5

7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


A consistency result on weak reection
James Cummings

[email protected]
Mirna Dzamonja

[email protected]
Saharon Shelah

[email protected]
Hebrew University of Jerusalem
October 6, 2003
Abstract
In this paper we study the notion of strong non-reection, and its
contrapositive weak reection. We say strongly non-reects at i
there is a function F : such that for all < with cf() =
there is C club in such that F C is strictly increasing. We prove
that it is consistent to have a cardinal such that strong non-reection
and weak reection each hold on an unbounded set of cardinals less
than .
1
1 Introduction
In this paper we study the notion of strong non-reection, which was intro-
duced in [4] and is further studied in [3]. We prove that for a xed we can

Supported by a Postdoctoral Fellowship at the Hebrew University.

Partially supported by the Basic Research Fund of the Israel Academy of Science, and
a Postdoctoral Fellowship from the Hebrew University.

Partially supported by the Basic Research Fund of the Israel Academy of Science.
Paper number 571.
1
The research for this paper was done in the period July 1994 January 1995.
1
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


have an unbounded set of conalities at which strong non-reection holds,
and an unbounded set where it fails.
Denition 1: Let be a regular cardinal, and let be an ordinal with
.
S

= < [ cf() = .
S

<
= < [ cf() < .
If is uncountable, then strongly non-reects at i there is a func-
tion F : such that for all S

there is C club in such that

F C is strictly increasing. We will write SNR(, ) for this.
weakly reects at i does not strongly non-reect at .
In [4] Dzamonja and Shelah prove some theorems connecting weak club
principles, saturated ideals, and the ideal 1[, ) consisting of those A
such that there is h : increasing on a club at every point of AS

. In
particular Theorem 2.5 of that paper shows that a certain weak club principle
is incompatible with saturated ideals at successors of singulars, and Theorem
2.8 connects weak reection and the weak club principle.
In [3] strong non-reection is used as a tool to show that dierent instances
of stationary reection are independent. For example it is shown there that
every stationary subset of S

0
reects at a point in S

2
is consistent with
every stationary subset of S

1
has a non-reecting stationary subset.
We make a few remarks about the denitions. The next lemma is implicit
in Observation 1.2.3 from [4].
Lemma 1: Let be an ordinal, and
i
: i < ) a -sequence of ordinals
with cf(
i
) ,= . Let F : and F
i
:
i
witness strong non-
reection at for and each of the
i
. Then there is G :

i<

i

witnessing strong non-reection for

i<

i
.
2
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


Proof: Dene G(

i<j

i
) = F(j) and G(

i<j

i
+) = F
j
() for 0 < <
j
. It
is easy to check that this works.

It is proved in [4], using the previous lemma, that the least which weakly
reects at is a regular cardinal greater than .
As the terminology suggests, there is a connection between weak reection
and the more familiar notion of stationary reection.
Denition 2: Let < < be regular cardinals. Then Ref(, , ) i for
every stationary S S

there is S

such that S is stationary in .

We will use also Ref(, , < ) as a shorthand for < Ref(, , ).
The next fact shows that strong non-reection at is antithetical to
stationary reection to points of conality .
Lemma 2: Suppose that strongly non-reects at . Then for every sta-
tionary S there is T S stationary such that T is non-stationary
for all S

.
Proof: Use Fodors Lemma to nd T on which F : witnessing the
strong non-reection is constant. If C is a club on which F is strictly
increasing then C meets T in at most one point.

It is not hard to see that if C

: <
+
) is a 2

-sequence then the

function F : o.t.(C

) witnesses that
+
strongly non-reects at .
More is true, see 1.7 of [4]. The following remark is immediate from the
denition.
Lemma 3: If = cf() < <

, and weakly reects at , then

weakly
reects at .
We are now ready to state the main result.
3
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


Theorem 1: Let GCH hold, let be regular, and suppose that there are
cardinals
i
,
i
,
i
: i < ) such that for all i <
1.
i
= cf(
i
) <
+
i
<
i
= cf(
i
) <
i
, and
i
is measurable.
2.
i
> (sup
j
[ j < i )
++
.
Then there is a generic extension in which SNR(,
i
) and Ref(
i
,
i
, <
i
)
for every i. In particular, by Lemmas 2 and 3, weakly reects at
i
.
The proof will involve two stages. First we force functions that witness
the strong non-reection at the points
i
, via an iterated forcing with Easton
support. An important feature of the construction is that the individual
steps in the forcing have an increasing degree of strategic closure, and at any
stage a witness to the desired degree of strategic closure was added by the
previous stages.
We will show that the rst stage preserves the measurability of all the

