Chapter 1

The realm of the infinite

W. Hugh Woodin

Professor of Mathematics
Department of Mathematics
University of California, Berkeley
Berkeley, CA USA

1.1 Introduction
The 20th century witnessed the development and refinement of the mathematical no-
tion of infinity. Here of course I am referring primarily to the development of Set
Theory which is that area of modern mathematics devoted to the study of infinity. This
development raises an obvious question: Is there a non-physical realm of infinity?
As is customary in modern Set Theory, V denotes the universe of sets. The purpose
of this notation is to facilitate the (mathematical) discussion of Set Theory—it does not
presuppose any meaning to the concept of the universe of sets.
The basic properties of V are specified by the ZFC axioms. These axioms allow
one to infer the existence of a rich collection of sets, a collection which is complex
enough to support all of modern mathematics (and this according to some is the only
point of the conception of the universe of sets).
I shall assume familiarity with elementary aspects of Set Theory. The ordinals cal-
ibrate V through the definition of the cumulative hierarchy of sets, [17]. The relevant
definition is given below.

Definition 1. Define for each ordinal α a set Vα by induction on α.

(1) V0 = ∅.

(2) Vα+1 = P(Vα ) = {X | X ⊆ Vα }.

1
Infinity Book—woodin 2009 Oct 04 2
n o
(3) If β is a limit ordinal then Vα = ∪ Vβ | β < α . t
u

There is a much more specific version of the question raised above concerning the
existence of a non-physical realm of infinity: Is the universe of sets a non-physical
realm? It is this latter question that I shall focus on.
There are a number of serious challenges to the claim that the answer is yes. But
where do these issues arise? More precisely for which ordinals α is the conception of
Vα meaningful?
The first point that I wish to make is that for a rather specific finite value of n,
the claim that Vn exists is a falsifiable claim and moreover that this “possibility” is
consistent with our collective (formal) experience in Mathematics to date. The details
are the subject of the next section and this account is a variation of that given in [12]. I
will continue the narative bringing in the basic arguments of [13] and [15], ultimately
defining a position on mathematical truth which is the collective conclusion of these
three papers.

1.2 The realm of the finite

By Gödel’s Second Incompleteness Theorem any system of axioms of reasonable ex-
pressive power is subject to the possibility of inconsistency, including of course the
axioms for number theory. A natural question is how profound an effect could an
inconsistency have on our view of mathematics or indeed on our view of physics.
For each finite integer n, |Vn+1 | = 2|Vn | and so even for relatively small values of n,
Vn is quite large. Is the conception of V1000 meaningful? What about the conception
of Vn where n = |V1000 |?
By a routine Gödel sentence construction I produce a formula in the language of
set theory which implicitly defines a property for finite sequences of length at most
1024 . For a given sequence this property is easily decided; if s is a sequence with
this property then s is a sequence of non-negative integers each less than 1024 and the
verification can be completed (with appropriate inputs) in significantly fewer than 1048
steps.
If there exists a sequence with this property then the conception of Vn is mean-
ingless where n = |V1000 |. I use the bound 1024 in part because the verification that
a candidate sequence of length at most 1024 has the indicated property, is arguably
physically feasible. The question of whether or not there is a sequence of length 1024
concerns only the realm of all sequences of length 1024 and so is certainly a meaningful
question for this realm. The existence of such a sequence has implications for nonexis-
tence of Vn where n = |V1000 | which is a vastly larger realm. This is entirely analogous
to the well known and often discussed situation of the number theoretic statement,
“ZFC is formally inconsistent”.
Infinity Book—woodin 2009 Oct 04 3

This statement concerns only Vω (i.e., the realm of Number Theory) and yet its truth
has implications for the nonexistence of the universe of sets, again a vastly larger
realm.
The philosophical consequences of the existence of a sequence of length 1024 as
described above, are clearly profound for it would demonstrate the necessity of the
finiteness of the universe. Clearly such a sequence does not exist. However this prop-
erty has the feature that if arbitrarily large sets do exist then there is no proof of length
less than 1024 that no such sequence of length at most 1024 can have this property. I
shall make these claims more precise.
Is the existence of such a sequence a meaningful question for our actual physical
universe? A consequence of quantum mechanics (as opposed to classical mechanics)
is that one could really build (on Earth, today) a device with a nonzero (though ridicu-
lously small) chance of finding such a sequence if such a sequence exists, which is the
other reason for the explicit bound of 1024 . So the claim that no such sequence exists
is a prediction about our world.
Now the claim that there is no such sequence is analogous to the claim that there
is no formal contradiction in Set Theory or in Set Theory together with large cardinal
axioms. I do not see any credible argument at present for the former claim other than
26
the claim that the conception of Vn is meaningful where n = |V1000 | (though in 210
years there will be such a credible argument). But then what can possibly provide the
basis for the latter claim other than some version of the belief that the conception of
the universe of sets is also meaningful?

1.2.1 Preliminaries
I shall assume familiarity with set theory at a naive level and below list informally the
axioms. I do this because I will need a variation of this system of axioms and this
variation is not a standard one.

Axiom 0 There exists a set.

Axiom 1 (Extensionality) Two sets A and B are equal if and only if they have the
same elements.

Axiom 2 (Pairing) If A and B are sets then there exists a set C = {A, B} whose only
elements are A and B.

Axiom 3 (Union) If A is a set then there exists a set C whose elements are the elements
of the elements of A.

Axiom 4 (Powerset) If A is a set then there exists a set C whose elements are the
subsets of A.
Infinity Book—woodin 2009 Oct 04 4

Axiom 5 (Regularity or Foundation) If A is a set then either A is empty (i. e. A has

no elements) or there exists an element C of A which is disjoint from A.

Axiom 6 (Comprehension) If A is a set and P(x) formalizes a property of sets then

there exists a set C whose elements are the elements of A with this property.

Axiom 7 (Axiom of Choice) If A is a set whose elements are pairwise disjoint and
each nonempty then there exists a set C which contains exactly one element
from each element of A.

Axiom 8 (Replacement) If A is a set and P(x) formalizes a property which defines

a function of sets then there exists a set C which contains as elements all the
values of this function acting on the elements of A.

Axiom 9 (Infinity) There exists a set W which is nonempty and such that for each
element A of W there exists an element B of W such that A is an element of B.

I make some remarks. Axiom 6 and Axiom 8 are really infinite lists or schemata
corresponding to the possibilities of the acceptable properties. These axioms are vague
in that it may not be clear what an acceptable property is. Intuitively these properties
are those that can be expressed using only the fundamental relationships of equality
and set membership and is made mathematically precise through the use of formal
mathematical logic.
Axioms 0-8 are (essentially) a reformulation of the axioms of number theory. It
is the Axiom of Infinity that takes one from number theory to set theory. An exact
reformulation of the number theory is given by Axioms 0-8 together the negation of
Axiom 9. Mathematical constructions specify objects in the universe of sets, this is
the informal point of view I shall adopt. For example by using a property that cannot
be true for any set, x , x, one can easily show using Axiom 0 and Axiom 4 that there
exists a set with no elements. By Axiom 1 this set is unique, it is the emptyset and is
denoted by ∅.

1.2.2 Finite Set Theory

The formal versions of the axioms on page 3 are the ZFC axioms which is a specific
(infinite) theory in the formal first order language for Set Theory–a specific list is given
in [12] for the formal language, L(=, ˆ of Set Theory. This theory is too strong for
ˆ ∈),
my purposes. The following axioms describe the universe of sets under the assumption
that for some finite ordinal α, V = Vα+1 .

Axiom 0 There exists a set.

Axiom 1 (Extensionality) Two sets A and B are equal if and only if they have the
same elements.
Infinity Book—woodin 2009 Oct 04 5

Axiom 2 (Bounding) There exists a set C such that every set is a subset of C.

Axiom 3 (Union) If A is a set then there exists a set C whose elements are the elements
of the elements of A.

Axiom (4a) (Powerset) For all sets A either there exists a set B whose elements are
all the subsets of A, or there exists a set C such that every set is a subset of C
and such that A is not an element of C.

Axiom (4b) (Powerset) For all sets A, either every set is a subset of A, or there exists
a set B such that B is an element of A and such that A does not contain all the
subsets of B, or there is a set C whose elements are all the subsets of A.

Axiom 5 (Regularity) If A is a set then either A is empty or there exists an element C

of A which is disjoint from A.

Axiom 6 (Comprehension) If A is a set and P(x) formalizes a property of sets then

there exists a set C whose elements are the elements of A with this property.

Axiom 7 (Axiom of Finiteness) If A is a nonempty set then there is an element B of

A such that for all sets C, if C is an element of A then B is not an element of C.

The two forms of the Powerset Axiom are needed to compensate for the lack of the
Pairing Axiom and the Bounding Axiom eliminates the need for the Axiom of Re-
placement. Note that the set specified by the Bounding Axiom must be unique (by
the Axiom of Extensionality). Axiom 1–Axiom 6 imply that for some ordinal α,
V = Vα+1 . By Axiom 7, this ordinal is finite and so these axioms actually do imply the
assertion, “For some finite ordinal α, V = Vα+1 ”. As a consequence one can show that
these axioms also imply the Axiom of Choice.
The formal versions of these axioms above define the theory with which I shall be
working, it is our base theory and I denote it by ZFC0 .

