THE SPECTRUM OF CHARACTERS OF ULTRAFILTERS ON SH846

Saharon Shelah The Hebrew University of Jerusalem Einstein Institute of Mathematics Edmond J. Safra Campus, Givat Ram Jerusalem 91904, Israel Department of Mathematics Hill Center-Busch Campus Rutgers, The State University of New Jersey 110 Frelinghuysen Road Piscataway, NJ 08854-8019 USA
Abstract. We show the consistency of statement: the set of regular cardinals which are the character of some ultralter on is not convex. We also deal with the set of -characters of ultralters on .

revision:2008-06-10

modified:2008-06-11

Key words and phrases. characters, ultralter, forcing, set theory. I would like to thank Alice Leonhardt for the beautiful typing. First Typed - 03/Oct/1; moved from F605 Latest Revision - 08/June/11 Partially supported by the Binational Science Foundation and the Canadian Research Chair; 613-943-9382 Typeset by AMS-TEX

(846)

SAHARON SHELAH

0 Introduction Some cardinal invariants of the continuum are actually the minimum of a natural set of cardinals 20 which can be called the spectrum of the invariant. Such a case is Sp , the set of characters (D) of non-principal ultralters on (the minimal number of generators). On the history see [BnSh 642]; there this spectrum and others were investigated and it was asked if Sp can be non-convex (formally 0.2(2) below). The main result here is 1.1, it solves the problem (starting with a measurable). This was presented in a conference in honor of Juhasz, quite tting as he had started the investigation of consistency on (D). In 2 we note what we can say on the strict -character of ultralters. The investigation is continued in [Sh:915] trying to get more disorderly behaviours in smaller cardinals and in particular answering negatively the original question, 0.2(2). Recall 0.1 Denition. 1) Sp = Sp() is the set of cardinals such that: = (D) for some non-principal ultralter D on where 2) For D an ultralter on let = (D) be the minimal cardinality such that D is generated by some family of members, i.e. Min{|A | : A D and (B D)(A A )[A B]}, it does not matter if we use A B. Now, Brendle and Shelah [BnSh 642, Problem 5], asked the question formulated in 0.2(2) below, but it seems to me, at least now that the question is really 0.2(1)+(3). 0.2 Problem 1) Can Sp() Reg have gaps, i.e., can it be that < < are regular, Sp(), Sp(), Sp()? / 2) In particular does 1 , 3 Sp() imply 2 Sp()? 3) Are there any restrictions on Sp() Reg? We thank the referee for helpful comments and in particular 2.5(1). Discussion: This rely on [Sh 700, 4], there is no point to repeat it but we try to give a description. Let 0 < < < be regular cardinals, measurable. Let S = { < : cf() = } or any unbounded subset of it. We dene ([Sh 700, 4.3]) the class K = K,S of objects t approximating our nal forcing. Each t K consists mainly of a nite support iteration Pt , Qt : i < of c.c.c. forcing of i i cardinality with limit P = Pt = Pt , but also Qt -names t (i < ) so formally t i i

(846)

revision:2008-06-10

modified:2008-06-11

THE SPECTRUM OF CHARACTERS OF ULTRAFILTERS ON SH846

Pt -names, satisfying a strong version of the c.c.c. and for i S, also Dt , a Pt i i i+1 name of a non-principal ultralter on from which Qt is nicely dened and At , a i i Qt -name (so Pt -name) of a pseudo-intersection (and Qi , i S, nicely dened) of i+1 i
t Dt such that i < j S At Dj . So {Ai : i S} witness u in VPt ; not i i

modified:2008-06-11

necessarily we have to use nicely dened Qi , though for i S we do. The order K is natural order, we prove the existence of the so-called canonical limit. Now a major point of [Sh 700] is: for s K letting D be a uniform -complete ultralter on , (or just 1 -complete 0 < < ), we can consider t = s /D; by Los theorem, more exactly by Hanfs Ph.D. Thesis, (the parallel of) Los theorem for L, apply, it gives that t K, well if = /D; and moreover s K t under the canonical embedding. The eect is that, e.g. being a linear order having conality = is preserved, even by the same witness whereas having cardinality < is not necessarily preserved, and sets of cardinality are increased. As d is the conality (not of a linear order but) of a partial order there are complications, anyhow as d is dened by conality whereas a by cardinality of sets this helps in [Sh 700], noting that as we deal with c.c.c. forcing, names reals are represented by -sequences of conditions, the relevant thing are preserved. So we use a K -increasing sequence t : such that for unboundedly many < , t+1 is essentially (t ) /D. What does nice Q = Q(D), for D a non-principal ultralter over mean? We need that () Q satises a strong version of the c.c.c. () the denition commute with the ultra-power used () if P is a forcing notion then we can extend D to an ultralter D+ for every (or at least some) P-name of an ultralter D extending D we have Q(D) P Q(D+ ) (used for the existence of canonical limit).

