Professional Documents
Culture Documents
Ralf Schindler
Set Theory
Exploring Independence and Truth
Universitext
Universitext
Series editors
Sheldon Axler
San Francisco State University, San Francisco, CA, USA
Vincenzo Capasso
Universit degli Studi di Milano, Milan, Italy
Carles Casacuberta
Universitat de Barcelona, Barcelona, Spain
Angus MacIntyre
Queen Mary University of London, London, UK
Kenneth Ribet
University of California, Berkeley, CA, USA
Claude Sabbah
CNRS cole Polytechnique Centre de mathmatiques, Palaiseau, France
Endre Sli
University of Oxford, Oxford, UK
Wojbor A. Woyczynski
Case Western Reserve University, Cleveland, OH, USA
Ralf Schindler
Set Theory
Exploring Independence and Truth
123
Ralf Schindler
Institut fr Mathematische Logik und
Grundlagenforschung
Universitt Mnster
Mnster
Germany
ISSN 0172-5939
ISSN 2191-6675 (electronic)
ISBN 978-3-319-06724-7
ISBN 978-3-319-06725-4 (eBook)
DOI 10.1007/978-3-319-06725-4
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number: 2014938475
Mathematics Subject Classification: 03-01, 03E10, 03E15, 03E35, 03E45, 03E55, 03E60
Springer International Publishing Switzerland 2014
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publishers location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)
Preface
Set theory aims at proving interesting true statements about the mathematical
universe. Different people interpret interesting in different ways. It is well
known that set theory comes from real analysis. This led to descriptive set theory,
the study of properties of definable sets of reals, and it certainly is an important
area of set theory. We now know that the theory of large cardinals is a twin of
descriptive set theory. I find the interplay of large cardinals, inner models, and
properties of definable sets of reals very interesting.
We give a complete account of the Solovay-Shelah Theorem according to
which having all sets of reals to be Lebesgue measurable and having an inaccessible cardinal are equiconsistent. We give a modern account of the theory of 0#,
produce Jensens Covering Lemma, and prove the Martin-Harrington Theorem
according to which the existence of 0# is equivalent with R11 determinacy. We also
produce the Martin-Steel Theorem according to which Projective Determinacy
follows from the existence of infinitely many Woodin cardinals.
I started learning logic by reading a script of my Masters thesis advisor, Ulrich
Blau, on a nude beach by the Ammersee near Munich back in 1989. It was a very
enjoyable way of learning a fascinating and exciting subject, and I then decided to
become a logician (In the meantime, Blaus script appeared as [6]). We shall assume
in what follows that the reader has some familiarity with mathematical logic, to the
extent of e.g. [11]. We are not going to explain the key concepts of first order logic.
I thank David Asper, Fabiana Castiblanco, William Chan, Gabriel Fernandes,
Daisuke Ikegami, Marios Koulakis, Paul Larson, Stefan Miedzianowski, Haimanti
Sarbadhikari, Shashi Srivastava, Sandra Uhlenbrock, Yong Cheng, and the anonymous referees for their many helpful comments on earlier versions of this book.
I thank my father and my mother. I thank my academic teachers, Ulrich Blau,
Ronald Jensen, Peter Koepke, and John Steel. I thank all my colleagues, especially
Martin Zeman. And I thank my wife, Marga Lpez Arp, for all her support over
the last years.
Berkeley, Girona, and Mnster, February 2014
Ralf Schindler
vii
Contents
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
18
19
Ordinals. . . . . . . . . . . . . . . .
3.1 Ordinal Numbers . . . . .
3.2 Induction and Recursion
3.3 Problems . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
23
23
26
29
Cardinals . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1 Regular and Singular Cardinal Numbers
4.2 Stationary Sets . . . . . . . . . . . . . . . . . .
4.3 Large Cardinals . . . . . . . . . . . . . . . . .
4.4 Problems . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
33
33
40
46
62
Constructibility . . . . . . . . . . . .
5.1 The Constructible Universe
5.2 Ordinal Definability . . . . .
5.3 Problems . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
67
67
86
88
Forcing . . . . . . . . . . . . . . . . . . . . .
6.1 The General Theory of Forcing
6.2 Applications of Forcing. . . . . .
6.3 Problems . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
93
93
103
122
............
............
Constructibility .
............
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
127
127
141
143
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
7
ix
Contents
Solovays Model . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.1 Lebesgue Measurability and the Property of Baire
8.2 Solovays Theorem . . . . . . . . . . . . . . . . . . . . . .
8.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
147
147
156
163
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
165
165
173
180
10 Measurable Cardinals. . . . . . . .
10.1 Iterations of V . . . . . . . . .
10.2 The Story of 0] , Revisited .
10.3 Extenders . . . . . . . . . . . .
10.4 Iteration Trees . . . . . . . . .
10.5 Problems . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
183
183
191
209
221
229
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
235
235
262
269
275
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
279
279
292
294
300
13 Projective Determinacy. . . . . . .
13.1 Embedding Normal Forms .
13.2 The MartinSteel Theorem
13.3 Problems . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
303
303
307
322
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
325
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
327
Chapter 1
Georg Cantor (18451918) discovered set theory. Prior to Cantor, people often
took it to be paradoxical that there are sets which can be put into a bijective correspondence with a proper subset of themselves. For instance, there is a bijection from
N onto the set of all prime numbers. Hence, it seemed, on the one hand the set of all
primes is smaller than N, whereas on the other hand it is as big as N.
Cantors solution to this paradox was as follows. Let X and Y be arbitrary sets.
Define X is smaller than or of the same size as Y (or, Y is not bigger than X ) as:
there is an injection f : X Y . Write this as X Y . Define X is of the same size
as Y as: there is a bijection f : X Y . Write this as X Y . Obviously, X Y
implies X Y . The theorem of CantorSchrderBernstein (cf. Theorem 1.4)
will say that X Y follows from X Y and Y X . We write X < Y if X Y
but not Y X .
Notice that if X Y , i.e., if there is an injection f : X Y , then there is a
surjection g : Y X . This is clear if f is already bijective. If not, then pick a0 X
(we may assume X to be nonempty). Define g : Y X by g(b) = f 1 (b), if b is
in the range of f, and g(b) = a0 otherwise.
Conversely, if f : X Y is surjective then there is an injection g : Y X , i.e.,
Y X . This is shown by choosing for each b Y some a X with f (a) = b and
setting g(b) = a. This argument is in need of the Axiom of Choice, AC, which we
shall present in the next chapter and discuss in detail later on.
To a certain extent, set theory is the study of the cardinality of arbitrary sets, i.e.,
of the relations and as defined above. The proof of the following theorem may
be regarded as the birth of set theory.
Theorem 1.1 (Cantor)
N < R.
Proof N R is trivial. We show that R N does not hold.
Assume that there is an injection from R to N, so that there is then also a surjection
f : N R. Write xn for f (n). In particular, R = {xn : n N}.
x {xn : n N} = R. Contradiction!
It is not hard to verify that the sets of all integers, of all rationals, and of all
algebraic numbers are each of the same size as N (cf. Problem 1.1). In particular,
Theorem 1.1 immediately gives the following.
Corollary 1.2 There are transcendental numbers.
For arbitrary sets X and Y , we write Y X for: Y is a (not necessarily proper) subset
of X , i.e., every element of Y is also an element of X , and we let P(X ) = {Y : Y X }
denote the power set of X , i.e., the set of all subsets of X . Problem 1.2 shows that
P(N) R. The following is thus a generalization of Theorem 1.1.
Theorem 1.3 For every X , X < P(X ).
Proof We have X P(X ), because f : X P(X ) is injective where f (x) = {x}
for x X .
We have to see that P(X ) X does not hold true. Given an arbitrary f : X
P(X ), consider Y = {x X : x f (x)} X . If Y were in the range of f , say
Y = f (x0 ), then we would have that x0 Y x0 f (x0 ) = Y . Contradiction!
In particular, f cannot be surjective, which shows that P(X ) X is false.
f (x)
g 1 (x)
if x X 0 X 1 , and
if x X 2 .
1
1
, and bs 1 as 1
.
n+1
n+1
nN
(ax , bx ) A.
xA0
As Q N, there are at most countably many sets of the form (ax , bx ) A, and each
of them is at most countable. Hence A0 is at most countable (cf. Problem 1.4).
Suppose that P = . We first show that P P . Let x P. Let a < x < b. We
have that (a, b) A is uncountable. Suppose that (a, b) (P\{x}) = . For each
y ((a, b)\{x}) A there are then a y < y < b y with a y , b y Q so that (a y , b y ) A
is at most countable. But then we have that
(a y , b y ) A
(a, b) A {x}
y(a,b)\{x}
A2 as A , etc., i.e., An+1 as (An ) for n N. It is easy to see that each An is closed,
and
. . . An+1 An . . . A1 A.
If there is some n with An+1 = An then P = An and A0 = A\P are as in the statement of Theorem 1.9. Otherwise we have to continue this process into the transfinite.
Let
An , A+1 = (A ) , . . . , A+n+1 = (A+n ) ,
A =
nN
A+ =
A+n , . . . etc.
nN
It can be shown that there is a number so that A+1 = A . For such an , A\A
is at most countable, and if A = , then A is perfect.
Such numbers are called ordinal numbers (cf. Definition 3.3). We need an
axiomatization of set theory (to be presented in Chap. 2), though, in order to be able
to introduce them rigorously. With their help we shall be able to prove much stronger
forms of the Theorem of CantorBendixson (cf. Theorems 7.15 and 12.11).
Definition 1.11 A set A R is called nowhere dense iff R \ A has an open subset
which is dense in R. A set A R
is called meager (or of first category) iff there are
An R, n N, such that A = nN An and each An is nowhere dense. If A R
is not meager, then it is of second category.
It is not hard to see that A is nowhere dense iff for all a, b R with a < b there
are a , b R with a a < b b and [a , b ] A = (cf. Problem 1.8(a)).
Of course, every countable set of reals is meager, and in fact the countable union of
meager sets is meager, but there are nowhere dense sets which have the same size as
R (cf. Problem 1.8 (c)).
Theorem
1.12 (Baire Category Theorem) If each An R is open and dense, n N,
then nN An is dense.
Proof Let a < b, a, b R be arbitrary. We need to see that [a, b] nN An = . Let
us define [an , bn ], n N, recursively as follows. We set [a0 , b0 ] = [a, b]. Suppose
[an , bn ] is already chosen. As An is dense, (an , bn ) An = , say x (an , bn ) An .
As An is open, we may pick c, d with an < c < x < d < bn and (c, d) An . Let
such that c < an+1 < bn+1
an+1 , bn+1 be
< d, so that [an+1 , bn+1 ] An [an , bn ].
bn an
nN
1.1 Problems
1.1. Show that the sets of all finite sets of natural numbers, of all integers, of all
rationals, and of all algebraic numbers are each countable, i.e., of the same size
as N.
1.2. Show that R N {0, 1}, where N {0, 1} is the set of all infinite sequences of 0s
and 1s.
1.3. If A, B are sets of natural numbers, then A and B are called almost disjoint
iff A B is finite. A collection D of sets of natural numbers is called almost
disjoint iff any two distinct elements of D are almost disjoint. Show that there
is an almost disjoint collection D of sets of natural numbers such that D R.
[Hint: Use a bijection between the set of finite 01sequences and N.]
1.4. Let, for each n N, An be a countable set. Show that nN An is countable.
(This uses AC, the Axiom of Choice, cf. Theorem 6.69.)
1.5. Let n N. Construct a set A R such that An = , but An+1 = . Also
construct a set A R such that A+n = , but A+n+1 = .
1.6. Let A R be closed. Show that the pair (A0 , P) as in the statement of Theorem
1.9 of CantorBendixson is unique.
1.7. Show that if A [0, 1], A = nN [an , bn ], where the [an , bn ] are pairwise
disjoint, then (A) as defined above is independent
from the choice of the
pairwise disjoint intervals [an , bn ] with A = nN [an , bn ].
1.8. (a) Show that A R is nowhere dense iff for all a, b R with a < b there are
a , b R with a a < b b and [a , b ] A = .
(b) Show that R is not meager. In fact, the complement of a meager set A R
is dense in R.
(c) For a, b R with a < b let
2
2
1
1
2
[a, b] 3 = [a, a + b] [ a + b, b],
3
3
3
3
Finally, let, for a, b R with a < b, [a, b]0 = [a, b], [a, b]n+1 = ([a, b]n ) 3 ,
and
[a, b]n .
[a, b] =
nN
Chapter 2
Ernst Zermelo (18711953) was the first to find an axiomatization of set theory,
and it was later expanded by Abraham Fraenkel (18911965).
(Ext)
(Fund)
10
(Pair)
x.
(Union)
u(u z u x u y),
(Pair) and (Union) prove that xyz(z = x y), as x y = {x, y}.
The power set axiom, (Pow), says that for every set x, the set of all subsets of x
exists. We write x = P(y) for
z(z x z y)
and formulate
xy y = P(x).
(Pow)
The axiom of infinity, (Inf), tells us that there is a set which contains all of the
following sets as members:
, {}, {, {}}, {, {}, {, {}}}, . . . .
To make this precise, we call a set x inductive iff
x y(y x y {y} x).
11
We then say:
x(x is inductive).
(Inf)
(Aus )
u(u z u x u y),
y for
y).
The separation schema (Aus) is the set of all (Aus ). It says that we may separate
elements from a given set according to some well-defined device to obtain a new set.
Now let be a formula of L in which exactly the variables x, y, v1 , . . . , v p (all
different from b) occur freely. The replacement axiom corresponding to runs as
follows.
v1 . . . v p (xy y(y = y ) aby(y b x(x a ))).
(Rep )
The replacement schema (Rep) is the set of all (Rep ). It says that we may replace
elements from a given set according to some well-defined device by other sets to
obtain a new set.
We could not have crossed out x a in (Aus ). If we did cross it out in (Aus )
and let be x x, then we would get
12
bx(x b x x),
which is a false statement, because it gives b b b b. This observation
sometimes runs under the title of Russells Antinomy.
In what follows we shall write x
/ y instead of x y, and we shall write x = y
instead of x = y.
A trivial application of the separation schema is the existence of the empty set
which may be obtained from any set a by separating using the formula x = x as ,
in other words,
bx(x b x = x).
With the help of (Pair) and (Union) we can then prove the existence of each of the
following sets:
, {}, {{}}, {, {}}, . . . .
In particular, we will be able to prove the existence of each member of the intersection
of all inductive sets. This will be discussed in the next chapter.
The axiom of choice finally says that for each family of pairwise disjoint nonempty sets there is a choice set, i.e.
x(y(y x y = ) yy (y x y x y = y y y = )
zy(y x uu (u = u u z y))).
(AC)
13
Modulo ZF, (AC) has many equivalent formulations. In order to formulate some
of them, we first have to introduce basic notations of axiomatic set theory, though.
For sets x, y we write (x, y) for {{x}, {x, y}}. It is easy to verify that for all
x, y, x , y , if (x, y) = (x , y ), then x = x and y = y . The set (x, y) can be shown
to exist for every x, y by applying the pairing axiom three times; (x, y) is called the
ordered pair of x and y.
We also write {x1 , . . . , xn+1 } for {x1 , . . . , xn } {xn+1 } and (x1 , . . . , xn+1 ) for
), then x1 = x1 , . . ., and
((x1 , . . . , xn ), xn+1 ). If (x1 , . . . , xn+1 ) = (x1 , . . . , xn+1
xn+1 = xn+1 .
The Cartesian product of two sets a, b is defined to be
a b = {(x, y): x a y b}.
a .
a n = a
n -times
14
15
(a, a )
= (b, b ). Then there is a unique f with (a, a )
= (b, b ).
f
Proof If (a, a )
= (b, b ) and (a, a )
= (b, b ), then (a, a )
1
g f is the identity, so f = g.
g 1 f
(a, a ), so
16
If (a, ) is a partially ordered set, and if x a, then we write (a, ) x for the
partially ordered set
({y a: y < x}, {y a: y < x}2 ),
i.e., for the restriction of (a, ) to the predecessors of x.
Lemma 2.8 If (a, ) is a well-ordering, and if x a, then (a, )
= (a, ) x.
f
Proof If (a, )
= (a, ) x, then f : a a is order-preserving with f (x) < x.
This contradicts Lemma 2.5.
Theorem 2.9 Let (a, a ), (b, b ) be well-orderings. Then exactly one of the following statements holds true.
(1) (a, a )
= (b, b )
(2) x b (a, a )
= (b, b ) x
(3) x a (a, a ) x
= (b, b ).
Proof Let us define r a b by
(x, y) r (a, a ) x
= (b, b ) y.
By the previous lemma, for each x a there is at most one y b such that (x, y) r
and vice versa. Therefore, r is an injective function from a subset of a to b. We have
that r is order-preserving, because, if x <a x and
f
(a, a ) x
= (b, b ) y,
then
(a, a ) x
f {ya:y<x}
(b, b ) f (x),
(a, a ) x
= (b, b ) y,
so that (x, y) r after all. Contradiction!
The following Theorem is usually called Zorns Lemma. The reader will gladly
verify that its proof is performed in the theory ZC.
Theorem 2.10 (Zorn) Let (a, ) be a partial ordering, a = , such that for all
b a, b = , if x by b(x y y x), then b has an upper bound. Then a
has a maximal element.
17
else there is some v c such that (b, ) = (c, ) v. Let g: b c be such that
g
(b, )
= (c, ) or (b, )
= (c, ) v.
We aim to see that g is the identity on b.
Suppose not, and let u 0 b be -minimal in
{w b: g(w) = w}.
Writing g = g {w b: w < u 0 },
g
(b, ) u 0
= (c, ) g(u 0 ),
and g is in fact the identity on {w b: w < u 0 }, so that
{w b: w < u 0 } = {w c: w < g(u 0 )}.
) = g(u 0 ).
u 0 = f (Bu0 ) = f (Bg(u
0)
Contradiction!
We have shown that if , W , then or .
18
But now
Setting
our hypothesis on gives us that B = . Suppose that b does have a maximum with
respect to . We must then have B b = , and if we set
u 0 = f (B)
(2.1)
xX x = X
(2.2)
X ( Y X Y x x = X )
(2.3)
(Rep )
19
and for all such that is a formula of the language of BG, which contains exactly
x, X 1 , . . . , X k (but not Y ) as its free variables and which does not have quantifiers
ranging over classes (in other words, results from a formula of the language of
ZF by replacing free occurences of set variables by class variables), then
X 1 . . . X k Y x(x Y ).
(Comp )
(Comp ) is called the comprehension axiom for , and the collection of all (Comp )
is called the comprehension schema. The BernaysGdel class theory with choice,
BGC, is the theory BG plus the following version of the axiom of choice:
There is a (class) function F such that x(x = F(x) x).
(AC)
It can be shown that ZFC and BGC prove the same theorems in their common
language L (i.e., BGC is conservative over ZFC).
If is a formula as in (Comp ), then we shall write {x: } for the class given
by (Comp ). (Rep ) says that for all class functions F and for all sets a, F a =
{y: x (x, y) F} is a set.
We shall write V for the universe of all sets, i.e., for {x: x = x}. V cannot be a
set, because otherwise
R = {x V : x
/ x}
would be a set, and then R R iff R
/ R. This is another instantiation of Russells
antinomy.
If A is a class, then we write
A = {x: y A x y}
and
A = {x: y A x y}.
A and A always exist, and = and = V .
It may be shown that in contrast to ZFC, BGC can be finitely axiomatized. BGC
will be the theory used in this book.
The books [15, 18, 23] present introductions to axiomatic set theory.
2.3 Problems
2.1. Let k N. Show that there cannot be sets x1 , x2 , . . . , xk such that x1 x2
. . . xk x1 .
2.2. Show that for all x, y, (x, y) exists. Show that if (x, y) = (x , y ), then x = x
and y = y . Show that for all a, b, a b exists (cf. Lemma 2.1). Show that for
20
all d, b, d b exists (cf. Lemma 2.2). Which axioms of ZF do you need in each
case? Show that (Pair) may be derived from the rest of the axioms of ZF (from
which ones?).
2.3. Show that neither in (Aus ) nor in (Rep ), as formulated on p. 11, we could
have allowed b to occur freely in . Show that the separation schema (Aus)
can be derived from the rest of the axioms of ZF augmented by the statement
x x = .
2.4. Show that the following version of (AC) is simply false:
x(y x y = ) zy xu z y = {u}).
2.5. Show tht every partial order can be extended to a linear order. More precisely:
Let a be any set. Show that for any partial order on a there is a linear order
on a with .
2.6. Show that in the theory ZF, the following statements are equivalent.
(i) (AC).
(ii) For every x such that
y = for every y x there is a choice function,
i.e., some f : x x such that f (y) y for all y x.
2.7. (a) Let denote the natural ordering on N, and let m N, m 2. Let the
ordering m on N be defined as follows. n m n iff either n n (mod m)
and n n , or else if k < m, k N, is least such that n k(mod m) and
k < m, k N, is least such that n k (mod m), then k < k . Show that m
is a well-ordering on N.
(b) Let, for m N, m be any well-ordering of N, and let : N N N be
a bijection. Let us define on N by n n iff, letting (m, q) = (n) and
(m , q ) = (n ), m < m or else m = m and q m q . Show that is a
well-ordering of N.
2.8. (Cantor) Let (a, <) be a linear order. (a, <) is called dense iff for all x, y a
with x < y there is some z a with x < z < y. Show that if (a, <) is dense
(and a has more than one element), then < is not a well-ordering on a. (a, <)
is said to have no endpoints iff for all x a there are y, z a with y < x < z.
Let (a, <a ) and (b, <b ) be two dense linear orders with no endpoints such
that both a and b are countable. Show that (a, <a ) is isomorphic to (b, <b ).
[Hint. Write a = {xn : n N} and b = {yn : n N}, and construct f : a b
by recursively choosing f (x0 ), f 1 (y0 ), f (x1 ), f 1 (y1 ), etc.]
2.9. Show that there is a set A of pairwise non-isomorphic linear orders on N such
that A R.
2.10. Show that every axiom of ZFC is provable in BGC.
Let us introduce Ackermanns set theory, AST. The language of AST arises
The axioms of AST are (Ext),
from L by adding a single constant, say v.
(Fund), (Aus), as well as (Str) and (Refl) which are formulated as follows.
2.3 Problems
21
x v y ((y x y x) y v).
(Str)
(Refl )
Chapter 3
Ordinals
23
24
3 Ordinals
25
26
3 Ordinals
27
there is exactly one set y with (u, x, y, p). There is then a (class) function F with
domain B such that for all x in B, F(x) is the unique y with
(F { y B : y Rx}, x, y, p).
Proof Let us call a (set or class) function F good for x iff
(a) x dom(F) B,
(b) x dom(F) y B ( y Rx y dom(F)), and
(c) x dom(F) (F(x ) is the unique y with (F { y B : y Rx }, x , y, p).
If F, F are both good for x, then F(x) = F (x), as we may otherwise consider
the Rleast x0 dom(F) dom(F ) with F(x0 ) = F (x0 ) and get an immediate
contradiction. For all x for which there is a set function f V which is good for x,
{ f V : f is good for x},
which we shall ad hoc denote by f x , is then easily seen to be good for x.
We claim that for each x B, there is some set function g V which is good
for x. Suppose not, and let x0 be Rleast in the class of all x such that there is no
set function g V which is good for x. Then g x exists for all x Rx0 , and we may
consider
g=
{g x : x Rx0 }.
As R is set-like, g is a set by the appropriate axiom of replacement. Moreover g is a
function which is good for each x B with x Rx0 . Now let y be unique such that
(g {x B : x Rx0 }, x0 , y, p),
and set g = g {(x0 , y)}. Then g V is good for x0 . Contradiction!
We may now simply let
F=
{ f x : x B}.
Then F is a (class) function which is good for all x B.
Lemma 3.14 For every set x there is a transitive set y such that x y and y y
for all transitive sets y with x y .
Proof We use the recursion Theorem
such
3.13 to construct a function with domain
that f (0) = {x} and f (n + 1) =
f (n) for n < , and we consider ran( f ).
Definition 3.15 Let x be a set, and let y be as in Lemma 3.14. Then y is called the
transitive closure of {x}, denoted by TC({x}).
The following Lemma says that the relation, restricted to any class, is well
founded.
28
3 Ordinals
Lemma 3.16 Let A be a non-empty class. Then A has an -minimal member, i.e.,
there is some a A with a A = .
Proof Let x A be arbitrary, and let y be a transitive set with x y. As y A =
is a set, the axiom of foundation gives us some a y A which is minimal, i.e.,
a (y A) = . Then a A, and if z a, then z y (as y is transitive), so z
/ A.
That is, a A = .
Lemma 3.17 Let B be a class, and let R B B be set-like. Then R is well
founded if and only if there is some (unique) which is either an ordinal or else
= OR and some (unique) : B such that (x) = sup({(y) + 1 : y Rx}) for
all x B.
Proof Let us first suppose that R is wellfounded. We may then apply the recursion
theorem 3.13 to the formula (u, x, y) which says that y = sup({u(y) + 1 : y
dom(u)}) if u is a function whose range is contained in OR and y = otherwise.
We then get an and a function as desired.
On the other hand, if : B is such that (x) = sup({(y) + 1 : y Rx})
for all x B, then in particular y Rx implies that (y) < (x), so that R must be
wellfounded.
Definition 3.18 If R B B is wellfounded and set like, and if and : B
are as in Lemma 3.17, then (x) is called the Rrank of x B, written rk R (x) or
||x|| R , and is called the rank of R, written ||R||.
Definition 3.19 The hierarchy (V : OR) is recursively defined by
V =
{P(V ) : < }.
(3.1)
29
Proof Apply the recursion theorem to the formula (u, x, y) y = ran(u). We then
get a function F with domain B such that for all x B, F(x) = {F( y ) : y Rx}.
Notice that F is injective, because R is extensional. We may then set X R = ran(F)
and R = F 1 .
In particular, we get that wellorderings are wellfounded relations whose transitive collapse is an ordinal. Notice that if R B B is a wellordering, then R
is automatically extensional, so that we may indeed apply Mostowskis theorem to
R. The reason is that if {z B : z Rx} = {z B : z Ry} and x = y, then x Ry, say,
and so x Rx; but then R would not be wellfounded.
Definition 3.22 If R is a wellordering on B, then the unique ordinal such that
)
whenever
,
K
.
Then
K
F,
as
K is a function, dom( K ) =
{dom( ) : K } is an ordinal, and
K is injective. Hence F satisfies the
hypothesis of the Hausdorff Maximality Principle, Corollary 2.11, and there is
some F such that for no F, .
But now we must have ran( ) = A. Otherwise let x A\ ran( ), and set
= {(dom( ), x)}. Then F (with dom( ) = dom( ) + 1), .
Contradiction!
If f : A, where is an ordinal (or = OR), then f is also called a sequence
and we sometimes write ( f ( ) : < ) instead of f .
3.3 Problems
3.1 Use the recursion theorem 3.13 to show that there is a sequence (V : OR)
which satisfies (3.1). Show that every V is transitive and that V V for .
30
3 Ordinals
<
V for every
3.2 Show that for every set x there is some with x V (and thus x V+1 ).
For any set x, let rk (x) be as in Definition 3.18 for B = V and R = =
{(x, y) : x y}. Show that for every set x, the least such that x V is equal
to rk (x).
rk (x) is called the (set, or ) rank of x, also just written rk(x).
3.3 If M is transitive, then we may construe (M; M) as a model of L . Which
axioms of ZFC hold true in all (V ; V ), where > is a limit ordinal?
Which ones hold true in (V ; V )?
3.4 For a formula , let (Fund ) be the following version of the axiom of foundation.
p(x (x, p) x((x, p) y x (y, p))).
(Fund )
(Coll )
The collection principle is the set of all (Coll ). Show that in the theory Z, the
collection principle is equivalent to the replacement schema (Rep).
3.6 Let be an ordinal. Use the recursion Theorem 3.13 to show that there are
functions + , , and with the following properties.
(a) +0 = , +( +1) = ( +)+1 for all , and + = sup({ + : <
}) for a limit ordinal.
(b) 0 = 0, ( + 1) = ( ) + for all , and = sup({ : < })
for a limit ordinal.
(c) 0 = 1, +1 = ( ) for all , and = sup({ : < }) for a limit
ordinal.
3.7 Show that + and are associative. Show also that = 1 + = + 1 and
= 2 = 2. Show that if is a limit ordinal = 0, then + is a limit
ordinal. Show that if is a successor ordinal, then + is a successor ordinal.
Show that if is a limit ordinal = 0, then and are limit ordinals.
3.8 Show that if > 0 is an ordinal, then there are unique positive natural numbers
k and c1 , . . . , ck and ordinals 0 1 < . . . < k such that
= k ck + . . . + 1 c1 .
The representation (3.2) is called Cantor normal form of .
(3.2)
3.3 Problems
31
3.9 Use (AC) to show the following statement, called the principle of dependent
choice, DC. Let R be a binary relation on a set a such that for every x a
there is some y a such that (y, x) R. Show that there is some function
f : a such that for every n , ( f (n + 1), f (n)) R. Use DC to prove
Lemma 3.11.
3.10 Show that in the theory ZF, the following statements are equivalent.
(i)
(ii)
(iii)
(iv)
(AC).
Zorns Lemma, i.e., Theorem 2.10.
The Hausdorff Maximality Principle, i.e., Corollary 2.11.
Zermelos WellOrdering Theorem 3.23.
3.11 Show in ZC that for every set a there is some r such that r is a wellordering
of a.
3.12 (F. Hartogs) Show in ZF that for every set x there is an ordinal such that
there is no injection f : x. [Hint. Consider the set W of all wellorders of
subsets of a, and wellorder W via Theorem 2.9.]
Chapter 4
Cardinals
We here use the notation for ordinal arithmetic from Problem 3.6.
33
34
4 Cardinals
0 , 1 , 2 , . . . , , +1 , . . . , 2 , . . . , ,
etc.2
An easy induction shows that for every ordinal . In particular, if is
an infinite cardinal, then , so that there is some ordinal with = .
Every infinite cardinal is thus of the form , where is an ordinal.
We define cardinal addition, multiplication, and exponentiation as follows. By
tradition, these operations are denoted the same way as ordinal addition, multiplication, and exponentiation, respectively, (cf. Problem 3.6) but it is usually clear from
the context which one we refer to.
Definition 4.5 Let , be cardinals. We set
+ = Card(( {0}) ( {1})),
= Card( ), and
= Card( ) = Card({ f : f is a function with dom( f ) = and
ran( f ) }).
It is easy to verify that + = Card(X Y ), whenever X, Y are disjoint sets with
X = and Y = . It is also easy to verify that if , 2, then + .
Cf. Problem 4.3. Moreover, we have the usual rules for addition, multiplication and
exponentiation.
2
35
36
4 Cardinals
(4.1)
37
Corollary 4.17 Let be a singular limit cardinal and assume that there are 0 <
and such that 2 = whenever 0 < . Then 2 = .
Proof Let 0 be such that cf(). Then, using Lemma 4.16, 2 =
cf() = (2 )cf() = 2cf() = 2 = .
The expression cf() will reappear in the statement of the Singular Cardinals
Hypothesis, cf. (4.2) below.
38
4 Cardinals
Definition 4.18 Let be an infinite cardinal. We then say that a set x is hereditarily
smaller than iff TC({x}) < . We let
H = {x: x is hereditarily smaller than } .
We also write HF (hereditarily finite) instead of H0 and HC (hereditarily
countable) instead of H1 .
It is not hard to show that H is a set for every infinite cardinal and that Card(H ) =
2< , cf. Poblem 4.10.
The following is often referred to as the Hausdorff Formula.
Theorem 4.19 (Hausdorff) For all infinite cardinals , , ( + ) = + .
Proof Suppose first that + . Then ( + ) = 2 = + .
Let us now assume that + > . Then, as + is regular by Lemma 4.15, every
< + with ran( f ) . Therefore
f : + is bounded, i.e., there
is some
+
( ) = Card( ( )) = Card( < + ) = + .
We may define infinite sums and products as follows. Let f be a function with
dom( f ) = I (where I is any non-empty set) and such that f (i) is a cardinal for
every i I . Let us write i for f (i), where i I . We then define
and
iI
i = Card
(i {i}) ,
iI
iI
39
As i < i , this set must be non-empty, so that we may let i be the least such that
for all i , f (, i)(i) = .
Now let g X iI i be defined by g(i) = i for i I . If i I and i , then
f (, i)(i) = i = g(i), i.e., f (, i) = g. Therefore, g ran( f ).
Corollary 4.21 For all infinite cardinals , cf(2 ) > and cf() > .
iI i = Card( (2 )) = (2 ) = 2 .
cf()
> , let f : cf()
To see that
i = f (i) for i I . Set i = for all i
Card(cf() ) = cf() .
The Singular Cardinal Hypothesis, abbreviated by SCH, is the statement that for
all singular limit cardinals ,
cf() = 2cf() + .
(4.2)
Notice that cf() 2cf() + holds for all infinite by Corollary 4.21, so that
SCH says that cf() has the minimal possible value. Moreover, if is regular, then
2cf() + = 2 + = 2 = = cf() , so that (4.2) is always true for regular . A
deep theorem of R. Jensen will say that the negation of SCH implies the existence
of an object called 0# , cf. Corollary 11.61.
If GCH holds true and is a singular limit cardinal, then cf() = 2 = + =
2cf() + using Lemma 4.16. Therefore, GCH implies SCH.
Lemma 4.22 Let be a limit cardinal, and suppose that SCH holds below , i.e.,
cf() = 2cf() + for every (infinite) < . Then for every (infinite) < and
for every infinite ,
2 if 2 ,
+
if > 2 is a limit cardinal of cofinality , and
=
if > 2 is a successor cardinal or a limit cardinal
of cofinality > .
Proof by induction on , fixing . If 2 , then (2 ) = 2 , and thus
= 2 . If = + > 2 , < , then = ( + ) = + = + = by the
Hausdorff Formula 4.19 and the inductive hypothesis.
Now let < , > 2 , be a limit cardinal, and let (i : i < cf()) be cofinal in
, where i > 2 for all i < cf(). By the inductive hypothesis (cf. also Problem
4.16) and as we assume SCH to hold below , we have that
i<cf()
i<cf()
(i )
i<cf()
i+
i<cf()
= cf() = 2cf() + .
40
4 Cardinals
Therefore, if cf() , then with the help of Corollary 4.21 we get that +
cf() 2 + = + , so that = + . On the other hand, if < cf(),
then every f : is bounded, so that by the inductive hypothesis
i<cf()
i+ cf() = ,
i<cf()
so that = .
41
{X : < \ X }
is a non-trivial < cf()-closed filter on , called the Frchet filter (on ). Every
filter can be extended to an ultrafilter, cf. Problem 4.11.
Lemma 4.25 Let be an ordinal such that cf() > , and let F be the set of all
A such that there is some B A which is club in . Then F is a (non-trivial)
< cf()-closed filter on .
Proof We need to see that if
< cf() and Ai is club in for every i , then
in
.
Well,
(
i Ai is club
i Ai ) {} is certainly closed, so that it suffices
to verify that i Ai is unbounded in . Let < . We define f : as
follows.3 For n and i we let f ( n + i) be the least element of Ai which
is bigger than sup({ f () : < n + i} { }). Notice that this is welldefined as
< cf() and each Ai is unbounded in .
Let = sup{ f () : < }. We have that < and in fact i Ai ,
because every Ai is closed; notice that for each i , = sup{ f ( n + i) : n
} = sup(Ai ).
F as in this lemma is called the club filter on .
Definition 4.26 Let be regular, and let X for all < . The diagonal inter
section of X , < , abbreviated by < X , is the set < : < X .
Definition
be regular, and let F be a filter on . F is called normal iff
4.27 Let
for all X : < F, < X F.
Lemma 4.28 Let be regular, and let F be the club filter on . Then F is normal.
Proof We need to see that if
A is club in for every < , then < A is
club in . By replacing A by A if necessary, we may and shall assume that
A A whenever . (Notice that every A is again club in by Lemma
4.25.) In order to see that ( < A ) {} is closed, let < be a limit ordinal such
that = sup(( < A ) ). We want to argue that < A , i.e., A for all
< . Well,
if < , then there is some with < < and < A ,
i.e., < A , in particular A . This shows that = sup(A ) for all
< ; hence A for all < .
In order to see that < A is unbounded in , let < . We construct a
sequence n , n , as follows. Let 0 = . If n is defined, then let n+1 be the
least > n such that An . We claim that sup{n : n } < A . Set
= sup{n : n }. We need to see that A for all < . Let < . Then
< n for some n . We have that m+1 Am An for all m n, so that
= sup{m+1 : m n} A xn A .
Definition 4.29 Let be an ordinal such that cf() > . A is called stationary
(in ) iff A C = for all C which are club in .
3
42
4 Cardinals
(4.3)
X F +.
Y = < :
/
<
43
The following theorem is a strong from of saying that the club filter is not an ultrafilter,
i.e. for any regular uncountable there are X + which neither contain nor are
disjoint from a club.
Theorem 4.33 (Solovay) Let be a regular uncountable cardinal, and let S be
stationary. Then S may be written as a disjoint union of stationary sets, i.e., there
in for every < , S S =
is (S : < ) such that S S is stationary
for all , < with = , and S = < S .
Proof Let us first write S = S0 S1 , where S0 = { S : cf() < } and S1 =
{ S : cf() = }. At least one of S0 , S1 must be stationary.
such
Claim 4.34 There is some stationary S S and some sequence (A : S)
us pick (A : S), where A is club in , otp(A ) = , and cf( ) < for all
A S A S1 = for each S.
let ( : i < otp(A )) be the monotone enumeration of A .
For S,
i
Claim 4.35 There is some i < such that for all < ,
44
4 Cardinals
I.e., i < for all i < . This yields , as A is club; however, clearly
Contradiction!
i i for every i < , so that in fact = . But then S.
45
as being defined by FYX (i) = gi ( f Y (i)) for i < cf(). In particular, FYX (i) < i for
every i SYX . Let D be the club of limit ordinals in below cf(). Then the map which
sends i SYX D to the least j < cf() with FYX (i) < j is regressive. As SYX D
is still stationary, by Fodors Theorem 4.32 there is a stationary set S YX SYX and
some i YX < cf() such that FYX (i) < i X for all i SYX .
Y
If Y, Z X , S YX = S ZX , i YX = i ZX , and FYX S YX = FZX S ZX , then Y = Z . But
there are only
2cf() cf() (sup< cf() ) =
many possible triples ( S YX , i YX , FYX SY ), so that there are at most many Y X .
In order to finish the proof of the Theorem, it thus remains to be shown that there
is some A P() of cardinality + such that P() = {Y : X A Y X }.
Let us recursively construct X for < + as follows. Given < + , having
constructed X for all < , we pick X such that for no < , X X . Notice
that this choice is possible, as {Y : < Y X } has size at most by
Claim 4.37. Set A = {X : < + }. We must have that P() = {Y : X
A Y X }, as otherwise there would be some Y with X Y for all < + ;
but X = X for = , so this is impossible by Claim 4.37.
