Professional Documents
Culture Documents
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
A COMMENT ON p < t
SAHARON SHELAH
Abstract. Dealing with the cardinal invariants p and t of the continuum we
prove that
m= p =
2
t =
2
.
In other words, if MA
1
(or a weak version of this) holds, then (of course
2
p t and) p =
2
p = t. The proof is based on a criterion for p < t.
0. Introduction
We are interested in two cardinal invariants of the continuum, p and t. The
cardinal p measures when a family of innite subsets of with nite intersection
property has a pseudointersection. A relative is t, which deals with towers, i.e.,
families well ordered by almost inclusion. These are closely related classical cardinal
invariants. Rothberger [7], [8] proved (stated in our terminology) that p t and
p =
1
p = t,
and he asked if p = t.
Our main result is Corollary 2.5 stating that
m p =
2
p = t,
where m is the minimal cardinal such that Martin Axiom for dense sets fails
(i.e. MA
). Considering that m
1
is a theorem (of ZFC), the parallelism
with Rothbergers theorem is clear. The reader may conclude that probably m =
p p = t; this is not unreasonable but we believe that eventually one should
be able to show CON(m = + p = + t =
+
). In the rst section we present a
characterization of p < t which is crucial for the proof of 2.5, and which also sheds
some light on the strategy to approach the question of p < t presented in [9].
We thank Andreas Blass for detailed comments on an earlier version and David
Fremlin for historical information.
Notation Our notation is rather standard and compatible with that of clas-
sical textbooks (like Bartoszy nski and Judach [3]). In forcing we keep the older
convention that a stronger condition is the larger one.
(1) Ordinal numbers will be denoted be the lower case initial letters of the Greek
alphabet (, , , . . .) and also by i, j (with possible sub- and superscripts).
(2) Cardinal numbers will be called ,
i
, .
Date: December 2006.
1991 Mathematics Subject Classication. Primary 03E17; Secondary: 03E05, 03E50.
The author acknowledges support from the United States-Israel Binational Science Foundation
(Grant no. 2002323). Publication 885.
1
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
2 SAHARON SHELAH
(3) A bar above a letter denotes that the considered object is a sequence;
usually
X will be X
i
: i < ), where is the length g(
X) of
X. Sometimes
our sequences will be indexed by a set of ordinals, say S , and then
X
will typically be X
: S).
(4) The set of all innite subsets of the set of natural numbers is denoted by
[]
0
and the relation of almost inclusion on []
0
is denoted by
. Thus
for A, B []
0
we write A
and <
. Thus, for f, g
,
f
g if and only if (
g if and only if (
, <
B for all B B.
(2) A sequence X
: < ) []
0
is a tower if X
(-centered)
fails,
Szyma nski proved that p is regular (see, e.g., Fremlin [5, Proposition 21K]),
Piotrowski and Szyma nski [6] showed that t add(/) (so also t b).
For more results and discussion we refer the reader to [3, 1.3, 2.2].
Denition 1.2. We say that a family B []
0
exemplies p if:
B is closed under nite intersections (i.e., A, B B A B B), and
B has no pseudointersection and [B[ = p.
Proposition 1.3. Assume p < t and let B exemplify p. Then there are a cardinal
= cf() < p and a
decreasing sequence A
i
: i < ) []
0
such that
(a) A
i
B is innite for every i < and B B, and
(b) if A is a pseudointersection of A
i
: i < , then for some B B the
intersection A B is nite.
Proof. Fix an enumeration B = B
i
: i < p. By induction on i < p we try to
choose A
i
[]
0
such that
() A
i
A
j
whenever j < i,
() B A
i
is innite for each B B,
() if i = j + 1, then A
i
B
j
.
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
A COMMENT ON p < t 3
If we succeed, then A
i
: i < p has no pseudointersection so t p, a contradiction.
So for some i < p we cannot choose A
i
. Easily i is a limit ordinal and let = cf(i)
(so i < p). Pick an increasing sequence j
decreasing and
(iii) for each i < and n < the intersection A
i
B
n
is innite, and
(iv)
_
i < j <
__
n <
__
A
j
B
n
A
i
B
n
_
.