i
. In the second stage we will force with a product of the Levy col-
lapses Coll(
i
, <
i
), and use Baumgartners argument from [2] to show that
Ref(
i
,
i
, <
i
) holds in the extension.
2 Forcing strong non-reection
Let and be regular cardinals with < . In this section we dene a
forcing P(, ) which adds a function from to witnessing strong non-
reection for at . We could make the same denition for an arbitrary
ordinal greater than , but for our purposes it will suce to restrict ourselves
to regular cardinals.
Denition 3: Conditions in P(, ) are functions p such that dom(p) < ,
rge(p) , and for every dom(p) if S

then there is is a club C in

such that p C is strictly increasing.
The condition p extends the condition q i p q. We write this as p q.
4
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


Clearly this forcing has at most
<
conditions, so enjoys the (
<
)
+
-c.c.
There are several pieces of information about the closure properties of the
forcing that we will need later.
Lemma 4: Let p

: < ) be a strictly decreasing sequence of conditions

in P(, ), where S

=
. Then p =

is the greatest lower bound for

the sequence.
Proof: Notice that cf(dom(p)) = cf() ,= , so that if dom(p) and
cf() = then dom(p

) for some < . There is C club in such that

p C = p

C is strictly increasing.

We remind the reader of the notion of strategic closure.

Denition 4: Let P be a partial ordering, and let be an ordinal.
1. The game G(P, ) is played by two players I and II, who take turns to
play elements p

of P for 0 < < , with player I playing at odd stages

and player II at even stages (Nota bene: limit ordinals are even).
The rules of the game are that the sequence that is played must be
decreasing (not necessarily strictly decreasing), the rst player who
cannot make a move loses, and player II wins if play proceeds for
stages.
2. P is -strategically closed i player II has a winning strategy in G(P, ).
3. P is < -strategically closed i for all < P is -strategically closed.
We say that a forcing notion P is < -distributive i it does not add any
< -sequence of ordinals to the ground model (equivalently, the intersection
of fewer than dense open sets is nonempty). The following lemma is easy.
Lemma 5: If P is < -closed it is -strategically closed, and if P is < -
strategically closed it is < -distributive.
5
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


Notice that P(, ) will only contain conditions of lengths unbounded in
if SNR(, ) holds for all < . This condition is actually enough to make
P(, ) be < -strategically closed.
Lemma 6: Suppose that all [, ) are strongly non-reecting at . Then
P(, ) is < -strategically closed.
Proof: Let < . If < then player II can win with the following
strategy; he plays p
2
=
def

<2
p

.
If < then by hypothesis there is a function F : witnessing
strong non-reection. Player II will play p
2
=
def
(

<2
p

) F(). We
check that this is a winning strategy.
Let 2 be an even stage of conality in G(P(, ), ). There is D club in
such that F D is strictly increasing. If we dene C = lh(p
2
) [ D
then C witnesses that II does not lose at stage 2.

We will be interested in forcing strong non-reection to several values of

simultaneously. For this we will use a certain dense subset of the < -support
product of the appropriate P(, ).
Denition 5: Let A REG. Then P(A, ) is the set of functions p such
that
1. dom(p) = (A ) for some < .
2. If dom(p) = (A ) and A then < p(, ) is a
condition in P(, ).
If p, q P(A, ) then p q i p extends q.
Clearly [P(A, )[
<
, so the forcing has the (
<
)
+
-c.c. We also record
some information about the closure of the forcing.
Lemma 7: Let p

: < ) be a strictly decreasing sequence of conditions in

P(A, ), where cf() / A. Then the condition p given by p(i) =
def

<
p

(i)
is the greatest lower bound for the sequence.
6
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


