A MODEL IN WHICH THERE ARE JECHKUNEN TREES BUT THERE ARE NO KUREPA TREES 1

Saharon Shelah
2

and Renling Jin

Abstract
By an 1 tree we mean a tree of power 1 and height 1 . We call an 1 tree a JechKunen tree if it has many branches for some strictly between 1 and 21 . In this paper we construct the models of CH plus 2 1 > 2 , in which there are JechKunen trees and there are no Kurepa trees.

An partially ordered set, or poset for short, T, <T is called a tree if for every t T the set {s T : s <T t} is wellordered under <T . The order type of that set is called the height of t in T , denoted by htT (t). We will not distinguish a tree from its base set. For every ordinal , let T , the th level of T , = {t T : htT (t) = } and T = < T . Let ht(T ), the height of T , is the smallest ordinal such that T = . By a branch of T we mean a linearly ordered subset of T which intersects every nonempty level of T . Let B(T ) be the set of all branches of T . T is called a subtree of T if T T and <T =<T T T (T inherits the order of T ). T is called an 1 tree if |T | = 1 and ht(T ) = 1 . An 1 tree T is called a Kurepa tree if |B(T )| > 1 and for every 1 , |T | < 1 . An 1 tree is called a JechKunen tree if 1 < |B(T )| < 21 . The independence of the existence of Kurepa trees was proved by J. H. Silver (see [K2, 3 of Chapter VIII]). T. Jech in [Je1] constructed by forcing a model of CH plus 21 > 2 , in which there is a JechKunen tree. In fact, it is a Kurepa tree with fewer than 21 many branches. The independence of the existence of JechKunen trees under CH plus 21 > 2 was given by K. Kunen [K1]. In his paper he gave an equivalent form of JechKunen trees in terms of compact Hausdor spaces. The detailed proof can be found in [Ju, Theorem 4.8]. In both Silver and Kunens proofs, the existence of a strongly inaccessible cardinal was assumed (the assumption is also necessary). The technique they used to kill all Kurepa trees or JechKunen trees is to show that if an 1 tree T has a new branch in an 1 closed forcing extension, then T must have a subtree which is isomorphic to 2<1 , , a complete binary tree of height 1 . So in Kunens model not only all JechKunen trees are killed, but also all Kurepa trees are killed. R. Jin in [Ji1] started discussing the dierences between Kurepa trees and Jech Kunen trees. He showed that it is independent of CH plus 21 > 2 that there
1980 Mathematics Subject Classication (1985 Revision). Primary 03E35. The research of the rst author was partially supported by the Basic Research Fund, Israeli Acad. of Science Publ. nu. 466.
2 1

revision:1993-08-29

modified:2002-07-16

466

exists a Kurepa tree which has no JechKunen subtrees. He also showed that it is independent of CH plus 21 > 2 that there exists a JechKunen tree which has no Kurepa subtrees. In his proofs some strongly inaccessible cardinals were assumed and later, Kunen eliminated the large cardinal assumption for one of the proofs. In [Ji2] Jin proved that assuming the existence of two inaccessible cardinals, it is consistent with CH plus 21 > 2 that there exist Kurepa trees and there are no JechKunen trees. The problem whether CH plus 21 > 2 is consistent with that there exist Jech Kunen trees and there are no Kurepa trees, was posed in [Ji2]. We will answer the question in this paper by assuming naturally the existence of a strongly inaccessible cardinal. Before proving our results we need more notations and denitions. A tree T is called normal if, (1) every t T has at least two immediate successors, (2) for every t T and an ordinal such that htT (t) < < ht(T ), there exists t T such that t <T t . A tree C = {cs : s 2< } is called a Cantor tree if the map s cs is an isomorphism from 2< , to C. For convenience we assume, from now on, that every tree considered in this paper is a subtree of 2<1 , with the unique root . By that way we can dene the least upper bound of an increasing sequence in a tree by taking its union. Let lim(1 ) be the set of all limit ordinals in 1 . Let T be a tree and lim(1 ). A subtree C of T is called conal in T if for every B B(C), the set {htT (t) : t B} is conal in . T is called complete at level if for every B B(T ), B T . T is called properly pruned at level if for every Cantor subtree C = {cs : s 2< } of T which is conal in T , there exist f, g 2 such that n cf n T and n cg n T Let S lim(1 ). A tree is called Sproperly pruned if for every lim(1 ), S implies that T is complete at level , and S implies that T is properly pruned at level . Let I be an index set and T be a tree. For every F T I , let supt(F ), the support of F , be the set {i I : F (i) = }. Let F, G T I . Dene F G i for every i I, F (i) G(i). We call F T I uniform at for some 1 if for every i supt(F ), htT (F (i)) = . Let C = {Fs : s 2< } T I be a Cantor tree (under ). C is called uniformly conal in (T )I for some 1 if for every s 2< , there is a s < such that Fs is uniform at s and for every i s2< supt(Fs ), the subtree {Fs (i) : s 2< } of T is conal in T . We use for the word incompatible. For example, for any s, t 2< , s t means s t is not a function. For any F, G T I , we call that F and G are completely incompatible if for any i supt(F ) and any j supt(G), F (i) G(j) (F (i) and G(j) have no common upper bound in T ). Now C is called separated if for any s, s 2< , s s implies that Fs and Fs are completely incompatible. 2

