Introduction To Analysis

Introdu
tion to Analysis
Irena Swanson
Reed College
Spring 2016
Table of contents
Preface 7
The briefest overview, motivation, notation 9
Chapter 1: How we will do mathematics 13
Section 1.1: Statements and proof methods 13
Section 1.2: Statements with quantifiers 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 45
Chapter 2: Concepts with which we will do mathematics 49
Section 2.1: Sets 49
Section 2.2: Cartesian product 57
Section 2.3: Relations, equivalence relations 59
Section 2.4: Functions 64
Section 2.5: Binary operations on sets 73
Section 2.6: Fields 79
Section 2.7: Order on sets, ordered fields 82
Section 2.8: Increasing and decreasing functions 89
Section 2.9: The Least upper bound property of R 91
Section 2.10: Absolute values 93
Chapter 3: The field of complex numbers, and topology 96
Section 3.1: Complex numbers 96
Section 3.2: Absolute value in C and polar coordinates 100
Section 3.3: Topology on the fields of real and complex numbers 107
Section 3.4: Closed and bounded sets, and open balls 113
Chapter 4: Limits of functions 116
Section 4.1: Limit of a function 116
Section 4.2: When something is not a limit 127
4
Section 4.3: Limit theorems 130

Section 4.4: Infinite limits (for real-valued functions) 137
Section 4.5: Limits at infinity 141
Chapter 5: Continuity 144
Section 5.1: Continuous functions 144
Section 5.2: Topology and the Extreme value theorem 149
Section 5.3: Intermediate Value Theorem 152
Section 5.4: Radical functions 155
Section 5.5: Uniform continuity 160
Chapter 6: Differentiation 163
Section 6.1: Definition of derivatives 163
Section 6.2: Basic properties of derivatives 167
Section 6.3: The Mean Value Theorem 175
Section 6.4: LHopitals rule 178
Section 6.5: Higher-order derivatives, Taylor polynomials 182
Chapter 7: Integration 187
Section 7.1: Approximating areas 187
Section 7.2: Computing integrals from upper and lower sums 196
Section 7.3: What functions are integrable? 199
Section 7.4: The Fundamental Theorem of Calculus 204
Section 7.5: Natural logarithm and the exponential functions 212
Section 7.6: Applications of integration 217
Chapter 8: Sequences 224
Section 8.1: Introduction to sequences 224
Section 8.2: Convergence of infinite sequences 228
Section 8.3: Divergence of infinite sequences and infinite limits 236
Section 8.4: Convergence theorems via epsilon-N proofs 240
Section 8.5: Convergence theorems via functions 245
Section 8.6: Bounded sequences, monotone sequences, ratio test 249
Section 8.7: Cauchy sequences, completeness of R, C 252
Section 8.8: Subsequences 255
Section 8.9: Liminf, limsup for real-valued sequences 259
Chapter 9: Infinite series and power series 264
Section 9.1: Infinite series 264
Section 9.2: Convergence and divergence theorems for series 268
5
Section 9.3: Power series 274

Section 9.4: Differentiation of power series 279
Section 9.5: Numerical evaluations of some series 282
Section 9.6: Specialized uses of power series 283
Section 9.7: Taylor series 287
Section 9.8: A special function 291
Section 9.9: A special function, continued 293
Section 9.10: Trigonometry 298
Section 9.11: Examples of LHopitals rule 301
Section 9.12: Further exotic uses of trigonometry 302
Appendix A: Advice on writing mathematics 309
Appendix B: What you should never forget 313
Preface
These notes were written expressly for Mathematics 112 at Reed College, with first us-
age 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., En-
glewood Cliffs, NJ, 1986, 4th edition).
In Math 112, students learn to write proofs while at the same time learning about bi-
nary operations, orders, fields, ordered fields, complete fields, complex numbers, sequences,
and series. We also review limits, continuity, differentiation, and integration. My aim for
these notes is to constitute a self-contained book that covers the standard topics of a
course in introductory analysis, that handles complex-valued functions, sequences, and se-
ries, 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, Carte-
sian products, relations, functions, binary operations, fields, ordered fields, Archimedean
property for the set of real numbers. In particular, we assume that the set of familiar real
numbers forms an ordered field with the Least upper bound property. In Chapter 3 we
construct the very useful field of complex numbers, and introduce the study of topology
that is indispensable for the rigorous treatment of limits. I cover topology more lightly
than what is in the notes. Subsequent chapters cover standard material for introduction
to analysis: limits, continuity, differentiation, integration, sequences, series, ending with
P k
the development of the power series k=0 xk! , the exponential and the trigonometric func-
tions. Since students have seen limits, continuity, differentiation and integration before, I
go through chapters 4 through 7 quickly. I slow down for sequences and series (chapters 8
and 9).
An effort is made throughout to use only what had been proved. In particular, trigono-
metric functions appear properly only in the last chapter, with occasional appearances in
the exercises earlier (with careful listing of the needed assumptions for these functions). 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 currently maintain two versions of these notes, one in which the natural, rational and real numbers are
constructed and the Least upper bound theorem is proved for the ordered field of real numbers, and one version
in which the Least upper bound property is assumed for the ordered field of real numbers. See my Math 112
webpage www.reed.edu/~iswanson/112.html for links to the two versions.
8 Preface
I acknowledge and thank the support from the Dean of Faculty of Reed College to
fund exercise and proofreading support in the summer of 2012 for Maddie Brandt, Munyo
Frey-Edwards, and Kelsey Houston-Edwards. I further thank Mark Angeles, Josie Baker,
Zachary Campbell, Laura Dallago, Andrew Erlanger, Palak Jain, Ya Jiang, Albyn Jones,
Mason Kennedy, Christopher Keane, Michael Keppler, Oleks Lushchyk, Molly Maguire,
Benjamin Morrison, Samuel Olson, Kyle Ormsby, Angelica Osorno, Shannon Pearson,
David Perkinson, Jeremy Rachels, Marika Swanberg, Simon Swanson, Matyas Szabo, Emer-
son Webb, Qiaoyu Yang, and Dean Young for their feedback. If you have further comments
or corrections, please send them to [email protected].
The briefest overview, motivation, notation
What are the meanings of the following:

5+6
79
84
4/5

2
48
1/3 = 0.333 . . . 1 = 3 1/3 = 0.999 . . .
(a b)2 = a2 b2
(a + b) (c + d) = ac + ad + bc + bd
(a + b) (a b) = a2 b2
(a + b)2 = a2 + 2ab + b2

a b = ab (for which a, b?)
p
What is going on: 4 9 = (4)(9) = 36 = 6,

4 9 = 2i 3i = 6
You know all of the above except possibly the complex numbers in the last two rows,
where obviously something went wrong. We will not resolve this last issue until later in
the semester, but the point for now is that we do need to reason carefully.
The main goal of this class is to learn to reason carefully, rigorously. Since one cannot
reason in a vacuum, we will (but of course) be learning a lot of mathematics as well:
sets, logic, various number systems, fields, the field of real numbers, the field of complex
numbers, sequences, series, some calculus, and that eix = cos x + i sin x.
We will make it all rigorous, i.e., we will be doing proofs. A proof is a sequence of
steps that logically follow from previously accepted knowledge.
But no matter what you do, never divide by 0. For further wise advice, turn to
Appendix A.
[Notational convention: Text between square brackets in this font 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
Chapter 1: How we will do mathematics
1.1 Statements and proof methods

Definition 1.1.1 A statement is a reasonably grammatical and unambiguous sentence
that can be declared either true or false.
Why do we specify reasonably grammatical? We do not disqualify a statement just

because of poor grammar, nevertheless, we strive to use correct grammar and to express the
meaning clearly. And what do we mean by true or false? For our purposes, a statement
is false if there is at least one counterexample to it, and a statement is true if it has been
proved so, or if we assume it to be true.
Examples and non-examples 1.1.2

(i) The sum of 1 and 2 equals 3. (This is a statement; it is always true.)
(ii) Seventeen. (This is not a statement.)
(iii) Seventeen is the seventh prime number. (This is a true statement.)
(iv) Is x positive? (This is not a statement.)
(v) 1 = 2.* (This is a false statement.)
(vi) For every real number > 0 there exists a real number > 0 such that for all x,
if 0 < |x a| < then x is in the domain of f and |f (x) L| < . (This is a
statement, and it is (a part of) the definition of the limit of a (special) function f
at a being L. Out of context, this statement is neither true or false, but we can
prove it or assume it for various functions f .) )
(vii) Every even number greater than 4 can be written as a sum of two odd primes.
(This statement is known as Goldbachs conjecture. No counterexample is
known, and no proof has been devised, so it is currently not known if it is true or
false.)
These examples show that not all statements have a definitive truth value. What makes
them statements is that after possibly arbitrarily assigning them truth values, different
consequences follow. For example, if we assume that (vi) above is true, then the graph
* This statement can also be written in plain English as One equals two. In mathematics it is acceptable
to use symbolic notation to some extent, but keep in mind that too many symbols can make a sentence hard to
read. In general we avoid starting sentences with a symbol. In particular, do not make the following sentence.
= is a verb. Instead make a sentence such as the following one. Note that = is a verb.
14 Chapter 1: How we will do mathematics
of f near a is close to the graph of the constant function L. If instead we assume that
(vi) above is false, then the graph of f near a has infinitely many values at some vertical
distance away from L no matter how much we zoom in at a. With this in mind, even I
am good is a statement: if I am good, then I get a cookie, but if I am not good, then you
get the cookie. On the other hand, if Hello were to be true or false, I would not be able
to make any further deductions about the world or my next action, so that Hello is not
a statement, but only a sentence.
A useful tool for manipulating statements is a truth table: it is a table in which the
first few columns may set up a situation, and the subsequent columns record truth values of
statements applying in those particular situations. Here are two examples of truth tables,
where T of course stands for true and F for false:
f (x) f is a constant function. f is continuous. f is differentiable everywhere.

x2 F T T
|x| F T F
7 T T T
x y xy > 0 xy 0 xy < 0
x>0 y >0 T F F
x>0 y 0 F T F
x<0 y >0 F T T
x<0 y 0 F F F
Note that in the second row of the last table, in the exceptional case y = 0, the
statement xy < 0 is false, but in the majority of the cases in that row xy < 0 is true.
The one counterexample is enough to declare xy < 0 not true, i.e., false.
Statements can be manipulated just like numbers and variables can be manipulated,
and rather than adding or multiplying statements, we connect them (by compounding the
sentences in grammatical ways) with connectors such as not, and, or, and so on.
Statement connecting:
(1) Negation of a statement P is a statement whose truth values are exactly opposite
from the truth values of P (under any specific circumstance). The negation of P
is denoted 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.
Now go back to the last truth table. Note that in the last line, the truth
values of xy > 0 and xy 0 are both false. But one should think that
xy > 0 and xy 0 are negations of each other! So what is going on,
why are the two truth values not opposites of each other? The problem is of
course that the circumstances x < 0 and y 0 are not specific enough. The
statement xy > 0 is under these circumstances false precisely when y = 0,
but then xy 0 is true. Similarly, the statement xy 0 is under the
given circumstances false precisely when y < 0, but then xy > 0 is true.
Thus, once we make the conditions specific enough, then the truth values of
xy > 0 and xy 0 are opposite, so that the two statements are indeed
negations of each other.
(2) Conjunction of statements P and Q is a statement that is true precisely when
both P and Q are true, and it is false otherwise. It is denoted P and Q or
P Q. We can record this in a truth table as follows:
P Q P Q
T T T
T F F
F T F
F F F
(3) Disjunction of statements P and Q is a statement that is false precisely when

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

plies Q, or variants thereof, such as If P then Q, Given P , Q follows, Q

whenever P . A symbolic abbreviation of implication is P Q. An implication
is true when a true conclusion follows a true assumption, or whenever the assump-
tion is false. In other words, P Q is false exactly when P is true and Q is false.
In an implication P Q, P is called the antecedent and Q the consequent.
The truth table for implication is as follows:
P Q P Q
T T T
T F F
F T T
F F T
It may be counterintuitive that a false antecedent always makes the implication
true. Bertrand Russell once lectured on this and claimed that if 1 = 2 then he
(Bertrand Russell) was the pope. An audience member challenged him to prove
it. So Russell reasoned somewhat like this: If I am the pope, then the consequent
is true. If the consequent is false, then I am not the pope. But if I am not the
pope, then the pope and I are two different people. By assumption 1 = 2, so we
two people are one, so I am the pope. Thus no matter what, I am the pope.
Furthermore, if 1 = 2, then Bertrand Russell is similarly also not the pope.
Namely, if he is not the pope, the consequent is true, but if he is the pope, then
the pope and he are one, and since one equals two, then the pope and he are two
people, so Russell cannot be the pope.
A further discussion about why false antecedent makes the implication true is in
the next discussion (5).
We list here most possible rephrasings of P implies Q:
P implies Q.
If P then Q.
P is a sufficient condition for Q.
P only if Q.
Q if P .
Q provided P .
Q given P .
Q whenever P .
Q is a necessary condition for P .
Unfortunately, the implication statement is not used consistently in informal spo-

ken language. For example, your grandmother may say: You may have ice cream
if you eat your broccoli when she means You may have ice cream only if you eat
your broccoli. Be nice to your grandmother and eat that broccoli even if she does
not express herself precisely because you know precisely what she means. But
in mathematics you do have to express yourself precisely! (Well, read the next
paragraph.)
Even in mathematics some shortcuts in precise expressions are acceptable. Here
is an example. The statements An object x has property P if somethingorother
holds and An object x has property P if and only if somethingorother holds (see
(5) below for if and only if) in general have different truth values and the proof
of the second is longer. However, the definition of what it means for an object
to have property P in terms of somethingorother is usually phrased with if, but
if and only if is meant. For example, the following is standard: Definition:
A positive integer strictly bigger than 1 is prime if whenever it can be written as
a product of two positive integers, one of the two factors must be 1. The given
definition, if read logically precisely, since it said nothing about numbers such as
4 = 2 2, would allow us to call 4 prime. However, it is an understood shortcut
that only the numbers with the stated property are called prime.
(5) Equivalence or the logical biconditional of P and Q stands for the compound
statement (P Q) (Q P ). It is abbreviated P Q or P iff Q, and is
true precisely when P and Q have the same truth values.
For example, for real numbers x and y, the statement x y+1 is equivalent
to x 1 y. Another example: 2x = 4x2 is equivalent to x = 2x2 ,
but it is not equivalent to 1 = 2x. (Say why!)
We now backtrack on the truth values of P Q. We can certainly fill in some
parts without qualms, leaving some unknown truth values x and y:
P Q P Q QP P Q
T T T T T
T F F x F
F T x F F
F F y y T
Since the last column above is the conjunction of the previous two, the last line
forces the value of y to be T . If x equals F , then the truth values of P Q are the
same as the truth values of P Q, which would say that the statements P Q
and P Q are logically the same. But this cannot be: If r > 0 then r 0 is
true whereas If r 0 then r > 0 is false. So this may convince you that the
truth values for the third and the fifth column have to be distinct, and this is only
possible if x is T .
Here is the truth table for all the connectives so far:
P Q P P Q P Q P xor Q P Q P Q
T T F T T F T T
T F F F T T F F
F T T F T T T F
F F T F F F T T
One can form more elaborate truth tables if we start not with two statements P
and Q but with three or more. Examples of logically compounding P, Q, and R
are: P 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, An odd integer is of the form 1 plus 2
times an integer, or, A positive integer is prime if whenever it can be written
as a product of two positive integers, one of the two factors is 1.
Another type of proof consists of filling in a truth table. For example, P (P ) is
always true, no matter what the truth value of P is, and this can be easily verified
with the truth table:
P P P P
T F T
F 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,
and it is proved below with the truth table:
P Q P Q (P Q) P ((P Q) P ) Q
T T T T T
T F F F T
F T T F T
F F T F T
This particular tautology is called modus ponens, and its most famous example
is the following:
Every man is mortal. (If X is a man/human, then X is mortal.)
Socrates is human.
Therefore, Socrates is mortal.
Here is a more mathematical example of modus ponens:
Every differentiable function is continuous.
f is differentiable.
Therefore, f is continuous.
Another tautology is modus tollens: ((P Q) (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
positive, and the assumption y < x < y implies that y < y, which also says
that y must be positive. So, with either assumption, we can draw the following
part of the real number line:
y 0 y
Now, by drawing, the real numbers x with |x| < y are precisely those real num-
bers x with y < x < y. A fancier way of saying this is that |x| < y if and only if
y < x < y.
Similarly, for all real numbers x and y, |x| y if and only if y x y. (Here,
the word similarly is a clue that I am invoking the proof-similarity principle,
and a reader who wants to practice proofs or is not convinced should at this point
work through a proof by mimicking the steps in the previous one.)
Some (or actually most) proofs invoke previous results without re-doing the previ-
ous work. In this way we prove the triangle inequality, which asserts that for all
real numbers x and y, |x y| |x| + |y|. (By the way, we will use the triangle in-
equality intensely, so understand it well.) Proof: Note that always |x| x |x|,
|y| y |y|. Since the sum of smaller numbers is always less than or equal to
the sum of larger numbers, we then get that |x| |y| x y |x| + |y|. But
|x| |y| = (|x| + |y|), so that (|x| + |y|) x y |x| + |y|. But then by the
previous result, |x y| |x| + |y|.
Most proofs require a combination of methods. Here we prove that whenever x is

a real number with |x 5| < 4, then |x3 3x| < 1030. Proof: The following is
standard formatting that you should adopt: first write down the left side of the
desired inequality (|x3 3x|), then start manipulating it algebraically, in interme-
diate steps add a clever 0 here and there, multiply by a clever 1 here and there,
rewrite, simplify, make it less than or equal to something else, and so on, every
step should be either obvious or justified on the right, until at the end you get the
quantity on the right (1030):
|x3 3x| |x3 | + |3x| (by the triangle inequality)
= |x|3 + 3|x|
= |x 5 + 5|3 + 3|x 5 + 5| (by adding a clever 0)
(|x 5| + 5)3 + 3(|x 5| + 5) (by the triangle inequality
and since a b implies that a3 b3 )
(4 + 5)3 + 3(4 + 5) (since by assumption |x 5| < 4)
= 93 + 3 9
< 103 + 3 10
= 1030.
Here is another pictorial proof, this one establishing the basis of trigonometry and
A
the definition of slope as rise over run: with the picture below, B = ab .
B
b
a
| {z }
A
Proof: The areas of the big and small triangles are 12 AB and 21 ab, and the area of
the difference is base times the average height, i.e., it is (A a) b+B
2 . Thus
1 1 b+B
AB = ab + (A a) .
2 2 2
By multiplying through by 2 we get that AB = ab + (A a)(b + B) = ab + Ab +
AB ab aB, so that, after cancellations, Ab = aB. Then, after dividing through
A
by bB we get that B = ab .
Another common method of proof of a statement P is proof by contradiction:

you assume that 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 1.1.15
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.
Exercises for Section 1.1

1.1.1 Determine whether the following statements are true or false, and justify:
i) 3 is odd or 5 is even.
ii) If n is even, then 3n is prime.
iii) If 3n is even, then n is prime.
iv) If n is prime, then 3n is odd.
v) If 3n is prime, then n is odd.
vi) (P 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 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.
1.1.9 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.
1.1.10 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?
1.1.11 Suppose that x is any real number such that |x+2| < 3. Prove that |x3 3x| < 200.
1.1.12 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.
1.1.13 (The quadratic formula) Let a, b, c be real numbers with a non-zero. Prove
(with algebra) that the solutions x of the quadratic equation ax2 + bx + c = 0 equal

b b2 4ac
x= .
2a
1.1.14 (Logic circuits) Logic circuits are simple circuits which take as inputs logical
values of true and false (or 1 and 0) and give a single output. Logic circuits are composed
of logic gates. Each logic gate stands for a logical connective you are familiar with it could
be and, or, or not (more complex logic circuits incorporate more). The shapes for logical
and, or, not are as follows:
Given inputs, each of these logic gates outputs values equal to the values in the associated
truth table. For instance, an and gate only outputs on if both of the wires leading
into it are on. From these three logic gates we can build many others. For example, the
following circuit is equivalent to xor.
input
output
input
A circuit that computes xor. The output on the right is on when

exactly one of the inputs on the left is on.
Make logic circuits that complete the following tasks. (It may be helpful to make logic
tables for each one.)
i) xor in a different way than the circuit above.
ii) P implies Q implication.
iii) xor for three inputs.
iv) Is a 3-digit binary number greater than 2?
v) Is a 4-digit binary string a palindrome?
1.1.15 (Odd-even integers)

i) Prove that the sum of two odd integers is an even integer.
ii) Prove that the product of two integers is odd if and only if the two integers are
both odd.
iii) Suppose that the product of two integers is odd. Prove that the sum of those two
integers is even.
iv) Suppose that the sum of the squares of two integers is odd. Prove that one of the
two integers is even and the other is odd.
v) Prove that the product of two consecutive integers is even. Prove that the product
of three consecutive integers is an integer multiple of 6.
vi) Prove that the sum of two consecutive integers is odd. Prove that the sum of three
consecutive integers is an integer multiple of 3.

1.1.16 Prove that 3 is not a rational number.
1.2 Statements with quantifiers
The number x equals 1 is true for some x and false for some x. Thus x = 1 is
not true or false universally. For determining a 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 P (x)
x with a certain specification, P (x) holds x P (x)
Section 1.2: Statements with quantifiers 25
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
[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 abbre-
viations. 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:
a possible situation x P (x) x P (x) xP (x) xP (x)
there are no x of specified type T vacuously T vacuously F F
there are x of specified type, T F F T

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

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

true for some, false for some
Notice that among the columns with truth values, one and three have opposite values,
and two and four have opposite values. This proves the following:
Proposition 1.2.1 The negation of x P (x) is x 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 prime number. Every
whole number divisible by 3 is divisible by 2 is false because 3 is divisible by 3 and is not
divisible by 2.
Remark 1.2.2 The statement For all whole numbers x between 1/3 and 2/3, x2 is
irrational is true vacuously. Another reason why For all whole numbers x between 1/3
and 2/3, x2 is irrational is true is that its negation, There exists a whole number x
between 1/3 and 2/3 for which x2 is rational, is false because there is no whole number
between 1/3 and 2/3: since the negation is false, we get yet more motivation to declare the
original statement true.

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.
iii) Everybody loves Raymond.
iv) Spring break is always in March.
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) There exists an integer m such that for all integers n, m + n is even.

xi) For every integer n, n2 n is even.
xii) Every list of 5 consecutive integers has one element that is a multiple of 5.
xiii) The element in the previous part is a multiple of 3.
xiv) 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 Let xLy represent the statement that x loves y. Rewrite the following statements
by using quantifiers: Everybody loves somebody, Somebody loves everybody, Some-
body 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 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.)
1.3 More proof methods, and negation
When statements are compound, they can be harder to prove. Fortunately, proofs can
be broken down into simpler statements. Here is an essential chart of this breaking down.
Statement How to prove it

P (via contradiction). Suppose P . Establish some nonsense (that
makes P thus impossible so that P must
hold).
P and Q. Prove P . Prove Q.
P or Q. Suppose that P is false. Then prove Q.

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

Contrapositively: Suppose that Q is false.
Prove that P is false.
P Q. Prove P Q. Prove Q P .
For all x of a specified type, prop- Let x be arbitrary of the specified type. Prove
erty P holds for x. that property P holds for x.
(Possibly break up into a few subcases.)
There exists x of a specified type Find/construct an x of the specified type.

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

with property P is unique. type and satisfy property P . Prove that
x = x .
Alternatively, show that x is the only solution
to an equation, or the only element on a list,
or ....
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) = 6, by the Intermediate Value Theorem (Theorem 5.3.1 in these
notes) [invoking a theorem rather than constructing x, as opposed to in the
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.
Example 1.3.7 (Mixture of methods) For every real number x strictly between 0 and 1
there exists a positive real number y such that x1 + y1 = xy
1
.
Proof. [We have to prove that for all x as specified some property holds.]
Let x be in (0, 1). [For this x we have to find y ...] Set y = 1 x. [Was this
a lucky find? No matter how we got inspired to determine this y, we now
verify that the stated properties hold for x and y.] Since x is strictly smaller
than 1, it follows that y is positive. Thus also xy is positive, and y + x = 1. After dividing
the last equation by the positive number xy we get that x1 + y1 = xy 1
.
Furthermore, y in the previous example is unique: it has no choice but to be 1 x.
Example 1.3.8 (The Fundamental theorem of arithmetic) Any positive integer

n > 1 can be written as pa1 1 pa2 2 pakk for some positive prime integers p1 < < pk and
some positive integers a1 , . . . , ak . (Another standard part of the Fundamental theorem of
arithmetic is that the pi and the ai are unique, but we do not prove that. Once you are
comfortable with proofs you can prove that part yourself.)
Proof. (You may want to skip this proof for now.) Suppose for contradiction that the
conclusion fails for some positive integer n. Then on the list 2, 3, 4, . . . , n let m be the
smallest integer for which the conclusion fails. If m is a prime, take k = 1 and p1 = m,
a1 = 1, and so the conclusion does not fail. Thus m cannot be a prime number, and so m =
m1 m2 for some positive integers m1 , m2 strictly bigger than 1. Necessarily 2 m1 , m2 < m.
By the choice of m, the conclusion is true for m1 and m2 . Write m1 = pa1 1 pa2 2 pakk and
m2 = q1b1 q2b2 qlbl for some positive prime integers p1 < < pk , q1 < < qk and some
positive integers a1 , . . . , ak , b1 , . . . , bl . Thus m = pa1 1 pa2 2 pakk q1b1 q2b2 qlbl is a product
of positive prime numbers, and after sorting and merging the pi and qj , the conclusion
follows also for m. But we assumed that the conclusion fails for m, which yields the
desired contradiction. Hence the conclusion does not fail for any positive integer.
Example 1.3.9 Any positive rational number can be written in the form ab , where a and b
are positive whole numbers, and in any prime factorizations of a and b as in the previous
example, the prime factors for a are distinct from the prime factors for b.
Proof. [We have to prove that for all ...] Let x be a(n arbitrary) positive rational
number. [We rewrite the meaning of the assumption next in a more concrete
and usable form.] Thus x = ab for some whole numbers a, b. If a is negative, since x
is positive necessarily b has to be negative. But then a, b are positive numbers, and
x = a
b . Thus by possibly replacing a, b with a, b we may assume that a, b are positive.
[A rewriting trick.] There may be many different pairs of a, b, and we choose a pair for
which a is the smallest of all possibilities. [A choosing trick. But does the smallest
a exist?] Such a does exist because in a collection of given positive integers there is always
a smallest one. Suppose that a and b have a (positive) prime factor p in common. Write
a = a0 p and b = b0 p for some positive whole numbers a0 , b0 . Then x = ab00 , and since
0 < a0 < a, this contradicts the choice of the pair a, b. Thus a and b could not have had a
prime factor in common.
In order to be able to prove statements effectively, we often have to suppose the

negation of a part, say for proving statements with or and for proofs by contradiction.
Work and think through the following negations:
Statement Negation
P P
P and Q (P Q) = (P ) (Q)
P or Q (P Q) = (P ) (Q)
P Q (P Q) = P (Q)
P Q (P Q) = P (Q) = (P ) Q
For all x of a specified type, prop- There exists x of the specified type such that
erty P holds for x. P is false for x.
There exists x of a specified type For all x of the specified type, P is false for x.
such that property P holds for x.
Warning: The negation of a conditional statement is not another conditional statement!

Practice this one!
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 all: p1 , p2 , . . . , pn . Let a = (p1 p2 pn ) + 1. Since we know
that 2, 3, 5 are primes, necessarily n 3 and so a > 1. By the Fundamental theorem of
arithmetic (Example 1.3.8), a has a prime factor p. Since p1 , p2 , . . . , pn are all the primes,
necessarily p = pi for some i. But then p = pi divides a and p1 p2 pn , whence it divides
1 = a (p1 p2 pn ), which is a contradiction. So it is not the case that there are only
finitely many prime numbers, so there must be infinitely many.

1.3.1 Prove that every whole number is either odd or even.
1.3.2 Prove that the successor of any odd integer is even.
1.3.3 Prove that if n is an even integer, then either n is a multiple of 4 or n/2 is odd.

1.3.4 Prove that if 0 < x < 1, then x2 < x < x.

1.3.5 Prove that if 1 < x, then x < x < x2 .
5
1.3.6 Prove that there exists a real number x such that x2 + 12 x = 16 .
1.3.7 Prove that the following pairs of statements are negations of each other:
i) P Q.
(P 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 func-
tions 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.
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 9.9.3.) Draw a unit circle and a line segment from the center to the
circle. Any real number x uniquely determines a point P on the circle at angle x radians
from the line. Draw the line from that point that is perpendicular to the first line. The
length of this perpendicular line is called sin(x), and the distance from the intersection of
the two perpendicular lines to the center of the circle is called cos(x). This is our definition
of cos and sin.
i) Consider the following picture inside the circle of radius 1:
y x
a) Label the line segments of lengths sin(x), sin(y), cos(y).

b) Use ratio geometry (from page 21) to assert that the smallest vertical line
in the bottom triangle has length sin(x) cos(y).
c) Use trigonometry and ratio geometry to assert that the vertical line in the
top triangle has length sin(y) cos(x).
d) Prove that sin(x + y) = sin(x) cos(y) + sin(y) cos(x).
1.3.21 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 21 xr 2 .
r
x
1.3.23 (Invoked in Theorem 9.9.4.) Let x be a small positive real number. Consider the
following picture with inside the drawn circle of radius 1:
i) Assert that the area of the small triangle is strictly smaller than the area of the
wedge, which in turn is strictly smaller than the area of the big triangle.
ii) Using the previous part and ratio geometry (from page 21), prove that
1 1 1
2 sin(x) cos(x) < 2 x < 2 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 the sum of the first hundred numbers
in full length with 100 numbers and 99 plus signs: it would be too long, it would not
be any clearer, we would probably start doubting the intelligence of the writer, it would
waste paper and ink... The clear and short way of writing such a long sum is with the
summation sign :
100
X 100
X
k or n.
k=1 n=1
The counters k and n above are dummy variables, they vary from 1 to 100. We could use any
other name in place of k or n. In general, if f is a function defined at m, m+1, m+2, . . . , n,
we use the summation sign for shortening as follows:
n
X
f (k) = f (m) + f (m + 1) + f (m + 2) + + f (n).
k=m
This is one example where a good notation saves space (and it even sometimes clarifies
the concept). Typographically, when summation is displayed, the two bounds (m and n)
Section 1.4: Summation 35
appear below and above the summation sign, but when the summation is in-line, the two
Pn
bounds appear to the right of the sign, like so: k=m f (k). (This prevents lines jamming
into each other.)
Now is a good time to discuss polynomials. A polynomial function is a function of
the form f (x) = a0 + a1 x + + an xn for some non-negative integer n and some numbers
a0 , a1 , . . . , an . The largest d for which ad is non-zero is called the degree of the polynomial.
It is convenient to write this polynomial with the shorthand notation
n
X
n
f (x) = a0 + a1 x + + an x = ak xk .
k=0
Here, of course, x0 stands for 1. When we evaluate f at 0, we get a0 = a0 +a1 0+ +an 0n

P
= nk=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 00 equal to 0 or 1 or to something else entirely? Well, it turns
out that 00 is not equal to that zero limit you surely know of other functions f for which
lim f (x) exists but the limit is not equal to f (c). (Consult also Exercise 7.5.7.)
xc
Examples 1.4.2
5
X
(1) 2 = 10.
k=1
X5
(2) k = 15.
k=1
X4
(3) k 2 = 12 + 22 + 32 + 42 = 30.
k=1
X12
(4) cos(k) = cos(10) + cos(11) + cos(12) = 1 1 + 1 = 1.
k=10
X2
(5) (4k 3 ) = 4 + 0 + 4 + 4 8 = 32.
k=1
Xn
(6) 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 2
X 4
X 1
X 4
X 0
X 4
X
ak = ak + ak = ak + ak = ak + ak ,
k=1 k=1 k=3 k=1 k=2 k=1 k=1
or explicitly written out:
a1 + a2 + a3 + a4 = (a1 + a2 ) + (a3 + a4 ) = (a1 ) + (a2 + a3 + a4 ) = () + (a1 + a2 + a3 + a4 ),

from which we deduce that this empty sum must be 0. Similarly, every empty sum equals 0.
Similarly we can shorten products with the product sign :

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

P4 P4
1.4.1 Compute k=0 (2k + 1), k=0 (k 2 + 2).
Pn Pn
1.4.2 Determine all non-negative integers n for which k=0 k= k=0 n.
1.4.3 Prove that

n
X n
X
i) c f (k) = cf (k).
k=m k=m
n
X n
X n
X
ii) f (k) + g(k) = (f (k) + g(k)).
k=m k=m k=m
Xn Xn m1
X
iii) f (k) = f (k) f (k).
k=m k=1 k=1
Pn Pn Pn
1.4.4 Show that ( k=m f (k)) ( k=m g(k)) = k=m f (k) g(k) is false (in general).
P0
1.4.5 Prove that k=1 k = 0(1)
2 .
1.4.6 Prove that the empty product equals 1. In particular, it follows that we can declare
0! = 1. This turns out to be very helpful notationally.
1.4.7 Prove:
5
Y
i) 2 = 32.
k=1
Y5
ii) k = 120.
k=1
1.5 Proofs by (mathematical) induction
So far we have learned a few proof methods. There is another type of proofs that
deserves special mention, and this is proof by (mathematical) induction, sometimes
referred to as the principle of mathematical induction. This method can be used when
one wants to prove that a property P holds for all integers n greater than or equal to an
integer n0 . Typically, n0 is either 0 or 1, but it can be any integer, even a negative one.
Induction is a two-step procedure:

(1) Base case: Prove that P holds for n0 .
(2) Inductive step: Let n > n0 . Assume that P holds for all integers n0 , n0 + 1, n0 +
2, . . . , n 1. Prove that P holds for n.
Why does induction succeed in proving that P holds for all n n0 ? By the base case
we know that P holds for n0 . The inductive step then proves that P also holds for n0 + 1.
So then we know that the property holds for n0 and n0 + 1, whence the inductive step
implies that it also holds for n0 + 2. So then the property holds for n0 , n0 + 1 and n0 + 2,
whence the inductive step implies that it also holds for n0 + 3. This establishes that the
property holds for n0 , n0 + 1, n0 + 2, and n0 + 3, so that by inductive step it also holds for
n0 + 4. We keep going. For any integer n > n0 , in n n0 step we similarly establish that
the inductive step holds for n0 , n0 + 1, n0 + 2, . . . , n0 + (n n0 ) = n. Thus for any integer
n n0 , we eventually prove that P holds for it.
The same method can be phrased with a slightly different two-step process, with the
same result, and the same name:
1. Base case: Prove that P holds for n0 .
2. Inductive step: Let n > n0 . Assume that P holds for integer n 1. Prove that P
holds for n.
Similar reasoning as in the previous case also shows that this induction principle
succeeds in proving that P holds for all n n0 .
Pn n(n+1)
Example 1.5.1 Prove the equality k=1 k= 2 for all n 1.
P1
Proof. Base case n = 1: The left side of the equation is k=1 k which equals 1. The right
side is 1(1+1)
2 which also equals 1. This verifies the base case.
Inductive step: Let n > 1 and we assume that the equality holds for n 1. [We
want to prove the equality for n. We start with the expression on the
left side of the desired and not-yet-proved equation for n (the messier of
the two) and manipulate the expression until it resembles the desired right
side.] Then
n 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
We can even prove the equality k=1 k = n(n+1) 2 for all n 0. Since we have already
P0
proved this equality for all n 1, it remains to prove it for n = 0. The left side k=1 k is
an empty sum and hence 0, and the right side is 0(0+1)2 , which is also 0.
Pn n(n+1)(n+2)(n+3)(n+4)
Example 1.5.2 Prove the equality k=1 k(k + 1)(k + 2)(k + 3) = 5
for all n 1.
Proof. Base case n = 1:

1
X
k(k + 1)(k + 2)(k + 3) = 1(1 + 1)(1 + 2)(1 + 3) = 1 2 3 4
k=1
12345 1(1 + 1)(1 + 2)(1 + 3)(1 + 4)
= = ,
5 5
which verifies the base case.
Inductive step: Let n > 1 and we assume that the equality holds for n 1. [We
want to prove the equality for n. We start with the expression on the
left side of the desired and not-yet-proved equation for n (the messier of
the two) and manipulate the expression until it resembles the desired right
side.] Then
n 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
Section 6.1.) Prove that for all n 1, dx (xn ) = nxn1 . [ Or should induction start
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
Inductive step: Suppose that equality holds for [0, ]1, 2, . . . , n 1. Then
d n d
(x ) = (x xn1 )
dx dx
d d
= (x) xn1 + x (xn1 ) (by the product rule of differentiation)
dx dx
= 1 xn1 + (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
r(x) = f (x). If n = m, then set q(x) = abnn and (necessarily) r(x) = f (x) abnn g(x) =
a0 + a1 x + + an xn abnn (b0 + b1 x + + bn xn ) = (a0 abnn b0 ) + (a1 abnn b1 )x + (a2
an 2 an n1
bn b2 )x + + +(an1 bn bn1 )x . These are the base cases.
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) such that h(x) = q1 (x) g(x) + r1 (x) and such
that the degree of r1 (x) is strictly smaller than m. Then xh(x) = xq1 (x)g(x)+xr1(x), and
by induction there exist polynomials q2 (x) and r2 (x) such that xr1 (x) = q2 (x)g(x) + r2 (x)
and such that the degree of r2 (x) is strictly smaller than m. Now set q(x) = xq1 (x) + q2 (x)
and r(x) = r2 (x) + a0 . Then the degree of r(x) is strictly smaller than m, and
q(x)g(x) + r(x) = xq1 (x)g(x) + q2 (x)g(x) + r2 (x) + a0

= xq1 (x)g(x) + xr1 (x) + a0
= xh(x) + a0
= f (x).

Example 1.5.6 For all positive integers n, n n < 2.

Proof. Base case: n = 1, so n n = 1 < 2.

Inductive step: Suppose that n is an integer with n 2 and that n1 n 1 < 2. This
means that n 1 < 2n1 . Hence n < 2n1 + 1 < 2n1 + 2n1 = 2 2n1 = 2n , so that
n
n < 2.
Remark 1.5.7 There are two other equivalent formulations of mathematical induction for
proving a property P for all integers n n0 :
Mathematical induction, version III:
(1) Base case: Prove that P holds for n0 .
(2) Inductive step: Let n n0 . Assume that P holds for all integers n0 , n0 + 1, n0 +
2, . . . , n. Prove that P holds for n + 1.
Mathematical induction, version IV:
1. Base case: Prove that P holds for n0 .
2. Inductive step: Let n n0 . Assume that P holds for integer n. Prove that P holds
for n + 1.
Convince yourself that these two versions of the workings of mathematical induction
differ from the original two versions only in notation.
Exercises for Section 1.5: Prove the following properties for n 1 by induction.
Xn
n(n + 1)(2n + 1)
1.5.1 k2 = .
6
k=1
Xn 2
3 n(n + 1)
1.5.2 k = .
2
k=1
1.5.3 The sum of the first n odd integers is n2 .
Xn
1.5.4 (2k 1) = n2 .
k=1
1.5.5 (Triangle inequality) For all real numbers a1 , . . . , an , |a1 + a2 + + an | |a1 | +
|a2 | + + |an |.
X n
1.5.6 (3k 2 k) = n2 (n + 1).
k=1
1.5.7 1 2 + 2 3 + 3 4 + + n(n + 1) = 31 n(n + 1)(n + 2).
1.5.8 7n + 2 is a multiple of 3.
1.5.9 3n1 < (n + 1)!.
1 1 1 1
1.5.10 + + + + n.
1 2 3 n
1 1 1 1 1
1.5.11 2 + 2 + 2 + + 2 2 .
1 2 3 n n
1.5.12 Let a1 = 2, and for n 2, an = 3an1 . Formulate and prove a theorem giving an
in terms of n (no dependence on other ai ).
1.5.13 8 divides 5n + 2 3n1 + 1.
1.5.14 1(1!) + 2(2!) + 3(3!) + + n(n!) = (n + 1)! 1.
1.5.15 2n1 n!.

Yn
1 1
1.5.16 1 = .
k n
k=2
Yn
1 n+1
1.5.17 1 2 = .
k 2n
k=2
Xn
1.5.18 2k (k + 1) = 2n+1 n + 1.
k=0
P2n 1 n+2
1.5.19 k=1 k 2
.
P2n 1 1
Pn 1
1.5.20 (This is invoked in Example 9.1.8.) k=1 k2 k=0 2k .
1.5.21 (This is invoked in the proof of Theorem 9.4.1.) For all numbers x, y, xn y n =
Pn1
(x y)(xn1 + xn2 y + xn3 y 2 + + y n1 = (x y) k=0 xn1k y k .
1.5.22 (This is invoked in the Ratio tests Theorems 8.6.5 and 9.2.4.) Let r be a positive
real number. Suppose that for all positive integers n n0 , an+1 < ran . Prove that for all
positive integers n n0 , an+1 < r nn0 an0 . (The strict inequality < can also be replaced
by , >, throughout.)
1.5.23 Let An = 12 + 22 + 32 + + (2n 1)2 and Bn = 12 + 32 + 52 + + (2n 1)2 .
Find formulas for An and Bn , and prove them (by using algebra and previous problems,
and possibly not with induction).
1.5.24 (From the American Mathematical Monthly 123 (2016), page 87, by K. Gaitanas)
Pn1 k 1
Prove that for every n 2, k=1 (k+1)! = 1 n! .
1 1 1 1
1.5.25 Let An = 12 + 23 + 34 ++ n(n+1) . Find a formula for An and prove it.
1.5.26 There are exactly n people at a gathering. Everybody shakes everybody elses
hands exactly once. How many handshakes are there?
1.5.27 Pick a vertex V in a triangle. Draw n distinct lines from V to the opposite edge
of the triangle. If n = 1, you get the original triangle and two smaller triangles, for a total
of three triangles. Determine the number of distinct triangles obtained in this way with
arbitrary n.
1.5.28 Find, with proof, an integer n0 such that n2 < 2n for all integers n n0 .
1.5.29 Find, with proof, an integer n0 such that 2n < n! for all integers n n0 .
1.5.30 (Via the grapevine, based on ideas of Art Benjamin, Harvey Mudd College, and
Dan Velleman, Amherst College) A tromino is a plane figure composed of three squares in
L-shape:
Prove that for every positive integer n, any 2n 2n square grid with exactly one of the
squares removed can be tiled with trominoes.
1.5.31 (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, . . .. (Many parts below are taken from the
book Fibonacci Numbers by N. N. Vorobev, published by Blaisdell Publishing Company,
1961, translated from the Russian by Halina Moss.)
i) Fibonacci numbers are sometimes motivated as follows. You get the rare gift of
a pair of newborn Fibonacci rabbits. Fibonacci rabbits are the type of rabbits who
never die and each month starting in their second month produce another pair of
rabbits. At the beginning of months one and two you have exactly that 1 pair of
rabbits. In the second month, that pair gives you another pair of rabbits, so at
the beginning of the third month you have 2 pairs of rabbits. In the third month,
the original pair produces another pair of rabbits, so that at the beginning of the
fourth month, you have 3 pairs of rabbits. Justify why the number of rabbits at
the beginning of the nth month is sn . n
n
1 1+ 5
ii) Prove that for all n 1, sn = 5
2
5 12 5 . (It may seem amazing
1
that these expressions with square roots of 5 always yield positive integers.) Note
that the base case requires proving this for n = 1 and n = 2, and that the inductive
step uses knowing the property for the previous two integers.
iii) Prove that s1 + s3 + s5 + + s2n1 = s2n .
iv) Prove that s2 + s4 + s6 + + s2n = s2n+1 1.
v) Prove that s1 + s2 + s3 + + sn = sn+2 1.
vi) Prove that s1 s2 + s3 s4 + + s2n1 s2n = 1 s2n1 .
vii) Prove that s1 s2 + s3 s4 + + s2n1 s2n + s2n+1 = s2n + 1.
viii) Prove that s1 s2 + s3 s4 + + (1)n+1 sn = (1)n+1 sn1 + 1.
ix) Prove that for all n 3, sn > ( 1+2 5 )n2 .
x) Prove that for all n 1, s21 + s22 + + s2n = sn sn+1 .
xi) Prove that sn+1 sn1 s2n = (1)n .
xii) Prove that s1 s2 + s2 s3 + + s2n1 s2n = s22n .
xiii) Prove that s1 s2 + s2 s3 + + s2n s2n+1 = s22n+1 1.
xiv) Prove that ns1 + (n 1)s2 + (n 2)s3 + + 2sn1 + sn = sn+4 (n + 3).
xv) Prove that for all n 1 and all k 2, sn+k = sk sn+1 + sk1 sn .
xvi) Prove that for all n, k 1, skn is a multiple of sn . (Use the previous part.)
xvii) Prove that s2n+1 = s2n+1 + s2n .
xviii) Prove that s2n = s2n+1 s2n1 .
xix) Prove that s3n = s3n+1 + s3n s3n1 .
n
xx) Prove that sn+1 = 1+2 5 sn + 12 5 .
s3n+2 1
xxi) Prove that s3 + s6 + s9 + + s3n = 2 . (Use the previous part.)
s3n+2 +(1)n+1 6sn1 +5
xxii) Prove that s31 + s32 + s33 + + s3n = 10 .
xxiii) Prove that

(1 + 5)n 1

sn < .
2n 5 2
(xxiv)* If you know a bit of number theory, prove that for all positive integers m, n, the
greatest common divisor of sm and sn is sgcd(m,n) .
(xxv)* Prove that sn is even if and only if n is a multiple of 3. Prove that sn is divisible
by 3 if and only if n is a multiple of 4. Prove that sn is divisible by 4 if and only
if n is a multiple of 6. Prove that sn is divisible by 5 if and only if n is a multiple
of 5. Prove that sn is divisible by 7 if and only if n is a multiple of 8. Prove that
there are no Fibonacci numbers that have the remainder of 4 when divided by 8.
Prove that there are no odd Fibonacci numbersthat are n divisible
by 17.
1 1 sn+1 sn
xxvi) If you know matrices, prove that for all n 2, = .
1 0 sn sn1
1.5.32 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
1x . Compare with the proof by induction in
Example 1.5.4.
1.5.33 (Spiral of Theodorus) Draw a triangle with vertices at (0, 0), (1, 0), (1, 1). The

hypotenuse has length 2.
i) One of the vertices of the hypotenuse is at (0, 0). At the other vertex of the
hypotenuse, draw an edge of length 1 at a right angle away from the first triangle.
Make a triangle from the old hypotenuse and this new edge. What is the length
of the hypotenuse of the new triangle?
ii) Repeat the previous step twice.
iii) Argue by induction that one can draw the square root of every positive integer.
1.5.34 (Tower of Hanoi) There are 3 pegs on a board. On one peg, there are n disks,
stacked from largest to smallest. The task is to move all of the disks from one peg to a
different peg, given the following constraints: you may only move one disk at a time, and
you may only place a smaller peg on a larger one (never a larger one on a smaller one).
Let Sn be the least number of moves to complete the task for n disks.
i) If n = 1, then what is the least number of moves it takes to complete the task?
What if there are 2 disks? Repeat for 3, 4, 5 disks.
ii) Make a recursive formula (defining Sn based on Sn1 ) for this Sn . Then, make a
guess for a non-recursive formula for Sn (defining Sn based on n without invoking
Sn1 ). Prove your guess using induction and the recursive formula that you wrote.
1.5.35 What is wrong with the following proof by induction? I will prove that 5n + 1
is a multiple of 4. Assume that this is true for n 1. Then we can write 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 besides the fact that the
conclusion is false for many n? I will prove that all horses are of the same color. This is the
same as saying that for any integer n 1 and any set of n horses, all the horses belonging
to the set have the same color. If n = 1, of course this only horse is the same color as
itself, so the base case is proved. Now let n > 1. If we remove one horse from this set, the
remaining n 1 horses in the set are all of the same color by the induction assumption.
Now bring that one horse back into the set and remove another horse. Then again all of
these horses are of the same color, so the horse that was removed first is the same color as
all the rest of them.
1.6 Pascals triangle
Pascals triangle is very useful, so read this section with the exercises.
The following is rows 0 through 8 of Pascals triangle, and the pattern is obvious
for continuation into further rows:
row 0: . . . . . . . . . 1
row 1: . . . . . . . . 1 1
row 2: . . . . . . . 1 2 1
row 3: . . . . . . 1 3 3 1
row 4: . . . . . 1 4 6 4 1
row 5: . . . . 1 5 10 10 5 1
row 6: . . . 1 6 15 20 15 6 1
row 7: . . 1 7 21 35 35 21 7 1
row 8: . 1 8 28 56 70 56 28 8 1
Note that the leftmost and rightmost numbers in each row are all 1, and each of the other
numbers is the sum of the two numbers nearest to it in the row above it. We number the
slanted columns from left to right starting from 0: the 0th slanted column consists of all 1s,
the 1st slanted column consists of consecutive numbers 1, 2, 3, 4, . . ., the 2nd slanted column
consists of consecutive numbers 1, 3, 6, 10, . . ., and so on for the subsequent columns.

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

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

What would it take to compute 100 5
? It seems like we would need to write down rows
0 through 100 of 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
k+1
. In other words, prove that
k
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+1
.
1.6.3 Prove that every integer n 0 has the property that for all k = 0, 1, 2, . . . , n,
n
n!
k = k!(nk)! .

