Bachelorarbeit

Suslin’s Problem

Verfasser
Joaquín Padilla Montani

angestrebter akademischer Grad

Bachelor of Science (BSc)

Wien, 2018

Studienkennzahl lt. Studienblatt: A 033 621

Fachrichtung: Mathematik
Betreuerin: Vera Fischer, Privatdoz., PhD
 CONTENTS

Contents
Introduction 1
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1 A Characterization of (R, <R ) 3

2 Suslin’s Hypothesis 7

3 Trees 10

4 Jensen’s Diamond 14

5 Gödel’s Constructible Universe 17

6 Martin’s Axiom 23

References 27
 INTRODUCTION

Introduction
Motivation1
The first volume of the Polish journal Fundamenta Mathematicae, published in 1920,
ended with a list of open problems. Problem 3 [17], posthumously attributed to the
Russian mathematician Mikhail Yakovlevich Suslin (1894-1919), read as follows:

Un ensemble ordonné (linéairement) sans sauts ni lacunes et tel que tout en-
semble de ses intervalles (contenant plus qu’un élément) n’empiétant pas les
uns sur les autres est au plus dénumerable, est-il nécessairement un continue
linéaire (ordinaire)?

Which roughly translates to: “Is a (linearly) ordered set, with no jumps and no gaps,
and such that every collection of its intervals (containing more than one element) not
overlapping each other is at most countable, necessarily an (ordinary) linear continuum?”.
In modern terms, the absence of jumps refers to the ordering being dense, and the lack
of gaps amounts to asking for the ordering to be complete. If we compare this to the
usual characterization of (R, <R ) where the real numbers are that unique unbounded
linearly ordered set which is dense in itself, complete and separable, Suslin’s Problem
asks whether the hypothesis of separability can be replaced by the topological countable
chain condition (c.c.c.): that every collection of nonempty pairwise disjoint open intervals
is at most countable.
It is not difficult to see that any separable linear ordering must satisfy the countable
chain condition. In the opposite direction, the question about the existence of a Suslin
line, that is, an unbounded linearly ordered set which is dense, complete and ccc, but
not separable, proved to be much harder to settle. The existence of a Suslin line would
constitute a negative answer to Suslin’s Problem, and the assertion “there are no Suslin
lines” came to be known as Suslin’s Hypothesis or SH, although Suslin simply posed the
question, without any discussion.
Other problems in that original list of 1920, mostly dealing with topics from the
incipient field of descriptive set theory, would eventually be solved. Suslin’s Problem,
however, remained unsolved for half a century, and so it grew in importance through the
years. Around 1940, the first advance towards settling Suslin’s Problem was indepen-
dently produced by Djuro Kurepa [11], Edwin Miller [14] and Wacław Sierpiński [15].
Their contribution reduced SH, an eminently topological problem, to a problem purely
in the realm of combinatorial set theory. They found that the existence of a Suslin line is
equivalent to the existence of a combinatorial object called a Suslin tree, a tree of height
ω1 without uncountable chains or antichains.
Much later, in a paper published in 1968 [18], Stanley Tennenbaum proved that ¬SH
is consistent with ZFC. That is, he showed that there are models of set theory where
a Suslin line exists, hence rendering impossible the task of proving Suslin’s Hypothesis.
His proof made use of Paul Cohen’s celebrated forcing technique, and was done in the
summer of 1963, only months after Cohen had developed his method for extending
models of set theory. As such, after Cohen’s initial applications of forcing (most notably
on the non-provability of the Continuum Hypothesis) SH was one of the first problems
which were solved using this groundbreaking technique.
1
This historical outline is mostly based on [10] and [12].

1
 INTRODUCTION

Thomas Jech independently proved [8], a few years later, the same results on the
non-provability of SH. His use of forcing was more complicated than Tennenbaum’s and
also made, along with SH, the Continuum Hypothesis (CH) true in that model, whereas
Tennenbaum’s technique could both produce models of ¬SH + CH and of ¬SH + ¬CH.
It was, however, Jech’s approach which would generalize to prove results on higher
cardinality versions of Suslin trees.
Close in time to Tennenbaum and Jech’s result on the non-provability of SH, another
major discovery was made by Ronald Jensen [9]. He showed that in Kurt Gödel’s famous
Constructible Universe L, a model of ZFC where the Generalized Continuum Hypothesis
holds true, a Suslin tree (and hence also Suslin line) can be found. So Gödel’s L is a
witness for the non-provability of SH. Jensen then extracted from his proof a combina-
torial principle known as Diamond (♦), and showed that ♦ implies that there is a Suslin
tree. This initiated his study of combinatorial principles in L. Combinatorial principles
are focal axioms which hold in one or another model of set theory, out of which one can
draw, by purely classical methods, strong consequences unavailable in ZFC alone.
In a paper published in 1971 [16], Robert Solovay and Stanley Tennenbaum finally
settled Suslin’s Problem. They showed that SH is consistent with ZFC, meaning that
there is a model of set theory were no Suslin line exists. Together with Tennenbaum’s
earlier result on the consistency of ¬SH, this proved that SH is independent of ZFC, i.e.,
using the usual axioms for set theory, SH can neither be proven nor disproven. Solovay
and Tennenbaum’s proof was actually developed in 1965, and again used the forcing
technique, but this time in an iterative fashion. Nowadays a standard tool in set theory,
iterated forcing is considered to have originated in this proof.
Donald Martin and Robert Solovay then showed in [13] that, using the same tech-
niques as in [16], one can establish the consistency of a combinatorial principle nowadays
known as Martin’s Axiom, which has many consequences; among those is SH, if one takes
the additional assumption that 2ℵ0 > ℵ1 , i.e. that CH does not hold.2
Suslin’s Problem, for decades a famous open question, permeated the development of
key areas in set theory, most notably in forcing and constructibility.

Outline
The aim of this thesis is to show, assuming the consistency of Jensen’s Diamond Principle
and of Martin’s Axiom, the independence of Suslin’s Hypothesis. Therefore the key
results we want to prove are ♦ → ¬SH (Theorem 4.12) and MA(ℵ1 ) → SH (Theorem
6.11). We also present Gödel’s constructible universe L and show that ♦ is true in L,
therefore fully proving the non-provability of SH.
Section 1 discusses, via the Dedekind Completion, the characterization of (R, <R )
mentioned in the introduction. Section 2 formally states Suslin’s Problem, related
definitions, and shows that the product of a Suslin line with itself is not ccc. Section 3
reduces SH to a problem in infinitary combinatorics: the proof that a Suslin line exists
if and only if a Suslin tree exists is presented. In Section 4 the Diamond Principle is
defined, and the section is dedicated to proving that ♦ implies ¬SH. Section 5 deals
with Gödel’s Constructible Universe, and the main result is the proof that ♦ holds in L.
Section 6 discusses Martin’s Axiom, and it is shown in two ways that under MA(ℵ1 ),
SH is true.
2
The consistency of CH+SH was established by Ronald Jensen shortly thereafter (c.f. [1]).

2
 1 A CHARACTERIZATION OF (R, <R )

Prerequisites
The main prerequisite is familiarity with the basics of ordinals and cardinals, transfinite
induction and transfinite recursion. Excellent references for this are [3], [4], [5] and [6].
For Section 5 on the Constructible Universe, model theory, especially in the context
of set theory, is required; in particular the Downward Löwenheim-Skolem Theorem,
Mostowski’s Collapsing Lemma and absoluteness. For this, see [5] or [6]. Also, basic
ideas from topology are needed in a few proofs.

1 A Characterization of (R, <R)

In the next section we will formally define Suslin’s Problem. To do this, one starts with
a characterization of (R, <R ) as the unique (up to isomorphism) linearly ordered set
which is unbounded, dense in itself, complete and separable. The aim of this section is
to formally define the notions involved in this characterization and prove it via Dedekind
cuts, starting from any ordered set resembling the rationals.
Definition 1.1 (Partial/Linear Ordering). Let P be a nonempty set. A partial ordering
of P is a binary relation < on P which is irreflexive, i.e. ∀p ∈ P (p ≮ p), and transitive,
meaning ∀pqr ∈ P (p < q ∧ q < r → p < r). The pair (P, <) is called a partially ordered
set. If, moreover, < satisfies trichotomy, that is ∀pq ∈ P (p = q ∨ p < q ∨ q < p), we say
that < is a linear ordering of P and call (P, <) a linearly ordered set. The notation p ≤ q
is an abbreviation for p < q ∨ p = q.
Definition 1.2 (Dense Linear Ordering). Let (P, <) be a linearly ordered set with at
least two elements and let D ⊆ P . Then we say that
(i) D is a dense subset of P or, alternatively, D is dense in P iff
∀pq ∈ P (p < q → ∃d ∈ D(p < d < q));
(ii) (P, <) is dense iff P is dense in itself, that is ∀pq ∈ P (p < q → ∃x ∈ P (p < x < q)).
Note that the existence of some D ⊆ P dense in P already ensures that (P, <) is dense.
Remark 1.3. If (P, <) is a dense linearly ordered set, then P is always infinite. More-
over, D ⊆ P being dense in P in the sense defined above is equivalent to asking for D
to be dense in P in the topological sense, given the order topology induced by <.
Definition 1.4. Let (P, <) be a partially ordered set, and let a ∈ X ⊆ P . Then
(i) a is a maximal element of X iff ∀x ∈ X(a ≮ x);
(ii) a is the greatest element of X iff ∀x ∈ X(x ≤ a).
Similarly,
(iii) a is a minimal element of X iff ∀x ∈ X(x ≮ a);
(iv) a is the least element of X iff ∀x ∈ X(a ≤ x).
Definition 1.5 (Unbounded Linear Ordering). Let (P, <) be a linearly ordered set. We
say that (P, <) is unbounded or, alternatively, that (P, <) has no endpoints iff P has
neither a least nor a greatest element. In symbols: ∀p ∈ P ∃a, b ∈ P (a < p < b).
It is well known that the rational numbers (Q, <Q ) are a countable linearly ordered
set, that there is no greatest or least rational, and that between any two rationals we
can always find another one, i.e. (Q, <Q ) is dense. Moreover, it turns out that any other
countable dense unbounded linearly ordered set looks just like (Q, <Q ). To prove this,
we first need a notion of isomorphism between ordered sets.

3
 1 A CHARACTERIZATION OF (R, <R )

Definition 1.6. Let (P, <) and (Q, ≺) be two partially ordered sets. An order-embedding
is a function f : P → Q such that ∀p1 p2 ∈ P (p1 < p2 ↔ f (p1 ) ≺ f (p2 )). If f is also
bijective, then we call it an order-isomorphism. A function g : X → Q for X ( P is
called a partial isomorphism from P to Q, provided that g behaves as an order-embedding
on its domain of definition.
Lemma 1.7. Let (P, <) and (Q, ≺) be two dense linearly ordered sets without endpoints.
If f is a partial isomorphism from P to Q such that dom(f ) is finite, then for any choice
of p ∈ P and q ∈ Q there is a partial isomorphism fp,q ⊇ f such that p ∈ dom(fp,q ) and
q ∈ ran(fp,q ).
Proof. Let p ∈ P and q ∈ Q. Let f = {(p1 , q1 ), . . . , (pk , qk )}, where p1 < · · · < pk , and
thus also q1 ≺ · · · ≺ qk .
We first extend f by p. If p ∈ dom(f ), there’s nothing to do, so assume that p 6∈
dom(f ). If p < p1 , then we find q̃ ∈ Q such that q̃ ≺ q1 , which is possible since the
orderings we are considering have no endpoints. Then define fp = f ∪ {(p, q̃)}; clearly
this is still a partial isomorphism. Similarly if pk < p, we find q̃ ∈ Q such that qk ≺ q̃
and define fp in the same way. If otherwise there is some i such that pi < p < pi+1 , since
our orderings are dense we can find some q̃ ∈ Q such that qi ≺ q̃ ≺ qi+1 and define fp
again as before.
To now extend fp by q, note that fp−1 is a finite partial isomorphism in the oppo-
site direction, so we can add q to the domain of fp−1 by arguing in the same way as
above. Hence we get a partial isomorphism fp,q extending f such that p ∈ dom(fp,q ) and
q ∈ ran(fp,q ).
Theorem 1.8. Any two countable dense linearly ordered sets without endpoints are
isomorphic.
Proof. Let (P, <) and (Q, ≺) be two unbounded dense linearly ordered sets such that P =
{pn : n ∈ ω} and Q = {qn : n ∈ ω}. We construct a sequence of partial isomorphisms
by recursion. Let f0 = ∅ and given fn , let fn+1 = (fn )pn ,qn using the previous lemma.
Notice that fn+1 ⊇ fn for any n ∈ ω, so the partial isomorphisms are compatible with
each other, in the sense that given any p ∈ P , it is not possible to find two functions
in our sequence which map p to different values in Q. For this reason we can now let
f = n∈ω fn . It is immediate to check that f : P → Q is an isomorphism between (P, <)
S

and (Q, ≺).

