Sie sind auf Seite 1von 25

MAD SATURATED FAMILIES AND SANE PLAYER SH935

SAHARON SHELAH Abstract. We throw some light on the question: is there a MAD family (= a maximal family of innite subsets of N, the intersection of any two is nite) which is saturated (= completely separable i.e. any X N is included in a nite union of members of the family or includes a member (and even continuum many members) of the family). We prove that it is hard to prove the consistency of the negation: (a) if 20 < , then there is such a family (b) if there is no such family then some situation related to pcf holds whose consistency is large; and if a > 1 even unknown (c) if, e.g. there is no inner model with measurables then there is such a family.

0. Introduction We try to throw some light on Problem 0.1. Is there, provably in ZFC, a completely separable MAD family A []0 , see Denition 0.3(1),(4). Erds-Shelah [ErSh:19] investigates the ZFC-existence of families A P() o with separability properties, continuing Hechler [Hec71] which mostly uses MA; now 0.1 is Problem A of [ErSh:19], pg.209, see earlier Miller [Mil37], and see later Goldstern-Judah-Shelah [GJSh:399] on existence for larger cardinals. It seemed natural to prove the consistency of a negative answer by CS iteration making the continuum 2 but this had not worked out; the results here show this is impossible. The celebrated matrix-tree theorem of Balcar-Pelant-Simon [BPS80], BalcarSimon [BS89] is related to our starting point. In Gruenhut-Shelah [GhSh:E64] we try to generalize it, hoping eventually to get applications, e.g. there is a subgroup of Z which is reexive (i.e. canonically isomorphic to the dual of its dual) and less, see Problem D7 of [EM02], no success so far. We then had tried to use such constructions to answer 0.1 positively, but this does not work. Simon [CK96] have proved (in ZFC), that there is an innite almost disjoint A []0 such that B and ( A A)[B A innite] (A A)(A B). Shelah-Steprans [ShSr:931] try to continue it with dealing with Hilbert spaces. Here s and ideals (formally J OB) are central. Originally we have a unied proof using games between the MAD and the SANE players but with some parameters for the properties. As on the one hand it was claimed this is unreadable and on the other hand we have a direct proof, which was presented (for s < a ), in the
Date: June 18, 2010. Research supported by the United States-Israel Binational Science Foundation (Grant No. 2006108). Publication 935. The author thanks Alice Leonhardt for the beautiful typing.
1

{0z.2}

(935)

revision:2010-06-18

modified:2010-06-20

SAHARON SHELAH

Hebrew University and Rutgers, we use the later one. A minor price is that the proof in 2 are saying - repeat the earlier one with the following changes. The major price is that some information is lost: using smaller more complicated cardinal invariants as well as some points in the proof which we hope will serve other proofs (including covering all cases) so we shall return to the main problem and relatives in [Sh:F1047] which continue this work. A related problem of Balcar and Simon is: given a MAD family B we look for such A rening it, i.e. (B id+ )(A A)(A B). At present there is no A dierence between the two problems (i.e. in 1.1, 2.1, 2.6 we cover this too)
{0z.3}

Anyhow Conclusion 0.2. 1) If 20 < then there is a saturated MAD family. 2) Moreover in (1) for any dense J []0 we can nd such a family J . We thank Shimoni Garti and the referee for helpful corrections. Denition 0.3. 1) We say A is an AD (family) for B when A [B]0 is innite, almost disjoint (i.e. A1 = A2 A A1 A2 nite). We say A is MAD for B when A is AD for B and is -maximal among such As. 2) If B = we may omit it. 3) For A []0 , idA is the ideal generated by A []<0 . 4) A MAD family A is saturated when: if B id+ (see 0.7(3)) then B almost A contains some member of A (equivalently: if B id+ then B almost contains A continuum many members of A because if B id+ then there is an AD family A B [B]0 id+ of cardinality 20 ). A Denition 0.4. (1) Let a be the minimal cardinality of a MAD family (2) Let a be the minimal such that there is a sequence A : < + of pairwise almost disjoint (=with nite intersection) innite subsets of satisfying: there is no innite set B almost disjoint to A for < but B A+n is innite for innitely many n-s. Observation 0.5. We have b a a. Remark 0.6. 1) Note that if there is a MAD family A []0 such that B id+ A 0 (2 A A) (B A is innite), then there is a MAD family A []0 such that 0 B id+ (2 A A)(A B) equivalently B id+ (A A)(A B); just A A list our tasks and full them by dividing each member of A to two innite sets to full on task. 2) So the four variants of there is A . . . in 0.3(4), 0.6(1) are equivalent. Notation 0.7. 1) For A let A[] be A if = 1 and \A if = 0. 2) For J []0 let J = {B : B []0 and [A J A B nite]} and also for A = As : s S let A = {As : s S} . 3) idA (B) is the ideal of P(B) generated by (AB)[B]<0 and id+ (B) = [B]0 \ idA (B), A for AB see 7) below; if B = we may omit it. 4) Let A B means that A\B is nite. 5) If C P() and C 2 then IC, (B) is {C B : C A[(A)] for every A C}; if B = we may omit it. 6) In part 5), if is a function extending then let IC, = IC, .

{0z.4}

{0z.3d}
modified:2010-06-20

{0z.3f} {0z.5}

revision:2010-06-18

{0z.6}

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

7) For A P(B2 ) and B1 B2 let AB1 = {A B1 : A A satises A B1 is innite}. Denition 0.8. 1) Let OB = {I []0 : I []<0 is an ideal of P()}. 2) For A let ob(A) = {B : B []0 and B A} so ob() = []0 . 3) means ( ) ( ). 4) We say A is AD in J []0 when A is AD and A J. 5) We say A is MAD in J []0 when A is AD in J and is -maximal among such As. 6) J []0 is hereditary when A []0 A B J A J. 7) J []0 is dense when (B []0 )(A J)[A B].

{0z.8}

(935)

revision:2010-06-18

modified:2010-06-20

SAHARON SHELAH

1. The simple case: s < a


{4d.3}

We here give a proof for the case s < a . Theorem 1.1. 1) If s < a then there is a saturated MAD family A []0 . 2) Moreover, given a dense J []0 we can demand A J . Proof. Stage A: Let = s, so cf() > 0 . For part (1) let J []0 be a dense (and even hereditary) subset of []0 , i.e. as in part (2) and in both cases without loss of generality every nite union of members of J is co-innite, i.e. idJ . / Choose a sequence C : < of subsets of exemplifying s = , i.e. (B []0 ) (B C B \C ). For i < and i 2 let C = Ci , the aim of

this notation is to simplify later proofs where we say repeat the present proof but .... Stage B: For 20 let AP , the set of -approximations, be the set of t consisting of the following objects satisfying the following conditions: 1 (a) (b) T = Tt is a subtree of > 2, i.e. closed under initial segments let suc(T ) = { T : g() is a successor ordinal} and1 c(T ) = { 2: if i < g() then i T } (c) 1 |T | 0 + || t (d) I = It = I : c(T ) = I : c(Tt ) (e) A = At = A : suc(T ) = At : suc(Tt )

such that
modified:2010-06-20

(f ) (g)

A I J or2 A = and St = { suc(Tt ) : A = }


I = {A []0 : if i < g() then A (Ci )[(i)] and if i + 1 < g() then A A(i+1) is nite}, so I is well dened also when c(T ).

