Академический Документы
Профессиональный Документы
Культура Документы
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
-Entangled Linear Orders and Narrowness of
Products of Boolean Algebras
Saharon Shelah
Institute of Mathematics
The Hebrew University
91 904 Jerusalem, Israel
and
Department of Mathematics
Rutgers University
New Brunswick, NJ 08854, USA
August 17, 2011
Abstract
We investigate -entangled linear orders and narrowness of Boolean
algebras. We show existence of -entangled linear orders in many
cardinals, and we build Boolean algebras with neither large chains nor
large pies. We study the behavior of these notions in ultraproducts.
/D (1.7).]
Section 2: Constructions for =
<
.
[In 2.3, assuming = 2
=
+
(and
of cardinality
+
, the derived linear order is -entangled (of cardinality
+
). It provides an example of Boolean algebras B
) = , inc((B
/D) =
+
for each uniform ultralter D on .]
Section 3: Constructions Related to pcf Theory.
[We give sucient conditions for Ens
i
/D),
i
> max pcf(
j
: j < i) (see 3.1). If 2
sup
i
i
(and more)
we can get -entangled linear order (3.2). Also we can utilize Ens
(
i
,
i
)
(see 3.3, 3.4). Now relaying on a generalization of <
pp(
) <
||
+4, we prove that if =
<
then for many [,
+4) we have
Ens
(
+
, ) and if 2
+
(see 3.6). Hence for each for a class of successor cardinals there is a
-entangled linear order of cardinality
+
(see 3.7).]
Section 4: Boolean Algebras with neither pies nor chains.
[Rening results in Section 3, we get Boolean algebras (again derived from
trees
i
j<i
j
using = tcf(
i
/D), but not as interval Boolean algebras),
which have neither large chains nor large pies. For this we need more on
how = tcf(
i
/D).]
Section 5: More on entangledness.
[In 5.1, 5.4 we deal with cases 2
<
< 2
(a) =
pcf(a b) : b a, [b[ < proving for it parallel of
the old theorem and connecting it to entangledness, mainly: if each a is
1
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(, , 2)-inaccessible, then pcf
ex
(a) Ens(, 2
/D >
inc(B
/D) it suces to nd B = BA
inter
(1) such that [B[ > (inc(B))
0
.]
0 Introduction
In the present paper we investigate -entangled linear orders and narrow-
ness of Boolean algebras (if B is the interval Boolean algebra of a linear
order 1, then the algebra B is narrow if and only if 1 is entangled). On
entangled =
0
entangled (= narrow interval Boolean algebra) linear orders
(Denition 1.1(4)) see Bonnet [Bo], Abraham, Shelah [AbSh 106], Abraham,
Rubin, Shelah [ARSh 153], Bonnet, Shelah [BoSh 210], Todorcevic [To] and
[Sh 345, 4], [Sh 345b], [Sh 355, 4.94.14], [Sh 410, 4].
We prove that for many cardinals there is a entangled linear order
of cardinality (see 3.7). For example, if is a limit cardinal, =
<
,
2
>
+
+4
then for some singular cardinal [,
/D, and we may ask whether this holds for but for
no smaller
/D) =
+
(on suciency see
1.7; on existence see 2.3(3), 3.2, 3.6(3)). Similarly for the entangledness of
a linear order. Unfortunately, though we know that there are entangled
linear orders of cardinality
+
for many cardinals (as needed), we do not
know this for cardinals satisfying =
(even
0
= ), and <
/D
+
. Still, the unresolved case requires quite
2
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
peculiar cardinal arithmetic (everywhere): usually 2
is not so large in
the aleph sequence, and there are additional strong restrictions on the power
structure in V. For instance, for every
= 2
<
+4
and
is strong limit of conality > 2
<
+
& ()( <
= 2
)
and
>
+
.
To make the paper more self-contained we give fully the straight general-
izations of [Sh 345], [Sh 355] and [Sh 410]. The research is continued in
Magidor Shelah [MgSh 433], Shar Shelah [SaSh 553], Roslanowski Shelah
[RoSh 534], [RoSh 599], and lately [Sh 620].
We thank Andrzej Roslanowski and Opher Shar for reading, correcting,
pointing out various aws and writing down signicant expansions.
Notation Our notation is rather standard. We will keep the following
rules for our notation:
1. , , , , , , i, j . . . will denote ordinals,
2. , , , , . . . will stand for cardinal numbers,
3. a bar above a name indicates that the object is a sequence, usually
X
will be X
i
: i < g(
X)), where g(
X) denotes the length of
X,
4. for two sequences , we write whenever is a proper initial
segment of , and when either or = . The length of a
sequence is denoted by g().
For a set A of ordinals with no last element, J
bd
A
is the ideal of bounded
subsets of A.
1 Basic Properties
In this section we formulate basic denitions and prove fundamental depen-
dencies between the notions we introduce.
Denition 1.1 Let , , , be cardinal numbers.
3
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
1. A sequence
1 = 1
(for
u v), there are < < such that
u t
<
I
t
and v t
>
I
t
.
Ens(, , , ) = Ens
(, , ) means:
there is a (, )entangled sequence
1 = 1
: < ) of
linear orders, each of cardinality .
2. If we omit , this means = (i.e. [1
[ = ), if we omit it means
=
0
.
3. A linear order 1 is (, )entangled if (1 has cardinality and)
for every () < and a partition (u, v) of () and pairwise distinct
t
<
I
t
and v t
>
I
t
.
4. We omit if [1[ = (and so we write 1 is entangled instead 1
is ([1[, )entangled); we omit also if it is
0
.
5. A sequence 1
)entangled
if
(a) each of 1
is of cardinality ,
(b) if u, v are disjoint subsets of , [u v[ < 1 + , () <
for
u v and t
,
1
,
,= t
,
)
then for some < < we have:
u ( < ())(t
,
< t
,
) and
v ( < ())(t
,
< t
,
).
Proposition 1.2 1. Assume
1
1
,
1
and
1
.
Then Ens
(, , ) implies Ens
1
(
1
,
1
,
1
).
4
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
2. If 1 is a (, )entangled linear order, 1, and [1[ [[
1
,
1
then is (
1
,
1
)entangled.
3. If a linear order 1 has density ,
<
< , = cf() and 2
then in Denition 1.1(3) of 1 is (, )entangled we can add to the
assumptions
() there is a sequence [a
, b
(a
, b
).
4. Moreover, if a linear order 1 has density ,
<
< = cf()
then for each () < and sequences
t
= t
,= t
, b
(a
, b
)) or ( A)(t
= a
).
5. If 2 and a linear order 1 is (, )entangled then 1 has density
< .
6. If there exists a (, )entangled linear order of size then Ens
(, , ).
7. In Denition 1.1(3), if is innite, we can weaken < < to
,= , < , < .
8. If there is a (, )entangled linear order of size and ()
below holds
then Ens
(, , ), where:
()
< t
there exists s
,
such
that t
s
,
< t
h
1
() = s
,
: , ) h
0
())
h
2
() = , , , TV(t
= s
,
)) : , ) h
0
(), < ())
h
3
() = , , , TV(t
< s
,
)) : , ) h
0
(), < ()),
(where TV() is the truth value of ). Now, for each < 4, dom(h
) =
and [rang(h
)[ [[
|()|
3
<
< . Since cf() = , there exists A []
= s
,
for A. As the t
/ s
,
: , ) h
0
().
Dene for < ():
1
s
,
and t s
,
t
s
,
.
Note that the value of is immaterial.
Now, easily 1
, [
+
, [
). Let
=
<()
(
+
realizes in
, and t
.
Now we can easily choose (a
, b
): a
is any member of
which is < t
for A
and b
is any member of
+
which is > t
for A
.
4) Included in the proof of 1.2(3).
5) By 1.2(2), without loss of generality = 2. Suppose that 1 has density
at least . By induction on < we try to choose t
0
, t
1
such that
(i) t
0
< t
1
,
(ii) t
0
, t
1
/ t
0
, t
1
: < ,
(iii) ( < )( 0, 1)(t
0
< t
t
1
< t
).
6
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Continue to dene for as long as possible. There are two possible outcomes.
Outcome (a): One gets stuck at some < .
Let = t
0
, t
1
: < . Then
(t
0
< t
1
1 )(s )(t
0
< s [t
1
< s]).
Since t
0
, t
1
/ , it follows that t
0
< s < t
1
. So is dense in 1 and is of
cardinality 2[[ < a contradiction.
Outcome (b): One can dene t
0
, t
1
, t
1
: < )
of pairwise disjoint subsets of 1, each of power . This sequence witnesses
Ens
= 2 : u 2 + 1 : v, v
= 2 : v 2 + 1 : u,
s
2
= t
2
, s
2+1
= t
2+1
.
Now we apply Denition 1.1(3) the 1.2(7) version to u
, v
, s
:
u
,=
as required there. If
<
then
= 2
, = 2
>
and then
= 2
+ 1, = 2
: <
+
[1]
such that
( < <
+
)([A
= 1 A
(for <
+
). We claim
that the sequence
: <
+
) witnesses Ens
(, ,
+
). Why? Clearly
[J
[ = = . Suppose that u, v
+
are disjoint, [u v[ < 1 + and for
u v let t
: < )
0
,= t
1
)
(remember the choice of the A
: u v, ((), )) 1.
If > then we can choose a family A
: <
+
of pairwise disjoint sets
from [1]
inter
(1) is the closure of the family of
subsets of 1 of the form [, s) (for s 1), by complementation and
unions and intersections of < members
1
.
Denition 1.4 Let B be an innite Boolean algebra.
1. A set Y B is a pie if any two members of Y are incomparable (in
B; pie comes from a set of pairwise incomparable elements).
2. inc(B) = sup[Y [ : Y B is a pie .
3. inc
+
(B) = sup[Y [
+
: Y B is a pie .
4. The algebra B is narrow if there is no pie of cardinality .
5. Length(B) = sup[Y [ : Y B is a chain ,
Length
+
(B) = sup[Y [
+
: Y B is a chain .
Proposition 1.5 Suppose that 1 is a linear order and that the regular car-
dinals
0
< satisfy ( < )[
<
< ]. Then the following conditions
are equivalent:
(a) The order 1 is (, )entangled.
(b) If () < , and u, v () are disjoint and t
I
t
and v t
I
t
.
(Note: if the t
inter
(1) is narrow.
1
Equivalently, the Boolean algebra generated by {xt : t I} freely except xs xt
when I |= s < t.
8
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proof (a) (c). By 1.2(5) the order 1 has density < .
Let A
inter
(1).
We know that for each there are: an ordinal
: <
) 1 such that A
(. . . , t
, . . .)
<
. By 1.2(4),
without loss of generality for some () and pairwise disjoint intervals [a
, b
]
we have:
< t
< b
) or
( < )(t
= a
: < , where x
=: t
: < ().
Consequently, we may assume that x
= t
) and if t
,= t
(for < )
are pairwise distinct or they are pairwise equal. Since = cf() > and
<
< for < , without loss of generality
= . Let
w = < () : t
, b
] we have
A
= A
uv
(a
, t
, b
), where we let
(x, y, z) =
_
[x, y) if u,
[y, z) if v.
Since 1 is (, )entangled, we can nd < such that
( u v)(t
< t
u).
Clearly this implies that A
, so we are done.
(c) (a). First we note that the linear order 1 has density < .
[Why? Easily 1 has no well ordered subset of power nor an inverse well
ordered subset of power . Assume 1 has density . First we show that
there are disjoint closedopen intervals 1
0
, 1
1
of 1 with density . To
prove the existence of 1
0
, 1
1
dene the relation E on 1 by:
a E b if and only if
a = b or [a < b and density([a, b)) < ] or
[a > b and density([b, a)) < ].
Clearly E is an equivalence relation and its equivalence classes are convex.
Moreover, the density of each Eequivalence class is less than (as there
9
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
is no monotonic sequence of length of members of 1). Consequently we
nd a, b 1 such that a < b, a E b. Next we can nd c, d (a, b),
c < d such that neither a E c nor d E b. Thus we may put 1
0
= [a, c),
1
1
= [d, b). Now for each 1
m
we choose by induction on < elements
a
m
< b
m
from 1
m
such that [a
m
, b
m
] is disjoint from a
m
, b
m
: < . So
< [a
m
, b
m
) , [a
m
, b
m
). Now, [a
0
, b
0
) (1
1
[a
1
, b
1
)) : < )
shows that the algebra BA
inter
(1) is not -narrow, a contradiction].
By 1.2(7) it is enough to prove that if () < and t
< t
, and v t
> t
.
By 1.2(4), without loss of generality for some pairwise disjoint intervals
[a
, b
] of 1, we have t
(a
, b
). Let x
=: x
1
x
2
, where
x
1
=:
_
[a
, t
) : u, x
2
=:
_
[t
, b
) : v.
So for < , x
BA
inter
(1). The algebra BA
inter
(1) is narrow, so for
some ,= (< ) we have x
, and v 1 [= t
.
This is as required.
(a) (b). It is included in the proof of (a) (c).
(b) (a). Trivial.
1.4
Proposition 1.6 Let B be a Boolean algebra.
1. If inc
+
(B) is a successor cardinal then [inc(B)]
+
= inc
+
(B).
2. If inc
+
(B) is a limit cardinal then inc(B) = inc
+
(B).
3. B is narrow if and only if inc
+
(B) .
4. If B is narrow then so is every homomorphic image of B.
5. If D is a lter on and the product algebra B
/D is narrow.
1.6
Conclusion 1.7 Assume
,,,,
(B)
,,,,
, using B
j
= BA
inter
(1
j
+
j
), where
(A)
,,,,
there are linear orders 1
j
,
j
(for j ) of cardinality
, such that each 1
j
+
j
is (, )entangled and for any uniform
ultralter D on , the linear orders
j<
1
j
/D and
j<
j
/D have
isomorphic subsets of cardinality
;
(B)
,,,,
there are interval Boolean algebras B
j
(for j < ) which
are narrow and of cardinality such that for any uniform ul-
tralter D on the algebra B =
i<
B
i
/D is not
narrow.
2. Also (A)
+
,,,,
(B)
+
,,,,
(using B = BA
inter
(1 +)), where
(A)
+
,,,,
there are linear orders 1, of cardinality , such that 1+
is (, )entangled and for any uniform ultralter D on the
linear orders 1
/D,
.
(B)
+
,,,,
there is a -narrow interval Boolean algebra B of cardi-
nality such that
< inc
+
[B
narrow).
3. We can replace uniform ultralter D by regular ultralter D or
x a lter D on .
Proof Just note that
if B is a Boolean algebra, 1, are linear orders, a
t
B for
t 1 + are such that t < s a
t
<
B
a
s
and f is an (order)
isomorphism from 1 to
then a
f(t)
a
t
: t 1 is a pie of B.
1.7
Conclusion 1.8 Assume that <
/D,
/D)
and even
inc
+
((BA
inter
(1 +))
/D) >
(so inc((BA
inter
(1 +))
/D).
11
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Remark: See an example in 3.2(3).
Denition 1.9 We say that a linear order 1 has exact (, , )density if
for every 1 of cardinality we have density(J) [, ).
If =
+
we may omit ; if = [1[ we may omit it. We may also say 1
has exact density (, , ) or (, , ) is an exact density of 1 (and replace
(, , ) by (, ) or (, ) or ).
Denition 1.10 1. A linear order 1 is positively -entangled if for each
() < 1 + , u , () and an indexed set t
,= t
)
there exist < < [1[ such that ( < ())( u t
< t
).
2. Similarly we dene when
1 = 1
i
: i < i
) is positively -entangled
and PosEns
(, , ), PosEns
(, ).
For more on entangledness in ultraproducts see section 6.
2 Constructions for =
<
In this section we will build entangled linear orders from instances of GCH.
Our main tool here is the construction principle presented in [HLSh 162] and
developed in [Sh 405]. The main point of the principle is that for standard
+
semiuniform partial orders (see 2.1 below) there are suciently generic
lters G, provided
= p T : dom(p) .
We say that (T, ) is a standard
+
semiuniform partial order if the
following conditions are satised:
12
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(a) If p q then dom(p) dom(q).
(b) If p T, <
+
is either a successor ordinal or cf() = then
there exists q T such that q p and dom(q) = dom(p) ;
moreover there is a unique maximal such q which will be denoted
p .
(c) If p = (, u) T, h : u
11
v
+
is an order isomorphism onto
v such that ( u)(cf() = cf(h()) = ) and v is closed
then h[p]
def
= (, v) T; moreover, q p implies h[q] h[p].
(d) If p, q T, <
+
is either a successor ordinal or cf() = and
p q T
i
) and p
i
: i <
) T is increasing. Then the family p
i
: i < q has an
upper bound r T such that (i < )(p
i
r
i
).
(h) Suppose that
1
,
2
< are limit ordinals and
j
: j <
2
)
+
is a strictly increasing sequence of ordinals, each
j
either a
successor or of conality . Let
p
i,j
: (i, j) (
1
+ 1) (
2
+ 1) (
1
,
2
)) T
be such that
p
i,j
T
j
, i i
p
i,j
p
i
,j
, j j
p
i,j
p
i,j
j
.
Then the family p
i,j
: (i, j) (
1
+1) (
2
+1) (
1
,
2
) has
an upper bound r T such that (j <
2
)(r
j
= p
1
,j
).
Notation 2.2 Let , be cardinals and T be a tree.
1. For an ordinal , the -th level of the tree T is denoted by T
; for
x T, lev(x) is the unique such that x T
.
13
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
2. We say that the tree T is normal if for each y, z T we have:
if (x T)(x <
T
y x <
T
z) and lev(y) = lev(z) is a limit ordinal
then y = z.
Usually we assume that T is normal.
3. We say that the tree T is
+
Aronszajn if it has
+
levels, each level
is of size , there is no
+
-branch in T, T is normal, and
y T, lev(y) < <
+
(z T)[y <
T
z & lev(z) = ].
4. For ordinals and let T
[]
. We let T
[]
=
T
[]
.
6. For x, y T
[]
, let x < y mean ( < )(x
<
T
y
); similarly for
x y.
7. Let x T
[]
. We dene T
[]
x
def
= y T
[]
: x <
T
y.
8. T
T
[]
x
= T
[]
x
T
.
