On closed P -sets with ccc in the space R.Frankiewicz(Warsaw) S.Shelah(Jerusalem) P.Zbierski(Warsaw) Abstract.

t. It is proved that consistently there can be no ccc closed P -sets in the remainder space . In this paper we show how to construct a model of set theory in which there are no P -sets satisfying ccc (countable antichain condition) in the ultralter space = | \ . The problem of the existence of such sets (which are generalizations of P -points) was known since some time and occurred explicitly in vM-R. In the proof we follow the construction from S of a model in which there are no P -points. A particular case of P -sets, which are supports of approximative measures has been settled in M, where the author shows that there can be no such measures on P ()/f in. (Under CH, e.g. the Gleason space G(2 ) of the Cantor space is a ccc P -set in which carries no approximative measure). Sec.1. Closed P -sets in the space can be identied with P -lters F on . Thus, the dual ideal I = { \ A : A F } has the property: If An I, for n , then there is an A I such that An A, for each n .

(1.1)

modified:1993-08-27

Further, the countable chain condition imposed upon F implies that I is fat in the following sense (see F-Z): if An I, for n and limn min An = , then there is an innite Z such that
nZ

(1.2)

An I.

Indeed, let en = An \ A, where A I is as in (1.1). Since min An are arbitrarily large, we can nd an innite set Y such that the family {en : n Y } is disjoint. If {Y : < c} is an almost disjoint family of subsets of Y , then the unions

revision:1993-08-27

S =

{en : n Y },

<c

are almost disjoint and hence the closures S in the space are disjoint. By ccc we have S

{B : B F } = ,

for some and consequently S I. It follows that th e union An =

nY nY

(An A)
nY

(An \ A)

396

is in I as a subset of S A. Let us x a given ccc P -lter F and its dual I. We shall dene a forcing P = P(F ). A partial ordering (T, T ), where T , will be called a tree, if for each i T the set of predecessors {j T : j T i} is linearly ordered and i T j implies i j, for all i, j T . We dene a partial ordering for trees T t S i (S, S ) is a subordering of (T, T ) and each branch of T contains conally a (unique) branch of S. There is a tree T0 such that T0 I and T0 is order isomorphic to the full binary tree of height . (n) Deleting the numbers n from T0 we obtain a subtree denoted by T0 (we have (n) (m) T0 t T0 , for n m). Let T consist of all the trees T I such that T t T0 , for some n . Note that each tree T T has nitely many roots. Definition. Elements of the forcing P are of the form p =< Tp , fp >, where Tp T and fp : Tp {0, 1}. The ordering of P is dened thus p q i Tp t Tq and fp fq .
(n)

modified:1993-08-27

Let {b : < c} be a xed enumeration of all the branches of T0 in V . For a generic G P let TG = pG Tp and fG = pG fp . Each branch B of TG contains conally a unique b . Let us write B = B and dene X = {i : i B and fG (i) = 1} Since Tp I, for any p P, hence \ Tp A is innite, for each A F . It follows that the sets
A Dn = {p P : i > n i bp and fp (i) = |}

revision:1993-08-27

are dense, for each A F , n , < c and = 0, 1 (here bp denotes the branch of Tp extending b ). Thus, P adds uncountably many almost disjoint Gregorie-like sets. Sec.2. Let Q = Q(F ) be a countable product of P = P(F ). Thus the elements q Q can be written in the form
q q q =< f0 , f1 , . . . >, where < dm(fiq ), fiq > P, for each i < .

By q (n) we denote the condition < gi : i < > where 2

396

gi =

fiq | dm(fiq )(n) , fiq

for i < n for i n

Here T (n) is a tree obtained from T by deleting the numbers n. Lemma 2.1.For each decreasing sequence p0 p1 . . . there is a q and an innite Z such that q pn , for each n Z. Proof. Let Tni = dm(fipn ), where pn =< fipn : i < >. Since min Tni n, we may use (1.2) to dene by induction a descending sequence Z0 Z1 . . . of innite subsets of such that
nZi (n) (n)

Tni is in I, for each i < .

(n)

There is an innite Z , such that Z Zi , for each i < . Dene Ti = Tii

nZ

Tni

(n)

and fiq = fipi

nZ
modified:1993-08-27

fipn | Tni .
QED

(n)

Then,

dm(fiq )

= Ti and q =<

fiq

: i < > is as required.

For q Q and n let S(q, n) be the set of all sequences s =< s0 , . . . , sn1 > satisfying the following properties 1. s0 , . . . , sn1 are nite zero-one functions. 2. The domains t0 = dm(s0 ), . . . , tn1 = dm(sn1 ) are nite trees such that t0 T 0
(n)

= . . . = tn1 Tn1 = ,

(n)

revision:1993-08-27

q q where T0 = dm(f0 ), . . . , Tn1 = dm(fn1 ) (n) (n) 3. Ordered sums t0 T0 , . . . , tn1 Tn1 are trees belonging to T . Note that from the denition of T it follows that S(q, n) is always nite. Let us denote q q q s q (n) =< s0 f0 , . . . , sn1 fn1 , fn , . . . >

for q, n, s as above. Obviously, we have (2.2) the set {s q (n) : s S(q, n)} is predense below q (n)
sS(q,n) s

(i.e. the boolean sum

q (n) = q (n) ). 3

396

Now, we obtain easily an analogue of VI, 4.5 in S. For arbitrary p Q, n < and V (Q) such that Q is an ordinal there is a q p and ordinals {(s) : s S(p, n)} so that q (n)
s

(2.3)

