Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/297319798

Set theory

Chapter · January 2010

CITATIONS READS
0 10,446

1 author:

Joan Bagaria
Catalan Institution for Research and Advanced Studies (ICREA) and Universitat de Barcelona
64 PUBLICATIONS   508 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Reflection principles View project

Ramsey methods for coordinate systems in separable Banach spaces View project

All content following this page was uploaded by Joan Bagaria on 07 November 2018.

The user has requested enhancement of the downloaded file.


Set Theory1 1, followed by all new solutions of equations of in-
dex 2, and so on.
J. Bagaria
Two sets A and B have the same size, or cardi-
ICREA (Catalan Institution for Research and nality, if there is a bijection between them, that is,
Advanced Studies) and University of Barcelona the elements of A can be put into one-to-one cor-
respondence with the elements of B. Thus, N and
1 Introduction the set of algebraic numbers have the same cardi-
nality. It follows that Z and Q have also the same
Set theory is the theory of abstract sets. It occu- cardinality as N. All these sets are countable.
pies a special place among all mathematical disci- But Cantor discovered that, surprisingly, the set
plines, for it is both an area of mathematics and, R of real numbers is not countable. Here is Can-
at the same time, it provides mathematics with its tor’s original proof: suppose, aiming for a con-
foundation. Moreover, insofar as set theory is a tradiction, that r0 , r1 , r2 , . . . is an enumeration of
source of the axioms of mathematics and consti- R. Let a0 = r0 . Choose the least k such that
tutes a sound basis for the existence of infinite a0 < rk and put b0 = rk . Given an and bn ,
mathematical objects, it is also of philosophical choose the least l such that an < rl < bn , and
significance. put an+1 = rl . And choose the least m such that
an+1 < rm < bn , and put bn+1 = rm . Thus, we
2 The Theory of Transfinite Num- have a0 < a1 < a2 < · · · · · · < b2 < b1 < b0 . Now
let a be the limit of the an . Then a is a real num-
bers
ber different from rn , for all n, contradicting our
Set theory began with the work of Georg Can- assumption that the sequence r0 , r1 , r2 , . . . enume-
tor. In 1874, he proved that the set of algebraic rates all real numbers.
numbers is countable. This means they can be So, for the first time it was established that there
enumerated, that is, put into one-to-one correspon- are at least two kinds of infinite sets: the count-
dence with the set N of natural numbers. Recall able and the uncountable. Cantor also showed that
that the algebraic numbers are the complex solu- there are bijections between any two of the sets Rn ,
tions of polynomial equations of the form: n ≥ 1, and even RN , the set of all infinite sequences
r0 , r1 , r2 . . . of real numbers, and therefore all have
an X n + an−1 X n−1 + · · · + a1 X + a0 = 0 the same uncountable cardinality.
From 1879 to 1884 Cantor published a series of
where the coefficients ai are integers and an is non-
works that constitute the origin of set theory. In
zero (and so may as well be assumed to be strictly
them he introduced the notion of infinite, or trans-
positive). Given such an equation, let the number
finite, ordinals. When we use the natural num-
|an | + |an−1 | + · · · + |a0 | + n bers to count a collection of objects we assign a
number to each counted object, starting with 1,
be its index. It is clear that for every k > 0 there continuing with 2, 3, etc., and stopping when we
are only a finite number of equations of index k. have counted each object exactly once. When this
For instance, there are only 4 equations of index 3 process is over we have done two things. On one
with strictly positive an , namely, X 2 = 0, 2X = 0, hand we have obtained a number n, the last one
X + 1 = 0, and X − 1 = 0, which have as solutions in the sequence, that tells us how many objects
0, −1, and 1. It follows that for all k > 0 there are there are in the collection. On the other hand
only finitely-many solutions of equations of index we have at the same time ordered the objects be-
k. Thus, we can enumerate all algebraic numbers ing counted in accordance with the ordering of the
by enumerating all solutions of equations of index natural numbers. Thus, natural numbers are both
1 This is an expanded version of the article published in:
cardinal numbers, which indicate a quantity, and
The Princeton Companion to Mathematics. Edited
ordinal numbers, which indicate a position in an
by Timothy Gowers. June Barrow-Green and Imre Leader, ordered sequence. Suppose now we want to count
associate editors. Princeton University Press, 2008. the points in the unit interval [0, 1]. Cantor’s ar-

1
2

gument given above shows that no matter how the


points are counted we will run out of natural num-
bers before we have counted all the points in the
interval. That is, a one-to-one function from N
into R can never be onto. Nothing prevents us,
however, from introducing new ordinal numbers Figure 1: ω and ω + 1 have the same cardinality.
and continuing ‘counting’. For this, we first need
an ordinal number that represents the position in
the ordinal sequence that comes right after all the
natural numbers. This is the first infinite ordi- instance, the natural number n is represented by
nal number, which Cantor denoted by ω. In other the set {0, 1, 2, . . . , n − 1}, which has cardinality
words, after 0, 1, 2, 3, . . . comes ω. It is also the n, and the ordinal ω + 3 is identified with the set
first limit ordinal, for it is not the immediate suc- {0, 1, 2, . . . , ω, ω + 1, ω + 2}. Thinking of ordinals
cessor of any of its predecessors. But once we have this way, the ordering on the ordinals becomes set
ω we can continue the ordinal sequence simply by membership: if α comes before β in the ordinal
adding 1 repeatedly. Thus, the sequence of ordinal sequence, then α is one of the predecessors of β
numbers begins as follows: and therefore an element of β. A critically impor-
tant property of this ordering is that each ordinal
0, 1, 2, 3, 4, 5, 6, 7, . . . , ω, ω + 1, ω + 2, ω + 3, . . . is a well-ordered set, which means that every non-
empty subset of it has a least element.
Then comes the next limit ordinal, which it seems
As we said earlier, cardinal numbers are used
natural to call ω + ω, and which we can write as
for measuring the sizes of sets, while ordinal num-
ω · 2. The sequence continues as:
bers indicate the position in an ordered sequence.
ω · 2, ω · 2 + 1, ω · 2 + 2, . . . , ω · n, . . . , ω · n + m, . . . This distinction is much more apparent for infinite
numbers than for finite ones, because it is possible
As this discussion indicates, there are two basic for two infinite ordinals to have the same size. For
rules for generating new ordinals: adding 1 and example, ω and ω + 1 are different ordinals, but, as
passing to the limit. What we mean by “passing Figure 1 shows, the corresponding sets {0, 1, 2, . . .}
to the limit” is “assigning a new ordinal number and {0, 1, 2, . . . , ω} have the same cardinality. In
to the position in the ordinal sequence that comes fact, all sets that can be counted using the infinite
straight after all the ordinals obtained so far”. For ordinals we have described so far are countable. So
example, after all the ordinals ω ·n, comes the next in what sense are different ordinals different? The
limit ordinal, which we write as ω · ω, or ω 2 , and point is that although two sets such as {0, 1, 2, . . .}
we obtain: and {0, 1, 2, . . . , ω} have the same cardinality, they
are not order isomorphic: that is, you cannot find
ω 2 , ω 2 + 1, . . . , ω 2 + ω, . . . , ω 2 + ω · n, . . . , ω 2 · n, . . . a bijection f from one set to the other such that
f (x) < f (y) whenever x < y. Thus they are the
Eventually, we reach ω 3 and the sequence contin-
same “as sets” but not “as ordered sets”.
ues as:
Informally, the cardinal numbers are the possi-
ω 3 , ω 3 + 1, . . . , ω 3 + ω, . . . , ω 3 + ω 2 , . . . , ω 3 · n, . . . ble sizes of sets. A formal definition of a cardi-
nal number is that it is an ordinal number that is
The next limit ordinal is ω 4 , etc. After all the ω n , bigger than all its predecessors. Thus, all natural
for n a natural number, comes ω ω . And after ω ω , numbers are cardinal numbers, and so is ω. Can-
ω ωω
ω ω , ω ω ,... comes the limit ordinal denoted by tor used the first letter of the Hebrew alphabet ℵ
0 . And on and on it goes. (aleph) to represent the infinite cardinal numbers,
In set theory, one likes to regard all mathemat- ℵ0 denoting the first one, so that ℵ0 = ω.
ical objects as sets. For ordinals this can be done Cantor called numbers of the second class– the
in a particularly simple way: we represent 0 by natural numbers being the numbers of the first
the empty set and the ordinal number α is repre- class – those infinite ordinal numbers that are ob-
sented by the set of all its predecessors. Thus, for tained starting from ω by means of the two gen-
2. THE THEORY OF TRANSFINITE NUMBERS 3

erating rules of adding 1 and passing to the limit, b ∈ B. (For this set, which is called the Cartesian
and are countable. The set of all such ordinals is product, we do not need A and B to be disjoint.)
a new ordinal, which Cantor called ω1 , that rep- The point of these definitions is that they apply
resents the position in the ordinal sequence that just as well to infinite cardinal numbers: the sum
comes right after all the countable infinite ordi- of two infinite cardinals κ and λ is the cardinal of
nals. He also observed that ω1 is not countable, the union of any two disjoint sets, one of cardi-
for it cannot contain itself as an element, and rep- nality κ and the other of cardinality λ. And the
resented the cardinal number corresponding to this product of κ and λ is the cardinal of the Cartesian
set by ℵ1 . Thus ℵ1 = ω1 is the first uncountable product of any two sets, one of cardinality κ and
cardinal. the other of cardinality λ. The resulting arithmetic
This process can be repeated. The ordinals of of transfinite cardinals is very simple, however. It
the third class are those obtained starting from turns out that for all transfinite cardinals ℵα and
ω1 by the two generating rules of adding 1 and ℵβ ,
passing to the limit, and have cardinality ℵ1 (or
equivalently the set of all ordinals that can be ℵα + ℵβ = ℵα · ℵβ = max(ℵα , ℵβ ) = ℵmax(α,β) .
put in one-to-one correspondence with ω1 ). The
set of all such ordinals, called ω2 , is the small- It is also possible to define exponentiation for
est ordinal having cardinality greater than ℵ1 , and transfinite cardinals, but for this the picture
so the corresponding cardinal is then called ℵ2 . changes completely. If κ and λ are cardinals, then
We can continue generating a sequence of ordi- κλ is defined as the cardinal of the Cartesian prod-
nals ω1 , ω2 , ω3 , . . . of larger and larger cardinaltiy. uct of λ copies of any set of cardinality κ. Equiva-
Moreover, using limits as well, e can continue the lently, it is the cardinality of the set of all functions
sequence transfinitely. For example, the ordinal from a set of cardinality λ into a set of cardinal-
obtained as the limit of the ωn , n ∈ N, is called ωω , ity κ. Again, if κ and λ are finite, this gives the
and the corresponding limit cardinal ℵω . In this usual definition: for instance the number of func-
way we obtain the sequence of infinite, or transfi- tion from a set of size 3 to a set of size 4 is 43 .
nite, ordinals, What happens if we take the simplest nontrivial
transfinite example, 2ℵ0 ? Not only is this question
ω
0, 1, . . . , ω, ω+1, . . . , ω , . . . , ω1 , . . . , ω2 , . . . , ωω , . . . extremely hard, there is a sense in which it cannot
be resolved, as we shall see later.
as well as the sequence of infinite, or transfinite, The most obvious set of cardinality 2ℵ0 is the set
cardinals: of all functions from N into {0, 1}. If f is such a
function, then we can regard it as giving the binary
ℵ0 , ℵ1 , . . . , ℵω , ℵω+1 , . . . , ℵωω , . . . ,
expansion of the number
ℵω1 , . . . , ℵω2 , . . . , ℵωω , . . . X
x= f (n)2−(n+1) ,
where for each ordinal α we have a cardinal ℵα . n∈N
Observe that the alephs give the cardinality of the
infinite initial segments of the ordinal sequence, which belongs to the closed interval [0, 1]. (The
namely, ℵα is the cardinality of the set of all ordi- power is 2−(n+1) rather than 2−n because we are
nals less than β, for any β greater or equal than using the convention, standard in set theory, that
ωα and strictly less than ωα+1 . 0 is the first natural number rather than 1.) Since
Given two natural numbers, we can calculate every point in [0, 1] has at most two different bi-
their sum and product. A convenient set-theoretic nary representations, it follows easily that 2ℵ0 is
way to do define these binary operations is as fol- also the cardinality of [0, 1], and therefore also the
lows. Given natural numbers m and n, take any cardinality of R. Thus, 2ℵ0 is uncountable, and so
two disjoint sets A and B of size m and n, respec- greater or equal than ℵ1 . Cantor’s conjecture that
tively; m + n is then the size of the union A ∪ B. it is exactly ℵ1 is the famous Continuum Hypoth-
As for the product, it is the size of the set A × B, esis, which will be discussed at length in section 5
the set of all ordered pairs (a, b) with a ∈ A and below.
4

