2.

Ordinal Numbers

In this chapter we introduce ordinal numbers and prove the Transfinite Re-
cursion Theorem.

Linear and Partial Ordering

Definition 2.1. A binary relation < on a set P is a partial ordering of P if:
(i) p < p for any p ∈ P ;
(ii) if p < q and q < r, then p < r.
(P, <) is called a partially ordered set. A partial ordering < of P is a linear
ordering if moreover
(iii) p < q or p = q or q < p for all p, q ∈ P .
If < is a partial (linear) ordering, then the relation ≤ (where p ≤ q if either
p < q or p = q) is also called a partial (linear) ordering (and < is sometimes
called a strict ordering).

Definition 2.2. If (P, <) is a partially ordered set, X is a nonempty subset

of P , and a ∈ P , then:
a is a maximal element of X if a ∈ X and (∀x ∈ X) a < x;
a is a minimal element of X if a ∈ X and (∀x ∈ X) x < a;
a is the greatest element of X if a ∈ X and (∀x ∈ X) x ≤ a;
a is the least element of X if a ∈ X and (∀x ∈ X) a ≤ x;
a is an upper bound of X if (∀x ∈ X) x ≤ a;
a is a lower bound of X if (∀x ∈ X) a ≤ x;
a is the supremum of X if a is the least upper bound of X;
a is the infimum of X if a is the greatest lower bound of X.
The supremum (infimum) of X (if it exists) is denoted sup X (inf X).
Note that if X is linearly ordered by <, then a maximal element of X is its
greatest element (similarly for a minimal element).
If (P, <) and (Q, <) are partially ordered sets and f : P → Q, then f is
order-preserving if x < y implies f (x) < f (y). If P and Q are linearly ordered,
then an order-preserving function is also called increasing.
18 Part I. Basic Set Theory

A one-to-one function of P onto Q is an isomorphism of P and Q if

both f and f −1 are order-preserving; (P, <) is then isomorphic to (Q, <).
An isomorphism of P onto itself is an automorphism of (P, <).

Well-Ordering
Definition 2.3. A linear ordering < of a set P is a well-ordering if every
nonempty subset of P has a least element.

The concept of well-ordering is of fundamental importance. It is shown be-

low that well-ordered sets can be compared by their lengths; ordinal numbers
will be introduced as order-types of well-ordered sets.

Lemma 2.4. If (W, <) is a well-ordered set and f : W → W is an increasing

function, then f (x) ≥ x for each x ∈ W .

Proof. Assume that the set X = {x ∈ W : f (x) < x} is nonempty and let z
be the least element of X. If w = f (z), then f (w) < w, a contradiction. 


Corollary 2.5. The only automorphism of a well-ordered set is the identity.

Proof. By Lemma 2.4, f (x) ≥ x for all x, and f −1 (x) ≥ x for all x. 


Corollary 2.6. If two well-ordered sets W1 , W2 are isomorphic, then the

isomorphism of W1 onto W2 is unique. 


If W is a well-ordered set and u ∈ W , then {x ∈ W : x < u} is an initial

segment of W (given by u).

Lemma 2.7. No well-ordered set is isomorphic to an initial segment of itself.

Proof. If ran(f ) = {x : x < u}, then f (u) < u, contrary to Lemma 2.4. 


Theorem 2.8. If W1 and W2 are well-ordered sets, then exactly one of the
following three cases holds:
(i) W1 is isomorphic to W2 ;
(ii) W1 is isomorphic to an initial segment of W2 ;
(iii) W2 is isomorphic to an initial segment of W1 .

Proof. For u ∈ Wi , (i = 1, 2), let Wi (u) denote the initial segment of Wi

given by u. Let

f = {(x, y) ∈ W1 × W2 : W1 (x) is isomorphic to W2 (y)}.

Using Lemma 2.7, it is easy to see that f is a one-to-one function. If h is

an isomorphism between W1 (x) and W2 (y), and x
< x, then W1 (x
) and
W2 (h(x
)) are isomorphic. It follows that f is order-preserving.
 2. Ordinal Numbers 19

If dom(f ) = W1 and ran(f ) = W2 , then case (i) holds.

