You are on page 1of 335

Universitext

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

Universitext is a series of textbooks that presents material from a wide variety of


mathematical disciplines at masters level and beyond. The books, often well
class-tested by their author, may have an informal, personal, even experimental
approach to their subject matter. Some of the most successful and established
books in the series have evolved through several editions, always following the
evolution of teaching curricula, into very polished texts.
Thus as research topics trickle down into graduate-level teaching, first textbooks
written for new, cutting-edge courses may make their way into Universitext.

For further volumes:


http://www.springer.com/series/223

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)

To Julia, Gregor, and Joana, with love

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

Naive Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .


1.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Axiomatic Set Theory . . . . . . . . . .


2.1 ZermeloFraenkel Set Theory
2.2 GdelBernays Class Theory .
2.3 Problems . . . . . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

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

Descriptive Set Theory . . . . . . .


7.1 Definable Sets of Reals . . .
7.2 Descriptive Set Theory and
7.3 Problems . . . . . . . . . . . . .

............
............
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

The Raisonnier Filter . . . . . .


9.1 Rapid Filters on x . . . .
9.2 Mokobodzkis Theorem.
9.3 Problems . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

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

11 0# and Jensens Covering Lemma .


11.1 Fine Structure Theory . . . . . .
11.2 Jensens Covering Lemma . . .
11.3 hj and Its Failure . . . . . . . .
11.4 Problems . . . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

235
235
262
269
275

12 Analytic and Full Determinacy .


12.1 Determinacy . . . . . . . . . .
12.2 Martins Theorem . . . . . . .
12.3 Harringtons Theorem . . . .
12.4 Problems . . . . . . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

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

Naive Set Theory

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}.

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_1,


Springer International Publishing Switzerland 2014

1 Naive Set Theory

Let us now recursively define a sequence of closed intervals [an , bn ] = {x : an


x bn } as follows. Put [a0 , b0 ] = [0, 1]. Suppose [an , bn ] has already been defined.
1
and
Pick [an+1 , bn+1 ] so that an an+1 < bn+1 bn , bn+1 an+1 n+1
xn  [an+1
,
b
].
 n+1
Now nN [an , bn ] = {x} for some x R by the Nested Interval Principle.
Obviously, x = xn for every n, as xn  [an+1 , bn+1 ] and x [an+1 , bn+1 ]. Hence

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.

Theorem 1.4 (CantorSchrderBernstein) Let X and Y be arbitrary. If X Y


and Y X , then X Y .
Proof Let both f : X Y and g : Y X be injective. We are looking for a
bijection h : X Y . Let x X . An X orbit of x is a finite or infinite sequence of
the form
g 1 (x), f 1 (g 1 (x)), g 1 ( f 1 (g 1 (x))), . . .
For each n N {} there is obviously at most one X orbit of x of length n. Let
n(x) be the maximal n N {} so that there is an X orbit of x of length n. We
put x X 0 iff n(x) = , x X 1 iff n(x) N is even, and x X 2 iff n(x) N is
odd.
For y Y we define the concept of a Y orbit in an analoguous way, i.e., as a
finite or infinite sequence of the form
f 1 (y), g 1 ( f 1 (y)), f 1 (g 1 ( f 1 (y))), . . .
We write n(y) for the maximal n N {} so that there is a Y orbit of y of length
n. We set y Y0 iff n(y) = , y Y1 iff n(y) N is odd, and y Y2 iff n(y) N
is even.

1 Naive Set Theory

Let us now define h : X Y by



h(x) =

f (x)
g 1 (x)

if x X 0 X 1 , and
if x X 2 .

The function h is well-defined, as X is the disjoint union of X 0 , X 1 , and X 2 , and


because for every x X 2 there is an X orbit of x of length 1, i.e., g 1 (x) is defined.
The function h is injective: Let x1 = x2 with h(x1 ) = h(x2 ). Say x1 X 0 X 1
and x2 X 2 . Then obviously h(x1 ) = f (x1 ) Y0 Y1 and h(x2 ) = g 1 (x2 ) Y2 .
But Y is the disjoint union of Y0 , Y1 , and Y2 . Contradiction!
The function h is surjective: Let y Y0 Y1 . Then y = f (x) for some x X 0 X 1 ;

but then y = h(x). Let y Y2 . Then g(y) X 2 , so y = g 1 (g(y)) = h(g(y)).


Cantors Continuum Problem is the question if there is a set A of real numbers
such that
N < A < R.
This problem has certainly always been one of the key driving forces of set theory.
A set A is called at most countable if A N. A is called countable if A N, and
A is called finite iff A < N. A is called uncountable iff N < A.
Cantors Continuum Hypothesis says that the Continuum Problem has a negative
answer, i.e., that for every uncountable set A of real numbers, A R.
Cantor initiated the project of proving the Continuum Hypothesis by an induction on the complexity of the sets A in question. There is indeed a hierarchy of
sets of reals which we shall study in Chap. 7. The open and closed sets sit at the very
bottom of this hierarchy.
Let A R. A is called open iff for every a A there are c < a and b > a with
(c, b) = {x : c < x < b} A. A is called closed iff R\A is open.
It is easy to see that if A R is any nonempty open set, then R A. As A R
is trivial for every A R, we immediately get that A R for every nonempty
open A R with the help of the Theorem 1.4 of CantorSchrderBernstein.
Theorem 1.9 of CantorBendixson will say that A R for every uncountable
closed set A R, which may be construed as a first step towards a realization
of Cantors project. We shall later prove much more general results (cf. Theorem
12.11 and Corollary 13.8) which have a direct impact on Cantors project.
Lemma 1.5 Let A R. The following are equivalent:
(1) A is closed.
(2) For all x R, if a < x < b always implies (a, b) A = , then x A.
Proof (1) = (2): Let x
/ A. Let a < x < b be such that (a, b) R\A. Then
(a, b) A = .
(2) = (1): We prove that R\A is open. Let x R\A. Then there are a < x < b
so that (a, b) A = , i.e., (a, b) R\A.

1 Naive Set Theory

Let A R. x is called an accumulation point of A iff for all a < x < b,


(a, b) (A\{x}) = (here, x itself need not be an element of A). The set of all
accumulation points of A is called the (first) derivative of A and is abbreviated by
A . Lemma 1.5 readily gives:
Lemma 1.6 Let A R. The following are equivalent:
(1) A is closed.
(2) A A.
Let A R. A set B A is called dense in A iff for all a, b R with a < b
and [a, b] A = , [a, b] B = . B R is called dense iff B is dense in R. It is
wellknown that Q is dense.
Definition 1.7 A set A R is called perfect iff A = and A = A.
Theorem 1.8 Let A R be perfect. Then A R.
Proof A R is trivial. It thus remains to be shown that R A. We shall make use
of the fact that R N {0, 1}, where N {0, 1} is the set of all infinite sequences of 0s
and 1s. (Cf. Problem 1.2.) We aim to see that N {0, 1} A.
Let {0, 1} be the set of all non-empty finite sequences of 0s and 1s, i.e., of all
s : {0, . . . , n} {0, 1} for some n N. Let us define a function from {0, 1} to
closed intervals as follows.
Let s0 : {0} {0} and s1 : {0} {1}. As A = and A A we easily find
as0 < bs0 < as1 < bs1
so that
(as0 , bs0 ) A = and (as1 , bs1 ) A = .
Set (s0 ) = [as0 , bs0 ] and (s1 ) = [as1 , bs1 ].
Now let s {0, 1} and suppose that (s) is already defined, where (s) =
[as , bs ] with as < bs and (as , bs ) A = .
Let s : {0, . . . , n} {0, 1}. For h = 0, 1 write s h for the unique t : {0, . . . , n +
1} {0, 1} with t (i) = s(i) for i n and t (n + 1) = h. Because A A , we
easily find
as < as 0 < bs 0 < as 1 < bs 1 < bs ,
so that
(as 0 , bs 0 ) A = , (as 1 , bs 1 ) A = ,
bs 0 as 0

1
1
, and bs 1 as 1
.
n+1
n+1

Set (s h) = [as h , bs h ] for h = 0, 1.

1 Naive Set Theory

We may now define an injection F : N {0, 1} A. Let f N {0, 1}. Then




[a f {0,...,n} , b f {0,...,n} ] = {x}

nN

for some x R by the Nested Interval Principle. Set F( f ) = x. Obviously,


F( f ) A, as F( f ) is an accumulation point of A and A A. Also, F is certainly
injective.

Theorem 1.9 (CantorBendixson) Let A R be closed. Then there are sets A0 R


and P R so that:
(1) A is the disjoint union of A0 and P,
(2) A0 is at most countable, and
(3) P is perfect, unless P = .
Corollary 1.10 Let A R be closed. Then A N or A R.
Proof of Theorem 1.8. An x R is called a condensation point of A iff (a, b) A
is uncountable for all a < x < b.
Let P be the set of all condensation points of A, and let A0 = A\P. As A is
closed, P A A. It remains to be shown that (2) and (3) both hold true. We shall
make use of the fact that Q N (cf. Problem 1.1) and that Q is dense, so that that
for all x, y R with x < y there is some z Q with x < z < y.
Let x A0 . Then there are ax < x < bx with ax , bx Q and such that
(ax , bx ) A is at most countable. Therefore,
A0

(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}

is at most countable. Contradiction!


Let us finally show that P  P. Let x P  . Then (a, b) (P\{x}) = for
all a < x < b. Let y (a, b) (P\{x}), where a < x < b. Then (a, b) A is
uncountable. Hence x P.

There is a different proof of the Theorem of CantorBendixson which brings


the concept of an ordinal number into play. Let A R be closed. Define A1 as A ,

1 Naive Set Theory

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 ].

But now = nN [an , bn ] [a, b] nN An , as desired.


The Baire Category Theorem implies that R is of second category, in fact that
the complement of a meager set is dense in R (cf. Problem 1.8 (b)).
If a, b R, a < b, then we call b a the length of the closed interval [a, b]. As
Q is dense in R, any union of closed intervals may be written as a union of closed
intervals with rational endpoints (cf. the proof of Theorem 1.9) and thus as a union

1 Naive Set Theory

of countably many closed intervals


which in addition may be picked to be pairwise

disjoint. If A [0, 1], A = nN [an , bn ], where an < bn for each n N and the
[an , bn ] are pairwise disjoint, then we write
(A) =

bn an

nN

and call it the measure of A. One can show that (A)


is independent from the choice
of the pairwise disjoint intervals [an , bn ] with A = nN [an , bn ] (cf. Problem 1.7).
Definition 1.13 Let A
 [0, 1]. Then A is called a null set iff for all > 0 there is
a countable union A = nN [an , bn ] of closed intervals [an , bn ] [0, 1] such that
(A) .
Of course, every countable subset of [0, 1] is null, and in fact the countable union of
null sets is null, but there are null sets which have the same size as R (cf. Problem
1.8(b)).

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 = .

1 Naive Set Theory

(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

and for a0 , b0 , . . ., ak , bk R with ai < bi for all i k let


2

([a0 , b0 ] . . . [ak , bk ]) 3 = [a0 , b0 ] 3 . . . [ak , bk ] 3 .


2

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

[a, b] is called Cantors Discontinuum. Show that for all a, b R with


a < b, [a, b] is dense in [a, b], and [a, b] is perfect, nowhere dense, and a
null set.

Chapter 2

Axiomatic Set Theory

Ernst Zermelo (18711953) was the first to find an axiomatization of set theory,
and it was later expanded by Abraham Fraenkel (18911965).

2.1 ZermeloFraenkel Set Theory


The language of set theory, which we denote by L , is the usual language of first
order logic (with one type of variables) equipped with just one binary relation symbol,
. The intended domain of set theoretical discourse (i.e., the range of the variables)
is the universe of all sets, and the intended interpretation of is is an element of.
We shall use x, y, z, . . ., a, b, . . ., etc. as variables to range over sets.
The standard axiomatization of set theory, ZFC (ZermeloFraenkel set theory
with choice), has infinitely many axioms. The first one, the axiom of extensionality,
says that two sets are equal iff they contain the same elements.
xy(x = y z(z x z y)).

(Ext)

A set x is a subset of y, abbreviated by x y, if z(z x z y). (Ext) is then


logically equivalent to xy(x y y x x = y). We also write y x for
x y. x is a proper subset of y, written x  y, iff x y and x = y.
The next axiom, the axiom of foundation, says that each nonempty set has an
-minimal member.
x(y y x y(y x z(z y z x))).

(Fund)

This is easier to grasp if we use the following abbreviations: We write x = for


y y x (and x = for y y x), and x y = for z(z x z y). (Fund)
then says that
x(x = y(y x y x = )).

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_2,


Springer International Publishing Switzerland 2014

10

2 Axiomatic Set Theory

(Fund) plays an important technical role in the development of set theory.


Let us write x = {y, z} instead of
y x z x u(u x (u = y u = z)).
The axiom of pairing runs as follows.
xyz z = {x, y}.

(Pair)

We also write {x} instead of {x, x}.


In the presence of (Pair), (Fund) implies that there cannot be a set x with x x:
if x x, then x is the only element of {x}, but x {x} = , as x x {x}. A similar
argument shows that there cannot be sets x1 , x2 , . . ., xk such that x1 x2
2.1).
xk x1 (cf. Problem
Let us write x = y for
z(z x u(u y z u)).
The axiom of union is the following one.
xy y =
Writing z = x y for

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).

2.1 ZermeloFraenkel Set Theory

11

We then say:
x(x is inductive).

(Inf)

We now need to formulate the separation and replacement schemas.


A schema is an infinite set of axioms which is generated in a simple (recursive)
way.
Let be a formula of L in which exactly the variables x, v1 , . . . , v p (which
all differ from b) occur freely. The axiom of separation, or Aussonderung, corresponding to runs as follows.
v1 . . . v p abx (x b x a ).

(Aus )

Let us write b = {x a: } for x(x b x a ). If we suppress v1 , . . . , v p ,


(Aus ) then says that
ab b = {x a: }.
Writing z = x y for

u(u z u x u y),

(Ausxc ) proves that acb b = a c. Writing z = x\y for


u(u z u x u y),
(Ausxc ) proves that acb b = a\c. Also, if we write x =

y for

z(z x u(u y z u)),


then (Ausu(uyzu) ), applied to any member of y proves that
y(y = x x =

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

2 Axiomatic Set Theory

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)

In what follows we shall always abbreviate y(y x ) by y x and


y(y x ) by y x . We may then also formulate (AC) as
x(y x y = y xy x(y = y y y = )
zy xu z y = {u}),
i.e., for each member of x, z contains exactly one representative.
One may also formulate (AC) in terms of the existence of choice functions (cf.
Problem 2.6).
The theory which is given by the axioms (Ext), (Fund), (Pair), (Union), (Pow),
(Inf) and (Aus ) for all is called Zermelos set theory, abbreviated by Z. The
theory which is given by the axioms of Z together with (Rep ) for all is called
ZermeloFraenkel set theory, abbreviated by ZF. The theory which is given by
the axioms of ZF together with (AC) is called ZermeloFraenkel set theory with
choice, abbreviated by ZFC. This system, ZFC, is the standard axiomatization of set
theory. Most questions of mathematics can be decided in ZFC, but many questions
of set theory and other branches of mathematics are independent from ZFC. The
theory which is given by the axioms of Z together with (AC) is called Zermelo set
theory with choice and is often abbreviated by ZC. We also use ZFC to denote
ZFC without (Pow), and we use ZFC to denote ZFC without (Inf).

2.1 ZermeloFraenkel Set Theory

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}.

Lemma 2.1 For all a, b, a b exists, i.e., abc c = a b.


Proof a b may be separated from P(P(a b)).
We also define a1 an+1 to be (a1 an ) an+1 and

a .
a n = a 
n -times

An n-ary relation r is a subset of a1 an for some sets a1 , . . ., an . The


n-ary relation r is on a iff r a n . If r is a binary (i.e., 2-ary) relation, then we often
write x r y instead of (x, y) r , and we define the domain of r as
dom(r ) = {x: y x r y}
and the range of r as
ran(r ) = {y: x x r y}.
A relation r a b is a function iff
x dom(r )yy (y = y x r y ).
If f a b is a function, and if x dom( f ), then we write f (x) for the unique
y ran( f ) with (x, y) f .
A function f is a function from d to b iff d = dom( f ) and ran( f ) b (sic!),
which we also express by writing
f : d b.
The set of all functions from d to b is denoted by d b.

14

2 Axiomatic Set Theory

Lemma 2.2 For all d, b, d b exists.


Proof d b may be separated from P(d b).

If f : b d and g: d e, then we write g f for the function from b to e which
sends x b to g( f (x)) e.
If f : d b, then f is surjective iff b = ran( f ), and f is injective iff
x dx d( f (x) = f (x ) x = x ).
f is bijective iff f is surjective and injective.
If f : d b and a d, then f  a, the restriction of f to a, is that function
g: a b such that g(x) = f (x) for every x a. We write f a for the image of a
under f , i.e., for the set {y ran( f ): x a y = f (x)}. Of course, f a = ran( f 
a).
If f : d b is injective, and if y ran( f ), then we write f 1 (y) for the unique
x dom( f ) with f (x) = y. If c b, then we write f 1 c for the set {x
dom( f ): f (x) c}.
A binary relation on a set a is called a partial order on a iff is reflexive (i.e.,
x x for all x a), is symmetric (i.e., if x, y a, then x y y x x = y),
and is transitive (i.e., if x, y, z a and x y y z, then x z). In this case
we call (a, ) (or just a) a partially ordered set. If is a partial order on a, then
is called linear (or total) iff for all x a and y a, x y or y x.
If (a, ) is a partially ordered set, then we also write x < y iff x y x = y.
Notice that x y iff x < y x = y. We shall also call < a partial order.
Let (a, ) be a partially ordered set, and let b a. We say that x is a maximal
element of b iff x b y b x < y. We say that x is the maximum of b,
x = max(b), iff x b y b y x. We say that x is a minimal element of b iff
x b y b y < x, and we say that x is the minimum of b, x = min(b), iff
x b y b x y. Of course, if x = max(b), then x is a maximal element of
b, and if x = min(b), then x is a minimal element of b. We say that x is an upper
bound of b iff y x for each y b, and we say that x is a strict upper bound of b
iff y < x for each y b; x is the supremum of b, x = sup(b), iff x is the minimum
of the set of all upper bounds of b, i.e., if x is an upper bound and
y a(y b y y x y).
If x = max(b), then x = sup(b). We say that x is a lower bound of b iff x y for
each y b, and we say that x is a strict lower bound of b iff x < y for all y b; x
is the infimum of b, x = inf(b), iff x is the maximum of the set of all lower bounds
of b, i.e., if x is a lower bound and
y a(y b y y y x).
If x = min(b), then x = inf(b). If is not clear from the context, then we also say
-maximal element, -supremum, -upper bound, etc.

2.1 ZermeloFraenkel Set Theory

15

Let (a, a ), (b, b ) be partially ordered sets. A function f : a b is called


order-preserving iff for all x, y a,
x a y f (x) b f (y).
If f : a b is order-preserving and f is bijective, then f is called an isomorphism,
also written
f
(b, b ).
(a, a ) =
(a, a ) and (b, b ) are called isomorphic iff there is an isomorphism f : a b,
written
(a, a )
= (b, b ).
The following concept plays a key role in set theory.
Definition 2.3 Let (a, ) be a partial order. Then (a, ) is called a well-ordering
iff for every b a with b = , min(b) exists.
The natural ordering on N is a well-ordering, but there are many other well-orderings
on N (cf. Problem 2.7).
Lemma 2.4 Let (a, ) be a well-ordering. Then is total.
Proof If x, y a, then min({x, y}) x and min({x, y}) y. Hence if min({x, y}) =

x, then x y, and if min({x, y}) = y, then y x.
Lemma 2.5 Let (a, ) be a well-ordering, and let f : a a be order-preserving.
Then f (x) x for all x a.
Proof If {x a: f (x) < x} = , set
x0 = min({x a: f (x) < x}).
Then y0 = f (x0 ) < x0 and so f (y0 ) < f (x0 ) = y0 , as f is order-preserving. But
this contradicts the choice of x0 .

f

Lemma 2.6 If (a, ) is a well-ordering, and if (a, )


= (a, ), then f is the
identity.
Proof By the previous lemma applied to f as well as to f 1 , we must have f (x) x

as well as f 1 (x) x, i.e., f (x) = x, for every x a.
Lemma 2.7 Suppose that (a, a ) and (b, b ) are both well-orderings such that
f

(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

2 Axiomatic Set Theory

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),

so that r (x) = f (x) < y = r (x ).


If both a\ dom(r ) as well as b\ ran(r ) were nonempty, say x = min(a\ dom(r ))
and y = min(b\ dom(r )), then
r

(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.

2.1 ZermeloFraenkel Set Theory

17

Proof Fix (a, ) as in the hypothesis. Let


A = {{(b, x): x b}: b a, b = }.

Notice that A exists, as it can be separated from P(P(a) P(a)). (AC), the
axiom of choice, gives us some set f such that for all y A there is some z with
y f = {z}, which means that for all b a, b = , there is some unique x b such
that (b, x) f . Therefore, f is a function from P(a)\{} to a such that f (b) b
for every b P(a)\{}.
Let us now define a binary relation on a as follows.
We let W denote the set of all well-orderings of subsets b of a such that for all
u, v b, if u v, then u v, and for all u b, writing

Bu = {w a: w is a upper bound of {v b: v < u}},

Bu = and u = f (Bu ). Notice that W may be separated from P(a 2 ).


Let us show that if , W , then or else . Let W be a
well-ordering of b a, and let W be a well-ordering of c a.
By Theorem 2.9, we may assume by symmetry that either (b, )
= (c, ) or

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 )}.

But then Bu0 = Bg(u


= and thus
0)

) = g(u 0 ).
u 0 = f (Bu0 ) = f (Bg(u
0)

Contradiction!
We have shown that if , W , then or .

18

2 Axiomatic Set Theory

But now
Setting

W , call it , is easily seen to be a well-ordering of a subset b of a.


B = {w a: w is a upper bound of b},

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)

and = {(u, u 0 ): u b} {(u 0 , u 0 )}, then B = Bu0 . It is thus easy to see


that W . This gives u 0 b, a contradiction!
Thus b has a maximum with respect to . Zorns Lemma is shown.

The following is a special case of Zorns lemma (cf. Problem 3.10).
Corollary 2.11 (Hausdorff Maximality Principle) Let a = , and let A P(a) be
such that for all B A, if x y y x for all x, y B, then there is some z A
such that x z for all x B. Then A contains an -maximal element.
In the next chapter, we shall use the Hausdorff Maximality Principle to show that
every set can be well-ordered (cf. Theorem 3.23).
It is not hard to show that in the theory ZF, (AC) is in fact equivalent with Zorns
Lemma, with the Hausdorff Maximality Principle, as well as with the assertion
that every set can be well-ordered, i.e., that for every set x there is some well-order
< on x (cf. Problem 3.10).

2.2 GdelBernays Class Theory


There is another axiomatization of set theory, BGC, which is often more convenient
to use. Its language is the same one as L , except that in addition there is a second
type of variables. The variables x, y, z, . . ., a, b, . . . of L are supposed to range over
sets, whereas the new variables, X , Y , Z , . . ., A, B, . . . are supposed to range over
classes. Each set is a class, and a given class is a set iff it is a member of some class
(equivalently, of some set). Classes which are not sets are called proper classes.
Functions may now be proper classes. The axioms of the BernaysGdel class
theory BG are (Ext), (Fund), (Pair), (Union), (Pow), (Inf) exactly as before together
with the following ones:
X Y x((x X x Y ) X = Y )

(2.1)

xX x = X

(2.2)

X ( Y X Y x x = X )

(2.3)

If F is a (class) function, then F a is a set for each set a,

(Rep )

2.2 GdelBernays Class Theory

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

2 Axiomatic Set Theory

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)

Let be any formula of L in which exactly v1 , . . ., vk occur freely. Then


v results from by replacing every occurence of x by x v and every
occurence of x by x v.
Then
v1 v . . . vk v ( v ).

(Refl )

(Refl) is the schema of all (Refl ), where is a formula of L (in which v


does not occur). (Str) states that v is supertransitive, and (Refl) states (as a
schema) that v is a fully elementary submodel of V , the universe of all sets.
2.11. (W. Reinhardt) Show that every axiom of ZF is provable in AST.
AST is also conservative over ZF, cf. Problem 5.15.

Chapter 3

Ordinals

3.1 Ordinal Numbers


The axiom of infinity (Inf) states there is an inductive set. Recall that a set x is called
inductive iff x and for each y x, y {y} x. Let us write 0 for and y + 1
for y {y}. The axiom of infinity then says that there is a set x such that 0 x and
for each y x, y + 1 x. We shall also write 1 for 0 + 1, 2 for (0 + 1) + 1, etc.
Each inductive set therefore contains 0, 1, 2, etc. We shall write for

{x : x is inductive}.
This set exists by (Inf) plus the separation scheme: if x0 inductive, then
= {y x0 : x(x is inductive y x)}.
Clearly, is inductive. Intuitively, the set contains exactly 0, 1, 2, etc.
We have the following principle of induction.
Lemma 3.1 Let A be such that 0 A and for each y A, y + 1 A. Then
A = .
Proof A is inductive, hence A, and thus A = .

In particular, if is a statement such that (0) and y ((y) (y + 1))


both hold true, then y (y) holds true as well. We shall call elements of
natural numbers and itself the set of natural numbers. All natural numbers as well
as will be ordinals according to Definition 3.3.
Definition 3.2 A set x is transitive iff for each y x, y x.
We shall see later (cf. Lemma 3.14) that every set is contained in a transitive set.

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_3,


Springer International Publishing Switzerland 2014

23

24

3 Ordinals

The following concept of an ordinal was isolated by Jnos Neumann (1903


1957) which is why ordinals are sometimes called Neumann ordinals.
Definition 3.3 A set x is called an ordinal number, or just an ordinal, iff x is transitive and for all y, z x we have y z y = z z y.
Ordinals will typically be denoted by , , , . . ., i, j, . . . We shall write OR for the
class { : is an ordinal} of all ordinals.
By (Fund), if is an ordinal, then  = {(x, y) 2 : x y} is a well-order
of .
Lemma 3.4 Each natural number is an ordinal.
Proof by induction, i.e., by using Lemma 3.1: 0 is trivially an ordinal. Now let be
an ordinal. We have to see that + 1 is an ordinal. + 1 = {} is transitive: let
y {}; then either y and hence y {} because is transitive,
or else y = and hence y {}. Now let y, z + 1 = {}. We have to
see that y z y = z z y. If y, z , then this follows from the fact that is
an ordinal; if y, z {}, then this is trivial; but it is also trivial if y and z {}
or vice versa.

Lemma 3.5 is an ordinal.
Proof We first show y y by induction. This is trivial for y = 0. Now fix
y with y . Then {y} , hence y + 1 = y {y} .
We now show y z (y z y = z z y) by a nested induction. Let
us write (y, z) for y z y = z z y. In order to prove y z (y, z)
it obviously suffices to show the conjunction of the following three statements:
(a) (0, 0),
(b) z ((0, z) (0, z + 1)),
(c) y (z (y, z ) z (y + 1, z))
This is because if (a) and (b) hold true, then z (0, z) holds true by induction.
This, together with (c), yields y z (y, z) again by induction.
(a) and (b) are trivial.
As to (c), let us fix y , and let us suppose that z (y, z ). We aim to
show z (y + 1, z), and we will do so by induction. We already know that
z (0, z), which in particular gives (0, y + 1) and thus also (y + 1, 0) by
symmetry. Let us assume that (y + 1, z) holds true to deduce that (y + 1, z + 1)
holds true as well.
We have that y + 1 z y + 1 = z z y + 1 by hypothesis. If y + 1 z, then
y + 1 z + 1 = z {z}. If y + 1 = z, then y + 1 z + 1 = z {z} as well. Now
let z y + 1 = y {y}. If z {y}, then y + 1 = z + 1. So suppose that z y. We
have that y z + 1 y = z + 1 z + 1 y by our hypothesis z (y, z ).
But z y z + 1 = z {z} contradicts the axiom of foundation (consider {z, y}).
Therefore, z y implies y = z +1 z +1 y, and therefore z +1 y {y} = y +1
as desired.


3.1 Ordinal Numbers

25

Lemma 3.6 The following statements are true.


(a)
(b)
(c)
(d)

0 is an ordinal, and if is an ordinal, then so is + 1.


If is an ordinal and x , then x is an ordinal.
If , are ordinals, and  , then .
If , are ordinals, then or (and hence = ).

Proof (a) is given by the proof of Lemma 3.4 above.


(b) is easy.
To show (c), let be a proper subset of . Let \ such that (\) = .
(There is such a by the axiom of foundation.) If , then by the transitivity
of , so , as otherwise (\). If , then = ,
because is an ordinal. But = implies , because and is
an ordinal; however, \. Therefore if , then . We have shown that
= . Hence .
(d): Suppose not. Let OR be such that there is some OR with (
). Let 0 be minimal in + 1 = {} such that there is some OR
with (0 0 ), and let 0 OR be such that (0 0 0 0 ).
Clearly, 0 0 is transitive, and if , 0 0 , then or by the
choice of 0 , so that = by (b) and (c). Hence 0 0 is an
ordinal, call it 0 . We claim that 0 = 0 or 0 = 0 . Otherwise by (c), 0 0 and
0 0 , so that one of 0 0 , 0 0 , 0 0 0 holds true, which contradicts
the axiom of foundation. We have shown that 0 0 or 0 0 which contradicts

the choice of 0 and 0 .
By Lemma 3.6 (b) and (d), OR cannot be a set, as otherwise OR OR.
If , OR, then we shall often write instead of (equivalently,
= ) and < instead of . We shall also write (, ), [, ),
(, ], and [, ] for the sets { : < < }, { : < }, { : < },
and { : }, respectively.
Lemma 3.7 The following statements are true.

(a) If X = is a set of ordinals, then
X is the minimal element of X .

(b) If X is a set of ordinals, then X is also an ordinal.


Proof To show (a), notice that
 X is
certainly an ordinal. If X is a proper subset
of every element of X , then X X . Contradiction!
(b) is easy by the previous lemma.


If X is a set ofordinals, then we also write min(X ) for X (provided that X = )
and sup(X ) for X .
Definition 3.8 An ordinal is called a successor ordinal iff there is some ordinal
such that = + 1. An ordinal is a limit ordinal iff is not a successor
ordinal.

26

3 Ordinals

3.2 Induction and Recursion


Definition 3.9 A binary relation R B B on a set or class B is called well
founded iff every nonempty b B has an R-least element, i.e., there is x b such
that for all y b, y Rx. If R is not wellfounded, then we say that R is illfounded.
We have the following principle of induction for wellfounded relations:
Lemma 3.10 Let R B B be well-founded, where B is a set. Let A B be such
that for all x B, if {y B : y Rx} A, then x A. Then A = B.
Proof Suppose that B\A = . Let x B\A be Rleast, i.e., for all y B\A, y Rx.
In other words, if y Rx, then y A. Then x A by hypothesis. Contradiction! 
If B is a set, then  B = {(x, y) B B : x y} is wellfounded by the axiom
of foundation.
Lemma 3.11 R B B is wellfounded iff there is no f : B such that
f (n + 1)R f (n) for all n .
Proof Suppose there is an f : B such that f (n + 1)R f (n) for all n . Then
ran( f ) B doesnt have an Rleast element.
Now suppose that R is not wellfounded. Pick b B, b = with no Rleast
element; i.e.; for all x b, {y b : y Rx} = . Apply the axiom of choice to
the set {{(y, x) : y b y Rx} : x b} to get a set u such that for all x b,
u {(y, x) : y b y Rx} = {(y , x)} for some y ; write yx for this unique y .
We may now define f : B as follows. Pick x0 b, and set f (0) = x0 . Set
f (n) = y iff there is some g : n + 1 b such that g(n) = y, g(0) = x0 , and for all
i n, g(i + 1) = yg(i) .
Obviously, for each n there is at most one such g, and an easy induction
yields that for each n , there is at least one such g. But then f is welldefined,
and of course f (n + 1)R f (n) for all n .

If R B B, then the wellfounded part wfp(R) of B is the class of all
x B such that there is no infinite sequence (xn : n < ) such that x0 = x and
(xn+1 , xn ) R for all n < .
The previous proof gave an example of a recursive definition. There is a general
recursion theorem. We state the NBG version of it which extends the ZFC version.
Definition 3.12 Let R B B, where B is a class. R is then called setlike iff
{x : (x, y) R} is a set for all y B.
Theorem 3.13 (Recursion) Let R B B be wellfounded and setlike, where B
is a class. Let p be a set,1 and let (v0 , v1 , v2 , p) be such that for all sets u and x

p will play the role of a parameter in what follows.

3.2 Induction and Recursion

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)

We call V a rank initial segment of V .


Cf. Problem 3.1.
Definition 3.20 A binary relation R B B on a class B is called extensional iff
for all x, y B,
{z B : z Rx} = {z B : z Ry} x = y.
By the axiom(s) of extensionality (and foundation),  B is (wellfounded and)
extensional for every set B.
The function R as in the following theorem is often called the transitive collapse.
Theorem 3.21 (Mostowski Collapse) Let B be a class. Let R B B be well
founded, extensional, and setlike. There is then a unique pair (X R , R ) such that
X R is transitive, R : X R B is bijective, and for all x, y X R , x y
R (x)R R (y).

3.2 Induction and Recursion

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

there is some isomorphism (; < )


= (B; R) is called the length or the order
type of R, denoted by otp(R). If A is a set of ordinals, then we also denote by
otp(A) the order type of < A and call it the order type of A. The isomorphism

(otp(A); < otp(A))


= (A; < A) is also called the monotone enumeration of A.
Theorem 3.23 (Zermelo) Let A be any set. There is then a wellordering on A.
There is even an ordinal and some bijection : A.
Proof We use the Hausdorff Maximality Principle 2.11 to show that there is a
bijection : A for some ordinal . Let F be the set of all injections : A,
where is an ordinal.
F is indeed a set by the following argument. For each : A, R A A
is a wellordering on ran( ), where we define x Ry 1 (x) 1 (y) for
x, y ran( ); but any such wellordering is in P(A A). Conversely, any well
ordering R on a subset B of A induces a unique injection : A with B = ran( )
and x Ry 1 (x) 1 (y) for x, y ran( ) by Mostowskis Theorem 3.21.
Therefore, as P(A A) is a set, F is a set by the appropriate axiom of replacement.
Let K F be such that 
or (i.e.,
  dom( ) = or  dom
(
)
=

)
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

Show that V0 = , V+1 = P(V ) for every , and V =


limit ordinal .

<

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 )

Show that every instance of (Fund ) is provable in ZFC.


3.5 Let be a formula of L in which exactly the variables x, y, v1 , . . . , v p (all
different from b) occur freely. The collection principle corresponding to ,
(Coll ) runs as follows.
v1 . . . v p ((xy ) (abx ay b )).

(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

4.1 Regular and Singular Cardinal Numbers


We know by Zermelos Theorem 3.23 that for each set x there is an ordinal such
that x , i.e., there is a bijection f : x .
Definition 4.1 Let x be a set. The cardinality of x, abbreviated by x, or Card(x), is
the least ordinal such that x .
Notice that Card(x) exists for every set x. Namely, let x . Then either =
Card(x), or else Card() is the least < such that x .
To give a few examples, Card(n) = n for every n ; Card() = = Card( +
1) = Card( + 2) = . . . = Card( + ) = . . . = Card( ) = . . . = Card( ) =
. . .1 We shall see more examples later.
Definition 4.2 An ordinal is called a cardinal iff = .
Obviously, is a cardinal iff there is some set x such that = Card(x). We shall
typically use the letters , , , . . . to denote cardinals.
By Cantors Theorem 1.3, if x is any set, then there is no surjection f : x
P(x). Therefore, if is a cardinal, then there is a cardinal > , and there is thus
also a least cardinal > which may also be identified as the least cardinal with
< P().
The Pigeonhole Principle says that if and are cardinals with > and if
f : , then f cannot be injective.
Definition 4.3 Let be a cardinal. The least cardinal > is called the cardinal
successor of , abbreviated by + . A cardinal is called a successor cardinal iff
there is some cardinal < with = + ; otherwise is called a limit cardinal.
All positive natural numbers are therefore successor cardinals, is a limit cardinal,
+ , ++ , . . . are successor cardinals, etc.
1

We here use the notation for ordinal arithmetic from Problem 3.6.

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_4,


Springer International Publishing Switzerland 2014

33

34

4 Cardinals

The cardinal successor + of a given cardinal my also be characterized as the


set of all ordinals of at most the same size as (cf. Problem 4.2).
As + exists for each cardinal , there are arbitrarily large successor cardinals.
But there are also arbitrarily large limit cardinals.

Lemma 4.4 Let X be a set of cardinals. Then X is a cardinal.

Proof By Lemma 3.7(b) we already know
We have
 that X is an ordinal.

 to show
that there is no < X such that X . Well, if < X , i.e., X , then
there is some X with , i.e., < . As is a cardinal, there is no surjection
from 
onto . But X gives X , so that there is also no surjection from
onto X .

+ ++
In particular, if is any cardinal,
then X = {, , , . . .} exists by the replacement schema and we have that
X is a limit cardinal > . There are therefore
arbitrarily large limit cardinals.

If X = is a set of ordinals (or cardinals), then wealso write sup(X ) for X
and min(X ) for the least element of X , i.e., min(X ) = (X ).
We may now recursively, i.e., by exploiting Theorem 3.13, define the -sequence
as follows. 0 = , the least infinite cardinal, and for > 0, = the least cardinal
such that > for all < .
The first infinite cardinals are therefore

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

Again, we use the notation from Problem 3.6.

4.1 Regular and Singular Cardinal Numbers

35

If is a cardinal and A is a set,


write [A] for 
{x A: Card(x) =
 then we
<

= < [A] and [A] = [A] . Trivially,


}, and we also write [A]
[A]Card(A) = P(A). It is not hard to verify that = Card([] ) for cardinals ,
(cf. Problem 4.4).
We now want to verify that = for every (cf. Theorem 4.6). For this
purpose we need the Gdel pairing function.
We define an ordering on OR OR as follows. We set (, ) ( , ) iff either
(a) max{, } < max{ , }, or else
(b) max{, } = max{ , }, and < , or else
(c) max{, } = max{ , }, = , and .
We claim that is a well-ordering on OR OR. We need to see that each non-empty
X OR OR has a -least element. Let X = be given, X OR OR. We let
X 0 = {(, ) X : ( , ) X max{, } max{ , }}; we let X 1 = {(, )
X 0 : ( , ) X 0 }; finally we let X 2 = {(, ) X 1 : ( , ) X 1
}. Obviously, X 2 contains exactly one element and it is -least in X .
It is easy to see that is set-like. Using Theorem 3.21 (cf. also Definition 3.22),
we may therefore let : OR OR OR be the transitive collapse of . I.e., is a
bijection such that (, ) ( , ) iff ((, )) (( , )) (where the latter
denotes the usual well-ordering on ordinals).
Notice that  ( ): ( ) is bijective for every . It is easy to
verify that ( ) for every . In what follows, we shall sometimes write ,
for ((, )). The map ,  , is called the Gdel pairing function.
Theorem 4.6 (Hessenberg) For every , = .
Proof We use the notation from the preceeding paragraphs. One easily shows that
(0 ) = 0 , so that  (0 0 ) witnesses that 0 0 0 , i.e., 0 0 = 0 .
Now suppose that there is some with > , and let us fix the least such
. We then must have > 0 and ( ) > . Say ((, )) = , where , < .
Let < be such that , < . Then ran(  (( + 1) ( + 1))) , so that
in particular there is a surjection f : ( + 1) ( + 1) . Now + 1 < ,
say Card( + 1) = , where < . We have = by the choice of
, so that there is a surjection g : , and hence also a surjection
g : ( + 1) ( + 1). But then f g : is surjective, contradicting

the fact that < and is a cardinal.
Hessenbergs Theorem 4.6 yields that cardinal addition and multiplication are
trivial.
Corollary 4.7 For all , , + = = max{,} .
Proof Assume without loss of generality that . Then
+ = ,
the last equality being true by Hessenbergs Theorem 4.6.

36

4 Cardinals

Cardinal exponentiation is a different matter.


Lemma 4.8 For all , 2 = P().
Proof 2 = the cardinality of the set of all functions f : 2 = {0, 1}, which is
the same as the cardinality of the set P().

Corollary 4.9 + 2 .
Cantors Continuum Hypothesis, abbreviated by CH, may now be restated as 20 =
1 (= +
0 ). The assertion that
2 = +1

(4.1)

is called the Generalized Continuum Hypothesis and is abbreviated by GCH. We


shall see that GCH as well as CH are consistent with ZFC (cf. Theorems 5.31 and
6.33).
Definition 4.10 Let be an ordinal. A function f : A is called cofinal in iff
for all < there is some a A such that f (a) . The cofinality of , written
cf(), is the least such that there is a cofinal f : .
Notice that cf() is defined for all , as the identity on is always cofinal in .
cf( + 1) = 1 for all , so that cf() is only interesting for limit ordinals .
For instance, cf() = = cf( + ) = cf( 3) = . . . = cf( ) = . . . =
cf( ).
The fact that = cf() is witnessed by a monotone function as follows. Let
= cf(). Let f : be cofinal. Define f : as follows: f ( ) =
sup{ f (): < } for < . Notice that in fact for every < , f ( ) < , as
otherwise f  would witness that cf() < = cf(). f is thus well-defined
and cofinal, and of course if , then f ( ) f ( ). If : ran( f )
is the monotone enumeration of ran( f ), then cf() and thus = cf(). Of
course, is then strictly monotone. is also continuous in the sense that for all limit
ordinals dom( ), ( ) = sup({( ) : < }).
Definition 4.11 Let be an ordinal. is called regular iff cf() = , and is called
singular iff cf() < .
Examples of singular cardinals are , + , etc., or more generally all where
is a limit ordinal with < . However, = does not imply that is regular,
cf. Problem 4.5.
Lemma 4.12 For every ordinal , cf() is regular.
Proof Let = cf(). We need to see that cf() = . Let f : be cofinal, and
let g: cf() be cofinal. By the above observation, we may and shall assume
that f is monotone. Consider f g: cf() . If < , then there is some
< with f () , and then there is some < cf() with g( ) . But then
f g( ) = f (g( )) f () by the monotonicity of f . I.e., f g is cofinal, so
that = cf().


4.1 Regular and Singular Cardinal Numbers

37

Lemma 4.13 Let be regular; then is a cardinal.


Proof Every bijection (or just surjection) f : is cofinal.

Corollary 4.14 For every ordinal , cf() is a regular cardinal.


Lemma 4.15 Let be an infinite successor cardinal. Then is regular.
Proof Let = + . Suppose that cf() < , i.e., cf() by Lemma 4.13. Let
f : be cofinal. Let (g : < ) be such that for each < , g : f ( )
is surjective. (Here we use AC, the axiom of choice.) Let h : be bijective
(cf. Theorem 4.6). We may then define a surjection F : as follows. Let
< . Let (, ) = h(), and set F() = g (). But because < and is a
cardinal, there cant be such a surjection.

We thus get that 0 , 1 , 2 , . . . are all regular, whereas cf( ) = < . +1
is again regular, etc.
Felix Hausdorff (18681942) asked whether every limit cardinal is singular.
This question leads to the concept of large cardinals, which will be discussed in
detail below and in later chapters, cf. Definitions 4.41, 4.42, 4.48, 4.49, 4.54, 4.60,
4.62, 4.68, and 10.76. They are ubiquitous is current day set theory.
We now want to look at ,  .
Notice that 1 = 1, but for every infinite cardinal , 2 (2 ) = 2 =

2 , i.e., 2 = . Therefore, = 2 for all infinite and 2 .


If is a limit cardinal, then we write 2< for sup< 2 . More generally, we write
<
for sup< .
Lemma 4.16 If is a limit cardinal, then 2 = (2< )cf() . In particular, if is a
limit cardinal with 2< = , then 2 = cf() .
Proof Let be an arbitrary limit cardinal, let f : cf() be cofinal, and let us
write i for f (i), where i < cf(). For i < cf(), let gi : P(i ) 2< be an
injection. We may define
: P() cf() (2< )
by letting (X )(i) = gi (X i ), where X and i < cf(). Obviously, is
injective. Therefore,
2 (2< )cf() (2 )cf() = 2cf() = 2 ,
so that in fact 2 = (2< )cf() .

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

i = the cardinality of the set of all functions g


with dom(g) = I and g(i) i for all i I.

This generalizes the earlier definitions of + and .


It is not hard to verify that if is a limit cardinal, then cf() may be characterized
as theleast such that there is a sequence (i : i < ) of cardinals less than with
= i< i (cf. Problem 4.6).
If Ai is a set for each i I (where I is any index set), then we write X iI Ai for
the set of all functions g with dom(g) = I and g(i) Ai for all i I . The axiom
of choice 
says that X iI Ai = provided that I = and Ai = for all i I . We
have that iI i = Card(X iI i ).
Theorem 4.20 (Knig) Let I = ,
and suppose
 that for every i I , i and i are
cardinals such that i < i . Then iI i < iI i .

Proof Let f : iI (i {i}) X iI i . We need to see that f is not surjective.
Let i I . Look at
{ i : i f (, i)(i) = } .

4.1 Regular and Singular Cardinal Numbers

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() > .

Proof Let I = , and let


i < 2 for all i
it suffices to show that iI i < 2 . Set i


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() .

I . In order to show that cf(2


 ) > ,

= 2 for all i I . Then iI i <

be cofinal, andwrite I =cf() and


I . Then iI i < iI i =


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 = .

In order to prove more powerful statements in cardinal arithmetic, we need the


concept of a stationary set.

4.2 Stationary Sets


Definition 4.23 Let A be a set of ordinals. A is called closed iff for all ordinals
, sup(A ) A. If is an ordinal, then A is unbounded in iff for all <
, (A )\ = . A is called -closed, where is an infinite regular cardinal, iff for all
ordinals with cf() = and such that A is unbounded in , = sup(A) A.
A is called club in iff A {} is closed and A is unbounded in . A is called -club
in , where is an infinite regular cardinal, iff A{} is -closed and A is unbounded
in .
A set A of ordinals is closed iff it is -closed for every , iff it is a closed subset of
sup(A) in the topology generated by the non-empty open intervals below . For any
set A of ordinals we usually denote by A the set of limit points of A, where is a
limit point of A iff for all < there is some A with < < . A is always
closed. Also, e.g., A is closed iff A A {sup(A)}.
For any , the cofinality of is the least size of a subset of which is unbounded
in (cf. Problem 4.7). If C is club, where cf() > , then C is also club in .
If : cf() is strictly monotone, continuous, and cofinal, then ran( ) is club
in , and if cf() > , then the set of limit points of ran( ) is club in and consists
of points of cofinality strictly less than cf().
Definition 4.24 Let X = be a set. F P(X ) is called a filter on X iff
(1) F = ,
(2) ab(a F b F a b F), and
(3) ab(a F a b X b F).
F is called non-trivial iff F. F is called an ultrafilter iff for every a X , either
a F or else X \ a F.
Let be a cardinal.
 Then F is called < -closed iff for all < and for all
{X i : i < } F, {X i : i < } F.
Notice that every filter is < -closed. If is a limit ordinal, then

4.2 Stationary Sets

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

Here, denotes ordinal multiplication.

