section and derivatives of inverse functions (Theo-
rem 6.2.7) to prove that (ln)′ (x) = x1 .
9.8.3 Prove that {f : R → R differentiable | f ′ = f } = {cE(restricted to R) : c ∈ R}.
Similarly, prove that {f : C → C differentiable | f ′ = f } = {cE : c ∈ C}.
296 Chapter 9: Infinite series and power series
P∞
9.8.4 Find a0 , a1 , a2 , . . . ∈ R such that the power series k=0 ak xk converges for all x ∈ R
2
to ex . (Hint: Use the function E to do so easily.) With that, determine a power series
2
whose derivative is e(x ) . (This problem is related to Exercise 9.4.2; it turns out that there
2
is no simpler, finite-term antiderivative of ex .)
2
9.8.5 Determine a power series whose derivative is 2(x ) . (It turns out that there is no
simpler, finite-term antiderivative of this function – see the no-closed-form discussion on
page 206).
P∞ k
9.8.6 Express k=0 (−1) k
4 k!
in terms of e.
P∞ 2 k
9.8.7 Express k=0 3k (k+1)! in terms of e.
9.8.8 Compute a Taylor polynomial of degree n of the function E centered at 0. (Do as
little work as possible, but do explain your reasoning.)
9.8.9 Numerically evaluate e2 from the power series for ex = E(x) to 5 significant digits.
Prove that you have achieved desired precision. (Hint: Theorem 6.5.5.)
9.8.10 Compute limx→a E(x). (Do as little hard work as possible, but do explain your
reasoning.)
9.8.11 With proof, for which real numbers α does the sequence {E(nα)}n converge?

9.9 A special function, continued

In this section we restrict E from the previous section to the imaginary axis. Thus we
look at E(ix) for x being real. Note that
∞
X (ix)k
E(ix) =
k!
k=0
ix (ix)2 (ix)3 (ix)4 (ix)5 (ix)6 (ix)7 (ix)8 (ix)9
=1+ + + + + + + + + + ···
1! 2! 3! 4! 5! 6! 7! 8! 9!
x x2 x3 x4 x5 x6 x7 x8 x9
=1+i − −i + +i − −i + + i + ···.
1! 2! 3! 4! 5! 6! 7! 8! 9!
We define two new functions (their names may be purely coincidental, but pronounce them
for the time being as “cause” and “sin”):
x2 x4 x6 x8
COS (x) = Re E(ix) = 1 − + − + + ···,
2! 4! 6! 8!
x x3 x5 x7 x9
SIN (x) = Im E(ix) = − + − + + ···.
1! 3! 5! 7! 9!
Since E(ix) converges for all x, so do its real and imaginary parts. This means that COS
and SIN are defined for all x (even for complex x; but in this section all x are real!).
 Section 9.9: A special function, continued 297

Wencan q practice the root test on these functions: the α q for theqfunctiono COS is
  q
2  1 
  q
4  1 
 
6  1 
o n q      
2  1  4  1  6  1 
lim sup 1, 0, 2! , 0, 4! , 0, 6! , . . . = lim sup 1, 2! , 4! , 6! , . . . , and by
Example 8.3.6 this limsup α is 0. Thus the radius of convergence of COS is ∞. The proof
for the infinite radius of convergence of SIN is similar.

Remarks 9.9.1
(1) E(ix) = COS (x) + i SIN (x). This writing is not a random rewriting of the sum-
mands in E(ix), but it is the sum of the real and of the imaginary parts.
(2) COS (0) = Re E(i · 0) = Re 1 = 1, SIN (0) = Im E(i · 0) = Im 1 = 0.
(3) By the powers appearing in the power series for the two functions, for all x ∈ R,

COS (−x) = COS (x), SIN (−x) = −SIN (x).

Thus E(−ix) = COS (x) − i SIN (x), which is the complex conjugate of E(ix).
(4) For all x ∈ R,

(COS (x))2 + (SIN (x))2 = (COS (x) + i SIN (x)) · (COS (x) − i SIN (x))
= E(ix)E(−ix)
= E(ix − ix)
= E(0)
= 1.

Thus |E(ix)| = 1 for all x ∈ R.

(5) We conclude that for all x ∈ R, the real and imaginary parts of E(ix) have absolute
value at most 1. In other words,

−1 ≤ COS (x), SIN (x) ≤ 1.

(6) Since E is differentiable,

(E(ix))′ = E ′ (ix) i = E(ix) i = (COS (x) + i SIN (x)) i = i COS (x) − SIN (x).
It follows by Theorem 6.2.6 that

(COS (x))′ = (Re(E(ix)))′ = Re((E(ix))′ ) = −SIN (x),

and
(SIN (x))′ = (Im(E(ix)))′ = Im((E(ix))′) = COS (x).
298 Chapter 9: Infinite series and power series
√
Theorem 9.9.2 There exists a unique real number s ∈ (0, 3) such that E(is) = i, i.e.,
COS (s) = 0 and SIN (s) = 1.

Proof. The function t − SIN (t) is differentiable and its derivative is 1 − COS (t), which is
always non-negative. Thus t − SIN (t) is non-decreasing for all real t, so that for all t ≥ 0,
2
t −SIN (t) ≥ 0 −SIN (0) = 0. Hence the function t2 +COS (t) has a non-negative derivative
2 2
on [0, ∞), so that t2 + COS (t) is non-decreasing on [0, ∞). Thus for all t ≥ 0, t2 + COS (t)
2 3
≥ 02 +COS (0) = 1. It follows that the function t6 −t +SIN (t) is non-decreasing on [0, ∞).
3
[How long will we keep going like this???] Thus for all t ≥ 0, t6 − t + SIN (t)
3
t4 2
≥ 06 − 0 + SIN (0) = 0. Thus 24 − t2 − COS (t) is non-decreasing on [0, ∞), so that for all
t4 2
04 2
t ≥ 0, 24 − t2 − COS (t) ≥ 24 − 02 − COS (0) = −1. We conclude that for all t ≥ 0,
t4 t2
COS (t) ≤ − + 1.
24 2
√ √ 4 √ 2
In particular, COS ( 3) ≤ 243 − 23 + 1 = − 18 < 0. We also know that COS (0) =
1 > 0. Since COS is differentiable, it is continuous, so by the Intermediate value theorem
√
(Theorem 5.3.1) there exists s ∈ (0, 3) such that COS (s) = 0. Since (COS (s))2 +
3
(SIN (s))2 = 1, we have that SIN (s) = ±1. We proved that for all t ≥ 0, t6 −t+SIN (t) ≥ 0,
3 √
so that SIN (t) ≥ t − t6 = 6t (6 − t2 ). If t ∈ (0, 3], then SIN (t) ≥ 6t (6 − t2 ) > 0. Hence
SIN (s) = 1. This proves that E(is) = i.

