You are on page 1of 76

A COMBINATORIAL FORCING FOR CODING THE UNIVERSE BY A REAL WHEN THERE ARE NO SHARPS

Saharon Shelah1,2,4,5 and Lee J. Stanley3,4,5


Abstract. Assuming 0 does not exist, we present a combinatorial approach to Jensens method of coding by a real. The forcing uses combinatorial consequences of ne structure (including the Covering Lemma, in various guises), but makes no direct appeal to ne structure itself.

0. INTRODUCTION.

In [5], S. Friedman calls the original proof in [1] of the Coding Theorem one of the hardest in all of set theory. The technical considerations are
modified:2004-08-01

extremely elaborate and the proof draws heavily on Jensens profound ne structure theory. In addition to providing an excellent overview, both of the general approach and of the particulars of that proof, [5] presents certain simplications. R. David, [2], [3], and in subsequent work, Friedman, [7], [8], [9], [10] and the forthcoming [12], and David, [4], have shown how

revision:2004-08-01

to integrate additional structure into the forcing conditions to obtain yet


(1) Research partially supported by NSF, the Basic Research Fund, Israel Academy of Science and MSRI (2) Paper number 340 (3) Research partially supported by NSF grant DMS 8806536, MSRI and the Reidler Foundation. (4) We are very grateful to the referee for his rare combination of diligence and patience, as well as for many helpful observations. (5) We would also like to thank many colleagues for their interest and the administration and sta of MSRI and for their hospitality during 1989-90. 1

340

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

stronger results. In particular, [8] has generalized the coding method to coding over ground models where there are measurable cardinals, while preserving the measurability of a designated measurable cardinal (in the extension, V = L[ , R], where R is a real and is a normal measure on , extending a designated normal measure, , of the ground model). Also, ignoring all references to measures and mice, pp. 1147 - 1154 of [8] provides a the skeleton of a highly general and concise version of coding over L (obtaining V = L[R], in the extension, where R is a real), with no hypotheses on the ground model (other than GCH), which is fully developed in the forthcoming [11]. Nevertheless, the fundamental features of Jensens approach remain unchanged: the obstacles (which we shall discuss shortly) in the path of a
modified:2004-08-01

naive attempt to piece together the Building Blocks of Chapter 1 of [1] are overcome by integrating ne-structural considerations into the very denition of the forcing conditions. As a result, questions of uniformity, eectiveness and absoluteness of notions involved in locally dened approximations to the forcing must be faced.

revision:2004-08-01

Our approach will be radically dierent, drawing upon the great simplication aorded by the hypothesis of the non-existence of 0 . One of the two main obstacles will be overcome by a preliminary forcing. The heart of the proof of Lemma 3, in [16], which states that this preliminary forcing behaves as needed, involves an appeal to the Covering Lemma.

340

[SHST:340]

The other main diculty is to prove a strategic closure property of the class of coding conditions. This will be done in (5.1) below; in (5.2) and (5.3) it is shown how this yields distributivity properties of the class of coding conditions. The material of 5 (and the material of [17], upon which it draws) appeals to strong combinatorial properties of L, developed in [17]. A sketch of the results required for this paper is presented in (1.2), (1.4), (1.5) below. Here again, the Covering Lemma plays a role, this time by guaranteeing that the L-combinatorics give us a handle on the situation in V . The obvious downside to our approach is the need for the hypothesis that 0 does not exist. The main advantage is that the role of ne structure is modular: it is crystallized in the Covering Lemma itself, and in the Lmodified:2004-08-01

combinatorics. This is quite analogous to the approach that the rst author took in his proof of Strong Covering (an early version appears as [13], while a revised version will appear in the forthcoming [14]). Indeed, in some ways, this paper is an outgrowth of that work. This allows for a simpler denition of the coding conditions, involving the combinatorial apparatus,

revision:2004-08-01

but making no direct reference to ne-structural nor denability notions. It is our hope that non-experts will nd this easier to use and to customize for particular applications they have in mind. There are two other drawbacks to our approach. The rst is that it seems to preclude obtaining sharp denability-type or minimal-degree type of re-

340

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

sults. In fact, this is a rather natural consequence of our seeking a combinatorial forcing, intended for use in obtaining combinatorial consequences. We recognize that this is a signicant departure from the tradition created by Jensens original treatment, and that, in the eyes of those steeped in that tradition, the entire approach may seem somewhat unattractive. The second is a consequence of our treatment of inaccessible cardinals. We treat them in a way which is much closer to our treatment of successor cardinals than to our treatment of singular cardinals, in that, if is inaccessible, then, in any condition p, except for fewer than many [, + ), if is mentioned in p, then p says nothing about a tail of the coding area for , whereas, if is singular, [, + ), is a multiple of 2 (non-multiples of 2 are treated totally dierently), is mentioned in p, then p says somemodified:2004-08-01

thing about a tail of the coding area for . Jensen has informed us that in early, unpublished, versions of the coding paper (which evolved into [1]), he treated inaccessible cardinals as we do, but that he later shifted to treating them in a way similar to his treatment of singular cardinals, in order, e.g., to be able to prove the preservation of small large cardinals, as in 4.3 of

revision:2004-08-01

[1]. Thus, though we have not yet investigated the question, our approach may preclude analogues of some of the results there. We should point out that this way of dealing with inaccessibles (essentially by requiring that the set of cardinals mentioned in a condition is an Easton set) has two main uses. The rst has to do with dealing

340

[SHST:340]

with contamination, see below (2.3), (3.3) - (3.4), and the portion of the SUMMARY AND INTRODUCTION section, below, which deals with these items. The second involves the results of [17] and will be more fully discussed in (1.3). Apparently, the second use is really an essential feature of using ground model scales as the main coding areas at singular cardinals (see the discussion leading up to Lemma 5, below), while the need for the rst use is a result of treating inaccessibles very dierently than singulars. Recently, S. Friedman has circulated a preprint (A Short Proof of Jensens Coding Theorem, Assuming Not 0#) which draws, in part, on ideas of this paper and [17]. Finally, we would like to thank the referee for pointing out that the methods of this paper are compatible with the existence of generics, but that,
modified:2004-08-01

unlike the more classical coding methods, this requires the techniques of [6].

DISCUSSION. The coding theorem we prove is:

revision:2004-08-01

Theorem 1. If V |= ZF C + GCH + 0 does not exist, then there is a denable class forcing, P which preserves ZFC, conalities, GCH and such that in V P , (there is a real r, such that V = L[r]) holds. We should immediately point out that the conditions, P = P2 of 3 add a subset B3 3 , rather than a real. However, since, in the generic

340

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

extension V = L[B3 ] holds, it is an easy matter to code B3 into a subset of 2 , which, in turn, is coded into a subset of 1 , which, nally, is coded into a real, using, e.g., almost disjoint coding. It may be necessary to intersperse the forcings of 1.3 of [1] to reshape the intervals (i , i+1 ), for i = 0, 1, but this is not problematical, since when this is called for, the subset of i+1 which we already have codes the universe. Before embarking on the promised discussion of the obstacles to a naive attempt to piecing together the Building Blocks of [1], Chapter 1 (or some variation on them), and how these obstacles are overcome in Lemmas 3, 5, below, we should note that we follow [5] for the general strategy for proving such a coding theorem, and especially pp. 1005-1006, middle. In particular, it will suce, by the arguments presented in [5], to prove the
modified:2004-08-01

four main properties of P presented there: Extendability, Distributivity, Factoring, and Chain Condition. The versions of these properties which we prove reect the dierences between the detailed denition of our P and that considered in [5], but they are suciently similar that the general arguments for their suciency go over to the setting of this paper. These

revision:2004-08-01

properties are proved in (6.1), (5.3), (4.4) and (6.2) respectively. The Factoring property states that for all regular > 2 , there are P , P such that P P P . The Distributivity property states that = P is (, )-distributive. The Chain Condition property states that, in V P , P has the + -chain condition. The proof of Distributivity in (5.3)

340

[SHST:340]

is based on a strategic closure property of P established in (5.1), together with result of (5.2), which proves that the BAD player need not lose the game for trivial reasons. Though the proof of (5.1) has been rendered rather short and easy, by the introduction of the very tidy conditions, and the preliminary results of (4.3) and (4.5), in many ways, this result is the main lemma of the entire paper. It shows that, by the use of deactivators and generic scales (in addition to the ground model scales used in setting up the main coding apparatus), we can overcome the second of the two main obstacles to a naive attempt to piece together the building blocks. We turn now to a discussion of these obstacles. The rst main obstacle simply involves the possibility of coding R + into a subset of , when is regular. In order to use almost disjoint set
modified:2004-08-01

coding (or, as below, in 2, 3, almost inclusion coding, a variant used in [15], (1.3)), we seem to need extra properties of the ground model, or of the set R, since, in order to carry out the decoding recursion across [, + ) we need, e.g., an almost disjoint sequence satisfying: () : for all (, + ), (b | ) L[R ], and is canonically denable there. Such a b is called decodable. It is easy to obtain a decodable b if R satises: () : for all (, + ), (card )L[R] = . If () holds, we say that R promptly collapses fake cardinals.

340

revision:2004-08-01

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

Of course, typically () fails, and the reshaping conditions of 1.3 of [1], the F B of [5], are introduced to obtain () in a generic extension. Unfortunately, the distributivity argument for the F B seems to require not merely that H + = L + [B], but that H ++ = L ++ [B], where B + . This will be the case if B is the result of coding as far as + , but that is another story, which leads to the original approach to the Coding Theorem. Instead, in [16], we showed, assuming GCH and that 0 does not exist: Proposition 2. Let > 1 be a cardinal, let Z + be such that for all cardinals with + , H = L [Z]. Then, there is a conalitypreserving, GCH-preserving forcing, S(), which adds a W (, + ) such that Z L[W, Z ] and, for all < + , (card )L[W ,
Z]

= .

modified:2004-08-01

Then, starting from A OR such that for all innite cardinals , H = L [A] and taking S to be the product, with Easton supports, of the S() for = 2 or a limit cardinal, we have: Lemma 3. In V S , there is A OR, such that letting be the class of limit cardinals together with 2 :

revision:2004-08-01

A = (A 2 )

{A (, + ) : },

such that for all innite cardinals , H = L [A] and such that for = 2 or inaccessible, for all < + , (card )L[A] = (for singular , the last property is true with L in place of L[A ], in virtue of Covering).

340

[SHST:340]

In virtue of the preceding discussion, we clearly have: Corollary 4. In V S , letting A be as in Lemma 3, for all regular 2 , there is decodable b = (b | (, + )) of conal almost disjoint subsets of as above. In order to discuss the diculty in proving the strategic closure properties of the P , we need to say a bit about the coding apparatus for singular cardinals. This material is discussed at somewhat greater length in (1.2), (2.2) - (2.4) and formally presented in (3.4), (3.5), so the reader who nds the present discussion insuciently informative is encouraged to look ahead to these items. If is singular and < < + , a multiple of 2 , then the main coding
area for will be a conal subset of which is the range of a function, f .
modified:2004-08-01

f is part of a scale between and + . The domain of f is a xed club subset, D , of the cardinals below , and for each D , f () is of the

form 2 , where is even, 0 < < + . If is a limit of singular cardinals, the D are all singular cardinals, while if is of the form + , the are all of the form , > 1, odd, where = + .

revision:2004-08-01

In a condition, p, in which is mentioned, an initial segment, (, p ()) of ordinals from (, + ) will be mentioned, and a tail of D will be mentioned. We shall require that p () is a multiple of 2 . If < < p (),
a multiple of 2 , then for a tail of D , f () is mentioned in p (i.e., f () < p ()). It is natural to expect, and will, in fact, be true in tidy

340

10

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

conditions that () if p () < + , and is a multiple of 2 , then on a tail of


D , f () is not mentioned in p (i.e. p () f ()).

