Analysis

Introdu
tion to Analysis
Irena Swanson
Reed College
Spring 2015
Table of contents
Preface
The briefest overview, motivation, notation
Chapter 1: How we will do mathematics
13
Section 1.1: Statements and proof methods
13
Section 1.2: Statements with quantifiers
24
Section 1.3: More proof methods, and negation
27
Section 1.4: Summation
34
Section 1.5: Proofs by (mathematical) induction
36
Section 1.6: Pascals triangle
43
Chapter 2: Concepts with which we will do mathematics
47
Section 2.1: Sets
47
Section 2.2: Cartesian product
55
Section 2.3: Relations, equivalence relations
57
Section 2.4: Functions
62
Section 2.5: Binary operations on sets
71
Section 2.6: Fields
77
Section 2.7: Order on sets, ordered fields, absolute values
79
Section 2.8: Increasing and decreasing functions
86
Chapter 3: Construction of the basic number systems
90
Section 3.1: Inductive sets and a construction of natural numbers
90
Section 3.2: Arithmetic on N0
93
Section 3.3: Order on N0
97
Section 3.4: Cancellation in N0
102
Section 3.5: Construction of Z, arithmetic, and order on Z
103
Section 3.6: Construction of the field Q of rational numbers
109
Section 3.7: Construction of the field R of real numbers
114
Section 3.8: Order on R, the least upper/greatest lower bound theorem
121
Section 3.9: Complex numbers
124
4
Section 3.10: Absolute value in C
128
Section 3.11: Topology on the constructed fields
134
Section 3.12: Closed and bounded sets, and open balls
140
Chapter 4: Limits of functions
143
Section 4.1: Limit of a function
143
Section 4.2: When something is not a limit
152
Section 4.3: Limit theorems
156
Section 4.4: Infinite limits (for real-valued functions)
162
Section 4.5: Limits at infinity
166
Chapter 5: Continuity
169
Section 5.1: Continuous functions
169
Section 5.2: Intermediate Value Theorem, Extreme Value Theorem
173
Section 5.3: Radical functions
179
Section 5.4: Uniform continuity
183
Chapter 6: Differentiation
187
Section 6.1: Definition of derivatives
187
Section 6.2: Basic properties of derivatives
191
Section 6.3: The Mean Value Theorem
198
Section 6.4: LHopitals rule
202
Section 6.5: Higher-order derivatives, Taylor polynomials
205
Chapter 7: Integration
210
Section 7.1: Approximating areas
210
Section 7.2: Computing integrals from upper and lower sums
216
Section 7.3: What functions are integrable?
220
Section 7.4: The Fundamental Theorem of Calculus
226
Section 7.5: Natural logarithm and the exponential functions
232
Section 7.6: Applications of integration
237
Chapter 8: Sequences
244
Section 8.1: Introduction to sequences
244
Section 8.2: Convergence of infinite sequences
248
Section 8.3: Divergence of infinite sequences and infinite limits
254
Section 8.4: Convergence theorems via functions
258
Section 8.5: Convergence theorems via epsilon-N proofs
261
Section 8.6: Bounded sequences, monotone sequences, ratio test
267
Section 8.7: Cauchy sequences, completeness of R, C
270
5
Section 8.8: Subsequences
272
Section 8.9: Liminf, limsup for real-valued sequences
275
Chapter 9: Infinite series and power series
279
Section 9.1: Infinite series

Section 9.2: Convergence and divergence theorems for series
279
283
Section 9.3: Power series
288
Section 9.4: Taylor series
294
Section 9.5: Differentiation of power series
299
Section 9.6: Numerical evaluations of some series
302
Section 9.7: A special function

Section 9.8: A special function, continued
303
305
Section 9.9: Trigonometry
310
Section 9.10: Examples of LHopitals rule
312
Section 9.11: Further exotic uses of trigonometry
313
Appendix A: Advice on writing mathematics

Appendix B: What you should never forget
319
323
Index
327
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, prerequisite
is calculus. Recently used textbooks have been Ray Mayers in-house notes Introduction to
Analysis (2006, available at http://www.reed.edu/~mayer/math112.html/index.html),
and Steven R. Lays Analysis, With an Introduction to Proof (Prentice Hall, Inc., Englewood Cliffs, NJ, 1986, 4th edition).
In Math 112, students learn to write proofs while at the same time learning about binary 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 undergraduates.
Chapter 1 is about how we do mathematics: basic logic, proof methods, and Pascals
triangle for practicing proofs. Chapter 2 introduces foundational concepts: sets, Cartesian
products, relations, functions, binary operations, fields, ordered fields. In Chapters 1 and 2
we assume knowledge of high school mathematics so that we do not practice abstract
concepts and methods in a vacuum. Chapter 3 through Section 3.8 takes a step back:
we forget most previously learned mathematics, and we use the newly learned abstract
tools to construct natural numbers, integers, rational numbers, real numbers, with all the
arithmetic and order. I do not teach these constructions in great detail; my aim is to
give a sense of them and to practice abstract logical thinking. The remaining sections
in Chapter 3 are new material for most students: the field of complex numbers, and
some topology. I cover the last section of Chapter 3 very lightly. Subsequent chapters
cover standard material for introduction to analysis: limits, continuity, differentiation,
P k
integration, sequences, series, ending with the development of the power series k=0 xk! ,
the exponential and the trigonometric functions. 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. In particular, trigonometric functions appear properly only in the last chapter, with occasional light appearances
in the exercises earlier. For this reason, the chapters on differentiation and integration do
not have the usual palette of examples of other books where differentiability and derivatives
of trigonometric functions are assumed.
I acknowledge and thank the support from the Dean of Faculty of Reed College to
fund proofreading help in the summer of 2012 from Maddie Brandt, Munyo Frey-Edwards,
Preface
and Kelsey Houston-Edwards. I further thank Mark Angeles, Andrew Erlanger, Ya Jiang,
Mason Kennedy, Christopher Keane, Michael Keppler, Oleks Lushchyk, Molly Maguire,
Benjamin Morrison, Samuel Olson, Simon Swanson, and Qiaoyu Yang for their feedback.
If you have further comments or corrections, please send them to iswanson@reed.edu.
The briefest overview, motivation, notation

What are the meanings of the following:
5+6
79
84
4/5
48
1/3 = 0.333 . . .
(a b)2 = a2 b2
1 = 3 1/3 = 0.999 . . .
(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 row, 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 and in
red color 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
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 it is not written in flawless grammar, nevertheless, we strive to use correct
grammar and to express the meaning clearly.
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) I am five years old. (This is a statement, and it is false for me.)
(iii) Hello. (This is not a statement.)
(iv) It is raining now. (This is a statement.)
(v) How did you do that? (This is not a statement.)
(vi) 1 = 2. (This is a statement, which in plain English would be One equals two.*)
(vii) For every real number > 0 there exists a real number > 0 such that for all x,
if 0 < |x a| < then |f (x) L| < . (This is a statement.)
(viii) Every even number greater than 4 can be written as a sum of two odd primes.
(This is a statement for which it is currently not known if it is true. This particular
statement is known as the Goldbachs conjecture.)
Sentences such as All functions are expressible as polynomials, The Riemann hypothesis is true, P = NP are statements. We may not agree on or we may not know
their veracity, but we can start with assuming that one is true or false, and then proceed
from there. 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. If all functions were expressible as
polynomials, then we would not have to deal with trigonometric, exponential and logarithmic functions. If the Riemann hypothesis is true, then prime numbers are distributed fairly
* 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
regularly among all integers. If P = NP, then many computational problems would become tractable, spurring advances in logistics, biology, etc. If P 6= NP, then many common
problems would not be solvable efficiently, and efforts should be made instead in partial
and heuristic solutions. 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)
x2
|x|
7
f is a constant function.
F
F
T
x
x>0
x>0
x<0
x<0
y
y
y
y
y
>0
0
>0
0
f is continuous.
T
T
T
xy > 0
T
F
F
F
xy 0
F
T
T
F
f is differentiable.
T
F
T
xy < 0
F
F
T
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 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.
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
T
T
F
F
Q
T
F
T
F
P Q
T
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 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
T
T
F
F
Q
T
F
T
F
P Q
T
T
T
F
P xor Q
F
T
T
F
(4) Implication or a conditional statement is a statement of the form P im-
16

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 assumption 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
T
T
F
F
Q
T
F
T
F
P Q
T
F
T
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 .
17
Unfortunately, the implication statement is not used consistently in informal spoken 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) (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
T
T
F
F
Q
T
F
T
F
P Q
T
F
x
y
QP
T
x
F
y
P Q
T
F
F
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

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
T
T
F
F
Q
T
F
T
F
P
F
F
T
T
P Q
T
F
F
F
P Q
T
T
T
F
P xor Q
F
T
T
F
P Q
T
F
T
T
P Q
T
F
F
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 Q R, (P 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 mathematicians 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 (Quod erat demonstrandum,
which is Latin for that which was to be proved).
The most trivial proofs simply invoke a definition or axiom, such as An even
integer is of the form 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 (P ) is
always true, no matter what the truth value of P is, and this can be easily verified
with the truth table:
P
T
F
P
F
T
P P
T
T
A formula using logical statements that is always true is called a tautology. So

P P is a tautology. Here is another example of tautology: ((P Q) P ) Q,
19
and it is proved below with the truth table:

P
T
T
F
F
Q
T
F
T
F
P Q
T
F
T
T
(P Q) P
T
F
F
F
((P Q) P ) Q
T
T
T
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) (Q)) (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
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
20

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
Now, by drawing, the real numbers x with |x| < y are precisely those real numbers 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 with 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 previous 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 inequality 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 intermediate 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 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 triangle inequality

and since a b implies that a3 b3 )
(4 + 5)3 + 3(4 + 5) (since by assumption |x 5| < 4)

= 93 + 3 9
21
< 103 + 3 10
= 1030.
Here is another pictorial proof, establishing the basis of trigonometry: with the
A
= ab .
picture below, B
{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
= ab .
by bB we get that B
Another common method of proof of a statement P is proof by contradiction:
you assume that P is true and by using only correct logical and mathematical
steps you derive some nonsense (contradiction). This then says that P is false,
so that (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 1.1.11
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
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.
22
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 6 is prime or 7 is odd, then 23 = 7 implies that 7 is prime.
iv) is a very large real number.
v) (P 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 (3 is prime and 5 is even), or it may
be saying ((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 (Q) (P ).
1.1.5 Fill in the following extra columns to the truth table in (5): P Q, (P ) (Q),
(P ) (Q), P Q, (P ) 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 P ) P .
ii) P P .
iii) (P Q) (P Q).
1.1.7 Prove, using truth tables, that the following statements are tautologies:
i) (P Q) [(P Q) (Q P )].
ii) (P Q) (Q P ).
iii) (P Q) [P (P Q)].
iv) [P (Q R)] [(P Q) R].
1.1.8 Suppose that x is any real number such that |x + 2| < 3. Prove that |x3 3x| < 200.
1.1.9 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.
1.1.10 Prove that the sum of two odd integers is an even integer. (An integer is odd
precisely if it can be written as 2m + 1 for some integer m.)
1.1.11 Prove that the product of two integers is odd if and only if the two integers are
both odd.
23
1.1.12 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.
1.1.13 Prove that the product of two consecutive integers is even. Prove that the product
of three consecutive integers is an integer multiple of 6.
1.1.14 Prove that the sum of two consecutive integers is odd. Prove that the sum of three
consecutive integers is an integer multiple of 3.
1.1.15 (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?
24
1.1.16 Is the following correct or false reasoning: If 1 = 2, then seven times each side gives
equality 7 = 14. By subtracting 4 from both sides we get equality 3 = 10. Thus 1 = 2
implies 3 = 10.
1.1.17 Prove that 3 is not a rational number.
1.1.18 For which integers n is n not a rational number?

1.1.19 Assume that P 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.
1.1.20 Assume that (P Q) R is false. Determine, with proof, the truth values of
P, Q, R, or explain if there is not enough information.
1.1.21 Suppose that my shirt is either brown or red, but it is not the case that it is either
red or green. What color is my shirt?
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 statements 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 saying: For every
real number x, x = 1.
Certainly the first statement is true and the second is false.
For shorthand and for symbolic statement computations we abbreviate for all with
the symbol , and there exists with . Here are the same two examples in abbreviated
form:
a real number x such that x = 1.
real number x, x = 1.
Note that after x, the phrase such that is almost automatic when reading but is often
omitted in writing.
The general forms of abbreviated and more abbreviated quantifier statements are:
x with a certain specification, P (x) holds
x with a certain specification, P (x) holds
x P (x)
x P (x)
where P is some property that can be applied to objects in question. The forms on the
right have an implicit understanding of the restrictions on x. We read for all x, P of x
Section 1.2: Statements with quantifiers
25
[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, and can be replaced with
for which, but note that for some specific properties P it can 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.
For ease of readability it may be better to use full words rather than symbolic abbreviations. The reader is invited to read the following symbolic statement (that defines the
limit of the function f at a to be L; see Definition 4.1.1):
> 0 > 0 x, 0 < |x a| < |f (x) L| < .
When are the truth values of statements with quantifiers? A for all statement is
true precisely when all x with the given description have the property P , and a there
exists statement is true precisely when one or more 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, et cetera.
What is the negation of x P (x)? We first write a truth table with all the possible
situations in the first column with regards to P , and other columns giving the truth values
of various quantifier statements:
x P (x)
T vacuously
x P (x)
T vacuously
xP (x)
F
xP (x)
F
there are x of specified type,

P true for all x
there are x of specified type,

P false for all x
there are x of specified type, P

true for some, false for some
a possible situation
there are no x of specified type
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:
26
Proposition 1.2.1 The negation of x P (x) is x P (x). The negation of x P (x)

is x P (x).
For this reason, 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 odd 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.
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 Rewrite the following statements using quantifiers:
i) 7 is prime.
ii) There are infinitely many prime numbers. (Hint: The number of prime numbers
is greater than any real number.)
iii) Everybody loves Raymond.
iv) Spring break is always in March.
v) Sam never sleeps in late.
1.2.3 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.
vi) There exists a real number x such that x3 = 8.
vii) For all real numbers x there exists a positive integer n such that xn > 0.
viii) For all real numbers x and integers n, |x| < xn .
ix) For all integers m there exists an integer n such that m + n is even.

x)
xi)
xii)
xiii)
xiv)
27
There exists an integer m such that for all integers n, m + n is even.

For every integer n, n2 n is even.
Every list of 5 consecutive integers has one element that is a multiple of 5.
The element in the previous part is a multiple of 3.
Every odd number is a multiple of 3.
1.2.4 Explain why the following statements have the same truth values:
i) There exists x such that there exists y such that xy = 1.
ii) There exists y such that there exists x such that xy = 1.
iii) There exists a pair (x, y) such that xy = 1.
1.2.5 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.
1.2.6 In contrast to the switching of quantifiers in the previous exercises, in this one we
show that we cannot arbitrarily switch the order of distinct quantifiers. Namely, 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.
(Another example of this is: y x, y > x versus x y, y > x.)
1.2.7 Rewrite the following statements by using quantifiers: Everybody loves somebody,
Somebody loves everybody, Somebody is loved by everybody, Everybody is loved by
somebody. At least one of your sentences should be of the form x y, xLy. Find its
pair x y, xLy. Compare truth values of the four statements.
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 for all x of its kind, property P holds for x. Under what
conditions is it true that there exists x of the same kind with property P ?
1.2.10 Suppose that it is true that for 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 travelling and they 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.)
28
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 a small chart of this breaking down:
Statement
How to prove it
P (via contradiction).
Suppose P . Establish some nonsense (that

makes P thus impossible).
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, property P holds for x.
Let x be arbitrary of the specified type. Prove

that property P holds for x.
(Possibly break up into a few subcases.)
There exists x of a specified type

such that property P holds for x.
Find/construct an x of the specified type.

Prove that property P holds for x.
Alternatively, invoke a theorem guaranteeing
that such x exists.
An element x of a specified type

with property P is unique.
Suppose that x and x are both of specified 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 ....
29
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.
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) = 4, by the Intermediate Value Theorem (Theorem 5.2.1 in these
notes) [invoking a theorem rather than constructing x, as opposed to in the
previous example]* there exists a real number x strictly between 0 and 2 such that
f (x) = 1.
* Recall that this font in brackets and in red color indicates the reasoning that should go on in the background
in your head; these statements are not part of a proof.
30
Example 1.3.7 (Mixture of methods) For every real number x strictly between 0 and 1
1
there exists a positive real number y such that x1 + y1 = xy
.
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
1
the last equation by the positive number xy we get that x1 + y1 = xy
.
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 = n,
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.
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. Thus x = ab for some whole numbers a, b. [Rewriting the meaning of
assumptions.] If a is negative, since x is positive necessarily b has to be negative. But
then a, b are positive numbers, and x = a
. Thus by possibly replacing a, b with
b
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 among
31
a collection of 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
P and Q
(P Q) = (P ) (Q)
P or Q
(P Q) = (P ) (Q)
P Q
(P Q) = P (Q)
There exists x of the specified type such that

P is false for x.

For all x of the specified type, P is false for x.
Example 1.3.10 (Due to Euclid.) There are infinitely many (positive) prime numbers.
Proof by contradiction: Suppose that there are only finitely many prime numbers. Then
we can enumerate them: 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.
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.
32
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 .

1.3.6 Prove that there exists a real number x such that x2 +
5
12 x
= 16 .
1.3.7 Prove that the following pairs of statements are negations of each other:
i) P Q.
(P Q) (P Q).
ii) P (P Q).
P Q.
iii) (P Q) (R Q).
(P Q) (R Q).
iv) P (Q R).
P (Q 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 functions are continuous not negations of each other?
1.3.10 Why is P 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 > 0, there exists > 0 such that f () = .
iv) For every > 0, there exists > 0 such that for all x, f (x ) = x.
v) For every > 0, there exists > 0 such that for all x, 0 < |x a| < implies
|f (x) L| < .
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.
33
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.
1.3.20 (Cf. Exercise 9.8.3.) 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:
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).
1.3.21 In the spirit of the previous exercise, draw another picture to prove that cos(x+y) =
cos(x) cos(y) sin(x) sin(y).
1.3.22 Assuming that the area of the circle of radius r is r 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
34
1.3.23 (Invoked in Theorem 9.8.3.) Let x be a small positive real number. Consider the
following picture with inside the drawn circle of radius 1:
x
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, prove that 12 sin(x) cos(x) < 21 x < 12 tan x.
iii) Using the previous part, prove that 0 < cos(x) < sinx x < cos1 x .
1.4
Summation
There are many reasons for not writing out 1 + 2 + 3 + + 100 in full length: 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 shorter way is with the summation
sign :
100
100
X
X
n.
k
or
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,
we use the summation sign for shortening as follows:
n
X
k=m
f (k) = f (m) + f (m + 1) + f (m + 2) + + f (n).
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
f (x) = a0 + a1 x + + an x =
n
X
k=0
ak xk .
Section 1.4: Summation
35
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
x0+
equal to 0. But then, is 0 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.)
xc
Examples 1.4.2
5
X
2 = 10.
(1)
(2)
(3)
(4)
(5)
(6)
k=1
5
X
k=1
4
X
k = 15.
k 2 = 12 + 22 + 32 + 42 = 30.
k=1
12
X
k=10
2
X
cos(k) = cos(10) + cos(11) + cos(12) = 1 1 + 1 = 1.

(4k 3 ) = 4 + 0 + 4 + 4 8 = 32.
k=1
n
X
3 = 3n.
k=1
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
ak =
k=1
2
X
k=1
or explicitly written out:
ak +
4
X
k=3
ak =
1
X
k=1
ak +
4
X
k=2
ak =
0
X
ak +
k=1
4
X
ak ,
k=1
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 :
n
Y
k=m
f (k) = f (m) f (m + 1) f (m + 2) f (n).
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 1.4.5 for the fact that 0! = 1.
36

P4
P4
1.4.1 Compute k=0 (2k + 1), k=0 (k 2 + 2).
1.4.2 Prove that

n
n
X
X
i) c
f (k) =
cf (k).
k=m
ii)
iii)
n
X
f (k) +
k=m
n
X
f (k) =
k=m
1.4.3
1.4.4
k=m
n
X
g(k) =
k=m
n
X
f (k)
n
X
(f (k) + g(k)).
k=m
m1
X
f (k).
k=1
k=1
Pn
P
Show that ( k=m f (k)) ( nk=m
P0
.
Prove that k=1 k = 0(1)
2
g(k)) =
Pn
k=m
f (k) g(k) is false (in general).
1.4.5 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.
1.4.6 Prove:
5
Y
2 = 32.
i)
ii)
k=1
5
Y
k = 120.
k=1
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
37
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 .
Example 1.5.1 Prove the equality
Pn
k=1
k=
n(n+1)
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)
which also equals 1. This verifies the base case.
2
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 right side.] Then
!
n
n1
X
X
k=
k +n
k=1
k=1
(n 1)(n 1 + 1)
+ n (by induction assumption)
2
n2 n 2n
=
+
(by algebra)
2
2
n2 + n
=
2
n(n + 1)
,
=
2
as was to be proved.
Pn
for all n 0. Since we have already
We can even prove the equality k=1 k = n(n+1)
2
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)
, which is also 0.
2
38
Example 1.5.2 Prove the equality

for all n 1.
Pn
k=1
k(k + 1)(k + 2)(k + 3) =
n(n+1)(n+2)(n+3)(n+4)
5
Proof. Base case n = 1:

1
X
k=1
k(k + 1)(k + 2)(k + 3) = 1(1 + 1)(1 + 2)(1 + 3) = 1 2 3 4

=
1(1 + 1)(1 + 2)(1 + 3)(1 + 4)

12345
=
,
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 right side.] Then
!
n
n1
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.
=
Example 1.5.3 (This is a basic derivative problem. We introduce derivatives formally in

d
(xn ) = nxn1 . [ Or should induction start
Section 6.1.) Prove that for all n 1, dx
at n = 0? In Section 1.4 we have determined to agree that 00 equals 1, but
was that simply a notational convenience rather than truth?]
Proof. [If n = 0, then x0 = 1 and its derivative is 0 = 0 x01 . At least when
x 6= 0, the conclusion follows, but at x = 0, it looks like we are dividing
by 0. So the equality holds most of the time in this case. What kind of
mathematical statement is most of the time? We should not like this!]
We start the induction at n = 1, as instructed. By calculus we know that the derivative
of x is 1 = 1 x0 = 1 x11 , so equality holds in this case.
1
39
Inductive step: Suppose that equality holds for [0, ]1, 2, . . . , n 1. Then
d
d n
(x ) =
(x xn1 )
dx
dx
d
d
=
(x) xn1 + x (xn1 ) (by the product rule of differentiation)
dx
dx
n1
=x
+ (n 1)x xn2 (by induction assumption for 1 and n 1)
= xn1 + (n 1)xn1
= nxn1 .
Note that if in the previous example we only proved the base case n = 0, the inductive
step there does not prove the case n = 1. The moral of the story is that we have to be
careful about how we use induction (or proofs) in general. See Exercise 1.5.31, for example,
for another case where one has to be careful.
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 + + xn1 + xn ) = (1 x) (1 + x + x2 + + xn1 ) + xn
= (1 x)(1 + x + x2 + + xn1 ) + (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
degree n of f (x) is strictly smaller than the degree m of g(x), then set q(x) = 0 and
40
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
an
2
n1
. These are the base cases.
bn b2 )x + + +(an1 + bn bn1 )x
Now suppose that n > m. Set h(x) = a1 + a2 x + a3 x2 + + an xn1 . By induction
on n, there exist polynomials q1 (x) and r1 (x) 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 a positive integer and that n n < 2. This means
that n < 2n . Hence n + 1 < 2n + 1 < 2n + 2n = 2 2n = 2n+1 , so that n+1 n + 1 < 2.

Exercises for Section 1.5: Prove the following properties for n 1 by induction.
n
X
n(n + 1)(2n + 1)
.
1.5.1
k2 =
6
k=1
2

n
X
n(n + 1)
3
.
1.5.2
k =
2
k=1
1.5.3 The sum of the first n odd integers is n2 .

n
X
1.5.4
(2k 1) = n2 .
k=1
1.5.5 For all real numbers a1 , . . . , an , |a1 + a2 + + an | |a1 | + |a2 | + + |an |.

n
X
(3k 2 k) = n2 (n + 1).
1.5.6
k=1
1.5.7 1 2 + 2 3 + 3 4 + + n(n + 1) = 31 n(n + 1)(n + 2).
1.5.8 Let x be a real number. For any positive integer n define Sn = 1 + x + x2 + + xn .

i) Prove that for any n 2, xSn1 + 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 = 1x
. Compare with the proof by induction in
1x
Example 1.5.4.
1.5.9 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 probably not with induction).

1.5.10 Let An =
1
12
1
23
1
34
++
1
n(n+1) .
41
Find a formula for An and prove it.
1.5.11 There are exactly n people at a gathering. Everybody shakes everybody elses
hands exactly once. How many handshakes are there?
1 xn+1
1.5.12 If x 6= 1, then 1 + x + x2 + + xn =
1x
n
1.5.13 7 + 2 is a multiple of 3.
1.5.14 3n1 < (n + 1)!.
1
1
1
1
1.5.15 + + + + n.
n
1
2
3
1
1
1
1
1
1.5.16 2 + 2 + 2 + + 2 2 .
1
2
3
n
n
1.5.17 Let a1 = 2, and for n 2, an = 3an1 . Formulate and prove a theorem giving an
in terms of n (no dependence on other ai ).
1.5.18 Every integer greater than 2 is a finite product of primes.
1.5.19 8 divides 5n + 2 3n1 + 1.
1.5.20 1(1!) + 2(2!) + 3(3!) + + n(n!) = (n + 1)! 1.

1.5.21 2n1 n!.

n
Y
n+1
1
.
1.5.22 (n 2?)
1 2 =
k
2n
k=2
1.5.23
n
X
2k (k + 1) = 2n+1 n + 1.
k=0
1.5.24 Prove that
P2n
1
k=1 k
n+2
2 .
1.5.25 (This is invoked in Example 9.1.7.) Prove that
P2n 1
k=1
1
k2
Pn
1
k=0 2k .
1.5.26 (This is invoked in the Ratio tests Theorems 8.6.4 and 9.2.4.) Suppose that for
all positive integers n n0 , an+1 < ran . Prove that for all positive integers n n0 ,
an+1 < r nn0 an0 . (The strict inequality < can also be replaced by , >, throughout.)
1.5.27 (This is invoked in the proof of Theorem 9.5.2.) Prove that for all numbers x, y
and all positive integers n, xn y n = (x y)(xn1 + xn2 y + xn3 y 2 + + y n1 =
Pn1
(x y) k=0 xn1k y k .
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 Prove that for all n 2,

n
Y
1
1
= .
1
k
n
k=2
42
1.5.31 (Fibonacci numbers) Let s1 = 1, s2 = 1, and for all n 2, let sn+1 = sn + sn1 .
This sequence starts with 1, 1, 2, 3, 5, 8, 13, 21, 34, . . ..
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 .
n
n
15 12 5 . Note that the base
ii) Prove that for all n 1, sn = 15 1+2 5
case requires proving this for n = 1 and n = 2, and that the inductive step uses
knowing the property for the previous two integers. (It may seem amazing that
these expressions with square roots
of 5 always yield positive integers.)
1+ 5 n2
.
iii) Prove that for all n 3, sn > ( 2 )
2
2
iv) Prove that for all n 1, s1 + s2 + + s2n = sn sn+1 .
v) Prove that s1 + s3 + s5 + + s2n1 = s2n .
vi) Prove that sn+1 sn1 s2n = (1)n .
vii) Prove that s1 s2 + s2 s3 + + s2n1 s2n = s22n .
viii) Prove that s1 s2 + s3 s4 + + s2n1 s2n = 1 s2n1 .
ix) Prove that s2n+1 = s2n+1 + s2n .

n
x) Prove that for all n 1, sn+1 = 1+2 5 sn + 12 5 .
xi) Prove that for all n 1 and all k 2, sn+k = sk sn+1 + sk1 sn .
xii) Prove that for all n, k 1, skn is a multiple of sn . (Use the previous part.)
(xiii)* 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) .

n

1 1
sn+1
sn
xiv) If you know matrices, prove that for all n 2,
.
=
1 0
sn
sn1
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.
43
1.5.33 (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 Sn1 ) for this Sn . Then, make a
guess for a non-recursive formula for Sn (defining Sn based on n without invoking
Sn1 ).
iii) Prove your guess using induction and the recursive formula that you wrote.
*1.5.34 Let a1 , . . . , an+1 be positive integers not exceeding 2n. Prove that some element
on this list divides another one.
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 5n1 + 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? 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 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.
44
1.6
Pascals triangle
The following is rows 0 through 8

for continuation into further rows:
row 0: . . . . . .
.
row 1: . . . . . .
.
row 2: . . . . . .
.
row 3: . . . . . .
1
row 4: . . . . . 1
row 5: . . . . 1
5
row 6: . . . 1
6
row 7: . . 1
7
21
row 8: . 1
8
28
of Pascals triangle, and the pattern is obvious

.
.
1
.
1
1
1
2
3
4
6
10
15
1
4
10
20
35
56
1
3
5
15
35
70
1
1
6
21
56
1
7
28
1
8
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.
Pascals triangle is defined so that for all n 0 and all k = 0, 1, . . . , n 1,

n+1
n
n
.
=
+
k+1
k+1
k

What would it take to compute 100
? It seems like we would need to write down rows
5
0 through 100 of Pascals 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 Pascals triangle in general, including shortcuts for

computing 100
5 . We will accomplish this through exercises, most of which can be proved
by mathematical induction.
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
. In other words, prove that
k+1
k+1 k+2

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

n!
.
1.6.3 Prove that nk = k!(nk)!

1.6.4 Prove that
n
k
45
n(n1)(n2)(nk+1)
.
k!
1.6.5 Prove that for any non-negative integer k,

n
X
j=1
j(j + 1)(j + 2) (j + k) =
n(n + 1)(n + 2) (n + k + 1)
.
k+1
(Hint: induction on n or instead use Exercises 1.6.2 and 1.6.4.)

1.6.6 Use Exercise 1.6.5 to get succinct simplifying formulas for
n
X
j=1
j,
n
X
j=1
j(j + 1),
n
X
j(j + 1)(j + 2),
j=1
n
X
j(j + 1)(j + 2)(j + 3).
j=1
i) Note that j 2 = j(j + 1) j. Use the simplifications from above to prove that
Pn
n(n+1)(2n+1)
2
.
j=1 j =
6
3
ii) From j = j(j + 1)(j + 2) 3j 2 2j = j(j + 1)(j + 2) 3j(j + 1) + j develop the
Pn
formula for j=1 j 3 .
Pn
iii) Mimic the previous work to develop the formula for j=1 j 4 .

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

1.6.8 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.
1.6.9 (This is invoked in Theorem 2.8.2, Example 8.2.7, Theorem 6.2.3.) Prove that for
all non-negative integers n,
n
X
n k nk
n
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.)
1.6.10 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? (You may want to use the previous two
exercises.)

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

large integers n, Cnk nk .
46

1.6.12 Give reasons why we should have nk = 0 for n < k or if either k or n is negative. (It is convenient to have the enlargement of the Pascals triangle where we allow all
combinations of rows and columns, even negatively indexed ones.)
1.6.13 Let d be a positive integer. This is about summing entries in Pascals 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
*1.6.14 Prove that for all non-negative integers n,

!2
!2
n1
n
X 2n
X
2n k
2k
1=2
2
2k + 1
2k
k=0
k=0
and
!2
2n + 1 k
.
2
+1=2
2k + 1
k=0
k=0
k
Pon
Pen n k
n
For notations 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 En1 , On1 . Proceed with induction.)

n
X
2n + 1 k
2
2k
!2
n1
X
2.1
Sets
What is a set? Dont 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.
(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.
We occasionally allow also non-intervals such as (2, 3), which consist of no
members at all.
(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
with the property that x2 > 4x 2.
48

(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 of 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 written 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 3.9, and until then do not
worry when C appears in the text). (Comment: at this point I rely on your basic
training to understand the examples N, N0 , N+ , Z, Q, R, C; in the next chapter
we construct these sets from basic set theory and another axiom.)
(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 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).
(iiii) Q = { ab : a, b Z, b 6= 0}.
Section 2.1: Sets
49
(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} equals {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 3.9).
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.
If A is a subset of B and A is not equal to B (so B contains at least one element that
50

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
\
k=1
n
[
k=1
Ak = A1 A2 An = {a : a Ak for all k = 1, . . . , n},

Ak = A1 A2 An = {a : there exists k = 1, . . . , n such that a Ak },
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},
kI
kI
Ak = {a : there exists k I such that a Ak }.
Section 2.1: Sets
51
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 = .
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.
52
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. From the first part it follows 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),
kN
kN
kN
Ak = [1, ),
kN
Bk = N+ ,
kN
Ck = R.
kN
Example 2.1.6 For each real number r, let Ar = {r}, Br = [0, |r|]. Then
\
\
[
[
Ar = ,
Br = {0},
Ar = R,
Br = [0, ).
rR
rR
rR
rR
Set operations can be represented with a Venn diagram, especially in the presence
of a universal set U . Here is an example:
A
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
Section 2.1: Sets

b
A
b
y
b
t
b
u
b
53
z
b
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
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 SoutheastNorthwest 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
A
U
B
We prove this also algebraically. We have to prove that A(B C) (AB)(AC)

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
54
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.
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).
Section 2.2: Cartesian product
55
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.)
\
\
\
{1/k, 1/k}.
[1/k, 1/k],
(1/k, 1/k),
2.1.9 Compute
kN+
kN+
2.1.10 Compute
(1/k, 1/k),
kN+
2.2
kN+
kN+
[1/k, 1/k],
kN+
{1/k, 1/k}.
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:
56
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) from high school and calculus
classes really stands for {{2}, {2, 3}}, (3, 2) stands for {{3}, {2, 3}}, and (2, 2) stands for
{{2}, {2, 2}} = {{2}, {2}} = {{2}}.
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 ordered pairs.
In general, one can think of A B as the rectangle with A on 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:
57
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 R R.
(The three-dimensional space is R(RR) or (RR)R. In the former case we write
elements in the form (a, (b, c)), 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).)
2.2.1 Let A have m elements and B have n elements. Prove that A B has mn elements.
Prove that A B has 2mn subsets.
2.2.2 How many elements are in R?
2.2.3 Prove that A (B C) = (A B) (A C).
2.2.4 Prove that A (B C) = (A B) (A C).
2.2.5 Prove that (A C) (B D) = (A B) (C D).
2.2.6 Prove that (A C) (B D) 6= (A B) (C D).
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 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 alternately write aRb.
58
Examples 2.3.2
(1) Some relations on R are , <, =, , >. For example, is a subset of the real plane
consisting of all points on or above the line y = x.
(2) Some further relations on R: R = {(a, b) : a, b R, a2 < b + 1}.
(3) 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)},
{}.
Remark 2.3.3 Most relations do not have good names, such as or cousin; think
of relation on A and B simply as a possibly contrived subset of A B.
Definition 2.3.4 Let R be 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.
59
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.
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 = iI 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.
60
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 ab = x n and b c = y n for
some integers x and y. Then ac = a+(b+b)c = (ab)+(bc) = xn+yn = (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. (Normally in the literature
this is written as a b mod n.) We denote the set of all equivalence classes with Z/nZ.
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, . . . , n1 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.)
In everyday life we use congruence modulo 12 (or sometimes 24) for hours, congruence modulo 7 for days of the week, congruence modulo 4 for seasons of the year, congruence modulo 12 for months. In the latter congruence, the equivalence class of 1 is
{1, 13, 25, 37, . . .} {11, 23, 35, . . .}. The equivalence class of 12 is all multiples of 12
(including 0).
There are exactly two equivalence classes for the congruences modulo 2: one consists
of 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.
61
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. In Definition 3.5.2 and beyond we will identify the set of equivalence
classes with the usual integers 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, and furthermore, we will develop the arithmetic on Z from the arithmetic on N0 .
2.3.1 Let A = {a, b} and B = {b, c, d}.
i) How many elements are in the set AA?
ii) How many elements are in the set AB?

iii) How many elements are in the set BB?
iv) How many relations are there on set 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.
62
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.

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 points 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 the relation is perpendicular to is not an equivalence relation.
2.3.10 (Invoked in Section 3.6.) For (a, b), (a, b ) Z Z \ {0}, define (a, b) (a , b ) if
a b = a b.
(1) Prove that is an equivalence relation. (Possibly mimic Example 2.3.11.)
(2) What are the equivalence class of (0, 1), (1, 1), (2, 3)?
(3) Find a natural identification between the equivalence classes and elements of Q.
2.4
63
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 of f is
Range(f ) = {b B : b = f (a) for some a A}.
But in the spirit of introducing few new axioms, lets 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
n1
2 , if n is odd;
f (n) =
n2 , if n is even.
(4) Let f : N+ R be given by the description that f (n) equals the nth prime.
By Euclids theorem (proved on page 31) 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 computers 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.
64

(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
1
2
3
f (x)
1
1
2
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.

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.
65
Definition 2.4.6 Let A, B, B , C be sets with B B . If f : A B and g : B C,

then the composition g composed with f is a function g f : A C 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 R0 , and the domain
of g is R. It is possible to compose g f : R0 R to obtain (g f )(x) = x, but we cannot
compose f and g in the other order. If instead h : R R is defined by h(x) = x2 , then
g f = h f , and f h : R R is the function (f h)(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 : R0 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
any non-negative real number. The function f : R0 R0 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
1
2
f (x)
1
1
x
1
2
f (x)
1
2
x
1
2
f (x)
2
1
x
1
2
f (x)
2
2
66
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
1
2
3
f (x)
1
1
1
x
1
2
3
f (x)
1
1
2
x
1
2
3
f (x)
1
2
1
x
1
2
3
f (x)
1
2
2
x
1
2
3
f (x)
2
1
1
x
1
2
3
f (x)
2
1
2
x
1
2
3
f (x)
2
2
1
x
1
2
3
f (x)
2
2
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. The composition
of two surjective functions is surjective. The composition of two bijective functions is
bijective.
Proof. Let f : A B and g : B C.
Suppose that f and g are injective. 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.
Suppose that f and g are surjective. 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 part follows immediately.
Definition 2.4.11 We have seen polynomials 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 .
Polynomial and rational functions are very special, and are a workhorse of analysis.
(The reader has of course encountered trigonometric, exponential, and logarithmic functions, which are not polynomial or rational.) The following are some special properties:
67
Theorem 2.4.12 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.
(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, is a proof that these four trigonometric functions
*
c1
c2 4c0 c2
1
. There are root
If n = 2, then by the quadratic formula the roots of c0 + c1 x + c2 x are
2c2
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 Niccol`
o Fontana Tartaglia (15001557) and for quartic ones
by Lodovico Ferrari (15221565). Both formulas were popularized in a book by Gerolamo Cardano (15011576).
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 (18021829) and Evariste

Galois (18111832) proved that no formulas exist using
only radicals, sums, products, quotients.
68
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.)
2.4.1 Fix a positive integer n. Define f : Z {0, 1, 2, 3, . . . , n1} 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,
= 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, = 4, 1 = 1,
1.5 = 1. Graph this function.
2.4.4 Find a set A and functions f, g : A A such that f g 6= g f .
2.4.5 Is the composition of functions associative? That is, does f (g h) = (f g) h?
Prove or give a counterexample.
2.4.6 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?

2.4.7 Let A = {1, 2, 3}, B = {2, 4, 6}, 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, 3), (3, 2)}.
iii) h = {(1, 6), (2, 2)}.
2.4.8 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, 3), (4, 2)}.
ii) {(9, 6), (3, 2)}.
iii) {(2, 6)}.
69
2.4.9 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?
2.4.10 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.
2.4.11 Let A be a set A 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?
2.4.12 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?
2.4.13 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.
70
2.4.14 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.
2.4.15 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.
2.4.16 Prove that if f and g are both injective functions, then f g is also injective.
2.4.17 Prove that if f and g are both surjective functions, then f g is also surjective.
2.4.18 Prove that if f and g are both bijective functions, then f g is also bijective.
2.4.19 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)?
2.4.20 For the following polynomial and rational functions, determine the domains (largest
sets on which the function is defined):
1
x3 x .
x4 2x3 1
.
x6 +2
1
x5 17x2 +x4 .
i) f (x) = x2
ii) f (x) =
(iii)* f (x) =
(Gottcha! We do not know the roots of this denominator.)
2.4.21 Prove that f : R R given by f (x) = x3 1 is surjective and injective.

2.4.22 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 g.
2.4.23 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.
2.4.24 Let f : R R be given by f (x) = x3 5. Prove that f is invertible and find its
inverse.
2.4.25 (Lagrange interpolation) Let c1 , . . . , cn be distinct real (or complex) numbers.

(xc1 )(xc2 )(xcj1 )(xcj+1 )(xcn )
Set gj (x) = (cj c1 )(c
.
j c2 )(cj cj1 )(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 .
71
2.4.26 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).
(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, i.e., if [f ] = [g], then F ([f ]) = F ([g]).
(iii)* Prove that F is bijective.
2.4.27 Let f : N+ Z and g : N+ Q+ be defined by
n1
2 , if n is odd;
f (n) =
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 245.)
2.5
Binary operations on sets
Definition 2.5.1 A binary operation on a set A is a function : A A A. Binary

operations are usually denoted by special symbols such as:
+, , , /, , , , , or, and.
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).
(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) on Z, Q, R;
(3) / on Q \ {0}, R \ {0};
(4) is not a binary operation on N.
(5) Let S be a set, and let A be the set of all functions f : S S. Then composition
(of functions) is a binary operation on A.
(6) , , \ are binary operations on the set of all subsets of a set S.
(7) and is a binary operation on the set of all logical statements.
72
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 theorem below resolves the problem of an identity versus the identity.
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).
(1) Let = + on Z. Then 0 is the identity element and every element has an additive
inverse.
(2) Let = on Q \ {0}. Then 1 is the identity element and every element has a
multiplicative inverse.
(3) 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 .
(4) Here is a new binary operation on the set S = {a, b, c, d} presented via its
73
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(bc) =
(a b) c.
(1) + and are associative.
(2) , / are not associative.
(3) Composition of functions is an associative operation.
(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 x1 .
Be careful! What is the number 51 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!
74
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.
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 = an1 a = a an1 .
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 1.4.5 for the first occurrence of an empty product.)
If a has a multiplicative inverse, then a1 is that inverse, and in that case if is in
also associative,
an = a(n1) a1 .
To prove this by induction on n, we multiply
(a(n1) a1 ) an = (a(n1) a1 ) (a an1 )
= a(n1) (a1 a) an1
= a(n1) an1 = e,
and similarly an (a(n1) a1 ) = e, which proves that an = a(n1) a1 .

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
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 x1 .
e, and similarly
75
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, x1 x = e = x x1 , which also reads as the inverse
of x1 is x.
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=xe
= x (z z 1 )
= (x z) z 1 (by associativity)
= (y z) z 1
= y (z z 1 ) (by associativity)
=ye
= y.
Definition 2.5.17 A binary operation on A is commutative if for all a, b A, a b =

b a.
(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 manipulations below when n = 12 or n = 24 for hours of the day (we do not say 28 oclock),
when n = 7 for days of the week, when n = 3 for standard meals of the day, et cetera.
76
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
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] = [ab]. 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
0 1
0 0
0 1
Z/3Z:
0
1
2
0
0
0
0
1
0
1
2
2
0
2
1
Z/4Z:
0
1
2
3
0
0
0
0
0
1
0
1
2
3
2
0
2
0
2
3
0
3
2
1
Note that for multiplication in Z/4Z, [3] = [1] is the inverse of itself, and [2] has no
inverse.
2.5.1 Let A be the set of all logical statements.
i) Verify that , and xor are binary operations on A.
ii) Find the identity elements, if they exist, for each of the three binary operations.
iii) For each binary operation above with identity, which elements of A have inverses?
2.5.2 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.
Section 2.6: Fields
77
2.5.3 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 .
2.5.4 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?)
2.5.5 Refer to Example 2.5.19: Write addition tables for Z/2Z, Z/3Z, Z/4Z.
2.5.6 Write multiplication tables for Z/5Z, Z/6Z, Z/7Z.
2.5.7 Determine all [a] Z/12Z that have a multiplicative inverse.
*2.5.8 Determine all [a] Z/pZ that have a multiplicative inverse if p is a prime number.
2.5.9 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.)
2.6
Fields
Definition 2.6.1 A non-empty 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).
(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.
Note that 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
78
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 will take it for granted (until Chapter 3) that Q and R are fields. In Section 3.9
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.
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 m1 , or also 1/m. The sum n + (m) of n and m is
also written as n m, and the product n m1 of n and m1 is also written as n/m. The
latter two operations are also called subtraction and division.
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 2.6.2 (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).
Theorem 2.6.3 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.

Note that 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
79
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 2.6.3, n = 0. So, if we were to divide by
0, we could at most divide 0 by 0. Furthermore continuing the horrible detour then
0/0 = (0 0)/0 = 0 (0/0) = 0 by (2), but 0/0 should perhaps also be 1? In any case,
division by 0 gets us into a muddle. In a mathematics paper, simply never write x0 or
x/0.
At this stage of your mathematical life, you will of course never write something like
3/0 (oh, my eyes hurt seeing this!), but a common college mistake that is essentially
division by 0 is when one is solving an equation such as x2 = 3x, cancels x, and obtains
the solution x = 3. However, that cancellation was division by 0 when x was the other
possible solution! Avoid this hidden division by 0, so that you can find all the solutions.
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.
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.
80
2.7
Order on sets, ordered fields, absolute values

If < is a relation on a set S, we can define relations , >, on S by
a b means that 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 , . 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 strictly younger than on the set of all people, in which
case is is younger than or of equal age, and >, have the obvious meanings.
(2) < can be the relation is a proper subset of on a set S of all subsets of some universal set U . In this case, means is a subset of, > means properly contains,
and means contains.
Definition 2.7.2 Let be a relation on a set S. A subset T of S is bounded above

(resp. bounded below)* (in S) if there exists b S such that for all t T , t b (resp.
b t). The element b is called an upper bound (resp. lower bound) of T (in S).
A set that is bounded above and below is called bounded.
An upper bound c S of T is a least upper bound, or supremum, if for all upper
bounds b of T , c b. If c T , then c is also called the maximum of T .
A lower bound c S of T is a greatest lower bound, or infimum, 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 .
* A sentence of the form P is Q (resp. Q ) if R (resp. R ) is a shorthand for two sentences: P is Q if R

and P is Q if R .
81
Examples 2.7.3
(1) The set N0 in R has minimum 0. It is not bounded above.
(2) The set {1/n : n N+ } has maximum 1, it is bounded below, the infimum is 0,
and there is no minimum.
(3) The set {sin(x) : x R} has maximum 1 and minimum 1.
(4) The set {ex : x R} has no upper bound, it is bounded below with infimum 0,
but there is no minimum.
(5) The empty subset obviously has no minima nor maxima. Every element of S is
vacuously an upper and a lower bound of the empty set.
(6) The set {x R : |x 5| < 3} has no minimum and maximum, but the infimum
is 2 and the supremum is 8. This is easy to see once we rewrite
{x R : |x 5| < 3} = {x R : 3 < x 5 < 3} = {x R : 2 < x < 8} = (2, 8).
In the sequel we restrict < to relations that satisfy the trichotomy property:
Definition 2.7.4 A relation < on a set S satisfies trichotomy if for all s, t S, exactly
one of the following relations holds:
s = t,
s < t,
t < s.
Theorem 2.7.5 Let < on S satisfy 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 2.7.6 A set S with relation is well-ordered if inf (T ) = min (T ) for every
non-empty subset T of S that is bounded below. The element min (T ) is called the least
element of T .
82
Examples 2.7.7
(1) Any finite set with relation is well-ordered (simply check the finitely many
pairings for which element is smaller).
(2) Z is not well-ordered as there is no smallest whole number.
(3) The set of all positive rational numbers is not well-ordered as it itself does not
have a minimum.
(4) N0 is well-ordered because for any non-empty subset T of N0 , by the fact that the
set is not empty there exists an element n T , and after that one has to check if
any of the finitely many numbers 0, 1 through n is the smallest one in T .
(5) Similarly, N+ is well-ordered.
Definition 2.7.8 Let F be a set with binary operations +, , with additive identity 0 F
and multiplicative identity 1 F , and with a relation < satisfying trichotomy (recall
Definition 2.7.4). 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}.
We say that F is ordered if in addition the following properties hold for all x, y, z F :
(1) (Transitivity of <) If x < y and y < z then x < z.
(2) (Compatibility of < with addition) If x < y then x + z < y + z.

(3) (Compatibility of < with multiplication by positive elements) If x < y
and 0 < z then xz < yz.
If F is a field with these properties, it is called an ordered field.
83
Theorem 2.7.9 Let F be an ordered set as in Definition 2.7.8.