Remark 9.9.3 The proof above establishes the following properties for all t ≥ 0:
t2 t2 t4
COS (t) ≤ 1, COS (t) ≥ 1 − , COS (t) ≤ 1 − + ,
2 2 24
t3
SIN (t) ≤ t, SIN (t) ≥ t − .
6
The pattern may be obvious at this point: Taylor polynomials of COS and SIN alternate
between over- and under-estimating. The limits of the Taylor polynomials, namely the
Taylor series, are precisely the given power series of COS and SIN .
Finally, we connect COS and SIN to trigonometric functions. First we need to specify
the trigonometric functions: For any real number t,
cos(t) + i sin(t)

is the unique complex number on the unit circle centered at the origin that is on the half-
ray from the origin at angle t radians measured counterclockwise from the positive x-axis.
In terms of ratio geometry (see page 21), this says that in a right triangle with one angle t
radians, cos(t) is the ratio of the length of the adjacent edge divided by the length of the
hypotenuse, and sin(t) is the ratio of the length of the opposite edge divided by the length
of the hypotenuse. Geometrically it is clear, and we do assume this fact, that cos and sin
 Section 9.9: A special function, continued 299

are continuous functions. We will use continuity in proving that the two functions are also
differentiable.

Theorem 9.9.4 The trigonometric functions cos and sin are differentiable. Furthermore,
COS and SIN are the functions cos and sin, s = π/2, and for any real number x, E(ix)
is the point on the unit circle centered at 0 at angle x radians measured counterclockwise
from the positive real axis.

Proof. We know that E(ix) is a point on the unit circle with coordinates (COS (x), SIN (x)).
What we do not yet know is whether the angle of this point counterclockwise from the
positive real axis equals x radians.
Let s be as in Theorem 9.9.2. The angle of E(is) = 0 + i · 1 = i measured in radians
counterclockwise from the positive real axis is π/2. The proof of Theorem 9.9.2 showed that
for t ∈ (0, 1], the imaginary part SIN (st) of E(ist) is positive, and the real part COS (st)
of E(ist) is strictly decreasing from 1 to 0. Thus E(ist) is in the first quadrant.
Let n be an integer strictly greater than 1. We will be analyzing s of the form 1/n.
Let α be the angle of E(is/n) counterclockwise from the positive real axis. I claim that
π
α = 2n . For every integer j, (E(ijs/n)) = E(is)j , so by Theorem 3.4.2, (E(ijs/n)) has
angle j · α counterclockwise from the positive real axis. By the previous paragraph, for
j = 1, . . . , n, since j/n ∈ (0, 1], E(ijs/n)) is in the first quadrant. Thus α, 2α, 3α, . . . nα
are in (0, π/2]. When j = n, the angle nα coincides with π/2. The angle π2 coincides with
π π
2 + 2πk for all integers k, so that α = 2n (1 + 4k) for some integer k. We know that for all
j = 1, . . . , n,
π π
0 < jα = j (1 + 4k) ≤ .
2n 2
The first inequality says that k is not negative, and the second one says that k is not
π
positive. Thus k = 0 and α = 2n .
For all positive integers n and all integers j, by Theorem 3.4.2, the angle of E(isj/n) =
(E(is/n))j is jα = π2 · nj . Since j and n are arbitrary integers with n ≥ 1, this says that
for all rational numbers r,
π  π 
COS (sr) + i SIN (sr) = E(isr) = cos r + i sin r .
2 2
Now let x be any real number. Then x is a limit point of Q, and since COS and cos
are continuous, by comparing the real parts above we get that
π  π 
COS (sx) = lim COS (sr) = lim cos r = cos x .
r∈Q,r→x r∈Q,r→x 2 2


Similarly, SIN (sx) = sin π2 x . In other words, for all x ∈ R,
   
2s 2s
cos(x) = COS x , sin(x) = SIN x .
π π
300 Chapter 9: Infinite series and power series

Since COS and SIN are differentiable, these equalities and the chain rule say that cos and
sin are differentiable as well and that
   
′ 2s 2s 2s ′ 2s 2s 2s
cos (t) = − SIN t = − sin(t), sin (t) = COS t = cos(t).
π π π π π π
By geometry (see Exercise 1.1.17), for small positive real x, 0 < cos(x) < sinx x < cos1 x .
Since cos is differentiable, it is continuous. Since cos(0) = COS (0) = 1, by the Squeeze
theorem (Theorem 4.3.11), limx→0+ sinx x = 1. Hence
2s 2s sin x − sin 0 sin x
= cos(0) = sin′ (0) = lim = lim+ = 1,
π π x→0 x x→0 x
whence s = π/2.

Theorem 9.9.5 Every complex number x can be written in the form rE(iθ), where r = |x|
is the length and θ the angle of x counterclockwise from the positive real axis.

Proof. Let r = |x|. Then x lies on the circle centered at 0 and of radius x. If r = 0, then
x = 0, and the angle is irrelevant. If instead r is non-zero, it is necessarily positive, and
x/r is a complex number of length |x|/r = 1. By Theorem 3.4.1, x/r and x have the same
angle. Let θ be that angle. Then x/r = E(iθ), so that x = rE(iθ).

Notation 9.9.6 It is common to write E(x) = ex for any complex number x. We have
seen that equality does hold if x is real, but we adopt this notation also for other numbers.
With this, if x, y ∈ R, then
ex+iy = ex eiy ,
and ex is the length and y is the angle of ex+iy counterclockwise from the positive x axis.

Exercises for Section 9.9

9.9.1 Let α and r be real numbers with r > 0. Prove that for any complex number x,
reiα · x is obtained from x by rotating x counterclockwise around the origin by angle α
radians and by stretching the result by a factor of r. (Compare with Theorem 3.4.1.)
9.9.2 Prove that eiπ + 1 = 0. (This has been called one of the most beautiful equalities,
because it connects the fundamental numbers 0, 1, e and π.)
9.9.3 Let x, y ∈ R. Expand both sides of E(i(x + y)) = E(ix)E(iy) in terms of sin and
cos to prove that cos(x + y) = cos(x) cos(y) − sin(x) sin(y), and sin(x + y) = sin(x) cos(y) +
cos(x) sin(y). (Cf. Exercise 1.1.13.)
9.9.4 Prove de Moivre’s formula: for all n ∈ Z and all real numbers x,
(cos(x) + i sin(x))n = cos(nx) + i sin(nx).
 Section 9.10: Trigonometry 301

9.9.5 Express the following complex numbers in the form ex+iy : i, −i, 1, −1, e, 2 + 2i.
Can 0 be expressed in this way? Justify.
9.9.6 Express (3 + 4i)5 in the form ea+bi for some a, b ∈ R. (Do not give numeric approx-
imations of a and b.)
9.9.7 Prove that for any integer m, the sequence {E(2πimn)}n converges. Prove that the
√
sequence {E( 2πin)}n does not converge. Determine all real numbers α for which the
sequence {E(inα)}n converges. Prove your conclusion.
9.9.8 For which real numbers α and β does the sequence {E(n(α+iβ))}n converge? Prove
your conclusion.
9.9.9 Determine a power series whose derivative is sin(x2 ). Repeat for cos(x2 ). (It turns
out that there are no simpler, finite-term antiderivatives of the functions sin(x2 ), cos(x2 ).)
*9.9.10 For the moment pretend that we do not know that COS = cos and SIN = sin.
√
Let f : [0, 1] → R be given by f (x) = 1 − x2 . Then the graph of f is a quarter of the
circle of radius 1 in the first quadrant.
i) Justify that
s  2
Z 1
π −2x
= 1+ √ dx.
2 0 2 1 − x2

(Hint: This is an improper integral; use lengths of curves; π is the perimeter of

half the circle of radius 1.)
ii) Compute the improper integral via substitution u = COS (x) and further other
steps. The integral should evaluate to s from Theorem 9.9.2.