The next lemma is easy, with a proof almost identical to that of Lemma
6.
Lemma 8: Let A and be as above. Suppose that for all A, all
[, ) are strongly non-reecting at . Then P(A, ) is < -strategically
closed.
3 The iteration
The idea of the construction is now to dene A =
i
[ i < (where the
i
are as in the statement of Theorem 1) and to iterate P(A, ) for all regular
. A crucial point will be that the forcing at stage is < -strategically
closed, using Lemma 8 and the fact that in the iteration we have already
arranged the required instances of non-reection below .
We will do a Reverse Easton iteration, that is to say an iteration where
direct limits are taken at strongly inaccessible limit stages and inverse limits
are taken at other limit stages. We will refer to [1] for details about this sort
of iteration, and we will also follow the notation of that paper (in particular
P

is the forcing up to stage and

Q

V
P
is the forcing at ).
Formally, we will dene

Q

to be 0 if is not a regular cardinal, and

to be P(A , )
V
P if is regular. We will collect some information about
the iteration in the following lemma.
Lemma 9: Let P

and

Q

be as above, and let

R
,
be the canonical iteration
in V
P

such that P

R
,
has a dense subset isomorphic to P

. Then for all

regular
1. [P

[ .
2. V
P
GCH, so in particular V
P
[

[ = .
3. P
+1
has the
+
-c.c. In addition, if is Mahlo, then P

has the -c.c.

4. V
P

is < -strategically closed.

5. For all regular < , R
,
is < -strategically closed in V
P

.
7
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


6. P

preserves all cardinals and conalities.

Proof: The proof will be by induction on . Most of the proof is straight-
forward, using the results of Section 2 in [1] to power the induction. The
distinctive point here is in showing that clauses 4 and 5 hold at , given that
we have proved the lemma for regular cardinals less than .
By construction, P

forces that for every A and every regular cardinal

[, ) we have SNR(, ). As we remarked after Lemma 1, this implies
that for every ordinal [, ) we have SNR(, ). By Lemma 8 this means
that P(A , ) is < -strategically closed in V
P
.
Finally, to see that clause 5 holds one should check that Theorem 2.5
from [1] is still true if -closed is replaced by < -strategically closed.
This is routine, the point is that a term for a strategy can be applied to a
term for a condition to get a term for a stronger condition.

We make some remarks about this construction.

1. Since cardinals and conalities are preserved, a witness to strong non-
reection added at some stage by some P(, ) will remain a witness at
all subsequent stages.
2. At stage we forced with P(A, ), so added witnesses to all the strong
non-reection that is claimed in Theorem 1.
4 Preserving measurability
As we mentioned in the rst section, we want to show that for each i the
measurability of
i
is preserved by the iteration P

. It is enough to argue that

i
is measurable in the extension by P

+
i
+1
, because the rest of the forcing
is
+
i
-strategically closed, so that the power set of
i
does not change and a
measure remains a measure. For brevity, we will denote
i
by throughout
this section.
8
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


Let G be P

-generic over V , let g be P(A, )-generic over V [G], and let

h be P(A
+
,
+
)-generic over V [G][g]. Let j : V M be the ultrapower
map arising from a normal measure U on . We list some facts about j and
M, all of whose proofs can be found in [5].
1. crit(j) = .
2.

M M.
3. H

+ M.
4.
+
=
+
M
.
5.
+
< j() < j(
+
) <
++
.
6. M = j(F)() [ F V and dom(F) = .
The strategy of the proof will be to dene, in V [G][g][h], an extension
of j : V M to a new embedding j : V [G][g][h] N V [G][g][h].
The existence of such an extension will imply that is still measurable in
V [G][g][h].
We start by comparing the iterations P

+
+1
and j(P

+
+1
). The forcing
j(P

+
+1
) is an iteration dened in M, forcing strong non-reection at co-
nalities in the set j(A) j(
+
) for all M-regular cardinals up to j(
+
). Since
A is bounded in and
i+1
>
+
, we see that
A (
+
+ 1) = j(A) (j(
+
) + 1) = A .
By the resemblance between V and M, if we compute the iteration
j(P

+
+1
) up to stage
+
we get P

+
+1
. We can therefore compute a generic
extension M[G][g][h] of M by using the V -generic lters, and observing that
V -generic lters are M-generic.
We claim that V [G][g][h]

(M[G][g][h]) M[G][g][h]. Since P
+1
is
+
-
c.c. every canonical P
+1
-name for a -sequence of ordinals is in M, so that
easily V [G][g]

(M[G][g]) M[G][g]. The forcing Q

+ is <
+
-strategically
closed in V [G][g], so it adds no -sequence of ordinals, and we are done.
9
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


In M[G][g][h] let R = R

+
+1,j()
be the canonical factor forcing to prolong
G g h to a j(P

)-generic. We claim that R is

+
-strategically closed in
V [G][g][h]. This follows from the fact that R is <
++
M
-strategically closed in
M[G][g][h] and the fact that V [G][g][h]