If () failed, then it might be impossible to extend p to a condition which mentions and which codes correctly at , since the portion of p below may have already imposed an unbounded amount of information on the main coding area for . However, () is quite hard to maintain, when trying to construct an upper bound for an increasing sequence of length = cf of conditions from P . So, rather than require the property, we drop the requirement that p has to code correctly at all . Instead, we allow certain to be deactivated, not used for coding. We inherit another problem though: how to detect
modified:2004-08-01

deactivated ordinals. For this we are led to introduce two auxiliary coding areas. The rst is simply the set of multiples of between and + 2 . This area is used for coding an ordinal hp () . The idea is that not only but all the ordinals in [, hp ()) will also be deactivated. For singular , we have, associated with each such , a function p, with

revision:2004-08-01

domain D and we have that hp () is the least such that f p, ;

in the notation introduced at the end of this section, hp () = scale( p, ). The p, are the generic scale functions, as opposed to the ground model
scale functions, f . We thank the referee for insisting on the point of view

that what we are really doing is forcing a generic scale since the ground

340

[SHST:340]

11

model scale is not adequate for dealing with deactivation. In fact, for singular , hp () is decoded as scale( p, ) rather than being read directly in s . For of the form where > 1 is odd, we also have hp () for multiples, , of 2 which are mentioned in p. Here, however, there is no associated function and the hp () are directly decoded from s . When is a limit of multiples of 2 , the second auxiliary coding area will be a club subset, C . This will be used to help us detect deactivation. The C will be part of a square system between and 2 . Returning to (), we have mentioned that we do require it in tidy conditions, and we require something even stronger in very tidy conditions. In (4.3), we show that the latter are dense. However, by dropping the requirement (), we make it easier to construct upper bounds which might
modified:2004-08-01

not be tidy, as we do in (4.5). In (1.1), we dene games G(, N , p0 ), where > 1 is regular, N is a certain kind of sequence (of length + 1) of models, and p0 P . The two players, GOOD and BAD, alternately pick conditions, pi P . GOOD plays at non-zero even stages, and BAD plays at odd stages. BAD also picks

revision:2004-08-01

a subsequence of N by choosing an increasing sequence ((i)|i < ) from ; of course, (i) is chosen at stage 2i+1. We require that the pi are increasing, and that p2i , p2i+1 |N(i) |. In the cases of interest, N will satisfy a more technical condition, introduced in (1.3), called supercoherence. This will guarantee that at limit stages, we will have the hypotheses of (4.5). GOOD

340

12

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

wins if she succeeds in playing p . BAD wins if at some even stage j , GOOD has no legal move. In (5.1) we prove: Lemma 5. For , p0 as above, and for supercoherent N , GOOD has a winning strategy in G(, N , p0 ). In (5.3), it is argued (using results of (5.2) and [17]) that this gives that P is (, )-distributive. What is at issue here is whether BAD always loses because of his inability to play super-coherent sequences. The results of [17], summarized in (1.4), below, show that this is not the case: there are enough supercoherent sequences. In (1.4), this is presented as a property of the combined squares and scales system, introduced in (1.2).

modified:2004-08-01

SUMMARY AND ORGANIZATION. In 1, we present the coding apparatus for singular cardinals and the related results from [17] notably the result about the existence of supercoherent sequences. In (1.1), we introduce the models sequences and the games G(, N , p0 ). In (1.2) we introduce the combinatorial apparatus of

revision:2004-08-01

squares and scales. In (1.3), we introduce the notion of supercoherence. In (1.4) we state the main result of [17], presented as an additional property of the combinatorial apparatus. In (1.5), we state a small combinatorial result about the system of scales which we use in (4.3). The result is clearly closely related to the denition of very tidy condition. This is also proved

340

[SHST:340]

13

in [17]. 2 takes care of some other preliminaries. In (2.1), we recapitulate some of the material of Lemma 3, above, and (1.2), by giving a complete discussion of coding areas for various kinds of ordinals. In particular, in (2.1.1) we cite an additional result from [17] which shows that, without loss of generality, we can assume that the system of b , for such that card eta is inaccessible has an additional property called tree-like. In (2.2), we give a preliminary idea of the nature of conditions, by introducing the class of protoconditions, P (0). In (2.3), we discuss the phenomenon of contamination at limit cardinals, and the devices for dealing with it, namely the sets X of candidates for coding which are not multiples of . We also introduce the weak deactivator, ?, and the component, p , of conditions
modified:2004-08-01

which provides bounds for contamination. The X are also useful in the context of the strong deactivator, !, which we discuss in (2.4), along with the generic scales. In (2.5), we give a very brief sketch of the decoding procedure, which we complete in (4.6). In 3, we give the formal denition of the class of coding conditions.

revision:2004-08-01

(3.1) recalls some notation, terminology and conventions. In (3.2) we de ne a sub-class P of the proto-conditions of (2.2). These still incorporate none of the sophisticated properties intended to deal with contamination and deactivators. In (3.3), we formally dene the notions associated with contamination, and in (3.4) we cut down P still further by imposing ve

340

14

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

additional properties. The rst four of these deal with contamination. The last deals with the use of the auxiliary coding areas, s , and thus foreshadows (3.5), where we deal with the strong deactivator, !, the generic scales, p, and nally dene the class of coding conditions by imposing four additional properties related to these. In (3.6) we give the (very simple) denition of the partial ordering of conditions. In 4, we prove some basic Lemmas which will greatly facilitate our work in 5 and 6. In (4.1) we introduce the tidy and very tidy conditions. We develop some of their properties in (4.1) and (4.2), and in (4.3) we prove the crucial result that the very tidy conditions are dense. In (4.4) we develop the Factoring Property. In (4.5) we show that certain increasing sequences have least upper bounds. Taken together, (4.3) and (4.5) do most of the
modified:2004-08-01

groundwork for (5.1). In (4.6), we provide a fully detailed discussion of the decoding procedure, completing the sketch of (2.4). In 5, we rst prove Lemma 5, above, in (5.1). In (5.2), we show that the results of [17] really do mean that BAD need not lose for trivial reasons, and in (5.3), we show that this yields the (, )-distributivity of P . We

revision:2004-08-01

close with two remarks, in (5.4). The rst has to do with iterations of P . The second concerns a variant of the games G(, N , p0 ), which we use in case (c) of (6.1)(7). In (6.1), we establish the Extendability properties of P, and in (6.2) we establish the Chain Condition property. NOTATION AND TERMINOLOGY.

340

[SHST:340]

15

Our notation and terminology is intended to be standard, or have a clear meaning, e.g., o.t. for order type, card for cardinality. A catalogue of possible exceptions follows. Also, the index of notation at the end of this section summarize what follows but also some of the important denitions and notation which is introduced in later sections. When forcing, p q means q gives more information. Closed unbounded sets are clubs. The set of limit points of a set X of ordinals is denoted by X . AB is the symmetric dierence of A and B, and A \ B is the relative complement of B in A. Notions like =, , , etc., when decorated with a superscript *, mean on a tail. For ordinals, , [, ) is the half-open interval { : < }. The notation for the other three intervals is clear. It should be clear from context whether the open interval or the ordered pair
modified:2004-08-01

is meant. For ordinals , , we write >> to mean that is MUCH greater than ; the precise sense of how much greater we must take it to be is supposed to be clear from context. For innite cardinals, , H is the set of all sets hereditarily of cardinality < , i.e. those sets x such that if t is the transitive closure of x, then

revision:2004-08-01

card t < . We regard as a successor cardinal, by ignoring the positive nite cardinals. Thus, for us, = 0+ . We say that a cardinal is slike if it singular or of the form where > 1 is odd, and that it is i-like if is inaccessible or of the form where > 0 is even. For s-like cardinals, , we dene U () to be the set of multiples of 2 in (, + ),

340

[SHST:340]

11

while for i-like cardinals, we dene U () to be the set of multiples of in (, + ). We dene E to be the class of ordinals, , such that letting = card , U (), is regular and either is inaccessible or ( is s-like and is an even multiple of 2 ). For models, M, SkM denotes the Skolem hull operator for M, where the Skolem functions are obtained in some reasonable xed fashion. We often suppress mention of the membership relation as a relation of a model, but it is usually intended that it is one. Thus, (M, A) frequently denotes the same model as (M, , A). When we have a -increasing sequences of functions ( | X), where X is a set of ordinals, and is a function which is one of the , we let scale() denote the least X such that . All other notation is
modified:2004-08-01

introduced as needed (we hope). 1. SINGULAR COMBINATORICS: RESULTS FROM [17].

(1.1) MODEL SEQUENCES AND THE GAMES G(, M, p0 ). Let > 1 be regular. Let M = (H + , , , ), where is a

revision:2004-08-01

singular cardinal, >> and (H , ) models a suciently rich fragment of ZFC. Let and let (Ni : i ) be an increasing continuous elementary tower of elementary substructures of M. We say that (Ni |i ) is (M, )-standard of length + 1 if, letting Ni := |Ni |, for all i , card Ni = , + 1 N0 , for i < , [Ni+1 ]<

Ni+1 and, if

340

12

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

i is even, Ni Ni+1 . Let Q be a partial ordering (of course, in 5, Q will be P , the upper part of P at ). Let X be dense in Q (in 5, below, X will be the dense subclass of very tidy conditions, see (4.1) and (4.3)). Let N = (Ni |i ) be -standard with each Ni M (in 5, below, N will be super-coherent (see below)),

and let q0 Q (:= |Q|) M (:= |M|). The game G(, N , Q, X, q0 ) is dened as follows. Two players, GOOD and BAD alternate plays. GOOD plays at positive even stages (including limit stages); BAD plays at odd stages. GOODs moves are conditions, q2i Q M , where 0 < i . For 0 i < , BADs move at stage 2i + 1 is a pair, (q2i+1 , (i)), where q2i+1 X, q2i q2i+1 , (i) > sup {(j)|j < i}, q2i , q2i+1 N(i) .
modified:2004-08-01

We require that at all stages , (qi : i ) is increasing. BAD loses if GOOD succeeds in playing q . GOOD loses if at some stage i , she has no legal move, i.e., there is no upper bound to the sequence (qj : j < i). Of course, this can only occur if i is a limit ordinal.

revision:2004-08-01

We have already hinted at the diculty for GOOD at limit stages in the discussion in the Introduction, preceding the statement of Lemma 5. (1.2) THE SQUARES AND SCALES.

From [17] (and for singular cardinals of the form + , using Lemma 3 of

340

[SHST:340]

13

the Introduction, above, as well), we have the following combinatorics for singular cardinals. (A) A Square on Singular Limits of Limit Cardinals: we have a sequence, (D : is a singular cardinal), where D is a club subset of the singular cardinals below satisfying: (1) o.t. D < min D , (2) if is a limit point of D , D = D . (3) if D is not a limit point of D then is not a limit of limit cardinals. (4) suppose that Di , i = 1, 2, and let ji be such that is
th the ji member of Di . Then, j1 = j2 .

modified:2004-08-01

If is not a successor ordinal, and = + , conventionally, we let D := { +n |n is odd, + n = 1}. For such we set := D . For which are singular limits of limit cardinals we set := {{} D | D }.

(B) Squares on (U ()) + , where is a singular cardinal for each such , we have a sequence (C | (U ()) + ) such

revision:2004-08-01

that each C is a club subset of the set of even multiples of 2 below , of order type less than , and such that if C but is not a limit point of C , then is not a limit point of U (), and with the usual coherence property: if is a limit point of C , C = C . (C) Scales on (, + ), where is a singular cardinal):

340

14

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5


for such , we have a sequence (f : U ()), where dom f = D , for D , f () is an even multiple of 2 and: (1) if < < , , U () then f < f , i.e., for some 0 < , whenever D \ 0 , f () < f (); further, if

C , then the preceding holds for all D , (2) whenever g is a function with dom g = D and for all
D , g() < + , for some U (), g < f ,

(3) if is a singular limit of limit cardinals, D , U (), =


f () and D , then f ( ) = f ( ), and if is

not a limit of limit cardinals and , U (), D and


f () = f (), then f | = f |, (4) for limit points, , of U (), and D , (, ) := {f ()|
modified:2004-08-01
C } is a nal segment of Cf () ; further, on a tail of D , (, )

has limit order type. Regarding (3), the property given in the second clause follows from the property given in the rst. Unfortunately, we needed two dierent clauses,
since we do not have any f where card is a successor cardinal. However,

revision:2004-08-01

the property of the second clause of (3) in fact allows us to dene these according to the following convention. Once this is done, in virtue of this denition, we will have the property of the rst clause of (3) even for which are not limits of limit cardinals: suppose that = , where > 1 is odd. Let = + . Suppose

340

[SHST:340]
that = f () for some U (). For D , we dene f ( ) to be f ( ). By the second clause of (3), this does not depend on

15

our choice of . Property (4) is the crucial condensation coherence property. It plays an important role in the proof, in [17], of the existence of super-coherent sequences. We state this in (1.4), below, as an additional property of the above combinatorial system, (A) - (C). Strictly speaking, we never appeal directly to (4), only to the property of (1.4), but we do appeal to the following more obvious consequence of (4):
(4 ) on a tail of D , is conal in f ().

