57 views

Uploaded by ozgan

ISBN: 3319067249

- Introduction to Axiomatic Set Theory (1971) - Krivine
- Judith Roitman - Introduction to Modern Set Theory - Wiley
- Handbook of Philosophical Logic Second Edition 11
- ContinuumHypothesis
- Modal Logic for Philosophers ---- James W. Garson
- Hrbacek-Introduction to Set Theory PDF
- Handbook of Philosophical Logic Second Edition 10
- Geoffrey Hunter Metalogic an Introduction to the Metatheory of Standard First Order Logic 1996 (1)
- Leila Haaparanta, Jaakko Hintikka (Auth.), Leila Haaparanta, Jaakko Hintikka (Eds.) Frege Synthesized_ Essays on the Philosophical and Foundational Work of Gottlob Frege 1986
- Drake, Set Theory
- Krzysztof Ciesielski - Set Theory for the Working Mathematician
- Bertrand Russell and the Origins of the Set-Theoretic 'Paradoxes'. Alejandro R. Garciadiego
- Hermann Weyl - The Continuum: A Critical Examination
- Epistemic Logic by Rescher
- Modal Logics and Philosophy (Girle)
- Paradoxes
- A Course on Set Theory
- Dauben Georg Cantor
- Kleene - Introduction to Metamathematics
- Elements of Set Theory

You are on page 1of 335

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

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.

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)

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

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

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

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

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

11.1 Fine Structure Theory . . . . . .

11.2 Jensens Covering Lemma . . .

11.3 hj and Its Failure . . . . . . . .

11.4 Problems . . . . . . . . . . . . . . .

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

235

235

262

269

275

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

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}.

Springer International Publishing Switzerland 2014

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.

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.

h(x) =

f (x)

g 1 (x)

if x X 0 X 1 , and

if x X 2 .

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 ;

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.

(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

nN

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

injective.

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}

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.

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

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 ].

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

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

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 = .

(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

2

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, [a, b] is dense in [a, b], and [a, b] is perfect, nowhere dense, and a

null set.

Chapter 2

Ernst Zermelo (18711953) was the first to find an axiomatization of set theory,

and it was later expanded by Abraham Fraenkel (18911965).

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)

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)

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 = )).

Springer International Publishing Switzerland 2014

10

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)

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).

11

We then say:

x(x is inductive).

(Inf)

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 )

(Aus ) then says that

ab b = {x a: }.

Writing z = x y for

u(u z u x u y),

u(u z u x u y),

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

y for

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

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)

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).

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}.

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

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

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.

15

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

= (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

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),

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.

17

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

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 )}.

= and thus

0)

) = g(u 0 ).

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

0)

Contradiction!

We have shown that if , W , then or .

18

But now

Setting

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)

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).

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)

(Rep )

19

and for all such that is a formula of the language of BG, which contains exactly

x, X 1 , . . . , X k (but not Y ) as its free variables and which does not have quantifiers

ranging over classes (in other words, results from a formula of the language of

ZF by replacing free occurences of set variables by class variables), then

X 1 . . . X k Y x(x Y ).

(Comp )

(Comp ) is called the comprehension axiom for , and the collection of all (Comp )

is called the comprehension schema. The BernaysGdel class theory with choice,

BGC, is the theory BG plus the following version of the axiom of choice:

There is a (class) function F such that x(x = F(x) x).

(AC)

It can be shown that ZFC and BGC prove the same theorems in their common

language L (i.e., BGC is conservative over ZFC).

If is a formula as in (Comp ), then we shall write {x: } for the class given

by (Comp ). (Rep ) says that for all class functions F and for all sets a, F a =

{y: x (x, y) F} is a set.

We shall write V for the universe of all sets, i.e., for {x: x = x}. V cannot be a

set, because otherwise

R = {x V : x

/ x}

would be a set, and then R R iff R

/ R. This is another instantiation of Russells

antinomy.

If A is a class, then we write

A = {x: y A x y}

and

A = {x: y A x y}.

A and A always exist, and = and = V .

It may be shown that in contrast to ZFC, BGC can be finitely axiomatized. BGC

will be the theory used in this book.

The books [15, 18, 23] present introductions to axiomatic set theory.

2.3 Problems

2.1. Let k N. Show that there cannot be sets x1 , x2 , . . . , xk such that x1 x2

. . . xk x1 .

2.2. Show that for all x, y, (x, y) exists. Show that if (x, y) = (x , y ), then x = x

and y = y . Show that for all a, b, a b exists (cf. Lemma 2.1). Show that for

20

all d, b, d b exists (cf. Lemma 2.2). Which axioms of ZF do you need in each

case? Show that (Pair) may be derived from the rest of the axioms of ZF (from

which ones?).

2.3. Show that neither in (Aus ) nor in (Rep ), as formulated on p. 11, we could

have allowed b to occur freely in . Show that the separation schema (Aus)

can be derived from the rest of the axioms of ZF augmented by the statement

x x = .

2.4. Show that the following version of (AC) is simply false:

x(y x y = ) zy xu z y = {u}).

2.5. Show tht every partial order can be extended to a linear order. More precisely:

Let a be any set. Show that for any partial order on a there is a linear order

on a with .

2.6. Show that in the theory ZF, the following statements are equivalent.

(i) (AC).

(ii) For every x such that

y = for every y x there is a choice function,

i.e., some f : x x such that f (y) y for all y x.

2.7. (a) Let denote the natural ordering on N, and let m N, m 2. Let the

ordering m on N be defined as follows. n m n iff either n n (mod m)

and n n , or else if k < m, k N, is least such that n k(mod m) and

k < m, k N, is least such that n k (mod m), then k < k . Show that m

is a well-ordering on N.

(b) Let, for m N, m be any well-ordering of N, and let : N N N be

a bijection. Let us define on N by n n iff, letting (m, q) = (n) and

(m , q ) = (n ), m < m or else m = m and q m q . Show that is a

well-ordering of N.

2.8. (Cantor) Let (a, <) be a linear order. (a, <) is called dense iff for all x, y a

with x < y there is some z a with x < z < y. Show that if (a, <) is dense

(and a has more than one element), then < is not a well-ordering on a. (a, <)

is said to have no endpoints iff for all x a there are y, z a with y < x < z.

Let (a, <a ) and (b, <b ) be two dense linear orders with no endpoints such

that both a and b are countable. Show that (a, <a ) is isomorphic to (b, <b ).

[Hint. Write a = {xn : n N} and b = {yn : n N}, and construct f : a b

by recursively choosing f (x0 ), f 1 (y0 ), f (x1 ), f 1 (y1 ), etc.]

2.9. Show that there is a set A of pairwise non-isomorphic linear orders on N such

that A R.

2.10. Show that every axiom of ZFC is provable in BGC.

Let us introduce Ackermanns set theory, AST. The language of AST arises

The axioms of AST are (Ext),

from L by adding a single constant, say v.

(Fund), (Aus), as well as (Str) and (Refl) which are formulated as follows.

2.3 Problems

21

x v y ((y x y x) y v).

(Str)

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 )

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

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 = .

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.

Springer International Publishing Switzerland 2014

23

24

3 Ordinals

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.

25

(a)

(b)

(c)

(d)

If is an ordinal and x , then x is an ordinal.

If , are ordinals, and , then .

If , are ordinals, then or (and hence = ).

(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

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

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)

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).

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

= (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

= (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

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 )

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

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.

Springer International Publishing Switzerland 2014

33

34

4 Cardinals

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

35

write [A] for

{x A: Card(x) =

then we

<

}, 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

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)

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().

37

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

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 =

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

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

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) = } .

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() > .

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 . 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 = .

concept of a stationary set.

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

41

{X : < \ X }

is a non-trivial < cf()-closed filter on , called the Frchet filter (on ). Every

filter can be extended to an ultrafilter, cf. Problem 4.11.

Lemma 4.25 Let be an ordinal such that cf() > , and let F be the set of all

A such that there is some B A which is club in . Then F is a (non-trivial)

< cf()-closed filter on .

Proof We need to see that if

< cf() and Ai is club in for every i , then

in

.

Well,

(

i Ai is club

i Ai ) {} is certainly closed, so that it suffices

to verify that i Ai is unbounded in . Let < . We define f : as

follows.3 For n and i we let f ( n + i) be the least element of Ai which

is bigger than sup({ f () : < n + i} { }). Notice that this is welldefined as

< cf() and each Ai is unbounded in .

Let = sup{ f () : < }. We have that < and in fact i Ai ,

because every Ai is closed; notice that for each i , = sup{ f ( n + i) : n

} = sup(Ai ).

F as in this lemma is called the club filter on .

Definition 4.26 Let be regular, and let X for all < . The diagonal inter

section of X , < , abbreviated by < X , is the set < : < X .

Definition

be regular, and let F be a filter on . F is called normal iff

4.27 Let

for all X : < F, < X F.

Lemma 4.28 Let be regular, and let F be the club filter on . Then F is normal.