(1) For x F , x F + if and only if x F , and x F if and only if x F + .
(2) 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) We know that 1 6= 0. If 1 6 F + , then by trichotomy 1 < 0, and by above, 0 < 1.
Thus by compatibility of < with multiplication by positive numbers, 0 = 0 (1) <
(1) (1) = (1) = 1, which says that 1 < 0 and 0 < 1, contradicting the trichotomy.
Thus 1 F + .
Definition 2.7.10 Let F be an ordered set.
a function defined as
(
0;
|x| = x;
x;
The absolute value function | | : F F is

if x = 0;
if x F + ;
if x F .
The following theorem lists the familiar properties of absolute values, and the reader
may wish to prove it without reading the given proof.
Theorem 2.7.11 Let F be an ordered set.
(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) If x, y F + , then x + y, xy F + .
(7) 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|.
(3) If x 0, then |x| = x, and if x < 0, then x < 0 < |x|. Thus by transitivity for
all x, x |x|. Hence 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).
84
(6) By assumption, 0 < x and 0 < y. Thus by compatibility of multiplication by

positive elements, 0 = 0 y < x y = xy, so that xy F + . By compatibility of addition,
0 < x = x + 0 < x + y, so that by transitivity, x + y F + .
(7) We may choose and {1, 1} such that x 0 and y 0. Then
|xy| = |(x)(y)| (by Exercises 2.6.8 and 2.6.9 and (1))
= (x)(y) (by (6))
= |x||y| (by definition of , ).

In the next theorem we use the standard notation when we mean that the result
should hold in place either for + or for .
Theorem 2.7.12 (Triangle and reverse triangle inequalities) Let F be an ordered
set. Then the following inequalities hold.
(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.
Theorem 2.7.13 Let F be an ordered field. Let r F .
(1) If r F + , then r 1 F + .
(2) If r < for all F + , then r 0.
(3) If r > for all F + , then r 0.
(4) If |r| < for all F + , then r = 0.
85
Proof. Suppose that r F + . Then r 6= 0, so that r 1 F . If r 1 = 0, then 1 = r 1 r =

0 r = 0, which contradicts the axioms of a field. If r 1 F , then by Theorem 2.7.9,
r 1 F + , so that then 1 = r 1 r F + by compatibility of < with multiplication by
positive numbers, which is impossible by Theorem 2.7.9. Thus r 1 F + . This proves (1).
Proof of (2): By Theorem 2.7.9, 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 (1),
21 F + . Since r > 0, by compatibility of < with multiplication by positive numbers,
= r/2 is a positive number. By assumption, r < = 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 (3) is similar.
For (4), if |r| < for all F + , then < r < . Then by (2) and (3), 0 r 0.
Since F + F = by trichotomy, it follows that r = 0.
2.7.1 Prove that if < is transitive then is transitive.
2.7.2 Prove that < (resp. ) is transitive if and only if > (resp. ) is transitive.
2.7.3 In computing limits (see Section 4.1), especially for finding , one in general needs
to work with minima of finite sets. For example, under what conditions is the minimum of {0.2, /7} equal to 0.2, and when is the minimum /7? Similarly determine
min {0.2, /7, 2/4}.
2.7.4 Does the set {2n /n : n N+ } have a lower (resp. upper bound)? Is it well-ordered?
Justify. Repeat with {n/2n : n N+ }.
2.7.5 For each of the sets below, determine its minimum, maximum, infimum, supremum,
if applicable. Justify all answers.
i) {1, 2, , 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 R : x2 < 2}.
vii) {x Q : x2 < 2}.
viii) {x R : x2 + x 1 = 0}.
ix) {x Q : x2 + x 1 = 0}.
86
2.7.6 Suppose that a subset T of an ordered field has a minimum (resp., maximum, infimum, supremum) b. Prove that the set T = {t : t T } has a maximum (resp.,
minimum, supremum, infimum) b.
2.7.7 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 }?
2.7.8 (Invoked in Theorem 7.3.6.) 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 .
2.7.9 Let F be an ordered field and let x be non-zero element of F . Prove that x F + if
and only if x1 F + .
2.7.10 Let F be an ordered field. Prove that 2, 3 are positive (and so not zero).
2.7.11 Let F be an ordered set and x F . Prove that x2 = 0 if and only if x = 0.
2.7.12 Let F be an ordered set and x F . Let x, y F be non-negative (resp. nonpositive) such that x + y = 0. Prove that x = y = 0.
2.7.13 Prove that Z/pZ is not an ordered field for any prime integer p. (Cf. Exercise 2.6.5.)
2.7.14 Let F be an ordered set. 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.
2.7.15 Let F be an ordered set 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.
2.7.16 Let F be an ordered field.
i) Prove that if 0 < x < y, then 1/y < 1/x.
ii) Suppose that x < y and that x, y are non-zero. Does it follow that 1/y < 1/x?
2.7.17 (In-betweenness in an ordered set) Let F be an ordered set. Let x, y F with
x < y. Prove that if 2 is a unit in F , then x < (x + y)/2 < y.
2.7.18 Find an ordered set F with a subset S without a minimum.
2.7.19 Let F be an ordered set. Prove that any non-empty finite subset S of F has a
maximum and a minimum (cf. Definition 2.7.2). These are denoted max (S) and min (S),
respectively.
2.7.20 Let F be an ordered set and S a finite subset S. Prove that for all s S, min (S)
s max (S).
2.8
87
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 set. 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 1.6.9,
n
f (y) f (x) = y x = (x + (y x)) x =

n
k
n1
X
k=0

n k
x (y x)nk .
k
But x, y x are positive, and

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)n0 = (y x)n , which is positive, so that
f (y) f (x) > 0.
Corollary 2.8.3 Let n be a positive integer and F an ordered set. 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 increasing (resp. decreasing), then there exists a strictly increasing (resp. decreasing) 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 . Since f is injective, 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.
88
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 increasing 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 transitivity 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 decreasing. The proof in the case where both f and g are strictly decreasing is similar.
2.8.1 Let n be an odd positive integer and F an ordered set. 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 set. Prove that the function
f : F F defined by f (x) = xn is strictly decreasing on F {0}.
2.8.3 Let n be an odd positive integer and F an ordered set. Suppose that a, b F and
that an < bn . Prove that a < b.
2.8.4 Let F be an ordered set, 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 set, a F and f : F F defined by f (x) = ax. Prove that
f is a strictly decreasing function.
2.8.6 Let F be an ordered set. Prove that the absolute value function on F is increasing
on F + {0} and decreasing on F {0}.
2.8.7 Prove that the composition of (strictly) increasing functions is (strictly) increasing.
Prove that the composition of (strictly) decreasing functions is (strictly) increasing.
2.8.8 Prove that the composition of a (strictly) increasing function with a (strictly) decreasing function, in any order, is (strictly) decreasing.
89
2.8.9 Suppose that f : F G is strictly monotone.

i) Prove that f is injective.
ii) Let B be a subset of F , and define g : B G by g(x) = f (x). Prove that f is
strictly monotone.
iii) Define h : B Range(g) by h(x) = f (x). Prove that h is bijective.
2.8.10 Give an example of a non-decreasing function f : R R that is not injective.
In Chapter 1 we laid the basic groundwork for how we do mathematics: how we reason
logically, how we prove further facts from established truths, and we learned some notation.
In Chapter 2 we introduced sets, and from those derived functions and binary operations
that play a big role in mathematics. In this chapter, we use set theory to derive numbers
and basic arithmetic: we will do so on N0 , Z, Q, R, C, in this order.
When I teach this course, I go lightly through the first eight sections of this very
long chapter, but I do spend time on Theorems 3.8.5, 3.8.3, 3.8.4. Dedekind cuts give
straightforward proofs, but if one wishes to avoid Dedekind cuts, then these theorems can
(and should) be simply accepted as facts. Section 3.9 plays a central role in the rest of
the book, so this section and Section 3.10 should be studied thoroughly. Sections 3.11 and
3.12 introduce topology, and they contain more information than strictly necessary for the
rest of the book.
3.1
Inductive sets and a construction of natural numbers
We start with the most basic set: . We can make the empty set be the unique element
of a new set, like so: {}. The sets and {} are distinct because the latter contains the
empty set and the former contains no elements. With these two distinct sets we can form
a new set with precisely these two elements: {, {}}. This set certainly differs from the
empty set, and it also differs from {} because {} is not an element of {} (but it is a
subset think about this). So now we have three distinct sets, and we can form another
set from these: {, {}, {, {}}}, giving us distinct sets
,
{},
{, {}},
{, {}, {, {}}}.
We could call these Zeno, Juan, Drew, Tricia, but more familiarly these sets can be called
zero, one, two, three, and written 0, 1, 2, 3 for short. At this point, we are simply giving
them names, with no assumption on any arithmetic properties. We could say and so on,
but that is not very rigorous, is it.
Section 3.1: Inductive sets and a construction of natural numbers
91
Definition 3.1.1 For any set S, its successor S + is defined as S + = S {S}.

By our naming convention, 0 is not the successor of any set, 0+ = 1, 1+ = 2, 2+ = 3.
We can also write 0++ = (0+ )+ = 1+ = 2 and 0+++ = ((0+ )+ )+ = (1+ )+ = 2+ = 3.
At this point I rely on your basic training for the naming conventions of 3+ = 4,
3++ = 5, . . ..
The definition of successors of successors of successors... is recursive. We can see
informally that the set of all successive successors of 0 will be infinite, in fact we hope that
this set would be the familiar N0 , but to get to that we would have to take a union of
infinitely many sets that at this point need not be elements of some universal set. Taking
unions of infinitely many sets is subject to pitfalls, just like infinite sums have pitfalls.
Definition 3.1.2 A set J is called inductive if it satisfies two conditions:
(1) J .
(2) For any n J , the successor n+ of n is also in J .
We accept as a given (we take it as an axiom) that there exists a set that contains
and all its successors:
Axiom 3.1.3 There exists an inductive set.
An inductive set has to contain the familiar 0, 1, 2, 3, . . ., but possibly it can also contain
the familiar numbers 1.3 or , which are not part of the familiar (inductive set) N0 .
Theorem 3.1.4 Let S be a set that contains inductive subsets. The intersection of any
collection of inductive subsets is inductive.
Proof. Each inductive set contains , hence their intersection contains as well; for any
n in the intersection, that n is in all the inductive sets, hence n+ is also in those same
inductive sets, which says that n+ is in the intersection, thus proving that the intersection
of inductive sets is inductive.
Definition 3.1.5 (Definition of N0 ) Let J be any inductive set. Define N0 as the intersection of all inductive subsets of J . Elements of N0 are called natural numbers. (Beware:
in some books, this set is written as N, whereas in some other books N stands for all
non-zero natural numbers. In these notes, we always write N0 for clarity.)
92
Remark 3.1.6 By Theorem 3.1.4, N0 is an inductive set, and by definition it is a subset

of every inductive subset of J . It appears that N0 depends on the choice of J . However,
Exercise 3.1.4 shows that this is not so.
Some of the elements of N0 are: , {}, {, {}}, {, {}, {, {}}}, or written alternatively: 0, 1, 2, 3. In fact, we prefer to write elements of N0 numerically, but for proofs we
will often need sets to resort to the notion of successors of sets.
Theorem 3.1.7 0 is not the successor of any element of N0 , and every element of N0 other
than 0 is the successor of some element of N0 .
Proof. 0 stands for the empty set, which by definition cannot be the successor of any set.
Let n N0 \ {0}. If n is not the successor of any element in N0 , then N0 \ {n} is an
inductive set which is strictly smaller than N0 . But this contradicts the definition of N0 .
Thus there is no such n, which proves that every element of N0 \ {0} is the successor of
some element of N0 .
At least this says that if were to be in N0 , then 1 perhaps would have to be in
N0 as well?
Recall mathematical induction from Section 1.5. The notion of inductive sets gives
us a new perspective at induction and it will enable us to eventually rigorously define the
(usual) arithmetic. We will use the following new look at induction over and over:
Theorem 3.1.8 (Induction Theorem) Let P be a property applicable to (some) elements of N0 . Suppose that
(1) P (0) is true;
(2) For all n N0 , (P (n) = P (n+ )) is true.
Then P (n) is true for all n N0 .
Proof. [To use the new concept of inductive sets, we have to define a set
that is hopefully inductive.] Let T = {n N0 : P (n) is true}. [We will prove
that this T is inductive.] By assumption (1) we have that 0 T . [So the first
property of inductive sets holds for T .] If n T , then P (n) is true, and since
(P (n) = P (n+ )) is true, it follows that P (n+ ) is true. Thus n+ T . This means that
T is an inductive set. But N0 is the smallest inductive set, which means that N0 T , so
that P (n) is true for all n N0 .
We illustrate below how the inductive theorem can be used. Many more examples are
in the next three sections.
93
Theorem 3.1.9 For all n N0 , 0 n+ .

Proof. Set T = {n N0 : 0 n+ }. Since 0+ = {} contains 0 = , we have that 0 T .
Now assume n T . This means that 0 n n {n} = n+ , so that 0 n+ . Thus n+ T .
This proves that T is an inductive set, and it is a subset of N0 . Since N0 is the smallest
inductive set and contained in all others, it follows that N0 T . Hence T = N0 , which
means that for all n N0 , 0 n.
In more familiar language, this is saying that 0 is an element of 1, 2, 3, et cetera.

3.1.1 Prove that for every set S, S S + and S S + .
3.1.2 Let T = {, {}, {{}}}. Prove that T is not the successor of any set S.
3.1.3 Prove that for all n N0 , if n 6= 0 and n 6= 1, then 1 n. (Hint: set T = {n N0 :

1 (n+ )+ }.)
*3.1.4 (The goal of this exercise is to prove that N0 as defined in Definition 3.1.5 is
independent of the ambient inductive set.) Let J and K be inductive sets. Let N be the
intersection of all inductive subsets of J , and let M be the intersection of all inductive
subsets of K. Let f : N M be a function defined by f (0) = 0 and for any n N ,
f (n+ ) = (f (n))+.
i) Verify that f is indeed a function with domain N . (Hint: Let T be the subset of
all n N at which f is defined. Prove that T is inductive and hence equal to N .)
ii) Prove that the image of f is an inductive subset of M .
iii) Prove that the image of f is M .
iv) Prove that via this natural function f that takes 0 to 0, 1 to 1, et cetera, the
elements of N can be identified with the corresponding elements of M .
3.2
Arithmetic on N0
We can apply the Induction Theorem (Theorem 3.1.8) to define addition and multiplication on N0 . For example, for any m N0 we define the adding m function
Am : N0 N0 as follows:
Am (0) = m; Am (n+ ) = Am (n)+ for n N0 .
The induction theorem says that Am (n) is defined for all m and n: Am (0) is in N0 , and if
Am (n) N0 , then Am (n)+ is in N0 , hence Am (n+ ) N0 , so that Am is a function from N0
to N0 . Similarly, we define multiplication by m as a function Mm : N0 N0 given by
Mm (0) = 0;
Mm (n+ ) = Am (Mm (n)).
94
With this functions we define binary operations + and on N0 :

m + n = Am (n),
m n = Mm (n).
Example 3.2.1 We will prove that 2 + 2 = 4, and you can verify that 2 2 = 4.
2 + 2 = A2 (2)
= A2 (1+ )
= A2 (1)+
= A2 (0+ )+
= (A2 (0)+ )+
= (2+ )+
= 3+
= 4.
Remark 3.2.2 For all m, n N0 ,
m + 1 = m+ ,
m + (n+ ) = (m + n)+
and
m (n+ ) = m + (m n).
The last equality is simply rewriting the meanings of , Mm and Am , the second equality
follows by the definitions with m + n+ = Am (n+ ) = (Am (n))+ = (m + n)+ , and when
n = 0, this produces m + 1 = m + 0+ = (m + 0)+ = (Am (0))+ = m+ .
Theorem 3.2.3 The following identities hold for all m, n, p N0 :
(1) (Additive identity) m + 0 = m = 0 + m.
(2) m 0 = 0 = 0 m.
(5) (Multiplicative identity) m 1 = m = 1 m.
(Distributivity) (n + p) m = (n m) + (p m).
Proof. (1) Let T = {m N0 : 0 + m = m}. By definitions of + and Am , m + 0 = Am (0) =
m. In particular, 0 T . If m T , then
0 + (m+ ) = (0 + m)+ (by Remark 3.2.2)
= m+ (as m T ),
so that m+ T . It follows that T is an inductive subset of N0 , so that as N0 is the smallest
inductive set, N0 = T .
95
(2) By definition of , m 0 = Mm (0) = 0. Let T = {m N0 : 0 m = 0}. By m 0 = 0,

0 T . If m T , then
0 (m+ ) = 0 + (0 m) (by Definition of )
= 0 + 0 (as m T )
= 0 (by (1)),
so that m+ T . It follows that T is an inductive set, so that as N0 is the smallest inductive

set, N0 = T .
(3) Fix m, n N0 , and let T = {p N0 : m + (n + p) = (m + n) + p}. Since
m + (n + 0) = m + n = (m + n) + 0, it follows that 0 T . If p T , then
m + (n + (p+ )) = m + ((n + p)+ ) (by Remark 3.2.2)
= (m + (n + p))+ (by Remark 3.2.2)
= ((m + n) + p)+ (as p T )
= (m + n) + (p+ ) (by Remark 3.2.2),

so that p+ T . So T is an inductive set, and as N0 is the smallest inductive set, necessarily
N0 T , so (3) holds for all m, n, p N0 .
(4) You prove this. First prove that n + 1 = 1 + n for all n N0 .
(5) By definition of , m 1 = m (0+ ) = m + m 0. In (2) we already proved that

m 0 = 0, and in (1) we proved that m + 0 = m. Thus m 1 = m 0 + m = 0 + m = m.
Let T = {m N0 : 1 m = m}. By (2), 0 T . If m T , then
1 (m+ ) = 1 + (1 m) (by Definition of )
= 1 + m (as m T ),
= m + 1 (by (4)),
= m+ (by Remark 3.2.2),

so that m+ T . So T is an inductive set, and as N0 is the smallest inductive set, necessarily
N0 = T , so (5) holds for all m N0 .
(6) Fix m, n N0 . Let T = {p N0 : m (n + p) = (m n) + (m p)}. By (2), m 0 = 0,

and by (1), (m n) + 0 = m n, so that 0 T . If p T , then
m (n + (p+ )) = m ((n + p)+ ) (by Remark 3.2.2)
= m + (m (n + p)) (by Definition of )
= m + ((m n) + (m p)) (as p T )
= (m + (m n)) + (m p) (by (3))
= ((m n) + m) + (m p) (by (4))
96

= (m n) + (m + (m p)) (by (3))
= (m n) + (m (p+ )) (by Remark 3.2.2),

so that p+ T . So T is an inductive set, and as N0 is the smallest inductive set, necessarily
N0 T , so the first part of (6) holds for all m, n, p N0 .
Fix m, n N0 . Let T = {p N0 : (m + n) p = (m p) + (n p)}. By (2) and (1),
(m + n) 0 = 0 + 0 = (m 0) + (n 0), so that 0 T . If p T , then
(m + n) (p+ ) = Mm+n (p+ ) (by Definition of )
= (m + n) + ((m + n) p) (by Definition of Mm+n )
= (m + n) + ((m p) + (n p)) (as p T )
= ((m + n) + (m p)) + (n p) (by (3))
= ((n + m) + (m p)) + (n p) (by (4))
= (n + (m + (m p))) + (n p) (by (3))
= (n + Mm (p+ )) + (n p) (by Definition of Mm )

= (Mm (p+ ) + n) + (n p) (by (4))
= Mm (p+ ) + (n + (n p)) (by (3))
= Mm (p+ ) + (n + Mn (p+ ) (by Definition of Mn )

= (m p+ ) + (n p+ ) (by Definition of ),
so that p+ T , and the second part of (6) holds for all m, n, p N0 .

(7) You prove (7).
(8) Fix m N0 . Let T = {n N0 : n m = m n for all m N0 }. By (2), 0 T . If
n T , then
m (n+ ) = m + m n
(by Definition of )
= m + n m (since n T )
= 1 m + n m (by (5))
= (1 + n) m (by (6))
= (n + 1) m (by (4))
= n+ m,
so that n+ T . Thus T is an inductive set, so T = N0 , which proves (8).

This gives us the familiar N0 with the familiar arithmetic properties. In particular,
N0 has a binary operation + with identity 0 and a binary operation with identity 1. We
already proved abstractly in Theorem 2.5.5 that such identities are unique.
97

3.2.1 Verify with the new definitions that 3 2 = 6 and that 2 3 = 6. Was one easier?
3.2.2 Prove (4) and (7) in Theorem 3.2.3.
3.2.3 Prove that for all m, n N0 , m+ + n = n+ + m.
3.2.4 Prove that for all m, n, p N0 , (m+ ) (p + n) = m (p + n) + (p + n).
3.3
Order on N0
Definition 3.3.1 For a, b N0 , we write a are

defined from as in Section 2.7. In particular, a b means that a are transitive on N0 .
Proof. Define T = {m N0 : whenever n m and p n, then p m}. Since there are no
elements n in the empty set, it follows that is vacuously in T . In other words, 0 T .
Suppose that m T . [We want to prove that m+ T ; i.e., we have to prove
that whenever n m+ and p n, then p m+ .] Let n m+ and p n. [We have
to prove that p m+ .] Since m+ = m {m}, the assumption n m+ says that n m
or n = m. In the former case, since m T , it follows that p m, and in the second case
also since p n = m, it follows that p m. But then p m {m} = m+ . This proves
that T is an inductive subset of N0 . But N0 is the smallest inductive set, so that T = N0 .
Thus by the definition of <, whenever p < n and n < m, we conclude that p < m.
Thus < is transitive. It follows by Exercise 2.7.2 that , , > are transitive.
Theorem 3.3.4 For all n N0 ,
n 6< n, n+ 6 n, n+ 6< n,
or, in other words,
n 6 n, n+ 6 n, n+ 6 n.
In particular, n 6= n+ .
Proof. In the proof we use the set notation rather than the order notation.
Let T be the subset of N0 containing all n for which n 6 n, n+ 6 n, and n+ 6 n.
98
Since the empty set contains no elements, it follows immediately that 0 6 0 and 0+ 6 0.
Furthermore, since 0+ = {} = {}, this set is not a subset of the empty set. Thus
0 T.
Now suppose that n T .
Suppose (for contradiction) that n+ n+ . By the definition of n+ , this says that

n+ n {n}, so that either n+ n or n+ = n. But both of these contradict the
assumption that n T . Thus we have proved that n+ 6 n+ .
Suppose (for contradiction) that (n+ )+ n+ . By the definition of (n+ )+ , this says
that n+ {n+ } n+ . In particular, n+ n+ {n+ } n+ , which contradicts the previous
paragraph, and thus proves that (n+ )+ 6 n+ .
Finally, suppose (for contradiction) that (n+ )+ n+ . Since n+ (n+ )+ , by Theorem 3.3.3 it follows that n+ n+ . But we have already proved that this is impossible.
Thus (n+ )+ 6 n+ .
We just proved that n+ T . Thus T is an inductive subset of N0 . Since N0 is the
smallest inductive subset of itself, necessarily T = N0 , which proves the theorem.
Theorem 3.3.5 The relation on N0 is symmetric.

Proof. Suppose that m n and n m. If m 6= n, then necessarily m < n and n < m,
and by Theorem 3.3.3, n < n. This means that n n, which contradicts Theorem 3.3.4.
So necessarily m = n.
Lemma 3.3.6 For all m, n N0 , if m ( n+ , then n 6 m.
Proof. Set T = {m N0 : for all n N0 , if m ( n+ then n 6 m}. Since 0 = has no
elements, the conclusion in the defining statement for T is true for , so that T . Now
suppose that m T . Let n N0 such that m+ ( n+ . This implies that m 6= n, and since
m+ = m {m}, it also implies that m ( n+ . As m T , it follows that n 6 m. Thus
n 6 m {m} = m+ , which proves that m+ T . Thus T is an inductive subset of N0 . Since
N0 is the smallest inductive subset of itself, necessarily T = N0 , which proves the lemma.
Lemma means a helpful theorem, possibly not interesting in its own right, but useful later. There are
examples of so-called lemmas that have turned out to be very interesting in their own right.
99
Theorem 3.3.7 Let m, n N0 .
(1) m < n if and only if m ( n.

(2) m n if and only if m n.
Proof. Set T = {n N0 : for all m N0 , m < n if and only if m ( n}. Note that m <
(i.e., m ) and m ( are both impossible, so that n = = 0 T vacuously.
Now let n T , and let m N0 . Suppose that m < n+ . This means that m n {n}.
If m n, this means that m < n, and since n T , we have that m ( n n {n} = n+ ,
and if m {n}, then m = n, which is a proper subset of n+ by Theorem 3.3.4. This proves
one half of what we need to show for concluding n+ T . It remains to prove that m < n+
if m ( n+ . So suppose that m ( n+ . By Lemma 3.3.6, n 6 m, so that the assumption
m ( n+ = n {n} means that m n. If m = n, then certainly m n+ , so we may
assume that m ( n. Since n T , it follows that m < n, and as n < n+ , by transitivity
(Theorem 3.3.3), m < n+ . This proves that n+ T , so that T is an inductive set, and so
that T = N0 , and (1) is proved.
Now let m, n N0 . We have that m n if and only if m = n or m < n, which by (1)
is true if and only if m = n or m ( n, which holds if and only if m n. This proves (2)
and hence the theorem.
Theorem 3.3.8 For m, n N0 , m n if and only if m+ n+ .
Proof. Suppose that m n. Then by Theorem 3.3.7, m n, and by definition, either
m = n or m < n, i.e., either m = n or m n. In all cases, by Theorem 3.3.7, m+ =
m {m} n n {n} = n+ , so that again by Theorem 3.3.7, m+ n+ .
Now suppose that m+ n+ . This says that m {m} n {n}. In particular,

m n {n}. If m n, then m < n so that m n. If instead m {n}, then m = n, so
that again m n.
Proof. Suppose that m n. If m = n then certainly m+ = n+ , so that m+ n+ .

Now suppose that m < n. By definition, m n, and by Theorem 3.3.7, m n. Thus
m+ = m {m} n n {n} = n+ , so that by Theorem 3.3.7, m+ n+ .
Now suppose that m+ n+ . This says that m {m} n {n}. In particular,

m n {n}. If m n, then m < n so that m n. If instead m {n}, then m = n, so
that again m n.
100
Lemma 3.3.9 For all m, n N0 , either m n or n m.

Proof. Let T = {m N0 : for all n N0 , m n or n m}. By Theorem 3.1.9, 0 T .
Suppose that m T . We want to prove that m+ T . Let n be arbitrary in N0 . We

have to prove that either m+ n or n m+ . By Theorem 3.1.9 this is true if n = 0.
So we may assume that n 6= 0. Then by Theorem 3.1.7, n = p+ for some p N0 . Since
m T , it follows that either m p or p m, and by Theorem 3.3.8, this says that either
m+ n or n m+ . Thus m+ T , so that T is inductive, and so T = N0 .
Theorem 3.3.10 (Trichotomy on N0 ) For all m, n N0 , exactly one of the following
conditions hold:
(1) m < n,
(2) n < m,
(3) m = n.
Proof. By Lemma 3.3.9 and definition of < in terms of , at least one of the following
three conditions holds: m < n, m = n, n < m. In the language of sets, by Theorem 3.3.7,
this says that at least one of the following conditions holds: m n, m = n, n m. By
Theorem 3.3.4, for all p N0 , p 6 p, so that the middle condition cannot be simultaneously
satisfied with any of the other two conditions. If the other two conditions are satisfied, then
m n and n m, so that by Theorem 3.3.3, n n, which also contradicts Theorem 3.3.4.
So no two conditions can be satisfied simultaneously. This proves the trichotomy.
Lemma 3.3.11 Let m, n N0 be such that n m n+ . Then either m = n or m = n+ .
+
+
Proof. Suppose n = 0. If m 6= 0, then m = m+
0 for some m0 N0 , so m0 0 , so
that by Theorem 3.3.8, m0 0, which is only possible if m0 = 0, which then implies that
m = m+
0 = 1. Thus the lemma holds for n = 0.
Suppose that the lemma holds for n. We will prove that it holds for n+ . So suppose
that n+ m (n+ )+ . Since n+ m, necessarily m 6= 0. Thus m = m+
0 for some
+
+
+ +
m0 N0 . Then n m0 (n ) , so that by Theorem 3.3.8, n m0 n+ . By
assumption on n, either m0 = n or m0 = n+ , so that either m = n+ or m = (n+ )+ . Thus
the lemma also holds for n+ . This proves that the lemma holds for 0, and that when it
holds for n, it also holds for n+ . Thus it holds for all n N0 , which proves the lemma.
101
Proposition 3.3.12 N0 is well-ordered (see Definition 2.7.6). In other words, every nonempty subset S of N0 has a least element, that is, an element r such that for all t S,
r t.
Proof. We will prove that the following set is inductive:
T = {n N0 : If S N0 and n S, then S has a least element}.
Note that 0 T because 0 n for all n N0 (and hence 0 n for all n S). Suppose
that n T . We need to prove that n+ T . Let S N0 and n+ T . By assumption that
n T , the set S {n} has a least element. Thus there exists r S {n} such that for all
t S, r t and r n. If r S, then we just showed that for all t s, r t, so that S has
a least element. Now suppose that r 6 S. So necessarily r = n. Then we claim that n+ is
the least element of S. Namely, let t S. If t < n+ , then since n t, we have n t < n+ ,
which is a contradiction. So necessarily t n+ .
This shows that every non-empty subset of N0 has a least element.
So far the order on N0 was given in terms of set inclusions. Now we relate it to
arithmetic more directly:
Theorem 3.3.13 If m, n N0 and m n, there exists r N0 such that m + r = n.
Proof. Let T = {n N0 : for all m N0 , if m n then r N0 such that m + r = n}.
If m, n N0 , m n and n = 0, then since 0 stands for the empty set, necessarily
m = 0, and hence we can set r = 0 to get m + r = n. Thus 0 T .
Suppose that n T . We want to prove that n+ T . Let m N0 such that m n+ .
If m = 0, set r = n+ and m+r = n+ . If instead m > 0, then by Theorem 3.1.7, m = p+ for
some p N0 . Then by assumption p+ = m n+ , and so by Theorem 3.3.8, p n. Since
n T , there exists r N0 such that p + r = n. Hence m + r = p+ + r = (p + r)+ = n+ .
Thus in all cases, whether m = 0 of m > 0, there exists r N0 such that m + r = n+ ,
so that n+ T . Thus T is an inductive set, and since N0 is the smallest inductive set, it
follows that T = N0 .
3.3.1 Prove that for all n N0 , n < n+ .
3.3.2 Prove that if m < n and n p, then m < p. Verify also other standard combinations
of and <. In particular, prove that < is a transitive relation.
3.3.3 Let m, n N0 and m < n. Prove that there exists r N>0 such that m + r = n.
3.3.4 (Compatibility of and < with addition) Let m, n, p N0 . Prove that m n

if and only if m + p n + p and that m > n if and only if m + p > n + p.
102
3.3.5 (Compatibility of with multiplication by non-zero numbers) Let

m, n, p N0 .
i) Suppose that m n. Prove that mp np for all p.
ii) Suppose that m < n. Prove that mp np for all p and that mp < np for all p 6= 0.
3.3.6 Suppose that m, p, n N0 and m n. Prove that m + p < n + p+ .
3.3.7 Prove that for all n N0 , n < 2n . (Recall Notation 2.5.13. Use inductive sets.)
3.3.8 Let a, b, c N+ . If ab = c, prove that a c and b c.
3.3.9 Let a, b, c N0 . Is it true that ab = c implies that a c?
3.3.10 Let m, n, r N0 such that m = n + r. Prove that m n.
3.4
Cancellation in N0
The familiar subtraction cannot be applied to arbitrary pairs of elements in N0 : recall

that elements of N0 are sets, and we have no way of identifying the familiar negative
numbers with sets. However, we can cancel addition of equal terms, as we prove below in
Theorem 3.4.2. We need an intermediate result:
Lemma 3.4.1 Let m, n N0 . Then m = n if and only if m+ = n+ .
Proof. If m = n, then certainly the successor m+ of m is the successor n+ of n.
Suppose that m+ = n+ . Then m m+ = n+ = n {n}, so that if m 6= n, necessarily
m n. Similarly, n m. Then by Theorem 3.3.3, m m, but this is impossible by
Theorem 3.3.4. So necessarily m = n.
Theorem 3.4.2 (Cancellation theorem) Let m, n, p N0 .
(1) m + p = n + p if and only if m = n.
(2) m + n = 0 if and only if m = 0 and n = 0.
(3) m n = 0 if and only if m = 0 or n = 0.
(4) If p 6= 0 and p m = p n, then m = n.
Proof. (1) Certainly if m = n, then m + p = Am (p) = An (p) = n + p. Let T = {p N0 :
whenever m + p = n + p for some m, n N0 , then m = n}. By Theorem 3.2.3 (1), 0 T .
Suppose that p T . We will prove that p+ is also in T . For this, suppose that for
some m, n N0 , m + (p+ ) = n + (p+ ). By the definition of + and by commutativity and
associativity of addition (Theorem 3.2.3 (4) and (3)), m+(p+ ) = m+(p+1) = m+(1+p) =
(m + 1) + p = (m+ ) + p, and similarly n + (p+ ) = (n+ ) + p, so that (m+ ) + p = (n+ ) + p.
As p T and as m+ , n+ N0 , it follows that m+ = n+ . Hence by Lemma 3.4.1, m = n.
103
This proves that p+ T , whence T is an inductive set and necessarily equal to N0 . This
proves the additive cancellation.
(2) Certainly 0 + 0 = 0. Now suppose that m + n = 0. This says that Am (n) = 0.
If n 6= 0, then n = q + for some q N0 . Thus by the definition of addition, Am (n) =
Am (q + ) = (Am (q))+ , which cannot be the empty set. So necessarily n = 0. But addition
is commutative, so similarly m = 0. This proves (2).
(3) In Theorem 3.2.3 (3) we proved that if m = 0 or n = 0, then m n = 0. Now
suppose that m n = 0. If n 6= 0, then n = p+ for some p N0 (by Theorem 3.1.7), hence
0 = m (p+ ) = m p + m. If m 6= 0, then m is a successor of some element k of N0 . Thus
m p + m = m p + k + = m p + (k + 1) = (m p + k) + 1 = (m p + k)+ is the successor
of some element of N0 as well, which means that m p + m 6= 0. So necessarily m = 0. We
just proved that n 6= 0 implies m = 0, or in other words, we just proved that either n = 0
or m = 0. This proves (3).
(4) Fix p N0 \ {0}. Let
T = {m N0 : whenever p m = p n for some n N0 , then m = n}.
If p 0 = p n, then by (3), 0 = p n and n = 0, so that 0 T . Suppose that m T and
that for some n N0 , p (m+ ) = p n. By (3), n 6= 0, so by Theorem 3.1.7, n = r + for
some r N0 . By Definition of , p (m+ ) = p m + p, and p n = p (r + ) = p r + p, so
that p m + p = p r + p. By (1), p m = p r, so that as m T , necessarily m = r. Then
m+ = r + = n, which proves that m+ T . Thus T is an inductive set, so T = N0 .
3.4.1 Suppose that m, p N0 satisfy m (p + m) = 0. Prove that m = 0.
3.4.2 Suppose that m, p, n N0 satisfy m + p = m (n+ ). Prove that p = m n.
3.5
Construction of Z, arithmetic, and order on Z
How are we supposed to think of the familiar 5? It does not seem to be possible to
represent negative numbers with sets as we did for non-negative numbers. We will go about
it by using the rigorous construction of N0 and all the arithmetic on it. The following is a
rehashing of Example 2.3.11.
Definition 3.5.1 Consider the Cartesian product N0 N0 . Elements are pairs of the form
(a, b), with a, b N0 . If a, b, a , b N0 , we will write (a, b) (a , b ) if a + b = a + b .
In Example 2.3.11 it was proved that the relation is an equivalence relation on
N0 N0 . Thus we can talk about the equivalence class [(a, b)] of (a, b).
104
Definition 3.5.2 We define Z to be the set of all equivalence classes of elements of N0 N0

under the equivalence relation . Elements of Z are called whole numbers, or integers.
We have a natural inclusion of N0 into Z by identifying n N0 with [(n, 0)] in Z.
Note that if n, m N0 are distinct, then so are [(n, 0)] and [(m, 0)], so that N0 is indeed
identified with a natural subset of Z. But Z is strictly larger: [(0, 1)] is not equal to [(m, 0)]
for all m N0 .
To define arithmetic on Z, we first define arithmetic on N0 N0 . Start with addition:
for a, b, c, d N0 , set
(a, b) + (c, d) = (a + c, b + d).
(There are three + here: the last two are the already well-studied addition in N0 , and
the first one is the one we are defining now on N0 N0 . Yes, we are using the same name
for two different operations; but thats ok: many families have a child named Pat, and the
many Pats are not all equal.) Since + on N0 is binary, so is + on N0 N0 . We show next
that + is compatible with :
Theorem 3.5.3 Let a, a , b, b, c, c , d, d N0 such that (a, b) (a , b ) and (c, d) (c , d ).
Then (a + c, b + d) (a + c , b + d ).
Proof. By assumption, a + b = a + b and c + d = c + d. Then by associativity and
commutativity of + on N0 , (a + c) + (b + d ) = (a + b ) + (c + d ) = (a + b) + (c + d) =
(a + c) + (b + d ), so that (a + c, b + d) (a + c , b + d ).
This shows that + is well-defined on the equivalence classes of , hence + makes

sense on Z:
[(a, b)] + [(c, d)] = [(a + c, b + d)].
For example, identifying 5, 6 N0 with [(5, 0)], [(6, 0)] Z/0Z, we get 5 + 6 = 11 and
[(5, 0)] + [(6, 0)] = [(5 + 6, 0)] = [(11, 0)], as expected. Also, [(5, 0)] + [(0, 6)] = [(5, 6)] =
[(0, 1)].
Multiplication on Z is a little more complicated: if (a, b) and (c, d) are in N0 N0 ,
then we declare
[(a, b)] [(c, d)] = [(a c + b d, a d + b c)].
We next prove that is well defined on equivalence classes. Suppose that [(a, b)] = [(a , b )]
and [(c, d)] = [(c , d )]. This means that a+b = a +b and c+d = c +d. By the established
associativity, distributivity, and commutativity properties in N0 , we then have that
(ac+bd + a d + b c ) + (a + b )d + (a + b)c + b(c + d ) + a(c + d)
= ac + bd + a d + b c + ad + b d + a c + bc + bc + bd + ac + ad
= ad + bc + a c + b d + a d + bd + ac + b c + bc + bd + ac + ad (by inspection)
105
= (ad + bc + a c + b d ) + (a + b)d + (a + b )c + b(c + d) + a(c + d )

= (ad + bc + a c + b d ) + (a + b )d + (a + b)c + b(c + d ) + a(c + d),
so that by cancellation Theorem 3.4.2 in N0 , ac + bd + a d + b c = ad + bc + a c + b d . It
follows that [(ac + bd, ad + bc)] = [(a c + b d , a d + b c )], so that multiplication on Z is
well-defined.
Consequently, [(5, 0)] [(6, 0)] = [(5 6 + 0 0, 5 0 + 0 6)] = [(30, 0)], as expected.
Similarly compute [(5, 0)] [(0, 6)].
We next prove the basic arithmetic laws on Z.
Theorem 3.5.4 The following identities hold for all m, n, p Z:
(1) (Additive identity) m + [(0, 0)] = m = [(0, 0)] + m.
(2) m [(0, 0)] = [(0, 0)] = [(0, 0)] m.
(5) (Multiplicative identity) m [(1, 0)] = m = [(1, 0)] m.
(9) (Existence of additive inverses) There exists r Z such that m + r = r + m =
[(0, 0)].
Proof. Let a, b, c, d, e, f N0 such that m = [(a, b)], n = [(c, d)] and p = [(e, f )]. With this,
the proofs follow easily from the definitions of + and . For example,
m + [(0, 0)] = [(a, b)] + [(0, 0)] = [(a + 0, b + 0)] = [(a, b)] = m,
and similarly [(0, 0)] + m = m. This proves (1).
Properties (2)-(4) are equally easy to prove, and (5) follows from
m [(1, 0)] = [(a, b)] [(1, 0)] = [(a 1 + b 0, a 0 + b 1)] = [(a + 0, 0 + b)] = [(a, b)] = m,
and similarly [(1, 0)] m = m.
We check (6), and you explain each step:
(m (n + p)) = ([(a, b)] ([(c, d)] + [(e, f )]))
= [(a, b)] [(c + e, d + f )]
= [(a (c + e) + b (d + f ), a (d + f ) + b (c + e))]
= [(a c + a e + b d + b f, a d + a f + b c + b e)]
(why may we omit parentheses in addition?)
106

= [(a c + b d, a d + b c)] + [(a e + b f, a f + b e)]
= [(a, b)] [(c, d)] + [(a, b)] [(e, f )]
= m n + m p.
You verify (7) and (8).

(9) Set r = [(b, a)]. Then m + r = [(a, b)] + [(b, a)] = [(a + b, b + a)] = [(a + b, a + b)] =
[(0, 0)], and similarly r + m = [(0, 0)].
Notation 3.5.5 The additive inverse of the equivalence class [(a, b)] of (a, b) is the equivalence class [(b, a)] of (b, a). The additive inverse of an element m is always denoted m.
Notation 3.5.6 Furthermore, for any a, b N0 , by trichotomy on N0 (Theorem 3.3.10),
exactly one of the following three conditions is satisfied:
a < b,
a = b,
b < a.
In the first two cases, by Theorem 3.3.13, there exists r N0 such that a + r = b, and then
[(a, b)] = [(a, a + r)] = [(0, r)], and in the last case there exists r N0 such that b + r = a,
and then [(a, b)] = [(b + r, b)] = [(r, 0)]. This proves that every element of Z can be written
as the equivalence class of [(a, b)] where a or b is 0.
Recall the identification of elements of N0 with elements in Z: n N0 is identified
with [(n, 0)]. Then with the last two paragraphs,
[(a, b)] = [(a, 0)] + ([(0, b)]) = a + (b),
and it is standard shorthand to write this as ab. Thus in the future we will write elements
[(a, b)] in Z as a b.
Theorem 3.5.7 Let m, n Z. Then (m) n = m (n) = (m n).
Proof. Observe that
m (n) + m n = m ((n) + n) (by distributivity: Theorem 3.5.4 (6))
= m 0 (by notation for additive inverses)
= 0 (by Theorem 3.5.4 (2)).
Thus m(n) is an additive inverse of mn, and by uniqueness of inverses (Theorem 2.5.10),
m (n) is the additive inverse of m n, i.e., m (n) = (m n).
With that,
(m) n = n (m) (by commutativity of multiplication)
= (n m) (by the previous paragraph).
107
Theorem 3.5.8 Let m, n Z. Then m n = 0 if and only if m = 0 or n = 0.

Proof. We already proved that if m = 0 or n = 0, then m n = 0.
Now suppose that m n = 0. Write m = [(a, b)] with a, b N0 . By Notation 3.5.6 we
may assume that either a = 0 or b = 0. Write n = [(c, d)] for c, d N0 .
If a = 0, then [(0, 0)] = 0 = m n = [(0, b)] [(c, d)] = [(b d, b c)]. This means (by
definition of ) that b d = b c. If b = 0, then m = [(0, 0)] = 0, and if b 6= 0, then by
Theorem 3.4.2, d = c, so that n = [(c, d)] = [(0, 0)] = 0.
Similarly, if b = 0, then either m = 0 or n = 0. But a = 0 or b = 0 covers all the
possible cases for elements of Z, so weve proved the theorem.
Definition 3.5.9 (Order on Z) Define a relation on Z as follows: [(a, b)] [(c, d)] if
a + d b + c, where the latter is the already known relation on N0 (see Section 3.3).
Similarly define <, , >.
It is left to the reader to verify the following:
Lemma 3.5.10
(1) on Z is well-defined (this means that if one picks different representatives of the
equivalence classes, for one pair holds if and only if it holds for the other pair).
(2) , are reflexive and transitive but not symmetric.
(3) <, > are transitive but not reflexive and not symmetric.
(4) If m Z and 0 m, then m N0 .
(5) If m Z and m 0, then m N0 .
Theorem 3.5.11 The following properties hold on Z:
(1) (Trichotomy) For any m, n Z, exactly one of the following three conditions
holds:
(i) m = n.
(ii) m < n.
(iii) n < m.
(2) (Transitivity) For all m, n, p Z, m < n and n < p imply m < p.
(3) (Compatibility of < with addition) For all m, n, p Z, m < n implies m+p <
n + p.
(4) (Compatibility of < with multiplication by positive numbers) For all
m, n, p Z, m < n and 0 < p imply mp < np.
Proof. Write m = [(a, b)], n = [(c, d)], and p = [(e, f )] for some a, b, c, d, e, f N0 .
By Theorem 3.3.10, exactly one of the following three conditions holds:
a + d are compatible with the choice of representatives by
Lemma 3.5.10, this says that exactly one of the following three conditions on Z:
[(a, b)] < [(c, d)],
[(a, b)] = [(c, d)],
[(c, d)] < [(a, b)].
This proves (1).