(M[G][g][h]) M[G][g][h]. The
point is that if II plays for
+
steps in V [G][g][h] using the strategy from
M[G][g][h], then every initial segment of the play is in M[G][g][h], so that
player II does not get stuck at any stage below
+
.
The previous claim explains why we are working in V [G][g][h] rather than
V [G][g]. If we truncate j(P

) at + 1 then the rest of the forcing will be

<
+
-strategically closed in V [G][g], but the following stage of the proof will
demand
+
-strategic closure.
Recall that
+
< j() <
++
. In M[G][g][h] the forcing R is j()-c.c. and
has size j(), so there are at most j() maximal antichains in that model. In
V [G][g][h] let us enumerate these antichains as A

: <
+
). Now consider
a run of the game G(R,
+
) in which player I plays the following strategy;
in response to p
2
player I will choose some element q

of A

such that
p
2
is compatible with q

, and then will play p

2+1
which is some common
renement. Player II will play according to some winning strategy; after

+
steps we have built a decreasing sequence of conditions which clearly
generates an M[G][g][h]-generic lter H.
Now we will start to extend j. Dene G
+
=
def
G g h H, which
will be j(P

)-generic over M. We attempt to dene j : V [G] M[G

+
] by
j(
G
) =
def
j( )
G
+
. We check that this is a well-dened elementary embed-
ding, using the following well-known fact.
Lemma 10: Let k : M N be an elementary embedding between two
transitive models of ZFC. Let P M be some forcing, let k(P) = Q, and
suppose that we have G which is P-generic over M and H which is Q-generic
over N. Suppose also that kG H. Then dening k(
G
) = k( )
H
for every
M
P
gives a well-dened elementary embedding k : M[G] N[H],
which extends k : M N and has k(G) = H.
Proof: Easy, using the Truth Lemma and the elementarity of k.

10
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


By Lemma 10, it is enough to check that jG G
+
. G is generic for
P

which was constructed as a direct limit, so every condition p in G has

support bounded in . Since crit(j) = , the condition j(p) contains the
same information as p, and since G = G
+
we conclude that j(p) G
+
.
Since G
+
V [G][g][h], we see that M[G
+
] V [G][g][h]. Also, we know
that H is generic for a forcing which adds no -sequences of ordinals over
M[G][g][h], so that V [G][g][h]

(M[G
+
]) M[G
+
].
Now we aim to lift j further to get a map with domain V [G][g]. In V [G]
the forcing Q

has cardinality , and is < -strategically closed with at most

2

(that is
+
) many maximal antichains. Since j : V [G] M[G
+
] is
elementary, in M[G
+
] the forcing Q
j()
is < j()-strategically closed with at
most j(
+
) maximal antichains.
Arguing as before, Q
j()
is
+
-strategically closed in V [G][g][h]. Since
j(
+
) <
++
we can repeat the argument from the construction of H to
build g
+
which is Q
j()
-generic over M[G
+
]. But it is not clear at this point
that we can lift j onto V [G][g], because it may not be the case that jg g
+
.
We will use Silvers master condition idea. Observe that g M[G
+
],
and that g is equivalent to a function p where dom(p) = (A ) and
p(, ) : < p(, ) witnesses SNR(, ) for each A .
Recall that Q
j()
is dened to be P(j(A) j(), j()). We claim that
p Q
j()
. The support condition is satised because < j() and (as we
saw before) j(A) j() = j(A ) = A . It is enough to show that for
each we have p(, ) P(, j()), which is to say that for all of
conality there is a club in on which p(, ) is increasing. This is easy
because V , M and M[G
+
] agree about cardinals and conalities up to
+
.
Since p is a condition in Q
j()
, when we construct g
+
we can arrange that
g
+
p. We claim that this suces to guarantee that jg g
+
. This follows
from the observation that p j(q) for every q g, which is true because q
has size less than and so j(q) is just a copy of q.
We can now build j : V [G][g] M[G
+
][g
+
], using Lemma 10. Before
we can nish the construction, we need one piece of information about this
embedding. We claim that
M[G
+
][g
+
] = j(F)() [ F V [G][g] and dom(F) = .
11
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


To see this let
G
+
g
+
be some element of M[G
+
][g
+
], where is a P
j()+1
-
name in M. We know that = j(f)() for some f V , and we may
as well assume that f() is a P
+1
-name for every < . Let us dene
F V [G] by F() =
def
f()
Gg
. Since j(G g) = G
+
g
+
, we see that
j(F)() = (j(f)())
G
+
g
+
=
G
+
g
+
as required.
We will now dene a lter h
+
on Q
j(
+
)
, by setting
h
+
=
def
q [ (p h) j(p) q .
It is easy to see that h
+
is in fact a lter, and certainly jh h
+
and
h
+
V [G][g][h]. We claim that h
+
is generic. To see this, let D M[G
+
][g
+
]
be a dense subset of Q
j(
+
)
. We know D = j(F)() for some F V [G][g].
Dene E Q