1.2.3 The formula

I first discuss the standard example of a Gödel sentence modified to our context (in
the language L(=, ˆ and relative to the theory ZFC0 ). This is the sentence, Ξ0 , which
ˆ ∈)
asserts that its negation, (¬Ξ0 ), can be proved from the theory ZFC0 .
By the usual arguments it follows (within our universe of sets) that the theory ZFC0
does not prove Ξ0 and ZFC0 does not prove (¬Ξ0 ); i. e. the sentence Ξ0 is independent
of the theory ZFC0 . I give the argument.
For trivial reasons, ZFC0 cannot prove Ξ0 . This is because

({∅} , ∈)  ZFC0
Infinity Book—woodin 2009 Oct 04 6

and clearly
({∅} , ∈)  (¬Ξ0 )

since ∅ is not a formal proof. Therefore I have only to show that

ZFC0 0 (¬Ξ0 ).

Assume toward a contradiction that

ZFC0 ` (¬Ξ0 ).

Then for all sufficiently large finite ordinals, n,

(Vn , ∈)  “ ZFC0 ` (¬Ξ0 ) ”,

and so for all sufficiently large finite ordinals, n, (Vn , ∈)  Ξ0 . But for all finite ordinals
n > 0, (Vn , ∈)  ZFC0 , and so for all finite ordinals n > 0,

(Vn , ∈)  (¬Ξ0 )

which is a contradiction.
The sentence Ξ0 is too pathological even for my purposes, a proof of (¬Ξ0 ) cannot
belong to a model of ZFC0 with any extent beyond the proof itself. The sentence I
seek is obtained by a simple modification of Ξ0 which yields the sentence, Ξ.
Informally, the sentence Ξ asserts that there is a proof from ZFC0 of (¬Ξ) of length
less than 1024 and further that Vn exists where n = |V1000 |. As I have already indicated,
the choice of 1024 is only for practical reasons. There is no corresponding reason for
my particular choice of n, one could quite easily modify the definition by requiring
that the choice of n be larger.
The formal specification of Ξ is a completely standard (though tedious) exercise
using the modern theory of formal mathematical logic; this involves the formal notion
of proof defined so that proofs are finite sequences of natural numbers, etc. [12].
In our universe of sets (¬Ξ) is true and so there is a proof of (¬Ξ) from ZFC0 . It
is not clear just how short such a proof can be. This is a very interesting question.
The witness for Armageddon (though with the end of time comfortably distant in the
future) is a proof of (¬Ξ) from ZFC0 of length less than 1024 .
It is important to emphasize that while ZFC0 is a very weak theory, in attempting
to prove (¬Ξ) from ZFC0 , one is free to augment ZFC0 with the axiom that Vn exists
where n = |V1000 |. This theory is not weak, particularly as far as the structure of binary
10
sequences of length 1024 or even of length 1010 is concerned.
The sum total of human experience in mathematics to date (i. e., the number of
manuscript pages written to date) is certainly less than 1012 pages. The shortest proof
from ZFC0 that no such sequence exists must have length greater than 1024 . This
is arguably beyond the reach of our current experience but there is an important issue
which concerns the compression achieved by the informal style in which mathematical
arguments are actually written. This is explored a little bit further in [12].
Infinity Book—woodin 2009 Oct 04 7

With proper inputs and global determination one could verify with current technol-
ogy that a given sequence of length at most 1024 is a proof of (¬Ξ) from ZFC0 . But,
obviously we do not expect to be able to find a sequence of length less than 1024 which
is a proof of (¬Ξ) from ZFC0 . This actually gives a prediction about the physical uni-
verse since one can code any candidate for such a sequence by a binary sequence of
length at most 1026 . The point is that assuming the validity of the quantum view of
the world, it is possible to build an actual physical device which must have a nonzero
chance of finding such a sequence if such a sequence can exist. The device simply
contains (a suitably large number of independent) modules each of which performs an
independent series of measurements which in effect flips a quantum coin. The point
of course is that by quantum law any outcome is possible. The prediction is simply
that any such device must fail to find a sequence of length less than 1024 which is a
proof of (¬Ξ) from ZFC0 . One may object that the belief that any binary sequence
of length 1026 is really a possible outcome of such a device, requires an extraordinary
faith in quantum law. But any attempt to build a quantum computer which is useful
(for factoring) requires the analogous claim where 1026 is replaced by numbers at least
as large as 105 .
This of course requires something like quantum theory. In the universe as de-
scribed by Newtonian laws, the argument described above does not apply since truly
random processes would not exist. One could imagine proving that for a large class of
chaotic (but deterministic) processes (“mechanical coin flippers”), no binary sequence
of length 1024 which actually codes a formal proof, can possibly be generated. In other
words, for the non-quantum world, the prediction that no such sequence (as above) can
be generated may not require that the conception of Vn is meaningful where n = |V1000 |.
Granting quantum law, and based only on our collective experience in Mathematics
to date, how can one account for the prediction (that one cannot find a sequence of
length less than 1024 which is a proof of (¬Ξ) from ZFC0 ) unless one believes that the
conception of Vn is meaningful where n = |V1000 |?
Arguably (given current physical theory) this is already a conception of a nonphys-
ical realm.

1.3 Beyond the finite realm

In this section we briefly summarize the basic argument of [15] though here our use of
this argument is for a different purpose.

Skeptic’s Attack: The mathematical conception of infinity is meaning-

less and without consequence because the entire conception of the uni-
verse of sets is a complete fiction. Further, all the theorems of Set Theory
Infinity Book—woodin 2009 Oct 04 8

are merely finitistic truths, a reflection of the mathematician and not of

any genuine mathematical “reality”.

Throughout this section, the “Skeptic” simply refers to the meta-mathematical position
which denies any genuine meaning to a conception of uncountable sets. The counter-
view is that of the “Set Theorist”.

The Set Theorist’s Response: The development of Set Theory, after

Cohen, has led to the realization there is a robust hierarchy of strong ax-
ioms of infinity.

Elaborating further, it has been discovered that in many cases, very different lines
of investigation have led to problems whose degree of unsolvability is exactly cali-
brated by a notion of infinity. Thus the hierarchy of large cardinal axioms emerges
an intrinsic, fundamental, conception within Set Theory. To illustrate this I discuss an
example from modern Set Theory which concerns infinite games.
Suppose A ⊂ P(N) where P(N) denotes the set of all sets σ ⊆ N and N is the set
of all natural numbers; N = {1, 2, . . . , k, . . .}.
Associated to the set A is an infinite game involving two players, Player I and
Player II. The players alternate declaring at stage k whether k ∈ σ or k < σ:

Stage 1: Player I declares 1 ∈ σ or declares 1 < σ;

Stage 2: Player II declares 2 ∈ σ or declares 2 < σ;

Stage 3: Player I declares 3 ∈ σ or declares 3 < σ; . . .

After infinitely many stages a set σ ⊆ N is specified. Player I wins this run of the game
if σ ∈ A; otherwise Player II wins. (Note: Player I has control of which odd numbers
are in σ, and Player II has control of which even numbers are in σ.).
A strategy is simply a function which provides moves for the players given just the
current state of the game. More formally a strategy is a function

τ : [N]<ω × N → {0, 1}

where [N]<ω denotes the set of all finite subsets of N. At each stage k of the game the
relevant player can choose to follow τ by declaring “k ∈ σ” if

τ(a, k) = 1

and declaring “k < σ” if τ(a, k) = 0, where

a = i < k | “i ∈ σ” was declared at stage i” .



The strategy τ is a winning strategy for Player I if by following the strategy at each
stage k where it is Player I’s turn to play (i.e., for all odd k), Player I wins the game
no matter how Player II plays. Similarly τ is a winning strategy for Player II if by
Infinity Book—woodin 2009 Oct 04 9

following the strategy at each stage k where it is Player II’s turn to play (i.e., for all
even k), Player II wins the game no matter how Player I plays.
The game is determined if there is a winning strategy for one of the players. Clearly
it is impossible for there to be winning strategies for both players.
It is easy to specify sets A ⊆ P(N) for which the corresponding game is determined,
however, the problem of specifying a set A ⊆ P(N) for which the corresponding game
is not determined, turns out to be quite a bit more difficult. The Axiom of Determinacy,
AD, is the axiom which asserts that for all sets A ⊆ P(N), the game given by A,
as described above, is determined. This axiom was first proposed by Mycielski and
Steinhaus, [9], and contradicts the Axiom of Choice, so the problem here is whether
the Axiom of Choice is necessary to construct a set
A ⊆ P(N)
for which the corresponding game is not determined. Clearly if the Axiom of Choice
is necessary then the existence of such set A is quite a subtle fact.
The unsolvability of this problem is exactly calibrated by large cardinal axioms.
The relevant large cardinal notion is that of a Woodin cardinal which I shall not define,
[5]. The ZF axioms are the ZFC axioms but without the Axiom of Choice. The issue
of whether the Axiom of Choice is needed to construct a counterexample to AD is
exactly the question of whether the theory, ZF + AD, is formally consistent.

Theorem 2. The two theories,

(1) ZF + AD

(2) ZFC + “There exist infinitely many Woodin cardinals”

are equiconsistent. t
u

1.3.1 A prediction and a challenge for the Skeptic

Is the theory, ZF + AD, really formally consistent? The claim that it is consistent is a
prediction which can be refuted by finite evidence (a formal contradiction). Taking an
admittedly extreme position, I claim in [15] the following.

It is only through the calibration by a large cardinal axiom in conjunction

with our understanding of the hierarchy of such axioms as true axioms
about the universe of sets, that this prediction; the formal theory ZF+AD
is consistent, is justified.

