5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
Lusin sequences under CH and under Martins
Axiom
Uri Abraham
Departments of Mathematics and Computer Science,
Ben-Gurion University, Beer-Sheva, Israel
Saharon Shelah
Institute of Mathematics
The Hebrew University, Jerusalem, Israel
October 6, 2003
Abstract
Assuming the continuum hypothesis there is an inseparable se-
quence of length
1
that contains no Lusin subsequence, while if Mar-
tins Axiom and CH is assumed then every inseparable sequence (of
length
1
) is a union of countably many Lusin subsequences.
2000 Mathematics Subject Classication . 03E05 (Combinatorial set theory), 03E50
(Continuum hypothesis and Martins axiom), 03E35 ( Consistency and independence re-
sults).
Key words and phrases. Lusin, Martins Axiom, Continuum hypothesis.
The work was carried out in the meeting in Lyon in 1996.
Email: [email protected]
Publication # 537. The author would like to thank the Israel Science Foundation,
founded by the Israel Academy of Science and Humanities.
1
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
1 Preface
We rst x some notations and denitions. The set of natural numbers is
denoted , and for A, B we write A
B i AB is nite, and A B i
AB is nite (almost inclusion, almost disjointness). Let / = A
[
1
)
be a sequence of pairwise almost disjoint, innite subsets of . So A
and
A
1
A
1
for
1
,=
2
. We say that B separates / if
1
[ A
B
and
1
[ A
B are both uncountable. If no B separates / then
/ is said to be inseparable. That is / is inseparable if it is an almost disjoint
family of innite subsets of such that there is no B for which
(
1
A /)(A
B) & (
1
A /)(A
B).
An inseparable family of size
1
can be constructed in ZFC alone (Lusin
[1], cited by [2]). We say that / is a Lusin sequence if for every i <
1
and
n
j < i [ A
i
A
j
n is nite.
A seemingly stronger property is the following. We say that / is a Lusin*
family if for every i <
1
and n
j < i : [A
i
A
j
[ < n
is nite.
It is not dicult to prove that every Lusin sequence is inseparable, and
Lusin constructed a Lusin sequence in ZFC. Is this the only way to build
inseparable families? The answer depends on set-theoretical assumptions
as the following two results show (obtained by the rst and second author
respectively).
Theorem 1.1 (1) CH implies that there is an inseparable family which con-
tains no Lusin subsequence. (2) Martins Axiom+CH implies that every
inseparable sequence is a countable union of Lusin* sequences.
2 Proofs
2.1 CH gives an inseparable non-Lusin sequence
Assume the continuum hypothesis (CH) throughout this subsection. We
shall dene a sequence / = A
[
1
) of almost disjoint subsets of
2
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
which is inseparable by virtue of the following property denoted P.
For every innite X one of the following three possibilities holds:
P1 X is nitely covered by / (which means that for some nite set u
1
X
[ u.
P2 X is nitely covered by /.
P3 For some
0
<
1
for all
0
<
1
X splits A
(which means that
both X A
and A
X are innite).
It is quite obvious that if / satises this property then it is inseparable,
and so we describe the construction assuming CH of a sequence that satises
property P, but does not contain any Lusin subsequence.
Let X
[
1
) be an enumeration of all innite subsets of , and let
e
i
[ i
1
) be an enumeration of all countable subsets of
1
of order-type
a limit ordinal. The sequence / = A
[
1
) is dened by induction on
. First A
i
[ i ) are dened as some almost disjoint family of innite
subsets of . Each A
, for , is required to satisfy the following three
conditions.
C1 A
is innite and A
for all < .
C2 For every < one of the following possibilities holds:
p1 X
is nitely covered by A
[ < ), or
p2 X
is nitely covered by A
[ < ), or else
p3 X
splits A
(i.e. both X
and A
are innite).
C3 For every i < such that e
i
there are two possibilities:
1. For some m A
m+1 for an innite number of indices
e
i
. (This is the good possibility.)
2. For some m A
there is
0
< sup(e
i
) such that for every ,
0
< e
i
,
min(A
m+ 1) < min(A
m+ 1).
Or, equivalently, if n is the rst member of A
above m then
A
(m, n) ,= .
3
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
If we succeed then C1 and C2 clearly imply that / is pairwise almost
disjoint and inseparable. We are going to show that C3 implies that / con-
tains no Lusin subsequences. Suppose that L = A
i
[ i I) is a subsequence
of /, where I
1
is uncountable . We want to nd some I and m
for which I [ A
m is innite.
Consider the structure on I with predicates for , I, , and the binary
relation m A
i
(for m , i I). Let e I be the universe of a countable
elementary substructure. Then e = e
i
for some i
1
. Let I be any
ordinal such that > i and > sup(e). We want to prove that possibility
C3(1) holds for . This shows that L is not a Lusin sequence. Suppose
instead that C3(2) holds. Then there are m A
and
0
< sup(e) as in
C3(2). Namely, if n is the rst member of A
above m then
A
(m, n) ,= (1)
for every >
0
in e. However, since e
i
is an elementary substructure, we
actually have (1) for every
0
< in I. But this is clearly impossible for
= itself!
