7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH
SH794
Saharon Shelah
Institute of Mathematics
The Hebrew University
Jerusalem, Israel
Rutgers University
Mathematics Department
New Brunswick, NJ USA
Abstract. We prove that, e.g., if > cf() =
0
and > 2
0
and every stationary
family of countable subsets of
+
reect in some subset of
+
of cardinality
1
then
the SCH for
+
holds (moreover, for
+
, any scale for
+
has a bad stationary set
of conality
1
). This answers a question of Foreman and Todorcevic who gets such
conclusion from the simultaneous reection of four stationary sets.
2000 Mathematics Subject Classication. 03E04, 03E05.
Key words and phrases. reection, stationary sets, Singular Cardinal Hypotheses, pcf, set
theory.
The author would like to thank the Israel Science Foundation for partial support of this research
(Grant No. 242/03)
I would like to thank Alice Leonhardt for the beautiful typing.
First Typed  01/12/18
Done  Dec. 2001
Latest Revisions  07/Oct/1
Typeset by A
M
ST
E
X
1
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
2 SAHARON SHELAH
0 Introduction
In 1 we prove that the strong hypothesis (pp() =
+
for every singular )
hence the SCH (singular cardinal hypothesis, that is
+
+ 2
) holds when:
for every
1
every stationary S []
0
reect in some A []
1
.
This answers a question of Foreman and Todorcevic [FoTo] where they proved
that the SCH holds for every
1
when: every four stationary S
[]
0
, =
1, 2, 3, 4 reect simultaneously in some A []
1
. They were probably motivated
by Velickovic [Ve92a] which used another reection principle: for every stationary
A []
0
there is A []
1
such that A [A]
0
contains a closed unbounded
subset, rather than just a stationary set.
The proof here is selfcontained modulo two basic quotations from [Sh:g], [Sh:f];
we continued [Sh:e], [Sh 755] in some respects. We prove more in 1. In particular
if > cf() =
0
and pp() >
+
then some A [
+
]
0
reect in no uncountable
A [
+
]
, we
can use [B]
: < A where
0 < < and
1
<
2
< x
1
x
2
, we have
<
x
A
(b) A is unbounded in [B]
<
, if for any y [B]
<
we can nd x A, such
that x y
(c) A is a club in [B]
<
, if A [B]
<
and (a)+(b) hold for A
(d) A is stationary in [B]
<
, or is a stationary subset of [B]
<
when A [B]
<
and A C ,= for every club C of [B]
<
(e) similarly for [B]
or [B]
or consider S [B]
<
as a subset of [B]
.
0.2 Remark. Note: if B = then A [B]
<
is stationary i A is a stationary
subset of .
0.3 Denition. Let A [B
1
]
<
and B
2
[B
1
]
]
<
.
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 3
0.4 Denition. Let be a regular uncountable cardinal, and assume A is a
stationary subset of [B]
<
. We dene
A
(i.e., the diamond principle for A) as
the following assertion:
there exists a sequence u
a
: a A), such that u
a
a for any a A, and for every
B
B the set a A : B
a = u
a
is stationary in [B
]
<
.
0.5 Notation. 1) For regular > let S
= < : cf() = .
2) H () is the set of x with transitive closure of cardinality < .
3) <
) <
(iii) C
= C
(iv) E S = sup(C
).
0.7 Claim. (By [Sh 420] or see [Sh:E12]).
1) If , are regular and >
+
then there is a stationary S S
such that
S
I[].
2) In 0.6 we can add E S otp(C
) = cf().
0.8 Denition/Observation. Let A []
0
and the strong hypothesis
1.1 The Main Claim. Assume
(A) =
+
and > cf() =
0
and
2
(e.g.,
=
2
which implies
that below always =
1
)
(B)
=
n
: n < ) is an increasing sequence of regular cardinals >
1
with
limit and = tcf(
n
, <
J
bd
)
(C)
f = f
: < ) is <
J
bd
increasing conal in (
n<
n
, <
J
bd
)
(D) the sequence
f is
, A []
0
is stationary
(recall 0.2) and
i
: i < ) is an increasing continuous sequence of ordinals
< then for some stationary subfamily A
1
of A (A
1
is stationary in []
0
of course) letting R
1
= prj
(A
1
), see 0.8 we have f
i
: i R
1
) is free. See
0.9 and by 0.10 we can assume that i R
1
n
i
= n() so f
i
(n) : i R
1
)
is strictly increasing for every n [n(), ).