9. For x, y T let xy be their maximal lower bound (in T, exists when
T is normal).
10. For x T and an ordinal lev(x) let x is the unique y
T
x
such that lev(y) = .
11. For x = x
: < ) T
[]
and an ordinal lev( x) let x = x
: < ).
12. Let H
1
(
+
+
) : < and such that for < , x
(
+
+
)
we have: h( x)
(
+
) (if dened, and then) there are
+
members
of h( x) with pairwise disjoint ranges.
If h H
1
: dom(h) =
(
+
+
).
14
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
13. Let H
0
,
1
) :
< )) does not depend on
0
: <
))).
14. Let H
3
: < )
+
and every <
+
there is
1
: <
) (,
+
) with
(
<
+
)(
2
: < ) (
,
+
))((
0
,
1
,
2
) : < ) h()).
15. H
2
,
1
,
2
) :
< ) h()does not depend on
0
,
2
) : < ) h()).
Theorem 2.3 Suppose =
+
= 2
and
H
1
and H
3,
H
3
+
Aronszajn tree T
+
>
in which each node has immediate
successors and there are two functions c,
d such that:
(a) c is a function from T to ,
(b) for every x T
[]
and a function h H
1,
H
3,
we have d
x,h
:
T
[]
x
such that if y, z T
x
are distinct, d
x,h
( y) = d
x,h
( z)
then for some
t T
[]
x
we have
() t
i
= y
i
z
i
, and the values of lev(t
i
) do not depend on i,
() lev(
t) then ( < )(
i < )(c(t
i
) = ),
() if < and h H
1,
t)) : < ))
(ii) or y
i+
(lev(
and d
x,h
( y) < d
x,h
(
t)), z
i+
(lev(
t)), y
i+
(lev(
x
and not dom(d
x,h
)? As in proving amalgamation
we should compare x
i
: i < ), x
i
: i < ); necessary when =
sup(w
q
). However, working a little bit harder we may wave this.
2. Why e.g in clause (b)() we demand d
x,h
( y) < d
x,h
(
t)? Otherwise we
will not be able to prove the density of
T
x,h, y
=: p /T : y dom(d
x,h
) ( or x < y).
3. Why do we have clauses (b)() and (b)()? For the application here
(b)() suces, if this is enough for the reader then clause (J) in the
denition of /T may be omitted. But they both look local maximal.
Proof of 2.3(1) We will use 2).
Let T
+
>
and c,
d be as there and let all the functions h
,u
H
0
dened
in the continuation of the proof of 2.3(1) below be in H
1,
. We may assume
that if x T and < then x
) T. Let <
be a linear order on
such that (, <
i
<
j
for i < i
0
< , j < j
0
< then for some ,
i
j
).
We dene the order <
I
on 1 = T
1
= x T : lev(x) is a successor or of
conality :
y <
I
z if and only if
either z = (y z) y or y()) <
= y
i
: i < ) (for <
+
) are such that a
i
<
I
y
i
<
I
b
i
and y
i
s are pairwise distinct for <
+
. Let u . Take () <
+
such
that (i < )(lev(a
i
), lev(b
i
) < ()). As y
i
s are pairwise distinct we may
assume that
( <
+
)(i < )(() < lev(y
i
) and lev(y
i
)).
Note that if i < j < , and , <
+
then y
i
() ,= y
j
(). Now the
following claim (of self interest) is applicable to y
i
: i < ) : [(), ))
and as we shall see later this nishes the proof of 2.3(1) shortly.
Claim 2.3.1 Assume (for the objects constructed in 2.3(2)):
(a) < ,
(b) for each <
+
we have a sequence y
= y
i
: i < ) such that y
i
T
and either
() (
1
, i
1
) ,= (
2
, i
2
) y
1
i
1
,= y
2
i
2
or
() lev(y
i
) ,
(c) h H
1,
H
3,
,
(d) for some () <
+
,
, <
+
, i < j < y
i
() ,= y
j
().
Then we can nd
1
<
2
<
+
such that clause (b)()(i) (or (b)()(i),
respectively) of 2.3(2) holds with ( y
1
, y
2
) standing for ( y, z), i.e.:
() if h H
1,
then
(i) y
(lev(
), y
(lev(
t)) : < ),
17
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() if h H
3,
then
(i) (c(t
), y
(lev(
t)), y
(lev(
1
j
y
2
j
etc.
Proof of the claim: First note that, by easy thining (as either (b)()
or (b)() holds; remember clause (e)) we can assume (b)() & (b)(). As
=
i
() : i < ) is the same for all
+
.
Let
Z = [(),
+
) : (Y [T
]
<
)([ <
+
: (i < )(y
i
/ Y )[ ).
First we are going to show that Z ,= [(),
+
). If not then for each
[(),
+
) we nd a set Y
[T
]
<
and an ordinal () <
+
such that
<
+
: (i < )(y
i
/ Y ) ().
For [(),
+
), choose x
T
[]
such that
(i) (i < )(y
i
= x
i
),
(ii) Y
i
: i < .
For each [(),
+
) with cf() = we can nd
=
x
i
T
are pairwise
distinct, i < < and the tree T is normal). By Fodors lemma, for some
the set
S
0
def
= [(),
+
) : cf() = &
is stationary. For S
0
there are at most
= possibilities for x
:
i < ) and hence for some x
= x
i
: i < ) T
[]
the set
S
1
def
= S
0
: x
= x
: i < )
is stationary. Hence the set
S
2
def
= [(),
+
) S
1
: (i < )(x
= x
i
) & ( < )(() < )
is stationary. Look at d
x
,h
(really any h
H
,
dom(d
x
,h
) for S
2
(remember cf() = ), and therefore we nd
1
,
2
S
2
,
1
<
2
such that d
x
,h
( x
1
) = d
x
,h
( x
2
). As (
1
) <
2
we
can nd i < such that y
2
i
1
Y
1
. Thus for some j < necessarily
y
2
i
1
= x
1
j
and hence x
2
i
1
= x
1
j
and consequently x
2
i
= x
1
j
.
This implies i = j (as
1
,
2
S
2
S
0
and hence
1
=
2
and now
18
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
apply the denition of
1
,
2
, S
2
) and thus x
1
i
x
2
i
, what contradicts
clause () of 2.3(2b). So we have proved Z ,= [(),
+
), but by its denition
Z is an initial segment of [(),
+
). Hence for some () [(),
+
) we
have Z = [(), ()).
Let [(),
+
) be a successor ordinal.
By induction on < choose
3
pairwise disjoint x
= x
i
: i < ) T
[]
def
= [,
+
) : (i < )(y
i
= x
i
)
are of the size
+
. Suppose we have dened x
for <
. Let Y =
x
i
: i < , <
. Then Y [T
]
<
and by the choice of the set
<
+
: (i < )(y
i
/ Y ) is of the size
+
. As
= we can nd
x
T
[]
such that
[ <
+
: (i < )(y
i
= x
i
/ Y )[ =
+
.
Now for < and i < let x
+i
= x
i
and let x = x
j
: j < ). Thus
x T
[]
. For each [,
+
) choose
,
: < ) such that
,
Z
and
1
<
2
implies
1
<
1
,
<
2
,
. For < , i < and [,
+
)
put z
+i
= y
,
i
. Then z
= z
j
: j < ) T
[]
, x < z
. Consider
the function d
x,h
. Let S
3
def
= (,
+
) : cf() = and note that for
each S
3
the restriction d
x,h
z
: < is a one-to-one
function (since cf() = and we have clause () of 2.3(2b)). Consequently,
for S
3
, we nd () such that
() < d
x,h
( z
) > d
x,h
( z
); () > .
Next, applying Fodor Lemma, we nd a stationary set S
4
S
3
and
>
such that d
x,h
( z
) > d
x,h
( z
) for S
4
, (
, ) and z
= z
for all
1
,
2
S
4
. Let
S
4
def
= S
4
: ( < )([ S
4
: z
= z
[ =
+
).
If S
4
is not stationary then for S
4
S
4
choose
4
S
4
we have
for S
: S
5
is
constant. Choose
1
,=
2
from S
5
. By clauses (), () of 2.3(2b) we get
3
We could have done it for < , but no need here.
19
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
that if t
j
= z
1
j
z
2
j
(for j < ) then
t T
[]
x
and lev( x) lev(
t) < lev( z
1
),
lev(
t) < lev( z
2
). Moreover by the denition of S
4
we have lev(
t)
and
d
x,h
(
t) > d
x,h
( z
1
). Hence by the clause () of 2.3(2b) we nd i < divisible
by such that either (i) or (ii) of clause () of 2.3(2b) holds with ( y, z) there
standing for ( z
1
, z
2
) here. By the symmetry wlog (i) of 2.3(2b)() holds.
Let lev(
t) <
<
2
S
5
S
4
) and by the character of the requirement on
1
,
2
wlog
1
<
2
, so we are done.
2.3.1
Continuation of the proof of 2.3(1): Remember we have < ,
(a
i
, b
i
) (for i < ), u and we have y
: [(),
+
)), y
= y
i
: i < ).
Dene h = h
,u
H
0
):
h(
1
i
: i < )) =
2
i
: i < )
: (i < )(
1
i
<
2
i
i u).
By the choice of <
, there are
1
<
2
as there and we are done.
We still have to prove ()
1
.
Suppose that a, b 1, a <
I
b. By the denition of the order there is t T
such that t s T s (a, b)
I
. As the tree T is
+
Aronszajn we
nd x T
[]
T
[]
. Take any
h H
1,
H
3,
) :
S) is constant. Consequently, elements y
0
for S are pairwise
incomparable (in the tree T). Hence y T : y
0
y : S is a family
of pairwise disjoint convex subsets of (a, b)
I
(each with
+
elements), so we
have nished.
2.2(1)
Proof of 2.3(2): We want to apply [Sh 405, Appendix]. For this we
have to dene a set /T of approximations and check the conditions of 2.1.
Denition 2.3.2 The set /T of approximations consists of all tuples p =
t, w, , D,
d, e, f, c) (we may write t
p
, w
p
etc) such that
(A) t is a subset of
+
of cardinality < , t [0, ) = 0,
(B) w = <
+
: [, ( + 1)) t ,= is such that
( <
+
)( w + 1 w), and
the set <
+
: w is closed,
20
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(C) =
t
is a partial order on t such that t = t
p
= (t, ) is a normal tree
(so xy is well dened) and for each w the set t [, (+1))
is the otp(w )-th level of t (but possibly < are in w, and for
some x in the -th level of t there is no y in the -th level, x
t
y)
[obviously, the intention is: t approximates T; we may use t for (t,
t
)],
(D) D is a set of < pairs ( x, h) such that x t
[]
and h H
1,
H
3,
,
(E)
d = d
x,h
: ( x, h) D), each d
x,h
is a partial function from t
[]
x
to
with domain of cardinality < such that:
[ y t
[]
x
& x <
t
y <
t
z & z dom(d
x,h
)] y dom(d
x,h
),
(F) e = e
x,h
: ( x, h) D), each e
x,h
is a partial function from t
[]
x
to
0, 1, 2 of size < and such that:
(i) ( y, ) dom(e
x,h
) y dom(d
x,h
), and
dom(e
x,h
) ( y, ): y dom(d
x,h
) & ( z dom(d
x,h
))( d
x,h
( z)),
(ii) if y t
[]
x
, x <
t
y <
t
z dom(d
x,h
) and e
x,h
( z, ) is dened then
e
x,h
( y, ) is dened and e
x,h
( y, ) e
x,h
( z, ) and if y t
x
then
at most one of them is 1,
[here, we interpret t
x
as the set of those y t
[]
x
that lev( y) is
either a successor ordinal or is otp(w ) for some w such
that cf() = ],
(iii) d
x,h
( y) = e
x,h
( y, ) = 1,
[the intention: e
x,h
is not explicitly present in 2.3(2b), but e
x,h
( y, ) =
will mean that: if = 0 then for some z we have y < z and d
x,h
( z) =
; if = 1 then d
x,h
( y) = and if = 2 then none of these],
(G) f is a function from t
+
= t : is of a successor level in t to
such that:
if ,= are immediate successors (in t) of some
then f() ,= f(),
[the intention is that if represents T
i+1
, then f() = (i)],
(H) c is a function from t to ,
21
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(I) if ( x, h) D, e
x,h
( y, ), e
x,h
( z, ) 1, [ y
t
z], [ z
t
y]
then clauses ()() of 2.3(2b) hold (with f(y
(i + 1)), f(z
(i + 1))) replacing y
(i), z
and
y
0
, y
1
dom(d
x,h
), lev( y
0
) = , lev( y
1
) = + 1, e
x,h
( y
1
, ) 1 and
< d
x,h
( y
0
)
then in looking at y
0
, y
1
as candidates for
t, y (or
t, z) in clause (I)
(i.e. in clauses ()() from 2.3(2b)) in () there, for each < ,
for ordinals i < divisible by , the values we have i.e. c
p
(y
0
i+
),
f
p
(y
1
i+
) for < , are compatible with the demand (i) there, i.e. for
every < there are
2
) : < ) h(),
(K) if x < y are in t
[]
, ( x, h) D, y dom(d
x,h
), < d
x,h
( y) and
e
x,h
( y, ) = 0 then ( < )(
i < )(c(y
i
) = ) (i.e. looking at y as
a candidate for
t in 2.3(2b)() the values we have are compatible with
the demand there).
The set /T of approximations is equipped with the natural partial order.
We will want to apply the machinery of [Sh 405, Appendix] to the par-
tial order (/T, ). For this we have to represent it as a standard
+
= (u
= p /T : w
p
.
Now we have to check that (with this representation) /T satises the
demands 2.1(2)(a)(h). Clauses (a) and (c) there should be clear.
To deal with the clause (b) of 2.1(2), for an approximation p /T and
<
+
such that either is a successor ordinal or cf() = , we dene
q = p by:
t
q
= t
p
( ), w
q
= w
p
,
q
=
p
t
q
,
22
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
D
q
= ( x, h) D
p
: x t
q
,
if ( x, h) D
q
then d
q
x,h
= d
p
x,h
(t
q
)
[]
x
and e
q
x,h
= e
p
x,h
((t
q
)
[]
x
),
f
q
= f
p
(t
q
dom(f
q
)), c
q
= c
p
t
q
.
Observation 2.3.3 If p /T is an approximation and <
+
is either a
successor or of conality then p /T is a unique maximal approxi-
mation such that p p and w
p
= w
p
.
Thus p corresponds to p as required in 2.1(2c). The main diculty
of the proof is checking the amalgamation property 2.1(2d). Before we deal
with this demand we will check that some sets are dense in /T (which will
allow us to simplify some arguments and will be of importance in drawing
conclusions) and we will deal with existing of some upper bounds.
Claim 2.3.4 In /T, if p
i
: i < ) is increasing, < and for i
0
< i
1
<
we have w
p
i
0
= w
p
i
1
then its union (dened naturally) is its least upper
bound in /T.
Claim 2.3.5 (Density Observation) Assume p /T.
1. Suppose that w
p
, u [, +) t
p
, [u[ < and for i u we
are given a full branch A
i
of (t
p
,
p
) (i.e. A
i
is linearly ordered
by
p
and w
p
[, + ) A
i
,= ), i ,= j A
i
,= A
j
.
Furthermore, assume that if is limit then
t
p
[, + ) & i u A
i
,= y t
p
: y <
p
.
Then there is q /T, p q such that t
q
= t
p
u, <
q
=<
p
(y, i) :
y A
i
, i u, and the rest is equal (i.e. w
q
= w
p
, D
q
= D
p
,
d
q
=
d
p
,
e
q
= e
p
, f
q
f
p
, c
q
c
p
naturally).
2. If w
p
, i [, + ) then there is q /T, p q such that
i t
q
, w
p
= w
q
, D
q
= D
p
,
d
q
=
d
p
, e
q
= e
p
, and naturally f
q
f
p
,
c
q
c
p
.
3. If x (t
p
,
p
)
[]
and h H
1,
H
3,
: x < y
y, y
(t
p
,
p
)
[]
,
t
p
= t
q
, w
p
= w
q
,
p
=
q
, D
p
= D
q
, d
q
x
,h
= d
p
x
,h
, e
q
x
,h
= e
p
x
,h
for
( x
, h
) D
p
( x, h), and f
q
= f
p
, c
q
= c
p
.
23
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proof of the claim: 1) Check.
2) Iterate (1) (in the j-th time on the j-th level of t) using 2.3.4 for limit
stage. More elaborately, for each w
p
0 choose i
[ , +)t
p
such that i
= i. Next by induction on w
p
(+1) choose an increasing
sequence p
: w
p
( + 1)) /T of approximations such that p
0
= p
and t
p
= t
p
i
: 0 < w
p
( + 1) (so w
p
= w
p
) and the sequence
i
: 0 < w
p
( + 1)) is
p
increasing.
3) We just put: t
q
= t
p
, w
q
= w
p
, f
q
= f
p
, c
q
= c
p
, d
q
x
,h
= d
p
x
,h
if
( x
, h
) D
p
, d
q
x,h
is empty if ( x, h) / D
p
, and similarly for e
q
x
,h
.
4) Let y
= sup(rang(d
p
x,h
) : ( y
)(( y
, ) dom(e
p
x,h
))). Now put
dom(e
q
x,h
) = ( y, ) : y dom(d
q
x,h
) & <
+ 1 + ,
declare that e
p
x,h
e
q
x,h
and
if y dom(d
q
x,h
), y < y
, < then e
q
x,h
( y,
+ 1 + ) = 0,
e
q
x,h
( y
+ 1 + ) = 1,
e
q
x,h
( y, ) = 2 in all other instances.
It should be clear that this denes correctly an approximation q /T and
that it is as required. [Note that clauses (), () of 2.3(2b) relevant to
clauses (J), (K) in the denition of /T hold by the requirement d
x,h
(
t) <
d
x,h
( y) = d
x,h
( z).]
2.3.5
Claim 2.3.6 If p /T and an ordinal
+
w
p
is divisible by then
for some q /T, p q we have w
q
= w
p
[, + ), D
p
= D
q
and
w
p
t
q
[, + ) = t
p
[, + ).