466

revision:1993-08-29

modified:2002-07-16

Let T be a tree and lim(1 ). We call that T is properly pruned in countable products at level if for every Cantor tree C = {Fs : s 2< } T I , which is separated and uniformly conal in (T ) , there exist f, g 2 such that for every i n supt(Ff n ), n Ff n (i) T and for every i n supt(Fg n ), n Fg n (i) T . Let S lim(1 ). A tree is called Sproperly pruned in countable products if for every lim(1 ), S implies that T is complete at level , and S implies that T is properly pruned in countable products at level . Lemma 1 Let T be a tree and I be an index set. For any Cantor tree C = {Fs : s 2< } T I , if C is separated, then for any f, g 2 , f = g implies that n Fg n (i) iI are completely incompatible. n Ff n (i) iI and Proof: Let i n supt(Ff n ) and j n supt(Fg n ). Let m such that i supt(Ff m ), j supt(Fg m ) and f m = g m. Then n Ff n (i) and n Fg n (j) are compatible implies that Ff m (i) and Fg m (j) are compatible, a contradiction. 2 Lemma 2 (CH). For any S lim(1 ), there exists a normal 1 tree which is S properly pruned in countable products. Proof: We construct T 2 recursively on < 1 and T = tree we want. Case 1. = + 1 for some 1 . Let T = {t l : t T , l = 0, 1}.
modified:2002-07-16

<1

T will be the

Case 2. lim(1 ) S. Let T = { B : B B(T )}. Case 3. S. Let C be the set of all Cantor trees which are separated and uniformly conal in (T ) . By CH we have that |C| (1 ) = 1 . Let C = {C : 1 } be an enumeration, where C = {Fs : s 2< }. We now want to nd a set X { B : B B(T )} such that for every 1 , there are f, g 2 such that
n

Ff n (i) : i } X

{}

revision:1993-08-29

and
n

Fg n (i) : i }

X = .

If X is found, we let T = X. We now build X and Y recursively such that, (1) X and Y are countable, (2) < < 1 implies that X X and Y Y , 3

466

(3) X Y = for every 1 , (4) for every 1 , there exist f, g 2 such that { n Ff n (i) : i } X+1 and { n Fg n (i) : i } Y+1 . Let X0 = Y0 = . Let X = < X and Y = < Y if lim(1 ). For + 1, since X and Y are countable and C is separated, by Lemma 1, there exist f, g 2 , f = g such that
n n

(X Hence let

Y )

({

Ff n (i) : i }

Fg n (i) : i }) = .

n supt(Fg n )}. n

and
n

Y+1 = Y Then X =
1

Fg n (i) : i

X is the set we want.

Lemma 3 Let S lim(1 ). T is Sproperly pruned in countable products implies that T is Sproperly pruned. Proof: If C = {cs : s 2< } T is a Cantor tree which is conal in T for some S, then the Cantor tree D = {Fs : s 2< } T , where Fs (0) = cs and Fs (i) = for every i = 0, is separated and uniformly conal in (T ) . 2
modified:2002-07-16

Lemma 4 Let S lim(1 ) and T be Sproperly pruned in countable products. Let C = {Fs : s 2< } be a separated and uniformly conal Cantor subtree in (T ) for some S. Then there are uncountably many f 2 such that for every i n supt(Ff n ), n Ff n (i) T . Proof: Suppose that the lemma is not true. Then we can nd a Cantor subtree C = {Fs : s 2< } C such that for every f 2 , there exists i , n Ff n (i) T . Since C is a subtree of C, C itself is also separated and uniformly conal in (T ) . That contradicts the denition of the Sproperly prunedness in countable products. 2 Next we shall use the forcing method to construct desired models. For the terminology and basic facts of forcing, see [K2] and [Je2]. We always assume the consistency of ZFC and let M be always a countable transitive model of ZFC. In the forcing arguments, we always let a be a name of a if a is not in the ground model. For every element a in the ground model, we will not distinguish a from its canonical name. Let I, J be two sets. Let F n(I, J, 1 ) = {p : p I J is a function and |p| < 1 } 4

