CARDINAL PRESERVING IDEALS

by

Moti Gitik School of Mathematical Sciences Tel Aviv University Tel Aviv, Israel

and

Saharon Shelah Department of Mathematics Hebrew University of Jerusalem Jerusalem, Israel

Abstract We give some general criteria, when -complete forcing preserves largeness properties like -presaturation of normal ideals on (even when they concentrate on small conalities).
modified:1996-05-06

Then we quite accurately obtain the consistency strength N S is 1 -preserving, for > 2 .

revision:1996-05-05

310

We consider the notion of a -presaturated ideal which was basically introduced by Baumgartner and Taylor [B-T]. It is a weakening of presaturation. It turns out that this notion can be preserved under forcing like the Levy collapse. So in order to obtain such an ideal over a small cardinal it is enough to construct it over an inaccessible and then just to use the Levy collapse. The paper is organized as follows. In Section 1 the notions are introduced and various conditions on forcing notions for the preservation of -presaturation are presented. Models with N S cardinal preserving are constructed in Section 2. The reading of this section requires some knowledge of [G1,2,4]. The results of the rst section are due to the second author and second section to the rst. Notation.
N S denotes the nonstationary ideal over a regular cardinal > 0 , N S

denotes the N S restricted to conality , i.e. {X |X { < |cf = } N S }, D denotes a normal lter over a regular cardinal = (D) > 0 , D + = {A |A = mod D, i.e. A D}. D Q for a forcing notion Q, such that
Q

(D) is regular is the

Q-name of the normal lter on (D) generated by D. By forcing with D + we mean the forcing with D-positive sets ordered by inclusion.
modified:1996-05-06

1. -Presaturation and Preservation Conditions The denitions and the facts 1.1-1.5 below are basically due to Baumgartner-Taylor [B-T]. Denition 1.1. A normal lter (or ideal) D over is -presaturated if
D+

every

set of ordinals of cardinality < can be covered by a set of V of cardinality .

revision:1996-05-05

Let us formulate an equivalent denition which is much easier to use. Denition 1.2. A normal lter D on is -presaturated, if for every 0 < , and

maximal antichains I = {A : i < i }, i.e. i A D + , [i = j A A D + ], i i j (A D + )(i < i )A A D + ) i

(for < 0 ) for every B D + there is A D + such that A B and ( < 0 )|{i < i : A A D + }| . i 2

310

Fact 1.3. If < , ( < )[ < ], {i < : cf i } D, D is -presaturated, then in 1.2 we can nd A such that < !i < i A A D + . i Proof:
First pick A as in the denition. For every < 0 let Br | < be

an enumeration of {A A |i < i , i otherwise. Dene f : by f () =

A A D + }. Without loss of generality min i (). Then there are < and a D-positive
<0

Br > . For < , < 0 let () be the least s.t. B if such exists and -1

subset A of A s.t f (A ) = {}. Since < , using once more the Fodor Lemma, we can nd A A as required. Denition 1.4. A normal ideal or lter on is -preserving i

(a) I is precipitous (b)

I+

is a cardinal.

Remarks. (1) If = + , then such an ideal is called presaturated. This notion was introduced by Baumgartner-Taylor [B-T]. (2) It is unknown if for 1 (b) may hold without (a). But if 2 = + , then (b)
modified:1996-05-06

(a). (3) Every precipitous ideal is |I + |+ -preserving. Proposition 1.5. Suppose that < are regular, 2 = + and D is a normal ideal over . Then the following conditions are equivalent: (a) D is not -presaturated; (b) cf + < ;
D

revision:1996-05-05

(c) the forcing with D + collapses all the cardinals between + and some cardinal < ; (d) I is not -preserving; (e) a generic ultrapower of D is not well founded or it is well founded but it is not closed in V D under less than -sequences of its elements; If in addition 2< < then also (f) the forcing with D + adds new subsets to some ordinal < . 3
+

310

Proof:

(b) (c) since |D + | = + . The direction (c) (a), (b), (d), (e), (f) are trivial.

If (b) holds, then, as in [B-T], also (a) holds. (a) implies (d), (e) and (f). We are now going to formulate various preservation conditions. Lemma 1.6. Suppose that < are regular cardinals, D a normal lter on = (D), Q a forcing notion
Q

regular and D Q denotes the normal lter on in V Q which

D generates. If D is -presaturated (in V ) then D Q is -presaturated in V Q provided that for some K

()1 D,K,Q,

K is a structure with universe |K|, partial order =K unitary function T =

T K , and partial unitary functions pi = pK such that i (a) for t K, T (t) D + ; (b) K| t s implies T (s) T (t) mod D; = (c) pi (t) is dened i i T (t); (d) pi (t) Q and [K| t s, i T (s) pi (t) pi (s)]; = (e) = K = {i T (t) : pi (t) G } is not forced to be mod D Q ;
t t Q

(f) if q
Q s

T (D )

Q +

for some q Q then there is s, s K, pi (s) q and

T mod D Q ;

Q
modified:1996-05-06

(g) if () < , for < () Y K is such that {T (t) : t Y } is a maximal antichain of D + such that [t Y , s Y , > (C D)(i C T (t) T (s))pi (s) pi (t)], and T D + then there is s, s K, T (s) T and there are y Y , |y | for < () such that [ < (), x Y \y T (x) T (s) = mod D] and

< ()(x y )(C D)(i C T (x) T (s))[pi (x) pi (s)]. Proof: Let () < , and I = {A : i < i } (for < ()) be Q-names of maximal
i Q + Q

revision:1996-05-05

antichains of (D ) (as in Denition 1.2) (i.e.

is a maximal antichain.

We dene by induction on < () Y h such that (i) Y K, {T (t) : t Y } a maximal antichain of D + ; t A

(ii) for every < , t Y , s Y for some C D i C T (s) T (t) pi (t) pi (s)); (iii) h : Y ordinals, and
h (t)

for t Y .
s

Using (g) pick s and y | < () . Then will be as required in Denition 1.2. 4

310

Denition 1.7. 1) Gm(D, , a) (where a ) is a game which lasts moves, in the ith move if i a player I, and if i a player II, choose a set Ai D + , Ai Aj mod D for j < i. If at some stage there is no legal move, player I wins, otherwise player II wins. 2) If we omit a, it means a = {1 + 2i : 1 + 2i < }. 3) Gm+ (D, , a) is dened similarly, but for player II to win, Ai = has to hold as well. By Galvin-Jech-Magidor [G-J-M], the following holds. Proposition 1.8. D is precipitous if player I has no winning strategy in Gm+ (D, ). Lemma 1.9. 1) Suppose that < are regular cardinals, D a normal lter on = (D), Q a forcing notion
Q

regular , = (D).