+ by
E =

F() [ F() is a dense subset of Q

+ .
Q

+ is <
+
-distributive, so that E is dense, and clearly E V [G][g]. There-
fore there is some p E h. Certainly j(p) h
+
, and by elementarity
D = j(F)() j(E) so that j(p) h
+
D.
In conclusion, we can dene j : V [G][g][h] M[G
+
][g
+
][h
+
] in the
model V [G][g][h], so that is still measurable in V [G][g][h].
5 The collapse
To save on notation, we will now denote the model V
P

constructed in Section
3 by V . In this model we have the following situation. For all i
1. GCH holds.
2.
i
= cf(
i
) <
+
i
<
i
= cf(
i
) <
i
, and
i
is measurable.
3.
i
> (sup
j
[ j < i )
++
.
4. SNR(,
i
) holds for every i.
12
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


We still have to get the reection property Ref(
i
,
i
, <
i
) for every i.
We will do this by collapsing the measurable cardinals
i
, using an idea from
Section 7 of [2]. We will also check that this collapse does not destroy the
strong non-reection.
Let S
i
=
def
Coll(
i
, <
i
). We dene S to be the Easton product of the
S
i
, to be precise p S i p is a function with
1. dom(p) .
2. p(i) S
i
for all i dom(p).
3. If is an inaccessible cardinal and i < =
i
< , then
dom(p) is bounded in .
The ordering is the natural one.
For each i, the forcing S factorises as S
l
i
S
i
S
u
i
, where S
l
i
talks about the
coordinates below i and S
u
i
talks about those above. Using Eastons Lemma
and the GCH, it is easy to see that S collapses cardinals in the interval [
i
,
i
)
to
i
and preserves all other cardinals. In particular SNR(,
i
) still holds
in V
S
, because
i
is still regular and there are no new points of conality
i
(this is easy, because by our assumptions on
i
we have [S
l
i
[ <
i
, and S
i
S
u
i
is
+
i
-closed).
For the reection, it will suce to check that Ref(
i
,
i
, <
i
) holds in
V
S
l
i
S
i
, because this model agrees with V
S
up to
i+1
. We will look at V
S
l
i
S
i
in a slightly dierent way, by writing it as (V
S
i
)
S
l
i
.
Since GCH holds in V and
i
is measurable there, the results of [2] show
that Ref(
i
,
i
, <
i
) holds in V
S
i
. Of course,
i
is now
+
i
. We claim that
Ref(
i
,
i
, <
i
) still holds in (V
S
i
)
S
l
i
. Observe that [S
l
i
[ <
i
<
i
, so that if S
is a stationary subset of S

i
<
i
in (V
S
i
)
S
l
i
then there is T S stationary with
T V
S
i
. By the reection which holds in V
S
i
, there is S

i
such that
T is stationary in V
S
i
. Since [S
l
i
[ <
i
, we see that T is still stationary
in (V
S
i
)
S
l
i
(and of course still has conality
i
).
We have shown that SNR(,
i
) and Ref(
i
,
i
, <
i
) hold in V
S
for all
i < . This concludes the proof of Theorem 1.
13
5
7
1


r
e
v
i
s
i
o
n
:
1
9
9
5
-
0
4
-
1
8







m
o
d
i
f
i
e
d
:
1
9
9
5
-
0
4
-
2
2


References
[1] J. Baumgartner, Iterated forcing, in Surveys in set theory (ed. A. Math-
ias), London Mathematical Society Lecture Notes 87, Cambridge Uni-
versity Press (1983), 159.
[2] J. Baumgartner, A new class of order types, Annals of Mathematical
Logic 9 (1976), 187222.
[3] J. Cummings, Weak reection and strong non-reection (in prepara-
tion).
[4] M. Dzamonja and S. Shelah, Saturated lters at successors of singulars,
weak reection and yet another weak club principle (submitted to Annals
of Pure and Applied Logic).
[5] A. Kanamori, W. Reinhardt and R. Solovay, Strong axioms of innity
and elementary embeddings, Annals of Mathematical Logic 13 (1978),
73-116.
14