THE CHARACTER SPECTRUM OF (N) SH915

SAHARON SHELAH Dedicated to Kenneth Kunen Abstract. We show the consistency of: the set of regular cardinals which are the character of some ultralter on N can be quite chaotic, in particular can have many gaps.

0. Introduction The set of characters of non-principal ultralters on N, that we call the character spectrum and denote by Sp , is naturlly of interest to topologists and set theorists alike, see Denition 0.1 below. A natural question is what can this set of cardinals be? The rst result on Sp is Posp proof that c Sp . sil It is consistent that Sp = {20 }, since Martins Axiom implies Sp = {20 }. Nevertheless, Sp = {20 } is not a theorem of ZFC. Juhas (see [Juh80]) proved the consistency of the existence of a non-principal ultralter D so that (D) < 20 . Kunen (in [Kun]) mentions that 1 Sp in the side-by-side Sacks model. Those initial resulsts show that (D) is not a trivial cardinal invariant. But we may wonder whether Sp is an interesting set. For instance, can Sp include more than two members? Does it have to be a convex set? It is proved in [BnSh:642, 6] that |Sp | large is consistent, e.g. 20 is large and all regular uncountable 20 (or just of uncountable conality) belong to it. It was asked there: among regular cardinals is it convex? Now (proved in [Sh:846]) Sp does not have to be convex. In the model of [Sh:846], there is a triple of cardinals (, , ) such that < < , , Sp but Sp . In the present paper we show that Sp may / exhibit much more chaotic behavior. To be specic, starting from two disjoint sets 1 and 2 of regular uncountable cardinals we produce a forcing notion P which forces the following properties: (a) no cardinal (of V) is collapsed in VP (b) 20 is an upper bound for the union of 1 and 2 (c) 1 Sp whereas 2 Sp = . The proof requires that each element of 2 be measurable and that 1 satisfy < = . This means that in the extension all members of 2 are weakly inaccessible
Date: July 15, 2011. 1991 Mathematics Subject Classication. MSC 2010: Primary 03E35; Secondary: 03E17, 54A25. Key words and phrases. set theory, ultralters, cardinal invariants, characer, forcing. The author thanks Alice Leonhardt for the beautiful typing. First Typed - 06/Sept/11; created from Sh846 (i.e. 1); 2 done 7/2007. Partially supported by the Binational Science Foundation.
1

(915)

revision:2011-07-15

modified:2011-07-17

SAHARON SHELAH

{z2}

and hence that we also do not know for certain that there are successor cardinals outside Sp . In the last section we show that we can, e.g. specify Sp , basically at will: if we have innitely many measurable cardinals then we can make the intersection be {n : n u} for any subset of [1, ) that has no large gaps, i.e. for every n at least one of n and n + 1 belongs to u. If we assume innitely many compact cardinals then we can realize any ground model subset of [1, ), e.g. Sp can be even {p : p prime}. Let us try to explain how do we do this. A purpose of [BnSh:642] is to create a large Sp . It provides a way to ensure many cardinals are in Sp . On the other hand, [Sh:846] provides a way for guaranteeing a cardinal is not in Sp . Here we try to combine the methods, hence creating a large set with many prescribed gaps which establishes Sp in VP . For adding cardinals we use systems of lters, so we deal with them and with the one step forcing in 1; we use such systems indexed, e.g. by -trees, and in the end force by a suitable product of those trees, not adding reals. In this direction we do not need large cardinal assumptions. For eliminating cardinals we need, essentially, measurables in the ground model. After the forcing with P, our measurable cardinals become weakly inaccessible, and we show that they do not belong to Sp . We emphasize that for adding a cardinal to 1 Sp , we have to assume = < . Moreover, 2 consists (in the ground model) of measurable cardinals which remain weakly inaccessible (= regular limit) cardinals in VP . Consequently, in 2 we do not know for certain that there are successor cardinals outside Sp . As in many other cases, to deal with small, e.g. successor cardinals we have also to collapse. The last section of the paper is devoted to the set Sp . Let u be any set (e.g., u = {p : p is a prime number}). If we assume that there are innitely many compact cardinals in the ground model, then we can force Sp = {n : n u}. Assuming just the existence of innitely many measurable cardinals, we can prove a similar result with some restrictions on u. We need that |u {n, n + 1}| 1 for every n . We thank the referee and Shimoni Garti for helpful comments. Recall Denition 0.1. 1) For an ultralter D on N let (D), the character of D be min{|A | : A D and every member of D include some member of A }. 2) The character spectrum of non-principal ultralters on N is Sp := {(D) : D a non-principal ultralter on N}.

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

1. Preliminaries This section is devoted to denitions and facts, needed for proving the main results of the paper. We present lter systems D = Dt : t I and we deal with the one step forcing QD where D = D : > , D a lter on N containing the co-nite subsets of N; when P1 QD1 P2 QD2 , and with frames d = (Dd , Fd ) for analyzing Qd -name of A of subsetsof N modulo the lter on N which Fd generated, in particular, a derived QD -name of an ideal idd . {z4} Denition 1.1. For forcing notion P1 , P2 (i.e. quasi orders). 1) P1 P2 i p P1 p P2 and for every p, q P1 we have P1 |= p q i P2 |= p q. 2) P1 ic P2 i P1 P2 and for every p, q P1 we have p, q are compatible in P1 i p, q are compatible in P2 . 3) P1 P2 i 1 P1 P2 and every maximal antichain of P1 is a maixmal antichain of P2 , equivalently 2 P1 ic P2 and for every p2 P2 for some p1 P1 we have p1 P1 p (p2 , p are compatible in P2 ). Denition/Observation 1.2. 1) For A P(N) let l(A ) = {B :
<n

{cn.1}

modified:2011-07-17

B for some n < and A0 , . . . , An1 A }; so if A is empty then l(A ) is the lter of co-nite sets. We may forget to distinguish between A and l(A ). 2) l(A ) is a lter on N extending the lter of co-bounded subsets of N but possibly l(A ) = P(N), equivalently l(A ). 3) For a lter D on X let D+ = {Y X : Y = mod D}. Denition 1.3. Let I be a partial order or just a quasi order. 1) We say D is an I-lter system when : (a) D = Dt : t I (b) Dt P(N) but l(Dt ) / (c) if s I t then l(Ds ) l(Dt ). 2) We say D is an ultra I-lter system when in addition:
{cn.7}

revision:2011-07-15

(d) if s I, A N and A = mod Ds then for some t we have s I t and A l(Dt ). 3) If D is an I -lter system for = 1, 2 then we let (D = D,t : t I and): (a) D1 D2 means I1 I2 (as quasi orders, so possibly I1 = I2 ) and s I1 D1,s D2,s (b) D1 D2 means I1 I2 and s I1 l(D1,s ) l(D2,s ) (c) D1 D2 means I1 I2 and s I1 l(D1,s ) = l(D2,s ) 1 2 (d) D = D means I1 = I2 and s I1 l(D1,s ) = l(D2,s ).

(915)

SAHARON SHELAH

Observation 1.4. Let I be a partial order. 0) , and quasi order the set of I-lter systems and l(Dt ) : t I is an I-lter system for any I-lter system D and D1 D2 D1 D2 and D1 = D2 D1 D2 D1 D2 and D1 D2 D1 D1 = D2 and D1 D2 D1 D1 = D2 . 1) If As [N]0 for each s I and At As for s I t then there is an I-lter system D such that s I Ds = {As }. 2) If D is an I-lter system then for some ultra I-lter system D we have D D . 3) If D is an I-lter system, s I and A and (t)[s I t A = mod l (Dt )], then for some I-lter system D we have D D and A Ds . 4) If D : < is an -increasing sequence of I-lter systems then some I-lter system D is an upper bound of the sequence; in fact, one can use the limit, i.e. D,s = {D,s : < }; similarly for -increasing. 5) If D is an I-lter system and D = l(Dt ) : t I then D D . 6) If D is an I-lter system and each Dt is an ultralter on then D is an ultra I-lter system and necessarily s I t Ds = Dt . 7) If D1 is an ultra I-lter system and D2 is an I-lter system such that D1 D2 then D1 D2 . 8) Assume P1 P2 and P1 D is an I-lter system for = 1, 2. If P1 D1 D2 1 D2 ; also if P1 D1 D2 then P2 D1 D2 . then P2 D 9) If P1 P2 and P1 D is an I -lter system for = 2 and P1 D1 is ultra 1, D D and (l(D )+ )V[P1 ] l(D )+ . and D1 D2 then P2 1,t 2,t 1,t 2,t Proof. 0) Easy. 1) Check. 2) Use parts (3),(4), easy, but we elaborate. We try to choose D by induction on + 0 is an I-lter system, < D D and for < (2 + |I|) such that D = D, for limit use part each = + 1 for some t, D,t = D,t . For = 0 let D (4) and for = + 1 if D is not ultra, use part (3). By cardinality consideration is dened but we cannot dene D+1 so necessarily D is ultra as for some , D required. 3)-9) Easy, too. 1.4
{cn.17}

{cn.14}

modified:2011-07-17

revision:2011-07-15

Claim 1.5. 1) Assume the quasi-order I as a forcing notion adds no new reals. An I-lter system D is ultra i I {l(Dt ) : t GI } is an ultralter on . 2) Assume the quasi-order I as a forcing notion adds no new 1 -sequences of ordinals and P is a c.c.c. forcing notion, (or just I is 1 -complete or just P forcing with I add no new real). If P Dt : t I is an I-lter system then P I {l(D t ) : t GI } is an ultralter on N i P D t : t I is an ultra I-lter system. Proof. Easy.
1.5

{cn.19}

Discussion 1.6. An I-lter system D may be degenerated, i.e. Dt = D is an ultralter, the same for every t I. But in this case adding a generic set to I will not add naturally a new ultralter, which is our aim here. Denition 1.7. 1) For D = D : > , each D a lter on N let QD be

{c20}

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

T : T > is closed under initial segments, and for some tr(T ) > , the trunk of T, we have : (i) g(tr(T )) T = {tr(T ) } (ii) tr(T ) > {n : n T } D ordered by inverse inclusion. 2) For p QD let wfst(p, D) be the set of pairs (S, ) such that: (a) () S { p : tr(p) p} () tr(p) S () tr(p) S S (b) () is a function from S into 1 () if are from S then () > () () if S and () > 0 then {k : k S} = mod D . 3) If p QD and p then we let p[] = { p : or }. 4) If D = D : > , D = D for > then let QD = QD and wfst(p, D) = wfst(p, D); we may write instead of p when this holds for some p QD with tr(p) = ; wfst stands for well founded sub-tree. Claim 1.8. Assume > , D is a lter on N for > and Y is a subset of = = { : > }. Then exactly one of the following clauses holds: (a) there is q QD such that () = tr(q) () Y q = , equivalently q + = q\{tr(q(t)) : < g(tr(q))}) is disjoint to Y (b) there is a function such that (Dom(), ) wfst( , D) and max(Dom()) Y ; that is: () Dom() is a set satisfying (i) { : > } (ii) (iii) if and then () (i) Rang() 1 (ii) () < () () for every at least one of the following holds: (i) Y + (ii) the set {n : n } belongs to D . Proof. Similar to [Sh:700, 4.7] or better [Sh:707, 5.4]. In full, recall = { : > }. We dene when dp() for by induction on the ordinal : = 0: always a limit ordinal: dp() i rk() for every < = + 1: dp() i both of the following occurs:

{c30}

(915)

revision:2011-07-15

modified:2011-07-17

SAHARON SHELAH

(i) Y / + (ii) the following set belongs to D : {n : dp( n ) }. We dene dp() Ord {} such that = dp() i ( Ord)[dp() i ]. Easily for every , dp() 1 {}. Case 1: dp( ) = . For each such that dp() = clearly there is A D such that n A dp( n ) = . Let q be { p : or and if then (g()) A }.