If y1 < y2 and y2 ∈ ran(f ), then y1 ∈ ran(f ). Thus if ran(f ) = W2 and
y0 is the least element of W2 − ran(f ), we have ran(f ) = W2 (y0 ). Necessarily,
dom(f ) = W1 , for otherwise we would have (x0 , y0 ) ∈ f , where x0 = the least
element of W1 − dom(f ). Thus case (ii) holds.
Similarly, if dom(f ) = W1 , then case (iii) holds.
In view of Lemma 2.7, the three cases are mutually exclusive. 


If W1 and W2 are isomorphic, we say that they have the same order-type.
Informally, an ordinal number is the order-type of a well-ordered set.
We shall now give a formal definition of ordinal numbers.

Ordinal Numbers

The idea is to define ordinal numbers so that

α<β if and only if α ∈ β, and α = {β : β < α}.

Definition 2.9.
A set T is transitive if every element of T is a subset of T .

(Equivalently, T ⊂ T , or T ⊂ P (T ).)

Definition 2.10. A set is an ordinal number (an ordinal ) if it is transitive

and well-ordered by ∈.

We shall denote ordinals by lowercase Greek letters α, β, γ, . . . . The class

of all ordinals is denoted by Ord.
We define
α < β if and only if α ∈ β.

Lemma 2.11.

(i) 0 = ∅ is an ordinal.
(ii) If α is an ordinal and β ∈ α, then β is an ordinal.
(iii) If α = β are ordinals and α ⊂ β, then α ∈ β.
(iv) If α, β are ordinals, then either α ⊂ β or β ⊂ α.

Proof. (i), (ii) by definition.

(iii) If α ⊂ β, let γ be the least element of the set β − α. Since α is
transitive, it follows that α is the initial segment of β given by γ. Thus
α = {ξ ∈ β : ξ < γ} = γ, and so α ∈ β.
(iv) Clearly, α ∩ β is an ordinal, α ∩ β = γ. We have γ = α or γ = β,
for otherwise γ ∈ α, and γ ∈ β, by (iii). Then γ ∈ γ, which contradicts the
definition of an ordinal (namely that ∈ is a strict ordering of α). 

20 Part I. Basic Set Theory

Using Lemma 2.11 one gets the following facts about ordinal numbers
(the proofs are routine):

(2.1) < is a linear ordering of the class Ord.

(2.2) For each α, α = {β : β < α}.
 
is a nonempty class of ordinals, then C is an ordinal, C ∈ C
(2.3) If C 
and C = inf C.



(2.4) If X is a nonempty set of ordinals, then X is an ordinal, and X =
sup X.
(2.5) For every α, α ∪ {α} is an ordinal and α ∪ {α} = inf{β : β > α}.

We thus define α + 1 = α ∪ {α} (the successor of α). In view of (2.4), the

class Ord is a proper class; otherwise, consider sup Ord + 1.
We can now prove that the above definition of ordinals provides us with
order-types of well-ordered sets.

Theorem 2.12. Every well-ordered set is isomorphic to a unique ordinal

number.

Proof. The uniqueness follows from Lemma 2.7. Given a well-ordered set W ,
we find an isomorphic ordinal as follows: Define F (x) = α if α is isomorphic
to the initial segment of W given by x. If such an α exists, then it is unique.
By the Replacement Axioms, F (W ) is a set. For each x ∈ W , such an α
exists (otherwise consider the least x for which such an α does not exist). If
γ is the least γ ∈/ F (W ), then F (W ) = γ and we have an isomorphism of W
onto γ. 


If α = β + 1, then α is a

successor ordinal. If α is not a successor ordinal,
then α = sup{β : β < α} = α; α is called a limit ordinal. We also consider 0
a limit ordinal and define sup ∅ = 0.
The existence of limit ordinals other than 0 follows from the Axiom of
Infinity; see Exercise 2.3.

Definition 2.13 (Natural Numbers). We denote the least nonzero limit

ordinal ω (or N ). The ordinals less than ω (elements of N ) are called finite
ordinals, or natural numbers. Specifically,

0 = ∅, 1 = 0 + 1, 2 = 1 + 1, 3 = 2 + 1, etc.