It is not immediately obvious, but there are 3 The Universe of All Sets
many mathematical contexts in which transfinite
ordinals occur naturally. Cantor himself devised In the discussion so far we have taken for granted
his theory of transfinite ordinals and cardinals as that every set has a cardinality, or in other words
a result of his attempts, which were eventually that for every set X there is a unique cardinal num-
successful, to prove the continuum hypothesis for ber that can be put into one-to-one correspondence
closed sets of real numbers. he first defined the with X. If κ is such a cardinal and f : X → X is a
derivative of a set X of real numbers to be the set bijection (recall that we identify κ with the set of
you obtain when you throw out all the “isolated” its predecessors), then we can define an ordering
points of X. These are points x for which you can on X by taking x < y if and only if f (x) < f (y).
find a small neighborhood around x that contains Since κ is a well-ordered set, this makes X into a
no other points in X. For example, if X is the set well-ordered set. But it is far from obvious that
{0}∪{1, 12 , 13 , . . .}, then all points in X are isolated, every set can be given a well-ordering. Indeed, it
except for 0. So the derivative of X is {0}. is not obvious even for the set R. (If you need
convincing of this, then try to find one.)
In general, given a set X, we can take its deriva- Thus, to make full use of the theory of transfi-
tive repeatedly. If we set X = X 0 , then we obtain nite ordinals and cardinals and solve some of the
a sequence X 0 ⊇ X 1 ⊇ X 2 ⊇ · · · , where X n+1 fundamental problems – such as computing where
is the derivative of X n . It may happen that this in the aleph hierarchy of infinite cardinals the car-
sequence does not become constant at any finite dinal of R is – one needs to appeal to the well-
stage, that is, new isolated points appear at each ordering principle: the assertion that every set can
step. If this is the case, then we can take the inter- be well-ordered. Without this assertion, one can-
section of all the X n , call it X ω , and continue with not even make sense of the questions. The well-
the derivation. Then we can define X ω+1 to be ordering principle was introduced by Cantor, but
the derivative of X ω , and so on. Thus, the reason he was unable to prove it. David Hilbert listed
that ordinals appear naturally is that we have two proving that R could be well-ordered as part of
operations, taking the derivative and taking the the first problem in his celebrated list of twenty-
intersection of everything obtained so far, which three unsolved mathematical problems presented
correspond to successors and limits in the ordinal in 1900 at the Second International Congress of
sequence. Cantor initially regarded superscripts Mathematicians, in Paris. Four years later, Ernst
such as ω + 1 as tags that marked the transfinite Zermelo gave a proof of the well-ordering principle
stages of the derivation. These tags later became that drew a lot of criticism for its use of the Axiom
the countable ordinal numbers. of Choice (AC), a principle that had been tacitly
Cantor was then able to prove that for every used for many years, but was now brought into fo-
closed set X of real numbers there is a count- cus by Zermelo’s result. AC states that for every
able ordinal α (which could be finite) such that set X of pairwise-disjoint non-empty sets there is
X α = X α+1 . It is easy to show that each X β in the a set that contains exactly one element from each
sequence of derivatives is closed, and that it con- set in X. In a second, much more detailed proof
tains all but countably many points of the original published in 1908, Zermelo spells out some of the
set X. Therefore, X α is a closed set that contains principles or axioms involved in his proof of the
no isolated points. Such sets are called perfect sets well-ordering principle, including AC.
and it is not too hard to show that they are either In that same year, Zermelo published the first
empty or have cardinality 2ℵ0 . From this it follows axiomatization of set theory, the main motivation
that X is either countable or of cardinality 2ℵ0 . being the need to continue with the development
of set theory while avoiding the logical traps, or
The intimate connection, discovered by Cantor, paradoxes, that originated in the careless use of
between transfinite ordinals and cardinals and the the intuitive notion of set. For instance, it seems
structure of the continuum was destined to leave intuitively clear that every property determines a
its mark on the entire subsequent development of set, namely, the set of those objects that have it.
set theory. But then consider the property of being an ordi-
3. THE UNIVERSE OF ALL SETS 5

nal number. If this property determined a set, this


would be the set of all ordinal numbers. But a mo-
ment of reflection shows that there cannot be such
a set, since it would be well-ordered and would
therefore correspond to an ordinal greater than all
ordinals, which is absurd. Similarly, the property
of being a set that is not an element of itself cannot
determine a set, for otherwise we fall into Russell’s
paradox, that if A is such a set, then A is an ele-
ment of A if and only if A is not an element of A,
which is absurd. Thus, not every collection of ob-
jects, not even those that are defined by some prop-
erty, can be taken to be a set. So, what is a set?
Zermelo’s 1908 axiomatization provides the first
attempt to capture our intuitive notion of set in a
short list of basic principles. It was later improved
through contributions from Thoralf Skolem, Abra- Figure 2: The universe V of all pure sets.
ham H. Fraenkel, and John von Neumann, becom-
ing what is now known as Zermelo-Fraenkel set
theory with the Axiom of Choice, or ZF C. sets already obtained in the previous stages. Once
again, this can be continued transfinitely: at limit
The basic idea behind the axioms of ZF C is that stages we collect into a set all the sets obtained
there is a “universe of all sets” that we would like to so far, and keep going. The universe of all (pure)
understand. The naturalness of the ZF C axioms sets, represented by the letter V and usually drawn
is best appreciated when viewing them as a list as a V -shape with a vertical axis representing the
of principles that produce new sets out of already ordinals (see Figure 2), therefore forms a cumula-
given ones in successive stages along the ordinal tive well-ordered hierarchy, indexed by the ordinal
sequence, eventually generating the universe of all numbers, beginning with the empty set ∅. That is,
sets. In usual mathematical practice we take sets we let
of integers, sets of real numbers, sets of functions,
V0 = ∅
etc., but also sets of sets, like sets of open sets in
a topological space, sets of sets of sets, like sets of Vα+1 = P(Vα ), the set of all subsets of Vα .
open covers, and so on. Thus, the universe of all [
sets should consist not only of sets of objects, but Vλ = Vβ , the union of all the Vβ , β < λ,
also of sets of sets of objects, etc. Now, it turns β<λ
out that it is much more convenient to dispense
if λ is a limit ordinal.
with “objects” altogether and consider only sets
whose elements are sets, whose elements are also The universe of all sets is then the union of all
sets, etc. Let us call those sets “pure sets”. The the sets Vα such that α is an ordinal. More con-
restriction to pure sets is technically advantageous cisely,
[
and yields a more elegant theory. Moreover, it is V = Vα
possible to model traditional mathematical con- α

cepts such as real numbers using pure sets, so one


does not lose any mathematical power. Pure sets
are built from nothing, i.e., the empty set, by suc- 3.1 The Axioms of ZF C
cessively applying the “set of” operation. A simple The ZF C axioms, stated informally, are the fo-
example is {∅, {∅, {∅}}}: to build this we start by llowing:
forming {∅}, then {∅, {∅}}, and putting these two
sets together gives us {∅, {∅, {∅}}}. Thus, at ev- 1. Extensionality. If two sets have the same
ery stage we form all the sets whose elements are elements, they are equal.
6

2. Power Set. For every set x there is a set and the “cumulative hierarchy of Vα ’s” need not
P(x) whose elements are all the subsets of x. appear in the formulation of the axioms of ZF C.
In ZF C, the ordinals are identified, according
3. Infinity. There is an infinite set.
to the definition given by John Von Neumann,
4. Replacement. If x is a set and f is a with the set of their predecessors, starting with
function-class 1 restricted to x, then there is 0 = ∅. Thus, 1 = {0}, . . . , n + 1 = n ∪ {n}, . . . , ω =
a set y = {f (z) : z ∈ x}. {0, 1, 2, 3, . . .}, α+1 = α∪{α}, etc. And the cardi-
nals are identified with the ordinals for which there
5. Union. For every set x, there is a set x is no bijection with any of their elements. Thus,
S
whose elements are all the elements of the ele- S ℵ0 = ω, ℵ1 = ω1 , . . . , ℵn = ωn , . . . , ℵω = ωω =
ments of x. n ℵn , etc. The existence and the basic properties
of transfinite ordinal and cardinal numbers can be
6. Regularity. Every set x belongs to Vα , for proved from the ZF C axioms. For example, one
some ordinal α. can prove that there is no largest cardinal, or that
2κ > κ for every cardinal κ.
7. Axiom of Choice (AC). For every set X of
pairwise-disjoint nonempty sets there is a set The axioms of ZF C lead a kind of double life.
that contains exactly one element from each On the one hand, they tell us the things we can
set in X. do, based on what we tend to do, with sets. In
this sense, ZF C is just like any other collection of
Usually a further axiom appears on this list, axioms for algebraic structures, e.g., the axioms for
called the Pairing axiom. It asserts that for any groups, or fields. So, in a similar manner as one
given two sets A and B, the set {A, B} exists. In studies abstract groups, i.e., algebraic structures
particular, {A} exists. Applying the Union axiom that satisfy the axioms for groups, one can also
to the set {A, B} one then gets the union A ∪ B study those structures that satisfy the axioms of
of A and B. But Pairing can be derived from the ZF C. These are called models of ZF C. Since, for
other axioms. Another important axiom that ap- reasons to be explained below, models of ZF C are
peared in Zermelo’s original list, one that is both not easy to come by, one is also interested in mod-
natural and very useful, is the axiom of Separa- els of fragments of ZF C. Thus, a model of some
tion. It states that for every set A and every de- fragment A of ZF C is just a pair hM, Ei, where
finable property P , the set of elements of A that M is a non-empty set and E is a binary relation
have the property P is also a set. But this axiom is on M so that all axioms of A are true when the
a consequence of the axiom of Replacement, and elements of M are interpreted as the sets and E
so there is no need to include this axiom in the is interpreted as the membership relation. For in-
list. Using the axiom of Separation one can eas- stance, hVω , ∈i is a model of all the axioms of ZF C
ily prove the existence of the empty set ∅, as well except Infinity, and hVω+ω , ∈i is a model of ZF C
as the intersection A ∩ B and difference A − B of except Replacement. But one can also look at a
any two sets A and B. The axiom of Regularity completely different relation E on a set M , and see
is also known as the axiom of Foundation and it whether it happens to satisfy some of the axioms of
is usually stated as follows: every non-empty set ZFC. For example, the pair hN, Ei, where E is the
X has an ∈-minimal element, i.e., an element that relation given by: mEn iff the m-th digit (counting
no element of X belongs to. In the presence of the from right to left) in the binary expansion of n is
other axioms the two formulations are equivalent. 1, is a model of ZF C minus Infinity.
We chose the formulation in terms of the Vα ’s to But on the other hand we can view the ZF C
stress the fact that this is a natural axiom based on axioms as telling us how to build up the hierar-
the construction of the universe of all sets. But it is chy of the Vα ’s. Indeed, axiom 1, Extensionality,
important to notice that the notions of “ordinal” just states that a set is something entirely deter-
1 A function-class can be thought of as a function that is mined by its elements. Axioms 2-5 are tailored to
given as a definition rather than an object that exists as a construct V . The Power Set axiom takes us from
set. The concept will be made precise in section 3.2. Vα to Vα+1 . The Infinity axiom allows the con-
3. THE UNIVERSE OF ALL SETS 7

struction to go into the transfinite. Indeed, in the the formula:


context of the other ZF C axioms, Infinity is equiv-
alent to the assertion that ω exists. Replacement ∀x∀y(∀z(z ∈ x ↔ z ∈ y) → x = y)
is used to continue the construction of V at limit
stages λ. To see this, consider the function defined which reads as: for every set x and every set y, if
by: F (x) = y if and only if x is an ordinal and every set z belongs to x if and only if it belongs to
y = Vx . Then the range of F restricted to λ con- y (i.e., if x and y have the same elements), then x
sists of all Vβ , β < λ, which by Replacement form and y are equal.
a set. Now by an application of the axiom of Union Conversely, any formula of the formal language
to this set one obtains Vλ . Finally, the axiom of can be interpreted as a sentence (in English) about
Regularity states that all sets are obtained in this sets, and it makes sense to ask whether the inter-
way, that is, the universe of all sets is precisely V . preted sentence is true or not. Usually, by ’true’
This rules out pathologies, such as sets that belong we mean ‘true in the universe V of all sets’, but
to themselves. The point is that for every set X it also makes sense to ask for the truth or falsity
there is a first α such that X ∈ Vα+1 . This α is of a formula in any structure of the form hM, Ei,
called the rank of X and it marks the stage of the where E is a binary relation on M . For exam-
cumulative hierarchy where X was formed. So, we ple, the formula ∀x∃y x ∈ y is true in all models
couldn’t possibly have X ∈ X, since all elements hM, Ei of ZF C, while the formula ∃x∀y y ∈ x is
of X must have a rank strictly smaller than the false (because of Regularity). Now, a formula may
rank of X. AC is equivalent, in the context of the have free variables, namely, variables that are not
other ZF C axioms, to the Well-Ordering Princi- under the scope of a quantifier. For instance, in
ple, i.e., to the assertion that every set A can be the formula ∃y y ∈ x the variable x is free. Given
well-ordered and, therefore, can be put into one- a formula with a free variable x, we can ask: what
to-one correspondence with an ordinal. are the solutions? That is, what are the sets a
that make the sentence true, if any? This is en-
tirely analogous to the case when we ask for the
3.2 Formulas and models solutions to a given algebraic equation with vari-
able x. Thus, for example, every non-empty set a
The ZF C axioms can be formalized using the is a solution to the formula ∃y y ∈ x. To keep track
language of first-order logic for sets, that is, for- of the free variables in a given formula ϕ, we usu-
mal logic with quantifiers that range over objects ally write ϕ(x, y, . . .) to indicate that the variables
(in our case the objects being sets), with equal- x, y, . . . are free in ϕ. And we write ϕ(a, b, . . .) to
ity, and with the binary relation symbol ∈ as express that the sets a, b, . . . are solutions to the
the unique non-logical symbol. ∈ represents, of formula ϕ(x, y, . . .). Notice that, as in the last ex-
course, the membership relation. More precisely, ample, the solutions to a given formula may not
the symbols of the formal language are: variables form a set. So, in general, instead of talking about
x, y, z, . . . , X, Y, Z, . . .; quantifiers ∀ (for all) and ∃ ‘the set of solutions to ϕ(x, y, . . .)’ we talk about
(there exists); logical connectives ¬ (not), ∧ (and), ‘the class defined by ϕ(x, y, . . .)’. Notice that if
∨ (or), → (if ..., then ...), and ↔ (if and only if); ϕ has n free variables, then the class defined by
equality =, and membership ∈, as well as paren- ϕ will have n-tuples as elements, if non-empty.
theses. The formulas of the language are built up This is fine, since n-tuples can also be regarded
from the atomic formulas x = y and x ∈ y by as sets (see the beginning of next section). If the
means of the logical connectives and the quanti- formula has more than one free variable, for in-
fiers. Namely, if ϕ and ψ are formulas, then so are stance, ϕ(x1 , . . . , xn ), with n > 1, then we may
¬ϕ, (ϕ ∧ ψ), (ϕ ∨ ψ), (ϕ → ψ), (ϕ ↔ ψ), ∀xϕ, and ‘fix the value’ of one of the variables, say xn , by
∃xϕ. Thus, the formulas are the formal counter- taking a fixed set a and look for the n − 1-tuples
part of the sentences in English (or in any other ha1 , . . . , an−1 i such that ha1 , . . . , an−1 , ai is a solu-
natural language) that talk only about sets and the tion of ϕ(x1 , . . . , xn ). This is usually referred to as
membership relation. For instance, in this formal ‘the class defined by ϕ(x1 , . . . , xn−1 , a)’, and a is
language the axiom of Extensionality is written as called a parameter. Notice that every set a is also
8

