MODELS OF SET THEORY

STEFAN GESCHKE

Contents
1. First order logic and the axioms of set theory 2
1.1. Syntax 2
1.2. Semantics 2
1.3. Completeness, compactness and consistency 3
1.4. Foundations of mathematics and the incompleteness theorems 3
1.5. The axioms 4
2. Review of basic set theory 5
2.1. Classes 5
2.2. Well-founded relations and recursion 5
2.3. Ordinals, cardinals and arithmetic 6
3. The consistency of the Axiom of Foundation 8
4. Elementary Submodels, the Reflection Principle, and the Mostowski
Collapse 10
5. Constructibiltiy 13
5.1. Definability 13
5.2. The definition of L and its elementary properties 14
5.3. ZF in L 15
5.4. V = L? 15
5.5. AC and GCH in L 17
6. Forcing 20
6.1. Partial orderings, dense sets and filters 20
6.2. Generic extensions 22
6.3. The forcing relation 24
6.4. ZFC in M [G] 28
7. CH is independent of ZFC 30
7.1. Forcing CH 30
8. Forcing ¬CH 31
8.1. The countable chain condition and preservation of cardinals 32
8.2. Nice names and the size of 2ℵ0 33
9. Martin’s Axiom 35
9.1. Martin’s Axiom and Souslin’s Hypothesis 35
9.2. The Baire Category Theorem and Martin’s Axiom 37
9.3. More consequences of MA 40
9.4. Iterated forcing 42
9.5. Long iterations 45
References 47

1
2 STEFAN GESCHKE

1. First order logic and the axioms of set theory

1.1. Syntax. The language L of set theory is the first-order language with the
binary relation-symbol ∈. That is, the language L consists of the formulas over the
alphabet {∧, ¬, ∃, (, ), ∈, =} ∪ Var, where Var is a countably infinite set of variables.
By recursion on the length, we define what a formula is.
If x and y are variables, then x ∈ y and x = y are formulas. Such formulas
are called atomic. If ϕ and ψ are formulas, then so are ¬ϕ and (ϕ ∧ ψ). If ϕ is a
formula and x is a variable, then ∃xϕ is a formula, too.
We will frequently omit parentheses in a formula if this does not lead to any am-
biguity. Also, we will add parentheses if this improves readability. ∀xϕ abbreviates
¬∃x¬ϕ and (ϕ ∨ ψ) abbreviates ¬(¬ϕ ∧ ¬ψ). As usual, we will freely use other
abbreviations like ⊆, →, and ∃x ∈ y inside a formula.
Exercise 1.1. Let x and y be variables and let ϕ and ψ be formulas. Write out
what the following abbreviations stand for:
x ⊆ y, ϕ → ψ, ∃x ∈ yϕ.
This obviously depends on your intuition of what formulas are supposed to mean.
We discuss the meaning of formulas in Subsection 1.2. You may use all the previ-
ously defined abbreviations.
A free occurrence of a variable x in a formula ϕ is an occurence of x outside the
scope of any quantifier ∃x. (Recall that ∀x is just an abbreviation.) If x occurs
freely in ϕ, then x is a free variable of ϕ. We write ϕ(x1 , . . . , xn ) instead of just ϕ
in order to indicate that all the free variables of ϕ are listed among x1 , . . . , xn and
that the variables x1 , . . . , xn are pairwise distinct. A sentence is a formula without
free variables.

1.2. Semantics. We briefly discuss the intended meaning of formulas. A structure

for the language of set theory is a set X together with a binary relation E. In
the following, let X be a set and E a binary relation on X. Small letters from
the beginning of the alphabet, possibly with index, such as a, a1 , . . . , an , b denote
elements of X. Small letters from the end of the alphabet, possibly with index,
such as x, x1 , . . . , xn , y are typically variables of the language of set theory.
Let ϕ(x1 , . . . , xn ) be a formula. The intended meaning of ϕ[a1 , . . . , an ] is the
statement about a1 , . . . , an that is obtained by replacing every free occurence of
xi by ai , i = 1, . . . , n, and interpreting the symbol = by actual equality and the
symbol ∈ by the relation E.
To make this precise, by recursion on the length of formulas we define whether
(X, E) satisfies ϕ[a1 , . . . , an ]. We write (X, E) |= ϕ[a1 , . . . , an ] for “(X, E) satisfies
ϕ[a1 , . . . , an ]”. We abbreviate (x1 ∈ x2 )[a1 , a2 ] by a1 ∈ a2 and (x1 = x2 )[a1 , a2 ] by
a1 = a2 .
(1) (X, E) |= a1 ∈ a2 iff a1 Ea2 .
(2) (X, E) |= a1 = a2 iff a1 = a2 . Here the first = is an element of the alphabet
of the language of set theory, the second = has the usual meaning.
(3) (X, E) |= ¬ϕ[a1 , . . . , an ] iff (X, E) 6|= ϕ[a1 , . . . , an ].
(4) (X, E) |= (ϕ ∧ ψ)[a1 , . . . an ] iff
(X, E) |= ϕ[a1 , . . . , an ] and (X, E) |= ψ[a1 , . . . , an ].
(5) Let ϕ(x, y1 , . . . , yn ) be a formula. Then the free variables of ∃xϕ are among
y1 , . . . , yn . Hence we can write (∃xϕ)[y1 , . . . , yn ] instead of ∃xϕ. Now
(X, E) |= (∃xϕ)[b1 , . . . , bn ]
 MODELS OF SET THEORY 3

iff there is an element a of X such that

(X, E) |= ϕ[a, b1 , . . . , bn ].
1.3. Completeness, compactness and consistency. A set of sentences is a
theory. If Σ is a theory and (X, E) satisfies each of the sentences in Σ, we say that
(X, E) is a model of Σ. Similarly, if (X, E) satisfies a sentence ϕ, we say that (X, E)
is a model of ϕ. A sentence ϕ follows from Σ if every model of Σ is a model of ϕ.
In this case we write Σ |= ϕ.
There is also a notion `. Σ ` ϕ means that ϕ is formally provable from Σ.
The following statement about the connection between |= and ` is easily the most
important theorem about first order logic.
Theorem 1.2 (Completeness Theorem). If Σ is a theory and ϕ is a sentence, then
Σ |= ϕ iff Σ ` ϕ.
We are going to use the Completeness Theorem freely, without referring to it
explicitly.
Since formal proofs have a finite length, a formal proof of ϕ from Σ only uses
finitely many sentences from Σ. Hence, using the completeness theorem, we obtain
Corollary 1.3 (Compactness Theorem). If Σ is a theory and ϕ a sentence, then
Σ |= ϕ iff there is a finite theory Σ0 ⊆ Σ such that Σ0 |= ϕ.
A theory Σ is consistent if it does not lead to a contradiction, i.e., if there is no
sentence ϕ such that both ϕ and ¬ϕ are provable from Σ.
Corollary 1.4. A theory Σ is consistent iff it has a model.
Proof. Suppose Σ is not consistent. Let ϕ be a sentence such that both ϕ and ¬ϕ
follow from Σ. Now, if M = (X, E) is a model of Σ, then M |= ϕ and M |= ¬ϕ.
But this contradicts the definition of M |= ¬ϕ.
Now suppose that Σ does not have a model and let ϕ be any sentence. Since Σ
does not have a model, every model of Σ satisfies both ϕ and ¬ϕ. Hence, ϕ and
¬ϕ both follow from Σ and thus, Σ is inconsistent. 
We will frequently use
Corollary 1.5. A theory Σ is consistent iff every finite subset of Σ has a model.
Exercise 1.6. Give a proof of Corollary 1.5.
1.4. Foundations of mathematics and the incompleteness theorems. Prac-
tically all of mathematics can be formulated in the language of set theory. Hence,
all one has to do in order to come up with a sound foundation of mathematics is
to find a system of axioms, i.e., a theory, with the following properties:
(1) The axioms reflect our intuitive understanding of the concept “set”.
(2) Practically all mathematical statements that are generally regarded as prov-
able actually follow from the axioms.
(3) The axioms are consistent.
Note that the demands (1) and (2) are slightly vague, but there is common
agreement that there is a system of axioms that satisfies (1)–(3), namely Zermelo-
Fraenkel Set Theory (ZF) together with the Axiom of Choice (AC, ZF+AC=ZFC).
Since (1) and (2) are open to different interpretations, all we can say is that most
mathematicians agree that ZFC is a suitable system of axioms. Even the mathe-
maticians who are not familiar with ZFC almost never exceed the strength of ZFC
in their proofs. In fact, most of mathematics can be done in far weaker systems of
arithmetic.
4 STEFAN GESCHKE

(3) is a very precise statement. In particular, if (3) holds for a specific system of
axioms, as a true mathematical statement it should be provable from the axioms.
However, if the system of axioms was inconsistent, then it would prove its own
consistency. It follows that a proof of (3) would actually not increase our believe
in the consistency of the system of axioms.
In fact, Gödel has shown that for every system of axioms satisfying (1) and
(2) (in a precise sense), (3) is not provable from the axioms. This is the Second
Incompleteness Theorem.
Gödels First Incompleteness Theorem says that if ZFC (or any similar theory)
is consistent, then there are sentences ϕ such that neither ϕ nor ¬ϕ follow from it.
Such a sentence ϕ is called independent over ZFC. The main goal of this course is
to show that the Continuum Hypothesis (CH) is independent over ZFC.
1.5. The axioms. The axioms of ZFC are the following (we use some obvious
abreviations):
(1) Set Existence.
∃x(x = ∅)
(2) Extensionality.
∀x∀y(∀z(z ∈ x ↔ z ∈ y) → x = y)
(3) Foundation.
∀x(x 6= ∅ → ∃y ∈ x∀z ∈ x(z 6∈ y))
(4) Separation (really a scheme of axioms). For every formula
ϕ(x, y1 , . . . , yn ) without a free occurrence of y,
∀y1 . . . ∀yn ∀z∃y∀x(x ∈ y ↔ x ∈ z ∧ ϕ(x, y1 , . . . , yn ))
is an axiom.
(5) Pairing.
∀x∀y∃z(x ∈ z ∧ y ∈ z)
(6) Union.
∀F ∃A∀Y ∈ F (Y ⊆ A)
(7) Replacement (again a scheme of axioms). For every formula
ϕ(x, y, y1 , . . . , yn ), without a free occurence of y,
∀y1 . . . ∀yn ∀A(∀x ∈ A∃!yϕ(x, y, y1 , . . . , yn )
→ ∃Y ∀x ∈ A∃y ∈ Y ϕ(x, y, y1 , . . . , yn ))
is an axiom. Here ∃! means “there is exactly one”.
(8) Infinity.
∃x(∅ ∈ x ∧ ∀y ∈ x(y ∪ {y} ∈ x))
(9) Power set.
∀x∃y∀z(z ⊆ x → z ∈ y)
(10) Choice.
∀x∃R(R is a well-ordering of x)
For the Axiom of Choice recall that a binary relation ≤ on a set X is well-
ordering if ≤ is a linear order on X and every non-empty set Y ⊆ X has a minimal
element with respect to ≤. (See also Definition .)
Exercise 1.7. Write out the formulas ∃!xϕ, x = ∅, and x = y ∪ z.
Exercise 1.8. Write out the formula “R is a well-ordering of x”.
Note that since Separation and Replacement are schemes and not just single
axioms, ZFC consists of infinitely many axioms.
 MODELS OF SET THEORY 5

2. Review of basic set theory

2.1. Classes. Assume that we live in a universe of sets that satisfies all the axioms
of ZFC. This is the usual framework in which mathematics takes place. A class C
is a collection of all sets with a certain property. More precisely, if ϕ(x, y1 , . . . , yn )
is a formula and b1 , . . . , bn are sets, then C = {a : ϕ[a, b1 , . . . , bn ]} is a class, the
class defined by ϕ(x, y1 , . . . , yn ) and the parameters b1 , . . . , bn . The class C = {a :
ϕ[a, b1 , . . . , bn ]} is identified with a set c iff ϕ[a, b1 , . . . , bn ] is equivalent to a ∈ c. A
class that does not correspond to a set in this way is a proper class.
We now consider the same situation, but inside a structure M = (X, E) for the
language of set theory that we can look at from the outside. The variables of our
language range over elements of the structure, i.e., they are interpreted by elements
of the structure, the sets of M . If b1 , . . . , bn are elements of X and ϕ(x, y1 , . . . , yn )
is a formula, then C = {a ∈ X : M |= ϕ[a, b1 , . . . , bn ]} is a class of M , the class
defined by the formula ϕ(x, y1 , . . . , yn ) with the parameters b1 , . . . , bn . Note that
looking at M from the outside, C is a set.
C can correspond to an element of X in the following way: It might happen that
for some c ∈ X we have that for all a ∈ M ,
M |= a ∈ c ⇔ M |= ϕ[a, b1 , . . . , bn ],
i.e., C = {a ∈ X : aEc}. This is the situation in which we identify C with c. If R
is a binary relation and b is a set, then extR (b) = {a : aRb} is the extension of b
with respect to R or the R-extension of b. Note that the Axiom of Extensionality
just says that two sets are the same iff they have the same ∈-extension. This can
be written as ∀x(ext∈ (x) = x). In other words, every set is uniquely determined
by its ∈-extension.
The most important class is V , the class of all sets. V is a proper class. We will
soon define other important proper classes.
2.2. Well-founded relations and recursion. Let C be a class and R a binary
relation on C. R is set-like if for all a ∈ C, extR (a) is a set. R is well-founded if
every non-empty subset S of C has an R-minimal element, i.e., there is a ∈ S such
that S ∩ extR (a) = ∅.
The Axiom of Foundation says that ∈ is well-founded. The Axiom of Extension-
ality implies that ∈ is set-like.
Exercise 2.1. Let R be a set-like well-founded relation on a class C. Show that
every non-empty subclass of C has an R-minimal element.
Hint: This might be harder than it seems at first sight. For a set S ⊆ C define
[
ExtR (S) = S ∪ {extR (a) : a ∈ S}
and [
Ext∞
R (S) = {ExtnR (S) : n ∈ N}.
Now, if D is a non-empty subclass of C, pick a ∈ D and consider the set D ∩
Ext∞R ({a}). (Why is this a set? You may take recursion on the natural numbers
for granted.)
Because of the usefulness of this exercise, whenever we consider a well-founded
relation, we will automatically assume that it is set-like.
Theorem 2.2 (Transfinite Induction). Let C be a class, R a well founded relation
on C, ϕ(x, y1 , . . . , yn ) a formula and b1 , . . . , bn ∈ C. If for all a ∈ C it holds that
if ϕ[b, b1 , . . . , bn ] for all b ∈ extR (a), then ϕ[a, b1 , . . . , bn ],
then for all a ∈ C, ϕ[a, b1 , . . . , bn ].
6 STEFAN GESCHKE

Exercise 2.3. Give a proof of Theorem 2.2.

Theorem 2.4 (Recursion Theorem). Let C be a class, R a well-founded relation
on C and F : V → V a function. Then there is exactly one function G : C → V
such that for all a ∈ C,
(∗) G(a) = F (a, G  extR (a)).
Proof. We first show uniqueness. Let G and G0 be functions on C that both satisfy
(∗) and assume that they are not the same. By Exercise 2.1 there is a ∈ C which
is R-minimal with the property that G(a) 6= G0 (a). But now by (∗),
G(a) = F (a, G  extR (a)) = G0 (a),
a contradiction.
We now show the existence of G. An initial segment of C is a set S ⊆ C such
that if aRb and b ∈ S, then a ∈ S. Note that for any a ∈ C, Ext∞ R ({a}) is an initial
segment of C. In fact, Ext∞ R ({a}) is the smallest initial segment containing a.
Let G be the collection (class) of all functions g that are defined on an initial
segment of C such that for all a ∈ dom(g) the equation (∗) holds for g. Note that
trivially, the empty function is an element of G. By the proof of uniqueness of
G, any two functions g, g 0 ∈ G agree on the intersection of their domains, which
happens to be an initial Ssegment of C.
It follows that G = G is a function. If a ∈ dom(G), then there is g ∈ G
such that a ∈ dom(g). Now G  extR (a) = g  extR (a) and hence G satisfies (∗)
for all a ∈ dom(G). Now assume that dom(G) 6= C. Let a be R-minimal with
a 6∈ dom(G). Let
g = G  Ext∞ ∞
R (extR (a)) = G  (ExtR ({a}) \ {a}).
We extend g to g 0 : Ext∞ 0 0
R ({a}) → V by letting g (a) = F (a, g). Now g ∈ G and
0
a ∈ dom(g ). Therefore a ∈ dom(G), a contradiction.
Note that if C is a proper class, then so is G. But a class is a collection of sets
that is definable by a formula. What is the formula that defines G?
A pair (b, c) of sets is in G iff there is a set g that happens to be a function that
is defined on an initial segment of C and satisfies (∗) for all a in its domain such
that b ∈ dom(g) and g(b) = c. This is tedious but certainly possible to express in
our formal language of set theory. 
2.3. Ordinals, cardinals and arithmetic. A set a is transitive if for all b ∈ a
and all c ∈ b, c ∈ a.
Exercise 2.5. Let a be transitive. Show that a ∪ {a} and P(a) are transitive.
Exercise 2.6. Let X be a transitive set. Show that (X, ∈) satisfies the Axiom
of Extensionality. Find a set X such that (X, ∈) does not satisfy the Axiom of
Extensionality.
An ordinal is a transitive set that is linearly ordered by ∈. The (proper) class
of all ordinals is denoted by Ord. If α and β are ordinals, then α < β iff α ∈ β.
Ord is linearly ordered by <, i.e., by ∈. The Axiom of Foundation implies that <
is a well-ordering of Ord. If α is an ordinal, then α ∪ {α} is the smallest ordinal
that is bigger than α. α ∪ {α} is frequently denoted by α + 1, where the + is the +
of ordinal arithmetic, not of cardinal arithmetic, which we will discuss below. An
ordinal of the form α + 1 is a successor ordinal. An ordinal different from 0 = ∅
that is not a successor ordinal is a limit ordinal.
Two sets a and b are of the same size if there is a bijection between them. A
cardinal is an ordinal α that is not of the same size as any smaller ordinal. The
(proper) class of cardinals is denoted by Card. The cardinality of a set a is the
 MODELS OF SET THEORY 7

least ordinal that is of the same size as a. It follows from the Axiom of Choice that
every set has a cardinality, and the cardinality of a set is always a cardinal. Every
infinite cardinal is a limit ordinal.
If κ and λ are cardinals, then κ + λ is the size of ({0} × κ) ∪ ({1} × λ) and κ · λ
is the size of κ × λ. κλ is the size of the set of all functions from λ to κ.
Theorem 2.7. If κ and λ are cardinals and at least one of them is infinite, then
κ × λ = κ + λ = max(κ, λ).
The finite ordinals happen to be cardinals and are denoted by 0, 1, 2, . . . . Recall
that 0 = ∅, 1 = {0}, 2 = {0, 1} and so on. The set of all finite ordinals is ω, the
first infinite ordinal. For every ordinal α the α-th infinite cardinal is denoted by
ℵα . In other words, the first infinite cardinals are ω = ℵ0 , ℵ1 , ℵ2 and so on.
ℵ0 is the size of the set N = ω. 2ℵ0 is the size of the set R of real numbers which
is the same as the size of P(ω).
Theorem 2.8. For every cardinal κ, 2κ > κ.
By this theorem, the first candidate for 2ℵ0 is ℵ1 . The statement 2ℵ0 = ℵ1 is
known as the Continuum Hypothesis (CH). The main goal of this course is to show
that CH can be neither proved nor refuted from the axioms of ZFC.
Note that Theorem 2.8 implies that for every cardinal there is a larger cardinal.
If κ is a cardinal, then the smallest cardinal (strictly) larger than κ is denoted by
κ+ . The Generalized Continuum Hypothesis (GCH) is the statement that for every
infinite cardinal κ we have 2κ = κ+ .
Let α be an ordinal. A ⊆ α is cofinal (in α) iff for all β < α there is γ ∈ A such
that β ≤ γ. The cofinality cf(α) of an ordinal α is the least size of a cofinal subset
of α. A cardinal κ is regular if κ = cf(κ), otherwise it is singular. Cofinalities of
ordinals turn out to be regular cardinals.
Exercise 2.9. Show that ℵ1 is regular S and ℵω is singular.
Hint: If A Sis cofinal in ℵ1 , then ℵ1 = A. (Why?) If A ⊆ ℵ1 is countable, what
is the size of A? For the singularity of ℵω use the fact that the union of a set of
cardinals, i.e., the supremum of the set of cardinals, is again a cardinal.
The only limitation to the possible values of 2ℵ0 is the following strengthening
of Theorem 2.8.
Theorem 2.10. Let κ be an infinite cardinal. Then κ < cf(2κ ).
8 STEFAN GESCHKE

3. The consistency of the Axiom of Foundation

In this section we show that ZF is consistent provided that ZF without the
Axiom of Foundation is consistent. This relative consistency proof is technically
very easy but nevertheless already uses important ideas that can be used for other
consistency proofs as well.
Let ZF− denote ZF without the Axiom of Foundation.