Corollary 4.38 Let be a singular cardinal of uncountable cofinality. If GCH holds
below , then it holds at , i.e., if 2 = + for every < , then 2 = + .
Proof If 2 = + , then cf() = + , so that cf() = 2cf() + by Silvers
Theorem 4.36. But 2cf() < , so that cf() = + . By Lemma 4.16, 2 = cf() ,
which gives 2 = + .
Problems 4.17 and 4.18 produce generalizations of Theorem 4.36 and Corollary
4.38.
There is a generalization of stationarity which we shall now briefly discuss.
Definition 4.39 We say that S [ ] is stationary in [ ] iff for every A and
on A with at most many functions
for every algebra A = (A; ( f i : i < ))
there is some X S which is closed under all the f i , i < ,
from A, i.e.,
f i , i < ,
46
4 Cardinals
, 2 , 2(2 ) , . . .
is a strong limit cardinal above . There are thus arbitrarily large strong limit cardinals. Also, any limit of strong limit cardinals is a strong limit cardinal.
Definition 4.41 A cardinal is called weakly inaccessible iff is an uncountable
regular limit cardinal. A cardinal is called (strongly) inaccessible iff is an uncountable regular strong limit cardinal.
Trivially, every (strongly) inaccessible cardinal is weakly inaccessible. It can be
shown that there may be weakly inaccessible cardinals which are not strongly inaccessible (cf. Problem 6.13). Moreover, the existence of weakly inaccessible cardinals
cannot be proven in ZFC (cf. Problem 5.16). Hausdorffs question (cf. p. 35) as to
whether every uncountable limit cardinal is singular thus does not have an answer in
ZFC.
Large cardinals may be used to prove true statements which are unprovable in
ZFC, cf. e.g. Theorems 12.20 and 13.7. They may also be used for showing that
certain statements are consistent with ZFC, cf. e.g. Theorem 8.23.
Definition 4.42 A cardinal is called weakly Mahlo iff is weakly inaccessible
and the set
{ < : is regular}
is stationary. A cardinal is called (strongly) Mahlo iff is (strongly) inaccessible
and the set
{ < : is regular}
is stationary.
Again, every (strongly) Mahlo cardinal is weakly Mahlo, and there may be weakly
Mahlo cardinals which are not (strongly) Mahlo. If is weakly/strongly Mahlo,
then there are weakly/strongly inaccessible cardinals below (cf. Problem 4.21).
In the proof of Claim 1 of the proof of Solovays Theorem 4.33, S1 can only be
stationary if is weakly Mahlo.
Definition 4.43 Let be an infinite cardinal. A partially ordered set (T, <T ) is
called a tree iff for all s T , {t T : t <T s} is well-ordered by <T . In this case,
we write lvT (s) for the order-type of {t T : t <T s} and call it the level of s in T .
We also write ht(T ) for sup({lvT (s) + 1 : s T }) and call it the height of T .
47
48
4 Cardinals
If lvT2 (u ) : < were bounded in , then by (5) for T2 there would be some
X of size such that u = u for all , X . But then {s T : X s T
t } would be a cofinal branch through T . Contradiction! Hence {lvT2 (u ) : < }
is unbounded in , and (1) and (3) are shown for T2 .
(4) is clear by construction.
Lemma 4.46 (Knig) There is no 0 -Aronszajn tree.
Proof Cf. Problem 4.20.
49
Because T U will be closed under initial segments and ordered by endextension, we will have that lvT (s) = dom(s) for every s T . We shall construct
T = {s T : lvT (s) = dom(s) < }
by induction on . We maintain the following conditions.
(1) \ ran(s) = for all s T .
(2) If s T , \ ran(s) = A B, where A B = and A = B = , and if
lvT (s) < < + , then there is some t T with lvT (t) = , s t, and
ran(t) ran(s) A.
(3) T for all < + .
(4) Let < + be a limit ordinal with cf() < . Let C be club in with
otp(C) = cf(), and let (i : i < cf()) be the monotone enumeration of C. Let
= i<cf() Ai B, where Ai A j = , Ai B = , and B = Ai = for
i = j, i, j cf(). Let s : be such that
s i Ti +1 s i
Aj
ji+1
ji+1
Aj.
50
4 Cardinals
(4.4)
51
52
4 Cardinals
b = t (T ) : t <(T ) s .
We may assume without loss of generality that (T, <T ) V+1 , so that (T ) V =
T and <(T ) T =<T . But then b M V is a cofinal branch through T .
A cardinal is called Reinhardt iff there is a non-trivial elementary embedding
: V V with = crit( ). The following Theorem shows that there are no
Reinhardt cardinals (in ZFC).
Theorem 4.53 (K. Kunen) There is no non-trivial elementary embedding
: V V.
Proof Let 0 = crit( ), and recursively define n+1 = (n ). Set = supn< n .
By Lemma 4.52 (c), () = , and therefore also (+ ) = ()+ = + .
Let S = { < + : cf() = }. Because S is a stationary subset of + , S may be
partitioned into
0 stationary sets by Theorem 4.33, i.e., we may choose (Si : i < 0 )
such that S = i<0 Si , Si S j = for i = j, i, j < 0 , and each Si , i < 0 , is
stationary in + .
Set (Ti : i < 1 ) = ((Si : i < 0 )). We have that Ti T j = for i = j, i,
j < 1 , and each Ti , i < 1 , is stationary in + . Then T0 is a stationary subset of
+ . By Lemma 4.52 (c) and + + ,
C = < + : cf() = () =
53
+
is
an -club in . There is hence some T0 C (cf. Problem 4.12). As C S =
i<0 Si , there must be some i < 0 with Si . But then = () (Si ) =
T(i) , so that T(i) T0 = . But 0 = crit( ), so that 0 is not in the range of
and therefore (i) = 0 . Contradiction!
The proof of Theorem 4.53 in fact shows that there can be no non-trivial elementary
embedding : V+2 V+2 with crit( ) < . We remark that the proof of
Theorem 4.53 uses Theorem 4.33 which in turn makes use of the Axiom of Choice.
It is open whether Theorem 4.53 can be proven in ZF alone; this question leads to
Woodins HOD-conjecture, cf. [45, Section 7].
Large cardinal theory studies the question which fragments of Reinhardt
cardinals are consistent with ZFC.
Definition 4.54 Let be a cardinal. A filter F on is called uniform iff X =
for every X F. An uncountable cardinal is called measurable iff there is a
< -closed uniform ultrafilter on . Such a filter is also called a measure on
It is easy to see that if U is a < -closed ultrafilter on , then U is uniform iff for
no < , { } U , i.e., iff U is not generated by a singleton (cf. Problem 4.26).
If we didnt require a measurable cardinal to be uncountable, then 0 would be a
measurable cardinal.
Theorem 4.55 Let be a cardinal. The following are equivalent.
(1) is measurable.
(2) There is a normal < -closed uniform ultrafilter on .
(3) There is an inner model M and an elementary embedding : V M with
critical point .
Proof (3) = (2): Let : V M be an elementary embedding with critical point
. Let us set
U = U = {X : (X )}.
(4.5)
54
4 Cardinals
55
otherwise.
By the choice of a we now have that { < : V |= ( f 0 ( ), f 1 ( ), . . . , f k ( ))}
U , which inductively implies that ult(V ; U ) |= ([ f 0 ], [ f 1 ], . . . , [ f k ]), and hence
that ult(V ; U ) |= v0 (v0 , [ f 1 ], . . . , [ f k ]).
This verifies the Claim.
We now prove that E is well-founded, using Lemma 3.11. If ([ f n ] : n < ) were
a sequence such that [ f n+1 ]E[ f n ] for all n < , then
{ < : f n+1 ( ) f n ( )} U
. . . f 2 ( ) f 1 ( ) f 0 ( ),
a contradiction.
Therefore, by Theorem 3.21 there is an inner model N and some such that
(N ; )
= ({[ f ] : f V }; E). By os Theorem, we have that
N |= ( 1 ([ f 1 ]), . . . , 1 ([ f k ]))
if and only if
{ < : V |= ( f 1 ( ), . . . , f k ( ))} U
for all formulae and f 1 , . . ., f k V . This implies that we may define an elementary
embedding U : V N by setting U (x) = 1 (x) = 1 ([cx ]), where again
cx : {x} is the constant function with value x.
It remains to be shown that is the critical point of U . We first prove that
=
U () = for all < by induction on . Fix < and suppose that U ()
for all < . Let 1 ([ f ]) < U (), i.e., { < : f ( ) < } U . As U is
< + -closed, there is then some 0 < such that { < : f ( ) = 0 } U . But
then [ f ] = [c0 ], so that by the inductive hypothesis 0 = U (0 ) = 1 ([ f ]).
This shows that U () , so that in fact U () = .
56
4 Cardinals
(4.6)
(4.7)
(4.8)
V
U
M
k
ult(V ;U)
57
(4.9)
ult(M; U ) = UM ( f )( ) : f : M f M ,
(4.10)
where = [id].
Lemma 4.58 Let be a measurable cardinal, and let U be a normal < -closed
ultrafilter on . If R U , then R is ineffable.
Proof Let (A : R) be such that A for every R. Let
= UV : V M,
and let ( A : (R)) = ((A : R)). As (R), we may set A = A .
By the os Theorem, for every < there is some X U such that for all
X , A iff A . Let X = < X . It is then easy to verify that for every
X , A = A .
Theorem 4.59 (Rowbottom) Let be a measurable cardinal, and let U be a normal
measure on . Let < , and let F : []< . There is then some X U such
that for every n < , F [X ]n is constant.
Proof Fix < . It suffices to show that for every n < ,
F : []n < X U F [X ]n = { }.
(4.11)
58
4 Cardinals
(4.12)
Xa.
(4.13)
a[X ]n
59
where M is an inner model, crit( ) = , and 2 M M. In particular, is measurable, and if U is a measure on , then U M. Therefore,
({ < : V |= is measurable}),
so that if U is the measure on derived from as in the proof of Theorem 4.55, then
{ < : is measurable} U as desired.
Definition 4.64 Let be a regular cardinal, and let F be a filter on . F is called
weakly normal iff for all f : with { < : f ( ) < } F + there is some
< and some X F + such that f ( ) < for every X .
By Lemma 4.31, every normal filter is weakly normal.
60
4 Cardinals
(4.14)
Theorem 4.66 (Solovay) Let be supercompact. Then < = for every regular
cardinal .
Proof Let us fix , let again
: V M
61
S : f () < U.
(4.15)
< S : C [ , +1 ) = U.
Let us write
(4.16)
I = < : C [ , +1 ) =
(4.17)
Corollary 4.67 Let be supercompact. Then SCH holds above , i.e., if > is
singular, then cf() = 2cf() + .
Proof By (the proof of) Silvers Theorem 4.36 (cf. Problem 4.17 (1)) it suffices to
prove that 0 = + for every > with cf() = . However, for every such ,
0 (+ )0 = + by Theorem 4.66.
The following large cardinal will play a role for the failure of , cf. Definition
11.62 and Lemma 11.69.
Definition 4.68 A cardinal is called subcompact iff for every A H + there
is some < and some B H+ such that there is an elementary embedding
: (H+ ; , B) (H + ; , A).
Notice that in the situation of Definition 4.68, () = .
62
4 Cardinals
: V M
2 M
M. We have that
4.4 Problems
4.1. Let < 1 . Show that there is some X Q such that (; <)
= (X ; <Q X ).
(Here, < denotes the natural order on and <Q denotes the natural order on
Q.) [Hint. Use induction on .]
4.2. Let be a cardinal. Let y = x, where x = { : is an ordinal with }.
Show that y = + .
4.3. Let and be cardinals. Show that + = Card(X Y ), whenever X, Y are
disjoint sets with X = and Y = . Also show that + whenever
, 2.
4.4. Show that if and are cardinals, then = Card([] ).
4.5. Show that the least with = is singular of cofinality . Show that for
every regular cardinal there is some with = and cf() = .
4.6. Show that if is a limit cardinal, then cf() may be characterized as the least
such
that there is a sequence (i : i < ) of cardinals less than with
= i< i .
4.7. Let be an ordinal. Show that the cofinality of is the least size of a subset
of which is unbounded in . Show also that there is a club C such that
Card(C) = otp(C) = cf().
4.4 Problems
63
64
4 Cardinals
4.19. Use the axiom of choice to show that there is some A R such that neither
A nor R contains a perfect subset. [Hint. Show that there is an enumeration
e : 20 C, where C is the collection of perfect sets. If < is a well-ordering
of R, then construct (a , b : < 20 ) by letting a be the <-least element
of e( ) \ ({a : < } {b : < }) and b be the <-least element of
e( ) \ ({a : } {b : < }). Show that A = {a : < 20 } works.]
4.20. Prove Lemma 4.46.
4.21. Show that if is weakly Mahlo, then { < : is weakly inaccessible} is
stationary. Also show that if is Mahlo, then { < : is inaccessible} is
stationary.
4.22. Let 1 be a cardinal. If (T, <T ) is a -Souslin tree, then (T, <T ) is a
-Aronszajn tree.
4.23. Let be a cardinal. Show that the following are equivalent.
(a) is weakly compact.
(b) If X P(), Card(X ) , then there are transitive models H and H
with X H , Card(H ) = , H H and H H for every <
and there is some elementary embedding : H H such that is the
critical point of .
4.24. Let be a regular uncountable cardinal. Show:
(a) If < is an infinite regular cardinal, then { < : cf( ) = } is not
ineffable.
(b) If itself is ineffable, then is weakly compact and the set
{ < : is weakly compact}
is stationary in .
(c) Let be a measurable cardinal, and let U be a normal < -closed ultrafilter on . Then { < : is ineffable } U .
4.25. (JensenKunen) Show that if is ineffable, then there is no -Kurepa tree.
4.26. Let be a cardinal. Show that if U is a < -closed ultrafilter on , then U
is uniform iff for no < , { } U . Show also that if U is a < -closed
uniform ultrafilter on , then U is normal iff = 1 [id], where id is the
identity function and is the (inverse of) transitive collapse as in the proof
of Theorem 4.55.
Let U be a < -closed normal ultrafilter on , and let (X s : s []< ) be a
family such that X s U for every s []< . Let us define
s[]< X s = { :
s[ ]<
X s }.
4.4 Problems
65
Chapter 5
Constructibility
Models of the language L of set theory are of the form (M; E), where M = is a
set and E M M interprets . We shall also consider class models (M; E) of
L where M is a proper class rather than a set.
67
68
5 Constructibility
69
follows directly
Proof The first part, that R is a well-founded relation be ZFC
1
from (the proofs of) Lemmas 3.11 and 3.17. The second part is then a consequence
of Lemma 5.5.
The following easy lemma will be used to verify that fragments of ZFC hold in
a given transitive model.
Lemma 5.7 Let M be a transitive set or class.
(1)
(2)
(3)
(4)
Recall from p. 12 that ZFC is ZFC without the power set axiom.
70
5 Constructibility
x,
x can be written as
z y u x z u u x z u z y.
f (x) = x E
f is called rudimentary (or, rud) if f is rud .
easy to verify that for
We often write x for (x1 , . . . , xk ) in what follows. It is
instance the following functions are rudimentary: f (x) =
xi , f (x) = xi x j ,
71
72
5 Constructibility
n<
U
n . We
U for U ().
shall also write n/
n/
U
U
The notions n (X ), U
n , n , etc. are defined in an entirely analoguous fash
U
U
ion. A relation is U
n iff it is both n and n .
(5.1)
(U ;,E)
(U {U };,E)
= P(U ) 0
, so that we have to
(U {U };,E)
there are x1 , ..., xk U {U } such that y x iff y U and R(y, x1 , ..., xk ) holds.
But then x = U {y: R(y, x1 , ..., xk )} rud E (U {U }) by Lemma 5.10 (g).
: Call a function f : V k V , where k < , simple iff the following holds true:
if (v0 , v1 , . . . , vm ) is 0 in the language L,E , then ( f (v1 , . . . , vk ), v1 , . . . , vm )
is equivalent over transitive rud E closed structures to a 0 formula in the same
73
language. It is not hard to verify inductively that every rud E function is simple
(cf. Problem 5.8).
Now let x P(U ) rud E (U {U }), say x = f (x1 , . . . , xk ), where x1 ,
, xk U {U } and f is rud E . Then, as f is simple, v0 f (v1 , ..., vk ) is
(equivalent over rud E (U {U }) to) a 0 formula in the language L,E , and hence
(U {U },,E)
x = {y U : y f (x1 , ..., xn )} is in 0
({x1 , ..., xn }).
Of course Lemma 5.11 also holds with P(U ) being replaced by the set of all relations
on U .
Let U be rud E closed, and let x U be transitive. Suppose that
(U ;,E)
B 0
L[E] =
OR
J [E]
74
5 Constructibility
Obviously, every J [E] is rud E closed and transitive. We shall also denote by
J [E] the model (J [E]; J [E], E J [E]).
An important special case is obtained by letting E = in Definition 5.14. We
write J for J [], and L for L[]. L is Gdels Constructible Universe. Other
important examples which are studied in contemporary set theory are obtained by
letting E be a set or proper class with certain condensation properties or by letting
E code a (carefully chosen) sequence of extenders (cf. Definition 10.45 and also
Problem 10.5).
The next lemma is an immediate consequence of Lemma 5.11.
Lemma 5.15 Let E be a set or proper class. Assume that3 E Lim V . Let us
write
E = {x: (, x) E}
and
E = E ( V )
for limit ordinals . Let us assume that E J [E] and that (J [E]; , E ) is
amenable for every limit ordinal .
Then
[E];,E ,E ) .
P(J [E]) J+ [E] = P(J [E]) (J
The hypothesis of Lemma 5.15 is satisfied for all present-day canonical inner
models. (Cf. also Problem 5.13.) We will assume from now on, that E always satisfies
the hypothesis of Lemma 5.15.
The following is easy to verify inductively (cf. Problem 5.9).
Lemma 5.16 For every limit ordinal, J [E] OR = , and Card(J [E]) =
Card().
The following can be easily proved by induction on , with a subinduction on the
rank according to Definition 5.8. We shall produce a much stronger statement later
on, cf. (5.2).
Lemma 5.17 Let be a limit ordinal. If x J [E], then there is a transitive set
y J [E] such that x y.
Lemmas 5.7 and 5.17 (and Problem 5.5) immediately give the following.
Corollary 5.18 Let E be a set or a proper class. Let be a limit ordinal, > . Then
J [E] is a model of the following statements: (Ext), (Fund), (Inf), (Pair), (Union),
the statement that every set is an element of a transitive set, and 0 -comprehension.
Also, J [E] is a model of xy x y exists.
Theorem 5.19 Let E be a set or a proper class. L[E] |= ZF.
3
75
Proof We have to verify that the following axioms hold in L[E]: the power set
axiom, the separation schema and the replacement schema.
Let us start with the power set axiom (Pow). Fix a L[E]. By the replacement
schema in V , there is some such that
P(a) L[E] J [E].
But then
P(a) L[E] = {x J [E]: J [E] |= y x y a} J [E],
as J [E] satisfies 0 comprehension. This shows that L[E] |= z(z = P(a)).
In order to show that the separation schema (Sep) holds in L[E], let (x1 , . . . , xk )
be a formula, and let a, a1 , . . . , ak L[E]. As (J [E]: O R) is a continuous
cumulative hierarchy, we may pick some with a, a1 , . . . , ak J [E] such that for
all b J [E],
J [E] |= (b, a1 , . . . , ak ) L[E] |= (b, a1 , . . . , ak ).
(Cf. Problem 5.14.) But then
{b a: L[E] |= (b, a1 , . . . , ak )}
= {b a: J [E] |= (b, a1 , . . . , ak )} J+ [E] L[E],
using Lemma 5.15. This verifies that the separation schema holds in L[E].
That the replacement schema (Rep) holds in L[E] can be shown similarily by
using the replacement schema in V .
It is often necessary to work with the auxiliary hierarchy S [E] which is defined
as follows:
S0 [E] =
S+1 [E] = S E (S [E])
S [E] for limit
S [E] =
<
i{3,4,5,16}
Fi (U {U })
15
i=0, i=3,4,5
Fi (U {U })2 ,
76
5 Constructibility
where
F0 (x, y) = {x, y}
F1 (x, y) = x\y
F2 (x, y) = x y
x
F3 (x) =
F4 (x) = {a: b (a, b) x}
F5 (x) = (x x) = {(b, a): a, b x a b}
F6 (x, y) = {{b: (a, b) x}: a y}
F7 (x, y) = {(a, b, c): a x (b, c) y}
F8 (x, y) = {(a, c, b): (a, b) x c y}
F9 (x, y) = (x, y)
F10 (x, y) = {b: (y, b) x}
F11 (x, y) = (x, (y)0 , (y)1 )
F12 (x, y) = ((y)0 , x, (y)1 )
F13 (x, y) = {((y)0 , x), (y)1 }
F14 (x, y) = {(x, (y)0 ), (y)0 }
F15 (x, y) = {(x, y)}
F16 (x) = E x.
(Here, (y)0 = u and (y)1 = v if y = (u, v) and (y)0 = 0 = (y)1 if y is not an
ordered pair.) It is not difficult to show that each Fi , 0 i 15, is rud E , and that
S E is rud E as well (cf. Problem 5.5).
Lemma 5.20 The ten functions F0 , . . . , F8 , and F16 form a basis for the rud E functions in the sense that every rud E function can be generated as a composition of
F0 , . . . , F8 , and F16 .
Proof (Cf. also Problem 5.6.) It obviously suffices to prove that the nine functions
F0 , . . ., F8 form a basis for the rud functions. Let us write C for the class of all
functions which can be obtained from F0 , . . ., F8 via composition. We aim to see
that every rud function is in C.
If (v1 , . . . , vk ) is a formula of L with the free variables among v1 , . . . , vk , then
we write4
g,k (x) = {(y1 , . . . , yk ) x k : (x; x 2 ) |= (y1 , . . . , yk )}.
77
x)) X k (x),
which may be generated with the additional help of F0 and F3 , and if (v1 , . . . , vk )
vi v j , where 1 i < j k, then
g,k (x) = F4 (F4 (gvk+2 vk+1 (vk+2 =vi vk+1 =v j vk+2 vk+1 ),k+2 (x))).
Also,
gvi ,k (x) = F4 (gvk+1 (vi =vk+1 ),k+1 (x)),
where results from by replacing each (free) occurence of vi by vk+1 .
We have shown Claim 5.21.
The proof of Claim 5.21 made use of all of F0 , . . . , F8 except for F6 . The role of
F6 is to verify the following.
Claim 5.22 If f is rud and k-ary, then the function
h f,k (x) = f x k = {z: y x k f (y) = z}
is in C.
5
a b = a\(a\b).
78
5 Constructibility
Proof We use the obvious induction along the schemata from Definition 5.8.
Let f (x) = xi \x j . Let (v1 , v2 , v3 ) v3 v1 \v2 . Then
h f,k (x) = F6 (g,3 (x
x) (x x
x), x x).
If f (x) = {xi , x j }, then h f,k (x) = (x x).
Let f (x) = g0 (g1 (x), . . . , g (x)). As every rud function is simple (cf. the proof
of Lemma 5.11), we may let (v0 , v) be a formula expressing that6
(w1 , . . . , w )v1 . . . vk (v = (v1 , . . . , vk )
w1 = g1 (v1 , . . . , vk ) . . . w = g (v1 , . . . , vk ) v0 = g0 (w1 , . . . , w )).
Let us write H1 = h g1 ,k (x) h g ,k (x), H2 = h g0 , (H1 ), and H3 = H1
(H1 )2 . . . (H1 ) H2 x x 2 . . . x k . Then
h f,k (x) = F4 (g,2 (H3 ) (H2 x k )).
Finally, if f (x) = yx1 g(y, x2 , . . . , xk ), then we may argue analogously. Claim
5.22 is thus proven.
Claim 5.22 now immediately implies that every rud function is in C. Let f be rud,
say f is k-ary. Let f be defined by
f(u) =
f (x1 , . . . , xk ), if u = (x1 , . . . , xk )
otherwise.
f{(x1 , . . . , xk )} = f (x1 , . . . , xk )
is in C as well.
S E (U )
is transitive
It is now straightforward to verify that if U is transitive, then
as well, cf. Problem 5.5.7 We thus inductively get that every S [E] is transitive.
Moreover, by Lemma 5.20 and the definition of the S-hierarchy,
S [E] J [E] = S [E]
(5.2)
for all limit ordinals and all < . It is easy to see that there is only a finite jump
in -rank from S [E] to S+1 [E].
Lemma 5.12 together with (5.2) readily gives the following.
We assume w.l.o.g. that every gi , 1 i , is k-ary.
The reason why the functions F9 through F15 were added to the above list is in fact to guarantee
that if U is transitive, then S E (U ) is transitive as well.
6
7
79
Lemma 5.23 Let E be a set or proper class, and let be a limit ordinal. Let
B ( 0 ) J [E] . Then (J [E]; , E, B) is amenable, i.e., J [E] is a model of 0
E and B.
comprehension in the language L with ,
E, B
/ S [E], or else
x S [E] y
E
E
/ S [E] and (i, u x , vx ) <,lex
( j, u y , v y )
x < y x, y
E
where (i, u x , vx ) is <
minimal with x = Fi (u x , vx )
,lex
,lex
If M = J [E], then we shall also write < M for <E .
Using Lemma 5.25 and by a proof similar to the one for Lemma 5.25, one may
show the following. (Cf. Problem 5.10.)
8
80
5 Constructibility
Lemma 5.28 then immediately gives M = (J [ E],
The Condensation Lemma 5.29 leads to the following natural concept.
81
Definition 5.30 Let E be a set or a proper class. Then E is said to satisfy full
1 J [E],
condensation iff for every limit ordinal and for every : M = J [ E]
E M = E M (in particular, M = J [E]).
Trivially, E = (or more generally, E ) satisfies full condensation. There
are non-trivial examples, though (cf. Problem 8.9).
The following Theorem was shown by Kurt Gdel (19061978).
Theorem 5.31 (Gdel) Let E be a set or a proper class which satisfies full condensation. Then L[E] |= GCH. In fact, if is an infinite cardinal in L[E] and
= +L[E] , then
P() L[E] J [E].
Proof Let x P() L[E], and pick some such that x J [E]. Let us work in
L[E]. Pick
: M1 J [E],
where M is transitive, ( + 1) {x} ran( ), and Card(M) = . By the Con for some and some E.
We must have
densation Lemma 5.29, M = J [ E]
Card()
< + by Lemma 5.16. Also, E M = E M, as E satisfies full conden = J [E], where < = + . We have shown
sation, so that in fact M = J [ E]
that x = 1 (x) J [E]. As x was arbitrary, P() J [E]. Finally, because
Card(J [E]) = again by Lemma 5.16, Card(P()) = = + . Because we
worked in L[E], the Theorem is shown.
Corollary 5.32 If ZFC is consistent, then so are ZFC + V = L as well as ZFC
+ GCH.
Proof Let (M; E) be a model of ZFC. Construct L inside (M; E). This yields a
model of ZFC plus GCH which thinks that V = L, by Theorems 5.27, 5.28, and
5.31.
We aim to study refinements of Theorem 5.31. For one thing, we may localize
condensation for L[E], cf. Definition 5.33. We will obtain the combinatorial principle
(and its variants) to hold in L[E], cf. Definitions 5.34 and 5.37 and Theorems
5.35 and 5.39. For another thing, we may localize GCH in L[E]; this will lead to
the concept of acceptability and the fine structure theory of L[E], cf. Definition
11.1.
Definition 5.33 Let E be a set or a proper class. E is said to satisfy local condensa 1 J [E] with critical
tion iff for every limit ordinal , for every : M = J [ E]
point , and for all limit ordinals < ,
if
= ,
(P() J+ [ E])\J
[ E]
= E J+ [ E]
(in particular, J+ [ E]
= J+ [E]).
then E J+ [ E]
82
5 Constructibility
(5.3)
Let (B, C) J [E]. We have that R C is stationary, so that we may pick some
J [E],
: J [ E]
< , {R, B, C} ran( ),
where = 1 () is the critical point of , Card(J [ E])
and RC (cf. Problem 4.14). Say R = 1 (R), B = 1 (B), and C = 1 (C).
We have that R = R , B = B , and C = C is club in .
As <E is an initial segment of < L[E] , (B, C) is also the <E -least pair such that
B , C is club in , and (5.3) holds true. By the elementarity of , (B , C )
83
{ R : B = B } (C ) = .
(5.4)
{R , B , C } J+ [ E].
= E J+ [ E],
and hence
As E satisfies local condensation, E J+ [ E]
= J+ [E]. This gives that (B , C ) is the < E -least (and thus
J+ [ E]
+
also the <E -least) pair such that B , C is club in , and (5.4) holds true.
Therefore, (B , C ) = (B , C ) by the choice of (B , C ). Therefore,
{ R: B = B } C,
A = { , i!: Ai }.
84
5 Constructibility
(5.6)
But i 0 < D = i< Ci implies that Ci0 , so that (5.6) contradicts (5.5).
We now aim to characterize for which R we have that (R) holds true in
models of the form L[E], where E satisfies local condensation. It turns out that the
notion of ineffability (cf. Definition 4.49) is the relevant concept.
Theorem 5.39 (Jensen) Let E be a set or a proper class which satisfies local condensation. The following is true inside L[E].
If is regular and uncountable and if R is not ineffable, then (R) holds
true.
In particular, inside L, if is regular and uncountable and if R is not
ineffable, then (R) holds true.
Proof Let us work inside L[E]. Let (A : R) witness that R is not ineffable,
i.e., A for every R and whenever S R is stationary, then there are
with , S and A = A . For every R, let ( ) be the least > such
that
A J [E],
and set
A = P( ) J( ) [E].
We claim that (A : R) witnesses (R).
By Problem 5.17 and Lemma 5.16, Card(A ) for every R. Let us fix
B . We aim to find some club C such that B A for every C R.
Let > be least such that B J [E]. By Problem 5.17, < + . Using
Lemma 5.16, let us pick some bijection g: J [E]. The set
D = { < : B g g J [E] = (g ) }
is easily verified to be club in . (Cf. Problem 4.14.) For every D there is some
( ) and some with critical point such that
E,
J( ) [ E]
= g J [E].
and hence if ( ) = + , where is a
Notice that = + , some limit ,
[ E].
As E satisfies local condensation,
limit, then B = 1 (B) J( ) [ E]\J
= J( ) [E]. I.e.,
85
B = 1 (B) J( ) [E]
(5.7)
for all D.
Suppose that there were no club C D such that B A for every C R.
This means that
{ D R: B
/ A } is stationary.
(5.8)
By (5.7), B
/ A = P( ) J( ) [E] implies that ( ) < ( ), which in turn
gives A J( ) [E] J( ) [E]. Hence (5.8) yields that
S = { D R: A J( ) [E]} is stationary.
(5.9)
(5.10)
86
5 Constructibility
By Lemma 5.43 and Problem 5.3, ODz = {x: (x, z)} for some formula which
is is 2 . This implies that HODz = {x: (x, z)} for some formula which is is 2 .
Theorem 5.44 (Gdel) Let z be a set such that z ODz . Then HODz |= ZF.
Proof Notice that HODz is trivially transitive. (Ext) and (Fund) are therefore true in
HODz . It is easy to see that OR HODz , so that (Inf) is also true in HODz . (Pair),
(Union), and (Sep) are also straightforward.
87
Let us show that the Power Set Axiom (Pow) holds in HODz . We need to see
that if x HODz , then P(x) ODz ODz . Fix x HODz . By Lemma 5.43 and
Replacement in V , there is some such that y P(x) ODz iff there is a formula
, there are 1 , . . . , n , z V and there are y1 , . . . , ym z such that for all u,
u y V |= (u, 1 , . . . , n , y1 , . . . , ym ).
Because z ODz , this shows that P(x) ODz ODz .
In order to show the Replacement Schema (Rep) in HODz , it is easy to see that
it suffices to prove HODz V HODz for all . However, y HODz V iff
y V and for all x T C({y}) there is a formula , there are ordinals 1 , . . . , n ,
and elements y1 , . . . , ym of z such that 1 , . . . , n , z V and for all u,
u x V |= (u, 1 , . . . , n , y1 , . . . , ym ).
This shows that HODz V ODz . Trivially, HODz V HODz , and therefore
in fact HODz V HODz .
Theorem 5.45 (Gdel) Let z be a set with z ODz . If there is a well-order of z in
ODz , then HODz |= ZFC. In particular, HOD |= ZFC.
Proof By Theorem 5.44, we are left with having to verify that HODz |= (AC). Let
z be a well-order of z which exists in ODz .
We write ab for the symmetric difference (a\b)(b\a) of a and b. For finite sets
u, v of ordinals, i.e., u, v OR< , let us write u v iff u = v or else max(uv) v.
It is easy to show that is a well-ordering on OR< , cf. Problem 5.19. The wellorder z induces a well-order z of finite subsets of z in the same fashion: for
u, v [z]< , let u v iff u = v or else y is largest (in the sense of z ) in uv, then
y v. Notice that , z O Dz . For formulae (v0 , v1 , . . . , vn , v1 , . . . , vn ),
= (v0 , v1 , . . . , v p , v1 , . . . , vq ), ordinals , , finite sets = {1 , . . . , n }, =
{1 , . . . , p } of ordinals, and finite vectors y = (y1 , . . . , ym ), w = (w1 , . . . , wq ) of
elements of z, we may then set
(, , , y) (, , , w)
iff (the Gdel no. of) is smaller than (the Gdel no. of) , or else if < , or
else < , or else if y z w. We have that is a well-order.
Now if x O D z , we may let (x , x , x , y x ) be the -least tuple (, , , y)
), = ( , . . . , ), and y =
such that if (v0 , v1 , . . . , vn , v1 , . . . , vm
1
n
(y1 , . . . , ym ) then for all u,
u x V |= (u, 1 , . . . , n , y1 , . . . , yn ).
For x, y ODz we may then set x y iff (x , x , x , y x ) ( y , y , y , y y ).
We have that is a well-order of ODz . For any ordinal , the restriction of
88
5 Constructibility
to sets in ODz V is an element of ODz . But this implies that for any ordinal ,
the restriction of to HODz V is in HODz . It follows that HODz |= (AC).
We refer the reader to [39] for an outline of the status quo of current day inner
model theory.
5.3 Problems
5.1. Show that the relation (M; E) |= (x) is definable in the language L by a
1 - as well as by a 1 -formula, i.e., there is a 1 -formula and a 1 -formula
such that for all models (M; E) of L and for all x M,
(M; E) |= (x) (M, E, , x) (M, E, , x).
Here, is the Gdel number of . We shall produce a stronger statement
in Section 11.1, cf. Lemma 10.14.
.
relation R HF, where (n, a) R (n) = a, is ZFC
1
5.3. Show that a formula (v) of L is 2 iff there is a formula (v) of L such
that
ZFC v((v) V |= (v)).
5.4. Let be an infinite cardinal, and let be a 2 -sentence. Show that if H |= ,
then V |= .
5.5. Show that the following functions are rudimentary: f (x) =
xi , f (x) =
xi x j , f (x) = {x1 , ..., xk }, f (x) = (x1 , ..., xk ), and f (x) = xi x j .
5.6. Let E be a set or a proper class. Let Fi , 0 i 16, be the collection of
functions from p. 73 which produce the S [E] hierarchy.
Show that each Fi , 0 i 15, is rud, and that S E is rud E as well. Also, fill
in the details in the proof of Lemma 5.20. Show that if U is transitive, then
S E (U ) is transitive as well.
5.7. Show that if U is a transitive set and E is a set or a proper class, then rud E (U
{U }) is transitive as well.
5.8. Show that every rud E function is simple.
5.9. Prove Lemma 5.16!
5.10. Prove Lemma 5.26!
5.3 Problems
89
5.11. Show that for every x V there is some A such that x L[A]. Show also
that it need not be the case that x L[x]. Show in BGC that there is a class
A of ordinals such that V = L[A].
5.12. Let M and N be two transitive models of ZFC. Show that if for all sets x of
ordinals, x M x N , then M = N .
5.13. Let A OR. Set
E = {( + , ): A [, + )}
for every limit ordinal . Show that E satisfies the hypotheses of Lemma
5.15 and that if is a limit of limit ordinals, then J [E] = J [A] (and hence
L[E] = L[A]).
5.14. Let (M : OR) be a continuous cumulative hierarchy of transitive
sets,
M M for , and M = < M for
i.e., every M is transitive,
limit ordinals . Set M = M . Let (x1 , . . . , xk ) be a formula, and let
a, a1 , . . . , ak M.
Show that there is then some with a, a1 , . . . , ak M such that for all
b M ,
M |= (b, a1 , . . . , ak ) M |= (b, a1 , . . . , ak ).
In particular, the Reflection Principle holds true: If is a formula, then there
is a club class of such that for all x1 , . . ., xk V ,
(x1 , . . . , xk ) V |= (x1 , . . . , xk ).
5.15. (A. Levy) Use Problem 5.14 to show that Ackermanns set theory AST is
conservative over ZF, i.e., if is a formula of L which is provable in AST,
then is provable in ZF. [Hint. Use the compactness theorem.]
5.16. Let be weakly inaccessible. Show that J |= ZFC. Conclude that the existence of weakly inaccessible cardinals cannot be proven in ZFC.
5.17. Show that the conclusion of Theorem 5.31 also holds if E is just assumed to
satisfy local condensation. I.e., if E is a set or a proper class which satisfies
local condensation, then L[E] |= GCH, and in fact, if is an infinite cardinal
in L[E] and = +L[E] , then P() L[E] J [E].
5.18. Let M = J [E] be a J -structure. Show that there is a (partial) surjection
h : []< such that h 1M .
OR< ,
let u
5.19. For u, v
ordering on OR<
5.20. Let be an infinite cardinal such that + holds true. Show that 2 = + .
90
5 Constructibility
<
,
let
M
be
a
transitive model of ZFC
(A : < ) witness +
5.3 Problems
91
Chapter 6
Forcing
The method of forcing was invented by Paul Cohen (19342007) to show the
independence of the Continuum Hypothesis from the axioms of ZFC, using Cohen
forcing (cf. Definition 6.5 and Theorem 6.33).
93
94
6 Forcing
Dn = { p C : n dom( p)}.
If n < , then,
as Dn is dense, there is some p G with n dom( p). Therefore,
if n < , then, as Dn is dense, there
as G is a filter, G . Also,
is some p G
with p = xn dom( p), so that G = xn . We have seen that G \ X . In
particular, \ X = . We have shown that 20 > 0 .
In what follows, we aim to produce generic extensions M[G] of given (countable) transitive models M of ZFC.
Lemma 6.6 Let P = (P; ) M, where M is a transitive model of ZFC. Then P
is a partial order M |= P is a partial order. Also, if D P, where D M,
and if p P, then D is dense in P M |= D is dense in P and D is dense
below p M |= D is dense below p.
