Sie sind auf Seite 1von 53

Lecture 1: Fundamentals

Matthew Rognlie
August 12, 2013
Contents
1 Logic 3
1.1 Basic denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Some rules of logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Quantiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Proof techniques and concepts 9
2.1 Direct proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Proving the contrapositive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Proof by contradiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Equivalence proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Proof by exhaustion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.6 Proof by induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.7 Constructive and non-constructive proofs . . . . . . . . . . . . . . . . . . . . 13
2.8 Necessity and sufciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 Set theory 15
3.1 Set basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Algebra of sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Orders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.5 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.6 Axiom of choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.7 Cardinality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Fields 31
4.1 Fields, orders, and ordered elds . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Rational numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Properties of the rational numbers . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Real numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5 Properties of the real numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.6 Complex numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1
A Dedekind cut construction of the reals. 53
2
1 Logic
1.1 Basic denitions
Denition 1.1 (Statement). If a sentence can be classied as true or false, it is called a
statement. (For a sentence to be a statement, it is not necessary that we actually know
whether it is true or false, but it must be true that the sentence is clearly either one or the
other.)
Examples.
2 +2 = 4. (True statement)
Every continuous function is differentiable. (False statement)
x
2
5x +6 = 0. (Not a statement, since we havent dened x, and it could be either
true or false depending on the value of x. We will later be able to adapt this into a
statement by using quantiers.)
Denition 1.2 (Negation). Let p stand for a given statement. Then p (read as not p)
represents the logical opposite. This operation is called the negation of p. When p is true,
then p is false, and when p is false, then p is true.
Example.
p : Today is Monday
p : Today is not Monday
Note that double negation leaves us with the original statement: (p) p.
Denition 1.3 (Conjunction). If p and q are statements, then the statement p and q is
called the conjunction of p and q. It is denoted by p q.
Example. Today is Monday
. .
p
and
..

the Math Camp is fun


. .
q
.
Denition 1.4 (Disjunction). If p and q are statements, then the statement p or q is called
the disjunction of p and q. It is denoted by p q.
Example. The pencil is blue
. .
p
or
..

red
..
q
.
Denition 1.5. [Implication] Astatement of the formif p, then q is called an implication
or conditional statement. It is denoted by p q.
1
Formally, it is dened to be equivalent
to the statement (p) q, so that the only way the implication p q can fail to be true is
if p is true and q is false.
1
Sometimes p q is used instead, but in math we like to use the notation to denote limits and for
various other purposes.
3
Example. If I regularly attend Math Camp
. .
p
, then
..

in three weeks I will be an expert at math


. .
q
.
Denition 1.6 (Equivalence). The statement p if and only if q is the conjunction of the
two implications p q and q p; formally, it is (p q) (q p). A statement of this
form is called an equivalence and is denoted by p q. It is also sometimes written as
p iff q.
Example.
You are the best basketball player in the world
. .
p
iff
..

your name is Lebron James and you play for the Miami Heat
. .
q
The following table, called a truth table, shows when the preceding ve concepts yield
values of true or false, depending on the truth or falsity of p and q. (The left two columns
show combinations of values of p and q, and the right columns show the values taken by
each formula given by the values p and q in the same row on the left.)
p q p p q p q p q p q
T T F T T T T
F T T F T T F
T F F F T F F
F F T F F T T
1.2 Some rules of logic
It is possible to derive many nice rules characterizing the logical language described in
the previous section. Here we mention a few of the most useful.
Proposition 1.7. An implication p q is equivalent to its contrapositive q p. The
converse q p is equivalent to the inverse p q:
(p q) (q p) (1)
(q p) (p q) (2)
Proof. We prove (1), the equivalence of the original implication and its contrapositive,
using a truth table:
p q p q p q q p
T T T F F T
F T T F T T
T F F T F F
F F T T T T
4
To prove (2), the equivalence of converse and inverse, replace p and q in (1) with p
and q. Then cancel all double negations to obtain (2):
((p) (q)) (((q)) ((p)))
((p) (q)) q p
We will discuss the contrapositive further in Section 2.2, where it will be used as a
method of proof.
Proposition 1.8 (De Morgans laws). The negation of a conjunction is equivalent to the dis-
junction of the negations, and the negation of a disjunction is equivalent to the conjunction of the
negations:
(p q) (p) (q) (3)
(p q) (p) (q) (4)
Proof. We prove (3) using truth tables:
p q p q (p q) p q (p) (q)
T T T F F F F
F T F T T F T
T F F T F T T
F F F T T T T
To prove (4), replace p and q in (3) with p and q. Take the negation of both sides
2
,
then cancel all double negations to obtain (4):
((p) (q)) ((p)) ((q))
(((p) (q))) (((p)) ((q)))
((p) (q)) (p q)
These laws are quite intuitive: they say that it is not true that the car is fast and blue is
equivalent to the car is not fast or not blue. Similarly, it is not true that the car is fast or blue is
equivalent to the car is not fast and not blue.
De Morgans laws in logic have a generalization using quantiers, as we will see in the
next section, as well as a natural counterpart in set theory, as we will see in Proposition
3.8. These rules turn out to be essential in dealing with sets.
Proposition 1.9 (Distributivity). Conjunction and disjunction are distributive over both them-
selves and each other:
(p (q r)) (p q) (p r) (5)
(p (q r)) (p q) (p r) (6)
(p (q r)) (p q) (p r) (7)
(p (q r)) (p q) (p r) (8)
2
Note that here we are implicitly using the fact that (a b) ((a) (b)). We could easily
prove this, and many other such obvious facts, using truth tables. Ideally, we would develop and prove
a rich vocabulary of rules that can be combined to obtain virtually any other rule; but were teaching a
math-for-economists course here, not Logic 1, and we dont have the time.
5
Equivalence (6), for instance, states that the car is blue and the car is fast or broken is
equivalent to the car is blue and fast or the car is blue and broken. (Try to think about the
intuition behind the other lines.) Like De Morgans laws, the distributivity property has
both a generalization using quantiers and a natural counterpart in set theory.
Finally, we have the following unsurprising proposition.
Proposition 1.10. Conjunction and disjunction are both commutative:
(p q) (q p) (9)
(p q) (q p) (10)
and also both associative:
((p q) r) (p (q r)) (11)
((p q) r) (p (q r)) (12)
1.3 Quantiers
Earlier we observed that the sentence
x
2
5x +6 = 0
is not a statement, because x is not dened. We need to consider it in a particular context
for it to become a statement. We usually write:
p(x) : x
2
5x +6 = 0
For a specic value of x, p(x) becomes a statement that is either true or false. For example,
p(2) is true and p(4) is false.
We may remove the ambiguity by using a quantier. For instance, we may write:
For every x, x
2
5x +6 = 0
This is a statement, since it is false. In symbols, we write
x : p(x)
where the universal quantier is read for every....
The sentence
There exists an x such that x
2
5x +6 = 0
is also a statement, and it is true. In symbols we write
x : p(x)
where the existential quantier is read there exists.... (The colon : is shorthand for
the phrase such that. The vertical bar [ is also often used for this purpose.)
6
Occasionally, we need to express uniqueness along with existence. The sentence
There exists a unique x such that x
2
5x +6 = 0
is a statement, and it is false, since the equation holds for both x = 2 and x = 3. In
symbols, we write:
!x : p(x)
There are also many ways to write the statement above without making use of ! notation.
For instance, the following statement is equivalent to the statement above (think about
why):
x y (p(y) x = y)
If we switch the order of a universal or existential quantier and the negation opera-
tion, we get the opposite kind of quantier. (This is the quantier generalization of De
Morgans laws from Proposition 1.8.)
Proposition 1.11 (De Morgans laws, quantier version). The following equivalences hold:
x : p(x) (x : p(x)) (13)
x : p(x) (x : p(x)) (14)
Example. Let p(x) be student x loves macro. Then (13) says that there exists a student
who does not love macro is equivalent to it is not true that all students love macro. Similarly,
(14) says that all students do not love macro is equivalent to it is not true that there exists a
student who loves macro. This is quite intuitive.
Additionally, if a statement does not depend on the variables being quantied, we
can freely move it in and out of the quantiers. (This is the quantier generalization of
distributivity from Proposition 1.9.)
Proposition 1.12 (Distributivity, quantier version). The following equivalences hold:
q (x : p(x)) (x : q p(x)) (15)
q (x : p(x)) (x : q p(x)) (16)
q (x : p(x)) (x : q p(x)) (17)
q (x : p(x)) (x : q p(x)) (18)
Finally, the order of multiple quantiers is an extremely common point of confusion
in mathematical proofs, and indeed logical reasoning in general. Consider the following
sentence:
y x Student x hates day y of math camp
This says that there exists some specic day y of math camp that every student x hates.
This is as opposed to the following sentence, created from switching the order of the two
quantiers:
x y Student x hates day y of math camp
7
This says that for every student x, there exists some day y of math camp that x hates.
But unlike before, there could be a different day y for each student x! This statement,
beginning with xy, is in fact weaker than the rst statement, which begins with y x. It
is implied by the rst statement, but not vice versa. Such problems do not exist when we
are dealing with two of the same kind of quantier, as switching x with y (or x with
y) leaves the meaning of a statement unchanged. Formally:
The following implication holds for any p(x, y):
y x : p(x, y) x y : p(x, y) (19)
The opposite implication does not always hold for p(x, y). In particular, for some p(x, y)
we have:
x y : p(x, y) y x : p(x, y) (20)
The following equivalences hold for any p(x, y):
x y : p(x, y) y x : p(x, y) (21)
x y : p(x, y) y x : p(x, y) (22)
As your lecturer, I am more worried by the rst statement about math camp than the
second one. Its frightening to imagine that I will give a single talk that everyone dislikes.
On the other hand, its perfectly reasonable that everyone will dislike some talk that I give.
After all, there are 11 days of this stuff.
8
2 Proof techniques and concepts
We will now discuss various proof techniques. The goal here is not to dene distinct
proof techniques in some rigorous way. (Indeed, in a few instances the boundaries
are fuzzy, or one technique is arguably just a special case of the other.) Instead, the idea
is to convey various patterns for mathematical argumentspatterns that you will nd
yourself reusing over and over again.
2.1 Direct proof
Suppose we already know that A A
1
, A
1
A
2
, . . . , A
n1
A
n
. We may conclude
that A A
n
. This direct proof combines if, then statements (implications) that we
already know to be true and produces a new if, then statement.
Example. For instance, suppose we want to prove the following statement.
If a divides b
. .
A
1
and
..

b divides c
. .
A
2
then
..

a divides c
. .
A
3
Proof.
By denition, a dividing b means that b = ak
1
for some natural number k
1
, and b di-
viding c means that c = bk
2
for some integer k
2
. Combining these results, c = bk
2
= ak
1
k
2
.
Let k = k
1
k
2
. Now k is a natural number and c = ak, so by the denition of divisibility, a
divides c.
Even this very simple proof combines several if, then statements, some of themimplicit,
to arrive at its conclusion. We apply the denition of divisibility three times (x divides y
m : y = xm), along with the associative law of arithmetic when we dene k. Note that
we make no attempt here to write the actual proof in the language of formal logic; that is a
pain even when a proposition is simple, and becomes extremely impractical whenever we
are proving anything complicated. It is important, however, to write proofs sufciently
clear-cut that a reader could translate them into formal terms if necessary.
2.2 Proving the contrapositive
Sometimes, in order to prove an implication p q, we prove its contrapositive q
p instead. This is possible thanks to the equivalence of p q and q p, as we
demonstrated in Proposition 1.7.
For example, the implication
If an apple is tasty
. .
p
, then
..

it is green
. .
q
has contrapositive
If an apple is not green
. .
q
, then
..

it is not tasty
. .
p
9
The equivalence of these two statements probably seems obvious, and indeed people of-
ten pass between an implication and its contrapositive without ever having seen the for-
mal notion of contrapositive. In mathematical settings, however, sometimes the equiv-
alence between an implication and its contrapositive is not so obvious, and it is useful to
remind yourself of the idea.
Example. Suppose we want to prove that
If xand yare two integers for which x + y is even
. .
p
, then
..

x and y both are either even or odd.