Suppose that m < n and n < p. This means that a + d < b + c and c + f < d + e.
Then by associativity and commutativity of addition in N0 , and by transitivity of < on N0 ,
(a + f ) + (c + d) = (a + d) + (c + f ) < (a + d) + (d + e) < (b + c) + (d + e) = (b + e) + (c + d),
so that by Exercise 3.3.4, a + f < b + e. This says that m < p, which proves that < is
transitive. This proves (2).
The other properties follow similarly and are left to the reader.
With this, we can use the adjectives positive, negative, non-positive, nonnegative from Definition 2.7.8 for elements of Z, and in general we can use the properties
developed in Section 2.7, such as absolute values, triangle inequality, and reverse
triangle inequality. We can also generalize Proposition 3.3.12 as follows:
Proposition 3.5.12 Let S be a non-empty subset of Z that is bounded above (resp. below). Then S contains its largest (resp. least) element. (For definitions see Definition 2.7.2.)
Proof. Let s S, and let b Z be an upper bound on S. Let T = {n N0 : b n S}.
Then T is a subset of N0 . Note that b s N0 and that s = b (b s) S, so that T is not
empty. Thus by Proposition 3.3.12, T has a least element, call it t. Thus b t S. We will
prove that b t is an upper bound for S. Let r S. We need to prove that b t r. Since
b is an upper bound on S and r S, we have that b r N0 . Since r = b (b r) S,
we have that b r T . But then by the choice of t, t b r, which proves that r b t,
as desired. This proves the first part.
Now let s S, and let b Z be a lower bound on S. Let T = {n Z : b + n S}.
Since b is a lower bound, we conclude that T is a subset of N0 . Note that s b N0 and
that s = b + (s b) S, so that T is not empty. Thus by Proposition 3.3.12, T has a least
element, call it t. Thus b + t S. We will prove that b + t is a lower bound for S. Let
r S. We need to prove that b + t r. Since b is a lower bound on S and r S, we have
that r b N0 . Since r = b + (r b) S, we have that r b T . But then by the choice
of t, t r b, which proves that r b + t, as desired. This proves the second part, and
hence the theorem.
3.5.1 Prove that for all a, b, c N0 , (a, b) (a + c, b + c).
109
3.5.2 Suppose that a, b, c, d N0 such that (a, b) (c, d) and a c. Prove that there
exists e N0 such that a = c + e and b = d + e.
3.5.3 Convince yourself that Z as defined in Definition 3.5.2 is the same as the familiar
set of all integers.
3.5.4 (Cancellation in Z) Let m, n, p Z.
i) If m + p = n + p, prove that m = n.
ii) If p 6= 0 and m p = n p, prove that m = n.
3.5.5 Prove that if (a, b) (c, d) then (a, c) (b, d).
3.5.6 Let a, b, c, d N0 . Suppose that [(b, a)] < [(c, d)]. Prove that [(a, b)] > [(d, c)] and
[(a, b)] + [(c, d)] > [(0, 0)].
3.5.7 Prove (2), (3), (4), (7), (8) in Theorem 3.5.4.
3.5.8 Prove Lemma 3.5.10.
3.5.9 Prove (3) and (4) in Theorem 3.5.11.
3.6
Construction of the field Q of rational numbers
We next define the set Q, arithmetic, and order on it, and we show that Q is a field.
Elements of course correspond to the familiar rational numbers.
The following is a rehashing of the construction of Q from Z first given in Exercise 2.3.10. The relation on Z (Z \ {0}), given by
(m, n) (m , n )
if
m n = m n,
is an equivalence relation. The reader may wish to go through the details of the proof, to
check that only those properties of Z are used that we have proved in the previous section,
and that no well-known knowledge creeps into the proof.
Remark 3.6.1 For any a Z and b, c Z \ {0}, the equivalence class [(ca, cb)] equals
the equivalence class [(a, b)]. In particular, by possibly taking a negative c, we may always
choose a representative of any equivalence class in Z N>0 .
Definition 3.6.2 We define the set of all equivalence classes of this relation to be Q. The
elements of Q are called rational numbers.
We have a natural inclusion of Z into Q (and thus of N0 into Q) by identifying m Z
with [(m, 1)]. Note that if m, n Z are distinct, so are [(m, 1)] and [(n, 1)], so that Z is
indeed identified with a natural subset of Q. But Q is strictly larger: [(1, 2)] is not equal
to [(m, 1)] for any m Z. More about the familiar rational numbers is in Notation 3.6.5.
110

We declare the following binary operations + and on elements of Z (Z \ {0}):
(m, n) + (p, q) = (m q + p n, n q),
(m, n) (p, q) = (m p, n q).
(The + and inside the pair entries are the operations on Z; the other + and are what
we are defining here.) We show next that the new + and are compatible with :
Theorem 3.6.3 Let m, m , n, n, p, p , q, q N0 such that (m, n)(m , n ) and (p, q)(p, q ).
Then
((m, n) + (p, q)) ((m , n ) + (p , q )) ,
and
((m, n) (p, q)) ((m , n ) (p , q )) .
Proof. By assumption, m n = m n and p q = p q. Then (you explain each step):

(m q + p n) (n q ) = (n q ) (m q + p n)
= (n q ) (m q) + (n q ) (p n)
= (q n ) (m q) + n (q (p n))
= q (n (m q)) + n ((q p) n)
= q ((n m) q) + n ((p q ) n)
= q ((m n ) q) + n ((p q) n)
= q ((m n) q) + n (p (q n))
= q (m (n q)) + n (p (n q))
= (q m ) (n q) + (n p ) (n q)
= (n q) (q m ) + (n q) (n p )
= (n q) (m q ) + (n q) (p n )
= (n q) (m q + p n )
= (m q + p n ) (n q).
This says that

(m q + p n, n q), = (m q + p n , n q ),
which in turn says that ((m, n) + (p, q)) ((m , n ) + (p , q )).
It is even easier to prove that ((m, n) (p, q)) ((m , n ) (p , q )) (verify).
The last theorem proves that that the following binary operations + and on Q are
well-defined:
[(m, n)] + [(p, q)] = [(m, n) + (p, q)] = [(m q + p n, n q)],
[(m, n)] [(p, q)] = [(m, n) (p, q)] = [(m p, n q)].
111
The + and on the left are what we are defining here; the + and in the middle are binary
operations on Z (Z \ {0}), and the + and on the right are binary operations on Z.
We now move to arithmetic laws on Q.
Theorem 3.6.4 Q is a field (see Definition 2.6.1), or explicitly, the following identities
hold for all m, n, p Q:
(1) (Additive identity) m + [(0, 1)] = m = [(0, 1)] + m.
(2) m [(0, 1)] = [(0, 1)] = [(0, 1)] m.
(5) (Multiplicative identity) m [(1, 1)] = m = [(1, 1)] m.
(9) [(1, 1)] 6= [(0, 1)].
(10) (Existence of additive inverses) There exists r Q such that m + r = r + m =
[(0, 1)].
(11) (Existence of multiplicative inverses) If m 6= [(0, 1)], there exists r Q such
that m r = r m = [(1, 1)].
Proof. Let a, b, c, d, e, f Z such that m = [(a, b)], n = [(c, d)] and p = [(e, f )]. With this,
the proofs follow easily from the definitions of + and . For example,
m + [(0, 1)] = [(a, b)] + [(0, 1)] = [(a + 0, b 1)] = [(a, b)] = m,
and similarly [(0, 1)] + m = m. This proves (1).
Properties (2)-(5) are equally easy to prove.
We check (6), and you explain each step:
m (n + p) = [(a, b)] ([(c, d)] + [(e, f )])
= [(a, b)] [(c f + e d, d f )]
= [(a (c f + e d), b (d f ))]
= [(a (c f ) + a (e d), b (d f ))]
= [(a, b)] [(c, d)] + [(a, b)] [(e, f )]
= m n + m p.
You verify (7) and (8).

(9) By the properties of Z, 0 1 = 0 6= 0+ = 1 = 1 1, so that (0, 1) 6 (1, 1), so that
[(0, 1)] 6= [(1, 1)] in Q.
112
(10) By definition of m and Q, [(a, b)] Q, and m+[(a, b)] = [(ab+(a)b, bb)] =
[(b a + b (a), b b)] = [(b (a + (a)), b b)] = [(b 0, b b)] = [(0, b b)] = [(0, 1)], and
similarly [(a, b)] + m = [(0, 1)]. Thus additive inverses exist in Q.
(11) By assumption [(a, b)] 6= [(0, 1)], This means that a 1 6= 0 b, so that a 6= 0.
Then [(b, a)] Q, and [(a, b)] [(b, a)] = [(a b, b a)] = [(a b, a b)] = [(1, 1)]. Similarly
[(b, a)] [(a, b)] = [(1, 1)].
Notation 3.6.5 The additive inverse of the equivalence class [(a, b)] of (a, b) is [(a, b)],
and the multiplicative inverse of non-zero [(a, b)] is [(b, a)]. For an element m, its additive
inverse is always denoted m, and the multiplicative inverse, when it exists, is written
m1 . Thus for any n Z, by identifying n with [(n, 1)] in Q, the additive inverse is n =
[(n, 1)] = [(n, 1)], and if n 6= 0, the multiplicative inverse is n1 = [(n, 1)]1 = [(1, n)].
Thus, under our identifications, any element [(m, n)] in Q is equal to
[(m, n)] = [(m, 1)] [(1, n)] = m n1 ,
and it is standard shorthand to write this as m/n. Thus in the future we will write elements
[(m, n)] in Q as m/n, and whenever m, n Z and n 6= 0, then [(m, n)] = m/n Q. By
Remark 3.6.1 we may even assume that n > 0.
Theorem 3.6.6 (Cancellation laws in Q) Let m, n, p Q.
(1) If m + p = n + p, then m = n.
(2) If m p = n p and if p 6= 0, then m = n.
Proof. For part (1),
m = m + 0 (by additive identity)
= m + (p + (p))
= (m + p) + (p) (by associativity of addition)
= (n + p) + (p) (by assumption)
= n + (p + (p)) (by associativity of addition)
= n + 0 (by additive identity)
= n (by additive identity),
which proves (1). The proof of part (2) follows the proof above with the following changes:
replace all + by , replace 0 by 1, replace p by p1 , and correspondingly
modify all explanations.
113
Theorem 3.6.7 Q is an ordered field (see Definition 2.7.8) with the order < given by:

ad < bc; if bd > 0;
[(a, b)] < [(c, d)] if
ad > bc; if bd < 0.
Proof. Theorem 3.6.4 proves that Q is a field. Now we prove that Q is ordered.
First we need to prove that < is well-defined. Let m, n Q. Write m = [(a, b)] =

[(a , b )], n = [(c, d)] = [(c , d )]. By definition of Q, a b = ab and c d = cd . By associativity
and commutativity of multiplication in Z, (bd)(b d )(b c a d ) = bb dd (b c a d ) =
b(b )2 d (dc ) b d(d )2 (ba ) = b(b )2 d (cd ) b d(d )2 (ab ) = (b )2 (d )2 (bc ad). Thus the
sign of (bd)(bd )(b c a d ) is the same as the sign of bc ad. In particular, if bd > 0,
b d > 0, and ad < bc, then necessarily a d 0, b d < 0, and ad < bc, then
necessarily a d > b c ; if bd < 0, b d > 0, and ad < bc, then necessarily a d > b c ; if
bd < 0, b d < 0, and ad < bc, then necessarily a d < b c . All these cases confirm that the
order is independent of the representative of the equivalence class.
Trichotomy follows from trichotomy on Z (Theorem 3.5.11). The proof of transitivity
of <, and of compatibility of < with addition and with multiplication by positive rational
numbers is left as an exercise.
Thus the following terms and facts, developed in Section 2.7, apply to elements of Q:
positive, negative, non-positive, non-negative, intervals, absolute value, triangle
inequality, reverse triangle inequality.
Theorem 3.6.8 (Archimedean property) Let m, n Q. If m > 0, there exists p N+
such that n < pm.
Proof. If n 0, then take p = 1, and n 0 < m = pm. So we may assume that n > 0.
Write m = a/b and n = c/d for some positive a, b, c, d Z. Set p = bc + 1. Then p is a
positive integer, as is ad. By compatibility of < with addition on Z and multiplication by
positive integers, we have that pad = bcad + ad > bcad bc, and since b, d > 0, it follows
that pm = p(a/b) > c/d = n.
3.6.1 Compute (5/6) (3/4), (5/6) + (3/4).
3.6.2 Prove that for all (m, n) Z (Z \ {0}) and all non-zero p Z, (m, n) (m p, n p).
In other words, m/n = (m p)/(n p).
3.6.3 Prove (2), (3), (4), (5), (7), (8) in Theorem 3.6.4.
3.6.4 Finish the proof of Theorem 3.6.7.
3.6.5 Let m, n, p, q Z with n, q non-zero. Prove that m/n + p/q and m/n p/q are
rational numbers.
114
3.6.6 Let m Q. Prove that there exists a positive integer N such that m < 2N . (Hint:
If m 0, this is easy. Now suppose that m > 0. Write m = a/b for some positive integers
a, b. Prove that a/b a < 2a ; perhaps apply or reprove Exercise 3.3.7.)
3.6.7 Find a non-empty subset S of Q that is bounded above such that S does not contain
its least upper bound. (Contrast with Proposition 3.5.12.)
3.7
Construction of the field R of real numbers
In this section we construct real numbers from Q via Dedekind cuts: a Dedekind cut
stands for a real number and vice versa. We establish the familiar arithmetic on R, and
in the subsequent section we show that R is an ordered field with further good properties.
To go through these two sections carefully would take many hours. I typically spend 1.5
class hours going through them, asking the students not to take notes but instead to nod
vigorously in agreement and awe at all the constructions and claims. I expect the students
to get the gist of the construction, a conviction that any arithmetic and order property
of real numbers can be verified if necessary, but I do not expect the students to have
gone through all the details. Furthermore, I do not use the standard Dedekind cuts, as
arithmetic with them is messier (see Exercise 3.7.15).
Definition 3.7.1 A Dedekind cut is a pair (L, R) with the following properties:
(1) L, R are non-empty subsets of Q.
(2) For all l L and all r R, l < r.
(3) For all l L, there exists l L such that l < l .
(4) For all r R, there exists r R such that r < r.
(5) L R is either Q or Q \ {m} for some m Q. In case L R = Q \ {m}, then for
all l L and all r R, l < m < r.
By condition (2) and trichotomy property of < on Q, L R = .
We can visualize a Dedekind cut (L, R) as the separation of the rational number
line into the left and right parts, and the separation comes either at a rational number m,
or possibly at something else. The letters L and R are mnemonics for left and
right.
An easy consequence of the definition is that if l L and x Q with x < l, then
x L, and similarly that if r R and x Q with x > l, then x R. (See Exercise 3.7.1.)
115

(1) (Q \ {0}, {0}) is not a Dedekind cut: it violates conditions (2) and (4).
(2) ({x Q : x2 < 2}, {x Q : x2 > 2}) is not a Dedekind cut: it violates condition (2).
(3) ({x Q : x < 0}, {x Q : x > 0}) is a Dedekind cut.
(4) ({x Q : x 0}, {x Q : x > 0}) is not a Dedekind cut: it violates condition (3).
Theorem 3.7.3 Let (L, R) be a Dedekind cut. Let x Q.

(1) Suppose that x < l for some l L. Then x L.
(2) Suppose that x > r for some r R. Then x R.

Proof. We only prove the first part. By condition (2), x is not in R, and by condition (5),
x L.
Definition 3.7.4 We define R to be the set of all Dedekind cuts. Elements of R are
therefore Dedekind cuts, but more normally, they are called real numbers.
The first goal is to prove that Q is a subset of R, at least under an identification of
elements of Q with Dedekind cuts.
Proposition 3.7.5 For any m Q, Dm = ({l Q : l < m}, {r Q : r > m}) is a
Dedekind cut.
Proof. Let L = {l Q : l < m}, R = {r Q : r > m}. By Theorem 3.6.8, there exists
p N+ such that m < p 1 = p. Since p N+ Z Q, we have p R. Similarly, there
exists q Q such that m < q, which says that the rational number q is strictly smaller
than m and is thus in L. Thus L and R are not empty.
By trichotomy on Q, for every r Q, exactly one of the following holds: r < m,
r = m, r > m. So L R = Q \ {m}. By definition, for any l L and any r R, l < m < r,
and by transitivity, we conclude that l < r.
If l L, then l < m, and by Exercise 2.7.17, l = (l +m)/2 is a rational number strictly
larger than l and strictly smaller than m. Thus, l < l L, and condition (3) of Dedekind
cuts holds. Similarly, for any r R, (r + m)/2 is in R and strictly smaller than r, so that
condition (4) of Dedekind cuts is satisfied.
116
Remark 3.7.6 If m, n are distinct in Q, say m < n, then by Exercise 2.7.17, (m + n)/2 is
in the right side of Dm and the left side of Dn , so that Dm and Dn are also distinct. This
gives a natural inclusion of Q into R (and thus of N0 , N+ , and Z into R) by identifying
m Q with the Dedekind cut Dm . Thus we can think of Q as a subset of R.
Many differences between Q and R are in Section 3.8. Here we point out that R is
strictly bigger than Q. Namely, ({x Q : x2 < 2 or x < 0}, {x Q : x2 2 and x 0})
is a Dedekind cut that is not of the form Dm for any m Q (Exercise 3.7.4).
Proposition 3.7.7 Let L be a non-empty proper subset of Q such that for all l L and
all x Q, if x < l then x L. Then there exists a unique subset R of Q such that (L, R)
is a Dedekind cut.
Let R be a non-empty proper subset of Q such that for all r R and all x Q, if
x > r then x R. Then there exists a unique subset L of Q such that (L, R) is a Dedekind
cut.
Proof. We only prove the first part. Set R = Q \ L.
If condition (4) of Dedekind cuts holds for (L, R ), let R = R . Otherwise, there exists
m R such that for all r R , r m. In this case set R = R \ {m}. Now let r R.
Then r > m. By Exercise 2.7.17, (r + m)/2 is a rational number strictly larger than m
and strictly smaller than r, so that by assumption (r + m)/2 R. Thus condition (4) of
Dedekind cuts holds for (L, R).
Since L is a proper non-empty subset of Q, so is R , and even the modified R in
the previous paragraph. If l L and r R, by construction l 6= r and even l < r. Thus
conditions (1) and (2) hold, and condition (3) holds by assumption. Furthermore, condition
(5) holds by construction as well.
Now we turn to arithmetic on R.
Definition 3.7.8 Let A and B be subsets of Q. Define

A + B = {a + b : a A, b B},
A = {a : a A},
A B = {a b : a A, b B} = A + (B),
A B = {a b : a A, b B}.
117
Lemma 3.7.9 Let (L1 , R1 ) and (L2 , R2 ) be Dedekind cuts. Then (L1 + L2 , R1 + R2 ) is a
Dedekind cut.
Proof. Since L1 , L2 , R1 , R2 are non-empty subsets of Q, then L1 + L2 , R1 + R2 are nonempty subsets of Q as well. Let l L1 + L2 and r R1 + R2 . Then l = l1 + l2 and
r = r1 + r2 for some li Li and ri Ri (i = 1, 2). Then l1 < r1 and l2 < r2 , so that
l = l1 + l2 < l1 + r2 < r1 + r2 = r, and by transitivity of < on Q, l < r. Thus conditions
(1) and (2) of Dedekind cuts hold.
Conditions (3) and (4) hold similarly. Furthermore, if l L1 + L2 and x Q with
x < l, then we claim that x L1 + L2 . Namely, write l = l1 + l2 for some li Li ,
i = 1, 2. By assumption we have that x l + l1 < l1 , and since (L1 , R1 ) is a Dedekind cut,
by Exercise 3.7.1, we conclude that xl+l1 L1 . Thus x = xl+l = xl+l1 +l2 L1 +L2 .
Similarly, whenever r R1 + R2 and x Q with r < x, then x R1 + R2 .
It remains to prove Condition (5) of Dedekind cuts for (L1 + L2 , R1 + R2 ). Let x Q
and suppose that x 6 (L1 + L2 ) (R1 + R2 ). We will prove that (L1 + L2 , R1 + R2 ) is the
Dedekind cut Dx . For this we need to prove that any y Q not in (L1 + L2 ) (R1 + R2 )
p
equals x. By Remark 3.6.1 and Notation 3.6.5 we may write x = m
n , y = n for some
m, n, p Z, with n > 0. By the definition of Dedekind cuts the sets Si = {r Z : nr Ri }
are bounded below, so that Si has a least element ai Z by Proposition 3.5.12. Thus
c
c
c
n Ri for all integers c ai , and n 6 Ri for all integers c < ai . In particular, n
R1 + R2 for all integers c a1 + a2 . If (ai 1)/n Li for i = 1, 2, then nc L1 + L2
for all integers c a1 + a2 2, so that necessarily x = y = a1 +an2 1 . So say that
(a1 1)/n 6 L1 . Then by the definition of Dedekind cuts we have that L1 = D(a1 1)/n . If
also (a2 1)/n 6 L2 , then L2 = D(a2 1)/n , nc L1 + L2 for all integers c a1 + a2 4,
a1 +a2 3
1 3
2 3
1 1
2 1
= 2a2n
+ 2a2n
L1 + L2 , a1 +an2 1 = 2a2n
+ 2a2n
R1 + R2 , so that
n
a1 +a2 2
x=y =
. The remaining case is where L1 = D(a1 1)/n and L2 6= D(a2 1)/n . If
n
L2 = Dr for some rational number r, then by possibly taking a multiple of n we may
assume that r is of the form nq for some integer q, but then we are in the previous case.
So we may suppose that (L2 , R2 ) is not a rational number. Then 2a2n1 is either in R2
1 1
2 1
or in L2 . If it is in R2 , then a1 +an2 1 = 2a2n
+ 2a2n
R1 + R2 , so that necessarily
a1 +a2 2
2a1 3
a1 +a2 2
2 1
, and if it is in L2 , then
= 2n + 2a2n
L1 + L2 , so that
x= y =
n
n
a1 +a2 1
. Thus Condition (5) holds.
x=y=
n
Thus we can define addition on R by
(L1 , R1 ) + (L2 , R2 ) = (L1 + L2 , R1 + R2 ).
There are two different meanings to + above: on the right side, + acts on two sets of
rational numbers, and on the left, + stands for the new sum.
The proof of the following is left as an exercise.
118
Theorem 3.7.10 Addition on R, as defined above, is commutative and associative. The

additive identity is D0 = (Q , Q+ ), and the additive inverse of any (L, R) R is (R, L)
(recall Definition 3.7.8).
Naturally we will write the additive identity D0 of R as 0, and for any x R, we write
its additive inverse as x.
Multiplication on R is more complicated, but if one keeps in mind the expected facts
about real numbers, the following definition of multiplication on Dedekind cuts is natural
(see Exercise 3.7.11 for an alternate formulation):
( , R R ), if R and R contain only positive elements of Q;
(R L , ), if R and L contain only positive elements;

(L, R) (L , R ) =
only positive elements;
(L R , ), if R and L contain
( , L L ), if neither R nor R contains only positive elements,
where in each case is the unique subset of Q that makes the result a Dedekind cut. It is
only possible to find such a if the non-blank part on the right side satisfies the conditions
in Proposition 3.7.7. It is left to Exercise 3.7.9 to prove that.
Clearly is a commutative operation, and it is easy to show that D1 = ({x Q : x <
1}, {x Q : x > 1}) is a multiplicative identity (and therefore the unique multiplicative
identity by Theorem 2.5.5). The following are multiplicative inverses for (L, R) 6= D0 :

( , {1/r : r R}), if R contains only positive elements of Q;
1
(L, R) =
({1/l : l L}, ), if R does not contain only positive elements.
If all elements of R are positive, we are certainly not dividing by 0 in the first case. In the
second case, by Exercise 3.7.1, L contains only negative elements, and so again we are not
dividing by 0. It is now straightforward to verify that the listed Dedekind cut is indeed
the multiplicative inverse.
Proofs of associativity of multiplication and of distributivity of multiplication over
addition are more involved. With the definition above, we would have to verify eight cases
for associativity, and many cases for distributivity, depending on whether the left parts
of the various Dedekind cuts contain positive elements. Yes, this can be done, but it is
tedious work. We provide a different proof in these notes that is not a brute-force attack
but is more conceptual and gives side results as a bonus. We first need a lemma.
Lemma 3.7.11 For any Dedekind cuts x, y, (x) y = (x y) = x (y).
Proof. Write x = (L, R) and y = (L , R ).
If L, L contain positive elements, then R, R do not contain positive elements, and
by definition of multiplication,
(x y) + ((x) y) = ((L, R) (L , R )) + ((R, L) (L , R ))
119
= ( , R R ) + ((R) R , )
= ( , R R ) + ((R R ), ).
By Proposition 3.7.7, the left side U of ( , R R ) is either Q \ (R R ) or Q \ ((R R ) {m})
for some rational number m. Since R R consists of positive numbers strictly greater
than numbers in Q \ (R R ) it follows that the left side is contained in the set Q of all
negative rational numbers. Furthermore, let r be any negative rational number. We want
to find x U and y R R such that x y = r. Let x U and y R R . Write
r = r0 /n, x = x0 /n and y = y0 /n for some integers r0 , x0 , y0 , n, with n positive. By
possibly changing x and y but not n, we may assume that x0 is the largest possible and y0
the smallest possible. If (y0 1)/n U , then y0 = x0 +1, and y x = y0 /nx0 /n = 1/n. If
instead (y0 1)/n 6 U , then (y0 1)/n = m, so that (2y0 1)/2n RR , (2y0 3)/2n U ,
and (2y0 3)/2n (2y0 1)/2n = 1/2n. Thus we have found a representation of r as r0 /n
for some positive integer n and some negative integer r0 such that for some integers x0 , y0 ,
x0 /n U , y0 /n R R , and x0 /n y0 /n = 1/n. But then (y0 + (r 1))/n R R and
x0 /n (y0 + (r0 1)/n = r0 /n = r. This proves that the left side of the Dedekind cut in
the display equals Q . Thus (x y) + ((x) y) = D0 . By commutativity of addition this
also gives ((x) y) + (x y) = D0 , ((x) y) = (x y).
If L contains positive elements but L does not, then
(x y) + ((x) y) = ((L, R) (L , R )) + ((R, L) (L , R ))
= (R L , ) + ( , (R L )),
and similarly to the previous case, this is D0 , so again ((x) y) = (x y).

The remaining cases of ((x) y) = (x y) are proved similarly. Then by commutativity of multiplication, x (y) = (y) x, and by the established case this equals
(y x) = (x y).
Theorem 3.7.12 R is a field containing Q.
Proof. The only parts left to prove are that multiplication is associative and that it distributes over addition. Let x = (L, R), y = (L , R ), z = (L , R ) be in R.
To prove that x (y z) = (x y) z, it suffices to prove that the additive inverses of
the two products are identical. By Lemma 3.7.11, we may then in each product replace
x = (L, R) by x = (R, L), and similarly for y by y and z by z. Note that R does
not contain only positive numbers if and only if L contains only positive numbers, so via
these replacements we may assume that R, R , R all contain only positive elements. But
then x (y z) = ( , L (L L )), which is by associativity of multiplication on Q the same
as ( , (L L ) L ) = (x y) z. This proves associativity of multiplication.
To prove that x (y + z) = (x y) + (x z), we may similarly assume that R and R + R
120
contain only positive numbers. Then either R or R contains only positive elements, and
by commutativity of addition we may assume that R contains only positive elements. From
assumptions we have x (y + z) = ( , R (R + R )). If a R, b R and c R , then
a(b+c) = ab+ac by distributivity in Q, which proves that R(R +R ) RR +RR .
Suppose that in addition R contains only positive numbers. Let r1 , r2 R, r R ,
r R . Then (r1 r +r2 r )/(r +r ) is a rational number. If (r1 r +r2 r )/(r +r ) < r1 ,
then by compatibility of multiplication with < on Q, r1 r + r2 r < r1 r + r1 r ,
whence 0 < (r1 r2 ) r , which forces r1 > r2 . By the same reasoning then necessarily
(r1 r + r2 r )/(r + r ) r2 , whence in any case (r1 r + r2 r )/(r + r ) min {r1 , r2 }.
It follows by Exercise 3.7.1 that (r1 r + r2 r )/(r + r ) R, so that the arbitrary element
r1 r +r2 r of RR +RR is in R(R +R ). This proves that R(R +R ) = RR +RR .
Thus in this case x (y + z) = ( , R R + R R ) = ( , R R ) + ( , R R ) x y + x z.
Now suppose that R does not contain only positive numbers. Then L contains
only positive numbers. Thus by the previous case x (y + z) + x (z) = x ((y + z) + (z)).
By the already established associativity of addition, this is x y. But by Lemma 3.7.11 this
says that x (y + z) = x y + x z.
Certainly Q is a subset of R, and the addition and multiplication on R restrict to the
familiar addition and multiplication on Q.
The Dedekind cuts as in Proposition 3.7.5 are the familiar rational numbers, and if
we establish some good order on R, we should be able to order the other Dedekind cuts
among the rational ones as expected, to get that R consists of the familiar real numbers.
3.7.1 Let (L, R) be a Dedekind cut, l L, r R, and x Q. Prove the following:
i) If x l, then x L.
ii) If x r, then x R.
3.7.2 Prove that there exists no R Q such that (Q {0}, R) is a Dedekind cut.
Similarly, there exists no L Q such that (L, Q+ {0}) is a Dedekind cut.
3.7.3 Let (L, R) be a Dedekind cut.
i) If m Q is the supremum of L, prove that R = Q \ (L {m}). If L does not have
a supremum in Q, prove that R = Q \ L.
ii) If m Q is the infimum of R, prove that L = Q \ (R {m}). If R does not have
an infimum in Q, prove that L = Q \ R.
3.7.4 Prove that ({x Q : x2 < 2 or x < 0}, ({x Q : x2 2 and x 0}) is a Dedekind
cut. Prove that it is not of the form Dm = ({x Q : x < m}, {x Q : x > m}) for
any rational number m. (Hint: 2 is not the square of a rational number, as was proved
on page 21.)
121
3.7.5 If m and n are rational numbers, prove that Dm + Dn = Dm+n .

3.7.6 Let a = ({x Q : x < 0 and x2 > 2}, {x Q : x2 2 or x > 0}), b = ({x Q : x <
0 or x2 < 3}, {x Q : x2 3 and x > 0}). Compute a + b, a b.
3.7.7 Prove Theorem 3.7.10.
3.7.8 Prove that if (L1 , R1 ), . . . , (Ln, Rn ) are finitely many Dedekind cuts, then
(nk=1 Lk , nk=1 Rk ) is a Dedekind cut. Furthermore, this Dedekind cut equals (Lk , Rk ) for
some k {1, . . . , n}.
3.7.9 Prove that multiplication on Dedekind cuts is well-defined.
3.7.10 Let (L, R) be a Dedekind cut. Prove that there exists z Q such that z R and
z L. Prove in addition that for any a Q, there exists such z with z |a|.
3.7.11 For any set S Q, define S+ = S Q+ and S = S Q . Prove that for any
Dedekind sets (L, R), (L , R ),
(L, R)(L, R ) = (L R+
+R+ L +L+ L+ +R R
, L L +R L+ +L+ R
+R+ R+
),
where it is understood that A + = A and A = .
3.7.12 Prove that every positive real number is the square of some real number. (Use
Dedekind cuts.)
3.7.13 Let x, y R.
i) Prove that x2 = (x)2 .
ii) Suppose that x2 = y 2 . Prove that either x = y or x = y.
*3.7.14 Let x, y R such that x3 = y 3 . Prove that x = y.
3.7.15 The standard Dedekind cut in the literature is defined as the pair (L, R) with
the following properties:
(1) L, R are non-empty subsets of Q.
(2) For all l L and all r R, l < r.
(3) For all l L, there exists l L such that l < l .
(4) L R = Q.
What would be the definition of sum of two standard Dedekind cuts? Why does (L1 , R1 ) +
(L2 , R2 ) = (L1 + L2 , R1 + R2 ) not work? Why does (L, R) = (R, L) not work?
3.8
Order on R, the least upper/greatest lower bound theorem
Definition 3.8.1 We impose the following order on R: (L, R) (L , R ) if L L , and

(L, R) < (L , R ) if L ( L .
122
Theorem 3.8.2 R is an ordered field.

Proof. By Theorem 3.7.12, R is a field. Let (L, R), (L, R ), (L, R ) R.
Sets L, L can be in exactly on of the following relationships: L = L , L ( L , L 6 L .
If L = L , then by Proposition 3.7.7, (L, R) = (L , R ). If L ( L , then by definition
(L, R) < (L , R ). If L 6 L , let l L \ L . Let l L . By Exercise 3.7.1 applied to the
Dedekind cut (L, R), l > l and l L. This proves that L L, and since L 6= L, it
follows that L ( L, whence (L , R ) < (L, R). This proves trichotomy of <.
Suppose that (L, R) < (L , R ) and (L , R ) < (L , R ). Then by definition, L(L (L ,
so that L ( L , which says that (L, R) < (L , R ). Thus transitivity holds for <.
Suppose that (L, R) < (L , R ). Then certainly L + L L + L . Suppose for
contradiction that L + L = L + L . Then L + L + (R ) = L + L + (R ), or in other
words, by associativity of addition on Q and by Theorem 3.7.10, L + (Q = L + Q+ . Let
m L \ L. By condition (3) of Dedekind cuts, there exists n L such that m < n. But
then (m + n)/2 = n + (m n)/2 L + Q = L + Q+ , so that (m + n)/2 is smaller than
an element in L and hence is in L. However, (m + n)/2 = m + (n m)/2 is strictly larger
than m, so necessarily in R. This contradicts the disjointness of L and R, and thus proves
that < is compatible with addition.
Finally, suppose that (L, R) < (L , R ) and that D0 < (L , R ). Then D0 < (L , R )
(L, R) = (L , R ) + (R, L) = (L R, R L), so that by definition of multiplication,
((L , R ) (L, R)) (L, R ) = ( , (R L) R ), so that D0 < ((L , R ) (L, R)) (L, R ).
But then by associativity, D0 < ((L , R )(L, R ))((L, R)(L, R )), so that by compatibility of < with addition, (L, R) (L , R ) = D0 + ((L, R) (L , R )) < (L , R ) (L , R ),
which proves that < is compatible with multiplication by positive numbers.
Thus the following terms and facts, developed in Section 2.7, apply to elements of R:
positive, negative, non-positive, non-negative, intervals, absolute value, triangle
inequality, reverse triangle inequality.
The following is a rephrasing of Theorem 2.7.13 in this context: Suppose that a, b R
have the property that |a b| < for all real numbers > 0. Then a = b.
Theorem 3.8.3 (Archimedean property) Let m, n R. If m > 0, there exists p N+
such that n < pm.
Proof. Write m = (L, R), n = (L , R ). Since m > 0, L contains a positive rational
number x. Let y R . Then by the Archimedean property for Q, Theorem 3.6.8, there
exists p N+ such that y < px. Then
n = (L , R )
< Dy (since y R )
123
< Dpx (since y < px)

= pDx (by multiplication in Q and R)
p (L, R) (since x L and compatibility of with multiplication by positive p)
= pm.
Theorem 3.8.4 For any two distinct real numbers there is a rational number strictly
between them.
Proof. Let a, b R with a < b. Write a = (L, R) and b = (L , R ) as Dedekind cuts. Since
a < b, it means that L ( L . Let r L \ L. Then a < Dr < b, which by identification
r = Dr proves the theorem.
One of the main properties that distinguishes R from Q is the following theorem:
Theorem 3.8.5 (Least upper/greatest lower bound theorem) Let T be a non-empty
subset of R that is bounded above (resp. below). Then sup (T ) (resp. inf (T )) exists in R.
Proof. First suppose that T is bounded above. Let U = (L,R)T L. In other words, U is
the union of all the left parts of all the Dedekind cuts in T . Since T is not empty, neither
is U . Now let (U0 , V0 ) R be an upper bound on T . Then by definition L U0 for all
(L, R) T . Thus U U0 . This implies by Proposition 3.7.7 that there exists V Q be
such that (U, V ) is a Dedekind cut. By definition, (U, V ) is an upper bound for T , and this
upper bound is less than or equal to an arbitrary upper bound (U0 , V0 ). This proves that
sup (T ) exists in R.
The proof of infimum is left for Exercise 3.8.2.
Remark 3.8.6 In contrast, note that the non-empty bounded set {x Q : x2 < 2}
has no infimum nor maximum in Q. But this same set, as a subset of R, has infimum
({x Q : x2 > 2}, ) and supremum ( , {x Q+ : x2 > 2}) in R.
3.8.1 Prove that the following conditions are equivalent on Dedekind cuts (L, R), (L , R ):
i) (L, R) (L , R ).
ii) For all l L there exists l L such that l l .
iii) For all l L there exists l L such that l < l . (Hint: condition (3) of Dedekind
cuts.)
3.8.2 Let T be a non-empty subset of R that is bounded below. Let (U, V ) be the Dedekind
cut with U = (L,R)T (R). Prove that (U, V ) is the infimum of T .
124
3.8.3 Let m R be a positive number. Prove that there exists a positive integer N such
that 1/2N < m. (Hint: Exercise 3.6.6.)
3.8.4 Compute the least upper bounds and greatest lower bounds, if they exist, of the
following subsets of R:
i) {1/n : n a positive integer}.
ii) {1/n : n a positive real number}.
iii) {1/n : n > 4}.
iv) {(1)n : n Z}.
v) {2n : n N0 }.
vi) {2n /n : n a positive integer}.
vii) {n/(n + 1) : n N0 }.
3.9
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 3.9.1 Let C = R R (the Cartesian product). Define two binary operations
on C:
(a, b) + (c, d) = (a + c, b + d),
(a, b) (c, d) = (ac bd, ad + bc).
Elements of C are called complex numbers.
The following are all easy to verify, and the reader is encouraged to do so:
(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).
(4) For all (a, b) C, (a, b) + (a, b) = (0, 0). In other words, every element has an
additive inverse. 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).
(6) (1, 0) 6= (0, 0). i.e., 1 6= 0.
b
a
b
a
(7) For all (a, b) 6= (0, 0), ( a2 +b
2 , a2 +b2 ) C and ( a2 +b2 , a2 +b2 )(a, b) = (1, 0). In other
words, every non-zero element has a multiplicative inverse. By Theorem 2.5.10,
the multiplicative inverse is unique.
(8) C is a field. As usual for fields, the additive inverse of x is denoted x, the
multiplicative inverse as x1 = x1 .
125
(9) (0, 1)2 = (1, 0) and (0, 1)2 = (1, 0). In other words, i2 = 1, (i)2 = 1.
(10) If (a, b)2 = (1, 0), then (a, b) = (0, 1) or (a, b) = (0, 1).
Notation 3.9.2 We now switch to the more standard notation for elements of C: (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. In this notation, the additive identity
in C is 0, the multiplicative identity is 1, and i = 0 + 1 i is that number for which
i2 = (i)2 = 1. The additive inverse of a + bi is a bi, and the multiplicative inverse of
35i
35i
3
5i
= 34
34
.
a + bi is aabi
2 +b2 . For example, the multiplicative inverse of 3 + 5i is 32 +52 =
34
Any real number r can be uniquely identified with the complex number (r, 0) = r +
i 0 = r, so that we can think of R as a subset of C. The addition and multiplication on
R are compatible with those on C. In total we have found 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 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). The field C has many excellent properties*
some of which we explore in these notes.
Since C = R R, we can represent complex numbers in the real plane, and we can
label them either with the pair-notation or with the new notation, like so:
(0, 1) = i
(3, 1) = 3 + i
(0, 0) = 0
b
(2, 0.5) = 2 + 0.5i
(1, 0) = 1
b
(1, 0) = 1
(0, 1) = i
Because of this two-dimensional quality of C it is perhaps intuitively obvious that C is not

an ordered field in the sense of Definition 2.7.8. A rigorous proof is left for Exercise 3.9.6.
* 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, quotients.
126
The real numbers all lie on the horizontal axis. This axis is called the real axis. The
vertical axis of numbers whose first entry is 0 is called the imaginary axis.
Definition 3.9.3 The real part of a + bi is Re(a + bi) = a, and the imaginary part is
Im(a + bi) = b. The complex conjugate of a + bi, is a + bi = a bi C.
Note that the complex conjugage is a complex number, but the real and imaginary
parts are both real numbers.
Theorem 3.9.4 Let x, y C. Then
(1) 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).
Facts 3.9.5 So far we have established existence and properties of several functions of
complex numbers:
(1) Inclusion of R into C: f (r) = r.
(2) Identity function on C: idC (x) = x. (Say why this function is different from the
one in the previous part.)
(3) Complex conjugation : C C.
(4) Real part Re( ) : C R.
(5) Imaginary part Im( ) : C R.
(6) Additive inverse ( ) : C C.
(7) Multiplicative inverse ( 1 ) : (C \ {0}) C.
127
(8) Addition + : C C C.
(9) Multiplication : C C C.
(10) Scalar multiplication functions: for any z C, multiplication by z is a function
with domain and codomain C: f (x) = zx.
(11) Polynomial functions: for any a0 , a1 , . . . , an C, the polynomial function f (x) =
a0 + a1 x + a2 x2 + + an xn has domain and codomain C.
(12) Note that division by x2 + 1 is fine if the domain consists of real numbers, but that
this would not be defined for all complex numbers x.
Remark 3.9.6 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.
3.9.1 Write the following elements in the form a + bi with a, b R:
1
.
i) 1+i
.
ii) (4+i)(73i)
(1i)(3+2i)
5
iii) (2 3i) . (Hint: Exercise 1.6.9.)
3.9.2 Let =
3
2
3.9.3 Draw the

i) {x : the
ii) {x : the
iii) {x : the
iv) {x : the
+ 2i . Draw , 2 , 3 , 4 , 5 , 6 , 7 . What do you observe?
following sets in C:
real part of x is 3}.
imaginary part of x is 3}.
product of the real and imaginary parts of x is 3}.
product of the real and imaginary parts of x is 0}.
3.9.4 Let x C such that x2 = 1. Prove that either x = i or x = i.

3.9.5 Prove that the only x C with x2 = 0 is x = 0.
3.9.6 Prove that C is not ordered in the sense of Definition 2.7.8. Justify any facts (such
as that 0 < 1).
3.9.7 Let x, y
i) Prove
ii) Prove
iii) Prove
C.
that x + x = 2 Re x.
that x x = 2i Im x.
that x y + x y = 2 Re x.
3.9.8 Let x be a complex number. Prove that x x is real.
128
3.9.9 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.
3.9.10 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.
3.10
Absolute value in C
We have seen the absolute value function in ordered fields. Even though the field C is
not ordered (see Exercise 3.9.6), nevertheless, the Pythagorean theorem in the plane R R
motivates the natural definition of distance in C:
bi
b
(a, b) = a + bi
Definition 3.10.1 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 and is | |.
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 3.9.6: 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.
Theorem 3.10.2 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 x1 = x/|x|2 .
(6) |xy| = |x| |y|.
(7) (Triangle inequality) |x y| |x| + |y|.
(8) (Reverse triangle inequality) |x y| ||x| |y||.
129
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 2.7.11, a2 , b2 0, so by Exercise 2.7.12, 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| |a|. Similarly | Im x| |a|. This proves (4).
If x 6= 0, then by (4), |x| is a non-zero (real, complex) number, and by (3), 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 3.9.4)

= xxyy (by associativity and commutativity of in C)
= xyxy (by Theorem 3.9.4)
= 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.

Finally,
2
|x y| = (x y)(x y)
= (x y)(x y)
= xx xy yx + yy
= |x|2 xy xy + |y|2
= |x|2 2 Re(xy) + |y|2
|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 2.7.12 also (8).
130
So far we have expressed complex numbers with pairs of real numbers either in orderedpair notation (a, b) or in the form a + bi. We refer to this formulation as Cartesian
coordinates, named after Rene Descartes (15961650). We next introduce an alternate
expression, namely the polar coordinates: every point x in C is uniquely determined by
its distance |x| from 0 = (0, 0) and the angle measured counterclockwise from the positive
real axis to the point along the circle of radius |x| and center (0, 0). We draw a few
examples:
radius 1, angle /2
b
b
radius 1, angle /4
radius 2, angle
b
radius 2, angle 0
radius 1, angle 3/2

b
radius 2, angle 5/3
For further examples, 1+i2 3 is on the unit circle centered

at the origin and is at angle
1i 3
/3 counterclockwise from the positive real axis, 2 is on the same unit circle and at
3
is on the same circle
angle /3 counterclockwise from the positive real axis, and 1+i
2
and at angle 2/3 counterclockwise from the positive real axis.
It may be worth remarking that while the complex number is uniquely determined by
its length and the angle from the positive real axis, the number does not uniquely define the
angle: namely, angles and + 2 counterclockwise from the positive real axis determine
the same complex number. So the angle 3/2 = 2 /2 on the picture above can also be
labelled simply as /2, and the angle 5/3 as /3.
Note that the angle is /2 precisely when Re x = 0, that the angle is 0 when x is a
positive real number, that it is when x is a negative real number, et cetera. Furthermore,
if the angle is not /2, then the tangent of this angle is precisely Im x/ Re x.
Theorem 3.10.3 (Fun fact) Let z be a complex number, of length |z| and with angle
counterclockwise from the positive real axis. (If z = 0, arbitrarily set = 0.) Define
functions M, S, R : C C as follows:
M (x) = zx = Multiply x by z,
S(x) = |z|x = Stretch x by a factor of |z|,
R(x) = Rotate x by angle counterclockwise with center at (0, 0).
131
Then
M = S R = R S,
or in other words, multiplication by z is the same as stretching by |z| followed by or preceded
by rotating by the angle 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(|z|x) = |z|R(x) = S R(x). So it suffices to prove that
R S = M.
The angle of M (1) = z is and its length is |z|. But (R S)(1) = R(|z|) also has
length |z| and angle , 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:
(c1 , d1 )
b
(d1 , c1 )
(c2 , d2 )
b
(d4 , c4 )
(d2 , c2 )
b
(c3 , d3 )
b
(c4 , d4 )
(d3 , c3 )
The complex number M (i) = (d, c) has length equal to |(c, d)| = |z|. The angle between
(c, d) and (d, c) is 90 , or /2 radians, and more precisely, to get from z = (c, d) to
M (i) = (d, c) we have to rotate counterclockwise by /2. Thus the angle formed by M (i)
counterclockwise from the positive real axis is + /2. But (R S)(i) = R(|z|i) 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)
= |z|R(x)
= |z|R(a + bi)
= |z|(aR(1) + bR(i))
= a|z|R(1) + b|z|R(i)
= aS(R(1)) + bS(R(i))
132

= aM (1) + bM (i) (by previously proved cases)
= az1 + bzi
= z(a + bi)
= M (x).
Theorem 3.10.4 For any complex non-zero number x and any integer n, the angle of xn
counterclockwise away from the positive x-axis is n times the angle of x.
Proof. If n = 1, this is trivially true. Now suppose that the theorem is true for some
positive integer n. Then the angle of xn1 counterclockwise away from the positive x-axis
is n 1 times the angle of x, and by Theorem 3.10.3, the angle of xn = xxn1 is the sum
of the angles of x and xn1 , so that it is n times the angle of x. Thus by induction the
theorem is proved for all positive n.
Still keep n positive. Since 1 = xn xn has angle 0 and xn has has angle n times the
angle of x, by Theorem 3.10.3, xn must have angle n times the angle of x. 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.
Thus for example, since 1+2 3i is on the unit circle at angle 2/3 away from the
positive x-axis, we have that the second power of 1+2 3i is on the unit circle at angle 4/3
away from the positive
x-axis, and the cube power is on the unit circle at angle 2, i.e., at
1+ 3i 3
angle 0, so that ( 2 ) = 1.
Corollary 3.10.5 Let n be a positive integer. Let A be the set of all complex numbers
2
2
2
on the unit circle at angles 0, 2
n , 2 n , 3 n , . . . , (n 1) n . Then A equals the set of all the
complex number solutions to the equation of xn = 1.
Proof. Let a A. By the previous theorem, an has length 1 and angle an integer multiple
of 2, so that an = 1. If b C satisfies bn = 1, then |b|n = |bn | = 1, so that the nonnegative real number |b| equals 1. Thus b is on the unit circle. If is its angle, then the
angle of bn = 1 is by the previous theorem equal to n, so that n must be an integer
multiple of 2. It follows that is an integer multiple of 2
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.
The absolute value allows the definition of bounded sets in C (despite not having an
order on C):
133
Definition 3.10.6 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 /3} is not bounded.
(Draw these sets.)
3.10.1 Write the following elements in the form a + bi with a, b R:
i) Complex number of length 1 and at angle /4 counterclockwise from the positive
real axis.
ii) Complex number of length 1 and at angle /4 counterclockwise from the positive
real axis.
iii) The product and the sum of the numbers from the previous two parts.
3.10.2 Draw the following points in the real plane (= complex line), and think about
Theorem 3.10.3:
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).
3.10.3 Draw the following sets in C:
i) {x : the angle of x counterclockwise from the positive real axis is /3}.
ii) {x : |x| = 3}.
iii) {x : |x 2 + i| = 3}.
3.10.4 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.
3.10.5 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.
3.10.6 Let a, b C. Suppose that for all real numbers > 0, |a b| < . Prove that a = b.
(Hint: Theorem 2.7.13.)
134
3.10.7 Prove that for any non-zero z C there exist exactly two elements in C whose
square equals z. (Hint: Exercise 3.7.12 and Theorem 3.10.3.)
3.10.8 Square the pentagon drawn below. Namely, estimate the coordinates (real, imaginary 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 3.10.3.)
3.10.9 (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.
3.11
Topology on the constructed fields
When reading this section, absorb the following main points: open ball, open set, limit
point, closed set. Everything else is for extra enjoyment. 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, series, and so on.
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
135
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 3.11.1 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 3.11.1).
Examples 3.11.2
(1) B(a, r) is open.
(2) F = aF 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 50).
(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 : Re x < 1, Im x < 2} is open in C. Namely, this set is the
union aA B(a, min {1 Re x, 2 Im x}.
Theorem 3.11.3 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 3.11.4 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.
136
Definition 3.11.5 Let A be an arbitrary subset of F and a F . If there exists r > 0

such that B(a, r) A, then a is called an interior point of A. If for every r > 0,
B(a, r) A 6= and B(a, r) (F \ A) 6= , then a is called a boundary point of A.
The set of all interior points of A is denoted Int A, and the set of all boundary points
of A is denoted Bd A.
Note that Int A is automatically a subset of A.
Examples 3.11.6
(1) Int F = F , Bd F = .
(2) Int = , Bd = .
(3) Int {a} = , Bd {a} = {a}.
(4) Int Z = , Bd Z = Z.
(5) Int Q = , Bd Q = R. This shows that Bd A need not be a subset of A.
(6) Int (R \ Q) = , Bd (R \ Q) = R.
Theorem 3.11.7 Int B(a, r) = B(a, r).
Proof. It suffices to prove that every point b of B(a, r) is an interior point. Namely,
|b a| < r. Let x B(b, r |b a|). Then by triangle inequality |x a| = |x b + b a|
|x b| + |b a| < (r |b a|) + |b a| = r, so that x B(a, r). We just proved that
B(b, r |b a|) B(a, r), which proves that b Int A.
Definition 3.11.8 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.
Theorem 3.11.9 Any interior point of A is a limit point of A.
Proof. Let a Int A. Then there exists r > 0 such that B(a, r) A. Now let s be
an arbitrary positive real number. By Theorem 3.11.3, B(a, min {r, s}) contains infinitely
many points, and they are all elements of A, so that B(a, min {r, s}) contains at least one
point of A different from a. But B(a, min {r, s}) B(a, s), so that B(a, s) contains at least
one point of of A different from a. Thus a is a limit point of A.
Theorem 3.11.10 If a is a limit point of A, then a A Bd A.
Proof. To prove that a A Bd A, it suffices to prove that if a 6 A, then a Bd A. So
assume that a 6 A. Since a is a limit point of A, by definition for all positive real numbers
s > 0, B(a, s) contains a point of A other than a. Thus B(a, s) A is not empty. But also
B(a, s) (F \ A) is not empty because a is in this intersection. Thus a Bd A.
137
Examples 3.11.11
(1) If A = {a}, then Bd A = {a}, but 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.
In the rest of the section we develop more topology. Most of it is not needed in the rest
of the book, but since I started to talk about topology, a section would not be complete
without the rest of this section.
Theorem 3.11.12 A set A is open if and only if A = Int A.
Proof. Suppose that A is open. We need to prove that A = Int A. Since we already know
that Int A is always a subset of A, we only need to prove that A Int A. Let a A.
Since A is open, it is a union of open balls, so that a is in one of those balls. Namely,
a B(b, r) A for some b A and positive real number r. Then |a b| < r. Set
d = r |a b|. Then d is a positive real number. If y B(a, d), then by the triangle
inequality, |y b| = |y a + a b| |y a| + |a b| < d + |a b| = r, so that y B(b, r).
Thus B(a, d) B(b, r) A, which proves that the arbitrary element a of A is in Int A.
Now suppose that A = Int A. Then for all a A there exists ra > 0 such that
B(a, ra ) A. Then A aA B(a, ra ) A, so that equality holds and A is a union of
open balls, so that A is open.
Theorem 3.11.13 (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 = aF 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. We need to prove that A1 An is open, i.e.,
by using Theorem 3.11.12 we have to prove that every point in this intersection is in the
interior of the set. Let a A1 An . Since each Ai is open, there exists a positive
real number ri such that B(a, ri ) Ai . Set r = min {r1 , . . . , rn }. Then r is a positive real
number, and by construction B(a, r) B(a, r1 ) B(a, rn ) A1 An . Thus a is
in the interior of the intersection of these finitely many sets. Since a was arbitrary in the
intersection, this proves that the intersection is open.
Note that an arbitrary intersection of open sets need not be open:
n=1 B(a, 1/n) =
{a}, which is not open.
138
Definition 3.11.14 A set A is a closed set if Bd A A.

