The Cantor Set

A winter course at the Nesin Matematik Köyü

David Pierce

January –, 

Edited January , 

Matematik Bölümü
Mimar Sinan Güzel Sanatlar Üniversitesi
İstanbul
mat.msgsu.edu.tr/~dpierce/
dpierce@msgsu.edu.tr
 Abstract
We show how the Cantor set, with the order topology induced
from R, is homeomorphic to the power set of the natural num-
bers, as equipped with the Tychonoff topology. We obtain an
analogy between the Heine–Borel Theorem and the Tychonoff
Theorem—and with the Compactness Theorem of model the-
ory.
Preface
These notes are based on a course of seven lectures, 2 × 50
minutes each. I have thoroughly revised the notes from a sim-
ilar course in January . Those notes were divided by day;
the present notes are divided by topic and may differ more
from what actually happened in class. The Summary on page
 lists the numbered theorems of these notes (other results are
called lemmas and have their own numbering); what I did day
by day is roughly as follows.
Monday Theorem .
Tuesday Theorem , Cantor’s Theorem, Russell Paradox.
Wednesday Schröder–Bernstein, Cantor Intersection.
Thursday Theorems , , and .
Friday Theorem , Heine–Borel.
Saturday König’s Lemma, Tychonoff Theorem.
Sunday Compactness Theorem, Stone Representation.
The section on topological rings is an afterthought that con-
nects the course to the following week’s course, on finite fields.
Two students attended at least parts of both courses. For the
Cantor set course, I had thirty or forty students the first day;
down to about fifteen on Thursday, six on Friday, and four on
Sunday.


Summary
The Cantor Set is the intersection C of an infinite family of
sets, as in (.). The family forms a decreasing chain, as in
(.). Each set in the family is a union, as in (.), of finitely
many sets, which closed intervals of R, as in (.).
. C is the set of points in [0, 1] that can be written in base
3 without the digit 1 (page ).
. C is equipollent with the set P(ω) of subsets of the set
ω of natural numbers (page ).
. P(ω) is uncountable, By Cantor’s Theorem (page ).
. The proof is as for the Russell Paradox (page ).
. P(ω) and R are equipollent, by the Cantor–Schröder–
Bernstein Theorem (page ).
. As the intersection of a decreasing chain of nonempty
bounded closed sets, C is nonempty, by the Cantor In-
tersection Theorem (page ).
. R is a topological space with open intervals as a basis
(page ).
. C is closed in this topology (page ).
. R has only itself and the empty set as clopen subsets
(page ).
. C has many clopen subsets, namely its intersection with
the intervals mentioned above (page ).
. C is compact by the Heine–Borel Theorem (page ), so
every clopen subset is a finite union of the clopen subsets
just mentioned.


 . Our proof of the Heine–Borel Theorem is analous to that
of König’s Lemma (page ).
. —Also to that of the Tychonoff Theorem (page ).
. —And of the Compactness Theorem for propositional
logic (page ).
. This Compactness Theorem is a special case of the Stone
Representation Theorem (page ), though the Com-
pactness Theorem of first-order logic is not.
With the topology induced from R, the Cantor set is a topolog-
ical ring in two different ways, as (isomorphic to) the Boolean
ring P(ω) and the ring Z(2) of dyadic integers.


Contents

 Cardinality 
. The Cantor Set . . . . . . . . . . . . . . . . . . 
. Natural numbers . . . . . . . . . . . . . . . . . 
. Comparison of sets . . . . . . . . . . . . . . . . 

 Topology 
. Open and closed sets . . . . . . . . . . . . . . . 
. Clopen sets . . . . . . . . . . . . . . . . . . . . 
. Compactness . . . . . . . . . . . . . . . . . . . 
. Continuity . . . . . . . . . . . . . . . . . . . . . 
. König’s Lemma . . . . . . . . . . . . . . . . . . 
. Tychonoff topology . . . . . . . . . . . . . . . . 

 Algebra 
. Logic . . . . . . . . . . . . . . . . . . . . . . . . 
. Boolean algebras . . . . . . . . . . . . . . . . . 
. Stone spaces . . . . . . . . . . . . . . . . . . . . 
. Topological rings . . . . . . . . . . . . . . . . . 


List of Figures
. Some sets Cn . . . . . . . . . . . . . . . . . . . 
. The Cantor set . . . . . . . . . . . . . . . . . . 
x − 1/2
. Graph of y = . . . . . . . . . . . . . 
x · (1 − x)
. Cantor–Schröder–Bernstein Theorem . . . . . . 

. Ring operations on 2 . . . . . . . . . . . . . . . 


 Cardinality
. The Cantor Set
The Cantor set is a certain subset of R. For the sake of defining
it, we let

C0 = [0, 1],
C1 = [0, 1/3] ∪ [2/3, 1],
C2 = [0, 1/9] ∪ [2/9, 1/3] ∪ [2/3, 7/9] ∪ [8/9, 1],

and so on, as in Figure .. Thus each Cn is a union of bounded

closed intervals, and we obtain Cn+1 from Cn by removing the
middle third of each component interval. Presently we shall
find a “closed” expression for each set Cn , ultimately as in (.).
For now, we make three observations.

1/9 2/9 1/3 2/3 7/9 8/9

0 1
C0

C1

C2

C3

Figure .: Some sets Cn


 . The sets Cn compose a decreasing chain:

C0 ⊇ C1 ⊇ C2 ⊇ · · · (.)

. Each Cn is the union of 2n -many disjoint bounded closed

intervals.
. Each of those intervals has length 3−n .
Thus the measure of Cn is (2/3)n ; we may write
 n
2
µ(Cn ) = ,
3

where µ is the minuscule Greek letter mu. By definition, the

Cantor Set is the subset C of R given by

C = C0 ∩ C1 ∩ C2 ∩ · · ·

of R. We may write the intersection here as

∞
\
Cn .
n=0

If this has a measure, it ought to be limn→∞ (2/3)n , which is

0. Nonetheless, C is not empty. This will turn out to be a
special case of the Cantor Intersection Theorem, page . For
example, C contains the endpoints of the component intervals
of each Cn . To specify these points, we observe
 
1
C0 = 0, 0 ,
3
    (.)
1 2 2 1
C1 = 0, 1 ∪ 1 , 1 + 1 ,
3 3 3 3

 Cardinality 
and
   
1 2 2 1
C2 = 0, 2 ∪ 2 , 2 + 2
3 3 3 3
   
2 2 1 2 2 2 2 1
∪ 1 , 1 + 2 ∪ 1 + 2 , 1 + 2 + 2 . (.)
3 3 3 3 3 3 3 3

In general, each Cn is the union of all of the intervals of the

form  
d1 dn d1 dn 1
+ ··· + n, 1 + ··· + n + n ,
31 3 3 3 3
where each dk is either 0 or 2. In Sigma-notation, the intervals
are " n #
X dk X n
dk 1
k
, k
+ n .
k=1
3 k=1
3 3

. Natural numbers

In seeking to understand the Cantor set C, we shall make use,
at least notationally, of the natural numbers according to the
definition of von Neumann. We shall denote the set of natural
numbers by ω, which is the minuscule Greek letter omega.
Informally,
ω = {0, 1, 2, . . . }.

We make a recursive definition as follows.

. 0 is ∅, the empty set, and this is in ω.
. If n is in ω, then the successor of n, namely n + 1, is
n ∪ {n}, and this is in ω.
. Nothing else is in ω.