. .
q
Proof. It is easier to think about the contrapositive.
If one of x and y is even and one is odd
. .
q
, then
..

x + y is odd.
. .
q
Now, either x or y is even, and the other is odd. There is no loss of generality to suppose
that x is even and y is odd. Then by denition there are integers k and m for which
x = 2k + 1 and y = 2m. Then we compute the sum x + y = 2k + 2m + 1 = 2(k + m) + 1,
which is odd by denition.
2.3 Proof by contradiction
Often, in order to prove an implication p q, we show that p q is impossible. Gener-
ally in this situation, we think of ourselves as assuming p and q and trying to obtain
a contradiction from that assumption. This is permissible because, as the following truth
table demonstrates, (p q) (p q):
p q p q q p q
T T T F F
F T T T F
T F F F T
F F T T F
Example. There are innitely many primes.
Proof. Suppose to the contrary that there are only nitely many primes, which we list
as r
1
, r
2
, . . . , r
n
. We construct their product plus one, s = r
1
r
2
r
n
+ 1. s is not divisible
by any of the primes r
1
, r
2
, . . . , r
n
and therefore must be a distinct prime itself, which
contradicts the assumption that r
1
, r
2
, . . . , r
n
was the full list of primes.
3
3
This example is slightly confusing, because at rst glance we dont seem to be proving any implication
p q; rather, we are simply showing that the statement q there are innitely many primes is true. In a
sense, proof by contradiction here is just proof by double negationshowing that it is impossible that q is
not true, or (q). But we can interpret p as a statement consisting of the other axioms or facts of number
10
2.4 Equivalence proofs
When asked to prove an equivalence p q, there are two implications to prove, and
usually we will provide separate demonstrations for each of them, showing that p q
and q p. Sometimes one step is much easier than the other, and proofs of the two might
be very different.
Often we must prove that a set of more than two statements, p
1
, . . . , p
n
, are all equiv-
alent. Many different approaches can be fruitful depending on the circumstances. For
instance, if there are three statements p
1
, p
2
, p
3
, we might show that p
1
p
2
(by prov-
ing each implication separately) and then p
2
p
3
(again by proving each implication
separately), nally combining the two to obtain p
1
p
3
. Another common approach
is to establish a circular chain of implications, proving that p
1
p
2
, p
2
p
3
, and that
p
3
p
1
.
Even more elaborate combinations are sometimes useful. For instance, if A, B, C, D,
and E below are statements, proving the six implications represented by the arrows is one
way to show that the ve statements are equivalent. (This format for representing impli-
cations is a special case of a more general concept called a directed graph. All we need to
check is that we can follow the arrows from any statement to any other statement.)
A
B
C
D E
2.5 Proof by exhaustion
A proof by exhaustion, sometimes called a proof by cases, splits up the statement to
be proved into a nite number of cases and then shows that the statement holds in each
particular case. Generally, this method of proof involves two stages:
1. A proof that the cases are truly exhaustivethat every possible instance of the state-
ment to be proved matches the conditions of (at least) one of the cases.
2. A proof of the statement in each of the cases.
theory. Then the contrast between direct proof and proof by contradiction is very clear: with direct proof,
we start with the truth of p and combine already-proven implications to show the truth of q, while with
proof by contradiction we start with the truth of p and the falsity of q and show that something must go
wrong.
11
Example. Show that for any integer square n, we can write (for some integer k) either
n
2
= 4k or n
2
= 4k +1.
Proof. We split into cases. (In this instance, the exhaustiveness of the cases is obvious.)
n is even. Then we can write n = 2m for some m, and we obtain n
2
= 4m
2
. Then the
proposition is true for k = m
2
.
n is odd. Then we can write n = 2m + 1 for some m, and we obtain n
2
= 4(m
2
+
m) +1. Then the proposition is true for k = m
2
+ m.
2.6 Proof by induction
In the simplest case, a proof by induction may be used to show that some property P(n)
is true for all natural numbers.
Algorithm.
First stage: Prove that the statement P(1) is true. (Base case)
Second stage: Assume that P(n) is true, and use this to prove that P(n + 1) is also
true. This completes the proof. (Induction step)
The idea here is analogous to showing that a line of dominos will fall. First you need
to prove that the rst domino will fall, and second you need to prove that when each
domino falls, it topples the next domino.
Example. Show that n < 2
n
for any integer n 1.
Proof. The base case 1 < 2 is easy. For the induction step, we assume that n < 2
n
for arbitrary n 1, and then obtain 2
n+1
= 2 2
n
> 2 n n + 1. This completes the
induction, and we conclude that n < 2
n
for all n.
We also sometimes use a slight twist on this concept, called strong induction. (Some-
times our original denition of induction is called weak induction to emphasize the dif-
ference.
4
) Here the algorithm is
First stage: Prove that the statement P(1) is true. (Base case)
Second stage: Assume that P(k) is true for all k = 1, . . . , n, and use this to prove that
P(n +1), is also true. (Induction step)
Example. Show that each positive integer can be written as a sum of distinct powers of
2.
4
If youre very clever, you might realize that strong induction is actually a special case of weak induction,
if we apply weak induction to the statement Q(n) : P(k) is true for all k n.
12
Proof. Clearly this is true for n = 1 = 2
0
. Now suppose that it is true for all k = 1, . . . , n,
and consider n + 1. There exists some m such that 2
m
n + 1 < 2
m+1
. By the strong
induction assumption, we can represent (n + 1) 2
m
as a sum of distinct powers of 2.
Moreover, none of these powers will be 2
m
, since (n + 1) 2
m
< 2
m
. Therefore we can
also represent n + 1 as a sum of distinct powers of 2, appending 2
m
to the representation
of (n +1) 2
m
. This completes the induction, and we conclude that any n can be written
as a sum of distinct powers of 2.
Although proving that some P(n) is true for all natural numbers N is the canonical
case of induction, the principle may be adapted to other, related settings as well. For
instance, we can start from a number other than 1 (say, 3) and after proving the n = 3
base case and the induction step for any n 3, conclude that a statement is true for any
n 3. We can also move downward, establishing a base case k and an induction step
P(n) P(n 1), and concluding that P(n) is true for all n k.
More generally, we can do induction on a sequence a
1
, a
2
, . . ., proving that P(a
1
) is
true for the base case a
n
and that P(a
n
) P(a
n+1
) for the induction step, and concluding
that P(a
n
) is true for any a
n
in the sequence. (This is an immediate consequence of the
basic induction principle, tantamount to doing induction on Q(n) P(a
n
).) Induction
can be nite; perhaps we start at base case n = k, and our induction step P(n) P(n 1)
only works when n 1. In this case, induction still shows that P(n) holds for all 0 n
k. There are many other clever ways to adapt the principle of induction as well.
2.7 Constructive and non-constructive proofs
A constructive proof, or a proof by construction, demonstrates the existence of a math-
ematical object with certain properties by explicitly creating or providing a method to
create that object. In contrast, a non-constructive proof proves the validity of a proposi-
tion without considering a example.
It may not initially be obvious that there is any way to demonstrate the existence of
a mathematical object with certain properties without actually providing such an object.
In fact, this is a very common idea.
Example 1. Prove that there exists some x > 0 solving f (x) = x
2
4x +1 = 0.
Constructive proof. Evaluate the expression for x = 2 +

3.
Non-constructive proof. f (2) = 3 and f (4) = 1. Since f is a continuous function,
the Intermediate Value Theorem from calculus (which we will formally cover later) then
implies that there is some x (2, 4) such that f (x) = 0.
This proof is non-constructive because it tells us that some x satisfying the conditions
in the statement exists, but does not specify (or offer a way to specify) a particular x.
Example 2 (taken from Wikipedia). Prove that there exist irrational numbers a and b
such that a
b
is rational.
13
Constructive proof. Take a =

2 and b = log
2
9, and observe that a
b
= 3. We know
that a =

2 is irrational. (We will prove this in a subsequent section.) log


2
9 is irrational
because if it were equal to
m
n
, then we would have 9
n
= 2
m
, which is impossible because
the left is odd while the right is even. 3 is clearly rational, making this a suitable example.
Non-constructive proof. Again use the fact that

2 is irrational. Consider the number
q =

2
. Either it is rational or it is irrational. If q is rational, then the theorem is true for
a =

2, b =

2. If q is not rational, then the theorem is true for a =

2
and b =

2,
since
_

2
_

2
=

2
=

2
2
= 2
This proof is non-constructive because we havent constructed a single example a, b of the
theorems correctness. Instead, we have demonstrated that one of two possible examples
must be valid.
5
2.8 Necessity and sufciency
The concepts of necessity and sufciency are not proof techniques per se, but they are
very important for mathematical reasoning.
Denition 2.1 (Necessary and sufcient conditions). A necessary condition for a state-
ment must be satised for the statement to be true. Formally, a statement p is a necessary
condition for q if q p. A sufcient condition for a statement is one that, if satised,
ensures that the statement is true. Formally, a statement p is a sufcient condition for q if
p q.
For instance, n 0 is a necessary condition for n to be a square. Conversely, n being a
square is a sufcient condition for n 0.
We will often talk about necessity and sufciency when discussing optimization in
economics. Often a certain set of conditions will be necessary for optimality, and a strictly
larger set will be both necessary and sufcient for optimality. Other times, some conditions
will be merely sufcient.
5
For the curious, it turns out that

2
is irrational.
14
3 Set theory
3.1 Set basics
We will not try to formally or axiomatically dene sets. (That becomes very complicated,
very quickly.) Instead, we will intuitively think of a set as any collection of well dened
and distinct objects. The objects in a set are called its elements. If x is an element of set
A, we write
x A
and otherwise we write
x / A
Some details of set notation:
1. We describe a set by writing its elements between two curly brackets. Order and
repetition do not matter.
A = x, y = y, x
B = x, x = x
x ,= x
2. We can specify a set by a property
S = x : P(x) is a true statement
A = x : x Nand even
= 2, 4, 6, . . . ,
= 2n : n N
Some useful sets:
N = 0, 1, 2, 3, . . . or 1, 2, 3, . . . set of natural numbers
Z = 0, +1, 1, +2, 2, . . . set of integers
Q = p/q : p Z, q N set of rational numbers
R = set of real numbers, R
+
= [0, ), R
++
= (0, )
= empty set
Denition 3.1 (Subset). If every member of A is also a member of B, then we say that
A is a subset of B and write A B. Formally, to prove that A B, we must show
x A x B.
15
Denition 3.2 (Set equality). Two sets A and B are equal if A B and B A. We write
this as A = B. Equivalently, two sets A and B are equal if x A x B.
Denition 3.3 (Power set). The set of all subsets of S is called the power set of S and is
denoted by
2
S
= T : T S
Denition 3.4 (Index set). Suppose that for each element in a nonempty set I there
corresponds an object x

. Then we say that the collection of all objects x

,
x

: I
is an indexed collection of objects, with I as the index set.
Often the objects x

under consideration are themselves sets. (Usually, in this case,


we use notation of the form A

rather than x

to denote sets.) If so, we say we have an


indexed collection of sets.
Formally, we may view an indexed collection of objects as a function from the index
set I to the set of possible objects. (This will be clearer once we dene the concept of
function, later in this section.)
Recall that when we are dealing with nitely many objects, we often number them
x
1
, . . . , x
n
. In doing so, we are using 1, . . . , n as an index set. The formal concept of
index set allows us to generalize this notation to cover collections of arbitrarily many
objects, even when there are innitely many, possibly too many for us to even use the
natural numbers Nas an index set.
3.2 Algebra of sets
Denition 3.5 (Intersection). If A and B are two sets, then the intersection A B consists
of those elements that are in both A and B.
x A B x A x B
If A B = , we say that A and B are disjoint.
More generally, if A

is a collection of sets indexed by I, we dene the intersection


to consist of those elements that are in all A

:
x

I
A

I : x A

Denition 3.6 (Union). If A and B are two sets, then the union A B consists of those
elements that are either in A or in B.
x A B x A x B
More generally, if A

is a collection of sets indexed by I, we dene the union to consist


of those elements that are in at least one A

:
x

I
A

I : x A

16
Denition 3.7 (Complement). If A and B are two sets, then the complement of B in A,
also called the set-theoretic difference of A and B, is denoted by A B and consists of
those elements that are contained in A but not B:
x A B x A x / B
Some useful properties for the algebra of sets follow from logical rules.
Proposition 3.8 (De Morgans laws, set version). Let S, T
1
and T
2
be sets. Then:
S (T
1
T
2
) = (S T
1
) (S T
2
) (23)
S (T
1
T
2
) = (S T
1
) (S T
2
) (24)
More generally, if T

is a collection of sets, then:


S
_

_
=
_

S T

(25)
S
_
_

_
=

S T

(26)
Proof. We will content ourselves with proving (23) and (24). (To prove (25) and (26), we
would use the relevant quantier rules (13) and (14) instead.)
For (1), we observe that by denition x S (T
1
T
2
) if (x S) (x T
1
x T
2
).
This is equivalent to:
(x S) ((x T
1
) (x T
2
)) (by (3) in Proposition 1.8)
((x S) (x T
1
)) ((x S) (x T
2
)) (by (6) in Proposition 1.9)
x (S T
1
) (S T
2
) (by denition of and )
Similarly, for (2), we observe that by denition x S (T
1
T
2
) if (x S) (x
T
1
x T
2
). This is equivalent to:
(x S) ((x T
1
) (x T
2
)) (by (4) in Proposition 1.8)
((x S) (x T
1
)) ((x S) (x T
2
)) (by (5) in Proposition 1.9)
x (S T
1
) (S T
2
) (by denition of and )
Proposition 3.9 (Distributivity, set version). Let A, B, and C be sets. Then:
A (B C) = (A B) (A C) (27)
A (B C) = (A B) (A C) (28)
More generally, if B

is a collection of sets, then:


A
_
_

_
=
_

(A B

) (29)
A
_

_
=

(A B

) (30)
17
Proof. For (27), by denition x A (B C) if (x A) ((x B) (x C)), which
by (6) in Proposition 1.9 is equivalent to ((x A) (x B)) ((x A) (x B)). By
denition, this is equivalent to x (A B) (A C), as desired.
For (28), by denition x A (B C) if (x A) ((x B) (x C)), which by
(7) in Proposition 1.9 is equivalent to ((x A) (x B)) ((x A) (x C)). By
denition, this is equivalent to x (A B) (A C), as desired.
The proofs of (29) and (30) are similar, except that we use the quantier distributivity
rules from Proposition 1.12.
Finally, we dene the concept of a partition.
Denition 3.10 (Partition). A partition of a set A is a set I of non-empty subsets of A
whose union is A and that are pairwise disjoint. Formally, I is a partition of A iff
a ,= for all a I.

aI
a = A.
a b = for any a, b I where a ,= b.
A partition J is ner than partition I if a J, a
/
I such that a a
/
. J is coarser than I
if I is ner than J.
Note that a : a A, the partition of A into a subset for each element of A, is the
nest partition. A is the coarsest partition.
Partitions are important in studies of information and measure, and sets are beneath
most of the mathematics used in ecnoomics. You will encounter them directly in classes
like 14.122 and 14.126.
3.3 Relations
When we dened sets, we said that the order does not matter:
a, b = b, a
There are times, however, when the order is important. For example, in analytic geometry,
the coordinates of a point (x, y) represent an ordered pair.
(1, 3) ,= (3, 1)
When we wish to indicate that the order matters, we enclose the elements in parentheses:
(a, b).
Denition 3.11 (Ordered pairs and tuples). An ordered pair (a, b) is an ordered list of two
(not necessarily distinct) objects. More generally, an n-tuple (a
1
, a
2
, . . . , a
n
) is an ordered
list of n objects.
6
(An ordered pair is a 2-tuple.) Two n-tuples are equal if corresponding
elements are equal:
(a
1
, a
2
, . . . , a
n
) = (b
1
, b
2
, . . . , b
n
) iff a
i
= b
i
for all i = 1, 2, . . . , n
6
In more axiomatic treatments these concepts are built up from the foundation of set theory, with an
ordered pair dened as a certain kind of set and an n-tuple dened as a nested sequence of ordered pairs.
This formalism is unlikely to be useful to us.
18
Denition 3.12 (Cartesian product). If A and B are sets, then the Cartesian product of A
and B, written A B, is the set of all ordered pairs (a, b) such that a A and b B. In
symbols,
A B = (a, b) : a A and b B
Denition 3.13 (Relation). Let A and B be sets. A relation between A and B is any subset
R of A B. We say that a A and b B are related by R if (a, b) R, and we often
denote this by writing aRb. If B = A, then we can speak of a relation R A A being
a relation on A.
Denition 3.14 (Properties of relations). A relation R on a nonempty set X is said to be
reexive if xRx for each x X
total if for any x, y X, either xRy or yRx holds
symmetric if for any x, y X, xRy implies yRx
antisymmetric if for any x, y X, xRy and yRx imply x = y
transitive if xRy and yRz imply xRz for any x, y, z X
Denition 3.15 (Equivalence relation). A relation R on a set S is an equivalence relation
if it has the following properties for all x, y, z in S.
Reexive: xRx
Symmetric: If xRy, then yRx
Transitive: If xRy and yRz then xRz
Denition 3.16 (Equivalence class). An equivalence class of x X is given by
[x]
R
= y X : yRx
The set of all equivalence classes in X forms a partition of X. (Try proving this.)
Example. Suppose that we dene X and the relation R on X to be
X = (a, b) : a, b 1, 2, . . .
(a, b)R(c, d) ad = bc
The equivalence class of some (a, b) X is the set
[(a, b)]
R
=
_
(c, d) X :
c
d
=
a
b
_
(In fact, we will dene the rational numbers in a very similar manner.)
19
3.4 Orders
Denition 3.17 (Preorder). A relation R on X is called a preorder if it is reexive and
transitive.
Denition 3.18 (Partial order). A relation R on X is called a partial order if it is a preorder
that is also antisymmetric. In other words, R is a partial order if it is reexive, transitive,
and antisymmetric.
Denition 3.19 (Total order). A relation R on X is called a total order if it is a partial order
that is also total. In other words, R is a partial order if it is transitive, antisymmetric, and
total.
7
If we denote a total order R by , then we often write x < y to mean that x y
and x ,= y.
Examples:
In individual choice theory, a preference relation _ on a nonempty set X is dened
as a total preorder on X. For any two alternatives x, y X we must either weakly
prefer x to y (x _ y) or y to x (y _ x), but we are also allowed to be indifferent
between x and y (x _ y and y _ x, written as x y) even if x ,= y. Preferences are
transitive: if we weakly prefer x to y and y to z, then we weakly prefer x to z as well.
The inclusion relation on sets is a partial order. It is reexive (A A), transitive
(A B and B C imply A C), and antisymmetric (A B and B A imply
A = B). It is generally not, however, a total order, because often we cannot write
either A B or B A.
Anatural partial order on pairs (a, b) of real numbers is given by (a
1
, b
1
) _ (a
2
, b
2
)
(a
1
a
2
) (b
1
b
2
). This is not a total order because we cannot compare, say, (1, 4)
and (2, 3).
The less-than-or-equal relation on the integers Z is a total order.
Sometimes rather than dening a total order, we dene a strict total order, which re-
places the antisymmetric and total requirements with a slightly modied version called
trichotomy.
Denition 3.20 (Strict total order). A relation R on X, which we will denote by <, is
called a strict total order if it is transitive and trichotomous (meaning that for any x and
y, exactly one of the following is true: x < y, y < x, or x = y).
It is easy to go back and forth between total orders and strict total orders, and we will
frequently do this.
Proposition 3.21. If is a total order, then the relation < dened by x < y ((x
y) (x ,= y)) is a strict total order. If < is a strict total order, then the relation dened by
x y ((x < y) (x = y)) is a total order. This denes a relation that associates each total
order with a unique strict total order, and vice versa.
7
Note that reexivity is implied by totality, so we can leave it off the list.
20
Upper and lower bounds.
Denition 3.22 (Upper and lower bounds). Let be a partial order on X. Then a subset
A X has an upper bound in X if there exists some x X such that a x for all
a A. Similarly, A X has a lower bound if there exists some x X such that a x
for all a A. In these cases, x itself is often called a upper bound and a lower bound,
respectively.
Denition 3.23 (Least upper bound and greatest lower bound). Let be a partial order
on X, and A be a subset of X. Then a least upper bound of A, or supremum, is an element
x X such that a x for all a A, and also y x for any y X such that a y for all
a A. If a least upper bound of A exists, we denote it by sup A.
A greatest lower bound of A, or inmum, is an element x X such that a x for all
a A, and also y x for any y X such that a y for all a A. If a greatest lower
bound of A exists, we denote it by inf A.
Note that when sup A or inf A exists, the denition plus antisymmetry implies that it
is unique.
Denition 3.24 (Least upper bound and greatest lower bound properties). Let be a
partial order on X. has the least upper bound property if every nonempty subset A
that has an upper bound also has a least upper bound. has the greatest lower bound
property if every nonempty subset A that has a lower bound also has a greatest lower
bound.
Usually we only hear about the least upper bound property, not its greatest lower
bound counterpart. Why? It turns out that these two properties are equivalent, and the
convention is to use the name least upper bound property when referring to this pair
of equivalent properties.
Proposition 3.25. Let be a partial order on X. Then if has the least upper bound property,
it has the greatest lower bound property, and vice versa.
Proof. Suppose has the least upper bound property on X, and let A be a set with a
lower bound. Dene B = y : y a for all a A as the set of all lower bounds of A. B
is nonempty and has an upper bound (take any element in A), and therefore it has a least
upper bound, which we denote by b = sup B. We will show that b is, in fact, the greatest
lower bound of A.
By construction, b y for all y B: if b is a lower bound, it is the greatest lower
bound. Moreover, since every a A is an upper bound for B, we must have b a for all
a A by the denition of least upper bound. Thus b is indeed a lower bound, completing
the proof.
The converse is proven by modifying the argument in the obvious way: ipping all
instances of with and all instances of upper and least upper with lower and
greatest lower, respectively.
21
Examples.
If A = (0, 1) (i.e. the set of all real numbers between 0 and 1, excluding 0 and 1
themselves), then any real number x 1 is an upper bound for A, and any real
number x 0 is a lower bound for A. The least upper bound and greatest lower
bound, however, are unique: sup A = 1 and inf A = 0.
The integers Z satisfy the least upper bound property. As we will show, the real
numbers R also satisfy the property, while the rationals Q do not.
3.5 Functions
Denition 3.26 (Function). Let X and Y be any two nonempty sets. By a function f that
maps X into Y, denoted as f : X Y, we mean a relation f X Y such that
1. for every x X, there exists a y Y such that x f y
2. for every y, z Y with x f y and x f z, we have y = z.
We generally write the relation x f y as y = f (x).
Some further details:
Here X is called the domain of f and Y the codomain of f . the range of f is dened
as
f (X) = y Y : y = f (x) for some x X
The set of all functions that map X into Y is denoted by Y
X
. For example, R
[0,1]
is
the set of all real-valued functions on [0, 1].
When a function consists of just a few ordered pairs, it can be described simply by listing
them. Typically, however, there are too many pairs to list, and we must describe the
function by specifying the domain and codomain and providing a rule for obtaining the
second element in the ordered pair from the rst element.
For instance, if we want to dene the function f corresponding to all ordered pairs
(x, x
2
) : x R, we cant possibly list the innite number of pairs. Instead we specify
a rule that f takes each x R to its square, i.e. f (x) = x
2
.
Denition 3.27 (Surjective function). A function f : A B is called surjective (or onto)
if B = f (A).
Denition 3.28 (Injective function). A function f : A B is called injective (or one-to-
one) if for all a and a
/
in A, f (a) = f (a
/
) implies that a = a
/
.
Denition 3.29 (Bijective function). A function f : A B is called bijective if it is both
surjective and injective.
Denition 3.30 (Image). Suppose that f : A B. If C A, then we dene the image
f (C) to be the subset f (C) = f (x) : x C of B.
22
Surjective
not injective
A B
1
2
3
4
w
x
y
Injective
not surjective
A B
1
2
3
w
x
y
z
Bijective
A B
1
2
3
4
w
x
y
z
Not surjective
or injective
A B
1
2
3
4
w
x
y
z
Figure 3.1: Surjective, injective, and bijective functions.
Denition 3.31 (Inverse image). Suppose that f : A B. If D B, then we dene the
inverse image f
1
(D) (also called preimage), to be the subset y : f (y) D of A.
The concept of inverse image is distinct from the concept of an inverse function. For
instance, given some point y B, the inverse image f
1
(y) of the singleton set con-
taining y may be empty, or it may have more than one member. Only when f is bijective
can we dene a true inverse function.
Note that f
1
( f (B)) = B is not necessarily true when f is not injective. (For instance,
consider B = 1, 2, 3 in the surjective, not injective example in Figure 3.1.) We can only
say that B f
1
( f (B)). Similarly, f ( f
1
(B)) = B is not necessarily true when f is not
surjective. Indeed, there are many such subtleties when taking the images and inverse
images of sets. We list several relevant facts in the following proposition.
Proposition 3.32. Let f : X Y be a function, and let A

be a collection of subsets of X
indexed by I and B

be a collection of subsets of Y indexed by J. Also let A


1
, A
2
be subsets of X
23
and B
1
, B
2
be subsets of Y. Then
f
1
( f (B)) B (31)
f ( f
1
(B)) B (32)
f (A) B A f
1
(B) (33)
A
1
A
2
f (A
1
) f (A
2
) (34)
B
1
B
2
f
1
(B
1
) f
1
(B
2
) (35)
f
_
_
I
A