As a consequence of my belief in this claim, I also made a prediction:

In the next 10,000 years there will be no discovery of an inconsistency in

this theory.
Infinity Book—woodin 2009 Oct 04 10

This is a specific and unambiguous prediction about the physical universe just as is
the case for the analogous prediction in the previous section. Further it is a predic-
tion which does not arise by a reduction to a previously held truth (as for example is
the case for the prediction that no counterexample to Fermat’s Last Theorem will be
discovered). This is a genuinely new prediction which I make in [15] based on the de-
velopment of Set Theory over the last 50 years and on my belief that the conception of
the transfinite universe of sets is meaningful. I make this prediction independently of
all speculation of what computational devices might be developed in the next 10,000
years (or whatever new sources of knowledge might be discovered) which increase the
effectiveness of research in Mathematics.
Now the Skeptic might object that this prediction is not interesting or natural be-
cause the formal theories are not interesting or natural. But such objections are not
allowed in Physics, the ultimate physical theory should explain all (physical) aspects
of the physical universe, not just those which we regard as natural. How can we apply
a lessor standard for the ultimate mathematical theory? Of course, I also predict:

There will be no discovery ever of an inconsistency in this theory;

and this prediction, if true, is arguably a physical law.

Skeptic’s Retreat: OK, I accept the challenge noting that I only have to
explain the predictions of formal consistency given by the large cardinal
axioms. The formal theory of Set Theory as given by the axioms, ZFC, is
so “incomplete” that: Any large cardinal axiom, in the natural formulation
of such axioms, is either consistent with the axioms of Set Theory, or there
is an elementary proof that the axiom cannot hold.

To examine the Skeptic’s Retreat and to assess how this too might be refuted I need to
briefly survey the basic template for large cardinal axioms in Set Theory.

1.3.2 Large cardinal axioms within Set Theory

A set N is transitive if every element of N is a subset of N. Transitive sets are frag-
ments of V which are analogous to initial segments. For each ordinal α the set Vα is a
transitive set.
The simplest (proper) class is the class of all ordinals. This class is a transitive
class where a class M ⊆ V is defined to be a transitive class if every element of M is a
subset of M. The basic template for large cardinal axioms is as follows.

There is a transitive class M and an elementary embedding

j:V→M
which is not the identity.
Infinity Book—woodin 2009 Oct 04 11

With the exception of the definition of a Reinhardt cardinal which I shall come to
below, one can always assume that the classes, M and j, are classes which are logically
definable from parameters by formulas of a fixed bounded level of complexity (Σ2 -
formulas). Moreover the assertion that j is an elementary embedding—that is the
assertion:

• For all formulas φ(x) and for all sets a,

V  φ[a]

if and only if M  φ[ j(a)];

—is equivalent to the assertion:

• For all formulas φ(x), for all ordinals α, and for all sets a ∈ Vα ,

Vα  φ[a]

if and only if j(Vα )  φ[ j(a)].

Therefore this template makes no essential use of the notion of a class. It is simply for
convenience that I refer to classes (and this is the usual practice in Set Theory).
Suppose that M is a transitive class and that

j:V→M

is an elementary embedding which not the identity. Suppose that j(α) = α for all
ordinals α. Then one can show by transfinite induction that for all ordinals α, the
embedding, j, is the identity on Vα . Therefore since j is not the identity, there must
exist an ordinal α such that j(α) , α. The least such ordinal is the critical point of j.
This must be a cardinal. The critical point of j is the large cardinal and the existence
of the transitive class M and the elementary embedding j are the witnesses for this.
A cardinal κ is a measurable cardinal if there exists a transitive class M and an
elementary embedding,
j:V→M

such that κ is the critical point of j.

It is by requiring M to be closer to V that one can define large cardinal axioms far
beyond the axiom, “There is a measurable cardinal”. In general the closer one requires
M to be to V, the stronger the large cardinal axiom. The natural maximum axiom was
proposed (M = V) by Reinhardt in his Ph.D thesis, see [10]. The associated large
cardinal axiom is that of a Reinhardt cardinal.

Definition 3. A cardinal κ is a Reinhardt cardinal if there is an elementary embedding,

j:V→V

such that κ is the critical point of j. t

u
Infinity Book—woodin 2009 Oct 04 12

The definition of a Reinhardt cardinal makes essential use of classes, but the fol-
lowing variation does not and this variation (which is not a standard notion) is only
formulated in order to facilitate this discussion. The definition requires a logical no-
tion. Suppose that α and β are ordinals such that α < β. Then

Vα ≺ Vβ

if for all formulas, φ(x), for all a ∈ Vα ,

Vα  φ[a]

if and only if Vβ  φ[a]. Thus Vα ≺ Vβ if and only if

I : Vα → Vβ

is an elementary embedding where I is the identity map.

Definition 4. A cardinal κ is a weak Reinhardt cardinal if there exist γ > λ > κ such
that

(1) Vκ ≺ Vλ ≺ Vγ ,

(2) there exists an elementary embedding,

j : Vλ+2 → Vλ+2

such that κ is the critical point of j. t

u

The definition of a weak Reinhardt cardinal only involves sets. The relationship
between Reinhardt cardinals and weak Reinhardt cardinals is unclear but one would
naturally conjecture that at least in terms of consistency strength, Reinhardt cardinals
are stronger than weak Reinhardt cardinals and hence my choice in terminology. The
following theorem is an immediate corollary of the fundamental inconsistency results
of Kunen, [6].

Theorem 5 (Kunen). There are no weak Reinhardt cardinals. t

u

The proof is elementary so this does not refute the Skeptic’s Retreat. But Kunen’s
proof makes essential use of the Axiom of Choice. The problem is open without this
assumption. Further there is really no known interesting example of a strengthening
of the definition of a weak Reinhardt cardinal that yields a large cardinal axiom which
can be refuted without using the Axiom of Choice. The difficulty is that without the
Axiom of Choice it is extraordinarily difficult to prove anything about sets.
Kunen’s proof leaves open the possibility that the following large cardinal ax-
iom might be consistent with the Axiom of Choice. This therefore is essentially the
strongest large cardinal axiom not known to be refuted by the Axiom of Choice, see
[5] more on this as well as for the actual statement of Kunen’s theorem.
Infinity Book—woodin 2009 Oct 04 13

Definition 6. A cardinal κ is a strongly (ω + 1)-huge cardinal if there exist γ > λ > κ

such that

(1) Vκ ≺ Vλ ≺ Vγ ,

(2) there exists an elementary embedding,

j : Vλ+1 → Vλ+1
such that κ is the critical point of j. t
u

The issue of whether the existence of a weak Reinhardt cardinal is consistent with
the axioms, ZF, is an important issue for the Set Theorist because by the results of
[16], the theory
ZF + “There is a weak Reinhardt cardinal”
proves the formal consistency of the theory
ZFC + “There is a proper class of strongly (ω + 1)-huge cardinals”.
This number theoretic statement is a theorem of Number Theory. But as indicated
above, the notion of a strongly (ω + 1)-huge cardinal is essentially the strongest large
cardinal notion which is not known to be refuted by the Axiom of Choice.
Therefore the number theoretic assertion that the theory
ZF + “There is a weak Reinhardt cardinal”
is consistent is a stronger assertion than the number theoretic assertion that the theory
ZFC + “There is a proper class of strongly (ω + 1)-huge cardinals”
is consistent. More precisely, the former assertion implies, but is not implied by, the
latter assertion; unless of course the theory
ZFC + “There is a proper class of strongly (ω + 1)-huge cardinals”
is formally inconsistent. This raises an interesting question.

How could the Set Theorist ever be able to argue for the prediction that
the existence of weak Reinhardt cardinals is consistent with axioms of Set
Theory without the Axiom of Choice?

Moreover this one prediction implies all the predictions (of formal consistency) the Set
Theorist can currently make based on the entire large cardinal hierarchy as presently
conceived (in the context of a universe of sets which satisfies the Axiom of Choice).
My point is that by appealing to the Skeptic’s Retreat, one could reasonably claim that
the theory
ZF + “There is a weak Reinhardt cardinal”,
is formally consistent and in making this single claim one would subsume all the
claims of consistency that the Set Theorist can make based on our current understand-
ing of the universe of sets (without abandoning the Axiom of Choice).
Infinity Book—woodin 2009 Oct 04 14

Before presenting a potential option to deal with this, I describe an analogous op-
tion of how the Set Theorist can claim that the theory

ZF + AD

is consistent even though as I have indicated AD also refutes the Axiom of Choice. The
explanation requires some definitions which I shall require anyway. Gödel defined a
very special transitive class L ⊆ V and showed that all the axioms of ZFC hold when
interpreted in L. The definition of L does not require the Axiom of Choice and so one
obtains the seminal result that if the axioms, ZF, are consistent then so are the axioms
ZFC. Gödel also proved that the Continuum Hypothesis holds in L thereby showing
that one cannot formally refute the Continum Hypothesis from the axioms ZFC (unless
of course these axioms are inconsistent).
The definition of L is simply given by replacing the operation P(X) in the definition
of Vα+1 by the operation PDef (X) which associates to the set X the set of all subsets
Y ⊆ X such that Y is logically definable in the structure, (X, ∈), from parameters in X.
For any infinite set X, PDef (X) ⊂ P(X) and PDef (X) , P(X).
Thus one defines Lα by induction on the ordinal α; setting L0 = ∅, setting

Lα+1 = PDef (Lα ),

and taking unions at limit stages. The class L is defined as the class of all sets a such
that a ∈ Lα for some ordinal α. It is perhaps important to note that while there must
exist a proper class of ordinals α such that

Lα = L ∩ Vα ,

this is not true for all ordinals α.