If player II wins the game Gm(D, , a) in V , then it wins in Gm(D, , a) in V Q provided that for some , , the following principle holds: ()2 D,K,Q : K is a structure as in Lemma 1.6,
modified:1996-05-06

(a) for t K, T (t) D + ; (b) K| t s implies T (s) T (t) mod D; = (c) pi (t) is dened i i T (t); (d) pi (t) Q and [K| t s, =
t Q

i T (s) pi (t) pi (s)];

Q t

(e) = {i T (t) : pi (t) G } and pi (t) (f) if, q

Q

(D Q )+

for i T (t);

T (D Q )+
s

for some q Q then there is s, s K, pi (s) q and

revision:1996-05-05

T mod D Q ;

(g) K is -complete, i.e. if ti : i < < is increasing and T (ti ) D + then there is t K, ti t; (h) Q is -complete and preserves stationarity of all A D + . 2) The same holds with Player I having no winning strategy.

310

Proof: 1) Let us provide a winning strategy for player II. Let G Q be generic over V without loss of generality the players choose Q-names for their moves. We will now describe the strategy of player II in V [G]. Player II also chooses Ti , tj (j a, i < ), according to the moves. Player II preserves the following (for a xed winning strategy F of player II in Gm(D, , a) in V ). (*) the plays A : i < and Ti : i < so far satisfy
i

(a) Ti : i < is a beginning of a play of Gm(D, , a) (in V ) in which player II uses the strategy F ; (b) for i a, A = , T (ti ) = Ti , ti K;
i

(c) for j < i < , tj ti ; (d) V [G]| A [G] D + . =

i

ti

Proposition 1.10. Suppose < are regular cardinals, < [< < ], D a normal lter on , Q a forcing notion,
Q

is a regular and D Q denotes the normal lter on

in V Q which D generates. Let ()3 D,K,Q denote the following:

modified:1996-05-06

(a) K is a partial order, elements of K are of the form p = pi : i T , where T = Tp is D-positive and pi Q for i T . For p, q K, p q i Tp Tq and for every i Tp , pi qi in the ordering of Q; (b) if p K, then p = {i T : pi G } is not forced to be empty mod D;
Q

(c) if T is a Q-name, T p not forced to be empty mod D, then for some q K,

q p and
Q

q T ;

(d) (I) if

p ( i

< () < ) is an increasing sequence in K and

<()

T D + then

revision:1996-05-05

there is an upper bound in K or (II) if p ( < () < ) is a sequence from K, T D + , T =

T , [ < , i T T p p ] then there is i i

<()

p K, Tp T , [i T Tp p p ]; i i (e) for every g Q, s.t. q

(D Q )+

there is p K, pi q and

p ;

(f) Q |i < K, where Q is the minimal element of Q. 6

310

If ()3 D,K,Q holds then also the following statements are true: (1) If D is -presaturated in V , { : cf (in V )} D, then DQ is -presaturated in

V Q. (2) If player II wins in Gm(, D, ) in V ( ), Q -complete then he wins in Gm(, D, ) in V Q provided that (a) p K pi
Q

()4 D,K,Q

= mod D Q

for i Tp

(b)

()3 D,K,Q

(3) The same holds if we replace The player II wins by Player I does not have winning strategy or the game Gm(, D, ) is replaced by Gm+ (, D, ). (4) In (2) we can replace ()3 D,K,Q by: D1 a normal lter over in V, K1 like K, player II wins also in Gm(, D, ) and

()5 D,K,D1 ,K1 ,Q :

modified:1996-05-06

() () ()

()4 1 ,K1 ,Q D ()4 D,K,Q for every p = pi : i Tp K there are q = q : S K 1 p = p : i T K i T Tp pairwise disjoint [i T pi p ], q = mod D Q . i
p

Proof:

Let us prove (1). The proof of (2) is similar. Let () < , I = {A : i < i }
i

( < ()) be () Q-names of maximal antichains of D + as mentioned in the denition of -presaturitivity (i.e. it is forced that they are like that). We dene by induction on (), Y , j() and p ( Y ) such that (i) Y is a set of sequence of ordinals of length ; (ii) p K; (iii) < , Y implies | Y , p| p (i.e. T T | ), [i T p|i p ]; (iv) Y0 = {<>}, p<> = p<> : i < , i p<> = Q (the minimal element); i

revision:1996-05-05

(v) for limit, Y = { : a sequence of ordinals g() = , (i < )|i Yi }; (vi) {T : Y } is a maximal antichain; 7

310

(vii) for limit, Y , T =

<

T | ; i A .
Q j()

(viii) for i T , Y+1 , p i

There is no problem to do this for limit, (vi) is preserved as < [ < ] and { : cf } D by an assumption and Fact 1.3. Now as D is -presaturated there are B D + , y Y , |y | , such that [ Y y B T D + ]. So there is C D such that ( < ())(1 = 2 y ) [T 1 T 2 C = ]. Let B =
<() y

(T C), then B B mod D, hence B D + .

Apply (d) and get p . The following proposition shows that it is possible to remove the assumptions on conality used in Proposition 1.10(1). Proposition 1.11. Suppose < are regular cardinals, Q a -complete forcing notion, is regular cardinal. Let D Q be the Q-name of the normal lter on which D
Q

generates in V Q . Assume that the following principle holds: ()6 D,K,Q : (a) K is a set, its elements are of the form p = pi : i T , pi Q, T = Tp D + ;
modified:1996-05-06

(b) if q Q, T D + then there is p K, i q pi , T Tp ; (c) let = {i : pi G} we assume pi

p p

(D Q )+ (or just

/ (D Q )+ );
p

(d) if p K, T Tp , T D = ri : i T

and iT pi qi Q then for some T (D Q )+ , there is p K, pi q, q

T and

K iT qi ri ;

(e) for every q Q, s.t. q

p ;

(f) Q : i < K, where Q is the minimal element of Q.

revision:1996-05-05

If D is -presaturated then D Q is -presaturated.

Remark. p K. Proof:

We can omit (b) and get only

/ D Q | T
i

is not -presaturated if

Let () < , I = {A : i < i } for < () are () Q-names of maximal

i

antichains of (D Q )+ which form a counterexample to -presaturativity (i.e. at least some q0 P forces this). 8

310

We dene by induction on () Y and the function j|Y : Y ordinals and p (v Y ) s.t. (i) Y a set; (ii) pv K let pv = pv : i T v , i T v D+ ;

(iii) for every v Y and i T v q0 pv ; i (iv) T v : v Y } is a maximal antichain of D + ; (v) for v Y , i T v , pv i Tv T Cv, ) [p pv ]; i i (vii) < Y Y = ; (viii) for < , v Y |{ Y : T T v D + }| . If , v
<

i A ;

(vi) if Y , < , v Y , and T T D + then for some Cv, D, (i

j(v) v

Y and T T v D + , then we choose C,v D disjoint to T T v . (It

occurs when = v Y .) Arriving at , let {pv : v Y } be maximal such that (i), (ii), (iii), (v), (vi), (vii), (viii), holds and {T v : v Y } is an antichain on D + (not necessarily maximal). It suces to prove that it is a maximal antichain, so let T D + , T T v = mod D
modified:1996-05-06

for every v Y . As D is -presaturated, there is T T , T D + s.t. for every < , { Y : T T D + } has cardinality , and let it be {,j : j < j }. Let C = { < : if 1 2 < , j1 < , j2 < , then C1 ,j1 ,2 ,j2 }. Now for each < , let T = { T : (j < j )[ T ,j ]}. Then T T and T T D + (as {T v : v Y } is a maximal antichain). Let T = C T
<