Then some stationary A []
0
does not reect in any A []
1
or even in any
uncountable A []
<
i
(n) < f
j
(n) (so f
i
(n) : i < cf() and n n
i
: n < ) is a <
J
bd
eub of
f ).
If we use another ideal J say on < , the n
i
is replaced by s
i
J.
2) Recall that by using the silly square ([Sh:g, II,1.5A,pg.51]), if cf() < , J
an ideal on (e.g. =
0
, J = J
bd
) and pp
J
() > = cf() > then we can
nd a sequence
i
: i < ) of regulars < such that = lim
J
i
: i < ) and
tcf(
i<
i
, <
J
) = and some
f = f
: u) such that s
) f
() < f
: < ) exemplies =
+
=
tcf(
i<
i
, <
J
bd
Reg
+
the set S
gd(
f)
is stationary. This is preserved by e.g. Levy(
1
, < ).
4) For part of the proof (mainly subclaim 1.5) we can weaken clause (D) of the
assumption, e.g. in the end demand f
i
(n) ,= f
j
(n) only. The weakest version
of clause (D) which suces there is: for any club C of the set Rang(f
) : C
has cardinality .
Before proving 1.1 we draw some conclusions.
1.3 Conclusion. 1) Assume > 2
0
then
0
=
+
provided that
(A)
> cf() =
0
(B)
every stationary A [
+
]
0
reects in some A [
+
]
1
.
2) Assume
2
. We can replace (B)
by
(B)
,
every stationary A [
+
]
0
reects in some uncountable A [
+
]
<
.
Proof. 1) Easily if
1
then (B)
n<
0
n
/J
bd
: <
++
)
witness this. Hence by [Sh:g, II,1.5A,p.51] there is
f as required in 1.1 even a
+
free one and also the other assumptions there hold so we can conclude that
there exists A [
+
]
0
which does not reect in any A [
+
]
1
, so we get a
contradiction to (B)
.
2) The same proof.
1.3
1.4 Conclusion. 1) If for every >
1
, every stationary A []
0
reects in some
A []
1
, then
(a) the strong hypothesis (see [Sh 410], [Sh 420], [Sh:E12]) holds, i.e. for every
singular , pp() =
+
and moreover cf([]
cf()
, ) =
+
which follows
(b) the SCH holds.
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
6 SAHARON SHELAH
2) Let
0
. We can restrict ourselves to >
+
, A []
+
(getting the strong
hypothesis and SCH above ).
Proof. 1) As in 1.3, by 1.1 we have > cf() =
0
pp() =
+
, this implies
clause (a) (i.e. by [Sh:g, VIII,1], > cf() pp() =
+
). Hence inductively by
[Sh:g, IX,1.8,pg.369], [Sh 430, 1.1] we have < cf([]
, ) = if cf() >
and is
+
if > cf(). This is a consequence of the strong hypothesis.) The
SCH follows.
2) The same proof.
1.4
Proof of 1.1. Let M
+
,
f) and are functions from to or just
the functions f
(n), + 1.
1.5 Subclaim. There are
S, S
,
D such that:
()
1
S = S
: <
1
) is a sequence of pairwise disjoint stationary subsets of
S
0
()
2
(i) S
1
= < : cf() =
1
is stationary and belongs to
I[]
(ii) if S
: <
1
)
of ordinals with limit such that for some sequence
=
: R) of
ordinals <
1
, the set R
1
is stationary, R
and
is
with no repetitions
()
3
(i)
D = (D
1,
, D
2,
) : <
1
)
(ii) D
,
is a lter on containing the lter of cobounded subsets of
(iii) if R
1
1
is unbounded and A D
1,
: R
1
then for some R
1
we have A ,= mod D
2,
(iv) for each <
1
for some A we have A D
1,
& A D
2,
.
1.6 Remark. 1) For 1.5 we can assume (A), (B), (C) of 1.1 and weaken clause (D):
because (inside the proof below) necessarily for any stationary S
1
, which
belongs to
I[], we can restrict the demand in (D) of 1.1 for any
i
: i < ) with
limit in S
= in ()
2
(ii). See the proof.