_
=
_
I
f (A

) (36)
f
_

I
A

I
f (A

) (37)
f
1
_
_
_
J
B

_
_
=
_
J
f
1
(B

) (38)
f
1
_
_

J
B

_
_
=

J
f
1
(B

) (39)
f
1
(Y B
1
) = X f
1
(B
1
) (40)
In (31), (32), and (37), where there are inclusion signs rather than equalities =, there exist
examples where there is strict inclusion, so that the statement with an equality sign would be false.
Furthermore, there is no analog of (40) with images f rather than inverse images f
1
: there are
cases both where f (X A
1
) , Y f (A
1
) (when f is not injective) and where f (X A
1
) ,
Y f (A
1
) (when f is not surjective).
Denition 3.33 (Inverse function). Suppose that f : A B is bijective. The inverse
function of f is the function f
1
given by f
1
(y) = x y = f (x).
8
Denition 3.34 (Composition). Suppose that g : A B and f : B C are functions. The
composition of f and g is denoted by f g and given by ( f g)(x) = f (g(x)).
The concepts of inverse function and composition can be combined nicely: both ( f
f
1
)(x) and ( f
1
f )(x) equal simply x.
Proposition 3.35 (Associativity of function composition). Let h : A B, g : B C, and
f : C D be functions. Then (h g) f = h (g f ).
Proof. This follows from the denition of composition. For any x A, ((h g) f )(x) =
(h g)( f (x)) = h(g( f (x)) = h((g f )(x)) = (h (g f ))(x).
8
The requirement that f is bijective is necessary and sufcient for this function to be uniquely dened
for every y B.
24
3.6 Axiom of choice
9
In Denition 3.12, we dened the Cartesian product of two sets A
1
and A
2
as the set of
all ordered pairs (a
1
, a
2
), with a
1
A
1
and a
2
A
2
. What if we want to extend this
notion and dene the Cartesian product of an arbitrary collection A

of sets, indexed
by I? The natural way to generalize the Cartesian product is to dene it as the set of all
functions f , called choice functions, from the index set I to the union
I
A

such that
f () A

for every I. Our earlier denition of Cartesian product for two sets A
1
and
A
2
is a special case of this general denition, for index set I = 1, 2.
Yet it is not immediately clear that this denition is meaningful. Without additional
assumptions, we cannot guarantee that such a function f will even exist. This implies that
we cannot guarantee that the Cartesian product of A

is nonempty, even when each set


A

is nonempty.
Now, when the index set is nite (for instance, I = 1, 2, . . . , n), there is a clear pro-
cedure to construct a choice function f . We just list the value of f (i) for each i = 1, . . . , n,
picking some element of A
i
each time. Indeed, for the nite case we can formally prove
the existence of a choice function by induction, assuming the existence of a choice func-
tion for index sets of size n 1, then dening the function for one additional element to
complete the proof for index sets of size n. But when the index set is innite, no such
proof is possible. The situation seems particularly hopeless when the set has too many
elements to count
10
, like R.
Although this may seem like a technical nitpick, it turns out to be a remarkably deep
issue in set theory, and mathematicians ultimately decided that the best route is simply
to take the existence of f as an axiom, called the axiom of choice. They even proved that
there is an entirely consistent way of dening set theory without this axiom; the axiom of
choice can therefore never be proven from the other axioms of set theory. The axiom of
choice is a key step in many non-constructive proofs, since the axiom of choice tells us that
a choice function exists, but does not tell us how to nd it.
Denition 3.36 (Axiom of choice). Let A

be a collection of nonempty sets indexed


by I. Then there exists a function, called a choice function, f : I
I
A

such that
f () A

for every I.
Now we can safely dene the generalization of Cartesian product, knowing that the
Cartesian product of nonempty sets will always be nonempty.
Denition 3.37 (Cartesian product). Let A