Then for some A []
0
we have
_
i <
__
A
A
i
_
and
_
n <
__
A
B
n
_
.
Proof. Without loss of generality B
n+1
B
n
and =
B
n
: n < (as we may
use B
n
=
n
B
f
_
and n < f(n) < f(n + 1)
for n < . Let
B
=
_
(B
n+1
[n, f(n + 1)) : n < .
Then B
[]
0
as for n large enough,
min[B
n+1
[0, n) A
0
] f
0
(n + 1) < f(n + 1).
Clearly for each n < we have B
[0, f(n)) B
n
and hence B
B
n
. Moreover,
(i < )(A
i
B
[]
0
) (as above) and (i < j < )(A
j
B
A
i
B
)
(remember assumption (iv)). Now applying t > to A
i
B
: i < ) we get a
pseudointersection A
which is as required.
Denition 1.6. (1) Let S be the family of all sequences =
n
: n B)
such that B []
0
, and for n B,
n
[n,k)
2 for some k (n, ).
We let dom( ) = B and let set( ) =
set(
n
) : n dom( ), where
set(
n
) = :
n
() = 1.
(2) For
A = A
i
: i < ) []
0
let
S
A
=
_
S :
_
i <
__
set( )
A
i
_
and
_
n dom( )
__
set(
n
) ,=
__
.
(3) For , S let
n)(
n
).
Observation 1.7. (1) If
, then
which implies C
C
.
(2) For every S and a meagre set B
2 there is S such that
and C
B = .
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
4 SAHARON SHELAH
Lemma 1.8. (1) If
A = A
i
: i < i
) []
0
has nite intersection property
and i
< p, then S
A
,= .
(2) Every
-
upper bound.
(3) If
A = A
i
: i < i
) []
0
is
decreasing and i
-
increasing sequence of members of S
A
of length < p has an
-upper bound
in S
A
.
Proof. (1) Let A []
0
be such that (i < i
)(A
A
i
) (exists as i
: < ) be a
=: dom(
)
for < . Then A
: < ) is a
[]
0
such that < A
=
_
: for some n dom(
) we have
n
_
.
One easily veries that the assumptions of Proposition 1.5 are satised (well, re-
placing by B
0
!!). Let A B
0
be given by the conclusion of 1.5, and put
A
=
_
n : for some A we have
_
[n,k)
2 : k (n, )
_
.
Plainly, the set A
is innite. We let
=
n
: n A
) where
n
is any member of
A B
n
B
n+1
.
(3) Assume that
A = A
i
: i < i
) []
0
is
decreasing, i
< p, and
: <
) S
A
is
is nite.
The order
P
= of P is given by p q if and only if (p, q P and)
(i) u
p
u
q
, w
p
w
q
, a
p
a
q
, and
q
u
p
=
p
,
(ii) max(u
p
) < min(u
q
u
p
) and for n u
q
u
p
we have
(iii) ( w
p
)(n dom(
n
q
n
),
(iv) (i a
p
)(set(
q
n
) A
i
).
Plainly, P is a centered forcing notion and the sets
1
,i
m
=
_
p P : w
p
i a
p
[u
p
[ > m
_
(for < , i < i
[ +
0
< p, we may
choose a directed set G P meeting all the sets 1
,i
n
. Putting =
p
: p G
we will get an upper bound to
: < ) in S
A
.
Lemma 1.9. Assume that
(i) p < t and B = B
: p) such that
(a)
S
A
for < p and
p
S (sic!),
(b)
: p) is
-increasing,
(c) if < p and n dom(
+1
) is large enough, then set(
+1
n
) B
,=
(hence (
n dom(
))(set(
n
) B
,= and set(
+1
n
) B
,= for
n dom(
+1
), and the truth values of
min(set(
+1
n
) B
) < min(set(
+1
n
) B
)
are the same for all n dom(
+1
),
(e) in (d), if < we can replace B
by A
[]
0
such that B
and
(i < )(B
A
i
). Next, for n dom(
) we choose
n
such that
n
and
,= :
n
() = 1 and g(
n
) < g(
n
) B
.