We let (h)
t C = C (for generalizations)

revision:2010-06-18

2 AP = {AP : 20 } 3 s AP t i (both are from AP and) (a) Ts Tt (b) Is = It c(Ts ) (c) As = At suc(Ts ). Stage C: We assert various properties of AP; of course s, t denote members of AP: 4 (a) AP partially orders AP t t (b) c(Tt ) I I (c)
t t if c(Tt ) then I OB, i.e. I []<0 is an ideal of P()

1so c({<>}) = {<>, < 0 >, < 1 >} 2the case A = is not needed in this proof

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

(d)

At : St is almost disjoint (so At ob() and = St At At nite; recall that here we can assume St = suc(Tt ))
t if c(Tt ) and g() = then I = s t if s AP t then c(Ts ) c(Tt ) and c(Ts ) I = I (and clause (b) of 3 follow from clauses (a),(c))

(e) (f ) (g)

s if c(Ts )\Ts and Ss and B I then B A is nite

if Ts and Ss but ( then B A is nite.

s ) and B I

[Why clause (d)? Let 0 = 1 St , if 0 1 let = 0 1 hence for some t t t {0, 1} we have 0 , 1 1 so Ak Ik I<k> ob((C )[k] ) t t for k = 0, 1 hence A0 A1 ob((C )[] ) ob((C )[1] ) = . If 0 1 note t t t that A1 I1 ob(\A0 ) by clause 1 (g). Also if 1 0 similarly so clause (d) holds indeed. t Why Clause (e)? Recall the choice of C : < and C : > 2 hence t t < C = C . So if B I , then B I(+1) hence (B C B \C ) for every < , a contradiction to the choice of C : < .] 5 (a) (b) (c)
modified:2010-06-20

< 20 AP AP AP0 = (e.g. use t with Tt = {<>}) if ti : i < is AP -increasing, ti APi for i < , i : i < is increasing, a limit ordinal and = {i : i < } then t = {ti : i < } naturally dened belongs to AP and i < ti AP t

6 let Jt be the ideal on P() generated by {At : St } []<0 . For s AP and B ob() we dene:
s 1 2 ()1 SB = SB := SB SB where s,1 s 1 (a) SB = SB := { c(Ts ) : [B\A]0 I = for every A Js } s,2 2 (b) SB = SB := { c(Ts ): for innitely many , set B A is innite} 3,s 3 (c) SB = SB := SB

Ss and the

revision:2010-06-18

s, s, ()2 SP = SPs, := { Ts : 0 SB and 1 SB } for = 1, 2, 3 and B B s 3 SB = SB = SB .

Note ()3 for = 1, 2, 3 (a) SB is a subtree of c(Ts ) (b)


+ SB B Js 1 SB

(c) SP Ts B
(d) if B A are from []0 then SB SA , SP SP . A B

(935)

SAHARON SHELAH

[Why? For = 1, the rst statement holds by recalling 4 (b), the second, + s SB B Js , holds as I<> = ob(), the third, SP Ts as by the denition of B c(Ts ) we have c(Ts ) Ts . Also the fourth is obvious. For = 2 this is even easier and for = 3 it follows.] ()4 if SB and 0 1 . . . n1 list { : SPB } so this set is nite s and we let Cs (, B) := {(C )[(g( ))] : < n}, then SBCs (,B) = { SB : or }.
s s [Why? Clearly (A I )(A Cs (, B)) by the denition of I , see 1 (g) but s (A I )(|A B| = 0 ) hence B Cs (, B) ob(). As B Cs (, B) B clearly SBCs (,B) SB . Also as SB and as (A s I )(A Cs (, B)) clearly SBCs (,B) and moreover { SBCs (,B) : } = { SB : } by 4 (b). Also as SB and SBCs (,B) are subtrees clearly { : } SB SBCs (,B) and c(Ts ) [ SB SBCs (,B) ]. So to prove the equality it suces to assume < g(), SB , g( ) = , g() > and SBCs (,B) and get a contradiction. If < n and = g( ) s s then (A I )[A (C )[1()] ], so an easy contradiction. If {g( ) : < / n} we can get contradiction to SPB . So we are done proving ()4 .] /

()5 (a) (b)

for every c(Tt ) the set {B ob() : SB } belongs to OB /


if = 1, 2, 3 and B = B0 . . . Bn then SB = SB0 . . . SBn

(c) if A Jt and B2 = B1 \A then SB2 = SB1 , SP 2 = SP 1 for B B = 1, 2, 3

modified:2010-06-20

(d) (e) (f )

2 SB Tt for B ob()

if c(Tt ) then Tt if B ob() and s AP t then s,1 t,1 1 SB SB c(Ts ) 2 3 4


s,2 t,2 SB SB Ts (inclusion in dierent direction? yes!) s,3 t,3 t,2 s,2 s,1 SB SB c(Ts ), in fact (SB c(Ts )\SB SB

revision:2010-06-18

s,1 t,1 if SB \Tt then SB (clause (f) is not used here) ()6 for = 1, 2, 3 we have t (a) if B , < 2 and SB Tt and B (C )[] then SB but 1 SB / (b) if B and SB Tt then for some < 2 we have SB

(c) Jt . / [Why? Read the denitions recalling ()5 (c). For clause (c) recall that Ss As Js and by 1 (f ) we have Js J and by Stage A, ob(J ).] / Stage D: 7 if < 20 , s AP and B ob()\Js then we can nd t AP+1 such that s AP t and B contains A for some St \Ts .