Proof We need to see that if

A is club in for every < , then < A is

club in . By replacing A by A if necessary, we may and shall assume that

A A whenever . (Notice that every A is again club in by Lemma

4.25.) In order to see that ( < A ) {} is closed, let < be a limit ordinal such

that = sup(( < A ) ). We want to argue that < A , i.e., A for all

< . Well,

if < , then there is some with < < and < A ,

i.e., < A , in particular A . This shows that = sup(A ) for all

< ; hence A for all < .

In order to see that < A is unbounded in , let < . We construct a

sequence n , n , as follows. Let 0 = . If n is defined, then let n+1 be the

least > n such that An . We claim that sup{n : n } < A . Set

= sup{n : n }. We need to see that A for all < . Let < . Then

< n for some n . We have that m+1 Am An for all m n, so that

= sup{m+1 : m n} A xn A .

Definition 4.29 Let be an ordinal such that cf() > . A is called stationary

(in ) iff A C = for all C which are club in .

3

42

4 Cardinals

{ < : 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)

some < X such that f ( ) < . This contradicts (4.3).

/ F, then

(b) = (a): Let X F, < . If < X

X F +.

Y = < :

/

<

/ 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.

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)

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

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 < ,

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

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.

< as follows.

Fix < , and suppose that S and have already been chosen for all < .

The set

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 ,

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 < ,

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 < ,

< 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

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 .

47

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

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.

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

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.

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

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

T+1 = {t (s) : s T } .

It is easy to verify that (1) through (4) remain true.

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)

(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 () > .

51

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

( 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() = () =

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)

/ 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

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 ]).

55

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.

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

. . . 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

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)

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)

X U (X ).

(4.8)

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)

57

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)

} 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)

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

(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)

X,

Xa.

(4.13)

a[X ]n

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 ).

59

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

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)

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

Theorem 4.66 (Solovay) Let be supercompact. Then < = for every regular

cardinal .

Proof Let us fix , let again

: V M

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

61

Then f is a regressive function, and because U is weakly normal there is

for S.

some > , < , such that

S : f () < U.

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

(4.15)

such that 0 = 0 and

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

Let us write

(4.16)

I = < : C [ , +1 ) =

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)

We have shown that

P(I ),

[]<

S

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

Proof Let A H + , and let

: V M

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

2 M

M. We have that

crit( ) = ).

By the elementarity of ,

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

= ).

We have shown that is subcompact.

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

4.9. 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

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.

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

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 ).

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 .

69

. 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 (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

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.

(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 ,

71

(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

72

5 Constructibility

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 .

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 ,

(5.1)

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

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)

prove that

(U {U };,E)

= P(U ) 0

, so that we have to

(U {U };,E)

/ 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

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

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

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

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] =

<

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 )}.

77

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

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.

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

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

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

/ 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

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

80

5 Constructibility

(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.

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.

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.

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

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 )

83

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

(5.4)

{R , B , C } J+ [ E].

= E J+ [ E],

and hence

As E satisfies local condensation, E J+ [ E]

= J+ [E]. This gives that (B , C ) is the < E -least (and thus

J+ [ E]

+

also the <E -least) pair such that B , C is club in , and (5.4) holds true.

Therefore, (B , C ) = (B , C ) by the choice of (B , C ). Therefore,

{ R: B = B } C,

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 .

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

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 }.

for every C R.

84

5 Constructibility

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.,

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)

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)

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

, cf.

Problem 5.22.

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.

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.

= (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<

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

90

5 Constructibility

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 +

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

J |= (Aus ).

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

= (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).

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

( 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.

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)}.

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.

95

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

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

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.

97

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

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:

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)

{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)

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 )}

101

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

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.

103

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 ).

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

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):

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)

(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 .

105

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)

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

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

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

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

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

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

Working in M, for each

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)

<

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 = ,

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.

109

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 }

<

X.

110

6 Forcing

is easily

seen to be open dense. As < D is open dense, we may pick q

G < D . As q|| p, this means that for every < , there is a (unique) y x

with q PM ( ) = y . Setting

g = {(, y) x : q PM ( ) = y },

Therefore, f = G = g M.

we then get q PM = g.

Problem 6.10 shows that the converse to Lemma 6.38 is true also. The following

just generalizes Definition 6.27.

Definition 6.39 Let be an ordinal, and let be an infinite regular cardinal. For

p (C ) let supp( p) (the support of p) be the set of all < with p( ) = . Let

C () = { p (C ) : Card(supp( p)) < }.

For p, q C (), let us write p q iff for all < , p( ) q( ) in the sense of

C , i.e., p( ) = q( ).

Lemma 6.40 Let M be a transitive model of ZFC, let be a regular cardinal of

M, let be an ordinal in M, and let (C ()) M be Ms version of C (). Let G be

(C ()) M -generic over M. Then exactly the cardinals of M which are not in the

half-open interval (, ( < ) M ] remain cardinals of M[G].

Proof Of course, C () is < -closed. The -Lemma 6.31 implies that M |=

(C ()) M has the (( < )+ ) M -c.c., so that no M-cardinals outside the half-open

interval (, ( < ) M ] will get collapsed. On the other hand, Problem 6.8 shows that

all the M-cardinals inside the half-open interval (, ( < ) M ] will get collapsed

to .

If < = in M, (C ()) M will therefore not collapse any M-cardinals. Moreover, we may have 2 = in a forcing extension (cf. Problem 6.9).

We shall now study forcings which collapse cardinals.

Definition 6.41 Let be a regular cardinal, and let . We let Col(, ) = < ,

i.e., the set of all functions f such that there is some < with f : . For

p, q Col(, ), let p q iff p q (iff p = q). The partial order

(Col(, ); ) is called the collapse of to .

Notice that Col(, ) = C .

Lemma 6.42 Let M be a transitive model of ZFC, let be a regular cardinal of M,

let be a cardinal of M, and let P = (Col(, )) M be Ms version of Col(, ).

Let G be P-generic over M. Then every M-cardinal is still a cardinal in M[G],

and in M[G], Card() = = 2< . Moreover, cardinals above (( < )+ ) M remain

cardinals in M[G].

111

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

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

by setting

G)( + ) = 1} = X.

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(, )}.

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 .

112

6 Forcing

Lemma 6.44 Let be a regular cardinal, and let > be a regular cardinal such

that < for all < and < . Then Col(, < ) has the -c.c.

Proof This immediately follows from the -Lemma 6.31.

As Col(, < ) is certainly < -closed, this immediately implies the following.

Lemma 6.45 Let M be a transitive model of ZFC, and let < be regular

cardinals of M such that inside M, < for all < and < . Let

P = (Col(, < )) M be Ms version of the Levy collapse of to . Then all

M-cardinals strictly between and will have size in M[G], and all M-cardinals

outside of the open interval (, ) will remain cardinals in M[G]. In particular,

= + in M[G].

The Levy collapse Col(, < ) will play a crucial role in Chap. 8.

We may force which was shown to be true in L, cf. Definition 5.34 and Theorem

5.35.

Theorem 6.46 Let M be a transitive model of ZFC, let be an uncountable regular

cardinal in M, and let P M be defined inside M as follows. P = {(c : ) : <

c }, ordered by end-extension. Let G be P-generic over M. Let

S , S M, be stationary in M. Then (S) holds true in M[G].

Proof

An easy density argument shows

that there are C , < , such that

G = (C : < ). We claim that G S = (C : S) witnesses that (S)

holds true in M[G].

Let , M P and p G be such that

p is club in ,

and .

Let p0 p be arbitrary. It suffices to show that there is some q p0 such that if

q = (c : ), then

q S = c .

(6.9)

Let us work inside M. Notice that P is < -closed. We may thus easily construct a

sequence ( pi : 1 i < ) of conditions in P such that there are F = (c : < ) and

(i : i < ), such that for all 0 i < j < , pi = (c : i ), supk< j k < j

(in particular, p j < pi ), and there is some j \ supk< j k and some a j M

such that, writing j for supk< j k ,

p j j = aj .

As S is stationary and {i : i < } is club in , we may pick some limit ordinal

i 0 such that i0 S. Set

A=

113

ai and q = (

i<i 0

pi ) {(i0 , A)}.

i<i 0

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

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

r : is surjective.

(6.10)

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)

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).

= 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

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

(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

(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

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

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].

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

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

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.

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

121

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

(6.15)

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)

, 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 <

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

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

J [(

G)

) = 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)) .

), ( 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

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,

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

{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

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

(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.

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<

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

is bijective. Let be such that (, ) =

Show that

G = { < : A ( )}

is V-generic over HOD and A HOD[G].]

126

6 Forcing

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.

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 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}.

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

/ 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)

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.

coanalytic).

129

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

(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 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 .

x n

u n

C i,k

j,l separates A y n , Bv n ,

i, j k,l

i.e.,

x n

A y n

u n

C i,k

j,l \Bv n .

i, j k,l

(x n i, 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

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

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

. 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

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

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

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

1

n+1

n+1

set.

Definition 7.10 A set A ( )k , some k < , is called projective iff there is some

n < such that A is 1 .

n

133

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

1

which cannot be 1 .

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

(7.4)

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

1

1

A = {x : (x, z) B}.

