64 3.

Relations and functions

3.3. Finite and Infinite Sets

Broadly speaking (that is, with very little detail contained in what I will say
next) we have sets that are finite —intuitively meaning that we can count all
their elements in a finite amount of time (but see the -remark 3.3.3 below)—
and those that are not, naturally called infinite!
What is a mathematical way to say all this?
Any counting process of the elements of a finite set A will have us say out
loud —every time we pick or point at an element of A— “0th”, “1st”, “2nd”,
etc., and, once we reach and pick the last element of the set, we finally pronounce
“nth”, for some appropriate n that we reached in our counting (Again, see 3.3.3.)
Thus, mathematically, we are pairing each member of the set with a member
from {0, . . . , n}.
So we propose,

3.3.1 Definition. (Finite and infinite sets) A set A is finite iff it is either
empty, or is in 1-1 correspondence with {x ∈ N : x ≤ n}. This “normalized”
small set of natural numbers we usually denote by {0, 1, 2, . . . , n}.
If a set is not finite, then it is infinite. 

3.3.2 Example. For any n, {0, . . . , n} is finite since, trivially, {0, . . . , n} ∼

{0, . . . , n} using the identity (∆) function on the set {0, . . . , n}. 

 3.3.3 Remark. One must be careful when one attempts to explain finiteness
via counting by a human.
For example, Achilles† could count infinitely many objects by constantly
accelerating his counting process as follows:
He procrastinated for a full second, and then counted the first element. Then,
he counted the second object exactly after 1/2 a second from the first. Then
he got to the third element 1/22 seconds after the previous, . . . , he counted the
n th item at exactly 1/2n−1 seconds after the previous, and so on forever.

Hmm! It was not “forever”, was it? After a total of 2 seconds he was done!

You see (as you can easily verify from your calculus knowledge (limits)),‡
1 1 1 1
1+ + 2 + . . . + n−1 + . . . = =2
2 2 2 1 − 1/2
So “time” is not a good determinant of finiteness!  

3.3.4 Theorem. If X ⊂ {0, . . . , n}, then there is no onto function f : X →

{0, . . . , n}.
 I am saying, no such f , whether total or not; totalness is immaterial. 
† OK, he was a demigod; but only “demi”.
1 1−1/2n
‡1 + 2
+ 212 + . . . + 2n−1
1
= 1−1/2 . Now let n go to infinity at the limit.

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 3.3. Finite and Infinite Sets 65

Proof. First off, the claim holds if X = ∅, since then any such f equals ∅ and
its range is empty.
Let us otherwise proceed by way of contradiction, and assume that the theo-
rem is wrong: That is, assume that it is possible to have such onto functions,
for some n and well chosen X.
Since I assume there are such n > 0 values, suppose then that the smallest
n that allows this to happen is, say, n0 , and let X0 be a corresponding set “X”
that works, that is,

Assume that we have an onto f : X0 → {0, . . . , n0 } (1)

Thus X0 6= ∅, by the preceding remark, and therefore n0 > 0, since otherwise

X0 = ∅.

Let us call H be the set of all x such that f (x) = n0 , for short, H =
f −1 ({n0 }). ∅ =
6 H ⊆ X0 ; the 6= by ontoness.

Case 1. n0 ∈ H. Then removing all pairs (a, n0 ) from f —all these have a ∈
H— we get a new function f 0 : X0 − H → {0, 1, . . . , n0 − 1}, which is
still onto as we only removed inputs that cause output n0 .
This contradicts minimality of n0 since n0 − 1 works too!

0
1

Case 2. n0 ∈
/ H.
If n0 ∈
/ X0 , then we argue exactly as in Case 1 and we just remove the
base “H” of the cone (in the picture) from X0 .
Otherwise, we have two subcases:

• f (n0 ) ↑. Then (almost) we act as in Case 1: The new “X0 ” is

(X0 − H) − {n0 }, since if we leave n0 in, then the new “X0 ” will
not be a subset of {0, 1, . . . , n0 − 1}. We get a contradiction per
Case 1.

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 66 3. Relations and functions

• The picture below —that is, f (n0 ) = m for some m.