 Cantor Set 

It is a set-theoretical challenge to establish that ω is a well-
defined set; we accept that it is.∗ We have

0 = ∅, 1 = {0}, 2 = {0, 1},

and for all n in ω,

n = {0, . . . , n − 1} = {x ∈ ω : x < n}. (.)

Here the ordering < of ω is ⊂, and one can show that this
is also ∈. Easily ω is well-ordered, that is, each nonempty
subset has a least element; for the T
least element of a nonempty
subset A of ω is the intersection A.
We shall denote the set of functions from a set B to a set A
by
B
A.
Putting everything together, we have
n

2 = functions from {0, . . . , n − 1} to {0, 1} .

A typical element of n 2 can be written as either of

(e0 , . . . , en−1 ), e
∗
Each element of ω includes, as a subset, each of its elements; it
is also well-ordered, in the sense to be defined presently; therefore, by
definition, each element of ω is an ordinal. There may be other ordinals,
such as ω itself. Still, ω would be a limit ordinal, in the sense of being
neither 0 nor a successor. By definition, ω is the class, in the sense of
page , of ordinals that neither are limits nor contain limits. Though
it is not usually expressed this way, the Axiom of Infinity is that the
class ω is a set. In this way, the Axiom is parallel to most of the other
Zermelo–Fraenkel Axioms, namely those whereby certain classes are sets.

 Cardinality 
—in handwriting, ~e —, where each ek is in 2. We can now
say precisely \
C= Cn , (.)
n∈ω

where " #
[ X 2ek X 2ek 1
Cn = k+1
, k+1
+ n . (.)
e∈n 2 k∈n
3 k∈n
3 3
In ternary (base-3) notation,
X 2ek
k+1
= 0.d1 · · · dn ,
k∈n
3

where dk+1 = 2ek . In this notation then, from (.) and (.)
we have
C0 = [0, 1],
C1 = [0, 0.1] ∪ [0.2, 1],
C2 = [0, 0.01] ∪ [0.02, 0.1] ∪ [0.2, 0.21] ∪ [0.22, 1],
and likewise

C3 = [0, 0.001] ∪ [0.002, 0.01] ∪ [0.02, 0.021] ∪ [0.022, 0.1]

∪ [0.2, 0.201] ∪ [0.202, 0.21] ∪ [0.22, 0.221] ∪ [0.222, 1].
Again, C contains all of the endpoints here. Some of these are
shown in Figure .. Since
X 2
n+1
= 1,
n∈ω
3

we can replace terminal digits 1 with 0 followed by repeating

2:
0.d1 · · · dn 1 = 0.d1 · · · dn 0222 · · · = 0.d1 · · · dn 02.

 Cantor Set 

 ω b
b 1
b

{0, 1, 2} b
0.222
ω r {2} b
b 0.221
b

{0, 1} b
0.22

ω r {1} b
b 0.21
b

{0, 2} b
0.202
ω r {1, 2} b
b 0.201
b

{0} b
0.2

X 2
X 3k+1
k∈X

ω r {0} b
b 0.1
b

{1, 2} b
0.022
ω r {0, 2} b
b 0.021
b

{1} b
0.02

ω r {0, 1} b
b 0.01
b

{2} b
0.002
ω r {0, 1, 2} b
b 0.001
b

∅ b
0
Figure .: The Cantor set

 Cardinality 
Thus each of the endpoints of a component interval of Cn is a
series
∞
X 2ek
,
k=0
3k+1
where each ek is in the set 2. Then the series is just
X 2
k+1
,
k∈A
3

where A is the subset {k ∈ ω : ek = 1} of ω.

The set of subsets any set A is the called the power set of
A, for reasons to be suggested by (.). We shall denote the
power set of A by P(A). Thus

P(A) = {X : X ⊆ A} = {X : ∀y (y ∈ X =⇒ y ∈ A)}.
(.)

 Theorem. The Cantor set consists precisely of the points

of the interval [0, 1] that can be written in ternary notation
without use of the digit 1. That is,
( )
X 2
C= : X ∈ P(ω) .
k∈X
3k+1

Proof. Using (.) and (.), and applying an infinitary De

Morgan law, we compute
\ [
[0, 1] r C = C0 r Cn+1 = (C0 r Cn+1 )
n∈ω n∈ω
[
= (Cn r Cn+1 ). (.)
n∈ω

 Cantor Set 

In ternary notation, by (.),

C0 r C1 = (0.1, 0.2),

which is just the set of numbers in [0, 1] that need the digit 1
in the first place after the point. Likewise,

C1 r C2 = (0.01, 0.02) ∪ (0.21, 0.22),

consisting of numbers needing the digit 1 not in the first place,

but in the second place after the point. In general,
!
[ X 2ek 1 X 2ek 2
Cn r Cn+1 = + , + ,
e∈n 2 k∈n
3k+1 3n+1 k∈n 3k+1 3n+1

and so [0, 1] r C consists of the points of [0, 1] that do need 1

at any place in their ternary expansions.

We now define a function g from P(ω) to R by

X 2
g(A) = . (.)
k∈A
3k+1

Then
C = g[P(ω)] = {g(X) : X ∈ P(ω)},
which is the range of g or the image of P(ω) under g. Note
that g is increasing, in the sense that

A ⊂ B =⇒ g(A) < g(B). (.)

Also g is additive in the sense that

A ∩ B = ∅ =⇒ g(A ∪ B) = g(A) + g(B). (.)

 Cardinality 
Here we use the result from analysis that the sum of an ab-
solutely convergent series is independent of the order of the
terms.
The symbol △ denotes the operation of taking the sym-
metric difference of two sets, so that

A △ B = (A r B) ∪ (B r A).

Then
A = B ⇐⇒ A △ B = ∅. (.)

 Theorem. The function g given by (.) is injective.

Proof. Suppose A and B are distinct subsets of ω. Then A △

B 6= ∅ by (.), and so, since ω is well-ordered, we may let

m = min(A △ B).

Then

A ∩ m = {x ∈ A : x < m} = {x ∈ B : x < m} = B ∩ m.

We may assume m ∈ B r A. Then

A ⊆ (A ∩ m) ∪ {x ∈ ω : x > m},
(A ∩ m) ∪ {m} ⊆ B,

and so, by (.) and (.),

1 2
g(A) 6 g(A ∩ m) + < g(A ∩ m) + 6 g(B).
3m+1 3m+1
In short, if A 6= B, then g(A) 6= g(B). Thus g is injective.

 Cantor Set 

. Comparison of sets
If there is a bijection from a set A to a set B, we shall write
A ≈ B,
and we shall say that A and B are equipollent (words like
“equipotent” are also used). By Theorem  then,
P(ω) ≈ C. (.)
A set is called countable if it is equipollent with a subset of
ω (possibly ω itself). We shall show presently that C is not
countable.
If there is an injection from A to B, that is, an embedding
of A in B, we may write simply
A 4 B.
By Theorem  then,
P(ω) 4 R.
By definition
A ≺ B ⇐⇒ A 4 B & A 6≈ B.
 Cantor’s Theorem. For all sets A,
A ≺ P(A).
Proof. The map x 7→ {x} embeds A in P(A). Suppose some
function f embeds A in P(A). Let
B = {x ∈ A : x ∈
/ f (x)}.
Then B ∈ P(A), and for all c in A we have
c ∈ B ⇐⇒ c ∈
/ f (c).
Thus B 6= f (c). This means f is not surjective onto P(A).