466

revision:1993-08-29

X+1 = X

Ff n (i) : i

supt(Ff n )}

be a poset ordered by reverse inclusion. Let I be a subset of a cardinal . Let Lv(I, 1 ) = {p : p (I 1 ) is a function, |p| < 1 and , dom(p)(p(, ) )} be a poset ordered by reverse inclusion. Let T be a tree and I be an index set. Let (T, I, 1 ) = {F : F T I and |supt(F )| < 1 }.

The order of (T, I, 1 ) is dened as the reverse order of T I , or F G F.

(T,I,1)

G i

Lemma 5 Let T be a normal 1 tree and I be an index set. For any p, q (T, I, 1 ), there exist p , q (T, I, 1 ) such that p < p and q < q, p , q are uniform at for some 1 , and p is completely incompatible with q . Proof: Let 1 be large enough so that p, q (T )I ( exists because p, q both have countable supports). Let > be countable such that for every i supt(p) |{t T : p(i) <T t}| , and for every j supt(q) |{t T : q(j) <T t}| . exists because T is normal. Let
modified:2002-07-16

supt(p) = {in : n } and supt(q) = {jn : n }. We now dene p (in ) and q (jn ) such that p (in ), q (jn ) T , p (in ) > p(in ), q (jn ) > q(jn ), p (in ) = q (jn ) and

revision:1993-08-29

p (in ), q (jn ) {p (im ), q (jm ) : m < n}. Let p (i) = if i supt(p) and let q (j) = if j supt(q). Then p and q are the desired elements. 2 Let be a poset and D . D is called dense in if for every p there is d D such that d p. D is called open in if for every p and d D, p d implies that p D. is called 1 Baire if for any countable sequence Dn : n of dense open subsets of , n Dn is dense in . 5

466

Lemma 6 In M let be a poset which is 1 Baire. Let G be a generic lter over M . Then M M [G] M . Proof: Let h M [G] be a function from to A, where A M . We work in M and let p G such that p | (h is a function from to A). For every n , let

Dn = {q

: q p or a A(q | h(n) = a)}.

n

Then Dn is dense open in . Let p

Dn such that p p. Then

h = { n, a A : p | h(n) = a} M. 2 Lemma 7 Let S lim(1 ) and T be an 1 tree which is Sproperly pruned in countable products. Then for any index set I, the poset (T, I, 1 ) is 1 Baire. Proof: For each n , let Dn be a dense open subset of (T, I, 1 ). Let p (T, I, 1 ). We now construct ps (T, I, 1 ) for every s 2< inductively on the length of s such that, (1) p0 p, (2) s t i pt ps , (3) there is an increasing sequence n : n of countable ordinals such that for every s 2n , ps is uniform at n . (4) for every s 2< , ps 0 and ps 1 are completely incompatible, (5) for every s 2n , ps Dn . Assume that we have already had ps for every s 2<n . Let s 2n1 and q s Dn such that q s ps . Let l = 0, 1. By Lemma 5, there are s s s ql < q s such that q0 and q1 are completely incompatible. Let n = {htT (qls (i)) : i I, s 2n1 , l = 0, 1} + 1.

modified:2002-07-16

revision:1993-08-29

n is countable because the support of every qls is countable. Let ps l (T, I, 1 ) such that ps l qls and all ps l are uniform at n . ps l Dn because ps l q s . Let I = s2< supt(ps ). Then I is countable. C I = {ps I : s 2< } is now a Cantor tree in T I , which is separated by (4) and uniformly conal in (T )I , where = n n . Since T is Sproperly pruned in countable products, there exists f 2 such that for every i I , n pf n (i) T {}. 6