Relativizing the definition of L to Vω+1 we obtains the class L(Vω+1 ) which is more
customarily denoted by L(R); here one defines,

L0 (R) = Vω+1

and proceeds by induction exactly as above to define Lα (R) for all ordinals α. The
class L(R) is the class of all sets a such that a ∈ Lα (R) for some ordinal α.
Unlike the case for L, one cannot prove that the Axiom of Choice holds in L(R),
though one can show that all of the other axioms of ZFC hold in L(R). The following
theorem which is related to Theorem 2, not only establishes the consistency of ZF+AD
from simply the existence of large cardinals, it also establishes that L(R)  AD (as a
new truth about sets)–see [5] for more on the history of this theorem and the attempts
to establish that L(R)  AD from large cardinal axioms.

Theorem 7 (Martin, Steel, Woodin). Suppose there is a proper class of Woodin car-
dinals. Then
L(R)  AD. t
u
Infinity Book—woodin 2009 Oct 04 15

For the reasons I have indicated one cannot hope to argue for the consistency of
the theory,
ZF + “There is a weak Reinhardt cardinal”,

on the basis of any large cardinal axiom not known to refute the Axiom of Choice. The
experience with the theory, ZF + AD, suggests that as an alternative, one should seek
both a generalization of L(R) and some structural principles for this fragment such
that the axiom that asserts both the existence of this fragment and that the structural
principles hold in this fragment, implies the formal consistency of the axiom which
asserts the existence of a weak Reinhardt cardinal, or even better that implies that the
latter axiom actually holds in this fragment.
In fact there are compelling candidates for generalizations of L(R) and axioms for
these fragments generalizing AD. But at present there is simply no plausible candidate
for such a generalization of L(R) in which the axiom that there is a weak Reinhardt
cardinal can even hold; nor is there a plausible candidate for a fragment together with
structural principles for that fragment which would imply the formal consistency of
the existence of a weak Reinhardt cardinal. This is explored more fully in [16].
There is another potential option which is suggested by a remarkable theorem of
Vopenka. But to explain this further I must give another definition which I shall also
require in the subsequent discussion. This is the definition of the class, HOD, which
originates in remarks of Gödel at the Princeton Bicentennial Conference in December,
1946. The first detailed reference appears to be [7] (see the review of [7] by G. Kreisel).

Definition 8 (ZF). (1) For each ordinal α, HODα is the set of all sets a such that
there exists a transitive set M ⊂ Vα such that a ∈ M and such for all b ∈ M, b is
definable in Vα from ordinal parameters.

(2) HOD is the class of all sets a such that a ∈ HODα for some α. t
u

I caution that just as is the case for Lα , in general

HODα , HOD ∩ Vα ,
though for a proper class of ordinals α it is true that HODα = HOD ∩ Vα .
The class HOD is quite interesting for a number of reasons, one of which is illus-
trated by the following observation of Gödel which as indicated is stated within just
the theory ZF, in other words without assuming the Axiom of Choice.

Theorem 9 (ZF). HOD  ZFC. t

u

This theorem gives a completely different approach to showing that if the theory, ZF,
is formally consistent then so is the theory, ZFC.
One difficulty with HOD is that the definition of HOD is not absolute, for example
in general HOD is not even the same as defined within HOD. As a consequence almost
Infinity Book—woodin 2009 Oct 04 16

any set theoretic question one might naturally ask about HOD, is formally unsolvable.
Two immediate such questions are whether V = HOD and more simply, whether HOD
contains all the real numbers. Both of these questions are formally unsolvable but are
of evident importance because they specifically address the complexity of the Axiom
of Choice. If V = HOD then there is no mystery as to why the Axiom of Choice holds
but of course one is left with the problem of explaining why V = HOD.
I end this section with the remarkable theorem of Vopenka alluded to above. The
statement involves Cohen’s method of forcing adapted to produce extensions in which
the Axiom of Choice can fail, these are called symmetric generic extensions. For each
ordinal α, there is a minimum extension of the class HOD which contains both HOD
and Vα and in which the axioms ZF hold. This minimum extension is denoted by
HOD(Vα ).

Theorem 10 (ZF; Vopenka). For all ordinals α, HOD(Vα ) is a symmetric generic ex-
tension of HOD. t
u

The alternative conception of truth for Set Theory which is suggested by this the-
orem and which could provide a basis for the claim that weak Reinhardt cardinals are
consistent is the subject of the next section.

1.4 The generic-multiverse of sets

The challenge presented in the previous section—the challenge to account for the pre-
diction that the existence of a weak Reinhardt cardinal is formally consistent with ZF
axioms—suggests that one should consider a multiverse conception of the universe of
sets. The point of course is that while the existence of a weak Reinhardt cardinal is not
possible, granting the Axiom of Choice, this does not rule out that there may be (sym-
metric) generic extensions of V in which there are weak Reinhardt cardinals. Such a
multiverse approach to the conception of the universe of sets would also mitigate the
difficulties associated to the formal unsolvability of fundamental problems such as that
of the Continum Hypothesis, and this latter feature is the primary motivation for such
an approach. This section is based on [13].
Let the multiverse (of sets) refer to the collection of possible universes of sets.
The truths of the Set Theory according to the multiverse conception of truth are the
sentences which hold in each universe of the multiverse. Cohen’s method of forcing
which is the fundamental technique for constructing non-trivial extensions of a given
(countable) model of ZFC suggests a natural candidate for a multiverse; the generic-
multiverse is generated from each universe of the collection by closing under generic
extensions (enlargements) and under generic refinements (inner models of a universe
which the given universe is a generic extension of). To illustrate the concept of the
Infinity Book—woodin 2009 Oct 04 17

generic-multiverse, suppose that M is a countable transitive set with the property that

M  ZFC.

Let V M be the smallest set of countable transitive sets such that M ∈ V M and such that
for all pairs, (M1 , M2 ), of countable transitive sets such that

M1  ZFC,

and such that M2 is a generic extension of M1 , if either M1 ∈ V M or M2 ∈ V M then

both M1 and M2 are in V M . It is easily verified that for each N ∈ V M ,

VN = V M ,

where VN is defined using N in place of M. V M is the generic-multiverse generated in

V from M.
The generic-multiverse conception of truth is the position that a sentence is true if
and only if it holds in each universe of the generic-multiverse generated by V. This
can be formalized within V in the sense that for each sentence φ there is a sentence
φ∗ , recursively depending on φ, such that φ is true in each universe of the generic-
multiverse generated by V if and only if φ∗ is true in V. The sentence φ∗ is explicit
given φ and does not depend on V. For exmple, given any countable transitive set, M,
such that M  ZFC,
M  φ∗

if and only if N  φ for all N ∈ V M (the proof is given in [14]). This is an important
point in favor of the generic-multiverse position since it shows that as far as assessing
truth is concerned, the generic-multiverse position is not that sensitive to the meta-
universe in which the generic-multiverse is being defined.
Is the generic-multiverse position a reasonable one? The refinements of Cohen’s
method of forcing in the decades since his initial discovery of the method and the
resulting plethora of problems shown to be unsolvable, have in a practical sense almost
compelled one to adopt the generic-multiverse position. This has been reinforced by
some rather unexpected consequences of large cardinal axioms which I shall discuss
later in this section.
The purpose of this section is not to argue against any possible multiverse posi-
tion but to more carefully examine the generic-multiverse position within the context
of modern Set Theory. In brief I shall argue that modulo the Ω Conjecture (which I
shall define in the next section), the generic-multiverse position outlined above is not
plausible. The essence of the argument against the generic-multiverse position is that
assuming the Ω Conjecture is true (and that there is a proper class of Woodin cardi-
nals) then this position is simply a brand of formalism that denies the transfinite by
a reducing truth about the universe of sets to truth about a simple fragment such as
the integers or, in this case, the sets of real numbers. The Ω Conjecture is invariant
Infinity Book—woodin 2009 Oct 04 18

between V and any generic extension of V and so the generic-multiverse position must
either declare the Ω Conjecture to be true or declare the Ω Conjecture to be false.
It is a fairly common (informal) claim that the quest for truth about the universe
of sets is analogous to the quest for truth about the physical universe. However I am
claiming an important distinction. While physicists would rejoice in the discovery
that the conception of the physical universe reduces to the conception of some simple
fragment or model, the set theorist rejects this possibility. I claim that by the very
nature of its conception, the set of all truths of the transfinite universe (the universe of
sets) cannot be reduced to the set of truths of some explicit fragment of the universe
of sets. Taking into account the iterative conception of sets, the set of all truths of an
explicit fragment of the universe of sets cannot be reduced to the truths of an explicit
simpler fragment. The latter is the basic position on which I shall base my arguments.
An assertion is Π2 if it is of the form,
“ For every infinite ordinal α , Vα  φ ”,
for some sentence, φ. A Π2 assertion is a multiverse truth if the Π2 assertion holds in
each universe of the multiverse. A key point:

Remark 11. Arguably, the generic-multiverse view of truth is only viable for Π2 -
sentences and not in general even for Σ2 -sentences (these are sentences expressible
as the negation of a Π2 -sentence). This is because of the restriction to set forcing in
the definition of the generic-multiverse. Therefore one can quite reasonably question
whether the generic-multiverse view can possibly account for the predictions of con-
sistency given by large cardinal axioms. At present there is no reasonable candidate
for the definition of an expanded version of the generic-multiverse which allows class
forcing extensions and yet which preserves the existence of large cardinals across the
multiverse. t
u

In the context where there is a Woodin cardinal, let us use “δ0 ” to denote the least
Woodin cardinal. So I am just fixing a notation just as “ω1 ” is fixed as the notation for
the least uncountable ordinal. Both ω1 and δ0 can change in passing from one universe
of sets to an extension of that universe.
The assertion,
“δ is a Woodin cardinal”
is equivalent to the assertion,
Vδ+1  “δ is a Woodin cardinal”
and so δ = δ0 if and only if
Vδ+1  “δ = δ0 ”.
Therefore assuming there is a Woodin cardinal, for each sentence φ, it is a Π2 assertion
to say that
Vδ0 +1  φ
Infinity Book—woodin 2009 Oct 04 19

and it is a Π2 assertion to say that Vδ0 +1 6 φ. Thus in any one universe of the multiverse,
the set of all sentences φ such that Vδ0 +1  φ—that is, the theory of Vδ0 +1 as computed
in that in that universe—is recursive in the set of Π2 sentences (assertions) which hold
in that universe. Further by Tarski’s Theorem on the undefinability of truth the latter
set cannot be recursive in the former set.
These comments suggest the following multiverse laws which I state in reference
to an arbitrary multiverse position (though assuming that the existence of a Woodin
cardinal holds throughout the multiverse).