T . So T T D + . Hence T D + ,
,j

T T . For every T and < , there is a unique j < , T

(if j1 , j2 are

revision:1996-05-05

candidates use the denition of C,j1 ,,j2 (there is one as T ). Now for 1 < 2 < (for our ) by the choice of C1 j1 ,2 j2 p

1 ,j 1

2 ,j 2

. Hence using -completeness of

Q, we can nd p , < p p . Now p : T satises much, almost contradicting the maximality of {pv : v Y } and non-maximality of {T v : v Y } (and the choice of a j). But repairing this is easy. Without loss of generality T T 0,0 . Using (d) of ()6 D,K,Q we obtain p = p | T a K s.t. T a T and iT a pi pi . We have to take care of (v). So we have p = p : T a K vY T a T v = , p satises everything except 9

310

(v). Now

/ p (D + )Q so for some r Q r

p (D + )Q . Hence for some r1 ,

r r1 Q, and j the following holds: () 1 p A D + .

j

For each i T choose if possible pi , pi pi Q pi

i A . Set T b = {i : pi j

is dened}. It is necessarily in D + (otherwise this contradicts ()). Applying (d) of

v ()6 K,Q,D we get a nal p contradicting Y maximal but {T : v Y } is not.

For = 0, use (b) with our q0 .

Proposition 1.12. Let D, D1 are normal lters over a regular cardinal . Suppose that the following holds CD,D1 : for every T D+ ,
+ F (T ) = { < : T is stationary} D1 .

Let Q be a -closed forcing for a regular < . Then in V Q CD,D1 is not forced to fail provided that ()7 D,K,D1 ,Q ()()6 D,K,Q (1 ) for
p

or ()3 D,K,Q

and

pK
+ is forced to be stationary} D1

{ < :
modified:1996-05-06

or

(2 )
p

for

pK
+ is not forced to be nonstationary} D1

{ < : and Proof: Q

satises

c.c .
1

Suppose otherwise. Let

S (D Q )+ and F (S ) (D Q )+ . Set T = {i <

| some pi forces i S }. Using (d) for p = : i < (see (f)) nd r K, r = ri |i T s.t. T T and ri pi . Then

revision:1996-05-05

S . If (1 ) holds, then we are done. If (2 ) is true,

then by -c.c. of Q and normality of D1 nd C D1 so that also is in

+ D1 .

C F (S ) = . Then

C F ( ) = . The set { C| for some q , q

r

Hence it is nonempty. So there is C s.t. q

r

Lemma 1.13. Suppose that is a regular cardinal, > is an inaccessible Q = Col(, < ), and D is a normal lter on . Let T D + and p = pi |i T be a sequence of conditions 10

310

then q

F ( ) and C which contradicts the choice of C.

is stationary. But

is stationary}

in Q. Then there is C D such that for every i T C pi

p

(D Q )+ , where

= {i : pi G}.

Proof:

Set S = {i T : pi

/ p (D Q )+ }. If S D, then let C = S. It will

be as required since ( p

and

p C)

i.e and p are the same mod D Q where = {i T C|pi G}. Let us now assume that S D . For every i S there is qi pi qi
Q + + p

(D Q )+ .

Set = {i S : qi G}. Since Q satises -c.c. and D is -complete, there exists q forcing
Q +

(D )

. Then for some q q, i0 S q qi0 . So q

p

{i S : qi G

} (D ) (D )
Q +

. Hence q

(D Q )+ . Which is impossible since qi0

. Contradiction. So S D + . It is possible to replace the Levy collapse by any -c.c. forcing.

Remark 1.13A.

Proposition 1.14. Suppose that is a regular cardinal > is an inaccessible, Q is the Levy collapse Col(, < ) and D, D1 are normal lters over . Then
+ 1) ()3 D,K,Q holds if { < |cf } D and we let K = { pi |i T |T D , pi Q};
modified:1996-05-06

+ 2) ()6 D,K,Q holds if we let K = { pi |i T C |T D , pi Q, C is as in Lemma 1.13};

3) ()4 D,K,Q holds if { < |cf } D and K is as in 2); 4) ()7 D,K,D1 ,Q holds if { < |cf < } D, { < | is an inaccessible } D1 and K is as in 2). Proof: 1), 2), 3) easy checking. 4). Suppose that in V CD,D1 holds. Let us show the is stationary.

condition (2 ) of ()7 D,K,D1 ,Q . Let p = pi |i T K and C D1 . We should nd some

revision:1996-05-05

C and q Q q

Using CD,D1 nd an inaccessible C such that C, T is stationary and for every i < pi Col(, < ). Now, as in Lemma 1.13, there is a stationary S T such that every pi (i S) forces in Col (, < ) that is stationary. If D concentrates on
p

ordinals of conality < , then without loss of generality all elements of T are of some xed conality < . The forcing Col(, < ) is -closed, so it would preserve the stationarity of

(p )Col(,<) . 11

310

Corollary 1.15. Let be an inaccessible, < be a regular cardinal and Q = Col(, < ). Then the following properties are preserved in the generic extension (i.e. for any normal lter D on ) (1) -presaturatedness, for ; (2) the existence of winning strategy for II in games dened in Denition 1.7; (3) the nonexistence of winning strategy for I; (4) reection of stationary subsets of ordinals of conality < provided that is Mahlo and in V the reecting ordinals are inaccessibles. It is possible to use (4) in order to give an alternative to the Harrington-Shelah [H-S] proof of every stationary subset of 2 consisting of ordinals of conality reects from a Mahlo cardinal. Let be a Mahlo cardinal. Use the Backward Easton iteration in order to add + Cohen subsets to every regular < . Now iterate the forcing for shooting clubs through compliments of nonreecting subsets of { < |cf = } as it is done in [H-S]. Such forcing will preserve all the cardinals. Since it is possible to pick a submodel N s.t. = N is an inaccessible cardinal and use Cohen subsets of for the denition of N -generic clubs. We refer to [S1] for similar construction with .
modified:1996-05-06

Let us now give an example of a forcing notion satisfying the preservation conditions but not -c.c. The forcing notion we are going to consider was introduced by J. Baumgartner [B] and in a slightly dierent form by U. Avraham [A]. Proposition 1.16. Suppose < are regular cardinals, < (< < ), D is the closed unbounded lter on restricted to some conalities (or to a stationary set consisting of ordinals of such conalities) and S D is a stationary subset of containing

revision:1996-05-05

all ordinals of conality < . Let Q = {A|A is a set of pairwise disjoint closed intervals of cardinality < , and [, ] A implies S} ordered by inclusion (so
Q Q Q

contains a club C = {| for some [, ] G ). Dene K = { pi |i T |T D + , T { S|cf }, pi Q for some j i, [i, j] pi and pi i T }.
3 4 Then ()6 D,K,Q , ()D,K,Q , ()D,K,Q hold.

p (D Q )+ for every

12

310

Proof:

Let p = pi |i T be so that T D + , T { S|cf }, pi Q and for some / is not stationary. Suppose otherwise. Let
p

j i, [i, j] pi . Let us show that E be a name of a club such that

E = . For every i T pick qi pi forcing

p

i E . Pick an increasing continuous sequence Mi |i < of elementary submodels of some H( ) for big enough so that |Mi | < and Mi . Set C = {|M = }. Clearly, C is a club. Pick C T . Then q < , r q r since r r Q. So for every p as above some q Q forces is stationary. Now the arguments of
p

E since otherwise for some

E . But since cf and |r| < , for some , < <

there exists r Q M , r r| forcing E . This leads to the contradiction

Lemma 1.13 apply. So for every p = pi |i T as above there exists C D such that for every i T C pi
p

(D Q )+ .