0
1

We simply transform the picture to the one below, “correcting” f

to have f (a) = m and f (n0 ) = n0 , that is defining a new “f ” that
we will call f 0 by
 
f 0 = f − {(n0 , m), (a, n0 )} ∪ {(n0 , n0 ), (a, m)}

0
1

We get a contradiction per Case 1. 

3.3.5 Corollary. (Pigeon-Hole Principle) If m < n, then {0, . . . , m} 6∼

{0, . . . , n}.
Proof. If the conclusion fails then we have an onto f : {0, . . . , m} → {0, . . . , n},
contradicting 3.3.4. 
 Important!

3.3.6 Theorem. If A is finite due to A ∼ {0, 1, 2, . . . n} then there is no justi-

fication of finiteness via another canonical set {0, 1, 2, . . . m} with n 6= m.
Proof. If {0, 1, 2, . . . n} ∼ A ∼ {0, 1, 2, . . . m}, then {0, 1, 2, . . . n} ∼ {0, 1, 2, . . . m}
by 3.2.13, hence n = m, otherwise we contradict 3.3.5. 

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 3.3. Finite and Infinite Sets 67

3.3.7 Definition. Let A ∼ {0, . . . , n}. Since n is uniquely determined by A we

say that A has n + 1 elements and write |A| = n + 1. 


3.3.8 Corollary. There is no onto function from {0, . . . , n} to N.

 “For all n ∈ N, there is no. . . ” is, of course, implied. 

Proof. Fix an n. By way of contradiction, let g : {0, . . . , n} → N be onto. Let
Def
Y = {x ≤ n : g(x) > n + 1}

Now let
Def
X = {0, . . . , n} − Y
and
Def
g0 = g − Y × N
 The “g − Y × N” above is an easy way to say “remove all pairs from g that have
their first component in Y ”. 
Thus, g 0 : X → {0, . . . , n, n + 1} is onto, contradicting 3.3.4 because X ⊆
{0, . . . , n} ⊂ {0, . . . , n, n + 1}. 

3.3.9 Corollary. N is infinite.

Proof. By 3.3.1 the opposite case requires that there is an n and a function
f : {0, 1, 2, . . . , n} → N that is a 1-1 correspondence. Impossible, since any such
an f will fail to be onto. 
 Our mathematical definitions have led to what we hoped they would: That N
is infinite as we intuitively understand, notwithstanding Achilles’s accelerated
counting! 
N is a “canonical” infinite set that we can use to index the members of many
infinite sets. Sets that can be indexed using natural number indices

a0 , a1 , . . .

are called countable.

In the interest of technical flexibility, we do not insist that all members of
N be used as indices. We might enumerate with gaps:

b5 , b9 , b13 , b42 , . . .

Thus, informally, a set A is countable if it is empty or (in the opposite case) if

there is a way to index, hence enumerate, all its members in an array, utilizing
indices from N. Cf. 3.1.40.

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 68 3. Relations and functions

It is allowed to repeatedly list any element of A, so that finite sets are

countable. For example, the set {42}:
42 forever
42, 42, 42, z}|{
...

We may think that the enumeration above is done by assigning to “42” all of
the members of N as indices, in other words, the enumeration is effected, for
example, by the constant function f : N → {42} given by f (n) = 42 for all
n ∈ N. This is consistent with our earlier definition of indexing (3.1.40).
Now, mathematically,

3.3.10 Definition. (Countable Sets) We call a set A countable if A = ∅, or

there is an onto function f : N → A. We do NOT require f to be total. This
means that some or many indices from N need not be used in the enumeration
If f (n) ↓, then we say that f (n) is the nth element of A in the enumeration f .
We often write fn instead of f (n) and then call n a “subscript” or “index”. 
 Thus a nonempty set is countable iff it is the range of some function that has
N as its left field.
BTW, since we allow f to be non total, the hedging “nonempty” is unnec-
essary: ∅ is the range of the empty function that has N as its left field. 
We said that the f that proves countability of a set A need not be total.
But such an f can always be “completed”, by adding pairs to it, to get an f 0
such that f 0 : N → A is onto and total. Here is how:

3.3.11 Proposition. Let f : N → A 6= ∅† be onto. Then we can extend f to

f 0 so that f 0 : N → A is onto and total.
Proof. Pick an a ∈ A —possible since A 6= ∅— and keep it fixed. Now, our
sought f 0 is given for all n ∈ N by cases as below:
(
0 f (n) if f (n) ↓
f (n) =
a if f (n) ↑


Some set theorists also define sets that can be enumerated using all the
elements of N as indices without repetitions.