Theorem 3.1. If ZF− is consistent, then so is ZF.

Proof. Assume that ZF− is consistent. Then this theory has a model (V, ∈). We
use (V, ∈) in order to construct a model of ZF. To do this, we pretend to live inside
V . The ordinals are the ordinals of V , the cardinals are the cardinals of V , and so
on. Note that we should adjust our definition of ordinals so that we arrive at the
same notion as in the context of Foundation: an ordinal is a transitive set that is
well-ordered (rather than just linearly ordered) by ∈. It can be shown without the
Axiom of Foundation that the class Ord of ordinals is well-ordered by ∈.
For α ∈ Ord let Vα be defined by
(i) V0 := ∅
S P(Vα )
(ii) Vα+1 :=
(iii) Vα := β<α Vβ , if α is a limit ordinal.
Using transfinite induction we show that (Vα )α∈Ord is a strictly ⊆-increasing se-
quence of transitive
S sets.
Let WF = α∈Ord Vα , i.e., let WF be the class consisting of all sets a such that
for some ordinal α, a ∈ Vα . For every a ∈ WF let rk(a) denote the least ordinal
such that a ∈ Vα+1 . The ordinal rk(a) is the rank of a. Note that every ordinal α
is an element of W F with rk(α) = α.
We show that (W F , ∈) is a model of ZF. Set Existence is obviously satisfied. It
follows from the transitivity of the Vα that W F is transitive. Hence, W F satisfies
the Axiom of Extensionality. The Axiom of Separation follows directly from the
construction of the Vα . Namely, if a is an element of Vα , then so is every subset of
a. The Axioms of Infinity, Union, and Pairing are easily checked.
For the Axiom of Replacement let F be a function, i.e., a class of pairs in W F
such that for all a ∈ W F there is at most one b ∈ W F such that (a, b) ∈ F . Now
consider the function rk ◦F . For every a ∈ W F , F [a] und (rk ◦F )[a] are sets in
V since (V, ∈) satisfies Replacement. We have to find a superset of F [a] in W F .
It follows from Separation the F [a] is a set in W F . Let α := sup((rk ◦F )[a]).
Now F [a] ⊆ Vα+1 and Vα+1 ∈ Vα+2 ⊆ W F . It follows that (W F , ∈) satisfies
replacement.
It remains to show that (W F , ∈) satisfies the Axiom of Foundation. Let a ∈ W F .
Let b ∈ a be of minimal rank. Then b is an ∈-minimal element of a since for all
c, d ∈ W F with c ∈ d we have rk(c) < rk(d). 

While this proof looks convincing at first sight, it does have a couple of problems.
First of all, it is a bit confusing to denote a model of ZF− by (V, ∈), since we don’t
assume ∈ to be the real ∈-relation and also the real V is not a set, while structures
for the language of set theory are assumed to be sets. On the other hand, for
someone living inside the structure, the elements of the structure do form the class
of all sets and the binary relation of the structure is simply the ∈-relation. Looking
at the structure from the outside, it should be denoted by M = (X, E) or something
similar. This is what we will do from now on.
The class W F of M , really a formula with a single free variable defining the
class, has been tacitly identified with the subset of X the consists of all a ∈ X that
 MODELS OF SET THEORY 9

satisfy this formula in the structure M . Then we showed that this subset of X with
the restriction of E to it is a model of ZF.
It is possible to completely ingnore the world outside M in the proof of Theorem
3.1. In this case the notation (V, ∈) for M is again appropriate. We again define
the class W F . But now the statement “W F is a model of ZF” does not make sense
anymore, since W F is a class and not a set. Still, it is possible to formalize that
(W F , ∈) is a model of a certain sentence ϕ.
Let C be a class and ϕ a formula. The relativization of ϕ to C is the formula
ϕC that is obtained by replacing every quantifier ∃x that occurs in ϕ by ∃x ∈ C.
(Recall that ∀x is just an abbreviation and therefore we do not have to consider
this quantifier here.) Of course, here ∃x ∈ Cψ is just an abbreviation for a more
complicated formula that depends on the formula that defines C. Now “(W F , ∈)
is a model of ZF” can be expressed by saying that for every axiom ϕ of ZF, ϕW F
holds (in V ).
There is one subtle problem if we want to show ϕW F for every axiom ϕ of ZF.
Consider for example the axiom Pairing. It looks like we have to show for all
a, b ∈ W F that {a, b} is an element of W F . This is easy since by Pairing in V ,
{a, b} is a set in V as well and it easy to check that in fact, {a, b} ∈ W F . Something
similar is true for Power Set and Union.
Exercise
S 3.2. Let a, b ∈ W F . Show that {a, b} ∈ W F . Also, show that P(a) and
a are elements of W F .
It is, however, not totally obvious that the set that satisfies the definition of
{a, b} in V also satisfies this definition
S in W F . Similarly, we have to show that the
sets that satisfy the definition of a and of P(a) in V satisfy the same definition in
W F . Luckily, these definitions are simple enough that an element of W F satisfies
the respective definition in V if and only if it satisfies that definition relativized to
WF.
We introduce absoluteness in order to deal with this problem explicitly.
Definition 3.3. Let C be a class and ϕ(x1 , . . . , xn ) a formula. The formula ϕ is
absolute over C if the following holds:
∀x1 , . . . , xn ∈ C(ϕ(x1 , . . . , xn ) ↔ ϕC (x1 , . . . , xn ))
A function (possibly a proper class) is absolute over C if the formula that defines
it is absolute over C.
Obviously, every formula that does not have any quantifiers (i.e., Boolean com-
binations of atomic formulas) of is absolute over every class. Quantifiers of the form
∃x ∈ y are bounded. A formula ϕ is ∆0 if all quantifiers of ϕ are bounded.
Lemma 3.4. ∆0 -formulas are absolute over transitive classes.
Let C be a class that satisfies a certain fragment ZF∗ of ZF, i.e., assume that
ϕ holds for every ϕ ∈ ZF∗ . Let ϕ be a formula and suppose that ZF∗ implies that
C

ϕ is equivalent to a ∆0 -formula. Then ϕ is absolute over transitive classes that

satisfy ZF∗ .
It follows that intersections, unions, unordered pairs, ordered pairs, the empty
set and so on are absolute over transitive classes that satisfy all the axioms of ZF−
except possibly Power Set and Infinity.
Now suppose that C is a transitive class that satisfies enough of ZF to show the
recursion theorem. Then the function α 7→ Vα is absolute over C in the sense that
the formula ϕ(x, y) that says that y is an ordinal α and x is an element of Vα is
absolute over C.
10 STEFAN GESCHKE

From this absoluteness of the Vα it follows that W F is a model of V = W F .

By transfinite induction over ∈ it can be shown that assuming ZF− , V = W F is
equivalent to the Axiom of Foundation.
Exercise 3.5. Show that ZF implies that V = W F

4. Elementary Submodels, the Reflection Principle, and the

Mostowski Collapse
Let (M, E) be a set with a binary relation E. We will often identify M with
(M, E). If E is clear from the context, this is not going to lead to confusion. This is
actually quite common throughout mathematics: we seldomly distinguish between
the field (R, 0, 1, +, ·) of real numbers and the set R.
Now N ⊆ M is an elementary substructure or an elementary submodel of M if
for every formula ϕ(x1 , . . . , xn ) in the language of set theory and all a1 , . . . , an the
following holds:
(N, E) |= ϕ(a1 , . . . , an ) ⇔ (M, E) |= ϕ(a1 , . . . , an )
Note that if we write (N, E) we really mean (N, E ∩ N 2 ).
The Incompleteness Theorems imply that there cannot be a set N (in V ) such
that (N, ∈) is an elementary submodel of (V, ∈). This is Tarski’s “undefinability
of truth”. In other words, there are no sets over which all formulas are absolute.
However, we will soon see that for a given finite sets of formulas it is possible to
construct even transitive sets such that all of the finitely many formulas under
consideration are absolute over the transitive set.
We will use this in the following way: In order to prove the consistency of
ZFC+¬CH we pretend that there is a transitive set M such that (M, ∈) is a model
of ZFC. Using M we construct another transitive set N such that (N, ∈) satisfies
ZFC but not CH.
Now, if ZFC+¬CH failed to be consistent, then, by the Compactness Theorem,
there is already a finite subset of ZFC+¬CH that leads to a contradiction. But in
order to prove that these finitely many sentences hold in (N, ∈), we only need that
(M, ∈) satisfies a certain finite subset of ZFC. And for finitely many sentences there
are transitive sets M such that these finitely many sentences are absolute over M .
We start with a lemma that will save us some work.
Lemma 4.1. Let C be a class and ϕ1 , . . . , ϕn a sequence of formulas that is closed
under taking subformulas. Recall that only ∃ is part of our language, ∀ is an abbre-
viation. Then the formulas ϕ1 , . . . , ϕn are absolute over C if for all ϕi (y1 , . . . , ym )
of the form ∃xϕj (x, y1 , . . . , yn ) we have
∀y1 , . . . , ym ∈ C(∃xϕj (x, y1 , . . . , ym ) → ∃x ∈ Cϕj (x, y1 , . . . , ym )).
Exercise 4.2. Give the proof of Lemma 4.1. Use induction over the length of ϕi .
Theorem 4.3 (Reflection Principle). Let W : Ord → V be a function with the
following properties:
(i) For all α, β ∈ Ord with α < β we have S W (α) ⊆ W (β).
(ii) If γ is a limit ordinal, then W (γ) = α<γ W (α).
S
(iii) V = α∈Ord W (α).
For every sequence ϕ1 , . . . , ϕn of formulas and all α ∈ Ord there is β ∈ Ord with
β > α such that the ϕj are absolute over W (β).
Proof. Let ϕ1 , . . . , ϕn be formulas and α ∈ Ord. We may assume that the sequence
ϕ1 , . . . , ϕn is closed under taking subformulas. By Lemma 4.1 it is sufficient to find
β > α such that for all ϕi (y1 , . . . , ymi ) of the form ∃xϕj (x, y1 , . . . , ymi ) we have
(∗) ∀y1 , . . . , ymi ∈ W (β)(∃xϕj (x, y1 , . . . , ymi ) → ∃x ∈ W (β)ϕj (x, y1 , . . . , ymi )).
 MODELS OF SET THEORY 11

For every i ∈ {1, . . . , n} we define a function Gi as follows: Suppose ϕi (y1 , . . . , ymi )

is of the form ∃xϕj (x, y1 , . . . , ymi ). Let Gi (b1 , . . . , bmi ) = 0 if there is no a such
that ϕj (a, b1 , . . . , bmi ) holds in V . If there is a such that ϕj (a, b1 , . . . , bmi ) holds in
V , then let Gi (b1 , . . . , bmi ) be the least ordinal α such that there is such an a in
W (α). Note that the existence of α follows from (iii).
Now we choose a sequence (βk )k∈ω as follows: Let β0 = α. Suppose we have
defined βk for some k ∈ ω. Let βk+1 be the least ordinal > βk such that for all
ϕi (y1 , . . . , ymi ) and all b1 , . . . , bmi in W (βk ), Gi (b1 , . . . , bmi ) < βk+1 .
Let β = supk∈ω βk . Now it is easily checked that for each ϕi (y1 , . . . , ymi ) and
all b1 , . . . , bmi ∈ W (β), Gi (b1 , . . . , bmi ) < β. Hence, by the choice of the Gi and by
Lemma 4.1, all the ϕi are absolute over W (β). 

Observe that the function α 7→ Vα satisfies all the assumptions of Theorem 4.3.
In particular, if V satisfies ZFC, then for every finite list ϕ1 , . . . , ϕn of axiom in
ZFC there are arbitrarily large α such that (Vα , ∈) satisfies ϕ1 , . . . , ϕn .
A proof very similar to the proof of Theorem 4.3 yields
Theorem 4.4 (Löwenheim-Skolem Theorem, downward). Let (M, E) be a struc-
ture for the language of set theory. For every X ⊆ M there is an elementary
submodel N ⊆ M of M such that X ⊆ N and |N|≤|X| +ℵ0 .
Proof. As in the proof of Theorem 4.3 it is enough to find N ⊆ M with X ⊆ N
and |N |=|X | +ℵ0 such that for every existential formula ϕ(x, y1 , . . . , yn ) and all
b1 , . . . , bn ∈ N with
M |= ∃xϕ(b1 , . . . , bn )
there is a ∈ N such that
M |= ϕ(a, b1 , . . . , bn ).
For every formula ϕ(x, y1 , . . . , yn ) we define a function fϕ : M n → M such that
for all b1 , . . . , bn ∈ M we have
M |= ϕ(fϕ (b1 , . . . , bn ), b1 , . . . , bn )
provided
M |= ∃xϕ(b1 , . . . , bn ).
The fϕ are called Skolem functions.
For every X ⊆ M let sk(X) be the Skolem hull of X, i.e., the closure of X
under all the functions fϕ . Since there are only countably many formulas and
hence only countably many Skolem functions, for every X ⊆ M , |sk(X)|≤|X| +ℵ0 .
Clearly, N = sk(X) satisfies all existential statements with parameters in N that
are satisfied in M . Hence, N is an elementary submodel of M . 

Now, if V satisfies all axioms of ZFC, we can carry out the following construc-
tion. Let Σ be a finite collection of axioms of ZFC. By the Reflection Principle
there is α such that all ϕ ∈ Σ are absolute over Vα . Since (V, ∈) satisfies Σ, (Vα , ∈)
is a model of Σ as well. By the Löwenheim-Skolem Theorem, Vα has a count-
able elementary submodel N . This shows that every finite collection of axioms of
ZFC has a countable model (assuming that V satisfies ZFC). We will proceed by
constructing countable transitive models of finite parts of ZFC.
Theorem 4.5 (Mostowski Collapse). Let C be a class and R a well-founded re-
lation on C (recall that we assume well-founded relations to be set-like). If R is
extensional, i.e., if two elements a and b of C agree iff extR (a) = extR (b), then
there are a transitive class D and an isomorphism µ : (C, R) → (D, ∈).
12 STEFAN GESCHKE

Proof. We define µ by recursion on R. For every a ∈ C let

µ(a) = {µ(b) : b ∈ extR (a)}.
By the Recursion Theorem, µ is a well-defined function from C to V . Let D =
{µ(a) : a ∈ C}. The function µ is the Mostowski collapsing function and D is the
Mostowski collapse of (C, R).
We first show that µ is 1-1. Suppose it is not and let a ∈ C be R-minimal
such that for some b ∈ C we have a 6= b and µ(a) = µ(b). By the definition of µ,
µ(a) = {µ(c) : c ∈ extR (a)}. Since µ(a) = µ(b),
{µ(c) : c ∈ extR (a)} = {µ(c) : c ∈ extR (b)}.
But since a is an R-minimal counter-example to the injectivity of µ, we can conclude
that extR (a) = extR (b). Now the extensionality of R implies a = b, a contradiction.
Clearly, if a, b ∈ C are such that aRb, then µ(a) ∈ µ(b). On the other hand, if
µ(a) ∈ µ(b), then, since µ is 1-1, a ∈ extR (b), i.e., aRb. This shows that µ is an
isomorphism.
It remains to show that D is a transitive class. Let a ∈ D and b ∈ a. Choose
c ∈ C such that a = µ(c). By the definition of µ, b is of the form µ(d) for some
d ∈ extR (c). But then d ∈ C and thus b = µ(d) ∈ D. 
Our main application of the Mostowski Collapse is
Corollary 4.6. Every finite fragment of ZFC has a countable transitive model.
Proof. Let Σ be a finite fragment of ZFC. We may assume that Σ contains the
Axiom of Extensionality. We have already argued that Σ has a countable model of
the form (N, ∈). Since N satisfies the Axiom of Extensionality, ∈ is extensional on
N . Therefore (N, ∈) is isomorphic to a structure (M, ∈) where M is transitive. But
since (N, ∈) and (M, ∈) are isomorphic, they satisfy the same sentences. Clearly,
M is countable. 
Exercise 4.7. Let C be a class on which ∈ is extensional. Suppose T is a transitive
(in V ) subclass of C. Let D be the Mostowski collapse of (C, ∈) and let µ be the
collapsing function. Show that µ is the identity on T . In particular, T ⊆ D.
 MODELS OF SET THEORY 13

5. Constructibiltiy
In ZF we will define a class L that satisfies ZFC+GCH. Recall that GCH is the
statement
∀κ ∈ Card(κ ≥ ℵ0 → 2κ = κ+ ).
L is Gödel’s universe of constructible sets. Since we want to show the consistency
of the Axiom of Choice (AC) with ZF, we have to be careful not to use AC in the
following arguments. We will explicit indicate uses of AC in certain places that are
not essential for the theory.

