Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Trees

Author(s): Thomas J. Jech


Source: The Journal of Symbolic Logic, Vol. 36, No. 1 (Mar., 1971), pp. 1-14
Published by: Association for Symbolic Logic
Stable URL: https://www.jstor.org/stable/2271510
Accessed: 13-11-2018 00:01 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend
access to The Journal of Symbolic Logic

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
Tim JouRNAL OF SYMBOLIC LoGic
Volume 36, Number 1, March 1971

TREES

THOMAS J. JECH

This is an attempt to give a survey of recent results concerning tr


is an extended version of our talk in Oberwolfach (Schwarzwald) last May; the
forests surrounding the Forschungsinstitut turned out to be a good inspiration.
A tree is a partially ordered set T = (T, <) such that for every x E T, the set
x = {y E T: y < x} is well-ordered. The order type of x is called the order of x,
o(x), and the length of T is sup {o(x) + 1: x e T}; an a-tree (where a is an ordinal)
is a tree of length a. The ath level of T is the set Ups of all elements of T whose order
is ca. TJla is the union of all U0, fi < a; its length is ca. A tree (T2, <2) is called an
extension of (T1, <?1) if <l = ?2 fn (T1 x T1); T2 is an end-extension of T1 if
T1- = T2 ca for some a. A maximal linearly ordered subset of a tree T is called a
branch of T; an a-branch is a branch of length a.
The organization of the paper is as follows: In ?1, we present a classical con-
struction of an Aronszajn tree and discuss the problem of its generalization to
higher cardinals-the problem known as "the tree property" of cardinals. In
?2, we mention the relationship between trees and the famous Suslin's Problem
and give consistency proofs for both Suslin's hypothesis and its negation. ?3 dea
with Kurepa's conjecture and again, models are presented both for the conjecture
and its negation. The last section contains recent results which settle Suslin's and
Kurepa's problem in the constructible universe.
Most of the proofs in this paper are only sketched. Our intention is to show
the ideas of the proofs rather than provide all details. We believe that an interest
reader can recover the proofs himself. The consistency proofs in ??2 and 3 requir
of course, a certain amount of knowledge of basic techniques of forcing. (We
refer here to [2], [12] or [21].) Throughout the paper we use the notation standard
in set-theoretical circles.

?1. Aronszajn trees. Let us begin with the following trivial fact, sometimes
quoted as "Kbnig's Lemma":
If T is an co-tree and if every level is finite then T has an co-branch.
A natural generalization of this fact would be: If T-is an co,-tree whose
are countable then T has an wol-branch. This is however not true and a c
result due to Aronszajn is that wol-trees with countable levels and withou
branches exist. It should be perhaps remarked that an cn-tree T whose levels
are only finite, does have an col-branch. For, consider the set T' of all elements
x of T such that {y e T: y > x} is uncountable. T' is an wn-tree and only finitely

Received September 15, 1969.

1-J.S.L. 1

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
2 THOMAS T. JECH

many elements of T' split, i.e. have more than one immediate successor. To con-
struct an w1-branch in T' is then easy.
We shall now construct an cwl-tree T whose levels are countable and which has
no wu-branches. The elements of Twill be certain increasing transfinite sequences
of rational numbers, in natural ordering: if x = <q:: t < a> and y = <re:
e < 8> then x < y means that a ?, P and q4 = rz for all e < a. We sh
following properties of the set Q of all rational numbers:
(1.1) Q is countable;
(1.2) every ac < col can be embedded in any interval of Q.
The property (1.1) ensures that any tree constructed from increasing sequences
of rational numbers has no co,-branch. The tree T cannot, however, consist of all
sequences, since then already the coth level would be uncountable. Thus we must
make a selection among the sequences; we shall construct T by recursion:
having constructed Tla, we construct the ath level Us, (consisting of a-sequences
of rationals). It turns out that the following condition makes the construction work:
(1.3) (Vx TJa)(Vq > sup x)(3y E U,)[y v x & sup y ? q].
The induction step from a to a + I is easy; we let Use + 1 be the set of all xrq
where x E Ua and q > sup x. Clearly U, + 1 is countable and (1.3) holds for a + 1.
If A is a limit ordinal, having constructed TjA such that (1.3) holds for all a < A
then first we observe that TJ A has A-branches; moreover, for each x e TJA and
for each q > sup x there is a A-branch y going through x and such that sup y < q
(here we do not distinguish between y and the corresponding A-sequence of ra-
tionals which y represents). This is proved as follows: let A = lim A an and
q 2 limqC qn (both increasing); by (1.3) we obtain x C Yo C Yi C ... such that
sup yn < q,, and we let y =Un=O oYn, Now we construct Use as follows: for
x e TJ A and each q > sup x we choose one y which we put into UA. Again, UA
is countable and (1.3) holds for A. The tree T = Ua ,,, TIla is then an cw,-tree wi
countable levels and without cw-branches-an Aronszajn tree.
We can ask whether this result generalizes to higher cardinals. Let K be a card-
inal, let us call an Aronszajn K-tree a K-tree whose levels are of power less than K
and which has no K-branches. If there is no Aronszajn K-tree we say that K has the
tree property (TP). It is clear that singular cardinals have not TP. A classical
result [17] says that if the generalized continuum hypothesis (GCH) holds then
for every regular K there exists an Aronszajn K+-tree. However, not much is
known in absence of GCH and even assuming GCH, not much is known about
TP of successors of singular cardinals.' Tarski and his school proved that if K
is strongly inaccessible then TP is equivalent to several other properties (Erdbs'
K -* (K)2, HIl-indescribability, etc.); cardinals with these properties are commonly
known as weakly compact cardinals and are rather large (cf. [6]).
We shall show now how to generalize the construction of Aronszajn trees and
why GCH is needed.2 Let K be a cardinal, we want to construct an Aronszajn