3) If we like to demand that each D
,
is an ultralter (or just have A D
2,
in
the end of ()
3
(iii) of 1.5), use [Sh:E3].
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 7
Proof of the subclaim 1.5. How do we choose them?
Let A
i
: i
1
) be a sequence of innite pairwise almost disjoint subsets of .
Let D
1,i
= A : A
i
A is nite, D
2,i
= A : A
j
A is nite for all but
nitely many j <
1
, so D
2,i
does not depend on i. Clearly (D
1,i
, D
2,i
) : i <
1
)
satises ()
3
.
Recall, 0.7, that by 0.7 and the fact that >
>
2
, there is a stationary
S
1
from
I[], and so every stationary S
1
).
Let N (H ((2
)
+
), , <
, ,
f
belongs to N. Let C
= C : C N is a club of , so clearly C
is a club of .
For each h
(
1
) we can try
S
h
= S
h
: <
1
) where S
h
= < : cf() =
0
and h() = , so it is enough to show that for some h N, the sequence
S
h
is as
required. As N < , for this it is enough to show that for every S
1
C
(or
just for every S
) the demand
holds for
S
h
for some h (
(
1
)) N. That is, S
h
satises ()
1
and ()
2
(ii) of
subclaim 1.5. Given any S
1
C
let
: <
1
) be an increasing continuous
sequence of ordinals with limit , without loss of generality <
1
cf(
) =
0
,
and by assumption (D) of 1.1 for some
1
stationary R
1
and n = n() < , the
sequence f
. So
and cf(
) =
1
but + 1 N hence
N.
Note that
()
1
for every
<
the set S
0
: f
(n()) [
) is a stationary
subset of .
[Why? So assume that
<
= S
0
: f
(n()) [
)
is not a stationary subset of . As
+1 N and
f N clearly S
N hence there
is a club C
of disjoint to S
) too is a club of
which belongs to N hence C
acc(C
) hence acc(C
). So = sup(C
),
so C
()
(n()), hence
R() f
(n()) [
hence disjoint to C
, contradiction.]
()
2
for every
<
there is
) such that S
0
: f
(n())
[
) is a stationary subset of .
[Why? Follows from ()
1
as
1
< .]
1
Note that if we require just that f
of R the sequence f
(n) : R
is increasing.
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
8 SAHARON SHELAH
As
: <
1
) N of
ordinals with limit
. So by ()
2
()
3
for every
1
<
1
for some
2
(
1
,
1
) the set S
0
: f
(n())
[
1
,
2
) is stationary.
Hence for some unbounded subset u of
1
we have
()
4
for every u the set S
0
: f
(n()) [
,
+1
) is a stationary
subset of .
If 2
1
then u =
1
recalling we demand
: <
1
) N.
We dene h :
1
by h() = i for some u we have = otp(u
f
(n()) sup(u).
Clearly h N is as required. So
S =
S
h
as required exists. But maybe 2
1
> ,
then after ()
3
we continue as follows. Let
C = C
: S
1
) be such that C
1
we have C
acc(C
) has conality
3
. Again
belongs to N hence
some increasing continuous sequence
: <
3
) N has limit
. Now for
each S
1
we could choose above =
< we have
(
<
)(
<
)[
<
(
stat
S
0
)(
(n
) <
]. So for some
n
:= <
3
: cf() =
1
and n
= n
is a stationary subset of S
1
.
It follows that ( <
3
)( <
3
)[ < (
stat
S
0
)(
(n
) <
+1
)].
Let
1
we have
C
()
C. Let u :=
: C
()
so clearly
: u) belongs to N.
1.5
1.7 Remark. Why cant we, in the proof of 1.5, after ()
3
, put the instead assuming
2
1
use as N (H (2
)
+
), , <
1
still u N, but otherwise
the without loss of generality u N does not seem to be justied.
Continuation of the proof of 1.1. Let S := S
: <
1
. For <
1
, S
let
A
= a :a []
0
is M
closed, sup(a) = ,
otp(a) and
(
D
1,
n)(a
n
f
(n))
and (
D
2,
n)(a
n
f
(n))
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 9
A
= A
: S
A = A
: <
1
.
So
A []
0
.
As the case
=
2
was the original question and its proof is simpler we rst prove
it.
1.8 Subclaim. A does not reect in any A []
1
.