A set X is finite if there is a one-to-one mapping of X onto some n ∈ N .

X is infinite if it is not finite.

We use letters n, m, l, k, j, i (most of the time) to denote natural numbers.

 2. Ordinal Numbers 21

Induction and Recursion

Theorem 2.14 (Transfinite Induction). Let C be a class of ordinals and

assume that :
(i) 0 ∈ C;
(ii) if α ∈ C, then α + 1 ∈ C;
(iii) if α is a nonzero limit ordinal and β ∈ C for all β < α, then α ∈ C.
Then C is the class of all ordinals.

Proof. Otherwise, let α be the least ordinal α ∈

/ C and apply (i), (ii), or (iii).



A function whose domain is the set N is called an (infinite) sequence

(A sequence in X is a function f : N → X.) The standard notation for
a sequence is
an : n < ω
or variants thereof. A finite sequence is a function s such dom(s) = {i : i < n}
for some n ∈ N ; then s is a sequence of length n.
A transfinite sequence is a function whose domain is an ordinal:

aξ : ξ < α.

It is also called an α-sequence or a sequence of length α. We also say that

a sequence aξ : ξ < α is an enumeration of its range {aξ : ξ < α}. If
s is a sequence of length α, then s x or simply sx denotes the sequence of
length α + 1 that extends s and whose αth term is x:

s x = sx = s ∪ {(α, x)}.

Sometimes we shall call a “sequence”

aα : α ∈ Ord 

a function (a proper class) on Ord .

“Definition by transfinite recursion” usually takes the following form:
Given a function G (on the class of transfinite sequences), then for every θ
there exists a unique θ-sequence

aα : α < θ

such that
aα = G(aξ : ξ < α)
for every α < θ.
We shall give a general version of this theorem, so that we can also con-
struct sequences aα : α ∈ Ord .
22 Part I. Basic Set Theory

Theorem 2.15 (Transfinite Recursion). Let G be a function (on V ),

then (2.6) below defines a unique function F on Ord such that

F (α) = G(F α)

for each α.

In other words, if we let aα = F (α), then for each α,

aα = G(aξ : ξ < α).

(Note that we tacitly use Replacement: F α is a set for each α.)

Corollary 2.16. Let X be a set and θ an ordinal number. For every func-
tion G on the set of all transfinite sequences in X of length < θ such that
ran(G) ⊂ X there exists a unique θ-sequence aα : α < θ in X such that
aα = G(aξ : ξ < α) for every α < θ. 


Proof. Let

(2.6) F (α) = x ↔ there is a sequence aξ : ξ < α such that:

(i) (∀ξ < α) aξ = G(aη : η < ξ);
(ii) x = G(aξ : ξ < α).

For every α, if there is an α-sequence that satisfies (i), then such a sequence is
unique: If aξ : ξ < α and bξ : ξ < α are two α-sequences satisfying (i), one
shows aξ = bξ by induction on ξ. Thus F (α) is determined uniquely by (ii),
and therefore F is a function. It follows, again by induction, that for each α
there is an α-sequence that satisfies (i) (at limit steps, we use Replacement
to get the α-sequence as the union of all the ξ-sequences, ξ < α). Thus F is
defined for all α ∈ Ord. It obviously satisfies

F (α) = G(F α).

If F
is any function on Ord that satisfies

F
(α) = G(F
α)

then it follows by induction that F
(α) = F (α) for all α. 



Definition 2.17. Let α > 0 be a limit ordinal and let γξ : ξ < α be
a nondecreasing sequence of ordinals (i.e., ξ < η implies γξ ≤ γη ). We define
the limit of the sequence by

limξ→α γξ = sup{γξ : ξ < α}.

A sequence of ordinals γα : α ∈ Ord  is normal if it is increasing and

continuous, i.e., for every limit α, γα = limξ→α γξ .
 2. Ordinal Numbers 23

Ordinal Arithmetic

We shall now define addition, multiplication and exponentiation of ordinal

numbers, using Transfinite Recursion.

Definition 2.18 (Addition). For all ordinal numbers α

(i) α + 0 = α,
(ii) α + (β + 1) = (α + β) + 1, for all β,
(iii) α + β = limξ→β (α + ξ) for all limit β > 0.