42

4 Cardinals

If , are infinite regular cardinals with < , then the set


{ < : cf() = }
is stationary in (cf. Problem 4.12). This immediately implies that for any regular
2 , F is not an ultrafilter. We shall prove a stronger statement below, cf.
Theorem 4.33.
Definition 4.30 Let X = , and let F be a filter on X . Then we write
F + = {a X : b F b a = } .
The elements of F + are called the positive sets (with respect to F).
The stationary sets are therefore just the positive sets with respect to the club filter.
Lemma 4.31 Let be regular, and let F be a filter on . The following statements
are equivalent.
(a) F is normal.
(b) Let f : be such that Y = { < : f ( ) < } F + . There is then some
< and some X F + , X Y , such that f ( ) = for all X .
Proof (a) = (b): Let f be as in (b). If there is no < and X F + , X Y ,
such that f ( ) = for all X , then for every < we may pick some X F
such that f ( ) = for all X Y . (Here we use AC, the axiom of choice.) By
definition
Y < X = f ( ) .

(4.3)

By (a), < X F, so that by Y = { < : f ( ) < } F + , we may pick


some < X such that f ( ) < . This contradicts (4.3).
/ F, then
(b) = (a): Let X F, < . If < X


X F +.
Y = < :
/

<

Let f : be such that f ( ) < and


/ X f ( ) for all Y . By (b), there is
then some X F + , X Y , and some < such that
/ X for all X . But

X X = , as X F. Contradiction!
In the light of Lemma 4.28, Lemma 4.31, applied to the club filter, immediately
gives the following.
Theorem 4.32 (Fodor) Let be regular and uncountable, and let S be stationary. Let f : S be regressive in the sense that f ( ) < for all S. Then there
is a stationary T S such that f  T is constant.

4.2 Stationary Sets

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)

that A is club in and A S = for every S.


Proof Suppose first that S0 is stationary. By Theorem 4.32, there is then some sta Let
tionary S S0 and some regular < such that cf() = for all S.

us pick (A : S), where A is club in , otp(A ) = , and cf( ) < for all

A . Notice A S = for each S.


Now suppose S0 to be non-stationary, so that S1 is stationary. Let
S = { S1 : S1 is non-stationary} .
We must have that S is stationary. To see this, let C ( \ ( + 1)) be club, and let
C be the club of all limit points of C. If = min(S1 C ), then C is club
in , so that C is still club in , as > is regular; but (C ) S1 = ,
where A is club in and
We may thus pick (A : S),
so that C S.


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 < ,


S : i < cf() i >


is stationary in .
Proof Suppose first that S S0 , and let again < be such that cf() = for all
If Claim 4.35 fails, then for every i < there is some i < and some club
S.
Ci such that for all S Ci , i i . But then if = sup({i : i < }) <

and ( S i< Ci ) \ ( + 1), then i for all i < , so that A would be
bounded in . Contradiction!
If Claim 4.35 fails,
Now suppose that S S1 , so that cf() = for every S.
then for every i < there is some i < and some club Ci such that for all
S Ci , either i cf() or else i i . Let D be the club of all <
such that i < implies i < . By Lemma 4.28 we may pick
, S i< Ci D,

44

4 Cardinals

with < . If i < , then S Ci , so that i i ; but D, so that i < .

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.

Now fix i 0 < such that for all < ,

S : i 0 < cf() i0 >

is stationary in . Let us recursively define stationary sets S and ordinals for


< as follows.
Fix < , and suppose that S and have already been chosen for all < .
The set

S : i0 > sup <


is stationary in , and we may thus use Fodors Theorem 4.32 to pick some >
sup < and some stationary S S such that for all S , i0 = . Let us set
S = S and = .
The rest is straightforward.

The Singular Cardinals Hypothesis SCH cannot first fail at a singular cardinal of
uncountable cofinality:
Theorem 4.36 (Silver) Let be a singular cardinal of uncountable cofinality. If
SCH holds below , then it holds at , i.e., if cf() = 2cf() + for every < ,
then cf() = 2cf() + .
Proof Suppose first that 2cf() . Then < 2cf() , as cf(2cf() ) > cf() by
Corollary 4.21. I.e., + 2cf() , and hence cf() (2cf() )cf() = 2cf() =
2cf() + cf() . Therefore, SCH holds at .
We may thus assume that 2cf() < . Let C be club in with otp(C) = cf(),
let C be the set of all limit points of C. Let (i : i < cf()) be the monotone
enumeration of C \ (2cf() )+ . As cf(i ) < cf() for every i < cf() (as being
witnessed by C i ), Lemma 4.22 gives that icf() = i+ for each i < cf(). So
for each i < cf(), we may pick a bijection gi : [i ]cf() i+ .
We now have to count []cf() . To each X []cf() we may associate a function
f X : cf() by setting f X (i) = gi (X i ). Obviously, X  f X is injective.
If X, Y []cf() , then we shall write X Y iff {i: f X (i) f Y (i)} is stationary.
We must have X Y or Y X for any two X, Y []cf() .
Claim 4.37 Let X []cf() . Then Card({Y []cf() : Y X }) .
Proof Let us fix X []cf() for a moment. For each i < cf(), let us pick an injection
gi : f X (i) + 1 i . If Y X , then the set SYX = {i < cf(): f Y (i) f X (i)} is
stationary, and we may look at
FYX : SYX ,

4.2 Stationary Sets

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 < ,

f i [X ]< X for all i < .


If S + , then S \ [ + ] . It is easy to verify that if S + is stationary in
in the sense of Definition 4.29, then S \ is stationary in [ + ] in the sense of
Definition 4.39.
We may call a set X [ ]< unbounded in [ ]< iff for all Y [ ]< there
is some X X with X Y , and we may call X [ ]< closed in [ ]< iff
such that < and X i X and X i X j for i < j < ,

for all (X i : i < )



< closed and unbounded (club) in [ ]<
i< X i X . We may then call C [ ]
iff C is both unbounded and closed in [ ]< .
A set X [ ]< is stationary in [ ]< iff X C = for all C which are club
+
in [ ]< . Then S [ ] is stationary in [ ]< iff S is stationary in the sense of
Definition 4.39, cf. Problem 4.15.
+

46

4 Cardinals

4.3 Large Cardinals


Definition 4.40 A cardinal is called a strong limit cardinal iff for all cardinals
< , 2 < .
Trivially, every strong limit cardinal is a limit cardinal. 0 is a strong limit cardinal,
and if is an arbitray cardinal, then

sup

, 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 .

4.3 Large Cardinals

47

A set c T is a chain in T iff for all s, t c, s <T t or s = t or t <T s. A chain


c T is called a branch through T iff for all s c and t <T s, t c. A branch b
through T is called maximal iff there is no branch b  b through T , and a branch b
through T is called cofinal iff for all < ht(T ) there is some s b with lvT (s) = .
A set a T is an antichain iff for all s, t a with s = t neither s <T t nor
t <T s.
Definition 4.44 Let be an infinite cardinal, and let (T, <T ) be a tree. We call
(T, <T ) a -tree iff the following hold true.
(1) ht(T ) = ,
(2) there is a unique r T with lvT (r ) = 0 (the root of T ),
(3) for every s T and every > lvT (s), < , there is some t T with s <T t
and ht(T ) = ,
(4) for every s T there are t, t T , t = t , with s <T t, s <T t and lvT (t) =
lvT (t ) = lvT (s) + 1, and
(5) for every < , Card({s T : lvT (s) = }) < .
A -tree (T, <T ) is called -Aronszajn iff there is no cofinal branch through T .
A -tree (T, <T ) is called -Souslin iff T has no antichain of size .
A -tree (T, <T ) is called -Kurepa iff T has at least + cofinal branches.
Notice that if (T, <T ) is -Souslin, then (T, <T ) is -Aronszajn. (Cf. Problem
4.22.)
We may turn any tree (T, <T ) with properties (1) and (5) from Definition 4.44 into
a -tree without adding cofinal branches or antichains of size as follows, provided
that be regular.
Lemma 4.45 Let 0 be a regular cardinal. If there is a tree with properties (1)
and (5) from Definition 4.44 which has no cofinal branch and no antichain of size ,
then there is a -Souslin tree.
Proof Let (T, <T ) be a tree with properties (1) and (5) from Definition 4.44. Let
T0 = {s T : < ( > lvT (s) t T (lvT (t) = s <T t))} .
Then T0 = (T0 , <T  T0 ) is a tree with property (5) of Definition 4.44. Of course,
T0 T and lvT0 (s) = lvT (s) for all s T0 .
Suppose that (1) of Definition 4.44 failed, and let = ht(T0 ) < . For each s T
with lvT (s) = we must then have that (s) = sup({lvT (t) : s <T t}) < , so that
sup({(s) : s T lvT (s) = }) < , as is regular and T satisfies (5). But then
T cannot have satisfied (1). Contradiction!
Therefore, T0 satisfies (1) and (5). Let us show that T0 satisfies (3). Let s T0 and
> lvT0 (s), < . As s T0 , for every > we may pick some t T , s <T t
with lvT (t ) = , and we may let u T be unique such that s <T u <T t ,
lvT (u ) = . As is regular and T satisfies (5), there is some cofinal X such
that u = u for all , X . Write u = u , where X . We must then have
that u T0 , where s <T u and lvT (u) = . This shows that T0 satisfies (3).

48

4 Cardinals

Now pick r T0 with rk T0 (r ) = 0, and let


T1 = {s T0 : r <T s} .
Then T1 = (T1 , <T  T1 ) is a tree which satisfies (1), (2), (3), and (5) from Definition
4.44. We are left with having to arrange (2).
Now let us set
T2 = {s T1 : t <T s r T1 (t T r lvT1 (r ) lvT1 (s) r T s)}.
Then (T2 , <T  T2 ) is again a tree.
T2 trivially satisfies (2). As for (5), let < . If L = {s T2 : lvT2 (s) = } has
size , then L cannot be an antichain in T , so that there are s, t L with s <T t
or t <T s. But then lvT2 (s) = lvT2 (t). Contradiction! So T2 satisfies (5).
As for (1) and (3), let us fix s T2 . For each > lvT1 (s), let us pick some t T1
such that lvT1 (t ) = . Let u T t be such that u T2 and


r T1 u T r lvT1 (r ) lvT1 (t ) r T t .


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.

The following Lemma is in some sense a special case of Lemma 11.68.


Theorem 4.47 (Aronszajn) Let be an infinite cardinal with < = . There is
then a + -Aronszajn tree.
0
= 0 , so that Theorem 4.47 yields the existence of an 1 Notice that <
0
Aronszajn tree. As cf() > by Corollary 4.21, < = implies that is a
regular cardinal. By Theorem 4.47, if there is no 2 -Aronszajn tree, then CH fails.
Proof of Theorem 4.47. Let

U = s : < + s is injective \ ran(s) =


We construe U as a tree by having it ordered under end-extension. The tree T which
we are about to construct will be a subtree of U and also closed under initial segments
and ordered by end-extension. Notice that U cannot have any cofinal branch, as this
would yield an injection from + into , so that the tree T we are about to construct
cannot have a cofinal branch either.

4.3 Large Cardinals

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

for every i < cf(). Then s T+1 .



Well, T1 = {}, and T = < T for limit ordinals < + . If = +1 < + ,
where = + 1 is a successor ordinal, then we just let T consist of all injections
s such that s  T . Notice that (1) through (4) for the tree constructed so
far follows from (1) through (4) for the ealier levels of the tree.
Now suppose that = + 1 < + , where is a limit ordinal, and T already
has been constructed.
Let us first assume that cf() < . We then let T+1 consist of all s which
need to be there in order to satisfy (4). Notice that there are cf() many clubs
in of order type cf(), and for each such club C there are cf() = possible
choices of s such that for all C, s  T +1 . Hence T+1 , and
(3) is maintained. (1) is ensured by the fact that for s T+1 as being given by (4),
ran(s) B = .
Let us now assume that cf() = . Let us fix C club in with otp(C) = , and
let (i : i < ) be the monotone enumeration of C. To each s T we shall assign
some t (s) U with dom(t (s)) = as follows. By (1), \ ran(s) has
size , and we
may hence pick sets Ai , i < , and B such that ran(s) A0 , = i< Ai B,
Ai A j = , Ai B = , and B = Ai = for i, j < , i = j. Using (2) and (4),
we may construct some t : extending s such that for every i < ,
t  i Ti +1 t i


ji+1

Aj.

50

4 Cardinals

Let us write t (s) for this t. We may then set


T+1 = {t (s) : s T } .
It is easy to verify that (1) through (4) remain true.

It is much harder to construct a + -Souslin tree. For = 0 , this will be done


from a principle called (cf. Lemma 5.36), and for 1 we shall need + and
 (cf. Lemma 11.68).
Definition 4.48 A cardinal is said to have the tree property iff there is no Aronszajn tree. A cardinal is called weakly compact iff is inaccessible and
has the tree property.
The following large cardinal concept will be needed for the analysis of the combinatorial principle , cf. Definition 5.37.
Definition 4.49 Let be a regular uncountable cardinal. Then R is called
ineffable iff for every sequence (A : R) such that A for every R
there is some S R which is stationary in such that A = A whenever ,
S, .
Trivially, if R is ineffable, then R is stationary. On the other hand, if < is
an infinite regular cardinal, then { < : cf( ) = } is stationary but not ineffable.
If itself is ineffable, then is weakly compact and the set
{ < : is weakly compact }
is stationary in (cf. Problem 4.24). On the other hand, every measurable cadinal
(cf. Definition 4.54 below) is ineffable, cf. Lemma 4.58.
The study of (non-trivial) elementary embeddings between transitive structures
plays a key role in set theory.
Definition 4.50 Let M, N be transitive sets or classes. We say that : M N is
an elementary embedding from M to N iff (ran( ); ) is an elementary substructure
of (N ; ), i.e., if for all formulae of the language of set theory and for all a1 , . . .,
ak M,
(M; ) |= (a1 , . . . , ak ) (N ; ) |= ((a1 ), . . . , (ak )).

(4.4)

The elementary embedding is called non-trivial iff there is some x M with


(x) = x. The least ordinal with () = (if it exists) is called the critical point
of , abbreviated as crit( ).
Notice that if : M N is an elementary embedding between transitive sets
or classes, then () for all M. This is because otherwise there would be
a least with () < ; but then (()) < (), a contradiction! Therefore, if
crit( ) exists, then it is the least M with () > .

4.3 Large Cardinals

51

We will mostly be concerned with non-trivial elementary embeddings : V


M from V to M, where M is some transitive class. In this situation, M is of course
also a model of ZFC, as by (4.4), the validity of any given axiom of ZFC is moved
up from V to M.
Definition 4.51 An inner model is a transitive proper class model of ZFC.
Let M be a transitive model of (some fragment of) ZFC. For terms like V , rk(x),
+ , etc., we shall denote by (V ) M , rk M (x), +M , etc., the respective objects as
defined in M, e.g. +M = the unique such that
(M; ) |= is the cardinal successor of ,
etc.
Lemma 4.52 Let : V M be a non-trivial elementary embedding, where M is
a transitive class. The following hold true.
(a) (V ) = (V() ) M and (rk(x)) = rk((x)) for all and all x.
(b) There is some ordinal with ( ) = .
Let be the least ordinal with ( ) = . The following hold true.
(c) is continuous at every ordinal of cofinality less than , i.e., if is a limit
ordinal and cf() < , then () = sup( ).
(d) is regular and uncountable.
(e) (V+1 ) M = V+1 .
(f) is an inaccessible cardinal.
(g) is a Mahlo cardinal.
(h) is weakly compact.
Proof (a) This is easy.
(b) Let be least such that there is some x with rk(x) = and (x) = x. We
show that ( ) = .
Suppose that ( ) = , i.e., = rk(x) = rk((x)). Then by the choice of ,
(y) = y for all y x (x). This means that
y x y = (y) (x)
for all y x (x), so that (x) = x. Contradiction!
(c) Let f : cf() be cofinal in , where cf() < . Then ( f ) : cf()
() is cofinal in () by the elementarity of . But ran(( f )) , so that
() = sup( ).
(d) If were singular, then () = using (c). It is easy to see that > 0 .
(e) (V ) M = V follows from (a) and the choice of . Therefore, if X V , then
X = (X ) V M, so that V+1 (V+1 ) M . Trivially, (V+1 ) M V+1 .
(f) If is not inaccessible, then by (d) we may choose < and a surjective
f : P() . By (e) and the elementarity of , ( f ) : P() () is

52

4 Cardinals

surjective. However, for all X , f (X ) < and thus f (X ) = ( f (X )) =


( f )((X )) = ( f )(X ). This shows that in fact ( f ) = f , so that ( f ) cannot be
a surjection onto () > after all. Contradiction!
(g) Let C be club in . Then (C) is club in () by the elementarity of .
Also, (C) = C. Therefore, (C). By (e),
(M; ) |= is inaccessible,
so that
(M; ) |= ( is inaccessible and (C)).
By the elementarity of ,
(V ; ) |= ( is inaccessible and C).
As C was arbitrary, this shows that is a Mahlo cardinal.
(h): Let (T, <T ) be a -tree. Writing ((T ), <(T ) ) = ((T, <T )), we have that
(M; ) |= ((T ), <(T ) ) is a ()-tree.
Let s (T ) be such that (M; ) |= rk (T ) (s) = , and set


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() = () =

4.3 Large Cardinals

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)

We claim that U is a normal < -closed uniform ultrafilter on . Well, { }


/ U
for any < , and U , as is the critical point of . U is easily seen to be
an ultrafilter, as (X Y ) = (X ) (Y ) for all sets X , Y , (X ) (Y ) for all
X Y , and () = (X ) ( \ X ) for all X . Moreover, if < , then
({X i : i < }) = {(X i ) : i < } for all {X i : i < }, which yields that U is
< -closed. Hence U witnesses that is a measurable cardinal.
It remains to be shown that U is normal. Let (X i : i < ) be such that X i U
for all i < . We need to see that i< X i U , i.e., (i< X i ). Writing
(Yi : i <()) = ((X i : i < )), we have that Yi = (X i ) for every i < , so
that i< Yi . This just means that (i< X i ).
(2) = (1) is trivial.
(1) = (3): This will be shown by an ultrapower construction which is wellknown from model theory and which will be refined later (cf. the proof of Theorem
10.48).

54

4 Cardinals

Let U be a < -closed uniform ultrafilter on . We aim to construct an inner


model M and an elementary embedding : V M with critical point . We shall
first construct M.
If f , g V , we write f g iff { < : f ( ) = g( )} U . It is easy to
verify that is an equivalence relation. For any f V , we write [ f ] for the equivalence class of f , massaged by Scotts trick, i.e., for the set {g V : g
f h V (h f rk(h) rk(g)}. If f , g V , then we write [ f ]E[g] iff
{ < : f (x) g(x)} U . It is easy to check that E is extensional. Also, for all
f V , {[g] : [g]E[ f ]} is a set.
Let us write ult(V ; U ) for the structure ({[ f ] : f V }; E). We may define a
map from V into ult(V ; U ) by setting
(x) = [cx ],
where cx : {x} is the constant function with value x. The following statement
shows that is elementary and it is referred to as the os Theorem.
Claim 4.56 (os Theorem) Let (v1 , . . . , vk ) be a formula, and let f 1 , . . ., f k V .
Then
ult(V ; U ) |= ([ f 1 ], . . . , [ f k ])
{ < : V |= ( f 1 ( ), . . . , f k ( ))} U.
Proof of Claim 4.56 by induction on the complexity of : The atomic case is immediate from the definition, as for f , g V we have that ult(V ; U ) |= [ f ] [g] iff
[ f ]E[g] iff { < : f (x) g(x)} U and ult(V ; U ) |= [ f ] = [g] iff [ f ]E[g] iff
{ < : f (x) = g(x)} U .
As for the sentential connectives, it suffices to discuss and .
As for , if (v1 , . . . , vk ) and (v1 , . . . , v ) are formulae, if f 1 , . . ., f k ,
g1 , . . . , g V , and if the Claim holds for and , then ult(V ; U ) |=
(([ f 1 ], . . . , [ f k ]) ([g1 ], , [g ])) iff ult(V ; U ) |= ([ f 1 ], . . . , [ f k ]) and
ult(V ; U ) |= ([g1 ], . . . , [g ]) iff { < : V |= ( f 1 ( ), . . . , f k ( ))} U and
{ < : V |= (g1 ( ), . . . , g ( ))} U iff { < : V |= (( f 1 ( ), . . . , f k ( ))
(g1 ( ), . . . , g ( )))} U , as U is a filter.
As for , if (v1 , . . . , vk ) is a formula, if f 1 , . . ., f k V , and if the Claim
holds for , then ult(V ; U ) |= ([ f 1 ], . . . , [ f k ]) iff ult(V ; U ) is not a model of
/ U iff { < : V |=
([ f 1 ], . . . , [ f k ]) iff { < : V |= ( f 1 ( ), , f k ( ))}
( f 1 ( ), . . . , f k ( ))} U , as U is an ultrafilter.
Let us finally suppose that (v1 , . . . , vk ) v0 (v0 , v1 , . . . , vk ) for some formula for which the Claim holds true. Let f 1 , . . ., f k V . If ult(V ; U ) |=
v0 (v0 , [ f 1 ], . . . , [ f k ]), then there is some f 0 V such that
ult(V ; U ) |= ([ f 0 ], [ f 1 ], . . . , [ f k ]).

4.3 Large Cardinals

55

By induction, { < : V |= ( f 0 ( ), f 1 ( ), . . . , f k ( ))} U , which also gives


that { < : V |= v0 (v0 , f 1 ( ), . . . , f k ( ))} U , as U is a filter.
Conversely, let us assume that { < : V |= v0 (v0 , f 1 ( ), . . . , f k ( ))} U .
By the replacement schema in V , there is a set a such that for all < , if there
is some x with V |= (x, f 1 ( ), . . . , f k ( )), then there is some x a with V |=
(x, f 1 ( ), . . . , f k ( )). Let <a be a well-ordering of a. Let us define f 0 : V
as follows.

the <a -smallest x a with


f 0 ( ) = V |= (x, f 1 ( ), . . . , f k ( )) if some such x exists,

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

because U is < 1 -closed, and then if

{ < : f n+1 ( ) f n ( )},

. . . 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

Finally, if id denotes the identity function on , then for all < , = U () =


1 ([c ]) < 1 ([id]) by the uniformity of U . Also 1 ([id]) < 1 ([c ]) =

U (). Hence is indeed the critical point of U .
In the situation of the proof of (1) = (3) of Thorem 4.55, we usually also write
ult(V ; U ) for the inner model which was called N there. Let = 1 ([id]). We
must have that
1 ([ f ]) = U ( f )( ) for all f : V.

(4.6)

This is because U ( f )( ) may be written in a cumbersome way as




1 ([c f ]) 1 ([id]) ,
which with the help of os Theorem is easily seen to be equal to 1 ([ f ]). In
particular,
N = ult(V ; U ) = {U ( f )( ) : f : V } .

(4.7)

is often called the generator of U . It is also easy to see that for X ,


X U (X ).

(4.8)

Now let : V M be any elementary embedding with critical point , where


M is an inner model. Let U = U be derived as in (4.5) in the proof of (3) = (2)
of Theorem 4.55, i.e., U = {X : (X )}. Let U : V ult(V ; U ) be as
constructed in the proof of (1) = (3). We may then define a factor map
k : ult(V : U ) M
by setting k(U ( f )()) = ( f )() for f : V . This map k is well-defined and
elementary because we have that
ult(V ; U ) |= (U ( f 1 )(), . . . , U ( f k )())
{ < : V |= ( f 1 ( ), . . . , f k ( ))} U
({ < : V |= ( f 1 ( ), . . . , f k ( ))}) =
{ < () : M |= (( f 1 )( ), . . . , ( f k )( ))}
M |= (( f 1 )(), . . . , ( f k )()).

V
U

M
k

ult(V ;U)

4.3 Large Cardinals

57

If : V M is itself an ultrapower map, i.e., = U for some < -closed


uniform ultrafilter U on and if = [g] in the sense of the ultrapower ult(V ; U ),
then
X U g 1 X U

(4.9)

for every X . We will have that U is normal iff U = U iff { < : g( ) =


} U . (Cf. problem 4.26.)
We remark that M. Gitik has shown that AC is needed to prove (1) = (2) in
the statement of Theorem 4.55 (cf. [7]).
Definition 4.57 Let M be a transitive model of ZFC, and let M |= U is a measure.
Then we shall ambiguously write ult(M; U ) for ult(V ; U ) from the proof of Theorem
4.55 as defined inside M or for its transitive collapse. Also, we shall ambiguously
write UM for the map from the proof of Theorem 4.55 or for the map from the
proof of Theorem 4.55. ult(M; U ) is called the ultrapower of M by U , and UM is
called the associated ultrapower embedding.
By (4.7), applied inside M, if U is a measure on , then

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)

This is because if F : []<


 and for each n < , X n U is such that
F  [X n ]n is constant, then n< X n U is as desired.

58

4 Cardinals

We prove (4.11) by induction on n. For n = 0, (4.11) is trivial. So let us assume


(4.11) for n and show it for n + 1.
Let F : []n+1 be given. Let
: V U M = ult(V ; U )
be the ultrapower map given by U , where M is an inner model, = crit( ), and for all
X , X U iff (X ). Let F : []n be defined by F (a) = (F)(a
{}) for a []n . By the inductive hypothesis, there is some < and some
X U such that F [X ]n = { }. That is, for every a [X ]n , (F)(a {}) = ,
or equivalently,
X a = { < : > max(a) F(a {}) = } U.

(4.12)

By the normality of U, we may pick some Y U , Y X , such that for every


X,

Xa.

(4.13)

a[X ]n

We then have for a {} [Y ]n+1 with > max(a) that Y , hence, as Y X ,


X a by (4.13), and so F(a {}) = by (4.12). We have shown that F [Y ]n+1 =
{ }, where Y U .

Definition 4.60 Let be a cardinal, and let > . Then is called -strong iff
there is some non-trivial elementary embedding : V M, where M is an inner
model and crit( ) = , such that V M. is called strong iff is -strong for all
> .
Lemma 4.61 If is measurable, then is ( + 1)-strong. If is ( + 2)-strong,
then is measurable and there exists a measurable cardinal < .
Proof The first part immediately follows from Lemma 4.52. As for the second part,
let be ( + 2)-strong, and let
: V M
be an elementary embedding, where M is an inner model, crit( ) = , and V+2
M. By Lemma 4.55, is measurable, and there is hence some < -closed uniform
ultrafilter U on . But U V+2 M, which gives that
M |= < ()U0 (U0 is a < -closed uniform ultrafilter on > 0 ).
By the elementarity of , this gives that
V |= < U0 (U0 is a < -closed uniform ultrafilter on > 0 ).

4.3 Large Cardinals

59

There is thus a measurable cardinal < .

Definition 4.62 Let be a cardinal, and let be a regular cardinal. Then is


called -supercompact iff there is some non-trivial elementary embedding : V
M, where M is an inner model, crit( ) = , and () > , such that M M.
is called supercompact iff is -supercompact for all regular .
Measures cannot witness suprcompactess, cf. Problem 4.27. We will see later, cf.
Lemmas 10.58 and 10.62, that extenders may be used to witness that is strong
or supercompact in much the same way as a measure witnesses that a given cardinal
is measurable.
Lemma 4.63 If is measurable, then is -supercompact. If is 2 -supercompact,
then is measurable and there is a measure U on such that
{ < : is measurable } U.
Proof We use Theorem 4.55. Let be measurable, let U be a normal measure on ,
and let
: V U ult(V ; U ) = M
be the ultrapower embedding, where we assume M to be transitive. We need to see
that M M. Let (xi : i < ) be a sequence with xi M for all i < . Say
xi = ( f i )(), where f i : V , for i < . Let us define g : V as follows.
For each < , g( ) : V and for i < , g( )(i) = f i ( ). We then get that
(g)() : M (here we use Problem 4.26), and for every i < ,
{ < : g( )(i) = f i ( )} = \ (i + 1) U,
so that (g)()(i) = ( f i )() = xi . Thus (xi : i < ) = (g)() M, as desired.
If is 2 -supercompact, then we may pick some
: V M,

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

Lemma 4.65 Let be -supercompact, where is a regular cardinal. There is


then a < -closed uniform weakly normal ultrafilter on .
Proof Let
: V M
be an elementary embedding, where M is an inner model, crit( ) = , () > ,
and M M. Let us set
U = {X : sup( ) (X )}.

(4.14)

It is not hard to verify that U is a < -closed uniform ultrafilter on .


To show that U is weakly normal, let f : . As U is an ultrafilter, U + = U .
If
sup( ) ({ < : f ( ) < }) = { < () : ( f )( ) < },
then we may pick < such that
( f )(sup( )) < (),
so that

sup( ) ({ < : f ( ) < }).




U is thus weakly normal.

Theorem 4.66 (Solovay) Let be supercompact. Then < = for every regular
cardinal .
Proof Let us fix , let again

: V M

be an elementary embedding, where M is an inner model, crit( ) = , () > ,


and M M. Let U be the < -closed uniform weakly normal ultrafilter on which
is given by (4.14).
Let us write
S = { < : cf() < } .
Notice that S U . This is because M M yields that M |= cf(sup ) = ,
which together with < () then gives that sup { < () : M |= cf() <
()} = (S).
For S, let us pick C cofinal in with otp(C ) = cf().
Let < be arbitrary. Because S U and U is uniform,
S = { S : C \ = } U.
Let f : S be defined by

4.3 Large Cardinals

61

f () = the least element of C \


Then f is a regressive function, and because U is weakly normal there is
for S.
some > , < , such that


S : f () < U.

We have shown that


< < ( > { S : C [, ) = } U ).

(4.15)

By (4.15), there is now a continuous and strictly increasing sequence ( : < )


such that 0 = 0 and


< S : C [ , +1 ) = U.
Let us write

(4.16)


I = < : C [ , +1 ) =

for S, so that I C = cf() < for every S.


Let X []< . For all X , { S : I } U by (4.16) and the definition
of I . Therefore, as X < and U is < -closed,
{ S : X I } = { S : X I } U.

(4.17)

In particular, there is some S such that X I .


We have shown that

P(I ),
[]<
S

so that < 2< = < , i.e., < = .

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

Lemma 4.69 Suppose that is 2 -supercompact. Then is subcompact.


Proof Let A H + , and let

: V M

be such that M is an inner model, crit( ) = , and


(H + ) M = H + M and  H + M, and therefore

2 M

M. We have that

M |= < ()B H+ ( : (H+ ; , B) (H()+ ; , (A)) is elementary


crit( ) = ).
By the elementarity of ,
V |= < B H+ ( : (H+ ; , B) (H + ; , A) is elementary crit( )
= ).
We have shown that is subcompact.

Amazing results in cardinal arithmetic were obtained by Saharon Shelah via


his pcf-theory, cf. e.g. [1]

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

4.8. Show that 1 = 0 21 . Show also that if < 1 , then 1 = 0 21 .


4.9. Use the recursion theorem 3.13 to show that there is a sequence ( : OR)
such that 0 = 0 , +1 = 2 for all , and  = sup<  for every limit
ordinal . Show that Card(V+ ) =  for every .
4.10. (a) Let be an infinite cardinal. Show that H is a set. [Hint. Show e.g. that
H V by induction on .] Also show that Card(H ) = 2< .
(b) Show that HF = V and (HF;  HF) |= ZFC .
(c) (W. Ackermann)
Let us define E A as follows. n E A m iff: if

i
m =
i ki 2 , where ki {0, 1} for all i, then kn = 1. Show that
(; E A )
= (HF; ).
(d) Show that if is uncountable and regular, then (H ;  H ) |= ZFC .
4.11. (A. Tarski) Let X be a set, and let F be a filter on X . Show that there is
an ultrafilter U on X with U F. [Hint. Use the Hausdorff Maximality
Principle 2.11.]
4.12. Show that if , are infinite regular cardinals with < , then the set
S = { < : cf() = }
is stationary in . Show also that if T S is stationary in and C is
-club in , then T C = .
4.13. Let S 1 be stationary, and let < 1 . Show that S has a closed subset of
order type .
4.14. Let be regular and uncountable, and let R be stationary. Let (U ;
, A1 , . . . , An ) be a model such that U is transitive and U . Show that
there is some X (U ; , A1 , . . . , An ) such that X R. Show also that
if Card(U ) = and f : U is surjective, then { < : f (U ;
, A1 , . . . , An )} is club in .
4.15. Show that set X [ ] is stationary in [ ] iff X C = for all C which
are club in [ ] according to the definition on p. 44.
 

= iI (i ).
4.16. Show that
iI i
4.17. Let be a singular cardinal of uncountable cofinality.
(1) Suppose that there is some cardinal < with cf() = and SCH holds
for every singular (, ), i.e., if is a singular cardinal, < < ,
then cf() = 2cf() + . Show that SCH holds at , i.e., cf() =
2cf() + .
(2) Suppose that cf() < for all < , and SCH holds on a stationary set
below , i.e., { < : cf() = 2cf() + } is stationary in . Show that
SCH holds at , i.e., cf() = 2cf() + .
4.18. Let be a singular cardinal of uncountable cofinality. If { < : 2 = + }
is stationary in , then 2 = + .

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

Show that s[]< X s U .


4.27. Let be a measurable cardinal.
(a) Let U be a measure on . Show that +ult(V ;U ) = +V and that 2 <
UV () < ((2 )+ )V . [Hint: If < UV (), then is represented by
some f : , and there are 2 functions from to .] Conclude that
U
/ ult(V ; U ).
(b) Let U , U be measures on . We define U <M U iff U ult(V ; U ).
ult(V ;U )
Show that UV () = U
() < UV (). Conclude that <M is wellfounded, and that the rank of <M is always less than or equal to (2 )+ .
<M is called the Mitchell order.
4.28. Let be a measurable cardinal, let U be a measure on , and let U : V U
M be the ultrapower map, where M is an inner model. Let > be a
cardinal, cf() = , and < for every < . Show that U () = .
4.29. (Magidor) Show that if is supercompact, then for every > there are
< < together with an elementary embedding : V V such
that crit( ) = and () = . [Hint: Let : V M, where M is an
inner model, crit( ) = , and V M M. Show that in M, there is some
: V (V() ) M such that crit( ) = and () = (). Pull this
statement back via .]
4.30. Let be supercompact. Show that for every cardinal there is a < closed ultrafilter U on []< such that {a}
/ U for all a []< , {a
<
[] : a} U for all < , and if (A : < ) is such that A U
for all < , then there is some A U such that whenever a A,
a A . [Hint. Let U be derived from morally as in (4.14).]
Problems 10.21 and 10.22 will show that the conclusions of Problems 4.29
and 4.30 actually both characterize the supercompactness of .
4.31. Use the necessary criterion for supercompactness provided by Problem 4.29
to show that if is supercompact, then is subcompact (cf. Lemma 4.69) and
in fact there is a measure U on such that { < : is subcompact } U .

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.

5.1 The Constructible Universe


Definition 5.1 A formula of L is called 0 (or 0 , or 0 ) iff is contained in
each set for which the following hold true.
(a) Every atomic formula is in ,
(b) if 0 , 1 are in then so are 0 , (0 1 ), (0 1 ), (0 1 ), and
(0 1 ), and
(c) if is in and x, y are variables, then x(x y ) and x(x y ) are
in .
For n \{0}, a formula of L is called n iff is of the form
x1 . . . xk ,
where x1 , . . . , xk are variables and is n1 , and is called n iff is of the form
x1 . . . xk ,
where x1 , . . . , xk are variables and is n1 .
If M is a transitive set or class and is a sentence of L , then we write
M |= for (M;  M) |= (where  M = M 2 ). If is a formula, being
(x1 , . . . , xk ) with all free variables shown, and if a1 , . . . , ak M, then we write
M |= (a1 , . . . , ak ) for the assertion that holds in (M;  M) for an assignment
which sends vl to al (1 l k).
R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_5,
Springer International Publishing Switzerland 2014

67

68

5 Constructibility

We have the following absoluteness properties.


Lemma 5.2 Let M be transitive, let be a 0 -formula, and let a1 , . . . , ak M.
Then
M |= (a1 , . . . , ak ) V |= (a1 , . . . , ak ).
Proof by induction on the complexity of . Let us only consider the case where is
of the form x0 x1 (x0 , x1 , . . . , xk ).
=: If V |= (a1 , . . . , ak ), let a a1 be such that V |= (a, a1 , . . . , ak ). As
M is transitive, a a1 M gives a M, and V |= a a1 (a, a1 , . . . , ak ) gives
M |= a a1 (a, a1 , . . . , ak )
by the inductive hypothesis, and so M |= (a1 , . . . , ak ).
=: If M |= (a1 , . . . , ak ), let a M be such that M |= a a1
(a, a1 , . . . , ak ). Then
V |= a a1 (a, a1 , . . . , ak )
by the inductive hypothesis, and so V |= (a1 , . . . , ak ).

This proof also shows the following.


Lemma 5.3 Let M be transitive, let (v1 , . . . , vk ) be a 1 -formula,
let (v1 , . . . , vk ) be a 1 -formula, and let a1 , . . . , ak M. Then is upward
absolute, i.e.
M |= (a1 , . . . , ak ) = V |= (a1 , . . . , ak ),
and is downward absolute, i.e.
V |= (a1 , . . . , ak ) = M |= (a1 , . . . , ak ).
Definition 5.4 Let T be a theory in the language L , and let be a formula of L .
Then is called 1T iff there are L -formulae and such that
is 1 ,
is 1 , and
T 
Lemma 5.3 immediately implies:
Lemma 5.5 Let T be a theory in the language of L , and let be a formula of L
which is 1T . Let M be a transitive model of T . Then is absolute between V and
M, i.e.,
V |= (a1 , . . . , ak ) M |= (a1 , . . . , ak )
for all a1 , . . . , ak M making an assignment to all the free variables of .

5.1 The Constructible Universe

69

The following Lemma expresses the absoluteness of well-foundedness.

Lemma 5.6 The statement1 R is a well-founded relation is ZFC


. In particular,
1
if M is a transitive model of ZFC such that R M is a binary relation, then
V |= R is well-founded M |= R is well-founded.

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)

M is a model of (Ext), the axiom of extensionality.


M is a model of (Fund), the axiom of foundation.
If M then M is a model of (Inf), the axiom of infinity.
If M is closed under x, y  {x, y} (i.e., a, b M = {a, b} M), then M is
a model of (Pair), the pairing
 axiom.

(5) If M is closed under x  x (i.e., a M = a M), then M is a model
of (Union), the axiom of union.
Proof We shall use Lemma 5.2.
(1) The axiom of extensionality holds in V and it is 1 , as it says
xy((z x z y z y z x) x = y).
(2) The axiom of foundation holds in V and is 1 , as it says
x(x = y x y x = ).
Notice that x = can be written as y x y = y, and y x = can be written
as z y z x.
(3) The axiom of infinity is 1 . It says that x (x), where (x) is
x y x z x z = y {y}.
Here, x can be written as y x z y z = z, and z = y {y} can be
written as
u z (u y u = y) u y u z y z.
Thus (x) is 0 , V |= () holds, and thus M |= () holds provided that
M.
1

Recall from p. 12 that ZFC is ZFC without the power set axiom.

70

5 Constructibility

(4) The pairing axiom is of the form


x y z z = {x, y},
where z = {x, y} can be written as x z y z u z(u = x u = y).
(5) The union axiom is of the form
x y y =
where y =

x,

x can be written as
z y u x z u u x z u z y.


Lemma 5.7 is shown.

It is more delicate verifying that a given transitive model M is a model of (Pow),


(Aus) , (Ers) , and (AC).
Recall (cf. Definition 4.51) that an inner model is a transitive proper class model
of ZFC. If E is a set or a proper class, then L[E] is the least inner model which is
closed under the operation x  E x. An important example will be L = L[],
Gdels constructible universe, which we shall study in detail.
In order to show that L[E] indeed always exists, we need to define it in a way that
is different from saying the least inner model which is closed under the operation
x  E x. Any model of the form L[E] may be stratified in two ways: into
levels of the L-hierarchy and into levels of the J -hierarchy. The former approach
was Gdels original one, but it turned out that the latter one (which was discovered
by Ronald B. Jensen, cf. [16]) is much more useful.
In order to define the J -hierarchy we need the concept of rudimentary functions.
Definition 5.8 Let E be a set or a proper class. A function f : V k V , where
k < , is called rudimentary in E (or, rud E ) if it is generated by the following
schemata:
f (x1 , . . . , xk ) = xi
f (x1 , . . . , xk ) = xi \x j
f (x1 , . . . , xk ) = {xi , x j }
f (x1 , . . . , xk ) = h(g1 (x11 , . . . , xk11 ), , g (x1 , . . . , xk ))

g(y, x2 , . . . , xk )
f (x1 , . . . , xk ) =
yx1

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 ,

5.1 The Constructible Universe

71

f (x) = xi x j , f (x) = {x1 , . . . , xk }, f (x) = (x1 , . . . , xk ), and f (x) = x1 . . .xk .


(Cf. Problem 5.5.) Lemma 5.10 below will provide more information.
If U is a set and E is a set or a proper class then we shall denote by rud E (U ) the
rud E closure of U , i.e., the set
U { f ((x1 , ..., xk )); f is rud E and x1 , ..., xk U }.
It is not hard to verify that if U is transitive then so is rud E (U {U })
(cf. Problem 5.7). We shall now be interested in P(U ) rud E (U {U })
(cf. Lemma 5.11 below).
Definition 5.9 Let E be a set or a proper class. A relation R V k , where k < , is
called rudimentary in E (or, rud E ) if there is a rud E function f : V k V such that
R = {x: f (x) = }. R is called rudimentary (or, rud) if R is rud .
Lemma 5.10 Let E be a set or a proper class.
(a) The relation
/ is rud.
(b) Let f , R be rud E . Let g(x) = f (x) if R(x) holds, and g(x) = if not. Then
g is rud E .
(c) If R, S are rud E then so is R S.
(d) Membership in E is rud E .
(e) If R is rud E , then so is its characteristic function R .2
(f) R is rud E iff R is rud E .
(g) Let R be rud E . Let f (y, x) = y {z: R(z, x)}. Then f is rud E .
(h) If R(y, x) is rud E , then so is z y R(z, x).
Proof (a) x
/ y iff {x}\y = .

(b) If R(x) r (x) = , where r is rud E , then g(x) = yr (x) f (x).
(c) Let R(x) f (x) = , where f is rud E . Let g(x) = f (x) if S(x) holds,
and g(x) = if not. g is rud E by (b), and thus g witnesses that R S is rud E .
(d) x E iff {x} E = .
(e): by (b).
(f) R (x) = 1\ R (x).
(g) Let g(z, x) = {z} if R(z,
 x) holds, and g(z, x) = if not. We have that g is
rud E by (b), and f (y, x) = zy g(z, x).
(h) Set f (y, x) = y {z; R(z, x)}. f is rud E by (g), and thus f witnesses that

z y R(z, x) is rud E .
We shall often be concerned with models of the form (U ; , A1 , . . . , Am ), where
A1 , . . ., Am U < and stands for  U = U 2 . Each such structure comes
A 1 , . . ., A m . We shall mostly restrict
with a language L,A1 ,...,Am with predicates ,
ourselves to discussing the cases where m = 0 (i.e., where there is no Ai around) or
m = 1 or m = 2.
Let U = (U ; , A1 , . . . , Am ) be a model as above. The notions of n -and n formulae of the language of L,A1 ,...,Am is defined as in Definition 5.1, where x Ai ,
2

I.e., R (x) = 1 iff R(x) and = 0 otherwise.

72

5 Constructibility

0 < i m, count as atomic formulae. If X U , and n < , then we let nU (X )


denote the set of all relations which are n definable over U from parameters in X ,
i.e., the set of all R such that there are k, l < , some n -formula (x1 , . . . , xk+l ) of
L,A1 ,...,Am and a1 , . . ., al X such that R U k and for all z = (z 1 , . . . , z k ) U k ,
z R U |= (z 1 , . . . , z k , a1 , . . . , al ).
U
U
We shall also write U
n for n (U ), and we shall write for


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 .

Let U = (U ; , A1 , . . . , Am ) and U = (U ; , A1 , . . . , Am ) be models as


above. Generalizing Definition 4.50, we say that : U U is a n -elementary
embedding, written : U n U , where n {} iff ran( ) is a n -elementary
substructure of U in the common language L,A1 ,...,Am , i.e., if for all n -formulae
of the language L,A1 ,...,Am and for all a1 , . . . , ak U ,

U |= (a1 , . . . , ak ) U |= ((a1 ), . . . , (ak )).

(5.1)

If is the identity on U , then we also write this as U n U .


The following lemma says that rud E (U {U }) is just the result of stretching
(U ;,E)
without introducing additional elements of P(U ). By (U ; , E) we shall

always mean the model (U ;  U, E U ). A set U is rud E closed iff rud E (U ) U .


Lemma 5.11 Let U be a transitive set, and let E U . Then P(U ) rud E (U
;,E)
.
{U }) = P(U ) (U

(U ;,E)

Proof Notice that P(U )


prove that

(U {U };,E)

= P(U ) 0

, so that we have to

(U {U };,E)

P(U ) rud E (U {U }) = P(U ) 0

: By Lemma 5.10 (a) and (d),


/ and membership in E are both rud E . By
Lemma 5.10 (f), (c), and (h), the collection of rud E relations is closed under complement, intersection, and bounded quantification. Therefore we get inductively that
every relation which is 0 in the language L,E with and E is also rud E .
(U {U };,E)
Now let x P(U ) 0
. There is then some rud E relation R and

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

5.1 The Constructible Universe

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

({x1 , ..., xk }),

, and hence B x rud E (x {x})


where x1 , , xk x. Then B x (x;,Ex)
0
by Lemma 5.11. But rud E (x {x}) U , and therefore B x U . We have shown
the following.
Lemma 5.12 Let U be a transitive set such that for every x U there is some
transitive y U with x y, let E be a set or a proper class, and suppose that U
is rud E closed. Then (U ; , E) is a model of 0 comprehension in the sense that if
;,E)
B (U
and x U , then B x U .
0

In the situation of Lemma 5.12, (U ; , E, B) is therefore amenable in the sense


of the following definition.
Definition 5.13 A structure (U ; , A1 , . . . , Am ), where U is transitive and A1 , . . .,
Am U < , is called amenable if and only if Ai x U whenever 0 < i m and
x U.
Later on, cf. Definition 11.4, we will study possible failures of 1 comprehension
in rud E closed structues. Lemma 5.12 provides the key element for proving that
(all but two of) the structures we are now about to define are models of basic set
theory, a theory which consists of 0 comprehension together with extensionality,
foundation, pairing, union, infinity, and the statement that Cartesian products and
transitive closures exist.
We are ready to define the J [E] hierarchy as follows. For later purposes it is
convenient to index this hierarchy by limit ordinals.
Definition 5.14 Let E be a set or a proper class.
J0 [E] =
J+ [E] = rud E (J [E] {J [E]})

J [E] for limit
J [E] =
<

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

Here, Lim denotes the class of all limit ordinals.

5.1 The Constructible Universe

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] =
<

where S E is an operator which, applied to a set U , adds images of members of


U {U } under rud E functions from a certain carefully chosen fixed finite list. We
may set
S E (U ) =


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 )}.

Here, x is not assumed to be transitive.

5.1 The Constructible Universe

77

Claim 5.21 g,k C for every (v1 , . . . , vk ).


Proof First let (v1 , . . . , vk ) v j vi , where 1 i < j k. If k = 2, then we
simply have g,k (x) = F5 (x), but in general we also need F2 , as well as F7 and F8
for reshuffling. Let us write X 1 (z, x) = z and X n+1 (z, x) = F2 (X n (z, x), x) =
X n (z, x) x for n 1, and let us also write F81 (z, x) = F8 (z, x) and F8n+1 (z, x) =
F8 (F8n (z, x), x) for n 1. Say 2 i < i + 1 < j k; then
ji1

g,k (x) = X k j+1 (F8

(F7 (X i1 (x, x), F5 (x)), x), x).

The case i = 1 or j = i + 1 is similar.