Clearly q is as required in clause (a) of 1.8. Case 2: dp( ) < . We dene = { : and if k [g( ), g()) then k Y and dp(k) > dp((k + 1))}. / We dene : 1 by () = dp(). Now check. Claim 1.9. P1 QD1 P2 QD2 when :
modified:2011-07-17

{cn.28}

1.8

revision:2011-07-15

P1 P2 and D = D , : > for = 1, 2 D1, is a P1 -name of a lter on N D2, is a P2 -name of a lter on N (d) P2 D1, D2, and moreover (l(D1, )+ )V[P1 ] l(D2, )+ , i.e. for every A P(N)V[P1 ] we have A l(D1, ) A l(D 2, ). Proof. Like [Sh:700, 4] more [Sh:707, 5] but we elaborate. Without loss of generality P1 and P2 p for every p P2 . Clearly P1 QD1 P2 QD2 by clause (d) of the assumption and moreover P1 P2 P2 QD2 recalling Denition 1.1(1),(2). Now we can force by P1 so without loss of generality it is trivial, hence we have to prove that QD1 P2 QD2 identifying q QD1 with (, q) P2 QD2 . By clause (d) of the assumption, this identication is well dened and Q 1 ic P2 QD2 because for p1 , p2 QD1 , p1 , p2 are compatible i D (tr(p1 ) p2 ) (tr(p2 ) p1 ). It suces to verify 1.1(3), requirement 2 . So let (p2 , q 2 ) P2 QD2 ; without loss of generality for some from V we have p2 = tr(q 2 ), so > and of course: ()1 P2 q 2 QD2 . By 1.1(3), it suces to nd q QD1 such that (a) (b) (c) ()2 q q QD1 (p2 , q 2 ), q are compatible; that is, (p2 , q 2 ), (, q ) are compatible in P2 QD2 .

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

Now we shall apply Claim 1.8 in V with , D1 here standing for , D there. Still Y is missing, so let Y = { : > and there is r QD1 such that = tr(r) and (, r), (p2 , q 2 ) are incompatible in P2 QD2 equivalently p2 P2 q2 , r are incompatible in QD2 }. By Claim 1.8 below we get clause (a) or clause (b) there. Case 1: Clause (a) holds, say as witnessed by q QD1 . We shall prove that in this case q is as required, i.e. q QD1 and [q QD1 r QD1 (p2 , q 2 ) P2 QD2 and r are compatible (in P2 QD2 )]. Why? Let = tr(r). Clearly ( q) hence by the choice of q, i.e. 1.8(a)() we have Y so r cannot witness Y hence r, (p2 , q 2 ) are compatible in / P2 QD as required.

Case 2: Clause (b) holds as witnessed by the function . By the denition of Y , in V, we can choose q such that: (a) q = q : Y (b) q QD1 and tr(q ) = (c) q witness Y , i.e. p2

q , q 2 are incompatible in QD2 .

modified:2011-07-17

We dene a P2 -name q as follows: q = { : or q 2 and if g( ) k < g() and k Y then qk , hence k g() qk }. Clearly q QD2 and tr(q ) = and QD2 |= q2 q . ()3 if Y then and p2 P2 ( q ). [Why? Otherwise there is p3 P2 such that p2 p3 and p3 P2 q , as tr(q ) is forced to be and tr(q ) = , necessarily p3 P2 q , q are compatible. p2 P q2 q , we get a contradiction to the choice of q .] But 2 Now we know that Dom() and q hence S := { : Dom() hence and p2 q } is not empty. So as S Dom() the set / and by the choice of the function we have U = {() : S} is not empty, U 1 , hence there is a minimal U and let Dom() be such that () = . By the denition, if = 0 then by clauses () and () of 1.8(b), i.e. the choice of () we have Y and, of course, S. By ()3 , p2 P2 ( q ) we get easy contradiction to S, hence we can assume > 0. By the denition of S there is p P2 such that P2 |= p2 p and p P2 q hence q 2 and, of course, S. By the choice of the function , in V we have A := {n : n Dom()} = mod D1, , hence by clause (d) of the assumption of the claim P2 A = mod D2, and, of course, p P2 {n : n q } D2, . Together p P2 there is nsuch that n q Dom(), so let n and P2 be such that P2 |= p p p and p P2 n q Dom().
P2

(915)

revision:2011-07-15

SAHARON SHELAH

{c31}

So ( n ) is well dened, i.e. n belongs to Dom() hence ( n ) < () = and easily n S and ( n ) U , so we get a contradiction to the choice of . 1.9 Denition 1.10. 1) We say d = (D, F ) is a frame when : / (a) D = D : > and D [N]0 , l(D ) for > 0 / (b) F [N] and l(F ). 1A) Above let Dd = Dd, : > , Dd, = l(D ), Fd = l(F ), Qd = QDd and if D = D for > we may write D, Dd instead of D, Dd , respectively. 2) We say A is a d-candidate when (d is a frame and): (c) A is a Qd -name of a subset of N. 3) We say A is d-null when it is a d-candidate and is not d-positive, see below. 4) We say is d-positive when for some p Qd , for a dense set of p p some A quadruple (p, A, S, ) is a local witness1 for (A, d) or for (, A, d) when = tr(p) or for (p, A, d) or for A being d-positive, which means: (a) p Qd + (b) A Fd (c) S = Sn : n A and = n : n A (d) (Sn , n ) wfst(p, D) for n A recalling Denition 1.7(2) (e) if Sn and n () = 0 then p[] n A. Denition 1.11. 1) For a frame d = (D, F ) let idd = id(d) = {A N : A is a Qd -name which is d-null}. 2) If P d is a frame then idd [P] is the P Qd -name of idd . Claim 1.12. For a frame d, QD idd is an ideal on N containing the nite sets and N idd ; moreover, for every A P(N) from V, we have A = mod Fd i / Qd A idd . Proof. It suces to prove the following 1 4 . if A1 A2 and A2 idd then A1 idd . [Why? If (p, A, S, ) is a local witness for (A1 , d) then obviously it is a lcoal witness for (A2 , d).] 2 if Qd if A1 , A2 idd then A1 A2 idd . Why? It suces to prove: if Qd A1 A2 = A N and A is d-positive then A ) is d-positive for some {1, 2}. Let (p, A, S, be a local witness for (A, d) and by the we shall prove that there are {1, 2} and a local witness for (tr(p), A , d); dense in Denition 1.10(4) this suces. For any n A and Sn such that n () = 0 we choose (n, , n, , Sn, ) such that: 1 If
QD

{c33}

modified:2011-07-17

{c35}

revision:2011-07-15

()2.1 (a)

,n {1, 2}

1An equivalent version is when we weaken clause (e) to: if S and () = 0 then there n n is q Qd such that tr(q) = , p q and q n A, see ()2.2 in the proof. Moreover, we can omit p q; hence actually only tr(p) is important so we may write tr(p) instead of p.

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

(Sn, , n, ) wfst(p[] , Dd ) if n, () = 0 so Sn, then there is q Qd such that p q, tr(q) = and q n An, ; let qn, be such q. [Why (n, , n, , Sn, ) exists? We shall use 1.8; that is for {1, 2} let Yn,, = { : p and there is r QD such that tr(r) = and p r and r n A }. We apply for = 1, 2 Claim 1.8 with Dd , , Yn,, here standing for D, , Y there. If for some {1, 2} clause (b) there holds as witness by the function , easily the desired ()2.1 holds. If for both = 1, 2 clause (a) there holds then for = 1, 2 there is q Qd such that tr(q ) = and q Yn,, = . Necessarily q := q1 q2 p belongs to Qd and has trunk and is disjoint to Yn,,1 Yn,,2 . But Qd |= p[] q and q [] Qd n A = A1 A2 , hence there are {1, 2} and r Qd such that q r and r Q n A , but then d () holds tr(r) Yn,, and tr(r) q q , contradicting the choice of q . So 2.1 indeed.] (b) (c) ()2.2 without loss of generality n, () = 0 g() > n. [Why? Obvious.]
()2.3 for n A there are n , Sn , n such that (a) (Sn , n ) wfst(p, Dd ) (b) Sn Sn and max(Sn ) = Sn max(Sn ) (c) n {1, 2} and max(Sn ) n, = n .

[Why? Easy.]
()2.4 for n A letting Sn = {Sn, : max(Sn )} Sn , for some n and qn we have: (Sn , n ) wfst(p, D) { : n () = 0} = {: for some we have Sn , n () = 0, Sn, and n, () = 0} q = qn, : n () = 0 n () = 0 p qn, tr(qn, ) = qn, n An [Why? Think.]

modified:2011-07-17

revision:2011-07-15

()2.5 there is {1, 2} such that A := {n A : n = } = mod Fd .

+ [Why? Obvious as A Fd .] We now consider the quadruple (p , A , S , ) dened by: p = { p: if tr(p) , n g() and max(Sn ) then qn, } where Sn , qn, are from ()2.4 .

[Why p Qd with tr(p ) = tr(p)? Recall()2.2 .] So together we have:

+ A is from ()2.5 , so A Fd S = Sn : n A where Sn is from ()2.4

(915)

10

SAHARON SHELAH = n : n A where n is from ()2.4 .

Now check that (p , A , S , ) is a witness for (tr(p), A , D) hence 2 holds as said in the beginning of its proof. idd ; moreover if A = mod Fd is from V then A idd . Why? Because of clause (b) in Denition 1.10(4). 3
Qd + N idd , moreover if B Fd and B V then B idd . / / Why? This means that B is d-positive which is obvious: use the witness (p, A, S, ) where p is any member of Qd , A = B, Sn = {tr(p)}, n (tr(p)) = 0. 1.12

QD [d]

{c36}

{c37}

Observation 1.13. Assume d1 , d2 are frames and Dd1 = D = Dd2 and Fd1 Fd2 then QD idd1 idd2 . Proof. Should be clear. 1.13 Claim 1.14. We have
P2

idd1 idd2 and (idd1 )+ [P1 ] (idd2 )+ [P2 ] when :

(a) P1 P2 (b) P d is a frame for = 1, 2 (c) P2 Dd1 , Dd2 , for > (d) if A (D+1 , )V[P1 ] then A (D+2 )V[P2 ] d d (e) P2 Fd1 Fd2
+ + (f ) if A (Fd1 )V[P1 ] then A (Fd2 )V[P2 ] .

modified:2011-07-17

{c39}

Proof. Should be clear by 1.15 below recalling 1.9.

1.14

{c40} {c41}

Claim 1.15. Let d be a frame and A a QDd -name of a subset of N. We have A is d-null i for a pre-dense set of p Qd we have tr(p) p there is no local witness for (p[] , A, d) equivalently, for (, A, d). Proof. Straight. 1.15 Remark 1.16. The point of 1.15 is that the second condition is clearly absolute in the relevant cases by 1.9, i.e. in 1.14. Denition 1.17. 1) n(I) is the set of nite functions from I to H (0 ). 2) Let K be the set of forcing notions Q such that some pair (I, f ) witness it, i.e. (I, f, Q) K+ which means:

revision:2011-07-15

(a) f is a function from Q to n(I) (b) if p1 , p2 Q and the functions g(p1 ), g(p2 ) are compatible then p1 , p2 have a common upper bound p with g(p) = g(p1 ) g(p2 ). 2) We dene K =wk by: (I1 , f1 , Q1 ) wk (I2 , f2 , Q2 ) means that: K K (a) (b) (c) (d) (I , f ) witness Q K for = 1, 2 I1 I2 f1 f2 Q1 ic Q2 .