Proof of the claim: Let = min(w
p
) (if is undened then it is
much easier; of course > as / w
p
and therefore + ). Let
t
p
[, + ) = y
i
: i < i
[( + n), ( + n) + )
be with no repetition. Let t
q
= t
p
y
+n
i
: n < , i < i
, and
q
=
p
(y
+n
i
, y
+m
i
) : n m < , i < i
(y
+n
i
, x) : n < , i < i
, y
i
p
x
(x, y
+n
i
) : n < , i < i
, x <
p
y
i
.
24
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
By 2.3.5 we may assume that i
= . For ( x, h) D
p
= D
q
we let
dom(d
q
x,h
) = dom(d
p
x,h
) y (t
q
)
[]
: ( y
dom(d
p
x,h
))( x < y < y
) and
(n < )( < )(y
[ ( + n), ( + n + 1))),
and let
x,h
= sup(rang(d
p
x,h
) : ( y
)(( y
, ) dom(e
p
x,h
))). Fix an
enumeration y
: < of dom(d
q
x,h
) dom(d
p
x,h
) such that y
0
< y
0
<
1
. We put d
q
x,h
( y
) =
x,h
+1+ for < (and we declare d
q
x,h
d
p
x,h
).
Next we dene e
q
x,h
similarly as in 2.3.5(4) putting the value 2 whenever
possible (so dom(e
q
x,h
) = ( y, ) : y dom(d
q
x,h
) & <
x,h
+ 1 + ).
Now comes the main point: we have to dene functions f
q
, c
q
(extending
f
p
, c
p
, respectively) such that clauses (I) + (J) + (K) hold. But it should be
clear that each instance of clause (I) in t
q
can be reduced to an instance of
this clause in t
p
(just look at the denitions of t
q
, d
q
x,h
, e
q
x,h
). Thus what we
really have to take care of are instances of (J) and (K). For this we dene
c
q
y
+n
i
: i < and f
q
y
+n+1
i
: i < by induction on n < . At
the rst stage (for n = 0) we let
P = (, x, h, y, z) : ( x, h) D
q
and < and x < y < z dom(d
q
x,h
),
and z y
+1
i
: i < and y y
i
: i < .
Take a list (X
, x
, h
, y
, z
) : < ) of
(X, , x, h, y, z) : X J, K & (, x, h, y, z) P
in which each 6-tuple appears times and
1 + . Next by induction
on < choose a sequence c
, f
: ) such that
() c
: dom(c
) , dom(c
) y
i
: i < , [dom(c
)[ <
0
+[[
+
,
() f
: dom(f
) , dom(f
) y
+1
i
: i < , [dom(f
)[ <
0
+[[
+
,
() c
: ), f
< divisible by
such that
rang( z
[i
, i
)) dom(f
+1
) dom(f
) and
rang( y
[i
, i
)) dom(c
+1
) dom(c
),
() if X
= J and h
H
3,
, h
, y
,
z
with i = i
,
() if X
= K then c
+1
(( y
)
i
) =
,
25
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() y
dom(c
+1
), y
+1
dom(f
+1
).
There are no diculties with carrying out the construction: the only possible
troubles could come from demand () above. But look at the denition
2.2(14) of H
3
: <
),
1
: <
[i
, i
)) and f
+1
( z
[i
, i
)) in clause ().
The functions c
, f
i
: i < and
f
q
y
+1
i
: i < . Next, arriving to a stage n + 1 of the denition we
repeat the above procedure with no changes. Note that at this stage we
know c
q
y
+n
i
: i < , f
q
y
+n+1
i
: i < but they have no inuence
on dening c
q
, f
q
at levels + n + 1, + n + 2.
2.3.6
Claim 2.3.7 (The Amalgamation Property) Assume that <
+
is
either a successor ordinal or cf() = , p, q /T and p q /T
.
Then there is r /T such that p, q r.
Proof of the claim: First try just the r dened by:
w
r
= w
p
w
q
, t
r
= t
p
t
q
,
r
=
p
q
, D
r
= D
p
D
q
,
d
r
x,h
is: d
p
x,h
if ( x, h) D
p
D
q
d
q
x,h
if ( x, h) D
q
D
p
d
p
x,h
d
q
x,h
if ( x, h) D
q
D
p
and e
r
x,h
e
p
x,h
e
q
x,h
(dened naturally, i.e. with dom(e
r
x,h
) minimal possible
to satisfy demand (F) and value 2 whenever possible), f
r
= f
p
f
q
, c
r
=
c
p
c
q
. Clearly p r, q r, w
r
= w
p
w
q
, but does r belong to /T? The
things that might go wrong are:
there is y t
p
t
q
which has nothing below it in some levels, or
y z not dened for some y, z, or
the relevant cases of clauses (I)(K) fail.
Let
0
=
: w
p
= sup(dom(p) ). Note that
0
dom(p) (as is either successor or of conality ) and
() if w
q
0
+ then r /T.
So we assume from now on that w
q
,
0
+ (by () above). Then
necessarily
0
+1 < (as w
q
). Without loss of generality dom(p) ,=
(as if w
p
we can let r = q) and q (
0
+ 1) = p . Let
26
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
def
= min(dom(p) ) and
def
= min(dom(q) (
0
+ 1)). By 2.3.6 we
may assume that
=
0
+ 1 (i.e.
0
+ w
q
). Let x
i
: i < i
list t
p
[ (
), (
), ((
i
: i < i
of distinct members of t
q
[ (
), (
) + )
such that (z t
p
t
q
)[z <
p
x
i
z <
q
x
i
]. Let x
i
[, + )
(for w
q
(
:
t
r
= t
r
x
i
: w
q
(
+ 1), i < i
, w
r
= w
r
,
=
r
(x
i
, x) : i < i
, x t
p
, x
i
p
x, w
q
(x, x
i
) : i < i
, x t
q
, x
q
x
i
, w
q
(x
0
i
, x
1
i
) :
0
,
1
w
q
,
0
1
, i < i
.
Put D
r
= D
r
. If ( x, h) D
r
D
p
then we can let d
r
x,h
= d
r
x,h
, but if
( x, h) D
p
then we rst let
x,h
= sup(rang(d
r
x,h
) : ( y
)(( y, ) dom(e
p
x,h
) dom(e
q
x,h
))) and
dom(d
r
x,h
) = y (t
r
)
[]
: y dom(d
r
x,h
) or for some
z = x
(j)
: j <) (t
p
)
[]
otp(w
p
)
and w
q
(
+ 1)
we have z dom(d
p
x,h
) and y = x
(j)
: j < ).
Choose d
r
x,h
in such a manner that d
r
x,h
d
r
x,h
and the values d
r
x,h
( y), if not
dened before, are distinct ordinals from (
x,h
, ). Thus, in particular,
d
r
x,h
( y) = d
r
x,h
( z) & y ,= z y, z dom(d
r
x,h
).
Next we dene e
r
x,h
extending e
r
x,h
to satisfy clause (F) we put the value 2
whenever it is possible. [Note that this is the place in which the assumption
that y t
x
in clause (F)(ii), and so the respective assumption in 2.3(2b),
play role: the values of e
q
x,h
at the level
=
0
+ do not interfere
with the values of e
p
x,h
at the level
since
is a successor, not of
conality .] Now we have to dene c
r
c
r
, f
r
f
r
, i.e. to dene
c
r
i
: i < i
, w
q
(
+ 1), and
f
r
i
: i < i
, w
q
is a successor,
such that clauses 2.3.2(G)(K) hold. This is done like in 2.3.6, but now the
clause (I) is active too. Of course, the point is that we have commit-
ments, each has disjoint chances, so we list them in a list of length
27
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
and inductively we can easily do it (for singular at place i of the list
there may appear only a commitment of size [i[ +
0
). More fully, let
P
1
= (, x, h, y, z) : ( x, h) D
p
D
q
and < and for some
( z, ) dom(e
r
x,h
) dom(e
r
x,h
) and ( y, ) dom(e
q
x,h
) and
e
r
x,h
( z, ) 1 and e
q
x,h
( y, ) 1 and
y [(
+ 1), (
+ 1) +) and z x
+1
i
: i < i
.
Dening f we have to take care of condition (I) for all (, x, h, y, z) P
1
.
We also have to take care of conditions (J), (K) for
P
2
= (, x, h, y, z) : ( x, h) D
p
D
q
and < and
z dom(d
r
x,h
), z x
+1
i
: i < i
, y = z , w
q
.
So we use a list (X
, x
, h
, y
, z
) : < ) of (X, , x, h, y, z) : X
1, 2 and (, x, h, y, z) P
1
P
2
in which each 6-tuple appears times
and
1 +. Let x
: < list t
r
t
r
. Now we dene by induction
on functions c
, f
such that
() c
is a function extending c
r
, rang(c
) ,
() dom(c
) dom(c
r
) is a subset of x
i
: i < i
, w
q
(
+ 1) of
cardinality <
0
+[[
+
,
() f
is a function extending f
r
, rang(f
) ,
() dom(f
) dom(f
r
) is a subset of
x
i
: i < i
+ 1)
of cardinality <
0
+[[
+
,
() the sequences c
: ), f
< divisible by
such that:
if X
= I then
rang( z
[i
, i
)) [dom(c
+1
)dom(c
)][dom(f
+1
)dom(f
)],
and if X
J, K, (
, x
, h
, y
, z
) P
2
then
rang( z
[i
, i
)) dom(f
+1
) dom(f
), and
rang( y
[i
, i
)) (dom(c
+1
) dom(c
)) c
r
,
28
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() if X
= I and (
, x
, h
, y
, z
) P
1
then condition 2.3.2(I) holds
for (
, x
, h
, y
, z
, i
),
() if X
= J, (
, x
, h
, y
, z
) P
2
and h
H
3,
then condition
2.3.2(J) holds for ( x
, h
, y
, z
, i
),
() if X
= K, y
, dom(c
r
) then c
+1
(( y
)
i
) =
,
() x
dom(c
+1
) and if x
then x
dom(f
+1
),
() rang(f
+1
i
: i < i
) rang(f
q
) = .
The denition is carried out as in 2.3.6. The new points are clause () and
instances of clause () for such that y
i
: i < i
. In the second
case a potential trouble could be caused by the fact that the function c
is
dened on y
, i
)).
To deal with clause () note that if h
H
3,
H
1,
then the
denition 2.2(12) of H
1
works.
The functions c
, f
are as required.
2.3.7
The demands (e)(h) of 2.1(2) are easy now:
Claim 2.3.8 1. If a sequence p
i
: i < ) /T is increasing, < then
it has an upper bound q /T such that dom(q) = cl(
i<
dom(p
i
))).
2. Assume <
+
, cf() = , < . Let p
i
: i < ) /T
+1
be an
increasing sequence and let q /T
be an upper bound to p
i
:
i < ). Then the family p
i
: i < q has an upper bound r such
that r q.
3. Assume that
i
: i < )
+
is strictly increasing, each
i
is either
a successor or has conality , < is limit. Suppose that q /T
and p
i
: i < ) /T is an increasing sequence such that
(i < )(q
i
p
i
/T
i
).
Then the family p
i
: i < q has an upper bound r /T such
that (i < )(p
i
r
i
).
29
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
4. Suppose that
1
,
2
< are limit ordinals and
j
: j <
2
)
+
is a
strictly increasing sequence of ordinals, each
j
either a successor or
of conality . Let
p
i,j
: (i, j) (
1
+ 1) (
2
+ 1) (
1
,
2
)) /T
be such that
p
i,j
/T
j
, i i
p
i,j
p
i
,j
, j j
p
i,j
p
i,j
j
.
Then the family p
i,j
: (i, j) (
1
+ 1) (
2
+ 1) (
1
,
2
) has an
upper bound r /T such that (j <
2
)(r
j
= p
1
,j
).
Proof of the claim: 1) The rst try may be to take the natural union of
the sequence p
i
: i < ). However, it may happen that we will not get a legal
approximation, as
i<
dom(p
i
) does not have to be closed. But we may take
its closure cl(
i<
dom(p
i
)) and apply a procedure similar to the one described
in 2.3.6 (successively at each element of cl(
i<
dom(p
i
))
i<
dom(p
i
)) and
construct the required q.
2)4) Similarly as 1) above plus the proof of 2.3.7.
2.3.8
Now we apply [Sh 405, Appendix]: we nd a suciently generic G
/T which gives the T, c, d we need (remember 2.3.5):
T =
: t
p
for some p G
where for [, + ) we dene
by:
=
() if and only if
(p G)(
[( + 1), ( + 1) + ))(t
p
[=
< & f
p
(
) = ).
This nishes the proof.
2.3
Remark 2.4 1. Theorem 2.3 is close to [Sh 50], which is a strengthening
of the construction of special Aronszajn trees. There essentially we
replace () + () by
()
y
i
(lev(t
i
)), z
i
(lev(t
i
)) do not depend on i.
2. By the proof of 2.3(2), T is -complete.
3. We may add to 2.3(2):
30
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(c) T is special, i.e. there is a function d :
such that
d() = d() [ <
T
].
[Just in the denition of p /T (see 2.3.2) add such d
p
: t
p
.]
4. The reader may wonder why we need the condition h( x) has pair-
wise disjoint members. The point is that when we amalgamate p
and q when p q, it may happen that p gives information on
levels
n
<
n+1
(for n < ), =
n<
n
< , q gives information
on the level , and when amalgamating the function f
q
gives infor-
mation on f on this level and f is supposed to be one-to-one on every
succ
T
(). So considering x (t
q
)
+1
, y (t
p
)
, 1
of the
cardinality
+
such that for any uniform ultralter D on the linear
orders (1
/D, (1
+1
is -entangled.
Hence there is a Boolean algebra B which is
+
-narrow but B
/D is
not
+
-narrow for any uniform ultralter D on .
3. There are a set R
, [R[ =
+
and functions c,
d such that, letting
T
+
= (
>
R, ) ( is being initial segment), we have:
(a) c is a function from
>
to ,
(b) R =
: <
+
(with no repetition), <
R
= (
) : < ;
dene
R
i
: i < ) :
i
<
+
,
i
: i <
+
) is increasing,
(c) for every x T
[]
<
, < , and h
( ), d
x,h
is a function
from y R
appears in y , = sup :
appears in z
and for some
t T
[]
x
T
[]
<
and i
< we have:
31
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() t
i
= y
i
z
i
, lev(t
i
) = lev(t
i
) for i i
,
() ( < )(
i < )(c(t
i
) = ),
() for ordinals i < divisible by we have
(i) either there are
0
<
1
< such that
( < )(
0
y
i+
(lev(
t)) <
1
z
i+
(lev(
t))) and
h = (c(t
i+
), y
i+
(lev(
t))
0
, z
i+
(lev(
t))
1
) : < ),
(ii) or a symmetrical requirement interchanging y and z.
Proof 1) We apply 2.5(3): let R, c,
d be as there (of course we are in
the universe V
P
all time). We dene the order <
I
on 1 = R by
y <
I
z if and only if
either c(y z) = 0 and y() < z()
or c(y z) ,= 0 and y() > z(),
where = lev(y z).
Clearly <
I
is a linear order of the density , [1[ =
+
. To show that it is
-entangled suppose that y
R (for <
+
, < () < ) are pairwise
distinct, u (). Let y
=
(,)
(for <
+
, < ()). We may
assume that the truth value of (,
1
) < (,
2
) does not depend on
<
+
. For simplicity, we may assume that for each <
+
, <
< ()
implies y
<
R
y
<
+
and ,
< () then y
<
R
y
(i.e. (, ) < (
)). For
< (), i < and <
+
let z
i()+
= y
+i()
and z
= z
i
: i < ).
Clearly each z
is in R
i
() (for i < ) are pairwise distinct. Without loss of generality we
may assume that () = for <
+
. There are
= possibilities for
z
i
: i < ), so we may assume that for all <
+
z
i
: i < ) = x
i
: i < ) = x T
[]
.
Let h
def
= (0,
u
, 1
u
) : < ())
()
(), where
u
is 0 if u and 1
otherwise. For some distinct
1
,
2
<
+
we have d
x,h
( z
1
) = d
x,h
( z
2
). By
the properties of d
x,h
, possibly interchanging
1
,
2
, we nd i < , ordinals
0
<
1
< and
t T
[]
x
such that
( < ())(z
1
i()+
z
2
i()+
= t
i()+
) and
( < ())(()
0
z
1
i()+
() < ()
1
z
2
i()+
() and
h = (c(t
i()+
), z
1
i()+
() ()
0
, z
2
i()+
() ()
1
) : < ()),
32
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
where = lev(
t). Then
1
+ i() ,=
2
+ i() (for i < ) and
( < ())(y
1
+i()
<
I
y
2
+i()
i u)
(by the denition of the order <
I
and the choice of h), so using 1.2(7) we
are done.
2) As in 1) above, we work in V
P
and we use 2.5(3). Suppose . For
a set A we dene the order <
A
on R by
y <
A
z if and only if
either c(y z) A and y() < z()
or c(y z) / A and z() < y(),
where = lev(y z).
[Note that the order <
I
from part 1) above is just <
{0}
).]
Clearly <
A
is a linear order. As in the proof of 2.5(1) one can show that it
is -entangled. As
<
we may choose sets A
i
for i such that
(i) for each < the set i < : A
i
/ A
=
i<
1
i
, 1
= 1
, 1 = 1
+1
=
i
1
i
.
So it is notationally clearer to let 1
i
= (i R, <
i
), (i, y
1
) <
i
(i, y
2
) i
y
1
<
A
i
y
2
, and 1 = (( + 1) R, <
), (i
1
, y
1
) <
(i
2
, y
2
) i i
1
< i
2
or
(i
1
= i
2
& y
1
<
A
i
1
y
2
).
Claim 2.5.1 If y
0
, y
1
R, y
0
<
A
y
1
then that the set i < : y
0
<
A
i
y
1
is co-bounded.
Proof of the claim: Let t = y
0
y
1
, y
0
<
A
y
1
. The set
u =: i < : c(t) A
i
c(t) A
then y
0
(lev(t)) < y
1
(lev(t)). Hence
y
0
<
A
i
y
1
for i u. If c(t) / A
then y
0
(lev(t)) > y
1
(lev(t)) and y
1
<
A
i
y
0
for i u.
2.5.1
Let : 1
i<
1
(1
/D is an embedding.
Thus both (1
/D and (1
/D contain a copy of 1
+1
is -entangled.