3.3.12 Definition. (Enumerable or denumerable sets) A set A is enumer-

able iff A ∼ N. 

 3.3.13 Example. Every enumerable set is countable, but the converse fails.
For example, {1} is countable but not enumerable due to 3.3.8. {2n : n ∈ N} is
enumerable, with f (n) = 2n effecting the 1-1 correspondence f : N → {2n : n ∈
N}.  
† Since we are constructing a total onto function to A we need to assume the case A 6= ∅

as we cannot put any outputs into ∅.

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 3.3. Finite and Infinite Sets 69

3.3.14 Theorem. If A is an infinite subset of N, then A ∼ N.

Proof. We will build a 1-1 and total enumeration of A, presented in a finite
manner as a (pseudo) program below, which enumerates all the members of A
in strict ascending order and arranges them in an array

a(0), a(1), a(2), . . . (1)

n ←0
while A 6= ∅
a(n) ← min A Comment. Inside the loop ∅ =
6 A ⊆ N, hence min exists.
A ← A − {a(n)}
n ←n+1
end while

 Note that the sequence {a(0), a(1), . . . , a(m)} is strictly increasing for any m,
since a(0) is smallest in A, a(1) is smallest in A−{a(0)} and hence the next higher than
a(0) in A, etc. 

Will this loop ever exit? Say, yes, when it starts (but does not complete)
the k-th pass through the loop. Thus A became empty when we did A ←
A − {a(k − 1)} in the previous pass, that is A = {a(0), a(1), . . . , a(k − 1)} and
thus, since
a(0) < a91) < . . . < a(k − 1)
we have that the function f : {0, . . . , k − 1} → A given by

f = {(0, a(0)), (1, a(1)), . . . (k − 1, a(k − 1))}

is total, 1-1 and onto, thus, A ∼ {0, . . . , k − 1} contradicting that A is

infinite!
Thus, we never exit the loop!
Thus, by the remark in the  paragraph above, (1) enumerates A in strict
ascending order, that is,

if we define f : N → A by f (n) = a(n), for all n

then f is 1-1 (by strict increasing property: distinct inputs cause distinct out-
puts), and is trivially total, and onto. Why the latter? Every a ∈ A is reached
in ascending order, and assigned an “n” from N. 

3.3.15 Theorem. Every infinite countable set is enumerable.

Proof. Let f : N → A be onto and total (cf. 3.3.11), where A is infinite. Let
g : A → N such that (f g) = 1A (3.2.23). Thus, if we let B = ran(g), we have
that g is onto B, and thus by 3.2.18 is also 1-1 and total. Thus it is a 1-1
correspondence g : A → B, that is,

A∼B (1)

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 70 3. Relations and functions

B must be infinite, otherwise (3.3.1), for some n, B ∼ {0, . . . , n} and by (1) via
Exercise 3.2.13 we have A ∼ {0, . . . , n}, contradicting that A is infinite. Thus,
by 3.3.14, B ∼ N, hence (again, Exercise 3.2.13 and (1)) A ∼ N. That is, A is
enumerable. 

 So, if we can enumerate an infinite set at all, then we can enumerate it without
repetitions. 

We can linearise an infinite square matrix of elements in each location (i, j)

by devising a traversal that will go through each (i, j) entry once, and will not
miss any entry!
In the literature one often sees the method diagrammatically, see below,
where arrows clearly indicate the sequence of traversing, with the understanding
that we use the arrows by pick the first unused chain of arrows from left to right.

(0, 0) (0, 1) (0, 2) (0, 3) . . .

