Lecture Notes:

Axiomatic Set Theory

Asaf Karagila

Last Update:
September 17, 2023
Contents

1 Introduction 3
1.1 Why do we need axioms? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Classes and sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 The axioms of set theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Ordinals, recursion and induction 7

2.1 Ordinals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Transfinite induction and recursion . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Transitive classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 The relative consistency of the Axiom of Foundation 12

4 Cardinals and their arithmetic 15

4.1 The definition of cardinals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 The Aleph numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3 Finiteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Absoluteness and reflection 21

5.1 Absoluteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.2 Reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 The Axiom of Choice 25

6.1 The Axiom of Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.2 Weak version of the Axiom of Choice . . . . . . . . . . . . . . . . . . . . . . . . . 27

7 Sets of Ordinals 31
7.1 Cofinality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7.2 Some cardinal arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7.3 Clubs and stationary sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
7.4 The Club filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

8 Inner models of ZF 37
8.1 Inner models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
8.2 Gödel’s constructible universe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

1
 8.3 The properties of L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.4 Ordinal definable sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

9 Some combinatorics on ω1 43
9.1 Aronszajn trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.2 Diamond and Suslin trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

10 Coda: Games and determinacy 46

2
Chapter 1

Introduction

1.1 Why do we need axioms?

In modern mathematics, axioms are given to define an object. The axioms of a group define
the notion of a group, the axioms of a Banach space define what it means for something to be a
Banach space. In these examples, however, we are not interested in the objects residing within
the objects we define, but rather the properties of the objects we defined.
Sets formalize the notion of a collection of mathematical objects. This is a primitive and
foundational notion, whose roots lie in the roots of counting. We cannot count a collection of
objects, without first defining this collection.
Set theory was developed in the late 19th century, and along with logic, it was proposed as
a way to formalize mathematics into a uniform language. This meant that we are interested
both in the structure in which sets exist, as well as the properties of the sets that exist there.
One of the naive properties expected of sets in the late 19th century was that every well-defined
collection is in fact a set.

Theorem 1.1 (Russell’s paradox). Not every well-defined collection defines a set.
Proof. Let X be the collection of {A | A is a set and A ∈
/ A}. If X is a set, then either X ∈ X,
in which case the defining property requires that X ∈ / X; and if X ∈ / X, then the defining
property of X requires that X ∈ X. In either case we arrive to the conclusion that X ∈ X and
X∈ / X. This is a contradiction, so X cannot be a set.

Zermelo first gave an axiomatization for what properties sets should have. These axioms were
later developed by Fraenkel, Skolem and cemented by von Neumann in his Ph.D. dissertation.
Why does the choice of axioms even affect us? Well, proofs do not live in vacuum. After the
foundational crises began to resolve themselves, people became increasingly more aware to the
importance of formal definitions and the necessity of axioms. Cantor famously tried to prove
(and on occasion disprove) the continuum hypothesis. Later Gödel and Cohen proved that if set
theory is consistent, then both the continuum hypothesis and its negation are consistent with
set theory. Later in the course we will see how Gödel proved the consistency of the continuum
hypothesis. It turns out that the choice of axioms could have implications on problems in
mathematics outside of set theory. Here are two examples for such questions.

Question (Sierpiński sets). Recall that A ⊆ R is a null set if for every ε > 0 there is a sequence
of intervals (an , bn ) for n < ω such that A ⊆ n (an , bn ) and n (bn − an ) < ε.
S P

Is there an uncountable X ⊆ R, such that for every null set A, A ∩ X is countable?

3
Question (Productivity of ccc spaces). We say that a topological space X satisfies the countable
chain condition, or that X is ccc, if every family of pairwise disjoint open sets is countable.
Let X and Y be compact Hausdorff spaces which are ccc. Is X × Y ccc as well?

1.2 Classes and sets

The objects of the universe of set theory are called sets, so when we say that something exist
we mean to say that it is a set. But the universe of set theory is sometimes just a set inside
a larger universe. The language of set theory has only one extralogical symbol,1 the binary
relation symbol ∈.
If M is a set and E is a binary relation on M , we could ask whether or not the structure
⟨M, E⟩ satisfies the axioms of set theory, as these are just first-order axioms. We can talk about
subsets of M , as collections of “M -sets”. We say that A ⊆ M is a class of M if there is some
formula in the language of set theory φ(x, p1 , . . . , pn ) and there are parameters p1 , . . . , pn ∈ M
such that A = {x ∈ M | ⟨M, E⟩ |= φ(x, p1 , . . . , pn )}.2 Some classes correspond to a set of M ,
for example, {x ∈ M | x ̸= x} corresponds to the empty set of M ; but some classes do not
correspond to sets, as we saw with Russell’s paradox. These are called proper classes. If A is a
class which corresponds to a set, we will “confuse” A with the set it defines.
Exercise 1.1. Show that every set is a class.
We can conservatively extend the language by adding symbols for various definable objects,
predicates and functions, and that will ease our writing. For example a ⊆ b is a shorthand for
∀x(x ∈ a → x ∈ b). We will also have a few logical abbreviations.

• If φ(u) is a formula, (∃u ∈ x)φ(u) is a shorthand for ∃u(u ∈ x ∧ φ(u); (∀u ∈ x)φ(u) is a
shorthand for ∀u(u ∈ x → φ(u)).
• If φ(u) is a formula, ∃!uφ(u) is a shorthand for ∃u(φ(u) ∧ ∀x(φ(x) → x = u)).
• We will denote by a ⊆ b the formula ∀x(x ∈ a → x ∈ b).
• We will denote by ∅ the class {x | x ̸= x}.
• We will write {a1 , . . . , an } for the class {x | x = a1 ∨ . . . x = an }.
• We will denote by P (a) the class {u | u ⊆ a}.
• We will denote by a the class {u | (∃b ∈ a)u ∈ b}.
S

• We will write a ∪ b for the class {a, b}.

S

• We will denote by a the class {u | (∀b ∈ a)u ∈ b}.

T

• We will write a ∩ b for the class {a, b}.

T

We will use these symbols, and define others as we go along, as freely as we would like,
understanding that we can always translate every statement into a statement involving only
∈. We will often use the above abbreviations when a is a class, which we will understand as a
different class, definable from the definition of a. For example, if a is the class defined by φ(u),
then P (a) is the class {x | (∀y ∈ x)φ(y)}.
1
We take = to be a symbol of the underlying logic, although we can do without this.
2
There are places where every subset of M is a class, but we will only use the term class to denote a definable
collection.

4
1.3 The axioms of set theory
These are the axioms of the set theory commonly called the Zermelo–Fraenkel axioms, and
denoted by ZF. Some of these might not make a lot of sense right now, and we will have to
justify them in one way or another.

Extensionality: Two set are equal if and only if they have the same elements,

∀x∀y(x = y ↔ x ⊆ y ∧ y ⊆ x).

Empty set: The empty set exists,

∃x∀y(y ∈
/ x).

Union: Every set has a union set,

∀x∃y(y =
[
x)).

Power: Every set has a power set,

∀x∃y(y = P (x)).

Foundation: The ∈ relation is well-founded,

∀x(x ̸= ∅ → (∃y ∈ x)(x ∩ y = ∅)).

Infinity: There is an infinite set,

∃x(∅ ∈ x ∧ (∀z ∈ x)(z ∪ {z} ∈ x)).

Separation: If φ(u, p1 , . . . , pn ) is a formula in the language of set theory, then for every choice
of parameters p1 , . . . , pn and every set x there is a subset y composed of those elements
satisfying φ,
∀p1 , . . . , pn ∀x∃y∀u(u ∈ y ↔ u ∈ x ∧ φ(u, p1 , . . . , pn )).

Replacement: If φ(u, v, p1 , . . . , pn ) is a formula in the language of set theory, for every choice
of parameters p1 , . . . , pn , if for some x we can prove that for u ∈ x there exists exactly one
v such that φ(u, v, p1 , . . . , pn ) holds, namely that φ defines a function on x, then there is
y which is the range of this function,

∀p1 , . . . , pn ∀x((∀u ∈ x)(∃!vφ(u, v, p1 , . . . , pn ))

→ ∃y∀v(v ∈ y ↔ (∃u ∈ x)φ(u, v, p1 , . . . , pn ))).

The first four axioms are self-explanatory. We will gracefully ignore the axiom of foundation
for the time being, and return to it later. The axiom of infinity postulates the existence of a
set with a certain property. We will later see that under a reasonable definition of “finite”, any
such set cannot be finite. The axioms of Separation and Replacement are in fact schemata, and
not a single axiom. As we are working with first-order logic, we cannot quantify over arbitrary
collection of objects; we can only quantify over objects. These schemata tell us that for each
formula we add an axiom which has a particular syntactic structure which we can identify.
The axiom schema of Separation tells us that while not every well-defined collection defines
a set, it is certainly the case that every definable subcollection of an existing set defines a set
on its own. The axiom schema of Replacement tells us that if we can define a function, then

5
the range of that function applied to any set x is also a set. Namely, if we can define a rule for
replacing the elements of a setx, then the collection of “replaced elements” is a set as well.
Sometimes we will be interested in theories that we obtain by removing some of the axioms
from ZF. For example, we will prove that if ZF without Foundation is consistent, then ZF is
consistent as well. Let us name some of the subtheories of ZF:

• Z is ZF without Replacement.

• ZF− is ZF without Power Set.

• ZF0 is ZF without Foundation.

Theorem 1.2. For every x, x ∈ / x.

Proof. If x is a set, then by Exercise 1.3 {x, x} = {x} is also a set. It follows that x ∩ {x} =
̸ ∅,
in contradiction to Foundation.

Exercise 1.2. Show that the axioms Empty Set and Separation are redundant.

Exercise 1.3. Show that if x, y are sets, then {x, y} is a set.

Exercise 1.4 (*). For every x and y, if there is a sequence x0 = x, xn = y and xk+1 ∈ xk for
0 ≤ k < n, then x ∈
/ y.

Exercise 1.5. Show that if x ̸= ∅, then

T
x is a set.

Exercise 1.6. For every x, P (x) ⊈ x.

Exercise 1.7 (*) (Hilbert’s paradox). Show that there is no set S with the properties: (a) If
x ∈ S, then P (x) ∈ S; and (b) if T ⊆ S, then T ∈ S. (Hint: Consider S, and reduce the two
S S

properties to obtain a contradiction to the previous exercise.)

Exercise 1.8. Recall that we can encode the ordered pair ⟨a, b⟩ as {{a}, {a, b}}. Show that if a
and b are sets, then ⟨a, b⟩ is a set, and that a × b = {⟨u, v⟩ | u ∈ a ∧ v ∈ b} is also a set.

Exercise 1.9. Write an explicit formula in the language of set theory stating that y is a linear
ordering of the set x.

Exercise 1.10. Write a formula φ(x) in the language of set theory stating that x is an injective
function.

Exercise 1.11 (**). We say that φ(x, y, z, p) satisfies the ordered pair property for a parameter p,
if we can prove without Replacement ∀x∀y∃!zφ(x, y, z, p). Let a ×φ b denote the Cartesian product
defined with φ as defining ordered pairs, namely {z | (∃x ∈ a)(∃y ∈ b)φ(x, y, z, p)}.
Suppose that for every φ satisfying the ordered pair property, a ×φ b exists. Prove that Replace-
ment holds.

Exercise 1.12 (**). Suppose that Replacement holds only for parameter-free formulas. Prove that
the full schema of Replacement holds.

6
Chapter 2

Ordinals, recursion and induction

Definition 2.1. A class A is transitive if for every b ∈ A, b ⊆ A.

Exercise 2.1. If A is a class such that for every a ∈ A, a is transitive, then

S T
A and A are also
transitive classes.

Exercise 2.2. If A is transitive set, then P (A) is transitive, A ∪ {A} is also a transitive set.

Definition 2.2. Let A be a class, and let R be a relation on A.

1. We say that R is a well-founded relation if for every b ⊆ A, if b is non-empty, then there

is some x ∈ b such that for all y ∈ b, ⟨y, x⟩ ∈
/ R. In order words, every non-empty subset
of A has an R-minimal element.

2. We say that R is extensional if for every a, b ∈ A, a = b if and only if ∀x(⟨x, a⟩ ∈ R ↔

⟨x, b⟩ ∈ R).

3. We say that R is set-like if for every a ∈ A, the class {b ∈ A | ⟨b, a⟩ ∈ R} is a set.

The Axiom of Foundation states that ∈ is a well-founded relation; the Axiom of Extension-
ality states that ∈ is in fact an extensional relation. Note that if A is a set, then every relation
on A is indeed set-like.
For the remainder of the course, a partial order will be an irreflexive and transitive relation.
A partial order < of a class A is total (or linear) if for every a, b ∈ A one of the mutually
exclusive statements hold: a = b or a < b or b < a. Finally, a well-order is a well-founded linear
order.

Definition 2.3. If A and B are partially ordered classes, an embedding is a function F : A → B,

such that F (a) <B F (a′ ) if and only if a <A a′ . A surjective embedding is called an isomorphism.

Remark. Note that the definition makes sense for an arbitrary relation, and not necessarily a partial
order.

Exercise 2.3. Show that an embedding of partial orders is injective.

Exercise 2.4. Given two well-ordered sets (X, <X ) and (Y, <Y ), we can either embed X into an
initial segment of Y or embed Y into an initial segment of X. Moreover such embedding is unique.

7
2.1 Ordinals
We say that a set x is an ordinal if it is a transitive set such that ∈ is a well-founded and linear
ordering of x. We will always use Greek letters to denote ordinals, with the exception of finite
ordinals which we will denote by Latin letters such as k, n, m and so on.
Proposition 2.4. α is an ordinal if and only if α is a transitive set linearly ordered by ∈.
Proof. If α is an ordinal then by definition it is a transitive set which is well-ordered by ∈, and
in particular it is linearly ordered. If α is a transitive set which is linearly ordered by ∈, then
by the Axiom of Foundation, ∈ is well-founded and therefore ∈ is a well-order of α.

Proposition 2.5. α is an ordinal if and only if it is a transitive set of transitive sets.

Proof. Suppose that α is an ordinal, let x ∈ α, y ∈ x and z ∈ y. By transitivity of α we have
that y ∈ α and therefore z ∈ α. Since ∈ is a well-ordering of α it is a transitive relation on α,
so z ∈ x. Therefore x is transitive.
In the other direction, suppose that α is a transitive set of transitive sets, then if ∈ is not
a well-order of α, by Foundation it means that ∈ is not a linear order of α, namely there are
x, y ∈ α such that x ∈/ y and y ∈/ x. Let A be the set {x ∈ α | (∃y ∈ α)y ̸= x ∧ x ∈ / y∧y ∈ / x},
then by Foundation there is some x ∈ A such that x ∩ A = ∅, fix such x. Then A(x) = {y ∈
α | x ̸= y ∧ x ∈
/ y∧y ∈ / x} is non-empty, and by Foundation there is some y ∈ A(x) such that
y ∩ A(x) = ∅. Therefore every z ∈ y satisfies either z ∈ x or x ∈ z. By transitivity of y, if
z ∈ y and x ∈ z we get that x ∈ y, so y ∈ / A(x); and therefore z ∈ x for every z ∈ y. By the
∈-minimality of x in A, if z ∈ x it follows that for every y ∈ α either z = y or z ∈ y or y ∈ z;
but by transitivity of x if y ∈ z, then y ∈ x which is impossible, so z ∈ y. Therefore for every
z ∈ x we get that z ∈ y which means that we proved that y ⊆ x and x ⊆ y, so x = y.