Remark 1.9. In model theory books, the statement of Theorem 1.8 would read: “the
theory DLO (the theory of dense linear orders without endpoints) is ℵ0 -categorical”.
The method of proof given above is commonly referred to as the back-and-forth method.
Cantor was the first to prove this theorem, although using a different technique.
Definition 1.10. Let (P, <) be a partially ordered set, p ∈ P , and let ∅ 6= X ⊆ P .
Then
(i) p is an upper bound of X iff ∀x ∈ X(x ≤ p);
(ii) p is the supremum of X iff p is the least upper bound of X, i.e. p is the least element
of the set {p̃ ∈ P : p̃ is an upper bound of X}.
Similarly,
(iii) p is a lower bound of X iff ∀x ∈ X(p ≤ x);
(iv) p is the infimum of X iff p is the greatest lower bound of X, i.e. p is the greatest
element of the set {p̃ ∈ P : p̃ is a lower bound of X}.

4
 1 A CHARACTERIZATION OF (R, <R )

Definition 1.11 (Complete Linear Ordering). A linearly ordered set (P, <) is complete
iff every nonempty subset of P which has an upper bound also has a supremum.

Remark 1.12. For a linearly ordered set (P, <), being dense and complete as defined
above is equivalent to asking for (P, <) to be connected in its order topology.

Definition 1.13 (Dedekind Cut). Let (P, <) be a linearly ordered set. A Dedekind cut is
a pair (A, B) of disjoint nonempty subsets of P satisfying the three following conditions:
(i) A ∪ B = P ;
(ii) A does not have a greatest element;
(iii) ∀a ∈ A ∀b ∈ B(a < b).

Example 1.14. Given some p ∈ P , the pair ({p̃ ∈ P | p̃ < p}, {p̃ ∈ P | p̃ ≥ p}) is a
Dedekind cut; such a cut will be denoted by [p].

Theorem 1.15 (Dedekind Completion). Let (P, <) be a dense linearly ordered set
without endpoints. Then there is a complete linearly ordered set without endpoints
(C, ≺) and an order-embedding ϕ : P → C such that ϕ[P ] = {c ∈ C | ∃p ∈ P (ϕ(p) = c)}
is dense in C. Moreover, (C, ≺) is unique, in the sense that given further (C̃, @) and
ϕ̃ : P → C̃ as before, there is an order-isomorphism f : C → C̃ satisfying f ◦ ϕ = ϕ̃. We
call (C, ≺) the Dedekind completion of (P, <).

Proof. Existence
We define C to be the set of all Dedekind cuts of P . Given (A1 , B1 ) ∈ C and
(A2 , B2 ) ∈ C, the ordering ≺ is characterized by (A1 , B1 ) ≺ (A2 , B2 ) iff A1 ( A2 .
First we will show that (C, ≺) is a linearly ordered set. The irreflexivity of ≺ is
immediate since the relation ≺ is defined in terms of strict inclusion, and transitivity
follows from the transitivity of (strict) set inclusion. It remains to show that ≺ satisfies
trichotomy. Let (A1 , B1 ) ∈ C and (A2 , B2 ) ∈ C with (A1 , B1 ) 6= (A2 , B2 ). Without
loss of generality we can assume that there is a1 ∈ A1 such that a1 ∈ / A2 , and hence
a1 ∈ B2 . By Definition 1.13 it is clear that A2 ⊆ {p ∈ P | p < a1 } ( A1 , therefore
(A2 , B2 ) ≺ (A1 , B1 ), i.e. trichotomy holds.
Next consider the function ϕ : P → C defined by p 7→ [p]. Note that for any a, b ∈ P ,
a < b if and only if {p ∈ P | p < a} ( {p ∈ P | p < b}, which in turn holds iff [a] ≺ [b],
which is equivalent to ϕ(a) ≺ ϕ(b), so ϕ is indeed an order-embedding.
Having defined the linearly ordered set (C, ≺) and the order-embedding ϕ : P → C,
it remains to show that
(a) ϕ[P ] is dense in C, hence (C, ≺) is dense;
(b) (C, ≺) is unbounded;
(c) (C, ≺) is complete.
For (a), let (A1 , B1 ) ∈ C and (A2 , B2 ) ∈ C with (A1 , B1 ) ≺ (A2 , B2 ), i.e. A1 ( A2 .
Then A2 \ A1 6= ∅, so there is some b ∈ A2 \ A1 . We can assume that b is not the least
element of B1 , if not we could take b̃ > b with b̃ ∈ A2 by the fact that A2 has no greatest
element. It would still be the case that b̃ ∈ / A1 , otherwise b̃ > b would be contradicting
Definition 1.13. This guarantees that (A1 , B1 ) ≺ [b]. Moreover, since b ∈ A2 , we have
that p ∈ A2 for any p ∈ P satisfying p < b, using Definition 1.13. This shows that
[b] ≺ (A2 , B2 ), and (a) is now proven.
For (b), fix (A, B) ∈ C and take any b ∈ B which is not the least element of B. Then
we have that (A, B) ≺ [b], and of course [b] ∈ C. Now given any a ∈ A, it holds that
[a] ∈ C and [a] ≺ (A, B). This shows that C has no endpoints.

5
 1 A CHARACTERIZATION OF (R, <R )

It remains to show (c), the completeness of (C, ≺). Let D be a nonempty subset
of C and let (A0 , B0 ) ∈ C be an upper bound for D, i.e. for any (A, B) ∈ D we
have that A ⊆ A0 . We need to come up with a supremum for D. To this end, let
AD = {A | (A, B) ∈ D} and BD = X \ AD . Notice that BD = {B | (A, B) ∈ D} using
S T

De Morgan’s laws.
We first verify that (AD , BD ) is a Dedekind cut. By construction AD and BD are
disjoint subsets of P . Moreover, since D 6= ∅, we know that AD 6= ∅, and BD 6= ∅
is clear from B0 ⊆ BD . It remains to check the three conditions from Definition 1.13.
First, AD ∪ BD = P is clear by construction, and the same is true for condition (iii).
Furthermore, AD cannot have a greatest element, since none of the A’s have one.
Now given that A ⊆ AD for any (A, B) ∈ D, we have that (AD , BD ) is an upper
bound of D. It remains to show that (AD , BD ) is the supremum, i.e. that (AD , BD ) is
 than any other upper bound of D. For any upper bound (Ã, B̃) ∈ C of D it holds
that A ⊆ Ã for any (A, B) ∈ D, so AD = {A | (A, B) ∈ D} ⊆ Ã; it follows that
S

(AD , BD )  (Ã, B̃), which is what we needed to show.

Uniqueness
Let (C, ≺) and ϕ : P → C be the Dedekind completion of (P, <) and its corresponding
order-embedding as defined above. Let (C̃, @) be another complete and unbounded
linearly ordered set, with ϕ̃ : P → C̃ an order-embedding and ϕ̃[P ] dense in C̃. We want
to show that there is an order-isomorphism f : C → C̃ satisfying f ◦ ϕ = ϕ̃.
For a Dedekind cut c = (A, B) ∈ C we define f (c) as supC̃ (ϕ̃[A]), i.e. the supremum
of ϕ̃[A] as a subset of C̃. It is clear that f is a mapping from C to C̃, since for (A, B) ∈ C
we have that A is bounded above in P by any element of B, and since ϕ̃ is an order-
embedding we have that ϕ̃[A] also has an upper bound in C̃, so its supremum in C̃ exists
(and is unique, as suprema always are).
It remains to check that f is an order-isomorphism and that f ◦ ϕ = ϕ̃. To see that
f is onto, let c̃ ∈ C̃ be arbitrary. Notice that, by the density of ϕ̃[P ] in C̃ and the fact
that ϕ̃ is an order-embedding, the pair

(Ac̃ , Bc̃ ) := ({p ∈ P | ϕ̃(p) @ c̃} , {p ∈ P | ϕ̃(p) w c̃})

is a Dedekind cut, and therefore an element of C. It is easy to see that then f ((Ac̃ , Bc̃ )) =
supC̃ (ϕ̃[Ac̃ ]) = c̃, which shows that f is onto.
Now let (A1 , B1 ) ∈ C and (A2 , B2 ) ∈ C with (A1 , B1 ) ≺ (A2 , B2 ). By definition this
is equivalent to A1 ( A2 , which is equivalent to ϕ̃[A1 ] ( ϕ̃[A2 ] since ϕ̃ is injective. We
are done if we can prove that this is equivalent to supC̃ ϕ̃[A1 ] @ supC̃ ϕ̃[A2 ], but this is
true by the fact that A1 and A2 are both initial segments of P that don’t have a greatest
element. We conclude that f is an order-isomorphism.
Finally we have that

(f ◦ ϕ)(p) = f ([p]) = supC̃ ϕ̃[{q ∈ P | q < p}] = ϕ̃(p)

which proves that f ◦ ϕ = ϕ̃.

Definition 1.16. Let (Q, <Q ) be the rational numbers with their usual ordering. The
completion of (Q, <Q ) is denoted by (R, <R ). The elements of R are called real numbers.

Definition 1.17 (Separable Linear Ordering). A linearly ordered set (P, <) is separable
iff there is a countable D ⊆ P that is dense in P .

6
 2 SUSLIN’S HYPOTHESIS

We can now state and prove the characterization of (R, <R ) that we mentioned at the
beginning of this section.

Theorem 1.18. Let (L, <L ) be a dense linearly ordered set without endpoints which is
complete and separable. Then (L, <L ) is order-isomorphic to (R, <R ).

Proof. Let (L, <L ) be a complete linearly ordered set without endpoints and let D ⊆ L
be countable and dense in L. Then (D, <L  D) is a countable dense linearly ordered
set without endpoints, so it is isomorphic to (Q, <Q ) by Theorem 1.8. By the unique-
ness of the Dedekind completion (Theorem 1.15), (L, <L ) is isomorphic to (R, <R ), the
completion of (Q, <Q ).

2 Suslin’s Hypothesis
Using the characterization of (R, <R ) from the previous section, we will now discuss
Suslin’s Problem. Further, we will define the related concepts of a Suslin line and
of Suslin’s Hypothesis, and prove that a Suslin line can always be made to be dense,
unbounded and complete. We will also show that the negation of Suslin’s Hypothesis,
i.e. the existence of a Suslin line, implies that the product of ccc topological spaces it
not always ccc.

Definition 2.1 (Countable Chain Condition). A topological space X has the countable
chain condition (or the ccc) iff every family of disjoint non-empty open subsets of X is
at most countable.

Remark 2.2. In the context of a linearly ordered set (P, <), we will call (P, <) ccc if
it satisfies the countable chain condition given its order topology. This is equivalent to
the statement: every family of pairwise disjoint non-empty open intervals of the form
(a, b) = {p ∈ P | a < p < b} is at most countable.

Let (L, <L ) be a dense linearly ordered set without endpoints which is complete and
satisfies the countable chain condition. Is (L, <L ) order-isomorphic to (R, <R )? This is
known as Suslin’s Problem, i.e. asking if in Theorem 1.18 the condition of separability
can be weakened to the countable chain condition. The next lemma shows that this
condition is indeed weaker.

Lemma 2.3. Every separable linearly ordered set satisfies the ccc.

Proof. Let (P, <) be a separable linearly ordered set, with Q ⊆ P countable and dense
in P . Let A be a family of pairwise disjoint open intervals in P . Consider the set
X := Q ∩ A. Since Q is dense in P , every open interval a ∈ A contains at least one
S

x ∈ X and since the open intervals in A are pairwise disjoint, each x ∈ X is an element
of a unique open interval a ∈ A. Hence the function f : X → A, which assigns to an
element of X the unique open interval a ∈ A to which it belongs, is well-defined and
onto. Since X is at most countable, we conclude that A is at most countable, witnessed
by f . This shows that (P, <) satisfies the ccc.