5.1. Definability. The construction of L will be very similar to the construction

of W F in that we define a hierarchy (Lα )α∈Ord using some restricted power set
operation.
First of all we recall that formulas can be regarded as certain sets, namely finite
sequences of elements from our alphabet where the alphabet consists of sets. I.e.,
we code the characters of the alphabet by certain sets, say “(” by 0, “)” by 1, “=”by
2 and so on.
For a finite sequence ϕ of characters in the real world let dϕe denote its character-
by-character translation into the finite sequence (a set) of sets that we used to
code the respective characters. dϕe is the Gödelization of ϕ. Using the Recursion
Theorem we can now write down a formula fml(x) in the real world that expresses
the fact that x is a finite sequence of characters from our alphabet that happens to
be a formula in the language of set theory. We can then prove for every formula ϕ
in the real world that dϕe actually satisfies fml.
Again using the Recursion Theorem we can write down a formula (in the real
world) sat(x, y, z) that says:
(i) There is n ∈ ω such that y is a formula with n free variables,
(ii) there are a1 , . . . , an ∈ x such that z = (a1 , . . . , an ) and
(iii) the structure (x, ∈) satisfies the formula y if the free variables of y are
interpreted by a1 , . . . , an .
By induction over the complexity of formulas we can show that for every formula
ϕ(x1 , . . . , xn ), every set M and all a1 , . . . , an ∈ M it holds that
ϕM (a1 , . . . , an ) ↔ sat(M, dϕe, (a1 , . . . , an ))
Hence the relation |= can now be regarded as a definable class (where we consider |=
only for structures of the form (M, ∈), but the general case (M, E) can be handled
practically in the same way).
Note that in the definition of the formula sat we used the fact that we can code
the finitely many sets a1 , . . . , an into a single set, namely the n-tuple (a1 , . . . , an ).
This is necessary since the formula sat has to have a fixed number of free variables.
We can easily write down a formula that defines the set ZFC of formulas. We can
then prove that every sentence in the real-world ZFC has its Gödelization in V ’s
version of ZFC. If we are a little bit more general in the definition of sat, so that sat
applies to structures of the form (M, E), we can write down the formula Con(x) that
says that x is a set of sentences that has a model. Gödel’s Second Incompleteness
Theorem now states that Con(ZFC) is not provable from ZFC. Note that the first
ZFC is the set satisfying the definition of ZFC in V where the second ZFC is the
real-world (meta-mathematical) ZFC.
Now let M be a set. A set P ⊆ M is definable (in (M, ∈)) if there is a formula
ϕ(x, y1 , . . . , yn ) such that for some parameters b1 , . . . bn ∈ M we have
P = {a ∈ M : (M, ∈) |= ϕ(a, b1 , . . . , bn )}.
14 STEFAN GESCHKE

Using the formula sat we can define a function D that assigns to each set M the
set D(M ) of subsets of M that are definable in (M, ∈). For every set M and every
real-world formula ϕ(x, y1 , . . . , yn ) we can show that for all b1 , . . . , bn ∈ M we have
{a ∈ M : ϕM (a, b1 , . . . , bn )} ∈ D(M ).
Lemma 5.1. Let M be a set. Then the following hold:
(1) D(M ) ⊆ P(M )
(2) If M is transitive, then D(M ) and M ⊆ D(M ).
(3) ∀X ⊆ M (|X|< ω → X ∈ D(M ))
(4) (AC) |M|≥ ω →|D(M )|=|M|
Proof. (1) is obvious. For (2) let M be transitive and a ∈ M . Then a = {x ∈
M : x ∈ a}. Hence a is a definable subset of M and thus a ∈ D(M ). Since a was
arbitrary, this shows M ⊆ D(M ). Now let a ∈ D(M ) and let b ∈ a. Since a ⊆ M ,
b ∈ M . But since M ⊆ D(M ), b ∈ D(M ). This shows that D(M ) is transitive.
(3) In V we use induction over the size of X. Clearly, the empty set is definable.
Now let n ∈ ω and let f : n + 1 → X be a bijection. By our inductive hypothesis,
f [n] is a definable subset of M , for instance f [n] = {x ∈ M : M |= ϕ(x, b1 , . . . , bm )}
for some b1 , . . . , bm ∈ M . Now
f [n + 1] = {x ∈ M : M |= ϕ(x, b1 , . . . , bm ) ∨ x = bm+1 },
where we choose bm+1 = f (n).
(4) easily follows from the facts that the language of set theory has only countably
many formulas and that for every n ∈ ω there are only |M| n-tuples of parameters
from M . 
5.2. The definition of L and its elementary properties.
Definition 5.2. For each ordinal α we define Lα recursively as follows:
(i) L0 := ∅
S D(Lα )
(ii) Lα+1 :=
(iii) Lα := β<α Lβ if α is a limit ordinal.
S
Let L := α∈Ord Lα be the class of constructible sets.
Using Lemma 5.1 by transfinite induction we easily get
Lemma 5.3. For all α ∈ Ord the set Lα is transitive. For all α, β ∈ Ord with
α ≤ β we have Lα ⊆ Lβ .
In particular, the Reflection Principle relativized to L applies to the function
α 7→ Lα .
Definition 5.4. For every a ∈ L we define the L-rank ρ(a) of a to be the least
α ∈ Ord such that a ∈ Lα+1 .
Lemma 5.5. For all α ∈ Ord we have the following:
(1) Lα = {a ∈ L : ρ(a) < α}
(2) Lα ∩ Ord = α
(3) α ∈ L and ρ(α) = α
(4) Lα ∈ Lα+1
(5) Every finite subset of Lα is an element of Lα+1 .
 MODELS OF SET THEORY 15

Proof. (1) is immediate from the definition of ρ. We show (2) by induction on α.

If α = 0, then Lα = ∅ = 0 ∩ Ord. If α is a limit ordinal, then
[ [
α= β= Lβ ∩ Ord = Lα ∩ Ord .
β<α β<α

Now let α = β + 1. Then β = Lβ ∩ Ord. But the definition of the class of ordinals
is ∆0 and hence absolute over the transitive class Lβ . Thus, β is a definable subset
of Lβ , namely β = {a ∈ Lβ : (Lβ , ∈) |= a is an ordinal}. Hence β ∈ Lβ+1 . On the
other hand, α is not a subset of Lβ , simply because β 6∈ Lβ . The same holds for
every ordinal ≥ α. Hence α = Lα ∩ Ord.
(3) follows immediately from (2). (4) is obvious. 

Lemma 5.6. a) For all α ∈ Ord we have Lα ⊆ Vα .

b) For all n ∈ ω, Ln = Vn . Moreover, Lω = Vω .
c) Assuming AC, for all α ≥ ω, |Lα|=|α|.
Proof. a) follows from the fact that D(M ) ⊆ P(M ) for all sets M .
Since all finite subsets of a set are definable, for every finite set M we have
D(M
S ) = P(M S ). This implies Ln = Vn for every finite ordinal n. Now Lω =
n∈ω L n = n∈ω Vn = Vω . This shows b).
For c) observe that since all the Ln are finite, |Lω |= ℵ0 . Now, since for every
infinite set M we have |D(M )|=|M| transfinite induction on α shows |Lα|=|α| for
all α ≥ ω. 

5.3. ZF in L. We assume that V satisfies ZF and show that L satisfies ZF as well.

That AC and GCH are satisfied in L requires some more work.
Theorem 5.7. L satisfies ZF.
Proof. Being an increasing union of transitive sets, the class L itself is transitive.
Hence it satisfies the Axiom of Extensionality. Clearly, ∅ ∈ L and hence Set Exis-
tence is satisfied. Foundation holds in L because it holds in V . L satsifies Infinity
since ω ∈ L.
We now show Separation in L. Let ϕ(x, y1 , . . . , yn ) be a formula and let a, b1 , . . . , bn ∈
L. We have to show that {x ∈ a : ϕL (x, b1 , . . . , bn )} ∈ L. Choose α ∈ Ord such
that a, b1 , . . . , bn ∈ Lα . By the Reflection Principle there is β > α such that ϕ is
absolute between Lβ and L. Now
{x ∈ a : ϕL (x, b1 , . . . , bn )} = {x ∈ a : ϕLβ (x, b1 , . . . , bn )}.
Clearly, {x ∈ a : ϕLβ (x, b1 , . . . , bn )} ∈ D(Lβ ) = Lβ+1 ⊆ L.
Since the Separation Axiom holds in L, in order to prove Pairing, Union, Power
Set and Replacement in L we only have to prove the existence of sufficiently large
sets. Good candidates are the Lα . We give the proof of the Power Set Axiom. The
proofs of the other axioms are similar but rather easier.
Let a ∈ L and let α := sup{ρ(b) : b ∈ L ∧ b ⊆ a} + 1. Now for all b ∈ L with
b ⊆ a we have b ∈ Lα ∈ L. Now the collection of all elements of L that are subsets
of a can be obtained from Lα using Separation. 

5.4. V = L? It is a natural question whether V can be the same as L, i.e., whether

ZF+V = L is consistent. Intuitively, one would think that L satisfies V = L.
While this is true, it is an non-trivial fact. It could happen that, when we repeat
the construction of L inside L itself, we end up with some class LL which is a proper
subclass of L. Fortunately, this is not the case.
Theorem 5.8. L satisfies V = L.
16 STEFAN GESCHKE

Proof. We already know that L contains all the ordinals. Hence, in order to show
(V = L)L , it suffices to show that for all α ∈ Ord we have Lα = LL α . Clearly
L0 = ∅ = LL 0 and the limit stages of this inductive proof are easy.
Now let α ∈ Ord and assume Lα = LL L
α . Now Lα+1 = Lα+1 is equivalent to
L
D(Lα ) = D (Lα ). S The definition of D is a formula that uses the parameter ω,
respectively ω <ω = n∈ω ω n , because that is the set that we use to code formulas
as sets. Moreover, the definition of D uses recursion over ω, but the individual
steps in the recursion only use ∆0 -formulas.
It is easily checked that recursively defined functions in which the definitions of
the individual steps of the recursion are absolute are again absolute over models
of sufficiently large fragments of ZF. This shows that D is absolute over transitive
classes that satisfies ZF without Power Set. (We need some fragment of ZF that
allows us to prove the recursion theorem. ZF without Power Set is convenient,
works for our purposes and is satisfied by many sets.)
This shows that L and V agree on whether a subset of Lα is definable or not.
Hence D(Lα ) = DL (Lα ). This finishes the proof of the theorem. 
These absoluteness considerations show this: If M is a transitive class that
contains all the ordinals and satisfies ZF, then L ⊆ M and L = LM . In other
words, L is the smallest transitive class that contains all the ordinals and satisfies
ZF.
We will see later that ZF is consistent with V 6= L.
 MODELS OF SET THEORY 17

5.5. AC and GCH in L.

Theorem 5.9. V = L implies the Axiom of Choice
The proof of this theorem uses
Lemma 5.10. Let @ be a well ordering of a set X.
a) For all a1 , . . . , an , b1 , . . . , bn ∈ X let ā = (a1 , . . . , an ) @n b̄ = (b1 , . . . , bn ) iff
ā 6= b̄ and for the minimal i with ai 6= bi it holds that ai @ bi . Then @n is a
well-ordering on X n .
b) For all ā, b̄ ∈ X <ω = n∈ω X n let ā @<ω b̄ if either ā is a shorter finite
S

sequence than b̄ or for some n ∈ ω, ā, b̄ ∈ X n and ā @n b̄. Then @<ω is a well-
ordering of X <ω .
Exercise 5.11. Prove Lemma 5.10.
Proof of Theorem 5.9. We show a statement stronger than AC. We show that V =
L implies that there is a definable binary relation / that is a well-ordering of L.
This clearly implies that every constructible set X can be well-ordered, namely by
the restriction of / to X.
Recursively, for all α ∈ Ord we define well-orderings Cα of Lα such that for
α < β, (Lα , Cα ) is an initial segment of (Lβ , Cβ ), i.e., Cα =Cβ  Lα and for all
a ∈ Lα and b ∈ Lβ with b Cβ a we have b ∈ Lα . S
Obviously, we have to put C0 = ∅. If α is a limit ordinal, let Cα = β<α Cβ .
S
Similarly, let C= α∈Ord Cα . The only difficult step is the definition of Cα if
α = β + 1 for some ordinal β.
Assume Cβ is a wellordering of Lβ . Since there are only countably many formulas
in the language of set theory, the set of all formulas can be well-ordered. Let ≺ be
a well-ordering of the set of formulas.
Now let a, b ∈ Lα = Lβ+1 . If a, b ∈ Lβ , let a Cα b iff a Cβ b. If a ∈ Lβ and
b ∈ Lα \ Lβ , let a C b. If a, b ∈ Lα \ Lβ , let ϕa (x1 , . . . , xn ) and ϕb (y1 , . . . , ym ) be
≺-minimal formulas defining a, respectively b in (Lβ , ∈) with suitable parameters.
Let (a1 , . . . , an ) ∈ Lnβ be Cnβ -minimal with
a = {c ∈ Lβ : (Lβ , ∈) |= ϕa [c, a1 , . . . , an ]}
and let (b1 , . . . , bm ) ∈ Lm m
β be Cβ -minimal with

b = {c ∈ Lβ : (Lβ , ∈) |= ϕb [c, b1 , . . . , bm ]}.

Now let a Cα b iff either ϕa ≺ ϕb or ϕa = ϕb and (a1 , . . . , an ) C<ω
β (b1 , . . . , bm ).
This finishes the definition of the Cα and hence of C. It is straight forward to
verify that the construction works (using the proof of Lemma 5.10). 
18 STEFAN GESCHKE

In order to show that GCH holds in L we need

Lemma 5.12. Let κ > ℵ0 be a regular cardinal. Then Lκ is a model of ZF without
the Power Set Axiom.
Proof. Since κ is an infinite cardinal, κ is a limit ordinal. The proofs of Set Exis-
tence, Extensionality, Pairing and Union in L actually go through for all Lα with
α a limit ordinal. In order to show the Separation Scheme, we have to use some
form of the Reflection Principle for Lκ .
Let if α < κ. For each existential formula ∃xϕ(x, y1 , . . . , yn ) and all parameters
b1 , . . . , bn ∈ Lα with
Lκ |= ∃xϕ[b1 , . . . , bn ]
choose γ < κ such that Lγ already contains a witness to this existential statement.
To avoid trivialities, we choose γ > α.
Let B be the set of all γ’s chosen in this way. Since there are only countably
many formulas and since Lα is of size < κ, B is of size < κ. Since κ is regular,
β0 = sup(B) < κ. We now replace Lα by Lβ0 and go through the same process,
obtaining β1 < κ and so on.
Finally, let β = supn∈ω βn . Again by the regularity of κ, β < κ. Clearly, Lβ is
an elementary submodel of Lκ .
We are now ready to show that the Separation Scheme holds in Lκ . Let ϕ(x, y1 , . . . , yn )
be a formula and a, b1 , . . . , bn ∈ Lκ . We have to show that
d = {c ∈ a : Lκ |= ϕ[c, b1 , . . . , bn ]} ∈ Lκ .
Let α < κ be such that a, b1 , . . . , bn ∈ Lα . By the argument above, there is
β < κ such that α < β and Lβ is an elementary submodel of Lκ . Now d = {c ∈ a :
Lβ |= ϕ[c, b1 , . . . , bn ]} is a definable subset of Lβ and hence d ∈ Lβ+1 ⊆ Lκ . This
shows Separation.
The proof of Replacement is very similar to the proof of Replacement in L and
uses again the regularity of κ and the fact that all elements of Lκ are of size < κ. 
Theorem 5.13. V = L implies GCH.
Proof. Assume V = L and let κ be an infinite cardinal. We show that P(κ) ⊆ Lκ+
and therefore 2κ ≤ κ+ .
Let a ⊆ κ. By V = L there is a regular cardinal λ > ℵ0 such that a ∈ Lλ . By the
Löwenheim-Skolem Theorem, there is an elementary submodel M of Lλ such that
|M|= κ and κ ∪ {a} ⊆ M . Let N be the transitive collapse of M . Since a ⊆ κ ⊆ κ
and κ is transitive, κ ∪ {a} is transitive. Since Mostowski collapsing functions are
the identity on transitive classes, a ∈ N .
Since N is isomorphic to an elementary submodel of Lλ , N |= V = L. Since N is
a transitive model of S
ZF without the Power Set Axiom, the absoluteness properties
of L give that N = α∈N ∩Ord Lα = Lβ where β = sup(N ∩ Ord). Since M and
therefore N are of size κ, |β|=|Lβ |= κ. In particular, β < κ+ .
It follows that a ∈ Lκ+ , finishing the proof of the theorem. 
Corollary 5.14. If ZF is consistent, then so is ZFC+GCH.
Exercise 5.15. Assume V = L. Given an infinite α < ℵ1 , give an explicit example
of a subset a of ω such that a ∈ Lℵ1 \ Lα .
Hint: This actually requires some thought. First of all observe that if α < ℵ1
is large enough, then Lα contains a bijection between ω and ω × ω. Using this
bijection, relations on ω, i.e., subsets of ω × ω, can be coded by subsets of ω. Now,
if R is a well-ordering on ω and α < ℵ1 is suitably chosen (say, for example, Lα is
an elementary submodel of Lℵ1 ), then Lα will also contain the unique ordinal that
is isomorphic to (ω, R), because we can perform the Mostowski collaps of (ω, R)
 MODELS OF SET THEORY 19

inside Lα . What is an example of a set (not necessarily a subset of ω) that is in