Proof of the claim: Suppose that () < and (j
, y
) 1 for <
+
,
< () are pairwise distinct and u (). As 2
wlog j
= j
. Let
z
: <
be an enumeration of Y
=: y
: <
enumerate v =: j
: <
) are
<
R
-increasing and pairwise disjoint (for <
+
) as each z may appear in at
most () of these sequences. Moreover we may assume that (
, , , ) :
(j
, y
) = (i
, z
, y
, z
) : <
, <
) and
so u
)[c[] A
i
(, ) u]
and we proceed as in earlier cases (considering h = (c[], 0, 1) : <
( ))
4
.
2.5.2
3) The denition of the forcing notion P is somewhat similar to that of the
approximations is 2.3(2).
Denition 2.5.3 A condition in P is a tuple p = t, , w, , D,
d, e, c) (we
may write t
p
, w
p
, etc) such that:
(A) w
+
is a set of cardinality < , is a limit ordinal < ; let w
[]
be
the family of all increasing sequences y w of length ,
(B) t is a non-empty closed under initial segments subset of
of cardi-
nality < ,
(C) is such that (t w, ) is a normal tree,
p
t is , and for w,
b
p
def
=
t :
p
t
, and ,= b
p
,= b
p
,
(D) c is a function from t to ,
(E) D is a set of < pairs ( x, h) such that x t
[]
and h
<
(),
(F)
d = d
x,h
: ( x, h) D), each d
x,h
is a partial function from w
[]
to
p
with domain of cardinality < ,
(G) e = e
x,h
: ( x, h) D), each e
x,h
is a partial function from t
[]
x
p
to
0, 2 such that for ( x, h) D:
4
Alternatively, let R be the disjoint union of Ri (i ) and use I
i
= Ii ({i} Ri).
34
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(i) dom(e
x,h
) ( y, ) : for some z
0
, z
1
dom(d
x,h
) we have
x < y z
0
and = d
x,h
( z
1
)),
(ii) if y t
[]
x
, x <
t
y <
t
z and e
x,h
( z, ) is dened then e
x,h
( y, ) is
dened and e
x,h
( y, ) e
x,h
( z, ),
(iii) if z dom(d
x,h
) and x < y < z then e
x,h
( y, d
x,h
( z)) = 0,
and letting h = (
0
,
1
,
2
) : < ) we have
(iv) if ( y, ) dom(e
x,h
), lev( y) =
p
and e
x,h
( y, ) = 0 then
( < )(
i < )(c(u
i
) = ) and
for ordinals i < divisible by , for some <
( < )(c(y
i+
) =
0
),
(v) if x < y
0
< y
1
, lev( y
0
)+1 = lev( y
1
) <
p
and e
x,h
( y
1
, ) = 0,
then for ordinals i < divisible by , for some <
( < )(c(y
0
i+
) =
0
& +
1
= y
1
i+
(lev( y
0
))),
(H) if e
x,h
( y, ) = e
x,h
( z, ) = 0, [ y
t
z], [ z
t
y] then clauses (),
(), () of 2.5(4c) hold.
P is equipped with the natural partial order:
p q if and only if
t
p
t
q
,
p
q
, w
p
w
q
, c
p
c
q
, D
p
D
q
, and
( x, h) D
p
d
p
x,h
d
q
x,h
& e
p
x,h
e
q
x,h
.
For a condition p P and an ordinal <
+
we dene q = p by:
q
=
p
, t
q
= t
p
, w
q
= w
p
,
q
=
p
t
q
, D
q
= D
p
, c
q
= c
p
,
if ( x, h) D
q
then d
q
x,h
= d
p
x,h
(w
q
)
[]
and e
q
x,h
= e
p
x,h
((t
q
)
[]
x
q
).
Observation 2.5.4 If p P and <
+
is either a successor or of conal-
ity then p /T is the unique maximal condition such that p p
and w
p
= w
p
.
Claim 2.5.5 (Density Observation) Assume p P.
1. Suppose
>
. Then there is q P such that
p q, t
q
, w
p
= w
q
, D
p
= D
q
,
d
q
=
d
p
.
35
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
2. For each
+
w
p
there is q P such that p q and w
q
.
3. If x (t
p
,
p
)
[]
, < and h
( ) then there is q P such
that p q and D
q
= D
p
( x, h).
4. If ( x, h) D
p
, x < y (w
p
)
[]
then there is q /T, p q such that
w
q
= w
p
and y dom(d
x,h
).
Proof of the claim: 1) We may assume that [t
p
p
[ = , as otherwise
it is even easier. So let
: < ) enumerate t
p
p
. Put
q
= max
p
+
, g() + and x
q
such that . Next, by induction on
(
p
,
q
] dene sequences
such that
(a) c
: < ,
.
(b)
p
<
0
<
1
q
.
(c) Suppose that ( x, h) D
p
, h = (
0
,
1
,
2
) : < )
( ),
x < y =
(i)
: i < ) t
p
p
, ( y, ) dom(e
p
x,h
) and e
p
x,h
( y, ) = 0.
Then
() for ordinals i < divisible by , for some <
(j < )(c
p
(
(i+j)
) =
0
j
&
p
+1
(i+j)
(
p
) = +
1
j
),
() for each (
p
,
q
) we have
( < )(
i < )(c
(i)
) = ) and
for ordinals i < divisible by there is < such that
(j < )(c
(i+j)
) =
0
j
&
+1
(i+j)
= +
1
j
),
() ( < )(
i < )(c
q (
q
(i)
) = ) and for ordinals i <
divisible by
(j < )(c
q (
q
(i+j)
=
0
j
).
The construction is easy and can be done like the one in 2.3.6. Next we put
t
q
= t
p
:
q
: < ,
p
<
q
,
w
q
= w
p
, D
q
= D
p
,
d
q
=
d
p
.
The function c
q
is dened as any extension of c
p
c
:
p
<
q
(note
that the possibly non-dened values are that at some initial segments of ).
36
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
For w
q
let b
q
be
(i)
: i < ) > y
=
(i)
: i < ),
(
p
,
q
], ( y
, dom(e
p
x,h
) ,
if x < y
< y, ( y
, ) dom(e
p
x,h
), lev( y
) =
p
then e
q
x,h
( y, ) = e
p
x,h
( y
, ).
Now one easily checks that the condition q dened above is as required.
2) Choose
p
such that 1 / t
p
. Let
q
=
p
, t
q
= t
p
: <
p
and w
q
= w
p
.
Dene
q
by letting b
q
= b
p
for w
p
and b
q
= : <
q
. Finally
put D
q
= D
p
, d
q
x,h
= d
p
x,h
and e
q
x,h
= e
q
x,h
.
3) Let t
q
= t
p
, w
q
= w
p
, c
q
= c
p
, d
q
x
,h
= d
p
x
,h
if ( x
, h
) D
p
, d
q
x,h
is
empty if ( x, h) / D
p
, and similarly for e
q
x
,h
.
4) Let h = (
0
,
1
,
2
: < ) of t
p
p
. Let y
=
(i)
: i < ) be such that
y
< y ( w
[]
). Next, like in (1) above build c
for (
p
,
q
], <
satisfying demands (a)(c) there plus:
(c) () for each (
p
,
q
)
( < )(
i < )(c
(i)
) = ) and
for ordinals i < divisible by there is < such that
(j < )(c
(i+j)
) =
0
j
&
+1
(i+j)
() = +
1
j
),
() ( < )(
i < )(c
q (
q
(i)
) = ) and for ordinals i <
divisible by
(j < )(c
q (
q
(i+j)
=
0
j
).
Next we put dom(d
q
x,h
) = dom(d
p
x,h
) y, d
q
x,h
d
p
x,h
, d
q
x,h
( y =
p
+ 2,
d
q
x
,h
= d
p
x
,h
for ( x
, h
) D
p
( x, h), and D
q
= D
p
. The functions c
q
and e
q
x
,h
(for ( x
, h
) D
q
) are dened as in (1), but dealing with ( x, h)
we take into account the new obligation: d
q
x,h
( y) =
p
+ 2 (and we put the
value 2 whenever possible). There is no problem with it as we demanded
clauses (c)(, ). Now one easily checks that q is as required.
2.5.5
37
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Claim 2.5.6 The forcing notion P is (< )complete, i.e. if p = p
i
: i <
i
) P is increasing, i
t
p
i
does not
have the last level. But this is not a problem as we may use the procedure
of 2.5.5(1) to add it.
Claim 2.5.7 (The Amalgamation Property) If <
+
is either a suc-
cessor ordinal or of conality , p, q P, p q and w
q
then there
is r P such that p r, q r and w
r
= w
p
w
q
.
Proof of the claim: By 2.5.5(1) we may assume that
p
<
q
. Moreover
we may assume that [w
p
[ = (as otherwise everything is easier). Let
r
=
q
and w
r
= w
p
w
q
. By induction on [
p
,
q
] choose sequences
,
: w
p
) and functions c
such that
()
,
are increasing with ,
()
,
p = b
p
,
,
p
+1
/ t
q
[note that
,
p : w
p
) is with no repetition],
() c
:
,
: w
p
, [
p
, ) are continuously increasing
with , c
p
+1
is c
p
restricted to
,
p : w
p
,
and for each ( x, h) D
p
, h = (
0
,
1
,
2
) : < ), z dom(d
p
x,h
)
dom(d
q
x,h
) and i
we have
() for each [
p
,
q
), for ordinals i [i
&
z
i+
,+1
() = +
1
),
() for ordinals i [i
, ) divisible by
( < )(c
q
+1
(
z
i+
,
q ) =
0
),
() if x < y < z, lev( y) =
p
, y < y
, lev( y
) =
p
+ 1, ( y
, d
p
x,h
( z))
dom(e
q
x,h
) and e
q
x,h
( y
, d
p
x,h
( z)) = 0 then for ordinals i [i
, ) divis-
ible by there are
0
<
1
< such that
( < )(
0
y
i+
(
p
) <
1
y
i+
(
p
)) and
( < )(c
p
(
z
i+
,
p) =
0
, y
i+
(
p
) =
0
+
1
, y
i+
(
p
) =
1
+
2
,
38
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() for each (
p
,
q
], for every < there are ordinals i < such that
c
+1
(
z
i
,
) = .
[Our intension here is that b
r
=
,
q and c
r
c
q .] We have actually
demands, each of which can be satised by pairwise disjoint cases of size
< . So we may carry out the procedure analogous to that from the end
of the proof of 2.3.7. Note that in handling instances of clause () we use
demand 2.5.3(G)(v) for q (applicable as d
p
x,h
( z) <
p
) and for clause () we
use 2.5.3(G)(iv). After the construction is carried out we easily dene a
condition r as required.
2.5.7
Claim 2.5.8 The forcing notion P satises the
+
cc.
Proof of the claim: Suppose that p
: <
+
) is an antichain in P.
By passing to a subsequence we may assume that otp(w
p
) is constant and
that the order isomorphism of w
p
, w
p
to p
(so
t
p
= t
p
, D
p
= D
p
(remember = 2
=
+
).
Now we may nd an ordinal
<
+
of conality and
0
<
1
<
+
such
that w
p
0
, w
p
1
= w
and w
0
. So applying 2.5.7 we conclude
that the conditions p
0
, p
1
have a common upper bound, a contradiction.
2.5.8
To nish the proof note that if G P is a generic lter over V then, in
V[G] we may dene the tree T by:
T = (
>
)
: <
+
by
=
_
>
: (p G)( b
p
)
(and c, d are dened similarly). By 2.5.5 and 2.5.6 (no new sequences of
ordinals are added) we easily conclude that these objects are as needed.
2.5
We may want to improve 2.3(2) so that it looks more like 2.5(4); we can
do it at some price.
Proposition 2.6 Let
be a linear order,
=
<
+
1
, each 1
a -
dense linear order of cardinality (as in the proof of 2.3(1)). Then
39
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
is a
+
-like linear order such that every j 1 which is neither successor
nor the rst element (under <
I
) satises
cf(i : i <
J
j, <
J
) =
and each member of has an immediate successor.
Denition 2.7 For a
+
-like linear order , a -Aronszajn tree is T =
(T,
T
, lev) such that
(a) T is a set of cardinality
+
,
(b) (T,
T
) is a partial order which is a tree, i.e. for every y T the set
x : x
T
y is linearly ordered by
T
,
(c) lev is a function from T to , T
j
=: y T : lev(y) = j,
(d) for every y T, lev is a one-to-one order preserving function from
x : x <
T
y onto j : j <
J
lev(y), so y j is naturally dened,
(e) for y T and j , lev(y) <
J
j there is z such that y <
T
z T,
lev(z) = j,
(f ) y : lev(y) = j has cardinality ,
(g) normality: if y ,= z, both in T
j
, j is neither successor nor the rst
element of (, <
J
) then x : x <
T
y , = x : x <
T
z,
(h) for y ,= z T there is j such that y j = z j and
(i)(j <
J
i y i ,= z i)
[we write y j = z j for z y].
Theorem 2.8 Assume that =
+
= 2
and
,
, < we have d
x,h
:
T
[]
x
such that:
if d
x,h
( x) = d
x,h
( y) then for some
t T
[]
x
we have
40
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() t
i
= y
i
z
i
,
() lev(
i < )(c(t
i
) = ),
() for ordinals i < divisible by we have
h() = (c(t
i
), f(y
i
( +
J
1)), f(z
i
( +
J
1))),
where = lev(
t).
Proof Like 2.3.
3 Constructions Related to pcf Theory
Lemma 3.1 1. Suppose that
(A)
i
: i < ) is a strictly increasing sequence of regular cardinals,
[[ <
i
< = cf() for i < and D is a -complete lter on
containing all co-bounded subsets of (follows by clause (D);
hence cf() ),
(B) tcf(
i<
i
/D) = , i.e. there is a sequence f
: < )
i<
i
such that
(i) < < implies f
<
D
f
,
(ii) (f
i<
i
)( < )(f <
D
f
),
(C) sets A
i
(for i < ) are such that the family A
i
: i <
is -independent in T()/D (i.e. if u, v are disjoint subsets of ,
[u v[ < then
iu
A
i
jv
A
j
,= mod D),
(D) [f
i : < [
<
<
i
for each i < .
Then Ens
(, ).
2. The linear orders in part (1) have exact density =:
i<
i
(see Def-
inition 1.9) and they are positively -entangled (see Denition 1.10).
Moreover, if f
i
: i < (but
/ pcf(
i
: i < ) for < ) then we can nd a lter D as required in
(A) + (B). So if > cf() = and < [[
of 1. It is <
A
by:
f
<
A
f
if and only if
(i < )(f
(i) ,= f
(i) & f
i = f
i & [f
(i) < f
(i) i A]).
Let u, v be disjoint subsets of , [u v[ < and for each u v let
t
= f
(,)
be pairwise distinct (for < ). We should nd < < as
in 1.1(1). Let
g
(i) =: minf
(,)
(i) : u v,
i
=: mini < : f
(,)
i : u v) are pairwise distinct.
Since [u v[ < cf(), i
< . Clearly g
i<
i
. Without loss of
generality i
= i
< )(g
(i) > ).
Claim 3.1.1 B D.
Proof of the claim: Assume not, so B ,= modD. For i B
let
i
<
i
,
i
< exemplify i / B, i.e. [
i
, ) g
(i)
i
. Dene
h
i<
i
by:
h(i) =:
_
i
+ 1 if i B;
0 if i B.
Now f
/D : < ) is conal in
i<
i
/D (i.e. clause (ii) of (B)), so there
exists < such that h < f
i
<
(remember that B , [[ < = cf() and (i B)(
i
< )). Since,
for each u v, (, ) (for < ) are pairwise distinct and < ,
there exists < such that ( u v)((, ) > ). Without loss of
generality < and hence sup
i\B
i
< . Now by the choice of we have
( u v)(f
< f
(,)
modD) and for every i B, g
(i)
i
. Hence
E
=: i < : f
(i) < f
(,)
(i) D and as D is -complete and > [uv[
42
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
we get
uv
E
D. By g
(i) , = mod D.
Choosing i in this (non-empty) intersection, one obtains g
(i)
i
<
i
+1 =
h(i) < g
(i) (the rst inequality see above, the third equality see choice
of h, the last inequality see choice of i), a contradiction. So B D,
proving the claim.
3.1.1
Remember that [f
i : < [ <
i
for each i < , and cf(
i<
i
/D) =
, D contains all co-bounded subsets of . By our hypothesis,
A =:
u
A
v
( A
) ,= modD,
so C =: i < : i
1
<
2
S &
1
,
2
u v f
(
1
,
1
)
(i) < f
(
2
,
2
)
(i).
Without loss of generality the sequence f
(,
)
i : u v) : S) is
constant (by hypothesis (D) of 3.1(1)). The conclusion should be clear now
(look at the denition of <
v
A
).
2) We will state the requirements and prove them one by one.
Claim 3.1.2 The linear orders constructed in the rst part have exact den-
sity .
Proof of the claim: Let us consider 1 = (1, <
A
). For each i < choose
a set X
i
such that [X
i
[
i
and f
i : < = f
i : X
i
.
Then f
:
i<
X
i
is a dense subset of (1, <
A
) (and its size is ).
Suppose now that 1, [[ = and assume that
0
is a dense
subset of , [
0
[ < . The set X = < : f
has cardinality , so
it is unbounded in . Let i() = mini < :
i
> [
0
[. Then
(i i())(
i
=: supf
(i) + 1 : f
0
<
i
)
(as
i
is a supremum of a set of [
0
[ <
i
= cf(
i
) ordinals <
i
). Let
i
= 0 for i < i(). Then
i
: i < )
i<
i
and, as f
: < ) is conal
43
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
in (
i<
i
, <
D
), for some () < we have
i
: i < ) <
D
f
()
. Since
( X ())(f
()
D
f
), we have
( X ())(i < :
i
< f
(i) D).
Consequently, for each X() we nd i
(i
) >
i
. As = cf() > [[, there is j (i(), ) such that the set
X
=: X : > () & i
= j
is unbounded in . Since [f
j : < [ <
j
< = cf(), for some
unbounded set X
)(f
j = ).
But now note that the convex hull of f
: X
in (1, <
A
) is disjoint
from
0
, a contradiction.