a class, namely, the class defined by the formula countable or having a certain cardinality, with re-
x ∈ a. spect to different models of ZF C is an important
One advantage of this formalization, but by no phenomenon which, even if a bit disconcerting at
means the only one, as we shall have the chance first, may be used to a great advantage in consis-
to see later on, is that it makes it possible to for- tency proofs (see Section 5 below).
mulate in a completely precise way the axiom of It is not difficult to see that all the axioms of
Replacement. Indeed, in order to specify what is ZF C are true in V , which is hardly surprising
meant by a definable function, we just say that since they were designed for that to happen. But
a definable function is a class function definable, the ZF C axioms may conceivably hold in some
possibly with parameters, by a formula of the for- smaller universes. That is, there may be some class
mal language. That is, there is a formula ϕ(x, y, w) M properly contained in V , or even some set M ,
and a set a such that ϕ(x, y, a) defines a class that and therefore by the Löwenheim-Skolem Theorem
is a function, namely, for all sets b, c, and d, if also some countable set M , which is a model of
ϕ(b, c, a) and ϕ(b, d, a) hold, then c = d. Thus, ZF C. As we shall see, while the existence of mod-
Replacement is not really an axiom, but an axiom els of ZF C cannot be proved in ZF C, the fact that
scheme, namely, a collection of infinitely-many ax- one can consistently assume they exist – provided
ioms, one for each formula of the formal language ZF C is consistent, of course – is of the greatest
with at least two free variables. A typical formal- importance for set theory.
ized instance of the axiom is:

∀w(∀x∀y∀z((ϕ(x, y, w) ∧ ϕ(x, z, w)) → y = z) → 4 Set theory and the foundation of


mathematics
∀X∃Y ∀y(y ∈ Y ↔ ∃x(x ∈ X ∧ ϕ(x, y, w))))
This says that for every set a, if ϕ(x, y, a) defines ZF C not only allows us to develop the theory of
a class function, then for every set A, there is a set transfinite numbers but also, being an extremely
B that is the range of the function restricted to A. general and abstract theory based on the funda-
An important consequence of the fact that ZF C mental notion of set, it allows us to develop all
can be formalized in first-order logic is that it is standard mathematics. This means that all mathe-
subject to the Löwenheim-Skolem Theorem. This matical objects may be viewed as sets, and all clas-
is a general theorem about first-order formal lan- sical mathematical theorems can be proved from
guages which, in the particular case of ZF C, says ZF C using the usual logical rules of proof.
that if ZF C has a model, then it has a countable Regarding mathematical objects as sets is com-
one. More precisely, given any model M = hM, Ei mon practice in modern mathematics. The na-
of ZF C, there is a model N of ZF C contained tural numbers are the finite ordinals, so each one
in M which is countable and satisfies exactly the of them may be identified with the set of its pre-
same sentences as M . At first, this may seem para- decessors. An ordered pair ha, bi is defined as the
doxical, for how can ZF C have a countable model set {{a}, {a, b}}. The binary relations on a set
if one can prove in ZF C that there are uncount- A are identified with sets of ordered pairs of ele-
able sets? Doesn’t the Theorem imply, in fact, that ments of A. For instance, the divisibility relation
there are no models of ZF C? Not quite. There is between natural numbers is identified with the set
nothing paradoxical in having a countable model of ordered pairs hm, ni of natural numbers such
N of ZF C and a set a in N such that the state- that m divides n. An n + 1-ary relation on A
ment saying ’a is uncountable’ is true in N . For is a set of n + 1-tuples of elements of A, i.e., a
’a is uncountable’ means ’There is no map from ω set of ordered pairs whose first component is a n-
onto a’. Such a map may exist in V , or in some tuple of elements of A and the second component
model M larger than N , but not in N . Thus, from is an element of A. The integers are constructed
the point of view of N , a is uncountable, but in as equivalence classes, i.e., equivalence sets, of or-
fact, i.e., from the point of view of V , a may be dered pairs of natural numbers, and the rationals
countable. Far from presenting a problem, the rel- as equivalence classes of ordered pairs of integers.
ativity of certain set-theoretic notions, like being Real numbers are just Dedekind cuts, which are
4. SET THEORY AND THE FOUNDATION OF MATHEMATICS 9

non-empty bounded sets of rationals closed under 4.1 Undecidable statements.


predecessors and with no greatest element. An n-
Let ϕ be a mathematical statement. We may ask
ary function on a set A is just a special kind of
if ϕ is true or not. But in mathematics the truth of
n + 1-ary relation on A, since such a function f
a mathematical statement is established by means
may be identified with the set of ordered n + 1-
of a proof from basic principles or axioms. Thus,
tuples hx1 , ..., xn , yi such that f (hx1 , ..., xn i) = y.
to establish the truth of ϕ a proof of it should
An algebraic structure, e.g., a group or a ring, is
be found. But, if such a proof doesn’t exist, how
a tuple of the form hA, R, S, . . . , f, g, . . .i, where A
can we know that? One way would be to find a
is a non-empty set, R, S, . . . are relations on A and
proof of the negation of ϕ. This would tell us that
f, g . . . are functions on A of various n-arities. A
there cannot be a proof of ϕ, unless mathematics is
topological space X is precisely a set with a topol-
inconsistent! But, what if there is no proof of the
ogy, which is a set of subsets of X closed under ar-
negation of ϕ either? Then ϕ would be undecidable
bitrary unions and finite intersections, and so on.
from the axioms.
By viewing them as sets, the existence of all usual
Kurt Gödel proved in 1931 his famous in-
mathematical objects: numbers, algebraic struc-
completeness theorems (see Kurt Gödel in sec-
tures, topological and vector spaces, smooth man-
tion V). The first one says that in every axiomatic
ifolds and dynamical systems, etc., can be proved
formal system that is consistent and rich enough
in ZF C.
to develop basic arithmetic there are undecidable
statements, that is, statements such that neither
As for mathematical theorems, take for instance
they nor their negations are provable in the system.
the classical Bolzano Theorem, which is a special
In particular, there are statements of the formal
case of the Intermediate Value Theorem found at
language of set theory that are neither provable
the beginning of any Analysis course. It states
nor disprovable from the ZF C axioms, supposing,
that if a < b are real numbers, f : [a, b] → R is
that is, that ZF C is consistent.
a continuous function, and f (a) < 0 < f (b), then
But is ZF C consistent? The statement that as-
there exists some c ∈ [a, b] such that f (c) = 0.
serts the consistency of ZF C, usually written as
We have just observed that all the mathemati-
CON (ZF C), is the translation into the language
cal objects that appear in this theorem may be
of set theory of:
viewed as sets. The existence and basic properties
of these objects can be formally proved from ZF C, 0 = 1 is not provable in ZFC
for example, the completeness of the real numbers.
Then, the usual proof of the theorem may be seen This is an arithmetical statement, because it as-
as a formal proof from ZF C, obtained by the ap- serts that the sequence of symbols 0 = 1 is not
plication of the deduction rules of first-order logic. the last step of any formal proof from ZF C, and
Of course, writing the complete proof using the formal proofs may be coded by finite sequences of
formal language is both laborious and inadvisable natural numbers. Gödel’s second incompleteness
for practical reasons, for the proof would not only theorem is even more surprising than the first. It
be very long but also intuitively incomprehensi- says that in any consistent axiomatic formal sys-
ble. It is important, however, to convince oneself tem that is rich enough to develop basic arithmetic,
that in principle this can be done. Indeed, the the arithmetical statement that asserts the consis-
fact that all standard mathematics can be formu- tency of the system cannot be proved. Thus, if
lated and developed within the axiomatic system ZF C is consistent, then its consistency can nei-
of ZF C makes meta-mathematics possible, that ther be proved nor disproved in ZF C.
is, the mathematical study of mathematics itself. ZF C is currently accepted as the standard for-
Questions like the possibility or impossibility of mal system in which to develop mathematics.
proving a mathematical statement make now per- Thus, the truth of a mathematical statement is
fect sense, for they become mathematical questions firmly established if its translation into the lan-
once we understand by mathematical statement a guage of set theory is provable in ZF C. But what
statement in the formal language of set theory, and about undecidable statements? Since ZF C em-
by proof a formal proof from ZF C. bodies all standard mathematical methods, the
10

fact that a given mathematical statement ϕ is un- 5 The Continuum Hypothesis


decidable in ZF C means that the truth or fal-
sity of ϕ cannot be established by means of usual Cantor’s Continuum Hypothesis (CH), first
mathematical practice. If all undecidable state- formulated in 1878, states that every infinite set of
ments were like CON (ZF C) this would probably real numbers is either countable or has the same
not be a cause of worry, since they seem not to cardinality as R. In ZF C, since AC implies that
directly affect the kind of mathematical problems every set, and in particular every infinite set of
that people are usually interested in. But for bet- real numbers, can be put into one-to-one and onto
ter or worse this is not so. As we will see, there correspondence with a cardinal number, one can
are many statements of mathematical interest that easily see that CH is equivalent to the assertion
are undecidable in ZF C. that the cardinality of R is ℵ1 , or equivalently, that
2ℵ0 = ℵ1 .
Solving CH was the first problem in Hilbert’s
famous list of 23 unsolved problems, and has been
How can it be possible to prove, mathematically,
one of the main driving forces for the development
that a given mathematical statement ϕ is unde-
of set theory. In spite of many attempts at proving
cidable in ZF C? It is possible because ZF C is
CH by Cantor himself and by many leading math-
a formal theory in a first-order language, and the
ematicians of the first third of the twentieth cen-
correctness of the rules of proof for first-order logic
tury, no major progress was made until, 60 years
tells us that in order to show that ϕ cannot be
after its formulation, Gödel was able to prove its
proved in ZFC it is enough to find a model of ZF C
consistency with ZF C.
in which ϕ is false. Recall that a model M of ZF C
is an ordered pair hM, Ei where M is a non-empty
set, E is a binary relation on M , and all axioms of 5.1 The constructible universe
ZF C are true in hM, Ei when the elements of M
In 1938, Gödel found a way to construct, start-
are interpreted as the sets and E is interpreted as
ing with a model M of ZF C, another model of
the membership relation. Therefore, if we can find
ZF C, contained in M , where CH holds, thereby
models M and N of ZF C with ϕ true in M and
proving the relative consistency of CH with ZF C.
false in N , we can conclude that ϕ is undecidable.
This model is known as the constructible universe
Unfortunately, a consequence of Gödel’s second in-
and is represented by the letter L. Since M is
completeness theorem is that it is not possible to
a model of ZF C, we may view M as the uni-
prove in ZF C the existence of a model of ZF C.
verse V of all sets. Then L is built inside M si-
This is because, by Gödel’s completeness theorem
milarly to the way in which we built V , but un-
for first-order logic, ZF C is consistent if and only if
like in the construction of V , where when pass-
it has a model. Thus, a proof of the undecidability
ing from Vα to Vα+1 one takes all subsets of Vα ,
of ϕ splits into two relative consistency proofs: a
to go from Lα to Lα+1 one only takes those sub-
proof that if ZF C is consistent, then so is ZF C
sets of Lα that are definable in Lα (with parame-
plus ϕ, and a proof that if ZF C is consistent, then
ters). That is, Lα+1 consists of all sets of the form
so is ZF C plus the negation of ϕ. That is, one as-
{a : a ∈ Lα and ϕ(a) holds in Lα }, where ϕ(x) is
sumes that there is a model M of ZF C and proves
a formula of the language of set theory (or, more
the existence of two models of ZF C: one where ϕ
precisely, a formula of M ’s version of the language
holds, and one where it fails. The assumption of
of set theory), that may mention elements of Lα .
the existence of a model of ZF C is harmless, for if
If λ is a limit ordinal, then Lλ is just the union of
there is no such model then ZF C must be incon-
all the Lα , α < λ, and L is the union of all the
sistent and therefore everything is provable.
Lα , α an ordinal. Of course, we can also build L
inside V . This is the real L, the universe of all
constructible sets.
One of the most surprising results of twentieth- One important observation is that to build L it is
century mathematics is that the Continuum Hy- not necessary to use AC, and so we do not require
pothesis is undecidable in ZF C. AC to hold in M . But once L is constructed it can
5. THE CONTINUUM HYPOTHESIS 11