Then we let
+1
=
n
: n dom(
: <
2
)
_
of sequences of functions in
such
() (i < j <
1
)(f
j
<
f
i
),
() ( < <
2
)(f
<
),
() (i <
1
)( <
2
)(f
<
f
i
),
() if f : is such that (i <
1
)(f
f
i
), then f
for some
<
2
,
() if f : is such that ( <
2
)(f
f), then f
i
f for some
i <
1
.
Proposition 1.11. If
2
< b, then there is no (
0
,
2
)peculiar cut.
Proof. Assume towards contradiction that b >
2
but there is an (
0
,
2
)peculiar
cut, say
_
f
i
: i < ), f
: <
2
)
_
is such a cut. Let S be the family of all
increasing sequences n = n
i
: i < ) with n
0
= 0. For n S and g
we say
that n obeys g if (i < )(g(n
i
) < n
i+1
). Also for n S dene h
n
by
h
n
[n
i
, n
i+1
) = f
i
[n
i
, n
i+1
) for i < .
Now, let g
be an increasing function such that for every n < and m g
(n)
we have
f
n+1
(m) < f
n
(m) < . . . < f
1
(m) < f
0
(m).
Note that
()
1
if n S obeys g
f
i
).
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
6 SAHARON SHELAH
Now, for <
2
dene g
by
()
2
g
(n) = min
_
k < : k > n + 1
_
i n
__
[n, k)
__
f
() < f
i
()
__
.
Since
2
< b, we may choose g
such that
g
<
g).
Pick n S which obeys g and consider the function h
n
. It follows from ()
1
that
h
n
<
f
i
for all i < , so by the properties of an (
0
,
2
)peculiar cut there is
<
2
such that h
n
[n
i
, n
i+1
), and
n
i
< g
(n
i
) < g(n
i
) < n
i+1
.
The latter implies that for some [n
i
, n
i+1
) we have f
() < f
i
(), contradicting
the former.
Theorem 1.12. Assume p < t. Then for some regular cardinal there exists a
(, p)peculiar cut in
and
1
< p.
Proof. Use Proposition 1.3 and Lemma 1.9 to choose B, ,
A, pr and
: p)
so that:
(i) B = B
: < p exemplies p,
(ii)
A = A
i
: i < ) []
0
is
: < p)
A
(as by renaming, A
is ). Let
(v) A
= dom(
p
),
(vi) for i < we let f
i
: A
be dened by
f
i
(n) = min
_
: [
p
n
(n +) = 1 n + / A
i
] or dom(
p
n
) = [n, n +)
_
,
(vii) for < p we let f
: A
be dened by
f
(n) = min
_
+ 1 : [
p
n
(n +) = 1 n + B
] or dom(
p
n
) = [n, n +)
_
.
Note that (by the choice of f
i
, i.e., clause (vi)):
(viii)
[n, n +f
i
(n)) set(
p
n
) : n A
A
i
for every i < .
Also,
()
a
1
f
j
f
i
for i < j < .
[Why? Let i < j < . Then A
j
A
i
and hence for some n
we have that
A
j
n
A
i
. Therefore, for every n A
, in the denition of f
i
, f
j
in clause
(vi), if can serve as a candidate for f
i
(n) then it can serve for f
j
(n), so (as we use
the minimum there) f
j
(n) f
i
(n). Consequently f
j
f
i
.]
Now, we want to argue that we may nd a subsequence of f
i
: i < ) which is
<
f
i
.
So assume towards contradiction that for some i() < we have
(j)(i() < j < (f
j
<
f
i()
)).
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
A COMMENT ON p < t 7
For j < put B
j
=: n A
: f
j
(n) f
i()
(n). Then B
j
[A
]
0
is
of B
j
: j < ) (so B
[A
]
0
and (j < )(B
j
)). Now, let A
set(
p
n
) [n, n +f
i()
(n)) : n B
.