be a collection of nonempty sets indexed


by I. Then the Cartesian product
I
A

is the set of all choice functions f on A

.
Despite its apparent simplicity, the axiom of choice has a number of remarkable impli-
cations, some of which are deeply unintuitive.
11
Two of the most useful are the following,
which are actually equivalent to the axiom of choice (they can be proven from the axiom
of choice, or used to prove the axiom of choice).
9
This is more technical than other sections (though it does not go into much detail), and it is possibly to
lead a perfectly satisfying life as an economist without internalizing it.
10
We will dene precisely what this means in the next section.
11
Look up the Banach-Tarski paradox
25
Proposition 3.38 (Zorns lemma). Let (P, ) be a partially ordered set, and suppose that for
any subset T P such that T is totally ordered (i.e. x, y : x y y x), T has an upper
bound. Then P has at least one maximal element, dened as being an element x P such that
x y for all y P.
Zorns lemma is a key step for proving foundational results in many subelds of math-
ematics. In linear algebra, it is used to prove that every vector space has a basis; in
topology, it is used to prove that the product space of compact spaces is always compact
(Tychonoffs theorem); and in functional analysis, it is used to prove the Hahn-Banach
theorem, which will make a disguised cameo in 14.121 as the separating hyperplane
theorem.
Denition 3.39 (Well-order). A well-order on a set S is a total order such that every
non-empty subset T S has a least element. (In other words, for any subset T S, there
exists some x T such that x y for all y T. A least element of a set is a lower bound
contained within the set itself.) We say that the set S is well-ordered by .
The fact that the usual order on the natural numbers N is a well-order is fairly
obvious (given a set of positive integers, there is always a smallest one), and it turns out
to be the key property that allows induction on N. Less obvious is the fact that given any
set S, there exists some order that is a well-order on S .
Proposition 3.40 (Well-ordering theorem). Any set can be well-ordered.
This sometimes allows us to do induction on arbitrary sets (called transnite induc-
tion), although we must be careful because the order that well-orders a set may be very,
very different from the order that we usually use on that set. The well-ordering theorem
is another example of a non-constructive result: we know from the theorem that a well-
ordering exists on R, yet mathematicians have never been able to actually nd such an
ordering. Most of them have trouble even imagining one.
3.7 Cardinality
Suppose that we want to compare the size of two sets A and B. If A and B are both nite,
this is easy: we just look at the number of elements m in A and n in B, and see which
number is larger. But what if A and B are both innite? There isnt any clear rule to
decide when one innity is larger than another innity.
Mathematicians like to generalize the notion of comparing size using functions. If
there is a bijective function f between A and B, they say that A and B have the same
size. If, on the other hand, there is an injective function from A to B but not an injective
function from B to A, they say that A is smaller than B.
This notion of size is consistent with the standard denition for nite sets. For nite A
and B, there is a bijective function between A and B iff A and B have the same number of
elements, and there is an injective function from A to B iff A has weakly fewer elements
than B. For instance, there is a bijection between 1, 2, 3 and 2, 3, 4 ( f (n) = n +1), since
these sets both have 3 elements. There is an injective function from 1, 2, 3 to 1, 2, 3, 4,
26
but not an injective function from1, 2, 3, 4 to 1, 2, 3; in the latter case, a function f will
inevitably send at least two elements in 1, 2, 3, 4 to the same element in 1, 2, 3.
There are also unintuitive aspects. We might think that the set of positive even num-
bers 2, 4, 6, . . . is clearly smaller than the set of all positive integers 1, 2, 3, . . ., since
the former is a strict subset of the latter. But since there is an obvious bijection between
the sets ( f (n) = n/2), these sets are actually the same size according to our new denition.
Above all, if this is to be a rigorous notion of comparative size, we need to show that
it satises the properties of an order. First, we dene what it means for two sets to be the
same size, and show that it is an equivalence relation.
Denition 3.41. Two sets A and B are said to be equinumerous, to have the same cardi-
nality, if there is a bijection f : A B. We then write A B.
Proposition 3.42. is an equivalence relation.
Proof. To prove that is an equivalence relation, we must show that it is reexive, sym-
metric, and transitive. For reexivity, note that the identity function f : A A dened
by f (x) = x is a bijection. For symmetry, recall that for any bijective function f : A B
we can dene an inverse f : B A, which is also bijective. For transitivity, observe that
the composition f g : A C of any two bijections g : A B and f : B C is also a
bijection.
Now that we have show that is an equivalence relation, we dene the cardinality
[A[ of a set A to be its equivalence class under .
1213
We then dene an order that
compares cardinalities.
Denition 3.43 (Denition of order on cardinalities). Let A and B be sets. Then we
dene [A[ [B[ if there exists an injection from A to B.
First, we need to be sure that this denition is consistent. We dene [A[ [B[ if there
exists an injection from A to B, but A is only one member of the equivalence class [A[ and
B is only one member of the equivalence class [B[. What if we took some other member
C of the equivalence class [A[ (i.e. C A) and D of the equivalence class [B[ (i.e. D B)?
For the denition to be consistent, we should get the same answer. There should be an
injection from C to D iff there is an injection from A to B.
Fortunately, this is true. If C A and D B, then let g
1
: A C and g
2
: B D be
the bijection between A and C and the bijection between B and D, respectively. Now, if
f : A B is an injection, h : C D dened by h(x) = g
2
f g
1
1
(x) is also an injection.
Similarly, we can take an injection from C to D and obtain an injection from A to B.
12
Recall that any element x in a set with an equivalence relation is a member of an equivalence class,
which consists of all members of the set that are equivalent to it. The equivalence classes of a set form a
partition of that set. In this case, the elements x are themselves sets, which are members of the set of all
sets. (Dening the set of all sets is difcult, and we run into paradoxes unless we limit the sets that are
allowed to be members. For now we simply assume that this has been done properly. Look up the Russell
paradox for more details if you are interested.)
13
In more in-depth treatments, objects called cardinals are dened that serve as representatives of these
equivalence classes. This doesnt really matter for us, though; we just care about the fact that we can
compare the cardinality of any two sets.
27
Although we have shown that the denition of is consistent, we still have not shown
that it satises the properties of an order, either partial or total. As it turns out, is a total
order. To prove this, we need to verify the three properties of a total order.
Proposition 3.44. is a total order on cardinalities.
Proof. We need to verify transitivity, antisymmetry, and totality.
Transitivity. This is the easiest of the three. Suppose that [A[ [B[ and [B[ [C[.
Then there exist injections g : A B and f : B C. The composition f g : A
C is then an injection from A to C, so that [A[ [C[.
Antisymmetry. We need to show that if [A[ [B[ and [B[ [A[, then [A[ = [B[.
In other words, we need to show that if there is an injection from A to B and an
injection from B to A, then there is a bijection from A to B. This is not too hard to
prove, but it is enough that we will not do it here. This result is called the Cantor-
Bernstein-Schroeder theorem, and a proof is available in its Wikipedia article.
Totality. We need to show that any A and B are comparable; that for any A and
B, either [A[ [B[ or [B[ [A[. Like antisymmetry, this is highly nonobvious; in
fact, it is even more difcult to prove, and unlike transitivity and antisymmetry it
requires us to assume the axiom of choice. The usual method of proof is to show
that any two well-ordered sets are comparable (making use of the nice properties
of a well-ordering), and then to use the well-ordering theorem (Proposition 3.40) to
well-order arbitrary A and B.
What practical use is all this? Although it seems unnecessarily complicated, showing
that [S
1
[ ,= [S
2
[ is sometimes one of the easiest ways to show that S
1
,= S
2
. (This is a
very nice way to show that Q ,= R, as we will see.) Also note that behaves nicely with
respect to set inclusion:
Proposition 3.45. Suppose A B. Then [A[ [B[.
Proof. Let f : A B be the inclusion function dened by f (x) = x, which sends an
element x in A to the same element x in the larger set B. f is injective, and therefore by
denition [A[ [B[.
We can equivalently dene in terms of surjections rather than injections:
Proposition 3.46. [A[ [B[ iff there is a surjection from B to A.
Another very important application is to compare the cardinality of arbitrary sets S
with the cardinality of the natural numbers N. Often the validity of an argument depends
on how [S[ compares to [N[; for instance, a proof strategy may work if [S[ [N[, but not
otherwise. This comparison is so important that we use special terms to discuss it.
28
Denition 3.47 (Countable set). A set S is countable if it has cardinality weakly less than
that of the natural numbers: [S[ [N[. In other words, S is countable if there exists an
injective function from S to N.
Denition 3.48 (Countably innite set). A set S is countably innite if it has cardinality
equal to that of the natural numbers: [S[ = [N[. (Sometimes the term countable is used
specically with this meaning as well.) In other words, S is countably innite if there
exists a bijection between S and N.
Denition 3.49 (Uncountable, or uncountably innite, set). A set S is uncountable (also
called uncountably innite) if [S[ > [N[. In other words, S is uncountable if there exists
an injective function fromNto S, but no bijection between Nand S.
The following is a surprising and useful result about countable sets. We will use it in
the next section to show that the set of rational numbers is countable!
Proposition 3.50. The union of countably many countable sets is countable.
Proof. Suppose that we have countably innitely many countable sets. In particular, let
A
n
be a collection of countable sets, indexed by n N. Let us arrange the elements of
each set A
n
in a sequence x
nk
, where k Nas well, and consider the array
A
1
: x
11
x
12
x
13
x
14
. . .
A
2
: x
21
x
22
x
23
x
24
. . .
A
3
: x
31
x
32
x
33
x
34
. . .
A
4
: x
41
x
42
x
43
x
44
. . .
By going up the diagonals from bottom-left to top-right, we can arrange these elements
in a sequence (note that since we did not assume that the A
n
were disjoint, there might be
repeats):
x
11
, x
21
, x
12
, x
31
, x
22
, x
13
, x
41
, . . . (41)
This sequence is a surjective mapping fromNto
nN
A
n
. Therefore by Proposition 3.46,
[
nN
A
n
[ [N[. Yet for any n, [A
n
[ = N, and since A
n
is a subset of
nN
A
n
, Proposi-
tion 3.45 implies that [N[ [
nN
A
n
[. We conclude that
[
nN
A
n
[ = [N[ (42)
as desired.
The result for nitely many countable sets A
1
, . . . , A
m
(rather than countably innitely
many, as above) is implied by the above result, where we simply take A
n
to be the empty
set for all n m +1.
Above we found a way to show that for certain set unions A, [A[ = [N[. When can we
show that [A[ > [N[, or more generally [A[ > [B[ for some B? The following proposition
provides one situation where this is possible.
29
Proposition 3.51. For any set X, [X[ < [2
X
[: the cardinality of X is strictly less than the
cardinality of its power set.
Proof. First, note that there is an injection f : X 2
X
that takes each x to the singleton
set x in 2
X
. Therefore we have [X[ [2
X
[, and we now need to rule out [X[ = [2
X
[.
Suppose that [X[ = [2
X
[, i.e. that there is a bijection between X and 2
X
. Then, in
particular, there is a surjection g from X onto 2
X
. Let us dene the element Y of 2
X
(i.e. a
subset Y of X) as follows:
x Y x / g(x)
Since g is surjective, there must exist some y X such that g(y) = Y. Substituting y for x
above, and using g(y) = Y, we obtain:
y Y = y / Y
which is clearly impossible. Therefore our assumption was wrong, and there cannot be a
bijection between X and 2
X
. We conclude that [X[ < [2
X
[, as desired.
30
4 Fields
4.1 Fields, orders, and ordered elds
We now dene special kinds of sets, called elds, on which the basic operations of arith-
metic are dened. The formal denition of a eld is important in its own right, but it
will be especially important in the denition of vector spaces, which we will cover in the
linear algebra lecture. Although many different elds exist, there are three specic elds
that we will emphasize in these notes: the rational numbers Q, the real numbers R, and
the complex numbers Q.
Denition 4.1 (Field). A eld is a triple (F, +, ) consisting of a set F, an addition opera-
tion + that maps a pair of elements of F to another element of F:
F F F
(x, y) x + y
and a multiplication operation that maps a pair of elements of F to another element of
F:
F F F
(x, y) x y
such that addition satises the following 4 properties:
1. Addition is associative: (x + y) + z = x + (y + z) for all x, y, z F.
2. Addition is commutative: x + y = y + x for all x, y F.
3. Zero (additive identity): There is a unique element 0 F such that x + 0 = x for all
x F.
4. Additive inverse: For each element x F, there is a unique element x F such that
x + (x) = 0.
multiplication satises the following 4 properties:
1. Multiplication is associative: (x y) z = x (y z) for all x, y, z F.
2. Multiplication is commutative: x y = y x for all x, y F.
3. One (multiplicative identity): There is a unique element 1 F such that x 1 = x for
all x F.
4. Multiplicative inverse: For each element x F other than 0, there is a unique element
1/x F such that x 1/x = 1.
and addition and multiplication jointly obey the distributive law
x (y + z) = x y + x z
for all x, y, z F.
31
Like other sets, elds can be preordered, partially ordered, or totally ordered. The to-
tally ordered case (see Denition 3.19) is particularly interesting, because in special cases
we can dene an ordered eld.
Denition 4.2 (Ordered eld). An ordered eld is a 4-tuple (F, , +, ) such that (F, +, )
is a eld, (F, ) is a totally ordered set, and the following additional two properties hold:
1. For any x, y, z F, y z implies x + y x + z.
2. For any x, y F, x 0 and y 0 together imply x y 0.
Keep in mind that an ordered eld is both a eld and a totally ordered set, but the
converse is not true: it is possible for a eld to have a total order that does not satisfy
properties 1 and 2 in Denition 4.2. These properties link the structure of the arithmetic
operations in the eld to the order on that eld.
Examples.
The integers Z = . . . , 1, 0, 1, . . . are a totally ordered set under the usual order
but not a eld, because no elements except 1 and -1 have multiplicative inverses.
The set 0, 1 is a eld if 0 and 1 have their usual meanings and we complete the
denition of addition by writing 1 + 1 = 1. (Note: the outcome of all other opera-
tions is specied by the eld axioms. For instance, it can be proven that in any eld,
0 x = 0 for all x.) It is not an ordered eld, however, since property 1 in Denition
4.2 is violated regardless of how we dene the order.
Both the rationals Q and the reals R (which we will dene rigorously soon, though
you have probably already encountered them) are ordered elds under the usual
order. The complex numbers C are a eld, and can be made into a totally ordered
set by dening the order in various ways, but cannot be made into an ordered eld
no matter how the order is dened.
Although the reals R are an ordered eld under the usual order, we can dene the
order in other ways that do not satisfy the requirements for an ordered eld. For
instance, suppose we dene a new order
R
by reversing the usual order of R, so
that x
R
y y x. Then (R,
R
) is still an ordered set, but it is not an ordered
eld because it fails property 2 of Denition 4.2. (Remember: an ordered eld is
more than just a eld with some order!)
All the usual rules of arithmetic can be derived fromthe properties of a eld. For instance,
we have:
14
Proposition 4.3. Suppose that (F, +, ) is a eld. Then the properties of addition imply:
D(a): If x + y = x + z, then y = z.
14
All these examples are taken from Rudin
32
D(b): If x + y = x, then y = 0.
D(c): If x + y = 0, then y = x.
D(d): (x) = x.
The properties of multiplication similarly imply:
M(a): If x ,= 0 and xz = yz, then y = z.
M(b): If x ,= 0 and xy = x, then y = 1.
M(c): If x ,= 0 and xy = 1, then y = 1/x.
M(d): If x ,= 0, then 1/(1/x) = x.
We also have the following four statements:
F(a): 0 x = 0.
F(b): If x ,= 0 and y ,= 0, then xy ,= 0.
F(c): (x) y = (x y) = x (y).
F(d): (x) (y) = x y.
Proof. We will prove D(a), M(d), F(a), and F(b) as examples, leaving the rest as exercises.
(All these statements are very simple to prove, but it is good practice to work through
them formally.)
For D(a), we add x from the left on both sides of x + y = x + z to obtain x +
(x + y) = x + (x + z). We apply the associative property to make this (x + x) + y =
(x + x) + z, and then apply the denition of an additive inverse to obtain 0 + y = 0 + z.
Finally, we apply the denition of the additive identity 0 to obtain y = z.
For M(d), we observe that 1/x x = x 1/x = 1, where the rst step follows from the
commutative property for multiplication and the second step follows from the denition
of 1/x. Therefore x is an additive inverse of 1/x (and additive inverses are unique by the
properties of a eld); we write this as 1/(1/x) = x.
For F(a), we note that 0 x equals (0 +0) x by the denition of 0, and further that this
equals 0 x + 0 x by the commutativity of multiplication and the distributive property.
Now that we have 0 x +0 x = 0 x, we apply D(b) to obtain 0 x = 0.
For F(b), if x y = 0 and x ,= 0, then we multiply both sides by 1/x to obtain 1/x (x
y) = 1/x 0. Using associativity on the left and F(a) on the right, we have (1/x x) y = 0,
and from the denitions of additive inverse and identity this reduces to y = 0.
Similarly, all the usual rules for arithmetic and inequalities can be derived from the
properties of an ordered eld:
Proposition 4.4. Suppose that (F, , +, ) is an ordered eld. Then we have:
O(a): If x 0, then x 0, and vice versa.
33
O(b): If x 0 and y z, then x y x z.
O(c): If x 0 and y z, then x y x z.
O(d): x x 0. (We usually write this as x
2
0)
O(e): If 0 1/x 1/y, then 0 y x.
4.2 Rational numbers
Weve seen the abstract properties of ordered elds, but lets try to imagine an ordered
eld. What elements might it contain, and what might it look like? By denition, we must
have the elements 0 and 1, and using the addition operation we obtain 1 + 1, 1 + 1 + 1,
and so on.
For elds in general, its possible that a sum of n 1s will equal 0 (1 + 1 + + 1 = 0),
but for ordered elds this is impossible. Why? By point O(d) in Proposition 4.4 above, we
know that 1 = 1
2
0. Then property 1 of Denition 4.2 implies 1 + 1 1, 1 + 1 + 1
1 +1, and so on. If
1 +1 + +1
. .
n ones
= 0
for some n, then we nd:
1 +1 + +1
. .
n1 ones
0
= 1 +1 + +1
. .
n ones
1
=0 1
This is not consistent with antisymmetry: we cant have both 0 1 and 1 0 without
0 = 1. But as F(a) in Theorem 4.3 demonstrates, we cant have 0 = 1 either. We conclude
that its impossible to have 1 +1 + +1 = 0, for any number of 1s on the left.
This has a useful consequence: it implies that 1 +1 + +1
. .
n ones
and 1 +1 + +1
. .
m ones
are
not equal whenever m ,= n. Otherwise (if, say, m < n) we could subtract the sum of
m 1s from both sides, and conclude that 1 +1 + 1
. .
nm ones
= 0, which we just proved to be
impossible.
We therefore have found a hierarchy of distinct elements in any ordered eld: 1, 1 +1,
1 + 1 + 1, and so on. In the usual manner, we denote these values by 1, 2, 3, etc. By the
additive inverse property, the ordered eld must also contain 1, 2, 3, etc. as well.
Essentially, weve demonstrated that any ordered eld must contain the integers.
Thats not all. Multiplicative inverses must exist as well. We must have 1/2, 1/3, etc.,
as well as the products of these values with the integers: say, 3 1/2, or 5 1/7. And we
need to dene the sums and products of these values (say, the sum of 3 1/2 and 5 1/7)
in a way thats consistent with the properties of an ordered eld. And so on...
34
Although we wont prove it here, it turns out that there is really only one way to
dene the sums and products of these values consistently, and this gives us the system
of numbers known as the rationals. Every ordered eld must contain the rationals as
a subset, though beyond this there are many possibilities for additional elements and
structure.
Denition 4.5 (Rational numbers). The set of rational numbers Q is an ordered eld
consisting of equivalence classes of pairs of integers (p, q) where q > 0, which we write
as p/q. Two pairs p
1
/q
1
and p
2
/q
2
are in the same equivalence class (i.e. represent the
same rational number) if p
1
q
2
= p
2
q
1
.
15
We dene the order on Q by p
1
/q
1
p
2
/q
2
p
1
q
2
p
2
q
1
. The additive identity
is the equivalence class of pairs of the form 0/q, and the multiplicative identity is the
equivalence class of pairs of the form q/q. Addition, multiplication, additive inverses,
and multiplicative inverses are dened by:
p
1
q
1
+
p
2
q
2