Theorem 3.11.15 A is open if and only if F \ A is closed.
Proof. Suppose that A is open. Let x Bd (F \ A). We need to prove that x 6 A. By
assumption for all r > 0, B(x, r) (F \ A) is not empty, so that no ball centered at x is
contained in A. Hence x 6 Int A, and since A is open, then x 6 A.
Now suppose that F \ A is closed. Let x A. We need to prove that x Int A.
If x 6 Int A, then for every positive real number r, B(x, r) is not contained in A. Thus
B(x, r) contains both points of F \ (F \ A) = A (namely x) and points of F \ A. Thus
x Bd (F \ A), so by assumption x F \ A, which contradicts the assumption x A.
Thus x must be in Int A.
The following is now almost immediate from previous results:
Theorem 3.11.16 (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 ).
kI
kI
kI
kI
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 3.11.2). Some sets are neither open or closed (see
Exercise 3.11.1).
Definition 3.11.17 For any set A, its closure A is the smallest closed set containing A.
Theorem 3.11.18 A = A Bd A.
Proof. Certainly A A by definition. If x 6 A, then x is in the open set F \ A, so there
exists r > 0 such that B(x, r) F \ A F \ A. Thus x 6 Bd A and A Bd A A.
Now we prove that A Bd A is closed. It suffices to prove that the complement is
open. So let x F \ (A Bd A). Since x 6 Bd A, r > 0 such that either B(x, r) A =
or B(x, r) (F \ A) = . The latter case is impossible because x B(x, r) (F \ A). So
necessarily B(x, r) A = . Let y B(x, r). Then by triangle inequality B(y, r |y
139
x|) B(x, r), so that y 6 Bd A. This proves that B(x, r) (A Bd A) is empty, so that
B(x, r) F \ (A Bd A). Since x was arbitrary, this proves that F \ (A Bd A) is open,
so that A Bd A is closed. Thus we have found a closed set A Bd A that contains A
and is contained in every closed set containing A. Thus A Bd A is the smallest closed set
containing A, and thus by the definition A Bd A equals A.
Example 3.11.19 The boundary of the open ball B(a, r) is {x C : |x a| = r}. Thus
the closure of B(a, r) is {x C : |x a| r}.
Proof. Suppose that x C satisfies |x a| = r. Let > 0. Set = min {, r}. Set
y = x + (a x)/(2r), z = x (a x)/(2r). Then |y x| = |z x| = | (a x)/(2r)| =
| /2| < , |y a| = |xa+ (ax)/(2r)| = |(xa)(1 /(2r))| < |xa| = r, |z a| =
|x a (a x)/(2r)| = |(x a)(1 + /(2r))| > |x a| = r, so that y B(x, ) B(a, r),
and z B(x, ) (C \ B(a, r)). This proves that x Bd B(a, r).
Now let x C. Suppose that |xa| < r. Then r |xa| > 0. Let y B(x, r |xa|).
Then by triangle inequality |ya| = |yx+xa| |yx|+|xa| < (r|xa|)+|xa| = r,
so that y B(a, r). This proves that B(x, r |x a|) B(a, r), which means that
B(x, r |x a|) (C \ B(a, r)) 6= , so that x 6 Bd B(a, r).
Now suppose that |x a| > r. Then |x a| r > 0. Let y B(x, |x a| r). Then by
reverse triangle inequality |y a| = |y x + x a| |x a| |y x| > r (r |x a|) =
|x a| > r, so that y C \ B(a, r). This proves that B(x, |x a| r) C \ B(a, r), which
means that B(x, |x a| r) B(a, r) 6= , so that x 6 Bd B(a, r).
We just proved that if x C such that |x a| =
6 r, then x 6 Bd B(a, r), so that
Bd B(a, r) {x C : |x a| = r}. By the first paragraph then Bd B(a, r) = {x C :
|x a| = r}.
3.11.1 Let A be B(0, 1) be an open ball in R. 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.
3.11.2 Let F be either R or C, and let A be a closed and open subset of F . Prove that
A = or A = F .
3.11.3 Prove that Q is neither an open nor a closed subset of R or C.
3.11.4 Sketch the following subsets of C. Determine their interior and boundary sets, the
set 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+ }.
140
3.11.5 For each of the following intervals as subsets of R, determine the interior and
boundary sets, the set of limit points, and whether the sets are open, closed, or neither:
(0, 1), [0, 1], [3, 5), [3, ).
3.11.6 Find the boundary and the set of limit points of {x C : Re x < 0}.
3.11.7 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.
3.11.8 Give examples of sets A B C and a C such that a is a limit point of B but
not of A.
3.11.9 Prove that the closure of B(a, r) is {x F : |x a| r}.
3.11.10 Prove that the closure of { n1 : n N+ } is {0} { n1 : n N+ }.

3.11.11 Prove that the closure of Q in R and in C is R.
3.11.12 Let B = {a + bi : a, b Q}. Determine the closure of B in C.
3.11.13 Let A B F .
i) Prove that Int A Int B.
ii) Prove that (Bd A) \ B Bd B.
iii) Find examples of A and B such that Bd A 6 Bd B.
3.11.14 Recall the definition of bounded sets Definition 2.7.2. Let A R be bounded
below (resp. above). Prove that inf A A (resp. sup A A.
*3.11.15 Prove that a set A is closed if and only if A = A.
3.11.16 Let A be a subset of F and a F \ Int A. Prove that a is a limit point of A if

and only if a is a boundary point of A \ {a}.
3.11.17 Let A be a subset of R and let a complex number a be a limit point of A. Prove
that a R.
3.12
Closed and bounded sets, and open balls
Closed and bounded sets in C and R have many excellent properties we will for
example see in Section 5.2 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 5.4) needs the fairly
technical Heine-Borel theorems proved in this section. (I go very lightly through this
section.)
Section 3.12: Closed and bounded sets, and open balls
141
Theorem 3.12.1 (Heine-Borel theorem in R) Let A be a closed and bounded subset

of R, and for each c A let c be a positive number. Then there exists a finite subset S of
S
A such that A cS B(c, c ).
Proof. First suppose that A = [a, b]. Let D be the subset of all x A such that [a, x] is
contained in the union of only finitely many B(c, c ). Note that [a, a] = {a} B(a, a ),
so that D contains a and is not empty. Also, all elements of D are in A, so that D is
bounded above by b. Thus by the Upper bound theorem (Theorem 3.8.5), d = sup D is a
real number. Necessarily d [a, b], so that by assumption there is a positive number d .
Since d = sup D, there exists e D such that d e < d . By the definition of D, the
interval [a, e] is contained in the union of finitely many B(c, c ), so that by throwing in also
the one ball B(d, d ), the interval [a, d + d /2] [a, b] is contained in the union of finitely
many balls. Thus all the points of [a, d + d /2] [a, b] are in D, and since d = sup D, it
follows that necessarily d = b C. This proves the theorem in case A = [a, b].
Now let A be an arbitrary closed and bounded set in R. Since A is bounded, there
exists M in R such that A [M, M ]. For each x [M, M ] \ A, since A is closed, there
exists x > 0 such that B(x, x ) A = . Thus we now have x for all x [M, M ]. By
the first case, there exists a finite subset S of [M, M ] such that [M, M ] sup B(c, c ).
cS
If x A, there exists c S such that x B(c, c ). But then by the choice of c for
c [M, M ] \ A, necessarily c A. Thus A sup B(c, c ). Since S is finite, so is S A,
cSA
and this proves the theorem.

The next result is a generalization of the previous one: we consider intervals in R as
closed and bounded subsets of C, and we even allow vertical translations of such intervals,
and we then prove a more general result:
Theorem 3.12.2 Let A be a closed and bounded subset of R, let y R, and think of
B = A {y} as a subset of C. For each b B let b be a positive number. Then there exist
> 0 and a finite subset S [M1 , M2 ] {u} such that A [y , y + ] cS B(c, c ).
Proof. To each c A we set c = (c,y) /2. By Theorem 3.12.1, there exists a finite subset
S of A such that A cS B(c, c ). Let = 12 min {c : c S }. Then > 0 since S is
finite. Let x A [y , y + ]. Write x = (a, z) for some a A and z [y , y + ]. By
the choice of S there exists c S such that a B(c, c ). Then
|x (c, y)| = |(a, z) (c, y)|
p
= (a c)2 + (z y)2
p
< c2 + 2
p
c2 + c2
142

=
2c
< (c,y) ,
which proves that a [y , y + ] cS B((c, y), (c,y)). Now set S = S {y}.
Theorem 3.12.3 (Heine-Borel theorem in C) Let A be a closed and bounded subset
of C. For each c A let c be a positive number. Then there exists a finite subset S of A
such that A cS B(c, c ).
Proof. First assume that A = [M1 , M2 ] [N1 , N2 ]. Let C be the subset of all c [N1 , N2 ]
for which the conclusion of the theorem holds for [M1 , M2 ] [N1 , c]. By Theorem 3.12.2,
C contains N1 , so that C is not empty. Since C is bounded above by N2 , by the Upper
bound theorem (Theorem 3.8.5), c = sup C is a real number. Necessarily c [N1 , N2 ].
By Theorem 3.12.2, there exist 1 > 0 and a finite subset S1 [M1 , M2 ] {c} such
that [M1 , M2 ] [c 1 , c + 1 ] aS1 B(a, a ). Similar reasoning as in the proof of
Theorem 3.12.1 shows that c = N2 . This proves the theorem in case A is a rectangle.
Now let A be an arbitrary closed and bounded set in C. By boundedness A
[M1 , M2 ] [N1 , N2 ] for some numbers M1 , N2 , N1 , N2 R. For each x in this rectangle that is not in A, since A is closed, there exists x > 0 such that B(x, x ) A = .
Thus we now have x for all x [M1 , M2 ] [N1 , N2 ]. By the first case, there exists a finite
set S [M1 , M2 ] [N1 , N2 ] such that for all [M1 , M2 ] [N1 , N2 ] aS B(a, a ). Similar
reasoning as in the proof of Theorem 3.12.1 shows that A aSA B(a, a ), and S A is
a finite subset of A.
3.12.1 Let A = [2, 3] R. For each a A let a = 1/a. Does there exist a finite set S as
in Theorem 3.12.1? If yes, find it, if no, explain why not.
3.12.2 Let A = {1/n : n N+ } R. For each a A let a = a/3. Does there exist as in
Theorem 3.12.1? If yes, find it, if no, explain why not. Repeat with A = N+ .
3.12.3 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 kI Uk . Prove that there exists a finite
S
subset K of I such that A kK Uk . (Hint: Prove that for each a A there exists
a > 0 such that B(a, a ) is in some Uk . Apply Theorem 3.12.3.)
*3.12.4 Let A be a closed and bounded subset of R or C, and for each a A let a be a
positive number. Prove that there exist a finite subset S of A and a positive real number
such that A cS B(c, c ) and such that for all x A there exists c S such that
B(x, ) B(c, c ).
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 3.11.8). The limit of f (x) as x
approaches a is the complex number L if for every real number > 0 there exists a
real number > 0 such that for all x A, if 0 < |x a| < , then |f (x) L| < .
When this is the case, we write lim f (x) = L. Alternatively, in order to not make
xa
lines crowded with subscripts, we can write limxa 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 behaviour of the function f at points near and nearer a.
We can give a simple geometric picture of this in case the domain and codomain are
subset of R (refer to Remark 3.9.6 for why we cannot draw functions when domains and
codomains are subsets of C). Here is the picture: on the graph of y = f (x) we cover the
vertical line x = a, and with that information, we conclude that limxa f (x) = L.
L
a
The function f itself might be any of the following:
144
L
a
L
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 = 10 of the predicted value. Or when instruments get better, perhaps gets
smaller, say one thousandth. Or a new material is discovered which allows us to get
even smaller. But no matter what is determined ahead of time, for the prediction to be
believable, we need to determine a range x within a of a but not equal to a, for which
the f -values are within the given 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 there exists a positive such
that for all x in the domain with 0 < |x a| < (the x 6= a in the vertical gray band), the
value of f (x) is within of L (in the horizontal grey band):
L+
L
L
a a a+
Clearly if gets smaller, has to get smaller too. (But if gets larger, we may keep
the old .)
While these pictures can help our intuition, they do not constitute a proof: the definition 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
145
want to replace them with some shortcuts. We will have to prove that those shortcuts
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.
x3
Proof. The function that takes x to 4x 5 is a polynomial function, so it is defined for all
complex numbers. So the domain of the function is C and all points of the domain are limit
points of the domain. Let > 0. [We are proving that for all real numbers > 0
something-or-other holds. Recall that all proofs of this form start with
Let 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-orother claims that there exists a real number > 0 with a certain property.
Thus we have to construct such a . In this first example we simply present
a that works, but in subsequent examples we will show how to find a
working . In general, the proof should contain a specification of and
that it works, but how one finds that is in general left for scratch work
or inspiration. ] Set = /4. [Again: never mind how this magic /4 appears
here; wait until the next example where we present the process of finding
.] Then is a positive real number. [Now we have to prove that for all x, if
0 < |x 3| < , then |f (x) L| < . 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| < , then |f (x) L| < . The proof of If P then Q starts with
Assume P .] Assume that 0 < |x 3| < . [Now we have to prove Q, i.e., we
have to prove that |f (x) L| = |(4x 5) 7| < . We do not simply write
|(4x 5) 7| < because we do not know that yet. We write the left
side of this inequality, and manipulate it algebraically, with triangle
inequalities... until we get < . ] Then
|(4x 5) 7| = |4x 12|
= 4|x 3|
< 4
= 4 [Wasnt this a clever choice of ?]

4
= .
In the next example we have no magic appearing, but we do the work in deriving a
working form for it. It suffices to find any that works.
146
Example 4.1.3
lim (4x2 5x + 2) = 28.
x2
Proof. The function that takes x to 4x2 5x + 2 is a polynomial function, so it is defined

for all complex numbers, so that 2 is a limit point of the domain. Let > 0. Set
[ to be determined still; the final write-up will have
=
this filled in, but as we are writing the proof, we do not know yet!]. Then
is a positive real number. [Hoping, anyway.] Let x be any complex number. Suppose
that 0 < |x + 2| < . Then
|(4x2 5x + 2) 28| = |4x2 5x 26| [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 the coefficient 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 : lets make sure that is at most 1 (or 2, or 15,

it does not matter what positive number you pick in this example).]
[So on the line write: = min {1,
}.]
= |4(x + 2 2) 13| |x + 2| (adding a clever zero)
= |4(x + 2) 21| |x + 2|
(|4(x + 2)| + 21) |x + 2| (by triangle inequality |a b| |a| + |b|)

< (4 1 + 21) |x + 2| (since |x + 2| < 1)
= 25 |x + 2| [Goal #2 accomplished: constant B times |x a|]
[Now go back to and finish with: = min {1, /B}.]

< 25
25 /25
= .
The final version of the proof of lim (4x2 5x + 2) = 28 then looks like this:
x2
The function that takes x to 4x2 5x + 2 is a polynomial function, so 2 is a limit

point of the domain. Let > 0. Set = min {1, /25}. Then is a positive real number.
Let x be any complex number. Suppose that 0 < |x + 2| < . Then
|(4x2 5x + 2) 28| = |4x2 5x 26|
= |(x + 2)(4x 13)|
= |4x 13| |x + 2|
147
= |4(x + 2 2) 13| |x + 2| (adding a clever zero)
= |4(x + 2) 21| |x + 2|
(|4(x + 2)| + 21) |x + 2| (by triangle inequality |a b| |a| + |b|)
< (4 1 + 21) |x + 2| (since |x + 2| < 1)
= 25 |x + 2|
< 25
25 /25
= .
We repeat the same type of discovery work on the next example, with a few fewer
explicit comments:
Example 4.1.4
lim (4/x2 ) = 4.
x1
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 > 0. Set =
. Then is a positive
real number. Let x be any complex number that satisfies 0 < |x + 1| < . Then

4 4x2
2

|4/x 4| =
x2

(1 x)(1 + x)

= 4

x2

1 x
= 4 2 |1 + x|[Goal #1 accomplished: something times |x a|.]
x
[Now we want something to be at most a 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 of 1, in order to avoid 0 we need to make sure that

is strictly smaller than 1. For example, make sure that 0.4. Thus, on
the line write: = min {0.4,

}.]

2 (x + 1)
|1 + x| (by rewriting 1 x = 2 (x + 1))
= 4

|x|2
2 + |x + 1|
4
|1 + x| (by triangle inequality)
x2
2.4
4 2 |1 + x| (since 0.4)
|x|
9.6
=
|1 + x| (by rewriting x = x + 1 1)
|x + 1 1|2
148
9.6
|1 + x| (by reverse triangle inequality
0.62
|x + 1 1| 1 |x + 1| > 1 1 0.4 = 0.6,
so that 1/|x + 1 1|2 < 1/0.62)
[On the line now write: = min {0.4, 0.62/9.6}.]

9.6
<
0.62
9.6
0.62 /9.6
0.62
= .
And here is another example for good measure, with already filled in:
4x3 +x
x2i 8xi
Example 4.1.5 lim
= 2.
+x
Proof. The domain of 4x
8xi consists of all complex numbers different from i/8, so 2i is a
limit point of the domain. Let > 0. Set = min {1, /9}. Let x be any complex number
different from i/8 such that 0 < |x 2i| < . [Here we merged: Let x be any
complex number different from i/8. Let x satisfy 0 < |x 2i| < . into one
shorter and logically equivalent statement Let x be any complex number
different from i/8 such that 0 < |x 2i| < .] Then
3

3
4x + x

4x + x

=
(2)
+
2
8x i

8x i
3

4x + x + 16x 2i

=

8x i
3

4x + 17x 2i

=

8x i

(4x2 + 8ix + 1)(x 2i)
[If x 2i were not a factor, then
=

8x i
either our limit or our algebra
would have been wrong.]

(4x2 + 8ix + 1)
|x 2i| (Goal #1: x a is a factor.)
=

8x i
|4x2 | + |8ix| + 1
|x 2i| (by triangle inequality)
|8x i|
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 triangle and reverse triangle inequality)
149
4(1 + 2)2 + 8(1 + 2) + 1

|x 2i| (since |x 2i| < 1)
8 + 15
61
|x 2i|
=
7
< 9
9/9
= .
We now give an example where the limit is at a non-specific a.

Example 4.1.6 lim (x2 2x) = a2 2a.
xa
Proof. Any a is a limit point of the domain of the given polynomial function. Let > 0.
Set = min {1, /(1 + |2a 2|)}. Let x satisfy 0 < |x a| < . 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 triangle inequality)
(1 + |2a 2|) |x a| (since |x a| < 1)
< (1 + |2a 2|)
(1 + |2a 2|)/(1 + |2a 2|)

= .
Remark 4.1.7 Note that depends on and a, which are constants in the problem; is
not allowed to depend on x, as the definition goes:
for all > 0 there exists > 0 such that for all x, etc
so that x depends on , but not on x.
Another reason why is not allowed to depend on x is that by the definition of limits, is
supposed to be a positive real number, not a function of x. (See also Exercise 4.1.9.)
Example 4.1.8 lim 2x 6 = 0.

x3
Proof. The domain here is all x 3. So 3 is a limit point of the domain. Let > 0. Set
= 2 /2. Let x > 3 satisfy 0 < |x 3| < . Then

2x 6 = 2 x 3

< 2
150

=
= .
p
2 /2
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 4.1.9 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 rightsided (resp. left-sided) limit of f (x) as x approaches a is L if for every real number
> 0 there exists a real number > 0 such that for all x A, if 0 < x a < (resp.
if 0 < a x < ) then |f (x) L| < . When this is the case, we write limxa+ f (x) =
lim+ f (x) = L (resp. limxa f (x) = lim f (x) = L).
xa
xa
With this, Example 4.1.8 can be phrased as lim
x3+
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 > 0. Set = 2 /2. Let x satisfy 0 < x 3 < . Then

2x 6 = 2 x 3

< 2
p
= 2 2 /2
= .
Thus, the two proofs are almost identical. Note that lim
x3
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 limxa f (x) does not exist. Below
is one example.

x2 + 4, if x > 1;
Example 4.1.10 Let f : R R be given by f (x) =
Then
x 2, if x 1.
lim f (x) = 5, lim f (x) = 3.
x1+
x1
Proof. Let > 0.

Set = min {1, /3}. Let x satisfy 0 < x 1 < . Then
f (x) 5 = x2 + 4 5 (since x > 1)
= x2 1
= (x + 1)(x 1)
< (x + 1) (since x > 1, so x + 1 is positive)
151
= (x 1 + 2) (by rewriting)
< (1 + 2) (since x 1 < 1)

3/3
= .
This proves that lim+ f (x) = 5.

x1
Set = . Let x satisfy 0 < 1 x < . Then

f (x) (3) = x 2 + 3 (since x < 1)
=x1
<
= .
This proves that lim f (x) = 3.
x1

4.1.1 Fill in the blanks of the following proof that limx2 (x2 3x) = 2. Explain why
none of the inequalities can be changed into equalities.
so
. Let > 0. Set =
. Let x satisfy
that 2 is a
0 < |x 2| < . Then

|(x2 3x) (2)| = x2 3x + 2
= |x 1| |x 2|
(because
< (3 + 1) |x 2|
(because
(|x 2| + 1) |x 2|
< 4
4
4
= .
(because
(because
(because
4.1.2 Determine the following limits and prove them with the epsilon-delta proofs.
i) lim (x3 4).
x1
1
.
x2 x
lim xx4
2 +2 .
x3
ii) lim
iii)
iv) lim
x4
x + 5.
x2 9
.
x3 x3
v) lim
152
4.1.3 Rework Example 4.1.3 with choosing to be at most 2 rather than at most 1.
4.1.4 Let b C and f, g : C C with

x3 4x2 , if x 6= 5;
f (x) =
b,
if x = 5,
g(x) =
x3 4x2 ,
b,
if x = 5;
if x =
6 5.
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.
4.1.5 Suppose that a is a limit point of {x A : x > a} and of {x A : x < a}. Prove
that limxa f (x) = L if and only if limxa+ f (x) = L and limxa f (x) = L.
4.1.6 Suppose that a is a limit point of {x A : x > a} but not of {x A : x < a}. Prove
that limxa f (x) = L if and only if limxa+ f (x) = L.
4.1.7 Prove that limxa (mx + l) = ma + l, where m and l are constants.
4.1.8 Let f : R R be given by f (x) =
limx0 f (x) = 1.
x
.
|x|
Prove that limx0+ f (x) = 1 and that
4.1.9 Below you will find an attempt at a proof that limx3 x2 = 9. Explain how the
two starred steps below contain three mistakes.
Let > 0. Set = /|x + 3|. Then
|x2 9| = |x 3| |x + 3|
< |x + 3|
= .
4.1.10 Find a function f : R R such that limx0 f (x) = 2 and such that
limx0+ f (x) = 5.
4.1.11 Find a function f : R R such that limx0 f (x) = 2, limx0+ f (x) = 5,

limx1 f (x) = 3, and limx1+ 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 limxa f (x) = L means that a is the limit point of the domain of f , and
that for all real numbers > 0 there exists a real number > 0 such that for all x in the
domain of f , if 0 < |x a| < then |f (x) L| < . [Think of limxa 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
Q R.]
153
Thus if limxa f (x) 6= L, then either a is not the limit point of the domain of f or else
it is not true that for all real numbers > 0 there exists a real number > 0 such that for
all x in the domain of f , if 0 < |x a| < then |f (x) L| < . [This simply says that
P is the same as (Q) (R).]
In particular,
lim f (x) 6= L and a is a limit point of the domain of f
xa
means that it is not true that for all real numbers > 0 there exists a real number > 0
such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < . [This says
that (P ) Q is the same as 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. By the usual negation rules of compound
statements (see chart on page 324), we successively rewrite this last negation into a form
that is easier to handle:
For all real numbers > 0 there exists a real number > 0 such that

for all x in the domain of f , if 0 < |xa| < then |f (x)L| < .
[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 > 0 such that there exists a real number
> 0 such that for all x in the domain of f , if 0 < |x a| <

then |f (x) L| < .
[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 > 0 such that for all real numbers
> 0, for all x in the domain of f , if 0 < |x a| < then

|f (x) L| < .
[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 > 0 such that for all real numbers > 0,
there exists x in the domain of f such that if 0 < |x a| <

then |f (x) L| < .
[Negation of If P then Q is P and not Q. Hence the following rephrasing:]
there exists x in the domain of f such that 0 < |x a| < and

|f (x) L| < .
154

there exists x in the domain of f such that 0 < |x a| < and
|f (x) L| .
In summary, we just proved that:
Theorem 4.2.1 If a is a limit point of the domain of f , then limxa f (x) 6= L means that
there exists a real number > 0 such that for all real numbers > 0, there exists x in the
domain of f such that 0 < |x a| < and |f (x) L| .
x
as x approaches 0 does not exist. In other words, for all
Example 4.2.2 The limit of |x|
x
complex numbers L, limx0 |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
is the set of all non-zero complex
Proof. The domain of the function that takes x to |x|
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 = 1. Let > 0 be an
arbitrary positive number. Let x = /2 if Re(L) 0, and let x = /2 otherwise. Then
0 < |x| = |x 0| < . If Re(L) 0, then

x
/2
Re
L = Re
L = 1 Re(L) 1,
|x|
| /2|
x
L| 1 = , and if Re(L) < 0, then
so that | |x|

x
/2
Re
L = Re
L = 1 Re(L) > 1,
|x|
|/2|
x
L| > 1 = . This proves the claim of the example.
so that again | |x|
i
Example 4.2.3 For all L C, limx2 x2
6= L.
A geometric reason for the non-existence of this limit is that as x gets closer to 2 (but
i
gets larger and larger.
not equal to 2), the size of x2
Proof. Set = 1. Let > 0 be an arbitrary positive number. Set = min {, 1/(|L| + 1)}.
Let x = 2 /2. Then 0 < |x 2| < , and

2i

i
= L

L

x 2
155

2i
|L| (by reverse triangle inequality)
2(|L| + 1) |L| (since 1/(|L| + 1))

1
= .
Example 4.2.4 For f : R R given by the graph below, lim f (x) does not exist because
x2
of the jump in the function at 2.
2
1
Here is an epsilon-delta proof. Say that the limit exists. Call it L. Set = 41 . Let

be an arbitrary positive number. If L 32 , set x = 2 + min 41 , 2 , and if L < 32 , set
x = 2 min { 41 , 2 }. In either case,

1
0 < |x 2| = min
< .
,
4 2
2
If L 32 , by our choice x = 2 + min { 41 , 2 }, so that f (x) = 1 + min { 41 , 2 } 1 + 14 ,
whence |f (x) L| 41 = . Similarly, if L < 23 , by our choice x = 2 min { 41 , 2 }, so that
f (x) = 2 min { 41 , 2 } 2 14 = 1 + 43 , whence |f (x) L| 14 = . Thus no L works, so
the limit of f (x) as x approaches 2 does not exist.
4.2.1 Prove that limx3 (3x 4) 6= 3.
4.2.2 Prove that limx1 (x2 + 4) 6= 5.
4.2.3 Prove that limx3
x3
x2 9
does not exist.
156
4.2.4 Prove
that for all a R, limxa 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 limx0 ( x x) does not exist. (Hint: 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 for many functions
efficiently establish the limits, and remove the need for the time-consuming epsilon-delta
proofs. Nevertheless, to prove these theorems we will need epsilon-delta methods.
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. If L1 6= L2 ,
then |L1 L2 | is a positive real number. Thus for each i = 1, 2, there exists i > 0 such
that for all x in the domain of f , if 0 < |x a| < i , then |f (x) Li | < |L1 L2 |/2.
Set = min {1 , 2 }. Then is a positive real number. Let x in the domain of f satisfy
0 < |x a| < . Then
|L1 L2 | = |L1 f (x) + f (x) L2 | (by adding a clever 0)
|L1 f (x)| + |f (x) L2 | (by triangle inequality)
< 2|L1 L2 |/2 (since 1 , 2 )

= |L1 L2 |,
which says that |L1 L2 | is strictly smaller than itself. But this contradicts trichotomy
on R. Thus the assumption L1 6= L2 must have been wrong, which proves the theorem.
Theorem 4.3.2 Let a be a limit point of the domain of a function f with limxa f (x) = L.
Suppose that L 6= 0. Then there exists > 0 such that for all x in the domain of f , if
0 < |x a| < , then |f (x)| > |L|/2.
In particular, there exists > 0 such that for all x in the domain of f , if 0 < |xa| < ,
then f (x) 6= 0.
Proof. Set |L|/2 > 0, there exists > 0 such that for all x in the domain of f , if
0 < |x a| < , 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.
157
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 limxa f (x) and limxa g(x) both exist. Then
(1) (Constant rule) lim c = c.
xa
(2) (Linear rule) lim x = a.

xa
(3) (Sum/difference rule) lim (f (x) g(x)) = lim f (x) lim g(x).
xa
xa
xa
(4) (Scalar rule) lim cf (x) = c lim f (x).

xa
xa
(5) (Product rule) lim (f (x) g(x)) = lim f (x) lim g(x).
xa
xa
xa
lim f (x)
f (x)
= xa
.
xa g(x)
lim g(x)
(6) (Quotient rule) If lim g(x) 6= 0, then lim

xa
xa
Proof. Both (1) and (2) were proved in Exercise 4.1.7.

Set L = limxa f (x) and K = limxa g(x).
(3) Let > 0. Then /(|c| + 1) is a positive number. (Note that we did not divide
by 0.) Since L = limxa f (x), there exists > 0 such that for all x A, 0 < |x a| <
implies that |f (x) L| < /(|c| + 1). Hence for the same x,
|cf (x) cL| = |c| |f (x) L| < |c| /(|c| + 1) < ,
which proves that limxa cf (x) = cL = c limxa f (x).
(4) Let > 0. Since L = limxa f (x), there exists 1 > 0 such that for all x A, if
0 < |xa| < 1 , then |f (x)L| < /2. Similarly, since K = limxa g(x), there exists 2 > 0
such that for all x A, if 0 < |x a| < 2 , then |g(x) L| < /2. Set = min {1 , 2 }.
Then is a positive number. Let x A satisfy 0 < |x a| < . Then
|(f (x) g(x)) (L K)| = |(f (x) L) (g(x) K)| (by algebra)
|f (x) L| + |g(x) K| (by triangle inequality)

< /2 + /2 (because 0 < |x a| < 1 , 2 )
= ,
which proves (4).

(5) Let > 0. Then min {/(2|L| + 2), 1}, min {/(2|K| + 1), 1} are positive numbers.
Since L = limxa f (x), there exists 1 > 0 such that for all x A, if 0 < |x a| < 1 ,
then |f (x) L| < min {/(2|K| + 1), 1}. Similarly, since K = limxa g(x), there exists
2 > 0 such that for all x A, if 0 < |x a| < 2 , then |g(x) L| < /(2|L| + 2). Set
= min {1 , 2 }. Then is a positive number. Let x A satisfy 0 < |x a| < . Then by
triangle inequality,
|f (x)| = |f (x) L + L| |f (x) L| + |L| < 1 + |L|,
158
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 triangle inequality)
= |f (x)| |g(x) K| + |f (x) L| |K| (by factoring)
+
|K| (since 1 , 2 )
< (1 + |L|)
2|L| + 2 2|K| + 1
< /2 + /2
= .
(6) Let > 0. Since K 6= 0, by Theorem 4.3.2, there exists 0 > 0 such that for all
x A, if 0 < |x a| < 0 , then |g(x)| > |K|/2.
The numbers |K|/4, |K|2 /(4|L| + 1) are positive numbers. Since L = limxa f (x),
there exists 1 > 0 such that for all x A, if 0 < |x a| < 1 , then |f (x) L| < |K|/4.
Similarly, since K = limxa g(x), there exists 2 > 0 such that for all x A, if 0 < |xa| <
2 , then |g(x)L| < |K|2 /(4|L|+1). Set = min {0 , 1 , 2 }. Then is a positive number.
Let x A satisfy 0 < |x a| < . Then

f (x)K Lg(x)
f (x)
L

(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 triangle inequality)
Kg(x)

f (x) L |L| K g(x)
+

(by factoring)
=
g(x) |K| g(x)
|K| 2
|L| |K|2
2
<
(since 0 , 1 , 2 )
4
|K| |K| 4|L| + 1 |K|
< /2 + /2
= .
This proves (6) and thus the theorem.
Theorem 4.3.4 (Power rule for limits) Let n be a positive integer. If limxa f (x) = L,
then limxa 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)n1 f (x) (by algebra)
xa
xa
n1
=L
= Ln .
L (by induction assumption and by the product rule)
159
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, limxa 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,
limxa x = a. Hence by the power rule, for all i = 1, . . . , n, limxa xi = ai . By the
constant rule, limxa ci = ci , so that by the product rule limxa 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).
xa
xa
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 , limxa 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 2.4.12, 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, limxa g(x) = g(a) and limxa h(x) = h(a) 6= 0. Thus by the
quotient rule, limxa f (x) = g(a)/h(a) = f (a).
Theorem 4.3.7 (Absolute value rule for limits) For all a C, limxa |x| = |a|.
Proof. This function is defined for all complex numbers, and so every a C is a limit point
of the domain. Let > 0. Set = . Then for all x C, by the reverse triangle inequality,

|x| |a| |x a| < = .
Theorem 4.3.8 (Real and imaginary parts of limits) Let f : A C be a function, let a be a limit point of A, and let L C. Then limxa f (x) = L if and only if
limxa Re f (x) = Re L and limxa Im f (x) = Im L.
Proof. First suppose that limxa f (x) = L. Let > 0. By assumption there exists > 0
such that for all x A, 0 < |x a| < implies that |f (x) L| < . Then for the same x,
| Re f (x) Re L| = | Re(f (x) L)| |f (x) L| < ,
and similarly | Im f (x) Im L| < which proves that limxa Re f (x) = Re L and
limxa Im f (x) = Im L.
Now suppose that limxa Re f (x) = Re L and limxa Im f (x) = Im L. Let > 0. By
assumptions there exist 1 , 2 > 0 such that for all x A, 0 < |x a| < 1 implies that
160
| Re f (x) Re L| < /2 and 0 < |x a| < 2 implies that | Im f (x) Im L| < /2. Let
= min {1 , 2 }. Then > 0, and if 0 < |x a| < , then
|f (x) L| = | Re f (x) + i Im f (x) Re L i Im L|
| Re f (x) Re L| + |i|| Im f (x) Im L|

< /2 + /2 = ,
which proves that limxa f (x) = L.

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
xa
and that lim g(x) = g(L). Then lim h(x) = lim (g f )(x) = g(L).
xa
xL
xa
Proof. Let > 0. Since lim g(x) = g(L), there exists 1 > 0 such that for all x in the
xL
domain of g, if 0 < |xa| < 1 then |g(x)g(L)| < . Since lim f (x) = L, there exists > 0
xa
such that for all x in the domain of f , if 0 < |x a| < then |f (x) L| < 1 . Thus for the
same , if x is in the domain of h and 0 < |xa| < , then |h(x)g(L)| = |g(f (x))g(L)| <
because |f (x) L| < 1 .
Theorem 4.3.10 Suppose that f, g : A R, that a is a limit point of A, that limxa f (x)
and limxa g(x) both exist, and that for all x A, f (x) g(x). Then
lim f (x) lim g(x).
xa
xa
Proof. Let L = lim f (x), K = lim g(x). Let > 0. By assumptions there exists > 0
xa
xa
such that for all x A, if 0 < |x a| < , then |f (x) L|, |g(x) K| < /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 triangle inequality)
> /2 /2 = .
Since this is true for all > 0, by Theorem 2.7.13, 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 limxa f (x) and
limxa h(x) both exist and are equal, then limxa g(x) exists as well and
lim f (x) = lim g(x) = lim h(x).
xa
xa
xa
161
Proof. [If we knew that limxa g(x) existed, then by the previous theorem,
limxa f (x) limxa g(x) limxa h(x) = limxa f (x) would give that the three
limits are equal. But we have yet to prove that limxa g(x) exists.]
Let L = limxa f (x) = limxa h(x). Let > 0. Since limxa f (x) = L, there
exists 1 > 0 such that for all x, if 0 < |x a| < 1 then |f (x) L| < . Similarly,
since limxa h(x) = L, there exists 2 > 0 such that for all x, if 0 < |x a| < 2 then
|h(x) L| < . Now set = min {1 , 2 }. Let x satisfy 0 < |x a| < . Then
< f (x) L g(x) L h(x) L < ,
where the first inequality holds because 1 , and the last inequality holds because 2 .
Hence < g(x) L < , which says that |g(x) L| < , so that limxa g(x) = L.
4.3.1 Determine the following limits by invoking appropriate results:
i) lim (x3 4x 27), lim (x2 + 5).
x2
x2
(x3 4x 27)3
.
ii) lim
x2
(x2 + 5)2
|x3 4x 27|
iii) lim
.
x2
(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.
xa
xa
n
xa
4.3.3 Prove that lim x = a for any non-zero complex number a and any integer n.
xa
4.3.4 Prove that for any positive real number a and any whole number n, lim xn = an .
xa
4.3.5 Prove that if limxa f (x) = L, then limxa |f (x)| = |L|. Give an example of a
function such that limxa |f (x)| = |L| and limxa f (x) does not exist.
4.3.6 Find functions f, g such that limxa (f (x) + g(x)) exists but limxa f (x) and
limxa g(x) do not exist. Does this contradict the sum rule?
4.3.7 Find functions f, g such that limxa (f (x)g(x)) exists but limxa f (x) and
limxa g(x) do not exist. Does this contradict the product rule?

x
x
x
x
and lim 1 |x|
do not exist, but that lim |x|
+ 1 |x|
4.3.8 Prove that lim |x|
x0
x0
x0
does exist. Does this contradict the sum rule for limits? Justify.
x
x0 |x|
4.3.9 Prove that lim
|x|
x0 x
and lim
do not exist, but that lim
this contradict the product rule for limits? Justify.
x0
x
|x|

|x|
does exist. Does
x
162
4.3.10 The following information is known about functions f and g:

a
0
1
2
3
4
f (a)
1
1
not defined
4
5
lim f (x)
xa
2
0
3
4
1
g(a)
6
5
6
not defined
4
lim g(x)
xa
4
5
6
5
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)).
x0
ii) lim (f (x) + g(x)).

x4
iii) lim (f (x) g(x)).

x3
f (x)
.
x2 g(x)
lim g(x) .
x4 f (x)
iv) lim
v)
vi) lim (g f )(x).

x1
vii) lim (g f )(x).

x4
4.3.11 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 limxa g(x) = L, then limxa f (x) = L.
In particular, if a R is a limit point of A = B R and if limxa g(x) = L, then the
restriction of g to A has the same limit point at a.
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 limxa f (x) exists if and only if limxa g(x) exists,
and in they both exist, then the two limits are equal.
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.
163
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 > 0 such that for all x A, if 0 < |x a| < then
f (x) > M . We write this as lim f (x) = limxa f (x) = . Similarly we say that the
xa
limit of f (x) as x approaches a is if for every real number M < 0 there exists a
real number > 0 such that for all x A, if 0 < |x a| < then f (x) < M . We write this
as lim f (x) = limxa f (x) = .
xa
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 > 0
such that for all x A, if 0 < x a < then f (x) < M . This is written as lim f (x)
xa+
= limxa+ 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) = .
xa
xa+
xa
Note that we cannot use epsilon-delta proofs: no real numbers are within 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.)
Example 4.4.2 limx0
1
|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 = 1/ M . Let x
satisfy 0 < |x 0| < , i.e., let x satisfy 0 < |x| < . Then
1
1
> 2 (because 0 < |x| < )
2
|x|
= M.
Example 4.4.3 limx5+
x+2
x2 25
= .
Proof. Certainly 5 is the limit point of the domain of the given function. Let M > 0. Set
7
= min {1, 11M
}. Let x satisfy 0 < x 5 < . Then
5+2
x+2
> 2
(because x > 5 and x2 25 > 0)
2
x 25
x 25
164

1
7
(by algebra)
(x + 5) x 5
1
7
(because 0 < x 5 < 1, so 0 < x + 5 < 11, and x 5 > 0)

>
11 x 5
7 1
>
(because 0 < x 5 < )
11
1
7
7
7
7 (because 11M
, so 1/ 1/( 11M
))
11 11M
= M.
=
Example 4.4.4 limx0+
1
x
= .
Proof. Let M > 0. Set = 1/M . Then for all x with 0 < x 0 < ,
Example 4.4.5 limx0
1
x
1
x
> 1/ = M .
= .
Proof. Let M < 0. Set = 1/M . Then is a positive number. Then for all x with
0 < 0 x < ,
1
< 1/ = M.
x
Example 4.4.6 We conclude that limx0 x1 cannot be a real number, and it cannot be
either or . Thus limx0 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
limxa f (x) = L R, limxa g(x) = , limxa 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;
xa
xa
0,
if c = 0;
0,
if c = 0.
(2) (Sum/difference rule)
lim (f (x) g(x)) = ,
xa
lim (f (x) h(x)) = ,
xa
lim (g(x) h(x)) = ,
xa
lim (h(x) g(x)) = ,
xa
lim (g(x) + h(x)) = not enough information.
xa
(3) (Product rule)

lim (f (x) g(x)) =
xa
,
,
not enough information,
if L > 0;
if L < 0;
if L = 0,

(
,
lim (f (x) h(x)) = ,
xa
lim (g(x) h(x)) = .
165
if L > 0;
if L < 0;
if L = 0,
xa
(4) (Quotient rule)

f (x)
= 0,
g(x)
f (x)
lim
= 0,
xa h(x)
(
,
g(x)
lim
= ,
xa f (x)
(
,
h(x)
= ,
lim
xa f (x)
g(x)
lim
= not enough information.
xa h(x)
lim
xa
if
if
if
if
if
if
L > 0;
L < 0;
L = 0,
L > 0;
L < 0;
L = 0,
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
limx0 g(x) = , and similarly that limx0 h1 (x) = limx0 h2 (x) = limx0 h3 (x) =
limx0 h4 (x) = . However,
lim (g(x) + h1 (x)) = lim 0 = 0,
x0
x0
lim (g(x) + h2 (x)) = lim 17 = 17,
x0
x0
lim (g(x) + h3 (x)) = lim (1/x4 ) = ,
x0
x0
but limx0 (g(x) + h4 (x)) = limx0 (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 limxa f (x) = 0, if f is
a constant function, then limxa (f (x)g(x)) = 0, but if f is not a constant function, then
we do not have enough information for limxa (f (x)g(x)) = 0.
4.4.1 Write the definitions of
i) limxa+ f (x) = .
ii) limxa f (x) = .
iii) limxa f (x) = .
166
4.4.2 Prove that limx0+

1 = .
x
1
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 limx0 g(x) = , limx0 f1 (x) = limx0 f2 (x) = limx0 f3 (x) = limx0 f4 (x) = 0.
i) Compute the limits limx0 (f1 (x)g(x)), limx0 (f2 (x)g(x)), limx0 (f3 (x)g(x)),
limx0 (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 > 0 there exists a real
number N > 0 such that for all x > N , |f (x) L| < .
For f : (, a) C, we say that limx = L if for every > 0 there exists a real
number N < 0 such that for all x < N , |f (x) L| < .
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 .
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 x > N M 1/4 )
= M.
Section 4.5: Limits at infinity

Example 4.5.3 limx
x5 16x4
x5 +4x2
167
= 1.
Proof. Let > 0. Set N = max {1, 20/}. 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)
5
x
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/)
20/
= .
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) = .
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 3x2x4x+1
= 3/2.
2 +2
3x3 4x+1
= .
2x2 +2
2
limx 3x2x4x+1
= 0.
3 +2
9x2 +4
limx x+2 = 3.
2 +4
= 3.
limx 9x
x+2
ii) limx
iii)
iv)
v)
4.5.5 For which rational functions f does limx f (x) = 0 hold? Justify.
168
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.
Continuous functions from an interval in R to R are the ones that we can graph
without 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 functions 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 5.2),
existence of radical functions (Section 5.3), and the new notion of uniform continuity
(Section 5.4).
5.1
Continuous functions
Definition 5.1.1 A function f : A B is continuous at a A if for all real numbers

> 0 there exists a real number > 0 such that for all x A, |x a| < implies that
|f (x) f (a)| < . 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 limxa f (x) = f (a).
Proof. In the first case, there exists > 0 such that B(a, ) A = {a}. Thus by definition,
the only x A with |x a| < is x = a, whence |f (x) f (a)| = 0 is strictly smaller than
an arbitrary positive . 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 limxa f (x) = f (a). Let > 0. Since f is continuous at a, there exists > 0
such that for all x A, if |x a| < , then |f (x) f (a)| < . Hence for all 0 < |x a| < ,
|f (x) f (a)| < , and since a is a limit point of A, this proves that limxa f (x) = f (a).
Finally suppose that limxa f (x) = f (a). We need to prove that f is continuous at a.
Let > 0. By assumption there exists > 0 such that for all x A, if 0 < |x a| < then
170
|f (x) f (a)| < . If x = a, then |f (x) f (a)| = 0 < , so that for all x A, if |x a| < ,
then |f (x) f (a)| < . Thus f is continuous at a.
Easy applications 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, yield the next theorem easy.
When the points in question are not limit points of the domain, the results below need
somewhat different 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 function 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.
(7) (The composite function rule) Suppose that f is continuous at a, g is continuous
at f (a), and that a is a limit point of the domain of g f . Then g f is continuous
at a.
Now suppose that f, g : A C, and that f and g are continuous at a. Then
(8) (Scalar rule) For any c C, cf is continuous at a.
(9) (Sum/difference rule) f + g is continuous at a.
(10) (Product rule) f g is continuous at a.
(11) (Quotient rule) If g(a) 6= 0, then f /g is continuous at a.
(12) (Power rule) For any positive integer n, the function that takes x to (f (x))n is
continuous at a.
The theorem covers many continuous functions, but the following function for example
has to be verified differently.
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
f (x) =
x1,Re x>1
x1,Re x1
lim
(x2 4) = 12 4 = 3,
lim
3x3 = 3 13 = 3,
x1,Re x>1
x1,Re x1
Section 5.1: Continuous functions
171
so that limx1 f (x) = 3 = 3 13 = f (1).

Note however that this function is not continuous at 1 + i:
lim
f (x) =
lim
f (x) =
x1+i,Re x>1
x1+i,Re x1
lim
(x2 4) = (1 + i)2 4 = 4 + 2i,
lim
3x3 = 3 (1 + i)3 = 6 6i.
x1+i,Re x>1
x1+i,Re x1
so that even limx1+i f (x) does not exist.

