Examples for Souslin Forcing

Haim Judah Dept. of Mathematics and Computer Science Bar-Ilan University 52900 Ramat-Gan, Israel Andrzej Roslanowski Dept. of Mathematics and Computer Science Bar-Ilan University 52900 Ramat-Gan, Israel and Mathematical Institute of Wroclaw University 50384 Wroclaw, Poland Saharon Shelah Institute of Mathematics The Hebrew University of Jerusalem Jerusalem, Israel and Department of Mathematics Rutgers University Rutgers, USA October 6, 2003
modified:1996-03-17

revision:1996-03-17

The authors would like to thank Israel Academy of Sciences BRF and MSRI for partial support.

373

Introduction

In this paper we continue with study of forcing notions having a simple denition. We began this study in [JS1] and [JS2]. In [BJ] we gave more results about Souslin forcing notions and in this paper we will give some examples of Souslin forcing notions answering a question of [JS1] and a question of H.Woodin. A forcing notion P is Souslin if P R is a 1 -set, {(p, q) : p P q} is a 1 1 -set and {(p, q) : p is incompatible with q} is a 1 -set. 1 1 More information on Souslin forcing notion can be found in [JS1]. A related work is [BJ]. In [JS1] we prove that if P is Souslin ccc and Q is any forcing notion then VQ |=P satises ccc. A natural question was: does P is Souslin ccc imply P has Knaster property. Recall that P satises Knaster property if and only if (A [P]1 )(B [A]1 )(p, q B)(p is compatible with q). In the second section we will give a model where there is a ccc Souslin forcing which does not satisfy the Knaster condition. Recall that under the assumption of MA every ccc notion of forcing has Knaster property. Many simple forcing notions P satisfy the following condition:
modified:1996-03-17

P is -centered.

revision:1996-03-17

This property is connected with the homogeneity of the forcing notion. The example of a totally nonhomogeneous Souslin forcing will be constructed in the third section. In the next section we present a model where there is a -linked not centered Souslin forcing such that all its small subsets are -centered but Martin Axiom fails for this order. In section 5 we will give an example of a -centered Souslin forcing notion and a model of the negation of CH in which the union of less than continuum meager subsets of R is meager but Martin Axiom fails for this notion of forcing. In the last session of the MSRI Workshop on the continuum (October 1989) H.Woodin asked if P has a simple denition and does not satisfy cccimplies that there exists a perfect set of mutually incompatible conditions. Clearly Mathias forcing satises such a requirement. In section 6 we 2

373

will nd a Souslin forcing which is Proper but not ccc that does not contain a perfect set of mutually incompatible conditions. The last section will show that ccc 1 -notions of forcing may not be 2 indestructible ccc. Our notation is standard and derived from [Je]. There is one exception, however. We write p q to say that q is a stronger condition then p.

On the Knaster condition

In this section we will build a Souslin forcing satisfying the countable chain condition but which fails the Knaster condition. Fix a sequence < i : i > of functions from into such that (*) if N < , i : N (for i < N ) then there are distinct n0 , . . . , nN 1 such that (i, j0 , j1 < N )(i (j0 ) = j1 i (nj0 ) = nj1 ). Note that there exists a sequence < i : i > satisfying (*):
modified:1996-03-17

Suppose we have dened i |m0 : m0 m0 for i < m0 . We want to ensure (*) for n0 + 1, i (i n0 ). Dene i (m0 + j0 ) = m0 +i (j0 ) for i, j0 n0 . Take large m1 and extend all i (i n0 ) on m1 in such a way that rng(i ) m1 . Next we dene functions fi : for i by fi (x)(k) = x(k) if k < i i (x(k)) otherwise

revision:1996-03-17

Clearly all functions fi are continuous. Put F (x) = {fi (x) : i } for x . Lemma 2.1 Suppose that x , y are such that there is no repetition in {x , y : 1 }. Then there exists A [1 ]1 such that (, A)( < x F (y )).

373

Proof For < 1 let n = min{n : x (n) = y (n)}. We nd n , and s, t n such that the set A0 = { < 1 : n = n + 1 & x |n = s & y |n = t} is stationary in 1 . Clearly s = t and s|(n 1) = t|(n 1). Thus , A0 and x F (y ) imply x {fi (y ) : i n}. Consequently the set { A0 : x F (y )} is nite for each A0 . We dene the regresive function : A0 1 by () = max{ A0 : x F (y )} (with the convention that max = 0). By Fodors lemma there are < 1 and a stationary set A1 A0 such that () = for all A1 . Put A = A1 \( + 1). Now, if , A, < then () < and hence x F (y ). Lemma 2.2 Suppose that {W : < 1 } is a family of disjoint nite . Then there exist < and an innite set A such that subsets of 1 ( A)(x W )(y W )(x F (y)). Proof We may assume that all sets W are of the same cardinality, say |W | = n for < 1 . For = + k, where < 1 is a limit ordinal and k we dene X = W+2k and Y = W+2k+1 . Let X = {x : i < n}, i Y = {yi : i < n}. Choose by the induction on l = l1 n + l2 < n2 , l1 , l2 < n uncountable sets Al 1 satisfying
modified:1996-03-17