Proof. So assume A []
1
, let a
i
: i <
1
) be an increasing continuous sequence
of countable subsets of A with union A, and let R = i <
1
: a
i
A, and
assume toward contradiction that R is a stationary subset of
1
. As every a A
is M
closed, necessarily A is M
closed.
For each i R as a
i
A by the denition of A we can nd
i
<
1
and
i
S
i
such that a
i
A
i
hence by the denition of A
i
we have otp(a
i
)
i
. But
as A = a
i
: i <
1
with a
i
countable increasing with i and [A[ =
1
, clearly
for some club E of
1
the sequence otp(a
i
) : i E) is strictly increasing, hence
i E otp(i E) otp(a
i
) so without loss of generality i E i otp(a
i
)
and without loss of generality i < j E
i
< j otp(a
j
).
Now j E R j otp(a
j
)
j
so
i
: i E R) is strictly increasing
but S
: <
1
) are pairwise disjoint and
i
S
i
so
i
: i E R) is without
repetitions; but
i
= sup(a
i
) and for i < j from R E we have a
i
a
j
which
implies that
i
= sup(a
i
) sup(a
j
) =
j
so necessarily
i
: i R E) is strictly
increasing.
As sup(a
i
) =
i
for i R E, clearly sup(A) =
i
: i E R and let
i
= Min(A
i
) for i <
1
, it is well dened as
j
: j R E) is strictly
increasing. Thinning E without loss of generality
1
i < j E R
i
<
j
&
i
a
j
.
Note that, by the choice of M
2
i E R i < j E R
i
a
j
n
(f
i
(n) a
j
)
n
(f
i
(n) + 1
a
j
).
As
i
: i E R) is (strictly) increasing continuous and R E is a stationary
subset of
1
clearly by clause (D) of the assumption of 1.1 we can nd a stationary
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
10 SAHARON SHELAH
R
1
E R and n() such that i R
1
j R
1
i < j n() n < f
i
(n) <
f
j
(n).
Now if i R
1
, let j(i) =: Min(R
1
(i + 1)), so f
i
J
bd
i
<
J
bd
j(i)
so for
some m
i
< we have n [m
i
, ) f
i
(n) f
i
(n) < f
j(i)
(n). Clearly for some
stationary R
2
R
1
we have i, j R
2
m
i
= m
j
= m(), so possibly increasing
n() without loss of generality n() m(); so we have (where Ch
a
n<
n
is
dened by Ch
a
(n) = sup(a
n
) for any a []
<
0
):
3
for i < j from R
2
we have j(i) j and
() f
i
[n(), ) f
i
[n(), )
() f
i
[n(), ) < f
j(i)
[n(), ) f
j
[n(), )
() f
i
[n(), ) < Ch
a
j
[n(), ), by
2
.
Now by the denition of A
i
as a
i
A
i
A
i
we have
4
if i R
2
then
() Ch
a
i
D
1,
i
f
i
() f
i
<
D
2,
i
Ch
a
i
.
Let f
n<
n
be f
(n) = f
i
(n) : i R
2
if n n() and zero otherwise. As
f
i
(n) a
j(i)
for i R
2
by
3
() clearly n n() f
(n) supCh
a
i
(n) : i
R
2
= sup(A
n
) = Ch
A
(n) and by
3
() we have n n() cf(f
(n)) =
1
.
Let B
1
=: n < : n n() and f
(n) = sup(A
n
) and B
2
=: [n(), )B
1
. As
A + 1 A we have n B
1
A
n
f
(n) = sup(A
n
). Also as by
the previous sentence f
[n(), ) Ch
A
[n(), ) clearly n B
2
A
n
f
(n). As a
i
: i R
2
) is increasing with union A, clearly there is i() R
2
such
that: n B
2
a
i()
n
f
(n), so as i R
2
& a
i
+ 1 a
i
we have i() i R
2
Ch
a
i
B
2
> f
i
B
2
hence by clause
4
() we have
i R
2
i() B
2
= mod D
1,
i
B
1
D
1,
i
. Also by
3
and the choice
of f
and B
1
, for each n B
1
for some club E
n
of
1
we have i E
n
R
2
sup(a
i
n
) = supf
j
(n) : j R
2
i = supf
j
(n) : j R
2
i f
i
(n), hence
R
3
= R
2
E
n
i() : n < is a stationary subset of
1
. So n B
1
& i
R
3
a
i
n
f
i
(n) hence i R
3
Ch
a
i
B
1
f
i
B
1
hence by
4
() we
have i R
3
B
1
= mod D
2,
i
hence i R
3
B
2
D
2,
i
.