First Multiverse Law

The set of Π2 assertions which are multiverse truths is not recursive in the
set of multiverse truths of Vδ0 +1 . t
u

The motivation for this multiverse law is that if the set of Π2 multiverse truths is
recursive in the set of multiverse truths of Vδ0 +1 then as far as evaluating Π2 assertions is
concerned, the multiverse is equivalent to the reduced multiverse of just the fragments
Vδ0 +1 of the universes of the multiverse. This amounts to a rejection of the transfinite
beyond Vδ0 +1 and constitutes in effect the unacceptable brand of formalism alluded to
earlier. This claim would be reinforced should the multiverse position also violate a
second multiverse law which I now formulate.
A set Y ⊂ Vω is definable in Vδ0 +1 across the multiverse if the set Y is definable
in the structure Vδ0 +1 of each universe of the multiverse (possibly by formulas which
depend on the parent universe). The second multiverse law is a variation of the First
Multiverse Law.

Second Multiverse Law

The set of Π2 assertions which are multiverse truths, is not definable in

Vδ0 +1 across the multiverse. t
u

Again, by Tarski’s Theorem on the undefinability of truth, this multiverse law is

obviously a reasonable one if one regards the only possibility for the multiverse to be
the universe of sets so that set of multiverse truths of Vδ0 +1 is simply the set of all
sentences which are true in Vδ0 +1 and the set of Π2 assertions which are multiverse
truths is simply the set of Π2 assertions which are true in V. Likewise the Second
Multiverse Law would have to hold if one modified the law to simply require that the
set of Π2 assertions which are multiverse truths, is not uniformly definable in Vδ0 +1
across the multiverse (i.e., by a single formula).
Assuming both that Ω Conjecture and the existence of a proper class of Woodin
cardinals hold in each (or one) universe of the generic-multiverse generated by V,
then both the First Multiverse Law and the Second Multiverse Law are violated by
Infinity Book—woodin 2009 Oct 04 20

the generic-multiverse position. This is the basis for the argument I am giving against
the generic-multiverse position in this paper. In fact the technical details of how the
generic-multiverse position violates these multiverse laws provides an even more com-
pelling argument against the generic-multiverse position since the analysis shows that
in addition the generic-multiverse position is truly a form of formalism because of the
connections to Ω-logic. The argument also shows that the violation of the First Mul-
tiverse Law is explicit; i.e. assuming the Ω Conjecture, there is an explicit recursive
reduction of the set of Π2 assertions which are generic-multiverse truths is to the set of
generic-multiverse truths of Vδ0 +1 .
There is a special case which I can present without any additional definitions and
which is not contingent on any conjectures.

Theorem 12. Suppose that M is a countable transitive set

M  ZFC + “There is a proper class of Woodin cardinals”

and that M ∩ Ord is as small as possible. Then V M violates both multiverse laws. t
u

1.4.1 Ω-logic
The generic-multiverse conception of truth declares the Continum Hypothesis to be
neither true nor false and declares, granting large cardinals, that assertion,

L(R)  AD,

to be true (see Theorem 7). I note that for essentially all current large cardinal axioms,
the existence of a proper class of large cardinals holds in V if and only if it holds in V B
for all complete Boolean algebras, B. In other words, in the generic-multiverse position
the existence of a proper class of, say, Woodin cardinals is either true or false since it
either holds in every universe of the generic-multiverse or it holds in no universe of the
generic-multiverse, [4].
I am going to analyze the generic-multiverse position from the perspective of Ω-
logic which I first briefly review. I will use the standard modern notation for Co-
hen’s method of forcing; potential extensions of the universe, V, are given by complete
Boolean algebras B, V B denotes the corresponding boolean valued extension and for
each ordinal α, VαB denotes Vα as defined in that extension.

Definition 13. Suppose that T is a countable theory in the language of Set Theory, and
φ is a sentence. Then
T Ω φ

if for all complete Boolean algebras, B, for all ordinals, α, if VαB  T then VαB  φ. t
u
Infinity Book—woodin 2009 Oct 04 21

If there is a proper class of Woodin cardinals then the relation T Ω φ, is generically

absolute. This fact which arguably was a completely unanticipated consequence of
large cardinals, makes Ω-logic interesting from a meta-mathematical point of view.
For example the set
VΩ = {φ | ∅ Ω φ}

is generically absolute in the sense that for a given sentence, φ, the question whether
or not φ is logically Ω-valid; i.e., whether or not φ ∈ VΩ , is absolute between V and all
of its generic extensions. In particular the method of forcing cannot be used to show
the formal independence of assertions of the form ∅ Ω φ.

Theorem 14. Suppose that T is a countable theory in the language of Set Theory, φ
is a sentence and that there exists a proper class of Woodin cardinals. Then for all
complete Boolean algebras, B, V B  “T Ω φ” if and only if T Ω φ. t
u

There are a variety of technical theorems which show that one cannot hope to
prove the generic invariance of Ω-logic from any large cardinal hypothesis weaker
than the existence of a proper class of Woodin cardinals—for example if V = L then
definition of VΩ is not absolute between V and V B , for any non-atomic complete
Boolean algebra, B.
It follows easily from the definition of Ω-logic, that for any Π2 -sentence, φ,

∅ Ω φ

if and only if for all complete Boolean algebras, B,

V B  φ.

Therefore by the theorem above, assuming there is a proper class of Woodin cardi-
nals, for each sentence, ψ, the assertion

For all complete Boolean algebras, B, V B  “Vδ0 +1  ψ”

is itself absolute between V and V B for all complete Boolean algebras B. This re-
markable consequence of the existence of a proper class of Woodin cardinals actually
seems to be evidence for the generic-multiverse position. In particular this shows that
the generic-multiverse position, at least for assessing Π2 assertions, and so for assess-
ing all assertions of the form,
Vδ0 +1  φ,

is equivalent to the position that a Π2 assertion is true if and only if it holds in V B for
all complete Boolean algebras B. Notice that if R 1 L and if V is a generic extension
of L then this equivalence is false. In this situation the Π2 sentence which expresses
R 1 L holds in V B for all complete Boolean algebras, B, but this sentence fails to hold
across the generic-multiverse generated by V (since L belongs to this multiverse).
Infinity Book—woodin 2009 Oct 04 22

To summarize, suppose that there exists a proper class of Woodin cardinals in

each universe of the generic-multiverse (or equivalently that there is a proper class of
Woodin cardinals in at least one universe of the generic-multiverse). Then for each Π2
sentence φ; the following are equivalent:

(1) φ is holds across the generic-multiverse;

(2) “∅ Ω φ” holds across the generic-multiverse;

(3) “∅ Ω φ” holds in at least one universe of the generic-multiverse.

For any Σ2 -sentence φ, the assertion, “∅ 6 Ω (¬φ)”, is by the definitions equivalent to

the assertion that for some complete Boolean algebra, B, V B  φ. Therefore assuming
that there exists a proper class of Woodin cardinals, for any Σ2 -sentence, if in one
universe of the generic-multiverse generated by V, the sentence φ is true, then in every
universe of the generic-multiverse generated by V, the sentence φ can be forced to
be true by passing to a generic extension of that universe. This same remarkable fact
applies to symmetric forcing extensions as well.
Therefore assuming there is a proper class of Woodin cardinals, it seems that the
generic-multiverse view of truth can account for the prediction that weak Reinhardt
cardinals are consistent with ZF; the relevant Σ2 -sentence is the sentence which asserts
that there exists a complete Boolean algebra B and there exists a term τ ∈ V B for a
transitive set such that (with Boolean value 1),
V(τ)  ZF + “There is a weak Reinhardt cardinal”.
This sentence if true in one universe of the generic-multiverse generated by V must
be true in every universe of the generic-multiverse generated by V. Further by Theo-
rem 10, if there is a weak Reinhardt cardinal then in the generic-multiverse generated
by HOD this Σ2 -sentence is actually declared as true.
But on close inspection one realizes this is not really a justification at all. The
sentence above is meaningful in the generic-multiverse view of truth but there is no
explanation of why it is true. This is exactly as is the case for the sentence which
asserts that the formal theory,
ZF + “There is a weak Reinhardt cardinal.”,
is consistent.
To more fully evaluate the generic-multiverse position one must understand the
logical relation, T Ω φ. In particular a natural question arises: is there a corresponding
proof relation?