Proof (P; ) is a partial order, D is dense in P, and D is dense below p, may
all be written as 0 -formulae.
Definition 6.7 Let M be a transitive model of ZFC, and let P = (P; ) M be a
partial order. A filter G P is called P-generic over M (or, M-generic for P) iff G
is D-generic, where D = {D M : D is dense in P}.
By Lemma 6.4, if M is a countable transitive model of ZFC and P M is a
partial order, then for each p P there is a P-generic filter G over M with p G.
95
96
6 Forcing
<MOR
97
x. The result
In order to verify M[G] to satisfy (Aus), (Rep), and (Pow), we need the forcing
language.
Definition 6.18 Let M be a transitive model of ZFC, and let P M be a partial
order. Let p P, let (v1 , . . . , vn ) be a formula of the language of set theory, and
let 1 , . . . , n M P . We say that p forces (1 , . . . , n ) (over M), abbreviated by
p PM (1 , . . . , n ),
iff for all G which are P-generic over M and such that p G we have that M[G] |=
(1G , . . . , nG ).
We also write P or just instead of PM . Notice that for a fixed ,
{( p, 1 , . . . , n ) : p PM (1 , . . . , n )} P (M P )n M.
We shall verify that this relation is in fact definable over M (from the parameter P).
In order to do that, we now define a relation by working in M, and then prove that
and have the same extension.1
1
will be used as a symbol for this purpose only temporarily, until p. 97.
98
6 Forcing
99
(1) p (1 , . . . , n )
(2) q p q (1 , . . . , n )
(3) {q p : q (1 , . . . , n )} is dense below p.
Theorem 6.22 (Forcing Theorem, part 1) Let M be a transitive model of ZFC, let
P M be a partial order, and let G P be P-generic over M. Let (v1 , . . . , vn ) be
a formula, and let 1 , . . . , n M P .
(1) If p G and p PM (1 , . . . , n ), then M[G] |= (1G , . . . , nG ).
(2) If M[G] |= (1G , . . . , nG ), then there is some p G such that p PM
(1 , . . . , n ).
Proof We prove (1) and (2) simultaneously. We first prove (1) and (2) for v1 = v2
by induction on (rk (1 ), rk (2 )), ordered lexicographically.
(1): Suppose that p G and p 1 = 2 . Let us verify that 1G 2G . By
symmetry, this will also show that 2G 1G , and therefore 1G = 2G .
Let x 1G , say x = 1G , where (1 , s1 ) 1 for some s1 G. We need to see
that x 2G . Pick r G such that r p, r s1 . We still have r 1 = 2 by
Lemma 6.21, so that there is some q G, q r , such that
q s1 (2 , s2 ) 2 (q s2 q 1 = 2 ).
As r s1 , we have that q s1 . Hence we may pick some (2 , s2 ) 2 with
q s2 q 1 = 2 . As q G, we also have that s2 G, and moreover we
have that 1G = 2G by our inductive hypothesis. But then x = 1G = 2G 2G , as
(2 , s2 ) 2 and s2 G.
(2): Now suppose that 1G = 2G . Consider the following statement about a condition r :
1 (r ) : (1 , s1 ) 1 (r s1 (2 , s2 ) 2
q (q s2 q 1 = 2 qr )).
Assume we had 1 (r ) for some r G, and let (1 , s1 ) 1 be a witness. As r s1 ,
we also have s1 G, so that 1G 1G = 2G . Pick (2 , s2 ) 2 such that s2 G and
1G = 2G . By our inductive hypothesis, there is some q0 G with q0 1 = 2 .
Pick q G such that q q0 and q s2 . Still q 1 = 2 by Lemma 6.21. By
1 (r ), we must then have qr . However, q G as well as r G. Contradiction!
Therefore, we cannot have 1 (r ) for r G. The same argument shows that we
cannot have 2 (r ) for r G, where
2 (r ) : (2 , s2 ) 2 (r s2 (1 , s1 ) 1
q (q s1 q 1 = 2 qr )).
Now let us consider
D = {r : 1 (r ) 2 (r ) r 1 = 2 }.
100
6 Forcing
We claim that D is dense. To this end, let r be given. Suppose that r 1 = 2 does
not hold true. By the definition of , there is then some (1 , s1 ) 1 such that
{q r : q s1 (2 , s2 ) 2 (q s2 q 1 = 2 )}
(6.1)
(6.2)
is not dense below r . Let us assume (6.1) to be true. Well then show that there is
some p r such that 1 ( p) holds true. (By symmetry, if (6.2) holds true, then 2 ( p)
holds true for some p r .) Let (1 , s1 ) 1 witness that (6.1) holds true. There is
some p r such that
q p(q s1 (2 , s2 ) 2 (q s2 q 1 = 2 )).
(6.3)
101
102
6 Forcing
= {(, p) : (, p) ( p p
p PM (, , 2 , . . . , n )
( p PM (, , 2 , . . . , n ) rk ( ) rk ()) )}.
Notice that M P , again by Theorem 6.23 (1). (Without the clause [. . .] in the
bottom line of this definition of , would have ended up being a proper class in
M rather than a set in M.)
Suppose that x G , say x = G , where (, p) for some p G.
Let be such that M[G] |= ( G , G , 2G , . . . , nG ), and let p G be such
that p PM (, , 2 , . . . , n ). We may as well assume that for all , if p PM
(, , 2 , . . . , n ), then rk ( ) rk (), and p p.
Then G G , and
G
a = M[G] is as desired.
103
104
6 Forcing
(6.4)
Let us now just assume that ZFC is consistent. This means that there is a (not
necessarily well-founded) model (N ; E) of ZFC. We need the following statement
about (N ; E):
There is a function M( ) which maps finite subsets of ZFC to elements
M = M( ) of N such that for all ,
(N ; E) |= M = M( ) is a countable transitive model of .
(6.5)
105
(6.6)
106
6 Forcing
For p A, let supp( ( p)) = { < 1 : ( ) supp( p)}, and if supp( ( p)),
then let ( p)( ) = p(( )). It now suffices to verify that there is some B A
of size 1 such that any two conditions in B are compatible, because then any two
conditions in B = 1 B A are compatible as well.
Let us write D = A. By the Pigeonhole Principle, there is some n <
such that { p D : Card(supp( p)) = n} has size 1 . Let D0 D be { p D :
Card(supp( p)) = n} for the least such n. For p D0 , let us write supp( p) =
p
p
p
p
{1 , . . . , n }, where 1 < . . . < n .
A simple application of the Pigeonhole Principle yields the following.
p
107
not be collapsed in M[G]. We will later study forcings which do collapse cardinals,
cf. Definitions 6.41 and 6.43.
Lemma 6.32 Let M be a transitive model of ZFC, and let be a cardinal of M.
Let P = (P; ) M be a partial order such that M |= P has the -c.c. Let G be
P-generic over M. Let X M, X M[G], and write = Card M[G] (X ). There is
then some Y M such that Y X and
, if ,
+
M |= Card(Y ) < , if and < cf(), and
(6.7)
, if and cf().
In particular, if is a cardinal in M such that = and is regular in M or
else > , then remains a cardinal in M[G].
Proof Let X M[G] be given, X M. Pick f M[G], f : X bijective. Let
f = G . Pick p G such that
(6.8)
<
108
6 Forcing
We first prove the first part of the theorem. Let M be a countable transitive model
over
of ZFC. Let M, where 2M . Then C() M. Let G be C()-generic
M. Inside M[G] we may define F : by setting F( ) = { p( ) : p G}
for < . For all n < , <
Dn, = { p C() : n dom( p( ))} M
is dense in C(), so that F is well-defined. For all , < with = ,
E n = { p C() : m < p M
(n)
= m}.
Each E n is dense, and we may pick some maximal antichain An E n for n < .
Let
C()
= m}.
= ( ) = {(n, m, p) : p An , p M (n)
We claim that G = G .
m,
p) , which
First let (n, m) G . There is then some p G such that (n,
()
G . Now let (n, m) G . There
(
n)
=
m,
and
hence
(n,
m)
implies that p C
M
C()
is then some p G such that p M (n)
= m.
As An is a maximal antichain,
()
there is also some q G An , which implies that q C
(n)
= s for some
M
s < . But as p and q are both in G, p and q are compatible, so that we must have
s = m. Therefore (n, m, q) , where q G, i.e., (n, m) G . = ( ) is
often referred to as a nice name for x (or, for ).
We have shown that for every x J [G] there is some nice name
C()
J
such that x = G , each element of is of the form (n,
m,
p), and for
all n there are at most countably many m, p such that (n,
m,
p) . By the
Hausdorff Formula 4.19 we may compute inside J that there are 0 = such
names. In J [G] we may hence define an injection from J [G] into , so that
20 = 2 in J [G], as desired.
109
M M[G] = < M M.
Proof Let f M[G], f : M for some < . We may pick some x M such
that ran( f ) x, and by the Forcing Theorem we may pick some M P and some
p G such that
p PM : x.
For each < ,
D = {q P : q p y x q PM ( ) = y }
<
X.
110
6 Forcing
is easily
seen to be open dense. As < D is open dense, we may pick q
G < D . As q|| p, this means that for every < , there is a (unique) y x
with q PM ( ) = y . Setting
g = {(, y) x : q PM ( ) = y },
Therefore, f = G = g M.
we then get q PM = g.
Problem 6.10 shows that the converse to Lemma 6.38 is true also. The following
just generalizes Definition 6.27.
Definition 6.39 Let be an ordinal, and let be an infinite regular cardinal. For
p (C ) let supp( p) (the support of p) be the set of all < with p( ) = . Let
C () = { p (C ) : Card(supp( p)) < }.
For p, q C (), let us write p q iff for all < , p( ) q( ) in the sense of
C , i.e., p( ) = q( ).
Lemma 6.40 Let M be a transitive model of ZFC, let be a regular cardinal of
M, let be an ordinal in M, and let (C ()) M be Ms version of C (). Let G be
(C ()) M -generic over M. Then exactly the cardinals of M which are not in the
half-open interval (, ( < ) M ] remain cardinals of M[G].
Proof Of course, C () is < -closed. The -Lemma 6.31 implies that M |=
(C ()) M has the (( < )+ ) M -c.c., so that no M-cardinals outside the half-open
interval (, ( < ) M ] will get collapsed. On the other hand, Problem 6.8 shows that
all the M-cardinals inside the half-open interval (, ( < ) M ] will get collapsed
to .
If < = in M, (C ()) M will therefore not collapse any M-cardinals. Moreover, we may have 2 = in a forcing extension (cf. Problem 6.9).
We shall now study forcings which collapse cardinals.
Definition 6.41 Let be a regular cardinal, and let . We let Col(, ) = < ,
i.e., the set of all functions f such that there is some < with f : . For
p, q Col(, ), let p q iff p q (iff p = q). The partial order
(Col(, ); ) is called the collapse of to .
Notice that Col(, ) = C .
Lemma 6.42 Let M be a transitive model of ZFC, let be a regular cardinal of M,
let be a cardinal of M, and let P = (Col(, )) M be Ms version of Col(, ).
Let G be P-generic over M. Then every M-cardinal is still a cardinal in M[G],
and in M[G], Card() = = 2< . Moreover, cardinals above (( < )+ ) M remain
cardinals in M[G].
111
D = { p P : ran( p)}.
For all
< and < , D M and D M are both dense in P. Therefore,
f = G is a surjective function from onto , so that Card() = in M[G].
To show that 2< = in M[G], let us fix < . If X P() M =
P() M[G], then4
G)( + ) = 1} = X.
112
6 Forcing
Lemma 6.44 Let be a regular cardinal, and let > be a regular cardinal such
that < for all < and < . Then Col(, < ) has the -c.c.
Proof This immediately follows from the -Lemma 6.31.
As Col(, < ) is certainly < -closed, this immediately implies the following.
Lemma 6.45 Let M be a transitive model of ZFC, and let < be regular
cardinals of M such that inside M, < for all < and < . Let
P = (Col(, < )) M be Ms version of the Levy collapse of to . Then all
M-cardinals strictly between and will have size in M[G], and all M-cardinals
outside of the open interval (, ) will remain cardinals in M[G]. In particular,
= + in M[G].
The Levy collapse Col(, < ) will play a crucial role in Chap. 8.
We may force which was shown to be true in L, cf. Definition 5.34 and Theorem
5.35.
Theorem 6.46 Let M be a transitive model of ZFC, let be an uncountable regular
cardinal in M, and let P M be defined inside M as follows. P = {(c : ) : <
c }, ordered by end-extension. Let G be P-generic over M. Let
S , S M, be stationary in M. Then (S) holds true in M[G].
Proof
An easy density argument shows
that there are C , < , such that
G = (C : < ). We claim that G S = (C : S) witnesses that (S)
holds true in M[G].
Let , M P and p G be such that
p is club in ,
and .
Let p0 p be arbitrary. It suffices to show that there is some q p0 such that if
q = (c : ), then
q S = c .
(6.9)
Let us work inside M. Notice that P is < -closed. We may thus easily construct a
sequence ( pi : 1 i < ) of conditions in P such that there are F = (c : < ) and
(i : i < ), such that for all 0 i < j < , pi = (c : i ), supk< j k < j
(in particular, p j < pi ), and there is some j \ supk< j k and some a j M
such that, writing j for supk< j k ,
p j j = aj .
As S is stationary and {i : i < } is club in , we may pick some limit ordinal
i 0 such that i0 S. Set
A=
113
ai and q = (
i<i 0
pi ) {(i0 , A)}.
i<i 0
114
6 Forcing
Lemma 6.49 Let be an infinite cardinal, and let P be an atomless partial order
such that
= 0 .
1P Card()
Then for every p P there is an antichain A {q P : q P p} of size .
Proof Let us fix p P.
Let us first assume that = . Let us construct a sequences ( pn : n < ) and
(qn : n < ) of conditions at follows. Set p0 = p. Given pn , let qn and pn+1 two
incompatible extensions of pn . We then have that {qn : n < } is an antichain of size
0 .
Let us now assume that cf() = < . Let (n : n < ) be a sequence of
uncountable regular cardinals which is cofinal in . We construct a sequence ( pn : n <
) of conditions and a sequence (An : n < ) of antichains in P as follows. Set
p0 = p. Given pn , notice that
= 0 ,
pn Card()
so that by Lemma 6.32 there must be an antichain A {q P : q pn } of size n .
Let An be some such antichain, and let pn+1 An be arbitrary. It is now easy to see
that
(6.10)
{An : n < }) .
(6.11)
However,
we may define a function f as in (6.11) by setting, for n < and
p {An : n < },
f ((n, p)) =
, if p (n)
= ,
0, if there is no < such that p (n)
= .
(6.12)
115
If < , then by (6.10) there is some r r and some n < such that r (n)
= .
= . Therefore,
But then r ||q for some q An by the choice of An , so that q (n)
f is surjective. Contradiction!
Definition 6.50 Let P = (P, ) be a partial order. P is called separative iff whenever
p is not stronger than q then there is some r p such that r and q are incompatible.
Every separative partial order P such that for every p P there is some q P
with q <P p is easily seen to be atomless.
Lemma 6.51 Let be an infinite cardinal, and let P be a separative partial order
such that Card(P) = and
1P is countable.
Then there is a dense homomorphism : Col(, ) P.
Proof Let be a name such that
1P : G is onto.
Let us construct ( p) by recursion on lh( p), where p Col(, ). Set () = 1P .
Let us now suppose that p Col(, ) and ( p) has been defined, where n = lh( p).
As Card(P) = , by Lemma 6.49, we may let A P be a maximal antichain of size
i.e., there is some
consisting of q P such that q P ( p) and q decides (n),
< such that
q (n)
= .
We may write A = {qi : i < }, where qi is different from (and thus incomaptible
with) q j for i = j. We may then set ( p {(n, i)}) = qi .
It is easy to see that is a homomorphism. Also, an easy induction shows that
for each n < ,
An = {( p) : p Col(, ) lh( p) = n}
. . . , ((n 1)).
is a maximal antichain of q P such that q decides (0),
there is some s P r and
Let us show that is dense. Pick r P. As r r G,
. . . , (n).
116
6 Forcing
Proof Let us first show that C is homogenous. Let us fix p, q C. Let us then define
: C C as follows. If r C, then dom((r )) = dom(r ), and if n dom(r ),
then
r (n) otherwise.
Then if n dom( p) dom(q), ( p)(n) = q(n), so that ( p) q. It is easy to see
that is a dense endomorphism.
Now if is an ordinal, and if p, q C(), then for each supp( p) supp(q)
there is a dense endomorphism : C C such that ( p( )) q( ) in the
sense of Cohen forcing. These endomorphisms then easily induce an endomorphism
: C() C() such that ( p) q in the sense of C(). Again, will be
dense.
The endomorphism constructed in the previous proof is actually an automorphism,
i.e. bijective.
In much the same way as Lemma 6.53 we may prove the following.
Lemma 6.54 Let be a regular cardinal, and let X be a set of ordinals which are
all of size . Then Col(, X ) is homogeneous.
Proof Let p, q Col(, X ) be given. We may then define : Col(, X )
Col(, X ) as follows. Given r Col(, X ), let supp((r )) = supp(r ) and
dom((r )()) = dom(r ()) for all X , and if X and dom(r ()),
then let
r ()( ) otherwise.
(6.14)
It is easy to see that is a dense automorphism of Col(, X ) such that
( p)||q.
Definition 6.55 Let M be a transitive model of ZFC, and let P = (P; ) M be
a partial order. Let : P P be a dense endomorphism, M. The induces a
map
: M P M P
as follows:
( ) = {( ( ), ( p)) : (, p) }.
Lemma 6.56 Let M be a transitive model of ZFC, let P = (P; ) be a partial
order, and let : P P be a dense endomorphism with M. Let p P, let
(v1 , . . . , vn ) be a formula, and let 1 , . . . , n M P Then
117
p PM (1 , . . . , n ) ( p) PM ( (1 ), . . . , (n )).
Proof We first show:
Claim 6.57 Let G P be P-generic over M, and let H = { p : q G (q) p}.
Then for all M P , G = ( ) H .
The proof is an easy induction on the rank of . Notice that G G iff (, p)
for some p G iff ( ( ), ( p)) ( ) for some p G (i.e., ( p) H ) iff
( ) H ( ) H .
The same argument shows:
Claim 6.58 Let G P be P-generic over M, and let H = { p : ( p) G}. Then
for all M P , H = ( )G .
Now suppose that p PM (1 , . . . , n ). Let G P be P-generic over M such
that ( p) G.6 Setting H = { p : ( p) G}, H is P-generic over M by
Lemma 6.48, and p H . By p PM (1 , . . . , n ), M[H ] |= (1H , . . . , nH ).
But M[H ] = M[G] by Lemma 6.48 and 1H = (1 )G , . . . , nH = (1 )G
by Claim 6.58, so that M[G] |= ( (1 )G , . . . , (n ) H ). We have shown that
( p) PM ( (1 ), . . . , (n )).
Conversely suppose that ( p) PM ( (1 ), . . . , (n )). Let G P be P-generic
over M by Lemma 6.48, and p G. Setting H = { p : q G (q) p},
H is P-generic over M such that ( p) H . By ( p) PM ( (1 ), . . . , (n )),
M[H ] |= ((
1 ) H , . . . , (n ) H ). But M[G] = M[H ] by Lemma 6.48 and 1G =
G
n ) H by Claim 6.57, so that M[G] |= (1G , . . . , nG ). We
(1 ) , . . . , nG = (
have shown that p PM (1 , . . . , n ).
Definition 6.59 Let M be a transitive model of ZFC, and let P = (P; ) be a
partial order. Then M P is called homogenous iff for all dense endomorphisms
: P P with M, (
) = . If 1 , . . . , n M P , then P is called homogenous
with respect to 1 , . . . , n iff for all p, q P there is some dense endomorphism
: P P such that ( p) q, and (1 ) = 1 , . . . , (n ) = n .
Hence P is homogenous iff P is homogenous with respect to the empty sequence of
names. Moreover, if P is homogenous with respect to 1 , . . . , m and 1 , . . . , n
M P are homogenous, then P is homogenous with respect to 1 , . . . , n , 1 , . . . , m .
Lemma 6.60 Let M be a transitive model of ZFC, and let P = (P, ) M be a
separative partial order. For every x M, x is homogenous.
Proof We must have (1) = 1 for every dense homomorphism : P P. This is
because if (1) < 1, then there is some r 1 such that (1), r are incompatible.
By density, there would be some s such that (s) r . Then (s) and (1) are
incompatible, which is nonsense.
6
We may assume without loss of generality that such a G exists, as otherwise we might work with
the transitive collapse of a countable (sufficiently) elementary substructure of M.
118
6 Forcing
119
Proof First let G be P-generic over M and H be Q-generic over M[G]. It is clear that
G H is a filter. Let us show that G H is P Q-generic over M. Let D P Q
be dense. We need to see that D (G H ) = .
Let D = {q Q : p G ( p, q) D}. D is dense in Q: Given q Q, let
{supp( p) : (, p) } is at
C()
= { p X C : Card(supp( p)) < 0 } { p \X C : Card(supp( p)) < 0 }.
It is easy to verify that
{ p X C : Card(supp( p)) < 0 }
=C
120
6 Forcing
and if 1M , then
{ p \X C : Card(supp( p)) < 0 }
= C().
The rest is then immediate by the Product Lemma 6.65.
{ J [G ] : < },
and set
M = HOD(N ) {( ) } .
By Theorem 5.44, M |= ZF. We aim to verify that
121
= 1M and M |= cf(1 ) = .
(6.15)
M = ( ) .
(6.16)
122
6 Forcing
Let us now verify that Col(, < ) is homogenous with respect to . Let p, q
Col(, < ) be given. We may then define : Col(, < ) Col(, < ) as in
(6.14) in the proof of Lemma 6.54 (where = and X = [, )).
= {((
We have that (G)
p),
( p)) : p Col(, < )} = {( p,
( p)) : p
G = { p : ( p) G} =
Col(, < )}, as is an automorphism, so that (G)
1 G, where ( 1 G) Col(, < ) is Col(, < ) generic over J for every
< by Lemma 6.48 (and the definition of ). Also, J [( 1 G) Col(, <
)] = J [G Col(, < )] for every < by Lemma 6.48, which is certainly true
independently from the particular choice of G, so that in fact
Col(,<)
1Col(,<) J
(, p) J < p J
J [G Col(, < )]
Col(,<)
(
) J < ( p) J
Col(,<)
(
) J < ( p) J
Col(, < )]
( ) J [(
G)
( ) J [G Col(, < )]
((
), ( p)) .
6.3 Problems
6.1. Let (P; ) be a partial order, and let D be dense in P. Use the Hausdorff
Maximality Principle 2.11 to construct an antichain A D such that p
Dq A q p. Conclude that A is a maximal antichain in P.
6.2. Prove Lemma 6.21!
In what follows, we shall always assume that M is a (countable, if convenient)
transitive model of ZFC, P = (P, ) M is a partial order, and G is P-generic
over M.
6.3. Let > Card(P) be a regular cardinal in M (and hence in M[G]). Show that
(M, M[G]) has the -approximation property which means that if A ,
A M[G], is such that A M for all < , then A M.
6.4. Show that if N is a transitive model of ZFC and if H is Q-generic over N ,
where Q N is a partial order, then N [H ] |= ZFC .
6.3 Problems
123
M M[G] = < M M,
124
6 Forcing
6.3 Problems
125
F G,
if X , Y G, then X Y G,
if X G and Y X , Y P() M, then Y G,
/ G, and
if X P() M, then either X G or \ X G.
Ai }
i<
126
6 Forcing
g F,
if H F, and H is a subgroup of G with H H , then H F,
if H , H F, then H H F, and
if H F and g, then H 1 = { 1 : H } F.
Chapter 7
X \Us =
{Ut : lh(t) = lh(s) t = s}.
127
128
If A , then
{B : B A B is closed}
is the smallest closed set in which A is contained, called the closure of A.
A -algebra on a set Y is a collection S P(Y ) which is closed under relative
complements as well as countable unions and intersections.
Definition 7.2 A set A is called Borel iff A is in the smallest -algebra
containing all closed (open) subsets o f .
The simplest Borel sets are the open and closed sets. Countable intersections of
open sets are often called G - and countable unions of closed sets F -sets. The
Borel sets form a natural hierarchy, cf. Problem 7.3.
Let . A tree T on is a set of pairs (s, t) with s < , t < ,
and lh(s) = lh(t), such that T is closed under initial segments, i.e. if (s, t) T and
n lh(s) then (s n, t n) T . If T is a tree on , then we write [T ] for the
set of all (x, y) such that (x n, y n) T for all n < . If T is a tree
on then p[T ], the projection of T , is the set of all x such that there is
some f so that for all n, (x n, f n) T . If x then we let Tx denote
the set of all t < such that (x lh(t), t) T . Obviously,
x p[T ] y (x, y) [T ] (Tx , ) is ill-founded.
(7.1)
129
130
(s, t) T = Ast =
{Ast mn : (s n, t m) T }
and
(s, t) U = Bts =
{Bts mn : (s n, t m) U }.
u n k
A y n j and Bv n l ,
i.e.,
x n i
u n k
A y n j C i,k
j,l \Bv n l .
x n
u n
C i,k
j,l separates A y n , Bv n ,
i, j k,l
i.e.,
x n
A y n
u n
C i,k
j,l \Bv n .
i, j k,l
x n i
u n k
A y n j and Bv n l
Bv n G for all but finitely many n. In particular, there is some n < such that
131
x n
u n
A y n F \G \Bv n .
x n
u n
So A y n and Bv n can be separated by a Borel (in fact, open) set after all. Contradiction!
Definition 7.7 Let A , and let be an ordinal. If A = p[T ], where T is
a tree on , then A is called -Souslin.
The 0 -Souslin sets are hence exactly the analytic sets.
Lemma 7.8 If A is coanalytic, then A is 1 -Souslin.
Proof Let \A = p[T ], where T is a tree on . Therefore, x A iff (Tx , )
is well-founded. If Tx = (Tx , ) is well-founded, then, as Tx = 0 , it can be ranked
by some function f : Tx 1 such that
s t = f (s) < f (t)
(7.2)
and vice versa, cf. Lemma 3.17. Therefore, x A iff there is some f : Tx 1 with
(7.2).
Now we construct S to be a tree searching for some such ranking. Let e :
< be a bijection such that if n < lh(s), then e1 (s n) < e1 (s). We let (s, h) S
iff s < and, setting
Ts = {t < : lh(t) lh(s) (s lh(t), t) T },
h: lh(s) 1 is such that
k < lh(s)l < lh(s) (e(k) Ts e(l) Ts e(k)e(l) = h(k) < h(l)).
For (s , h ), (s, h) S we write (s , h ) S (s, h) iff s s and h h.
It is easy to verify now that x A iff (S, ) is ill-founded.
The tree S constructed in the previous proof is called the Shoenfield tree for A.
Corollary 7.9 Every coanalytic set A is the union of 1 many Borel sets.
Proof Let A be coanalytic. Let S be the Shoenfield tree on 1 for
A, as being constructed in the proof of Lemma 7.8, so that x A iff x p[S]. If
< 1 , let us write S for the set of all (s, t) S with ran(t) . Obviously,
p[S] =
p[S ].
(7.3)
<1
132
We then have A =
<1
p[S ]
A=
<1
B A, so that in fact
<1
as desired.
Let us now consider the spaces ( )k , where 1 k < , equipped with the
product topology. It is not hard to see that ( )k is actually homeomorphic to the
Baire space .
Let . A tree T on k is a set of (k + 1)-tuples (s0 , s1 , . . . , sk ) with
s0 , s1 , . . . , sk1 < , sk < , and lh(s0 ) = = lh(sk ), such that T is closed
under initial segments, i.e., if (s0 , . . . , sk ) T and n lh(s0 ) then (s0 n, . . . , sk
n) T . We shall write [T ] for the set of all (x0 , x1 , . . . , xk1 , f ) such that for all n,
(x0 n, x1 n, . . . , xk1 n, f n) T .
A set A ( )k can easily be verified to be closed iff there is a tree T on
k1
with A = [T ]. A ( )k is perfect iff there is a tree T = on k1
n+1
iff
n+1
set.
Definition 7.10 A set A ( )k , some k < , is called projective iff there is some
n < such that A is 1 .
n
133
which cannot be 1 .
(7.4)
where z(n) = 1 iff there is (s, t) T such that e(s, t) = n. Then for all x, y ,
(x, y, z) [U ] iff (x, y) [Uz ] = [T ], so that A = p[Uz ]. On the other hand, if
z and A = p[Uz ], then A is clearly 1 .
1
A = {x : (x, z) B}.
(7.5)
(This uses (7.4).) A set B with these properties is called a universal 1 -set.
1
Otherwise
A = {x : (x, x)
/ B}
would be a 1 -subset of , and there would thus be some z such (7.5) holds.
1
In particular, (z, z)
/ B iff z A iff (z, z) B.
We have shown:
Lemma 7.11 There is an analytic set of reals which is not coanalytic.
We may also think of the universal 1 -set B constructed above, in fact of any
1
134
Let n, m n, m be the Gdel pairing function, cf. p. 35. Every real x
induces a binary relation Rx on as follows:
(n, m) Rx x(n, m) = 1.
(7.6)
We let
WF = {x : Rx is well-founded } , and
WO = {x : Rx is a well-ordering}.
The sets WF and WO are coanalytic, cf. Problem 7.6.
WF and WO are in fact complete coanalytic sets in the sense that if B is
coanalytic, then there are continuous (in fact, Lipschitz) functions f :
and g: such that for all x ,
x B f (x) WF g(x) WO.
We may construct such a function g for WO as follows. (Then f = g will also work
for WF.) Let B be coanalytic. There is then a tree T on with x B
iff (Tx ; ) is well-founded. With the help of some bijection e : < which is
such that for all s < and i < we have that e1 (s i) e1 (s), Tx induces
an order R x on as follows2 :
n R x m [(e(n) Tx e(m) Tx e(n) e(m))
(e(n) Tx e(m) Tx e(n)e(m) e(n) <lex e(m))
(e(n)
/ Tx (e(m) Tx (e(m)
/ Tx n < m))].
Let us then define g(x) to be such that Rg(x) = R x , i.e., g(x) is that real y
such that
1 iff (n, m) R x , and
y(n, m) =
0 otherwise.
This defines g: . Notice that g is continuous. In fact, if x n = y n,
then Tx and Ty agree upon the first n levels, so that R x n = R y n and hence
g(x) n = g(y) n.
It is easy to see that if g(x) WO, then (Tx ; ) must be well-founded, so that
x B. On the other hand, suppose that g(x)
/ WO, and let (n i : i < ) be such
that (n i+1 , n i ) R x for all i < . Clearly, e(n i ) Tx for all i < . Moreover,
We here write st iff s and t are incomparable, i.e., s (lh(s) lh(t)) = t (lh(s) lh(t)). Also,
<lex is the lexicographic ordering.
135
k<
Contradiction!
Similar to our proof of Lemma 7.11 one may show that for all n < , 1 is
different from 1 , in fact 1 \ 1 = and 1 \ 1 = . (Cf. Problem 7.6.)
n
n n
n n
1 = 1 1
n
for n < . By Souslins Theorem 7.5, 1 is the family of all Borel sets.
Let us consider a 1 set A
2
i<n
Ai , where
Proof By induction on n: There is nothing to prove for n = 0. Now let n > 0 and
suppose the statement to be true for n 1. Let A = p[T ], where T is on n . For
let T be the set of all (s, t) T with ran(t) . Because cf(n ) > ,
< n ,
p[T ] = <n p[T ].
136
<n i<n1
If U is a tree on
the tree
i<n1
Ai , where
Ai ,
a representation as desired.
k
(7.7)
Let us now fix a tree T on such that A = p[T ]. Let us inductively define
trees Ti , i OR, as follows.
T0 = T ,
Ti+1 =
(Ti ) , and
T = i< Ti for limit ordinals .
Notice that, inductively, each Ti is in fact a tree on . Moreover, Ti T j
whenever i j.
As Card(T ) , there must be some < + such that T+1 = T . Let us write
{ p[(Ti )(s,t) ]: (s, t) Ti \Ti+1 },
where p[(Ti )(s,t) ] has exactly one element in case (s, t) Ti \Ti+1 . Because
Card(T ) , this shows that Car d(A) .
137
Case 2: T = .
]) > 1 for all (s, t) T .
Then Card( p[T(s,t)
Let us recursively construct (su , tu ) T , where u < 2. Set (s , t ) = (, ).
Suppose that (su , tu ) has been chosen. As p[T(su ,tu ) ] has at least two elements, we
may pick (su 0 , tu 0 ) T , (su 1 , tu 1 ) T such that
lh(su 0 ) = lh(su 1 ) > lh(su ) and su 0 = su 1 .
For z 2, let x z = {sz n : n < }. We have x z A = p[T ], as being witnessed
by (tz n : n < ), for each z 2. Moreover, {x z : z 2} is a perfect set.
Corollary 7.16 Every uncountable analytic set of reals has a perfect subset.
We now need to define the effective projective hierarchy.
Let x . A set A ( )k is called 11 (x) iff A = p[T ], where T is a tree
on k+1 which is definable over the structure (V ; , x). A is called n1 (x) iff A is
1 (x) if A is the projection of a
the complement of a n1 (x) set, and A is called n+1
n1 (x) set. We also set 1n (x) = n1 (x) n1 (x). Notice that
1 =
1
x
11 (x),
and therefore analogous facts hold for the other projective pointclasses 1 and 1
n
n+1
(7.8)
Proof By induction on n. The only non-trivial step of this induction is the base,
n = 1. We first verify
Claim 7.18 11 (x) is closed under and , i.e., if B ( )k is 11 (x),4
then so are
{y : n (y, n) B} and
{y : n (y, n) B}.
Proof : Let us assume that k = 1. Let (y, n) B iff (y, n) p[T ], where T
is on 3 . We may then define U on 2 by setting (s, t) U iff for all n < lh(s),
In what follows, Q is or depending on whether n is odd or even.
By identifying n < with the constant function cn : with value n, we may construe
( )k as a subset of ( )k+1 .
3
4
138
(s, n, t) T . It is then easy to see that y p[U ] iff for all n < , (y, n) p[T ] =
B. The proof for is easy.
Let us now prove Lemma 7.17 for n = 1. First let A 11 (x), say A = p[T ],
where (s, t) T iff (V ; , x) |= (s, t) for some formula . Then
y A z (V ; , x, y, z) |= n < ((y n, z n)).
Now we prove by induction on the complexity of that if A be as in (7.8) with
y, z, u),
y A z n (V ; , x, y, z) |= u(u = (n) (x,
so that the result follows from the inductive hypothesis and Claim 7.18.
Finally, let u ,
where is n . Then
y, z, u),
y A z n (V ; , x, y, z) |= u(u = (n) (x,
so that the result also follows from the inductive hypothesis and Claim 7.18.
Lemma 7.19 Let z . Then L[z] as well as < L[z] ( L[z]) are 21 (z).
Proof The proof in the general case is only notationally different from the proof in
the case z = , so let us assume that z = . We have (using Theorem 5.31) that
x L iff x J for some < 1 iff (using Lemma 5.28) x M
for some countable transitive model of V = L, which is true if and only if there
is some z such that, setting (n, m) E z(n, m) = 1, E is well-founded (i.e.,
there is no y such that for all n < , (y)n+1 E(y)n ),
(; E) |= V = L ,
and x is a real number in the transitive collapse of (; E). It is easy to verify that
this can be written in a 21 fashion. (Cf. Problem 7.10.)
This shows that L is 21 .
An entirely analoguous argument shows that < L ( L) is 21 .
In general, the complexity of L given by Lemma 7.19 is optimal, as we aim
to show now.
Lemma 7.20 Let , and let A be -Souslin, say A = p[T ], where T
is a tree on . If A = , then A L[T ] = . Moreover, if A does not contain a
perfect subset, e.g., if Card(A) < 20 , then A L[T ].
139
(7.9)
(7.10)
Now let us suppose that A does not contain a perfect subset, so that T = . We
aim to show that A L[T ]. Let x A, say x p[Ti ]\ p[Ti+1 ]. By construction,
there is then some n < and t < with lh(t) = n such that
Card( p[(Ti )(x n,t) ]) = 1.
(7.11)
By Lemma 5.6, there must be some (x , y ) [(Ti )(x n,t) ] L[T ]. However, by
(7.11), any such x must be equal to x, and thus x is easily definable from Ti and
(x n, t), so that x L[T ]. We have shown that A L[T ] and in fact A L[T ].
Corollary 7.21 (Shoenfield absoluteness) Let x , and let A be 21 (x).
If A = , then A L[x] = . Moreover, if A does not contain a perfect subset, then
A L[x]
Proof Let S be the Shoenfield tree for A, cf. p. 135. An inspection of the construction of S, cf. the proof of Lemma 7.8, shows that S L[x] follows from the
assumption that A be 21 (x). The conclusion then follows from Lemma 7.20.
This implies that L[x] can in general not be better than the complexity
given by Lemma 7.19, namely 21 (x), unless L[x]. This is because if
L[x] were 21 (x), then \L[x] would be 21 (x), hence if \L[x] = , then
( \L[x]) L[x] = by Corollary 7.21, which is nonsense.
Also, if (20 ) L[x] = 1L[x] < 20 , then by Lemma 7.19 there is a largest 21 (x)-set
of reals, namely L[x].
Definition 7.22 Let A ( )2 . We say that a partial function F :
uniformizes A iff for all x , if there is some y such that (x, y) A, then
x dom(F) and (x, F(y)) A.
Theorem 7.23 (Kondo, Addison) Let A be 11 . Then A can be uniformized by a function whose graph is 11 .
140
Proof We shall prove that each nonempty 11 (z) set A has a member x such
that {x} is 11 (z). As the definition of {x} will be uniform in the parameter z, this
proof will readily imply the theorem. For notational convenience, we shall assume
that z = 0. Hence let A be given such that A is 11 and A = . Let T be a
tree on such that
x A Tx is well-founded.
Let us fix an enumeration (sn : n < ) of < such that if sn sm , then n < m. x
We first define maps n : A 1 by setting
n (x) =
Here, ||s||Tx is the rank of s in (Tx , ) in the sense of Definition 3.18, which is
well-defined for x A.
Claim 7.24 If lim xk = x, where each xk is in A, and for all n, n (xk ) is eventually
k
constant, i.e.
n kn k kn n (xk ) = n ,
then x is in A and n (x) n .
Proof Suppose that (xk : k < ) is as described, but x = lim xk
/ A. Then Tx is
k
141
says that there is no order-preserving embedding f : (Ty )sn (Tx )t for some t sn ,
and is hence 11 by Lemmas 7.17 and 7.18. We may thus rewrite x B in a 11
fashion.
Let A, B . We say that A and B reduce A and B iff A A, B B,
A B = A B, and A B = . If P( ), then we say that has the
B such that A and B reduce
reduction property iff for all A, B there are A,
A and B.