Definition 2.4 (Suslin Line / Suslin’s Hypothesis). A linearly ordered set that satis-
fies the countable chain condition and is not separable is called a Suslin line. Suslin’s
Hypothesis (SH) is the statement, “there are no Suslin lines”.

7
 2 SUSLIN’S HYPOTHESIS

Thus SH is equivalent to a “yes” answer to Suslin’s Problem, and the existence of a

Suslin line, i.e. ¬SH, is equivalent to a “no” answer.
The given definition of a Suslin line omits some of the properties that are mentioned
in Suslin’s original question. The next two lemmas show that given a Suslin line, we can
always find another one with these additional requirements.
Lemma 2.5. Assume that there exists a Suslin line (L, <). Then there is a Suslin
line (L∗ , ≺) that is dense, unbounded and where no non-empty open interval in L∗ is
separable.
Proof. Let (L, <) be a Suslin line. We define a relation ∼ on L by: x ∼ y if and only if
x = y or the open interval (min(x, y), max(x, y)) is separable, meaning that it contains a
countable dense subset. It is straightforward to check that ∼ is an equivalence relation.
Let L∗ = L/∼ be the set of all equivalence classes. Note that every I ∈ L∗ is convex,
in the sense that if x, y ∈ I and x < y, then I ⊇ (x, y) = {` ∈ L | x < ` < y}. Another
easy observation is that under ∼, two points x, y ∈ L such that x < y and (x, y) = ∅ get
identified to the same point in L∗ .
Next we define a linear ordering ≺ on L∗ as follows: I ≺ J iff I 6= J and there is
x ∈ I and y ∈ J with x < y. It is easy to check that (L∗ , ≺) is indeed a linear ordering,
and I ≺ J implies that x < y for any x ∈ I and y ∈ J.
We claim that every I ∈ L∗ is separable as a subset of L. Indeed, let A be a ⊆-maximal
collection of disjoint non-empty open intervals of the form (x, y) ⊆ L with x, y ∈ I. Since
(L, <) is a Suslin line, it has the ccc, hence A is at most countable and we can write
A = {(xn , yn ) | n ∈ ω}. For each n ∈ ω we have that xn ∈ I and yn ∈ I, i.e. xn ∼ yn
which by definition means that (xn , yn ) is separable, so for each n ∈ ω let Dn ⊆ (xn , yn )
be a countable subset dense in (xn , yn ). Then D := n∈ω Dn is dense in n∈ω (xn , yn ),
S S

and D is countable. Let z ∈ I be arbitrary. If z is not the first or last element of I,

then there is a non-empty open interval (x, y) ⊆ I with x, y ∈ I and z ∈ (x, y). By the
maximality of A, the interval (x, y) intersects (xn , yn ) for some n ∈ ω. This implies that
S S
z is an element of the closure of n∈ω (xn , yn ), and since D is dense in n∈ω (xn , yn ), we
have that z is an element of the closure of D. Therefore D together with the first and
last element of I (if I has a first or last element) is a countable dense subset of I.
We list the facts we will prove next:
(a) (L∗ , ≺) is dense;
(b) no non-empty open interval in L∗ is separable;
(c) (L∗ , ≺) satisfies the ccc.
We first show (a). Towards a contradiction, assume we have I, J ∈ L∗ with I ≺ J but
{K ∈ L∗ | I ≺ K ≺ J} = (I, J) = ∅. Fix x ∈ I and y ∈ J. The assumption (I, J) = ∅
implies that (x, y) = ((x, y) ∩ I) ∪ ((x, y) ∩ J) ⊆ I ∪ J, and I ∪ J is separable by the
previous paragraph, so (x, y) is also separable and we get x ∼ y, which is a contradiction
to I ≺ J.
Moving on, we will now prove (b). Say it does not hold, i.e. there are I, J ∈ L∗ with
I ≺ J such that (I, J) is separable. Let {Kn ∈ (I, J) | 2 ≤ n < ω} be dense in (I, J),
and set K0 = I and K1 = J. Using the previous claim (the fact that every element of
L∗ is separable as a subset of L) we define, for each n ∈ ω, Dn to be a countable dense
subset of Kn . Then we have that D := n∈ω Dn is dense in n∈ω Kn , but also n∈ω Kn
S S S

is dense in {K ∈ L∗ | I  K  J}, so D is dense in {K ∈ L∗ | I  K  J} and D is

S S

countable. Using the definition of ∼, this is a contradiction to the fact that I ≺ J, i.e.
to I 6= J.

8
 2 SUSLIN’S HYPOTHESIS

To prove (c), assume that we have an uncountable collection A = {(Iα , Jα ) | α ∈ ω1 }

of pairwise disjoint non-empty open intervals in L∗ . For each α ∈ ω1 pick xα ∈ Iα
and yα ∈ Jα . Then B = {(xα , yα ) | α ∈ ω1 } would be a collection of pairwise disjoint
non-empty open intervals in L, which contradicts the fact that (L, <) is ccc.
To recapitulate, we have shown that (L∗ , ≺) is a dense linearly ordered set. What’s
more, property (b) implies that (L∗ , ≺) is not separable. This, together with (c), shows
that (L∗ , ≺) is indeed a Suslin line. To finish the proof of our Lemma, it remains to argue
that we can get (L∗ , ≺) to also be unbounded. This is not hard, since if L∗ happens to
have a least element or a greatest element (or both), we can just remove it from L∗ , and
all the properties we have shown so far would remain true.
Lemma 2.6. Let (P, <) be a dense linearly ordered set without endpoints. Let (C, ≺)
be its Dedekind completion and ϕ : P → C the corresponding order-embedding, as
constructed in Theorem 1.15. If (P, <) is a Suslin line, then (C, ≺) is also a Suslin line.
Proof. We first show that if (P, <) is not separable, then (C, ≺) is also not separable.
Assume it is, so there is D ⊆ C countable and dense in C. Then (C, ≺) is a dense linearly
ordered set without endpoints that is complete and separable, so by Theorem 1.18 it is
isomorphic to (R, <R ). Then, since P is an infinite set and ϕ is an order-embedding,
we know that ϕ[P ] is an infinite subset of C, and using the fact that infinite subsets of
R always have a countable dense subset we have that also ϕ[P ] has a countable dense
subset, which would give us a countable dense subset for P , contradicting the fact that
P is not separable.
Now assume that (P, <) satisfies the ccc, but (C, ≺) does not. So there is a family A
of pairwise disjoint open intervals in C which is uncountable. Since ϕ[P ] is dense in C,
this would immediately give us an uncountable family of pairwise disjoint open intervals
in P , contradicting the fact that (P, <) satisfies the ccc.
To finish this section we show that ¬SH implies that the product of ccc topological
spaces is not always ccc.
Lemma 2.7. If (L, <) is a Suslin line, then L × L with the product topology is not ccc.
Proof. Given x, y ∈ L with x < y, let (x, y) denote the open interval {` ∈ L | x < ` < y}
as usual. Notice that (x, y) could be empty even though x < y, since (L, <) is not
necessarily dense.
By recursion on α < ω1 , we will construct sequences haα | α ∈ ω1 i, hbα | α ∈ ω1 i and
hcα | α ∈ ω1 i of elements of L so that, for every α ∈ ω1 , the following properties hold:
(i) aα < bα < cα ;
(ii) (aα , bα ) 6= ∅ and (bα , cα ) 6= ∅;
(iii) β < α → bβ ∈ / (aα , cα ).
First, assume that we have already constructed such three sequences as above. Then,
for every α ∈ ω1 , the open set Uα := (aα , bα ) × (bα , cα ) ⊆ L × L is non-empty by
(ii). Moreover, given β < α < ω1 , we have that Uβ ∩ Uα = ∅, since, by (iii), we have
that either bβ ≤ aα , in which case (aβ , bβ ) ∩ (aα , bα ) = ∅, or bβ ≥ cα , in which case
(bβ , cβ ) ∩ (bα , cα ) = ∅. Therefore {Uα | α ∈ ω1 } is an uncountable collection of pairwise
disjoint non-empty open subsets of L × L, i.e. L × L does not satisfy the ccc.
To construct the sequences haα | α ∈ ω1 i, hbα | α ∈ ω1 i and hcα | α ∈ ω1 i, let W be
the set of all isolated points of L, i.e. W := {` ∈ L | {`} is open}. Since (L, <) sat-
isfies the ccc, we have that W is at most countable. Now assume we have picked

9
 3 TREES

aβ , bβ , cβ for all β ∈ α and choose aα and cα such that aα < cα and (aα , cα ) 6= ∅
and (aα , cα ) ∩ (W ∪ {bβ | β < α}) = ∅. This can be done since the set W ∪ {bβ | β < α}
is countable and (L, <) is not separable. Then, since (aα , cα ) is non-empty and contains
no isolated points, it must be infinite and hence we can choose bα ∈ (aα , cα ) such that
(aα , bα ) 6= ∅ and (bα , cα ) 6= ∅.

3 Trees
Trees are combinatorial objects that abound in set theory. The purpose of this section
is to show a connection between Suslin’s Hypothesis and a certain type of trees called
Suslin trees.
Definition 3.1 (Tree). A tree is a partially ordered set (T, <) such that, given any x ∈ T ,
the set x↓:= {y ∈ T | y < x} of all predecessors of x is well-ordered by <. Moreover, we
say that
• the height ht(x) of an element x ∈ T is the order type of x↓;
• if ht(x) is a successor ordinal, then x is called a successor node; otherwise x is called
a limit node;
• the set Lα (T ) = {x ∈ T | ht(x) = α} is the αth level of T ;
• the least ordinal α such that Lα (T ) = ∅ is called the height ht(T ) of the tree T ;
• T is rooted iff |L0 (T )| = 1, in which case the root of T is the element 1T ∈ L0 (T );
• we write T (α) for {x ∈ T | ht(x) ∈ α}, which is the same as the set β∈α Lβ (T );
S

• a branch in T is a maximal chain (i.e., a maximal linearly ordered subset) of T ;

• the length `(b) of a branch b ⊆ T is the order type of b;
• an antichain in T is a subset A ⊆ T such that any two distinct elements x, y ∈ A
are incomparable, i.e., neither x < y nor y < x;
• y ∈ T is an immediate successor of x ∈ T iff x < y and ht(y) = ht(x) + 1;
Example 3.2. Every well-ordered set (W, <) is a tree. For this reason trees can be seen
as generalizations of well-ordered sets, or generalizations of ordinals. The height of W is
the order type of W , and the only branch is W itself.
Example 3.3. Let α be an ordinal number and X a nonempty set. We denote by
X <α the set of transfinite sequences of length less than α, i.e. X <α = β<α X β =
S
<α
β<α {f : β → X}. We define an ordering < on X by f < g :⇐⇒ f ( g, so f < g iff
S
<α
g extends f . It is easy to verify that (X , <) is a tree of height α.
Given a tree (T, <), any subset X ⊆ T is a tree itself under the induced order, but the
levels of the T and X need not coincide. For this reason we define subtrees as follows:
Definition 3.4 (Subtree). Let (T, <) be a tree. A subset S ⊆ T is a subtree iff s↓⊆ S
for any s ∈ S, i.e. iff for any s ∈ S and x ∈ T , x < s implies x ∈ S.
Example 3.5. Let (T, <) be a tree and α ∈ ht(T ) + 1 an ordinal. Then the set T (α) =
(α)
β∈α Lβ (T ) is a subtree of T and ht(T
S
) = α.
Having stated the basic definitions regarding trees, we can now define what Suslin
trees are and proceed to show the relationship between Suslin lines and Suslin trees.
Definition 3.6 (Suslin Tree). A tree (T, <) is called a Suslin tree if it has the following
properties:

10
 3 TREES

(i) ht(T ) = ω1 ;
(ii) every branch in T is at most countable;
(iii) every antichain in T is at most countable.

Definition 3.7 (Normal α-Tree). Let α ≤ ω1 be an ordinal number. A tree (T, <) is
called a normal α-tree if it has height α and satisfies the following properties:
(iv) T is rooted;
(v) if β ∈ ht(T ) is a limit ordinal and x, y ∈ Lβ (T ) and if x↓ = y↓, then x = y;
(vi) for every x ∈ T there is some y > x at each higher level less than ht(T );
(vii) the set of immediate successors of x, succT (x), is countably infinite for every x ∈ T .
(viii) each level of T is at most countable.

Lemma 3.8. If there exists a Suslin tree, then there exists a normal Suslin tree.