Lℵ1 , but not in Lα ?
20 STEFAN GESCHKE

6. Forcing
In this section we will find a way to enlarge the universe of all sets.

6.1. Partial orderings, dense sets and filters.

Definition 6.1. Let P be a set and ≤ a binary relation on P. Then (P, ≤) is a
partial ordering or a partial order if the following hold:
(i) (Transitivity) ∀x, y, z ∈ P(x ≤ y ∧ y ≤ z → x ≤ z)
(ii) (Reflexivity) ∀x ∈ P(x ≤ x)
(iii) (Antisymmetry) ∀x, y ∈ P(x ≤ y ∧ y ≤ x → x = y))
The elements of P are conditions. If p, q ∈ P and p ≤ q we say that p is stronger
than q, respectively, p extends q. Two conditions p and q are compatible if they
have a common extension r ∈ P, i.e., if there is r ∈ P such that r ≤ p and r ≤ q. If
p and q are not compatible, they are incompatible and we write p ⊥ q. For technical
reasons we will only consider partial orders P with a largest element, denoted by
1 or 1P . Similarly, when dealing with several partial orders at the same time, we
might denote the relation ≤ on P by ≤P . In this case we might also write ⊥P instead
of just ⊥.
Let us consider a couple of examples. (N, ≤) is not a partial order in our sense,
because it does not have a largest element (but of course, it is a perfectly nice
partial order in other contexts). (N, ≥) is a partial order in our sense, but boring,
because any two elements are compatible.
Let O be the collection of all non-empty open subsets of R. (O, ⊆) is a very nice
partial order in our sense. Two condition U, V ∈ O are compatible iff U ∩ V 6= ∅.
Clearly, 1O = R.
Let M denote the collection of all measurable subsets of R of positive measure.
Then (M, ⊆) is another nice partial order. A, B ∈ M are compatible iff A ∩ B is
not of measure zero.
Observe that O and M are not standard names for these partial orders. Typically
one considers closely related Boolean algebras that are, for our purposes, equivalent
to O and M, namely the Cohen-algebra C and the measure algebra B. We will see
later what the connection between partial orders and Boolean algebras is.
Another important family of partial orders arises as follows: Let X be a set.
Consider
Fn(X, 2) := {p : p is a function from a finite subset of X to 2}.
Fn(X, 2) is partially ordered by the reverse inclusion ⊇. The largest element of
Fn(X, 2) is the empty function. Two conditions p, q ∈ Fn(X, 2) are compatible
iff p ∪ q is a function. The elements of Fn(X, 2) should be considered as finite
approximations of a function from X to 2.
For our purposes, Fn(ω, 2) is equivalent to O and to C. We will later see why
this is.
Definition 6.2. Let (P, ≤) be a partial order. Then D ⊆ P is dense in P if for all
p ∈ P there is q ∈ D such that q ≤ p. For p ∈ P we say that D ⊆ P is dense below
p if for all q ≤ p there is r ∈ D such that r ≤ q. A set A ⊆ P is an antichain if no
two distinct elements of A are compatible. A set G ⊆ P is a filter if the following
hold:
(i) ∀p ∈ P∀q ∈ G(q ≤ p → p ∈ G)
(ii) ∀p, q ∈ G∃r ∈ G(r ≤ p ∧ r ≤ q)
If D is a family of dense subsets of P and G ⊆ P a filter, then G is D-generic iff
for all D ∈ D, G ∩ D 6= ∅.
 MODELS OF SET THEORY 21

S
What do dense subsets of O look like? If D ⊆ O is any set, then X = D is
a union of open sets and therefore open. If D is dense, then for every non-empty
open set U ⊆ R there is V ∈ D such that V ⊆ U . In particular, X ∩ U 6= ∅. It
follows that X is dense in R in the topological sense.
On the other hand, if X ⊆ R is dense and open, consider D = {U ⊆ R :
U ⊆ X is open}. Now for every V ∈ O, X ∩ V is non-empty and open and thus
U = X ∩ V ∈ D. It follows that D is dense in O in the order-theoretic sense.
Note however that not every dense subset of O is the collection of all non-empty
open subsets of a dense open set X ⊆ R. An example is the set of all open intervals
with rational endpoints. Another example is for ε > 0 the set Dε of all open
intervals of length < ε.
Now let x be a real number and consider Gx := {U ∈ O : x ∈ U }. It is easily
checked that Gx is a filter in O. Moreover, Gx is (Dε )ε>0 -generic. On the other
hand, if G ⊆ OTis a (Dε )ε>0 -generic filter, then thereTis exactly one real number x
such that x ∈ U ∈G cl(U ). Note, however, that x ∈ G can fail.
Exercise 6.3. Consider the partial order Fn(ω, 2). Show:
S
(1) If G ⊆ Fn(ω, 2) is a filter, then G is a function.
(2) For all n ∈ ω the setDn := {p ∈ Fn(ω, 2) : n ∈ dom(p)}S is dense Fn(ω, 2).
(3) If G ⊆ Fn(ω, 2) is a {Dn : n ∈ ω}-generic filter, then G is a function from
ω to 2.
(4) For every function f : ω → 2 the set Gf := {p ∈ Fn(ω, 2) : p ⊆ f } is a
{Dn : n ∈ ω}-generic filter.
The next theorem is a version of the Baire Category Theorem and guarantees
the existence of sufficiently generic filters.
Theorem 6.4. (Rasiowa-Sikorsky) Let (P, ≤) be a partial order and let D be a
countable family of dense subsets of P. Then there is a D generic filter G ⊆ P.
Proof. Let {Dn : n ∈ ω} be an enumeration of D. Choose a sequence (pn )n∈ω of
conditions in P as follows: Let p0 ∈ D0 . Suppose for n ∈ ω we have chosen pn
already. Choose pn+1 ≤ pn such that pn+1 ∈ Dn+1 . This can be done since Dn+1
is dense.
Now let
G := {p ∈ P : ∃n ∈ ω(pn ≤ p)}.
It is straight forward to check that G is a D-generic filter. 
Exercise 6.5. Let (A, ≤A ) and (B, ≤B ) countably infinite, dense linear orders
without endpoints. (Recall that a linear order is dense if strictly between any two
elements there is another element of the linear order.) Show that (A, ≤A ) and
(B, ≤B ) are isomorphic.
Hint: Use the Rasiowa-Sikorski Theorem. Consider the partial P order of iso-
morphisms between finite subsets of A and B, ordered by reverse inclusion. Find a
S D of dense subsets of P such that for every D-generic filter G ⊆ P
countable family
the function G is an isomorphism from A to B. It might help to take another
look at Exercise 6.3.
22 STEFAN GESCHKE

6.2. Generic extensions. Our general assumption is that ZFC is consistent. In

order to show that ZFC+¬CH is consistent it is enough to show that every finite
set of sentences from ZFC+¬CH is consistent.
We assume that V satisfies all axioms of ZFC. We show that for every finite
subset F of ZFC+¬CH there is a finite subset F 0 of ZFC such that is M is a
countable transitive model of F 0 , then there is a countable transitive model M [G]
of F . M and M [G] will have the same ordinals and M ⊆ M [G].
The construction of M [G] is independent of the particular set F , i.e., we will
simply assume that M is a countable transitive model of all of ZFC and then
construct a model M [G] of all of ZFC+¬CH from it. In order to verify that M [G]
satisfies F we will have used the fact, that M satisfies a certain finite set F 0 of
axioms of ZFC.
Now let M be a countable transitive model of ZFC and let P be a partial order
in M . For every D ∈ M being a dense subset of P is absolute over M . Since M is
countable, M contains only countably many dense subsets of P. By the Rasiowa-
Sikorski Theorem, there is a filter G ⊆ P such that G intersect all the dense subsets
of P that are elements of M . Such a filter is P-generic over M .
Exercise 6.6. Let M be a countable transitive model of ZFC and let (P, ≤) ∈ M
be a partial order with the property that every p ∈ Phas at least two incompatible
extensions. Show that no filter G ⊆ P that is P-generic over M is actually an
element of M .
Hint: Given a filter G ∈ M , find a dense subset D ∈ M that is disjoint from G.
The model M [G] will be the smallest transitive model of ZFC such that M ∪
{G} ⊆ M [G]. The elements of M [G] coded by names in M that will be decoded
into elements of M [G] using the generic filter G.
Definition 6.7. A set τ is a P-name if it is a set of pairs and moreover, for every
pair (σ, p) ∈ τ , p is an element of P and σ is a P-name. Let M P be the class of all
P-names in M .
Note that the definition of names is done by recursion over ∈ and is simple
enough to be absolute over M . I.e., τ ∈ M is a P-name iff it is a P-name in V .
Definition 6.8. Let G be a P-generic filter over M . For every τ ∈ M P let
τG = {σG : ∃p ∈ G((σ, p) ∈ τ )}.
be the evaluation of τ with respect to G. Let
M [G] = {τG : τ ∈ M P }.
The evaluations τG are defined by recursion over ∈. This definition takes place
in V . However, the definition is absolute over transitive models of a sufficiently
large fragment of ZFC.
Lemma 6.9. If N is a transitive model of ZFC such that M ∪ {G} ⊆ N , then
M [G] ⊆ N .
In order to show that M is a subset of M [G], for every x ∈ M we have to find a
name τ ∈ M such that x = τG .
Definition 6.10. For every x let x̌ = {(y̌, 1P ) : y ∈ x} be the canonical name for
x.
Lemma 6.11. For all x ∈ M , x̌G = x. In particular, M ⊆ M [G].
Proof. Observe that 1P ∈ G. Now using induction over ∈,
x̌G = {y̌G : (y̌, 1P ) ∈ x̌} = {y : y ∈ x} = x.

 MODELS OF SET THEORY 23

Next we show that G ∈ M [G]. We have to come up with a name for G.

Lemma 6.12. Let Γ = {(p̌, p) : p ∈ P}. Then ΓG = G.
Proof.
ΓG = {p̌G : p ∈ G} = {p : p ∈ G} = G

Let us collect some properties of M [G].
Lemma 6.13. a) M [G] is transitive.
b) For all τ ∈ M P , rk(τG ) ≤ rk(τ ).
c) Ord ∩M = Ord ∩M [G]
Proof. For a) let y ∈ M [G] and x ∈ y. Then there is some name ẏ ∈ M P such that
ẏG = y. By the definition of ẏG , there is (ẋ, p) ∈ ẏ such that x = ẋG and p ∈ G.
But that means that x = ẋG ∈ M [G].
b) follows by a straight forward ∈-induction.
For c) first observe that Ord ∩M ⊆ Ord ∩M [G] by Lemma 6.11. On the other
hand, If α ∈ Ord ∩M [G], then there is a name α̇ ∈ M P such that α̇G = α. By b),
α = rk(α) ≤ rk(α̇) ∈ Ord ∩M.
Since M [G] is transitive, α ∈ M . It follows that Ord ∩M [G] ⊆ Ord ∩M . 
The easier part of the proof that M [G] satisfies ZFC is
Lemma 6.14. M [G] satisfies Foundation, Extensionality, Infinity, Pairing and
Union.
Proof. Foundation is automatic since ∈ is well-founded in V . Extensionality is
satisfied since M [G] is transitive. Infinity holds since ω ∈ M ⊆ M [G]. For Pairing
let a, b ∈ M [G]. Choose names σ, τ ∈ M P such that a = σG and b = τG . Now
η = {(σ, 1P ), (τ, 1P )} is a name and ηG = {σG , τG } = {a, b}.
For Union let F ∈ M [G] and let τ ∈ M P be a name such that τG = F . Let
σ = {(η, 1P ) : ∃p, q ∈ P∃π ∈ M P ((π, p) ∈ τ ∧ (η, q) ∈ π)}.
Obviously, σ ∈ MS P
. S
We show that F ⊆ σG . Let a ∈ F . Then there is b ∈ F such that a ∈ b. It
follows that there are π, η ∈ M P and p, q ∈ G such that b = πG , a = ηG , (η, q) ∈ π
and (π, p) ∈ τ . By the definition of σ, (η, 1P ∈ σ and hence a ∈ σG . This shows
Union. 
Exercise 6.15. Assume we already know that M [G] satisfies ZF. Show that M [G]
satisfies AC.
Hint: Since M satisfies AC, every name can be well-ordered.
Before we introduce the tools needed to verify the rest of ZFC in M [G], we
collect some additional data on generic filters.
Definition 6.16. A subset D of a partial order P is predense if the set {p ∈ P : ∃q ∈
D(p ≤ q)} is dense. D is predense below p ∈ P if the set {q ∈ P : ∃r ∈ D(q ≤ r)} is
dense below p. A set O ⊆ P is open if for all p ∈ O and all q ≤ p we have q ∈ O.
A good example of predense sets are maximal antichains (maximal with respect
to ⊆). Recall that A ⊆ P is an antichain if any two elements of A are incompatible.
Exercise 6.17. Use Zorn’s Lemma to show that every antichain of P is contained
in a maximal one.
24 STEFAN GESCHKE

Now let A ⊆ P be a maximal antichain. We show that A is predense. For this

we have to show that the set D = {p ∈ P : ∃q ∈ A(p ≤ q)} is dense in P. So, let
r ∈ P. Since A is a maximal antichain, either r ∈ A and hence r ∈ D or A ∪ {r}
fails to be an antichain. In the latter case, there is p ∈ A such that r and p are
compatible. Let q be a common extension of r and p. Now q ≤ r and q ∈ D. This
shows that indeed, D is dense in P. Hence A is predense in P.
Exercise 6.18. Let D ⊆ P be dense and let A ⊆ D be an antichain that is maximal
among all antichains that are subsets of D. Show that A is a maximal antichain in
P.
Lemma 6.19. Let M be a countable transitive model of ZFC, P ∈ M a partial
order and G ⊆ P a filter.
a) If G is generic over M and p ∈ P, then the following holds:
p∈G ↔ ∀q ∈ G(p 6⊥ q)
b) The following are equivalent:
(1) G is generic over M .
(2) For every D ∈ M that is predense in P, G ∩ D 6= ∅.
(3) For every D ∈ M that is dense and open in P, G ∩ D 6= ∅.
c) For every p ∈ P the following are equivalent:
(1) G is generic over M and p ∈ G.
(2) For every D ∈ M that is dense below p, G ∩ D 6= ∅.
(3) For every D ∈ M that is predense below p, G ∩ D 6= ∅.
Proof. a) Obviously, if p ∈ G, then p is compatible with all elements of G. Now
assume that p is compatible with all elements of G. Consider the set D = {q ∈ P :
q ≤ p ∨ q ⊥ p}. D is dense in P:
Let r ∈ P. If r ⊥ p, then r ∈ D. If r 6⊥ p, then there is q ∈ P such that q ≤ r, p.
But now q ∈ D and hence r has an extension in D. This shows that D is dense.
Since p, P ∈ M , D ∈ M . Hence G intersects D. But since all elements of G are
compatible with p, the only way that G can intersect D is that G contains some q
with q ≤ p. But then p ∈ G.
b) (1) ⇒ (3) and (2) ⇒ (1) are trivial. Now assume (3) and let D ∈ M be
predense in P. Let E = {p ∈ P : ∃q ∈ D(p ≤ q)}. Note that E is definable from D
and P and thus E ∈ M . Since D is predense, E is dense in P. It is easily checked
that E is open. By (3), there is p ∈ G ∩ E. By the definition of E, there is q ∈ D
such that p ≤ q. Since G is a filter, q ∈ G. Hence G intersects D. This shows (2).
c) The equivalence of (2) and (3) follows using the same arguments as for the
equivalence of (1) and (2) in b). Now assume either (2) or (3). Consider the set
D = {q ∈ P : q ≤ p}. Clearly, D is dense below p. Hence G intersects D. Hence G
contains some condition ≤ p. Hence G contains p. This shows (1).
Now assume (1) and let D ∈ M be dense below p. Let E = {q ∈ P : (q ∈ D ∧ q ≤
p) ∨ q ⊥ p}. It is easily checked that E is dense in P. Since p, P, D ∈ M , E ∈ M .
By genericity of G there is q ∈ G ∩ E. Since p ∈ G and since G is a filter, q 6⊥ p.
By the definition of E, q ∈ D. Hence G intersects D. This shows (2). 
6.3. The forcing relation. We need a method to talk about truth in M [G] from
the perspective of M . This is provided by the forcing relation (“forces”). On the
left hand side of this relation we have conditions in our fixed partial order P. On
the right hand side we have formulas of the forcing language. The forcing language
consists of expressions of the form ϕ(τ1 , . . . , τn ) where ϕ(x1 , . . . , xn ) is a formula in
the language of set theory and at every free occurence of xi , xi has been substituted
by the P-name τi .
 MODELS OF SET THEORY 25

Definition 6.20. Let p ∈ P, let ϕ(x1 , . . . , xn ) be a formula in the language of set

theory and let τ1 , . . . , τn be P-names. Then
p ϕ(τ1 , . . . , τn )
iff for all P-generic filters G over M with p ∈ G,
M [G] |= ϕ((τ1 )G , . . . , (τn )G ).
The relation turns out to be a definable class in M . This is surprising since
M does not have any knowledge about generic filters over M . The goal of this
subsection is to show the definability of in M .
Definition 6.21. For a formula ϕ(τ1 , . . . , τn ) of the forcing language of M let
[[ϕ(τ1 , . . . , τn )]] = {p ∈ P : p ϕ(τ1 , . . . , τn )}
be the truth value of ϕ(τ1 , . . . , τn ).
Exercise 6.22. a) Show that for every formula ϕ(τ1 , . . . , τn ) in the forcing language
the truthvalue [[ϕ(τ1 , . . . , τn )]] is an open subset of P.
b) Let ϕ(τ1 , . . . , τn ) be as above, p ∈ P and assume that [[ϕ(τ1 , . . . , τn )]] is pre-
dense below p. Show that p ∈ [[ϕ(τ1 , . . . , τn )]].
Hint: For b) use Lemma 6.19.
Note that our definition is not the standard definition of truth values in forcing.
Typically, truth values are only defined if the partial order P is a complete Boolean
algebra.
Definition 6.23. A Boolean algebra is is a partial order B with the following
properties:
(1) B has a largest element 1 and a smallest element 0.
(2) Any two elements a, b ∈ B have a largest common lower bound a ∧ b and a
smallest common upper bound a ∨ b.
(3) Every a ∈ B has a complement ¬a such that a ∨ ¬a = 1 and a ∧ ¬a = 0.
(4) For all a, b, c ∈ B, a ∧ (b ∨ c) = (a ∧ b) ∨ (a ∧ c).
W
A Boolean algebra B is complete
V if every set A ⊆ B has a least upper bound A
and a greatest lower bound A.
If B is a Boolean algebra, then A ⊆ B is a subalgebra of B if it contains 0 and 1
and is closed under ∨, ∧ and ¬.
The simplest examples of Boolean algebras are the power set algebras (P(X), ⊆)
where ∧ is intersection, ∨ union and ¬ actual complementation relative to the set
X. It can be shown that every Boolean algebra is isomorphic to a subalgebra of a
power set algebra.
If a Boolean algebra B is to be used for forcing purposes, we consider the partial
order P = B \ {0} instead. The traditional definition of the truth value of a formula
ϕ(τ1 , . . . , τn ) in a complete Boolean algebra B is
_
[[ϕ(τ1 , . . . , τn )]] = {a ∈ B : a ϕ(τ1 , . . . , τn )}.
We will however stick to our definition of truth values in Definition 6.21 and
show later that our truth values actually are the appropriate elements of a certain
complete Boolean algebra, namely the completion of the partial order P that we are
forcing with.
26 STEFAN GESCHKE