3) We dene st similarly adding: K

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

11

(d)+ Q1 Q2 . 4) If q K+ let q = (Iq , fq , Qq ). Remark 1.18. We can use much less in Denition 1.17.

(915)

revision:2011-07-15

modified:2011-07-17

12

SAHARON SHELAH

2. Consistency of many gaps We prove the rst result promised in the introduction. Assume = < > 1 and we like to build a c.c.c. forcing notion P of cardinality , such that VP is as required: Sp includes 1 and is disjoint to 2 ; really we force by P T , the T quite complete and translate P-names of ultra systems of lters to ultra-lters. In order to have 1 Sp , we shall represent P as an FS iteration P , Q : , < , |P | and T is, e.g. > 2 and for each 2 we have a D = D ,s : s T a P -name of a ultra system of lters for unboundedly many < , increasing with ; in the end we force by P {T : 1 }. Toward this for each s T , 1 we many times force by Q+ ,s from 1. D But in order to have 2 Sp = , we intend to represent P as the union of a -increasing sequence P : < and for each 2 for stationarily many < , cf() = and P +1 is essentially the ultrapower (P ) /E , E a -complete ultra-lter on , so is a measurable cardinal. To accomplish both we dene a set Q, each x Q consist of a FS-iteration of P , Q : + , < + with Ds, : s {T : 1 } for many < + , . increasing with and Q = QDt(), In the end for suitable x, we shall use P for some < + of conality << , (e.g. = 1 ). So why go so high as + ? It helps in the construction toward the other aim; we shall construct x : increasing in Q such that for each 2 for < of conality , x+1 is essentially (x ) /E . In particular, we have to prove Q = , the existence of the ultrapower and the existence of limit which happens to be a major proof here. For this we have to choose the right denition, in particular using id(D,s ,D,t ) from Denition 1.11. For this section we assume Hypothesis 2.1. 1) We now x two cardinals and as well as two sets, 1 and 2 , of regular cardinals in the interval [, ] and let = 1 2 . Our assumptions are (a) is regular and uncountable, = 0 and < (b) 1 and 2 are disjoint sets of regular cardinals < from the interval [, ) but 2 / (c) Each 1 we have < = (d) each 2 carries a normal ultralter E , hence 2 consists of measurable cardinals (e) for all 2 the cardinal satises cf() > and = /E . 2) Furthermore (and see 2.4) below so it is not a burden (f ) T = T : 1 , T is a tree of cardinality with levels, such that above any element there are elements of any higher level (may add T is 2 -complete and even T is -complete, then clause (g) follows) (g) for every 1 , forcing by T := {T : 1 \}, the product with Easton support, adds no sequence of ordinals of length < and, for simplicity, collapses no cardinal and changes no conality; if = 2 add T is 1 -complete; let T = Tmin(1 )

modified:2011-07-17

{cn.42}

(915)

revision:2011-07-15

THE CHARACTER SPECTRUM OF (N)

SH915

13

(h) if 1 then |T | is (1 \) except when sup(1 ) is strongly inaccessible and then the value is sup(1 ) Choice 2.2. 1) Without loss of generality T : 1 is a sequence of pairwise disjoint trees. 2) Let T be the disjoint sum of {T : }, so it is a forest. 3) Let t = ti : i S be a sequence of members of T where S = { < + : cf() = cf()} such that if t T then { S : t = t} is a stationary subset of + ; let t(i) = ti . 4) Furthermore choose () () () () () S0 { < + : cf() = 0 } is stationary = ,t,n : S0 , t T , n N ; let (, t, n) = ,t,n ,t,n : n < is an increasing -sequence of ordinals with limit ,t,n { S : t = t} guess clubs, i.e. if E is a club of + then the set { S0 : C := {,t,n : t, n} E} is stationary.

{c45}

Remark 2.3. If |T | < we can nd such , but in general it is easy to force such . Claim 2.4. Assuming 2.1(1) only, a sequence T as in 2.1, clauses (f ),(g),(h) (and , s, S , as in 2.2) exists, provided that 1 { : = < } and G.C.H. also t holds (or just = sup(1 ) 2 = + ). Proof. Straight, e.g. T = (> 2, ).
modified:2011-07-17
2.4 +

{cn.47}

Denition 2.5. Let Q be the set of objects x consisting of (below , ): (a) P H (++ ) and I< , f H (++ ) witnessed P K for + , all in H (+ ) if < + (b) I H (+ ) and Q , g H (+ ) are P -names such that P Q K as witnessed by I g for < + , (c) Q = P , Q : < + H (++ ) is an FS iteration except that () P ={p : p a nite function with domain such that if dom(p) then g (p()) n(I ) is an object (not just a P -name)} (d) I< = {I : < } is disjoint to I and P = P+ = {P : < + } and f () = {g (p()) : dom(p)} (e) E is a club of + and for S E: () D = D,s : s T is a P -name of an ultra T -lter system (equiv alently each D, = D T is a P -name of an ultra T -lter system), and for simplicity l(D,s ) = D,s () D,s : S E, is -increasing continuous for each s T (f ) if E then Q is QD,t() see Denition 1.7 and calling the generic S , we have I = {0}, g (p) = tr(p) (g) () if S E and s, t T then P l(D ,s ) l(D ,t ) i s T t actually follows from (e)() )[P ]; () if < are from S E and s T then P if A id(d t(),s then A = mod D,s where d = (D ,t , D,s ) t,s

{c49}

(915)

revision:2011-07-15

14

SAHARON SHELAH

{c50}

if S0 E then Q is Ql() with the generic if S0 E and C Ex , see 2.2(4)() then u,t,n D,t , see below, whenever t T , n N and S \( + 1) Ex () in clause () we let u,t,m = { (,t,n) (k) : n N, n m and k (n)}. Discussion 2.6. 1) Later we shall use an increasing continuous sequence x : . Where and how will conality reappear? Well, we shall use P() [x ] for some () Ex of conality . So why not replace + by above? We have a problem in proving the existence of a (canonical) upper bound to x : < , specically in nding the Di in the proof of Claim 2.11, i.e. completing an appropriate T -lter system to an ultra one, e.g. in Case 3 in the proof of 2.11. To help we carry a strong induction hypothesis, see clause (i)()2 in there and then rst nd an Rj ,+ [Pj x]-name, then reect it to a i . 2) Note that it helps to have not only Q = QD , but possibly some related forcing notions. First in proving there is a limit, see 2.11, in proving the reection discussed above lead us to use some unions. Second, using ultrapower by E , see 2.13, for limit of conality , the ultrapower naturally leads us to use some iterations. 3) We may in 2.1 demand 1 , equivalently < min(), but let T be a / singleton {t } and T is Tmin(1 ) T . In this case in 2.17 we get PT {} 1 Sp . (h) () () Denition 2.7. 1) For x Q, of course we let Qx = Qx = Q[x] = Q, Px = x x P [x] = P , Px = P = P = P+ , etc. 2) We dene a two-place relation Q on Q : x Q y i:
y y x x (a) (I< , f , Px ) st (I< , f , Py ) for + , see Denition 1.17(3) K x x x wk y (b) Py (I , g , Q ) K (I , gy , Qy ) for < + , see Denition 1.17(2) (c) Ey Ex for S Ey and t T (d) P [y] Dx Dy ,t(i) ,t(i) )+ , really follows by )+ )V[P[x]] then A (Dy (e) P [y] if A ((D x ,t(i) ,t(i) clause (d) and 2.5(e)(), the ultra.

modified:2011-07-17

{cn.51}

Claim 2.8. Q is non-empty, in fact there is x Q such that Px has cardinality x for [1, + ) and in VP1 we have 20 = .
Proof. For i = 0, rst letting D0,s = for s T , clearly D = D0,s : s T 0 is a T -lter system hence by 1.4(2) we can choose D0 = D0,s : s T , an ultra P0 T -lter system (in V = V ). Second, we choose Qi as adding Cohen reals, say : < so Ii = , fi is the identity, so fi (p)() = p() > 2. Third, let , t ) : < be such that s , t T are T -incomparable and any such pair (s appears. 1 We dene a P1 -name D = D : t T by D = {1, {} : s I t = 0 or t t is an T -lter system, so by 1.4(2) there t I t = 1} D0,t . Clearly P1 D is D1 such that P1 D1 is an ultra T-lter satisfying D D1 hence D0 D1 . P , D by induction on + \S suchthat Now we shall choose also for the relevant demands from Denition 2.5 hold, in particular, P , Q : , <

(915)

revision:2011-07-15

THE CHARACTER SPECTRUM OF (N)

SH915

15

is a FS-iteration but dom(p), p P implies that P forces a value to tr(p()) and also P D is a T -lter system such that D D for < is ultra when S0 ; recall that in Denition 2.5 D is dened only for and D / E S, but no harm in dening D in more cases. For = 0, 1 this was done above. = For limit let P = {P : < } and D = D : t T where D ,t ,t(i) {D,t(i) : < }. It is easy to see that P : is a -increasing continuous sequence of c.c.c. forcing notions and P D is an T -lter system. If S0 = D , otherwise by 1.4(2) we can nd D such that P D is an ultra let D T -lter system and D D . For = + 1 such that S S0 let Q be trivial. Now let P = P Q and / let D,t be D,t . Easily P D : t T is an T -lter system and choose D ,t . as above, i.e. (a P -name of an) ultra T -lter system above D P = P Q . Now Next, assume = + 1, S; we let Q = QD,t() and for s T , let D = D,s {N\A : A iddt(),s } where t() = t is from 2.2(3). ,s Note that P l(D ,s ) l(D,t ) i s t by the choice of the D1,s s and the T D,s s, so the denition of iddt(),s depend on the truth value of t() I s. Now (pedantically working in VP ): D [N]0 by its denition ,s D,s D,s , by 1.12 l(D,s ) by 1.12 /

revision:2011-07-15

+ if A (D,s )V[P ] then A ((D,s )+ )V[P+1 ] by 1.12 s I t D,s D,t by 1.13 and the choice of the D,t s. We continue as in the previous case. Lastly, assume = + 1, S0 and we shall dene for . We let Q = Ql() in VP and so is dened as the generic and P+1 = P Q . Note that u,t,n is well dened, (see clause (h) of Denition 2.5). By Claim 2.9 below letting D = ,t D,t {u,s,n : n N and s T satises s T t} we have D = D : t T is ,t a P -name of a T -lter system above D and let D be (a P -name of) an ultra . T -lter system above D Let I = {} for < + , I< = for + and if S S0 then we let P if p Q then g (p) is tr(p), the trunk and if + \(S S0 ) then g (p) = 0. x x Naturally, wedene x by: Px = P , Qx = Q , Ex = , I = I , I< = I< , gx = Dx = D for S, + , . + + x g for < , (and so f is dened), < is easily to check that x Q is as required. It 2.8

modified:2011-07-17

{c52}

Claim 2.9. If (A) then (B) where (A) (a) S0 (b) P ( ), Q ( ), E , etc., are as in Denition 2.5 except that all up to is (c) Q , , u,t are as in clause (h) of Denition 2.5 (d) D := {D,t : S E} {u,t,n : n N and s T satises ,t t} so a P Q s T l() -name (B) (a) P Qfil() D,t : t T is a T -lter system

(915)

16

SAHARON SHELAH

(b) (c)