Remark. Both these proofs rely heavily on the Axiom of Foundation, and for a good reason. It is
consistent, for example, that in the absence of Foundation there exist x such that x = {x}, in which
case x is a transitive set of transitive sets; or that there exists an infinite sequence xn such that
xn = {xk | k > n}, in which case x0 is a transitive set linearly ordered by ∈, but it is not an ordinal.

We shall denote the class of the ordinals by Ord. If α and β are ordinals, we will write α < β
to denote that α ∈ β, and α ≤ β to denote that α ∈ β or α = β which translates to α ⊆ β.
S S
Exercise 2.5. If A is a set of ordinals, then A is an ordinal. Moreover, A is the least ordinal α
such that for all β ∈ A, β ≤ α. In other words, A = sup A.
S

A is an ordinal α and for every β ∈ A,

T
Exercise 2.6. Let A be a non-empty class of ordinals, then
α ≤ β. In other words, A = min A.
T

Exercise 2.7. The class of ordinals is transitive and well-ordered by ∈. Therefore the class of all the
ordinals is a proper class.
Definition 2.6. If α is an ordinal, we say that α is a successor ordinal if there exists β ∈ Ord
such that α = β ∪ {β}, and we will write α = S(β). If α is a non-empty ordinal which is not
a successor we say that α is a limit ordinal. We shall write Lim to denote the class of limit
ordinals.

Let ω denote the class {α ∈ Ord | α ∈

/ Lim ∧ ∀β < α, β ∈/ Lim}. We will denote by 0 the
ordinal ∅ and for every natural number n, we will identify the nth successor of 0 as n. So
1 = S(0) and so on.
Exercise 2.8. Prove in ZF without Infinity that the Axiom of Infinity holds if and only if ω is a set.

8
2.2 Transfinite induction and recursion
Theorem 2.7 (Transfinite Induction). Suppose that A is a class of ordinals such that when-
ever β ⊆ A, β ∈ A. Then A = Ord.
Proof. Note that ∅ ∈ A, since ∅ ⊆ A. Let β be an ordinal, by Separation β \ A is a subset of
β. It β \ A is empty, then β ⊆ A and therefore β ∈ A. Otherwise, there is a least γ ∈ β \ A. By
virtue of being minimal, if ξ ∈ γ, then ξ ∈ A. Therefore γ ⊆ A, so γ ∈ A. It follows that β \ A
is indeed the empty set, so β ∈ A for all β ∈ Ord.

Theorem 2.8 (Transfinite Recursion). Suppose that G is a class function defined on all sets,
then there is a unique class function F with domain Ord such that for every α, F (α) = G(F ↾α).
Proof. We say that f is an α-approximation if f is defined on α, and for all ξ < α, f (ξ) = G(f ↾ξ).
By induction, if f is an α-approximation, then f is unique; and again by induction we can prove
that for every α, there exists an α-approximation. Therefore define F (α) = x if and only if f is
the unique α + 1-approximation and x = f (α).
The uniqueness of F is proved similarly by induction: if F ′ is another function with the
same property, then {α ∈ Ord | F (α) = F ′ (α)} = Ord by transfinite induction.

Exercise 2.9 (**). Prove in Z that Replacement is equivalent to the Transfinite Recursion theorem.

Definition 2.9 (Ordinal arithmetic). Let α and β be ordinals. We define by recursion the
following operations.

Addition:

1. α + 0 = α.
2. α + S(β) = S(α + β).
3. α + β = sup{α + γ | γ < β} for β ∈ Lim.

Multiplication:

1. α · 0 = α.
2. α · S(β) = α · β + α.
3. α · β = sup{α · γ | γ < β} for β ∈ Lim.

Exponentiation:

1. α0 = 1.
2. αS(β) = αβ · α.
3. αβ = sup{αγ | γ < β} for β ∈ Lim.

Exercise 2.10. Show that ordinal addition and multiplication are associative, and that α · (β + γ) =
α · β + α · γ.

Exercise 2.11. Show that if α, β and γ are ordinals, then αβ · αγ = αβ+γ and (αβ )γ = αβ·γ .

9
2.3 Transitive classes
Definition 2.10. Let a be a set, the transitive closure of a is the smallest transitive set x such
that a ⊆ x. We denote this set by tcl(a).
Theorem 2.11. For every set a, tcl(a) exists.
Proof. We define by induction, F (0) = a and F (n + 1) = F (n) for n < ω, by Replacement if
S

{F (n) | n < ω} is a set, define F (α) = {F (n) | n < ω} for all α ≥ ω. We claim that F (ω) is
S

tcl(a).
To see that F (ω) is transitive, suppose that x ∈ F (ω), then there is some n < ω such
that x ∈ F (n) and therefore x ⊆ F (n + 1) and so x ⊆ F (ω). Moreover, since F (0) = a we
automatically have that a ⊆ F (ω).
Suppose that x is a transitive set such that a ⊆ x, we will show that for all n < ω, F (n) ⊆ x
and therefore F (ω) ⊆ x. For n = 0 this is just the assumption that a ⊆ x. Suppose that
F (n) ⊆ x, then for every y ∈ F (n + 1) there is some u ∈ F (n) such that y ∈ u by the definition
of F (n + 1) as F (n). By the assumption that x is transitive and that F (n) ⊆ x we get that
S

y ∈ x and therefore u ⊆ x, so y ∈ x as well.

Theorem 2.12 (Generalized Induction). Let R be a well-founded and set-like relation on

a class A, and let B ⊆ A such that whenever {b ∈ A | b R a} ⊆ B, then a ∈ B as well. Then
A = B.
Proof. Suppose that B ⊆ A is a class with the above property. Let a ∈ A, define by recursion
F (0) = {a} and F (n + 1) = {b ∈ A | ∃y ∈ F (n) ∧ b R y ∧ b ∈ / B} for n < ω; finally,
F (α) = {F (β) | β < α} for α ≥ ω. By the assumption that R is set-like, F is a well-defined
S

function, so F (ω) is a subset of A. By well-foundedness, if F (ω) is non-empty, then there is an

R-minimal element b there. So for some n < ω, b ∈ F (n). This means that there is no y ∈ F (ω)
such that y R b and y ∈ / B. But this means exactly that b ∈ B which is a contradiction to the
assumption that b ∈ F (ω), except if n = 0 and b = a. But then it means that F (n + 1) = ∅ for
n > 0, so a ∈ B by the defining property of B.

It shouldn’t come as a great surprise that the idea of the proof is not very different from
the one of the basic transfinite induction on the ordinals. This leads us to a theorem and proof
similar to the transfinite recursion theorem.
Theorem 2.13 (Generalized Recursion). Let R be a well-founded and set-like relation on
A, and suppose that G is a function whose domain is {⟨a, x⟩ | a ∈ A}. Then there is a unique
function F whose domain is A and F (a) = G(a, F ↾ a) (here F ↾ a is the restriction of F to
{b ∈ A | b R a}).
Theorem 2.14. Suppose that R is a well-founded and set-like relation on A. Then there exists
a unique function rankR : A → Ord such that rankR (a) = sup{rankR (b) + 1 | b R a}.
Exercise 2.12. Show that R is a well-founded relation on A if and only if there exists a function
F : A → Ord such that whenever a R b, F (a) < F (b).

Under the Axiom of Foundation, ∈ is a well-founded relation, and it is certainly set-like.

This leads us to the specific case of the generalized induction and recursion.
Theorem 2.15 (∈-Induction). If A is a class such that x ⊆ A → x ∈ A, then ∀x(x ∈ A).
Theorem 2.16 (∈-Recursion). Suppose that G(x) is a function defined for all x, then there
is a unique function F such that F (x) = G({F (y) | y ∈ x}).

10
Theorem 2.17 (Mostowski’s Collapse Lemma). Suppose that R is an extensional, well-
founded and set-like relation on A. Then there exists a unique transitive class A′ and a unique
isomorphism π : A → A′ such that a R b if and only if π(a) ∈ π(b). In particular, if R was ∈
and A is transitive, then A = A′ and π(a) = a for all a.
Proof. Define by recursion, π(a) = {π(b) | b R a}.

Exercise 2.13. Complete the proof of Mostowski’s Collapse Lemma.

Exercise 2.14. Use the collapse lemma to prove that every well-ordered set is isomorphic to a unique
ordinal.

Exercise 2.15. There is no function f whose domain is ω, and for all n < ω, f (n + 1) ∈ f (n).

11
Chapter 3

The relative consistency of the

Axiom of Foundation

So far we have taken the Axiom of Foundation for granted. And while the previous chapter
should have given us sufficient motivation, we still would like to know that if ZF0 does not prove
any false statement, then ZF will not prove false statement either.
In this section we only assume ZF0 . The definition of ordinals, mind you, stays the same,
although the proofs of the equivalent definitions will no longer work. Transfinite induction and
recursion also stay the same, although ∈-induction fails.

Definition 3.1. We say that a is a well-founded set, if {⟨x, y⟩ ∈ a × a | x ∈ y} is a well-founded

relation on a.

Exercise 3.1. The following are equivalent over ZF0 :

1. The Axiom of Foundation.

2. Every set is well-founded.

3. Every transitive set is well-founded.

4. Every set is a subset of a well-founded set.

Definition 3.2. The von Neumann hierarchy is defined by recursion on the ordinals:

1. V0 = ∅.

2. Vα+1 = P (Vα ).

3. Vα = {Vβ | β < α} for α ∈ Lim.

S

We define V = {Vα | α ∈ Ord} and call V the von Neumann universe.

S

Note that if we write x ∈ V , what we write is really that ∃α(x ∈ Vα ).

Exercise 3.2. Show that for every α, Vα is a transitive set and conclude that V is a transitive class.

Exercise 3.3. Show that for every α, Vα is a well-founded set.

Exercise 3.4. Show that if α < β, then Vα ⊆ Vβ .

12
Definition 3.3 (Relativization). Suppose that θ(x, p̄) is a formula in the language of set
theory. We define the relativization of a formula φ to θ and p̄ by recursion on the structure of
φ:

• If φ is atomic, e.g. x ∈ y or x = y, we define (x ∈ y)(θ,p̄) as x ∈ y ∧ θ(x, p̄) ∧ θ(y, p̄), and

similarly (x = y)(θ,p̄) as x = y ∧ θ(x, p̄) ∧ θ(y, p̄).

• If φ = φ1 ∗ φ2 for some connective, we define φ(θ,p̄) as (φ0 )(θ,p̄) ∗ (φ2 )(θ,p̄) .

• If φ = ¬ψ, we define φ(θ,p̄) as ¬(ψ (θ,p̄) ).

• If φ is ∃xψ, we define φ(θ,p̄) as ∃x(θ(x, p̄) ∧ ψ (θ,p̄) ).

• If φ is ∀xψ, we define φ(θ,p̄) as ∀x(θ(x, p̄) → ψ (θ,p̄) ).

If θ has no parameters, we will omit them, and write φθ . Moreover, if we denote by M the class
defined by θ (and p̄), we will write φM for the relativization of φ to θ.

Theorem 3.4. If φ is an axiom of ZF, then φV holds. In other words, V satisfies ZF, that is
ZF0 and the Axiom of Foundation.
Proof. Extensionality is easy to verify, and Infinity holds since Vω ∈ V and it is a witness for
the existence of an inductive set. Power set and Union can be proved by transfinite induction:
if x ∈ Vα , then x ∈ Vα , and P (x) ∈ Vα+1 .1
S

For readability, we will prove Replacement without parameters. Suppose that φV (u, v) is a
formula such that for x ∈ V it holds that (∀u ∈ x)∃!(v ∈ V )φV (u, v). We want to show that
there is some y ∈ V such that

y = {v ∈ V | (∃u ∈ x)φ(u, v)}.

Define ψ(u, v) as v ∈ V ∧ φV (u, v). Then by the assumption, (∀u ∈ x)∃!vψ(u, v). Therefore, by
Replacement we have that y = {v | (∃u ∈ x)ψ(u, v)} is a set in the universe. It remains to show
that y ∈ V . Note that y ⊆ V , the function f (v) = min{α | v ∈ Vα } is a well-defined function
on y, and therefore there is a set of ordinals A such that α ∈ A if and only if α = f (v) for some
v ∈ y. Let α = sup A, then for every v ∈ y we get that v ∈ Vα , and therefore y ⊆ Vα , which
means that y ∈ Vα+1 , so y ∈ V as wanted.
Finally, the Axiom of Foundation holds because every Vα is a well-founded set, and every
x ∈ V satisfies that x ⊆ Vα for some α. Therefore, in V every set is a subset of a well-founded
set and Foundation holds.

Proposition 3.5. For every α, α ⊆ Vα and there is no β < α such that α ⊆ Vβ .

Proof. We prove this by induction α. For α = 0 this is true vacuously. Suppose that the
assumption holds for all β < α. It follows that if β < α, then β ⊆ Vβ , so β ∈ Vβ+1 ⊆ Vα .
Therefore α ⊆ Vα . Suppose that β was the least such that α ⊆ Vβ . If β < α, then β + 1 ⊆ Vβ ,
which is a contradiction to the induction hypothesis.

Exercise 3.5 (*). a ∈ V if and only if tcl(a) is well-founded.

Exercise 3.6. The Axiom of Foundation is equivalent to the statement that every set lies in V .
In other words, if we started from ZF (rather than ZF0 ), then V = {x | x = x}. In other words,
constructing V in a model of ZF gives us the model again. In other other words, every model of ZF
is its own V .
1
In fact, we get more here: V “computes” power sets and unions correctly.

13
Exercise 3.7. Give an alternative proof to the following statement: Every x ∈ V has a transitive
closure.

Exercise 3.8 (*). Show that ∈-Induction is equivalent to the Axiom of Foundation.

Exercise 3.9. Show that W is a transitive class satisfying Foundation, then W ⊆ V .

Exercise 3.10. Formulate and prove an analogous theorem for Theorem 3.4 for ZF0 without Infinity.
Moreover, prove that in V as you defined it, Infinity holds if and only if it held in the outset of the
theorem.

Exercise 3.11. Show that ⟨Vω , ∈⟩ is a model for ZF without Infinity. Therefore ZF0 proves the
consistency of ZF without Infinity.

Exercise 3.12. Show that if δ is a limit ordinal, then Vδ satisfies all the axioms of ZF without
Replacement.

Proposition 3.6. Vω+ω does not satisfy Replacement.

Proof. Let φ(x, y) be the formula stating: x and y are ordinals and y = ω + x. Then φ defines
a function on ω in Vω+ω . However the image of this function is the set {ω + n | n < ω} which
does not lie in Vω+ω . Therefore Replacement fails.

Remark. One of the consequences of Gödel’s second incompleteness theorem is that ZF does not
prove its own consistency. Therefore ZF cannot prove that there exists a set M and a relation E
such that ⟨M, E⟩ satisfy all the axioms of ZF. The last two exercises prove, therefore, that assuming
Infinity or Replacement increases the power of our theory.

Remark. We have seen that the Axiom of Foundation is consistent. But maybe it is outright
provable? It turns out that the answer is negative, but due to time constraints we will not see this in
details. This idea has even been extended to “Anti-Foundation Axioms” which posits the existence
of non-well founded sets in various ways.

From here on end, we shall denote by V the universe of set theory in which we are working.

14
Chapter 4

Cardinals and their arithmetic

4.1 The definition of cardinals

As you may recall from the basic set theory course, we are interested in “measuring” the size
of sets. For finite sets, we can just count their elements, but generally we need to devise a way
that will work for infinite sets as well.