 Cardinality 
 By (.) now,
ω ≺ C,
and so C is uncountable. Note also that, for any set A,

P(A) ≈ A 2, (.)

since there is a bijection X 7→ χX from P(A) to A 2 given by

(
1, if x ∈ B,
χB (x) = (.)
0, if x ∈ A r B,

where B ranges over P(A). Here χ is the minuscule Greek

letter khi or chi, standing for characteristic, since χB can be
called the characteristic function of B; strictly this function
depends on A as well. The inverse of X 7→ χX is σ 7→ σ −1 (1),
where, if i ∈ 2,

σ −1 (i) = {k ∈ ω : σ(k) = i}. (.)

By Cantor’s Theorem,

A ≺ A 2.

When n ∈ ω, the size of the set n 2 is just 2n . This, with

(.), is why the power set is so called. Cantor’s Theorem is
a generalization of the theorem that n < 2n .
The proof of Cantor’s Theorem is similar to one proof that
there are classes that are not sets. In set theory, a class is the
collection
{x : ϕ(x)}
of sets that satisfy a formula ϕ(x) of the language of set theory.
We shall not define formulas now, but shall only give examples.

 Cantor Set 

 • For each set A,

A = {x : x ∈ A},

the class of elements of A. Variables stand for sets; thus

the language of set theory requires all elements of sets to
be sets. Von Neumann’s definition of the natural num-
bers conforms to this convention: numbers are sets, and
their elements (which are other numbers) are sets.
• We let
V = {x : x = x},
the universal class, consisting of all of the sets.
• For an arbitrary class D, generalizing (.), we let

P(D) = {x : x ⊆ D},

the class of all subsets of D.

We now have
P(V) = V. (.)

 Exercise. Why does (.) not contradict Cantor’s Theo-

rem?

 Russell Paradox. The class {x : x ∈ / x} of sets that are

not elements of themselves cannot be a set.

Proof. If we denote the class by R, then for every set b we

have
b ∈ R ⇐⇒ b ∈ / b,
and so b 6= R. Thus R is not a set.
The conclusion of the Russell Paradox is that the class called
R in the proof is a proper class: a class that is not a set.

 Cardinality 
The Separation Axiom is that every sub-class of a set is a
set. Since every class is a sub-class of the universal class V, it
follows that this too is a proper class. But we already knew
this from (.) and Cantor’s Theorem.
We now define a function f , like g in (.), from P(ω) to
R. We let X 1
f (A) = k+1
.
k∈A
2
Then f (∅) = 0 and f (ω) = 1, and in general
f (A) ∈ [0, 1].
Indeed,
f [P(ω)] = [0, 1],
since every number in the interval has a binary expansion.
However, some of these numbers have more than one binary
expansion. For example, the equation f (X) = 1/2 has two
solutions, {0} and ω r {0}; so f is not injective. By the
Axiom of Choice, f has a right inverse h, which means


f h(a) = a (.)
for all y in [0, 1]. Then h is injective, so
[0, 1] 4 P(ω). (.)
In fact we can make an explicit definition of h once for all,
without needing the Axiom of Choice. If a ∈ [0, 1], we define
h(a) = {k ∈ ω : ak = 1},
where the ak are defined recursively by
 X ai 1

 0, if a < i+1
+ k+1
,
2 2


i∈k
ak = X ai 1

 1, if a > i+1
+ k+1
.
2 2


i∈k

 Cantor Set 

This ensures (.). We have defined the an so that, for all
k in ω, for some n in ω, we have k 6 n and an = 0, unless
a = 1.
We have
R ≈ (0, 1) (.)
because the function
x − 1/2
x 7→ ,
x · (1 − x)

restricted to (0, 1), is a bijection onto R, as in Figure ..

Indeed, the inverse can be computed as
p
1 1 1 + y2 − 1
y 7→ + ·
2 2 y
away from 0. This is by the computations
x − 1/2
y= ,
x · (1 − x)
yx − yx2 = x − 1/2,
yx2 + (1 − y)x − 1/2 = 0,
p p
y − 1 ± (1 − y)2 + 2y y − 1 ± 1 + y2
x= = ,
2y 2y
and we let the ambiguous sign ± be + to put x in the interval
(0, 1). In sum, by (.), (.), and (.),

P(ω) ≈ C ⊆ R ≈ (0, 1) ⊆ [0, 1] 4 P(ω). (.)

By the next theorem, we shall be able to conclude

R ≈ P(ω). (.)

 Cardinality 
 3

−2 −1 1 2 3

−1

−2

−3

x − 1/2
Figure .: Graph of y =
x · (1 − x)

 Cantor Set 

 A B

A0 B0
g[B] f [A]

A0 B0
A B
 1   1 
g f [A] f g[B]

A0 B0
A1 B1
A B
h  2 i h  2 i
g f g[B] f g f [A]

Figure .: Cantor–Schröder–Bernstein Theorem

 Cantor–Schröder–Bernstein Theorem. If A 4 B and

B 4 A, then
A ≈ B.
Proof. Suppose f is an injection from A to B; and g, from B
to A. By recursion, we define
A0 = A r g[B], B0 = B r f [A],
An+1 = g[Bn ], Bn+1 = f [An ].
See Figure .. By induction, we prove
[ [
An ⊆ A r Ak , Bn ⊆ B r Bk .
k<n k<n

 Cardinality 
This is clear when n = 0, and if true when n = m, then for
example [
Bm+1 ⊆ f [A] r Bk+1 ,
k<m
S
but the latter is B r k<m+1 Bk , since

f [A] = B r B0 . (.)

Then also, not by induction,

A2n ∪ A2n+1 ≈ B2n+1 ∪ B2n ,

and therefore [ [
An ≈ Bn .
n∈ω n∈ω

Finally, [ [
Ar An ≈ B r Bn ,
n∈ω n∈ω

since
" # " #
[ [ [
f Ar An = f A r An
n∈ω n∈ω k<n
" !# " #
\ [ \ [
=f Ar An = f Ar Ak
n∈ω k<n n∈ω k<n
!
\ [ [
= Br Bk =Br Bn ,
n∈ω k6n n∈ω

by an infinitary De Morgan law as in (.), and (.).

 Cantor Set 

 Topology
. Open and closed sets
We shall study how to generalize the following into the Cantor
Intersection Theorem.

 Lemma. If a sequence (Fn : n ∈ ω) of closed, bounded in-

tervals compose a decreasing chain, then
\
Fn 6= ∅.
n∈ω

Proof. We can write each Fn as [an , bn ]. Then the sequence

(an : n ∈ ω) is bounded and increasing, so it converges. Let

lim an = c. (.)
n→∞

Since

a0 6 a1 6 a2 6 · · · 6 c 6 · · · 6 b2 6 b1 6 b0 , (.)

c belongs to each Fn .
In the proof, we still get (.), provided each Fn is bounded,
so that we can define

an = inf Fn , bn = sup Fn .


If we also have an ∈ Fn , then because the Fn compose a de-
creasing chain, we can conclude

{an : n > k} ⊆ Fk .

What property do the Fk need so that, from (.), we can

conclude c ∈ Fk ? It will be the property of being closed.
Suppose A ⊆ R and b ∈ A. We may denote the comple-
ment, R r A, of A in R by

Ac .

There are two possibilities.

. If, for some positive ε,

(b − ε, b + ε) ⊆ A,

then b is called an interior point of A.