Let us show that for any q Q, T s.t. q and q p T mod(D )

Q +

T (D Q )+ there exists p K, pi q

. The set T = {|for some p q, p

T C }
Q

is in D + . But if p

C , then for some [, ] p . Now x p | T as above

Q

in order to obtain p K as required.

modified:1996-05-06

The checking of the rest of the conditions is routine.

What happens if D concentrates on small conality? Does forcing with Q of Proposition 1.6 preserve < -presaturedness? Strengthening the assumptions it is possible to obtain a positive answer. Namely the following holds. Proposition 1.17. Let , , S, Q be as in Proposition 1.16, assume that D is a club lter restricted to some conality < (or to a stationary set of such conality) and there exists a set S D consisting of ordinals of conality < and for every S there is a set

revision:1996-05-05

A , |A | < consisting of ordinals of conality , and the following holds: () for every club C { S | sup(A C) = } D. Let K = { pi |i T |T D + , pi Q, T S , for every i T ( < i)( Ai )()([, ] pi )}. Then ()6 ,K,Q holds, where S = { S | sup(A C ) < }. D|S

Remarks. (1) The property () is true in L and also it can be easily forced. 13

310

(2) If we are interested only in a condition forcing < -presaturation then there is no need in S .

Proof:

Let us check that for p = pi |i T K

p

/ p is not stationary. Suppose

otherwise, let C be a name of club disjoint to . Pick qi pi forcing i C . Let M | < and C be as in the previous proposition. Pick C T such that C . So for

otp(C ) = and sup (A C ) = . By the assumption q some r q and < r

C . But it is possible to nd (A C )

such that r| M . So some r r|, r M forces C . But r r Q.

Contradiction. 2. Constructions of Cardinal Preserving Ideals.

By Jech-Magidor-Mitchell-Prikry [J-M-Mi-P] it is possible to construct a model with a

normal -preserving ideal over + for any regular from one measurable. Actually N S+ r can be such ideal. We shall examine here N S+ for < and N S+ .

In order to formulate the results we need the following denition of [G4]: Denition 2.0.
modified:1996-05-06

Let F =< F (, )| < ) be a sequence of ultralters over . Let

< , > 0 be ordinals and be a regular cardinal. Then (a) is an up-repeat point for F if for every A F (, ) there is , < < such that A F (, ); (b) is a (, )-repeat point for F if (a1) cf = and (a2) for every A {F (, )| < + } there are unboundedly many s in such that A {F (, )| < + }. If F is the maximal sequence of measures of the core model we will simply omit it.

revision:1996-05-05

Theorem 2.1. The exact strength of (1) N S+ is -preserving for a regular > 2 + GCH or;
0 (2) N S+ is -preserving for a regular > 2 + GCH or;

(3) N S+ is 2 -preserving for a regular > 2 + GCH or;

r (4) N S+ is 2 -preserving for a regular > 2 , < + GCH

is the existence of an up-repeat point. 14

310

Proof:

Use the model of [G4] with a presaturated N S over an inaccessible and the

Levy collapse Col(, < ). By the results of Section1, N S will be -preserving in the

generic extension. an (, + + 1)-repeat point. The following theorem will then follow. Theorem 2.2.

The rest of the section will be devoted to the construction of N S 1 -preserving from

(1) the strength of N S 0 is 1 -preserving + is an inaccessible is an (, + + 1)-repeat

point; (2) the existence of an (, + + 1)-repeat point is sucient for N S is 1 -preserving + is an inaccessible +GCH;
0 (3) the strength of N S+ is 1 -preserving +GCH for a regular > 2 is (, )-repeat

point; (4) the existence of an (, + 1)-repeat point is sucient for N S+ is 1 -preserving +GCH, where is a regular cardinal. Suppose that U is a coherent sequence of ultralters with an (, + + 1)-repeat point at . Dene the iteration P for in the closure of {|( = ) or ( < and is an inaccessible or = + 1 and is an inaccessible)}.
modified:1996-05-06

On the limit stages use the limit of [G1]. For the benet of the reader let us give a precise denition. Let A be a set consisting of s such that < and > 0() > 0. Denote by A the closure of the set { + 1 | A} A. For every A dene by induction P to be the set of all elements p of the form p | g where (1) g is a subset of A.

revision:1996-05-05

(2) g has an Easton support, i.e. for every inaccessible , > | dom g |; (3) for every dom g p = p | g P and p

p Q .
P

Let p = p | g , q = q | f be elements of P . Then p q (p is stronger than q) if the following holds: (1) g f (2) for every f p

p q in the forcing Q
P

15

310

(3) there exists a nite subset b of f so that for every f \b, p Q .

p q

in

If b = , then let p q. Suppose that is an inaccessible and P is dened. Dene P +1 . Let C( + ) be the forcing for adding + -Cohen subsets to , i.e. {f V P |f is a partial function from + into , |f |V
P

< and for every < + { |(, ) dom f } is an ordinal}. P +1 will be

P C( + ) P(, O U ( )), where P(, O U ( )) is the forcing of [G1,2] with the slight change described below. The change is in the denition of U (, , t) the ultralter extending U (, ) for < O U ( ) and a coherent sequence t or more precisely in the denition of the master conditions
sequence. Let j : V N V /U (, ) for < O U ( ). Pick some well ordering W of =

V , for a big enough so that for every inaccessible < , W |V : V . Let be some
xed ordinal below O U ( ). Let us for a while drop the indices , in j , N .

Let <D

| < + > be the j(W )-least enumeration of all E-dense open subsets

of j(P C( + )) which are in N . Where a subset D of a forcing notion P E-dense is if for every p P there exists q D which is an Easton extension of p. Dene an Eincreasing sequence <p
modified:1996-05-06

| < + > of elements of Pj( ) C(j( )+ )/P +1 so that for every | < > in D

<

there will be a j(W )-least E-extension of <p

compatible

with j (G C( + )). Now as in [G1,2] set A U (, , t) if for some r in the generic subset of P C( + ), some < + , a name A of A and a P C( ) name T , in N

r {< t,

T >} p

j(A) .

Note that the set D = {q|q||( j(A)} is E-dense and it belongs to N . So it appears in

revision:1996-05-05

the list <D

| <

>. Hence some p

is in D. Recall that by [G2] the sequence <

Let G be a generic subset of P C(+ ).