9.10 Trigonometry

In this section we review the semester’s concepts of limits, continuity, differentiability,

and integrability on the newly derived trigonometric functions from the previous section.

Definition 9.10.1 For any x, eix is the complex number that is on the unit circle at angle
x radians counterclockwise from the positive horizontal axis.
(1) sin(x) is the imaginary part of eix .
(2) cos(x) is the real part of eix .
sin(x)
(3) tan(x) = cos(x) , cot(x) = cos(x)
sin(x)
1
, sec(x) = cos(x) 1
, csc(x) = sin(x) .
302 Chapter 9: Infinite series and power series

Remarks 9.10.2 The following follow easily from the previous section:
(1) sin and cos are differentiable and hence continuous functions whose domain is R.
(2) For all x ∈ R, sin(x + 2π) = sin(x) and cos(x + 2π) = cos(x). This is not obvious
from the power series definition of E(ix), but if follows from Theorem 9.9.5.
(3) The functions sin and cos are continuous, i.e., for all a ∈ R, lim sin(x) = sin(a),
x→a
lim cos(x) = cos(a). (In the proof of Theorem 9.9.4 we assumed that cos is
x→a
continuous, but we did not assume that sin is continuous.)
(4) For all a ∈ R in the domain of tan (resp. cot, sec, csc), tan (resp. cot, sec, csc) is
continuous and differentiable at a.
ix −ix
(5) sin(x) = e −e 2i
.
eix +e−ix
(6) cos(x) = 2 .
(7) (sin(x))2 + (cos(x))2 = 1. (Recall from Remark 2.4.7 that for trigonometric func-
tions we also write this as sin2 (x) + cos2 (x) = 1, but for an arbitrary function f ,
f 2 (x) refers to f (f (x)) rather than for (f (x))2.)
(8) (tan(x))2 + 1 = (sec(x))2 .
(9) (cot(x))2 + 1 = (csc(x))2 .
(10) sin′ (x) = cos(x).
(11) cos′ (x) = − sin(x).
(12) For all x ∈ R, sin(x) = − sin(−x) and cos(x) = cos(−x).
(13) sin and cos take on non-negative values on [0, π/2]. By the previous part, cos takes
on non-negative values on [−π/2, π/2].
(14) sin is increasing on [−π/2, π/2].
(15) sin, when restricted to [−π/2, π/2], has an inverse. The inverse is called arcsin.
The domain of arcsin is [−1, 1].
(16) Geometrically, sin takes on non-negative values on [0, π], so that cos, when re-
stricted to [0, π], has an inverse, called arccos. The domain of arccos is [−1, 1].
(17) Verify the details in the following. The derivative of tan is always non-negative, so
that on (−π/2, π/2), tan is invertible. Its inverse is called arctan, and the domain
of arctan is (−∞, ∞).

Proposition 9.10.3 Refer to Theorem 6.2.7.

(1) For x ∈ (−π/2, π/2), arcsin′ (x) = √1−x
1
2
.
′ −1
(2) For x ∈ (0, π), arccos (x) = √1−x2 .
1
(3) arctan′ (x) = 1+x 2.

Proof.
1
arcsin′ (x) = ′
sin (arcsin(x))
 Section 9.10: Trigonometry 303

1
=
cos(arcsin(x))
1
= p (because cos is non-negative on [−π/2, π/2])
(cos(arcsin(x)))2
1
= p
1 − (sin(arcsin(x)))2
1
= √ .
1 − x2
The proof of the second part is similar:
1
arctan′ (x) = ′
tan (arctan(x))
1
=
(sec(arctan(x)))2
1
=
1 + (tan(arctan(x)))2
1
= .
1 + x2

Exercises for Section 9.10

9.10.1 Compute the derivatives of the standard trigonometric functions tan(x), cot(x),
sec(x), csc(x).
9.10.2 Use integration by substitution (Exercise 7.4.5) to compute the antiderivatives of
tan(x) and cot(x).
9.10.3 Use integration by substitution (Exercise 7.4.5) and a rewriting trick to compute
an antiderivative of sec(x) = sec(x)(sec(x)+tan(x))
sec(x)+tan(x) . Compute an antiderivative of csc(x).
9.10.4 Prove properly Example 6.3.6 that for all positive x, x > sin(x).
9.10.5 Let f : [0, π/2] → R be defined by f (x) = (π/2) sin(x) − x.
i) Prove that there is exactly one c ∈ [0, π/2] such that f ′ (c) = 0.
ii) Prove that (π/2) sin(x) ≥ x for all x ∈ [0, π/2]. (Compare with Exercise 9.10.4.)
9.10.6 Prove that for all x ∈ R,
1 − cos(2x) 1 + cos(2x)
(sin(x))2 = , (cos(x))2 = .
2 2
R R
9.10.7 Compute (sin(x))2 dx (or (cos(x))2 dx) twice, with two different methods:
i) Use the previous exercise.
ii) Use integration by parts twice (Exercise 7.4.6).
304 Chapter 9: Infinite series and power series

9.10.8 Let k, j ∈ Z. Show all of the following (possibly by multiple integration by parts
(Exercise 7.4.6)):
Rπ
i) −π sin(kt) cos(jt) dt = 0.
ii)
Z π (
0, if j 6= k,
sin(kt) sin(jt) dt = 0, if jk = 0,
−π π, otherwise.
iii) Z π 
0, if j 6= k or jk 6= 0,
cos(kt) cos(jt) dt =
−π π, otherwise.
9.10.9 Prove that for any non-zero integer k,
Z
1 1
x sin(kx) dx = − x cos(kx) + 2 sin(kx) + C,
k k
and that Z
1 1
x cos(kx) dx = x sin(kx) + 2 cos(kx) + C.
k k
9.10.10 Let t ∈ R and n ∈ N+ .
i) Prove that sin((n + 12 )t) = sin((n − 12 )t) + 2 cos(nt) sin( 21 t). (Hint: Exercise 9.9.3,
multiple times.)
Pn

ii) Prove that sin((n + 12 )t) = sin( 21 t) 1 + 2 k=1 cos(kt) .
Pn
iii) Give a formula or formulas for simplifying k=1 cos(kt).
9.10.11 (Fourier analysis, I) Let f : [−π, π] → C be a differentiable function (or pos-
sibly not quite differentiable, but that would take us too far). For n ∈ N0 define an =
Rπ Rπ
−π
f (x) cos(nx) dx, and for n ∈ N+ define bn = −π f (x) sin(nx) dx. The Fourier series
of f is
∞
a0 X
+ (an cos(x) + bn sin(x)).
2 n=1
i) Compute the Fourier series for f (x) = sin(3x). (Hint: Exercise 9.10.8.)
ii) Compute the Fourier series for f (x) = (sin(x))2 . (Hint: Exercises 9.10.6 and
9.10.8.)
iii) Compute a0 , a1 , a2 , b1 , b2 for f (x) = x. (Hint: Exercise 9.10.9.)
9.10.12 (Fourier analysis, II) Let f : [−π, π] → C be a differentiable function (or pos-
sibly not quite differentiable, but that would take us too far). For n ∈ Z define an =
Rπ
−π
f (x)einx dx. The Fourier series of f is
∞
X
an e−inx .
n=−∞

Compute these Fourier series for the functions in the previous exercise.
 Section 9.11: Examples of L’Hôpital’s rule 305

9.11 Examples of L’Hôpital’s rule

L’Hôpital’s rule was proved in many versions in Section 6.4. Now that we have all
the necessary facts about trigonometric functions we can show more interesting examples,
including counterexamples if some hypothesis is omitted. All work in this section is in the
exercises.