Proof. Let (T, <) be a Suslin tree. Notice that, by property (iii), every level of T is count-
able, so property (viii) doesn’t need to be discussed in the context of a Suslin tree. For
x ∈ T , let Tx := {y ∈ T | y ≥ x}. Now consider the tree T1 = {x ∈ T | Tx is uncountable},
i.e. we are using the induced ordering on the subset T1 . If x ∈ T1 and α is such that
ht(x) < α < ht(T1 ) = ω1 , then there must be some y ∈ Lα (T1 ) with y > x: otherwise
Tx = y∈Lα (T ) ({t ∈ T | x ≤ t ≤ y} ∪ Ty ) would be countable. Hence T1 satisfies (vi) and
S

is clearly still a Suslin tree.

Now we define T2 by inserting nodes to T1 as follows: for every limit node y ∈ T1
we add an extra node ay such that x < ay ⇐⇒ x < y and ay < x ⇐⇒ y ≤ x.
We can think of ay as a node between y ↓ and y. So we have that T2 := T1 ∪
{ay | y ∈ T1 and y is a limit node}, and since every level of T1 is countable and there
are only ω1 countable limit ordinals, T2 is still a Suslin tree. Moreover, T2 still satisfies
(vi) and now also (v).
We call a node a branching point if it has at least two immediate successors. Since
T2 has no uncountable chain (by the fact that it has no uncountable branch), and since
T2 satisfies (vi), there must be uncountably many branching points in T2 . Define T3 to
be the set of all branching points of T2 . By the previous observation, ht(T3 ) = ω1 , so T3
is still a Suslin tree and it is not hard to see that it still satisfies (v) and (vi).
Let T4 = {x ∈ T3 | x is a limit node}. Since each node of T3 has at least two imme-
diate successors, keeping only limit nodes ensures that each node has infinitely many
immediate successors. By (iii), sets of immediate successors are in particular countably
infinite. So T4 satisfies (vii) and is still a Suslin tree satisfying (v) and (vi).
Finally, let T5 be T4 but with the initial level forced to have only one element. So
we just pick one of the elements in L0 (T4 ) and set L0 (T5 ) to be the set consisting of
solely that node. The rest of the levels of T5 are the excepted restrictions of the levels of
T4 . None of the relevant properties of T4 are broken, and moreover now T5 also satisfies
(iv).
The next theorem is the main purpose of this section. It reduces Suslin’s Problem to
a purely combinatorial question regarding trees.

Theorem 3.9. There exists a Suslin line if and only if there exists a Suslin tree.

Proof. (⇒) Let (S, <) be a Suslin line. Assume our line has the properties given by
Lemma 2.5 and Lemma 2.6. We will construct a Suslin tree T from certain subsets of

11
 3 TREES

S. Namely, given a, b ∈ S with a < b, we call [a, b] := {s ∈ S | a ≤ s ≤ b} ⊆ S, a non-

degenerate closed interval in S. The intervals will be ordered by reverse inclusion, i.e. if
I, J ⊆ S are two such intervals, then I ≺ J iff J ( I. This is clearly a partial ordering.
The construction of our tree T will be done by recursion on α < ω1 . Let I0 =
[a0 , b0 ] be an arbitrary non-degenerate closed interval. Assume that for some α ∈ ω1 we
have already constructed non-degenerate closed intervals Iβ = [aβ , bβ ] for every β ∈ α.
Consider the set Cα := {aβ | β ∈ α} ∪ {bβ | β ∈ α} consisting of the endpoints of the
intervals Iβ for β ∈ α. Note that, since α is countable, Cα ⊆ S must also be countable.
Since S is a Suslin line, no countable subset of S is dense in S. In particular, Cα is
not dense in S, so there exists a non-degenerate closed interval Iα = [aα , bα ] disjoint
from Cα . Define T = {Iα | α ∈ ω1 }, which is clearly uncountable, and partially ordered
by ≺ as defined above. To say that (T, ≺) is a tree, we still need to check that for an
element of T , the set of its predecessors is well-ordered by ≺. If α ∈ β ∈ ω1 , then by the
previous construction either Iα ≺ Iβ (meaning Iβ ( Iα ) or Iα and Iβ are incomparable
(i.e. Iα ∩ Iβ = ∅). Hence for every α ∈ ω1 we have that Iα ↓= {I ∈ T | I ≺ Iα } is
well-ordered by ≺ and (T, ≺) is a tree.
To prove that (T, ≺) is a Suslin tree, it remains to show that
(a) every antichain in T is at most countable;
(b) every branch in T is at most countable;
(c) ht(T ) = ω1 .
For (a) note that, by construction, I, J ∈ T are incomparable if and only if they are
disjoint intervals of S. Therefore, since S satisfies the ccc, every antichain in T is at
most countable.
For (b), we argue by contradiction: if b = {[aα , bα ] | α ∈ ω1 } ⊆ T is a branch of
T of length ω1 , then the left endpoints of these intervals form an increasing sequence
haα | α ∈ ω1 i of points of S. Then {(aα , aα+1 ) | α ∈ ω1 } is an uncountable collection of
nonempty pairwise disjoint open intervals in S, which contradicts the fact that S is ccc.
So every branch of T must be at most countable, i.e. T satisfies (b).
To show (c): since all branches of T are at most countable, ht(T ) is at most ω1 and
since every level of T is an antichain (and hence at most countable), and T is uncountable,
ht(T ) is at least ω1 . These two observations show (c), i.e. that ht(T ) = ω1 .
(⇐) Let (T, ≺) be a Suslin tree. By Lemma 3.8, we can assume that (T, ≺) is a
normal Suslin tree, i.e. T satisfies properties (i) to (viii) of Definitions 3.6 and 3.7. We
will construct a Suslin line (S, <).
Let S := {b ⊆ T | b is a branch in T }. Having defined S, we now proceed to construct
an ordering < on its elements, one which resembles a lexicographical ordering. By
property (vii), every x ∈ T has countably many immediate successors, so for each x ∈ T
we can fix a bijection σx : succT (x) → Q. Let a, b ∈ S, with a 6= b, a = {aξ | ξ ∈ α}
and b = {bξ | ξ ∈ β}. Note that a ⊆ b or b ⊆ is impossible, since branches are maximal
chains. By property (v), if a 6= b and γ is the least level at which a and b differ, then γ is
a successor ordinal and aγ and bγ are both successors of the same node c := aγ−1 = bγ−1 ,
accordingly we define
a < b :⇐⇒ σc (aγ ) <Q σc (bγ )
To show that (S, <) is a Suslin line, it remains to prove that
(a) S is linearly ordered by <;
(b) (S, <) satisfies the countable chain condition;
(c) (S, <) is not separable.

12
 3 TREES

We first show (a). As before, let a, b ∈ S, with a 6= b, a = {aξ | ξ ∈ α} and b =

{bξ | ξ ∈ β}. Again we denote by γ the least level at which a and b differ, and set
c := aγ−1 = bγ−1 . Then since aγ 6= bγ and σc is injective, we have that σc (aγ ) 6= σc (bγ ),
i.e. either σc (aγ ) <Q σc (bγ ) or σc (bγ ) <Q σc (aγ ). This proves that S is linearly ordered
by <. Moreover, it is easy to see that (S, <) is also dense and unbounded by using the
density and unboundedness of (Q, <Q ).
For (b), define Ix := {b ∈ S | x ∈ b} ⊆ S for each x ∈ T . Given (a, b) an arbitrary
nonempty open interval in S, i.e. (a, b) = {s ∈ S | a < s < b}, we claim that there is
x ∈ T such that Ix ⊆ (a, b). To see this, again let γ denote the least level at which a and
b differ, and set c := aγ−1 = bγ−1 . Since a < b, we have that σc (aγ ) <Q σc (bγ ). By the
density of <Q , we can find x ∈ succT (c) such that σc (aγ ) <Q σc (x) <Q σc (bγ ), hence we
have Ix ⊆ (a, b). Now notice that if Ix and Iy for x 6= y are such that Ix ∩ Iy = ∅, then
neither x ≺ y nor y ≺ x, otherwise we would have a branch in T which passes through
both x and y.
Let A be a family of pairwise disjoint nonempty open intervals in S. Using the
observations from the above paragraph, we can obtain an antichain in T with the same
cardinality as A. Since (T, ≺) is a Suslin tree, we conclude that A is at most countable,
i.e. (S, <) satisfies the ccc.
Finally let C ⊆ S be a countable subset of S, i.e. C is a countable set of branches of
T . Let α := sup{`(b) | b ∈ C}. Note that since T is a Suslin tree, `(b) ∈ ω1 for any b ∈ C.
This together with the fact that C is countable and ω1 is regular ensures that α ∈ ω1 .
Therefore we can let x ∈ Lα (T ). Since x has infinitely many immediate successors, we
can choose a, b ∈ S such that a 6= b and both contain x. Without loss of generality, we
can assume that a < b. Then the nonempty open interval (a, b) is disjoint from C, and
thus C is not dense in S. This proves (c).
Definition 3.10 (Aronszajn Tree). A tree (T, <) is called an Aronszajn tree if it has
the following properties:
(i) ht(T ) = ω1 ;
(ii) every branch in T is at most countable;
(iii) every level in T is at most countable.
Suslin trees are in particular also Aronszajn trees. We have discussed that the exis-
tence of a Suslin line cannot be proven or disproven, i.e., by the previous theorem, the
existence of a Suslin tree cannot be proven or disproven. In contrast, the next result
shows that Aronszajn trees can always be found.
Theorem 3.11. There exists an Aronszajn tree.
Proof. Let ω <ω1 denote the set of all transfinite sequences of elements of ω of length
less than ω1 , or in symbols ω <ω1 = α∈ω1 ω α = α∈ω1 {s : α → ω}. We start by set-
S S

ting T := {s ∈ ω <ω1 | s is injective} and ordering the elements of T by inclusion as in

Example 3.3. It is easy to see that T is a subtree of ω <ω1 , since restricting the domain
of an injective function always yields an injective function. Moreover, T still has height
ω1 , since for every α ∈ ω1 we can construct an injective function from α into ω, by the
fact that either |α| ∈ |ω| or |α| = |ω|. Moreover, if b would be an uncountable branch in
S
T , then b would be an injective function from ω1 to ω, which is impossible. Therefore
branches in T are at most countable.
The previous observations show that T satisfies (i) and (ii) of the definition of Aron-
szajn tree; T is, however, not an Aronszajn tree since, for example, there are uncountably

13
 4 JENSEN’S DIAMOND

many functions (in fact, c many) from ω to ω, so Lω (T ) is uncountable. We will fix this
by constructing a subtree S of T which is Aronszajn.
First, for α ∈ ω1 and given s, t ∈ ω α we define s =∗ t iff the set {β ∈ α | s(β) 6= t(β)} is
finite. Now consider a transfinite sequence hsα | α ∈ ω1 i satisfying the following properties
for every α, β ∈ ω1 :
(i) sα ∈ ω α = {t : α → ω};
(ii) sα is injective;
(iii) α ∈ β → sα =∗ sβ  α;
(iv) ω \ ran(sα ) is infinite.
Suppose that such a sequence exists. Then we can define S := α∈ω1 {t ∈ Lα (T ) | t =∗ sα }.
S

Note that S ⊆ T . Property (iii) above implies that S is a subtree of T . The definition of
S and properties (i) and (ii) above mean that sα ∈ S for every α < ω1 , so the levels of S
are nonempty, i.e. S also has height ω1 . Since S ⊆ T and we are simply considering the
ordering of T restricted to S, the property that every branch is at most countable is still
satisfied, but now S also has no uncountable levels, since {t ∈ ω α | t =∗ sα } is countable
for every α ∈ ω1 . Hence S is an Aronszajn tree.
To finish the proof, it remains to construct the sequence hsα | α ∈ ω1 i described above.
We will do this by recursion on α < ω1 . Start with s0 = ∅. For the successor steps of
the construction, assume that sα is given. Then, by (iv), it is always possible to pick
some n ∈ ω \ ran(sα ) and set sα+1 = sα ∪ {(α, n)}. For the limit steps, suppose we
have constructed sα for every α < λ with λ < ω1 a limit ordinal. Fix hαn | n ∈ ωi a
strictly increasing sequence of ordinals satisfying supn∈ω αn = λ. Using this, we define a
sequence of functions htn : αn → ω | n ∈ ωi as follows: let t0 = sα0 and having constructed
tn , choose tn+1 such that is is injective, tn+1 =∗ sαn+1 and tn+1 αn = tn . This is possible
by condition (iii) above. Now let t = n∈ω tn ; then t : γ → ω and t is injective. If we
S

would set sγ = t, then (i), (ii) would hold for α = γ, and (iii) would hold for every
α ∈ β = γ, which is promising. However, condition (iv) could fail. To fix this, we will
remove ℵ0 elements from ran(t). Let sγ (αn ) = t(α2n ) for every n ∈ ω and sγ (ξ) = t(ξ)
for ξ ∈ γ \ {αn | n ∈ ω}. Then we have that ω \ ran(sγ ) ⊇ {t(α2n+1 ) | n ∈ ω}, so condition
(iv) is preserved, and we have constructed hsα | α ∈ ω1 i satisfying (i), (ii), (iii) and (iv)
as desired.