U (, , )| < O U () > is commutative Rudin-Keisler increasing sequence of ultralters, for every . Now let j : V [G] M V [G] /U (, , ). Then M = M [j (G)] for a model M = of ZF C which is contained in M . We would like to have the exact description M . 16

310

The next denition is based on the Mitchel notion of complete iteration see [Mi 1,2,3]. Denition 2.3. Suppose that N is a model of set theory, V a coherent sequence in N ,

is a cardinal. The complete iteration j : N M of V at will be the direct limit of j : N M , <

for some ordinal

where

j , M are dened as follows. Set

M0 = N , j0 = id, V0 = V , C0 (, ) = for all and . If O V () = 0, then Suppose otherwise: Case 1. O V () is a limit ordinal.

= 1.

If j , M , V , C are dened then set j+1 : M M+1 M /V ( , ) , = where is the minimal ordinal so that (i) ; (ii) is less than the rst > with OV ( ) > 0; (iii) for some < O V () C (, ) is bounded in ; and is the minimal satisfying (iii) for . If there is not such an then set
modified:1996-05-06

= , j = j , M = M , C = C . Dene

C+1 |( , ) = C |( , ), +1 = j+1 ( ) C+1 (+1 , j+1 ()) = C ( , ), if = C ( , ) { }, if =

Case 2.

O V () = + 1.

Dene j , M , V and C as in Case 1, only in (ii) let be less than the image of

revision:1996-05-05

under the embedding of N by V (, ). Theorem 2.4. Let j : V [G] M be (a) the ultrapower of V [G] by the ultralter U (, , ) or (b) the direct limit of the ultrapowers with the Rudin-Kiesler increasing sequence of the ultralters < U (, , )| < >. 17

310

Then M is a generic extension of M , where M is the complete iteration of U |(, +1) at , if (a) holds, and of U |(, ) at , if (b) holds. Main Lemma. Let M be as in the theorem and let i be the canonical embedding of V into M . Then there is G i(P C(+ )) so that G V [G] and G is M [G] generic. Proof: Let us prove the lemma by induction on the pairs (, ) ordered lexicographically.

Suppose that it holds for all (, ) < (, ). Let us prove the lemma for (, ). Case 1. = + 1.

Let N be the ultrapower of V by U (, ) and j : V N the canonical embedding.

We just simplify the previous notations, where N = N , j = j . Then M is the complete

iteration of j(U )| + 1 at in N if is a limit ordinal, or for successor , the above iteration should be performed -times in order to obtain M . Let us concentrate on the rst case; similar and slightly simpler arguments work for the second one. Denote by k : N M the above iteration. Then the following diagram is commutative N j V i M By the inductive assumption there exists G N [G] M -generic subset of Pk() C((k())+ ). If = 0 then k( ) = 0i(U ) (k()) = 0 and C(k()+ ) is only the forcing used over k(). Suppose now that > 0. Then k( ) > 0 and the forcing over k() is C(k()+ ) followed by P(k(), k( )). Let us dene in N [G] a M [G ]-generic subset of P(k(), k( )). Recall that a generic subset of P(k(), k ()) can be reconstructed from a generic sequence bk() to k(), where bk() is a combination of a conal in k() sequences so that b1 ({n}) k() is a sequence appropriate for the ultralters i(U )(k(), ) with on the depth n. We refer to [G2] for detailed denitions. 18 k

310

revision:1996-05-05

modified:1996-05-06

Let us use the indiscernibles of the complete iteration C in order to dene bk() . Namely, only < C(k(), )| < k( ) > will be used. Set bk() = {< , n, > |n < , < k()+ , < k(), for some < k( ) coded by < n, > in sense of [G2] C(k(), ))}. Let G be the subset of P(k(), k( )) generated by bk() . Let us show that G is M [G ]-generic subset. Let D M [G ] be a dense open subset of P(k(), k( )) and D its

canonical name. Since M is the direct limit for some less than the length of the complete iteration, for some D M , D is the image of D . Let us work in M . Denote by G

the appropriate part of G and by t a coherent sequence which generates bk() | . Let k : M M be the part of the complete iteration k on the step . Case 1.1. cf M (k ( )) < k ( ).

Then k ( ) changes its conality to cf M (k ( )) after the forcing with P(k (), k ( )). For every T so that < t, T > P(k (), k ( )), there exists i < cf M (k ( )) and T so that < t, T > is a condition stronger than < t, T > and < t, T > forces some < t , T > D with t on the level i is in the generic set. In order to nd such T just use the k ( )-completeness of the ultralters involved in the forcing P(k (), k ( )) and
modified:1996-05-06

the Prikry property. Now for every t T which is appropriate for i or some j i there exists T such that < t , T > D . The same property remains true for k (T ) for every < length of the iteration k. Pick to be large enough in order to contain elements of bk() appropriate for i. Let t be a coherent sequence generating bk() | . It is possible to pick such t in k (T ). But then for some T < t , T k (T ) > k (D ). The image of < t , T k (T ) > under the rest of the iteration will be in G . Case 1.2. cf M (k ( )) = k ( ).

revision:1996-05-05

Then k ( ) changes its conality to . For every T so that < t, T > P(k (), k ( )) there exist n < and T so that < t, T > is a condition stronger than < t, T > and < t, T > forces some < t , T > D , with t containing the rst n elements of the canonical -sequence to k (), is in the generic set. Pick in order to rst reach n elements of the canonical -sequence to k(). 19

310

Then proceed as in Case 1.1. Case 1.3. cf M (k ( )) = (k ( ))+ in M .

Let D be a dense open subset of P(k (), k ( )). Set D = {< p, T > P(k (), k ( ))| for some level < (k ())+ , for every < 1 , . . . , n > s.t. n SucT, (p < 1 , . . . , n1 ) for some , < p < 1 , . . . , n >, Tp <1 ,...,n > > D}. Claim. D is a dense open. The proof is similar to Lemma 3.11 of [G1]. Dene D1 = D , for every n < set Dn+1 = Dn and nally D = n< Dn . Claim. For every condition < p, T > there exists a stronger condition < p, T > in D . We refer to Lemma 1.4 [G1] for the proof. Just replace there by D . Let < t, T > D . Then for some n < t, T > Dn . Now by simple induction it is possible to show that there is < (k ())+ so that for every SucT , (t) < t < >, Tt <> > D. Now pick a large enough part of the iteration k to reach the level . Continue as in Case 1.1.
modified:1996-05-06

It completes the denition of a generic subset of Pk()+1 . Let us refer to it as Gk()+1 . We now turn to the construction of the generic object for the forcing between k() + 1 and i() + 1. Let us dene it by induction on , k() + 1 i() + 1. Suppose that for every < a M -generic subset G of P is dened in V [G]. Dene G . If there is some > and an indiscernible for it , then use the inductive assumptions to produce G . Suppose now that there is no , as above. Notice, that then = k( ) for some . Let us split the proof according to the following two cases.

revision:1996-05-05

Case A.

= + 1.