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

This is a major point and we shall prove it in Stage F below. Stage E: We prove the theorem. Let B : < 20 list P() each appear 20 times. By induction on 20 we choose t such that (a) t AP (b) < t AP t (c) if = + 1 then either B Jt or B contains A , for some St \Tt . For = 0 use 5 (b). For limit use 5 (c). For = + 1 use 7 . Now let t AP be {t : < 20 } and recalling ()6 (c) it is easy to check that t is a saturated MAD family, enough for 1.1(1) and recalling that by 1 (f ) it is A J also enough for 1.1(2). Stage F: The rest of the proof is dedicated to the proof of 7 so , s and B are given. The proof is now split into cases. Case 1: Some SB is such that c(Ts )\Ts . s 1 By ()5 (d) we have SB . Clearly as SB there is B1 [B]0 I . Note that g() > 0 as Ts by clause (c) of 1 . s Note that A I Ss A As is nite, e.g. by the proof of 4 (d) or better by 4 (g). Subcase 1A: Assume g() is a successor ordinal. Let B2 B1 be such that B2 J and B1 \B2 are innite. Now dene t as follows: Tt = Ts {}, At is As if suc(Ts ) and is B2 if = , lastly dene t I for Tt as in clause (g) of 1 . Easy to check that t is as required; actually B2 = B1 is O.K., too. Subcase 1B: Assume g() is a limit ordinal. s s Clearly g() < by 4 (e), as I = because B1 I , clearly there is {0, 1} s [] such that B1 := (C ) B1 is innite, let B2 B1 be such that B2 , B1 \B2 are t s innite and B2 J . We dene t by Tt = Ts {, }, A is A if suc(Ts ) t and is B2 if = and I for Tt is dened as in clause (g) of 1 . Easy to check that t is as required. Case 2: SPB = but not case 1. Let B := { : SB }.
1 Subcase 2A: B SB . As SB c(Ts ) by the denition of SB , as we are assuming not case 1 neces sarily SB Ts hence B Ts so g(B ) < . We dene B2 as B AB if AB is well dened and B2 = otherwise; and for = 0, 1, let B := B (Cg( ) )[] \B2 . b So

(935)

revision:2010-06-18

modified:2010-06-20

SAHARON SHELAH ()7 B0 , B1 , B2 is a partition of B

hence by ()5 (b) for some = 0, 1, 2


1 ()8 B SB

easily ()9 = 2, and


()10 := B SB .

[Why? By the denitions noting c(Ts ).] Also as B B clearly


()11 SB SB .

But ()10 + ()11 contradicts the choice of B . Subcase 2B: B SB . / 1 By ()3 (b) + ()6 (c) and the assumption of 7 we have SB and by ()6 (b) clearly 0 SB or 1 SB hence B = . If B = by the denition of B we have B SB , contradiction to the subcase assumption. Hence necessarily g(B ) is a limit ordinal , call it . So < B SB but c(Ts ) Ts hence < B Ts . Now for every < let B, := (B ) 1 B () , 2 2 so clearly B, c(Ts )\SB hence as B, SB , by the denition of SB SB / the set A = { Ss : B, and B A is innite} is nite, so we can nd n = n() < and A , . . . , A / 1 ,0 ,n()1 enumerating A but also B, SB SB , + hence ob(B\ A ) = ob(B\ {A : < n()} is disjoint to IB, Js and by , 0 is disjoint the choice of A and ()4 , ob(B\ A ) = [B\(A . . . A ,0 ,n()1 )] s to I . Let A,n() be AB when dened and otherwise. By the denitions of B, s s I , I we have (for < of course):
B, B

modified:2010-06-20

1 (a) (b)

s s 0 is disjoint to I [B (C )[1B ()] \(A . . . A ,0 ,n() )]


B B

s A {A : St and B, (hence A, (C )[1B ()] )} , B for < n() (not needed presently)

revision:2010-06-18

[Why clause (b)? By the choice of A .] Let A = {B A : A = A for some < , k n() and B A is innite}. ,k So A is a family of pairwise almost disjoint innite subsets of B and if A is nite, still B\ {A : A A } is innite because A Js and we are assuming B Js . / Let := { Ss : B , |A B| = 0 }. Now 2 there is a set B1 such that: (a) B1 B is innite (b) B1 is almost disjoint to any A A (c) if is nite then |B1 A | < 0 (d) if is innite then for innitely many we have |B1 A | = 0

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

[Why? First assume is nite, so without loss of generality it is empty. If A is nite use the paragraph above on A . Otherwise as |A | || + 0 = s and by the theorems assumption s < a a and by the denition of a it follows that 2 holds. Second, assume that is innite, and choose pairwise distinct n for n < . Now we recall that we are assuming s < a and apply the denition 0.4 of a to A and n : n < and get an innite B1 B as required.] 3 (a) (b)
B1 Js if (B

) and Ss then B1 A is nite.

[Why? For clause (a), note that rst B1 B , second B1 is innite by clause (a) of 2 , third Bs Js is proved by dividing to two cases. If is nite use clause / (b) of 3 proved below and clause (b) of 2 ; and if is innite use 2 (d)). So let us turn to proving clause (b); we should prove that [ suc(Ts ) (B ) B1 A nite. If A {A : < , n n()} then either A B is nite hence A B1 ,n A B is nite or A B is innite hence A B A hence B1 (A B) is nite by the choice of B1 but B1 B hence B1 A is nite. So assume A {A : < , n n()}, so by the choice of A,n() for < necessarily / ,m ( B ). Recall that we are assuming that (B ). Together for some < we have = g(B ) < and B and we get contradiction by the choice of A = {A : < n()} and A , ,n() .] s We shall now prove by induction on that B1 I . For = 0 recall B s s s I = []0 , for limit I = {I : < } and use the induction hypothB B B esis. For = + 1 rst note that B1 is almost disjoint to AB if B Ss s [1B ()] suc(Ts ) by 2 (b) and, second, B1 is almost disjoint to (C ) otherwise B recalling 3 (b) we get contradiction to the present case assumption SPB = by s s ()6 (a) + the induction hypothesis; together by the denition of I , I we B B s s have B1 I . Having carried the induction, in particular B1 I = IB ; B B now recalling rst B1 Js by 3 (a), second B1 B by 2 (a) and third, the choice / of A , Js , together they contradict the subcase assumption B SB . / 1 Case 3: None of the above. Without loss of generality

modified:2010-06-20

revision:2010-06-18

1 if B1 B but B1 Js then none of the two cases above holds. / We try to choose n = : n 2 by induction on n such that: (a) SPB (b) if = then (c) { : and SPB } = {k : k < g()}. For n = 0, note that SPB = as not case 2 (and not case 1) so we can choose SPB with minimal length. If n = m+1 and m 2 by the induction hypothesis SPB , hence Ts and by the denition of SPB for = 0, 1 the sequence belongs to SB .

(935)

10

SAHARON SHELAH

First assume { SPB : } = . So B1 := B Cs ( , B) Js / [(k)] noting Cs ( , B) = {C k : k m}, recalling it is dened in ()4 from Stage C using SB ; hence SB1 . or } so case 2 or Now by ()4 we know SB1 = { SB : case 1 holds for B1 , contradiction to 1 . Second, assume we have ()( SPB ) so choose such of minimal length. Hence we have carried the inductive choice of n : n < . For each 2 let = {n : n < }, clearly c(Ts ). Also : 2 is without repetitions and each belongs to c(Ts ), so as |Ts | < 20 there is 2 such that Ts . By clause (c) above we have { : and SPB } = {n : / n < }. Note that 2 Cs (k , B) : k < is -decreasing. Let W = { < g( ): for some Ss we have g( ) = and A B is innite}. First, assume W is an unbounded subset of g( ). In this case choose n W such that n+1 > n g(n ) for n < and we choose n Ss such that g(n ) = n and An B is innite. So can choose an innite B0 B such that n < implies B0 \ {Ak : k < n} Cs (n , B) and (B0 An ob()). So 3 B0 B, B0 Js ; / 4 the set SPB0 is empty.
modified:2010-06-20

n n [Why? By ()4 for each n < we have SB0 { : / SB0 Ts { : or } but Ts so SPB0 = .] 5 SB0 is not empty.