466

Let pf (T, I, 1 ) dened by letting

n

pf I = and Then pf (T, I, 1 ) and pf pf 2

pf n (i) : i I

for every n . So pf p and pf

Theorem 8 Assume the existence of a strongly inaccessible cardinal. It is consistent with CH plus 21 > 2 that there exists a JechKunen tree and there are no Kurepa trees. Proof: Let M be a model of GCH plus that there is a strongly inaccessible cardinal . In M , let T be an 1 tree which is lim(1 )properly pruned in countable products and let and be two regular cardinals such that < . Again in M let 1 = Lv(, 1 ), 2 = (T, , 1 ) and 3 = F n(, 2, 1 ). Let G = G1 G2 G3 be a 1 2 3 generic lter over M . We will show that M [G] is a model of CH plus = 21 > 2 = , in which there are no Kurepa tree and T is a JechKunen tree with many branches. Claim 8.1. M M [G] M . Proof of Claim 8.1: We rst force with 2 . By Lemma 6 and Lemma 7, 2 is 1 Baire and forcing with 2 will not add any new countable sequences. Hence 1 3 is still 1 closed in M [G2 ]. Then forcing with 1 3 will also not add any new countable sequences because it is 1 closed. Claim 8.2. 1 2 Proof of Claim 8.2: Let

modified:2002-07-16

has the c.c..

revision:1993-08-29

By the system lemma, we can assume that the domains of all p , the domains of all q and the domains of all r form three systems with roots 1 , 2 and 3 respectively. Since there are less than many ps in 1 with domains = 1 , there are 1 many qs in 2 with domains = 2 , and there are 1 many rs in 3 with domains = 3 , then there exist 1 and 2 in such that p1 1 = p2 1 , q1 2 = q2 2 and r1 3 = r2 3 . Obviously p1 , q1 , r1 and p2 , q2 , r2 are compatible. Remark: By Claim 8.1 and Claim 8.2, 1 and all the cardinals greater than or equal to in M are preserved and CH is true in M [G]. In M [G], = 2 because forcing 7

466

{ p , q , r : < }

3.

pf I

I .
n

Dn .

with 1 collepses all the cardinals between 1 and in M . Also in M [G], 21 = because forcing with 3 adds many subsets of 1 . Claim 8.3. There are no Kurepa trees in M [G]. Proof of Claim 8.3: Suppose that is not true. Let K be a normal Kurepa tree in M [G]. Since |K| = 1 , there are < , I with |I| 1 and J with |J| 1 such that K M [G ] = M [G1 G2 G3 ], where G1 = G 1

Lv(, 1 ), (T, I, 1 ),

G2 = G 2 G3 = G 3 and

F n(J, 2, 1 )

G = G1 G2 G3 . Let

and G = G1 G2 G3 .
modified:2002-07-16

Since M [G ] |= 21 < , there exists

b B(K) Furthermore

M [G]

b M [G ][G1 ][G3 ] because Lv( , 1 ) and F n( J, 2, 1 ) are 1 closed in M [G ]. We now work in M [G ][G1 ][G3 ] and let p G2 such that

p | (b B(K)

M [G]

revision:1993-08-29

We construct and D = {ks : s 2< } K such that, (1) s s i ps ps i ks ks , 8

C = {ps : s 2< } (T,

466

G3 = G 3

F n(

G2 = G 2

(T,

G1 = G 1

Lv(

, 1 ), I, 1 ), J, 2, 1 )

M [G ].

M [G ][G1 ][G3 ]).

I, 1 )

Assume that we have already had ps and ks for all s 2<n . Let n = and pick s 2n1 . Let l = 0, 1. First nd ps ps such that x Kn (ps | x b). Since ps | b M [G ][G1 ][G3 ], {htK (ks ) : s 2<n } + 1

there exist qls ps and xl > x > ks such that x0 x1 and qls | xl b. By Lemma s 5, we can extend qls to rls such that rls are uniform at s < 1 and r0 is completely s incompatible with r1 . Let n = {s : s 2n1 } + 1,

ps l be an extension of rls such that supt(ps l ) = supt(rls ) and ps l be uniform at n . This ends the construction. Let = n n , = n n and I = s2< supt(ps ). Then I is countable. Since T is lim(1 )properly pruned in countable products and C I is a Cantor tree which is separated and uniformly conal in (T )I , then there are uncountably many f 2 such that pf dened by letting
n

modified:2002-07-16

pf (i) =

pf n (i)

for every i I is a lower bound of {pf n : n } in (T, , 1 ). (Note that C is in M because no new countable sequences are added.) For every such f there exists kf K such that pf | kf b and for dierent f , kf are dierent. That contradicts that K is a Kurepa tree. Claim 8.4. M [G] |= (|B(T )| = ). Proof of Claim 8.4: |B(T )| is trivial because forcing with 2 adds at least many new branches of T . Since in M [G1 ][G2 ], 21 = , then we need only to show that forcing with 3 will not add any new branches of T . Suppose that is not true and let b be a branch of T , which is in M [G] M [G1 ][G2 ]. We now work in M [G1 ][G2 ] and let p G3 such that

revision:1993-08-29

p | b B(T )

M [G] 9

M [G1 ][G2 ].