By the choice of (D
1,i
, D
2,i
) : i <
1
) in 1.5 as B
1
B
2
is a conite subset of
, B
1
B
2
= (by the choice of B
1
, B
2
, clearly) and R
3
1
is stationary we get
a contradiction, see ()
3
(iii) of 1.5.
1.8
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 11
1.9 Subclaim. A is a stationary subset of []
0
.
Remark. See [RuSh 117], [Sh:f, XI,3.5,pg.546], [Sh:f, XV,2.6].
We give a proof relying only on [Sh:f, XI,3.5,pg.546]. In fact, also if we are
interested in Ch
N
= sup( N) :
0
< N Reg), N (H (), ) we have full
control, e.g., if
S = S
:
1
Reg ), S
0
stationary we can demand
1
= cf() N Ch
N
() S
be an expansion of M
1
is from 1.5 so it belongs to
I[] and let a = a
: < )
witness it (see 0.6, 0.7) so otp(a
)
1
and a
= a
, and omitting
a nonstationary subset of S
we have S
otp(a
) =
1
& = sup(a
).
Let
T
=
n
if g() = 2n + 1 and let I
be the non
stationary ideal on
for T
, so (T
: T
).
For T
, let M
be the M
is
countable and T
) =
: n T
= M
n
: n < , so it is enough to prove that M
A
for some lim(T
), more exactly [M
[ A recall M
as M
: T
such that
(T
<
1
we have lim(T) otp(M
) =
(recalling (T
, I) (T, I) means T T
, ( T
)( < g())(
T
), <> T
: ) T
,= mod I
, i.e. is
stationary)).
Why? As lim(T
) = B
: <
1
, see
4
below, and by
1
below each B
, I)
(T, I) lim(T) B
be the universe of M
, a
countable set of ordinals.
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
12 SAHARON SHELAH
So we use
1
for any <
1
the set B
= lim(T) : otp(u
) = is a Borel set.
[Why? Without loss of generality u
,= and let
,n
: n < ) enumerate the
members of u
and for n
1
, n
2
< and m
1
, m
2
< let B
n
1
,n
2
,m
1
,m
2
:=
lim(T
) :
n
1
,m
1
<
n
2
,m
2
.
Clearly
2
B
n
1
,n
2
,m
1
,m
2
is an open subset of lim(T
3
there is a L
1
,
sentence
4
lim(T
) = B
: <
1
.
[Why? As otp(M
) < M

+
=
1
. Together should be clear.]
Note that for every T
of length 2n + 2 we have T
is
+
n
complete. As we can shrink T further by [Sh:f, XI,3.5,pg.346] without loss of
generality
for every n < and T
2n+2
for some =
<
n
we have: if
lim(T) then
= sup(
n
M
).
[Why? As above applied to each T
=
>
: T.]
Let = (2
)
+
and N
B = (H(), , <
 = , N
, N
: ) N
+1
and (T, I,
M, a,
f,
, ) N
, clearly
possible and E = < : N
E,
so a
()
is well dened. Let
N
= N
: < ). Let C
()
be the closure of a
()
as
a subset of () in the order topology and let
: <
1
) list C
()
in increasing
order, so is increasing continuous.
We dene N
by induction on <
1
by:
()
0
N
: < N
: < ),
N
: < (T, I,
M, a,
f,
, ).
Let
()
1
g
n<
n
be dened by g
(n) = sup(N
n
).
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 13
Clearly
()
2
(a) N
: ) N
()
and even N
for every [ + 1,
1
)
(b) C
()
(
+ 1) and a
belongs to N
for [ + 1,
1
).
[Why? For clause (a), N
. For
clause (b) because a N
()
and a
()
a
= a
()
and C
()
(
+1) =
the closure of a
+1
(
+ 1).]
Let e = <
1
: is a limit ordinal and N
1
= . So
()
3
(a) e is a club of
1
,
(b) if e then sup(N
) =
= N
, N
1
N
hence
()
4
if + 2 < e then g
, g
+1
N
+2
N
.