Definition 2.19 (Multiplication). For all ordinal numbers α

(i) α · 0 = 0,
(ii) α · (β + 1) = α · β + α for all β,
(iii) α · β = limξ→β α · ξ for all limit β > 0.

Definition 2.20 (Exponentiation). For all ordinal numbers α

(i) α0 = 1,
(ii) αβ+1 = αβ · α for all β,
(iii) αβ = limξ→β αξ for all limit β > 0.

As defined, the operations α + β, α · β and αβ are normal functions in the

second variable β. Their properties can be proved by transfinite induction.
For instance, + and · are associative:

Lemma 2.21. For all ordinals α, β and γ,

(i) α + (β + γ) = (α + β) + γ,
(ii) α · (β · γ) = (α · β) · γ.

Proof. By induction on γ. 


Neither + nor · are commutative:

1 + ω = ω = ω + 1, 2 · ω = ω = ω · 2 = ω + ω.

Ordinal sums and products can be also defined geometrically, as can sums
and products of arbitrary linear orders:

Definition 2.22. Let (A, <A ) and (B, <B ) be disjoint linearly ordered sets.
The sum of these linear orders is the set A ∪ B with the ordering defined as
follows: x < y if and only if
(i) x, y ∈ A and x <A y, or
(ii) x, y ∈ B and x <B y, or
(iii) x ∈ A and y ∈ B.
24 Part I. Basic Set Theory

Definition 2.23. Let (A, <) and (B, <) be linearly ordered sets. The product
of these linear orders is the set A × B with the ordering defined by

(a1 , b1 ) < (a2 , b2 ) if and only if either b1 < b2 or (b1 = b2 and a1 < a2 ).

Lemma 2.24. For all ordinals α and β, α + β and α · β are, respectively,

isomorphic to the sum and to the product of α and β.

Proof. By induction on β. 


Ordinal sums and products have some properties of ordinary addition and
multiplication of integers. For instance:

Lemma 2.25.

(i) If β < γ then α + β < α + γ.

(ii) If α < β then there exists a unique δ such that α + δ = β.
(iii) If β < γ and α > 0, then α · β < α · γ.
(iv) If α > 0 and γ is arbitrary, then there exist a unique β and a unique
ρ < α such that γ = α · β + ρ.
(v) If β < γ and α > 1, then αβ < αγ .

Proof. (i), (iii) and (v) are proved by induction on γ.

(ii) Let δ be the order-type of the set {ξ : α ≤ ξ < β}; δ is unique by (i).
(iv) Let β be the greatest ordinal such that α · β ≤ γ. 


For more, see Exercises 2.10 and 2.11.

Theorem 2.26 (Cantor’s Normal Form Theorem). Every ordinal α >

0 can be represented uniquely in the form

α = ω β1 · k1 + . . . + ω βn · kn ,

where n ≥ 1, α ≥ β1 > . . . > βn , and k1 , . . . , kn are nonzero natural

numbers.

Proof. By induction on α. For α = 1 we have 1 = ω 0 · 1; for arbitrary α > 0

let β be the greatest ordinal such that ω β ≤ α. By Lemma 2.25(iv) there
exists a unique δ and a unique ρ < ω β such that α = ω β · δ + ρ; this δ
must necessarily be finite. The uniqueness of the normal form is proved by
induction. 


In the normal form it is possible to have α = ω α ; see Exercise 2.12. The

least ordinal with this property is called ε0 .
 2. Ordinal Numbers 25

Well-Founded Relations

Now we shall define an important generalization of well-ordered sets.

A binary relation E on a set P is well-founded if every nonempty X ⊂ P
has an E-minimal element, that is a ∈ X such that there is no x ∈ X with
x E a.
Clearly, a well-ordering of P is a well-founded relation.
Given a well-founded relation E on a set P , we can define the height of E,
and assign to each x ∈ P an ordinal number, the rank of x in E.

Theorem 2.27. If E is a well-founded relation on P , then there exists

a unique function ρ from P into the ordinals such that for all x ∈ P ,

(2.7) ρ(x) = sup{ρ(y) + 1 : y E x}.