Exercises for Section 9.11

The next two exercises are taken from R. J. Bumcrot’s article “Some subtleties in
L’Hôpital’s rule” in Century of Calculus. Part II. 1969–1991, edited by T. M. Apostol,
D. H. Mugler, D. R. Scott, A. Sterrett, Jr., and A. E. Watkins. Raymond W. Brink
Selected Mathematical Papers. Mathematical Association of America, Washington, DC,
1992, pages 203–204.
9.11.1 Let f (x) = x2 sin(x−1 ) and g(x) = sin x.
f (x)
i) Prove that limx→0 g(x) = 0.
′
ii) Prove that limx→0 fg′ (x)
(x)
does not exist.
iii) Why does this not contradict L’Hôpital’s rule Theorem 6.4.3?
9.11.2 Let f (x) = x sin(x−1 ) and g(x) = sin x.
i) Compute limx→0 f (x) and limx→0 g(x).
ii) Compute f ′ and g ′ .
f (x) f ′ (x)
iii) Prove that neither limx→0 g(x) nor limx→0 g ′ (x) exist.
*9.11.3 (This is from R. P. Boas’s article “Counterexamples to L’Hôpital’s rule”, American
Rx f (x)
Mathematical Monthly 93 (1986), 644–645.) Define f (x) = 0 cos2 (t) dt, g(x) = 2+sin(x) .
i) Prove that lim f (x) = ∞ = lim g(x).
x→∞ x→∞
f (x)
ii) Prove that lim 6= 0.
x→∞ g(x)
iii) Compute f ′ (x), g ′ (x).
iv) Write f ′ (x) = f1 (x) · cos x and g ′ (x) = g1 (x) · cos x for some functions f1 , g1 . Prove
f1 (x)
that lim = 0.
x→∞ g1 (x)
f ′ (x)
v) Prove that lim ′ 6= 0.
x→∞ g (x)
vi) Why does this not contradict L’Hôpital’s rule at infinity (proved in Exercise 6.4.5)?
306 Chapter 9: Infinite series and power series

*9.11.4 (This is from R. C. Buck’s “Advanced Calculus”, McGraw-Hill Book Company,

Inc., 1956, page 48.) Let f (x) = 2x + sin(2x), g(x) = (2x + sin(2x))esin(x) , and h(x) =
e− sin(x) .
i) Prove that limx→∞ h(x) does not exist.
ii) Prove that limx→∞ fg(x)
(x)
does not exist.
iii) Verify that f ′ (x) = 4 cos2 (x) and that g ′ (x) = (4 cos(x)+2x+sin(2x)) cos(x)esin(x) .
4 cos(x)
iv) Prove that limx→∞ (4 cos(x)+2x+sin(2x))e sin(x) = 0.

v) Why does L’Hôpital’s rule not apply here?

2
*9.11.5 (Modification of the previous exercise.) Let f (x) = x + sin( x2 ), and g(x) =
1 f ′ (x)
( x2 + sin( x2 ))esin( x ) . Prove that limx→0+ fg(x)
(x)
does not exist but that limx→0+ g ′ (x) = 0.
Why does L’Hôpital’s rule not apply here?

9.12 Further specialized uses of trigonometry

Exercises for Section 9.12

9.12.1 (This is taken from the Monthly Gem article by D. Velleman in the American
Mathematical Monthly 123 (2016), page 77. The gem is a simplification of the article
P∞ −2
by Y. Matsuoka, An elementary proof of the formula k=1 k = π 2 /6, in American
Mathematical Monthly 68 (1961), 485–487.) For every non-negative integer n define In =
R π/2 2n
R π/2 2
0
(cos(x)) dx and J n = 0
x (cos(x))2n dx.
i) Prove that I0 = π/2 and J0 = π 3 /24.
ii) Prove that for all n ≥ 1, In = (2n − 1)(In−1 − In ). (Hint: integration by parts
(Exercise 7.4.6) and dv = cos(x) dx.)
iii) Prove that In = 2n−1
2n · In−1 .
iv) Use integration by parts twice, first with dv = dx and then with dv = 2xdx, to
2
prove that for all n ≥ 1, In = n(2n
 − 1)Jn−1− 2n In .
1 Jn−1 Jn
v) Prove that for all n ≥ 1, n2 =2 In−1 − In .
vi) Prove that for all integers N ,
XN  
1 J0 JN π2 JN
2
= 2 − = −2 .
n=1
n I0 IN 6 IN

vii) Use the inequality x ≤ (π/2) sin(x) for x ∈ [0, π/2] (see Exercise 9.10.5) to estimate
Z
π 2 π/2 π2 π2 1
0 ≤ JN ≤ (sin(x))2 (cos(x))2N dx = (IN − IN +1 ) = · IN .
4 0 4 4 2N + 2
 Section 9.12: Further specialized uses of trigonometry 307

viii) Prove that

X∞
1 π2
= .
n=1
n2 6

*9.12.2 (This exercise is used in Exercise 9.12.3.) Use de Moivre’s formula (see Exer-
cise 9.9.4) to prove that for all x ∈ R and all positive integers n,
Xn  
n n n n k
cos(nx) + i sin(nx) = (sin(x)) (cot(x) + i) = (sin(x)) i (cot(x))n−k .
k
k=0

If n = 2m + 1, this says that

X 
2m+1 
2m + 1 k
2m+1
cos((2m + 1)x) + i sin((2m + 1)x) = (sin(x)) i (cot(x))2m+1−k .
k
k=0