Next, notice that
g,k (x) = F1 (X k (x, x), g,k (x))
and5
g,k (x) = F1 (g,k (x), F1 (g,k (x), g,k (x))).
Also
gvk ,k (x) = F4 (g,k+1 (x)).
Finally, if (v1 , . . . , vk ) vi = v j , where 1 i, j k, i = j, then
g,k (x) = F4 (gvk+1 (vk+1 vi vk+1 v j ),k+1 (x

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.

Then f is rud, so that by Claim 5.22 the function u  fu is in C. But then


x1 , . . . , xk 

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

5.1 The Constructible Universe

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

Definition 5.24 A J -structure is an amenable structure of the form (J [E], B) for


a limit ordinal and predicates E, B, where E satisfies the hypothesis of Lemma
5.15.
Here, (J [E], B) denotes the structure (J [E];  J [E], E J [E], B J [E]).
Of course, every J [E] is a J -structure.
Lemma 5.25 Let J [E] be a J -structure, where is a limit ordinal.
(1) For all < , (S [E]: < ) J [E]. In particular, S [E] J [E] for all
< .
(2) (S [E]: < ) is uniformly 1J [E] . I.e., x = S [E] is 1 over J [E], as
being witnessed by a formula which does not depend on .
Proof (1) and (2) are shown simultaneously by induction on (, ), ordered lexicographically. Fix a limit ordinal and some < . If is a limit ordinal, then
J [E]
inductively by (2), (S [E]: < ) is 1 , and hence (S [E]: < ) J [E]
by Lemma 5.15. If = + 1, then 
inductively by (1), (S [E]: < ) J [E].
If is a limit ordinal, then S [E] = < S [E] J [E]; and if = + 1 then
S [E] = S E (S [E]) J [E] as well, as S E is rud E (cf. Problem 5.6). It follows that
(S [E]: < ) = (S [E]: < ) {(, S [E])} J [E], which proves (1). (2) is
then not hard to verify.

In order to show that L[E] satisfies the Axiom of Choice (even locally), we may
inductively define a well-ordering <E of S [E] as follows. If is a limit ordinal

then we let <E = < <E . Now suppose that = + 1. The order <E induces
E
, of 17 S [E] S [E]. We may then set8
a lexicographical order, call it <,lex

x, y S [E] and x <E y, or else

/ 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

and ( j, u , v ) is < E minimal with y = F (u , v ).


y y
j y y

,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

we pretend that all Fi , i < 17, are binary.

80

5 Constructibility

Lemma 5.26 Let J [E] be a J -structure.


(1) For all < , (<E : < ) J [E]. In particular, <E J [E] for all < .
(2) (<E : < ) is uniformly 1J [E] . I.e., x =<E is 1 over J [E], as being
witnessed by a formula which does not depend on .
Theorem 5.27 (Gdel) Let E be a set or a proper class. Then L[E] |= ZFC.
Proof This is an immediate consequence of Theorem 5.19 and Lemma 5.26.

we abbreviate the L -sentence


By V = L[ E]
, E
x y),
xy (y = S [ E]
stands for the 1 formula given by Lemma 5.25 (2). By Lemma
where y = S [ E]
is 2 , uniformly over J [E] (including = ). As a special
5.25 (2), V = L[ E]
case, by V = L we abbreviate the L -sentence
xy (y = S [] x y).
We then get:
Lemma 5.28 Let M be a transitive model. Then M |= V = L iff M = J for some
. (In particular, for any transitive model M, L M = J , where = M OR.) More
for some and E,
where x E
iff M = J [ E]
generally, M |= V = L[ E]
M |= x E for all x M. (Here, OR if M is a set and = if M is a proper
class.)
Proof We prove the first assertion; the proof of the second one is basically identical.
= immediately follows from Lemma 5.25 (1). To see =, let M |= V = L.
If x M, then M |= y (y = S [] x y). But this is 1 , so that by Lemma
5.3, every x M isreally contained in some S [] for M. Thus, setting
= M OR, M < S [] = S [] = J . By the same reasoning, if < ,

then M |= y (y = S [] y) and J = < S [] M.

The Condensation Lemma for the constructible hierarchy is the following statement.

Theorem 5.29 Let M = (J [E], B) be a J -structure, and let : M


1 M, where
B)
is a
and B such that M = (J [ E],
M is transitive. Then there are , E,
J -structure.
Proof Set = OR M , E = 1 "E, and B = 1 "B. The sentence
is inherited from M |= V = L[ E],
and
is 2 , so that M |= V = L[ E]
V = L[ E]
B).


Lemma 5.28 then immediately gives M = (J [ E],
The Condensation Lemma 5.29 leads to the following natural concept.

5.1 The Constructible Universe

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

Trivially, full condensation implies local condensation, and E = satisfies


full condensation. Theorem 5.31 also holds if E just satisfies local condensation
(cf. Problem 5.17).
The definitions of variants of as well as the results on them which we are about
to prove are all due to Ronald Jensen.
Definition 5.34 Let be a regular uncountable cardinal, and let R . By (R)
we mean the following statement. There is a sequence (A : R) such that for all
R, A , and for every A , the set
{ R: A = A }
is stationary in . We also write for () and for 1 .
Trivially, (R) implies that R be stationary. It is not hard to show that +
implies that 2 = + (cf. Problem 5.20).
Theorem 5.35 (Jensen) Let E be a set or a proper class which satisfies local condensation. Then inside L[E], for every regular uncountable cardinal and every
stationary R , (R) holds true.
In particular, inside L, for every regular uncountable cardinal and every stationary R , (R) holds true.
Proof Let us work inside L[E], and let us fix a stationary set R . Let us recursively
construct (B , C : R) as follows. Having constructed (B , C : R ), where
R, let (B , C ) be the <E -least pair (B, C) J [E] such that B , C is club
in , and
{ R : B = B } C = ,
provided such a pair exists; otherwise we set (B , C ) = (, ).
We claim that (B : R) witnesses that (R) holds true. If not, then let
(B, C) L[E] be the < L[E] -least pair such that B , C is club in , and
{ R: B = B } C = .

(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 )

is then the <E -least pair such that B , C is club in , and

5.1 The Constructible Universe

83

{ R : B = B } (C ) = .

(5.4)

Let < be least such that

{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,


which contradicts (5.3).


Lemma 5.36 (Jensen) If holds true, then there is an 1 -Souslin tree.
We shall prove a more general statement, cf. Lemma 11.68, later.
We also discuss a strengthening of , called .

Definition 5.37 Let be a regular uncountable cardinal, and let R . By (R)


we mean the following statement. There is a sequence (A : R) such that for all
R, A P( ) and Card(A ) , and for every A there is some club
C such that A A for every C R. We also write for () and
for 1 .
Lemma 5.38 (Kunen) Let be a regular uncountable cardinal, and let R be
stationary. Then (R) implies (R).
Proof Let (A : R) witness (R). Say A = {A,i : i < } for R. If i,
j < , R, then let us write
j

A,i = { < : , j! A,i },


where , j  , j! is the Gdel pairing function (cf. p. 33). We claim that there
is some i < such that (Ai,i : R) witnesses (R).
Suppose not. Then for every i < there is some Ai and some club Ci
such that
(5.5)
{ R: Ai = Ai,i } Ci = .
Let D = i< Ci , and let

A = { , i!: Ai }.

As (A : R) witnesses (R), there is some club C D such that A A


for every C R.

84

5 Constructibility

Let C R, where we may assume without loss of generality that is closed


under the Gdel pairing function (  , ! is incresing and continuous). There is
then some i 0 < such that A = A,i0 , and we also have that A = { , i!:
Ai , i < }. This yields that
0
= Ai0 .
Ai,i
0

(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.,

we therefore have that E = E J( ) [ E] and J( ) [ E]

5.1 The Constructible Universe

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)

If S, then (A ) g , so that there is some f ( ) < with (A ) = g( f ( )).


By Fodors Theorem 4.32, there is some stationary T S and some < such
that for all T , f ( ) = .
Let us write A = g(). If T , then (A ) = g() = A, which means that
A = A . But then A = A whenever , , T . This contradicts

the fact that (A : R) witnesses that R is not ineffable.
Lemma 5.40 Let be an uncountable regular cardinal, and assume R to be
ineffable. Then (R) fails.
Proof Suppose (A : R) were to witness (R). For R, say A =
{A,i : i < }, and set
A = { , i!: A,i },
where , i  , i! again is the Gdel pairing function. Applying the ineffability
of R to (A : R), we find some stationary S R such that for all with
, S,
A = A .
For i < , let us write
Ai = { < : S , i! A }.
By the properties of S,
A,i = Ai

(5.10)

for all i < and S.


Let us now pick any A P()\{Ai : i < }. As (A : R) is supposed to
witness (R), there is some club C such that A A for all R C.
By Fodors Theorem 4.32 there is some stationary T S C and some i 0 <
such that for all T ,
A = A,i0 .

86

5 Constructibility

But this implies that A = Ai0 by (5.10). Contradiction!

There is a principle which is slightly stronger than and which is called +


, cf.
Problem 5.22.

5.2 Ordinal Definability


We also need to introduce HOD.
Definition 5.41 Let Y be a set or a proper class. We say that x is hereditarily in Y
iff T C({x}) Y .
Definition 5.42 Let z be a set or a proper class. We write ODz for the class of all x
which are ordinal definable from elements of z, i.e., such that there is a formula ,
there are ordinals 1 , . . . , n and elements y1 , . . . , ym of z such that for all u,
u x (u, 1 , . . . , n , y1 , . . . , ym ).
We also write HODz for the class of all x which are hereditarily in ODz , i.e.,
HODz = {x : T C({x}) ODz }.
If x HODz , then we say that x is hereditarily ordinal definable from elements of
z. If z = , then we write OD instead of OD and HOD instead of HOD .
Lemma 5.43 Let z be a set. Then x ODz iff 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 ).
Proof By the reflection principle, cf. Problem 5.14, given , 1 , . . . , n and
y1 , . . . , ym z there is some with 1 , . . ., n , z V and for all u V ,
(u, 1 , . . . , n , y1 , . . . , ym ) V |= (u, 1 , . . . , n , y1 , . . . , ym ).
The Lemma then follows using Problem 5.1.

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.

5.2 Ordinal Definability

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.

5.2. Let E A be as in Problem 4.10 (c). Let (; E A )


= (HF; ). Show that the

.
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<

v iff max(uv) v. Show that is a well-

5.20. Let be an infinite cardinal such that + holds true. Show that 2 = + .

90

5 Constructibility

Let be a regular uncoutable cardinal. R is called subtle iff for every


sequence (A : R) such that A for every R and for every club
C there are , R C with < such that A = A .
5.21. (K. Kunen) Let be a regular uncoutable cardinal, and let R . Show that
if R is ineffable, then R is subtle. Show also that if R is subtle, then (R)
holds true. [Hint. Follow the proof of Theorem 5.35.]
Let be a regular uncountable cardinal, and let R . By +
(R) we mean
the following statement. There is a sequence (A : R) such that for all
R, A P( ) and Card(A ) , and for every A there is some
+
club C such that for every C R, {A , C } A . +
is ().
5.22. (R. Jensen) Assume V = L[E], where E satisfies local condensation. Let be
a regular uncountable cardinal, and let R . Show that if R is not ineffable,
then +
(R) holds true. [Hint. Follow the proof of Theorem 5.39. Pick such
that R J , say R = g(0). Towards the end, after (5.7), let C = {( ): D}.
then B 0 , C 0 A0 .
Verify as follows that if 0 R is a limit point of C,
Otherwise A0 J(0 ) . Set S = { R 0 : A = A0 }. By the choice
of (A : R) and the elementarity of 0 , S cant be stationary in J(0 ) ,
so that J(0 ) has a club I disjoint from S. But setting I = 0 (I ) and
A = 0 (A0 ), 0 R I and A0 = A 0 . Contradiction!]
Kurepas Hypothesis at , KH , is the statement that there is some set B
P() of size + such that for all < , {X : X B} has size at most
Card( ).
5.23. (R. Jensen) Let be regular and uncoutable. Then +
implies KH . [Hint. Let
.
For

<
,
let
M
be
a
transitive model of ZFC
(A : < ) witness +

of size Card( ) such that A M . Let B = {X : < X M }.]


5.24. (R. Jensen, K. Kunen) Show that if is ineffable, then KH fails. Conclude
that if is ineffable, then +
fails.
Kripke- Platek set theory, KP, for short, is the theory in the language of
L which has the following axioms. (Ext), (Fund ) for every formula (cf.
Problem 3.4), (Pair), (Union), (Inf), (Aus ) for all 0 -formulae , and (Coll )
for all 0 -formulae (cf. Problem 3.5).
5.25. (a) Show that there is a proof of Lemma 2.1 in KP, i.e., for all a, b, a b
exists is provable in KP.
(b) Show also that KP proves (Coll ) for all 1 -formulae .
(c) Also show that if and are both 1 and such that KP proves ,
then KP proves (Aus ).
5.26. Let > be a limit ordinal. Show that the following statements are equivalent.
(a) J |= KP.
(b) J |= (Coll ) for all 0 -formulae .

5.3 Problems

91

(c) if and are both 1 and such that J |= xv( ), then


J |= (Aus ).
(d) there is no total f : a J , where a J and f 1J .

Let (M; E) be a model of KP, and let (N ; )


= (wfp(E); E  wfp(M)), i.e.
N is the transitive collapse of the well-founded part of (M; E). Then (M; E)
is called an -model iff N . A set N is called admissible iff N is transitive
and (N ;  N ) |= KP.
5.27. (Villes Lemma) Let (M; E) |= KP be an -model. Show that if N is the
transitive collapse of the well-founded part of (M; E), then N is admissible.
5.28. Show in KP that if R M M is well-founded, where M is a set, then the
function : R OR is a set, where (x) = {(y): y Rx}, i.e., (x) is the Rrank of x for every x R. [Hint. The relevant in the Recursion Theorem 3.13
is 1 .] Show also in KP that if R M M is well-founded and extensional,
where M is a set, then the transitive collapse as defined in Theorem 3.21 is
a set.
Conclude the following. Let N be admissible. Let R N be a binary relation which is
well-founded in V . Then ||R|| < N OR. If R N is well-founded and extensional
in V , then the transitive collapse of R is an element of N . Also, if R N is any
relation, then ||wfp(R)|| N OR.
Let z . We call z-admissible iff J [z] is an admissible set. We write 1z for
the least z-admissible ordinal. We also write 1CK for 10 (0 = the constant function
wth value 0, CK = ChurchKleene).

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).

6.1 The General Theory of Forcing


Recall that P = (P; ) is a partial order iff is reflexive, symmetric, and transitive
(cf. p. 14). In what follows, we shall always assume that P = . As before, we write
p < q for p q q p (which, by symmetry, is equivalent to p q p = q).
Definition 6.1 Let P = (P; ) be a partial order. We also call P a notion of forcing
and the elements of P the forcing conditions. For p, q P we say that p is stronger
than q iff p q, and we say that p is strictly stronger than q iff p < q.
Definition 6.2 Let P = (P; ) be a partial order. A set D P is called dense (in
P) iff p P q D q p. If p P, then D P is called dense below p iff
p p q D q p . A set G P is called a filter iff (a) if p, q G, then there
is some r G with r p r q, and (b) p G q P( p q q G).
If P = (P; ) is a partial order, and if p, q P, then we write p q for
r P(r p r q), in which case p, q are called compatible, and we write
pq for p q, in which case p, q are called incompatible. If G is a filter, then
any two p, q G are compatible (as being witnessed by an element of the filter).
Definition 6.3 Let P = (P; ) be a partial order, and let D be a family of dense
sets. A filter G P is called D-generic iff G D = for all D D.
Lemma 6.4 Let P = (P; ) be a partial order, and let D be family of dense sets
such that D is at most countable. Then for every p P there is a D-generic filter G
with p G.
R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_6,
Springer International Publishing Switzerland 2014

93

94

6 Forcing

Proof Say D = {Dn : n < }. Let p P be given, and recursively construct


( pn : n < ) as follows. Set p0 = p. If pn is constructed, where n < , then
pick some q pn with q Dn (this is possible because Dn is dense in P), and set
pn+1 = q. It is then easy to see that
G = {r P : n pn r }
is a D-generic filter.

This lemma produces a forcing proof of Cantors Theorem 1.1 as follows.


Definition 6.5 Let C = < , i.e., the set of all finite sequences of natural numbers.
For p, q C, let p q iff p q (iff n p  n = q). The partial order C = (C; )
is called Cohen forcing.
Now let X be a countable subset of , the set of all infinite sequences of natural
numbers. Say X = {xn : n < }. Set
Dn = { p C : p = xn  dom( p)},
and

Dn = { p C : n dom( p)}.

It is easy to verify that each Dn as well as each Dn is dense in C.


Set D = {Dn : n < } {Dn : n < }, and let G be D-generic, via Lemma 6.4.

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.

6.1 The General Theory of Forcing

95

Lemma 6.8 Let M be a transitive model of ZFC, let P = (P; ) M be a partial


order, and let G be P-generic over M. If p G, and if D P, D M, is dense
below p, then G D = .
Proof Set
D = {q P : r D q r } {q P : r D qr }.
Then D M, and D is easily be seen to be dense. Let q G D , and let s G
be such that s p and s q. As D is dense below p, there is some t s with
t D, so that in particular q||t. But q D , so that we must now have some r D
with q r , which gives r D G as desired.

Definition 6.9 Let P = (P; ) be a partial order. A P is called an antichain iff
for all p, q A, if p = q, then pq. A P is called a maximal antichain iff A is
an antichain and for all p P there is some q A with p q. D P is called open
iff for all p D, if q p, then q D.
The Hausdorff Maximality Principle 2.11 easily gives that every antichain is
contained in a maximal antichain.
Lemma 6.10 Let M be a transitive model of ZFC, and let P = (P; ) M be a
partial order. Let G P be a filter. The following are equivalent.
(1) G is P-generic over M.
(2) G A = for every maximal antichain A P, A M.
(3) G D = for every open dense set D P, D M.
Proof (1) = (2): Let A P, A M, be a maximal antichain. Let D = { p P :
q A p q}. D is easily seen to be dense, and of course D M. Let p G D.
There is some q A with p q. But then q G A, as G is a filter.
(2) = (1): Let D P, D M be dense. Working in M, let A D be an
antichain such that for every p D there is some q A with p q. It is easy to see
that A is then a maximal antichain. (Cf. Problem 6.1.) But then p A G implies
p D G.
(1) = (3): This is trivial.
(3) = (1): Let D P, D M be dense. Let D = { p : q D p q}. D
is then open dense, and of course D M. Let p D G. There is then some
q D with p q. But then q D G, as G is a filter.

P is called atomless iff p P q P r P (q p r p qr ). Cohen
forcing is atomless as are all the other forcings considered in this book.
Lemma 6.11 Let M be a transitive model of ZFC, let P = (P; ) be an atomless
partial order, and let G P be P-generic over M. Then G
/ M.
Proof Suppose that G M. Then D = P \ G M, and D is dense: if p P
and if q p and r p are incompatible, then at most one of q, r can be
in G, i.e., at least one of q, r must be in D. But then D G = , which is
nonsense.


96

6 Forcing

Definition 6.12 Let M be a transitive model of ZFC, and let P M be a partial


order. For < M OR, we recursively define the sets MP of P-names (of M) of
rank as follows. Set
MP = { M : is a binary relation and
(, p) ( p P < MP )}
We also write M P =

<MOR

MP for the class of P-names (of M).

Of course M P M, so that if M is a set, then M P is a set as well.


Definition 6.13 Let M be a transitive model of ZFC, and let P M be a partial order.
Let G P be P-generic over M. For M P we write G for the G-interpretation
of , which is defined to be
{ G : p G (, p) }.
We also write M[G] = { G : M P } and call it a (the) generic extension of M
(via P, G).
Of course, the definition of G is by recursion on the rank of in the sense of
Definition 6.12. We aim to prove that any generic extension of a transitive model of
ZFC is again a transitive model of ZFC.
In what follows, we want to assume that our partial order P always has a least
(weakest) element 1 = 1P , i.e., p 1 for all p P. (Most often, 1P = .)
By recursion on the -rank of x M, we define x = {( y , 1) : y x} for every
x M. A trivial induction shows x M P for every x M. We also define G to be
the P-name {( p,
p) : p P}; obviously, G M P .
Lemma 6.14 Let M be a transitive model of ZFC, let P M be a partial order,
and let G P be P-generic over M. M[G] is transitive, and M {G} M[G].
Proof The transitivity of M[G] is trivial. In order to verify M M[G], we show
x G = x for every x M by induction on the rank of x. We have x G = { G : p
G (, p) x}
= { y G : y x} (notice that 1 G) = x by the inductive hypothesis.
In order to verify G M[G], we show that G G = G. Well, G G = { G : p
= { p G : p G} = G, as p G = p for all p P.

G (, p) G}
It is easy to verify that if N is a transitive model of ZFC with M {G} N ,
then M[G] N . Therefore, once we showed that M[G] is indeed a model of ZFC,
we know that it is the smallest ZFC-model which contains M {G}. To begin, the
ordinal height of M[G] is the same as the one of M:
Lemma 6.15 Let M be a transitive model of ZFC, let P M be a partial order,
and let G P be P-generic over M. M[G] O R = M O R.

6.1 The General Theory of Forcing

97

Proof By Lemma 6.14, we only need to see that M[G] O R M.


A straightforward induction yields that the -rank rk ( G ) of G is at most the
-rank of , for every M P . Now let M[G] O R, say = G , where
M P . Then = rk ( ) = rk ( G ) r k( ) < M OR, as desired.

Because M[G] is transitive and M[G], we know by Lemma 5.7 (1) through
(3) that M[G] is a model of the axioms (Ext), (Fund), and (Inf).
Lemma 6.16 Let M be a transitive model of ZFC, let P M be a partial order,
and let G P be P-generic over M. M[G] |= (Pair).
Proof Let x, y M[G], say x = G , y = G , where , M P . Let
= {(, 1P ), (, 1P )}.
Of course, M P . But it is easy to see that G = { G , G } = {x, y}, so that
{x, y} M[G]. The result then follows via Lemma 5.7 (4).

Lemma 6.17 Let M be a transitive model of ZFC, let P M be a partial order,
and let G P be P-generic over M. M[G] |= (Union).
Proof Let x M[G], say x = G , where M P . Let
= {(, p) : qq ( p q p q (, q) (, q ) )}.
Of course, M P , and it is straightforward to verify that G =
then follows via Lemma 5.7 (5).

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

Definition 6.19 Let M be a transitive model of ZFC, and let P M be a partial


order. Let p P.
(1) Let 1 , 2 M P . We define p PM 1 = 2 to hold iff: for all (1 , s1 ) 1 ,
{q p : q s1 (2 , s2 ) 2 (q s2 q PM 1 = 2 )}
is dense below p and for all (2 , s2 ) 2 ,
{q p : q s2 (1 , s1 ) 1 (q s1 q PM 1 = 2 )}
is dense below p.
(2) Again let 1 , 2 M P . We define p PM 1 2 to hold iff
{q p : (, s) 2 (q s q PM = 1 )}
is dense below p.
) be formulae, and let , . . . , , , . . . ,
(3) Let (v1 , . . . , vn ), (v1 , . . . , vm
1
n 1
m
P
P
M . We define p M (1 , . . . , n ) (1 , . . . , m ) to hold iff both
p PM (1 , . . . , n ) as well as p PM (1 , . . . , m ) hold. We define p PM
(1 , . . . , n ) to hold iff for no q p, q PM (1 , . . . , n ) holds.
(4) Let x(x, v1 , . . . , vn ) be a formula, and let 1 , . . . , n M P . We define p PM
x(x, 1 , . . . , n ) to hold iff
{q p : M P q PM (, 1 , . . . , n )}
is dense below p.
In what follows, we shall often write rather than PM .
The definition of p 1 = 2 is by recursion on (rk (1 ), rk (2 )), ordered
lexicographically. p 1 2 is then defined with the help of p = 1 , where
(, s) 2 for some s. Moreover, the definition of p (1 , . . . , n ) for nonatomic
is by recursion on the complexity of . The relation is thus well-defined. We
obviously have:
Lemma 6.20 Let M be a transitive model of ZFC, and let P M be a partial order.
Let be a formula. Then {( p, 1 , . . . , n ) : p (1 , . . . , n )} is definable over M
(from the parameter P).
We say that p decides (1 , . . . , k ) iff p (1 , . . . , k ) or p (1 , . . . , k ).
Definition 6.19 (3) trivially yields that for a given (1 , . . . , k ), there are densely
many p which decide (1 , . . . , k ). The following is also straightforward to verify,
cf. Problem 6.2.
Lemma 6.21 Let M be a transitive model of ZFC, and let P M be a partial order.
Let p P, let be a formula, and let 1 , . . . , n M P . Equivalent are:

6.1 The General Theory of 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)

is not dense below r , or there is some (2 , s2 ) 2 such that


{q r : q s2 (1 , s1 ) 1 (q s1 q 1 = 2 )}

(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)

In particular, p s1 . Also, (6.3) gives that if (2 , s2 ) 2 , q s2 , q 1 = 2 ,


then q p. That is, p r and 1 ( p) holds true. We have shown that D is dense.
But now there must be some p G D. As p G, we have seen that 1 ( p) and
2 ( p) must both fail, so that p 1 = 2 holds true, as desired.
We now prove (1) and (2) for v1 v2 , exploiting the fact that (1) and (2)
hold true for v1 = v2 .
(1): Suppose that p G and p 1 2 . By definition,
D = {q p : (, s) 2 (q s q = 1 )}
is then dense below p. Pick q D G. Let (, s) 2 be such that q s and
q = 1 . As q G, G = 1G . But s G, too, and hence 1G = G 2G .
(2): Suppose that 1G 2G . There is then some (, s) 2 such that s G and
G
1 = G . We therefore have some r G with r 1 = . Let p G be such that
p s, r . Then q p(q s q 1 = ). Hence p 1 2 .
Let us finally prove (1) and (2) for nonatomic formulae.
) be formulae, and let , . . . , , , . . . ,
Let (v1 , . . . , vn ), (v1 , . . . , vm
1
n 1
m
M P . Suppose that (1) and (2) hold for (1 , . . . , n ) and for (1 , . . . , n ). It is then
trivial that (1) and (2) also hold for (1 , . . . , n ) (1 , . . . , m ). Let us show that
(1) and (2) hold for (1 , . . . , n ).
(1): Let p G, p (1 , . . . , n ). Suppose that M[G] |= (1G , . . . , nG ).
There is then some q G such that q (1 , . . . , n ). Pick r G, r p, q. Then
r p and r (1 , . . . , n ). Contradiction! Hence M[G] |= (1G , . . . , nG ).
(2): Let M[G] |= (1G , . . . , nG ). It is easy to see that
D = {q : q (1 , . . . , n ) q (1 , . . . , n )}

6.1 The General Theory of Forcing

101

is dense. But if q D G, then q (1 , . . . , n ). This is because otherwise


q (1 , . . . , n ) and then M[G] |= (1G , . . . , nG ). Contradiction!
Finally, let x(x, v1 , . . . , vn ) be a formula, and let 1 , . . . , n M P . Suppose
that (1) and (2) hold for (, 1 , . . . , n ) whenever M P . We aim to show (1)
and (2) for x(x, 1 , . . . , n ).
(1): Suppose that p G, p x(x, 1 , . . . , n ). By definition,
D = {r p : M P r (, 1 , . . . , n )}
is then dense below p. Pick r D G. Then there is some M P such that,
using the inductive hypothesis, M[G] |= ( G , 1G , . . . , nG ). But we then have that
M[G] |= x(x, 1G , . . . , nG ).
(2): Let M[G] |= x(x, 1G , . . . , nG ). Pick M P witnessing this, i.e., such
that M[G] |= ( G , 1G , . . . , nG ). By our inductive hypothesis, there is some p G
with p (, 1 , . . . , n ). But then r (, 1 , . . . , n ) for all r p, which

trivially implies that p x(x, 1 , . . . , n ) by definition.
One can in fact show that if p x(x, 1 , . . . , n ), then there is some M P
such that p (, 1 , . . . , n ). This property is called fullness, cf. Problem 6.5.
Theorem 6.23 (Forcing Theorem, part 2) Let M be a transitive model of ZFC, and
let P M be a partial order such that for every p P there is some G with
p G such that G is P-generic over M. Let (v1 , . . . , vn ) be a formula, and let
1 , . . . , n M P .
(1) For all p P,
p PM (1 , . . . , n ) p PM (1 , . . . , n ).
(2) Let G P be P-generic over M. Then
M[G] |= (1G , . . . , nG ) p G p PM (1 , . . . , n ).
Proof Let us first show (1):
=: Let p (1 , . . . , n ). Let G be any P-generic filter over M such that
p G. Then M[G] |= (1G , . . . , nG ) by Theorem 6.22 (1). Therefore, p PM
(1 , . . . , n ).
=: Suppose that p PM (1 , . . . , n ). We need to see that {q p : q
(1 , . . . , n )} is dense below p. If not, then there is some q p such that for all
r q, r (1 , . . . , n ) does not hold true, i.e., q (1 , . . . , n ). But then
q PM (1 , . . . , n ) by =. This contradicts p PM (1 , . . . , n ), as q p.
Let us show (2). Well, = follows from (1) above plus Theorem 6.22 (2).
= is just by definition.

By Theorem 6.23 (1), we shall not have any use for the notation any more.

102

6 Forcing

Theorem 6.24 Let M be a transitive model of ZFC, let P M be a partial order,


and let G P be P-generic over M. M[G] |= ZFC.
Proof We are left with having to verify that M[G] |= (Aus), (Rep), (Pow), and (AC).
(Cf. also the remark before Lemma 6.17.)
Let us begin with (Aus), i.e., separation. Let (v0 , v1 , . . . , vn ) be a formula, and
let G , 1G , . . ., nG M[G]. We aim to see that
{x G : M[G] |= (x, 1G , . . . , nG )} M[G].
Well, consider
= {(, p) : q( p q (, q) p PM (, 1 , . . . , n ))}.
Notice that M P by 6.23 (1).
Let x M[G], say x = G . Then G G iff (, p) for some p G iff
there is some q p G with (, q) and p PM (, 1 , . . . , n ) iff G G
and M[G] |= (, 1G , . . . , nG ). Hence
{x G : M[G] |= (x, 1G , . . . , nG )} = G M[G].
A similar argument is used to show (Rep) in M[G]. Let (v0 , v1 , v2 , . . . , vn ) be
a formula, let G , 2G , . . ., nG M[G], and suppose that
M[G] |= x G y (x, y, 2G , . . . , nG ).
We aim to see that there is some a M[G] such that
M[G] |= x G y a (x, y, 2G , . . . , nG ).
Consider

= {(, 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.

6.1 The General Theory of Forcing

103

In order to verify (Pow) in M[G], let G M[G]. As we already verified (Aus)


in M[G], it suffices to see that there is some b M[G] such that {x M[G] : x
G } b.
Set N = { : p (, p) }, and let = {(, 1P ) : M P N P}
P
M . Let G G . We want to see that G G . Let = {(, p) : N p PM
}. By Theorem 6.23 (1), M P . We have ( , 1P ) , so G G , and
hence it suffices to verify that G = G .
If G G , then (, p) for some p G, which implies that N
and p PM , where p G. But then G G . On the other hand, let
G G . There is then some p G with (, p) , which implies that N
and p PM . But then (, p) , where p G, and hence G G .
We have shown that b = G is as desired.
Let us finally verify (AC) in M[G]. Let x M[G], say x = G . Let f M,
f : bijective (for some < M O R). We aim to see that M[G] has a
surjection g : x.
In order to define a name for g, we shall use the following notation Let y, z
M[G], say y = G , z = G . Write
, = {({(, 1), (, 1)}, 1), ({(, 1)}, 1)}.
it is easy to see that , M P and in fact , G = (y, z).
Now set
= {( , , 1) : < p f ( ) = (, p)}.
Obviously, M P . Moreover, G is easily seen to be a function with domain .
Let us verify that x ran( G ).
Let y x, say y = G , where (, p) for some p G. There is then some
< with f ( ) = (, p), and hence ( , , 1) . But then (, G ) G , i.e.,

G ( ) = G = y, so that y ran( G ).

6.2 Applications of Forcing


Let us now turn towards applications of forcing. By Corollary 5.32, V = L is
consistent with ZFC. We now show that forcing may be used to show that V = L,
the negation of V = L, is also consistent with ZFC.
Our proof will make use of the concept of a 01 statement: is 10 iff it can be
written in the form n , where is recursive (cf. e.g. [11]). 10 statements
are downward absolute between models of ZFC (compare Lemma 5.3); therefore,
as every model of ZFC has an isomorphic copy of the standard natural numbers ,
we have that if is 10 and there is some ZFC-model which thinks that is true,
then is really true (in V ). In particular, this holds for T is consistent for any

104

6 Forcing

recursively enumerable theory T , because T does not prove 0 = 1 may be written


in a 10 fashion.
Theorem 6.25 If ZFC is consistent, then so is ZFC + V = L.
Proof Let us first prove the following.
Claim 6.26 If M is a transitive model of ZFC and if G is C-generic over M, where
C is Cohen forcing (cf. Definition 6.5), then M[G] is a model of ZFC + V = L.
Proof By Theorem 6.24, we just need to verify M[G] |= V = L.
Let us suppose that M[G] |= V = L. By Lemma 5.28, we must then have
M[G] = J , where = M[G] O R. However, M[G] O R = M O R by Lemma
6.15, so that M[G] = J = L M M by Lemma 5.28 again. But G M[G] \ M
by Lemma 6.11, because C is certainly atomless. Contradiction!

An inspection of the proofs of Theorem 6.24 and of Claim 6.26 shows that we may
define a function  () which maps finite subsets of ZFC {V = L}
to finite subsets = () of ZFC such that the following holds true (provably in
ZFC):


If M is a transitive model of (), and if G is C-generic over M,


then M[G] is a transitive model of .

(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)

M( ) may be obtained from as follows. Given , by Problem 5.14 there is an


(N ; E)least which is an ordinal from the point of view of (N ; E) such that
(N ; E) |= V |= .
We may then apply the Lwenheim- - Skolem Theorem and Theorem 3.21 inside
(N ; E) to find an M as desired. Set M( ) = M. Of course, in general the function
 M( ) will not be in N , but we will only need that M( ) N for every
individual .
Let us now show that ZFC does not prove V = L. Let ZFC be finite, and
set = {V = L}. Setting = () and M = M( ), (6.5) gives that
(N ; E) |= M is a countable transitive model of .

6.2 Applications of Forcing

105

By Lemma 6.4 there is then some G N such that (N ; E) |= G is C-generic over


M, and because (6.4) holds inside (N ; E) we get that2
(N ; E) |= M[G] is a transitive model of .
This means that
(N ; E) |= is consistent.

(6.6)

However, the consistency of is a 10 statement, so that (6.6) yields that is really


consistent. But then does not prove that V = L.
Assuming that ZFC is consistent we have shown that ZFC does not prove that
V = L.

All relative consistency results which use forcing may be produced in this fashion.
In proving them, we thus may and shall always pretend to have a countable transitive
model of ZFC at hand.
In order to prove the consistency of ZFC + CH (relative to the consistency of
ZFC), we need finite support products of Cohen forcing.
Definition 6.27 Let C be Cohen forcing, cf. Definition 6.5. Let be an ordinal. For
p C let supp( p) (the support of p) be the set of all < with p( ) = . Let
C() = { p C : Card(supp( p)) < 0 }.
For p, q C(), let us write p q iff for all < , p( ) q( ) in the sense of
Cohen forcing, i.e., n p( )  n = q( ).
C() is often referred to as the finite support product of Cohen forcings.
Definition 6.28 Let P = (P; ) be a partial order, and let be an uncountable
cardinal. P is said to be -Knaster iff for all A P of size there is some B A
of size such that if p, q B, then p q. P is said to have the -chain condition
(-c.c., for short) iff A < whenever A P is an antichain. P is said to have the
countable chain condition (c.c.c., for short) iff P has the 1 -c.c.
Trivially, if P is -Knaster, then P has the -c.c. Also, if Card(P) = , then
trivially P has the + -c.c.
Lemma 6.29 Let be an ordinal. Then C() is 1 -Knaster.
Proof If is at most countable, then C() = 0 , so that C() is trivially 1 -Knaster
in this case.
Now let be uncountable. Let A C() have size 1 . We shall verify that there
is some B A of size 1 such that any two conditions in B are compatible.
Let X = {supp( p) : p A}. We must have that X has size 1 . Let us pick
some bijection : 1 X . This naturally induces : A C(1 ) as follows.
2

Here, M[G] is M[G] as computed inside (N ; E).

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

Claim 6.30 Let 1 k n, and suppose {k : p D0 } to be bounded in 1 . There


is then some D1 D0 of size 1 , some set {1 , . . . , k } and some set {s1 , . . . , sk } C
p
p
such that for all p D1 , 1 = 1 , . . . , k = k , p(1 ) = s1 , . . . , p(k ) = sk .
p

There is therefore some k n, k 1, such that {k : p D0 } is unbounded in


1 . Let k0 be the least such.
p
Set = sup{k : p D0 k < k0 } < 1 . (If k0 = 1, then set = 0. If k0 > 1,
p
then actually = sup{k0 1 : p D0 }.)
If k0 > 1, then we may apply Claim 6.30 to D0 to get some D1 D0 of size 1 ,
some set {1 , . . . , k0 1 }, and some set {s1 , . . . , sk0 1 } C such that for all p D1 ,
p
p
1 = 1 , . . ., k0 1 = k0 1 , p(1 ) = s1 , . . ., p(k0 ) = sk0 . If k0 = 1, then we just
set D1 = D0 .
Let us now recursively define (i : i < 1 ) and ( pi : i < 1 ) as follows. Let
< 1 and suppose (i : i < ) and ( pi : i < ) have already been defined.
p
Set = sup{n i : i < }. (If = 0, then set = 0 = .) By the choice of
p
k0 , there is then some p D1 such that k0 > ; let p be some such p. Now
p
write B = { pi : i < 1 }. We have that if pi B, then i < k0i < . . . <
pi
n i+1 . We therefore have that any two conditions in B are compatible: if
p, q B, then we may define r C(1 ) by: supp(r ) = supp( p) supp(q) =
p
p q
q
{1 , . . . , k0 1 , k0 , . . . , n , k0 , . . . , n }, r (k ) = p(k ) = q(k ) for 1 k < k0 ,
p
p
q
q
r (k ) = p(k ) for k0 k n, and r (k ) = q(k ) for k0 k n. B is thus as
desired.

The combinatorial heart of this latter argument leads to the following lemma
which is very useful for the analysis of many forcings. The proof is pretty much the
same as the proof of the previous lemma. (Cf. Problem 6.6.)
Lemma 6.31 (-Lemma) Let be an uncountable regular cardinal, and let <
be an infinite cardinal such that < for all < and < . Let A []<
with A = . There is then some B A with B = which forms a -system , which
means that there is some r []< (the root of B) such that for all x, y B with
x = y, x y = r .
Notice that if M is a transitive model of ZFC and G is P-generic over M for some
P M and if is a cardinal of M[G], then is also a cardinal of M. The following
lemma provides a covering fact and gives a criterion for when cardinals of M will

6.2 Applications of Forcing

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

p PM is a function with domain .


For each < , let B = { : q p q PM ( ) = }.
Working in M, for each

B we may pick q p such that q M ( ) = .


If = , then q q , so
) < , as M |= P has the -c.c.
that M |= Card(B
Now set Y = {B : < }. Of course, Y M and Y X . If is the
cardinality of B inside M, then
M |= Card(Y )

(6.8)

<

It is straightforward to verify that (6.8) yields (6.7).


Now suppose that is a cardinal in M such that if = , then is regular.
Suppose that = Card M[G] () < . We may then cover the set X = by a set
Y M such that (6.7) holds true. Let us now argue in M. If + , or +
and < cf(), or < , then M |= Card(Y ) < , which is nonsense. Otherwise
cf() and = , so that is regular by hypothesis and hence in fact
= = , which contradicts < .

In the light of Theorem 5.31, the following result shows that CH cannot be decided
on the basis of ZFC.
Theorem 6.33 (P. Cohen) If ZFC is consistent, then so is ZFC + CH. In fact, if
ZFC is consistent, then so is ZFC + 20 = 2 .
Proof In the light of the discussion above (cf. the proof of Theorem 6.25), we may
argue under the hypothesis that there be a transitive model of ZFC.

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 = ,

D , = { p C() : n dom( p( )) dom( p( )) p( )(n) = p( )(n)} M


is dense in C(), so that F( ) = F( ) for = . F M[G] is therefore an
injection from into .
In order to verify M[G] |= C H , we need to verify 2M[G] . For this it will
be enough to show that M, M[G] have the same cardinals. However, as M |= C()
has the c.c.c. by Lemma 6.29, this immediately follows from Lemma 6.32.
We now prove the second part of the theorem. By our hypothesis that there be
a transitive model of ZFC, we have as in the proof of Theorem 6.25 that there is
J
some < 1 such that J is a (countable transitive) model of ZFC. Let = 2 .
Then C() J , and we may pick some G which is C()-generic over J . By the
argument for the first part of the theorem, 20 2 in J [G]. We are hence left with
having to verify that 20 2 in J [G].
Let x J [G], say x = G . For n < let
C()

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.

6.2 Applications of Forcing

109

The proof of Theorem 6.33 shows that if we assume that M |= ZFC + 0 = ,


then 20 = holds true in M[G] whenever G is C()-generic over M. Hence by
Lemma 4.22 we get that 20 may be any cardinal with cf() > (cf. Problem
6.9).
We now consider variants of Cohen forcing for cardinals above .
Definition 6.34 Let be a cardinal. Let C = < , i.e., the set of all f such
that there is some < with f : . For p, q C , let p q iff p q (iff
p  = q). The partial order (C , ) is called Cohen forcing at .
Of course, C = C. If < = (which is true for = and only possible for
regular ), then Card(C ) = , so that in this case forcing with C preserves all
cardinals above + by Lemma 6.32 (though cf. Problem 6.8). We now develop a
technique for showing that forcing with C never collapses cardinals below .
Definition 6.35 Let P = (P, ) be a partial order, and let be an infinite regular
cardinal. P is called < -closed iff for all < and for all sequences ( p : < )
of conditions in P such that p p for all there is some condition q P
for every < and for
with q p for all < . P is called < -distributive iff
every collection (D : < ) of open dense subsets of P, < D is open dense.
The proofs of the following two lemmas are trivial.
Lemma 6.36 Let be an infinite regular cardinal. Then C is < -closed.
Lemma 6.37 Let P = (P, ) be a partial order, and let be an infinite regular
cardinal. If P is < -closed, then P is < -distributive.
Not every < -distributive forcing is < -closed, cf. Problem 6.16.
Lemma 6.38 Let M be a transitive model of ZFC, let P = (P; ) be a partial order
in M, and let be a regular cardinal of M such that M |= P is < -distributive.
Let G be P-generic over M. Then3
<

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 }

If is an ordinal and X is any set or class, then we write < X for

<

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].

6.2 Applications of Forcing

111

Proof P is certainly < -closed inside M, so that every M-cardinal is still a


cardinal in M[G]. For < let
D = { p P : dom( p)},
and for < let

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

D X = { p P : < ( ( + 1) dom( p) { < : p( + ) = 1} = X )}


is in M and is dense in P, so that we may define F M[G], F : P() M[G] ,
by setting

F(X ) = the least with { < : (

G)( + ) = 1} = X.

F is certainly injective, so that Card(2 ) in M[G]. This shows that 2< =


in M[G].
P has size < in M, so that cardinals above (( < )+ ) M remain cardinals in
M[G].

If M is a transitive model of ZFC, and if G is (Col(1 , 20 )) M -generic over M,
then CH holds is M[G] by Lemmas 6.38 and 6.42. More generally, if is regular in
M and H is (Col( + , 2 )) M -generic over M, then 2 = + holds M[H ]. Theorem
6.46 will produce a stronger result.
Definition 6.43 Let be a regular cardinal, and let X be a set of ordinals which are
all of size . We let
Col (, X ) = { p : p is a function with domain X and X p( ) Col(, )}.

For p Col (, X ), let supp( p) = { X : p( ) = }. We let


Col(, X ) = { p Col (, X ) : Card(supp( p)) < }.
For p, q Col(, X ) we write p q iff for all X we have that p( ) q( ) in
the sense of Col(, ). If > , then we also write Col(, < ) for Col(, [, )).
The partial order (Col(, < ); ) is called the Levy collapse of to .

In what follows, we use ordinal arithmetic.

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

6.2 Applications of Forcing

A=

113

ai and q = (

i<i 0

pi ) {(i0 , A)}.

i<i 0

Write = i0 . We have that q P, q < p0 , and q by the choice of


As S, (6.9) is shown.
= A.

( pi : i < i 0 ). Also, q ( )
The forcing P used in the previous proof is forcing eqivalent to Col(, ) in the
sense of Lemma 6.48 below.
We now work towards showing that 1 may be singular in ZF, cf. Theorem 6.69.
Definition 6.47 Let P = (P; P ), Q = (Q; Q ) be partial orders. We call a map
: P Q a homomorphism iff for all p, q P,
(a) p P q = ( p) Q (q) and
(b) pP q = ( p)Q (q).
A homomorphism : P Q is called dense iff ran( ) is dense in Q, i.e., for every
q Q there is some p P such that ( p) Q q.
If : P Q is a homomorphism, then (a) implies that p P q = ( p) Q
(q), so that (b) gives p P q ( p) Q (q) for all p, q P.
Lemma 6.48 Let M be a transitive model of ZFC, let P = (P; P ), Q = (Q; Q )
M be partial orders and let : P Q be a dense homomorphism, where M.
If G P is P-generic over M, then H = { p Q : q G (q) Q p} is Q-generic
over M, G = { p P : ( p) H }, and M[G] = M[H ]. Also, if H Q is Q-generic
over M, then G = { p P : ( p) H } is P-generic over M and M[H ] = M[G].
Proof First let G P be P-generic over M, and set H = { p Q : q G (q) Q
p}. To see that H is a filter, let p, p H . Then there are q, q G with (q) Q p
and (q ) Q p . If r P q, q , then (r ) Q p, p . Now let D M be dense in
Q. We need to see D H = . Let D = {s P : r D (s) Q r } M. D
is dense in P: given p P, there is some r D with r Q ( p), and because
is dense there is some s P with (s) Q r ; in particular, (s) Q ( p), so that
s P p, and if q P s, p, then (q) Q (s) Q r ; i.e., q D and q P p. Now
let p D G. Then ( p) Q r for some r D, where p G, so that r D H .
Let us now show that G = { p P : ( p) H }. If ( p) H , then there is some
q G with (q) P ( p). As D = {r P : r P p r P p} is dense in P, we
may pick r D G. There is some s G with s P r , q; then (s), ( p) H ,
hence (s)||Q ( p), and hence s||P p, so that r ||P p. But then r P p, as r D, and
so p G.
Conversely, let H Q be Q-generic over M, and set G = { p P : ( p)
H }. It is again easy to see that G is a filter. Now let D M be dense in P. Let
D = {( p) : p D}. D is dense in Q: given p Q, there is some q P with
(q) Q p, as is dense, and there is some r D such that r P q, as D is dense;
but then (r ) D and (r ) Q p. Now let p D H . Then p = (q) for some
q D G.


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

{An \ { pn+1 } : n < } {q P : q p}

is an antichain of size n< n = .


Finally, let us assume that cf() > . Let r p and V P be such that
r : is surjective.