The range of ρ is an initial segment of the ordinals, thus an ordinal num-

ber. This ordinal is called the height of E.

Proof. We shall define a function ρ satisfying (2.7) and then prove its unique-
ness. By induction, let

P0 = ∅, Pα+1 = {x ∈ P : ∀y (y E x → y ∈ Pα )},


Pα = Pξ if α is a limit ordinal.
ξ<α

Let θ be the least ordinal such that Pθ+1 = Pθ (such θ exists by Replacement).
First, it should be easy to see that Pα ⊂ Pα+1 for each α (by induction).
Thus P0 ⊂ P1 ⊂ . . . ⊂ Pθ . We claim that Pθ = P . Otherwise, let a be
an E-minimal element of P − Pθ . It follows that each x E a is in Pθ , and
so a ∈ Pθ+1 , a contradiction. Now we define ρ(x) as the least α such that
x ∈ Pα+1 . It is obvious that if x E y, then ρ(x) < ρ(y), and (2.7) is easily
verified. The ordinal θ is the height of E.
The uniqueness of ρ is established as follows: Let ρ
be another function
satisfying (2.7) and consider an E-minimal element of the set {x ∈ P : ρ(x) =
ρ
(x)}. 


Exercises
2.1. The relation “(P, <) is isomorphic to (Q, <)” is an equivalence relation (on
the class of all partially ordered sets).

2.2. α is a limit ordinal if and only if β < α implies β + 1 < α, for every β.
T
2.3. If a set X is inductive, then X ∩ Ord is inductive. The set N = {X : X is
inductive} is the least limit ordinal = 0.
26 Part I. Basic Set Theory

2.4. (Without the Axiom of Infinity). Let ω = least limit α = 0 if it exists, ω = Ord
otherwise. Prove that the following statements are equivalent:
(i) There exists an inductive set.
(ii) There exists an infinite set.
(iii) ω is a set.
[For (ii) → (iii), apply Replacement to the set of all finite subsets of X.]
2.5. If W is a well-ordered set, then there exists no sequence an : n ∈ N  in W
such that a0 > a1 > a2 > . . ..
2.6. There are arbitrarily large limit ordinals; i.e., ∀α ∃β > α (β is a limit).
[Consider limn→ω αn , where αn+1 = αn + 1.]
2.7. Every normal sequence γα : α ∈ Ord  has arbitrarily large fixed points, i.e.,
α such that γα = α.
[Let αn+1 = γαn , and α = limn→ω αn .]
2.8. For all α, β and γ,
(i) α · (β + γ) = α · β + α · γ,
(ii) αβ+γ = αβ · αγ ,
(iii) (αβ )γ = αβ·γ .
2.9. (i) Show that (ω + 1) · 2 = ω · 2 + 1 · 2.
(ii) Show that (ω · 2)2 = ω 2 · 22 .
2.10. If α < β then α + γ ≤ β + γ, α · γ ≤ β · γ, and αγ ≤ β γ ,
2.11. Find α, β, γ such that
(i) α < β and α + γ = β + γ,
(ii) α < β and α · γ = β · γ,
(iii) α < β and αγ = β γ .
2.12. Let ε0 = limn→ω αn where α0 = ω and αn+1 = ω αn for all n. Show that
ε0 is the least ordinal ε such that ω ε = ε.
A limit ordinal γ > 0 is called indecomposable if there exist no α < γ and β < γ
such that α + β = γ.
2.13. A limit ordinal γ > 0 is indecomposable if and only if α + γ = γ for all α < γ
if and only if γ = ω α for some α.
2.14. If E is a well-founded relation on P , then there is no sequence an : n ∈ N 
in P such that a1 E a0 , a2 E a1 , a3 E a2 , . . . .
2.15 (Well-Founded Recursion). Let E be a well-founded relation on a set P ,
and let G be a function. Then there exists a function F such that for all x ∈ P ,
F (x) = G(x, F {y ∈ P : y E x}).

Historical Notes
The theory of well-ordered sets was developed by Cantor, who also introduced
transfinite induction. The idea of identifying an ordinal number with the set of
smaller ordinals is due to Zermelo and von Neumann.