1 D. Kurepa's result in [22] that Aronszajn K+-trees exist for every K, is false.
2 (For model theorists.) F. Rowbottom and J. Silver observed that the existence of Arons-
zajn K+-trees follows immediately from the existence of -WI-trees, using the two cardinals
conjecture (W1, W) -4. (Ki, K). The conjecture is known to be true for regular K, assuming
GCH.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
TREES 3

K+-tree. Instead of rationals, we take a set Q such that


(1.4) IQI = K,
(1.5) every a < K+ can be embedded in any interval of Q.
E.g. Q can be the set of all finite sequences of ordinals less than K, ordered
lexicographically.
The constructed K +-tree consists of increasing a-sequences of elements of Q,
for a < K +. We proceed again by recursion and try to preserve the analogue of
(1.3). The step from a to a + 1 is easy; if A is a limit ordinal and if cf(A) < K
then we put every branch of TI A into UA (*); if cf(A) = Kthen we observe that for
each x e TIA and each q > sup x there is a A-branch y of TJ A going through x
and such that sup y c q (as above, we let A = l and q > lim_.,<q:; in
construction of y we of course use also (*)). Everything is O.K.; our only con-
cern is whether I UAI < K for all A and (*) is the only vulnerable place. The number
of all A-branches in TjA is KCf(A); hence we must assume that KO = K (where Ka =

This excludes automatically every singular K, since then K > K. In general, it


gives a sufficient condition for the existence of Aronszajn K +-trees, for regular K;
e.g. if the continuum hypothesis (CH) holds then there exists an Aronszajn W2-tree.3
Not much is known about TP except the cases stated above. However, there
are several recent results which throw more light at this problem. In his thesis
[14], Silver observed that every real-valued measurable (RVM) cardinal has TP.
(A cardinal K is RVM if there is a K-additive nontrivial nonnegative real-valued
measure on K.) For, having a K-tree T, consider the set T' of all x e T such that
{y e T: y > x} has a positive measure. T' is a K-tree whose every level is count-
able. Necessarily, T' has a K-branch (the same argument as for uncountable trees
with finite levels). The observant reader can certainly see that a K-complete A-
saturated (A < K) nonprincipal ideal on K will do as well (instead of RVM). Since
RVM cardinals are not necessarily strongly inaccessible it turns out that not every
cardinal with TP must be weakly compact. Moreover, since "2KD is RVM" is
consistent (relative to some large cardinal assumption), we obtain the consistency
of "2Ko has TP".
Another consistency result is due to Pnikry [10]. He started with a measurable
cardinal K (MC) in the ground model and constructed a Cohen extension where
K is XM and all cardinals above K are preserved. In the ground model there is an
Aronszajn K+-tree T, e.g. that constructed above using sequences in Q. In the
extension, T is an M, + l-tree, whose levels have power XH,. Also, there
branch in T, since every branch represents an increasing sequence in Q; but
I QI = xM in the extension. Thus we have the consistency of " ,+ has n
relative to a large cardinal assumption.

3 The tree constructed here satisfies the following normality conditions: (a) every point
splits into K successors; (b) for limit A, each A-branch in TIA has at most one extension in
Tj (A + 1); (c) for each linearly ordered subset C of power less than K and for each a < K',
there is a branch B ' C of length at least cc.
It should be mentioned that, assuming KG = K, there are 2'" + nonisomorphic normal Arons-
zajn K+-trees (23]. In particular, the number of isomorphism-types of normal Aronszajn
w1-trees is 2K1.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
4 THOMAS J. JECH

Recently, Silver and Mitchell constructed a model (using a weakly compact


cardinal in the ground model), where 21o = X2 and X2 has TP. Thus we may
sum up what is known so far about the tree property:4
THEOREM. (a) X, has not TP;
(b) if K- = K then Kc has not TP;
(c) if K is strongly inaccessible then K has TP if and only if K is weakly compact.
The following theories are consistent relative to ZFC (Zermelo-Fraenkel + Choice)
(+ some large cardinal assumption):
(d) ZFC + 2Ro has TP;
(e) ZFC + X.c+ has not TP.