1.4.2 The Ω Conjecture

I define the proof relation, T `Ω φ. This requires a preliminary notion that a set of
reals be universally Baire, [1]. In fact I shall define T `Ω φ, assuming the existence of
Infinity Book—woodin 2009 Oct 04 23

a proper class of Woodin cardinals and exploiting the fact that there are a number of
(equivalent) definitions. Without the assumption that there is a proper class of Woodin
cardinals, the definition is a bit more technical, [13]. Recall that if S is a compact
Hausdorff space then a set X ⊆ S has the property of Baire in the space S if there
exists an open set O ⊆ S such that symmetric difference,
X4O,
is meager in S (contained in a countable union of closed sets with empty interior).

Definition 15. A set A ⊂ R is universally Baire if for all compact Hausdorff spaces,
S , and for all continuous functions,
F : S → R,
the preimage of A by F has the property of Baire in the space S . t
u

Suppose that A ⊆ R is universally Baire. Suppose that M is a countable transitive

model of ZFC. Then M is strongly A-closed if for all countable transitive sets N such
that N is a generic extension of M,
A ∩ N ∈ N.

Definition 16. Suppose there is a proper class of Woodin cardinals. Suppose that T is
a countable theory in the language of Set Theory, and φ is a sentence. Then T `Ω φ if
there exists a set A ⊂ R such that:

(1) A is universally Baire,

(2) for all countable transitive models, M, if M is strongly A-closed and T ∈ M,

then
M  “T Ω φ”. t
u

Assuming there is a proper class of Woodin cardinals, the relation, T `Ω φ, is

generically absolute. Moreover Soundness holds as well.

Theorem 17. Assume there is a proper class of Woodin cardinals. Then for all (T, φ)
and for all complete Boolean algebras, B, T `Ω φ if and only if V B  “T `Ω φ”. t
u

Theorem 18 (Soundness). Assume that there is a proper class of Woodin cardinals

and that T `Ω φ. Then T Ω φ. t
u

I now come to the Ω Conjecture which in essence is simply the conjecture that the
Gödel Completeness Theorem holds for Ω-logic; see [13] for a more detailed discus-
sion.

Definition 19 (Ω Conjecture). Suppose that there exists a proper class of Woodin car-
dinals. Then for all sentences φ, ∅ Ω φ if and only if ∅ `Ω φ. t
u
Infinity Book—woodin 2009 Oct 04 24

Assuming the Ω Conjecture one can analyze the generic-multiverse view of truth
by computing the logical complexity of Ω-logic. The key issue of course is whether
the generic-multiverse view of truth satisfies the two multiverse laws. This is the sub-
ject of the next section. We end this section with a curious connection between the
Ω Conjecture, HOD, and the universally Baire sets. This requires a definition.

Definition 20. A set A ⊆ R is OD if there exists an ordinal α and a formula φ such

that
A = {x ∈ R | Vα  φ[x]} . t
u

Theorem 21. Suppose that there is a proper class of Woodin cardinals and that for
every set A ⊆ R, if A is OD then A is universally Baire. Then
HOD  “Ω Conjecture” t
u

1.4.3 The complexity of Ω-logic

Let (as defined on page 21) VΩ be the set of sentences φ such that
∅ Ω φ,
and (assuming there is a proper class of Woodin cardinals) let VΩ (Vδ0 +1 ) be the set of
sentences, φ, such that
ZFC Ω “Vδ0 +1  φ”.

Assuming there is a proper class of Woodin cardinals then the set of generic-
multiverse truths which are Π2 assertions is of the same Turing complexity as VΩ
(i.e., each set is recursive in the other). Further (assuming there is a proper class of
Woodin cardinals) the set, VΩ (Vδ0 +1 ), is precisely the set of generic-multiverse truths
of Vδ0 +1 . Thus the requirement that the generic-multiverse position satisfies the First
Multiverse Law, as discussed on page 19, reduces to the requirement that VΩ not be
recursive in the set VΩ (Vδ0 +1 ).
The following theorem is a corollary of the basic analysis of Ω-logic in the context
that there is a proper class of Woodin cardinals.

Theorem 22. Assume there is a proper class of Woodin cardinals and that the Ω Con-
jecture holds. Then the set VΩ is recursive in the set VΩ (Vδ0 +1 ). t
u

Therefore, assuming the existence of a proper class of Woodin cardinals and that
the Ω Conjecture both hold across the generic-multiverse generated by V, the generic-
multiverse position violates the First Multiverse Law. What about the Second Multi-
verse Law (on page 19)? This requires understanding the complexity of the set VΩ .
From the definition of VΩ it is evident that this set is definable in V by a Π2 formula:
if V = L then this set is recursively equivalent to the set of all Π2 sentences which are
Infinity Book—woodin 2009 Oct 04 25

true in V. However in the context of large cardinal axioms the complexity of VΩ is

more subtle.

Theorem 23. Assume there is a proper class of Woodin cardinals and that the Ω Con-
jecture holds. Then the set VΩ is definable in Vδ0 +1 . t
u

Therefore if the Ω Conjecture holds and there is a proper class of Woodin cardinals
then the generic-multiverse position that the only Π2 assertions which are true are
those which are true in each universe of the generic-multiverse also violates the Second
Multiverse Law—for this set of assertions is itself definable in Vδ0 +1 across the generic-
multiverse. I make one final comment here. The weak multiverse laws are the versions
of the two multiverse laws I have defined where Vδ0 +1 is replaced by Vω+2 . Assuming
the Ω Conjecture and that there is a proper class of Woodin cardinals then the generic-
multiverse position actually violates the Weak First Multiverse Law, and augmented
by a second conjecture, it also violates the Weak Second Multiverse Law, more details
can be found in [14]. The example of Theorem 12 violates both weak multiverse laws.

1.5 The infinite realm

If the Ω Conjecture is true then what is the only plausible alternative to the conception
of the universe of sets is arguably ruled out. The point is that any multiverse conception
based on a (reasonable) multiverse not smaller than the generic-multiverse also violates
the two multiverse laws. Here a multiverse, V∗M , associated to a countable transitive
model M is not smaller than the generic-multiverse if for each N ∈ V∗M , VN ⊆ V∗M .
Similarly, V∗M is smaller than the generic multiverse (generated by M) if V∗M ⊆ V M .
The following theorem gives a more precise version of this claim.

Theorem 24. Suppose that M is a countable transitive set

M  ZFC + “There is a proper class of Woodin cardinals.”,

and that
M  “The Ω Conjecture.”

Suppose V∗M is a multiverse generated by M which is not smaller than the generic-
multiverse and which contains only transitive sets N such that

N  ZFC + “There is a proper class of Woodin cardinals.”,

and such that

N  “The Ω Conjecture.”

Then the multiverse view of truth given by V∗M violates both multiverse laws. t
u
Infinity Book—woodin 2009 Oct 04 26

Finally, there is no compelling candidate for a multiverse view of truth based on

a multiverse generated by V which is smaller than the generic-multiverse; other than
the multiverse which just contains V. But there certainly are candidates. For example,
restricting the generic-multiverse to the multiverse generated by only allowing forcing
notions which are homogeneous could (as far as I know) give a multiverse view which
does not violate the multiverse laws. Further given one motivation I cited for a generic-
multiverse view (based on Theorem 10), such a restriction is a natural one.
At this point, I just do not see any argument for such a restricted multiverse view.
The fundamental problem with a multiverse view based on a multiverse which is
smaller than the generic-multiverse is that there must exist Π2 sentences which are
declared to be true and which are not true in the generic-multiverse view. At present
there is simply no natural candidate for such a collection of Π2 sentences which is not
simply the set of all Π2 sentences true in V.
The determined advocate for the generic-multiverse conception of truth might sim-
ply accept the failure of the multiverse laws as a deep fact about truth in Set Theory,
and seek salvation in the multiverse truths which are beyond Π2 sentences. But then
this advocate must either explain the restriction to set forcing extensions in the defini-
tion of the generic-multiverse or specify exactly how the multiverse is to be defined.
Restricting to set generic extensions resurrects the specter of unsolvable problems. For
example consider the question of whether there exists a cardinal λ such that the Gen-
eralized Continuum Hypothesis holds for all cardinals above λ. The answer to this
question is invariant across the generic-multiverse. Alternatively, specifying the class
forcing extensions which are to be allowed seems utterly hopeless at the present time if
the corresponding multiverse conception of truth is to declare the question above to be
meaningless and yet preserve large axioms as meaningful. Further there is the issue of
whether truth in this expanded generic-multiverse is reducible to truth in the universes
of that multiverse as is the case of the generic-multiverse.
Of course one could just conclude from all of this that the Ω Conjecture is false and
predict that the solution to specifying the true generic-multiverse will be revealed by
the nature of the failure of the Ω Conjecture. But the Ω Conjecture is invariant across
the generic-multiverse. Thus it is not unreasonable to expect both that the Ω Conjec-
ture has an answer and further if that answer is that it is false, then the Ω Conjec-
ture be refuted from some large cardinal hypothesis. Many of the meta-mathematical
consequences of the Ω Conjecture follow from the nontrivial Ω-satisfiability of the
Ω Conjecture; this is the assertion that for some universe V ∗ of the generic-multiverse
generated by V, there exists an ordinal α such that

Vα∗  ZFC + “There is a proper class of Woodin cardinals”

and such that

Vα∗  “The Ω Conjecture”.
Infinity Book—woodin 2009 Oct 04 27

This assertion is itself a Σ2 -assertion and so assuming there is a proper class of Woodin
cardinals, this assertion must also be invariant across the generic-multiverse generated
by V. While the claim that if the Ω Conjecture is false, then the Ω Conjecture must
be refuted from some large cardinal hypothesis, is debatable, the corresponding claim
for the nontrivial Ω-satisfiability of the Ω Conjecture (in the sense just defined) is
much harder to argue against. The point here is that while there are many examples
of sentences which are provably absolute for set forcing and which cannot be decided
by any large cardinal axiom, there are no known examples where the sentence is Σ2 .
In fact if the Ω Conjecture is true then there really can be no such example. Finally
it seems unlikely that there is a large cardinal axiom which proves the nontrivial Ω-
satisfiability of the Ω Conjecture and yet there be no large cardinal axiom which proves
the Ω Conjecture.