The imaginary part of this is:

m
X  
2m+1 k 2m + 1
sin((2m + 1)x) = (sin(x)) (−1) (cot(x))2m−2k .
2k + 1
k=0
Pm

k 2m+1 m−k
Let P (x) = k=0 (−1) 2k+1 x . Prove that P is a polynomial of degree m with
leading coefficient 2m + 1. Prove that for k = 1, . . . , m, (cot(kπ/(2m + 1)))2 is a root of P .
Prove that P has m distinct roots. Prove that
Ym   2 !
kπ
P (x) = (2m + 1) · x − cot .
2m + 1
k=1
m−1
Pm
By equating the coefficient of x in the two forms of P prove that k=1 (cot(kπ/(2m +
1)))2 = m(2m−1)
3 .
*9.12.3 (See also Exercises 9.12.1 and 9.12.5.) (This is from the article I. Papadimitriou,
P∞ −2
A simple proof of the formula k=1 k = π 2 /6, American Mathematical Monthly 80
(1973), 424–425.) Let x be an angle measured in radians strictly between 0 and π/2. Draw
the circular wedge with angle x on a circle of radius 1. The largest right triangle in this
wedge whose hypotenuse is one of the wedge sides of length 1 has area 12 sin(x) cos(x), and
the smallest right triangle containing this wedge whose side is one of the wedge sides (of
length 1) has area 12 tan(x).
i) Prove that sin(x) cos(x) < x < tan(x), and that (cot(x))2 < x12 .
ii) Use Exercise 9.10.4 to prove that x12 < 1 + (cot(x))2 .
Pm   kπ 2
iii) Prove that for all integers k, m with 1 ≤ k ≤ m, k=1 cot 2m+1 <
  
(2m+1)2 Pm Pm 2
1 kπ
π 2 k=1 k 2 < m + k=1 cot 2m+1 .
2 P
m
iv) Use Exercise 9.12.2 to prove that m(2m−1)3 < (2m+1)
π 2
1
k=1 k2 < m +
m(2m−1)
3 .
P∞ 1 π 2
2 2
v) Prove that k=1 k2 = 6 . (Hint: multiply the previous part by π /4m .)
308 Chapter 9: Infinite series and power series

9.12.4 Fill in the explanations and any missing steps in the two double improper integrals.
(While integrating with respect to x, think of y as a constant, and while integrating with
respect to y, think of x as a constant.)
Z ∞ Z ∞  Z ∞ Z ∞ 
1 1 1 1
· dx dy = dx dy
0 0 1 + y 1 + x2 y 0 1+y 0 1 + x2 y
Z ∞ Z N !
1 1
= lim dx dy
0 1 + y N →∞ 0 1 + x2 y
Z ∞ √  !
1 arctan( yx) N
= lim √  dy
0 1 + y N →∞ y x=0
Z
π ∞ 1
= √ dy
2 0 y(1 + y)
Z
π ∞ 2u
= du (by substitution y = u2 )
2 0 u(1 + u2 )
π2
= ,
2
Z ∞ Z ∞  Z ∞ Z ∞   
1 1 1 1 x2
· dy dx = − dy dx
0 0 1 + y 1 + x2 y 0 0 1 − x2 1 + y 1 + x2 y
Z ∞  
1 1
= ln dx
0 1 − x2 x2
Z ∞
ln x
=2 2
dx
0 x −1
Z 1 Z ∞ 
ln x ln x
=2 2
dx + dx
0 x −1 1 x2 − 1
Z 1 Z 1 
ln x ln u
=2 2
dx + 2
du
0 x −1 0 u −1
Z 1
ln x
=4 2
dx. (Stop here.)
0 x −1
2
*9.12.5 (This is from the article D. Ritelli, Another proof of ζ(2) = π6 , American Mathe-
matical Monthly 120 (2013), 642–645.) We proved in Section 9.4 that derivatives and (def-
inite) integrals commute with infinite sums for power series. There are other cases where
integrals commute with infinite sums, but the proofs in greater generality are harder. Ac-
R 1 ln x R1 P∞ 2k P∞ R 1
cept that 0 − 1−x 2 dx = 0
(− ln x) k=0 x dx = 2k
k=0 0 (− ln x)x dx. Also accept that
the two integrals in Exercise 9.12.4 are the same (order of integration matters sometimes).
P∞ 2
Use Exercises 7.4.7 and 9.2.1 to prove that k=1 k12 = π6 .
Appendi es
Appendix A. Advice on writing mathematics

Make your arguments succinct and straightforward to read

Write preliminary arguments for yourself on scratch paper: your first attempt may
yield some dead ends which definitely should not be on the final write-up. In the final
write-up, write succinctly and clearly; write what you mean and mean what you write;
write with the goal of not being misunderstood. Use good English grammar, punctuate
properly. And above all, use correct logical reasoning.
Do not allow yourself to turn in work that is half-thought out or that is produced in
a hurry. Use your best writing, in correct logical order, with good spatial organization on
paper, with only occasional crossing out of words or sections, on neat paper. Represent
your reasoning and yourself well. Take pride in your good work.

Process is important
Perhaps the final answer to the question is 42. It is not sufficient to simply write “42”,
“The answer is 42,” or similar, without the process that led to that answer. While it is
extremely beneficial to have the intuition, the smarts, the mental calculating and reason-
ing capacity, the inspiration, or what-not, to conclude “42”, a huge part of learning and
understanding is to be able to explain clearly the reasoning that lead to your answer.
I encourage you to discuss the homework with others before, during or after completing
it: the explanations back-and-forth will make you a better thinker and expositor.
Write your solutions in your own words on your own, and for full disclosure write the
names of all of your collaborators on the work that you turn in for credit. I do not take
points off, but you should practice full honesty.
Sometimes you may want to consult a book or the internet. Again, on the work that
you turn in disclose the help that you got from outside sources.
Keep in mind that the more you have to consult outside sources, the more fragile your
stand-alone knowledge is, the less well you understand the material, and the less likely you
are to be able to do satisfactory work on closed-book or limited-time projects.

Do not divide by 0
Never write “1/0”, “0/0”, “02 /0”, “∞/0”. (Erase from your mind that you ever saw
this in print! It cannot exist.)
312 Appendix A: Advice on writing mathematics

Sometimes division by zero creeps in in subtler ways. For example, to find solutions
2
to x = 3x, it is wrong to simply cancel the x on both sides to get only one solution x = 3.
Yes, x = 3 is one of the solutions, but x = 0 is another one. Cancellation of x in x2 = 3x
amounts to dividing by 0 in case is the solution is x = 0.

Never plug numbers into a function that are not in the domain of the function
By design, the only numbers you can plug into a function are those that are in the
domain of the function. What else is there to say?
I will say more, by way of examples. Never plug 0 into the function f (x) = x1 (see
previous admonition). Even never plug 0 into the function f (x) = xx : the latter function
is undefined at x = 0 and is constant 1 at all other x.
√
Never plug −1 into or into ln. 
3x − 4, if x > 3;
Do not plug x = 0 or x = 1.12 into f that is defined by f (x) =
2x + 1, if x < −1.

Order of writing is important

The meanings of “Everybody is loved by somebody” and “Somebody loves everybody”
are very different. Another way of phrasing these two statements is as follows: “For every x
there exists y such that y loves x,” and “There exists y such that for every x, y loves x.”
In crisp symbols, without distracting “such that”, “for”, and commas, these are written as
“ ∀x ∃y, y loves x” and “ ∃y ∀x, y loves x.”
Conclusion: order of quantifiers matters.
Certainly also the order matters in implications; simply consider the truth values of
“If x > 2 then x > 0” and “If x > 0 then x > 2.”
The statement “if A then B” can be written also as “B if A”, but in general it is better
to avoid the latter usage. In particular, when writing a long proof, do not write out a very
long B and only at the very end add that you were assuming A throughout; you could have
lost the doubting-your-statements reader before the disclaimer.
Here is a fairly short example where “B if A” form is not as elegant. With the statement
2
“ x is ... if x 6= 0” you might have gone over the abyss of dividing by 0 and no ifs can make
you whole again. It is thus better to first obtain proper assurances, and write “if x 6= 0
then x2 is ... .”