?2. Suslin trees. Here we show how the famous Suslin's Problem can be
formulated in terms of trees and what has been done about it.
Suslin's hypothesis (SH), cf. [19], reads as follows:
(2.1) If Q is a linearly ordered set such that every disjoint family of open intervals
is countable then Q contains a countable dense subset (and is embeddable in the
real line).
Several people have observed, cf. [7], [9], [13] that SH can be formulated in terms
of trees. Let us call a subset A of a tree T an antichain if any two distinct elements
of A are incomparable (e.g. every level is an antichain). A Suslin tree (ST) is an
wl-tree which has no col-branches and no uncountable antichains.
LEMMA 1. SH holds if and only if there is no ST.
PROOF. (a) If T is a ST then let Q be the set of all branches of T. We order
Q as follows: we assume that every level of T is linearly ordered and if bl, b2 E Q.
we let b1 < b2, if Us, is the least level where b1 and b2 differ and if the ath element
of b1 precedes the ath element of b2 in the ordering of U4,. Q satisfies the assump-
tions of (2.1) but no countable subset is dense in Q.
(b) Suppose that Q is a linearly ordered set which satisfies the assumptions of
(2.1) and has no countable dense subset. We construct, by recursion, a sequence
Io a < col, of nonempty open intervals of Q: 10 is arbitrary; having constructed
I4,, P < a, let A be the set of all endpoints of all 4; since A is countable, there
an interval, It., disjoint from A. The set of all I4, a < wo,, ordered by inverse
clusion, is a ST.
In constructions later in this section, it will be more convenient to deal with
Suslin trees which satisfy the following normality conditions:
(2.2) (a) every element splits;
(b) for every- x, the set {y: y > x} is uncountable.
It is not difficult to construct a ST satisfying (2.2) if we have some ST. From now
on, we shall deal only with normal Suslin trees; it enables us, among other things,
to omit the condition concerning w1-branches. For, if an co,-tree T satis
and has an col-branch b then it has an uncountable antichain A: for every x E b
let y(x) > x be an element not in b and let A = {y(x): x E b}.
In recent years, Suslin's Problem has been settled by showing that SH is neither

4 By a result of Jensen [26], if V = L then no K+ has TP; thus (e) is consistent relative to
ZFC.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
TREES 5

provable nor refutable in ZFC. The consistency of -SH was established indepen-
dently by Tennenbaum (earlier) and ourselves; the consistency of SH is a joint
result of Solovay and Tennenbaum.
A model where there is a ST [20]. Here and also several times thereafter we
shall establish a consistency result by producing a model of ZFC. We shall present
a partially ordered set (P, <), the set of forcing conditions, and then taking a
P-generic set G, we obtain a Cohen extension V[G] of the universe (the ground
model) V. We say that a forcing condition p is stronger than q if p < q.
To come back to Tennenbaum's model, we shall consider the following set
(P. <) of forcing conditions:
(2.3) a forcing condition is a finite tree T = (T, <), for our convenience satisfy-
ing the conditions
(a) T cwl,
(b) if c< 8 in the tree T then cc < f as ordinals;
a condition T1 is stronger than T2 if the tree T2 is an extension of T1.
If G is a generic set of conditions then the "generic" tree .- = UTeGT is ob-
viously an wl-tree; to prove that Y is a ST, we have to show that
(a) wl is a cardinal in V[G];
(b) V[G] I every antichain in .Y. is countable.
Both proofs follow the same pattern. The crucial fact is a classical combinatorial
lemma due to Marczewski [8]:
LEMMA 2. Let W be an uncountable family of finite subsets of WC. Then there is
an uncountable Z c W and a finite S c w1 such that X n Y = S for any distinct
elements X, Y of Z.
PROOF. For Xe W, let X1 be the nth element of X (in the natural order).
There exists n such that {X: Xe- W} is uncountable; let no be the least one. We
can construct an uncountable W1 c W such that for any distinct X, Ye W1,
the noth element of X is greater than all elements of Y (or conversely). The family
{{Xo0., 4On - 1} Xe W1} is at most countable; hence there is an uncountable
Z c W1 such that {X0, * - , XO - 1} is the same for each X e Z.
To show that co, is preserved in the Cohen extension V[G], it is sufficient to
verify that the set of forcing conditions satisfies the countable chain condition
(c.c.c.): every set of pairwise incompatible conditions is at most countable. Given
an uncountable set W of conditions, a Marczewski-type argument provides an
uncountable Z c W and a finite S c co, such that T1 n T2 = S and
(T1, ? 1)IS = (2, ? 2)JS for any pair of distinct elements of Z; cons
any two elements of Z are compatible.
To prove that the generic tree has no uncountable antichain, suppose that
some condition To forces "there is an uncountable antichain A". It follows that
for every e < co, there exists a condition T 2 To and an ordinal aT E T such
that T I Fac e A. We apply a Marczewski-type argument to this uncountable set of
conditions and find an uncountable subset Z such that: if T1, T2 are distinct ele-
ments of Z then there exists a condition T3 stronger than both T, and T2 and
aT1 is comparable in 3 with aTs. This means that T3 forces " aT1 E A, aTl e A and
a. is comparable with aT", which is contradictory.
This model establishes the consistency of ZFC + -SH. Moreover, it estab-