(6.10)

Let us suppose that every antichain A {q P : q r } is smaller than . For


every n < , let An be a maximal antichain in
{q P : q r q (n)
= }.
By our hypothesis, Card(An ) < for every n < , so that by cf() > there
cannot be a surjective function
f : (

{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)

6.2 Applications of Forcing

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).

some n < such that s (n)


= r . Let t P s be such that t decides (0),
There is then some p Col(, ) such that ( p) An+1 and ( p)||t. We must then
As P is separative,
have that ( p) (n)
= r , which implies that ( p) r G.

this gives that ( p) P r .
Definition 6.52 Let P = (P; ) be a partial order. P is called homogenous iff for
all p, q P there is some dense endomorphism5 : P P such that ( p) q.
Lemma 6.53 C is homogeneous. If is an ordinal, then C() is homogeneous.
5

i.e., a homomorphism to itself.

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

q(n), if n dom( p) dom(q) and r (n) = p(n),


(6.13)
(r )(n) = p(n), if n dom( p) dom(q) and r (n) = q(n), and

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

q()( ), if dom( p()) dom(q()) and r ()( ) = p()( ),


(r )()( ) p()( ), if dom( p()) dom(q()) and r ()( ) = q()( ), and

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

6.2 Applications of Forcing

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

Lemma 6.61 Let M be a transitive model of ZFC, and let P = (P, ) M be a


partial order. Let (v1 , . . . , vn ) be a formula, and let 1 , . . . , n M P be such that
P is homogenous with respect to 1 , . . . , n . Then either 1 PM (1 , . . . , n ) or else
1 PM (1 , . . . , n ).
Proof Otherwise there are p, q P such that p PM (1 , . . . , n ) and q PM
(1 , . . . , n ). Pick a dense endomorphism : P P such that ( p) q
and (1 ) = 1 , . . . , (n ) = n . By Lemma 6.56, we then have ( p) PM
( (1 ), . . . , (n )), i.e., ( p) PM (1 , . . . , n ), and q PM (1 , . . . , n ), so
that ( p), q cannot be compatible. Contradiction!

Corollary 6.62 Let M be a transitive model of ZFC, and let P = (P; ) be a partial
order. Let G P be P-generic over M. Let x M[G], where x M. Suppose also
that
M[G] |= y(y x (y, 1G , . . . , nG ))
for some formula and 1 , . . . , n M P such that P is homogenous with respect to
1 , . . . , n . Then x M.
such that
In particular, if P is homogenous, then every x M[G] OD M[G]
M
x M is an element of M. In particular, if P is homogenous, then HOD M[G]
M.
M
Proof Let y M. Then y x iff p G p PM ( y , 1 , . . . , n ). But because P
is homogenous with respect to y , 1 , . . . , n , p PM ( y , 1 , . . . , n ) is equivalent
to 1 PM ( y , 1 , . . . , n ). We may therefore compute x inside M as {y : 1 PM
( y , 1 , . . . , n )}.

As an example, we get that a Cohen real is not definable in the generic extension:
Corollary 6.63 Let M 
be a transitive model of ZFC, and let G be C-generic over
M. Then neither G nor G is definable in M[G] from parameters in M.
Definition 6.64 Let P = (P; P ), Q = (Q; Q ) be partial orders. The product
P Q of P, Q is defined to be P Q = (P Q; PQ ), where for ( p, q), ( p , q )
P Q we set ( p, q) PQ ( p , q ) iff p P p and q Q q .
Lemma 6.65 (Product Lemma) Let M be a transitive model of ZFC, and let P =
(P; P ) and Q = (Q; Q ) be partial orders in M. If G is P-generic over M and H
is Q-generic over M[G], then G H is P Q-generic over M. On the other hand,
if K P Q is P Q-generic over M, then, setting
G = { p P : q Q ( p, q) K }, and
H = {q Q : p P ( p, q) K }
G is P-generic over M and H is Q-generic over M[G].

6.2 Applications of 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

D = { p P : q Q q ( p, q ) D}. D M and D is clearly dense in P, so that


there is some p D G. But then there is some q Q q with ( p, q ) D, i.e.,
q D and q Q q.
Now D M[G], and thus there is some q D H . This means that there is
some p G with ( p, q) D and ( p, q) G H .
Now let K P Q be P Q-generic over M, and set G = { p P : q
Q ( p, q) K } and H = {q Q : p P ( p, q) K }. Let D P be dense in P,
where D M. Then D = {( p, q) P Q : p D} is clearly dense in P Q and
D M, so that there is some ( p, q) D K , i.e., p D G. This shows that G
is P-generic over M.
Now let D Q be dense in Q, where D M[G]. Let D = G , and p PM
where p G. Let
is dense in Q,
D = {( p, q) P Q : p p p PM q }.
D is dense below ( p , 1Q ): Given p P p , q Q, we have p PM is dense
There is then some p P p and some q Q q with p P q . Then
in Q.
M

( p , q ) D and ( p , q ) PQ ( p , 1Q ). Now let ( p, q) D K . Then p G



and p PM q , so that q G = D. Therefore q H D.
In the situation of Lemma 6.65, G and H are called mutually generic.
Lemma 6.66 Let M be a transitive model of ZFC, and let M. Let G be C()generic over M, and let x M[G]. Then x is C-generic over M in the following
sense: there is some C-generic H M[G] over M such that x M[H ]. In addition,
if 1M , then there is also some C()-generic K M[G] over M[H ] such that
M[G] = M[H ][K ].
Proof Fix x M[G], say x = G . Because M |= C() has the c.c.c, there is
some M C() and there is a sequence (An : n < ) M of countable antichains
in C() such that G = G and
(, p) = = n,
m
and p An for some n, m < ,
cf. the proof of Theorem 6.33. In particular, X =
most countable. Obviously,

{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.

Corollary 6.67 Let M be a transitive model of ZFC, and let M, 1M . Let G


be C()-generic over M. Then in M[G] there is no ODR -wellordering of the reals.
Proof Suppose that there is a formula (v0 , v1 , v2 , . . . , vn , vn+1 , . . . , vn+m ) and
there are 2 , . . . , n M OR and xn+1 , . . . , xn+m M[G] such that
M[G] |={(u, v) : (u, v, 2 , . . . , n , xn+1 , . . . , xn+m )}
is a wellordering of .
Let H M[G] be C-generic over M such that xn+1 , . . . , xn+m M[H ], and let
K M[G] be C()-generic over M[H ] such that M[G] = M[H ][K ]. The choice
of H and K is possible by Lemma 6.66.
M[H ][K ]
, so that by
Then every x M[G] = M[H ][K ] is OD{x
n+1 ,...,x n+m }
the homogeneity of C() in M[H ] (cf. Lemma 6.53) every such x is in M[H ] (cf.
Corollary 6.62). But this is nonsense!

Theorem 6.68 (P. Cohen) If ZFC is consistent, then so is ZF + AC.
Proof Let M be a countable transitive model of ZFC, and let G be C(1M )-generic
over M. Let
N = HODM[G]
M[G] .
We have that N |= ZF by Theorem 5.44. However, by Corollary 6.67 there is no
wellorder of the reals in N .

The following is a strengthening of Theorem 6.68. (Cf. also Problem 11.11.)
Theorem 6.69 (FefermanLevy) If ZFC is consistent, then so is ZF + cf(1 ) = .
Proof Let J be a countable model of ZFC. We aim to find a symmetric extension
of J in which ZF + cf(1 ) = holds true.
Let = J . Let G be Col(, < )-generic over J , and write N = J [G]. With
the help of Lemma 6.65, it is straightforward to see that 1N = +J . Now let
( ) =

{ J [G  ] : < },

and set
M = HOD(N ) {( ) } .
By Theorem 5.44, M |= ZF. We aim to verify that

6.2 Applications of Forcing

121

= 1M and M |= cf(1 ) = .

(6.15)

Notice that for every < , J [G  + 1] |= is countable, so that there is


some bijective f : with f ( ) . In particular, every < is countable
in M, i.e., 1M . Moreover, because (nJ : n < ) J ( ) , we have
that M |= cf() = . In order to verify (6.15) it thus suffices to show that is a
cardinal in M.
If were not a cardinal in M, then M |= is countable, and there would then
be some bijection f : with f M. Such a bijection cannot be an element
of ( ) . This is because if f J [G  ], say, where < , then G  is
Col(, < )-generic over J by the Product Lemma 6.65 and every J -cardinal
above (in particular, ) will remain a cardinal in J [G  ] by Lemma 6.32.
In order to show (6.15), it thus suffices to verify that

M = ( ) .

(6.16)

To this end, let f M. There is then a formula (v0 , v1 , v2 , . . . , vn , v1 , . . . ,


, v), there are ordinals , . . . , < , and there are f , . . . , f ( ) such
vm
2
n
1
m
that for all (n, ) ,
f (n) = N |= (n, , 2 , . . . , n , f 1 , . . . , f m , ( ) ).
Let < be such that f 1 , . . . , f m J [G  ]. By the Product Lemma 6.65,
G  is Col(, < )-generic over J , and G  [, ) is Col(, [, ))-generic over
J [G  ].
Claim 6.70 There is some J [G  ] such that G [,) = ( ) and is
homogenous for Col(, [, )).
Proof The proof for > is only notationally different from the proof for = ,
so let us assume that = . I.e., we assume that f 1 , . . ., f m J .
Let = +J = (+1 ) J , and let
= {(, p) : J <

J [G Col(, < )]}.


p Col(,<)
J
Let us verify that G = ( ) . First let f ( ) , say f = G . We may assume
that J , cf. the proof of Theorem 6.33. As f ( ) , there is some <
such that f J [G Col(, < )], and there is then some p G such
that p Col(,<)
J [G Col(, < )]. But then (, p) , so that
J
G
G
f = . Now let f G , say f = G , where (, p) for some p G.
Col(,<)
and
There is then some < with p J
J [G Col(, < )],
G

hence J [G Col(, < )], i.e., f = ( ) .

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

Col(, < )] = J [G Col(, < )]. (6.17)


J [(
G)

But we may now show (


) = as follows. With the help of (6.17), we have
Col(,<)

(, p) J < p J

J [G Col(, < )]

Col(,<)

(
) J < ( p) J

Col(,<)

(
) J < ( p) J

Col(, < )]
( ) J [(
G)
( ) J [G Col(, < )]

((
), ( p)) .

Therefore = {(, p) : (, p) } = {((


), ( p)) : (, p) } = ( ), as
desired. This shows Claim 6.70.

By Claim 6.70 and Corollary 6.62 we get that in fact f J [G  ]. Hence

f ( ) . We verified (6.16).
Elaborate forcings are studied e.g. in [9, 24, 37] and [44].

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

6.5. Suppose that p PM x(x, 1 , . . . , n ). Show that there is some M P


such that p (, 1 , . . . , n ). (This is called fullness.) Let X (H ) M ,
where is regular in M and P X . Let X [G] = { G : M P X } and
(H ) M [G] = { G : M P (H ) M }. Show that (H ) M [G] = (H ) M[G]
and (using fullness)
X [G] (H ) M[G] .
6.6. Prove Lemma 6.31!
6.7. Let H be C-generic over M. Let s < , and let Hs = {( p  [dom(s),
dom( p))) (s  dom( p)) : p H }. Show that Hs is C-generic over M.
6.8. Let be an infinite cardinal of M, and write = ( < ) M . Let H be (C ) M generic over M. Show that in M[H ], there is a surjection f : . (Cf. the
proof of Lemma 6.42.) Conclude that (C ) M collapses exactly the M-cardinals
in the half-open interval (, ].
6.9 Assume M to satisfy GCH. Let M be an M-cardinal such that M |=
cf() > . Show that if H is C()-generic over M, then M[H ] |= 20 = .
More generally, show that if is an infinite regular cardinal in M, M is
an M-cardinal with M |= cf() > , and if H is C ()-generic over M then
M and M[H ] have the same cardinals and M[H ] |= 2 = .
6.10. Show that the converse to Lemma 6.38 is also true, i.e., if P is separative and
<

M M[G] = < M M,

then P is < -distributive in M.


6.11. Let M be a transitive model of ZFC such that if = M OR, then Card() =
1 . Show that there is a transitive model M of ZFC with M OR = and
M = M.
6.12 (Solovay) Let us assume G and K to be mutually P-generic over M. Show that
M[G] M[K ] = M. [Hint. Let G = K , where and are P-names. We
may also construe and as (P P)-names, and we may pick ( p, q) P P
such that ( p, q) = . Show that for every y M, p decides y , i.e.,
/ .]
p P y or p P y
6.13. Let be inaccessible in M. Show that if H is C()-generic over M, then is
weakly inaccessible in M[H ].
6.14. (R. Solovay) Suppose that M |= 1L = 1 and A 1M , A M. Show that
there is some poset R which has the c.c.c. such that if H is R-generic over M,
then in M[H ] there is some x with M[H ] |= A L[x]. [Hint. First work
in M. Let {xi : i < 1 } L be an almost disjoint collection of subsets of ,
cf. Problem 1.3. Let
R = {(s, t) : s < 2, t [1 ]< },

124

6 Forcing

ordered by (s , t ) (s, t) iff s s, t t, and if i t A, then


{n dom(s ) \ dom(s) : s (n) = 1} xi = .
Show that R has the c.c.c. Now stepping out of M, if H 
is R-generic over M
and if x is such that its characteristic function is {s : t (s, t) G},
then i A iff x and xi are almost disjoint.]
6.15. (a) Suppose that M |= S 1 is stationary. Let M |= P has the c.c.c. or
is -closed. Show that M[G] |= S is stationary.
(b) Let M |= S is stationary, where is uncoutable and regular, and
cf() = for all S. Suppose also that M |= P is -closed. Then
M[G] |= S is stationary. [Hint. Fix p P such that p PM is club
in . In M pick some nice X H with p X and sup(X ) S.
Pick (n : n < ) cofinal in = sup(X ). Construct ( pn : n < ), a
decreasing sequence of conditions in X , such that pn \ n = . Let

q be stronger than all pn . Then q S.]


6.16. Assume M |= S 1 is stationary. Show that there is some -distributive
forcing Q M such that if H is Q-generic over M, then S contains a club in
M[H ]. [Hint. In M, let
Q = { p : < 1 ( p is a closed subset of S, otp( p) = + 1)},
ordered by end-extension.] Show that in fact if T S is stationary in M, then
T is still stationary in M[H ].
More generally, let be an infinite regular cardinal in M, and let M |= S +
is stationary and < -closed. Show that there is a < -closed -distributive
forcing Q M such that if H is Q-generic over M, then S contains a club in
M[H ].
6.17. (J. Silver) Let H , H be transitive models of a sufficiently large fragment
of ZFC, let P H be a partial order, let : H H be an elementary
embedding, let G be P-generic over H , and let K be (P)-generic over H such
that G H . There is then an elementary embedding : H [G] H [K ]
such that .
6.18. Let X be a large cardinal concept, e.g., X = inaccessible, measurable, etc. We
say that is an X -cardinal is preserved by small forcing iff the following
holds true. Let be an X -cardinal in M, and assume P (V ) M to be a poset.
Then is still an X -cardinal in M[G]. Show that the following statements are
preserved by small forcing. is inaccessible, is Mahlo, is weakly
compact, and is measurable. [Hint: To prove that is measurable is
preserved by small forcing, let U be any measure on in M. Show that
U = {Y : X U Y X },
as defined in M[G], witnesses that is still measurable in M[G].]

6.3 Problems

125

6.19. In M, let be regular, and let F be a non-trivial filter on . For X , Y F +


(in the sense of Definition 4.30) let us write X F + Y iff
\ (X \ Y ) = ( \ X ) Y F.
Let G be (F + , F + )-generic over M. Show that in M[G], G is a non-trivial
M-ultrafilter which extends F in the following sense.
(a)
(b)
(c)
(d)
(e)

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.

We may then make sense of Ult(M; G) M[G].


If F is -closed in M ( < ), then G is M-
-closed in the sense that if
(X i : i < ) M, X i G for all i < , then i< X i G. If F is normal
in M, then G is M-normal in the sense that if (X i : i < ) M, X i G for
all i < , then i< X i G.
6.20. Let 0 . Recall that F + is the club filter on + , and write F = F + .
Assume that (F + , F + ) has the ++ -c.c. Let
: H
= X H(2 + )+ ,
where + 1 X , Card(X ) = , (F + , F + ) X , and H is transitive. Let
() = + and (Q) = (F + , F + ). Write g = {X + : (X )}. Show
that g is Q-generic over H . [Hint. Show that if {Ai : i < + } ran( ) is a
maximal antichain in (F + , F + ), then
{ < + :

Ai }

i<

contains a club C in ran( ), so that C.] Conclude that if M |= (F + ,


F + ) M has the ++ -c.c., and if H is (F + , F + ) M -generic over M, then
ult(M; H ) is well-founded.
6.21. (Petr Vopenka) Show that for every there is some partial order V = V()
HOD such that for every A , A V , there is some G V such that G is
V-generic over HOD and A HOD[G].
[Hint. Let D = {Y P() : Y is OD}, and let f be OD such that f : D

(D, ), so that V = (, ) HOD.


is bijective. Let be such that (, ) =
Show that
G = { < : A ( )}
is V-generic over HOD and A HOD[G].]

126

6 Forcing

6.22. Let g be a group of automorphisms of P. Let F be a collection of subgroups


of g which satisfies the following.
(a)
(b)
(c)
(d)

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.

(Such an F is called a filter on g.) For V P let us write


sym g ( ) = { g : ( ) = }.
A name V P is called symmetric iff sym g ( ) F, and is called hereditarily symmetric iff for every (finite) sequence ((i , pi ) : 0 i n) such that
0 = and (i+1 , pi+1 ) i for 0 i < n we have that all i , 0 i n, are
symmetric.
Now let
N = { G : M |= M P is hereditarily symmetric}.
Show that N is a model of ZF.

Chapter 7

Descriptive Set Theory

7.1 Definable Sets of Reals


Descriptive set theory is the study of definable sets of real numbers. However, rather
than working with R, descriptive set theorists often work with a space which can be
shown to be homeomorphic to the space of all irrational numbers.
Let X = be an arbitrary set. If s < X then we declare Us = {x X : s x}
to be a basic open 
set. A X is declared to be open iff A is the union of basic
open sets. (As = , is also open.) Complements of open sets are called closed.
Notice that each Us is also closed, because

X \Us =


{Ut : lh(t) = lh(s) t = s}.

Here, lh(t) = dom(t) = t is the length of t. If X = , then the space , together


with the topology just defined, is called the Baire space. We shall often refer to
the elements of as reals. If X = {0, 1}, then the space 2, together with the
topology just defined, is called the Cantor space. In this chapter, we shall focus our
attention on the Baire space, but most statements carry over, mutatis mutandis, to
the Cantor space.
If x, y , x = y, then their distance d(x, y) is defined to be 21n , where n is
least such that x(n) = y(n). It is easy to see now that the topology we defined on
is exactly the one which is induced by the distance function d, so that is a Polish
space, i.e. a complete seperable metric space. (Cf. Problem 7.1.)
A tree T on X is a subset of < X which is closed under initial segments, i.e.,
if s T and n lh(s), then s  n T . Then (T,  T ) is a tree in the sense of
Definition 4.43. If T is a tree on X , then we write [T ] for the set of all x X such
that x  n T for all n < . A tree T on X is called perfect iff T = and whenever
s T , then s has X pairwise incompatible extensions t in T , i.e., there is (ti : i < X )
such that for all i, j < X , s ti , ti T , ti
/ t j , and t j
/ ti . Perfect trees admit a
CantorBendixson analysis (cf. p. 5), cf. Problem 7.5.
R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_7,
Springer International Publishing Switzerland 2014

127

128

7 Descriptive Set Theory

If T = is a tree on , then T is perfect iff whenever s T , then there are t,


/ t , and t
/ t. Recall that a set of reals A is called
t T with s t, s t , t
perfect iff A = , A is closed, and every element of A is an accumulation point of
A, cf. Definition 1.7.
Lemma 7.1 If A is closed, then A = [T ] for some tree T on . On the other
hand, if T is a tree on then [T ] is closed.
Moreover, A is perfect iff A = [T ] for some tree T = on which is
perfect.
Proof We show the first part of the lemma. Let A be closed. Let T = {s
x A s x}. It is easy to see that A [T ]. Let x [T ]. For each n <
there is some xn A with x  n xn . But then x = limn xn A, because A is
closed. Therefore, A = [T ].
It is easy to verify that [T ] is closed whenever T is a tree on .
The second part of the lemma is easy to check.

< :

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)

Definition 7.3 A set A is called analytic iff there is a tree T on with


A = p[T ]. A is called coanalytic iff \A is analytic.
We will show below, cf. Lemma 7.11, that there are analytic sets which are not
Borel. A classical result of Souslin says that a set of reals is Borel if and only if
it is analytic as well as coanalytic, cf. Theorem 7.5.

7.1 Definable Sets of Reals

Lemma 7.4 Let A


coanalytic).

129

be a Borel-set. Then A is analytic (and hence also

Proof Let us first show the following two statements.



1. If every An , n < , is analytic, then so is n< An .
2. If every An , n < , is analytic, then so is n< An .
(1): Let An = p[Tn ], where Tn is a tree on . Let T on be defined by1
(s, t) T iff s = t =
n < t (t = n t (s  (lh(s) 1), t ) Tn ).

It is straightforward to verify that p[T ] = n< An .
(2): Again let An = p[Tn ], where Tn is a tree on , and let e:
be bijective such that e(n, k) e(n, l) whenever k l. If t < , say t =
(m 0 , . . . , m i1 ), and n < , then we write t n for (m e(n,0) , . . . , m e(n,k1) ), where k
is least with e(n, k) i. (If e(n, 0) i, then t n = .) We now let T on be
defined by
(s, t) T iff lh(s) = lh(t)
n < (s  lh(t n ), t n ) Tn .

It is straightforward to verify that p[T ] = n< An .
Now Lemma 7.1 quite trivially yields that every closed set is analytic. In particular,
every basic open set Us is analytic, and hence by (1) every open set is analytic. (1)
and (2) then imply that every Borel set is analytic.

Theorem 7.5 (Souslin) Let A . Then A is Borel if and only if A is analytic
as well as coanalytic.
This theorem readily follows from Lemma 7.4 and the following one, Lemma 7.6.
If A and B are disjoint sets, then we say that C separates A and B iff C A and
C B = .
Lemma 7.6 Let A, B disjoint analytic sets. Then A, B are separable by a
Borel set C.
Proof Let A = p[T ], B = p[U ], where T, U are trees on 2 . For s, t < , let
Ast = {x A: s x y(t y (x, y) [T ])},
and let
Bts = {x B: s x y(t y (x, y) [U ])}.
We have that A = A and B = B, and we always have
Here, n t is that sequence which starts with n, followed by t (0), . . . , t (lh(t )1). This notation
as well as self-explaining variants thereof will frequently be used in what follows.

130

7 Descriptive Set Theory

(s, t) T = Ast =


{Ast mn : (s n, t m) T }

and
(s, t) U = Bts =

{Bts mn : (s n, t m) U }.

Let us assume that A, B are not separable. We aim to derive a contradiction.


Let us define four reals x, y, u, v. We shall define x(n), y(n), u(n), v(n) recursively. We shall inductively maintain that (x  n, y  n) T , (u  n, v  n) U ,
x n
u n
and A y n , Bv n are not separable, which is true for n = 0.
Now suppose that x  n, y  n, u  n, v  n have been chosen in such a way that
x n
u n
(x  n, y  n) T , (u  n, v  n) U , and A y n , Bv n are not separable.
Assume that for all i, j, k, l < such that (x  n i, y  n j) T and
(u  n k, v  n l) U there is a Borel set C i,k
j,l separating
x n i

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 .

It is then easy to see that the Borel set




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

There must hence be i, j, k, l < such that


(x  n i, y  n j) T, (u  n k, v  n l) U,
and

x n i

u n k

A y n j and Bv n l

cannot be separated by a Borel set. Set x(n) = i, y(n) = j, u(n) = k, v(n) = l.


 u n
 x n
Now of course n A y n = {x} and n Bv n = {u}. We have that x A, as witnessed by (x, y) [T ], and we have that u B as witnessed by (u, v) [U ]. As A, B
are disjoint, x = u, and we may pick two disjoint open sets F, G such that x F and
x n
y G. Because F is open, A y n F for all but finitely many n. For the same reason,
u n

Bv n G for all but finitely many n. In particular, there is some n < such that

7.1 Definable Sets of Reals

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

Let < 1 . Using any bijection of with , we may construe S  as a tree on


. The sets p[S  ] and \A are then disjoint analytic sets and may hence by
Lemma 7.6 be separated by a Borel set B .

132

7 Descriptive Set Theory

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

with A = [T ] and T is perfect, i.e., whenever (s0 , . . . , sk ) T then there are


extensions (t0 , . . . , tk ), (t 0 , . . . , t k ) T of (s0 , . . . , sk ) with lh(t0 ) = lh(t 0 ) and
(t0 , . . . , tk ) = (t 0 , . . . , t k ) (Cf. Lemma 7.1).
We may now define the projective hierarchy. Let T be a tree on k . The
projection of T onto the first l k many coordinates, written pl [T ] or just p[T ], is
the set of all (x0 , . . . , xl1 ) such that there are (xl , . . . , xk1 , f ) with
(x0 , . . . , xl1 , xl , . . . , xk1 , f ) [T ].
A set A ( )k is analytic iff there is a tree T on k with A = pk [T ]. (Hence
A is analytic iff
A = {(x0 , . . . , xk1 ): xk (x0 , . . . , xk1 , xk ) B}
for some closed set B ( )k+1 .)
A ( )k is coanalytic iff A is the complement of an analytic set, A = ( )k \B,
where B is analytic. The analytic sets are also called 1 , the coanalytic sets 1 .
A ( )k is 1

n+1

iff

A = {(x1 , . . . , xk1 ): xk (x1 , . . . , xk1 , xk ) B}


for some 1 set B ( )k+1 . A ( )k is 1
1
n+1

n+1

iff A is the complement of a

set.

Definition 7.10 A set A ( )k , some k < , is called projective iff there is some
n < such that A is 1 .
n

7.1 Definable Sets of Reals

133

Of course there are exactly 20 projective sets.


We now aim to verify that 1 = 1 by showing that there is a universal 1 -set
1

which cannot be 1 .

Let us pick a bijection


e: {(s, t): s < , t < , lh(s) = lh(t)}
such that for all s, t and i < , e(s  i, t  i) e(s, t). Let us say that u <
codes a finite tree iff Tu = {(s, t): s < , t < , lh(s) = lh(t), u(e(s, t)) = 1}
is a (finite) tree, i.e., is closed under initial segments.
We may now define a tree U on as follows. We set (s, t, u) U iff
s, t, u < , lh(s) = lh(t) = lh(u), u codes a finite tree, and if i < is such that
e(s  i, t  i) dom(u), then u(e(s  i, t  i)) = 1.
We claim that for all A , A is 1 iff there is some z such that
1

A = p[Uz ], where Uz = {(s, t) : u ((s, t, u) U u = z  lh(u))}.

(7.4)

Well, let A be 1 , and let A = p[T ], where T is on . Let z ,


1

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

The set B = {(x, z) : y (x, y, z) [U ]} is easily be seen to be 1 , and A


1

is 1 iff there is some z with


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

Let B ( )2 be a universal 1 -set. We claim that B cannot be also a 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

B ( )2 , as a subset of , in the following way. If x, y , then let x y


denote that z such that z(2n) = x(n) and z(2n + 1) = y(n) for all n < .
(Clearly, (x, y) x y is a continuous, in fact Lipschitz, bijection between ( )2

134

7 Descriptive Set Theory

and .) If B ( )2 , then B = {x y : (x, y) B} codes B in the sense that


B may be easily read off from B .
Let us define two important sets which are in 1 \ 1 , namely WF and WO.
1 1

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.

7.1 Definable Sets of Reals

135

k i(k) i i(k) (lh(e(n i )) k e(n i )  k = e(n i(k) )  k).


This gives that


e(n i(k) )  k [Tx ],

k<

so that x B. We have verified that x B iff g(x) WO.


If x WO and Rx is defined as in (7.6), then we write ||x|| for the order type
of Rx , i.e., for the (countable) ordinal such that (; )  (; Rx ), cf. Definition
3.22.
Lemma 7.12 (Boundedness Lemma) Let A WO be analytic. Then {||x|| : x A}
is bounded below 1 .
Proof Suppose not, i.e., let A WO be analytic such that {||x||: x A} is
unbounded in 1 . Let B be an arbitrary coanalytic set. There is a tree T
on such that x B iff (Tx ; ) is well-founded. We may thus write x B
iff there is a ranking f : Tx , where = ||z|| for some z A. It is straightforward to verify that B is then 1 . Therefore, every coanalytic set would be analytic.
1

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

One also defines

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

As every coanalytic set B ( )2 is of

the form p2 [T ], where T is on 2 1 , by the proof of Lemma 7.8, we get that


A = p1 [T ]. Via some bijection g: 1 1 , we thus see that A = p[S], where
S is on 1 . This tree S is also called the Shoenfield tree for A.
This argument shows:
Lemma 7.13 If A is 1 then A is 1 -Souslin.
2

Lemma 7.14 Let A be n -Souslin, where n < . Then A =


each Ai is analytic.


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

7 Descriptive Set Theory

By the inductive hypothesis, for each < n , p[T  ] =


each Ai is analytic. Therefore,
A = p[T ] =

<n i<n1

If U is a tree on
the tree

i<n1

Ai , where

Ai ,


a representation as desired.
k

and (s0 , s1 , . . . , sk ) U , then we write U(s0 ,s1 ,...,sk ) for

{(t0 , t1 , . . . , tk ) U : (s0 t0 s1 t1 . . .sk tk )(s0 t0 s1 t1 . . .sk tk )}.

Theorem 7.15 (Souslin, Mansfield) Let A be -Souslin. Then either A has


at most elements or else A contains a perfect subset.
Proof This is shown by a Cantor Bendixson analysis of A, cf. the proof of
Theorem 1.9.
If U is a tree on then we set
U = {(s, t) U : Card( p[U(s,t) ]) > 1}.

(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

T for this tree.


The argument now splits into two cases.
Case 1: T = .
Let x A = p[T ]. Pick g such that (x  n, g  n) T for all n. As T = ,
there must be a largest i such that (x  n, g  n) Ti for all n. Let n be maximal
such that (x  n, g  n) Ti+1 . Then p[(Ti )(x n+1,gn+1) ] has exactly one element,
namely x, as
(x  n + 1, g  n + 1) Ti \Ti+1 .
We have seen that
A=


{ 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) .

7.1 Definable Sets of Reals

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

as well. We write n1 instead of n1 (0), n1 instead of n1 (0), and 1n instead of 1n (0).


The following is often very useful.
Lemma 7.17 Let x . A set A ( )k is n1 (x) iff there is a formula such
that3 for all y ,
y A z 1 z 2 . . . Qz n
(V ; , x, y, z 1 , . . . , z n ) |= (x, y, z 1 , . . . , z n ).

(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

7 Descriptive Set Theory

(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

n = 1, then A is 11 (x). We shall use Problem 5.2. Let in what follows (; E A ) =

(HF; ) be as in Problem 5.2. The relation (n) = u is ZFC


by Problem 5.2.
1
First let be 1 . We let (s, t) T iff s, t < , lh(s) = lh(t), and (Vlh(s) ;
, x Vlh(s) , s, t) |= (x Vlh(s) , s, t). It is easy to see that A = p[T ].
If u ,
where is n , then

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 ].

7.1 Definable Sets of Reals

139

Proof Suppose that A = , i.e., [T ] = . This means that the relation


(T, ) L[T ]

(7.9)

is ill-founded in V . By the absoluteness of well-foundedness, cf. Lemma 5.6, the


relation (7.9) is then ill-founded in L[T ], which implies that [T ] L[T ] = , i.e.,
A L[T ] = p[T ] L[T ] = .
Now let (Ti : i ), where < + , be the CantorBendixson analysis of
A as in the proof of Theorem 7.15. If U L[T ] is a tree on , and if U is
defined as in (7.7), then U computed in V is the same as U computed in L[T ];
this is because Card( p[U(s,t) ]) > 1 is absolute between V and L[T ] by Lemma
5.6. We therefore in fact get that the construction producing (Ti : i ) is absolute
between V and L[T ], so that
(Ti : i ) L[T ].

(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

7 Descriptive Set Theory

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) =

||sn ||Tx , ifsn Tx


0
, else.

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

ill-founded, and we may pick some y [Tx ]. Let y  i = sn i for i < . If k kn i ,


kn i+1 is large enough, then
n i+1 = ||sn i+1 ||Txk < ||sn i ||Txk = n i .
Hence (n i : i < ) is a descending sequence of ordinals. This contradiction shows

that x A after all. It is easy to see that n (x) n .
We shall now pick some x A. Let x B iff x A and for all y and for all n,
[x  n = y  n m < n(y A m (x) = m (y))]
[x(n) < y(n) (x(n) = y(n) (y
/ A (y A n (x) n (y))))].
A moment of reflection shows that B = {x} A for some x.
It remains to be shown that B is 11 . Well, m < n(y A m (x) = m (y))
says that for all m < n there are order-preserving embeddings f : (Tx )sm (Ty )sm
and g: (Ty )sm (Tx )sm , and is hence 11 by Lemmas 7.17 and 7.18. Similarily,
y
/ A (y A n (x) n (y))

7.1 Definable Sets of Reals

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

(b) 1 has the separation property.


1

(c) 1 does not have the reduction property.


1

Proof (a) Let A, B 1 , A, B . Let C = (A {0}) (B {1}) 1 , and


1

let F : , F 1 uniformize C. Then A = {x A : F(x) = 0} 1 and


1

B = {x B : F(x) = 1} 1 reduce A and B.

(b) This easily follows from (a).


(c) Let us assume that 1 has the separation property. Let U ( )2 be a univer1

sal 1 -set, and define A = {x : ((x)0 , x) U } and B = {x : ((x)1 , x)


1

B 1 such that A and B reduce A and B,


U }.5 As A, B 1 , we may pick A,
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

7.2 Descriptive Set Theory and Constructibility


Definition 7.26 We say that 1 is inaccessible to the reals iff 1L[x] < 1 for every
x .
Lemma 7.27 1 is inaccessible to the reals iff 1V is an inaccessible cardinal in
L[x] for every x .
5

Here, (x)0 and (x)1 are defined to be the unique reals such that (x)0 (x)1 = x.

142

7 Descriptive Set Theory

Proof Suppose that 1 is inaccessible to the reals, and let x . We have to


show that 1V is not a successor cardinal in L[x]. Suppose that 1V = +L[x] . Let
f : , f V be a bijection, and define z by

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

following statement gives a characterization of when every 1 -set of reals has a


1

perfect subset in terms of inner model theory, cf. Corollary 7.29.


Theorem 7.28 Let x . The following statements are equivalent.
(1) Every uncountable 21 (x)-set of reals has a perfect subset.
(2) Every uncountable 11 (x)-set of reals has a perfect subset.
(3) 1L[x] < 1 .
Let P( ). We say that has the perfect subset property iff every uncountable A has a perfect subset.
Corollary 7.29 The class of coanalytic sets has the perfect subset property if and
only if 1 is inaccessible to the reals.
Proof of Theorem 7.28. Let us suppose that x = 0. The proof relativizes to any real
different from 0.
(1) = (2) is trivial. Let us prove (2) (3). Suppose that 1L = 1 . Let x A
iff
/ WO ||y|| = ||x||)).
x L x WO y ( L) (y < L x (y
A is 21 by Lemma 7.19 and Problem 7.6. By the Boundedness Lemma 7.12, if
B A is analytic, {||x||: x B} is bounded below 1 , and hence B is countable.
(In particular, A does not contain a perfect subset.)
As A is 21 , there is a coanalytic set B ( )2 such that
A = {x : y (x, y) B}.
By the Uniformization Theorem 7.23, let F B be a uniformizing function whose
graph is 11 .

7.2 Descriptive Set Theory and Constructibility

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

of all closed A . Having defined 0 and 0 for all < , let 0 be



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

the set of all Borel subsets of .


A 0 , . . . , A k , B 0 , . . . , B l ) and (M; A0 , . . . , Ak , B0 , . . . , Bl ) be mod7.4. Let ( M;
els of the same type, let N be a transitive model of ZFC, and assume
A 0 , . . . , A k ) N , (M; A0 , . . . , Ak , B0 , . . . , Bl ) N , M is countable
( M;
in N , and in V there is some elementary embedding
A 0 , . . . , A k , B 0 , . . . , B l ) (M; A0 , . . . , Ak , B0 , . . . , Bl ).
: ( M;
Show that in N there are B 0 , . . . , B l and an elementary embedding

144

7 Descriptive Set Theory

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

7.7. Show that A


<s , where s < ,
<
such that for all s, t with s t, <
t is an order on lh(t) which extends
<s , and for all x , x A iff <x = sx <s is a well-ordering. (Hint:
Proof of Lemma 7.8.)
Let (M; E), (M , E ) be models of the language L of set theory. We say that
(M ; E ) is an end-extension of (M; E) iff M M , E E , and if x M,
y M , and y E x, then y M.
7.8. Show that for all countable transitive M there is a end-extension (M ; E ) of
(M;  M) such that (M ; E ) |= V = L. [Hint. This holds in L. Then use
Corollary 7.21.] If M\L = , then (M ; E ) cannot be well-founded.
7.9. Show that if there is a transitive model of ZFC + there is a supercompact
cardinal, then some such model exists in L. [Hint. Corollary 7.21.]
7.10. Fill in the details in the proof of Lemma 7.19! Show that there is a real x
which is not an element of L may be written in a 31 fashion. Conclude that
it is consistent to have a non-empty 21 set A such that A L = .
(Compare Corollary 7.21.)
1 -formula is equivalent to a HC -formula in
7.11. Let n < . Show that every n+1
n
1 (z), where z . There is then
the following sense. Let A be n+1
a n -formula (v, w) such that for all x , x A HC |= (x, z).
Conclude that if z , (v) is 1 , and V |= (z), then L[z] |= (z). [Hint.
Corollary 7.21.] Show also that it is consistent to have some a HC and
1 -formula (v) such that V |= (a), but L[a] |= (a). [Hint. a = 1L ,
(v) v is countable, and V is Col(, 1L )-generic over L.]

7.3 Problems

145

Let A . A function : A OR is called a norm on A.


7.12. Let A be 11 . Show that there is a norm : A OR on A such that
there are R, S ( )2 , R 11 and S 11 , such that for all y A,
(x, y) R (x A (x) (y)) and
(x, y) S (x A (x) (y)).
[Hint. Proof of Theorem 7.23.] Use this to show that 11 has the reduction
property.
Let A . A sequence (n : n < ) of norms on A is called a scale on A iff
the following holds true. Let {xk : k < } A be such that x = limk xk
and such that for every n < there are k(n) < and n such that n (xk ) = n
for every k k(n). Then x A and n (x) n for all n < .
7.13. Let A , and let (n : n < ) be a scale on A. Let (s, f ) T iff s < ,
f is a finite sequence of ordinals of the same length as s, and there is some
x A such that s = x  lh(s) and f (n) = n (x) for all n < lh(s). Show that
A = p[T ].
7.14. Let A be 11 . Show that there is a scale (n : n < ) on A such that
there are R, S ( )2 , R 11 and S 11 , such that for all y A,
(n, x, y) R (x A n (x) n (y)) and
(n, x, y) S (x A n (x) n (y)).
[Hint. Proof of Theorem 7.23.] Use this to derive the conclusion of Theorem
7.23.
Let x, z . Then x is called a 11 (z) singleton iff {x} is a 11 (z), and hence
11 (z), set.
7.15. Let x, z . Show that x is a 11 (z) singleton iff x J1z . [Hint. =: Let
{x} = p[T ] where T on is in J1z [z]. If s < , s = x  lh(s), then
Ts = {(s , t ) T : s s s s} is well-founded and hence has a ranking
in J1z [z] by Problem 5.28. Then use Problem 5.25 (c).]
7.16. Show that for every z there is a 11 (z)-set A such that A does not contain
any 11 (z) singleton. Show that there is actually such an A which is 10 (z),
i.e., A is closed and A = p[T ] for some tree T on which is definable over
(V ; , z).
Conclude that Theorem 7.23 is false for 11 and that in fact there is a closed
R ( )2 which cannot be uniformized by an analytic function. [Hint. Suppose
that every 10 (z) set contains a 11 (z) singleton. Say A is 11 (z). Then
x A y 11 (x z) (x, y) B,

(7.12)

146

7 Descriptive Set Theory

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

In this chapter we shall construct a model of ZF in which every set of reals is


Lebesgue measurable and has the property of Baire.

8.1 Lebesgue Measurability and the Property of Baire


Definition 8.1 Let s < , and let Us = {x : s x}. We recursively define
the measure (Us ) of Us , for s < as follows. Set (U ) = ( ) = 1. Having
1

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

If B is arbitrary, then we define the outer measure (B) of B to be


inf{(A): B A A is open}.
A set B is called a null set, or just null, iff (B) = 0.

It is easy to verify that if every Bn , n < , is null, then so is n< Bn .
Usually, a set B is called Lebesgue measurable iff for all X ,
(X ) = (X B) + (X \ B).

(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

Definition 8.2 Let A . We say that A is Lebesgue measurable iff there is a


Borel set B such that
AB = (A \ B) (B \ A) is null.
The definitions given carry over, mutatis mutandis, to the Cantor space 2, with
1
.
the difference that if s < 2, then for Us = {x 2 : s x}, (Us ) = 2lh(s)
Let B be the -algebra of all Borel sets B . For A, B B, let us write
A B iff A B modulo a null set, i.e., iff A\B is null. Write A B iff A B
and B A (i.e., iff AB is null), and let [A] denote the equivalence class of A with
respect to , i.e. [A] = {B: B A}. The order on B induces an order, which we
shall also denote by , on the set of of all equivalence classes by setting [A] [B]
iff A B. (Notice that if A [A] and B [B], then A B iff A B .)
We shall write B/null for the set {[A]: A B (A) > 0}, equipped with the
order . The partial order B/null is called the measure algebra, or, random algebra.
Well see later, cf. Lemma 8.8, that forcing with B/null amounts to adding a single
real, a random real.
Recall Definition 1.11 for R which, mutatis mutandis, carries over to . Let
A . Then A is nowhere dense iff for all nonempty open B there is some
nonempty open B B such that B A = . A set B is called meager (or
of first category) iff B is the countable union of nowhere dense sets.
Usually, a set B is said to have the Baire property iff there is some open
set A such that BA is meager. It is not hard to verify that the family of sets
which have the Baire property forms a -algebra containing all the open sets, so
that in particular all Borel sets have the Baire property. (Cf. Problem 8.2.)
For our purposes, we may then define the Baire property as follows.
Definition 8.3 Let A . We say that A has the Baire property iff there is a
Borel set B such that
AB = (A \ B) (B \ A) is meager.
Again the definitions given carry over, mutatis mutandis, to the Cantor space 2.
We may now define a partial order B/meager in exactly the same way as we
defined the measure algebra B/null, except that we start with declaring A B iff
A B modulo a meager set, i.e., iff A\B is meager, for A, B B. It turns out,
though, that forcing with B/meager is tantamount to forcing with Cohen forcing,
which is why we refer to B/meager as the Cohen algebra.
Lemma 8.4 There is a dense homomorphism
i: C B/meager.
Proof Let [B] B/meager, so that B is a nonmeager Borel set. There is a nonempty open set A such that A\B is meager, and there is hence a nonempty

8.1 Lebesgue Measurability and the Property of Baire

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

for i < 1 . Notice that


 Bi B j is null whenever i = j, as {[Bi ]: i < 1 } is
an antichain, so that j<i (Bi B j ) is null whenever i < 1 . But then Bi =

Bi \ j<i (Bi B j ) Bi , i.e., [Bi ] = [Bi ] for all i < 1 , Bi B j = whenever
i = j, and (Bi ) > 0 for all i < 1 . By the Pigeonhole Principle, there will be an
n < such that (Bi ) > n1 for 1 many i < 1 . This gives a contradiction!

1
1
We now want to verify that all 1 as well as all 1 -sets are Lebesgue measurable

and have the Baire property.


Let P be an atomless partial order. By Lemma 6.11, there is then no P-generic filter
over V . As in the following definition, it is often very convenient, though, to pretend
that there is and say things like pick G which is P-generic over V . To make such
talk rigorous, we should instead talk about filters which are generic over collapses of
countable elementary substructures of rank initial segments of V , or have the letter
V not denote the true universe V of all sets but rather e.g. a countable transitive
model of ZFC.
Definition 8.6 Let A , and let be an uncountable cardinal. We say that A is
-universally Baire iff A = p[T ], where T is a tree on for some ordinal ,
and there is some tree U on for some ordinal such that p[U ] p[T ] =
and for all posets P H ,
1P  p[U ] p[T ] = .
A is called universally Baire iff A is -universally Baire for all uncountable cardinals .
Notice that if p[U ] p[T ] = in V , then p[U ] p[T ] = in V [G] for all
generic extensions1 V [G] of V : if p[U ] p[T ] = in V [G], then, setting U T =
{(s, f, g): (s, f ) U (s, g) T }, p[U T ] = in V [G] and hence p[U T ] =
in V by absoluteness of wellfoundedness, cf. Lemma 5.6, and thus p[U ] p[T ] =
in V . Therefore, if T, U witness that A is -universally Baire, then p[T ] and p[U ]
project to complements of each other in all V [G], where G is P-generic over V for
some P H . Also, if T, U as well as T , U both witness that A is -universally
1

cf. the remark before Definition 8.6.

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.

For s < , ((Us ) M ) = (Us ) N .


( M\A) =( N )\A .
( n< An ) = n< (An ) .

8.1 Lebesgue Measurability and the Property of Baire

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 ]}

is dense below [Usn ]. There is hence some k < such that


 [Usn k ] G, and we
may set sn+1 = sn k for this k. Let us also set x = x G = n< sn .

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

/ B for all B B M which are null sets


Now suppose x to be such that x
in M. Let


G = [B]: B B M x B .
It suffices to verify that G is a filter which is generic over M.
Well, G is easily seen to be a filter. Now let A M be a maximal antichain.
As M |= B/null has the c.c.c. by Lemma 8.5, A is countable
in M, say A =

{[Bn ]: n < }, where (Bn : n < ) M. Also ( M)\ n< B
n must be a null
set in M, asA is a maximal
But then x
/ (( M)\ n< Bn ) , and
 antichain.

hence x ( n< Bn ) = n< Bn , i.e., x Bn for some n < , so that [Bn ] G


for some n < , as desired.
The proof of (2) is entirely analogous.

Lemma 8.10 Let M be a transitive model of ZFC.
(1) If (20 ) M is countable, then A = {x : x is not random over M} is a null set.

8.1 Lebesgue Measurability and the Property of Baire

153

(2) If (20 ) M is countable, then B = {x : x is not Cohen over M} is a meager


set.
Proof There are (provably in ZFC) 20 Borel sets. Therefore, as (20 ) M is countable, there are only countably many B M such that B is a Borel set from the
point of view of M.
(1) By Lemma 8.9 (1), x A iff
x


{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}.

As (20 ) M is countable, B is thus a countable union of meager sets, and hence B is


meager.