Now
f is increasing and conal in
n<
n
hence
()
5
if < e then g
<
J
bd
and f
<
J
bd
.
Also easily
()
6
if < e then g
< g
.
For n < , <
1
let N
,n+1
be the Skolem hull inside B of N
n
and let
N
,0
= N
. Easily
()
7
if n m < and <
1
then g
(m) = sup(N
,n
m
).
Recall that
for some [,
1
) (possible by
subclaim 1.5 particularly clause ()
2
(ii)) and choose
k
e for k < such that
k
<
k+1
< =
< n
k
< for < k
(b) g
k+1
[n
k
, ) < f
[n
k
, ).
[Why is this choice possible? By ()
5
.]
Stipulate n
1
= 0.
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
14 SAHARON SHELAH
Let B
1
D
1,
be such that B
2
= B
1
D
2,
, exists by clause ()
3
(iv) of
subclaim 1.5.
Now we choose
n
by induction on n < such that
(a)
n
T and g(
n
) = n
(b) m < n
m
n
(c) if n [n
k1
, n
k
) then
2n
,
2n+1
N
k
,n
(d) if n [n
k1
, n
k
) then
2n+1
(2n) = Min < :
2n
) T and
k1
if k > 0
(e) if n [n
k1
, n
k
) and n B
1
then
2n+2
(2n + 1) = Min <
n
:
2n+1
) T
(f) if n [n
k1
, n
k
) and n B
2
then
2n+2
(2n + 1) = Min <
n
:
2n+1
) T and > f
(n).
No problem to carry the induction.
[Clearly if
n
is well dened then
n+1
(n) is well dened (by clause (c) or (d) or
(e) according to the case; hence
n+1
T
n+1
is well dened by why clause (c)
holds, i.e. assume n [n
k1
, n
k
), why
2n
,
2n+1
N
k,n
?
Case 1: If n = 0, then
2n
=<> N
k
,n
trivially.
Case 2:
2n
is O.K. hence N
k,n
and show
2n+1
N
k,n
.
[Why? Because N
k
,n
B, if k = 0 as
2n+2
(2) is dened from
2n
and T both of
which belongs to N
k
,n
. If k > 0 we have to check that also
k1
N
k
,n
which
holds by ()
0
.
Case 3:
2n+1
is O.K. so N
k
,n
and we have to show
2n+2
N
k
,n+1
.
As
2n+2
(n) <
n
N
k
,n+1
this should be clear.]
Let =
n
: n < . Clearly lim(T) hence u =: [M
[ []
0
and
M
1
sup(u)
[Why? As
n
belongs to the Skolem hull of N
hence M
n
N
and N
as
E.]
2
sup(u)
n
, for every n <
[by clause (d) of ]
3
sup(u) =
[Why? By
1
+
2
]
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 15
and >
= otp(u)
[Why? By the choice of ]
5
if n n
0
, n > 0 and n B
1
then u
n
f
(n)
[Why? By the choice of
2n+2
(2n+1), i.e., let k be such that n [n
k1
, n
k
),
so
2n+1
N
k
,n
by clause (c) and by clause (e) of we have
2n+2
(2n+1)
n
N
k
,n
hence by above, as lim(T) we have
(2n+2)
=
2n+2
=
sup(u
n
) and as
M N
k
,n
we have
2n+2
N
k
,n
so sup(u
n
) =
2n+2
< sup(N
k
,n
n
) but the latter is equal to sup(N
k
n
) by ()
7
which is equal to g
k
(n) which is < f
(n) by ()
8
, as required.]
6
if n n
1
and n B
2
then u
n
f
(n)
[Why? By the choice of
2n+2
(2n + 1).]
So we are done.
1.9
This (i.e., 1.8 + 1.9) is enough for proving 1.1 in the case
=
2
. In general we
should replace 1.8 by the following claim.
1.10 Claim. The family A does not reect in any uncountable A []
<
.
Proof. Assume A is a counterexample.
Trivially
0
A is M
closed.
For a A let ((a), (a)) be such that a A
(a)
(a)
hence (a) = sup(a), otp(a)
(a). Let A
= A [A]
0
and let = (a) : a A
. Of course, ,= .