Example 5.1.5 When we restrict the function in the previous examples to R, we get
f : R R defined by

x2 4, if x > 1;
f (x) =
3x3 , if x 1.
As before, f is continuous at 1. Explicitly, by the polynomial rules,
lim f (x) = lim (x2 4) = 12 4 = 3,
x1+
x1+
lim f (x) = lim 3x3 = 3 13 = 3,
x1
x1
so that limx1 f (x) = 3 = 3 13 = f (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;
f (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
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.
Compare the example above to Theorem 5.2.4.
5.1.1 Prove that for any c C, the functions f, g : C C given by f (x) = c + x and
g(x) = cx are continuous.
172
5.1.2 Give an 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
0
1
2
3
4
f (c)
1
1
not defined
4
2
lim f (x)
xc
2
0
4
3
g(c)
3
5
6
not defined
3
lim g(x)
xc
4
5
6
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 Consider the continuous function f in Example 5.1.6. Prove that f has an inverse
function f 1 : (1, 1) (1, 0) (0, 1]. Graph f 1 and prove that f 1 is not continuous.
5.1.6 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.
5.1.7 Assume that the trigonometric functions sine and cosine are continuous everywhere.
(We will prove this in Theorem 9.8.3.) 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 9.8.3.) Define a function g : R R as

sin(1/x), if x 6= 0;
g(x) =
0,
if x = 0.
i) Prove that f is not continuous at 0.
ii) Prove that f is continuous at all other points.
5.1.9 Assume that the trigonometric function sine is continuous everywhere. (We will
prove this in Theorem 9.8.3.) Define a function g : R R as

x sin(1/x), if x 6= 0;
g(x) =
0,
if x = 0.
i) Is f continuous at 0? Prove.
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.
173
ii) Prove that f is not continuous anywhere else.

The exercises below modify the definition of continuity to get very different types of functions. The moral is that the order of quantifiers and implications is very important!
5.1.11 (This is from page 1177 of the Edward Nelsons article Internal set theory: a new
approach to nonstandard analysis. Bull. Amer. Math. Soc. 83 (1977), no. 6, 11651198.)
A function f : A B is suounitnoc at a A if for all real numbers > 0 there exists a
real number > 0 such that for all x A, |x a| < implies that |f (x) f (a)| < .
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.
5.1.12 A function f : A B is ticonnuous at a A if there exists a real number
> 0 such that for all real numbers > 0 and for all x A, |x a| < implies that
|f (x) f (a)| < .
i) Suppose that f is ticonnuous at some a A. Prove that there exist a real number
> 0 and b B such that for all x A, |x a| < 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.
5.1.13 A function f : A B is connuousti at a A if for all real numbers > 0 there
exists a real number > 0 such that for all x A, |f (x) f (a)| < implies that |x a| < .
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.
174
5.2
Intermediate Value Theorem, Extreme Value Theorem
In this section, all functions are real-valued. The reason is that the content is about
numbers between other numbers or about largest possible numbers; whereas R is an ordered
field (Theorem 3.8.2), C is not (Exercise 3.9.6).
Theorem 5.2.1 (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 yaxis 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.
y
f (b)
b
k
a
b
c1
c2
f (a)
Proof of Theorem 5.2.1: The idea of the proof is to construct a1 , a2 , . . . and b1 , b2 , . . ., such
that a a1 a2 an bn b2 b1 b, bn an = (b a)/2n , and either
k is f applied to one of these numbers, or else for any positive integer n, k is (strictly)
between f (an ) and f (bn ). In the latter case we define c = sup {a1 , a2 , . . .}, and we prove
that f (c) = k.
Define a0 = a, b0 = b. Once we have an1 , bn1 [a, b] with k strictly between f (an1 )
and f (bn1 ), we construct either c (an1 , bn1 ) with f (c) = k or else we construct a new
pair an , bn using the following directions. Let d = (an1 + bn1 )/2. So d is the midpoint
of the interval [an1 , bn1 ]. If f (d) = k, we set c = d and we are done. Instead if f (d) 6= k,
then by trichotomy either f (d) < k or f (d) > k. We know that the signs of f (an1 ) k
and f (bn1 ) k are opposite; let e be that one among an1 , bn1 for which the sign of
f (e) k is opposite from the sign of f (d) k. Then k is strictly between f (d) and f (e).
Now let the smaller of the numbers d, e be called an , and the larger of the two be called
bn . Then by construction an1 an < bn bn1 and |bn an | = 21 |bn1 an1 |.
With this procedure, either we find c in finitely many steps, or else we construct
175
intervals
[a, b] = [a0 , b0 ] [a1 , b1 ] [a2 , b2 ] [an1 , bn1 ] [an , bn] [an+1 , bn+1 ]
such that for all n, bn an = (b a)/2n and k is strictly between f (an ) and f (bn ).
By the inclusions of the intervals, a a1 a2 an bn b2 b1 b,
so that each b1 , b2 , . . . , bn , . . . is an upper bound on the set {a1 , a2 , . . .}. Thus by the Upper
bound theorem Theorem 3.8.5, the set {a1 , a2 , a3 , . . .} has a supremum c in R, and
a a1 a2 an c bn b2 b1 b.
Let > 0. Since f is continuous, it is continuous at c, so there exists > 0 such that
for all x [a, b], if |x c| < , then |f (x) f (c)| < /3. By Exercise 3.8.3, there exists a
positive integer n such that 1/2n < /(b a). Then |an c| |an bn | = (b a)/2n < , so
that |f (an ) f (c)| < /3. Similarly, |f (bn) f (c)| < /3. Hence by the triangle inequality,
|f (bn) f (an)| < 2/3. But k is between f (an ) and f (bn ), so that |f (an) k|, |f (bn) k| <
2/3. But then
|f (c) k| |f (c) f (an )| + |f (an ) k| < ,
and since this true for all > 0, it follows by Theorem 2.7.13 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. The next section remedies this gap.)
Example 5.2.2 There exists a real number c such that c5 4 =
c2 2
c2 +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 5.2.1 there exists c in (0, 2) such that
2
f (c) = 0. In other words, c5 4 = cc2 2
+2 .
Theorem 5.2.3 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 (Theorem 5.2.1), any real number between c and d is in the image of f , which proves the
theorem.
Compare the theorem below to Exercise 5.1.5.
176
Theorem 5.2.4 Let I be an interval in R, and let f : I R be continuous. If f has

an inverse, then f 1 is continuous and f, f 1 are either both strictly increasing or both
strictly decreasing.
Proof. Let a f (b). For now we assume that f (a) < f (b).
Let x < y < z be in I. If f (y) is not between f (x) and f (z), then by the order on R,
either f (x) or f (z) is between the other two. In the former case, an application of the
Intermediate Value Theorem gives c (y, z) (so c 6= x) such that f (c) = f (x), and the
latter case gives c (x, y) (so c 6= y) such that f (c) = f (z), and both of these contradict f s
invertibility. So necessarily f (y) is between f (x) and f (z). In particular, since f (a) < f (b),
it follows that f (x) < f (y) < f (z), so that f is strictly increasing. By Theorem 5.2.3, the
domain of f 1 is an interval, and clearly f 1 is strictly increasing.
Now we prove that f 1 is a continuous function at an arbitrary y be in the domain
of f 1 . Let > 0. Let c = f 1 (y) (so that y = f (c)). Let z (c, c + ) I. (If c is the
right endpoint of I, then there is no such z, and the discussion below is vacuous.) Since f
is increasing, y = f (c) < f (z). Let 1 = f (z) y. Let x (y, y + 1 ) = (f (c), f (z)). By the
Intermediate Value Theorem x is in the range of f , i.e., 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 + . Thus |f 1 (x) f 1 (y)| < .
This proves that for the given > 0 there exists a positive number 1 such that for any
x (y, y + 1 ) I, |f 1 (x) f 1 (y)| < . Similarly, there exists 2 > 0 such that for all
x (y 2 , y) I, |f 1 (y) f 1 (x)| < . Now set = min {1 , 2 }. By what we just
proved, for all x in the domain of f 1 , if |x y| < , then |f 1 (x) f 1 (y)| < . Thus
f 1 is continuous at y, and since y was arbitrary, f 1 is continuous.
Now 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.
Theorem 5.2.5 (Extreme Value Theorem) Let A be a closed and bounded subset of
C, and let f : A R be a continuous function. Then there exists 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. Since f is continuous, for each a A there exists a > 0 such that for all x
A B(a, a ), |f (x) f (a)| < 1. By Theorem 3.12.1 or Theorem 3.12.3 there exists a finite
S
subset S of A such that A aS B(a, a ).
177
Thus {|f (a)| + 1 : a S} is a finite subset of R+ , so it has an upper bound M . We

claim that the image of f lies in the bounded interval [M, M ]. Namely, if x A, then
x B(a, a ) for some a S, so that by triangle inequality, |f (x)| |f (x) f (a)| + |f (a)| <
1 + |f (a)| M . This proves the claim that for all x A, M f (x) M .
Thus the set {f (x) : x A} is a bounded subset of R, so that its infimum L and
supremum U are real numbers. Suppose that for all a A, f (a) 6= L. Then for every
a A, f (a)L > 0, so there exists a > 0 such that for all x AB(a, a ), |f (x)f (a)| <
(f (a) L)/2. By Theorem 3.12.1 or Theorem 3.12.3 there exist a finite subset S of A such
S
that A aS B(a, a ). Since S is finite, the set {(f (a) + L)/2 : a S} is finite, hence
bounded, so that K = min {(f (a) + L)/2 : a S} is a real number. Furthermore, for each
a, f (a) > L so that K > L. For all x A, there exists a S such that x B(a, a ). By
definition of a , |f (x) f (a)| < (f (a) L)/2, so that
f (x) = f (x) f (a) + f (a)
f (a) L
+ f (a)
2
f (a) + L
=
2
K.
Since this is true for all x A, it follows that K is a lower bound for {f (x) : x A} so
that K L. But L < K, which gives a contradiction. Thus L = f (l) for some l A.
Similarly, there exists u A such that f (u) = U .
Example 5.2.6 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.
5.2.1 Let f (x) = x3 + 4, g(x) = x4 + x2 . Prove that there exists c [4, 4] such that
f (c) = g(c).
5.2.2 Assuming that cos is a continuous functions, prove that there exists a real number x
such that cos x = x.
5.2.3 Let f : R Q be a continuous function. Prove that f is a constant function. Let
g : C Q be a continuous function. Prove that g is a constant function.
178
5.2.4 (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.
5.2.5 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).
5.2.6 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 |, . . . , |an1 |, |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 + + an1 (b)n1 |
|a0 | + |a1 b| + |a2 b2 | + + |an1 bn1 | (by triangle inequality)

|an |b |an |b2
|an |b3
|an |bn
+
+
++
(by the choice of b)
n
n
n
n

|an |b
1 + b + b2 + + bn1
n

|an |b n1
b
+ bn1 + bn1 + + bn1 (because b 1)
n
0
n1
(place markers)
|an |b n1
nb
n
= |an |bn
= |an bn |.
iii) Justify all steps below:

|f (b)| = |a0 + a1 (b) + a2 (b)2 + + an1 (b)n1 + an (b)n |
|an (b)n | |a0 + a1 (b) + a2 (b)2 + + an1 (b)n1 |

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).
5.2.7 Let f : (0, 1) R be defined by f (x) = x1 . Prove that f is continuous and that f
does not achieve either its minimum or its maximum. Explain why this does not contradict
the Extreme Value Theorem (Theorem 5.2.5).
179
5.2.8 (Outline of the 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). 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 > 0 such that for all x [a, b]B(c, ), |f (x)f (c)| <
(f (y) f (c))/2. (Hint: use continuity at c.)
Prove that there exists x L B(c, min {, 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 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, xn1 ), . . ., such that > f (xn) > f (xn1 ) > > 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].
5.3
Radical functions
Let n be a positive integer. Define the function f (x) = xn with domain R if n is odd
and domain R0 otherwise. 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 5.2.4, f 1 is strictly increasing and continuous.
180
Definition 5.3.1 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 n is a strictly increasing continuous function for any positive
integer n. By continuity we immediately get the following:
Theorem 5.3.2 (Radical rule for limits) For any positive integer n and any a in the
domain of n ,
lim n x = n a.
xa
As a consequence of the Intermediate Value Theorem, and more specifically, by The

orem 5.2.3, the range of n is an interval in R. We can be more specific about the range:
Theorem 5.3.3 The domain of
Proof. If a = 0, then
is R if n is odd and is R0 otherwise.
n
a = 0. If a > 0,
0n = 0 a < a + 1 (a + 1)n ,
so that since exponentiation by n is continuous, by the Intermediate Value Theorem (The

orem 5.2.1), there exists r (0, a + 1) such that a = r n . Thus n a = r. If a < 0 and n is
odd, then
(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 . Thus n a = r.
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
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 5.3.4 Let r Q. If r 0 let A = R0 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
181
exponentiation by m with exponentiation by 1/n, in either order. Exponentiation by nonnegative 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 5.3.2.
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 composition of a strictly increasing and strictly decreasing function, hence strictly decreasing by
Exercise 2.8.8. Monotonicity of inverses follows from Theorem 5.2.4.
The last part is obvious.
We prove in Theorem 7.5.4 that more generally exponentiation by arbitrary real numbers (not just by rational numbers) is continuous.
5.3.1 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 an = 1/an .
5.3.2 Let f (x) = x x.
i) Prove that for all a in the domain of f , limxa f (x) is inapplicable.
ii) Prove that f is continuous.

5.3.3 Determine the following limits, and justify all steps by invoking the relevant theorems/rules:
p
i) lim x2 3x + 4.
x2
x 2
.
ii) lim
x2 x2 + 4
x 2
iii) lim
.
x2 x2 4
x a
.
iv) If a 0, lim 2
xa x a2
5.3.4 Determine the domain of the function f given by f (x) = x2 .
182
5.3.5 Recall from Exercise 3.10.7 that for every non-zero complex number a there exist
exactly two complex numbers whose squares are a. Lets try to create a 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
limxa f (x) = f (a) for all a C?
p
(4)(9) = 36 = 6,
iii) Explain away the problematic claims 4 9 =
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 be any real number, and let D be the set of all complex numbers whose
counterclockwise angle from the positive real axis is . Prove that we can define
a continuous square root function f : D C.
5.3.6 (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 5.3.4.)
iii) Let > 0. Justify the following:
( + 1)
n+1
1=
n+1
X
i=1
( + 1) ( + 1)
i1
n+1
X
i=1
( + 1)
i1
n+1
X
= (n + 1).
i=1
Use the Archimedean property of R (Theorem 3.8.3) to prove that the set {(+1)n :
n N0 } is not bounded above.
iv) Prove that there exists a positive integer n such that c1/n < + 1.
v) Prove that there exists 1 > 0 such that for all x (0, 1 ) Q, |cx 1| < .
vi) Prove that there exists 2 > 0 such that for all x (2 , 0) Q, |cx 1| < .
vii) Prove that limx0,xQ cx = 1.
viii) Prove that for any r R, limxr,xQ cx exists and is a real number.
183
5.3.7 (Related to the previous exercise.) Let c R+ .

i) Prove that for any r R, limxr,xQ 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
, (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.)
cr =
5.3.8 Here is an alternate proof of Theorem 5.3.2. 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 = n . Let x A satisfy 0 < |x a| < . Since nth root
function is an increasing function on A, it follows that
p

n x n a = n x = n |x| < n = .
iii) Suppose that a > 0. First let = min {, n a}. So is a positive number. Set
n
= min {( + n a)n a, a ( n a )n }. Note that
( + n a)n a > ( n a)n a = 0,
0 n a < n a,
n
a ( n a )n > 0,
so that is positive. Let x A satisfy 0 < |x a| < . Then < x a < and
n
a < x < + a. Since = min {( + n a)n a, a ( n a )n }, it follows that

n
a a ( n a )n a < x < a + a + ( + n a)n a ,
or in other words, ( n a )n < x < ( + n a)n . Since the nth radical function is
increasing on R+ , it follows that
n
a n a < n x < + n a,
whence < n x n a < , so that | n x n a| < , 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 > 0 such

that for all x R, if 0 < |(x) (a)| < , then n x n a < . But then
for x R with 0 < |x a| < , since n is odd,

n x n a = (1) n x n a = n x n a < .
184
5.4
Uniform continuity
Definition 5.4.1 A function f is uniformly continuous if for all real numbers > 0
there exists a real number > 0 such that for all x and y in the domain, if |x y| < ,
then |f (x) f (y)| < .
For example, constant functions are uniformly continuous.
In uniform continuity, given a real number > 0, there exists > 0 that depends only
on that makes some conclusion true, whereas in the definition of continuity at a, given a
real number > 0, there exists > 0 that depends on and on a for the same conclusion
to be true (with y = a). Thus the following is immediate:
Theorem 5.4.2 Every uniformly continuous function is continuous.
The converse is false in general:
Example 5.4.3 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 = 1 there
exists > 0 such that for all x, y A, if |xy| < then |f (x)f (y)| < 1. Set d = min {, 1}.
Then 1/d > d/4. Set x = 1/d d/4 + 1, y = x + d/2. Then |x y| = d/2 < d , and

d
d 2
+ 1 = 1 + > 1,
|f (x) f (y)| = |x y||x + y| =
2 d
2
which contradicts the assumption that f is uniformly continuous. Thus f is not uniformly
continuous.
(With your calculus background, uniform continuity for differentiable functions essentially says that the range of the derivative function is bounded. Examine the veracity of
this on the examples and non-examples so far.)
The converse of Theorem 5.4.2 does hold with some extra assumptions.
Theorem 5.4.4 Let A be a closed and bounded subset of R or C. Let f : A B be
continuous. Then f is uniformly continuous.
Proof. Let > 0. Since f is continuous, for each a A there exists a > 0 such that for
all x A, |x a| < a implies that |f (x) f (a)| < /2. By applying Theorem 3.12.1 when
A R or applying Theorem 3.12.3 otherwise, we get that for a = a , there exists > 0
such that for all x A there exists a A such that B(x, ) B(a, a ). Let x, y A with
|x y| < . Since x, y B(x, ) B(a, a ), it follows that |x a|, |y a| < a . 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)

< /2 + /2 = .

Example 5.4.5 The continuous function
185
is uniformly continuous.
Proof. Let > 0. By the previous theorem there exists 1 > 0 such that for all a, x [0, 2],
if |x a| < 1 then | x a| < .

Now let a, x 1 such that |x a| < . Then

xa

x
+
a
=

| x a| = ( x a)
x + a x + a

|x a|
(because x, a 1)
2
< .
Finally, set = min {1 , , 1}. Let x, a 0 satisfy |x a| < . Since |x a| < 1,
necessarily either x, a 1 or x, a 2. But we have analyzed both cases, and we conclude
that | x a| < .
5.4.1 Let f : C C be given by f (x) = mx + l for some constants m, l. Prove that f is
uniformly continuous.
5.4.2 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|
otherwise.
0,
vi) f : R R, f (x) =
0, otherwise.
vii) Re, Im : C R.
viii) The absolute value function from C to R.
5.4.3 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.
5.4.4 Let g : [a, b] C be continuous. Prove that the function f : (a, b) C defined by
f (x) = g(x) for all x (a, b) is uniformly continuous.
5.4.5 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).
5.4.6 If the composition of two uniformly continuous functions uniformly continuous?

186
5.4.7 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.
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)):
y
f (a)
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
xa
xa
exists. In this case, we call this limit the derivative of f at a, and we use either Newtons

(x)
.
notation f (a) = (f (x))x=a or Leibnizs notation dfdx
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 h0
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 6.2.1.)
188
Example 6.1.2 Let m, l be complex numbers, and let f (x) = mx + l. Then

(mx + l) (ma + l)
m(x a)
f (x) f (a)
= lim
= lim
= m.
xa
xa
xa
xa
xa
xa
f (a) = lim
Alternatively,
f (a + h) f (a)
(m(a + h) + l) (ma + l)
mh
= lim
= lim
= m.
h0
h0
h0 h
h
h
f (a) = lim
Example 6.1.3 Let f (x) = x2 . Then

f (x) f (a)
xa
xa
2
x a2
= lim
h0 x a
(x a)(x + a)
= lim
h0
xa
= lim (x + a)
f (a) = lim
h0
= 2a.
Alternatively,
f (a + h) f (a)
h0
h
2
(a + h) a2
= lim
h0
h
a2 + 2ah + h2 a2
= lim
h0
h
2
2ah + h
= lim
h0
h
(2a + h)h
= lim
h0
h
= lim (2a + h)
f (a) = lim
h0
= 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)
h0
h
1
1
= lim a+h a
h0
h
a
a+h
a(a+h) a(a+h)
= lim
(common denominator in the fractions)
h0
h
f (a) = lim
Section 6.1: Definition of derivatives

= lim
189
aah
a(a+h)
h
h
= lim
h0 a(a + h)h
1
= lim
h0 a(a + h)
1
= 2
a
by the quotient rule for limits.
h0
Example 6.1.5 Let f (x) =
x. The domain of f is R0 , and for all a > 0,
f (a + h) f (a)
h0
h
a+h a
= lim
h0
h
a+h a
a+h+ a
= lim
h0
h
a+h+ a
(a + h) a
(since (x y)(x + y) = x2 y 2 )
= lim
h0 h( a + h +
a)
h
= lim
h0 h( a + h +
a)
1
= lim
h0
a+h+ a
1
=
2 a
f (a) = lim
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 R0 , and for all a 0,
f (a + h) f (a)
h
3/2
(a + h) a3/2
= lim
h0
h
3/2
(a + h) a3/2 (a + h)3/2 + a3/2
= lim
h0
h
(a + h)3/2 + a3/2
(a + h)3 a3
= lim
(since (x y)(x + y) = x2 y 2 )
h0 h((a + h)3/2 + a3/2 )
a3 + 3a2 h + 3ah2 + h3 a3
= lim
h0
h((a + h)3/2 + a3/2 )
f (a) = lim
h0
190
3a2 h + 3ah2 + h3
h0 h((a + h)3/2 + a3/2 )
(3a2 + 3ah + h2 )h
= lim
h0 h((a + h)3/2 + a3/2 )
3a2 + 3ah + h2
= lim
h0 (a + h)3/2 + a3/2
(
3a2
if a > 0;
3/2 +a3/2 ,
a
=
2
limh0 hh3/2 , if a = 0;
3
= a1/2
2
by the linear, radical, composite, and quotient rules for limits. Note that this f is differentiable even at 0, whereas the square root function (previous example) is not differentiable
at 0.
= lim
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 labelled x rather than a:
f (x + h) f (x)
f (z) f (x)
df (x)
= lim
= lim
.
zx
h0
dx
h
zx
The h-limit is perhaps preferable to the last limit, where it is z that varies and gets closer
and closer to x.
f (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|
as h goes to 0 does not exist. Namely, if h varies over positive real
h
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 continuous function with a jagged graph is not differentiable.)
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.
6.1.4 Prove that the square root function is not differentiable at 0.

6.1.5 Let f : R R be given by
f (x) =
Prove that f is differentiable at 1.
f (x) =
x2 1,
x3 x,
191
if x > 1;
if x 1.
x2 1,
if x > 1;
x4 4x, if x 1.
Prove that f is continuous but not differentiable at 1.

f (x) =
x2 ,
x3 x,
if x > 1;
if x 1.
Prove that f is not continuous and not differentiable at 1.

6.1.8 Determine if the following function is differentiable at 1: f (x) =
6.1.9 Let f, g : R R be f (x) =
1
2
2x ,
1 2
2x ,
if x 0;
and g(x) =
if x < 0,
1
x2 , if x > 1;
2x, if x 1.
2
2 x + 3,
1 2
2x ,
if x 0;
if x < 0.
i) Prove that f is differentiable everywhere and that for all x R, f (x) = |x|.
ii) Prove g is not differentiable at 0.
iii) Prove g is differentiable at any non-zero real number x with g (x) = |x|.
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
(a)
exists, by the product rule for limits
domain of f . Furthermore, since limh0 f (a+h)f
h
f (a+h)f (a)
also limh0 (h
) exists and equals 0 f (a) = 0. In other words, limh0 (f (a +
h
h) f (a)) = 0, so that by the sum rule for limits, limh0 f (a + h) = limh0 (f (a + h)
f (a) + f (a)) = 0 + f (a) = f (a). Thus by Exercise 6.2.1, limxa f (x) = f (a), so that f is
continuous at a.
192
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
lim
h0
cf (a + h) cf (h)
f (a + h) f (h)
(cf )(a + h) (cf )(h)
= lim
= c lim
= cf (a),
h0
h0
h
h
h
and (4) follows from the sum rule for limits and from
f (a + h) + g(a + h) f (h) g(h)
(f + g)(a + h) (f + g)(h)
= lim
h0
h0
h
h
f (a + h) f (h) + g(a + h) g(h)
= lim
h0
h

f (a + h) f (h) g(a + h) g(h)
= lim
+
h0
h
h
= f (a) + g (a).
lim
The following proves the product rule (5):

f (a + h) g(a + h) f (a) g(a)
(f g)(a + h) (f g)(a)
= lim
h0
h0
h
h
f (a + h) g(a + h) f (a) g(a + h) + f (a) g(a + h) f (a) g(a)
= lim
(adding 0)
h0
h

(f (a + h) f (a))g(a + h) + f (a)(g(a + h) g(a))
= lim
h0
h

f (a + h) f (a)
g(a + h) g(a))
= lim
g(a + h) + f (a)
h0
h
h
= f (a)g(a) + f (a)g (a),

lim
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
193
to provide the commentary on the steps:

lim
h0
f
g (a
+ h) fg (a)
f (a+h)
g(a+h)
f (a)
g(a)
= lim
h0
h
h
f (a + h)g(a) f (a)g(a + h)
= lim
h0
hg(a + h)g(a)
f (a + h)g(a) f (a)g(a) + f (a)g(a) f (a)g(a + h)
= lim
h0
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
h0
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
h0
hg(a + h)g(a)
hg(a + h)g(a)

g(a + h)
f (a)
g(a + h) g(a)
f (a + h) f (a)
= lim
h0
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 ) = nxn1 .

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 ) = (xn1 x)
= (xn1 ) x + xn1 x (by the product rule)
= (n 1)xn2 x + xn1 (by induction assumption)
= (n 1)xn1 + xn1
= nxn1 ,
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 1.6.9:
(x + h)n xn
h0
h

Pn
n k nk
xn
k=0 k x h
= lim
h0
h
Pn1 n k nk
x h
= lim k=0 k
h0
h
Pn1 n k nk1
h k=0 k x h
= lim
h0
h
(xn ) = lim
194
= lim
h0
n1
X
k=0

n1
X

n k nk1
x h
k
n k nk1
x 0
(by the polynomial rule for limits)
k
k=0

n
xn1
=
n1
= nxn1 .
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 . Then g f is differentiable at a, and (g f ) (a) = g (f (a)) f (a).
Proof. Let > 0. Since f is differentiable at a, there exists 1 > 0 such that for all a + h
(a)
f (a)| < min {1, /(2|g (f (a))| + 2)}.
in the domain of f , if 0 < |h| < 1 then | f (a+h)f
h
(a)
For all such h, by triangle inequality, | f (a+h)f
| < |f (a)| + 1. By assumption g is
h
differentiable at f (a), so that there exists 2 > 0 such that for all x in the domain of g, if
(a))
g (f (a))| < /(2|f (a)| + 2). Since f is differentiable
0 < |x f (a)| < 2 , then | g(x)g(f
xf (a)
and hence continuous at a, there exists 3 > 0 such that for all x in the domain of f , if
|x a| < 3 , then |f (x) f (a)| < 1 . Set = min {1 , 2 }. Let a + h be arbitrary in
the domain of g f such that 0 < |h| < . 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)) f (a)
+ g (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))|
f (a)
h
(|f (a)| + 1) + |g (f (a))|
2|f (a)| + 2
2|g (f (a))| + 2
< +
2 2
= .
195
Thus if there exists as above but possibly smaller such that f (a + h) 6= f (a) for all a + h
in the domain with 0 < |h| < , the above proves the theorem.
Now suppose that for all > 0 there exists h as above such that f (a + h) = f (a).
(a)
Then f (a) = limh0 f (a+h)f
, so that in particular when h varies over those infinitely
h
many h getting closer to 0 with f (a + h) f (a), we get that f (a) = 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)
= 0 < .

g
(f
(a))
f
(a)

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,
Re(f (a + h) f (a)) + i Im(f (a + h) f (a))
f (a + h) f (a)
=
h
h

f (a + h) f (a)
f (a + h) f (a)
+ i Im
,
= Re
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 theorem with Exercises 6.2.5 and 6.2.6.
Theorem 6.2.7 (Derivatives of inverses) Let A, B C, and let f : A B be an
invertible differentiable function whose derivative is never 0. Then for all b B, f 1 is
invertible at b, and
1
(f 1 ) (b) = 1
,
f (f (b))
or in other words, for all a A, f 1 is invertible at f (a) and
(f 1 ) (f (a)) =
f (a)
196
Proof. 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. Since f is never 0, the conclusion follows.
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):
( x) = (f 1 ) (x) =
1
f (f 1 (x))
1
2f 1 (x)
1
= .
2 x
Theorem 6.2.9 Let n be a positive integer. Then for all non-zero x in the domain of
1
( n x) = x1/n1 .
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) = nxn1 , which is never 0. Thus by the previous theorem, f 1 = n is differentiable

with
1 (n1)/n
1
1
1 1/n1
1
=
( n x) =
=
=
x
=
x
.
n
n
f ( x)
n( x)n1
n
n
nx(n1)/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) = rxr1 .
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/n1 m (m1)/n+1/n1
x
= x
= rxm/n1/n+1/n1 = rxr1 .
n
n
This proves the theorem for all rational r.
f (x) = m(x1/n)m1
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 5.3.6 and 5.3.7, and if we were to continue that kind of laborious 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.
197

6.2.1 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
limxa f (x) = L if and only if limzg(a) f (g 1 (z)) = L.
6.2.2 Provide the commentary for the proof of the quotient rule for derivatives in Theorem 6.2.2.
6.2.3 Prove the general power rule for derivatives: If f is differentiable at a, then

(f (x))nx=a = n(f (a))n1f (a).
6.2.4 Prove the general power rule for derivatives: If f is differentiable at a, then for every
positive integer n, f n is differentiable at a, and (f n ) (a) = n(f (a))n1f (a).
6.2.5 Prove that the functions Re, Im : C R are not differentiable at any a.
6.2.6 (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.)
6.2.7 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) =
6.2.8 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.
198
6.2.9 Let f (x) = x1

x2 .
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.)
6.2.10 We assume the following familiar properties of the trigonometric functions: sin is
differentiable on R, sin = cos, and when restricted to [, ], sin has an inverse. Use
Theorem 6.2.7, the fact that (sin(x))2 + (cos(x))2 = 1, and that on [, ] cosine is non1
.
negative to prove that (sin1 ) (x) = 1x
2
6.2.11 Make the requisite assumptions about the trigonometric functions to prove that
1
(cos1 ) (x) = 1x
.
2
6.2.12 Make the requisite assumptions about the trigonometric functions to prove that
1
(tan1 ) (x) = 1+x
2.
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
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.
(x)
. Thus for all x very
Suppose that f (c) > 0. By definition, f (c) = limxc f (x)f
xc
near c but larger than c,
f (x)f (c)
xc
> 0, so that f (x) f (c) > 0, so that f does not achieve
(c)
> 0, so that
its maximum at c. Also, for all x very near c but smaller than c, f (x)f
xc
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
199
that 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 by 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.
Theorem 6.3.2 (Darbouxs 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 5.2.5).
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 (Rolles 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 Extreme Value Theorem (Theorem 5.2.5) 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). Suppose in addition that f (l) 6= f (a). Then
(l)
0
a < l < b. For all x [a, b], f (x) f (l), so that in particular for all x (a, l), f (x)f
xl
(l)
(l)
and for all x (l, b), f (x)f
0. Since f (l) = limxl f (x)f
exists, it must be both
xl
xl
non-negative and non-positive, so necessarily it has to be 0. Now if instead f (l) = f (a),
then by the assumption that f (l) 6= f (u), it follows that f (u) 6= f (u). Then a < u < b.
200
For all x [a, b], f (x) f (u), so that in particular for all x (a, u),
f (x)f (u)
xu
f (x)f (u)
xu
(u)
limxu f (x)f
xu
0 and
for all x (u, b),

0. Since f (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.
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)
(a)
such that f (c) = f (b)f
.
ba
(a)
Here is an illustration of this theorem: the slope f (b)f
of the line from (a, f (a)) to
ba
(b, f (b)) also equals the slope of the tangent line to the graph at some c between a and b:
(a)
(x a). By the sum and
Proof. Let g : [a, b] R be defined by g(x) = f (x) f (b)f
ba
scalar rules for continuity and differentiability, g is continuous on [a, b] and differentiable
(a)
on (a, b). Also, g(a) = f (a) and g(b) = f (b) f (b)f
(b a) = f (b) (f (b) f (a))
ba
= f (a) = g(a). Thus by Rolles theorem, there exists c (a, b) such that g (c) = 0. But
g (x) = f (x)
so that 0 = g (c) = f (c)
f (b)f (a)
,
ba
f (b) f (a)
,
ba
whence f (c) =
f (b)f (a)
.
ba
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.
201
Proof. We only prove part (2). Let x, y [a, b] with x < y. By Theorem 6.3.4 there exists
(y)
c (x, y) such that f (c) = f (x)f
. Since f (c) > 0 and x < y, necessarily f (x) < f (y).
xy
Since x and y were arbitrary with x < y, then f is strictly increasing on [a, b].
Example 6.3.6 Assuming that sin (x) 1 for all x, 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 (Cauchys 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)).
In particular, if g (c) 6= 0 and g(b) 6= g(a), this says that
f (c)
g (c)
f (b)f (a)
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 continuous 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 (Cauchys 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 (b) f (a)
f (c)
=
.
g (c)
g(b) g(a)
Proof. By the Mean Value Theorem (Theorem 6.3.4) there exists c (a, b) such that
. By assumption, g (c) 6= 0, so that g(b) 6= g(a). The rest follows by
g (c) = g(b)g(a)
ba
Cauchys Mean Value Theorem (Theorem 6.3.7).
6.3.1 (Bernoullis inequality) Prove that for all x R0 and all n N0 , (1 + x)n
1 + nx.
6.3.2 Let f be a polynomial 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 2.4.12.)
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?
202
6.3.3 Let f : [0, 1] C be given by f (x) = x3 + ix2 . Prove that there exists no c between
(0)
0 and 1 such that f (c) = f (1)f
. Does this contradict the Mean Value Theorem?
10
6.3.4 Let f : [a, b] C be continuous on [a, b] and differentiable on (a, b). Prove that
(a)
f (b)f (a)
there exist c, d (a, b) such that Re(f (c)) = Re f (b)f

and
Im(f
(d))
=
Im
.
ba
ba
Give an example showing that c may not equal to d. Does this contradict the Mean Value
Theorem?
6.3.5 Prove the remaining parts of Theorem 6.3.5.
6.4
LH
opitals rule
As in the previous section, the domains and codomains of all functions here are subsets
of R.
Theorem 6.4.1 (LH
opitals rule) Let a < b be real numbers. Let f, g : [a, b) R be
continuous with the following properties:
(1) f (a) = g(a) = 0.
(2) f and g are differentiable at all points in (a, b).
(3) For all x (a, b), g (x) 6= 0.
f (x)
= L.
(4) lim
xa+ g (x)
f (x)
Then lim
= L.
+
xa g(x)
f (x)
= L, there exists > 0 such that for all x in the domain,
xa g (x)
(x)
then | fg (x)
L| < . Let x be one such element. By Theorem 6.3.8 there
(c)
(x)f (a)
(x)
c (a, x) such that fg (c)
= fg(x)g(a)
= fg(x)
. But then 0 < c a < , so
Proof. Let > 0. Since lim

if 0 < x a <
exists a number
that

f (c)

f (x)

g(x) L = g (c) L < .
A very similar proof shows the other one-sided version of the LHopitals rule:
Section 6.4: LHopitals rule
203
Theorem 6.4.2 (LH

opitals rule) Let f, g : (a, b] R be continuous with the following
properties:
(1) f (b) = g(b) = 0.
(2) f and g are differentiable at all points in (a, b).
(3) For all x (a, b), g (x) 6= 0.
f (x)
= L.
(4) lim
xb g (x)
f (x)
Then lim
= L.
xb g(x)
An immediate consequence is of course the two-sided version of LHopitals rule. I
should note that many books omit hypothesis (3). A counterexample if we omit that
hypothesis is given in Exercise 9.10.5.
The three versions above of LHopitals rule deal with limits of the form zero over
zero. There are similarly three versions for the form infinity over infinity, and I write
and prove only the right-sided version:
Theorem 6.4.3 (LH
opitals 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) = .
xb
xb
(2) For all x (a, b), g (x) 6= 0.

f (x)
= L.
(3) lim
xb g (x)
f (x)
Then lim
= L.
xb g(x)
Proof. Let > 0. By assumption

there exists 1 > 0 such that for all x (a, b), if

f (x)

0 a.
Set a0 = b 1 . Fix x such that 0 < b x < 1 . Then x (a0 , b) (a, b). By
(x)f (a0 )
(c)
= fg(x)g(a
, and so
Theorem 6.3.8 there exists c (a0 , x) such that fg (c)
0)

f (x) f (a0 )
f (c)

< /4.
=
L
g(x) g(a0 )
g (c)

Since lim f (x) = lim g(x) = , there exists 2 > 0 such that for all x with
xb
xb
0 0 such that for all x, if
0 21 . Set = min {1 , 2 , 3 }. Then for all x with 0 1/2)
g(x) g(a0 )

f (x) f (a0 )

2
L + |L Lh(x)| (by triangle inequality)
g(x) g(a0 )

f (c)
L + |L| |1 h(x)|
= 2
g (c)

+
<2
4 4
= .
x2 1
.
3
x1 x 1
Example 6.4.4 Compute lim
x2 1
x3 1
lim 2x+1
x1 x +x+1
Proof #1: By Example 1.5.4,

2
x 1
3
x1 x 1
rem 4.3.6, lim
=
=
(x1)(x+1)
,
(x1)(x2 +x+1)
1+1
2
12 +1+1 = 3 .
so that by Exercise 4.3.11 and Theo-
Proof #2: By Theorem 4.3.5, lim (x2 1) = 0 = lim (x3 1), and lim (x2 1) = lim 2x = 2,
x1
x1
x1
x1
x2 1
3
x1 x 1
lim (x3 1) = lim 3x2 = 3, so that by LHopitals rule (Theorem 6.4.3), lim
x1
x1
2/3.
equals
More forms of LHopitals rule are in the exercises.

x2 64
.
3
x+2
x8
6.4.1 Compute lim
6.4.2 What is wrong with the following: Since limx2

x2 +3
that limx2 2x3
= 2.
6.4.3 Let n be a positive integer.
i) Use LHopitals rule to prove that limx1
ously, say how.
n
an
ii) Repeat for limxa x xa
= nan1 .
2x
2
xn 1
x1
= 2, by LHopitals rule we have
= n. We have proved this previ-
6.4.4 Explicitly state the assumptions about properties of trigonometric functions that
= 1 and lim 1cos(x)
= 0.
are necessary to prove lim sin(x)
x
x
x0
x0
205
6.4.5 (Here the goal is to prove another version of LHopitals 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,
and that
(x)
lim fg (x)
x
= L.
(x)
L| < .
i) Let > 0. Prove that there exists N > a such that for all x > N , | fg (x)
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 ).
(c)
iv) Prove that for all x > N there exists c (N , x) such that fg (c)
=
f (x) 1f (N )/f (x)
g(x) 1g(N )/g(x) .
f (x)
x g(x)
v) Prove that lim
= L.
6.4.6 (Yet another version of LHopitals 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
f (x)
g (x)
x
f (x)
x g(x)
= L. Prove that lim
= L.
Many interesting exercising on LHopitals rule are in Section 9.10, after the exponential and trigonometric functions have been covered.
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 limxa f (x) = f (a). Among all constant functions, the function y = f (a) approximates f at a best.
If f is differentiable near a and the derivative is continuous at a, then for all x near a but
(a)
not equal to a, f (x) = f (x) f (a) + f (a) = f (x)f
(x a) + f (a)
= f (a)(x a) + f (a),
xa
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 let the derivative of f
be f , or f (2) . If f (n1) is differentiable, we denote its derivative as f (n) . For completeness,
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)xmn .
206
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
n
X f (k) (a)
f (n) (a)
f (a)
(xa)2 + +
(xa)n =
(xa)k .
Tn,f,a (x) = f (a)+f (a)(xa)+
2!
n!
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 2)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 ,
..
.
f (k) (a) = k!ek if k d,
207
f (k) (a) = 0 if k > d.

But then for n d,
Tn,f,a (x) =
n
X
f (k) (a)
k=0
k!
d
X
k!ek
(x a) =
k=0
k!
(x a) =
d
X
k=0
ek (x a)k = f (x).
Theorem 6.5.5 (Taylors 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 continuous higher order
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)
(x a)(x c)n ,
n!
(n+1)
f
(d)
(x a)n+1 .
f (x) = Tn,f,a (x) +
(n + 1)!
f (x) = Tn,f,a (x) +
Proof. Let g : B(a, r) R be defined by

g(t) =
n
X
f (k) (t)
k=0
k!
(x t)k .
Then g is differentiable on B(a, r), and
g (t) =
n
X
f (k+1) (t)
k=0
k!
(x t)
n1
n
X
kf (k) (t)
k=0
k!
(x t)k1
n
X f (k+1) (t)
X f (k) (t)
f (n+1) (t)
(x t)n +
(x t)k
(x t)k1
=
n!
k!
(k 1)!
k=0
(n+1)
(t)
k=1
(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)
(x
n!
c)n (x a) = f (x) Tn,f,a (x), which proves the first formulation.

By Cauchys Mean Value Theorem (Theorem 6.3.7) applied to functions g(t) and
h(t) = (xt)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,
(n + 1)(x d)n (f (x) Tn,f,a (x)) =
f (n+1) (d)
(x d)n (0 (x a)n+1 ).
n!
Since d is not equal to x, we get that (n + 1)(f (x) Tn,f,a (x)) =

which proves the second formulation.
f (n+1) (d)
((x a)n+1 ),
n!
208
Theorem 6.5.6 (Taylors remainder theorem over C) Let A be a subset of C or

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 > 0 there exists
> 0 such that if x B(a, ), then
|f (x) Tn,f,a (x)| < .
Proof. Let > 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 1 > 0 such that for all
|x a| < 1 ,
qx B(a, r), ifq
}.
then |f (x)f (a)| < /(n+1). Set = min {r, 1 , /(M (n+1)), 2 (M (n+1)) , . . . n (M (n+1))
Let x satisfy |x a| < . Then x is in the domain of f and Tn,f,a , and

(n)

f
(a)
f
(a)
|f (x) Tn,f,a (x)| = f (x) f (a) f (a)(x a)
(x a)2
(x a)n
2!
n!
(n)

f (a)

|x a|2 + + f (a) |x a|n
|f (x) f (a)| + |f (a)| |x a| +
n!
2!
+M
+M
++M
<
n+1
M (n + 1)
M (n + 1)
M (n + 1)
= .
More on Taylor polynomials and Taylor series is in the exercises in this section, in
Exercise 7.4.13, and in Section 9.3.
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
n
X
n (k)
(n)
f (a)g (nk) (a).
(f g) (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.
6.5.5 Compute the Taylor polynomial of f (x) = 1 x of degree 10 centered at a = 0.
6.5.6 Let n N0 be given.

1
i) Compute the Taylor polynomial of f (x) = 1x
of degree n centered at a = 0.
ii) Compute Tn,f,0 (0.5).
iii) Use Theorem 6.5.5 to determine n such that |Tn,f,0 (0.5) f (0.5)| < 0.001.
209
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 such that f has derivatives of orders 1, 2, 3, 4, but f (4)
is not differentiable.
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
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 of course 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:
211
y = f (x)
y
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:
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 (Theorem 3.8.5), for any subset B of the domain, sup {f (x) : x B} and
inf {f (x) : x B} are real numbers.
212
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 <
< xn1 < 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
L(f, P ) =
n
X
inf {f (x) : x [xi1 , xi ]}(xi xi1 ).
n
X
sup {f (x) : x [xi1 , xi]}(xi xi1 ).
i=1
The upper sum of f with respect to P is

U (f, P ) =
i=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
Rb
are c(b a), so that a f = 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 ) < , then either U (f, P ) or L(f, P ) serves as an approximation of the true signed area within 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, xi = 1 + i 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 [xi1 , xi]} = f (xi1 ) and
sup {f (x) : x [xi1 , xi ]} = f (xi). 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
213
partitions Pn into n equal subintervals:

n
10
100
1000
10000
100000
1000000
L(f, Pn)
5061.2757
6038.727
6141.5217
6151.8517
6152.885
6152.988
U (f, Pn)
7358.0757
6268.407
6164.48967
6154.148
6153.1148
6153.011
Notice how the lower sums get larger and the upper sums get smaller as we take finer
partitions. We 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 [xi1 , xi ] = {y0 =

xi1 , y1 , . . . , ym1 , ym = xi }. By set inclusion, inf {f (x) : x [yj1 , yj ]} inf {f (x) : x
[xi1 , xi ]}, so that
L(f, {y0, y1 , . . . , ym1 , ym }) =
m
X
j=1
m
X
j=1
inf {f (x) : x [yj1 , yj ]}(yj yj1 )

inf {f (x) : x [xi1 , xi ]}(yj yj1 )
m
X
= inf {f (x) : x [xi1 , xi ]}
(yj yj1 )
j=1
= inf {f (x) : x [xi1 , xi ]}(xi xi1 ),
and hence that

n
n
X
X
inf {f (x) : x [xi1 , xi]}(xi xi1 ) = L(f, P ).
L(f, R [xi1 , xi ])
L(f, R) =
i=1
The proof for upper sums is similar.
i=1
214

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 [xi1 , xi]} = 0 and sup {f (x) : x
Pn
[xi1 , xi ]} = 1, so that L(f, P ) = 0 and U (f, P ) =
i=1 1 (xi xi1 ) = 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
Theorem 7.1.8 shows that L(f ) U (f ). Again, if f is a constant function f (x) = c

Rb
for all x, then all lower and all upper sums are 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 in the next section we will
prove that every continuous function is integrable over a closed bounded interval.
215
Notation 7.1.10 In the definition of integral we sometimes write dx or dt, and sometimes 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
x dx = 8,
x dt = 4x.
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
independent, we can get further values too. Say if x = 3t, then
Z 4
Z 4
x dt =
3t dt = 24.
0
In short, use notation precisely.

7.1.1 Prove that if P = {x0 , . . . , xn } is a partition of [a, b] into n equal parts, then xi =
a + i ba
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).
216
7.1.5 Use geometry to compute the following integrals:

Z a
i)
f=
a
Z rp
r 2 x2 dx =
ii)
Z0 5
iii)
(4x 10) dx =
3
5,
if x < 7;
Z 5
2,
if 7 x < 1;
f , where f (x) =
iv)
if 1 x < 3;
1
3x,
9 x, if 3 x.
7.1.6 Compute the following integrals (t and x do not depend on each other):
Z rp
i)
r 2 x2 dt =
Z0 5
ii)
(4x 10) dt =
3
5,
if t < 7;
Z 5
2,
if 7 t < 1;
f , where f (t) =
iii)
3t,
if
1 t < 3;
9 x, if 3 t.
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.
217
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,
xi = 2 + i n4 . Since f is increasing, on each subinterval [xi1 , xi ] the minimum is xi1 and
the maximum is xi . Thus
U (f, Pn) =
n
X
xi (xi
i=1
n
X
xi1 )

4
n
i=1
2
n
n
X
4
4 X
i
=
2 +
n i=1
n
i=1
2
4
n(n + 1) 4
=2 n+
n
2
n
8(n + 1)
=8+
.
n
4
2+i
n
It follows that for all n,

U (f ) = inf {U (f, P ) : P varies over partitions of [2, 6]} U (f, Pn) = 8 +
so that U (f ) 8 + 8 = 16. Similarly,
L(f, Pn) =
n
X
xi1 (xi
i=1
n
X
8(n + 1)
,
n
xi1 )
4
2 + (i 1)
n
4
n
i=1
2

n
n
X
4 X
4
=
2 +
(i 1)
n
n
i=1
i=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,
218
so that L(f ) = U (f ) = 16, and finally that
R6
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 integer
n let Pn = {x0 , . . . , xn } be the partition of [1, 7] into n equal parts. By Exercise 7.1.1,
xi = 1 + i n6 . Since f is increasing on [1, 7], on each subinterval [xi1 , xi ] the minimum is
x2i1 and the maximum is x2i . Thus, by using Exercise 1.5.1 in one step below:
U (f, Pn) =
n
X
x2i (xi xi1 )
i=1
n
X
2
6
6
=
1+i
n
n
i=1

n
X
12
6
2 36
1+i +i 2
=
n
n
n
i=1
3
n
n
n
X
6 X 12 6 X 2 6
=
+
i
+
i
n
n
n
n
i=1
i=1
i=1
n(n + 1) 12 6
n(n + 1)(2n + 1)
6
+
= n+
n
2
n n
6
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.
3
6
n
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,
xi = 2i
n . The square root function is increasing, so that
n
X
U (f, Pn) =
xi (xi xi1 )
i=1
n r
X
i=1
2i 2
n n
3/2 X
n
2
=
i.
n
i=1

But now we are stuck: we have no simplification of
3/2 Pn
diate tricks to compute inf { n2
i : n}.
i=1
219
i=1 i, and we have no other imme-
Pn
But it is possible to compute enough upper and lower sums for this function to get the
2(n1)2 2n2
29
integral. Namely, for each positive integer n let Qn = {0, n22 , 24
,
,
.
.
.
,
, n2 =
2
2
n
n
n2
2i2
is an
2}. This is a partition of [0, 2] into n (unequal) parts, with xi = n2 . Since
increasing function, on each subinterval [xi1 , xi ] the minimum is achieved at xi1 and the
maximum at xi , so that

n
X
2 i 2i2
2(i 1)2
U (f, Qn) =
n
n2
n2
i=1
n

2 2X 2
= 3
i i (i 1)2
n i=1
n
2 2X
= 3
i (2i 1)
n i=1
n
n
4 2X 2 2 2X
= 3
i 3
i
n i=1
n i=1
4 2 n(n + 1)(2n + 1) 2 2 n(n + 1)

3
(by Exercise 1.5.1)
= 3
n
6
n
2
2 2
=
(n + 1)(2(2n + 1) 3) (by factoring)
2
6n
2
=
(n + 1)(4n 1).
3n2
Thus
U (f ) inf {U (f, Pn) : n} = inf { 3n22 (n + 1)(4n 1) : n} =
R2
4 2
,
so
that
x dx = 4 3 2 .
3
0
4 2
.
3
Similarly, L(f )

R2
x3 dx. (You may need Exercise 1.5.2.)