4 Jensen’s Diamond
This section introduces the combinatorial principle ♦. The main objective is to prove
that under ♦ we can always construct a Suslin line, i.e. we show that ♦ → ¬SH.

Definition 4.1. Let α be a limit ordinal. A subset X ⊆ α is called bounded in α iff

sup(X) < α, and unbounded in α iff sup(X) = α.

Remark 4.2. The condition that the set X is unbounded in α can be restated as: for
every β < α there is some γ ∈ X such that γ > β.

Definition 4.3 (Closed Unbounded Set). A set C ⊆ ω1 is club or closed unbounded if it

satisfies the following properties:
(i) C is unbounded in ω1 , i.e. sup(C) = ω1 ;
(ii) C is closed: for all limit α < ω1 , if C ∩ α is unbounded in α, then α ∈ C.

14
 4 JENSEN’S DIAMOND

Remark 4.4. Condition (ii) can be restated as: for every countable sequence hcn | n ∈ ωi
of elements of C we have that sup{cn | n ∈ ω} ∈ C.
Definition 4.5 (Stationary Set). A set S ⊆ ω1 is said to be stationary iff for every
closed unbounded set C ⊆ ω1 we have that S ∩ C 6= ∅.
Lemma 4.6. Let S ⊆ ω1 be stationary. Then S is unbounded in ω1 .
Proof. Assume S ⊆ ω1 is stationary and consider the family of nonempty open intervals
A = {(α, ω1 ) ⊆ ω1 | α < ω1 }. Each a ∈ A is closed unbounded, so S ∩ a 6= ∅. Hence for
every α < ω1 we can find some β ∈ S with β > α.
Definition 4.7 (Diamond Sequence). A sequence of sets hSα | α < ω1 i is called a ♦-
sequence if it satisfies the following properties:
(i) Sα ⊆ α for every α < ω1 ;
(ii) for every X ⊆ ω1 the set {α ∈ ω1 | X ∩ α = Sα } ⊆ ω1 is stationary.
Definition 4.8 (Jensen’s ♦ Principle). There exists a ♦-sequence.
Theorem 4.9 (♦ → CH). If ♦ holds, then 2ℵ0 = ℵ1 .
Proof. Let hSα | α < ω1 i be a ♦-sequence. By definition, for every X ⊆ ω (⊆ ω1 ) the set
{α ∈ ω1 | X ∩ α = Sα } is stationary. By Lemma 4.6, this set is unbounded, and so we
have that for every such X ⊆ ω there exists an α with ω < α < ω1 satisfying X = Sα .
This induces a function f : P(ω) → ω1 , defined as f (X) := min {ω < α < ω1 | X = Sα }.
We claim that f is injective. Indeed, given X, Y ⊆ ω, if f (X) = f (Y ) then X = Sf (X) =
Sf (Y ) = Y . Since we constructed an injective function from P(ω) into ω1 , we conclude
that 2ℵ0 = ℵ1 .
Let (T, <) be a tree and A an antichain in T . Then A is called a maximal antichain
if there is no antichain B in T such that A ( B, hence every x ∈ T is comparable with
some a ∈ A. If A is a maximal antichain in T and T̃ ⊇ T is a tree extending T , then A
is not necessarily maximal in T̃ .
Lemma 4.10. n Let (T, <) be a normal ω1 -tree. Let A be a maximal
o antichain in T . Then
(α) (α)
the set C := α ∈ ω1 | A ∩ T is a maximal antichain in T is closed unbounded.

Proof. Since A is a maximal antichain in T , for any α ∈ ω1 the set A∩T (α) is an antichain
in T (α) . We first show that C is unbounded. We construct a sequence hαn | n ∈ ωi by
recursion. Start with an arbitrary α0 < ω1 . Then, since T (α0 ) is countable and every
element of T (α0 ) is comparable to some element of A, there exists α1 with α0 < α1 < ω1
such that every element of T (α0 ) is comparable with some element of A ∩ T (α1 ) . Then
we find α2 with α1 < α2 < ω1 such that every element of T (α1 ) is comparable with some
element of A ∩ T (α2 ) , and so on. Setting α := sup {αn | n ∈ ω} ∈ ω1 , we have that every
element of T (α) is comparable to some element of A∩T (α) , therefore A∩T (α) is a maximal
antichain in T (α) , i.e. α ∈ C. Our choice of α0 < α was arbitrary, so we conclude that
C ⊆ ω1 is unbounded.
Now we prove that C is closed. Let hαn | n ∈ ωi be a countable sequence of elements
of C. We want to show that α := {αn | n ∈ ω} ∈ C. Any element of T (α) belongs to
T (αn ) for some n ∈ ω, so any element of T (α) is comparable so some element of A ∩ T (αn )
for some n ∈ ω. It follows that any element of T (α) is comparable to some element of
A ∩ T (α) , which shows that A ∩ T (α) is a maximal antichain in T (α) , i.e. α ∈ C.

15
 4 JENSEN’S DIAMOND

Lemma 4.11. Let (T, <) be a tree where each node has at least two immediate succes-
sors. If T has no uncountable antichains then T has no uncountable branches.
Proof. Let (T, <) be a tree where each node has at least two immediate successors. Let
{xα | α ∈ ω1 } be an uncountable chain in T , i.e. a linearly ordered subset of T . For each
α ∈ ω1 , the set of immediate successors of xα has at least two elements, so we can define
yα+1 to be an immediate successor of xα with yα+1 6= xα+1 . Then {yα+1 | α ∈ ω1 } is an
uncountable antichain in T .
Theorem 4.12 (♦ → ¬SH). If ♦ holds, then Suslin lines exist.
Proof. We will use ♦ to construct a Suslin tree (T, <). This implies the existence of a
Suslin line, by Theorem 3.9.
We set T = ω1 , i.e. the nodes of our tree are the countable ordinals. The construction
will be done by recursion on the levels of T . Each T (α) = {x ∈ T | ht(x) ∈ α} will be
some countable ordinal, i.e. an initial segment of T = ω1 . So every time we add a node,
it will be the least countable ordinal not yet used. Moreover, our construction will ensure
that each T (α) is a normal α-tree.
Let L0 (T ) = {0} = T (1) . Now if α is a limit ordinal, then T (α) will simply be the
union of the trees T (β) for β < α. If α is a successor ordinal, then T (α+1) is obtained
from T (α) by adding ℵ0 immediate successors to each node at level α − 1. Recall that we
always pick the new nodes in order, i.e. in such a way that each T (α) for α < ω1 is an
initial segment of ω1 . It is easy to check that the steps of the construction described in
this paragraph preserve the property that each T (α) is an α-tree, in particular property
(viii) is preserved because we are either taking a countable union of countable sets or
adding countable successors to each node in a countable level.
We still need to describe the construction of T (α+1) in the case that α is a limit ordinal.
Let hSα | α ∈ ω1 i be a ♦-sequence. Recall that Sα ⊆ α, and, moreover, α ⊆ T (α) . If Sα
is a maximal antichain in T (α) , then for each x ∈ T (α) there is a ∈ Sα such that either
x < a or a < x. Since T (α) is a normal α-tree, we can choose a branch bx of height α
that contains both x and a. Hence for each x ∈ T we add one node at level α to lie
above the elements of bx . This means that Sα is still a maximal antichain in T (α+1) , and
the construction ensures that T (α+1) is a normal (α + 1)-tree. On the other hand, if Sα
is not a maximal antichain in T (α) , we can extend T (α) in any way such that T (α+1) is a
normal (α + 1)-tree, i.e. we simply choose for each x ∈ T a branch bx of height α and
add one node at level α to lie above the elements of bx .
The construction of our tree T = α∈ω1 T (α) ensures that T is a normal ω1 -tree. We
S

now want to argue that T is a Suslin tree. By Lemma 4.11, we only need to show that
T has no uncountable antichains.
n Let A ⊆ T = ω1 be a maximal antichainoin T . By
Lemma 4.10, the set C := α ∈ ω1 | A ∩ T (α) is a maximal antichain in T (α) is closed
unbounded. Moreover, the set of α ∈ C such that T (α) = α is also closed unbounded3 .
Now since hSα | α ∈ ω1 i is a ♦-sequence, we find a limit α ∈ ω1 such that T (α) = α and
A ∩ T (α) is a maximal antichain in T (α) and A ∩ α = Sα . It follows that we find limit
α ∈ ω1 such that A ∩ T (α) = Sα and A ∩ T (α) is a maximal antichain in T (α) . Our
construction of T now ensures that A ∩ T (α) remains a maximal antichain in T , so it
must be that A = A ∩ T (α) . By fact that T (α) is countable, we have that A is countable,
which is what we wanted to show.
3
Closure is clear. For unboundedness, note that α ⊆ T (α) for any α ∈ ω1 , and construct a sequence
hαn | n ∈ ωi of elements of C such that T (αn ) ⊆ αn+1 . This can be done since C is club, and we get
equality T (α) = α at the supremum of the sequence.

16
 5 GÖDEL’S CONSTRUCTIBLE UNIVERSE

5 Gödel’s Constructible Universe

The aim of this section is to present Gödel’s class L, also known as Gödel’s Constructible
Universe, and show that ♦ holds in L. This completes the proof of the non-provability
of SH, since ♦ → ¬SH as we saw in Theorem 4.12. To do this, we will of course have
to show that all the ZFC axioms hold in L. The proof that the Generalized Continuum
Hypothesis holds in L will also be presented.

Definition 5.1. Let A be a set. We say that X ⊆ A is definable over the model
(A, ∈) iff there exists a formula ϕ in the language of set theory L = {∈} and some
a1 , . . . , an ∈ A such that X = {x ∈ A | (A, ∈) |= ϕ[x, a1 , . . . , an ]}. Then let def(A) :=
{X ⊆ A | X is definable over (A, ∈)}, i.e. def(A) is the set of all subsets of A that are
definable over (A, ∈).

Lemma 5.2. Let A be a set. Then we have that

(a) A ∈ def(A);
(b) def(A) ⊆ P(A);
(c) A transitive → A ⊆ def(A);
(d) every finite subset of A is in def(A);
(e) assuming AC, |def(A)| = |A| whenever |A| ≥ ω.

Proof. (a) In Definition 5.1, take ϕ to be a tautology.

(b) Clear, since the elements of def(A) are subsets of A, i.e. elements of P(A).
(c) Since A is transitive, for a ∈ A we have that a = {x ∈ A | x ∈ a} ∈ def(A).
(d) Let a1 , . . . , an ∈ A for some n ∈ ω. We want to show that {a1 , . . . , an } ∈ def(A).
This is immediate if we let ϕ in Definition 5.1 to be the formula x = y1 ∨ · · · ∨ x = yn .
(e) Assume AC and |A| ≥ ω. On the one hand we have that |def(A)| ≥ |A|, since
{a} ∈ def(A) for every a ∈ A by part (d). On the other hand it is also true that
|def(A)| ≤ |A|, since there are only countably many formulas (since formulas are
finite sets of symbols) and only countably many finite subsets of A to be used as
parameters in a given formula.
We can now define L. The definition is identical to the definition of the von Neumann
hierarchy of sets V (α), except that we replace the power set operation ‘P’ by ‘def’.

Definition 5.3. Define L(α) via transfinite recursion on α ∈ ON as:

• L(0) = ∅;
• L(α + 1) = def(L(α));
• L(λ) = α<λ L(α) when λ is a limit ordinal.
S
S
Then set L = α∈ON L(α). We call the proper class L the constructible universe.

Below we show some basic properties of L.