P Qfil() P Qfil()

l(D ) = l({u,s,n : s T t and n N}) ,t if t T and A {D,t : S} then u,t,n A for every large enough n.
P Qfil()

Proof. Straight; the point is

l(D ) for t T , which holds as ,t ()1 if A D(,t,n) then for every large enough k, (,t,n) (k) A P ()2 if A D+ then for innitely many k, (,t,n) (k) A (,t,n) in V ()3 is a dominating real.
2.9

{cn.53}

Observation 2.10. 1) Q partially orders Q. 2) Px satises the c.c.c. and even is locally 1 -centered2 when x Q and + . Proof. Easy.
{cn.56}
2.10

{cn.57}

Claim 2.11. The upper bound existence claim If x : < is Q -increasing and is a limit ordinal < + then there is x which is a canonical limit of x : < , see below. Denition 2.12. We say x = x is a canonical limit of x = x : < when x is Q -increasing, is a limit ordinal < + and (for every < + ): x Q x Q x for < and Ex {Ex : < } I [x ] = {I [x ] : < } if has uncountable conality then () Px = {Px : < } () Px Dx = {Dx : < } for t T if Ex S ,t ,t () Qx = {Qx : < } () gx = {gx : < }. (e) if has conality 0 , then () if + \(S Ex ) or S0 C Ex then P [x ] Q [x ] = {Q [x ] : < } and similarly g [x ] = {g [x ] : < } () if S Ex then P [x ] D,t [x ] {D,t [x ] : < } (f ) in fact |Px | ({|Px | : < })0 . (a) (b) (c) (d) Proof. Let 0 (a) I = {I [x ] : < } for < + (b) I< = {I : < } for + (c) E := {E[x ] : < }. So E {E[x ] : < } and clearly E is a club of + (but in general this will not be E[x ]). If + and Q satises < P [x ] Q and for transparency q Q Q q then R = R, [Q, x] is dened as follows:
2meaning that any elements can be divided to sets such that any nitely many members 1 0 of one sets has a common upper bound

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

17

1 (a)

p R i p = (p1 , p2 ) and some pair (, p0 ) witenss it which means < and p0 P [x ], p1 P [x ], p2 Q and one of the following occurs () p1 = or p2 = recalling clause (c) of 2.5 () p0 P [x ] p1 P [x ]/P [x ] and p2 Q/P [x ] (b) for p R let (p) be the minimal < such that (, p0 ) witness p R for some p0 (c) R |= p q i letting = max{(p), (q)} we have P [x ] |= p1 q1 and Q |= p2 q2 .

We note that: 2 (a) R, [Q, x] is a partial order (b) above R [Q, x] is a dense subset of R, [Q, x] where R [Q, x] is , , dened like R, [Q, x] when in 1 (a) we omit subclause (). [Why? Clause (a) by 3 below and clause (b) is easy.] So below we may ignore the dierence between R, [Q, x] and R [Q, x] , 3 for (, , Q) as above; if (, p0 ) is a witness for p = (p1 , p2 ) R, [Q, x] and (, ) then for some q0 P [x ] the pair (, q0 ) is a witness for (p1 , p2 ) R, [Q, x]. [Why? As we can increase p0 in P [x ], without loss of generality (p1 ) p0 , where on recall Denition 2.5, clause (c). As (, p0 ) is a witness for (p1 , p2 ) R, [Q, x] necessarily p0 , p2 are compatible in Q hence they have a common upper bound q2 Q. As P [x ]Q, there is q0 P [x ] such that q0 q P [x ] q, q2 are compatible in Q. As we can increase q0 in P [x ] and p0 q2 without loss of generality p0 q0 but (p1 ) p0 hence (p1 ) q0 . As x x and P [x ], Q [x ] : < + is FS iteration and p1 P [x ] P [x ], clearly q0 q P [x ] q, p1 are compatible. So clearly (, q0 ) is a witness for p R, [Q, x] as required in 3 .] 4 if , , Q are as above and (1) + then R, [Q, x] R,(1) [Q, x]. [Why? We check the conditions from Denition 1.1(3), the second alternative. First, if p = (p1 , p2 ) R, [Q, x] we shall prove p R,(1) [Q, x]; as p R, [Q, x], some (, p0 ) witness it, easily it witnesses p R,(1) [Q, x] as P [x ] P(1) [x ]. Second, assume R, [Q, x] |= p q and we should prove R,(1) [Q, x] |= p q, this is obvious by the denition of the orders for those forcing notions. Together R, [Q, x] R,(1) [Q, x]. Third, we should prove R, [Q, x] ic R,(1)[Q, x] so assume p, q R, [Q, x] has a common upper bound r = (r1 , r2 ) in R,(1) [Q, x]. Now easily (r1 , r2 ) is a common upper bound of p, q in R, [Q, x] as required. Fourth, for p R,(1)[Q, x] we should nd q R, [Q, x] such that if R, [Q, x] |= q q then q , p are compatible in R,(1) [Q, x]. Now let p = (p1 , p2 ) R,(1) [Q, x] and let (, p0 ) witness it; without loss of generality P [x ] |= (p1 ) p0 . Let q1 = p1 P [x ], now q := (q1 , p2 ) satises q R, [Q, x].

(915)

revision:2011-07-15

modified:2011-07-17

18

SAHARON SHELAH

Why? The pair (, p0 ) witness it because if p0 q P [x ] then rst p1 , q has a common upper bound r P(1)[x ] hence r P [x ] is a common upper bound of q , q1 ; second q , p2 has a common upper bound in Q as (, p0 ) witness (p1 , p2 ). So indeed (, p0 ) witness q = (q1 , p2 ) R, [Q, x]. If q q R, [Q, x] then q , p are compatible in P(1) [x ].
Why? Let q = (q1 , q2 ) and let r1 = (p1 [, (1)))q1 , easily (r1 , q2 ) R,(1)[Q, x] is a common upper bound of q , p. This nishes checking the last demand for R, [Q, x]R,(1) [Q, x] so 4 holds.]

5 if Q satises the c.c.c. then R, [Q, x] satises the c.c.c.. [Why? Let pi = (p1,i , p2,i ) R, [Q, x] for i < 1 . Let (i , p0,i ) be a witness for (p1,i , p2,i ). As before let qi Q be such that p0 , p1,i , p2,i are below it. We can nd an uncountable S such that f [xi ](p1,i ) : i S are pairwise compatible functions and i : i S is non-decreasing. As Q satises the c.c.c., for some i < j from S there is a common upper bound q Q of qi , qj ; let { : < n} list in increasing order {} dom(p1,i ) dom(p1,j )\ and let n = . By induction on n we choose r P [xj ] such that: if = 0 so = then r0 r P [xj ] r, q are compatible in Q if = m + 1 then rm r P [xj ] |= (p1,i ) r and (p1,j ) r .
modified:2011-07-17

For = 0 use q Q and P [xj ]Q. For = m+1, we shall choose r Pm +1 [xi ] as follows: if dom(p1,i ) then r = rm {( , p1,j ( ))}; if dom(p1,j ) / / similarly; otherwise, i.e. if dom(p1,i ) dom(p1,j ) use the demands on g recalling () of clause (c) and end of clause (d) of Denition 2.5. Having carried the induction, (rm , q) is well dened. Now let r P [xj ] be above r0 such that r r P [xj ] rm , r are compatible. Also r r P [xj ] r0 r P [xj ] r, q are compatible in Q. So (j , r ) witness (rm , q) R, [Q, x] and easily (rm , q) is above pi = (p1,i , p2,i ) and above pj = (p1,j , p2,j ), so 5 holds indeed.] 6 for , , Q as above, Q R, [Q, x] when we identify p2 Q with (, p2 ).

revision:2011-07-15

[Why? Again, rst p Q p R, [Q, x] by the identication, and for p, q Q we have Q |= p q R, [Q, x] |= p q by the denition of the order of R, [Q, x]. So Q R, [Q, x] holds, moreover Q ic R, [Q, x] by the denition of the order. Lastly, let q R, [Q, x], so by 2 without loss of generality q = (q1 , q2 ) R [Q, x] and we shall nd p Q such that p p Q p , (q1 , q2 ) are compati, ble. Let p = q2 , i.e. (, q2 ), and the rest should be clear.] 7 for , , Q as above we have P [x ]R, [Q, x] when we identify p1 P [x ] with (p1 , ).

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

19

[Why? Similarly.]

Now by induction on i + we choose i and P , f (when i and j < i j < ), Q , g (when < i and j < i j ) and3 also Di (when i S) such that the relevant parts of clauses (a)-(e) of Denition 2.12 and of the denition of x Q holds, in particular (all when dened): (a) P H (+ ) is a c.c.c. forcing notion (b) () Px P and Px P = Px for < () (I< [x ], f [x ], P [x ]) st (I< , f , P ) K (c) Di is a Pi -name of an I-lter system; ultra when i S; see (i)()1
Dxi ,t Di ,t (e) P : i is -increasing continuous (f ) if = + 1 then P = P Q , in fact, P , Q : i , < i is as in clause (c) of Denition 2.5

(d) if i S, < and t T then

P i

revision:2011-07-15

(g) if (j)( = j S) then P Q = {Q [x ] : < }, g = {g [x ] : < }; note that Q [x ], g [x ] are Px -names hence P [x ], [x ], Q [x ]) wk (I , f , Q ) by clause (b) and name P (I f K (h) () if j < i then Pi Dj Di () if i is a limit ordinal and t T then Pi Di ,t = {Dj ,t : j < i} (i) () j : j i is increasing continuous () if i = 0 then i = 0 () if i = j + 1 then 1 i S E 2 if [i , + ] (S E) {+ } and t T , then / Ri , [Pi , ] l({D ,t [x ] : < } D i ,t ) x 3 if j S E then clause (g) of Denition 2.5 holds 4 if j S0 and Cj { : < j} then Qj = Ql() , and so the relevant case of clause (h)() of Denition holds 2.5 + () if i is a limit ordinal, (i , ] (S E) {+ } and t T then Ri , [Pi , ] l({D,t [x ] : < } / x {D,t : < i }). Note that as D (when (j i)( = j S j = 0)) is an ultra T -lter system, + we do not have to bother proving A (D+ [GP ]) A (D,s [GP ]) (when < ,s are from {j : j i, j S}). Also ()1 if t T , < , i and S Ex then
3so we dene some D not used in x . i P

modified:2011-07-17

Dx Di ,t . ,t

(915)