Having shown the usefulness of the three conditions C1C3, we return
now to the inductive construction. At the -th stage of this construction, to
construct A
, it is convenient to dene a poset P = (P, ) and a countable
collection of dense subsets of P, and then to dene a lter G P such that
G intersects each of the dense sets in the countable collection. With this
we shall dene A
=
a [ E (a, E) G, and A
will satisfy all three
conditions because of the choice of the dense sets. In this fashion one does
not have to over specify the construction.
A condition p = (a, E) P consists of:
1. A nite set a (which will grow to become A
).
2. A nite set E (p promises that A
= A
a for all E.
Following the tradition that p
1
p
2
means that p
2
gives more information
than p
1
, the partial order on P is dened by
(a
1
, E
1
) (a
0
, E
0
) i E
0
E
1
and a
1
is an end-extension of a
0
such that, for every E
0
, (a
1
a
0
) A
= .
We say that an end-extension a
1
of a
0
respects E (where E
1
is nite)
if (a
1
a
0
) A
= for every E. So (a
1
, E
1
) extends (a
0
, E
0
) if and only
if E
0
E
1
and a
1
is an end-extension of a
0
that respects E
0
.
4
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
Now we shall dene a countable collection of dense subsets of P. First,
to ensure that A
is innite, for every k and p P observe that there
is an extension (a
, E
) of p with k < sup(a
). Then to ensure that A
for all < observe that (a, E ) extends (a, E). These dense sets take
care of C1.
For every X such that neither X nor X are nitely covered by
A
[ < ), and for every k , dene D
X,k
P by:
D
X,k
= (a, E) P [ both a X and a X contain k members .
Claim 2.1 D
X,k
is dense (open) in P.
Proof. Take any (a
0
, E
0
) P. Since neither X nor its complement are
-included in A =
[ E
0
, both X A and ( X) A are innite.
We can nd an end-extension a
1
of a
0
such that (a
1
a
0
) A = and both
a
1
X and a
1
X contain k members. Thus (a
0
, E
0
) < (a
1
, E
0
) D
X,k
.
So add to the countable list of dense sets all sets D
X
,k
for k and
< such that neither X
nor X
are nitely covered by A
[ < ).
This ensures C2.
The main issue of the proof is to take care of C3. What dense sets will do
the job? Fix e = e
i
for any i < such that e . We say that a condition
p = (a, E) P is of type (a) for e if for some m a the following holds.
For every end-extension a
of a that respects E and for every
0
e there is some e,
0
, such that A
m+ 1.
(2)
If p is of type (a) then the least m a that satises (2) is denoted m
p
.
Observe that if p is of type (a) then any extension of p is also of type (a)
(and with the same m).
We say that p = (a, E) P is of type (b) for e if there are two adjacent
members of a, m and n (i.e. m, n a and (m, n) a = ) such that for some
0
e for every
0
e A
(m, n) ,= .
Claim 2.2 Any condition in P has an extension of type (a) or an extension
of type (b).
Proof. Given p = (a, E) let m = max(a). Is p of type (a) by virtue of m?
If yes, we are done, and if not then there are
1. a
an end-extension of a, respecting E, and
5
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
2.
0
e,
such that for every e with
0
, A
m+1 ,= . Let n > max a
be
such that n ,
A
[ E and consider the condition p
= (a n, E)
extending p. Then for every
0
e, A
(m, n) ,= . That is, p
is of
type (b).
For every
0
e dene D
0
,e
by p = (a, E) D
0
,e
i either p is of type (b)
or p is of type (a) and there exists some eE above
0
with A
a m+1
(where m = m
p
).
Claim 2.3 D
0
,e
is dense in P.
Proof. Suppose p
0
P is given. If p
0
is extendible into a condition of type
(b) then we are done. Otherwise there is p
1
= (a
1
, E
1
) p
0
of type (a).
By the denition of type (a), there is some e, with
0
such that
A
a
1
m+ 1. Hence (a
1
, E
1
) D
0
,e
is as required.
Add to the countable list of dense sets all sets D
0
,e
where e = e
i
for some
i < such that e
i
and
0
e
i
. We claim that if A
is dened from a
lter G that intersects all the above dense sets, then condition C3 is ensured.
Given i < such that e
i
= e , we ask if there is (a, E) G of type (b)
for e. If yes, then possibility C3(2) holds for A
.
So we assume that G contains no condition of type (b) for e. Since any
two conditions in G are compatible, it follows that if p, q G are of type (a),
then m
p
= m
q
. Let m denote this common value. We claim that there is an
unbounded set of e such that A
m+1. To see this, consider any
0
e and pick p = (a, E) D
0
,e
G. Then p is of type (a) and there is
E e above
0
with A
a m+ 1. But then A
m+1 follows.