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
6 THOMAS J. JECH

lishes both the consistency of ZFC + -SH + CH and of ZFC + -qSH + -CH.
This is because the continuum hypothesis holds in the above extension if and only
if it holds in the ground model. (The reader familiar with forcing will certainly
see this.)5
Another model where there is a ST [4]. We shall consider the following forcing
conditions:
(2.4) a forcing condition is a countable tree T = (T, <), for our convenience
normalized as follows:
(a) T ' wl, or better, T = {e: e < f} for some P < w1;
(b) every element (not in the last level) splits;
(c) for each x E T there exists an a-branch going through x (a is the length of T);
a condition T1 is stronger than T2 if T1 is an end-extension of T2.
If G is a generic set of conditions then the generic tree 9 = UTEGT is obviously
an wl-tree; to prove that - is a ST, we have again to show that:
(a) w, is a cardinal in V[G];
(b) V[G] k every antichain in Y is countable.
To show that w, is preserved, it is sufficient to realize that the set of forcing con-
ditions is countably closed: every decreasing countable sequence of forcing con-
ditions has a lower bound. Certainly, if To ? T1 ? T2 ? ... is such a sequence
then T = U= oT, is a condition which is stronger than each T,.
To prove that the generic tree has no uncountable antichain, suppose that
some condition To forces "A is a maximal antichain". We construct, by recursion,
a transfinite sequence of conditions To ? T1 ? ... ? T ? . .. and of ordinals
fl: having constructed Tz, there exist (by maximality of A) a condition T,+j < T:
and an ordinal flz such that Tz + 1 F "c- eA and Pz is comparable in Y with
aI; if A is a limit ordinal, we simply let TA = U< ATe. Since 6 5 Tz is an increasing
continuous function, there exists a countable limit ordinal a such that Ta = a =
{f: e < al. For a similar reason, a can be taken even such that the length of Ta
is a. Clearly,

Ta IF7ja = Ta &A n a = Aa
where Aa = e < a}; since every 6 < a is comparable with some ,B e Ta, Aa
is a maximal antichain in Ta. The crucial step now is
LEMMA 3 (KILLING A MAXIMAL ANTICHAIN). If Aa is a maximal antichain in Ta
then there is a tree T 2 Ta of length a + 1 such that Aa is a maximal antichain in
T; consequently, Aa is a maximal antichain in every end-extension of T.
PROOF. For each x e T,, there is y e Aa comparable with x. Let bX be an a-
branch of Ta going through x and y. We let T be the end-extension of T,, obtained
by extending all the branches b., x e Ta (and omitting all other branches).
Since Aa is a maximal antichain in every T' < T, we have T I F A = A,,. Thus
we have found a condition T < To such that T I F "A is countable". Hence the
generic tree .7 has no uncountable antichain.

5 An interesting property of Tennenbaum's model is that it is the same model as the model
obtained by adjunction of a Cohen-generic real. This follows from the fact that both Tennen-
baum's and Cohen's forcing conditions correspond to the same complete Boolean algebra
(R. Solovay).

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
TREES 7

There is another interesting thing about this forcing: For every condition T
there are exactly 2Ko pairwise incompatible extensions of T. Using this and the
fact that the set of forcing conditions is countably closed, we may identify the
forcing conditions with functions whose domain is a countable ordinal and taking
values in 2"o (the ordering is by inverse inclusion). This is the set of forcing con-
ditions known as "collapsing" forcing: it makes 12Xol = M, and preserves
to, and all cardinals greater than 2Ko. Thus in our model CH always holds, reg
less of the size of 2Xo in the ground model.
Suslin's Problem can be generalized to higher cardinals. A Suslin K-tree is a
K-tree which has no K-branch and no antichain of cardinality K. It has been shown
in [3] that the above method can be used to get the following:
If Oc = K then there is a model where there is a Suslin K+-tree.6'7
The forcing conditions are trees of length less than K+ which satisfy conditions
similar to (2.4); the condition (c) is replaced by
(c') if C is a linearly ordered subset of T and if j Cl < K then there is a branch
B (as long as T) such that B 2 C.
A model where there is no Suslin tree. We are going to sketch the construction
described in [16]. Unlike the preceding constructions we use the Boolean algebraic
version of forcing. An algebra means here a complete Boolean algebra. V is the
(real) universe and V(B) the Boolean universe given by an algebra B. V C V(B)
denotes the canonical embedding of V in V(B); similarly, V(B1) c V(B2) denotes
the embedding given by B1 c B2, i.e. by a complete monomorphism of B1 into B2.
To begin with, we state (without proof) the fact that "a Cohen extension of a
Cohen extension of V is a Cohen extension of V". A little more precisely:
If B is an algebra and if V(B) I (C is an algebra), then there is an algebra D 2 B
such that V(D) is (isomorphic to) V(B)(C).
As a matter of fact we shall need some more information about the "double
forcing":
(1) If, moreover, B is c.c.c. and V(B) I= (C is c.c.c), then D is c.c.c.
(2) If, in addition, 2'1 = 92, JBj < 92 and V(B) I jCj < X2, then |DI <
Another thing we shall use is the following Boolean algebraic lemma proved
in [16]:
LEMMA 4. If Bo B1 C B, B,, c *--, x < K is a sequence of c.c.c. alge-
bras such that BA = lima .ABa for each limit ordinal A < K, then lima<, B,, is a
c.c.c. algebra.
PROOF. Let us mention that BA = limrA B, means that U.<,AB,, is dense in
Be. Suppose that lima?Bg is not c.c.c., so that there exists W ' Ua<t,,B, of pair-
wise incompatible elements, I WI = M1. It follows that cf(K) = wl; hence we may
suppose K = W1. For every x E Ua<,,,liB<, let p(x) be the least a( such that x E B,.
For every x # 0, p(x) > 0, we define j(x) # 0 as follows: if p(x) = ac + 1 then
j(x) = inf {r e B,,: r ? x} if p(x) = A (a limit ordinal) then j(x) is some r # 0,
r E Uca<,ABat, such that r < x.

6 It can be required that the tree satisfies the normality conditions stated in Footnote 3.
' J. Silver has found how to make the construction work also for K singular (assuming
GCH). Thus we have one more consistency-proof for (e) in Theorem, Section 1.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
8 THOMAS J. JECH

SUBLEMMA. If p(y) < p(x) and x-y = 0 then ](x)-y = 0.