(7.5)

(This uses (7.4).) A set B with these properties is called a universal 1 -set.

1

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

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

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.

135

This gives that

k<

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

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

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

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

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

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

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) .

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

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

(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

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 ].

139

(T, ) L[T ]

(7.9)

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

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) =

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

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))

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

1

1

1

1

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

1

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

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

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

1

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 .

143

We have

A = {x : y y = F(x)}.

As A is uncountable, (the graph of) F is an uncountable 11 subset of ( )2 .

Suppose that F has a perfect subset, say P F, where P is perfect. Write

Q = {x : y (x, y) P}.

As F is a function, Card(Q) = Card(P) = 20 , so that Q is an uncountable analytic

subset of A. Contradiction!

Hence F is an uncountable coanalytic set without a perfect subset.

Finally, (3) = (1) is given by Corollary 7.21. Let A be an uncountable

1

2 set. If A does not have a perfect subset, then A L by Corollary 7.21. However,

1L < 1 implies that A is then countable. Contradiction!

There is a more cumbersome argument of proving (3) = (1) of Theorem 7.28,

using forcing, cf. the proof of Lemma 8.18.

Excellent textbooks on classical descriptive set theory are [27] and [20]. Modern

variants of descriptive set theory are dealt with e.g. in [4, 12, 14, 19, 21].

7.3 Problems

7.1. Show that the topology we defined on the Baire space is exactly the one

which is induced by the distance function d. Conclude that is a Polish

space.

7.2. Show that there is a continuous bijection f : 2.

7.3. (Borel hierarchy) Let 0 denote the set of all open A and 0 the set

1

the set of all n< An , where {An : n < } < 0 , and let 0 be the

set of all \A, where A 0 . Show that 0

1 +1

= 0 , and that 0 is

1

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

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

<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

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

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

Lebesgue measurable and has 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

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

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

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

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

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

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.

( M\A) =( N )\A .

( n< An ) = n< (An ) .

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 ]}

[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

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.

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.

153

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}.

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

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

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

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 ).

M such that by (8.1), gx

/ M, gx [T ], and gx A. By construction, the set

g

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)

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.

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.

157

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

= },

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 ]

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.

159

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)

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],

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)

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

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)

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)

/ 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

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.

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)

(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 .

163

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

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

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||).

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

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.

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)

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

3

Springer International Publishing Switzerland 2014

165

166

Proof Let us fix a such that 1V = 1L[a] , cf. the proof of Lemma 7.27. We

have L[a] = 2 L[a] = 1 . Let us write X = 2 L[a].

If x, y 2, x = y, let us write h(x, y) for the distance of x and y, i.e., h(x, y)

is the least n < such that x n = y n (Hence h(x, y) > 0). For Y 2, let us

write H (Y ) for

{h(x, y): x, y Y x = y};

H (Y ) is thus a set of positive integers.

Definition 9.4 We define F

X P() by setting a FX iff there is a covering

(X n :n < ) of X , i.e., X n< X n , where X n 2 for each n < , such that

H (X n ) a.

n<

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<

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 .

167

Proof

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

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.

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

2

Set

{x: (x, y) A}

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

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

168

(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

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

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

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

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 =

k=1 ((C k1 ) (C k ))

1

2

1

1

(C0 ) k=1 4k 3 3 = 3 > 0.

169

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)

(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

G=

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

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

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

171

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 ,

(9.5)

Let again m < and s < 2 be such that e(m, s) < n. Notice that if x, x , y

X, x f (n) = x f (n), and h(x, y) < f (n), then h(x , y) = h(x, y). This implies

that

(Card({t

f (n)

(9.6)

{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

G t,n,m

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

n m pn

Card({t

f (n)

(9.7)

f (n)

2: C Us G t,n,m = .

We have that

C Us

As (G t,n,m ) =

Claim 9.8,

1

,

2n+m

2\G t,n,m : t

f (n)

2, C Us G t,n,m = .

(C Us )

k

q

q

q

1

1

n+m

= 1 n+m

.

k

2

2

k=0

1

1

1

By the choice of C, (C Us ) 23lh(s)+1

, so that 23lh(s)+1

1 2n+m

. By (9.4)

and the properties of e, we have that m < n and lh(s) < n and therefore

1

23n+1

1

23lh(s)+1

1 q

1 2n

.

2

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.

monotone, and let f : be defined by

f (n) = g((n + 1) (3n + 4)2 24n+4 ).

173

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

as desired.

This finishes the proof of Theorem 9.3.

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) =

x(k)

otherwise .

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

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 }.

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

1

2n ,

(9.8)

Set

Fn = {s n 2: km s = smk }.

With the help of (9.8),

A {x 2: x n Fn for infinitely many n < }.

Also

inf m<

n=m

({x 2 : x n Fn

= 0.

(9.9)

175

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 < ,

and set

Jk = {s Ik 2 : i [m k+1 , n k+1 ]t Fi s [n k , i) = t [n k , i)}

and

Jk = {s Ik 2 : i [n k , m k+1 ]t Fi s [m k , i) = t [m k , i)}

We claim that both (I, J) and (I , J ) satisfy (1) and (2) of Definition 9.12.

As for (2),

x 2: x Ik Jk

= x 2: i [m k+1 , n k+1 ]t Fi x [n k , i) = t [n k , i)

n k+1

2n k

i=m k+1

x 2: t Fi x i = t

< k .

176

Symmetrically,

x 2: x Ik Jk

= x 2: i [n k , m k+1 ]t Fi x [m k , i) = t [m k , i)

m k+1

mk

x 2: t Fi x i = t

< k .

i=n k

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

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 < .

(k )2

8

F (I, J) (I , J ),

(9.12)

where (I, J) and (I , J ) are exactly as constructed in the proof of Lemma 9.13. We

are also going to use the notations n k , m k , Ik , Ik , Jk , Jk for k < from the proof of

Lemma 9.13; in particular, ({x : x Ik Jk }) < k and ({x : x Ik Jk }) < k

for every k < . We aim to find

(I , J ) F

(9.13)

such that for every k < , x : x Ik Jk < k .

For k < , let

Hk = t [n k ,m k+1 ) 2 : x 2 : t x [m k+1 , n k+1 ) Jk k .

177

({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)

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 }.

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

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

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) =

x0 (n) otherwise.

(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!

/ Jk , then

Say z N1 . Consider Ik+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,

,

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.

Proof Let n =

1

2n+1

F (I, J),

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

179

Jn = {s Jn : t

In

We claim that

F (I, J ).

(9.16)

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)

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

and

#(n) n + 1.

(9.18)

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

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

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

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)

: 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

: 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( ),

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

( )

=

( )

( ) =

( )

( ) =

Now let > 0.

/ ran(),

Let us first suppose to be a successor ordinal, say = + 1. If

us thus assume ran(), which implies that is a successor ordinal as well, say

= + 1, and () = . By (4.8), we have that

M = {, ( f )( ): f : M , f M }.

( ( f )( )) = ( f )( ),

where f M , f : M . This is well-defined because if is a formula and if

f 1 , . . . , f k M M , then

M |= ( ( f 1 )( ), . . . , f k ( ))

{ < : M |= ( f 1 ( ), . . . , f k ( ))} U

{ < : M |= ( ( f 1 )( ), . . . , ( f k )( ))} U , by using ,

M |= ( ( f 1 )( ), . . . , ( f k )( )).

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

. Let

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

188

10 Measurable Cardinals

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

Notice that each y M is of the form (x) for some < and x M .

Moreover, if (x) = (x ), where < , then

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

= ( ) ( ) (x ),

so that the definition of (y) is independent from the choice of < and x M

with y = (x). It is easy to verify the inductive hypothesis.

We now aim to make a measurable cardinal singular in a generic extension

without collapsing cardinals. The following definition is reminiscent of the definition

of Mathias forcing, cf. p. 176.

Definition 10.5 Let M be a transitive model of ZFC, and let , U M be such

that M |= U is a normal < -complete uniform ultrafilter on . We let PU =

P = (P; ) denote the following poset, called Prikry forcing. We let p P iff

p = (a, X ) where a []< and X U, min(X ) > max(a) (if a = ). We let

(b, Y ) (a, X ) iff b a, Y X , and b\a X (in particular, b is an end-extension

of a).

If p = (a, X ) P then a is called the stem of p. Notice that any two conditions with

the same stem are compatible, so that P has the + -c.c. in M. Hence no M-cardinal

strictly above will be collapsed by .

Assume G to be P-generic over M. Let

A=

{a: X (a, X ) G}.

It is clear that A hast order-type ; in fact, an easy density argument shows that

otp(A) = and A is cofinal in . In particular, will have cofinality in M[G].

We shall now prove that no M-cardinal will be collapsed in M[G]. As is

a limit cardinal, it suffices to prove that no M-cardinal < will be collapsed in

M[G]. For this in turn it is (more than) enough to prove the following.

Lemma 10.6 Let M be a transitive model of ZFC, let , U M be such that U

witnesses that is measurable in M, let P = PU , and let G be P-generic over M.