Recall that if A, B are disjoint, then we say that C separates A and B iff
A C and C B = . If P( ), then we say that has the separation
property iff for all A, B there is some C such that also \C and C
separates A and B.
Lemma 7.25 The following hold true.
(a) 1 has the reduction property.
1
Let a, b
and we may then pick a Borel set C such that C separates A and B.
be such that C = {x : (a, x) U } and \C = {x : (b, x) U }. It is
easy to verify that then b a C iff b a \C. Contradiction!
It follows from Lemma 7.25 (c) and the proof of Lemma 7.25 (a) that Theorem
7.23 is false with 11 replaced by 1 .
1
Here, (x)0 and (x)1 are defined to be the unique reals such that (x)0 (x)1 = x.
142
x( 2 ) if n is even
z(n) = 1
if n = 3k 5l and f (k) < f (l)
0
otherwise.
Then x, f L[z], and thus 1V = 1L[z] . Contradiction!
We shall later see a model in which 1 is inaccessible to the reals, cf. Theorem
8.20.
By Theorem 7.15, every uncountable 1 -set of reals has a perfect subset. The
1
143
We have
A = {x : y y = F(x)}.
As A is uncountable, (the graph of) F is an uncountable 11 subset of ( )2 .
Suppose that F has a perfect subset, say P F, where P is perfect. Write
Q = {x : y (x, y) P}.
As F is a function, Card(Q) = Card(P) = 20 , so that Q is an uncountable analytic
subset of A. Contradiction!
Hence F is an uncountable coanalytic set without a perfect subset.
Finally, (3) = (1) is given by Corollary 7.21. Let A be an uncountable
1
2 set. If A does not have a perfect subset, then A L by Corollary 7.21. However,
1L < 1 implies that A is then countable. Contradiction!
There is a more cumbersome argument of proving (3) = (1) of Theorem 7.28,
using forcing, cf. the proof of Lemma 8.18.
Excellent textbooks on classical descriptive set theory are [27] and [20]. Modern
variants of descriptive set theory are dealt with e.g. in [4, 12, 14, 19, 21].
7.3 Problems
7.1. Show that the topology we defined on the Baire space is exactly the one
which is induced by the distance function d. Conclude that is a Polish
space.
7.2. Show that there is a continuous bijection f : 2.
7.3. (Borel hierarchy) Let 0 denote the set of all open A and 0 the set
1
the set of all n< An , where {An : n < } < 0 , and let 0 be the
set of all \A, where A 0 . Show that 0
1 +1
= 0 , and that 0 is
1
144
A 0 , . . . , A k , B 0 , . . . , B l ) (M; A0 , . . . , Ak , B0 , . . . , Bl ).
: ( M;
[Hint. Construct a tree of height seaching for some such B 0 , . . . , B l , and
.] For the conclusion to hold it actually suffices that N is a transitive model
A 0 , . . . , A k )
which contains an admissible set N which in turn contains ( M;
and (M; A0 , . . . , Ak , B0 , . . . , Bl ) such that M is countable in N , cf. Problem
5.28.
7.5. Let be a regular infinite cardinal, and let T = be a tree on . Let
/ t j )))},
T = {s T : {ti : i < } T (i j (s ti (i = j ti
and define T 0 = T , T +1 = (T ) , and T = {T : < } for limit
ordinals . Show that there is some < + with T +1 = T , call it T . If
s T \T +1 , then we say that is the CantorBendixson rank of s, and
if s T , then we say that is the CantorBendixson rank of s. Show
that if T = , then T is perfect.
7.6. Show that the sets WF and WO are both coanalytic and in fact 11 . Show also
that for every n 1, 1 \ 1 = and 1 \ 1 = .
n n
n n
is coanalytic iff there is some map s
7.3 Problems
145
(7.12)
146
where B is 10 (z). Using Problem 5.27 , show that the right hand side of (7.12) can
be written in a 11 (z) fashion, so that every 11 (z) set would also be 11 (z).]
Chapter 8
Solovays Model
defined
(Us ), we let (Us n ) = 2n+1 (Us ). Now let A be open, say
A = {Us : s X }, where Us Ut = for all s = t, s, t X . We define the
measure (A) of A to be
(Us ).
sX
(8.1)
If B is Lebesgue measurable, then one also writes (B) for (B) and calls
it the Lebesgue measure of B. It is not hard to verify that the family of sets which
are Lebesgue measurable forms a -algebra containing all the open sets, so that
in particular all Borel sets are Lebesgue measurable. (Cf. Problem 8.2.) For our
purpose, well define Lebesgue measurability as follows.
R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_8,
Springer International Publishing Switzerland 2014
147
148
8 Solovays Model
149
basic open set Us , s < , with Us \B being meager, i.e., [Us ] [B]. But this
means that i: C B/meager defined by i(s) = [Us ] is dense.
Lemma 8.5 Both B/null as well as B/meager have the c.c.c.
Proof For B/meager this immediately follows from the preceding lemma. Now
suppose {[Bi ]: i < 1 } to be an antichain in B/null. Set
Bi = Bi \
Bj
j<i
150
8 Solovays Model
Baire, then p[T ] = p[T ] in all V [G], where G is P-generic over V for some
P H : if, say x ( p[T ] V [G])\ p[T ], then x p[U ], so that by the argument
just given p[T ] p[U ] = in V ; but p[T ] = A and p[U ] = \A in V .
Therefore, if A is -universally Baire and if G is P-generic over V , where
P H , then we may unambiguously define the new version A G of A in V [G] as
p[T ] V [G], where T, U witness A is -universally Baire. We often just write A
rather then A G , provided that G is clear from the context.
Let us consider a closely related situation. Let M and N be inner models such that
M N . (We allow M and N to exists in V [G], a generic extension of V .) Let A be
a Borel set in M, say M |= A = p[T ] \A = p[U ], where T and U are trees
on 2 , T , U M. By the absoluteness of wellfoundedness, cf. Lemma 5.6, we must
have that N |= p[T ] p[U ] = . Let us define a simple variant of the Shoenfield
tree S on 1N by (s, h) S iff s < and, setting
Ts = {t < : lh(t) lh(s) (s lh(t), t) T } and
Us = {t < : lh(t) lh(s) (s lh(t), t) U },
h: lh(s) 1N is such that
k <
lh(s) 1
lh(s) 1
l <
[(e(k) Ts e(l) Ts e(k) e(l) h(2k) < h(2l))
2
2
(e(k) Us e(l) Us e(k) e(l) h(2k + 1) < h(2l + 1))].
(Here, e: < is a bijection such that if n < lh(s), then e1 (s n) < e1 (s).)
It is straightforward to verify that S L[T, U ] M, and that both in M and N ,
p[T ] p[U ] = [S] = .
By the absoluteness of wellfoundedness, cf. Lemma 5.6, we must then have that
N |= p[T ] p[U ] = . (In particular, every Borel set is universally Baire, as
being witnessed by a pair of trees on 2 , cf. Lemma 8.7.)
As above, we may now also show that if M |= A = p[T ] \A = p[U ],
where T and U are trees on 2 , T , U M, then N |= p[T ] = p[T ]. We may
therefore now unambiguously write A N for p[T ], as computed in N . If M, N are
clear from the context, we often just write A rather than A N . Of course if M = V
and N = V [G] is a generic extension of V , then A N = A G , so that A has an
unambiguous meaning.
Still let M and N be inner models such that M N (possibly in V [G] rather
than V , e.g. M = V , N = V [G]). Suppose that A is a Borel set in M, or A is
universally Baire in M and N is a generic extension of M. The following facts are
easy to verify, cf. Problem 8.4:
1.
2.
3.
151
Lemma 8.7 Every analytic (and hence also every coanalytic) set is universally
Baire.
Proof Let A be analytic, say A = p[T ], where T is on . Let be any
uncountable cardinal. Let S on be the version of the Shoenfield tree (cf.
the proof of Lemmas 7.8 and 7.13): (s, h) S iff s < and, setting
Ts = {t < : lh(t) lh(s) (s lh(t), t) T },
h: lh(s) is such that
k < lh(s)l < lh(s)(e(k) Ts e(l) Ts e(k) e(l) h(k) < h(l)).
(Again, e: < is a bijection such that if n < lh(s), then e1 (s n) < e1 (s).)
It is straightforward to see that T, S witness that A is -universally Baire.
The previous proof shows that in fact if A is coanalytic, then A is universally
Baire in a strong sense: Say A is 11 (x) with x , then the trees witnessing
that A is universally Baire may be taken as elements of L[x].
By Lemma 8.7, if A is analytic or coanalytic (for instance, if A is just Borel)
and if G is P-generic over V , where P V , then A G (i.e., A ) is well-defined.
Lemma 8.8 (1) Let G be B/null-generic over V . Then there is a unique x G
V [G] such that for all B B,
x G B [B] G.
(2) Let G be B/meager-generic over V . Then there is a unique x G V [G]
such that for all B B,
x G B [B] G.
Proof The same proof works for (1) and (2). Let us first show uniqueness. Let
x, y V [G] be such that x B [B] G y B for all B B. In
particular, x Us [Us ] G y Us for all s < . If x = y, then there is
some s < with lh(s) > 0 and x Us but y Us ; this gives a contradiction!
Let us now show existence. Working in V [G], let us recursively construct {sn : n <
} < with lh(sn ) = n and [Usn ] G for all n < as follows. Set s0 = . Of
course, [U ] = [ ] G. Given sn with lh(sn ) = n and [Usn ] G, we pick sn+1
as follows. Let [B] [Usn ], where [B] B/null (or [B] B/meager), i.e., B is
not null (or not meager). Therefore, one of B Usn k , k < , is not null (or not
meager). This argument shows that
D = {[B] B/null (or meager): k [B] [Usn k ]}
152
8 Solovays Model
We claim that
for all B B, x B [B] G.
(8.2)
Well, (8.2) is true for each basic open set by the construction of x. An easy density
argument shows that for each B B, exactly one of [B], [ \B] has to be in
G. Therefore, if (8.2) is true for B B, then it is also true for \B. Now let
Bn B, n < , such that for each n < , x Bn [Bn ] G. Another
easy
density arguement (similar to the one above) yields that at least one element
\B ] : n < } must be in G. We therefore must have that
of {[ n< Bn ]} {[
n
x
( n< Bn ) = n< Bn iff x Bn for all n < iff [Bn ] G for all n <
iff [ Bn ] G. We have shown (8.2).
If M is an inner model, if G is (B/null) M -generic over M and if x G is unique
with x G B [B] G for all Borel sets B of M, then x G is called a random
real over M. (Here, B is computed in V .)
If, on the other hand, G is (B/meager) M -generic over M, and if x G is unique
with x G B [B] G for all Borel sets B of M, then Lemma 8.4 above
shows that x is a Cohen real over M. (Again, B is computed in V .)
Lemma 8.9 Let M be a transitive model of ZFC.
(1) x is a random real over M iff x
/ B for all B B M which are null sets
in M.
/ B for all B B M which are meager
(2) x is a Cohen real over M iff x
sets in M.
Proof (1) First let x = x G be random over M. Let B B M be a null set
in M. As [( M)\B] = [ M] G, we have that x (( M)\B) =
/ B.
( M[G])\B , i.e., x
153
{B : B B M B is null in M}.
As (20 ) M is countable, A is thus a countable union of null sets, and hence A is null.
(2) By Lemma 8.9 (2), x B iff
x
{B : B B M B is meager in M}.
[D] M
(, a 1 , . . . , a k ).
As (B/ null) has the c.c.c. by Lemma 8.5, E is at most countable in M, say
154
8 Solovays Model
E = {[Dn ]: n < },
where Dn B for every n < . Set
B=
Dn
n<
Dn
n<
B is a countable union of Borel sets, and hence Borel. We claim that for all x
which are random over M, x A x B.
Let x = x G be random over M. Then x A iff M[x] = M[G] |= (x, a1 , . . . , ak )
iff there is some [D] E G such that
(B / null) M
[D] M
(, a 1 , . . . , a k )
iff
there is some n < with [Dn ] G iff [
( n< Dn ) = B, as desired.
The proof of (2) is entirely analogous.
n<
Dn ] G iff x = x G
155
f n M for each n < , which is trivial.) In the same way, x p[U ] implies
that x p[U ].
This shows that x A x p[T ], as desired.
Corollary 8.15 Every analytic as well as every coanalytic set is Lebesgue measurable and has the Baire property.
Definition 8.16 Let A . A is said to have the Bernstein property iff for every
perfect set P , P A or P\ A contains a perfect subset.
Lemma 8.17 Let M be a transitive model of ZFC such that (20 ) M is countable.
Let A be Solovay over M. Let P be a perfect set such that P = [T ]
for some perfect tree on with T M. Then P A or P\ A contains a perfect
subset.
Proof Let be a formula, and let a1 , . . . , ak M be such that for all x which
are generic over M, x A M[x] |= (x, a1 , . . . , ak ). As (20 ) M is countable,
there is some G V which is C-generic over M. As T is perfect, we may pick
some (ts : s < 2) such that for all s < 2, ts T and ts
0 and ts 1 are two
incompatible extensions of ts in T of the same length. Let x G = G be the Cohen
real over M given by G. Let y 2 be defined by
y(n) =
It is easy to verify that then
t y n [T ]\M.
n<
Write z = n< t y n .
Let us suppose that z A. Let M C be a canonical name for n< t y n , and
let p C be such that
p C
M [T ]\( M) (, a1 , . . . , ak ).
(8.3)
Let (Dn : n < ) V enumerate the sets in M which are dense in C. By recursion on
lh(s), where s < 2, we now construct conditions ps p and sequences ts <
such that ps lh(ts ) = ts as follows.
Put p = p and t = . Now suppose ps and ts have been constructed. There
must be extensions p 0 , p 1 ps , m lh(ts ) and n 0 = n 1 such that
= n0 and p 1 (m)
= n1 ,
p 0 (m)
because otherwise ps ( M) by the homogeneity of C, cf. Lemma 6.53.
We may thus pick ps 0 = ps 1 , both in Dlh(s) , and ts 0 = ts 1 with lh(ts 0 ) =
lh(ts 1 ) such that
156
8 Solovays Model
C
ps 0 C
M lh((ts 0 )) = (ts 0 )and ps 1 M lh((ts 1 )) = (ts 1 ).
: M H
be an elementary embedding where M is countable and transitive and S, T, U
ran( ), say (T ) = T and (U ) = U . Notice that (S) = S.
Let g V be C-generic over M. As in the proof of Lemma 8.17,
(P\M) M[g] = .
(8.4)
157
a perfect subset. On the other hand, if every uncountable coanalytic set of reals has
a perfect subset, then 1V is inaccessible in L.
Theorem 8.23 (Solovay) Let be an inaccessible cardinal, and let G be Col
(, <)-generic over V . Then in V [G], every set of reals which is OD is
Lebesgue measurable and has the Baire property, and every uncountable set of
reals which is OD has a perfect subset.
Proof We continue from where we left off the proof of Theorem 8.20. The following
is the key technical fact.
V [G ]
158
8 Solovays Model
With the help of Claim 8.24, the proof of Solovays Theorem 8.23 may be finished
as follows.
Let us fix A V [G] which is OD in V [G]. Let be a formula, let
1 , . . . , k be ordinals, and let x1 , . . . , xl V [G] such that for all x
V [G],
x A V [G] |= (x, 1 , . . . , k , x1 , . . . , xl ).
By Claim 8.21, we may pick some < such that G is Col(, <)-generic over
V and x1 , . . . , xk V [G ]. Let x V [G] be generic over V [G ].2 By
Claim 8.24, there is some H V [G] which is Col(, <)-generic over V [G ][x]
such that
V [G] = V [G ][x][H ].
We then have
x A V [G] |= (x, 1 , . . . , k , x1 , . . . , xl )
V [G ][x][H ] |= (x, 1 , . . . , k , x1 , . . . , xl )
Col(, <)
1 , . . . , k , x1 , . . . , xl ).
p H V [G ][x] (x,
However, Col(, <) is homogeneous by Lemma 6.54, and therefore by Lemma
6.61 the last line is equivalent to
Col(, <)
1 , . . . , k , x1 , . . . , xl ).
1Col(, <) V [G ][x] (x,
By the definability of over V [G ][x], cf. Theorem 6.23 (1), A is thus in fact
Solovay over V [G ]. However, (20 )V [G ] is certainly countable from the point
of view of V [G], so that in V [G], A is Lebesgue measurable and has the Baire
property by Corollary 8.13.
Now let us assume that A is uncountable in V [G]. We use the proof of Lemma
8.17 to show that A contains a perfect subset in V [G]. As A is uncountable in V [G],
A\V [G ] = . By Claim 8.21, there is hence some with < < such that
(A V [G ])\V [G ] = .
By the Product Lemma 6.65, G [, ) is Col(, [, ))-generic over V [G ].
Setting P = V [G] and replacing C with Col(, [, )), the argument for
Lemma 8.17 now proves that A contains a perfect subset.
In order to finish the proof of Solovays theorem, it therefore remains to show
Claim 8.24.
As the case > 0 is only notationally different frome the case = 0, let us
assume that = 0.
It can be shown that every real in V [G] is generic over V [G ] in the sense of Definition 8.11,
but we wont need that.
159
(8.5)
160
8 Solovays Model
Then H0 is Col( < + 2)-generic over V [s] by Lemma 6.48 and V [s][H ] =
V [s][H0 ]. If we finally set
H = { p Col(, <): p + 2 H0 p [ + 2, ) G [ + 2, )},
then H is Col(, <)-generic over V [s] and
V [s][H ] = V [s][H0 ][G [ + 2, )]
= V [s][H ][G [ + 2, )]
= V [G],
which is
(8.6)
p r = r s
p r
/ = r
/ s.
(8.7)
Having defined Q , set p Q+1 iff for all open dense sets D Col(, ), D V ,
p p ( p D Q ).
If is a limit ordinal, and Q is defined for every < , then we set Q = < Q .
For each , if p Q and p p , p Col(, ), then p Q . This gives that
if , then Q Q . Let be least such that Q+1 = Q . Set
Col(, ).
= Q and Q = Q
Q
and Q as partial orders, with the order relation given by the restriction
We construe Q
and Q, respectively.
of the order relation of Col(, ) and Col(, )Col(, ) to Q
(8.8)
161
Subclaim 8.26 G 0 Q.
Proof Suppose that p G 0 \Q , where is minimal such that G 0 \Q = . We
cannot have = 0, by (8.6) and the definition of Q0 . Also, cannot be a limit ordinal,
so that = + 1 for some . We may pick some open dense set D Col(, ),
D V , such that
/ Q ).
p p ( p D p
(8.9)
s = G0 .
Let D Col(, ), D V , be open dense in Col(, ). By
Proof Let p Q.
In V [G], there are only
may work in V [G] and produce some G 0 which is Col(, )-generic over V such
that p G 0 and G 0 Q.
Q0 , we
Let r P, and suppose that p G 0 decides r . As p G 0 Q
must have (8.7). Therefore, G 0 is as desired.
With Subclaim 8.26, G 0 G 1 Q is a filter. We now show:
Subclaim 8.29 G 0 G 1 is Q-generic over V [s].
Proof Let D V [s] be dense in Q. We need to see that D (G 0 G 1 ) = . Suppose
that D (G 0 G 1 ) = .
Let V P be such that s = D. Recall (8.8), and let (v0 , v1 , v2 , v3 , v4 ) be
a formula such that whenever G 0 G 1 is Col(, ) Col(, )-generic over V ,
then
V [G 0 G 1 ] |= (G 0 , G 1 , , P, )
iff the following holds true.
162
8 Solovays Model
D (G 0 G 1 ) = .
By hypothesis, V [G 0 G 1 ] |= (G 0 , G 1 , , P, ). Let ( p, q) (G 0 G 1 ) be
such that
Col(,)Col(,)
( p, q) V
),
(G 0 , G 1 , ,
P,
(8.10)
where G h V Col(,)Col(,) is the canonical name for G h , h {0, 1}. By Subclaim 8.26, ( p, q) Q. As D is dense in Q, there is some ( p , q ) ( p, q) such
that ( p , q ) D.
By Subclaim 8.28, there is some G 0 G 1 inside V [G] which is Col(, )
Col(, )-generic over V such that ( p , q ) G 0 G 1 and
s = G0 .
(8.11)
163
8.3 Problems
8.2. Let L be the set of all B such that for all X (8.1) holds true. Let
C be be the set of all B such that there is some open set A with
BA being meager. Show that both L and C form a -algebra containing all
the open sets.
8.3. A set A 2 is called a flip set iff for all x, x 2 such that Card({n <
/ A. Show that if A 2 is a flip
: x(n) = x (n)}) = 1, x A x
set, then A is not Lebesgue measurable and A does not have the property of
Baire. Show in ZF + there is a uniform ultrafilter on that there is a flip
set.
8.4. Verify the statements (1) through (3) from p. 150.
8.5. Show that Claim 8.21 implies that 1 is inaccessible to the reals. [Hint. Use
Problem 6.18.]
8.6. Let be weakly compact, let G be Col(, <)-generic over V , and let H be
Q-generic over V [G], where Q V [G] and V [G] |= Q has the c.c.c. Show
V and some g which is
that if x V [G][H ], then there is some Q
Q-generic
over V such that x V [g].
8.7. A set A 1 is called reshaped iff for all < 1 ,
L[A ] |= is countable.
Show that if 1V is not Mahlo in L, then there is a reshaped A 1 . Show
also that if A 1 is reshaped, then there is some poset P which has the c.c.c.
such that if G is P-generic over V , then in V [G] there is a real x with A L[x].
Conclude that if 1V is inaccessible in L, then 1 need not be inaccessible to
the reals. [Hint: There is an almost disjoint collection {xi : i < 1 } of subsets
164
8 Solovays Model
V
x R (x p[T ] = ||x||).
Chapter 9
Our proof will make use of Fubinis Theorem as well as the 01Law of Hewitt
Savage; we refer the reader to any standard textbook on Measure theory, e.g. [38].
In order to prove this theorem, we need the concept of a rapid filter.
Definition 9.2 Let F P() be a filter on . We say that F is rapid iff F is
non-trivial, F extends the Frchet filter, and for every monotone f: there is
some b F such that
n < b f (n) n.
(9.1)
165
166
Proof Let us fix a such that 1V = 1L[a] , cf. the proof of Lemma 7.27. We
have L[a] = 2 L[a] = 1 . Let us write X = 2 L[a].
If x, y 2, x = y, let us write h(x, y) for the distance of x and y, i.e., h(x, y)
is the least n < such that x n = y n (Hence h(x, y) > 0). For Y 2, let us
write H (Y ) for
{h(x, y): x, y Y x = y};
H (Y ) is thus a set of positive integers.
Definition 9.4 We define F
X P() by setting a FX iff there is a covering
(X n :n < ) of X , i.e., X n< X n , where X n 2 for each n < , such that
H (X n ) a.
n<
X=X
X na = X
n<
b
(X na X m
)=X
n< m<
X p,
p<
H (X p ) a b,
p<
so that a b FX .
Also, FX is non-trivial: if X n< X n , then at least one X n has two (in fact
uncountably many) elements, because X is uncountable; therefore H (X n ) = , and
hence
/ FX .
To show that FX extends the Frchet filter, let (X n : n < 2m ) be an enumeration
of all
Us = {x 2: x s},
where s m 2. We have that
Claim 9.6 FX is 1 .
3
n<2m
H (X n ) = \ m FX .
167
Proof
Let us first verify that a FX iff there is a covering (Yn : n < ) of X with
1.
3
In order to finish
the proof of Theorem 9.3, we now need to see that FX is rapid. Let f : be
monotone. We need to find some b FX such that (9.1) hold true.
Claim 9.7 For every f : , L[a, f ] is a null set.
Proof Because L[a, f ] is 1 , cf. Lemma 7.19, it is Lebesgue measurable by
2
Set
{x: (x, y) A}
168
1
2n k
for all k < and such that (G k : k < ) is independent in that if N is finite,
(G k ).
{G k : k N } =
kN
k n , and put
Proof Write r1 = 0. For k , set rk = l=0
l
2n k =
(G k ),
{G k : k N } = 2 kN n k =
kN
kN
Claim 9.9 Let Y 2 be null. There is then some closed set C 2 such that
Y C = , (C) > 0, and in fact for all s 2, if Us C = , then
(Us C)
1
23lh(s)+1
169
(Ck Us ) l=k+1
((Cl Us ) (Cl1 Us )) 81k l=k+1
2lk 81l =
1
1
1
1
1
1
1
81k
p=1 4 p = 8k (1 3 ) > 8k 2 = 23k+1 .
8k
Claim 9.10 For every monotone f : , there is some b FX such that
n < b f (n) n (3n + 1)2 24n .
(9.2)
{G s,m,n : s, m, n) N } =
(G s,m,n ).
(s,m,n)N
n< n n mn
2)2 . Let x 2. Setting G x = {y 2: (x, y)
Obviously, G is a G subset
of (
x
G}, we have that G n n mn G x f (m),m,n for every n < . However,
1
(G x f (m),m,n ) = m+n
, and for each > 0 there is some n < such that
2
1
x
n n
mn 2m+n < . Thus, G is null for every x 2.
Let us define
{G x : x X }.
G =
We aim to see that G is null. Well, for each y G we may let x(y) be the < L[a] -least
x X such that y G x , and we may set
A = {(y, z) (G )2 : x(y) < L[a] x(z)}.
A is 1 by Lemma 7.19 and hence Lebesgue measurable by our hypothesis. For
2
170
Each Onx , n < , is open. Let x X . Suppose that for every n < and every
s < 2, if C Us = , then C Us Onx = . We may then define (z n : n < ) 2
and a monotone (kn : n < ) with C Uz n kn = and z n+1 kn = z n kn
for all n < as follows. Let z 0 C and k0 = 0. If z n and kn have been defined such
that z n C, then C Uz n kn Onx = , so that as Onx is open we may pick z n+1 and
n k+1 > n k such that z n+1 kn = z n kn , Uz n+1 kn+1 Onx , and z n+1 C. Then
n<
z n kn
Onx C = G x C G C = .
n<
Contradiction! There is thus for each x X , a pair (n(x), s(x)) such that n(x) <
, s(x) < 2, and
x
= .
C Us(x) = , yet C Us(x) On(x)
(9.3)
Let e: < 2 be bijective such that e(n, s) n and e(n, s) lh(s) for all
n < and s < 2, and let (X m : m < ) be an enumeration of the set of all
{x X : n(x) = n s(x) = s x f (e(n, s)) = t},
where n < , s < 2, and t < 2. We may write X =
b=
m<
H (X m ),
m<
X m , so that, setting
171
(9.4)
by the monotonicity of f .
Let m < and s < 2 be such that e(m, s) < n. Let us write
bm,s = {h(x, y) < f (n): x, y X, x = y, n(x) = n(y) = m,
s(x) = s(y) = s, and
x f (e(m, s)) = y f (e(m, s))}.
As b f (n) =
show that
e(m,s)<n bm,s ,
(9.5)
Let again m < and s < 2 be such that e(m, s) < n. Notice that if x, x , y
X, x f (n) = x f (n), and h(x, y) < f (n), then h(x , y) = h(x, y). This implies
that
(Card({t
f (n)
(9.6)
= C Us Omx = C Us
G x| f ( p), p,n .
n m pn
172
G t,n,m
G x| f ( p), p,n .
n m pn
f (n)
(9.7)
f (n)
2: C Us G t,n,m = .
We have that
C Us
As (G t,n,m ) =
Claim 9.8,
1
,
2n+m
2\G t,n,m : t
f (n)
2, C Us G t,n,m = .
(C Us )
k
q
q
q
1
1
n+m
= 1 n+m
.
k
2
2
k=0
1
1
1
By the choice of C, (C Us ) 23lh(s)+1
, so that 23lh(s)+1
1 2n+m
. By (9.4)
and the properties of e, we have that m < n and lh(s) < n and therefore
1
23n+1
1
23lh(s)+1
1 q
1 2n
.
2
22n
22n 1
22n
, thus 1
2n 1
2
q (3n + 1) log2
(3n + 1) 22n ,
2n
2 1
as we had wished.
173
Let n < , and let n n be largest such that n (3n + 1)2 24n n. Then
174
Definition 9.12 Let I = {In : n < } be a partition of into intervals, and let
J = (Jn : n < ) be such that Jn In 2 for n < . We write
(I, J) = {x 2: x In Jn for infinitely many n < }.
A set N 2 is called small iff for every sequence (n : n < ) of positive reals
there is a partition I = {In : n < } of into intervals and there is a sequence
J = (Jn : n < ) with Jn In 2 for n < such that
(1) N (I, J), and
(2) ({x 2 : x Ik Jk }) < k for every k < .
If N 2 is small and if (I, J) is as in (1) and (2) of Definition 9.12, then for every
n 0 < ,
(I, J) = {x 2 : x In Jn for infinitely many n n 0 }.
Usmn ,
m<
where smn < 2 and Usmn Us n = for smn = smn . Notice that by (On ) <
m
1
2n ,
(9.8)
Set
Fn = {s n 2: km s = smk }.
With the help of (9.8),
A {x 2: x n Fn for infinitely many n < }.
Also
inf m<
n=m
({x 2 : x n Fn
= 0.
(9.9)
175
nk
x 2: x n Fn < k
m k+1 = min m > n k : 2
n=m
and
n k+1 = min m > m k+1 : 2
m k+1
x 2: x n Fn
< k .
n=m
Ik
nk+1
nk
mk
mk+1
Ik
Let, for k < ,
and set
Jk = {s Ik 2 : i [m k+1 , n k+1 ]t Fi s [n k , i) = t [n k , i)}
and
Jk = {s Ik 2 : i [n k , m k+1 ]t Fi s [m k , i) = t [m k , i)}
We claim that both (I, J) and (I , J ) satisfy (1) and (2) of Definition 9.12.
As for (2),
x 2: x Ik Jk
= x 2: i [m k+1 , n k+1 ]t Fi x [n k , i) = t [n k , i)
n k+1
2n k
i=m k+1
x 2: t Fi x i = t
< k .
176
Symmetrically,
x 2: x Ik Jk
= x 2: i [n k , m k+1 ]t Fi x [m k , i) = t [m k , i)
m k+1
mk
x 2: t Fi x i = t
< k .
i=n k
[m k+1 , n k+1 ]
(9.10)
k=0
or with
x n Fn and n
[n k , m k+1 ].
(9.11)
k=0
(k )2
8
F (I, J) (I , J ),
(9.12)
where (I, J) and (I , J ) are exactly as constructed in the proof of Lemma 9.13. We
are also going to use the notations n k , m k , Ik , Ik , Jk , Jk for k < from the proof of
Lemma 9.13; in particular, ({x : x Ik Jk }) < k and ({x : x Ik Jk }) < k
for every k < . We aim to find
(I , J ) F
(9.13)
such that for every k < , x : x Ik Jk < k .
For k < , let
Hk = t [n k ,m k+1 ) 2 : x 2 : t x [m k+1 , n k+1 ) Jk k .
177
1
k = k .
<
k
4
(9.14)
Hk = t [n k ,m k+1 ) 2 : x 2 : x [m k , n k ) t Jk k ,
so that in analogy with (9.14)
({x : x [n k , m k+1 ) Hk }) <
Now let us write
where
k
.
4
(9.15)
N0 = (I, J),
N1 = (I , J ), and
N2 = (I , J ),
Jk = {s t : s [m k ,n k ) 2 t Hk Hk }.
178
As x0
/ N2 , we must have x0 [m k , m k+1 )
/ Hk and x0 [m k , m k+1 )
/ Hk and
hence
0
}) < 2k = k
({x : x [m k+1 , n k+1 ) J2k+1
for all but finitely many k < . Therefore, the proof of the following Claim will
provide a covering of F as in (9.13) and finish the proof of Lemma 9.14.
Claim 9.15 F (I , J ).
Proof Suppose not, and pick y F\(I , J ). Define z 2 by
z(n) =
y
/ (I , J ), so there can be only finitely many such k. Contradiction!
for finitely many k. But z N1 , so we must have z Ik+1 Jk+1 and hence
x0 [n k , n k+1 ) Jk for infinitely many k. For such k,
1
2n+1
({ 2 : x In Jn }) < n .
179
In
We claim that
F (I, J ).
(9.16)
tn (k) if n X \l and k In
x(k) otherwise.
Obviously, y
/ (I, J). But F is a filter, so that x F implies y F (I, J).
Contradiction! We have shown that (9.16) holds true.
As Jn Jn for every n < , we still have that
({ 2 : x In Jn }) < n =
1
2n+1
(9.17)
(9.18)
180
9.3 Problems
Let F P() be a non-trivial filter on extending the Frchet filter. We say that
F is a p-point iff for every f there is some X F such that f X is constant
or finite-to-one (by which we mean that {n : f (n) = m} is finite for every m < ).
We say that F is a q-point iff for every f which is finite-to-one there is some
X F such that f X is injective. F is called selective, or Ramsey, iff F is both
a p-point as well as a q-point. F is called nowhere dense iff for every f : R
there is some X F such that f X is nowhere dense.
9.1. Let F be a p-point.
(a) Show that if (X n : n < ) is such that X n F for all n < , then there is
n < .
some Y F such that Y \X n is finite for all
/ F for all
(b) Show that if {X n : n < } is such that n< X n = , X n
n < , and X n X m = for all n = m, then there is some X F such
that X X n has finitely many elements for every n < . If F is assumed to
be selective, then we may in fact pick X F in such a way that X X n has
exactly one element for every n < .
9.2. (a) Show that if F is a p-point, then F is nowhere dense. In fact, if F is a
p-point, then F is discrete (by which we mean that for every f : R there
is some X F such that for every x f X there are a < x < b such that
(a, b) f X = {x}).
(b) Show that if F is a q-point, then F is rapid.
9.3. Let U be a selective ultrafilter on .
(a) Let (X n : n < ) be such that X n F for all n < . Show that there
is some Y U such that for all {n, m} Y with n < m, m X n . [Hint.
First use Problem 9.1 (a) to get some Z U and some g : such that
Z \g(n) X n for all n < . Suppose w.l.o.g. that g is strictly inceasing, and
write
f (n) = g . . . g (0).
n times
By Problem 9.1 (b), let Z U be such that for every n < , there is exactly
one m Z with g(n) m < g(n + 1), call it m n . One of {m 2n : n < },
{m 2n+1 : n < } is in U , call it Z . Verify that Y = Z Z is as desired.]
9.3 Problems
181
(b) Let (X s : s < ) be such that X s F for all s < . Show that there
is some Y U such that for all strictly increasing s < with ran(s) Y ,
s(n) X s n for all n lh(s).
9.4. Show that if CH holds, then there is a selective ultrafilter.
[Hint. Let ({X n : n <
} : < 1 ) enumerate all {X n : n < } such that n< X n = and
X n X m = for all n = m. Recursively construct a sequence (Y : < 1 ) of
infinite subsets of such that if < , then Y \Y is finite and Y+1 = Y X n
for some n for which Y X n is infinite, if such an n exists, and otherwise
Card(Y+1 X n ) 1 for all n. Set F = {X : X \X is finite }.]
9.5. Show that if CH holds, then there is a q-point which is not selective. [Hint.
Let U , U0 , U1 , . . . be non-isomorphic selective ultrafilters, and let X U iff
{m : {n : m, n X m } U .]
Let P V be a partial order, an let G be P-generic over V . Then z V [G]
is called unbounded iff for every x V , {n < : x(n) < z(n)} is
infinite. z V [G] is called dominating iff for every x V ,
{n < : x(n) < z(n)} is cofinite, i.e., there are only finitely many n < with
z(n) x(n).
9.6. Let z be a Cohen real over V . Then z is unbounded.
Let be any ordinal, and let G be C()-generic over V . Show that V [G]
does not contain a dominating real. [Hint. Use Lemma 6.29 and the proofs of
Lemmas 6.53 and 6.61.]
Let
b = min{Card(F) : x z F {n : x(n) < z(n)} is infinite}, and
d = min{Card(F) : x z F {n : x(n) < z(n)} is cofinite}.
9.7. b d. Let 2 be a cardinal, and let G be C()-generic over V . Suppose
that V |= CH. Show that in V [G], 1 = b < d.
Let D consist of all (x, n), where x and n < , ordered by (x , n )
(x, n) iff n n, x n = x n, and x (k) x(k) for all k n.
9.8. If (x, n), (x , n ) D, where n = n and x n = x n, then (x, n) is
compatible with (x n ). Conclude that D has the c.c.c. Show that if G is Dgeneric over V , then V [G] contains a dominating real.
Let F P(), and let
M F = {(s, X ) : s []< X F (s = min(X ) > max(s))},(9.19)
ordered by (s , X ) (s, X ) iff s s, X X , and s \s X . M F is called
Mathias forcing for F.
9.9. Let F be a filter on .
(a) Show that M F has the c.c.c.
182
{s : X (s, X )
9.10. Let F be a non-trivial filter on extending the Frchet filter. Asuume that
either (a) F is not an ultrafilter, or else (b) is an ultrafilter, but not selective.
Show that if G is M F -generic over V , then there is a Cohen real over V in
V [G].
9.11. Let U be a non-trivial filter on extending the Frchet filter such that U is
not a p-point. Show that if G is MU -generic over V , then V [G] contains a
dominating real.
Chapter 10
Measurable Cardinals
10.1 Iterations of V
Theorem 10.3 and Lemma 10.4 of this section will be used in the proof of Theorem
13.3.
Definition 10.1 Let be a measurable cardinal, and let U be a measure on , i.e., a
< -closed uniform ultrafilter on . Let be an ordinal, or = . Then the system
I = (M , : < )
is called the (linear) putative iteration of V of length given by U iff the following
hold true.
(1) M0 = V , and if + 1 < , then M is an inner model.
(2) If < , then : M M is an elementary embedding, and
= .
(3) If + 1 < , then M+1 = ult(M ; 0 (U )) and +1 is the canonical
ultrapower embedding.
(4) If < is a limit ordinal, then (M , : < ) is the direct limit of
(M , : < ).
The system I is called the (linear) iteration of V of length given by U if either
is a limit ordinal or else the last model M 1 is well-founded (and may therefore be
identified with an inner model).
Notice that by (2), = id for all < . Also, if we write = 0 () and
U = 0 (U ), then
R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_10,
Springer International Publishing Switzerland 2014
183
184
10 Measurable Cardinals
M |= U is a measure on .
Therefore, (3) makes sense and is to be understood in the sense of Definition 4.57.
The requirement that the M for + 1 < be inner models is tantamount to
requiring that they be transitive. For < a limit ordinal, the requirement that
) is the direct limit of (M , : < ) means, by virtue of
(M , : <
(2), that M = {ran( ): < }.
Definition 10.2 Let be a measurable cardinal, and let U be a measure on . Then
V is called iterable by U and its images iff for every , if
I = (M , : < + 1)
is the (linear) putative iteration of V of length + 1 given by U , then I is an
iteration, i.e., M is well-founded (and may therefore be identified with an inner
model).