In order to show the definability of in M it is certainly enough to show that

the map
ϕ(τ1 , . . . , τn ) 7→ [[ϕ(τ1 , . . . , τn )]]
is definable in M . In M , we will define an approximation [[ϕ(τ1 , . . . , τn )]]∗ of
[[ϕ(τ1 , . . . , τn )]] and then show that the two sets are actually the same.
For the following definitions we pretend to live inside M .
Definition 6.24. For A ⊆ P let
reg(A) = {p ∈ P : A is predense below p}
be the regularization of A. A is regular open if A = reg(A).
For A, B ⊆ P let
A ∨ B = reg(A ∪ B), A ∧ B = reg(A) ∩ reg(B) and ¬A = {p ∈ P : ∀q ∈ A(p ⊥ q)}.
For p ∈ P and A ⊆ P let p ∧ A = A ∧ p = {p} ∧ A and ¬p = ¬{p}. For F ⊆ P(P)
let _ [  ^ \
F = reg F and F = {reg(A) : A ∈ F}.
Observe that for every A, reg(A) is open and contains every p such that reg(A)
is predense below p. In other words, reg(A) is regular open. In fact, reg(A) is the
smallest regular open superset of A.
The collection of all regular open subsets of P is denoted by ro(P) and is a com-
W order ⊆. The algebraic operations
plete Boolean algebra with respect to the partial
on ro(P) areV just ∨, ∧ and ¬. For F ⊆ ro(P), F is indeed the supremum of F in
ro(P) and F is the infimum. The smallest element of ro(P) is just ∅, the largest
element is P.
It is tempting to believe that ro(P) is a subalgebra of P(P), which in fact it
typically is not because ∨, ∧ and ¬ are not the same as ∩, ∪ and complementation
in P(P).
The Boolean algebra ro(P) is the completion of P. Via the map
e : P → ro(P); p 7→ reg({p})
every element of P can be considered as an element of ro(P). This map, however,
sometimes fails to be 1-1.
T
Exercise 6.25. a) Let F ⊆ ro(P) be a family of regular open sets. Show that F
is regular open.
b) Let A ⊆ P. Show that ¬A is a regular open subset of P.
Exercise 6.26. Let e : P → ro(P) be as above.
a) Show that the range of e is dense in ro(P). Here a subset D of a Boolean
algebra B is dense in B if it is dense in B \ {0} in the partial order sense.
b) P is separative if for p, q ∈ P we have
p=q ↔ ∀r ∈ P(r ⊥ p ↔ r ⊥ q).
Show that e is 1-1 iff P is separative.
Hint: For b), first show that e(p) = e(q) iff ¬p = ¬q.
Definition 6.27. Let σ and τ be P-names. We define
_
[[σ ∈ τ ]]∗ = {[[σ = η]]∗ ∧ p : (η, p) ∈ τ },
^
[[σ ⊆ τ ]]∗ = {¬p ∨ [[η ∈ τ ]]∗ : (η, p) ∈ σ}
and
[[σ = τ ]]∗ = [[σ ⊆ τ ]]∗ ∧ [[τ ⊆ σ]]∗ .
 MODELS OF SET THEORY 27

For formulas ϕ(σ1 , . . . , σn ) and ψ(τ1 , . . . , τm ) in the forcing language let

[[ϕ(σ1 , . . . , σn ) ∧ ψ(τ1 , . . . , τm )]]∗ = [[ϕ(σ1 , . . . , σn )]]∗ ∧ [[ψ(τ1 , . . . , τm )]]∗ ,
[[ϕ(σ1 , . . . , σn ) ∨ ψ(τ1 , . . . , τm )]]∗ = [[ϕ(σ1 , . . . , σn )]]∗ ∨ [[ψ(τ1 , . . . , τm )]]∗
and
[[¬ϕ(σ1 , . . . , σn )]]∗ = ¬[[ϕ(σ1 , . . . , σn )]]∗ .
For a formula ϕ(x, y1 , . . . , yn ) in the language of set theory and P-names τ1 , . . . , τn
let _
[[∃xϕ(x, τ1 , . . . , τn )]]∗ = {[[ϕ(σ, τ1 , . . . , τn )]]∗ : σ is a P-name}.
Note the very subtle recursion in this definition. For P-names σ0 , σ1 , τ0 and τ1
let (σ0 , τ0 )R(σ1 , τ1 ) if rk(σ0 ) < rk(σ1 ) and rk(τ0 ) ≤ rk(τ1 ) or if rk(σ0 ) ≤ rk(σ1 ) and
rk(τ0 ) < rk(τ1 ). The relation R is well-founded. [[σ = τ ]]∗ , [[σ ∈ τ ]]∗ and [[σ ⊆ τ ]]∗
are defined by recursion over R, where we first define [[σ ⊆ τ ]]∗ and [[τ ⊆ σ]]∗ and
only then [[σ = τ ]]∗ . The rest of Definition 6.27 is a typical recursion over the
complexity of a formula. Here we consider σ ⊆ τ as an atomic formula.
Lemma 6.28. Let ϕ(τ1 , . . . , τn ) be a formula in the forcing language of M and let
G be P-generic over M . Then
M [G] |= ϕ[(τ1 )G , . . . , (τn )G ] ⇔ G ∩ [[ϕ(τ1 , . . . , τn )]]∗ 6= ∅.
Proof. We will first prove this lemma for atomic formulas including formulas of the
form x ⊆ y. We use induction over a well-founded relation, the same relation R
that we used in the definition of the truth values of atomic formulas in the forcing
language.
Let σ, τ ∈ M P and suppose that σG ∈ τG . Then there are η ∈ M P and p ∈ G
such that (η, p) ∈ τ and σG = ηG . Now (σ, η)R(σ, τ ) and therefore, by the inductive
hypothesis, there is q ∈ [[σ = η]]∗ ∩ G. Since p, q ∈ G, there is r ∈ G such that
r ≤ p, q. We have r ∈ [[σ = η]]∗ ∧ p ⊆ [[σ ∈ τ ]]∗ , showing that G ∩ [[σ ∈ τ ]]∗ 6= ∅.
Now suppose that G intersects the set
_
[[σ ∈ τ ]]∗ = {[[σ = η]]∗ ∧ p : (η, p) ∈ τ }.
By the definition of , {[[σ = η]]∗ ∧ p : (η, p) ∈ τ } is predense below every element
W S
of [[σ ∈ τ ]]∗ . By the genericity of G, there is (η, p) ∈ τ such that G∩[[σ = η]]∗ ∧p 6= ∅.
Now p ∈ G and G ∩ [[σ = η]]∗ 6= ∅. By the inductive hypotesis, σG = ηG . Since
p ∈ G we have ηG ∈ τG . It follows that σG ∈ τG .
Now let σG ⊆ τG . Suppose G ∩ [[σ ⊆ τ ]]∗ = ∅. By the genericity of G, there is
q ∈ G ∩ ¬[[σ ⊆ τ ]]∗ . We have
^
¬[[σ ⊆ τ ]]∗ = ¬ {¬p ∨ [[η ∈ τ ]]∗ : (η, p) ∈ σ} =
_ _
{¬(¬p ∨ [[η ∈ τ ]]∗ ) : (η, p) ∈ σ} = {p ∧ ¬[[η ∈ τ ]]∗ : (η, p) ∈ σ}.
Therefore {p ∧ ¬[[η ∈ τ ]]∗ : (η, p) ∈ σ} is predense below q. Hence, there is
S
(η, p) ∈ σ such that G ∩ p ∧ ¬[[η ∈ τ ]]∗ 6= ∅. In particular, p ∈ G. Hence ηG ∈ σG .
Moreover, G ∩ [[η ∈ τ ]]∗ = ∅. By the inductive hypothesis, ηG 6∈ τG . It follows that
σ 6⊆ τ .
Assume σG 6⊆ τG . Then there is (η, p) ∈ σ with p ∈ G and ηG 6∈ τG . By
the inductive hypothesis, G does not intersect the set [[η ∈ τ ]]∗ . But then, by the
genericity of G, G intersects ¬[[η ∈ τ ]]∗ . It follows that G intersects
p ∧ ¬[[η ∈ τ ]]∗ = ¬(¬p ∨ [[η ∈ τ ]]∗ ).
Hence G is disjoint from
^
[[σ ⊆ τ ]]∗ = {¬p ∨ [[η ∈ τ ]]∗ : (η, p) ∈ σ}.
28 STEFAN GESCHKE

Now suppose that σG = τG . Then σG ⊆ τG and τG ⊆ σG . If G intersects

[[σ ⊆ τ ]]∗ and [[τ ⊆ σ]]∗ , then it intersects [[σ = τ ]]∗ .
On the other hand, if G interects [[σ = τ ]]∗ , then it also intersects [[σ ⊆ τ ]]∗ and
[[τ ⊆ σ]]∗ . It follows that σG = τG . This finishes the argument for atomic formulas.
For non-atomic formulas we use a straight forward induction on the complexity.
This yields no hidden traps and relies on arguments of the kind we have already
used. 
Using this lemma it is not hard to show
Theorem 6.29. For all formulas ϕ(τ1 , . . . , τn ) in the forcing language of M and
for all p ∈ P we have p ∈ [[ϕ(τ1 , . . . , τn )]] iff p ∈ [[ϕ(τ1 , . . . , τn )]]∗ holds in M . In
particular, the relation is definable in M .
Proof. Suppose p ϕ(τ1 , . . . , τn ). We show that [[ϕ(τ1 , . . . , τn )]]∗ is predense below
p. Suppose not. Then there is some q ≤ p that is not compatible with any ele-
ment of [[ϕ(τ1 , . . . , τn )]]∗ . By the proof of the Rasiowa-Sikorski Theorem, there is
a P-generic filter G over M that contains q. Because of the choice of q, G is dis-
joint from [[ϕ(τ1 , . . . , τn )]]∗ . By Lemma 6.28, M [G] satisfies ¬ϕ((τ1 )G , . . . , (τn )G ),
contradicting the fact that with q also p is an element of G. This shows
[[ϕ(τ1 , . . . , τn )]] ⊆ [[ϕ(τ1 , . . . , τn )]]∗ .
Now let p ∈ [[ϕ(τ1 , . . . , τn )]]∗ . Suppose there is a P-generic filter G over M
with p ∈ G such that ϕ((τ1 )G , . . . , (τn )G ) fails in M [G]. By Lemma 6.28, G ∩
[[¬ϕ(τ1 , . . . , τn )]]∗ 6= ∅. However, the condition p is incompatible with all elements
of [[¬ϕ(τ1 , . . . , τn )]]∗ , a contradiction. If follows that p ϕ(τ1 , . . . , τn ). This shows
[[ϕ(τ1 , . . . , τn )]]∗ ⊆ [[ϕ(τ1 , . . . , τn )]].

Corollary 6.30. Let G be P-generic over M . For a formula ϕ(τ1 , . . . , τn ) in the
forcing language of M we have M [G] |= ϕ((τ1 )G , . . . , (τn )G ) iff there is p ∈ G such
that p ϕ(τ1 , . . . , τn ).
6.4. ZFC in M [G]. Having defined the forcing relation in M , we are now ready
to verify the rest of ZFC in M [G].
Theorem 6.31. M [G] satifies ZFC.
Proof. The only axioms that we have not verified yet are Separation, Replacement
and Power Set.
For Separation let ϕ(x, y1 , . . . , yn ) be a formula in the language of set theory and
let a, b1 , . . . , bn ∈ M [G]. We show that
{x ∈ a : M [G] |= ϕ(x, b1 , . . . , bn )} ∈ M [G].
Let σ, τ1 , . . . , τn ∈ M P be such that σG = a and for all i ∈ {1, . . . , n}, (τi )G = bi .
Set
η = {(π, p) : ∃q ∈ P((π, q) ∈ σ ∧ p ≤ q) ∧ p ϕ(π, τ1 , . . . , τn )}.
Since is definable in M , we have η ∈ M . We show that ηG = {x ∈ a : M [G] |=
ϕ(x, b1 , . . . , bn )}. Let x ∈ ηG . The there is (π, p) ∈ η such that πG = x and p ∈ G.
By the definition of η we have p ϕ(π, τ1 , . . . , τn ) and p ≤ q for some q ∈ P with
(π, q) ∈ σ. It follows that x = πG ∈ σG and
M [G] |= ϕ(x, b1 , . . . , bn ).
Now let x ∈ a be such that M [G] |= ϕ(x, b1 , . . . , bn ). Then there is (π, q) ∈ σ
such that x = πG and q ∈ G. Moreover, there is p ∈ G with p ϕ(π, τ1 , . . . , τn ).
Since any two elements of G have a common extension in G, we may actually
 MODELS OF SET THEORY 29

assume p ≤ q. Hence (π, p) ∈ η, and thus x = πG ∈ ηG . This shows Separation in

M [G].
In order to show Replacement let ϕ(x, y, z1 , . . . , zn ) be a formula and let a, b1 , . . . , bn ∈
M [G] such that
M [G] |= ∀x ∈ a∃!y(ϕ(x, y, b1 , . . . , bn )).
Choose σ, τ1 , . . . , τn ∈ M P such that σG = a and for all i ∈ {1, . . . , n}, (τi )G = bi .
Choose a set S ⊆ M P in M such that the following holds: for all (π, p) ∈ σ and all
q ≤ p with
q ∃!y(ϕ(π, y, τ1 , . . . , τn ))
the set
{r ≤ q : ∃η ∈ S(r ϕ(π, η, τ1 , . . . , τn ))}
is dense below q. Consider the P-name S × {1}. We show that
M [G] |= ∀x ∈ a∃y ∈ (S × {1})G (ϕ(x, y, b1 , . . . , bn )).
Let x ∈ a. Then there is (π, p) ∈ σ such that πG = x and p ∈ G. Moreover, there
is q ∈ G such that q ≤ p and
q ∃!y(ϕ(π, y, τ1 , . . . , τn )).
By the choice of S there are η ∈ S and r ∈ G such that r ≤ q and r
ϕ(π, η, τ1 , . . . , τn ). Hence
M [G] |= ϕ(x, ηG , b1 , . . . , bn ),
and therefore
M [G] |= ∃y ∈ (S × {1})G (ϕ(x, y, b1 , . . . , bn )).
For Power Set let a ∈ M [G]. Choose σ ∈ M P with σG = a. Let A = {π : ∃p ∈
P((π, p) ∈ σ)}. For every function f : A → P(P) let τf = {(π, p) : p ∈ f (π)}. Put
η = {(τf , 1) : f is a function from A to P(P)}.
This definition is a definition in M .
We show
M [G] |= ∀x(x ⊆ a → x ∈ ηG ).
Let x ∈ M [G] with x ⊆ a. Choose τ ∈ M P with τG = x. Define f : A → P(P) as
follows: For π ∈ A let f (π) = {p ∈ P : p π ∈ τ }. Observe that f is a function in
M . Clearly, (τf )G ∈ ηG . We show that x = τG = (τf )G .
Let y ∈ x. Since x ⊆ σG there is (π, p) ∈ σ with p ∈ G and y = πG . Choose
q ≤ p with q ∈ G and q π ∈ τ . By the definition of f , y = πG ∈ (τf )G . Now let
y ∈ (τf )G . Then there is (π, p) ∈ τf with y = πG and p ∈ G. By the definition of
f , p π ∈ τ , and hence y = πG ∈ τG . It follows that τG = (τf )G . This shows that
ηG is a superset of (P(a))M [G] . 
Corollary 6.32. If ZFC is consistent, then so is ZFC+V6=L.
Proof. Let M be a countable transitive model of ZFC and let P ∈ M a partial
order such that every condition in P has two incompatible extensions, for instance
P = Fn(ω, 2). Let G be P-generic over M . By the previous theorem, M [G] is a
model of ZFC. Moreover, M [G] is transitive and has the same ordinals as M . By
the absoluteness of the definition of L, LM [G] = LM ⊆ M . Sice G 6∈ M we have
LM [G] 6= M [G] and thus M [G] is a model of ZFC+V6=L. 
30 STEFAN GESCHKE

7. CH is independent of ZFC
In this section we show that ZFC neither implies nor refutes CH. We already
know that ZFC does not refute CH, but we will give a forcing argument of this fact.
7.1. Forcing CH. Let M be a countable transitive model of ZFC. We define a
partial order P ∈ M such that 1P CH.
Definition 7.1. In M let
P = {f : A → P(ω) : A is a countable subset of ℵ1 }.
P is ordered by ≤=⊇.
Let G be P-generic over M . In order to show that CH holds in M [G], we first
observe the following (in M [G]):
Lemma 7.2. Let fG = G. Then fG is a function from (ℵ1 )M onto P(ω) ∩ M .
S
S
Proof. Since any two elements of G have a common extension in G, G is a
function. To see that fG is defined on all of (ℵ1 )M and onto P(ω) ∩ M , in M
we define the following dense sets:
For every α < ℵ1 let Dα := {p ∈ P : α ∈ dom(p)}. For every A ⊆ ω let
DA := {p ∈ P : A ∈ rng(p)}. It is easily checked that the Dα and the DA are dense
in P. Since G is generic over M , G intersects all Dα . Hence dom(fG ) = (ℵ1 )M .
Since G also intersects all DA , we have rng(fG ) = (P(ω))M . 
In order to show that M [G] satisfies CH, it is enough to show that (ℵ1 )M =
(ℵ1 )M [G] and (P(ω))M = (P(ω))M [G] . The first equality could fail since (ℵ1 )M
might be countable in M [G]. I.e., M [G] could contain a bijection between ω and
(ℵ1 )M . The second equality could fail since M [G] might contain new subsets of ω.
In both cases M [G] would have to contain new functions from ω into the ordinals.
We show that this cannot happen.
Lemma 7.3. Let f ∈ M [G] be a map from ω into the ordinals. Then f ∈ M .
Proof. Let f˙ ∈ M P be a name such that f˙G = f . Moreover, let p ∈ G be such that
p f˙ is a function from ω to Ord .
In M let
D = {q ≤ p : ∃g : ω → Ord(q f˙ = ǧ)}.
By the genericity of G it is enough to show that D is dense below p.
We first observe the following: Let q ≤ p and let F be P-generic over M with
q ∈ F . Then p ∈ F . By the choice of p, f˙F is a function from ω to the ordinals.
Let n ∈ ω. Then f˙F (n) = α for some ordinal α. Since M and M [F ] have the same
ordinals, α ∈ M . Hence there is r ∈ F such that r f˙(ň) = α̌. (From now on
we will drop severalˇ’s in order to improve readability.) Since G is a filter, we can
choose r ≤ q. This shows that the set of all r ∈ P for which there is some α ∈ Ord
with r f˙(n) = α in dense below p.
From now on we argue in M . (This saves us several M ’s.) A partial order Q
is σ-closed if for every descending sequence (qn )n∈ω of conditions in Q there is a
common extension q ∈ Q of the qn .
S P is σ-closed. Namely, let (pn )n∈ω be a descending sequence in P. Then p :=
n∈ω pn is a partial function from ℵ1 to P(ω) with countable domain and hence
an element of P. Clearly, p is a common extension of all the pn .
We are now in the position to show that D is dense below p. Let q ≤ p. Choose
q0 ≤ q and α0 ∈ Ord such that q0 f˙(0) = α0 . This is possible by the remark
at the beginning of this proof. Suppose we have chosen qn . As above, there is
 MODELS OF SET THEORY 31

qn+1 ≤ qn and αn+1 ∈ Ord such that qn+1 f˙(n + 1) = αn+1 . By the σ-closedness
of P, there is a common extension r of all the qn . We show that r ∈ D.
Let g : ω → Ord; n 7→ αn . For all n ∈ ω we have r ≤ qn . In particular,
r f˙(n) = αn for all n ∈ ω. Since ω is the same in all transitive models of set
theory, r f˙ = ǧ. Hence r ∈ D. It follows that D is dense below p. 