Write parentheses
·− is not a recognized binary operator. Do not write “5 · −2”; instead write “5 · (−2)”.
lim 4 − 3x = 4 − 3x, whereas lim (4 − 3x) = 7.
x→−1 x→−1
R R
“ 4 − 3x dx” is terrible grammar; instead write “ (4 − 3x) dx”.
 Appendix A: Advice on writing mathematics 313

π does not equal 3.14159

√
If the answer to a problem is π 17/59, leave it at that. This is an exact number from
which one can get an approximation to arbitrary precision, but from 0.21954437870195
√
one cannot recover further digits. Never write “π 17/59 = .21954437870195”, but it is
√
fine to write “π 17/59 ∼ = .21954437870195”. Usually it is not necessary to write out the
numerical approximation, but sometimes the approximation helps us get a sense of the
size of the answer and to check our derivation with any intuition about the problem. The
answer to how far a person can run in one minute certainly should not exceed a kilometer
or a mile.

To prove an (in)equality, manipulate one side in steps to get to the other

Pn
If you have to prove that k=1 k 2 = n(n+1)(2n+1)
6 for n = 1, do NOT do the following:
1
X 1(1 + 1)(2 · 1 + 1)
k2 =

DO THIS!
DO NOT
6
k=1
1·2·3
12 =
6
1=1
The reasoning above is wrong-headed because in the first line you are asserting the equality
that you are expected to prove, and in subsequent lines you are simply repeating your
assumptions more succinctly. If you add question marks over the three equal sums and a
check mark on the last line, then you are at least acknowledging that you are not yet sure
of the equality. However, even writing with question marks over equal signs is inelegant
and long-winded. That kind of writing is what we do on scratch paper to get our bearings
on how to tackle the problem. But a cleaned-up version of the proof would be better as
follows:
X 1
1·2·3 1(1 + 1)(2 · 1 + 1)
k 2 = 12 = 1 = = .
6 6
k=1
Do you see how this is shorter and proves succinctly the desired equality by transitivity of
equality, with each step on the way sure-footed?
Another reason why the three-line reasoning above is bad is because it can lead to the
following nonsense:
?
1=0
?
add 1 to both sides: 2=1
? √
multiply both sides by 0: 0=0

But we certainly cannot conclude that the first line “1 = 0” is correct.

314 Appendix A: Advice on writing mathematics

Do not assume what is to be proved

If you have to prove that a is positive, do not assume that a is positive. (For a proof by
contradiction suppose that a ≤ 0, do correct logical reasoning until you get a contradiction,
from which you can conclude that a ≤ 0 is impossible, so that a must have been positive.)

Limit versus function value

Asking for limx→a f (x) is in general different from asking for f (a). (If the latter is
always meant, would there be a point to developing the theory of limits?)

Do not start a sentence with a mathematical symbol

This admonition is in a sense an arbitrary stylistic point, but it helps avoid certain
confusions, such as in “Let ǫ > 0. x can be taken to be negative.” Here one could read
or confuse the part “0. x” as “0 · x = 0”, but then ǫ would have to be both positive and
negative. Do not force a reader to have to do a double-take: write unambiguous and correct
sentences.
Appendix B. What you should never forget

Logic
You should remember the basic truth tables, correct usage of “or” and of implications,
how to justify/prove a statement, and how to negate a statement.
If A implies B and if A is true, we may conclude B.
If A implies B and if B is false, we may conclude that A is false.
If A implies B and if A is false, we may not conclude anything.
If A implies B and if B is true, we may not conclude anything.

Truth table:

P Q not P P and Q P or Q P xor Q P ⇒Q P ⇔Q

T T F T T F T T
T F F F T T F F
F T T F T T T F
F F T F F F T T

Statement Negation
not P P

P and Q not (P and Q) = ( not P ) or ( not Q)

P or Q not (P or Q) = ( not P ) and ( not Q)

P ⇒Q not (P ⇒ Q) = P and ( not Q)

P ⇔Q not (P ⇔ Q) = P ⇔ ( not Q) = ( not P ) ⇔ Q

For all x of a specified type, prop- There exists x of the specified type such that
erty P holds for x. P is false for x.

There exists x of a specified type For all x of the specified type, P is false for x.
such that property P holds for x.
316 Appendix B: What you should never forget

Statement How to prove it

P (via contradiction). Suppose not P . Establish some nonsense
(that makes not P thus impossible so that P
must hold).

P and Q. Prove P . Prove Q.

P or Q. Suppose that P is false. Then prove Q.

Alternatively: Suppose that Q is false and
then prove P .
(It may even be the case that P is true always.
Then simply prove P . Or simply prove Q.)

If P then Q. Suppose that P is true. Then prove Q.

Contrapositively: Suppose that Q is false.
Prove that P is false.

P ⇔ Q. Prove P ⇒ Q. Prove Q ⇒ P .

For all x of a specified type, prop- Let x be arbitrary of the specified type. Prove
erty P holds for x. that property P holds for x.
(Possibly break up into a few subcases.)

There exists x of a specified type Find/construct an x of the specified type.

such that property P holds for x. Prove that property P holds for x.
Alternatively, invoke a theorem guaranteeing
that such x exists.

An element x of a specified type Suppose that x and x′ are both of specified

with property P is unique. type and satisfy property P . Prove that
x = x′ .
Alternatively, show that x is the only solution
to an equation, or the only element on a list,
or ....
 Appendix B: What you should never forget 317

Mathematical induction
The goal is to prove a property for all integers n ≥ n0 . First prove the base case,
namely that the property holds for n0 . For the inductive step, assume that for some
n − 1 ≥ n0 , the property holds for n − 1 (alternatively, for n0 , n0 + 1, . . . , n − 1), and then
prove the property for n.

The limit definition of derivative

f (a + h) − f (a) f (x) − f (a)

f ′ (a) = lim = lim .
h→0 h x→a x−a

The limit-partition definition of integrals

A partition of [a, b] is a finite set P = {x0 , x1 , . . . , xn } such that x0 = a < x1 < x2 <
· · · < xn−1 < xn = b. Let f : [a, b] → R be a bounded function. For each j = 1, . . . , n, let
mj = inf{f (x) : x ∈ [xj−1 , xj ]}, Mj = sup{f (x) : x ∈ [xj−1 , xj ]}.
The lower sum of f with respect to P is
n
X
L(f, P ) = mj (xj − xj−1 ).
j=1

The upper sum of f with respect to P is

n
X
U (f, P ) = Mj (xj − xj−1 ).
j=1

The lower integral of f over [a, b] is

L(f ) = sup{L(f, P ) : as P varies over partitions of [a, b]},

and the upper integral of f over [a, b] is

U (f ) = inf{U (f, P ) : P varies over partitions of [a, b]}.
We say that f is integrable over [a, b] when L(f ) = U (f ). We call this common value
the integral of f over [a, b], and we write it as
Z b Z b Z b
f= f (x) dx = f (t) dt.
a a a
318 Appendix B: What you should never forget

The Fundamental theorems of calculus

I: Let f, g : [a, b] → R such that f is continuous and g is differentiable with g ′ = f . Then
Z b
f = g(b) − g(a).
a

II: Let f : [a, b] → R be continuous. Then for all x ∈ [a, b], f is integrable over [a, x], and
Rx
the function g : [a, b] → R given by g(x) = a f is differentiable on (a, x) with
Z x
d
f = f (x).
dx a

Geometric series
P∞ k r
r diverges if |r| ≥ 1 and converges to if |r| < 1.
Pk=1
∞ k
1−r
1
k=0 r diverges if |r| ≥ 1 and converges to 1−r
if |r| < 1.
Pn n+1
For all r ∈ C \ {1}, k=0 r k = r r−1−1 .