% % %
(1, 0) (1, 1) (1, 2)
% %
(2, 0) (2, 1)
%
(3, 0)
..
.

So the linearisation induces a 1-1 correspondence between N and the linearised

sequence of matrix entries, that is, it shows that N × N ∼ N. For short,

3.3.16 Theorem. The set N × N is countable. In fact, it is enumerable.

Is there a “mathematical” way to do this? Well, the above IS mathematical,

don’t get me wrong, but is given in outline. It is kind of an argument in
geometry, where we rely on drawings (figures).
Here are the algebraic details:
Proof. (of 3.3.16 with an algebraic argument). Let us call i + j + 1 the “weight”
of a pair (i, j). The weight is the number of elements in the group:

(i + j, 0), (i + j − 1, 1), (i + j − 2, 2), . . . , (i, j), . . . , (0, i + j)

Thus the diagrammatic enumeration proceeds by enumerating groups by in-

creasing weight
1, 2, 3, 4, 5, . . .
and in each group of weight k we enumerate in ascending order of the second
component.
Thus the (i, j) th entry occupies position j in its group —the first position
in the group being the 0 th, e.g., in the group of (3, 0) the first position is the
0 th— and this position globally is the number of elements in all groups before

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 3.3. Finite and Infinite Sets 71

group i + j + 1, plus j. Thus the first available position for the first entry of
group (i, j) members is just after this many occupied positions:

(i + j)(i + j + 1)
1 + 2 + 3 + . . . (i + j) =
2
That is,
(i + j)(i + j + 1)
global position of (i, j) is this: +j
2
The function f which for all i, j is given by

(i + j)(i + j + 1)
f (i, j) = +j
2
is the algebraic form of the above enumeration. 
 There is an easier way to show that N × N ∼ N without diagrams:
By the unique factorisation of numbers into products of primes (Euclid) the
function g : N × N → N given for all m, n by g(m, n) = 2m 3n is 1-1, since Euclid
0 0
proved that 2m 3n = 2m 3n implies m = m0 and n = n0 . It is not onto as it
never outputs, say, 5, but ran(g) is an infinite subset of N (Exercise!).
Thus, trivially, N × N ∼ ran(g) ∼ N, the latter “∼” by 3.3.14. 

3.3.17 Exercise. If A and B are enumerable, so is A × B.

Hint. So, N ∼ A and N ∼ B. Can you show now that N × N ∼ A × B? 
n
With little additional effort one can generalise to the case of ×A .
i=1
i 

3.3.18 Remark.

1. Let us collect a few more remarks on countable sets here. Suppose now
that we start with a countable set A. Is every subset of A countable? Yes,
because the composition of onto functions is onto.

2. 3.3.19 Exercise. What does composition of onto functions have to do

with this? Well, if B ⊆ A then there is a natural onto function g : A →
B. Which one? Think “natural”! Get a natural total and 1-1 function
f : B → A and then use f to get g. 
3. As a special case, if A is countable, then so is A ∩ B for any B, since
A ∩ B ⊆ A.

4. How about A ∪ B? If both A and B are countable, then so is A ∪ B.

Indeed, and without inventing a new technique, let

a0 , a1 , . . .

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
72 3. Relations and functions

be an enumeration of A and

b0 , b1 , . . .

for B. Now form an infinite matrix with the A-enumeration as the 1st
row, while each remaining row is the same as the B-enumeration. Now
linearise this matrix!
Of course, we may alternatively adapt the unfolding technique to an infinite
matrix of just two rows. How?

5. 3.3.20 Exercise. Let A be enumerable and an enumeration of A

a0 , a1 , a2 , . . . (1)

is given.
So, this is an enumeration without repetitions.
Use techniques we employed in this section to propose a new enumeration
in which every ai is listed infinitely many times (this is useful in some
applications of logic). 