3.1.2
Claim 3.1.3 (1, <
A
) is positively -entangled.
Proof of the claim: Like in part (1).
Claim 3.1.4 If the sequence f
= f
1
, f
2
and
f
1
i = f
i = f
2
i & f
1
<
A
f
<
A
f
2
).
Plainly [
[ = , so [
[
+
. Since =: [
0
[ < and is a
limit cardinal, we have = ( + [[)
+
< . Let X = : f
and
choose X
1
X of size . Now we may nd B
: X
1
) D such
that for each j < the sequence f
(j) : X & j B
) is strictly
increasing (or just without repetitions). Then for some i() < the set
X
2
= X
1
: i() B
0
)(f
(i() + 1) = f
(i() + 1))
is of size (remember [
0
[ < = cf()), a contradiction with the choice of
.
3.1.4
This nishes the proof of the lemma.
3.1
44
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Lemma 3.2 Suppose that a is a set of regular cardinals satisfying
[a[
+
< min(a), = max pcf(a) and [ a > (max pcf( a))
<
].
1. Assume that = [a[, =
<
and for < , J is a -complete ideal
on a extending J
<
[a] and a
or just 2
: < ) be
J
<
[a]
-increasing conal in
a/J
<
[a] with
[f
[f
) : a) such that
for each a:
a,
< , and
for each b, < the set a :
= ,
= is ,= mod J.
Now we dene a linear order <
et
on f
: < as follows:
f
<
et
f
(a ) = f
(a ), f
() ,= f
() and
f
() < f
() f
.
45
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Readily <
et
is a linear order on the set 1 = f
<
et
f
we have
f
= f
as
= f
(,)
(for < (), < ) and (, )s are pairwise distinct. For each
< take
b such that f
(,)
b. We have
that [f
))
<
<
and
hence we may assume that for some g
: < ()):
( < )( < ())(f
(,)
= g
).
Let X = g
: u, Y = g
,
(X Y ) = X. Like in the proof of 3.1 one can show that
a : ( < )([ < : g
() > [ = ) = a mod J,
where g
() = minf
(,)
() : < (). Thus we can nd a such that
, =
(), ( < )(
<
) and
( < ())( < )
_
f
(,
)
() < g
()
_
and the sequence
_
f
(,
)
: <
) : S
_
is constant. Choose any
1
,
2
S,
1
<
2
and note that for every < () we have
f
(,
1
)
= f
(,
2
)
, f
(,
1
)
() < f
(,
2
)
(),
and f
(,
1
)
= g
= f
(,
2
)
. Thus
1
<
2
satisfy the condition
given by entangledness for t
s.
3) Let f
: < as follows:
f
<
pet
f
(a ) = f
(a ) and f
() < f
().
The rest is even simpler than in the proof of part (2) after dening <
et
(remember 1.2(6)).
4) It is similar to the proof of 3.1(2), noting that
if
f
= f
: < ) is <
J
increasing conal in
a for = 1, 2,
= cf() and
f
1
is free
then for some X []
,
f
2
X is free.
3.2
46
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proposition 3.3 1. Assume
(a) Ens
(
i
,
i
) for i < ,
(b)
i
are regular cardinals for i < ,
i
: i < ) is strictly increasing,
<
0
,
(c) J is a complete ideal on extending J
bd
,
(d) < T
+
J
(
i
: i < )) (=: sup[F[
+
: F
i<
i
and f ,= g F
f ,=
J
g),
(e)
i<
i
/J has the true conality as exemplied by f
: <
and for each i < ,
i
> [f
i : < [
<
(if for each i, max pcf(
j
: j < i) <
i
then we have such f
s).
Then Ens
(, ).
2. Assume that in part (1) we omit (d) but in addition we have
(f ) for each i < ,
i
[f
i : < [
[and we have such f
s e.g. if
i
max pcf(
j
: j < i)], or at least
liminf
J
(
i
) = sup
i<
i
, or
(f ) can be partitioned
5
to [[ many J-positive sets and for each
i < for J-almost all j < we have
j
[f
i : < [
(if
j
strictly increasing this means every large enough j).
Then there is a -entangled linear order 1 of cardinality .
3. Assume (f ) or (f ) + (g), where
(g) there is a decreasing sequence B
: < ) of elements of J
+
with
empty intersection.
Then in (2) we can get: 1 = 1
1
+1
2
such that for any uniform ul-
tralter D on the orders (1
1
)
/D, (1
2
)
i
, if f
: < ) is free
even exact density (, . . .).
5. In 3.3(1) we can weaken clause (d) to:
5
If j is non-decreasing then partition to cf() sets suces.
47
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(d)
for some F
i<
i
, [F[ = and for every F
F of cardinality
< we have
i < : g(i) : g F
) is with no repetition J
+
.
6. In 3.3(2) we can replace (f ) by
() there is h : such that i h(i),
i
[f
h(i) : < [
and for every i < , j < : h(j) i J
+
.
Proof Similar. (About part (3) look at the proof of 2.3(3)). However
we will give some details.
1) As Ens
(
i
,
i
) (by clause (a)), we can nd linear orders <
i
of
i
(for
<
i
) such that the sequence (
i
, <
i
) : <
i
) is -entangled. By clause
(d) we can nd g
i<
i
(for < ) such that
< < i < : g
(i) = g
(i) J.
Now for each < we dene a linear order 1
= (F, <
: < as follows:
f
<
(i) ,= f
(i), f
i = f
i and f
(i) <
i
g
(i)
f
(i).
It is easy to check that <
i
. As
in the proof of 3.1,
A =: i < : (f
j<i
j
)(
i
= [f
i[) = mod J
and for i A let f
i
exemplify i A. If G F is dense, [G[ cannot be <
as then it is <
i()
for some i() A and so for some <
i()
for no <
is
f
i() = f
i()
& f
(i()) <
i()
,
thus proving = dens(1
). The part if f
) : <
i
) is -entangled.
2) First we assume clause (f). As Ens
(
i
,
i
) and
i
[
i
[ where
i
=:
f
of
i
(for
i
) such
that (
i
, <
i
) :
i
) is -entangled. We dene the linear order <
of
F =: f
: < as follows:
f
<
(i) ,= f
(i), f
i = f
i, and f
(i) <
i
fi
f
(i).
The rest is as in 3.3(1).
Next we assume clause (f) instead of (f). So let A
i
: i < ) be a partition
of with every A
j
in J
+
and so necessarily
A
i
=: j A
i
:
j
[
i
[ = A
i
mod J.
Then we can choose
6
a function h such that
() h : , h(i) i,
i
[
h(i)
[ and for every i < we have
j < : h(j) i J
+
.
Let (
i
, <
i
) :
h(i)
) be a -entangled sequence of linear orders. Now
we dene a linear order on F =: f
: < :
f
<
(i) ,= f
(i), f
i = f
i, and f
(i) <
i
fh(i)
f
(i).
3) Without loss of generality, for each < + the set J
=: < :
f
: < & f
(0) < + . We
can nd B
J
+
(for < ), decreasing with and such that
<
B
=
and in the proof from (f) replace () by
(
) h : , h(i) i,
i
[
h(i)
[ and for every i < , < + we
have
j B
: h(j) i J
+
,
6
Actually we can replace the assumption (g) (in 3.3(3)) by the existence of such h.
49
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
and dene <
as
f
<
(i) ,= f
(i), f
i = f
i, and
[i = 0 f
(0) < f
(0)], [0 < i B
f(0)
f
(i) <
i
fh(i)
f
(i)] and
[0 < i / B
f(0)
f
(i) < f
(i)].
The rest is as before (we can replace by other cardinal but
0
).
4) For 3.3(1) see in its proof, other cases similar.
5) Really included in the proof of 3.3(1).
6) Really included in the proof of 3.3(2).
3.3
Proposition 3.4 1. Assume that:
(a) Ens
(
i
,
i
,
i
) for i < ,
(b)
i
: i < ) is a strictly increasing sequence of regular cardinals,
2
||
<
0
,
(c) J is a -complete ideal on extending J
bd
,
(d) < T
+
J
(
i
: i < )),
i<
i
< , = cf() and
( < )([[
<
< ),
(e) F = f
: <
i<
i
, f
,=
J
f
for ,= , and
[f
i : < [
<
<
i
,
(f ) if
i
= cf(
i
) <
i
, A J
+
then tcf(
iA
i
/J) = is impossible,
(g) ( < )([[
<
< ).
Then Ens
(, , ).
2. Assume in addition
(h)
i
[f
i : < [,
or at least
(h) cf() = and liminf
J
i
= sup
i<
i
,
50
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
or at least
(h) there is h : such that i h(i),
i
[f
h(i) : <
and = limsup
i<
h(i).
Then there is a (, )-entangled linear order 1 of cardinality .
3. Suppose also that
(i) we can partition into sets from J
+
(or clause (g) from 3.3(3)
holds).
Then we can get: for any uniform ultralter D on , 1
/D has two
isomorphic subsets with disjoint convex hulls of cardinality .
Proof Similar proof but for readers convenience we will present some
details.
1) We repeat the proof of 3.3(1) up to proving entanglness. To show
1
(
2
, ) u).
Claim 3.4.1 Assume that
()
i
: i < ) is a strictly increasing sequence of regular cardinals,
() J is a -complete ideal on extending J
bd
,
() a sequence
f = f
: < )
i<
i
is <
J
increasing,
()
i<
i
< = cf() and ( < )([[
<
< ),
() one of the following occurs:
(i) 2
||
<
0
, or
(ii)
f is free,
() if
i
<
i
for i < and A J
+
then tcf(
iA
i
/J) = is impossible,
51
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() a sequence (, ) : < , < () < ) of ordinals < satises
(
1
,
1
) ,= (
2
,
2
) (
1
,
1
) ,= (
2
,
2
).
Then there are a set X []
and a sequence h
: < ())
i<
i
such
that
(a) for each < () the sequence (, ) : X) is strictly increasing,
(b) ( X)( < ())(f
(,)
<
J
h
),
(c) h
(i))
i
) = mod J.
Proof of the claim: Since ( < )([[
<
< ) and cf() = we know
that for some X []
we have
( < ())(the sequence (, ) : X) is strictly increasing).
[Why? For < , < () dene f(, ) as follows:
if there exists such that (, ) > (, ) and [(, ), (, )]
(, ) : < = then f(, ) is this unique ,
otherwise f(, ) = 1.
By Fodor Lemma, there is a stationary set S such that supf[S()] <
. Since [[
()
< , on a stationary set X S the sequence f(, ) : <
()) does not depend on . This X is as required.]
By renaming we may assume that X = . Consequently, for each < ()
the sequence f
(,)
: < ) is <
J
increasing. Since = cf() > 2
||
or the sequence is -free, we may use [Sh 355, 1] to conclude that it has
a <
J
eub, call it h
(i) is a
limit ordinal. Since h
<
J
i
: i < ), wlog (i < )(h
(i) <
i
). Also
= tcf(
i<
cf(h
(i)) h
(i) <
i
, so by the assumption
() we have
i < : cf(h
(i)) <
i
or max pcf(
j
: j < i < cf(h
(i))) = mod J.
Since J is -complete we conclude
B
(i))
i
) = mod J,
nishing the proof of the claim.
3.4.1
52
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Note that we may assume that for some i
and h
<()
h
(i), and
B = i B
: for each
: < ())
<()
h
(i)
there are many ordinals < such that
( < ())(
< f
(,)
(i)).
We have to show the following.
Claim 3.4.2 B = mod J.
Proof of the claim: Assume not. Then, as B
= mod J, necessarily
B
B choose a sequence
i
: < ())
<()
h
B and [
i
, ) then ( < ())(
i
f
(,)
(i)).
For < () dene a function h
i<
h
(i) by
h
(i) =
_
i
+ 1 if i B
B,
0 if i B
or i B.
Now, for each , for every suciently large < we have h
<
J
f
(,)
.
Consequently, we nd
, ) h
<
J
f
(,)
.
But the ideal J is -complete, so for each [
, )
B
(i) < f
(,)
(i)) = mod J.
Now we may take [
(B
B) and get a
contradiction as in the proof of 3.1.1.
3.4.2
Remember that
[f
(,)
i : < ()) : < [ [f
i : < [
<
<
i
(not just <
i
), see clause (e) of the assumptions of 3.4(1). Hence there is
, )[ = ).
53
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
For each i B (dened above) we know that ( < ())(cf (h
(i))
i
),
and hence
<()
h
(i) is
i
-directed and the set
f
(,)
(i) : < ()) : [
, )
is conal in
<()
h
j<i
j
(for
< ()) such that
for every
: < ())
<()
h
, ) we
have ( < ())(f
(,)
i =
i
&
< f
(,)
(i)).
Now take any i B, i > i
(i) is
i
directed we can choose by
induction on <
i
a sequence
: <
i
)
i
such that for each <
i
sup
< <
and (
<)(<())(f
(
,)
(i) < f
(,)
(i)).
Now remember that the sequence <
i
: <
i
) exemplies Ens
(
i
,
i
,
i
).
So we apply this to g
(i)
f
(
2
,)
(i),
v f
(
2
,)
(i) <
i
g
(i)
f
(
1
,)
(i),
so we are done.
2) The proof is exactly like that of 3.4(1) except of two points. First we
have to dene a linear order 1 (rather than 1
on
i
(for T
i
= f
) : T
i
) is
a (
i
, )-entangled sequence of linear orders. This does no aect the proof
except in the very end when we use the entangledness.
3) Combine the proofs above.
3.4
Remark 3.5 1. We can also vary .
2. The 2
||
<
0
rather than just [[ <
0
is needed just to have
<
J
-eub (to use that if
i
= cf(
i
) <
i
. . .) so if f
: < is
(
i<
i
)-free, we can weaken 2
||
<
0
to [[ <
0
.
54
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
3. Instead of T
+
J
(
i
: i < )) we may use any = [G[, G
i<
i
such
that for every sequence g
(
++1
, 2
cf ()
) or Ens
(
++1
, 2
+
)
contains a club of
+4
.
2. If in addition 2
+4 (or 2
contains a club of
+4
and
+4
itself. (We can weaken the assump-
tions.)
3. We can add in part (2) the conditions needed for 1.7. Also in parts (1),
(2) the exact density of the linear orders is
+
provided cf()
+
.
Proof 1) By [Sh 400, 4], for some club C of
+4
,
() < C
+
> cov(
+
,
+
,
+
, 2),
and hence, if cf() and C then (
+
)
<
=
+
.
Let be an accumulation point of C of conality
1
and let A be
a closed unbounded set such that
A
++1
/J
bd
A
has the true conality
++1
and otp(A) = cf() (exists by [Sh 355, 2.1]). Now for A we have
max pcf(
++1
: A) is <
+
<
++1
if cf() by () (and
[Sh 355, 5.4]). Hence for some closed unbounded set B A we have
nacc(B) cov(
+sup(A)+1
,
+
,
+
, 2) <
+
.
Hence, if nacc(B) then max pcf(
+(+1)
: B) <
++1
. Wlog
otp(B) = cf(), B = A. Now if cf()
+
then we may apply 3.1 to
++1
: nacc(A) (for
) and get Ens
(
++1
, 2
cf ()
). We still have to
deal with the case cf() >
+
. We try to choose by induction on i ordinals
i
A
j<i
(
j
+ 1) such that
+
i
> max pcf(
+
j
: j < i) < , cf(
i
) =
+
.
55
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
For some the ordinal
i
is dened if and only if i < . Necessarily
is limit and by () + [Sh 371, 1.1] we have cf()
+
. Now for each
i < , Ens
(
+
i
+1
, 2
+
) holds by the part we have already proved (as
cf(
i
) =
+
). So we may apply 3.3(1).
2) Let C be as in the proof of part (1) and exemplifying its conclusion. If
= sup(C ) <
+4
, cf()
+
, 2
cf()
+
then we can apply 3.2(1)
(and the proof of part (1)). So we are left with the case cf() >
+
. Now
we can repeat the proof of part (1) the case cf() >
0
. Choose A as there
and also
i
A, but demand in addition Ens
(
+(+1)
, 2
+
),
i
acc(C)
and cf(
i
) =
+
, hence Ens
(
+(+1)
, 2
+
). In the end apply 3.3(2) to
+(
i
+1)
: i < ).
3) Similar to the proof of 3.3(4), 3.3(5).
3.6
Conclusion 3.7 Assume
0
. Then for arbitrarily large cardinals
there is are -entangled linear orders of cardinality
+
.
Proof Let > be given. We choose by induction on i < regular
cardinals
i
> such that Ens
(
i
,
0
+
j<i
j
) holds and
i
>
j<i
j
.
The inductive step is done by 3.6. Now for some -complete ideal I on
extending J
bd
,
j<i
j
/I has a true conality, say . By 3.3 there is a -
entangled linear order of cardinality , so if is a successor cardinal we are
done (as > ). If not, necessarily is inaccessible and letting =
j<i
j
clearly =
<
<
+4) such
that there is an entangled linear order in
+
i
, so in any case we are done.
3.7
4 Boolean Algebras with neither pies nor chains
Let us recall the following denition.
Denition 4.1 Let B be a Boolean algebra.
(a) We say that a set Y B is a chain of B if
(x, y Y )(x ,= t x <
B
y or y <
B
x).
(b) We say that a set Y B is a pie of B if
(x, y Y )(x ,= y x , y and y , x).
56
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(c) (B), the (algebraic) density of B, is
min[X[ : X B 0 and (y)(x X)(0 <
B
y B x
B
y).
Lemma 4.2 1. Suppose that
(a)
i
: i < ) is a strictly increasing sequence of regular cardinals,
is a regular cardinal,
(b) J is a -complete ideal on extending J
bd
,
(c) f
: < ) is a <
J
-increasing sequence of functions from
i<
i
,
conal in (
i<
i
, <
J
),
(d) for every i < , [f
i : < [
<
<
i
,
(e) A
B,
(f ) 2
=: sup
i<
i
and =
<
(so [[ ) and cf() .
Then there is a Boolean algebra B of cardinality such that:
()
B
() A
B and [/e[ n.