Definition 4.1. We say that two sets x and y are equipotent if there exists a bijection f : x → y.

Theorem 4.2 (Cantor–Bernstein theorem). If x is equipotent with a subset of y, and y is

equipotent with a subset of x, then x and y are equipotent.

Theorem 4.3 (Cantor’s theorem). There is no surjection from x onto P (x).

Equipotence gives rise to an equivalence relation on the sets, but this equivalence relation is
such that with the exception of ∅, the class of sets equipotent with x is always a proper class.
We would like to find objects which can be used to represent the equipotency classes.

Definition 4.4 (Scott’s trick). Suppose that E is an equivalence relation on V . Let x be any
set, and let α be the least ordinal such that for some y ∈ Vα , ⟨x, y⟩ ∈ E; we define the partial
equivalence class x/E to be {y ∈ Vα | ⟨x, y⟩ ∈ E}.

Proposition 4.5. Let E be an equivalence relation on V , then x E y if and only if x/E =

y/E.

Exercise 4.1. Show that Scott’s trick always gives rise to sets namely x/E is a set, and show that
{x/E | x ∈ V } is a class as well.

We can use Scott’s trick to define the cardinals in ZF. However, if we define |x| to be the
Scott cardinal for x, namely x/E where E denotes the equipotence relation, then we will not
have the property that |x| = |x|. This is somewhat upsetting, since we do expect a finite set
to have a cardinal number which is somewhat related to its actual size.
We digress from this discussion to recall a few facts about well-ordered sets.

Definition 4.6. We say that a set x is well-orderable if it is equipotent with an ordinal.

Equivalently, if there is some linear order < on x which is a well-order.

Theorem 4.7. If x is well-orderable, then there is a least ordinal α equipotent with x.

Proof. By assumption, the class {α | α is equipotent with x} is non-empty.

15
Definition 4.8. We say that an ordinal α is an initial ordinal, if there is no β < α such that
α and β are equipotent.

Exercise 4.2. Show that a well-ordering is isomorphic to an initial ordinal, if and only if every proper
initial segment has a strictly smaller cardinality.

Theorem 4.9. If α ≤ ω, then α is an initial ordinal.

Proof. We prove by induction. For 0 this is trivial, as is the case n = 1 (for the only smaller
ordinal is empty, and there is no bijection between a non-empty set and an empty set); assume
for n ≥ 1, then if there is a bijection f : n + 1 → n, then there is one such that f (n) = n − 1.
Therefore by restricting f to n we obtain a bijection between n and n − 1, contrary to the
induction hypothesis.
The case ω follows directly, if n < ω is equipotent with ω, then by the Cantor–Bernstein
theorem, n + 1 and n are equipotent which is a contradiction.

Definition 4.10. We say that x is finite if it is equipotent with a finite ordinal. By the previous
theorem, it is a unique ordinal. We say that x is countable, if it is equipotent with a subset of
ω.

Exercise 4.3. The ordinal ω + ω is equipotent with ω, therefore not every limit ordinal is an initial
ordinal.

Definition 4.11. Let x be a set. We define the cardinal of x to be either the unique initial
ordinal equipotent with x in the case that x is well-orderable, or the Scott cardinal of x in case
x cannot be well-ordered. We denote this as |x|.

Now we have that |x| = |y| if and only if x and y are equipotent. Moreover, if x can be
well-ordered, then |x| is the least ordinal to which x is equipotent, and in particular if x is finite
then |x| faithfully represents the number of elements in x. We will write |x| ≤ |y| to denote
that x is equipotent with a subset of y, and |x| < |y| to mean that |x| ≤ |y| and |x| = ̸ |y|.

Definition 4.12 (Cardinal arithmetic). Suppose that x and y are two sets.

Addition We define |x| + |y| to be |x × {0} ∪ y × {1}|.

Multiplication We define |x| · |y| to be |x × y|.

Exponentiation We define |x||y| to be |{f : y → x}|.

Exercise 4.4. Show that cardinal addition is commutative and associative. Moreover, show that if
n, m are finite ordinals, then ordinal and cardinal arithmetic coincide.

Exercise 4.5. Show that |x| · (|y| + |z|) = (|x| · |y|) + (|x| · |z|); |x||y| · |x||z| = |x||y|+|z| ; and
(|x||y| )|z| = |x||y|·|z| .

Exercise 4.6. Show that |x| + |x| = |x| · 2 (by induction conclude for every finite ordinal in place of
2), and |x| · |x| = |x|2 (again, conclude by induction for every finite ordinal in place of 2).

Exercise 4.7. Show that for every x, there is some y such that |x| ≤ |y|, and |y|2 = |y|.

Exercise 4.8. If α is an infinite ordinal, then |α| + |1| = |α + 1| = |α|.

16
4.2 The Aleph numbers
Theorem 4.13 (Hartogs’ theorem). If x is a set, then there is some ordinal α such that
|α| ≰ |x|.
Proof. Let W denote the set {A ⊆ P (x) | ⟨A, ⊊⟩ is a well-ordered set}, consider the function
defined on W such that F (A) = α if and only if ⟨A, ⊊⟩ ∼= α, this is a well-defined function, since
a well-ordered set is isomorphic to a unique ordinal. Note that if an ordinal β is in the range
of F , then every γ < β is also in the range of F : if A has order type β, then it has an initial
segment which have order type γ. Therefore rng F is an ordinal, denote it by α. If f : α → x is
an injection, define aβ = rng f ↾ β for β ≤ α. We get that A = {aβ | β ≤ α} is well-ordered by
strict inclusion, and therefore A ∈ W. But at the same time A is isomorphic to α, and therefore
α ∈ rng F which is a contradiction since rng F = α and α ∈ / α. Therefore |α| ≰ |x|.

Exercise 4.9. Show that if α is the least ordinal which does not inject into a set x, then α is an
initial ordinal (i.e. a cardinal).

Definition 4.14 (Hartogs’ number). Let x be a set, we write ℵ(x) as the least initial ordinal
α such that |α| ≰ |x|.

Definition 4.15. We define by recursion, ω0 = ω; ωα+1 = ℵ(ωα ) and if ωβ were defined for all
β < α for α ∈ Lim, then ωα = sup{ωβ | β < α}.

Note that by Exercise 4.8 every ωα is a limit ordinal.

Exercise 4.10. η ≥ ω is an initial ordinal if and only if there is some α such that η = ωα .

In order to discern cardinality from order type, and to make it easier to understand the
context in which arithmetic operations are interpreted, we write ℵα as the cardinal of ωα .
While these are formally the same object, we will understand ω2 + ω1 as ordinal addition,
whereas ℵ2 + ℵ1 as cardinal addition.

Theorem 4.16. For every α, ℵα · ℵα = ℵα .

Proof. We will define a well-ordering on Ord × Ord and show that restricting it to ωα × ωα has
order type ωα . This will provide us with a bijection between the two sets, thus proving the
theorem.
We define the following ordering on Ord × Ord:

max{β, γ} < max{β , γ } or
 ′ ′

′ ′
⟨β, γ⟩ <gp ⟨β , γ ⟩ ⇐⇒ max{β, γ} = max{β ′ , γ ′ } ∧ β < β ′ or

max{β, γ} = max{β ′ , γ ′ } ∧ β = β ′ ∧ γ < γ ′


Easily this is an irreflexive order, and verifying transitivity and linearity is straightforward.
To see that this is a well-ordering, suppose that A is a non-empty subset of Ord × Ord. The
set {max{β, γ} | ⟨β, γ⟩ ∈ A} is a non-empty set of ordinals, let A′ ⊆ A be those pairs in A
mapped to its minimum. Among the pairs in A′ , let β be the least ordinal appearing in the left
coordinate of an ordered pair in A′ ; and let γ be the least ordinal such that ⟨β, γ⟩ ∈ A′ . It is
not hard to check that ⟨β, γ⟩ is the minimum element of A.
Let ↓(β, γ) denote the set {⟨ε, δ⟩ | ⟨ε, δ⟩ <gp ⟨β, γ⟩} this is a set, since every such ⟨ε, δ⟩ must
satisfy that max{ϵ, δ} < max{β, γ} + 1. By the fact that <gp is a well-ordering, every proper
initial segment of the form ↓(β, γ). Moreover, note that if η is any ordinal, then η × η = ↓(0, η).

17
 We will now prove by induction that ↓(0, ωα ) has the same order type as ωα . For α = 0,
we get that every proper initial segment is a subset of ↓(0, m) for some m < ω. Since the
underlying set of ↓(0, m) is m × m, which is finite, we get that <gp up to ω × ω is of order type
ω (note that this is the initial segment ↓(0, ω) as remarked above).
Suppose that α is infinite and for all α′ < α the order type of ↓(0, ωα′ ) is ωα′ . Let η < ωα ,
and we may assume η ≥ ω, since ωα × ωα = {η × η | η < ωα } it is enough to show that ↓(0, η)
S

has cardinality strictly less than ℵα . By the fact that η < ωα , |η| < ℵα , then |η| = ℵα′ for some
α′ < α. By the induction hypothesis, then, |η × η| = |η|, and therefore of ↓(0, η) has cardinality
ℵα′ . In particular, every proper initial segment of ↓(0, ωα ) has cardinality strictly less than ℵα ,
so it means that ωα × ωα = ↓(0, ωα ) is isomorphic to ωα as wanted.

Corollary 4.17. If α ≤ β, then ℵα + ℵβ = ℵα · ℵβ = ℵβ .

Exercise 4.11. Let α and β be ordinals such that at least one of them is infinite. Then α + β, α · β
and αβ all have the same cardinality as max{|α|, |β|}.

Exercise 4.12. There is no set x such that | P (x)| = ℵ0 .

Let us define the following relation on cardinals |x| ≤∗ |y| if and only if there is a surjective
function from a subset of y onto x (alternatively, either x = ∅, or there is a surjection from y
onto x). This relation is reflexive and transitive, but not provably anti-symmetric.

Exercise 4.13 (Lindenbaum’s theorem). Prove that for every x there is some ordinal α such
that |α| ≰∗ |x|. Let ℵ∗ (x) denote the least such ordinal, show that ℵ∗ (x) is a cardinal and that
ℵ(x) ≤ ℵ∗ (x). We shall refer to ℵ∗ (x) as the Lindenbaum number of x.

Exercise 4.14. Show that ℵ(x) < ℵ(P (P (P (x)))).

Exercise 4.15. Show that if |x| ≤∗ |y|, then | P (x)| ≤ | P (y)|.

Exercise 4.16 (**) (Specker trees). For a set x, let us define the Specker tree on x, S(x) by
recursion: S(x) is {|x|} ∪ {S(y) | | P (y)| = |x|}, with |y| <S |y ′ | if and only if |y ′ | = | P (y)|.
S

Prove that S(x) is a set, that <S is a well-founded partial order on S(x). (Hint: Use the previous
exercise to justify that <S is well-founded.)

4.3 Finiteness
We defined a set to be finite if it was in bijection with a finite ordinal, and that is certainly one
way of defining finiteness. But we can examine what other properties finite ordinals have which
“familiar infinite sets” do not, and try to extrapolate these for our definitions.

Definition 4.18 (Finiteness). Let x be a set.

1. We say that x is finite if it can be put in bijection with a finite ordinal.

2. We say that x is amorphous if it cannot be written as the disjoint union of two infinite
sets.

3. We say that x is Tarski-finite if every ⊆-chain in P (x) is finite.

4. We say that x is strongly ∗-finite if there is no surjection from x onto ω.

5. We say that x is ∗-finite if there is no surjection from x onto x ∪ {x}.

18
 6. We say that x is Dedekind-finite if every injection from x into x is a bijection.
Remark. The fourth definition is sometimes called by the terrible name “weakly Dedekind-finite”
(which is probably due to the fact that its negation is a weakening of Dedekind-infinite sets), and
the term Tarski-finite is sometimes squandered on a definition equivalent to (true) finiteness.
Exercise 4.17 (*). Show that the above definitions form a hierarchy. Namely, if x satisfies a
definition, it must satisfy all those that follow it.
Exercise 4.18. Show that if x is well-ordered, then being Dedekind-finite implies being finite.
Exercise 4.19 (*). Show that the union and product of two strongly ∗-finite sets is again a strongly
∗-finite set.
Exercise 4.20. Show that x is Dedekind-finite if and only if ℵ(x) ≤ ℵ0 .
Exercise 4.21 (*). Show that if there exists an infinite Dedekind-finite set x, then there is a
Dedekind-finite set y which is not ∗-finite. (Hint: show that the set I(x) of all finite injective
sequences from x is Dedekind-finite, and then show that there is a surjection from I(x) \ {∅} onto
I(x).)
Theorem 4.19 (Kuratowski). x is strongly ∗-finite if and only if P (x) is Dedekind-finite. In
other words,
ℵ0 ≤∗ |x| ⇐⇒ ℵ0 ≤ | P (x)|.
Proof. In the one direction, by Exercise 4.15 if ℵ0 ≤∗ |x|, then ℵ0 < 2ℵ0 ≤ | P (x)|. So if x is
not strongly ∗-finite, P (x) is Dedekind-infinite.
In the other direction, suppose that P (x) is Dedekind-infinite and an is a sequence of sets in
P (x). Our goal is to show that there is a sequence bn of pairwise disjoint and non-empty sets,
in which case we can define the following function:
(
n a ∈ bn
f (y) =
0 a∈/ {bn | n < ω}
S

which is easily a surjection onto ω.

If the an ’s are pairwise disjoint, we finished the proof. And if there is a ⊆-decreasing
subsequence (without loss of generality, the sequence itself) we can define bn+1 = an \ an+1 to
obtain this. So we may assume that every ⊆-decreasing subsequence is finite. We define bn by
induction. Let s0 = {an | n < ω}, and for simplicity we may also assume x = a0 . Suppose that
bm was defined for m < n, and suppose that
n [ o
ak \ {bm | m < n} | k ≥ n

is infinite. Define n∗ to be the least k, if it exists, such that both ak and x \ ak do not
cover {bm | m < n}. If n∗ exists define bn = an∗ \ {bm | m < n} in the case where
S S

{ak \ ( {bm | m < n} ∪ an∗ | k ≥ n∗ } is infinite; or bn = x \ (an∗ \ {bm | m < n}) (in which
S S

case the set defined similarly must have infinite many elements).
In case that n∗ does not exist, define sj+1 = {ak \ {bm | m < n} | k ≥ n} and restart the
S

process described above. If for some j, we managed to define bn for all n < ω from sj , then we
finished because we have a sequence of pairwise disjoint non-empty sets. Otherwise, for every
j we got stuck, then taking each finite sequence of bn ’s but by the definition of sj+1 , the next
finite sequence is pairwise disjoint from it. Therefore in either case we end up with a sequence
of pairwise disjoint sets, as wanted.

19
 The idea behind the proof, ultimately, is at each step take either a subset of some an or
a subset of its complement, which is never empty, and the union of everything we have thus
far—including the new set—will not cover everything. If we happened to run into a dead-end,
we refine the sequence of sets and continue the construction from those refined sets. In either
case, this produces a sequence of pairwise disjoint sets, from which we can define a surjection
as wanted.

20
Chapter 5

Absoluteness and reflection

5.1 Absoluteness
Definition 5.1. If φ is a formula in the language of set theory, we say that it is a bounded
formula if every quantifier appearing in φ is of the form (∃x ∈ y) or (∀x ∈ y).

Definition 5.2 (The Levy Hierarchy). Let φ be a formula in the language of set theory.