Theorem 10.3 Let be a measurable cardinal, and let U be a measure on . Then
V is iterable by U and its images.
Proof Let be an ordinal, and let
(M , : < + 1)
(10.1)
10.1 Iterations of V
185
M0
M1
M2
(10.2)
V
We set 0 = . Now let , and suppose all , < , are already construed
such that (10.2) holds true for all < .
Let us first suppose to be a limit ordinal, so that ( M , ( : < )) is the direct
limt of (( M : < ), ( : < )). We then define : M V by setting
(x) = 1
(x),
whenever x ran( ). For every x M there is some < with x ran( ),
(x ) = ( 1
(x)) =
(x) = (x ).
This means that is well-defined, and it is easy to verify that is 1000 -elementary
and (10.2) holds true for all .
Now suppose to be a successor ordinal, say = + 1. Set = 0 () and
U = 0 (U ). We have that M = ult( M ; U ), which is given by equivalence
relations (mod U ) of functions f M M .
If is a 1000 -formula, and f 1 , . . . , f k M M , then we write X , f1 ,..., fk
for
{ < : M |= ( f 1 (), . . . , f k ())}.
By os Theorem 4.56, X , f1 ,..., fk U iff
186
10 Measurable Cardinals
M |= ([ f 1 ], . . . , [ f k ]).
U is a countable
subset of U . As U is
Because M and hence U is countable,
< 1 -closed, we thus have that U = . Say U .
We may now define : M V by setting
([ f ]) = ( f )().
This is well-defined and 1000 -elementary, because if is a 1000 -formula and
f 1 , . . . , f k M M , then
M +1 |= ([ f 1 ], . . . [ f k ]) iff
X , f1 ,..., fk U iff
(X , f1 ,..., fk ) = { < : V |= ( ( f 1 )(), . . . , ( f k )())} iff
V |= ( ( f 1 )(), . . . , ( f k )()).
We use that uniformly over M and V , bounded quantification in front of a 1000 formula may be rewritten in a 1000 way. It is also easy to verify that =
and hence (10.2) holds true for all .
But now the last model M of (M , : < +1) cannot be ill-founded, as
: M 1000 V.
By the elementarity of , the last model M of (M , : < + 1) cannot be
ill-founded either. This means that (10.1) is in fact a (linear) iteration of V of length
+ 1 given by U , as desired.
Lemma 10.4 (Shift Lemma) Let be a measurable cardinal, and let U be a normal
measure on . Let
(M , : OR)
be the (linear) iteration of V = M0 which is given by U . For OR, set U =
0 (U ) and = crit(U ) = 0 (). Let , and let : be order
preserving. There is then a natural elementary embedding
: M M ,
called the shift map given by such that ( ) = , and for all < , ( ) =
() and in fact
=
for all with ran( ) < .
(10.3)
10.1 Iterations of V
187
( )
=
( )
( ) =
( )
( ) =
us thus assume ran(), which implies that is a successor ordinal as well, say
= + 1, and () = . By (4.8), we have that
M = {, ( f )( ): f : M , f M }.
( ( f )( )) = ( f )( ),
where f M , f : M . This is well-defined because if is a formula and if
f 1 , . . . , f k M M , then
M |= ( ( f 1 )( ), . . . , f k ( ))
{ < : M |= ( f 1 ( ), . . . , f k ( ))} U
{ < : M |= ( ( f 1 )( ), . . . , ( f k )( ))} U , by using ,
M |= ( ( f 1 )( ), . . . , ( f k )( )).
188
10 Measurable Cardinals
( (x)) = () () (x).
Notice that each y M is of the form (x) for some < and x M .
Moreover, if (x) = (x ), where < , then
= ( ) ()( ) () (x)
= ( ) ( ) (x ),
so that the definition of (y) is independent from the choice of < and x M
with y = (x). It is easy to verify the inductive hypothesis.
We now aim to make a measurable cardinal singular in a generic extension
without collapsing cardinals. The following definition is reminiscent of the definition
of Mathias forcing, cf. p. 176.
Definition 10.5 Let M be a transitive model of ZFC, and let , U M be such
that M |= U is a normal < -complete uniform ultrafilter on . We let PU =
P = (P; ) denote the following poset, called Prikry forcing. We let p P iff
p = (a, X ) where a []< and X U, min(X ) > max(a) (if a = ). We let
(b, Y ) (a, X ) iff b a, Y X , and b\a X (in particular, b is an end-extension
of a).
If p = (a, X ) P then a is called the stem of p. Notice that any two conditions with
the same stem are compatible, so that P has the + -c.c. in M. Hence no M-cardinal
strictly above will be collapsed by .
Assume G to be P-generic over M. Let
A=
{a: X (a, X ) G}.
It is clear that A hast order-type ; in fact, an easy density argument shows that
otp(A) = and A is cofinal in . In particular, will have cofinality in M[G].
We shall now prove that no M-cardinal will be collapsed in M[G]. As is
a limit cardinal, it suffices to prove that no M-cardinal < will be collapsed in
M[G]. For this in turn it is (more than) enough to prove the following.
Lemma 10.6 Let M be a transitive model of ZFC, let , U M be such that U
witnesses that is measurable in M, let P = PU , and let G be P-generic over M.
Then
(V ) M[G] = (V ) M .
Proof Let us first verify the following.
10.1 Iterations of V
189
Claim 10.7 (Prikry-Lemma) For all p P, for all formulae , and for all names 1 ,
. . ., k M PU there is some q p with the same stem as p deciding (1 , . . . , k ).
Proof This is an application of Rowbottoms Theorem 4.59. Fix p = (a, X )
and . (We shall supress the parameters 1 , . . ., k .) Let us define F: [X ]< 3 as
follows. Let b [X ]< . We set F(b) = 0 iff there is no X such that (a b, X )
(a, X ) and (a b, X ) decides ; otherwise we set F(b) = 1 (resp., 2) iff there is
some X such that (a b, X ) (a, X ) and (a b, X ) forces that holds (resp., does
not hold).
As f M, let Y U be given by Rowbottoms Theorem 4.59, i.e., for each
n < , F is constant on [Y ]< . We claim that (a, Y ) decides .
Well, if not, then there are (b1 , Y1 ) and (b2 , Y2 ) such that (b1 , Y1 ) (a, Y ),
(b2 , Y2 ) (a, Y ) and (b1 , Y1 ) and (b2 , Y2 ) . By extending one of these
two conditions if necessary we may assume that
Card(b1 ) = Card(b2 ) = Card(a) + n
for some n < . But then
F(b1 \a) = 1 = 2 = F(b2 \a),
although
b1 \a, b2 \a [Y ]n .
Contradiction!
2
In order to prove Lemma 10.6, it now suffices to show that if < , then
It suffices to find
M[G] M. Let f 2 M[G], f = G . Let p : 2.
some q p and some g M with q = g.
Let a be the stem of p. In virtue of the Prikry Lemma we may let for each <
is the stem of q and q ( ) = (h ). If
be q p and h 2 such that a
where g 2
q = (a, X ) for < then q = (a, < X ) p and q = g,
and g( ) = h for all < .
There is a version of PU , called tree Prikry forcing, where we dont need to
assume that U is normal (in M) in order to verify the Prikry Lemma; cf. Problem
10.24, which produces a generalization of tree Prikry forcing.
Definition 10.8 Let M be a transitive model of ZFC, and let , U M be such that
U witnesses that is measurable in M. A strictly increasing sequence (n : n < )
which is cofinal in is called a Prikry sequence over M (with respect to U ) iff for
all X P() M.
X U {n : n < }\X is finite.
190
10 Measurable Cardinals
(0n ()) X.
0n () X 0m () =
We have shown that either {0n (): n 0 < n < } X or else {0n (): n 0 < n <
} X = .
We shall prove the converse to the fact that if (n : n < ) is the sequence given
by the first coordinates of elements of G then (n : n < ) is a Prikry sequence. By
virtue of Lemma 10.9, this will mean that iterations produce Prikry generics.
Definition 10.10 Let (n : n < ) be a Prikry sequence (with respect to U ). Define
G (n : n<) to be the set of all ({n : n < n 0 }, X ) where n 0 < , X U , min(X ) >
n 0 1 (if n 0 > 0), and {n : n n 0 } X .
Theorem 10.11 (A. Mathias) Let M be a transitive model of ZFC such that M |=
U is a normal measure on . Let (n : n < ) be a Prikry sequence over M with
respect to U . Then G (n : n<) is P-generic over M.
The proof will be given in Chap. 12, cf. p. 274 ff.
Theorem 10.12 Let M be a transitive model of ZFC, and suppose that there is some
M such that M |= is a measurable cardinal, as being witnessed by the normal
measure U , and 2 ++ . Let G be PU -generic over M. Then in M[G], SCH,
the Singular Cardinal Hypothesis, fails.
Proof M and M[G] have the same cardinals, and (V ) M = (V ) M[G] . As is
measurable in M, this implies that is certainly a strong limit cardinal in M[G].
Therefore, by Lemma 4.16
10.1 Iterations of V
191
M[G] |= cf() = 2 .
(10.4)
(10.5)
(10.6)
192
10 Measurable Cardinals
f mN {(m, T )} M.
193
{n} v(n) M) M,
n<
where h may be partial, such that whenever n = vi0 j and a: v(n) M, then
y M M |= j [a {(vi0 , y)} v( j)]
= M |= j [a {(vi0 , h(n, a))} v( j)].
Theorem 10.16 There is a 1 Skolem function h M which is uniformly 1M over
J -structures M, i.e., there is a 1 -formula such that whenever M = J [E] is a
J -structure, then defines h M over M in that
y = h M (n, a) M |= (n, a, y).
Proof The idea here is to let y = h M (n, a) be the first component of a minimal
witness to the 1 statement in question (rather than letting y be minimal itself). We
may let y = h M (n, a) iff
N R b, all in M, (vi0 , . . . , vik , j), some i 0 , . . . , i k , j (N = S [E] R =<E
n = vi0 . . . vik j j is 0 a, b are functions dom(a) = v(n)
0
dom(b) = v( j) a = b v(n) ran(b) N ( j, b) |=
M
= v( j) a = b v(n)
b N ((b is a function dom(b)
N b R b) ( j, b) |=0 ) y = b(vi ) ).
ran(b)
0
M
Here, N = S [E] and R =<E are uniformly 1M by Lemmas 5.25 (2)
0
M
and 5.26 (2), respectively, and |=
M is uniformly 1 by Lemma 10.14. The rest is
straightforward.
194
10 Measurable Cardinals
h M (X ) for h M (
n<
195
In what follows, x and U are predicates which are supposed to be interepreted by x and
U , respectively (or, more generally, by z and U , respectively, over (J [z]; , U )).
Lemma 10.21 Let x , and let M = (J [x]; , U ) be an x-pm with largest cardinal , and suppose that ult 0 (M ) be well-founded. Let = UM : M ult 0 (M )
be the ultrapower map. Then:
.
196
10 Measurable Cardinals
(a)
(b)
(c)
(d)
197
It is easy to verify that if (Mi , i j : i j < ) is the putative iteration of the x-pm
M of length , then every Mi (for i + 1 < ) has to be an x-pm (for successor
stages this uses Corollary 10.23; it is easy for limit stages).
Definition 10.25 Let x , let M be an x-pm, and let OR {}. A putative
iteration (Mi , i j : i j < ) is called an iteration iff either is a limit ordinal or
= or is a successor ordinal and the last model M1 is transitive.3
Definition 10.26 Let x , and let M be an x-pm. M is called (0 -)iterable iff
every putative iteration of M is an iteration.
Lemma 10.27 Let x , and let M be an x-pm. Suppose that for all < 1 , if
T is a putative iteration of M of length + 1, then T is an iteration. Then M is
iterable.
For the proof of Lemma 10.27, cf. Problem 10.1.
Definition 10.28 Let M
= (J [x]; , U ) be an x-pm. U is called -complete iff
for all {X n : n < } U , n< X n = .
The point is that not necessarily (X n : n < ) J [x].
Lemma 10.29 Let x , and let M = (J [x]; , U ) be an x-pm such that U is
-complete. Then M is iterable.
Proof This proof is similar to the proof of Theorem 10.3.) Let T be a putative iteration
of M of length + 1. Let : V V , where is large enough, V is countable and
transitive, and M , T ran( ). Set M, T = 1 (M , T ). Then T is a putative
iteration of M of length + 1 < 1 . Assuming that T is not an iteration, T is not
an iteration either.
We shall now recursively construct maps
Let T = (Mi , i j : i j ).
{ < :
Mi |= ( f 0 ( ), . . . , f k1 ( ))} U
{ < : M |= (i ( f 0 )( ), . . . , f k1 ( ))} i U
M |= (i ( f 0 )( ), . . . , i ( f k1 )( ))
whenever is 0 .
This finishes the construction. Now notice that : M M witnesses that M
is well-founded. This gives a contradiction.
3
By our convention, cf. Definition 10.20, this is tantamount to saying that M1 is well-founded.
198
10 Measurable Cardinals
: M M ,
called the shift map given by such that ( ) = , and for all < , ( ) =
() and in fact
(10.7)
M = (J [x]; , U )
= h M () 1 M .
Then M
/ J [x].
Proof To begin with, with the help of Lemma 5.28 it is straightforward to verify that
M is again an x-premouse.
Claim 10.34 M is an x-mouse.
199
(10.8)
(10.9)
(10.10)
200
10 Measurable Cardinals
(J [x]; , U )
= h M (0 ) 1 M .
(10.11)
It is easy to see that must be the identity, so that 0 is also the largest cardinal of
J [x] and 0 < . As in the proof of Lemma 10.33, we get that = (0 )+L[x] .
But certainly (0 )+L[x] , as 0 is the largest cardinal of J [x]. Therefore =
= (0 )+L[x] , so that (10.11) in fact gives that
M = h M (0 ).
The same argument shows that in fact for every i OR,
Mi = h Mi (i ), i = (i )+L[x] , Ni = h Ni (i ), and i = (i )+L[x] .(10.12)
Moreover, the maps i j and i j are 1 -elementary by Lemma 10.21 (c), so that they
preserve the 1 -Skolem function. Therefore, if i j OR, then
i j
i j
Mi
= h M j (i ) 1 M j and Ni
= h N j (i ) 1 N j .
(10.13)
(10.14)
(10.15)
201
ran(ii+1 ) = h Mi+1 (i ),
which together with (10.15) yields that for all j > i,
i+1 j
Mi+1 = h Mi+1 (i {i })
= h M j (i {i })
and therefore
i+1 h M j (i {i }).
(10.16)
(10.17)
(10.18)
(10.19)
0i
N = N0
= h Ni (0 ) = h Mi (i )
= Mi ,
so that N = Mi . Symmetrically, if 0 0 , then there is some i with M = Ni .
Corollary 10.36 Let x , and assume that there is an x-mouse. There is then
exactly one x-mouse M such that
M = h M (),
(10.20)
202
10 Measurable Cardinals
M
= h M (),
where M is transitive. As is 1 -elementary and thus respects the 1 -Skolem
function, we must have that (10.20) holds true.
Let N be any x-mouse, and let be the largest cardinal of N . Suppose that
> 0 would be such that if (Ni , i j : i j ) is the iteration of N of length
/ h N ()
+ 1, then M = N . Then as in (10.14) in the proof of Lemma 10.35,
and thus
/ h M () = M , which is nonsense. By Lemma 10.35 we must therefore
have some such that if (Mi , i j : i j ) is the iteration of M0 = M of length
+ 1, then M = N . Moreover, let be the largest cardinal of M . If > 0,
/ h N ().
then as in (10.14) in the proof of Lemma 10.35,
/ h M (), and thus
This means that if we assume in addition that N = h N (), then = 0, so that
N = M0 = M .
Definition 10.37 Let x . By x # (x-sharp) we denote the unique x-mouse M
with M = h M (), if it exists. We also write 0# (zero-sharp) for # .
In the light of Lemma 10.35, it is easy to verify that x # is that x-mouse whose
largest cardinal is smallest possible among all x-mice.
Via Gdelization and as x # = h x # (), the 1 -theory of x # , call it Th1 (x # ),
may be construed as a set of natural numbers. By Lemma 10.27 and with the help of
Lemma 7.17, it is not hard to verify that {Th1 (x # )} is 21 (x). In this sense we may
construe x # itself as a 21 (x)-singleton, cf. Problem 10.7.
Definition 10.38 Let W be an inner model. W is called rigid iff there is no non-trivial
elementary embedding : W W .
By Theorem 4.53, V is rigid.
Theorem 10.39 (K. Kunen) Let x . Suppose that L[x] is not rigid. Then x #
exists.
Proof We shall make use of Problem 10.10. Let us fix an elementary embedding
: L[x] L[x], = id. Let be the critical point of , and let U = {X
P() L[x] : (X )} be the L[x]-ultrafilter derived from as in Lemma 10.19.
Setting = +L[x] , we know that (J [x]; , U ) is an x-pm (cf. Lemma 10.19). In
order to prove that x # exists, it suffices to verify that (J [x]; , U ) is iterable.
In much the same way as on p. 55, we may factor : L[x] L[x] as = k U ,
where U : L[x] ult(L[x]; U ) is the ultrapower map (cf. Problem 10.10) and
k([ f ]U ) = ( f )() for every f L[x] L[x]. We thus have ult(L[x]; U )
= L[x],
and we may and shall as well assume that k = id and U = .
203
L[x]
= h L[x] ( ) L[x]
(10.21)
and
/ h L[x] ( ). For the purpose of this proof, let us call an x-pm M =
(J [x]; , U ) certified iff the following holds true. If denotes the critical point of
/ h L[x] ( ) and U = U ( ).
U , and if is then as in (10.21), then
If : L[x] L[x] is an elementary embedding with critical point , then
UL[x]
( ) : L[x] U ( ) L[x]
is the associated ultrapower map (cf. also Problem 10.10); we also write U ( ) rather
than UL[x]
( ) , and we write k( ): L[x] L[x] for the canonical factor map obtained
as on p. 55 such that
= k( ) U ( ) ,
(10.22)
(10.23)
204
10 Measurable Cardinals
(i) M is certified.
(ii) For all i < and a [ ]< , h L[x] (i, a) U h L[x] (i, a).
(iii) = +L[x] and ult 0 (L[x]; U ) is well-founded (equivalently, transitive and
thus equal to L[x]).
Proof Let be as in (10.21).
(i) = (ii) Using Claim 10.40, let i < and a [ ]< . Let X = h L[x] (i, a)
. We have that X U = U ( ) iff (X ) ( + 1), which by (10.23) is
equivalent to U (X ) ( + 1). But U (X ) = h L[x] (i, a) U (), as the
elements of a are fixed points under U (cf. Problem 10.10 (d)), so that X U is
equivalent to h L[x] (i, a), as desired.
(ii) = (i) It is easy to see that (ii) implies that = crit( ): If = h L[x] (i, a),
where i < and a [ ]< , then = h L[x] (i, a) U would give
h L[x] (i, a) = , a contradiction!
Now let X P()L[x]. By Claim 10.40 there is some i < and some a [
]< such that X = h L[x] (i, a). We get that X U iff h L[x] (i, a)(+1),
which is equal to U ( ) (h L[x] (i, a))(+1), as the elements of a are fixed points
under U ( ) ; but by (10.23), U ( ) (h L[x] (i, a) ) ( + 1) = U ( ) (X ) ( + 1)
is equal to (X ) ( + 1), so that X U is equivalent to X U ( ), as desired.
In particular, U ( ) J [x] (as M is a J -structure), which implies that
+L[x] and thus = +L[x] .
(i) = (iii) If M is certified, then U = U ( ), which immediately gives that
ult 0 (L[x]; U ) must be well-founded.
(iii) = (ii) Let U : L[x] U L[x] be the ultrapower map, and let i <
and a [ ]< . As the elements of a are fixed points under U , we get
that h L[x] (i, a) U iff U (h L[x] (i, a) ) = h L[x] (i, a) U () iff
h L[x] (i, a), as desired.
Let us now return to the x-pm (J [x]; , U ) isolated from = U above. Let
be a countable ordinal, and let us suppose (Mi , i j : i j ) to be the putative
iteration of M0 = (J [x]; , U ) of length + 1. In the light of Lemma 10.27, it
suffices to show that M is transitive. We show:
Claim 10.42 Every Mi , i , is transitive and in fact certified.
Proof Of course, if i < , then Mi is trivially transitive by the definition of
putative iteration.
That M0 = (J [x]; , U ) be certified follows immediately from (iii) = (i) of
Claim 10.41, applied to = U .
Let us now suppose that i < and that Mi is certified. We aim to verify that
Mi+1 is transitive and certified.
Let us write Mi = (Ji [x]; , Ui ), and let i be the critical point of Ui . By
(i) = (iii) of Claim 10.41, ult 0 (L[x]; Ui ) is equal to L[x], so that the ultrapower
= Ui (cf. Problem 10.10) maps L[x] to L[x]. It is easy to verify that
map UL[x]
i
the universe of Mi+1 = ult 0 (Mi ) is isomorphic to Ui (Ji [x]). This is because
if [ f ]Ui Ui (Ji [x]), where f i L[x] L[x], then [ f ]Ui = [ f ]Ui for some
205
(10.24)
as desired.
Now let be a limit ordinal, and suppose that every Mi , i < , is certified.
We first need to see that M is well-founded.
For i < , let us write Mi = (Ji [x]; , Ui ), and let i = crit(Ui ). Because every
Mi , i < , is certified, by (i) = (iii) of Claim 10.41 we know that the iteration
map i j : Mi M j extends to an elementary embedding from L[x] to L[x], which
we shall denote by i j , for all i j < . (cf. Problem 10.10.)
Let us also write = supi< i . For i , let let
i
L[x]
= h L[x] (i ) L[x].
(10.25)
( : i j < )).
Moreover, if i j < , then = i j .
of ((L[x]: i < ),
ij
ij
206
10 Measurable Cardinals
As for the second part, it therefore suffices to prove this for j = i + 1 < .
hL[x] ( i { i } )
hL[x] ( i+1 )
L[x]
i,i+1
i+1
L[x]
hL[x] ( )
i+1,
L[x]
L[x]
hL[x] ( i )
i,i+1
(10.26)
To show this, let < i+1 . There is some f : i i , f L[x], such that =
[ f ]Ui = ii+1 ( f )(i ). As Mi is certified, (the proof of) Claim 10.40 yields some j <
and some a [i ]< such that f = h L[x] ( j, a) i . As the elements of a are
not moved by ii+1 , ii+1 ( f ) = h L[x] ( j, a) i+1 . But then = ii+1 ( f )(i ) =
h L[x] ( j, a)(i ) h L[x] (i + 1 ). This shows (10.26).
Let us now define
( f )(i ),
ii+1 ( f )(i ) ii+1
(10.27)
( f )(i )Eii+1
(g)(i ) i ii+1
({ < i : f ( )Eg( )})
ii+1
1
i i+1
i ({ < i : f ( )Eg( )})
207
(10.28)
for i, k < .
/ h L[x] (i { j : j > i}) for every i OR.
(4) i
(5) Let : L[x] L[x] be an elementary embedding. Then there is some orderpreserving e: OR OR such that = e , where e is defined as in (10.28).
Proof (1) Suppose that L[x] = h L[x] ({i : i OR}). Let be the least ordinal such
that
/ h L[x] ({i : i OR}), so that is the critical point of
: L[x]
= h L[x] ({i : i OR}) L[x].
If M = (J [x]; , U ) is derived from as in the proof of Lemma 10.19, then the
proof of Theorem 10.39 shows that M is an x-mouse. By Lemma 10.35 and the
definition of x # , cf. Definition 10.37, we must have = i for some i OR. But
then trivially h L[x] ({i : i OR}). Contradiction!
(2) For i j, i j i = id and i j (i ) = j , which gives that Ji [x]
i j (Ji [x]) = J j [x]. This immediately yields
Ji [x] L[x]
(10.29)
208
10 Measurable Cardinals
/ h L[x] ({ j : j = i}).
which is nonsense. Therefore, i
But then if is the least ordinal with
/ h L[x] ({ j : j = i}), then = j for
some j OR as in the proof of (1), and hence j = i, i.e., = i . We have shown
that
/ h L[x] ({ j : j = i}) = h L[x] (i { j : j > i}).
i
(5) By (1), it suffices to prove that for every i OR there is some j OR with
(i ) = j . Let us fix i OR, and let us write = (i ). Let
L[x]
= h L[x] (( + 1) ran( )) = h L[x] (( + 1) {( j ): j OR}) L[x].
Let be a limit ordinal with > . Let i < , [i + 1]< , and i < 1 <
. . . < n < n+1 < . . . < k . If h L[x] (i , (, 1 , . . . , k )) < , then by (2)
we must have that
h L[x] (i , (, 1 , . . . , k )) < n +1 ,
and therefore
h L[x] ((i + 1) {1 , . . . , k }) n +1 ,
so that by the elementarity of ,
209
(10.30)
(10.31)
(10.32)
for some i , k < and []< . By (2), we may assume here that every p ,
p = 1, . . . , k, is a cardinal (in V ) above with p p , so that ( p ) = p .
As (i ) = , (10.32) and the elementarity of then yield that
< i i = h L[x] (i , ( , 1 , . . . , k )).
This contradicts (4)!
In the light of Corollary 10.44, the club class {i : i OR} is called the class
of Silver indicernibles for L[x]. Theorem 10.39 and Corollary 10.44 give that the
existence of x # is equivalent to the non-rigidity of L[x]. Jensens Covering Lemma,
cf. Theorem 11.56, will produce a much deeper equivalence.
10.3 Extenders
We need to introduce extenders which generalize measures on measurable cardinals, cf. Definition 4.54, and which allow ultrapower constructions which generalize
the one from the proof of Theorem 4.55, (1) = (3), cf. Theorem 10.48. They will
be used in the proof of Jensens Covering Lemma, cf. Theorem 11.56, as well as in
the proof of projective determinacy, cf. Theorem 13.7.
For the sake of this section, by a transitive model M of a sufficiently large
fragment of ZFC we mean a transitive model M of the statements listed in Corollary
5.18, i.e., (Ext), (Fund), (Inf), (Pair), (Union), the statement that every set is an
element of a transitive set, 0 -comprehension, and xy x y exists, together
with (AC) in the form that every set can be well-ordered. We allow M to be a model of
the form (M; , A1 , . . . , An ), where Ai M < , 1 i n, are predicates, in which
210
10 Measurable Cardinals
} E a{} .
10.3 Extenders
211
/ Ea .
It is easy to see that X E a and Y X , Y P([a ]Card(a) ) M imply Y E a ,
and that
/ E a . If X P([a ]Card(a) ) M, then (X )(([a ]Card(a) M)\ X ) =
Card(a)
) M), so that X E a or ([a ]Card(a) M) \ X E a .
([a ]
Let
(X i : i < ) M, where < and X i E a for each i < . Then
a i< (X i ) = ( i< X i ), i.e., i< X i E a .
(2) Let a b []< and X P([a ]Card(a) ) M. Then X E a iff a (X )
iff b (X ab ) iff X ab E b .
(3) Whereas {} (), {} is not a subset of () = whenever < . Thus
{} = .
(4) Let a []< and f : [a ]Card(a) a with f M. Suppose that
a ({u [a ]Card(a) : f (u) < max(u)}),
which means that
( f )(a) < max(a).
Set = ( f )(a). Then
a{}
}),
212
10 Measurable Cardinals
Let us first argue that N and are unique up to isomorphism. Suppose that N ,
and N , are both as in the statement of the Theorem. We claim that
( f )(a) ( f )(a),
(10.33)
A([a,
f ]) {u [a ]Card(a) : f (u) A} E a
Notice that the relevant sets are members of M, as M is a model of a sufficiently
are
large fragment of ZFC. Moreover, by (1) and (2) of Definition 10.45, and A,
10.3 Extenders
213
A).
N = ( D;
Claim 10.49 ( os Theorem) Let (v1 , ..., vk ) be a 0 formula (in the language of
M), and let (a1 , f 1 ), ..., (ak , f k ) D. Then
N |= ([a1 , f 1 ], ..., [ak , f k ])
{u [c ]
Card(c)
; M |=
E c for c = a1 . . . ak .
Proof Notice again that the relevant sets are members of M. Claim 10.49 is shown
by induction on the complexity of , by exploiting (1) and (2) of Definition 10.45.
Let us illustrate this by verifying the direction from right to left in the case that, say,
v0 v1 for some 0 formula .
We assume that, setting c = a1 . . . ak ,
{u [c ]Card(c) : M |= v0 f 1a1 ,c (u) (v0 , f 1a1 ,c (u), . . . , f kak ,c (u))} E c .
Let us define f 0 : [c ]Card(c) ran( f 1 ) {} as follows, where <ran( f1 ) M is a
well-ordering of ran( f 1 ).
as desired.
Given Claim 10.49, we may and shall from now on identify, via the Mostowski
collapse, the well-founded part wfp(N ) of N with a transitive structure. In particular,
if [a, f ] wfp(N ) then we identify the equivalence class [a, f ] with its image under
the Mostowski collapse.
Let us now define : M N by
(x) = [0, cx ], where cx : {} = [0 ]0 {x}.
214
10 Measurable Cardinals
We aim to verify that N , satisfy (a), (b), (c), and (d) from the statement of Theorem
10.49.
Claim 10.50 If < and [a, f ] [{}, pr] then [a, f ] = [{}, pr] for some < .
Proof Let [a, f ] [{}, pr]. Set b = a {}. By the os Theorem 10.49,
{u [b ]Card(b) : f a,b (u) pr {},b (u)} E b .
By (4) of Definition 10.45, there is some < such that, setting c = b {},
{u [c ]Card(c) : f a,c (u) = pr {},c (u)} E c ,
and hence, by the os Theorem again, [a, f ] = [{}, pr].
(10.34)
10.3 Extenders
215
Claim 10.52 readily implies (c) from the statement of Theorem 10.48.
Claim 10.53 = crit( ).
Proof Let us first show that = id. We prove that ( ) = for all < by
induction on .
Let < . Suppose that [a, f ] ( ) = [0, c ]. Set b = a { }. Then
{u [b ]Card(b) : f a,b (u) < } E b .
As E b is < -closed with respect to sequences in M (cf. (1) of Definition 10.45),
there is hence some < such that
{u [b ]Card(b) : f a,b (u) = } E b ,
and therefore [a, f ] = ( ) which is by the inductive hypothesis. Hence ( ) .
It is clear that ( ).
It remains to prove (d) from the statement of Theorem 10.48. Let X E a . By (1)
of Definition 10.45,
X = {u [a ]Card(a) : u X } E a ,
which, by the os Theorem and Claim 3, gives that a = [a, id] [0, c X ] = (X ).
On the other hand, suppose that X P([a ]Card(a) ) M and a (X ). Then
by Claim 3, [a, id] = a (X ) = [0, c X ], and thus by the os Theorem
X = {u [a ]Card(a) : u X } E a .
We have shown Theorem 10.48.
216
10 Measurable Cardinals
10.3 Extenders
217
Proof We may pick some E such that for every inaccessible cardinal ,
( ; E ( ))
= (V ; ).
Identifying N with ult 0 (V ; F), we let
i F : V F N
be the ultrapower map, and we let
k: N M
be the factor map which is defined as in the proof of Lemma 10.56 by k(i F ( f )(a)) =
( f )(a) for a []< and f : []Card(a) V . We have that k i F = and
k = id.
By the elementarity of ,
( ; (E) ( ))
= ((V ) M ; )
for every () which is inaccessible in M. In particular,
(; (E) ( ))
= ((V ) M ; ).
(10.35)
(10.36)
N N.
Proof Let {xi : i < } N , say xi = E ( f i )(ai ) for i < . We aim to prove that
(xi : i < ) N . Let b and g be as in Definition 10.59. We may assume that b.
218
10 Measurable Cardinals
10.3 Extenders
219
(10.37)
(10.38)
Now let (xi : i < ) N . We aim to show that (xi : i < ) N . Let xi = [ai , f i ]
for i < , where ai []< and f i : [ai ]Card(ai ) V . Let us write G for the
function with domain and G(i) = ai for i < , i.e., G = (ai : i < ). By (10.37),
(ai : i < ) N , so that we may pick some b []< and g: [b ]Card(b) V with
G = (ai : i < ) = E (g)(b).
(10.39)
(10.40)
Here, we understand that if f a,c (u) is an injective function with i in its domain, then
( f a,c (u))1 (i) is the preimage of i under that function, and ( f a,c (u))1 (i) =
otherwise.
We get that E (H )(c): E () N , and for each i < ,
E (H )(c)( E (i)) = E ( f i )( E (g b,c )(c)(( E ( f a,c )(c))1 )( E (i)))
= E ( f i )( E (g)(b)(( E ( f )(a))1 )( E (i)))
= E ( f i )( E (g)(b)( E )1 )( E (i)))
= E ( f i )(G(i))
= E ( f i )(ai ) = xi .
Using E N once more, we then get that the function with domain which
maps
i E (H )(c)(( E )(i)) = xi
also exists inside N . We have shown that (xi : i < ) N .
The following concepts and techniques will be refined in the next section.
Definition 10.63 Let M be a transitive model of a sufficiently large fragment of
ZFC, and let E = (E a : a []< ) be a (, )-extender over M. Let < Card()
be an infinite cardinal (in V ). Then E is called -complete provided the following
220
10 Measurable Cardinals
1 ,c
(u), ..., f ka
k ,c
(u))} E c .
Let : i< ai (E) be order-preserving such that "ai X i for every i < .
Let us define : U M by setting ([a, f ]) = f ( "(a)) for (a, f ) U .
We get that is well-defined and 0 -elementary by the following reasoning. Let
(v1 , ..., vk ) be 0 , and let [a j , f j ] U , 1 j k. Set c = a 1 ... a k . We then
get that
U |=([a 1 , f 1 ], ..., [a k , f k ])
{u [c ]
1
k
: M |=( f 1a ,c (u), ..., f ka ,c (u))} E c
1
k
[c ]Card(c) : M |= ( f 1a ,c (u), ..., f ka ,c (u))}
M |=( f 1 ( a 1 ), ..., f k ( a k )).
Card(c)
c {u
10.3 Extenders
221
Proof Let ((ai , X i ): i < ) be such that X i E ai , and hence ai (X i ), for all
i < . As H H , (X i : i < ) H . Let : i< ai
= = otp( i< ai ) be the
transitive collapse; notice that < + < . For each i < let a i = ai . We have
that (a i : i < ) H . But now
H |= order-preserving : OR i < a i ((X j ; j < ))(i),
as witnessed by 1 . Therefore,
H |= order-preserving : OR i < a i X i .
Hence, if H is a witness to (10.41), then :
ai X i for every i < .
i< ai
(10.41)
OR is such that
0 )
M M V M (M, , x)),
(10.42)
i.e., X Fa or ([]Card(a) ) \ X Fa .
222
10 Measurable Cardinals
then M also witnesses V |= (, x), as (2 0 ) P P.
Definition 10.70 Let F be a short (, )-extender over V . Then F is called certified
iff is also the strength5 of F, is inaccessible, and for every U 1+ V of size 20
there is some : U 1+ V witnessing that F is 20 -complete with respect to U .
Lemma 10.71 Let F be a short (, )-extender over V , and suppose that is also
the strength of F and is inaccessible. Then F is certified.
Proof By Lemma 10.69, is a reflection point. Let us fix U 1+ V of size 20 . We
need to find some : U 1+ V such that ( U ) = id and for all a [ U ]<
and for all X P([]Card(a) ) U , X Fa iff a X .
Let : V F M = ult(V, F), where M is transitive. Notice that V M (i.e.,
M
V M), and U 1+ V = VM 1+ V()
1+ M by Lemma 10.69 applied
Recall that the strength of an extender F is the largest ordinal such that V Ult(V ; F).
223
224
10 Measurable Cardinals
(10.43)
for all n < m < , for some monotone i: (which depends on b). Let
: V 1002 V
be such that V is transitive, Card(V ) = 20 , and {E i : i < } {nb : n < , b
an infinite branch through T } ran( ). We let U = (U, U ) be the tree of
attempts to find an infinite branch b through T together with a proof of the
well-foundedness of the direct limit along b.
More precisely, U is defined as follows. Let us set E i = 1 (E i ) for i < .
Obviously, there is a unique
T = (( M i , i j : i T j < ), ( E i : i < ), T )
such that
(10.44)
(10.45)
If there is any. The current proof does not presuppose that there be some such branch. Rather, it
will show the existence of some such b such that the direct limit along b is well-founded.
225
for some (all) large enough n < . But as {nb : n < } ran( ), and hence if
(Mb , (ib : i b)) is the direct limit of (Mi , i j : i T j < b), then (10.43) yields
that
i(n)b ( 1 (nb )) = i(m)b ( i(n)i(m) ( 1 (nb )))
= i(m)b (i(n)i(m) (nb ))
b
)
> i(m)b (m
b
))
= i(m)b ( 1 (m
for all n < m < , so that ( i(n)b ( 1 (n )): n < ) witnesses that dir lim( M i , i j :
i T j b) must be ill-founded. This contradicts the existence of , cf. (10.45).
We therefore must have that U = (U ; U ) is well-founded. In order to finish the
proof of the theorem, it now suffices to derive a contradiction.
In order to work towards a contradiction, we need generalized versions of U
as well as realizations and enlargements. Let R be a transitive model of ZC
plus replacement for 1000 -formulae. We then call the triple (, Q, R) a realization
of M i , where i < , iff : M i 1000 Q, Q is a (not necessarily proper) rank initial
2 0 R R. If (, Q, R) is a realization of M
i , then we may define a tree
U ( 0i , Q) = (U ( 0i , Q), U ( 0i ,Q) )
in the same fashion as U was defined above: we set (, j) U ( 0i , Q) iff
j < and : M j 1000 Q is such that 0 j = 0i , and if (, j), ( , k)
U ( 0i , Q) then we write ( , k) U ( 0i ,Q) (, j) iff j T k and jk = .
Hence U = U (, V ).
Vi j ,
226
10 Measurable Cardinals
i
V (i ) , VM
= , and if U = U ( 0i+1 , Q), then U is well-founded
i
and there are (in order type) at least ||( , i + 1)||U many < R O R such that
( , Q, VR ) is a realization of M i+1 . We aim to verify that
i ).
Ri |= (i VM
i
(10.46)
Ri
i
of size 20 such that ran(i VM
1+ V witnessing
i ) U and some : U
0
that i ( E i ) is 20 -complete with respect to U . Notice that 2 Q i Q i Ri , and
i
hence ran(i VM
i ) Ri .
In order to verify (10.46) it then remains to verify that
i ).