20

SAHARON SHELAH

[Why? This follows from clause (i) of .] Let us carry the induction, this clearly suces. Case 1: i = 0. Trivial. Case 2: i is a limit ordinal. Let = i be {j : j < i}, clearly j : j i is increasing continuous and i E. Below vary on . Let P = {P : < } and f = {f : < } and from 0 recall I< = {I : < }. Clearly P K as witnessed by (I< , f ) and < P P . Note that P satises the c.c.c. as P : < is -increasing continuous and the induction hypothesis; alternatively using f . Now ()2 P [x ] P for < ; hence R, [P , x] is well dened for [, + ]. [Why? Again we shall use 1.1(3). First, P [x ] = {Pj [x ] : j < i} but j < i Pj [x ] Pj P so clearly P [x ] P . Second, P [x ] ic P , because if p, q P [x ] are incompatible in P [x ] then for some j < i we have p, q Pj [x ] hence p, q are incompatible in Pj [x ], so as Pj [x ] ic Pj they are incompatible in Pj , but Pj P so they are incompatible in P as required. Third, if q P then for some (0) < we have q P(0) and so there is p P(0) [x ] such that p p P(0) [x ] p , q are compatible in P(0) . So it suces to prove p p P [x ] p , q are compatible in P , so x such p . As is a limit ordinal, P = {P : < } hence there is (1) such that (0) (1) < and p P(1) [x ]. Now p := p (0) is well dened and belong to P(0) [x ] and is above p, so by the choice of p there is a common upper bound q + P(0) of q and p . As P , Q : < is FS iteration, q + P(0) , p P(1) [x ] P(1) and p (0) q + , clearly there is a common upper bound r P(1) P of p , q + so r exemplies p , q are compatible in P . So we have nished proving ()2 .] Let D = {D,t : = j for some j < i so < }. Clearly s T t ,t D,s D so the main point is to prove not just P l(D ), but that / ,t ,t moreover [, + ] (S E) {+ } R, [P , ] l(D,,t ) where / x D,,t = {D,t [x ] : < } D,t = {D,t [x ] : < } {D,t : = j for some j < i}.Fixing such , again as D x < is increasing and D,t : = j ,t : to prove x for some j < i is increasing, it suce R, [P , ] l(D ,t D ,t ), x for any < and = j , j < i. For this it suces to prove: ()3 if (A) then (B) where (A) (a) p = (p1 , p2 ) R, [P , x] (b) t T (c) = j < and A D,t a P -name of a subset of N (d) < and B Dx a Px -name of a subset of N ,t (e) n N (B) p R, [P , ] A B [0, n ). x

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

21

Proof of ()3 : Let (0 , p0 ) be a witness for (p1 , p2 ) R, [P, x]; as we can increase 0 , by 3 , and we can increase , without loss of generality 0 = . Without loss of generality p0 , p2 P , as we can increase , moreover as < i +1 E S, similarly without loss of generality S E. Let p P be 2 a common upper bound of p0 , p2 . We dene a Px -name A by: ()3.1 if G Px is generic over V then A [G] = {n: some q P /G forces n A and if p P /G then P |= p q}. 2 2 Easily ()3.2 A is a Px -name of a subset of N ()3.3 P A A . As x Q and S E S Ex and Px P and follows that A Dx . ,t But P [x ] P [x ] hence, recalling (A)() of ()3 : ()3.4
P [x ]

Dx D,t , it ,t

()3.5 hence

P [x ]

A D,t [x ] and B D,t [x ]

A B D,t [x ]. Let p P [x ] be such that p p P [x ] p , p are compatible and without 2 0 0 loss of generality p0 p . Let p3 P [x ] be above p1 and p ; by ()3.6 there are 0 0 q1 and n such that: p3 q1 P [x ] and n n and q1 P [x ] n A B . Let q0 = q1 , it belongs to to P [x ]; clearly q0 q P [x ] p , q are 2 compatible in P [x ]. Also clearly p0 q0 P [x ] so there is r1 such that q0 r1 P [x ] and r1 forces a truth value to n A so as r1 is compatible with q1 , necessarily r1 n A . So p0 q0 r1 P [x ]. By the denition of A and the choice of p0 , there is q2 P [x ] such that: ()3.7 (a) P |= p q2 and q0 r1 q2 2 ()3.6
P [x ]

modified:2011-07-17

revision:2011-07-15

n A. Let (1) < be such that q1 P(1) [x ]; as (q1 ) = q0 r1 q2 and as P , Q : < is a FS iteration, clearly q1 , q2 are compatible in P(1) and let q4 (1) be a common upper bound of (q1 ), q2 . Let q0 P(1) [x ] be such that P q0 q P(1) [x ] q, q4 are compatible in P(1) , so as (q1 ) q4 , without loss of generality (q1 ) q0 . (b) q2
P ()3.8 q0 P [x ] and (, q0 ) witness (q1 , q4 ) R, [P , x]. [Why? As x Q and q1 = q0 q0 clearly q0 P [x ] q1 P [x ]/GP [x ] . For proving q0 P [x ] q4 P /GP [x ] recall the choice of q0 .] ()3.9 (q1 , q4 ) R, [P , ] n A B \[0, n ). x

(915)

22

SAHARON SHELAH

[Why? First, q1 P [x ] n B by the choice of q1 hence (q1 , q4 ) R, [P , ] n x B recalling P [x ] R, [P x] by 7 . , Second, q 4 P n A because q2 P n A and q2 q4 , P P and so (q1 , q4 ) R, [P , ] n A because P R, [P , x] by 6 . x the choice of n. So () holds.] Third, n n recalling 3.9 Together we have proved ()3 . / Lastly, clearly i E and let D = D . If = i S we are done. So assume S; by the induction hypothesis = j < Pj+1 Dj+1 is ultra T -lter increases with , also necessarily cf() = hence P {D,t : system, and D < } : t T is ultra hence D is ultra so we are done. Case 3: i = j + 1, j S S0 . / Let (j , + ] and R = Rj , [Pj , x], recalling 5 we know R satises the c.c.c., by 6 we know Pj R and by 7 we know < P [x ] R. For t T , let D j ,,t = {D,t [x ] : < } Dj ,t , noting Dj ,t = {D ,t : j}, so by the induction hypothesis, R l(D j ,,t ) so D j ,,t = D j ,,t : t T / is a Rj , [Pj ]-name of a T -lter system. Hence there is Di , = Di ,,t : t T , a R-name of an ultra T -lter system above Di , , without loss of generality Di ,,t = l(Di ,,t ) for t T . In particular this holds for = + hence Ei is a + club of where
()4 Ei = { < + : is a limit ordinal from E and if < then Di ,+ ,t P(N)V[P ] : t T is a Rj ,1 [Pj , x]-name for some 1 < } . So we can choose i = (i) Ei E S\(j + 1). Let Pi = Rj ,i [Pj , x] and similarly P = Rj , [Pj , x] for (j , i ) and Di = Di ,+ ,t P(N)V[P(i) ] : t T . Also the choice of Q , g for [j , i ) is dictated by clause (g) of hence also of f and it is easy to check that all the clauses in the induction hypothesis are satised.

modified:2011-07-17

revision:2011-07-15

Case 4: i = j + 1, j S. So Pj Dj is an ultra Pj -lter system. Let = j . Let Q = QD , P+1 = P QD . By Claim 1.9, P+1 [x ] = P [x ] QD [x ] P QD = Pj +1 for < . So R+1, [P+1 , x] is well dened for [+ 1, + ]. be the dual of iddt(),s [P ], a P+1 -name. For t T let D +1,s ()5 R+1, [P+1 , ] l({D,s [x ] : < } D for (, + ]. / x ,s Note that for (, ) we know the parallel statements. ()6 convention: we write (p1 , p2 , p3 ) = (p1 , (p2 , p3 )) for members of R+1, [P+1 , x], where we treat P+1 as P QD , so p2 P and P p3 QD and tr(p3 ) is an object not just a name. We need ()7 if (A) then (B) where (A) (a) p = (p1 , p2 , p3 ) R+1, [P+1 , x] (b) t T

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

23

(c) A is a P+1 -name of a member of D that is, ,t P+1 A is dt(),t -null (d) < and B is a P [x ]-name of a memer of D,t [x ] (e) n N (B) p R+1, [P+1 , ] A B [0, n ). x First note let (, (p0 , p )) where (p0 , p ) P [x ] QD [x ] witness 3 3 p R+1, [P+1 , x] (b) let q2 P be above p0 , p2 (c) let q0 P be such that q0 q P [x ] q2 , q are compatible (d) let B be the following P+1 [x+1 ]-name {n: there is q P [x ]/GP+1 [x ] forcing n B above p1 when p1 P [x ]/GP+1 ] }. [x Next consider ()7.1 (a) A B [0, n ). Why is ()7.2 true? Note that P+1 [x ] B (idd(t ,s) )+ by clause (g) of Denition 2.5, as P [x ] B D,s and B B . Now apply Claim 1.14 for P+1 [x ] = P [x ] QD,t() ] and P+1 = QD,t() . P [x Why is ()7.2 enough for proving ()7 ? As in the proof of Case 2, only much easier. ()7.2
P+1

modified:2011-07-17

Case 5: i = j + 1, j S0 . Let = j ; and let P+1 = P Ql() so Q = Ql() , and again P [x ]Ql() is P Ql() by 1.9. Clearly R+1, [P+1 , x] well dened for [ + 1, + ]. We let D = {D,t : S E} {u,s,n : s T t and n N}, a P+1 -name. +1,t We have to prove the parallel of ()5 , i.e. l(D ) for [ + 1, + ] and t T . / ,t By 2.9 it suces to prove ()8
R+1, [P+1 , ] x

revision:2011-07-15

l({u,s : s T t}) for t T . / Now it is like Case 4 only easier. ()9

R+1, [P+1 , ] x

2.11

{cn.77}

Claim 2.13. If x Q and 2 then we can nd a pair (y, j ) such that (a) x Q y (b) j is an isomorphism from (Px ) /E onto Py extending j1 where j is the canonical embedding of Px into (Px ) /E (c) j maps (Px ) /E onto Py for any < + satisfying cf() = (d) note that j maps j (Px ) to a -subforcing of Py for + satisfying cf() = . Before proving 2.13 recall

(915)

24

SAHARON SHELAH

{cn.81}

Denition 2.14. 1) For a c.c.c. forcing notion P and P-name A of a subset of N we say that p = (pn,m , tn,m ) : m, n < represents A when : (a) pn,m P and tn,m is a truth value (b) for each n, pn,m : m < is a maximal antichain of P (c) for n, m < we have pn,m P n A i tn,m . 2) For p as in part (1) let Ap be the canonical P-name represented by p. Fact 2.15. 1) If P is a c.c.c. forcing notion and A is a P-name of a subset of N then some (pn,m , tn,m ) : n, m < represents A. 2) If P is a c.c.c. forcing notion and A , A are P-names of subsets of , both then represented by (pn,m , tn,m ) : n, m < P A = A . = tn,m : n, m < of truth values, for some formula = 3) For a sequence t 0 () L1 ,1 ( ), = {} where x = xn,m : n < we have: for every c.c.c. t x forcing notion P and pn,m P (n, m < ) we have: P |= ( pn,m : n, m < ) i (pn ,m , tn,m ) : n, m < represents a P-name of a non-empty subset of . 4) For k < , sequences = t : n, m < of truth values for k for some t n,m L1 ,1 ( )-formula = k0 ,...,tk (y, x0 , . . . , xk ) where x = x n,m : n, m < we t have: for every q, p P (n, m < , k), P a c.c.c. forcing notion we have: n,m 1 k P |= [q, p0 n,m : n, m < , pn,m : n, m < , . . . , pn,m : n, m < ] i (pn,m , tn,m ) : n, m < represents a P-name of a subset of which we call A , for k and q P Ak and N\Ak do not almost include A0 A1 . . . Ak1 . Proof. Easy. 2.15 Remark 2.16. In 2.15 we can treat any other relevant properties of such P-names. Proof. Proof of 2.13 Let be large enough, x H () and B = (H (), ) /E and let j the canonical embedding of (H (), ) into B. We now dene (a) P is (Pj() )B if + , cf() = and P = {P : < } if < + cf() = I< = {I<j(+1) )B : < } and E = E y = E x
j(x) j(x)

{p.1b}

modified:2011-07-17

{cn.83}

revision:2011-07-15

(b) (c)

f , a function with domain P is dened by: () f (p) is a function with domain {a : B |= a Dom(j(f (p))} () f (p)(a) = j1 ((f (p)(a))B ) (d) (I , g , Q ) is dened naturallly for < + : () if cf() = as (j(Qx (p)))B () if cf() = , it is {p P+1 : dom(p) [j(), j( + 1))B },
<

etc.