3.4. Diagonalisation and uncountable sets

3.4.1 Example. Suppose we have a 3 × 3 matrix

1 1 0
1 0 1
0 1 1

and we are asked: Find a sequence of three numbers, using only 0 or 1, that
does not fit as a row of the above matrix —i.e., is different from all rows.
Sure, you reply: Take 1 1 1. Or, take 0 0 0.
That is correct. But what if the matrix were big, say, 10350000 × 10350000 , or
even infinite?
Is there a finitely describable technique that can produce an “unfit” row for
any square matrix, even an infinite one? Yes, it is Cantor’s diagonal method or
technique.
He noticed that any row that fits in the matrix as the, say, i-th row, inter-
sects the main diagonal at the same spot that the i-th column does.

That is, at entry (i, i).

Thus if we take the main diagonal —a sequence that has the same length as
any row— and change every one of its entries, then it will not fit anywhere as
a row! Because no row can have an entry that is different than the entry at the
location where it intersects the main diagonal!

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 3.4. Diagonalisation and uncountable sets 73

This idea would give the answer 0 1 0 to our original question. While
1000 11 3 also follows the principle “change all the entries of the diagonal”
and works, we are constrained here to “use only 0 or 1” as entries. More
seriously, in a case of a very large or infinite matrix it is best to have a simple
technique that works even if we do not know much about the elements of the
matrix. Read on! 

3.4.2 Example. We have an infinite matrix of 0-1 entries. Can we produce an

infinite sequence of 0-1 entries that does not match any row in the matrix? Yes,
take the main diagonal and flip every entry (0 to 1; 1 to 0).
If we think that, yes, it fits as row i, then we get a contradiction:
Say the original row has an a as entry (i, i). But, by our construction, the
new row has an 1 − a in as entry (i, i), so it will not fit as row i after all. So it
fits nowhere, i being arbitrary. 

 3.4.3 Example. (Cantor) Let S denote the set of all infinite sequences of 0s
and 1s.
Pause. What is an infinite sequence? Our intuitive understanding of the
term is captured mathematically by the concept of a total function f with left
field (and hence domain) N. The n-th member of the sequence is f (n).J
Can we arrange all of S in an infinite matrix —one element per row? No,
since the preceding example shows that we would miss at least one infinite
sequence (i.e., we would fail to list it as a row), for a sequence of infinitely many
0s and/or 1s can be found, that does not match any row!
But arranging all members of S as an infinite matrix —one element per
row— is tantamount to saying that we can enumerate all the members of S
using members of N as indices.
So we cannot do that. S is not countable!  

3.4.4 Definition. (Uncountable Sets) A set that is not countable is called

uncountable. 
So, an uncountable set is neither finite, nor enumerable. The first observa-
tion makes it infinite, the second makes it “more infinite” than the set of natural
numbers since it is not in 1-1 correspondence with N (else it would be enumer-
able, hence countable) nor with a subset of N: If the latter, our uncountable
set would be finite or enumerable (which is absurd) according as it is in 1-1
correspondence with a finite subset or an infinite subset (cf. 3.3.14 and Exercise
3.2.13).
Example 3.4.3 shows that uncountable sets exist. Here is a more interesting
one.

 3.4.5 Example. (Cantor) The set of real numbers in the interval

Def
(0, 1) = {x ∈ R : 0 < x < 1}

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
74 3. Relations and functions

is uncountable. This is done via an elaboration of the argument in 3.4.3.

Think of a member of (0, 1), in form, as an infinite sequence of numbers
from the set {0, 1, 2, 3, 4, 5, 6, 7, 8, 9} prefixed with a dot; that is, think of the
number’s decimal notation.
Some numbers have representations that end in 0s after a certain point.
We call these representations finite. Every such number has also an “infinite
representation” since the non zero digit d immediately to the left of the infinite
tail of 0s can be converted to d−1, and the infinite tail into 9s, without changing
the value of the number.
Allow only infinite representations.
Assume now by way of contradiction that a listing of all members of (0, 1)
exists, listing them via their infinite representations

.a00 a01 a02 a03 a04 . . .

.a10 a11 a12 a13 a14 . . .
.a20 a21 a22 a23 a24 . . .
.a30 a31 a32 a33 a34 . . .
..
.