1. φ is a Σ0 or Π0 if it is a bounded formula.

2. φ is a Σn+1 if there is a Πn formula ψ such that φ = ∃xψ.

3. φ is a Πn+1 if there is a Σn formula ψ such that φ = ∀xψ.

n (Πn ) if ZF proves that φ is equivalent to a Σn (Πn ) formula, and we

We say that φ is a ΣZF ZF
ZF
n and Πn .
say that φ is ∆n if it is both ΣZF ZF

From this point onward, we will omit the ZF superscript, and write just Σn , Πn and ∆n .

Exercise 5.1. Σn and Πn are closed under conjunction, disjunction and bounded quantification; if
n > 0 then Σn formulas are closed under existential quantifiers and Πn are closed under universal
quantifiers. Finally, Πn formula is the negation of a Σn formula (and therefore vice versa).

Exercise 5.2. Show that following statements are ∆0 : “x is an ordinal”, “x is transitive”, “x is a

function”, “x is a finite ordinal”, “x is ω”, “x is a function and y is in the domain of x”.

Exercise 5.3. Show that “x is a finite set” is a ∆1 statement.

Theorem 5.3. Suppose that φ(u1 , . . . , un ) is a ∆0 formula. Then for every x1 , . . . , xn and for
every transitive class A it holds that ⟨A, ∈⟩ |= φ(x1 , . . . , xn ) if and only if x1 , . . . , xn ∈ A and
φ(x1 , . . . , xn ) holds in V .

Remark. From here on end, we will write A |= φ to mean that ⟨A, ∈⟩ |= φ.

Proof. We prove this by structural induction on φ. For φ an atomic formula this is obviously
true, and the proof for connectives is as usual. If φ is (∀y ∈ x)ψ and x ∈ A, then by transitivity
x ⊆ A. Therefore A |= ψ(y) if and only if ψ(y) holds in V , so A |= φ(x) if and only if φ(x)
holds in V . The proof for (∃y ∈ x)ψ is similar.

21
Definition 5.4. We say that φ is downwards absolute if whenever A is a transitive class such
that A |= φ, and B is a transitive subclass of A (possibly a set), then B |= φ. Similarly, φ is
upwards absolute if whenever A is a transitive class such that A |= φ, then whenever B is a
transitive class with A ⊆ B, then B |= φ. If φ is both upwards and downwards absolute, we
just say it is absolute.

The above theorem, then, states that ∆0 formulas are absolute between any two transitive
[sets or] classes which include the relevant assignments.

Theorem 5.5. Every Σ1 formula is upwards absolute and every Π1 formula is downwards
absolute. Consequently, every ∆1 formula is absolute.
Proof. Let ∃xφ(x) be a Σ1 formula with φ a ∆0 formula. Assume that A |= ∃xφ(x) and B is
a transitive superclass of A. Then there is some a ∈ A such that A |= φ(a), by the previous
theorem B |= φ(a) and therefore B |= ∃xφ(x).
The proof for the Π1 case is similar: if A |= ∀xψ(x) with ψ a ∆0 formula, and B ⊆ A
is a transitive class, then for every b ∈ B we get that A |= ψ(b), so B |= ψ(b). Therefore
B |= ∀xψ(x). The consequence for ∆1 now follows.

Exercise 5.4. Show that φ is upwards absolute if and only if ¬φ is downwards absolute.

Definition 5.6 (Elementary submodel). Suppose that A is a structure in a language L. We

say that B is an elementary substructure of A if it is a substructure, and the following holds:
For every formula φ(u1 , . . . , un ) in L, and for every b1 , . . . , bn : A |= φ(b1 , . . . , bn ) if and only if
B |= φ(b1 , . . . , bn ).
We denote this by B ≼ A. If L is the language of set theory, we write ≺Σ1 , ≺Πn , etc. to
mean that the above definition holds for the specified class of formulas.

As we have seen, every transitive class is a ∆1 -elementary submodel of V .

Theorem 5.7 (Tarski–Vaught criterion). Let A be a structure in the language L and let
B be a substructure of A. Then B ≼ A if and only if for every formula φ(x, u1 , . . . , un ) and
b1 , . . . , bn ∈ B, if A |= ∃xφ(x, b1 , . . . , bn ) then there is b ∈ B such that A |= φ(b, b1 , . . . , bn ).

Theorem 5.8. Suppose {Mα | α < β} is a sequence of structures in some fixed language L,
and for α < α′ , Mα ≼ Mα′ . Then Mβ defined as {Mα | α < β} is an L-structure and for all
S

α < β, Mα ≼ Mβ .

Theorem 5.9. If M is a well-orderable structure, and A ⊆ M , then there is an elementary

submodel N ≼ M such that A ⊆ N and |N | = |A| + ℵ0 .

Exercise 5.5. Verify that “x is countable” is a Σ1 formula.

Definition 5.10. We say that x is Σn -definable (Πn -definable) in a transitive class A, if there
is a Σn (Πn ) formula φ(u) such that A |= φ(u) ↔ u = x. If we allow parameters in the formula,
in which case we say that x is Σn -definable (Πn -definable) in p1 , . . . , pn (and require them to
be in A).

Exercise 5.6. Show that “α is an initial ordinal” is a Π1 formula. Show that ω1 is Π2 -definable.

Exercise 5.7. If M ≼ N then every member of N which is definable, is an element of M .

Theorem 5.11. Let H(ω1 ) denote the set {x | tcl(x) is countable}. If M is a countable ele-
mentary submodel of H(ω1 ), then M is transitive.

22
Proof. Suppose that M is a countable elementary submodel of H(ω1 ). Let x ∈ M , then in
H(ω1 ) there exists a bijection between x and a subset of ω. By elementarity the same must
hold in M . However, ω is ∆0 -definable, so ω ∈ M and ω ⊆ M by similar arguments. Let f ∈ M
be an injective function f : x → ω, then every member of the range of f is in M and therefore
every member of the domain of M must also be in M . In other words, x ⊆ M .

Exercise 5.8 (*). Let H(ω2 ) denote the set {x | | tcl(x)| ≤ ℵ1 } and assume that M is a countable
elementary submodel of H(ω2 ). Prove that M cannot be transitive, and use Mostowski’s collapse
lemma to prove that ω1 is not Σ1 -definable. (Hint: Assume by contradiction, collapse M and use
Theorem 5.5.)

5.2 Reflection
We saw in the previous section that the Vα ’s which are transitive satisfy that they are ∆1 -
elementary submodels of V . But what about formulas which are not ∆1 ? Can we find a way
to reflect them in some transitive set?

Theorem 5.12 (The Reflection Theorem). Let φ(u1 , . . . , un ) be a formula in the language of
set theory. Then ZF proves that for every α there exists β > α such that for all x1 , . . . , xn ∈ Vβ ,
φ(x1 , . . . , xn ) ↔ φVβ (x1 , . . . , xn ).

In this case, we say that Vβ reflects φ. Note that φVβ holds if and only if Vβ |= φ. We will
first prove two lemmas; and for readability we will assume that φ has one free variable.

Lemma 5.13. Let φ(x, u) be a formula in the language of set theory, then for every α there is
some β ≥ α such that for all a ∈ Vβ , ∃xφ(x, a) → (∃x ∈ Vβ )φ(x, a).
Proof of Lemma 5.13. We define by recursion a sequence βn . Take β0 = α. Suppose that βn
was defined, then the function f be defined on Vβn as follows f (a) = min{γ | ∃xφ(x, a) →
(∃x ∈ Vγ )φ(x, a)}, by Replacement rng f is a set of ordinals, let βn+1 = sup rng f . Finally, let
β = sup{βn | n < ω}.
If a ∈ Vβ , then there is some n < ω such that a ∈ Vβn , then if there exists x such that
φ(x, a), then by definition there is such x in Vβn+1 and therefore in Vβ .

Lemma 5.14. Suppose that {αn | n < ω} is a set of ordinals such that for each n, Vαn reflects
φ. Let α = sup{αn | n < ω}, then Vα reflects φ.
Proof of Lemma 5.14. We prove this by induction on the structure of φ. If φ is atomic, then
absoluteness implies that every Vγ reflects φ. Connectives and negations are easily verified using
truth tables.
Suppose that φ has the form ∃xψ(x, u) and let a ∈ Vα . There is some n such that a ∈ Vαn ,
and therefore V |= ∃xψ(x, a) if and only if (∃x ∈ Vαn )ψ Vαn (x, a), by the induction hypothesis
Vα reflects ψ and therefore V |= ∃xψ(x, a) if and only if (∃x ∈ Vα )ψ Vα (x, a).
Proof of Theorem 5.12. We prove this by induction on the structure of φ. For atomic formulas
this follows from absoluteness. Negation and connectives follow by verifying truth tables. For
φ(u) defined as ∃xψ(x, u), we recursively define an intertwined sequence: β0 = α + 1; for odd
indices, β2n+1 is the least ordinal obtained from Lemma 5.13 such that Vβ2n+1 is closed under
∃xψ; for even indices, we take β2n+2 to be the least ordinal above β2n+1 such that Vβ2n+2 reflects
ψ, such ordinal exists by the inductive hypothesis on ψ.
Let β be sup{βn | n < ω}, and let a ∈ Vβ . If V |= ∃xψ(x, a), then by the same argument as
Lemma 5.13 we get that (∃x ∈ Vβ )ψ(x, a), but since Vβ is the limit of points which reflects ψ,

23
this is the same as saying that (∃x ∈ Vβ )ψ Vβ (x, a), or φVβ (a). In the other direction, if φVβ (a)
holds, then there is some x ∈ Vβ such that ψ Vβ (x, a) holds, but again by reflection ψ(x, a) holds
so ∃xψ(x, a) holds. Therefore Vβ reflects φ as wanted.

Remark. For n > 0 we can prove there exists a Σn -truth predicate. Reflecting it means that there is
a proper class of ordinals which are Σn -elementary submodels of V ; and moreover that ZF cannot be
given a finite axiomatization, as we could reflect such finite list of axioms and obtain that ZF proves
its own consistency. This would be a contradiction to Gödel’s second incompleteness theorem.
Question: why do compactness and reflection not prove the consistency of ZF?

Exercise 5.9 (*). The axiom schema of Replacement is equivalent to the Reflection theorem over
Z with the assumption that every set belongs to some Vα .

Exercise 5.10 (*). Suppose that {Dα | α ∈ Ord} is a sequence of transitive sets such that ZF
proves that: Dα ⊆ Dβ for all α < β; for α ∈ Lim, Dα = {Dβ | β < α}; and every set lies inside
S

some Dα . Then for every φ in the language of set theory, ZF proves that for every α there is some
β > α for which φ ↔ φDβ .
Namely, every continuous filtration of V satisfies Reflection.

Exercise 5.11. x is Σn -definable if and only if there is an ordinal α such that x is Σn -definable in
Vα .

Exercise 5.12. Let φ be a formula in the language of set theory. Show that the class of ordinals
β such that Vβ reflects φ is closed. Namely, if {β < δ | Vβ reflects φ} is unbounded in δ, then Vδ
reflects φ.

24
Chapter 6

The Axiom of Choice

6.1 The Axiom of Choice

Definition 6.1. We say that a function f is a choice function if for all x ∈ dom f , f (x) ∈ x.
We say that x admits a choice function if there is a choice function on x \ {∅}.

Exercise 6.1. Show that “f is a choice function” is a ∆0 statement, and “x admits a choice
function” is a Σ1 statement.

Theorem 6.2. If x is finite, then x admits a choice function.

Proof. Suppose that |x| = n, then for n = 0 the statement is trivially true. Suppose that
for |x| = n there is a choice function, let |x ∪ {xn }| be a set of size n + 1, then |x| admits a
choice function f , and either xn = ∅ in which case f is a choice function on x ∪ {xn }, or xn
is non-empty, in which case let yn ∈ xn be some element and then f ∪ {⟨xn , yn ⟩} is a choice
function on x ∪ {xn }.

Remark. The proof seems as though it should work for |x| = ℵ0 . However, when removing one
element from a set of size ℵ0 , we remain with a set of size ℵ0 so there is no way we can appeal to
an induction hypothesis. It is true, however, that if x and y admit a choice function, then x ∪ y
admits a choice function as well.

Definition 6.3. The Axiom of Choice is the axiom stating that for every x admits a choice
function.

Definition 6.4. If X is a set, X is the set of all choice functions that X admits.
Q

Exercise 6.2. Suppose that X = {xi | i ∈ I}, then {f : I → X | f (i) ∈ xi } is non-empty if and
S
Q
only if X is non-empty.

Theorem 6.5 (Zermelo’s theorem). The Axiom of Choice holds if and only if for every x
is well-orderable.

In other words, the Axiom of Choice is equivalent to stating that every cardinal is an ordinal.

Exercise 6.3. Let α be an initial ordinal. Show that if x ≤ α or x ≤∗ α, then x is well-orderable.

Theorem 6.6. The following are equivalent:

1. AC.

25
 2. (Zorn’s lemma) If ⟨P, <⟩ is a partial order where every chain C ⊆ P has an upper
bound, then there is a maximal element in P .

3. (Weak Zorn’s lemma) If ⟨P, <⟩ is a partial order where every well-ordered chain C ⊆ P
has an upper bound, then there is a maximal element in P .

4. (Hausdorff’s Maximality Principle) If ⟨P, <⟩ is a partial order, then there is a max-
imal chain C ⊆ P .

5. For every x, P (x) admits a choice function.

6. For every x whose members are pairwise disjoint with ∅ ∈

/ x, there is C such that for
every y ∈ x, C ∩ y is a singleton.

7. ≤ is a linear ordering of the cardinals.

8. ≤∗ is a linear ordering of the cardinals.

Exercise 6.4. If ⟨P, ≤⟩ is a partial order satisfying the conditions of the Zorn’s lemma, if P is
well-orderable, then P has a maximal element.

Exercise 6.5 (*). The Axiom of Choice is equivalent to the statement “There is some n < ω, such
that given any n cardinals, at least two of them are comparable by ≤ or by ≤∗ ”.

Remark. If we change the above to be “Given any infinitely many cardinals, two of them are
comparable [in either order]” is not known to imply the axiom of choice, or even the fact that any
Dedekind-finite set is finite, which means we do not even know if this is equivalent to “Given any
countably infinite set of cardinals, two of them are comparable”.

Theorem 6.7. AC is equivalent to the statement “If x is well-orderable, then P (x) is well-
orderable”.
Proof. One direction follows from Zermelo’s theorem. We shall prove the second direction. It
is enough, of course, to show that every Vα is well-orderable.
We prove by induction that for every α, Vα can be well-ordered. We proved that Vω is
countable, so for α ≤ ω the claim is true. If Vα can be well-ordered, by the assumption
P (Vα ) = Vα+1 can be well-ordered.
Suppose that α is a limit ordinal, and for every β < α, Vβ can be well-ordered. Let λ
be ℵ(Vα ), then λ × λ can be well-ordered, and therefore P (λ × λ) can be well-ordered, fix a
well-ordering ◁ of P (λ × λ). Since every β < α satisfies that Vβ can be well-ordered, it has
to be the case that |Vβ | < λ. Therefore, for every β < α there is some Rβ ⊆ λ × λ such that
Rβ is an extensional and well-founded relation on its domain, and the Mostowski collapse of
⟨dom Rβ , Rβ ⟩ is exactly Vβ . Moreover, by the fact that ◁ is a well-ordering of P (λ × λ) we can
choose the least such relation. Note that the isomorphism between Rβ and Vβ is unique, so
once Rβ was chosen (using ◁) there is no choice when we identify between Vβ and dom Rβ . Let
πβ : Vβ → dom Rβ be that unique isomorphism.
Finally, we define a well-ordering on Vα as follows: x ≺ y if and only if rank(x) < rank(y),1
or rank(x) = rank(y) = β and πβ (x) < πβ (y). Easily, ≺ is a well-order of Vα as wanted.