Al+1 Al and if l = l1 n + l2 , l1 , l2 < n, , Al and < then x F (yl ). l1 2 Since there is no repetition in {x , yl : Al1 } we may apply lemma 2.1 l1 2 to get Al from Al1 . Consider An2 1 . Choose 0 An2 1 such that the set A = { + 2k < 0 : + k An2 1 & k & is a limit ordinal }

revision:1996-03-17

is innite. Let 0 = 0 +k0 where k0 and 0 is limit. Put = 0 +2k0 +1. Since 0 < we have A . Suppose = + 2k A. Let x W , y W . Then + k An2 , W = X+k and W = Y0 +k0 = Y0 . Thus for 1 some l1 , l2 < n we have x = x+k and y = yl0 . Since + k, 0 Al1 n+l2 , l1 2 + k < 0 we get x F (y). The lemma is proved. Let relations R on < be dened by
i

373

sRi t if and only if i < |s| = |t|, s|i = t|i and (l [i, |s|))(s(l) = i (t(l))). Note that if x, y are such that (n > i)(x|nRi y|n) then x = fi (y). We dene the following forcing notion Q. A member q of Q is a nite function such that: ) dom(q) [1 ]< , rng(q) < , ) (, dom(q))( = q() = q()), ) there is n(q) such that q() n(q) for all dom(q). The order is dened as follows: q p if and only if 1. dom(q) dom(p) and 2. ( dom(q))(q() p()) and 3. if , dom(q), < , i < n(q) and q()Ri q() then p()Ri p().
modified:1996-03-17

Lemma 2.3

Q satises ccc.

Proof Suppose {q : < 1 } Q. We nd < 1 and A [1 ]1 such that for each , A, < we have n(q ) = n(q ), dom(q ) = dom(q ) , (dom(q )\) (dom(q )\) = , q |(dom(q ) ) = q |(dom(q ) ).

revision:1996-03-17

Suppose , A. Clearly q = q q is a function. The only problem is that there may exist 0 dom(q ) and 1 dom(q ) such that q (0 ) = q (1 ). Therefore to get a condition above both q and q we have to extend all q (). Let dom() = {j : j < N } be an increasing enumeration. For i < n(q ) q choose i : N such that

373

if j1 < j0 < N and q (j1 )Ri q (j0 ) and either j0 , j1 dom(q ) or j0 , j1 dom(q ) then i (j0 ) = j1 . Note that q (j1 )Ri q (j0 ), q (j2 )Ri q (j0 ) imply q (j1 ) = q (j2 ). Hence if j2 < j1 < j0 are as above then j2 and consequently only one pair (j1 , j0 ) or (j2 , j0 ) will be considered in the denition of i . Apply condition (*) to nd distinct n0 , . . . , nN 1 such that (i < n)(q )(j0 , j1 < N )(i (j0 ) = j1 i (nj0 ) = nj1 ). Put q(j ) = q (j )nj for j < N . Clearly q Q. Suppose j1 , j0 dom(q ), j1 < j0 and q (j1 )Ri q (j0 ) for some i < n(q ). Then i (j0 ) = j1 and hence i (nj0 ) = nj1 . Thus q(j1 )Ri q(j0 ). It shows that q q. Similarly q q. Thus we have proved that Q satises Knaster condition. Let G Q be generic over V. In V[G] we dene xG = {q() : q G & dom(q)} for < 1 . Obviously each xG is a sequence of integers. As in the proof of lemma 2.3 we can show that for each q Q there is p q such that n(p) = n(q) + 1. Consequently xG for every < 1 . Moreover xG = xG for < < 1 (recall that q() = q() for distinct , dom(q)). Note that if < , , dom(q), q Q and i < n(q) then
modified:1996-03-17

q()Ri q() implies q x = fi (x ) and q()Ri q() implies q x = fi (x ). Lemma 2.4 Suppose G Q is generic over V. Then

V[G] |= (A [1 ]1 )(, A)( < & xG F (xG )). Proof Let A be a Q-name for an uncountable subset of 1 . Given p Q. Find A0 [1 ]1 and q p for A0 such that dom(q ) and q A. We may assume that for each , A0 we have n = n(q ) = n(q ) and q () = q (). Now we repeat the procedure of lemma 2.3 with one small change. We choose suitable A1 [A0 ]1 , < 1 and we take , A1 , < . Dening integers n0 , . . . , nN 1 we consider functions i : N (for i < n) as in 2.3 and a function n : N such that n (k) = l, where = l , = k . Then we get a condition q Q above both q and q and such that n (q()(n)) = q()(n). Since q()|n = q()|n and n(q) = n + 1 6

373

revision:1996-03-17

we have q()Rn q() and consequently q get q , A & x F (x ).