Then

(V ) M[G] = (V ) M .

Proof Let us first verify the following.

10.1 Iterations of V

189

Claim 10.7 (Prikry-Lemma) For all p P, for all formulae , and for all names 1 ,

. . ., k M PU there is some q p with the same stem as p deciding (1 , . . . , k ).

Proof This is an application of Rowbottoms Theorem 4.59. Fix p = (a, X )

and . (We shall supress the parameters 1 , . . ., k .) Let us define F: [X ]< 3 as

follows. Let b [X ]< . We set F(b) = 0 iff there is no X such that (a b, X )

(a, X ) and (a b, X ) decides ; otherwise we set F(b) = 1 (resp., 2) iff there is

some X such that (a b, X ) (a, X ) and (a b, X ) forces that holds (resp., does

not hold).

As f M, let Y U be given by Rowbottoms Theorem 4.59, i.e., for each

n < , F is constant on [Y ]< . We claim that (a, Y ) decides .

Well, if not, then there are (b1 , Y1 ) and (b2 , Y2 ) such that (b1 , Y1 ) (a, Y ),

(b2 , Y2 ) (a, Y ) and (b1 , Y1 ) and (b2 , Y2 ) . By extending one of these

two conditions if necessary we may assume that

Card(b1 ) = Card(b2 ) = Card(a) + n

for some n < . But then

F(b1 \a) = 1 = 2 = F(b2 \a),

although

b1 \a, b2 \a [Y ]n .

Contradiction!

2

In order to prove Lemma 10.6, it now suffices to show that if < , then

It suffices to find

M[G] M. Let f 2 M[G], f = G . Let p : 2.

some q p and some g M with q = g.

Let a be the stem of p. In virtue of the Prikry Lemma we may let for each <

is the stem of q and q ( ) = (h ). If

be q p and h 2 such that a

where g 2

q = (a, X ) for < then q = (a, < X ) p and q = g,

and g( ) = h for all < .

There is a version of PU , called tree Prikry forcing, where we dont need to

assume that U is normal (in M) in order to verify the Prikry Lemma; cf. Problem

10.24, which produces a generalization of tree Prikry forcing.

Definition 10.8 Let M be a transitive model of ZFC, and let , U M be such that

U witnesses that is measurable in M. A strictly increasing sequence (n : n < )

which is cofinal in is called a Prikry sequence over M (with respect to U ) iff for

all X P() M.

X U {n : n < }\X is finite.

190

10 Measurable Cardinals

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.

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)

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

(10.5)

so that with (10.4)

M[G] |= cf() ++ .

(10.5) and (10.6) yield a failure of SCH in M[G].

(10.6)

with the help of iterated forcing starting from a model with a measurable cardinal

whose Mitchell order (cf. Problem 4.27) has rank (2 )+ .

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

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.

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).

193

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.

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.

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<

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 .

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

195

f g

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:

.

(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

(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)

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 ).

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

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)

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.

199

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)

(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)

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)

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

/ 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)

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)

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)

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)

/ h M j ()},

{i : i < j} = { < j : 0

(10.17)

/ h N j ()}.

{i : i < j} = { < j : 0

(10.18)

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)

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

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 = .

203

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)

k( ) ( + 1) = id,

(10.23)

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

205

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)

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

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

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)

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( )})

()

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]).

207

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)

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

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

/ h L[x] ({ j : j = i}).

which is nonsense. Therefore, i

But then if is the least ordinal with

/ h L[x] ({ j : j = i}), then = j for

some j OR as in the proof of (1), and hence j = i, i.e., = i . We have shown

that

/ h L[x] ({ j : j = i}) = h L[x] (i { j : j > i}).

i

(5) By (1), it suffices to prove that for every i OR there is some j OR with

(i ) = j . Let us fix i OR, and let us write = (i ). Let

L[x]

= h L[x] (( + 1) ran( )) = h L[x] (( + 1) {( j ): j OR}) L[x].

Let be a limit ordinal with > . Let i < , [i + 1]< , and i < 1 <

. . . < n < n+1 < . . . < k . If h L[x] (i , (, 1 , . . . , k )) < , then by (2)

we must have that

h L[x] (i , (, 1 , . . . , k )) < n +1 ,

and therefore

h L[x] ((i + 1) {1 , . . . , k }) n +1 ,

so that by the elementarity of ,

209

This implies that if , then

h L[x] (( + 1) {( ): OR}) ( ) .

(10.30)

otp(h L[x] (( + 1) {( ): OR}) ( )) = .

(10.31)

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

(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{}

} E a{} .

(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

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{}

as desired.

}),

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)

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 ).

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)

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}.

[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

A).

N = ( D;

Claim 10.49 ( os Theorem) Let (v1 , ..., vk ) be a 0 formula (in the language of

M), and let (a1 , f 1 ), ..., (ak , f k ) D. Then

N |= ([a1 , f 1 ], ..., [ak , f k ])

{u [c ]

Card(c)

; M |=

E c for c = a1 . . . ak .

Proof Notice again that the relevant sets are members of M. Claim 10.49 is shown

by induction on the complexity of , by exploiting (1) and (2) of Definition 10.45.

Let us illustrate this by verifying the direction from right to left in the case that, say,

v0 v1 for some 0 formula .

We assume that, setting c = a1 . . . ak ,

{u [c ]Card(c) : M |= v0 f 1a1 ,c (u) (v0 , f 1a1 ,c (u), . . . , f kak ,c (u))} E c .

Let us define f 0 : [c ]Card(c) ran( f 1 ) {} as follows, where <ran( f1 ) M is a

well-ordering of ran( f 1 ).

f 0 (u) =

M |= (x, f 1a1 ,c , . . . , f kak ,c (u))

otherwise.

{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].

[{}, pr] = for < .

(10.34)

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 ( ).

/ 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

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) : 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)

, 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)

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

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 .

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

E () = () > . Also E (e) = (e). Therefore,

[] V ran( E (e)).

(10.37)

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)

[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

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 ])

{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

=: 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

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)

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

: P 2 V,

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

M

1+ M,

: U 1+ VM = V 1+ V()

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 )

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).

223

(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

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)

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)

(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.

225

for some (all) large enough n < . But as {nb : n < } ran( ), and hence if

(Mb , (ib : i b)) is the direct limit of (Mi , i j : i T j < b), then (10.43) yields

that

i(n)b ( 1 (nb )) = i(m)b ( i(n)i(m) ( 1 (nb )))

= i(m)b (i(n)i(m) (nb ))

b

)

> i(m)b (m

b

))

= i(m)b ( 1 (m

for all n < m < , so that ( i(n)b ( 1 (n )): n < ) witnesses that dir lim( M i , i j :

i T j b) must be ill-founded. This contradicts the existence of , cf. (10.45).

We therefore must have that U = (U ; U ) is well-founded. In order to finish the

proof of the theorem, it now suffices to derive a contradiction.

In order to work towards a contradiction, we need generalized versions of U

as well as realizations and enlargements. Let R be a transitive model of ZC

plus replacement for 1000 -formulae. We then call the triple (, Q, R) a realization

of M i , where i < , iff : M i 1000 Q, Q is a (not necessarily proper) rank initial

2 0 R R. If (, Q, R) is a realization of M

i , then we may define a tree

U ( 0i , Q) = (U ( 0i , Q), U ( 0i ,Q) )

in the same fashion as U was defined above: we set (, j) U ( 0i , Q) iff

j < and : M j 1000 Q is such that 0 j = 0i , and if (, j), ( , k)

U ( 0i , Q) then we write ( , k) U ( 0i ,Q) (, j) iff j T k and jk = .

Hence U = U (, V ).

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

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)

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

with respect to U , by the proof of Lemma 10.64 we may define : M i+1 1000 Q j

by setting

(10.47)

Card(a)

where a is a finite subset of the length of E i and writing = crit( E i ), f : []

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

227

{u: M j |= ( f a1 ,c (u), . . . , f ak ,c (u)} ( E i )c

1

()

{u: Q j |= ( j ( f a1 ,c )(u), . . . , j ( f ak ,c )(u)} i ( E i ) (c)

1

()

Q j |=

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

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

Rj

VRi = V

|= ( i ),

We end this section by defining the concept of Woodin cardinals. The following

definition strengthens Definition 4.60.

Definition 10.75 Let be a cardinal, let > be a limit ordinal, and let A V .

We say that is strong up to with respect to A iff for all < there is some

elementary embedding : V M such that M is transitive, crit( ) = , V M,

and (A) V = A V .

Definition 10.76 A cardinal is called a Woodin cardinal iff for every A V

there is some < such that is strong up to with respect to A.

Lemma 10.77 Let be a Woodin cardinal. Then is a Mahlo cardinal. In fact,

for every A V there is a stationary subset S of such that for all S, for all

228

10 Measurable Cardinals

< there is a certified extender F with critical point and lh(F) such that

if F : V Ult(V ; F) is the ultrapower map, then F (A) V = A V .7

Proof Fix A V . Let C be club in . Let f : C be the monotone

enumeration of C, and let g: A be a surjection such that

g = A V for every inaccessible cardinal .