Then P /G = C( + ) P(, 0k(U ) ()) if 0k(U ) () > 0 and P/G = C( + ) otherwise. So G will be G (G G ), where G may be empty. Let us dene rst G C( + ). We use as inductive assumption that k(p ) | G for every < + where < p | < + > is the master condition sequence for U (, , ). Set G = {k(p )()|C( + )| < + }. 20

310

Let us check that G is M [G ]-generic subset of C( + ). Suppose that D M [G ] is a dense open subset of C( + ). Let D be a canonical P -name of D. By the assumption

on there are indiscernibles < 1 < < n < such that the support of D is a

subset of {, 1 , . . . , n }. Since the forcing C( ) is -closed D can be replaced by it

dense open subset with support alone. Let us assume that D is already such a subset.

Then D = k(D ) for dense open D N subset of C(( )+ ). By the choice of the master

condition sequence for some < p | But this implies

p ( )|C(( )+ ) D .

k(p )()|C( + ) D . So G D = . Suppose now that 0k(U ) () > 0. We need to dene G a M [G G ]-generic subset of P(, 0k(U ) ()). In this case is a limit of indiscernibles, i.e. C(, ) is unbounded in for every < 0k(U ) ()). So we are in the situation considered above. The only dierence is that some p s may contain information about the generic sequence b to . In order to preserve k(p )| + 1 in the generic set, we need to start b according to k(p )(). Notice, that
modified:1996-05-06

further elements of the master condition sequence do not increase the coherent sequence given by p . Case B. is a limit ordinal.

Set G = {p P | for every < p| G }. Let us show that G is M -generic subset of P . Consider two cases. Case B.1. There are unboundedly many in indiscernibles for ordinals .

revision:1996-05-05

Since = k( ), is a limit of indiscernibles for . Then is measurable in M and the direct limit is used on the stage . Let < i |i < > be a conal sequence of indiscernibles. Let D M be a dense open subset of P . Then { < |D H( ) is a dense open subset of
<

P } contains a club in M . The sequence < i |i < > is almost contained

in every club of M so there is i0 s.t. D P0 is dense. But then G0 D = . Hence G D = . 21

310

Case B.2.

There are only boundedly many in indiscernibles for ordinals .

Then, since = k( ), there is no indiscernibles for ordinals below . So there are unboundedly many in ordinals which are in the range of k. Let D be a dense open subset of P in M . Dene D = {p P | for some < p| Dn .
n<

p\ D/ G }. Set D0 = D, Dn+1 = Dn for every n < and let D =

Claim B.2.1. D is E-dense subset of P . Proof: Let p P dene p p as in Lemma 1.4 [G1] where is replaced by belonging E

to D . Suppose that p D . Then there is p D , p p . Let domp domp be the last on which an information about conal sequences is added. But then for some p P p

p ()

p \ D1 /G . Then, as in Lemma 1.4 [G1], it is possible to go

down until nally for some n and some p Pmin(domp ) p which contradicts the denition of p .
modified:1996-05-06

p Dn /Gmin

(domp )

of the claim.

It is enough to show that G D = . Since then, for some n, G Dn = . Now use the fact that all initial segments of G are M -generic. So let us prove for every E-dense set D that G D = . Without loss of generality we can assume that D is in the range of k, since it is always possible to nd some < in the range of k above the support of D and the intersection of E-dense open subsets of P /G is E-dense open.

revision:1996-05-05

Let D N be an E-dense open subset of P so that k(D ) = D. By the denition of the master condition sequence < p | < + >, for some 0 < + p0 | D . Then k(p0 )| D. But, by the choice of < G | < > k(p0 )| G for every < . So k(p0 )| G . So G is an M -generic subset of P . It completes Case B and hence also Case 1 of the lemma. 22

310

Case 2.

is a limit ordinal.

Use the inductive assumption and the denitions of the generic sets of Case 1. Dene a generic subset of Pk() as in Case B.1. of the lemma.

Let G V [G] be the M -generic subset of Pi() C(i()+ ) dened in the Main Lemma. Then G P C(+ ) = G and for every < + i(p ) G . Dene the elementary embedding i : V [G] M [G ] by i (a [G]) = (i(a ))[G ]. Then i |V = i,

i (G) and, if U = {A | i (A)} then U = U (, , ). So the following diagram is commutative M [G ] i V [G]

M V [G] /U = where ([f ]U ) = i (f )().

modified:1996-05-06

It remains to show that

= id. Notice, that it is enough to prove that every indis-

cernible in C is of the form i (f )() for some f V [G]. Examining the construction of G , it is not hard to see that every indiscernible is an element of the generic sequence b for some k (N ). So indiscernibles are interpretations of forcing terms with parameters in k (N ). But such elements can easily be represented by functions on in V [G]. Let C and

revision:1996-05-05

= (t())[G ] for = k( ), where t is a term. Let = [f ]U (,) for some f : , f V . Dene a function g V [G] on as follows: g( ) = t(f ( ))[G]. Then i (g)() =

t(t(f )())[G ] = (t())[G ] = .

Let us now turn to the construction of 1 -preserving ideal. Suppose that is an (, + + 1)-repeat point for U in V . Let < + + be an ordinal. Denote by M
the complete iteration of j (U )|((j (), (j ( + + )) in N . Let i , k be the canonical

23

310

elementary embeddings making the diagram

N

jk

V i

commutative. As in the Main Lemma nd in V [G] a P(, )-name of a M -generic subset G of

Pi () C(i (+ ))/P+1 so that each k (p ) G for < + , where < p | < + > is the

master condition sequence for U (, , ). Let G be obtained from G by removing all the

information on the generic subset of C(i (+ )) except i (G C(+ )). We shall dene a presaturated lter U (, ) on in V [G] extending U (, ) so that (i) all generic ultrapowers of U (, ) are generic extensions of M ; (ii) every set U (, , ) is U (, )-positive. Property (i) will insure that the forcing for shooting clubs over i () will be -closed
modified:1996-05-06

forcing. Property (ii) is needed for the iteration of forcings for shooting clubs over in V [G]. Denote G P by G and G C(+ ) by G. Let G() denote the -th function of G for < + . Let j be the embedding of V [G] into the ultrapower of V [G] by U (, , ). Let < | < + > be the list of all the indiscernibles for i () of the complete iteration used to dene M . For every < + extend G to G by adding to G

conditions

revision:1996-05-05

< + + 1, , > for every < + , i.e. the + + 1 generic function moves to . Dene a lter U (, ) on in V [G] as follows: A belongs to U (, ) i for some r G some r s.t.

r G , for every

p C

(j (+ ))(0)

s.t. in

N [j (G)]

p j (G(0)), and for some

s.t.

r G

(p )

in M r 1P(,) p r
P i ()+1

i (A)

24

310

where (p) < + is the minimal s.t. p| j (G(0)), for the (p) member of some xed
enumeration of j ().

Claim 1. U (, , ) U (, ). Proof:
Let A U (, , ), then for some < + some r, < , T > P(, ), in N j (A) .

r {< , T >} p

Let us split p into three parts p =p | Pj , p =p C(j (+ ))(0) and p = the rest of

p . Then, using k , in M

r {< , T >} k (p )

i (A) .