1.6.4 Prove that nk = n(n1)(n2)(nk+1)
k! .
4
5 6 7 8 100 100 100 100
1.6.5 Compute 2 , 2 , 2 , 2 , 2 , 2 , 3 , 4 , 5 .

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

members. For this reason nk is read n choose k.
1.6.7 (This is invoked in Theorem 2.8.2, Example 8.2.8, Theorem 6.2.3.) Prove that for
all non-negative integers n,
Xn
n n k nk
(a + b) = a b .
k
k=0
(Since a + b contains two summands, it is called a binomial, and the expansion of (a + b)n

is called the binomial expansion, with coefficients ni being called by yet another name
in this context: binomial coefficients.)

1.6.8 Express each of the following as a + b 2 for some integers a, b:

i) 2 1, ( 2 1)2 , ( 2 1)3 , ( 2 1)4 , ( 2 1)5 .

ii) Write each of the five expressions in the previous part in the form c d for
some positive integers c, d.
(iii)* Do you see a relation between c and d for each expression in the previous part?
Is there a general rule? Can you prove it? (You may want to use the previous two
exercises.)
Pn Pn
1.6.9 Prove that for all positive integers n, k=0 (1)k nk = 0. Compute k=0 (1)k nk
in case n = 0.
1.6.10 Prove that for any non-negative integer k,
n
X n(n + 1)(n + 2) (n + k + 1)
j(j + 1)(j + 2) (j + k) = .
j=1
k+1
(Hint: induction on n or instead use Exercises 1.6.2 and 1.6.4.)

1.6.11 Use Exercise 1.6.10 to get succinct simplifying formulas for
n
X n
X n
X n
X
j, j(j + 1), j(j + 1)(j + 2), j(j + 1)(j + 2)(j + 3).
j=1 j=1 j=1 j=1
i) Note that j 2 = j(j + 1) j. Use the simplifications from above to prove that
Pn 2 n(n+1)(2n+1)
j=1 j = 6 .
ii) From j 3 = 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 .
k
1.6.12 Prove that for all non-negative integers n and all k = 0, 1, . . . n, nk nk! .
1.6.13 Fix a positive integer k. Prove that there exists a positive number C such that for

all sufficiently large integers n, Cnk nk .

1.6.14 Give reasons why we should have nk = 0 for n < k or if either k or n is nega-
tive. (It is convenient to have the enlargement of the Pascals triangle where we allow all
combinations of rows and columns, even negatively indexed ones.)
1.6.15 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

k=0 k = d+n+1
n .
*1.6.16 Prove that for all non-negative integers n,

Xn !2 n1
X !2
2n k 2n
2 1=2 2k
2k 2k + 1
k=0 k=0
and !2
n
X X
n1 !2
2n + 1 k 2n + 1 k
2 +1=2 2 .
2k 2k + 1
k=0 k=0
Pen n k Pon n
k
For notations sake you may want to label En = k=0 2k 2 and O n = 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.)
Chapter 2: Concepts with which we will do mathematics
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. (From the context you
should be able to distinguish between the interval (0, 1) and a point (0, 1) in the
plane.)
(0, 1] is the interval from 0 to 1 that does not include 0 but includes 1.
[0, 1) is the interval from 0 to 1 that does not include 1 but includes 0.
[0, 1] is the interval from 0 to 1 that includes 0, 1.
(5, ) is the interval of all real numbers strictly bigger than 5.
[5, ) is the interval of all real numbers bigger than or equal to 5.
(, 5) is the interval of all real numbers strictly smaller than 5.
(, 5] is the interval of all real numbers smaller than or equal to 5.
(, ) is the set of all real numbers.
[a, a] is the interval consisting of precisely one number a.
(2) Descriptively we can say for example that a set A is the set of all prime numbers.
The description may be more loaded with symbols: A is the set of all real numbers x
50 Chapter 2: Concepts with which we will do mathematics
with the property that x2 > 4x 3. (Use the quadratic formula to prove that A
consists of the numbers that are strictly smaller than 1 or strictly larger than 3.)
(3) We can list the elements and surround them with curly braces to define a set:
(i) {1, 2, 3} is the set consisting of precisely 1, 2 and 3.
(i) {1, 2, 3, 2} is the set consisting of precisely 1, 2 and 3. Thus {1, 2, 3, 2} =
{1, 2, 3}.
(iii) {blue, hello, 5} is the set consisting precisely of the color blue, of the word
hello, and of number 5.
(iv) {1, 2, {1, 2}} is the set consisting of precisely of numbers 1 and 2 and of the
set {1, 2}. This set has exactly three distinct elements, it is not the same as {1, 2},
and it is not the same as {{1, 2}}.
(4) The set with no elements is called the empty set and is denoted or {}.
The set {} is not empty because it contains the empty set.
(5) When the list of elements is not small enough for reasonable explicit listing but
the pattern of elements is clear, we can start the list and then add , . . . when
the pattern is clear:
(i) {1, 2, 3, . . . , 10000} is the set of all positive integers that are at most 10000.
(ii) {1, 4, 9, . . . , 169} is the set of the first 13 squares of integers.
(iii) {0, 1, 2, 3, . . .} is the set consisting of all non-negative whole numbers. This
set is often denoted by N. WARNING: in some books, N stands for the set of
all positive whole numbers. To distinguish between the two, we will write N0 for
{0, 1, 2, 3, . . .} and N+ for {1, 2, 3, . . .}.
(6) Warning: {3, 5, 7, . . .} or {3, 5, 7, . . . , 101} could stand for the set of all odd primes
(up to 101), or possibly for the set of all odd whole numbers strictly greater
than 1 (up to 101). Avoid ambiguities: write more elements, or write an explicit
description of the elements instead.
(7) The set of all whole numbers is written Z, the set of all rational numbers is writ-
ten Q, the set of all real numbers is written R, and the set of all complex numbers is
written C (complex numbers are defined in Section 3.1, and until then do not worry
when C appears in the text). (I rely on your basic training to understand the
examples N, N0 , N+ , Z, Q, R, C.)
(8) We can also define sets propositionally: the set of all x (or x A) for which
property P holds can be we written as
{x : P (x)} or {x A : P (x)}.
Here are some explicit examples:
(i) {x R : x2 = x}, and this happens to be the set {0, 1}.
(ii) {x R : x > 0 and x < 1}, and this happens to be the interval (0, 1).
(iiii) Q = { ab : a, b Z, b 6= 0}.
(iv) {n N : n > 1 and if n = pq for some integers p, q then |p| = 1 or |q| = 1} is
the same as the set of all positive prime numbers.
(v) A = {x : x is a positive integer that equals the sum of its proper factors}.
Who are the elements of A? They have to be positive integers, and it is easy to
verify that 1, 2, 3, 4, 5 are not in A. But 6 has factors 1, 2, 3, 6, and the sum of
the factors 1, 2, 3 other than 6 equals 6. Thus 6 is an element of A. Verify that
28, 496, 8128 are also in A (and if you have a lot of time, program a computer to
verify that no other number smaller than 33 million is in A).
(9) Proving that a property P holds for all integers n n0 is the same as saying that
the set A = {n Z : P holds for n} contains {n0 , n0 + 1, n0 + 2, . . .}. By the
principle of mathematical induction, P holds for all integers n n0 is the same as
saying that n0 A and that n 1 A implies that n A.
Summary of example sets, and their notation

= {}: the set with no elements.
{a, b, c}, {a, b, . . . , z}.
{x : x can be written as a sum of three consecutive integers}.
N: the set of all Natural numbers. Depending on the book, this could be the set of all
positive integers or it could be the set of all non-negative integers. The symbols below are
unambiguous:
N0 : the set of all non-negative integers;
N+ : the set of all positive integers.
Z: the set of all integers (Zahlen in German).
Q: the set of all rational numbers (Quotients).
R: the set of all Real numbers.
C: the set of all Complex numbers (more about them starting in Section 3.1).
Just like numbers, functions, and logical statements, sets and their elements can also
be related and combined in meaningful ways. The list below introduces quite a few new
concepts that may be overwhelming at first, but in a few weeks you will be very comfortable
with them.
Subsets: A set A is a subset of a set B if every element of A is an element of B. In that

case we write A B. For example, N+ N0 Z Q R. The non-subset relation
is expressed with the symbol 6: R 6 N.
Every set is a subset of itself, i.e., for every set A, A A.
The empty set is a subset of every set, i.e., for every set A, A.
If A is a subset of B and A is not equal to B (so B contains at least one element that
is not in A), then we say that A is a proper subset of B, and we write A ( B. For
example, N+ ( N0 ( Z ( Q ( R.
Equality: Two sets are equal if they consist of exactly the same elements. In other words,
A = B if and only if A B and B A.
Intersection: The intersection of sets A and B is the set of all objects that are in A
and in B:
A B = {x : x A and x B}.
When A B = , we say that A and B are disjoint.
Union: The union of sets A and B is the set of all objects that are either in A or in B:
A B = {x : x A or x B}.
Intersections and unions of arbitrary families of sets: We have seen intersections

and unions of two sets at a time. We can also take intersections and unions of three,
four, five, and even infinitely many sets at a time. Verify the equalities below:
(A B) C = A (B C),
(A B) C = A (B C),
(A B) (C D) = A (B C D), etc
(A B) (C D) = (A (B C)) D, etc
(Verification of the first equality above: Let x (A B) C. This holds if and only
if x A B and x C, which holds if and only if x A, x B and x C, which in
turn holds if and only if x A and x (B C), i.e., if and only if x A (B C).)
Thus having established that parentheses above are irrelevant, we simply write the
four sets above as A B C, A B C, A B C D, A B C D, respectively.
More generally, given sets A1 , A2 , . . . , An , we write
n
\
Ak = A1 A2 An = {a : a Ak for all k = 1, . . . , n},
k=1
[n
Ak = A1 A2 An = {a : there exists k = 1, . . . , n such that a Ak },
k=1
or even more generally, when Ak are sets as k varies over a possibly infinite index set I,
then
\
Ak = {a : a Ak for all k I},
kI
[
Ak = {a : there exists k I such that a Ak }.
kI
When I is the empty index set, one can argue as for empty sums and empty products
in Section 1.4 that
\
Ak = the universal set that contains all the Ak ,
k
[
Ak = .
k
We return to this theme in Section 2.5. Note that the k in the subscripts are referred
to as indices of unions or intersections.
Complement: The complement of A in B is
B \ A = {b B : b 6 A}.
Some authors write B A, but that has another meaning as well: B A : {b a : b

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

a A: a is an element of a set A.
A B: A is a subset of set B; every element of A is an element of B.
A ( B: A B and A 6= B; A is a proper subset of B.
A = B: A B and B A.
A B: the set of all elements that are in A and in B.
A and B are disjoint: A B = .
A B: the set of all elements that are either in A or in B.
A \ B: the set of all elements of A that are not in B.
Ac : the set of all elements in the universal set that are not in A.
Example 2.1.3 We prove that Z = {3m + 4n : m, n Z}. Certainly for any integers m
and n, 3m + 4n is also an integer, so that {3m + 4n : m, n Z} Z. Now let x Z. Then
x = 1 x = (4 3) x = 3(x) + 4x,
so that x {3m + 4n : m, n Z}, whence Z {3m + 4n : m, n Z}. Since we already

proved the other inclusion, the proof is done.
Example 2.1.4 We prove that A = {6m + 14n : m, n Z} equals the set B of all even
integers. Certainly for any integers m and n, 6m + 14n is an even integer, so that A B.
Now let x B. Then x is even, so x = 2n for some integer n. Write
x = 2n = (14 2 6)n = 6(2n) + 14n,
so that x {6m + 14n : m, n Z} = A. Thus B A. Together with the first part this
implies that A = B.
Example 2.1.5 For each i N+ , let Ai = [i, ), Bi = {i, i + 1, i + 2, . . . , }, and

Ci = (i, i). Think through the following:
\ \ \
Ak = , Bk = , Ck = (1, 1),
kN kN kN
[ [ [
Ak = [1, ), Bk = N+ , Ck = R.
kN kN 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 B U
On this Venn diagram, sets are represented by the geometric regions: A is the set
represented by the left circle, B is represented by the right circle, A B is the part of the
two circles that is both in A and in B, A B can be represented by the region that is
either in A or in B, A \ B is the left crescent after B is chopped out of A, etc. (There is
no reason why the regions for sets A and B are drawn as circles, but this is traditional.)
Sometimes we draw a few (or all) elements into the diagram. For example, in
b
w U
A B
b
y b
t b
u
b
x b
z
b
v
we read that A = {x, y, t}, A B = {t}, et cetera.

Two disjoint sets A and B are represented by a Venn diagram as follows:
A B U
Proposition 2.1.7 For all sets A, B, C, we have A (B C) = (A B) (A C).
Proof. With the Venn diagram below, B C is the region filled with either horizontal
or vertical lines, A is the region filled with the Southeast-Northwest slanted lines, and so
A (B C) is the region that has simultaneously the Southeast-Northwest slanted lines
and either horizontal or vertical lines. Also, A C is the region that has horizontal and
Southeast-Northwest slanted lines, A B is the region that has vertical and Southeast-
Northwest slanted lines, so that their union (A B) (A C) represents the total region
of Southeast-Northwest slanted lines that either have horizontal or vertical cross lines as
well, which is the same as the region for A (B C).
C U
A B
We prove this also algebraically. We have to prove that A(B C) (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
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 57
2.1.6 Let A, B U .
i) Prove that there exist at most 16 distinct subsets of U obtained from A, B, U by
intersections, unions, and complementation.
ii) If A and B are disjoint, prove that there exist at most 8 such distinct subsets.
iii) If A = B, prove that there exist at most 4 such distinct subsets.
iv) If A = B = U , prove that there exist at most 2 such distinct subsets.
v) If A = B = U = , prove that there exists at most 1 such subset.
2.1.7 Let A, B, C U . Prove the following statements:
i) (A C) \ B = (A \ B) (C \ B).
ii) (A \ B) (B \ A) = (A B) \ (A B).
iii) (A B) (U \ (A B)) = (U \ (A \ B)) \ (B \ A).
iv) U \ (A \ B) = (U \ A) B.
v) If U = A B, then A = B = U .
2.1.8 Let A, B U .
i) Prove that (U \ A) (U \ B) = U \ (A B). (The intersection of the complements
is the complement of the union.)
ii) Prove that (U \ A) (U \ B) = U \ (A B). (The union of the complements is the
complement of the intersection.)
\ \ \
2.1.9 Compute (1/k, 1/k), [1/k, 1/k], {1/k, 1/k}.
kN+ kN+ kN+
[ [ [
2.1.10 Compute (1/k, 1/k), [1/k, 1/k], {1/k, 1/k}.
kN+ kN+ kN+
2.2 Cartesian product
The set {a, b} is the same as the set {b, a}, as any element of either set is also the
element of the other set. Thus, the order of the listing of elements does not matter. But
sometimes we want the order to matter. We can then simply make another new notation
for ordered pairs, but in general it is not a good idea to be inventing many new notations
and concepts; it is much better if we can reuse and recycle old ones. We do this next:
Definition 2.2.1 An ordered pair (a, b) is defined as the set {{a}, {a, b}}.
So here we defined (a, b) in terms of already known constructions: (a, b) is the set one
of whose elements is the set {a} with exactly one element a, and the remaining element of
(a, b) is the set {a, b} that has exactly two elements a, b if a 6= b and has exactly one element
otherwise. Thus for example the familiar ordered pair (2, 3) 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:
b b b b b
B b b b b b
b b b b b
b b b b
A
If instead A and B are intervals as above, then A B is the indicated rectangle.
When A and B extend infinitely far, then A B is correspondingly a large rectangle:
The familiar real plane is 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.
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 Relation names such as or cousin are familiar, but most relations
do not have such familiar or even good names; think of relation on A and B simply as a
possibly contrived subset of A B.
Definition 2.3.4 Let R be a relation on A.

(1) R is reflexive if for all a A, aRa.
(2) R is symmetric if for all a, b A, aRb implies bRa.
(3) R is transitive if for all a, b, c A, if aRb, bRc, then aRc.
(4) R is an equivalence relation if it is reflexive, symmetric and transitive.
Examples 2.3.5
(1) on R is reflexive and transitive but not symmetric.
(2) < on R is transitive but not reflexive or symmetric.
(3) = on any set A is reflexive, symmetric, and transitive.
(4) Being a cousin is symmetric but not reflexive or transitive.
(5) Being taller than is ...
Definition 2.3.6 Let R be an equivalence relation on a set A. For each a A, the set
of all elements b of A such that aRb is called the equivalence class of a. We denote the
equivalence class of a with the shorthand [a].
For example, if R is the equality relation, then the equivalence class of a is {a}. If
R = A A, then the equivalence class of any a is A. If A is the set of all students in
Math 112 this year, and aRb if students a and b are in the same section of Math 112, then
[a] is the set of all students that are in the same section as student a.
Theorem 2.3.7 Let R be an equivalence relation on a set A. Two equivalence classes are
either identical or they have no elements in common.
Proof. Let a, b A, and suppose that their equivalence classes have an element in common.
Call the element c.
We now prove that the equivalence class of a is a subset of the equivalence class of b.
Let d be any element in the equivalence class of a. Then aRd, aRc and bRc imply by
symmetry that dRa and cRb, so that by transitivity dRc. Then dRc, cRb and transitivity
give dRb, so that by symmetry bRd, which says that d is in the equivalence class of b. Thus
the equivalence class of a is a subset of the equivalence class of b.
A symmetric proof shows that the equivalence class of b is a subset of the equivalence
class of a, so that the two equivalence classes are identical.
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.
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, congru-
ence modulo 7 for days of the week, congruence modulo 4 for seasons of the year, con-
gruence 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.
Example 2.3.11 (Construction of Z from N0 .) Consider the Cartesian product N0 N0 .

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

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.
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 lines in the plane.
i) Prove that the relation is parallel to is an equivalence relation. Note that the
equivalence class of a non-vertical line can be identified by the (same) slope of the
lines in that class. Note that the vertical lines are in their own equivalence class.
ii) Prove that the relation is perpendicular to is not an equivalence relation.
2.3.10 For (a, b), (a, b ) Z Z \ {0}, define (a, b) (a , b ) if a b = a b.
i) Prove that is an equivalence relation. (Possibly mimic Example 2.3.11.)
ii) Describe the equivalence classes of (0, 1), (1, 1), (2, 3)?
iii) Find a natural identification between the equivalence classes and elements of Q.
2.4 Functions
Here is the familiar definition of functions:
Definition 2.4.1 Let A and B be sets. A function from A to B is a rule that assigns to
each element of A a unique element of B. We express this with f : A B is a function.
The set A is the domain of f and B is the codomain of f . The range or image of f is
Image(f ) = Range(f ) = {b B : b = f (a) for some a A}.
But in the spirit of introducing few new 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
f (n) = 2 , if n is odd;
n2 , if n is even.
(4) Let f : N+ R be given by the description that f (n) equals the nth prime.
By 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.
(6) The function f : R R given by f (x) = 1 for all x is not the identity function.
(7) Let b B. A function f : A B given by f (a) = b for all a is called a constant
function.
(8) A function can be given by its graph:
From this particular graph we surmise that f (0) = 0, but with the precision of
the drawing and our eyesight it might be the case that f (0) = 0.000000000004.
Without any further labels on the axes we cannot estimate the numerical values
of f at other points. If numerical values are important, the graph should be filled
in with more information.
(9) A function may be presented by a table:
x f (x)
1 1
2 1
3 2
(10) A function may be presented with Venn diagrams and arrows:

r
r
r
Remark 2.4.4 Two functions are the same if they have the same domains, the same
codomains, and if to each element of the domain they assign the same element of the
codomain.
For example, f : R R and g : R [0, ) given by f (x) = x2 and g(x) = x2 are
not the same! On the other hand, the functions f, g : R R given by f (x) = |x| and

g(x) = x2 are the same.
Notation 2.4.5 It is common to not specify the domain, in which case the domain is
implicitly the largest possible subset of R (after we introduce complex numbers we will
take the largest possible subset of C) on which the function is defined. For example, the

domain of f (x) = x1 is the set of all non-zero real numbers, and the domain of f (x) = x
is the set of all non-negative real numbers.
Definition 2.4.6 Let A, B, 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 when
composing f and g in the other order, (f g)(x) is not always equal to x. Namely, for any
negative number x, (f g)(x) = x.
We just demonstrated that in composing two functions, the order matters.
Definition 2.4.9 A function f : A B is injective (or one-to-one) if for all a, a A,

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

f (x) = x2 is not injective because f (1) = 1 = f (1). The function f : 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 f (x) x f (x) x f (x) x f (x)

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

1 1 1 1 1 1 1 1 1 2 1 2 1 2 1 2
2 1 2 1 2 2 2 2 2 1 2 1 2 2 2 2
3 1 3 2 3 1 3 2 3 1 3 2 3 1 3 2
In this case no functions are injective, and all non-constant ones are surjective.
Theorem 2.4.10 The composition of two injective functions is injective.
Proof. Let f : A B and g : B C be injective functions, and suppose that B B so

that we can compose the two.
If (g f )(a) = (g f )(a ), then g(f (a)) = g(f (a )). Since g is injective, it follows that
f (a) = f (a ), and since f is injective, it follows that a = a . Thus g f is injective.
Theorem 2.4.11 If f : A B and g : B C are surjective functions, so is g f : A C.

If f and g are both bijective, so is g f .
Proof. Let c C. Since g is surjective, there exists b B such that g(b) = c. Since f is
surjective, there exists a A such that f (a) = b. Thus (g f )(a) = g(f (a)) = g(b) = c, so
that g f is surjective.
The last statement follows from the first part and Theorem 2.4.10.
Definition 2.4.12 We have seen polynomial functions in Section 1.4: recall that for any
subset S R, a function f : S R is a polynomial function if there exist a non-negative
integer n and c0 , c1 , . . . , cn R such that for all x S, f (x) = c0 + c1 x + c2 x2 + + cn xn .
If cn 6= 0, we say that the degree of f is n. A function f : S R is a rational function
if there exist polynomial functions f1 , f2 : S R such that for all x S, f2 (x) 6= 0 and
f (x) = f1 (x)/f2 (x).
Similarly there are polynomial and rational functions if all occurrences of R above
are replaced by Q or C .
Polynomial and rational functions are very special, and are a workhorse of analysis.
(The reader has of course encountered trigonometric, exponential, and logarithmic func-
tions, which are not polynomial or rational.) The following are some special properties:
Theorem 2.4.13 If a polynomial function is not constant zero, it has only finitely many
roots. In fact, the number of roots is at most the degree of the polynomial.
The domain of a rational function is the complement of a finite subset of R (or C).
Proof. Suppose that f (x) = c0 + c1 x + c2 x2 + + cn xn is not constant 0. Thus at least

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

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

roots. If b is a root of f , then
0 = f (b) = q(b)(b a),
so that either q(b) = 0 of b a = 0. Thus b is either a root of q or b = a. Thus the roots of

f are a or any of the at most n 1 roots of q, so that f has at most n roots. This proves
the first part.
A rational function is a quotient of two polynomial functions, and the rational function
is defined everywhere except where the denominator is 0. By the first part this excludes
only finitely many numbers.
This theorem is useful in that it assures us that the domain of a rational function
contains infinitely many points, or even better, that the domain is all except finitely many
real (or complex) numbers.
c1
qc2 4c0 c2
* 2
If n = 2, then by the quadratic formula the roots of c0 + c1 x + c2 x are 1
2c2 . There are root
formulas for cases n = 3, 4, but executing them is very time-consuming, they require familiarity with complex
numbers, the solutions involve sums of cube roots with a few square roots thrown in for good measure, and
furthermore it can be hard to identify that the ensuing long expression simplifies to a nice number such as 2.
This, and the existence of computer capabilities, are the reasons that we do not teach such formulas. The formula
for solutions of cubic polynomials was discovered by 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.
(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
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}, B = {2, 4}, and C = {3, 6, 9}. Mark each of the following as a
function A B, B C, or C A.
i) f = {(2, 6), (4, 3)}.
ii) g = {(9, 1), (6, 2), (3, 2)}.
iii) h = {(1, 4), (2, 2)}.
2.4.8 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, 2), (4, 2)}.
ii) {(9, 4), (3, 2), (6, 2)}.
iii) {(2, 6), (1, 3)}.
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 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.
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
i) f (x) = x2 x3 x .
x4 2x3 1
ii) f (x) = x6 +2 .
1
(iii)* f (x) = x5 17x2 +x4 . (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 f .
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 )(cj 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 .
Section 2.5: Binary operations on sets 73
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 225.)
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 that we have seen are:
+, , , /, , , , , , , xor, =, .
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).
Section 2.8: Increasing and decreasing functions 89
2.7.17 (In-betweenness in an ordered field) Let F be an ordered field. Let x, y F with

x < y. Prove that x < (x + y)/2 < y. (Why are we allowed to divide by 2?)
2.7.18 Find an ordered set 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. Prove that for all s S, min(S) s max(S).
2.7.20 Let m R be a positive number. Prove that there exists a positive integer N such
that 1/2N < m.
2.7.21 Let n > 1 be an integer and F = Z/nZ. (This was defined in Example 2.3.10.)
Prove that F is not an ordered set. In particular, using Exercise 2.6.5, for any prime integer
p, Z/pZ is a field that is not ordered.
2.8 Increasing and decreasing functions

Definition 2.8.1 Let F, G be ordered sets (as in the previous section), and let A F . A
function f : F G is increasing (resp. decreasing) on A if for all x, y A, x < y implies
that f (x) f (y) (resp. f (x) f (y)). If furthermore f (x) < f (y) (resp. f (x) > f (y) for
all x < y, then we say that f is strictly increasing (resp. strictly decreasing) on A. A
function is (strictly) monotone if it is (strictly) increasing or (strictly) decreasing.
Theorem 2.8.2 Let n be a positive integer and F an ordered field. Then the function
f : F F defined by f (x) = xn is strictly increasing on F + {0}.
Proof. Let x, y F + with x < y. Then by Exercise 1.6.7,
f (y) f (x) = y n xn = (x + (y x))n xn

Xn X n
n1
n k nk n
= x (y x) x = xk (y x)nk .
k k
k=0 k=0

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

of the non-negative summands is n0 x0 (y x)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 field. Suppose that x, y
F + {0} have the property that xn < y n . Then x < y. (In other words, the nth radical
function is strictly increasing on F + {0}.)
Proof. If x = y, then xn = y n , which contradicts the assumption and trichotomy. If x > y,

then by Theorem 2.8.2, xn > y n , which also contradicts the assumption. So by trichotomy
x < y.
Theorem 2.8.4 If F, G are ordered sets and f : F G is strictly monotone, then f is

injective, and there exists a strictly monotone function g : Range(f ) F such that for all
x F , (g f )(x) = x and for all y G, (f g)(y) = y. In other words, g is the inverse of
the function f : F Range(f ).
Proof. Let y Range(f ). Then y = f (x) for some x F . If also y = f (z) for some z F ,
since f is strictly monotone, x = z. So f is injective and x is unique. Thus we define
g : Range(f ) F by g(y) = x. Then by definition for all x F , g(f (x)) = x and for all
y Range(f ), f (g(y)) = y. If f is increasing and y1 , y2 Range(f ) such that y1 < y2 , then
g(y1 ) < g(y2 ) for otherwise by the increasing property of f , y1 = f (g(y1)) f (g(y2 )) = y2 ,
which is a contradiction. Thus if f is increasing, so is g = f 1 . Thus if g = f 1 is
increasing, so is f = (f 1 )1 . The same reasoning goes for the decreasing property.
Remark 2.8.5 Let F be an ordered set and G an ordered field. Let f, g : F G be

functions. We define functions f + g, f g : F G by (f + g)(x) = f (x) + g(x) and
(f g)(x) = f (x) g(x). (Instead of f g we simply write f g.)
(1) If f, g are both strictly increasing (resp. decreasing), then f +g is strictly increasing
(resp. decreasing).
(2) If f, g are both strictly increasing (resp. decreasing) and always take on positive
values, then f g is strictly increasing (resp. decreasing).
(3) If f, g are both strictly increasing (resp. decreasing) and always take on negative
values, then f g is strictly decreasing (resp. increasing).
Proof of (3): Let x, y F with x < y. Suppose that f and g are both increas-
ing functions, so that f (x) < f (y) < 0 and g(x) < g(y) < 0. Then f (y), g(x)
are positive numbers, so by compatibility of < with multiplication by positive numbers,
f (x)(g(x)) < f (y)(g(x)) and (f (y))g(x) < (f (y))g(y). By Exercise 2.6.8, this
says that (f (x)g(x)) < (f (y)g(x)) and (f (y)g(x)) < (f (y))g(y)). By transitiv-
ity of < then (f (x)g(x)) < (f (y))g(y)). By compatibility of < with addition, by
adding f (x)g(x) + f (y)g(y) we get that f (y)g(y) < f (x)g(x). With function notation,
(f g)(y) < (f g)(x), and since x and y were arbitrary, this says that f g is strictly decreas-
ing. The proof in the case where both f and g are strictly decreasing is similar.

2.8.1 Let n be an odd positive integer and F an ordered field. Prove that the function
f : F F defined by f (x) = xn is strictly increasing.
2.8.2 Let n be an even positive integer and F an ordered field. Prove that the function
f : F {0} F defined by f (x) = xn is strictly decreasing.
Section 2.9: The Least upper bound property of R 91
2.8.3 Let n be an odd positive integer and F an ordered field. Suppose that a, b F and
that an < bn . Prove that a < b.
2.8.4 Let F be an ordered field, a F + and f : F F defined by f (x) = ax. Prove that
f is a strictly increasing function.
2.8.5 Let F be an ordered field, a F and f : F F defined by f (x) = ax. Prove that
f is a strictly decreasing function.
2.8.6 Prove that the composition of (strictly) increasing functions is (strictly) increasing.
Prove that the composition of (strictly) decreasing functions is (strictly) increasing.
2.8.7 Prove that the composition of a (strictly) increasing function with a (strictly) de-
creasing function, in any order, is (strictly) decreasing.
2.8.8 Suppose that f : F G is strictly monotone.
i) Let B be a subset of F , and define g : B G by g(x) = f (x). Prove that g is
strictly monotone.
ii) Define h : B Range(g) by h(x) = f (x). Prove that h is bijective.
2.8.9 Give an example of a non-decreasing function f : R R that is not injective.
2.9 The Least upper bound property of R
In this section we formalize and analyze the important Least upper bound property
of R that we accept without proof.
Axiom 2.9.1 (Least upper/greatest lower bound property) For any non-empty
subset T of R that is bounded above, sup(T ) exists in R.
Similarly, for any non-empty subset T of R that is bounded below, inf(T ) exists in R.
Moreover, by Theorem 2.7.6, sup(T ) and inf(T ) are unique.

For example, the non-empty bounded set {x Q : x2 < 2} in R has infimum 2

and supremum 2 in R. In contrast, the same set has no infimum nor maximum in Q. So,
R satisfies the least upper bound property but Q does not.
We will use this Least upper bound property in the proof of the Intermediate value
theorem Theorem 5.3.1, in the proof of the Extreme value theorem Theorem 5.2.2, and in
many limit tricks for functions and sequences. Here is the first example of its usage.
Theorem 2.9.2 Every positive real number is the square of a positive real number.
Proof. Let y be a positive real number. Let S = {r R : 0 r and r 2 y}. Then S

contains 0, so it is non-empty. If for some r S, r > y, then r > y r 2 , so that by
compatibility of multiplication/division by positive numbers, 1 r. But y is positive, so
y is bigger than or equal to 1, so that 1 r > y 1, which by transitivity contradicts
the trichotomy. So necessarily for all r S, r y. This means that S is bounded above.
By the Least Upper Bound property of R, there exists a real number s that is the least
upper bound on S.
We claim that s2 = y. Suppose first that s2 < y. [We want to find a small
positive number q such that (s + q)2 < y, which would say that s + q S and
thus give a contradiction to s being an upper bound on S. The small number
q might as well be of the form p1 for some (large) positive integer p, so that
we really want to find some p N+ such that s2 + 2s 1
p + p2 < y, and for that it
suffices to find p N+ such that 2s + 1 < p(y s2 ).] By the Archimedean property
there exists a positive integer p such that 2s + 1 < p(y s2 ). We know that 1p > 0. Since
p 1, by compatibility of < with multiplication by the positive number 1p we have that
1 1p . Thus 2s + p1 2s + 1 < p(y s2 ). By compatibility of multiplication again we get
that 2s/p+1/p2 < ys2 , so that by compatibility with addition, (s+ p1 )2 = s2 + 2s 1
p + p2 < y.
This means that s + p1 S, so that s is not an upper bound on S, which is a contradiction.
We conclude that necessarily s2 y. So for all r S, r 2 y s2 . Suppose next that
s2 > y. [We want to find a small positive number q such that (s q)2 > y. This
would prove that for all r S, r 2 y < (s q)2 . We need to guarantee that
s q is still positive, and then by the increasing property of the squaring
function on R+ , for all r S, r s q, which contradicts the fact that
s is the supremum of S. This q might as well be of the form p1 for some
large positive integer p.] By the Archimedean property there exists a positive integer
p1 such that 2s < p1 (s2 y), and there exists a positive integer p2 such that 1 < p2 s.
Let p = max{p1 , p2 }. Then 2s < p(s2 y) and 1 < ps. The assumption 1 < ps implies
that s 1p is positive. Similarly to before, we derive that 2s p1 < p(s2 y) and with
compatibilities we derive that y < (s p1 )2 . Then for any r S, r 2 < y < (s p1 )2 . But
then by Corollary 2.8.3, r < s 1p . Hence s p1 is an upper bound on S, contradicting
the assumption that s is the least upper bound on S. This proves that s2 y. Since we
already proved that s2 y, the trichotomy on R gives that s2 = y.

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

Definition 2.10.1 Let F be an ordered field. The absolute value function | | : F F
is a function defined as (
0; if x = 0;
|x| = x; if x F + ;
x; if x F .
This defines the absolute value function on the ordered fields Q and R.
The following theorem lists the familiar properties of absolute values, and the reader
may wish to prove them without reading the given proof.
Theorem 2.10.2 Let F be an ordered field.

(1) For all x F , |x| 0. Furthermore, x 0 if and only if x = |x|; and x 0 if and
only if x = |x|. (In particular, |1| = 1.)
(2) For all x F , |x| = | x|.
(3) For all x F , |x| x |x|.
(4) For all x, a F , |x| a if and only if a x a.
(5) For all x, a F , |x| < a if and only if a < x < a.
(6) 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|. In particular, when applied to x, this says that x | x| = |x|, and by
adding x |x| to both sides we get that |x| x.
(4) Suppose that |x| a. Then by (3) and transitivity, x a, and x | x| = |x|
a, so that by transitivity and adding x a to both sides, a x.
(5) The proof of (5) is similar to that of (4).
(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 2.10.3 The following inequalities hold for an ordered field F .

(1) Triangle inequality: For all x, y F , |x y| |x| + |y|.
(2) Reverse triangle inequality: For all x, y F , |x y| ||x| |y|| = ||y| |x||.
Proof. (1) By the first part of the previous theorem, |x| x |x| and |y| y |y|.
Thus
(|x| + |y|) = (|x|) + (|y|) (by Theorem 2.5.14)

x + (|y|) (by compatibility of with addition)
x + y (by compatibility of with addition)
|x| + y (by compatibility of with addition)
|x| + |y| (by compatibility of with addition),
so that by transitivity, (|x| + |y|) x + y |x| + |y|. Thus by part (4) of the previous
theorem, |x + y| |x| + |y|. It follows that |x y| = |x + (y)| |x| + | y|, and by the
second part again this is equal to |x| + |y|.
(2) By (1), |x| = |x y y| |x y| + |y|, so that |x| |y| |x y|. Similarly,
|y| |x| |y x|. But by the second part of the previous theorem, |y x| = | (y x)| =
|x y| and |y + x| = |x + y|, so that |x y| |x| |y| and |x y| |y| |x| = (|x| |y|).
Since ||x| |y|| is either |x| |y| or |y| |x|, (2) follows.
Theorem 2.10.4 Let F be an ordered field. Let r F .

(1) If r < for all F + , then r 0.
(2) If r > for all F + , then r 0.
(3) If |r| < for all F + , then r = 0.
Proof. Proof of (1): By Theorem 2.7.10, 1 F + , so that 0 < 1, and by compatibility of

< with addition, 1 < 2. Thus by transitivity of <, 0 < 2, so that 2 is positive, and by
Theorem 2.7.10 (3), 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 (2) is similar.
For (3), if |r| < for all F + , then < r < . Then by (1) and (2), 0 r 0.
Since F + F = by trichotomy, it follows that r = 0.

2.10.1 Let F be an ordered field. Prove that the absolute value function on F is increasing
on F + {0} and decreasing on F {0}.
2.10.2 Let F be an ordered field.

i) Prove that for all a, b F , |a| + |b| = |a| + |b|.

ii) Prove that for all a, b F + , |a| + |b| = a + b.

iii) Prove that for all a, b F , |a| + |b| = a b.
2.10.3 Let F be an ordered field, a F and r F + . Express the sets {x F : |xa| < r}
and {x F : |x a| r} in interval notation.
2.10.4 (Triangle inequality) Let F be an ordered field and a1 , . . . , an F . Prove that
|a1 + a2 + + an | |a1 | + |a2 | + + |an |.
2.10.5 (Reverse triangle inequality) Let F be an ordered field and a1 , . . . , an F . Prove

that
|a1 + a2 + + an | |a1 | |a2 | |an |.

Give an example of real numbers a1 , a2 , a3 with |a1 + a2 + a3 | 6 |a1 | |a2 | |a3 |.
2.10.6 Let F be an ordered field and S a subset of F .
i) Prove that S is bounded if and only if {|s| : s S} is bounded.
ii) Give an example of a set S in R that is bounded above but {|s| : s S} is not.
Chapter 3: The field of complex numbers, and topology
In this chapter we construct complex numbers from the real numbers. Even if we
want to measure only real numbers in the real world, the more general complex numbers
streamline many constructions, so they are an important tool. We prove the important
basic properties of complex numbers and their field C in the first two sections. As a starting
point we take it as fact that R is an ordered field with the Least upper bound property
and the Archimedean property.
The last two sections are about the Euclidean topology on R and C, which is crucial
in the subsequent chapters on limits. I cover those two sections lightly but invoke them as
needed.
3.1 Complex numbers
Note that there is no real number x such that x2 = 1. In this section we build the
smallest possible field containing R with an element whose square is 1.
Definition 3.1.1 Let C = R R (the Cartesian product). Elements of C are called

complex numbers. Define binary operations + and on C:
(a, b) + (c, d) = (a + c, b + d),
(a, b) (c, d) = (ac bd, ad + bc).
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.
a b a b
(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 .
(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 3.1.2 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
a + bi is aabi 35i
2 +b2 . For example, the multiplicative inverse of 3 + 5i is 32 +52 =
35i 3 5i
34 = 34 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 the following natural inclusions that are
compatible with addition and multiplication:
N0 Z Q R C.
These number systems progressively contain more numbers and more solutions of more
equations. For example, the equation 1 + x = 0 does not have any solutions in N0 but it
does have one in Z; the equation 1 + 2x = 0 does not have any solutions in Z, but it does
have one in Q; the equation 2 x2 = 0 does not have any solutions in Q, but it does have

two in R (namely 2 and 2); the equation 2 + x2 = 0 does not have any solutions in

R, but it does have two in C (namely 2i and 2i). 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:
b b
(0, 1) = i
(3, 1) = 3 + i b
(2, 0.5) = 2 + 0.5i
(0, 0) = 0
b b b
(1, 0) = 1 (1, 0) = 1
b
(0, 1) = i
* One of the excellent properties of C is the Fundamental Theorem of Algebra: every polynomial with
coefficients in C or R has roots in C. The proof of this fact is proved in a junior-level class on complex analysis or
in a senior-level class on algebra. The theorem does not say how to find those roots, only that they exist. In fact,
there is another theorem in Galois theory that says that in general it is impossible to find roots of a polynomial
by using radicals, sums, differences, products, and quotients.
98 Chapter 3: The field of complex numbers, and topology
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 3.1.6.
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 3.1.3 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 3.1.4 Let x, y C. Then

(1) The complex conjugate of the complex conjugate of x equals x. In symbols: x = x.
(2) x + y = x + y.
(3) x y = x y.
(4) If y 6= 0, then y 6= 0 and (x/y) = x/y.
(5) x = x if and only if x R; and x = 0 if and only if x = 0.
Proof. Write x = a + bi and y = c + di for some a, b, c, d R. Certainly x = a bi =

a + bi = x, proving (1). Addition in (2) is straightforward. The following proves (3):
x y = (a + bi) (c + di)
= ac bd + (ad + bc)i
= ac bd (ad + bc)i
= (a bi) (c di)
= x y.
If y 6= 0, then by (3), x = (x/y)y = x/y y, and so (4) follows.

Certainly x = x if and only if a + bi = a bi, i.e., if and only if (a, b) = (a, b), and
that holds if and only if b = 0. Thus x = x if and only if x = a R. Also, x = 0 if and only
if (a, b) = (0, 0), which is the same as saying that x = (a, b) = (0, 0). This proves (5).
Facts 3.1.5 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.
(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 3.1.6 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.

3.1.1 Write the following elements in the form a + bi with a, b R:
1
i) 1+i .
ii) (4+i)(73i)
(1i)(3+2i) .
iii) (2 3i)5 . (Hint: Exercise 1.6.7.)

3
3.1.2 Let = 2 + 2i . Draw , 2 , 3 , 4 , 5 , 6 , 7 . What do you observe?
3.1.3 Draw the following sets in C:
i) {x : the real part of x is 3}.
ii) {x : the imaginary part of x is 3}.
iii) {x : the product of the real and imaginary parts of x is 3}.
iv) {x : the product of the real and imaginary parts of x is 0}.
3.1.4 Let x C such that x2 = 1. Prove that either x = i or x = i.
3.1.5 Prove that the only x C with x2 = 0 is x = 0.
3.1.6 Prove that C is not ordered in the sense of Definition 2.7.8. Justify any facts (such
as that 0 < 1).
3.1.7 Let x, y C.
i) Prove that x + x = 2 Re x.
ii) Prove that x x = 2i Im x.
iii) Prove that x y + x y = 2 Re(xy) = 2 Re x Re y + 2 Im x Im y.
3.1.8 Let x be a complex number. Prove that x x is real.
3.1.9 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.
3.1.10 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.
3.2 Absolute value in C and polar coordinates
We have seen the absolute value function in ordered fields. Even though the field C is
not ordered (see Exercise 3.1.6), nevertheless, the Pythagorean theorem in the plane R R
motivates the natural definition of distance in C:
bi b
(a, b) = a + bi
Definition 3.2.1 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 3.1.6: C
is not an ordered field.)
The absolute value of (a, 0) = a or (0, a) = ia is |a|; the absolute value of (1, 1) = 1 + i

is 2; the absolute value of (1, 2) = 1+i 2 is 3; the absolute value of (1, 3) = 1+i 3

is 4 = 2, et cetera.
Theorem 3.2.2 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||.
p
Proof. Write x = a+bi for some a, b R. Then |x| = a2 + b2 = a2 + (b)2 = |x|, which
proves (1). Also, x x = (a + bi)(a bi) = a2 (bi)2 = a2 + b2 = |x|2 , and this is the sum
of two non-negative real numbers, and is thus non-negative. This proves (2). Furthermore,
since R is an ordered field, by Exercise 2.7.11, 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| |x|. Similarly | Im x| |x|. This proves (4).

If x 6= 0, then by (3), |x| is a non-zero (real, complex) number, and by (2), x/|x|2 is
the multiplicative inverse of x. This proves (5).
We could prove (6) with straightforward but laborious algebra by using that x = a+bi
and z = c + di for some real numbers a, b, c, d, and expanding the relevant sides, but the
following proof is better:
|xy|2 = (xy)(xy) (by (3))
= xyx y (by Theorem 3.1.4)
= xxyy (by associativity and commutativity of in C)
= xyxy (by Theorem 3.1.4)
= 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) (by (2))
= (x y)(x y) (by Theorem 3.1.4 (2))
= xx xy yx + yy (by algebra)
= |x|2 xy xy + |y|2 (by Theorem 3.1.4 (2))
= |x|2 2 Re(xy) + |y|2 (by Exercise 3.1.7 iii))

|x|2 + 2| Re(xy)| + |y|2 (comparison of real numbers)
|x|2 + 2|xy| + |y|2 (by (4))
= |x|2 + 2|x||y| + |y|2 (by (6))
= |x|2 + 2|x||y| + |y|2 (by (1))
= (|x| + |y|)2,
and since the squaring function is strictly increasing on the set of non-negative real numbers,
it follows that |x y| |x| + |y|. This proves (7), and by Theorem 2.10.3 also (8).
So far we have expressed complex numbers with pairs of real numbers either in ordered-
pair 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 polar coordinates: from every point x in C we can read off 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). That angle is not unique: if is an
angle for x, so are + 2n for any integer n. So for example the angle 3/2 = 2 /2 on
the picture below can also be labeled as /2, and the angle 5/3 as /3. In addition,
if x = 0, any choice of will do.
radius 1, angle /2
b
b
radius 1, angle /4
radius 2, angle
b 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
angle /3 counterclockwise from the positive real axis, and 1+i 2
3
is on the same circle
and at angle 2/3 counterclockwise from the positive real axis.
As pointed out, a complex number x does not determine the angle uniquely, but the
length of a number and an angle uniquely determine the number.
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 3.2.3 (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 around (0, 0).
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:
b
(c1 , d1 )
(c2 , d2 )
(d1 , c1 ) b b b
(d4 , c4 )
(d2 , c2 ) b b b
(d3 , c3 )
(c4 , d4 )
(c3 , d3 ) b
The complex number M (i) = (d, c) has length equal to |(c, d)| = |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))
= aM (1) + bM (i) (by previously proved cases)
= az1 + bzi
= z(a + bi)
= M (x).
Theorem 3.2.4 For any non-zero complex number x and any integer n, the angle of xn
counterclockwise away from the positive x-axis is n times the angle of x.
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 3.2.3, 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 3.2.3, 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 3.2.5 Let n be a positive integer. Let A be the set of all complex numbers
on the unit circle at angles 0, 2 2 2 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 non-
negative 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):
Definition 3.2.6 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.)

3.2.1 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.
3.2.2 Draw the following points in the real plane, and think about Theorem 3.2.3:
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).
3.2.3 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}.
3.2.4 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.
3.2.5 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.
3.2.6 Let a, b C. Suppose that for all real numbers > 0, |a b| < . Prove that a = b.
(Hint: Theorem 2.10.4.)
3.2.7 Square the pentagon drawn below. Namely, estimate the coordinates (real, imag-
inary or length, angle) of various points on the pentagon, square the point, and draw its
image on a different real plane. You need to plot the image not only of the five vertices, but
of several representative points from each side. (Hint: You may want to use Theorem 3.2.3.)
b
i
3.2.8 Prove that for any non-zero z C there exist exactly two elements in C whose
square equals z. (Hint: Theorem 2.9.2 and Theorem 3.2.3.)
3.2.9 Let z be non-zero in C and n N+ . Set r = |z| and and let be the angle of z
measured counterclockwise from the positive x axis. For and integer k, let zk be the complex
number whose absolute value equals r 1/n and whose angle measured counterclockwise from
the positive x axis is k/n.
i) Prove that zk is uniquely determined.
ii) Prove that for all k, zkn = z. (Hint: Theorem 3.2.3.)
iii) Prove that there the set {zk : k N} contains exactly n elements.
(iv)* (You cannot do this part yet, we have not yet studied continuity.) Thus we can talk
about nth roots of z in C, but no choice of these roots would create a continuous
nth root function from C to C.
3.2.10 Discuss correctness/incorrectness issues with the following equalities:

p
i) 6 = ( 3i)( 12i) = 3 12 = (3)(12) = 36 = 6.
ii) (R. Bombelli, 1560, when solving the equation x3 = 15x + 4.)
q q
3 3
4 = 2 + 121 + 2 121.
iii) (G. Leibniz, 1675)

q q

1+ 3 + 1 3 = 6.
3.3 Topology on the fields of real and complex numbers
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
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 3.3.1 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 3.3.2).
1 0 1 1 0 1
Examples 3.3.2
(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 52).
(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 3.3.3 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 3.3.4 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.
Definition 3.3.5 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 3.3.6
(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 3.3.7 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,
|ba| < r. Let x B(b, r |ba|). Then by the triangle inequality |xa| = |xb+ba|
|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 3.3.8 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 3.3.9 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 3.3.3, 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 3.3.10 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.
Examples 3.3.11
(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 3.3.12 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 3.3.13 (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 3.3.12 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.
Definition 3.3.14 A set A is a closed set if Bd A A.
Theorem 3.3.15 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 3.3.16 (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 3.3.3). Some sets are neither open or closed (see
Exercise 3.3.2).
Definition 3.3.17 For any set A, its closure A is the smallest closed set containing A.
Theorem 3.3.18 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 the triangle inequality B(y, r |y
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 3.3.19 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 the triangle inequality |y a| = |y x + x a| |y x| + |x a| < (r |x a|) +
|x a| = 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
the 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}.

3.3.1 Prove that a subset of R or C is closed if and only if it contains all of its own limit
points.
3.3.2 Let A be the open ball in R of radius 1 and centered at 0. Since R is a subset of C,
then A is also a subset of C. Prove that A is neither a closed nor an open subset of C.
3.3.3 Let F be either R or C, and let A be a closed and open subset of F . Prove that
A = or A = F .
3.3.4 Prove that Q is neither an open nor a closed subset of R or C.
3.3.5 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+ }.
3.3.6 For each of the following intervals as subsets of R, determine the interior and bound-
ary sets, the set of limit points, and whether the sets are open, closed, or neither:
(0, 1), [0, 1], [3, 5), [3, ).
3.3.7 Find the boundary and the set of limit points of {x C : Re x < 0}.
3.3.8 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.
3.3.9 Give examples of sets A B C and a C such that a is a limit point of B but
not of A.
3.3.10 Prove that the closure of B(a, r) is {x F : |x a| r}.
3.3.11 Prove that the closure of { n1 : n N+ } is {0} { n1 : n N+ }.
3.3.12 Prove that the closure of Q in R and in C is R.