}, hence

(935)

revision:2010-06-18

[Why? By 3 .] By 4 + 5 for the set B0 , case 2 or case 1 hold, so we get contradiction to 1 . Second, assume sup(W) < g( ), so we can choose n() < such that sup(W) < g(n() ). Now [ Ss n() B As is nite] as otherwise recalling c(Ts )\Ts necessarily = g( ) < g( ) and of course g(n() ), 1 but see the choice of n(); so n() S hence n() SB , so we can choose / 2 s an innite B1 B such that B1 In() . So checking by cases, B1 ob() is s almost disjoint to any A , Ss . Obviously B1 I , so for it case 1 holds as exemplied by again contradiction to 1 . 1.1

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

11

2. The other cases Theorem 2.1. 1) If = s = a and cf([s]0 , ) = s then there is a saturated MAD family. 2) If = s = a and U() = , see Denition 2.2 below and J []0 is dense then there is a saturated MAD family J . Recall Denition 2.2. 1) For cardinals (also the case < is OK) let U,, () = Min{|P| : P [] such that for every X [] for some u P we have |X u| }. If = we may omit ; if = = 0 we may omit them both, and if = = 0 = we may omit , , . In the case of our Theorem, it means: U() = Min{|P| : P []0 and (X [] )(u P)(|X u| 0 )}. 2) If in addition J is an ideal on then U,,J () = Min{|P| : P [] such that for every function f : for some u P the set {i < : f (i) u} does not belong to J}. 3) Let Pr(, , , ) mean: and we can nd (E, P) such that (if = 0 we may omit , if = = 0 we may omit them, if = = 0 = we may omit , , ): P = P : E E is a club of and E || divide if u P then u [] has no last member 1 P is -increasing 2 |P | < (e) if w is bounded and otp(w) = and sup(w) acc(E) then for some u, j we have: 1 |u w| 2 j acc(E) 3 u Pj 4 |w j| < ,i.e. j < sup(w) (f ) if i {0} E and j = min(E\(i + 1)), w [i, j), otp(w) = then for some set u 1 u Pj and u (i, j) 2 |u w| . (a) (b) (c) (d) Explanation 2.3. The proof of 2.1 is based on the proof of 1.1. The dierence is that in the proof of 2 of subcase 2B of stage F, if g(B ) = it does not follow that we have |A | < a , so we have to do something else when |A | = a = s. By the assumption U() = there is a sequence u : < of members of []0 such that u and for every X [] for some , u X is innite. Now if e.g. g() = we can use u and apply 2.5 below to appropriate B and get P and add it to the family {C : < } witnessing s = the family P as in 2.5. So s now we really need to use C rather than C . Observation 2.4. If Pr(, , , ) is satised by (E, P) then we can nd (E , P ) as in 2.2(3) but

{g.3}

{g.5}

modified:2010-06-20

revision:2010-06-18

{g.7}

{g.10}

(935)

12

SAHARON SHELAH (d) 2 if j > sup(j E ) then |Pj | j (e) as above but sup(w) E.

{g.19}

Proof. Use any club E acc(E) of such that E |P | |min(E \( + 1))| and nacc(E ) cf() = cf() and let P be P if acc(E ) and be {P : E } if nacc(E ). 2.4
Observation 2.5. Assume B = Bn : n < satises Bn []0 , Bn+1 Bn and |Bn \Bn+1 | = 0 for innitely many ns. Then we can nd P such that

() (a) (b)

P []0 is of cardinality b if A []0 is an AD family, B and (n)(B Bn idA ) then for some countable (innite) P P for 20 function / P 2 we have: for some idA -positive set A B we have: A C [(C)] for every C P and A Bn for every n.

Proof. Proof of 2.5 Let B = {B : B = Bn : n < where Bn is innite, Bn Bn+1 and Bn \Bn+1 is innite for innitely many n < }, i.e. the set of B . satisfying the demands on B / For B B and A []0 let pos(B, A) = {B : B Bn idA for every n}. B there is P []0 of cardinality b such that So the claim says that for every B if A []0 is an AD family and B pos(B, A) then there is a countable innite P P as there. Consider the statement: if B B then we can nd B such that b b (a) B = B : S0 recalling S0 = { < b : cf() = 0 } (b) S b B B
0

modified:2010-06-20

(c) if A is an AD family and B pos(B, A), then for some club E of b, b for every E S0 we have ( n)[B (B,n \B,n+1 ) id+ ] A
b (d) if 1 < 2 are from S0 then for some n < the set B1 ,n B2 ,n is nite.

revision:2010-06-18

Why is this statement enough? By it we can nd a subset B of B of cardinality b b such that B B and for every B B for some B = B : S0 as in b we have S0 B B . Now P, the closure by Boolean operations of B and n < } is as required. {Bn : B Why? Let B B (e.g. B ) and an AD family A []0 and assume B A) be given. pos(B, We choose by induction on n < a sequence B : n 2 such that B B moreover ( n)(B,n \B,n+1 id+ ) for n 2 A B = B if = so n = 0 B pos(B , A) if n 2 if 0 , 1 n 2 then for some k < the set B 0 ,k B

1 ,k

is nite.

For n = 0 this is trivial and for n = m + 1 we use (c), i.e. the construction of B . For every n < , n 2 let B = {Bk,m : k n, m n}. So B id+ A and m < g() B Bm and if 1 = 2 n 2 then for some k < , for every 1 n+k 2, 2 n+k 2 satisfying 1 1 , 2 2 we have B1 B2 is nite.