We close by stating the decodability property of the above system. As


modified:2004-08-01

usual, A is as given by Lemma 3 of the Introduction.

(D) Decodability of (A) - (C): for all singular , D and the systems
(C | < + is a limit point of U ()), (f | U ()) are canoni-

cally denable in L[A ].

revision:2004-08-01

The decodability property is an easy consequence of the fact that the systems of (B) and (C) are rather simple modications of systems which are canonically constructed in L, for singular limits of limit cardinals, and for which are not limits of limit cardinals, in L[A ], while the system of (A) is a simple modication, also given in [17], of a constructible system.

340

16

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(1.3) COHERENCE AND SUPERCOHERENCE. Let > 1 be regular. Let > cf >> be such that (H , ) |= a suciently rich fragment of ZFC. Let M = (H + , , ). Suppose that N M, where, letting N := |N |, card N = , and let be a cardinal

with , N . Let N () = sup(N (, + )). Recall that an Easton set of ordinals is one which is bounded below any inaccessible cardinal. For such N and singular cardinals, , with < , we say that is N controlled if there is an Easton set d with d N . The Easton sets we have in mind are those consisting of the sets of cardinals mentioned in some condition in N . We dene pN , an analogue of N , dened on all singular cardinals, , which are N controlled. The denition makes sense for all cardinals
modified:2004-08-01

[, ], but we will only use it for the singulars which are N controlled. If N , then of course is N controlled and in this case, pN () := N (). Otherwise, pN () := sup (+ SkM ({} N )). The reason that we only consider controlled is that one of the results of [17] gives an alternative characterization of pN () which is central in

revision:2004-08-01

proving the main result about the existence of supercoherent sequences (see below). The alternative characterization is equivalent only for controlled . The restriction to such is benign, for our purposes, since it allows us to handle any cardinal mentioned in any condition in N . This is the essential use, alluded to in the Introduction, just prior to the DISCUSSION sec-

340

[SHST:340]

17

tion, of the fact that the set of cardinals mentioned in any condition is an Easton set. Now, let (Ni | i ) be (M, )-standard of length + 1. For i , let i = Ni , pi = pNi . Let N = N = , p = p , so dom = {Ni : i < }, and let =

{dom i : i < }, and for dom , =

sup {i () : Ni }. Also, for singular cardinals, [, ], which are N -controlled, p() = sup {pi () : i < & is Ni -controlled}. Let be a singular cardinal, dom . Note that since cf = > , there is a club D such that for all i D, i () C() . This motivates the following.

Denition.
modified:2004-08-01

Let M, be as above, and let (Ni | i ) be (M, )-standard

of length + 1. Let N = N . Let N, Ni , , p, i , pi be as above. Let be a singular cardinal, N . (Ni : i ) is M-coherent at i for all limit ordinals with N , there is a club D such that for all i D, i () C () . (Ni : i ) is M-coherent if for all singular cardinals N \ , (Ni : i ) is M-coherent at . (Ni : i )

revision:2004-08-01

is strongly

M-coherent i for all i < and all singular cardinals

Ni , i () C() . Finally, (Ni : i ) is super M-coherent i (Ni : i ) is strongly M-coherent and for all limit ordinals, and all singular cardinals, which are N -controlled, for suciently large i < , pi () Cp () .

340

18

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(1.4) THE EXISTENCE OF SUPER-COHERENT SEQUENCES. Here is the statement of the main result of [17] which is the crucial additional property of the combinatorial system of (1.2). Lemma. Let , , M be as in (1.3). Let C [H + ] be club. There

there is super M-coherent (Ni |i ) with each |Ni | C.

(1.5) AN ADDITIONAL RESULT ABOUT THE SCALES. The following small combinatorial result concerning the scales of (1.2) is also proved in [17] and will be quite useful in (4.3), below. Proposition. Let > 1 and let , M be as in (1.4). Let d [, ) be an Easton set of cardinals, and let be a function with domain d such that for
modified:2004-08-01

all d, () < + . Then, there is a function with domain d such that for all s-like d, () > () and such that for all singular d, letting
= (), f = |D . Further, if N

M with ( + 1) {} |N |,

then |N |. 2. PRELIMINARIES ABOUT CONDITIONS.

revision:2004-08-01

(2.1) CODING AREAS FOR (, + ). We recapitulate, here, some of what we have done in Corollary 4 of the Introduction and (1.2), and provide some insight into how the coding will work. First, suppose that , fall under one of the following cases.

340

[SHST:340]

19

(1) (2) (3)

2 is a successor cardinal, (, + ), is inaccessible, U (), is a singular cardinal, U ().

Then, by Corollary 4 of the Introduction, for cases (1), (2), and by (1.2), for case (3), we have associated to an unbounded subset b . In cases (1) and (2), this is the coding area for . In case (3), it is the main coding area for , but we also have one, and sometimes two auxiliary coding areas for as well, see below.

(2.1.1) let U :=

In case (2), we shall need an additional property of the b . So, {U ()| is inaccessible}. We say that the system (b | U ) is

tree-like i whenever 1 , 2 U , if b1 b2 , then b1 = b2 .


modified:2004-08-01

In [17] we also prove the rather simple observation that without loss of generality, we can assume that (b | U ) is tree-like and has the following additional property: b = range g , where g is a funtion, dom g = { |t au < card & is an i-like successor cardinal}; further, for all b , is a multiple of 4 but not of 8.

revision:2004-08-01

In case (1), if = + , it is easy to see that we can, without loss of generality, assume that the b have the following additional properties: b = and the members of b are even ordinals but not multiples of 4; further, if is s-like, then the members of b are never of the form + 2, where E.

340

20

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(2.1.2)

In case (3), (1.2) already gives us b which have the following

properties. Once again, the b are ranges of functions, f , with domain

D . Case (3) subdivides according to whether is a limit of singular cardinals, or of the form + . In the rst subcase, D is a club subset of singular cardinals below , whose order-type is less than its least element. In the second subcase, D is the set of such that > 1 is odd and such
that = + . In both subcases of case (3), the f () are even multiples

of 2 , i.e., they are of the form 2 where > 0 is even. (2.1.3) Recall that a cardinal is s-like if is singular or = , where > 1, is odd. If is s-like, U (), we set s := the set of multiples of in (, + 2 ); s is an auxiliary coding area for discussed in in the Introduction, above, and at greater length in (2.4), (3.4) (E) and (3.5),
modified:2004-08-01

below. Finally, if is singular and (U ()) , we have an additional auxiliary coding area for , namely C , from (1.2) (and so also, implicitly, all of the s for C ). This will be used for detecting deactivation of in a way that is rather important for the limit case of GOODs winning strategy in the games of (1.1), and for determining p, . This will also be

revision:2004-08-01

discussed more fully in (2.4), below. It should be noted that unlike the previous coding areas which are essentially unique to , this last is not, since if (C ) then this coding area for is an initial segment of this coding area for , and if (C ) then this coding area for is a subset of this coding area for .

340

[SHST:340]

21

It is worth recalling that if is a limit of singular cardinals and D is not a limit point of D , then is not a limit of singular cardinals, while if is a limit point of D then D = D . It is also worth recalling (3)
of (1.2), and the related convention whereby we regard f as dened when = f (), U ((card )+ ),and card is s-like but regular.

(2.1.4)

For regular and as in (1) or (2), above, b will be used for

coding as follows. If, on a tail of b , we read value 1, then we will decode value 1 for . Any condition will mention at most a bounded subset of b , so we will guarantee a tail of 1s on b by making promises of the form (, ), where < . Such a pair is a promise to have value 1 at all members of b above . By a density argument, (6.1) (2), except for is inaccessible in a bounded subset of U () (the bound is p (), see (2.2), (2.3)), we will
modified:2004-08-01

have made such a promise whenever gets value 1. If is a successor cardinal, and we do not have value 1 on a tail of b , then we will have value 0 on an unbounded subset of b . This is also by a density argument, (6.1) (5). We then decode value 0 for . This will essentially be the procedure when is inaccessible, again, except for a

revision:2004-08-01

bounded subset of U (). The situation regarding the in the bounded set will be discussed more fully in (2.3), and (3.3), (3.4). (2.1.5) For singular cardinals , and as in case (3), the situation is more complicated. Here, any condition which mentions will mention a tail of b . In the simplest situation, we will have an i {0, 1} and a tail of b on

340

22

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

which we have value i. It is natural to expect that when this occurs, we will decode value i for . However, it could still occur that is deactivated, and that we therefore decode value !, or sometimes value ? for . We discuss this in (2.4) and (3.3) - (3.5). If there is no such tail, then either we will decode value ? or value ! for . Again this will be discussed more fully in (2.3), (3.3) - (3.5). (2.2) PROTOCONDITIONS. We dene P (0), the class of protoconditions. Denition. p P (0) i p = (g, , ) = (g p , p , (p)), and (2.2.1) (2.2.5), below, hold; g is the main component, the approximation to the class function G which are seeking to add (and code down to a subset of
modified:2004-08-01

3 ). (2.2.1) There is an Easton set, d = dp , of cardinals 2 , and a function, = p with dom = d, such that for all d, < () < + and we will have g : dom g {0, 1, ?, !}, with dom g = {(, ())| d}.

d will have the following additional property: for singular cardinals d,

revision:2004-08-01

there is a tail of D d. In addition to the usual characters, 0, 1, we have the strong deactivator, !, and the weak deactivator, ?, whose roles will be discussed in (2.3), (2.4), (3.4) and (3.5). (2.2.2) g() {0, 1} unless is singular and U (). However, we

have a convention for systematically abuse of notation for certain .

340

[SHST:340]

23

(2.2.3)

Recall that E i, letting = card , > 1 is regular,

U () and either is inaccessible or is s-like and is an even multiple of 2 . In either case, for i = 0, 1, we take g() = i as an abbreviation for (g(), g( + 2)) = (0, i), and we take g() = ?, as an abbreviation for (g(), g( + 2)) = (1, 0). In the second case only, we take g() = ! as an abbreviation for (g(), g( + 2)) = (1, 1); thus it is our intent that we can have g() = ? for any E, but that we have g() = ! only for those E whose cardinalities are s-like. (2.2.4) is a function with dom = d, such that for d, <

() (). For successor cardinals, d, () = + 1. The role of () for limit cardinals will be made clearer in (2.3) when we discuss contamination. For now, we will just say that () is a bound on the
modified:2004-08-01

contamination in (, + ), not only in p, but in all stronger conditions, q. For cardinals, d which are either s-like or inaccessible, we will have that (), () U (), and if is inaccessible, we will have that () 2 . (2.2.5) Finally, is the system of promises which we discussed in

(2.1), above. is a set of ordered pairs (, ) such that g() = 1, card is

revision:2004-08-01

regular, 2 < < card and if card is inaccessible then ( (card ) and U (card )). We shall also require that if card = + then > . Let W (p) = dom (p). We let R(p) := {b \ )|(, ) (p)}. We then

require g() = 1 for all R(p); thus (, ) is the promise to put all 1s in b from on.

340

24

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(2.2.6) In 3 we will build to the denition of P , by imposing additional restrictions on the protoconditions. If > 2 , is regular, then we shall dene P in (4.4). It is only slightly inaccurate and not at all misleading, at this point, to say that the main idea is that d = . The real point is that P is the class of conditions for coding down to a subset of + . The partial ordering of protoconditions is dened in the most obvious way: p q i g p g q , p q , and (p) (q). This is identical to the denition of the partial ordering of conditions, in 3.