466

(2) C is separated and uniformly conal in (T ) (3) D is conal in K for some lim(1 ), (4) for every s 2< , ps | ks b.

for some lim(1 ),

We can then easily construct C = {ps : s 2< } 3 and D = {ts : s 2< } T such that, (1) s s i ps ps i ts ts , (2) D is a Cantor tree which is conal in T for some lim(1 ), (3) for every s 2< , ps | ts b. Since T is lim(1 )properly pruned by Lemma 3, there exists g 2 such that n tg n T . But 3 is 1 closed in M [G1 ][G2 ] because no new countable sequences have been added. Hence there exists pf 3 such that pg pg n for every n . This implies that there exists t T such that pf | t b. Hence
n

t= a contradiction. 2

tg

T ,

In the model constructed above, there is only one JechKunen tree. Next we will build a model of CH plus 21 > 2 , in which there are no Kurepa trees and there are many JechKunen trees with dierent numbers of branches. Theorem 9 Assume the existence of a strongly inaccessible cardinal. It is consistent with CH plus 21 > 2 that there are no Kurepa trees and there are JechKunen trees T for 1 such that = implies |B(T )| = |B(T )|. Proof: Let M be a model of GCH and that there exists a strongly inaccessible cardinal . In M , let = { : 1 } [, ) be a set of dierent regular cardinals, where is also a regular cardinal. Again in M , let {S : 1 } be a partition of lim(1 ) such that every S is a stationary, and let T be an 1 tree which is S properly pruned in countable products for every 1 . In M , let 1 = Lv(, 1 ), 2 be the product of { (T , , 1 ) : 1 } with countable supports, and 3 = F n(, 2, 1 ). Let G = G1 G2 G3 be a 1 2 3 generic lter over M . Then M [G] is the model we are looking for. Claim 9.1. M M [G] M .
2

modified:2002-07-16

Claim 9.2.

has the c.c..

revision:1993-08-29

Claim 9.3. There are no Kurepa trees in M [G]. All the proofs of above three claims are similar to the proofs of corresponding claims in Theorem 8. By Claim 9.1 and Claim 9.2, 1 and all the cardinals greater than or equal to are preserved. Besides, forcing with 1 collapses all the cardinals between 1 and . So in M [G], CH is true, = 2 < = 21 and { : 1 } [, ) is still a set of dierent cardinals. Claim 9.4. M [G] |= (|B(T )| = ) for every 1 . 10

466

Proof of Claim 9.4: Pick an 1 . Let = (T , , 1 ) and be the 2 2 product of { (T , , 1 ) : 1 {}} with countable supports. Then 2 = 2 2 . Let p 2 . We let SUPT (p) = { 1 : supt(p()) = }.

Notice the dierences between supt and SUPT . We call an element p at for some 1 if for every SUPT (p), p() is uniform at .

uniform

Subclaim 9.4.1. Forcing with will not add any new branches to T . 2 Proof of Subclaim 9.4.1: Let G2 = G G . Suppose that Subclaim 2 2 2 2 1 is not true and let b be a branch of T shch that

b M [G1 ][G2 ]

M [G1 ][G ]. 2

We now work in M [G1 ][G ] and let p G such that 2 2 p | b B(T )

M [G1 ][G ]. 2

modified:2002-07-16

We construct recursively a normal subtree T of T with every level countable, and a subset C = {pt : t T } of such that, 2 (1) for every 1 there is such that T T , (2) if lim(1 ), then = < , (3) p p, and for any t, t T , t t i pt pt , (4) for every t T , there is , +1 such that pt is uniform at , (5) if t T for some lim(1 ), then pt is uniform at htT (t) = , (6) t t implies that pt () and pt () are completely incompatible for every SUPT (pt ) SUPT (pt ), (7) for every t T , pt | t b. Case 1. = + 1 for some 1 . Pick t T and let l = 0, 1. Since pt | b M [G1 ][G ], 2 there exist tl T , tl > t and qlt pt such that t0 t1 and qlt | tl b. Without loss of generality we can pick tl such that htT (tl ) = for every t T and l = 0, 1, where > { : pt is uniform at for some t T }. 11

Assume that we have already had T

and C = {pt : t T

}.