() A
is an innite subset of .
[Why? Recall that by 1.9(a) we have
0
S
A
and hence set(
0
)
A
i()
and
(n dom(
0
)(set(
0
n
) ,= ) (see Denition 1.6(2)). By 1.9(b) we know that for
every large enough n dom(
p
) we have n dom(
0
) and
0
n
p
n
. For every large
enough n dom(
0
) we have set(
0
) 0, . . . , n 1 A
i()
, and hence for every
large enough n dom(
p
) we have
0
n
p
n
and ,= set(
0
n
) A
i()
. Consequently,
for large enough n B
, [n, n +f
i()
(n)) set(
p
n
) ,= and we are done.]
() A
A
j
for j (i(), ) (and hence for all j < ).
[As f
j
B
f
i()
B
,= and set(
+1
n
) A
i()
.]
Properties ()( ) contradict (b) of 1.3, nishing the proof of ()
b
1
.
Thus passing to a subsequence if necessary, we may assume that
()
c
1
the demand in () of 1.10 is satised, i.e., f
j
<
f
i
for i < j < .
Now,
()
2
(i < )( < p)(f
<
f
i
).
[Why? Let i < , < p. For large enough n A
,= and
+1
n
p
n
. Then for those n we have f
(n) f
i
(n).
Now, remembering ()
1
, we may conclude that actually f
<
f
i
.]
()
a
3
The set (of functions) f
i
: i < f
,
()
b
3
in fact, if f
, f
or f
<
or
f
<
.
[Why? By ()
1
, ()
2
and clauses (d) + (e) of 1.9.]
Choose inductively a sequence = () : <
) p such that:
() is the minimal p () : < satisfying ( < )(f
()
<
),
and
we cannot choose (
).
We ignore (till (
7
)) the question of the value of
. Now,
()
4
f
i
: i < ), f
()
: <
and
_
i <
__
f
f
i
_
but
_
<
__
(f
f
()
)
_
.
Clearly, without loss of generality, we may assume that [n, n + f(n)) dom(
p
n
)
for n A
. Let A
=
_
[n, n + f(n)) set(
p
n
) : n A
_
. Now for every i < ,
A
A
i
because f
f
i
and by the denition of f
i
. Also, for every < p, the
intersection A
we have (f
()
<
) , and thus f
f
()
(remember
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
8 SAHARON SHELAH
()
3
). Hence, if n A
(n) f
()
(n) and for innitely
many n A
we have f
(n) f
()
(n) < f(n) f
0
(n) [dom(
p
n
)[. For every
such n we have n+f
(n) 1 A
. Together, A
, and
_
<
__
f
()
f
_
but
_
i <
__
(f
i
f)
_
.
It follows from ()
1
(and the assumption above) that we may choose an innite set
A
such that
_
i <
__
(f A
) <
(f
i
A
)
_
. Let
A
=
_
[n, n +f(n)) set(
p
n
) : n A
.
Since (f A
) <
(f
i
A
A
i
for all i < (remember
(viii)). Like in the justication for ()
5
above, if < p then for some <
we
have f
f
()
and we may conclude from our assumption towards contradiction
that f
: < p).]
()
7
= p.
[Why? The sequence () : < p) is an increasing sequence of ordinals < p, hence
p. If
: <
2
) of open dense subsets of Q, there is no
directed G Q such that ( <
2
)(G 1
,= ). Hence MA
2
(centered) fails
and thus p
2
.
Proof. Let
_
f
i
: i <
1
), f
: <
2
)
_
be a (
1
,
2
)peculiar cut in
. Dene a
forcing notion Q as follows.
A condition in Q is a pair p = (, u) such that
>
and u
1
is nite.
The order
Q
= of Q is given by (
1
, u
1
) (
2
, u
2
) if and only if (both are in
Q and)
(a)
1
2
,
(b) u
1
u
2
,
(c) if n [g(
1
), g(
2
)) and i u
1
, then f
i
(n)
2
(n).
Plainly, Q is a forcing notion of cardinality
1
. It is centered as for each
, the set (, u) Q : = is directed.