(2.3) X , CONTAMINATION, () AND THE DEACTIVATOR, ?. In (2.1), no coding areas were dened for such that = card is a limit cardinal, and is not a multiple of . Strictly speaking, for singular
modified:2004-08-01

and which are multiples of but which are not in U (), there was also no coding area dened, but, except for the multiples of [, 2 ), these ordinals are in s , where is the largest member of U () below . The multiples of in [, 2 ) are simply ignored. This is because such ordinals, , are not coded directly. Instead, each

revision:2004-08-01

such has a set, X , of surrogates, for coding . Each of the surrogates, , will have a coding area b associated with it. X will have size + , so we have many tries at coding correctly. When is singular, this is not entirely unexpected, since possibly some of the surrogates have been deactivated with the strong deactivator, !, as in the discussion in the In-

340

[SHST:340]

25

troduction leading up to Lemma 5. Here we discuss the weak deactivator, ?, and the phenomenon of contamination which is one of the contexts in which it arises. The reasons for calling ? weak and ! strong are discussed at the beginning of (2.4) and in (2.3.5), below, where we also discuss another context in which ? arises, for singular . Before doing this, we present the X . (2.3.1) For limit cardinals, , and (, + ) which are not multiples
+

of , we have sets, X [U ()] . If is singular, the X are all odd multiples of 2 , i.e., of the form = 2 , where is odd. If is singular, the system (X : (, + ), 0 (mod )) L, while if is inaccessible, each X = X + , where the X are classes of ordinals and the relation X is canonically 1 denable over L. When is inaccessible, we
modified:2004-08-01

shall also require the following property of the X : (*) if < < < + and is a cardinal in L, then = sup(X ). Finally, for inaccessible , we take the X to partition U (), while if is singular, we take the X to partition the set of odd multiples of 2 in

revision:2004-08-01

(, + ). (2.3.2) Contamination is most easily understood in the context of inaccessible . For such and U (), it can occur that for some inaccessible > and some U ( ), b b is unbounded in . Further, it could also occur that in some condition p, g p ( ) = 1, and in fact that

340

26

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

( , ) (p) for some < . This will either prevent us from having g p () = 0 or from coding this correctly. When, for other reasons, we are required to have g p () = 0, is said to be contaminated (by ) in p. We cannot prevent such contamination, but we will dene conditions in such a way (see (3.2) (A), below) that (): fewer than many (, + ) are contaminated. Typically, contamination occurs here because was added to dp after the promise ( , ) had already been made. (2.3.3) When is singular, we shall also have the phenomenon of con-

taminated ordinals. It may occur, for singular , and conditions, p, with dp , that for some U () p (), one of the following holds:
modified:2004-08-01

(1) There are x1 = x2 and Y1 , Y2 , conal subsets of b , such that for i = 1, 2 and Yi , g p () = xi , (2) i {0, 1}, and for other reasons we are required to have g p () = i, but on a conal set of b , g p () = 1 i (see (3.3) below, the

revision:2004-08-01

denition of forced to be i, for what these other reasons are).

If g q () = !, then will be deactivated anyway. If, however, this fails, then is contaminated in q. Once again, contamination will occur only for a bounded set of , though here this is a simple observation which does not require a special property of the conditions, as in the inaccessible case. Here

340

[SHST:340]

27

again, typically, contamination arises due to the fact that an unbounded set of information below was part of a condition before was mentioned. In both the inaccessible and the singular case, p () is the sup of the contaminated ordinals (, + ) (see (3.4) (C)). Once p () has been specied in a condition, no further contamination is allowed in any stronger condition, q, since q () = p (). In both the inaccessible and the singular case, we require, (3.4) (B), that if is contaminated then g p () = ?. (2.3.4) We can now specify how the X are used to code . If

dom g p , then we will have g p () {0, 1}. In the inaccessible case, we will have that if X \ p (), then g p () = g p () (see (3.4) (A), and one clause of the denition of forced to be i). In the singular case, things are somewhat more complicated, since even for X \ p (), we
modified:2004-08-01

can have g p () {?, !}. However, as part of the denition of condition ((3.4) (A), again), we will have that for such , g p ()=1 g p (). We will show, by a density argument, (6.1) (4), that when I is generic and G is the union of the g p for p I, there will be a conal set of X such that G() = G(). Thus, in decoding, there is a common denition: decode for

revision:2004-08-01

the unique value i {0, 1} such that we have value i on a conal subset of X . (2.3.5) To conclude, we should mention the other way the weak deactivator, ?, can occur. For singular , in addition to occurring at contaminated , it can occur at other U (), but only if on a tail of b the value ?

340

28

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

occurs. The reason for this has exactly to do with the density argument we just mentioned. As will become clearer in (2.4), the strong deactivator at can contribute to deactivating larger ordinals. This is not the case for the weak deactivator, ?. Thus, the weak deactivator, ?, can play the role of safe, neutral ller, and does not present the potential danger of forcing us to deactivate ordinals we want to preserve as active, to get value in {0, 1}, such as the conally many X we need for the preceding.

(2.4) THE STRONG DEACTIVATOR, !, AND THE GENERIC SCALES. Suppose that is singular and U (). We have already mentioned most of the elements of this discussion: (1)
modified:2004-08-01

when g p () = !, not only , but also all members of U () in the open interval (, hp ()) are deactivated (if g p () = ! and hp () = , itself is still deactivated); this is one of the senses in which ! is the strong deactivator.

(2) (3)

hp () = scale( p, ). if g p () = ! this can contribute to making g p () = !, for certain larger U (); this is the other sense in which ! is the strong deactivator.

revision:2004-08-01

(4)

when is a limit point of U (), C is an additional, auxiliary coding area used for detecting strong deactivation.

(5)

detecting strong deactivation and lengths of deactivated intervals

340

[SHST:340]

29

(by (2), above, this amounts to the same thing as decoding the p, ) is one of our major preoccupations.

(2.4.1)

We now put these elements together and lay the groundwork for

(3.5), omitting, for now, some of the ner points related to certain p () for which p, will nevertheless be dened. We should say, at the outset, that p, will be dened whether or not we end up having g p () = !, but that this is just for convenience, since the only case in which it has any signicance is when this occurs; when g p () = !, we ignore p, and take hp () to be . As we have already mentioned, we are grateful to the referee for emphasizing the point of view that the p, are really potential members of generic scales which we are forcing as we do the coding. In almost all
modified:2004-08-01

cases, we will have p, f ; the exceptions are discussed in (3.5), (3.6).

(2.4.2)

We will have two other functions, p, and p, and that, in

most cases, for D dp , we take p, () := max( p, (), p, ()). Looking at p, amounts to considering what happens from below, on b . Looking at p, amounts to considering what happens to the left, on

revision:2004-08-01

C . These are two of the ways in which could be strongly deactivated, and are two of the places we have to look to detect strong deactivation. (2.4.3) Before developing this, however, there is a third way in which

can be strongly deactivated, and we deal with this rst, since it is simplest, and directly related to (1), above. is p-interval-strongly-deactivated if it

340

30

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

is in a deactivated interval, (, hp ()), for some p () < . When this occurs, we take least possible and set p, := p, , without considering the p, , p, . Thus, when is p-interval-stronly-deactivated, only this counts, even if it turns out that it is also deactivated in one of the two other ways we now discuss. In terms of Lemma 4.3, this corresponds to the between and the tp () of (4.2), and, roughly speaking, to limit stages 2 of GOODs winning strategy. (2.4.4) is p-strongly-deactivated on b (from below) i on a tail of

b , g p () = !. Typically, this occurs when didnt have to be deactivated, when we are strongly deactivating intentionally, to be sure that we are able to strongly deactivate other, larger, which will be more problematical, see the discusssion of () in the Introduction. This corresponds
modified:2004-08-01

to some of the work in (4.3) (beyond the tp (), of (4.3)) and all of the work 2 of (6.1), and, roughly speaking, to successor stages in GOODs winning strategy. (2.4.5) is p-strongly-deactivated on C i it is a limit point of U ()

and on a conal subset of C , g p () = !. Typically, this occurs in

revision:2004-08-01

situations where we really needed to deactivate and we are happy to nd that we prepared for this by strongly deactivating enough members of C . This corresponds to the portion of the work in (4.3) dealing with p () and to the situation of = () in (4.5), and roughly speaking, to limit stages in GOODs winning strategy.

340

[SHST:340]

31

(2.4.6)

It remains only to give the main idea of the denitions of the

p, and the p, (there are some ne points which can be deferred until the ocial denition in (3.5)). The main idea for the p, () is that this
should be hp (f ()). The ne points arise when f () p (). The main

idea for the p, () is that this should be sup { p, ()| C }. The ne


points arise because we want this sup to be f (), but p ().

(2.5) OVERVIEW OF THE DECODING PROCEDURE. Let = A be the (class) characteristic function of A. Our forcing will produce a generic class function G with domain OR and range {0, 1, ?, !}. We will code A into G on odd ordinals, i.e., we shall have that for non-successor ordinals and n < , G( + 2n + 1) = ( + n).
modified:2004-08-01

Of course, we want to recover G from G|3 by decoding. This is done by recursion on cardinals, . The basic recursion step is to go from G| to G|(, + ), when CARD. This will involve a nested recursion across (, + ). The procedure for obtaining G|(, + ) from G| will be uniform within each of the following classes of cardinals: inaccessibles, singulars,

revision:2004-08-01

and successors. Thus, at limit cardinals, , we can piece together G| from the G|, < , and continue. The recursion step for successor cardinals is provided by (2.1.4). As noted there, for inaccessibles, this also essentially gives the way we obtain G0 , which we now discuss. For limit cardinals, , it will simplify matters if, in decoding G|(, + ),

340

32

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

we have available not only G|, but also an auxiliary function, G0 , which represents the rst stage in dening G|(, + ). The role of G0 can best be understood by discussing the broad outline of how we nally obtain G|(, + ). For inaccessible cardinals, this is atwo-pass process. For singular cardinals, it is a three-pass process. The rst pass involves decoding the information provided by G| on b without regard to the analogous information for the (, ). G0 represents the outcome of this rst pass. For inaccessibles, even this rst pass involves a recursion, since we have to decode the b as we go. For singulars, however, there is no recursion involved in the rst pass, but there denitely is a recursion involved in the second pass for singulars, where we deal with the strong deactivator, !, and the generic scales. The second pass
modified:2004-08-01

for inaccessibles and the third pass for singulars are analogous, in that this is where we deal with contamination, and dene G on the non-multiples of (, + ).

3. THE CODING CONDITIONS: DEFINITIONS.

revision:2004-08-01

We build to the denition of the class of coding conditions, P, in (3.5) (3.6). In our original treatment we had stronger properties, which appear below as (4.1) (A) and (B+), in place of (3.5) (C) and (D). The latter are technical weakenings of the properties of (4.1), which are designed to allow us to prove, in (4.3), that the very tidy conditions, those with the properties

340

[SHST:340]

33

of (4.1) are dense.

(3.1) We recall some terminology and conventions from the Introduction and (2.2). A cardinal is s-like if it is singular or of the form , with > 1 and odd. Next, let be a regular uncountable cardinal and let (, + ). Recall that E if U () and either is inaccessible or is s-like. Formally, for conditions p and E we shall have g p () {0, 1}, but recall the convention from (2.2.3) involving the use of + 2 as an extra bit for E. Naturally, we have taken care not to assign any other coding duties to the + 2 where E.

modified:2004-08-01

(3.2) Denition. Suppose p = (g, , ) = (g p , p , (p)) P (0), where P (0) is as in (2.2). p P i the following properties, (A) and (B) are satised.

(A) For all regular , there are fewer than many such that for some

revision:2004-08-01

< , (, ) (p) (note: need not be a member of d), (B) If dom g p , = card is singular and is a multiple of 2 :
(1) f < p |D ,

(2) if g p () {0, 1}, then on a tail of b , g p () = g p (), (3) if p () and g p () = ?, then on a tail of b , g p () = ?.

340

34

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(3.3) Suppose p P . First, consider such that = card is a limit cardinal, and suppose that X . We say that g p () is forced to be 0 (resp. 1) if g p () = 0 (resp. 1). We also say that g p () is forced to be 1 if for some R(p), X . Finally, drop the restriction on . If = 2 + 1, then we say that g p () is forced to be 0 (resp. 1) if A (resp. A). If = card is inaccessible and is a multiple of , then is contaminated by if W (p), (, ) (p) for some < , b b is conal in and g p () is forced to be 0; is contaminated i for some it is contaminated by . Because the system of b is tree-like for U (see (2.1.1)), it is easy to see that any W (p) contaminates at most one (, + ). Therefore, (3.2) (A) gives that there are fewer than many
modified:2004-08-01

(, + ) which are contaminated. If = card is singular, dp D is conal in , then is contaminated i is a multiple of 2 , g p () = ! and one of the following holds: (1) there are x1 = x2 and conal subsets Y1 , Y2 b dom g p such

revision:2004-08-01

that for Yi , g p () = xi , (2) g p () is forced to be 0 (resp. 1) but on a conal subset of b dom g p , g p () = 1 (resp. 0). Here, it is easy to see that at most many (, + ) are contaminated, since if > scale( p |D ) then cannot be contaminated. Also, note that

340

[SHST:340]

35

which are contaminated because of (2) are odd multiples of 2 since they are members of some X .