x = fn (x ). Since q q , q we

Fix a Borel isomorphism (0 , 1 , 2 ) : ( ) 2 . Thus if x then 1 (x) is a relation on and 0 (x) is a sequence of reals. Let consists of all reals x such that 1. (n = m)(0 (n) = 0 (m)) 2. 1 (x) is a linear order on 3. 2 (x) Ax = {0 (n) : n } and it is the 1 (x)-last element of Ax . Note that in 3 we think of 1 (x) as an order on Ax . We dene relations < and on by x < y if and only if Ax is a proper 1 (y)-initial segment of Ay and 1 (y)|Ax = 1 (x). x y if and only if Ax = Ay and 1 (y) = 1 (x) (we treat 1 (x), 1 (y) as orders on Ax , Ay , respectively). Clearly is a Borel subset of , < is a Borel transitive relation on and is a Borel equivalence relation on . Now we dene a forcing notion P1 . Conditions in P1 are nite subsets p of such that if x, y p, x < y then 2 (x) F (2 (y)). P1 is ordered by the inclusion. Lemma 2.5 P1 is a ccc Souslin forcing.

modified:1996-03-17

revision:1996-03-17

Proof P1 is Souslin since it can be easily coded as a Borel subset in such a way that the order is Borel too. We have to show that P of 1 satises the countable chain condition. First let us note some properties of the incompability in P1 . Suppose p, q P1 are incompatible. Clearly p\q and q\p are incompatible. If x p and x x then (p\{x}) {x } and q are incompatible. Suppose now that {p : < 1 } is an antichain in P1 . By the -lemma and by the above remarks we may assume that 7

373

1) p p = for < < 1 2) if x, x

<1

p , x = x then x x .

Note that if p P1 then the set {[y] : y & (x p)(y < x)} is countable. Hence, due to 2, we may assume that 3) ( < < 1 )(x p )(y p )(y < x) Claim: Let d [ ]< . Then d = {2 (x) : x p } for at most countably many < 1 . Indeed, assume not. Then we nd < 1 such that {2 (x) : x p } = d and the set B = { < : {2 (x) : x p } = d} is innite. Note that if x , x < x and 2 (x ) = 2 (x ) then x x . Hence if x p then for at most |d| elements x of A p we have x < x. Thus we nd A such that (x p )(x p )(x < x). It follows from 3 that (x p )(x p )(x < x ) and hence conditions p and p are compatible a contradiction. Let d = {2 (x) : x p }. By the above claim we may assume that d = d for all < < 1 . Applying the -lemma we may assume that
modified:1996-03-17

4) {d : < 1 } forms a -system with the root d. Since the set

wd

F (w) is countable w.l.o.g.

wd

5) ( < 1 )(v d \d)(v

F (w)).

Apply lemma 2.2 for the family {d \d : < 1 } to get < 1 and an innite set A such that 6) ( A)(v d \d)(w d \d)(v F (w)).

revision:1996-03-17

Let y p . As in the claim the set {x

A

p : 2 (x) d & x < y}

is nite. Consequently we nd A such that 7) (x p )(y p )(2 (x) d x < y). 8

373

We claim that p and p are compatible. Let x p and y p . By 7 we have that if 2 (x) d then x < y. If 2 (x) d and 2 (y) d then 6 applies and we get 2 (x) F (2 (y)). Finally if 2 (x) d and 2 (y) d then we use 5 to conclude that 2 (x) F (2 (y)). Hence x p , y p and x < y imply 2 (x) F (2 (y)). Consequently p p P1 . Lemma 2.6 Assume that there exists a sequence {x : < 1 } of ele such that ments of (A [1 ]1 )(, A)( < & x F (x )). Then the forcing notion P1 does not satisfy the Knaster condition. Proof For < 1 choose y such that

Ay = {0 (y )(n) : n } = {x : } 1 (y ) is the natural order on Ay , x <1 (y ) x i < . 2 (y ) = x . Let p = {y } for < 1 . Then {p : < 1 } does not have an uncountable subset of pairwise compatible elements. Putting together lemmas 2.5, 2.6 and 2.4 we get Theorem 2.7 It is consistent that there exists a ccc Souslin forcing notion which does not satisfy the Knaster condition. It is not dicult to see that this example does not satisfy the following requirement: The generic object is encoded by a real

revision:1996-03-17

modified:1996-03-17

The next theorem says that also we can require such a condition. This answers a question of J.Bagaria. Theorem 2.8 It is consistent that there exists a ccc Souslin forcing notion Q such that Q V[G] = V[r] for some Q-name r for a real and Q does not satisfy the Knaster condition. 9

373

Proof We follow the notation of the previous results. We work in the model of 2.7. Let Q = {(p, w) : p P & w [ < ]< } be ordered by (p, w) (q, v) if and only if p q, w v and x|n v\w for every x p and n . Q may be easily represented as a Souslin forcing notion (remember that for each x in p is a quantication on natural numbers). Note that if p q, p, q P and w [ < ]< then (p, w) (q, w). Hence Q satises the countable chain condition. If Q satised the Knaster condition then P would have satised it. We show that the Q-generic object is encoded by a real. Let r be a Q-name for a subset of < (a real) such that for any Q-generic G we have r G = {w : (p)((p, w) G)}. Now in V[r G ] dene H = {(p, w) Q : w r G & (x p)(n )(x|n r G x|n w)}. Note that H includes G since x p, x|n w imply (p, w) x|n r. H is a lter - suppose (p0 , w0 ), (p1 , w1 ) H. For each x p0 p1 we nd (px , wx ) G such that (px , wx ) (n N )(x|n r) for some N . If x px we could take large n and add x|n to wx . Then we would have (px , wx ) (px , wx {x|n}) and (px , wx {x|n}) x|n r. Thus x px for all x p0 p1 . Let p = xp0 p1 px , w = xp0 p1 wx . Then (p, w) G H and (p0 , w0 ), (p1 , w1 ) (p, w). Consequently H = G and the theorem is proved. In the same time when the forcing notion P1 was constructed S.Todorcevic found another example of this kind. Let F be the family of all converging sequences s of real numbers such that lim s s. Todorcevics forcing notion P consists of nite subsets p of / 1 F with property that (s, t p)(s = t lim s t). / Todorcevic proved that P satises ccc and that if b = 1 then P does not 1 1 have Knaster property (see [To]).