PROOF. Case I. p(x) = a + 1: x-y = 0 means -y > x, so that -y > j(x), by
definition of j, since -y e Baz.
Case II. p(x) a limit ordinal, is trivial.
For every x, the sequence x, jxj2x,**_ is finite, since p(x) > p(jx) > p(j2x) >
Let s(x), a finite subset of wo, be the set {p(x), p(jx), p(j2x),.. .}. By Lemma 2
there is an uncountable Z c W and a finite S ' w1 such that s(x) n s(y) = S
for any distinct x, y E Z. Let a = max S (S is nonempty since 0 E s(x) for each x);
if x E Z, let x(a) be the element of {x,]x,12x,} whose p is a. Using the Sublemma,
it can be proved that x(a) y(a) = 0 for any distinct x, y E Z. Thus we have an
uncountable set {x(a): x E Z} of pairwise incompatible elements of Ba, contrary
to our assumption.
Finally, here is the key device for killing Suslin trees:
LEMMA 5. If T is a Suslin tree then there is a c.c.c. algebra B (of power 2Vo) s
that V(B) 1 (T is not ST).
PROOF. Take the elements of T as forcing conditions with the inverse ordering,
i.e. consider the set (P, <) of forcing conditions defined by (P. <) = (T, >). It
is easy to see that any generic G c P is an wl-branch in T and that P is c.c.c. (
T is a ST). If we let B be the algebra determined by P., B is c.c.c. and V(B) I= (T
an w1-branch). By c.c.c. we have JBj = -Plo = 2$o.
Everything is now prepared to construct the model. For simplicity, let us as-
sume that 2K1 = X2 in V. We shall construct a sequence Bo C B1 ' .-. '
Ba c ..., a < co)2 of c.c.c. algebras such that IBal ? 2 for each a < C02; then
we let B = MiX-_-2 Ba. The intention is that there is no ST in V(B). As a matter
of fact, it will be sufficient if

V(B) k (w1, R) is not ST

for any relation R e V(B) on co,.


Let a -* (u(a), v(ac)) be some reasonable 1-1 correspondence between o2 and
C2 X C02- Since every Ba has power at most X2, every V(Ba) has K2 subsets
thus let Rax be the u(a)th relation on coi in the model V(BV(,)).
Having constructed Ba, we let BX+1 =2 Ba be a c.c.c. algebra of power :52
such that

V(Ba+i) k (col, Ra) is not ST;


such an algebra exists by Lemma 5 and the remark on "double forcing". If a
is a limit ordinal then we simply let B, = lim,_A B.; B., is c.c.c. and clearly,
B,\| < X2-
Finally we let B = lima. 2 B.. Since B is c.c.c., the cardinals are absolute. If
R E V(B) is a relation on co, then, again by c.c.c., there is fi < (02 such that
R e V(BB). Hence R = Ra for some a < ?02, and so (co,, R) is not ST in V(Ba+1).
But, if a tree is not Suslin in V(Ba+ ) then it cannot be Suslin in V(B), for V(B) O
V(Ba + 1).
Thus we have constructed a model where there is no ST. Using the above con-
struction, it is possible to establish the consistency of ZFC + SH + 2Ko = K2.
In [25], Jensen proved that ZFC + SH + 2Ko = Ml is also consistent.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
TREES 9

?3. Kurepa trees. Among cow-trees which have countable levels there are some
which have cwl-branches and some which have not; we may ask the question, how
many col-branches can these trees have. Certainly, 241 is an upper bound. But,
although it is not difficult to construct a tree with Ml co,-branches, nothing w
known until recently about trees with more than Ml col-branches. Such trees a
called Kurepa trees (KT). Kurepa's conjecture (KC) is that Kurepa trees exist.
The problem has been settled by showing that both KC and -KC are con-
sistent (the latter relative to the consistency of inaccessible cardinals). Yet more
information about KC has been given recently by Solovay; we shall discuss it later.
Before giving more details about the work concerning KC, let us mention that
the existence of a KT is equivalent to the existence of a Kurepa family, a family
W of subsets of co, such that
(a) I WI > X1,
(b) {S r a: S e W} is countable for all a < col.
In fact, we shall use only one implication: if there is a Kurepa family then there
is a KT. For, consider the characteristic functions f: cwl -{. (0, I} of elem
W and build up a tree of their restrictions flec.
The first result about KC was when several people (e.g. Levy, Rowbottom [11],
Bukovsk' [1]) realized the following:
If cwl is inaccessible in L and if no ordinal between col and W2 is a cardinal in
then KC is true.
(For, the family W = {S c co : S e L} is then a Kurepa family.)
This proves the consistency of KC relative to ZFC + "there exists an inac-
cessible cardinal". Later, Rowbottom and Stewart established the consistency of
KC relative to ZFC. Silver then proved that -KC is consistent relative to ZFC +
"there is an inaccessible cardinal". Recently, Solovay proved that KC is true
whenever V = L[X] for some X c cwl. (We shall give the proof in ?4.) This shows
that the assumption of inaccessibility in Silver's proof cannot be dropped; for,
an immediate consequence of Solovay's result is the following observation of
Silver:
If W2 is not inaccessible in L then there exists a Kurepa tree.
PROOF. We shall deduce the theorem from the above-mentioned Solovay's
result. To begin with, there is an ordinal a, co, < a < cw2, such that no
between a and W2 is a cardinal in L. Also, there is a set X ' w1 such that every
countable ordinal is countable in L[X] and a. has power XI in L[X]. This m
that w1LEXJ = co, and w2LCXJ = C02. There exists a tree Te-L[X] which is
in L[X]. It follows that T is a KT in V, since its length is Wlo1t] = wo, every level
is (countable)LEX3 = countable and T has at least XL[X] = X2 w1-branches.
A model where there is a KT (Stewart [18]). We assume that CH holds in the
ground model and consider the following forcing conditions:
(3.1) a forcing condition is a pair (T, ) where T is a countable tree, as in (2.4),
and f is a 1-1 function whose domain is a countable subset of wt2 and
whose values are a-branches in T (a is the length of T); a condition (T1, fl)
is stronger than (T2,f2) if T1 is an end-extension of 7T2, if dom (f])
dom (J2) and iff(i) 2 f2(i) for each i E dom (f2).