Assume that
n
for n < so let
n
= (a
n
) where a
n
A so necessarily
n
S
(a
n
)
. As A is uncountable we can nd a countable b such that a
n
b A and
(a
n
) < otp(b) for every n < and as A
[A]
0
is stationary we can nd c such
that b c A
n
sup(a
n
) sup(c) = (c) for each n < . So if (a
n
) =
n
= (c), n <
necessarily (a
n
) = (c) contradiction so
n
,= (c); hence (c) > (a
n
) and, of
course, (c) so n <
n
< (c) . As
n
for n < were any members
of , clearly has no last element, and let
) =
0
is
impossible, so clearly cf(
) >
0
and let = cf(
) so [A[ <
and is a
regular uncountable cardinal.
As a A
sup(a) = and A
[A]
0
is stationary clearly A
= sup(A) =
sup(). Let
i
: i < ) be increasing continuous with limit
and if
i
S
then
we let
i
= .
For i < let
i
= Min(A
i
), so
i
i
<
,
i
A and i < j <
i
j
.
But i <
i
<
:=
i
: i < and let H : []
0
be H(b) = supi :
i
b and let
J := R : the family b A
: H(b) R = b A
: sup(i < :
i
b)
R is not a stationary subset of [A
0
.
Clearly
1
J is an
1
complete ideal on extending the nonstationary ideal and / J
by the denition of the ideal
2
if B J
+
(i.e., B P()J) then a A
: H(a) B is a stationary
subset of []
0
.
By clause (D) of the assumption of 1.1, for some stationary R
1
J
+
and n
i
<
for i R
1
we have
3
if i < j are from R
1
and n n
i
, n
j
(but n < ) then f
i
(n) < f
j
(n).
Recall that
4
i < j R
1
i
<
j
.
Now if i R
1
, let j(i) = Min(R
1
(i + 1)), so f
i
J
bd
i
<
J
bd
j(i)
hence for
some m
i
< we have n [m
i
, ) f
i
(n) f
i
(n) < f
j(i)
(n). Clearly for some
n() satisfying
n()
> and R
2
R
1
from J
+
we have i R
2
n
i
, m
i
n(),
so
5
for i < j in R
2
we have
() f
i
[n(), ) f
i
[n(), )
() f
i
[n(), ) < f
j
[n(), ).
Let f
n<
n
be dened by f
(n) = f
i
(n) : i R
2
if n n() and zero
otherwise. Clearly f
(n) sup(A
n
) for n < .
Let A
= a A
: (i < )(i a
i
a
i
i
) supi R
2
:
i
a =
supi :
i
a = sup(a ) R
2
and sup(A
n
) > f
(n) a
n
f
(n). As
R
2
J
+
clearly A
0
.
Let R
3
= i R
2
: i = sup(i R
2
) so R
3
R
2
, R
2
R
3
is a nonstationary subset
of (hence belongs to J) and a A
sup(a)
i
: i R
3
.
Let
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
REFLECTION IMPLIES THE SCH SH794 17
A
a [A]
0
:(a)
min(R
2
)
a and a is M
closed
(b) if i R
2
& j = Min(R
2
(i + 1)) then [a
i
a
j
] and
n [n(), ) & a
n
f
i
(n) a
n
f
j
(n) ,=
(c) if i < & n [n(), ) then ()(
i
a)
(j)(i < j < &
j
a) ()(f
i
(n) a f
(n)) and
(d) if A
n
f
(n) then a
n
f
(n)
but ( a)( + 1 a)
hence sup (a
n
) > f
(n)
.
Clearly A
is a club of [A]
0
(recall that A is M
closed). But if a A
, then
for some limit ordinal i R
3
we have a sup(a) =
i
and n [n(), )
sup(a f
(n)) = sup(a f
j
(n) : j R
2
).
Let
B
1
= n : n() n < and A
n
f
(n) = sup(A
n
).
B
2
= n : n() n < and f
0
.
By the denition of A
, for every a A
, we have
6
n B
2
Ch
a
(n) f
(n) > f
(a)
(n)
7
n B
1
Ch
a
(n) = f
(n) : R
2
(a) f
(a)
(n).
But this contradicts the observation below.
1.11 Observation. If B , then for some <
1
we have:
if a A is M
,
then otp(a) < .
Proof. Read the denition of A (and A
, A
1.11
,
1.10
,
1.1
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
18 SAHARON SHELAH
Remark. Clearly 1.11 shows that we have much freeness in the choice of A
s.