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

13

Obviously [ 2 (n < )(k < )(B,n \B,k id+ )] hence for each 2 A there is C id+ such that C Bn for n < . A [Why? We try by induction on k < to choose A,k , A ob() such that ,k A,k A, A,k A and m < k A = A ,m and A,k Bk . Now rst, if ,k ,k we succeed then we can nd C ob() such that for every n < we have C An is innite and C \ {A : m < n} Bkn . If there is an innite C C almost ,m disjoint to every member of A, then C = C is as required. If there is no such C then we can nd pairwise distinct A A \ {A n ,m : m < } such that C An is innite for every n < . Clearly An C Bm for every n, m < and there is an innite C C such that C Bm and C A is innite for every n, m < , n so C is as required. Second, if k < and we cannot choose A,k then we can choose C ob() such that n < C Bn and C A,m = for m < k, and C is as required, so we are done.] So P = {Bk,m : k, m < } is as required. So proving is enough. Why does this statement hold? Let f = f : < b be a sequence of members of witnessing b and without loss of generality f is increasing and < < b f <J f . bd For < b let C := {Bn [0, f (n)) : n < } so clearly ()1 (a) < C C (b) < b n < C Bn . We choose = () < b by induction on < b, increasing with as follows: for = 0 let = min{ < b : C is innite}, for = + 1 let = min{ < b : > and C \C() is innite} and for limit let = { : < }. By the choice of f every is well dened, see the proof of below. b So : < b is increasing continuous with limit b. For each S0 let (, n) : C((,m)) : n < n < be increasing with limit and, lastly, let B = C() \
mn

modified:2010-06-20

so B,n = C() \
mn

C((,m)) hence B,n+1 B,n and B,n \B,n+1 is innite

by the choice of (,n)+1 . Clearly B B (also follows from the proof below). b Why is B : S0 as required in ? Clauses (a) + (b) are obvious and clause (d) is easy (as if 1 < 2 then for some n we have 1 < ((2 , n)) hence B1 ,n B2 ,n B((1 ,n)) (B(2 ) \C((2 ,n)) ) B(1 ) (B(2 ) \B(1 ) ) = . Lastly, to check clause (c) of let A be an AD family and B be such that ()2 u = uB := {n < : B Bn idA } is innite, equivalently is . / It is enough to prove that for every < b there is (, b) such that B C \C id+ . A
b [Why is it enough? As then for some club E of b, for every E S0 we + have ( < )( < ) and ( < )()( < < C \C idA ) hence ( n)((C((,n+1)) \C((,n)) ) id+ ) which means ( n)(B,n \B,n+1 ) id+ ) A A as required.] So let us prove .

(935)

revision:2010-06-18

14

SAHARON SHELAH

If fails, for every (, b) there are n = n() and A,0 , . . . , An()1 A such that B C \C A,0 . . . A,n()1 . Without loss of generality n() is minimal hence by ()1 the sequence n() : [, b) is non-decreasing, but b = cf(b) > 0 , hence, for some [, b), the sequence n() : [ , b) is constant and let n( ) = n . As A is AD and B C \C A ,0 . . . A ,n 1 and ( , b) B C \C B C \C A,0 . . . A,n 1 , using A is almost disjoint and the minimality of n = n it follows that {A , : < n } {A, : < n } hence they are equal. So (, b) B C \C A ,0 . . . A ,n 1 . For each n u = uB as B Bn \C id+ , and A ,0 , . . . , A ,n 1 are from A idA , clearly there is kn (B Bn \C )\A ,0 \ . . . \A ,n 1 \{k0 , . . . , kn1 }. By the choice of f there is ( , b) such that u1 := {n < : kn < f (n)} is innite. As f is increasing, clearly n u1 kn < f (n) kn C \C . So {kn : n u1 } []0 is innite and is a subset of B C \C \A ,0 , . . . , A ,n 1 , so indeed holds, so we are done. 2.5 Proof. Proof of 2.1 We prove part (2), and part (1) follows from it. We immitate the proof of 1.1. Stage A: Let = s. Let P []0 witness U() = , for transparency we assume P and u P otp(u) = , this holds without loss of generality as b a = s = . [Why? It is enough to show that for every countable u there is a family Pu of cardinality b of subsets of u each of order type such that every innite subset of u has an innite intersection with some member of P. Without loss of generality u is a countable ordinal and we prove this by induction on . For successor ordinal or not divisible by 2 this is trivial so let n : n < be an increasing sequence of limit ordinals with limit but 0 = 0. Let n,k : k < list [n , n+1 ) with no repetitions and let f : < b exemplies b, each f increasing and let P = {P : < } {{n,k : n < , k < f (n)} : < b}. Clearly P has the right form and cardinality. Lastly, assume v u is innite, if for some < , u is innite use the choice of P . Otherwise let f be dened by f (n) = min{k : (m)[n mm,k v]}, and use < b large enough.] Let u : < list P possibly with repetitions, without loss of generality n un = and > u . For < let (, k) : k < list u in increasing order and ,k = (, k). Let U : < be a partition of , to sets each of cardinality such that min(U1+ ) sup(u ) + 1 and U0 . Let C : U0 list a subset of P() witnessing s = and as in Stage A of the proof of 1.1, the set J ob() is dense and idJ . / If B is as in the assumption of 2.5 and (0, ) let PB be as in the conclusion of 2.5 and for < let CB,,i : i U list PB .

(935)

revision:2010-06-18

modified:2010-06-20

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

15

s Stage B: As in the proof of 1.1 but we use C ( Ts ) which may really depend t t on s and where C , B, are dened in clauses 1 (e), (g), (h), (i), (j) below (so the (e), (g), (h) from 1.1 are replaced) and depend just on Tt , At and It , too3 where

1 (a)-(d) and (f) as in 1.1 of course and (e) 1 as before, i.e. A = At : suc(Tt ) , t 2 C = Ct = C : t where t = { : i 2 and i U0 or > 0, i U and (sup(u )) Tt or (for 2.6) Tt } 3 we stipulate A = if T \ suc(T ) t (g) as in 1.1 but replacing Ci by Ci (h) (i)
s if i U0 and Tt i 2 then C = Ci t if (0, ) and sup(u ) 2 and both Ci : i u and t t t Ai : i u are well dened then we let B, = B,,n : n < t t be dened by B,,n = {(C(,k) )[((,k)] \At (,k) : k < n} if (0, ), i U hence i sup(u ) and i 2 and B t t is well dened then, recalling stage A, C = CB t sup(u ), ,,i

(j)

sup(u ),

t Note that Tt , At determine t, i.e. It , t , Ct and B, : , as above . Stage C: As in 1.1 we just add: 4 (h)
modified:2010-06-20

(i) (j)

s t if s AP t and B, is well dened then B, is well dened and equal to it s t if s AP t and C is well dened then C is well dened and equal to it, so s t s C is well dened when c(Ts ) > 2.

s In the proof of 4 (e) use the choice of C : i 2, i U0 , i.e. of C : U0 in Stage A.

Stages D,E: As in 1.1. Stage F: The only dierence is in the proof of 2 in subcase(2B). Recall

revision:2010-06-18

Case 2: SPB = but not Case 1, i.e. SB Ts , recall B , B Js . /


Subcase 2B: B SB where B = { : SB } /

2 there is a set B1 such that (a) B1 B is innite (b) B1 is almost disjoint to any A A (c) if is nite then |B1 A | < 0 (d) if is innite then for innitely many we have |B1 A | = 0 .
3also here we require suc(T ) A = s

(935)

16

SAHARON SHELAH

Why 2 holds? If |A | < then A has cardinality < = s hence by the theorems assumption |A | < s = a , so 2 follows as in the proof of 1.1. So we can assume |A | = but |A | 0 + |g(B )| hence necessarily g(B ) = follows and let for some n() we have A B A , (equivalently A B is innite) but , A {A 1 ,1 : 1 < and 1 n(1 )}}. , / For W choose () n() such that B A / ,() is innite and A,() {A1 ,1 : 1 < and 1 n(1 )}, in fact by 1 (b) the last condition follows. As n() < for < , clearly |W| = because |A | = , hence by the choice of P there is u P such that |W u | is innite; let () [, ) be such that u() = u and let = B sup(u ); recall that otp(u ) = ; note that W := { < :
s s 2.1 k < B,(),k+1 B,(),k .