R2
7.2.2 Mimic examples in this section to compute 0 3 x dx.
7.2.1 Mimic examples in this section to compute
7.2.3 In EasyLand they define Entegrals as follows: first they divide the interval [a, b] into
. By Exercise 7.1.1, the ith subinterval given in
n equal subintervals, so each has length ba
n

ba
ba
this way is a + (i 1) n , a + i n . Rather than finding the minimum and maximum
of f on this subinterval, they simply Epproximate f on the subinterval by plugging in the
right endpoint a + i ba
n , so that the signed area over the ith subinterval is Epproximately

ba ba
f a+i n
n . Thus the Epproximate signed area of f over [a, b] via this partition is
Pn
ba ba
i=1 f a + i n
n . If this sum has a limit as n goes to infinity, then EasyLanders
220
declare the Entegral of f over [a, b] to be

n
b
X
ba
b
a
f = lim
.
f a+i
n
n
n
a
i=1
b
Suppose that f : [a, b] R is integrable. Prove that the Entegral f exists and equals
a
Rb
the integral a f .

7.2.4 Use Entegrability from previous exercise. Let f (x) =
2
i) Prove that f is Entegrable over [0, 2] and find its Entegral f .
0
2
2
2
2 2
ii) Compute f, f , and prove that f + f 6= f .
0
iii) Prove that f is not integrable over [0, 2] (cf. Example 7.1.7)).
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.
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 2.7.13 it suffices to prove that for all > 0, U (f ) L(f ) .
So let > 0. By Theorem 5.4.2, f is uniformly continuous, so there exists > 0 such
that for all x, c [a, b], if |xc| < then |f (x)f (c)| < /(ba). Let P = {x0 , x1 , . . . , xn}
be a partition of [a, b] such that for all i = 1, . . . , n, xi xi1 < . (For example, this can
be accomplished as follows: by Theorem 3.8.3, there exists n N+ such that (b a) < n,
and then P can be taken to be the partition of [a, b] into n equal parts.) Then
U (f, P ) L(f, P ) =
n
X
i=1
n
X
i=1
(sup {f (x) : x [xi1 , xi ]} inf {f (x) : x [xi1 , xi]}) (xi xi1 )
(xi xi1 ) (by uniform continuity since xi xi1 < )

(b a)
n
(xi xi1 )
=
(b a) i=1
= .
221
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 ) .

Thus U (f ) = L(f ) by Theorem 2.7.13, so that f is integrable over [a, b].
Furthermore we get the following formulation of integrals that looks very technical
but is fundamental for applications (see Section 7.6). The proof is simply a more careful
rephrasing of the previous proof:
Theorem 7.3.2 Let f : [a, b] R be continuous. For each real number r > 0 let
(r)
(r)
(r)
Pr = {x0 , x1 , . . . , xnr } be a partition of [a, b] such that each subinterval has length at
(r)
(r)
(r)
most r. For each i = 1, 2, . . . , nr , let ci [xi1 , xi ]. Then
Z b
nr
X
(r)
(r)
(r)
f (ci )(xi xi1 ) =
f.
lim+
r0
i=1
The superscripts tell us the dependence on the real number r, and the limit is over such r.
Somewhat sloppier but perhaps notationally saner is the following formulation:
Z b
n
X
f.
lim
f (ci )(xi xi1 ) =
P ={x0 ,x1 ,...,xn }finer and finer,ci [xi1 ,xi ]
i=1
Theorem 7.3.3 Let g : [a, b] R be continuous, and let f : [a, b] R be such that for
all x (a, b), f (x) = g(x). Then f is integrable over [a, b], and
Z b
Z b
g.
f=
a
Proof. Since g is continuous, by the Extreme Value Theorem (Theorem 5.2.5) there
exists a positive real number M such that for all x [a, b], |g(x)| < M . Let M =
max {M , |f (a)|, |f (b)|. Then for all x [a, b], |f (x)| < M .
Let > 0. Since f is bounded, L(f ) and U (f ) exist, and we need to prove that
L(f ) = U (f ). By Theorem 2.7.13 it suffices to prove that for all > 0, U (f ) L(f ) .
By Theorem 7.3.1, g is integrable over [a, b], so there exists a partition P of [a, b] such
that U (g, P ) L(g, P ) < /3. Let w = min {/(4M + 1), (b a)/3}. Let P = {a, a +
w, b w, b} be a partition of [a, b]. Let P be a common refinement of P and P . Then
U (g, P ) L(g, P ) < /2, and furthermore, if P = P [a + w, b w], then by necessity
U (f, P ) L(f, P ) = U (g, P ) L(g, P ) < /2.
Since M is an upper bound on |f |, it follows that M m[a,a+w] M[a,a+w] M and

M m[bw,b] M[bw,b] M . Thus U (f, P ) L(f, P ) is at most this difference on
[a, a + w] plus this difference on [a + w, b w] plus this difference on [b w, b], so that
U (f, P ) L(f, P ) 2M w + /2 + 2M w < . It follows that U (f ) U (g) . Thus f is
integrable.
222
Furthermore, U (f, P ) and U (g, P ) differ by at most 4M w, namely the difference on

[a, a + w] and [b w, b], which is at most /2. As we take finer partitions, the difference is
still at most /2, so that U (f ) and U (g) differ by at most /2. Thus by Theorem 2.7.13,
U (f ) = U (g), and so the two integrals are the same.
Theorem 7.3.4 Suppose that f is integrable over [a, b] and [b, c]. Then f is integrable
Rc
Rb
Rc
over [a, c], and a f = a f + b f .
More generally, if for each i = 1, . . . , m, f is integrable on [ai1 , ai ], then f is integrable
R ai
Ra
P
f.
on [a0 , am ], and a0m f = ni=1 ai1
Proof. Let > 0. Since f is integrable over [a, b] there exists a partition P of [a, b]
such that U (f, P ) L(f, P ) < /2. Similarly there exists a partition Q of [b, c] such
that U (f, Q) L(f, Q) < /2. Let R = P Q. Then R is a partition of [a, c], and
U (f, R) = U (f, P ) + U (f, Q) and L(f, R) = L(f, P ) + L(f, Q). Thus U (f, R) L(f, R) < ,
and so the upper integral of f over [a, c] minus the lower integral of f over [a, c] is between
0 and . Thus by Theorem 2.7.13, U (f ) = L(f ), so that f is integrable over [a, c]. Finally,
Z c
f = sup {L(f, R) : R a partition of [a, c]}
a
sup {L(f, R) : b R a partition of [a, c]}
= sup {L(f, P ) + L(f, Q) : P a partition of [a, b], Q a partition of [b, c]}
= (L(f ) on [a, b]) + (L(f ) on [b, c])
(as the partitions of [a, b] and [b, c] are independent of each other)
= (U (f ) on [a, b]) + (U (f ) on [b, c])
= inf {U (f, P ) + U (f, Q) : P a partition of [a, b], Q a partition of [b, c]}
= inf {U (f, R) : b R a partition of [a, c]}
inf {U (f, R) : R a partition of [a, c]}

Z c
f.
=
a
This proves the first part. The second part follows by induction on m.
Ra
f if a < b? In our definition

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
xi1 xi = (xi xi1 ), so that all partial sums and both the lower and upper integrals
would get the negative value. Thus it seems reasonable to declare
Z a
Z b
f =
f.
Notation 7.3.5 What could possibly be the meaning of
223
In fact, this is exactly what makes Theorem 7.3.4 work without any order assumptions
Rb
Rc
Rb
on a, b, c. For example, if a < c < b, by Theorem 7.3.4, a f = a f + c f , whence
Rc
Rb
Rb
Rb
Rc
f
=
f
f
=
f
+
f , which is exactly what Theorem 7.3.4 says.
a
a
c
a
b
Ra
Rb
Ra
Rb
Furthermore, from a f = a f + a f we then deduce that a f = 0.
Theorem 7.3.6 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]}
= 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).
This proves that cg is integrable with
Rb
a
(cg) = c
Rb
a
g.
Let > 0. By integrability of f and cg there exist partitions P, Q of [a, b] such that
U (f, P ) L(f, P ) < /2 and U (cg, Q) L(cg, Q) < /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) < /2 and U (cg, R)L(cg, R) <
/2. By Exercise 2.7.8, for every partition of [a, b], and in particular for the partition R,
224
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)

< .
Thus U (f + cg) L(f + cg) = 0 by Theorem 2.7.13, 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),
so that finally
Z b
Z b
Z b
Z b
Z b
g.
f +c
(cg) =
f+
(f + cg) =
a
Theorem 7.3.7 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
g.
f
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)
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.
7.3.1 The following is known:
R1
R3
R2
3g.
5f
,
(3f
4g),
2
1
1
R1
0
f = 5,
R3
2
f = 6,
R3
0
f = 15,
R2
1
g = 3. Compute
225
x, if x is rational;
i) Prove that f is not integrable over [0, 1].
ii) Does this contradict Theorem 7.3.4? Justify.
7.3.2 Let f (x) =
7.3.3 (There is no integration in this exercise.) 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.
Ra
7.3.4 Let f : [a, a] R be an odd function. Prove that a f = 0.
Ra
Ra
7.3.5 Let f : [a, a] R be an even function. Prove that a f = 2 0 f .
7.3.6 Let f : [a, b] R be piecewise continuous as in Theorem 7.3.4.
i) Prove that |f | isintegrable over [a, b].
R b R b
ii) Prove that a f a |f |.
7.3.7 Find a function f : [0, 1] R that is not integrable over [0, 1] but such that |f | is
integrable over [0, 1].
7.3.8 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.
7.3.9 (Improper integral, I) Let f : [a, ) R be a continuous function.
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.

R
3x + 1, if x < 10;
iv) Let f (x) =
. Compute 0 f .
0,
otherwise.
R
Rb
7.3.10 Similarly to part (iii) in the previous exercise, formulate f and f .

1,
if x (the floor of x) is even;
7.3.11 Let 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 (, ) in the extended sense
from the previous two exercises.
226
7.3.12 (Improper integral, II) Let f : (a, b] R be a continuous function.

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.
*7.3.13 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 continuous 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 [xi1 , xi] and differentiable on each (xi1 , xi). Thus by the Mean
Value Theorem (Theorem 6.3.4), there exists ci (xi1 , xi ) such that f (ci ) = g (ci ) =
g(xi )g(xi1 )
. By the definition of lower and upper sums,
xi xi1
L(f, P )
But
n
X
i=1
n
X
i=1
f (ci )(xi xi1 ) U (f, P ).
f (ci )(xi xi1 ) =

=
n
X
g(xi ) g(xi1 )
i=1
n
X
i=1
xi xi1
(xi xi1 )
(g(xi) g(xi1 ))
= 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)
227
inf {U (f, P ) : P a partition of [a, b]}
= U (f ).
Since f is continuous on [a, b], it is integrable by Theorem 7.3.1, so that L(f ) = U (f ), and
Rb
so all inequalities above 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
b
Z b

f = g(x) = g(b) g(a).
a
For example, assuming that cos (x) = sin(x), we have that

Z

sin x dx = cos(x) = cos() ( cos(0)) = (1) + 1 = 2.
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:
r+1
r+1

Z b
b
a
br+1
ar+1
r
x dx =
+C
+C =
,
r+1
r+1
r+1 r+1
a
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 5.3.6 also allows real exponents, 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
R x
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 (18091882). What is the
228
meaning of closed-form? Here is an oblique answer: Exercise 9.7.5 claims that there ex2
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 Liouvilles 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 9.5).
The simplest method for finding more antiderivatives is to first find a differentiable
function and compute its derivative, and voila, 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, x) with
Z x
d
f = f (x).
dx a
Proof. f is integrable over [a, x] by Theorem 7.3.1, thus g is a well-defined function.
Let c (a, x). Let > 0. Since f is continuous on [a, b], there exists > 0 such that
for all x [a, b], if |x c| < then |f (x) f (c)| < . Thus on [c , c + ] [a, b],
f (c) < f (x) < f (c) + , so that by Theorem 7.3.7,
Z max {x,c}
Z max {x,c}
Z max {x,c}
(f (c) + ).
f
(f (c) )
min {x,c}
min {x,c}
min {x,c}
But by integrals of constant functions,

Z max {x,c}
(f (c) ) = (max {x, c} min {x, c})(f (c) ) = |x c|(f (c) ).
min {x,c}
Thus
|x c|(f (c) )
max {x,c}
min {x,c}
f |x c|(f (c) + ).
If x 6= c, dividing by |x c| and rewriting the middle term gives this says that
Rx
f
f (c) c
f (c) + ,
xc

229
Rx
R x

f

f
c
c
whence xc f (c) , and xc f (c) < . Then for all x [a, b], if 0 < |xc| < ,
then

R
R x

a f ac f
g(x) g(c)

f (c) =
f (c)

xc
xc
Rx

c f

=
f (c) (by Theorem 7.3.4 and Notation 7.3.5)
xc

< .
g(x) g(c)
exists and equals f (c), i.e., g (c) = f (c).
xc
xc
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
Thus lim
f=
f (t) dt =
f (x) dx.
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
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
The following are then immediate generalizations of the two versions of the fundamental theorem of calculus:
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
230
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
7.4.1 Compute the integrals below. Clever guessing is fine.
Z 1
i)
x3 dx =
Z0 1
x dx =
ii)
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 =
Z0 t
iv)
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)
x3 dx =
dx Z0
x
d
ii)
t3 dt =
dx 0
Z t
d
x3 dx =
iii)
dx 0
Z t
d
iv)
t3 dt =
dx Z0
x
d
t3 dx =
v)
dx 0
Z t
d
vi)
x3 dt =
dx 0
231
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 =
(2x2 + 3x)10
1
Z 3
4x + 3
p
dx =
2x2 + 3x)10
1
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 =
Z1
1
4x + 3
dx =
2x + 4
1
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 =
.
(2n + 1)2
0
7.4.8 Compute the following derivatives (hint: the Fundamental Theorem of Calculus and
the chain rule):
Z 3x q
d
x4 + 5 x dx.
i)
dx 2
Z x
d
x+5 x
ii)
dx.
dx x2 x1 00 2x5 0 + x7 2
Z h(x)
d
f (t) dt.
iii)
dx g(x)
Z h(x)
d
iv)
f.
dx g(x)
232
7.4.9 Consult Exercise 7.3.9.

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 7.3.12.
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.
f (c) =
ba a
7.4.12 Work out Exercises 7.3.4, 7.3.5, and 7.3.6 for complex-valued functions.
7.4.13 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 (t) dt.
f (x) = f (a) +
a
ii) Integrate the integral above by parts, and rewrite, to get that
Z x
(x t)f (t) dt.

f (x) = f (a) + (x a)f (a) +
a
iii) Use induction, integration by parts, and rewritings, to get that

Z x
f (a)
f (n) (a)
(x t)n (n+1)
n
f (x) = f (a) +
(x a) + +
(x a) +
f
(t) dt.
1!
n!
n!
a
iv) Say why you cannot apply the Fundamental Theorem of Calculus, II, to compute
Rx
d
(x t)n+1 f (n+1) (t) dt.
dx a
v) (Taylor remainder formula in integral form) Consult Section 6.5 for Taylor polyRx
n
(n+1)
nomials to prove that f (x) = Tn,f,a (x) + a (xt)
(t) dt.
n! f
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 7.3.5, for all x > 0,
Rx 1
dx is well-defined. This function has a familiar name:
1 x
233
Definition 7.5.1 The natural logarithm is the function

Z x

Z x
1
1
dx =
dt
ln x =
1 t
1 x
for all x > 0.
We prove below all the familiar properties of this familiar function.

Remark 7.5.2
R1
(1) ln 1 = 1 x1 dx = 0.
Rx
Rx
(2) By geometry, for x > 1, ln x = 1 x1 dx > 0, and for x (0, 1), ln x = 1 x1 dx =
R1
x x1 dx < 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
c = x1 = ln (x). Thus the function g ln has constant derivative 0. It
g (x) = cx
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(c) ln(x).
ln
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(cn1 c) =
ln(cn1 ) + 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/cr ) =
ln(1) ln(cr ) = 0 (r) ln(c) = r ln(c), which proves the claim for all integers.
r
m/n
)=
Now write r = m
n for some integers m, n with n 6= 0. Then n ln(c ) = n ln(c
m
m
r
ln(c ) = m ln(c), so that ln(c ) = n ln(c) = r ln(c).
234
(10) The range of ln is R = (, ). Here is a proof. By geometry, ln(0.5) < 0 < ln(2).
Let y R+ . By Theorem 3.8.3, 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 5.2.1), 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(x1 ). 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 ln1 : R R+ . By Theorem 5.2.4, ln1 is
increasing and continuous.
(12) By Theorem 6.2.7, the derivative of ln1 is
(ln1 ) (x) =
1
= ln1 (x).
1
ln (ln (x))
(13) For all x, y R,

1

ln1 (x)
ln (x)
1
1
1
1
=
ln
ln
ln
(x)
ln
ln
(y)
= ln1 (x y) .
=
ln
ln
1
1
ln (y)
ln (y)
We have proved that for all c R+ and r Q, ln1 (r ln(c)) = ln1 (ln(cr )) = cr , and
we have proved that for all r R, ln1 (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 = ln1 (r ln(c)).
This definition gives rise to two functions:
(1) The generalized power function with exponent r when c varies and r is constant;
(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 differentiable, with f (x) = rxr1 . This function is increasing if r > 0 and decreasing if r < 0.
Proof. By definition, f (x) = ln1 (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
ln(x))
= r ln1 (r ln(x) ln(x)) = r ln1 ((r 1) ln(x)) =
f (x) = ln1 (r ln(x)) xr = r lnln1(r(ln(x))
rxr1 . The monotone properties then follow from Theorem 6.3.5.
235
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) = ln1 (x ln(c)), which is differentiable by the chain and scalar
rules and the fact that ln1 is differentiable. Furthermore, the derivative is f (x) =
ln1 (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 ln1 .
Definition 7.5.6 Let e = ln1 (1) (so that ln(e) = 1). The constant e is called Eulers
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) 2. We conclude that e is a number strictly between 2 and 3.
1
1
1
1
+ 1.5
+ 1.75
+ 21 + 2.25
)=
Note that U (f, {1, 1.25, 1.5, 1.75, 2, 2.25, 2.5}) = 0.25(1 + 1.25
R
2.5
2509
f is strictly smaller than this upper sum. It follows that
2520 < 1, so that ln 2.5 = 1
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 different and perhaps easier computation is in Exercise 7.5.13.
Theorem 7.5.7 ln1 (x) = ex .
Proof. By definitions,
ex = ln1 (x ln(e)) = ln1 (x ln(ln1 (1))) = ln1 (x 1) = ln1 (x).
We have already proved in page 234 that the derivative of ln1 is ln1 :
Theorem 7.5.8 (ex ) = ex .
236

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
(cb ca ).
ii) Prove that if c =
6 1, then a cx dx = ln(c)
7.5.3 Use integration by substitution (Exercise 7.4.5) to compute
x
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 LH

opitals rule.
ln x
i) Prove that lim x = 0.
x
ii) Prove that lim+ x ln x = 0. (Hint: Use one of x ln x =

x0
x
1/ ln x
or x ln x =
ln x
1/x ,
and
perhaps not both work.)

iii) Compute lim+ ln(xx ).
x0
iv) Compute lim xx .

x0+
7.5.8 (Logarithmic differentiation) Sometimes it is hard or even impossible to compute

3
the derivativeof a function. Try for example f (x) = xx , f (x) = (x2 + 2)x +4 , or f (x) =
x2 (x1)3 (x+4)3 x2 +1
. There is another way if the range of the function consists of positive

3
x+2(x+7)3 (x2)4
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
x
f (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 the derivative of f (x) = (x2 + 2)x +4 .
2
(x1)3 (x+4)3 x2 +1
.
ii) Compute the derivative of f (x) = x
3
x+2(x+7)3 (x2)4
1
ln(1 + x) = 1. LHopitals rule are in Section 9.10,
x0+ x

1
= 1.
i) Prove that lim x ln 1 +
x
x

x
1
ii) Prove that lim 1 +
= e.
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.)

rem 6.4.3).
ex 1
x
237
= 1. You may want to apply LHopitals rule (Theo-
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 Taylors remainder theorem (Theorem 6.5.5) to show that for any x R and
any > 0 there exists n N such that |f (x) Tn,f,0 (x)| < .
iii) Use the Taylors remainder theorem (Theorem 6.5.5) to prove that |eT8,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 235. 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 . (Hint: Rewrite for LHopitals rule.) Use LHopitals rule (Theorem 6.4.3) and induction on n to prove that
ex Tn,f,0 (x)
1
=
.
n+1
x0
x
(n + 1)!

1/x2
, if x 6= 0;
7.5.15 Let f : R R be given by f (x) = e
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
)
e
, if x 6= 0;
h
(
(n)
n
x
f (x) =
0,
if x = 0.
lim
(At non-zero x you can use the chain rule, the derivative of the exponential function, and the power rule for derivatives. However, f (n+1) (0) is but of course
(n)
(n)
(0)
computed as limh0 f (h)f
, and then you have to use LHopitals rule.
h
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 ) Use calculus (not a calculator!) to
determine which number is bigger, e or e . You may assume that 1 < e < . (Hint:
take the natural logarithm of both numbers, cross divide, compute the derivative of some
function.)
238
7.6
Applications of integration
The Fundamental theorems of calculus relate integration with differentiation. In parRb

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 functions 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
[xi1 , xi ] we approximate the curve with the line (xi1 , f (xi1)) to (xi , f (xi)), compute
p
the length of that line as (xi xi1 )2 + (f (xi) f (xi1 ))2 , and sum up all the lengths:
n p
X
(xi xi1 )2 + (f (xi) f (xi1 ))2 .
i=1
Whether this is an approximation of the true length depends on the partition, but geometrically it makes sense that the true length of the curve equals
lim
n p
X
(xi xi1 )2 + (f (xi) f (xi1 ))2 ,
i=1
as the partitions {x0 , x1 , . . . , xn } get finer and finer. But this is not yet in form of Theorem 7.3.2. 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 i = 1, . . . , n there exists
ci (xi1 , xi ) such that f (xi) f (xi1 ) = f (ci )(xi xi1 ). Then the true length of the
curve equals
lim
n p
X
(xi xi1 )2 + (f (ci )(xi xi1 ))2
i=1

=
lim
239
n p
X
1 + (f (ci ))2 (xi xi1 ),
i=1
and by Theorem 7.3.2, if f is continuous, this equals

We just proved:
Rbp
a
1 + (f (x))2 dx.
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
(f (a))2(b a). For a general f we let P = {x0 , x1 , . . . , xn} be a partition of [a, b]; on each
subinterval [xi1 , xi] we approximate the curve with the horizontal line y = f (ci ) for
some ci [xi1 , xi], we compute the volume of the solid of revolution obtained by rotating
that approximated line over the interval [xi1 , xi ] around the x-axis, and sum up all the
volumes:
n
X
(f (ci))2 (xi xi1 ).
i=1
Geometrically it makes sense that the true volume equals

lim
and by Theorem 7.3.2 this equals
Rb
a
n
X
i=1
(f (ci))2 (xi xi1 ),
(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
(f (x))2 dx.
a
240
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 (d c)(b2 a2 ). For general f and g we let P = {y0 , y1 , . . . , yn} be a partition of
[a, b]; on each subinterval [yi1 , yi ] we approximate the curve f, g with the horizontal
line f (ci), g(ci ) for some ci [yi1 , yi ], we compute the volume of the solid of revolution
obtained by rotating that approximated region over the interval [yi1 , yi ] around the x-axis,
and sum up all the volumes:
n
X
i=1
2
(g(ci) f (ci ))(yi2 yi1
).
Geometrically it makes sense that the true volume equals

lim
P ={y0 ,y1 ,...,yn }finer and finer,ci [yi1 ,yi ]
lim
n
X
i=1
P ={y0 ,y1 ,...,yn }finer and finer,ci [yi1 ,yi ]
and by Theorem 7.3.2 this equals
Rb
a
2
(g(ci) f (ci ))(yi2 yi1
)
n
X
i=1
(g(ci) f (ci ))(yi + yi1 )(yi yi1 ),
(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
2
y(g(y) f (y)) dy.
a
Example 7.6.4 The volume of the sphere of radius r is 43 r 3 .

Proof. We rotate the upper-half circle of radius r centered at the origin around the xaxis. 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
r

Z r p

4
2
2
3/2
= 4 (r 2)3/2 = 4 r 3 .
y r 2 y 2 dy = (r y )
4

3
3
3
0
0
241
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 2|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 2|m|b
radius
p
b2 + (mb)2
p
Without the clip in the disc, the perimeter would be 2 b2 + (mb)2 , but our perimeter
is only 2|m|b. Thus the angle subtended by the wedge is by proportionality equal to
2|m|b
. The area of the full circle is radius squared times one half of
2 = 2|m|
2
2
1+m2
2
b +(mb)
the full angle, and so proportionally the area of our wedge is radius squared times one half
p
2|m|
of our angle, i.e., the surface area of this surface of revolution is ( b2 + (mb)2 )2 2
1+m2
2
2
= |m|b 1 + m . Note that even if b < 0, the surface area is the absolute value of
mb2 1 + m2 .
Thus the surface area of revolution obtained by rotating the line y = mx with non
zero m from x = a to x = b the absolute value of m(b2 a2 ) 1 + m2 if 0 a < b or
a < b 0. Note the geometric requirement that at a and b the line is on the same side of
242
the x-axis.
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
m((b + l/m)2 (a + l/m)2 ) 1 + m2
p
= m b2 + 2bl/m + l2 /m2 (a2 + 2al/m + l2 /m2 )
1 + m2
p

1 + m2
= m b2 a2 + 2(b a)l/m
p
= m(b a)(b + a + 2l/m) 1 + m2
p
= (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)2|l|, which is the absolute value of (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
(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] R0 be a differentiable (and not necessarily linear) function. Let
P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on each subinterval [xi1 , xi ] we approximate
the curve with the line from (xi1 , f (xi1)) to (xi , f (xi)). By the assumption that f takes
on only non-negative values we have that both xi1 , xi are both on the same side of the
(xi1 )
x-intercept of that line. The equation of the line is y = f (xxi i)f
(xxi )+f (xi ), so that
xi1
m=
f (xi )f (xi1 )
xi xi1
(xi1 )
and l = f (xxi i)f
xi + f (xi). Since f is differentiable, by the Mean
xi1
(xi1 )
Value Theorem (Theorem 6.3.4) there exists ci (xi1 , xi ) such that f (ci ) = f (xxi i)f
.
xi1
Thus m = f (ci ) and l = f (xi) f (ci )xi .

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:
n

X
p

2
(xi xi1 ) (f (ci )(xi + xi1 ) + 2 (f (xi) f (ci )xi )) 1 + (f (ci ))
i=1
n
X
i=1
|(f (ci )(xi + xi1 ) + 2f (xi))|
By Theorem 7.3.2 then we get the following:
p
1 + (f (ci ))2 (xi xi1 ).
243
Theorem 7.6.5 If f : [a, b] R0 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
Z b
p
p
2
f (x) 1 + (f (x))2 dx.
|f (x)(x + x) + 2f (x)| 1 + (f (x)) dx = 2

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.
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 A tiny particle of mass m is rotating around a circle of radius r. The moment of
inertia of the particle is I = mr 2 . Instead of a tiny particle we have a thin circular plate
of radius 5 m and mass 4 kg. Let b be the thickness of the plate and 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 [5, 5]. Prove that the moment of inertia of the sliver of the
p
plate between xi1 and xi is approximately c2i 2 25 c2i b, where ci [xi1 , xi ].
iii) Prove that the moment of inertia of the rotating circular plate is 25 kg m2 .
7.6.8 If a constance 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.
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 submerged 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.
In this chapter, Sections 8.4 and 8.5 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 s : N+ C. 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 }n1 = {sn }nN+ = {sn+3 }n=2 = {sn4 }n>4 = {sn }.
The nth element in the ordered list is called the nth term of the sequence.
We note that while 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 a 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}n4 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 }n1 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
c+d
n
more concisely as { dc
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}.
245
(4) The sequence s with s(n) = n + 4 for all n 1 can be written as {n + 4} =

{n+4}n1 = {5, 6, 7, . . .} = {n}n5 . Note that {n} stands for a different sequence,
namely for the sequence {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}n1 = {2n + 1}n0 (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 of course sn = 2n1 has a non-recursive form. 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 +sn1 .
This sequence starts with 1, 1, 2, 3, 5, 8, 13, 21, 34, . . ., and obviously keeps growing.
(For more on these numbers, see
Exercise
where
1.5.31,
in particular it is proved
that for all integers n, sn =
1
5
1+ 5
2
1
5
1 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 31.)
(10) Note that {n}nZ 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
if n = 1;
0,
if n is even;
sn = n/2,
(n 1)/2, otherwise.
(Otherwise applies to odd integers 3 or bigger.)
(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
246
in the first row, and then proceed to the second row, well, actually, we never get
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, 3/4,
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.
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 Exercise 2.4.27.
(13) A trick incorporating the last two sequencing strategies yields also a scrambling
of Q into a sequence (see Exercise 8.1.6).
Incidentally, it is impossible to scramble R or C into a sequence. This can be proved
with a so-called Cantors 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
247
a plot of the sequence {1/n}.

sn
1
b
b
b
b
b
b
b
b
b
b
b
b
b
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
0
s1
b
s3
b
b b
b b bb
b
i/2
s2
s4
bb b b
b
b b
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 },
{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}n3 + {n}n2 = {n + 1}n2 + {n}n2 = {2n + 1}n2 .
Here are a few further examples of arithmetic on sequences:
{2n } + {2n } = {0},
{2n } + {(2)n } = {0, 8, 0, 32, 0, 128, 0, 512, 0, . . .},
2{2n } = {2n+1 },
{2n } {2n } = {1},
248
{2n 1} + {1} = {2n},
{(1)n n} + {2/n} = {(1)nn + 2/n},
{in }/{(i)n} = {(1)n }.

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(n)}.
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 Q can be scrambled into a sequence. (Hint: Use the scrambling of Q+ .)
8.1.7 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.
8.1.8 Prove that + is a binary operation on the set of all infinite sequences. Prove that
{0} is the identity for this operation. Prove that every infinite sequence has an inverse for
this operation.
8.1.9 Prove that is a binary operation on the set of all infinite sequences. Prove that
{1} is the identity for this operation. Show that not every infinite sequence has an inverse
for this operation. What infinite sequences have an inverse for this operation?
8.1.10 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.
8.2
Convergence of infinite sequences
Definition 8.2.1 A sequence s = {sn } converges to L C if for every real number > 0
there exists a positive real number N such that for all integers n > N , |sn L| < .
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,
249
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.
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 .
The sequence s = {300, 5, , 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 .
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 8.4
in terms of limits of functions. More epsilon-N proofs are in Section 8.5. 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 text between square brackets in this
font and in red color 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 > 0 some property
holds. All proofs of this form start with:] Let 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 = 1 .
Then N is a positive real number. [Now we have to show that for all integers
n > N , |sn 0| < . All proofs of statements of the form for all integers
n > N start with:] Let n be an (arbitrary) integer with n > N . [Finally, we have to
prove the inequality |sn 0| < . We do that by algebraically manipulating
the left side until we get the desired final < :]
|sn 0| = |1/n 0|
= |1/n|
= 1/n (because n is positive)
< 1/N (because n > N > 0)

1
(because N = 1/) [Ah, what a clever guess that was!]
=
(1/)
= .
250
So we conclude that |sn 0| < , which proves that lim sn = 0.
Just as in the epsilon-delta proofs where one has to find a , 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 bb
b
bb b b b
b b
b
b
s4
b
s2
We prove that lim sn = 0. Let > 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 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

n
n+1
|sn 0| = ((1) + i(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
|(1)n | + |i(1)n+1 | (by triangle inequality)
n
1
= (1 + 1)
n
1
2 (because n > N )
<
N
[Aside: we want/need 2/N , and 2/N = is a possibility, so set N = 2/.
Now go ahead, write that missing information on N in line 1 of this proof!]
251
2
(because N = 2/)
2/
= ,
=
which proves that for all n > 1/, |sn 0| < . Since 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 > 0. Set N = 2/. Then N
is a positive real number. Let n be an integer strictly bigger than N . Then

1

n
n+1

|sn 0| = ((1) + i(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
(|(1)n| + |i(1)n+1 |) (by triangle inequality)
n
1
= (1 + 1)
n
1
<
2 (because n > N )
N
2
(because N = 2/)
=
2/
= ,
which proves that for all n > 1/, |sn 0| < . Since is arbitrary, this proves that
lim sn = 0.
2
2n+3n
Example 8.2.4 Claim: lim 3+4n+n
. Let n be
2 = 3. Proof: Let > 0. Set N =
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
252
11
n
11
(because n > N )
<
N
11
(because N 11/ so 1/N 1/(11/)) [Assuming this.]

11/
= ,
=
which was desired. Now (on scratch paper) we gather all the information we used about N :
N > 0, N 8, N 11/, and that is it. Thus on the first line we fill in the blank part:
Set N = max {8, 11/}, which says that N is either 8 or 11/, whichever is greater, so that
N 8 and N 11/.
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 > 0. Set N = max {8, 11/}. Then
N is a positive real number. Let n be an integer strictly bigger than N . Then

2
2n + 3n2
2n + 3n2
3(3
+
4n
+
n
)

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
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/ so 1/N 1/(11/))
11/
= ,
which proves that for all n > N , |sn 3| < . Since is arbitrary, this proves that the limit
of this sequence is 3.

Example 8.2.6 lim
1+

1 n
n
253
= e.
Proof. All the hard work for this has been done already in Exercise 7.5.9. Let > 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 < . In particular, for any integer n > N , 1 + 1 n e < .
x
n
Example 8.2.7 lim n1/n = 1.
Proof. For all n 2, by Exercise 1.6.9,

n = (n
1/n n
) = (n
1/n
n
X
n
(n1/n 1)k .
1 + 1) =
k
n
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 thatn 1 12 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 n2 . Now let > 0. Set
N = max {2, 2/2}. Then
N is a positive real number. Let n > N be an integer. Then
2
2
1/n
0n
1 n < N = , which proves that |n1/n 1| < , and hence proves this
limit.
Example 8.2.8 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 > 0. By the previous example, there exists N > 0 such
that for all integers n > N , 0 < n1/n 1 < . Then for all integers n > max {M, N },
0 M 1/n 1 n1/n 1 < , which implies that |M 1/n 1| < . This proves the example
in case M 1.
Now suppose that M < 1. By the Archimedean property of R (Theorem 3.8.3) there
exists a positive integer B such that 1 < B. By assumption, 1/M > 1, so by the previous
case there exists N > 0 such that for all integers n > N , 0 1/M 1/n 1 < 1/B. By
adding 1 to all three parts in this inequality we get that 1 1/M 1/n < 1 + 1/B = B+1
B ,
B
1/n
1.
so that by compatibility of < with multiplication by positive numbers, B+1 < M
Hence by compatibility of < with addition,
0 1 M 1/n < 1
B
1
1
=
<
< ,
B+1
B+1
B
so that |1 M 1/n | < , which proves that lim M 1/n = 1.
254

8.2.1 Let sn = 1/n2 . Fill in the blanks of the following proof that lim{sn } = 0.
. Then if n > N ,
Let > 0. Set N =

1

|sn s| = 2 0
n
1
)
= 2
(because
n
1
< 2
(because
)
N
(because
)
=
= .
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)

= 0.
n
8.2.5 Prove that lim 2n+1
n2 4 = 0.
3
8.2.6 Prove that lim 3n+4
2n = 2 .

n + 1 n = 0.
Pn
8.2.8 Suppose that the terms of a sequence are given by sn = k=1
Pn
1
1
= 1 n+1
.
i) Using induction on n, prove that k=1 k(k+1)
ii) Use this to find and prove the limit of {sn }.
1
.
k(k+1)
8.2.9 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.
8.2.10 Let r R. By Theorem 3.8.4, for every positive integer n, there exists a rational
number sn (r n1 , r + n1 ). Prove that {sn } converges to r.
8.2.11 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 = , 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.)
8.2.12 Prove that limn { n! n! } = 1.
8.3
255
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.
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 324), lim{sn } =
6 L means:
For all real numbers > 0 there exists a positive real number N such

that for all integers n > N , |sn L| < .
= There exists a real number > 0 such that there exists a positive

real number N such that for all integers n > N , |sn L| < .
= There exists a real number > 0 such that for all positive real numbers

N , for all integers n > N , |sn L| < .

N there exists an integer n > N such that |sn L| < .
N there exists an integer n > N such that |sn L| .
Example 8.3.2 {(1)n} is divergent. Namely, for all complex numbers L, lim sn 6= L.
Proof. Set = 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 = .
The sequence in the previous example has no limit, whereas the sequence in the next
example has no finite limit:
Example 8.3.3 For all complex numbers L, lim{n} =
6 L.
Proof. Set = 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 reverse triangle inequality,
|n L| |n| |L| 53 = .
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:
256
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.
Theorem 8.3.6 (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 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 8.3.7 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 8.3.8 lim n n1 = .

2
Proof. Note that for all integers n > 2, n n1 = 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 n1 = .
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
1
n
N
n2 1
=n >
M.
n
n
2
2

Example 8.3.9 lim
257
n! = .
Proof. Let N be any positive integer. Then for all integers n N ,
N!
n! = n(n 1) (N + 1) N ! N nN N ! = N n N .
N
q
Thus n n! N n NNN! . By Example 8.2.8 then lim n n! N , and since N is arbitrary, the
conclusion follows.
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 > 0. By the definition of infinite limits, there
exists a positive number N such that for all integers n > N , sn > 1/. Then for the same
n, 0 s1n < , so that | s1n | < . 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 = .
8.3.1 Prove that the following sequences diverge:
i) { n}.
ii) {2
}.
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.
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.
258
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)
n3 }.
n
n
iv) { (1)
n+1 }.
n
v) { 45n }.
3
8n
vi) { nn2 +8n
}.
2
vii) { 1n
n }.
2n
viii) { n! }.
n!
}.
ix) { (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:

i) A sequence {sn } of non-zero terms such that lim{sn /sn+1 } = 0.
ii) A sequence {sn } of non-zero terms such that lim{sn+1 /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 } = .
8.4
Convergence theorems via functions
In this section we exploit the connection between sequences and functions to prove
theorems about limits of sequences more easily. Because of having to keep in mind these
connections as well as theorems about limits of functions to get at theorems about limits of
sequences, a reader may find this section fairly abstract. In Section 8.5 we give epsilon-N
proofs of the same results. The reader may omit one of these two sections. (Exercises are
almost identical in the two sections.)
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 limx0 f (x) with standard theorems from Section 4.3,
Section 8.4: Convergence theorems via functions
259
Theorem 8.4.1 Let s, f be as above. Then lim sn = L if and only if limx0 f (x) = L.
Proof. () Suppose that lim sn = L. We have to prove that limx0 f (x) = L. Let > 0.
By assumption lim sn = L, there exists a positive real number N such that for all integers
n > N , |sn L| < . Let = 1/N . Then is a positive real number. Let x be in the
domain of f such that 0 < |x 0| < . Necessarily x = 1/n for some positive integer
n. Thus 0 < |x 0| < simply says that 1/n < = 1/N , so that N < n. But then by
assumption |f (1/n) L| = |sn L| < , which proves that limx0 f (x) = L.
() Now suppose that limx0 f (x) = L. We have to prove that lim sn = L. Let
> 0. By assumption limx0 f (x) = L there exists a positive real number such that for
all x in the domain of f , if 0 < |x 0| < then |f (x) L| < . Set N = 1/. Then N
is a positive real number. Let n be an integer greater than N . Then 0 < 1/n < 1/N = ,
so that by assumption |f (1/n) L| < . Hence |sn L| = |f (1/n) L| < , which proves
that lim sn = L.
Example 8.4.2 (Compare the reasoning in this example with the epsilon-N proofs of
52/n
5n2 1/n
Section 8.2.) Let sn = 5n2
3n+4 . We note that sn = 3n+4 1/n = 3+4/n . The corresponding
function f : {1/n : n N+ } is f (x) = 52x
3+4x , and by the scalar, sum, difference, and
quotient rules for limits of functions, limx0 f (x) = 50
3+0 = 5/3, so that by Theorem 8.4.1,
lim sn = 5/3.
Example 8.4.3 Suppose sn =
3n+2
n2 3 .
1/n2
3/n+2/n2
3n+2
n2 3 1/n2 =
13/n2 .
2
3x+2x
13x2 , and by the scalar,
Note that sn =
The
corresponding function f : {1/n : n N+ } is f (x) =

sum,
difference, product, and quotient rules for limits of functions, lim sn = limx0 f (x) =
0+0
= 0.
10
Theorem 8.4.4 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 8.4.1,
limx0 f (x) = L and limx0 f (x) = L . By Theorem 4.3.1, L = L . This proves uniqueness
of limits for sequences.
Theorem 8.4.5 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 limx0 f (x) = L, by Theorem 4.3.2
there exists > 0 such that for all x in the domain of f , if x < then |f (x)| > |L|/2.
Set N = 1/. Let n be an integer strictly greater than N . Then 1/n < 1/N = , so
|sn | = |f (1/n)| > |L|/2.
260
Theorem 8.4.6 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.
Proof. Let f, g : {1/n : n N+ } C be given by f (1/n) = sn , g(1/n) = tn . By
Theorem 8.4.1, limx0 f (x) = L and limx0 g(x) = K. Theorem 4.3.3 proves parts (3),
(4), (5), (6) for f, g, hence via Theorem 8.4.1 also for s, t.
For part (1), declare f (1/n) = c and for part (2), declare f (1/n) = 1/n. Again
Theorems 4.3.3 and 8.4.1 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 8.4.7 (Power, polynomial, rational rules for sequences) Let f be a polynomial 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 8.4.8 (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 8.4.1, limx0 f (x) = L, and by assumption limxL g(x) = g(L). Thus by the composite function theorem (Theorem 4.3.9),
limx0 g(f (x)) = g(L), so that by Theorem 8.4.1, 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 8.4.9 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
have:
261
Theorem 8.4.10 A sequence {sn } of complex numbers converges if and only if the sequences {Re sn } and {Im sn } of real numbers converge.
Proof. Let f (1/n) = sn . By Theorem 4.3.8, limx0 f (x) = L if and only if
limx0 Re f (x) = Re L and limx0 Im f (x) = Im L, which is simply a restatement of
the theorem.
Theorem 8.4.11 (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 8.4.1, limx0 f (x) and limx0 g(x)
both exist, by Theorem 4.3.10, limx0 f (x) < limx0 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 8.4.12 (The squeeze theorem for sequences) Suppose that s, t, u are sequences 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 8.4.1 that limx0 f (x) = limx0 h(x). Then by the squeeze theorem
for functions (Theorem 4.3.11), limx0 f (x) = limx0 g(x) = limx0 h(x). Hence by
Theorem 8.4.1, lim s = lim t = lim u.
Example 8.4.13 lim(n + 1)1/n = 1.
1/n

3 1/n
n1/n . Thus by the previous

Proof. Observe that for n 2, 1 n n+1
n
2
o
1/n
theorem and by Example 8.2.7, lim n+1
= 1. Hence by Theorem 8.4.6 and by
n
o
n
n

o
1/n
n+1 1/n
n+1 1/n 1/n
n
= lim
lim{n1/n} = 1.
Example 8.2.7, lim{(n + 1) } = lim
n
n
Exercises for Section 8.4 are at the end of the Section 8.5.
262
8.5
Convergence theorems via epsilon-N proofs
All of the theorems below were proved in the previous section (Section 8.4) with
a different method; here we use the epsilon-N formulation for proofs without explicitly
resorting to functions whose domains have a limit point. The reader may omit one of these
two sections. (Exercises are almost identical in the two sections.)
Theorem 8.5.1 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 > 0, there exists an N such that |sn L| < /2 for all n > N . Likewise, for
any > 0, there exists an N such that |sn L | < /2 for all n > N . Then by the triangle
inequality, |L L | = |L sn + sn L | |L sn | + |sn L | < /2 + /2 = . Since is
arbitrary, by Theorem 2.7.13 it must be the case that |L L | = 0, i.e., that L = L .
Theorem 8.5.2 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 2.7.12),
|sn | = |sn L + L| = |(sn L) + L| |L| |sn L| > |L| |L|/2 = |L|/2.
Theorem 8.5.3 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 t 6= 0, then lim{sn /tn } = L/K.
Proof. Part (1) was proved immediately after Definition 8.2.1. Part (2) was Example 8.2.2.
Part (3): Let > 0. Since lim sn = L, there exists a positive real number N1 such
that for all integers n > N1 , |sn L| < /2. Since lim tn = K, there exists a positive real
number N2 such that for all integers n > N2 , |tn K| < /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 triangle inequality)
263
< /2 + /2 (since n > N N1 , N2 )

= .
This proves (3).

Part (4): Let > 0. Note that /(|c| + 1) is a positive number. Since lim sn = L, there
exists N such that for all integers n > N , |sn L| < /(|c| + 1). Then for the same n,
|c|
< . Since is any positive number, we
|csn cL| = |c| |sn L| |c|/(|c| + 1) = |c|+1
conclude that csn converges to cL.
Part (5): Let > 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| < /(2|K| + 1). By assumption
lim{tn } = K there exists N3 such that for all integers n > N3 , |tn K| < /(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 triangle inequality)

= |sn (tn K)| + |(sn L)K|
= |sn | |tn K| + |sn L| |K|
+
|K|
< (|L| + 1)
2|L| + 2 2|K| + 1
< +
2 2
= ,
which proves (5).

Part (6): Let > 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|
2 . By assumption
2
|K|
lim{tn } = L there exists N3 such that for all integers n > N3 , |tn K| < 4(1+|L|)
. Since
K 6= 0, by Theorem 8.5.2 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 triangle inequality)

K tn
K tn
264
1
1 1
+ |L sn |
|K| |tn |
K
2
|K|
1 2
|K| 1
<
(1 + |L|)
+
4(1 + |L|)
|K| |K|
2 |K|
= +
2 2
= ,
= |K tn ||sn|
which proves (6).