revision:1996-03-17

modified:1996-03-17

A nonhomogeneous example

In this section we give an example of ccc Souslin forcing notion which is very nonhomogeneous. Our forcing P2 will satisfy the following property: 10

373

there is no p P2 such that p

P2 |p is -centered.

Recall that if Q is the Amoeba Algebra for Measure or the Measure Algebra then Q Q is -centered (see [BJ]). The Todorcevic example P has this 1 property too. Proposition 3.1
P 1

1 P is -centered

Proof For a rational number d Q let d : P P be the trans1 1 lation by d. Thus d (p) = {s + d : s p}. Note that d is an automorphism of P . Moreover if p1 , p2 P then {s r : s p1 , r p2 } is a nowhere 1 1 dense set. Hence we nd a rational d such that d {a b : a s{lim s}, b r{lim r}, s p1 , r p2 }. / Then the conditions d (p1 ) and p2 are compatible. Thus we have proved that for each p P the set {d (p) : d Q} is predense in P . This implies that 1 1
P 1

P = 1
dQ

d [] and each d [] is centered,

where is the canonical name for a generic lter.

modified:1996-03-17

We do not know if
P1

P1 is -centered.

One can easily construct a ccc Souslin forcing P which does not force that P is -centered. An example of such a forcing notion is the disjoint union of Cohen forcing and the measure algebra, P = ({0} C) ({1} B). In this order we have (0, ) {1} B is not -centered. But in this example we can nd a dense set of conditions p P such that p
P

revision:1996-03-17

P|p = {q P : q p} is -centered.

Dene T < , f, g : T in such a way that: () T is a tree, () if T then succT () = f (), 11

373

() if lh < lh or lh = lh but (k < lh)(|k = |k & (k) < (k)) then f () < f (), () g() > |T lh | ( ) f () > g() {f () : f () < f ()} (100 + lh) ,

{2f () : f () < f ()}

For T and and a set A succT () = f () we dene a norm of A: nor (A) = g() . |f ()\A|

Lemma 3.2 Suppose that T and Al f () for l < m. Let = min{nor (Al ) : l < m}. Then 1) nor ( l<m Al ) /m, 2) if 1 and m {2f () : f () < f ()} then l<m Al = . Proof 1) Note that Al | = |
l<m l<m

|f ()\ Hence
modified:1996-03-17

f ()\Al |
l<m

|f ()\Al | m g()/

nor (
l<m

Al ) =
l<m

g() /m. |f ()\ l<m Al | Al ) 1/m. Hence {2f () : f () < f ()} < f ()

2) Applying 1) we get nor ( |f ()\

l<m

Al | g() m g()

(the last inequality is guaranteed by condition ( )). Consequently the set l<m Al is nonempty. Let P2 consists of all trees T T such that lim min{nor (succT ()) : T n } = .

revision:1996-03-17

The order is the inclusion. Recall that a forcing notion Q is -k-linked if there exist sets Rn Q (for n ) such that n Rn = Q and each Rn is k-linked (i.e. any k members of Rn has a common upper boud in Q). 12

373

Proposition 3.3 Proof

For every k < the forcing notion P2 is -k-linked.

Let n be such that for each T n k< {2f () : f () < f ()}.

Note that the set {T P2 : lh(rootT ) n & ( T )(rootT nor (succT ()) 1)} is dense in P2 . For T , lh n dene D = {T P2 : rootT = & (nu T )( nor (succT ()) 1)}. Since {D : lh n} is dense in P2 it is enough to show that each D is k-linked. Suppose T0 , . . . , Tk1 D . Since k < {2f () : f () < f ()} we may apply lemma 3.2 2) to conclude that if T = T0 . . . Tk1 , then succT () = . By 3.2 1) we get T P2 . For T we dene the forcing notion Q : Q = {t T : t is a nite tree of the height n , roott = and ( t < n )( nor (succt ()) lh)}. Since Q is countable and atomless it is isomorphic to Cohen forcing C. Let P = {Qi, : i < 1 , T } be the nite support product such that each Qi, is a copy of Q . Theorem 3.4 Let G P be a generic lter over V. Then, in V[G], there is no S P2 such that S
P2

modified:1996-03-17

P2 |S is -centered.

revision:1996-03-17

Proof We work in V[G]. Assume S P2 |S is -centered. Let Rn (n ) be P2 -names for subsets of P2 such that S
P2

P2 |S
n

Rn & each Rn is directed. 13

373

Take n such that ( S)(n lh nor (succS ()) 1). Fix any S n and choose l, m succS (), l < m. For i < 1 put Ti = {t : {((i, m), t)} G}.