Then there is a Boolean algebra B of cardinality as in part (1) but
()
B
is strengthened to
()
B
,n
if:
() a
,
() B
n
is the nite Boolean algebra of subsets of n(n+1), and
for n, f
(i) ,= f
m
(i) and
x
(i)
then for some
0
< <
n1
< , the quantier free type
which a
0
, . . . a
n1
) realizes in B is equal to the quantier free
type which x
0
, . . . , x
n1
) realizes in B
.
Remark: 1) The case = sup
i<
i
is included.
2) Of course, no order Boolean Algebra of cardinality can satisfy ()
B
.
3) Again, B has density and if f
i :
< for i < . For each i < we can nd subsets F
2i,
(for < ) of
Y
2i
such that for any disjoint subsets X
1
, X
2
of Y
2i
, each of cardinality < ,
for some < we have F
2i,
(X
1
X
2
) = X
1
(possible as =
<
and
2
2i+1
> [Y
2i
[, by [EK] or see [Sh:g, Appendix 1.10]). We can nd a
sequence (j
i
,
i
) : i < ) such that
j
i
i < ,
i
< and for an unbounded set of j < for every
< for some < we have
2i, 2i+1 : 2i < & 2i, 2i+1 A
Y :
f Z
(2i),
() f (2i + 2) ,= f
(2i + 2),
() f (2j
i
) F
2j
i
,
i
f(2i) f
(2i + 1),
() f (2j
i
) / F
2j
i
,
i
f(2i) f
(2i + 1).
Let B be the -complete Boolean algebra of subsets of Y generated by the
family Z
0
, . . . , Z
n1
, say Z
= [Z
0
, . . . , Z
n1
] (where is a Boolean term)
and i < is such that f
m
(2i) : m < n) is with no repetition and
(m < n)(f
1
m
(2i) = f
2
m
(2i)) then:
(a) Z
1
m<n
[f
1
m
(2i)] = Z
2
m<n
[f
2
m
(2i)],
(b) for each m < n:
() either Z
1
[f
1
m
(2i)] = [f
1
m
(2i)] Z
1
m
and
Z
2
[f
2
m
(2i)] = [f
2
m
(2i)] Z
2
m
,
() or Z
1
[f
1
m
(2i)] = [f
1
m
(2i)] Z
1
m
and
Z
2
[f
2
m
(2i)] = [f
2
m
(2i)] Z
2
m
.
Proof of the claim: Check the denition of Z
.
4.2.1
Clearly B is a Boolean algebra of cardinality . Now the proof of B
has no chain of cardinality is similar to the proof of 3.1, 3.2 noting that
for each i:
() if ()
2i
2i+1
and for arbitrarily large <
2i
, for arbitrarily
large <
2i+1
we have (, )
then we can nd (
1
,
1
) , (
2
,
2
) such that
1
<
2
&
1
<
2
.
Up to now, the use of the pairs 2i, 2i +1 was not necessary. But in the proof
of B has no pie of cardinality , instead of () we use:
() if () of () holds
then we can nd (
1
,
1
) and (
2
,
2
) such that
1
<
2
&
1
>
2
,
easily nishing the proof
4.2
Conclusion 4.3 For a class of cardinals , there is a Boolean algebra B of
cardinality
+
, with no chain and no pie of cardinality
+
. [We can say, in
fact, that this holds for many .]
Proof For any regular , if 2
>
.
59
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Next we choose by induction on i < cardinals
i
Reg [
+4
,
+4],
i
> max pcf(
j
: j < i). Let be minimal such that sup
i<
i
and
pcf(
i
: i ). Then (by [Sh 345a, 1.8], replacing
i
: i < by a
subset of the same cardinality, noting
i
: i A J
<
[
i
: i < ] when
A []
<
) we have = tcf(
i<
i
/J
bd
). Also, as [,
+4] is regular,
it is a successor cardinal; now the conclusion follows by 4.2. Generally,
for any let
0
= 0,
0
=
+
. Choose by the induction on n,
n+1
and
i
:
n
i <
n+1
), u
n
and regular
n+1
such that
n+1
=
n
+
n
,
i
:
n
i <
n+1
) is a strictly increasing sequence of regular cardinals in
[
n
,
(
n
)
+4] such that
i
> max pcf(
j
:
n
j < i) (possible by [Sh 400,
2] as above). Let u
n
[
n
,
n+1
), [u
n
[ =
n
be such that
iun
i
/J
bd
un
has a true conality which we call
n+1
. Lastly, for some innite v ,
nv
n
/J
bd
v
has a true conality, which we call . By renaming v = ,
u
n
= [
n
,
n+1
). Then =: sup
n
n
, ,
i
: i < ) are as required in 4.2, if
we let:
J = u : for every large enough n, sup(u
n
) <
n
.
One point is left: why is a successor cardinal? Because it is in [sup
n<
n
,
n<
n
]
and either
n<
n
[sup
n<
n
]
0
2
n
n
<
(
n<
n)
+4,
or the rst attempt succeeds for =
n<
n
.
4.3
We have actually proved the existence of many such objects. If we waive
some requirements, even more.
Proposition 4.4 For any regular cardinal we can nd , J, ,
i
(for
i < ) as in 4.2 and such that:
(x) is a successor cardinal,
(y) for each i for some regular cardinal
i
we have
i
=
+
i
and (
i
)
=
i
,
(z) one of the following occurs:
(i) is a regular cardinal <
0
, > 0 and J = J
bd
,
(ii) =
j
J
bd
cf ()
60
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
and
= (
)
sup{
i
:i<}
.
Lemma 4.5 Assume
i
: i < ), , J, , f
(
i
,
i
) or at least for some club C of ,
if i < and i = supj C : [i C j[ 1 then Ens
(
2i
, [f
(2i) :
< [).
Then the conclusion of 4.2(1) holds.
[We can weaken (f ) as in 3.3(6).]
Proof For each i < let (
2i
, <
2i
) : = f
we let:
f Z
(2i), f (2i + 2) ,= f
(2i + 1).
4.5
Discussion: Now instead of using on each set
) : <
g()
a
linear order we can use a partial order; we can combine 4.6 below with
4.2(2) or with any of our proofs involving pcf for the existence of entangled
linear order.
Lemma 4.6 1. Assume
i
: i < ), , J, , f
: < ) are as in
(a)(d) of 4.2(1) and
(e) for each i < there is a sequence
P = P
: <
i
) where
i
= [f
i : < [, each P
is a partial order,
P is (, )-
entangled which means:
if u
0
, u
1
, u
2
are disjoint subsets of
i
of cardinality < and for
u
0
u
1
u
2
, t
[= t
< t
,
u
1
P
[= t
> t
,
u
2
P
[= t
, t
are incomparable.
61
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Then the conclusion of 4.2(1) holds.
2. Assume as in (1) but
(e) for some A , A / J, A / J and
(e)
1
like (e) of 4.6(1) for i A with u
2
= ,
(e)
2
like (e) of 4.6(1) for i A, with u
0
= u
1
= , u
2
= 1 (so we
can use P
= (
i
, =)).
Then the conclusion of 4.2(1) holds.
3. We can weaken
i
= [f
i : < [ as in 4.5(f ).
Proof Similar to earlier ones.
4.6
5 More on Entangledness
Proposition 5.1 Suppose that
i
: i < i()) is a strictly increasing se-
quence of regular cardinals, T
i
i
>
2 is closed under initial segments,
i + 1 < i() [T
i
[ <
i+1
and the set
B
i
=
i
2 : for every <
i
, T
i
and 2
<
< 2
[ then
Ens(, ), see [Sh 430, 3.4].
Proof Clearly [T
i
[
i
(as B
i
,= ). So let n < , i
0
< i
1
< . . . <
i
n1
< i(), and
B
i
,=
,
and let u n. We should nd < < such that ( < n)(
<
x
<
x
_
if (u)
.
62
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Note that ()
n
is exemplied by w =: and ()
0
says (more than) that the
conclusion holds, so this suces. Hence assume ()
m+1
is exemplied by
w
and we shall nd w w
exemplifying ()
m
, [w[ =
im
. Without loss
of generality m u (otherwise replace each T
im
m
: w
by
1 () : < g())). Let for <
im
and T
i
2:
w
=: w
: =
m
, ( w
)( < &
m
= &
m
<
x
m
),
w
=
_
w
: T
im
2.
As in (B
im
, <
x
) there is no monotonic sequence of length
+
im
, clearly [w
im
. Moreover,
[w
[ [ T
im
: g() = [ sup[w
[ : T
im
2 [T
im
[
im
,
and hence [
<
im
w
[
im
+ [T
im
[. But [T
im
[ <
i
m+1
and
im
<
i
m+1
.
Hence we nd () w
<
im
w
exemplify () / w
()
w
, so
< (),
m
=
m
()
and
m
<
x
m
()
.
Then some
<
im
, we have
m
()
=
m
,
m
()
(
) = 1,
m
) = 0.
So for some unbounded set A
i
the sequence
: A) is strictly
increasing in and also
: A) is increasing. Let w =:
: A
w
. It exemplies ()
m
, hence we nish.
5.1
Proposition 5.3 1. Assume that
(a) = max pcf(a
[ <
min(a
),
(c) a
is (
,
+
, 2)-inaccessible,
(d) for n < and distinct
0
,
1
, . . . ,
n
< () we have
a
0
n
_
=1
a
/ J
<
[a
0
].
Then Ens(, , ()).
63
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
2. Assume in addition that
(d) if u [()]
<
, () u then a
u
a
/ J
<
[a
],
(e) if a
) < ,
(f ) ( < )([[
<
< ).
Then Ens
(, , ()).
Proof As in 3.
5.3
Proposition 5.4 For any cardinal satisfying ( < )(2
< 2
) there is
a successor cardinal [, 2
] be any cardinals
(we shall try to nd
+
[, 2
= 2
) and < 2
< 2
.
First assume 2
<
= 2
: < )
is eventually constant so for some < we have 2
= 2
<
but 2
<
= 2
,
a contradiction to the choice of . Thus we have 2
<
< 2
= 2
. Assume
= + 2
<
or just 2
> 2
<
. The proof splits to cases: if cf(2
) is
a successor, use cases B or C or D, if cf(2
.
By 3.6(2).
Case B: cf(2
) (as in
Case B). But on the other hand, if (iii) holds, the conclusion of [Sh 410,
4.3] gives more than we asked for. In both cases there is an entangled linear
order in cf(2
) (which is a successor).
64
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Case D: is singular, not strong limit.
By [Sh 430, 3.4] there are regular cardinals
i
(for i < cf()) such that the
sequence
i
: i < cf()) is increasing with limit , < 2
i
, 2
i
: i < cf())
is strictly increasing and
for all such that = cf() 2
i
there is a tree T
i
, [T
i
[ =
i
or at least 2
|T
i
|
= 2
i
and T
i
has
i
-branches.
If for some i either 2
i
is a successor or cf(2
i
) is a successor we
nish as in Case B, if cf(2
i
) is a limit cardinal, we nish as in Case A (or
use 5.5).
5.4
Proposition 5.5 Assume 2
)
+
.
Proof Let be the rst singular cardinal > such that ( < )(pp
() >
2
(as 2
is singular so
pp(2
) (2
)
+
> 2
(2
= 2
).
Lastly apply 3.2(1).
5.5
Proposition 5.6 If
+
<
0
< and
+4
< cf() < 2
, then (a) or
(b) holds:
(a) there is a strictly increasing sequence
i
: i < ) of regular cardinals
from (, ), = cf() [, cf ()] Reg and
i
> max pcf(
j
: j <
i), such that
+
= tcf
i<
i
/J
bd
,
(b) there is a strictly increasing sequence
i
: i < ) of regular cardinals
from (, ) such that
i
> max pcf(
j
: j < i) and Ens(
i
,
i
) and
+
= tcf
i<
i
/I, I a proper ideal on extending J
bd
.
Moreover for some (
0
, ), <
0
and there is a sequence b
i,j
:
i < , j <
+
) such that b
i,j
Reg
0
, [b
i,j
[ , each
i,j
b
i,j
is (
0
,
+
,
0
)-inaccessible (i.e. a Reg
0
, [a[
max pcf(a) < ) and
j
1
< j
2
<
+
b
i,j
1
b
i,j
2
= ,
i
= max pcf(b
i,j
),
= sup(b
i,j
), J
bd
b
i,j
J
<
i
[b
i,j
].
(This implies Ens(
i
,
i
).)
65
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Remark 5.7 1. Why
+
instead of
= cf(
) (, pp
+
J
bd
cf()
())? To
be able to apply [Sh 410, 3.3] in case III of the proof of [Sh 410, 4.1].
So if
i
: i < cf()) ts in such a theorem we can get
+
.
2. We could have improved the theorem if we knew that always
cf([]
, ) = + suppp
+
i
, <
J
bd
cf()
) =
+
and
0
>
0
. The proof will split to cases. Wlog
0
> cf().
Case I: max pcf(
+
j
: j < i) < for i < cf().
So for some unbounded A cf() we have
(i A)(max pcf(
+
j
: j A i) <
+
i
).
Consequently a =
+
i
: i A satises the demands of 3.6 and hence (a)
holds true with = cf(),
i
=
i
.
Thus assume that Case I fails. So there is such that
0
< < , cf() <
cf() and pp
<cf()
() > . Choose a minimal such . Then, by [Sh 410,
3.2], we have:
() [a Reg
0
& supa < & [a[ < cf()] max pcf(a) < .
By [Sh 355, 2.3] in the conclusion of () we may replace < by <
and we get
()
[a Reg
0
& supa < & [a[ < cf()] max pcf(a) < .
Let = cf(). Then pp() = pp
<cf()
() (and pp
<cf()
() > ). Wlog
<
0
.
Case II: > (and not Case I).
By [Sh 371, 1.7], if >
0
and by [Sh 430, 6.x] if =
0
we nd a strictly
increasing sequence
i
: i < ) of regular cardinals, =
i<
i
and an ideal
J on extending J
bd
(if >
0
then J = J
bd
) such that
+
= max pcf(
i
: i < ) = tcf
i<
i
/J.
66
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
If >
0
, since we may replace
i
: i < ) by
i
: i A) for any
unbounded A , we may assume that
i
> max pcf(
j
: j < i). If
=
0
this holds automatically, so in both cases we can apply 3.6. So we
get (a), as > .
Case III: .
So
+4
cf(). Let
T =: C cf() : otp(C) =
+3
, C is closed in sup(C) and
max pcf(
+
i
: i C) < .
[Why there are such Cs? For any < cf(), cf() =
+3
we have a club C
of
+
/J
bd
+3
) =
+
. Now C
i
: i < ) will do.]
For each C T try to choose by induction on i <
+
, b
i
= b
i,C
and
i
=
i,C
such that:
(i) b
i
Reg
j<i
b
j
0
,
(ii)
i
C
j<i
(
j
+ 1),
(iii)
+
i
pcf(b
i
),
(iv) [b
i
[ ,
(v) all members of b
i
are (
0
,
+
,
0
)-inaccessible,
(vi)
i
is minimal under the other requirements.
Let (b
i,C
,
i,C
) be dened if and only if i < i
C
(). So success in dening
means i
C
() =
+
, failure means i
C
() <
+
.
Subcase IIIA: For some j <
+
, for every C T with min(C) j we
have i
C
() <
+
, so we cannot dene b
i
C
(),C
,
i
C
(),C
.
Let C, i
C
() be as above. Let
C
=
i<i
C
()
i,C
, so
C
C. Now, if C
C
then (by [Sh 355, 1.5B]) as pp
()
+
>
+
, there is a
Reg (, ),
[a
[ such that
+
pcf(a
Reg (, )
of cardinality consisting of (,
+
,
0
)-inaccessible cardinals such that
pcf(c
). Now , c
i<i
C
()
b
i,C
cannot serve as
i
C
(),C
, b
i
C
(),C
, so
necessarily
+
/ pcf(c
i<i
C
()
b
i,C
). Hence wlog c
i<i
C
()
b
i,C
. So
+
i
: i C
C
pcf(
_
i<i
C
()
b
i,C
) and [
_
i<i
C
()
b
i,C
[ .
67
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
By the proof of [Sh 400, 4.2] (or see [Sh 410, 3]) we get a contradiction
(note that cf() >
+4
does not disturb).
Subcase IIIB: For every j < cf() there is C T such that min(C) > j
and i
C
() =
+
, i.e. b
i,C
,
i,C
are dened for every i <
+
.
We will show
() for every j() < cf() there is
pcf(
+
j
: j < cf())
j()
such that
() Ens(
= max pcf(b).
Moreover, b can be divided to
+
subsets of cardinality , no
one in J
<
[b] and
( b)( > max pcf(b ))
(even is (
0
,
+
,
0
)-inaccessible).
Why does () suce?
Suppose that we have proved () already. So for i < cf() we can choose
i
,
i
<
i
= cf(
i
) pcf(
+
j
: j < cf()) as required in (). Since
(i)(
i
< ), wlog the sequence
i
: i < cf()) is strictly increasing. By
induction on < cf() choose strictly increasing i() < cf() such that
i()
> max pcf(
i()
: < ).
Let i() be dened if and only if < (). So () is limit,
+
= max pcf(
i()
: < ()), and
i()
> max pcf(
i()
: < ),
i()
is strictly increasing and Ens(
i()
,
i()
). Thus applying [Sh 355, 4.12]
we nish, getting clause (b) of 5.6.
Why does () hold?
Choose C (j(), cf ()) of order type such that
i
: i <
+
), b
i
: i <
+
)
are well dened and
max pcf(
+
i
: i C) <
(possible by our being in subcase IIIB, see the denition of T). Let d =:
i
: i <
+
and let b
C
(for < ) such that
: B
[d], otp(B
) = .
68
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Clearly max pcf(
+
i
:
i
B
) = , since
+
i
:
i
B
[d] but it is
not a subset of any nite union of b
[c],
=:
jC
b
j
,
we nd (by [Sh 371, 2.6]) a subset a of a
: B
),
+
pcf(b
) we have
(by [Sh 345a, 1.12]) pcf(
B
b
). Let
c
=: a
_
jB
b
j
,
= .
We can apply 3.2 and get that there is an entangled linear order of cardinality
j()
<
pcf(
j
: j < cf()).
The assumptions of 3.2 hold as the c
i
:
i
B
), pcf(
_
jB
b
j
) = pcf(c
) and
1
a max pcf(a
1
) <
1
,
as
1
is (
0
,
+
,
0
)-inaccessible and
=
sup
+
i
: i C >
j()
>
0
.