1.5.1 Probing the universe of sets; the Inner Model Program

The Inner Model Program is the detailed study of large cardinal axioms. The first
construction of an inner model is due to Gödel, [2] and [3]. This construction founded
the Inner Model Program, the transitive class constructed is denoted by L and I have
given the definition. The question of whether V = L is an important one for Set Theory.
The answer has profound implications for the conception of the universe of sets.

Theorem 25 (Scott, [11]). Suppose there is a measurable cardinal. Then V , L. t

u

The Axiom of Constructibility is the axiom which asserts, “V = L”; more precisely
this is the axiom which asserts that for each set a there exists an ordinal α such that
a ∈ Lα . Scott’s theorem provided the first indication that the Axiom of Constructibil-
ity is independent of the ZFC axioms. At the time there was no compelling reason
to believe that the existence of a measurable cardinal was consistent with the ZFC
axioms, so one could not make the claim that Scott’s theorem established the formal
independence of the Axiom of Constructibility from the ZFC axioms. Of course it
is an immediate corollary of Cohen’s results that the Axiom of Constructibilit is for-
mally independent of the ZFC axioms. The modern significance of Scott’s theorem
is more profound; I would argue that Scott’s theorem establishes that the Axiom of
Constructibility is false. This claim (that V , L) is not universally accepted but in my
view no one has come up with a credible argument against this claim.
The Inner Model Program seeks generalizations of L for the large cardinal axioms,
in brief it seeks generalizations of the Axiom of Constructibility which are compatible
with large cardinal axioms (such as the axioms for measurable cardinals and beyond).
It has been a very successful program and its successes have led to the realization
that the large cardinal hierarchy is a very “robust” notion. The results which have
been obtained provide some of our deepest glimpses into the universe of sets. Despite
Infinity Book—woodin 2009 Oct 04 28

the rather formidable merits as indicated above, there is a fundamental difficulty with
the prospect of using the Inner Model Program to counter the Skeptic’s Retreat. The
problem is in the basic methodology of the Inner Model Program. But to explain this
I must give a (brief) description of the (technical) template for inner models.
The inner models which are the goal and focus of the Inner Model Program are
defined layer by layer working up through the hierarchy of large cardinal axioms,
which in turn is naturally revealed by the construction of these inner models. Each
layer provides the foundation for the next and L is the first layer.
Roughly (and in practice) in constructing the inner model for a specific large car-
dinal axiom, one obtains an exhaustive analysis of all weaker large cardinal axioms.
There can be surprises here in that seemingly different notions of large cardinals can
coincide in the inner model. Finally as one ascends through the hierarchy of large
cardinal axioms, the construction generally becomes more and more difficult.
However there is a fundamental problem with appealing to the Inner Model Pro-
gram to counter the Skeptic’s Retreat. Suppose (for example) that a hypothetical large
cardinal axiom “Φ” provides a counterexample to the Skeptic’s Retreat and this is ac-
complished by the Inner Model Program. To use the Inner Model Program to refute
the existence of an “Φ-cardinal” one first must be able to successfully construct the
inner models for all smaller large cardinals and this hierarchy would be fully revealed
by the construction.
Perhaps this could happen, but it can only happen once. This is the problem. Hav-
ing refuted the existence of an “Φ-cardinal” how could one then refute the existence
of any smaller large cardinals, for one would have solved the inner model problem for
these smaller large cardinals. The fundamental problem is that the Inner Model Pro-
gram seems inherently unable, by virtue of its inductive nature, to provide a framework
for an evolving understanding of the boundary between the possible and the impossible
(large cardinal axioms).
Thus it would seem that the Skeptic’s Retreat is in fact a powerful counter-attack.
But there is something wrong here and the answer lies in understanding large cardinal
axioms which are much stronger than those within reach of the current hierarchy of
inner models.

1.5.2 Supercompact cardinals and beyond

Paraphrasing a standard definition: an extender is an elementary embedding
E : Vα+1 ∩ M → Vβ+1 ∩ M
where M is a transitive class such that M  ZFC. Necessarily α ≤ β and E(α) = β. It
is not difficult to show that if for all ordinals ξ ≤ α, E(ξ) = ξ then for all a ∈ Vα+1 ,
E(a) = a. Thus if E is non-trivial there must exist an ordinal γ ≤ α such that E(γ) , γ
and the least such ordinal γ is the critical point of E and is denoted CRT(E).
Infinity Book—woodin 2009 Oct 04 29

Extenders are the building blocks for the Inner Model Program which seeks en-
largements of L which are transitive classes N such that N contains enough extenders
to witness that the targeted large cardinal axiom holds in N. The complication is in
specifying just which extenders are to be included in N.
The definition of a supercompact cardinal is due to Reinhardt and Solovay—see
[5] for more on the history of the axiom. Below is a reformulation of the definition
due to Magidor in terms of extenders.

Definition 26. A cardinal δ is a supercompact cardinal if for each ordinal β > δ there
exists an extender
E : Vα+1 → Vβ+1

such that E(κ) = δ where κ = CRT(E). t

u

Slightly stronger is the notion that δ is an extendible cardinal: for all α > δ there exists
an extender,
E : Vα+1 → Vβ+1

such that CRT(E) = δ.

As I have already indicated, the strongest large cardinal axioms not known to be
inconsistent with the Axiom of Choice are the family of axioms asserting the existence
of strongly (ω + 1)-huge cardinals. These axioms have seemed so far beyond any
conceivable inner model theory that they simply are not understood.
The possibilities for an inner model theory at the level of supercompact cardinals
and beyond, has been essentially a complete mystery until recently. The reason lies in
the nature of extenders. Again for expository purposes, let me define an extender

E : Vα+1 → Vβ+1

to be a suitable extender if E(CRT(E)) > α. Thus an extender

E : Vα+1 ∩ M → Vβ+1 ∩ M

is a suitable extender if it is not too long and if Vβ+1 ⊂ M. For example suppose that κ
is a strongly (ω + 1)-huge cardinal as defined in Definition 6 on page 13. Then there
exists there exist γ > λ > κ such that

(1) Vκ ≺ Vλ ≺ Vγ ,

(2) there exists an elementary embedding,

j : Vλ+1 → Vλ+1

such that κ is the critical point of j.

Thus j is an extender but j is not a suitable extender. In particular the existence of a

strongly (ω + 1)-huge cardinal cannot be witnessed by a suitable extender.
Infinity Book—woodin 2009 Oct 04 30

The Inner Model Program at the level of supercompact cardinals and beyond seeks
enlargements N of L such that there are enough extenders E such that E|N ∈ N to
witness the targeted large cardinal axiom holds in N. For the weaker large cardinal
axioms this has been an extremely successful program. For example, Mitchell-Steel
[8], have defined enlargements at the level of Woodin cardinals. In fact they define
the basic form of such enlargements of L up to level of superstrong cardinals which
are just below the level of supercompact cardinals. The Mitchell-Steel models are
constructed from sequences of extenders. The basic methodology is to construct N
from a sequence of extenders which includes enough extenders to directly witness that
the targeted large cardinal axiom holds in N.
For the construction of the Mitchell-Steel models there is a fundamental require-
ment that the extenders on sequence from which the enlargement of L is constructed
be derived from extenders

E : Vα+1 ∩ M → Vβ+1 ∩ M

such that Vα+1 ⊂ M. Following this basic methodology the enlargements of L at

the level of supercompact cardinals and beyond must be constructed from extender
sequences which now include extenders which are restrictions of extenders of the form,

E : Vα+1 → Vβ+1 ,

and Steel has shown that the basic methodology of analyzing extender models encoun-
ters serious obstructions once there are such extenders on the sequence, particularly if
the extenders are not suitable.
But by some fairly recent theorems something completely unexpected and remark-
able happens. Suppose that N is a transitive class, for some cardinal δ,

N  “δ is a supercompact cardinal”,

and that this is witnessed by class of all E|N such that E|N ∈ N and such that E is a
suitable extender. Then the transitive class N is close to V and N inherits essentially all
large cardinals from V. The amazing thing is that this must happen no matter how N is
constructed. This would seem to undermine my earlier claim that inner models should
be constructed from extender sequences which contain enough extenders to witness
that the targeted large cardinal axiom holds in the inner model. It does not and the
reason is that by simply requiring that E|N ∈ N for enough suitable extenders from
V to witness that the large cardinal axiom, “There is a supercompact cardinal”, holds
in N, one (and this is the surprise) necessarily must have E|N ∈ N for a much larger
class of extenders, E : Vα+1 → Vβ+1 . So the principle that there are enough extenders
in N to witness the targeted large cardinal axiom holds in N is preserved (as it must
be). The change, in the case that N is constructed from a sequence of extenders which
includes restrictions of suitable extenders, is that these extenders do not have to be on
the sequence from which N is constructed. In particular in this case, large cardinal
Infinity Book—woodin 2009 Oct 04 31

axioms can be witnessed to hold in N by “phantom” extenders, these are extenders of

N which are not on the sequence, which cannot be witnessed to hold by any extender
on the sequence. This includes large cardinal axioms at the level of strongly (ω + 1)-
huge cardinals. As a consequence of this, one can completely avoid the cited obstacles
because: One does not need to have the kinds of extenders on the sequence which give
rise to the obstacles. Specifically, one can restrict consideration to extender sequences
of just extenders derived from suitable extenders and this is a paradigm shift in the
whole conception of inner models.
The analysis yields still more. Suppose that there is a positive solution (in ZFC) to
the inner model problem for just one supercompact cardinal. Then as a corollary one
would would obtain a proof of the following conjecture.