VRi |= ( i VM
i
(10.47)
Card(a)
where a is a finite subset of the length of E i and writing = crit( E i ), f : []
227
()
Q j |=
i
= j Vi j , and () holds true as witnesses
Here, () holds true as i VM
i
that i ( E i ) is 20 -complete with respect to U . (10.47) immediately yields that
0i+1 = ji+1 0 j = j 0 j ,
and hence ( , i + 1) U ( j 0 j , Q j ), moreover, clearly,
= ||( , i + 1)||U ( j 0 j ,Q j ) < ||( j , j)||U ( j 0 j ,Q j ) ,
R
R
of M j . As 2 0 R j R j , the triple ( , Q j , V j ) R j witnesses that
i ).
R j |= ( i VM
i
VRi = V
|= ( i ),
We end this section by defining the concept of Woodin cardinals. The following
definition strengthens Definition 4.60.
Definition 10.75 Let be a cardinal, let > be a limit ordinal, and let A V .
We say that is strong up to with respect to A iff for all < there is some
elementary embedding : V M such that M is transitive, crit( ) = , V M,
and (A) V = A V .
Definition 10.76 A cardinal is called a Woodin cardinal iff for every A V
there is some < such that is strong up to with respect to A.
Lemma 10.77 Let be a Woodin cardinal. Then is a Mahlo cardinal. In fact,
for every A V there is a stationary subset S of such that for all S, for all
228
10 Measurable Cardinals
< there is a certified extender F with critical point and lh(F) such that
if F : V Ult(V ; F) is the ultrapower map, then F (A) V = A V .7
Proof Fix A V . Let C be club in . Let f : C be the monotone
enumeration of C, and let g: A be a surjection such that
g = A V for every inaccessible cardinal .
(10.48)
Define h: V by setting
h( ) =
f ( + n)
g( + n)
(10.49)
(10.50)
229
10.5 Problems
10.1. Prove Lemma 10.27, using the method from the proof of Lemma 10.29.
10.2. Let be a measurable cardinal, let U be a measure on , and let (Mi , i, j : i
j < ) be the iteration of V = M0 of length OR given by U . Let i = 0i ()
for i < .
(a) Show by induction on i OR that for all x Mi there are k < ,
i 1 , . . . , i k < i, and a function f : [0 ]k M0 , f M0 , such that x =
0,i ( f )(i1 , . . . , ik ).
(b) Show that {i : i OR} is club in OR.
(c) Let i 0 > 0, and let X P(i0 ) Mi0 . Show that X 0,i0 (U ) iff there
is some k < i 0 such that {i : k i < i 0 } X . (Cf. Lemma 10.9.)
(d) Conclude that if > 2 is a regular cardinal, then 0, (U ) = F M
(where F is the club filter on , cf. Lemma 4.25).
230
10 Measurable Cardinals
10.3. Let be a measurable cardinal, let U be a measure on . (a) Show that, setting
U = U L[U ], L[U ] = L[U ] and L[U ] |= U is a measure on .
Let (Mi , i, j : i j < ) be the iteration of L[U ] of length OR given by
U . (b) Show that if > 2 is a regular cardinal, then M = L[F ], where
F is the club filter on . [Hint. Problem 10.2 (d).]
10.4. Let us call a J -structure J [U ] an L -premouse iff there is some < such
that U J [U ] and J [U ] |= ZFC + U is a measure on . If M = J [U ]
is an L -premouse, then we may define (putative) iterations of M in much
the same way as putative iterations of V , cf. Definition 10.1. In the spirit of
Definition 10.26, an L -premouse is called an L -mouse iff every putative
iteration of M is an iteration.
(a) Let M = J [U ] be an L -mouse and let (Mi , i, j : i j < ) be the
iteration of M of length OR. Show that there is some and some such
that M = J [F ]. [Hint. Problem 10.3 (b).] Conclude that any two L -mice
M, N may be coiterated, i.e., there are iterates M of M and N of N ,
respectively, such that M = J [F] and N = J [F] for some , , F.
Show also that if M |= U is a measure on , then for all i, M = M0 and
Mi have the same subsets of .
(b) Let : J [U ] J [U ] be an elementary embedding, where J [U ] is an
L -premouse and J [U ] is an L -mouse. Show that J [U ] is an L -mouse
also. [Hint. Cf. the proof of Lemma 10.33.]
10.5. Show in ZF + there is a measurable cardinal that there is an inner model
M such that M |= GCH + there is a measurable cardinal. [Hint. Let U
witness that is measurable, and set M = L[U ]. To show that M satisfies
GCH, verify that in M, for each infinite cardinal and every X there
are at most many Y with Y < M X (with < M being as on p. 77),
as follows. Fix X . Pick : J [U ] J [U ] such that Card(J [U ]) = ,
( + 1) = id, X ran( ) and J [U ] is an L -premouse. We claim that
if Y with Y <W X , then Y J [U ]. For this, use Theorem 10.3 and
Problem 10.4.]
10.6. Prove oss Theorem 10.22. Show also Lemma 10.21 (d).
10.7. Show that {Th1 (x # )} is 12 (x), where Th1 (x # ) is the set of all Gdel
numbers of 1 -sentences which hold true in x # .
10.8. Let A be 21 (x), A = . Show that A x # = . [Hint. Corollary 7.21.]
10.9. Let be any ordinal. A cardinal is called -Erds iff for every F: []<
2 there is some X with otp(X ) = such that for every n < , F [X ]n
is constant.
Show that if is 1 -Erds, then x # exists for every x .
Show also that if < 1L and if is -Erds, then L |= is -Erds.
Let x , and let M = (J [x]; , U ) be an x-pm. Let be the critical point
of U , and let us assume that = +L[x] . We define the (0 -)ultrapower,
written ult0 (L[x]; U ) or just ult(L[x]; U ), of L[x] as follows. For f, g
10.5 Problems
231
J [x]
iff
L[x], set f g iff { < : f ( ) = g( )} U , and write f g
{ < : f ( ) g( )} U . We let [ f ] denote the -equivalence class of f .
(10.51)
232
10 Measurable Cardinals
{X [V ]0 : X V , X , and V
= X}
is stationary in [V ]0 .
Show that if 0# exists, then every Silver indiscernible is remarkable in L.
Show also that if is remarkable in V , then is remarkable in L. [Hint.
Problem 7.4.]
10.15. Show that if is -Erds, then there are < < such that V |= ZFC
+ is remarkable. Show also that every remarkable cardinal is ineffable.
Col(,)
10.16. (Martin-Solovay) We say that V is closed under sharps iff for all , V
x # exists for all x . Let be any ordinal, and let G be Col(, )-generic
over V . Let z V , and let A V [G] be such that V [G] |= A
is 31 (z), A = . Show that A V = . (Compare Corollary 7.21.) [Hint.
For any X V , we may make sense of X # . Let A = {x: y (x, y, z)},
where is 12 . Let T be a tree of attempts to find x, y, , H , and g such that
: H # (H )# is an elementary embedding, H is countable, z H , g is
Q-generic over H for some Q H , and H # [g] |= (x, y, z).]
10.17. Let E = (E a : a []< ) be a (, )-extender over V . Show that ult(V ; E)
is well-founded iff E is -complete.
10.18. Let E be a short (, )-extender over V .
(1) If is a limit ordinal with cf() = , then E is continuous at . (Compare
Lemma 4.52 (c).)
(2) If > is a cardinal such that cf() = and < for every < ,
then is a fixed point of E , i.e., E () = . (Compare Problem 4.28.)
10.19. Let E = (E a : a []< ) be a (, ) extender over V . Let P V be a poset,
and let G be P-generic over V . Set
E a = {Y []Card(a) : X E a Y X },
as defined in V [G]. Show that (E a : a []< ) is a (, ) extender over V [G].
Conclude that is a strong cardinaland is supercompact are preserved
by small forcing in the sense of Problem 6.18.
10.5 Problems
233
10.20. Show that is a Woodin cardinal is preserved by small forcing in the sense
of Problem 6.18.
10.21. (Magidor) Show that the conclusion of problem 4.29 yields that is supercompact, i.e., if for every > there are < < together with an elementary embedding : V V such that crit( ) = and () = , then is
supercompact. [Hint: Let be least such that there is no (, )-extender
over V witnessing that is -supercompact. Pick < < <
together with some elementary embedding : V V such that crit( ) =
and () = . Then ran( ) and one can derive from a (, )-extender
F V over V witnessing that is 1 ()-supercompact in V . Lift this
statement up via .]
10.22. Show that the conclusion of problem 4.30 yields that is supercompact.
[Hint. Design an ultrapower construction la Theorem 10.48.]
10.23. Show that if is subcompact, then is a Woodin cardinal.
10.24. (Supercompact tree Prikry forcing) Let be supercompact, and let > .
Let M be an inner model with M M, and let : V M be an elementary
embedding with critical point such that () > . Let U be derived from
as in Problem 4.30. Let P be the set of all trees T on P () (in the sense
of the definition given on p. 123) such that there is some (stem) s T such
that t s s t for all t T and for all t s, t T ,
{x P (): t x T } U.
P, ordered by U P T iff U T , is called supercompact tree Prikry forcing. Let G be P-generic over V . Show that cf V [G] () = for every [, ]
with cf V () . Show also that the Prikry-Lemma 10.7 holds true for the
supercompact tree Prikry forcing P and conclude that V and V [G] have the
same V .
10.25. Let E, E be certified extenders on . We define E < M E iff E ult(V ; E ).
Show that < M is well-founded. (Compare Problem 4.27.) [Hint. Use Theorem 10.74.] Again, <M is called the Mitchell order (this time on certified
extenders).
Chapter 11
235
236
Lemma 11.3 Let M = J [E] be an acceptable J-structure. If is an infinite cardinal of M (or = ), then
(H ) M = J [E].
Proof It suffices to prove that if M and = +M , then (H ) M = J [E].
That J [E] (H ) M follows from Lemma 5.16. Let us prove (H ) M J [E].
Suppose not, and let x be -minimal in (H ) M \ J [E]. Then x J [E], and
there is some surjection g: x, g M. For < , let < be least such that
g( ) J [E]. By (the proof of) Lemma 4.15, = sup({ : < }) < .
By Lemma 11.2, there is some < , such that
(P() J + [E]) \ J [E] = ,
which by acceptability (and Lemma 10.17) yields some surjective f : J [E],
/ M. Let A = f 1 A. Then A
/ M, since
A P() 1M be such that A
Let us now show that is in fact a 1 -cardinal in M. Suppose not, and let
f : be a possibly partial function from onto , f 1M . We know that
J [E]
there is a 1
A = { :
/ g( )}.
/ J [E] by the proof of Theorem 1.3. We
Then A is clearly in P( ) 1M , and A
get that A
/ M by Lemma 11.2. But < , so that we must have that A M by
the definition of . Contradiction!
237
M p = (J [E], A M )
is called the reduct determined by p.
p
We shall often write A M (n, x) instead of (n, x) A M , and we shall write A p rather
p
than A M if there is no danger of confusion.
Definition 11.8 Let M be an acceptable structure, and write = 1 (M).
PM = the set of all p [, OR M)< for which
there is aB 1M ({ p}) such that B
/ M.
The elements of PM are called good parameters.
Lemma 11.9 Let M be an acceptable J-structure, p [1 (M), OR M)< , and
p
A = A M . Then
/ M.
p PM A ( 1 (M))
238
(11.1)
p
AM .
(11.2)
239
(11.3)
f S +n [E] f : f surjective).
Clearly, if M |= , then M is acceptable. To show the converse, let M be acceptable,
and let < M OR. Let 0 be the least such that
(P( ) J + [E]) \ J [E] = .
(11.4)
240
for
some n and z [1 ( M)]
Let x M.
M
241
. . . , h M (n , (z , p))),
M |= n (h M (n 1 , (z1 , p)),
(11.5)
(11.6)
(11.7)
(11.8)
(11.9)
(11.10)
(11.11)
242
Mp
p !
M =N
M
= h M (ran( ) { p}) 1 M,
(11.12)
M = (J [ E],
Let us first show that
ran( )
ran( ) = ran( ).
(11.13)
It is easy to see of (11.13). To show of (11.13), suppose that y =
h M (n, (z, p)) x, where n < and z, x ran( ). Since y, z M p , the 1
statement y = h M (n, (z, p)) can be equivalently expressed in the form A(k, (y, z))
for some k . As also x M p , we thus have that
v x A(k, (v, z)),
a 0 -statement which is true in M p , so that
(v, z )),
v x A(k,
holds true in N , where x = 1 (x) and z = 1 (z). Let y x be such that
( y , z )). Then A(k, (( y ), z)), so that in fact ( y ) = h M (n, (z, p)) = y, i.e.,
A(k,
y ran( ). We have shown (11.13).
Equation (11.13) now immediately implies that
and J [ E ] = J [ E].
(11.14)
243
p = 1 ( p).
1 ( M)
(11.15)
there is a
Well, by (11.12) and (11.14) and as there is a 1M map of onto J [ E],
let P be M ({q})
and let < .
To show that 1 ( M),
for some q M,
1
By (11.12) we can find an n < and some
We aim to see that P M.
such that
x N = J [ E]
z P M |= n ((z, x), p)
(11.16)
A(k,
y) A(k, (y)) M |= k ( (y), p) M |= k (y, p).
In particular,
(z, x)) M |= n ((z, x), p)
A(n,
(11.17)
We now aim to prove a dual result, the upward extension of embeddings lemma.
Definition 11.19 Let M and M be acceptable J-structures. A map : M M is
called a good embedding iff
(a) : M 1 M
(b) For all R and R such that R M 2 is rudimentary over M and R M 2 is
rudimentary over M by the same definition,
if R is well-founded, then so isR.
244
M
p
M
p
N
245
(n, x) E(m,
y) h (n, (x, p))
h (m, (y, p))
[n, x] W / I (n, x) W .
We obviously have
I, F/
I, W / I)
S = (M/ I; E/
= (J [F]; , F, W ),
(n, x) E(m,
y) (n 3 , ((n, x), (m, y))) A M
p
(n, x) F (n 4 , (n, x)) A M
p
(n, x) W (n 5 , (n, x)) A M
246
(11.18)
F,
and W are I-invariant
The fact that I is an equivalence relation and that E,
may easily be formulated in a 1 fashion over M p . As is assumed to be 1 elementary, I is thus also an equivalence relation, and E, F, and W are I -invariant.
We may therefore write, for (n, x), (m, y) M ,
[n, x] = {[m, y]: (n, x)I (m, y)}
M /I = {[n, x] : (n, x) M }
(11.19)
and
(e( ), ((m 1 , x1 ), . . . , (m k , xk ))) A
[ (e(), ((m 1 , x1 ), . . . , (m k , xk ))) A
(e(), ((m 1 , x1 ), . . . , (m k , xk ))) A ].
(11.20)
Equations (11.19) and (11.20) just follow from the corresponding facts for A M and the
1 -elementarity of . We also need versions of (11.19) and (11.20) for quantification.
In order to arrive at these versions, we are going to use the 1 -Skolem function for
M to express 2 -truth over M in a 1 fashion over M p . For the sake of readability,
let us pretend in what follows that if n is 1 but not 0 , then n has only one free
variable, w, and that it is in fact of the form v (v, w), where is 0 . We may
pick a (partial) computable map e:
such that for all n < in the domain of
is 0 and
e,
e(n)
247
m x (e(e(n)),
(11.21)
(e(e(n)),
(11.22)
M |= v e(n)
(v, h M (m, (x, p)))
M |= e(n)
h M (m, (x, p)))
(e(e(n)),
(11.23)
We have shown that for all formulae = n which are 1 but not 0 and for all
(m, x) N ,
(e(n), (m, x)) A
(m , x ) N (e(e(n)),
(11.24)
248
(11.25)
by
: M M
([n, (x)] ),
h M (n, (x, p))
249
Moreover, is 2 -elementary, by the following reasoning. Let n < , and say that
say z = h (, (y, p)),
where
n has two free variables, v and w. Then for all z M,
M
< and y M p ,
M |= v n (v, z) (m, x) M p M |= n (h M (m, (x, p)),
h M (, (y, p)))
p
M p |= m, x (e(n), ((m, x), (, y))) A
that A = A M .
For x M p and n < , we have that
p
rl(n, x) A M M |= n (x, p)
250
yields that
p
(v1 = h(v2 , (v3 , v4 )), ((n, x), (k1 , n), ((k1 , x), (k2 , 0)))) A M ,
so that for all (n, x) N ,
(v1 = h(v2 , (v3 , v4 )), ((n, x), (k1 , n), ((k1 , x), (k2 , 0)))) A ,
(11.26)
(11.27)
x B (x, z) B,
which in turn for some fixed n (namely, the Gdel number of the defining formula)
is equivalent to (n, (x, z)) A . But (N , A ) is amenable, so that if < , then
{z} B N , and hence also B N . This shows that 1 (M). Thus,
= 1 (M) and p R M .
R be binary relations which are
It only remains to show that is good. Let R,
n+1, p
for p Mn+1 , A M
251
We also set (M) = min{n (M); n < }. The ordinal (M) is called the ultimate
projectum of M.
We remark that if M is not 1-sound (cf. Definition 11.28) then it need not be the case
that 2 (M) is the least such that P() 2M M.
n+1
If n (M)
1 (M), then we may identify p = ( p(0), . . . , p(n)) M
with the (finite) set ran( p) of ordinals; this will play a rle in the next section.
n =
R nM PM
n,q
Let p R nM . If q Mn then A M is rud M n, p in parameters from M n, p .
n
n,
p
Let p R M . Then 1 (M ) = n+1 (M).
n , then for all i < n, p(i) P
n
If p PM
M i, pi . If p R M , then for all i < n,
n1
p(i) R M i, pi . Moreover, if p (n 1) R M , then p(n 1) PM n1, pn1
n and p(n 1) R
n
implies that p PM
M n1, pn1 implies that p R M .
Proof (a) This is easily shown inductively by using Lemma 11.11 and amalgamating
parameters.
(b) By induction on n < . The case n = 0 is trivial. Now let n > 0, and suppose (b)
holds for n 1. Write m = n 1. Let p R nM and q Mn . We have to show that
q(m), (M)
is rud M n, p in parameters from M n, p . Inductively, M m,q m is rud M m, pm
A M m,qmn
in a parameter t M n, p . As p(m) R M m, pm , there are e0 and e1 and z M n, p
such that
q(m) = h M m, pm (e0 , (z, p(m)))
and
t = h M m, pm (e1 , (z, p(m))).
For i < and x M n, p , we have that
q(m), (M)
(i, x) A M m,qmn
252
the parameter z. (c) Let n+1 (M) = 1 (M n,q ), where q Mn . By (b), M n,q is
n,q
n, p
rud M n, p in parameters from M n, p , which implies that 1M 1M . But then
(11.28)
n+1, p
i+1, p
for i > 0, h M
hM
n+1, p
(X ) for h M
( < X ).
n+1, p
M = hM
"(n+1 (M)).
(11.30)
Lemma 11.26 Let 0 < n < . Let M be an acceptable J-structure, and let p R nM .
n, p
Then M P(M n, p ) = M .
n, p
x A M |= x1 x2 /xk (x, y, x1 , x2 , , xk ),
253
Lemma 11.25.
n, p
, as h nM is definable over M by
A more careful look at the proofs of Lemmata 11.25 and 11.26 shows the following.
Lemma 11.27 Let n < . Let M be an acceptable J-structure, and let p R nM . Let
n, p
M . Then A is
A M n, p be n+1
1M .
n.
Definition 11.28 Let M be an acceptable J-structure. M is n-sound iff R nM = PM
M is sound iff M is n-sound for all n < .
We shall prove later (cf. Lemma 11.53) that every J is sound. It is in fact a crucial
requirement on L-like models that there proper initial segments be sound.
We may now formulate generalizations of the downward and the upward extension
of embeddings Lemmas 11.16, 11.18, and 11.20.
Lemma 11.29 (General Downward Extension of Embeddings Lemma, Part 1) Let
n > 0. Let M and M be acceptable J-structures, and let : M n, p 0 M n, p , where
( p)
= M,
= p
p R nM . Then there is a unique map such that dom()
and, setting i = M i, p i ,
i : M i, p i 0 M i, pi for i n.
For i < n, the map i is in fact 1 -elementary.
In particular,
n, p
n, p
h M (k, x) = h M (k, (x))
for every k < and x M n, p .
Lemma 11.30 (General Downward Extensions of Embeddings Lemma, Part 2) Let
M be an acceptable J-structure, and let p M. Let N be a J-structure, and let
p such that p R n and N = M n, p .
: N 0 M n, p . Then there are unique M,
M
The general upward extension of embeddings lemma is the conjunction of the
following lemma together with Lemmas 11.29 and 11.30.
Lemma 11.31 (General Upward Extensions of Embeddings Lemma) Let :
M n, p 1 N be good, where M is an acceptable J-structure and p R nM . Then
there are unique M, p such that M is an acceptable J-structure, p R nM and
M n, p = N . Moreover, if is as in Lemma 11.29, then is good.
254
255
pn (M) = pn+1 (M) n. Then we set p(M) = n< pn (M). p(M) is called the
standard parameter of M.
We shall often confuse p(M) with ran( p(M)). Lemma 11.35 readily gives the
following.
Corollary 11.37 Let M be an acceptable J-structure which is also sound. Then
p(M) exists.
Lemma 11.38 Let M be an acceptable J-structure. M is sound iff pn (M) R nM
for all n .
Proof We shall prove the non-trivial direction =. We need to see that for each
n > 0,
n
pn (M) R nM = R nM = PM
.
(11.32)
n \ R n = by Lemma
Suppose n > 0 to be least such that (11.32) fails. Hence PM
M
n
n
11.24 (a). Let q be the < -least element of PM \ R M . This means that p < q, where
p = pn (M).
n1
n1,q (n1) ) =
= R n1
We have that q (n 1) PM
n+1 (M), and
M , 1 (M
q(n) PM n1,q(n1) \ R M n1,q(n1) . Let
: N
= h M n1,q(n1) (n (M) {q(n 1)}) 1 M n1,q (n1) ,
where N is transitive. By the Downward Extension of Embeddings Lemma 11.29
q,
and 11.30, there are unique M,
and such that
, N = M n1,q ,
q R n1
M
:
M M is r n elementary, and
and (
q)
= q (n 1).
(11.33)
(11.34)
256
Because
M i,q i |= r < q(i) x [i+1 (M)]< e < q(i) = h M i,q i (e , (x, r )), (11.35)
q(k)
and A M k, p k and A M k,qk are easily computable from each other. (11.35) and (11.36)
n . Also, p < q. However, p ran(
then give that p PM
), whereas q
/ R nM
implies that p
/ ran( ), and hence p
/ Rn . This contradicts the choice of q.
Solidity witnesses are witnesses to the fact that a given ordinal is a member of the
standard parameter. We shall make use of witnesses in the proof of Theorem 11.64.
Definition 11.40 Let M be an acceptable J-structure, let p OR M < , and let
p. Let W be an acceptable J-structure with W , and let r OR W < . We
say that (W, r ), or just W , is a witness for p with respect to M, p iff for every
1 formula (v0 , . . . , vl+1 ) and for all 0 , . . ., l <
M |= (0 , . . . , l , p \ ( + 1)) = W |= (0 , . . . , l , r ).
(11.37)
By the proof of the following Lemma, if a witness exists, then there is also one
where = may be replaced by in (11.37).
Lemma 11.41 Let M be an acceptable J-structure, and let p PM . Suppose that
for each p there is a witness W for p with respect to M, p such that W M.
Then p = p1 (M).
Proof Suppose not. Then p1 (M) < p, and we may let p \ p1 (M) be such that
p \ ( + 1) = p1 (M) \ ( + 1). Let us write q = p \ ( + 1) = p1 (M) \ . Let
(W, r ) M be a witness for p with respect to M, p. Let A 1M ({ p1 (M)}) be
such that A 1 (M)
/ M.
257
(11.38)
(11.39)
for every < 1 (M) and every 1 -formula . In particular, 11.39 holds for
and every < 1 (M) . As W M, this shows that in fact A 1 (M) M.
Contradiction!
Definition 11.42 Let M be an acceptable J-structure, let p On M < , and let
, p
p. We denote by W M the transitive collapse of h M ( ( p \ ( + 1))). We call
, p
W M the standard witness for p with respect to M, p.
Lemma 11.43 Let M be an acceptable J-structure, and let p PM . The
following are equivalent.
, p
(1) W M M.
(2) There is a witness (W, r ) for p with respect to M, p such that W M.
, p
258
WM
Let (W , r ) M be such
p OR M , and set = ( ) and p = ( p).
259
Lemma 11.50 Let M and M be acceptable J-structures, let n > 0, and let
: M M be r n elementary as being witnessed by pn1 (M). If M is n-solid and
and M is n-solid.
( p1 ( M n1 )) PM n1 then pn (M) = ( pn ( M))
Lemma 11.51 Let M and M be acceptable J-structures, let n > 0, and let : M
M be r n elementary as being witnessed by 1 ( pn1 (M)). Suppose that M is n, p (M k )
n1
solid, and in fact W M k 1
ran( ) for every k < n. If 1 ( pn1 (M)) PM
The ultrapower maps we shall deal with in the next section shall be elementary
in the sense of the following definition. (Cf. [30, Definition 2.8.4].)
Definition 11.52 Let both M and N be acceptable, let : M N , and let n < .
Then is called an n-embedding if the following hold true.
(1)
(2)
(3)
(4)
(11.40)
(J )n
1
260
(11.42)
Suppose that this is false, and let n be least such that p = pn+1 (J )
/ R n+1
J . Let us
consider
(J , A)
= h (J )n (n+1 (J ) { p(n)}) 1 (J )n .
(11.43)
By the Downward Extension of Embeddings Lemma 11.29 and 11.30 we may extend
to a map
: J 1 J
=
such that p n ran( ) and writing p = 1 ( p n), p R nJ and (J , A)
n,
p
1
1
(J ) . Let us also write p = ( p(n)) = ( p(n)).
(J )n
Let B be 1 ({ p(n)}) such that B n+1 (J )
/ J , say
B = {x (J )n : (J )n |= (x, p(n))},
where is 1 . Let
B = {x (J )n, p : (J )n, p |= (x, p )}.
As n+1 (J ) = id,
/ (J )n .
B n+1 (J ) = B n+1 (J )
(11.44)
that p(i)
p(i), as ( p(i))
= (J )n .
This yields that in fact p = p n, and therefore (J , A)
(J )n, p
261
W(J )1n
(J )n .
(11.45)
Let us thus fix n < and p1 ((J )n ). Let us write p = p(J ), so that
(J )n = (J )n, pn . Let us consider
(J , A)
= h (J )n ( { p(n) \ ( + 1)}) 1 (J )n ,
(11.46)
, p ((J )n )
(J ) .
In order to verify (11.45), it suffices to prove that < . This is because
= (J )n, p J . But it is clear from (11.46) that
if < , then (J , A)
Let B 1
B = {x (J )n : (J )n |= (x, (r ))},
then
/ (J )n .
B 1 ((J )n ) = B 1 ((J )n )
(11.47)
262
(11.48)
(J )n
such that (r ) = h (J )n (m, s), so that B 1 ({s}). But (11.48) gives that
s < p(n), so that this contradicts the choice of p(n).
263
The most difficult part here is (4) (1) which is due to Ronald Jensen, cf. [10],
and which is called Jensens Covering Lemma. There is, of course, a version of
Theorem 11.56 for L[x], x , which we leave to the readers discretion.
It is not possible to cross out +1 in (1) or replace 1 by 0 in (2) of
Theorem 11.56, cf. Problem 11.7 (cf. Problem 11.9, though).
(4) = (3) of Theorem 11.56 was shown as Theorem 10.39. (2) = (1) is trivial.
If 0# exists, then every uncountable cardinal of V is a Silver indiscernible, so that
{n : n < } cannot be covered in L by a set of size less than . This shows (1)
= (4). We are left with having to verify (3) = (2).
Let us first observe that it suffices to prove (3) = (2) of Theorem 11.56 for the
case that be regular. This follows from:
Lemma 11.57 Let W be an inner model. Let be a singular cardinal, and suppose
that for all < , 1 , and for all , [ ] W is stationary in [ ] . Then
for all , [ ] W is stationary in [ ] .
Proof Let A = ( ; ( f : < )) be any algebra on . We need to see that there is
some X [ ] W such that for all < , if f is n-ary, n < , then f [X ]n X ,
i.e., X is closed under f . Let (i : i < cf()) be monotone and cofinal in , 0 1 .
If i < cf(), 1 k < , and X 1 , . . . , X k [ ]i W , then by our hypothesis
there is some X [ ]i such that X X 1 X k and X is closed under all the
functions f with < i . We may thus pick, for every i < cf() and 1 k <
some
ik : [[ ]i W ]k [ ]i W
(11.49)
264
Proof of Theorem 11.56, (3) = (2). Let us fix , where 1 is regular. Let
> be a regular cardinal. Let : H H be elementary s.t. H is transitive. Let
(i : i < ) = (i : i < )
enumerate the transfinite cardinals of L H = J H O R , and set = = H O R.
For i let i = i O R {} be largest such that i is a cardinal in Ji .
Hence by Lemma 11.53, if i < , then (Ji ) < i , whereas (J ) i for
all [ , i ).
If i j then j i , so that {i : i } is finite. For each i with i <
we let n i = n i be such that
n i +1 (Ji ) < i n i (Ji ).
(11.50)
265
H
J i
ultni (J i , EJ )
i
J i
ultn j (J j , EJ )
j
J j
( j )
( i)
LH
(crit( ))
j
i
crit( )
i
(11.51)
be the ultrapower map. Let p = p(J j ) (n j + 1). By Lemmas 11.53 and 11.25,
266
n +1, p
J j = h Jj
(n j +1 (J j )).
(11.52)
By Lemma 11.29, the map from (11.51) has the property that
n +1, p
(h
Jj
n +1, ( p)
(k, x)) = h Jj
(k, (x))
for every k < and x [n j +1 (J j )]< , so that (11.52) immediately gives that
n +1, ( p)
ran( ) = h Jj
( (n j +1 (J j ))).
(11.53)
as desired.
Suppose that ult n i (Ji ; E Ji ) is not well-founded. Then by Lemma 11.31, either
ult0 ((Ji )n i ; E Ji ) is ill-founded or else the ultrapower map
: (Ji )n i 1 ult 0 ((Ji )n i ; E Ji )
(11.54)
is not good in the sense of Definition 11.19. In both cases, there is a wellfounded relation R ((Ji )n )2 which is rudimentary over (Ji )n such that if
R (ult 0 ((Ji )n i ; E Ji ))2 is rudimentary over ult0 ((Ji )n i ; E Ji ) via the
same definition, then R is ill-founded. We may then pick ([ak , f k ]: k < ) with
[ak , f k ] ult 0 ((Ji )n i ; E Ji ) and [ak+1 , f k+1 ]R[ak , f k ] for every k < .
In what follows, we shall refer to the fact that ult0 ((Ji )n i ; E Ji ) is illfounded or that the ultrapower map as in (11.54) is not good by saying that
ult n i (Ji ; E Ji ) is bad, and we shall call
R, ([ak , f k ]: k < ))
( R,
(11.55)
is not bad.
267
Yi <
, for all i <
Y = i< Yi for all limit ordinals
Yi+1 f j Yi< for all j < i <
i : H i
= Yi , where H i is transitive, and
Suppose that j I i and ult n ij (J i ; E i J
j
j i
j i
j i
),
)
i, j
Lemma 11.31. By Claim 11.58, this then finishes the proof of (3) = (2) of Theorem
11.56, as obviously Y is closed under all functions from A. Let us assume is not
as claimed and work towards a contradiction.
We write Y = Y and H = H . We also write I = I , i = i , etc.
By assumption, there is some j I such that ult n j (J j ; E J j ) is bad. Let
R, ([ak , f k ] : k < )) be a badness witness for ult n j (J j ; E J ).
( R,
1
1 Yi
i : H i H and Y i =
=
Let us write i =
ran( i ). Hence
Y i H , Y i Y l for i l, Card(Y i ) < for i < , Y = i< Y i for limit
ordinals and H = i< Y i . Let > be some sufficiently large regular
cardinal. As is regular, we may pick some Z H such that (Y i : i ) Z and
Z . If i 0 = Z , then Z H = Y i0 . We may thus assume that Z H is
such that
(1) Z < ,
H } Z , and
(2) { f k : k < } { R,
(3) Z H = Y i0 for some i 0 < .
Let us write : H
= Z , where H is transitive. Let us also write R = 1 ( R)
1
1
and f k = ( f k ) for k < . Obviously, (H ) = H i0 and H i0 = i0 . Let
k < . We then have [ak+1 , f k+1 ]R[ak , f k ] in ult0 ((J j )n j ; E J j ), and hence
268
and n = nl 0 . Therefore,
ult
i
0
nl
(J
i
0
; E i0 J
i
l 0
(11.56)
is bad. Hence by (6) there is a badness witness (R , R , ([aki0 ,l , f ki0 ,l ] : k < )) for
(11.56) such that {aki0 ,l : k < } Yi0 +1 Y .
i 0 ,l
i 0 ,l
, f k+1
]R [aki0 ,l , f ki0 ,l ] gives that
Let k < . Then [ak+1
i 0 ,l
i 0 ,l
ak+1
, aki0 ,l i0 ({(u, v) : f k+1
(u)R f ki0 ,l (v)})
i 0 ,l
(u)R f ki0 ,l (v)})
= i0 ({(u, v) : f k+1
i 0 ,l
(u)R f ki0 ,l (v)})
= ({(u, v) : f k+1
i 0 ,l
(u) R
f ki0 ,l (v) .
= (u, v) : f k+1
i 0 ,l
However, ak+1
, aki0 ,l Y = ran( ), so that this gives that
i 0 ,l
i 0 ,l
i 0 ,l
f k+1
i 0 ,l
1 ak+1
R
f ki0 ,l 1 aki0 ,l .
Corollary 11.60 (Weak covering for L ) Suppose that 0# does not exist. If 2
is a cardinal in L, then in V , cf( +L ) Card(). In particular, +L = + for every
singular cardinal .
269
Proof Assume that 0# does not exist, and let 2 be a cardinal in L. Suppose that
in V , cf( +L ) < Card(), and let X +L be cofinal with Card(X ) < Card().
By Theorem 11.56 (1), there is some Y L such that Y X and Card(Y )
Card(X ) + 1 . As Card() 2 , Card(Y ) < Card(). This implies that otp(Y ) <
(Card(Y ))+ Card() < +L . Contradiction!
If is singular, then cf( +L ) = , so that cf( +L ) Card() = implies
+L
= +.
Corollary 11.61 Suppose that 0# does not exist. Then SCH, the Singular Cardinal
Hypothesis, holds true.
Proof Assume that 0# does not exist. Let be a (singular) limit cardinal. We need
to see that cf() + 2cf() (which implies that in fact cf() = + 2cf() ). If
X []cf() , then by Theorem 11.56 (1) there is some Y []cf()1 L such
that Y X . On the other hand, for any Y []cf()1 there are (cf() 1 )cf() =
2cf() many X Y with Card(X ) = cf(). Moreover, as the GCH is true in L,
[]cf()1 L has size at most + . Therefore, cf() + 2cf() .
270
Proof This proof is in need of the fact that every level of the L-hierarchy is solid,
cf. Definition 11.40 and Theorem 11.54.
Let us set C = { < + : J J + }, which is a club subset of + consisting
of limit ordinals above .
Let C. Obviously, is the largest cardinal of J . We may let () be the largest
such that either = or is a cardinal in J . By Lemma 5.15, (J() ) = .
Let n() be the unique n < such that = n+1 (J() ) < n (J() ).
If C, then we define D as follows. We let D consist of all C such
that n( ) = n() and there is a weakly r n()+1 elementary embedding
: J( ) J() .
such that = id, ( pn( )+1 (J( ) )) = pn()+1 (J() ), and if J( ) , then
J() and ( ) = . It is easy to see that if D , then by Lemma 11.53 there
is exactly one map witnessing this, namely the one which is given by
n( )+1, pn( )+1 (J( ) )
h J( )
(i, x) h J()
(i, x),
(11.57)
J () = h J()
so that if D , then
n()+1, pn()+1 (J() )
ran( , ) h J()
which means that there must be i < and < such that the left hand side of
(11.57) is undefined, whereas the right hand side of (11.57) is defined.
Notice that the maps , trivially commute, i.e., if D and D , then
D and
, = , , .
Claim 11.65 Let C. The following hold true.
(a) D is closed.
(b) If cf() > , then D is unbounded in .
(c) If D then D = D .
Proof (a) is easy. Let be a limit point of D . If < , , D , then
ran(, ) ran( , ).
We then have that the inverse of the transitive collapse of
271
ran(, )
proves that D .
Let us now show (b). Suppose that cf() > . Set = () and n = n(). Let
< . We aim to show that D \ = .
Let : J n+1 J be such that is countable, ran( ), and
, p1 (Jk )
{W J k
: p1 (Jk ), k n} ran( ).
h J( )
where i < and x []< . We need to see that if i < , x []< , and h (i, x)
exists, then h (i, x) exists as well.
Let e < and y []< be such that = h (e, y). We obviously cannot have
that h (e, y) exists, as otherwise
, (h (e, y)) = h (e, y) = , (h (e, y)) = , () = ,
but
/ ran(, ). Write n = n() = n( ) = n(). Let < n (J( ) ) be least such
that S contains a witness to v v = h (J( ) )n (u(e), (y, p(J( ) )(n))), cf. (11.29).
We claim that
ran(, ) (J() )n , (S ).
(11.58)
If (11.58) is false, then we may pick some < n (J() ) such that , ( )
, (). We may then write h (e, y) = as a statement over J() in the parameter
272
h J
(k, x)
/ ran(i , )
be the least D
for some k < and some x [ ]< . Given i , we let i+1
such that
n+1, p
(J )
h J n+1 (k, x) ran( , )
(J )
for all k < and x [i ]< such that h J n+1 (k, x) exists. Finally, given
(i : i < ), where is a limit ordinal, we set = sup({i : i < }). Naturally,
() will be the least i such that i = . We set C = {i : i < ()}.
n+1, p
Proof (a) is immediate, and it implies (b). (c) and (e) are trivial.
Let us show (d). Let C . We have C D , and D = D by Claim 1
(c). We may then show that (i : i < ( )) = (i : i < ( )) and (i : i < ( )) =
(i : i < ( )) by an induction. Say i = i , where i + 1 ( ) (). Write
= i = i . As , = , , , for all k < and x []< ,
n( )+1, pn( )+1 (J( ) )
h J( )
h J()
(k, x) ran(, ) =
(k, ) ran(, ) = .
.
so as to guarantee conversely that i+1
i+1
Now let f : + C be the monotone enumeration of C. For < + , let us set
B = f 1 C f () .
273
C \ ( + 1) if C and otp(C ) =
otherwise.
C
274
We now use at most the next Card(S) ordinals above T to construct T (+1)
in such a way that for every s S there is some t T ( + 1) \ T such that
s <T t for all s cs and for every t T ( + 1) \ T there is some s S
such that s <T t for all s cs .