Definition 8.11 Let M be a transitive model of ZFC. We say that x V is
generic over M if there is a poset P M and there is a P-generic filter G over M
such that M[G] is the -least transitive model N of ZFC with M {x} N . In this
situation, we also write M[x] instead of M[G].
Now let A . We say that A is Solovay over M iff there is a formula and
there are parameters a1 , . . . , ak M such that for all x , if x is generic over
M, then
x A M[x] |= (x, a1 , . . . , ak ).
Lemma 8.12 Let M be a transitive model of ZFC, and let A be Solovay
over M.
(1) There is a Borel set B such that for every x which is random over
M, x A x B.
(2) There is a Borel set C such that for every x which is Cohen over
M, x A x C.
Proof (1) Let be a formula, and let a1 , . . . , ak M be such that for all x
M
which are generic over M, x A M[x] |= (x, a1 , . . . , ak ). Let M (B / null)
be a (canonical) name for the random real which is added by forcing with (B/ null) M
over M, i.e., if G is (B/ null) M -generic over M, then G = x G . Let E (B/ null) M
be a maximal antichain of [D] (B/ null) M such that
(B / null) 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


Lemmas 8.10 and 8.12 now immediately give the following


Corollary 8.13 Let M be a transitive model of ZFC, such that (20 ) M is countable.
Let A be Solovay over M. Then A is Lebesgue measurable and has the
Baire property.
Theorem 8.14 (Feng, Magidor, Woodin) Let A be (20 )+ -universally Baire.
Then A is Lebesgue measurable and has the Baire property.
Proof Let T, U witness that A is (20 )+ -universally Baire. Let be a regular cardinal such that > (20 )+ and T, U H . Let
: M H
be an elementary embedding where M is countable and transitive and T, U
ran( ), say (T ) = T and (U ) = U . In order to prove the theorem, by the
proof of Corollary 8.13 it suffices to verify that for every x which is either
random over M or else Cohen over M,
x A M[x] |= x p[T ].
Well, let x be either random over M or else Cohen over M. As
M |= T , U witness that p[T ] is (20 )+ -universally Baire,
we must have that either M[x] |= x p[T ] or else M[x] |= x p[U ].
Suppose that x p[T ], say (x  n, f  n) T for all n < . Then (x  n, ( f 
n)) T for every n < , and hence x p[T ]. (We dont need f M here, just

8.1 Lebesgue Measurability and the Property of Baire

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

0 , if x(n) is even, and


1 , if x(n) is odd.

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 ).

In the end, for each x 2, { px n : n < } generates a C-generic filter gx V over


M such that by (8.1), gx
/ M, gx [T ], and gx A. By construction, the set
g

{t x : x 2} is thus a perfect subset of P A.


If z
/ A, then a symmetric argument yields that P\ A contains a perfect
subset.

Theorem 8.18 Let A be 1 -universally Baire. Then A has the Bernstein
property.
Proof We amalgamate the arguments for Theorem 8.14 and Lemma 8.17. Let P
witness that A is 1 -universally Baire, and let S be a perfect
tree on such that P = [S], cf. Lemma 7.1. Let be a regular cardinal such that
> (20 )+ and T, U H . Let
be perfect. Let T, U

: 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)

As in the proof of Theorem 8.14,


p[T ] M[g] = A M[g] and p[U ] M[g] = ( \A) M[g].
We may then finish off the argument exactly as in the proof of Lemma 8.17, with

x p[T ] playing the role of (x, a1 , . . . , ak ).
Corollary 8.19 Every analytic as well as every coanalytic set of reals has the Bernstein property.
We now want to start producing a model of ZF + every set of reals is Lebesgue
measurable and has the Baire property. Modulo of what has been done so far, the
remaining issue will be an analysis of forcing.

8.2 Solovays Theorem


We shall now analyze the situation after Levy collapsing an inaccessible cardinal,
cf. Definitions 4.41 and 6.43. We shall also need the concept of OD , cf. Definition
5.42. We may construe Theorem 8.23 as an ultimate extension of Corollaries 8.15
and 7.16.

8.2 Solovays Theorem

157

Theorem 8.20 Let be an inaccessible cardinal, and let G be Col(, <)-generic


over V . Then in V [G], 1 is inaccessible to the reals.
Proof Recall that Col(, <) has the -c.c., cf. Lemma 6.44. Therefore, V [G] and
V have the same cardinals and = 1V [G] , cf. Lemma 6.32.
If f OR V [G], then there is some V Col(, <) such that f = G and
= },

= {((n, ), p): n < OR p An p  (n)


where each An , n < , is a maximal antichain in V of p Col(, <) with
p  (n)
= .
As Col(, <) has the -c.c., (An : n < ) V = H , say
(An : n < ) H , where < . Let us write
G  = { p  : p G} and
G  [, ) = { p  [, ): p G}.
By the Product Lemma 6.65, G  is Col(, <)-generic over V and G  [, ) is
Col(, [, ))-generic over V [G  ]. We now have f (n) = iff p An G p 
= ,
so that in fact f V [G  ].
(n)
= iff p An G  p  (n)
We thus have shown the following.
Claim 8.21 For each f V [G] there is some < such that f V [G  ].
It is not hard to verify that Claim 8.21 implies that 1 is inaccessible to the reals, cf.
Problem 8.5.

Theorem 7.28 and Lemma 7.27 now immediately yields:
Corollary 8.22 (Specker) Let be an inaccessible cardinal, and let G be Col
(, <)-generic over V . Then in V [G], every uncountable 1 set of reals has
2

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 ]

Claim 8.24 Let < . Let P H


be a partial order, and let s V [G] be Pgeneric over V [G  ]. There is then some H V [G] which is Col(, <)-generic
over V [G  ][s] such that
V [G] = V [G  ][s][H ].

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.

8.2 Solovays Theorem

159

So let us fix P H , a partial order, and let s V [G] be P-generic over V . We


aim to construct some H V [G] which is Col(, <)-generic over V [s] such that
V [G] = V [s][H ].
Well, we have s V [G  + 1] for some < by Claim 8.21. As Col(, < + 1)
has the same cardinality as and
1 Col(, <+1) is countable,
by Lemma 6.51 there is a dense homomorphism
i: Col(, ) Col(, < + 1).
By Lemma 6.48,
G 0 = { p Col(, ): i( p) G  ( + 1)}
is a Col(, )-generic filter over V with V [G 0 ] = V [G  + 1].
Also, Col(, )
= Col(, { + 1}), so that if
j: Col(, ) Col(, { + 1})
is an isomorphism, then
G 1 = { p Col(, ): q G j ( p) = q( + 1)}
is a Col(, )-generic filter over V [G 0 ] = V [G  + 1] with V [G 0 ][G 1 ] = V [G 
( + 2)]. Recall that s V [G 0 ] = V [G  + 1].
Claim 8.25 There is some Col(, )-generic filter H over V [s] with
V [s][H ] = V [G 0 ][G 1 ].

(8.5)

Suppose Claim 8.25 to be true. We then have that


V [G] = V [G  ( + 2)][G  [ + 2, )]
= V [G 0 ][G 1 ][G  [ + 2, )]
= V [s][H ][G  [ + 2, )].
Let i : Col(, ) Col(, < + 2) be a dense homomorphism, and let us set
H0 = { p Col(, < + 2): q H i (q) p}.

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],


as desired. We have shown Claim 8.24, modulo Claim 8.25.


It thus remains to verify Claim 8.25. We aim to produce some
Col(, )-generic over V [s] such that (8.5) holds true.
As s V [G 0 ], we may pick some V Col(,) such that
s = G0 .

which is

(8.6)

Let us recursively define inside V [s] a sequence (Q : OR) of subsets of


Col(, ) as follows. Set p Q0 iff for all r P,

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

If p Q, p q, and q Col(, ), then q Q.


For the record, notice that Q was defined inside V [s], and the parameters we need
for this are , P, , and s. Let us write (v0 , v1 , v2 , v3 , v4 ) for the defining formula,
i.e.,
V [s] |= Q (Q = Q (Q , , P, , s)).

(8.8)

8.2 Solovays Theorem

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)

Let p D G 0 . If p p , p with p G 0 , then p D, as D is open, and hence


/ Q by (8.9). But then p G 0 \Q , hence G 0 \Q = , which contradicts the
p
choice of .

Subclaim 8.26 trivially implies that Q = , so that:
Subclaim 8.27 Q is separative and has the same cardinality as inside V [s].
In V [G], there is then some G which is Col(, )Subclaim 8.28 Let p Q.
0
generic over V such that p G 0 and

s = G0 .
Let D Col(, ), D V , be open dense in Col(, ). By
Proof Let p Q.
In V [G], there are only

p Q = Q+1 there is some p p with p D Q = Q.


we
countably many dense subsets of Col(, ) which are in V , so that given p Q

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.

If s = G 0 , then s is P-generic over V , and if


D = {( p, q) Col(, ) Col(, ) : r s r PV ( p, q) }
and inside V [s ], (Q , , P, , s ) holds true for exactly one Q , then D is dense
in Q and

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)

By (8.10), V [G 0 G 1 ] |= (G 0 , G 1 , , P, ). By (8.11), the s which


(G 0 , G 1 , , P, ) describes in V [G 0 G 1 ] is equal to s, which then also gives
that the D which (G 0 , G 1 , , P, ) describes in V [G 0 G 1 ] must be equal
to D and that the Q which (Q , , P, , s) describes in V [G 0 G 1 ] as part of
(G 0 , G 1 , P, ) must be equal to Q. Therefore, V [G 0 G 1 ] |= (G 0 , G 1 , ,
P, ) yields that D (G 0 G 1 ) = .

However, ( p , q ) (G 0 G 1 ) D. Contradiction!
Now by Subclaim 8.27, inside V [s], there is thus a dense homomorphism
k : Col(, ) Q.
By Subclaim 8.29, if we set
H = { p Col(, ) : k( p) G 0 G 1 },
then H is Col(, )-generic over V [s] and V [s][H ] = V [s][G 0 G 1 ] =

V [G 0 ][G 1 ]. Therefore, H is as desired.
This proof has the following corollary.
Theorem 8.30 (Solovay) Let be an inaccessible cardinal, and let G be Col
(, <)-generic over V . Set
N = HOD(V[G]
V [G]) .
Then in N , ZF + DC holds and every set of reals is Lebesgue measurable and has
the Baire property and every uncountable set of reals has a perfect subset.
Proof In the light of Theorem 8.23, we are left with having to prove that DC holds
true in N . If f N V [G], then f N V [G  ] for some < by the proof
of Theorem 8.20. It is easy to see that this implies f N .


8.2 Solovays Theorem

163

Definition 8.31 Let A [] be uncountable. We say that A is Ramsey iff there is


some x [] such that [x] A or [x] A = .
In the presence of (AC), it is not hard to construct an A [] with Card(A) =
20 which is not Ramsey, cf. Problem 8.11.
With the help of Mathias forcing, cf. p. 181, the arguments developed in this
Chapter may be used to show that every uncountable A [] is Ramsey in the
model of Theorem 8.30. Cf. Problem 12.13. Cf. also [13].

8.3 Problems

< such that A =


8.1. Let

 A be open. Show that there is some X
sX Us and Us Us = for all s  = s . Show also that
sX (Us ) is
independent of this representation of A, so that (A) is well-defined according
to Definition 8.1.

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

of such that for each i, xi is uniformly definable from (x j : j < i) and A i


inside L[A i]. Then use Problem 6.14.]
8.8. (R. Jensen) Show that if V = L[B], where B 1 , then there is an
distributive P such that if G is P-generic over V , then in V [G] there is
a reshaped A 1 . [Hint. Let p P iff p : 2, where < 1 and for
all , L[B , p  ] |= is countable, ordered by endextension.]
8.9. Let E OR be universally Baire in the codes in the following sense. There
are trees T , U witnessing that p[T ] is a universally Baire set of reals, and for
all ordinals , E iff
Col(, )

V

x R (x p[T ] = ||x||).

Show that E satisfies full condensation in the sense of Definition 5.30.


8.10. Show that if is a strong cardinal and A is -universally Baire, then
A is universally Baire.
8.11. Show in ZFC that there is some A [] with Card(A) = 20 which is not
Ramsey.

Chapter 9

The Raisonnier Filter

By Corollary 8.22, it is impossible to construct just from a model of ZFC a model


in which the statements from the conclusions of Solovays Theorems 8.23 and 8.30
hold true. We now aim to consider Lebesgue measurability and prove a theorem of
Saharon Shelah, Theorem 9.1.

9.1 Rapid Filters on


Theorem 9.1 (Shelah) Suppose that every 1 -set of reals is Lebesgue measurable.
Then 1V is inaccessible to the reals.

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)

We first want to construct, assuming that 1V is not inaccessible to the reals, an


interesting rapid filter.
Notice that by identifying any a P() with its characteristic function, we
may identify P() with the Cantor space 2. We construe 2
= P() as being
equipped with the natural topology, cf. 123. We shall verify that no rapid filter is
Lebesgue measurable, cf. Theorem 9.16.
Theorem 9.3 Assume that 1V is not inaccessible to the reals, but every 1 -set of
2

reals is Lebesgue measurable. There is then a rapid filter F on such that F is 1 .


3

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_9,


Springer International Publishing Switzerland 2014

165

166

9 The Raisonnier Filter

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<

FX is called the Raisonnier filter.


Claim 9.5 FX is a non-trivial filter extending the Frchet filter.
Proof Trivially, if a FX and b a, where b , then b FX . Also, FX = ,
because FX . Let us suppose that a FX and b FX , witnessed by (X na : n < )
and (X nb : n < ) respectively. Let : be bijective. Set, for n, m <
b . We have that
, X (n,m) = X na X m


X=X

X na = X

n<

 

b
(X na X m
)=X

n< m<

X p,

p<

so that (X p : p < ) is a covering of X . Let q H (X (n,m) ), say q = h(x, y), where


b , x = y. Then q = h(x, y) a, as H (X a ) a, and
x, y X (n,m) = X na X m
n
b
q h(x, y) b, as H (X m ) b. We have shown that


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 .

9.1 Rapid Filters on

167

Proof
Let us first verify that a FX iff there is a covering (Yn : n < ) of X with


n< H (Yn ) a, where Yn is a closed subset of 2 for each n < . Namely,


let a FX , as being witnessed by (X n : n < ). For n < , let Yn be the closure
of X n . Trivially, X n Yn , and therefore H (X n ) H (Yn ). We claim that in fact
H (X n ) = H (Yn ). Well, if n = h(x, y) H (Yn ), we may pick x , y X n with
x  n = x  n and y  n =y  n. But thenn = h(x, y) = h(x , y ), so that
n H (X n ). We now have that n< H (Yn ) = n< H (X n ) a.
This now gives the following characterization of FX . a FX iff (Tn : n < )
such that Tn is a tree on < 2 for each n < ,
x (x X nm x  m Tn )
x yn(x = y m x  m Tn m y  m Tn
m a\{0}(x  m 1 = y  m 1 x  m = y  m)).
Because X = 2 L[a] is 1 (a), cf. Lemma 7.19, this easily gives that FX is 1 (a)
by Lemma 7.17.

We have verified that FX is a nontrivial filter which is

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

hypothesis (Recall that we assume all 1 sets of reals to be Lebesgue measurable).


2

Set

A = {(x, y) ( )2 L[a, f ]: x < L[a, f ] y}.

For each y L[a, f ],

{x: (x, y) A}

is countable, and hence null. By Fubinis Theorem, we therefore first get A to be


null and then also
{x: {y:(x, y) A} is not null}
to be null. If L[a, f ] is not null, there is then some x0 such that
{y: (x0 , y) A} is null;
but then

L[a, f ] = {x L[a, f ]: x L[a, f ] x0 }


{y L[a, f ]: (x0 , y) A}

is the union of two null sets and hence null.

168

9 The Raisonnier Filter

We have shown that L[a, f ] is a null set.

Claim 9.8 Let (n k : k < ) be a sequence of positive integers. There is a family


(G k : k < ) of open subsets of 2 such that
(G k ) =

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

G k = {x 2: n(rk1 n < rk x(n) = 1)}.


It is easy to see that (G k ) = 2n k and if N is finite, then





2n k =
(G k ),
{G k : k N } = 2 kN n k =
kN

kN

so that (G k : k < ) is as desired.

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

Proof Let C0 2 be closed such that (C0 ) 23 and Y C0 = . Let C0 = [T0 ],


where T0 < 2 is a tree. Let us recursively define trees Tk < 2, k > 0, as
follows.
Tk = {t Tk1 : s k 2 x [Tk1 ] Us
1
([Tk1 ] Us ) k t = x  lh(t)}.
8

Set Ck = [Tk ] for k < . Also set T = k< Tk and C = [T ] = k< Ck . We


claim that C is as desired.
As T T0 , Y C = is trivial.
In the step from Ck1 to Ck we consider 2k many Us , s k 2, and throw out those
sets Ck1 Us such that (Ck1 Us ) < 81k . Therefore, (Ck ) (Ck1 )2k 81k =

(Ck1 ) 41k . This means that (C) (C0 )


k=1 ((C k1 ) (C k ))
1
2
1
1
(C0 ) k=1 4k 3 3 = 3 > 0.

9.1 Rapid Filters on

169

Now let s < 2 be such that Us C = . Then, setting k = lh(s), (Ck1


Us ) 81k and Ck Us = Ck1 Us , so that (Ck Us ) 81k . In the step from
Cl1 Us to Cl Us , l > k, we consider 2lk many Ut , t l 2, t s, and throw
out those sets Cl1 Ut such that (Cl1 Ut ) < 81k ; therefore, (C Us )

(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)

Proof Using Claim 9.8, we may pick a sequence


(G s,m,n : s < 2 m, n < )
1
for all s, m, n, and such that the
of open subsets of 2 such that (G s,m,n ) = 2m+n
<
sequence is independent in that if N 2 is finite,




{G s,m,n : s, m, n) N } =

(G s,m,n ).

(s,m,n)N

Fix f : monotone. Let


G=

  

(x, y) ( 2)2 : y G x  f (m),m,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

each z G , {x(y): (y, z) A} is at most countable, and for any x X, {y


G : x(y) = x} G x is null. Therefore, {y G : (y, z) A} is null for every
z G , so that A is null by Fubinis Theorem. Hence

170

9 The Raisonnier Filter

{y G : {z G : (y, z) A} is not null }


is a null set, again by Fubinis Theorem. If G were not null, then we could pick
some y0 G such that {z G : (y0 , z) A} is null. But then
G = {z G : (z, y0 ) A}
{z G : x(z) = x(y0 )}
{z G : (y0 , z) A}
would be null after all.
We have shown that G is a null set, so that by Claim 9.9, we may pick some
closed set C 2 such that G C = , (C) > 0, and in fact for all s < 2, if
1
.
Us C = , then (Us C) 23lh(s)+1
For x X and n < , let
 
G x  f (m),m,n .
Onx =
n n mn

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

9.1 Rapid Filters on

171

we have that b FX . We aim to verify that (9.2) holds true.


So let us fix n < . We have that
b f (n) = {h(x, y) < f (n): x, y X, x = y, n(x) = n(y),
s(x) = s(y), and
x  f (e(n(x), s(x))) = y  f (e(n(x), s(x)))}.
Obviously, if x, y X, x = y, witness that h(x, y) b f (n), then, setting
m = n(x) = n(y) and s = s(x) = s(y),
f (e(m, s)) < h(x, y) < f (n),
so that
e(m, s) < n

(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 ,

in order to show that (9.2) holds true it suffices to

bm,s (3n + 1)2 24n = ((3n + 1) 22n )2 .

(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)

2: x X (t x n(x) = m s(x) = s)}))2 bm,s .

(9.6)

But we have that


{t f (n) 2: x X (t x n(x) = m s(x) = s)}
{t f (n) 2: C Us = C Us G t,n,m = }.
This is because if x X, t x, n(x) = m, and s(x) = s, then by (9.3), C Us =
and

 
= C Us Omx = C Us
G x| f ( p), p,n .
n m pn

172

9 The Raisonnier Filter

But e(m, s) < n, so that m < n by the property of e, and thus

G t,n,m

 

G x| f ( p), p,n .

n m pn

By (9.6), in order to verify (9.5) it thus suffices to show that if C Us = , then


Card({t

f (n)

2: C Us G t,n,m = }) (3n + 1) 22n .

(9.7)

Suppose that C Us = , where s < 2. Let us write q for the cardinality of

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 = .

and because the G t,n,m s are independent in the sense of

(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

We always have log22n (2) 1, which gives us 2


 2n 
log2 22n2 1 22n , and hence

22n
22n 1

22n

, thus 1


 2n 1
2
q (3n + 1) log2
(3n + 1) 22n ,
2n
2 1


as we had wished.

Using Claim 9.10 it is now easy to prove that FX is rapid. Let g: be


monotone, and let f : be defined by
f (n) = g((n + 1) (3n + 4)2 24n+4 ).

9.1 Rapid Filters on

173

By Claim 9.10, there is some b FX such that for all n < ,


b f (n) n (3n + 1)2 24n .

Let n < , and let n n be largest such that n (3n + 1)2 24n n. Then

b g(n) b g((n + 1) (3n + 4)2 24n +4 )

= b f (n ) n (3n + 1)2 24n n,


as desired.
This finishes the proof of Theorem 9.3.

9.2 Mokobodzkis Theorem


Lemma 9.11 (Sierpinski) Let F 2 be a non-trivial filter which extends the
Frchet filter and is Lebesgue measurable. Then F is null.
Proof For s n 2, where n < , we may define a homomorphism s : 2 2 by

s (x)(k) =

1 x(k) if k < n s(k) = 1


x(k)
otherwise .

Because F is assumed to extend the Frchet filter, we have {s (x): x F} = F for


every s n 2. The 01Law of HewittSavage then implies that either (F) = 0
or (F) = 1.
Suppose that (F) = 1. Let us define a homeomorphism : 2 2 by
(x)(k) = 1 x(k)
for k < . It is easy to see that respects , i.e., {(x) : x X } is Lebesgue
measurable and ({(x) : x X }) = (X ) for all Lebesgue measurable X 2.
(F) = 1 yields that ({(x) : x F}) = 1. We may then pick some x0
F {x : (x) F}. As F is a filter, the characteristic function of the intersection
of the two sets for which x0 and (x0 ) are the respective characteristic functions is
then in F again, i.e.,
= {k < : x0 (k) = 1 (x0 )(k) = 1} F.
This contradicts the fact that F is assumed to be non-trivial.
We have shown that (F) = 0.

174

9 The Raisonnier Filter

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 }.

for every n 0 < . We may choose (n : n < ) in


Hence ((I, J))
n=n 0 n
such a way that n=0 n < , so that we get that (I, J) is a null set. Hence N is a
null set. We show that every null set can be covered by two small sets:
Lemma 9.13 If A 2 is null, then there are small sets N0 , N1 2 with A
N0 N1 .
Proof Let A 2 be null. For each n < , we may pick an open set On 2 such
that A On and (On ) < 21n . Let
On =

Usmn ,

m<

where smn < 2 and Usmn Us n = for smn = smn . Notice that by (On ) <
m

min{lh(smn ) : m < } > n.

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)

9.2 Mokobodzkis Theorem

175

Let (n : n < ) be a sequence of positive reals. In the light of


what we are supposed
to prove, we may assume without loss of generality that
n=0 n < . Let us
recursively construct (n k : k < ) and (m k : k < ) as follows.
Let n 0 = 0, m 0 = 0,






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 < ,

Ik = [n k , n k+1 ) and Ik = [m k , m k+1 ),

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

9 The Raisonnier Filter

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

To verify that A (I, J) (I , J ), let x A. By (9.9), there are infinitely many


n < with
x  n Fn and n

[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

If x  n Fn and n [m k+1 , n k+1 ], then x  Ik Jk ; hence if (9.10) holds true,


then x (I, J). If x  n Fn and n [n k , m k+1 ], then x  Ik Jk ; hence if (9.11)

holds true, then x (I , J ).
Lemma 9.14 Let F be a non-trivial filter on which extends the Frchet filter
and is Lebesgue measurable. Then F is small.
Proof By Lemma 9.11 we know that F is null. Let us fix a sequence (k : k < ) of
positive reals. We may assume without loss of generality that k < 1 for all k < .

Let k = min{ 2k , 2m k+1 n k+1 } < k and k =

(k )2
8

< k . We may write

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 .

9.2 Mokobodzkis Theorem

177

For all k < ,


({x : x  [n k , m k+1 ) Hk }) 2n k+1 m k+1 ({x : x  Ik Jk })
1
({x : x  Ik Jk })

1
k = k .
<
k
4

(9.14)

We also define, for k < ,



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 }.

With the help of (9.14) and (9.15), we have that




{x : x  [m k , m k+1 ) Jk Jk ) < k + 2 k < k .


4
Hence if F N1 N2 , then we found a covering of F as in (9.13). Let us thus
assume that F N1 N2 .
/ N1 N2 . Then x0 N0 by (9.12), so that for
Let x0 F be such that x0
infinitely many k < ,
x0  Ik Jk .
= [n , m

For k < , let I2k


k
k+1 ), I2k+1 = [m k+1 , n k+1 ), J2k = , and also J2k+1 =
, unless x0  Ik Jk in which case
0
J2k+1
= L 2k+1 L 2k+1 , where
L 2k+1 = {s [m k+1 ,n k+1 ) 2 : x0  [n k , m k+1 ) s Jk } and
}.
L 2k+1 = {s [m k+1 ,n k+1 ) 2 : s x0  [n k+1 , m k+2 ) Jk+1

178

9 The Raisonnier Filter

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(n) if k (x0  [n k , n k+1 ) Jk and n [m k+1 , n k+1 ))


x0 (n) otherwise.

As F is a filter and x0 (n) = y(n) = 1 implies z(n) = 1, we have that z F. By


(9.12), we must have z N0 or z N1 .
/ Jk , then z  Ik = x0  Ik
/ Jk .
Say z N0 . Consider Ik = [n k , n k+1 ]. If x0  Ik
But as z N0 , we must have z  Ik Jk for infinitely many k. We must then have
x0  Ik Jk and then
z  Ik = x0  [n k , m k+1 ) y  [m k+1 , n k+1 ) Jk
0
for any such k, which implies that y  [m k+1 , n k+1 ) L 2k+1 J2k+1
. However,

y
/ (I , J ), so there can be only finitely many such k. Contradiction!

= [m k+1 , m k+2 ). If x0  [n k , n k+1 )


/ Jk , then
Say z N1 . Consider Ik+1

/ N1 can only be in Jk+1


z  [m k+1 , m k+2 ) = x0  [m k+1 , m k+2 ), which by x0

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,

= y  [m k+1 , n k+1 ) x0  [n k+1 , m k+2 ) Jk+1


,
z  Ik+1
0
which implies that y  [m k+1 , n k+1 ) L 2k+1 I2k+1
. But again y
/ (I , J ), so
there can be only finitely many such k. Contradiction!

We have shown that F (I , J ).
In the light of Theorem 9.3, Shelahs Theorem 9.1 is now an immediate consequence of the following.

Theorem 9.16 (Mokobodzki) No rapid filter F 2 is Lebesgue measurable.


Proof Let n =

1
2n+1

for n < . By Lemma 9.14 we may write


F (I, J),

where for every n < ,

({ 2 : x  In Jn }) < n .

9.2 Mokobodzkis Theorem

179

For n < , let


Jn = {s Jn : t

In

2(k In s(k) t (k) t Jn )}.

We claim that
F (I, J ).

(9.16)

Suppose that x F\(I, J ). As x (I, J), X = {n < : x  In Jn } is infinite.


As x
/ (I, J ), there must be an l < such that for all n X \l we may pick some
I
n
/ Jn . Define y 2 by
tn 2 such that for all k In , x(k) tn (k), but tn

y(k) =

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)

for all n < . Let us write


#(n) = min{{k In : s(k) = 1} : s Jn },
and

Jn,min = {s Jn : {k In : s(k) = 1} = #(n)}.

We must have that


#(n) n + 1.

(9.18)

This is because if s Jn is such that {k In : s(k) = 1} n, then ({x 2 : x 


In Jn }) 21n , contradicting (9.17).
Let us now define f : by
f (n) = max{{max(k) : s(k) = 1} : s Jn,min }
for n < . If F is rapid, then we may pick some b F such that
n < {k : b(k) = 1} f (n) n.

180

9 The Raisonnier Filter

By (9.16), b (I, J ). If b  In Jn , then (9.18) gives that {k max(In ) : b(k) = 1}


is contained in In and has maximum f (n). Hence there can be at most one such n.
In particular, b
/ (I, J ). Contradiction!
We have shown that F cannot be rapid.

The book [3] contains exciting material extending the topic of the current chapter.
We also refer the reader to [5].

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

9 The Raisonnier Filter

(b) Show that if G is M F -generic over V , then, setting x G =


G}, x G \X is finite for all X F.

{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

Measurable cardinals (cf. Definition 4.54) and elementary embeddings induced by


them (cf. Theorem 4.55) play a crucial role in contemporary set theory. We here
develop the theory of iterated ultrapowers, of 0 , and of short and long extenders.

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)

be the (linear) putative iteration of V of length + 1 given by U . Let


: V
= X 1002 V,
where {, U, } X , X is countable, and V is transitive. Let = 1 (),
U = 1 (U ), and = 1 ( ). We may also set, for ran( ) ( + 1),
M 1 () = 1 (M ),
and for , , ran( ) ( + 1),1
1 (), 1 () = 1 ( ).
Then, from the point of view of V ,
(M , : < + 1)
is the (linear) putative iteration of V of length + 1 given by U .2

For a proper class X , we write 1 (X ) for { 1 (X V ): X V ran( )}.
We here use the fact that the ultrapower construction may also be applied with transitive models
of a sufficiently large fragment of ZFC. We leave the straightforward details to the reader.
1
2

10.1 Iterations of V

185

We shall now recursively, for < + 1, construct embeddings


: M 1000 V
such that whenever < + 1, then
= .

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( ),

, then, using (10.2),


and if x = (x ) =
(x ) with
1

(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

( )
=

(  )
(  ) =

(  )
(  ) =




Proof by induction on . The statement is trivial for = 0, setting = id.


Now let > 0.
/ ran(),
Let us first suppose to be a successor ordinal, say = + 1. If

then we may construe as a map from to and simply set = . Let

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 }.

We may thus define by setting




( ( 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 )( )).

It is easy to verify, using the inductive hypotheses, that is as desired.


Now suppose to be a limit ordinal. If is not cofinal in , say ran() < ,

then we may construe as a map from to and simply set =


. Let
us thus assume that is cofinal in , which implies that is a limit ordinal as well.

We then define by setting

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 and () () (x)




= ( ) ()( ) () (x)


= ( ) ( ) (x) by the inductive hypothesis,




= ( ) ( ) (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

By an easy density argument, if G is P-generic over M and if (n : n < ) is the


sequence given by the first coordinates of elements of G, then (n : n < ) is a
Prikry sequence. Cf. also Problem 9.9 (b).
Prikry sequences are also generated by iterated ultrapowers, cf. also Problem
10.2 (c):
Lemma 10.9 Let be a measurable cardinal, let U be a measure on , and let
(M , : )
be an iteration of V = M0 of length + 1 given by U . Then (0n (): n < ) is a
Prikry sequence over M with respect to 0 (U ).
Proof This is an immediate consequence of the Shift Lemma 10.4. Let X 0 (),
X M . Say X ran(n 0 ), where n 0 < . Let m n > n 0 . Let : be
defined by

k
if k < n
(k) =
k + (m n) if k n.

By Lemma 10.4, cf. (10.3),  ran(n 0 ) = id, so that in particular (X ) = X .


But then

(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)

As has countable cofinality in M[G],


M[G] |= 2cf() = 20 < .

(10.5)

On the other hand, by hypothesis in M there is a surjection from P() onto ++ ,


so that with (10.4)
M[G] |= cf() ++ .
(10.5) and (10.6) yield a failure of SCH in M[G].

(10.6)


A model M which satisfies the hypothesis of Theorem 10.12 may be produced


with the help of iterated forcing starting from a model with a measurable cardinal
whose Mitchell order (cf. Problem 4.27) has rank (2 )+ .

10.2 The Story of 0 , Revisited


0# is a countable structure which transcends Gdels Constructible Universe L
in a precise way, cf. Theorem 11.56. We first need to explain what we mean by
0# exists. For later purposes (cf. Theorem 12.27) we shall in fact introduce x # for
arbitray reals x.
We begin by introducing 1 -Skolem functions for J -structures.
In what follows, we shall only consider models of L, E , but what we shall say
easily generalizes to L, A 1 ,..., A m . Let us fix an enumeration (n : n < ) of all 1 formulae of the language L, E . We shall denote by  the Gdel number of ,
i.e.,  = n iff = n . We may and shall assume that (n : n < ) is recursive,
and if is a proper subformula of , then 
< . We shall write v(n) for the
set of free variables of n .
Let M be a model of L, E . We shall express by
M |= n [a]
the fact that a: v(n) M, i.e., a assigns elements of M to the free variables of
1
n , and n holds true in M under this assignment. We shall also write |=
M for the
0
set of all (n, a) such that M |= n [a], and we shall write |= M for the set of all
1
(n, a) |=
M such that n is a 0 -formula.
Lemma 10.13 Let M = J [E] be a J -structure. Let N M be transitive. For each
m < , there is a unique f = f mN M such that dom( f ) = m and for all n < m,
if n is not a 0 formula, then f (n) = , and if n is a 0 formula, then
f (n) = {a v(n) N : (N ; , E N ) |= n [a]}.

192

10 Measurable Cardinals

Proof As uniqueness is clear, let us verify inductively that f mN M. Well, f 0N =


N
= f mN {(m, )} M.
M. Now suppose that f mN M. If m is not 0 , then f m+1
v(m)
Now let m be 0 . We have that
N M (cf. Corollary 5.18), and if
T = {a v(m) N : (N ; , E N ) |= m [a]}
N
=
then T P(v(m) N ) ( 0 ) M , and thus T M by Lemma 5.23. Therefore, f m+1

f mN {(m, T )} M.

Now let ( f, N , m) denote the following formula.


N is transitive m < f is a function with domain m n < m
( ( n =vi0 vi1 , some i 0 , i 1 f (n) = {a v(n) N : a(vi0 ) a(vi1 )})
some i 0 f (n) = {a v(n) N : a(vi0 ) E})
( n =vi0 E,
( n =0 1 , some 0 , 1
f (n) = {a v(n) N : a  v(0 ) f (0 ) a  v(1 ) f (1 )})
( n =vi0 vi1 , some i 0 , i 1 , , where is 0
f (n) = {a v(n) N : x a(vi1 )
(a {(vi0 , x)})  v() f ()})
( n =, some , where is not 0 f (n) = ) ).
It is straightforward to check that ( f, N , m) holds (in M) if and only if f = f mN .
Now Lemma 10.13 and the fact that every element of M is contained in a transitive
element of M (cf. Lemma 5.25) immediately gives the following.
Lemma 10.14 Let M = J [E] be a J -structure. Let n be 0 , and let a: v(n) M.
Then M |= n [a] holds true if and only if
M |= f N (ran(a) N ( f, N , n + 1) a f (n)),
which in turn holds true if and only if
M |= f N ((ran(a) N ( f, N , n + 1)) a f (n)).
0
M
In particular, the relation |=
M is 1 .
1
M over
Theorem 10.15 The 1 -satisfaction relation |=
M is uniformly 1
J - structures M, i.e., there is a 1 -formula such that whenever M = J [E]
1
is a J -structure, defines |=
M in that
1
(n, a) |=
M M |= (n, a).

Proof We have that M |= n [a] iff

10.2 The Story of 0 , Revisited

193

b M (vi0 , . . . , vik , j), some vi0 , . . . , vik , j [n = vi0 . . . vik j 


j is 0 a, b are functions dom(a) = v(n)
0
dom(b) = v( j) a = b  v(n) ( j, b) |=
M ].
0
M
Here, |=
M is uniformly 1 by Lemma 10.14. The rest is easy.

If M is a J -structure, then h is a 1 -Skolem function for M if


h: (

{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.


If we were to define a 2 Skolem function for M in the same manner then we


would end up with a 3 definition. Jensen solved this problem by showing that
under favourable circumstances n over M can be viewed as 1 over a reduct of
M, cf. Definition 11.7.
We also want to write h M (X ) for the closure of X under h M , more precisely:
Let M = J [E] be a J -structure, and let X M. We shall write

194

10 Measurable Cardinals

h M (X ) for h M (

({n} v(n) X )).

n<

Using Theorem 10.16, it is easy to verify that h M (X ) 1 M. There will be no


danger of confusing the two usages of h M .
Lemma 10.17 Let M = J [E] be a J -structure. There is then some (partial) surjective f : []< M which is 1M , and there is a (partial) surjection g: M
which is 1M .

hence h M () = M. But it is straightforward


Proof We have that h M () 1 M, and 
to construct a surjective g : []< n< ({n} v(n) ) which is 1M . We may
then set f = h M g .
The existence of g now follows from Problem 5.18.

We now introduce mice.
Definition 10.18 Let x . We say that the J -structure M = (J [x]; , U ) is an
x-premouse (x-pm, for short) iff the following hold true.
(a) (J [x]; ) |= ZFC (i.e., ZFC without the power set axiom) + there is a largest
cardinal, and
(b) if is the largest cardinal of (J [x]; ), then M |= U is a non-trivial normal
< -closed ultrafilter on .
Notice that in (a) of Definition 10.18, we consider the reduct of M with U being
removed. Also recall (cf. Definition 5.24) that a J -structure has to be amenable, so
that in the situation of Definition 10.18, for all z J [x], z U J [x]. J -structures
are models of 0 -comprehension and more, cf. Corollary 5.18.
Lemma 10.19 Let x . Let : L[x] L[x] be s.t. = id. Let = crit( ). Set
= +L[x] , and let X U iff X P() L[x] (X ). Then (J [x]; , U )
is an x-pm.
Proof We show that (J [x]; , U ) is amenable, using what is sometimes referred
to as the ancient Kunen argument. Let z J [x]. Pick f J [x], f : z
onto. Then y z U iff there is some < such that y = f ( ) z and
(y) = ( f ( )) = ( f )( ). But ( f ) L[x]. Hence z U can be computed
inside L[x] from f and ( f ). But of course, z U z J [x], so that in fact

z U J [x] by Theorem 5.31.
It is easy to verify that if x, , are as in Lemma 10.19 then (J [x]; ) |= ZFC
(cf. also Lemma 10.21 (h)).
We now aim to iterate premice by taking ultrapowers in much the same way as in
the proof of Theorem 4.55 and in Definition 10.1.
Definition 10.20 Let x , and let M = (J [x]; , U ) be an x-pm. We define the
(0 -)ultrapower ult 0 (M ) of M as follows. Let = the largest cardinal of J [x]. For
f, g J [x] J [x] we write f g iff { < : f ( ) = g( )} U , and we write

10.2 The Story of 0 , Revisited

195

iff { < : f ( ) g( )} U . We let [ f ] denote the -equivalence class of f .


f g

and we also write [ f ] U iff { < : f ( ) U } U .


We write [ f ][g]
iff f g,
We let
U }.
ult 0 (M ) = {{[ f ]: f J [x] J [x]}; ,
We may define a natural map UM : M ult0 (M ) by setting UM (x) = [cx ], where
cx ( ) = x for all < . UM is called the (0 -)ultrapower map.
If ult0 (M ) is well-founded, then we identify it with its transitive collapse, we
identify [ f ] with the image under the transitive collapse, and we idenitify with the
composition of with the transitive collapse.
.

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:
.

(a) is 0 elementary with respect to , x, U ,


(b) is cofinal, i.e. for all x ult0 (M ) there is some (transitive) y M with
x (y),
. . .
(c) is 1 elementary with respect to ,. x, U ,
.
(d) is fully elementary with respect to , x,
(e) J [x] J() [x],
(f) P() M = P() ult0 (M ),
(g) ult0 (M ) |= = + , and
(h) is inaccessible (in fact Mahlo) in both M and ult0 (M ).
Proof (a) We have the following version of os Theorem 4.56.
Claim 10.22 (os Theorem) Let (v1 , ..., vk ) be a 0 -formula in the language of
M , and let f 1 , . . ., f k J [x] J [x]. Then
ult 0 (M ) |= ([ f 1 ], , [ f k ])
{ < : M |= ( f 1 ( ), , f k ( ))} U.
Notice that if is 0 , then { < : M |= ( f 1 ( ), , f k ( ))} J [x] by 0 comprehension, cf. Corollary 5.18, so that this makes sense. The proof of Claim
10.22 is analoguous to the proof of Claim 4.56, cf. Problem 10.6.
(b) Let x ult0 (M ), say x = [ f ], where f M . We have that, setting y =
TC({ran( f )}), y M and { < : f (x) y} = U , so that by oss Theorem
10.22 x = [ f ] [c y ] = (y). By (a), (y) will be transitive.
(c) Suppose that ult0 (M ) |= x(x, (a1 ), . . . , (ak )), where is 0 and a1 ,
. . . , ak M . Let x0 ult 0 (M ) be a witness. By (b) there is some y M with
x0 (y), so that in fact ult 0 (M ) |= x (y)(x, (a1 ), . . . , (ak )). By (a),
M |= x y(x, a1 , . . . , ak ), i.e., M |= x(x, a1 , . . . , ak ).

196

10 Measurable Cardinals

(d) This uses that J |= ZFC , cf. Problem 10.6.


(e) This is immediate by (d).
(f) If X P() M , then X = (X ) ult 0 (M ). On the other hand, let
X = [ f ] P() ult 0 (M ). Setting X = { < : f ( )} for < ,
(X : < ) M . But then X = { < : X U } M by 0 -comprehension,
cf. Corollary 5.18.
(g) As is the largest cardinal of J [x] and J [x] ult0 (M ), we must have that
+ in ult0 (M ). Suppose that in ult0 (M ), and let f : J [x] be
/ f ( )} ult0 (M ) and hence
surjective, f ult0 (M ). Then X 0 = { < :
X 0 J [x] by (f). But then if 0 < is such that X 0 = f (0 ), 0 X 0 iff
/ f (0 ) = X 0 . Contradiction!
0
(h) By the proof of Lemma 4.52.

Corollary 10.23 Let x , let M be an x-pm, and suppose that ult0 (M ) is wellfounded. Then ult0 (M ) is an x-pm again.
Proof We will prove a more general statement later, cf. Corollary 11.15. Let M =
(J [x]; , U ), where is the largest cardinal of J , and write
= UM : M ult0 (M ) = (P; U ),
where P is transitive.
We first aim to see that P = J [x] for some , and for this in turn it suffices to
show that P |= V = L[x],
cf. p. 77 .and Lemma 5.28 But by Lemma 10.21 (d),
.
is fully elementary with respect to and x, so that this follows from J [x] |=
V = L[x]
by elementarity.
Let us show that (J [x]; U ) is amenable. Let z J [x]. By Lemma 10.21 (b)
there is some transitive y J [x] such that z (y). We have that M |= u u =
y U . If u 0 is a witness to this fact, then ult0 (M ) |= (u 0 ) = (y) U . But then
z U = z ((y) U ) = z (u 0 ) J [x].
It is now straightforward to verify that () is the largest cardinal of J [x], and
that (J [x]; , U ) |= U is a non-trivial normal < ()-closed ultrafilter on ().

The following is in the spirit of Definition 10.1.
Definition 10.24 Let x , let M be an x-pm, and let OR {}. We call
T = (Mi ; i j : i j < ) a (the) putative iteration of M of length iff the
following hold true.
M0 = M ,
Mi+1 = ult 0 (Mi ), ii+1 is the ultrapower map, whenever i + 1 < ,
the maps i j , i j < , commute,
if < is a limit ordinal then (Mi , i j : i j ) is the direct limit of
(Mi , i j : i j < ), and
(e) for all i + 1 < , Mi is transitive.

(a)
(b)
(c)
(d)

10.2 The Story of 0 , Revisited

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 ).

i : Mi 0 M for i s.t. i = j i j whenever i j . We let


0 =  M. The construction of for limit is straightforward, cf. the proof
of Theorem 10.3.
Now suppose i has been constructed, i < . Let be the largest cardinal of Mi ,
and let be the largest cardinal of Mi . Let Mi = (J ; , U ). As U is -complete,


i U = . Let i U . We may then define i+1 : Mi+1 M by setting
i+1 ([ f ]) = i ( f )( ). Notice that we have
Mi+1 |= ([ f 0 ], . . . , [ f k1 ])

{ < :
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

Definition 10.30 Let x , and let M be an x-pm. Then M is called an x-mouse


iff M is iterable.
We should remark that the current-day inner model theory studies mice which are
much more complicated than those objects which Definition 10.30 dubs x-mice.
But as such more complicated objects wont play a role in this book, Definition 10.30
fits our purposes here.
Lemma 10.31 Suppose that there is a measurable cardinal. Then for every x
there is an x-mouse.
Proof Let be a measurable cardinal, and let U be a normal measure on . Let
: V M be as in Theorem 4.55 (3). Fix x . We may derive an x-pm (J [x];
, U ) from  L[x] L[x] as in the proof of Lemma 10.19. In particular, X U
iff X P() L[x] (X ). As (X ) is equivalent to X U , so that
U U , and as U is < -closed, U is clearly -complete in the sense of Definition

10.28. Then (J ; , U ) is iterable by Lemma 10.29.
The following Lemma is shown in exacty the same manner as is Lemma 10.4.
Lemma 10.32 (Shift Lemma) Let x , and let M = (J [x]; , U ) be an
x-mouse. Let
(M , : < )
be the iteration of M = M0 of length . 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


(10.7)

for all with ran(  ) < .


Lemma 10.33 Let x , and let M = (J [x]; , U ) be an x-mouse. Let

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.

10.2 The Story of 0 , Revisited

199

Proof Suppose that (Mi , i j : i j ) is the putative iteration of M0 = M of


length + 1. We aim to see that M is well-founded.
Let (M , i j : i j ) be the iteration of M0 = M of length + 1. Let
us write Mi = (Ji [x]; , U i ), Mi = (Ji [x]; , Ui ), and let us write i for the
largest cardinal of Mi and i for the largest cardinal of Mi . We aim to construct
0 -elementary embeddings i : Mi Mi , i , such that for all i j ,
i j i = j i j .

(10.8)

We set 0 = . The construction at limit stages is straightforward in the light of


(10.8), so let us assume that i is already constructued, i < . In much the same
way as in (4.6), cf. p. 54, we have that
Mi+1 = { ii+1 ( f )( i ): f i Mi Mi }and
Mi+1 = {ii+1 ( f )(i ): f i Mi Mi }.
We then set
i+1 ( ii+1 ( f )( i )) = ii+1 (i ( f ))(i )

(10.9)

for f i Mi Mi . We have that if is 0 and f 0 , . . ., f k1 i Mi Mi ,


Mi+1 |= ( ii+1 ( f 0 )( i ), . . . , ii+1 ( f k1 )( i ))
{ < i : Mi |= ( f 0 ( ), . . . , f k1 ( ))} U i
{ < i : Mi |= (i ( f 0 )( ), . . . , i ( f k1 )( ))} Ui
Mi+1 |= (ii+1 (i ( f 0 ))(i ), . . . , ii+1 ( ( f k1 ))(i )),
so that i+1 is well-defined and 0 -elementary. By (10.9), ii+1 i = i+1 ii+1 .

As : M M , M inherits the well-foundedness from M .
We thus have an iteration (Mi , i j : i j < ) of M of length OR, and still
using the notation from the proof of Claim 10.34 we have that { i : i OR} is a class
which is club in OR, and by Lemma 1.31 (g) and (e),
Ji+1 [x] |= i = i+ and Ji [x] Ji+1 [x]
for every i OR. This implies that
L[x] |= i = i+ and Ji [x] L[x]

(10.10)

for all i OR.


By Lemma 10.29 (h), i is inaccessible in Ji+1 [x], i.e., also in Ji+1 [x] and thus
in L[x] by (10.10). On the other hand, M is countable (in every transitive model of

200

10 Measurable Cardinals

ZFC which contains it). Thus M


/ L[x]. But this trivially implies that M
/ J [x],
as desired.

