October 6, 2003

Large Normal Ideals Concentrating on a Fixed Small Cardinality

Saharon Shelah Institut of Mathematics Hebrew University of Jerusalem 91904 Jerusalem, Israel and Department of Mathematics Rutgers University New Brunswick, NJ 08854, USA

modified:1994-06-19

The property on the lter in Denition 1, a kind of large cardinal property, suces for the proof in Liu Shelah [484] and is proved consistent as required there (see conclusion 6). A natural property which looks better, not only is not obtained here, but is shown to be false (in Claim 7). On earlier related theorems see Gitik Shelah [GiSh310].

revision:1994-06-19

1. Denition (1) Let be a cardinal and D a lter on and be an ordinal and < but 2 and . Let GM,,, (D) be there following game: Research supported by Basic Research Foundation of The Israel Academy of Sciences and Humanities. Publication 542. 1

542

a play lasts moves, in the s move the rst player chooses a function h from to some ordinal < and the second player chooses a subset B of of cardinality < . The second player wins a play if for every < the set } is = modD. (2) If = 2 we may omit it, if = 2 and = we omit and . 2. Denition: (P , pr ) K,,, i 1. is a regular cardinal. 2. (P, ) is a forcing notion with minimal element (if in doubt we use P , P ). 3. P satises the -c.c. 4. pr is a partial order on P such that: a. p pr q implies p q b. any pr -increasing chain of length < with rst element in P has an pr -upper bound.
modified:1994-06-19

{{ < : h () B } :

c. if < and is a P -name of an ordinal < and pr p P then for some q and B of cardinality < we have p pr q P and q forces B. 5. for any Y P of cardinality < there is P P of cardinality < such that P/P satises condition (4), i.e. if G P is generic over V and P/G = {p P : p compatible with every q G } then
def

revision:1994-06-19

a. in P/G , any pr -increasing sequences starting with of length < have an pr -upper bound in P/G . b. if p P/G and a P -name of an ordinal < where < then there is a subset B of of cardinality < and p , p pr p P/G such that p forces B. 2A Remark: The relation in clause 4(b) is not really stronger than having a winning 2

542

strategy in the corresponding play, see [Sh250, 2.43] (or [Sh-f, XIV 2.4]). 3. Lemma: Assume a. is a measurable cardinal with D a -complete ultralter on it b. (P , pr ) K,,, Then in V P the second player wins GM,,, (D) 3A Remark: 1. We can replace ultralter by a lter in which the rst player wins GM, (D) [see Lemma 5].
Proof: In V we dene a set R, its members are sequences p = p : Ap where Ap D and pr p P (for Ap ). On R we dene a partial order R as follows: p R q i Aq Ap and for every Aq we have p pr q .

Clearly, in V the partial order (R, R ) is -complete.

For G P generic over V we dene R[G] as { : p R and { Ap : p G} = p

modD (in V P , D is not a lter just a family of subsets of but it naturally generates a
modified:1994-06-19

lter- just closed upward and we refer to this lter in mod D}.
For G P generic over V and p R let w[, G] = { Ap : p G}. p def

So R[G] = { R : w[, G] = modD}. We now prove some facts. p p 3B. Fact: In V [G], (R[G], R ) is -complete. Proof: If not then there is a P -name of a sequence of length < , p : < and r P which forces this sequence to be a counter example, so < . So there are maximal antichains I for < of conditions in P forcing a value to p (note p is a P -name of a member of V ); let Y be the set of elements appearing in some I and r. As P satises the -c.c. clearly Y has cardinality < so there is P as required in condition (5) of Denition 2. Let G P be generic over V and r G . 3

542

revision:1994-06-19

p Now working in V [G ] we can (for each < ) compute and A , call it then p p def and A respectively and so A D and A = {A : < } belongs to D V [G ] (=the

ultralter which D generates in V [G ], remember |P | < , D a -complete ultralter); also letting w = { A : there is G P generic over P extending G to which p belongs} V [G ] we know that in V [G] we get a D-positive set w[p , G] (because r forces this) hence in V [G ] the set w is D-positive but in V [G ] we know D V [G ] is an ultralter so necessarily w belongs to D V [G ] ; clearly for < , w we have p P/G . Let B = A {w : < }, it is in D V [G ] . Now for any B the sequence p : <

def

is a pr -increasing sequence of member of P/G and by demand (5) (a) of Denition 2, the sequence has an pr -upper bound q (in P/G ). Let r G be above r and force that this holds and moreover force some specic q P is as above. So, still in V [G ], for some C D, C B and r G we have ( C)[r = r ] without loss of generality q P/Gp , r is compatible with every q + + + ( C). By 3D below for some q , r q P and q + P { : q GP }= mod D. + So q (which is above r r ) force that q = q : C is an upper bound as required.
+ (note: q V , r force it is an upper bound of { : < }; we need q() as we do not p know the value of . p 3B