3.3.13 Let B = {a + bi : a, b Q}. Determine the closure of B in C.
3.3.14 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.
3.3.15 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).
*3.3.16 Prove that a set A is closed if and only if A = A.
3.3.17 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}.
3.3.18 Let A be a subset of R and let a complex number a be a limit point of A. Prove
that a R.
3.4 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 5.3 that when a good (say continuous) real-valued function has a
closed and bounded domain, then that function achieves a maximum and minimum value,
et cetera. The concept of uniform continuity (introduced in Section 5.5) needs the fairly
technical Heine-Borel theorems proved in this section. (I used to not cover this section in
class, but with the later replacement of the proof with one using the powerful quartering
and halving construction, I do present this section.)
Construction 3.4.1 (Halving closed and bounded subsets of R and quartering closed and
bounded subsets of C) Let A be a bounded subset of of R or of C, and let P be a property
that applies to subsets of A. Boundedness of A guarantees that A fits inside a closed
bounded rectangle R0 in C. If A is a subset of R, we take the rectangle R0 to have
height zero. This rectangle can be halved lengthwise and crosswise to get four equal closed
subrectangles. In the next iteration we pick, if possible, one of these four closed quarter
subrectangles such that its intersection with A has property P . We call this subrectangle
R1 , and note that the side lengths are half the side lengths in R0 (and so the area of R1
is one quarter of the area of R0 ). In general, once we have Rn , we pick its closed quarter
subrectangle Rn+1 such that Rn+1 A has property P and the sides of Rn+1 are half the
lengths of the sides in Rn . Write Rn = [an , bn ] [cn , dn] for some real numbers an bn
and cn dn . By construction, for all n, bn an = (b0 a0 )/2n, and
a0 a1 a2 an bn b2 b1 b0 .
This means that {a1 , a2 , a3 , . . .} is a non-empty subset of R that is bounded above, so that
by the Least upper bound theorem (Axiom 2.9.1), a = sup{a1 , a2 , a3 , . . .} is a real number.
By definition, a b1 , b2 , b3 , . . .. Similarly, b = inf{b1 , b2 , b3 , . . .} is a real number that is
greater than or equal to a. Suppose that a < b. Then by Exercise 2.7.20, there exists a
positive integer N such that 1/2N < (b a)/(b0 a0 ). But aN a b bN , so that
0 b a bN aN = (b0 a0 )/2N < b a, which contradicts trichotomy. Thus a = b,
i.e., we just proved that
sup{a1 , a2 , a3 , . . .} = inf{b1 , b2 , b3 , . . .}.

Similarly,
c = sup{c1 , c2 , c3 , . . .} = inf{d1 , d2 , d3 , . . .}.
This means that the intersection of all the Rn equals the set {a + ci}, consisting of exactly
one complex number. By the shrinking property of the subrectangles, for every > 0 there
exists a positive integer N such that RN A B(a + ci, ).
In particular, as indicated above, if A = [a0 , b0 ] R, quartering of this closed and
bounded region (interval) means halving the rectangle (interval), and the intersection of
all the chosen closed half-rectangles is a set with exactly one element. That element is in
A, so a real number.
Theorem 3.4.2 (Heine-Borel theorem (in R, C)) Let A be a closed and bounded
subset of R or 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. We say that a subset B of A satisfies (property) P if there exists a finite subset S
of B such that B cS B(c, c ). We want to prove that A has P .
Suppose for contradiction that A does not have P . Since A is closed and bounded, it
fits inside a closed rectangle R0 . With Construction 3.4.1, we construct iteratively nested
subrectangles R0 R1 R2 . The quarter subrectangles are chosen so that each Rn
does not have P . This is true if n = 0 by assumption. Suppose that Rn has been chosen
to not have P . If each of the four quarter subrectangles of Rn has P , i.e., if each of the
four subrectangles (as in Construction 3.4.1) intersected with P is contained in the union
of finitely many balls B(c, c ), then Rn is covered by finitely many such balls as well, which
contradicts the assumption on Rn . Thus it is possible to choose Rn+1 to not have P . By
construction, n=1 Rn contains exactly one point. Let that point be x.
By the shrinking sizes of the Rn , there exists a positive integer N such that Rn
B(x, x ). But then Rn A has P , which contradicts the assumption that A does not have
P . Thus A has P .
Theorem 3.4.3 Let A be a closed and bounded subset of R or C, and for each a A
let a be a positive number. Then 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 ).
Proof. Note that c /2 is also a positive number. Then by Theorem 3.4.2, there exists a
finite subset S of A such that A cS B(c, c /2). Let = 12 min{c : c S}. Since S is a
finite set, is a positive real number.
Let x A. By the choice of S there exists c S such that x B(c, c /2). Let
y B(x, ). Then
|y c| = |y x + x c| |y x| + |x c| < + c /2 c ,
so that y B(c, c ). It follows that B(x, ) B(c, c ).

3.4.1 Let A = {x R : 2 x 3}. Prove that A aA B(a, 1/a). Does there exist a
finite set S of A such that A aS B(a, 1/a). If yes, find it, if no, explain why not.
3.4.2 Let A = {1/n : n N+ } R. Prove that A aA B(a, a/3). Does there exist a
finite set S of A such that A aS B(a, a/3). Repeat with A = N+ .
3.4.3 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 3.4.2.)
Chapter 4: Limits of functions
4.1 Limit of a function
Any calculus class teaches about limits, but the domains there are typically intervals
in R. Here we learn a more general definition for (more interesting) domains in C.
Definition 4.1.1 Let A be a subset of C and let f : A C be a function. Suppose that

a complex number a is a limit point of A (see Definition 3.3.8). 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 behavior of the function f at points near a.
We can give a simple geometric picture of this in case the domain and codomain are
subsets of R (refer to Remark 3.1.6 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.
The function f itself might be any of the following:

L L L
a a a
In an intuitive sense we are hoping that f (x) for x near a can predict a value of f as we
get arbitrarily close to a. For example, we may not be able to bring x to 0 Kelvin, but if we
can take measurements f (x) for x getting colder and colder, perhaps we can theoretically
predict what may happen at 0 Kelvin. But how believable is our prediction? Perhaps for
our theory to be satisfactory, we need to run experiments at temperatures x that give us
f (x) within = 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 def-
inition is an algebraic formulation, and as such it requires algebraic proofs. In the rest
of the section we examine many examples algebraically, with the goal of mastering the
epsilon-delta proofs. But epsilon-delta proofs are time-consuming, so in the future we will
118 Chapter 4: Limits of functions
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 3 is a limit point 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-or-other
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 a
demonstration that it works, but how one finds that is in general left to
scratch work or to inspiration. ] Set = /4. [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. [Sure! 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, often with triangle
inequalities... until we get < . ] Then
|(4x 5) 7| = |4x 12|
= 4|x 3|
< 4

= 4 [Wasnt this a clever choice of ?]
4
= .
The commentary in the proof above is describing the thought process behind the proof
but need not and should not be written in homework solutions. Below is the proof without
the commentary: 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 3 is a limit point
of the domain. Let > 0. Set = /4. Then is a positive real number. Let x be an
arbitrary complex number. Assume that 0 < |x 3| < . Then
|(4x 5) 7| = |4x 12|
= 4|x 3|
< 4

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

number depending on that made the limit proof work. Other numbers would have
worked as well, such as /5, or /(10000 + ), and so on. For good style choose simple
formulations over complicated ones. If some positive number works as , so does any
smaller positive, so there is no smallest possible and there are many different correct
choices of . It is not necessary to find the largest possible for a given , and it is not
even necessary to show how you derived or chose your . However, it is necessary to show
that your pick of does satisfy the rest of the defining property of limit.
Think through the comments in the previous paragraph on limx4 5 = 5: any positive
number works for .
In further examples we show all the necessary work for proofs of limits including
how to determine the . It is standard to use the proof-writing trick of partial filling-in:
when by the definition of limit it is time in the proof to declare what is, we write Set
= , and in subsequent reasoning we fill in the underlined blank
with whatever restrictions seem necessary. Study the proofs below for how this is done,
but let me just say that all restrictions put an upper bound on , say bound above
by 1 and by 2 /4, and this is accomplished by filling in Set = min{1, 2 /4 },
with room for possible further restrictions. (In Example 4.1.2 we only had one restriction,
namely /4, so that = min{/4} = /4.)
I find that the following two goals make these proofs more concrete and doable:
Goal #1: Write |f (x) L| as (something)|x a|, or actually it suffices to have |f (x)
L| (something)|x a|.
Goal #2: Find a positive constant upper bound B on the (something) found in Goal
#1. Then you make sure that /B.
You may want to keep these guidelines in mind as you read the proofs below.
Example 4.1.3 lim (4x2 5x + 2) = 28.

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

defined for all complex numbers. Thus 2 is a limit point of the domain. Let > 0.
Set = [ to be determined still; the final write-up will
have this filled in, but we do not yet know .]. 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 above 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, }. This guarantees that will be 1 or
smaller, depending on what gets filled in after the comma.]
= |4(x + 2 2) 13| |x + 2| (adding a clever zero)
= |4(x + 2) 21| |x + 2|

|4(x + 2)| + 21 |x + 2| (by the triangle inequality |a b| |a| + |b|)
< (4 1 + 21) |x + 2| (since |x + 2| < 1)

= 25 |x + 2|
[Goal #2 accomplished: we have a constant B times |x a|]
[Now go back to specifying at the beginning of the proof by filling in
with: = min{1, /B}. This means that is the smaller of 1 and /B, and in
particular 1 and /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 and it is defined

for all complex numbers. Thus 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|
= |4(x + 2 2) 13| |x + 2| (adding a clever zero)
= |4(x + 2) 21| |x + 2|
(|4(x + 2)| + 21) |x + 2| (by the triangle inequality |a b| |a| + |b|)
< (4 1 + 21) |x + 2| (since |x + 2| < 1)
= 25 |x + 2|
< 25
25 /25
= .
We repeat the same type of discovery work on the next example, with fewer 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 4x 2
|4/x2 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)
= 4 |1 + x| (by rewriting 1 x = 2 (x + 1))

|x|2
2 + |x + 1|
4 |1 + x| (by the triangle inequality)
x2
2.4
4 |1 + x| (since 0.4)
|x|2
9.6
= |1 + x| (by rewriting x = x + 1 1)
|x + 1 1|2
9.6
|1 + x| (by the reverse triangle inequality
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
Example 4.1.5 lim = 2.
x2i 8xi
3
Proof. The domain of 4x +x
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}. [It is so obvious that this
minimum of two positive numbers is positive that we skip the assertion Thus
is a positive real number. Do not omit the assertion or the checking of
its veracity for more complicated specifications of .] 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

8x i (2) = 8x i + 2
3
4x + x + 16x 2i
=
8x i
3
4x + 17x 2i
=

8x i

(4x2 + 8ix + 1)(x 2i)
=

8x i

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

8x i
|4x2 | + |8ix| + 1
|x 2i| (by the triangle inequality)
|8x i|
4|(x 2i + 2i)|2 + 8|x 2i + 2i| + 1

|x 2i|
|8(x 2i) + 15i|
4(|x 2i| + 2)2 + 8(|x 2i| + 2) + 1
|x 2i|
8|x 2i| + 15
(by the triangle and reverse triangle inequalities)
2
4(1 + 2) + 8(1 + 2) + 1
|x 2i| (since |x 2i| < 1)
8 + 15
61
= |x 2i|
7
< 9
9/9
= .
The next example is of a 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 the 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 does not depend on x.
By the definition of limits, is supposed to be a positive real number, not a function of x.
(See also Exercise 4.1.9.)
Remark 4.1.8 In all cases above, Goal #1 is to factor x a from f (x) L. In similar
examples, if x a is not a factor, check your limit or algebra for any mistakes. In the next

example, x a is not a factor, but x a is.

Example 4.1.9 lim 2x 6 = 0.
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
p
= 2 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 4.1.10 Let A R, a R, L C, and f : A C a function. Suppose that

a is a limit point of {x A : x > a} (resp. of {x A : x < a}). We say that the right-
sided (resp. left-sided) limit of f (x) as x approaches a is L if for every real number
> 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) = L
(resp. limxa f (x) = L).

With this, Example 4.1.9 can be phrased as limx3+ 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 limx3 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 4.1.11 Let f : R R be given by f (x) = Then
x 2, if x 1.
limx1+ f (x) = 5, limx1 f (x) = 1.
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)
= |x 1 + 2| (by adding a clever 0)
(|x 1| + 2) (by the triangle inequality)
< (1 + 2) (since 0 < x 1 < 1)
3/3
= .
This proves that limx1+ f (x) = 5.
Set = . Let x satisfy 0 < 1 x < . Then
|f (x) (1)| = |x 2 + 1| (since x < 1)

= |x 1|
<
= .
This proves that limx1 f (x) = 3.

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
that 2 is a . Let > 0. Set = . Let x satisfy
0 < |x 2| < . Then

|(x2 3x) (2)| = x2 3x + 2
= |x 1| |x 2| (because )
= |x 2 + 1| |x 2| (by )

|x 2| + 1 |x 2| (by )
< (3 + 1) |x 2| (because )
< 4 (because )

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

also with the following proof: Since |x 2| < 3, then 3 < x 2 < 3, so that
2 < x 1 < 4, which means that |x 1| < 4. Say why this argument does not
work if the domain is C.
4.1.2 Determine the following limits and prove them with the epsilon-delta proofs.
i) lim (x3 4).
x1
1
ii) lim .
x2 x
iii) lim xx4
2 +2 .
x3
iv) lim x + 5.
x4
x2 9
v) lim .
x3 x3
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; x3 4x2 , if x = 5;
f (x) = g(x) =
b, if x = 5, b, 6 5.
if x =
Prove that the limit of f (x) as x approaches 5 is independent of b, but that the limit of
g(x) as x approaches 5 depends on b.
4.1.5 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.
x
4.1.8 Let f : R R be given by f (x) = |x| . Prove that limx0+ f (x) = 1 and that
limx0 f (x) = 1.
4.1.9 Below is 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 a 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.]
Thus if limxa f (x) 6= L, then either a is not a limit point of the domain of f or else
it is not true that for all real numbers > 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 314), 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 rephras-
ing:]
there exists x in the domain of f such that 0 < |x a| < and

|f (x) L| < .
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
Example 4.2.2 The limit of |x| as x approaches 0 does not exist. In other words, for all
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
Proof. The domain of the function that takes x to |x| is the set of all non-zero complex
numbers, so that 0 is a limit point of the domain. [Thus if the limit is not L, then
it must be the epsilon-delta condition that fails.] Set = 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
so that | |x| L| 1 = , and if Re(L) < 0, then

x /2
Re L = Re L = 1 Re(L) > 1,
|x| |/2|
x
so that again | |x| L| > 1 = . This proves the claim of the example.
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
not equal to 2), the size of x2 gets larger and larger.
Proof. Set = 1. Let > 0 be an arbitrary positive number. Set = min{, 1/(|L| + 1)}.
Let x = 2 + /2. Then 0 < |x 2| < , and

i 2i
L = L
x 2

2i
|L| (by the 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.
1 2 3
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 14 , 2 , and if L < 32 , set
x = 2 min{ 14 , 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 + 41 ,

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 + 34 , whence |f (x) L| 41 = . 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.
x3
4.2.3 Prove that limx3 x2 9 does not exist.
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: The reason is different from
the reasons in all other examples.)
4.3 Limit theorems
While epsilon-delta proofs are a reliable method for proving limits, they do not help
in deciding what a limit may be. In this section we prove theorems that will efficiently
establish the limits for many functions. The proofs of these theorems require the epsilon-
delta machinery as this is the definition of limits, but subsequent applications of these
theorems allow us to omit the time-consuming epsilon-delta proofs.
Theorem 4.3.1 If a limit exists, it is unique.
Proof. Suppose that both L1 and L2 are limits of f (x) as x approaches a.

First I give a FALSE proof: by assumption, L1 = limxa f (x) = L2 , so that by
transitivity of equality, L1 = L2 .
What makes this proof false is that the equal sign in limxa f (x) = L signifies, at this
point, not numerical equality, but that a is a limit point of the domain of f and that for
every > 0 there exists > 0 such that for all x in the domain of f , if 0 < |x a| < then
|f (x) L| < . Thus, we need a proof that will use these definitions instead.
Real proof: Suppose that L1 and L2 are both limits. Let be an arbitrary positive
number. Then /2 is also positive, and by the definition of limits, for each i = 1, 2, there
exists i > 0 such that for all x in the domain of f , if 0 < |xa| < i , then |f (x)Li| < /2.
Set = min{1 , 2 }. Then is a positive real number. Let x in the domain of f satisfy
0 < |x a| < . Since a is a limit point of the domain, such x exists. Then
|L1 L2 | = |L1 f (x) + f (x) L2 | (by adding a clever 0)

|L1 f (x)| + |f (x) L2 | (by the triangle inequality)
< 2/2 (since 1 , 2 )
= ,
which says that |L1 L2 | < . Since was arbitrary, an application of Theorem 2.10.4
gives that |L1 L2 | = 0, so that L1 L2 = 0, and hence that L1 = L2 .
(If in the definition of limxa f (x) we did not require that a be a limit point of the
domain of f , then x as in the proof above need not exist and we would not be able to
guarantee that limits are unique; in fact, any complex number L would be a limit.)
Theorem 4.3.2 Let a be a limit point of the domain of a function f with 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.
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) (Scalar rule) lim cf (x) = c lim f (x).
xa xa
(4) (Sum/difference rule) lim (f (x) g(x)) = lim f (x) lim g(x).
xa xa xa
(5) (Product rule) lim (f (x) g(x)) = lim f (x) lim g(x).
xa xa xa
f (x) lim f (x)
xa
(6) (Quotient rule) If lim g(x) 6= 0, then lim = .
xa xa g(x) lim g(x)
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 < |x a| < 1 , then |f (x) L| < /2. Similarly, since K = limxa g(x), there exists
2 > 0 such that for all x A, if 0 < |xa| < 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 the 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
the triangle inequality,
|f (x)| = |f (x) L + L| |f (x) L| + |L| < 1 + |L|,
and so
|f (x) g(x) L K| = |f (x) g(x) f (x)K + f (x)K L K| (by adding a clever zero)
|f (x) g(x) f (x)K| + |f (x)K L K| (by the triangle inequality)
= |f (x)| |g(x) K| + |f (x) L| |K| (by factoring)

< (1 + |L|) + |K| (since 1 , 2 )
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) L f (x)K Lg(x)

g(x) K = Kg(x) (by algebra)

f (x)K LK + LK Lg(x)
= (by adding a clever zero)

Kg(x)

f (x)K LK LK Lg(x)
+
Kg(x) (by the triangle inequality)
Kg(x)

f (x) L |L| K g(x)
= + (by factoring)
g(x) |K| g(x)
|K| 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 L (by induction assumption and by the product rule)
= Ln .
So the result holds for n, and we are done by mathematical induction.
Theorem 4.3.5 (Polynomial function rule for limits) Let f be a polynomial function.
Then for all complex (or real) a, 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 2.4.13, the domain of f is the set of all except finitely
many numbers, so that in particular a is a limit point of the domain. By the polynomial
function rule for limits, 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 func-

tion, 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. By the scalar
and sum rules in Theorem 4.3.3, and by the definition of real and imaginary parts,
lim f (x) = lim (Re f (x) + i Im f (x)) = lim Re f (x) + i lim Im f (x) = Re L + i Im L = L.
xa xa xa xa
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).
xL xa 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 .
Perhaps the hypotheses on g in the theorem above seem overly restrictive, and you
think that the limit of g(x) as x approaches L need not be g(L) but an arbitrary K?
Consider the following example which shows that limxa g(f (x)) then need not be K. Let
3, if x 6= 5;
f (x) = 5 and g(x) = . Then
7, otherwise.
lim f (x) = 5, lim g(x) = 3, and lim g(f (x)) = 7.
xa x5 xa
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 the triangle inequality)
> /2 /2 = .
Since this is true for all > 0, by Theorem 2.10.4, K L, as desired.
Theorem 4.3.11 (The squeeze theorem) Suppose that f, g, h : A R, that a is a

limit point of A, and that for all x A \ {a}, f (x) g(x) h(x). If 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
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 xa
4.3.3 Prove that lim xn = an 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

4.3.8 Prove that lim |x| and lim 1 |x| do not exist, but that lim |x| + 1 |x|
x0 x0 x0
does exist. Does this contradict the sum rule for limits? Justify.
|x|
4.3.9 Prove that lim x
and lim do not exist, but that lim x
|x| does exist. Does
x0 |x| x0 x x0 |x| x
this contradict the product rule for limits? Justify.
4.3.10 The following information is known about functions f and g:
a f (a) lim f (x) g(a) lim g(x)

xa xa
0 1 2 6 4
1 1 0 5 5
2 not defined 3 6 6
3 4 4 not defined 5
4 5 1 4 7
For each part below provide the limit if there is enough information, and justify. (Careful,
the answers to the last two parts are different.)
i) lim (f (x) g(x)).
x0
ii) lim (f (x) + g(x)).
x4
iii) lim (f (x) g(x)).
x3
f (x)
iv) lim .
x2 g(x)
v) lim g(x) .
x4 f (x)
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 if they both exist, then the two limits are equal.
4.3.13 Let f, g : R R be defined by

x, if x Q; 1, if x 6= 0;
f (x) = g(x) =
0, otherwise, 0, otherwise.
i) Prove that lim f (x) = 0.
x0
ii) Prove that lim g(x) = 1.
x0
iii) Prove that lim (g f )(x) does not exist.
x0
iv) Let a, L, M R. Construct functions f, g : R R such that limxa f (x) = L,
limxL g(x) = M , but limxa (g f )(x) 6= M . Comment on Theorem 4.3.9.
4.4 Infinite limits (for real-valued functions)
When the codomain of a function is a subset of an ordered field such as R, the values
of a function may grow larger and larger with no upper bound, or more and more negative
with no lower bound. In that case we may want to declare limit to be or . Naturally
both the definition and how we operate with infinite limits requires different handling.
Definition 4.4.1 Let A C, f : A R a function, and a C. Suppose that a is a limit

point of A (and not necessarily in A).
We say that the limit of f (x) as x approaches a is if for every real number
M > 0 there exists a real number > 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.)
1
Example 4.4.2 limx0 |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
> (because 0 < |x| < )
|x|2 2
= M.
x+2
Example 4.4.3 limx5+ 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
x+2 5+2
2
> 2 (because x > 5 and x2 25 > 0)
x 25 x 25
7 1
= (by algebra)
(x + 5) x 5
7 1
> (because 0 < x 5 < 1, so 0 < x + 5 < 11, and x 5 > 0)
11 x 5
7 1
> (because 0 < x 5 < )
11
7 1 7 7
7 (because 11M , so 1/ 1/( 11M ))
11 11M
= M.
1
Example 4.4.4 limx0+ x = .
1
Proof. Let M > 0. Set = 1/M . Then for all x with 0 < x 0 < , x > 1/ = M .
1
Example 4.4.5 limx0 x = .
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
but we do not have enough information to determine limxa (g(x) + h(x)) or even
whether it exists.
(3) (Product rule)

, if L > 0;
lim (f (x) g(x)) =
xa , if L < 0;

, if L > 0;
lim (f (x) h(x)) =
xa , if L < 0;
lim (g(x) h(x)) = .
xa
We do not have enough information to determine the existence (or value) of

limxa (f (x) g(x)) and of limxa (f (x) h(x)) in case L = 0.
(4) (Quotient rule)

f (x)
lim = 0,
xa g(x)
f (x)
lim = 0,
xa h(x)

g(x) , if L > 0;
lim =
xa f (x) , if L < 0;

h(x) , if L > 0;
lim =
xa f (x) , if L < 0.
We do not have enough information to determine the existence (or value) of
limxa (g(x)
f (x)) and of limxa (h(x)
f (x)) in case L = 0. We also do not have enough in-
limxa (g(x)
formation to determine the existence (or value) of h(x)) .
Example 4.4.8 Define g, h1 , h2 , h3 , h4 : R \ {0} R by g(x) = 1/x2 , h1 (x) = 1/x2 ,

h2 (x) = 17 1/x2 , h3 (x) = 1/x2 1/x4 , h4 (x) = 1/x2 1/x3 . We have seen that
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 Give definitions for the following limits.
i) limxa+ f (x) = .
ii) limxa f (x) = .
iii) limxa f (x) = .
4.4.2 Prove that limx0+ 1 = .
x
1
4.4.3 Prove that limx0 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.
x5 16x4
Example 4.5.3 limx x5 +4x2 = 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
5
(because x5 + 4x2 x5 > 0)
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
2 +2 = 3/2.
3x3 4x+1
ii) limx 2x2 +2 = .
2
iii) limx 3x2x4x+1
3 +2 = 0.

9x2 +4
iv) limx x+2 = 3.

2 +4
v) limx 9x x+2 = 3.
4.5.5 For which rational functions f does limx f (x) = 0 hold? Justify.
4.5.6 For which rational functions f does limx f (x) exist and is not ? Justify.
4.5.7 Let f : (a, ) C and L C. Prove that limx f (x) = L if and only if
limx Re f (x) = Re L and limx Im f (x) = Im L.
4.5.8 Let f : (, b) C and L C. Prove that limx f (x) = L if and only if
limx Re f (x) = Re L and limx Im f (x) = Im L.
Chapter 5: Continuity
Continuous functions from an interval in R to R are the ones that we can graph
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 5.3),
existence of radical functions (Section 5.4), and the new notion of uniform continuity
(Section 5.5).
5.1 Continuous functions
Definition 5.1.1 A function f : A B is continuous at a A if for all real numbers

> 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
|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.
The next theorem is an easy application of Theorems 4.3.3, 4.3.4, 4.3.5, 4.3.6, 4.3.7,
4.3.8, and 4.3.9, at least when the points in question are limit points of the domain. When
the points in question are not limit points of the domain, the results below need somewhat
different but still easy proofs.
Theorem 5.1.3 We have:

(1) (Constant rule) Constant functions are continuous (at all real/complex numbers).
(2) (Linear rule) The function f (x) = x is continuous (at all real/complex numbers).
(3) (Absolute value rule) The function f (x) = |x| is continuous (at all real/complex
numbers).
(4) (Polynomial function rule) Polynomial functions are continuous (at all real/complex
numbers).
(5) (Rational function rule) Rational functions are continuous (at all points in the
domain).
(6) (Real and imaginary parts) f is continuous at a if and only if Re f and Im f are
continuous at a.
(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.
Suppose that f : A C is continuous at a A. Then
(8) (Scalar rule) For any c C, cf is continuous at a.
(9) (Power rule) For any positive integer n, the function that takes x to (f (x))n is
continuous at a.
(10) (Restriction rule) For any subset B of A that contains a and in which a is a limit
point, the restriction of f to B is continuous at a. In other words, if g : B C is
defined as g(x) = f (x) for all x B, then g is continuous at a. (Such g is called
the restriction of f to B. By this rule and by (1), the constant functions with
domain a subinterval in R are continuous.)
Now suppose that f, g : A C, and that f and g are continuous at a A. Then
(11) (Sum/difference rule) f + g and f g are continuous at a.
(12) (Product rule) f g is continuous at a.
(13) (Quotient rule) If g(a) 6= 0, then f /g is continuous at a.
The theorem covers many continuous functions, but the following function for example
has to be verified differently.
146 Chapter 5: Continuity
Example 5.1.4 Let f : C C be defined by

x2 4, if Re x > 1;
f (x) =
3x3 , if Re x 1.
Then f is continuous at 1, because by the polynomial rules,
lim f (x) = lim (x2 4) = 12 4 = 3,

x1,Re x>1 x1,Re x>1
lim f (x) = lim 3x3 = 3 13 = 3,

x1,Re x1 x1,Re x1
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 (x2 4) = (1 + i)2 4 = 4 + 2i,

x1+i,Re x>1 x1+i,Re x>1
lim f (x) = lim 3x3 = 3 (1 + i)3 = 6 6i.

x1+i,Re x1 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
g : R R defined by
x2 4, if x > 1;
g(x) =
3x3 , if x 1.
As before, g is continuous at 1. Explicitly, by the polynomial rules,
lim+ g(x) = lim+ (x2 4) = 12 4 = 3,

x1 x1
lim g(x) = lim 3x3 = 3 13 = 3,
x1 x1
3
so that limx1 g(x) = 3 = 3 1 = g(1). It is worth noting that precisely because of
continuity this function can be expressed in the following ways as well:

x2 4, if x 1; x2 4, if x 1;
g(x) = =
3x3 , if x < 1 3x3 , if x 1.
By the polynomial rules, this function is continuous at all other real numbers as well.

x + 1, if x < 0;
Example 5.1.6 Let f : (1, 0) (0, 1] R be defined by f (x) =
x 1, if x > 0.
y
This function is continuous at all a < 0 because there f is the polynomial/linear function
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.

5.1.1 Let c C.
i) Prove that the function f : C C given by f (x) = c + x is continuous.
ii) Prove that the function g : C C given by g(x) = cx is continuous.
5.1.2 Give an proof to show that the functions Re, Im : C R are continuous.
5.1.3 Give details of the proofs of Theorem 5.1.3.
5.1.4 The following information is known about functions f and g:
c f (c) lim f (x) g(c) lim g(x)
xc xc
0 1 2 3 4
1 1 0 5 5
2 not defined 6 6
3 4 4 not defined
4 2 3 3 2
i) At which points is f continuous?

ii) At which points is |g| continuous?
iii) At which points is f g continuous?
5.1.5 Define a function f : C \ {0} R such that for all non-zero x C, f (x) is the angle
of x counterclockwise from the positive real axis. Argue that f is not continuous.
5.1.6 Assume that the trigonometric functions sine and cosine are continuous everywhere.
(We will prove this in Theorem 9.9.4.) Define a function g : C \ {0} R such that for all
non-zero x C, g(x) is the sine (respectively cosine) of the angle of x counterclockwise from
the positive real axis. Prove that f is continuous. Compare with the previous exercise.
5.1.7 Assume that the trigonometric function sine is continuous everywhere. (We will
prove this in Theorem 9.9.4.) Define a function f : R R as

sin(1/x), if x 6= 0;
f (x) =
0, if x = 0.
Prove that f is continuous at all points, including at 0. (You need a separate proof for
continuity at 0.)
5.1.8 Assume that the trigonometric function sine is continuous everywhere. (We will
prove this in Theorem 9.9.4.) Define a function f : R R as

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

x, if x is rational;
5.1.9 Let f : R R be given by f (x) =
0, otherwise.
i) Prove that f is continuous at 0.
ii) Prove that f is not continuous anywhere else.
The exercises below modify the definition of continuity to get very different types of func-
tions. The moral is that the order of quantifiers and implications is very important!
5.1.10 (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.
5.1.11 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.
5.1.12 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.
5.2 Topology and the Extreme value theorem
Topology and continuity go hand in hand (see Exercise 5.2.3), but not in the obvious
way, as the next two examples show.
(1) If f : A B is continuous and A is open, it need not be true that the range of f
is open in B, even if A is bounded. For example, let f : R R or f : (1, 1) R
be the squaring function. Certainly f is continuous and R is an open set, but the
image of f is [0, ) or [0, 1), which is not open.
(2) If f : A B is continuous and A is closed, it need not be true that the range of
1
f is closed in B. For example, let f : R R be given by f (x) = 1+x 2 . This f is
continuous as it is a rational function defined on all of R. The domain A = R is

a closed set (and open), but the image of f is (0, 1], which is not closed (and not
open).
However, if f is continuous and its domain A is closed and in addition bounded, then
it is true that the image of f is closed. This fact is proved next.
Theorem 5.2.1 Let A be a closed and bounded subset of C or R, and let f : A C be

continuous. Then the range of f is closed and bounded.
Proof. We first prove that the range is closed. Let b be a limit point of the range. We want
to prove that b is in the range. Since b is a limit point, by definition for every positive real
number r, B(b, r) contains an element of the range (even an element of the range that is
different from b). In particular, for every positive integer m there exists xm A such that
f (xm ) B(b, 1/m). If for some m we have f (xm ) = b, then we are done, so we may assume
that for all m, f (xm ) 6= b. Thus there are infinitely many xm . As in Construction 3.4.1,
we can construct nested quarter subrectangles Rn that contain infinitely many xm . There
is a unique complex number c that is contained in all the Rn . By construction, c is the
limit point of the set of the xm , hence of A. As A is closed, c A. But f is continuous
at c, so that for all > 0 there exists > 0 such that for all x A, if |x c| < then
|f (x) f (c)| < /2. In particular, for infinitely many large m, |xm c| < , so that for
these same m, |f (xm ) f (c)| < /2. But for all large m we also have |f (xm ) b| < /2,
so that by the triangle inequality, for |f (c) b| < . Since this is true for all positive , it
follows that f (c) = b. Thus any limit point of the range is in the range, so that the range
is closed.
Next we prove that the range is bounded. If not, then for every positive integer m there
exists xm A such that |f (xm )| > m. Again we use Construction 3.4.1, and this time
we construct nested quarter subrectangles Rn that contain infinitely many of these xm .
As before, there is a unique complex number c that is contained in all the Rn and in A,
and as before (with = 2), we get that for infinitely many m, |f (xm ) f (c)| < 1. But
|f (xm)| > m, so that by the reverse triangle inequality,
|f (c)| |f (xm) (f (xm ) f (c))| |f (xm )| |f (xm ) f (c)| > m 1
for infinitely many positive integers m. Since |f (c)| is a fixed real number, it cannot be
larger than all positive integers. Thus we get a contradiction to the assumption that the
range is not bounded, which means that the range must be bounded.
Theorem 5.2.2 (Extreme Value Theorem) Let A be a closed and bounded subset
of C, and let f : A R be a continuous function. Then there exist l, u A such that for
all x A,
f (l) f (x) f (u).
In other words, f achieves its maximum value at u and its minimum value at l.
Proof. By Theorem 5.2.1, the range {f (x) : x A} of f is a closed and bounded subset
of R, so that its infimum L and supremum U are real numbers which are by closedness in
the range. Thus there exist u, l A such that L = f (l) and U = f (u).
Example 5.2.3 The function f : [2, 2] R given by f (x) = x2 6x + 5 achieves a

minimum and maximum. We can rewrite the function as f (x) = (x 3)2 4, from which
it is obvious that the minimum of the function is achieved at 3 but wait a minute, this
function is not defined at 3 and hence cannot achieve a minimum at 3. Here is a correction:
the quadratic function (x 3)2 4 achieves its minimum at 3 and is decreasing on (, 3),
so that f achieves its minimum on [2, 2] at 2 and its maximum at 2.
Exercises 5.2.1 and 5.2.2 below give examples of continuous invertible functions whose
inverses are not continuous. The next theorem contains instead some positive results of
this flavor.
Theorem 5.2.4 Let F = R or F = C. Let A, B be subsets of F , let A be closed and

bounded, and let f : A B be continuous and invertible. Then f 1 is continuous.
Proof. We need to prove that f 1 is continuous at every b B. Let > 0. Set a = f 1 (b).
The set B(a, ) is open, so its complement is closed. Therefore C = A \ B(a, ) = A
(F \ B(a, )) is closed. As C A, then C is also bounded. Thus by Theorem 5.2.1,
{f (x) : x C} is a closed and bounded subset of F . By injectivity of f , this set does not
contain b = f (a). The complement of this set, {f (x) : x 6 C} = {f (x) : x A B(a, )},
contains b and is open, so that there exists > 0 such that B(b, ) {f (x) : x AB(a, ).
This means that for all y B, if |y b| < , then y = f (x) for some x A B(a, ), which
means that f 1 (x) B(a, ). This proves that f 1 is continuous at b.

5.2.1 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.
5.2.2 Define f : {x C : |x| < 1 or |x| 2} C by

x, if |x| < 1;
f (x) = 1
2
x, otherwise.
Prove that f is continuous and invertible but that f 1 is not continuous.
*5.2.3 Let F be R or C, and let A and B be subsets of F . Prove that f : A B is
continuous if and only if for every open subset U of F there exists an open subset V of F
such that the set {x A : f (x) U } = V A.
5.2.4 Let f : (0, 1) R be defined by f (x) = x1 . Prove that f is continuous and that f
has neither a minimum nor a maximum. Explain why this does not contradict the Extreme
Value Theorem (Theorem 5.2.2).
5.2.5 (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 , if non-empty, has a minimum.
vii) Suppose that for some x [a, b], f (x) > f (u).
Prove that x > u and x 6 L.
Prove that there exists x1 (u, x) such that f (x1 ) > f (x).
Prove that there exist x1 (u, x), x2 (u, x1 ), x3 (u, x2 ), . . ., xn
(u, 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 Intermediate Value Theorem
In this section, all functions are real-valued. The reason is that we can only make
comparisons in an ordered field, and R is an ordered field (Axiom 2.7.9), and C is not
(Exercise 3.1.6).
Theorem 5.3.1 (Intermediate Value Theorem) Let a, b R with b > a, and let
f : [a, b] R be continuous. Let k be a real number strictly between f (a) and f (b). Then
there exists c (a, b) such that f (c) = k.
Here is a picture that illustrates the Intermediate Value Theorem: for value k on the y-
axis between f (a) and f (b) there happen to be two c for which f (c) = k. The Intermediate
Value Theorem guarantees that one such c exists but does not say how many such c exist.
f (b) b
a c1 c2 b x
b
f (a)
Proof of Theorem 5.3.1: Set a0 = a and b0 = b. Apply Construction 3.4.1 of halving

intervals, with the property P of intervals being that k is strictly between the values of f
at the two endpoints. If for any n > 0, f ( an1 +b 2
n1
) = k, then the theorem is proved and
we can immediately stop the proof (and the construction).
Note that if k is not f ( an1 +b
2
n1
) 6= k, then since k is between f (an1 ) and f (bn1 ),
necessarily k is between f (an1 ) and f ( an1 +b 2
n1
) or else between f ( an1 +b
2
n1
) and
f (bn1 ). Choose that half [an , bn] of [an1 , bn1 ] which says that k is strictly between
Section 5.3: Intermediate Value Theorem 153
f (an ) and f (bn ). Thus for all n, k is strictly between f (an ) and f (bn ). By construction
c = sup{a1 , a2 , a3 , . . .} = inf{b1 , b2 , b3 , . . .} is in [a, b]. So c is in the domain of f .
We will prove that for all > 0, |f (c) k| < . 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 2.7.20, there exists a positive integer n such that
1/2n < /(b a). As an c bn , we have |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), and thus both |f (an ) k| and
|f (bn) k| must be less than 2/3. Therefore
|f (c) k| |f (c) f (an )| + |f (an ) k| < ,
and since this true for all > 0, it follows by Theorem 2.10.4 that f (c) = k.
An important application of this theorem is in the next section, introducing the radical
functions. (So far we have sporadically used the square roots only, relying on some facts
from high school that we have not yet proved. The next section remedies this gap.)
c2 2
Example 5.3.2 There exists a real number c such that c5 4 = c2 +2 .
2
Proof. Let f : R R be defined by f (x) = x5 4 xx2 2 +2 . This function is a rational
function and defined for all real numbers, so that by Theorem 5.1.3, f is continuous. Note
that f (0) = 3 < 0 < f (2), so that by Theorem 5.3.1 there exists c in (0, 2) such that
2
f (c) = 0. In other words, c5 4 = cc2 2
+2
.
Theorem 5.3.3 Let I be an interval in R, and let f : I R be continuous. Then the

image of f is an interval in R.
Proof. For any c and d in the image of f , by the Intermediate Value Theorem (The-
orem 5.3.1), any real number between c and d is in the image of f , which proves the
theorem.
However, if f : A B is continuous and injective and A is open, then the range of f
is open in B. We prove this first for A and B subsets of R, and then for subsets of C.
Compare the next theorem to Example 5.1.6.
Theorem 5.3.4 Let I be an interval in R, B a subset of R, and let f : I B be

continuous and invertible. Then f 1 is continuous and f, f 1 are either both strictly
increasing or both strictly decreasing. Furthermore, if I is open, so is B.
Proof. Let a f (b).
For now we assume that f (a) < f (b). With that we prove that f is an increasing
function, i.e., that for any x, y I, if x < y then f (x) < f (y). First suppose that
x < y < z are in I. By invertibility, f (x), f (y), f (z) are distinct. If f (x) is between f (y)
and f (z), then an application of the Intermediate Value Theorem gives c (y, z) such
that f (c) = f (x). But x < y < c, so x and c are distinct, and f (c) = f (x) contradicts
invertibility of f . So f (x) is not between f (y) and f (z), and similarly f (z) is not between
f (x) and f (y). Thus necessarily f (x) < f (y) < f (z) or f (x) > f (y) > f (z). By setting
x = a and z = b we get that f is increasing on [a, b], by setting y = a and z = b we get that
f is increasing on I (, a], and by setting x = a and y = b we get that f is increasing
on I [b, ). Thus f is increasing on I.
By Theorem 5.3.3 we know that B is an interval. By definition of inverses, f 1 is
strictly increasing.
If I is in addition open, let c B and a = f 1 (c). Since I is open and f is increasing,
there exist b1 , b2 A such that b1 < a < b2 . Thus f (b1 ) < f (a) = c < f (b2 ), and by
Theorem 5.3.3 we know that (f (b1), f (b2)) is an open subset of B that contains c. Thus B
is open.
Now we prove that f 1 is a continuous function at an arbitrary y be in the domain
of f 1 . Let > 0. Let c = f 1 (y). Let z (c, c + ) I. (Of course, if c is the right
endpoint of I, there is no such z.) Since f is increasing, y = f (c) < f (z) and we set
1 = f (z) y. Then for any x (y, y + 1 ) = (f (c), f (z)), by the Intermediate Value
Theorem x is in the range of f , i.e., x is in the domain of f 1 , and since f 1 is increasing,
c = f 1 (y) < f 1 (x) < f 1 (f (z)) = z < c + . Thus |f 1 (x) f 1 (y)| < . This
proves that for the given > 0 there exists 1 > 0 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.
Finally, suppose that f (a) > f (b) for some a < b in I. Set g = f . Then g is
invertible and continuous, g(a) < g(b), and by the work so far we know that g is a strictly
increasing function with a continuous strictly increasing inverse g 1 . Thus f = g is a
strictly decreasing function, and f 1 = g 1 is a continuous strictly decreasing function.
In addition, if I is open, so is B.

5.3.1 Let f (x) = x3 + 4, g(x) = x4 + x2 . Prove that there exists c [4, 4] such that
f (c) = g(c).
5.3.2 Assuming that cos is a continuous function, prove that there exists a real number x
such that cos x = x.
5.3.3 Let f : R Q be a continuous function. Prove that f is a constant function. Let
g : C Q be a continuous function. Prove that g is a constant function.
5.3.4 (Fixed point theorem) Let f : [a, b] [a, b] be a continuous function. Prove that
there exists c [a, b] such that f (c) = c.
5.3.5 Find real numbers a < b and a continuous function f : [a, b] [a, b] such that
f (c) 6= c for all c (a, b).
5.3.6 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 the triangle inequality)
|an |b |an |b2 |an |b3 |an |bn
+ + ++ (by the choice of b)
n n n n
|an |b
1 + b + b2 + + bn1
n
|an |b n1
b + bn1 + bn1 + + bn1 (because b 1)
n
0 1 2 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).
5.4 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.
Theorem 5.4.1 The range of f is R if n is odd and is R0 otherwise.
Proof. Certainly 0 = 0n is in the range. If a > 0, then
0n = 0 < a < a + 1 (a + 1)n .

The last inequality is by Exercise 2.7.1. Since exponentiation by n is a polynomial function,
it is continuous, and so by the Intermediate Value Theorem (Theorem 5.3.1), there exists
r (0, a + 1) such that a = r n . Thus every non-negative real number is in the range of f .
If a < 0 and n is odd, then similarly
(a 1)n a 1 < a < 0 = 0n ,
and the Intermediate Value Theorem guarantees that there exists r (a 1, 0) such that
a = r n . So for odd n all real numbers are in the range.
By Theorem 2.8.2 and Exercise 2.8.1, f is strictly increasing. Thus by Theorem 2.8.4,
f has an inverse function f 1 . By the power rule, f is continuous, so that by Theorem 5.3.4,
f 1 is strictly increasing and continuous.
Definition 5.4.2 The function f (f (x) = xn ) is called exponentiation by n, where n is

called the exponent, or power, and x the base. The inverse of exponentiation by n is

the nth radical function, and is written as n . For each a in the domain, n a is called
the nth root of a.

We just established that for any positive integer n, n is defined on R0 , or even

on R if n is odd. We also just established that n is a strictly increasing continuous
function. (Compare to Theorem 2.9.2 in which we established with quite a bit of labor

that the square root function 2 is defined on R+ .) By continuity we immediately get the
following:
Theorem 5.4.3 (Radical rule for limits) For any positive integer n and any a in the

domain of n ,

lim n x = n a.
xa
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 5.4.4 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
exponentiation by m with exponentiation by 1/n, in either order. Exponentiation by non-
negative m is continuous by the constant or power rule, exponentiation by negative m is
continuous by the quotient rule, and exponentiation by 1/n is continuous by Theorem 5.4.3.
Thus f is continuous by the composite rule.
If r > 0, then m, n > 0, and then f is the composition of two strictly increasing
functions, hence strictly increasing. If r < 0, then m < 0 and n > 0, so f is the compo-
sition of a strictly increasing and strictly decreasing function, hence strictly decreasing by
Exercise 2.8.7. Monotonicity of inverses follows from Theorem 5.3.4.
The last part is obvious.
We prove in Theorem 7.5.4 that exponentiation by arbitrary real numbers (not just
by rational numbers) is continuous.

5.4.1 Let n, m Q, and suppose that a and b are in the domain of exponentiation by n
and m. Prove:
i) an bn = (ab)n .
ii) (an )m = amn .
iii) an am = an+m .
iv) If a 6= 0, then an = 1/an .

5.4.2 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.
5.4.3 Determine the following limits, and justify all steps by invoking the relevant theo-
rems/rules: p
i) lim x2 3x + 4.
x2

x 2
ii) lim .
x2 x2 + 4
x2
iii) lim 2 .
x2 x 4

x 2
iv) lim .
x2 x2 4

5.4.4 Determine the domain of the function f given by f (x) = x2 .
5.4.5 Recall from Exercise 3.2.8 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
iii) Explain away the problematic claims 4 9 = (4)(9) = 36 = 6,

4 9 = 2i 3i = 6 from page 9.
iv) Let D be the set of all complex numbers that are not on the negative real axis.
Prove that we can define a continuous square root function f : D C.
(v)* Let 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.
5.4.6 (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 5.4.4.)
iii) Let > 0. Justify the following:
n+1
X n+1
X n+1
X
n+1 i i1
i1
( + 1) 1= ( + 1) ( + 1) = ( + 1) = (n + 1).
i=1 i=1 i=1
Use the Archimedean property of R (Theorem 2.7.12) 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.
5.4.7 (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
cr = , (c1 c2 )r = cr1 cr2 , cr1 +r2 = cr1 cr2 , cr1 r2 = (cr1 )r2 .
cr
iii) Prove that the function g : R R given by g(x) = cx is continuous. (This is easy.)
5.4.8 Here is an alternate proof of Theorem 5.4.3. Study the proof, and provide any

missing commentary. Let A be the domain of n .
i) Prove that an element of A is a limit point of A.
ii) Suppose that a = 0. Set = n . Let x A satisfy 0 < |x a| < . Since the 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 < .
5.5 Uniform continuity
Definition 5.5.1 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 5.5.2 Every uniformly continuous function is continuous.
The converse is false in general:
Example 5.5.3 Let f (x) = x2 , with domain C. Since f is a polynomial function, it is

continuous. Suppose that f is also uniformly continuous. Then in particular for = 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 2 d
|f (x) f (y)| = |x y||x + y| = + 1 = 1 + > 1,
2 d 2
which contradicts the assumption that f is uniformly continuous. Thus f is not uniformly
continuous.
(It is possible to show using elementary calculus that a uniformly continuous differen-
tiable real-valued function must have a bounded derivative, i.e., there exists a positive real
number c such that for all x in the domain of f , |f (x)| < c.)
The converse of Theorem 5.5.2 does hold with some extra assumptions.
Section 5.5: Uniform continuity 161
Theorem 5.5.4 Let A, B be closed and bounded subsets of R or C. Let f : A B be

continuous. Then f is uniformly continuous.
Proof. Let > 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. Note that A aA B(a, a ).
By Theorem 3.4.3 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 5.5.5 The continuous function is uniformly continuous.

Proof. We established in Section 5.4 that is continuous. Let > 0. We divide the
domain into two regions, one closed and bounded so we can invoke the theorem above, and

the other unbounded but where has a bounded derivative.
The first region is the closed and bounded interval [0, 2]. By the previous theorem

there exists 1 > 0 such that for all a, x [0, 2], if |x a| < 1 then | x a| < .
The second region is the unbounded interval [1, ). For a, x [1, ) with |x a| <
we have

x + a xa
| x a| = ( x a) =
x + a x + a
|x a|
(because x, a 1)
2
< .

Finally, set = min{1 , , 1}. Let a and x be in the domain of such that |xa| < .
Since |x a| < 1, necessarily either x, a [0, 2] or x, a [1, ). We have analyzed

both cases, and we conclude that | x a| < .