(3.4) Denition. If p P , then p P i the following properties (A) - (E) hold.

(A) If g p () is forced to be i and g p () {0, 1} then g p () = i, (B) If dom g p and is contaminated, then g p () = ?, (C) For limit d, p () = sup { (, + )| is contaminated},
p (D) If dp , is inaccessible, we also dene 1 () := sup {

(, 2 )| is contaminated} and we require:


p p (1) g p (1 + ) = 0, g p (1 + ) = 1, for all limit ordinals
modified:2004-08-01

< ,
p (2) g p |(1 + , 2 ) codes a well-ordering of in type p () p on odd successor multiples of in (1 () + , 2 ), and codes p A p () on even successor multiples of in (1 () + , 2 ).

(E) Suppose d is s-like. Suppose that p () < p (), with

revision:2004-08-01

a multiple of 2 . Let ( , ) denote the Gdel pairing function. o Let H p () := {(, ) |g p ( + (1 + (, ))) = 1}. We then require that H p () is a well-ordering of a subset of which lies in L[A ], and, if is singular, we also require that it is the <L[A] -least well-ordering of a subset of in its order type.

340

36

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

We let hp () := the least multiple of 2 the order type of H p (). We further require: (1) hp () , (2) if is singular, p () < , and is a multiple of 2 then hp () hp (), and if hp () , then hp () = hp (), (3) If g p () = !, then we require that hp () has the smallest value consistent with (1) and (2) (which, it will be easy to see, from (3.5) (A), will be ).

(3.5) (3.5.1) Denition. Fix p P and singular dp . Suppose that U () \ p (). We are mainly interested in the case where p (), but it
modified:2004-08-01

will be useful to have the denition in the more general context. This results in somewhat more complicated denitions; we will also give the simpler denitions that result when we restrict to p ().
Let g = g p , h = hp , f = f , = p (). We rst dene some additional

functions, p, , p, , p, , with domain D dp .

revision:2004-08-01

First, for dom p, , if f () dom g, we set p, () = hp (f ()); otherwise, p, () = p (). Note that if dom g, then on a tail of D d, p, () = hp (f ()) f (). Thus, if it is not the case that f p, then p () and there is no tail of b dom g. We now dene p, , p, by simultaneous recursion on ; at the same

340

[SHST:340]

37

p p p time we dene three properties, P r1 , P r2 , P r3 , by dening, by recursion p p on when P ri () holds. We use P r p () as an abbreviation for P r1 () or p p P r2 () or P r3 (). We say that is p-interval-strongly-deactivated p i p () and P r1 () holds. We say that is p-strongly-deactivated p on b i p () and P r2 () holds. Finally, we say that is p-strongly-deactivated p on C i p () and P r3 () holds. We say that is p-strongly-deactivated

i it is p-strongly-deactivated on b or it is p-interval-strongly-deactivated or it is p-strongly-deactivated on C . Thus, is p-strongly deactivated i p () and P r p () holds. We turn, now, to the recursive denition of the two above-mentioned
p functions, and the three properties. P r1 () holds just in case there is p U () [, ), such that P r p () holds and scale( p, ) > . If P r1 ()
modified:2004-08-01

holds, let be the least witness to this. In this case, we set p, = p, , and p, := p, .
p Thus, for the denition of the two functions, we can assume P r1 () fails. p, p, In this case, for dom p, , 1 () := min( p (), f ()), 2 () := p, p, sup { p, ()| C }) and p, () := max(1 (), 2 ()). Finally,

revision:2004-08-01

for dom p, , p, () = max( p, (), p, ()).


p p We conclude by dening when the other two properties, P r2 (), P r3 () p p hold. P r2 () holds i on a tail of D , P r p (f ()) holds. P r3 ()

holds i is a limit of multiples of 2 , and Z is conal in , where Z = { C |P r p () holds}.

340

38

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5


p Of course, more than one of these may be true for . However, if P r 1 (),

only this counts, in terms of how p, is dened. It is also possible that, letting = p (), ) is p-deactivated. This is clear in the case of interval deactivation and deactivation on C . Deactivation on b is only possible if a tail of b dom g. The simplications which arise when we restrict to p () are mainly
p p that we can remove the denitions of P r2 () and P r3 () from the recursion

which gives us the denitions of p, and p, , by changing the denition


p of P r2 () to be: on a tail of D , f () dom g & g(f ()) = !, and p for P r3 (), by changing the denition of Z to: { C |g() = !}. The

reasons will be clear from (A), below. We can also drop from the denition
p of P r1 () the requirement that P r p () holds.
modified:2004-08-01

A disquieting possibility is that P r p () holds for all U () \ p (). In Remark 2 of (4.3) we shall show that this cannot occur. We are now ready for the denition of P . (3.5.2) Denition. p P i p P , property (D), below, holds, and

revision:2004-08-01

whenever , , etc., are as above, and = p (), the following properties (A) - (C) hold:

(A) If < , then g() = ! i is p-strongly-deactivated, (B) If < , and g() = !, then hp () = scale( p, ),
(C) If ( p |D f ), then is p-strongly-deactivated, p |D p, ;

340

[SHST:340]

39

further, letting = scale( p, ), whenever U () with < < , if b dom g p is conal in , then on a tail of b dom g p , hp ()
f (card ),

(D) If dp is s-like, the following set has power :

{ p, ()| p, () is dened}.

Remark. The substantive part of (D) concerns those p, () which are > p (). As indicated at the beginning of this section, our original denition of P required that all the p, () p () and that, with the notation of (C),
above, p |D = f . Instead, we have opted to relax this requirement and

show, in 4, that these properties hold on a dense set. With this in mind,
modified:2004-08-01

(D) is clearly a necessary condition to be able to extend p to a condition with these properties. (C) is a technical property, formulated with the same aim.

(3.6) Denition.

revision:2004-08-01

If p, q P , we set p q i g p g q , p q , (p) (q). Remark 1. Note that if p q then hp hq . Note, also, that if
p, q, p (), dom g p U () then p, = q,, i = i , i = 1, 2,

and therefore p, = q, and p, = q, . It is also easy to see that if


p, q, = p (), then 2 = 2 . It is possible that q, () > p, (); this will

340

40

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

p, q, occur exactly when p () < f () < q (). Similarly, 1 () < 1 () just in case p () < f () < q (). Thus, we could have q, () > p, ()

on a tail of . It is also clear that is p-interval-strongly-deactivated just in case it is q-interval-strongly-deactivated, and similarly for deactivation on C . However, it is possible that is q-strongly-deactivated on b without being p-strongly-deactivated on b . The situation is similar for U () \ + 1. It is easy to see that if
p q p, q, P ri () holds then P ri () holds, and that p, q, , i i and

therefore that p, q,, p, q, . Remark 2. In virtue of (3.5)(B), above, letting = p (), we dene hp () by hp () := scale( p, ). Remark 3. Let = p (), = scale( p |D ), and suppose that U ()
modified:2004-08-01

. Note that this occurs exactly when U (), < and b dom g p is conal in . Thus, for such there is already an unbounded set of information imposed by p on b , which might require us to deactivate , and the question arises of how far this deactivation should go. However, if this occurs, then we have the hypotheses of (3.5)(C), above, and so is

revision:2004-08-01

p-strongly-deactivated. Further, since (3.5)(C) gives us that p |D p, ,


p and < , < scale( p, ). Thus, P r1 () holds. Suppose, now that

q p and < q (). By Remark 1, above, p, q, . Thus, in such q p, will already be q-interval-strongly-deactivated by . Finally, since < scale( p |Dk ) and, by hypothesis, b dom g p is conal in , on a tail of

340

[SHST:340]

41

b dom g p , hp () f (card ). Thus, as far as such are concerned,

already provides the essentials of the deactivation information. 4. THE CODING CONDITIONS: BASIC LEMMAS.

(4.1) Denition. If p P, p is tidy i for all s-like dp , (A), below, holds and for all singular dp , (B), below holds. (A) if U () with p () < p (), then hp () p (),
(B) p |D f , where = p ().

If p P , then p is very tidy i for all s-like dp , (A), above, holds and for all singular dp , (B+), below, holds.
modified:2004-08-01

(B+) p |D = f , where = p ().

Remark 1. Suppose that p is tidy. We argue that for all singular d p , all U () [ p (), p (k) and all D dp , p, () p (). It suces, of course, to prove this for the hp, and the p, . For the hp, ,

revision:2004-08-01

if f () dom g, then hp, () = p (), so suppose that f () dom g. / Then, hp, () = hp (f ()), and by (A) above (with in place of and f () in place of ), the latter is p (), as required. For the p, , we

work by induction on , with the induction hypothesis being the statement of the remark, i.e., the statement for the p, , with < . But then, the

340

42

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

p, conclusion is immediate by the denition of p, : clearly, 1 () p (), p, and 2 () is the sup of things all p () and so the conclusion is clear.

Remark 2. If p is very tidy then for singular dp , it is easy to see that, with the convention of Remark 2 of (3.6), hp ( p ()) p (). This is clear from Remark 1 and the fact (which is just a restatement of (B+)) that p () = scale( p |D ). Remark 3. If p is very tidy, p q and for all s-like, regular dp , hq ( p ()) = p () then for all s-like dp , hq ( p ()) = p (). This is easily argued by induction on the rank of in the well-founded relation D . The basis is the hypothesis. Let := p (). By the induction hypothesis, we have
q, p, p, that hq, = p |D . Clearly 2 = 2 , and by Remark 2, 2 = p |D .
modified:2004-08-01

q, Clearly, 1 = p |D , and the conclusion is then immediate.

(4.2) The following material will be helpful in both (4.3) and 5. If p P , and t is a function with dp dom t, we say that t covers p i whenever dp is singular, U () [ p (), p ()] and D() dp , p, () < t().

revision:2004-08-01

If q P , we we say that q covers p i p q and q covers p. If q P and t is a function with dom t = dq , we say that q dominates t i for all s-like dq we have t() < q (). Next, we dene still more functions associated with a p P . For singular dp , and D dp , we let tp () := sup { p, ()| U () ,1

340

[SHST:340]

43

[ p (), p ()}. We note that by (3.5)(D), tp () < + . For regular, s-like ,1 dp , we let tp () := sup {hp ()| U () & < p ()}. Again, by 1 (3.5) (D), for regular, s-like dp , tp () < + . For singular dp , we 1 let tp () := scale(tp ). Clearly, for singular dp , tp () < + . We also 1 ,1 dene the tp and the tp analogously, but based on the function p ; thus, ,2 2 tp := p |D dp , and tp () := scale(tp ), for singular dp , while for ,2 2 ,2 s-like regular dp , tp () := p (). Note that whenever these functions 2 are dened, we have tp () tp () and tp () tp (). ,2 ,1 2 1 Now, let , M, N , etc., be as in (1.2), (1.3), and suppose that p N . Then it is obvious that the tp SkM (N ) and that the tp SkM (N {}), i ,i and therefore, that

modified:2004-08-01

() for all s-like dp , tp () < pN (), 1 since () holds for any t SkM (N {}) in place of tp . 1 (4.3) Lemma. If p P , t is a function with dom t = dp and for all dp , t() < + ,

revision:2004-08-01

then there is very tidy q P with p q and such that for all s-like dp , t() < q (). Proof. We shall prove this in the way that will be most useful for (5.1). Choose regular 2 , and let M, N be as above for this . We have just observed that since p N, pN is everywhere tp ; similarly, since t N , 1

340

44

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

if we construct very tidy q p such that

() for all s-like dp , q () > pN (), then q will be as required. This is the approach we shall take; we shall choose such an N , and construct q satisfying () and such that whenever dp is s-like, g q ( p ()) = !. Our approach to this will be to take = pN , and to let be as given by (1.5) for this , and to take dq = dp , (q) = (p), q = p , q () = (), for all s-like dp and q () = p () for all other dp . Recall that, for all singular dp , letting =
(), |D = f . This makes it clear that we will have (B+) of (4.1)

and that q will satisy (). Thus, in order to complete the proof, it will suce to verify that (A) of (4.1) holds and that q P , since it will then
modified:2004-08-01

be clear that q p. Before going further, it will be useful to exploit (4.1) (B+) further. Suppose that u is a function with domain = D . We make the following observations: (A) Suppose that for all dom u, u() (). Then, clearly, scale(u) . (B) Suppose further b is a conal subset of D and that for b, u() = (). Then, again clearly, scale(u) = .