Each Ti is the tree added by G Qi,m and it is an element of P2 . Moreover rootTi = m and for each Ti if m then nor (succTi ()) lh. Hence, by lemma 3.2, Ti S P2 for each i 1 . Now we work in V. We nd pi , Si , i , ni such that for each i 1 : pi P, ni , i T and Si is a P-name for a member of P2 , pi
P

( Si )(rootSi nor (succSi ()) 1),

P

l i = rootSi & Si

P2

Ti Rni ,

(i, m) dompi . Next we nd a set I [1 ]1 such that {dompi : i I} forms a -system with the root d and for each i I:
modified:1996-03-17

i = and ni = n , pi |d = p , pi (i, m) = t, (i, m) d. / Let n# be the height of the tree t. Clearly we may assume that n# > lh . # Fix an enumeration {k : k < k # } of t n . Put H = {(ak : k < k # ) : ak f (k ) & nork (ak ) n# }.

revision:1996-03-17

Choose distinct ia I for a H. We dene a condition q P extending all pia ( H): a domq = {dompia : a H}; if (i, ) dompia , (i, ) = (ia , m) then q(i, ) = pia (i, ); q(ia , m) = t {kc : k < k # , c a(k)}. 14

373

 Now we take r q such that r decides all Sia |(n# + 1). Thus we have nite ia |(n# + 1) = sa . trees sa (for a H) such that r P S Claim: : There exists H H such that n# + 1 ) = and (i) aH (sa (ii) for each k < k # the set {(k) : a H } is empty. a # Indeed, let H = { H : s } for T n + 1 , l . Clearly a
a aH sa , so it is enough to show that for some the family H satises # (ii). Suppose that for each T n + 1 , l we can nd k < k # and c such that c {(k ) : a H }. Put a # a (k) = f (k )\{c : T n + 1 & l }.

Let + T n be such that f (+ ) = max{f () : T n , l }. By condition () we get

# |{ T n + 1 : l }| #

{f () : f () f (+ )}.

modified:1996-03-17

Now, for each k < k # we have f (+ ) < f (k ) (recall that l + , m k and l < m so condition () works). Hence nork ( (k)) a g(k ) {f () : f () f (+ )} g(k ) > n# {f () : f () < f (k )}

revision:1996-03-17

# / Thus a H. Since c a (k ) we have a H for every T n + 1 , / # l . Since {H : T n + 1 , l } = H we get a contradiction. The claim is proved. Now let H H be a family given by the claim. Condition (ii) implies that

the family {Tia : a H } has no upper bound in P2 .

# {2f (k ) : k < k # } we have that for each T n + 1

Since |H|

|H |

{2f () : f () < f ()}. 15

373

Hence we may apply 3.2 2) to conclude that for every l , lh n# + 1: r Thus Since r
P P

if
aH

Sia then
aH

succSi () = .
a

the family {Sia : a H } has an upper bound. Sia P2 Tia Rn we get a contradiction.
P

Remark: 1) In the above theorem we worked in the model V[G] for technical reasons only. The assertion of the theorem can be proved in ZFC. 2) The forcing notion P2 is a special case of the forcing studied in [Sh1]. Problem 3.5 Does there exist a ccc Souslin forcing P such that 1) P is homegeneous (i.e. for each p P, P there exists a generic lter G over V such that p G) 2) P P is -centered?

4
modified:1996-03-17

On small subsets of P are -centered.

Our next example is connected with the following, still open, question: Problem 4.1 Assume that for each ccc Souslin forcing P every set Q [P]1 is -centered (in P). Does MA1 (Souslin) hold true? As an illustration of this subject let us recall a property of Random (Solovay) Algebra B (see [BaJ]): if every B [B]1 is -centered then the real line can not be covered by 1 null sets and consequently MA1 (B) holds true.

revision:1996-03-17

Our example shows that the above property of the algebra B does not extend for other forcing notions. Let P3 = {(n, T ) : n & T 2< is a tree & (t T 2n )(([Tt ]) > 0)}. The order is dened by 16

373

(n1 , T1 ) (n2 , T2 ) if and only if n1 n2 , T2 T1 and T1 |n1 = T2 |n1 . Lemma 4.2 P3 is a -linked Souslin forcing which is not -centered.