be verified that AC, and indeed all ZF C, holds in be defined here, enable us to carry out construc-
L. The verification of AC is based on the fact that tions of uncountable mathematical structures by
every element of L is defined at some stage α, and induction on the ordinals so that the construc-
so it is uniquely determined by a formula and some tion does not break down at limit stages. Both ♦
ordinals. So, any sensible well-ordering of all the and  have been widely used to prove the consis-
formulas will naturally yield a well-ordering of L, tency of mathematical statements without having
and therefore of every set in L. This shows that to construct models every single time, for since,
if ZF (i.e., ZF C minus AC) is consistent, then as Jensen showed, these principles hold in L, so
so is ZF C and, therefore, by adding AC to the does any statement implied by them. Thus, they
ZF axioms no contradiction is introduced into the are ready-made tools for proving consistency with
system. This is very reassuring, for although AC ZF C without requiring any knowledge of con-
has many desirable consequences it also has some structible sets.
that at first sight can appear counterintuitive, such Another important outgrowth of constructibility
as the Banach-Tarski Paradox. is inner model theory. Given any set A it is possi-
That CH holds in L is due to the fact that in L ble to build the constructible closure of A, namely,
every real number appears at some countable stage the smallest model of ZF that contains all ordinals
of the construction, i.e., in some Lα , α countable in and A. This model is called L(A) and is built in
L. This can be shown as follows: if r is a real, say the same way as L, but instead of beginning with
an infinite binary sequence, then it belongs to some the empty set, one begins with the transitive clo-
Lβ that satisfies those axioms of ZF C that are sure of A, namely, A together with the elements
needed to build L. Using the Löwenheim-Skolem of A, the elements of elements of A, etc. Models
theorem for first-order logic inside L, we can find of this sort are examples of inner models, that is,
a countable, in the sense of L, subset X of Lβ that models of ZF that contain all the ordinals and all
contains r and satisfies exactly the same sentences the elements of their elements. Especially promi-
as Lβ . But then X must be isomorphic to some nent are the inner models L(r), where r is a real
Lα , with an isomorphism that is the identity on the number, and L(R), the constructible closure of the
natural numbers, and therefore on r. And since Lα set of real numbers. Very important also are the
is countable in L, so is α. But since there are only inner models of large cardinal axioms, which will
ℵ1 -many countable ordinals, and for each ordinal α be discussed in section 6 below.
that is countable in L, Lα is also countable, there After the result of Gödel, and given the repeated
can only be ℵ1 -many real numbers in L. failed attempts to prove CH in ZF C, the idea
started to take shape that maybe it was undecida-
Since for each ordinal α, Lα contains only the ble. To prove this, it was necessary to find a way to
sets that are strictly necessary, namely, those that build a model of ZF C in which CH is false. This
were explicitly definable in one of the previous was finally accomplished 25 years later, in 1963, by
stages, L is the smallest possible model of ZF C Paul J. Cohen using a revolutionary new technique
containing all the ordinals, and in it the cardinal- called forcing.
ity of R is also the smallest possible, namely ℵ1 . In
fact, in L the Generalized Continuum Hypothesis
(GCH) holds, that is, for every ordinal α, 2ℵα has 5.2 Forcing
the smallest possible value, namely, ℵα+1 . The forcing technique is an extremely flexible and
The theory of constructible sets went through an powerful tool for building models of ZF C. It al-
extraordinary development in the hands of Ronald lows one to construct models with the most di-
Jensen [22]. He showed that in L Suslin’s Hypo- verse properties and with great control over the
thesis fails (see section 10 below) and isolated the statements that will hold in the model being con-
combinatorial principle known as ♦ (diamond). structed, thereby allowing one to prove the consis-
Moreover, using his fine-structural analysis of the tency with, and the undecidability from, ZF C of
constructible sets, he proved that another impor- many mathematical statements.
tant combinatorial principle, known as  (square), In a manner reminiscent of the way one passes
holds in L. These two principles, which will not from a field K to an algebraic extension K[a], one
12

goes from a model M of ZF C to a forcing ex- of large cardinals (see section 6 below). Since we
tension M [G] that is also a model of ZF C. How- want to build a model M [c] of ZF C that contains
ever, in spite of the analogy, the forcing method is, a new infinite binary sequence c and all the ele-
both conceptually and technically, far more com- ments of M , M [c] will have to contain Lα (c), i.e.,
plex, involving set-theoretic, combinatorial, topo- all sets that can be constructed in less than α steps
logical, logical and metamathematical aspects. starting with c. If we take M [c] = Lα (c), then c
To give an idea of how it works, let us con- might unfortunately code α by encoding a binary
sider Cohen’s original problem of obtaining, start- relation on the natural numbers order-isomorphic
ing from a model M of ZF C, a model where CH to α, so that α ∈ M [c]. Thus, if M [c] were a
fails. The only thing we know about M is that it model of ZF C, then Lα (c) would belong to itself,
is a model of ZF C, and so as far as we know CH which is impossible. If we try to circumvent the
may hold in it. In fact, for all we can tell M may problem by adding more sets to M [c] so that it
just be L, in the sense that when we build L inside becomes a model of ZF C, then we may end up
M we obtain the whole of M . Therefore, we need with M [c] = Lγ , for some ordinal γ greater than
to add to M new real numbers to ensure that in the α. And this is not good for our purposes, for in all
extension M [G] there will be at least ℵ2 -many of models of ZF C of the form Lγ , CH holds. The
them; more precisely, we need the extended model conclusion is that we cannot just pick any c that is
to satisfy the sentence that says that there are at not in M . We will have to choose c very carefully.
least ℵ2 -many real numbers. However, the ‘real The idea is that c should be generic, it the sense
numbers’ in M [G] may not be real numbers at all, that it should have no special property that sin-
and the element of M [G] that plays the rôle of ℵ2 gles it out. The reason for this is that if, as before,
need not be the real cardinal ℵ2 in V . M = Lα , and we want to ensure that M [c] = Lα (c)
To simplify the problem, suppose first that we is still a model of ZF C, then we don’t want c to
only want to add to M a single real number and, to have any special property that would interfere in
simplify it even further, think of the real number the construction of M [c] so that some ZF C ax-
to be added as just a binary representation of a iom would no longer hold. To accomplish this we
real in [0, 1], that is, an infinite binary sequence in build c little by little, i.e., by finite approxima-
the real world V . tions, while at the same time taking care of all
A first difficulty is that M may already contain possible properties that c could possibly have in
all infinite binary sequences, so there aren’t any to M [c]. More precisely, suppose ϕ(x) is a formula
be added. Fortunately, by the Löwenheim-Skolem with free variable x and p is a finite binary se-
Theorem, every model M of ZF C has a countable quence. Think of ϕ(x) as a property of the infinite
N ⊆ M that satisfies exactly the same sentences of binary sequence x and p as a finite approximation
the language of set theory as M . Let us emphasize to c. If we can extend p to a longer finite sequence
that N is countable in the real world, that is, in V ; q so that in M no infinite binary sequence a that
so there is, outside N , a function that enumerates begins with q has the property ϕ(x), i.e., ϕ(a) is
all its elements. Since M was a model of ZF C, false in M , then we do it. In this way, since c will
so is N . So, since we really don’t care about the extend q, we have a good chance, though by no
size of M , but only that it is a model of ZF C, means a certitude, that c will not have the prop-
we may as well assume that M is countable. Now, erty ϕ(x) in M [c]. Of course, things are actually
since there are uncountably many infinite binary more complicated. For example, when adding c
sequences, there are plenty of them that do not we also add many other new sets, e.g., the binary
belong to M . sequence obtained by switching zeros and ones in
c, so besides the properties of c one also needs to
So, can we just pick any one of them and add
keep track of all the properties these other new sets
it to M ? Well, no. For suppose M is of the form
might have in M [c]. Let us now see in some detail
Lα , as constructed in V , where α is a countable
how it really works.
ordinal in V . For all we know, M may be just
that. Indeed, the existence of models of ZF C of Let us begin by viewing finite binary sequences
this form follows, for instance, from the existence as approximations to the infinite sequence we want
5. THE CONTINUUM HYPOTHESIS 13

M (with parameters) and dense subsets of P that


belong to M .
Now suppose that c is a branch of P that in-
tersects all dense subsets of P that belong to M ,
that is, in every such subset there is some p that
is an initial segment of c. Such a c is called
generic or Cohen over M . We can easily see that
c cannot belong to M : if it did, then the set
D = {p : p is not an initial segment of c} would
be dense, and it would belong to M ; therefore,
c would have to intersect D, which is impossible.
But c being generic over M not only guarantees
that it is not in M , but also that it will not have
any avoidable property that is definable in M with
parameters.
Are there any generic branches over M ? Yes,
because M is countable. To find one, fix an enu-
meration hDn : n ∈ Ni of all dense subsets of P
Figure 3: The complete binary tree P and a generic
that belong to M . Then let p0 be any element of
branch.
D0 , and given pn in Dn pick pn+1 in Dn+1 such
that pn ≤ pn+1 . If we take c to be the union of all
pn , then c is generic over M .
to add, and consider the partial ordering P in M Now let M [c] be the set of all sets that can be
consisting of all finite binary sequences ordered by constructed, using c and the elements of M as pa-
extension: p ≤ q if and only if q extends p. Observe rameters, in as many steps as the ordinals of M .
that P is a tree, that is, for every q, the set of all p For instance, if M were of the form Lα , then M [c]
less than q is linearly ordered. We may then look would just be Lα (c). The model M [c] is called a
at infinite binary sequences as branches of this tree P-generic extension of M .
(see Figure 3). It turns out that, miraculously, M [c] is a model
Let us call a property of the branches of P avoid- of ZF C. Moreover, it has the same ordinals as M
able if given any p ∈ P, we can always find a q ∈ P and, therefore, it is not of the form Lγ , for any ordi-
extending p so that every branch that extends q nal γ. In particular, when we build L inside M [c],
does not have the property in question. For in- c does not belong to it. To prove that the ZF C
stance, the property: all terms in the sequence are axioms hold in M [c], Cohen introduces the rela-
zero is avoidable, while the property: there are ten tion p forces ϕ, written p ϕ, between elements
consecutive ones in the sequence is not avoidable. of P and formulas of the language of set theory
If we endow P with the order topology, that is, the augmented with names for the elements of of the
topology generated by the cones {q : p ≤ q}, it is P-generic extension. The point of using names is
easy to see that if a property is avoidable, then the that the truth or falsity in M [c] of a given formula
set of all q such that every branch that extends q depends not only on the sets in M and, of course,
does not have the property in question is a dense on c itself, but also on the sets that can be con-
(and open) set. Thus, every avoidable property structed from sets in M together with c. Since the
that is defined (with parameters) in M determines elements of M [c] can be constructed, using c and
a dense subset of P that belongs to M . Conversely, the elements of M as parameters, in ≤ α steps,
if D is a dense subset of P in M , then the property: where α is an ordinal in M , one defines in M , for
none of the initial segments of the sequence belongs each ordinal α, the class of names of rank ≤ α.
to D is avoidable and definable in M with D as a The idea is that if τ is a name of rank ≤ α, then
parameter. So, there is an exact correspondence the set in M [c] named by τ will be constructed
between avoidable properties that are definable in from τ and c in ≤ α steps and, conversely, all sets
14