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

25

(e)

E = Ex .

We like to choose Py = P , a pedantic objection is that j is not the identity, moreover P H (++ ); so Px P , by renaming we can overcome this. Also for E {+ } and t T the Py -name Dy are naturally dened such ,t that Dy = {Ap : p represents some Py -name of subset of N and p t, t, p j(D x ) for some p GPy }. t, Almost all the desired properties hold by Los theorem for L1 ,1 as in 2.15. A problem is to show clause (d)() of 2.5, being ultra which means ()
Py

if 1 , s T , E S then Py if A N and A = mod Dy then ,s for some t we have s T t and A Dy . ,t Toward this, as 2 , 1 we have = hence if be a generic branch of T over V so is a subset of T of order type by <T then () E is a -complete ultralter on even in V[] () (Px ) /E is the same in V and V[]. [Why? The proof by the division to two cases: First Case: < . The forcing T adds to V no sequence of length < so obvious.
modified:2011-07-17

Second Case: > . Note that j T is an isormorphism from T onto (j T )B as |T | < .] So by {Dy t : t } is an ultralter on N. , This suces for by 1.5 so we are done.
T

2.13

We lastly arrive to the desired conclusion. Conclusion 2.17. There is P such that (for our T see 2.1(g), 2.4): (a) P is a c.c.c. forcing notion of cardinality and P 20 = (b) T has cardinality 1 + , add no new sequence of length < min(1 ) of ordinals, collapse no cardinal, change no conality (c) P T has cardinality + 1 , collapse no cardinality, change no connality and forces 20 = (d) in VPT we have 1 Sp , i.e. for every 1 there is a non-principal ultralter D of character (e) in VPT we have 2 Sp = (f ) P = Px for some x Q and () Ex S. () Remark 2.18. 1) So if sup(1 ) is strongly Mahlo then |T | = sup(1 ). 2) Similarly in 3.2 for . Proof. We choose x Q by induction on such that

{cn.91}

revision:2011-07-15

{c93}

(915)

26

SAHARON SHELAH

() (a) x Q (b) < x x (c) if = + 1 and cf() = 2 or cf() 2 = min(2 ) then x / is gotten from x as y was gotten from x in 2.13 using E (d) < Ex Ex . For = 0 use 2.8, for successor use 2.13 and for limit use 2.11. Having carried the induction, let x = x . Let S0 = { S0 : C Ex } so a + stationary subset of . Let E = { Ex : = sup( S0 )}. Let () E be such that () has conality . Let () : < be an increasing sequence of members of Ex with limit () such that < ( + 1) S0 . x x x Now letting P = P() recalling P() = {P() : < }, it easily satises all the requirements but we give some details. We have P 20 and |P| by the choice of x0 as P1 [x0 ] P, see 2.8; also P satises the c.c.c. (see 2.10(2)) and P has cardinality , (see Denition 2.5, clause (a)) hence P 20 recalling = 0 . So we have shown clause (a) of the conclusion. Clause (b) holds by the choice of T (see end of clause (g) of the hypothesis 2.1). Now |P| = , |T | 1 hence |P T | + 1 and T P satises the c.c.c. by Hypothesis 2.1(g); hence forcing with P T collapse no cardinal which forcing with T does not collapse; but as 1 = < and the use of Easton support in the product T , forcing with T collapse no cardinal. Similarly forcing with P T changes no conality; together clause (c) of 2.17 holds. As for clause (d), as T is a product, forcing with T adds = : 1 , a -branch of T so in V[] we have {Dx : t 0 }, which is a P-name D of (),t an ultralter on N by 1.5(2), non-principal by 1.2(2). Now for each t T , the lter Dx is (forced to be) generated by the -decreasing u(+1),t,n : < (),t and n N , in the sense that u(+1),t,n+1 u(+1),t,n and for < for some n u u we have n1 N n2 N\n (+1),t,n2 (+1),t,n1 . So D is generated by || + = sets. Now under <T has order type and no Dx is an ultralter (),t and it increases with so clearly < sets do not suce. Hence PT Sp t, for every 1 , so clause (d) of 2.17 holds. Lastly, concerning clause (e), assume that (p, t) (P T ) forces that A P(N) generates a non-principal ultralter D, of character , = |A | and 2 . As cf() > and T T T< and T is + -complete, min(1 \) > (1 )+ 1 , without loss of generality A is a (PT< )-name. As cf() > (1 ) by 2.1(1)(e) for some < , A a (Px T< )-name. As we can increase without is () loss of generality cf() = . Now apply 2.13 recalling clause (c) of (). 2.17

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

27

3. The n s and collapsing A drawback of 2.17 is that V and VP have the same cardinals while the cardinals missing from Sp are ex-large cardinals so weakly inaccessible. In particular it gives no information on chaotic behaviour of Sp among the n s. This is resolved to a large extent below. However, here we do not improve the consistency strength, also we do not deal here with successor of singulars but deal little with singulars. So fullling the second promise from 0 (the rst was dealt with in 2, i.e. 2.17) the main result of this section is Conclusion 3.1. 1) If u {1, 2, . . . , n, . . .} and n 1 n u n + 1 u and in V there are innitely many measurable cardinals, then for some forcing notion P in VP we have Sp = {n : n u}. 2) Assume in V there are innitely many compact cardinals. Then in part (1) we can use any u [1, ). Proof. Straightforward from 3.2, 3.4 below.
3.1

{e.16}

{e.4}

Claim 3.2. Assume G.C.H. for simplicity, Hypothesis 2.1 and 2 > sup( ) and T is -complete for 1 , = cf() for simplicity; let f be a function with domain 2 such that > f () > sup( ), f () > 1 is regular (so f ()<f () = f ()) and f () 2 and let Q be the product {Levy(f (), < ) : / 2 } with Easton support (recall Levy(f (), < ) is collapsing each [f (), ) to f () by approximation of cardinality < f ()). Lastly, let x = x , () be as in the proof of 2.17. Then P = Px T Q () satises: (a) P is a forcing notion of cardinality and P 20 = (b) T has cardinality 1 , and as a forcing notion adds no new sequence of length < min(1 ) of ordinals, collapses no cardinal, changes no conality (c) P has cardinality + 1 , really + |T | + |Q|, collapses no cardinal except those in {(f (), ) : 2 }, changes no connality except that cfV () = (f (), ) cf V[P] () = f (). (d) In VP we have 1 Sp , i.e. for every 1 there is a non-principal ultralter D of character (e) in VP we have 2 Sp = . Discussion 3.3. 1) We may allow f () = sup( ) when sup( ) 2 . / 2) We may like to have successive members of 2 , see 3.4; together with 3.3(1) we get full answer for the n s. 3) We may in 3.2, if = < demand P MA< , for this we need in the inductive choice of the x s for < another case; we do not get MA as cf(()) = . 4) Similarly to part (3) in 1.6, 2.17, 3.6, 3.1. Proof. First, clause (c), on when cardinals and conalities are preserved should be clear. Second, note that forcing by T Q adds no new -sequence of members of V and even preserve Px satises c.c.c. (and even satises the Knaster condition and even being locally 1 -centered) all because T Q is 1 -complete. So P(N)V[P] and even ( Ord)V[P] is the same as the one in V[Px ]. () Third, note that for every 1 , in VT we have a Px -name D of an () ultralter on N with (D ) = , so there is a set D of Px -names of reals of ()
modified:2011-07-17

{e5}

(915)

revision:2011-07-15

28

SAHARON SHELAH

cardinality , or better a set of representations of such names, (see Denition 2.14), which generates D . Now D has the same properties in VT Q (see rst and second above) so V[P] we have Sp so VP |= 1 Sp . Fourth, the main point, we would like to prove that 2 Sp = in VP . So toward contradiction assume 1 2 and (p , r , q ) P forces D is an ultralter on N with (D ) = . Let Q< be {p Q : dom(q) } and similarly Q , Q> so essentially Q = Q Q> and Q = Q< Q where Q = Levy(f (), < ). Similarly T< = {r T : dom(r) }, etc. Now ()1 |T< Q< | < . [Why? Recalling |T< | (sup(1 ))+ (sup( ))+ f ()+ < by an assumption on f and Q< {Q : 2 } has cardinality sup(2 )+ f ()+ < .] ()2 there is a sequence p : < , p a (T Q)-name of a Px -representation () of a subset A of N such that (p , r , q ) P {A : < } generates D and An [0, n) = and (D) = ()3 without loss of generality (p , r , q ) P := Px T< Q and D, () moreover the sequence p : < are P -names. [Why? Because, rst, Q/Q is + -complete as we are assuming 2 \+ f () > . Second, recalling 1 as 1 , 2 are disjoint, forcing by T = T> / adds no new sequence of length of ordinals (by 2.1) and even is + -complete (by the claim assumptions). Third, T< Q has cardinality .] ()4 there are (r , q , q , A ) : < such that: (a) r T< and q Q (b) q is a canonical representation of a Px -name of a subset of N () (c) (p , r , q ) belongs to Px T< Q , is above (p , r , q ) and forces () that Aq , N\Aq are = mod l({A : < }) and {A : < } is included in this lter and the condition also forces p is q (d) A is the Px -name of a subset of N represented by q (e) for technical reasons dom(q ). [Why? As (p , r , q ) forces that {A : < } generates D but A l({A : < / }).] Easily ()5 there are representations q (i < ) of Px -names C i such that i () (a) (p , r , q ) P p {q : i < } for every < i (b) (p , r , q ) {C i : i < } includes {Ai : i < } and is closed under (the nitary) Boolean operations

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

29

(c) (p , r , q ) Px T< Q {C i : i < } D generated D and for () some club E of , if < then {p : < }, {C i : i < } D generate the same lter (d) E is actually a club of from V (e) E (p , r , q ) {C i : i < } is closed under the (nitary) Boolean operations, so even p forces this (for P() [x ] ) ()6 there are r , q from T< , Q respectively and U E such that (a) U r = r q = q so r T< r ; also q Q q (b) q () : U is a -system with heart q () Q (c) if 1 < 2 , 1 U , 2 U then q1 , q2 are compatible4 (d) U E where E is from ()5 (d) [Why? By the proof of Levy(f (), < ) |= -c.c.] ()7 for < < let D be the following Px -name: it is the lter on N () , generated by the family {(C i0 , . . . , C in 1 ) : (x0 , . . . , xn1 ) is a Boolean term and for some U \ we have < n i (, ) and A (C i0 , . . . , C in1 )} ()8 Px D : (, ] is increasing continuous for each < and () , / , D : < is decreasing for each < and D for < < and , if < U then A , N\A are = mod D . , Recall < = cf() and so P() [x ] : < is -increasing with union Px , () hence there is () < of conality such that for every < , q , q are repre sentations of P() [x() ]-name so A , C are P() [x() ]-names and let j() be the j from 2.13, so (j() , x() , x()+1 ) here stand for (j , x, y) there. Recall P() [x ] : < is increasing and is continuous for ordinals of conality > 0 . Let A be j() ( A : U /E ), well abusing our notation a little; you may prefer to use q = j() ( q : < /E ) and A be the P() [x()+1 ]-name represented by q . Now as (p , r , q ) {C i : i < } D generate an ultralter on N and (p , r , q ) is below (p , qmin(U ) , rmin(U ) ) so there is (p1 , r1 , q 1 ) Px T< Q () above it, n N, 0 , . . . , n1 < , Boolean term (x0 , . . . , xn1 ) and truth value t such that ()9 (p1 , r1 , q 1 ) forces (C 0 , . . . , C n1 ) D and is included in (A )[t] recalling A[1] = A, A[0] = N\A hence (C 0 , . . . , C n1 ) (A )[t] . Let p2 P() [x()+1 ] be such that p2 p P() [x()+1 ] p1 , p compatible, so clearly ()10 p1
P() [x]

revision:2011-07-15

modified:2011-07-17

()11 p2

P() [x()+1 ]

(C 0 , . . . , C n1 ) is (A )[t] .