We can get somewhat more, as in [Sh:e]
1.12 Claim. In Claim 1.1 we can add to the conclusion
() A satises the diamond, i.e.
A
.
Proof. In 1.5 we can add
()
5
2n + 1 : n < = mod D
,
for < 2, <
1
.
This is easy: replace D
,
by D
,
= A : n : 2n A D
,
. We can x
a countable vocabulary and for <
1
choose a function F
(A) = F
(B) if A = B mod
nite.
Case 1: > 2
0
.
Lastly, for a A let
a
,
a
be such that a A
a
and let A
a
= n : sup(a
2n+1
) < f
a
(2n), and let N
a
be the model with universe a such that the one
toone order preserving function from onto a is an isomorphism from F
(N) onto
N. Note that in the proof of A []
0
is stationary, i.e. of 1.9, given a 
model M with universe without loss of generality
0
> 2
0
and so can demand
that the isomorphism type of M
is the universe of M
(n : 2n + 1 B
1
) is isomorphic to M u
0
.
Similarly letting 2n + 1 : n < be the disjoint union of B
n
: n < ), each
B
n
innite. We use A
a
B
n
to code model with universe for some <
1
, by
a function F
n
. We then let N
a
be the model with universe a sucht hat the order
preserving function from a onto a countable ordinal is an isomorphism from N
a
onto F
n
(A
a
B
n
) : n < when the union is a model with universe .
Now we cannot demand them all M
)
(b) replace J
bd
1
, = lim
J
n
: n < ) but not necessarily n <
n
<
n+1
and add
P()/J is innite (hence uncountable).
Proof. In 1.5 in ()
3
we choose A
: <
1
), a sequence of subsets of such
that A
/J : <
1
) are pairwise distinct. This implies some changes and waiving
n
<
n+1
requires some changes in 1.9, in particular for each n using B
:
S
0
) with B
= lim(T
) : a
n
and the partition theorem [Sh:f,
XI,3.7,pg.549].
1.13
(
7
9
4
)
r
e
v
i
s
i
o
n
:
2
0
0
7

1
0

0
1
m
o
d
i
f
i
e
d
:
2
0
0
7

1
0

0
1
20 SAHARON SHELAH
REFERENCES.
[FoTo] Matthew Foreman and Stevo Todorcevic. A new L owenheim
Skolem theorem. Transactions of the American Mathematical Society,
357:16931715, 2005.
[RuSh 117] Matatyahu Rubin and Saharon Shelah. Combinatorial problems on
trees: partitions, systems and large free subtrees. Annals of Pure
and Applied Logic, 33:4381, 1987.
[Sh:E3] S. Shelah. On some problems in general topology. In Set Theory, Boise
ID, 19921994, volume 192 of Contemporary Mathematics, pages 91
101. 0708.1981.
[Sh:E12] Saharon Shelah. Analytical Guide and Corrections to [Sh:g].
math.LO/9906022.
[Sh:e] Saharon Shelah. Nonstructure theory, accepted. Oxford University
Press.
[Sh 420] Saharon Shelah. Advances in Cardinal Arithmetic. In Finite and In
nite Combinatorics in Sets and Logic, pages 355383. Kluwer Academic
Publishers, 1993. N.W. Sauer et al (eds.). 0708.1979.
[Sh 410] Saharon Shelah. More on Cardinal Arithmetic. Archive for Mathemat
ical Logic, 32:399428, 1993. math.LO/0406550.
[Sh:g] Saharon Shelah. Cardinal Arithmetic, volume 29 of Oxford Logic
Guides. Oxford University Press, 1994.
[Sh 430] Saharon Shelah. Further cardinal arithmetic. Israel Journal of Mathe
matics, 95:61114, 1996. math.LO/9610226.
[Sh:f] Saharon Shelah. Proper and improper forcing. Perspectives in Mathe
matical Logic. Springer, 1998.
[Sh 755] Saharon Shelah. Weak Diamond. Mathematica Japonica, 55:531538,
2002. math.LO/0107207.
[Sh 775] Saharon Shelah. Middle Diamond. Archive for Mathematical Logic,
44:527560, 2005. math.LO/0212249.
[Ve92a] Boban Velickovic. Forcing axioms and stationary sets. Advances in
Mathematics, 94:256284, 1992.