in M [c] constructed in ≤ α steps can actually be α’s, pα is empty. Arguing as before, we can find
constructed from c and some name of rank ≤ α. A a sequence hcβ : β < ℵM 2 i of generic branches over
typical name is a set τ of pairs hp, σi, where p ∈ P M , one for each copy of P, which are all different.
and σ is a name of lower rank than τ . Then the But before we can conclude that in the resulting
element of M [c] named by τ , also called the inter- model M [hcβ : β < ℵM 2 i] the cardinality of R is
pretation of τ by c and denoted by ic (τ ), is just ℵM2 and therefore CH fails, there is a small point
the set of all ic (σ) such that hp, σi ∈ τ , for some that needs to be addressed. We have to make sure
initial segment p of c. To see a concrete example, that ℵ2 is preserved, i.e., when we add the sequence
let τ be the set of all pairs hp, qi, where p ∈ P and q hcβ : β < ℵM2 i to M , the ℵ2 of the expanded model
is obtained from p by switching 0’s and 1’s. Then M [hcβ : β < ℵM2 i], that is, the ordinal that in this
ic (τ ) is the set of all finite binary sequences ob- model is the second uncountable cardinal, is pre-
tained by switching the 0’s and 1’s on some initial cisely ℵM
2 . Otherwise, if CH held in M , it would
segment of c. Notice that ic (τ ) is not in M , since continue to hold in M [hcβ : β < ℵM 2 i] and our
c can be easily obtained from it. One can prove work would have been wasted. Fortunately, thanks
that, as desired, M [c] is precisely the set N of all to the topological properties of the product of ℵM 2
interpretations ic (τ ) of names τ ∈ M , although copies of P, ℵ2 is preserved, although checking this
this is by no means trivial. It follows from the fact requires, again, use of the forcing relation.
that N satisfies all ZF C axioms, and therefore so
The same kind of forcing argument allows one to
does M [c], but to prove this one needs to use the
construct models where the cardinality of R is ℵ3 ,
Forcing Theorem.
or ℵ27 , or any other cardinal of uncountable cofi-
The Forcing Theorem, proved by Cohen, asserts nality, i.e., any uncountable cardinal that is not the
that a formula ϕ, possibly with parameters, is true supremum (least upper bound) of countably many
in N , defined as the set of all interpretations ic (τ ) smaller cardinals. The cardinality of the contin-
of names τ ∈ M , if and only if there exists an uum is, therefore, undetermined by ZF C. In any
initial segment p of c that forces ϕ. More pre- case, since CH holds in Gödel’s constructible uni-
cisely, for all formulas ϕ(x1 , . . . , xn ) of the formal verse L and fails in the model constructed by Co-
language of set theory and all names τ1 , . . . , τn , hen using forcing, CH is undecidable in ZF C.
the formula ϕ(ic (τ1 ), . . . , ic (τn )) is true in N if and
We now turn to the independence of AC. Since
only if there is an initial segment p of c such that
AC holds in L, it only remains to prove the consis-
p ϕ(τ1 , . . . , τn ). Cohen also proves the crucial
tency, relative to ZF , of the failure of AC. So, as-
Definability Theorem, which says that the forc-
suming there is a model of ZF , one needs to build
ing relation is definable in M . Thus it is possible
a model of ZF where AC fails. Since AC is equiv-
to determine what formulas are true in M [c] just
alent to the Well-Ordering Principle, it is enough
by knowing the initial segments of c, for one only
to build a model of ZF in which there is a set of
needs to check that for some initial segment p of
reals that cannot be well-ordered. This was first
c the relation p forces ϕ holds in M to conclude
done by Cohen, using forcing, in the following way:
that ϕ is true in M [c].
let P be, as before, the partial ordering of all finite
Returning to the problem of building a model binary sequences. Given a countable model M of
where CH fails, we have to add to M at least ℵM 2 ZF , force over M with the product of ℵ0 copies of
generic branches over M , where ℵM 2 is the ordi- P to obtain a model M [hcn : n ∈ Ni] where each cn
nal that in M plays the role of ℵ2 , i.e., it is the is generic over M . Let N be the smallest submodel
second uncountable cardinal (of course, ℵM 2 need of M [hcn : n ∈ Ni] that contains all the ordinals
not be the real ℵ2 , as will happen, for instance, and the un-ordered set A = {cn : n ∈ N}. Thus, N
if M is of the form Lα , α a countable ordinal in is just L(A), as built inside M [hcn : n ∈ Ni]. One
V ). Adding ℵM2 generic branches can be done by can then show that N is a model of ZF , but that
working, instead of P, with the product, as defined in N there is no well-ordering of A. The reason is
in M , of ℵM2 copies of P. The elements of this that any well-ordering of A would be definable in
product are sequences hpα : α < ℵM 2 i of finite bi- L(A) with a finite number of ordinals and finitely-
nary sequences such that for all but finitely many many elements of A as parameters, and then each
6. LARGE CARDINALS 15

one of the cn would be in its turn definable by used it to prove the consistency of Suslin’s Hypoth-
indicating its ordinal position in the well-ordering. esis (see Section 10 below); Adrian R. D. Mathias
But since the whole sequence of cn ’s is generic over [32] proved the consistency of the infinitary form of
L, no formula can distinguish one of the cn ’s from the Ramsey Theorem; Saharon Shelah [37] proved
another, unless they appear as parameters in the the undecidability of the Whitehead Problem in
formula. So, if we choose cn that does not appear group theory; Richard Laver [28] proved the con-
as a parameter in the definition of the well-ordering sistency of the Borel Conjecture; to cite just a few
of A, any definition of the ordinal position of cn in remarkable examples from the seventies.
the well-ordering will also be satisfied by some cm , The forcing technique now pervades all of set
m > n, and so cn and cm would have to be equal. theory. It continues to be a research area of
But they are not. great interest, very sophisticated from the tech-
The forcing technique is much more general than nical point of view and of great beauty. It keeps
the previous examples might lead one to think. In- producing important results, with applications in
deed, by forcing it is possible to add to a countable most areas of mathematics (see some examples in
model M not only real numbers, but almost any [5], [9], and in the references below). Especially
other kind of set. For this, one has to design an influential has been the development over the last
appropriate forcing notion in M , i.e., a partial or- 25 years of the theory of Proper Forcing, intro-
dering P in M , so that if G is a generic subset of P duced by Shelah ([36]). Proper forcing has been
over M , then the model M [G] resulting from the very useful in the context of forcing iterations and
addition of G to M contains the required set. In their applications to the analysis of the cardinal
this general setting, a generic subset of P over M is characteristics of the continuum (see [2] for a sur-
a subset of P whose elements are compatible (they vey), as well as in applications to general topology
better be, since they are the bits that will make and combinatorics (see [3] and [45]). Other impor-
up the set to be added to M !) and intersects all tant developments have been the formulation and
dense subsets of P that belong to M . As before, study of forcing axioms like the Proper Forcing Ax-
the properties of the model M [G] are forced by the iom (see section 10 below), the use of class forcing
elements of P, and the crucial notion is the forcing for coding the universe into a single real number
relation p ` ϕ(τ1 , ..., τn ), where p is an element of P (see [16]), or the invention by W. Hugh Woodin of
and ϕ(τ1 , ..., τn ) is a formula of the language of set new powerful forcing notions associated to the the-
theory expanded with names τ1 , ..., τn for elements ory of large cardinals, like Pmax and the Stationary
of M [G]. Tower (see [27] and [46]).
The large number of independence results ob-
Immediately after Cohen’s proof of the indepen- tained by forcing made explicit the insufficiency of
dence of AC from ZF and of CH from ZF C, ZF C to answer many fundamental mathematical
a result for which he got the Fields Medal in questions. Thus, it became desirable to find new
1966, many set theorists started developing the axioms that, once added to ZF C, would provide a
forcing technique in its full generality (notably solution to some of those questions.
Azriel Lévy, Dana Scott, Joseph R. Shoenfield,
and Robert M. Solovay) and began to apply it to
other well-known mathematical problems. For in- 6 Large cardinals
stance, Solovay [43] constructed a model of ZF in
which every set of real numbers is Lebesgue mea- As we have already seen, the collection of all ordi-
surable, thereby showing that AC is necessary for nal numbers cannot form a set. But if it did, to it
the existence of non-measurable sets. He also con- would correspond an ordinal number, call it κ, that
structed a model of ZF C where every definable would coincide with the κ-th cardinal ℵκ . More-
set of real numbers is Lebesgue measurable, and so over, Vκ would be a model of ZF C. We cannot
non-measurable sets, although they can be proved prove in ZF C that there is a κ with these proper-
to exist (see the example in 6.1 below), cannot be ties, for then we would have proved in ZF C that
explicitly given; Solovay and Stanley Tennenbaum ZF C has a model, which is impossible by Gödel’s
[41] developed the theory of iterated forcing and second incompleteness theorem. So, why don’t we
16

add to ZF C the axiom that says that there is a 6.1 Measurable cardinals
cardinal κ such that Vκ is a model of ZFC?
Recall that a set of real numbers is Borel if it can
be obtained in countably-many steps starting from
This axiom, with the further requirement that
the open intervals and applying the two operations
κ be regular, that is, not the limit of less than κ-
of taking complements and countable unions. Re-
many smaller cardinals, was proposed in 1930 by
call also that a set A of real numbers is null, or has
Waclaw Sierpiński and Alfred Tarski, and it is the
measure zero, if for every ε > 0 there is a sequence
first of the large cardinal axioms. A cardinal κ S
with those properties is called inaccessible. P intervals I0 , I1 , I2 , . . . such that A ⊆ n In
of open
and n |In | < ε. A is Lebesgue measurable if it is
almost a Borel set, that is, if it differs from a Borel
Other notions of large cardinals, implying inac- set by a null set. To each measurable set A corre-
cessibility, kept appearing along the 20th century. sponds a number µ(A) ∈ [0, ∞], its measure, that
Some of them originated in generalizations to the is invariant under translation of A and is count-
uncountable of the Ramsey Theorem in its infinite ably additive, that is, the measure of a countable
version, which says that if each (unordered) pair of union of measurable pairwise disjoint sets is the
elements of ω (i.e., of natural numbers) is painted sum of their measures. Moreover, the measure of
either red or blue, then there is an infinite sub- an interval is its length.
set X of ω such that all pairs of elements of X On can prove in ZF C that there exist non-
have the same color. The natural generalization Lebesgue-measurable sets of real numbers, for ex-
of the theorem to ω1 turns out to be false. How- ample the set discovered in 1905 by Giuseppe Vi-
ever, on the positive side, Paul Erdős and Richard tali: let A be a subset of [0, 1] that is maximal
Rado ([7]) proved that for every cardinal κ > 2ℵ0 , with respect to the property that for every dis-
if each pair of elements of κ is painted either red tinct x, y ∈ A, x − y is not rational. The ex-
or blue, then there is a subset X of κ of size ω1 istence of A is guaranteed by AC. To see that
such that all pairs of elements of X have the same A is not measurable, consider for each rational p,
color. This is one of the landmark results of the the set Ap = {x + p : x ∈ A}. All these sets
partition calculus, an important area of combina- are pairwise disjoint. Let B be the union of all
torial set theory developed mainly by the Hun- Ap , where p is a rational number of the interval
garian school, led by Erdős and András Hajnal. [−1, 1]. A cannot have measure zero, for then B
The problem of whether the Ramsey Theorem can itself would have measure zero, and this is impos-
be generalized to some uncountable cardinal leads sible because [0, 1] ⊆ B. On the other hand, A
naturally to weakly-compact cardinals: a cardinal cannot have positive measure either, since then B
κ is weakly compact if it is uncountable and when- would have infinite measure, and this is impossible
ever all pairs of elements of κ are painted either because B ⊆ [−1, 2].
red or blue, there is a subset X of κ of size κ Trivially, all Borel sets are measurable. And in
such that all pairs of elements of X have the same 1905 Lebesgue showed that there are measurable
color. Weakly-compact cardinals are inaccessible sets that are not Borel. While reading Lebesgue’s
and, therefore, their existence cannot be proved work, Mikhail Suslin noticed that Lebesgue had
in ZF C. Moreover, it turns out that below the made a mistake in claiming that continuous images
first weakly-compact cardinal, assuming it exists, of Borel sets are Borel. Indeed, Suslin soon found
there are many inaccessible cardinals, and so the a counterexample, which led eventually to the dis-
existence of a weakly-compact cardinal cannot be covery of a new natural hierarchy of sets of reals
proved even assuming the existence of inaccessible beyond the Borel sets, the so-called projective sets.
cardinals. These are the sets that can be obtained from the
Borel sets by taking continuous images and com-
The most important large cardinals, the mea- plements (see section 9 below). In 1917 Nikolai N.
surable cardinals, are much larger than the weakly- Luzin showed that all continuous images of Borel
compact, and were discovered in 1930 by Stanislaw sets, the analytic sets, are also measurable. If a set
M. Ulam. is measurable, then so is its complement, hence all
6. LARGE CARDINALS 17