Never divide by 0
It bears repeating. Similarly do not plug 0 or negative numbers into ln, do not plug
negative numbers into the square root function, do not ascribe a function (or a person) a
task that makes no sense.
Index

SYMBOLS L(f, P ): lower sum 190

◦ abstract binary operation 73 N 49
C 49, 96 ± 94
n

Qn Yn
k , n choose k 45 , product 36, 75
k=1
Ac complement of A 52 k=1
\ complement of sets 52 Q 49
◦ composition of functions 66 // end of proof 18
cos 23 QED end of proof 18
e (identity element for binary operation) 73 end of proof 18
e: Euler’s constant 215 end of proof 18
{} empty set 49 R 49
∅ empty set 49 Range: range of a function 64
√
∃ there exists 26
n
157, 158
∀ for all 26 sin 23, 25
f n function composed with itself 66 ⊆ subset of a set 50
n
X
Pn
sinn power of trigonometric function (not
k=1 , 34, 75
composition) 66 k=1
sup: supremum 83, 261
> (strictly) greater than 82
∪ union 51
≥ greater than or equal to 82
U (f, P ): upper sum 190
e (identity element for binary operation) 73
Z 49
id (identity function) 64
00 35
⇔ if and only if 17
Image: image of a function 64
A
⇒ implies 16
Abel’s lemma 290
∈ in a set 48
absolute value 20, 93
inf: infimum 83
absolutely convergent 273
∩ intersection 51
advice on writing proofs 311
< (strictly) smaller than 82, 84, 85
alternating test for series 272
≤ less than or equal to 82
and (logical) 15
lim of sequence 230
antecedent 16
lim inf 261
antiderivative 206
lim sup 261
arcsin, arccos, arctan 302
ln: natural logarithm 213
area (signed or usual) 188
¬ logical not 14
associative 74, 79, 80
∨ logical or 15
∧ logical and 15
320 Index

B ceiling function 69
base (of exponential function) 214 chain rule 171
base case (of induction) 37, 40 chart
base of exponentiation 157 how to prove 29, 316
Benjamin, A. 44 negation 32, 315


Bernoulli’s inequality 87 choose: n choose k, nk 45
bijective function 66 closed and bounded—contained in open
composition 67 sets 114, 115
binary operation 73 closed set 111
associative 74 closed-form antiderivative 206
commutative 77 codomain (of a function) 64
binomial (coefficient) 46 commutative 77, 79, 80
Boas, R. P. 305 comparison
Bombelli, R. 104 sequences with infinite limits 239
bounded function 189 sequences 245, 250
bounded set 83 series 271
above 83 completeness of R, C 255
and open balls 113 complex conjugate 100
below 83 complex number 96
Buck, R. C. 306 with e 300
Bumcrot, R. J. 305 composition of functions 67
conditional statement 16
C antecedent, consequent 16
C 49, 96 beware 17
absolute value 101 congruent modulo n 61
arithmetic 96 congruent modulo n 61
Cartesian coordinates 105 conjunction 15
distance 101 connuousti 149
length = norm = absolute value 101 consequent 16
polar coordinates 105 construction of Q from Z 63
properties 97 construction of Z from N0 62
reverse triangle inequality 102 contain 48
triangle inequality 102 continuous 144
cancellation 77 and inverse function 154
Cartesian coordinates 105 and monotonicity 154
Cartesian product 57 connuousti 149
Cauchy’s criterion for series 270 exponential function 160, 161
Cauchy’s mean value theorem 179 exponentiation 158
 Index 321

image of interval 154 element 48

power series 290 empty product 36, 76
properties 145 empty set 49
suounitnoc 148 empty sum 36
ticonnuous 148 enumeration of Q+ 227
uniformly continuous 161, 180 epsilon-delta definition 116, 124
convergent sequence 230 epsilon-N definition 141
convergent equivalence (logical) 17
absolutely 273 equivalence class 60
cos: trigonometric function 23, 298, 299 equivalence relation 60
COS 296 Euclidean algorithm 39
Euclid 33
D Euler’s constant (e) 215, 295
Darboux’s theorem 177 even function, its integral 69
de Moivre’s formula 300, 307 even integer 25
definition: beware 17 definition 18
degree (of polynomial) 35, 67 exclusive or (xor) 15
derivative 164 existential quantifier ∃ 25, 26
and monotonicity 178 exponential function 214
complex-valued functions 172 continuity 161
inverse 173 without derivatives and integrals 160
Leibniz’s notation 164 exponentiation 157, 158
Newton’s notation 164 by real exponents 214
power rule for rational exponents 174 continuity 158
properties 169, 170, 171, 172, 173, 174 exponent 157, 214
differentiable 164 Extreme value theorem 150
hence continuous 168
Dirichlet’s test 276 F
disjunction 15 Ferguson, S. J. 152
distributivity 79 Fibonacci numbers 42, 226
division 80 field 79, 97
do not divide by zero 81 axioms 79
domain (of a function) 64 unusual 82
implicit 66 fixed point theorem 155
floor function 69
E Fourier series 304
e vs. π 217 Fueter, R. 227
Eantegral 197 function 64
322 Index

bijective 66 Goldbach’s conjecture 13

ceiling 69 Greatest lower bound theorem 91
codomain 64
composition 66 H
connuousti 149 halving intervals 113
constant 65 harmonic sequence 256
continuous 144 harmonic series 268
decreasing 89, 90 higher order derivatives 183
domain 64 how-to-prove chart 29, 316
exponential 160, 161, 214
floor 69 I
graph 64 identity element 73
identity 64 additive 79, 80
increasing 89, 90 multiplicative 79, 80
injective 66 identity function 64
monotone 89 if and only if, iff 16
inverse 90 imaginary axis 97
one-to-one 66 imaginary part of a complex number 97, 98
onto 66 implication 16
polynomial 35, 67 improper integral 211, 212
power 90 indefinite integral 206
product of 90 index (of unions, intersections) 52
radical 90, 157 induction 37
range 64 base case 37, 40
rational 67 inductive step 37, 40
suounitnoc 148 infimum (inf, glb) 83
surjective 66 −∞ 261
ticonnuous 148 infinite sequence: see sequence
fundamental theorem of arithmetic 31 infinite series: see series
fundamental theorem of calculus 205, 207 injective function 66
complex-valued function 209 composition 67
integrable 192, 201
G integral test for series 273
Gaitanas, K. 42 integral 192
generalized power function 214 addition 202
generalized power series 282 applications 218
geometric series 268 complex-valued function 208
glb, inf: greatest lower bound 83 double 308
 Index 323