(10.48)

Define h: V by setting

h( ) =

f ( + n)

g( + n)

if = + 2 n + 1 for some limit ordinal and n < .

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)

(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)

a limit and n < }, by (10.48) and the elementarity of F together with the

7

229

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 .

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)

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

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

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)}.

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.

Springer International Publishing Switzerland 2014

235

236

Lemma 11.3 Let M = J [E] be an acceptable J-structure. If is an infinite cardinal of M (or = ), then

(H ) M = J [E].

Proof It suffices to prove that if M and = +M , then (H ) M = J [E].

That J [E] (H ) M follows from Lemma 5.16. Let us prove (H ) M J [E].

Suppose not, and let x be -minimal in (H ) M \ J [E]. Then x J [E], and

there is some surjection g: x, g M. For < , let < be least such that

g( ) J [E]. By (the proof of) Lemma 4.15, = sup({ : < }) < .

By Lemma 11.2, there is some < , such that

(P() J + [E]) \ J [E] = ,

which by acceptability (and Lemma 10.17) yields some surjective f : J [E],

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.

1 (M) is a cardinal in M. In fact, 1 (M) is a 1 -cardinal in M, i.e., there is no 1M

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

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

A = { :

/ g( )}.

/ J [E] by the proof of Theorem 1.3. We

Then A is clearly in P( ) 1M , and A

get that A

/ M by Lemma 11.2. But < , so that we must have that A M by

the definition of . Contradiction!

237

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.

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

p

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

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 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 .

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)

read (11.2) as for cofinally many v j , (v j ). A map : M N which preserves

239

: 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)).

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)

240

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].

M be transitive structures.

Corollary 11.15 Let M,

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

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

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 .

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

241

. . . , h M (n , (z , p))),

M |= n (h M (n 1 , (z1 , p)),

(11.5)

(11.6)

M |= e(n) (z1 , . . . , z , p).

(11.7)

p

(11.8)

p

Since is 0 preserving, (11.8) and (11.9) are equivalent.

(11.9)

h M (n, ((z), p))

(h M (n, (z, p)))

(11.10)

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)

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.

242

Mp

p !

M =N

Let

let N = (J [ E ], A).

M

= h M (ran( ) { p}) 1 M,

(11.12)

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)

243

p = 1 ( p).

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],

,

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)

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,

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)

Because there is a 1M map of onto J [ E],

p R M .

We now aim to prove a dual result, the upward extension of embeddings lemma.

Definition 11.19 Let M and M be acceptable J-structures. A map : M M is

called a good embedding iff

(a) : M 1 M

(b) For all R and R such that R M 2 is rudimentary over M and R M 2 is

rudimentary over M by the same definition,

if R is well-founded, then so isR.

244

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 :

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

245

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

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) E(m,

y) h (n, (x, p))

h (m, (y, p))

F

W

F,

and W are IObviously, I is an equivalence relation, and the predicates E,

[n, x] = {[m, y]: (n, x) I(m, y)}

M/ I = {[n, x]: (n, x) 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 ),

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

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

I , E, F, and W as follows, where (n, x), (m, y) N .

246

(n, x)E(m, y) (n 3 , ((n, x), (m, y))) 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] 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)

247

(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)),

(11.21)

M p as follows.

p

M p |= mx ((e(n), (m, x)) A M

p

(e(e(n)),

(11.22)

v e(n)

(v, h M (m, (x, p))).

Equation (11.22) is true as

p

M |= v e(n)

(v, h M (m, (x, p)))

M |= e(n)

h M (m, (x, p)))

p

(e(e(n)),

The statement (11.22) will be transported upward via , and we thus have

N |=mx (e(n), (m, x)) A

(e(e(n)),

(11.23)

We have shown that for all formulae = n which are 1 but not 0 and for all

(m, x) N ,

(e(n), (m, x)) A

(m , x ) N (e(e(n)),

(11.24)

N , then

248

(m , x ) N (e(v1 v2 (v1 , v2 , v3 )),

((m , x ), (m , x ), (m, x))) A .

(11.25)

We know that E/

p

then there is some (m , x ) M with

p

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,

= (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))

= h M (m, (y, p))

(n, x) I(m, y)

h M (n, (x, p))

p

[n, (x)] = [m, (y)] .

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

()

M |= v n (v, (z)).

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)

p

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 (k1 , (x, p)),

h M (k2 , (0, p))))

250

yields that

p

(v1 = h(v2 , (v3 , v4 )), ((n, x), (k1 , n), ((k1 , x), (k2 , 0)))) A M ,

so that for all (n, x) N ,

(v1 = h(v2 , (v3 , v4 )), ((n, x), (k1 , n), ((k1 , x), (k2 , 0)))) A ,

(11.26)

([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)

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,

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

p(n)

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.

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.

(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, h M m, pm (e0 , (z, p(m))))

M m, pm |= j ((x, z), p(m))

p(m)

252

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

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

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

(11.29)

i, p

n+1, p

hM

n+1, p

(X ) for h M

( < X ).

n+1, p

M rather than h M

The following is straightforward.

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 .

P(M n, p ), say

n, p

x A M |= x1 x2 /xk (x, y, x1 , x2 , , xk ),

253

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

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.

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).

255

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)

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

(11.34)

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.

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)

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)

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.

257

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)

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

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.

258

, p1 (M)

WM

M be acceptable J-structures, and let : M 1 M. Let

Lemma 11.45 Let M,

<

p OR M , and set = ( ) and p = ( p).

(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.

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).

p.

p. Then (W , r ) is a witness for p with respect to 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.

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

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)

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)).

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)

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

J

(J )n

260

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)

must therefore have that = . Then p R nJ . For every i < n, we certainly have

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

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!

261

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 )

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)

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

/ (J )n , say

B = {x (J )n : (J )n |= (x, r )},

(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)

262

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).

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.

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.

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)

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 .

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)

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

265

H

J i

ultni (J i , EJ )

i

J i

ultn j (J j , EJ )

j

J j

( j )

( i)

LH

(crit( ))

j

i

crit( )

i

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,

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)

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)

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.

267

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

there is a badness witness ( R,

i

j i

there is a badness witness ( R,

i

k

k

j i

),

)

i, j

j

j

j

Lemma 11.31. By Claim 11.58, this then finishes the proof of (3) = (2) of Theorem

11.56, as obviously Y is closed under all functions from A. Let us assume is not

as claimed and work towards a contradiction.

We write Y = Y and H = H . We also write I = I , i = i , etc.

By assumption, there is some j I such that ult n j (J j ; E J j ) is bad. Let

R, ([ak , f k ] : k < )) be a badness witness for ult n j (J j ; E J ).

( R,

1

1 Yi

i : H i H and Y i =

=

Let us write i =

ran( i ). Hence

Y i H , Y i Y l for i l, Card(Y i ) < for i < , Y = i< Y i for limit

ordinals and H = i< Y i . Let > be some sufficiently large regular

cardinal. As is regular, we may pick some Z H such that (Y i : i ) Z and

Z . If i 0 = Z , then Z H = Y i0 . We may thus assume that Z H is

such that

(1) Z < ,

H } Z , and

(2) { f k : k < } { R,

(3) Z H = Y i0 for some i 0 < .

Let us write : H

= Z , where H is transitive. Let us also write R = 1 ( R)

1

1

and f k = ( f k ) for k < . Obviously, (H ) = H i0 and H i0 = i0 . Let

k < . We then have [ak+1 , f k+1 ]R[ak , f k ] in ult0 ((J j )n j ; E J j ), and hence

268

= ({(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

(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 .

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 .

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() .

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.

270

Proof This proof is in need of the fact that every level of the L-hierarchy is solid,

cf. Definition 11.40 and Theorem 11.54.

Let us set C = { < + : J J + }, which is a club subset of + consisting

of limit ordinals above .

Let C. Obviously, is the largest cardinal of J . We may let () be the largest

such that either = or is a cardinal in J . By Lemma 5.15, (J() ) = .

Let n() be the unique n < such that = n+1 (J() ) < n (J() ).

If C, then we define D as follows. We let D consist of all C such

that n( ) = n() and there is a weakly r n()+1 elementary embedding

: J( ) J() .

such that = id, ( pn( )+1 (J( ) )) = pn()+1 (J() ), and if J( ) , then

J() and ( ) = . It is easy to see that if D , then by Lemma 11.53 there

is exactly one map witnessing this, namely the one which is given by

n( )+1, pn( )+1 (J( ) )

h J( )

(i, x) h J()

(i, x),

(11.57)

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

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

= 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( )

where i < and x []< . We need to see that if i < , x []< , and h (i, x)

exists, then h (i, x) exists as well.

Let e < and y []< be such that = h (e, y). We obviously cannot have

that h (e, y) exists, as otherwise

, (h (e, y)) = h (e, y) = , (h (e, y)) = , () = ,

but

/ ran(, ). Write n = n() = n( ) = n(). Let < n (J( ) ) be least such

that S contains a witness to v v = h (J( ) )n (u(e), (y, p(J( ) )(n))), cf. (11.29).