5.5.1 Let f : C C be given by f (x) = mx + l for some constants m, l. Prove that f is
uniformly continuous.
5.5.2 Which of the following functions are uniformly continuous? Justify your answers.
i) f : B(0, 1) C, f (x) = x2 .
ii) f : (0, 1] R, f (x) = 1/x.
iii) f : R R, f (x) = x21+1 .
x
iv) f : R \ {0} R, f (x) = |x| .
x
, if x 6= 0;
v) f : R R, f (x) = |x|
0, otherwise.
vi) f : R R, f (x) =
0, otherwise.
vii) Re, Im : C R.
viii) The absolute value function from C to R.
5.5.3 Suppose that f, g : A C are uniformly continuous and that c C.
i) Prove that cf and f g are uniformly continuous.
ii) Is f g uniformly continuous? Prove or give a counterexample.
5.5.4 Let 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.
5.5.5 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).
5.5.6 If the composition of two uniformly continuous functions uniformly continuous?
Prove or give a counterexample.
5.5.7 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.
Chapter 6: Differentiation
The geometric motivation for differentiation comes from lines tangent to a graph of
a function f at a point (a, f (a)). For example, on the graph below are two secant lines
through (a, f (a)):
f (a)
a x
It appears that the line through (a, f (a)) and (x, f (x)) is closer to the tangent line to the
graph of f at (a, f (a)) if x is closer to a. Intuitively, the slope of the tangent line is the
limit of the slopes of the secant lines.
6.1 Definition of derivatives
Definition 6.1.1 Let A C, and let a A be a limit point of A. A function f : A C

is differentiable at a if
f (x) f (a)
lim
xa xa
exists. In this case, we call this limit the derivative of f at a, and we use either Newtons

notation f (a) = (f (x))x=a or Leibnizs notation dfdx
(x)
x=a
.
A function is differentiable if it is differentiable at all points in its domain.
An alternative way of computing the derivatives is

df (x) f (a + h) f (a)
f (a) = = lim ,

dx x=a 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.)
164 Chapter 6: Differentiation
Example 6.1.2 Let f (x) = mx + l, where m and l are complex numbers. Then for any
a C, Then
f (x) f (a) (mx + l) (ma + l) m(x a)
f (a) = lim = lim = lim = m.
xa xa xa xa xa xa
Alternatively,
f (a + h) f (a) (m(a + h) + l) (ma + l) mh
f (a) = lim = lim = lim = m.
h0 h h0 h h0 h
Example 6.1.3 Let f (x) = x2 . Then

f (x) f (a)
f (a) = lim
xa xa
x2 a2
= lim
xa x a
(x a)(x + a)
= lim
xa xa
= lim (x + a)
xa
= 2a.
Alternatively,
f (a + h) f (a)
f (a) = lim
h0 h
(a + h) a2
2
= lim
h0 h
a + 2ah + h2 a2
2
= lim
h0 h
2ah + h2
= lim
h0 h
(2a + h)h
= lim
h0 h
= lim (2a + h)
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)
f (a) = lim
h0 h
1 1

= lim a+h a
h0 h
Section 6.1: Definition of derivatives 165
a a+h
a(a+h)
a(a+h)
= lim (common denominator in the fractions)
h0 h
aah
a(a+h)
= lim
h0 h
h
= lim
h0 a(a + h)h
1
= lim
h0 a(a + h)
1
= 2
a
by the quotient rule for limits.

Example 6.1.5 Let f (x) = x. The domain of f is R0 , and for all a > 0,
f (a + h) f (a)
f (a) = lim
h0 h

a+h a
= lim
h0 h

a+h a a+h+ a
= lim
h0 h a+h+ a
(a + h) a
= lim (since (x y)(x + y) = x2 y 2 )
h0 h( a + h + a)
h
= lim
h0 h( a + h + a)
1
= lim
h0 a+h+ a
1
=
2 a
by the linear, radical, composite, and quotient rules for limits. It is left to Exercise 6.1.4
to show that f is not differentiable at 0.
Example 6.1.6 Let f (x) = x3/2 . The domain of f is R0 , and for all a 0,
f (a + h) f (a)
f (a) = lim
h0 h
(a + h) a3/2
3/2
= lim
h0 h
(a + h) a3/2 (a + h)3/2 + a3/2
3/2
= 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 )
3a2 h + 3ah2 + h3
= lim
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
a 3/2 +a3/2 , if a > 0;
= h2
limh0 h3/2 , if a = 0;
3
= a1/2
2
by the linear, radical, composite, and quotient rules for limits. (Note that this f is differen-
tiable even at 0, whereas the square root function (previous example) is not differentiable
at 0.)
Note that in all these examples, f is a function from some subset of the domain of f
to a subset of C, and we can compute f at a number labeled x rather than a:
df (x) f (x + h) f (x) f (z) f (x)
f (x) = = lim = lim .
dx h0 h zx zx
The h-limit is perhaps preferable to the last limit, where it is z that varies and gets closer
and closer to x.
Example 6.1.7 The absolute value function is not differentiable at 0.
Proof. This function is not differentiable whether the domain is C or R. The reason is
that the limit of |h||0|
h as h goes to 0 does not exist. Namely, if h varies over positive real
numbers, this limit is 1, and if h varies over negative real numbers, the limit is 1, so that
the limit indeed does not exist, and hence the derivative does not exist.
This gives an example of a continuous function that is not differentiable. (Any con-
tinuous function with a jagged graph is not differentiable.)

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

x2 1, if x > 1;
f (x) =
x3 x, if x 1.
Prove that f is differentiable at 1.

x2 1, if x > 1;
f (x) =
x4 4x, if x 1.
Prove that f is continuous but not differentiable at 1.

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

x2 , if x > 1;
6.1.8 Determine if the following function is differentiable at 1: f (x) =
2x, if x 1.
1 2 1 2
x , if x 0; x + 3, if x 0;
6.1.9 Let f, g : R R be f (x) = 2 1 2 and g(x) = 2 1 2
2 x , if x < 0, 2x , if x < 0.
i) Prove that f is differentiable everywhere and that for all x R, f (x) = |x|.
ii) Prove that g is not differentiable at 0.
iii) Prove that g is differentiable at any non-zero real number x with g (x) = |x|.
6.2 Basic properties of derivatives
Theorem 6.2.1 If f is differentiable at a, then f is continuous at a.
Proof. By definition of differentiability, a is a limit point of the domain of f and a is in the

domain of f . Furthermore, since limh0 f (a+h)f h
(a)
exists, by the product rule for limits
f (a+h)f (a)
also limh0 (h h ) exists and equals 0 f (a) = 0. In other words, limh0 (f (a +
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.
Theorem 6.2.2 (Basic properties of derivatives)

(1) (Constant rule) Constant functions are differentiable (at all real/complex numbers)
and the derivative is 0 everywhere.
(2) (Linear rule) The function f (x) = x is differentiable (at all real/complex numbers)
and the derivative is 1 everywhere.
Let A be a subset of C and let a A be a limit point of A. Suppose that f, g : A C are

differentiable at a. Then
(3) (Scalar rule) For any c C, cf is differentiable at a and (cf ) (a) = cf (a).
(4) (Sum/difference rule) f g is differentiable at a and (f g) (a) = f (a) g (a).
(5) (Product rule) f g is differentiable at a and (f g) (a) = f (a)g(a) + f (a)g (a).
(6) (Quotient rule) If g(a) 6= 0, then f /g is differentiable at a and (f /g)(a) =
f (a)g(a)f (a)g (a)
(g(a))2
.
Proof. Parts (1) and (2) were already proved in part (1) of Example 6.1.2. (3) follows from
(cf )(a + h) (cf )(h) cf (a + h) cf (h) f (a + h) f (h)
lim = lim = c lim = cf (a),
h0 h h0 h h0 h
and (4) follows from the sum rule for limits and from
(f + g)(a + h) (f + g)(h) f (a + h) + g(a + h) f (h) g(h)
lim = lim
h0 h h0 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).
The following proves the product rule (5):

(f g)(a + h) (f g)(a) f (a + h) g(a + h) f (a) g(a)
lim = lim
h0 h h0 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),
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
to provide the commentary on the steps:

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

f (a + h) f (a) g(a + h) f (a) g(a + h) g(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 1.6.7:
(x + h)n xn
(xn ) = lim
h0 h
Pn
n k nk
k=0 k x h xn
= 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
X
n1
n k nk1
= lim x h
h0 k
k=0
n1
X
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
in the domain of f , if 0 < |h| < 1 then | f (a+h)f h
(a)
f (a)| < min{1, /(2|g (f (a))| + 2)}.
For all such h, by the triangle inequality, | f (a+h)f h
(a)
| < |f (a)| + 1. By assumption g is
differentiable at f (a), so that there exists 2 > 0 such that for all x in the domain of g, if
0 < |x f (a)| < 2 , then | g(x)g(f (a))
xf (a) g (f (a))| < /(2|f (a)| + 2). Since f is differentiable
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)) g (f (a)) 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
= .
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 such that a + h is in the domain of f ,
0 < |h| < , and f (a + h) = f (a). Then f (a) = limh0 f (a+h)f h
(a)
, so that in particular
when h varies over those infinitely 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)
g (f (a)) f (a) = 0 < .

h
This analyzes all the cases and finishes the proof of the theorem.
Theorem 6.2.6 (Real and imaginary parts) Let A R, and let a A be a limit
point of A. Let f : A C. Then f is differentiable at a if and only if Re f and Im f are
differentiable at a, and in that case, f = (Re f ) + i(Im f ) .
Proof. Since all h are necessarily real,

f (a + h) f (a) Re(f (a + h) f (a)) + i Im(f (a + h) f (a))
=
h h
f (a + h) f (a) f (a + h) f (a)
= Re + i Im ,
h h
and by the definition of limits of complex functions, the limit of the function on the left
exists if and only if the limits of its real and imaginary parts exist.
Compare this last theorem with Exercises 6.2.4 and 6.2.5.
Exercises 5.2.1 and 5.2.2 each give an example of a continuous invertible function
whose inverse is not continuous. It is also true that the inverse of a differentiable invertible
function need not be differentiable. For example, the cubing function on R (f (x) = x3 ) is
differentiable and invertible,

but its inverse, the cube root function, is not differentiable at
3
0+h 3 0
0, as the limit of h
= h1/3 does not exist as h goes to 0. Nevertheless, we can
connect differentiability of an invertible function to the derivative of the inverse, as shown
in the next theorem.
Theorem 6.2.7 (Derivatives of inverses) Let A, B C, let A be open, and let f :

A B be an invertible differentiable function whose derivative is never 0. Then for all
b B, f 1 is differentiable at b, and
1
(f 1 ) (b) = ,
f (f 1 (b))
or in other words, for all a A, f 1 is differentiable at f (a) and
1
(f 1 ) (f (a)) = .
f (a)
Proof. Let b B. Then b = f (a) for some a A. Since A is open, there exists r > 0 such
that B(a, 2r) A. Let C be any closed and bounded subset of A that contains B(a, r).
For example, C could be {x C : |x a| r}. Define g : C D = {f (x) : x C} as
g(x) = f (x). Then g is invertible and continuous. By Theorem 5.2.4, g 1 is continuous.
In particular, g 1 is continuous at b = f (a) = g(a).
Let > 0. Since g is differentiable at a, and g(a) = f (a) = b and g (a) = f (a), by
the limit
definition of derivatives, there exists > 0 such that for all x C, if |x a| <
g(x)b
then xa f (a) < . By continuity of g 1 , there exists > 0 such that for all

1
y D, if |y b| < then |g (y) a| = |g 1 (y) 1
g (b)| < . In particular, for all
g(g (y))b
1
such y, g1yb
(y)a f (a) = g 1 (y)a f (a) < . This says that for all h C, if
1

h
|h| < and b + h D, then g1 (b+h)a f (a) = g(g (b+h))b
g 1 (b+h)a f
(a) < . Thus
h
limh0 g 1 (b+h)a = f (a), and by the quotient rule for limits,
g 1 (b + h) g 1 (b) g 1 (b + h) a 1 1
lim = lim = = 1 .
h0 h h0 h f (a) f (f (b))
But this holds for every C (and g which is f restricted to C), so that in particular,
f 1 (b + h) f 1 (b) 1
(f 1 ) (b) = lim = 1 .
h0 h f (f (b))
It should be noted that if we know that f 1 is invertible, the rest of the proof of the
theorem above goes as follows. For all x B, (f f 1 )(x) = x, so that (f f 1 ) (x) = 1,
and by the chain rule, f (f 1 (x)) (f 1 ) (x) = 1. Then if f is never 0, we get that
1
(f 1 ) (x) = f (f 1 (x)) .
Example 6.2.8 Let f : [0, ) [0, ) be the function f (x) = x2 . We know that f
is differentiable at all points in the domain and that f (x) = 2x. By Example 6.1.5 and
Exercise 6.1.4, the inverse of f , namely the square root function, is differentiable at all
positive x, but not at 0. The theorem above applies to positive x (but not to x = 0):
1 1 1
( x) = (f 1 ) (x) = = = .
f (f 1 (x)) 2f 1 (x) 2 x

Theorem 6.2.9 Let n be a positive integer. Then for all non-zero x in the domain of n
,
1
( n x) = x1/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 1 1 1 1
( n x) = = n1
= (n1)/n
= x(n1)/n = x1/n1 .
f ( x)
n
n( x)
n
nx n n
Theorem 6.2.10 (Generalized power rule) Let r be an arbitrary rational number and
let f : R+ R+ be given by f (x) = xr . Then for all x, f (x) = 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
f (x) = m(x1/n)m1 x = x = rxm/n1/n+1/n1 = rxr1 .
n n
This proves the theorem for all rational r.
The theorem also holds for all real r. But to prove it for all real r one first needs to
define exponentiation by non-rational numbers. Such exponentiation was worked through
laboriously in Exercises 5.4.6 and 5.4.7, and if we were to continue that kind of labori-
ous treatment, the proof of the form of the derivative of such exponentiation would also
be laborious. So we postpone the definition of such exponentiation and the proof of its
derivative to Theorem 7.5.4, where with the help of integrals the definition and proofs write
themselves elegantly.

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 Theo-
rem 6.2.2.
6.2.3 Prove yet another form of the general power rule for derivatives: If f is dif-
ferentiable at a, then for every positive integer n, f n is differentiable at a, and (f n ) (a) =
n(f (a))n1f (a).
6.2.4 Prove that the functions Re, Im : C R are not differentiable at any a.
6.2.5 (Compare with Theorem 6.2.6.) Prove that the absolute value function on R is
differentiable at all non-zero a R. Prove that the absolute value function on C is not
differentiable at any non-zero a C. (Hint: Let h = (r 1)a for r near 1.)
6.2.6 The following information is known: c f (c) f (c) g(c) g (c)
0 1 2 6 4
1 1 0 5 3
2 2 3 6 6
3 4 2 3 5
4 0 1 4 7
For each of the following, either provide the derivative or argue that there is not enough
information. In any case, justify every answer.
i) (f g) (1) =
ii) (f g) (2) =

iii) fg (3) =
iv) (g f ) (4) =
6.2.7 A function f is differentiable on (2, 5) and f (3) = 4, f (3) = 1. Let g(x) = 3x.
For each of the statements below determine whether it is true, false, or if there is not
enough information. Explain your reasoning.
i) f is constant.
ii) The slope of the tangent line to the graph of f at 3 is 4.
iii) f is continuous on (2, 5).
iv) The derivative of (f g) at 1 is 3.
v) (f + g) (3) = 2.
x1
6.2.8 Let f (x) = 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.)
6.2.9 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 non-
negative to prove that (sin1 ) (x) = 1x
1
2
.
6.2.10 Make the requisite assumptions about the trigonometric functions to prove that
1
(cos1 ) (x) = 1x 2
.
6.2.11 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.
Suppose that f (c) > 0. By the definition of derivative, f (c) = limxc f (x)f
xc
(c)
.
Thus for all x very near c but larger than c, f (x)f xc
(c)
> 0, so that f (x) f (c) > 0, so
that f does not achieve its maximum at c. Also, for all x very near c but smaller than c,
f (x)f (c)
xc
> 0, so that f (x) f (c) < 0, so that f does not achieve its minimum at c. This
is a contradiction, so that f (c) cannot be positive. Similarly, f (c) cannot be negative.
Thus f (c) = 0.
Thus to find extreme values of a function, one only has to check if extreme values
occur at the endpoints of the domain, at points where the function is not continuous or
non-differentiable, or where it is differentiable and the derivative is 0. One should be aware
that just because any of the five conditions is satisfied, we need not have an extreme value
of the function. Here are some examples:
(1) The function f : [1, 1] R given by f (x) = x3 x has neither the maximum
nor the minimum at the endpoints.
x, if x > 0;
(2) Let f : [1, 1] R be given by f (x) = Then f is not continuous
1/2, if x 0.
at 0 but f does not have a minimum or maximum at 0.
x, if x > 0;
(3) Let f : [1, 1] R be given by f (x) = Then f is continuous
2x, if x 0.
and not differentiable at 0, yet f does not have a minimum or maximum at 0.
(4) Let f : [1, 1] R be given by f (x) = x3 . Then f is differentiable, f (0) = 0, but
f does not have a minimum or maximum at 0.
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 5.2.2).
Note that g (a) < 0 if and only if f (a) < k, which holds if and only if f (b) > k, which
in turn holds if and only if g (b) > 0. Similarly g (a) > 0 if and only if g (b) < 0. Thus
for both choices of c, c cannot be a or b, so that c (a, b). By Theorem 6.3.1, g (c) = 0.
Hence f (c) = k.
Theorem 6.3.3 (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 5.2.2) there exist l, u [a, b] such that f
achieves its minimum at l and its maximum at u. If f (l) = f (u), then the minimum value
of f is the same as the maximum value of f , so that f is a constant function, and so
f (c) = 0 for all c (a, b).
Thus we may assume that f (l) 6= f (u). Suppose in addition that f (l) 6= f (a). Then
a < l < b. For all x [a, b], f (x) f (l), so that in particular for all x (a, l), f (x)f
xl
(l)
0
and for all x (l, b), f (x)f
xl
(l)
0. Since f (l) = limxl f (x)f
xl
(l)
exists, it must be both
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.
For all x [a, b], f (x) f (u), so that in particular for all x (a, u), f (x)f xu
(u)
0 and
for all x (u, b), f (x)f
xu
(u)
0. Since f (u) = limxu f (x)f
xu
(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)
such that f (c) = f (b)f
ba
(a)
.
Here is an illustration of this theorem: the slope f (b)f

ba
(a)
of the line from (a, f (a)) to
(b, f (b)) also equals the slope of the tangent line to the graph at some c between a and b:
a b x
Proof. Let g : [a, b] R be defined by g(x) = f (x) f (b)f ba

(a)
(x a). By the sum and
scalar rules for continuity and differentiability, g is continuous on [a, b] and differentiable
on (a, b). Also, g(a) = f (a) and g(b) = f (b) f (b)fba
(a)
(b a) = f (b) (f (b) f (a))
= f (a) = g(a). Thus by Rolles theorem, there exists c (a, b) such that g (c) = 0. But
f (b) f (a)
g (x) = f (x) ,
ba
f (b)f (a) f (b)f (a)
so that 0 = g (c) = f (c) ba , whence f (c) = 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.
Proof. We only prove part (2). Let x, y [a, b] with x < y. By Theorem 6.3.4 there exists
c (x, y) such that f (c) = f (x)f
xy
(y)
. Since f (c) > 0 and x < y, necessarily f (x) < f (y).
Since x and y were arbitrary with x < y, then f is strictly increasing on [a, b].
Example 6.3.6 Assuming that sin (x) 1 for all x R, we have that sin(x) x for all
x 0.
Proof. Let f (x) = x sin(x). Then f is differentiable on R, and f (x) = 1 sin (x) 0.
By the previous theorem, f is non-decreasing, so that for all x 0, x sin(x) = f (x)
f (0) = 0, whence x sin(x).
Theorem 6.3.7 (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)).
f (c) f (b)f (a)
In particular, if g (c) 6= 0 and g(b) 6= g(a), this says that g (c) = g(b)g(a) .
Proof. Define h : [a, b] R by h(x) = f (x)(g(b) g(a)) g(x)(f (b) f (a)). Then h is con-
tinuous on [a, b] and differentiable on (a, b). Note that h(a) = f (a)(g(b)g(a))g(a)(f (b)
f (a)) = f (a)g(b) g(a)f (b) = h(b). Then by the Mean Value Theorem (Theorem 6.3.4)
there exists c (a, b) such that h (c) = 0, i.e., 0 = f (c)(g(b) g(a)) g (c)(f (b) f (a)).
Theorem 6.3.8 (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 (c) f (b) f (a)

= .
g (c) g(b) g(a)
Proof. By the Mean Value Theorem (Theorem 6.3.4) there exists c (a, b) such that
g (c) = g(b)g(a)
ba . By assumption, g (c) 6= 0, so that g(b) 6= g(a). The rest follows by
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 2.4.13.)
iii) Give an example of f with n = 2, where f has no roots but f has 1 root. Why
does this not contradict the first part?
6.3.3 Let f : [0, 1] C be given by f (x) = x3 + ix2 . Prove that there exists no c between
0 and 1 such that f (c) = f (1)f
10
(0)
. Does this contradict the Mean Value Theorem?
6.3.4 Let f : [a, b] C be continuous on [a, b] and differentiable on (a, b). Prove that
(a) (a)
there exist c, d (a, b) such that Re(f (c)) = Re f (b)f
ba and Im(f (d)) = Im f (b)f
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)
(4) lim = L.
xa+ g (x)
f (x)
Then lim+ = L.
xa g(x)
f (x)
Proof. Let > 0. Since lim = L, there exists > 0 such that for all x in the domain,
xa g (x)

if 0 < x a < then | fg (x)
(x)
L| < . Let x be one such element. By Theorem 6.3.8 there

exists a number c (a, x) such that fg (c) (c)
= fg(x)g(a)
(x)f (a)
= fg(x)
(x)
. But then 0 < c a < , so
that
f (x) f (c)

g(x) L = g (c) L < .
A very similar proof shows the other one-sided version of the LHopitals rule:
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)
(4) lim = L.
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 9.11.5.
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)
(3) lim = L.
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

Theorem 6.3.8 there exists c (a0 , x) such that fg (c) (c)
= fg(x)g(a
(x)f (a0 )
0)
, and so

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

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

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

<2 +
4 4
= .
x2 1
Example 6.4.4 Compute lim 3 .
x1 x 1
x2 1 (x1)(x+1)
Proof #1: By Example 1.5.4, x3 1 = (x1)(x2 +x+1) , so that by Exercise 4.3.11 and Theo-
2
x 1
rem 4.3.6, lim 3 = lim 2x+1 = 1+1
12 +1+1
= 23 .
x1 x 1 x1 x +x+1
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
3 2 x2 1
lim (x 1) = lim 3x = 3, so that by LHopitals rule (Theorem 6.4.3), lim 3 equals
x1 x1 x1 x 1
2/3.
More forms of LHopitals rule are in the exercises.

x2 64
6.4.1 Compute and justify lim 3
x+2
.
x8
2x
6.4.2 What is wrong with the following: Since limx2 2 = 2, by LHopitals rule we have
x2 +3
that limx2 2x3 = 2.
6.4.3 Let n be a positive integer.
xn 1
i) Use LHopitals rule to prove that limx1 x1 = n. We have proved this previ-
ously, say how.
xn an
ii) Repeat for limxa xa
= nan1 .
6.4.4 Explicitly state the assumptions about properties of trigonometric functions that
are necessary to prove lim sin(x)
x = 1 and lim 1cos(x)
x = 0.
x0 x0
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,
x x

and that lim fg (x)
(x)
= L.
x

i) Let > 0. Prove that there exists N > a such that for all x > N , | fg (x)
(x)
L| < .
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 ).
f (c)
iv) Prove that for all x > N there exists c (N , x) such that g (c)
=

f (x) 1f (N )/f (x)
g(x) 1g(N )/g(x)
.
f (x)
v) Prove that lim = L.
x g(x)
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
x x x
f (x)
g (x) = L. Prove that lim f (x) = L.
x g(x)
Many interesting exercises on LHopitals rule are in Section 9.11, 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) approx-
imates f at a best.
If f is differentiable near a and the derivative is continuous at a, then for all x near a but
not equal to a, f (x) = f (x) f (a) + f (a) = f (x)f
xa
(a)
(x a) + f (a)
= f (a)(x a) + f (a),
so that for x near a (and possibly equal to a), f is approximately the linear function
f (a)(x a) + f (a), i.e., f is approximated by its tangent line.
This game keeps going, but for this we need higher order derivatives:
Definition 6.5.1 Let f be differentiable. If f is differentiable, we write the derivative of

f as f , or as f (2) . If f (n1) is differentiable, we denote its derivative as f (n) . Using this
notation we also write f (1) = f and f (0) = f .
For example, if f (x) = xm , then for all n m, f (n) (x) = m(m 1)(m 2) (m
n + 1)xmn .
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 (a) f (n) (a)
Tn,f,a (x) = f (a)+f (a)(xa)+ (xa)2 + + (xa)n = (xa)k .
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,
f (k) (a) = 0 if k > d.
But then for n d,

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

n
X f (k) (t)
g(t) = (x t)k .
k!
k=0
Then g is differentiable on B(a, r), and

n
X n
X
f (k+1) (t) k kf (k) (t)
g (t) = (x t) (x t)k1
k! k!
k=0 k=0
n1
X f (k+1) (t) X f (k) (t) n
f (n+1) (t)
= (x t)n + (x t)k (x t)k1
n! k! (k 1)!
k=0 k=1
(n+1)
f (t)
= (x t)n .
n!
Note that g(x) = f (x), g(a) = Tn,f,a (x). By the Mean Value Theorem (Theorem 6.3.4),
there exists c strictly between a and x such that g (c)(x a) = g(x) g(a). In other words,
f (n+1) (c)
n! c)n (x a) = f (x) Tn,f,a (x), which proves the first formulation.
(x
By 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,
f (n+1) (d)
(n + 1)(x d)n (f (x) Tn,f,a (x)) = (x d)n (0 (x a)n+1 ).
n!
f (n+1) (d)
Since d is not equal to x, we get that (n + 1)(f (x) Tn,f,a (x)) = n! ((x a)n+1 ),
which proves the second formulation.
Theorem 6.5.6 (Taylors remainder theorem over C) Let A be a subset of C or

of R, let a A, and let r R+ such that B(a, r) A. Let f : A C have higher order
derivatives of orders 1, 2, . . . , n on B(a, r) with n 1. Then for every > 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 qx B(a, r), ifq|x a| < 1 ,

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

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

< +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
Xn
(n) n (k)
(f g) (a) = f (a)g (nk) (a).
k
k=0
6.5.3 (Generalized quotient rule.) Prove the following generalization of the product rule:
Suppose that f and g have derivatives of orders 1, . . . , n at a and that g(a) 6= 0. Find and
prove a formula for the nth derivative of the function f /g.

6.5.4 Compute the Taylor polynomial of f (x) = 1 + x of degree 5 centered at a = 0.
Justify your work.

6.5.5 Compute the Taylor polynomial of f (x) = 1 x of degree 10 centered at a = 0.
Justify your work.
6.5.6 Let n N0 be given. Justify all work below.
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) Compute Tn,f,0 (0.5) f (0.5).
iv) Compute n such that |Tn,f,0(0.5) f (0.5)| < 0.001.
v) Try to use Theorem 6.5.5 to determine n such that |Tn,f,0(0.5) f (0.5)| < 0.001.
Note that this usage is not fruitful.
vi) Use Theorem 6.5.5 to determine n such that |Tn,f,0 (0.4) f (0.4)| < 0.001.
6.5.7 Explicitly assume the necessary properties of trigonometric functions to compute
the Taylor polynomials of degree n of sin and cos centered at a = 0.
i) Estimate the error term with Theorem 6.5.5.
ii) In fact, calculators and computers use Taylor polynomials to compute values of
trigonometric and also exponential functions. Compute sin(1) and sin(100) to
within 0.01 of true value. What degree Taylor polynomial suffices for each? Use
Theorem 6.5.5.
6.5.8 Let f : C R be the absolute value function.

i) Prove that f has derivatives of all orders at all non-zero numbers.
ii) Compute T2,f,i.
6.5.9 Let f : R R be given by f (x) = |x|3 .
i) Prove that f is differentiable, and compute f .
ii) Prove that f is differentiable, and compute f .
iii) Prove that f is not differentiable.
6.5.10 Find a differentiable function f : R R such that f is not differentiable.
6.5.11 Find a function f : R R such that f has derivatives of orders 1, 2, 3, 4, but f (4)
is not differentiable.
Chapter 7: Integration
The basic motivation for integration is computing areas of regions bounded by graphs
of functions. In this chapter we develop the theory of integration for functions whose
domains are subsets of R. The first two sections handle only codomains in R, and at the
end of Section 7.4 we extend integration to functions with codomains in C. We do not
extend to domains being subsets of C as that would require multi-variable methods and
complex analysis, which are not the subject of this course.
7.1 Approximating areas
In this section, domains and codomains of all functions are subsets of R. Thus we can
draw the regions and build the geometric intuition together with the formalism.
Let f : [a, b] R. The basic aim is to compute the signed area of the region bounded
by the x-axis, the graph of y = f (x), and the lines x = a and x = b. By signed area we
mean that we add up the areas of the regions above the x-axis and subtract the areas of
the regions below the x-axis. Thus a signed area may be positive, negative or zero.
y = f (x)
y
a b x
In the plot above, there are many (eight) regions whose boundaries are some of the
listed curves, but only the shaded region (comprising of two of the eight regions in the
count) is bounded as a subset of the plane.
The simplest case of an area is 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:
188 Chapter 7: Integration
y = f (x)
y
a b x
It may be hard to decide how close the approximation is to the true value. But we can
approximate the region more systematically, by having heights of the rectangles be either
the least possible height or the largest possible height, as below:
a b a b
Then clearly the true area is larger than the sum of the areas of the darker rectangles
on the left and smaller than the sum of the areas of the darker rectangles on the right.
We establish some notation for all this.
Definition 7.1.1 A function f : A C is bounded if the range {f (x) : x A} of f is a

bounded subset of C.
Remark 7.1.2 If f is a bounded function with codomain R, then by the Least upper
bound theorem (Axiom 2.9.1), for any subset B of the domain, sup{f (x) : x B} and
inf{f (x) : x B} are real numbers.
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
n
X
L(f, P ) = inf{f (x) : x [xk1 , xk ]}(xk xk1 ).
k=1
The upper sum of f with respect to P is

n
X
U (f, P ) = sup{f (x) : x [xk1 , xk ]}(xk xk1 ).
k=1
By the previous remark, L(f, P ) and U (f, P ) are real numbers.
Clearly if f is a constant function f (x) = c for all x, then all lower and all upper sums
are c(b a), so that every lower and every upper sum equals c(b a). If instead f is a
non-constant function, then for every partition P there exists at least one subinterval of
P on which the supremum of the values of f is strictly bigger than the infimum of such
values, so that L(f, P ) < U (f, P ).
By the geometric set-up for all partitions P of [a, b],
L(f, P ) the signed area U (f, P ). (7.1.4)

In particular, if U (f, P ) L(f, P ) < , then either U (f, P ) or L(f, P ) serves as an approx-
imation 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, xk = 1 + k 3/n.
Note that xn = 1 + n 3/n = 4, as needed. Since f (x) = 12x3 + 2 is positive on [1, 4],
it follows that f is increasing on [1, 4]. Thus necessarily for each i, the infimum of all
values of f on the ith subinterval is achieved at the left endpoint, and the supremum at
the right endpoint. In symbols, this says that inf{f (x) : x [xk1 , xk ]} = f (xk1 ) and
sup{f (x) : x [xk1 , xk ]} = f (xk ). For example, with n = 1, L(f, P1 ) = f (1) 3 = 96 and
U (f, P1) = f (4) 3 = 3 (30 44 + 2 4) = 23064. Thus the true area is some number between
96 and 23064. Admittedly, this is not much information. A computer program produced
the following better numerical approximations for lower and upper sums with respect to
partitions Pn into n equal subintervals:
n L(f, Pn) U (f, Pn)

10 5061.2757 7358.0757
100 6038.727 6268.407
1000 6141.5217 6164.48967
10000 6151.8517 6154.148
100000 6152.885 6153.1148
1000000 6152.988 6153.011
Notice how the lower sums get larger and the upper sums get smaller as we take finer
partitions. We 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 [xk1 , xk ] = {y0 =

xk1 , y1 , . . . , ym1 , ym = xk }. By set inclusion, inf{f (x) : x [yj1 , yj ]} inf{f (x) : x
[xk1 , xk ]}, so that
m
X
L(f, {y0, y1 , . . . , ym1 , ym }) = inf{f (x) : x [yj1 , yj ]}(yj yj1 )
j=1
Xm
inf{f (x) : x [xk1 , xk ]}(yj yj1 )
j=1
m
X
= inf{f (x) : x [xk1 , xk ]} (yj yj1 )
j=1
= inf{f (x) : x [xk1 , xk ]}(xk xk1 ),
and hence that

Xn n
X
L(f, R) = L(f, R [xk1 , xk ]) inf{f (x) : x [xk1 , xk ]}(xk xk1 ) = L(f, P ).
k=1 k=1
The proof for upper sums is similar.


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 [xk1 , xk ]} = 0 and sup{f (x) : x
Pn
[xk1 , xk ]} = 1, so that L(f, P ) = 0 and U (f, P ) = k=1 1 (xk xk1 ) = xn x0 =
4 (2) = 6. Thus in this case, changing the partition does not produce better
approximations.
Theorem 7.1.8 For any partitions P and Q of [a, b], and for any bounded function f :
[a, b] R,
L(f, P ) U (f, Q).
Proof. Let R = P Q. (Then R is a refinement of both P and Q.) By Theorem 7.1.6,

L(f, P ) L(f, R) and U (f, P ) U (f, R). Since always L(f, R) U (f, R), the conclusion
follows by transitivity of .
It follows that if f is bounded on [a, b], then the set of all lower sums of f as P varies
over all the partitions of [a, b] is bounded above, so that the set of all lower sums has a real
least upper bound. Similarly, the set of all upper sums is bounded below and has a real
greatest lower bound.
Definition 7.1.9 The lower integral of f over [a, b] is

L(f ) = sup{L(f, P ) : as P varies over partitions of [a, b]},
and the upper integral of f over [a, b] is

U (f ) = inf{U (f, P ) : P varies over partitions of [a, b]}.
We say that f is integrable over [a, b] when L(f ) = U (f ). We call this common
value the integral of f over [a, b], and we write it as
Z b Z b Z b
f= f (x) dx = f (t) dt.
a a a
Theorem 7.1.8 shows that L(f ) U (f ). Equality holds sometimes (for integrable
functions), but not always.
Ra
For example, every function f : [a, a] R is integrable over [a, a], and a f = 0.
If f is a constant function f (x) = c for all x, then all lower and all upper sums are
Rb
c(b a), so that a f = c(b a), and all constant functions are integrable. We have seen
before that if f is non-constant function, then for every partition P , L(f, P ) < U (f, P ),
yet for good functions it can happen that L(f ) U (f ).
Example 7.1.7 shows that sometimes strict inequality holds. Note that we have not
yet proved that the function in Example 7.1.5 is integrable, but Section 7.3 we will prove
that every continuous function is integrable over a closed bounded interval.
Theorem 7.1.10 Let f : [a, b] R be bounded. Then f is integrable over [a, b] if and
only if for all > 0 there exists a partition P of [a, b] such that U (f, P) L(f, P) < .
Proof. Suppose that f is integrable over [a, b]. Then L(f ) = U (f ). Let > 0. Since L(f )
is the supremum of all lower sums L(f, P ) as P varies over partitions of [a, b], there exists
a partition P1 of [a, b] such that L(f ) L(f, P1) < /2. Similarly, there exists a partition
P2 of [a, b] such that U (f, P2) U (f ) < /2. Let P = P1 P2 . Then P is a partition of
[a, b], and by Equation (7.1.4) and by Theorem 7.1.6,
L(f, P1) L(f, P ) U (f, P ) U (f, P2).

Thus since L(f ) = U (f ), we have that
U (f, P )L(f, P ) U (f, P2)L(f, P1) = U (f, P2 )U (f )+L(f )L(f, P1) < /2+/2 = .
Now suppose that for every > 0 there exists a partition P of [a, b] such that U (f, P)
L(f, P) < . By the supremum/infimum definitions of lower and upper integrals,
0 U (f ) L(f ) U (f, P) L(f, P) < .
Since the non-negative constant U (f ) L(f ) is strictly smaller than every positive number
, it means, by Theorem 2.10.4, that U (f ) L(f ) = 0, so that f is integrable.
Example 7.1.11 Let

1, if x = n1 for some n N+ ;
f (x) =
0, otherwise.
R1
Then f is integrable over [0, 1] and 0 f = 0. Namely, let > 0. By Theorem 2.7.12,
there exists a positive integer p such that 1 < p. Thus p1 < . Set d = 2p2 (p+1) 1
, x0 =
1 1
0, for k = 1, . . . , p, set x2k1 = pk+2 d, x2k = pk+2 + d, and x2p+1 = 1. Then
P = {x0 , x1 , . . . , x2p+1 } is a partition of [0, 1]. On each subinterval of the partition, the
infimum of all possible values of f on that subinterval is 0, so that L(f, P ) = 0. However,
M1 (f ) = M2k = 1 and for k 1, M2k+1 = 0. Thus
U (f, P ) L(f, P ) = U (f, P )
p
X p
X
= 1 (x1 x0 ) + 1 (x2k x2k1 ) = x1 x0 + 2d
k=1 k=1
1 1
= d + 2dp < + 2pd
p+1 p+1
1 2p 1 1 1
= + 2 = + =
p + 1 2p (p + 1) p + 1 p(p + 1) p
< .
Thus by the previous theorem, f is integrable over [0, 1].
The following theorem says that to compute (lower, upper) integrals of integrable
functions, we need not use all the possible partitions. It suffices to use, for example, only
partitions into equal-length subintervals. The second part of the theorem below gives a
formulation of integrals that looks very technical but is fundamental for applications (see
Section 7.6).
Theorem 7.1.12 Let f : [a, b] R be bounded and integrable. For each real number
(r) (r) (r)
r > 0 let Pr = {x0 , x1 , . . . , xnr } be a partition of [a, b] such that each subinterval
(r) (r)
[xk1 , xk ] has length at most r. Then
Z b
lim+ L(f, Pr ) = lim+ U (f, Pr ) = f.
r0 r0 a
(r) (r) (r)

Furthermore, if for each r > 0 and each k = 1, 2, . . . , nr , ck is arbitrary in [xk1 , xk ],
then
nr
X Z b
(r) (r) (r)
lim f (ck )(xk xk1 ) = f.
r0+ a
k=1
Proof. Let > 0. By Theorem 7.1.10, there exists a partition P of [a, b] such that
U (f, P ) L(f, P ) < /2. Let P = {y0 , y1 , . . . , yn1 }. Since f is bounded, there exists a
positive real number M such that for all x [a, b], |f (x)| < M . Let r be a positive real
number that is at most 8Mn1 . The elements of P other than a and b may be contained in
two subintervals, so that at most 2n1 1 subintervals in Pr contain the n1 + 1 elements
of P . I will refer to these subintervals of Pr special. Let Pr = {x0 , x1 , . . . , xmr }. Then
L(f, P Pr ) L(f, Pr ) (by Theorem 7.1.6)

mr
X
= mk (f )(xk xk+1 )
k=1
mr
X mr
X
= mk (f )(xk xk+1 ) + mk (f )(xk xk+1 )
k=1,special k=1,non-special
mr
X
(M )(xk xk+1 )
k=1,special
mr
X
+ (inf of f on the corresp. subinterval of P ) (xk xk+1 )
k=1,non-special
(since the non-special [xk1 , xk ] is a subset of a subinterval in P )
mr
X
(M (inf of f on the corresp. subinterval of P )) (xk xk+1 )
k=1,special
mr
X
+ (inf of f on the corresp. subinterval of P ) (xk xk+1 )
k=1
(by adding a 0)
mr
X
(2M )(xk xk+1 ) + L(f, P )
k=1,special
But the total length of the non-special subintervals is at most r(2n1 1) < 2rn1 < /4M,
so that for all r < 8Mn1 ,

L(f, P Pr ) L(f, P ) L(f, Pr ) L(f, P ) (2M ) = .
4M 2
Furthermore, both L(f, P ) and the lower sum of its refinement L(f, P Pr ) are within /2
of L(f ), and since they are both less than or equal to L(f ), it follows that L(f, P ) and
L(f, P Pr ) are within /2 of each other. Thus L(f, Pr ) is within /2 of L(f ) as well.
Similarly, for all r sufficiently small, U (f, Pr ) is within /2 of U (f ) = L(f ). Hence
U (f, Pr ) L(f, Pr ) < , which proves the first paragraph of the theorem.
Furthemore, by compatibility of order with multiplication by positive numbers and
addition, we have that for each r,
nr
X (r) (r) (r)
L(f, Pr ) f (ck )(xk xk1 ) U (f, Pr ),
k=1
Pnr (r) (r) (r) Rb
So that for all r sufficiently close to 0, k=1 f (ck )(xk xk1 ) is within of a f . Thus
Pnr (r) (r) (r) Rb
by the definition of limits, limr0+ k=1 f (ck )(xk xk1 ) = a f .
Notation 7.1.13 In the definition of integral we sometimes write dx or dt, and some-
times we do not. There is no need to write dx when we are simply integrating a function f ,
Rb
as in a f : we seek the signed area determined by f over the domain from a to b. For
this it does not matter if we like to plug x or t or anything else into f . But when we
write f (x) rather than f , then we add dx, and the reason is that f (x) is an element
of the codomain and is not a function. Why do we have to be pedantic? If x and t are
non-dependent variables, by the constant rule established above we then have
Z b
f (x) dt = f (x)(b a),
a
and specifically, by geometric reasoning,
Z 4 Z 4
x dx = 8, x dt = 4x.
0 0
Thus writing dx versus dt is important, and omitting it can lead to confusion: is
the answer the constant 8, or is it 4x depending on x? Furthermore, if x and t are not
independent, we can get further values too. Say if x = 3t, then

Z 4 Z 4
x dt = 3t dt = 24.
0 0
In short, use notation precisely.

7.1.1 Prove that if P = {x0 , . . . , xn } is a partition of [a, b] into n equal parts, then xk =
a + k 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).
7.1.5 Use geometry to compute the following integrals:
Z a
i) f=
Zar p
ii) r 2 x2 dx =
Z0 5
iii) (4x 10) dx =
3

Z 5 5, if x < 7;

2, if 7 x < 1;
iv) f , where f (x) =
1 3x,
if 1 x < 3;
9 x, if 3 x.
7.1.6 Compute the following integrals (t and x do not depend on each other). Justify all
work.
Z rp
i) r 2 x2 dt =
0
Z 5
ii) (4x 10) dt =
3

Z 5, if t < 7;
5
2, if 7 t < 1;
iii) f , where f (t) =
1
3t, if 1 t < 3;
9 x, if 3 t.
7.1.7 In EasyLand they define Eantegrals as follows: first they divide the interval [a, b]
into n equal subintervals, so each has length ba n . By Exercise 7.1.1, the kth subinterval
ba ba
given in this way is a + (k 1) n , a + k n . Rather than finding the minimum and
maximum of f on this subinterval, they simply Eapproximate f on the subinterval by
plugging in the right endpoint a + k ba n , so that the signed area over the kth subinterval

ba ba
is Eapproximately f a + k n n . Thus the Eapproximate signed area of f over [a, b]
Pn
ba ba
via this partition is k=1 f a + k n n . If this sum has a limit as n goes to infinity,
then EasyLanders declare the Eantegral of f over [a, b] to be
b Xn
f = lim b a ba
f a+k .
a n n n
k=1
b
Suppose that f : [a, b] R is integrable. Prove that the Eantegral f exists and equals
Rb a
the integral a f .

7.1.8 Let f (x) = By Example 7.1.7 we know that f is not inte-
grable over [0, 2]. Use Eantegrability from the previous exercise.
2
i) Prove that f is Eantegrable over [0, 2] and find its Eantegral f .
0
2 2 2 2 2
ii) Compute f, f , and prove that f + f 6= f .
0 2 0 2 0
7.2 Computing integrals from upper and lower sums
The definition of integrals appears daunting: we seem to need to compute all the
possible lower sums to get at the lower integral, all the possible upper sums to get at the
upper integral, and only if the lower and upper integrals are the same do we have the precise
integral. In Example 7.1.5 in the previous section we have already seen that numerically
we can often compute the integral to within desired precision by taking finer and finer
partitions. In this section we compute some precise numerical values of integrals, and
without computing all the possible upper and lower sums. Admittedly, the computations
are time-consuming, but the reader is encouraged to read through them to get an idea of
what calculations are needed to follow the definition of integrals. In Section 7.4 we will see
very efficient shortcuts for computing integrals, but only for easy/good functions.
Section 7.2: Computing integrals from upper and lower sums 197
Example 7.2.1 Let f (x) = x on [2, 6]. We know that the area under the curve between
Z 6
x = 2 and x = 6 is 16. Here we compute that indeed f = 16. For any positive integer n
2
let Pn = {x0 , . . . , xn } be the partition of [2, 6] into n equal parts. By Exercise 7.1.1,
xk = 2 + k n4 . Since f is increasing, on each subinterval [xk1 , xk ] the minimum is xk1
and the maximum is xk . Thus
n
X
U (f, Pn) = xk (xk xk1 )
k=1
Xn
4 4
= 2+k
n n
k=1
Xn n 2
4 X 4
= 2 + k
n n
k=1 k=1
2
4 n(n + 1) 4
=2 n+
n 2 n
8(n + 1)
=8+ .
n
It follows that for all n,
8(n + 1)
U (f ) = inf{U (f, P ) : P varies over partitions of [2, 6]} U (f, Pn) = 8 + ,
n
so that U (f ) 8 + 8 = 16. Similarly,
n
X
L(f, Pn) = xk1 (xk xk1 )
k=1
Xn
4 4
= 2 + (k 1)
n n
k=1
Xn n 2
4 X 4
= 2 + (k 1)
n n
k=1 k=1
2
4 (n 1)n 4
=2 n+
n 2 n
8(n 1)
=8+ ,
n
and
L(f ) = sup{L(f, P ) : P varies over partitions of [2, 6]} sup{L(f, Pn) : n N>0 } = 16.
All together this says that
16 L(f ) U (f ) 16,
R6
so that L(f ) = U (f ) = 16, and finally that 2
f = 16.
Note that we did not compute all the possible lower and upper sums, but we computed
enough of them. We knew that we computed enough of them because as n goes large,
sup{L(f, Pn) : n} = inf{U (f, Pn) : n}.
Example 7.2.2 We compute the integral for f (x) = x2 over [1, 7]. For any positive inte-
ger n let Pn = {x0 , . . . , xn } be the partition of [1, 7] into n equal parts. By Exercise 7.1.1,
xk = 1 + k n6 . Since f is increasing on [1, 7], on each subinterval [xk1 , xk ] the minimum is
x2k1 and the maximum is x2k . Thus, by using Exercise 1.5.1 in one of the steps below:
n
X
U (f, Pn) = x2k (xk xk1 )
k=1
Xn 2
6 6
= 1+k
n n
k=1
Xn
12 2 36 6
= 1+k +k 2
n n n
k=1
Xn n n 3
6 X 12 6 X 2 6
= + k + k
n n n n
k=1 k=1 k=1
3
6 n(n + 1) 12 6 n(n + 1)(2n + 1) 6
= n+ +
n 2 n n 6 n
36(n + 1) 36(n + 1)(2n + 1)
=6+ + ,
n n2
so that U (f ) inf{U (f, Pn) : n} = 6 + 36 + 72 = 114. Similarly,
L(f ) L(f, Pn) = 114,
R7
whence 114 L(f ) U (f ) 114 and 1 f = 114.
R2
Example 7.2.3 The goal of this exercise is to compute 0 x dx. For the first attempt,
let Pn = {x0 , . . . , xn } be the partition of [0, 2] into n equal intervals. By Exercise 7.1.1,
xk = 2k
n
. The square root function is increasing, so that
n
X
U (f, Pn) = xk (xk xk1 )
k=1
Xn r
2k 2
=
n n
k=1
3/2 X
n
2
= k.
n
k=1
Pn
But now we are stuck: we have no simplification of k=1 k, and we have no other
Pn
2 3/2
immediate tricks to compute inf{ n k=1 k : n}.
But it is possible to compute enough upper and lower sums for this function to get the
2(n1)2 2n2
integral. Namely, for each positive integer n let Qn = {0, n22 , 24 29
n2 , n2 , . . . , n2 , n2 =
2k2
2}. This is a partition of [0, 2] into n (unequal) parts, with xk = n2 . Since is an
increasing function, on each subinterval [xk1 , xk ] the minimum is achieved at xk1 and
the maximum at xk , so that
X n
2 k 2k 2 2(k 1)2
U (f, Qn) =
n n2 n2
k=1
n
2 2X
= 3 k k 2 (k 1)2
n
k=1
n
2 2X
= 3 k (2k 1)
n
k=1
n n
4 2X 2 2 2X
= 3 k 3 k
n n
k=1 k=1

4 2 n(n + 1)(2n + 1) 2 2 n(n + 1)
= 3 3 (by Exercise 1.5.1)
n 6 n 2
2 2
= (n + 1)(2(2n + 1) 3) (by factoring)
6n2

2
= (n + 1)(4n 1).
3n2

Thus U (f ) inf{U (f, Pn): n} = inf{ 3n22 (n + 1)(4n 1) : n} = 4 2
3 . Similarly, L(f )

4 2
R2
3 , so that 0 x dx = 4 3 2 .