Proof Note that P3 there exists a perfect set of random reals over V. Hence P3 is not -centered. To show that it is -linked dene sets U (W, n, m) for n < m < and nite trees W 2 m : U (W, n, m) = {(n, T ) P3 : T |m = W & (t T 2n )(([Tt ]) > W (t)/2m+1 )} where W (t) = |{s W 2m : t s}| (for t W 2n ). Clearly each set U (W, n, m) is linked (i.e. each two members of it are compatible in P3 ) and P3 = {U (W, n, m) : n < m < & W 2 m }. Since obviously P3 is Souslin we are done. Let B() stand for Random Algebra for adding many random reals. This is the measure algebra of the space 2 . Theorem 4.3 Assume V |= CH. Let G B(2 ) be a generic set over V. Then, in V[G]
modified:1996-03-17

(i) Martin axiom fails for P3 but (ii) each Q [P3 ]1 is -centered (in P3 ). Proof Cichon proved that one random real does not produce a perfect set of random reals (see [BaJ]). Hence in V[G] there is no perfect set of random reals over V. Consequently the rst assertion is satised in V[G]. Since V[G] |= each B [B]1 is -centered in B (compare section 3) it is enough to show the following Claim: Suppose that each B [B]1 is -centered. Then every set 1 is -centered. Q [P3 ] Indeed, let Q [P3 ]1 . For n and t 2n put B(t, n) = {[Tt ] : (n, T ) Q & t T }. By our assumption we nd sets B(t, n, k) for k, n , t 2n such that B(t, n) = k B(t, n, k) and for each A1 , A2 B(t, n, k) the set A1 A2 17

373

revision:1996-03-17

is of positive measure. Now dene sets Q(n, W, ) for n , a nite tree W 2 n and a function : W 2n : Q(n, W, ) = {(n, T ) Q : T |n = W & (t T 2n )([Tt ] B(t, n, (t)))}. Note that if (n, T1 ), (n, T2 ) Q(n, W, ) then for each t W 2n the set [(T1 )t ] [(T2 )t ] is of positive measure. Consequently each Q(n, W, ) is linked and we are done.

A -centered example

In this section we dene a very simple -centered Souslin forcing notion. Next we show that in any generic extension of some model of CH via nite support iteration of the Dominating (Hechler) Algebra, Martin Axiom fails for this forcing notion. Consequently we get the consistency of the following sentence: any union of less than continuum meager sets is meager + CH + MA fails for some -centered Souslin forcing.
modified:1996-03-17

Our example P4 consists of all pairs (n, F ) such that n , F [2 ]< and all elements of the list {x|n : x F } are distinct. P4 is ordered by (n, F ) (n , F ) if and only if n n , F F and {x|n : x F } = {x|n : x F }. Lemma 5.1 P4 is a -centered Souslin forcing.

revision:1996-03-17

Proof Clearly P4 is Souslin (even Borel). To show that P4 is centered note that if {x|n : x F0 } = . . . = {x|n : x Fk } then the conditions (n, F0 ), . . . , (n, Fk ) are compatible (if m is large enough then (m, F0 . . . Fk ) is a witness for this). Now we want to dene the model we will start with. At the beginnig we work in L. Applying the technology of [Sh] we can construct a sequence (P : 1 ) of forcing notions such that for each , < 1 , 1 : (1) if < then P is a complete suborder of P , 18

373

 (2) there is > such that P+1 = P D , where D is the P -name for nite support, in length, iteration of Hechler forcing, (3) P satises ccc, (4) if is limit then P =lim< P , (5) P1 every projective set of reals has Baire property

(for details see also [JR]). Recall that Hechler forcing D consits of all pairs (n, f ) such that n , f . These pairs are ordered by (n, f ) (n , f ) if and only if n n , f |n = f |n and f (k) f (k) for all k . Suppose G P1 is a generic set over L. We work in L[G]. For distinct x, y 2 we dene h(x, y) = min{n : x(n) = y(n)}. Easy calculations show the following Lemma 5.2 Let b . Then the following conditions are equivalent: (i) there exists a Borel equivalence relation R on 2 with countable many equivalence classes such that {h(x, y) : x, y 2 L & x = y & R(x, y)} b,
modified:1996-03-17

(ii) there exists an equivalence relation R on 2 with countable many equivalence classes such that {h(x, y) : x, y 2 L & x = y & R(x, y)} b, (iii) there exist sets Yn 2 (for n ) such that L 2 n Yn and n {h(x, y) : x = y & x, y Yn } b, (iv) (f : 2< 2)(x 2 L)(m )(n > m)(n b f (x|n) = x(n)). / The Raisonnier lter F consists of all sets b satisfying one of the conditions of 5.2 (cf [Ra]). F is a proper lter on . Directly from (iv) of 5.2 one can see that F is a 1 -subset of 2 . Consequently it has Baire 3 property (recall that we are in L[G]). Theorem 5.3 (Talagrand, [Ta]) conditions are equivalent: For any proper lter F on the following

revision:1996-03-17

(i) F does not have Baire property, 19

373

(ii) for every increasing sequence (nk : k ) of integers there exists b F such that ( )(b [nk , nk+1 ) = ). k Applying the above theorem we can nd an increasing function r L[G] such that (in L[G]) (b F )( )(b [r(k), r(k + 1)) = ) k Let r be the P1 -name for r and let 0 < 1 be such that r is a P0 -name. Our basic model will be L[r] Theorem 5.4 Let be a regular cardinal. Let D be the nite support iteration of Hechler forcing of the length . Suppose H D is a generic set over L[r]. Then L[r][H] |=there is no P4 -generic over L[r]. Proof Assume not. Let H L[r][H] be a P4 -generic over L[r]. Put T = {F : (n )((n, F ) H )}. Then in L[r][H ] we have: (6) T is a closed subset of 2 ,
modified:1996-03-17