. In the other case, for all positive ε,

(b − ε, b + ε) ∩ Ac 6= ∅,

and now b is a limit point of Ac . In this case, b is also

a limit point of Ac ∪ {b}.
As a special case, if the sequence (an : n ∈ ω) is not eventually
constant, and (.) holds, then c is a limit point of the set
{an : n ∈ ω} and therefore of any set that includes this one.
A subset of R is called
() open, if its every point is an interior point;
() closed, if it contains all of its limit points.
Now Lemma  generalizes, as discussed.
 Cantor Intersection Theorem. Every decreasing chain
of nonempty, closed, bounded subsets of R has nonempty in-
tersection.

 Cantor Set 

 We aim to characterize open and closed sets. Generalizing
notation already used, for any family A of sets, we define
[
A = {x : ∃Y (Y ∈ A & x ∈ Y )},
\
A = {x : ∀Y (Y ∈ A =⇒ x ∈ Y )};

these are the union and intersection of A , respectively. A

family is just a set whose elements are sets whose own ele-
ments will be of interest to us.
 Theorem. A subset of R is open if and only if it is the
union of a family of open intervals.
Proof. Open intervals contain all of their interior points, and
therefore a union of open intervals does the same.
Conversely, for every open set A, for every b in A, writing

EA,b = {ε : ε > 0 & (b − ε, b + ε) ⊆ A},

we have
[
A= (b − ε, b + ε) : b ∈ A & ε ∈ EA,b , (.)

a union of open intervals.

The set Q of rational numbers, and therefore the product
Q × Q, are countable, and Q is dense in R, in the sense that
every interval contains an element of Q.
 Exercise. Show that the family of open intervals in Theo-
rem  can be required to be countable, because, still assuming
A is open, we can rewrite (.) as
[
A= (b − ε, b + ε) : b ∈ A ∩ Q & ε ∈ EA,b ∩ Q .

 Topology 
 Theorem. In R,
() if A and B are closed, so is A ∪ B; T
() if F is a family of closed sets, then F is closed;
() every closed set is the intersection of a family, each of
whose elements is the union of finitely many closed in-
tervals.

 Exercise. Prove the theorem.

In the theorem, the intersection of the empty family should

be understood as R; the union of the empty family is the
empty set. Thus R and ∅ are closed. Each of them being the
complement of the other, they are also open.
According to Theorem , the closed intervals of R compose
a sub-basis, and their finite unions compose a basis, for a
topology of closed sets on R. By Theorem , the open inter-
vals compose a basis of open sets for the same topology. For a
precise definition, some writers define a topology on a set to be
the family of subsets that are to be called open; other writers,
the family of subsets called closed. In any case, a set with a
topology is called a topological space, and it has open and
closed subsets meeting the conditions discussed:
• finite unions of closed sets are closed,
• arbitrary intersections of closed sets are closed,
and complementarily,
• finite intersections of open sets are open,
• arbitrary unions of open sets are open,

. Clopen sets

In any topological space, being the union of the empty set of
open sets, the empty set is open; being the union of the empty

 Cantor Set 

set of closed sets, it is closed. Then the whole space is also
open and closed. We call such sets clopen.

 Theorem. The only clopen subsets of R are R and ∅.

Proof. Suppose A ⊆ R and b ∈ A, but c ∈ Ac . We may

assume b < c. If b is a limit point of Ac , then A is not open.
If b is an interior point of A, let

d = sup{ε : ε > 0 & [b, b + ε) ⊆ A}.

Then d is a limit point of both A and Ac , so one of these is

not closed.
The intersection of an open subset of R with C is called
open in C; it might not be open in R (in fact it will not be,
unless it is empty). The intersection of a closed subset of R
with C is called closed in C, but it is still closed in R anyway.
In this way we obtain a topology on C, namely the subspace
topology inherited from R.
Let us recall that C is the intersection of the sets Cn as in
(.), where [
Cn = Ie , (.)
e∈2 n

where  
1
Ie = ae , ae + n , (.)
3
where we have X 2ek
ae = . (.)
k∈n
3k+1
Let us write [
n
ω>
2= 2. (.)
n∈ω

 Topology 
If e ∈ ω> 2, let us define

De = Ie ∩ C. (.)

Each of these is immediately closed in C.

 Theorem. The subsets De of C are open and compose a

basis of open sets for the subspace topology on C inherited from
R.

Proof. For each n in ω,

[
C= De .
e∈n 2

Since n 2 is finite, and the sets De are disjoint from one another,
the complement of each of these is closed, so each is open as
well as closed.
Now let U be an open subset of R. For every a in C ∩ U,
for some positive ε,

(a − ε, a + ε) ⊆ U.

With g as in (.), we know by Theorem  that a is g(A) for

some subset A of ω. Thus
X 2
a= .
k∈A
3k+1

For any n in ω, by (.),

g(A ∩ n) 6 a 6 g(A ∩ n) + g({x ∈ ω : x > n})

1
= g(A ∩ n) + .
3n

 Cantor Set 

Now let e in n 2 be such that

g(A ∩ n) = ae

as in (.). With Ie as in (.), and now requiring n to be so

large that 3n > 1/ε, we have

Ie ⊆ (a − ε, a + ε),

and so
De ⊆ U ∩ C.
Since also a ∈ De , and a was an arbitrary element of U ∩ C,
this is a union of various sets De .
We now know that unions of finite subsets of

{De : e ∈ ω> 2}

are clopen subsets of C. We aim to show the converse. If E

is a clopen subset of C, so that, being open, E is a union of
someSfamily U of sets De , we shall show that, being closed,
E is U0 for some finite subset U0 of U .

. Compactness
If a topological space is the union of a certain family of open
subsets, this family is called an open cover of the space. If
some finite subset of the family also covers the space, that
subset is called a finite sub-cover. If every open cover of
a space has a finite sub-cover, then the space itself is called
compact.
Easily, every closed subset of a compact space is compact
in the subspace topology, since an open cover of the subset,

 Topology 
together with the complement of the subset, yields an open
cover of the whole space.
There is an equivalent definition of compactness in terms of
closed sets. A family of subsets of a topological space whose
every finite subset has nonempty intersection is said to have
the Finite Intersection Property or FIP. Then the space
is compact if and only if every family of subsets with FIP has
nonempty intersection.
 Heine–Borel Theorem. Every closed bounded subset of
R is compact in the subspace topology.
Proof. For reasons discussed, it is enough to show that an
interval [a, b] is compact. Writing this interval as I0 , we let
F be a collection of closed subsets of I0 that has FIP. Let
c = 12 (a + b), so that
a < c < b.
One of the intervals [a, c] and [c, b] is an interval I1 such that
the family F ∪ {I1 } has FIP. For, suppose I1 cannot be [a, c].
Then for some finite subset F0 of F ,
\
F0 ∩ [a, c] = ∅.

But then for every finite subset F1 of F , since

\ \ \
(F0 ∪ F1 ) = F0 ∩ F1 6= ∅

(note that the signs are correct, and we are not applying a De
Morgan law), we must have
\ \
F0 ∩ F1 ∩ [c, b] 6= ∅.

Thus F ∪ {[c, b]} must have FIP.

 Cantor Set 

 Continuing in this way, we obtain a decreasing chain of
closed intervals In such that
1
µ(In ) = n µ(I0 ), (.)
2
and each family F ∪ {In } hasT FIP. By the Cantor Intersection
Theorem, or just Lemma , n∈ω In contains a point d. By
(.) then, \
In = {d}.
n∈ω
T
Then d must be in F . For, if it is not, then some E in F
does not contain d. Since E is closed, d is not a limit point of
it, and so for some positive ε,
E ∩ (d − ε, d + ε) = ∅.
If n is so large that µ(In ) < ε, then
\
{E, In } = ∅,
contradicting that F ∪ {In } has FIP.
For reasons discussed, we now know that the clopen subsets
of C are just the finite unions of sets De as in (.).