This finishes the construction. It is not hard to verify that the construction in
fact never breaks down and produces a + -tree. Otherwise, for some limit ordinal
< + there would be some s S and some limit ordinal i < such that there is
no t T (i + 1) \ T i with s j <T t for all j < i (for S, , (i : i < ) as
/ R, so that Ci = C i is easily seen
above). However, i C gives that i
to yield that in the construction of T (i + 1) \ T i , we did indeed add some
t T (i + 1) \ T i such that s j <T t for all j < i.
Finally, suppose that T would not be + -Souslin, and let A T be an antichain
of size + . As R is stationary, we may then pick some limit ordinal R such that
A = S is a maximal antichain in T , so that also S = {s T : t
S (t = s t <T s)} (for S as above). The construction of T then gives that every
node in T \ T is above some node in S , so that in fact A = A = S . But
because T is a + -tree, this means that A has size at most . Contradiction!
Lemma 11.69 (Jensen) Let be subcompact. Then fails.
Proof Suppose (C : < + ) witnesses . As is subcompact, there must then
be some < together with a witness (D : < + ) to and an elementary
embedding
: (H+ ; , (D : < + )) (H + ; , (C : < + ))
with crit( ) = . Set = sup( + ) < + . As + is < -club in ,
C = C +
is also < -club in .
Let () < () < both be limit points of C. Then
C () () = C () = C () .
(11.59)
D ,
()C
275
A fine structure theory for inner models is developped e.g. in [30] and [47], and
fine structural models with significant large cardinals are constructed e.g. in [2, 31,
40]; cf. also [45] and [46]. Generalizations of Jensens Covering Lemma 11.56 are
shown in [29] and [28]. In the light of Theorem 11.70, an ultimate generalization of
Theorem 11.64 is shown in [35], and an application in the spirit of Corollary 11.67
is given in [17].
11.4 Problems
11.1. Assume GCH to hold in V . Show that there is some E OR such that
V = L[E] and L[E] is acceptable. [Hint. Use Problem 5.12.]
11.2. Let L[U ] be as in Problem 10.3. Show that L[U ] is not acceptable. Show also
that L[U ] is weakly acceptable in the following sense. If (P() J+ [U ])\
J [U ] = , then there is some surjection f : P() J [U ], f
J+ [U ]. [Hint. Problem 10.5.]
11.3. Prove Lemma 11.13!
M, p,
11.4. Let M,
p, , be as in Lemma 11.16. Suppose moreover that p R M ,
and that : M p n M p . Then : M n+1 M.
11.5. Let M be an acceptable J-structure, and let n < . Show that if is a
cardinal of M such that n+1 (M) < +M n (M), then cf( +M ) =
cf(n (M)).
Let be a regular uncountable cardinal, and let > be a cardinal. By
, we mean the following statement. There is a family (Ax : x []< )
such that for every x []< , Ax P(x) and Card(Ax ) Card(x), and
for every A there is some club C []< such that for every x C ,
A x Ax .
11.6. (R. Jensen) Assume V = L. Let be a regular uncountable cardinal, and
let > be a cardinal. Then , holds true. [Hint. If x = X , where
276
J
= X J , then let > be least such that (J ) < and set
Ax = { y x: y J P()}. Cf. the proof of Theorem 5.39.]
11.7. Let N be the set of all perfect trees on 2 . N, ordered by U N T iff
U T , is called Namba forcing. Show that if G is N-generic over V ,
then cf V [G] (2V ) = and 1V [G] = 1V . [Hint. Let G be N-generic over
V . Then
G is cofinal from into 2V . Let T : 1 . Choose
(ts , Ts , s : s < 2 ) such that T = T , t = , ts Ts , {ts : < 2 } Ts
is a set of 2 extensions of ts of the same length, ts = ts for = ,
Ts N Ts , Ts {t T : ts t ts t}, and Ts ran( (lh(s))) n ,
n < 1 . For < , set
T =
{Ts : lh(s) = n s < }: n < .
11.4 Problems
277
278
Chapter 12
12.1 Determinacy
E. Zermelo observed that finite two player games (which dont allow a tie) are
determined in that one of the two players has a winning strategy. Let X be any nonempty set, and let n < . Let A 2n X , and let players I and II alternate playing
elements x0 , x1 , x2 , . . ., x2n1 of X . Say that I wins iff (x0 , x1 , x2 , . . . , x2n1 ) A,
otherwise II wins. Then either I has a winning strategy, i.e.,
x0 x1 x2 . . . x2n1 (x0 , x1 , x2 , . . . , x2n1 ) A,
or else II has a winning strategy, i.e.,
x0 x1 x2 . . . x2n1 (x0 , x1 , x2 , . . . , x2n1 ) 2n X \A.
Jan Mycielski (* 1932) and Hugo Steinhaus (18871972) proposed studying
infinite games and their winning strategies, which led to a deep structural theory of
definable sets of reals. Let X be a non-empty set, and let A X . We associate to A
a game, called G(A), which we define as follows. In a run of this game two players,
I and II, alternate playing elements x0 , x1 , x2 , . . . of X as follows.
I x0
x2
...
II
x1
x3
...
After moves they produced an element x = (x0 , x1 , x2 , . . . ) of X . We say that I
wins this run of G(A) iff x A, otherwise II wins. A strategy for I is a function
:
2n
X X,
n<
279
280
2n+1
X X.
n<
12.1 Determinacy
281
(x : < 20 ) and (y : < 20 ) such that for all < 20 , x = z for some
/ {y : < }, and y = z for some z such that
z such that z
/ {x : }. It is then easy to see that
z z
A = {y : < 20 }
cannot be determined.
Now let (An : n < ) be a sequence of non-empty sets of reals, and let us consider
the following game (in which I just keeps passing after his first move).
In
...
II n 1 n 2 x3 n 4 . . .
I plays some n , and then II plays some x = (n 0 , n 1 , n 2 , . . .). We say that II
wins iff x An .
Of course, I cannot have a winning strategy in this game. Hence II has a winning
strategy, as our game may be construed as G( \A), where A = {x: (x(1), x(3), . . .)
A x(0) }, and IIs winning strategy then gives rise to a function f as desired.
With a some extra work beyond Theorem 13.7 one can construct models of
ZF + DC + AD, which are of the form of the models of Theorems 6.69 or 8.30.
The moral of this is that whereas AD is false it holds for definable sets of reals,
and the results of this section should be thought of being applied inside models of
ZF + AD which contain all the reals.
We first want to show that open games on are determined. Let still X be an arbitrary
non-empty set. Recall, cf. p. 123, that we may construe X as a topological space
as follows. For s < X , set Us = {x X : s x}. The sets Us are declared to be
< X
the basic open
sets, so that a set A X is called open iff there is some Y
with A = sY Us .
If , is a strategy for player I , II, respectively, then we say that x is according
to , iff there is some y such that x = y, x = y , respectively.
Theorem 12.3 (Gale, Stewart) Let A X be open. Then A is determined.
Proof Let us suppose I not to have a winning strategy in G(A). We aim to produce
a winning strategy for II in G(A). Let us say that I has a winning strategy in G s (A),
where s < X has even length, iff I has a winning strategy in G({x X : s x
A}). By our hypothesis, I doesnt have a winning strategy in G (A).
Claim 12.4 Let s < X have even length, and suppose I not to have a winning
strategy in G s (A). Then for all y X there is some z X such that I doesnt have
a winning strategy in G s y z (A).
Proof Otherwise there is some y X such that for all z X , I has a winning
strategy in G s y z (A). But then I has a winning strategy in G s (A): he first plays
such a y, and subsequently, after II played z, follows his strategy in G s y z (A).
282
I x0
x2
...
II
x1
x3
...
12.1 Determinacy
283
I Xk
X k+1
...
II
k
k+1
...
In round l k, I has to play some X l U . II has to reply with some l X l such
that l > l1 (if l > 0). II wins the game iff there is some n < and some X U
such that
({0 , . . . , n1 }, X ) D.
G s is a closed game. Moreover, G s M. Therefore, one of the two players has a
winning strategy in M which works for plays in V .
Claim 12.8 I does not have a winning strategy in M.
Proof Suppose M to be a winning strategy for I. We claim that there is some
Z U such that all k < k+1 < in Z are compatible with , by which we mean
that there is a play of G s in which I follows and II plays k , k+1 , . . .. In order to
get Z , define F: []< 2 by F({k < < n1 }) = 1 iff k < < n1 are
compatible with . As F M, we may let Z U be such that F is constant on [Z ]l
for every l < . It cannot be that F Z ]l = {0} for some l < ; this is because in
round m, where k m < k + l, if tells I to play X m then II can reply with some
m X m Z .
Now let us look at ({0 , . . . , k1 }, Z ) P. As D M is dense in P there is
some
({0 , . . . , n1 }, Z ) ({0 , . . . , k1 }, Z )
with
({0 , . . . , n1 }, Z ) D.
Because k < < n1 Z , k < < n1 are compatible with ; on the
other hand, II wins if he plays k , . . . , n1 in rounds k, . . . , n 1. Contradiction!
Let us still fix s = {0 < < k1 } for a while, and let = s M be a
winning strategy for II in G s (which also works for plays in V ).
We shall now for l k and t = {0 < < k1 < < l1 } s and
> l1 define sets Yst and X st, which are in U . The definition will be by recursion
on the length of t. We shall call t {l } realizable iff
I X ss,k
...
X st,l
II
k
...
l
is a position in the game G s in which I obeyed the rules and II played according
to .
t m,m
t m
and Ys have been defined for all
Now fix t and l and assume that X s
t
k m < l, where l k. We first aim to define Ys .
284
Claim 12.9 Suppose t to be realizable. There is then some Y U such that for all
Y there is an X such that
t l1
I X ss,k
...
Xs
II
k
...
X
l1
I X ss,k
...
Xs
II
k
...
X
l1
...
Xs
I X ss,
II
...
l1
\Y
Because II follows s in the play above, there is some m n and some X with
({0 , . . . , m1 }, X ) D. But then ({0 , . . . , m1 }, Z x m ) D. However, for any
l m, l X 0 , and hence l Z x m . We thus have that
1
12.1 Determinacy
285
({0 , . . . , m1 }, Z x m ) G.
Therefore, D G = .
We shall now prove that AD, the Axiom of Determinacy, proves the ultimate
generalization of the Cantor- Bendixson Theorem 1.9. If AD were not to contradict
the Axiom of Choice, it could be construed as providing a solution to Cantors
project of proving the Continuum Hypothesis, cf. p. 3. The following Theorem says
that all of P( ) has the perfect subset property (cf. p. 138).
Theorem 12.11 (M. Davis) Assume AD. Every uncountable A has a perfect
subset.
Proof Fix A . Let f : 2 be a continuous bijection (cf. Problem 7.2),
and write B = f A. It suffices to prove that B is either countable or else has a perfect
subset.
Let us consider the following game, G p (B).
I s0
s1
...
II n 0 n 1
...
In this game, I plays finite 01-sequences si < 2 (with si = being explicitly
allowed), and II plays n i 2 = {0, 1}. Player I wins iff
s0 n 0 s1 n 1 . . . B,
otherwise II wins. We may construe G p (B) as G(B ) for some B , so that
G p (B) is determined.
If I has a winning strategy in G p (B), , then
{s0 n 0 s1 n 1 . . . : (n 0 , n 1 , . . .) 2 i < si = (s0 n 0 . . . n i1 )}
is a perfect subset of B.
So let us suppose to be a winning strategy for II in G p (B). We say that a finite
sequence p = (s0 , n 0 , . . . , sk , n k ) is according to iff n i = (s0 n 0 . . . si ) for
every i k, and if x 2, then we say that p = (s0 , n 0 , . . . , sk , n k ) is compatible
with x iff s0 n 0 . . . si n i x. Because is a winning strategy for II, we
have that x
/ B follows from the fact that for all p = (s0 , n 0 , . . . , sk , n k ), if p
is according to and compatible with x, then there exists some s 2 such that
(s0 , n 0 , . . . , sk , n k , s, ((s0 , n 0 , . . . , sk , n k , s)) is compatible with x. In other words,
if x B, then there is some px = (s0 , n 0 , . . . , sk , n k ) which is according to and
compatible with x such that for all s 2, (s0 , n 0 , . . . ,sk , n k , s, ((s0 , n 0 , . . . , sk ,
n k , s)) is not compatible with x.
Notice that x px is injective for x B. This is because if x B, px =
(s0 , n 0 , . . . , sk , n k ), and
286
x = s0 n 0 . . . sk n k m 0 m 1 m 2 . . . ,
then we must have that m i = 1 (s0 n 0 . . . sk n k m 0 . . . m i1 ) for every
i < .
But as x px , x B, is injective, B is countable.
Lemma 12.12 Assume AD. Let A 2 be such that every Lebesgue measurable
D A is a null set. Then A is a null set.
Proof Let us fix A 2 such that every Lebesgue measurable D A is a null
set. Let > 0 be arbitrary. We aim to show that A can be covered by a countable
union B of basic open sets such that (B) . Let us consider the following game
G cov (A), the covering game for A.
n1
...
I n0
II
K0
K1
...
In this game, I plays n i {0, 1}, i < , and II plays K i , i < , where each K i is a
Ki .
i<
Notice that for each i < there are only countably many candidates for K i , so
that G cov (A) may be simulated by a game in which both players play just natural
numbers. Therefore, G cov (A) is determined.
We claim that I cannot have a winning strategy. Suppose not, let be a winning
strategy for I , and let D 2 be the set of all (n 0 , n 1 , . . .) such that there is a play
of G cov (A) in which II plays some sequence (K i : i < ) and (n i : i < ) is the
sequence of moves of I obtained by following the winning strategy in response
to (K i : i < ). The set D is then analytic and hence Lebesgue measurable by
Corollary 8.15. As is a winning strategy for I , D A, so that D is in fact a null
set by our hypothesis
on A.
In 23 .
Let D n< In , where each In is a basic open set and
n<
1
, so that by cutting and relabelling if necessary we may
Notice that 23 = i=0
22(i+1)
assume that there is a strictly
sequnce (i : i < ) of natural numbers
increasing
i+1 1
Ik 22(i+1)
. But now II can defeat I s alleged
with 0 = 0 such that
k=i
i+1 1
winning strategy by playing k=i Ik in her ith move. If I plays by following
, he will lose. Contradiction!
By AD, player II therefore has a winning strategy in G cov (A). For s < 2
with lh(s) = i + 1 \{0}, let us write K s for the i th move K i of II in a play in
which I s first i + 1 moves are s(0), . . ., s(i) and IIs first i + 1 moves K 0 , . . ., K i
are obtained by following in response to s(0), . . . , s(i). Write
12.1 Determinacy
287
B=
Ks .
s< 2\
1
,
22(lh(s))
so that
K s 2i+1
lh(s)=i+1
and hence
(B)
1
1
= i+1
2
22(i+1)
= .
i+1
2
i=0
is
is
Lemma 12.14 Assume AD. Let A be non-meager. There is then some s <
such that A Us is comeager in the space Us .
Proof Let us fix A . For s < , let us consider the following game, G bm
s (A),
called the Banach- Mazur game.
I s0 s2 . . .
II s1 s3
...
In this game, I and II alternate playing non-empty sn < , n < , and I wins iff
s s0 s1 s2 s3 . . . A.
Claim 12.15 II has a winning strategy in G bm
s (A) iff A Us is meager.
Proof This proof does
not use AD. Suppose first that A Us is meager, so that we
may write A Us = n< An , where each An is nowhere dense. Let us define a
strategy for II as follows. Let (s0 , s1 , . . . , s2n ) be some non-empty t such
that
Us s0 s1 ... s2n t An = .
288
This is always well-defined, as all An are nowhere dense, and is easily seen to be
a winning strategy for II in G bm
s (A).
Now let us assume that II has a winning strategy, , in G bm
s (A). We may define
what it means for a finite sequence p = (s0 , s1 , . . . , sn ) of non-empty elements of
< to be according to in much the same way as in the proof of Theorem 12.11.
Let x with s x. Because is a winning strategy for II, if for all sequences
(s0 , s1 , . . . , s2n1 ) of non-empty elements of < which are according to and such
that s p = s s0 s1 . . . s2n1 x there is some non-empty t < such that
/ A. Therefore,
s p t ( p t) x, then x
A Us
Bp,
G bm
(A).
Setting
s
=
(),
esily
induces
a winning strategy for II in G bm
s ( \A).
{Us : s < Us \A is meager}.
Trivially, O\A is meager. If A\O were non-meager, then by Lemma 12.14 there is
some s < such that (A\O) Us is comeager in Us . This means that Us \A
Us \(A\O) is meager in Us , so that by the definition of O, Us O. So (A\O)Us =
is not comeager in Us after all. Contradiction!
For x, y , we let x T y denote that x is Turing reducible to y, and we
write x T y for x T y and y T x. A set A is called Turing invariant iff
for all x A and y , if y T x, then y A. A set A is called a (Turing)
cone iff there is some x such that A = {y: x T y}, in which case x is also
called a base of the cone A.
Set S be a set of ordinals. A set A is called S-invariant iff for all x A and
y , if L[S, x] = L[S, y], then y A. We call A an S-cone iff there is
some x such that A = {y: x L[S, y]}, in which case x is also called a base
of the S-cone A.
12.1 Determinacy
289
Proof (a) Immediately follows from Theorem 12.11 via Corollary 7.29.
(b) Set B = {1L[x] : x }. By (a), B is a subset of 1 of size 1 . Let : 1 B
be the monotone enumeration of B. For X 1 , let A(X ) be the Turing invariant
set {x : 1L[x] X }. Let
U = {X 1 : A(X ) contains a cone}.
Notice that if x is a base for the cone A and y is a base for the cone B, then x y is
a cone for a base contained in A B. It is thus straightforward to verify that U is a
filter, and U is in fact an ultrafilter by Theorem 12.17. If {X n : n < } U , then by
AC (cf. Lemma 12.2) we may pick a sequence (xn : n < ) of reals such that for
each n < , xn is a base for a cone of reals
contained in A(X
n ). But then n< xn
is a base for a cone of reals contained in n< A(X n ) = A( n< X n ). This shows
that U is <1 -complete, so that U witnesses that 1 is a measurable cardinal.
(c) This follows from (b) and the proof of Lemma 10.31 which does not need
AC. If U is a <1 -complete measure on 1 , then for every x , U L[x] is
-complete, so that ult(L[x]; U L[x]) is well-founded by Lemma 10.29, and we
get x # by Theorem 10.39.
(d) Fix X 1 . Let us consider the following game, called the Solovay game.
I n0
n1
...
II
m1
m2
...
290
D(X i ).
i||x||
The Boundedness Lemma 7.12 implies as in (b) that I cannot have a winning
strategy in this game. Let be a winning strategy for player II. We aim to verify that
{z : T z} D(X i )
for every i < .
12.1 Determinacy
291
Let us fix z such that T z, and let us also fix i < 1 . By the existence
of z # , we may pick g V to be Col(, i)-generic over L[z]. Let x WO L[z][g]
be such that ||x|| = i. As x T x z, 1+L[xz] X i . However, 1+L[z]
+L[z][g]
1+L[xz] 1
, which is equal to 1+L[z] , as Col(, i) is smaller than 1V in
+L[z]
= 1+L[xz] X i , i.e., z D(X i ) as desired.
L[z]. Therefore 1
By OD S -determinacy we mean the statement that if A is OD S , cf. Definition
5.42, then A is determined.
Theorem 12.19 (A. Kechris) Assume AD. Let S O R. For an S-cone of reals x
we have
L[S, x] |= OD S -determinacy.
In particular, 1L[S,x] is measurable in HOD SL[S,x] .
Proof Let us assume that there is no S-cone of reals x such that in L[S, x], all OD S sets of reals are determined. By Theorem 12.17 there is thus an S-cone C such
that for every x C, in L[S, x] there is an non-determined OD SL[S,x] -set of reals.
Define, for x C, x A x by letting A x be the least OD SL[S,x] -set of reals which
is not determined in L[S, x]. (Least in the sense of a well-ordering of the OD S sets, cf. the proof of Theorem 5.45) I.e., if G(A x ) is the usual game with payoff
A x , as defined in L[S, x], then G A x is not determined in L[S, x]. Notice that A x
only depends on the S-constructibility degree of x, i.e., if L[S, x] = L[S, y], then
Ax = A y .
Let G be the game in which I , II alternate playing natural numbers so that if
n2
...
I n0
II
n1
n3
...
is a play of G, then I wins iff, setting x = (n 4i : i < ), a = (n 4i+2 : i < ),
y = (n 4i+1 : i < ), and b = (n 4i+3 : i < ) (which we shall also refer to by saying
that I produces the reals x, a and II produces the reals y, b), then
a b A xy .
Let us suppose that I has a winning strategy, , in G. Let L[S, z], where z is in
C. Let be a strategy for I in G Az played in L[S, z] so that if II produces the real b,
and if calls for I to produce the reals a, x in a play of G in which II plays b, z b,
then calls for I to produce the real a. Then for every b L[S, z], if a = b,
in fact if a, x = (b, z b), then
a b A x(zb) = A z .
So is a winning strategy for I in the game G(A z ) played in L[S, z]. Contradiction!
We may argue similarily if II has a winning strategy in G.
292
<s is a well-ordering
sx
293
where (i) = i for every i k (and < is the natural order on ordinals). The first
player to disobey one of the rules loses. If the play is infinite, then II wins.
Notice that what II has to do is playing a witness to the fact that <x is a well-order,
where x = (n 0 , n 1 , . . .).
Notice also that G (A) is an open game in the space 1 [(which we
identify with ( 1 )] and hence by Theorem 12.3, G (A) is determined in every
inner model which contains s <s .
Fix a real x such that the map s <s is in L[x]. E.g., let x be such that B is
11 (x).
Let us first assume that II has a winning strategy for G (A) in L[x], call it .
Obviously, L[x] is then also a winning strategy for G (A) for all plays in V
(not only the ones in L[x]). But then II will win G(B) in V by just following and
hiding her side moves 0 , 1 , . . .. If x = (n 0 , n 1 , . . .) is the real produced at the
end of a play, then
({0 , . . . , k }, <)
= ({0 , . . . , k }, <),
where (i ) = i for every i k, then
L[x] |= (, 0 , . . . , k ) L[x] |= (, 0 , . . . , k )
for every L -formula , cf. Corollary 10.44 (2). In particular, then,
(n 0 , n 1 , 0 , . . . , n 2k , n 2k+1 , k ) = (n 0 , n 1 , 0 , . . . , n 2k , n 2k+1 , k )
for all integers n 0 , n 1 , . . . , n 2k+1 . We may therefore define a strategy for I in G(B)
as follows. Let
(n 0 , n 1 , . . . , n 2k , n 2k+1 ) = (n 0 , n 1 , 0 , . . . , n 2k , n 2k+1 , k )
where 0 , . . . , k are countable x-indiscernibles with
k + 1, <(n 0 ,...,n k )
= ({0 , . . . , k }, <),
(i) = i for i k. We claim that is a winning strategy for I in G(B).
Let us assume that this is not the case, so that there is a play of G(B) in which I
follows and which produces x = (n 0 , n 1 , . . .) A. Then <x is a well-order and
294
(, <x )
= ({0 , 1 , . . .}, <),
where (i) = i for i < . I.e.,
(k+1)
k + 1, <(n 0 ,...,n k )
= ({0 , . . . , k }, <)
for all k < , and this means that for every k < ,
n 2k = (n 0 , n 1 , 0 , . . . , n 2k2 , n 2k1 , k1 ),
that is, n 0 , n 1 , 0 , n 2 , n 3 , 1 , . . . is a play of G (A) in which I follows .
Let us now define the tree T of attempts to find an infinite play of G (A) in which I
follows as follows. We set s T iff s = (n 0 , n 1 , 0 , . . . , n 2k2 , n 2k1 , k1 , n 2k )
for some n 0 , n 1 , . . . , n 2k and 0 , . . . , k1 1 such that for all l k,
n 2l = (n 0 , n 1 , 0 , . . . , n 2l2 , n 2l1 , l1 ).
If s, t T, then we let s t iff s t. Notice that (T ; ) L[x].
Now (T ; ) is ill-founded in V by what was shown above. Hence (T ; ) is illfounded in L[x] as well by the absoluteness of well-foundedness, cf. Lemma 5.6.
Therefore, in L[x] there is a play of G (A) in which I follows and loses. But
there cannot be such a play in L[x], as is a winning strategy for I in G (A).
Contradiction!
295
Any real code y for x # satisfies the hypothesis of Lemma 12.21, cf. Problem
12.11.
There is a proof of Lemma 12.21 which avoids the use of Corollary 11.60 and
which just makes use of the argument for Lemma 10.29. Cf. problem 12.9.
Theorem 12.22 (L. Harrington) If analytical determinacy holds then for every x
x # exists. In fact, if x and every 11 (x) set B is determined, then x #
exists.
,
Proof Forcing with Col(, ) adds reals coding ordinals below + 1. There is a
forcing which adds such reals more directly, namely Steel forcing which we shall
denote by TCol(, ).
Let be an infinite ordinal. We let TCol(, ) consist of all (t, h) such that t is a
finite tree on , i.e., t is a non-empty finite subset of < such that s t and n lh(s)
implies s n t, and h is a ranking of t in the following sense: h: t {}
is such that h() = , and if s t, n < lh(s), and h(s n) , then h(s) and
h(s) < h(s n). For (t, h), (t , h ) TCol(, ), we let (t , h ) (t, h) iff t t
and h h.
Let G be TCol(, )-generic over V , and set
T = {t: h(t, h) G} and
H = {h: t (t, h) G}.
(12.1)
296
g (s) = + k , where k > k , so g (s) < g (s n).
We shall now be interested in forcing with TCol(, ) over (initial segments of)
L[x], where x is a real. If G is TCol(, )-generic over L[x] and T and H are
defined from G as in (12.1), then truth about initial segments of L[x][T ] can be
decided by the right restrictions (t, h)| of elements (t, h) from G. In order to
formulate this precisely, we need to rank sentences expressing truths about initial
segments of L[x][T ] as follows.
Recall (cf. p. 70) that the rud x,T functions are simple in the sense that if
(v0 , . . . , vk1 ) is a 0 -formula (in the language for L[x, T ]) and f 0 , . . . , f k1 are
rud x,T functions, then there is a 0 -formula (again in the language for L[x, T ])
such that
( f 0 (x0 ), . . . , f k1 (xk1 )) (x0 , . . . , xk1 )
holds true over all transitive rud x,T -closed models which contain x0 , . . . , xk1 . In
particular, we may associate to each pair f, g of r udx,T functions a 0 -formula and
hence a -formula such that for all limit ordinals and for all x, y J [x, T ],
f (J [x, T ], x) g(J [x, T ], y) J+ [x, T ] |= (J [x, T ], x, y)
J [x, T ] |= (x, y).
We shall write ( f, g) for in what follows. The choice of and ( f, g)
can in fact be made uniformly in x, T .
Let us now pretend that the language for L[x, T ] has function symbols for rud x,T
functions available; we shall in fact confuse a given rud x,T function f with the
function symbol denoting it. We then define terms of rank recursively as follows.
A term of rank is an expression of the form
f (J [x, T ], y),
where f is (the function symbol for) a rud x,T function, y is a vector of terms of rank
< , and J [x, T ] stands for the term denoting J [x, T ]. Inductively, every element
of J+ [x, T ] is thus denoted by a term of rank .
297
(t, h) L[x]
J [x, T ] |= (1 , . . . , k )
TCol(,+(+2n))
J [x, T ] |= (1 , . . . , k ).
(t, h)| + ( + 2n) L[x]
Proof The proof is, of course, by induction on + n.
Let us first assume n = 0, i.e. that is an atomic formula. Let us assume that
1 f (J [x, T ], y) and 2 g(J [x, T ], z), where < , and that (1 , 2 )
f (J [x, T ], y) g(J [x, T ], z). Then for + and (t, h) TCol(, ),
TCol(, )
(t, h) L[x]
J [x, T ] |= (1 , 2 )
TCol(, )
(t, h) L[x]
J+ [x, T ] |= (1 , 2 )
TCol(, )
(t, h) L[x]
J [x, T ] |= ( f, g)( y, z).
Therefore, the desired statement easily follows from the inductive hypothesis.
Now let n > 0. Let us assume that v0 . The cases and 1 2
are similar and easier.
Let us assume that
TCol(,)
(t, h) L[x]
J [x, T ] |= v0 (v0 , 1 , . . . , k ).
TCol(,)
J [x, T ] |= (0 , 1 , . . . , k )
TCol(,+(+2n))
J [x, T ] |= (0 , 1 , . . . , k ).
We have shown that the set of (t, h) (t, h)| + ( + 2n) in TCol(, + ( +
2n) ) such that
TCol(,+(+2n))
(t, h) L[x]
J [x, T ] |= (1 , . . . , k ).
298
J [x, T ] |= (1 , . . . , k ).
< .
Let us consider the
But then II can easily defeat by playing a code for a transitive model of ZFC +
V = L[x] which contains .
By Claim 12.25 and 11 (x)-determinacy, we may let be a winning strategy for
player II in G. By Lemma 12.21, the following will produce x # .
2
299
Claim 12.26 Let be a countable ordinal such that J [x, ] |= ZFC . Then is
a cardinal of L[x]. In fact, if is countable and x -admissible (cf. p. 88), then
is a cardinal of L[x].
Proof Let < be an infinite cardinal of L[x]. It suffices to verify that if b
, b L[x], then b J [x, ] (cf. Problem 12.10). So let us fix some such b, and
let b J [x], where without loss of generality 1 > . Let > be such that
J [x, ] |= ZFC and is countable. Let G be TCol(, +1)-generic over J [x, ],
and let T and H be given by G as in (12.1). Obviously, there is a real z 0 in J [x, , T ]
(even in J+ [x, , T ]) which codes a well-founded tree S such that ||S|| = . E.g.,
let S = T s, where s T with H (s) = . Therefore, z 0 J [x, , T ] (in
fact J+ [x, , T ]) codes a model (; E) of ZFC + V = L[x] such that is
contained in the transitive collapse of the well-founded part of (; E). Let
: (J [x]; )
= wfp(; E)
be the transitive collapse of the well-founded part wfp(; E) of (; E), so that .
By E J+ [x, , T ], it is easy to verify inductively that J [x] is uniformly
J [x,,T ]
1
J [x] J+ [x, , T ].
(12.2)
(12.3)
By (12.2), there is some formula and terms 0 and 1 for E and J [x],
respectively, where 0 and 1 are of rank and such that
< ( b J+ [x, , T ] |= (, 0 , 1 )).
(12.4)
(t, h) L[x, ]
< (
b J+
[x, , T ] |= (, 0 , 1 )),
b (t, h) L[x, ]
J+
[x, , T ] |= (, 0 , 1 ).
(12.5)
J+
[x, , T ] |= (, 0 , 1 )
(12.6)
300
But we may replace L[x, ] by J [x, ] here (we could in fact replace it by
J+2 [x, ]), so that we may therefore define b over J [x, ] as follows:
TCol(,)
J+ [x, , T ] |= ( , 0 , 1 )}.
Therefore b J [ ] as desired.
We may now add the following to our list of equivalences to x # exists, cf.
Theorem 11.56.
Corollary 12.27 The following statements are equivalent.
(1) Every analytic A is determined.
(2) For every x , x # exists.
The paper [8] gives information on the Axiom of Determinacy. Cf. also [25].
12.4 Problems
12.1 Assume AD. Show that if (xi : i < ) is a sequence of pairwise different reals,
then < 1 [Hint. Let U witness that 1 is measurable, cf. Theorem 12.18
(b). Consider L[U, (xi : i < )], cf. proof of Problem 10.3 (a)].
12.2 Show (in ZF) that there is some A (1 ) which is not determined [Hint. If
AD holds, then ask for II to play some x with ||x|| = in response to
I playing < 1 ].
12.3. Assume AD. Show that for every set A, OD{A} is countable. Fixing A,
show that there is no f : such that f (x) \OD{x,A} for all
x and f OD {A} . Conclude that HOD {A} |= AD and there
is some (A x : x ) with = A x for all x with no choice
function.
12.4. Assume ZF plus x # exists for every real x. Show that there is some f :
such that f (x) \L[x] for all x and f OD. (In fact, we
may pick f to be 31 , cf. Problem 10.7.) Show also that there is a function
f : HC such that for every x , f (x) is a C-generic filter over L[x]
(Hint. Use x # to enumerate the dense sets of L[x]).
12.5. For A, B , write A Wadge B iff there is some continuous f :
such that for all x , x A f (x) B, or for all x ,
x A f (x)
/ B. Assume AD.
(a) (Wadge) Show for all A, B , A Wadge B or B Wadge A. [Hint.
Let G Wadge (A, B) be the game so that if I plays x and II plays y, then I
wins iff x B y A]. Show that Wadge is reflexive and transitive,
so that A Wadge B iff A Wadge B B Wadge A is an equivalence
relation. Show that Wadge is not symmetric.
12.4 Problems
301
(b) (Martin, Monk) Show that Wadge is well-founded [Hint. Otherwise there
are An , n < , such that for all n < , I has winning strategies n0
and n1 for G Wadge (An+1 , An ) and G Wadge ( \An+1 , An ), respectively.
z(n)
z
for all
For z 2, we get (xnz : n < ) such that xnz = n xn+1
n < . Let z, z 2, and n < be such that for all m, z(m) = z (m)
z
z
iff m = n. Then xn+1
= xn+1
, and xnz An xnz
/ An , . . .,
x0z A0 x0z
/ A0 . Hence {z 2: x0z A0 } is a flip set, cf.
Problem 8.3].
(c) Show that for all A there is some J (A) with A <Wadge J (A)
[Hint. For x write f x for the canonical continuous function given
/ B}].
by x. Let x B iff f x (x) A, and set J (A) = {x y: x B and y
(d) Let = sup({: surjective f : }). Show that || <Wadge || =
[Hint. To show that || <Wadge || , let f : be surjective, and let
(A : < ) be such that if < , then A = J ({x y: f (x) < y
A f (x) }), where J is as in (c)]. Let X = . If A ( ), then in a run
of the game G(A) players I and II alternate playing real numbers, i.e.,
elements of . The Axiom of Real Determinacy, abbreviated by ADR ,
states that G(A) is determined for every A ( ).
302
defined for all i < , then set = supi< i . We call the length of the
Solovay sequence.
12.7. Assume AD.
(a) Show that if A is such that ||A||<Wadge < 0 , then A OD [Hint.
Writing = ||A||<Wadge , pick an ODR surjection f : + 1, and let
(A : ) be as in Problem 12.5 (d). Then ||A ||<Wadge and A
OD , which yields A OD ]. Conclude that if = 0 , then P( )
HOD .
(b) Show also if A is such that ||A||<Wadge < i+1 , then A OD {B}
for some (all) B with ||B||<Wadge = i . Conclude that if the length of
the Solovay sequence is a successor ordinal, then there is some B
with P( ) HOD {B} .
12.8. Show that if ADR holds, then the length of the Solovay sequence is a limit
ordinal and there is no B with P( ) HOD {B} . Conclude that
AD does not imply ADR [Hint. Use Problems 12.3 and 12.7 (b)].
12.9. Show Lemma 12.21 by using the argument for Lemma 10.29.
12.10. (a) Show that there is a transitive model M of ZFC with (M L P()) \
J = , where = M OR, so that J = L M (Hint: Let be countable
in L such that J |= ZFC and pick G L which is C-generic over J ).
(b) Show that if M is admissible with = M OR and P() L M for
every < , then is a cardinal in L [Hint: Let < +L , let f : J
be bijective, f L, and let n Em iff f (n) f (m). Then E M, and
hence J M by Problem 5.28].
12.11. Suppose that 0# exists, and let x be a real code for 0# . Let be an infinite
L-cardinal. Show that if > is x-admissible, then P() L J [x]
(Hint. The th iterate of 0# exists in J [x], cf. Problem 5.28). Conclude that
every x-admissible is a cardinal of L. In fact, every x-admissible is a Silver
indiscernible.
12.12. (A. Mathias) Let M be a transitive model of ZFC such that M |= U is a
selective ultrafilter on . Let M be (MU ) M , i.e., Mathias forcing for U , as
being defined in M, cf. p. 176. Let x [] be such that x\X is finite for
every X U , and let
G = {(s, X ) M: n < s = x n}
(cf. Definition 10.10.). Show that G is M-generic over M (This is the converse
to Problem Problem 9.9.) [Hint. Use Problem 9.3 (b) and the proof of Theorem
10.11, cf. p. 274 ff].
12.13. (A. Mathias) Show that in the model of Theorem 8.30, every uncountable
A [] is Ramsey, cf. Definition 8.17 (Hint. Imitate the proof of Lemma
8.17, replacing C by Mathias forcing for some selective ultrafilter on , cf.
Problem 9.4. Then use Problem 12.12).
Chapter 13
Projective Determinacy
We shall now use large cardinals to prove stronger forms of determinacy. We shall
always reduce the determinacy of a complicated game in to the determinacy
of a simple (open or closed) game in a more complicated space as in the proof of
Martins Theorem 12.20.
< )
303
304
13 Projective Determinacy
We first construct a natural tree T , which we call the Windus tree for A, such
/ A, so
that A = p[T ] as follows. We first define a sequence (s : s < ). If x
that Mx is illfounded, we pick a sequence (xn : n < ) of ordinals witnessing that
Mx is illfounded in the sense that
x n,x (n+1) (xn ) > xn+1
(13.1)
for all n < . If x A, then we let xn = 0 for all n < . We then let, for s < ,
s : OR be defined by
s (x) =
lh(s)
x
0
if x A
/ s = x lh(s)
otherwise.
((
),
305
be a position in G(A), so that its IIs turn to play. Notice that, because each Ms is
20 closed, s Ms for every s < . Moreover, setting s = (n 0 , . . . , n k ),
((n 0 , . . . , n k ), (,s k ( ), . . . , s k1,s k (s k1 ), s k )) ,s k (T )
(13.2)
sx
<s is a wellordering.
(13.3)
306
13 Projective Determinacy
s,t = lh(s)lh(t) ,
(s,t)
where lh(s)lh(t) is the shift map given by (s, t), cf. the Shift Lemma 10.4. For
x , let
(Mx , (s,x : s x)) = dir lim(Ms , s,t : s t x).
Notice that {sx (n ): n < , s x, lh(s) = n + 1}, ordered by the relation of
Mx , is always isomorphic to , ordered by <x . By (13.3), this readily implies that if
x
/ A, i.e., if <x is illfounded, then Mx is illfounded.
But it also implies that if x A, i.e., <x is wellfounded, then Mx is wellfounded
as follows. Let x A and let = otp(<x ). We may define maps (s, x): lh(s)
by setting
(s, x)(n) = ||n||<x
for n < lh(s), s x. Notice that (s, x) = (t, x) lh(s) whenever s t x. By
Lemma 10.4, we have, for s x,
(s,x)
lh(s), : Ms M ,
where for s t x,
(t,x)
(s,t)
(s,x)
307
there are infinitely many Woodin cardinals, then Projective Determinacy holds.