Lemma 10.35 Let x , and let M = (J [x]; , U ) and N = (J [x]; , U ) be
two x-mice. Let (Mi , i j : i j < ) and (Ni , i j : i j < ) be the iterations
of M0 = M and N0 = N , respectively, of length . There is then some i OR
such that
N = Mi or M = Ni .
Proof Let us write Mi = (Ji [x]; , Ui ) and Ni = (Ji [x]; , Ui ), and let us also
write i for the largest cardinal of Ji [x] and i for the largest cardinal of Ji [x].
Let

(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)

In particular, i j is the same as the inverse of the collapsing map of h M j (i ) and i j


is the same as the inverse of the collapsing map of h N j (i ). (10.13) implies that if
i < j OR, then
/ h M j (i ) and i
/ h N j (i ),
i

(10.14)

as i is the critical point of i j and i is the critical point of i j .


By the proof of (4.6), cf. p. 54, we have that every element of Mi+1 is of the form
ii+1 ( f )(i ) where f i Ji Ji , so that
Mi+1 = h Mi+1 (ran(ii+1 ) {i }).
By (10.13)

(10.15)

10.2 The Story of 0 , Revisited

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)

Combining (10.14) and (10.16), we get that for every j,


/ h M j ()},
{i : i < j} = { < j : 0

(10.17)

and by the same reasoning


/ h N j ()}.
{i : i < j} = { < j : 0

(10.18)

Now let us pick sequences (i k : k < ) and ( jk : k ) such that for


all k < , ik < jk < ik+1 and if is a limit ordinal, then i =
supk< i k = supk< jk = j . Notice that i = i for every limit ordinal .
Write i = i = j and = i = i . We have i = +L[x] = i by (10.12).
Set
A = {i : < is a limit ordinal}.
By the proof of Lemma 10.9, if X P() Ji [x], then
X Ui A \ X is finite X Ui .
This shows that
Mi = (Ji [x]; , Ui ) = (Ji [x]; , Ui ) = Ni .

(10.19)

Let us assume that 0 0 . By (10.17) and (10.18) applied to j = i there is


then some i such that 0 = i . But then by (10.13)
ii

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

and if N is any x-mouse, then there is some + 1 such that if (Mi , i j : i j )


is the iteration of M0 = M of length + 1, then M = N .
Proof Let M be any x-mouse, and let

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 = .

10.2 The Story of 0 , Revisited

203

Let be the class of all strong limit cardinals of cofinality (2 )+ . By Problem


10.10 (d), we shall have that () = for all (this is why we opted for
= U ).
Let be any ordinal, and let : L[x] L[x] be any elementary embedding with
= crit( ). Then we write
U ( ) = {X P() L[x]: (X )}
for the L[x]-ultrafilter derived from . (So U = U ( ).)
An example for how to obtain such a situation is when

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)

i.e., k( )([ f ]U ( ) ) = ( f )() for f L[x] L[x]. Notice that


k( )  ( + 1) = id,

(10.23)

because k( )( ) = k( )(U ( ) ( )) = ( ) = for all < and k( )() =


k( )([id]U ( ) ) = (id)() = .
Claim 10.40 Let be as in (10.21). For every X P() L[x] there is some
i < and some a [ ]< such that X = h L[x] (i, a) .
Proof If X P() L[x], then there is some i < and some a [ ]<
such that (X ) = h L[x] (i, a). But X = (X ) , so that we must have that X =
h L[x] (i, a)
.

In (c) of the following Claim, ult 0 (L[x]; U ) is as defined in Problem 10.10.
Claim 10.41 Let M = (J [x]; , U ) be an x-pm, and let is the critical point of
U , where < (2Card() )+ in V . The following are equivalent.

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

10.2 The Story of 0 , Revisited

205

f i Ji [x] Ji , as i = i+L[x] is regular in L[x]. (cf. Problem 10.10 (b).) In


particular, Mi+1 is transitive.
Let us now write Mi+1 = (Ji+1 [x]; , Ui+1 ) and let i+1 denote the critical point
of Ui+1 . We aim to use (ii) = (i) of Claim 10.41 to verify that Mi+1 is certified.
Let us fix j < and a [i+1 ]< . Let a0 [i+1 ]< and a1 [ ]< be
such that a = a0 a1 . Let a0 = [ f ]Ui = Ui ( f )(i ), where f : i [i ]Card(a0 ) ,
f L[x].
Let g: i P(i ), g L[x], be defined by
g() = h L[x] ( j, f () a1 ) i
for < i . Using the os Theorem and the fact that the elements of a1 are fixed
points under Ui , it is straightforward to verify that [g]Ui = h L[x] ( j, a) i+1 .
We now get that h L[x] (i, a) i+1 Ui+1 iff { < i : g() Ui } Ui .
Applying (i) = (ii) of Claim 10.41 yields that this is equivalent to
{ < i : i h L[x] ( j, f () a1 )} Ui ,
which by X Ui iff i Ui (X ) for all X P(i ) L[x] and the fact that the
elements of a1 are fixed points under Ui is in turn equivalent to
i+1 h L[x] ( j, Ui ( f )(i ) a1 ) = h L[x] ( j, a),

(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)

If i < , then Mi is certified, so that i = crit(i ) and Ui = U (i ).


For i j we may define ij : L[x] L[x] by
ij (x) = j1 i (x).
This is well-defined and elementary: notice that if i j , then ran(i ) =
h L[x] (i ) h L[x] ( j ) = ran( j ).
is the direct limit
Claim 10.43 If is a limit ordinal, then (L[x], (i : i < ))

( : i j < )).
Moreover, if i j < , then = i j .
of ((L[x]: i < ),
ij
ij

Proof The first part trivially


follows from the fact that
 for a limit ordinal ,
ran( ) = h L[x] ( ) = i< h L[x] (i ) = i< ran(i ).

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

We first verify that


h L[x] (i+1 ) = h L[x] (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)

where f : i L[x], f L[x]. Let E be either = or , and let f : i L[x],


g: i L[x], f , g L[x]. We then get that

( f )(i )Eii+1
(g)(i ) i ii+1
({ < i : f ( )Eg( )})
ii+1
1
i i+1
i ({ < i : f ( )Eg( )})

i i ({ < i : f ( )Eg( )})


()

i ii+1 ({ < i : f ( )Eg( )})


ii+1 ( f )(i )E ii+1 (g)(i )
The equivalence marked by (*) holds true as i witnesses that Mi is certified.
Every element of L[x] is of the form ii+1 ( f )(i ) where f : i L[x], f L[x].
By (10.26), ran(i+1 ) = h L[x] (ran(i ) {i }), so that every element of L[x] is also
( f )( ), where again f : L[x], f L[x]. By the above
of the form ii+1
i
i
computation, , as given by (10.27), is thus a well-defined -isomorphism of L[x]
.

with L[x]. But this now implies that ii+1 = ii+1
We now easily get that M is transitive, as its universe is equal to 0 (J [x]).

10.2 The Story of 0 , Revisited

207

In order to show that M is certified, we use (ii) = (i) of Claim 10.41.


Let us fix j < and a [ ]< . Let a0 [ ]< and a1 [ ]< be such
that a = a0 a1 . Let i < be sufficiently big such that a0 [i ]< . The elements
of a are then fixed points under ii+1 . We then get that h L[x] ( j, a) U iff
h L[x] ( j, a) i Ui by applying i i , iff i h L[x] ( j, a), by (i) = (ii)

applied to Mi , iff h L[x] ( j, a) by applying i , as desired.
It is not hard to see now that Lemmas 10.32 and 10.35 and the proof of Theorem
10.39 yields the following.
Corollary 10.44 (J. Silver) Let x , and suppose that x # exists. Let (Mi , i j : i
j OR) denote the iteration of x # of length , and let i be the largest cardinal of
Mi , i OR. The following hold true.
(1) L[x] = h L[x] ({i : i OR}).
(2) Let be a formula, let k < , and let i 1 < . . . < i k and j1 < . . . < jk . Write
i = min({i 1 , j1 }), and let z Ji [x]. Then
L[x] |= (z, i1 , . . . , ik ) L[x] |= (z, j1 , . . . , jk ).
(3) Let e: OR OR be order-preserving. Then e : L[x] L[x] is an elementary
embedding, where e is defined by
e (h L[x] (i, (i1 , . . . , ik ))) = h L[x] (i, (e(i1 ) , . . . , e(ik ) )),

(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)

for every i OR.


Now let  = max{i k , jk }, and write =  (k + 1). It is easy to see that there
is an order preserving map : i k + 1 with  i 1 = id and (i p ) =  p for

208

10 Measurable Cardinals

p = 1, . . . , k, e.g.  i 1 = id and (i p + ) = ( p) + for p = 1, . . . , k 1


and i p + < i p+1 , or p = k and = 0. Analoguously, there is an order preserving
map : jk + 1 with  j1 = id and ( j p ) =  p for p = 1, . . . , k. With
the help of (10.29) and the Shift Lemma 10.32 we then get that
L[x] |= (z, i1 , . . . , ik ) Jik +1 [x] |= (z, i1 , . . . , ik )
J [x] |= (z, 1 , . . . , k )
J jk +1 [x] |= (z, j1 , . . . , jk )
L[x] |= (z, j1 , . . . , jk ),
as desired.
(3) This is immediate by (2).
(4) Fix i OR. Suppose that i h L[x] ({ j : j = i}), say
i = h L[x] (i , (i1 , . . . , i p , i p+1 , . . . ik )),
where i < and i 1 < < i p < i < i p+1 < < i k . Using (2), we may then
derive both
i = h L[x] (i , (i1 , . . . , i p , i p+1 +1 , . . . ik +1 ))
and

i+1 = h L[x] (i , (i1 , . . . , i p , i p+1 +1 , . . . ik +1 )),

/ 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 ,

10.2 The Story of 0 , Revisited

209

h L[x] (( + 1) {(1 ), . . . , (k )}) ( ) (n +1 ).


This implies that if , then
h L[x] (( + 1) {( ):  OR}) ( ) .

(10.30)

If in addition is a cardinal (in V ), so that { j : j < } = , then


otp(h L[x] (( + 1) {( ):  OR}) ( )) = .

(10.31)

(10.30) and (10.31) together imply that 1 ( ) = .


Let = 1 : L[x] L[x], so that (i ) = . If = j for all j OR,
then by (1),
= h L[x] (i , (, 1 , . . . , k ))

(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

case we understand 0 -comprehension to be 0 -comprehension in the language of


(M; , A1 , . . . , An ) (which implies that (M; , A1 , . . . , An ) be amenable, cf. p. 71).
In practice, the results of this section will be applied to models M which are
either acceptable J -structures (cf. Definitions 5.24 and 11.1) or to inner models (cf.
Definition 4.51).
Definition 10.45 Let M be a transitive model of a sufficiently large fragment of
ZFC. Then E = (E a : a []< ) is called a (, )-extender over M with critical
points (a : a []< ) provided the following hold true.
(1) (Ultrafilter property) For a []< we have that E a is an ultrafilter on the set
P([a ]Card(a) ) M which is < -closed with respect to sequences in M, i.e.,
every i < , then
if
 < and (X i : i < ) M is such that X i E a for
Card(a) E .
a
i< X i E a . Moreover, a is the least such that []
(2) (Coherency) For a, b []< with a b and for X P([a ]Card(a) ) M we
have that
X E a X ab E b .
(3) (Uniformity) {} = .
(4) (Normality) Let a []< and f : [a ]Card(a) a with f M. If
{u [a ]Card(a) : f (u) < max(u)} E a
then there is some < max(a) such that
a{}

{u [a ]Card(a{}) : f a,a{} (u) = u

} E a{} .

We write (E) = sup{a + 1: a []< }. The extender E is called short if


(E) = + 1; otherwise E is called long.
This definition as well as the discussion to follow makes use of the following
notational conventions. Let b = {1 < ... < n }, and let a = { j1 < ... < jm } b.
If u = {1 < ... < n } then we write u ab for { j1 < ... < jm }; we also write u bi for
i . If X P([a ]Card(a) ) then we write X ab for {u [b ]Card(b) : u ab X }. Finally,
if f has domain [a ]Card(a) then we write f a,b for that g with domain [b ]Card(b)
such that g(u) = f (u ab ) if u ab [a ]Card(a) and g(u) = otherwise. Finally, we
write pr for the function which maps {} to (i.e., pr = ).
Notice that if E is a (, )-extender over M with critical points a , a []< ,
and if N is another transitive model of a sufficiently large fragment of ZFC such that
P(a ) N = P(a ) M for all a []< , then E is also an extender over N .
Lemma 10.46 Let M and N be transitive models of a sufficiently large fragment of
ZFC, and let : M 0 N be cofinal with critical point . Let N OR. For
each a []< let a be the least M OR such that a (), and set
E a = {X P([a ]Card(a) ) M: a (X )}.

10.3 Extenders

211

Then E = (E a : a []< ) is a (, )-extender over M.


Proof We have to verify conditions (1) through (4) of Definition 10.45.
(1) Fix a []< . As a (a ), a ([a ]Card(a) ) M) = [(a )]Card(a) )
/ ([]Card(a) ) M) = [()]Card(a) )
N . If < a , then a \ () = , so that a
Card(a)
N , i.e., []

/ 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{}

a {} ({u [a ]Card(a{}) : f a,a{} (u) = u


as desired.

}),


Definition 10.47 If : M N , E, , and are as in the statement of Lemma 10.46,


then E is called the (, )-extender derived from . If = N OR, then we shall
denote this extender by E .
Theorem 10.48 Let M be a transitive model of a sufficiently large fragment of ZFC,
and let E = (E a : a []< ) be a (, )-extender over M. There are then N and
such that the following hold true.
(a)
(b)
(c)
(d)

: M 0 N is cofinal and has critical point ,


the well-founded part wfp(N ) of N is transitive and wfp(N ),
N = {( f )(a): a []< f : [a ]Card(a) M f M}, and
for a []< we have that X E a if and only if X P([a ]Card(a) ) M and
a (X ).

Moreover, N and are unique up to isomorphism.


Proof We do not construe (c) in the stament of this Theorem to presuppose that
N be well-founded; in fact, this statement makes perfect sense even if N is not
well-founded.

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)

where a []< and f : [a ]Card(a) M, f M, defines an -isomorphism from


N onto N . Notice that for a, b []< and f : [a ]Card(a) M, g: [b ]Card(a)
M, f , g M, we have that ( f )(a) (g)(b) in N if and only if, setting c = a b,
c ({u [c ]Card(c) : f a,c (u) g b,c (u)}),
which by (d) yields that
{u [c ]Card(c) : f a,c (u) g b,c (u)} E c ,
and hence by (d) once more that
c ({u [c ]Card(c) : f a,c (u) g b,c (u)}),
i.e., ( f )(a) (g)(b) in N . The same reasoning applies with = instead of
, so that we indeed get that (10.33) produces an -isomorphism from N onto N .
The existence of N and is shown by an ultrapower construction, similar to the
proof of Theorem 4.55, (1) = (3).
Let us assume that M is of the form (M; , A). Let us set
D = {(a, f ): a []< f : [a ]Card(a) M f M}.
For (a, f ), (b, g) D let us write
(a, f ) (b, g) {u [c ]Card(c) : f a,c (u) = g b,c (u)} E c , for c = a b.
We may easily use (1) and (2) of Definition 10.45 to see that is an equivalence
relation on D. If (a, f ) D then let us write [a, f ] = [a, f ] M
E for the equivalence
class {(b, g) D: (a, f ) (b, g)}, and let us set
D = {[a, f ]: (a, f ) D}.

Let us also define, for [a, f ], [b, g] D,


[a, f ] [b, g] {u [c ]Card(c) : f a,c (u) g b,c (u)} E c for c = a b and

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

well-defined. Let us set

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 |=

( f 1a1 ,c (u), ..., f kak ,c (u))}

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 ).

the <ran( f1 ) -smallest x ran( f 1 ) with


f 0 (u) =
M |= (x, f 1a1 ,c , . . . , f kak ,c (u))

if some such x exists,


otherwise.

The point is that f 0 M. But we then have that


{u [c ]Card(c) : M |= f 0 (u) f 1a1 ,c (u)( f 0 (u), f 1a1 ,c (u), . . . , f kak ,c (u))} E c ,
which inductively implies that
N |= [c, f 0 ] [a1 , f 1 ] ([c, f 0 ], [a1 , f 1 ], . . . , [ak , f k ]),
and hence that
N |= v0 [a1 , f 1 ] (v0 , [a1 , f 1 ], ..., [ak , f k ]),


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].

Claim 10.50 implies, via a straightforward induction, that


[{}, pr] = for < .

(10.34)

In particular, (b) from the statement of Theorem 10.48 holds true.


Claim 10.51 If a []< then [a, id] = a.
Proof If [b, f ] [a, id] then by the os Theorem, setting c = a b,
{u [c ]Card(c) : f b,c (u) u ac } E c .
However, as E c is an ultrafilter, there must then be some a such that
{u [c ]Card(c) : f b,c (u) = u c } E c ,
and hence by the os Theorem and (10.34)
[b, f ] = [{}, pr] = .
On the other hand, if a then it is easy to see that [a, id].
Claim 10.52 [a, f ] = ( f )(a).
Proof Notice that this statement makes sense even if [a, f ]
/ wfp(N ).
Let b = a {0}. We have that
{u [b ]Card(b) : f a,b (u) = ((c f ),b (u))(ida,b (u))} = [b ]Card(b) E b ,
by (1) of Definition 10.45, and therefore by the os Theorem and Claim 10.51,
[a, f ] = [0, c f ]([a, id]) = ( f )(a).

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 ( ).

We now prove that () > (if ()


/ wfp(N ) we mean that ())
which
will establish Claim 10.53. Well, {} = , and
{u []1 : pr (u) < } = []1 E {} ,
from which it follows, using the os Theorem, that = [{}, pr] < [0, c ] = ().
The following, together with the previous Claims, will establish (a) from the
statement of Theorem 10.48.
Claim 10.54 For all [a, f ] N there is some y M with [a, f ] (y).
Proof It is easy to see that we can just take y = ran( f ).

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

Definition 10.55 Let M, E, N , and be as in the statement of Theorem 10.48. We


shall denote N by ult 0 (M; E) and call it the (0 -) ultrapower of M by E, and we
call : M N the (0 -) ultrapower map (given by E). We shall also write EM or
E for .
We now turn towards criteria for Ult 0 (M; E) being well-founded. (This will also
be a big issue in the proof of Theorem 11.56.) The easiest such criterion is given by
when E is a derived extender.
Lemma 10.56 Let : M N , , and be as in the statement of Theorem 10.46, and
let E be the (, )-extender derived from (cf. Definition 10.47). Then ult 0 (M; E)
is well-founded, and in fact there is an embedding k: ult0 (M; E) N such that
= k E and k  = id.
Proof We define k: ult0 (M; E) N by setting k([a, f ]) = ( f )(a) for a []<
and f : [a ]Card(a) M, f M. We have that k is a well-defined 0 -elementary
and a j []< and f j : [a j ]Card(a j ) M,
embedding. To see this let be 0 ,
f j M, for j {1, . . . , k}. Set a = j{1,...,k} a j . We then get that
ult0 (M; E) |= ([a1 , f 1 ], . . . , [ak , f k ])
{u [a ]Card(a) : M |= ( f 1a1 ,a (u), . . . , f kak ,a (u))} E a

a ({u [a ]Card(a) : M |= ( f 1a1 ,a (u), . . . , f kak ,a (u))})


a {u [(a )]Card(a) : N |= (( f 1 )a1 ,a (u), . . . , ( f k )ak ,a (u))})
N |= (( f 1 )(a1 ), . . . , ( f k )(ak )).
We have that k( E (x)) = k([, cx ]) = (cx )() = c(x) () = (x) for all x M,
so that = k E . As k() = k([pr, {}]) = (pr)({}) = for every < , we
have that k  = id.

Let us consider extenders over V .
Definition 10.57 Let F be a (, )-extender over V , and suppose that ult0 (V ; F) is
well-founded. Say ult0 (V ; F)
= M, where M is transitive. The strength of F is then
the largest ordinal such that V M.
In the situation of Definition 10.57, the strength of F is always at least + 1. If
F is derived from : V M (in the sense of Definition 10.47), where = crit( )
and V M for some > + 1, i.e., witnesses that is -strong (cf. Definition
4.60), then the strength of F may be ; more precisely:
Lemma 10.58 Let : V M be an non-trivial elementary embedding, where M
is transitive. Let = crit( ) < (), and suppose that V M. Let ()
be least such that and is inaccessible in M, and let F be the short (, )extender over V derived from . Then ult 0 (V ; F) is well-founded, and the strength
of F is at least .

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)

By k  = id, i F (E) ( ) = (E) ( ). Hence (E) ( ) N . As



, this gives V = (V ) M (V ) M N by (10.35).
Lemma 10.58 says that short extenders may be used to witness that a given cardinal
is strong. On the other hand, Lemma 10.62 below will tell us that long extenders may
be used to witness the supercompactness (cf. Definition 4.62) of a given cardinal.
Definition 10.59 Let F be a short (, )-extender over V , and let . Then F is
called -closed iff for all {ai : i < } []< there are b []< and g: []Card(b)
V such that for every i < ,
i} E
{u []Card(bai ) : g bai ,b (u)(i) = u aba
bai .
i

(10.36)

The following Lemma may be construed as strengthening of Lemma 4.63


Lemma 10.60 Let F be a short (, )-extender over V , let , and suppose that
F is -closed and N = ult0 (V ; F) is transitive. Then N is -closed, i.e.,

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

Let H : []Card(b) V be such that for every u []Card(b) , H (u) is a function


with domain u b and for every i < ,
H (u)(i) = f i (g(u)(i)).
Then E (H )(b) is a function with domain , and by (10.36) and the os Theorem,
for every i < ,
E (H )(b)(i) = E ( f i )( E (g)(b)(i)) = E ( f i )(ai ) = xi .
This shows that (xi : i < ) N .

Lemma 10.61 Let : V M be an non-trivial elementary embedding, where M


is transitive. Let = crit( ) < (), and suppose that is an inaccessible
cardinal in V and V M. Let F be the short (, )-extender over V derived from
. Then ult 0 (V ; F) is well-founded and F is -closed.
Proof This is an immediate consequence of Lemma 10.58, and we may in fact
just continue the proof of Lemma 10.58, where now = and is inaccessible in
V (not only in M). We have that V N = ult 0 (V ; F). Let {ai : i < } []< .
Then (ai : i < ) N , and hence there are b []< and f : []Card(b) V such
that (ai : i < ) = E (g)(b)(i) for every i < . The os Theorem 10.49 then yields
that b and f are as in (10.36).

Lemma 10.62 Let be -supercompact, where . There is then a long extender
E witnessing that is -supercompact, i.e., ult0 (V ; E) is transitive and if E : V
N = ult0 (V ; E) is the ultrapower embedding, then E () > and N N .
Proof Let us fix an elementary embedding
: V M,
where M is an inner model, = crit( ), () > , and M M. We aim to derive
a long extender E from in such a way that the ultrapower of V by E is also closed
under -sequences.
Set = 2 , so that = . Set = ( ), and let E be the (, )-extender over
V derived from . Let
E : V ult0 (V ; E)
be the ultrapower embedding, and let k: ult0 (V ; E) M be the factor map which is
defined as in the proof of Lemma 10.56 by k([a, f ]) = ( f )(a) for a []< and
f : [a ]Card(a) V . We may identify ult 0 (V ; E) with its transitive collapse, and we
will denote it by N . We have that = k E , and k  = id.
Let e: [ ] be a bijection. Then (e): []() M is bijective.
As M M, we have that [] V M, so that [] V ran((e)). But

10.3 Extenders

219

k  = id and = k E , which implies that E  =  , so in particular


E () = () > . Also E (e) = (e). Therefore,
[] V ran( E (e)).

(10.37)

By (10.37), E = N = ult0 (V ; E), which gives that E  = 


N , and we may pick some a []< and f : [a ]Card(a) V with
E  =  = E ( f )(a).

(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)

Set c = a b, and let us define H : [c ]Card(c) V as follows. For each u


[c ]Card(c) , we let H (u) be a function with domain such that for i < ,
H (u)(i) = f i (g b,c (u)(( f a,c (u))1 (i))).

(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

holds true. Suppose that ((ai , X i ): i < )


 is such that X i E ai for all i < . Then
there is some order-preserving map : i< ai (E) such that "ai X i for
every i < . E is called countably complete iff E is 0 -complete, and E is called
continuum-complete iff E is 20 -complete.
Lemma 10.64 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. Then E is -complete if and only if for every U 0 Ult0 (M; E) of
size there is some : U 0 M such that E (x) = x whenever E (x) U .
Proof =: Let U 0 Ult0 (M; E) be of size . Write U = {[a, f ]: (a, f ) U }
for some U of size . Let ((ai , X i ); i < ) be an enumeration of all pairs (c, X )
such that there is a 0 formula and there are (a 1 , f 1 ), ..., (a k , f k ) U with
c = a 1 ... a k and
X = {u [c ]Card(c) : M |= ( f 1a

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 ])

Ult0 (M; E) |=([a , f 1 ], ..., [a , 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

We also get that E (x) = ([, cx ]) = cx () = x.


=: Let ((ai , X i ): i < ) be such that X i E ai for all i < . Pick U 0
Ult 0 (M; E) with {(ai , X i ): i < } U , Card(U ) = , and let :
 U 0 M
be such that E (x) = x whenever E (x) U . Set =  i< ai . Then

ai = (ai ) E (X i ) = X i for all i < . Clearly, ran( ) (E).
Corollary 10.65 Let M be a transitive model of a sufficiently large fragment of
ZFC, and let E be a countably complete (, )-extender over M. Then Ult0 (M; E)
is well-founded.
Lemma 10.66 Let be an infinite cardinal, and let be regular. Let : H H ,
where H is transitive and H H . Suppose that = id, and set = crit( ). Let
M be a transitive model of a sufficiently large fragment of ZFC, let be regular in
M, and suppose that HM H . Set = sup ", and let E be the (, )-extender
over M derived from  HM . Then E is -complete.

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


10.4 Iteration Trees


Iteration trees are needed for the proof of projective determinacy, cf. Theorem 13.6.
In order to prove a relevant technical tool, Theorem 10.74, we need a strengthening
of the concept of countable completeness. All the extenders in this section will be
short, though.
Definition 10.67 Let F be a short (, )-extender over V , and let U be any set. We
say that F is complete with respect to U iff there is a map such that U
dom( ),  ( U ): U is order preserving,  ( U ) = id, and for all
a [ U ]< and for every X P([]Card(a) ) U which is measured by Fa ,4 we
have that
X Fa a X.
Hence if is an infinite cardinal, then F is -complete iff whenever U has size ,
F is complete with respect to U .
We shall be interested in a strengthening of continuum-completeness, cf. Definition 10.63.
Definition 10.68 A formula (, x) is said to be 1+ iff (, x) is of the form
M(M is transitive (2

0 )

M M V M (M, , x)),

(10.42)

where is 1 . An ordinal is called a reflection point iff V 1+ V .


It is not hard to verify that every 1+ formula is 2 .
Lemma 10.69 Let be an inaccessible cardinal. Then is a reflection point.
4

i.e., X Fa or ([]Card(a) ) \ X Fa .

222

10 Measurable Cardinals

Proof Let be 1+ , and let , x V . Let us pick some


: P 2 V,

where P is transitive, {V , TC({x})} P, (2 0 ) P P, and Card(P) < . There is


some such P, as is inaccessible.
Because every 1+ -formula can be written as a 2 -formula, so that if V |=
(, x), then P |= (, x). But if M P is as in (10.42) to witness P |= (, x),


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

inside M. Let : U = U , where U is transitive. Then, using Lemma 10.69,


M
1+ M,
: U 1+ VM = V 1+ V()

and U and are both elements of M by Lemmas


10.60 and 10.61.

Let (X i : i < 20 ) be an enumeration of {P([]n ): n < } U , and let (ai : i <
20 ) be an enumeration of [ U ]< . Let be the set of all (i, j) (20 ) (20 )
such that X j Fai . Let < be such that U V = U V . Of course, U V M
and 1 (U V ) M.
Let a i = 1 (ai ) for i < 20 . Notice that (a i : i < 20 ) M.
Now M witnesses that in M, the following holds true.
M k 
k(k: U 1+ V()
1 (U V ) =  1 (U V )

i, j < 2 0 (k(a i ) (X j ) (i, j) )).

Therefore, in V we have that


k(k: U 1+ V k  1 (U V ) =  1 (U V )
i, j < 20 (k(a i ) X j (i, j) )).
Let : U 1+ V be a witness, and set = 1 . Obviously, : U 1+ V ,
and  (U V ) =  (U V ) = id. Moreover, if a [ U ]< and
5

Recall that the strength of an extender F is the largest ordinal such that V Ult(V ; F).

10.4 Iteration Trees

223

X P([]Card(a) ) U , say a = ai and X = X j , then X j Fai iff (i, j) iff



(a i ) X j iff (ai ) X j .
Definition 10.72 Let 0 < N . A system T = ((Mi , i j : i T j < N ), (E i : i +
1 < N ), T ) is called a putative iteration tree on V of length N iff the following
holds true.
(1) T is a reflexive and transitive order on N such that if i T j, then i j in the
natural order, and if i < N , then 0 T i.
(2) M0 = V , and if i + 1 < N , then Mi is a (transitive) inner model.
(3) If i T j T k < N , then i j : Mi M j is an elementary embedding, and
ik = jk i j .
(4) If i + 1 < N , then Mi |= E i is a short extender, and if = crit (E i ) and j i
M
Mi
is maximal such that j T i + 1, then V+1j = V+1
, Mi+1 = ult(M j ; E i ), and
ji+1 is the canonical ultrapower embedding.
If N < , then we say that T is well-behaved iff M N 1 is well-founded (i.e.,
transitive).
If N = and if b is cofinal, then we say that b is an infinite branch through
T iff for all i, j b, if i j, then i T j, and if i b and k T i, then k b.
Lemma 10.73 Let n < , and let
T = ((Mi , i j : i T j n), (E i : i < n))
be a putative iteration tree on V such that for all i < n,
Mi |= E i is certified.
Then T is well-behaved.
Proof For each i < n, Mi Mi , and E i is countably complete (from the point of
view of V ) by Lemmas 10.60 and 10.61. This implies that Mn is well-founded (i.e.,
transitive) by Corollary 10.65.

The following is a key result on the iterability of V (cf. also Theorem 10.3)
which will be used in the proof of Theorem 13.6.
Theorem 10.74 Let
T = ((Mi , i j : i T j < ), (E i : i < ), T )
be an iteration tree on V such that for all i < ,
Mi |= E i is certified.
Then there is some cofinal b , an infinite branch through T such that the direct
limit

224

10 Measurable Cardinals

dir lim (Mi , i j : i T j b)


is well-founded.
Proof Suppose not. For each cofinal b , which is an infinite branch through T ,6
we may pick a sequence (nb : n < ) witnessing that
dir lim (Mi , i j : i T j b)
is ill-founded, say
b
i(n)i(m) (nb ) > m

(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)

V |= T is a putative iteration tree onV of length .

Of course, M 0 = V . We now let (, i) U iff i < and : M i 1000 V is such


that 0i = , and if (, i), ( , j) U then we write ( , j) U (, i) iff i T j
and i j = .
Suppose U = (U ; U ) to be ill-founded. It is straightforward to see that each
witness ((n , i n ): n < ) to the ill-foundedness of U gives rise to an infinite branch
b through T together with an embedding
: dir lim ( M i , i j : i T j < b) 1000 V.

(10.45)

If ( M b , ( ib : i b)) is the direct limit of ( M i , i j : i T j < b), then is defined by


(x) = n (( in b )1 (x))

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.

10.4 Iteration Trees

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

segment of R, Q |= ZC plus replacement for 1000 -formulae, and 2 0 Q Q and

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 ).

Let X = {: V 1002 V ran( ) V 2 0 V V }. Let = min(X ). We


have : V 1002 V 1002 V and U (, V ) inherits the well-foundedness from
U = U (, V ). We may thus write = ||(, 0)||U (,V ) , and we may let 0 be the th
element of X . So there are (in order type) = ||(, 0)||U (,V ) many < 0 such
that (, V , V ) is a realization of M 0 = V just by the choice of 0 .
We shall now construct an enlargement sequence
(i , Q i , Ri : i < )
such that for each i < the following holds true.
(a) (i , Q i , Ri ) is a realization of M i ,
i
Q
Q
= j 
(b) if i j and i is the length of E i , then Vi (i i ) = V j (j i ) and i  VM
i
M

Vi j ,

226

10 Measurable Cardinals

(c) if U = U (i 0i , Q i ), then U is well-founded and there are (in order type)


at least ||(i , i)||U many < Ri O R such that (i , Q i , VRi ) is a realization
of M i , and
(d) if i > 0, then Ri+1 Ri .
The last condition will give the desired contradiction.
To commence, we let (0 , Q 0 , R0 ) = (, V , V0 ), where 0 and are as above.
Now suppose ( j , Q j , R j : j i) to be constructed. Let ( ) abbreviate the
following statement: There is a realization ( , Q, R) of M i+1 such that V (i ) =

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)

An inspection shows that ( ) is a 1+ -statement in the parameter T , cf. (10.44).


Because
Ri |= i ( E i ) is certified,
we may pick, working inside Ri and setting = crit(i ( E i )), some
i
U 1+ VRi (
1+ Ri
i)

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

Let j i be largest such that j <T i + 1. As witnesses that i ( E i ) is 20 -complete


with respect to U , by the proof of Lemma 10.64 we may define : M i+1 1000 Q j
by setting

ji+1 ( f )(a) j ( f )( (i (a))),

(10.47)

Card(a)
where a is a finite subset of the length of E i and writing = crit( E i ), f : []

M j , f M j . is well-defined and 1000 -elementary by the following reasoning.


Let be 1000 , and let a ,  = 1, . . ., k, be finite subsets of the length of E
i , and
Card(a ) M j , f  M j . Write c =  a .
let f  ,  = 1, . . ., k, be such that f  : []
Then

10.4 Iteration Trees

227

M i+1 |= ( ji+1 ( f 1 )(a1 ), . . . , ji+1 ( f k )(ak ))


{u: M j |= ( f a1 ,c (u), . . . , f ak ,c (u)} ( E i )c
1

i ({u: M j |= ( f 1a1 ,c (u), . . . , f kak ,c (u)}) i ( E i )i (c)


()

j ({u: M j |= ( f 1a1 ,c (u), . . . , f kak ,c (u)}) i ( E i )i (c)


{u: Q j |= ( j ( f a1 ,c )(u), . . . , j ( f ak ,c )(u)} i ( E i ) (c)
1

()

Q j |=

( j ( f 1a1 ,c )( (i (c))), . . . , j ( f kak ,c )( (i (c)))

Q j |= ( j ( f 1 )( (i (a1 ))), . . . , j ( f k )( (i (ak ))).


M

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

so that we may let be the th < R j O R such that ( j , Q j , V j ) is a realization

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

But this implies that


Rj

VRi = V

|= ( i ),

because VRi 1+ R j . This finishes the proof of Theorem 10.74.

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)

if = + 2 n for some limit ordinal and n < , and


if = + 2 n + 1 for some limit ordinal and n < .

Let < be strong up to with respect to h.


Claim 10.78 h() = f () = . In particular, C.
Proof Suppose that h() > , and let < be least such that is even and
h( ) . Pick an elementary embedding : V M, where M is an inner model,
crit( ) = , and
Vh( )+1 M and (h) Vh( )+1 = h Vh( )+1 .

(10.49)

By elementarity, (h)( ) = (h( )) () > , whereas on the other hand


(h)( ) = h( ) by (10.49). Contradiction!

As C is arbitrary, Claim 10.78 proves that { < : is strong up to } is stationary
in , so that in particular is a Mahlo cardinal. Again as C is arbitrary, in order to
finish off the proof of Lemma 10.77 it suffices to verify the following.
Claim 10.79 For all < there is a certified extender F with crit(F) = and
lh(F) such that if F : V ult(V ; F) is the ultrapower map, then F (A)V =
A V .
Proof Fix < . As is a Mahlo cardinal, we may pick some inaccessible
cardinal with max(, ) < < . As is a Woodin cardinal, there is some
elementary embedding : V M, where M is an inner model, crit( ) = ,
V M, and h V = (h) V . Let F be the (, )-extender over V derived from
. By Lemmas 10.56, 10.58, and 10.71, F is certified, (V )ult(V ;F) = (V ) M = V ,
and there is an elementary embedding k: ult(V ; F) M with crit(k) and hence
k  V = id. In particular,
F (h) V = h V .

(10.50)

Now we have that for x V , x (A) iff x F (h){ + 2n + 1: <


a limit and n < }, by (10.48) and the elementarity of F together with the
7

The fact that F is certified implies that V Ult(V ; F).

10.4 Iteration Trees

229

choice of h, iff x h{ + 2n + 1: < a limit and n < }, by (10.50), iff x A,


by the choice of h.

We have shown that F (A) V = A V .
Definition 10.80 Let M be an inner model, let < , and let {x0 , , xk1 }
(V ) M . Then we write
type M (V ; , V , {x0 , , xk1 })
for the type of {x0 , , xk1 } in (V ) M with respect to the first order language of
set theory with parameters in (V ) M , i.e., for
{(x, v0 , , vk1 ) : x (V ) M (V ) M |= (x, x0 , , xk1 )}.
The following is an immediate consequence of Lemma 10.77.
Lemma 10.81 Let be a Woodin cardinal. Then for all > and for all
{x0 , . . . , xk1 } V there is a stationary subset S of such that for all S,
is strong up to with respect to
type V (V ; , V , {x0 , , xk1 }),
in fact for all < there is a certified extender F with critical point and lh(F)
such that if : V Ult(V ; F), then
typeUlt(V ;F) (V() ; , V , {(x0 ), , (xk1 )})
= type V (V ; , V , {x0 , , xk1 }).

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 .

and we also write [ f ] U iff { < : f ( )


We write [ f ][g]
iff f g,
U } U . We let
U }.
ult(L[x]; U ) = {{[ f ]: f J [x] J [x]}; ,
We may define a natural map UL[x] : L[x] ult(L[x]; U ) by setting
UL[x] (z) = [cz ], where cz ( ) = x for all < . UL[x] is called the
(0 -)ultrapower map. If ult(L[x]; U ) is well-founded, then we identify it
with its transitive collapse, which will be equal to L[x], and we identify [ f ]
with the image under the transitive collapse, and we idenitify with the
composition of with the transitive collapse.
For OR, we may now define the putative iteration
((Wi : i ), ( i j : i j ))

(10.51)

of W0 = L[x] by U and its images in much the same way as in Definition


10.24, with L[x] and U playing the role of M , cf. also Definition 10.1. If
i < , then Wi = L[x], and if W is transitive, then also W = L[x].
10.10. Let x, M = (J [x]; , U ), and be as above. In particular, = +L[x] . Let
UM be as in Definition 10.20.
(a) Show that UM = UL[x]  J [x].
(b) Suppose that ult(L[x]; U ) is transitive. Show that ult 0 (M ) is then also
transitive, and if ult0 (M ) = (J [x]; , U ), then J [x] = UL[x] (J [x]).
Let OR, and let Wi and i j be as in (10.51). Let Mi and i j be as in Definition 10.24 for a putative iteration of length + 1, say Mi = (Ji [x]; , Ui )
for i < .
(c) Show that if i j < , then Ji [x] = 0i (J [x]) and i j = i j  Ji [x].
(d) Show that if ult(L[x]; U ) is transitive and is any limit ordinal with
cf( ) = cf(), then UL[x] is continuous at , i.e., UL[x] ( ) = sup< UL[x]
(cf. Lemma 4.52 (c)). Conclude that if is a strong limit cardinal with
cf( ) (2Card() )+ and if (2Card() )+ , then 0i ( ) = for every i < .
10.11. Let x , and suppose that x # exists. Let A P(1V ) L[x]. Show that
either A or 1V \ A contains a club of L[x]-inaccessibles. [Hint. Consider the
countable Silver indiscernibles, and exploit the arguments for Lemmas 10.9
and 10.35.]
10.12. Assume that 0# exists. Show that for every < 1L there is some premouse
(J ; , U ) L such that there is a putative iteration of (J ; , U ) of length
+ 1 of which the th model is ill-founded.
10.13. Let x , and suppose that x # exists. Show that there is some G V (!)
such that G is Col(, < 1V )-generic over L[x]. [Hint. Recursively construct

232

10 Measurable Cardinals

initial segments of G along a club of L[x]-inaccessibles from Problem 10.11,


exploiting the Product Lemma 6.65. As limit stages, use Lemma 6.44.]
A cardinal is called remarkable iff for every > there are < < such
that if G is Col(, < )-generic over V , then in V [G] there is an elementary
embedding : V V such that crit( ) = and () = . (Here, V
and V refer to the respective rank initial segments of V rather than V [G].
Compare Problems 4.29 and 10.21.)
10.14. Show that if is remarkable and if G is Col(, < )-generic over V , then in
V [G] for every > the set

{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

0# and Jensens Covering Lemma

11.1 Fine Structure Theory


Definition 11.1 Let M = J [E] be a J-structure. Then M is called acceptable iff
for all limit ordinals < and for all , if
(P() J+ [E]) \ J [E] = ,
then there is some f J+ [E] such that f : is surjective.
Acceptability is a strong local form of GCH. If M = J [E] is a J-structure and if
M, then we write
+M = sup{ + : M f M ( f : is surjective)}.

Lemma 11.2 Let M = J [E] be an acceptable J-structure. Let M, and


set = +M . Then P() M J [E]. Moreover, is in fact the least with
P() M J [E].
Proof That P() M J [E] follows immediately from Definition 11.1. Now
suppose that there were some < with P() M J [E]. As Card( )
in M, Lemma 10.17 produces some surjective f : J [E], f M. Then
A = { < :
/ f ( )} M, but A
/ J [E] as in the proof of Theorem 1.3.
Contradiction!

If M = J [E] is a J-structure and is a cardinal of M (or = ), then by
(H ) M we mean the set of all sets which are hereditarily smaller than in M (or
(H ) M = M in case = ). Recall that for all x M, TC({x}) M (cf. Corollary
5.18), so that this makes sense.

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_11,


Springer International Publishing Switzerland 2014

235

11 0# and Jensens Covering Lemma

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],

f J + [E] J [E]. But now f 1 x P() J [E] by Lemma 11.2, and

hence x = f ( f 1 x) J [E]. Contradiction!



The following definition introduces a key concept of the fine structure theory.
Definition 11.4 The 1 -projectum (or, first projectum) 1 (M) of an acceptable
J-structure M = J [E] is defined by
1 (M) = the least OR such that P() 1M M.

Lemma 11.5 Let M = J [E] be an acceptable J-structure. If 1 (M) M, then


1 (M) is a cardinal in M. In fact, 1 (M) is a 1 -cardinal in M, i.e., there is no 1M

partial map from some < 1 (M) onto 1 (M).

Proof Write = 1 (M). Let us first show that is a cardinal in M. Suppose


not, and let f M be such that f : is surjective for some < . Let

/ M. Let A = f 1 A. Then A
/ M, since
A P() 1M be such that A

otherwise A = f A M. On the other hand, A M by the definition of , since


A and A 1M . Contradiction!

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

map from onto J [E] (cf. Lemma 10.17). Hence there is a 1M

map g: J [E] which is surjective. Set

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!


11.1 Fine Structure Theory

237

The following is an immediate consequence of Lemmas 11.5 and 11.3.


Corollary 11.6 Let M = J [E] be an acceptable J-structure, and let = 1 (M).
(a) (H ) M = J [E].
(b) If A J [E] is 1M , then (J [E], A) is amenable.

Recall our enumeration (n : n < ) of all 1 formulae from p. 185. In what


follows it will often be convenient to pretend that a given n has fewer free variables
than it actually has. E.g., we may always contract free variables into one as follows:
if n n (vi1 , . . . , vi ) with all free variables shown, then we may identify, for the
purposes to follow, n with
vi1 . . . vi (u = (vi1 , . . . , vi ) n (vi1 , . . . , vi )).
If a: v(n) M assigns values to the free variable(s) vi1 , . . . , vi of n then, setting
x1 = a(vi1 ), , x = a(vi ), we shall in what follows use the more suggestive
M |= i (x1 , . . . , x ) rather than the notation M |= i [a] from p. 185. We shall also
write h M (i, x) instead of h M (i, a), where x = (x1 , . . . , x ).
Definition 11.7 Let M = (J [E], B) be an acceptable J-structure, write =
1 (M), and let p M. We define
p

A M = {(n, x) (H ) M : M |= n (x, p)}.


p

A M is called the standard code determined by p. The structure


p

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 0# and Jensens Covering Lemma

Proof =: Pick some B which witnesses that p PM . Suppose B is defined by


n , i.e., B (n, ) A. As B 1 (M) is not in M, A ( 1 (M)) can
then not be in M either.
/ M. Let f : 1 (M) 1 (M) be
=: Suppose A ( 1 (M))
defined by f (n, + i) = + 2n 3i , where n, i < and < is a limit
ordinal. Clearly f is 1M , and if 1 (M) M then f M. Let B = f A. Then

B is 1M ({ p}), A ( 1 (M)) = f 1 (B 1 (M)), and it is easy to see that


B 1 (M)
/ M.

Definition 11.10 Let M be an acceptable J-structure, and write = 1 (M). We set
R M = the set of all r [, OR M)< such that h M ( {r }) = M.
The elements of R M are called very good parameters.
Lemma 11.11 Let M be an acceptable J-structure. R M PM = .
Proof That PM = easily follows from h M (OR M) = M, cf. the proof of Lemma
10.17.
As to R M PM , let p R M , and define A OR M by
(n, ) A (n, )
/ h M (n, (, p)).
/ M by a diagonal argument.
We have that A is 1M ({ p}), and A 1 (M)
Using the map f from the proof of Lemma 11.9 it is easy to turn the set A into some
/ M.

B OR M such that B is 1M ({ p}) and B 1 (M)
It is not hard to see that there is a computable map e: such that for all
n < , for all acceptable J-structures M, for all p M, and for all m 1 , . . . , m k <
and x1 , . . . , xk M p ,
M |= n (h M (m 1 , (x1 , p)), . . . , h M (m k , (xk , p)))
M |= e(n) (((m 1 , x1 ), . . . , (m k , xk )), p)
(e(n), ((m 1 , x1 ), . . . , (m k , xk )))

(11.1)

p
AM .

If M is an acceptable J-structure and p R M , then we may express in a uniform


p
1 fashion over M p that A M codes la (11.1) the 1 -theory of some acceptable
J-structure N which is given by applying the 1 -Skolem function h N to elements of
M p . This will play a crucial role in the proof of the Upward Extension of Embeddings
Lemma 11.20.
Definition 11.12 Let be a formula in a first order language. We say that is a
Q-formula iff is (equivalent to a formula) of the form
vi v j vi (v j ),

(11.2)

where is 1 and does not contain vi . We also write Qv j instead of vi v j vi and


read (11.2) as for cofinally many v j , (v j ). A map : M N which preserves

11.1 Fine Structure Theory

239

Q-formulae is called Q-preserving, in which case we write


: M Q N .
A map : M N is called cofinal iff for all y N there is some x M such that
y (x).
Lemma 11.13 Let : U 0 U , where U and U are transitive structures.
(a) If is cofinal, then is 1 -elementary.
(b) Let be 1 -elementary. Let be a 2 -formula, and let x U . If U |= ((x)),
then U |= (x).
(c) Let is cofinal. Let be a Q-formula, and let x U . If U |= (x), then
U |= ((x)).


Proof Problem 11.3.