466

revision:1993-08-29

t t Besides, we can require that q0 () and q1 () are uniform and are completely incomt t patible for every t T and SUPT (q0 ) SUPT (q1 ). Let 1 such that > and > { : qlt is uniform at for some t T and l = 0, 1}.

Let T = {tl : t T , l = 0, 1} and let ptl qlt such that ptl is uniform at . Case 2. lim(1 ). First cant be in S because otherwise every T for 1 {} is complete at level . But in M [G1 ][G ], T is still properly pruned at level because forcing 2 with 1 adds no new countable sequences, so that there exists B B(T ) 2 such that B has no upper bound in T . On the other hand, {pt : t B} has a lower bound pB in . Then 2 pB | t T (t b) implies that B has an upper bound in T , a contradiction. Assume that S for some = . Since in M [G1 ][G ], T is properly pruned 2 at level , then for every t T there exists Bt B(T ) such that t Bt and t Bt pt ()(i) i (T , , 1 ). Now every T is complete at level for = . We can dene pBt by 2 letting pBt ()(i) = pt ()(i)
t Bt

for every 1 {} and i . Let T = { Bt : t T } and let p

modified:2002-07-16

Bt

= pBt . This ends the construction.

Since S is stationary and by (2), { : 1 } is a club set, then there exists 1 such that S . But this has been shown impossible. Subclaim 9.4.2. Forcing with 3 will not add any new branches to T . Proof of Subclaim 9.4.2: Similar (but much easier) to the proof of Subclaim 9.4.1. By Subclaim 9.4.1 and Subclaim 9.4.2, all the branches of T in M [G] are already in M [G1 ][G ]. But in M [G1 ][G ] 21 = . So |B(T )| = . 2 2 2 Concluding remarks. (1) , and are not necessarily regular. (2) In Theorem 9, we can also have larger number of trees. For this we use S s which are only almost disjoint. (3) In the proof of Theorem 9, if we do not want to use stationary sets, we can force the trees as part of the forcing, and then prove that they are pruned together, so using the stationary sets simplies the matter. (4) We have used (lim(1 ) S)complete tree T (i. e. every branch of T for lim(1 ) S has an upper bound in T ). Our consideration leads naturally to S Kurepa trees. T is called an SKurepa tree if T = 1 T (), where ht(T ()) = ,

revision:1993-08-29

12

466

T () = < T () if lim(1 ) and |T { B : B B(T ())}| if S. So we may well consider SKurepa and (lim(1 ) S)complete trees. (5) The T we build are not only (lim(1 ) S)complete, but also strongly proper (see [S1] or/and [S2]).

References
[Je1] T. Jech, Trees, The Journal of Symbolic Logic, 36 (1971), pp. 114. [Je2] , Set Theory, Academic Press, New York, 1978.

[Ji1] R. Jin, Some independence results related to the Kurepa tree, Notre Dame Journal of Formal Logic, 32, No 3 (1991), pp. 448457. [Ji2] , A model in which every Kurepa tree is thick, Notre Dame Journal of Formal Logic, to appear.

[Ju] I. Juhsz, Cardinal functions II, pp. 63110 in Handbook of Set Theoretic a Topology, ed. by K. Kunen and J. E. Vaughan, NorthHolland, Amsterdam, 1984. [K1] K. Kunen, On the cardinality of compact spaces, Notices of The American Mathematical Society, 22 (1975), 212. [K2]
modified:2002-07-16

, Set Theory, an introduction to independence proofs, North Holland, Amsterdam, 1980.

[S1] S. Shelah, Proper Forcing, SpringerVerlag, 1982. [S2] , New version of Proper Forcing, to appear.

[T] S. Todorevi, Trees and linearly ordered sets, pp. 235293 in Handbook of c c Set Theoretic Topology, ed. by K. Kunen and J. E. Vaughan, NorthHolland, Amsterdam, 1984. Institute of Mathematics, The Hebrew University, Jerusalem, Israel. Department of Mathematics, Rutgers University, New Brunswick, NJ, 08903, USA. Department of Mathematics, 13

466

revision:1993-08-29

University of Wisconsin, Madison, WI 53706, USA. Sorting: The rst two addresses are the rst authors; the last one is the second authors.

revision:1993-08-29

modified:2002-07-16

14

466