p
1
q
2
+ p
2
q
1
q
1
q
2
p
1
q
1

p
2
q
2

p
1
p
2
q
1
q
2

p
1
q
1

p
1
q
1
1
_
p
1
q
1

_
q
1
p
1
p
1
> 0
q
1
p
1
p
1
< 0
It is straightforward but extremely tedious to verify that the ordered eld thus dened
satises all the properties laid out for elds, totally ordered sets, and ordered elds in Def-
initions 4.1, 3.19, and 4.2. Among other things, we must demonstrate that our denitions
are consistent with the equivalence relation weve used to dene Q. For instance, if p
/
1
/q
/
1
and p
/
2
/q
/
2
are equal to p
1
/q
1
and p
2
/q
2
, respectively, then our rule for addition had bet-
ter produce the same results for p
/
1
/q
/
1
+ p
/
2
/q
/
2
and p
1
/q
1
+ p
2
/q
2
. More concretely, we
should have 2/4 + 1/3 = 1/2 + 3/9. (And we do: applying the rule for addition, this
becomes 10/12 = 15/18.)
This all sounds like basic arithmetic, because it is. But there are several ideas here
that will recur in the future. The idea that we dene a set by partitioning a larger set
into equivalence classes, for instance, is extremely common. Perhaps we have a space of
15
Recall that in the previous lecture, we mentioned that an equivalence relation induces a partition on a
set; each subset in the partition consists of elements that are equivalent to each other. The rational num-
bers are simply the subsets in this partition. The rational number 1/2, for instance, is really the partition
1/2, 2/4, 3/6, 4/8, . . .. This is very important, because we dont want to dene the rationals in a way
that, say, 1/2 and 2/4 are treated as different; they are really just two ways of writing the same element in
Q.
35
functions, and we want to say that f = g as long as f (x) = g(x) for almost all x, even if
they disagree at a few points. This requires that we dene some notion of an equivalence
class, and show that our rules for the space of functions are compatible with it.
More generally, well often need to dene a set along with a set of operations on that
set, and verify that these operations satisfy some basic axioms. Dening the rationals may
seem like unnecessary formalism to an economist, but the concepts involved certainly are
not.
4.3 Properties of the rational numbers
The following property says, intuitively speaking, that Q has no very small or very
large elements: any positive element of Q, added to itself sufciently many times, can
be made greater than any other element. It is one of the important properties of Q (and,
as well soon see, R).
Proposition 4.6 (Archimedean property for rationals). If x Q, y Q, and x > 0, then
there exists some positive integer n such that n x > y.
Proof. Suppose x = p
1
/q
1
and y = p
2
/q
2
. Then one such n is 2q
1
p
2
.
A related property is that between any two rational numbers, we can nd another
rational number.
Proposition 4.7. If x Q, y Q, and x < y, then there exists a z Q such that x < z and
z < y.
Proof. Take z = 1/2 (x + y).
One surprising feature of the rationals is that they are countably innite. As stated in
Denition 3.48, this means that the rationals have the same cardinality as N, i.e. that there
is a bijective mapping f : N Q from the natural numbers Nto the rationals Q.
Proposition 4.8. Q is countably innite.
Proof. First, there is an injection from N into Q dened by f (x) = x/1. Therefore the
cardinality of Nis weakly less than the cardinality of Q: [N[ [Q[.
We now show that [Q[ [N[ as well, which in conjunction with [N[ [Q[ implies
that [Q[ = [N[. To do so, we use Proposition 3.50, which states that the union of countably
many countable sets is countable. For n N, we dene the set A
n
Q as the set containing
all rationals with denominator Q.
A
n
= m/n : m Z (43)
(Note that there will be some overlap between the various A
n
: for instance, 1/2 = 2/4 =
3/6 = . . ., so that 1/2 A
n
for each even n.)
Since any rational must be expressible as m/n with some denominator n, we can write
the rationals as the union of all the A
n
: Q =
nN
A
n
. Thus Q can be written as the union
of countably many countable sets. By Proposition 3.47 from the previous lecture, this implies
that Q itself is countable, i.e. [Q[ [N[.
36
The countability of the rationals is surprising in part because were accustomed to
counting the elements of a set in a natural order. For instance, if we were asked to show
that the set of positive squares 1, 4, 9, . . . was countable, wed simply write f (1) = 1,
f (2) = 4, f (3) = 9, and so on. But we cant do this with rationals: any bijective mapping
f : N Q cannot always have f (n) < f (n + 1), where < is the standard order on Q.
To see why, just consider Proposition 4.7: if f (n) < f (n + 1), there must be some other
rational z in between f (n) and f (n + 1). But if f is strictly increasing, it cant possibly
map to z; we know, therefore, that well miss at least one rational number by counting in
this manner. To show that Q is countable, we need to map Nto Q in a less natural way.
The fact that the rationals Q are countable is frequently useful, particularly because
(as we will soon show) the reals R are not countable. Sometimes an argument only works
for a countable set; a common strategy is to use this argument to prove a proposition over
Q, then nd some way to extend it to R.
Unfortunately, all is not well in the land of rational numbers. Although they possess
ddesirable features like the Archimedean property and countability, they fail in other
ways. For instance, we cannot always take square roots within the set of rational numbers.
Proposition 4.9. x
2
= 2 does not have a solution x Q.
Proof. Suppose to the contrary that there was some solution x = p/q. Take p and q to be
such that not both p and q are even. (If p and q are both even, we can divide them both
by 2 to get an equivalent representation of x.) Then we have:
_
p
q
_
2
= 2
p
2
= 2q
2
Therefore p
2
must be even, which implies that p is even as well. But then we can write
p = 2p
1
, which implies:
4p
1
1
= 2q
2
2p
2
1
= q
2
This implies that q
2
and therefore q must also be even. But this contradicts our assumption
that not both p and q were even. We conclude that the premise was false, and there cannot
be a solution x Q to x
2
= 2.
The strange part about this is that we can come arbitrarily close to obtaining a solution
x Q to x
2
= 2. For instance, we can take x = 1.4, x = 1.41, x = 1.414, x = 1.4142, and
so on. All these x are rational, and their squares seem to be approaching 2 from below.
Yet weve proven that we cant nd a x that actually solves the equation.
We recall the least upper bound property given in Denition 3.24. When this property
is true for a set X, for any nonempty subset A X that has an upper bound, there must
be a least upper bound sup A. As you can probably guess, the rationals do not satisfy this
property.
37
Proposition 4.10. Q does not satisfy the least upper bound property.
Proof. Suppose to the contrary that Q does satisfy the least upper bound property.
Consider the set X of all x Q such that x
2
2. This set has an upper bound (for
instance, 2), and by assumption it must have a least upper bound c. Since we cannot have
c
2
= 2, either c
2
< 2 or c
2
> 2.
Suppose c
2
< 2. Choose some positive r Q sufciently small that r 1/8 (2 c
2
).
(It follows from the Archimedean property that this is possible by picking r = 1/n for
some n.) Then
(c + r)
2
= c
2
+2rc + r
2
c
2
+
2 c
2
2
+
(2 c
2
)
2
64
< 2
c + r is strictly greater than c and satises (c + r)
2
< 2, which contradicts cs status as the
least upper bound of X.
We can show that the case c
2
> 2 leads to a contradiction by similar logic, and we
conclude that the set X cannot have a least upper bound.
16
Therefore Q does not satisfy
the least upper bound property.
In one sense, rational numbers are everywhere: you can nd rational numbers that
come closer and closer to solving an equation like x
2
= 2. But in another sense, there are
gaps: you cant actually solve x
2
= 2, and the set of rational numbers with x
2
2 has
no least upper bound. To close these gaps, we need a richer number system. This system
will be R, the reals.
4.4 Real numbers
After dening the concept of an ordered eld, we realized that any ordered eld must
contain a subset with certain properties. Focusing our attention on this subset as an or-
dered eld of its own, we called it Q, the rationals.
It turns out that when we add the requirement that an ordered eld must satisfy the
least upper bound property, only one ordered eld remains: the real numbers R. We can,
therefore, simply dene R to be an ordered eld with the least upper bound property.
Denition 4.11 (R, axiomatic denition). The eld of real numbers R is the unique or-
dered eld with the least upper bound property. (In the context of R, the latter property
is often called the completeness axiom.)
This denition is extremely useful: armed with only the properties of an ordered
eldincluding the fact that it must contain Qand the completeness axiom, we can
verify many facts about the reals. This is what we will do in the next section.
16
Observe that if we were in a eld that we knew did satisfy the least upper bound property, then we
would have just proven that c
2
= 2, by ruling out the alternatives c
2
< 2 and c
2
> 2! But
38
At the same time, this denition is also slightly unsatisfying. It is not easy to prove that
that real numbers are the unique eld satisfying these properties. Worse, it is not obvious
that it is even possible for an ordered eld with the completeness axiom to exist. To show
that it is possible, we would need to explicitly construct the real numbers, dening them
in terms of more fundamental constituents.
One standard method of doing this is to dene real numbers as Dedekind cuts. Es-
sentially, we dene each real number r R to be the subset of rationals x Q where
x < r. (Or, equivalently, the subset of rationals x Q where x > r.) The idea is that
each real number r is equivalent to a cut in Q, which splits Q into a subset of numbers
greater than r a subset of numbers less than r, possibly along with r itself.
This idea is motivated by the failure of x Q : x
2
< 2 to have a least upper
bound in Q; if we want some r to be the least upper bound, why not just dene r by the
property r > x x
2
< 2? This is extremely unintuitive at rst: its weird to say that
were dening a real number to be a certain subset of rational numbers. But it turns out
to be a clever and very useful way to construct the reals. We discuss this a little more in
Appendix A.
4.5 Properties of the real numbers
Like the rationals, the reals satisfy the Archimedean property.
Theorem 4.12. If x R, y R, and x > 0, then there exists some positive integer n such that
n x > y.
Proof. Consider the set X = n x : n Z.
17
. If the Archimedean property is false, then
y is an upper bound for X. Then by the completeness axiom, this set must have a least
upper bound, which we call z.
Now consider z x. Since z x < z = sup X, there must be n x X such that
z x < n x, or else z would not be the least upper bound. But then we have z <
(n + 1) x, which is a contradiction. We conclude that X must not have an upper bound
after all, and therefore that the Archimedean property holds.
Again like the rationals, between any two real numbers there is another real number.
(If we have x, y R and x < y, we just take (x + y)/2.) More surprisingly, however,
there is also a rational number between any two real numbers. This a critical fact about the
relationship between Q and R. It follows almost immediately from the Dedekind cut con-
struction of the reals in Appendix A, but here we will prove it without the construction,
which takes a little more work.
Proposition 4.13. If x R and y R, then there is some z Q such that x < z and z < y.
Proof. From Theorem 4.12, there exists some n such that n (y x) 1.
Applying Theorem 4.12 once more, there exist some positive integers m
1
and m
2
such
that m
1
> nx and m
2
> nx. This implies m
1
< nx < m
2
, which then implies that there
exists some m satisfying m
1
m m
2
such that m1 nx < m.
17
This is the set . . . , 2x, 1x, 0x, 1x, 2x, . . .
39
We now take z = m/n. We observe that z = m/n > nx/n > x, and also z = m/n <
(m +1)/n < x +1/n y, as desired.
One major difference between R and Q is that R is not countable. There are many
proofs of this fact, but arguably the simplest and most generalizable is the diagonal argu-
ment given below, which is closely related to the argument we used to prove Proposition
3.51. It involves the notion of decimal expansions, which we havent formally developed
in these notes (although certainly we are all familiar with it on an intuitive level!). Its so
elegant that Ill use it anyway.
Proposition 4.14. R is uncountable.
Proof. Certainly if the interval [0, 1) is uncountable, then R is uncountable; we will prove
the former.
Suppose to the contrary that [0, 1) is countable, and that there is a bijection f : N
[0, 1). We look at the decimal expansions
18
of f (1), f (2), and so on:
f (1) = 0.a
11
a
12
a
13
. . .
f (2) = 0.a
21
a
22
a
23
. . .
f (3) = 0.a
31
a
32
a
33
. . .
Now we will construct a new real number 0.b
1
b
2
b
3
. . . such that b
n
,= a
nn
for all n. In other
words, we are constructing a new number whose nth digit will disagree with the nth digit
in bold above. Clearly, this number cannot be in the list f (1), f (2), f (3), . . ., because it will
disagree with f (n) in the nth digit. But this is impossible, becasue we assumed that f was
a bijection, meaning that every real number in [0, 1) should equal f (n) for some n. We are
forced to conclude that such a bijection is impossible, and that [0, 1) (and thus R) is not
countable after all.
Alternatively, we can adapt Proposition 3.51 itself to show that R is uncountable,
though this is slightly less transparent. The idea is that each element S of 2
N
can be
represented as an innite sequence of 0s and 1s, with the nth number in the sequence
indicating whether n is a member of S. For instance, (1, 0, 1, 1, 0, . . .) represents S 2
N
such that 1 S, 2 / S, 3 S, 4 S, 5 / S, etc. We can interpret this sequence of 0s and
1s as base-3 notation for a real number in the interval [0, 1).
19
This is an injection from 2
N
into R, and therefore [2
N
[ R. Since we know from Proposition 3.51 that [N[ < [2
N
[, it
follows that [N[ < [R[.
Existence of roots. In Q, we couldnt nd a solution to x
2
= 2. As the following propo-
sition shows, the situation is much better in R.
Proposition 4.15. Let n N, y R, and y > 0. Then there exists a solution x R to x
n
= y.
18
Technically decimal expansions are not unique, since 0.999 . . . = 1. We get around this problem by
banning decimal expansions with innite strings of repeating 9s.
19
This gets around the problem in binary where, for instance, 0.0