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
A COMMENT ON p < t 9
For j <
1
let 1
j
= (, u) Q : j u, and for = +n <
2
let
1
=
_
(, u) Q :
_
m < g()
__
m n (m) > f
(m)
__
.
Clearly 1
j
, 1
, 1
j
for j <
1
, <
2
. Put g =
:
(u)((, u) G). Then
g is a function, its domain is (as G 1
n
,= for n < ), and
g
f
i
(as G 1
i
,= ), and
n < : f
: <
2
).
Corollary 2.2. If there exists an (
0
,
2
)peculiar cut, then cov(/)
2
.
Theorem 2.3. Let cf(
2
) =
2
>
1
. Assume MA
1
holds. Then there is no
(
1
,
2
)peculiar cut in
.
Proof. Suppose towards contradiction that cf(
2
) =
2
>
1
,
_
f
i
: i <
1
), f
:
<
2
)
_
is an (
1
,
2
)peculiar cut and MA
1
holds true. We dene a forcing
notion Q as follows.
A condition in Q is a pair p = (u, ) = (u
p
,
p
) such that
(a) u
1
is nite, =
i
: i u) =
p
i
: i u),
(b) for some n = n
p
, for all i u we have
i
n
,
(c) for each i u and m < n
p
we have
i
(m) f
i
(m),
(d) if i
0
= max(u) and m n
p
, then f
i0
(m) > 2 [u
p
[ + 885.
(e) f
i
[n
p
, ) : i u) is <decreasing.
The order
Q
= of Q is given by p q if and only if (p, q Q and)
(f) u
p
u
q
,
(g)
p
i
q
i
for every i u
p
,
(h) if i < j are from u
p
, then
q
i
[n
p
, n
q
) <
q
j
[n
p
, n
q
),
(i) if i < j, i u
q
u
p
and j u
p
, then for some m [n
p
, n
q
) we have
f
j
(m) <
q
i
(m).
Claim 2.3.1. Q is a ccc forcing notion of size
1
.
Proof of the Claim. Plainly, the relation
Q
is transitive and [Q[ =
1
. Let us
argue that the forcing notion Q satises the ccc.
Let p
Q for <
1
. Without loss of generality p
: <
1
) is without
repetition. Applying the Lemma we can nd an unbounded set |
1
and
m() < n() < and n
; let u
p
=
,
: < n() and
,
increases
with , and
(ii)
,
=
> n
such that
for each k k
we have
the sequence f
(k) :
,
: < n()
,
: < n()) is strictly
decreasing,
f
,n()1
(k) > 885 (n() + 1), and
f
,m()
(k) +n() + 885 < f
,n()1
(k).
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
10 SAHARON SHELAH
(The choice is possible because f
i
: i <
1
) is <
,n()1
(k) f
,m()
(k)
_
= .)
Now dene q = (u
q
,
q
) as follows:
u
q
= u
p
u
p
, n
q
= k
+ 1,
if n < n
, i u
p
, then
q
i
(n) =
p
i
(n),
if n < n
, i u
p
, then
q
i
(n) =
p
i
(n),
if i =
,
, < n(), n [n
, k
), then
q
i
(n) = , and if j =
,
, m()
< n(), then
q
j
(n) = n() + + 1,
if j =
,
, < n(), then
q
j
(k
) = , and if i =
,
, m() < n(),
then
q
i
(k
) = f
,m()
(k
) + + 1.
It is well dened (as
p
,
=
p
,
for < m()). Also q Q. Lastly, one easily
veries that p
Q
q and p
Q
q, so indeed Q satises the ccc.
For i <
1
and n < let
1
i,n
=
_
p Q :
_
u
p
i or for no q Q we have p
Q
q u
q
i
and n
p
n
_
.
Plainly,
() the sets 1
i,n
are open dense in Q.
Also,
() for each i <
1
there is p
i
Q such that u
pi
= i.
It follows from 2.3.1 that
() for some i(), p
i()
Q
j <
1
: p
j
G
is unbounded in
1
.