We formulate the following lemma just for models of L,


A , but of course it
E,
also holds for models of different types.
Lemma 11.14 There is a Q-sentence (of L,
A ) such that for every transitive
E,
)
which
is
closed under pairing, M is an
model M = (M; , E, A) (of L,
A
E,
acceptable J-structure iff M |= .
(cf. p. 77) may be written as
Proof The statement V = L[ E]
Qy y = S [E].
Here, y = S [E] is the 1 -formula from Lemma 5.25 (2). The fact that (M, A)
is amenable can be expressed by
Qy z z = A y,
as A x M iff there is some y x with A y M.
It remains to be checked that being acceptable can be written in a Q-fashion. Let
be the sentence
n < m < <
((P( ) S +m [E]) \ J [E] = ) =

(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)

11 0# and Jensens Covering Lemma

240

By acceptability, there is some n 0 < and some surjective f : 0 , f


S +n 0 [E]. But then there is some n n 0 such that for every with (11.4) there is
some surjective f : , f S +n [E]. Therefore, M |= .
We may now express that M be acceptable by saying that for cofinally many y,
y = S +m [E] (for some ,m < ) and if
(P( ) S +m [E]) \ J [E] = ,
then there is some surjective f : , f S +m [E].

Lemmas 11.13 and 11.14 now immediately give:


M be transitive structures.
Corollary 11.15 Let M,

(a) If : M 1 M and M is an acceptable J-structure, then so is M.

(b) If : M Q M (e.g., if is a 0 preserving cofinal map) and M is an acceptable


J-structure, then so is M.
We may now turn to the downward extension of embeddings lemma.
M be
Lemma 11.16 (Downward Extension of Embeddings Lemma, Part 1) Let M,
p

acceptable J-structures. Let p R M and p M. Let : M 0 M . Then there


p)
= p. Moreover, is in fact
is a unique : M 0 M such that and (
1 elementary.
Proof By Lemma 10.16, the 1 -Skolem function h N is uniformly definable over
J-structures N , i.e., there is a 1 -formula such that x = h N (n, y) iff N |=
(n, y, x) for every J-structure N . Say (v1 , v2 , v3 ) w1 . . . wk (w1 , . . . , wk ,
v1 , v2 , v3 ).
Let us first show the uniqueness of .
Suppose that has the above properties.
< . Pick
Then x = h (n, (z, p))

for
some n and z [1 ( M)]
Let x M.
M

x). Since is 0 preserving, this


z 1 , . . . , z k M such that (z 1 , . . . , z k , n, (z, p),
implies ( (z 1 ), . . . , (z k ), n, ( (z), p), (x)), so that we must have
(x) = h M (n, ( (z), p)) = h M (n, (( ), p)).
Hence, there can be at most one such .

Let us now show the existence of .


Claim 11.17 Suppose that (v1 , . . . , v ) is a 1 -formula. For 0 < i let xi =
where n i < and z i [1 (M)]< , and let xi = h M (n i , (zi , p))
h M (n i , (zi , p))
where zi = (zi ). Then
M |= (x1 , . . . , x ) iff M |= (x1 , . . . , x ).
Proof We shall use the map e from p. 230. Let n , n < . Then M |=
(x1 , . . . , x ) is equivalent to

11.1 Fine Structure Theory

241

. . . , h M (n , (z , p))),

M |= n (h M (n 1 , (z1 , p)),

(11.5)

which may be written as

M |= e(n) (z1 , . . . , z , p).

(11.6)

This also works over M, i.e., M |= (x1 , . . . , x ) is equivalent to


M |= e(n) (z1 , . . . , z , p).

(11.7)

Now (11.6) is equivalent to


p

A M (e(n), (z1 , . . . , z )),

(11.8)

and (11.7) is equivalent to


p

A M (e(n), (z1 , . . . , z )).


Since is 0 preserving, (11.8) and (11.9) are equivalent.

(11.9)


Now let us define by


 h M (n, ((z), p))
(h M (n, (z, p)))

(11.10)

< . Here,  is understood as saying that the left hand


for n and z [1 ( M)]
side is defined iff the right hand side is. Notice that is indeed well defined by (11.10);
= h M (n 2 , (z2 , p))
where n 1 , n 2 < and z 1 , z 2
this is because if h M (n 1 , (z1 , p))
< , then by Claim 11.17, h M (n 1 , ((z1 ), p)) = h M (n 2 , ((z2 ), p)). Claim
[1 ( M)]
11.17 then also yields that is 1 preserving.
To see that and ( p)
= p, pick k1 , k2 < such that
x = h N (k1 , (x, q)) and q = h N (k2 , (x, q)),

(11.11)

, z = h (k1 , (z, p)),

uniformly over all J-structures N . Then for all z [1 ( M)]


M
hence (z) = h M (k1 , ((z), p)) = (z). This gives . Also, p =
hence ( p)
= h M (k2 , (0, p)) = p.

h M (k2 , (0, p)),
If in addition the hypothesis of Lemma 11.16 we assume that p R M and
: M p n M p , then we may show that : M n+1 M (cf. Problem 11.4).
Lemma 11.18 (Downward Extension of Embeddings Lemma, Part 2) Let M be
an acceptable J-structure, and let p M. Suppose that N is a J-structure and
: N 0 M p . Then there are unique M and p such that p R M and N = M p .
Proof The uniqueness of M and p is easy to verify, arguing as in the proof of Lemma
11.16. Let us show the existence.

11 0# and Jensens Covering Lemma

242

Mp

p !

M =N

Let M = (J [E], B), M p = (J [E], A), where = 1 (M) and A = A M , and


Let
let N = (J [ E ], A).

M
= h M (ran( ) { p}) 1 M,

(11.12)

where M is transitive. By Corollary 11.15 (a), M is an acceptable J-structure, say


B).

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)

11.1 Fine Structure Theory

Let us now set

243

p = 1 ( p).

We aim to verify that M and p are as desired.


We claim that
= .

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],

This gives that 1 ( M)


,
cf. the proof of Lemma 11.11.
1M map of onto M.

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)

But for all k < and y N ,


for all z M.

(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,

for all z N . As N is a J-structure, A ({n} ( {x})) N , and thus P is


in N , too. This proves (11.15).
As an immediate consequence of (11.16) and (11.15) we get that
p
A = A M .

(11.17)

(11.12) implies that


Because there is a 1M map of onto J [ E],
p R M .


The proof is complete.

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.

11 0# and Jensens Covering Lemma

244

Lemma 11.20 (Upward Extension of Embeddings Lemma) Let M be an acceptable


J-structure, and let p R M . Suppose that N is an acceptable J-structure, and
: M p 1 N is a good embedding. Then there are unique M, p such that
N = M p and p R M . Moreover, is good, where and : M 1 M
with ( p)
= p is given by Lemma 11.16.
Notice that Lemma 11.16 in fact applies to the situation of Lemma 11.20.
Proof of Lemma 11.20. We shall make frequent use of the map e from p. 230.
Let us first show that M and p are unique. Suppose 1 : M M1 and 2 :

M M2 are two extensions of satisfying the conclusion of Lemma 11.20, and


p
p
that p1 , p2 are the corresponding parameters. Then A M11 = A M22 , call it A, and if
k {1, 2} and x Mk , then x is of the form h Mk (n, (z, pk )) for some n <
and z [N OR]< . Let : M1 M2 be the map sending h M1 (n, (z, p1 )) to
h M2 (n, (z, p2 )), where n < and z [N OR]< . Then is a well defined
surjection since
z z = h M1 (n, (z, p1 )) A(m, (n, )) z z = h M2 (n, (z, p2 ))
for an appropriate m < (namely, m = e(z z = v)). Also, respects , since
if x = h M1 (n 1 , (z1 , p1 )) and y = h M1 (n 2 , (z2 , p1 )), where n 1 , n 2 < and z1 ,
z2 [N OR]< then
x y h M1 (n 1 , (z1 , p1 )) h M1 (n 2 , (z2 , p1 ))
A(m, ((n 1 , z1 ), (n 2 , z2 )))
h M2 (n 1 , (z1 , p2 )) h M2 (n 2 , (z2 , p2 ))
(x) (y)
for an appropriate m < (namely, m = e(v1 v2 )). Therefore, is an isomorphism, so that is the identity, i.e., M1 = M2 , 1 = 2 and p1 = p2 .

M
p
M

p
N

11.1 Fine Structure Theory

245

Let us now verify the existence of M, p. Suppose that M = (J [F], W ). We first


represent M as a term model which is definable over M p , and then unfold the term
model which is defined in the corresponding fashion over N so as to obtain M and
p. Let

M = {(n, z) M p : y y = h M (n, (z, p))},


F,
and W over M as follows, where (n, x), (m, y)
and let us define relations I, E,
M.
= h M (m, (y, p))

(n, x) I(m, y) h M (n, (x, p))

(n, x) E(m,
y) h (n, (x, p))
h (m, (y, p))

(n, x) F h M (n, (x, p))


F

(n, x) W h M (n, (x, p))


W
F,
and W are IObviously, I is an equivalence relation, and the predicates E,

invariant. Let us write, for (n, x), (m, y) M ,


[n, x] = {[m, y]: (n, x) I(m, y)}
M/ I = {[n, x]: (n, x) M}

I[m, y] (n, x) E(m,


y)
[n, x] E/
I (n, x) F
[n, x] F/

[n, x] W / I (n, x) W .
We obviously have

I, F/
I, W / I)
S = (M/ I; E/
= (J [F]; , F, W ),

where sends [n, x] to h M (n, (x, p)).

and that we may thus choose


Notice that M , I , E, F, and W are all 1M ({ p}),
n 1 , n 2 , n 3 , n 4 , and n 5 < such that for all (n, x), (m, y) M p ,
p

(n, x) M (n 1 , (n, x)) A M


p
(n, x) I(m, y) (n 2 , ((n, x), (m, y))) A M
p

(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

Let us write N = (J [E ], A ), so that : M p 1 (J [E ], A ). We define M ,


I , E, F, and W as follows, where (n, x), (m, y) N .

246

11 0# and Jensens Covering Lemma

(n, x) M (n 1 , (n, x)) A

(n, x)I (m, y) (n 2 , ((n, x), (m, y))) A


(n, x)E(m, y) (n 3 , ((n, x), (m, y))) A

(n, x) F (n 4 , (n, x)) A

(n, x) W (n 5 , (n, x)) A

(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 }

[n, x] E/I [m, y] (n, x)E(m, y)


[n, x] F/I (n, x) F
[n, x] W/I (n, x) W.
Let us consider
S = (M /I ; E/I, F/I, W/I ).
In addition to (11.18), we shall need four more facts about A which are inherited
p
from A M and which will eventually enable us to show that A is a standard code. For
one thing, for all 1 -formulae and and for all (m 1 , x1 ), . . ., (m k , xk ) N ,
(e( ), ((m 1 , x1 ), . . . , (m k , xk ))) A
(e(), ((m 1 , x1 ), . . . , (m k , xk )))
/ A

(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)

11.1 Fine Structure Theory

247

n (w) is equivalent to v e(n)


(v, w)
over rudclosed structures.
Now let n be 1 but not 0 . We then have that
p
M p |= mx ((e(n), (m, x)) A M
p

m x (e(e(n)),

((m , x ), (m, x))) A M ).

(11.21)

Equation (11.21) is not 1 by itself, but it may be rephrased in a 1 fashion over


M p as follows.
p
M p |= mx ((e(n), (m, x)) A M
p

(e(e(n)),

((k(n), (m, x)), (m, x)) A M ).

(11.22)

Here, k: is a natural computable function such that


v e(n)

M |= v k(n) (v, ((m, x), p))


(v, h M (m, (x, p))).
Equation (11.22) is true as
p

((e(n), (m, x)) A M M |= n (h M (m, (x, p)))

M |= v e(n)
(v, h M (m, (x, p)))

M |= e(n)
h M (m, (x, p)))

(h M (k(n), ((m, x), p)),


p

(e(e(n)),

((k(n), (m, x)), (m, x)) A M .


The statement (11.22) will be transported upward via , and we thus have
N |=mx (e(n), (m, x)) A
(e(e(n)),

((k(n), (m, x)), (m, x)) A ).

(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)),

((m , x ), (m, x))) A .

(11.24)

In a similar fashion, we may prove that if (v1 , v2 , v3 ) is 0 and if (m, x), (m , x )


N , then

11 0# and Jensens Covering Lemma

248

(e(v1 v2 (v1 , v2 , v3 )), ((m, x), (m , x )) A


(m , x ) N (e(v1 v2 (v1 , v2 , v3 )),
((m , x ), (m , x ), (m, x))) A .

(11.25)

I is extensional, which may be formulated as saying that if


We know that E/

(n, x), (m, y) M are such that


p

(e(v = w), ((n, x), (m, y))) A M ,


then there is some (m , x ) M with
p

(e(u vw), ((m , x ), (n, x), (m, y))) A M .


Equation (11.24) will now give that E/I is also extensional. Because E is rudimentary
over N via the same definition as the one which gives E as being rudimentary over
M p , the relation E is actually well-founded by the goodness of . Therefore,

S = (M /I ; E/I, F/I, W/I )


= (J [E ]; , E , W )
for some , E , and W . We shall also write M = (J [E ], W ) instead of
(J [E ]; , E , W ).
It is now straightforward to use (11.1811.24) and prove the following by induction of the complexity of .
Claim 11.21 For all n < and (m 1 , x1 ), . . ., (m k , xk ) N ,
(e(n), ((m 1 , x1 ), . . . , (m k , xk ))) A
M |= n ( ([m 1 , x1 ] ), . . . , ([m k , xk ] )).
We may now define

by

: M M
([n, (x)] ),
h M (n, (x, p))

where (n, x) M. Notice that is well-defined, as for (n, x), (m, y) M,


= h M (m, (y, p))
(n, x) I(m, y)
h M (n, (x, p))
p

(n 2 , ((n, x), (m, y))) A M

(n 2 , ((n, (x)), (m, (y)))) A

(n, (x))I (m, (y))


[n, (x)] = [m, (y)] .

11.1 Fine Structure Theory

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

N |= m, x (e(n), ((m, x), (, (y)))) A


()

(m, x) N M |= ([m, x] , [, (y)] )


M |= v n (v, (z)).

Here, () holds true by Claim 11.21.


Let us write p = ( p).
We claim that M, ,
and p are as desired.
Let k1 , k2 be as in (11.11). Then
x = ([k1 , x]) for all x M p and p = ([k2 , 0]).
Let us first observe that N M. If x M p and ([n, y]) x = ([k1 , x]), then
([n, y]) = ([k1 , w]) for some w x. This may be written in a 1 fashion over
M p , so that if x N and ([n, y] ) ([k1 , x] ), then ([n, y] ) = ([k1 , w ]) for
some w . It follows by -induction that ([k1 , x] ) = x for all x N . Furthermore,
given any x M p ,
(x) = (
([k1 , x])) = ([k1 , (x)] ) = (x),
and hence .
= ([k2 , 0] ). We now prove
We also have that p = ( p)
= (h
M (k2 , (0, p)))
p

that A = A M .
For x M p and n < , we have that
p

rl(n, x) A M M |= n (x, p)

h M (k2 , (0, p)))

M |= n (h M (k1 , (x, p)),


p

(e(n), ((k1 , x), (k2 , 0)) A M ,


which implies that for all x N and n < ,
(n, x) A (e(n), ((k1 , x), (k2 , 0)) A M |= n (x, p).
To see that A = A M , it then suffices to show that N OR = = 1 (M). Our
computation will also yield that p R M .
The reader will gladly verify that for all (n, x) M p ,
p

= h M (h M (k1 , (n, p)),


(h M (k1 , (x, p)),
h M (k2 , (0, p))))

h M (n, (x, p))

11 0# and Jensens Covering Lemma

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)

which in turn implies that


([n, x] ) = h M ([k1 , n] , ([k1 , x] , [k2 , 0] )) = h M (n, (x, p)).
As there is a 1M map from onto N and M = { ([n, x] ): (n, x) M }, (11.26)
implies that
M = h M ( { p}), and hence also 1 (M)

(11.27)

by the proof of Lemma 11.11. On the other hand, if B 1M ({a}), where a M,


then by (11.27) there is some z []< and some B 1M ({ p}) such that for all
x N,

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,

rudimentary over M, M, respectively, by the same rudimentary definition. Define


R , R as follows
(n, x) R (m, y) (n, x), (m, y) M ([n, x]) R ([m, y])
(n, x)R (m, y) (n, x), (m, y) M ([n, x] )R ([m, y] ).
Then R is well-founded since R is, and R , R are rudimentary over M p , N ,
respectively, by the same rudimentary definition. As is good, R must then be
well-founded. Hence R must be well-founded as well.

Definition 11.22 Let M be an acceptable J-structure. For n < we recursively
n, p
define the n-th projectum n (M), the n-th standard code A M and the n-th reduct
n,
p
as follows:
M
0,
(1) 0 (M) = M OR, M0 = {}, A0,
M = , and M = M, and
n+1
[i+1 (M), i (M)) , and
(2) n+1 (M) = min{1 (M n, p ): p Mn }, M =
in

n+1, p

for p Mn+1 , A M

= A M n, pn , and M n+1, p = (M n, pn ) p(n) .


p(n)

11.1 Fine Structure Theory

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.

Definition 11.23 Let M be an acceptable J-structure. We the set


0
= {},
PM
n+1
n
PM
= { p Mn+1 : p  n PM
1 (M n, pn ) = n+1 (M) p(n) PM n, pn }, and
n+1
n
n, p n
R n+1
) = n+1 (M) p(n) R M n, pn }.
M = { p M : p  n R M 1 (M
n good parameters and the elements of R n very
As before, we call the elements of PM
M
good parameters.

Lemma 11.24 Let M be an acceptable J-structure.


(a)
(b)
(c)
(d)

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

M m,q m |= i (x, q(m))


M m,q m |= i (x, h M m, pm (e0 , (z, p(m))))
M m, pm |= j ((x, z), p(m))
p(m)

( j, (x, z)) A M m, pm ,

11 0# and Jensens Covering Lemma

252

for some J which is recursively computable from i, as M m,q m is rud M m, pm in


q(m), (M)
is rud A p(m)
in
the parameter t = h M m, pm (e1 , (z, p(m))). Therefore, A M m,qmn
M m, pm

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

1 (M n, p ) 1 (M n,q ) = n+1 (M), and hence 1 (M n, p ) = n+1 (M). (d) This


follows inductively, using (c).

The following is given just by the definition of R n+1
M . Let M be acceptable, and
.
Then
let p R n+1
M
M = h M (h M 1, p1 (. . . h M n, pn (n+1 (M) { p(n)}) . . .) { p(0)}).

(11.28)

n+1, p

basically as the iterated


We thus can, uniformly over M, define a function h M
composition of the 1 Skolem functions of the ith reducts of M, 0 i n, given
n+1, p
by p such that M is the h M -hull of n+1 (M) whenever p R n+1
M .
More precisely, let M be acceptable, and let p Mn+1 . Let us inductively define
i, p
h M , for 1 i n + 1, as follows. For k < , let g(k) = the largest m such that 2m
divides k, and let u(k) = the largest m such that 3m divides k. Let
h M (k, x) = h M (k, (x, p(0))) for x M 1, p1 , and
1, p

i+1, p

for i > 0, h M

(k, x) = h M (g(k), h M i, pi (u(k), (x, p(i)))) for x M i+1, pi+1 .


(11.29)
i, p

If X M n+1, p , then we shall write


n+1, p

hM

n+1, p

(X ) for h M

( < X ).
n+1, p

If p is clear from the context, then we may write h n+1


M rather than h M
The following is straightforward.

Lemma 11.25 Let n < , and let M be an acceptable J-structure. If p Mn+1 ,


n+1, p
then h M
is in M ({ p}), and if p R n+1
M , then
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 .

Proof It is easy to verify that M


P(M n, p ), say

n, p

M P(M n, p ). Now let A M

x A M |= x1 x2 /xk (x, y, x1 , x2 , , xk ),

11.1 Fine Structure Theory

253

where is 0 and y M. By Lemma 11.25, we may write


x A x1 M n, p x2 M n, p /xk M n, p
(x, h nM (y , p), h nM (x1 , p), h nM (x2 , p), h nM (xk , p)),
where y M n, p . But then A M

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

11 0# and Jensens Covering Lemma

If and are as in Lemma 11.29 then is often called the n-completion of .


Following [30, Sect. 2] we shall call embeddings arising from applications of
the Downward and Upward Extension of Embeddings Lemma r n+1 elementary.
Here is our official definition, which presupposes that the structures in question
possess very good parameters.
Definition 11.32 Let M, N be acceptable, let : M N , and let n < . Then
is called r n+1 elementary provided that there is p R nM with ( p) R nN , and for
all i n,
(11.31)
 M i, pi : M i, pi 1 N i,( p)i .
The map is called weakly r n+1 elementary provided that there is p R nM with
( p) R nN , and for all i < n, (11.31) holds, and
 M n, p : M n, p 0 N n,( p) .
If : M N is (weakly) r n+1 elementary then typically both M and N will
be n-sound, cf. Lemma 11.38; however, neither M nor N has to be (n + 1)-sound. It
is possible to generalize this definition so as to not assume that very good parameters
exist (cf. [30, Sect. 2]).
With the terminology of Definition 11.32, Lemma 11.29 says that the map can
be extended to its n-completion which is weakly r n+1 elementary, and if is
1 elementary to begin with, then the n-completion in fact be r n+1 elementary.
Moreover, if a map : M N is r n+1 elementary, then respects h n+1 by
Theorem 10.16:
Lemma 11.33 Let n < , and let M and N be acceptable J-structures. Let : M
N be r n+1 elementary. Let p Mn+1 be such that p  n R nM and ( p  n) R nN .
Then for all k < and x M n+1, p ,


n+1, p
n+1,( p)
h M (k, x) = h N
(k, (x)).
Recall the well-ordering < of finite sets of ordinals from Problem 5.19: if u,
v OR< , then u < v iff max(uv) v. If M is an acceptable J-model and
n < , then
the well-ordering < induces a well-ordering of Mn by confusing
n
p M with ran( p). We shall denote this latter well-ordering also by < .
n is called
Definition 11.34 Let M be an acceptable J-structure. The < -least p PM
the nth standard parameter of M and is denoted by pn (M). We shall write M n for
M n, pn (M) . M n is called the nth standard reduct of M.

Lemma 11.35 Let M be an acceptable J-structure, let n < , and let p R nM .


n+1
with p  n = p. In particular, if n > 0 and M is
Then there is some p PM
n-sound, then pn1 (M) = pn (M)  (n 1).
Proof This follows immediately from Lemma 11.24 (c).

11.1 Fine Structure Theory

255

Definition 11.36 Let M be an acceptable J-structure.


Suppose that for all n < ,

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)

Let q Mn be such that q  (n 1) = q and q (n 1) = 1 (q(n 1)),


Because p = pn (M) R nM , there are e < and
so that (q ) = q ran().
n, p
<
z [n (M)] such that q = h M (e, z), i.e.,
n, p

p < q e < z [n (M)]< (q = h M (e, z)).

(11.34)

We aim to verify that (11.34) also holds true in ran():

p < q, for which there are e < and


Claim 11.39 There is some p ran(),
n, p
<
z [n (M)] such that q = h M (e, z).
Proof Let i n 1 be least such that p(i) < q(i). (So p  i = q  i.) Let us
recursively define (( p (k), x(k)): i k n 1) as follows.

11 0# and Jensens Covering Lemma

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)

as being witnessed by p(i), and because  M i,q i is 1 -elementary, we may let


p (i) be the < -least r ran( ) as in (11.35), and we let x(i) be some x ran( )
as in (11.35) such that q(i) = h M i,qi (e , (x, p (i))) for some e < . For the record,
p (i) < q(i).
Having defined ( p (k 1), x(k 1)), where i < k n 1 and { p (k 1), x(k
1)} ran( ), we will have that
M k,q k |= r x [k+1 (M)]< e < (x(k 1), q(k)) = h M k,qk (e , (x, r )), (11.36)

as being witnessed by p(k), and because  M k,q k is 1 -elementary, we may let


p (k) be the < -least r ran( ) as in (11.36), and we let x(k) be some x ran( )
as in (11.36) such that (x(k 1), q(k)) = h M k,qk (e , (x, p (k))) for some e < .
We may now set p = p  i {(k, p (k)): i k n 1}.
It is straightforward to verify that for each k n 1,
h M k, p k (k+1 (M) { p (k)}) = h M k,qk (k+1 (M) {q(k)}),
p (k)

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.

11.1 Fine Structure Theory

257

Let k < and 1 < < k be such that p1 (M) = {1 , . . . , k }. Let be a


1 formula such that for every < 1 (M),
A M |= (, 1 , . . . , k , q).
Because (W, r ) M is a witness for p with respect to M, p, we have that
M |= (, 1 , . . . , k , q) = W |= (, 1 , . . . , k , r )

(11.38)

for every < 1 (M) and every 1 formula .


Say W = J [E]. Let = sup(h W ( {r })OR) , and write W = J [E]. Let
us define : h M ( {q}) W by setting h M (e, (, q)) h W (e, (, r )), where
e < and []< . By (11.38), is well-defined and 0 -elementary. By the
choice of , is cofinal and hence 1 -elementary by Lemma 11.13 (a). Therefore,
M |= (, 1 , . . . , k , q) W |= (, 1 , . . . , k , r )

(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

Proof We have to show (2) = (1). Let : W M M be the inverse of the


transitive collapse. As in the proof of Lemma 11.41, say W = J [E], set =
sup(h W ( {r }) OR) , and write W = J [E]. We may define a 1 -elementary
, p
embedding : W M W by setting
1 (h M (e, (, p \ ( + 1)))) h W (e, (, r )),
where e < and []< .
, p
Now if () = then a witness to 1 (M) is definable over W M , and hence over
W . But as W M, this witness to 1 (M) would then be in M. Contradiction!
We thus have that must be the critical point of . Thus we know that () is
regular in M, and hence writing M = J [E ], J () [E ] |= ZFC . We may code
, p
, p
W M by some a , definably over W M . Using , a is definable over W , so that
a M. In fact, a J () [B] by acceptability. We can thus decode a in J () [B],
, p

which gives W M J () [B] M.

11 0# and Jensens Covering Lemma

258

Definition 11.44 Let M be an acceptable J-structure. We say that M is 1-solid iff


, p1 (M)

WM

for every p1 (M).


M be acceptable J-structures, and let : M 1 M. Let
Lemma 11.45 Let M,
<

Let (W , r ) be a witness for


p OR M , and set = ( ) and p = ( p).

with respect to M, p such that W M, and set W = (W ) and r = (r ). Then


(W, r ) is a witness for with respect to M, p.
Proof Let be a 1 -formula. We know that
M |= 0 < . . . l < ((0 , . . . , l , p \ ( + 1)) W |= (0 , . . . , l , r )).
As is 1 -elementary, this yields that
M |= 0 < . . . l < ((0 , . . . , l , p \ ( + 1)) W |= (0 , . . . , l , r )).
We may thus conclude that (W, r ) is a witness for with respect to M, p.

M be acceptable J-structures, and let : M 1 M.


Corollary 11.46 Let M,
PM . Then p1 (M) = ( p1 ( M)),
and M
Suppose that M is 1-solid and ( p1 ( M))
is 1-solid.
The following lemma is a dual result to Lemma 11.45 with virtually the same
proof.
M be acceptable J-structures, and let : M 1 M. Let
Lemma 11.47 Let M,
<

Let (W , r ) M be such
p OR M , and set = ( ) and p = ( p).

that, setting W = (W ) and r = (r ), (W, r ) is a witness for with respect to M,


p.
p. Then (W , r ) is a witness for p with respect to M,

M be acceptable J-structures, and let : M 1 M.


Corollary 11.48 Let M,
, p (M)
ran( ) for every p1 (M).
Suppose that M is 1-solid, and that in fact W M 1
= 1 ( p1 (M)), and M is 1-solid.
Then p1 ( M)
We now generalize Definition 11.44.
Definition 11.49 Let M be an acceptable J-structure. If 0 < n < then we say
that M is n-solid if for every k < n, p1 (M k ) = pk+1 (M)(k) = pn (M)(k) and M k
is 1-solid, i.e.,
, p (M k )
Mk
WMk 1
for every p1 (M k ). We call M solid iff M is n-solid for every n < , n > 0.

11.1 Fine Structure Theory

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

= 1 ( pn (M)) and M is n-solid.


then pn ( M)

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)

Both M and N are n-sound,


is r n+1 elementary,
( pk (M)) = pk (N ) for every k n, and
(k (M)) = k (N ) for every k < n and n (N ) = sup( "n (M)).

Other examples for n-embeddings are typically obtained as follows. Let M be


acceptable, and let, for n , Cn (M) denote the transitive collapse of h nM "(n (M)
{ pn (M)}). Cn (M) is called the nth core of M. The natural map from Cn+1 (M) to
Cn (M) will be an n-embedding under favourable circumstances.
Lemma 11.53 For each limit ordinal , J is acceptable and sound.
Proof by induction. Suppose that for every limit ordinal < , J is acceptable
and sound.
Let us first verify that J is acceptable. By our inductive hypothesis, this is trivial
if is a limit of limit ordinals, so let us assume that = + , where is a limit
ordinal. We need to see that if < is such that
(P( ) J+ ) \ J = ,

(11.40)

then there is some surjection f : with f J+ . Let be least with (11.40).


We claim that
(11.41)
= (J ).
To see (11.41), note first that if n is such that n (J ) = (J ), then there is a

(J )n
1

subset of (J ) which is not in J . Such a set is by Lemma 11.26

and the soundness of J , and it is hence in J+ \ J by Lemma 5.15. Therefore,


J

(J ). On the other hand, let a such that a J+ \ J . Then a

by Lemma 5.15. As a (J ) and J is sound, Lemma 11.27 yields that a is


(J )n

for some n < . Hence (J ) and (11.41) follows.

11 0# and Jensens Covering Lemma

260

Now by the soundness of J again and by Lemma 11.25 there is some f

such that f : = (J ) J is surjective. By Lemma 5.15, f J+ . We have


verified that J is acceptable.
We are now going to show that J is sound. We shall make use of Lemma 11.38
and verify that for every n < ,
pn (J ) R nJ .

(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)

If < , then (J )n, p J , so that B n+1 (J ) J , contradicting (11.44). We


must therefore have that = . Then p R nJ . For every i < n, we certainly have

= p(i). By the choice of p = pn+1 (J ), p(i) p(i).

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

But B 1 ({ p }) and (11.44) then yield that p P(J )n . We must also


have p p(n), as ( p ) = p(n). By the choice of p(n), p(n) p , so that
p = p(n).
But now we must have that = id, and therefore p(n) R(J )n , i.e., p R n+1
J .
Contradiction!


11.1 Fine Structure Theory

261

Lemma 11.54 For each limit ordinal , J is solid.


Proof By Lemmas 11.53 and 11.35, it suffices to prove that if n < and
p1 ((J )n ) = p(J )(n), then
, p ((J )n )

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 )

= W 1n . By the Downward Extension of Embeddings Lemma


so that (J , A)
(J )
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

(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)

Card(T C({(J , A)})) = Card() inside J . As < n (J ) and Jn (J ) =


Jn (J ) , where
(Hn (J ) ) J by Corollary 11.6, we then get that in fact (J , A)
n
Jn (J ) is the universe of (J ) . Hence (11.45) follows.
We are left with having to prove that < . Suppose that = . Then, as
( p)
= p  n, p p  n. However, p R nJ = R nJ PJn , so that by the choice
of p  n we must actually have that p = p  n. That is,
= (J )n,q = (J )n, pn .
(J , A)
(J )n

Let B 1

be such that B 1 ((J )n )


/ (J )n , say
B = {x (J )n : (J )n |= (x, r )},

where r (J )n and is 1 . As p1 ((J )n ) [1 ((J )n ), (J )n OR)< and


(J )n
 = id, we have that  1 ((J )n ) = id. Therefore, if we let B 1 be
defined as

B = {x (J )n : (J )n |= (x, (r ))},

then
/ (J )n .
B 1 ((J )n ) = B 1 ((J )n )

(11.47)

11 0# and Jensens Covering Lemma

262

As ran( ) = h (J )n ( ( p(n) \ ( + 1))), there is m < and


s ( ( p(n) \ ( + 1)))<

(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).

11.2 Jensens Covering Lemma


We are now going to prove Jensens Covering Lemma, cf. Theorem 11.56. For this,
we need the concept of a fine ultrapower.
Definition 11.55 Let M be an acceptable J-structure, and let E be a (, )-extender
over M. Let n < be such that n (M) (E). Suppose that M is n-sound, and set
p = pn (M). Let
: M n, p N
be the 0 ultrapower map given by E. Suppose that
: M N
is as given by the proof of Lemmas 11.20 and 11.31. Then we write ultn (M; E) for
N and call it the r n+1 ultrapower of M by E, and we call the r n+1 ultrapower
map (given by E).
Lemmas 11.20 and 11.31 presuppose that is good (cf. Definition 11.19). However, the construction of the term model in the proof of Lemma 11.20 does not
require to be good, nor does it even require the target model N to be well-founded.
Consequently, we can make sense of ult n (M; E) even if is not good or N is not
well-founded. This is why we have the proof of Lemmata 11.20 and 11.31 in the
statement of Definition 11.55, as it does not assume anything about or N which is
not explicitly stated. We shall of course primarily be interested in situations where
ult n (M; E) is well-founded after all. In any event, as usual, we shall identify the
well-founded part of ultn (M; E) with its transitive collapse.

Recall that S [ ] is called stationary iff for every algebra A = ( ; ( f i : i < ))


with at most many functions f i , i < , there is some X S which is closed
from A, cf. Definition 4.39.
under all the f i , i < ,
Theorem 11.56 The following statements are equivalent.
(1) Jensen Covering holds, i.e., for all sets X of ordinals there is some Y L such
that Y X and Y X + 1 .
(2) Strong Covering holds, i.e., if 1 is a cardinal and , then [ ] L is
stationary in [ ] .
(3) L is rigid, i.e., there is no elementary embedding : L L which is not the
identity.
(4) 0# does not exist.

11.2 Jensens Covering Lemma

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)

such that for all X 1 , . . . , X k [ ]i , ik (X 1 , . . . , X k ) X 1 X k and


ik (X 1 , . . . , X k ) is closed under all functions f , < i .
Let us now consider the algebra
A = ([ ]< W ; (ik : i < cf(), 1 k < )).
It is easy to see that our hypothesis yields that if A is any set in W , then [A]cf()+1 W
size cf() + 1
is stationary in [A]cf()+1 . In particular, we find some Y W of
such that Y is closed under all the functions from A . Set X =
Y . Of course,
X W . Moreover X , as X is the union of cf() + 1 < many sets of size
<. We claim that X is closed under all functions f , < .
Let < and let f be n-ary, n < . We aim to see that f [X ]k X .
Let xl X l Y , 1 l k, and let i < cf() be such that i > and also
i > Card(X l ), 1 l k. Then
f (x1 , . . . , xk ) ik (X 1 , . . . , X k ) Y ,
and therefore f (x1 , . . . , xk ) X .

11 0# and Jensens Covering Lemma

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)

If i and i = , then we let n i = n i = 0.


In what follows we shall make frequent use of the notation introduced by Definition 10.47. E.g., for i , E  Ji is the (long) (crit( ), sup i )extender
derived from  Ji . Notice that by (11.50), we have that
ult n i (Ji ; E  Ji )
makes sense for all i . If i j and i = j then n j n i . Hence {n i +1 (Ji ) :
i } is finite. Let I = I be such that
{i : i I = I } = {n i +1 (Ji ) : i } { },
cf. Lemma 11.5.
The following Claim is the key point.
Claim 11.58 Suppose that for all i I , ult n i (Ji ; E  Ji ) is well-founded. Then
either L is not rigid or else ran( ) L.
Proof We may that assume = id, as otherwise the conclusion is trivial. Let us
assume that L is rigid and show that ran( ) L.
There must be some i I such that i crit( ). This is because otherwise,
H
letting < be such that = (crit( ))+L , = and the ultrapower map
: L ult0 (L; E  J )
=L

11.2 Jensens Covering Lemma

265

would witness that L is not rigid.

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

We now aim to show by induction on i I that i L. This is trivial for


i I such that i crit( ), as then i = i L. Now suppose inductively that
i L, where i I . If i = , then we are done. Otherwise let J be the least
element of I \(i + 1). We may assume that j > crit( ), as otherwise again trivially
j = j L. We must then have that j < , as otherwise the ultrapower map
: L ult 0 (L; E  J j )
=L
would witness that L is not rigid. But then n j +1 (J j ) i . Let
: J j ult n j (J j ; E  J j )
= J

(11.51)

be the ultrapower map. Let p = p(J j )  (n j + 1). By Lemmas 11.53 and 11.25,

11 0# and Jensens Covering Lemma

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)

But as n j +1 (J j ) i and i L, (11.53) says that ran()


L, and hence
j = j = ( j ) ran( ) L,


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)

R, and ([ak , f k ]: k < )


a badness witness for ultn i (Ji ; E  Ji ), provided that R,
are as in the preceding paragraph.
If X H then we let X denote the inverse of the transfinite collapse of X . In
the light of Claim 11.58, in order to finish the proof of (3) = (2) of Theorem 11.56
it suffices to verify the following.
Claim 11.59 There is a stationary set S [ ] such that wherever X S , then
for all i I X ,
ultniX (J X ; E X  J X )
i

is not bad.

11.2 Jensens Covering Lemma

267

Proof Let A = ( ; ( f i : i < )) be given. We recursively define sequences (Yi :


i ), (H i : i ) and (i : i ) such that the following hold true.
(1) Yi H , for all i
(2)
(3)
(4)
(5)
(6)

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

) is bad. Then for every R such that

R, ([ak , f k ] : k < )) for ultn ij (J i ; E  J


there is a badness witness ( R,
i

j i

R, ([a i, j , f i, j ] : k < )) for ultn ij (J i ; E  J


there is a badness witness ( R,
i
k
k

j i

),
)

i, j

with the property that {ak : k < } Yi+1 .

Let us write = . We claim that for all j I , ult 0 ((Ji )n j ; E  J ) is


j

well-founded and the ultrapower map

: (Ji )n j ult 0 ((Ji )n j ; E  J )


j

is good. This implies that for all j I , ult n j (Ji ; E  J ) is well-founded by


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

11 0# and Jensens Covering Lemma

268

(ak+1 , ak ) ({(u, v) : f k+1 (u) R f k (v)})


= ({(u, v) : ( f k+1 )(u) (R ) ( f k )(v)})
= ({(u, v) : f k+1 (u)R f k (v)})
= i0 ({(u, v) : f k+1 (u)R f k (v)}), as  H i0 = i0 ,
= i0 ({(u, v) : f k+1 (u)R f k (v)}).
We may assume that j ran( ) if j < . Let = 1 ( j ) if j < ,
= H OR otherwise. Write n = n j , = 1 ( j ).
We have that {[ak , f ] : k < } ult 0 ((J )n ; E i0  J ), and R is defined over
i
i
(J )n in the same way as R is defined over (J j )n j . Also, if = 0 , then 0

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

, 1 aki0 ,l (u, v) : f k+1


(u) R
f ki0 ,l (v) ,
1 ak+1
and therefore

i 0 ,l
f k+1








i 0 ,l

1 ak+1
R
f ki0 ,l 1 aki0 ,l .

Because this holds for all k < , R is ill-founded. Contradiction!


This finishes the proof of Theorem 11.56.




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 .

11.2 Jensens Covering Lemma

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() .

11.3  and Its Failure


We now aim to prove  in L. This is the combinatorial principle the proof of which
most heavily exploits the fine structure theory.
Definition 11.62 Let be an infinite cardinal, and let R + . We say that  (R)
holds if and only if there is a sequence (C : < + ) such that if is a limit ordinal,
< < + , then C is a club subset of with otp(C ) and whenever is a
/ R and C = C . We write  for  ().
limit point of C then
In order to prove  in L, we need the Interpolation Lemma. The proof is very
similar to the proof of Lemma 10.56, and we omit it. Recall the concept of a weakly
r n+1 elementary embedding, cf. Definition 11.32.
M be an acceptable J-structure, and let
Lemma 11.63 Let n < . Let M,
: M M
be r n+1 elementary. Let M OR, and let E be the (, )-extender derived
from .
There is then a weakly r n+1 elementary embedding
E) M
: ult n ( M;
such that  = id and E = .
Theorem 11.64 (R. Jensen) Suppose that V = L. Let 1 be a cardinal. Then
 holds.

11 0# and Jensens Covering Lemma

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( )

n()+1, pn()+1 (J() )

(i, x) h J()

(i, x),

(11.57)

where i < and x []< . We shall denote this map by , .


Notice that if C, then again by Lemma 11.53
n()+1, pn()+1 (J() )

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

11.3  and Its Failure

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( ).

Let = 1 () (if = , we mean = ).


Let
= E  J : J r n+1 ult n (J ; E  J ).
By Lemma 11.63, we may define a weakly r n+1 elementary embedding
k: ult n (J ; E  J ) J
with k = . In particular, ult n (J ; E  J ) is well-founded and we may identify
it with its transitive collapse. Let us write J = ult n (J ; E  J ). As ran( ),
k 1 () > . Moreover, k 1 () = sup " < , as cf() > . Therefore <
k 1 () D . This shows (b).
Let us now verify (c). If D , then D D . To show (c), we thus let <
be such that and are both in D . We need to see that D , i.e., that , is
well-defined. For the purpose of this proof, let us, for {, ,
}, abbreviate
n( )+1, pn( )+1 (J( ) )

h J( )

(i, x) by h (i, x),

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

11 0# and Jensens Covering Lemma

272

J, ( ) in a way that it is preserved by the weak r n+1 elementarity of , to yield


that h (e, y) exists after all. Contradiction! Therefore, (11.58) holds true.
But now if i < , x []< , and h (i, x) exists, then v v = h (i, x) may
be expressed by a statement over J() in the parameter J , () in a way that it is
preserved by the weak r n+1 elementarity of , to give that h (i, x) also exists.
Claim 11.65 is shown.

Now let C. We aim to define C . Set = () and n = n(). Recursively, we
define sequences (i : i ()) and (i : i < ()) as follows. Set 0 = min(D ).
Given i with i < , we let i be the least < such that
n+1, pn+1 (J )

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

Claim 11.66 Let C. The following hold true.


(a)
(b)
(c)
(d)
(e)

(i : i < ()) is strictly increasing.


otp(C ) = () .
C is closed.
If C then C = C .
If D is unbounded in then so is C .

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( )

n()+1, pn()+1 (J() )

h J()

(k, x) ran(, ) =

(k, ) ran(, ) = .

, ) contains the relevant witness


i+1
On the other hand, ran(i+1
This gives i+1

.
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 () .

11.3  and Its Failure

273

Of course, otp( B ) for every < + , and because C D C for every


C we have that every B is closed, if cf() > , then B is unbounded in ,
and if B , then B = B . For every < + such that cf() = and C is
not unbounded in , let us pick some B of order type which is cofinal in . For
< + , let

B if B is defined, and
B =
B otherwise.
It is easy to see now that (B : < + ) witnesses that  holds true.

Corollary 11.60 and Theorem 11.64 immediately imply the following.


Corollary 11.67 Let be a singular cardinal, and suppose that  fails. Then 0#
exists.
Let (C : < + ) witness that  holds true. By Fodors Theorem 4.32, there
must be some stationary R + such that otp(C ) is constant on R, say = otp(C )
for all R. For any < + , C can have at most one (limit) point such that
C = C has order type . We may then define
C =

C \ ( + 1) if C and otp(C ) =
otherwise.
C

(C : < + ) then witnesses  (R), cf. also Problem 11.17.


The following result generalizes Lemma 5.36, as 0 is provable in ZFC.
Lemma 11.68 (Jensen) Let be an infinite cardinal. Suppose that there is some
stationary set R + such that both + (R) and  (R) hold true. There is then a
+ -Souslin tree.
Proof Let R + be stationary such that + (R) and  (R) both hold true.
Let (S : R) witness + (R), and let (C : < + ) witness  (R). We will
construct T  by induction on < + in such a way that the underlying set of
T  will always be an ordinal below + (which we will also denote by T  ).
We set T  1 = {0}. Now let < be such that T  has already been
constructed. If is a successor ordinal, then we use the next (T  T  ( 1)) 2
ordinals above T  to provide each top node of T  with an immediate successor
in T  ( + 1).
Let us then suppose to be a limit ordinal. Let us set S = {s T  : t
S (t = s t <T s)} if R and S happens to be a maximal antichain of T  ,
and let us otherwise set S = T  .
Let us for a moment fix s S. We aim to define a chain cs through T 
as follows. Let C be least such that in fact s T  . Let (i : i < ) be
the monotone enumeration of C \ . We then let s0 be the least ordinal such that
s0 T  (0 + 1) \ T  0 and s <T s0 , and for i > 0, i < , we let si be the least
ordinal such that si T  (i + 1) \ T  i and s j <T si for all j < i, if there is one
(if not, then we let the construction break down). We set cs = {si : i < }.

11 0# and Jensens Covering Lemma

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)

But C () = (D ) and C () = (D ), so that by (11.59)


D = D .
Setting
D=

D ,

()C

we then have that D is cofinal in + and D = D for every () C. Pick



() C such that otp(D ) > . Then otp(D ) > . Contradiction!

11.3  and Its Failure

275

Lemma 11.70 (Solovay) Let be + -supercompact, where . Then  fails.


In particular, if is supercompact, then  fails for all .
Proof Let be + -supercompact, where , and suppose that (D : < + )
witnesses that  holds. Let
: V M
+

be such that M is an inner model, crit( ) = , () > and M M. This


implies that = sup( + ) < (+ ). Let (C : < (+ )) = ((D : < + )).
As + is < -club in ,
C = C +
is also < -club in . The rest is virtually as in the proof of Lemma 11.69.

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

11 0# and Jensens Covering Lemma

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 < .