. Continuity
One way that notion of compactness arises in mathematics is
as follows. On a subset A of R, a function f is continuous if
 
∀ε ε > 0 =⇒ ∀x x ∈ A =⇒ ∃δ δ > 0 & ∀y



(y ∈ A & |x − y| < δ =⇒ |f (x) − f (y)| < ε) ,

 Topology 
while f is absolutely continuous if
 
∀ε ε > 0 =⇒ ∃δ δ > 0 & ∀x x ∈ A =⇒ ∀y



(y ∈ A & |x − y| < δ =⇒ |f (x) − f (y)| < ε) .

Note that I write

∀ε (ε > 0 =⇒ . . . rather than ∀(ε > 0) . . . ,

∃δ (δ > 0 & . . . rather than ∃(δ > 0) . . .

 Exercise. Using the Heine–Borel Theorem, prove that, if A

is closed and bounded, then every function that is continuous
on A is uniformly continuous on A.
It will be useful to note that continuity of f on A is equiva-
lent to the requirement that, for every open set U, f −1 (U) be
open, and similarly for closed sets.

. König’s Lemma

If e ∈ n 2, so that the domain of e is n, we may write

dom(e) = n.

If m 6 n, then
e ↾ m = (e0 , . . . , em−1 ), (.)
and we may write
e ↾ m ⊆ e.
When understood to be ordered by inclusion in this way, the
set ω> 2 defined by (.) is a binary tree of height ω. Note

 Cantor Set 

that 0 2 has the single element ∅. For elements of the binary
tree, we may use notation as in (.), so that if σ ∈ ω 2 and
n ∈ ω, we write
σ ↾ n = (σ(0), . . . , σ(n − 1)).
If e ∈ n 2 and i ∈ 2, we may write
(e, i) = (e0 , . . . , en−1 , i),
an element of n+1 2. In our proof of the Heine–Borel Theorem,
we work with the binary tree
(Ie : e ∈ ω> 2),
where I∅ is the closed interval [a, b], and each Ie is divided at
its midpoint into the closed intervals I(e,0) and I(e,1) . To be
precise, if again e ∈ n 2, then
 
b−a
Ie = ce , ce + n ,
2
where X ek
ce = a + (b − a) · ,
k∈n
2k+1
so that
b−a
Ie = I(e,0) ∪ I(e,1) , c(e,0) + = c(e,1) .
2n+1
We find σ in ω>
2 such that, for each n in ω, the family
F ∪ {Iσ↾n }
has FIP. We can do this because, if F ∪ {Ie } has FIP, then
so does one of F ∪ {I(e,0) } and F ∪ {I(e,1) }. The same idea
yields the following.

 Topology 
 König’s Lemma. If T is an infinite subset of ω
2, then
for some σ in ω 2, the set

{n ∈ ω : σ ↾ n ∈ T }

is infinite. More generally, every infinite finitely branching

tree has an infinite branch.

Proof. If, for some n in ω, for some e in n 2, the set

{x ∈ T : e ⊆ x}

is infinite, then, for some i in 2, the set

{x ∈ T : (e, i) ⊆ x}

must be infinite, since

[
{x ∈ T : e ⊆ x} ⊆ {e} ∪ {x ∈ T : (e, i) ⊆ x}.
i∈2

If perhaps not 2, but a finite number of elements of the tree

are immediately above a given element, the same argument
works.

In the same way, we can prove directly that C is compact.

For suppose F is a family of subsets of C having FIP. In the
notation of (.), we have

C = D∅ ,

and for all n in ω, for all e in n 2,

De = D(e,0) ∪ D(e,1) .

 Cantor Set 

Then trivially F ∪ {D∅ } has FIP, and if F ∪ {De } has FIP,
then so must F ∪ {D(e,i) } for some i in 2. Hence for some σ
in ω 2, for each n in ω, each set F ∪ {Dσ↾n } has FIP. Since,
with g as in (.), with σ −1 as in (.) we have
 
\  X 2  
Dσ↾n = k+1
= g(σ −1 (1)) ,
n∈ω
 3 
σ(k)=1

we can conclude as before

\
g(σ −1 (1)) ∈ F.

. Tychonoff topology

As in (.) we defined the map g from P(ω) to R that turned
out, by Theorems  and , to be a bijection onto C, so now we
let G be the bijection from ω 2 to C given by
X 2σ(k)
G(σ) = .
k∈ω
3k+1

For each n in ω, for each i in 2, let us define

En i = {σ ∈ ω 2 : σ(n) = i},

so that [
G[En i ] = D(e,i) ,
e∈n 2

and also, for each e in n 2,

\
G−1 (De ) = {σ ∈ ω 2 : σ ↾ n = e} = Ek ek .
k∈n

 Topology 
Therefore the sets En i compose a sub-basis of closed sets for
a topology on ω 2 in which G is a homeomorphism onto
C. This means G is a bijection, and both G and G−1 are
continuous. Generalizing from the description of continuity
on page , we define a function f from one topological space
to another to be continuous if f −1 (F ) is closed for every
closed set F .
We give the set 2 the topology in which every subset is
closed. Then the topology on ω 2 is the product topology or
Tychonoff topology for the following reason.
If (Ωi : i ∈ I) is an indexed family of topological spaces, we
define the product Y
Ωi
i∈I
S
as the set of functions f from I to i∈I Ωi such that, for each
i in I,
f (i) ∈ Ωi .
If we denote the product by Ω, we can define, for each i in I,
the function πi from Ω to Ωi given by

πi (f ) = f (i).

Here Ω is the capital Greek letter Omega.∗ The Tychonoff

topology on Ω is the weakest topology in which each function
πi is continuous. This means that the sets πi −1 (F ), where F
ranges over the closed subsets of Ωi , where i ranges over I,
compose a sub-basis of closed subsets of Ω. In the special case
when I is ω, and each Ωi is 2, we recover the topology on ω 2
that we have already defined.
Often the Capital letter X is used for a topological space, but I
∗

prefer to use that letter as a variable for sets.

 Cantor Set 

 Tychonoff Theorem. Every product of compact spaces is
compact in the Tychonoff topology.
Proof. In the notation above, let F be family of closed subsets
of Ω that has FIP. Each element F of F is the intersection of
a family EF of sub-basic closed subsets. Now let
[
F∗ = EF .
F ∈F

This too has FIP, and

\ \
F∗ = F.
Suppose now F ∗ contains
[
πi(k) −1 (Fk )
k∈n

for some (i(k) : k ∈ n) in n I and some closed subset Fk of Ωi(k)

for each k in n. As before, for some k in n, the set
F ∗ ∪ {πi(k) −1 (Fk )}
must have TFIP; moreover, the intersection of this set is in-
cluded in F ∗ itself. By continuing in this way—strictly
we use Zorn’s Lemma, which follows from the Axiom of
Choice)—, we obtain a set G with FIP such that
\ \
G ⊆ F ∗,
and whose every element is, for some i in I, of the form
πi −1 (F ), where F is a closed subset of Ωi . For each i in I,
let Gi consist of the closed subsets FTof Ωi such that πi −1 (F )
belongs to G . Then Gi has FIP, so Gi has an element ai , if
we assume Ωi is compact. Then
\
(ai : i ∈ I) ∈ G.