So clause () of () holds and clause () was done along the way. Thus we
nish subcase IIIb and hence case III.
5.6
Conclusion 5.8 For as in 5.6 there is a Boolean algebra B of cardinality
+
satisfying ()
B
+
, ()
B
sup
i<
i
from 4.1 (and also there is an entangled linear
order in
+
).
Proof If (a) of 5.6 holds, apply 4.2. If (b) of 5.6 holds use 4.5.
5.8
Denition 5.9 1. pcf
ex
[a] b[ ).
2. J
ex,
<
[a] =: b : b a and for some c a we have: [c[ < and
b c J
<
[a].
69
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proposition 5.10 Assume that is a singular cardinal which is a x point
(i.e. =
) and
< .
1. For some successor cardinal
+
(, pp
+
()) there is an entangled
linear 1 order of cardinality
+
and density (
, ].
2. If
0
,
+
+4
0
pp
+
() then we can nd an entangled linear order
1, [1[ =
+
(
0
,
+
+4
0
) of density (
, ].
3. In both parts we get also a Boolean Algebra B satisfying ()
B
+
, ()
B
of 4.2.
4. In both parts 1), 2), if is an interval of 1 or [1]
+
then
dens() = dens(1). This applies to 5.10.1, 5.10.3, 5.10.2, too.
Proof Let
i
: 1 i < cf()) be a strictly increasing continuous
sequence with limit . Wlog
1
>
0
>
+ cf(),
0
cf(
i
) <
maxcf(),
1
.
1) We try to choose by induction on i < cf() regular cardinals
i
such
that
i
<
i
< , max pcf(
j
: j < i) <
i
,
and there is an entangled sequence of linear orders each of cardinality
i
of
length max pcf(
j
: j < i) (i.e., Ens(
i
, max pcf(
j
: j < i))). For some
,
i
is dened if and only if i < . Clearly, is a limit ordinal cf(),
and =: max pcf(
j
: j < ) is > [as otherwise < (as is regular
by [Sh 345a, 1.x]), so there is
as required among ( +
)
+
: <
( +
)
+4
]. So clearly < = cf() < pp
+
() and by 3.3(2) there is an
entangled linear order of cardinality and density
i<
max pcf(
j
: j <
i) . If is a successor cardinal then we are done. Otherwise, clearly
+
+4
pp
+
(), and hence we can apply part (2).
2) It follows by the claims below, each has the conclusion of 5.10(2) from
assumptions which are not necessarily implied by the assumption of 5.10(2),
but always at least one applies.
Claim 5.10.1 1. Assume cf()
< 2
, and
+
+4
pp
.
2. If in addition ( < )([[
of (
,
+
, 2)-inaccessible
cardinals, [b[ , b is the disjoint union of b
) is
the same for < and J
bd
b
J
<
++1[b] and
+(+1)
pcf(b
).
Proof of the claim: 1) Of course we can decrease as long as
< ,
cf() ,
+
+4
pp
,
(ii)
i
<
+4
is a successor ordinal,
(iii)
+
i
pcf(b
i
),
(iv)
i
is the rst successor ordinal for which
+
i
/ pcf(
j<i
b
j
),
(v) all members of b
i
are (
0
,
+
, 2)-inaccessible
(i.e. b
i
& a (
0
, ) & [a[ max pcf(a) < ),
(vi) b
i
has cardinality .
Note that this is possible, since if [b[ then pcf(b) cannot contain the
interval [,
+
+4
]Reg (see [Sh 410, 3]). Let d =:
+
i
: i < , let b
[d] :
pcf(d)) be as in [Sh 371, 2.6]. Note that we know pcf(d) [,
+
+4
]
(by [Sh 400, 4.2]). Let pcf(d) be minimal such that otp(b
[d]) , so
necessarily is a successor cardinal. Let d
: < ) be a partition of b
[d]
to pairwise disjoint subsets of order type . Let b
=
b
i
:
+
i
d
(still
i
> .
2. In addition
i
pcf
ex
(a
i
) for some sets a
i
Reg
i
j<i
j
of
(,
+
, 2)-inaccessible cardinals of cardinality . If cf() >
0
then
i<cf()
i
/J
bd
cf ()
has true conality.
Proof of the claim: Choose
(, ) such that
, ) & cf(
) pp
) <
(exists, as is (,
+
, 2)-inaccessible; see [Sh 355, 2.3]). Let b
i
,
i
(for i < )
be as in the proof of 5.10.1; so min(b
i
) >
, d =:
+
i
: i < , b
[d] :
pcf(d)) be as in [Sh 371, 2.6]. Let pcf(d) be minimal such that
otp(b
. Let a =:
i<
b
i
and let b
pcf
ex
(a
) and c
, [c
e
b
[a] c
i<
b
i
).
So by renaming = 0 (so c
satises
Ens(, ). As in the proof of [Sh 371, 1.5] and [Sh 430, 6.5] one of the
following holds:
()
1
cf() >
0
and for some S cf() unbounded, otp(S) = cf(), and
for S, = tcf(
/J
bd
S
),
()
2
cf() =
0
and for some increasing sequence
/I).
If ()
1
holds then (by the choice of
0
) without loss of generality S
: < ) is
1
so we can apply 3.1. In both cases we get the
desired conclusion.
5.10.2
72
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Claim 5.10.3 Assume that is a singular strong limit cardinal, or at least
that it is (, , 2)-inaccessible for every < , cf() < and
+
+4
pp(). Then
1. We can nd
+
(,
+
+4
) in which there is an entangled linear
order with density .
2. Moreover we can nd a strictly increasing sequence
i
: i < cf())
with limit , max pcf(
j
: j < i) <
i
,
+
= max pcf(
i
: i <
cf()). Letting
i
= (
j<i
j
+
)
+
we can also nd a set a
i
Reg
i
j<i
j
of (
+
j
i
,
+
j
i
, 2)-inaccessible cardinals of cardinality
j
i
such that
i
pcf
ex
i
(a
i
) and j
i
i. Also cf() >
0
implies =
tcf(
i<cf ()
i
/J
bd
cf ()
) and (i < cf())(j
i
= i).
Proof of the claim: Choose
i
: i < cf()) be strictly increasing contin-
uous with limit . By induction on < cf() we choose
(,
+
+4
)
and
=
+
and
7
= sup(
pcf(
: < cf()) [
+4
) such that
= max pcf(
>
<
and <
<
+
+4
, and
without loss of generality
+4
i
<
i+1
. Let
+
be
, b
=:
: < cf()
(for < cf()), = max pcf(
: < cf()), a =:
b
: < cf()]
[a] :
pcf(a)) be as in [Sh 371, 2.6].
First assume cf() >
0
. For every limit < cf() let e
be a nite
subset of pcf(a
) such that
() for some
< , (a
)
e
b
[a]
i
: i < cf() and
i
<
,
7
We could have asked = max{max pcf({
:
< cf()} are pairwise disjoint (while omitting few
for each )
73
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
() for every < and for every e
, we have
pcf(
i
: i < cf() &
<
i
<
).
(Exist by [Sh 430, 6.x].)
So for every e
we
have = tcf
S
/J
bd
S
. Note that without loss of generality
>
<
: < ), so by 3.3(2)
we can get the desired conclusion.
Now assume cf() =
0
. Use [Sh 430, 6.7] to nd nite e
pcf(a
)
for < such that
< < max(e
0
) < min(e
)
(the
0
, are for empty e
n
s), and so otp(e) = , sup(e) = where
e =
and a
= b
[a] :
n
e
n
; hence pcf(e). Dene
h : pcf(a) by h() = maxn < : Ens(,
n
) or n = 0. By 3.4(2) it
suces to prove, for each n < , that e : h() n J
<
[e]. This can
be easily checked.
5.10.3
3), 4) Left to the reader.
5.10
Remark 5.11 1. Under the assumptions of 5.10.2 we can get
() there are a successor cardinal
+
(,
+
+4
) and an increasing
sequence
=
i
: i < ) with limit such that
<
+
,
+
= tcf(
i<
i
/J), J an ideal on extending J
bd
i
is (
,
+
, 2)-inaccessible (where
= min
is (
,
+
, 2)-
inaccessible) and
there is an entangled linear order of cardinality
+
.
2. In the proof of 5.10.3 we can have a =
i<cf()
a
i
, = sup(a
i
) for each
i < cf(), and let a =
: , < cf(),
a
i
=
a : a
i
but for no < does
j
: j < cf()
and apply 3.2(1).
Proposition 5.12 Let b
a
i
for i < such that [a
i
e
i
b
[a
i
][ < , and pcf(
i<
e
i
).
[And if cf() >
0
, a
i
increasing with i and S = supS then
pcf(
iS
e
i
).]
4. Assume < min(a), and each a is (, , 2)-inaccessible and >
0
. If pcf
ex
(a).
2) By [Sh 345a, 1.16].
3) Easy, as in [Sh 371, 1].
4) Without loss of generality = max pcf(a), a has no last element and
a / pcf
ex
(a ),
so J
bd
a
J
ex,
<
[a] (see Denition 5.9). We are going to prove the statement
by induction on .
If a (= b
a & pcf
ex
(a
) sup( pcf
ex
(a
)) = sup( pcf
ex
(a)).
75
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
We can try to choose by induction on i,
i
pcf
ex
<
+1
< max pcf(
i
: i < [a[
+
max pcf(
i
: i < ) =
, a
contradiction). So for some < [a[
+
,
i
is dened if and only if i < . If
max pcf(
i
: i < ) < we can get a contradiction to () (by 5.12(1)). If
the equality holds, we get contradiction to ().
5.12
We want to state explicitly the pcf theorems behind 5.10.
Proposition 5.13 Assume is a singular cardinal which is a x point and
0
< .
1. There are , (
i
,
i
) : i < ) such that:
(a)
i
: i < ) is a strictly increasing sequence of regular cardinals,
is a limit ordinal cf(),
i
Reg
0
,
i
= max pcf(
j
:
j < i) <
i
, Reg (, pp
+
()), = tcf(
i<
i
/J
bd
),
(b)
i
=
i
j
: j < (
i
)
+
) is strictly increasing,
i
j
Reg
i
i
,
i
= tcf(
i
j
/J
bd
(
j
)
+
),
i
j
> max pcf(
i
: < j)
2. Assume in addition
0
and
+
+4
0
pp
+
(). Then for some
<
+4
, letting =
++1
, we can nd a strictly increasing sequence
i
: i < ) of regular cardinals of the length = cf() < cf(),
i
Reg
,
i
> max pcf(
j
: j < i), = max pcf(
i
:
i < ) and letting
i
= (
j<i
j
+
)
+
we can also nd sets a
i
Reg
i
j<i
of (
j
i
,
+
j
i
, 2)-inaccessible cardinals of cardinality
j
i
such that
i
pcf
ex
i
(a
i
), j
i
i. Also if cf() >
0
then =
tcf(
i<cf ()
i
/J
bd
cf ()
).
3. If part 2) does not apply then in 1) is a successor.
Proof Let
i
: 1 i < cf()) be increasing continuous with limit ,
1
>
0
and wlog if is (, cf(), 2)-inaccessible then
[
0
, ) & cf(
) cf() pp(
) < ,
and hence
a [
0
, ) Reg & [a[ cf() & sup(a) < max pcf(a) < .
1) Try to choose by induction on i < cf(),
i
and a
i
,
i
,
i
,
+
i
, j
i
such
that
76
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
()
i
j
: j
i
,
i
(
j
i
, ) Reg,
() >
j
i
,
()
i
,
i
,
i
are as required in 1).
So for some , (j
i
,
i
,
i
,
i
) is dened if and only if i < , in fact is limit
and pcf(
i
: i < ) , (as in the proof of 5.10.1). For some unbounded
set A we have max pcf(
i
: i A) > but it is minimal under this
restriction. Now restricting ourselves to A (and renaming) we nish.
2) It follows from 5.10.1, 5.10.3.
3) Should be clear.
5.13
Let us nish this section with stating some results which will be devel-
oped and presented with all details in a continuation of the present paper.
Proposition 5.14 Assume:
(A) (a) T
j
i<j
i
,
(b) T closed under initial segments,
(c) T
j
=: T
i<j
i
,= for j ,
(d) for j < , T
j
we have (
j
<
j
)(
) T),
(e) [T
[ = = cf() >
i<
[T
i
[ > ,
(f ) for T
i
, I
is a -complete ideal on
i
,
(g) is a regular cardinal, cf() ,
(B) (a) J is an ideal on extending J
bd
,
(b) g
i
is a function from T
i
to
i
,
(c) for i < , <
i
, I
i
is a -complete ideal on
i
,
(d) if
,
: <
, ( < )(
=
,
i()) and
: <
T
i
for
<
such that:
() g
i
(
) : <
I
i
g
i
()
(for <
)(
,
) and
,
(i) / B
,
77
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(C) for i < , 1
i
: <
i
) is a sequence of linear orders with universe
i
such that:
if () < ,
<()
1
)( <
)( < ())(
,
/ B
)
and () = u v, u v =
then for some
1
,
2
< () we have
u 1
i
[=
1
,
<
2
,
,
v 1
i
[=
1
,
>
2
,
.
Then there is a (, )-entangled linear order of cardinality .
Remark 5.15 1. The proof is derived from the proof of 3.3 (and so from
[Sh 355, 4.10]).
2. Are the assumptions reasonable? At least they are not so rare, see
[Sh 430, 5].
Proof The desired linear order 1 has the universe T
< ,
i() < and
( < ())( < )(
=
,
i())
and
: <
I
i
g
i
(
,
i)
for some < we have
( <
)(
,
i =
,
i) and ( <
)(
,
(i) B
) = mod J,
78
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(d2) if
( <
i) : <
i<
i
/J),
i
>
i
=: max pcf(
j
: j < i),
(B) J a -complete ideal on the limit ordinal ,
i
= cf(
i
) > ,
(C) g
i,j
:
i
i,j
for j < j
i
are such that for any w [
i
]
<
for some
j < j
i
the restriction g
i,j
w is one to one,
(D) a = a
i
: i < ), a
i
(
<i
j
we have
i : (, j
) a
i
, = mod J,
(E) Ens
(
i
,
(,j)a
i
,j
).
Then for some T the assumptions of 5.14 hold for
i
=
i
,
i
=:
(,j)a
i
,j
.
Proposition 5.18 In 5.17, (C) + (D) + (E) holds if
(C) Ens
(
i
, ),
(D) ()
||
i<
i
but for i < ,
|i|
= (so is a regular cardinal) or
()
||
i<
i
and there is a regular ultralter E on disjoint from
J A : otp(A) < .
Proposition 5.19 Suppose (A), (B) as in 5.14 and
(C) there is a sequence 1
i
: <
i
) of partial orders with universe
i
such
that
if () < ,
<()
1
)( <
)( < ())(
,
/ B
)
79
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
and () = u v, u v = ,
then for some
1
,
2
< () we have:
u 1
i
[=
,
<
2
,
,
v 1
i
[=
,
<
2
,
.
Then there is a Boolean Algebra B, [B[ = with neither chain nor pie of
cardinality ; moreover for < , B
1
,= g
2
mod D and
if < (), A E then g
1
[A] D,
(c) 1 is a linear order of the cardinality .
Then there exists a sequence f
/D < f
/D < f
/D).
In particular, the linear order 1
/D is not -narrow.
Proof Choose pairwise distinct a
:
1 be given by f
(i) = a
g(i)
. Note that if
1
,=
2
then i < : f
1
(i) =
f
2
(i) = . If
1
=
2
= but
1
,=
2
then
i : f
1
(i) = f
2
(i) = i : g
1
(i) = g
2
(i) / D
80
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(as g
1
,= g
2
mod D). Consequently, if (
1
,
1
) ,= (
2
,
2
) then f
1
/D ,=
f
2
/D. Suppose now that < < . Let A = < : a
< a
. Assume
that A E and let < (). Then
i < : f
(i) < f
(i) = i < : a
g(i)
< a
g(i)
= g
1
[A] D
and hence f
/D < f
/D > f
/D.
Now clearly 1
/D.
So if the cardinality of BA
inter
(1) is regular we can just quote 1.5 (a) (c)
(here =
0
, see Denition 1.3). Otherwise, just note that for a linear
order , if a
< b
(for = 0, 1) and [a
, b
) BA
inter
() are comparable
and a
0
, b
0
, a
1
, b
1
is with no repetition then
a
0
<
J
a
1
(b
0
<
J
b
1
);
this can be applied by the statement above.
6.1
Remark 6.2 Proposition 6.1 shows that entangledness can be destroyed
by ultraproducts. Of course, to make this complete we have to say how one
can get D, E, g
:
(for < 2) are given by g
0
(i, j) = i, g
1
(i, j) = j. Then E, D, (), g
. Let g
[A] D T,
81
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
and the conclusion by
< () : f
/D < f
/D T.
3. A kind of entangledness can be preserved by ultraproducts, see 6.4
below. More entangledness is preserved if we put additional demands
on the ultralter, see 6.8.
4. Let us explain why we introduced positive entangledness in 1.10.
The proof of 6.1 excludes not only full entangledness but many vari-
ants (for the ultrapower). Now the positive -entangledness seems to
be the maximal one not excluded. Rightly so by 6.4.
5. So if we can nd an linear order 1 which is
+
-entangled for some
such that
0
= (or at least
0
< [1[) then we can answer Monks
problem from the introduction: if D is a non principal non separative
ultralter on (see Denition 6.5; they exist by 6.2(1)), then 1
/D is
not
+
-entangled (by 6.1). Thus if B is the interval Boolean algebra
of 1 then
inc(B) ,
0
< [1[, but inc(B
/D) [1[
0
(in fact inc
+
(B
/D) = [1
/D[
+
= ([1[
/D)
+
, inc
+
(B)
+
). In fact
for any innite Boolean algebra B and a non principal ultralter D
on we have inc(B
/D) (inc(B))
/D (as for
n
inaccessible the
linear order
n<
(
n
, <)/D cannot be
+
-like (see [MgSh 433]).
Proposition 6.4 Suppose that < < are regular cardinals such that
( < )(
<
< ). Assume that D is an ultralter on . Then:
if 1 is a positively -entangled linear order of the size
then 1
/D is positively -entangled.