Example 8.5.4 (Compare with Example 8.4.2.) Suppose sn =
5n2 1/n
3n+4 1/n
5n2
.
3n+4
To prove that
52/n
3+4/n .
=
By the linear rule, lim 1/n = 0, so by
lim sn = 5/3, we note that sn =
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.
Example 8.5.5 (Compare with Example 8.4.3.) Let sn =
2
1/n
1/n2
3/n+2/n
13/n2
3n+2
.
n2 3
Note that sn =
3n+2
n2 3
=
. 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 8.5.6 (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, limn {1/n} = 0. If limn {1/nm1 } = 0, then by the product
rule, limn {1/nm} = limn {1/nm1 1/n} = limn {1/nm1 } limn {1/n} = 0.
Thus by mathematical induction (on m), 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 just-proved power rule, limn 1/nm = 0 for all
positive integers m, so that by the scalar rule, for all m = 1, . . . , k, limn am (1/n)m = 0.
Thus by the constant rule and the repeated sum rule,

lim f (1/n) = lim a0 + a1 (1/n) + a2 (1/n)2 + + ak (1/n)k = a0 = f (0).
n
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).
265
Theorem 8.5.7 (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 > 0. Since g is continuous at L, there exists a positive number > 0 such
that for all x in the domain of g, if |x L| < then |g(x) g(L)| < . Since lim sn = L,
there exists a positive number N such that for all integers n > N , |sn L| < . Hence for
the same n, |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 8.5.8 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 8.5.9 A sequence {sn } of complex numbers converges if and only if the sequences {Re sn } and {Im sn } of real numbers converge.
Proof. By Theorem 8.5.8 it suffices to prove that if lim{Re sn } = a and lim{Im sn } = b,
then lim{sn } = a + bi. Let > 0. By assumptions there exist positive real numbers N1 , N2
such that for all integers n > N1 | Re sn a| < /2 and such that for all integers n > N2
| Im sn b| < /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| < /2 + /2 = .
Theorem 8.5.10 (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 8.5.8, lim |sn | = |L|, lim |tn | = |K|.
Suppose that |L| > |K|. Set = (|L| |K|)/2|. By the definition of convergence, there
exist N1 , N2 > 0 such that if n is an integer, n > N1 implies that < |sn | |L| < , and
n > N2 implies that < |tn | |K| < . 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
266
have that |tn | < |K| + = (|L| + |K|)/2 = |L| < |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.
Theorem 8.5.11 (The squeeze theorem for sequences) Suppose that s, t, u are sequences 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 > 0. Since lim s = L, there exists a positive N1 such
that for all integers n > N1 , |sn L| < . Since lim u = L, there exists a positive N2 such
that for all integers n > N2 , |un L| < . Set N max {N1 , N2 }. Let n be an integer
strictly greater than N . Then < sn L tn L un L < , so that |tn L| < .
Since was arbitrary, this proves that lim t = L.
Example 8.5.12 lim(n + 1)1/n = 1.

n+1 1/n
3 1/n
n1/n . Thus by the previous

n
2

n+1 1/n
= 1. Hence by Theorem 8.5.3 and by
)
o
n
nn
n
o
n+1 1/n
n+1 1/n 1/n
= lim
lim{n1/n} = 1.
n
lim
n
n
Proof. Observe that for n 2, 1

theorem and by Example 8.2.7, lim{
1/n
Example 8.2.7, lim{(n + 1)
}=
Exercises for Sections 8.4 and 8.5

8.5.13 Compute the following limits:
i) lim n212 .
2
3n +2000
ii) lim 2n
3 2000 .
2
iii) lim 5n2n

2 n1 .
2
+4n2
iv) lim n 4n
2 +2 .
3
+2
v) lim 3n
2n3 +n .
4
2n +2
vi) lim 3n2
+n1 +2 .
8.5.14 Give an example of a sequence {sn } and a number L such that lim |sn | = |L| but
{sn } does not converge.
o
n
(1)n
= 0.
n
8.5.16 Let {sn } be a convergent sequence of positive real numbers. Prove that lim sn 0.
m
8.5.17 Prove that for all integers m, limn n+1
= 1.
n
8.5.18 Prove that limn {3 n2 } = 3.
Section 8.6: Bounded sequences, monotone sequences, ratio test
267

3

n + 1 3 n = 0.
n

= e. Determine the following limits:
8.5.20 By Example 8.2.6, lim 1 + n1
n+1 n
i) n n
. o

1 2n
.
ii) 1 + 2n
o
n

2n
.
iii) 1 + n1

1 n
.o
iv) n 1 + 2n

n+1
v) 1 + n1
.
8.5.21 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 .)
8.6
Bounded sequences, monotone sequences, ratio test
Definition 8.6.1 A sequence {sn } is bounded if there exists a positive real number M
such that for all integers n, |sn | < M .
n

4+i
n
}, {(1)n }, {34in}, { 3n4n
n} are bounded
Sequences {4+5i}, {1/n}, { 5+2i
2 4 }, {
2
(the latter by Example 1.5.6), but {n}, {(1)nn}, { n n+1 }, { n n!}, {2 n } are not bounded.
Definition 8.6.2 A sequence {sn } of real numbers is called non-decreasing (resp. nonincreasing, 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 }n1 is neither increasing nor decreasing, but
2
{ n n+5 }n2 is strictly increasing.
Theorem 8.6.3 (Bounded monotone sequences) Let {sn } be a bounded sequence
of real numbers such that for some integer N , {sn }nN is non-decreasing (resp. nonincreasing). Then lim sn exists, and equals the least upper bound (resp. greatest lower
bound) of the set {s1 , s2 , s3 , . . .}.
Proof. Suppose that for all n N , sn sn+1 . By the Least upper theorem (Theorem 3.8.5), the least upper bound of the set {sN , sN +1 , sN +2 , . . .} exists. Call it L.
Let > 0. Since L is the least upper bound, there exists a positive integer N N
such that 0 L sN < . Hence for all integers n > N , sN sn , so that
0 L sn L sN < ,
which proves that for all n > N , |sn L| < . Thus lim sn = L.
The proof of the case of sn sn+1 for all n N is similar.
268
Theorem 8.6.4 (Ratio

test
for sequences) Let {sn } be a sequence of non-zero complex
sn+1
numbers such that lim sn = L. Suppose that L < 1. Then lim sn = 0.
Proof. Let > 0. Let r be a real number strictly between L and 1. Then r and r L are

positive numbers. Since lim sn+1
sn = L, there exists a positive number N1 such that for all

L
integers n > N1 , sn+1
< r L. This means that

sn

sn+1 sn+1
sn+1

sn = sn L + L sn L + L r L + L = r.
Let n0 be the smallest integer strictly greater than N1 . Then by easy induction (see
Exercise 1.5.26), for all n n0 , |sn | r nn0 |sn0 |.
For all positive integers n, r n is positive, so that r < 1 implies that r n+1 < r n . Thus
the sequence {r nn0 }nn0 is a non-increasing sequence that is bounded below by 0 and
above by 1. By Theorem 8.6.3, K = inf {r nn0 : n n0 } is the limit of {r nn0 }nn0 . Since
0 is a lower bound and K is the greatest of lower bounds of {r nn0 : n n0 }, necessarily
0 K.
Suppose that K > 0. Then K(1r)/(2r) is a positive number. Since K is the infimum,
there exists a positive integer N > n0 such that 0 r N K < K(1r)/(2r).
By multiplying

by r we get that r N +1 < rK + K(1 r)/2 = K r +
K(1r)
2
= K (1+r)
< K (1+1)
= K,
2
2
and since K is the infimum of all high powers of r, we get that K r N +1 < K, which is a
contradiction. So necessarily K = 0.
Thus for all > 0, there exists N2 R such that for all integers n > N2 , r n = |r n 0| <
|sn |+1 . Now set N = max {N1 + n0 , N2 }, and let n > N be an integer. Then
0
|sn 0| r nn0 |sn0 |

This proves that lim sn = 0.
|sn | < .
|sn0 | + 1 0
Corollary 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
n+1
n+1
suppose that r 6= 0. Then r rn = r, so that lim r rn = r. By assumption |r| < 1, the
previous theorem applies, and proves that lim r n = 0.
269

8.6.1 For each of the following sequences {sn }, you may wish to compute the limit of {|sn |}
via some comparison, ratio test, decreasing property, or other tricks, to prove the indicated.
i) limn 21n = 0.
ii) limn
iii) limn
iv) limn
v) limn
vi) limn
(1)n
= 0.
3n
(4)n
= 0.
7n
n
k
= 0 for all k C.
n!
n
kn = 0 for all non-zero k C with |k| > 1.
nm
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 =
cise 1.5.31.)
1+ 5
2 .
(Hint: Exer-
8.6.5 Let r C satisfy |r| > 1. Prove that lim r n diverges. (Hint: Check the Cauchy
property.)
8.6.6 (Ratio test again, and compare to Theorem 8.6.4) 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

is strictly increasing. (Hint: Exercise 7.5.9.)
8.6.9 Prove that the sequence 1 + n1
n
8.6.10 Prove that the sequence { n n! } is strictly increasing.
8.6.11 Let {sn } be a non-decreasing sequence.

bounded below.
Prove that the set {s1 , s2 , s3 , . . .} is
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 increasing and that
lim sn = .
ii) Suppose that {sn } is not bounded below. Prove that {sn } is decreasing and that
lim sn = .
iii) Prove that lim sn is a real number if and only if {sn } is bounded.
270
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 > 0 there exists a positive real
number N such that for all integers m, n > N , |sn sm | < .
Theorem 8.7.2 Every Cauchy sequence is bounded.
Proof. Let {sn } be a Cauchy sequence. Thus for = 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 > 0. Since lim sn = L,
there exists a positive real number N such that for all n > N , |sn L| < /2. Thus for all
integers m, n > N ,
|sn sm | = |sn L + L sm | |sn L| + |L sm | < /2 + /2 = .
Remark 8.7.4 It follows that every convergent sequence is bounded. But not every
bounded sequence is convergent or Cauchy. For example, the bounded sequence {(1)n}n
is not Cauchy as it diverges (see Example 8.3.2).
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 > 0, let N be a positive
integer such that 1/10N < . 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 < . But {sn }
does not have a limit in Q, so that {sn } is a Cauchy but not convergent sequence.
But over R and C, all Cauchy sequences are convergent, as we prove next.
271
Theorem 8.7.5 (Completeness of R, C) Every Cauchy sequence in R or C is convergent.

Proof. First let {sn } be a Cauchy sequence in R. 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 (Theorem 3.8.5), 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 8.6.3, the monotone sequence
{un } has a limit L = inf {u1 , u2 , u3 , . . .}.
We claim that L = lim{sn }. Let > 0. Since {sn } is Cauchy, there exists N1 > 0
such that for all integers m n > N1 , |sn sm | < /2. Thus if we fix n > N1 , then for
all m n, we have that sm < sn + /2. But un is the least upper bound on all sm for
m n, so that sm un < sn + /2, and in particular, sn un < sn + /2. It follows
that |sn un | < /2. Since L = inf {u1 , u2 , u3 , . . .}, there exists an integer N2 such that
0 uN2 L < /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 < /2. Hence
|sn L| = |sn un + un L| |sn un | + |un L| < /2 + /2 = .
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 8.4.10 or
Theorem 8.5.9, {sn } converges to a + bi.
Example 8.7.6 The sequence {1 +
is monotone, and is not bounded.
1
2
1
3
+ + n1 }n is not Cauchy, does not converge,
Proof. Let sn be the nth term of the sequence. If m > n, then

sm sn =
1
1
1
mn
n
1
+
+
+ +
=1 .
m m1 m2
n+1
m
m
Thus if m = 1000n, then sm sn > 1 1/1000 = 0.999. In particular, if = 1/2, then

|sm sn | <
6 . 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.)
272

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) Under what conditions is {sn /tn } a Cauchy sequence?
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 Give examples of sequences with the listed properties, if they exist. If they do not
exist, justify.
i) A convergent sequence that is not Cauchy.
ii) A Cauchy sequence that is not convergent.
iii) A bounded sequence that is not Cauchy.
iv) An real increasing sequence that is not Cauchy.
v) A real Cauchy sequence that is increasing.
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 }
n1 ,
{skn }n1 ,
{skn }nN+ .
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} are subsequences of {(1)n}. The
constant sequence {1} is not a subsequence of {1/n}, because the latter sequence does not
have infinitely many terms equal to 1. If {sn } is the sequential enumeration of Q+ on
n
n
} are subsequences, but { n+2
} = {1/3, 1/2, . . .} is not.
page 245, then {1/n}, {n}, { n+1
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 > 0. By assumption there exists a positive number N such that for all integers
n > N , |sn L| < . Since n kn , it follows that |skn L| < . Thus {skn } converges.
The proof of the second part is similar.
Section 8.8: Subsequences
273
Theorem 8.8.4 Every bounded sequence has a Cauchy subsequence.

Proof. First suppose that {sn } is a bounded sequence of real numbers. Let M be a positive
real number such that for all n, |sn | M . Let a0 = M and b0 = M . The sequence {sn }
has infinitely many (all) terms on the interval [a0 , b0 ]. Set l0 = 0.
We prove that for all m N+ there exists a subsequence {skn } all of whose terms
are in the interval [am , bm ], where bm am = 2m (b0 a0 ) = 2m 2M and [am , bm ]
[am1 , bm1 ]. Furthermore, we prove that there exists lm > lm1 such that slm [am , bm ].
Namely, given a subsequence {skn } on [am1 , bm1 ], we define
h
i
h
i
am1 +bm1
am1 +bm1
, if infinitely many sn lie in am1 ,

;
am1 ,
2
2
[am , bm ] = h
i
am1 +bm1 , bm1 , otherwise.

2
In other words, [am , bm ] is the first half of the interval [am1 , bm1 ] if that interval contains
infinitely many terms of the subsequence, and otherwise [am , bm ] is the second half of the
interval [am1 , bm1 ]. By the choice of [am , bm ], the subsequence {skn } has an infinite
subsequence that lies in [am , bm ], and that subsequence is also a subsequence of the original
{sn }. The facts bm am = 2m 2M and [am , bm ] [am1 , bm1 ] follow easily from the
construction. Furthermore, since [am , bm ] contains infinitely many terms of {sn }, there
exists lm > lm1 such that slm [am , bm ].
By construction {sln }n is a subsequence of {sn }n . We next prove that {sln }n is a

Cauchy sequence. Let > 0. By Exercise 3.8.3 there exists a positive integer N such
that 1/2N < /(2M ). Let m, n > N be integers. Then slm , sln are in [aN , bN ], so that
|slm sln | bN aN = 2N 2M < .
This proves that every bounded sequence {sn } of real numbers has a Cauchy subsequence.
Now suppose that {sn } is a bounded sequence of complex numbers. Then {Re sn }
and {Im sn } are bounded sequences of real numbers. Thus by the first part there exists a
subsequence {Re skn } that is Cauchy. But {Im skn } is still bounded, and by the first part
this subsequence has a Cauchy subsequence {Im sln }. Since {Re sln } is a subsequence of the
Cauchy sequence {Re skn }, it follows that {Re sln } is Cauchy. But then by Exercise 8.7.1,
{sln } is Cauchy.
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.
274
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 [4, 4]. On this
subinterval the second term equals 0. Infinitely many terms lie on [4, 0], in particular the
third term 2. Infinitely many terms lie on [4, 2], in particular the fifth term 2. After
this all terms are 2, so that we have built the Cauchy subsequence {0, 2, 2, 2, . . .}.
We could have built the Cauchy subsequence {0, 0, 2, 2, . . .}. If we started with the
interval [8, 8], we could have built the Cauchy subsequence {0, 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:
{sn }
convergent {sn }
{(1)n}
{(1)n 1}
{(1)n + 1/n}
{(1)nn + n + 3}
{in }
{n}
{(1)nn}
the set of subsequential limits of {sn }

{lim sn }
{1, 1}
{0, 2}
{1, 1}
{3, }
{i, 1, i, 1}
{}
{, }
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 kn1 has been
chosen, choose an integer kn > kn1 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 = .
275

8.8.1 Find the subsequential limit sets of the following sequences:
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 245. Prove that the set of its
subsequential limits equals R0 .
n 2
o
n
n(n+1)
8.8.3 Prove that the real-valued sequence n +(1)
is unbounded. Find a subsen+1
quence that diverges to and a subsequence that converges to 1.
8.8.4 Let c be the complex number of absolute value 1 at angle 2/n 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.
8.9
Liminf, limsup for real-valued sequences
(I do not go into great detail in this section.)

Recall that by Theorem 3.8.5, 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.

We extend the same definitions also to sequences (thought of as sets). Thus 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.
276
We combine infima and suprema of tail ends of sequences:
Definition 8.9.1 Let {sn } be a real-valued sequence. We define limit superior lim sup
and limit inferior lim inf of {sn } as follows:
lim sup sn = inf {sup {sn : n m} : m 1},
lim inf sn = sup {inf {sn : n m} : m 1}.
In other words, lim sup is the infimum of the set of all the suprema of all the tail-end
subsequences of {sn }, and analogously, lim inf is the supremum of the set of all the infima
of all the tail-end subsequences of {sn }. In the plot below, the sequence oscillates between
positive and negative values but with peaks getting smaller and smaller. The red dots
denote the sequence {sup {sm : m n}}n and the blue dots the sequence {inf {sm : m
n}}n :
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 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 (Theorem 3.8.5), lim inf sn and
lim sup sn are real numbers.
For lim inf sn to be a real number it is not enough that the sequence 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
that the sequence be bounded above.
277
Theorem 8.9.2 If {sn } converges to L, then lim inf sn = lim sup sn = L.

Proof. Let > 0. Then there exists N > 0 such that for all integers n > N , |sn L| < .
Thus for all m N , sup {sn : n m} L + , so that L lim sup sn L + . Since
this is true for all > 0, it follows by Theorem 2.7.13 that L = lim sup sn . The other part
is left to the reader.
It turns out that there is an important connection between limsup, liminf, and subsequential limits:
Theorem 8.9.3 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 subsequential limits of {sn }, and let U = sup (S). Since the sequence is bounded, A and U are
real numbers.
Let > 0. Since A = inf {sup {sn : n m} : m 1}, there exists m0 1 such that
sup {sn : n m0 } A < . Thus for all n m0 , sn A < . But then any subsequential
limit of {sn } is a subsequential limit of {sn }nm0 , so that this limit must be at most A + .
Thus A + is an upper bound on all subsequential limits of {sn }, so that U A + . Since
is an arbitrary positive number, by Theorem 2.7.13 this means that U A.
By definition of U , there exists a convergent subsequence {skn } such that U
lim{skn }n < /2. Let L = lim{skn }n . So U L < /2, and there exists a positive real
number N such that for all integers n > N , |skn L| < /2. Thus for all n > N ,
skn > L /2 > (U /2) /2 = U .
Thus for any integer m N , the supremum of {sn : n > m} must be at least U , so
that A, the limsup of {sn } must be at least U . Hence by Theorem 2.7.13 this means
that A U .
It follows that A = U .
Theorem 8.9.4 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 > 0. Recall that a = inf {sup {sn : n m} : m 1}. Then there exists a positive
integer m such that
sup {sn : n m} < a sup {sn : n m}.
In particular for all n > m, sn < a. By possibly increasing m, we similarly get that
for all n > m in addition tn < b. Thus for all n > m, sn + tn 2 < a + b. Thus
278
c = inf {sup {sn + tn : n m} : m 1} a + b + 2. Since is arbitrary, it follows that

c a + b, which proves the first part. The rest is left as an exercise.
Theorem 8.9.5 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 > 0. Set = min {L/2, }. By assumption there exists N > 0 such that for
all integers n > N , |sn L| < . Then L < sn < L + . Thus each sn is positive, and
in fact |sn | > L/2. It follows that (L )tn sn tn (L + )tn . But then since L 0,
(L ) lim sup tn = lim sup(L )tn lim sup sn tn lim sup(L+ )tn = (L+ ) lim sup tn .
The proof of the liminf part is similar.
8.9.1 Suppose that {sn } converges to L. Finish the proof of Theorem 8.9.2, namely prove
that lim inf sn = L.
8.9.2 Find bounded real-valued sequences {sn }, {tn} such that lim sup sn + lim sup tn >
lim sup(sn + tn ). (Compare with Theorem 8.9.4.)
8.9.3 Let {sn }, {tn } be bounded sequences in R.
i) Finish the proof of Theorem 8.9.4, 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 ).
8.9.4 Suppose that lim sn = . Prove that the set of subsequential limits of {sn } is
empty.
8.9.5 Compute lim inf and lim sup for the following sequences.
i) { sinn n }.
ii) {(1)n n!}.
iii) {2n }.
iv) {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}.
v) The sequence of all positive prime numbers.
vi) The sequence of all multiplicative inverses of positive prime numbers.
8.9.6 Finish the proof of Theorem 8.9.3, namely prove that the infimum of the set of all
subsequential limits of a bounded sequence equals the liminf of the sequence.
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 Exercises 9.2.11 and 9.2.12.
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 corresponding 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 correP
sponding to the sequence {an } is k=1 ak (whether this infinite sum makes sense or
not).
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 a1a }n . In parn
ticular, the sequence of partial sums of {(1)n} is {{ (1)2
}n = {1, 0, 1, 0, 1, 0, . . .}.
280
P
Pn
Definition 9.1.3 The series k=1 ak converges to L C if the sequence { k=1 ak }n
P
converges to L. We say then that L is the sum of the series and we write k=1 ak = L.
If the series does not converge, it diverges.
Since convergence of series depends on the convergence of sequences, a series may
diverge to or to , or it may simply have no limit.
The following follows immediately from the corresponding results for sequences:
Theorem 9.1.4 Let A =
(1) If A, B C, then
k=1
ak , B =
k=1 bk ,
and c C.
X
(ak + cbk ) = A + cB.
k=1
(2) If all ak , bk are real numbers and A, B R {, }, then
,
if A = , B R0 {} and c 0;
,
if A = , B R0 {} and c 0;
,
if
A = , B R0 {} and c 0;
, if A = , B R0 {} and c 0;
X
(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=0 ak } and { k=0 bk } converge to A and B in
C, respectively. By the theorem on the convergence of sums of sequences (Theorem 8.5.3)
Pn
Pn
Pn
then k=0 (ak + cbk ) = k=0 ak + c k=0 bk converges to A + cB.
Other parts are proved similarly.
It is hard to immediately present examples of this because we know so few limits of
infinite series. There are examples in the exercises.
Theorem 9.1.5 If r C satisfies |r| < 1, then the geometric series
P
r
1
, so k=1 r k = 1r
.
to 1r
k=1
r k1 converges
Pn
n
Proof. By Example 1.5.4, k=1 r k1 = 1r
. By Corollary 8.6.5, lim r n = 0. Thus by the
1r
scalar and sum rules for limits of sequences (Theorem 8.4.6 or Theorem 8.5.3),

n
X
1
1
rn
1 rn
k1
=
= lim
.
lim
r
= lim
n 1 r
n
n 1 r
1r
1r
k=1
Pn
Pn
r
.
Thus limn k=1 r k = r limn k=1 r k1 = 1r
Section 9.1: Infinite series
281
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,
k1

X
3
3 10
3
3 X 1
1
1
=
=
=
= .
1
k
10
10
10
10 1 10
10 9
3
k=1
k=1
A little more work is expressing 5.523 as a fraction:
X
523
523
523
5.523 = 5 +
+
+
+
=
5
+
523
1000 10002 10003
k=1
= 5 + 523
=
1
1000
1
1000
5518
.
999
= 5 + 523
1
1000k
1
523
4995 + 523
=5+
=
1000 1
999
999
P
Example 9.1.6 The harmonic series k=1 k1 diverges to by Example 8.7.6.
P
Example 9.1.7 The series k=1 k12 converges.
Proof. By Exercise 1.5.25 and Theorem 9.1.5,
0
P2n 1
n
2X
1
k=1
n+1
X 1
X
1
1
1
=
= 2,
2
k1
k1
k
2
2
1 1/2
k=1
k=1
Pn
so that { k=1 k12 }n and { k=1 k12 }n are bounded increasing sequences of real numbers.
P
By Theorem 8.6.3, the sequence has a limit that is at most 2, so that k=1 k12 converges.
P
2
It turns out that k=1 k12 = 6 , but this is much harder to prove. (An indication of
proof is in Exercise 9.11.2.)
P
Theorem 9.1.8 If k=1 ak converges, then lim an = 0 and the sequence {an } is bounded.
P
Proof. Let L = k=1 ak . Let > 0. By assumption there exists a positive number N such
Pn
that for all integers n > N , |L k=1 ak | < /2. In particular, for integers n > N + 1,

n1
n1

X
X

ak L
ak + L
|an | = an +

k=1
k=1

n1

n1

X
X

ak + L
ak L +
an +

k=1
k=1

n

n1
X

X

ak + L
ak L +
=

k=1
< /2 + /2 = .
k=1
282
In particular, {an } is convergent, so by Theorem 8.7.3, it is Cauchy, hence by Theorem 8.7.2

it is bounded.
The converse of this theorem is of course false; see Example 9.1.6.
Theorem 9.1.9 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 + + am1 ) + k=m an .
Proof. Let sn = a1 + a2 + + an , and tn = am + am+1 + + an . By the constant

and sum rules for sequences (Theorem 8.4.6 or Theorem 8.5.3), the sequence {sn }n =
{a1 + a2 + + am1 }n + {tn }n converges if the sequence {tn } converges, and similarly,
{tn }n = {sn }n {a1 + a2 + + am1 }n converges if {sn } converges.
P
9.1.1 Let r C satisfy |r| 1. Prove that k=1 r k diverges.
P 1 P 1 P 2
9.1.2 Compute the following sums:
k=6 3k ,
k=8 5k .
k=1 2k ,
P
1
9.1.3 Prove that k=1 k(k+1)
converges, and find the sum. (Hint: Exercise 8.2.8.)
P
2k+1
9.1.4 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.)
P
P 1
1
2
2
9.1.5 Take it as fact that k=0 (2k+1)
2 = 8 . Prove that
k=1 k2 = 6 .
P xk
9.1.6 For each k N+ let xk be an integer between 0 and 9. Prove that
k=1 10k
converges.
i) What does this say about decimals?
ii) Find the sum if xk = 4 for all k.
iii) Find the sum if {xn } = {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}.
P xk
iv) Prove that whenever the sequence {xn } is eventually periodic, then k=1 10
k is a
rational number.
P
P
P
9.1.7 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
9.1.8 Suppose that lim an 6= 0. Prove that k=1 ak diverges.
9.1.9 Determine, with proof, which series converge.

P
i) k=1 k1k .

P
ii) k=1 k13 + ik .
P
1
iii)
k=1 k! .
283
9.1.10 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 (Cauchys criterion for series) The infinite series
ak converges if
and only if for all real numbers > 0 there exists a real number N > 0 such that for all
integers n m > N ,
|am+1 + am+2 + + an | < .
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
> 0 there exists N > 0 such that for all integers m, n > N , |sn sm | < . In particular
for n m > N , |am+1 + am+2 + + an | = |sn sm | < .
Now suppose that for all real numbers > 0 there exists a real number N > 0 such
that for all integers n m > N , |am+1 + am+2 + + an | < . 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. Let sn = a1 + a2 + + an , tn = b1 + b2 + + bn . Then for all integers n m,

|tn tm | = |bm+1 + bm+2 + + bn |
|bm+1 | + |bm+2 | + + |bn |

am+1 + am+2 + + an
= sn sm = |sn sm |.
Since {sn } is convergent, it is Cauchy, and the inequality above shows that {tn } is Cauchy
in C. By Theorem 8.7.5, {tn } has a limit L C, so that by definition of convergence of
P
series,
bk = L.
284
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. Let sn = a1 + a2 + + an , tn = b1 + b2 + + bn . Then for all integers n m,

|tn tm | = bm+1 + bm+2 + + bn
|am+1 | + |am+2 | + + |an |

|am+1 + am+2 + + an |
= |sn sm |.
Since {sn } is divergent, by Theorem 8.7.5 it is not Cauchy, so that by the inequality above
P
{tn } is not Cauchy. Hence
bk diverges.
Theorem 9.2.4 (Ratio
{an } be a sequence of non-zero complex numbers.

test) LetP
P
an+1
(1) If lim sup an < 1, then
|ak | and
ak converge.

P
P

|ak | and
ak diverge.
(2) If lim inf aan+1
> 1, then
n
|. Suppose that L < 1. Let r be a real number in the

Proof. Let L = lim sup | aan+1
n
|
: n m} : m 1} and r > L, it
open interval (L, 1). Since L = inf {sup { |a|an+1
n|
|
follows that there exists m 1 such that r > sup { |a|an+1
: n m}. Thus for all n
n|
m, |an+1 | < r|an |. Thus by Exercise 1.5.26, |am+n | < r n |am |. The geometric series
P k
P
k
k r converges by Theorem 9.1.5, and by Theorem 9.1.4,
k am r converges. Thus by
P
P
P
Theorem 9.2.2, k=1 |am+k | and k=1 am+k converge. Hence by Theorem 9.1.9,
|ak |
P
and
ak converge. This proves (1).
|, and suppose that L > 1. Let r be a real number in the open
Now let L = lim inf | aan+1
n
|
: n m} : m 1} and r < L, it follows that
interval (1, L). Since L = sup {inf { |a|an+1
n|
|
there exists m 1 such that r < inf { |a|an+1
: n m}. Thus for all n m, |an+1 | > r|an |.
n|
Thus by a straightforward modification of Exercise 1.5.26, |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
P
P
P
by Theorem 9.2.2, k=1 |am+k | and k=1 am+k diverge. Hence by Theorem 9.1.9,
|ak |
P
and
ak diverge. This proves (2).
| = 1 or
This ratio test for convergence of series does not apply when lim sup | aan+1
Pn
an+1
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,
| = lim inf | aan+1
| = 1, and the two series diverge; whereas if an = 1/n2 for all
lim sup | aan+1
n
n
n, then lim sup | aan+1
| = lim inf | aan+1
| = 1, and the two series converge.
n
n
285
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 |. As in the proof of of Theorem 9.2.4, the conclusion in (1) follows. 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+2m1 an+2m ) an+2m+1
an ,
and similarly
0 an an+1 + an+2 an+3 + + an+2m an .
P
Thus by Cauchys criterion Theorem 9.2.1, k (1)k ak converges.
P
Example 9.2.7 Recall from Example 9.1.6 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 Theorem 9.4.6.)
Theorem 9.2.8 (Integral test for series convergence) Let f : [1, ) [0, ) be a
Rn
P
decreasing function. Suppose that for all n, 1 f exists. Then k=1 f (k) converges if and
Rn
only if limn 1 f exists and is a real number.
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
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) +
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.
286
Rn
Conversely, suppose that limn 1 f exists. Let L R be this limit. By the disR3
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 8.6.3, 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 9.2.9 (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 n1 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+1
f exists. By the Fundamental Theorem of Calculus, 1 f = 1 xp dx = n p+11 . By the
1
composite rule for sequences (either Theorem 8.4.8 or Theorem 8.5.7), 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
P p
1
.
Thus
by
the
Integral
test
(Theorem
9.2.8),
k k converges.
p+1
9.2.1 For each of the following series, determine, with proof, whether they converge or
diverge. You may need to use Examples 8.2.7 and 8.3.9.
X
3i
.
i)
2
k + k4
k=1
X
1
.
ii)
k
k=1
X
1
.
iii)
k3
k=1
X
2k
iv)
.
k!
k=1
X
2k
v)
.
k3
k=1
X
1
.
vi)
k 2
k=1

9.2.2 Find a convergent series
lim sup |bn |1/n = 1.
ak and a divergent series
k bk
287
with lim sup |an |1/n =
9.2.3 Make a list of all encountered criteria of convergence for series.

9.2.4 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
| < 1. Prove that lim sup |an |1/n < 1.
n
| > 1. Prove that lim sup |an |1/n > 1.

ii) Suppose that lim inf | aan+1
n
P k
9.2.5 Apply the ratio test (Theorem 9.2.4) and the root test (Theorem 9.2.5) to k=1 5k! .
P
Was one test easier? Repeat for k=1 k1k .
P
9.2.6 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
P
9.2.7 Prove that if k=1 ak converges then k=1 akk converges.
P
9.2.8 Let an = (1)n /n, bn = 2(1)n. Then k=1 ak converges, and for all n, |bn | = 2,
P
and k=1 abkk diverges.
P
9.2.9 (Compare with Exercise 9.2.8.) 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.
9.2.10 The following guides through another proof of the p-series convergence test for
p < 1. (Confer Theorem 9.2.9 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 <
(n + 1)p+1 np+1
np+1 (n 1)p+1
and np <
.
p+1
p+1
iii) Prove by induction on n 1 that

n
X
k=1
kp
p
1
np+1 +
.
p+1
p+1
Pn
iv) Prove that the sequence { k=1 k p }n of partial sums is an increasing sequence
bounded above by p/(p + 1).
Pn
P
v) Prove that the sequence { k=1 k p }n converges, and so that k=1 k p converges.
288
9.2.11 (Order of summation in infinite sums is important.)

P
P 1
1
and k=1 2k+1
diverge. Refer to Example 9.2.7.
i) Prove that k=1 2k
Pn
P
P2n+1 (1)k
1
1
ii) Observe that k=1 2k
nk=1 2k+1
= k=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?
P
9.2.12 Let a1 , a2 , a3 , . . . be complex numbers such that k=1 |ak | converges. (The series
P
+
+
k=1 ak is then called absolutely convergent.) Let s : N N be a bijective function.
P
Prove that k=1 as(k) converges. In other words, order of summation does not matter.
Compare with Exercise 9.2.11, and comment what the difference is.
9.2.13 (Raabes test) Let a1 , a2 , . . . be positive real numbers such that for some > 1
P
for
all
n
N
.
Prove
that
and for some N N, aan+1
n an converges. (Hint: Let
n
n
f (x) = 1 x . Use the Mean Value Theorem to get c (x, 1) such that f (c)(1 x) =
P
f (1) f (x). Conclude that 1 x (1 x). Apply this to x = 1 n1 . Use that
n
converges.)
9.3
Power series
P
So far all infinite series were sums of terms of a sequence:
k=1 ak stands for the limit
of the sequence {a1 , a1 + a2 , a1 + a2 + a3 , . . .} of partial sums. Thus to obtain the sum, we
add the first term to the second term, the obtained sum to the third term, the new sum
to the fourth term, that sum to the fifth term, and so on ad infinitum. But the meaning is
clear when the infinite sum shifts the index by a finite number, such as in the following:
ak =
k=5
k=5
ak+4 ,
k=1
ak =
5
X
k=5
ak +
ak .
k=1
In this section we will be dealing with sums where the index varies through N0 , and
furthermore, the terms of the sequence will be special functions rather than constants:
Definition 9.3.1 A power series is an infinite series of the form
k=0
ak xk =
k0
ak xk = a0 + a1 x + a2 x2 + a3 x3 + ,
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 34, 00 = 1.)
289
Think through the following table of reading off the (fixed) coefficients an of the
variable power xn :
power series
X
xk
k=0
k=0
k=0
k=0
2k+1
an
an = 1
3
= x+x +x +x +
an =
1, if n is odd;
0, if n is even.
kxk
an = n
xk
k!
an =
(kx)k
an = nn
1
n!
k=0
The most important question that we address in this section is: for which x does such
an infinite sum make sense. This is the same as asking for the x-domain of the function
P
Pn
k
k
k=0 ak x . Certainly when x = 0, the partial sums are
k=0 ak 0 = a0 for all n 0, so
P
that k=0 ak xk converges to a0 . Thus 0 is always in the domain of a power series function.
P
Or course, when 1 is in the domain, evaluation of the power series k ak xk at x = 1
P
is the (ordinary) series k ak .
P
Example 9.3.2 Let f (x) = k=0 xk . By Theorem 9.1.5, the domain of f contains all
complex numbers with absolute value strictly smaller than 1, and by Theorem 9.1.8, the
domain of f contains no other numbers, so that the domain equals {x C : |x| < 1}. For
P
1
k
all x in the domain of f , by Theorem 9.1.5,
k=0 x = 1x . But note that the domain of
1
is strictly larger than the domain of f .
1x
P
P
P
1
In particular, k0 21k = 11 1 = 2, k0 31k = 11 1 = 32 , k0 0.6k = 10.6
= 2.5.
2
P
Theorem 9.3.3 (Root test for convergence of power series) Let ak xk be a power
series, and let = lim sup |an |1/n . Define R by
1/, if 0 < < ;

R = 0,
if = ;
,
if = 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 definition of limits, 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|. If
290
|x| < 1, then both of the series converge, and if |x| > 1, then the two series diverge. If
= 0, then |x| < 1 is true for all x C, so R = has the stated property. If = 0,
then |x| < 1 is true only for x = 0, so R = 0 has the stated property. If 0 < < , then
|x| < 1 is true only for all x C with |x| < 1/ = R.
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 conP
P
verges and outside of the circle the series diverges. Note that the series
ak xk ,
|ak |xk ,
P
and
|ak ||x|k have the same radius of convergence.
Theorem 9.3.5 (Ratio test for convergence of power series) Suppose that all an
are non-zero complex numbers.

P
P
n
, then
|ak ||x|k and
ak xk converge.
(1) If |x| < lim inf aan+1
a
P
P
n
, then
|ak ||x|k and
ak xk diverge.
(2) If |x| > lim sup an+1
a
P
n
exists, it equals the radius of convergence of
ak xk .
Thus if lim an+1
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.
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
an+1 xn+1

lim sup
= |x| lim sup
an xn
an

an+1
: n m : m 1}
= |x| inf {sup
an
)
(
1
a
:m1
= |x| inf
n

inf { an+1 : n m}
=
|x|
o
n
a
n
: n m} : m 1
sup inf { an+1
|x|

lim inf
an .
an+1
If this is strictly smaller than 1, then the two series converge, which proves (1). Similarly,

an+1 xn+1
|x|
=

,
lim inf

n
n
an x
lim sup aan+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.
291
Examples 9.3.6
P k P k P 2 k P
P 2k+1
(1)
x ,
kx ,
k x ,
k(k 1)xk ,
x
all have radius of convergence 1.
Proof of the last part: The root test gives lim sup |an |1/n = lim sup{1, 0, 1, 0, 1, 0, 1, 0, . . .}
= 1, so that the radius of convergence is 1/1 = 1. The ratio test for power series is
P 2k+1
P
inapplicable here. But note 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.
k
P
(2) The radius of convergence of k=1 xk is . For this we apply the root test:
1/n
= lim sup n1n
= lim sup n1 = 0.
1
P k
=
(3) By the ratio test, the radius of convergence of k=1 xk! is lim n!1 = lim (n+1)!
n!
(n+1)!
1/n
lim(n + 1) = . The root test gives = lim sup |1/n!| | = lim sup 1/n!1/n, and
by Example 8.3.9 this is 0. Thus the radius of convergence if also by the root
test.
We finish this section with an obvious result on products of power series whose proof
is nevertheless a bit involved.
P
P
k
k
Theorem 9.3.7 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)
P
P
the product sequence {( nk=0 ak xk ) ( nk=0 bk xk )}n converges to the power series
P Pk
k
k=0 (
i=0 ai bki )x .
P
P
P Pk
We write this as ( k=0 ak xk ) ( k=0 bk xk ) = k=0 ( i=0 ai bki )xk on B(0, R).
Proof. Let x B(0, R) \ {0}. By Theorem 9.1.8, {an xn }, {bnxn } are bounded, so there
Pn
exists M > 0 such that for all n, |an | < M/|xn | and |bn | < M/|xn |. Let cn = i=0 ai bni .
Pn
Hence |cn | i=0 |ai bni | (n + 1)M 2 /|xn|. Therefore by Examples 8.2.8 and either
8.4.13 or 8.5.12, lim sup |cn |1/n lim sup(n + 1)1/n (M 2 )1/n/|x| 1/|x| < 1/R. Hence by
P
Theorem 9.3.3, the power series k ck xk converges on B(0, R).
Let > 0. By the derivation in the previous paragraph there exists N R such that
Pn

P
for all integers n > N , k=0 ck xk k=0 ck xk < /2.
By Theorem 9.3.3
and by repetition
of the arguments above we also get that

Pn Pk
k
the sequence { k=0
i=0 |ai | |bki | |x| }n converges. In particular, the latter is a
Cauchy sequence,
so that there exists N R such that for all integers m n > N ,
P
Pm
k
k
i=0 |ai | |bki | |x| < /2.
k=n+1
By the theorem on the convergence of products of sequences (Theorem 8.5.3), the fol Pn

Pn
Pn
Pn
k
k
lowing sequence converges: { k=0 ak xk } { k=0 bk xk } = {
k=0 ak x
k=0 bk x } =
292

Pn
Pn
Pn Pn
l
{( m=0 am xm )
b
x
} = { m=0 l=0 am bl xm+l }. The mth term of this is
l
l=0
!
m X
m
m
2m
m
X
X
X
X
ak bl xk+l =
ck x k +
ai bki xk .
k=0 l=0
k=0
k=m+1
i=km
Thus for all m k, the coefficient of xk in the mth term of the product equals ck . (The
Pn
Pn
definition of a power series k=0 dk xk is as the limit of the sequence { k=0 dk xk }n , but
the sequence in the display above changes the coefficients dk with n. Thus, we cannot yet
P
conclude that the product of the two sequences converges to k ck xk .)
Thus for all n > max {N, N },

!
k
X
n
2n
k
n X

X
X
X
X

ai bki xk
ai bki xk
ai bki xk +

k=0 i=0
k=n+1 i=kn
k=0 i=0

n X
k
X
k
2n
n

X
X
X
X

ai bki xk
ai bki xk +
ai bki xk
=

k=0 i=0
k=0 i=0
k=n+1 i=kn
n k
2n

X
k
n
X X
X X

X

k
k
k
ai bki x
ai bki x +
ai bki x

k=0 i=0
k=0 i=0
k=n+1 i=kn
!
2n
n
X
X
< /2 +
|ai | |bki | |x|k
k=n+1
/2 +
< .
2n
X
k=n+1
Thus the product sequence {(

P Pk
k
k=0 (
i=0 ai bki )x .
i=kn
k
X
i=0
Pn
k=0
|ai | |bki | |x|k

ak xk ) (
Pn
k=0 bk x
)}n converges to the power series

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 (cak )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=0 ak (cx)k is convergent with radius of convergence equal to if
c = 0 and radius R/|c| otherwise.
P
P
9.3.2 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.)
293
9.3.3 Compare with the previous exercise: find the radii of convergence of the power series
P k P
P
P k P
P k P
k
k
k
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 the radius of convergence for the following series:
P k
i)
x .
P k
ii)
3x .
P
iii) (3x)k .
P k
iv)
kx .
P
v)
3kxk .
P
vi)
k(3x)k .
P xk
vii)
k3 .
P 3x
k
.
viii)
k3
P (3x)k
ix)
k3 .
9.3.5 Compute the radius of convergence for the following series:

P xk
i)
k.
P kxk
ii)
2k .
P k xk
.
iii)
(2k)k
P
k
9.3.6 What would be a sensible definition for generalized power series
k=0 ak (x a) ?
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.
P
9.3.7 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 convergence 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.
P
P
9.3.8 Let k=0 ak xk , k=0 bk xk be power series such that for all x B(0, R) for some
P

P
k
k
positive R,
= 1. In other words, at each x in B(0, R), the two

k=0 ak x
k=0 bk x
power 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 i=1 ai bki . (Hint: Theorem 9.3.7.) Suppose that there exists M R such that for all non-negative inteM +|a0 | k1
.
gers n, |an xn | < M . Prove that for all integers k 1, |bk xk | |aM
2(
|a0 | )
0|
(Hint: induction.)
294
*9.3.9 Let f (x) =
k=0
ak xk and g(x) =
k=0 bk x
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?
9.4
Taylor series
A common way of generating power series comes from approximations of functions.
Definition 9.4.1 Let a be in the domain of a function f , and assume that f has derivatives
P
f (k) (a)
of all orders at a. The Taylor series of f (centered) at a is the series
(xa)k .
k=0
k!
If a = 0, a Taylor series is a power series as defined in the previous section, and for
other a this is also a power series, but of a more general kind.
Remark 9.4.2 Let f : A C have derivatives of all orders at a. Let B = A {a} =

{x a : x A}, and let g : B C be defined by g(x) = f (x + a). By straightforward
P
P
calculus, k ak (x a)k is a Taylor series for f at a 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, a few are left for the exercises, and one exercise gives an illustration
of how Taylor series do not converge to the function for which we found the Taylor series.
Example 9.4.3 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, this series converges for all x C, not only for x R. (More on this
series is in Sections 9.7 and 9.8, where we learn more about exponentiation by complex
numbers.) Let x R. By Theorem 6.5.5 there exists d between 0 and x such that
x
n
X
xk
k=0
k!
ed
xn+1 .
(n + 1)!
295
Let > 0. Set M to be any integer bigger than or equal to |x|. By Corollary 8.6.5, there
!
. Then
exists N > M such that for all integers n > N , 21n < 22M 1M(3M
)M
e|x|
3M
|x|n+1
M n+1
(n + 1)!
(n + 1)!
(3M )M
M M M n+12M
=
M ! (M + 1)(M + 2) (2M )(2M + 1) n(n + 1)
M n+12M
(3M )M
<
M ! (2M + 1) n(n + 1)
n+12M
(3M )M 1
<
M!
2
n
M
1
2M 1 (3M )
=2
M!
2
< .
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
e =
X
xk
k=0
k!
In practice, this is how one (a human or a computer) computes values ex to arbitrary precision. For example, to compute e0.1 to within 0.0001, we take M = 1, and the smallest inteP16 0.1k
!
0.1
is
16.
Thus
to
within
0.0001,
e
is
ger N for which 21N 0.0001 22M 1M(3M
M
k=0 k! =
)
21021144462931669584965107651
0.1
, the same N works
19020718080000000000000000000 = 1.10517091807565. To compute e
P16 (0.1)k
7011749326061779176954720883
0.1
for desired precision, giving e
to be
=
=
k=0
k!
7749181440000000000000000000
0.90483741803596. To compute e to within 0.0001, we need bigger M , say M = 4,

P31 k
and then N = 31, giving e to be k=0 k! = 23.1406926327793 0.0001.
Example 9.4.4 Let f (x) = ln(x + 1), where the domain of f is (1, ). It is straightforward to compute the Taylor series for f centered at 0:
X
(1)k1 xk
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 9.1.6, 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),
k1 k
P
ln(x + 1) = k=1 (1) k x .
296
Example 9.4.5 It might be fun to look at another Taylor series. (This one is harder,
and the reader may wish to skip it.) 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
3
2n
3
1
(1 x)(2n1)/2.
f (n) (x) =
2 2 2
2
Thus the Taylor series for f centered at 0 is
X 1 3 5 (2k 3)
1
1 x
xk .
k
2
k!2
k=2
For large n, the quotient of the (n + 1)st coefficient divided by the nth coefficient equals
2n1
2(n+1) , whose limsup equals 1. Thus by the Ratio test for power series (Theorem 9.3.5),
the Taylor series converges absolutely on B(0, 1), and in particular it converges absolutely
4/3
2n1
= 2(n+1)3
on (1, 1). Furthermore, the quotient 2(n+1)
2(n+1) above is at most 1 n for all
n 4. Thus by Raabes test (Exercise 9.2.13), 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 Taylors remainder theorem (Exercise 7.4.13): the difference
between Tn,f,0 (x) f (x) is
n
Z x
Z x
1 3 5 (2n 1)
xt
2n 1
(x t)n 1 1 3
(2n+1)/2

(1 t)
dt =
dt.
n!
2 2 2
2
1t
n!2n+1 1 t
0
0
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). An application of Exercise 9.4.5 shows that
the Taylor series is continuous on [1, 1], and as f also continuous there, necessarily the
Taylor series converges to f on [1, 1].
P
Theorem 9.4.6 Let f (x) = k=0 ak xk have radius of convergence a positive real numP
ber R. Let a C with |a| = R such that k=0 ak 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(a, R) B,
then f (a) = g(a).
Proof. Let > 0.