(846)

revision:2008-06-10

Such a forcing is combining Laver forcing and Mathias forcing for an ultralter D on , that is: if p D i p is a subtree of with trunk tr(p) p such that for p we have g() < g(tr(p)) (!n)( n p) and g() g(tr(p)) {n : n p} D.

SAHARON SHELAH

1 Using measurables and FS iterations with non-transitive memory We use [Sh 700] in 1.1 heavily. We use measurables (we could have used extenders to get more). The question on 1 , 2 , 3 , i.e. Problem 0.2(2) remains open. 1.1 Theorem. There is a c.c.c. forcing notion P of cardinality such that in VP we have a = , b = d = , u = , {, } Sp but 2 Sp() if / 1 , 2 are measurable and 1 < = cf() < 2 < = = 2 = cf(). Proof. Let D be a normal ultralter on for = 1, 2. Repeat [Sh 700, 4] with (1 , , ) here standing for (, , ) there, getting t K for which is K -increasing and letting P = Pt we have Q = P : < is a -increasing i i continuous sequence of c.c.c. forcing notions, P = P = Pt := Lim(Q ) = {P : < }; in fact P , Q : < is an FS iterations, etc., but add the demand that for unboundedly many < 1 P+1 is isomorphic to the ultrapower (P )2 /D2 , by an isomorphism ex tending the canonical embedding. More explicitly we choose t by induction on such that
modified:2008-06-11

1 (a) t K, see Denition [Sh 700, 4.3] so the forcing notion Pt for i i is well dened and is -increasing with i (b) t : is K -increasing continuous which means that: () t K t , see Denition [Sh 700, 4.6](1) so Pi Pi for i () if is a limit ordinal then t is a canonical K -u.b. of t : < , see Denition [Sh 700, 4.6](2) (c) if = + 1 and cf() = 2 then t is essentially t1 /D1 (i.e. we have to identify P with its image under the t canonical embedding of it into (P )1 /D1 , in particular this holds for = , see Subclaim [Sh 700, 4.9]) (d) if = + 1 and cf() = 2 then t is essentially t2 /D2 . So we need 2 Subclaim [Sh 700, 4.9] applies also to the ultrapower t2 /D. [Why? The same proof applies as 2 /D2 = , i.e., the canonical embedding of into 2 /D2 is one-to-one and onto (and 1 /D1 = 2 /D2 = , of course).]
t t t

(846)

revision:2008-06-10

THE SPECTRUM OF CHARACTERS OF ULTRAFILTERS ON SH846

Let P = Pt for so P = {P : < } and P = P . It is proved in [Sh 700, 4.10] that in VP , by the construction, Sp(), a and u = , 20 = . By [Sh 700, 4.11] we have a hence a = , and always 20 Sp() hence = 20 Sp(). So what is left to be proved is 2 Sp(). Assume toward / contradiction that p D is a non-principal ultralter on and (D ) = 2 and

let it be exemplied by A : < 2 .

Without loss of generality p

A D does not belong to the lter on

generated by {A : < } {\n : n < }, for each < 2 and trivially also \A does not belong to this lter. As is regular > 2 and the forcing notion P satises the c.c.c., clearly for some < we have p P and < 2 A is equivalent to a P -name.

So for every [, ) we have 2 p

P

for each i < 2 the set Ai []0 is not in the lter on which

{Aj : j < i} {\n : n < } generates, and also the complement of Ai is not in this lter (as D exemplies this).

modified:2008-06-11

But for some such , the statement 1 holds, i.e. 1 (d) apply, so in P+1 which 2 essentially is a (P ) /D2 we get a contradiction. That is, let j be an isomorphism from P+1 onto (P )2 /D2 which extends the canonical embedding of P into j (P )2 /D2 . Now j induces a map from the set of P+1 -names of subsets of 2 j into the set of (P ) /D2 -names of subsets of , and let A = 1 ( Ai : i < 2 /D2 )

so p

P+1

A []0 and the sets A , \A do not include any nite intersection

P+1

of some members of {A : < 2 } {\n : n < }. So p

{A : < 2 }
1.1

does not generate an ultralter on but P+1 P, contradiction.

revision:2008-06-10

1.2 Remark. 1) As the referree pointed out we can in 1.1, if we waive u < a we can forget 1 (and D1 ) so not taking ultra-powers by D1 , so = 0 is allowed, but we have to start with t0 such that Pt0 is adding 2 -Cohen. 0 2) Moreover, in this case we can demand that Qt = Q(Dt ) and so we do not need the t . Still this way was taken in [Sh:915, 1]. But this gain in simplicity has a price in lack of exibility in choosing the t. We use this mildly in 2; mildly as only for P1 . See more in [Sh:915, 2,3].