 Topology 
 Algebra
. Logic
The Tychonoff topology on ω 2 arises independently in propo-
sitional logic as follows.
Starting with a collection {Pk : k ∈ ω} of (propositional)
variables, we define (propositional) formulas recursively:
. Each variable is a formula, namely an atomic formula.
. If F is a formula, then so is ¬G, the negation of F .
. If F and G are formulas, then so is (F ∧ G), the con-
junction of F and G.
Let us denote by
L
the set of propositional formulas so defined. Then for example,
if L contains F and G, then it contains

¬(¬F ∧ ¬G),

which is the disjunction of F and G; we may denote this by

(F ∨ G). Then the expression

(¬F ∨ G)

stands for another formula in L, denoted by F → G. Officially

though, L is the smallest set that includes {Pk : k ∈ ω} and is
closed under the operations F 7→ ¬F and (F, G) 7→ (F ∧ G).


 Lemma. Every formula in L is uniquely readable:
) no atomic formula is also a negation or a conjunction;
) no negation is also a conjunction;
) every conjunction is uniquely so.

Proof. Only the last claim is not entirely clear. By induction

in L, we show that for all formulas F ,
(a) no proper initial segment of F is a formula, and
(b) F is not a proper initial segment of any formula.
. The claim is clearly true when F is atomic.
. Suppose the claim is true when F is a formula G. Then
the claim must be true when F is ¬G. For if H is a proper
initial segment of ¬G, then H is of the form ¬K for some K,
which is a proper initial segment of G, so, by hypothesis, K
cannot be a formula, and therefore H cannot be a formula.
There is a similar argument if ¬G is a proper initial segment
of H.
. Similarly, if the claim is true when F is G or H, then it
must be true when F is (G ∧ H).
By induction, which is made possible by the recursive defi-
nition of L, the claim holds for all formulas F . The last part
of the theorem now follows.

Lemma  allows us to make recursive definitions of functions

on L. For example, for all subsets A of ω, we recursively define
which formulas are true in A. We shall express that a formula
F is true in A by writing

A  F.

 Algebra 
Then by definition

A  Pk ⇐⇒ k ∈ A, (.a)
A  ¬F ⇐⇒ A 2 F, (.b)
A  (F ∧ G) ⇐⇒ A  F & A  G. (.c)

Note that the expressions ⇐⇒ and & here, as throughout

these notes, are just abbreviations of ordinary language.
Without recursion, if F ∈ L, we define

Mod(F ) = {X ∈ P(ω) : X  F }. (.)

If Γ ⊆ L, we define
\
Mod(Γ) = {Mod(F ) : F ∈ Γ}; (.)

this is the set of models of Γ.

 Lemma. The family

{Mod(Γ) : Γ ⊆ L}

is the family of closed subsets of P(ω) in the topology whereby

the function X 7→ χX , which is given by (.) and whose
inverse σ 7→ σ −1 (1) is given by (.), is a homeomorphism
with ω 2 in the Tychonoff topology. The clopen subsets of the
topology on P(ω) compose the family

{Mod(F ) : Γ ∈ L}.

Proof. The family of sets Mod(F ) is a basis of closed sets for a

topology, since the family is closed under taking finite unions.
Indeed the family contains the empty set as

Mod(F ∧ ¬F ),

 Cantor Set 

and the family is closed under binary unions since

Mod(F ) ∪ Mod(G) = Mod(F ∨ G). (.)

Moreover,

{X ∈ P(ω) : χX ∈ Ek 1 }
= {X ∈ P(ω) : k ∈ X} = Mod(Pk ),

and likewise

{X ∈ P(ω) : χX ∈ Ek 0 } = Mod(¬Pk ),

so X 7→ χX is continuous. Since in addition to (.) we have

generally

Mod(F ) ∩ Mod(G) = Mod(F ∧ G), (.)

Mod(F )c = Mod(¬F ), (.)

by means of the De Morgan laws we obtain that each set

Mod(F ) is a finite union of finite intersections of sets Mod(Pk )
and Mod(¬Pk ). This yields that σ 7→ σ −1 (1) is continuous,
and the rest follows.

If every finite subset of Γ has a model, we shall say that Γ

is consistent. This is another way of saying that the family
{Mod(F ) : F ∈ Γ} has FIP. We shall show that every con-
sistent set of formulas has a model. This is another way of
saying P(ω) is compact. We already know this compactness,
but now we shall prove it in a new way, or at least in new
language.

 Algebra 
 To do so, we make one more recursive definition, parallel to
(.).

V (Pk ) = {k}, (.a)

V (¬F ) = V (F ), (.b)


V (F ∧ G) = V (F ) ∪ V (G). (.c)

We do not really need recursion here: we can just say

V (F ) = {k ∈ ω : Pk occurs in F }.

We are defining a formal logic. Logic lets us do mathematics

with logical precision. Such precision may be illusory when
used to define the logic in the first place. The same consider-
ation arises in the proof of the following.

 Lemma. Let F be a formula, and let A and B be subsets of

ω such that
V (F ) ∩ A = V (F ) ∩ B.
Then
A  F ⇐⇒ B  F.

Proof. The theorem is that whether F is true in A depends

only on whether k ∈ A when Pk actually occurs in F . This is
obvious when F is an atomic formula, by (.a). The remain-
ing rules (.b) and (.c) maintain the claim, since they do
not involve variables explicitly. In all formal detail though, we
can use induction to show

A  F ⇐⇒ V (F ) ∩ A  F

as follows.

 Cantor Set 

 . Supposing first that F is an atomic formula Pk , we have
V (F ) = {k} by (.a), and then
A  F ⇐⇒ k ∈ A [by (.a)]
⇐⇒ k ∈ V (F ) ∩ A [by (.a)]
⇐⇒ V (F ) ∩ A  F. [by (.a) again]
. Suppose the claim is true when F is a formula G. Then
A  ¬G ⇐⇒ A2G [by (.b)]
⇐⇒ V (G) ∩ A 2 G [by hypothesis]
⇐⇒ V (¬G) ∩ A 2 G [by (.b)]
⇐⇒ V (¬G) ∩ A  ¬G, [by (.b) again]
so the claim holds when F is ¬G.
. Suppose finally the claim is true when F is either of G
and H. Since



V (G) ⊆ V (G ∧ H) , V (H) ⊆ V (G ∧ H) (.)
by (.c), so that


V (G) ∩ V (G ∧ H) ∩ A = V (G) ∩ A, (.)


V (H) ∩ V (G ∧ H) ∩ A = V (H) ∩ A,
we have

A  G ⇐⇒ V G ∧ H) ∩ A  G,

(.)
A  H ⇐⇒ V G ∧ H) ∩ A  H,
since for example
A  G ⇐⇒ V (G) ∩ A  G [by hyp.]


⇐⇒ V (G) ∩ V (G ∧ H) ∩ A  G [by (.)]


⇐⇒ V (G ∧ H) ∩ A  G. [by hyp.]

 Algebra 
This gives us

A  (G ∧ H) ⇐⇒ A  G & A  H [(.c)]


⇐⇒ V (G ∧ H) ∩ A  G


& V (G ∧ H) ∩ A  H [(.)]


⇐⇒ V (G ∧ H) ∩ A  (G ∧ H). [(.c)]

This completes the induction.

 Compactness Theorem. Every consistent set of propo-

sitional formulas has a model.