Extend p to some p , still remaining in the master condition sequence, in order to make its domain above all the indices of the generic sequences listed in p . Extend p2 to some
3 2 2

p2

C(j (+ ))(0)

which is incompatible with a member of the master condition sequence.

Then, using k , in M r {< , T >} k (p ) k (p2 ) k (p )

1 3

modified:1996-05-06

i (A) .

By the denition of U (, ), it implies that A is a U (, )-positive set. So every member of U (, , ) is U (, )-positive. The fact that U (, , ) is an ultralter completes the proof of the claim. Claim 2. U (, ) is normal precipitous cardinal preserving lter and its generic ultrapower is a generic extension of M .

revision:1996-05-05

Proof:

Let U = U (, )V [G, G(0)]. The arguments of Theorem 1 show that a generic

ultrapower by U is isomorphic to a generic extension of the complete iteration of N with j (U )|j () above . Actually the forcing with U is isomorphic to P(, ) followed by C(j (+ ))(0) (in the sense of this iteration) over V [G, G(0)]. Clearly, the generic function form j (+ ) into j (+ ) produced by this forcing is incompatible with j (G(0)) which

belongs to V [G, G(0)]. 25

310

Now the forcing with U (, ) over V [G] does the following: First it picks p C
(j (+ ))(0)

incompatible with j (G(0)) and then

G ) (p

is added.

G ) (p

insures that a

generic ultrapower is a generic extension of M . So the forcing with U (, ) is isomorphic

to P(, ) followed by a portion of C(i (+ )), which is -closed.

Force over V [G] with the forcing Q which is the Backward-Easton iteration of the forcings adding + -Cohen subsets to every regular < with 0U () > 0. Fix a generic subset H of Q . All the lters U (, ) extend in the obvious fashion in V [G, H]. Let us use the same notations for the extended lters. Set F = {U (, )| + + }. Then F is a -preserving lter in V [G, H] and forcing with it is isomorphic to P(, ) (for some , + + ) followed by -closed forcing. Let us shoot clubs through elements of F , then through the sets of generic points and so on, as was done in [G 3,4]. Denote this forcing by B. Let R be its generic subset over V [G, H]. We shall show that N S is 1 -preserving ideal in V [G, H, R]. The forcing with N S consists of two parts: (a) embedding of B into P(, ); (b) (-closed forcing) (i (B)/i
modified:1996-05-06

(R)).

By the choice of i (), the forcing i (B) is a shooting club through sets containing a club (the club of indiscernibles for i ()). So it is -closed forcing and part (b) does not cause any problem. Let us examine part (a) and show that this forcing preserves 1 . Recall that B is the direct limit of < B | < + > where each B is of cardinality and for a limit , B = the direct limit of B ( < ), if cf = and B = the inverse limit of B ( < ) otherwise. The forcing of (a) is

revision:1996-05-05

P = { V [G, H]| for some < + is an embedding of B into P(, )} . For 1 , 2 P let 1 2 if 1 | dom 2 = 2 . Claim. N S is an 1 -preserving ideal in V [G, H, R]. Proof: Suppose otherwise. Then for some + + some condition in the

forcing P(, ) (the forcing isomorphic to the adding of + -Cohen subsets of + ) P over 26

310

V [G, H] forces 1 to collapse. Let us assume that the empty condition already forces this. Consider the case when P(,) cf = . The remaining cases are simpler. Let S be a V [G, H]-generic subset of P(, ) (+ -Cohen subsets of + ). Denote
V V [G, H, S] by V . Let f be a P-name in V of a function from to 1 . Pick an elementary

submodel N of < H(), , B, P, , f >, for big enough, satisfying the following three

conditions: (1) |N | = ; (2) N H V (); (3) N + = for some s.t. cf V [G,H] = . Then B N = B . Also B is a direct limit of < B | < >. Pick in V [G, H] a conal sequence < | < > to and in V a conal sequence < n |n < > to . Consider the subsets of , <A | < > s.t. B +1 /B is the forcing for shooting club

into A . Assume for simplicity that all A s are in V . Let A = < A = { < | for

every < ,

A }. The A s and A are in {U (, )| + + }. Pick

0 P N deciding the value of f (0). Without loss of generality dom 0 = B0 +1 for some 0 < . Let n0 be the least n < such that n > 0 . Denote n0 by 0 . Let C0 be the generic club through A0 dened by 0 and S. As in Lemma 3.6 [G4], it is
modified:1996-05-06

possible to nd an element of the generic sequence to , (0) A n0 such that C is a V [G| (0), H| (0)]-generic club through A . Choosing (0) more carefully, it is possible to also satisfy the following A (0) = < (0) (A (0)). Now pick 0 P N to be an extension of 0 with domain Br such that the clubs of 0 intersected with (0) are V [G| (0), H| (0)]-generic for B | (0). It is possible since N satises condition (2). Now nd in N an extension 1 of 0 deciding f (1). Dene 1 and (1) as above.

revision:1996-05-05

Continue the process for all n < . Finally set =

n<

n . It is enough to show that P,

since then

f V1 . Let us prove that and S produce a V [G, H]-generic subset

of B . Suppose that D V [G, H] is a dense subset of B . Then X = { < |D H( ) is a dense subset of Br | } contains a club in V [G, H]. Since the generic sequence to is almost contained in every club of V [G, H], for some n < , (n) X. But then the

generic subset produced by n intersects D. So the same is true for .

Let us now turn to successor cardinals. We would like to make N S 1 preserving for 27

310

 = + for a regular > 1 . It is possible to use the model constructed above, collapse to + and apply the results of Section 1. But an (, + + 1)-repeat point was used in the construction of the model. It turns out that an (, + 1)-repeat point suces for N S+ and an (, )-repeat point for N S+ |{ < + |cf < }. On the other hand,
0 precipitousness of N S+ implies an (, )-repeat point, by [G4].

Let us preserve the notations used above. Assume that < is a regular cardinal and some < 0U () is an (, + 1)-repeat point. Let G be a generic subset of P C(+ ). Over V[G] instead of the forcing Q , in the previous construction, use Col(, ) the Levy collapse of all the cardinals , < < on . Let H be a generic subset of Col(, ). Denote by H( ) the generic function from on where (, ). Now the forcing for shooting clubs should come. In order to prevent collapsing

cardinals by this forcing, j() was made a limit of + indiscernibles for the measures {U (, )| < + + }. But now we have only measures. So the best we can do is to make j() a limit of indiscernibles and then its conality in V will be < . It looks slightly paradoxical since usually cf j() = + , but it is possible by [G5]. In order to explain the idea of [G5] which will be used here let us give an example of a precipitous ideal I on 1 so that:
modified:1996-05-06

()
I+

cf V (j( 1 )) = .

Example.