monotonicity 204 infinite 138

notation 192, 195, 206, 208 M-delta definition 138
integration by parts 210 M-N definition 141
integration by substitution 210 of a function 116, 124, 138, 141
intermediate value theorem 153, 177 of sequence 230
interval 48, 85 polynomial rule 133
inverse 74, 90 power rule 133
additive 74, 80 rational rule 134
derivative 173 real/imaginary part 134
multiplicative 74, 80 right/left-handed 124, 138
of a function 71 squeeze theorem 135
of a strictly monotone function 90 subsequence 259
of product 76 theorems for infinite limits 139
unique 75 theorems 131, 133, 134, 135
invertible (element) 75 uniqueness 130
invertible (function) 71 ln, natural logarithm 213
inverse 295
J logic circuits 24
joke 28 logical biconditional 17
logic 13, 25, 29
L lower bound of set 83
L’Hôpital’s rule at infinity 183, 216, 305, lower integral 192
306 lower sum (integrals) 190
L’Hôpital’s rule 180, 181, 217, 305 lub, sup: see least upper bound
Lagrange interpolation 72
Least upper bound theorem 91 M
least upper bound: lub, supremum, sup 83 Matsuoka, Y. 306
Leibniz, G. 104 mean value theorem for integrals 211
length of curve 218 mean value theorem 178
limit inferior, liminf 261 member (of a set) 48
limit point (of set) 111 Meštrović, R. 33
limit superior, limsup 261 modulo n 61
limit modus ponens 19
absolute value 134 monotone function 89
at infinity 141 and derivative 178
composite function theorem 135 monotone sequence 251, 254
epsilon-delta definition 116, 124 multiplication table 74
epsilon-N definition 141
324 Index

N pigeonhole principle 70
N 49 polar coordinates 105
natural logarithm, ln 213 polynomial function 35, 67, 68, 69, 99, 133,
negation 14, 32 184, 244, 249
chart 32, 315 degree 35, 67
negative 85 Taylor polynomial 184
Nelson, E. 148 popcorn function 148
non-negative 85 positive 85
non-positive 85 power function 89, 90, 91, 133, 157, 244,
not: see negation 249
notation generalized 158, 214
integral 206, 208 power notation 35, 66, 76
set vs. sequence 225 00 35
number of elements in Q+ 72 for functions 66, 79
on sets with binary operation 76
O special in trigonometry 66
odd function, its integral 69 power rule for derivatives 170
odd integer 25 power rule
definition 18 generalized 174
one-to-one 66 power series—generalized 282
onto 66 power series 277, 282
open ball 109 continuous 290
open set 109 derivative 284
or (logical) 15 inverse 290
ordered pair 56 numerical evaluation 285
ordered set 85 product of two 287
field 85 radius of convergence 279
order 82 ratio test 280
root test 278
P Taylor series 291
p-test for series 274 uniqueness 284
Papadimitriou, I. 307 prime 13, 30, 31, 33
partial sum (of a sequence) 266, 267 definition 17, 18
partition (of an interval) 190 principle
refinement 191 of mathematical induction 37
Pascal’s triangle 45 pigeonhole 70
Qn
π vs. e 217 product k=1 36, 75
π2
6 306, 308 product of functions 90
 Index 325

product rule for derivatives 169 real axis 97

product rule for higher derivatives 186 real part of a complex number 97, 98
proof 18, 29 recursive 226
by contradiction 21 refinement (of a partition) 191
by induction 37 relation on sets 58
deductive reasoning 18 equivalence 60
how-to-prove chart 30, 316 reflexive 60
pigeonhole principle 70 symmetric 60
√
2 is not rational 21 transitive 60
proper subset 51 restriction of a function 145
Pythagoras 21 reverse triangle inequality 94, 102
Pólya, G. 227 Ritelli, D. 308
Rolle’s theorem 177
Q root of a number 157, 158
Q 49 root test for power series 278
QED (quod erat demonstrandum) 18 root test for series 272
quadratic formula 23, 68
quantifier S
existential 25 Sagher, Y. 72
universal 25 sequence vs. set notation 225
quartering subsets 113 sequence 225
quotient rule for derivatives 169 arithmetic on sequences 228
quotient rule for higher derivatives 186 bounded (below/above) 251
Cauchy 254
R comparison for infinite limits 239
R 49 comparison 245, 250
Archimedean property 87 completeness of R, C 255
Raabe’s test 277 composite rule 244, 249
radical function 90, 157 constant 225
radius of convergence (of power series) 279 convergence 230, 247
range (of a function) 64 properties 242, 244, 248, 249
ratio test divergence to ±∞ 239
for power series 280 divergence 238
for sequences 252, 262 Fibonacci 226
for series 271 finite 229
rational function 99 limit 230
rational numbers between real numbers 87 unique 242, 248
rational number 21 monotone 251, 254
326 Index

of partial sums 266, 267 statement 13

ratio test 252, 262 conditional statement 16
squeeze theorem 246, 250 conjunction 15
subsequence 258 disjunction 15
term 225 equivalence 17
series if and only if 16
alternating test 272 implication 16
arithmetic with 267 logical biconditional 17
comparison 271 negation 14
converges 267 xor 15
diverges 267 subsequence 258
geometric 268 subsequential limit 259
harmonic 268 subset 50
integral test 273 subtraction 80
numerical evaluation 285 sum of sets 88
Pn
p-series test 274 summation k=1 34, 75
power 277, 282 summation by parts 276
ratio test 271, 280 suounitnoc 148
root test 272, 278 supremum (sup, lub) 83
set vs. sequence notation 225 ∞ 261
set 48 surface area of a surface of revolution 221
complement 52 surjective function 66
disjoint 51 composition 67
empty set 49
equality 51 T
intersection 51 tautology 19
member 48 Taylor polynomial 184, 217
notation 48 Taylor series 291
propositionally 49 not approximating 294
subset 50 Taylor’s remainder theorem 185, 186, 211
sum of 88 Thomae function 148
union 51 ticonnuous 148
universal 52 topology 109, 111, 112
signed area 188, 190 closed set 111
sin: trigonometric function 23, 25, 298, 299 limit point 111
SIN 296 open ball 109
socks-and-shoes 78 open set 109
spiral of Theodorus 44 tower of Hanoi 44
 Index 327

triangle inequality 20, 94

over C 102
reverse 94, 102
trichotomy (of <) 84
trigonometry 298
tromino 44
truth table 14, 315

U
uniform continuity 161
uniformly continuity 180
universal quantifier ∀ 25, 26
universal set 52
unusual field 82
upper bound of set 83
upper integral 192
upper sum (integrals) 190

V
vacuously true 27
Velleman, D. 44, 306
Venn diagram 53
volume of a surface of revolution 219, 220
Vorob’ev, N. N. 42

W
well-defined
not 72

X
xor 15

Z
Z 49
zero
do not divide by it 9, 81, 311
to zeroth power 35
zeroth derivative 183