1 = 0.1.
40
Proof. Consider the set of all x R such that x
n
y. This set is nonempty and has an
upper bound (for instance, it is bounded by 1 + y). Let us denote the upper bound by c.
We argue that c
n
= y. Suppose to the contrary that c
n
< y; we will show that for a
judiciously chosen r, we then have (c + r)
n
y as well, contradicting the denition of c.
First, for any r > 0, we obtain the following factorization and inequality:
(c + r)
n
c
n
= r((c + r)
n1
+ (c + r)
n2
c + + (c + r)c
n2
+ c
n1
)
rn(c + r)
n1
If r c, then this becomes (c + r)
n
c
n
rn2
n1
c
n1
, and setting r = min(c, (y
c
n
)/(n2
n1
c
n1
)), we obtain:
(c + r)
n
= c
n
+ ((c + r)
n
c
n
)
c
n
+ rn2
n1
c
n1
c
n
+ (y c
n
)
= y
This contradicts our denition of c as the least upper bound of the set x
n
y. We can rule
out the case c
n
> y using similar arguments.
Although this proposition exhibits a nice feature of R, it was a pain to prove; we had
to do a factorization and explicitly choose an r of the right size.
The intuition for the existence of a solution is much simpler. As a function of x,
f (x) = x
n
is continuous. Furthermore, f (0) = 0 and f (y + 1) > y. Surely somewhere in
between 0 and y + 1, there was a point x where f (x) = y. Otherwise, there would be a
discontinuity in the graph of x
n
, and that doesnt seem likely.
There is a powerful formalization of this intuition called the Intermediate Value The-
orem. First, however, we need to develop more basic analytic machinery, like the formal
concept of continuity itself; and before we do any of this, we will think carefully about
the notion of distance in order to dene the general concept of a metric space. Once these
ideas have been developed (in the lectures on analysis), propositions like Proposition 4.15
will be much easier to prove.
Although it improves over Q in many ways, R has some aws as well. For instance,
we cannot take nth roots of every number. x
2
= 1 has no real solution. Indeed, it has
no solution in any ordered eld, as we can see from property O(d) in Proposition 4.4. To
remedy this, well need to expand the reals to make an even larger eld, the complex
numbers C. But again well need to make some sacrices: since x
2
= 1 cannot have a
solution in any ordered eld, we will not be able to dene any order on C that behaves so
nicely with respect to arithmetic operations.
4.6 Complex numbers
The complex numbers arise from the real numbers via a straightforward approach to
solving the equation x
2
= 1. Since there does not exist any x R that solves this
equation, why not just add another element that is dened to be a solution? In particular,
41
why not create a eld containing R and also some new number i, called the imaginary
unit, which satises i
2
= 1?
20
It turns out that it is possible to dene such a eld, which
we call the eld of complex numbers.
Denition 4.16 (Complex numbers). The eld of complex numbers consists of ordered
pairs of reals (a, b), which we will typically write as a + b i, where i denotes the imag-
inary unit. (Sometimes when a = 0 we simply write b i, and when b = 0 we write
a.)
The sumof a
1
+b
1
i and a
2
+b
2
i is dened to be (a
1
+a
2
) + (b
1
+b
2
) i. The additive
identity is 0 +0 i, the additive inverse of a +b i is a + (b) i. The product of a
1
+b
1
i
and a
2
+ b
2
i is dened to be (a
1
a
2
b
1
b
2
) + (a
1
b
2
+ a
2
b
1
) i, the mutiplicative identity is
1 +0 i, and the mutiplicative inverse of a + b i is
1
a + b i
=
a b i
a
2
+ b
2
Again, it is tedious to verify that all the properties of a eld formally hold. It is easier
for this case than either the rationals or reals before it, however, because the denition
is simpler. For one, there is a bijection between complex numbers and ordered pairs of
real numbers. Unlike the rationals, we are not muddying the waters with an equivalence
relation; and unlike the construction of the reals in Appendix A, the basic object we are
studying is just a pair of objects we have already dened, not some weird subset obeying
particular conditions. And since C cannot possibly be an ordered eld, we dont have to
worry about verifying those axioms anymore.
Denition 4.17 (Notation for complex numbers). Let z = a + bi be an element of C. Then
we dene:
1. The real part Re z = a.
2. The imaginary part Im z = b.
3. The complex conjugate z = a bi.
4. The absolute value [z[ =

a
2
+ b
2
. (Also called modulus.)
Complex conjugation. The complex conjugate is especially interesting, since it is based
on a deep relationship between i and i. All arithmetic operations on C can be derived
from the fact that i
2
= 1; yet i satises the exact same equation, with (i)
2
= 1.
How, then, can we possibly distinguish between i and i? The answer is that we cant;
if we have some equation in C, and we replace every i in the equation with i, then
we obtain an equivalent equation. A few implications of this principle are listed in the
following proposition.
20
This sounds like a wild approach, but in fact the idea of extending a eld with new elements in order
to satisfy some equation is very fundamental in eld theory, and happens all the timeits called a eld
extension. We wont go into this, since abstract algebra (outside linear algebra) doesnt matter very much to
economists.
42
Proposition 4.18. The complex conjugate z has the following properties:
1. z + w = z + w
2. zw = zw
3. z is a root of the polynomial p(z) = z
k
+ a
1
z
k1
+ . . . + a
k1
z + a
k
iff z is a root of the
conjugate polynomial p(z) = z
k
+ a
1
z
k1
+ . . . + a
k1
z + a
k
.
Proof. We let z = a
1
+ b
1
i and w = a
2
+ b
2
i.
1. z + w = (a
1
+ a
2
) + (b
1
+ b
2
)i, so z + w = (a
1
+ a
2
) (b
1
+ b
2
)i = (a
1
b
1
i) + (a
2

b
2
i) = z + w.
2. zw = (a
1
a
2
b
1
b
2
) + (a
1
b
2
+ a
2
b
1
)i, so zw = (a
1
a
2
b
1
b
2
) (a
1
b
2
+ a
2
b
1
)i. We
compute zw = (a
1
b
1
i)(a
2
b
2
i) = (a
1
a
2
b
1
b
2
) (a
1
b
2
+ a
2
b
1
)i as well.
3. If z
k
+ a
1
z
k1
+ . . . + a
k1
z + a
k
= 0, then taking the conjugate of both sides and
repeatedly applying both properties 1 and 2, we have z
k
+ a
1
z
k1
+. . . + a
k1
z +
a
k
= 0.
Since the complex conjugate has such nice properties, it is fortunate that we can re-
trieve the real and imaginary parts of any complex number, as well as its modulus, solely
with formulas involving the conjugate.
Proposition 4.19. The real part, imaginary part, and modulus of any complex number z satisfy
Re z =
z + z
2
Im z =
z z
2
[z[ =

zz
We also have some useful formulas involving the modulus.
Proposition 4.20. The modulus [z[ has the following properties:
1. [z[ 0, with [z[ = 0 iff z = 0.
2. [Re z[ [z[ and [Im z[ [z[.
3. [z[ = [z[
4. [zw[ = [z[[w[
5. [z + w[ [z[ +[w[
Proof. Writing z = a + bi:
1. [z[ = a
2
+ b
2
0, and a
2
+ b
2
= 0 iff a = b = 0, which is equivalent to z = 0.
2. [Re z[ = [a[

a
2
+ b
2
, and [Im z[ = [b[

a
2
+ b
2
.
43
3. [z[ =

z z =

zz = [z[
4. [zw[ =

zw zw =

zz

ww = [z[[w[
5. [z + w[ =
_
(z + w)(z + w) =

zz + wz + zw + ww =
_
[z[
2
+2 Re zw +[w[
2

_
[z[
2
+2[z[[w[ +[w[
2
= [z[ +[w[
Geometric interpretation of complex numbers: the complex plane We sometimes plot
real numbers on a one-dimensional axis, which depicts the concepts of order and distance
between reals. Since complex numbers are dened as pairs of real numbers, it is natural
to plot them on a two-dimensional plane, called the complex plane.
Generally, if z = a + bi, we plot the real part a horizontally and the imaginary part b
vertically. The horizontal axis, which consists of real numbers a, is called the real axis;
and the vertical axis, which consists of purely imaginary numbers of the form bi (real
multiples of i) is called the imaginary axis. Then the modulus [z[ is the length of the line
segment connecting a and b, and the conjugate z is the reection of z across the horizontal
(real) axis. This is depicted in Figure 4.1.
Something very interesting happens when we depict complex numbers in this manner.
It is a fact from elementary geometry that if we rotate the ray from the origin to (a, b)
by 90 degrees counterclockwise, we get a ray from the origin to (b, a). Since i (a +
bi) = b + ai, therefore, multiplication by i is equivalent to a 90 degree counterclockwise
rotation. This gives us a clear geometrical interpretation of the identity i
2
= 1: rotating
twice by 90 degrees is equivalent to rotating once by 180 degrees, which is equivalent to
reecting across the origin, or multiplying by -1. See Figure 4.2.
In other words, we managed to solve the equation z
2
= 1 by adding numbers that
correspond to rotations. This is in contrast to real numbers, which correspond to dilations:
if we multiply a +bi by some real number c, we get ca +cbi, which is equivalent to dilating
the ray from the origin to (a, b) by a factor of c, obtaining the ray to (ca, cb).
21
This geometrical view is so useful that we will extend it to all complex numbers z.
First, it is a fact that as we vary from 0 to 2 (in radians), (x(), y()) = (cos , sin )
traces out the unit circle, the locus of points (x, y) such that x
2
+ y
2
= 1.
22
Therefore,
for any z = a + bi satisfying [z[ = a
2
+ b
2
= 1, there is a unique [0, 2) such that
z = cos +i sin . More generally, for any complex z, it follows from Proposition 4.20 that
z/[z[ has modulus 1, and then there is a unique [0, 2) such that z/[z[ = cos +i sin .
This is so important that we introduce a name for it: the argument of z.
Denition 4.21 (Argument, trigonometric form). Let z be any complex number, and let
be the unique value in [0, 2) such that z/[z[ = cos + i sin . We say that is the
argument of z, and denote this relationship by = arg z.
21
If c < 1, we might call this squeezing, while if c > 1 we might call this stretching.
22
We must depart a little from our careful, formal approach here, since to prove facts about sin and cos we
need to dene them rst; and unfortunately, there is not a good way to do this rigorously without analysis,
which we will cover later. I assume that you are already familiar with the geometric meaning of sin and
cos.
44
z
z
Re z
Im z
[z[
Figure 4.1: The complex plane
z
iz
i
2
z = z
Figure 4.2: Multiplication by i
45
This allows us to write any z in trigonometric form z = r(cos +i sin ), where r = [z[
is the modulus and = arg z is the argument of z.
The argument has a remarkable property: the argument of a product is equal to the sum
of the arguments of each factor. This supplements the result we have already obtained in
Proposition 4.20, that the modulus of a product is equal to the products of each factor.
Proposition 4.22. Let z
1
= r
1
(cos
1
+ i sin
1
) and z
2
= r
2
(cos
2
+ i sin
2
) be complex
numbers expressed in trigonometric form. Then
z
1
z
2
= r
1
r
2
[cos(
1
+
2
) + i sin(
1
+
2
)]
and, up to a multiple of 2,
arg z
1
z
2
= arg z
1
+arg z
2
Proof. Directly multiplying z
1
and z
2
we have
z
1
z
2
= r
1
r
2
((cos
1
cos
2
sin
1
sin
2
) + i(sin
1
cos
2
+cos
1
sin
2
))
From trigonometry we know that cos(
1
+
2
) = cos
1
cos
2
sin
1
sin
2
and sin(
1
+