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
10 THOMAS J. JECH

A generic set of conditions determines an wl-tree $7- with countable levels and a
1-1 mapping $7' of2 onto a set of cwl-branches of S-. So $7 is a KT, provided the
cardinals are absolute. The set of forcing conditions is obviously countably closed,
so that co1 is absolute. To prove that all other cardinals are absolute, it suffices
to know that every family of pairwise incompatible conditions has power at most
k1. This can be proved using the following combinatorial lemma, which is a
straightforward generalization of Lemma 2.
LEMMA 6. Assume CH. If W is a family of countable subsets of w2 and if I
X2 then there is a subfamily Z c W of power g2 and a countable set S C w2
that X n Y =S for any pair X, Y of distinct elements of Z.
Thus, given a family W of conditions, I WI = X2, we can find Z c W, IZ
and S C @2 such that dom (fA) n dom (f2) = S whenever (Tl,fi), (T2,f2) are dis-
tinct conditions in Z. By CH, the number of countable trees as well as of branches
in these trees is Ml and hence there is Z1 ' Z, JZ1 = I 2, such that T1
and f1JS =f21S whenever (T1,f1), (T2,f2) eZ1. Such conditions, however, are
compatible. Thus any set of pairwise incompatible conditions has power at most
N1.
It should be remarked that if we change the definition of forcing conditions as
to take the domain of functions f to be a subset of a given cardinal K (instead of
w2), then we get a KT which has at least K w,-branches. (Again, any set of pair-
wise incompatible conditions has power at most b1.)
While writing this survey, we observed that the following holds in the above
model: The generic Tree $-f has no w1-branches other than those which are the
values of the generic function A*.
For, suppose that b is a branch in $7 which is not a value of -I and let (To, fo)
be a condition. We construct a sequence (To0,o) > (Tl,fJ) > (T2,f2) > *-- of
conditions and a sequenceal, a2, O3,*** of ordinals. Let n ~-* (u(n), v(n)) be a 1-1
correspondence between w and co x a. Having constructed (T1,fn) and a,, we
let an+l and (Tn+l,fn+1) be such that (a) an +1 0 T,,, (b) an+l e n+1 (c)
(Tn + 1, fn + 1) I Ban + 1 E b, (d) an + 1 fn + 1(i), where i is the v(n)th element of
domain of fu(nT). We let (T.,f,) limnc (Tnfn) and let b denote the unique a-
branch of T0,, (a is the length of T.,) containing all the points an, n E co. Since b
is not among the values Off.D, there is (Tf) < (T.JJ such that b "disappears"
just before the (a + 1)th level of T. Hence (T,f) I Fb = b, and consequently,
b is countable.
Thus, we can construct a model where there is a KT with a prescribed number
of w1-branches. In particular, we can arrange things so that the number of w
branches is greater than Ml but less than 2K1.
A model where there is no KT (Silver [15]). We assume that there exists a
strongly inaccessible cardinal in the ground model. We use Levy-type (collapsing)
forcing (P. <?) which makes K be w2 in the extension:
A forcing conditioned P is a function whose domain is a countable sub-
set of K X col and such thatf(a, f3) < a for all a < Kf, < Wl;
the ordering c of conditions is by inverse inclusion.

Let G be a P-generic set. In V[G], wol is absolute, since the set of forcing condit

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
TREES 11

is countably cl
Ml in the exte
incompatible conditions has power less than K (the K-chain condition); hence
K = gXV(G]
We first state the key lemma:
LEMMA 7. If a set (P, <) of forcing conditions is countably closed and if T is
an w1-tree in the ground model V and has countable levels, then in the extension, T
has no co,-branches other than those which are in V.
PROOF. Suppose that b is an u,-branch of T which is not in V and let Po be
a condition. We shall construct, by recursion, conditions ps < po and elements
of x, of T, for all finite sequences s of O's and ls. Having constructed ps, we can
find two points x, -o # x,-l, at the same level of T and two conditions p,-^o and
p.,, both stronger than p,. such that pa, I Fxi e b. There exists a < co, such that
x, e Tia for all s. For each f: co -* {0, 1}, there exists pf, a lower bound of all
pflns n e co; since b is uncountable, there exists Xf at the ath level such that q I Ixf X b
for some q < pf. It is clear that xf # x, iff #A g, so that we have 21o points at the
ath level of T; a contradiction.
By the lemma, we see immediately that no tree T e V can be a KT in the exten-
sion. For, the number of col-branches of T in V is at most 2Ko, which is less than
K, hence less than X2 in V[G]. Fortunately enough, the same argument can be
used also for trees which are not in V. Let T be an co,-tree in V[G], with counta
levels. T may be considered as a relation on Co1. By the K-chain condition, there
exists a < K such that T e V[G n Pj where Pa is the set of all functions in P
whose domain is a subset of a x co1; similarly, P" consists of all functions fe P
such that dom (f) c (Ka- a) x co,. Now we use a fact from the theory of f
which says that V[G] = V[G n Paj[G fir Pa] and G nm Pa is Pa-generic over
V[G Cn Pal. Moreover, Pa is countably closed in V[G Cn Pa] (this is easy to see).
Hence we may use the lemma, considering V[G n Pa] as the ground model. It
follows that the number of co,-branches of T is at most the size of 2K1 in V
but this is still not enough, since K is inaccessible in V[G r) Pa]. Hence there is
no KT in V[G].