C V . As for C B, r = r

modified:1994-06-19

3C Fact: Let G P be generic over V . In V [G], if < and p R[G] and h a function

revision:1994-06-19

from to , then for some q we have: a. q R[G] b. p R q c. on w[, G] the range of the function h is of cardinality < . p Proof: Assume the conclusion fails then some r G forces that it fails for a specic p 4

542

 and P -name h ( so in particular r forces that w[, G] = modD.) Let w =: { Ap : p the conditions r, p are compatible in P (equivalently, r does not force w[, G])} (so / p w V ) and w D. Now let P be as in condition (5) of Denition 2 for Y = {r} (so in

particular r P ). Now:
() for every w there are r and q and B such that: a. r r P .

b. p pr q . q P/GP . d. q forces (for P ) that h() B and for some set B (B V ), we have |B | < .
P c. r

[Why? for every in w we can nd G P generic over V to which r and p

belong (as w ); hence p P/(G P ) hence some r G P force this (for P ) so without loss of generality r r (as G P is directed). Now apply
modified:1994-06-19

condition (5) of Denition 2 to G P , p and h() and we get some B . |B| < and q P/(G P ) such that p pr q P/(G P ) and q forces
h() B. Now increasing again r we get ()]. So we can nd for w , r , q and B as in (), (all in V ); let A w be such that

A D and B : w is constant on A and also r is constantly r (note: D is

revision:1994-06-19

-complete w D, and is strongly inaccessible hence ||< < and |P | < . Now some q + , satisfying r q + P , forces that q : A is in R[G] by fact 3D below and so clearly is as required in the Fact 3C.

3C

3D. Observation Assume p = p : A R and r P is compatible (in P ) with every p (for A). Then some r , r r P , force that p R[GP ]. 5

542

Proof: Let I be a maximal antichain of P above r such that for every q I we have w[, GP ] is a subset of Aq where Aq and \ Aq D or q P p w[, GP ] = modD. p So I has cardinality < and if the conclusion fails then always the rst possibility either q
P

holds; now we let B =

def

{ \ Aq : q I}, clearly it belongs to D. Now there is B A

(as B A D) and there is r P above r and above p (exist by assumption); now r force that w[, Gp ] Aq \ B, contradiction. p 3E.Continuation of the Proof of Lemma 3: immediate for the Facts 3B, 3C. Now we shall redo it all in another version:

3D

4. Lemma: (From Gitik [Gi] 3, relaying on 1 there, in dierent terminology). Assume < , < a regular cardinal, is a measurable cardinal of order + 1 (i.e. there is a coherent sequence of ultralters on of length + 1, see [Gi, 3 p.293], with D an
modified:1994-06-19

ultralter on appearing in the th place in the appropriate sequence. Then for some forcing notion P we have (a) P of cardinality , (b) { :
P P

is strongly inaccessible.

cf() = } D

(c) P K,,,2 (in particular P satises the -c.c., pr for P is called E in [Gi]

revision:1994-06-19

(called Easton) (d) For some pr Condition (4) of Denition 2 is satised by P (for = 2). Moreover,
given any < and Y P of cardinality < we can nd P P as in clause

(5) of Denition 2 replacing and by . 5. Claim: Under the assumptions of lemma 4, if + = cf() < let Q = 6

542

P (Levy(, < ))V Levy(, < )V

P

dening (p1 , q1 ) pr (p2 , q2 ) i p1 P p2 and p2

q1 q2

Then Q K,,,2 and in V Q , = + = 2 . Proof: Easy. 5A Remark: Actually in the conclusion of Claim 5 we can weaken + to + + + hence in the conclusion = + (= ) is o.k. This applies also to conclusion 6. 5B Remark: Of course Claim 5 and Denition 2 are formulated so that we get consistency results justifying the name of the paper. We formulate below (conclusion 6) the one used in Liu Shelah [LiSh484].

6. Conclusion: Assume 0 = n0 < n1 < n2 < . . . < n , n +1 < n

+1 ,

+1

is a measurable
+1

of order + 1 and for simplicity GCH holds and stipulate 0 = 0 and regular for
modified:1994-06-19

<

+1

is

< , moreover

+(n +1

+1 n

.
< +1

Then there is a forcing notion P of cardinality 2

+1 ,

which preserves cf(

+1 )

makes

+1

to n

+1

and preserves ( )+i if i < n

+1 ,n +1 1 , +1 ,2

+1 , +1 )

preserves G.C.H. and for < for some D

+1

in V P the second player wins GM ultralter on

+1

(D

V , a normal

of order + 1.