/ (7) ( )(x, y T )(x = y h(x, y) [r(k), r(k + 1))) and k (8) (x 2 L)(q Q)(q + x T ) (Q stands for the set of all sequences eventually equal 0, + denotes the addition modulo 2). Since both (7) and (8) are absolute (1 ) sentences they 2 are satised in L[r][H] too. Let T L[r] be a P -name for T . Since T is a closed subset of 2 we can think of T as a name for a real. Now we work in L[r]. Let p D L[r] be such that p T satises (6), (7) and (8).

revision:1996-03-17

By Souslin forcing properties (see 1 of [JS1]) we nd a (closed) countable set S such that: (9) T is a D |S-name, p D |S and (10) D |S is a complete suborder of D . 20

373

Since (6)-(8) are absolute we get (11) p

D |S T

satises (6), (7) and (8).

But D |S is isomorphic to nite support iteration of Hechler forcing of the countable length ( < 1 ). Thus we can treat T as a D -name and p as a condition in D . Then, in L[r] (12) p
D |S T

satises (6), (7) and (8).

By (2) we nd > 0 such that (13) P+1 = P D and (14) p (1, p) interpreted as a member of P+1 belongs to G. By Souslin forcing properties (12) holds true in L[G P ] and hence (15) L[G P+1 ] |=T G satises (6), (7) and (8) (we treat here T as a P+1 -name). Let b = {h(x, y) : x = y & x, y T G } < L[G]. By (15) and by Shoeneld absoluteness we have [] (16) L[G] |=T G satises (6), (7) and (8).
modified:1996-03-17

Since {h(x, y) : x = y & x, y T G } = {h(x, y) : x = y & x, y T G + q} we conclude that (17) L[G] |=sets T G + q (for q Q) witness that b F and (18) L[G] |= ( )(b [r(k), r(k + 1)) = ). k The last condition contradicts our choice of r. Since
D any

union of less than meager sets is meager we get

revision:1996-03-17

Corollary 5.5 The following theory is consistent: ZFC + CH + Martin Axiom fails for some -centered Souslin forcing + any union of less than continuum meager sets is meager.

21

373

On Souslin not ccc

In this section we will give a negative answer to the following question of Woodin: If P is a Souslin forcing notion which is not ccc then there exists a perfect set T P such that each distinct t1 , t2 T are incompatible. Recall that in the case of non-ccc partial orders we do not require Souslin forcings to satisfy the condition: the set {(p, q) : p is incompatible with q} is 1 . 1 Thus a forcing notion P is is Souslin not ccc if both P and P are analytic sets. The reason for this is that we want to cover in our denition various standard forcing notions with simple denitions for which incompatibility is not analytic (e.g. Laver forcing). Let Q be the following partially ordered set: W Q if W is a nite set of pairs (, ), < 1 such that if (1 , 1 ), (2 , 2 ) are in W , then 1 < 2 or 2 < 1 . Q is ordered by the inclusion.
modified:1996-03-17

It follows from [Je1] that Q is proper. Clearly |Q| = 1 . Next dene a forcing notion P5 . It consists of all r such that r codes a pair (E r , w r ) where 1. E r is a relation on such that (, E r ) |= ZFC and E r encodes all elements of {}. 2. w r and E r |=w r Q. We say that a one-to-one function f interprets E r1 in E r2 if there exists n such that rng(f ) = {k : E r2 (k, n)} and E r1 (l, k) E r2 (f (l), f (k)). If f interprets E r1 in E r2 then E r2 may discover that some of the ordinals of E r1 are not ordinals (i.e. not well-founded). Let w(r1 , r2 , f ) = w r1 {(, ) : are ordinals in E r2 }. Then, in E r2 , w(r1 , r2 , f ) is an r initial segment of w r1 and it is in QE 2 . Now we can dene the order on P5 : 22

373

revision:1996-03-17

r1 r2 if and only if r1 = r2 or there exists f which interprets E r1 in E r2 and such that (, E r2 ) |= w(r1 , r2 , f ) wr2 . Obviously both P5 and the order are 1 -sets. 1 For r P5 we dene W (r) as w r {(, ) : are well founded }. Note that W (r1 ) = W (r2 ) implies r1 and r2 are equivalent in P5 (i.e they have the same compatible elements of P5 ). Consequently Q may be densely embedded into the complete Boolean algebra determined by P5 . It follows from [Je1] that P is proper, it is Souslin and it does not satisfy the countable chain condition. Moreover, if 1 < 2 then P does not contain a perfect set of pairwise incompatible elements (recall |Q| = 1 ). An interesting question appears here: Suppose P is -proper and Souslin. Does there exists a perfect set of pairwise incompatible elements of P? The negative answer to this question is given by the following result.
modified:1996-03-17

Theorem 6.1 Assume 1 < cf(2 ). There exists an -proper Souslin not ccc forcing notion P with no perfect set of pairwise incompatible elements. 5 Proof dened by: Let 1 be additively indecomposable. Let Q be the order