2
) = sin
1
cos
2
+cos
1
sin
2
. Therefore this becomes
z
1
z
2
= r
1
r
2
[cos(
1
+
2
) + i sin(
1
+
2
)]
as desired. By denition, then, arg z
1
z
2
=
1
+
2
up to a multiple of 2; or equivalently
arg z
1
z
2
= arg z
1
arg z
2
.
In the complex plane, the argument of z is the angle the positive real axis and the ray
from the origin to z. What Proposition 4.22 means geometrically, therefore, is that the
angle between the positive real axis and z
1
z
2
is the sum of the angles between the positive
real axis and z
1
and z
2
. Together with the fact that [z
1
z
2
[ = [z
1
[[z
2
[, from Proposition 4.20,
this is depicted in Figure 4.3.
Eulers formula. Suppose that we want to extend the usual exponential function e
x
on
the reals so that it takes complex values as well. How would we dene this new function
e
z
? One way is to observe that w(x) = e
x
is a solution to the differential equation w
/
= w:
the rate of change of the function equals the value of the function.
23
Suppose that the extension to the complex plane, w(z) = e
z
, obeys the same relation-
ship. Let us rst consider the case where z is purely imaginary, so that z = iy for real y.
Then dening v(y) w(iy) = e
iy
, applying the chain rule
24
we have v
/
(y) = ie
iy
= iv.
The derivative of v equals v times the imaginary unit i. Since multiplication by i is equiv-
alent to counterclockwise rotation by 90 degrees in the complex plane, this means that v
23
I apologize for bringing in calculus and differential equations before we formally cover them in math
campthis is the only way I know to obtain a clean intuition for complex exponentiation.
24
Again, since we have not yet covered analysis, here we are sloppy and assume that some version of the
chain rule holds for functions with real arguments and complex output. This turns out to be reasonable,
since we can identify the space of complex outputs with the space of two-dimensional real outputs, and
then our results from multivariable calculus will go through.
46
z
1
z
2
z
1
z
2
[z
1
[
[z
2
[
[z
1
z
2
[ = [z
1
[[z
2
[
arg z
1
arg z
2
arg z
1
z
2
= arg z
1
+arg z
2
Figure 4.3: arg z
1
z
2
= arg z
1
+arg z
2
47
moves in a direction perpendicular to its displacement fromthe origin. Visually, this corresponds
to rotation around the originsatellites travelling in circular orbits about a planet, for in-
stance, have velocity perpendicular to their position relative to the planet. Indeed, it turns
out that v(y) traces out exactly the unit circle in the complex plane.
Proposition 4.23. Let g(y) = cos y + i sin y be the function tracing out the unit circle in the
complex plane. Then v(y) = g(y); in other words, v(y) is the position on the unit circle that is at
an angle of y radians counterclockwise from the positive real axis. This result, summarized as
e
iy
= cos y + i sin y
is known as Eulers formula.
Proof. Since
v(0) = e
i0
= e
0
= 1 = cos 0 + i sin0 = g(0)
it sufces using the uniqueness of solutions to rst order ordinary differential equations
25
to show that v and g satisfy the same differential equation, namely v
/
= iv and g
/
= ig.
We have already demonstrated this for v, and can easily do the same for g:
g
/
(y) = sin y + i cos y = i(cos y + i sin y) = ig(y)
We conclude that v = g.
This proposition denes the exponential for purely imaginary inputs; we can com-
bine this with the conventional denition for real inputs and the usual properties of the
exponential function to obtain a denition of e
z
for any complex z = x + iy:
e
z
= e
x+iy
= e
x
e
iy
(This is the same as the result obtained from solving the differential equation w
/
(z) =
w(z) for all complex z, with the initial condition w(0) = 1.)
An alternative derivation of the exponential e
z
, which arrives at the same result, ex-
tends the characterization of the exponential as a limit:
e
z
= lim
n
_
1 +
z
n
_
n
The plots in Figure 4.4, inspired by an animation on the Wikipedia page for Eulers for-
mula, shows the progression of this limit in a special case z = i.
Another common use of the exponential is to rewrite the trigonometric form given in
Denition 4.21 in essentially identical polar form instead.
26
Denition 4.24. Let z be any complex number, and let z = r(cos +i sin ) be its trigono-
metric form. Then we dene the polar form of z to be z = re
i
, using the identity in
Proposition 4.23 to replace cos + i sin in the trigonometric form with e
i
.
In polar form, the additivity result from Proposition 4.23 becomes even simpler, as
it is merely another instance of the standard properties of the exponential function. If
z
1
= r
1
e
i
1
and z
2
= r
2
e
i
2
, then z
1
z
2
= r
1
r
2
e
i
1
e
i
2
= r
1
r
2
e
i(
1
+
2
)
.
25
Again, apologies for the bad ordering of the course here; we will formally cover ODEs, including
uniqueness of solutions, later.
26
Terminology here is used somewhat inconsistently.
48
Figure 4.4:
_
1 +
i
n
_
n
for n = 1, 2, 4, 8, 16 and = 7/18
Computations with complex numbers. For many computations involving complex num-
bers, it is far simpler to use trigonometric or polar form. This is true in particular for roots
and logarithms.
Denition 4.25. For any positive integer n, we dene an nth root of unity to be some
complex number satisfying
n
= 1.
Proposition 4.26. For any positive integer n, the nth roots of unity are the n numbers taking the
form = e
2im/n
for m = 0, . . . , n 1.
Proof. Writing = re
i
, is an nth root of unity if
n
= 1 r
n
e
i(n)
= 1 e
i0
. Clearly,
for this to be true we must have r = 1. We must also have n = 0 up to a multiple of 2,
or equivalently n = 2m = 2m/n for some integer m. If m
1
m
2
is a multiple
of n, then the corresponding are equal up to a multiple of 2; therefore, there are only
n values of m that lead to distinct angles , corresponding to the n possible values of m
modulo n. We can pick any n of these, and for convenience pick m = 0, . . . , n 1.
Proposition 4.27. Let z = re
i
be any complex number, and n be any positive integer. Then there
are n complex nth roots w of z, which take the form w =
n

z, where can be any of the nth


roots of unity, and
n

z r
1/n
e
i/n
.
Proof. Clearly
n

z is an nth root of z, since (


n

z)
n
= (r
1/n
e
i/n
)
n
= re
i
. Furthermore, the
ratio of any two nth roots of z must be an nth root of unity, since if both (z
1
)
n
= z and
(z
2
)
n
= z , then (z
1
/z
2
)
n
= (z
1
)
n
/(z
2
)
n
= 1. Therefore all nth roots of z must equal some
nth root of unity times
n

z, as desired.
Proposition 4.28. For any complex number z = re
i
, there are innitely many values of the
natural logarithm log z, taking the form log z = log r + i( + 2k), where k ranges over all
integers.
Proof. Let w = x + iy. Then w = log z e
x
e
iy
= e
w
= z = re
i
, which is equivalent
to e
x
= r and e
iy
= e
i
. This is true iff x = log r, under the usual denition of natural
logarithm for real numbers, and y = +2k for some k.
Since it is hard to deal with multiple-valued roots or multiple-valued logarithms, it is
sometimes convenient to pick a single value for each root or logarithm. The choice for
49
logarithms is, in fact, more fundamental, since if we have a single value for the logarithm
there is a corresponding choice for any root as well:
w = log z =
w
n
= log(z
1/n
) = e
w/n
= z
1/n
Unfortunately, it is impossible to pick a single value of log z for each z such that the
function log z is continuous on the complex plane. This is because there is no way to
pick the angle of each complex number z = e
i
such that is a continuous function of
z everywhere in the complex plane. If we use angles in the range [0, 2), for instance,
then there will be a discontinuity at the positive real axis, as the angle changes from just
below 2 to 0.
We must make an arbitrary choice; roughly speaking, once that choice is made, we
throw away the points at which the angle would be discontinuous. Such a choice of a
single value for log z, which denes log z everywhere except the origin and a ray going
outward from the origin, is called a branch of log z. The most popular branch, called the
principal branch, chooses angles in the range (, ), and throws away the negative
real axis.
27
Fundamental theorem of algebra. Finally we have the result that makes the eld of
complex numbers so special.
Proposition 4.29 (Fundamental theorem of algebra). Let p(z) be a nonconstant polynomial
with complex coefcients, in the single complex variable z. Then p has at least one complex root:
there exists some a C such that p(a) = 0.
There are many proofs of the fundamental theorem of algebra, each of which provides
some algebraic or analytic insight. Unfortunately, such proofs are slightly beyond the
scope of this course. Intuitively, however, the key contribution of C is that it offers rota-
tions in addition to dilations; multiplication by a number e
i
on the unit circle corresponds
to a rotation, while multiplication by a real number r corresponds to a dilation. There are
many polynomials, like z
2
+ 1, that do not have solutions in R because multiplication in
R consists only of dilation, and the solution must involve rotation as well.
It will be important to keep this intuition in mind when we study linear algebra, where
there are many direct analogies between certain types of linear maps on a vector space
and corresponding types of complex numbers. (For instance, rotation in a vector space is
closely related to complex numbers e
i
on the unit circle, which represent rotation in the
complex plane.)
Using polynomial division, we can go one step further than the fundamental theorem
of algebra, and write a degree n polynomial as the product of n linear factors.
Proposition 4.30. Let p(z) be a polynomial of degree n with complex coefcients, in the single
complex variable z. Then we can write p(z) as the product
p(z) = c (z a
1
) . . . (z a
n
)
27
This is all a little sloppy; if you have questions, please ask me or consult your friendly complex analysis
textbook.
50
where c C is a constant, and a
1
, . . . , a
n
C are (not necessarily distinct) roots of C. This
factorization is unique up to the ordering of a
1
, . . . , a
n
.
Proof. We prove this using induction on polynomial degree n. In the base case of degree
0, a polynomial is simply a constant c, and the result is immediate.
For arbitrary degree n > 0, from the fundamental theorem of algebra we know that
p(z) has a root a
n
. Using polynomial division, we can divide p(z) by (z a
n
) to obtain a
polynomial q(z) of degree n 1. Invoking the induction hypothesis, we know that q(z)
can be written in the form q(z) = c (z a
1
) . . . (z a
n1
), where a
1
, . . . , a
n1
are roots
of q(z). Therefore p(z) = q(z) (z a
n
) = c (z a
1
) . . . (z a
n
), with a
1
, . . . , a
n
roots
of p(z), as desired.
To prove that the factorization is unique up to ordering of the roots, we use a similar
induction on degree. Clearly the factorization is unique in the base case of constant poly-
nomials. Now suppose that we have two factorizations p(z) = c
1
(z a
1
) . . . (z a
n
)
and p(z) = c
2
(z b
1
) . . . (z b
n
) of the degree-n polynomial p. a
n
is a root of p, and
therefore we must have b
k
= a
n
for some k. Dividing p(z) by (z a
n
), we have a degree
n 1 polynomial q(z) that, by the induction hypothesis, has a unique factorization. But
then our two factorizations of p must be the same; if they were distinct, then factors other
than (z a
n
) = (z b
k
) would have to be distinct, and this would imply two distinct
factorizations of q(z), which is impossible.
Corollary 4.31. Let p(x) be a polynomial of degree n with real coefcients, in the single real
variable x. Then we can write p(x) as the product
p(x) = c (x
2
+ b
1
) . . . . (x
2
+ b
k
) (x d
1
) . . . (x d
m
)
where d
1
, . . . , d
m
are the real roots of p(x), and 2k + m = n.
Proof. Write p as a polynomial in a complex variable z, or p(z), and apply Proposition
4.30 to obtain a factorization c (z a
1
) . . . (z a
n
). Since the coefcients of p are real,
p(z) = p(z), and we can write:
p(z) = p(z) = c(z a
1
) . . . (z a
n
)
p(z) = c(z a
1
) . . . (z a
n
)
By the unique factorization result in Proposition 4.30, p(z) must have a unique factoriza-
tion, and therefore c = c, and a
1
, . . . , a
n
is a rearrangement of a
1
, . . . , a
n
. This means that
we can group the nonreal a
i
(where Im a
i
,= 0) into conjugate pairs.
Let d
1
, . . . , d
m
be the real members of the list a
1
, . . . , a
n
; and let e
1
, e
1
, . . . , e
k
, e
k
be the
nonreal members. Then
p(z) = c(z a
1
) . . . (z a
n
)
= c(z e
1
)(z e
1
) . . . (z e
k
)(z e
k
) (z d
1
) . . . (z d
m
)
= c(z
2
+[e
1
[
2
) . . . (z
2
+[e
k
[
2
) (z d
1
) . . . (z d
m
)
Now, writing b
i
[e
i
[
2
, and assuming that z is real and writing x z = z, this reduces to
p(x) = c (x
2
+ b
1
) . . . . (x
2
+ b
k
) (x d
1
) . . . (x d
m
)
as desired.
51
Both the complex factorization in Proposition 4.30 and the real factorization in Corol-
lary 4.31 will have important consequences when we cover canonical forms in linear alge-
bra. In general, when writing a real matrix in a canonical form, it will be possible either to
write it in a simpler complex canonical form, or a slightly more complicated real canoni-
cal form. As we will see, these are the direct analogues of the factorizations in Proposition
4.30 and Corollary 4.31.
52
A Dedekind cut construction of the reals.
Denition A.1 (R, denition by Dedekind cuts). Each member of R is dened to be a
certain subset of Q, called a cut. We dene a cut to be a subset Q with the following
properties:
1. is not empty, and ,= Q.
2. If p , q Q, and q < p, then q .
3. If p , then p < r for some r .
We dene the order on cuts by set inclusion: if . We dene the sumof two cuts
+ to be the cut consisting of all rationals of the form r + s, where r and s . We
dene the additive identity 0 to be the set of all negative rationals. We dene the additive
inverse to be the cut consisting of all r such that there exists s > 0 where r s / .
We dene the product of two positive cuts , where > 0 and > 0, to be the
cut consisting of all rationals of the form r s, where r and s . When either or
is 0, we dene the product to also be 0. In other cases, we dene multiplication by the
following rule:
=
_

_
(() ) < 0, > 0
( ()) > 0, < 0
() () < 0, < 0
We dene the multiplicative identity 1 to be the set of all rationals less than 1. For > 0,
we dene the multiplicative inverse 1/ to be the cut consisting of all r such that there
exists s > 0 where (1/r) (1/s) / . For < 0, we dene 1/ to be 1/().
It is very lengthy and tedious to verify that this construction actually produces an
ordered eld with the least upper bound property. First, we have to verify that set inclu-
sion is a legitimate order on cuts (in particular that it is a total order), and that it gives
us the least upper bound property. We then have to verify that all the properties of a
eld hold: that the denition of addition, the additive identity, and the additive inverse
are consistent with the properties of addition; that the denition of multiplication, the
multiplicative identity, and the multiplicative inverse are consistent with the properties
of multiplication; and that these denitions jointly give us the distributive property. We
also must show that we have an ordered eld. If you have a spare day at your disposal
and no sense of boredom, you can try this yourself. Otherwise, we will content ourselves
by proving that the order satises the least upper bound property; for the rest, if you are
still curious, you can check an analysis textbook.
We can see here, however, the notion of building up a new system (the reals) from a
simpler one (the rationals). This is a recurrent theme in mathematics. In fact, we did it for
the rationals too, which we dened as pairs of integers satisfying certain properties.
53

Das könnte Ihnen auch gefallen