complements of analytic sets, the coanalytic sets,


are also Lebesgue measurable. The natural ques- be L, that is, there are non-constructible sets, even
tion is then if the continuous images of coanalytic non-constructible real numbers. In fact, if there is
sets, the so-called Σ12 sets, are measurable. But the a measurable cardinal, then V is much larger than
answer is undecidable in ZF C. In L it turns out L. For instance, the first uncountable cardinal, ℵ1 ,
that there are Σ12 sets that are not Lebesgue mea- is an inaccessible cardinal in L.
surable, and with forcing one can construct models After the invention of forcing and the subsequent
where all Σ12 sets are measurable. avalanche of independence results, the hope arose
that axioms asserting the existence of large cardi-
The proof given above of the existence of a non- nals, like measurable cardinals, would settle some
Lebesgue measurable set of reals hinges on the of the questions that, thanks to the forcing tech-
fact that Lebesgue measure is translation invari- nique, had been proved undecidable in ZF C. It
ant. In fact, the proof shows that there cannot was shown soon, however, by Lévy and Solovay,
be any countably additive translation invariant that large cardinal axioms could not settle CH, as
measure that extends Lebesgue measure and mea- one could easily change by forcing the cardinality
sures all sets of reals. Thus, a natural question, of the continuum and make CH hold or fail with-
known as the measure problem, is whether there out destroying the large cardinals. But Solovay
can exist some countably additive measure extend- proved in 1969 that, surprisingly, if there exists a
ing Lebesgue measure and measuring all sets of measurable cardinal, then all Σ12 sets of real num-
reals. If such a measure exists, then the cardinal- bers are Lebesgue measurable. So, while the axiom
ity of the continuum cannot be ℵ1 , nor ℵ2 , nor that asserts the existence of a measurable cardinal
any ℵn with n < ω, etc. In fact, Ulam proved in cannot settle the size of the continuum, it has a
1930 that a positive solution to the measure prob- profound effect on its structure. It is indeed aston-
lem implies that the cardinality of R is extremely ishing that measurable cardinals, so far away from
large, namely, greater or equal than the least un- the sets of real numbers in the universe V , have
countable regular cardinal that is a limit of smaller such a strong influence on their basic properties.
cardinals. He also proved that the existence of a While the relationship between large cardinals and
nontrivial countably additive measure on any set the structure of the continuum is not yet fully un-
implies either a positive solution to the measure derstood, great progress has been made in the last
problem, or that there exists an uncountable car- 30 years through the work done in Descriptive Set
dinal κ with a (nontrivial) {0, 1}-valued κ-additive Theory and Determinacy, described in sections 8
measure that measures all its subsets. Such a car- and 9 below.
dinal is called measurable. If κ is measurable, then
it is weakly-compact, and therefore inaccessible. In
fact, the set of weakly-compact cardinals smaller 6.2 Even larger cardinals
than κ has measure one, and so κ is itself the κ-th
Curiously, the existence of a measurable cardinal
weakly-compact cardinal. It follows that the exis-
is equivalent to the existence of an elementary em-
tence of a measurable cardinal cannot be proved in
bedding (different from the identity) of the uni-
ZF C, even if one adds the axiom that inaccessi-
verse V into a subuniverse M that contains all the
ble, or weakly-compact, cardinals exist (unless, of
ordinals and is transitive, meaning that it contains
course, ZF C plus the existence of such cardinals
all elements of its elements. An elementary embed-
is inconsistent!). A complete clarification of the
ding is a function j : V → M that respects truth,
measure problem was finally provided by Solovay
namely, if x is a set and P is a property definable
[42] by showing that if the solution is positive, then
in the formal language of set theory, then x has
there is an inner model with a measurable cardinal.
property P in V if and only if j(x) has property
On the other hand, if there is a measurable cardi-
P in M . For instance, if κ is a cardinal, then j(κ)
nal, then one can build a forcing extension where
is a cardinal in M (see Figure 4). In particular, V
the measure problem has a positive solution.
and M satisfy the same sentences of the language
An unexpected consequence of the existence of a of set theory, and therefore all the axioms of ZF C
measurable cardinal is that the universe V cannot are true in M . The measurable cardinal corre-
18

and are transitive (i.e., they contain all elements


of their elements), and in which certain large car-
dinals exist. The larger the cardinal, the more dif-
ficult it is to build the model. This work is known
as the inner model programme (see [48]).

One of the striking consequences of the inner


model program is that the consistency strength
of virtually any set-theoretic statement ϕ can be
measured by large cardinal axioms. That is, there
are large cardinal axioms A1 and A2 such that the
consistency of ZFC plus ϕ implies that of ZFC plus
A1 and is implied by the consistency of ZFC plus
A2 . We refer to A1 as a lower bound for the con-
sistency of ϕ and A2 as an upper bound. In the
fortunate cases when the lower and upper bound
Figure 4: The elementary embedding arising from a coincide, we obtain an exact measure of the con-
measurable cardinal κ. sistency strength of ϕ. An upper bound A2 is usu-
ally obtained by forcing over a model of ZFC plus
A2 , whereas a lower bound A1 is obtained by in-
sponding to this embedding is the so-called critical
ner model theory. In section 6.1 we saw that the
point of the embedding, that is, the first ordinal κ
consistency strength of a positive solution to the
such that j(κ) 6= κ. On the other hand, an elemen-
measure problem is exactly that of the existence
tary embedding can be obtained from a two-valued
of a measurable cardinal. We shall see another im-
κ-additive measure U on a measurable cardinal κ
portant example in the next section.
by building M using a model-theoretic construc-
tion known as the κ-ultrapower of V modulo U. A
famous result of Kenneth Kunen from 1971 says Knowing upper and lower bounds for the consis-
that there cannot be any non-trivial elementary tency strength of set-theoretic statements, or even
embedding of V into V . But by demanding that better, knowing their exact consistency strength,
M is very close to V one obtains ever stronger large is extremely useful for comparing them. Indeed, if
cardinal axioms. For instance, if we demand that the lower bound for a sentence ϕ is greater than
for each ordinal α there is an elementary embed- the upper bound for another sentence ψ, then we
ding j : V → M with critical point κ and Vα ⊆ M , can conclude, via Gödel’s incompleteness, that ψ
then κ is a so-called strong cardinal. If we demand does not imply ϕ.
that for every α there is an elementary embed-
ding j : V → M with critical point κ and all α- The construction of canonical inner models for
sequences of elements of M contained in M , then large cardinals has now reached the level of Woodin
κ is what is known as a supercompact cardinal, etc. cardinals ([34]). Woodin cardinals lie in the large
The existence of a supercompact cardinal implies cardinal hierarchy between strong cardinals and
the existence of many strong cardinals below it, supercompact cardinals. These inner models have
and the existence of a strong cardinal implies the structural properties similar to L. For example,
existence of many measurable cardinals below it. they satisfy the covering property (see next sec-
For more on large cardinals, see [23]. tion), and Jensen’s combinatorial principle  holds
Some of the deepest and most technically diffi- in them ([35]). The techniques developed for the
cult work in set theory is currently devoted to the construction of the models and for the verifica-
construction and analysis of canonical inner mo- tion of their structural properties have made pos-
dels for large cardinals. These are analogs of L for sible to establish lower bounds on the consistency
large cardinals, that is, they are models built in strength of an astounding number of set-theoretic
some canonical way that contain all the ordinals statements.
7. CARDINAL ARITHMETIC 19

7 Cardinal arithmetic below κ, then it also holds on κ. That this is also


the case for singular cardinals of countable cofinal-
Beyond the Continuum Hypothesis, the behavior ity is a consequence of the Singular Cardinal Hypo-
of the exponential function 2κ for arbitrary infi- thesis (SCH), a general principle weaker than the
nite cardinals κ has been a motivating force in setGCH that completely determines singular cardinal
theory. Cantor had proved that 2κ > κ for all κ, exponentiation, relative to exponentiation for reg-
and Dénes König that the cofinality of 2κ is al- ular cardinals. A special case of SCH is: Suppose
ways greater than κ, that is, 2κ is not the limit that 2ℵn < ℵω for all finite n. Then 2ℵω = ℵω+1 .
of ≤ κ-many smaller cardinals. The GCH, which, So, in particular, if the GCH holds below ℵω , then
as we saw, holds in the constructible universe L, it must hold at ℵω . Shelah used his powerful pcf
states precisely that 2κ has the least possible value,
theory ([38]) to obtain the unexpected result that
namely, the least cardinal greater than κ, usually if 2ℵn < ℵω for all n, then 2ℵω < ℵω4 . So, if
denoted by κ+ . One might think that, as in the GCH holds below ℵω , then there is a bound (in
case of 2ℵ0 , by forcing it should be possible to build
ZF C!) on the possible values of 2ℵω . But can this
models of ZF C where 2κ takes any prescribed value actually be greater than the least possible
value, subject only to the necessary requirement one, namely ℵω+1 ? In particular, can the GCH
that its cofinality be greater than κ. This is truefirst fail at ℵω ? The answer is yes, but large cardi-
for cardinals κ that are regular, that is, not the nals are needed. Indeed, on one hand Magidor [29]
limit of less than κ smaller cardinals. Indeed, proved the consistency of the first failure of GCH
William B. Easton [8] showed that for any function at ℵω , assuming the consistency of the existence
F on the regular cardinals such that κ ≤ λ im- of a supercompact cardinal. Thus, the existence
plies F (κ) ≤ F (λ) and F (κ) has cofinality greater
of a supercompact cardinal is an upper bound for
than κ, there is a forcing extension of L in which the failure of SCH. On the other hand, using in-
2κ = F (κ), for all regular κ. So, for instance, ner model theory, Anthony J. Dodd and Ronald
one can build a model of ZF C where 2ℵ0 = ℵ7 , B. Jensen (see [6]) showed that large cardinals are
2ℵ1 = ℵ20 , 2ℵ2 = ℵ20 , 2ℵ3 = ℵ101 , etc. This shows
required for this to happen: they built a model K,
that the behavior of the exponential function for called the core model for a measurable cardinal,
infinite regular cardinals is totally undetermined which satisfies the GCH, and with the property
in ZF C, and anything possible can be attained by that either there is an inner model (of V ) with
forcing. a measurable cardinal in it, or V has the cover-
But how about non-regular cardinals? Non- ing property relative to K, which means that V is
regular cardinals are called singular. Thus, an infi-
close to K in the sense that every uncountable set
nite cardinal κ is singular if it is the supremum of
of ordinals is covered by, i.e., is a subset of, a set
less than κ-many smaller cardinals. For instance, of the same cardinality which belongs to K. Now
ℵω , being the supremum of the ℵn , n ∈ N, is the if V has the covering property relative to K, then
first singular cardinal. Determining the possible it follows that 2ℵω = ℵω+1 . And similarly with ℵω
values of the exponential function at singular car-replaced by any singular cardinal. So we can con-
dinals is a very hard problem that has generated clude: If SCH fails, then there is an inner model
much important research and involves, quite sur- with a measurable cardinal, and therefore the exis-
prisingly, the necessary use of large cardinals. tence of a measurable cardinal is a lower bound for
Using a supercompact cardinal, Matthew Fore- the consistency strength of a failure of the SCH.
man and W. Hugh Woodin [13] built a model of An exact measure of the consistency strength of
ZF C in which GCH fails everywhere, i.e., 2κ > κ+ the failure of SCH was later established by Moti
for all cardinals κ. But curiously, the value of Gitik ([17]).
the exponential function at a singular cardinal of
uncountable cofinality is somehow determined by
its values at smaller regular cardinals. Indeed, in 8 Determinacy
1975, Jack H. Silver [40] proved that if κ is singu-
lar of uncountable cofinality and 2α = α+ for all It turns out that the existence of very large car-
α < κ, then 2κ = κ+ . That is, if the GCH holds dinals, such as Woodin or supercompact, has a
20

sets in the class have properties similar to those of


the Borel sets. Thus, the Axiom of Determinacy
(AD), which asserts that all sets of reals are deter-
mined, implies that every set of reals is Lebesgue
Figure 5: A run of the infinite game associated to a set measurable, has the property of Baire (i.e., it dif-
A ⊆ [0, 1]. fers from an open set by a set of first category), and
has the perfect set property (i.e., if uncountable,
it contains a perfect set). For instance, to show
that every set A of reals, i.e., of infinite binary se-
dramatic effect on the properties of sets of real quences, is Lebesgue measurable it is enough to
numbers, especially on those that can be defined show that if all measurable subsets of A are null,
in some simple way. then A itself must be null. And for this one plays,
The reason for this can be seen through the anal- for every  > 0, the covering game for A and .
ysis of certain infinite two-player games with per- In this game, player I plays so that the sequence
fect information associated to sets of real numbers. a = hn0 , n2 , n4 , . . .i is in A, and player II plays
Given A ⊆ [0, 1], or equivalently A being a collec- binary codes for finite unions of rational intervals,
tion of binary sequences, consider the following in- with measure converging to 0, while attempting to
finite game associated to A: there are two players, cover a. Now, it can be shown that the assumption
I and II, who alternately play some ni ∈ {0, 1}. that every measurable subset of A is null implies
I plays n0 , then II plays n1 , to which I answers that player I cannot have a winning strategy. So
by playing n2 , and so on. A run of the game is by AD there must be a winning strategy for II.
displayed in Figure 5. At the end of the run, the Using this strategy one can show that the outer
players have produced an infinite binary sequence: measure of A is at most . And since this works
n0 , n1 , n2 , . . .. Player I wins the game if the se- for all  > 0, A must be null.
quence n0 , n1 , n2 , . . . belongs to A, and II wins
otherwise. While AD rules out the existence of badly-
behaved sets of reals, it implies the negation of
For example, if A is the interval [0, 12 ], then
AC, and so AD is inconsistent with ZF C. How-
player I wins the game by starting playing 0.
ever, weaker versions of AD are compatible with,
Whereas if A = [0, 14 ), then player II wins the
and even follow from, ZF C. Indeed, Donald A.
game by playing 1 in her first move. But for most
Martin proved in 1975 that ZF C implies that ev-
games, the question of who wins is not decided af-
ery Borel set is determined. Moreover, if there
ter any finite number of moves. For instance, if A
exists a measurable cardinal, then every analytic
is the set of rational points of [0, 1], then one can
set, and therefore also every co-analytic set, is de-
easily see that player II has a strategy for winning
termined. A natural question is, therefore, if the
the game, but she will not win at any finite stage
existence of larger cardinals implies the determi-
of the run.
nacy of more complex sets, like the Σ12 sets, and
The game is determined if one of the two play-
more.
ers has a winning strategy. A winning strategy
for player II is a function f that assigns 0 or 1 The intimate connection between large cardinals
to each finite binary sequence in such a way that and the determinacy of simple sets of reals was
if II plays in accordance with f by always play- first made explicit by Leo Harrington [19], who
ing f (n0 , n1 , ..., n2k ) in the k-th turn, she wins the showed that the determinacy of all analytic sets
game independently of what I plays. Similarly for is in fact equivalent to a large-cardinal principle
a winning strategy for I. We say that the set A is slightly weaker than the existence of a measurable
determined if so is the game associated to A. One cardinal. As we shall shortly see, large cardinals
might guess that every game is determined, but imply the determinacy of certain simply definable
actually it is quite easy, using AC, to find a game sets of reals, the so-called projective sets, while
that is not determined. the determinacy of those sets implies in turn the
Now, the determinacy of the games associated existence of the same kind of large cardinals in
to certain classes of sets of reals implies that all some inner models.
9. PROJECTIVE SETS AND DESCRIPTIVE SET THEORY 21