Let us write ||t||CT B for the Cantor-Bendixson rank of t in T , cf. Problem


7.5. It suffices to prove that there is some < 1 such that ||||CT B = , as
Otherwise we may construct
then T N, T N T , and T  ran( ) .
some x (2 ) such that for all n < and for all (sufficiently big) < 1 ,
||tx (n+1) ||CT B < ||tx n ||CT B .] Show also that if C H holds in V , then forcing
with N does not add a new real.
Conclude that it is not possible to cross out +1 in (1) or replace 1
by 0 in (2) of Theorem 11.56.
11.8. Show that 0# is not generic over an inner model which does not contain 0# ,
/ W , if P W is a poset, and if G V
i.e., if W is an inner model with 0#
/ W [G].
is P-generic over W , then 0#
11.9. Assume that 0# does not exist. Let W be any inner model such that (2 )W =
2 . Show that for every , [ ] W is stationary in [ ] .
11.10. Suppose V = W [x], where x is P-generic over W for some P W .
Suppose that W and V have the same cardinals, W |= CH, but V |= CH.
Show that 0# W . [Hint. Use Strong Covering.]
11.11. (M. Magidor) Assume ZF plus both 1 and 2 are singular. Show that 0#
exists. (Compare Theorem 6.69.)
11.12. Show that if 0# does not exist and if is weakly compact, then +L = + .
[Hint. Use Problem 4.23.]
11.13. Let be limit ordinals, and let : J J be 0 -cofinal. Assume
that cf()
= cf() > . Let be such that is a cardinal in J ,
Show that ult n (J ; E ) is welland let n < be such that n (J ) .
founded. [Hint. Let : J J be elementary, where is countable, and
let : J ult n (J ; E  1 (J ) ). By Lemma 11.63, this ultrapower is
well-founded, say equal to J , and there is an embedding k: J J .
By hypothesis, < .
We may assume that : H H , where

11.4 Problems

277

is sufficiently big and H is (countable and) transitive. We may then embed


1 (ult n (J ; E )) into (J ).]
11.14. Assume that 0# does not exist. Let be a cardinal in L, and let X J
be such that X 2V 2V and if 2V < is an L-cardinal, then
cf(X +L ) > . Show that we may write X = n< X n , where X n L
for each n < . [Hint. Problem 11.13.]
11.15. (M. Foreman, M. Magidor) Assume V = L, and let X J be such that
cf(X n+1 ) > for all n < . Show that there is some n 0 < such that
for all n n 0 , cf(X n ) = cf(X n 0 ). [Hint. Problems 11.13 and 11.5.]
11.16. Assume V = L, and 1 be a cardinal. Let C and n() (for C)
be defined as in the proof of Theorem 11.64. Show that for every n < ,
{ C: n() = n} is stationary in + .
11.17. Show that if  holds, then there is some stationary R + such that  (R)
holds. Also, if  holds, then there is some stationary S + such that for
no < + , S is stationary in .
11.18. (R. Jensen) (Global ) Assume V = L. Let S be the class of all ordinals
such that cf() < . Show that there is a sequence (C : S) such that
for every S, C is club in , otp(C ) < , and if is a limit point of C ,
then S and C = C . [Hint. Imitate the proof of Theorem 11.64.]
Let be a limit ordinal. A sequence (C : < ) is called coherent iff for all
< , C is a club subset of and whenever is a limit point of C , then
An ordinal is called threadable iff every coherent sequence
C = C .
(C : < ) admits a thread C, i.e., C is club and for every limit point
of C, C = C .
11.19. Let be a limit ordinal with cf() > . Show that is threadable iff cf() is
threadable.
11.20. Show that if + is threadable, then  fails. Also, if is weakly compact,
then is threadable.
11.21. (R. Jensen) Assume V = L, and let be not weakly compact. Show that
there is an unthreadable coherent sequence (C : < ). [Hint. Let T be a
-Aronszajn tree. Let
S = { < : > J |= ZFC and is regular and
there is a cofinal branch through the -tree T J }.
Notice that if
/ S, then is singular. Now imitate the argument from
Theorem 11.64 or rather from Problem 11.18, working separately on S and
on \ S.]
11.22. Let be an uncountable regular cardinal which is not threadable. Show that
if G is Col(1 , )-generic over V , then 1 is not threadable in V [G]. [Hint.
Use Problem 11.19. Let (C : < ) V witness that is not threadable

278

11 0# and Jensens Covering Lemma

in V . If 1 were threadable in V [G], we could pick H such that G, H are


mutually Col(1 , )-generic over V such that both V [G] and V [H ] contain
a thread. As cf() = 1 in V [G][H ], such a thread would then also be in
V [G] V [H ] and hence in V by Problem 6.12.]
Let 2 be regular, and let S be stationary. We say that S reflects iff
there is some < with cf() > such that S is stationary in .
11.23. Let be regular. Show that S = { < + : cf() = } does not reflect.
11.24. Show that if is weakly compact, then every stationary S reflects.
11.25. Assume V = L. Let 2 be regular but not weakly compact. Suppose
that S is stationary. Show that there is some stationary T S which
does not reflect. [Hint. If = + , then use  , cf. Theorem 11.64. If is
inaccessible, then exploit the argument from Problem 11.21.]
11.26. (J. Baumgartner) Let be weakly compact, and let G be Col(1 , < )generic over V . Show that the following is true in V [G]. Let S = 2 be
stationary such that cf() = for all S; then S reflects. [Hint. In V let
us pick : H H exactly as in Problem 4.23, and let : H [G] H [K ]
be as in Problem 6.17. We may assume that our given S is in H [G]. We have
1 H V [G] = 1 H H [G] H [G] V [G]. It suffices to prove that S
is stationary in H [K ], which follows from Problem 6.15(b).]

Chapter 12

Analytic and Full Determinacy

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<

and a strategy for II is a function

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_12,


Springer International Publishing Switzerland 2014

279

280

12 Analytic and Full Determinacy

2n+1

X X.

n<

If is a strategy for I and z X , then we let z be the unique x X with


x(2n) = (x(0), x(1), x(2), . . . x(2n 1)) and
x(2n + 1) = z(n)
for n < , i.e., x is the element of X produced by a run of G(A) in which I follows
and II plays z. We say that is a winning strategy for I in G(A) iff
{ z: z X } A.
Symmetrically, if is a strategy for II and z , then we let z be the unique
x X with
x(2n + 1) = (x(0), x(1), x(2), . . . x(2n)) and
x(2n) = z(n)
for n < , i.e., x is the element of X produced by a run of G(A) in which II follows
and I plays z. We say that is a winning strategy for II in G(A) iff
{z : x X } \A.
Of course, at most one of the two players can have a winning strategy.
Definition 12.1 Let X be a non-empty set, and let A X . We say that G(A) is
determined iff player I or player II has a winning strategy in G(A). In this case, we
also call A itself determined. The Axiom of Determinacy, abbreviated by AD, states
that every A is determined.
We also refer to A as the payoff of the game G(A).
We shall mostly be interested in the case where X is countable, in fact X =
in which A is a set of reals. It can be shown in ZFC that if A is Borel, then
A is determined, cf. [20, Chap. 20, pp. 137148]. We here aim to show that every
analytic set is determined, cf. Theorem 12.20. It turns out that this cannot be done
in ZFC, though, cf. Corollary 12.27. We shall prove later (cf. Theorem 13.7) that
in fact every projective set of reals is determined. The full Axiom of Choice, AC,
though, is incompatible with AD.
Lemma 12.2 Assume ZF + AD. Then AC is false, but AC holds for sets of reals,
i.e., if (An : n < ) is a sequence of non-empty sets of reals, then there is some
f : with f (n) An for every n < .
Proof In the presence of AC, we may enumerate all strategies for I as ( : < 20 )
and all strategies for II as ( : < 20 ). Using AC again, let us pick sequences

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

12 Analytic and Full Determinacy

Let us now define a strategy for II in G(A) as follows. Let s = (x0 , . . . , xn ) be


a position in G(A) where its IIs turn to play, i.e., n is odd. Then let (s) be some
z X such that I doesnt have a winning strategy in G s z (A), if some such z exists;
otherwise we let (s) be an arbitrary element of X .
Claim 12.5 is a winning strategy for II in G(A).
Proof Let

I x0
x2
...
II
x1
x3
...

be a play of G(A) which is according to . Claim 12.4 can easily be used to


show inductively that for each even n < , I does not have a winning strategy
in G (x0 ,...,xn1 ) (A).
Now suppose that I I looses, i.e., x = (x0 , x1 , x2 , . . .) A. Because A is open,
there is some basic open set Us such that x Us A. We may assume lh(s) to be
even. Then I has a trivial winning strategy in G s (A): he may play as he pleases, as

every s x , x X , will be in A. But this is a contradiction!
Lemma 12.6 Let A X . Suppose that for every y X, {x X : y x A} is
determined. Then X \ A is determined.
Proof Let us first suppose that there is a y X such that I I has a winning strategy
in G({x X : y x A}). We claim that in this case I has a winning strategy
in G( X \A). We let () = y, and we let (y s) = (s), where lh(s) is odd.
It is easy to see that if x X is produced by a run which is according to , then
x X \A.
Now let us suppose that for all y X , I has a winning strategy y in G({x
X : y x A}). We claim that in this case II has a winning strategy in G( X \A).
We let (y s) = y (s), where lh(s) is even. It is easy to see that if x X is
produced by a run which is according to , then x A.

Corollary 12.7 Let A X be closed. Then A is determined.
As an application, we now give a
Proof of Theorem 10.11, which uses an argument from a paper by I. Neeman. Recall
that Theorem 10.11 says that if M is a transitive model of ZFC such that M |= U
is a normal measure on , and if (n : n < ) is a Prikry sequence over M with
respect to U , then G (n : n<) is PU -generic over M. We shall write P = PU .
Set G = G (n : n<) , cf. Definition 10.10. It is easy to see that G is a filter. We
shall prove that G is generic over M.
To this end, fix D M which is open and dense in P for the rest of this proof.
We aim to show that D G = .
To each s []< we shall associate a game G s . Let s = {0 < < k1 }
where k < . The game has two players, I and II, and starts with round k.

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

12 Analytic and Full Determinacy

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

is a position in G s in which I obeyed the rules and II played according to .


Proof Let Y denote the set of such that there is some X such that
t l1

I X ss,k
...
Xs
II
k
...

X
l1

is a position in G s in which I obeyed the rules and II played according to . We


want to see that Y U . Suppose that Y
/ U , i.e., \Y U . Consider the following
position in G s in which is as dictated by .
t l1

...
Xs
I X ss,
II

...

l1

\Y

As II follows and thus obeys the rules, \Y . On the other hand, Y by


definition of Y . Contradiction!

We now let Yst be as given by Claim 12.9. For each Yst we let X st, be a witness
to the fact that Yst , i.e., some X as in the statement of Claim 12.9 for Y = Yst .
We finally also assign some Z s U to s as follows. If there is some Z with
(s, Z ) D then we let Z s be some such Z ; otherwise we set Z = .
We now let1
X 0 = s Z s st Yst .
By our hypothesis on (i : i < ), there is some n < such that {n , n+1 , . . .} X 0 .
Set s = (l : l < n).
Claim 12.10 The following is a play of G s in which I obeys the rules and II follows
s .
x m,m
...
Xs
...
I X ss,kn
II
n
...
m
...
x m

Proof by induction on m. Notice that m X 0 and hence m Ys

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

cf. Problem 4.26 on s X s .

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

12 Analytic and Full Determinacy

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

. Player I wins iff


finite union of basic open sets such that (K i ) 22(i+1)
(n 0 , n 1 , n 2 , . . .) 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\

Because is a winning strategy for II, A B. Moreover, (K s )


for every i < ,

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

We covered A by a countable union B of closed intervals such that (B) .


Theorem 12.13 (Mycielski-Swierczkowski) Assume AD. Every set A
Lebesgue measurable.


is

Proof Fix A 2. Let B A, B 2 be Lebesgue measurable such that for all


Lebesgue measurable D B\A, D is a null set. Then B\A is a null set by Lemma
12.12, so that A is Lebesgue measurable.

The hypothesis of Theorem 12.13 therefore also gives that every A
Lebesgue measurable.

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

12 Analytic and Full Determinacy

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,

where, for a sequence p = (s0 , s1 , . . . , s2n1 ) of non-empty elements of < which


is according to , B p is the set of all x such that s p = s s0 s1 . . . s2n1
x, but for all non-empty t < , s p t ( p t) is not an initial segment of x,
i.e.,
B p Us p t ( p,t) = .
Every B p is thus nowhere dense, so that A Us is in fact meager.

Let us now assume AD. Let us suppose A to be non-meager. By Claim 12.15, II


does not have a winning strategy in G bm
(A), so that I has a winning strategy in

G bm
(A).
Setting
s
=

(),

esily
induces
a winning strategy for II in G bm
s ( \A).

Again by the Claim, ( \A) Us is now meager, so that A Us is comeager



in Us .
Theorem 12.16 (Mazur) Assume AD. Every A has the property of Baire.
Proof Let us fix A , and set
O=


{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

Theorem 12.17 (D. A. Martin) Assume AD. Let A be Turing invariant.


Then either A or \A contains a (Turing) cone. Also, if S is a set of ordinals and
A is S-invariant, then either A or \A contains an S-cone.
Proof We prove the first part. Let A be Turing invariant. If is a winning
strategy for I in G(A), then x T x A whenever T x, and if is a winning
strategy for II in G(A), then x T x \A whenever T x. The proof of
the second part is the same.

In what follows it will be very convenient to use the -notation. If x, y ,
then we write x y for that z with z(2n) = x(n) and z(2n + 1) = y(n).
If xn , n < , then xn is that z such that z(n, k) = z n (k), where
,  ,  is the Gdel pairing function (cf. p. 33).
Theorem 12.18 Assume AD. The following hold true.
(a)
(b)
(c)
(d)
(e)
(f)

1 is inaccessible to the reals.


1 is a measurable cardinal.
For every x , x # exists.
For every X 1 there is a real x such that X L[x].
The club filter on 1 is an ultrafilter.
2 is a measurable cardinal.

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

12 Analytic and Full Determinacy

Players I and II alternate playing natural numbers. Let us write x = (n 0 , n 1 , . . .)


and yi = (m 2i+1 3k+1 : k < ). We say that II wins iff
x WO = { X ||x|| + 1} {||yi ||: i < yi WO} X.
If I had a winning strategy, , then A = { x: x } would be analytic
(cf. the proof of Lemma 12.12). Also A WO, and by the Boundedness Lemma
7.12 there would be some countable such that {|| x||: x } . But then
II can easily defeat by playing some y such that yi WO for all i < and
{||yi ||: i < } = X ( + 1).
Therefore, II has a winning strategy, . Let G V be Col(, < 1V )-generic
over L[ ] (Cf. Problem 10.13). Then X iff L[ ][G] |= there is some x WO
such that = ||x|| and {||yi ||: i < yi WO}, where y is the result of
having II play according to in a play in which I plays x. By the homogeneity of
Col(, < 1V ), cf. Lemma 6.54, X is therefore in L[ ], cf. Corollary 6.62.
(e) Let X 1 , and let, using (d), x be such that X L[x]. By (c), x #
exists, so that X either contains a club or is disjoint from a club. (cf. Problem 10.11.)
(f) Let C = {(1V )+L[x] : x }. As x # exists for every x , C 2 . By
(d), C is cofinal in 2V , so that we may let : 2 C be the monotone enumeration
of C. In a fashion similar to (a), for X 2 we may let D(X ) be the Turing invariant
set {x : 1+L[x] X }, and we may define
F = {X 2 : D(X ) contains a cone}.
Using Theorem 12.17 and AC as in (a), F can be verified to be a < 1 -complete
ultrafilter on 2 .
It remains to be shown that F is <2 -complete. Let us fix a sequence (X i : i < )
such that X i F for every i < 1 . Let us consider the following game.
I n0
n1
...
II
m1
m2
...
Players I and II alternate playing natural numbers. Let us write x = (n 0 , n 1 , . . .)
and y = (m i : i < ). We say that II wins iff
x WO = {z : y T z}

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

12 Analytic and Full Determinacy

We have shown that for an S-cone of x, L[S, x] |= OD S -determinacy. Let x


be such that L[S, x] |= OD S -determinacy. Working inside L[S, x], we may then
define a filter on 1L[S,x] as follows.
For reals x, let |x| = sup{||y||: y T x y WO}. Let S = {|x|: x R}. Let
: 1 S be the order isomorphism. Now if A 1 , then we put A iff
{x: |x| A}
contains an S-cone of reals. It is easy to verify that HOD S witnesses that 1 is

measurable in HOD S .

12.2 Martins Theorem


Theorem 12.20 (D. A. Martin) Suppose that x # exists for every x . Then every
analytic set B is determined. In fact, if x and x # exists, then every 11 (x)
set B is determined.
Proof Let us fix an analytic set B, set A = \B. Recall that a set A is
coanalytic iff there is some map s <s , where s < , such that for all s, t <
with s t, <t is an order on lh(t) which extends <s , and for all x ,
x A <x =

<s is a well-ordering

sx

(Cf. Lemma 7.8 and Problem 7.7).


We have to consider the game G(B),
I n0
n2
...
II
n1
n3
...
in which I and II alternate playing integers n 0 , n 1 , . . ., and I wins iff x =
(n 0 , n 1 , . . .) B. We have to prove that G(B) is determined.
The key idea is to first consider the following auxiliary game, G (A).
I n0
n2
n4
...
II
n 1 , 0
n 3 , 1
n 5 , 2
...
In this game, I and II also alternate playing integers n 0 , n 1 , . . .. In addition, II has
to play countable ordinals 0 , 1 , . . . such that for all k < ,


k + 1, <(n 0 ,...,n k )
= ({0 , . . . , k }, <) ,

12.2 Martins Theorem

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 , 1 , 2 , . . .}, <),


(, <x ) =
where (i) = i for all i < , so that <x must be a well-order, and thus x A, i.e.
x
/ B.
Let us now suppose that I has a winning strategy for G (A) in L[x], call it .
Whenever 0 , . . . , k and 0 , . . . , k are countable x-indiscernibles with

({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

12 Analytic and Full Determinacy

there is certainly a set {0 , 1 , . . .} of countable x-indiscernibles such that

(, <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!


12.3 Harringtons Theorem


We now aim to prove the converse to the previous theorem. Well first need the
following
Lemma 12.21 Let x . Suppose that there is a real y such that whenever is
a countable ordinal with J [x, y] |= ZFC , then is a cardinal of L[x]. Then x #
exists.
Proof Suppose not. Let y be such that whenever is a countable ordinal
with J [x, y] |= ZFC , then is a cardinal of L[x]. Let be a singular cardinal,
and let be such that < < + and J [x, y] |= ZFC . As we assume that x #
does not exist, Weak Covering, Corollary 11.60, yields that < +L[x] = +V .
Let : J [x, y] J + [x, y] be elementary, where is countable and ran( ).
Write = 1 (). Obviously, J [x, y] |= ZFC , but is not an L[x]-cardinal (

is not even a cardinal in J [x]). Contradiction!

12.3 Harringtons Theorem

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)

By easy density arguments, T must be an infinite tree on , and H : T {} is


surjective. For s T , write T  s = {s : s s T }. Straightforward density
arguments also yield that T  s is a well-founded tree on iff H (s) , and if
H (s) , then H (s) is the rank of in T  s (i.e., the rank of s in T ). If <
and H (s) = , then T  s codes in the sense that T  s, ordered by , is a
well-founded relation of rank .
If (t, h) TCol(, ) and , then we may construe (t, h) as an element of
TCol(, ) by identifying ordinals in [, ) with . We define (t, h)| as (t, h ),
/ .
where, for s t, h (s) = h(s) if h(s) and h (s) = if h(s)
The following combinational fact will be crucial for later purposes.
Claim 12.23 Let < + , , and let (t, h) TCol(, ) and (t , h )
TCol(, ) be such that
(t, h)| + = (t , h )| + .
Let (u, g) (t, h) in TCol(, ). Then there is (u , g ) (t , h ) in TCol(, )
such that
(u, g)| = (u , g )|.
Proof The hypothesis implies that t = t. Set u = u. We now define g . Set u =
t {s u\t: g(s) }, and let g = h g  {s u\t: g(s) }. We are forced to

296

12 Analytic and Full Determinacy

let g  u = g . For s u \ u , we let g (s) = + k, where k < is the rank of


in the tree u  s = {s : s s u} (i.e., the rank of s in u).
Let s u and n < lh(s). We need to see that if g (s  n) , then g (s) <

g (s  n). This is clear if s  n and s are both in t or both in {s u\t: g(s)


}. If s  n t and s u\t with g(s) , then g (s) = g (s) = g(s) and
h(s  n) = g(s  n) > g(s). By (t, h)| + = (t , h )| + , we must then have
g (s  n) = g (s  n) = h (s  n) > g (s).
Finally let s  n u\t and hence s u\t. If g(s  n) , then clearly g(s) ,
/
too, so g (s) = g (s) = g(s) < g(s  n) = g (s  n) = g (s  n). If g(s  n)
and g(s) , then g (s  n) [, + ) and g (s) = g (s) = g(s), so clearly
/ and g(s)
/ , then g (s  n) = + k and
g (s) < g (s  n). If g(s  n)


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 .

12.3 Harringtons Theorem

297

The following Claim will be crucial.


Claim 12.24 Let be a limit ordinal, let be a formula of complexity n
(in the language for L[x, T ]), and let 1 , . . . , k be terms of rank < . If
+ ( + 2n) , then for all (t, h) TCol(, ),
TCol(,)

(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 ).

Let (t , h ) (t, h)| + ( + 2n) in TCol(, + ( + 2n)). By Claim 12.23,


there is (t , h ) (t, h) in TCol(, ) such that (t , h )| + ( + 2n 1) =
(t , h )| + ( + 2n 1) . Let (t , h ) (t , h ) in TCol(, ) be such that
(t , h )  L[x]

TCol(,)

J [x, T ] |= (0 , 1 , . . . , k )

for some term 0 of rank < . Let (t , h ) (t , h ) in TCol(, + ( + 2n))


such that (t , h )| + ( + 2n 2) = (t , h )| + ( + 2n 2) , which
may again be chosen by Claim 12.23. By the induction hypothesis,
(t , h )  L[x]

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 ).

is dense below (t, h)| + ( + 2n) , so that in fact

298

12 Analytic and Full Determinacy


TCol(,+(+2n))

(t, h)| + ( + 2n)  L[x]

J [x, T ] |= (1 , . . . , k ).

The converse direction is shown in exactly the same fashion.


Let us now assume analytic determinacy.
Let us fix x and a natural bijection e:
following game G.
I n0
n2
...
II
n1
n3
...

< .


Let us consider the

Let us write z 0 = (n 0 , n 2 , . . . ) and z 1 = (n 1 , n 3 , . . . ). We say that player II wins iff


the following holds true: if z 0 codes a well-founded tree T , i.e.,
T = {e(n 2i ): i < }
is a well-founded tree on , then z 1 codes a model (; E) of ZFC + V = L[x],
say E = {(k, l): z 1 (k, l) = 1},2 such that ||T || is contained in the transitive collapse
of the well-founded part of (; E). It is straightforward to verify that the payoff set
for G is analytic, in fact 11 (x), so that G is determined.
Claim 12.25 I does not have a winning strategy in G.
Proof Suppose that is a winning strategy for I . Let D be the set of all real codes
for well-founded trees, and let
D = { z 1 : z 1 }.
Then D is an analytic set, D D . It is easy to define a continuous function
f : such that for all z ,
z WO f (z) D .
By the Boundedness Lemma 7.12 there is some < 1 with
{||z||: f (z) D} ,
i.e.,

{||T ||: z 1 ( z 1 codes T )} .

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

Here and in what follows we use the notation ,  from p. 33.

12.3 Harringtons Theorem

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

({E}) and  J [x] J+


[x, , T ] for all , so that in particular

 J [x] J+ [x, , T ].

(12.2)

As b J [x] J [x], there is some n 0 such that for all < ,


b (; E) |= m n 0 , where m = (  J [x])( ).

(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)

Let (t, h) TCol(, + 1) force (12.4) to hold true, i.e,


TCol(,+1)

(t, h)  L[x, ]

< (
b J+
[x, , T ] |= (, 0 , 1 )),

which may be rewritten as saying that for all < ,


TCol(,+1)

b (t, h)  L[x, ]

J+
[x, , T ] |= (, 0 , 1 ).

(12.5)

The point is now that because 0 and 1 are of rank , letting = ( + 2) =


+ ( + 2) , we may use Claim 12.24 to rewrite (12.5) further to say that for
every < ,
TCol(,)

b (t, h)|  L[x, ]

J+
[x, , T ] |= (, 0 , 1 )

(12.6)

300

12 Analytic and Full Determinacy

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(,)

b = { < : (t, h)|  J [x, ]

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 ( ).

12.6. Assume ADR .


(a) Show that for all (A x : x ) such that = A x for every x ,
there is a choice function.
(b) Show that there is a < 1 -closed ultrafilter U on [ ]0 such that every
member of U is uncountable, {a [ ]0 : x a} U for every x ,
and U is normal in the following sense: if (A x : x ) is such that A x U
for every x , then there is some A U such that whenever x a A,
then a A x (Compare Problem 4.30) [Hint. For A [ ]0 , let A U iff I
has a winning strategy in G({ f ( ): ran( f ) A}). To show normality,
argue as follows. Let (A x : x ) be such that A x U for every x .
Let x be a winning strategy for I in the game corresponding to A x , x .
Let be a strategy for I such that if
I x0
x2
...
II
x1
x3
...
is a play, then for each n < there is an infinite X n \(n + 1), say
X n = {m(n, 0) < m(n, 1) < . . .}, such that xm(n,i) is according to xn in
a play where so far I played xm(n,0) , . . ., xm(n,i1) , and II played the first i
many reals from x0 , x1 , . . . that were not played by I ].
The Solovay sequence (i : i ) is defined as follows. Let be as in
Problem 12.5 (d). Let 0 = sup({: surjective f : , f OD }).
If i has been defined, then set = i provided that i = ; otherwise
let i+1 = sup({: surjective f : , f OD {A} }) for some (all)
A with ||A||<Wadge = i . If > 0 is a limit ordinal and i has been

302

12 Analytic and Full Determinacy

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.

13.1 Embedding Normal Forms


Definition 13.1 Let A . We say that A has an embedding normal form,
(Ms , s,t : s t < ),
iff M = V , each Ms is an inner model, each s,t : Ms Mt is an elementary
embedding, t,r s,t = s,r whenever s t r , and for each x ,
x A dir lim(Ms , s,t : s t x) is well-founded.
Such an embedding normal form is -closed, where is an infinite cardinal, iff
M M for each s < .
s
s
Even though the following result had already been implicit in Martins proof of
Theorem 12.20 (cf. also [26]), it was first explicitly isolated and verified in [43].
Theorem 13.2 (K. Windus) Let A . If A has a 20 -closed embedding normal
form, then A is determined.
Proof Let (Ms , s,t : s t
For x , we shall write

< )

be a 20 -closed embedding normal form for A.

(Mx , (s,x : s x)) = dir lim(Ms , s,t : s t x).

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4_13,


Springer International Publishing Switzerland 2014

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.

Let be an ordinal which is bigger than all xn .


We define T by setting (s, f ) T iff s < , f = ( f i : i < lh(s)), where each
/
f i is a function from to , and for all i + 1 < lh(s) and for all x , if x
A s  i + 1 = x  i + 1, then f i+1 (x) < f i (x), and if x A s  i + 1 = x  i + 1,
then f i+1 (x) = 0. The order on T is reverse inclusion.
Now if ( f i : i < ) witnesses that x p[T ], then x A, because otherwise
f i+1 (x) < f i (x) for each i < . Hence p[T ] A. On the other hand, if x A,
then we may define a witness ( f i : i < ) to the fact that x p[T ] as follows: Let
/ A, then let k < be maximal with x  k = x  k and define
x . If x

f i (x ) = k + 1 i for i k and f i (x ) = 0 for i > k. If x A, then define


f i (x ) = 0 for all i < . This shows A p[T ], and hence A = p[T ].
Let us now consider the following game, called G (A):
n2 , f1
...
I n0 , f0
II
n1
n3
...
Here, each n i is a natural number, and each f i is a function from to . I wins iff
((n i : i < ), ( f i : i < )) [T ]. The payoff set is thus a closed subset of

((

),

and is hence determined by Theorem 12.3.


Let us first suppose I to have a winning strategy for G (A). Then a winning
strategy for I in G(A) is obtained by playing as according to , but hiding the side
moves f i . Recall that ((n i : i < ), ( f i : i < )) [T ] proves that (n i : i < ) A,
so that is indeed a winning strategy for I .
Now let us suppose I I to have a winning strategy in G (A). We aim to produce
a winning strategy for I I for G(A).
Let k < , and let
n2
...
n 2k
I n0
II
n1
n3
...

13.1 Embedding Normal Forms

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)

by the construction of the s i and of T . Equation (13.2) holds true because if x


/
A s = x  lh(s), then (13.1) yields that for all i < lh(s), s i+1 (x) < s i,s i+1 (x),
and thus s (i+1),s (xi+1 ) = s i+1,s (s (i+1) )(x) < s i+1,s s i,s i+1 (s i )(x) =
s i,s (s i )(x) = s i,s (xi ). Moreover, if x A or s = x  lh(s), then
s (i+1)s (s (i+1) )(x) = 0.
We may therefore define by letting ((n 0 , . . . , n 2k )) be the unique n < such
that
Ms |= n = ,s ( )((n 0 , ,s ( ), n 1 , . . . , n 2k , s )).
Suppose not to be a winning strategy for I I for G(A). There is then a play
(n 0 , n 1 , . . .) of G(A) in which I I follows , but I wins, i.e., setting x = (n 0 , n 1 , . . .),
x A. In particular, Mx is wellfounded (i.e., transitive), and by the elementarity
of ,x ,
Mx |= ,x ( )is a winning strategy
for I I in ,x (G (A)).
By (13.1) and the elementarity of ,x , in V there is a play of ,x (G (A)) in which
I I follows ,x ( ) and in which I I loses, namely
n 2 , x 1,x (x 1 )
...
I n 0 , ,x ( )
II
n1
...
We may now exploit the absoluteness of wellfoundedness between V and Mx , cf.
Lemma 5.6, and argue exactly as in the second last paragraph of the proof of Theorem
12.20 to deduce that there is hence a play of ,x (G (A)) in Mx in which I I follows

,x ( ) and in which I I looses. But this is a contradiction!
It is not hard to show that if there is a measurable cardinal, then every set of
reals has an embedding normal form, cf. Problem 13.1. It is much harder to get
embedding normal forms which are sufficiently closed. The proof of the following
result is similar to the proof of Theorem 12.20.
Theorem 13.3 Let be a measurable cardinal, and let A be coanalytic. Then
A has a closed embedding normal form.
Proof Recall again that a set A is coanalytic iff there is some map s  <s ,
where s < , such that for all s, t < with s t, <t is an order on lh(t) which
extends <s , and for all x ,
x A <x =


sx

<s is a wellordering.

(13.3)

306

13 Projective Determinacy

(Cf. Lemma 7.8 and Problem 7.7.)


Let s  t, where lh(t) = lh(s) + 1. Write n = lh(s). Suppose that n is the kth
element of n + 1 = {0, . . . , n} according to <t , i.e.,
m 0 <t . . . <t m k1 <t n <t m k <t . . . <t m n1 ,
where m l is the lth element of n = {0, . . . , n 1} according to <s , l < n. We then
define (s, t): n n + 1 by (s, t)(l) = l for l < k and (s, t)(l) = l + 1 for l k,
l < n.
If s  t, where lh(t) = lh(s) + m, then we define (s, t): lh(s) lh(t) by
(s, t) = (t  lh(t) 1, t) . . . (s, t  lh(s) + 1).
The map (s, t) then tells us how the lh(s), lh(s) + 1, . . . , lh(t) 1 sit inside
0, 1, . . . , lh(t) 1 according to <t .
Let us now define an embedding normal form (Ms , st : s t < ) for A as
follows. Let U be a measure on , and let
(M , : OR)
be the (linear) iteration of V = M0 given by U . Let us write U = 0 (U ) and
= 0 (), where OR. We set Ms = Mlh(s) and
(s,t)

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)

lh(t), lh(s),lh(t) = lh(s), .

13.1 Embedding Normal Forms

307

Therefore, we may define an elementary embedding k: Mx M by setting


(s,x)

k(sx (y)) = lh(s), (y)


for y Ms .
As Mn Mn for all n < by Lemma 4.63, we have thus shown that A has a
closed embedding normal form.

Theorems 13.2 and 13.3 reprove Martins Theorem 12.20.
We shall now turn towards proving Projective Determinacy.
Definition 13.4 Projective Determinacy, PD, is the statement that all projective subsets of are determined.
The key new ingredients to show that Projective Determinacy holds true are iteration
trees which are produced by Woodin cardinals.
Definition 13.5 Let A , and let
E = (Ms , st : s t < )
be an embedding normal form for A. If is an ordinal, then we say that the additivity
of E is bigger than iff st  ( + 1) = id for all s, t < , s  t.

13.2 The MartinSteel Theorem


The following seminal result was produced in [26].
Theorem 13.6 (D. A. Martin, J. Steel) Let be a Woodin cardinal, let B ,
and suppose that B has a 20 closed embedding normal form whose additivity is
bigger than . Then for every < ,
/ B}
{x : y x y
has a 20 closed embedding normal form whose additivity is bigger than .
Theorems 13.2 and 13.3 immediately give the following.
Corollary 13.7 Let n < , and suppose that there is a measurable cardinal above
n Woodin cardinals. Then every 1 subset of is determined. In particular, if
n+1

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

lh(s) = lh(t), s,t : OR be defined by

(13.6)
< ,

13.2 The MartinSteel Theorem


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)

n . We define T by setting (s, t, f ) T


Let be an ordinal which is bigger than all x,y
<
iff s, t , lh(s) = lh(t), f = ( f i : i < lh(s)), where each f i is a function from
to , and for all i + 1 < lh(s) and for all x, y , if (x, y)
/ As 
i + 1 = x  i + 1 t  i + 1 = y  i + 1, then f i+1 (x, y) < f i (x, y), and if
(x, y) A s  i + 1 = x  i + 1 t  i + 1 = y  i + 1, then f i+1 (x, y) = 0.
The order on T is again reverse inclusion.
Now if ( f i : i < ) witnesses that (x, y) p[T ], then we cannot have that
(x, y)
/ A, as otherwise f i+1 (x, y) < f i (x, y) for each i < . Hence p[T ] A.
On the other hand, if (x, y) A, then we may define a witness ( f i : i < ) to the
/ A, then let k < be
fact that (x, y) p[T ] as follows: Let x , y . If (x , y )
maximal with x  k = x  k y  k = y  k, and define f i (x , y ) = k + 1 i for
i k + 1 and f i (x , y ) = 0 for i k + 1. If (x , y ) A, then define f i (x , y ) = 0
for all i. This shows A p[T ], and hence A = p[T ].
Notice that, as each Ns,t is 20 closed, we have that s,t Ns,t for every s, t. In
fact, for all s, t < , lh(s) = lh(t),

((s, t, (s i,t i),(s,t) (s i,t i )) : i < lh(s)) (,),(s,t) (T )

(13.8)

To see this, notice that if x, y


/ A, s, t < , lh(s) = lh(t), x  lh(s) = s and
y  lh(s) = t, and i < j < lh(s), then by (13.6)
i )
(s i,t i),(s,t) (s i,t i )(x, y) = (s i,t i),(s,t) (x,y
j
> (s  j,t  j),(s,t) (x,y )
= (s  j,t  j),(s,t) (s  j,t  j )(x, y),

and if (x, y) A x  lh(s) = s y  lh(s) = t, then (s i,t i),(s,t) (s i,t i )(x, y) =


0.
Let us now fix < . We aim to construct a 20 closed embedding normal form
for {x: y(x, y)
/ A} whose additivity is bigger than . The embedding normal form
will be produced by iteration trees on V .
Let (sn : n < ) be an enumeration of < such that if sn  sm , then n < m. Let
 be the following order on : we set n  m iff n = 0, or n, m are both even and
n m, or n, m are both odd and
s n+1 s m+1 .
2

in the tree order .1


We intend to have sn correspond to the node 2n 1
<
For s , we shall produce objects

to denote k l, unless l > k in which case k l


= 0.
We here and in what follows use k l

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)),

such that for each k < 2 lh(s),


Ms,k |= E s,k is a 20 -closed and certified extender with
critical point s,i > and E s,k V ,
where

2m
i=
k

if k is even, say k = 2n and sm = sn+1  (lh(sn+1 ) 1), and


if k is odd.

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

with crit s,2n+1

Es,2n

with crit s,2m

k = 2n

sm 2m 1

(PD, 2) If s s , then Ts endextends Ts , i.e., Ms ,k = Ms,k for all k 2 lh(s)


and E s ,k = E s,k for all k < 2 lh(s) (of course, the latter implies the
former); also, s ,k = s,k for k 2 lh(s), s ,k = s,k for k lh(s), and
s ,k = s,k for k lh(s).

13.2 The MartinSteel Theorem

311

(PD, 3) Say n = lh(s). Then


 2n 1))

(s  lh(sn ), sn , (s,2l 1,2n


s,0,2n 1

(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)).

We will have that

s ,t
= (,),(s ,t ) (Ms,k ),
Ms,k
s ,t

s,k,l = (,),(s ,t ) (s,k,l ), and


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 .

Notice that we shall have


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)).

3 For a proper class X , we write


{(,),(s ,t ) (X V ) : OR}. Cf. the
(,),(s ,t ) (X ) =
s ,t
footnote on p. 184. As a matter of fact, in what follows we shall only need Ms,k
in case s and s
are compatible.

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,k,l )((s ,t ),(s ,t ) (x))


s ,t
(s ,t ),(s ,t ) (x).
= s,k,l

Our forth condition now runs as follows.


(PD, 4) If m < n < lh(s) and sm = sn+1  (lh(sn+1 ) 1), then
s lh(s

),sn+1

n+1
s,n+1 < s,2m,2n+2

((s lh(sm ),sm ),(s lh(sn+1 ),sn+1 ) (s,m ).

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

13.2 The MartinSteel Theorem

313

(N , ((s,t), : s x, t y)) = dir lim(Ns,t , (s,t),(s ,t ) : s s x, t t y),


(13.13)
so that N is transitive by (13.5).
For any s < , the sequence
(E s,k : k < 2 lh(s))
is easily seen to generate (Ts ) N such that
N |= (Ts ) N is an iteration tree on N of length 2 lh(s) + 1,
where we may write
x,y

x,y

(Ts ) N = ((Ms,k , s,k,l : k  l 2 lh(s)), (E s,k : k < 2 lh(s)),  (2 lh(s) + 1)).


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

(Mx, , (x,2k, : k < )) = dir lim(Mx,2k , x,2k,2l : k l < ).

(13.14)

Virtually the same proof as the one of (13.10) shows that


x k,y k

x,y

x k,y k

(x k,y k), x,2n, = x,2n, ((x k,y k),  Mx,2n

(13.15)

for all n, k < .


Again by (PD, 2) we also write x,k for x k,k (= x k ,k for all k k). By (PD,
4), if k m n < , sm = y  k and sn+1 = y  (k + 1), then
x (k+1),y (k+1)

x,n+1 < x,2m,2n+2

((x k,y k),(x (k+1)),y (k+1)) (x,m )).

which implies that


x (k+1),y (k+1)

x,2n+2,

x (k+1),y (k+1)

(x,n+1 ) < x,2m,

((x k,y k),(x (k+1)),y (k+1)) (x,m )).


(13.16)

Then (13.15) and (13.16) yield the following.


x,y

x,2n+2, ((x (k+1),y (k+1)), (x,n+1 ))


x (k+1),y (k+1)
(x,n+1 ))
= (x (k+1),y (k+1)), (x,2n+2,
x (k+1),y (k+1)
((x k,y k),(x (k+1)),y (k+1)) (x,m )))
< (x (k+1),y (k+1)), (x,2m,
x,y
= x,2m, ((x k,y k), (x,m )).

314

13 Projective Determinacy

This shows that the sequence


x,y

(x,2m, ((x lh(sm ),y lh(sm )), (x,m )) : m < , sm y)


x,y

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)||

i = ||(x n i ,2l 1,2n


(x n i ,l ): 2l 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

the node (x n i+1 2l 1,2n

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)||

= ||(x n i+1 ,2l 1,2n


(x n i+1 ,l ) : 2l 1

xni+1 ,0,2ni+1 1
(Tx,y )
i+1 1

> ||(x n i+1 ,2l 1,2n


(x n i+1 ,l ) : 2l 1  2n i+1 1)||xn ,0,2n 1

(Tx,y )
i+1 1
i+1
i+1
= i+1 .
This proves that M y is illfounded, as witnessed by (i : i < ).

13.2 The MartinSteel Theorem

315

We have verified (13.12).

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

(PD, 5) For each l < , l V

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

with crit s,2n+1

Es,2n

t  k 1 = sm 2m 1

k = 2n

As s
and write

<

with crit s,2m

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

(Vm +1 ; , V2m , {, , 0,2m


((,),(s k1,t k1) (T )),
type M2m
s k1,t k1
((s i,t i),(s k1,t k1) (s i,t i )): i k 1)})
(0,2m
.
M2m 1
. (T ),
(Vc0 +1 ; , V2m , {, , 0,2m 1
= type
.
.
.
(2i 1,2m

1 (i ): 2i 1  2m 1)}).
s k1,t k1

(B) Inside M2m

sk1,tk1

, 2m is strong up to with respect to


s k1,t k1

(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

(C) inside M2n , 2n+1 is strong up to with respect to

13.2 The MartinSteel Theorem

317

sk,t

type M2n (V 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

inside M2n , 2m is strong up to with respect to


sk,t

type M2n (V sk,tk (

(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

= type M2n (V sk,tk (


2m,2n

(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)})),

restricted to parameters in (V2n+1 +1 ) M2n+1 , which in turn by (A) is equal to


.

.
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

= type M2n (V sk,tk (


2m,2n

(sk1,tk1),(sk,t) (m )

; , V2n+1 , {, ,

s k,t

0,2n ((,),(s k,t) (T )),

(13.21)

s k,t
(0,2n ((s i,t i),(s k,t) (s i,t i )): i

k)}).

s k,t

By (C), we have that = 0,2n (s k,t ) witnesses that


s k,t

M2n

sk,t

|= V ( = type M2n (V sk,tk (


2m,2n

(sk1,tk1),(sk,t) (m )

; , V2n+1 ,

s k,t

{, , 0,2n ((,),(s k,t) (T )),


s k,t

(0,2n ((s i,t i),(s k1,t k1) (s i,t i )): i k 1) })


2n+1 is strong up to with respect to
type

sk,t
M2n

(V sk,tk (
2m,2n

(sk1,tk1),(sk,t) (m )

(13.22)
; , V ,

s k,t

{, , 0,2n ((,),(s k,t) (T )),


s k,t

(0,2n ((s i,t i),(s k1,t k1) (s i,t i )): i k 1) }) ).


As

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),

13.2 The MartinSteel Theorem

319

M2n+1 |= V ( = type M2n+1 (Vc0 ; , V2n+1 ,

(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

be a witness to (13.23), so that we shall now have that

(D)
s k,t

= type M2n (V s k,t k (


2m,2n

(s k1,t k1),(s k,t) (m )

s k,t

; , V2n+1 , {, , 0,2n ((,),(s k,t) (T )),

s k,t

(0,2n ((s i,t i),(s k,t) (s i,t i )): i k)})


= type M2n+1 (Vc0 ; , V2n+1 , {, , 0,2n+1 (T ),
.

(i ): 2i 1  2n + 1)})
(2i 1,2n+1

and
(C)

inside M2n+1 , 2n+1 is strong up to with respect to


type M2n+1 (Vc0 ; , V , {, , 0,2n+1 (T ),
.
(i ): 2i 1  2n + 1)}).
(2i 1,2n+1

M2n+1

Also, if we inductively assume (PD, 5) for l n, then n+1 V


for l = n + 1) yields that
.

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)}),

cf. Lemma 10.81.


Let
2n,2n+2 : M2n E s,2n+1 M2n+2 ,
where E s,2n+1 M2n+1 is a 20 closed certified extender which in M2n+1 and
witnesses that inside M2n+1 , 2n+1 is strong up to the least inaccessible cardinal
above 2n+2 with respect to
.

.
(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

type M2n+2 (V sk,t


s k,t

2m,2n+2 (sk1,tk1),(sk,t) (m )

; , V2n+2 +1 ,

{, , 0,2n+2 ((,),(s k,t) (T )),

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

This is because the left hand side of (F) is equal to


s k,t

sk,t

2n,2n+2 (type M2n (V sk,t

2m,2n (sk1,tk1),(sk,t) (m )

s k,t

; , V2n+1 ,

{, , 0,2n ((,),(s k,t) (T )),


s k,t
0,2n ((s i,t i),(s k,t) ((s i,t i)) : i k)})

sk,t

restricted to parameters in (V2n+2 +1 ) M2n+2


= E 2n+1 (type M2n+1 (Vc0 ; , V2n+1 , {, , 0,2n+1 (T ),
.
.
(i ) : 2i 1  2n + 1)})),
(2i 1,2n+1
restricted to parameters in (V2n+2 +1 ) M2n+1
by (D),
.

.
= 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

2n+2 is strong up to with respect to


type

M2n+1

(V+1 ; , V , {, , 0,2n+1 (T ),
.

.
(i ): 2i 1  2n + 1)}). ).
(2i 1,2n+1

(13.25)

13.2 The MartinSteel Theorem

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

( = type M2n+2 (V+1 ; , V2n+2 , {, , 0,2n+2 ((,),(s k,t) (T )),


s k,t

0,2n+2 ((s i,t i),(s k,t) ((s i,t i) : i k))})


s,2n+2 is strong up to with respect to
type

sk,t
M2n+2

(13.26)

s k,t

(V+1 ; , V , {, , 0,2n+2 ((,),(s k,t) (T )),

s k,t

0,2n+2 ((s i,t i),(s k,t) ((s i,t i) : i k))} ) ).


Let n+1 be a witness to this fact. In particular, with a brief show of the subscript s,
s k,t

s,n+1 < 2m,2n+2 ((s k1,t k1),(s k,t) (s,m )).

(13.27)

By (13.26) and the definition of , we now have the following.


(G)
sk,t

s k,t

type M2n+2 (Vn+1 +1 ; , V2n+2 , {, , 0,2n+2 ((,),(s k,t) (T )),


s k,t
(0,2n+2 ((s i,t i),(s k,t) (s i,t i )): i k)})
= type M2n+1 (Vc0 +1 ; , V2n+2 , {, , 0,2n+1 (T ),
.
.
(i ): 2i 1  2n + 1)}),
(2i 1,2n+1
and
s k,t

(H) inside M2n+2 , 2n+2 is strong up to with respect to


sk,t

s k,t

type M2n+2 (Vn+1 +1 ; , V , {, , 0,2n+2 ((,),(s k,t) (T )),


s k,t
(0,2n+2 ((s i,t i),(s k,t) (s i,t i )): i k)}).
We are back to where we started from, cf. (A) and (B).
It is now straightforward to verify that (PD, 1) through (PD, 4) hold true. Notice
that (13.27) above gives (PD, 4). Also, (PD, 3) follows from (A) (or, (G)) by virtue
of (13.8).
This finishes the proof of Theorem 13.6.

It can be shown that the conclusion of Thorem 13.6 implies the consistency of
Woodin cardinals, cf. e.g. [22]; in fact, PD turns out to be equivalent to the existence
of mice with Woodin cardinals, for a proof cf. [36].

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.

and let 0 . Show that the following are

(a) A has a 20 closed embedding normal form whose additivity is bigger


than .
(b) A is homogeneously Souslin.
Conclude that every homogeneously Souslin set of reals is determined. [Hint.
For (a) = (b), construct the Windus tree and define s , s < , via
(13.2).]
Let A , and let 0 . We say that A is weakly homogeneously
Souslin iff A = {x : y x y B}, where B is homogeneously

13.3 Problems

323

Souslin, and A is weakly homogeneously Souslin iff A is 0 weakly homogeneously Souslin.


13.4. (D.A. Martin, R. Solovay) Show that if A is weakly homogeneously Souslin, then A is < + universally Baire. [Hint. Let (sn : n < )
be a reasonable enumeration of < . For s < with k = lh(s), let
(s, (0 , . . . , k1 )) S iff for all i < j < k, if si  s j , then
j < slh(si )si ,slh(s j )s j (i ),
where 0 , . . . , k1 < for some sufficiently big .]
Conclude that if is a measurable cardinal, then all 1 sets of reals are <
2

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

R. Schindler, Set Theory, Universitext, DOI: 10.1007/978-3-319-06725-4,


Springer International Publishing Switzerland 2014

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

-universally Baire, 149


1 , 135
n

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

universally Baire, 149


weakly homogeneously Souslin, 324
Shift map, 186, 198

Shoenfield tree, 131


-algebra, 128
Silver indiscernibles for L[x], 209
Singular Cardinal Hypothesis, 39
Skolem function
1 , 193
Solid, 258
1-, 258
n-, 258
Solidity witness, 256
Solovay game, 289
Solovay sequence, 301
length of, 302
Sound, 253
n-, 253
 , 269
 (R), 269
Standard code, 237
nrmth , 250
Standard parameter, 255
nth, 254
Standard reduct
nth, 254
Standard witness, 257
Stationary, 41
Stationary set
reflecting, 278
Steel forcing, 295
Stem
of a Prikry condition, 188
Strategy, 279
Subset, 9
proper, 9
Support, 105, 110
Surjective, 14
Symmetric difference, 87
T
TC({x}), 27
||R||, 28
Transitive, 23
Transitive closure, 27
Transitive collapse, 28
Tree, 46
, 47
Aronszajn, 47
-Kurepa, 47
-Souslin, 47
on , 128
on X , 127
perfect, 127
Turing reducible, 288
2< , 37

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