We claim that

ran(, ) (J() )n , (S ).

(11.58)

If (11.58) is false, then we may pick some < n (J() ) such that , ( )

, (). We may then write h (e, y) = as a statement over J() in the parameter

272

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

(a)

(b)

(c)

(d)

(e)

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( )

h J()

(k, x) ran(, ) =

(k, ) ran(, ) = .

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 () .

273

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 11.67 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

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 < }.

274

We now use at most the next Card(S) ordinals above T to construct T (+1)

in such a way that for every s S there is some t T ( + 1) \ T such that

s <T t for all s cs and for every t T ( + 1) \ T there is some s S

such that s <T t for all s cs .

This finishes the construction. It is not hard to verify that the construction in

fact never breaks down and produces a + -tree. Otherwise, for some limit ordinal

< + there would be some s S and some limit ordinal i < such that there is

no t T (i + 1) \ T i with s j <T t for all j < i (for S, , (i : i < ) as

/ R, so that Ci = C i is easily seen

above). However, i C gives that i

to yield that in the construction of T (i + 1) \ T i , we did indeed add some

t T (i + 1) \ T i such that s j <T t for all j < i.

Finally, suppose that T would not be + -Souslin, and let A T be an antichain

of size + . As R is stationary, we may then pick some limit ordinal R such that

A = S is a maximal antichain in T , so that also S = {s T : t

S (t = s t <T s)} (for S as above). The construction of T then gives that every

node in T \ T is above some node in S , so that in fact A = A = S . But

because T is a + -tree, this means that A has size at most . Contradiction!

Lemma 11.69 (Jensen) Let be subcompact. Then fails.

Proof Suppose (C : < + ) witnesses . As is subcompact, there must then

be some < together with a witness (D : < + ) to and an elementary

embedding

: (H+ ; , (D : < + )) (H + ; , (C : < + ))

with crit( ) = . Set = sup( + ) < + . As + is < -club in ,

C = C +

is also < -club in .

Let () < () < both be limit points of C. Then

C () () = C () = C () .

(11.59)

D = D .

Setting

D=

D ,

()C

() C such that otp(D ) > . Then otp(D ) > . Contradiction!

275

In particular, if is supercompact, then fails for all .

Proof Let be + -supercompact, where , and suppose that (D : < + )

witnesses that holds. Let

: V M

+

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

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 < .

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

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

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

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<

Springer International Publishing Switzerland 2014

279

280

2n+1

X X.

n<

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

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

...

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

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

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

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

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

Because II follows s in the play above, there is some m n and some X with

({0 , . . . , m1 }, X ) D. But then ({0 , . . . , m1 }, Z x m ) D. However, for any

l m, l X 0 , and hence l Z x m . We thus have that

1

12.1 Determinacy

285

({0 , . . . , m1 }, Z x m ) G.

Therefore, D G = .

We shall now prove that AD, the Axiom of Determinacy, proves the ultimate

generalization of the Cantor- Bendixson Theorem 1.9. If AD were not to contradict

the Axiom of Choice, it could be construed as providing a solution to Cantors

project of proving the Continuum Hypothesis, cf. p. 3. The following Theorem says

that all of P( ) has the perfect subset property (cf. p. 138).

Theorem 12.11 (M. Davis) Assume AD. Every uncountable A has a perfect

subset.

Proof Fix A . Let f : 2 be a continuous bijection (cf. Problem 7.2),

and write B = f A. It suffices to prove that B is either countable or else has a perfect

subset.

Let us consider the following game, G p (B).

I s0

s1

...

II n 0 n 1

...

In this game, I plays finite 01-sequences si < 2 (with si = being explicitly

allowed), and II plays n i 2 = {0, 1}. Player I wins iff

s0 n 0 s1 n 1 . . . B,

otherwise II wins. We may construe G p (B) as G(B ) for some B , so that

G p (B) is determined.

If I has a winning strategy in G p (B), , then

{s0 n 0 s1 n 1 . . . : (n 0 , n 1 , . . .) 2 i < si = (s0 n 0 . . . n i1 )}

is a perfect subset of B.

So let us suppose to be a winning strategy for II in G p (B). We say that a finite

sequence p = (s0 , n 0 , . . . , sk , n k ) is according to iff n i = (s0 n 0 . . . si ) for

every i k, and if x 2, then we say that p = (s0 , n 0 , . . . , sk , n k ) is compatible

with x iff s0 n 0 . . . si n i x. Because is a winning strategy for II, we

have that x

/ B follows from the fact that for all p = (s0 , n 0 , . . . , sk , n k ), if p

is according to and compatible with x, then there exists some s 2 such that

(s0 , n 0 , . . . , sk , n k , s, ((s0 , n 0 , . . . , sk , n k , s)) is compatible with x. In other words,

if x B, then there is some px = (s0 , n 0 , . . . , sk , n k ) which is according to and

compatible with x such that for all s 2, (s0 , n 0 , . . . ,sk , n k , s, ((s0 , n 0 , . . . , sk ,

n k , s)) is not compatible with x.

Notice that x px is injective for x B. This is because if x B, px =

(s0 , n 0 , . . . , sk , n k ), and

286

x = s0 n 0 . . . sk n k m 0 m 1 m 2 . . . ,

then we must have that m i = 1 (s0 n 0 . . . sk n k m 0 . . . m i1 ) for every

i < .

But as x px , x B, is injective, B is countable.

Lemma 12.12 Assume AD. Let A 2 be such that every Lebesgue measurable

D A is a null set. Then A is a null set.

Proof Let us fix A 2 such that every Lebesgue measurable D A is a null

set. Let > 0 be arbitrary. We aim to show that A can be covered by a countable

union B of basic open sets such that (B) . Let us consider the following game

G cov (A), the covering game for A.

n1

...

I n0

II

K0

K1

...

In this game, I plays n i {0, 1}, i < , and II plays K i , i < , where each K i is a

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\

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

Theorem 12.13 (Mycielski-Swierczkowski) Assume AD. Every set A

Lebesgue measurable.

is

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

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,

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.

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).

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

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 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

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

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 .

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

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 }, <) ,

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

(, <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

(, <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!

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!

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)

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

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) <

}. 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 .

297

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 ).

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 )

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 ).

298

TCol(,+(+2n))

J [x, T ] |= (1 , . . . , k ).

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

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.,

But then II can easily defeat by playing a code for a transitive model of ZFC +

V = L[x] which contains .

By Claim 12.25 and 11 (x)-determinacy, we may let be a winning strategy for

player II in G. By Lemma 12.21, the following will produce x # .

2

299

Claim 12.26 Let be a countable ordinal such that J [x, ] |= ZFC . Then is

a cardinal of L[x]. In fact, if is countable and x -admissible (cf. p. 88), then

is a cardinal of L[x].

Proof Let < be an infinite cardinal of L[x]. It suffices to verify that if b

, b L[x], then b J [x, ] (cf. Problem 12.10). So let us fix some such b, and

let b J [x], where without loss of generality 1 > . Let > be such that

J [x, ] |= ZFC and is countable. Let G be TCol(, +1)-generic over J [x, ],

and let T and H be given by G as in (12.1). Obviously, there is a real z 0 in J [x, , T ]

(even in J+ [x, , T ]) which codes a well-founded tree S such that ||S|| = . E.g.,

let S = T s, where s T with H (s) = . Therefore, z 0 J [x, , T ] (in

fact J+ [x, , T ]) codes a model (; E) of ZFC + V = L[x] such that is

contained in the transitive collapse of the well-founded part of (; E). Let

: (J [x]; )

= wfp(; E)

be the transitive collapse of the well-founded part wfp(; E) of (; E), so that .

By E J+ [x, , T ], it is easy to verify inductively that J [x] is uniformly

J [x,,T ]

1

[x, , T ] for all , so that in particular

J [x] J+ [x, , T ].

(12.2)

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)

TCol(,+1)

(t, h) L[x, ]

< (

b J+

[x, , T ] |= (, 0 , 1 )),

TCol(,+1)

b (t, h) L[x, ]

J+

[x, , T ] |= (, 0 , 1 ).

(12.5)

+ ( + 2) , we may use Claim 12.24 to rewrite (12.5) further to say that for

every < ,

TCol(,)

J+

[x, , T ] |= (, 0 , 1 )

(12.6)

300

But we may replace L[x, ] by J [x, ] here (we could in fact replace it by

J+2 [x, ]), so that we may therefore define b over J [x, ] as follows:

TCol(,)

J+ [x, , T ] |= ( , 0 , 1 )}.

Therefore b J [ ] as desired.

We may now add the following to our list of equivalences to x # exists, cf.

Theorem 11.56.

Corollary 12.27 The following statements are equivalent.

(1) Every analytic A is determined.

(2) For every x , x # exists.

The paper [8] gives information on the Axiom of Determinacy. Cf. also [25].

12.4 Problems

12.1 Assume AD. Show that if (xi : i < ) is a sequence of pairwise different reals,

then < 1 [Hint. Let U witness that 1 is measurable, cf. Theorem 12.18

(b). Consider L[U, (xi : i < )], cf. proof of Problem 10.3 (a)].