Exercise 6.6. The naive proof that Vα is well-orderable, without considering well-ordering of previous
steps, fails. Why? Moreover, the proof does not imply that there is a class well-ordering of V . Why?
1
Note that rank(x) = α if and only if α = min{β | x ∈ Vβ+1 } = min{β | x ⊆ Vβ } is the rank function which
exists from Replacement and Foundation.

26
Exercise 6.7. Show that if AC holds, then for every δ ∈ Lim, Vδ |= AC.
Theorem 6.8 (Tarski). The Axiom of Choice holds if and only if for every infinite x, |x|2 =
|x|.
Proof. In the one direction, if the Axiom of Choice holds and x is infinite, then |x| = ℵα for
some α, and therefore |x|2 = ℵα · ℵα = ℵα = |x|.
In the other direction, let λ be ℵ(x), and without loss of generality x ∩ λ = ∅. Then the
following holds:
|x| + λ = (|x| + λ)2 = |x|2 + λ2 + |x| · λ · 2 = |x| + λ + |x| · λ.
Therefore it has to be the case that |x| · λ ≤ |x| + λ, and since the other inequality is trivial we
get |x| + λ = |x| · λ. Using Lemma 6.9, we get that |x| and λ are comparable, but by taking
λ = ℵ(x) we get that it is necessarily the case that |x| ≤ λ, so x can be well-ordered. Since every
finite set is well-orderable by definition, we have concluded that every set is well-orderable, so
the Axiom of Choice holds.

Lemma 6.9. If a is a set and λ is an initial ordinal, then |a| + λ = |a| · λ implies that |a| is
comparable with λ.
Proof. Without loss of generality we can assume that a ∩ λ = ∅ and that λ is infinite (if λ is
finite then it is comparable with |a| already). Let f : a × λ → a ∪ λ be a bijection. If there exists
some b ∈ a such that f ”{b} × λ ⊆ a,2 then f defines an injection from λ into a by α 7→ f (b, α).
Assume otherwise, then for every b ∈ a there is some α such that f (b, α) ∈ λ. This defines
an injection from a into λ × λ given by b 7→ ⟨α, f (b, α)⟩ where α is the least ordinal for which
f (b, α) ∈ λ. Since λ is infinite, |λ × λ| = λ, and therefore |a| ≤ |λ|.

Exercise 6.8 (*) (Abraham’s Lemma). If there is a surjection f : a ∪ b → a × b then either

|b| ≤∗ |a| or |a| ≤∗ |b|.
Theorem 6.10. Suppose that for every infinite set G there is a binary operation ⊙ such that
⟨G, ⊙⟩ is a group. Then the Axiom of Choice holds.
Proof. Let x be an infinite set, and let λ be an ordinal such that λ ≥ ℵ(x). Without loss of
generality, x ∩ λ = ∅. Let G = x ∪ λ and fix ⊙ to be a binary operation on G which makes it
into a group. Note that by the group axioms, if a ⊙ b = c ⊙ b, then a = c.
If there is some y ∈ x such that for all α < λ, α ⊙ y ∈ x, then α 7→ α ⊙ y is an injective
function from λ into x. By the choice of λ, this is of course impossible. Therefore for every
y ∈ x there is some α < λ such that α ⊙ y ∈ λ. Then y 7→ ⟨α, β⟩ such that α is the least ordinal
for which α ⊙ y = β is an injective function from x into λ × λ. Therefore x is well-orderable,
and therefore the axiom of choice holds.

6.2 Weak version of the Axiom of Choice

Definition 6.11. Let x, y be sets, we shall write [x]|y| as {a ⊆ x | |a| = |y|}, [x]<|y| and [x]≤|y|
are defined similarly by replacing |a| = |y| by the suitable inequality.
Definition 6.12. Let ACyx (z) denote the statement: Every a ⊆ [z]≤|y| such that |a| ≤ |x| admits
a choice function. If we omit one (or more) of the parameters, then we replace it by a universal
|y|
quantifier. So ACℵ0 is the same as ∀x∀yACℵ0 (x), or “Every countable family admits a choice
function”.
We also define versions with ≤ |x| and < |x| for the parameters x and y above.
2
We will use f ”X to denote the direct image of X, namely {f (x) | x ∈ X}.

27
Exercise 6.9. For all α < β, ACℵβ → ACℵα .

Exercise 6.10 (*). ACℵ0 implies that every Dedekind-finite set is finite.

Theorem 6.13. ∀α ACℵα if and only if for every x, ℵ(x) = ℵ∗ (x).

Proof. Suppose that every well-orderable family admits a choice function. We saw that ℵ(x) ≤
ℵ∗ (x); suppose that η < ℵ∗ (x), then there is a surjection f : x → η, but now Aα = {y ∈ x |
f (y) = α} is a family of non-empty sets, and by assumption admits a choice function. It is not
hard to verify that such a choice function implies that η < ℵ(x) and so equality follows.
In the other direction, suppose that ℵ(x) = ℵ∗ (x) for every x. We will prove by induction
that ACℵα holds. Suppose that α is an ordinal, such that AC<ℵα holds. Note that for α = 0 this
is a theorem of ZF. Let {aη | η < ωα } be a family of non-empty sets. For each γ < ωα , let xγ
be the set η<γ aη , then by our induction hypothesis xγ is non-empty. We define by induction
Q

two sequences:

• λγ = ℵ ( {Dη | η < γ}) + sup{ℵ(λη ) | η < γ}.

S

• Dγ = xγ × λγ .

Finally, let λ = sup{λγ | γ < ωα } and D = {Dγ | γ < ωα }. There is a natural surjection from
S

D onto λ, so λ < ℵ∗ (D) and therefore λ < ℵ(D). Fix an injection F : λ → D, then F cannot
be injective into any fixed Dγ , as λ ≥ ℵ(Dγ ). This means that for every γ < ωα there is some
β < λ such that F (β) ∈ Dγ ′ for some γ ′ > γ. Define βγ to be the least such β (this might not
be an injective assignment).
Finally, note that if F (βγ ) = ⟨fγ , ξγ ⟩, then fγ is a choice function whose domain includes
aγ . Which therefore defines a choice function from the entire family: h(aγ ) = fγ (aγ ).

Exercise 6.11 (*). ACℵ0 holds if and only if for every countable family {an | n < ω} there is an
infinite I ⊆ ω such that {ai | i ∈ I} admits a choice function.

Theorem 6.14. ACℵ0 holds if and only if whenever X is a metric space and A ⊆ X, then
x ∈ A if and only if there is a sequence ⟨an | n < ω⟩ ⊆ A such that lim an = x.
Proof. Let A = {An | n < ω} be a family of non-empty sets. Without loss of generality,
An ∩ Am = ∅ for n ̸= m. Let X = A ∪ {∞}, with ∞ some set not in A. We define the
S S

following metric on X:



 0 x=y
x ∈ An , y ∈ Am , n ̸= m

 1 − 1

d(x, y) = 1n m


 n x, y ∈ An

1 x ∈ An , y = ∞ or x = ∞, y ∈ An

n

We leave the reader with the task of verifying the definitions of a metric on X. Moreover, A is
dense in X: if ε > 0, then for some n < ω, n1 < ε, then A ∩ Bε (∞) contains Am for all m > n.
By the assumption, there is some sequence xn ∈ X such that xn ∈ A and lim xn = ∞.
This defines a choice function on {Am | ∃n(xn ∈ Am )} by taking the least such xn for each m.
Moreover, this family of sets has to be infinite, otherwise inf{d(xn , ∞) | n < ω} > m
1
for some
m, in contradiction to the convergence assumption. By Exercise 6.11 we get ACℵ0 as wanted.
In the other direction, assume ACℵ0 holds and X is an arbitrary metric space. Let A ⊆ X be
some non-empty subset, and a ∈ A. Then for every n < ω, A ∩ B 1 (a) is non-empty. Using ACℵ0
n
we can choose some xn ∈ A such that d(a, xn ) < 1/n, and it is clear now that lim xn = a.

28
Definition 6.15. We say that ⟨T, <T ⟩ is a tree if <T is a well-founded partial order on T , such
that for every t ∈ T , {s ∈ T | s <T t} is linearly ordered by <T . We write Tα as the elements
of T whose rank in <T is α, and the height of T is sup{α + 1 | Tα ̸= ∅}. We say that B ⊆ T is
a branch if it is a maximal chain; and it is a cofinal branch if it is a branch such that for every
α, B ∩ Tα ̸= ∅.
Definition 6.16. Let DCℵα (y) denote the statement that whenever T ⊆ y is a tree of height
ωα in which every chain of length less than ωα has an upper bound not in the chain, then T has
a cofinal branch. If we omit y, we do not restrict T to being a subset of any particular y; and
DC<ℵα denotes ∀β(β < α → DCℵβ ). If we omit the ℵα subscript we will always mean DCℵ0 .
Exercise 6.12. Show that if |X| = ℵα , then DCℵα (X) holds. In other words, if T is a well-orderable
tree satisfying the assumptions of DCℵα , then T has a cofinal branch.
Exercise 6.13. For every α and x, DCℵα (x) → ACℵα (x).
Exercise 6.14 (**). ∀α ACℵα → DC.
Remark. It was proved by Azriel Levy that the above implication cannot be extended even to DCℵ1 .
Theorem 6.17. The following are equivalent:

1. DC.

2. Every structure in a countable language has an elementary submodel of size ℵ0 .

3. For every α > ω, Vα has a countable elementary submodel.

The implication of (1) =⇒ (2) is beyond the scope of this course, but it can be proved
by tracing the usual proof of the Downwards Löwenheim-Skolem theorem and checking that we
only need DC to prove it for the countable case.
Proof. Clearly (2) =⇒ (3). It remains to show that (3) =⇒ (1). Let α be an arbitrary
infinite ordinal, we will show that if T ∈ Vα is a tree satisfying the conditions of DC, then T has
a branch. Consider the structure ⟨Vα , ∈, T, <T ⟩ with T and <T being constant symbols with
the axioms satisfying that <T is a tree order on T (as far as Vα is concerned) and T has height
ω and no maximal elements.
Let M ≺ Vα be a countable elementary submodel. Then ⟨T, <T ⟩ ∈ M and M |= “⟨T, <T ⟩
is a tree satisfying the conditions of DC”. Let T ′ be M ∩ T . Then T ′ is a countable subtree of
T , moreover by elementarity, if t ∈ T ′ , then Vα |= “t is not maximal” and therefore M satisfies
the same, so t is not maximal in T ′ . In other words, T ′ has height at least ω and no maximal
elements. But if t ∈ T , then it is impossible that t lies in Tω , since Tω = ∅, so Tω ∩ M = ∅.
Therefore T ′ is a countable subtree of T which also satisfies the assumptions of DC.
By Exercise 6.12 we get that T ′ has a branch B. But now B is also a branch in T , since it
meets every Tn , thus proving DC.
Definition 6.18. We say that W|x| holds if every cardinal is comparable with |x|. As before,
W<|x| means that for every y for which |y| < |x|, W|y| holds.
Exercise 6.15. Show that if W|x| holds, then |x| is an initial ordinal. Namely, x can be well-ordered.
Exercise 6.16. Wℵ0 holds if and only if every Dedekind-finite set is finite.
Exercise 6.17. Suppose that α is a limit ordinal such that for some η < α, there is a sequence
{αγ | γ < η} for which sup{αγ | γ < η} = α. Then DC<ℵα implies DCℵα , AC<ℵα implies ACℵα ,
and if ACℵβ holds and W<ℵα holds, then Wℵα holds.

29
Definition 6.19. Let x be a non-empty set. We say that F is a filter on x, if F ⊆ P (x)
satisfying the following properties:

1. x ∈ F,

2. if a, b ∈ F, then a ∩ b ∈ F,

3. if a ∈ F and a ⊆ b ⊆ x, then b ∈ F.

We will usually require that ∅ ∈ / F. In the case that ∅ ∈ F we say that F is the improper
filter, and we will always mention explicitly when we allow the improper filter in a statement.

Exercise 6.18. Let x be a non-empty set, then Ff in = {a ⊆ x | x \ a is finite} is a filter on x and

it is the improper filter if and only if x is finite.

Definition 6.20. Let F be a filter on x. We say that F is principal if F ∈ F. We say that

T

F is an ultrafilter if for every a ⊆ x either a ∈ F or x \ a ∈ F.

Exercise 6.19. If F is a filter on x and a ⊆ x, then there is a filter F ′ such that F ∪ {a} ⊆ F ′ if
and only if for every b ∈ F, a ∩ b is non-empty.

Exercise 6.20. The intersection of filters is a filter, and the union of a ⊆-increasing sequence of
filters is also a filter. Deduce that if F is a filter on x, with a ⊆ x such that a ∩ b ̸= ∅ for all b ∈ F,
then there is a smallest filter which contains F and a.

Exercise 6.21. F is an ultrafilter x if and only if there is no filter F ′ on x such that F ⊊ F ′ .

Theorem 6.21. If P (x) is well-orderable, then every filter on x can be extended to an ultrafilter.
Proof. Fix a filter F on x. Enumerate P (x) as {aα | α < η}, and by recursion define a ⊆-
increasing sequence of filters: F0 = F, and Fα + 1 is the smallest filter such that Fα ∪ {aα } if
there is such filter, or Fα otherwise. Then Fη is the increasing union of filters on x, therefore
itself is a filter on x, and given any a ⊆ x, there is some α such that a = aα , and therefore
either aα ∈ Fα+1 or its complement is there. Therefore Fη is indeed an ultrafilter.

Exercise 6.22 (*). If every filter can be extended to an ultrafilter, then every set can be linearly
ordered.

Exercise 6.23 (*). If A is an infinite amorphous set, then A cannot be linearly ordered.

Definition 6.22. The Partition Principle, abbreviated as PP states that |x| ≤∗ |y| if and only
if |x| ≤ |y|. In other words, if there is a surjective function from y onto x, then there is an
injective function from x into y.

Exercise 6.24. Show that PP implies that for every x, ℵ(x) = ℵ∗ (x).

Exercise 6.25. Show that PP implies that ≤∗ is anti-symmetric. Namely, there is a Cantor–Bernstein
theorem for the ≤∗ relation on the cardinal. Use this to prove there are no infinite Dedekind-finite
sets.

Remark. It is open whether or not PP implies the Axiom of Choice. As of 2016, this is the oldest
open problem in set theory.

30
Chapter 7

Sets of Ordinals

In this chapter we work in ZFC, that is ZF + AC. Unless stated otherwise.

7.1 Cofinality
Definition 7.1. Let α be an ordinal. A ⊆ α is cofinal (in α) if sup A = α. The cofinality of α
is the least ordinal δ such that there is a cofinal A ⊆ α such that otp(A) = δ. We denote this
as cf(α) = δ.

Using the Mostowski collapse, or rather its inverse, cf(α) = δ if and only if δ is the least
ordinal such that there is an increasing function from δ into α whose range is cofinal in α.

Definition 7.2. We say that α is a regular ordinal if cf(α) = α, and otherwise it is a singular
ordinal.

Exercise 7.1. cf(cf(α)) = cf(α). Namely, cf(α) is always regular.