R2
7.2.1 Mimic examples in this section to compute x3 dx. (You may need Exercise 1.5.2.)
0
R2
7.2.2 Mimic examples in this section to compute 0 3 x dx.
7.3 What functions are integrable?
In this section, we continue to assume that the domains and codomains of all functions
are subsets of R.
Theorem 7.3.1 Every continuous real-valued function on [a, b] is integrable over [a, b],
where a, b R with a < b.
Proof. Let f : [a, b] R be continuous. We need to prove that L(f ) = U (f ). By

Theorem 2.10.4 it suffices to prove that for all > 0, U (f ) L(f ) .
So let > 0. By Theorem 5.5.2, 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 k = 1, . . . , n, xk xk1 < . (For example, this can
be accomplished as follows: by Theorem 2.7.12, 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 )
Xn
= (sup{f (x) : x [xk1 , xk ]} inf{f (x) : x [xk1 , xk ]}) (xk xk1 )
k=1
Xn

(xk xk1 ) (by uniform continuity since xk xk1 < )
(b a)
k=1
X n

= (xk xk1 )
(b a)
k=1
= .
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 2.10.4, so that f is integrable over [a, b].
Both the proof above as well as Theorem 7.1.12 prove that for a continuous function f
on [a, b],
Z b Xnr
(r) (r) (r)
f = lim f (ck )(xk xk1 ),
a r0+
k=1
(r) (r) (r)
where for each r > 0, {x0 , x1 , . . . , xnr } is a partition of [a, b], and for each k = 1, . . . , nr ,
ck [xk1 , xk ].
Theorem 7.3.2 Let f : [a, b] R be bounded. Let S = {s0 , s1 , . . . , sm } be a finite subset

of [a, b] with s0 = a < s1 < < sm = b. Suppose that for all d > 0 and all k f is
integrable over [sk1 + d, sk d]. (By the previous theorem, Theorem 7.3.1, any f that is
continuous at all x [a, b] \ S satisfies this hypothesis.) Then f is integrable over [a, b],
and Z b Xm Z sk
f= f.
a k=1 sk1
Proof. Let > 0. Since f is bounded, there exists a positive real number M such that
for all x [a, b], |f (x)| < M . Let e = 31 min{s1 s0 , s2 s1 , . . . , sm sm1 } and d =
min{e, /(4M (m + 1)(2m + 1))}. By assumption, for each k = 1, . . . , m 1, f is integrable

on the interval [sk1 + d, sk d]. Thus by Theorem 7.1.10, there exists a partition Pk of
[sk1 + d, sk d] such that U (f, Pk ) L(f, Pk ) < /(2m + 1). Now let P = {a} P1 P2
Pm1 {b}. Then P is a partition of [a, b], and
m
X
U (f, P ) L(f, P ) = (U (f, P [sk1 + d, sk d]) L(f, P [sk1 + d, sk d]))
k=1
+ (U (f, P [a, a + d]) L(f, P [a, a + d]))
m1
X
+ (U (f, P [sk d, sk + d]) L(f, P [sk d, sk + d]))
k=1
+ (U (f, P [b d, b]) L(f, P [b d, b]))
m
X m1
X
(U (f, Pk ) L(f, Pk )) + 2M 2d + 2M 2d + 2M 2d
k=1 k=1
Xm

< + 4M (m + 1)d
2m + 1
k=1
< .
Rb Pm R sk
Thus by Theorem 7.3.1, f is integrable. Furthermore, a f k=1 sk1 f is bounded
Pm Pm
above by U (f, P ) k=1 L(f, Pk ) and below by L(f, P ) k=1 U (f, Pk ), and the upper
Rb P R sk
and lower bounds are within 2 of 0, so that a f m k=1 sk1 f is within 2 of 0. Since
Rb Pm R sk
is arbitrary, it follows by Theorem 2.10.4 that a f = k=1 sk1 f.
Ra
Notation 7.3.3 What could possibly be the meaning of f if a < b? In our definition
b Rb
of integrals, all partitions started from a smaller a to a larger b to get a f . If we did
reverse b and a, then the widths of the subintervals in each partition would be negative
(xk xk1 = (xk1 xk )), so that all the partial sums and both the lower and the upper
integrals would get the negative value. Thus it seems reasonable to declare
Z a Z b
f = f.
b a
In fact, this is exactly what makes Theorem 7.3.2 work without any order assumptions
Rb Rc Rb
on the sk . For example, if a < c < b, by Theorem 7.3.2, a f = a f + c f , whence
Rc Rb Rb Rb Rc
a
f = a f c f = a f + b f.
Theorem 7.3.4 Suppose that f and g are integrable over [a, b], and that c R. Then
Rb Rb Rb
f + cg is integrable over [a, b] and a (f + cg) = a f + c a g.
Proof. We first prove that L(cg) = U (cg) = cL(g) = cU (g). If c 0, then

L(cg) = sup{L(cg, P ) : P a partition of [a, b]}
= sup{cL(g, P ) : P a partition of [a, b]}

= c sup{L(g, P ) : P a partition of [a, b]}
= cL(g)
= cU (g)
= c inf{U (g, P ) : P a partition of [a, b]}
= inf{cU (g, P ) : P a partition of [a, b]}
= inf{U (cg, P ) : P a partition of [a, b]}
= U (cg),
and if c < 0, then
L(cg) = sup{L(cg, P ) : P a partition of [a, b]}

= sup{cU (g, P ) : P a partition of [a, b]}
= c inf{U (g, P ) : P a partition of [a, b]}
= cU (g)
= cL(g)
= c inf{L(g, P ) : P a partition of [a, b]}
= sup{cL(g, P ) : P a partition of [a, b]}
= inf{U (cg, P ) : P a partition of [a, b]}
= U (cg).
Rb Rb
This proves that cg is integrable with a
(cg) = c a
g.
Let > 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 2.7.9, for every partition of [a, b], and in particular for the partition R,
L(f + cg, R) L(f, R) + L(cg, R), and U (f + cg, R) U (f, R) + U (cg, R). Then
0 U (f + cg) L(f + cg)

U (f + cg, R) L(f + cg, R)
U (f, R) + U (cg, R) L(f, R) L(cg, R)
< .
Thus U (f + cg) L(f + cg) = 0 by Theorem 2.10.4, and so f + cg is integrable. The
inequalities L(f + cg, R) L(f, R) + L(cg, R), and U (f + cg, R) U (f, R) + U (cg, R)
furthermore prove that L(f )+L(cg) L(f +cg) U (f +cg) U (f )+U (cg) = L(f )+L(cg),
so that finally
Z b Z b Z b Z b Z b
(f + cg) = f+ (cg) = f +c g.
a a a a a
Theorem 7.3.5 Let a, b R with a < b. Let f, g : [a, b] R be integrable functions such
that f (x) g(x) for all x [a, b]. Then
Z b Z b
f g.
a a
Here is a picture that illustrates this theorem: the values of g are at each point in the
domain greater than or equal to the values of f , and the area under the graph of g is larger
than the area under the graph of f :
y = g(x)
y = f (x)
a b x
Proof. By assumption on every subinterval I of [a, b], inf{f (x) : x I} inf{g(x) : x I}.
Thus for all partitions P of [a, b], L(f, P ) L(g, P ). Hence L(f ) L(g), and since f and
Rb Rb
g are integrable, this says that a f a g.

R1 R3 R3 R2
7.3.1 The following is known: 0 f = 5, 2 f = 6, 0 f = 15, 1 g = 3. Compute and
R2 R3 R1
justify 1 (3f 4g), 1 5f , 2 3g.

x, if x is rational;
7.3.2 Let f (x) =
i) Prove that f is not integrable over [0, 1].
ii) Does this contradict Theorem 7.3.2? Justify.
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 integrable odd function. Prove that a
f = 0.
Ra Ra
7.3.5 Let f : [a, a] R be an integrable even function. Prove that a
f =2 0
f.
7.3.6 Let f : [a, b] R be piecewise continuous as in Theorem 7.3.2.
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 Let f : [a, b] R be monotone. Prove that f is integrable over [a, b].
7.4 The Fundamental Theorem of Calculus
Despite first appearances, it turns out that integration and differentiation are related.
For this we have two versions of the Fundamental Theorem of Calculus.
Theorem 7.4.1 (The Fundamental Theorem of Calculus, I) Let f, g : [a, b] R

such that f is integrable over [a, b] and g is differentiable with g = f . Then
Z b
f = g(b) g(a).
a
Proof. Let P = {x0 , x1 , . . . , xn } be a partition of [a, b]. Since g is differentiable on [a, b], it
is continuous on each [xk1 , xk ] and differentiable on each (xk1 , xk ). Thus by the Mean
Value Theorem (Theorem 6.3.4), there exists ck (xk1 , xk ) such that f (ck ) = g (ck ) =
g(xk )g(xk1 )
xk xk1 . By the definition of lower and upper sums,
n
X
L(f, P ) f (ck )(xk xk1 ) U (f, P ).
k=1
But
n
X n
X g(xk ) g(xk1 )
f (ck )(xk xk1 ) = (xk xk1 )
xk xk1
k=1 k=1
Xn
= (g(xk ) g(xk1 ))
k=1
= g(xn ) g(x0 )
= g(b) g(a),
so that L(f, P ) g(b) g(a) U (f, P ), whence

L(f ) = sup{L(f, P ) : P a partition of [a, b]}
g(b) g(a)
inf{U (f, P ) : P a partition of [a, b]}
= U (f ).
Since f is integrable over [a, b], by definition L(f ) = U (f ), and so all inequalities above
Rb
have to be equalities, so that necessarily a f = g(b) g(a).
The general notation for applying Theorem 7.4.1 is as follows: if g = f , then
Z b b

f = g(x) = g(b) g(a).
a a

For example, assuming that cos (x) = sin(x), we have that
Z

sin x dx = cos(x) = cos() ( cos(0)) = (1) + 1 = 2.
0 0
If we instead had to compute this integral with upper and lower sums, it would take us a
lot longer and a lot more effort to come up with the answer 2.
In general, upper and lower sums and integrals are time-consuming and we want to
avoid them if possible. The fundamental theorem of calculus that we just proved enables
us to do that for many functions: to integrate f over [a, b] one needs to find g with g = f .
Such g is called an antiderivative of f . For example, if r is a rational number different
r+1
from 1, then by the power rule (Theorem 6.2.10), an antiderivative of xr is xr+1 . By the
r+1
scalar rule for derivatives, for any constant C, xr+1 + C is also an antiderivative. It does
not matter which antiderivative we choose to compute the integral:
Z b r+1 r+1
r b a br+1 ar+1
x dx = +C +C = ,
a r+1 r+1 r+1 r+1
so that the choice of the antiderivative is irrelevant.
Definition 7.4.2 (Indefinite integral) If g is an antiderivative of f , we write also

Z
f (x) dx = g(x) + C,
where C stands for an arbitrary constant.

R R R R
For example, 3x2 dx = x3 + C, 3 dx = 3x + C, t dx = tx + C, x dx = 12 x2 + C,
and so on. (Study the differences and similarities of the last three.)
So far we have seen 2x for rational exponents x. Exercise 5.4.6 also allows real ex-
ponents, and proves that this function of x is continuous. Thus by Theorem 7.3.1 this
R
function is integrable. We do not yet know 2x dx, but in Theorem 7.5.5 we will see that
R x R 2
2 dx = ln12 2x + C. For 2(x ) dx instead, you and I do not know an antiderivative,
we will not know one by the end of the course, and there actually is no closed-form
antiderivative. This fact is due to a theory of Joseph Liouville (18091882). What is the
meaning of closed-form? Here is an oblique answer: Exercise 9.8.5 claims that there ex-
2
ists an infinite power series (sum of infinitely many terms) that is an antiderivative of 2(x ) .
2
Precisely because of this infinite sum nature, the values of any antiderivative of 2(x ) cannot
be computed precisely, only approximately. Furthermore, according to 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 9.4).
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, b) with
Z x
d
f = f (x).
dx a
Proof. Since f is continuous over [a, b], it is continuous over [a, x], so that by Theorem 7.3.1,
f is integrable over [a, x]. Thus g is a well-defined function. Let c (a, b). We will prove
that g is differentiable at c.
Let > 0. By continuity of f at c, 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 7.3.5,
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
Z max{x,c}
|x c|(f (c) ) f |x c|(f (c) + ).
min{x,c}
If x 6= c, dividing by |x c| and rewriting the middle term says that

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

g(x) g(c) a f ac f
f (c) = f (c)
xc xc
Rx
f

= c f (c) (by Theorem 7.3.2 and Notation 7.3.3)
xc
< .
g(x) g(c)
Thus lim 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 Z Z
x x x
f= f (t) dt = f (x) dx.
a a a
This is a function of x because x appears in the bound of the domain of integration. Note
Rx
similarly that a f (t) dz are functions of t and x but not of z. Thus by the Fundamental
Theorem of Calculus, II,
Z x Z x
d d
f = f (x), and f (t) dz = f (t).
dx a dx a
So far we have defined integrals of real-valued functions. By Theorem 7.4.1, if g = f ,

Rb Rb
then a f = g(b) g(a), i.e., a g = g(b) g(a). If g is complex-valued, we know that
Rb
g = (Re g) + i(Im g) , so that it would make sense to define a g as the integral of (Re g)
plus i times the integral of (Im g) . Indeed, this is the definition:
Definition 7.4.4 Let f : [a, b] C be a function such that Re f and Im f are integrable
over [a, b]. Define the integral of f over [a, b] to be
Z b Z b Z b
f= Re f + i Im f.
a a a
The following are then immediate generalizations of the two versions of the fundamen-
tal theorem of calculus:
Theorem 7.4.5 (The Fundamental Theorem of Calculus, I, for complex-valued

functions) Let f, g : [a, b] C such that f is continuous and g is differentiable with
g = f . Then
Z b
f = g(b) g(a).
a
Theorem 7.4.6 (The Fundamental Theorem of Calculus, II, for complex-valued

functions) Let f : [a, b] C be continuous. Then for all x [a, b], f is integrable over
Rx
[a, x], and the function g : [a, b] R given by g(x) = a f is differentiable on (a, x) with
Z x
d
f = f (x).
dx a

7.4.1 Compute and justify or verify the integrals below. Clever guessing is fine.
Z 1
i) x3 dx =
Z0 1

ii) x dx =
Z0 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 =
0
Z 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
iii) x3 dx =
dx 0
Z t
d
iv) t3 dt =
dx Z0
x
d
v) t3 dx =
dx 0
Z t
d
vi) x3 dt =
dx 0
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
Z1 3
4x + 3
p dx =
1 2x2 + 3x)10
Z 3 p
3
(8x + 6) 2x2 + 3x dx =
1
7.4.6 (Integration by parts) By the product rule for differentiation, (f g) (x) =
f (x)g(x) + f (x)g (x).
Rb Rb
i) Prove that a f (x)g(x)) dx = f (b)g(b) f (a)g(a) a f (x)g (x) dx.
R R
ii) Prove that f (x)g(x)) dx = f (x)g(x) f (x)g (x) dx.
iii) Compute the following integrals: explicitly state f, g in applying this rule:
Z 1
(4x + 3)(5x + 1)10 dx =
Z1
1
4x + 3
dx =
1 2x + 4
7.4.7 Prove by integration by parts the following improper integral value for a non-negative
integer n:
Z 1
1
(x2n ln x)dx = .
0 (2n + 1)2
7.4.8 Compute the following derivatives (hint: the Fundamental Theorem of Calculus and
the chain rule):
Z 3x q
d
i) x4 + 5 x dx.
dx 2
Z x
d x+5 x
ii) dx.
dx x2 x1 00 2x5 0 + x7 2
Z h(x)
d
iii) f (t) dt.
dx g(x)
Z h(x)
d
iv) f.
dx g(x)
7.4.9 Consult Exercise 7.4.14.
R
i) Compute 1 x12 dx.
R
ii) For any rational number p < 1, compute 1 xp dx. (The same is true for real
p < 1, but we have not developed enough properties for such functions.)
7.4.10 Consult Exercise 7.4.18.
R1
i) Compute 0 1x dx.
R1
ii) For any rational number p > 1, compute 0 xp dx. (The same is true for real
p > 1, but we have not developed enough properties for such functions.)
7.4.11 (Mean Value Theorem for integrals) Let f : [a, b] R be continuous. Prove
that there exists c (a, b) such that
Z b
1
f (c) = f.
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 (x) = f (a) + f (t) dt.
a
ii) Integrate the integral above by parts, and rewrite, to get that
Z x

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

Z x
f (a) f (n) (a) n (x t)n (n+1)
f (x) = f (a) + (x a) + + (x a) + f (t) dt.
1! n! a n!
iv) Say why you cannot apply the Fundamental Theorem of Calculus, II, to compute
d
Rx n+1 (n+1)
dx a (x t) f (t) dt.
v) (Taylor remainder formula in integral form) Consult Section 6.5 for Taylor poly-
Rx n
nomials to prove that f (x) = Tn,f,a (x) + a (xt)
n!
f (n+1) (t) dt.
7.4.14 (Improper integral, unbounded domain) Let f : [a, ) R be continuous.
i) Discuss how our construction/definition of integrals fails when the domain is not
bounded.
RN
ii) Prove that for all N [a, ), a f exists.
RN
iii) If limN a f exists, we call the limit the (improper) integral of f over [a, ).
R
We denote it a f . Observe that this is a limit of limits. Similarly formulate the
Rb
definition of g for a continuous function g : (, b] R.

3x + 1, if x < 10; R
iv) Let f (x) = . Compute and justify 0 f .
0, otherwise.
R 1
7.4.15 Let p > 1. Prove that 1 xp dx exists, and compute it. Comment on what happens
to the integral if p 1.
7.4.16 (Improper integral, unbounded domain) Let f : R R be bounded such
Rc R
that for some c R, f and c f exist in the sense of Exercise 7.4.14.
Re R
i) Prove that for all e R, f and e f exist and that
Z c Z Z e Z
f+ f= f+ f.
c e
R
We denote this common value as f .
R RN
ii) Prove that f = limN N f .
7.4.17 Let f : R R over R. This exercise is meant to show that integrability of f
RN
over R cannot be simply defined as the existence of the limit limN N f . Namely, let

1, if x (the floor of x) is even;
f (x) =
1, otherwise.
i) Sketch the graph of this function.
RN
ii) Prove that for every positive real number N , N f = 0.
iii) Prove that f is not integrable over [0, ) or over (, 0) in the sense of Exer-
cise 7.4.14.
7.4.18 (Improper integral, unbounded function) Let f : (a, b] R be continuous.

i) Discuss how our construction/definition of integrals fails when the domain does
not include the boundaries of the domain.
Rb
ii) Prove that for all N (a, b), N f exists.
Rb
iii) If limN a+ N f exists, we call this limit the (improper) integral of f over [a, b],
Rb Rb
and we denote it a f . Similarly formulate a f if the domain of f is [a, b) or (a, b).
Rb
iv) Let f : (a, b) R be given by f (x) = 1. Prove that a f = b a.
7.5 Natural logarithm and the exponential functions
The function that takes a non-zero x to 1/x is continuous everywhere on its domain
since it is a rational function. Thus by Theorem 7.3.1 and Notation 7.3.3, for all x > 0,
Rx 1
1 x
dx is well-defined. This function has a familiar name:
Definition 7.5.1 The natural logarithm is the function

Z x Z x
1 1
ln x = dx = dt
1 x 1 t
for all x > 0.
We prove below all the familiar properties of this familiar function.
Remark 7.5.2
R1
(1) ln 1 = 1 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
g (x) = cx c = x1 = ln (x). Thus the function g ln has constant derivative 0. It
follows by Theorem 6.3.5 that g ln is a constant function. Hence for all x R+ ,
ln(cx) ln(x) = g(x) ln(x) = g(1) ln(1) = ln(c) 0 = ln(c).
This proves that for all c, x R+ ,

ln(cx) = ln(c) + ln(x).
(7) By the previous part, for all c, x R+ ,

c
ln = ln(c) ln(x).
x
(8) For all non-negative integers n and all c R+ , ln(cn ) = n ln(c). We prove this
by mathematical induction. If n = 0, then ln(cn ) = ln(1) = 0 = 0 ln c = n ln c.
Now suppose that equality holds for some n 1. Then ln(cn ) = ln(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.
Now write r = m r
n for some integers m, n with n 6= 0. Then n ln(c ) = n ln(c
m/n
)=
m r m
ln(c ) = m ln(c), so that ln(c ) = n ln(c) = r ln(c).
(10) The range of ln is R = (, ). Here is a proof. By geometry, ln(0.5) < 0 < ln(2).
Let y R+ . By Theorem 2.7.12, there exists n N+ such that y < n ln(2). Hence
ln 1 = 0 < y < n ln(2) = ln(2n ), so that since ln is continuous, by the Intermediate
Value Theorem (Theorem 5.3.1), there exists x (1, 2n) such that ln(x) = y. If
y R , then by the just proved we have that y = ln(x) for some x R+ , so
that y = ln(x) = ln(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 5.3.4, ln1 is
increasing and continuous.
(12) By Theorem 6.2.7, the derivative of ln1 is
1
(ln1 ) (x) = 1 = ln1 (x).
ln (ln (x))
(13) For all x, y R,
1
ln1 (x) 1 ln (x) 1 1
1

1 = ln ln 1 = ln ln ln (x) ln ln (y) = ln1 (x y) .
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 con-
stant;
(2) The exponential function with base c when r varies and c is constant.
Theorem 7.5.4 Let r R. The function f : R+ R+ given by f (x) = xr is differen-

tiable, with f (x) = 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
f (x) = ln1 (r ln(x)) xr = r lnln1(r(ln(x))
ln(x))
= r ln1 (r ln(x) ln(x)) = r ln1 ((r 1) ln(x)) =
rxr1 . The monotone properties then follow from Theorem 6.3.5.
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
Note that U (f, {1, 1.25, 1.5, 1.75, 2, 2.25, 2.5}) = 0.25(1 + 1.25 + 1.5 + 1.75 + 21 + 2.25
1
)=
2509
R 2.5
2520 < 1, so that ln 2.5 = 1 f is strictly smaller than this upper sum. It follows that
ln 2.5 < 1 = ln e and 2.5 < e. If P is a partition of [1, 2.71828] into a million pieces of equal
length, a computer gives that U (f, P ) is just barely smaller than 1, so that 2.71828 < e.
If P is a partition of [1, 2.718285] into a million pieces of equal length, then L(f, P ) is just
barely bigger than 1, so that e < 2.718285. Thus e = 2.71828.
A reader may want to run further computer calculations for greater precision. A dif-
ferent and perhaps easier computation is in Exercise 7.5.13.
Theorem 7.5.7 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 213 that the derivative of ln1 is ln1 :
Theorem 7.5.8 (ex ) = ex .

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
ii) Prove that if c 6= 1, then a cx dx = ln(c) (cb ca ).
R x
7.5.3 Use integration by substitution (Exercise 7.4.5) to compute x2 +4 dx.
R 2
7.5.4 Let c R+ . Use integration by substitution (Exercise 7.4.5) to compute xc(x ) dx.
R
7.5.5 Let c R+ . Use integration by parts (Exercise 7.4.6) to compute xcx dx.
R
7.5.6 Use integration by parts (Exercise 7.4.6) to compute ln(x) dx.
7.5.7 This exercise is about applying various versions of the LH
opitals rule.
ln x
i) Prove that lim x = 0.
x
x ln x
ii) Prove that lim+ x ln x = 0. (Hint: Use one of x ln x = 1/ ln x or x ln x = 1/x , and
x0
perhaps not both work.)
iii) Compute and justify lim ln(xx ).
x0+
iv) Compute and justify 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

3
x+2(x+7)3 (x2)4
. There is another way if the range of the function consists of positive
real numbers: apply ln of both sides, take derivatives of both sides, and solve for f (x).
For example, if f (x) = xx , then ln(f (x)) = ln(xx ) = x ln(x), so that
f (x) x
= (ln(f (x))) = (x ln(x)) = ln(x) + = ln(x) + 1,
f (x) x
so that f (x) = xx (ln(x) + 1).
3
i) Compute and justify the derivative of f (x) = (x2 + 2)x +4 .
2
(x1)3 (x+4)3 x2 +1
ii) Compute and justify the derivative of f (x) = x
3
x+2(x+7)3 (x2)4
.
1
7.5.9 Prove that lim ln(1 + x) = 1. (Hint: LHopitals rule.)
x0+ x c
i) Prove that for any c R+ , lim x ln 1 + = c. (Hint: Change variables.)
x x
c x
+
ii) Prove that for any c R , lim 1 + = ec .
x x
7.5.10 Prove that f (x) = (1 + x1 )x is a strictly increasing function. (Hint: use logarithmic
differentiation and Exercise 7.5.1.)
7.5.11 Prove that for all x R, (1 + x1 )x < e. (Hint: previous two exercises.)
ex 1
7.5.12 Prove that limx0 x = 1. You may want to apply LHopitals rule (Theo-
rem 6.4.3).
7.5.13 Let f (x) = ex .
i) For any positive integer n, compute the Taylor polynomial Tn,f,0 for ex of degree n
centered at 0 (Definition 6.5.2).
ii) Use 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 215. On a computer, try to get
more digits of e with those methods, and with methods in this exercise. Which
method is faster? Can you streamline either method?
7.5.14 Let f (x) = ex . (Hint: Rewrite for LHopitals rule.) Use LHopitals rule (Theo-
rem 6.4.3) and induction on n to prove that
ex Tn,f,0 (x) 1
lim = .
x0 xn+1 (n + 1)!

1/x2
7.5.15 Let f : R R be given by f (x) = e , if x 6= 0;
0, if x = 0.
i) Prove by induction on n 0 that for each n there exists a polynomial function
hn : R R such that

1 1/x2
(n)
f (x) = hn ( x ) e , if x 6= 0;
0, if x = 0.
(At non-zero x you can use the chain rule, the derivative of the exponential func-
tion, and the power rule for derivatives. However, f (n+1) (0) is but of course
(n) (n)
computed as limh0 f (h)f h
(0)
, and then you have to use LHopitals rule.
You do not have to be explicit about the polynomial function.)
ii) Compute the nth Taylor polynomial for f centered at 0.
7.5.16 (A friendly competition between e and ) Use calculus (not a calculator!) to
determine which number is bigger, e or e . You may assume that 1 < e < . (Hint:
compute the derivative of some function.)
7.6 Applications of integration
The Fundamental theorems of calculus relate integration with differentiation. In par-

Rb
ticular, to compute a f , if we know an antiderivative g of f , then the integral is easy.
However, as already mentioned after the Fundamental Theorem of Calculus I, many func-
tions do not have a closed-form antiderivative. One can still compute definite integrals
up to a desired precision, however: we take finer and finer partitions of [a, b], and when
U (f, P ) and L(f, P ) are within a specified distance from each other, we know that the true
integral is somewhere in between, and hence up to the specified precision either L(f, P ) or
Rb
U (f, P ) stands for a f . In applications, such as in science and engineering, many integrals
have to be and are computed in this way because of the lack of closed-form antiderivatives.
In this section we look at many applications that exploit the original definition of
integrals via sums over finer and finer partitions. For many concrete examples we can then
solve the integral via antiderivatives, but for many we have to make do with numerical
approximation.
7.6.1 Length of a curve

Let f : [a, b] R be a continuous function. If the graph of f is a line, then by the
p
Pythagorean theorem the length of the curve is (b a)2 + (f (b) f (a))2 . For a general
curve it is harder to determine its length from (a, f (a)) to (b, f (b)). But we can do the
standard calculus trick: let P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on each subinterval
[xk1 , xk ] we approximate the curve with the line (xk1 , f (xk1)) to (xk , f (xk )), compute
p
the length of that line as (xk xk1 )2 + (f (xk ) f (xk1 ))2 , and sum up all the lengths:
n p
X
(xk xk1 )2 + (f (xk ) f (xk1 ))2 .
k=1
Whether this is an approximation of the true length depends on the partition, but geomet-
rically it makes sense that the true length of the curve equals
n p
X
lim (xk xk1 )2 + (f (xk ) f (xk1 ))2 ,
k=1
as the partitions {x0 , x1 , . . . , xn } get finer and finer. But this is not yet in form of The-
orem 7.1.12. For that we need to furthermore assume that f is differentiable on (a, b).
Then by the Mean Value Theorem (Theorem 6.3.4) for each k = 1, . . . , n there exists
ck (xk1 , xk ) such that f (xk ) f (xk1 ) = f (ck )(xk xk1 ). If in addition we assume
that f is continuous, then it is integrable by Theorem 7.3.1, and by Theorem 7.1.12 the
true length of the curve equals
n p
X
lim (xk xk1 )2 + (f (ck )(xk xk1 ))2
P ={x0 ,x1 ,...,xn }finer and finer,ck [xk1 ,xk ]
k=1
n p
X
= lim 1 + (f (ck ))2 (xk xk1 )
k=1
Z b p
= 1 + (f (x))2 dx.
a
We just proved:
Theorem 7.6.1 If f : [a, b] R is continuous and differentiable such that f is continuous

on [a, b], then the length of the curve from (a, f (a)) to (b, f (b)) is
Z bp
1 + (f (x))2 dx.
a
7.6.2 Volume of a surface area of revolution, disk method

Let f : [a, b] R be continuous. We rotate the region between x = a and x = b and
bounded by the x-axis and the graph of f around the x-axis. If the graph is a horizontal
line, then the rotated region is a disk of height b a and radius f (a), so that its volume
is (f (a))2(b a). For a general f we let P = {x0 , x1 , . . . , xn } be a partition of [a, b]; on
each subinterval [xk1 , xk ] we approximate the curve with the horizontal line y = f (ck )
for some ck [xk1 , xk ], we compute the volume of the solid of revolution obtained by
rotating that approximated line over the interval [xk1 , xk ] around the x-axis, and sum up
all the volumes:
Xn
(f (ck ))2 (xk xk1 ).
k=1
Geometrically it makes sense that the true volume equals

n
X
lim (f (ck ))2 (xk xk1 ),
k=1
Rb
and by Theorems 7.1.12 and 7.3.1, this equals a
(f (x))2 dx. This proves:
Theorem 7.6.2 If f : [a, b] R is continuous, then the volume of the solid of revolution
obtained by rotating around the x-axis the region between x = a and x = b and bounded
by the x-axis and the graph of f is
Z b
(f (x))2 dx.
a
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 [yk1 , yk ] we approximate the curve f, g with the horizontal
line f (ck ), g(ck ) for some ck [yk1 , yk ], we compute the volume of the solid of revolution
obtained by rotating that approximated region over the interval [yk1 , yk ] around the x-
axis, and sum up all the volumes:
n
X
(g(ck ) f (ck ))(yk2 yk1
2
).
k=1
Geometrically it makes sense that the true volume equals

n
X
lim (g(ck ) f (ck ))(yk2 yk1
2
)
P ={y0 ,y1 ,...,yn }finer and finer,ck [yk1 ,yk ]
k=1
n
X
= lim (g(ck ) f (ck ))(yk + yk1 )(yk yk1 ),
P ={y0 ,y1 ,...,yn }finer and finer,ck [yk1 ,yk ]
k=1
Rb
and by Theorems 7.1.12 and 7.3.1, this equals a
(g(y) f (y))2y dy. This proves:
Theorem 7.6.3 If 0 a b and f, g : [a, b] R are continuous, then the volume of the
solid of revolution obtained by rotating around the x-axis the region between y = a and
y = b and bounded by the graphs of x = f (y) and x = g(y), is
Z b
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 x-
axis. The circle of radius r centered at the origin consists of all points (x, y) such that
p p
x2 + y 2 = r 2 , so we have g(y) = r 2 y 2 and f (y) = r 2 y 2 . Thus the volume is
Z r p r
4 4 4
4 y r 2 y 2 dy = (r 2 y 2 )3/2 = (r 2)3/2 = r 3 .
0 3 0 3 3
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 2 = 2|m| . The area of the full circle is radius squared times one half of
2 2 2
b +(mb) 1+m2
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 21+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
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
= m b2 a2 + 2(b a)l/m 1 + m2
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 [xk1 , xk ] we approximate
the curve with the line from (xk1 , f (xk1)) to (xk , f (xk )). By the assumption that f
takes on only non-negative values we have that both xk1 , xk are both on the same side of
the x-intercept of that line. The equation of the line is y = f (xxkk)f (xk1 )
xk1
(x xk ) + f (xk ),
f (x )f (x ) f (x )f (x )
so that m = k k1
xk xk1 and l = xkk xk1k1 xk + f (xk ). Since f is differentiable, by
the Mean Value Theorem (Theorem 6.3.4) there exists ck (xk1 , xk ) such that f (ck ) =
f (xk )f (xk1 )
xk xk1 . Thus m = f (ck ) and l = f (xk ) f (ck )xk .
We rotate that line segment around the x-axis, compute its volume of that solid of
revolution, and add up the volumes for all the subparts:
X n p

(xk xk1 ) (f (ck )(xk + xk1 ) + 2 (f (xk ) f (ck )xk )) 1 + (f (ck ))2
k=1
n
X p
= |(f (ck )(xk + xk1 ) + 2f (xk ))| 1 + (f (ck ))2 (xk xk1 ).
k=1
By Theorems 7.1.12 and 7.3.1 we get the following:
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 p Z b p

2
|f (x)(x + x) + 2f (x)| 1 + (f (x)) dx = 2 f (x) 1 + (f (x))2 dx.
a a

7.6.1 Compute the perimeter of the circle with methods from this section.
7.6.2 Compute the volume of the sphere of radius r. Justify your work.
2 2 2
7.6.3 Compute the volume of the ellipsoid whose boundary satisfies xa2 + yb2 + zb2 = 1.
(We need multi-variable calculus to be able to compute the volume of the ellipsoid whose
2 2 2
boundary satisfies xa2 + yb2 + zc2 = 1.)
7.6.4 Compute the volume of a conical pyramid with base radius r and height h.
7.6.5 Compute the volume of a doughnut (you specify its dimensions).
7.6.6 Compute the surface area of the sphere of radius r.

7.6.7 If a tiny particle of mass m is rotating around a circle of radius r, its moment of
inertia of the particle is I = mr 2 . Likewise, if a (circular) hoop of radius r and of mass m
is rotating around its center, its moment of inertia of the particle is I = mr 2 . Now suppose
that we have a thin circular plate of radius 5a meters and mass 4 kilograms. (If you wish,
let b be the thickness of the plate and the mass divided by the volume.) The goal of this
exercise is to find its moment of inertia about its diameter.
i) Draw the circular plate with center at origin. Let the axis of rotation be the y-axis.
ii) Let P be a partition of [0, 5]. Prove that the moment q of inertia of the sliver of
the plate between xk1 and xk is approximately 542 x2k x2k1 c2k , where ck
[xk1 , xk ].
iii) Prove that the moment of inertia of the rotating circular plate is 50 kilogram
meters squared.
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 sub-
merged under water from depth a to depth b. At depth h, the cross section area of the
object is A(h). Compute the total force exerted on the object by the water.
Chapter 8: Sequences
In this chapter, Sections 8.5 and 8.4 contain identical results in identical order, but
the proofs are different. You may want to learn both perspectives, or you may choose to
omit one of the two sections.
8.1 Introduction to sequences
Definition 8.1.1 An infinite sequence is a function with domain N+ and codomain C.

If s is such a function, instead of writing s(n), it is common to write sn , and the sequence
s is commonly expressed also in all of the following notations (and obviously many more):
s = {s1 , s2 , s3 , . . .} = {sn }
n=1 = {sn }n1 = {sn }n = {sn+3 }n=2 = {sn4 }n>4 = {sn }.
The nth element in the ordered list is called the nth term of the sequence.
The notation {s1 , s2 , s3 , . . .} usually stands for the set consisting of the elements
s1 , s2 , s3 , . . ., and the order of a listing of elements in a set is irrelevant. Here, however,
{s1 , s2 , s3 , . . .} stands for the sequence, and the order matters. When the usage is not
clear from the context, we add the word sequence or set as appropriate.
The first term of the sequence {2n 1}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
more concisely as { dc n c+d
2 (1) + 2 }n .
(3) The sequence {(1)n } = {1, 1, 1, 1, . . .} has an infinite number of terms, and
its range is the finite set {1, 1}. The range of the sequence {in } is the set
{i, 1, i, 1}.
(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 . This sequence is different from the sequence
{n} = {1, 2, 3, . . .}.
(5) The sequence {2n} is an abbreviation for the sequence {2, 4, 6, 8, . . .}. This is not
the same as the sequence {4, 2, 8, 6, 12, 10, . . .}, namely, if we scramble the order
of the terms, we change the sequence.
(6) The sequence of all odd positive integers (in the natural order) can be written as
{2n 1} = {2n 1}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
1.5.31, where in particular it is proved
n n
that for all integers n, sn = 1 1+ 5
1 1 5
.)
5 2 5 2
(9) Some sequences, just like functions, do not have an algebraic expression for terms.
For example, let s be the sequence whose nth term is the nth prime number.
This sequence starts with 2, 3, 5, 7, 11, 13, 17, 19, 23, and we could write many more
terms out explicitly, but we do not have a formula for them. (This s is indeed an
infinite sequence since there are infinitely many primes, as proved on page 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
n/2, if n is even;
sn =
(n 1)/2, otherwise.
(Otherwise applies to odd (positive) integers.)
(12) One can scramble the set Q+ of all positive rational numbers into a sequence via
a diagonal construction as follows. First of all, each positive rational number can
be written in the form a/b for some positive integers a, b. Rather than plotting the
fraction a/b, we will plot the point (a, b) in the plane. Refer to Plot 8.1.2: the bold
points are elements of N+ N+ , and each such (a, b) is identified with the fraction
a/b. Every positive rational number appears in this way somewhere as a bold point,
and all appear multiple times because ab = ac bc . Now we want to systematically
enumerate these bold points/rational numbers. If we first enumerate all of them
in the first row, and then proceed to the second row, well, actually, we never get
226 Chapter 8: Sequences
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.
b b b b b b
b b b b b b
b b b b b b
b b b b b b
b b b b b b
1 b b b b b b
1 2 3 4 5 6
Plot 8.1.2 Beginning of the enumeration of the positive rational numbers.
It is important to note that every positive rational number appears on this list,
and because we are skipping any repetitions, it follows that every positive rational
number appears on this list exactly once. Thus this gives an enumeration of
positive rational numbers. *
A different enumeration of Q+ is given with an algebraic formulation in Exer-
cise 2.4.27.
(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
* Here is a fun exercise: look at the ordered list of positive rational numbers above, including the crossed-out
(n+m2)2
fractions. Verify for a few of them that n/m is in position 2 + 3n+m4
2 + 1 on the list. Namely, it is a
(x+y2)2 3x+y4 + 2 +
fact that f (x, y) = 2 + 2 + 1 gives a bijection of (N ) with N . This was first proved by Rudolf
Fueter and George P olya, but the proof is surprisingly hard, using transcendence of er for algebraic numbers r,
so do not attempt to prove this without much more number theory background.
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
a plot of the sequence {1/n}.
sn
1 b
b
b
b b
n
b b b b b b b b b b
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Plot 8.1.2 sn = 1/n
(15) Another way to plot a sequence is to simply plot and label each sn in the complex
plane or on the real line. We plot three examples below.
s1 , s3 , s5 , . . . s2 , s4 , s6 , . . . s3 s2 s1
1 1 0 1
s1
b
i/2
s3 s4 s2 b
b
b
b b
b b
b b bb b b
b bb b b b
1 1
Plot 8.1.2 Plots of the image sets of {(1)n}, {1/n} and {(1)n/n + 0.5i/n}.
There is an obvious arithmetic on sequences (just like there is on functions):

{sn } {tn } = {sn tn },
{sn } {tn } = {sn tn },
c{sn } = {csn },
{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},
{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,
n n n
and, to save vertical space, just like for limits of functions, we also use a variation on the
last three: limn sn = L, limn {sn } = L, limn (sn ) = L.
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 8.5
in terms of limits of functions. More epsilon-N proofs are in Section 8.4. Be aware that
this section is more concrete; the next section assumes greater ease with abstraction.
Example 8.2.2 Consider the sequence s = {1/n}. Plot 8.1.2 gives a hunch that lim sn =
0, and now we prove it. [Recall that 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/)
= .
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 b bb
b
bb b b
b b
b
s4
1
b
b
s2 1
b
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
|sn 0| = ((1) + i(1)
n n+1
) 0
n

1
= |(1)n + i(1)n+1 | (because |ab| = |a||b|)
n
1
= |(1)n + i(1)n+1 | (because n is positive)
n
1 p
= ((1)n )2 + ((1)n+1 )2
n
1
= 2
n
1
< 2 (because n > N )
N

[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!]

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
|sn 0| = ((1) + i(1)
n n+1
) 0
n

1
= |(1)n + i(1)n+1 | (because |ab| = |a||b|)
n
1
= |(1)n + i(1)n+1 | (because n is positive)
n
1 p
= ((1)n )2 + ((1)n+1 )2
n
1
= 2
n
1
< 2 (because n > N )
N
2
= (because N = 2/)
2/
= ,

which proves that for all n > 2/, |sn 0| < . Since is arbitrary, this proves that
lim sn = 0.
2
2n+3n
Example 8.2.4 Claim: lim 3+4n+n 2 = 3. Proof: Let > 0. Set N = . Let n be
an integer strictly bigger than N . Then

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

3 + 4n + n2 3 = 3 + 4n + n2 3 + 4n + n2

9 10n
=
3 + 4n + n2
9 + 10n
= (because n > 0) [Assuming that N > 0.]
3 + 4n + n2
n + 10n
(because n 9) [Assuming that N 8.]
3 + 4n + n2
11n
=
3 + 4n + n2
11n
2 (because 3 + 4n + n2 > n2 , so 1/(3 + 4n + n2 ) < 1/n2 )
n
11
=
n
11
< (because n > N )
N
11
(because N 11/ 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

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

3 + 4n + n2 3 = 3 + 4n + n2 3 + 4n + n2

9 10n
=
3 + 4n + n2
9 + 10n
= (because n > 0)
3 + 4n + n2
n + 10n
(because n 9)
3 + 4n + n2
11n
=
3 + 4n + n2
11n
(because 3 + 4n + n2 > n2 , so 1/(3 + 4n + n2 ) < 1/n2 )
n2
11
=
n
11
< (because n > N )
N
11
(because N 11/ 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.
Below is a polished proof of a very similar problem.
2
2n+3n
Example 8.2.6 Claim: lim 94n+n 2 = 3. Proof: Let > 0. Set N = max{2, 20/}. Then

2n + 3n2 2n + 3n2 3(9 4n + n 2
)

9 4n + n2 3 = 9 4n + n2 9 4n + n2

27 + 10n
=
9 4n + n2
27 + 10n
= (because n > N 2, so n 3, so 10n 27 > 0,
9 4n + n2
and n2 4n + 9 = (n 3)2 + 2n > 0)
10n
< (because 10n 27 < 10n)
9 4n + n2
10n
=
9 4n + 0.5n2 + 0.5n2
10n
< (because 2(9 4n + 0.5n2) = (n 4)2 + 2 > 0)
0.5n2
20
=
n
20
< (because n > N )
N
20
(because N 20/ so 1/N 1/(20/))
20/
= ,
which proves that for all n > N , |sn 3| < . Since is arbitrary, this proves that the limit
of this sequence is 3.

1 n
Example 8.2.7 lim 1+ n = 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 8.2.8 lim n1/n = 1.
Proof. For all n 2, by Exercise 1.6.7,

n
X
1/n n 1/n n n
n = (n ) = (n 1 + 1) = (n1/n 1)k .
k
k=0
Each of the summands is non-negative, and if we only use the summands with k = 0 and
k = 2, we then get that
1
n 1 + n(n 1)(n1/n 1)2 .
2
By subtracting 1we get that n1 21 n(n1)(n1/n 1)2 , so that for n 2, n2 (n1/n 1)2 ,
and hence that n2 n1/n 1. Certainly n1/n 1 0 for all n 1. It follows that for all

n 2, and even for all n 1, 0 n1/n 1 2 . Now let > 0. Set N = max{2, 2/2}.
n
Then N is a positive real number. Let n > N be an integer. Then 0 n1/n 1 2 <
n
2 = , which proves that |n1/n 1| < , and hence proves this limit.
N
Example 8.2.9 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 2.7.12) 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+1B ,
B 1/n
so that by compatibility of < with multiplication by positive numbers, B+1 < M 1.
Hence by compatibility of < with addition,
B 1 1
0 1 M 1/n < 1 = < < ,
B+1 B+1 B
so that |1 M 1/n | < , which proves that lim M 1/n = 1.

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

1
|sn 0| = 2 0
n
1
= (because )
n2
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)
n
= 0.
8.2.5 Prove that lim 2n+1
n2 4
= 0.
8.2.6 Prove that lim 3n+4
2n
= 23 .

8.2.7 Prove that lim n + 1 n = 0.
Pn 1
8.2.8 Suppose that the terms of a sequence are given by sn = k=1 k(k+1) .
Pn 1 1
i) Using induction on n, prove that k=1 k(k+1) = 1 n+1 .
ii) Use this to find and prove the limit of {sn }.
8.2.9 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 2.7.13, for every positive integer n, there exists a rational
number sn (r n1 , r + n1 ). Prove that {sn } converges to r.
8.2.11 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.)
p
8.2.12 Prove that limn { (n!) (n!) } = 1.
n
8.2.13 What is wrong with the following proof that limn 2n+1 = 12 .
1
2 1
Proof. Let > 0. Set N = 2 . Let n > N . Then

n 1 2n (2n + 1)

2n + 1 2 = 2(2n + 1)

1
=
2(2n + 1)
1
=
2(2n + 1)
1
< (because all terms are positive)
2(2N + 1)
= .
8.3 Divergence of infinite sequences and infinite limits
The sequence {(1)n} alternates in value between 1 and 1, and does not seem to
converge to a single number. The following definition addresses this situation.
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 314), 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:
6 L.
Example 8.3.3 For all complex numbers L, lim{n} =
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 the 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:
Definition 8.3.4 A real-valued sequence {sn } diverges to if for every positive real
number M there exists a positive number N such that for all integers n > N , sn > M . We
write this as lim sn = .
A real-valued sequence {sn } diverges to if for every negative real number M
there exists a positive real number N such that for all integers n > N , sn < M . We write
this as lim sn = .
Example 8.3.5 lim n = .
Proof. Let M > 0. Set N = M . (As in epsilon-delta or epsilon-N proofs, we must figure
out what to set N to. In this case, N = M works). Let n N+ with n > N . Since N = M ,
we conclude that n > M , and the proof is complete.

n
Example 8.3.6 lim n! = .
Proof. Let M > 0. Let M0 be an integer that is strictly greater than M . By Example 8.2.9
there exists N1 > 0 such that for all integers n > N1 ,
s
M0 ! M +1M
n 0
M
1 < .
(M0 + 1) 0 M0 + 1
Since M0 ! = M0 (M0 1)(M0 2) 3 2 1 < (M0 + 1)M0 , this means that

s
M0 ! M0 + 1 M M
0 1 n M
< =1 ,
(M0 + 1) 0 M0 + 1 M0 + 1
q
so that n (M0M 0!
+1)M0
> MM0 +1
. Set N = max{N1 , M0 }. Then for all integers n > N ,
M0 !
n! = n(n 1) (M0 + 1) M0 ! (M0 + 1)nM0 M0 ! = (M0 + 1)n ,
(M0 + 1)M0
so that s

n M0 ! M
n! (M0 + 1) n M
> (M0 + 1) = M.
(M0 + 1) 0 M0 + 1
Theorem 8.3.7 (Comparison theorem (for sequences with infinite limits)) Let
{sn }, {tn} be real-valued sequences such that for all sufficiently large n (say for n N for
some fixed N ), sn tn .
(1) If lim sn = , then lim tn = .
(2) If lim tn = , then lim sn = .
Proof. (1) By assumption lim sn = for every positive M there exists a positive N such
that for all integers n > N , sn > M . Hence by assumption tn sn for all n N we get
that for every positive M and for all integers n > max{N, N }, tn > M . Thus by definition
lim tn = .
Part (2) has an analogous proof.
2
Example 8.3.8 lim n n+1 = .
2
Proof. Note that for all n N+ , n n+1 = n + n1 n, and since we already know that
2
lim n = , it follows by the comparison theorem above that lim n n+1 = .
2
Example 8.3.9 lim n 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
n2 1 1 n N
=n > M.
n n 2 2
Theorem 8.3.10 Let {sn } be a sequence of positive numbers. Then lim sn = if and
only if lim s1n = 0.
Proof. Suppose that lim sn = . Let > 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 n }.
iii) {(n + 1)3 n3 }.
iv) {(1)n + 3/n}.
8.3.2 Suppose that {sn } diverges and {tn } converges. Prove that {sn tn } diverges.
8.3.3 Suppose that {sn } diverges and {tn } converges to a non-zero number. Prove that
{sn tn } diverges.
8.3.4 Give an example of two divergent sequences {sn }, {tn } such that {sn +tn } converges.
8.3.5 Let {sn } be a sequence of negative numbers. Prove that lim sn = if and only if
lim s1n = 0.
8.3.6 Given the following sequences, find and prove the limits, finite or infinite, if they
exist. Otherwise, prove divergence:
i) { nn+5
3 5 }.
2
n
ii) { 2n
3n2 5 }.
n
iii) { (1)
n3
}.
n
iv) { (1) n
n+1 }.
3
8n
v) { nn2 +8n }.
2
vi) { 1n
n }.
2n
vii) { n! }.
n!
viii) { (n+1)! }.
8.3.7 Prove or give a counterexample:

i) If {sn } and {tn } both diverge, then {sn + tn } diverges.
ii) If {sn } converges and {tn } diverges, then {sn + tn } diverges.
iii) If {sn } and {tn } both diverge, then {sn tn } diverges.
iv) If {sn } converges and {tn } diverges, then {sn tn } diverges.
8.3.8 Find examples of the following:
sn
i) A sequence {sn } of non-zero terms such that lim{ sn+1 } = 0.
sn+1
ii) A sequence {sn } of non-zero terms such that lim{ sn } = 1.
iii) A sequence {sn } of non-zero terms such that lim{ sn+1
sn } = .
iv) A sequence {sn } such that lim{sn+1 sn } = 0.
v) A sequence {sn } such that lim{sn+1 sn } = .
8.3.9 Suppose that the sequence {sn }n diverges to (or to ). Prove that {sn }n
diverges. (The point of this exercise is to parse the definitions correctly.)
8.4 Convergence theorems via epsilon-N proofs
All of the theorems in this section are also proved in Section 8.5 with a different
method; here we use the epsilon-N formulation for proofs without explicitly resorting to
functions whose domains have a limit point. I recommend reading this section and omitting
Section 8.5.
Theorem 8.4.1 If a sequence converges, then its limit is unique.
Proof. Let {sn } be a convergent sequence. Suppose that {sn } converges to both L and L .
Then for any > 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 2.10.4 it must be the case that |L L | = 0, i.e., that L = L .
Theorem 8.4.2 Suppose that lim{sn } = L and that L 6= 0. Then there exists a positive
number N such that for all integers n > N , |sn | > |L|/2. In particular, there exists a
positive number N such that for all integers n > N , sn 6= 0.
Proof. Note that p = |L|/2 is a positive real number. Since lim{sn } = L, it follows that
there exists a real number N such that for all integers n > N , |sn L| < |L|/2. Then by
the reverse triangle inequality (proved in Theorem 2.10.3),
|sn | = |sn L + L| = |(sn L) + L| |L| |sn L| > |L| |L|/2 = |L|/2.
Theorem 8.4.3 Suppose that lim sn = L and lim tn = K. Then

(1) (Constant rule) For any complex number c, lim{c} = c.
(2) (Linear rule) lim{1/n} = 0.
(3) (Sum/difference rule) lim{sn tn } = L K.
(4) (Scalar rule) For any complex number c, lim{csn } = cL.
(5) (Product rule) lim{sn tn } = LK.
(6) (Quotient rule) If tn 6= 0 for all n and K 6= 0, then lim{sn /tn } = L/K.
(7) (Power rule) For all positive integers m, lim{sm m
n}= L .
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 the triangle inequality)
< /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|
|csn cL| = |c| |sn L| |c|/(|c| + 1) = |c|+1 < . Since is any positive number, we
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 the triangle inequality)
= |sn (tn K)| + |(sn L)K|
= |sn | |tn K| + |sn L| |K|

< (|L| + 1) + |K|
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 8.4.2 there exists a positive number N4 such that for all integers
n > N4 , |tn | > |K|/2. Let N = max{N1 , N2 , N3 , N4 }. Then for all integers n > N ,

sn
L = K s n L tn
tn K K tn

K s n tn s n + tn s n L tn
= (by adding a clever 0)

K tn

K s n tn s n tn s n L tn
+

(by the triangle inequality)

K tn K tn
1 1 1
= |K tn ||sn | + |L sn |
|K| |tn | K
2
|K| 1 2 |K| 1
< (1 + |L|) +
4(1 + |L|) |K| |K| 2 |K|

= +
2 2
= ,
which proves (6).
Part (7): The proof is by induction on m, the base case being the assumption. 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. This proves (7).
5n2
Example 8.4.4 Suppose sn = 3n+4 . To prove that lim sn = 5/3, we note that sn =
5n2 1/n 52/n
3n+4 1/n = 3+4/n .By the linear rule, lim(1/n) = 0, so by the scalar rule, lim(2/n) =
lim(4/n) = 0. Thus by the constant, sum, and difference rules, lim{5 2/n} = 5 and
lim{3 + 4/n} = 3, so that by the quotient rule, lim sn = 5/3.
2 2
3n+2 1/n 3/n+2/n
Example 8.4.5 Let sn = n3n+2 2 3 . Note that sn = n2 3 1/n2 = 13/n2 . By the linear
rule, lim 1/n = 0, so that by the scalar rule, lim 3/n = 0, and by the product and scalar
rules, lim 2/n2 = lim 3/n2 = 0. Thus by the sum and difference rules, lim{3/n + 2/n2 } = 0
and lim{1 3/n2 } = 1. Finally, by the quotient rule, lim sn = 0/1 = 0.
Theorem 8.4.6 (Power, polynomial, rational rules for sequences) For any positive
integer m, limn {1/nm } = 0. If f is a polynomial function, then lim{f (1/n)} = f (0). If
f is a rational function that is defined at 0, then lim{f (1/n)} = f (0).
Proof. By the linear rule (in Theorem 8.4.3), limn {1/n} = 0, and by the power rule,
for all positive integers m, limn {1/nm } = 0.
Now write f (x) = a0 + a1 x + + ak xk for some non-negative integer k and some
complex numbers a0 , a1 , . . . , ak . By the constant, power and repeated sum rules,

lim f (1/n) = lim a0 + a1 (1/n) + a2 (1/n)2 + + ak (1/n)k
n n
= a0 + lim a1 (1/n) + lim a2 (1/n)2 + + lim ak (1/n)k

n n n
2 k
= a0 + a1 0 + a2 0 + + ak 0
= a0 = f (0).
This proves the polynomial rule.