Lemma 5.4. For every ordinal α we have that

(a) L(α) ⊆ V (α);
(b) L(α) is transitive;
(c) ∀β ≥ α(L(α) ⊆ L(β));
(d) L(α) ∩ ON = α, so ON ⊆ L;
(e) L(α) ∈ L(α + 1), which implies L(α) ∈ L;
(f) every finite subset of L(α) is in L(α + 1).

17
 5 GÖDEL’S CONSTRUCTIBLE UNIVERSE

Proof. (a) Follows easily by transfinite induction, using the fact that def(A) ⊆ P(A),
shown in Lemma 5.2 (b).
(b) Use transfinite induction on α. The steps where α is 0 or a limit ordinal are
easy. Now assume L(α) is transitive and note that L(α) ⊆ L(α + 1), since any
a ∈ L(α) can be written as a = {x ∈ L(α) | x ∈ a} ∈ L(α + 1). Then, since
L(α + 1) ⊆ P(L(α)), we have that x ∈ L(α + 1) → x ⊆ L(α) ⊆ L(α + 1), which
concludes the proof.
(c) We will prove the statement for an arbitrary fixed α by transfinite induction on β.
The cases where β is α or β is a limit ordinal are immediate, and the remaining
case is covered by L(α) ⊆ L(β) ⊆ L(β + 1), where the second inclusion was proven
in part (b).
(d) Once again we prove the statement using transfinite induction on α. The step α = 0
is trivial, and the step where α is a limit is also easy. Therefore we assume that
L(α) ∩ ON = α, and attempt to show L(α + 1) ∩ ON = α + 1. But L(α) ⊆ L(α + 1)
using part (c), and L(α + 1) = def(L(α)) ⊆ P(L(α)) using Lemma 5.2 (b), so α =
L(α)∩ON ⊆ L(α +1)∩ON ⊆ V (α +1)∩ON = α ∪{α}, which means that we only
need to show that α ∈ L(α + 1). Using the fact that being an ordinal can be written
as a ∆0 formula and part (b) which says that L(α)n is transitive, we have that oα=
L(α)
L(α) ∩ ON = {x ∈ L(α) | x is an ordinal} = x ∈ L(α) | (x is an ordinal) ∈
def(L(α))n = L(α + 1). o
(e) L(α) = x ∈ L(α) | (x = x)L(α) ∈ def(L(α)) = L(α + 1).
(f) This follows directly from Lemma 5.2 (d).
Given x ∈ L, by the definition of L it is clear that the least α such that x ∈ L(α) is
a successor ordinal. This justifies the following definition.

Definition 5.5. Let x ∈ L. Define ρ(x), the L-rank of x, as the least ordinal β such
that x ∈ L(β + 1).

Lemma 5.6. For any ordinal α the following properties hold:

(a) L(α) = {x ∈ L | ρ(x) < α};
(b) L(α + 1) \ L(α) = {x ∈ L | ρ(x) = α};
(c) ρ(L(α)) = ρ(α) = α.

Proof. (a) For x ∈ L, ρ(x) < α iff ∃β < α(x ∈ L(β + 1)) iff x ∈ L(α).
(b) Follows easily from (a).
(c) L(α) ∈/ L(α) is clear, and in the previous lemma we showed that L(α) ∈ L(α + 1),
hence ρ(L(α)) = α. Also by the previous lemma, α ∈ / L(α) and α ∈ L(α + 1), i.e.
ρ(α) = α.

Lemma 5.7.
(a) |L(α)| = |α| for every α ≥ ω;
(b) L(α) = V (α) for every α ≤ ω;
(c) L(ω + 1) ( V (ω + 1).

Proof. (a) By Lemma 5.4 (d) we have that α ⊆ L(α), so |L(α)| ≥ |α|. We prove
|L(α)| = |α| by transfinite induction on α ≥ ω. If α = ω, we just need to show that
L(ω) is countable, but this is clear since L(ω) is a countable union of finite sets. If
|L(α)| = |α|, then |L(α + 1)| = |L(α)| = |α| = |α + 1| using 5.2 (e), which covers
successor steps. Finally if |L(α)| = |α| for all α with ω ≤ α < β with β a limit

18
 5 GÖDEL’S CONSTRUCTIBLE UNIVERSE

ordinal, then |L(α)| ≤ |β| for all such α, therefore we have that L(β) is a union of
|β|-many sets of cardinality ≤ |β|, so |L(β)| ≤ |β| using AC, hence |L(β)| = |β|.
(b) Using Lemma 5.4 (f), it is immediate to prove L(n) = V (n) for any n ∈ ω using
S
induction on n. Then L(ω) = V (ω) is clear since L(ω) = n∈ω L(n) and V (ω) =
S
n∈ω V (n).
(c) Note that |V (ω + 1)| = |P(V (ω))| which is uncountable, whereas L(ω + 1) =
def(L(ω)) is countable using (a) and Lemma 5.2 (e).

Theorem 5.8. All axioms of ZF hold in L.

For proving Theorem 5.8, we will make use of two results, which we state here without
further discussion. For detailed proofs, see [5].

Lemma 5.9. Suppose that M is a transitive class such that the Comprehension Axiom
holds in M , and assume that for every subset x ⊆ M , there is a set y ∈ M such that
x ⊆ y. Then all the ZF axioms hold in M .

Theorem 5.10 (Reflection). Let ϕ0 , ϕ1 , . . . , ϕn−1 be any list of formulas for the language
L = {∈}. Assume that B is a non-empty class and A(ξ) is a set for each ξ ∈ ON , and
assume that:
(i) ξ < η → A(ξ) ⊆ A(η).
S
(ii) A(η) = ξ<η A(ξ) for limit η.
S
(iii) B = ξ∈ON A(ξ).
Then ∀ξ∃η > ξ(A(η) 6= ∅ ∧ i<n (A(η) 4ϕi B) ∧ η is a limit ordinal).
V

Proof of Theorem 5.8. In Lemma 5.4 (b) we showed that each L(α) is transitive, so
we know that L is a transitive class. By Lemma 5.9, we only need to verify that the
Comprehension Axiom holds in L, and that for every subset x ⊆ L there is a y ∈ L such
that x ⊆ y. This last condition is straightforward to prove, since given x ⊆ L we can let
y = L(α) for α = sup {ρ(z) + 1 | z ∈ x}.
Now we show that Comprehension Axiom holds in L. Fix a formula ϕ without y
among its free variables. The formula ϕ may have x, z and also possibly other variables
v0 , . . . , vn−1 free, therefore we write it as ϕ(x, z, v0 , . . . , vn−1 ). The Comprehension Axiom
reads in this case as:

∀z, v0 , . . . , vn−1 ∈ L ∃y ∈ L ∀x ∈ L(x ∈ y ↔ x ∈ z ∧ ϕL (x, z, ~v ))

n o
We begin by fixing arbitrary z, v0 , . . . , vn−1 ∈ L, and let y = x ∈ z | ϕL (x, z, ~v ) .
The task is now to verify that y ∈ L. Since z, v0 , . . . , vn−1 ∈ L, we can fix α such
that z, v0 , . . . , vn−1 ∈ L(α).
n By Theorem 5.10 o(Reflection), there is β > α such that
L(β) 4ϕ L. Then y = x ∈ L(β) | ψ L(β) (x, z, ~v ) ∈ def(L(β)) = L(β + 1) ⊆ L, where
ψ(x, z, ~v ) is the formula ϕ(x, z, ~v ) ∧ x ∈ z, so y ∈ L.

Definition 5.11. The Axiom of Constructibility, denoted V = L, is the statement that

every set is constructible, i.e. a member of L. In symbols: ∀x∃α(x ∈ L(α)).

Lemma 5.12. The function α 7→ L(α) is absolute for transitive M |= ZF \ (P ow).

Proof sketch. It is a basic fact that the notion of ordinal is absolute for transitive models
of ZF \ (P ow). Moreover, it can be shown that recursively defined notions such as “ϕ is

19
 5 GÖDEL’S CONSTRUCTIBLE UNIVERSE

an L-formula” and “A |= ϕ” are absolute for transitive models of ZF \ (P ow) (see, e.g.,
[5, pp. 124–125]). Therefore ‘def(A)’ is absolute.
The theorem of transfinite recursion only requires ZF \(P ow) to work, and this suffices
for the definition of L(α). So (L(α))M is defined for ordinals α ∈ M , and the equality
(L(α))M = L(α) can then be proven by transfinite induction using the absoluteness of
def(A) discussed above.

Corollary 5.13. The Axiom of Constructibility holds in L.

Proof. We need to check (V = L)L , i.e. (∀x∃α(x ∈ L(α)))L , which is the same as ∀x ∈
L∃α ∈ L(x ∈ L(α))L . Now using, from Lemma 5.4 (d), that ON ⊆ L, the absoluteness
of being an ordinal and Lemma 5.12 this reduces to ∀x ∈ L∃α ∈ ON (x ∈ L(α)), which
S
is clear from L = α∈ON L(α).

Definition 5.14. Let M be a transitive set. Define θ(M ) := M ∩ ON , i.e. θ(M ) is the
set of ordinals in M , or, equivalently, the first ordinal not in M .

Remark 5.15. Note that θ(M ) is always a limit ordinal whenever M is a model of
ZF \ (P ow), since α ∈ M → (α + 1) = α ∪ {α} ∈ M .

Lemma 5.16. Let M be a transitive set. Assume that M is a model of ZF \ (P ow).

Then M is a model of V = L if and only if M = L(θ(M )).

Proof. Let M be a transitive set with M |= ZF \ (P ow). Let γ = θ(M ). By Lemma

5.12, we have that L(α) ∈ M for every α < γ. Also recall from Remark 5.15 that γ is a
S
limit ordinal, so L(γ) = α<γ L(α). This two observations, together with the transitivity
of M, imply that L(γ) ⊆ M .
The statement (V = L)M , which is (∀x∃α(x ∈ L(α)))M , reduces in this case to
∀x ∈ M ∃α ∈ γ(x ∈ L(α)), and this is equivalent to M ⊆ L(γ) = α<γ L(α), i.e.
S

equivalent to M = L(γ). Therefore we have shown that (V = L)M is equivalent to

M = L(γ) = L(θ(M )).

Theorem 5.17. There is a relation <L that well-orders L. Therefore, assuming V = L,

the relation <L well-orders V and AC holds.

Proof sketch. For a complete formal proof see [4, pp. 188–190] or [5, pp. 139–140]. We
sketch here the main ideas involved in the construction of <L .
First, we argue why def(A) can be well-ordered, given a well-order <A of A. Using
this fact, by recursion we piece together a well-order of L(α), for each α. Finally we use
these to construct a well-order on L.
Recall, from Definition 5.1, that X ∈ def(A) iff X = {x ∈ A | (A, ∈) |= ϕ[x, a1 , . . . , an ]}
for a formula ϕ and some a1 , . . . , an ∈ A. First, notice that there are only countably
many possible formulas ϕ in the language of set theory {∈}, and we can well-order them
using the standard Gödel numbering. What’s more, the set A<ω of all possible tuples
of parameters from A can be well-ordered, assuming <A well-orders A, as follows: for
(a1 , . . . , an ) ∈ A<ω and (a01 , . . . , a0m ) ∈ A<ω , let (a1 , . . . , an ) < (a01 , . . . , a0m ) iff n < m or,
n = m and (a1 , . . . , an ) <lex (a01 , . . . , a0n ), where <lex is the lexicographic ordering based
on <A .
Now given X, Y ∈ def(A), we can let ϕX and ϕY be the least formulas, in the above
sense, such that there are some parameters witnessing X, Y ∈ def(A). Having fixed
ϕX and ϕY , let ~aX and ~aY be the least parameters, in the above sense, witnessing

20
 5 GÖDEL’S CONSTRUCTIBLE UNIVERSE

X, Y ∈ def(A) for ϕX and ϕY . Then we can give a well-ordering for def(A) as follows:
X <def(A) Y iff ϕX < ϕY or, ϕX = ϕY and ~aX < ~bX .
Proceeding with the construction we can now define, by recursion on α, a well-ordering
Cα ⊆ L(α) × L(α) as follows: x Cα y iff ρ(x) < ρ(y) or, ρ(x) = ρ(y) and x <def(L(ρ(x))) y.
Finally we define the relation <L on L by: x <L y iff ρ(x) < ρ(y) or, ρ(x) = ρ(y) and
x Cρ(x)+1 y.
The above argument actually shows that V = L implies global choice, i.e. there is one
relation <L that well-orders everything.