Corollary 7.4. M [G] |= CH

Proof. By Lemma 7.3 we have (P(ω))M = (P(ω))M [G] and (ℵ1 )M is uncountable in
M [G]. All ordinals below (ℵ1 )M are countable in M and hence in M [G]. It follows
that (ℵ1 )M = (ℵ1 )M [G] . By Lemma 7.2 in M [G] there is map from (ℵ1 )M [G] =
(ℵ1 )M onto (P(ω))M [G] = (P(ω))M . Hence M [G] |= |P(ω)| ≤ ℵ1 . This it is
provable in ZFC that P(ω) is uncountable and M [G] is a model of ZFC, we have
M [G] |= CH. 

Exercise 7.5. Let P be a partial order. Consider the following two player game
that lasts ω many rounds:
Let p0 = q0 = 1P . In the n-th round the first player choses a condition pn+1 ≤ qn
and the second player replies by chosing a condition qn+1 ≤ pn+1 . The second player
wins the game iff there is a common extension p ∈ P of the pn .
Suppose the second player has a winning strategy for this game. Show that for
every G that is P-generic over M , M [G] does not contain any new function from ω
into the ordinals.
Exercise 7.6. Let P = Fn(ω, 2). Show that the first player has a winning strategy
in the game described above.

8. Forcing ¬CH
In order to force the failure of CH, we start from a countable transitive model
M of ZFC+CH and generically add at least (ℵ2 )M new subsets of ω. This is rather
easily accomplished. What requires work is to show that the ℵ2 of the ground
model M actually remains ℵ2 in the extension.
In M , let κ be a cardinal > ℵ1 and let
P = Fn(κ × ω, 2) = {p : A → 2 : A is a finite subset of κ × ω}
be ordered by reverse inclusion. S
Let G be P-generic over M . Then fG = G is a function from κ × ω to 2. For
every α ∈ κ let aα = {n ∈ ω : fG (α, n) = 1}.
Lemma 8.1. If α, β ∈ κ are different, then so are aα and aβ .
Proof. Consider the set
Dα,β = {p ∈ P : ∃n ∈ ω((α, n), (β, n) ∈ dom(p) ∧ p(α, n) 6= p(β, n))}.
Let p ∈ P. Since the domain of p is finite, there is n ∈ ω such that neither
(α, n) not (β, n) are in the domain of p. Extend p to a condition q such that
(α, n), (β, n) ∈ dom(q) and q(α, n) 6= q(β, n). Then clearly, q ∈ Dα,β .
It follows that Dα,β is dense in P. Hence there is p ∈ G ∩ Dα,β . Now p ⊆ fG
and hence fG (α, n) 6= fG (β, n). This implies aα 6= aβ . 

This shows that forcing with P adds κ new subsets of ω. In order to show that κ
remains large in M [G], we need to discuss a crucial property of the forcing notion
P.
32 STEFAN GESCHKE

8.1. The countable chain condition and preservation of cardinals.

Definition 8.2. A partial order Q satisfies the countable (anti-) chain condition
(c.c.c.) if every antichain of Q is countable.
It turns out that forcing with a c.c.c. partial order does not collapse cardinals.
The following is the combinatorial foundation of the proof of this fact.
Lemma 8.3. In M , let Q be c.c.c. Then for every Q-generic filter F over M and
every function f ∈ M [F ] from some ordinal α ∈ M to the ordinals of M there is
a function f ∈ M from α to the countable sets of ordinals of M such that for all
β < α, f (β) ∈ f (β).
Proof. Let F be Q-generic over M and let f ∈ M [F ] be a function from some
ordinal α to Ord. Choose a name f˙ ∈ M Q such that f˙F = f . There is p ∈ F that
forces f˙ to be a function from α to Ord.
Let q ≤ p and let H be Q-generic over M with q ∈ H. For each β < α there is
an ordinal γβ such that f˙H (β) = γβ . This has to be forced by some r ∈ H. We say
that r decides f˙(β). Since q and r have a common extension in F , we may chose
r ≤ q to begin with.
This argument shows that for each β < α the set
Dβ = {r ∈ Q : r decides f˙(β)}
is dense below p. For each β < α let Aβ be a maximal antichain below p consisting
of conditions that decide f˙(β). Since Dβ is dense below p, Aβ is really maximal
among all the antichains below p. In particular, it is predense below p. Let
f (β) = {γ : ∃r ∈ Aβ (r f˙(β) = γ)}.
Since Q is c.c.c., Aβ and hence f (β) is countable. Since Aβ is predense below p
and p ∈ F , there is r ∈ F such that r forces f˙(β) to be an element of f (β). Hence
f˙F (β) ∈ f (β). 

Definition 8.4. A partial order Q preserves cardinals if for every cardinal κ,

1Q “κ is a cardinal”.
Q preserves cofinalities if for every ordinal α and κ = cf(α), 1Q κ = cf(α).
Lemma 8.5. If Q preserves cofinalities, then it preserves cardinals.
Proof. Let F be Q-generic over M and suppose there is a cardinal in M that ceases
to be a cardinal in M [F ]. By transfinite induction on the cardinals in M we show
that all the cardinals of M are still cardinals in M [F ].
Clearly, (ℵ0 )M = (ℵ0 )M [F ] . If κ is limit cardinal in M and all cardinals below κ
are preserved, then in M [F ], κ is still the supremum of a set of cardinals and hence
a cardinal.
If κ is a successor cardinal in M , then cf(κ) = κ in M . Since Q preserves
cofinalities, in M [F ], κ is still the cofinality of an ordinal and hence a cardinal. 

Lemma 8.6. If Q is c.c.c., then it preserves cofinalities and hence cardinals.

Proof. Let F be Q generic over M . Suppose in M [F ], α is an ordinal and κ = cf(α).
Let f : κ → α be cofinal.
If α is a successor ordinal in M [F ], then it is a successor ordinal in M as well.
Hence we may assume that α is a limit ordinal. If α is of countable cofinality in
M , then it will have the same cofinality in M [F ]. Hence we may assume that in M
the cofinality of α is uncountable.
 MODELS OF SET THEORY 33

By Lemma 8.3, in M there is a function f from κ to the countable subsets of α

such that for all β < κ, f (β) ∈ f (β). Since α is of uncountable cofinality in M , for
each β < α we have γ(β) = sup(f (β)) < α.
Now for each β < α, f (β) ≤ γ(β). Since f is cofinal in α, so is g. Hence in M ,
cf(α) ≤ κ. But since κ is the least size of a cofinal subset of α in M [F ], there is no
cofinal subset of α of size < κ in M . Therefore in M , κ = cf(α).
It follows that Q preserves cofinalities. 

We now have to show that P = Fn(κ×ω, 2) is actually c.c.c. This fact is also non-
trivial and can be shown using the ∆-System Lemma, which is pure combinatorics.

Lemma 8.7 (∆-System Lemma). Let F be an uncountable family of finite sets.

Then there are a finite set r and an uncountable family D ⊆ F such that for all
distinct s, t ∈ D, s ∩ t = r. D is called a ∆-system with root r.

Proof. After throwing

S away members of the family, we may assume that F is of
size ℵ1 . Now F is of size ℵ1 Hence we may also assume that F consists of subsets
of ℵ1 . Moreover, we may assume that all elements of F have the same size, say n.
Let (aα )α,ω1 be a 1-1 enumeration of F.
Let fi : ℵ1 → ℵ1 , i < n be functions such that for all α < ℵ1 , aα = {fi (α) : i < n}
and f0 (α) < · · · < fn−1 (α).
Clearly, if fi is unbounded, then for all j < n with i < j, fj is unbounded as well.
Let k ≤ n be maximal such that for all i < k, fi is bounded. Let α < ℵ1 be strictly
greater than all the elements of the ranges of the fi , i < k. Since α is countable,
α has only countably many finite subsets. This shows that at least one of fi is
unbounded and hence k < n. Since fk is unbounded, there is an uncountable set
I ⊆ ℵ1 such that fk is 1-1 on I. Moreover, there is a finite set r ⊆ α such that
J = {β ∈ I : {fi (β) : i < k} = r} is uncountable.
By recursion we now choose a strictly increasing sequence (βγ )γ<ℵ1 in J such
that max(r) < fk (β0 ) and for γ < γ 0 < ℵ1 we have fn−1 (βγ ) < fk (βγ 0 ). This is
possible since fk is unbounded on J.
It is easily checked that D = {aβγ : γ < ℵ1 } is a ∆-system with root r. 

Lemma 8.8. For every set X, Fn(X, 2) is c.c.c.

Proof. Let A ⊆ Fn(X, 2) be uncountable. By the ∆-System Lemma, A has an

uncountable subset B such that the supports of the elements of B form a ∆-system
with some root r ⊆ X. Since there are only finitely many functions from r to 2, B
contains two different conditions p and q that agree on r. Since the sets dom(p) \ r
and dom(q)\r are disjoint, p∪q is a function. It follows that p and q are compatible.
Hence A is not an antichain. 

Corollary 8.9. CH fails in M [G].

Proof. Recall that G is Fn(κ × ω, 2)-generic over M where κ is a cardinal > ℵ1 in

M . Since Fn(κ × ω, 2) is c.c.c., M and M [G] have the same cardinals. By Lemma
8.1, ω has at least κ subsets in M [G]. Hence M [G] |= ¬CH. 

8.2. Nice names and the size of 2ℵ0 . Let M , κ, P and G be as before. We want
to compute the actual value of 2ℵ0 in M [G]. We do this by finding a small set of
P-names such that every a ∈ P(ω) ∩ M [G] has a name in that set.

Exercise 8.10. Let a ∈ M [G]. Show that in M there is a proper class of names σ
with σG = a.
34 STEFAN GESCHKE

Definition 8.11. Let Q be a partial order. A Q-name σ is a nice name for a subset
of ω if there is a family (An )n∈ω of antichains of Q such that
[
σ = {(ň, p) : p ∈ An } = ({ň} × An ).
n∈ω

Lemma 8.12. Let F be Q-generic over M . Then for every a ∈ P(ω) ∩ M [F ] there
is a nice name σ ∈ M for a subset of ω such that a = σF .
Proof. Fix a name τ ∈ M such that a = τF . For each n ∈ ω let An be a maximal
antichain in the set {p ∈ Q : p n ∈ τ }. Clearly, σ = {(ň, p) : p ∈ An } is a nice
name for a subset of ω.
We have to verify that σF = a. Let n ∈ σF . Then for some p ∈ An , p ∈ F . By
the choice of An , p n ∈ τ and hence n ∈ τF = a. This shows σF ⊆ a.
On the other hand, if n ∈ a, then there is (π, p) ∈ τ such that p ∈ F and πF = n.
For some q ∈ F , q π = n. We can choose q ≤ p. Now q n ∈ τ . Since An is a
maximal antichain in the set of conditions that force n ∈ τ , An is predense below
q. Since F is generic and q ∈ F , there is r ∈ F ∩ An . Now r n ∈ σ and hence
n ∈ σF . This shows a ⊆ σF and hence a = σF . 
Exercise 8.13. Recall the definition of the sets aα , α < κ, mentioned in Lemma
8.1. Write down explicitly a nice name for each aα .
Lemma 8.14. Let Q be c.c.c. Then there are at most |Q|ℵ0 nice Q-names for
subsets of ω.
Proof. Since Q is c.c.c., all antichains of Q are countable. Hence Q has at most
|Q|ℵ0 antichains. A nice name for a subset of ω is essentially just a function from
ω into the set of antichains of Q. It follows that there are at most
(|Q|ℵ0 )ℵ0 = |Q|ℵ0
nice names for subsets of ω. 
Theorem 8.15. In M , let P = Fn(κ × ω, 2). Let G be P-generic over M . Then
M [G] |= 2ℵ0 = (κℵ0 )M .
Proof. Since κ is infinite, (κℵ0 )M ≥ (2ℵ0 )M . Since P is c.c.c., it preserves cardinals.
By Lemma 8.1, (2ℵ0 )M [G] is at least κ. Since (2ℵ0 )ℵ0 = 2ℵ0 , (2ℵ0 )M [G] is at least
(κℵ0 )M [G] . Clearly, (κℵ0 )M ≤ (κℵ0 )M [G] On the other hand, in M there are at most
κℵ0 nice names for subsets of ω and hence |P(ω)|M [G] ≤ (κℵ0 )M and the theorem
follows. 
Corollary 8.16. Let M , G and κ be as above. If M satisfies GCH and cf(κ) >
ℵ0 , then in M [G], 2ℵ0 = κ. If M satisfies GCH but cf(κ) = ℵ0 , then in M [G],
2ℵ0 = κ+ .
Proof. The corollary follows from the previous theorem and the fact that under
GCH we have κℵ0 = κ if cf(κ) > ℵ0 and κℵ0 = κ+ if κ is an infinite cardinal of
countable cofinality. 
Exercise 8.17. Prove the statement about the size of κℵ0 under GCH used in the
proof of the previous corollary.
 MODELS OF SET THEORY 35

9. Martin’s Axiom
Having produced models of ZFC in which CH fails, there are several natural
questions.
1. Let κ < 2ℵ0 be a cardinal. Is 2κ ≤ 2ℵ0 ?
2. How many measure zero sets are needed to cover the real line?
3. Let X be a topological space. A ⊆ X is nowhere dense if the closure of A
has an empty interior. The Baire Category Theorem says (in particular)
that R is not the union of countably many nowhere dense sets. But how
many nowhere dense sets are needed to cover all of R?
4. A family A ⊆ P(ω) is almost disjoint if all A, B ∈ A with A 6= B have a
finite intersection. An easy application of Zorn’s Lemma shows that every
almost disjoint family of subsets of ω is contained in a maximal one. What
is the minimal size of an infinite maximal almost disjoint family? (An
easy argument shows that no countably infinite almost disjoint family is
maximal.)
All of these questions have an obvious answer under CH. Martin’s Axiom also
answers all these questions, but is consistent with ¬CH. In fact, MA is only inter-
esting if CH fails since under CH it does not say anything new as it follows from
CH.
Definition 9.1. Let κ be a cardinal. MAκ says that for every c.c.c. partial order P
and every family D of size κ of dense subsets of P there is a D-generic filter G ⊆ P.
Martin’s Axiom (MA) is the statement
∀κ < 2ℵ0 (MAκ ).
9.1. Martin’s Axiom and Souslin’s Hypothesis. The original motivation to
introduce Martin’s Axiom lies in Souslin’s Hypothesis. A linear order (L, ≤) is
c.c.c. if there is no uncountable family of pairwise disjoint open intervals of L. L
is separable if it has a countable dense subset. L is connected if it is not the union
of two disjoint open subsets.
It is relatively easy to show that every connected separable linear order without
endpoints is isomorphic to R. Suslins Hypothesis (SH) is the statement that every
connected c.c.c. linear order without endpoints is isomorphic to R. CH neither
implies nor refutes SH. (Both directions of this independence result are due to
Jensen.) SH fails in L. The consistency proof of MA+¬CH is a relatively straight
forward generalization generalization of Solovay’s and Tennenbaum’s proof of the
consistency of SH. Martin’s name is attached to the axiom because he observed that
the consistency proof of SH could be adapted to show the consistency of something
much stronger, namely of MA with an arbitrarily large size of 2ℵ0
36 STEFAN GESCHKE