The proofs of Theorems 12.11, 12.13, and 12.16 also give the following.
308
13 Projective Determinacy
Corollary 13.8 Suppose that there are infinitely many Woodin cardinals. If A is
projective set of reals, then A is Lebesgue measurable and has the Baire property,
and if A is uncountable, then A has a perfect subset.
In particular, the collection of all projective sets of reals has the perfect subset property (cf. p. 142), i.e., here are no definable counterexamples to Cantors program,
cf. 3.
Proof of Theorem 13.6. Let us fix
(Ns , s,t : s t < ),
a 20 closed embedding normal form for B whose additivity is bigger than .
If s, t < are such that lh(s) = lh(t), then we define s t to be that r <
such that lh(r ) = 2 lh(s), and for all n < lh(s), r (2n) = s(n) and r (2n + 1) = t (n).
Let us write
A = {(x, y) ( )2 : x y B},
so that trivially {x : y x y
/ B} = {x : y : (x, y)
/ A}.
Let us also write Ns,t for Nst , and (s,t),(s ,t ) for st,s t , where s, s , t, t <
with lh(s) = lh(t) < lh(s ) = lh(t ). Then
(Ns,t , (s,t),(s ,t ) : s s < , t t < , lh(s) = lh(t), lh(s ) = lh(t ))
(13.4)
is a 20 closed embedding normal form for A whose additivity is bigger than in
the sense that for all x, y ,
(x, y) A dir lim(Ns,t , (s,t),(s ,t ) : s t x, t t y) is wellfounded,
(13.5)
every Ns,t is 20 closed, and (s,t),(s ,t ) ( + 1) = id for all relevant (s,t),(s ,t ) .
Let us first construct the Windus tree T for A in much the same way as in
the proof of Theorem 13.2, as follows. We start by defining a sequence (s,t : s, t
< , lh(s) = lh(t)). If (x, y)
/ A, so that
dir lim(Ns,t , (s,t),(s ,t ) : s s x, t t y)
n : n < ) of ordinals witnessing that this
is illfounded, we pick a sequence (x,y
direct limit is illfounded, i.e.,
n
n+1
) > x,y
(x n,y n),(x (n+1),y (n+1)) (x,y
n = 0 for all n < . Let, for s, t
for all n < . If (x, y) A, we let x,y
(13.6)
< ,
s,t (x, y) =
lh(s)
x,y
0
309
if (x, y)
/ A s = x lh(s) t = y lh(s)
otherwise
(13.7)
(13.8)
310
13 Projective Determinacy
Ts
s,k
s,k
s,k
for k 2 lh(s),
for k lh(s), and
for k lh(s).
(13.9)
We will arrange that the following statements (PD, 1) through (PD, 4) hold true.
(PD, 1) Each Ts is an iteration tree on V of length 2 lh(s) + 1,
Ts = ((Ms,k , s,k,l : k l 2 lh(s)), (E s,k : k < 2 lh(s)), (2 lh(s) + 1)),
Moreover, s,0 < s,1 < . . . < s,2lh(s) , and for all k < l 2 lh(s),
(V ) Ms,k = (V ) Ms,l ,
where is the least inaccessible cardinal of Ms,k which is bigger than
s,k+1 .
Es,2n+1
even branch
odd branches
2n + 2
sn+1 2n + 1
Es,2n
k = 2n
sm 2m 1
311
(s,l ) : 2l 1
(T ).
1
The forth condition which will guarantee that the even branches of Ts , s x,
where x , produce an embedding normal form. In order to state it, we need a
notation for the models and embeddings of Ts from the point of view of Ns ,t , the
models from (13.4). Let s < , and let s , t < , lh(s ) = lh(t ). Because for
M
all k < 2 lh(s), E s,k V s,k , and because the additivity of our given embedding
normal form (13.4) for A is bigger than , the sequence
(E s,k : k < 2 lh(s))
is easily seen to generate2 (Ts ) Ns ,t such that
Ns ,t |= (Ts ) Ns ,t is an iteration tree on Ns ,t of length 2 lh(s) + 1,
where we may write
s ,t
s ,t
(Ts ) Ns ,t = ((Ms,k
, s,k,l
: k l 2 lh(s)), (E s,k : k < 2 lh(s)),
(2 lh(s) + 1)).
s ,t
= (,),(s ,t ) (Ms,k ),
Ms,k
s ,t
s ,t
Ms,k
Ms,k
= V
s ,t
for all k l 2 lh(s).3 The models Ms,k
are the models Ms,k from the point of
s ,t
view of Ns ,t , and the embeddings s,k,l
are the embeddings s,k,l from the point
of view of Ns ,t .
s ,t
s ,t
s ,t
(s ,t ),(s ,t ) s,k,l
= s,k,l
((s ,t ),(s ,t ) Ms,k
),
(13.10)
There is a slight abuse of notation here, as Ts is not a set but rather a (sequence of) proper class(es).
By (Ts ) Ns ,t we mean that object which is defined over Ns ,t from the parameter (E s,k : k < 2lh(s))
by the very same formula which defines Ts over V from the same parameter (E s,k : k < 2 2lh(s)).
312
13 Projective Determinacy
s ,t
s ,t
where this is an embedding from Ms,k
to Ms,l
whenever s, s , s , t , t
< , k l 2 lh(s), lh(s ) = lh(t ) lh(s ) = lh(t ). Equation (13.10)
s ,t
holds true because for all x Ms,k
Ns ,t ,
s ,t
(x) = (s ,t ),(s ,t ) ((,),(s ,t ) (s,k,l )(x))
(s ,t ),(s ,t ) s,k,l
= (s ,t ),(s ,t ) ((,),(s ,t ) (s,k,l ))((s ,t ),(s ,t ) (x))
s ,t
(s ,t ),(s ,t ) (x).
= s,k,l
),sn+1
n+1
s,n+1 < s,2m,2n+2
Suppose that we manage producing objects Ts , s,k , s,k , and s,k with the properties (PD, 1) through (PD, 4). We may then verify the following
Claim 13.9
((Ms,2lh(s) : s < ), (s ,2lh(s),2lh(s ) : s s , s, s < ))
is a 20 closed embedding normal form for {x : y (x, y)
/ A}.
Proof 20 closedness is clear by (PD, 1), as every extender used is 20 closed in
the model where it is taken from. Cf. Lemmas 10.60 and 10.61.
Let us fix x . Let
(M Even , (2k, Even : k < )) = dir lim((Mx k,2k : k < ), (x l,2k,2l : k l < )).
We need to see that
y(x, y) A = M Even is ill founded
(13.11)
y(x, y)
/ A = M Even is wellfounded.
(13.12)
and
Let us first show (13.11). If s, t < with lh(s) = lh(t), then by the coherence
s,t
for Mxs,tk,k (= Mxs,tm,k for all sufficiently large
condition (PD, 2), we may write Mx,k
s,t
for xs,tl,k,l (= xs,tm,k,l for all sufficiently large m).
m) and x,k,l
Let us now pick some y such that (x, y) A. Let
313
x,y
x,y
x,y
By the coherence condition (PD, 2) we may write Mx,k for Mx k,k (= Mx m,k
x,y
x,y
x,y
for all sufficiently large m), x,k,l for x l,k,l (= x m,k,l for all sufficiently large
m), and
x,y
x,y
x,y
x,y
(13.14)
x,y
x k,y k
(13.15)
x,2n+2,
x (k+1),y (k+1)
314
13 Projective Determinacy
witnesses that Mx, is illfounded. However, the direct limit (13.14) which prox,y
duces Mx, can be formed within N , cf. (13.13), which is transitive. We may therefore use absoluteness of wellfoundedness, Lemma 5.6, to see that such a sequence
x,y
witnessing that Mx, be illfouded must also be an element of N . By the elementarity of (,), : V N , which maps (Mx,2m , x,2m,2k : m k < ) to
x,y
x,y
(Mx,2m , x,2m,2k : m k < ), there is hence a sequence which witnesses that
M Even = dir lim(Mx,2m , x,2m,2k : m k < )
is illfounded. We have verified (13.11).
Let us now show (13.12). For this, Theorem 10.74 is the key tool. Let us thus
/ A. As every extender used is certified in the
suppose that for all y , (x, y)
model where it is taken from, in order to verify that M Even is wellfounded, it suffices
by Theorem 10.74 to show that for all y , if
(M y , (k,y : k < )) = dir lim(Mx k,2k 1
, x l,2k 1,2l
: sk sl y),
1
then M y is illfounded.
/ p[T ],
To this end, let y be arbitrary. As (x, y)
Tx,y = { f : (x lh( f ), y lh( f ), f ) T }
is wellfounded. By (PD, 3), if k < and sn = y k, then
2n 1))
(x k, y k, (x n,2l 1,2n
x n,0,2n 1
(x n,l ): 2l 1
(T ).
1
Let (n i : i < ) be the monotone enumeration of {n < : sn y}, and let, for i < ,
2n i 1)||
xn ,0,2n 1
(Tx,y ) .
i 1
i
2n i+1 1)
extends
If i < , then the node (x n i+1 ,2l 1,2n
(x n i+1 ,l ) : 2l 1
i+1 1
(
)
:
2l
1
2n
1)
in
the
tree
i
x
n
,l
1
x
n
i+1
i+1
i+1 ,0,2n i+1 1
(Tx,y ), so that
x n i+1 ,2n i 1,2n
(i )
i+1 1
2n i 1)||
xni+1 ,0,2ni+1 1
(Tx,y )
i+1 1
(Tx,y )
i+1 1
i+1
i+1
= i+1 .
This proves that M y is illfounded, as witnessed by (i : i < ).
315
We are left with having to produce the objects Ts , s,k , s,k , and s,k such that
(PD, 1) through (PD, 4) hold true.
We first need a set of indiscernibles. Let us fix , a cardinal which is much
s ,t
(s,t ) will be in
bigger than ; in particular, we want that T V and that all 0,k
V also. Let c0 < c1 < be strong limit cardinals above of cofinality > such
that
(13.17)
type V (V ; , V , {, c0 })) = type V (V ; , V , {, c1 })).
An easy pigeonhole argument shows that such objects exist: e.g., let be a strong
+
limit cardinal of cofinality V+1 . In the construction to follow, we shall use the
descending chain of ordinals
c0 + 1 > c0 c1 > c0 + 1 > c0 c1 > . . .
and we may and shall assume that , c0 , c1 , and are fixed points of all the elementary
embeddings which we will encounter. Cf. Lemma 10.56 Problem 10.18. will always
be a fixed point anyway.
In order to keep our recursion going, we shall need a fifth condition.
M2l 1
The objects Ts , s,k , s,k , and s,k will be constructed by recursion on the length
of s. To get started, we let T be the trivial tree of length 1 which just consists of V .
We also set ,0 = c0 , and we pick ,0 < , ,0 > , such that in V , ,0 is strong
up to with respect to
type V (Vc0 +1 ; , V , {, , T, , }).
This choice of ,0 is certainly possible, as is a Woodin cardinal, cf. Lemma 10.81.
We also set ,0 = , .
Now let us fix s < of positive length throughout the rest of this proof. Write
n + 1 = lh(s). Let us suppose that
Ts n
s,k for k 2 n,
s,k for k n, and
s,k for k n
have already been constructed. We are forced to set s,k = s n,k for k 2 n,
E s,k = E s n,k for k < 2 n, Ms,k = Ms n,k for k 2 n, s,k,l = s n,k,l for
k l 2 n, s,k = s n,k for k n, and s,k = s n,k for k n. Let us write
t = sn+1 , k = lh(t) n + 1, and sm = t (k 1). We now need to define s,2n+1 ,
s,2n+2 , s,n+1 , and s,n+1 , and we also need to define Ms,2n+1 as an ultrapower of
316
13 Projective Determinacy
Ms,2m 1
by an extender E s,2n with critical point s,2m and Ms,2n+2 as an ultrapower
of Ms,2n by an extender E s,2n+1 with critical point s,2n+1 .
2n + 2
t = sn+1 2n + 1
Es,2n+1
Es,2n
t k 1 = sm 2m 1
k = 2n
As s
and write
<
0
will be fixed from now on, we shall mostly suppress the subscript s
Mk
for Ms,k ,
k,l for s,k,l ,
s ,t
,
Mks ,t for Ms,k
s ,t
k,l
k
k
k
s ,t
for s,k,l
,
for s,k ,
for s,k , and
for s,k .
Inductively, we shall assume that the following two statements, (A) and (B), are
satisfied. Here, s i,t i is as in (13.7).
(A)
sk1,tk1
s k1,t k1
1 (i ): 2i 1 2m 1)}).
s k1,t k1
sk1,tk1
(Vm +1 ; , V , {, , 0,2m
((,),(s k1,t k1) (T )),
s k1,t k1
((s i,t i),(s k1,t k1) (s i,t i )): i k 1)}).
(0,2m
type M2m
Notice that this is trivially true for n = 0 (in which case m = k 1 = 0) by the
choices of 0 = ,0 = c0 , 0 = ,0 , and 0 = , .
s k,t
Because is a Woodin cardinal inside M2n , we may pick some 2n+1 < ,
2n+1 > 2n , such that
s k,t
317
sk,t
; , V ,
2m,2n ((sk1,tk1),(sk,t) (m ))
s k,t
s k,t
{, , 0,2n ((,),(s k,t) (T )), (0,2n ((s i,t i),(s k,t) (s i,t i ): i
k)}),
s k,t
cf. Lemma 10.81. We may apply the map 2m,2n (s k1,t k1),(s k,t) to (B), which
by (13.10) and the fact that crit(2m,2n ) = 2m+1 > 2m produces the assertion that
s k,t
(sk1,tk1),(sk,t) (m ))+1
2m,2n
s k,t
; , V ,
(13.18)
s k,t
{, , 0,2n ((,),(s k,t) (T )), (0,2n ((s i,t i),(s k,t) (s i,t i )): i < k)}).
We may thus let
.
.
: M2m 1
E s,2n M2n+1 ,
2m 1,2n+1
where E s,2n M2n is a 20 closed certified extender in M2n which witnesses that
s k,t
inside M2n , 2m is strong up to the least inaccessible cardinal above 2n+1 with
.
respect to the type from (13.18); notice that (V2m +1 ) M2n = (V2m +1 ) M2m 1 , cf. (PD,
1), immediately gives that
(V ) M2n+1 = (V ) M2n .
(13.19)
We have that
type M2n+1 (Vc0 +1 ; , V2n+1 +1 , {, ,
.
.
0,2n+1 (T ), (2i 1,2m
1 (i ): 2i 1 2m 1) )
sk,t
(sk1,tk1),(sk,t) (m ))+1
s k,t
(13.20)
; , V2n+1 +1 , {, ,
s k,t
0,2n ((,),(s k,t) (T )), (0,2n ((s i,t i),(s k,t) (s i,t i )): i k 1)}).
This is because by the choice of E s,2n , the right hand side of (13.20) is equal to
sk,t
.
(type M2n (V sk,tk (
2m 1,2n+1
2m,2n
s k,t
(sk1,tk1),(sk,t) (m )+1
; , V2m , {, ,
s k,t
0,2n ((,),(s k,t) (T )), (0,2n ((s i,t i),(s k,t) (s i,t i )): i k 1)})),
restricted to parameters in (V2n+1 +1 ) M2n+1 = (V2n+1 +1 ) M2n , which by (13.10) and
s k1,t k1
is equal to
the elementarity of the map (s k1,t k1),(s k,t) 2m,2n
318
13 Projective Determinacy
sk1,tk1
.
2m 1,2n+1
(type M2m
(Vm +1 ; , V2m , {, ,
s k1,t k1
0,2m
((,),(s k1,t k1) (T )),
s k1,t k1
((s i,t i),(s k1,t k1) (s i,t i )): i
(0,2m
k 1)})),
.
2m 1,2n+1
(type M2m 1 (Vc0 +1 ; , V2m , {, ,
. (T ), (
.
0,2m 1
2i 1,2m
1 (i ): 2i 1 2m 1)})),
restricted to parameters in (V2n+1 +1 ) M2n+1 , and thus to the left hand side of (13.20).
Let us write
sk,t
(sk1,tk1),(sk,t) (m )
; , V2n+1 , {, ,
s k,t
(13.21)
s k,t
(0,2n ((s i,t i),(s k,t) (s i,t i )): i
k)}).
s k,t
M2n
sk,t
(sk1,tk1),(sk,t) (m )
; , V2n+1 ,
s k,t
sk,t
M2n
(V sk,tk (
2m,2n
(sk1,tk1),(sk,t) (m )
(13.22)
; , V ,
s k,t
sk,t
(V2n+1 +1 ) M2n ,
the statement V (. . .) in (13.22) can be written as an element of the type
from the right hand side of (13.20), so that by (13.20),
319
(i ): 2i 1 2m 1) })
{, , 0,2n+1 (T ), (2i 1,2n+1
(13.23)
2n+1 is strong up to with respect to
type M2n+1 (Vc0 ; , V ,
(i ): 2i 1 2m 1) }) ).
{, , 0,2n+1 (T ), (2i 1,2n+1
M2n1
Let n+1 V
(D)
s k,t
s k,t
s k,t
(i ): 2i 1 2n + 1)})
(2i 1,2n+1
and
(C)
M2n+1
M2n+1
.
(i ): 2i 1 2n + 1) V
(2i 1,2n+1
(i.e., (PD, 5)
(13.24)
Now again because is a Woodin cardinal inside M2n+1 , we may pick some
2n+2 < , 2n+2 > 2n+1 , such that
(E) inside M2n+1 , 2n+2 is strong up to with respect to
.
.
type M2n+1 (Vc0 +1 ; , V , {, , 0,2n+1 (T ), (2i 1,2n+1
(i ): 2i 1 2n + 1)}),
.
(i ): 2i 1 2n + 1)}).
type M2n+1 (Vc0 ; , V , {, , 0,2n+1 (T ), (2i 1,2n+1
320
13 Projective Determinacy
This choice is possible by (C); notice that (V2n+1 +1 ) M2n+1 = (V2n+1 +1 ) M2n , cf.
(13.19), which immediately gives that (V ) M2n+2 = (V ) M2n+1 . We shall have that
(F)
sk,t
2m,2n+2 (sk1,tk1),(sk,t) (m )
; , V2n+2 +1 ,
s k,t
0,2n+2 ((s i,t i),(s k,t) ((s i,t i)) : i
= type M2n+1 (Vc1 ; , V2n+2 +1 ,
k)})
{, , 0,2n+1 (T ),
.
.
(i ): 2i 1 2n + 1)}).
(2i 1,2n+1
sk,t
2m,2n (sk1,tk1),(sk,t) (m )
s k,t
; , V2n+1 ,
sk,t
.
= type M2n+1 (Vc0 ; , V2n+2 +1 , {, , 0,2n+1 (T ), (2i 1,2n+1
(i ): 2i 1 2n + 1)}),
by the choice of E s,2n+1 , which is equal to the right hand side of (F) by (13.24) and
the choice of c0 and c1 , cf. (13.17). We verified (F).
Let us write
.
.
= type M2n+1 (Vc0 +1 ; , V2n+2 , {, , 0,2n+1 (T ), (2i 1,2n+1
(i ): 2i 1 2n + 1)}).
With (E),
(Vc1 ) M2n+1 |= ( = type M2n+1 (V+1 ; , V2n+2 , {, , 0,2n+1 (T ),
.
.
(i ): 2i 1 2n + 1)})
(2i 1,2n+1
M2n+1
(V+1 ; , V , {, , 0,2n+1 (T ),
.
.
(i ): 2i 1 2n + 1)}). ).
(2i 1,2n+1
(13.25)
321
We have that
(V2n+2 +1 ) M2n+1 ,
so that the statement (. . .) in (13.25) can be written as an element of the type
from the right hand side of (F), and (F) yields the following.
sk,t
(V sk,t
2m,2n+2 ((sk1,tk1),(sk,t) (m ))
) M2n+2 |=
s k,t
sk,t
sk,t
M2n+2
(13.26)
s k,t
s k,t
(13.27)
s k,t
s k,t
322
13 Projective Determinacy
Results which are stronger than Theorem 13.6 but build upon its proof method
are presented e.g. in [32, 33] and [41].
The reader might also want to consult [34] and [42] on recent developments
concerning determinacy hypotheses and large cardinals.
13.3 Problems
13.1. Show that if there is a measurable cardinal, then every set of reals has an embedding normal form. [Hint. The embedding normal form will not be 20 closed.]
Let n m < , let be an ultrafilter on a set of functions with domain n, and
let be an ultrafilter on a set of functions with domain m. We say that ,
cohere iff for all X ,
X { f m : f n X } .
We may define : V ult 0 (V ; ) and : V ult 0 (V ; ), and we
may also define a canonical elementary embedding , : ult 0 (V ; )
ult 0 (V ; ).
Let A , and let 0 . We say that A is homogeneously Souslin
iff there is some and a tree T on such that A = p[T ] and there is
(s : s < ) such that for all s < , s is a < + closed ultrafilter on
Ts = {t : (s, t) T }, if s t < , then s and t cohere, and if x A,
then
dir limn< (ult 0 (V ; x n ), xn ,xm : n m < ) is wellfounded. (13.28)
A is called homogeneously Souslin iff A is 0 homogeneously.
13.2. Show that in the situation of the preceeding paragraph, if (13.28) holds true,
then x A. [Hint. If Tx is wellfounded, then look at ||[id]xn ||xn (Tx ) ,
n < .]
13.3. (K. Windus) Let A
equivalent.
13.3 Problems
323
universally Baire. [Hint: Use Problem 6.18 and the construction from Theorem
13.3.]
Conclude also that if is a measurable cardinal and if 1 < < n < are
Woodin cardinals, then all 1 sets of reals are < 1 universally Baire.
n+2
13.5. Suppose that is a measurable cardinal and 1 < < n < are Woodin
cardinals. Let A be 1 , so that by Problem 13.4, A is < 1 universally
n+2
Baire. Let P V1 , and let g be Pgeneric over V . Let A be the new version
of A in V [g] (cf. p. 150). Show that if V [g] |= A = , then V |= A = .
(Compare Problem 10.16.)
13.6. (W. H. Woodin) Let be a strong cardinal, and let A be universally
Baire. (Equivalently, A is universally Baire, cf. Problem 8.10.)
Let g be Col(, 2(2 ) )generic over V , and let A be the new version of A
in V [g] (cf. p. 150). Show that R ( V [g]) \ A = {x : y
x y
/ A } is universally Baire in V [g]. [Hint. In V , let T and U
on witness that A is universally Baire. In V [g], we construct T
and U by amalgamating setsized trees. We get T by rearranging stretched
versions of U . For every (short) 0 complete (, )extender E let us define
an approximation U E to U as follows. Let E : V M be the ultrapower
map, where M is transitive. In V [g], fix a reasonable enumeration (n : n < )
of all E (E a ), a []< , and write (n) for Card(a) in case n = E (E a ). If
i and j cohere [cf. Definition 10.45 (2)], then we write i j for the canonical
embedding from ult(M; i ) to ult(M; j ). For s < , say k = lh(s), we
set (s, (0 , . . . , k1 )) U E iff
i < j < k ( EV (Us (i) ) i EV (Us ( j) ) j j projects to i
i j (i ) > j ).
Show that this works.]
Conclude that if is the supremum of infinitely many strong cardinals and if
G is Col(, )generic over V [G], then in V [G] every projective set of reals
is Lebesgue measurable and has the property of Baire.
References
1. Abraham, U., Magidor, M.: Cardinal arithmetic. In: Foreman, M., Kanamori, A. (eds.) Handbook of Set Theory, vol. 2, pp. 11491228. Springer, Berlin (2010)
2. Andretta, A., Neeman, I., Steel, J.: The domestic levels of k c are iterable. Isr. J. Math. 125,
157201 (2001)
3. Bartoszynski, T., Judah, H.: Set Theory. On the Sructure of the Real Line. A K Peters, Wellesley,
MA (1995)
4. Becker, H., Kechris, A.: The descriptive set theory of Polish group actions. London Mathematical Society Lecture Notes Series, vol. 232 (1996)
5. Blass, A.: Combinatorial cardinal characteristics of the continuum. In: Kanamori, A., Foreman,
M. (eds.) Handbook of Set Theory, vol. 1, pp. 395489. Springer, Berlin (2010)
6. Blau, U.: Die Logik der Unbestimmtheiten und Paradoxien. Synchron-Verlag (2008)
7. Bilinsky, E., Gitik, M.: A model with a measurable which does not carry a normal measure.
Archive Math. Logic 51, 863876 (2012)
8. Caicedo, A., Ketchersid, R.: A trichotomy theorem in natural models of AD + . In: Babinkostova, L., Caicedo, A., Geschke, S., and Scheepers, M. (eds.) Set Theory and Its Applications,
Contemporary Mathematics, vol. 533, pp. 227258, American Mathematical Society, Providence, RI (2011)
9. Cummings, J.: Iterated forcing and elementary embeddings. In: Foreman, M., Kanamori, A.
(eds.) Handbook of Set Theory, vol. 2, pp. 775883. Springer, Berlin (2010)
10. Devlin, K., Ronald, B.: Jensen. Marginalia to a theorem of Silver. Lecture Notes in Mathematics
# 499, pp. 115142. Springer, Berlin (1975)
11. Enderton, H.B.: A Mathematical Introduction to Logic, 2nd edn. Harcourt Academic Press,
Burlington (2011)
12. Gao, S.: Invariant Descriptive Set Theory. CRC Press, Boca Raton (2009)
13. Halbeisen, L.: Combinatorial Set Theory. Springer, Berlin (2012)
14. Hjorth, G., Kechris, A.: New dichotomies for Borel equivalence relations. Bull. Symbolic Logic
3, 329346 (1997)
15. Jech, T.: Set Theory, 3rd edn. Springer, Berlin (2002)
16. Jensen, Ronald B.: The fine structure of the constructible hierarchy. Ann. Math. Logic 4, 229
308 (1972)
17. Jensen, R.B., Schimmerling, E., Schindler, R., Steel, J.: Stacking mice. J. Symb. Logic 74,
315335 (2009)
18. Kanamori, A.: The Higher Infinite, 2nd edn. Springer, Berlin (2009)
19. Kanovei, V., Sabok, M., Zapletal, J.: Canonical Ramsey theory on Polish spaces. Cambridge
Tracts in Mathematics. Cambridge University Press, Cambridge (2013)
R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4,
Springer International Publishing Switzerland 2014
325
326
References
20. Kechris, A.: Classical Descriptive Set Theory Graduate Texts in Mathematics # 156. Springer,
Berlin (1994)
21. Kechris, A., Miller, B.: Topics in orbit equivalence. Lecture Notes in Mathematics 1852.
Springer, Berlin (2004).
22. Koellner, P., Woodin, W.H.: Large cardinals from determinacy. In: Foreman, M., Kanamori, A.
(eds.) Handbook of Set Theory, vol. 3, pp. 19512119. Springer, Berlin (2010)
23. Kunen, K.: Set Theory. An Introduction to Independence Proofs. Elsevier, Amsterdam (1980)
24. Larson, P.: The Stationary Tower. In: Woodin, W.H. (ed.) Notes on a Course. AMS, New York
(2004)
25. Larson, P.: AD + . Monograph in preparation
26. Martin, D.A., Steel, J.: A proof of projective determinacy. J. Am. Math. Soc. 2, 71125 (1989)
27. Mochovakis, Y.: Descriptive Set Theory, 2nd edn. AMS, New York (2009)
28. Mitchell, W.J., Schimmerling, E.: Weak covering without countable closure. Math. Res. Lett.
2, 595609 (1995)
29. Mitchell, W.J., Schimmerling, E., Steel, J.: The covering lemma up to a woodin cardinal. Ann.
Pure Appl. Logic 84, 219255 (1997)
30. Mitchell, W.J., Steel, J.R.: Fine structure and iteration trees. Lecture Notes in Logic, vol. 3
(1994).
31. Neeman, Itay: Inner models in the region of a woodin limit of woodin cardinals. Ann. Pure
Appl. Logic 116, 67155 (2002)
32. Neeman, I.: The Determinacy of Long Games. de Gruyter, Berlin (2004)
33. Neeman, I.: Determinacy in L(R). In: Foreman, M., Kanamori, A. (eds.) Handbook of Set
Theory, vol. 3, pp. 18771950. Springer, Berlin (2010)
34. Sargsyan, G.: Descriptive inner model theory. Bull. Symbolic Logic 19, 155 (2013)
35. Schimmerling, E., Zeman, M.: A characterization of in core models. J. Math. Logic 4, 172
(2004)
36. Schindler, R., Steel, J.: The core model induction. Monograph in preparation. http://wwwmath.
uni-muenster.de/logik/Personen/rds/core_model_induction.pdf
37. Shelah, S.: Proper and Improper Forcing, 2nd edn. Springer, Berlin (1998)
38. Shiryaev, A.: Probability Theory, 2nd edn. Springer, New York (1996)
39. Steel, J.: An outline of inner model theory. In: Kanamori, A., Foreman, M. (eds.) Handbook of
Set Theory, vol. 3, pp. 15951684. Springer, Berlin (2010)
40. Steel, J.: The core model iterability problem. Lecture Notes in Logic, vol. 8 (1996).
41. Steel, J.: A stationary tower free proof of the derived model theorem. In: Gao, S., Zhang, J.
(eds.) Advances in Logic. The North Texas Logic Conference 810 Oct 2004, pp. 18, AMS,
New York (2007).
42. Steel, J.: Derived models associated to mice. In: Chong, C., Feng, Q., Slaman, T.A., Woodin,
W.H., Yong, Y. (eds.) Computational Aspects of Infinity, Part I: Tutorials pp. 105194, Singapore (2008).
43. Windus, K.: Projektive Determiniertheit. Diplomarbeit, Universitt Bonn (1993)
44. Woodin, W.H.: The Axiom of Determinacy, Forcing Axioms, and the Nonstationary Ideal, 2nd
edn. de Gruyter, Berlin (2010)
45. Woodin, W.H.: Suitable extender models i. J. Math. Logic 10, 101339 (2010)
46. Woodin, W.H.: Suitable extender models ii: beyond -huge. J. Math. Logic 11, 115436 (2011)
47. Zeman, M.: Inner models and large cardinals. de Gruyter Series in Logic and Its Applications,
vol. 5 (2002).
Index
A
Absolute, 68
Absolute between V and M, 68
Absoluteness of well-foundedness, 69
Accumulation point, 4
Ackermanns set theory, 20
AD, 280
ab, 87
ADR , 301
, 34
Almost disjoint, 7
Amenable, 73
n, p
A M , 250
Antichain, 95
maximal, 95
p
A M , 237
Approximation property, 122
AST, 20
At most countable, 3
Axiom
of choice, 12
of extensionality, 9
of foundation, 9, 30
of infinity, 10
of pairing, 10
of power set, 10
of replacement, 11
of separation, 11
of union, 10
Axiom of Determinacy, 280
Axiom of Real Determinacy, 301
B
Baire Category Theorem, 6
Baire space, 127
Banach-Mazur game, 287
Base, 288
Bernstein property, 155
, 63
BG, 18
BGC, 19
Bijective, 14
C
Cantor normal form, 30
Cantor space, 127
Cantors Discontinuum, 8
CantorBendixson
Theorem of, 5
CantorBendixson rank, 144
CantorSchrderBernstein
Theorem of, 2
Cardinal, 33
-Erds, 230
-strong, 58
-supercompact, 59
(strongly) Mahlo, 46
(strongly) inaccessible, 46
ineffable, 50
limit, 33
measurable, 53
regular, 36
Reinhardt, 52
remarkable, 232
singular, 36
strong, 58
strong limit, 46
strong up to with respect to A, 227
subcompact, 61
subtle, 90
successor, 33
supercompact, 59
327
328
threadable, 277
weakly compact, 50
weakly inaccessible, 46
weakly Mahlo, 46
with the tree property, 50
Woodin, 227
Cardinal successor, 33
Cardinality, 33
Cartesian product, 13
CH, 36
Closed, 40
-, 40
in []< , 45
Closure, 128
Club
-in , 40
in []< , 45
in , 40
Club filter, 41
Cofinal, 36, 239
Cofinality, 36
Cohen forcing, 94
at , 109
Collection principle, 30
Compatible, 93
Comprehension
0 , 73
axiom for , 19
schema, 19
Condensation
full, 81
local, 81
Condensation lemma, 80
Condensation point, 5
Cone
S, 288
Turing, 288
Constructible universe, 74
Continuum Hypothesis, 3, 36
Continuum Problem, 3
Countable, 3
Countable chain condition, 105
Covering game, 286
Critical point, 50
crit( ), 50
D
Decides
p (1 , . . . , k ), 98
Lemma, 106
1T , 68
-system, 106
Index
Dense, 93
below p, 93
Dependent choice, 31
Determinacy
OD S , 291
, 82
, 82
(R), 82
+
, 90
+
(R), 90
, 83
(R), 83
, 83
, , 275
Distance, 127
Domain, 13
Downward absolute, 68
E
Elementary embedding, 50
non-trivial, 50
Embedding normal form, 303
closed, 303
additivity, 307
Extender, 210
closed, 217
complete, 219
continuumcomplete, 220
countably complete, 220
derived from , E , 211
long, 210
short, 210
strength of an, 216
Extension of embeddings lemma
downward, 240
general downward, 253
general upward, 253
upward, 243
Extensional, 28
F
Filter, 40, 93
< -closed, 40
Frchet, 41
normal, 41
nowhere dense, 180
Ramsey, 180
rapid, 165
selective, 180
uniform, 53
weakly normal, 59
Index
Finite, 3
Forces, 97
Forcing
Mathias, 181
Namba, 276
notion of, 93
Prikry, 188
supercompact tree Prikry, 233
Forcing conditions, 93
Formula
n , 67
0 , 67
n , 67
F + , 42
Fullness, 101, 123
Function, 13
rudimentary (in E), 70
simple, 72, 296
uniformizing, 139
G
Gdel pairing function, 35
GCH, 36
Generalized Continuum Hypothesis, 36
Generic
P-over M, 94
D -, 93
Generic extension, 96
Good embedding, 243
H
Hausdorff formula, 38
HC, 38
Height of a tree T , 46
Hereditarily in, 86
Hereditarily ordinal definable from , 86
Hereditarily smaller than , 38
HF, 38
H , 38
h M , 193
n+1, p
hM
, 252
HOD, 86
HODz , 86
Homomorphism, 113
dense, 113
I
Illfounded, 26
Inaccessible to the reals
1 is -, 141
Incompatible, 93
329
Induction principle, 23, 26
Inductive, 10, 23
Ineffable, 50
Injective, 14
Inner model, 51
Interpretation
G-of , 96
Invariant
S-, 288
Turing-, 288
Isomorphic, 15
Isomorphism, 15
Iterability
of a ppm M , 197
Iterable by U and its images, 184
Iteration
linear of V of length given by U , 183
of a ppm M of length , 197
putative of a ppm M of length , 196
putative linear of V of length given
by U , 183
J
J , 74
J [E], 73
J -structure, 79
J-structure
acceptable, 235
K
-chain condition, 105
-Knaster, 105
KP, 90
KripkePlatek set theory, 90
Kurepas Hypothesis, 90
L
L, 74
< , 37
L[E], 73
Lebesgue measurable, 148
Length (of a wellordering), 29
< -closed, 109
< -distributive, 109
<M , 65
Level of s in T , 46
lvT (s), 46
Levy collapse, 111
Limit point, 40
os Theorem, 54, 195, 213
330
M
Measure, 53
Mitchell order, 65, 233
M n, p , 250
Monotone enumeration, 29
Mostowski collapse, 28
Mouse
x, 198
M p , 237
Mutual generics, 119
N
Names
P-, 96
n-completion, 254
n-embedding, 259
Nice name, 108
Norm on A , 145
Null set, 7, 147
Number
natural, 23
ordinal, 24
O
OD, 86
ODz , 86
complete, 197
1z , 91
Open, 95
x y, 133
Order
atomless, 95
linear, 14
partial, 14, 93
separative, 115
Order type, 29
Order-preserving, 15
Ordered pair, 13
Ordinal, 24
z-admissible, 91
limit, 25
successor, 25
Ordinal definable from , 86
Otp, 29
Outer measure, 147
P
Ppoint, 180
Parameters
good, 237, 251
very good, 238, 251
Index
PD, 307
Perfect subset property, 142
Pigeonhole Principle, 33
UM , 57
PM , 237
n , 251
PM
Polish space, 127
Positive sets, 42
Power set, 2
P (X ), 2, 10
Premouse
x, 194
Prikry sequence, 189
Product, 118
Product Lemma, 118
Projection of T , 128
Projective Determinacy, 307
Projectum
1 -, 236
nth, 250
Q
Qpoint, 180
Q-formula, 238
R
Range, 13
Rank, 28, 30
Rank initial segment of V , 28
rk R (x), 28
x R , 28
Real
Cohen over M, 152
dominating, 181
generic over M, 153
random over M, 152
unbounded, 181
Recursion theorem, 26
Reduct, 237
nrmth , 250
Reduction property, 141
Reflection Principle, 89
Regressive, 42
Relation, 13
setlike, 26
Reshaped subset of 1 , 163
Restriction, 14
n (M), 250
1 (M), 236
Rigid, 202
R M , 238
R nM , 251
r n+1 elementary, 254
Index
331
rud E closed, 72
Rudimentary relation, 71
Russells Antinomy, 12
S
S [E], 75
Scale on A , 145
SCH, 39
Sequence, 29
Set
admissible, 91
determined, 280
Set of reals
F , 128
G , 128
1n (x), 137
n1 (x), 137
11 (x), 137
1 (x), 137
n+1
-Souslin, 131
1 , 132
1
n+1
, 132
1 , 132
1
1 , 132
n+1
analytic, 132
basic open, 127
Borel, 128
closed, 3, 127
coanalytic, 132
complete coanalytic, 134
homogeneously Souslin, 323
weakly homogeneously Souslin, 323
dense, 4
determined, 280
homogeneously Souslin, 323
meager, 6, 148
nowhere dense, 6, 148
of first category, 6
of second category, 6
open, 3, 127
perfect, 4
projective, 132
small, 174
Solovay over M, 153
universal 1 , 133
1
332
U
ult(V ; U ), 57
Ultrafilter, 40
Ultrapower
0 , 194, 216
r n+1 -, 262
Ultrapower embedding, 57
Ultrapower map
0 , 195, 216
r n+1 -, 262
Ultrapower of M by U , 57
Unbounded
in []< , 45
in , 40
Uncountable, 3
Upward absolute, 68
V
V , 28
Villes Lemma, 91
Index
W
Weakly r n+1 elementary, 254
Wellfounded, 26
Wellfounded part, 26
Well-ordering, 15
Wfp(B), 26
Windus tree, 304
Winning strategy, 280
X
x # , 202
Z
Z, 12
ZC, 12
0# , 202
ZF, 12
ZFC, 12
ZFC , 12
ZFC , 12
Zorns Lemma, 16