Suppose that is a measurable cardinal. Let U be a normal measure on

and j : V N V /U the canonical embedding. Let P be the Backward-Easton = iteration of the forcings C(+ ) for all regular < . Let G H be a generic subset of V . Collapse to 1 by the Levy collapse. Let R be V [G H ]-generic subset of Col(1 , ).

revision:1996-05-05

Denote V [G H R ] by V1 . We shall dene a precipitous ideal satisfying () in V1 .

j () Let j0 = j, N0 = N , 0 = , U0 = U . Set N1 N0 0 /j0 (U ), 1 = j0 (), U1 = j0 (U ) =

and let j1 : N0 N1 be the canonical embedding. Continue the denition for all n < . Set j , N to be the direct limit of < jn , Nn |n < >. Then j () = = j (). Notice, that (j (+ )) = (+ )N0 and (jn+1 ((+ )Nn )) = n 1 (j (+ )) = (+ )N . 28 n . n< (+ )Nn +1 . n+1 Set So

310

Dene a lter U in V [G H R] as follows: A U i for some r G H R for some n < in N r pn j (A)

where pn is the name of condition in the forcing C(j (+ )) dened as follows: let h be the name of the generic function from onto (+ )V in Col(, j ())/G H R .

Set pn = {< j (h (0)), , 1 >, < j (h(1 )), , 2 >, . . . , < j (h (n)), , n+1 >}. The

meaning of the above is that the value on of j (h(m))-th function from j () to j () is forced to be m+1 . It is not dicult to see now that U is a normal precipitous ideal on 1 and a generic ultrapower with it is isomorphic to a V [G H R ]-generic extension of N [G , H , R ]. Also for a generic embedding j , j |V = j . Hence j () = which is of conality in V. As in [J-M-Mi-P], it is possible to extend U to the closed unbounded lter with the same property. Using the Namba forcing, it is possible to construct a precipitous ideal satisfying () on 2 . Starting from a measurable which is a limit of measurable, it is possible to build such an ideal over an inaccessible or even measurable. Since then it is
modified:1996-05-06

possible to change the conality of the ordinal of conality + to in N and that is what was needed to catch all n s in the above construction. We do not know if one measurable is sucient for a precipitous ideal satisfying () over > 2 . Note also that by Proposition 1.5 if I is -preserving, then cf V j() . In particular, if I is presaturated then cf V j() = + . Let us now return to the construction of N S+ 1 -preserving. Let < + be an ordinal. Dene M as in the construction of N S 1 -preserving for an inaccessible . Now cf V i () will be . We shall dene, in V [G, H], U (, ) extending U (, ) satisfying conditions (i) and (ii) from page 23. To do so simply combine the denition there with the denition of the example. We leave the details to the reader. The rest of the construction does not dier from an inaccessible cardinal case. The above results give equiconsistency for N S | (singular) or something close to equiconsistency for N S , but for > 2 . For = 2 , we do not know if the assump29

310

revision:1996-05-05

0 tion of the existence of an (, 1 + 1)-repeat point (or of an (, 1 )-repeat point for N S2 ) 0 can be weakened. By [G3], a measurable is sucient for the precipitousness of N S2 and

a measurable of order 2 for N S2 . Let us show that 1 -preservingness requires stronger assumptions. Lemma 2.5. Suppose that > 1 is a regular cardinal, 2 = + , 20 = 1 . I is a normal 1 -preserving ideal over so that { < |cf = } I. Then 0( ) 2 in the core model. Proof: Without loss of generality, assume that 0() = ++ . Denote by K(F) the

core model with the maximal sequence F . We refer to Mitchell papers [Mi 1,2] for the denitions and properties of K(F ) that we are going to use. The set A = { < | is regular in K(F ) and of conality in V } is I-positive. If the set A = { A| is not measurable in K(F )} is bounded in , then 0() 2. Suppose otherwise. Let j : V M be a generic elementary embedding so that the set { < |cf = } belongs to a generic ultralter GI . Then j(A ) is unbounded in j(). Pick the mimimal j(A ) . Claim. cf K(F) () = (+ )K(F ) . Proof: Suppose otherwise. Then, since is regular in K(F ) and j|K(F) is an iterated

modified:1996-05-06

ultrapower of K(F) by F , is a limit indiscernible of this iteration. By Proposition 1.5.

M V [GI ] M . Since is not a measurable in K(j(F)), it implies that cf K(F ) > .

But then, using M V [GI ] M and the arguments of [Mi 2] and [G4], we obtain some with 0F ( ) (1 )K(F ) . of the claim. By [Mi 2], (+ )K(F) = (+ )V . But in V [GI ], cf = and hence cf (+ )V = which

revision:1996-05-05

contradicts Proposition 1.5.

Lemma 2.6. Suppose that > 1 is a regular cardinal, 2 = + , 20 = 1 and I is a normal 1 -preserving ideal. If there exists an -club C so that every C is regular in K(F ), then 0F ( ) 1 in K(F). The proof is similar to Lemma 2.5; just consider the 1 -th member of j(C) . 30

310

0 Theorem 2.7. Assume GCH, if N S2 is 1 -preserving then 0F ( ) 1 in K(F ).

The proof follows from Lemma 2.6.

References [A] [B] U. Avraham, Isomorphism of Aronszajn trees, Ph.D. Thesis, Jerusalem, 1979. J. Baumgartner, Independence results in set theory, Notices Am. Math. Soc. 25 (1978), A 248-249. [B-T] J. Baumgartner and A. Taylor, Ideals in generic extensions. II, Trans. of Am. Math. Soc. 271 ( 1982), 587-609. [G-J-M] F. Galvin, T. Jech and M. Magidor, An ideal game, J. of Symbolic Logic 43 (1978), 284-292. [G1] M. Gitik, Changing conalities and the nonstationary ideal, Israel J. of Math. 56 (1986), 280-314. [G2]
modified:1996-05-06

M. Gitik, SCH from 0() = ++ , Ann. of Pure 43 (1989), 209-234. M. Gitik, The nonstationary ideal on 2 , Israel J. of Math. 48 (1984), 257-288. M. Gitik, Some results on the nonstationary ideal, Israel Journal of Math. M. Gitik, On generic elementary embeddings, J. Sym. Logic 54(3) (1989), 700-707.

[G3] [G4] [G5]

[H-S]

L. Harrington and S. Shelah, Equiconsistency results in set theory, Notre Dame J. of Formal Logic 26(2) (1985), 178-188.

revision:1996-05-05

[J-M-Mi-P] T. Jech, M. Magidor, W. Mitchell and K. Prikry, Precipitous ideals, J. Sym. Logic 45 (1980), 1-8. [Mi 1] W. Mitchell, The core model for sequences of measures I, Math. Proc. Camb. Phil Soc. 95 (1984), 229-260. [Mi 2] W. Mitchell, The core model for sequence of measures II, to appear. 31

310

[Mi 3]

W. Mitchell, Indiscernibles, skies and ideals, Contemp. Math. 31 (1984), 161-182.

[S1] [S2]

S. Shelah, Proper forcing, Springer-Verlag Lecture Notes in Math, 940 (1982). S. Shelah, Some notes on iterated forcing with 20 > 2 , Notre Dame J. of Formal Logic.

revision:1996-05-05

modified:1996-05-06

32

310