[Why? By their choice in 1 (i).] Recall also that (),k : k < list u in increasing order and so 2.2 v := {k < : (),k W} is innite;
s s 2.3 [B,(),k ]0 I((),k) for k < . s [Why? As for k(1) < k, (C((),k(1)) )[(()(),k(1))] and \A ((),k(1)) belongs s s 0 to {X : [X] I((),k) } hence by the denition of B,(),k in 1 (i) it satises 2.3 .] s s 2.4 k v B B,(),k \B,(),k+1 is innite.

modified:2010-06-20

[Why? For k v let = ((), k), n = n() and = (). On the one hand [B A ]0 [A ]0 IB . On the other hand A is disjoint to , , , s s [B ()] s (C ) \A if A, = A,n trivially and is almost disjoint to (C )[()]
s s otherwise (i.e. as [(C )[1()] ]0 I( ) 1() {A }). Hence (C )[B ()] \A , , B B is almost disjoint to B A, , an innite set from IB hence by 2.3 from s s s s s s [B,(),k ]0 . So BB,(),k \B,(),k+1 = BB,(),k \(B,(),k ((C )[B ()] \A ) ,k B almost contains this innite set hence is innite as promised.] So by the choice of PB ,() , i.e. 2.5 and clauses (i),(j) of 1 for some s s U() so () g() we have B\(C )[] Js for = 0, 1 hence B1 := / s s B\(C )[B ()] Js recalling that for U , = 0 and 2 the set C depends / B just on g() and sup(u ) (and our s). Now consider B1 instead of B, clearly SB1 is a subset of SB and B ( + 1) is not in it, but B1 ob(B) hence SB1 SB hence SB1 is {B : } and B1 fall under subcase (2A) as < = g(B ). 2.1
B B B B

revision:2010-06-18

{g.21}

Theorem 2.6. There is a saturated MAD family A J when a < = s, J ob() is dense and Pr(, a), see 2.2(3). Proof. Proof of 2.6 We immitate the proofs of 1.1, 2.1. Note that b a < s. Stage A: Similarly to stage A of the proof of 2.1; let (E, P ) be as in Denition 2.2(3) and Observation 2.4, as b < , without loss of generality u P otp(u) =

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

17

for [, ). As we can replace E by any appropriate club E of contained in acc(E) see 2.4 there, without loss of generality otp(E) = cf(), min(E) and E + 1 + b < min(E\( + 1)). Let i : i < cf() list E in increasing order. Let u : < be such that u : i < i+1 list Pi+1 (which includes Pi ) and uj = for j < 0 . Let U : < be a partition of {2i + 1 : i < } such that min(U1+ ) + , |U1+ | = b, |U0 | = , 1 < i U i . Let Ci : i U0 list a family of subsets of witnessing s = also J is as in the proof of 1.1. Let PB , CB,,i : i U be as in 2.1, Stage A. Stage B: As in 2.1, i.e. the case s = a, but we change 1 (f ) 1 (f ) A I J or A = and St := { Tt : A = } { Tt : g() = i + 1 for some i < }.

Stage C: As in the proof of 2.1. Stage D: Here there is a minor change: we replace 7 in 1.1, 2.1 by 7 , 8 , 9 below
+ 7 if < 20 , s AP and B Js then there are a limit ordinal \E t and t AP+1 such that s AP t and |SB 2| = 20 ; we may add St = Ss .

modified:2010-06-20

This is proved in Stage F. To clarify why this is O.K. recall 6 (f ) and note that
s t () if s AP t, B ob(), SB \Ts and Tt then SB . /

Now we need
t 8 if \E is a limit ordinal, < 20 , t AP , B ob() and |SB 0 2| = 2 and = min(E\) then for every t1 and + < 20 such that t AP t1 AP there is t2 , t1 AP t2 AP+1 satisfying / ( suc(Tt2 ))[ Tt1 At2 ob() At2 B].

(935)

revision:2010-06-18

The proof of 8 is like the proof of Case 1 in Stage F in the proof of 1.1 but we elaborate; we are given , , and t1 such that t AP t1 AP ; now we choose t t SB 2\Tt1 exists as |SB 2| = 20 > |Tt1 | so recalling ()5 (e) necessarily t1 ,1 t SB . Choose B1 such that B1 B, B1 I . t Note that for every [, + 1) either C1 is well dened for every 2 t such that and its value is the same for all such (when is odd) or C1 for t1 +1 t 2 is not well dened (when is even). So B = {C : 2 and C1 is well dened} is a family of || < = s subsets of B1 hence there is an innite t t t B2 B1 such that > 2 (C well dened) B2 C B2 \C and without loss of generality B2 J . We choose such that +1 2 and

18

SAHARON SHELAH

t t [g() < + 1 (C is well dened) B2 (C )[] {0, 1} () = ]. Let us dene t2 AP++2 := AP+1 (as + + 1 and |1 | = |2 | AP1 = AP2 ) as follows:

(a) Tt2 := Tt1 { : } (b) At2 is At1 if well dened, is B2 if = and is if suc(Tt2 ) but At2 is not already dened t2 t t (c) C2 is C1 if Tt1 and we choose C by induction on [, + 2] as follows: if it is determined by 1 we have no choice otherwise let it be [()] . The other objects of t2 are determined by those we have chosen. So 8 holds indeed. 9 if s AP and c(Ts ) then for some t, s AP t AP+3 and Ts s s t Tt Ts {, 0 , 1 } and I = Tt and I = < 2I<> = Tt . [Why? Easier than 8 .] Stage E: Similar to 1.1 with the changes necessitated by the change in Stage D. Stage F: We prove 7 , the proof splits to cases. Case 1: Some SB is such that c(Ts )\Ts . s Let B1 ob(B) I , there is such B1 as SB but SB as Ts . / s,2 / [(n)] Let C,n ob() for n < be such that {C,n : n < g()} B1 is innite for every > 2. t We choose Tt = Ts { : > 2}. For > 2, we choose C by induction t on g(): if g() = g() + n is even and n {2m, 2m + 1} then C = C,m , otherwise we act as in the proof of 8 . Lastly, let At = for > 2. Easily
s () if s AP s1 , |Ts1 | < 20 then |SB1 g()+ 2| = 20 .