Exercise 7.2. α is regular if and only if for all A ⊆ α, if |A| < |α| then sup A < α.

Exercise 7.3 (*). If α is regular, then α is a cardinal. But not every cardinal is regular.

As the consequence of these two exercises, cf(α) is always an infinite cardinal. We will
sometimes be interested in this cardinal in the context of cardinal arithmetic, so we will write
cf(ℵ )
things like ℵα α to hint that we are interested in the cardinal arithmetic of these sets, rather
than their ordinal arithmetic.

Exercise 7.4. If there is a function f : δ → α which is not decreasing and rng f is cofinal in α, then
cf(α) = cf(δ). In other terms, if A ⊆ α is cofinal, then cf(otp(A)) = cf(α).

Theorem 7.3. Let α be an ordinal. If α is not a limit ordinal, then ωα is regular; if α is a

limit ordinal, then cf(ωα ) = cf(α).
Proof. For α = 0 we get ω0 = ω, and of course that every finite set of finite ordinals is bounded
below ω. If α is a limit ordinal then {ωβ | β < α} is a cofinal subset of ωα , so by the previous
exercise cf(α) = cf(ωα ).
Finally, if α = β + 1 and A ⊆ ωα has order type η < ωα , then |A| ≤ ℵβ and for every
γ ∈ A, γ < ωα , |γ| ≤ ℵβ . Therefore we can choose suitable injections and prove that | sup A| ≤
ℵβ · ℵβ = ℵβ . And by the very definition of ωα as ωβ+1 we get that sup A < ωα .

31
Remark. It is consistent with ZF that ω1 , and indeed that every limit ordinal, has cofinality ω.
Definition 7.4. We define H(ωα ) to be the set {x | | tcl(x)| < ℵα }.
Exercise 7.5. Show that H(ωα ) is a continuous filtration of V , and conclude that it satisfies a
Reflection theorem.
Exercise 7.6. If ωα > ω is regular, then H(ωα ) satisfies ZFC− , namely ZFC without Power Set.

7.2 Some cardinal arithmetic

As we have seen before, for ℵ numbers the basic cardinal arithmetic is fairly simple:

ℵα + ℵβ = ℵα · ℵβ = ℵmax{α,β} .

Using the axiom of choice, we can make infinite arithmetic well-defined. The reason choice is
needed is that when we want to ensure two infinite unions have the same cardinality, we need
to choose bijections between the sets we unify. If there are finitely many, this is not an issue,
but for infinitely many this can become problematic.
Definition 7.5. We define |ai | as | {{i} × ai | i ∈ I}| and |ai | as | ai |.
P S Q Q
i∈I i∈I i∈I

Exercise 7.7. The definitions of infinite addition and multiplication are well-defined. Moreover, if
|ai | = |a| for all i, then i∈I |ai | = |I| · |a| and i∈I |ai | = |a||I| .
P Q
P
|bi |
Exercise 7.8. |a| = |a||bi | .
Q
i∈I
i∈I

Exercise 7.9 (*). For every sequence of sets if |I| ≥ ℵ0 , then |ai | = |I| · sup{|ai | | i ∈ I}.
P
i∈I

Exercise 7.10. cf(ωα ) = δ if and only if for all |I| < δ, and for all i ∈ I, |Ai | < ℵα , |Ai | < ℵα .
P
i∈I
ℵ
Proposition 7.6. If α ≤ β, then ℵαβ = 2ℵβ .
Proof.
ℵ
  ℵβ
2ℵβ ≤ ℵαβ ≤ 2ℵβ = 2ℵβ ·ℵβ = 2ℵβ .
ℵ ℵ
β
Theorem 7.7 (Hausdorff’s formula). For all α and β, ℵα+1 = ℵα+1 · ℵαβ .
Proof. If α ≤ β, then
ℵ ℵ
  ℵβ ℵ
ℵαβ ≤ ℵα+1
β
≤ 2ℵα ≤ 2ℵα ·ℵβ = 2ℵβ = ℵαβ .

If α > β, then every function from ωβ to ωα+1 is bounded, since ωα+1 is regular. For every
ℵ
η < ωα+1 there are at most ℵαβ functions from ωβ into η, and there are ℵα+1 such η’s and so
the calculation follows.

Theorem 7.8 (König’s lemma). If for all i ∈ I, λi < κi are cardinals. Then
X Y
λi < κi .
i∈I i∈I
Proof. For each i ∈ I, let Bi be a set of size κi and Ai ⊆ Bi a subset of size λi , we may further
assume that for i ̸= j, Bi ∩ Bj = ∅. Take any F : {Ai | i ∈ I} → i∈I Bi , then for every i, the
S Q

set Xi = {F (a)(i) | a ∈ Ai } has cardinality at most λi . Therefore {Bi \ Xi | i ∈ I} is a family

of non-empty sets. Let f be a choice function from this family, then for every i ∈ I, there is no
a ∈ Ai such that F (a) = f as the two must differ on i. Therefore F is not surjective, and the
conclusion follows.

32
Remark. It is consistent with ZF that R is a countable union of countable sets. In such situation
König’s lemma fails as we can take λn = ℵ0 and κn = 2ℵ0 (which is not well-orderable!). Of course
the problem is deeper there: infinite summation and products of cardinals are not well-defined.

Corollary 7.9. κ < κcf(κ) .

Proof. If κ is some ℵα+1 or ω, then this is just a consequence of Cantor’s theorem and basic
cardinal arithmetic. Otherwise, κ is a limit cardinal. Let {λi | i < cf(κ)} be a strictly increasing
sequence of cardinals such that sup{λi | i < cf(κ)} = κ. Let κi = λi+1 , then for all i < cf(κ),
λi < κi . Therefore
κ= κ = κcf(κ) .
X Y Y
λi < κi ≤
i<cf(κ) i<cf(κ) i<cf(κ)

Corollary 7.10. cf(2ℵ0 ) > ℵ0 . In particular ℵω ̸= 2ℵ0 .

Remark. It was shown by Cohen and Solovay that this is in fact the only restriction on the continuum
in ZFC.

Exercise 7.11. Prove that ℵℵω1 = ℵℵω0 · 2ℵ1 .

Definition 7.11. Let κ and λ be cardinals, we define the weak power as

κ<λ = sup{κµ | µ < λ}.1

Definition 7.12. We say that an infinite cardinal κ is a strong limit cardinal if for all λ < κ,
2λ < κ.

Exercise 7.12. Show that if κ is a strong limit cardinal, then it is a limit cardinal.

Definition 7.13. We say that κ is a weakly inaccessible cardinal if it is a regular limit cardinal.
We say that κ is a (strongly) inaccessible cardinal if it is a strong limit and weakly inaccessible
cardinal.

Exercise 7.13. Show that if κ<κ = κ, then κ is a regular cardinal. Show that if κ is inaccessible,
then κ<κ = κ.

Exercise 7.14. Show that if κ is a strong limit cardinal such that κ = ℵκ , then |Vκ | = κ.

Exercise 7.15. Show that if κ is inaccessible, then Vκ |= ZFC. In particular, show that ZFC does
not prove the existence of inaccessible cardinals.

7.3 Clubs and stationary sets

Definition 7.14. Let α be a limit ordinal, and let C ⊆ α. We say that C is a closed set if
whenever β < α and sup(C ∩ β) = β, then β ∈ C. We say that C is unbounded if sup C = α.
If C is both closed and unbounded we say that it is a club set.2

Definition 7.15. We say that S ⊆ α is a stationary set if whenever C ⊆ α is a club, then

S ∩ C ̸= ∅.

Exercise 7.16. If S ⊆ α is a stationary set, then S is unbounded.

1
Here µ is a cardinal, of course, but we can replace it by κ|µ| .
2
Club is an abbreviation for “CLosed and UnBounded”. In some places this is abbreviated as “cub” instead.

33
Exercise 7.17. Show that if C is a club in a limit ordinal α such that cf(α) > ω, then C ∩ Lim is
a club in α as well. And show that if λ is a regular cardinal, then Sλκ = {α < κ | cf(α) = λ} is
stationary. In particular, Sωω2 and Sωω12 are two disjoint stationary subsets of ω2 .

For the remainder of the section, we will always assume that κ is a regular uncountable
cardinal. Whenever we say club or stationary set without qualifications, we will mean as a
subset of κ.
Definition 7.16. We say that a function f : κ → κ is a normal function if it is increasing
and continuous. Namely, f (α) < f (β) whenever α < β and if δ is a limit ordinal, then
f (δ) = sup{f (α) | α < δ}.

Both the term “continuous” and “closed” that we use here are justified topologically when
considering an ordinal as a topological space, using the order topology.
Remark. This definition also makes sense in the context of class functions from Ord to itself. For
example, ordinal arithmetic, as well as the function α 7→ ωα .
Exercise 7.18. C is a club if and only if there is a normal function f such that C = rng f .
Theorem 7.17. S is stationary if and only if for every normal function f , there is some α ∈ S
such that f (α) = α.

We first prove the following lemma.

Lemma 7.18. If f is a normal function, then {α < κ | f (α) = α} is a club.
Proof of Lemma 7.18. The fact that the set is closed is easy. To see it is unbounded, take any
α0 < κ, and define αn+1 = f (αn ) and α = sup{αn | n < ω}. Then

f (α) = sup{f (αn ) | n < ω} = sup{αn+1 | n < ω} = α.

Therefore there is some α ≥ α0 such that f (α) = α.

Proof of Theorem 7.17. If S is stationary, then by the lemma, the set of fixed points is a club
so its intersection with S is non-empty. In the other direction, if S is a non-stationary set, then
there is some normal function f such that rng f ∩ S = ∅, and in particular S does not contain
any fixed points for f .

Exercise 7.19. Prove there is a cardinal µ such that µ = ℵµ . Moreover, show that there is one
which is in fact a strong limit cardinal.
Proposition 7.19. Suppose cf(α) = κ > ω, then there is a continuous function f : κ → α
whose range is cofinal.
Proof. Let g : κ → α be a function witnessing that cf(α) = κ. Define f by recursion:

f (γ) = sup{g(β) + 1, f (β) + 1 | β < γ}.

Note that for cf(α) = ω, the above proposition is trivial, since any cofinal ω sequence is
automatically a continuous function from ω into α. But what this means for the case where
cf(α) > ω is that we can translate statements about clubs in α to statements about clubs in
cf(α).
Exercise 7.20. Suppose that L is a countable first-order language, and let M be a structure in L
whose universe is κ, an uncountable regular cardinal. Then there is a club C such that for all α ∈ C
the substructure of M , Mα whose universe is α, satisfies Mα ≺ M .

34
7.4 The Club filter
Proposition 7.20. The intersection of two clubs is a club.
Proof. Suppose that C and D are clubs, if η < κ such that sup(C ∩D ∩η) = η, then in particular
sup(C ∩ η) = η = sup(D ∩ η). Therefore η ∈ C and η ∈ D, so C ∩ D is closed.
Suppose that η < κ is any ordinal, then we construct α0 = η, α2n+1 is the least ordinal in
C such that α2n < α2n+1 and α2n+2 is the least ordinal in D such that α2n+1 < α2n+2 . These
ordinals exist since neither C nor D is bounded. Let α = sup{αn | n < ω}, then α ∈ C ∩ D.
Therefore C ∩ D is unbounded as wanted.

Exercise 7.21. If S is a stationary set and C is a club, then S ∩ C is stationary.

Theorem 7.21. If γ < κ, and {Cα | α < κ} is a family of clubs, then C = {Cα | α < γ} is
T

also a club.
Proof. Suppose that η < κ is an ordinal, such that C ∩ η is unbounded in η. Then for every
α < γ, Cα ∩ η is unbounded in η. Therefore η ∈ Cα for all α < γ, so η ∈ C.
Suppose that η < κ. Similar to the previous proof, we construct an increasing sequence of
order-type γ · ω such that cγ·n+α ∈ Cα , and c0 > η. If β = sup{cγ·n+α | n < ω, α < γ}, then
easily β ∈ Cα for all α and therefore C is indeed unbounded.

Corollary 7.22. Let F be the filter generated by all the club sets, namely A ∈ F if and only if
A contains a club. Then F is closed under <κ-intersections. Such filter is called a κ-complete
filter.

At the same time, it is clear that the intersection of κ clubs need not be a club itself, just
consider Cα = κ \ α to see that {Cα | α < κ} = ∅. However, we can somewhat correct for
T

this problem.
Definition 7.23. Let γ ≤ κ, the diagonal intersection of {Cα | α < γ} is the following set:
△{Cα | α < γ} = {β < κ | β ∈
\
{Cα | α < β}}.

Easily, if γ < κ, then the diagonal intersection is just the intersection, at least above γ
itself. But for γ = κ this is no longer true. Nevertheless, the following theorem shows that the
situation is still under control.
Theorem 7.24. Suppose that Cα is a club for α < κ, then C = △{Cα | α < κ} is a club.
Proof. Suppose that η < κ such that C ∩ η is unbounded. Then for every α < η, the set
{β ∈ C | α < β < η} is a subset of Cα , therefore Cα ∩ η is unbounded for every α < η. As each
Cα is a club, η ∈ Cα for all α < η and so η ∈ C.
Suppose that η < κ is any ordinal, pick α0 ∈ C0 such that α0 > η. Suppose αn was chosen,
take αn+1 to be an ordinal in {Cβ | β < αn } such that αn+1 > αn . Let α = sup{αn | n < ω}.
T

Then for every β < α, Cβ ∩ α is unbounded below α, since for all large enough n, αn ∈ Cβ .
Therefore α ∈ C, and therefore C is unbounded.

Corollary 7.25 (Fodor’s lemma). Suppose that S is a stationary set and f : S → κ is

regressive, i.e. f (α) < α for all α ∈ S. Then there is some β such that {α | f (α) = β} is
stationary.
Proof. Suppose not, then for every α < κ, there is some Cα which is a club and disjoint from
{β | f (β) = α}. Let C = △{Cα | α < κ}, by the theorem, C is a club and in particular
non-empty. But if α ∈ C ∩ S, then for all β < α, f (α) ̸= β. This is impossible since f was
regressive.

35
Definition 7.26. We say that a filter F on κ is normal if whenever f is a regressive function
on κ, it is constant on some S such that κ \ S ∈
/ F.

Exercise 7.22. Let κ be a regular uncountable cardinal. If F is a normal filter such that for every
α, κ \ α ∈ F, then F contains the club filter.

Exercise 7.23 (Solovay’s theorem) (**). If S is a stationary subset of κ, then there is a partition
of S into {Sα | α < κ} such that Sα is stationary for all α.

Exercise 7.24. (*) Suppose that a train has ω1 + 1 stations. It embarks from station 0 empty.
When it stops at station α, if it has any passengers, one of them will get off. Then countably many
new passengers will get on the train, and it continues to the next station. How many passengers are
on the train when it reaches its final destination, station ω1 ?

Proposition 7.27. Suppose that κ > ℵ0 is weakly inaccessible, then κ = ℵκ .

Proof. Let α be such that κ = ℵα . Since α is a limit ordinal, we have that cf(κ) = cf(α) = κ. So
α ≥ κ, but since there is a club in κ of order type α, namely {ωη | η < α}, we get equality.

Exercise 7.25. Show that if κ > ℵ0 is weakly inaccessible, then the set {α | α = ℵα } is a club below
κ. Show that if κ is a strongly inaccessible cardinal, then the set {α | α is a strong limit cardinal}
is also a club below κ.