Proof. Let Γ be a consistent set of formulas. As in our earlier

proofs, one of Γ ∪ {P0 } and Γ ∪ {¬P0 } must be consistent.
In this way, by recursion, we obtain a sequence (Gk : k ∈ ω),
where each Gk is either Pk or ¬Pk , and each collection Γ ∪
{Gk : k < n} is consistent. Let

A = {k ∈ ω : Gk is Pk }.

For all F in Γ, the collection

{F } ∪ {Gk : k ∈ V (F )},

being finite, has a model B. Then for all k in V (F ), we have

B  Gk , and so
k ∈ B ⇐⇒ k ∈ A.

By the last lemma, since B  F , also A  F . Thus A ∈

Mod(Γ).

 Cantor Set 

. Boolean algebras
The relation ∼ of logical equivalence on L is given by
F ∼ G ⇐⇒ Mod(F ) = Mod(G). (.)
We can now define
[F ] = {X ∈ L : X ∼ F },
L/∼ = {[X] : X ∈ L}.
The definitions ensure that there is a well-defined injection
[X] 7→ Mod(X)
from L/∼ to P(ω). By Lemma , the map is also surjective
onto the collection B of clopen subsets of ω. Here B is closed
under
() the binary operations ∩ and ∪,
() the singulary operation c , and
() the nullary operations ∅ and ω (that is, B contains
these sets).
This makes B a Boolean subalgebra of P(ω). Because of
(.), (.), and (.), along with
Mod(P0 ∧ ¬P0 ) = ∅,
Mod(P0 ∨ ¬P0 ) = ω,
L/∼ is a Boolean algebra with respect to the (well-defined)
operations given by
[F ] ∧ [G] = [F ∧ G],
[F ] ∨ [G] = [F ∨ G],
¬[F ] = [¬F ],
⊥ = [P0 ∧ ¬P0 ],
⊤ = [P0 ∨ ¬P0 ].

 Algebra 
In general, an abstract Boolean algebra (like L/∼) is a set
B with operations ∧, ∨, ¬, ⊥, and ⊤ with the following prop-
erties.
. The binary operations ∨ and ∧ are commutative:

x ∨ y = y ∨ x, x ∧ y = y ∧ x.

. The elements ⊥ and ⊤ are identities for ∨ and ∧ re-

spectively:

x ∨ ⊥ = x, x ∧ ⊤ = x.

. ∨ and ∧ are mutually distributive:

x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z),
x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z).

. The element ¬x is a complement of x:

x ∨ ¬x = ⊤, x ∧ ¬x = ⊥.

Additional properties like associativity of ∧ and ∨ follow from

the given identities:

 Lemma (E. Huntington, ). In any Boolean algebra:

x ∨ x = x, x ∧ x = x,
x ∨ ⊤ = ⊤, x ∧ ⊥ = ⊥,
x ∨ (x ∧ y) = x, x ∧ (x ∨ y) = x, (.)
¬¬x = x,
¬(x ∨ y) = ¬x ∧ ¬y, ¬(x ∧ y) = ¬x ∨ ¬y,
(x ∨ y) ∨ z = x ∨ (y ∨ z), (x ∧ y) ∧ z = x ∧ (y ∧ z).

 Cantor Set 

Proof. We prove only (.):

x ∨ (x ∧ y) = (x ∧ ⊤) ∨ (x ∧ y)
= x ∧ (⊤ ∨ y) = x ∧ ⊤ = x.

We shall investigate how P(ω), or more precisely a space

homeomorphic with it, can be obtained from the Boolean al-
gebra L/∼. The same construction will work for any Boolean
algebra, and then the algebra can be recovered as being iso-
morphic to the algebra of clopen subsets of the space. This is
the Stone Representation Theorem, Theorem  below.
We shall need that, on any Boolean algebra, there is a partial
ordering ⊢ given by

x ⊢ y ⇐⇒ x ∧ y = x.

Note that, by (.),

x ∧ y = x ⇐⇒ x ∨ y = y.

If A ⊆ ω, we define

Th(A) = {X ∈ L : A  X}. (.)

This is the theory of A, and it has the following properties.

F ∈ Th(A) & G ∈ Th(A) =⇒ (F ∧ G) ∈ Th(A), (.)

F ∈ Th(A) & F ⊢ G =⇒ G ∈ Th(A), (.)
F ∈
/ Th(A) ⇐⇒ ¬F ∈ Th(A). (.)

As a result, Th(A), or more precisely the set {[X] : X ∈

Th(A)},
• is a filter of the algebra L/∼, by (.) and (.),

 Algebra 
 • by these and (.), is an ultrafilter.
Note that we can replace (.) and (.) with the single
equivalence

F ∈ Th(A) & G ∈ Th(A) ⇐⇒ (F ∧ G) ∈ Th(A).

We may blur the distinction between formulas in L and their

equivalence classes in L/∼, thus identifying the sets Th(A) and
{[X] : X ∈ Th(A)}. An ultrafilter is a maximal filter, if the
whole algebra is not counted as a filter; any larger filter than
Th(A) would contain some F not in Th(A); but then Th(A)
contains ¬F , so the larger filter contains F ∧ ¬F , which is
equivalent to ⊥, and ⊥ ⊢ G for all G in L.
The converse also holds:

 Lemma. Every ultrafilter of L/∼ is the theory of some el-

ement of P(ω).

Proof. Given an ultrafilter Φ of L/∼, we may let

A = {k ∈ ω : Pk ∈ Φ}.

By induction in L, Φ = Th(A), that is, for all F in L,

F ∈ Φ ⇐⇒ A  F. (.)

In detail:
. (.) holds by (.a) when F is atomic.
. If (.) holds when F is G, then

¬G ∈ Φ ⇐⇒ G ∈
/Φ [by (.)]
⇐⇒ A 2 G [by hypothesis]
⇐⇒ A  ¬G. [by (.b)]

 Cantor Set 

 . If (.) holds when F is either of G and H, then

(G ∧ H) ∈ Φ ⇐⇒ G ∈ Φ & H ∈ Φ [by (.) and (.)]

⇐⇒ A  G & A  H [by hypothesis]
⇐⇒ A  (G ∧ H). [by (.c)]

This completes the induction.

. Stone spaces

For any Boolean algebra B, we shall denote the set of ultrafil-
ters of B by
S(B);
this is the Stone space of B, because it will have a topology.
In our case, we have a bijection X 7→ Th(X) from P(ω) to
S(L/∼), and this will be a homeomorphism.
We have been working with the relation  from P(ω) to L.
We have used it to define, by (.), a map X 7→ Mod(X) from
L to P(P(ω)). This map has the properties given by (.)
and (.), so that, when we define the relation ∼ of logical
equivalence as in (.), the map X 7→ Mod(X) induces a
Boolean-algebra embedding of L/∼ in P(P(ω)). By Lemma
, the embedding is an isomorphism with the algebra of clopen
subsets of P(ω).
We have also defined by (.) a “dual” map, Y 7→ Th(Y ),
from P(ω) to P(L/∼). By Lemma , the map is a bijection
onto S(L/∼), the set of ultrafilters of L/∼.
As by (.) we define Mod(Γ) when Γ ⊆ L, so we can define
\
Th(A ) = Th(Y )
Y ∈A

 Algebra 
when A ⊆ P(ω). Then

Γ ⊆ ∆ =⇒ Mod(Γ) ⊇ Mod(∆),
A ⊆ B =⇒ Th(A ) ⊇ Th(B).