W Q if and only if W is a countable set of pairs (, ), < 1 such that (1 , 1 ), (2 , 2 ) W 1 < 2 or 2 < 1 , {(, ) W : = } is nite, the order type of the set { : ()((, ) W )} is less than . Q is ordered by the inclusion. It follows from Chapter XVII, 3 of [Sh 2] that Q is -proper for each < 1 . 23

373

revision:1996-03-17

Now we can repeat the coding procedure that we applied to dene the forcing notion P5 . Thus we get the Souslin forcing notion P such that Q 5 can be densely embedded in the Boolean algebra determined by P . 5 For W Q let heart(W ) = {(, ) W : = }. Assume that {(E r , w r : 2 } P is a perfect set of pairwise 5 incompatible elements. Let W be the well-founded part of w r . Since w.l.o.g we can assume that sup{ : ()((, ) W )} is constant and heart(W ) is constant we easily get a contradiction.

On ccc 1 2
Suppose P is a ccc Souslin notion of forcing. Let Q be a ccc forcing notion. Then Q P is ccc .

Souslin ccc notions of forcing are indestructible ccc (see [JS1]):

The above property does not hold true for more complicated forcing notions. In this section we show that there may exist two ccc 1 -notions of forcing 2 P6 and P such that P6 P does not satisfy ccc. 6 6 We start with V = L. Let Q be a ccc notion of forcing such that
modified:1996-03-17

MA + CH

Let G Q be a generic set over L and let r be a random real over L[G]. Recall that by theorem of Roitman (cf [Ro]) we have L[G][r] |= MA(-centered). 11 Fix a sequence (f : < 1 ) L of one-to-one functions f : and dene in L[r] sets E1 , E2 by Ei = {{, } [1 ]2 : < & r(f ()) = i} for i = 0, 1.

revision:1996-03-17

We dene forcing notions P6 , P : 6 P6 = {H [1 ]2 : [H]2 E0 }, P = {H [1 ]2 : [H]2 E1 }. 6 Orders are inclusions. 24

373

Both P6 and P are elements of L[r]. Moreover they can be thought of as 6 subsets of L[r] 2 . Applying MA(-centered) we get that (cf [Je]): L[G][r] |= any subset of L[r] 2 is a relative 0 -set. 2 Consequently L[G][r] |= any subset of L[r] 2 is 1 . 2 Thus P6 and P are 1 -notions of forcing in L[G][r] (i.e. both P6 , P and 6 2 6 orders and the relations of incompatibility are 1 -sets). Roitman proved the 2 following Theorem 7.1 (Roitman, Prop.4.6 of [Ro]) In L[G][r] both P6 and P sat6 isfy ccc and P6 P does not satisfy ccc. 6 Corollary 7.2 The following theory is consistent: ZFC + MA(-centered) + CH + there exist ccc 1 -notions of forcing 2 P6 , P such that P P6 is not ccc. 6 6 Problem 7.3 Is there a ccc Souslin forcing notion P such that MA(P) always fails after adding a random real?

revision:1996-03-17

modified:1996-03-17

25

373

revision:1996-03-17

References [Ba] J.Baumgartner, Iterated forcing in Surveys in Set Theory, ed. by A.R.D.Mathias, London Math. Soc., Lecture Notes 87. [BJ] J.Bagaria, H.Judah, Amoeba forcing, Souslin absoluteness and additivity of measure, Proceedings of the 1989 MSRI Workshop on Set Theory of the Reals (to appear). [BaJ] T.Bartoszynski, H.Judah, Jumping with random reals, [Je] T.Jech, Set Theory, Academic Press, New York 1978. [Je1] T.Jech, Multiple Forcing, Cambridge Tracts in Mathematics 88, Cambridge 1986. [JS1] H.Judah, S.Shelah, Souslin forcing, Journal of Symbolic Logic, 53(1988). [JS2] H.Judah, S.Shelah, Martins axioms, measurability and equiconsistency results, Journal of Symbolic Logic, 54(1989). [JR] H.Judah, A.Roslanowski, On Shelahs amalgamation, Proceedings of the Winter Institute on Set Theory of the Reals, Bar-Ilan 1991 (to appear). [Ra] J.Raisonnier, A mathematical proof of Shelahs theorem, Israel Journal of Mathematics 48(1984). [Ro] J.Roitman, Adding a random or a Cohen real: topological consequences and the eect of Martins axiom, Fundamenta Mathematicae vol CIII (1979). [Sh] S.Shelah, Can you take Solovays inaccessible away?, Israel Journal of Mathematics, 48(1984). [Sh1] S.Shelah, ???Vive la Dierence I (new version of models with no isomorphic ultra powers, Proceedings of the Conference in Set Theory, MSRI 10/89, [Sh2] S.Shelah, Proper and Improper Forcing, in preparation. [Ta] M.Talagrand, Compacts de functions mesurables et ltres non mesurables, Studia Mathematica, 67(1980). [To] S.Todorcevic, Two examples of Borel partially ordered sets with the countable chain conditions, Proceedings of the American Mathematical Society, 112(1991).

modified:1996-03-17

26

373