12.2 Show (in ZF) that there is some A (1 ) which is not determined [Hint. If

AD holds, then ask for II to play some x with ||x|| = in response to

I playing < 1 ].

12.3. Assume AD. Show that for every set A, OD{A} is countable. Fixing A,

show that there is no f : such that f (x) \OD{x,A} for all

x and f OD {A} . Conclude that HOD {A} |= AD and there

is some (A x : x ) with = A x for all x with no choice

function.

12.4. Assume ZF plus x # exists for every real x. Show that there is some f :

such that f (x) \L[x] for all x and f OD. (In fact, we

may pick f to be 31 , cf. Problem 10.7.) Show also that there is a function

f : HC such that for every x , f (x) is a C-generic filter over L[x]

(Hint. Use x # to enumerate the dense sets of L[x]).

12.5. For A, B , write A Wadge B iff there is some continuous f :

such that for all x , x A f (x) B, or for all x ,

x A f (x)

/ B. Assume AD.

(a) (Wadge) Show for all A, B , A Wadge B or B Wadge A. [Hint.

Let G Wadge (A, B) be the game so that if I plays x and II plays y, then I

wins iff x B y A]. Show that Wadge is reflexive and transitive,

so that A Wadge B iff A Wadge B B Wadge A is an equivalence

relation. Show that Wadge is not symmetric.

12.4 Problems

301

(b) (Martin, Monk) Show that Wadge is well-founded [Hint. Otherwise there

are An , n < , such that for all n < , I has winning strategies n0

and n1 for G Wadge (An+1 , An ) and G Wadge ( \An+1 , An ), respectively.

z(n)

z

for all

For z 2, we get (xnz : n < ) such that xnz = n xn+1

n < . Let z, z 2, and n < be such that for all m, z(m) = z (m)

z

z

iff m = n. Then xn+1

= xn+1

, and xnz An xnz

/ An , . . .,

x0z A0 x0z

/ A0 . Hence {z 2: x0z A0 } is a flip set, cf.

Problem 8.3].

(c) Show that for all A there is some J (A) with A <Wadge J (A)

[Hint. For x write f x for the canonical continuous function given

/ B}].

by x. Let x B iff f x (x) A, and set J (A) = {x y: x B and y

(d) Let = sup({: surjective f : }). Show that || <Wadge || =

[Hint. To show that || <Wadge || , let f : be surjective, and let

(A : < ) be such that if < , then A = J ({x y: f (x) < y

A f (x) }), where J is as in (c)]. Let X = . If A ( ), then in a run

of the game G(A) players I and II alternate playing real numbers, i.e.,

elements of . The Axiom of Real Determinacy, abbreviated by ADR ,

states that G(A) is determined for every A ( ).

(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

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.

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

< )

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.

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 ) = 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

((

),

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

...

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)

/

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

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)

307

(s,x)

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.

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

(13.6)

< ,

s,t (x, y) =

lh(s)

x,y

0

309

if (x, y)

/ A s = x lh(s) t = y lh(s)

otherwise

(13.7)

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),

(13.8)

/ 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),

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

We intend to have sn correspond to the node 2n 1

<

For s , we shall produce objects

= 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)),

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 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

Es,2n

k = 2n

sm 2m 1

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).

311

2n 1))

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)).

s ,t

= (,),(s ,t ) (Ms,k ),

Ms,k

s ,t

s ,t

Ms,k

Ms,k

= V

s ,t

for all k l 2 lh(s).3 The models Ms,k

are the models Ms,k from the point of

s ,t

view of Ns ,t , and the embeddings s,k,l

are the embeddings s,k,l from the point

of view of Ns ,t .

s ,t

s ,t

s ,t

(s ,t ),(s ,t ) s,k,l

= s,k,l

((s ,t ),(s ,t ) Ms,k

),

(13.10)

There is a slight abuse of notation here, as Ts is not a set but rather a (sequence of) proper class(es).

By (Ts ) Ns ,t we mean that object which is defined over Ns ,t from the parameter (E s,k : k < 2lh(s))

by the very same formula which defines Ts over V from the same parameter (E s,k : k < 2 2lh(s)).

{(,),(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 ,t ),(s ,t ) (x).

= s,k,l

(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

Suppose that we manage producing objects Ts , s,k , s,k , and s,k with the properties (PD, 1) through (PD, 4). We may then verify the following

Claim 13.9

((Ms,2lh(s) : s < ), (s ,2lh(s),2lh(s ) : s s , s, s < ))

is a 20 closed embedding normal form for {x : y (x, y)

/ A}.

Proof 20 closedness is clear by (PD, 1), as every extender used is 20 closed in

the model where it is taken from. Cf. Lemmas 10.60 and 10.61.

Let us fix x . Let

(M Even , (2k, Even : k < )) = dir lim((Mx k,2k : k < ), (x l,2k,2l : k l < )).

We need to see that

y(x, y) A = M Even is ill founded

(13.11)

y(x, y)

/ A = M Even is wellfounded.

(13.12)

and

Let us first show (13.11). If s, t < with lh(s) = lh(t), then by the coherence

s,t

for Mxs,tk,k (= Mxs,tm,k for all sufficiently large

condition (PD, 2), we may write Mx,k

s,t

for xs,tl,k,l (= xs,tm,k,l for all sufficiently large m).

m) and x,k,l

Let us now pick some y such that (x, y) A. Let

313

(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

x,y

x,y

x,y

By the coherence condition (PD, 2) we may write Mx,k for Mx k,k (= Mx m,k

x,y

x,y

x,y

for all sufficiently large m), x,k,l for x l,k,l (= x m,k,l for all sufficiently large

m), and

x,y

x,y

x,y

x,y

(13.14)

x k,y k

x,y

x k,y k

(13.15)

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 (k+1),y (k+1)

x,2n+2,

x (k+1),y (k+1)

(13.16)

x,y

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

x,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)||

(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

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 ,l ) : 2l 1

xni+1 ,0,2ni+1 1

(Tx,y )

i+1 1

(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 < ).

315

We are left with having to produce the objects Ts , s,k , s,k , and s,k such that

(PD, 1) through (PD, 4) hold true.

We first need a set of indiscernibles. Let us fix , a cardinal which is much

s ,t

(s,t ) will be in

bigger than ; in particular, we want that T V and that all 0,k

V also. Let c0 < c1 < be strong limit cardinals above of cofinality > such

that

(13.17)

type V (V ; , V , {, c0 })) = type V (V ; , V , {, c1 })).

An easy pigeonhole argument shows that such objects exist: e.g., let be a strong

+

limit cardinal of cofinality V+1 . In the construction to follow, we shall use the

descending chain of ordinals

c0 + 1 > c0 c1 > c0 + 1 > c0 c1 > . . .

and we may and shall assume that , c0 , c1 , and are fixed points of all the elementary

embeddings which we will encounter. Cf. Lemma 10.56 Problem 10.18. will always

be a fixed point anyway.

In order to keep our recursion going, we shall need a fifth condition.

M2l 1

The objects Ts , s,k , s,k , and s,k will be constructed by recursion on the length

of s. To get started, we let T be the trivial tree of length 1 which just consists of V .

We also set ,0 = c0 , and we pick ,0 < , ,0 > , such that in V , ,0 is strong

up to with respect to

type V (Vc0 +1 ; , V , {, , T, , }).

This choice of ,0 is certainly possible, as is a Woodin cardinal, cf. Lemma 10.81.

We also set ,0 = , .

Now let us fix s < of positive length throughout the rest of this proof. Write

n + 1 = lh(s). Let us suppose that

Ts n

s,k for k 2 n,

s,k for k n, and

s,k for k n

have already been constructed. We are forced to set s,k = s n,k for k 2 n,

E s,k = E s n,k for k < 2 n, Ms,k = Ms n,k for k 2 n, s,k,l = s n,k,l for

k l 2 n, s,k = s n,k for k n, and s,k = s n,k for k n. Let us write

t = sn+1 , k = lh(t) n + 1, and sm = t (k 1). We now need to define s,2n+1 ,

s,2n+2 , s,n+1 , and s,n+1 , and we also need to define Ms,2n+1 as an ultrapower of

316

13 Projective Determinacy

Ms,2m 1

by an extender E s,2n with critical point s,2m and Ms,2n+2 as an ultrapower

of Ms,2n by an extender E s,2n+1 with critical point s,2n+1 .

2n + 2

t = sn+1 2n + 1

Es,2n+1

Es,2n

t k 1 = sm 2m 1

k = 2n

As s

and write

<

0

will be fixed from now on, we shall mostly suppress the subscript s

Mk

for Ms,k ,

k,l for s,k,l ,

s ,t

,

Mks ,t for Ms,k

s ,t

k,l

k

k

k

s ,t

for s,k,l

,

for s,k ,

for s,k , and

for s,k .

Inductively, we shall assume that the following two statements, (A) and (B), are

satisfied. Here, s i,t i is as in (13.7).