Proof: We use iteration Pi , Qi : i < described as follows: Q =the forcing notion from

revision:1994-06-19

lemma 5 (for =

+1 ,

+1 ,

+(n

+1 n

and

+1

+(n +1

+1 n

1)

), the limit

is a full support for pure extensions of the and nite support otherwise (for the Levy collapse all conditions are pure extensions of ). The checking is standard.

Discussion: We shall now prove that for a natural strengthening of Denition 2, we cannot get consistency results. Specically we cannot, in the game in Denition 2, let 7

542

player I just decrease the present D-positive set.

7. Denition: (1) Let be a cardinal and D a lter on and be an ordinal . Let GM (D) be the following game: a play lasts moves; in the s move rst player chooses a subset A of , A = modD such that: if = 0, A and if = + 1 then A B and if is a limit ordinal then A = {A : < } and then the second player chooses a subset B of A satisfying B = modD. A player wins the play if he has no legal move (can occur only to the rst player in a limit stage), if the play lasts moves then the second player wins.

8. Denition: Let be regular countable, S ; we say that there is a ( )-square for S if: there is a set S + , and sequence C : S + such that:
modified:1994-06-19

a. S S + b. for C (so S + ) we have: S + and C = C . c. otp(C ) for S + . d. if S is a limit ordinal then = sup(C ) e. C is a closed subset of .

revision:1994-06-19

9. Claim; 1) Assume is regular > , D is a normal lter on + to which { : cf() = } belongs. Then in the game GM (D) (see Denition 8 below) the second player does not +1 have a winning strategy. 2) Assume is regular larger than ||+ , an ordinal, D is a normal lter on to which a set S belongs, and for S there is a ( )-square (as dened in Denition 7 above) (or just 8

542

every S , S = modD has a subset S for which there is a ( )-square. S = modD). Then in the game GM (D) (see Denition 8 below), the second player does not +1 have a winning strategy.

Proof: Part (1) follows form part (2) as the assumption of part (2) follows by [Sh 365, 2.14] (or [Sh 351, Th. 4.1]). So we concentrate on proving part (2). So let C : S + be as in Denition 8. So without loss of generality S + D. We divide { : < , cf() = 0 } to ||+ stationary sets Ti : i < ||+ . As D is a normal ideal on , ||+ < , clearly for each stationary subset S of S which is D-positive there are S S which is D-positive and ordinal j < ||+ such that for every S we have: C {} is disjoint to Tj() . Now suppose the second player has a winning strategy in GM (D) which we call +1 Sty. We can choose by induction on n < a sequence A , B , : n such that 1. for every n the sequence A k , B
modified:1994-06-19

: k n is an initial segment of a play

of the game in which the second player uses his winning strategy Sty 2. for some j < ||+ , for every A we have C {} is disjoint to Tj .

3. S + and for every n and A we have C . 4. for n we have: is larger than sup range (). There is no problem to carry the denition (for clause (3) remember D is a normal

revision:1994-06-19

lter on ); now let E = { < : for every

def

>

we have < }; clearly E is a club of

hence there is an ordinal E Tj ; so choose an increasing -sequence of ordinals <

with limit ; look at A k , B

: k < which is an initial segment of a play of the game

k

in which the second player uses his winning strategy Sty. Let now B = {B

: k < };
n

if sup(B) > (which holds if B = modD), B \ ( + 1) then for every n, 9

C .

542

necessarily > =
n<

n<

subset of + 1, contradicting to Sty is a winning strategy.

9A Remark: This continues the argument that e.g. not for every stationary S { < 3 : cf() = 0 }, there is a club E of 3 such that E & if () = 2 S stationary in (nd pairwise disjoint Si { < 3 : cf() = 0 }, for i < 3 , if for Si we have Ei , choose
i<2

Ei of conality 2 . References

[Gi] M. Gitik. Changing conalities and the non stationary ideal, Israel Journal of Mathematics 56(1986) pp.280-314 . [GiSh310] M.Gitik and S.Shelah, Cardinal preserving ideals, Journal of Symbolic Logic, to apmodified:1994-06-19

pear. [LiSh 484] K.Liu and S.Shelah, Connalities of elementary substructures of structures on , Israel Journal of Mathematics, to appear. [Sh 351] S.Shelah, Reecting stationary sets and successors of singular cardinals, Archive fur Math Logic 31 (1991) pp. 2534. [Sh 365] S.Shelah, There are Jonsson algebras in many inaccessible cardinals, Ch III Cardinal

revision:1994-06-19

Arithmetic, OUP. accepted.

542

). So C but Tj whereas B , contradiction. So B is a

10

so equality holds. Hence also = (n<

) C (as

Note: as , and E clearly

< ; so

n<

; as

sup range( n)