Conjecture (ZF) There are no weak Reinhardt cardinals. t

u

Suppose this conjecture is actually true and it is proved according to the scenario
that I have just described. This would in a convincing fashion refute the Skeptic’s
Retreat providing for the first time an example of a natural large cardinal axiom proved
to be inconsistent as a result of a deep structural analysis.
In fact, it is possible to isolate a specific conjecture which must be true if there is a
positive solution to the inner model problem for one supercompact cardinal and which
itself suffices for this inconsistency result. This conjecture (which is the HOD Con-
jecture of [16]) concerns HOD and we refer the interested (and dedicated) reader to
[16] for details. Actually a corollary of the HOD Conjecture suffices to prove the in-
consistency conjectured above and this corollary we can easily state. First, a cardinal
κ is a regular cardinal if every subset X ⊂ κ with |X| < κ, is bounded in κ. Thus ω
is a regular cardinal as is ω1 (assuming the Axiom of Choice). The corollary of the
HOD Conjecture is the following: There is a proper class of regular cardinals which
are not measurable cardinals in HOD.
I mention this for two reasons. First, this is a specific and precise conjecture which
does involve the Inner Model Program at all and so offers an independent route to
proving the conjectured inconsistency above. Second, it identifies specific combinato-
rial consequences of having a successful solution to the inner model problem for one
supercompact cardinal, and so provides a potential basis for establishing that there is
no solution to the inner model problem for one supercompact cardinal.
The extension of the Inner Model Program to level of one supercompact cardinal
(the definition of “ultimate L”) will come with a price. The successful extension of
the Inner Model Program to a large cardinal axiom can no longer serve as the basis
for the claim of the formal consistency of that axiom. The reason is that as the pre-
vious discussion indicates, in extending the Inner Model Program to the level of one
supercompact cardinal, one will have extended the Inner Model Program to essentially
all known large cardinals. The ramifications are discussed at length in [16]. In brief
Infinity Book—woodin 2009 Oct 04 32

further progress in understanding (and even discovering) large cardinal axioms would
have to depend on structural considerations of “ultimate L”.

1.6 Conclusions
The development of the mathematical theory of infinity has led to a number of spe-
cific predictions. These predictions assert that certain technical axioms concerning the
existence of large cardinals, are not formally inconsistent with the axioms of set the-
ory. As I have indicated these predications are actually predications about the physical
universe. To date there is no known (and credible) explanation for these predictions
except that they are true because the corresponding axioms are true in the universe of
sets. As the arguments of the first section indicate, these same issues arise even for the
conception of large finite sets.
As discussed in the second section, there is a serious challenge to this claim, even
ignoring the often cited challenge; the ubiquity of unsolvable problems in Set Theory.
The challenge arises from the fact that here are formal axioms of infinity which are ar-
guably a serious foundational issue for Set Theory for two reasons. First, these axioms
are known to refute the Axiom of Choice and second, these axioms are known to be
“stronger” than essentially all the notions of infinity believed to be formally consistent
with the Axiom of Choice. Here the metric for strength is simply the inference rela-
tion for the corresponding predictions (of formal consistency). The issues raised by
this are twofold. First (regarding the debate between the Set Theorist and the Skeptic),
there is no need to explain the success of a single prediction, it is a succession of ever
stronger successful predictions which demands explanation. But this one prediction
of consistency subsumes all the predictions made to date and so there is no series of
predictions which requires explanation. Second, for the Set Theorist to account for
this one prediction it would seem that a different conception of the Universe of Sets is
required.
The conception of a Universe of Sets in which the Axiom of Choice fails creates
more difficulties than it solves and so this does not seem to be a viable option. However
any large cardinal axiom (which is expressible by a Σ2 -sentence) which can hold in a
universe of sets satisfying all of the axioms except for the Axiom of Choice, can hold
in a generic extension of a universe of sets which does satisfy the Axiom of Choice.
Therefore this challenge, and the challenge posed by formally unsolvable problems
such as that of the Continuum Hypothesis, might both be addressed (but perhaps not
completely in a satisfactory manner) by adopting the conception of a multiverse of sets.
Here the Ω Conjecture emerges as a key conjecture. If this conjecture is true then what
is arguably the only candidate for a multiverse view for the infinite realm which can
address these challenges, also fails to be a viable alternative (accepting the requirement
Infinity Book—woodin 2009 Oct 04 33

that the multiverse laws of Section 3 be satisfied). Therefore if the multiverse view is
correct, the Ω Conjecture must be false.
The attempt to understand how the Ω Conjecture might be refuted leads directly
to the Inner Model Program. The Inner Model Program is the attempt to generalize
the definition of L to yield transitive classes M in which large cardinal axioms hold.
If the Inner Model Program as described in the fourth section can be extended to the
level of a single supercompact cardinal then no known large cardinal axiom can refute
the Ω Conjecture. Further one would also obtain as corollary the verification of a
series of conjectures. These conjectures imply that the large cardinal axioms—such
as the axiom which asserts the existence of a weak Reinhardt cardinal—which pose
such a challenge to the conception of the universe of sets, are formally inconsistent.
These inconsistency results would be the first examples of inconsistency results for
large cardinal axioms obtained only through a very detailed analysis.
Finally the extension of the Inner Model Program to the level of one supercom-
pact cardinal will yield examples (where none are currently known) of a single formal
axiom which is compatible with all the known large cardinal axioms and which pro-
vides an axiomatic foundation for Set Theory which is immune to independence by
Cohen’s method. This axiom will not be unique but there is the very real possibility
that among these axioms, there is an optimal one (from structural and philosophical
considerations). In which case we will have returned, against all odds or reasonable
expectation, to the view of truth for Set Theory which was present at the time when
the investigation of Set Theory began.
Bibliography

[1] Qi Feng, Menachem Magidor, and W. Hugh Woodin. Universally Baire sets of
reals. 26:203–242, 1992.

[2] Kurt Gödel. Consistency-Proof for the Generalized Continuum-Hypothesis.

Proc. Nat. Acad. Sci. U.S.A, 25:220–224, 1938.

[3] Kurt Gödel. The Consistency of the Continuum Hypothesis. Annals of Mathe-
matics Studies, no. 3. Princeton University Press, Princeton, N. J., 1940.

[4] Joel David Hamkins and W. Hugh Woodin. Small forcing creates neither strong
nor Woodin cardinals. Proc. Amer. Math. Soc., 128(10):3025–3029, 2000.

[5] Akihiro Kanamori. The higher infinite. Perspectives in Mathematical Logic.

Springer-Verlag, Berlin, 1994. Large cardinals in set theory from their begin-
nings.

[6] Kenneth Kunen. Elementary embeddings and infinitary combinatorics. J. Sym-

bolic Logic, 36:407–413, 1971.

[7] Azriel Lévy. Definability in axiomatic set theory. I. In Logic, Methodology

and Philos. Sci. (Proc. 1964 Internat. Congr.), pages 127–151. North-Holland,
Amsterdam, 1965.

[8] William J. Mitchell and John R. Steel. Fine structure and iteration trees.
Springer-Verlag, Berlin, 1994.

[9] Jan Mycielski and H. Steinhaus. A mathematical axiom contradicting the axiom
of choice. Bull. Acad. Polon. Sci. Sér. Sci. Math. Astronom. Phys., 10:1–3, 1962.

[10] William N. Reinhardt. Ackermann’s set theory equals ZF. Ann. Math. Logic,
2(2):189–249, 1970.

[11] Dana Scott. Measurable cardinals and constructible sets. Bull. Acad. Polon. Sci.
Sér. Sci. Math. Astronom. Phys., 9:521–524, 1961.

34
Infinity Book—woodin 2009 Oct 04 35

[12] Hugh Woodin. The Tower of Hanoi. In The Second International Meeting on
Truth in Mathematics, volume – of –, pages –. Oxford University Press, Oxford,
1998.

[13] W. Hugh Woodin. Set Theory after Russell; The journey back to Eden, volume 6
of de Gruyter Series in Logic and its Applications. Walter de Gruyter & Co.,
Berlin, 2004.

[14] W. Hugh Woodin. The Continuum Hypothesis, the generic-multiverse of sets,

and the Ω Conjecture. In press, 2009.

[15] W. Hugh Woodin. The Transfinite Universe. To appear in Gödel volume, 2009.

[16] W. Hugh Woodin. Suitable extender sequences. Preprint, pages 1–677, July,
2009.

[17] Ernst Zermelo. Über Grenzzahlen und Mengenbereiche: Neue Untersuchun-

genuberdie Grundlagen der Mengenlehre. Fundamenta Mathematicae, 16:29–
47, 1930.