(A)

sk1,tk1

s k1,t k1

((,),(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

sk1,tk1

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

317

sk,t

; , V ,

2m,2n ((sk1,tk1),(sk,t) (m ))

s k,t

s k,t

{, , 0,2n ((,),(s k,t) (T )), (0,2n ((s i,t i),(s k,t) (s i,t i ): i

k)}),

s k,t

cf. Lemma 10.81. We may apply the map 2m,2n (s k1,t k1),(s k,t) to (B), which

by (13.10) and the fact that crit(2m,2n ) = 2m+1 > 2m produces the assertion that

s k,t

sk,t

(sk1,tk1),(sk,t) (m ))+1

2m,2n

s k,t

; , V ,

(13.18)

s k,t

{, , 0,2n ((,),(s k,t) (T )), (0,2n ((s i,t i),(s k,t) (s i,t i )): i < k)}).

We may thus let

.

.

: M2m 1

E s,2n M2n+1 ,

2m 1,2n+1

where E s,2n M2n is a 20 closed certified extender in M2n which witnesses that

s k,t

inside M2n , 2m is strong up to the least inaccessible cardinal above 2n+1 with

.

respect to the type from (13.18); notice that (V2m +1 ) M2n = (V2m +1 ) M2m 1 , cf. (PD,

1), immediately gives that

(V ) M2n+1 = (V ) M2n .

(13.19)

We have that

type M2n+1 (Vc0 +1 ; , V2n+1 +1 , {, ,

.

.

0,2n+1 (T ), (2i 1,2m

1 (i ): 2i 1 2m 1) )

sk,t

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)})),

.

.

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

2m,2n

(sk1,tk1),(sk,t) (m )

; , V2n+1 , {, ,

s k,t

(13.21)

s k,t

(0,2n ((s i,t i),(s k,t) (s i,t i )): i

k)}).

s k,t

s k,t

M2n

sk,t

2m,2n

(sk1,tk1),(sk,t) (m )

; , V2n+1 ,

s k,t

s k,t

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

s k,t

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),

319

(i ): 2i 1 2m 1) })

{, , 0,2n+1 (T ), (2i 1,2n+1

(13.23)

2n+1 is strong up to with respect to

type M2n+1 (Vc0 ; , V ,

(i ): 2i 1 2m 1) }) ).

{, , 0,2n+1 (T ), (2i 1,2n+1

M2n1

Let n+1 V

(D)

s k,t

2m,2n

s k,t

s k,t

= type M2n+1 (Vc0 ; , V2n+1 , {, , 0,2n+1 (T ),

.

(i ): 2i 1 2n + 1)})

(2i 1,2n+1

and

(C)

type M2n+1 (Vc0 ; , V , {, , 0,2n+1 (T ),

.

(i ): 2i 1 2n + 1)}).

(2i 1,2n+1

M2n+1

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)}),

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

s k,t

2m,2n+2 (sk1,tk1),(sk,t) (m )

; , V2n+2 +1 ,

s k,t

0,2n+2 ((s i,t i),(s k,t) ((s i,t i)) : i

= type M2n+1 (Vc1 ; , V2n+2 +1 ,

k)})

{, , 0,2n+1 (T ),

.

.

(i ): 2i 1 2n + 1)}).

(2i 1,2n+1

s k,t

sk,t

2m,2n (sk1,tk1),(sk,t) (m )

s k,t

; , V2n+1 ,

s k,t

0,2n ((s i,t i),(s k,t) ((s i,t i)) : i k)})

sk,t

= 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

type

M2n+1

(V+1 ; , V , {, , 0,2n+1 (T ),

.

.

(i ): 2i 1 2n + 1)}). ).

(2i 1,2n+1

(13.25)

321

We have that

(V2n+2 +1 ) M2n+1 ,

so that the statement (. . .) in (13.25) can be written as an element of the type

from the right hand side of (F), and (F) yields the following.

sk,t

(V sk,t

2m,2n+2 ((sk1,tk1),(sk,t) (m ))

) M2n+2 |=

s k,t

sk,t

s k,t

s,2n+2 is strong up to with respect to

type

sk,t

M2n+2

(13.26)

s k,t

s k,t

Let n+1 be a witness to this fact. In particular, with a brief show of the subscript s,

s k,t

(13.27)

(G)

sk,t

s k,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

sk,t

s k,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.

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

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

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

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

weakly homogeneously Souslin, 324

Shift map, 186, 198

-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

- Introduction to Axiomatic Set Theory (1971) - KrivineUploaded bynaide2
- Judith Roitman - Introduction to Modern Set Theory - WileyUploaded byEdno Dos Santos Sousa
- Handbook of Philosophical Logic Second Edition 11Uploaded byobwohl
- ContinuumHypothesisUploaded byPremkiran Gopal Bhadane
- Modal Logic for Philosophers ---- James W. GarsonUploaded byJoel Alberto Aguilar Velazquez
- Hrbacek-Introduction to Set Theory PDFUploaded byhermahuitron
- Handbook of Philosophical Logic Second Edition 10Uploaded byobwohl
- Geoffrey Hunter Metalogic an Introduction to the Metatheory of Standard First Order Logic 1996 (1)Uploaded byLenin Vázquez
- Leila Haaparanta, Jaakko Hintikka (Auth.), Leila Haaparanta, Jaakko Hintikka (Eds.) Frege Synthesized_ Essays on the Philosophical and Foundational Work of Gottlob Frege 1986Uploaded byMaria Parra
- Drake, Set TheoryUploaded bybassoman
- Krzysztof Ciesielski - Set Theory for the Working MathematicianUploaded byKK-bookslover
- Bertrand Russell and the Origins of the Set-Theoretic 'Paradoxes'. Alejandro R. GarciadiegoUploaded byGustavo Ovatsug
- Hermann Weyl - The Continuum: A Critical ExaminationUploaded byPeisithanatos
- Epistemic Logic by RescherUploaded byKyle Foley
- Modal Logics and Philosophy (Girle)Uploaded byChristian Maura
- ParadoxesUploaded bystructuraldestruction
- A Course on Set TheoryUploaded byHassan Yossry Ibrahim
- Dauben Georg CantorUploaded byTomas 7
- Kleene - Introduction to MetamathematicsUploaded byBruno Saramago Monteiro
- Elements of Set TheoryUploaded byPeihong Jiang
- Modal Logics and Philosophy - Rod GirleUploaded byとなり のサターン
- The Blackwell Guide to Philosophical Logic.pdfUploaded byJeremy Neufeld
- Introducing Philosophy of Mathematics - Michele FriendUploaded bymykalachi1
- Paul Cohen - Set Theory and the Continuum HypothesisUploaded byChan Hin Lung
- Van Heijenoort, J (1967) From Frege To GÃ¶del A Source Book In Mathematical Logic, 1879-1931 Cambridge, Harvard University PressUploaded byLuciano Vaga
- Sider Theodore - Logic for PhilosophyUploaded byRodrigo Del Rio Joglar
- Handbook of Philosophical Logic Vol 1Uploaded byangel21znerolf
- Handbook of Philosophical Logic Second Edition 9Uploaded byobwohl
- Bochenski Joseph - La Logica FormaleUploaded bylsmokke

- Introduction to Automata Theory Languages and Computation by John Hopcroft, UllmanUploaded byHelton Danilo
- Schedules Cambridge Maths guideUploaded byNoob
- 2010 Measure TheoryUploaded byAravindVR
- Roussas g a Course in Mathematical StatisticsUploaded byKarlilla Cuéllar
- Fundamentals of AnalysisUploaded byHyuntae Kim
- UtilityUploaded bysamuelifamily
- Formal.languages.and.Their.relation.to.AutomataUploaded byAashaya Mysuru
- tocUploaded byPoovarasan Kumudha
- InfinityUploaded byJoshua Michael McKenna
- Infinite Games SofiaUploaded byrichard_dalaud
- A Friendly Introduction To AnalysisUploaded byclaudsk97
- Topology Without Tears - SolucionesUploaded byDeskartes
- aljabar modernUploaded bymuhammad soenarto
- Bridge to Abstract MathematicsUploaded bybohemianirc
- Diff Integral CalculusUploaded bykailash
- Countable & Uncountable NounsUploaded byMarlon Zambrano
- LACC 2013 Cantor NoPipeUploaded byOscar Velez
- COL100 Slides Updated 13nov2014Uploaded byKartar Singh
- ch1Uploaded bybond12314
- IntroRealAnal-ch01Uploaded byratchagar a
- Basic Exam Problems and SolutionsUploaded byemerson
- Analysis CohenUploaded byHarold Contreras
- Metric SpacesUploaded byNeelima Kandula
- mathematical.pdfUploaded bymuzaffer_91
- the Theory of Parsing, Translation, and Compiling.pdfUploaded bysonirichnavi
- Advanced CalculusUploaded byElliot Lu
- Joel G. Broida, Stanley Gill Williamson Comprehensive Introduction to Linear Algebra, Part I.pdfUploaded byZoran Loncarevic
- 321Uploaded byAlberto García
- LogicUploaded bySumeet Kalsi
- Natural, Uploaded byrustyy88