Finally, let f be a rational function. Write f (x) = g(x)/h(x), where g, h are polynomial
functions with h(0) 6= 0. By the just-proved polynomial rule, limn g(1/n) = g(0),
limn h(1/n) = h(0) 6= 0, so that by the quotient rule, limn f (1/n) = g(0)/h(0) =
f (0).
Theorem 8.4.7 (The composite rule for sequences) Suppose that lim sn = L. Let
g be a function whose domain contains L and all terms sn . Suppose that g is continuous
at L. Then lim g(sn ) = g(L).
Proof. Let > 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 8.4.8 Suppose that lim sn = L. Then

(1) lim |sn | = |L|.
(2) lim Re sn = Re L.
(3) lim Im sn = Im L.
Furthermore, since the real and imaginary parts determine a complex number, we
moreover get:
Theorem 8.4.9 A sequence {sn } of complex numbers converges if and only if the se-
quences {Re sn } and {Im sn } of real numbers converge.
Proof. By Theorem 8.4.8 it suffices to prove that if lim{Re sn } = a and lim{Im sn } = b,

then lim{sn } = a + bi. Let > 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 8.4.10 (Comparison of sequences) Let s and t be convergent sequences of

complex numbers. Suppose that |sn | |tn | for all except finitely many n. Then | lim sn |
| lim tn |.
If in addition for all except finitely many n, sn , tn are real numbers with sn tn , then
lim sn lim tn .
Proof. Let L = lim sn , K = lim tn . By Theorem 8.4.8, lim |sn | = |L|, lim |tn | = |K|.
Suppose that |L| > |K|. Set = (|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
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 8.4.11 (The squeeze theorem for sequences) Suppose that s, t, u are se-
quences of real numbers and that for all n N+ , sn tn un . If lim s and lim u both
exist and are equal, then lim t exists as well and
lim s = lim t = lim u.

Proof. Set L = lim s = lim u. Let > 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 8.4.12 lim(n + 1)1/n = 1.

1/n 1/n
Proof. For n 2, 1 n+1n
32 n1/n. Thus by the previous theorem and

by Example 8.2.8, lim{ n+1 )1/n n = 1. Hence by Theorem 8.4.3 and by Example 8.2.8,
n n o n o
1/n 1/n 1/n n+1 1/n
lim{(n + 1) } = lim n+1 n n = lim n lim{n1/n} = 1.
Exercises for Sections 8.4 and 8.5

8.4.1 Compute the following limits, and justify:
i) lim n212 .
2
3n +20
ii) lim 2n3 20 .
2
iii) lim 5n2n
2 n1 .
2
iv) lim n 4n
+4n2
2 +2 .
3
v) lim 3n +2
2n3 +n .
4
2n +2
vi) lim 3n2 +n1 +2 .
8.4.2 Give an example of a sequence {sn } and a number L such that lim |sn | = |L| but
{sn } does not converge. Justify your example.
n n
o
8.4.3 Prove that lim (1)n = 0.
8.4.4 Let {sn } be a convergent sequence of positive real numbers. Prove that lim sn 0.
m
8.4.5 Prove that for all integers m, limn n+1n = 1.
8.4.6 Prove that limn {3 n2 } = 3.

8.4.7 Prove that lim 3 n + 1 3 n = 0.
n
8.4.8 By Example 8.2.7, lim 1 + n1 = e. Determine the following limits:
n+1 n
i) .
n n o
1 2n
ii) 1 + 2n .
n o
2n
iii) 1 + n1 .
1
n
iv) 1 + 2n .
n o
1 n+1
v) 1 + n .

8.4.9 Let s1 be a positive real number. For each n 1, let sn+1 = sn . Prove that
lim sn = 1. (Hint: Prove that lim sn = lim s2n .)
8.5 Convergence theorems via functions
The results in this section are the same as those in Section 8.4, but here they are proved
with theorems about limits of functions that were proved in Chapter 4. So, a connection is
made between limits of functions and limits of sequences. The reader may omit this section
(or the previous one). This section is more abstract; one has to keep in mind connections
with functions as well as theorems about limits of functions to get at theorems about limits
of sequences. Exercises for this section appear at the end of Section 8.4.
For any sequence s we can define a function f : {1/n : n N+ } C with f (1/n) = sn .
Conversely, for every function f : {1/n : n N+ } C we can define a sequence s with
sn = f (1/n).
The domain of f has exactly one limit point, namely 0. With this we have the usual
notion of limx0 f (x) with standard theorems from Section 4.3,
Theorem 8.5.1 Let s, f be as above. Then lim sn = L if and only if 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 8.5.2 (Compare the reasoning in this example with the epsilon-N proofs of
5n2 1/n 52/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 8.5.1,
lim sn = 5/3.
3n+2 3n+2 1/n2 3/n+2/n2
Example 8.5.3 Suppose sn = n2 3 . Note that sn = 2
n 3 1/n2 = 13/n2 . The
+ 3x+2x2
corresponding function f : {1/n : n N } is f (x) = 13x2 , and by the scalar,
sum,
difference, product, and quotient rules for limits of functions, lim sn = limx0 f (x) =
0+0
10 = 0.
Theorem 8.5.4 The limit of a converging sequence is unique.
Proof. Let {sn } be a convergent sequence. Suppose that {sn } converges to both L
and L . Let f : {1/n : n N+ } be the function corresponding to s. By Theorem 8.5.1,
limx0 f (x) = L and limx0 f (x) = L . By Theorem 4.3.1, L = L . This proves uniqueness
of limits for sequences.
Theorem 8.5.5 Suppose that lim{sn } = L and that L 6= 0. Then there exists a positive
number N such that for all integers n > N , |sn | > |L|/2. In particular, there exists a
positive number N such that for all integers n > N , sn 6= 0.
Proof. Let f be the function corresponding to s. Then 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.
Theorem 8.5.6 Suppose that lim sn = L and lim tn = K. Then

(1) (Constant rule) For any complex number c, lim{c} = c.
(2) (Linear rule) lim{1/n} = 0.
(3) (Sum/difference rule) lim{sn tn } = L K.
(4) (Scalar rule) For any complex number c, lim{csn } = cL.
(5) (Product rule) lim{sn tn } = LK.
(6) (Quotient rule) If tn 6= 0 for all n and K 6= 0, then lim{sn /tn } = L/K.
(7) (Power rule) For all positive integers m, lim{sm n}= L .
m
Proof. Let f, g : {1/n : n N+ } C be given by f (1/n) = sn , g(1/n) = tn . By

Theorem 8.5.1, 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 8.5.1 also for s, t. Theorem 4.3.4 finishes the proof
of (7).
For part (1), declare f (1/n) = c and for part (2), declare f (1/n) = 1/n. Again
Theorems 4.3.3 and 8.5.1 easily finish the proofs of (1), (2).
The following theorem for sequences follows immediately from the corresponding
power, polynomial, and rational rules for functions:
Theorem 8.5.7 (Power, polynomial, rational rules for sequences) Let f be a poly-
nomial function. Then lim{f (1/n)} = f (0). In particular, for any positive integer m,
limn {1/nm } = 0. If f is a rational function that is defined at 0, then lim{f (1/n)} =
f (0).
Theorem 8.5.8 (The composite rule for sequences) Suppose that lim sn = L. Let g
be a function whose domain contains L and all terms sn . Suppose that g is continuous at
L. Then lim g(sn ) = g(L).
Proof. Let f (1/n) = sn . By Theorem 8.5.1, limx0 f (x) = L, and by assump-

tion 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 8.5.1, lim g(sn ) = g(L).
In particular, since the absolute value function, the real part, and the imaginary part
functions are continuous everywhere, we immediately conclude the following:
Theorem 8.5.9 Suppose that lim sn = L. Then

(1) lim |sn | = |L|.
(2) lim Re sn = Re L.
(3) lim Im sn = Im L.
Since the real and imaginary parts determine a complex number, we also have:
Theorem 8.5.10 A sequence {sn } of complex numbers converges if and only if the se-
quences {Re sn } and {Im sn } of real numbers converge.
Proof. Let f (1/n) = sn . By Theorem 4.3.8, 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 8.5.11 (Comparison of sequences) Let s and t be convergent sequences of

complex numbers. Suppose that |sn | |tn | for all except finitely many n. Then | lim sn |
| lim tn |.
If in addition for all except finitely many n, sn , tn are real numbers with sn tn , then
lim sn lim tn .
Proof. Let A be the set of those 1/n for which |sn | |tn | in the first case and for which
sn tn in the second case. Let f, g : A R be the functions f (1/n) = |sn |, g(1/n) = |tn |
in the first case, and f (1/n) = sn , g(1/n) = tn in the second case. By assumption
for all x in the domain, f (x) g(x). Since 0 is a limit point of the domain (despite
omitting finitely many 1/n) and since by Theorem 8.5.1, 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 8.5.12 (The squeeze theorem for sequences) Suppose that s, t, u are se-
quences of real numbers and that for all n N+ , sn tn un . If lim s and lim u both
exist and are equal, then lim t exists as well and
lim s = lim t = lim u.
Proof. Let f, g, h : {1/n : n N+ } C be functions defined by f (1/n) = sn , g(1/n) = tn ,

h(1/n) = un . The assumption is that for all x in the domain of f, g, h, f (x) g(x) h(x),
and by Theorem 8.5.1 that 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 8.5.1, lim s = lim t = lim u.
Example 8.5.13 lim(n + 1)1/n = 1.

n+1 1/n

3 1/n
Proof. For n 2, 1 n n o 2 n1/n. Thus by the previous theorem and

n+1 1/n
by Example 8.2.8, lim n = 1. Hence by Theorem 8.5.6 and by Example 8.2.8,
n o n o
1/n n+1 1/n
lim{(n + 1)1/n } = lim n+1
n
n 1/n
= lim n
lim{n1/n} = 1.
Exercises for this section appear at the end of Section 8.4.
8.6 Bounded sequences, monotone sequences, ratio test

Definition 8.6.1 A sequence {sn } is bounded if there exists a positive real number B
such that for all integers n, |sn | B.
If all sn are real numbers, we say that {sn } is bounded above (resp. below) if there
exists a real number B such that for all positive integers n, sn B (resp. sn B).
n
4+i
Sequences {4+5i}, {1/n}, { 5+2i }, {(1)n }, {34in}, { 3n4n 2 4 }, {
n
n} are bounded
2
(the latter by Example 1.5.6), but {n}, {(1)nn}, { n n+1 }, { n n!}, {2 n } are not bounded.
The next theorem provides many examples of bounded sequences.
Theorem 8.6.2 Every convergent sequence is bounded.
Proof. Let {sn } be a convergent sequence with limit L. Thus there exists a positive integer
N such that for all integers n > N , |sn L| < 1. Set B = max{|s1 |, |s2|, |s3|, . . . , |sN +1 |, |L|+
1}. Then for all positive integers n N , |sn | B by definition of B, and for n > N ,
|sn | = |sn L + L| |sn L| + |L| < 1 + |L| B.
Another proof of the theorem above is given in the next section via Cauchy sequences.
Definition 8.6.3 A sequence {sn } of real numbers is called non-decreasing (resp. non-
increasing, strictly increasing, strictly decreasing) if for all n, sn sn+1 (resp.
sn sn+1 , sn < sn+1 , sn > sn+1 ). Any such sequence is called monotone.
2
Sequences {1/n}, {n} are strictly decreasing, { n n+1 } is strictly increasing, {(1)n n}
2
is neither increasing nor decreasing, { n n+5 }n1 is neither increasing nor decreasing, but
2
{ n n+5 }n2 is strictly increasing.
Theorem 8.6.4 (Bounded monotone sequences) Let {sn } be a bounded sequence

of real numbers such that for some integer N , {sn }nN is non-decreasing (resp. non-
increasing). Then lim sn exists, and equals the least upper bound (resp. greatest lower
bound) of the set {sN , sN +1 , sN +2 , . . .}.
Proof. Suppose that for all n N , sn sn+1 . By the Least upper theorem (Axiom 2.9.1),
the least upper bound of the set {sN , sN +1 , sN +2 , . . .} exists. Call it L.
Let > 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.
Theorem 8.6.5 (Ratio test for sequences) Let {sn } be a sequence of non-zero complex
sn+1
numbers and let L be a real number in the interval [0, 1). Suppose that lim sn = L, or
more generally, that there exists a positive integer N1 such that for any integer n N1 ,
sn+1
sn L.
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. The general condition is that for all positive integers n > N1 ,
sn+1
sn L < r.

Under the first (limit) condition, since lim sn+1sn = L, there exists a positive number

N such that for all integers n > N , sn+1
sn L < r L. Set N1 = N + 1. Then for
all n N1 N ,

sn+1 sn+1 sn+1

sn = sn L + L sn L + L r L + L = r.
Thus all
hypotheses
are that there exists a positive integer N1 such that for all integers
sn+1
n N1 , sn r. Then by easy induction (see Exercise 1.5.22), for all n N1 ,
|sn | r nN1 |sN1 |.
For all positive integers n, r n is positive, so that r < 1 implies that r n+1 < r n . Thus
the sequence {r n } is a non-increasing sequence that is bounded below by 0 and above by 1.
By Theorem 8.6.4, K = inf{r n : n N+ } is the limit of {r n }. Since 0 is a lower bound
and K is the greatest of lower bounds of {r n : n N+ }, necessarily 0 K.
Suppose that K > 0. Then K(1 r)/(2r) is a positive number. Since K is the
infimum of the set {r, r 2 , r 3 , . . .}, there exists a positive integer N > N1 such that 0
r N K < K(1 r)/(2r). By multiplying by r we get that r N +1 < rK + K(1 r)/2
= K r + K(1r)2
= K (1+r)
2
< K (1+1)2
= K, and since K is the infimum of all 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 , N2 + N1 }, and let n > N be an integer. Then
1

|sn 0| r nN1 |sN1 | |sN1 | < .
|sN1 | + 1
This proves that lim sn = 0.
Corollary 8.6.6 (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.

8.6.1 For each of the following sequences {sn }, you may wish to compute the limit of {|sn |}
via a comparison, ratio test, decreasing property, or other tricks, to prove the indicated.
i) limn 21n = 0.
(1)n
ii) limn 3n = 0.
(4)n
iii) limn 7n = 0.
n
k
iv) limn n!
= 0 for all k C.
n
v) limn kn
= 0 for all non-zero k C with |k| > 1.
m
n
vi) limn kn = 0 for all non-zero k C with |k| > 1 and all integers m.
8.6.2 Give an example of a sequence {sn } of non-zero complex numbers such that
lim{sn+1 /sn } exists and has absolute value strictly smaller than 1.
8.6.3 Give an example of a sequence {sn } of non-zero complex numbers such that
lim{sn+1 /sn } = i/2.

8.6.4 Let sn be the nth Fibonacci number. Prove that lim sn+1
sn =
1+ 5
2 . (Hint: Exer-
cise 1.5.31.)
8.6.5 Let r C satisfy |r| > 1. Prove that the sequence {r n } is not bounded.
8.6.6 (Ratio test again, and compare to Theorem 8.6.5) Let {sn } be a sequence of non-zero
complex numbers such that lim sn+1
sn = L. Suppose that |L| < 1. Prove that lim sn = 0.
8.6.7 Consider sequences {(1)n } and {1}. For each, determine whether it converges, and
the limit of ratios | sn+1
sn |. Comment why the Ratio Test does not apply.
8.6.8 Let {sn } be a sequence of real numbers such that for all n 1, sn sn+1 (resp.
sn sn+1 ). Prove that {sn } is convergent if and only if it is bounded.
n
8.6.9 Prove that the sequence 1 + n1 n
is strictly increasing. (Hint: Exercise 7.5.9.)

8.6.10 Prove that the sequence { n n! } is strictly increasing.
8.6.11 Let {sn } be a non-decreasing sequence. Prove that the set {s1 , s2 , s3 , . . .} is
bounded below.
8.6.12 (Monotone sequences) Let {sn } be a monotone sequence of real numbers.
i) Suppose that {sn } is not bounded above. Prove that {sn } is non-decreasing and
that lim sn = .
ii) Suppose that {sn } is not bounded below. Prove that {sn } is non-increasing and
that lim sn = .
iii) Prove that lim sn is a real number if and only if {sn } is bounded.
8.7 Cauchy sequences, completeness of R, C
So far, to determine convergence of a sequence required knowing the limit. In this

section we prove an alternate machinery for deciding that a sequence converges without
knowing what its limit is. This will be used for subsequential limits and in comparison
theorems for convergence of series in the next chapter.
Definition 8.7.1 A sequence {sn } is Cauchy if for all > 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 .
Section 8.7: Cauchy sequences, completeness of R, C 253
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 (this was already
proved in Theorem 8.6.2). 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.
Theorem 8.7.5 (Completeness of R, C) Every Cauchy sequence in R or C is conver-

gent.
Proof. First let {sn } be a Cauchy sequence in R (as opposed to in C). By Theorem 8.7.2,
{sn } is bounded. It follows that all subsets {s1 , s2 , s3 , . . .} are bounded too. In particular,
by the Least upper bound theorem (Axiom 2.9.1), un = sup{sn , sn+1 , sn+2 , . . .} is a real
number. For all n, un un+1 because un is the supremum of a larger set. Any lower
bound on {s1 , s2 , s3 , . . .} is also a lower bound on {u1 , u2 , u3 , . . .}. Thus by Theorem 8.6.4,
the monotone sequence {un } has a limit L = inf{u1 , u2 , u3 , . . .}.
We claim that L = lim{sn }. Let > 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 for all integers n > N1 . 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 8.5.10 or

Theorem 8.4.9, {sn } converges to a + bi.
Example 8.7.6 (Harmonic sequence) The sequence {1 + 12 + 31 + + n1 }n is not Cauchy,

does not converge, is monotone, and is not bounded.
Proof. Let sn be the nth term of the sequence. If m > n, then

1 1 1 1 mn n
sm sn = + + + + =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.)
Here is a thought experiment related to this: we can stack finitely many books to form
a bridge to the Moon and beyond! All books weigh one unit of weight and are one unit
along the longest side. The books are stacked so that each one protrudes out from the
heap below as much as possible and with keeping the center of mass stable. The topmost
book, by the uniform assumption, protrudes out 12 of its length. Actually, for the book
to be stable this should actually be 21 minus a tad, but that tad can be taken to be say
1
, say, or much much less. We will ignore this tiny tad, but an interested reader
googolplex
may wish to work through the stacking below with (varied!) corrections being incorporated
in all the centers of mass and all the protrusions. This diagram shows the top two books.
The center of mass of the top two books is clearly at 43 from the right-hand edge of the
bottom book, so that the third book from the top down should protrude from underneath
the second one 41 .
21+1 12
The center of mass of this system, measured from the rightmost edge, is 3
= 56 ,
so that the fourth book has to protrude out 16 .
31+1 1
The center of mass of this system, measured from the rightmost edge, is 4
2
= 78 ,
so that the fifth book has to protrude out 18 units.
In general, the center of mass of the top n books is at 2n1
2n measured from the
1
rightmost edge, so that the (n + 1)st book should protrude out by 2n units.
Thus the total protrusion of the top n books equals

1 1 1 1
1+ + ++
2 2 3 n1
units. We just proved that this sum grows beyond limit. Thus we can reach the Moon
with enough books (and a platform to stand on).

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 Let r C satisfy |r| > 1. Prove that lim r n diverges. (Hint: Check the Cauchy
property.)
8.7.5 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) A 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+ .
Note that for all n, kn n.

Examples 8.8.2 Every sequence is a subsequence of itself. Sequences {1/2n}, {1/3n},

{1/(2n + 1)} are subsequences of {1/n}, and {1}, {1}, {(1)n+1 } are subsequences of
{(1)n }. The constant sequence {1} is not a subsequence of {1/n}, because the lat-
ter sequence does not have infinitely many terms equal to 1. If {sn } is the sequential
enumeration of Q+ on page 226, then {1/n}, {n}, { n+1n
}, { n12 } are subsequences, but
n
{ n+2 } = {1/3, 1/2, . . .} is not.
Theorem 8.8.3 A subsequence of a convergent sequence is convergent, with the same

limit. A subsequence of a Cauchy sequence is Cauchy.
Proof. Let {sn } be a convergent sequence, with limit L, and let {skn } be a subsequence.
Let > 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.
Theorem 8.8.4 Every bounded sequence has a Cauchy subsequence.
Proof. This proof uses the halving construction already encountered in Construction 3.4.1,
and here the property P is that the subset contains infinitely many elements of the sequence.
Let {sn } be a bounded sequence (of real or complex numbers). Let M be a positive
real number such that for all n, |sn | M . Let a0 = c0 = M and b0 = d0 = M . The
sequence {sn } has infinitely many (all) terms in the rectangle R0 = [a0 , b0 ] [c0 , d0 ]. Set
l0 = 0. (If all sn are real, we may take c0 = d0 = 0, or perhaps better, ignore the second
coordinates.)
We prove below that for all m N+ there exists a subsequence {skn } all of whose
terms are in the rectangle Rm = [am , bm ] [cm , dm ], where bm am = 2m (b0 a0 ),
[am , bm ] [am1 , bm1 ]. dm cm = 2m (d0 c0 ), [cm , dm ] [cm1 , dm1 ]. Furthermore,
we prove that there exists lm > lm1 such that slm Rm .
Namely, given the (m 1)st rectangle Rm1 , integer lm1 such that slm1 Rm1 ,
and a subsequence {skn } all of whose terms are in Rm1 , divide Rm1 into four equal-sized
subrectangles. Necessarily at least one of these four subrectangles contains infinitely many
elements of {skn }, so pick one such subrectangle, and call it Rm . Therefore there exists
a subsequence of {skn } that is contained in Rm , and that subsequence of {skn } is also a
subsequence of {sn }. We call it {skn }. Since we have infinitely many kn , in particular
there exists kn > lm1 , and we set lm = kn . Thus slm Rm .
By construction, {sln }n is a subsequence of {sn }n . We next prove that {sln }n is a
Cauchy sequence. Let > 0. Since the either side length of the mth subrectangle Rm
equals the corresponding side length of R0 divided by 2m , by Exercise 2.7.20 there exists
a positive integer N such that any side length of RN is strictly smaller than the constant
/2. Let m, n > N be integers. Then slm , sln are in RN , so that

p
|slm sln | (one side length of Rn )2 + (other side length of Rn )2 < .
The following is now an immediate consequence of Theorem 8.7.5:
Theorem 8.8.5 Every bounded sequence in C has a convergent subsequence.
Example 8.8.6 We work out the construction of a subsequence as in the proof on the
bounded sequence {(1)n 1}. For example, all terms lie on the interval [a0 , b0 ] = [4, 4].
Infinitely many terms lie on [a1 , b1 ] = [4, 0], and on this subinterval I arbitrarily choose the
second term, which equals 0. Infinitely many terms lie on [a2 , b2 ] = [4, 2], in particular,
I choose the third term 2. After this all terms of the sequence in [a2 , b2 ] are 2, so that
we have built the Cauchy subsequence {0, 2, 2, 2, . . .} (and subsequent [an , bn ] all have
bn = 2). We could have built the Cauchy subsequence {2, 2, . . .}, or, if we started
with the interval [8, 8], we could have built the Cauchy subsequences {0, 0, 2, 2, . . .}
or {2, 0, 0, 2, 2, . . .}, and so on.
Definition 8.8.7 A subsequential limit of a sequence {sn } is a limit of any subsequence

of {sn }. Thus a subsequential limit can be a complex number, or as in Definition 8.3.4, it
can be .
Here are a few examples in tabular form:
sequence {sn } the set of subsequential limits of {sn }

convergent {sn } {lim sn }
{(1)n} {1, 1}
{(1)n 1} {0, 2}
{(1)n + 1/n} {1, 1}
{(1)nn + n + 3} {3, }
{in } {i, 1, i, 1}
{n} {}
{(1)nn} {, }
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 = .

8.8.1 Find the subsequential limit sets of the following sequences, and justify:

i) {(1/2 + i 3/2)n }, {(1/2 i 3/2)n }.
ii) {1/2n }.

iii) {(1/2 + i 3/2)n + 1/2n }.

iv) {(1/2 + i 3/2)n 1/2n}.

v) {(1/2 + i 3/2)n + (1/2 i 3/2)n }.

vi) {(1/2 + i 3/2)n (1/2 i 3/2)n}.
8.8.2 Consider the sequence enumerating Q+ as on page 226. Prove that the set of its
subsequential limits equals R0 .
n 2 o
n +(1)n n(n+1)
8.8.3 Prove that the real-valued sequence n+1 is unbounded. Find a subse-
quence that diverges to and a subsequence that converges to 1.
8.8.4 Let m be a positive integer, and let c be the complex number of absolute value 1
at angle 2/m counterclockwise from the positive real axis. Find the set of subsequential
limits of {cn }n .
8.8.5 Give examples of sequences with the listed properties, if they exist. If they do not
exist, justify.
i) A sequence with no convergent subsequences.
ii) A sequence whose set of subsequential limits equals {1}.
iii) A sequence whose set of subsequential limits equals {1, 3}.
iv) A sequence whose set of subsequential limits equals {1, 2, 12}.
8.8.6 Suppose that a Cauchy sequence has a convergent subsequence. Prove that the
original sequence is convergent as well.
8.8.7 Prove that a sequence of real numbers contains a monotone subsequence.
8.9 Liminf, limsup for real-valued sequences
(I do not go into great detail in this section.)

Recall that by Axiom 2.9.1, every subset T of R bounded above has a least upper
bound sup T = lub T in R, and that every subset T bounded below has a greatest lower
bound inf T = inf T in R. We extend this definition by declaring
sup T = if T is not bounded above,

inf T = if T is not bounded below.
The same definitions apply to sequences thought of as sets: sup{1/n}n = 1, inf{1/n}n = 0,
sup{n}n = , inf{n}n = 1, sup{(1)n}n = 1, inf{(1)n }n = 1, et cetera.
Much analysis of sequences has to do with their long-term behavior rather than with
their first three, first hundred, or first million terms think convergence or the Cauchy
property of sequences. We will see further usage of such tail-end analysis in the next
chapter (for convergence criteria for series). Partly with this goal in mind, we apply infima
and suprema to sequences of tail ends of sequences:
Definition 8.9.1 Let {sn } be a real-valued sequence. The limit superior lim sup and
limit inferior lim inf of {sn } are:
lim sup sn = inf{sup{sn : n m} : m 1},

lim inf sn = sup{inf{sn : n m} : m 1}.
In other words, lim sup sn is the infimum of the set of all the suprema of all the tail-end
subsequences of {sn }, and analogously, lim inf sn is the supremum of the set of all the infima
of all the tail-end subsequences of {sn }. In the plot below, the sequence {sn }, drawn with
dots, oscillates between positive and negative values but with peaks and valleys getting
smaller and smaller. The connected red dots denote the sequence {sup{sm : m n}}n and
the connected blue dots represent 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 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 (Axiom 2.9.1), 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.
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 2.10.4 that L = lim sup sn . The other part
is left to the reader.
Remark 8.9.3 (Ratio test for sequences) With the new language, Theorem 8.6.5 can
be rephrased as follows: If {sn } is a sequence of non-zero complex numbers such that
lim sup{|sn+1 /sn |} < 1, then lim sn = 0. The proof there already accomplishes this. On
the other hand, the ratio test for divergence in Exercise 8.6.6 is not phrased in the most
general form. One generalization is that if lim sup{|sn+1 /sn |} > 1, then {sn } diverges. The
proof is simple. Let r (1, lim inf{|sn+1 /sn |}). By definition of liminf as supremum of some
infima, this means that there exists an integer m such that inf{|sn+1 /sn | : n m} > r.
Thus by an easy induction, for all n > m, |sn | > r nm |sm |, and then by the Comparison
test (Theorem 8.3.7), {|sn |} diverges to infinity, hence {sn } does not converge to a complex
number.
It turns out that there is an important connection between limsup, liminf, and subse-
quential limits:
Theorem 8.9.4 Let {sn } be a bounded sequence of real numbers. Then the supremum
of the set of all subsequential limits equals lim sup sn , and the infimum of the set of all
subsequential limits equals lim inf sn .
Proof. Proof of the limsup part only: Let A = lim sup{sn }, let S be the set of all subse-
quential limits of {sn }, and let U = sup(S). Since the sequence is bounded, A and U are
real numbers.
Let > 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 2.10.4 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 2.10.4 this means
that A U .
It follows that A = U .
Theorem 8.9.5 Let {sn }, {tn} be bounded sequences in R. Then lim sup sn +lim sup tn
lim sup(sn + tn ) and lim inf sn + lim inf tn lim inf(sn + tn ).
Proof. Let a = lim sup sn , b = lim sup tn , c = lim sup(sn + tn ). By boundedness, a, b, c R.

Let > 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
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 8.9.6 Let {sn } and {tn } be sequences of non-negative real numbers such that
lim sn is a positive real number L. Then lim sup(sn tn ) = L lim sup tn and lim inf(sn tn ) =
L lim inf tn .
Proof. Let > 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 Let {sn } and {tn } be sequences of real numbers such that lim sn is a positive real
number L. Prove that lim sup(sn +tn ) = L+lim sup tn and lim inf(sn +tn ) = L+lim inf tn .
8.9.2 Compute lim inf and lim sup for the following sequences. Justify your work.
i) { sinn n }.
ii) {(1)n n!}.
iii) {2n }.
iv) {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}.
v) {(4 + n1 )(1)n }.
vi) The sequence of all positive prime numbers.
vii) The sequence of all multiplicative inverses of positive prime numbers.
8.9.3 Prove the following for any sequence {sn }:
i) inf(sn ) = sup(sn ).
ii) lim inf(sn ) = lim sup(sn ).
8.9.4 Suppose that {sn } converges to L. Finish the proof of Theorem 8.9.2, namely prove
that lim inf sn = L.
8.9.5 For every positive integer n, let s2n1 = 1 and s2n = 1/2n .
i) Compute lim inf sn , lim inf sn .
ii) Prove that lim sup sn+1
sn
> 1.
iii) Does this contradict the ratio test as in Remark 8.9.3.
8.9.6 Find bounded real-valued sequences {sn }, {tn} such that lim sup sn + lim sup tn >
lim sup(sn + tn ). (Compare with Theorem 8.9.5.)
8.9.7 Let {sn }, {tn } be bounded sequences in R.
i) Finish the proof of Theorem 8.9.5, namely prove that lim inf sn + lim inf tn
lim inf(sn + tn ).
ii) Find such {sn }, {tn} so that lim inf sn + lim inf tn < lim inf(sn + tn ).
8.9.8 Suppose that lim sn = . Prove that the set of subsequential limits of {sn } is
empty.
8.9.9 Finish the proof of Theorem 8.9.4, namely prove that the infimum of the set of all
subsequential limits of a bounded sequence equals the liminf of the sequence.
8.9.10 Let {sn } be a sequence of positive real numbers. Prove that lim sup s1n = 1
lim inf sn .
Chapter 9: Infinite series and power series
In this section we will handle (some) infinite sums, and in particular functions that
arise as infinite sums of higher and higher powers of a variable x. The culmination of
the chapter and the course are the exponential and trigonometric function, with their
properties.
Warning: Finite sums are possible by the field axioms, but infinite sums need not
make any sense at all. For example,
1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1) + 1 + (1)

may be taken to be 0 or 1 depending on which consecutive pairs are grouped together in
a sum, or it could even be taken to be 3 by taking the first three positive 1s, and then
matching each successive 1 in the sum with the next not-yet-used +1. In this way each
1 in the expression is used exactly once, so that the sum can indeed be taken to be 3.
Similarly, we can make the limit be 4, 17, et cetera.
This should convince you in addition that in infinite sums, the order of addition
matters! For more on the order of addition, see Exercises 9.2.15 and 9.2.16.
Infinite sums require special handling, but limits of sequences prepared the ground for
that.
9.1 Infinite series
Definition 9.1.1 For an infinite sequence {an } of complex numbers, define the corre-
sponding sequence of partial sums
{a1 , a1 + a2 , a1 + a2 + a3 , a1 + a2 + a3 + a4 , . . .}.
Pn
We denote the nth term of this sequence sn = k=1 ak . The (infinite) series corre-
P
sponding to the sequence {an } is k=1 ak (whether this infinite sum makes sense or
not).
P P
When the range of indices is clear, we write simply k ak or ak .
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 par-
n
ticular, the sequence of partial sums of {(1)n} is {{ (1)2 1 }n = {1, 0, 1, 0, 1, 0, . . .}.
We have encountered shifted sequences, such as {an }nm , and similarly there are
P
shifted series: k=m ak stands for the limit of the sequence of partials sums, but in this
case, the nth partial sum is sn = am + am+1 + am+2 + + am+n1 .
P
Definition 9.1.3 (Most of the time and by default we take m = 1.) The series k=m ak
Pnm+1
converges to L C if the sequence { k=m ak }n converges to L. We say then that L
P
is the sum of the series and we write k=m ak = L.
If the series does not converge, it diverges.
Just like for sequences, when a series diverges, it may diverge to or to , or it

may simply have no limit.
Since a sequence {sn } converges if and only if {sn + c} converges (where c is any
P P
constant), it follows that k=1 ak converges if and only if k=m ak converges, and then

X
X
ak = a1 + a2 + + am1 + ak .
k=1 k=m
The following follows immediately from the corresponding results for sequences:
P P
Theorem 9.1.4 Let A = k=1 ak , B = k=1 bk , and c C.
(1) If A, B C, then
X
(ak + cbk ) = A + cB.
k=1
(2) If all ak , bk are real numbers and A, B R {, }, then

, if A = , B R {} and c 0;

, if A = , B R {} and c 0;

, if A = , B R {} and c 0;

X , if A = , B R {} and c 0;
(ak + cbk ) = , if A R, B = and c > 0;

k=1
, if A R, B = and c < 0;

, if A R, B = and c < 0;

,
if A R, B = and c > 0;
A, if c = 0.
Pn Pn
Proof. (1) By assumption, the sequences { k=1 ak } and { k=1 bk } converge to A and B
in C, respectively. By the theorem on the convergence of sums of sequences (Theorem 8.4.3)
Pn Pn Pn
then k=1 (ak + cbk ) = k=1 ak + c k=1 bk converges to A + cB.
Other parts are proved similarly.
266 Chapter 9: Infinite series and power series
Remark 9.1.5 This theorem justifies the binary operation of addition on the set of of
convergent infinite series:

!
!
X X X
ak + bk = (ak + bk ).
k=1 k=1 k=1
It is hard to immediately present examples of this because we know so few limits of

infinite series. There are examples in the exercises.
P
Theorem 9.1.6 If r C satisfies |r| < 1, then the geometric series k=1 r k1 converges
1
P r
to 1r , so k=1 r k = 1r .
Pn n
Proof. By Example 1.5.4, k=1 r k1 = 1r n
1r . By Corollary 8.6.6, lim r = 0. Thus by the
scalar and sum rules for limits of sequences (Theorem 8.5.6 or Theorem 8.4.3),
Xn
k1 1 rn 1 rn 1
lim r = lim = lim = .
n n 1 r n 1 r 1r 1r
k=1
Pn Pn r
Thus limn k=1 r k = r limn k=1 r k1 = 1r .
In particular, the familiar decimal expansion 0.33333 of 1/3 can be thought of as
P
the infinite sum k=1 103k . The sequence of its partial sums is {0.3, 0.33, 0.333, 0.3333, . . .},
and by the theorem above,
X k1
3 3 X 1 3 1 3 10 1
k
= = 1 = = .
10 10 10 10 1 10 10 9 3
k=1 k=1
A little more work is expressing 5.523 = 5.523523523523 (repeating 523) as a fraction:

X
523 523 523 1
5.523 = 5 + + + + = 5 + 523
1000 10002 10003 1000k
k=1
1
1000 1 523 4995 + 523
= 5 + 523 1 = 5 + 523 =5+ =
1 1000 1000 1 999 999
5518
= .
999
P 1
Example 9.1.7 The harmonic series k=1 k diverges to by Example 8.7.6.
P 1
Example 9.1.8 The series k=1 k2 converges.
Proof #1: By Exercise 1.5.20 and Theorem 9.1.6,

n
2X 1 n+1
X 1
X
1 1 1
0 2
k1
k1
= = 2,
k 2 2 1 1/2
k=1 k=1 k=1
P2n 1 Pn
so that { k=1 k12 }n and { k=1 k12 }n are bounded increasing sequences of real numbers.
P
By Theorem 8.6.4, the sequence has a limit that is at most 2, so that k=1 k12 converges.
Pn
Proof #2: Since k 2 > k(k 1), we have that for k > 2, k12 < k(k1) 1
. Thus k=1 k12 =
Pn Pn P P
1 + k=2 k12 < 1 + k=2 k(k1) 1 1
. By Exercise 9.1.3, the series k=2 k(k1) 1
= k=1 k(k+1)
Pn
converges, so that the increasing sequence { k=1 k12 } of partial sums is bounded above,
P
which means 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
the proof is in Exercise 9.12.5.)
P
Theorem 9.1.9 If k=1 ak converges, then lim an = 0, and the sequence {an } is bounded.
P
Proof. By assumption, k=1 ak converges, so that by definition, the sequence {sn } of
partial sums converges, and is thus Cauchy. In particular, for every > 0 there exists
N > 0 such that for all integers n > N , |an+1 | = |sn+1 sn | < . Thus by Theorem 8.6.2,
{an } is bounded.
The converse of this theorem is of course false; see Example 9.1.7.
Theorem 9.1.10 Let {an } be a sequence of non-negative real numbers. If the sequence
P
{a1 + a2 + + an } of partial sums is bounded above, then an converges.
Proof. The sequence {a1 + a2 + + an } of partial sums is monotone and bounded above,
so it converges by Theorem 8.6.4.
Theorem 9.1.11 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 8.5.6 or Theorem 8.4.3), 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=1r k diverges.
P 1 P 1 P 2
9.1.2 Compute and justify the following sums: k=1 2k , k=6 3k , k=8 5k .
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.)
9.1.5 Let an = (1)n . Prove that the sequence of partial sums {a1 + a2 + + an } is
bounded but does not converge. How does this not contradict Theorem 9.1.10?
9.1.6 For each k N+ let xk be an integer between 0 and 9.
P xk
i) Prove that k=1 10 k converges.
ii) What does this say about decimals?

iii) Find the sum if xk = 4 for all k.
iv) Find the sum if {xn } = {1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, . . .}.
P xk
(v)* Prove that whenever the sequence {xn } is eventually periodic, then k=1 10 k is
a rational number.
P P P
9.1.7 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! .
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. Note that all an are non-negative, and for all integers n m,
|bm+1 + bm+2 + + bn | |bm+1 | + |bm+2 | + + |bn |
am+1 + am+2 + + an
= |am+1 + am+2 + + an |.
P
Since ak converges, by Theorem 9.2.1, for every > 0 there exists N > 0 such that for all
integers n > m > N , |am+1 + am+2 + + an | < , and hence |bm+1 + bm+2 + + bn | < .
P
Thus again by Theorem 9.2.1, bk converges.
Theorem 9.2.3 (Comparison test (for series)) Let {an } be a complex and {bn } a real
P P
sequence. If ak diverges and if for all n, bn |an |, then bk diverges.
Proof. For all integers n m, |bm+1 + bm+2 + + bn | |am+1 | + |am+2 | + + |an |

P
|am+1 +am+2 + +an |. Since k ak is divergent, by Theorem 9.2.1, there exists > 0 such
that for all N > 0 there exist integers n > m > N , such that |am+1 + am+2 + + an | ,
P
and hence |bm+1 + bm+2 + + bn | . Thus again by Theorem 9.2.1, bk is divergent.
Theorem 9.2.4 (Ratio test) (Compare to Remark 8.9.3.) Let {an } be a sequence of
non-zero complex numbers.
P P
(1) If lim sup aan+1 < 1, then |ak | and ak converge.
n
P P

(2) If lim inf aan+1
n
> 1, then |ak | and ak diverge.
Proof. Let L = lim sup | aan+1

n
|. Suppose that L < 1. Let r be a real number in the
open interval (L, 1). Since L = inf{sup{ |a|an+1
n|
|
: n m} : m 1} and r > L, it fol-
lows that there exists m 1 such that r > sup{ |a|an+1n|
|
: n m}. Thus for all n m,
P
n
|an+1 | < r|an |. Thus by Exercise 1.5.22, |am+n | < r |am |. The geometric series k r k
P
converges by Theorem 9.1.6, and by Theorem 9.1.4, k am r k converges. Thus by Theo-
P P P
rem 9.2.2, k=1 |am+k | and k=1 am+k converge. Hence by Theorem 9.1.11, |ak | and
P
ak converge. This proves (1).
Now let L = lim inf | aan+1
n
|, and suppose that L > 1. Let r be a real number in the
open interval (1, L). Since L = sup{inf{ |a|an+1
n|
|
: n m} : m 1} and r < L, it follows that
there exists m 1 such that r < inf{ |a|an+1
n|
|
: n m}. Thus for all n m, |an+1 | > r|an |.
Thus by a straightforward modification of Exercise 1.5.22, |am+n | > r n |am |. The geometric
P P
series k r k diverges by Exercise 9.1.1, and by Theorem 9.1.4, k am r k diverges. Thus by
P P P
Theorem 9.2.2, k=1 |am+k | and k=1 am+k diverge. Hence by Theorem 9.1.11, |ak |
P
and ak diverge. This proves (2).
This ratio test for convergence of series does not apply when lim sup | aan+1 | = 1 or
an+1 Pn
lim inf | an | = 1. The reason is that under these assumptions the series k |ak | and
P
k ak sometimes converge and sometimes Pn 1
diverge. For example, if an = 1/n for all n,
an+1 an+1
lim sup | an | = lim inf | an | = 1, and k=1 k diverges; whereas if an = 1/n2 for all n,
Pn
then lim sup | aan+1
n
| = lim inf | an+1
an | = 1, and 1
k=1 k2 converges.
Theorem 9.2.5 (Root test for series) Let {an } be a sequence of complex numbers.
Let L = lim sup |an |1/n .
P P
(1) If L < 1, then k |ak |, k ak converge.
P P
(2) If L > 1, then k |ak |, k ak diverge.
Proof. If L < 1, choose r (L, 1). Since L = inf{sup{|an |1/n : n m} : m 1} and r > L,
there exists m 1 such that r > sup{|an |1/n : n m}. Thus for all n m, r n |an |.
P k
Thus by the Comparison test (Theorem 9.2.2), since the geometric series r converges,
P P
we have that ak and |ak | converge. The proof of (2) is similar, and is omitted here.
Theorem 9.2.6 (Alternating series test) If {an } is a sequence of positive real numbers
P
such that lim an = 0 and a1 a2 a3 , then k
k=1 (1) 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 k
Thus by Cauchys criterion Theorem 9.2.1, k (1) ak converges.
P
Example 9.2.7 Recall from Example 9.1.7 that the harmonic series k 1/k diverges.
P P
k
But the alternating series k (1) /k converges by this theorem. (In fact, k (1)k /k
converges to ln 2, but proving the limit is harder see the proof after Example 9.7.4.
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 n n
P
Suppose that k f (k) converges. Then by the definition limn (f (1) + f (2) + +
R n+1 R2 R3 R n+1
f (n)) exists. By the displayed inequalities, 1 f = 1 f + 2 f + + n f f (1) +
R n+1
f (2) + + f (n), so that { 1 f }n is a bounded increasing sequence of real numbers, so
R n+1 Rn
that limn 1 f exists, and hence that limn 1 f exists.
Rn
Conversely, suppose that limn 1 f exists. Let L R be this limit. By the dis-
R3 R n+1 R n+1
played inequalities, f (2) + f (3) + + f (n + 1) 2 f + + n f = 2 f . Since f
takes on only non-negative values, this says that f (2) + f (3) + + f (n + 1) L. Thus
{f (2) + + f (n + 1)}n is a non-decreasing sequence that is bounded above by L. Thus
by Theorem 8.6.4, this sequence converges. By adding the constant f (1), the sequence
P
{f (1) + + f (n + 1)}n converges, so that by the definition of series, k f (k) converges.
Theorem 9.2.9 (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 p+1
1
f exists. By the Fundamental Theorem of Calculus, 1
f = 1
x dx = n p+11 . By the
composite rule for sequences (either Theorem 8.5.8 or Theorem 8.4.7), since the function
that exponentiates by the positive (p +1) is continuous at all real numbers and lim n1 = 0,
(p+1) Rn
it follows that lim np+1 = lim n1 = 0(p+1) = 0, so that lim 1 f exists and equals
1
P
p+1
. Thus by the Integral test (Theorem 9.2.8), k k p converges.