Lemma 5.18. Let κ be a regular uncountable cardinal. Then ZF \ (P ow) and V = L

hold in L(κ).

Proof sketch. That V=L holds is a consequence of Lemma 5.16. Extensionality, Foun-
dation, Pairing, Union and Infinity are not hard to show, and we skip them here. For a
proof see [5]. For Comprehension, the proof is almost the same as the argument used in
Theorem 5.8, but replacing L with L(κ), and with α and β now ranging over κ instead
of over ON . Also, instead of using the Reflection Theorem, a “set version” of it is used,
where ON is replaced by the regular cardinal κ. We show Replacement in detail below.
Replacement: Let A ∈ L(κ) and suppose that

∀x ∈ L(κ)(x ∈ A → ∃!y ∈ L(κ) ϕL(κ) (x, y)).

We need to produce B ∈ L(κ) such that (∀x ∈ A ∃y ∈ B ϕ(x, y))L(κ) . Say A ∈ L(α) for
α < κ. Then, using Lemma 5.7 (a), |A| ≤ |L(α)| = |α| < κ. Write f for the function
with dom(f ) = A such that f (x) is the unique y ∈ L(κ) such that ϕL(κ) (x, y). Then
ρ(f (x)) < κ for each x ∈ A, so β := sup{ρ(f (x)) + 1 | x ∈ A} < κ since κ is regular and
|A| < κ. We can now let B = L(β), and we get B ∈ L(κ) because ρ(B) = β + 1 < κ.

Lemma 5.19. Assume V = L. Then L(κ) = H(κ) for all cardinals κ ≥ ω.

Proof. For any cardinal κ, recall that H(κ) denotes the set of all x such that |trcl(x)| < κ.
In Lemma 5.7 (b) we showed that L(ω) = V (ω), and it is a standard fact that H(ω) =
V (ω), so we only need to consider κ > ω.
We first show that L(κ) ⊆ H(κ) for all cardinals κ. Let x ∈ L(κ) for some cardinal
κ > ω. Since κ is a cardinal, we can fix α with x ∈ L(α) and ω ≤ α < κ. Then
trcl(x) ⊆ L(α) since L(α) is transitive, so |trcl(x)| ≤ |L(α)| = |α| < κ and we conclude
that x ∈ H(κ).
Note that, in order to show L(κ) = H(κ), it suffices to only consider successor cardi-
nals of the form λ+ for some infinite cardinal λ, since for an uncountable limit cardinal
κ we have L(κ) = λ<κ L(λ+ ) = λ<κ H(λ+ ) = H(κ).
S S

By the above remarks, we are left with showing H(λ+ ) ⊆ L(λ+ ) for all infinite
cardinals λ. Let b ∈ H(λ+ ) and let T = trcl({b}). So we have b ∈ T and |T | ≤ λ.
Since we are assuming V = L, we can consider ρ(T ) and fix some regular uncountable
cardinal γ > ρ(T ). Then, since T ∈ L(γ) and L(γ) is transitive, T ⊆ L(γ) holds. Also,
using Lemma 5.18, L(γ) |= ZF \ (P ow) + V = L. By the Downward Löwenheim-Skolem
Theorem and the fact that |T | ≤ λ, we can fix A such that A 4 L(γ) and T ⊆ A and
|A| ≤ λ. Then also A |= ZF \ (P ow) + V = L since A 4 L(γ).
Now let mos be the Mostowski collapsing function on A, and let B = ran(mos) the
collapse of A. Then mos is an isomorphism from (A, ∈) to (B, ∈) and B is transitive,

21
 5 GÖDEL’S CONSTRUCTIBLE UNIVERSE

and mos(x) = x for all x ∈ T (since T is transitive). Hence b = mos(b) ∈ B since

b ∈ T . Moreover, since mos is an isomorphism, B |= ZF \ (P ow) + V = L. This
yields, by Lemma 5.16 and the transitivity of B, that B = L(β) for β = θ(B). Then,
using Lemma 5.7 (a), |β| = |L(β)| = |B| = |A| ≤ λ, so β < λ+ , which implies that
b ∈ B = L(β) ⊆ L(λ+ ).
Theorem 5.20. Assume V = L. Then GCH holds.
Proof. Assume V = L. By Theorem 5.17, AC holds. Let λ be an infinite cardinal. The
inclusion P(λ) ⊆ H(λ+ ) is clear from the definitions. By Lemma 5.19, H(λ+ ) = L(λ+ ),
so P(λ) ⊆ L(λ+ ) and we get that 2λ ≤ |L(λ+ )| = λ+ using Lemma 5.7 (a), which
required AC. Therefore 2λ = λ+ , since 2λ ≥ λ+ by Cantor’s diagonal argument.
Our aim is now to show that under V = L the Diamond Principle is true. For this
purpose we state the following Lemma. For a proof see [5].
Lemma 5.21. Let γ be an uncountable cardinal, and let M be countable with M 4
H(γ). Then the following statements are true:
(a) If a ∈ M and a is countable, then a ⊆ M .
(b) M ∩ ω1 is a countable limit ordinal.
(c) If γ = ω1 , then M is transitive.
(d) If γ > ω1 , then ω1 ∈ M and ω1 ( M , and if also β = M ∩ ω1 and mos is
the Mostowski isomorphism from M onto a transitive T , then mos(ω1 ) = β and
mos(ξ) = ξ for all ξ < β. Moreover, T |= ZF C \ (P ow) if γ is regular, and
β = (ω1 )T .
Theorem 5.22. Assume V = L. Then ♦ is true.
Proof. Assume V = L throughout. Recall, from Theorem 5.17, that there is a relation
<L that well-orders L. Using similar methods to the ones used in the proof of Lemma
5.12, it can be seen that <L is absolute for transitive models of ZF \ (P ow) + V = L.
By recursion we construct a sequence of pairs (Sα , Cα ). For α a limit ordinal, let

P (α, S, C) ⇐⇒ S ⊆ α ∧ C ⊆ α ∧ C is club in α ∧ ¬∃ξ ∈ C(S ∩ ξ = Sξ )

and let (Sα , Cα ) be the <L -least pair such that P (α, S, C). If α is not a limit ordinal, or
if no such pair exists, then let Sα = Cα = ∅. Here we are using a generalized version of
Definition 4.3 of club set, were we replace ω1 for an arbitrary limit ordinal. Note that
(Sα , Cα ) is defined for all ordinals α, and the definition is absolute for transitive models
of ZF \ (P ow) + V = L.
We will show that hSα |α < ω1 i is a ♦-sequence. Assume it is not. By Definition 4.5
and Definition 4.7, this is equivalent to the existence of sets S, C such that P (ω1 , S, C)
holds. Therefore P (ω1 , Sω1 , Cω1 ) holds, by the construction of (Sα , Cα ).
Recall, from Lemma 5.19, that under V = L we have that L(κ) = H(κ) for all cardi-
nals κ ≥ ω, so L(ω2 ) = H(ω2 ) and using the Downward Löwenheim-Skolem Theorem we
can get a countable M 4 H(ω2 ) = L(ω2 ). Now let β = M ∩ ω1 , which is some countable
limit ordinal by Lemma 5.21 (b). Let mos be the Mostowski isomorphism from M onto
some transitive model T . Then, using Lemma 5.21 (c), mos(ω1 ) = β, mos(ξ) = ξ for all
ξ < β, and β = (ω1 )T . Also, combining M 4 H(ω2 ) = L(ω2 ) with Lemma 5.18 and the
regularity of ω2 , we have that T is a countable transitive model of ZF \ (P ow) and of
V = L. Therefore, by Lemma 5.16, T = L(γ) for some countable γ.

22
 6 MARTIN’S AXIOM

By the absoluteness of <L and of P (α, S, C), and by M 4 L(ω2 ) = H(ω2 ) we have
that, for each α ∈ M , Sα ∈ M and Cα ∈ M and mos(Sα ) = Smos(α) and mos(Cα ) =
Cmos(α) . So Sω1 ∈ M and Cω1 ∈ M and mos(Sω1 ) = Sβ and mos(Cω1 ) = Cβ . By the def-
inition of mos, we also have mos(Sω1 ) = {mos(ξ) | ξ ∈ Sω1 ∩ M } = {ξ | ξ ∈ Sω1 ∩ M } =
Sω1 ∩ β. Therefore Sβ = Sω1 ∩ β. Analogously, mos(Cω1 ) = Cβ = Cω1 ∩ β.
Proceeding similarly as above, by P (ω1 , Sω1 , Cω1 ) we know that Cω1 is club in ω1 , so
mos(Cω1 ) is club in mos(ω1 ), i.e. Cω1 ∩ β is club in β. Then β ∈ Cω1 since Cω1 is closed.
Now β ∈ Cω1 and Sβ = Sω1 ∩ β is in direct contradiction with P (ω1 , Sω1 , Cω1 ).

6 Martin’s Axiom
In this section we will present a second combinatorial principle called Martin’s Axiom
and show that one of its consequences is the nonexistence of Suslin lines.
Up to this point, our partial orders have been strict, i.e. taking < instead of ≤ as
basic. In the context of Martin’s Axiom, it is more common to use the non-strict form
(P, ≤). Moreover, in this section we will actually work with pre-orders, which are not
required to be anti-symmetric.
Definition 6.1. (Pre-Order) Let P be a non-empty set. A pre-order of P is a bi-
nary relation ≤ on P which is reflexive, i.e. ∀p ∈ P (p ≤ p), and transitive, meaning
∀pqr ∈ P (p ≤ q ∧ q ≤ r → p ≤ r). We call the pair (P, ≤) a pre-ordered set.
Definition 6.2. Let (P, ≤) be a pre-ordered set. Two elements p1 , p2 ∈ P are called
compatible (denoted p1 6⊥ p2 ) iff there exists q ∈ P such that p1 ≤ q ≥ p2 4 ; they are
called incompatible (denoted p1 ⊥ p2 ) otherwise. An antichain in P is a subset A ⊆ P
such that elements of A are pairwise incompatible, i.e. ∀pq ∈ P (p 6= q → p ⊥ q). If
every antichain in P is at most countable, we say that (P, ≤) satisfies the countable chain
condition or the ccc for short.
Remark 6.3. In the context of trees, we used the term antichain to refer to a subset
of incomparable elements, whereas here we are talking about incompatible elements.
These are two different notions in general. It is, however, not hard to see that when
talking about trees, these two definitions are equivalent, i.e. two elements of a tree are
comparable iff they are compatible, since the set of predecessors of any node is linearly
ordered.
Remark 6.4. Any linearly ordered set trivially satisfies the countable chain condition
as defined in the current section, hence this definition and the one given previously for
topological spaces are not equivalent.
Definition 6.5. Let (P, ≤) be a pre-ordered set. Let ∅ = 6 C ⊆ P . Then we say that
(i) C is cofinal iff ∀p ∈ P ∃q ∈ C(q ≥ p);
(ii) C is downwards closed iff ∀p ∈ C ∀q ∈ P (q ≤ p → q ∈ C);
(iii) C is directed iff ∀p1 p2 ∈ C ∃q ∈ C(p1 ≤ q ≥ p2 );
(iv) C is a filter iff C is directed and downwards closed.
Definition 6.6. Let (P, ≤) be a pre-ordered set. Let G ⊆ P be a filter and let C be a
family of cofinal subsets of P . Then G is called C-generic iff G ∩ C 6= ∅ for every C ∈ C.
4
This is called the Jerusalem convention. Many texts define compatibility in terms of p1 ≥ q ≤ p2 ,
which is called the American convention.

23
 6 MARTIN’S AXIOM

Lemma 6.7. (Rasiowa–Sikorski) Let (P, ≤) be a pre-ordered set and let C be a collection
of cofinal subsets of P with |C| ≤ ℵ0 . Then for every p ∈ P there exists a C-generic filter
G such that p ∈ G.

Proof. Let C = {Cn | n ∈ ω}. We construct a sequence hpn | n ∈ ωi by recursion. Let

p0 = p. Given pn , define pn+1 = q for some q ∈ Cn such that pn ≤ q; this is possible
since Cn is cofinal. Now we define G := {r ∈ P | r ≤ pn for some n ∈ ω} 3 p. Then G is
clearly directed and downwards closed, i.e. it is a filter. Moreover, G ∩ C 6= ∅ for every
C ∈ C, so G is also C-generic.
The previous lemma shows that it is only interesting to consider whether C-generic
filters exist in the case of uncountable C. The next lemma shows that, in that case, such
filters do not always exist.