4so even any < f () members are

(915)

30

SAHARON SHELAH

Let p2 : < (P() [x() ]) be such that j() ( p2 : < ) = p2 . Hence (C 0 , . . . , C n1 ) is (A )[t] } belongs to E . Without loss of generality p2 : U1 are pairwise compatible hence by Los theorem for some ()12 U1 = { U : p2 ()13 U1 so < and p2 , p2 has a common upper bound p3 Px()+1 , hence p1 , p3 has a common upper bound p4 Px . () So recalling q is from ()4 , ()14 (p4 , r , q ) forces (a) A D (b) (C 0 , . . . , C n1 ) D (c) (C 0 , . . . , C n1 ) (A )[t] , (A )[t] . Contradiction.

{e.8}

3.2

Claim 3.4. In 3.2 (and 1.6) instead of E is -complete (so is measurable) we may require that there is 2 such that: 2 (a) ( , f ) are as in 3.2 2 (b) dening Q we use if then E is -complete 2 2 (c) if 2 \ then = max( ) is well dened, [, ] 1 = and E 2 2 is a uniform -complete ultralter on so is a -compact cardinal.
modified:2011-07-17

{e.11} {e.30}

Proof. Similar to 3.2.

3.4

Remark 3.5. The situation is similar for any set { : u} of successor of regular cardinals. Claim 3.6. In 3.1 above the sucient conditions for Sp in VP are sucient / also for ()(cf()) = Sp ). / Proof. The same.
3.6

{e.34}

So we can resolve Problem (6) from Brendle-Shelah [BnSh:642, 8]. Conclusion 3.7. If GCH and 1 < = cf() < = , is measurable, then there is a forcing notion P of cardinality collapsing the cardinals in (, ) but no others such that in VP , for every cardinal (, ) of conality , we have Sp = sup(Sp ). /

(915)

revision:2011-07-15

THE CHARACTER SPECTRUM OF (N)

SH915

31

4. Private Appendix Moved 2011.7.04 from the old introduction: We investigate the Sp = character spectrum of non-principal ultralters on , see Denition 0.1 below. On background and early history, see [vD84], [Juh80]. The rst consistency result was of Juhasz who proves the consistency of the character may be < 20 . Here we continue Brendle-Shelah [BnSh:642] by which Sp can be very large and [Sh:846] by which it can be a non-convex set. We prove that if there are enough measurables then we can force that the character spectrum set, {(D) : D an ultralter on N} is quite chaotic. Concerning the proof, on the one hand, as in [BnSh:642] we use a product of a c.c.c. forcing and an Easton support product of a sequence T : , T a tree used to index a name in the c.c.c. forcing notion of a system of lters to get a witness for Sp , and on the other hand, as in [Sh:846] that is as in [Sh:700] we use ultrapower of a c.c.c. forcing notion by a -complete ultralter to get the non-existence of a witness for Sp . Concerning the latter, the reader may be helped by the articles of Brendle, [?], [Bre07] which include exposition of [Sh:700] and probably also by [Bre03]. However, we do not rely on those works, and generally within reason we try to make this work self contained. Moved 2010.11.30 from the proof of 1.8,pg.5: Case 1: There is S with dp(S) = . Let (sn , n ) : n < list {(s, ) : s I, s > } each appearing innitely many times. We choose Sn by induction on n such that:
modified:2011-07-17

(a) Sn moreover dp(Sn ) = (b) if n = m + 1 then Sm Sn (c) if (sn , n ) Sn and k is minimal such that (s , n k ) Sn but / dp(Sn {(sn , n k )) = , (sn , n k ) Sn+1 . For n = 0 let S0 = S by the case assumption the demands hold. For n = m + 1, if (sn , n ) Sn let Sn = Sm ; so assume (sn , n ) Sn ; if there is / k > n such that dp(S {(sn , n k )) 1 , then such k is as required, so assume note. Hence...SAHARON SORT OUT!
{cn.84}

revision:2011-07-15

Moved 2010.11.30,pg.14: Claim 4.1. In 2.13, recall T = Tmin(1 ) is the product {T : 1 } with Easton support and assume 2 and > sup(1 ). Then there is no pair (A , z) such that (a) A is a Px T -name of a family of subsets of [N]0 of cardinality (b) y z (c) Pz l(A ) is an ultralter on satisfying (l(A )) = . Proof. Let = sup( 1 ) and = (2 ), so < . Now we repeat the proof of [Sh:846, 1.1] and anyhow we do a more complicated proof in 3.2 below. 4.1

(915)

32

SAHARON SHELAH

Moved 2010.11.26 from the proof of 1.9,pg.4: Let S = Dom(). First, = because {(s, s ) : s I0 } belongs to Dom() by (b)() of 1.8 and as witnessed by p2 as if q QD1 I0 , tr(q(s)) : s I0 = then p2 P2 q2 , q are compatible in QD2 I0 by the denition of QD2 I0 . Hence there is S with the ordinal (S) minimal; so S satises at least one of the clauses in (b)() of 1.8. Subcase 2A: S satises (b)()(i). This means there is Y such that {(t, t ) : t I0 } S and let p+ witness S 2 QD1 I0 witness Y . So p+ forces that q 2 , q3 QD2 I0 , s I0 s q 2 (s) 2 and s I0 s = tr(q3 , (s)) hence p+ forces that q 2 , q3 are compatible in QD2 I0 . 2 + This implies that q3 (p2 , q 2 ) are compatible in P2 QD2 I hence also q3 , (p2 , q 2 ) are. But this contradicts the choice of q3 (witnessing Y ). Subcase 2B: S satises (b)()(ii) of 1.8. Let (s, ) S be as there. Let U1 = {n: there is p such that P2 |= p+ p 2 2 2 n q2 (s)}. By the denition of and as p+ witness S , and p 2 2 clearly n U1 S {(s, n )} . By the Denition of QD2 I0 we have p2 U l(D2,s ). But U1 V hence by clause (f) of the assumption of the claim 1.9 we have proved really U1 l(D1,s ). But by the present subcase assumption, U2 := {n : S { n } Dom() and (S { n }) < (S)} is = mod l(D1,s ). Together necessarily there is n U1 U2 . Let S1 = S {(s, )}, as U2 , clearly S1 Dom() and (S1 ) < (S) and as n U1 , S1 . Together we get contradiction to the choice of S, i.e. S has minimal (S). Together all cases lead either to the desired conclusion or to a contradiction, so we are done. Moved 2010.11.24 from 0:
{z8}

modified:2011-07-17

Denition 4.2. We say D is double-Ramsey when : (a) D is a lter on N (b) every co-nite subset of N belongs to D (c) in the games (D), (D+ ) the PO player does not lose where for X P(N) the game (X ) is dened as follows: () a play last -moves () in the n-th move the NU player chooses An X the PO player choose kn An

revision:2011-07-15

() in the end, the PO player wins the play if {kn : n < } X .

{z10}

Denition 4.3. 1) We say D is a double-Ramsey when : (a) D = Ds : s I (b) each Ds is a lter on N containing the co-bounded subset of N

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

33

(c) in the games (D) and (D+ ) the PO player does not lose where D+ = + = Xs : s I , Xs P(N) the game () is Ds : s I and for X X x dened as follows: () a player last moves () in the n-th move NU chooses a nite In I and An,s Xs for s In PO chooses kn,s An,s for s In () in the end PO wins when {kn,s : n satises s In } Xs for every s {Im : m < }. 2) We say D is strongly Ramsey when (a), (b) of part (1) holds and (c)+ in the game drn (D) the player PO does not lose where: () a play last moves () in the n-th move + NU chooses nite In I, An,s Ds (for s Is ) PO chooses kn An,s () in the end PO wins when g In such that
n

if t In An,t Dt for innitely many ns then {kn : t In , An,t Dt } Dt

+ if t In An,t Dt for innitely many ns then + {kn : t In , An,t Dt } Dt .

modified:2011-07-17

Remark 4.4. With enough Cohens or just cov(meagre) = 20 , it is easy to construct such Ds Moved 2010.11.22, from pgs.4,5: Remark 4.5. (Was in 1.6, pg. 4): 2) In 4.8(2) below, note that I2 may be the equality. Denition 4.6. For an I-lter system D let QD be dened by (A) p QD i (a) p is a function with domain a nite subset of I (b) for t Dom(p), p(t) is a perfect subtree of > with trunk tr(p(t)) p(t) such that tr(p(t)) p(t) sucp(t) () := {n : n p(t)} l(Dt ) (and, of course, p(t) g() g(tr(p(t)) tr(p(t))). (B) p Q q i (a) Dom(p) Dom(q) (b) p(t) q(t), i.e. p(t) q(t) for every t Dom(p). (C) for p Q let p+ be the function with domain Dom(p) such that p+ (t) = { : tr(p(t)) p(t)} for every t Dom(p). Denition 4.7. 1) For a partial order (or a quasi order) I let EI be the nest equivalence relation on I such that s I t sEI t. 2) For a quasi order I let EI+ = {(s, t) : s I t and t I s}.

{z12}

{cn.21}

revision:2011-07-15

{cn.23}

(915)

34

SAHARON SHELAH

{cn.28ya}

Observation 4.8. 1) If I is a quasi order and D = Dt : t I is an I-lter system then sEI+ t l(Ds ) = l(Dt ). 2) QD does not depend on the order of I, that is, if I1 , I2 are quasi-orders with the same set of elements, D an I -lter system, for = 1, 2 and D1,s = D2,s for s I1 then QD1 = QD2 . Claim 4.9. 1) The property D is an I-lter system is absolute (in order that this will hold we did not demand Ds is a lter on but just Ds P()). 2) We have P1 QD1 P2 QD2 when :

{cn.24}

Moved 2010.11.22, from pg.6:

{cn.31}

Denition 4.10. We say (P1 , D1 ) (P2 , D2 ) when the assumptions (a)-(f) of Claim 1.9(2) holds (hence P1 QD1 P2 QD2 ).

Moved 2010.11.22, from pg.7:

{c36.yajan}

Claim 4.11. If P1 P2 and D is a P -name of a nonprincipal ultralter on for = 1, 2 and P2 D1 D2 , then P1 Q(D1 ) P2 Q(D2 ). Moved 2010.11.22 from the end of the proof of 2.11, pg.11: Q is well dened by (f ), (g). We let P = P Q so by Claim 1.13 the forcing notion P satises the relevant parts of clauses of . But we have also to dene D . For t Ix let D = ,t x {D,t : < satises t Ix } D,t . Now clearly ()1 D is a P -name of a family of subsets of ,t ()2 if s <T t then P D D . ,s ,t The main point is proving l(D ). / ,t Why does this suce? Because by 1.4(2) there is a P -name D such that P is an ultra I-lter system such that D D . Why does ()3 holds? As D x D,t : < increases with , it suces for a given < to prove P x / P l(D ,t D,t ). Subcase 3A: is even as in Case 2. ()3
P

revision:2011-07-15

modified:2011-07-17

Subcase 3B: is odd. x Toward this let R = Px , R1 = P , E 0,t = D,t , E ,s = D,t a R -name, R+ := 1 x + + R QE for = 0, 1. So R0 = P , R1 = P and let E 2 = Dx an R0 -name. So ,t the assumption of Claim 1.8 hence its conclusion so weare easily done. Moved 2010.11.22, from pgs.11,12: We still owe: Claim 4.12.