revision:2004-08-01

An important property of the way we will dene q is:

340

[SHST:340]

45

(C) g q () = !, for all p () < q (), whether or not is singular. By (C) and the denition of tp if is singular, then, 2 (D) for tp () < q (), with U (), will be q-strongly-deactivated on b , 2 if for no other reason. It remains to see that for singular and p () < tp (), with 2 U (), we will still have that is q-strongly-deactivated. It is here that we
will appeal to (3.5) (C). Let := p (). If p |D f , then tp () = , 2

so there is nothing to verify in this case. If, on the other hand, the above fails, then, by (3.5) (C), is p-strongly-deactivated and p |D p, . Since it is clear that p, q, , this means that we will have that is q-strongly-deactivated and that we will have hq () scale( p |D ), and
modified:2004-08-01

scale( p |D ) is just tp (). This in turn means that if < < tp (), with 2 2 U (), then we will have that is q-interval-strongly-deactivated, as required. The preceding guarantees that we can carry out our plan of making (C) hold, while respecting (3.5). It remains to complete the denition of

revision:2004-08-01

the g q |[ p (), q ()) and to dene the hq (), for p () q (), with U (). This will be done by recursion on the rank of in the wellfounded relation: D , so the basis is when is regular. In view of (C), we must carry out the following: (1) dene hq () and g q |s for as above,

340

46

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(2) dene g q () for p () < < q () which are not covered by (C) nor by (1). As far as (2) is concerned, in all cases, we shall have g q () := 1 unless it is forced to be 0. As far as (1) is concerned, once we have computed hq (), (3.4) (E) tells us how to dene g q |s . Thus, it remains to compute the hq () and verify that the computed value is consistent with (3.4), (3.5) and (4.1) (A). It will then be clear that q is a very tidy condition, and of course, that p q, completing the proof of the Lemma. Of course, (3.5) (B) tells us that for singular , in order to compute the hp (), it suces to compute the p, . This will be done by recursion on , within the recursion on . We now appeal to (A) and (B). Our induction hypotheses are
modified:2004-08-01

(E) for all relevant < and p () < q (), hq ( ) (), (F) for all relevant p () < and all, not just on a tail of, D dp , q, () (). Now, (E) guarantees that for all D dp , hq, () (). Similarly,

revision:2004-08-01

q, (F) guarantees that for all D dp , 2 () (), and therefore,

for all such , q, () (). This preserves the induction hypothesis, (F), and then, by (A), scale( q, ) (), which preserves the induction hypothesis, (E), at least as far as . As indicated above, this completes the proof that q P and p q.

340

[SHST:340]

47

Remark 1. The q we obtain depends, obviously, on p and on the N we choose, so, we naturally denote it as q(p, N ). A very plausible choice for N is to take it to be some sort of Skolem hull in M of p, t and possibly some additional elements, e.g., all the members of , for some cardinal . We return to this in (5.1), (5.2), below. In this connection, it is also worth pointing out that if p |N |, N M, and, letting N := |N |, if

[N ]< N {N } |N |, card N = , then clearly q N . Remark 2. We are now in a position to show, as promised at the end of (3.5.1), that in no condition q P do we have that for a tail of U () \ q () + 1, P r q () holds, where dq is singular. What we show, in fact, is that if p P is very tidy, dp is singular, U ()\ p ()+1 then P r p () fails. This suces, since as noted in the last sentence of Remark 1
modified:2004-08-01

q p of (3.6), if q p, then for U (), if P ri () holds then P ri () holds.

This will also complements Remark 3 of (3.6). since if p P is very tidy, and dp be singular, then letting = p (), we have that = scale( p |D ), so there are no ordinals of the sort dealt with in Remark 3 of (3.6). We would like to know that no others share the property pointed

revision:2004-08-01

out there, of being q-strongly-deactivated in any q p with q () . Actually, this will follow from the extendability properties developed in (6.1), but showing that if P r p () holds then p () will in fact be quite useful for (6.1). If U () \ + 1, since p is very tidy, there is a tail of b disjoint from

340

48

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

p dom g p , so P r2 () fails. Again, since p is tidy, we cannot have hp () > ,

for any < p (), and by Remark 3 of (4.1), we cannot have hp () > . We conclude by induction on , so suppose that for all (, )U (), P r p ()
p p fails. Clearly, then P r3 () cannot hold, and P r1 () cannot be witnessed p by any (, ). But we have already argued that P r1 () cannot be

witnessed by any , so the proof is complete. (4.4) FACTORING. Let be a regular cardinal, 2 , and let p P . We set W (p) := W (p) \ + , and we set (p) := {(, max (, ))|(, ) (p)& W (p)}. We let W (p) := W (p) + , (p) := (p) W (p) and we let R (p) :=
modified:2004-08-01

{b [, )|(, ) (p)& p () < , + }. We are

now ready to dene the upper and lower parts of P, relative to , which give the Factoring Property of P. Denition. For p and as above, we set (p) = (g p |dp \ , p |dp \

, (p)). Note that (p) P . We let P = {(p) |p P }, with the restriction of . Thus, P is the class of conditions for coding down to a

revision:2004-08-01

subset of + . Note that P = P2 , since for p P, (p)2 = p. We also dene (p) , for p P : (p) = (g p |, p |, (p), R (p)), and we let P := {((p) , (p) )|p P }. Thus, P is a (proper class) P - name for a subset of {(p) |p P }. We have guaranteed that the latter is a set by replacing {(, )|(, ) (p) & < , + < } by

340

[SHST:340]

49

Rp (). Of course, our intention is to have P be the name of the underlying set of a partial subordering of ({(p) |p P }, S), where (p) S (q) i g p | g q , p | q , R (p) R (q), (p) (q). We let P be the name for this subordering. In fact we can cut P down to a set name, as follows. Let n < be such that all relevant notions about P are n . Let >> be such that (H , ) reects all n formulas. The set name is then simply {((p) , (p) )|p P H }. What makes P a name is the linkage between the top and the bottom, which is what guarantees that P will code down to a subset of 3 the subset B of + added by P . The R (p) is one feature of this linkage. The following is then clear: Lemma.
modified:2004-08-01

(FACTORING) P P P . =

(4.5) Lemma. Let be as in (4.4), and suppose that is a limit ordinal and (pi |i < ) is an increasing sequence from P . Let d := {g pi |i < }, := { pi |i < }, := {dpi |i < }, g :=

{(pi )|i < }. For d,

revision:2004-08-01

let () :=

{ pi ()|i < & dpi }. If d is singular, set Z()

i C() and g() = !. Let i I() i for some i j < and some Z(), pi () and for all D dpi , pi () pj , (). Suppose, further, that (pi |i < ) has the following properties. (1) {i < |pi is tidy} is conal.

340

50

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

(2) For all singular d, I() is conal in (this, of course, implies that Z() is conal in ()).

Then, p := (g, , ) P and is the least upper bound for (pi |i < ). Proof. We will concentrate on showing that (3.5) (C) holds. This is the heart of the matter for verifying that p P , as verifying that the other clauses hold is totally routine, and once we know that p P it is clear that it is the least upper bound. So, let be as in the statement of the Lemma, and adopt the other notation there. Also, let := (), and note that Z() is just the Z of (3.5). As observed in the parenthetical remark to hypothesis (2) of the statement of the Lemma, Z is therefore conal in , which means that
modified:2004-08-01

is p-strongly-deactivated on C . So, it remains to verify the last clause of (3.5) (C). For this, we rst note that for all D d, we have

p, () 2 () ().

revision:2004-08-01

This is because, since I() is conal in , if D d, () = sup , where := {(i)()|i I() & dpi }. Now, let i I() with dpi . Let i j < and Z() \ pi () be as guaranteed by the fact that i I(). Then, p, () pj , () pi (). We now complete the proof by verifying the last clause of (3.5) (C). So,

340

[SHST:340]

51

let := scale( p, ). Note that in virtue of (), and the fact that for all
p, p, relevant , p, () 2 (), we have that f p, 2 |D .

Now, if U () with < < , if b dom g p is conal in and b dom g p , then < (card ). But then there is i < with < pi (card ), and in virtue of (1), we can assume that pi is tidy. But then hpi () < pi (card ), by (4.1) (A), and since hpi () = hp (), the conclusion is clear. Remark. In addition to the hypotheses of the Lemma, suppose that < , that M is as in (1.2), (1.3), that N N M are such that, letting

N := |N |, N := |N |, we have that for all i < , pi N and that [N ]< N {N } N . Then p N .

modified:2004-08-01

(4.6) DECODING. We now complete the sketch of the decoding procedure given in (2.5) by supplying the details of the decoding for limit cardinals. For singulars, in addition to the case of decoding the generic, we also treat the case of decoding g|[, ) from a g dened on a large enough domain below . We

revision:2004-08-01

will give the details of the situation when we encounter it. This case is needed in (6.1) (7) (b), below. We treat the inaccessible case rst.

(4.6.1) Assume that is inaccessible, and that we are given G, a function with

340

52

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

range G {0, 1, ?, !} and dom G =

{(, + )|2 < , a cardinal} {g p | |p I}.

such that for some generic ideal I in P, G =

We will rst obtain an ordinal which will be the common value of the p () for dp , p I. Recall that, by Lemma 3 of the Introduction and the discussion preceding it, for all (, + ), if and in L[A ], card = , then we obtain (b | < ) canonically from in L[A ]. In particular, we have, in L, (b : < 2 ). So, we rst decode G on U () 2 . For such , we let G0 () = 1 i on a tail of b , G() = 1. Otherwise, we set G0 () = 0. Now, by (3.4) (D) (1), there is a largest (U ()) (, 2 ) such that G0 () = 0 and, further that this is of the form + , where {} U (). Also, by (3.4)(D)(2), we have that G0 |(, 2 ) codes a well-ordering of on odd successor multiples of in this
modified:2004-08-01

interval. We take := the order-type of this well-ordering. By (3.4)(D)(2), again, we have that is the common value of p () for dp , p I and that A is coded by G0 on even successor multiples of in this interval.

(4.6.2)

revision:2004-08-01

We can now dene G() by recursion on for U () \ . We rst dene () and obtain A(). We shall have that in L[A()], card = , so that we have b available. If is not a cardinal in L, we let () := cardL . Otherwise, we let () = . Recall from (2.3) that if < < + is a cardinal in L, then for all non-multiples of , with < < , X

340

[SHST:340]

53

is conal in . This allows us to dene A () as follows. If () , then we already have A . Otherwise, if < < (a), let := 2 + 1. Then, < (), so X (, ()) = , and we have A i for some (all) X (, ()), G() = 1. Thus, we have A (), and therefore, in L[A()], we have b . We set G() := 1 i on a tail of b , G() = 1; otherwise, G() = 0. This completes the recursion.

(4.6.3) We can then dene G on (, + ) \ U () by G() := i i for some (all) X \ , G() = i. Finally, we can go back and dene G on U () as follows: let U () be such that X . If G() = 1 and on a tail of b , G() = 1, G() := 1. If G() = 0 and on a conal subset of
modified:2004-08-01

b , G() = 0, then G() := 0. Otherwise, we set G() := ?. This completes the decoding procedure in the inaccessible case.

(4.6.4) So, assume next that is singular. We rst treat the generic case. Assume that G, I are as above. This time, there are three passes in the

revision:2004-08-01

denition of G. The rst is relatively straightforward: we ignore deactivated intervals, deactivation on C , we ignore the p, and the p, , and just organize the information from below provided by G. This is done simultaneously, for all members of U () at once. No recursion is involved. Recall here that we have A at our disposal, and that all of the coding

340

54

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

apparatus is present in L[A ]. The second pass proceeds by recursion on U (), and takes into account all of the above; we also dene G|s . At the end of the second pass, we will have dened a function G1 on all multiples of in (, + ). In the third pass, we then go back, dene G on the non-multiples of , detect contamination, dene (which, once again, will be the common value of the p () for p I with dp ) and revise the denition of G1 below . For the rst pass, if U () and there is no tail of b on which G is constant, set G0 () := ?; otherwise, if x {0, 1, ?, !} and G has constant value x on a tail of b , set G0 () := x. We also dene at this time, by
letting () := the order-type of the well-ordering coded by G|sf () ,

for D .
modified:2004-08-01

(4.6.5)

For the second pass, we assume that we have dened G1 on all multiples of below in such a way that for < < , with U (), G1 |s codes a well-ordering of in order type . We let H() := the order type of

revision:2004-08-01

this well-ordering. We also assume that we have dened functions , with domain D , for such , correctly (i.e. according to (3.5), and what now follows), so far. If there is such a < with H() > we take the least such and set G1 () := !, () = (), for all D , and we dene G1 |s to code the <L[A] -least well-ordering of in type H().