Theorem 7.28. Let κ be an ordinal such that cf(κ) > ℵ0 . Suppose that {α | cf(α) = α} is
a stationary set of κ. Then κ is a weakly inaccessible cardinal, and it is not the first weakly
inaccessible cardinal.
Proof. First note that κ is a cardinal, since it is a limit of cardinals; and it is in fact a limit
cardinal, since otherwise κ is the successor of some λ, and then {α | λ < α < κ} does not
contain any cardinals, but it is a club in κ.
Since κ is a limit cardinal, the set of cardinals below κ is a club, therefore the set of limit
cardinals is a club, and by the assumption, it contains a regular cardinal—in fact many regular
cardinals—which is to say that there is some weakly inaccessible cardinal below κ. Finally, if
δ = cf(κ) < κ, then there is a function f : δ → κ which is continuous and unbounded, so rng f
is a club. Look at the club C = rng(f ↾ Lim), if α ∈ C, then cf(α) < δ. Therefore every regular
cardinal in C must be at most δ, so the set of regular cardinals is not stationary after all.

Definition 7.29. An uncountable cardinal with the property that regular cardinals (equiva-
lently, inaccessible cardinals) below it form a stationary set is called a weakly Mahlo cardinal.

36
Chapter 8

Inner models of ZF

8.1 Inner models

Definition 8.1. We say that a class M is an inner model if it is transitive, Ord ⊆ M and for
every axiom φ of ZF, φM holds.

Definition 8.2. A class M is called almost universal if whenever x is a set, and x ⊆ M , then
there is some y ∈ M such that x ⊆ y.

Proposition 8.3. If M is almost universal, then M is a proper class.

Proof. Suppose otherwise, then M ⊆ M , and therefore for some y ∈ M we have that M ⊆ y.
This means that y ∈ y, a contradiction to Foundation.

Definition 8.4. Bounded Separation, or ∆0 -Separation, is the schema of Separation restricted

only for ∆0 formulas. Similar definitions can be made for Replacement as well as more complex
classes of formulas (e.g. Σ1 -Replacement).

Theorem 8.5. If M is a transitive class which is almost universal and satisfies ∆0 -Separation,
then M is an inner model of V .
Proof. First we claim that Ord ⊆ M , to see this let α be such that α ⊆ M , then by almost
universality there is some y ∈ M such that α ⊆ y. By ∆0 -Separation, and the fact that Ord is
definable by a ∆0 formula, y ∩ Ord ∈ M . As α ⊆ y, either y ∩ Ord = α in which case α ∈ M
or there is some γ ∈ y ∩ Ord such that α < γ, and then by the transitivity of M we get that
α ∈ M.
We start verifying the axioms: Extensionality, Empty Set, Infinity and Foundation follow
from the fact that M is a transitive class and ω ∈ M .
Next, we claim: If x ∈ M , then P M (x) = P (x)∩M ∈ M . Recall P M (x) = {u ∈ M | u ⊆ x},
so clearly P M (x) = P (x) ∩ M . Suppose now that x ∈ M , then by almost universality there is
some y ∈ M such that P M (x) ⊆ y. Consider the ∆0 formula, u ⊆ x (recall this is a shorthand
for ∀v(v ∈ u → v ∈ x)), then y ′ = {u ∈ y | u ⊆ x} ∈ M as it was obtained by ∆0 -Separation
from y, using x as a parameter. But as P M (x) ⊆ y, it means that y ′ = P M (x). Therefore M
satisfies the Power Set axiom.
It remains to prove that Replacement holds, which will imply that Separation holds as well.
For this we first prove that for all α, VαM = M ∩ Vα : for α = 0 this is just ∅; for successor steps
this holds from the Power Set axiom in M :
M
Vα+1 = P M (VαM ) = P (VαM ) ∩ M = P (Vα ∩ M ) ∩ M = P (Vα ) ∩ M = Vα+1 ∩ M ;

37
and for limit cases this follows from the fact that {Vβ ∩ M | β < α} = {Vβ | β < α} ∩ M .
S S

Let φ(u, v, p̄) be a formula such that for some p̄, x ∈ M , M |= (∀u ∈ x)∃!vφ(u, v, p̄). Then
 M
this means that V |= (∀u ∈ x)∃!vφ(u, v, p̄) . By the Reflection theorem there is some β
 M Vβ
large enough such that p̄, x ∈ Vβ and V |= (∀u ∈ x)∃!vφ(u, v, p̄) .
 B
It is not hard to check that if A and B are transitive classes, then ψ A is equivalent to
 VβM
ψ A∩B . Therefore, the Reflection theorem gives us that V |= (∀u ∈ x)∃!vφ(u, v, p̄) . But
being a ∆0 sentence where all the parameters (p̄, x and VβM ) are in M , removing the ∃!v gives
us the following ∆0 formula:

ψ(v, x, p̄, X) = x ∈ X ∧ p̄ ∈ X ∧ (∃u ∈ x)φX (u, v, p̄).

Placing VβM as X gives us the set {v | (∃u ∈ x)φ(u, v, p̄)} ∈ M as wanted.

Proposition 8.6. If P M (x) = P (x) for all x ∈ M , then M = V .

Proof. Since M ⊆ V , it is enough to prove that V ⊆ M , and for that it is enough to verify that
for all α, Vα ⊆ M . We will show that Vα = VαM , which is certainly enough.
For α = 0, this is trivial. At successor steps,
M
Vα+1 = P M (VαM ) = P M (Vα ) = P (Vα ) = Vα+1 .

Finally, for limit steps, VαM = {VβM | β < α} = {Vβ | β < α} = Vα .

S S

Proposition 8.7. If α is a cardinal in V , it is a cardinal in M . And ω1M ≤ ω1 .

Proof. Recall that φ(α) stating that α is cardinal is a Π1 formula, since α ∈ M , and M is
transitive, it is downwards absolute. In the other direction, if α < ω1M , then M |= α is countable.
But as we saw being a countable set is a Σ1 property, which is therefore upwards absolute, so
α is countable in V . So the least uncountable ordinal of M cannot be larger than the least
uncountable ordinal in V .

Theorem 8.8 (Balcar–Vopěnka). Suppose that M, N are two inner models such that for
every α, P M (α) = P N (α) and M |= AC, then M = N .
Proof. First note that having the same sets of ordinals means also having the same sets of pairs
of ordinals. And so on. This is because we can define a bijection between pairs of ordinals and
ordinals (using Theorem 4.16).
First we will show that M ⊆ N . For x ∈ M , first fix a bijection between tcl({x}) and
some ordinal α. This bijection induces a binary relation E on α which codes the ∈ relation on
tcl({x}). By the above remark, E ∈ N . Now ⟨α, E⟩ is a set-like, extensional and well-founded
structure, so we can collapse it. But its Mostowski collapse must be equal to tcl({x}). Therefore
tcl({x}) ∈ N , and so x ∈ N .
In the other direction, we prove by ∈-induction that every x ∈ N also lies in M . Let x ∈ N
such that x ⊆ M , let y ∈ M such that x ⊆ y (e.g. VαM for a suitable α). Let f : y → β be some
bijection in M , then it is also in N . But now f ”x is a set of ordinals in N , and therefore it lies
in M . Since both f and f ”x are in M , it follows that x ∈ M as well.

Remark. The Axiom of Choice plays a crucial role in this proof. It is consistent that there are two
models of ZF with the same sets of ordinals, but not with the same sets of sets of ordinals.

38
Theorem 8.9. Suppose that F is a function defined by recursion from a function G which was
Σn for n ≥ 1, then F is Σn as well.
Proof. Note that being an ordinal is a ∆0 formula, and F (α) = y if and only if there is a
function f whose domain is α coding the construction of F (α).

Corollary 8.10. Suppose that L is a first order language, then the formula φ(x, L) stating that
x is a term, a formula, or a sentence in the language L is ∆0 with parameters L. In particular
being a formula is absolute for infinite transitive classes. Moreover, if A is a structure for L
and σ is an assignment, then A |=σ x is a ∆0 formula with parameters A and σ.
Proof. If a transitive class (or set) is infinite, then it contains all the finite ordinals. Note that
ω is ∆0 -definable (it is either the set of ordinals, or an element). The proof above works for any
recursive definition such as being a term, etc.

We can agree that if L is a countable language, then we can code it using finite ordinals.
This means that terms, etc. are just elements of Vω , being recursively constructed as sequences
of sequences of sequences, etc.
Remark. This is an important place to make the distinction between the meta-theory and the theory.
Namely, when we write V |= φ, this is a statement made in the meta-theory, whereas when A is an L
structure and ψ some sentence in L, then A |= ψ is a statement about specific sets made inside V .
This issue was also present in the Reflection theorem, where we make the move from the formulas
in our meta-theory to formulas inside V , and the problem is that the meta-theory formulas (as well
as the satisfaction relation) are not objects of V , instead they are objects of the meta-theory.
We can do that, however, because we can faithfully “recreate” the formal logic of the meta-
theory inside the theory. While it is possible that V disagrees with its meta-theory on what are the
natural numbers, which may cause an excess of formulas, inference rules, and other objects which
are effectively coded by formulas, we still get a faithful copy of the meta-language inside V .

This means that if M is a transitive class such that Vω ⊆ M , and L, A ∈ M with A being
an L structure, then M |= “B is a definable subset of A” if and only if B is a definable subset
of A.

8.2 Gödel’s constructible universe

Definition 8.11. Let M be a transitive set,

Def(M ) = {B ⊆ M | B is definable with parameters in ⟨M, ∈⟩}.

Exercise 8.1. If M is finite, then Def(M ) = P (M ).

Exercise 8.2. If M is infinite and well-orderable, then |M | = | Def(M )|.
Exercise 8.3. If M is transitive, then Def(M ) is transitive, and M ∈ Def(M ).
Theorem 8.12. If M is a transitive set, then Def(M ) is the smallest transitive set such that
M ∈ Def(M ) and Def(M ) satisfies ∆0 -Separation.
Proof. It is clear that any transitive N satisfying ∆0 -Separation with M ∈ N will also include
very definable subset of M , so it is enough to show that Def(M ) indeed satisfies this property.
Suppose now that φ(x, p) is a ∆0 formula and let A ∈ Def(M ) and p ∈ Def(M ).1
1
We want to show the proof for the case where parameters are allowed, as it gives better insight, but one
parameter is plenty.

39
 Then there is some definition (over M ) for A, say φA (x, q̄) with q̄ ∈ M , and a definition
φp (x, q̄) for p, also with q̄ ∈ M . Note that we can assume that the parameters are the same by
allowing repetition and ignoring unneeded parameters. We prove by induction on the complexity
of φ that {x ∈ A | φ(x, p)} ∈ Def(M ).

• Suppose that φ is atomic, then it has the form x ∈ p or p ∈ x or x = p. All three are
easily translated to formulas defining subsets of M .
• For negation, conjunction, disjunction and implication this is just complement, intersec-
tion, union and subsets, and Def(M ) is clearly closed under all of these.
• Finally, for quantifiers we have (∀u ∈ p)φ(u, x) or (∃u ∈ p)φ(u, x, p) or (∀u ∈ x)φ(u, x, p)
or (∃u ∈ x)φ(u, x, p). Let ψ(u, x, ȳ) denote the formula which defines the set definable
from φ(u, x, p) (note that here x acts as a parameter). First we consider the case with
∃u ∈ p:
n o
A (x, q̄) ∧ (∃u(φp (u, q̄) ∧ ψ (u, x, ȳ)) ,
{x ∈ A | (∃u ∈ p)φ(u, x)} = x ∈ M φM M

clearly this results in a definable subset of M , and the case for ∀u ∈ p is similar. The
cases with p as a parameter of φ are proven using the induction hypothesis.
n o
{x ∈ A | (∃u ∈ x)φ(u, p)} = x ∈ M φM
A (x, q̄) ∧ (∃u ∈ x)ψ (u, x, ȳ) .
M

Again, the case for ∀u ∈ x is proved similarly.

Exercise 8.4 (**). The function Y = Def(X) is a ∆1 function.
Definition 8.13. We define by recursion Gödel’s Constructible hierarchy.

1. L0 = ∅.
2. Lα+1 = Def(Lα ).
3. Lα = {Lβ | β < α} for α ∈ Lim.
S

Let L be {Lα | α ∈ Ord}. By x ∈ L we mean ∃α(x ∈ Lα ), and V = L to mean that ∀x(x ∈ L).
S

Exercise 8.5. The function α 7→ Lα is a ∆1 function. And for all α ≥ ω, Lα satisfies ∆0 -Separation.
Theorem 8.14. L satisfies ZF.
Proof. It is enough to verify that the conditions of Theorem 8.5 hold. Easily Ord ⊆ L, and by
the existence of the constructible hierarchy, L is an almost universal class. It remains to check
that L satisfies ∆0 -Separation.
For readability purposes, we will prove ∆0 -Separation for formulas without parameters.
Suppose that φ(u) is a ∆0 formula, and let x ∈ L. Let y = {u ∈ x | φ(u)}, then y ⊆ L, so there
is some α such that x ∈ Lα and y ⊆ Lα . As Lα is transitive and φ is a ∆0 formula, φ(u) holds
if and only if Lα |= φ(u). But this means that y is definable over Lα using x as a parameter, so
y ∈ Lα+1 . Therefore L satisfies all the axioms of ZF.

Theorem 8.15. If M is an inner model of V , then LM = L.

Proof. By induction we will show that LMα = Lα . For α = 0 this is clearly true; as is the case
for α ∈ Lim:
α = β | β < α} = {Lβ | β < α} = Lα .
[ [
LM {LM

Suppose that α = β + 1 and LM

β = Lβ , then by the absoluteness of the Def function, we get
that Def (Lβ ) = Def(Lβ ).
M

40
Corollary 8.16. The following are quick and important corollaries from the theorem.

1. LL = L.

2. L |= V = L.

3. L is the smallest inner model.

4. If an inner model M satisfies V = L, then M = L.

Exercise 8.6. If M is a transitive set satisfying ∆0 -Separation, and Vω ∈ M , then LM = LM ∩Ord .

Exercise 8.7. For all α ∈ Lim, (Lα )L = Lα = LLα . So if α ∈ Lim, then Lα |= V = L.

Exercise 8.8. Show that for an unbounded class of ordinals α, Vα ̸= Lα . Prove that if V = L
holds, then there is a closed and unbounded class of ordinals for which Vα = Lα .

8.3 The properties of L

We wish to investigate the construction of L and the properties of the sets inside L, known as
constructible sets.

Definition 8.17. Let x be a set in L. We define rankL (x) = α if x ∈ Lα+1 but x ∈

/ Lα .

Exercise 8.9. Show that the formula φ(x, α) meaning that rankL (x) = α is a ∆1 formula.

Theorem 8.18. ACL .

Proof. We first define by induction a well-ordering ≺α on Lα . We begin by fixing a well-ordering
of the formulas in the language of set theory of order type ω, and let φn denote the nth formula
in the enumeration. For x ∈ Lα+1 , let n(x) denote the least n such that φn can be used to
define x over Lα .
For α + 1 we define, x ≺α+1 y if and only if n(x) < n(y), or n(x) = n(y) and the parameters
used to define x appear in the lexicographic ordering induced by ≺α before the parameters used
to define y.
For α ∈ Lim, define x ≺α y if and only if rankL (x) < rankL (y), or rankL (x) = rankL (y) = β
and x ≺β y.
Finally, define x <L y as we define ≺α for the limit case. This is a definable well-ordering
of L which defines a bijection between L and Ord, and therefore every set can be well-ordered,
so AC holds in L.