?4. Trees in L. Having described models where there are Suslin and Kurepa
trees and models where there are none, we present the cream of the article: we
shall show that in the constructible universe L both Suslin and Kurepa trees exist.
Moreover,
TiEOREM. If V = L[X] for some X c C1 then both Suslin trees and Kurepa
trees exist.
The result on Suslin trees is due to Jensen [5], the result on Kurepa trees is due
to Solovay [24] (to be published).
Construction of a ST in L (Jensen). We shall construct a (normal) Suslin tree
T by constructing its levels U.e, by recursion. For simplicity, the eleme
will be countable ordinals. It is obvious how to construct U,,+,,, having con-
structed Ua. The only problem is, how to construct Ua if a is a limit ordinal which
a-branches in Tla to extend and which to omit.
Specific properties of the constructible universe enable us to carry out the

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
12 THOMAS J. JECH

construction in such a way that T has no uncountable antichain. The idea is to kill
off eventually the maximal antichains in the partial trees Tla, for limit ordin
We have seen in Lemma 3 how to kill a single maximal antichain A: how to con-
struct U. such that A is still maximal in T I (a + 1) (and, consequently, in T). It
turns out that we can be more efficient and kill countably many maximal anti-
chains:
If S is a countable family of maximal antichains in TJ a then we can construct
U, such that each A E S is maximal in T j (a + 1) (and hence in T).
The proof goes as in Lemma 3: for every x E T, one easily finds an a-branch
through x which meets each A E S; these branches are to be extended.
Construction of limit levels of T. There exists a function f: L1 c,-* such
that for each a, < w1,
<Lf(<,,), e> = a is countable
(i.e. there exists g e Lf(a) which maps w onto a). We choose such a function f
and for every limit ordinal a we construct Ua such that we kill all maximal anti-
chains in Tla which are in Lf(a).
It remains to prove that T has no uncountable antichain. Suppose that A is a
maximal antichain in T. Having in mind that both T and A belong to LV2
sider the structure
X L= LC2S C,1, T. A>*
By Skolem-L6wenheim Theorem, there is a countable elementary submodel

X = <NEn, T, A>,
such that co, rl N E 1, i.e. a e N implies a c N for all a, < w1. Clea
sitive realization of X has the form

a' = <L, E, a, Tja, A n>,


where a = w1l n N., and /3 is some countable ordinal. Since
A" P ca is uncountable,
whereas
Lf(a,,) 0 a is countable,
it follows that 8 < f(a). Consequently, A n a eG Lf(c); this means that A n a has
been killed at the ath level. Therefore, A = A n a and A is countable.
The proof of -SH in L[X], X C co,, follows the same pattern; we only n
distinguish two cases:
Case I. If CLlXrna] is countable for all a < co, then we let f(a) = 2
(which is also countable).
Case II. If CwlL[Xrnao = co, for some ao then we define f(a,) for a 2 ao
that a is countable in <Lf(a,)[X ri a], e>. In either case, we kill the maximal anti-
chains which are in Lf(a,)[X n a], and in the proof that T is Suslin, we employ the
structure
.f = <Lw2[X], i, cov,,X, T. A>.
Construction of a KT in L (Solovay). The main idea is approximately the same
as above; some details are, however, more subtle. For the construction we use the
following function f: co,1 oil.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
TREES 13

(4.1) f(ca) is the least ordinal such that


(a) a ELf(,
(b) <Lf(,), E> is elementarily equivalent to <L.1. e>.
We shall prove that the family
W = cS C oil: (Va < il)[S fl a - Lf(aj]}
is a Kurepa family. To show that, it obviously suffices to show I WI =X
By way of contradiction, suppose that I WI < X1. A useful tool in the
the "elementary tower"