modified:2010-06-20

So we are done with Case 1.


s Case 2: SPB = but not Case 1 and let B = { : SB }.

revision:2010-06-18

Subcase 2A: B SB As in the proof 2.1 but end as in Case 1. Subcase 2B: B SB / Except 2 which we elaborate this is as in the proofs of 1.1, 2.1 but in the end replace Subcase 1B by Case 1. Recall from Stage 2B as in the proof of 1.1, := g(B ) is a limit ordinal and B, = (B ) 1 B () for < and A,n : n n() : < and A := {B A : A = A,k for some < , k n() and B A is innite}, = { Ss : B and A B is innite} are as in Stage (2B) of the proof of 1.1. We have to prove:

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

19

2 there is a set B1 such that: (a) B1 B is innite (b) B1 is almost disjoint to any A A (c) if is nite then |B1 A | < 0 (d) if is innite then for innitely many we have |B1 A | = 0 . Why 2 holds? If |A | < a then as in the proof of 1.1 the statement of 2 follows. So we can assume |A | a and let W := { < : and let W = { W : | W| < a }. Subcase 2B(): sup(W ) acc(E )
For W choose () n() such that B A / ,() is innite hence A, {A1 ,1 : 1 < , 1 n(1 )}. As n() < for < , clearly |W| = |A | a hence otp(W ) = a. So by Denition 2.2(3), i.e. the choice of (E, P), there is a pair (u , j ) as in clause (e) there. So u P hence u = u() for some () [j , j+1 ), j+1 < j sup(W ) and W sup(E j ) = W sup(E j ) has cardinality < a and let = B sup(u ); recall otp(u ) = . Recall also that (),n : n < list u in increasing order and so v := {n < s s : (),n W} is innite and clearly n v B,(),n \B,(),n+1 is innite as in the proof of 2.1. So by the choice of PB ,() , i.e. 2.5 and clauses (i),(j) of s s / 1 for some U() so g() we have B\(C )[] Js for = 0, 1 hence s s B1 := B\(C )[B ()] Js recalling that for U , = 0 and 2 the set C / B depends just on g() and sup(u ) (and our s).
B

for some n() we have A B A , (equivalently A B is innite)}. ,

modified:2010-06-20

We nish as in the proof of 2.1.


Subcase 2B(): sup(W ) (i , j ] where j = i + 1 so i , j E and let = sup(W ). Apply Denition 2.2(3), clause (f) to i , j , W \j we get u = u Pj so u = u() [i , j ) for some () (i , j ). Let = sup(u ), = so by 2.1 + 2.4 in stage 2B in the proof of 2.1, and by the choice of PB,() there is a Q PB s of cardinality 0 and Q 2
,()

revision:2010-06-18

+ of cardinality 20 such that for every there is B ob(B) Js such that


(),k A Q B A[(A)] and n < B,n := Bn \ {C B (),k : k < n} B s I (),n . B Clearly for some v U() (i , j ) of cardinality 0 , Ts g() s sup(v) {C : v} = Q. For analyzing SB and recalling j+1 < s clearly SB { Ts : g() sup(u )} is {B : < sup(u )} and sup(u ) > + 1, so there is no such that As is non-empty, Ts and sup(u ) g() < j , so SB { Ts : g(s) < j }

[ (

)]

(935)

20

SAHARON SHELAH

does not depend on As : suc(Ts ), g() j so we can nish easily as in case 1.

Case 3: As in the proof of 2.1.

2.6

(935)

revision:2010-06-18

modified:2010-06-20

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

21

3. Further Discussion The cardinal invariant s plays here a major role, so the claims depend on how s and a are compared; when s = a it is not clear whether the further assumption of 2.1(2) may fail. If s > a > 1 , it is not clear if the assumption of 2.6 may fail. Recall 1.1, dealing with s < a , the rst case is proved ZFC, but the others need pcf assumptions. All this does not exclude the case s = +1 , a = 1 hence b = 1 , as in [Sh:668]. Fullling the promise from 0 and the abstract. Claim 3.1. 1) If there is no inner model with a measurable cardinal (and even the non-existence of much stronger statements) then there is a saturated MAD family, A. 2) Also if s < there is one. 3) Moreover, if J ob() is dense then we can demand A J . Proof. As Theorems 1.1, 2.1, 2.6 cover them (using well known results). We now remark on some further possibilities. Denition 3.2. 1) We say S ob() is s-free when: (a) for every A ob() there is B ob(A) such that B induces an ultralter on S ; i.e. C S A C A (\C). 1A) We say S ob() is s-free in I when I OB and for every A I there is B ob(A) which induces an ultralter on S. 2) We say S ob() is s-richly free when clause (a) and
modified:2010-06-20
3.1

{4q.0}

{4q.1}

(b) if A ob() and the set {D S : D an ultralter on containing ob(A)} is innite, then it has cardinality continuum. 3) We say S ob() is s-anti-free if no B ob() induces an ultralter on S . 4) Let S be {: there is a -increasing sequence Si : i < of s-richly-free families such that {Si : i < } is not s-free. 5) Recall s = min{|S | : S ob() and no B ob() induces an ultralter on S }. 6) We say chdim (B) < when ( B 2)(I = IB, ) cf(I, ) < recalling 0.7(5). Observation 3.3. 1) If S is s-free and S S then S is s-free. 2) If S ob() and |S | < s then S is s-free. 3) If Sn ob() is s-free for n < then {Sn : n < } is s-free. 4) s S. 5) S i cf() S. 6) S 1 20 . 7) In Denition of S we can add {Si : i < } is s-anti-free. 8) cf(s) > 0 , in fact S cf() > 0 . Denition 3.4. 1) We say A ob() obeys f when : for every n1 < n2 from A we have f (n1 ) < n2 . 2) Let f = f : < be a sequence of members of . We say A = A : u obeys f when u and A obeys f for A. 3) af = Min{|u|: there are B ob() and A = A : u obeying f such that {A B : u} is a MAD of B}.

{4q.4}

revision:2010-06-18

{4q.7}

(935)

22

SAHARON SHELAH

Remark 3.5. 1) Also note that in 1.1, 2.1, 2.6 we can replace s by a smaller (or equal) cardinal invariant stree , the tree splitting number. 2) Let stree be the minimal such that there is a sequence C = C : > 2 such that C ob() for > 2 and there is no 2 and A ob() such that [(] < k A C . Note that the minimal for which there is such sequence > C : 2 has uncountable conality. 3) Also in 1.1 we may weaken s < a to s < a s a .