We now show how MAℵ1 implies SH. A Souslin line is a connected linear order
without endpoints that is c.c.c. but not separable. Since every separable connected
linear order without endpoints is isomorphic to the real line, SH holds and only if
there is no Souslin line.
Definition 9.2. A partial order (T, ≤) is a tree if for all t ∈ T the set {s ∈ T : s ≤ t}
is well ordered. If T is a tree and t ∈ T , then the height htT (t) of t in T is the
order type of the set {s ∈ T : s < t}, i.e., the unique ordinal α such that (α, ∈) is
isomorphic to ({s ∈ T : s < t}, <). For an ordinal α, the α-th level Levα (T ) of T is
the set of all nodes t ∈ T of height α.
A branch of a tree T is a maximal chain in T . An antichain in a tree T is a family
of pairwise incomparable elements of the tree. A tree is Souslin if it is uncountable
and has neither uncountable chains nor antichains.
Lemma 9.3. If there is a Souslin line, then there is a Souslin tree.
Actually the existence of a c.c.c. linear order that is not separable implies the
existence of a Souslin line by roughly the same proof as below. Connectedness
simplifies the proof a tiny little bit. The main information that connectedness gives
us is that L has to be a dense linear order, i.e., if a, b ∈ L are such that a < b, then
(a, b) is nonempty (and in fact infinite).
Proof. Let (L, ≤) be a Souslin line. The Souslin tree T that we are going to con-
struct will consist of nonempty open intervals in L ordered by reverse inclusion. By
recursion on α ∈ Ord we define the α-th level of the tree T .
Let Lev0 (T ) = {L}. We consider L as a nonempty open interval of L. Suppose
we have constructed the β-th level of T and α = β+1. For each interval I ∈ Levβ (T )
let AI be a an infinite maximal disjoint family of open intervals contained in I. This
is possible since by connectedness every nonempty open interval of L has at least
two disjoint nonempty open subintervals. Let
[
Levα (T ) = {AI : I ∈ Levβ (T )}.
Now assume that α is a limit ordinal and for all β < α we have already defined
Levβ (T ). Let Levα (T ) be a maximal of pairwise disjoint, nonempty open intervals
I with the property that for every β < α there is J ∈ Levβ (T ) with I ⊆ J. This
finishes the definition of T . Obviously, the levels of T are eventually empty.
Every antichain of T is a family of pairwise disjoint, nonempty open intervals of
L. Since L is c.c.c., every antichain of T is countable. In particular, every level of
T is countable.
Now suppose that T has an uncountable chain. Since T is a tree, every chain
of T is wellordered. Hence, if T has an uncountable chain, it has a chain of the
form (Iα )α∈ℵ1 , where Iβ ⊆ Iα if α < β. Now the sequence of left endpoints of the
Iα has an uncountable increasing subsequence or the sequence of right endpoints
of the Iα has an uncountable decreasing subsequence. Without loss of generality
assume the former and let (aα )α∈ℵ1 be a strictly increasing sequence in L. Then
((aα , aα+1 ))α∈ℵ1 is an uncountable family of pairwise disjoint, nonempty open in-
tervals in L, a contradiction. It follows that T has no uncountable chains. Note
that this implies that Levℵ1 (T ) is empty.
It remains to show that T is uncountable. Let D be the collection of all the
endpoints of the intervals I ∈ T that are not +∞ or −∞. We claim that D is dense
in L. Let a, b ∈ L be such that a < b. It is enough to find some interval I ∈ T that
intersects (a, b) but is not a superset of (a, b) since in this case (a, b) has to contain
an endpoint of I.
Let α be minimal with the property that (a, b) is disjoint from every element of
Levα (T ).
 MODELS OF SET THEORY 37

We first observe that α is a limit ordinal. Otherwise α = β + 1 for some ordinal

β and there is I ∈ Levβ (T ) such that I has a nonempty intersection with (a, b).
But now AI was chosen to be a maximal family of paiwise disjoint, nonempty open
subintervals of I. Since I ∩ (a, b) is a nonempty open subinterval of I, there is
J ∈ AI ⊆ Levα (T ) such that J intersects (a, b), a contradiction.
By the choice of α, for every β < α, there is I ∈ Levβ (T ) with (a, b) ∩ I 6= ∅. If
some of these I’s intersects (a, b) but is not a superset of (a, b), we are done. Hence
we may assume that for each β < α there is I ∈ Levβ (T ) such that (a, b) ⊆ I. But
by the definition of Levα (T ), this means that some J ∈ Levα (T ) intersect (a, b),
contradicting the choice of α.
We have thus found a dense subset of L of size at most |T |. Since L is not
separable, T is uncountable. It follows that T is a Souslin tree. 
The converse of this lemma is also true: if there is a Souslin tree, then there is
a Souslin line.
Theorem 9.4. MAℵ1 implies SH.
Proof. We show that under MAℵ1 there are no Souslin trees. Assume there is a
Souslin tree (T, ≤). We have to prune the tree a bit. Namely, we remove all those
nodes t ∈ T such that for some α < ℵ1 , there is no s ∈ Levα (T ) with t ≤ s. Let
T 0 denote the pruned tree. Now for every t in T and every α < ℵ1 with α > htT (t)
there is a node s ∈ T 0 of height α such that t ≤ s.
If not, then for some t ∈ T 0 and α < ℵ1 with α > htT (t), t has no successors
at the α’s level of T 0 . But this implies that all successors of t in Levα (T ) have
only countably many successors in T . It follows that t has only countably many
successors in T , a contradiction since t ∈ T 0 .
Now let P be the tree T 0 , but ordered by the relation ≥. Now antichains in the
tree T 0 correspond to antichains in the partial order P. Since with T also T 0 has
only countable antichains, P is c.c.c.
Since every t ∈ T 0 has successors of every countable height, for every α < ℵ1 the
set Dα = {t ∈ T 0 : htT 0 (t) > α} is dense in P. By MAℵ1 there is a (Dα )α<ℵ1 -generic
filter G ⊆ P. It is easily checked that G is an uncountable chain in T 0 and hence in
T . A contradiction. 
9.2. The Baire Category Theorem and Martin’s Axiom. In its strongest
form the Baire Category Theorem states that no complete metric space and no
compact space is the union of countably many nowhere dense sets. We concentrate
on compact spaces.
A topological space X is c.c.c. if every family of pairwise disjoint open subsets
of X is countable.
Theorem 9.5. MA is equivalent to the statement “no c.c.c. compact space is the
union of fewer than 2ℵ0 nowhere dense sets”.
The proof of this theorem needs a couple of lemmas. Before we start proving it,
we point out an important consequence.
Corollary 9.6. Under MA, R is not the union of fewer than 2ℵ0 nowhere dense
sets.
Proof. R is not compact, but [0, 1] is and R is homeomorphic to (0, 1). A subset of
(0, 1) is nowhere dense in (0, 1) iff it is nowhere dense in [0, 1]. The singletons {0}
and {1} are nowhere dense in [0, 1]. It follows that the number of nowhere dense
subsets of [0, 1] needed to cover [0, 1] is the same as the number of nowhere dense
subsets of (0, 1) needed to cover (0, 1). In particular, (0, 1) and therefore R cannot
be covered by fewer than 2ℵ0 nowhere dense sets. 
38 STEFAN GESCHKE

Our first step in the proof of Theorem 9.5 is to associate every Boolean algebra
with a compact space. This connection between Boolean algebras and certain
compact spaces is known as Stone Duality.
Definition 9.7. Let A be a Boolean algebra and F ⊆ A. We say that F is an
ultrafilter of A iff F is a maximal filter in the partial order A \ {0} or equivalently,
if F is a filter in A \ {0} and for each a ∈ A either a ∈ F or ¬a ∈ F .
Let Ult(A) denote the set of ultrafilters of A topologized by declaring the sets
of the form [a] = {F ∈ Ult(A) : a ∈ F }, a ∈ A, as open. I.e., a subset of Ult(A) is
open iff it is a union of sets of the form [a]. We call the sets [a] basic open.
Exercise 9.8. A set S ⊆ A has the finite intersection property if for all n and
all a1 , . . . , an ∈ A, a1 ∧ · · · ∧ an 6= 0. By Zorn’s Lemma, every set with the finite
intersection property is contained in a maximal set with the finite intersection
property. Show that a maximal set S ⊆ A with the finite intersection property is
an ultrafilter of A.
Lemma 9.9. For every Boolean algebra A, Ult(A) is a compact space, the Stone
space of A.
Proof. We first show that Stone spaces are Hausdorff, i.e., for two distinct points
x, y ∈ Ult(A) there are disjoint open sets U, V ⊆ Ult(A) such that x ∈ U and
y ∈V.
Let x, y ∈ Ult(A) be such that x 6= y. Then there is a ∈ A such that a ∈ x and
a 6∈ y or vice versa. Without loss of generality we assume the first. Since y is an
ultrafilter, ¬a ∈ y. Now [a] and [¬a] are disjoint open sets, the first containing x,
the second containing y. This shows Hausdorffness.
Now let O be an open cover of Ult(A). For every x ∈ Ult(A) choose S ax ∈ A such
that for some Ux ∈ O we have x ∈ [ax ] ⊆ Ux . Clearly, Ult(A) = x∈Ult(A) [ax ]. If
there are finitely many points x1 , . . . , xn ∈ Ult(A) such that Ult(A) = [ax1 ] ∪ · · · ∪
[axn ], then {Ux1 , . . . , Uxn } is a finite subcover of O and we are done.
Suppose there are not finitely many x such that the corresponding [ax ] union
up to Ult(A). In this case the family {Ult(A) \ [ax ] : x ∈ Ult(A)} has the finite
intersection property, i.e., no finite intersection of sets from the family is empty.
This easily translates to the finite intersection property of {¬ax : x ∈ Ult(A)}. By
the previous exercise there isSan ultrafilter y of A that extends {¬ax : x ∈ Ult(A)}.
It is easily checked that y 6∈ x∈Ult(A) [ax ], a contradiction. 
Exercise 9.10. Let A be a Boolean algebra. For each a ∈ A let Da = {b ∈ A : b ≤
a ∨ b ⊥ a}. Then each Da is a dense subset of A. If F ⊆ G is a (Da )a∈A -generic
filter, then F is an ultrafilter.
Lemma 9.11. Let A be a Boolean algebra and S ⊆ Ult(A). If S is nowhere dense
in Ult(A), then the set
DS = {a ∈ A \ {0} : [a] ∩ S = ∅}
is dense in A.
On the other hand, if D is a dense subset of A, then the set
[
{x ∈ Ult(A) : x 6∈ {[a] : a ∈ D}}
is nowhere dense in Ult(A).
Proof. Let S be nowhere dense. Taking the closure of S we can assume that S is
closed. Let a ∈ A \ {0}. Since S is closed and nowhere dense, [a] 6⊆ S. Hence [a] \ S
is a nonempty open subset of Ult(A). Since the topology on Ult(A) is generated
by the basic open sets, there is b ∈ A \ {0} such that [b] ⊆ [a] \ S and thus b ∈ DS
and b ≤ a. This shows the density of DS .
 MODELS OF SET THEORY 39

S
Now let D be a dense subset of A and let UD = {[a] : a ∈ D}. As a union
of open sets, UD is open. In order to show that Ult(A) \ UD is nowhere dense, it
is enough to show that UD is dense in Ult(A). Let V ⊆ Ult(A) be nonempty and
open. Then there is a ∈ A \ {0} such that [a] ⊆ V . By the density of D, there is
b ∈ D such that b ≤ a. Now clearly [b] ⊆ V ∩ UD . This shows that UD is dense in
Ult(A). 

We need yet another observation for the proof of Theorem 9.5 that will also be
useful when we show the consistency of Martin’s Axiom with the negation of CH.
Lemma 9.12. Let κ be a infinite cardinal. Then MAκ holds iff MAκ holds re-
stricted to ccc partial orders of size κ.
Proof. We show only the nontrivial direction. Let (P, ≤) be a ccc partial order and
let D be a family of κ dense subsets of P.
Let λ be sufficiently large and choose an elementary submodel M of Vλ of size λ
such that D ⊆ M and (P, ≤) ∈ M . We consider the partial order Q = M ∩ P. Like
P, Q is ccc. For each q ∈ Q and each D ∈ D, M believes that there is some p ∈ D
such that p ≤ q. It follows that for all q ∈ Q and all D ∈ D there is p ∈ M ∩ D
such that p ≤ q. Hence the sets D ∩ M , D ∈ D, are dense in Q. By MAκ restricted
to partial orders of size κ, there is a filter F ⊆ P that is generic for the collection
{D ∩ M : D ∈ D} of dense subsets of Q. F extends to a filter G ⊆ P. G is D
generic. 

Proof of Theorem 9.5. Assume MA and let X be a c.c.c. compact space. Let P
denote the collection of all nonempty open subsets of X ordered by inclusion. Since
X is c.c.c., so is P. Let F be a collection of fewer than 2ℵ0 nowhere dense subsets
of X. We may assume that each member of F is a closed set. For each S ∈ F let
DS = {U ∈ P : cl(U ) ∩ S = ∅}.
Let S ∈ F and let U ⊆ X be nonempty and open. Since S is closed and nowhere
dense, U \ S is nonempty and open. Let V ⊆ X be nonempty, open and such that
cl(V ) ⊆ U \ S. By the definition of DS , V ∈ DS . It follows thatTDS is dense in P.
By MA, there is a (DS )S∈F -generic filter G ⊆ P. Let T = {cl(U ) : U ∈ G}.
Since G has the finite intersection property and X is compact, T 6= ∅. By the
genericity of G, T is disjoint from every S ∈ F, showing that F does not cover X.
On the other hand, assume that no c.c.c. compact space is the union of fewer
than 2ℵ0 nowhere dense sets. By Lemma 9.12, in order to prove Martin’s Axiom,
it is enough to prove Martin’s Axiom restricted to partial orders of size < 2ℵ0 . Let
P be a partial order of size < 2ℵ0 . Let A = ro(P) and X = Ult(A).
Claim 9.13. X is c.c.c.
Let A be a family of pairwise disjoint, nonempty open subsets of X. We may
assume that every element of A is of the form [a] for some a ∈ A. We may further
assume that every element of A is of the form [a] where a = reg p for some p ∈ P.
Since the elements of A are pairwise disjoint, the corresponding elements of P are
pairwise incompatible. Since P is c.c.c., A is countable. This shows the claim.
Now let D be a collection of fewer than 2ℵ0 dense subsets of P. Since P is of size
< 2ℵ0 , we may assume that for all p, q ∈ P, the set
Dp,q = {r ∈ P : r ≤ p, q or r ⊥ p or r ⊥ q}
is in D. Let e : P → ro(P) be the natural embedding. Since e[P] is dense in ro(P),
the images of the SD ∈ D in ro(P) are also dense. It follows that for each D ∈ D
the set SD = X \ {[e(p)] : p ∈ D} is nowhere dense in X.
40 STEFAN GESCHKE

By our assumption, X is not the union of the SD , D ∈ D. It follows that

there is an ultrafilter F of A that is not in any S of the sets SD , D ∈ D. Let
G = {p ∈ P : e(p) ∈ F }. Hence, if D ∈ D, F ∈ {[e(p)] : p ∈ D}. But this implies
that there is some p ∈ D such that e(p) ∈ F . By the definition of G, p ∈ G. This
shows that G has a nonempty intersection with D. We still have to show that G is
in fact a filter. It is clear that if p ∈ G and p ≤ q, then q ∈ G.
Now let p, q ∈ G. If p ⊥ q, then e(p) and e(q) are incompatible in ro(P). Hence
e(p) and e(q) cannot both be elements of F . It follows that any two elements of G
are compatible. Since Dp,q ∈ D, there is r ∈ G ∩ Dp,q . Since any two elements of
G are compatible, r is compatible to both p and q. But in this case, r ≤ p, q. This
shows that G is a D-generic filter, proving that MA holds for ccc partial orders of
size < 2ℵ0 . 
9.3. More consequences of MA. We first observe that Martin’s Axiom implies
that the real line is not covered by less than continuum many sets of measure 0.
Theorem 9.14. MAκ implies that R is not the union of κ sets of measure 0.
Proof. First observe that it is enough to show that [0, 1] is not the union of κ sets of
measure 0. Let P be the partial order of all measurable subsets of [0, 1] of measure
> 0 ordered by inclusion. We first show that P is ccc.
Let A ⊆ P be an uncountable antichain. Then for distinct p, q ∈ A, p ∩ q is of
measure 0. For each n ∈ ω let
1
An = {p ∈ A : the measure of p is at least 2−n }.
S
Now A = n∈ω An and hence for some n ∈ ω, AnSis uncountable. Let (pi )i∈ω be a
1-1-sequence in An . For each k ∈ ω let qk = pk \ i<k pk . Now the qk are pairwise
disjoint and the measure of each qk is equal to the measure of pk since pk and qk
only differ by a set of measure 0.
1
The qk are pairwise disjoint and of measure at least 2−n . It follows that their
union is of infinite measure, contradicting the fact that each qk is a subset of [0, 1].
It follows that P is indeed ccc.
For a set N ⊆ [0, 1] of measure 0 let
DN = {p ∈ P : cl(p) ∩ N = ∅}.
Here cl(p) denotes the closure of the set p. Since N is of measure 0, for every p ∈ P,
p \ N is of positive measure. Every set of positive measure contains a closed set of
positive measure. It follows that DN is dense in P.
Now let G ⊆ P be a filter that intersects DN . Since G is a filter, it is closed
under finite intersections. Since [0, 1] is compact, every family of closed subsets
of
T [0, 1] with the finite intersection property has a nonempty
T intersection. Hence
{cl(p) : p ∈ G} is nonempty. Since G intersects DN , {cl(p) : p ∈ G} is disjoint
from N .
If N is a family of size κ of sets of measure 0 and
T MAκ holds, then there is a
filter G ⊆ P that intersects each DNS, N ∈ N . Now {cl(p) : p ∈ G} is a nonempty
subset of [0, 1] that is disjoint from N . In particular, the family N does not cover
[0, 1]. 
The next goal is to prove that for infinite κ, MAκ implies that 2κ = 2ℵ0 . Let
us we look at two similar forcing notions, one of which will be used for the above
statement.
Definition 9.15. a) Hechler forcing H is the partial order with conditions of the
form p = (fp , Fp ) where for some n ∈ ω, fp : n → ω and Fp is a finite set of functions
from ω to ω. For p, q ∈ H, let p ≤ q if fq ⊆ fp and for all n ∈ dom(fp ) \ dom(fq )
and all f ∈ Fq , fp (n) ≥ f (n).
 MODELS OF SET THEORY 41

b) Let A be an infinite set of subsets of ω that pairwise have finite intersection,

i.e., an almost disjoint family. Almost disjoint forcing AD is the partial order with
conditions of the form p = (fp , Ap ) where for some n ∈ ω, fp : n → 2 and Ap is a
finite subset of A. For p, q ∈ AD, let p ≤ q if fq ⊆ fp and for all A ∈ Aq and all
n ∈ (A ∩ dom(fp )) \ dom(fq ), fp (n) = 0. Observe that AD depends on the choice
of the almost disjoint family A.
We will show that both almost disjoint forcing and Hechler forcing are ccc. In
fact, they satisfy a stronger property.
Definition 9.16. A subset S of a partial order P is centered if any finitely many
elements of S have a common lower bound in P. A partial order P is σ-centered if
it is the union of countably centered set.
Lemma 9.17. Every σ-centered partial order is ccc.
Lemma 9.18. Both Hechler forcing and almost disjoint forcing are σ-centered.
Proof. In both cases it is enough to observe that there are only countably many
first components of conditions and the set of conditions with a fixed first component
is centered. 