Theorem 8.19 (Gödel’s Condensation Lemma). If α is a limit ordinal and M ≺ Lα , then

there is some β such that M ∼ = Lβ .
Proof. Since M is a well-founded structure, we can collapse it to a transitive set, N . We claim
that N = Lβ . Since Lα satisfies ∆0 -Separation, so must N . Therefore LN = Lβ for β = N ∩Ord.
But Lα |= V = L, so N |= V = L as well, and therefore N = Lβ .

Exercise 8.10. If α is infinite, then |Lα | = |α|.

41
 Recall Cantor’s Continuum Hypothesis (CH), is the statement 2ℵ0 = ℵ1 .
Theorem 8.20. CHL .
Proof. Assume V = L. Suppose that A ⊆ ω, then there is some α ∈ Lim such that A ∈ Lα .
Let M be a countable elementary submodel of Lα such that A ∈ M , and let π : M → Lβ be the
Mostowski collapse of M . Then π(A) = A, and β is a countable ordinal. Therefore if A ⊆ ω,
then A ∈ Lω1 . In particular, there are at most |Lω1 | = ℵ1 subsets of ω.
Exercise 8.11 (*). Show that L |= GCH, namely for every α, 2ℵα = ℵα+1 .
Exercise 8.12 (*). Show that H(ωαL )L = LωαL .
We finish this section with a remark about generalizations of L:
1. If A is any set, then L(A) is defined the same way as L, only L0 is now tcl({A}). It can be
shown that L(A) is the smallest inner model in which A is an element; and choice holds
there if and only if there is a definable well-ordering of A.
2. If A is any set, then L[A] is defined by augmenting the first-order structure over which
we take Def to include a predicate interpreted as A. Namely, L0 = ∅, and Lα+1 is the
set of all definable sets in the structure ⟨Lα , ∈, A ∩ Lα [A]⟩. We can show that L[A] is the
smallest inner model of ZFC satisfying A ∩ M ∈ M .
Both of these models have many uses throughout set theory. And one can show that if A ⊆ L,
then L(A) = L[A].

8.4 Ordinal definable sets

Definition 8.21. We say that x is ordinal definable if there is a formula in the language of set
theory φ such that x = {u | φ(u, α1 , . . . , αn )} where α1 , . . . , αn ∈ Ord.
Theorem 8.22. {x | x is ordinal definable} is a class. We shall denote it by by OD.
Proof. If x is ordinal definable and φ(u, α) is the formula defining it, by the reflection theorem
there is some β large enough such that x, α ∈ Vβ and Vβ reflects the fact that φ defines x (with
the parameter α). Therefore if x ∈ OD, then there is some β and a formula in the (internal)
language of set theory ψ(u, α) such that Vβ |= x = {u | ψ(u, α)}. On the other hand, if x is
ordinal definable in some Vβ by φ(u, α), then there is a formula ψ(u, α, β) stating that there
exists y which is a transitive set, closed under the power set operation, and its ordinals are
β—i.e. y = Vβ —such that in y, x is ordinal definable using some formula and α as a parameter.
So indeed x is in OD.
Definition 8.23. We denote by HOD the class {x | tcl({x}) ⊆ OD}, that is the class of all sets
which are hereditarily ordinal definable.
Exercise 8.13 (*). OD = {Def({Vβ | β < α}) | α ∈ Ord}.
S

Exercise 8.14. Show that HOD is an inner model.

Exercise 8.15 (*). Show there is a definable function from Ord onto HOD.
Remark. The model HOD is not robust as the model L. It can be that HODHOD ̸= HOD, for
example, or that M is an inner model, but HODV ⊊ HODM . It is always the case that if V is a
model of ZFC, then there is a larger model W such that V = HODW . And many other strange
properties which do not happen with the case of L. Interestingly, HOD is compatible with the failure
of CH.

42
Chapter 9

Some combinatorics on ω1

9.1 Aronszajn trees

Definition 9.1. We say that a tree T is an Aronszajn tree if it has height ω1 , every level of T
is countable, and there are no cofinal branches.

To avoid trivialities, we only consider normal trees, meaning given α < ω1 and t ∈ T , there
is an element in Tα comparable with t, and every node has at least two successors.

Theorem 9.2. There exists an Aronszajn tree.

Proof. Let T ∗ be the tree whose nodes are order embeddings of countable ordinals into bounded
sets of Q, ordered by end-extension. Then T ∗ has height ω1 , no cofinal branch; but almost all
the levels of T ∗ are in fact uncountable. We will refine T ∗ to an Aronszajn tree.
We define by recursion the levels of T . Suppose that the levels Tβ were defined for all β < α,
and that the following condition holds:

∀γ < β ∀x ∈ Tγ ∀q > sup x ∃y ∈ Tβ : x < y ∧ sup y ≤ q. (*)

The condition states, in other words, that every embedding of γ into Q, and any strict upper
bound of that embedding, can be extended to an embedding of β into Q with the same upper
bound. So Tβ is rich enough to have witnesses for extensions of embeddings into arbitrarily
small intervals.
Let T0 = {∅}, the only thing it can be. And if Tα was defined and t ∈ Tα , we define its
successors in Tα+1 to be t⌢ q for all q > sup t. Take Tα+1 to be the set of all these successors, for
all t ∈ Tα , then Tα+1 is a countable union of countable sets, and easily (∗) continues to hold.
Suppose that α is a limit ordinal, we need to decide which branches of the possible branches
we can add to Tα will be taken. For every x ∈ {Tβ | β < α}, and every q > sup x, we can
S

construct recursively a chain {xn | n < ω} such that x < xn , sup xn < q, and {dom xn | n < ω}
is cofinal in α. For every x ∈ {Tβ | β < α} and every q > sup x, choose a sequence like that,
S

and define Tα as the union of all these chosen sequences. It is easy to see that (∗) still holds,
and that Tα is still countable.

Exercise 9.1. Let T be the above tree as constructed in L. If there is a cofinal branch in T , then
ω1L < ω1 .

Exercise 9.2. In T as constructed above there is an uncountable antichain.

43
9.2 Diamond and Suslin trees
Definition 9.3. We say that T is a Suslin tree if it has height ω1 , but every antichain is
countable.

Exercise 9.3. Show that a Suslin tree is an Aronszajn tree. In other words, if every antichain is
countable, then every chain is countable.

We want to prove that there is a Suslin tree. However, ZFC cannot prove that a Suslin tree
exists. We need additional assumptions, one such assumption is the following axiom.

Definition 9.4. A ♢-sequence is a sequence ⟨Aα | α < ω1 ⟩ such that:

1. Aα ⊆ α.

2. For every A ⊆ ω1 , the set {α | A ∩ α = Aα } is stationary.

The axiom ♢ asserts that there exists a ♢-sequence.

Proposition 9.5. ♢ implies CH.

Proof. If A is a subset of ω, then there is some α > ω such that A = A ∩ ω1 = Aα . This defines
an injection from P (ω) into ω1 , and therefore CH holds.

Theorem 9.6. ♢ holds in L.

Proof. We work in L, and construct recursively for α < ω1 a sequence of pairs ⟨Aα , Cα ⟩ such that
Aα , Cα ⊆ α, and Cα is closed and unbounded in α. For α = 0 we can only take Aα = Cα = ∅.
For α + 1, let Cα+1 = Aα+1 = α + 1. For a limit α, define the pair as follows:

⟨Aα , Cα ⟩ is the least pair in <L such that Aα , Cα ⊆ α with Cα a club in α, and
for all β ∈ Cα , Aα ∩ β ̸= Aβ . If no such pair exists, take Aα = Cα = α.

We claim that ⟨Aα | α < ω1 ⟩ is a ♢-sequence. Assume otherwise, and let ⟨A, C⟩ be the <L -least
pair such that A ⊆ ω1 and C ⊆ ω1 is a club such that for all α ∈ C, A ∩ α ̸= Aα . Since the
sequence, A and C were all definable from <L , both our sequence and ⟨A, C⟩ are elements of
Lω2 , by condensation arguments. Let M be a countable elementary submodel of Lω2 ,1 then
by the virtue of definability, ⟨Aα | α < ω1 ⟩ and ⟨A, C⟩ are both elements of M . Let Lγ be
the transitive collapse of M and let π : M → Lγ denote the isomorphism. Note that ω1 ∩ M is
L
necessarily an ordinal δ, and that δ = ω1 γ , namely π(ω1 ) = δ; so π(A) = A∩δ and π(C) = C ∩δ.
Moreover, since for α < δ, π(α) = α, it follows that π(Aα ) = Aα . Therefore we get that
π(⟨Aα | α < ω1 ⟩) = ⟨Aα | α < δ⟩. Now Lγ satisfies that ⟨A ∩ δ, C ∩ δ⟩ is the <L -least pair
satisfying that C ∩δ is a club in δ and for all β ∈ C ∩δ, A∩δ ∩β = A∩β ̸= Aβ . By elementaritiy,
this is true in Lω2 , and therefore in L itself. But this means that Aδ = A ∩ δ. On the other
hand, δ ∈ C, since C is a club in ω1 and unbounded below δ. And this is a contradiction.

Theorem 9.7. If ♢ holds, then there is a Suslin tree.

Proof. Fix a ♢-sequence ⟨Aα | α < ω1 ⟩. We will construct a tree T by recursion, and for
simplicity we will assume the underlying set of the tree is ω1 itself, and that for every α,
{Tβ | β < α} is an ordinal (it matters very little which, but we can assume that it is α · ω).
S

The root of the tree, of course, is {0}. Suppose that we constructed Tα , let Tα+1 contain
some countably many ordinals in such way that every node in Tα has at least two successors.
1
M is necessarily not transitive, why?

44
Let α be a limit ordinal, and T<α = {Tβ | β < α} defined. If Aα is a maximal antichain in
S

T<α , let Tα be a suitable extension which preserves the maximality of Aα ; namely, every node
in Tα lies above an element of Aα . Otherwise, pick any suitable countable level, such that every
x ∈ T<α has an extension in Tα . In either case we can use Lemma 9.8.
Let T = {Tα | α < ω1 }. We claim that T is a Suslin tree. Suppose that A is a maximal
S

antichain in T , then ⟨T, <T , A⟩ is a first-order structure whose domain is ω1 . Therefore there
is a club C ⊆ ω1 such that for α ∈ C, T ∩ α = T<α = α and A ∩ α is a maximal antichain in
T<α . Using the ♢-sequence, there is a stationary subset S such that for α ∈ S, A ∩ α = Aα .
Pick α ∈ S ∩ C, then A ∩ α = Aα is a maximal antichain in T<α . But since we chose Tα to be
such that A ∩ α is still a maximal antichain in T<α+1 , we get that if t ∈ T , then t is comparable
with an element from Tα and therefore comparable with an element of A ∩ α. This means that
A ∩ α is in fact maximal in T , so A = Aα and therefore countable.

Lemma 9.8. Suppose that α is a countable limit ordinal and ⟨T, <T ⟩ is a countable tree of
height α which is normal. If A ⊆ T is a maximal antichain, then we can extend T by adding
one more countable level such that A remains a maximal antichain.
Proof. The added level must be obtained by realizing a point at the end of a cofinal branch
through T . For every t ∈ T such that there is some a ∈ A for which a <T t; for every such t,
choose a cofinal branch—which exists due to the normality assumption—and realize it.2 Then
Tα that was added is the realization of only countably many branches; every point in Tα extends
a point which extends some a ∈ A, so A is still maximal; and the extended tree is still normal
for obvious reasons.

2
Namely, add an upper bound to that cofinal branch.

45
Chapter 10

Coda: Games and determinacy

Definition 10.1. Suppose that A ⊆ ω ω , we define the game G(A) to be the game where two
players take turns choosing natural numbers for ω turns. This defines a sequence x ∈ ω ω such
that Player I played x(2n) and Player II played x(2n + 1). We say that Player I won if x ∈ A,
and otherwise Player II won.

Player I x(0) x(2) x(4) . . .

Player II x(1) x(3) ...
We call the sequence x the outcome of the game. And we will use xI and xII to denote the
sequences of the moves made by Players I and II respectively in the game.
Definition 10.2. We say that σ is a strategy for Player I for a game G(A) if σ is a function
from ω <ω to ω, such that if x is the outcome of the game, then x(2n) = σ(x ↾ 2n). We say that
a strategy is a winning strategy if it guarantees victory, namely if x(2n) = σ(x ↾ 2n) for all n,
then x ∈ A. A strategy for Player II is defined similarly (here the victory is when x ∈
/ A).

If A is a set such that G(A) has winning strategy for one of the players, we say that A is
determined.
Of course, at most one player can have a winning strategy. But is there always such a
strategy?
Proposition 10.3. If A ⊆ ω ω is countable, then Player II has a winning strategy.
Proof. Let A = {an | n < ω}, then on the 2n + 1-th move, Player II simply plays an (2n + 1) + 1,
thus guaranteeing that if x is the outcome of the game, then x ̸= an for all n < ω.

Theorem 10.4. There is a game without a winning strategy.

Proof. Let {σα , τα | α < 2ℵ0 } be an enumeration of all the winning strategies such that σα is a
winning strategy for Player I and τα is a winning strategy for Player II (not for the same game,
of course).
Define by recursion a set {aα , bα | α < 2ℵ0 }. Let aα be an outcome of a game where σα was
used by Player I, but aα ∈
/ {bβ | β < α}. We can find such aα , since the possible games where
Player I played using σα has cardinality 2ℵ0 . Similarly, let bα be an outcome obtained from a
game where Player II played using τα and bα ∈ / {aβ | β < α}.
Now we claim that X = {bα | α < 2ℵ0 } is not determined. For every α, aα is an outcome of
a game where Player I used σα , and aα ∈
/ X, therefore σα cannot be a winning strategy for I;
and bα is an outcome of a game where II used τα , but bα ∈ X, so τα is not a winning strategy
for II.

46
 Nevertheless, if A is an open, closed or even Borel, in the product topology on ω ω , then it
is determined even if we assume choice .This means that the set A defined in the proof above
is somewhat “pathological”.

Definition 10.5. The Axiom of Determinacy (AD) states that every set is determined.

Corollary 10.6. AD implies ¬AC.

But not all is lost, and we can still get some choice. The following is a very typical proof
using AD.

Theorem 10.7. Assume ZF + AD. Then ACℵ0 (ω ω ) holds.

Proof. Let {Xn | n < ω} be a countable family of non-empty sets such that Xn ⊆ ω ω . We
define the game where I first chooses n, and then II has to construct (in the odd-indexed turns)
an element of Xn . Namely, II wins if the outcome x is such that xII ∈ Xx(0) .
Clearly, I cannot possibly win, since once I played the first turn, II can choose a sequence
from the relevant Xn , and just play it. By AD it has to be the case that II has a winning
strategy, τ . Then we can define f (Xn ) to be xII , where x is the outcome of τ and I playing n,
and then only 0.

Proofs of this flavor are the staple of determinacy proofs. We can use such arguments to
show that every subset of ω ω is very “nice” from a topological and measure theoretic point of
view.

Remark. One might wonder about the consistency of AD. Unlike with the case of AC, where ZFC
is consistent if ZF is consistent, to prove that ZF + AD is consistent we need to assume additional
hypotheses which exceed what ZFC can prove by a lot. So for example, if we assume ZF + AD, then
ω1 is a strongly inaccessible cardinal in L, and in fact we can say much much more.

Exercise 10.1. AD implies that ω1 is regular.

Exercise 10.2 (*). AD implies that there are no free ultrafilters on ω.

Exercise 10.3 (**). AD implies that the club filter on ω1 is an ultrafilter.

47