P
By Cauchys criterion Theorem 9.2.1, since k=0 ak ak converges, there exists N
Pn
PN +m
N+ such that for all integers n N , | k=N ak ak | < /4. Let sm = k=N ak ak . By
assumption, for all m 1, |sm | < /4. Furthermore, aN = s0 /aN , and for n > N ,
aN +n = (sn sn1 )/aN +n .
297
Let r be a positive real number strictly smaller than 1. Then

N +n

X

ak (ra)k = r N s0 + (s1 s0 )r N +1 + (s2 s1 )r N +2 + + (sn sn1 )r N +n

k=N

= s0 (r N r N +1 ) + s1 (r N +1 r N +2 ) + + sn1 (r N +n r N +n ) + sn r N +n

= r N (1 r) s0 + s1 r + + sn1 r n1 + sn r n

r N (1 r) |s0 | + |s1 |r + + |sn1 |r n1 + |sn |r n

r N (1 r) 1 + r + + r n1 + r n
4
= r N (1 r n + r n )
4
= rN
4
< .
4
P

In particular,
ak (ra)k .
k=N
Since polynomial functions are continuous, there exists 1 > 0 such that for all x
PN 1
PN 1

B(a, 1 ), k=0 ak xk k=0 ak ak < /4.
Since g is continuous at a, there exists 2 > 0 such that for all x B(a, 2 ), |g(x)
g(a)| < /4. Let r (0, 1) such that 1 r < R1 min {1 , 2 }. Then |a ra| < 1 , 2 , so that

N
1
N
1

X
X

ak ak +
ak ak f (a)
|g(a) f (a)| = g(a) g(ra) + g(ra) f (ra) + f (ra)

k=0
k=0

1
N
1

NX

X

k
k
ak a f (a)
ak a +
|g(a) g(ra)| + |g(ra) f (ra)| + f (ra)

k=0
k=0

N
1
N
1

X
X
X
X

ak ak
ak (ra)k +
ak (ra)k
ak ak +
= |g(a) g(ra)| + 0 +

k=N
k=N
k=0
k=0

N 1

N
1
X
X

X
X

k
k
k
k
< +
ak (ra) +
ak (ra)
ak a +
ak a

4
k=N
k=0
k=0
k=N
< + + + = .
4 4 4 4
Since is arbitrary, by Theorem 2.7.13, f (a) = g(a).
Now we can continue Example 9.4.4: since ln(x + 1) is continuous on its domain, and
since its Taylor series converges at x = 1, it follows that
X
(1)k1
k=1
= ln 2.
298

1
centered at 0.
9.4.1 Compute the Taylor series of f (x) = 1x
i) Determine the radius of convergence.
ii) What is the domain of f and what is the domain of its Taylor series?
1
9.4.2 Compute the Taylor series for f (x) = 1x
centered at 2, and determine its radius
of convergence (this means that you take coefficients of powers of x 2 rather than powers
of x). Compare with the previous problem.
9.4.3 Let f (x) = ln(x+1), where the domain of f is (1, ) as in Example 9.4.4. 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.
f (x) =
e1/x ,
0,
if x 6= 0;
if x = 0
as in Exercise 7.5.15.
i) Compute 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.
P
9.4.5 Suppose that a0 , a1 , a2 , . . . are complex numbers and that
k ak converges. Let
P
f (x) = k ak xk . The goal of this exercise is to prove that f is defined and continuous on
B(0, 1) {1}.
P
Set c = k ak , and for n 0, set sn = a0 + a1 + + an L.
i) Prove that for every > 0 there exists N R such that for all integers n N ,
|sn | < .
Pn
ii) Prove that the sequence {(1 x) k=0 sk xk }n converges for all x B(0, 1) {1}.
P
iii) Prove that f (x) L = (1 x) k=0 sk xk and that therefore f is defined on
B(0, 1) {1}.
iv) Prove that the radius of convergence of f is at least 1.
v) Prove that f is continuous on B(0, 1). (Hint: Invoke a theorem.)
vi) Let > 0. Prove that there exists N R such that for all integers n N , and all

P
k
< .
x B(0, 1) {1}, (1 x)
k=n sk x
vii) Let > 0. Prove that for every positive integer n there exists > 0 such that for

Pn
all x (B(0, 1) {1}) B(1, ), (1 x) k=0 sk xk < .
viii) Let > 0. Prove that there exists > 0 such that for every x (B(0, 1) {1})

P
B(1, ), (1 x) k=0 sk xk < .
ix) Prove that f , as a function on B(0, 1) {1}, is continuous.
299
*9.4.6 In introductory analysis courses we typically handle long, even infinite, sums but
not so much long products. In Example 9.4.5 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
135(2n3)
{
}n converges to 0. Can you prove this directly without involving Raabes
n!2n
test?
9.5
Differentiation of power series
Power series are functions, and perhaps they are differentiable in their domains. Recall
(x)
, and any power series
that for any differentiable function f , f (x) = limh0 f (x+h)f
h
P
P
Pn
k
k
k
k=0 ak x is actually the limit of a sequence:
k=0 ak x = lim{
k=0 ak x }n . Thus
Pn
Pn
X
lim{ k=0 ak (x + h)k }n lim{ k=0 ak xk }n
k
ak x ) = lim
(
.
h0
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
(
ak x ) = lim lim
k=0
h0 n
n
X
k=0
ak
(x + h)k xk
.
h
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 xk1 .
k=0
n h0
k=0
In fact, it turns out that this is the correct derivative, but our reasoning 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, and we prove that the
radii of convergence of a series and its derivative series are the same.
P
P
P
k
k1
k1
Theorem 9.5.1 The series
=
have the
k=0 ak x and
k=0 kak x
k=1 kak x
same radius of convergence.
P
P
k1
k
Proof.
Certainly the radii of convergence of
, and
=
k=1 kak x
k=1 kak x
P
k
1/n
= 1, so that by Theorem 8.9.5,
k=0 kak x are the same. By Example 8.2.7, lim n
1/n
1/n
1/n
1/n
lim sup |nan |
= lim sup n |an |
= lim sup |an | . Thus by the root test for series
P
P
(Theorem 9.2.5), the radii of convergence of k=1 kak xk1 and k=0 ak xk are the same,
so that the radii of convergence of all the listed series are the same.
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.
300
P
Theorem 9.5.2 Let k ak xk have radius convergence R. Let c C satisfy |c| < R. Then
P
the function g(x) = k=1 ak (xk1 + cxk2 + c2 xk3 + + ck1 ) 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
k1
First of all,
converges on B(0, R) by Theorem 9.5.1. In particular,
k=1 kak x
P
P
k1
k1
= k=1 ak (c
+ cck2 + c2 ck3 + + ck1 ) is well-defined, but this is
k=1 kak c
simply g(c), so that c is in the domain of g.
Let > 0 and let d R satisfy |c| < d < R. Then again by Theorem 9.5.1, d is in the
P
P
P
P
domain of k ak xk , k kak xk1 , and also k k(k 1)ak xk2 and k k(k 1)|ak ||x|k2.
P
}. Then is positive. Let c C

Set D = 21 k k(k 1)|ak |dk2 , and = min {d |c|, D+1
with 0 < |x c| < . Then by the triangle inequality, |x| = |x c + c| |x c| + |c| <
P
+ |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 9.5.1 also the radius of convergence of k kak xk1 is R, so that k kak dk1
converges. Then from the Comparison theorem (Theorem 9.2.2) and from
|
n
X
k=1
ak (xk1 + cxk2 + c2 xk3 + + ck1 )|
n
X
k=1
n
X
k=1
n
X
k=1
|ak |(|x|k1 + |c||x|k2 + |c|2 |x|k3 + + |c|k1 ) (by triangle inequality)

|ak |(dk1 + ddk2 + d2 dk3 + + dk1 )
k|ak |dk1
we deduce that x is in the domain of g. Furthermore,

X
X
|g(x) g(c)| = |
ak (xk1 + cxk2 + c2 xk3 + + ck1 )
kak ck1 |
X
=|
ak (xk1 + cxk2 + c2 xk3 + + ck1 kck1 )|
X
=|
ak ((xk1 ck1 ) + (cxk2 ck1 ) + (c2 xk3 ck1 ) + + (ck1 ck1 )|
X
=|
ak ((xk1 ck1 ) + c(xk2 ck2 ) + c2 (xk3 ck3 ) + + ck2 (x c))|
X
=|
ak (x c)((xk2 + xk3 c + xk4 c2 + + ck2 )
+ c(xk3 + xk4 c + xk5 c2 + + ck3 )
+ c2 (xk4 + xk5 c + xk6 c2 + + ck4 ) + + ck2 )| (by Exercise 1.5.27)
|ak ||x c|((|x|k2 + |x|k3 |c| + |x|k4 |c|2 + + |c|k2 )
301
+ |c|(|x|k3 + |x|k4 |c| + |x|k5 |c|2 + + |c|k3 )
+ |c|2 (|x|k4 + |x|k5 |c| + |x|k6 |c|2 + + |c|k4 ) + + |c|k2 )

(by triangle inequality)
|x c|
|ak |((dk2 + dk3 d + dk4 d2 + + dk2 )
+ d(dk3 + dk4 d + dk5 d2 + + dk3 )
+ d2 (dk4 + dk5 d + dk6 d2 + + dk4 ) + + dk2 )

X
= |x c|
|ak |((k 1)dk2 + (k 2)dk2 + (k 3)dk2 + + dk2 )
X
k(k 1) k2
d
(by Example 1.5.1)
= |x c|
|ak |
2
= |x c|D
< D
< .
This proves that g is continuous at c.
P
Theorem 9.5.3 Let f (x) = k=0 ak xk have radius of convergence R. Then f is differenP
tiable on B(0, R), and for all c B(0, R), f (c) = k=1 kak ck1 . The derivative function
P
f (x) = k=1 kak xk1 also has radius of convergence R.
Proof. Let c, x B(0, R). Then
P
P
k
k
a
x
f (x) f (c)
k
k ak c
= k
xc
xc
P
ak (xk ck )
= k
xc
P
ak (x c)(xk1 + xk2 c + xk3 c2 + + ck1 )
(by Exercise 1.5.27)
= k
xc
X
=
ak (xk1 + xk2 c + xk3 c2 + + ck1 ),
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)
= lim g(x) = g(c) =
kak ck1 .
f (c) = lim
xc
xc
xc
k=1
Then the theorem follows from Theorem 9.5.1.
302

1
and its corresponding geometric series on B(0, 1).
9.5.1 Consider the function f (x) = 1x
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
9.5.2 Let f (x) =
k=0 (2k+1)k! .
i) Find the radius of convergence of the series.
ii) Compute the derivative of f .
P k
iii) By Example 9.4.3 we know that for every real x, ex = k=0 xk! . Find the series
2
for ex .
2
iv) Find a power series whose derivative is ex .
v) Address the no-closed-form discussion on page 227.
P
9.5.3 Let f (x) = k ak xk have radius of convergence R. Prove that for all c C with
|c| < R, f is continuous at c.
9.6
Numerical evaluations of some series
Differentiation of power series is a powerful tool. For all complex numbers x B(0, 1)
P
1
. Certainly it is much easier to compute
the geometric series k=0 xk converges to 1x
1
1x 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.
We also prove a theorem on what power series always take on value 0.
Example 9.6.1
k
k=1 2k1
= 4.
P
Proof. Let f (x) = k=0 xk . This is the geometric series with radius of convergence 1 that
P
P
1
k1
k1
(Example 9.3.2). By Theorem 9.5.3, f (x) =
=
,
converges to 1x
k=0 kx
k=1 kx
1
and by Theorem 9.5.1, the radius of convergence of f is also 1. Thus 2 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) =
1x
(1 x)2
k1
P
P
1
k
= f (1/2) = (11/2)
= k=0 k 21
From the two forms we deduce that k=0 2k1
2 = 4.
Section 9.7: A special function

Example 9.6.2
= 6, and
= ln 2.
k2
k=0 2k
k2
k=0 2k1
303
= 12.
P k
Proof. As in the previous example we start with the geometric series f (x) =
k=0 x
P
1
k1
that converges on B(0, 1). Its derivative f (x) =
= (1x)2 also converges
k=0 kx
P
P 2 k1
on B(0, 1). Then xf (x) =

also
k=0 kx and its derivative (xf (x)) =
k=0 k x
converge on B(0, 1). From

X
1
2x
x
2 k1
=
+
k x
= (xf (x)) =
2
2
(1 x)
(1 x)
(1 x)3
k=0

P 2
P k2
1
1
= 21 (11/2)
we deduce that k=0 2kk = 12 k=0 2k1
= 12 (4 + 8) = 6. The
+
2
(11/2)3
other part is also easy to verify.
Example 9.6.3
1
k=1 k2k
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
, so that f (x) = ln(1 x) + C for some constant C. In
Also, f (x) = k=1 xk1 = 1x
k
P
P
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).
P
Theorem 9.6.4 Let f (x) = k=0 ak xk . Suppose that there exists a positive R such that
f converges on B(0, R) and equals the zero function. Then ak = 0 for all k.
Proof. Since f has derivatives of all orders B(0, R), and since the derivative of the zero
function is 0, by induction on n we get that for all x B(0, R), f (n) (x) = 0. In particular,
n!an = f (n) (0) = 0. Thus an = 0 for all n 0.
P
9.6.1 Compute k=1 k21k .
P
1
9.6.2 Compute
k=1 k3k .
P
k
9.6.3 Consider k=1 k+23
k . According to my computer the partial sum of the first 1000
terms is a huge fraction that takes several screen pages that makes the number 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?
P
P
9.6.4 Suppose that the power series k=0 ak xk and k=0 bk xk both have radius of conP
P
vergence R > 0. Suppose that for all x B(0, R), k=0 ak xk = k=0 bk xk . Prove that
for all k, ak = bk .
304
9.7
A special function
Define the power series
E(x) =
X
xk
k=0
k!
By Examples 9.3.6 (3), the domain of E is C. Thus by Section 9.5, E is differentiable

everywhere.
Remarks 9.7.1
P
P xk1
P k
k1
(1) E (x) = k=1 kxk! = k=1 (k1)!
= k=0 xk! = E(x).
(2) E(0) = 1.
(3) Define f : C C by f (x) = E(x)E(x). Then f is a product of two differentiable
functions, hence is differentiable, and
f (x) = E (x) E(x) + E(x) E (x) (1) = E(x) E(x) E(x) E(x) = 0,
so that f is a constant function. But this constant has to equal E(0) E(0) = 1.
1
.
We conclude that E(x) is never 0, and that E(x) = E(x)
(4) 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) = 0,
(5)
(6)
(7)
(8)
so that g is a constant function. But this constant has to equal E(a)E(0) = E(a).
1
We conclude that E(x + a) = E(a) E(x)
= E(a)E(x).
Thus for all a, b C, E(a + b) = E(a)E(b).
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 .
By parts (3) and (6), E(na) = (E(a))n for all integers n and all a C.
Let a, b R. By part (5), 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 9.7.2 For all x R, E(x) = ln1 (x) = ex , the exponential function from Section 7.5. In particular, E(1) equals the Eulers constant e.
Proof. Define f : R R by f (x) =
f (x) =
E(x)
ex .
Then f is differentiable and
E (x)ex E(x)(ex )
E(x)ex E(x)ex
=
= 0.
(ex )2
(ex )2
1
Thus f is a constant function, so that for all x R, f (x) = f (0) = E(0)
e0 = 1 = 1. Thus
E(x) = ex . By Theorem 7.5.7, this is the same as ln1 (x). In particular, E(1) = e1 = e.
305
The following is now immediate:

Theorem 9.7.3 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
}n of partial sums for E(1) = e1 = e is convergent, and
The sequence { k=0 k!
163
1957
685
starts with 1, 2, 2.5, 83

= 2.66667, 65
24 = 2.70833, 60 = 2.71667, 720 = 2.71806, 252 =
98641
2.71825, 19601
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 9.7.1, we will then understand E completely.
9.7.1 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.
9.7.2 Use the definition of ln in this section to prove that (ln) (x) = x1 .
9.7.3 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}.
P
9.7.4 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 9.5.2; it turns out that there
2
is no simpler, finite-term antiderivative of ex .)
2
9.7.5 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 227).
P
(1)k
9.7.6 Express
k=0 4k k! in terms of e.
P
2k
9.7.7 Express k=0 3k (k+1)!
in terms of e.
9.7.8 Find a Taylor polynomial expansion of the function E centered at 0. (Do as little
work as possible, but do explain your reasoning.)
9.7.9 Numerically evaluate e2 from the power series expansion to 5 significant digits. Prove
that you have achieved desired precision. (Hint: Theorem 6.5.5.)
9.7.10 Compute limxa E(x). (Do as little hard work as possible, but do explain your
reasoning.)
9.7.11 With proof, for which real numbers does the sequence {E(n)}n converge?
306
9.8
A special function, continued
In this section restrict E from the previous section to the imaginary axis. Throughout,
all x are real, and we thus analyze the function E(ix). Note that
E(ix) =
X
(ix)k
k!
k=0
ix (ix)2 (ix)3
(ix)4 (ix)5
(ix)6
(ix)7 (ix)8
(ix)9
+
+
+
+
+
+
+
+
+
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):
=1+
x4
x6
x8
x2
+
+
+ ,
2!
4!
6!
8!
x
x3
x5
x7
x9
SIN (x) = Im E(ix) =
+
+ .
1!
3!
5!
7!
9!
Since E(ix) converges, so do its real and imaginary parts, so that the domains of COS and
SIN are both all of R.
COS (x) = Re E(ix) = 1
Remarks 9.8.1
(1) E(ix) = COS (x) + i SIN (x).
(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.
307
(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).
Theorem 9.8.2 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)
3
2
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)
2
3
t4
t2 COS (t) is non-decreasing on [0, ), so that for all
06 0 + SIN (0) = 0. Thus 24
2
2
04
t4
t2 COS (t) 24
02 COS (0) = 1. We conclude that for all t 0,
t 0, 24
COS (t)
t4
t2
+ 1.
24
2
2
4
In particular, COS ( 3) 243 23 + 1 = 18 < 0. Since COS is continuous and
COS (0) = 1 > 0, by the Intermediate Value Theorem there exists s (0, 3) such that
COS (s) = 0. Since (COS (s))2 + (SIN (s))2 = 1, we have that SIN (s) = 1. We proved
3
3
that for all t 0, t6 t + SIN (t) 0, so that SIN (t) t t6 = 6t (6 t2 ). If t (0, 3],
then SIN (t) 6t (6 t2 ) > 0. Hence SIN (s) = 1.
Is this s unique? Since SIN is positive in (0, 3] and since SIN is the derivative of
COS , it follows that COS is strictly increasing and that COS is strictly decreasing on
(0, 3]. But then there cannot be another s in this interval with the same COS value.
Theorem 9.8.3 The trigonometric functions cos and sin are differentiable and continuous.
Furthermore, COS and SIN are the functions cos and sin, and s = /2.
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 9.8.2. The angle (in radians, counterclockwise from the positive
real axis) of E(is) = 0+i1 = i is /2. The proof of Theorem 9.8.2 showed for t [0, 1], the
308
imaginary part SIN (st) of E(ist) is positive, and so by the derivative properties, the real
part COS (st) of E(ist) is decreasing. Thus for t in [0, 1], COS (st) decreases as a function
of t from 1 to 0, and SIN (st) increases from 0 to 1. Thus E(ist) is in the first quadrant.
It follows that for all positive integers n, E(is/n), E(is2/n), E(is3/n), . . . , E(is(n 1)/n),
E(isn/n) = E(is) = i are points on the unit circle in the first quadrant. Since E(ism/n) =
(E(is/n))m, by Theorem 3.10.4, the angle of E(ism/n) is m times the angle of E(is), so
that as the angle of E(isn/n) = E(is) = i is /2, it follows that the angle of E(is/n) is
/(2n), and that the angle of E(ism/n) is (/2)(m/n). Similarly, the angle of E(is/n) is
the angle of 1/E(is/n), which is /(2n). Thus, again by Theorem 3.10.4, for all rational
numbers t, the angle of E(ist) is t/2. Since E is a continuous function, it follows that for
all real t, the angle of E(ist) is t/2.
Thus
COS (st) + i SIN (st) = E(ist) = cos

t + i sin
t ,
2
2
so that by uniqueness of real and imaginary parts, COS (st) = cos( 2 t) and SIN (st) =

2s
t
and
sin(t)
=
SIN
t
. Since COS and
sin( 2 t), or in other words, cos(t) = COS 2s
SIN are differentiable, these equalities say that cos and sin are differentiable as well and
that

2s
2s
2s
2s
2s
2s
t = sin(t), sin (t) = COS

t =
cos(t).
cos (t) = SIN
This proves that cos and sin are differentiable. By Exercise 1.3.23, for small positive real x,
0 < cos(x) < sinx x < cos1 x . Since cos is continuous and cos(0) = 1, by the Squeeze theorem
(Theorem 4.3.11), limx0+ sinx x = 1. Since sin(x) = x, it follows that limx0 sinx x = 1.
Hence
2s
sin x sin 0
2s
sin x
=
cos(0) = sin (0) = lim
= lim+
= 1,
x0
x
x
x0
whence s = /2.
Theorem 9.8.4 Every complex number x can be written in the form rE(i), where r = |x|
is the length and 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 a complex number of length |x|/r = 1 and by Theorem 3.10.3, x/r and x have the
same angle. Let be that angle. Then x/r = E(i), so that x = rE(i).
309
Notation 9.8.5 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.
9.8.1 Let and r be real numbers with r > 0. Prove that for any complex number x,
rei x is obtained from x by rotating x counterclockwise around the origin by angle
radians and by stretching the result by a factor of r. (Compare with Theorem 3.10.3.)
9.8.2 Prove that ei + 1 = 0. (This has been called one of the most beautiful equalities.)
9.8.3 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 1.3.20.)
9.8.4 Prove de Moivres formula: for all n Z,
(cos(x) + i sin(x))n = cos(nx) + i sin(nx).
9.8.5 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.
9.8.6 Express (3 + 4i)5 in the form a + bi for some a, b R.
9.8.7 Prove that for any integer m, the sequence {E(2imn)}n converges. Prove that the
sequence {E( 2in)}n does not converge. Determine all real numbers for which the
sequence {E(in)}n converges. Prove your conclusion.
9.8.8 For which real numbers and does the sequence {E(n(+i))}n converge? Prove
your conclusion.
9.8.9 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 ).)
*9.8.10 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
2x
=
1+
dx.
2
2 1 x2
0
(Hint: This is an improper integral; use lengths of curves; the Greeks defined to
be the perimeter of half the circle of radius 1.)
310

ii) Compute the improper integral via substitution u = COS (x) and further other
steps. Your answer should be s from Theorem 9.8.2.
9.9
Trigonometry
In this section we review the semesters concepts of limits, continuity, differentiability,

and integrability on the newly derived trigonometric functions from the previous section.
Definition 9.9.1 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)
1
1
(3) tan(x) = cos(x)
, cot(x) = cos(x)
sin(x) , sec(x) = cos(x) , csc(x) = sin(x) .
Remarks 9.9.2 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 + 2) = sin(x) and cos(x + 2) = cos(x). This is not obvious
from the power series definition of E(ix), but if follows from Theorem 9.8.4.
(3) For all a R, lim sin(x) = sin(a), lim cos(x) = cos(a).
xa
xa
(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 +eix
.
(6) cos(x) =
2
2
(7) (sin(x)) + (cos(x))2 = 1. (Recall from Remark 2.4.7 that for trigonometric functions 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, /2]. By the previous part, cos takes
on non-negative values on [/2, /2].
(14) sin is increasing on [/2, /2].
(15) sin, when restricted to [/2, /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, ], so that cos, when restricted to [0, ], has an inverse, called arccos. The domain of arccos is [1, 1].
Section 9.9: Trigonometry
311
(17) Verify the details in the following. The derivative of tan is always non-negative, so
that on (/2, /2), tan is invertible. Its inverse is called arctan, and the domain
of arctan is (, ).
Proposition 9.9.3 Refer to Theorem 6.2.7.
1
.
(1) For x (/2, /2), arcsin (x) = 1x
2
1
(2) For x (0, ), arccos (x) = 1x2 .

1
(3) arctan (x) = 1+x
2.
Proof.
arcsin (x) =
=
=
=
=
1
sin (arcsin(x))
1
cos(arcsin(x))
1
p
(because cos is non-negative on [/2, /2])
(cos(arcsin(x)))2
1
p
1 (sin(arcsin(x)))2
1
.
1 x2
The proof of the second part is similar.
1
tan (arctan(x))
1
=
(sec(arctan(x)))2
1
=
1 + (tan(arctan(x)))2
1
.
=
1 + x2
arctan (x) =

9.9.1 Prove that for all x R,
1 cos(2x)
1 + cos(2x)
, (cos(x))2 =
.
2
2
R
R
9.9.2 Two methods for computing (sin(x))2 dx (and similarly (cos(x))2 dx):
i) Use the previous exercise.
ii) Use integration by parts twice.
(sin(x))2 =
312
9.9.3 Let k, j Z. You may want to use integration by

show all of the following.
R
i) sin(kt) cos(jt) dt = 0.
ii)
(
Z
0,
sin(kt) sin(jt) dt = 0,
,
iii)
cos(kt) cos(jt) dt =
parts or the previous exercise to
if j 6= k,
if jk = 0,
otherwise.
0, if j 6= k or jk 6= k,
, otherwise.
9.9.4 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
x cos(kx) dx =
1
1
x sin(kx) + 2 cos(kx) + C.
k
k
9.9.5 Compute the derivatives of the standard trigonometric functions tan(x), cot(x),
sec(x), csc(x).
9.9.6 Use integration by substitution (Exercise 7.4.5) to compute antiderivatives of tan(x)
and cot(x).
9.9.7 Use integration by substitution (Exercise 7.4.5) and a rewriting trick to compute an
. Compute an antiderivative of csc(x).
antiderivative of sec(x) = sec(x)(sec(x)+tan(x))
sec(x)+tan(x)
9.9.8 (Finally we can prove properly Example 6.3.6.) Prove that for all positive x, x >
sin(x).
9.9.9 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 9.8.3,
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
Examples of LH
opitals rule
LH
opitals 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 a hypothesis in the assumptions is omitted. All work in this
section is in the exercises.
313

The next two exercises are taken from R. J. Bumcrots article Some subtleties in
LH
opitals rule in Century of Calculus. Part II. 19691991, 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 203204.
9.10.1 Let f (x) = x2 sin(x1 ) and g(x) = sin x.
(x)
= 0.
i) Prove that limx0 fg(x)
(x)
ii) Prove that limx0 fg (x)
does not exist.
iii) Why does this not contradict LHopitals rule Theorem 6.4.3?
9.10.2 Let f (x) = x sin(x1 ) and g(x) = sin x. Prove that neither limx0
(x)
limx0 fg (x)
f (x)
g(x)
nor
exist.
*9.10.3 (This is from R. P. Boass article Counterexamples to LHopitals rule American

Mathematical Monthly 93 (1986), 644645. You will need certain facts about derivatives
Rx
f (x)
.
and even integrals of trigonometric functions.) Define f (x) = 0 cos2 (t) dt, g(x) = 2+sin(x)
i) Prove that lim f (x) = = lim g(x).
x
f (x)
6= 0.
ii) Prove that lim
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)
6= 0.
v) Prove that lim
x g (x)
vi) Why does this not contradict Exercise 6.4.5?
*9.10.4 (This is from R. C. Bucks Advanced Calculus, McGraw-Hill Book Company,

Inc., 1956.) 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.
(x)
ii) Prove that limx fg(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) .
(x)
= 0.
iv) Prove that limx fg (x)
v) Why does LHopitals rule not apply here?
*9.10.5 (Modification of the previous exercise.) Let f (x) =
1
2
x
+ sin( x2 ), and g(x) =
(x)
( x2 + sin( x2 ))esin( x . Prove that limx0+ fg(x)
does not exist but that limx0+
Why does LHopitals rule not apply here?
f (x)
g (x)
= 0.
314
9.11
Further exotic uses of trigonometry
All further exotic uses are presented through exercises.

9.11.1 Fill in the explanations and any missing steps in the two double improper integrals.
Z
Z
Z
Z

Z
1
1
1
1
dx dy =
dx dy
1 + y 1 + x2 y
1+y
1 + x2 y
0
0
!
Z
Z N
1
1
=
lim
dx dy
1 + y N 0 1 + x2 y
0
!

Z
arctan( yx) N
1
lim
dy
=

1 + y N
y
0
x=0
Z

1
=
dy
2 0
y(1 + y)
Z
2u

du (by substitution y = u2 )
=
2 0 u(1 + u2 )
2
=
,
2

Z Z
1
1
x2
1
1
dy dx
dy dx =
1 + y 1 + x2 y
1 x2 1 + y 1 + x2 y
0
0

Z
1
1
=
dx
ln
2
1x
x2
0
Z
ln x
dx
=2
2
x 1
0
Z 1

Z
ln x
ln x
=2
dx +
dx
2
x2 1
0 x 1
1

Z 1
Z 1
ln u
ln x
dx +
du
=2
2
2
0 u 1
0 x 1
Z 1
ln x
dx. (Stop here.)
=4
2
0 x 1
2
*9.11.2 (This is taken from the article D. Ritelli, Another proof of (2) = 6 , American
Mathematical Monthly 120 (2013), 642645.)
i) We proved in Section 9.5 that derivatives and (definite) 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. Accept that
315
R1
P
P R 1
= 0 ( ln x) k=0 x2k dx = k=0 0 ( ln x)x2k dx.
ii) Also accept that the two integrals in Exercise 9.11.1 are the same (order of integration matters sometimes).
P
2
iii) Use Exercises 7.4.7 and 9.1.5 to prove that k=1 k12 = 6 .
R1
ln x
dx
0 1x2
*9.11.3 (See also Exercise 9.11.2.) (This is taken from the article I. Papadimitriou, A
P
simple proof of the formula k=1 k 2 = 2 /6. American Mathematical Monthly 80 (1973),
424425.) Let x be an angle measured in radians strictly between 0 and /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 9.9.8 to prove that x12 < 1 + (cot(x))2 .
Pm k 2
<
iii) Prove that for all integers k, m with 1 k m,
k=1 cot 2m+1

2
Pm
(2m+1)2 Pm
1
k
.
k=1 k2 < m +
k=1 cot 2m+1
2
(2m+1)2 Pm
m(2m1)
1
iv) Use Exercise 9.11.4 to prove that m(2m1)
<
.
2
k=1 k2 < m +
3
3
P 1
2
2
2
v) Prove that k=1 k2 = 6 . (Hint: multiply the previous part by /4m .)
*9.11.4 Use de Moivres formula (see Exercise 9.8.4) to prove that for all x R and all
positive integers n,
n
X
n k
n
n
n
i (cot(x))nk .
cos(nx) + i sin(nx) = (sin(x)) (cot(x) + i) = (sin(x))
k
k=0
If n = 2m + 1, this says that
cos((2m + 1)x) + i sin((2m + 1)x) = (sin(x))
2m+1
2m+1
X
k=0
The imaginary part of this is:

sin((2m + 1)x) = (sin(x))
2m+1
m
X

2m + 1 k
i (cot(x))2m+1k .
k

2m + 1
(cot(x))2m2k .
(1)
2k + 1
k
k=0

k 2m+1 mk
Let P (x) =
. Prove that P is a polynomial of degree m with
k=0 (1)
2k+1 x
leading coefficient 2m + 1. Prove that for k = 1, . . . , m, (cot(k/(2m + 1)))2 is a root of P .
Prove that P has m distinct roots. Prove that P (x) = (2m + 1) (x (cot(/(2m + 1)))2 )
(x (cot(2/(2m + 1)))2 ) (x (cot(m/(2m + 1)))2 ). By equating the coefficient of
Pm
.
xm1 in the two forms of P prove that k=1 (cot(k/(2m + 1)))2 = m(2m1)
3
Pm
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 reasoning 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 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.)
Sometimes division by zero creeps in in subtler ways. For example, to find solutions
2
to x = 3x, it is wrong to simply cancel xs 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 x is the solution x = 0.
320
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 ... .
does not equal 3.14159
If the answer to a problem is 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 17/59 = .21954437870195, but it is
fine to write 17/59

= .21954437870195. Usually it is not necessary to write out the
numerical approximation, but sometimes it is helpful to write one down to get a sense of
the size of the answer, and to check that 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.
321
To prove an (in)equality, manipulate one side in steps to get to the other

If you have to prove that
Pn
k=1
k2 =
k=1
n(n+1)(2n+1)
6
for n = 1, do NOT do the following:
1(1 + 1)(2 1 + 1)
6
DO NOT
DO THIS!
1
X
k2 =
123
6
1=1
12 =
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:
1
X
1(1 + 1)(2 1 + 1)
123
=
.
k 2 = 12 = 1 =
6
6
k=1
Do you see how this is shorter and proves succinctly the desired 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:
add 1 to both sides:

multiply both sides by 0:
?
1=0
?
2=1
?
0=0
But we certainly cannot conclude that the first line 1 = 0 is correct.

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.
x1
R
R
4 3x dx is terrible grammar; instead write (4 3x) dx.
x1
322
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 you may want to suppose that a 0, hopefully you get a contradiction,
and hence a must have been positive.)
Limit versus function value
Asking for limxa 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 > 0. x can be taken to be negative. Here one could read
or confuse the part 0. x as 0 x = 0, but then 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
If
If
If
A
A
A
A
implies
implies
implies
implies
B
B
B
B
and
and
and
and
if
if
if
if
A
B
A
B
is true, we may conclude B.

is false, we may conclude that A is false.
is false, we may not conclude anything.
is true, we may not conclude anything.
Truth table:
P
T
T
F
F
Q
T
F
T
F
P
F
F
T
T
P Q
T
F
F
F
P Q
T
T
T
F
P xor Q
F
T
T
F
P Q
T
F
T
T
P Q
T
F
F
T
Statement
How to prove it
P (via contradiction).
Suppose P . Establish some nonsense (that

makes P thus impossible).
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 .
Let x be arbitrary of the specified type. Prove

that property P holds for x.
(Possibly break up into a few subcases.)
324

Find/construct an x of the specified type.

Prove that property P holds for x.
Alternatively, invoke a theorem guaranteeing
that such x exists.
An element x of a specified type

with property P is unique.
Suppose that x and x are both of specified 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 ....
A property P is false/does not

hold for elements of a specified
type.
Give a counterexample: find/construct an element x of specified type and prove that P

does not hold for that x.
Statement
Negation
P and Q
(P Q) = (P ) (Q)
P or Q
(P Q) = (P ) (Q)
P Q
(P Q) = P (Q)
There exists x of the specified type such that

P is false for x.

For all x of the specified type, P is false for x.
Geometric series
P k
k=1 r diverges if |r| 1 and converges to
P k
k=0 r diverges if |r| 1 and converges to
Pn
n+1
For all r C \ {1}, k=0 r k = r r11 .
r
1r
1
1r
if |r| < 1.
if |r| < 1.
325
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 (x) f (a)
f (a + h) f (a)
= lim
.
xa
h0
h
xa
f (a) = lim
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 <
< xn1 < xn = b. Let f : [a, b] R be a bounded function. For each i = 1, . . . , n, let
mi = inf {f (x) : x [xi1 , xi ]}, Mi = sup {f (x) : x [xi1 , xi]}.
The lower sum of f with respect to P is
L(f, P ) =
n
X
mi (xi xi1 ).
n
X
Mi (xi xi1 ).
i=1
The upper sum of f with respect to P is

U (f, P ) =
i=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
326
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
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
abstract binary operation 71
Bd boundary of a set 136
C 48, 124

n
k , n choose k 44
Ac complement of A 51
\ complement of sets 51
composition of functions 65
cos 33
e (identity element for binary operation) 72
e: Eulers constant 235
{} empty set 48
empty set 48
there exists 24
for all 24
f n function composed with itself 65
sinn power of trigonometric function (not
composition) 65
> (strictly) greater than 80
greater than or equal to 80
e (identity element for binary operation) 72
id (identity function) 63
if and only if 17
implies 16
in a set 47
inf : infimum 80
Int interior of a set 136
intersection 50
< (strictly) smaller than 80, 81, 82, 97
> (strictly) greater than 97
greater than or equal to 97
less than or equal to 80, 97
lim of sequence 248
lim inf 276
lim sup 276
ln: natural logarithm 233
logical not 14
logical or 15
logical and 15
L(f, P ): lower sum 212
N 48
N0 91
84
n
Y
Qn
product 35, 74
k=1 ,
k=1
Q 48, 109
// end of proof 18
QED end of proof 18
end of proof 18
end of proof 18
R 48, 115
Range
range of a function 63
n
180
sin 33, 34
subset of a set 49
n
X
Pn
,
34, 74
k=1
k=1
sup : supremum 80, 275

union 50
U (f, P ): upper sum 212
Z 48, 104
00 35
A
absolute value 19, 83
absolutely convergent 288
advice on writing proofs 319
alternating test for series 285
and (logical) 15
antecedent 16
antiderivative 227
arccos 310
328
Index
arcsin 310
arctan 311
area 210
signed 210
associative 73
addition 77
multiplication 77
axiom 91
B
base (of exponential function) 234
base case (of induction) 36
base of exponentiation 180
Bernoullis inequality 201
bijective 65
binary operation 71
associative 73
commutative 75
binomial coefficient 45
binomial 45
Boas, R. P. 313
boundary (Bd ) 136
bounded set 80
above 80
and open balls 140
below 80
bounded
function 211
Buck 313
Bumcrot, R. J. 313
C
C 48, 124
absolute value 128
arithmetic 124
Cartesian coordinates 130
distance 128
length 128
norm 128
polar coordinates 130

properties 124
reverse triangle inequality 128
triangle inequality 128
cancellation 75, 102, 109, 112
Cartesian coordinates 130
Cartesian product 56
Cauchys criterion for series 283
Cauchys Mean Value Theorem 201
ceiling function 68
chain rule 194
chart
how to prove 28, 324
negation 31, 324

choose: n choose k, nk 44
closed and boundedcontained in open
sets 141, 142
closed set 138
closed-form antiderivative 227
closure (of a set) 138
codomain (of a function) 63
commutative 75
addition 77
multiplication 77
comparison
sequences with infinite limits 256
sequences 261, 265
series 283, 284
completeness of R, C 271
complex conjugate 126
complex number 124
with e 308
composition of functions 66
conditional statement: beware 17
conditional statement 16
antecedent, consequent 16
congruent modulo n 60
congruent modulo n 60
conjunction 15
Index
connuousti 173
consequent 16
construction of N0 91
construction of Q from Z 62, 109
construction of R 115
construction of Z from N0 61
contain 47
continuous 169
and inverse function 176
and monotonicity 176
connuousti 173
exponential function 182, 183
exponentiation 180
hence integrable 221
image of interval 175
power series 298
properties 170
suounitnoc 173
ticonnuous 173
uniformly continuous 184
convergent
absolutely 288
cos: trigonometric function 33, 307
COS 306
D
Darbouxs theorem 199
de Moivres formula 309, 315
Dedekind cut 114
standard 121
definition: beware 17
degree (of polynomial) 34, 66
derivative 187
and monotonicity 200
complex-valued functions 195
inverse 195
Leibnizs notation 187
Newtons notation 187
power rule for rational exponents 196
329
properties 192, 193, 194, 195, 196

differentiable 187
hence continuous 191
disjunction 15
distributivity 77
division 78
do not divide by zero 78
domain (of a function) 63
implicit 64
E
e vs. 237
element 47
empty product 36, 74
empty set 48
empty sum 35
Entegral 219
enumeration of Q+ 246
epsilon-delta definition 143, 150
epsilon-N definition 166
equivalence (logical) 17
equivalence class 59, 103
equivalence relation 58, 103
Euclidean algorithm 39
Euclid 31
Eulers constant (e) 235, 304
even function, its integral 225
exclusive or (xor) 15
existential quantifier 24
exponential function 234
continuity 183
without derivatives and integrals 182
exponentiation 180
by real exponents 234
continuity 180
exponent 180, 234
Extreme Value Theorem 176
330
Index
F
Ferguson, S. J. 179
Fibonacci numbers 42, 245
field 77, 111, 119, 124
axioms 77
fixed point theorem 178
floor function 68
function 63
bijective 65
ceiling 68
codomain 63
composition 65
connuousti 173
constant 64
continuous 169
decreasing 87
domain 63
exponential 182, 183, 234
floor 68
graph 63
identity 63
increasing 87
inverse 87
injective 65
one-to-one 65
onto 65
polynomial 34, 66
power 87
product of 88
radical 87, 180
range 63
rational 66
suounitnoc 173
surjective 65
ticonnuous 173
Fundamental theorem of arithmetic 30
Fundamental Theorem of Calculus 226,
228
complex-valued function 229, 230
G
generalized power function 234
generalized power series 293
geometric series 280
glb , inf : greatest lower bound 80
Goldbachs conjecture 13
Greatest lower bound theorem 123
greatest lower bound: glb, infimum, inf 80
H
harmonic series 281
higher order derivatives 205
how-to-prove chart 28, 324
I
identity element 72
additive 77
multiplicative 77
identity function 63
if and only if, iff 16
imaginary axis 126
imaginary part of a complex number 126
implication 16
improper integral 225, 226, 232
indefinite integral 227
index (of unions, intersections) 51
induction 36, 92
base case 36
inductive step 36
inductive set 91
infimum (inf, glb) 80
275
infinite sequence: see sequence
infinite series: see series
injective function 65
composition 66
integer 104
integrable 214, 220
integral test for series 285
Index
integral 214
addition 223
applications 238
complex-valued function 229
double 314
monotonicity 224
notation 214, 215, 227, 229
integration by parts 231
integration by substitution 231
interior 136
intermediate value property 174, 199
Intermediate Value Theorem 174
interval 47, 82
inverse 72, 87
additive 72, 77
derivative 195
multiplicative 72, 77
of a function 70
of a strictly monotone function 87
of product 74
unique 73
invertible (element) 73
invertible (function) 70
L
LH
opitals rule 202, 203, 205, 236, 237, 312
Lagrange interpolation 70
least element 81, 101
Least upper bound theorem 123
least upper bound: lub, supremum, sup 80
lemma 98
length of curve 238
limit inferior, liminf 276
limit point (of set) 136
limit superior, limsup 276
limit
absolute value 159
at infinity 166
composite function theorem 160
epsilon-delta definition 143, 150

epsilon-N definition 166
infinite 163
M-delta definition 163
M-N definition 166
of a function 143, 150, 163, 166
of sequence 248
polynomial rule 159
power rule 158
rational rule 159
real/imaginary part 159
right/left-handed 150, 163
squeeze theorem 160
subsequence 274
theorems for infinite limits 164
theorems 157, 158, 159, 160
uniqueness 156
ln, natural logarithm 233
inverse 304
logic circuits 23
logical biconditional 17
lower bound of set 80
lower integral 214
lower sum (integrals) 212
lub , sup : see least upper bound
M
Mean Value Theorem for integrals 232
Mean Value Theorem 200
member (of a set) 47
modulo n 60
modus ponens 19
monotone function 87
and derivative 200
monotone sequence 267, 269
multiplication table 73
331
332
Index
N
N 48
N0 91, 94
natural logarithm, ln 233
natural number (in N0 ) 91
negation 14, 31
chart 31, 324
negative 82, 108, 113, 122
Nelson, E. 173
non-negative 82, 108, 113, 122
non-positive 82, 108, 113, 122
notation
integral 227, 229
set vs. sequence 244
number of elements in Q+ 71
O
odd function, its integral 225
odd integer 22
one-to-one 65
onto 65
open ball 135
open set 135
or (logical) 15
ordered pair 56
ordered set 82
field 82
order 80
N0 97
Q 113
R 121
Z 107
P
p-test for series 286
Papadimitriou, I. 315
partial sum (of a sequence) 279
partition (of an interval) 212
refinement 213
Pascals triangle 44
vs. e 237
2
6 314
pigeonhole principle 69
polynomial function 34, 66, 67, 68, 159,
206, 260, 264
Taylor polynomial 206
polynomial
degree 34, 66
positive 82, 108, 113, 122
power function 87, 88, 158, 179, 260, 264
generalized 181, 234
power notation 35, 65, 74
00 35
for functions 65, 77
on sets with binary operation 74
special in trigonometry 65
power rule for derivatives 193
power rule
generalized 196
power seriesgeneralized 293
power series 288, 293
continuous 298
derivative 301
inverse 293
numerical evaluation 302
product of two 291
radius of convergence 290
ratio test 290
root test 289
Taylor series 294
uniqueness 303
zero 303
prime 13, 29, 30, 31
definition 17, 18
principle
do not divide by zero
of mathematical induction 36
pigeonhole 69
Index
Qn
product k=1 35, 74
product of functions 88
product rule for derivatives 192
product rule for higher derivatives 208
proof 18, 28
by contradiction 21
by induction 36
deductive reasoning 18
how-to-prove chart 28, 324
pigeonhole principle 69
2 is not rational 21
proper subset 50
Pythagoras 21
Q
Q 48, 109
Archimedean property 113
cancellation 112
field 111
ordered 113
qed (quod erat demonstrandum) 18
also QED 18
quadratic formula 67
quantifier
existential 24
universal 24
quotient rule for derivatives 192
quotient rule for higher derivatives 208
R
R 48, 115
Archimedean property 122
field 119
ordered 122
properties 119, 122
Raabes test 288
radical function 87, 180
radius of convergence (of power series) 290
range (of a function) 63
333
ratio test
for power series 290
for sequences 268
for series 284
rational number (in Q) 109
rational numbers between real numbers
123
rational number 21
real axis 126
real number 115
real part of a complex number 126
recursive 245
refinement
partition 213
relation on sets 57
equivalence 58
reflexive 58
symmetric 58
transitive 58
reverse triangle inequality 84, 122, 128
over C 128
Ritelli, D. 314
Rolles theorem 199
root of a number 180
root test for power series 289
root test for series 285
S
Sagher, Y. 71
sequence vs. set notation 244
sequence 244
arithmetic of sequences 247
bounded 267
Cauchy 270
comparison for infinite limits 256
comparison 261, 265
completeness of R, C 271
composite rule 260, 265
constant 244
334
Index
convergence 248, 258

properties 260, 262, 264, 265
divergence to 256
divergence 255
Fibonacci 245
finite 248
limit 248
unique 259, 262
monotone 267, 269
of partial sums 279
ratio test 268
squeeze theorem 261, 266
subsequence 272
term 244
series
alternating test 285
arithmetic with 280
comparison 283, 284
converges 280
diverges 280
geometric 280
harmonic 281
integral test 285
numerical evaluation 302
p-series test 286
power 288, 293
ratio test 284, 290
root test 285, 289
set vs. sequence notation 244
set 47
complement 51
disjoint 50
empty set 48
equality 50
intersection 50
member 47
notation 47
propositionally 48
subset 49
sum of 86
union 50
universal 51
signed area 210, 212
sin: trigonometric function 33, 34, 307
SIN 306
socks-and-shoes 77
spiral of Theodorus 134
statement 13
conditional statement 16
conjunction 15
disjunction 15
equivalence 17
if and only if 16
implication 16
logical biconditional 17
negation 14
xor 15
subsequence 272
subsequential limit 274
subset 49
subtraction 78
successor 91
sum of sets 86
Pn
summation k=1 34, 74
suounitnoc 173
supremum (sup, lub) 80
275
surface area of a surface of revolution 241
surjective function 65
composition 66
T
tautology 18
Taylor polynomial 206, 237
Taylor series 294
not approximating 298
Taylors remainder theorem 207, 208, 232
term of a sequence 244
Index
ticonnuous 173
topology 134, 137, 138
boundary 136
closed set 138
closure 138
interior 136
limit point 136
open ball 135
open set 135
tower of Hanoi 43
triangle inequality 20, 84
over C 128
reverse 84, 122, 128
trichotomy (of <) 81, 100, 107
tromino 42
truth table 14, 323
U
uniform continuity 184
universal quantifier 24
universal set 51
upper bound of set 80
upper integral 214
upper sum (integrals) 212
V
vacuously true 25
Venn diagram 52
volume of a surface of revolution 239, 240
W
well-ordered 81, 101
whole number 104
X
xor 15
Z
Z 48, 104
properties 105, 107
zero
do not divide by it 9, 78, 319
to zeroth power 35
335

Analysis

Uploaded by

Copyright:

Available Formats

Analysis

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Analysis

Uploaded by

Copyright:

Available Formats

Introdu

The briefest overview, motivation, notation

Chapter 1: How we will do mathematics

Section 1.1: Statements and proof methods

Section 1.2: Statements with quantifiers

Section 1.3: More proof methods, and negation

Section 1.4: Summation

Section 1.5: Proofs by (mathematical) induction

Section 1.6: Pascals triangle

Chapter 2: Concepts with which we will do mathematics

Section 2.1: Sets

Section 2.2: Cartesian product

Section 2.3: Relations, equivalence relations

Section 2.4: Functions

Section 2.5: Binary operations on sets

Section 2.6: Fields

Section 2.7: Order on sets, ordered fields, absolute values

Section 2.8: Increasing and decreasing functions

Chapter 3: Construction of the basic number systems

Section 3.1: Inductive sets and a construction of natural numbers

Section 3.2: Arithmetic on N0

Section 3.3: Order on N0

Section 3.4: Cancellation in N0

Section 3.5: Construction of Z, arithmetic, and order on Z

Section 3.6: Construction of the field Q of rational numbers

Section 3.7: Construction of the field R of real numbers

Section 3.8: Order on R, the least upper/greatest lower bound theorem

Section 3.9: Complex numbers

Section 3.11: Topology on the constructed fields

Section 3.12: Closed and bounded sets, and open balls

Chapter 4: Limits of functions

Section 4.1: Limit of a function

Section 4.2: When something is not a limit

Section 4.3: Limit theorems

Section 4.4: Infinite limits (for real-valued functions)

Section 4.5: Limits at infinity

Section 5.1: Continuous functions

Section 5.2: Intermediate Value Theorem, Extreme Value Theorem

Section 5.3: Radical functions

Section 5.4: Uniform continuity

Section 6.1: Definition of derivatives

Section 6.2: Basic properties of derivatives

Section 6.3: The Mean Value Theorem

Section 6.4: LHopitals rule

Section 6.5: Higher-order derivatives, Taylor polynomials

Section 7.1: Approximating areas

Section 7.2: Computing integrals from upper and lower sums

Section 7.3: What functions are integrable?

Section 7.4: The Fundamental Theorem of Calculus

Section 7.5: Natural logarithm and the exponential functions

Section 7.6: Applications of integration

Section 8.1: Introduction to sequences

Section 8.2: Convergence of infinite sequences

Section 8.3: Divergence of infinite sequences and infinite limits

Section 8.4: Convergence theorems via functions

Section 8.5: Convergence theorems via epsilon-N proofs

Section 8.6: Bounded sequences, monotone sequences, ratio test

Section 8.7: Cauchy sequences, completeness of R, C

Section 8.9: Liminf, limsup for real-valued sequences