9 Projective sets and descriptive set tain a perfect set. However, as we already pointed
theory out, one cannot prove in ZF C that all Σ12 sets have
those properties, since in L there are counterex-
amples. In contrast, if there exists a measurable
The study of the structure of definable sets of re- cardinal, then they do have them. But what about
als is the subject of descriptive set theory. Ex- more complex projective sets?
amples of such sets are the Borel sets and the
projective sets, namely, those that are obtained The theory of projective sets is closely tied to
from Borel sets by taking continuous images and large cardinals. On the one hand, Solovay [43]
complements. Equivalently, those that can be ob- showed that if the existence of an inaccessible car-
tained from some closed subset of Rn , some finite dinal is consistent, then so is that every projective
n, by taking projections and complements; hence set of reals is Lebesgue measurable, has the Baire
the name projective. Thus, every projective set is property, etc. On the other hand, Shelah showed,
definable from a closed subset of Rn , since comple- quite unexpectedly, that the inaccessible cardinal
ments correspond to negations and projections to is necessary, in the sense that if all Σ13 sets are
existential quantification. The analytic sets, being Lebesgue measurable, then ℵ1 is an inaccessible
the continuous images of Borel sets, are projective. cardinal in L.
And so are the complements of the analytic sets, Nearly all the classical properties of Borel and
the coanalytic, and the continuous images of co- analytic sets are also held by the projective sets
analytic sets, the Σ12 . More complex projective assuming that they are determined. So, since the
sets are obtained by taking complements of Σ12 determinacy of all projective sets cannot be proved
sets, the so-called Π12 sets, their continuous im- in ZF C and allows for the extension of the the-
ages, called Σ13 , etc. The projective sets form a ory of Borel and analytic sets to all projective sets
hierarchy of increasing complexity, in accordance in a very elegant and satisfactory way, it consti-
with the number of steps (always finite) that are tutes an excellent candidate as a new set-theoretic
necessary to obtain them, starting from the Borel axiom. This axiom is known as Projective Deter-
sets. Many sets of reals that appear naturally in minacy (P D). It implies, for instance, that ev-
usual mathematical practice are projective. More- ery projective set is Lebesgue measurable, has the
over, the results and techniques of descriptive set Baire property, and has the perfect set property.
theory, although originally developed for the study In particular, since every uncountable perfect set
of sets of reals, also apply to definable sets in any has the same cardinality as R, it implies that there
Polish space (a separable and complete-metrizable is no projective counterexample to CH.
space), like Rn , C, separable Banach spaces, etc., One of the most remarkable advances in set the-
where projective sets arise in a very natural way. ory over the last 20 years has been the proof that
For example, in the space C[0, 1] of continuous P D follows from the existence of large cardinals.
real-valued functions on [0, 1] with the sup norm, Donald A. Martin and John R. Steel proved in
the set of everywhere differentiable functions is co- 1988 (see [31]) that if there exist infinitely many
analytic, and the set of functions that satisfy the Woodin cardinals, then P D holds. Subsequently,
Mean-Value Theorem is Π12 . More examples can Woodin showed that, surprisingly, the hypothesis
be found in [24]. Thus, since descriptive set the- that for each n, it is consistent that there exist
ory deals with rather natural sets in Polish spaces n Woodin cardinals is necessary in order to ob-
of general mathematical interest, it is not surpris- tain the consistency of P D. Thus the existence
ing that it has found many applications in other of infinitely many Woodin cardinals is a sufficient,
areas of mathematics such as: harmonic analysis, and essentially necessary, assumption for extend-
group actions, ergodic theory, and dynamical sys- ing the classical theory of Borel and analytic sets
tems (see [25], [20], and [11]). to all projective sets of reals, and more generally
Classical results of descriptive set theory are to all projective sets in Polish spaces.
that all analytic sets, hence also all coanalytic sets, In spite of the enormous success of the known
are Lebesgue measurable and have the Baire pro- large cardinal axioms, not only in descriptive set
perty; and that all uncountable analytic sets con- theory but also in many other areas of mathe-
22

matics, their status as true axioms of set theory is countable –, and takes so-called direct limits at
is still a matter of debate. This is more so in the limit stages, then in ω2 steps one can destroy
the case of very large cardinals, like the supercom- all counterexamples so that SH holds in the final
pact, the reason being that there is yet no inner model. On the other hand, Jensen proved in 1968
model theory available for them, and therefore not that a counterexample to SH exists in L, thereby
even strong evidence for their consistency. How- proving the undecidability of SH in ZF C. In fact,
ever, it should be noted that, as Harvey Friedman Jensen showed that one can build a counterexam-
has shown (see [15]), large cardinals are necessary ple to SH using the ♦ principle.
even for proving quite simple-looking and rather From the construction of Solovay and Tennen-
natural statements about finite functions on the baum, Martin isolated a new principle now known
integers, which provides evidence for their essen- as Martin’s axiom (M A) that is a generalization
tial rôle even for the most basic parts of mathe- of the well-known Baire category theorem, which
matics. Another shortcoming of the known large states that in every compact Hausdorff topologi-
cardinal axioms is that they cannot decide some cal space, the intersection of countably-many open
fundamental questions. The most conspicuous is dense sets is not empty. M A says the following:
CH, but there are others. In every compact Hausdorff ccc topological space,
the intersection of ℵ1 -many open dense sets is not
empty.
10 Forcing axioms
The condition that the space be ccc (i.e., every
An old and basic question about the continuum collection of pairwise disjoint open sets is count-
that the known large cardinal axioms cannot solve able) is necessary, for without it the statement is
either is Suslin’s Hypothesis (SH). Cantor had false. It is easy to see that M A implies the nega-
proved that every linearly ordered set that is dense tion of CH, for if there are only ℵ1 real numbers,
(i.e., any two distinct elements have another ele- then the intersection of the ℵ1 -many dense open
ment in-between), complete (i.e., every non-empty sets R \ {r}, r a real number, is empty. However,
subset with an upper bound has a supremum), M A does not decide the cardinality of R, for since
separable (i.e., contains a dense countable subset), there is no restriction on how long the Solovay and
and without endpoints is order-isomorphic to the Tennenbaum iteration can be, and since new re-
real line. In 1920 Mikhail Y. Suslin conjectured als are always added along the iteration and all
that if instead of being separable one assumes the the cardinals are preserved, one can obtain models
weaker countable chain condition, or ccc, that is, of M A with the size of the continuum as large as
every pairwise disjoint collection of open intervals wanted, although it will always be a regular cardi-
is at most countable, then it is also isomorphic to nal. M A has been used with great success to solve
R. The importance of SH for the development of many questions that are undecidable in ZF C. For
set theory is that it led to the discovery of a new example, it implies SH and that every Σ12 set is
class of axioms, the so-called forcing axioms. Lebesgue measurable (see [14] for many more con-
Robert M. Solovay and Stanley Tennenbaum sequences and applications). But is M A really an
[41] constructed in 1967 by forcing a model where axiom? In what sense, if any, is it a natural, or at
SH holds. The idea is to destroy any counterex- least plausible, assumption about sets? Does the
amples to SH by forcing. But by destroying coun- fact that it decides many ZF C undecidable ques-
terexamples new ones may be created, and so one tions suffice for it being accepted on a par with the
needs to force again and again. The iteration of ZF C axioms, or the axioms of large cardinals? We
forcing is technically cumbersome and difficult to shall come back to this.
control, for many unwanted things may happen at M A has many different equivalent formulations.
the limit stages. For instance ω1 may be collapsed, Martin’s original one is the following:
i.e., it may become countable. But fortunately, if For every ccc partial ordering P and every family
one starts with a model where GCH holds, uses {Dα : α < ω1 } of dense subsets of P, there is G ⊆
only forcing notions that are ccc – i.e., partial or- P whose elements are pairwise compatible and such
derings in which every set of incompatible elements that G ∩ Dα 6= ∅, for all α < ω1 .
11. GENERIC ABSOLUTENESS 23

Thus, M A is just saying that for forcing notions which have been used with great success by Stevo
P that are ccc there exist generic subsets G of P Todorčević to solve many outstanding problems in
over models M of size ℵ1 . (Recall from section 5.2 general topology (e.g., the S space problem) and
that for every P there are always generic subsets infinite combinatorics (see [45]; also [3] and [10]).
G of P over countable models.) This explains why As we already pointed out, forcing axioms are,
M A is referred to as a forcing axiom. apparently, not as intuitively evident as the ZF C
Stronger forcing axioms can be obtained from axioms, or even the axioms of large cardinals, and
this last formulation of M A by expanding the class so the question is to what extent they should be
of partial orderings to which it applies. Unfortu- considered as true axioms of set theory rather
nately, we cannot allow all partial orderings, for than just useful principles for showing that certain
if P is the partial ordering of all functions with statements are consistent with ZF C. In the case
domain a finite subset of ω and values in ω1 , the of M A and some weaker forms of P F A and M M ,
existence of a pairwise compatible G intersecting some justification for their being taken as true ax-
all the dense sets Dα = {f ∈ P : α ∈ range(f ioms is based on the fact that they are equivalent
S)}, to principles of generic absoluteness. That is, they
α < ω1 , would make ω1 countable. Indeed, G
would be a function from ω onto ω1 . So, in order assert, under certain restrictions which are neces-
to obtain useful stronger forms of M A one needs sary to avoid inconsistency, that everything which
to relax the ccc requirement while keeping the ax- might exist, does exist. More precisely, if some set
iom consistent. An important such strengthening having certain properties could be forced to ex-
of M A is the Proper Forcing Axiom (P F A), which ist over V , then a set having the same properties
is formulated for partial orderings that are proper. already exists (in V ). So, like the axioms of large
Properness is a property weaker than the ccc that cardinals, they are maximality principles, i.e., they
was discovered by Shelah and is particularly use- attempt to make V as large as possible.
ful when working with complicated forcing itera-
tions. Indeed, proper forcing notions can be iter-
ated without collapsing ω1 , although unlike the ccc 11 Generic absoluteness
forcing iterations, one needs to take so-called in-
Given a set X, its transitive closure is the set
verse limits at limit stages of countable cofinality.
whose elements are X, the elements of X, the el-
The strongest possible forcing axiom of this type
ements of elements of X, etc. For an infinite car-
was discovered by Matthew Foreman, Menachem
dinal κ, let H(κ) be the set of all sets whose tran-
Magidor, and Saharon Shelah in 1988 ([12]). It is
sitive closure has cardinality less than κ. Thus,
called Martin’s Maximum (M M ) and is consistent
H(ω) = Vω , and H(ω1 ) is the set of all count-
with ZF C assuming the consistency of a super-
able sets whose elements, the elements of its ele-
compact cardinal. M M is formulated as above
ments, etc. are also countable. We always have
but for partial orderings that have a property
H(κ) ⊆ Vκ , but not the converse, as, for instance,
even weaker than properness called stationary-set
Vω1 is not contained in H(ω1 ). If κ is inaccessible,
preservation.
then H(κ) = Vκ . The sets H(κ) also form a cumu-
Both M M and P F A have striking conse- lative hierarchy, similar to the Vκ , and the union
quences. For example, P F A, and therefore also of all the H(κ) is V .
M M , implies the axiom of projective determinacy Forcing is normally used to construct exten-
P D, the singular cardinal hypothesis SCH, and sions of countable models of ZF C. Each extension
that the cardinality of R is ℵ2 . (See [3] and [12] depends on a partial ordering P, which is, typi-
for more consequences of P F A and M M , respec- cally, the partial ordering of approximations to the
tively.) generic set one wants to add. But it is also pos-
An advantage of forcing axioms is that they are sible to consider forcing extensions of V . Clearly,
readily applicable without having to go into the these extensions are ideal – i.e., they do not exist
details of forcing. A very good example of this is – since all sets are in V . But given a partial or-
P F A and some combinatorial principles derived dering P, one can define the Boolean-valued model
from it, like the Open Coloring Axiom (OCA), V P , which contains names for all the elements of
24