Definition 9.19. A family F ⊆ ω ω is bounded if there is g : ω → ω such that for

all f ∈ F and all but finitely many n ∈ ω, f (n) ≤ g(n).
Theorem 9.20. Let κ be an infinite cardinal and assume MAκ . Then no almost
disjoint family of size κ of subsets of ω is maximal almost disjoint. Every family
of size κ of functions from ω to ω is bounded.
Proof. Let A be an almost disjoint family of size κ of subsets of ω. Let AD be
the almost disjoint forcing with respect to A. For each A ∈ A, DA = {p ∈ AD :
A ∈ Ap } is dense in AD. Also, for each n ∈ ω, the set Dn = {p ∈ AD : ∃m ∈
dom(fp ) \ n(fp (m) = 1)} is dense in AD.
By MASκ , let G ⊆ AD be a filter that intersects all DA ,A ∈ A, and all Dn , n ∈ ω.
Let f = p∈G fp . Then f : ω → 2 is the characteristic function of an infinite set
B ⊆ ω. We show that B is almost disjoint from all A ∈ A and hence A is not a
maximal almost disjoint family. It is enough to prove the following:
Claim 9.21. Let p ∈ G ∩ DA . Then A ∩ B ⊆ dom(fp ).
For the proof of the claim let n ∈ B \ dom(fp ). Then f (n) = 1 and hence there
is q ∈ G such that n ∈ dom(fq ) and fq (q) = 1. Wlog we may assume that q ≤ p.
By the definition of ≤ and since A ∈ Ap , n 6∈ A, finishing the proof of the claim.
The proof of the corresponding statement for families of functions is almost the
same using Hechler forcing. 

We now show that for every infinite cardinal κ, MAκ implies 2κ = 2ℵ0 . In order
to do this, we have to code subsets of κ by subsets of ω. This is done by almost
disjoint coding, which is due to Solovay.
Fix an almost disjoint family A of size κ of subsets of ω consisting of infinite
sets. Instead of subsets of κ, we code subsets of A by subsets of ω, in the following
way:
Lemma 9.22. Assume MAκ . Then for each B ⊆ A there is C ⊆ ω such that for
all B ∈ B, B ∩ C is finite and for all A ∈ A \ B, C ∩ A is infinite. In other words,
C is almost disjoint from every element of B and not almost disjoint from every
element of A \ B.
42 STEFAN GESCHKE

Proof. We use almost disjoint forcing for the almost disjoint family B. Let for n ∈ ω
and A ∈ B let Dn and DA be the dense sets of almost disjoint forcing defined in
the proof of Theorem 9.20. For all A ∈ A \ B and all n ∈ ω let
n
EA = {p ∈ AD : ∃m ∈ A \ n(m ∈ dom(fp ) ∧ fp (m) = 1)}.
n
Then each of the sets EA is dense in AD.
Now let G ⊆ AD be a filter that intersects all the dense sets Dn , n ∈ ω, the sets
DA , A ∈ B, and EA , n ∈ ω and A ∈ A \ B. Let f = p∈G fp and let C = f −1 (1).
n
S
Then C has the desired properties. 
Corollary 9.23. If κ is an infinite cardinal and MAκ holds, then 2ℵ0 = 2κ .
9.4. Iterated forcing. We are going to show the consistency of MA with ¬CH
by means of iterated forcing. The strategy is as follows: we start with a countable
transitive model M of GCH and construct an increasing sequence (Mα )α≤(ℵ2 )M
of models of set theory and a sequence (Gα )α<(ℵ2 )M such that M0 = M and for
all α < (ℵ2 )M , Gα is Qα -generic over Mα where Qα ∈ Mα is a partial order and
Mα+1 = Mα [Gα ]. For each α < (ℵ2 )M ,
Mα |= “Qα is a c.c.c. partial order of size ℵ1 ”.
Most of the Qα add new reals, all the Mα , α < (ℵ2 )M satisfy CH, the final model
M(ℵ2 )M satisfies 2ℵ0 = ℵ2 and no cardinals are collapsed in the process. We choose
Qα in such a way that in the final model the following holds: whenever P is a c.c.c.
forcing of size ℵ1 and D is a family of fewer than ℵ2 dense subsets of P, then there
is α < (ℵ2 )M = ℵ2 such that P = Qα and D ⊆ Mα . In particular, Gα , which is an
element of M(ℵ2 )M , intersects every D ∈ D, showing that there is a D-generic filter
of P. By Lemma 9.12, this is enough to show MA in M(ℵ2 )M .
Our strategy has one major problem: If α ≤ (ℵ2 )M is a limit ordinal, how do we
define Mα ? It is tempting to choose Mα simply as the union of the previous Mβ ,
β < α. However, there is no reason why this union should be a model of ZFC. A
strategy that works, however, is to define a single forcing notion Q in the ground
model M such that every Q-generic filter G over M codes a sequence (Gα )α<(ℵ2 )M
of filters and a sequence (Mα )α≤(ℵ2 )M of models of set theory as above.
Let us first consider the case where we want to iterate just two forcings. Let P
be a partial order in M and let G be P-generic over M . Let Q be a forcing notion
in M [G] and let F be Q-generic over M [G]. Then M [G, F ] = M [G][F ] is a model
of set theory. We want to represent M [G, F ] as a single generic extension of M
with respect to a certain forcing notion in M . The problem is that Q might not
be an element of M . Hence, from the point of view of M , we can access Q only in
terms of a P-name Q̇. For simplicity, we would like to assume that 1P forces that Q̇
is a partial order. Moreover, typically the partial order Q has a simple definition in
M [G] such as “Q is the partial order of all closed subsets of R of positive measure”.
We would like to have a name Q̇ that is forced by 1P to satisfy this definition.
All this is made possible by the Existential Completeness Lemma.
Lemma 9.24 (Existential Completeness Lemma, respectively Maximality Princi-
ple). Let ϕ(x, y1 , . . . , yn ) be a formula in the language of set theory. Let P be a
partial order, τ1 , . . . , τn P-names and p ∈ P. Suppose that
p ∃xϕ(x, τ1 , . . . , τn ).
Then there is a P-name σ such that
p ϕ(σ, τ1 , . . . , τn ).
The proof of this lemma uses the following sublemma:
 MODELS OF SET THEORY 43

Lemma 9.25. Let P be a partial order, A ⊆ P an antichain and (σp )p∈A a family
of P-names. Then there is a P-name σ such that for all p ∈ A, p σ = σp .
Proof. Let
σ = {(η, r) : ∃p ∈ A∃q ∈ P(r ≤ p ∧ r ≤ q ∧ (η, q) ∈ σp )}.
Now, if p ∈ A, we claim that p σ = σp .
Let G be generic over the ground model with p ∈ G. Let (η, r) ∈ σ be such
that r ∈ G and ηG = x. Since A is an antichain, p is the unique element of A that
is in G. Hence r ≤ p. By the definition of σ there is q ∈ P such that r ≤ q and
(η, q) ∈ σp . Now q ∈ G and hence x = ηG ∈ (σp )G . This shows σG ⊆ (σp )G .
On the other hand, let x ∈ (σp )G . Fix (η, q) ∈ σp such that q ∈ G and ηG = x.
since p and q are both in G, there is a common extension r ∈ G of both p and q. By
the definition of σ, (η, r) ∈ σ and hence x = ηG ∈ σG . This shows (σp )G ⊆ σG . 
Proof of the Existential Completeness Lemma. By the definition of the truthvalue
[[∃xϕ(x, τ1 , . . . , τn )]] the set
D = {q ≤ p : ∃η ∈ V P (q ϕ(η, τ1 , . . . , τn ))}
is predense below p. Since D is open, it is actually dense below p. Choose a maximal
antichain A in D. For each q ∈ A choose a name ηq such that q ϕ(ηq , τ1 , . . . , τn ).
By the previous lemma there is a name σ such that for all q ∈ A, q σ = ηq . Since
A is predense below p,
p ϕ(σ, τ1 , . . . , τn ).

Definition 9.26. Let P be a partial order and let Q̇ be a P-name for a partial
order, i.e., a P-name such that 1P forces Q̇ to be a partial order. The two-step
iteration of P and Q̇ is the partial order P ∗ Q̇ consisting of all pairs (p, q̇) where p
is a condition in P and q̇ is a P-name such that p q̇ ∈ Q̇.
Let ≤P denote the order on P and let ≤ ˙ Q be a P-name for the order on Q̇. We
define the order ≤ on P ∗ Q̇ as follows: (p1 , q̇1 ) ≤ (p2 , q̇2 ) iff p1 ≤P p2 and
p1 q̇1 ≤˙ Q q̇2 .
We will drop the subscripts P and Q if they are clear from the context.
Observe that with this definition of the iteration of P and Q̇, P ∗ Q̇ turns out to
be a proper class. Hence we redefine P ∗ Q̇ as
{(p, q̇) ∈ P ∗ Q̇ : rk(q̇) < rk(P) + ω, rk(Q̇) + ω}
even though this definition obviously lacks the elegance of the first definition. Now
whenever p ∈ P and q̇ is a P-name such that p q̇ ∈ Q̇, then there is a P-name ṙ
such that (p, ṙ) ∈ P∗ Q̇ and p ṙ = q̇. Whenever we choose a name for an element of
Q̇ we implicitly assume that we only consider names of rank < rk(P) + ω, rk(Q̇) + ω.
44 STEFAN GESCHKE

Now let 1̇Q be a name for the largest element of Q̇. Then obviously,
i : P → P ∗ Q̇ : p 7→ (p, 1̇P )
is an embedding of partial orders. If H is P ∗ Q̇-generic over the ground model M ,
We define G = i−1 [H] and F = {q̇G : ∃p ∈ G((p, q̇) ∈ H)}.
Lemma 9.27. G is P-generic over M and F is Q̇G -generic over M [G]
Proof. It is easily checked that G is a filter. Let D ∈ M be a dense subset of P.
Then the set D0 = {(p, q̇) ∈ P ∗ Q̇ : p ∈ D} is dense in P ∗ Q̇ and hence there is
(p, q̇) ∈ H ∩ D0 . Since (p, q̇) ≤ (p, 1̇Q ), (p, 1̇Q ) ∈ H and thus p ∈ G ∩ D.
We now show that F is a filter in Q̇G . Suppose that q̇G ∈ F and ṙG ∈ Q̇G is
such q̇G ≤ ṙG . Then there is p ∈ G such that p q̇ ≤ ṙ. Since p ∈ G, (p, 1̇Q ) ∈ H.
Since q̇G ∈ F , there is p0 ∈ P such that (p0 , q̇) ∈ H. Since H is a filter, (p, 1̇Q ) and
(p0 , q̇) have a common extension (p00 , ṡ) in H. Now p00 q̇ ≤ ṙ and p00 ṡ ≤ q̇. It
follows that p00 ṡ ≤ ṙ. Since p00 ≤ p0 we have (p00 , ṡ) ≤ (p00 , ṙ) and thus ṙG ∈ F .
A similar argument shows that any two elements of F have a common extension in
F.
Now let D ∈ M [G] be a dense subset of Q̇G . There is a P-name Ḋ ∈ M for D.
By the Maximality Principle we may assume that 1P forces Ḋ to be a dense subset
of Q̇. Consider the set
D0 = {(p, q̇) ∈ P ∗ Q̇ : p q̇ ∈ Ḋ}.
This set is dense in P ∗ Q̇:
Let (p, q̇) be an element of P ∗ Q̇. Since Ḋ is forced to be dense in Q̇ and by the
Maximality Principle, there is a name ṙ such that p forces ṙ to be an element of
Ḋ that is an extension of q̇. Now (p, ṙ) ≤ (p, q̇) and p ∈ D0 . It follows that D0 is
dense in P ∗ Q̇.
Since H is P ∗ Q̇-generic over M , there is (p, q̇) ∈ D0 ∩ H. Now (p, 1̇Q ) ∈ H and
thus p ∈ G. By the definition of D0 , p q̇ ∈ Ḋ. By the definition of F , q̇G ∈ F . It
follows that D ∩ F 6= ∅. Therefore F is generic over M [G]. 

Lemma 9.28. If M , P and Q̇ are as above, G is P-generic over M and F is

Q̇G -generic over M [G], then
H = G ∗ F = {(p, q̇) : p ∈ G ∧ q̇G ∈ F }
is P ∗ Q̇-generic over M .
Proof. Exercise 
Lemma 9.29. If P is c.c.c. and 1P “Q̇ is c.c.c.”, then P ∗ Q̇ is c.c.c.
Proof. Assume there is a subset {(pα , q̇α ) : α < ω1 } of P ∗ Q̇ such that (pα , q̇α ) is
incompatible with (pβ , q̇β ) if α 6= β. Consider the name
σ = {(α̌, pα ) : α < ω1 }.
Since P is c.c.c., the ω1 of the ground model remains ω1 in the generic extension
by P.
Now let G be P-generic over M and let α, β ∈ σG be distinct. Suppose (q̇α )G
and (q̇β )G are compatible. Then there is a P-name ṙ for a condition in Q̇ that is
a common extension of (q̇α )G and (q̇β )G . Some p ∈ G forces that ṙ is a common
extension of q̇α and q̇β . We may assume that p is a common extension of pα and
pβ . Now (p, ṙ) is a common extension of (pα , q̇α ) and (pβ , q̇β ), a contradiction.
It follows that the (q̇α )G are pairwise incompatible. Since Q̇G is c.c.c., it follows
that σG is countable. Hence the supremum of σG is a countable ordinal.
 MODELS OF SET THEORY 45

We now forget about the particular filter G again and choose a P-name β̇ ∈ M
for the supremum of σ. Since P is c.c.c., there is a countable set C ⊆ ω1 such that
1P forces β̇ to be an element Č.
Let γ be the supremum of C. By the choice of β̇, 1P β̇ ≤ γ̌. But by the
definition of β̇ this means that no pα with α > γ can be an element of a P-generic
filter over M , a contradiction. 
9.5. Long iterations. We will now define iterations of infinite length.
Definition 9.30. Let δ be an ordinal. ((Pα )α≤δ , (Q̇β )β<δ ) is a finite support iter-
ation length δ if the following conditions are satisfied:
(1) For every α ≤ δ, Pα is a partial order consisting of sequences of length α.
In particular, P0 is the trivial partial order {∅}.
(2) If α < β ≤ δ and p ∈ Pβ , then p  α ∈ Pα .
(3) For every α < δ, Q̇α is a Pα -name for a partial order and Pα+1 = Pα ∗ Q̇α .
Since Pα consists of sequences of length α, it is natural to consider Pα ∗ Q̇α
as consisting of sequences of length α + 1.
(4) If β ≤ δ is a limit ordinal, then Pβ consists of all sequences p of length β
such that the support
supt(p) = {α < β : p(α) 6= 1̇Qα }
of p is finite and for all α < β, p  α ∈ Pα .
(5) If ≤ denotes the order on Pβ and p, q ∈ Pβ , then p ≤ q if for all α < β,
p  α p(α)≤ ˙ α q(α) where ≤
˙ α is a name for the order on Q̇α .
Definition 9.31. Let P and Q be partial orders and let e : P → Q be an embedding.
I.e., let e be 1-1 and such that for all p0 , p1 ∈ P, p0 ≤ p1 iff e(p0 ) ≤ e(p1 ). Then e
is a complete embedding if it preserves ⊥ and for all q ∈ Q there is p ∈ P such that
whenever r ∈ P is an extension of p, then r is compatible with q.
Lemma 9.32. Let e : P → Q be an embedding of partial orders in the ground
model M . Then e is a complete embedding iff for every Q-generic filter F over M ,
G = e−1 [F ] is P-generic over M .
If e : P → Q is a complete embedding, then there is a P-name Q : P for a forcing
notion such that forcing with Q is equivalent to forcing with P ∗ (Q : P).
Lemma 9.33. Let ((Pα )α≤δ , (Q̇β )β<δ ) be a finite support iteration and let β < δ.
For every p ∈ Pβ let eβ (p) ∈ Pδ be such that eβ (p)  β = p and for all α < δ with
α ≥ β, eβ (p)(α) = 1̇Qα . Then e is a complete embedding.
Now let G be Pδ generic over the ground model M . Then for every β < δ,
G  β = e−1β [G] is Pβ -generic over M . Let α < δ. Then Pα+1 = Pα ∗ Q̇α and hence
there is a Q̇α -generic filter Gα over M [G  α] such that G  (α + 1) = (G  α) ∗ Gα .
Lemma 9.34. Let ((Pα )α≤δ , (Q̇β )β<δ ) be a finite support iteration. If for each
α < δ,
Pα “Q̇α is c.c.c.”,
then Pδ is c.c.c. In short, finite support iterations of c.c.c. forcings are c.c.c.
Theorem 9.35. Martin’s Axiom is consistent with the negation of CH.
Proof. Let M be a countable transitive model of GCH. In M we construct an
iteration
((Pα )α≤ℵ2 , (Q̇β )β<ℵ2 )
of c.c.c. forcing notions such that for all α < ℵ2 , Q̇α is forced to be of size ℵ1 and
for every Pℵ2 -generic filter G over M and every c.c.c. partial order Q of size ℵ1 in
46 STEFAN GESCHKE

M [G] there are cofinally many α < ℵ2 such that (Q̇α )Gα ∼
= Q. It can be shown
that M [G] |= MA + 2ℵ0 = ℵ2 . 
 MODELS OF SET THEORY 47

References
[1] T. Jech, Set Theory, Academic Press (1978)
[2] K. Kunen, Set Theory, An Introduction to Independence Proofs, North Holland (1980)

Department of Mathematics, Boise State University, 1910 University Drive, Boise,

ID 83725-1555
E-mail address: geschke@math.boisestate.edu