Proof Suppose f
/D 1
/D ,= f
/D).
Let u , (). For < let A
= f
<
A
, B
= A
A
<
. Note that [A
[ < , [A
<
[ < (for < ). If
< , cf() = then [B
A
<h()
. By Fodors lemma we nd
0
< and a stationary set
S such that
S cf() = & B
A
<
0
.
82
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
For S let Y
= (, i, f
(i) A
<
0
. As there are
[A
<
0
[
<
< possibilities for Y
= Y . Let , S
1
, < and < ().
If there is x such that (, i, x) Y then f
(i) = f
(i). Thus
E
(i) = f
then f
(i) ,= f
(i) : S
1
, < (), i E
and u
= (, i) : u, i E
.
Consequently we have < , both in S
1
and such that
( < ())(i E
)(f
(i) < f
(i) u).
Since E
/D < f
/D u).
6.4
Denition 6.5 An ultralter D on is called separative if for every ,
i
: i A
i
: i A = .
Remark 6.6 So 6.2(1) says that D D is not separative.
Proposition 6.7 Suppose that D is a separative ultralter on , n < and
(for < n) are such that
(
0
<
1
< n)(
0
/D ,=
1
/D).
Then there is A D such that the sets
i
: i A (for < n) are pairwise
disjoint.
Proof For < m < n, by 6.5, there is A
,m
D such that
i
: i A
,m
m
i
: i A
,m
= .
Now A =
<m<n
A
,m
is as required.
6.7
Proposition 6.8 Assume = cf(), ( < )([[
/D is (,
0
)-entangled.
83
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proof Let n < , u n and f
/D 1
(i) = f
(i))
(i.e. the heart of the -system). Note that for each < n the set i :
(i, ) A is not in D (as f
0
(i
0
) = f
1
(i
1
)
0
=
1
&
0
=
1
.
It is easy now to apply the (,
+
)-entangledness of 1 and nd < <
such that f
/D < f
/D u.
6.8
Proposition 6.9 1. If D is a selective ultralter on (i.e. for every
f : there is A D such that either f A is constant or f A
is one-to-one) then D is separative.
2. If no uniform ultralter on is generated by less than continuum sets
(i.e. u = 2
0
) then there exists a separative ultralter on .
Proof 1) Suppose that ,
are such that (i < )(
i
,=
i
). We
nd A D such that A,
A are either constant or one-to-one. If
at least one of them is constant then, possibly omitting one element from
A, the sets
i
: i A,
i
: i A are disjoint, so assume that both
sequences are one-to-one. Choose inductively m
i
0, 1, 2 (for i A) such
that
i, j A &
i
=
j
m
i
,= m
j
.
There are m
). Then
i
: i B
i
: i B = .
2) Straightforward.
6.9
Proposition 6.10 If D is a uniform not separative ultralter on , 1 is
a linear ordering of size then the linear order 1
/D is not (, 2)-
entangled.
84
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proof As D is not separative we have ,
witnessing it. This
means that ,=
mod D and the family
i
: i A,
i
: i A : A D
has the nite intersection property. Consequently we may apply 6.1.
6.10
Conclusion 6.11 1. If
/D > 2
2
>
2
then D is not separative.
Remark 6.12 If there is no inner model with measurable then every D is
regular or close enough to this to give the result.
Proof 1) For every f
the family E
f
= A : f
1
[A] D is an
ultralter on . For some g
0
, g
1
we have E
g
0
= E
g
1
but g
0
/D ,= g
1
/D
and hence we are done.
2) The regularity implies
0
/D = 2
,= t
I
t
,
(b) () u t
I
t
.
2. We say that a linear order 1 is strongly positively [positively
] (, )-
entangled if for every () < 1 + , t
I
t
,
(b) () u t
I
t
.
Remark 6.14 For positively
(, )-entangled.
3. In Denition 6.13 the properties are preserved when increasing and/or
decreasing and/or decreasing 1.
4. If = cf(), ( < )( < )(
||
< ) then:
(a) 1 is (, )-entangled if and only if 1 is strongly (, )-entangled,
(b) 1 is positively (, )-entangled if and only if 1 is strongly posi-
tively (, )-entangled.
5. If = cf(), ( < )(2
||
< ) then in Denition 6.13(1),(2) we can
assume
( < )( < < ())(t
,= t
).
6. (, <) and (, >) are not strongly (, 2)-entangled (even if 1 is strongly
(, 2)-entangled then there is no partial function f from 1 to 1 such
that x <
I
f(x), [dom(f)[ = and f preserves <
I
).
7. If ( < )([1[
||
< ) then 1 is strongly (, )-entangled.
8. Assume is a limit cardinal, 1 is a linear order. Then
(a) 1 is strongly (, )-entangled if and only if for every
1
< the
order 1 is strongly (,
1
)-entangled.
(b) Similarly for other notions of 1.1, 1.10, 6.13.
6.15
Proposition 6.16 Assume that 1 is a (, 2)-entangled linear order and =
cf() < . Then for some A 1, [A[ < we have:
x <
I
y & x, y , A [(x, y)
I
[ .
Proof Let E
I
be the following two place relation on 1:
x E
I
y if and only if
x = y or [x <
I
y & [(x, y)
I
[ < ] or [y <
I
x & [(y, x)
I
[ < ].
Obviously:
86
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
(a) E
I
is an equivalence relation,
(b) each equivalence class has cardinality ,
(c) if an equivalence class has cardinality then there is a monotonic se-
quence of length in it.
It is enough to show that the set A = x : [x/E
I
[ > 1 has cardinality
less than . Suppose that [A[ . Then there are at least equivalence
classes (as each class is of the size < ). Consequently we can nd
t
0
, t
1
< t
1
, t
0
I
t
1
/E
I
: < . For () = 2 and u = 0 we get contradiction to
Denition 6.13.
6.16
Remark: Easily, = cf() is redundant.
Proposition 6.17 If 1 is strongly (, )-entangled,
0
< , [1[
then
(a) 2
< and
(b) the cardinal =: [x/E
I
: x/E
I
not a singleton[ satises
<
(where E
I
is from the proof of 6.16).
Proof Take x
0
0
>
I
x
1
0
<
I
x
0
1
<
I
x
1
1
. For f
2 let t
2+
f
= x
f()
(for
< ). If 2
I
: i < ) be with no repetition, x
1
i
x
0
i
/E
I
, x
0
i
<
I
x
1
i
.
Let f
: <
() ,= f
()). Let t
2+
f()
and u = 2 : < (so () = ). If
as required.
6.17
Remark: See more in[SaSh 553].
Proposition 6.18 If 1 is a linear order, [1[ [
0
, ) then BA
inter
(1)
is not 2
-narrow.
87
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proof Choose 1, [[ = such that for every t for some
BA
inter
(1) we have = t ( =
[t, s) : t <
I
s ). Hence for
every
for some BA
inter
(1) we have =
. The conclusion
now follows.
6.18
Proposition 6.19 In Denition 6.13(1), if cf() = (or less) we can wlog
demand
( < )( < < ())(t
,= t
)
and for some linear order <
on ()
<
, , < () t
<
I
t
.
Proof Clearly the new version of the denition implies the old one.
So now assume the old denition and we shall prove the new one. Let
() < 1 + and t
<
I
t
then t
, t
<
I
t
, t
,= t
& (t
< t
<
)].
So for < , <
we can choose s
,,
<
I
s
,,1
<
I
s
,,2
<
I
t
88
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
and there are no repetitions in t
, s
,,
and 1, 2.
So for some < we have
u t
I
t
,
() u t
I
t
,
<
s
,,1
I
s
,,1
,
<
s
,,2
I
s
,,2
.
Now (c) follows immediately.
2) Use 6.16 + 6.17(b).
For (, )-entangledness straightforward.
6.20
Proposition 6.21 Assume
0
< , = dens(1) and: 1 is a strongly
(, )-entangled or BA
inter
(1) is -narrow. Then
< .
Proof First note the following fact:
Claim 6.21.1 Assume that 1
= dens(1
)
0
and =
<
inter
(1) is not -narrow.
Proof of the claim: Choose by induction on i <
, a
i
< b
i
from 1
such
that [a
i
, b
i
]
I
is disjoint to a
j
, b
j
: j < i. Let f
: < list
<
(with
no repetitions) and let () = , t
2+
be: a
f()
if = 0, b
f()
if = 1, and
u = 2 : < we get a contradiction to 1 is strongly (, )-entangled.
The proof for the Boolean version is similar.
6.21.1
By an argument similar to that of 6.21.1 one can show that (= dens(1)) <
. So the interesting case is when
> . As 2
. Let
1
= min :
, so cf(
1
) , and let
1
=
<()
+
1,
, () = cf(
1
) ,
1,
<
1
. For each < () we dene a
two place relation E
on 1:
x E
y if and only if
either x = y or x < y, dens(1 (x, y))
1,
or
y < x, dens(1 (y, x))
1,
.
It is an equivalence relation, each equivalence class has density
+
1,
, and
the number of E
i
: i <
+
) such that i ,= j
(x
i
E
j
).
89
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Without loss of generality, for all the monotonicity is the same, so wlog
i < j x
i
<
I
x
j
. Throughing away a long initial segment from each
x
i
: i <
+
) we may assume that for each , < () either
(i <
+
)(j <
+
)(x
i
<
I
x
j
& x
i
<
I
x
j
)
or
_
i<j
[x
i
, x
j
]
I
,
_
i<j
[x
i
, x
j
]
I
are disjoint.
Now it is easy to satisfy the assumptions of 6.21.1.
6.21
Conclusion 6.22 Assume that = cf() = cf()
0
, 1 is a linear
order of cardinality . Then in Denition 6.13(1) we can demand () of
1.2(3).
Proof By 6.21 (and see the proof of 1.2(3)).
6.22
Proposition 6.23 Assume = cf() > = cf()
0
, 1 is a linear
order, [1[ . Then the following conditions are equivalent:
(a) 1 is strongly (, )-entangled,
(b) BA
inter
(1) is -narrow.
Proof (a) (b) By 6.22 the situation is similar enough to the one
in 1.5 to carry out the proof.
(b) (a) By 6.21 we can apply the parallel of 1.2(3), so the situation
is similar enough to the one in 1.5 to carry out the proof as there.
6.23
Proposition 6.24 Assume =
+
>
0
, > .
1. If 1 is not strongly [or strongly positively] [or strongly positively
]
(
i
, )-entangled for i < then 1 is not strongly [or strongly posi-
tively] [or strongly positively
] (, )-entangled for =
i<
i
.
2. If is singular and 1 is strongly (, )-entangled then for some
< ,
1 is strongly (
< ).
3. The parallel of (2) holds for strongly (, )-entangled, strongly pos-
itively (, )-entangled and strongly positively
(, )-entangled.
90
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
Proof 1) First we deal with strongly (, )-entangled. We know
[1[ . Suppose that t
i,
: <
i
, <
i
()), u
i
form a counterexample
for
i
. As we can extend the sequences and u
i
wlog [u
i
[ = [
i
() u
i
[ = .
So by renaming
i
() = , u
i
= 2 : < . Let f
i<
i
for < be
pairwise distinct. Let us choose () = , t
i+
= t
i,
f
(i)
for i < , < ,
< and u = 2 : < . Now check.
For the cases strongly positively
()
(
i
, )-entangled rst wlog
i
() =
; then as u
i
has two values for some u
we have
i
: i < , u
i
= u
= .
Thus wlog u
i
= u and let u =
i<
i u
i
, <
i
< , = cf(), i < j
i
<
j
.
So for each i < we can nd a sequence t
i,
: <
i
(), <
+
i
), u
i
exemplifying the failure of 1 is (
+
i
, )-entangled. Thus for each i and for
every <
+
i
, there are no repetitions in t
i,
: <
i
() and [1[ . Now
let () = +, u = 2 : < +. For < ,
+
i
+
j
: 0 < j < i
we put t
= t
i,
, t
+
= t
0,
i
. Wlog for every < there are no repetitions
in t
i
, u
i
be as in
the proof of (2) (for the appropriate notion). Again wlog
i
() = , u
i
= u
,
but the choice of
i
does not transfer. But by 6.24(1) we have:
i
<
and hence wlog (
j<i
+
j
)
i
=
i
, so for some v
i
i
()
( v
i
[t
i,
= t
i,
= ]) and [
i
()v
i
t
i,
= t
i,
]
and t
i,
= t
i,
t
i,
= t
i,
. We can omit
i
() v
i
etc, so t
i,
: <
i
(), <
+
i
) is with no repetition and proceed as there.
6.24
Remark 6.25 In 6.24(2) we cannot replace strongly entangled by en-
tangled; see [SaSh 553].
Conclusion 6.26 If [1[ > =
+
>
0
then:
1 is (, )-entangled if and only if 1 is strongly (, )-entangled.
Conclusion 6.27 If =
+
0
, 1 is a strongly (, )-entangled
linear order then for some regular
we have:
91
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
, ( <
)([[
<
) and 1 is (strongly) (
, )-
entangled.
Consequently in 6.23 regularity is not needed.
References
[ARSh 153] Uri Abraham, Matatyahu Rubin, and Saharon Shelah. On the
consistency of some partition theorems for continuous colorings,
and the structure of
1
-dense real order types. Annals of Pure
and Applied Logic, 29:123206, 1985.
[AbSh 106] Uri Avraham (Abraham) and Saharon Shelah. Martins ax-
iom does not imply that every two
1
-dense sets of reals are
isomorphic. Israel Journal of Mathematics, 38:161176, 1981.
[Bo] Robert Bonnet. Sur les Alg`ebres de Boole rigides. PhD thesis,
Universite Lyon 1, 1978.
[BoSh 210] Robert Bonnet and Saharon Shelah. Narrow Boolean algebras.
Annals of Pure and Applied Logic, 28:112, 1985.
[EK] Ryszard Engelking and Monika Karlowicz. Some theorems of
set theory and their topological consequences. Fundamenta
Math., 57:275285, 1965.
[HLSh 162] Bradd Hart, Claude Laamme, and Saharon Shelah. Models
with second order properties, V: A General principle. Annals of
Pure and Applied Logic, 64:169194, 1993. math.LO/9311211.
[MgSh 433] Menachem Magidor and Saharon Shelah. Length of Boolean
algebras and ultraproducts. Mathematica Japonica, 48(2):301
307, 1998. math.LO/9805145.
[M1] J. Donald Monk. Cardinal Invariants of Boolean Algebras. Lec-
tures in Mathematics. ETH Zurich, Birkhauser Verlag, Basel
Boston Berlin, 1990.
[M2] J. Donald Monk. Cardinal Invariants of Boolean Algebras, vol-
ume 142 of Progress in Mathematics. Birkhauser Verlag, Basel
BostonBerlin, 1996.
92
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
[RoSh 534] Andrzej Roslanowski and Saharon Shelah. Cardinal invariants
of ultrapoducts of Boolean algebras. Fundamenta Mathemati-
cae, 155:101151, 1998. math.LO/9703218.
[RoSh 599] Andrzej Roslanowski and Saharon Shelah. More on cardinal
invariants of Boolean algebras. Annals of Pure and Applied
Logic, 103:137, 2000. math.LO/9808056.
[SaSh 553] Ofer Shar and Saharon Shelah. More on entangled or-
ders. Journal of Symbolic Logic, 65:18231832, 2000.
math.LO/9711220.
[Sh 50] Saharon Shelah. Decomposing uncountable squares to count-
ably many chains. Journal of Combinatorial Theory. Ser. A,
21:110114, 1976.
[Sh 345] Saharon Shelah. Products of regular cardinals and cardinal
invariants of products of Boolean algebras. Israel Journal of
Mathematics, 70:129187, 1990.
[Sh 410] Saharon Shelah. More on Cardinal Arithmetic. Archive for
Mathematical Logic, 32:399428, 1993. math.LO/0406550.
[Sh 371] Saharon Shelah. Advanced: conalities of small reduced prod-
ucts. In Cardinal Arithmetic, volume 29 of Oxford Logic Guides,
chapter VIII. Oxford University Press, 1994.
[Sh 355] Saharon Shelah.
+1
has a Jonsson Algebra. In Cardinal
Arithmetic, volume 29 of Oxford Logic Guides, chapter II. Ox-
ford University Press, 1994.
[Sh 345a] Saharon Shelah. Basic: Conalities of small reduced prod-
ucts. In Cardinal Arithmetic, volume 29 of Oxford Logic Guides,
chapter I. Oxford University Press, 1994.
[Sh:g] Saharon Shelah. Cardinal Arithmetic, volume 29 of Oxford
Logic Guides. Oxford University Press, 1994.
[Sh 400] Saharon Shelah. Cardinal Arithmetic. In Cardinal Arithmetic,
volume 29 of Oxford Logic Guides, chapter IX. Oxford Univer-
sity Press, 1994. Note: See also [Sh400a] below.
[Sh 345b] Saharon Shelah. Entangled Orders and Narrow Boolean Alge-
bras. In Cardinal Arithmetic, volume 29 of Oxford Logic Guides.
Oxford University Press, 1994. Appendix 2.
93
(
4
6
2
)
r
e
v
i
s
i
o
n
:
2
0
0
1
-
1
1
-
1
2
m
o
d
i
f
i
e
d
:
2
0
0
1
-
1
1
-
1
2
[Sh 405] Saharon Shelah. Vive la dierence II. The Ax-Kochen isomor-
phism theorem. Israel Journal of Mathematics, 85:351390,
1994.
[Sh 430] Saharon Shelah. Further cardinal arithmetic. Israel Journal of
Mathematics, 95:61114, 1996. math.LO/9610226.
[Sh 620] Saharon Shelah. Special Subsets of
cf()
, Boolean Algebras
and Maharam measure Algebras. Topology and its Applica-
tions, 99:135235, 1999. 8th Prague Topological Symposium
on General Topology and its Relations to Modern Analysis and
Algebra, Part II (1996). math.LO/9804156.
[Sh 460] Saharon Shelah. The Generalized Continuum Hypothesis re-
visited. Israel Journal of Mathematics, 116:285321, 2000.
math.LO/9809200.
[To] Stevo Todorcevic. Remarks on Chain Conditions in Products.
Compositio Math., 5:295302, 1985.
94