Simply on this basis,

Mod ◦ Th ◦ Mod = Mod,

Th ◦ Mod ◦ Th = Th,

so that there is a one-to-one correspondence, called a Ga-

lois correspondence, between the sets Mod(Γ) and the sets
Th(A ). In the original Galois theory, the correspondence is
between subfields of a field K and subgroups of the group of
automorphisms of K; one obtains this by using, in place of our
, the relation R from the field to the automorphism group
given by
x R σ ⇐⇒ xσ = x.
In our case, the sets Th(A ) are filters of L/∼, while the sets
Mod(Γ) can be called elementary classes (though all this
means is that they are the classes of models of sets of formulas).

 Stone Representation Theorem. Every Boolean alge-

bra embeds in an algebra P(Ω) for some set Ω. Indeed, when
the algebra is B, it embeds in P(S(B)) under the map x 7→ [x],
where
[a] = {U ∈ S(B) : a ∈ U};
and then the subsets [a] of S(B) are the clopen sets in a com-
pact topology on S(B).

Proof. In what we have done, if we replace L with an arbitrary

Boolean algebra B, then we can also replace P(ω) with S(B),

 Cantor Set 

and  with ∈. When we define the map x 7→ [x] from B to
P(S(B)) as indicated, then

[a] ∩ [b] = [a ∧ b],

[a]c = [¬a],

so the map is a homomorphism of Boolean algebras. It is an

embedding, by the Axiom of Choice, or rather by the weaker
axiom called the Prime Ideal Theorem (which is that every
proper ideal of a ring is included in a prime ideal; the Axiom of
Choice gives that every proper ideal is included in a maximal
ideal; maximal ideals are always prime, but in Boolean rings,
prime ideals are also maximal).
Since the image of B in P(S(B)) is closed under finite
unions, it is a basis for a topology of closed sets on S(B).
Thus the closed sets in this topology are the sets
\
[x],
x∈I
T
where I ⊆ B. The topology is compact, because if x∈I0 [x] is
never empty when I0 is a finite subset of I, then I is included
in a proper filter, and therefore (by the Prime
T Ideal Theorem)
an ultrafilter U; but this just means U ∈ x∈I [x].
The topology in the theorem is the Stone topology on
S(B).
A first-order logic, such as the logic of set theory, de-
fines formulas as in propositional logic, except that the atomic
formulas are not propositional variables, but (in the case of
set theory) formulas x ∈ y; also, if ϕ is a formula, and x is
a variable, then so is ∃x ϕ. One has the notion of a free
variable of a formula; if a formula has no free variable, the

 Algebra 
 + 0 1 × 0 1
0 0 1 0 0 0
1 1 0 1 0 1

Figure .: Ring operations on 2

formula is a sentence. Every sentence has a class of models,

which, considered in themselves, are structures. Defining
logical equivalence as before, one obtains a Boolean algebra
of sentences, called a Lindenbaum algebra after a student
of Tarski, murdered by the Nazis. The Stone space of this al-
gebra is automatically compact. The Compactness Theorem
of first-order logic can then be understood as corresponding
to Lemma : it is that every ultrafilter of the Lindenbaum
algebra is in fact the theory of some structure (and it is not
so easy to prove).
Finally, in the Stone Representation Theorem, instead of a
Boolean algebra, we may start with an arbitrary topological
space and extract its algebra of clopen subsets. However, the
Stone space of this algebra will not be homeomorphic with
the original space unless this is compact and totally discon-
nected (for any two points, some clopen set contains only one
of them).

. Topological rings

With respect to the operations of addition and multiplication
defined unsurprisingly as in Figure ., the set 2 becomes a

 Cantor Set 

(commutative) ring because the following identities hold:
x + y = y + x, xy = yx,
x + (y + z) = (x + y) + z, x(yz) = (xy)z,
x + 0 = x, x · 1 = x,
x(y + z) = xy + xz,
x + x = 0.
For an arbitrary commutative ring, the last identity would be
x − x = 0; but −x = x in 2. Also in 2,
x2 = x,
and therefore 2 is called a Boolean ring. From this and the
other identities, x + x = 0 and xy = yx follow, at least if
additive cancellation is assumed, so that
x2 + y 2 = x + y = (x + y)2 = x2 + xy + yx + y 2
and therefore
0 = xy + yx,
and hence as a special case
0 = x2 + x2 = x + x,
and consequently
xy = xy + yx + yx = yx.
A Boolean ring is a Boolean algebra, and conversely, by the
rules
x ∧ y = xy,
x ∨ y = x + y + xy,
¬x = x + 1,
(x ∨ y) ∧ (¬x ∨ ¬y) = x + y.

 Algebra 
Now ω 2 is a Boolean ring with respect to the pointwise oper-
ations given by
(xk : k ∈ ω) ∗ (yk : k ∈ ω) = (xk ∗ yk : k ∈ ω),
where ∗ is + or ×. By the Stone Representation Theorem, ω 2
is isomorphic to the Boolean algebra of clopen subsets in the
Stone topology on S(ω 2). This space consists of the principal
ultrafilters
{χX : a ∈ X},
where a ∈ ω, along with the nonprincipal ultrafilters,
which include the filter
{χX : X c ∈ Pω (ω)},
where Pω (A) consists of the finite subsets of A; but the
existence of the nonprincipal ultrafilters depends on the Prime
Ideal Theorem mentioned above.
The elements χ{n} (where n ∈ ω) are the atoms of the
Boolean algebra ω 2; but when this is given the Tychonoff
topology, the Boolean algebra of clopen subsets is atomless.
The Boolean ring ω 2 is a topological ring with respect to
the Tychonoff topology, because addition and multiplication
are continuous in each factor. Indeed, if A ∈ P(ω), and
• if f is x 7→ x + χA , then
(
En i+1 , if n ∈ A,
f −1 (En i ) =
En i , if n ∈ Ac ;
• if g is x 7→ x · χA , then

En i ,
 if n ∈ A,
−1
f (En i ) = ∅, if n ∈ Ac & i = 1;

2, if n ∈ Ac & i = 0.
ω

 Cantor Set 

 There is another ring structure on ω 2 that makes this a
topological ring with respect to the Tychonoff topology. In
the new ring structure, whether ∗ is + or ×, we have

(xk : k ∈ ω) ∗ (yk : k ∈ ω) = (zk : k ∈ ω)

if and only if, for all n in ω,

X X X
xk 2k ∗ y k 2k ≡ zk 2k (mod 2n ),
k∈n k∈n k∈n

the computations being performed in ω as usual. The element

(xk : k ∈ ω) can be denoted by
X
xk 2k ,
k∈ω

and the new ring itself by Z(2) . This is the ring of dyadic
integers (or 2-adic integers). Since
X
1+ 2k = 0
k∈ω

in the ring, negatives do exist. If a is a nonzero element

2
P
k∈ω ak 2 of Z(2) , we define

1
|a| = ,
2m
where
m = min{n ∈ ω : an 6= 0}.
Also |0| = 0. Then the triangle inequality holds in the
strong form
|x + y| 6 max{|x| , |y|},

 Algebra 
so the function (x, y) 7→ |x − y| is a metric on Z(2) (called
an ultrametric). The induced topology is just the Tychonoff
topology, since, if we define

B(a; r) = {x ∈ Z(2) : |x − a| < r},

so that these are a basis of open sets in Z(2) , then

!
n 1
[ X
k
En i = B ek 2 + i2 ; n ,
e∈n 2 k∈n
2
!
X 1 \
B ak 2k ; n = Ek ak .
k∈ω
2 k∈n

 Cantor Set 