2.2 Martins Axiom: Inseparable contains a Lusin*
subsequence
Assume Martins Axiom +2
0
>
0
. Let / = A
[
1
) be an inseparable
sequence of length
1
(any length below the continuum works). Dene the
following poset:
Q = (u, n) [ u
1
is nite and n <
6
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
ordered by
(u
1
, n
1
) (u
2
, n
2
) i u
1
u
2
& n
1
n
2
&
i u
1
j u
2
u
1
,
j < i [ A
i
A
j
[> n
1
.
(3)
This relation is easily shown to be transitive. We intend to prove that Q is
a c.c.c poset, and that for every <
1
and k < the set D
,k
of (u, n)
in Q for which sup(u) > and n > k is dense. So if G Q is a lter
provided by Martins Axiom which intersects each of these dense sets, then
U =
u [ n(u, n) G is uncountable and A
[ U) is a Lusin*
sequence. Because if i U and k < then
A
j
[ j U i and [A
i
A
j
[ k
is nite by the following argument. For some (u, n) G, i u and n k.
This implies that [A
i
A
j
[ > k for every j < i such that j U u.
The full result, concerning the decomposition of / into countably many
Lusin* subsequences, follows from the fact that (under Martins Axiom) if
Q is a c.c.c poset, then Q is a countable union of lters (each intersects the
required dense sets). (Consider the nite support product of copies of Q,
and remember that [Q[ =
1
.)
It is easy to see that if (u, n) Q and v is any end-extension of u then
(u, n) (v, n). Also, if n m then (u, n) (u, m). This shows that the
required sets D
,k
are indeed dense in Q, and so the main point of the proof
is to show that Q satises the countable chain condition.
Lemma 2.4 Q satises the c.c.c.
Proof. Let (u
, n
) [
1
) be an
1
-sequence of conditions in Q. We
may assume that for some xed n and k, n = n
and k = [u
[ for all
1
,
and that the sets u
form a system. That is, for some nite c
0
1
c
0
= u
1
u
2
for all
1
,=
2
and max(u
1
) < min(u
2
c
0
) for
1
<
2
.
We want to nd
1
<
2
such that (u
1
u
2
, n) extends both (u
1
, n)
and (u
2
, n). It is evident that (u
1
u
2
, n) extends (u
1
, n) (the lower part)
but the problem is the possibility that for some i u
2
and j u
1
c
0
[A
i
A
j
[ n.
We shall nd two uncountable sets K, L
1
such that for every
1
K
and
2
L, (u
1
, n) and (u
2
, n) are compatible. We start with K
0
= L
0
=
1
,
7
5
3
7
r
e
v
i
s
i
o
n
:
2
0
0
1
-
0
2
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
0
2
-
1
2
and dene K
i+1
K
i
and L
i+1
L
i
by induction, considering in turn each
pair 0 i, j [u
c
0
[ (any can be taken, as these sets have all the same
size). The denition of K
i
and L
i
depends on a nite parameter set, and it is
convenient to have a countable model in which the denition is carried on. So
let M H
1
, /, Q, (u
, n
) :
1
) be a countable elementary submodel
(where H
1
is the collection of all sets that are hereditarily countable). The
following lemma is used.
Lemma 2.5 Let U, V M be two uncountable subsets of
1
and n < .
There are uncountable subsets U
1
U and V
1
V (denable in M) such
that for every U
1
and V
1
, [A
[ > n (and hence (, n) and
(, n) are compatible in Q).
It should be obvious how successive applications of the lemma yield the c.c.c.,
and so we turn to the proof of the lemma. Let =
1
M be the set of
countable ordinals in our countable structure M.
Case 1: for some U and V [A
[ > n. In this case
pick a nite X A
with [X[ > n, and let U
1
= i U [ X A
i
,
V
1
= j V [ X A
j
. Both U
1
and V
1
are uncountable (for if U
1
is
countable then it would be included in M, but A
shows that this is not the
case).
Case 2: not Case 1. So, for every U and V , [A
[ n. Let
0 m
0
n be the maximal size of some intersection F = A
for indices
and as above. Then U
1
= i U [ F A
i
and V
1
= j V [ F A
j
are uncountable and for i U
1
and j V
1
, A
i
A
j
= F (by maximality
of [F[). So the set B =
A
i
[ i U
1
separates / (as B A
j
= F for every
j V
1
) which is a contradiction.
References
[1] N. Lusin, Sur les parties de la suite naturelle des nombres entiers,
Comptes Rendus (Doklady) de lAcademie des Sciences de lURSS 40
(1943), 175178.
[2] M. Scheepers, Gaps in
, in Set Theory of the Reals (Ramat Gan, 1991),
Israel Mathematical Conference Proceedings, Vol. 6, Bar-Ilan Univ., Ra-
mat Gan (1993), 439561.
8