{4q.12}

(935)

revision:2010-06-18

modified:2010-06-20

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

23

4. Private Appendix Discussion 4.1. Question: 1) Can we nd such MAD for every ? ( = (2 ) is the rst case and seems the problem). 2) Can we in [1][Sh:669], let: () CON(20 = 2 + = (or just limit of conality 0 ?) and every MAD A []0 , for some u []0 , |A P()| = 1 . E.g. (A) assume is 1.5 - huge as witness by the elementary embedding j, G.C.H. for simplicity 1 = , 2 = j(), = + . Now we need a MAD which is Cohen absolutely so. Discussion 4.2. 1) We may like to have only b levels. In this case our question is: () for regular (a) if B, B : < is -decreasing, = cf() let IB = {A : < A B } an ideal, (b) BB = {: in IB there is a -increasing sequence of length with no upper bound} (c) bB = min(BB ) (d) b = sup{bB : B seq as above} (e) b = Min{: if Reg then b }. Question 4.3. If < b is b b? I.e. b = b ? Or use sup(B)? 2) Even if b = b , to get a slender G G , we seem to need: (as in [Sh:F844]): = g() , F = Fi : restrict the f (generating the G2 , G2 ) such that f [1 , i+1 ) Fi i < is -increasing mod nite, |Fi | [ni , ni+1 ] so = ( i)({f (n), f (n) : n [ni , ni+1 )} = 0. The point is that we can continue of coures f (n) is divisible by i! if n ni or something like that. 3) The case B has no last member are more complicated (sup(B)) can be singular or weakly inaccessible. 4) Alternatively: if Discussion 4.4. We can dene for 1 or any variant a parallel of h. Dene a game.
modified:2010-06-20
0 +

{2q.13}

{2q.17}

{4q.19}

{4q.32}

revision:2010-06-18

{4q.26}

Discussion 4.5. (080215) We try to combine the ideas of present [S1][Sh:900] and last weak. The later one means that letting := s and A : < exemplies it we use the tree T , T = { 2: for some innite B , B idA }, id = {B : T =
<

we have < B A, }. For any C []0 , we can prove that there is a countable u such that for a <()> : } is innite. perfect set of u 2, we have (w [u]<0 )(B {A But cf() > 0 so () = { + 1 : u} < , so for there is () 2 such that [B]0 id<A< (B)> :<()> = .

<(B)>

(935)

24

SAHARON SHELAH

This is not enough for each (or just may) suc(T ) we should choose A id \[]0 which approximate the desired MAD (so almost disjoint along each version: incomparable s are automatically almost disjoint; let A = {At : At dened}, t = A : suc(T )}. A rst problem how can we make it disjoint. By the above it seemed each B []0 not in the ideal At , well this seems reasonable, i.e. for each such B will have 20 changes, so by reasonable bookkeeping each can be treated. But this leaves us with the problem of contradicting saturativity along some branch. This is the second idea: for each A , suc(T ) there is B such that () (a) suc(T ) A B nite (b) suc(T ) A B . This by using a maximal -increasing sequence in idt , and if (id , ) is not 1 -directed we terminate immediately. Denition 4.6. Denition/Choice: 1) = s. 2) A = A : < exemplies this, i.e. A for < (is innite co-innite) but for no innite B do we have < B A B \A . 3) [Here] for B let idB = {A : A B}. 4) [Here] is when g() = + 1.
{4q.34}

{4q.31}

modified:2010-06-20

Discussion 4.7. On Forcing: We may try to immitate [1][Sh:668] by building or probably forcing by P such that, e.g. |P| = , when = +1 , GCH holds and S1 I[]. / This may apply to a family of such problems, so we may consider the general forcing of this kind. If the approximations are p = (PP , , up + history) e.g. up []1 , p : up we need an algebra B controlling dependence or better the u = cB (u ) up pu AP p. E.g. p < AP q Pq /Pp is adding Cohens, [alternatively, try iterated creatures]. Moved from pg.18: We may use games, e.g. of length = 1 . This simplify. Note that in the games we can let the MAD player choose C {A : A []0 is non-empty countable and almost disjoint} such that (A ob())(A C )(A ob(A)] (or so) and the SANE player have to choose A from C . This leads to: for x AP, all members of Bi := {Ax : suc(Tx ), g() = i + 1} are fi -thin, i.e. A Ai ( n)(|A (0, fi (n))| n where fi : i < 1 exemplify B = 1 . Note: if A ob() is MAD, |A| = 1 then there is B ob(), |B| = 1 such that B id+ (continuum I WB )(ob(B) I = ] as if A = {A : < 1 }, for A each [, 1 ) deal with {B : (0 i < )(B A innite)}. But dene S.

revision:2010-06-18

References
[BPS80] Bohuslav Balcar, Jan Pelant, and Petr Simon, The space of ultralters on N covered by nowhere dense sets, Fundamenta Mathematicae CX (1980), 1124. [BS89] Bohuslav Balcar and Petr Simon, Disjoint renement, Handbook of Boolean Algebras, vol. 2, NorthHolland, 1989, Monk D., Bonnet R. eds., pp. 333388.

(935)

MAD SATURATED FAMILIES AND SANE PLAYER

SH935

25

[CK96] Petr Simon (CZ-KARL), A note on almost disjoint renement. (english summary), Acta. Univ. Carolin. Math. Phys. 37 (1996), 8899, 24th Winter School on Abstract Analysis (Beneova s Hora, 1996). [EM02] Paul C. Eklof and Alan Mekler, Almost free modules: Set theoretic methods, North Holland Mathematical Library, vol. 65, NorthHolland Publishing Co., Amsterdam, 2002, Revised Edition. [Hec71] Stephen H. Hechler, Classifying almost-disjoint families with applications to N N , Israel Journal of Mathematics 10 (1971), 413432. [Mil37] E.W. Miller, On a property of families of sets, Comptes Rendus Varsovie 30 (1937), 3138. [ErSh:19] Paul Erdos and Saharon Shelah, Separability properties of almost-disjoint families of sets, Israel Journal of Mathematics 12 (1972), 207214. [GhSh:E64] Esther Gruenhut and Saharon Shelah, Abstract matrix-tree. [GJSh:399] Martin Goldstern, Haim Judah, and Saharon Shelah, Saturated families, Proceedings of the American Mathematical Society 111 (1991), 10951104. [Sh:668] Saharon Shelah, Antihomogeneous Partitions of a Topological Space, Scientiae Mathematicae Japonicae 59, No. 2; (special issue:e9, 449501) (2004), 203255, math.LO/9906025. [Sh:669] , Non-Cohen Oracle c.c.c., Journal of Applied Analysis 12 (2006), 117, math.LO/0303294. , Constructing Abelian groups. [Sh:F844] [Sh:900] , Dependent theories and the generic pair conjecture, Communications in Contemporary Mathematics submitted, math.LO/0702292. [ShSr:931] Saharon Shelah and Juris Steprans, MASAS in the Calkin algebra without the continuum hypothesis, Journal of Applied Analysis 17 (2011), 6989. [Sh:F1047] Saharon Shelah, More MAD saturated families. 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

(935)

revision:2010-06-18

modified:2010-06-20