Lemma 6.8. There exists a non ccc pre-ordered set (P, ≤) and an uncountable family
C of cofinal subsets of P such that no filter G ⊆ P is C-generic.

Proof. Let P := {p ⊂ ω × ω1 | p is a function with finite domain}. Clearly (P, ⊆) is a

pre-ordered set, and the ordering is also anti-symmetric. Note also that P does not
satisfy the ccc, since, for example, {{(0, α)} | α ∈ ω1 } is an uncountable antichain.
Define, for each α ∈ ω1 , the set Cα := {p ∈ P | α ∈ ran(p)}. We claim that each Cα
is cofinal in P . Indeed, if p ∈ P and α ∈ ran(p), then p ∈ Cα and we are done. On the
other hand, if α ∈ / ran(p), then we have p ⊆ p ∪ {(n, α)} ∈ Cα , where we pick n such
that n ∈/ dom(p) (this is possible since dom(p) is finite).
Let G ⊆ P be a C-generic filter for C := {Cα | α ∈ ω1 }. Note that C is uncountable.
Then G is a family of functions with finite domain, and, being directed, any two elements
S
of G are compatible. For this reason f = G is a function from a subset of ω into ω1 .
Since ran(f ) is at most countable, there exists some β ∈ ω1 \ ran(f ). This contradicts
the fact that G ∩ Cβ 6= ∅.

Definition 6.9 (Martin’s Axiom). Let κ be a cardinal.

• MA(κ) is the statement: If (P, ≤) is a ccc pre-ordered set, and C is a family of
cofinal subsets of P with |C| ≤ κ, then there is a filter G ⊆ P which is C-generic.
• MA is the statement: MA(κ) holds for each κ < 2ℵ0 .

Having stated Martin’s Axiom, in the next lemma we show a few easy results related
to it. Most importantly, we illustrate that MA is only interesting when CH fails.

Lemma 6.10.
(a) If λ < κ, then MA(κ) → MA(λ).
(b) MA(2ℵ0 ) is false;
(c) MA(κ) is true whenever κ ≤ ℵ0 ;
(d) CH → MA;
(e) MA(κ) → κ < c = 2ℵ0 ;
(f) MA(ℵ1 ) → ¬CH.

Proof. Statement (a) is obvious, and (c) is immediate from Lemma 6.7, where we also
didn’t make use of the ccc hypothesis. We prove (b).
Let P := {p ⊂ ω × 2 | p is a function with finite domain}. Similarly to Lemma 6.8,
we have that (P, ⊆) is a pre-ordered set. It is easy to see that (P, ⊆) satisfies the ccc,
since P is countable.

24
 6 MARTIN’S AXIOM

Define, for each n ∈ ω, the set Cn := {p ∈ P | n ∈ dom(p)}. It is easy to see that,

given n ∈ ω, any p ∈ P can be extended to include n in its domain, so every Cn is cofinal
in P . Given h : ω → 2, define Eh := {p ∈ P | ∃n ∈ dom(p)(p(n) 6= h(n))}. Since h is
defined on all of ω, any p ∈ P can be extended in such a way that it disagrees somewhere
with h, so any Eh is cofinal.
Let C = {Cn | n ∈ ω} ∪ {Eh | h : ω → 2}. Then C is a family of cofinal subsets of P ,
and |C| = 2ℵ0 . Let G ⊆ P be a C- generic filter. Again, as in Lemma 6.8, f = G is a
S

function, since G is a filter. Having nonempty intersection with every Cn ensures that
dom(f ) = ω. Now, since for every h : ω → 2 we have that G ∩ Eh 6= ∅, f 6= h. Therefore
f is a function from ω to 2 which is different from every function from ω to 2. This
contradiction concludes the proof of (b).
For (d), assume CH. The condition κ < 2ℵ0 reduces to κ ≤ ℵ0 , so the result follows
from part (c). Part (e) is an immediate consequence of (a) and (b). Part (f) is immediate
from (e).
We can now show that, under MA(ℵ1 ), no Suslin lines exist. The next theorem proves
this directly.
Theorem 6.11 (MA(ℵ1 ) → SH). Assume MA(ℵ1 ); then there are no Suslin lines.
Proof. Assume that there is a Suslin line. Then, using Theorem 3.9 and Lemma 3.8, there
exists a normal Suslin tree (T, <). Note that (T, ≤) is, in particular, a ccc pre-ordered
set, since being Suslin doesn’t allow uncountable antichains.
For each α ∈ ω1 , let Dα := {x ∈ T | ht(x) > α}. By construction, and the fact
that (T, <) is a normal Suslin tree, the set Dα is cofinal in T for any α ∈ ω1 . Let
C := {Dα | α ∈ ω1 }. Clearly |C| = ℵ1 . Let G ⊆ T be C-generic. Then G is a branch in T
of length ω1 , which is a contradiction to the assumption that (T, <) is a Suslin tree.
Lemma 6.12. Assume MA(ℵ1 ). Let X be a ccc topological space and {Uα | α ∈ ω1 } a
family of nonempty open subsets of X. Then there is an uncountable I ⊆ ω1 such that
{Uα | α ∈ I} has the finite intersection property, meaning: if J ⊆ I is a finite subset of
I, then α∈J Uα 6= ∅.
T

Proof. Consider the sequence {Vα | α < ω1 } of nonempty open subsets of X given by
Vα = γ>α Uγ . Clearly Vβ ⊆ Vα for every α < β. We claim that there is an α0 such that
S

V β = V α0 for every β > α0 . Assume towards a contradiction that no such α0 exists.

Then we could, by recursion, find a strictly increasing sequence of ordinals hαξ | ξ < ω1 i
such that V αξ+1 6= Vnαξ for every ξ < ω1 ,o which implies that Vαξ \ V αξ+1 6= ∅ for every
ξ < ω1 . Now the set Vαξ \ V αξ+1 | ξ < ω1 is an uncountable family of nonempty subsets
of X. Moreover, the elements of the family are open, since each is the difference of an
open and a closed set, and pairwise disjoint by construction. Hence this uncountable
family of nonempty, pairwise disjoint subsets of X contradicts the fact that X satisfies
the topological ccc.
By the argument above, we can fix α0 < ω1 such that V β = V α0 for every β with
α0 < β < ω1 . Let P = {p ⊆ Vα0 | p is nonempty and open}. For pq ∈ P , define p ≤ q iff
q ⊆ p, i.e. we order P by reverse inclusion. Clearly (P, ≤) is a pre-ordered set. We claim
that (P, ≤) satisfies the ccc. Indeed, consider pq ∈ P such that p ⊥ q, i.e. such that
there is no element of P included in both p and q. This implies that p ∩ q = ∅, so any
antichain in the pre-ordered set P is also a family of pairwise disjoint nonempty open
subsets of X, so the topological ccc-ness of X implies that (P, ≤) is a ccc pre-ordered set.

25
 6 MARTIN’S AXIOM

For each β < ω1 , let Cβ := {p ∈ P | p ⊆ Uγ for some γ > β}. We claim that Cβ is
cofinal in P for each β < ω1 . Indeed, fix β < ω1 ; by the choice of α0 we have that
V α0 ⊆ V β , so if p ∈ P then p ∩ Vβ 6= ∅, since p is open. Therefore p ∩ Uγ 6= ∅ for some
γ > β, by the definition of Vβ , and we have that p ≤ p ∩ Uγ ∈ Cβ , so Cβ is cofinal in P .
Let C := {Cβ | β < ω1 }. Using MA(ℵ1 ), let G ⊆ P be a C-generic filter. Define AG :=
{γ < ω1 | ∃p ∈ G(p ⊆ Uγ )}. Then {Uγ | γ ∈ AG } has the finite intersection property, since
G has it by virtue of being a filter. It remains to argue that AG is uncountable. For any
β < ω1 we have that G ∩ Cβ 6= ∅, so AG contains some ordinal γ > β. Since β < ω1 was
arbitrary, this shows that AG is uncountable.

Theorem 6.13. Assume MA(ℵ1 ). If X and Y are topological spaces satisfying the ccc,
then X × Y with the product topology is also ccc.

Proof. Let X and Y be ccc and towards a contradiction suppose that X × Y is not
ccc. Let {Wα | α ∈ ω1 } be an uncountable family or pairwise disjoint nonempty open
subsets of X × Y . For each α ∈ ω1 , pick Uα ⊆ X and Vα ⊆ Y nonempty open subsets
such that Uα × Vα ⊆ Wα . Using Lemma 6.12 there is an uncountable I ⊆ ω1 such that
{Uα | α ∈ I} has the finite intersection property. Then given α, β ∈ I with α 6= β we
have that Uα ∩ Uβ 6= ∅ but (Uα × Vα ) ∩ (Uβ × Vβ ) = ∅, so it must be that Vα ∩ Vβ = ∅.
Hence the set {Vα | α ∈ I} contradicts the fact that Y is a ccc topological space.

Remark 6.14. In Theorem 6.11, we showed directly that MA(ℵ1 ) → SH. The above
result, i.e. Theorem 6.13, together with Lemma 2.7 gives an alternative proof of the
same fact.

26
 REFERENCES

References
Books
[1] Devlin, K. J. and Johnsbråten, H. The Souslin Problem. Lecture Notes in Mathe-
matics. Vol. 405. Springer, 1974, pp. viii+132.
[2] Halbeisen, L. J. Combinatorial Set Theory. With a Gentle Introduction to Forcing.
Second Edition. Springer Monographs in Mathematics. Springer, 2017, pp. xvi+594.
[3] Hrbacek, K. and Jech, T. J. Introduction to Set Theory. Third Edition, Revised
and Expanded. Monographs and Textbooks in Pure and Applied Mathematics. Vol.
220. CRC Press, 1999, pp. xii+291.
[4] Jech, T. J. Set Theory. The Third Millennium Edition, Revised and Expanded.
Springer Monographs in Mathematics. Springer, 2003, pp. xiv+769.
[5] Kunen, K. Set Theory. Studies in Logic. Mathematical Logic and Foundations. Vol.
34. College Publications, 2011, pp. viii+401.
[6] Kunen, K. The Foundations of Mathematics. Studies in Logic. Mathematical Logic
and Foundations. Vol. 19. College Publications, 2009, pp. viii+251.
[7] Weaver, N. Forcing for Mathematicians. World Scientific Publishing Co. Pte. Ltd.,
2014, pp. x+142.

Articles
[8] Jech, T. J. “Non-provability of Souslin’s hypothesis”. In: Commentationes Mathe-
maticae Universitatis Carolinae 8 (1967), pp. 291–305.
[9] Jensen, R. B. “Souslin’s Hypothesis is incompatible with V = L”. In: Notices of
the American Mathematical Society 15 (1968), p. 935.
[10] Kanamori, A. “Historical remarks on Suslin’s problem”. In: Set Theory, Arithmetic,
and Foundations of Mathematics: Theorems, Philosophies. Ed. by Kennedy, J. and
Kossak, R. Lecture Notes in Logic. Cambridge University Press, 2011, pp. 1–12.
[11] Kurepa, D. R. “H’hypothèse de ramification”. In: Comptes rendues hebdomadaires
des séances de l’Académie des Sciences 202 (1936), pp. 185–187.
[12] Malykhin, V. I. “The Suslin hypothesis and its significance for set-theoretic math-
ematics”. In: Russian Mathematical Surveys 51 (1996), pp. 419–437.
[13] Martin, D. A. and Solovay, R. M. “Internal Cohen extensions”. In: Annals of Math-
ematical Logic 2 (1970), pp. 143–178.
[14] Miller, E. W. “A note on Souslin’s Problem”. In: American Journal of Mathematics
65 (1943), pp. 673–678.
[15] Sierpiński, W. F. “Sur un problème de la théorie générale des ensembles équivalent
au problème de Souslin”. In: Fundamenta Mathematicae 35 (1948), pp. 165–174.
[16] Solovay, R. M. and Tennenbaum, S. “Iterated Cohen extensions and Souslin’s prob-
lem”. In: Annals of Mathematics 94 (1971), pp. 201–245.
[17] Suslin, M. Y. “Problème 3”. In: Fundamenta Mathematicae 1 (1920), p. 223.
[18] Tennenbaum, S. “Souslin’s Problem”. In: Proceedings of the National Academy of
Sciences of the United States of America 59 (1968), pp. 60–63.

27