Note also that
() if p is compatible with p
i()
and p 1
i,n
then u
p
i.
Since we have assumed MA
1
and Q satises the ccc (by 2.3.1), we may nd a
directed set G Q such that p
i()
G and 1
i,n
G ,= for all n < and i <
1
.
Note that then the set | :=
u
p
: p G is unbounded in
1
.
For i | let g
i
=
p
i
: p G. Clearly each g
i
(as G is directed,
1
i,n
G ,= for i <
1
, n < ). Also g
i
f
i
by clause (c) of the denition of
Q, and g
i
: i |) is <
g
j
f
j
<
f
i
. Thus by property 1.10()
of a peculiar cut, for every i | there is (i) <
2
such that g
i
<
f
(i)
. Let
() = sup(i) : i |. Then () <
2
(as
2
= cf(
2
) >
1
). Now, for each
i | we have g
i
<
f
()
<
f
i
, and thus for i | we may pick n
i
< such that
n [n
i
, ) g
i
(n) < f
()
(n) < f
i
(n).
For some n
the set |
= i | : n
i
= n
is unbounded in
1
. Let j |
be
such that |
(remember
G1
j,n
+1
,= and G is directed). Since u
p
is nite, we may choose i |
j u
p
,
and then q G such that q p and i u
q
. If follows from clause (i) of the denition
of the order of Q that there is n [n
p
, n
q
) such that f
j
(n) <
q
i
(n) = g
i
(n). Since
n > n
= n
i
= n
j
we get f
j
(n) < g
i
(n) < f
()
(n) < f
j
(n), a contradiction.
Remark 2.4. The proof of 2.3 actually used Hausdor gaps on which much is known
(see, e.g., Abraham and Shelah [1], [2]). More precisely, the proof could be presented
as a two-step argument:
8
8
5
r
e
v
i
s
i
o
n
:
2
0
0
7
-
0
5
-
1
0
m
o
d
i
f
i
e
d
:
2
0
0
7
-
0
5
-
1
0
A COMMENT ON p < t 11
(1) from MA
1
one gets that every decreasing
1
sequence is half-a-Hausdor
gap, and
(2) if
2
= cf(
2
) >
1
, then the
1
part of a peculiar (
1
,
2
)cut cannot be
half-a-Hausdor gap.
Corollary 2.5. If MA
1
then p =
2
t =
2
. In other words,
m = p =
2
t =
2
.
References
[1] Uri Abraham and Saharon Shelah. Lusin sequences under CH and under Martins Axiom.
Fundamenta Mathematicae, 169:97103, 2001. math.LO/9807178.
[2] Uri Abraham and Saharon Shelah. Ladder gaps over stationary sets. Journal of Symbolic
Logic, 69, 2:518532, 2004. math.LO/0404151.
[3] Tomek Bartoszy nski and Haim Judah. Set Theory: On the Structure of the Real Line. A K
Peters, Wellesley, Massachusetts, 1995.
[4] Murray G. Bell. On the combinatorial principle P(c). Fund. Math., 114:149157, 1981.
[5] David Fremlin. Consequences of Martins Axiom. Cambrdge U.P., 1984.
[6] Zbigniew Piotrowski and Andrzej Szyma nski. Some remarks on category in topological spaces.
Proc. Amer. Math. Soc., 101:156160, 1987.
[7] Fritz Rothberger. Sur un ensemble toujours de premiere categorie qui est depourvu de la
propriete . Fundamenta Mathematicae, 32:294300, 1939.
[8] Fritz Rothberger. On some problems of Hausdor and Sierpi nski. Fundamenta Mathematicae,
35:2946, 1948.
[9] Saharon Shelah. Large continuum, oracles, and a try at p < t.
[10] Stevo Todorcevic and Boban Velickovic. Martins axiom and partitions. Compositio Mathe-
matica, 63:391408, 1987.
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The Hebrew
University of Jerusalem, Jerusalem, 91904, Israel, and Department of Mathematics,
Rutgers University, New Brunswick, NJ 08854, USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at