(846)

SAHARON SHELAH

2 Remarks on -bases 2.1 Denition. 1) A is a -base if: (a) A []0 (b) for some ultralter D on , A is a -base of D, see below, note that D is necessarily non-principal 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 if: (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 |: there is a non-principal ultralter D on such that A is a strict -base of D}. 2.2 Denition. 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 }.
modified:2008-06-11

2.3 Observation. 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 (c) A is a -base i IdA . / 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 and D is an ultralter clearly there is < n such that B D hence by Denition 2.1(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. /

(846)

revision:2008-06-10

THE SPECTRUM OF CHARACTERS OF ULTRAFILTERS ON SH846

/ 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 2.1(1A) this means that A is a -base of D. Clause (c): Follows from clause (b).
2.4

2.4 Observation. 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 to nitely many sets, 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 2.2) so IdA and we are done. / For the if direction, use 2.2. 3) Easy. 2.4 2.5 Theorem. In VP as in 1.1, we have {, } Sp and 2 Sp . /

modified:2008-06-11

Proof. Similar to the proof of 1.1 but with some additions. Dening K in [Sh 700, 4.1] we allow Q0 = Qt = Pt to be any c.c.c. forcing notion of cardinality (this 0 1 makes no change). The main change is in the proof of P Sp . The main addition is that choosing t by induciton on we also dene A such that > (a), (b) as in 1 in the proof of ? 1 scite{1.1} undened (c) as in 1 (c) but only if = 2 mod (and = + 1)

revision:2008-06-10

(d) A is a Pt -name of an innite subset of 1 (e) (f ) (g) if = 2 mod then

Pt

A = (or do not dene A )

t P

if < are = 2 mod then

A A

if = + 1 and = 2 mod and B is a innite subset of then

P
t

Pt -name

of an

A .

(846)

SAHARON SHELAH

This addition requires that we also prove 3 if s K and D is a Ps -name of a lter on including all co-nite subsets 1

of (such that D) then for some (t, A) we have /

(a) s K t (b)
Pt 1

A is an innite subset of
Pt

(c) if B is a Ps -name of an innite subset of then

A.

[Why 3 holds? Without loss of generality nd a pair (P , A )

Ps 1

D is an ultralter on . We can

() P is a c.c.c. forcing notion () Ps P moreover P = Ps Q(D ) 1 1

() |P | ()
P

A is an almost intersection of D (i.e. A []0 and (B D)(A

B)) () is the generic of Q[D ] and A = Rang() so both are P -names.

modified:2008-06-11

Now we dene t : t K t and Pt = P , we do it by dening Qt by induciton on i 1 i as in the proof of [Sh 700, 4.8] and we choose t naturally. Let n : > 2 be i

a Pt -name listing the members of A. 0 Now we choose t such that t K t and for some Pt -name of a member of 2 0 we have
Pt

= for any Pt -name (clearly exists, e.g. when (t, t ) is like (t , s)

above, e.g. do as above with P , adding + such raels, and reect). Now A := {nk : k < } is forced to be an innite subset of A , and if it includes a member

revision:2008-06-10

of P()V[Ps ] or even P()V[Pt ] we get that is from ( 2)V[Pt ] , contradiction.]

()1 Sp , in VP , of course. [Why? As there is a -decreasing sequence B : < of sets which generates a (non-principle ultralter). We can use B as the generic of Qt = Pt /Pt .] +1 ()2 2 Sp . /

(846)

THE SPECTRUM OF CHARACTERS OF ULTRAFILTERS ON SH846

[Why? Toward contradiction assume p P and p a strict -base of D; so p

D is a non-principal

ultralter on and {U : < 2 } is a sequence of innite subsets of which is

P

{U : < } is not a -base of any ultralter on

for every < 2 , hence for some B , : < n we have p B , : < n is a partition of and < < n U

n < and

B , . We now as

in the proof of 1.1, choose suitable < and consider < n : < 2 /D2 ) so p
P+1

B : < n is a partition of to nitely many B . But this contradicts p

P

: < n = 1 ( B , : j {U : < 2 }

sets and < 2 < n U is a -base.] ()3 Sp . [Why? Clearly it is forced (i.e.

P )

that A+2 : < is a -decreasing

sequence of innite subsets of , hence there is an ultralter of D on including it. Now A+2 witness that P()V[Pt+2 ] is not a -base of D (recalling clause (g) of ). As is regular we are done.] 1
2.5

(846)

revision:2008-06-10

modified:2008-06-11

10

SAHARON SHELAH

REFERENCES. [BnSh 642] Jrg Brendle and Saharon Shelah. Ultralters on their ideals and o their cardinal characteristics. Transactions of the American Mathematical Society, 351:26432674, 1999. math.LO/9710217. [Sh:915] [Sh 700] Saharon Shelah. The character spectrum of (N ). Topology and its Applications, accepted. 1004.2083. Saharon Shelah. Two cardinal invariants of the continuum (d < a) and FS linearly ordered iterated forcing. Acta Mathematica, 192:187223, 2004. Also known under the title Are a and d your cup of tea?. math.LO/0012170.

(846)

revision:2008-06-10

modified:2008-06-11