.00 -< 1 < * *co,<*** e<@)


of submodels of ., = <L,,2, e>, constructed as follow
submodel of X' such that (A) n JO c co,, X<': + 1 is
of .X such that XX c :4'f+1 and cowl n W +,1 E co,;
A = U4<A\&. Let
U = {W(): 0 < w1},
where a(5) = c1 r at; U is a closed unbounded subset of co, (closed means:
sup (U Ci)e U for all a < wl).
Clearly, the function f is definable in X4 = <LC2, E>; since we suppose that
I WI < Xi, there is an enumeration C = <So: e < wj> of W, which is definable
in .,. We shall derive a contradiction by showing that the set
S = {a(0: a() 0 Se,
which is obviously not in C, belongs to W. We want to show that S n a E L
for every a < wl. By induction, this is easy if a is not a limit point of U; for then,
S n <a differs at most in one element from S n a(s), where a(e) is the greatest
element of U n a.
If a = ca(eo) is a limit point of U then
S n a = {a(): a() <a & a()S, a}.
This set is definable in Lf(a,,) from Cal = <So r' a: e < a> a
Thus it suffices to show that both C,. and Use are in Lf(a).
Let XA" = <LB, e> be the transitive realization of &X0. As before, since a is un-
countable in XA" and countable in Lf(,) we have /3 < f(a), and moreover, Lo e Lf(,X)
(by (4.1)). The set C is definable in Lc,,2, so that C E Xc,. Clearly, the image of
C under the transitive realization of &.X is C,,, and so Ca,, e AY. Since X c' Lf(a),
we have Cr e Lf(a). Use is also in Lf(,); we can obtain Use from a and Lo by con-
structing, in Lf(ar), the elementary tower of submodels of <LO, e>. Hence we have
S E W and S 0 C, a contradiction. Thus I WI = 92.
The proof of KC in L[XJ, X c co,, is analogous. Again, we have to distinguish
the same two cases as for SH. Now we define f(ca) either by setting f(a) = CO3LLAal]
or by taking the least ordinal such that Lf(,)[A l a] LO,)1[A n a]. The construc-
tion is then modified accordingly.

REFERENCES

(1] L. BUKOVSKY, Consistency theorems connected with some combinatorial problems, Com-
mentationes Mathematicae Universitatis Carolinae, vol. 7 (1966), pp. 495-499.

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms
14 THOMAS J. JECH

[2] P. J. COHEN, Set theory and


1966.
[3] K. M. HRBA6EK, A note on generalized Suslin's Problem, Commentationes Mathematicae
Universitatis Carolinae, vol. 8 (1967), pp. 307-309.
[4] T. J. JECH, Non-provability of Suslin's hypothesis, Commentationes Mathematicac Uniwer-
sitatis Carolinae, vol. 8 (1967), pp. 291-305.
[5] R. B. JENSEN, Notices of the American Mathematical Society, vol. 15 (1968), 935.
[6] H. J. KEISLER and A. TARSKI, From accessible to inaccessible cardinals, Fundamenta
Mathematicae, vol. 53 (1963-64), pp. 225-308.
[71 D. KUREPA, L'hypothese de ramification, Comptes Rendus Hebdomadaires des Seances
de l'Academie des Sciences. Series A et B, vol. 202 (1936), pp. 185-187.
[8] E. MARCZEWSKI (E. Szpilrajn), Separabilite et multiplication cartesienne des espaces
topologiques, Fundamenta Mathematicae, vol. 34 (1947), pp. 127-143.
[9] E. W. MILLER, A note on Suslin's Problem, American Journal of Mathematics, vol. 65
(1943), pp. 673-678.
[10] K. L. PAiKRi, Ph.D. Thesis, University of California, Berkeley, 1968.
[1 1] F. ROWBOTrOM, Ph.D. Thesis, University of Wisconsin, Madison, 1965 (notes for 1967).
[12] D. Scorr, Lectures on Boolean-valued models for set theory, Proceedings of the Summer
Institute on Set Theory, U.C.L.A., 1967 (to appear).
[13] W. SIERPI*sKm, Sur un problem de la theorie generate des ensembles equivalent au prob-
lkme de Souslin, Fundamenta Mathematicae, vol. 35 (1948), pp. 165-174.
[14] J. SILVER, Ph.D. Thesis, University of California, Berkeley, 1966.
[15] , The independence of Kurepa's conjecture and two-cardinal conjecture in model
theory, Proceedings of the Summer Institute on Set Theory, U.C.L.A., 1967 (to appear).
[16] R. M. SOLOVAY and S. TENNENBAUM, Iterated Cohen extensions and Suslin's Problem
(to appear).
[17] E. SPECKER, Sur un problem de Sikorski, Colloquium Mathematicum, vol. 2 (1951),
pp. 9-12.
[18] D. H. STEWART, M.Sc. Thesis, Bristol, 1966.
[19] M. SOUSLIN, Probkme 3, Fundamenta Mathematicae, vol. 1 (1920), p. 223.
[20] S. TENNENBAUM, Suslin's Problem, Proceedings of the National Academy of Sciences
of the United States of America, vol. 59 (1968), pp. 60-63.
[21] P. VOP9NKA and P. HAJEK, Sets, semisets, models, North-Holland (to appear).
[22] D. KUREPA, On A-trees, Publications de l'Institut Mathematique, vol. 8 (22), (1968),
pp. 153-161.
[23] H. GAIFMAN and E. P. SPECKER, Isomorphism types of trees, Proceedings of the American
Mathematical Society, vol. 15 (1964), pp. 1-7.
[24] R. SOLOVAY, forthcoming paper on Kurepa trees.
[25] R. JENSEN, SH is compatible with CH, (mimeographed).
[26] , SH = weak compactness in L, (mimeographed).

CHARLES UNIVERSITY
PRAGUE, CZECHOSLOVAKIA

This content downloaded from 143.107.231.30 on Tue, 13 Nov 2018 00:01:41 UTC
All use subject to https://about.jstor.org/terms

You might also like