9.2.1 For each of the following series, determine, with proof, whether they converge or
diverge. You may need to use Examples 8.2.8 and 8.3.6.
X
3i
i) .
k=1
k + k4
2
X
1
ii) .
k=1
k

X 1
iii) .
k3
k=1
X
2k
iv) .
k!
k=1
X
2k
v) .
k3
k=1
X
1
vi) .
k=1
k 2
P 1
P (1)k P 1 P (1)k
9.2.2 Prove that k=0 (2k+1) 2, k=0 (2k+1)2 , k=1 k2 and k=1 k2 all converge.
9.2.3 Prove that the following statements are equivalent. They are also all true, but you
do not have to prove that.
P 1 2
(1) k=0 (2k+1)2 = 8 .
P 1 2
(2) k=1 k2 = 6 .
P (1)k 2
(3) k=0 (2k+1)2 = 16 .
P (1)k 2
(4) k=1 k2 = 12 .
P (1)k P
(Hint for a part: Write out the first few summands of k=1 k2 + 12 k=1 k12
P k P
= k=1 (1) k 2 + 2 1
k=1 (2k)2 .)
P P
9.2.4 Find a convergent series k ak and a divergent series k bk with lim sup |an |1/n =
lim sup |bn |1/n = 1.
9.2.5 Make a list of all encountered criteria of convergence for series.
9.2.6 The goal of this exercise is to show that if the ratio test (Theorem 9.2.4) determines
the convergence/divergence of a series, then the root test (Theorem 9.2.5) determines it as
well. Let {an } be a sequence of non-zero complex numbers.
i) Suppose that lim sup | aan+1
n
| < 1. Prove that lim sup |an |1/n < 1.
ii) Suppose that lim inf | aan+1
n
| > 1. Prove that lim sup |an |1/n > 1.
P k
9.2.7 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
9.2.8 Let {an } be a complex sequence, and let c C. Is it true that k=1 ak converges
P
if and only if k=1 cak converges? If true, prove; if false, give a counterexample.
P
9.2.9 Let an = (1)n /n, bn = 2(1)n. Then k=1 ak converges, and for all n, |bn | = 2,
P
and k=1 abkk diverges.
P
9.2.10 (Compare with Exercise 9.2.9.) Suppose that k=1 |ak | converges. Let {bn } be a
P
sequence of complex numbers such that for all n, |bn | > 1. Prove that k=1 abkk converges.
9.2.11 (Summation by parts) Let {an }, {bn} be complex sequences. Prove that
n
X n
X n
X k
X
ak bk = an bk (ak+1 ak ) bj .
k=1 k=1 k=1 j=1
(Hint: Set a0 = b0 = 0. Let f, g : [1, ) C be defined by f on [n, n + 1) being the

constant function an an1 and g on [n, n + 1) being the constant function bn . Both f and
R n+1 R n+1 Pn
g are piecewise continuous, F (n) = 1 f = an , G(n) = 1 g = k=1 bk . The problem
R R
should remind you of integration by parts (F g) = F G (f G).)
P xk
9.2.12 Let x C with |x| = 1 and x 6= 1. Prove that k=1 k converges. (Hint:
Exercise 9.2.11, Example 1.5.4.)
P P
9.2.13 Prove that if k=1 ak converges then k=1 akk converges. (Hint: Exercise 9.2.11.)
9.2.14 The 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+1 np+1 np+1 (n 1)p+1
(n + 1)p < and np < .
p+1 p+1
iii) Prove by induction on n 1 that
n
X 1 p
kp np+1 + .
p+1 p+1
k=1
Pn
iv) Prove that the positive sequence { k=1 k p }n of partial sums is an increasing
sequence bounded above.
Pn P
v) Prove that the sequence { k=1 k p }n converges, and so that k=1 k p converges.
9.2.15 (Order of summation in infinite sums is important.)
P 1 P 1
i) Prove that k=1 2k and k=1 2k+1 diverge. Refer to Example 9.2.7.
Pn P n P2n+1 k
ii) Observe that k=1 2k1
k=1 2k+11
= k=1 (1) k .
P (1)k P 1 P 1
iii) Argue that k=1 k 6= k=1 2k k=1 2k+1 . Why does this not contradict
the expected summation and difference rules?
P
9.2.16 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 9.2.15, and comment what the difference is.
9.2.17 (Raabes test) Let a1 , a2 , . . . be positive real numbers such that for some > 1
P
and for some N N, aan+1
n
1
n for all n N . Prove that n an converges. (Hint: Let
f (x) = 1 x . Use the Mean Value Theorem to get c (x, 1) such that f (c)(1 x) =

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
In this section we deal with sums where the index varies through N0 , and furthermore,
the terms of the sequence are special functions rather than constants:
Definition 9.3.1 A power series is an infinite series of the form

X X
ak xk = ak xk = a0 + a1 x + a2 x2 + a3 x3 + ,
k=0 k0
where a0 , a1 , a2 , . . . are fixed complex numbers, and x is a variable that can be replaced by
any complex number. (By convention as on page 35, 00 = 1.)
Note that the partial sums of power series are polynomials, so a power series is a limit
of polynomials.
So a power series is a function of x, and the domain is to be determined. Clearly, 0
P
is in the domain of every power series: plugging in x = 0 returns k=0 ak 0k = a0 . If all
except finitely many an are 0, then the power series is actually a polynomial, and is thus
defined on all of C. Or course, when 1 is in the domain, evaluation of the power series
P k
P
k ak x at x = 1 is the (ordinary) series k ak .
The most important question that we address in this section is: which x are in the
domain of the power series, i.e., for which x does such an infinite sum make sense. We
prove that for every power series whose domain is not all of C there exists a non-negative
real number R such that the series converges for all x C with |x| < R and the series
diverges for all x C with |x| > R. What happens at x with |x| = R depends on the series.
Think through the following table of reading off the (fixed) coefficients an of the
variable power xn :
power series an
X
xk an = 1
k=0
X
2k+1 3 5 7 1, if n is odd;
x = x+x +x +x + an =
0, if n is even.
k=0
(
X 0, if n is odd;
x2k+12 = x12 + x14 + x16 + x18 + an = 0, if n is even and n < 12;
k=0 1, if n 12 is even.
X
kxk an = n
k=0
X
xk 1
an = n!
k!
k=0
X
(kx)k an = nn
k=0
P
Example 9.3.2 Let f (x) = k=0 xk . By Theorem 9.1.6, the domain of f contains all
complex numbers with absolute value strictly smaller than 1, and by Theorem 9.1.9, the
domain of f contains no other numbers, so that the domain equals {x C : |x| < 1}.
P
Moreover, for all x in the domain of f , by Theorem 9.1.6, k 1
k=0 x = 1x . Note that the
1
domain of 1x is strictly larger than the domain of f .
Whereas for general power series it is impossible to get a true numerical infinite sum,
P P
for geometric series this is easy: f ( 21 ) = k0 21k = 11 1 = 2, f ( 31 ) = k0 31k = 11 1 = 32 ,
P 1
2 3
f (0.6) = k0 0.6k = 10.6 = 2.5.
P
Theorem 9.3.3 (Root test for the convergence of power series) Let ak xk be a
power series, and let = 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
|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 = ,
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 con-
verges and outside of the circle the series diverges. Whether the power series converges at
points on the circle depends on the series; see Example 9.3.6.
P P P P P
Theorem 9.3.5 The series ak xk , |ak |xk , |ak ||x|k , kak |x|k , kak xk1 ,
P 2 P
k ak xk , k(k 1)ak xk , all have the same radius of convergence.
p
Proof. By Example 8.2.8, lim n |n| = 1. For any integer n 2, 1 n n 1 n n, so that
p
by the squeeze theorem, lim n n 1 = 1. Thus by Theorem 8.9.6, lim sup n |n(n 1)an | =
p p p
lim n n(n 1)lim sup n |an | = lim sup n |an |. This proves that the as in Theorem 9.3.3
P P
for ak xk is the same as the for k(k 1)ak xk , which proves that these two power
series have the same radius of convergence. The proofs of the other parts are similar.
P k
Example 9.3.6 We have seen that x has radius of convergence 1. By Theorem 9.1.9,
this series does not converge at any point on the unit circle. By the previous theorem,
P P
the radius of convergence of k=1 k1 xk and of k=1 k12 xk is also 1. By the p-series test
P
(Theorem 9.2.9) or by the harmonic series fact, k=1 k1 xk diverges at x = 1, and by the
P
alternating series test (Theorem 9.2.6), k=1 k1 xk converges at x = 1. By the p-series
P
test (Theorem 9.2.9), k=1 k12 converges, so that by the comparison test Theorem 9.2.2,
P 1 k
k=1 k2 x converges on the unit circle.
P
Example 9.3.7 Let the radius of convergence for ak xk be R. Then the radius of
P
convergence for ak x2k is if R = and it is R otherwise.
p
Proof. Let = lim sup n |an |. In the second power series, the nth coefficient is 0 if n is
odd, and it is an/2 xn if n is even. Then applying the root test to this power series gives
n p p p p o
1 3 5 7
lim sup 0, 2 |a1 |, 0, 4 |a2 |, 0, 6 |a3 |, 0, 8 |a4 |, . . .
np p p p o
2 4 6 8
= lim sup |a1 |, |a2 |, |a3 |, |a4 |, . . .
(0s do not contribute to limsup of non-negative numbers)
q qp qp qp
p
1 2 3 4
= lim sup |a1 |, |a2 |, |a3 |, |a4 |, . . .
r np o
1
p p p
= lim sup |a1 |, 2 |a2 |, 3 |a3 |, 4 |a4 |, . . .

= ,
and the conclusion follows.
P P
Example 9.3.8 Similarly to the last example and by Theorem 9.3.5, xk , kxk ,
P 2 k P P 2k+1
k x , k(k 1)xk , x all have radius of convergence 1.
Theorem 9.3.9 (Ratio test for the convergence of power series) Suppose that all
an are non-zero complex numbers.
P P
(1) If |x| < lim inf aan+1
n
, then |ak ||x|k and ak xk converge.
a P P
(2) If |x| > lim sup an+1 n
, then |ak ||x|k and ak xk diverge.
a P
Thus if lim n exists, it equals the radius of convergence of
an+1 ak xk .
Warning: Compare with the Ratio test for convergence of series (Theorem 9.2.4) where
fractions are different. Explain to yourself why that is necessarily so, possibly after going
through the proof below.
Proof. The two series converge in case x = 0, so that we may assume that x 6= 0. We may
then apply the Ratio test for convergence of series (Theorem 9.2.4):

an+1 xn+1 an+1
lim sup = |x| lim sup
an xn an

an+1
= |x| inf{sup : n m : m 1}
an
( )
1
= |x| inf :m1
n
inf{ aan+1 : n m}
|x|
= n a o
n
sup inf{ an+1 : n m} : m 1
|x|
= .
lim inf aan+1
n
If this is strictly smaller than 1, then the two series converge, which proves (1). Similarly,

an+1 xn+1 |x|

lim inf = ,
an x n lim sup aan
n+1
and if this is strictly larger than 1, then the two series diverge. The last part is then
immediate by the definition of radius of convergence.
Examples 9.3.10
P 2k+1
(1) In Example 9.3.8 we established via the root test that x has radius of
convergence 1. The ratio test is inapplicable for this power series. However, note
P 2k+1 P
that x = x (x2 )k , and by the ratio test for series (not power series), this
series converges for non-zero x if lim sup |(x2 )k+1 /(x2 )k | < 1, i.e., if |x2 | < 1, i.e.,
if |x| < 1, and it diverges if |x| > 1.
P k
(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.
P k 1
(3) By the ratio test, the radius of convergence of k=1 xk! is lim n!1 = lim (n+1)! n! =
(n+1)!
lim(n + 1) = . The root test gives = lim sup |1/n!| | = lim sup(1/n!)1/n, and
1/n
by Example 8.3.6 this is 0. Thus the radius of convergence is also by the root
test.

P
9.3.1 Let the radius of convergence of k
k=0 ak x 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
9.3.2 Suppose that k ak converges for some ak C.
P
i) Prove that the function defined as the power series k ak xk has radius of conver-
gence at least 1.
P
ii) Give an example of ak C for which the radius of convergence of k=0 ak xk is
strictly greater than 1. (Bonus points for easiest example.)
P
iii) Give an example of ak C for which the radius of convergence of k=0 ak xk is
equal to 1.
9.3.3 Compare with the previous exercise: find the radii of convergence of the power series
P k P k
P k
P k P k
P k P k
k=0 x , k=0 (1)x , k=0 (2x) , k=0 x + k=0 (1)x , k=0 x + k=0 (2x) .
9.3.4 Compute and justify the radius of convergence for the following series:
P k
i) 3x .
P
ii) (3x)k .
P
iii) 3kxk .
P
iv) k(3x)k .
P xk
v) k3
.
P 3x k
vi) k3 .
P (3x)k
vii) k3 .
9.3.5 Compute and justify the radius of convergence for the following series:
P xk
i) .
P kxkk
ii) .
P k2k
xk
iii) (2k)k
.
P P
9.3.6 Let k=0 ak xk and k
k=0 bk x be convergent power series with radii of convergence
P P
R1 and R2 , respectively. Let R = min{R1 , R2 }. Prove that k=0 ak xk k=0 bk xk =
P k
k=0 (ak bk )x is convergent with radius of convergence at least R. (Hint: Theorem 9.1.4.)
P
9.3.7 What would be a sensible definition for generalized power series k=0 ak (x a)k ?
P
What would be a sensible definition of the radius of convergence of k=0 ak (x a)k ? Draw
a relevant picture in C.
9.4 Differentiation of power series
Power series are functions. In this section we prove that they are differentiable at all
x inside the circle of convergence. Since a differentiable function is continuous, it follows
that a power series is continuous inside the circle of convergence.
Recall that for any differentiable function f , f (x) = limh0 f (x+h)f (x)
h P , and any
P P n
power series k=0 ak xk is actually the limit of a sequence: k=0 ak xk = lim{ k=0 ak xk }n .
Thus Pn Pn
X k k
lim{ ak (x + h) } n lim{ k=0 ak x }n
( ak xk ) = lim k=0
.
h0 h
k=0
P P
Certainly by the sum rule for convergent series, lim{ nk=0 ak (x+h)k }n lim{ nk=0 ak xk }n =
Pn
lim{ k=0 ak ((x + h)k xk )}n , and by the constant rule we get that

X n
X
k (x + h)k xk
( ak x ) = lim lim ak .
h0 n h
k=0 k=0
If we could change the order of limits, then we would get by the polynomial rule for
derivatives that

X Xn
k
( ak x ) = lim lim kak xk1 .
n h0
k=0 k=0
In fact, it turns out that this is the correct derivative, but our reasoning above was based
on an unproven (and generally false) switch of the two limits.
We give a correct proof of derivatives in the rest of the section. By Theorem 9.3.5
P P P
we already know that the series k=0 ak xk and k=0 kak xk1 = k=1 kak xk1 have the
same radius of convergence.
The following theorem is not necessarily interesting in its own right, but it is a stepping
stone in the proof of derivatives of power series.
P
Theorem 9.4.1 Let k ak xk have radius convergence R. Let c C satisfy |c| < R. Then
P
the function g(x) = k=1 ak (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, k=1 kak x converges on B(0, R) by Theorem 9.3.5. In particular,
P k1
P k1
k=1 kak c = k=1 ak (c + cck2 + c2 ck3 + + ck1 ) is well-defined, but this is
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 9.3.5, 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
Set D = 21 k k(k 1)|ak |dk2 , and = min{d |c|, D+1
}. Then is positive. Let c C
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 9.3.5 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
|ak (xk1 + cxk2 + c2 xk3 + + ck1 )|
|ak |(|x|k1 + |c||x|k2 + |c|2 |x|k3 + + |c|k1 ) (by the 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 1.5.21)
X
|ak ||x c|((|x|k2 + |x|k3 |c| + |x|k4 |c|2 + + |c|k2 )
+ |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 the triangle inequality)
X
|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
= |x c| |ak | d (by Example 1.5.1)
2
= |x c|D
< D
< .
This proves that g is continuous at c.
P
Theorem 9.4.2 Let f (x) = k=0 ak xk have radius of convergence R. Then f is differen-
P
tiable on B(0, R), and for all c B(0, R), f (c) = k=1 kak ck1 . The derivative function
P P
f (x) = k=0 kak xk1 = k=1 kak xk1 also has radius of convergence R.
Proof. Let c, x B(0, R). Then

P k
P k
f (x) f (c) k ak x k ak c
=
xc xc
P
ak (xk ck )
= k
xc
P
ak (x c)(xk1 + xk2 c + xk3 c2 + + ck1 )
= k (by Exercise 1.5.21)
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)
f (c) = lim = lim g(x) = g(c) = kak ck1 .
xc xc xc
k=1
Then the theorem follows from Theorem 9.3.5.

P
Theorem 9.4.3 Suppose that the power series k=0 ak xk has radius of convergence R
P
and that the power series k=0 bk xk has radius of convergence S. Suppose that for some
positive r R, S the two power series take on the same values at all z Br (0). Then for
all n 0, an = gn .
P P
In other words, if for all x Br (0), k=0 ak xk = k=0 bk xk , then the two series are
identical.
P P
Proof. Let f (x) = k=0 ak xk and g(x) = k=0 bk xk . By assumption a0 = f (0) = g(0) =
b0 . Let n N+ . By Theorem 9.4.2, f and g are differentiable on Br (0), and since they
are identical, their nth derivatives agree on Br (0) as well. By an n-fold application of the
derivative,
X
(n)
f (x) = ak k(k 1)(k 2) (k n + 1)xkn,
k=n
and similarly for g (n) . But then

an n(n 1)(n 2) (n n + 1) = f (n) (0) = g (n) (0)bnn(n 1)(n 2) (n n + 1),
so that an = bn .

1
9.4.1 Consider the function f (x) = 1x and its corresponding geometric series on B(0, 1).
i) Compute a derivative of f as a rational function and as a power series.
ii) Compute an antiderivative of f as a rational function and as a power series.
P x2k+1
9.4.2 Let f (x) = k=0 (2k+1)k! .
i) Find the radius of convergence of the series.
ii) Compute the derivative of f .
P xk
iii) By Example 9.7.3 we know that for every real x, ex = k=0 k! . Find the series
2
for ex .
2
iv) Find a power series whose derivative is ex .
v) Address the no-closed-form discussion on page 206.
9.5 Numerical evaluations of some series
Differentiation of power series is a powerful tool. For all complex numbers x B(0, 1)
P 1
the geometric series k=0 xk converges to 1x . Certainly it is much easier to compute
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.
P k
Example 9.5.1 k=1 2k1 = 4.
P
Proof. Let f (x) = k=0 xk . This is the geometric series with radius of convergence 1 that
1
P P
converges to 1x (Example 9.3.2). By Theorem 9.4.2, f (x) = k=0 kxk1 = k=1 kxk1 ,
and by Theorem 9.3.5, the radius of convergence of f is also 1. Thus 12 is in the domain
of f . Since we have two ways of expressing f (as power series and as a rational function),
there is similarly also a second form for f :

1 1
f (x) = = .
1x (1 x)2
P k
P
1 k1 1
From the two forms we deduce that k=0 2k1 = k=0 k 2 = f (1/2) = (11/2)2 = 4.
P k2
P k2
Example 9.5.2 k=0 2k = 6, and k=0 2k1 = 12.
P k
Proof. As in the previous example we start with the geometric series f (x) = k=0 x

P k1 1
that converges on B(0, 1). Its derivative f (x) = k=0 kx = (1x)2 also converges

P k
P 2 k1
on B(0, 1). Then xf (x) = k=0 kx and its derivative (xf (x)) = k=0 k x also
converge on B(0, 1). From
X
2 k1 x 1 2x
k x = (xf (x)) = = +
(1 x)2 (1 x)2 (1 x)3
k=0
P k 2 P k2
we deduce that k=0 2k = 2 k=0 2k1 = 2 (11/2)2 + (11/2)3 = 12 (4 + 8) = 6. The
1 1 1 1
other part is also easy to verify.

P 1
Example 9.5.3 k=1 k2k = ln 2.
P k
Proof. Let f (x) = k=1 xk . The radius of convergence of f is 1, so 1/2 is in its domain.
P
Also, f (x) = k=1 x
k1
= 1x 1
, so that f (x) = ln(1 x) + C for some constant C. In
P P k
particular, C = 0 + C = ln(1 0) + C = f (0) = 0, so that k=1 k21k = k=1 k1 12 =
f ( 21 ) = ln(1 21 ) = ln( 21 ) = ln(2).

P
9.5.1 Compute and justify k=1 k31k .
P k
9.5.2 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?
9.6 Specialized uses of power series
This section is meant as a reference, and should probably be skipped in a first class
on power series.
The result below seems obvious, but the proof is nevertheless a bit involved.
P k
P k
Theorem 9.6.1 Let k=0 ak x and k=0 bk x be convergent power series with radii
of convergence R1 and R2 , respectively. Let R = min{R1 , R2 }. Then on B(0, R)
Pn Pn
the product sequence {( k=0 ak xk ) ( k=0 bk xk )}n converges to the power series
P Pk k
k=0 ( j=0 aj bkj )x .
P P P Pk
We write this as ( k=0 ak xk ) ( k=0 bk xk ) = k=0 ( j=0 aj bkj )xk on B(0, R).
Proof. If R = 0, there is nothing to prove. So we may assume that R is a positive real

number or it is . Let x B(0, R).
P P
Set L = k=0 ak xk and K = k=0 bk xk (these limits exist as |x| < R1 , R2 ), sn =
Pn k
Pn k
Pn
k=0 ak x , tn = k=0 bk x , and cn = j=0 aj bnj . By the theorem on the convergence
of products of sequences (Theorem 8.4.3), lim(sn tn ) = LK. Thus we need to prove that
P k
k=0 ck x = LK.
Let > 0. By convergence of products, there exists N1 > 0 such that for all integers
n > N1 , |sn tn LK| < /2. By expanding (repeated foiling) and by grouping,
n
! n
X X
s n tn = ak xk bj xj
k=0 j=0
X k
n X n X
X n
= am bkm xk + am bn+km xn+k
k=0 m=0 k=1 m=k
Xn n X
X n
= ck x k + am bn+km xn+k .
k=0 k=1 m=k
P P
e K,
Since |x| < R R1 , R2 , the series k=0 |ak xk | and k=0 |bk xk | converge, to L, e respec-
tively. Thus there exists N2 > 0 such that for all integers n > N2 ,

X X n
X

|ak xk | = |ak xk | |ak xk | < ,
4Ke +1
k=n+1 k=0 k=0
P
and similarly there exists N3 > 0 such that for all integers n > N3 , k=n+1 |bk xk | < .
e+1
4L
Now let n > max{2N1 , N2 , N3 }. Then, with possibly adding more non-negative terms in
the third line below,
n X
X n

am bn+km xn+k
k=1 m=k
n n/2
X X n n
X X
= m
am x bn+km x n+km
+ am xm bn+km xn+km
k=1 m=k k=1 m=max{k,n/2+1}

n/2 n n n/2
X X X X
m
|am x | bk x k
+ |am xm | bk xk
m=1 k=n/2 m=n/2 k=1
e e
L + K
e+1
4L e +1
4K

< .
2
Thus

Xn X n
k k
ck x LK = ck x sn tn + sn tn LK

k=0 k=0
n
X

ck xk sn tn + |sn tn LK|

k=0
Xn X n

< am bn+km xn+k + /2

k=1 m=k
< /2 + /2 = .
P
Theorem 9.6.2 Let f (x) = k=0 ck xk have radius of convergence a positive real num-
P
ber R. Let a C with |a| = R such that k=0 ck ak converges. Let B be an open ball
centered at a, and let g : B C be continuous. If f (x) = g(x) for all x B(0, R) B,
then f (a) = g(a).
Proof. Let > 0. We want to show that |f (a) g(a)| < , which will via Theorem 2.10.4
then prove that f (a) = g(a). It suffices to prove the inequality |f (a) g(a)| < under the
additional assumption that < 1.
P
Since k=0 ck ak converges, by Cauchys criterion Theorem 9.2.1, there exists a positive
Pn PN +m
integer N such that for all integers n N , | k=N ck ak | < /4. Let sm = k=N ck ak .
By assumption, for all m 1, |sm | < /4 < 1. Furthermore, cN = s0 /aN , and for n > N ,
cN +n = (sn sn1 )/aN +n .
Let r be a positive real number strictly smaller than 1. First we do a simple expansion:
r N s0 +(s1 s0 )r N +1 + (s2 s1 )r N +2 + + (sn sn1 )r N +n
= r N cN aN + cN +1 aN +1 r N +1 + cN +2 aN +2 r N +2 + + cN +n aN +n r N +n
N
X +n
= ck (ra)k .
k=N
P
N +n
Thus by more rewriting and by the triangle inequality, k=N ck (ra)k equals

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

= s0 (r N r N +1 ) + s1 (r N +1 r N +2 ) + + sn1 (r N +n1 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
N
r (1 r) 1 + r + + r n1 + r n (since |sm | < /4 for sufficiently large m)
4

= r N (1 r n + r n ) (by Example 1.5.4)
4

= rN
4

< .
4
P
In particular, k=N ck (ra)
k
4.
Since polynomial functions are continuous, there exists 1 > 0 such that for all x
PN 1 PN 1
B(a, 1 ), k=0 ck xk k=0 ck 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

NX 1 NX1

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

k=0 k=0

N
X 1 NX 1
k k
|g(a) g(ra)| + |g(ra) f (ra)| + f (ra) ck a + ck a f (a)

k=0 k=0

X N
X 1 N
X 1 X

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

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

X NX X
1 N
X 1
k k k k
< + ck (ra) + ck (ra) ck a + ck a
4
k=N k=0 k=0 k=N

< + + + = .
4 4 4 4
Since is arbitrary, by Theorem 2.10.4, f (a) = g(a).

P P
9.6.1 Let k
k=0 ak x , bk xk be power series and R a positive real number such that
P k=0 k P k

for all x B(0, R), k=0 ak x k=0 bk x = 1. In other words, at each x in B(0, R),
the two infinite series are multiplicative inverses of each other.
P
i) Prove that for all x B(0, R), k=0 ak xk 6= 0.
ii) Prove that a0 6= 0.
Pk
iii) Prove that b0 = a10 , and that for all k > 0, bk = a10 j=1 aj bkj . (Hint:
Theorem 9.6.1.) Suppose that there exists M R such that for all non-negative
integers n, |an xn | < M .
M +|a0 | k1
iv) Prove that for all integers k 1, |bk xk | |aM 0|
2( |a0 | ) . (Hint: induction.)
P P
*9.6.2 Let f (x) = k=0 ak xk and g(x) = k=0 bk xk .
i) Express (f g)(x) as a power series in terms of the ai and bj .
ii) What is special for the power series of composition if a0 = 0?
iii) Assuming that a0 = 0, and given the radii of convergence for f and g, what would
it take to find the radius of convergence of the composition series?
P
9.6.3 (Abels lemma) Suppose that a0 , a1 , a2 , . . . are complex numbers and that k ak
P
converges. Let f (x) = k ak xk . The goal of this exercise is to prove that f is defined on
B(0, 1) {1}, that it is continuous on B(0, 1), and that as a function on the real interval
[0, 1], it is continuous.
i) Prove that the domain of f includes B(0, 1) {1}.
ii) Prove that f is continuous on B(0, 1). (Hint: Invoke a theorem.)
P
Set L = k ak , and for n 0, set sn = a0 + a1 + + an L.
Pn Pn
iii) Prove that k=0 sk (1 x)xk = sn (1 xn+1 ) + k=0 ak (xk 1). (Hint: Exer-
cise 9.2.11.)
iv) Prove that for every > 0 there exists N N such that for all integers n N ,
|sn | < .
Pn
v) Prove that the sequence { k=0 sk (1 x)xk }n converges for all x B(0, 1) {1}.
P
vi) Prove that f (x) L = k=0 sk (1 x)xk for all x B(0, 1) {1}.
vii) Let > 0. Prove that there exists N R such that for all integers m N and for
P
all real x in the interval [0, 1], k=m sk (1 x)xk < .
viii) Let > 0. Prove that for every positive integer n there exists > 0 such that for
Pn
all x (B(0, 1) {1}) B(1, ), k=0 sk (1 x)xk < .
ix) Let > 0. Prove that there exists > 0 such that for every real x in the interval
P
[max{0, 1 }, 1], k=0 sk (1 x)xk < .
x) Let g : [0, 1] C be defined by g(x) = f (x). Prove that g is continuous.
9.7 Taylor series
A common way of generating power series comes from approximations of functions.
Definition 9.7.1 Let a be in the domain of a function f , and assume that f has derivatives
P (k)
of all orders at a. The Taylor series of f (centered) at a is the series k=0 f k!(a) (xa)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 9.7.2 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, and a few examples are left for the exercises. One of the Taylor series
in the exercises does not converge to its function at any point other than at a.
Example 9.7.3 Let f (x) = ex , where the domain of f is R. It is easy to compute the
Taylor series for f :

X xk
.
k!
k=0
By the ratio test, this series converges for all x C, not only for x R. (More on this
series is in Sections 9.8 and 9.9, where we learn more about exponentiation by complex
numbers.) Let x R. By Theorem 6.5.5 there exists d between 0 and x such that
n
X
x xk ed
e = xn+1 .
k! (n + 1)!
k=0
Let > 0. Set M to be any integer bigger than or equal to |x|. By Corollary 8.6.6, there
exists N > 2M such that for all integers n > N , 21n < 22M 1M(3M
!
)M
. Then
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)
(3M )M M n+12M
<
M ! (2M + 1) n(n + 1)
(3M )M M M M M M
=
M ! 2M + 1 2M + 2 2M + 3 n n+1
n+12M
(3M )M 1
<
M! 2
(3M )M 1
= 22M 1
M ! 2n
< .
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 xk
ex = .
k!
k=0
In practice, the justification above is precisely how one (a human or a computer) com-
putes values ex to arbitrary precision. For example, to compute e0.1 to within 0.0001,
following the machinery above we take M = 1 > 0.1, and the smallest integer N
P16 0.1k
for which 21N 0.0001 22M 1M(3M
!
)M is 16. Thus to within 0.0001, e
0.1
is k=0 k! =
21021144462931669584965107651 0.1
19020718080000000000000000000 =
1.10517091807565. To compute e , the same N works
P 16 (0.1) k
7011749326061779176954720883
for desired precision, giving e0.1 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 9.7.4 Let f (x) = ln(x + 1), where the domain of f is (1, ). It is straightfor-
ward to compute the Taylor series for f centered at 0:

X (1)k1 xk
.
k
k=1
By the Ratio test for series, Theorem 9.2.4, the radius of convergence for this series is 1.
It is worth noting that the domain of the function f is all real numbers strictly bigger
than 1, whereas the computed Taylor series converges at all complex numbers in B(0, 1)
and diverges at all complex (including real) numbers whose absolute value is strictly bigger
than 1. By Example 9.1.7, the series diverges at x = 1, and by Theorem 9.2.6, it converges
at x = 1. You should test and invoke Theorem 6.5.5 to show that for all x (1, 1),
P k1 k
ln(x + 1) = k=1 (1) k x .
Incidentally, since ln(x + 1) is continuous on its domain and since its Taylor series
converges at x = 1 by the Alternating series test (Theorem 9.2.6), it follows from Theo-
rem 9.6.2 that
X
(1)k1
= ln 2.
k
k=1
With the help of Taylor series we can similarly get finite-term expressions for other
infinite sums. Below is another, harder example, and the reader may wish to skip it.

Example 9.7.5 Let f (x) = 1 x. The domain of f is the interval (, 1]. On the
subinterval (, 1) the function has derivatives of all orders:
1 1 1 1 1 3
f (x) = (1 x)1/2 , f (x) = (1 x)3/2 , f (x) = (1 x)5/2 , . . . ,
2 2 2 2 2 2
1 1 3 2n 3
f (n) (x) = (1 x)(2n1)/2.
2 2 2 2
Thus the Taylor series for f centered at 0 is

X 1 3 5 (2k 3)
1
1 x k
xk .
2 k!2
k=2
For large n, the quotient of the (n + 1)st coefficient divided by the nth coefficient equals
2n1
2(n+1) , whose limsup equals 1. Thus by the Ratio test for power series (Theorem 9.3.9),
the Taylor series converges absolutely on B(0, 1), and in particular it converges absolutely
on (1, 1). Furthermore, the quotient 2(n+1)2n1
= 2(n+1)3 4/3
2(n+1) above is at most 1 n for all
n 4. Thus by Raabes test (Exercise 9.2.17), this Taylor series converges at x = 1, and so
it converges absolutely on [1, 1]. But what does it converge to? Consider x (1, 1). We
use the integral form of the Taylors remainder theorem (Exercise 7.4.13): the difference
between Tn,f,0 (x) f (x) is
Z x Z x n
(x t)n 1 1 3 2n 1 (2n+1)/2 xt 1 3 5 (2n 1)
(1 t) dt = dt.
0 n! 2 2 2 2 0 1t n!2n+1 1 t
As the integrand goes to 0 with n, and as |x| < 1, the integral goes to 0 with n, so that
the Taylor series converges to f on (1, 1). (Incidentally, an application of Exercise 9.6.3
shows that the Taylor series is continuous on [1, 1], and as f is also continuous there,
necessarily the Taylor series converges to f on [1, 1].)

1
9.7.1 Let f (x) = 1x .
(n)
i) Compute f (x) for all x 0.
ii) Compute and justify the Taylor series of f centered at 0.
iii) Determine the radius of convergence.
iv) What is the domain of f and what is the domain of its Taylor series?
1
v) Compute the Taylor series for f (x) = 1x centered at 2. Determine its radius of
convergence (for this you have to take coefficients of powers of x 2 rather than
of powers of x).
1
vi) Compute the Taylor series for f (x) = 1x centered at 3. Determine its radius of
convergence. Compare with the previous parts.
9.7.2 Let f (x) = ln(x+1), where the domain of f is (1, ) as in Example 9.7.4. Compute
f (0.001) from one of its Taylor polynomials to four digits of precision. Show all work, and
in particular invoke Theorem 6.5.5.
Section 9.8: A special function 291

2
f (x) = e1/x , if x 6= 0;
0, if x = 0
as in Exercise 7.5.15.
i) Compute and justify the Taylor series for f centered at 0.
ii) Compute the radius of convergence for the series.
iii) Is it possible to compute f (0.001) from this series? Discuss.
*9.7.4 In introductory analysis courses we typically handle long, even infinite, sums but
not so much long products. In Example 9.7.5 we proved that the Taylor series for f (x) =
1 x centered at 0 converges on [1, 1]. This in particular means that the sequence
135(2n3)
{ n!2n }n converges to 0. Can you prove this directly without involving Raabes
test?
9.8 A special function
Define the power series

X xk
E(x) = .
k!
k=0
By Examples 9.3.10 (3), the domain of E is C. Thus by Section 9.4, E is differentiable

everywhere.
Remarks 9.8.1
P k1 P xk1 P k
(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,

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).
(5) Thus for all a, b C, E(a + b) = E(a)E(b).
(6) 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 .
(7) By parts (3) and (6), E(na) = (E(a))n for all integers n and all a C.
(8) 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 9.8.2 For all x R, E(x) = ln1 (x) = ex , the exponential function from Sec-
tion 7.5. In particular, E(1) equals the Eulers constant e.
E(x)
Proof. Define f : R R by f (x) = ex . Then f is differentiable and
E (x)ex E(x)(ex ) E(x)ex E(x)ex
f (x) = = = 0.
(ex )2 (ex )2
Thus f is a constant function, so that for all x R, f (x) = f (0) = E(0)
e0
= 11 = 1. Thus
E(x) = ex . By Theorem 7.5.7, this is the same as ln1 (x). In particular, E(1) = e1 = e.
The following is now immediate:
Theorem 9.8.3 The function E restricted to R has the following properties:

(1) The range is all of R+ .
(2) E restricted to R is invertible with E 1 (x) = ln(x).
Pn 1
The sequence { k=0 k! }n of partial sums for E(1) = e1 = e is convergent, and
starts with 1, 2, 2.5, 83
= 2.66667, 65 2.70833, 163
24 = 60 =
2.71667, 1957 2.71806, 685
720 = 252 =
19601 98641
2.71825, 40320 = 2.71828, 36288 = 2.71828. The last decimal approximation turns out
to be correct to 5 decimal places.
In the next section we analyze the case of E restricted to i times real numbers. By
(8) in Remarks 9.8.1, we will then understand E completely.

9.8.1 Use the definition of ln in this section to prove that for all x, y R+ , ln(xy) =
ln x + ln y, and that for all integers n, ln(xn ) = n ln x.
9.8.2 Use the definition of ln in this section and derivatives of inverse functions (Theo-
rem 6.2.7) to prove that (ln) (x) = x1 .
9.8.3 Prove that {f : R R differentiable | f = f } = {cE(restricted to R) : c R}.
Similarly, prove that {f : C C differentiable | f = f } = {cE : c C}.
P
9.8.4 Find a0 , a1 , a2 , . . . R such that the power series k=0 ak xk converges for all x R
2
to ex . (Hint: Use the function E to do so easily.) With that, determine a power series
2
whose derivative is e(x ) . (This problem is related to Exercise 9.4.2; it turns out that there
2
is no simpler, finite-term antiderivative of ex .)
2
9.8.5 Determine a power series whose derivative is 2(x ) . (It turns out that there is no
simpler, finite-term antiderivative of this function see the no-closed-form discussion on
page 206).
P k
9.8.6 Express k=0 (1) k
4 k!
in terms of e.
P 2k
9.8.7 Express k=0 3k (k+1)! in terms of e.
9.8.8 Find a Taylor polynomial expansion of the function E centered at 0. (Do as little
work as possible, but do explain your reasoning.)
9.8.9 Numerically evaluate e2 from the power series expansion to 5 significant digits. Prove
that you have achieved desired precision. (Hint: Theorem 6.5.5.)
9.8.10 Compute limxa E(x). (Do as little hard work as possible, but do explain your
reasoning.)
9.8.11 With proof, for which real numbers does the sequence {E(n)}n converge?
9.9 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 that takes real numbers x to E(ix). Note
that
X
(ix)k
E(ix) =
k!
k=0
ix (ix)2 (ix)3 (ix)4 (ix)5 (ix)6 (ix)7 (ix)8 (ix)9
=1+ + + + + + + + + +
1! 2! 3! 4! 5! 6! 7! 8! 9!
x x2 x3 x4 x5 x6 x7 x8 x9
=1+i i + +i i + + i + .
1! 2! 3! 4! 5! 6! 7! 8! 9!
We define two new functions (their names may be purely coincidental, but pronounce them
for the time being as cause and sin):
x2 x4 x6 x8
COS (x) = Re E(ix) = 1 + + + ,
2! 4! 6! 8!
x x3 x5 x7 x9
SIN (x) = Im E(ix) = + + + .
1! 3! 5! 7! 9!
Since E(ix) converges, so do its real and imaginary parts, so that the domains of COS and
SIN are both all of R.
Remarks 9.9.1
(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.
(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 9.9.2 There exists a unique real number s (0, 3) such that E(is) = i, i.e.,
COS (s) = 0 and SIN (s) = 1.
Proof. The function t SIN (t) is differentiable and its derivative is 1 COS (t), which is
always non-negative. Thus t SIN (t) is non-decreasing for all real t, so that for all t 0,
2
t SIN (t) 0 SIN (0) = 0. Hence the function t2 +COS (t) has a non-negative derivative
2 2
on [0, ), so that t2 + COS (t) is non-decreasing on [0, ). Thus for all t 0, t2 + COS (t)
2 3
02 +COS (0) = 1. It follows that the function t6 t +SIN (t) is non-decreasing on [0, ).
3
[How long will we keep going like this???] Thus for all t 0, t6 t + SIN (t)
3
t4 2
06 0 + SIN (0) = 0. Thus 24 t2 COS (t) is non-decreasing on [0, ), so that for all
t4 t2 04 02
t 0, 24 2 COS (t) 24 2 COS (0) = 1. We conclude that for all t 0,
t4 t2
COS (t) + 1.
24 2
4 2
In particular, COS ( 3) 243 23 + 1 = 18 < 0. We also know that COS (0) = 1 >
0. Since COS is differentiable, it is continuous, so by the Intermediate Value Theorem

(Theorem 5.3.1) there exists s (0, 3) such that COS (s) = 0. Since (COS (s))2 +
3
(SIN (s))2 = 1, we have that SIN (s) = 1. We proved that for all t 0, t6 t+SIN (t) 0,
3
so that SIN (t) t t6 = 6t (6 t2 ). If t (0, 3], then SIN (t) 6t (6 t2 ) > 0. Hence
SIN (s) = 1. This proves that E(is) = i.
Remark 9.9.3 The proof above establishes the following properties for all t 0:
t2 t2 t4
COS (t) 1, COS (t) 1 , COS (t) 1 + ,
2 2 24
t3
SIN (t) t, SIN (t) t
.
6
The pattern may be obvious at this point: Taylor polynomials of COS and SIN alternate
between over- and under-estimating. The limits of the Taylor polynomials, namely the
Taylor series, are precisely the given power series of COS and SIN .
Finally, we connect COS and SIN to trigonometric functions. First we need to specify
the trigonometric functions: For any real number t,
cos(t) + i sin(t)
is the unique complex number on the unit circle centered at the origin that is on the half-
ray from the origin at angle t radians measured counterclockwise from the positive x-axis.
In terms of ratio geometry (from page 21), this says that in a right triangle with one angle
t radians, cos(t) is the ratio of the length of the adjacent edge divided by the length of the
hypotenuse, and sin(t) is the ratio of the length of the opposite edge divided by the length
of the hypotenuse. Geometrically it is clear, and we do assume this fact, that cos and sin
are continuous functions. We will use continuity in proving that the two functions are also
differentiable.
Theorem 9.9.4 The trigonometric functions cos and sin are differentiable. Furthermore,
COS and SIN are the functions cos and sin, 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 9.9.2. The angle (in radians, counterclockwise from the positive
real axis) of E(is) = 0 + i 1 = i is /2. The proof of Theorem 9.9.2 showed for t [0, 1],
the imaginary part SIN (st) of E(ist) is positive, and the real part COS (st) of E(ist) is
strictly decreasing from 1 to 0. Thus E(ist) is in the first quadrant.
Let n be an integer strictly greater than 1. What is the angle of E(is/n) counter-

clockwise from the positive real axis? I claim that it is 2n . By the previous paragraph,
E(is/n) is in the first quadrant, so that 0 < < /2. Since (E(is/n))n = E(is) = i, by
Theorem 3.2.4 it follows that n equals the angle of i, i.e., that n = 2 + 2k for some
integer k. Thus

0<= (1 + 4k) < .
2n 2
The first inequality says that necessarily the integer k is non-negative, and the second
inequality says that 1 + 4k < n. We need to prove that k is 0.
n
Suppose that k 1. Let j = 1+4k , i.e., let j be the smallest integer greater than
or equal to n/(1 + 4k). With our assumptions of n 2 and k 1, necessarily j n 1.

By Theorem 3.2.4, the angle of E(ijs/n) = E(is/n)j is j = j 2n (1 + 4k) 2 . By the
second paragraph of this proof, for t [0, 1], E(ist) varies continuously counterclockwise
on the unit circle from 1 to i, so that the angle of E(ist) varies continuously from 0 to /2,
which means that for t = nj < 1, the angle of E(ijs/n) should be strictly smaller than /2,
but we computed it to be at least /2. This is a contradiction, so necessarily k = 0 and

= 2n .
Since E(ism/n) = (E(is/n))m by Theorem 3.2.4, it follows that for all m Z, the
angle of E(ism/n) is m = 2 m n . Since m and n are arbitrary integers with n 1, this
says that for all rational numbers r,

COS (sr) + i SIN (sr) = E(isr) = cos r + i sin r .
2 2
Now let x be any real number. Then x is a limit point of Q, and since COS and cos are
continuous, we get that

COS (sx) = lim COS (sr) = lim cos r = cos x .
rQ,rx rQ,rx 2 2

Similarly, SIN (sx) = sin 2 x . In other words, for all x R,

2s 2s
cos(x) = COS x , sin(x) = SIN x .

Since COS and SIN are differentiable, these equalities say that cos and sin are differentiable
as well and that

2s 2s 2s 2s 2s 2s
cos (t) = SIN t = sin(t), sin (t) = COS t = cos(t).

By Exercise 1.3.23, for small positive real x, 0 < cos(x) < sinx x < cos1 x . Since cos is
differentiable, it is continuous. Since cos(0) = 1, by the Squeeze theorem (Theorem 4.3.11),
x
limx0+ sin x = 1. Hence
2s 2s sin x sin 0 sin x
= cos(0) = sin (0) = lim = lim = 1,
x0 x x0 + x
whence s = /2.
Theorem 9.9.5 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 3.2.3, x/r and x have the same
angle. Let be that angle. Then x/r = E(i), so that x = rE(i).
Notation 9.9.6 It is common to write E(x) = ex for any complex number x. We have
seen that equality does hold if x is real, but we adopt this notation also for other numbers.
With this, if x, y R, then
ex+iy = ex eiy ,
and ex is the length and y is the angle of ex+iy counterclockwise from the positive x axis.

9.9.1 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 3.2.3.)
9.9.2 Prove that ei + 1 = 0. (This has been called one of the most beautiful equalities,
because it connects the fundamental numbers 0, 1, e and .)
9.9.3 Let x, y R. Expand both sides of E(i(x + y)) = E(ix)E(iy) in terms of sin and
cos to prove that cos(x + y) = cos(x) cos(y) sin(x) sin(y), and sin(x + y) = sin(x) cos(y) +
cos(x) sin(y). (Cf. Exercise 1.3.20.)
9.9.4 Prove de Moivres formula: for all n Z and all real numbers x,
(cos(x) + i sin(x))n = cos(nx) + i sin(nx).
9.9.5 Express the following complex numbers in the form ex+iy : i, i, 1, 1, e, 2 + 2i.
Can 0 be expressed in this way? Justify.
9.9.6 Express (3 + 4i)5 in the form a + bi for some a, b R.
9.9.7 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.
9.9.8 For which real numbers and does the sequence {E(n(+i))}n converge? Prove
your conclusion.
9.9.9 Determine a power series whose derivative is sin(x2 ). Repeat for cos(x2 ). (It turns
out that there are no simpler, finite-term antiderivatives of the functions sin(x2 ), cos(x2 ).)
*9.9.10 For the moment pretend that we do not know that COS = cos and SIN = sin.

Let f : [0, 1] R be given by f (x) = 1 x2 . Then the graph of f is a quarter of the
circle of radius 1 in the first quadrant.
i) Justify that s
Z 1 2
2x
= 1+ dx.
2 0 2 1 x2
(Hint: This is an improper integral; use lengths of curves; is the perimeter of
half the circle of radius 1.)
ii) Compute the improper integral via substitution u = COS (x) and further other
steps. The integral should evaluate to s from Theorem 9.9.2.
9.10 Trigonometry
In this section we review the semesters concepts of limits, continuity, differentiability,

and integrability on the newly derived trigonometric functions from the previous section.
Definition 9.10.1 For any x, eix is the complex number that is on the unit circle at angle
x radians counterclockwise from the positive horizontal axis.
(1) sin(x) is the imaginary part of eix .
(2) cos(x) is the real part of eix .
sin(x)
(3) tan(x) = cos(x) , cot(x) = cos(x) 1 1
sin(x) , sec(x) = cos(x) , csc(x) = sin(x) .
Remarks 9.10.2 The following follow easily from the previous section:
(1) sin and cos are differentiable and hence continuous functions whose domain is R.
(2) For all x R, sin(x + 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 9.9.5.
(3) The functions sin and cos are continuous, i.e., for all a R, lim sin(x) = sin(a),
xa
lim cos(x) = cos(a). (In the proof of Theorem 9.9.4 we assumed that cos is
xa
continuous, but we did not assume that sin is continuous.)
(4) For all a R in the domain of tan (resp. cot, sec, csc), tan (resp. cot, sec, csc) is
continuous and differentiable at a.
ix ix
(5) sin(x) = e e 2i .
eix +eix
(6) cos(x) = 2 .
Section 9.10: Trigonometry 299
(7) (sin(x))2 + (cos(x))2 = 1. (Recall from Remark 2.4.7 that for trigonometric func-
tions we also write this as sin2 (x) + cos2 (x) = 1, but for an arbitrary function f ,
f 2 (x) refers to f (f (x)) rather than for (f (x))2.)
(8) (tan(x))2 + 1 = (sec(x))2 .
(9) (cot(x))2 + 1 = (csc(x))2 .
(10) sin (x) = cos(x).
(11) cos (x) = sin(x).
(12) For all x R, sin(x) = sin(x) and cos(x) = cos(x).
(13) sin and cos take on non-negative values on [0, /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 re-
stricted to [0, ], has an inverse, called arccos. The domain of arccos is [1, 1].
(17) Verify the details in the following. The derivative of tan is always non-negative, so
that on (/2, /2), tan is invertible. Its inverse is called arctan, and the domain
of arctan is (, ).
Proposition 9.10.3 Refer to Theorem 6.2.7.