340

[SHST:340]

55
:= f and dene 1 := f

So, assume there is no such . Let f

and for D , dene 2 () := sup { ()| C } and () := max(1 (), 2 ()), () = max(H (), ()). If, on a tail of

b , G() = !, then we already had G0 () = ! and we maintain G1 () := !. However we also set G1 () := !, if is a limit of multiples of 2 and Z is conal in where Z := { C |G1 () = !}. In all other cases, we maintain G1 () := G0 (). If G1 () = !, we dene G1 |s to code the <L[A] -least well-ordering of in type scale( ). In all other cases we dene G1 |s to code the <LA] -least well-ordering of in type . This completes the recursive denition of G1 on the multiples of . The following statement (whose verication is now totally straightforward and is left to the reader) makes precise the claim that this decoding procedure correctly
modified:2004-08-01

decodes on a tail of U ():

() if p I, dp , p () < p (), and U (), then G1 () = g p ().

(4.6.6)

revision:2004-08-01

Before turning to the remainder of the denition in the generic case, we turn to the decoding of g|[, ) on the multiples of only, from a g dened on a large enough domain below , since in (6.1) (7) (b) we only need this for the multiples of . Here, rather than having G dened on {(, + )|2 = card < }, we have a tail t of D and a function

340

56

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

with domain t such that for t, () U () with < () < + , such that g is dened on {(, ())| t} (of course, g will also be

dened elsewhere, but only this is relevant for our decoding). We take = scale(). Finally, to complete the description of the context of (6.1) (7) (b), below, we have that there is very tidy p P such that t = dp D and that for t, p () (), g p |(, p ()) = g|(, p ()) and that for U () [ p (), ()) we will have g|s coding the <L[A] -least well-ordering of in type and g() = ? except possibly in the case of = p () when it is also possible that g( p ()) = !. Having described the context, the procedure is essentially identical to the above, with the obvious notational analogies, so we limit ourselves to describing the more substantial dierences. These deal only with the
modified:2004-08-01

denitions of the functions analogous to the H and 1 , above. In both cases, we impose that h (), 1 () () by taking them as dened to

be the min of () and the value dened as above. This completes the treatment of the singular non-generic case.

revision:2004-08-01

(4.6.7)

We complete the description of the decoding procedure by returning to the nal phase of the singular generic case: dening G on the non-multiples of in (, + ), detecting contamination, and revising the denition of G1 on the bounded initial segment of multiples of where there is contamina-

340

[SHST:340]

57

tion. In several places, our argument will appeal to density arguments from (6.1). We should emphasize that there is no circularity here, since we have already completed the portion of the argument (the singular non-generic case) needed in (6.1) (7) (b). So, let < < + with not a multiple of . We argue that there is a tail, T , of X and an i {0, 1} such that 1i G1 [T ] and for a conal set of T, G1 () = i. We rst show that X G1 [{0}], X G1 [{1}] cannot 1 1 both be conal. To this end, let p I such that dp and < p () (such a p exists, by (6.1) (4) and (7)). Suppose, now, towards a contradiction, that p () 0 , 1 X and that G1 (j ) = j. Let i := g p (). By (6.1) (4), again, there are q, r I with q () > 0 , r () > 1 . Clearly we can assume that p q r. By (), above, g r (j ) = j. But this contradicts
modified:2004-08-01

(3.4) (A), since g r (j ) is forced to be i. Finally, from (), above and (6.1) (4), it is immediate that if < + there is < and p q I with X q () such that g q ( ) = i and q () . Then, by (), above, again, G1 ( ) = i and we are nished. So, we dene G() := that i {0, 1} such that for a conal set of

revision:2004-08-01

X , G1 () = i. Finally, for U (), we say that is G1 -contaminated i (following (3.3)) G1 () = ! and either there are conal subsets, Yi b , i = 1, 2, and x1 = x2 such that for Yi , G() = xi , or, if is an odd multiple of 2 , letting be such that X , G1 () {0, 1} but G1 () = G(). We let := sup {| is G1 -contaminated}. It is

340

58

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

totally straightforward (and left to the reader to verify) that if p I and dp , = p (). Then we dene G() for < < + , a multiple of , by setting G() := G1 () if , while for U () , we set G() := ? and we dene G|s to code the <L[A] -least well-ordering of in type . This completes the decoding procedure. 5. STRATEGIC CLOSURE AND DISTRIBUTIVITY.

(5.1) THE WINNING STRATEGY. In this item we prove Lemma 5 of the Introduction. So, let 2 be regular and x p0 P , M and N as in (1.1) Further assume that N is super M coherent, and recall the denition of the game G(, N , p0 ) in (1.1) (where Q = P and X is the class of very tidy conditions). Recall that BAD
modified:2004-08-01

must play very tidy conditions. GOODs strategy will be to use (4.3) at successor stages, so that, in the notation of (4.3), she will have p2+2 := q(p2+1 , N ), for an N which we shall describe below. This will be chosen so as to guarantee that at limit stages, we have the hypotheses of (4.5), so that, at limit stages, , GOOD

revision:2004-08-01

will take p to be given by (4.5). The other implicit assumption is that at all stages so far, BAD has succeeded in catching p2i , p2i+1 inside |N(i) |. For the successor step, we take N := N(i) , p := p2+1 , N := the

Skolem hull in M of N {N } and q := q(p, N ). Note that we easily have that pN N , so that for all s-like d(p), q () > pN > p ().

340

[SHST:340]

59

This guarantees that we have g q (pN ()) = !. Now, let be a limit ordinal and let d(p) be singular, where p is as in (4.5). Let i0 be the least i < such that dpi . Since pi0 |N(j0 ) |, where j0 is least such that i 2j + 1, is N(j) - controlled for all j0 j < . Then, letting := p (), {pN(j) ()|j0 j < } is a subset of (g p )1 [{!}] which is conal in . Finally, the supercoherence of the model sequence N guarantees that it is also a subset of C , which means that it is a subset of Z . It is then routine to see that we have the hypotheses of (4.5), and therefore, that the strategy for GOOD is winning. To obtain the distributivity properties, we must see that BAD neednt lose due to inability to catch p2i , p2i+1 inside |N(i) |, and, more importantly, that there are enough supercoherent sequences. These points are addressed
modified:2004-08-01

in the next item; the latter draws on the work of our companion paper [17].

(5.2) Corollary. P is (, )-distributive. Proof. If p0 P and (Di |i < ) is a denable-in-parameters sequence of open dense subclasses of P , begin by picking singular with cf >> ,

revision:2004-08-01

such that all parameters in the denition of (Di |i < ) lie in H and such that (H , ) reects all n -formulas, where the denitions of (Di |i < ) and P are n and n is larger than the number of quarks in the physical universe. Let Di = Di H . We take M = (H + , , ), we let N0 M, with + 1 {(Di |i < ), P |H } N0 (:= |N0 |) and card N0 =

340

60

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

, [N0 ]<

N0 . By the main result of our companion paper, [17], see

(1.4), we can nd super M-coherent (Ni |i ) starting from N0 . We then play a run of the game G(, N , p0 ), where GOOD plays by her winning strategy. We argue that BAD can produce a subsequence of N which catches p2i , p2i+1 inside N(i) . For successor i, given that p2i2 , p2i1 N(i1) , the last sentence of Remark 1 immediately gives that if BAD chooses (i) > (i 1), p2i N(i) . Then, if he chooses p2i+1 from N(i) , this is as required. For limit i, letting := sup {(j)|j < i}, given that all the pj N , j < i, the Remark of (4.5) immediately gives that if BAD chooses (i) > , then pi , the p of (4.5), will lie in N(i) . Once again, if he then chooses pi+1 from N(i) , this will be required.
modified:2004-08-01

Thus, in a such a play, we will actually produce a p . If, in addition to the above, BAD chooses p2i+1 Di N(i) , then clearly we will have p {Di |i < }, as required.

(5.3) Remarks.

revision:2004-08-01

(1) It follows from (5.1) and (5.2) that any iteration of P with supports of cardinality is (, )-distributive. The reason is quite simply that (5.1), (5.2) and the results of [17] depend only on ground model properties. In fact, S. Friedman has informed us that prior to our work similar observations had been made about Jensens original

340

[SHST:340]

61

coding and the variants of it mentioned in the Introduction. (2) As mentioned in the Introduction, by a simple variant of the proof of (5.1), we can show that GOOD has a winning strategy for the version of the game where we start from a p0 P , which is not necessarily P . We freeze g p0 |, (and therefore, also p0 |) and never add any (, ), with < , to any (pi ). We use this in (6.1) (7) (c). 6. EXTENDABILITY AND CHAIN CONDITION. (6.1) Lemma. Let > 1 be a cardinal, p P . Then, in each of the cases 1 i 7, below, there is a q P, p q, satisfying the conclusion of (i) (which follows
modified:2004-08-01

the colon, in each case). (1) is a successor cardinal, dp : dq , (2) is a regular cardinal, dp , < p (), g p () = 1, W (p), and if is inaccessible, then letting 1 be as in (3.4) (D), either p () < p () or 1 < 2 : W (q),

revision:2004-08-01

(3) is regular, dp , p () < + : < q (), (4) is singular, dp , < < p (), is not a multiple of , p () < < + : there is X , < < q () and g q ( ) = g p (), (5) is regular, dp , < p (), g p () = 0, < and if

340

62

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

= + , where is a limit cardinal, then dp : there is b \ with card dp , q (card ) < q (card ) and g q () = 0, (6) is inaccessible, dp : dq , (7) is singular, dp : dq . Proof. The properties are given in order of increasing diculty and/or dependence on earlier properties. The analogue of (5), where = + , a limit cardinal, dp , is achieved by rst adding to dp , via (6) or (7), as appropriate, then using (5). We deal with the cases in the given order. In all cases, in virtue of (4.3), we can assume that p is very tidy. In all cases except (5), when we have to dene g q (), we shall always make g q () = 1 unless it is forced to be 0, in which case we make it 0, except in the following cases:
modified:2004-08-01

(a) (b)

card is s-like and is a multiple of card , = card is inaccessible, < q () and is a multiple of .

Thus, in what follows, except in case (5) we shall limit ourselves to treating the above cases. In case (5), we will nd a which is not forced to be 1

revision:2004-08-01

and which is not in R(p) and for this , we shall make g p () = 0. will not be a multiple of card . This will be the only exception to our general procedure, even in case (5). For (1), if = + , we simply set dq = dp {}, (q) = (p), we set q () = 2 , q () = + 1.

340

[SHST:340]

63

For (2), we shall have g q = g p . If = + , where dp , let = . If = + , where dp , let = p (). If is inaccessible and dp = , let = 2 . Finally, if is inaccessible and dp = , let = sup( dom g p ). In the last two cases, if < , is regular, we can always take , as well. Then, we let (q) = (p) {(, )}. For (3), we will have dq = dp , q = p , (q) = (p), and for all dp \ {}, q () = p (). We set q () = the least > which is a multiple of 2 . Now, suppose p () < q (). The only case we have to treat is when is s-like and regular, and is a multiple of 2 ; we set g q () = ?, and we take g q |s to be as required by (3.4)(E). Clearly q is as required. For (4), pick X , > . We shall have dq = dp , q = p , (q) =
modified:2004-08-01

(p). If dp , {} , we shall also have q () = p (). We set q () = + 2 . For a tail of D , dp and p ()


f p () () < f () < f q () (). So let 0 = 0 ( ) be suciently large

so that the preceding holds for D \ 0 . Clearly, we may assume


0 D . For D \ 0 , let q () = f q () (). If is a limit of singular

revision:2004-08-01

cardinals and > 0 is a successor point of D , let := q ( ) and let


= f ( ). Then, as for and p (), there is 0 ( ) 0 such that for all 0 D , p () f p ( ) () < f () < f (). For such we set q () := f (). For all other dp , q () = p ().