= (s)

Indeed, if S(q, n) = {s0 , . . . , sm1 }, then we dene inductively conditions p0 , . . . , pm (n) so that p0 = p and pk+1 sk pk is such that pk+1
(n)

= , for some ordinal = (sk )

Now, q = s pm , where s is such that p = s p(k) (we may assume s S(p, n)), satises (2.3). 2.4. Theorem.Q is -proper, for every < 1 , and has the strong P P -property. Proof. Let countable N H() for suciently large , be such that Q N and suppose that p Q N . To prove that Q is proper we have to nd a q p, which is N -generic. Let {n : n < } be an enumeration of all the Q-names for ordinals, such that n N , for n < . Using (2.3) we dene inductively a sequence p0 = p p1 . . . and ordinals (n, s) so that
modified:1993-08-27

p(n) n

in s

i = (n, s) for each n <

(i.e. in the n-th step we apply (2.3) for all names 0 , . . . , n ). Note that the p s and s can be found in N , since N H(). By Lemma 2.1 there is a q and an innite Z such that q pn , for each n Z. Hence also q (m) pn
(n) (n)

holds for arbitrarily large n and all m < and thus q (m) n N ,

revision:1993-08-27

for all n, m < . By III, 2.6 of S, each q (m) is N -generic. To see that Q is -proper let < N : > be a continuous sequence of elementary countable submodels of H() such that Q N0 and < N : > N+1 , for each < . 4

396

Assume that Q is -proper, for each < and let q0 Q N0 . If = + 1, we have a q q0 which is N -generic, for each and we may assume that the q (n) have the same property, for all n < . since N H() and all the parameters are in N , such a q can be found in N and as above we construct a q q which is N -generic and so are (n) (n) the q , for n < . Thus, q and all the q are N -generic for all . If is a limit ordinal, we x an increasing sequence < n : n < > such that = supn< n and by the inductive hypothesis there is a sequence q0 q0 q1 . . . such that, for each n < , (m) qn is N -generic, for each n and qn Nn +1 and that qn have the same property for each m < . By Lemma 2.1 there is a q Q such that q qn , for innitely many n < . Thus, q q0 and q is N -generic for each < and hence also for each . Finally, to prove the P P -property let h : diverge to innity and suppose that p f : . Dene kn = min{i : h(i) > 2n cardS(p, n)}, for n < and, using (2.3), dene inductively the sequence p = p0 p1 . . . such that p(n) n
i<kn sS(pi ,i) (n)

f (i) = (s, i) for each n < and some integers (s, i) < .

Let T be the tree built up of integers {(s, i) : i < and s S(pi , i)}
modified:1993-08-27

. (n) If q pn , for innitely many n, then we have q f Lim T and T kn has less elements than h(kn ), for all n < , which nishes the proof. QED The last point to be discussed is how does Q = Q(F ) act in the course of iteration. 2.5.Lemma.If R is -bounding (i.e. the set of old functions : dominates) and Q(F ) is a complete subforcing of R, then in V (R) the lter F cannot be extended to a ccc P -lter. n Proof. Let X be the -th set added by n-th factor of the product Q = P . Suppose that for some r R and a ccc P -lter E V (R) we have r F E

revision:1993-08-27

n Note that for each n < , the relation X E hold for at most countably many s, n since E is ccc. Hence, there is an such that for all n < we have \ X E and, since n E is a P -lter, there is an A E and a function g, so that A n< ( \ X ) 0, g(n)) i.e. for some r1 r we have

(2.6)

r1

n<

n ( \ X ) 0, g(n)) E

396

Since R is -bounding we may assume that g V . By the assumption, Q is a complete subforcing of R and hence there is a q Q such that r is compatible with each q q. q On the other hand, since Tn = dm(fn ) T , there is a set B I and an increasing sequence a0 < a1 < . . . such that Tn \ 0, an ) B, g(n) < an and an , an+1 ) \ B = , for each n < . Dene q q as follows. For a given n extend Tn by adding elements q of an , an+1 ) \ B on the -th branch bq and put fn (i) = 1, for each i an , an+1 ) \ B. Obviously, we have q and hence q Consequently q
n< n ( \ X ) 0, g(n)) \ B n< n< ( n \ X ) 0, g(n)) B, which contradicts (2.6). n ( \ X ) 0, g(n)) an , an+1 ) \ B = , for each n

an , an+1 ) =
QED

The rest of the proof is routine. Beginning with a model V of 2 = 1 and 21 = 2 we iterate with countable supports the forcings Q(F ), for all ccc P -lters F booked at each stage < 2 of the iteration. From S, V.4 we know that the resulting forcing R (obtained after 2 stages) is proper and -bounding. Hence, in V G| there are no ccc P -sets. References F-Z R.Frankiewicz, P.Zbierski Strongly discrete subsets in Fund. Math. 129 (1988) pp 173-180 MA.H.Mekler Finitely additive measures on N and the additive property Proc.AMS Vol.92 No3 Nov.1984 pp 439-444 vM-R J.vanMill, G.M.Reed editors Open problems in topology, North Holland Elsevier Science Publishers B.V 1990 S S.Shelah Proper forcing Lecture Notes in Mathematics 940 Springer Verlag Ryszard Frankiewicz Institute of Mathematics, Polish Academy of Sciences, Warsaw, Poland Saharon Shelah Institute of Mathematics, The Hebrew University, Jerusalem, Israel PawelZbierski Institute of Mathematics, Warsaw University, Warsaw, Poland

revision:1993-08-27

modified:1993-08-27

396