(1) For x (/2, /2), arcsin (x) = 1x
1
2
.
1
(2) For x (0, ), arccos (x) = 1x2 .
1
(3) arctan (x) = 1+x 2.
Proof.
1
arcsin (x) =
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
arctan (x) =
tan (arctan(x))
1
=
(sec(arctan(x)))2
1
=
1 + (tan(arctan(x)))2
1
= .
1 + x2

9.10.1 Compute the derivatives of the standard trigonometric functions tan(x), cot(x),
sec(x), csc(x).
9.10.2 Use integration by substitution (Exercise 7.4.5) to compute the antiderivatives of
tan(x) and cot(x).
9.10.3 Use integration by substitution (Exercise 7.4.5) and a rewriting trick to compute
an antiderivative of sec(x) = sec(x)(sec(x)+tan(x))
sec(x)+tan(x) . Compute an antiderivative of csc(x).
9.10.4 (We prove properly Example 6.3.6.) Prove that for all positive x, x > sin(x).
9.10.5 Let f : [0, /2] R be defined by f (x) = (/2) sin(x) x.
i) Prove that there is exactly one c [0, /2] such that f (c) = 0.
ii) Prove that (/2) sin(x) x for all x [0, /2]. (Compare with Exercise 9.10.4.)
9.10.6 Prove that for all x R,
1 cos(2x) 1 + cos(2x)
(sin(x))2 = , (cos(x))2 = .
2 2
R R
9.10.7 Compute (sin(x))2 dx (or (cos(x))2 dx) twice, with two different methods:
i) Use the previous exercise.
ii) Use integration by parts twice (Exercise 7.4.6).
9.10.8 Let k, j Z. Show all of the following (possibly by multiple integration by parts
(Exercise 7.4.6)):
R
i) sin(kt) cos(jt) dt = 0.
ii)
Z (
0, if j 6= k,
sin(kt) sin(jt) dt = 0, if jk = 0,
, otherwise.
iii)
Z
0, if j 6= k or jk 6= 0,
cos(kt) cos(jt) dt =
, otherwise.
Section 9.11: Examples of LHopitals rule 301
9.10.9 Prove that for any non-zero integer k,

Z
1 1
x sin(kx) dx = x cos(kx) + 2 sin(kx) + C,
k k
and that Z
1 1
x cos(kx) dx = x sin(kx) + 2 cos(kx) + C.
k k
9.10.10 Let t R and n N+ .

i) Prove that sin((n + 12 )t) = sin((n 12 )t) + 2 cos(nt) sin( 21 t). (Hint: Exercise 9.9.3,
multiple times.)
Pn
ii) Prove that sin((n + 12 )t) = sin( 21 t) 1 + 2 k=1 cos(kt) .
Pn
iii) Give a formula or formulas for simplifying k=1 cos(kt).
9.11 Examples of LH
opitals rule
LHopitals 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.

The next two exercises are taken from R. J. Bumcrots article Some subtleties in
LHopitals 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.
9.11.1 Let f (x) = x2 sin(x1 ) and g(x) = sin x.
f (x)
i) Prove that limx0 g(x) = 0.

ii) Prove that limx0 fg (x)
(x)
does not exist.
iii) Why does this not contradict LHopitals rule Theorem 6.4.3?
f (x)
9.11.2 Let f (x) = x sin(x1 ) and g(x) = sin x. Prove that neither limx0 g(x) nor

f (x)
limx0 g (x) exist.
*9.11.3 (This is from R. P. Boass article Counterexamples to LHopitals rule American

Rx f (x)
Mathematical Monthly 93 (1986), 644645.) Define f (x) = 0 cos2 (t) dt, g(x) = 2+sin(x) .
i) Prove that lim f (x) = = lim g(x).
x x
f (x)
ii) Prove that lim 6= 0.
x g(x)
iii) Compute f (x), g (x).
iv) Write f (x) = f1 (x) cos x and g (x) = g1 (x) cos x for some functions f1 , g1 . Prove
f1 (x)
that lim = 0.
x g1 (x)
f (x)
v) Prove that lim 6= 0.
x g (x)
vi) Why does this not contradict LHopitals rule at infinity (proved in Exercise 6.4.5)?
*9.11.4 (This is from R. C. Bucks Advanced Calculus, McGraw-Hill Book Company,
Inc., 1956, page 48.) Let f (x) = 2x + sin(2x), g(x) = (2x + sin(2x))esin(x) , and h(x) =
e sin(x) .
i) Prove that limx h(x) does not exist.
ii) Prove that limx fg(x)
(x)
does not exist.
iii) Verify that f (x) = 4 cos2 (x) and that g (x) = (4 cos(x)+2x+sin(2x)) cos(x)esin(x) .
4 cos(x)
iv) Prove that limx (4 cos(x)+2x+sin(2x))e sin(x) = 0.
v) Why does LHopitals rule not apply here?

2
*9.11.5 (Modification of the previous exercise.) Let f (x) = x + sin( x2 ), and g(x) =
f (x)
. Prove that limx0+ fg(x)
(x)
1
( x2 + sin( x2 ))e
sin( x
does not exist but that limx0+ g (x) = 0.
Why does LHopitals rule not apply here?
9.12 Further exotic uses of trigonometry
All further exotic uses are presented through exercises.

9.12.1 (This is taken from the Monthly Gem article by D. Velleman in the American
Mathematical Monthly 123 (2016), page 77. The gem is a simplification of the article
P 2
by Y. Matsuoka, An elementary proof of the formula k=1 k = 2 /6, in American
Mathematical Monthly 68 (1961), 485487.) For every non-negative integer n define In =
R /2 2n
R /2 2
0
(cos(x)) dx and J n = 0
x (cos(x))2n dx.
i) Prove that I0 = /2 and J0 = 3 /24.
ii) Use integration by parts (Exercise 7.4.6) to prove that for all n 1, In = (2n
1)(In1 In ). (Hint: dv = cos(x) dx.)
iii) Prove that In = 2n1

2n In1 .
iv) Use integration by parts twice, first with dv = dx and then with dv = 2xdx, to
prove that for all n 1, In = n(2n 1)Jn1 2n2 In .
v) Prove that for all n 1,
1 Jn1 Jn
=2 .
n2 In1 In
vi) Prove that for all integers N ,
XN
1 J0 JN 2 JN
2
= 2 = 2 .
n=1
n I0 IN 6 IN
vii) Use the inequality x (/2) sin(x) for x [0, /2] (see Exercise 9.10.5) to estimate
Z
2 /2 2 2 1
0 JN (sin(x))2 (cos(x))2N dx = (IN IN +1 ) = IN .
4 0 4 4 2N + 2
viii) Prove that
X
1 2
= .
n=1
n2 6
*9.12.2 (This exercise is used in Exercise 9.12.3.) Use de Moivres formula (see Exer-
cise 9.9.4) to prove that for all x R and all positive integers n,
Xn
n n n n k
cos(nx) + i sin(nx) = (sin(x)) (cot(x) + i) = (sin(x)) i (cot(x))nk .
k
k=0
If n = 2m + 1, this says that

X
2m+1
2m + 1 k
2m+1
cos((2m + 1)x) + i sin((2m + 1)x) = (sin(x)) i (cot(x))2m+1k .
k
k=0
The imaginary part of this is:

m
X
2m+1 k 2m + 1
sin((2m + 1)x) = (sin(x)) (1) (cot(x))2m2k .
2k + 1
k=0
Pm
k 2m+1 mk
Let P (x) = k=0 (1) 2k+1 x . Prove that P is a polynomial of degree m with
leading coefficient 2m + 1. Prove that for k = 1, . . . , m, (cot(k/(2m + 1)))2 is a root of P .
Prove that P has m distinct roots. Prove that
Ym 2 !
k
P (x) = (2m + 1) x cot .
2m + 1
k=1
Pm
By equating the coefficient of xm1 in the two forms of P prove that k=1 (cot(k/(2m +
1)))2 = m(2m1)
3 .
*9.12.3 (See also Exercises 9.12.1 and 9.12.5.) (This is taken from the article I. Papadim-
P
itriou, A 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).
1
i) Prove that sin(x) cos(x) < x < tan(x), and that (cot(x))2 < x2 .
1
ii) Use Exercise 9.10.4 to prove that x2 < 1 + (cot(x))2 .
Pm k 2
iii) Prove that for all integers k, m with 1 k m, k=1 cot 2m+1 <

(2m+1)2 Pm Pm 2
1 k
2 k=1 k2 < m + k=1 cot 2m+1 .
(2m+1)2 Pm
iv) Use Exercise 9.12.2 to prove that m(2m1)
3 < 2
1
k=1 k2 < m +
m(2m1)
3 .
P 1 2
v) Prove that k=1 k2 = 6 . (Hint: multiply the previous part by 2 /4m2 .)
9.12.4 Fill in the explanations and any missing steps in the two double improper integrals.
(While integrating with respect to x, think of y as a constant, and while integrating with
respect to y, think of x as a constant.)
Z Z Z Z
1 1 1 1
dx dy = dx dy
0 0 1 + y 1 + x2 y 0 1+y 0 1 + x2 y
Z Z N !
1 1
= lim dx dy
0 1 + y N 0 1 + x2 y
Z !
1 arctan( yx) N
= lim dy
0 1 + y N y x=0
Z
1
= dy
2 0 y(1 + y)
Z
2u
= 2
du (by substitution y = u2 )
2 0 u(1 + u )
2
= ,
2
Z Z Z Z
1 1 1 1 x2
dy dx = dy dx
0 0 1 + y 1 + x2 y 0 0 1 x2 1 + y 1 + x2 y
Z
1 1
= 2
ln dx
0 1x x2
Z
ln x
=2 2
dx
0 x 1
Z 1 Z
ln x ln x
=2 2
dx + dx
0 x 1 1 x2 1
Z 1 Z 1
ln x ln u
=2 2
dx + du
0 x 1 0 u2 1
Z 1
ln x
=4 dx. (Stop here.)
0 x2 1
2
*9.12.5 (This is taken from the article D. Ritelli, Another proof of (2) = 6 , American
Mathematical Monthly 120 (2013), 642645.)
i) We proved in Section 9.4 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
R 1 ln x R1 P P R 1
0 1x2
dx = 0 ( ln x) k=0 x2k dx = k=0 0 ( ln x)x2k dx.
ii) Also accept that the two integrals in Exercise 9.12.4 are the same (order of inte-
gration matters sometimes).
P 2
iii) Use Exercises 7.4.7 and 9.2.2 to prove that k=1 k12 = 6 .
Appendi es
Appendix A. Advice on writing mathematics
Make your arguments succinct and straightforward to read

Write preliminary arguments for yourself on scratch paper: your first attempt may
yield some dead ends which definitely should not be on the final write-up. In the final
write-up, write succinctly and clearly; write what you mean and mean what you write;
write with the goal of not being misunderstood. Use good English grammar, punctuate
properly. And above all, use correct logical reasoning.
Do not allow yourself to turn in work that is half-thought out or that is produced in
a hurry. Use your best writing, in correct logical order, with good spatial organization on
paper, with only occasional crossing out of words or sections, on neat paper. Represent
your reasoning and yourself well. Take pride in your good work.
Process is important
Perhaps the final answer to the question is 42. It is not sufficient to simply write 42,
The answer is 42, or similar, without the process that led to that answer. While it is
extremely beneficial to have the intuition, the smarts, the mental calculating and reason-
ing capacity, the inspiration, or what-not, to conclude 42, a huge part of learning and
understanding is to be able to explain clearly the reasoning that lead to your answer.
I encourage you to discuss the homework with others before, during or after completing
it: the explanations back-and-forth will make you a better thinker and expositor.
Write your solutions in your own words on your own, and for full disclosure write the
names of all of your collaborators on the work that you turn in for credit. I do not take
points off, but you should practice full honesty.
Sometimes you may want to consult a book or the internet. Again, on the work that
you turn in disclose the help that you got from outside sources.
Keep in mind that the more you have to consult outside sources, the more fragile your
stand-alone knowledge is, the less well you understand the material, and the less likely you
are to be able to do satisfactory work on closed-book or limited-time projects.
Do not divide by 0
Never write 1/0, 0/0, 02 /0, /0. (Erase from your mind that you ever saw
this in print! It cannot exist.)
310 Appendix A: Advice on writing mathematics
Sometimes division by zero creeps in in subtler ways. For example, to find solutions
2
to x = 3x, it is wrong to simply cancel the x on both sides to get only one solution x = 3.
Yes, x = 3 is one of the solutions, but x = 0 is another one. Cancellation of x in x2 = 3x
amounts to dividing by 0 in case is the solution is x = 0.
Never plug numbers into a function that are not in the domain of the function
By design, the only numbers you can plug into a function are those that are in the
domain of the function. What else is there to say?
I will say more, by way of examples. Never plug 0 into the function f (x) = x1 (see
previous admonition). Even never plug 0 into the function f (x) = xx : the latter function
is undefined at x = 0 and is constant 1 at all other x.

Never plug 1 into or into ln.
3x 4, if x > 3;
Do not plug x = 0 or x = 1.12 into f that is defined by f (x) =
2x + 1, if x < 1.
Order of writing is important

The meanings of Everybody is loved by somebody and Somebody loves everybody
are very different. Another way of phrasing these two statements is as follows: For every x
there exists y such that y loves x, and There exists y such that for every x, y loves x.
In crisp symbols, without distracting such that, for, and commas, these are written as
x y, y loves x and y x, y loves x.
Conclusion: order of quantifiers matters.
Certainly also the order matters in implications; simply consider the truth values of
If x > 2 then x > 0 and If x > 0 then x > 2.
The statement if A then B can be written also as B if A, but in general it is better
to avoid the latter usage. In particular, when writing a long proof, do not write out a very
long B and only at the very end add that you were assuming A throughout; you could have
lost the doubting-your-statements reader before the disclaimer.
Here is a fairly short example where B if A form is not as elegant. With the statement
2
x is ... if x 6= 0 you might have gone over the abyss of dividing by 0 and no ifs can make
you whole again. It is thus better to first obtain proper assurances, and write if x 6= 0
then x2 is ... .
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
Appendix A: Advice on writing mathematics 311
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.
To prove an (in)equality, manipulate one side in steps to get to the other

Pn
If you have to prove that k=1 k 2 = n(n+1)(2n+1)
6 for n = 1, do NOT do the following:
1
X 1(1 + 1)(2 1 + 1)
k2 =
DO THIS!
DO NOT
6
k=1
123
12 =
6
1=1
The reasoning above is wrong-headed because in the first line you are asserting the equality
that you are expected to prove, and in subsequent lines you are simply repeating your
assumptions more succinctly. If you add question marks over the three equal sums and a
check mark on the last line, then you are at least acknowledging that you are not yet sure
of the equality. However, even writing with question marks over equal signs is inelegant
and long-winded. That kind of writing is what we do on scratch paper to get our bearings
on how to tackle the problem. But a cleaned-up version of the proof would be better as
follows:
X 1
123 1(1 + 1)(2 1 + 1)
k 2 = 12 = 1 = = .
6 6
k=1
Do you see how this is shorter and proves succinctly the desired equality by transitivity of
equality, with each step on the way sure-footed?
Another reason why the three-line reasoning above is bad is because it can lead to the
following nonsense:
?
1=0
?
add 1 to both sides: 2=1
?
multiply both sides by 0: 0=0
But we certainly cannot conclude that the first line 1 = 0 is correct.
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 x1
312 Appendix A: Advice on writing mathematics
R R
4 3x dx is terrible grammar; instead write (4 3x) dx.
Do not assume what is to be proved

If you have to prove that a is positive, do not assume that a is positive. (For a proof by
contradiction suppose that a 0, do correct logical reasoning until you get a contradiction,
from which you can conclude that a 0 is impossible, so that a must have been positive.)
Limit versus function value

Asking for 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 A implies B and if A is true, we may conclude B.
If A implies B and if B is false, we may conclude that A is false.
If A implies B and if A is false, we may not conclude anything.
If A implies B and if B is true, we may not conclude anything.
Truth table:
P Q P P Q P Q P xor Q P Q P Q
T T F T T F T T
T F F F T T F F
F T T F T T T F
F F T F F F T T
Statement How to prove it

P (via contradiction). Suppose P . Establish some nonsense (that
makes P thus impossible so that P must
hold).
P and Q. Prove P . Prove Q.
P or Q. Suppose that P is false. Then prove Q.

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

Contrapositively: Suppose that Q is false.
Prove that P is false.
314 Appendix B: What you should never forget
P Q. Prove P Q. Prove Q P .
For all x of a specified type, prop- Let x be arbitrary of the specified type. Prove
erty P holds for x. that property P holds for x.
(Possibly break up into a few subcases.)
There exists x of a specified type Find/construct an x of the specified type.

such that property P holds for x. Prove that property P holds for x.
Alternatively, invoke a theorem guaranteeing
that such x exists.
An element x of a specified type Suppose that x and x are both of speci-

with property P is unique. fied 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 Give a counterexample: find/construct an el-
hold for elements of a specified ement x of specified type and prove that P
type. does not hold for that x.
Statement Negation
P P
P and Q (P Q) = (P ) (Q)
P or Q (P Q) = (P ) (Q)
P Q (P Q) = P (Q)
P Q (P Q) = P (Q) = (P ) Q
For all x of a specified type, prop- There exists x of the specified type such that
erty P holds for x. P is false for x.
There exists x of a specified type For all x of the specified type, P is false for x.
such that property P holds for x.
Mathematical induction
The goal is to prove a property for all integers n n0 . First prove the base case,
namely that the property holds for n0 . For the inductive step, assume that for some
n 1 n0 , the property holds for n 1 (alternatively, for n0 , n0 + 1, . . . , n 1), and then
prove the property for n.
The limit definition of derivative
f (a + h) f (a) f (x) f (a)

f (a) = lim = lim .
h0 h xa xa
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 j = 1, . . . , n, let
mj = inf{f (x) : x [xj1 , xj ]}, Mj = sup{f (x) : x [xj1 , xj ]}.
The lower sum of f with respect to P is
n
X
L(f, P ) = mj (xj xj1 ).
j=1
The upper sum of f with respect to P is

n
X
U (f, P ) = Mj (xj xj1 ).
j=1
The lower integral of f over [a, b] is

L(f ) = sup{L(f, P ) : as P varies over partitions of [a, b]},
and the upper integral of f over [a, b] is

U (f ) = inf{U (f, P ) : P varies over partitions of [a, b]}.
We say that f is integrable over [a, b] when L(f ) = U (f ). We call this common value
the integral of f over [a, b], and we write it as
Z b Z b Z b
f= f (x) dx = f (t) dt.
a a a
The Fundamental theorems of calculus

I: Let f, g : [a, b] R such that f is continuous and g is differentiable with g = f . Then
Z b
f = g(b) g(a).
a
II: Let f : [a, b] R be continuous. Then for all x [a, b], f is integrable over [a, x], and
Rx
the function g : [a, b] R given by g(x) = a f is differentiable on (a, x) with
Z x
d
f = f (x).
dx a
Geometric series
P k r
k=1 r diverges if |r| 1 and converges to 1r if |r| < 1.
P k 1
k=0 r diverges if |r| 1 and converges to 1r if |r| < 1.
Pn n+1
For all r C \ {1}, k=0 r k = r r11 .
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.
(m n ) n = m (n n )
= m (n n )
= (m n ) n
= (m n) n
= m (n n )
= m (n n)
= (m n ) n
= (m n ) n
= m (n n)
= m (n n )
= (m n) n .
Then m n = m n, so that (m, n) (m , n ).
2.4.1:
Appendix B: What
2 you should never forget
b b b b 317
b
1 b b b b
1 b b b
(i) b b b b b (ii) b b b b
2.4.2:
4 bc
3 bc
2 bc
1 bc
bc
bc
2.4.6: i) f g(x) = f (g(x)) = f (x, 0) = x, g f (x, y) = g(f (x, y)) = g(x) = (x, 0). ii) f is
bc
surjective but not injective, g is injective but not surjective, f g is the identity function
and so bijective, and g f is neither surjective nor injective.
2.4.11: i) Each of the m elements of A has n options in B for its value, so we get nm
functions. ii) By the pigeonhole principle, as every pigeon in A needs to be in its own hole
in B, necessarily m n. ii) By the pigeonhole principle, as every hole in B needs to have
its own pigeon from A, necessarily m n.
2.4.13: There are infinitely many answers here. Possible answers: (i) The identity function
f (x) = x for all x R. (ii) The function f (x) = 0 for all x R. (iii) The function
f (x) = ex for all x R. (iv) The function f (x) = x3 x for all x R. This function is
not injective because f (1) = 0 = f (0). It is surjective by the Intermediate Value Theorem
(Theorem 5.3.1 in these notes, but we have not yet seen that). It is easier to prove the
required properties for the following function:

x, if x 0;
f (x) =
x + 1, otherwise.
This f is not injective because f (0) = 0 = f (1), but it is surjective because for every
positive y, y = f (y), and for every non-positive y, y = f (y 1).
2.5.2: As seen on page 68, there are four functions from {1, 2} to {1, 2}, only two of
which are bijective: the identity function f = {(1, 1), (2, 2)} and the switch-1-2 function
g = {(1, 2), (2, 1)}. Thus A = {f, g} has two elements. Clearly f f = f , f g = g = g f ,
and g g = f , so that composition takes any pair of elements in A to A. Thus is a
binary operation on A. The multiplication table for is:
f g
f f g
g g f
Certainly f is the identity element; since f f = f , it follows that f is its own inverse, and
since g g = f , it follows that g is its own inverse. The multiplication (i.e., composition)
table for shows that f g = g f , so that is commutative.
2.5.3: A bijective function from {1, 2, 3, . . . , n} to {1, 2, 3, . . . , n} takes 1 to any of n el-

ements in the codomain, and once 1 has chosen where it maps to, 2 can map to any
of the remaining n 1 elements in the codomain, after which 3 can map to any of the
remaining n 2 elements in the codomain, et cetera, n 1 can map to any of the re-
maining 2 elements, and finally n can map to the remaining 1 element, the composition
of bijective functions is bijective, and the domains and codomains remain {1, 2, . . . , n}.
Thus is a binary operation on the set A of all bijective functions from {1, 2, 3, . . . , n} to
{1, 2, 3, . . . , n}. Let f, g A be the functions f = {(1, 2), (2, 1), (3, 3), (4, 4), . . . , (n, n)} and
g = {(1, 1), (2, 3), (3, 2), (4, 4), . . . , (n, n)}. Then f g(1) = f (1) = 2, and gf (1) = g(2) = 3,
so that f g 6= g f .
2.5.5: Z/2Z: + 0 1 Z/3Z: + 0 1 2 Z/4Z: + 0 1 2 3

0 0 1 0 0 1 2 0 0 1 2 3
1 1 0 1 1 2 0 1 1 2 3 0
2 2 0 1 2 2 3 0 1
3 3 0 1 2
2.5.9: Base case n = 0: (f g f 1 )0 = id = f f 1 = f id f 1 = f g 0 f 1 . Base

case n = 1: (f g f 1 )1 = f g f 1 = f g 1 f 1 . Inductive step: Suppose that
(f g f 1 )n1 = f g n1 f 1 . Then
(f g f 1 )n = (f g f 1 )n1 (f g f 1 ) (by Remark 2.4.7)

= (f g n1 f 1 ) (f g f 1 ) (by induction assumption)
= f g n1 f 1 f g f 1 (by associativity of function composition)
= f g n1 id g f 1 (by definition of function inverses)
= f g n1 g f 1 (by definition of identity function)
= f g n f 1 (by Remark 2.4.7).
2.7.15: If 1/y 1/x, then by compatibility of < with multiplication by positive numbers,
x = xy(1/y) xy(1/x) = y, which by transitivity of < contradicts the assumption that
x < y. So 1/y 1/x is false, so that by trichotomy, 1/y < 1/x.
1
3.1.1: i) 2 12 i. ii) 80
13 + 3
13 i. iii) 122 + 597i.
3.1.3: Draw the unit circle centered at the origin: is on the unit circle at angle /3
(or 60 ) counterclockwise from the positive real axis..., 6 = 1, 7 = , et cetera. All the
numbers are on the unit circle, consecutive ones at 60 from each other.
3.1.3: i) Vertical line at x = 3. ii) Horizontal line at y = 3. iii) Hyperbola in two parts:
first quadrant and third quadrant y = x3 . iv) The real and the imaginary axes.
3.1.4: Let x C have the property that x2 = 1. Write x = a + bi for some a, b R.

Then 1 = x2 = a2 b2 + 2abi, so that necessarily 2ab = 0 and a2 b2 = 1. We know
that any non-zero real number has a multiplicative inverse, so 2ab = 0 implies that either
a = 0 or b = 0. Together with a2 b2 = 1 we conclude that a = 0 and b2 = 1. We USE
that there are only two real numbers b with the property b2 = 1, namely b = 1. But then
x = i.
A geometric proof (but really, we only learn these geometric methods in the next
section): Let x C have the property that x2 = 1. Then |x|2 = |x2 | = 1, so that |x| = 1,
so that x is on the unit circle. Let be the angle of x counterclockwise from the positive
real axis. Then the angle of x2 is 2, and that has to be equal to the angle of x2 = 1,
i.e., 2 = + 2n for an integer n. But then = 2 + n, so that is / 3
2 or 2 + = 2 ,
and for other n the two angles just repeat themselves. In any case, the only numbers at
these two angles on the unit circle are i.
3.1.7: i) x + x = (Re x + i Im x) + (Re x i Im x) = 2 Re x. ii) x x = (Re x + i Im x)
(Re x i Im x) = 2 Im x. iii) x y + x y = x y + x y = x y + x y = 2 Re(x y) by i).

3.2.1: i) 22 + i 22 . ii) 22 i 22 . iii) Product is 1 and the sum is 2.
3.2.3: i) Ray (one-sided line) from the origin at angle /3 counterclockwise from the real
axis. Only ray, not the whole line through the origin. ii) Circle of radius 3 with center at
0. iii) Circle of radius 3 with center at 2 i.
3.2.7:
b
i
3.3.14: Let a Int A. By definition there exists r > 0 such that B(a, r) A. But then
B(a, r) B, so that a Int B.
Now let a (Bd A) \ B. Since a Bd A, for all r > 0 the sets B(a, r) A and
B(a, r) (F \ A) are not empty. Since A B, it follows that for all r > 0, B(a, r) B
is non-empty, and since a 6 B, it follows that and B(a, r) (F \ B) is not empty. Thus
a Bd B.
Let A = {0} and B = F . Then Bd A = A and Bd B = .
3.3.5:
A = {x C : Im x = 0, 0 < Re x < 1}:
1 Int(A) = ,
Bd(A) = {x C : Im x = 0, 0 Re x 1},
0
-2 -1 0 1 2
limits points of A = {x C : Im x = 0, 0 Re x 1},
A is not open and is not closed.
B = {x C : Im x = 0, 0 Re x 1}:
1 Int(B) = ,
Bd(B) = B,
0
-2 -1 0 1 2
limits points of B = B,
B is not open and is closed.
C = {x C : 2 Im x 2, 0 Re x 1}:
2
Int(C) = {x C : 2 < Im x < 2, 0 < Re x < 1},
1 Bd(C) = {x C : Im x = 2, 0 Re x 1}
{x C : 2 Im x 2, Re x = 0 or Re x = 1},
0
limits points of C = C,
C is not open and is closed.
-1
-2
-3
-1 0 1 2
D = {x C : 2 Im x 2, 0 Re x 1} = because 2 6 2. I produce no graph of

the empty set, its interior, boundary and limit point all form the empty set, and D is both
open and closed.
E = {1/n : n N+ }.
Int(E) = ,
Bd(E) = E {0},
bb bbbbb b b b b b b b b b
limits points of E = {0},
E is not open and is not closed.
4.1.1:
The domain of the function x2 2x is all complex (real) numbers so
that 2 is a limit point of the domain. Let > 0. Set = min{3, /4}. Let x satisfy
0 < |x 2| < . Then

|(x2 3x) (2)| = x2 3x + 2
= |x 1| |x 2| (because of factoring,
and abs. value of product is product of abs. values)
(|x 2| + 1) |x 2| (because by clever 0 and triangle inequality)
< (3 + 1) |x 2| (because < 3)
< 4 (because 3 + 1 = 4 and |x 2| < )

4 (because /4)
4
= .
4.1.2: i) lim (x3 4) = 3. The domain of x3 4 is all complex numbers, so 1 is a limit

x1
point of the domain. Let > 0. Set = min{1, /7}. Let x be in the domain and satisfy
0 < |x 1| < . Then
|x3 4 (3)| = |x3 1|
= |x2 + x + 1| |x 1|
(|x|2 + |x| + 1) |x 1|
= (|x 1 + 1|2 + |x 1 + 1| + 1) |x 1|
((|x 1| + 1)2 + |x 1| + 1 + 1) |x 1|
< ((1 + 1)2 + 1 + 1 + 1) |x 1|
= 7 |x 1|
7
7 /7
= .
1
ii) lim = 12 . The domain is all non-zero complex numbers, so 2 is a limit point of the
x2 x
domain. Let > 0. Set = min{1/2, }. Let x be in the domain and satisfy 0 < |x2| < .
Then

1 1 2 x
=
x 2 2x
|x 2|
=
2|x|
1
= |x 2|
2|x 2 + 2|
1
|x 2| (by the reverse triangle inequality because |x 2| < 1)
2 2|x 2|
1
< |x 2| (because |x 2| < 1/2, so 2 2|x 2| > 2 2 1/2 = 1)
2 2 21
= |x 2|
<

x4 1

iii) lim 2 = 11 . The domain is all complex numbers except 2i, so 3 is a limit
x3 x +2
point of the domain. Let > 0. Set = min{1, }. Let x be in the domain and satisfy
0 < |x 3| < . Then

x4 1 11x 44 + x2 + 2

x2 + 2 11 = 11(x2 + 2)
2
x + 11x 42
=
11(x2 + 2)

x + 14
= |x 3|
11(x2 + 2)
|x + 14|
= |x 3|
11 |x2 + 2|
|x 3 + 17|
= |x 3|
11 |(x 3 + 3)2 + 2|
|x 3| + 17
|x 3| (by the triangle ineq. and compat.)
11 |(x 3)2 + 6(x 3) + 9 + 2|
1 + 17
< |x 3| (since |x 3| < 1 and compat.)
11 |(x 3)2 + 6(x 3) + 11|
18
|x 3| (because
11 4
|(x 3)2 + 6(x 3)| = |x 3| |x 3 + 6| 1 (|x 3| + 6) < 7,
so that by the reverse triangle ineq.
|(x 3)2 + 6(x 3) + 11| 11 |(x 3)2 + 6(x 3)| > 11 7 = 4,
so that 1/|(x 3)2 + 6(x 3) + 11| 1/4),

< |x 3|
<

iv) lim x + 5 = 3. The domain of THE square root function is all non-negative real
x4
numbers, so that the domain of x + 5 is the (real) interval [5, ). Thus 4 is a limit point
of the domain. Let > 0. Set = 3. Let x be in the domain and satisfy 0 < |x 4| < .
Then

x + 5 3 = ( x + 5 3) x + 5 + 3
x + 5 + 3

x+59
=
x + 5 + 3
1
= |x 4|
x+5+3
1
|x 4| (because x + 5 + 3 3)
3
1
<
3
1
3
3
=
4.1.4: To determine limx5 f (x), we plug in x with x 6= 5, so that limx5 f (x) =
limx5 (x3 4x2 ), and there is no dependence on b.
4.1.8: 1) Let > 0. [Here does not matter, so we do not need to invoke it
explicitly, we will simply implicitly use that = 34.] Let x satisfy 0 < x 0 < 34.
Then x is positive, so

x
|f (x) 1| = 1 = |1 1| = 0 < .
|x|
2) Let > 0. Let x satisfy 0 < 0 x < 34. Then x is negative, and

x

|x| (1) = | 1 + 1| = 0 < .
4.3.11: Let > 0. Since limxa g(x) = L, there exists > 0 such that for all x B,
if 0 < |x a| < , then |g(x) L| < . But then for the same , for all x A with
0 < |x a| < , |f (x) L| = |g(x) L| < .
4.3.12: If limxa f (x) exists and equals L, then by an application of Exercise 4.3.11 with
A = B, we have that limxa g(x) exists and equals L. By symmetry, if limxa g(x) exists
and equals L, then limxa f (x) exists and equals L.

Similarly, to determine limx5 g(x), we plug in x with x 6= 5, so that limx5 g(x) =
limx5 b, and we have proved that this equals the constant b.
1 2
f (0+h)f (0) 2 h 0
6.1.9: i) limh0+ h = lim h0 +
h = limh0+ 21 h = 0, and
1 2
f (0+h)f (0) h 0
limh0 h
= limh0 2 h = limh0 ( 21 h) = 0, so that f
is differentiable at
0 and f (0) = 0 = |0|. Near x > 0, f equals half of the squaring function whose derivative
at x is x = |x|, and at negative x, f equals minus half of the squaring function whose
derivative at x is x = |x|. Thus f is differentiable at all x and f (x) = |x|.
1 2
2 h +30
ii) limh0+ g(0+h)g(0)
h = limh0+ h = limh0+ ( 21 h + 3/h) does not exist, so
that g is not differentiable at 0.
iii) Near x > 0, g equals half of the squaring function plus 3, whose derivative at x is
x = |x|, and at negative x, g equals minus half of the squaring function whose derivative
at x is x = |x|. Thus g is differentiable at all non-zero x with g (x) = |x|.
6.2.6: i) By the difference rule, (f g) (1) = f (1) g (1) = 0 3 = 3. ii) By the product
rule, (f g) (2) = f (2)g(2) + f (2)g (2) = (3) 6 + 2 (6) = 30. iii) By the quotient

(3)g (3)
rule ( fg ) (3) = f (3)g(3)f(g(3))2 = 2345
32 = 14
9 . iv) By the chain rule, (g f ) (4) =
g (f (4)) f (4) = g (0) 1 = 4.
6.5.9: i) For positive x, the function equals the cubing function whose derivative is 3x2 ,
for negative x the function equals minus the cubing function whose derivative is 3x2 , and
3 3
at 0, f (0) = limh0 |h| h0 = 0. Thus

3x2 , if x 0;
f (x) =
3x2 , if x < 0.
3h2 302
ii) f (0) = limh0 h = 0, so that clearly

6x, if x 0;
f (x) =
6x, if x < 0.
In other words, f (x) = 6|x|. iii) We know that the absolute value function is not differ-
entiable, or to really (re-) verify it:
f (0 + h) f (0) 6h 0
lim+ = lim+ = 6,
h0 h h0 h
f (0 + h) f (0) 6h 0
lim = lim = 6.
h0 h h0 + h
7.1.5: i) Zero: the function is constant on the interval of width 0. ii) This is the area of the
quarter circle, hence 14 r 2 . iii) The region is a union of a parallelogram of base 2 and height
4 3 10 = 2 with a triangle with base 2 and height (4 5 10) (4 3 10) = 10 2 = 8, so
that the integral is 24+ 21 28 = 16. iv) Between x = 1 and x = 1 the region is a rectangle
of width 2 and height 2, between x = 1 and x = 3 the region is a union of a parallelogram

of width 2 and height 3 1 and of a triangle of base 2 and height 3 3 3 1 = 6, and
between x = 3 and x = 5 the region is a union of a parallelogram of width 2 and height
9 5 = 4 and of a triangle of base 2 and height (9 3) (9 5) = 2. Thus, the integral
is: 2 2 + 2 3 + 12 2 6 + 2 4 + 12 2 2 = 26.

7.1.6: i) We areZ integrating the function r 2 x2 that is constant as t varies, so the
rp p
integral is easy: r 2 x2 dt = r r 2 x2 . ii) We are integrating a constant function
0 Z 5
4x 10 with respect to t, so that (4x 10) dt = 2(4x 10). iii) This was already
3
computed in Exercise 7.1.5 as 26.
8.1.8: Termwise sum of sequences still is still an ordered sequence, so that + is a binary
operation. Certainly the constant 0 sequence is the additive identity. The additive inverse
for the sequence {sn }n is {sn }n .
1
8.2.1: Let > 0. Set N =

.Then if n > N ,

1

|sn s| = 2 0
n
1
= 2 (because 1/n2 is positive)
n
1
< 2 (because n > N so 1/n2 < 1/N 2 )
N
1
= 2 (because N = 1 )
(1/ )
= .
1
8.2.3: Let > 0. Set N = 1/k . Then if n > N ,

1
0 = 1
nk nk
1
< k (because n > N so that nk > N k as k is positive)
N
= .
1
8.2.7: Let > 0. Set N = .
Then if n > N ,
2

n + 1 + n
n + 1 n 0 = ( n + 1 n)
n+1+ n
n+1n
=
n+1+ n
1
=
n+1+ n
1
< (because n + 1 > 0 so that n + 1 + n > n)
n
1
< (because n > N )
N
= .
2
8.3.8: Of course, there are infinitely many choices here. i) sn = 2(n ) . ii) sn = n. iii)
2
sn = 2(n ) (same as i). iv) sn = n. v) sn = n2 .
8.6.1: i) We can use N proof: Let > 0. Set N = log2 (). Then for any integer n > N ,
| 21n 0| = 21n < 21N = , where the second to the las step holds because exponentiation of
1/2 is a decreasing function.
i) A comparison proof: First prove by induction that for all integers n > 1, n < 2n ,
so that 1 1 1 1
n 2n < n . We know that { n } converges to 0, so that by the squeeze theorem,
{ 21n } converges to 0.
1
i) Ratio test: 2n+1
1 = 21 has limit 12 which has absolute value strictly smaller than 1,
2n
so that by the ratio test the conclusion follows.
(4)n
The proofs of ii) and iii) are similar, but observe that 7n = ( 4 n
7 ) .
kn+1
(n+1)! k
iv) Ratio test: kn = n+1 , and the limit of this sequence as n goes to infinity is 0.
n!
kn
As |0| < 1, it follows that n! converges to 0 by the ratio test.
n+1
kn+1 n+1 1
v) Ratio test: n = n k, and the limit of this sequence as n goes to infinity is
kn
1
1 k
As |k| > 1, it follows that the limit of the ratios has absolute value strictly smaller
.
n
than 1, and by the ratio test it follows that n! converges to 0.
(n+1)m
vi) Ratio test: kn+1

nm = ( n+1 m1
n ) k , and the limit of this sequence as n goes to infinity
kn
1 n
is k. As in v), it follows that n! converges to 0.
8.6.4: Recall that s1 = 1, s2 = 1, and for all n 2, let sn+1

=sn + sn1 . By Exer-
n2

1+ 5 1 5
n
cise 1.5.31, sn > ((1 + 5)/2) and sn+1 = 2 s n + 2 . Let > 0. Since

| 15 | < 1, by the ratio test (Corollary 8.6.6), there exists N R such that for all integers
1+ 5
15 n
n > N , | 1+ 5
| < . Now set N = N + 1. Then for all integers n > N ,

s 1 5 n
n+1 1 + 5 2
=
sn 2 sn
n
1 5 1 5 2 1 5 n2 1 5 1 5 n2
2
< n2 = =
1+ 5 2 1+ 5 1 + 5 1 + 5

2
< .
8.7.5: So many possible solutions..., here are some possibilities. (i) Every convergent
sequence is Cauchy by Theorem 8.7.3, so there is no example here. (ii) In R or C, every
Cauchy sequence is convergent, but in Q, we can take the sequence {10n /10n}n (this
is a sequence of decimal approximations of ). (iii) {(1)n } is bounded but not Cauchy.
(iv) {n} is real increasing but not Cauchy. (v) {1/n} is real increasing and Cauchy.

8.8.1: Let = 12 3 .
i) Then has length 1 and is at 60 = /3 counterclockwise from the positive real
axis. Thus the first sequence is a repetition of , 2 , 3 , 4 , 5 , 6 . These six numbers are
all distinct on the unit circle, and as they get repeated infinitely many times, the set of
subsequential limits of the first squence is precisely these six numbers.
The second sequence consists of powers of 5 = 1 = (complex conjugate), which
is on the unit circle but at angle 60 = /3 clockwise from the positive real axis, and
similarly it has the same six subsequential limits.
ii) 0 is the only subsequential limit as the sequence converges to 0, so all subsequences
converge to the same limit.
iii) This is a combination of i) and ii): by the sum rules for convergent (sub)sequences,
the subsequential limits are the same as in i) (plus i times the result in ii).
iv) This is a combination of i) and ii): by the product rules for convergent
(sub)sequences, the subsequential limits are the same as in i) times the result in ii),
thus there is only one subsequential limit, namely 0.
iv) The sequence is {n + n }. But n + n = 2 Re(n ), which equals the sequence
which is a repetition of the finite sequence 1, 1, 2, 1, 1, 2. Thus the subsequential limits
are 1, 2.
v) The sequence is {n n } = {1}, which has only one subsequential limit, namely 1.
8.8.5: So many possible solutions..., here are some possibilities. (i) {n} has no con-
vergent subsequences. (ii) The constant sequence {1} works. (iii) The sequence {2 +
(1)n } works. This sequence can also be written as {1, 3, 1, 3, 1, 3, . . .}. (iv) The sequence
{1, 2, 12, 1, 2, 12, 1, 2, 12, 1, 2, 12, 1, 2, 12, 1, 2, 12, . . .} works.
8.9.2: i) The sequence converges to 0, so liminf and limsup are both also 0.
ii) lim inf(1)n n! = and lim sup(1)n n! = .
iii) The sequence converges to 0, so liminf and limsup are both also 0.
iv) lim inf{1, 2, 3, 1, 2, 3, 1, 2, 3, . . .} = 1 and lim sup{1, 2, 3, 1, 2, 3, 1, 2, 3, . . .} = 3.
v) lim inf is 4 and lim sup is 4. vi) The sequence diverges to , so liminf and limsup
are both also .
vii) The sequence converges to 0 (by the previous part and Theorem 8.3.10).
9.1.1: If |r| 1, then |r|n 1 for all n, and so by the contrapositive to Theorem 9.1.9,
the series does not converge. Hence it diverges by definition.
9.1.2: The three series are modifications of geometric series:

X X k 1
1 1 2
= = = 1,
k=1
2k
k=1
2 1 21

!
X 1 1 1 1 1 X 1 1 1 1 3 1 1
k
= 6 + 7 + 8 + = 6 k
= 6 1 = 6 = 5
(= ),
3 3 3 3 3 3 3 1 3 3 2 23 486
k=6 k=0

X
2 2 X 1 2 1 1 1
k
= 8 k
= 8 1 = 7
(= ).
5 5 5 5 1 5
25 256, 250
k=8 k=0
q
n
n xn |xn | n
9 1
9.1.6: I use the root test: lim sup 10n = lim sup 10 lim sup 10 = lim sup 10 < 1,
so that the series converges.
i) (A possible answer:) Every decimal number is an infinite series of the stated form,
P 4 P 1 1 P 123
and it thus converges. ii) k=1 10k = 4 k=1 10k = 4 1 10
10
1 = 94 . iii) k=1 1000k =
P 1
1
123 41
123 k=1 1000 k = 123
1000
1 10001 = 999 = 333 . iv) I should not have assigned this. Give
everyone full credit. But here is a solution. Eventually periodic means that beyond a
certain point it is periodic, i.e., that there exists positive integers M and N and digits
c0 , c1 , . . . , cN 1 {0, 1, . . . , 9} such that for all n M , xn = cn mod N . Then

X M 1
NX
X
xk xk xk
k
= k
+
10 10 10k
k=1 k=1 k=N M
M 1
NX
X
xk c0 + 101 c1 + 102 c2 + 101N cN 1
= +
10k (10N )k
k=1 k=N M
M 1
NX
xk c0 + 101 c1 + 102 c2 + 101N cN 1 X 1
= k
+ 2M
10 10 N (10N )k
k=1 k=0
M 1
NX
xk c0 + 101 c1 + 102 c2 + 101N cN 1 1
= + ,
1 101N
k N 2M
10 10
k=1
which is a finite sum, product and quotient of rational numbers, and is thus rational.
9.2.1:

X X
3i 1
i) The series converges if and only if converges. But for
k 2 + k4 k 2 + k4
k=1 k=1 P
each n, n21+n4 1 4 = n12 . By Example 9.1.8, the series k k12 converges, so
n
by Theorem 9.2.2, the given series converges. Both the ratio and root tests fail to
reach a conclusion.
X
1
ii) . Use the p-series test Theorem 9.2.9 to conclude divergence of the series.
k=1
k
Alternatively, we know that 1n n1 , and by Example 9.1.7 that the harmonic
P
series k k1 diverges, so that by Theorem 9.2.3, the given series diverges. Both
the ratio and root tests fail to reach a conclusion.
X
1 P 1
iii) . Use the p-series test Theorem 9.2.9 or compare to k2
to conclude
k3
k=1
convergence. Both the ratio and root tests fail to reach a conclusion.
X
2k
iv) . Both the ratio and root tests prove that the series converges.
k!
k=1
X
2k
v) . Both the ratio and root tests prove that the series diverges.
k3
k=1
X
1
vi) . The p-series test proves convergence.
k=1
k 2
9.2.17: Let f (x) = 1x . Then f is differentiable for x > 0. By the Mean Value Theorem
applied to f on [x, 1] (for x (0, 1)), there exists c (x, 1) such that f (c)(1 x) =
f (1) f (x). This means that 1 x (1 x). In particular, if x = 1 n1 , we get
1 (1 n1 ) (1 (1 n1 )) =
n
. Thus 1 n
(1 n1 ) . Hence for n sufficiently
large, aan+1
n
1 1 n1
n (1 n ) = ( n ) . Say above holds for all n N for some fixed
n2 1
integer N . Then for all n N , an an1 ( n1 ) an2 ( n3 ) aN ( N
n1 ) , so
P PN P PN P n1 N 1
that n an = n=1 an + n=N +1 an n=1 an + aN n=N +1 ( n1 ) , which converges
by the p-series convergence test.
1
9.4.2: i) Note that a2n+1 = (2n+1)n! , and a2n = 0. Apply the root test:
q q
1 1
lim sup{ 2n 0, 2n+1 (2n+1)n! } = lim sup 2n+1 (2n+1)n! = 0, so that the radius of conver-
gence is . The ratio test cannot be applied in the current form because the coef-
ficients in front of even powers of x are 0. However, the power series can be rewrit-
P 1
ten as f (x) = x k=0 (2k+1)k! (x2 )k . We can apply the ratio test to the power series
P 1
g(x) = k=0 (2k+1)k! xk : the radius of convergence of g is
1
(2n+1)n! (2n + 3)(n + 1)
lim 1 = lim = ,
(2(n+1)+1)(n+1)!
(2n + 1)
so any complex number x can be plugged into g, so we can plug in x2 , and so f (x) converges
P 1 2k P 1 2 k
at all x C. ii) f (x) = k=0 k! x = k=0 k! (x ) . iii) Obviously the power series for
x2 2
f (x) is e . iv) Thus f (x) is an antiderivative of ex . v) This antiderivative is not a finite
sum of products, sums, quotients, roots.
P P
9.5.1: Let f (x) = k=1 k1 xk . The radius of convergence is 1, f (x) = k=1 xk1 = 1x 1
,
and so f (x) = ln(1x)+C for some constant C. Since 0 = f (x) = ln(10)+C = C, it

P
follows that f (x) = ln(1 x). Thus k=1 k31k = f ( 31 ) = ln(1 13 ) = ln( 32 ) = ln(3/2).
1
, if k is even;
9.8.4: This is exactly in Exercise 9.4.2: ak = (k/2)! .
0, otherwise.
P k P 1 1 k
9.8.6: k=0 (1) 4k k!
= k=0 k! 4 = e1/4 .
P P P 2 j P
2k 3 2k+1 3 3 2j
9.8.7: k=0 3k (k+1)! = 2 k=0 3k+1 (k+1)! = 2 j=1 3j j! = 2 j=0 3j j! 1 =
3 2/3

2 e 1 .
9.9.3:
cos(x + y) + i sin(x + y) = E(i(x + y)) = E(ix + iy)

= E(ix) E(iy)
= (cos x + i sin x)(cos y + i sin y)
= cos x cos y sin x sin y + i(cos x sin y + sin x cos y).
By reading off the real and imaginary parts we get the desired equalities.
9.10.10:
1

sin n + 2 t = sin nt cos 21 t + cos nt sin 21 t

= sin n 21 t cos 12 t + cos n 12 t sin 12 t cos 12 t + cos nt sin 21 t

= sin n 12 t cos2 21 t + cos n 12 t sin 12 t cos 12 t + cos nt sin 21 t

= sin n 12 t sin n 12 t sin2 12 t + cos n 12 t sin 12 t cos 12 t
+ cos nt sin 21 t

= sin n 12 t + cos nt sin 21 t + cos nt sin 21 t

= sin n 12 t + 2 cos nt sin 21 t.
1

Thus by repetition of the same expansion with n 1 in place of n, sin n + 2
t =
Pn
sin 21 t + 2 k=1 cos(kt) sin 21 t, and the claim follows.

Introduction To Analysis

Uploaded by

Copyright:

Available Formats

Introduction To Analysis

Uploaded by

Document Information

Original Description:

Original Title

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

Introduction To Analysis

Uploaded by

Copyright:

Available Formats

Introdu

Section 4.3: Limit theorems 130

Section 9.3: Power series 274

What are the meanings of the following:

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

1.1 Statements and proof methods

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

Examples and non-examples 1.1.2

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

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

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

plies Q, or variants thereof, such as If P then Q, Given P , Q follows, Q

We list here most possible rephrasings of P implies Q:

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

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

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

and it is proved below with the truth table:

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

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

Exercises for Section 1.1

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

1.1.15 (Odd-even integers)

1.2 Statements with quantifiers

> 0 > 0 x, 0 < |x a| < |f (x) L| < .

there are x of specified type, T F F T

there are x of specified type, F T T F

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

Proposition 1.2.1 The negation of x P (x) is x P (x). The negation of x P (x)

Exercises for Section 1.2

x) There exists an integer m such that for all integers n, m + n is even.

1.3 More proof methods, and negation

Statement How to prove it

P and Q. Prove P . Prove Q.

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

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

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

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

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

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

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 =

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

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

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

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

Exercises for Section 1.3

1.3.14 Prove that the following are false:

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

Here, of course, x0 stands for 1. When we evaluate f at 0, we get a0 = a0 +a1 0+ +an 0n

or explicitly written out:

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

Similarly we can shorten products with the product sign :

Exercises for Section 1.4

1.4.3 Prove that

1.5 Proofs by (mathematical) induction

Induction is a two-step procedure:

Proof. Base case n = 1:

n(n + 1)(n + 2)(n + 3)