{cn.58}

(915)

THE CHARACTER SPECTRUM OF (N)

SH915

35

(A) (a) R0 R1 (b) E is an R -name of a lter on N containing the co-nite sets for = 0, 1 (c) (d)
R1 E 0 + R = R

E 1 moreover (E + )V[R0 ] (E + )V[R1 ] 0 1 QE for = 0, 1

(e) R+2 = R QE for = 0, 1 (f ) E2 is a R2 -name of a lter extending E 0 moreover (E + )V[R ] (E + )V[R+2 ] 0 2 / (B) (a) R3 l(E E ). Proof. It is enough to prove (check both are stated!) that 2 holds assuming 1 when 1 (a) p3 R3 (b) A is an R -name of a member of E so subset of N for = 1, 2 2 p3 R3 A1 A2 = . So we shall assume p3 = (p1 , q 3 ) hence (b) p1 forces tr(q 3 ) = . As R0 R1 , there is p0 such that ()2 (a) (b)
modified:2011-07-17

()1 (a)

R1

q 3 QE 1 and without loss of generality

p 0 R0 if R0 |= p0 p then p, p1 are compatible in R1 .

We can work in V[G0 ] where ()3 (a) G0 R0 is generic over V (b) p0 G0 Let ()4 Y1 = {(n, ): for some q R1 /G0 we have R1 |= p1 q and q and q 3 }. n A1
4.12

(915)

revision:2011-07-15

Moved from pg.19: Alternative earlier assumption: Note that Px() Px()+1 Px . As (p , r , q ) forces that {A : < } generates D it also forces that D {A : < , i < } generates D. But Px {A : < ,i ,i i < } is closed under Boolean operations so (p , r , q ) there are (, , i) and B {A , \A } such that A B and A D so let (p1 , r1 , q 1 ) P and ,i ,i (1 , i1 ) and t be such that (p , r , q ) P (p1 , r1 , q 1 ) and (p1 , r1 , q 1 ) P (1 , i1 ) , [t = 1 B = A ], [t = 0 B = \A ] and A 1 ,i1 B . Now there is p2 Px()+1 such that [p2 p Px()+1 p, p2 are compatible in Px ]. Also there is p3 P() such that [p3 p Px() p, p2 are compatible in Px()+1 ]. Let q 2 = q {(, q ())} for any U . By absoluteness also 1 (p2 , q, r ) Px() T< Q (1 , i1 , t) is as above.

36

SAHARON SHELAH

By the choice of j() , (q , A ) it follows that there is a pair (B , p ) 4 () p Px() p4 , p3 are compatible in Px() B is a P() -name of a subset of (p4 , q 2 , r2 ) A D and [t = 1 B = A A 1 ,i1 ] and ,i [t = 0 B = \A A 1 ,i1 ]. Now U E where U := { U such that (r2 , q 2 ), (r2 , q ) T< ) T< Q 5 5 are compatible and let (r , q ) be a common upper bound}. Case 1: t = 1. Get a contradiction to the choice of A (not in l{. . .}) for every large enough U . Case 2: t = 0. Get A is forced to be included in the complement of a member of D, try forced to D ,contradiction.

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

37

5. Private Appendix Further results We may look at further properties in such forcing extensions. Claim 5.1. In all the results above (2.17, 3.2, 3.4) we can add: in VP we have: (a) u = Min(1 ) (b) there is no MAD family of cardinality [, cf()] (c) b = d. Proof. Same proof. More seriously in [Sh:700, 4] we have a more versatile framework. Hypothesis 5.2. As in 2.1, but only Min(1 ). Denition 5.3. 1) We dene Q = Q2 is as the class objects of x consisting of S, P , Q , for < , D,t , ,t (for < , t I) such that: (a) Q = P , Q : < H (+ ) is a FS iteration and P = {P : < } (b) D = D ,t : t I is a P -name of an I-lter system for (c) moreover it is an ultra I-lter system (d) P D D if (e) S { < : cf() 2 } is unbounded below / (f ) P Q = QD
modified:2011-07-17
5.1

{g.4}

{g.42} {g.49}

,t := {tr(p(t)) : p GQ } satises Rang( ,t ) D+1,t . x x 1A) For x Q as above we let Q = Qx , Px = P , Ix = I, D,t = D,t , ,t = . x x ,t , I = I . In 3 we use only Q2 so we write Q and = is the following two-place relation on Q : x y i 2) Q Q (g)
P+1

(a) Px Py for (b) Py Dx Dy for , t I ,t ,t (c) Py x y . (note: clause (b) restrict ourselves concerning the T ). 3) (070729) Use several S : each optional: (a) one for b = = d (b) Sp (c) A = {|A | : A [w]0 is MAD}. Claim 5.4. The upper bound existence claim If x : < is Q -increasing continuous and is a limit ordinal < + then there is x which is a canonical limit of x : < which means (a) (b) (c) (d) x Q x x for < Ix = {Ix : < } if has uncountable conality then
{g.56}

(915)

revision:2011-07-15

38

SAHARON SHELAH

() Px = {Px : < } and () Px Dx = {Dx : < and satises t I } ,t ,t (e) in fact |Px ({|Px | : < })0 .
{g.77}

Proof. Combine the proof of 2.11 and ??. Claim 5.5. If x Q and 2 then we can nd a pair (y, j ) such that (a) x y Q Q (b) j is an isomorphism from (Px ) /E onto Py extending the canonical em bedding of Px into (Px ) /E (c) j maps (Px ) /E onto Py for < is a -embedding (d) |Pg | |Px | /E . Remark 5.6. Saharon: 1) Deal with Sp ? Sp ? 2) Other invariants: b, d which now need: (A) in QD , many the t dominating by a suitable side condition.

(915)

revision:2011-07-15

modified:2011-07-17

THE CHARACTER SPECTRUM OF (N)

SH915

39

6. Assignments Moved 070726, pg.1: Saharon: (Oct. 2006) 1) Fill a proof of 1.8 by [Sh:707], pg.7 here. 2) Fill, like [Sh:509], proof of ()3 , case 2, proof of 2.11. 3) Check end of 4.1. 4) (July 2007) S . 5) S . 6) More invariants? Moved from end of old proof of 3.2,pg.10: ()6 for some representation q of a Px -name A of a subset of , we have: for is every < the set { < : Px A A innite i A A is innite} belongs to E . [Why? By the construction of x = x , the uses of ultra-power by E .] Letting G Px T< Q be generic over V such that (p , r , qmin(U ) ) G. Clearly ()7 { < : (p , r , q ) G} = mod D (the mod D means modulo the lter which D generates in the new universe). Together we get a contradiction.
modified:2011-07-17

(915)

revision:2011-07-15

40

SAHARON SHELAH

{p.2}

7. Remarks on -bases Denition 7.1. 1) A is a -base i : (a) A []0 (b) for some ultralter D on , A is a -base of D, see below. 1A) We say A is a -base of D if (B D)(A A )(A B). 1B) (D) = Min{|A | : A is a -base of D}. 2) A is a strict -base i : (a) A is a -base of some D (b) no subset of A of cardinality < |A | is a -base. 3) D has a strict -base when D has a -base A which is a strict -base 4) Sp = {|A | : D a non-principal ultralter on and A of D}. 5) We say A is a [strict] -base if it is a [strict] -base of D for some ultralter D on . Denition 7.2. For A []0 let IdA = {B : for some n < and partition B : < n of B for no A A and < n do we have A B }. Observation 7.3. For A []0 we have: (a) idA is an ideal on P() including the nite sets, though may be equal to P() (b) if B then: B []0 \ IdA i there is a (non-principal) ultralter D on to which B belongs and A is a -base of D.

{p.4} {p.5}

modified:2011-07-17

Proof. Clause (a): Obvious. Clause (b): The if direction: Let D be a non-principal ultralter on such that B D and A is a -base of D. Now for any n < and partition B : < n of B as B D as D is an ultralter clearly there is < n such that B < n hence by Denition 7.1(1A) there is A A such that A B . By the denition of IdA it follows that B IdA but []<0 IdA so we are done. / The only if direction: So we are assuming B IdA so as IdA is an ideal of / P() there is an ultralter D on disjoint to IdA such that B D. So if B D then B B IdA hence by the denition of idA it follows that / (A A )(A B ). By Denition 7.1(1A) this means that A is a -base of D. Observation 7.4. 1) If D is an ultralter on then D has a -base of cardinality (D). 2) A is a -base i for every n [1, ) and partition B : < n of for some A A and < n we have A B . 3) Min{(D) : D a non-principal ultralter on } = Min{|A | : A is a base} = Min{|A | : A is a strict -base}. Proof. 1) By the denition. 2) For the only if direction, assume A is a -base of D then IdA P()\D (see the proof of 7.2) so IdA and we are done. / For the if direction, use 7.2. 3) Easy. 7.4

{p.3}

(915)

revision:2011-07-15

THE CHARACTER SPECTRUM OF (N)

SH915

41

Claim 7.5. In VP as in [?, 1.1], we have {, } Sp and 2 Sp . / Proof. Similar to the proof of [?, 1.1].

{642.2a}

(915)

revision:2011-07-15

modified:2011-07-17

42

SAHARON SHELAH

References
[Bre03] Jrg Brendle, The almost-disjointness number may have countable conality, Transaco tions of the American Mathematical Society 355 (2003), no. 7, 26332649 (electronic). [Bre07] , Mad families and ultralters, Acta Universitatis Carolinae. Mathematica et Physica 48 (2007), no. 2, 1935. [Juh80] Istvan Juhsz, Cardinal functions in topologyten years later. Second edition, Mathea matical Centre Tracts, vol. 123, Mathematisch Centrum, Amsterdam, 1980. [Kun] Kenneth Kunen, Ultralters and independent sets, Trans. Amer. Math. Soc. 172, 299306. [vD84] Eric K. van Douwen, The integers and topology, Handbook of Set-Theoretic Topology (K. Kunen and J. E. Vaughan, eds.), Elsevier Science Publishers, 1984, pp. 111167. [Sh:509] Saharon Shelah, Vive la dirence III, Israel Journal of Mathematics 166 (2008), 6196, e math.LO/0112237. [BnSh:642] Joerg Brendle and Saharon Shelah, Ultralters on their ideals and their cardinal characteristics, Transactions of the American Mathematical Society 351 (1999), 26432674, math.LO/9710217. [Sh:700] Saharon Shelah, Two cardinal invariants of the continuum (d < a) and FS linearly ordered iterated forcing, Acta Mathematica 192 (2004), 187223, Also known under the title Are a and d your cup of tea?. math.LO/0012170. , Long iterations for the continuum, Archive for Mathematical Logic submitted, [Sh:707] math.LO/0112238. [Sh:846] , The spectrum of characters of ultralters on , Colloquium Mathematicum 111, No.2 (2008), 213220, math.LO/0612240. Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The Hebrew University of Jerusalem, Jerusalem, 91904, Israel, and, Department of Mathematics, Hill Center - Busch Campus, Rutgers, The State University of New Jersey, 110 Frelinghuysen Road, Piscataway, NJ 08854-8019 USA E-mail address: shelah@math.huji.ac.il URL: http://shelah.logic.at
modified:2011-07-17

(915)

revision:2011-07-15