The argument from 3.4.3 can be easily modified to get a “row that does not
fit”, that is, a representation
.d0 d1 d2 · · ·
not in the listing.
Well, just let (
2 if aii = 0 ∨ aii = 1
di =
1 otherwise
Clearly .d0 d1 d2 · · · does not fit in any row i as it differs from the expected digit
at the i-th decimal place: should be aii , but di 6= aii . It is, on the other hand,
an infinite decimal expansion, being devoid of zeros, and thus should be listed.
This contradiction settles the issue.  

3.4.6 Example. (3.4.3 Revisited) Consider the set of all total functions from
N to {0, 1}. Is this countable?
Well, if there is an enumeration of these one-variable functions

f0 , f1 , f2 , f3 , . . . (1)

consider the function g : N → {0, 1} given by g(x) = 1 − fx (x). Clearly, this

must appear in the listing (1) since it has the correct left and right fields, and
is total.
Too bad! If g = fi then g(i) = fi (i). By definition, it is also 1 − fi (i). A
contradiction.
This is just version of 3.4.3; as already noted there, an infinite sequence of
0s and 1s is just a total function from N to {0, 1}. 

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
 3.4. Diagonalisation and uncountable sets 75

The same argument as above shows that the set of all functions from N to
itself is uncountable. Taking g(x) = fx (x)+1 also works here to “systematically
change the diagonal” f0 (0), f1 (1), . . . since we are not constrained to keep the
function values in {0, 1}.

 3.4.7 Remark. Worth Emphasizing. Here is how we constructed g: We

have a list of in principle available f -indices for g. We want to make sure that
none of them applies.
A convenient method to do that is to inspect each available index, i, and
using the diagonal method do this: Ensure that g differs from fi at input i, by
setting g(i) = 1 − fi (i).
This ensures that g 6= fi ; period. We say that we cancelled the index i as a
possible “f -index” of g.
Since the process is applied for each i, we have cancelled all possible indices
for g: For no i can we have g = fi .  

 3.4.8 Example. (Cantor) What about the set of all subsets of N —P(N) or
2N ?
Cantor showed that this is uncountable as well: If not, we have an enumer-
ation of its members as
S0 , S1 , S2 , . . . (1)
Define the set
Def
D = {x ∈ N : x ∈
/ Sx } (2)
So, D ⊆ N, thus it must appear in the list (1) as an Si . But then i ∈ D iff i ∈ Si
by virtue of D = Si . However, also i ∈ D iff i ∈/ Si by (2). This contradiction
establishes that a legitimate subset of N, namely D, is not an Si . That is, 2N
cannot be so enumerated; it is uncountable.  

 3.4.9 Example. (Characteristic functions) Let S ⊆ N. We can represent

S as an infinite 0/1 array:
array position . . . i . . . j ...
array content . . . 0 . . . 1 ...
... ↑ ... ↑ ...
means . . .i ∈
/ S. . .j ∈ S. . .
This array faithfully represents S —tells all we need to know about what S
contains— since it contains a “1” in location x iff x ∈ S; contains “0” otherwise.
The array viewed as a total function from N to {0, 1} is called the charac-
teristic function of S, denoted by cS :
(
1 if x ∈ S
cS (x) =
0 if x ∈ N − S

Note that there is a 1-1 correspondence, let’s call it F , between subsets of N and
the total 0-1-valued functions from N simply given by F (S) = cS . (Exercise!)

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.
76 3. Relations and functions

Thus
{f : f : N → {0, 1} and f is total} ∼ 2N
In particular, the concept of characteristic functions shows that Example 3.4.8
fits the diagonalization methodology. Indeed, the argument in 3.4.8 sets cD (x) =
1 − cSx (x), for all x, because

cD (x) = 1 iff x ∈ D iff x ∈

/ Sx iff cSx (x) = 0 iff 1 − cSx (x) = 1

But then, the argument in 3.4.8 essentially applies the diagonal method to
the list of 0/1 functions cSx , for x = 0, 1, 2, . . ., to show that some 0/1 function,
namely, cD cannot be in the list.  

Fragments of “safe” set theory; from the EECS 1028 lecture notes c G. Tourlakis, W 2020.