Suppose now that dp and p () < q () (so, in particular

340

64

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

{} ). We deal rst with dening g q ( p ()), so let = p (). If is p-deactivated, we set g q () = !; otherwise, we set g q () = ?. In both cases we take hq () = , and dene g q |s to satisfy (3.4) (E). In virtue of the rest of the denition of g q , Remark 2 of (4.3) will guarantee that this is as required (recall that p is very tidy!). Next, suppose that is a multiple of 2 with p () < < q (). We shall have that g q () = ? unless one of the following occurs:

(c) (d) (e)

= and = ,
D \ 0 ( ) and = f (),

is a limit of singular cardinals and there is , a successor point


of D \ 0 ( ) such that, letting := f ( ), D \ 0 ( ) and = f ().

modified:2004-08-01

In these cases, we set g q () = g p (). In all cases, we will have hq () = , and well dene g q |s to satisfy (3.4) (E). By Remark 2 of (4.3), and the
fact that if dp is singular and p () < then f > p |D , it is then

clear that q P and is as required. In fact, it is easily veried that q is

revision:2004-08-01

very tidy, though we do not need this. For (5), suppose, rst, that = + , and dp . In this case, we shall have dq = dp , q = p , (q) = (p). By (3.2)(A) for p and = , we can nd 0 < such that whenever (, ) (p) and < , then b b 0 . Without loss of generality, 0 < p (). Now let b \ max (, 0 ). By

340

[SHST:340]

65

our choice of 0 , R(p). Also, since b and (, + ) where is a successor cardinal, is even, but not a multiple of . Thus, is not forced to be 1. Accordingly, we set g q () = 0. The remainder of the construction of q divides into cases, according to whether is regular or singular. If is regular, we proceed as in (3), with in place of and in place of and with the already-noted dierence that g q () = 0. If is singular, we proceed as in (4), with in place of , in place of with the alreadynoted dierence that g q () = 0; the argument here is simpler than in (4) since there are no nor involved. Then q is as required. If = + , dp , then, by hypothesis, is a successor cardinal, so we can use case (1) to obtain p q with dq , and then apply the immediately preceding argument to q instead of p, to obtain the required
modified:2004-08-01

q. Finally, suppose is inaccessible. As above, we can nd 0 < such that whenever (, ) (p) and < , then ba b 0 . Pick
max(, 0 ), = , even successor. We take := f ( ). We note,

once again, that by our choice of 0 , R(p), and that since is even and card is an even successor, is not forced to be 1. Then, we can proceed

revision:2004-08-01

as in (1) and (3), to add to dp , and make q ( ) > , EXCEPT that, as above, we can also make g q () = 0. Clearly q is as required. For (6), we shall have dq = dp {}, (q) = (p) and for dp , q () = p (). By (3.2)(A), for p with = , d = dp , we can compute q () according to (3.4)(C) and 1 , according to (3.4)(D) and they will be bounded,

340

66

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

in + , 2 , respectively. We take q () = the least multiple of q (). If < < 1 or 2 < q () and is a multiple of , we set g q () = ? i is contaminated. If it is not contaminated, we set g q () = 1, unless it is forced to be 0, in which case we make g q () = 0. If q () < q (), is a multiple of , we set g q () = 1 unless it is forced to be 0; in this case, we set g q () = 0. Finally, we dene g q on the multiples of in [1 , 2 ) to satisfy (3.4) (D), (1) and (2). Clearly this q is as required. Case (7) divides into subcases, as follows: (a) D dp is bounded in (in some sense, the simplest subcase: we must add to dp a tail of , but there is no contamination), (b) D dp ( will be the only new member of dq , but we must deal with contamination); we shall use the decoding procedure of (4.6);
modified:2004-08-01

(c) (a) and (b) both fail (the most complicated case: we must combine the methods used for (a) and (b), and appeal to (5.1), (5.2)). In case (a), let 0 < be such that D dp 0 . Clearly, we may assume 0 D , and, anticipating the argument for (c), if < , regular, we can take 0 . We shall have (q) = (p), dq = dp {} ( \ 0 ). For

revision:2004-08-01

dp , we will have q () = p (). For {} ( \ 0 ), we shall have q () = + 1 if is a successor cardinal and q () = 2 , if is singular.
We set q () = 2 . For D \ 0 , we set q () = f2 (). If is a

limit of singular cardinals, a successor point of D , > 0 , let = q ().


Then, if D \ 0 , we set q ( ) = f ( ). Then, for {} ( \ 0 ),

340

[SHST:340]

67

if < < q (), and is a multiple of 2 , we set g q () = ?, we take hq () = , as required by (3.4) (E), and we dene g q |s to code this, as required by (3.4) (E) (we can always nd R L[A ] as required, since either is singular, in which case (+ )L = + , or , in which case
(+ )L[A] = + ). If is s-like and regular, recall that f was dened at

the end of (1.2). This completes the proof in case (a) of (7). In case (b), we will have dq = dp {}, (q) = (p) and for dp , q () = p (). If dp , , we shall also have q () = p (). We set q () = = scale( p |D ). Let 0 < be such that if D \0 ,
then dp & p () f (). Clearly, we may assume 0 and,

anticipating the argument for (c) when is a limit of singular cardinals, below, if is regular, < we can take 0 . For D \ 0 we
modified:2004-08-01

set q () = f (). If is a limit of singular cardinals, for such , if is

a successor point of D and 0 = p () < q () = 1 , then, on a tail of


D , dp and p () = f0 () < f1 (). So, let 0 = 0 () be such that whenever 0 D , dp and p () = f0 () < f1 (). For such , set q () = f1 (). If D 0 , or (if is a limit of singular cardinals), for

revision:2004-08-01

some successor point, , of D 0 , D , or (if is a limit of singular cardinals) for some successor point, , of D \ 0 , D 0 (), set q ( ) = p ( ). For such that p () < q (), we handle the denition of g q |[ p (), q ()) as we did in case (4), except that, here again, as in (5), the argument is

340

68

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

simpler since there are no , involved. Thus, it remains to dene q () and g q |(, q ()). We dene g q in the usual way on the non-multiples of in (, ). We shall dene q () to satisfy (3.4)(C) with d = dp and q| in place of p|. Note that conceivably < q () < , where is the least multiple of 2 , , with <
such that (f p |D ), since instances of contamination could arise due

to the denition of the g q () for those D \ 0 with p () < q () (if there are conally many such). For a multiple of 2 , < < q (), we set g q () = ?, and we dene g q |s to satisfy (3.4) (E). For q () < , a multiple of 2 , we dene g q |({} s ) by recursion on , following the singular non-generic case of the decoding procedure of (4.6), with g = g q |. This completes the
modified:2004-08-01

construction of q in case (b). For case (c), our strategy is to obtain p p , p P such that the hypothesis of case (a), above, holds for p and , and then apply (a) to p . The construction of p diers according to whether = + or is a limit of singular cardinals. The former case is much easier, and we consider

revision:2004-08-01

it rst. Here, we obtain p by simultaneously adding to dp , following the procedure of (1), for any nal segment of D \ dp . In particular, anticipating the argument when is a limit of singular cardinals, the nal segment can be taken to lie above , if is regular, < . The simultaneity is emphasized to make clear that we are not yet appealing to any strategic

340

[SHST:340]

69

closure properties. We then proceed as in (a) with p in place of p. When is a limit of singular cardinals, as a rst step toward obtaining the desired p , we rst simultaneously add to dp all the D , for a successor point of D , dp , according to the procedure for (1). This is a condition, p0 , intermediate between p and p . To obtain p , we let (i : i < ) increasingly enumerate D \ dp . We let > 2 be regular, < . Let M be a master model, M = (H + , , ), singular, >> , such that p0 H and (H , ) models a suciently rich fragment of ZFC, etc., as in (5.1). As in (5.1), we can assume that we have (Ni : i ) which is super M-coherent, with p0 |N0 |. So, x such (Ni : i ). We then play the following run of the variant of G(, N , p0 ), mentioned in (5.3). GOOD
modified:2004-08-01

plays by the winning strategy of (5.1). BAD chooses (i) as in (5.2), and obtains p2i+1 |N(i) |, by adding i to dp2i , following the procedure of case (a) of (7) (note that the hypothesis of case (a) will always hold for i and p2i ). Then, p can be taken to be p . This completes the proof for (c), when is a limit of singular cardinals, and therefore completes the proof

revision:2004-08-01

of (7) and the Lemma.

(6.2) P HAS THE + -CHAIN CONDITION Let > 2 be regular. The crucial observation is: (6.2.1) Proposition. Suppose p, q P, (p) , (q) are compatible in

340

70

SAHARON SHELAH1,2,4,5 AND LEE J. STANLEY3,4,5

P , g p | = g q | and p | = q |. Then p, q are compatible in P . Proof. Let r P with (p) , (q) r. Note that, without loss of generality, we may assume that W (r) = W (p) W (q). We shall show that r P, p, q r , where r = (g r g p |, r p |, (r )), where (r ) = (p) (q) (r). Of course, p, q r is clear, once weve veried that r P . For this, all clauses of (2.2), (3.2) are clear, as are (3.4) (E) and all clauses of (3.5). We argue that there is no new contamination in r , from which it will follow readily that we also have all (3.4) (A) - (D). This will complete the proof. Clearly there is no new contamination at singulars, and there is no new contamination at inaccessibles above . So, suppose that is inaccessible, . Suppose that (, + ) and that is
modified:2004-08-01

contaminated by . Let ( , ) (r ) witness this, as in (3.3). Then, W (p) W (q), and since < clearly ( , ) {(p), (q)}. But then by the hypotheses of the Proposition, must be contaminated by either in p or in q according to whether ( , ) (p) or (q). This completes the proof.

revision:2004-08-01

(6.2.2) Corollary. In V P , P has the + -chain-condition. Proof. This is clear from (6.2.1) and the easy computation that {(g p |, p |)|p P } has power , for all regular > 2 .

340

[SHST:340]

71

References
1. A. Beller, R. Jensen and P. Welch, Coding the Universe,, London Mathematical Society Lecture Notes Series, vol. 47, Cambridge University Press, Cambridge, 1982. 2. R. David, Some applications of Jensens coding theorem, Annals of Math Logic 22 (1982), 177-196. 3. , 1 reals, Annals of Pure and Applied Logic 23 (1982), 121-125. 3 4. , A functorial 1 singleton, Advances in Math 74 (1989), 258-268. 2 5. S. Friedman, A guide to Coding the universe by Beller, Jensen, Welch, J of Symbolic Logic 50 (1985), 1002- 1019. 6. , An immune partition of the ordinals, Recursion theory week; proceedings Oberwolfach 1984, Lecture notes in math, vol 1121, H.-D. Ebbinghaus, et. al, eds., Springer-Verlag, 1985, pp. 141-147. , Strong Coding, Annals of Pure and Applied Logic 35 (1987), 1-98, 99-122. 7. 8. , Coding over a measurable cardinal, J. of Symbolic Logic 54 (1989), 11451159. 9. , Minimal Coding, Annals of Pure and Applied Logic 41 (1989), 233-297. 10. , The 1 -Singleton Conjecture, Journal of the AMS 3 (1990), 771-791. 2 11. , A Simpler Proof of Jensens Coding Theorem, accepted, Annals of Pure and Applied Logic (to appear). , A Large 1 Set Absolute for Set Forcings, accepted, Proceedings of the 12. 2 AMS (to appear). 13. S. Shelah, Proper forcing, Lecture Notes in Math, vol 940, Springer-Verlag, 1982. 14. , Cardinal arithmetic, accepted, Oxford University Press (to appear). 15. S. Shelah and L. Stanley, Corrigendum to Generalized Martins axiom and Souslins hypothesis for higher cardinals, Israel J. of Math. 53 (1986), 304-314. 16. , Coding and Reshaping When There Are No Sharps, Set Theory of the Continuum, Mathematical Sciences Research Institute Publications, 26, H. Judah, W. Just, H. Woodin, eds, Springer-Verlag, 1992, pp. 407-416. 17. , The Combinatorics of Combinatorial Coding by a Real, this journal.

340

revision:2004-08-01

modified:2004-08-01

You might also like