the ideal forcing extension, and is better thought of 12 Some recent advances
as the universe that would be obtained by forcing
with P over V , if that were possible. To finish, let us briefly mention some recent deep
There is always a certain degree of generic ab- results which establish strong underlying connec-
soluteness between V P and V . For instance, by tions between large cardinals, inner models, deter-
the so-called Levy-Shoenfield absoluteness theo- minacy, forcing axioms, generic absoluteness, and
rem, given an existential statement about sets in the continuum. Those results hold under the as-
H(ω1 ), if it holds in V P , for some P, then it also sumption that for every ordinal α there exists a
holds in V , i.e., it is true. That is, if a set X with Woodin cardinal greater than α.
certain properties that depend exclusively on sets The first result, due to Shelah and Woodin [39]
in its transitive closure and sets in H(ω1 ) could be is that the theory of L(R) is generically absolute.
forced to exist, then such a set X already exists. That is, all sentences with real numbers as pa-
Let us write this in a concise form as: For every rameters that would hold in the L(R) of any ideal
partial ordering P, generic extension V P of V , are already true in the
real L(R). This kind of generic absoluteness is suf-
H(ω1 ) ≺∃ V P . ficient for showing that all sets of reals in L(R),
and in particular the projective sets, are Lebesgue
This kind of generic absoluteness for H(ω1 ) is very measurable, have the Baire property, etc. Fur-
useful, especially in descriptive set theory, since the thermore, by refining the Martin-Steel result ([31])
projective sets of real numbers are precisely those that large cardinals imply P D, Woodin showed
that are definable in H(ω1 ) with parameters. But that in L(R) every set of reals is determined.
consider now the natural generalization to H(ω2 ), Another result of Woodin ([46]) is that there is
namely: Every existential statement about sets in an axiom, called (∗), which can be viewed as a gen-
H(ω2 ) that holds in V P , for some P, is true. i.e., eralization of BM M in the context of large cardi-
for every partial ordering P, nals, and is intended to play the role for H(ω2 )
that P D plays for H(ω1 ), in the sense that it de-
H(ω2 ) ≺∃ V P . cides ‘practically all’ questions about H(ω2 ). Of
course, no consistent axiom can decide ‘all’ ques-
Unfortunately, this generalization is inconsistent tions about H(ω2 ), since by Gödel’s incomplete-
with ZF C, unless we restrict the kind of forcing ness there will always be undecidable arithmetical
extensions. But as it turns out, M A is equiva- statements. So, to formulate precisely the notion
lent to H(ω2 ) ≺∃ V P for all ccc P. This char- of deciding practically all questions about H(ω2 ),
acterization of M A in terms of generic absolute- Woodin introduces a new logic, called Ω-logic (see
ness provides some justification for regarding M A [1] and [46]), which strengthens ordinary first-order
as a true axiom of set theory. The same ap- logic. A main feature of Ω-logic is that the valid
plies to the corresponding axioms for forcing ex- statements in Ω-logic are generically absolute. Un-
tensions with proper partial orderings and par- der suitable large cardinal hypotheses, (∗) is con-
tial orderings that are stationary-set preserving: sistent in Ω-logic and decides in Ω-logic the theory
the Bounded Proper Forcing Axiom (BP F A) is of H(ω2 ). The most important open problem is
H(ω2 ) ≺∃ V P for all P proper, and the Bounded the Ω-conjecture, whose formulation is quite tech-
Martin’s Maximum (BM M ) is H(ω2 ) ≺∃ V P for nical and beyond the scope of this article. If the
all P stationary-set preserving. Both BP F A and Ω-conjecture is true, then any axiom compatible
BM M , although weaker than P F A and M M , re- with the existence of large cardinals that decides
spectively, are strong enough to decide many ques- all questions about H(ω2 ) in Ω-logic must imply
tions that the large cardinal axioms are unable the negation of CH. Thus, the theories: ZF C
to settle. Indeed, as a culmination of a series plus CH, and ZF C plus the negation of CH are
of results by Woodin and David Asperó, in 2002 not equally reasonable from the point of view of
Todorčević [44] proved that BM M implies that Ω-logic, for in the presence of large cardinals, CH
the cardinality of R is ℵ2 . And recently, Justin T. puts an unnecessary limitation on the possibility
Moore [33] has shown that BP F A also implies it. of settling all natural questions about H(ω2 ).
13. FINAL REMARKS 25

13 Final remarks 3. Baumgartmer, J. E. 1984 Applications of the


Proper Forcing Axiom. Handbook of Set Theo-
In this short outline of set theory, we have re- retic Topology, K. Kunen and J. Vaughan, Edi-
viewed some of the key developments since its be- tors. Elsevier Science publishers B. V., 913-959.
ginnings in the late 19th century. What started 4. Cohen, P. J. 1966 Set Theory and the Continuum
in the hands of Cantor as a mathematical theory Hypothesis. W. A. Benjamin, Inc., New York.
of transfinite numbers has developed to become 5. Dales, H. G. and Woodin, W. H. 1987 An in-
troduction to independence for analysts. London
a general theory of infinite sets and a foundation Mathematical Society Lecture Note Series, 115.
of mathematics. The fact that it has been pos- Cambridge University Press, Cambridge.
sible to unify all classical mathematics into one 6. Dodd, A. J. 1982 The Core Model. London Math.
single theoretical framework, the ZF C axiom sys- Soc. Lecture Note Ser. 61. Cambridge University
tem, is certainly remarkable. But beyond this, and Press, Cambridge-New York.
most importantly, the techniques developed by set 7. Erdős, P. and Rado, R. 1956 A partition calculus
theory, such as constructibility, forcing, infinite in set theory. Bulletin of the American Mat. Soc.
combinatorics, the theory of large cardinals, de- 62, 427-489.
terminacy, the descriptive theory of definable sets 8. Easton, W. B. 1970 Powers of regular cardinals.
Annals of Mathematical Logic, Number 1. 139-
in Polish spaces, etc., have turned it into a dis-
178.
cipline of great depth and beauty, with fascinat-
9. Eklof, Paul C. and Mekler, Alan H. 1990 Almost
ing results that stimulate and challenge our imag- free modules. Set-theoretic methods. North-
ination, and with numerous applications in areas Holland Mathematical Library, 46. North-
such as algebra, topology, real and complex anal- Holland Publishing Co., Amsterdam.
ysis, functional analysis, and measure theory. In 10. Farah, I. 2000 Analytic Quotients. Memoirs of
the twenty-first century, the ideas and techniques the American mathematical Society, Number 2.
generated within set theory will surely continue to American Mathematical Society.
contribute to the solution of outstanding mathe- 11. Foreman, M., Kechris, A. S., Louveau, A. and
matical problems, old as well as new, and will help Weiss, B. (Editors) 2000 Descriptive set theory
and dynamical systems. Papers from the Interna-
mathematicians gain an ever deeper insight into
tional Workshop held in Marseille-Luminy, July
the complexities and vastness of the mathematical 1–5, 1996. London Mathematical Society Lecture
universe. Note Series, 277. Cambridge Univ. Press, Cam-
bridge.
12. Foreman, M., Magidor, M., and Shelah, S. 1988
14 Acknowledgements Martin’s maximum, saturated ideals and non-
regular ultrafilters. Part I. Ann. of Math. 127,
Thanks to David Asperó, Roger Bosch, Matt Fore- 1-47.
man, Sy D. Friedman, and Ignasi Jané for read- 13. Foreman, M. and Woodin, W. H. 1991 The gener-
ing early versions of the article and making valu- alized continuum hypothesis can fail everywhere.
able suggestions for improvement. I am especially Ann. of Math. 133, 1-35.
grateful to Timothy Gowers and Imre Leader, edi- 14. Fremlin, D. H. 1984 Consequences of Martin’s Ax-
tors of The Princeton Companion to Mathematics, iom. Cambridge tracts in Mathematics #84.
for their invitation to contribute to the Volume and Cambridge, Cambridge University press.
for their excellent work at editing the article. 15. Friedman, H. 1998 Finite Functions and the Nec-
essary Use of Large Cardinals. Ann. of Math.
148, 803-893.
Bibliography 16. Friedman, S. D. 2000 Fine Structure and Class
1. Bagaria, J., Castells, N., and Larson, P. 2006 An Forcing. de Gruyter Series in Logic and Its Appli-
Ω-logic Primer. In Set Theory, Centre de Recerca cations 3. Walter De Gruyter, Berlin-New York.
Matemàtica, Barcelona, 2003-2004. Joan Bagaria 17. Gitik, M. 1991 The strength of the failure of the
and Stevo Todorcevic, Editors. Trends in Mathe- singular cardinal hypothesis. Annals of Pure and
matics, Birkhäuser Verlag. Applied Logic 51, Number 3, 215-240.
2. Blass, A. (To appear) Cardinal Characteristics 18. Gödel, K. 1938 The consistency of the axiom of
of the Continuum. Handbook of Set Theory, M. choice and of the generalized continuum hypoth-
Foreman and A. Kanamori, Editors. esis. Proc. Nat. Acad. Sci. U.S.A. 24, 556-557.
26

19. Harrington, L. A. 1978 Analytic determinacy and 38. Shelah, S. 1994 Cardinal Arithmetic. Oxford
0] . Journal of Symbolic Logic 43, Number 43, Logic Guides 29, The Clarendon Press, Oxford
685-693. University Press, New York.
20. Hjorth, G. and Kechris, A. S. 2005 Rigidity theo- 39. Shelah, S. and Woodin, W. H. 1990 Large cardi-
rems for actions of product groups and countable nals imply that every reasonably definable set of
Borel equivalence relations. Mem. Amer. Math. reals is Lebesgue measurable. Israel J. Math 70,
Soc., 177. Number 3, 381-394.
21. Jech, T. 2003 Set theory. 3d Edition. Springer, 40. Silver, J. H. 1975 On the singular cardinals prob-
New York. lem. In Proceedings of the International Congress
22. Jensen, R. B. 1972 The fine structure of the of Mathematicians, Vancouver, B. C., 1974, Vol.
constructible hierarchy. Annals of Mathematical 1. Canad. Math. Congress, Montreal, Que., 265-
Logic 4, Number 3, 229-308. 268.
23. Kanamori, A. 2003 The Higher Infinite. Second 41. Solovay, R. and Tennenbaum, S. 1971 Iterated
Edition. Springer Monographs in Mathematics. Cohen extensions and Souslin’s problem. Ann.
Springer. of Math. (2) 94, 201-245.
24. Kechris, A. S. 1995 Classical Descriptive Set The- 42. Solovay, R. 1971 Real-valued measurable cardi-
ory. Graduate Texts in Mathematics. Springer nals. In: Axiomatic Set Theory (D. S. Scott,
Verlag. ed.), Proc. Sympos. Pure Math., Vol XIII, Part
I, Univ. California, Los Angeles, Calif., 1967.
25. Kechris, A. S. and Louveau, A. 1987 Descriptive Amer. Math. Soc., Providence, R.I., 397-428
set theory and the structure of sets of unique-
43. Solovay, R. 1970 A model of set theory in which
ness. London Mathematical Society Lecture Note
every set of reals is Lebesgue measurable. Ann.
Series, 128. Cambridge University Press, Cam-
of Math. 92, 1-56.
bridge.
44. Todorcevic, S. 2002 Generic absoluteness and the
26. Kunen, K. 1980 Set Theory, An Introduction to
continuum. Math. Research Letters 9, 1-7.
Independence Proofs. North-Holland, Amster-
dam. 45. Todorcevic, S. 1989 Partition Problems in Topol-
ogy. Contemporary Mathematics. American
27. Larson, P. B. 2004 The Stationary Tower. Notes Mathematical Society, Volume 84.
on a course by W. Hugh Woodin. University Lec-
46. Woodin, W. H. 1999 The Axiom of Determinacy,
ture Series, Vol. 32. American Mathematical So-
Forcing Axioms, and the Nonstationary Ideal.
ciety, Providence, RI.
DeGruyter Series in Logic and Its Applications
28. Laver, R. 1976 On the consistency of Borel’s con- 1. Walter de Gruyter, Berlin-New York.
jecture. Acta Mathematica 137, 151-169.
47. Woodin, W. H. 2001 The Continuum Hypothe-
29. Magidor, M. 1977 On the singular cardinals prob- sis, Part I. Notices of the AMS 48, Number 6,
lem, II. Ann. of Math. 106, 514-547. 567-576, and The Continuum Hypothesis, Part
30. Martin, D. A. and Solovay, R. 1970 Internal Co- II. Notices of the AMS 48, Number 7, 681-690.
hen Extensions. Annals of Mathematical Logic 2, 48. Zeman, M. 2001 Inner Models and Large Cardi-
143-178. nals. de Gruyter Series in Logic and Its Applica-
31. Martin, D. A. and Steel, J. R. 1989 A proof of tions 5. Walter de Gruyter, Berlin-New York.
projective determinacy. Journal of the American
mathematical Society 2, Number 1, 71-125.
32. Mathias, A. R. D. 1977 Happy families. Annals
of Mathematical Logic 12, 59-111.
33. Moore, J. T. 2005 Set mapping reflection. Journal
of Mathematical Logic 5, Number 1, 87-97.
34. Neeman, I. 2002 Inner models in the region of
a Woodin limit of Woodin cardinals. Annals of
Pure and Applied Logic 116, 67-155.
35. Schimmerling, E. and Zeman, M. 2001 Square in
core models. Bulletin of Symbolic Logic 7, Num-
ber 3, 305-314.
36. Shelah, S. 1998 Proper and improper forcing. 2nd
Edition. Springer-Verlag.
37. Shelah, S. 1974 Infinite abelian groups, White-
head problem and some constructions. Israel J.
Math 18, 243-256.

View publication stats

You might also like