Professional Documents
Culture Documents
Andrzej Roslanowski and Saharon Shelah - Reasonable Ultrafilters, Again
Andrzej Roslanowski and Saharon Shelah - Reasonable Ultrafilters, Again
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN
ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Abstract. We continue investigations of reasonable ultralters on uncount-
able cardinals dened in Shelah [12]. We introduce stronger properties of
ultralters and we show that those properties may be handled in support
iterations of reasonably bounding forcing notions. We use this to show that
consistently there are reasonable ultralters on an inaccessible cardinal with
generating systems of size less than 2
=
++
and there
is a very reasonable ultralter generated by a system of size
+
(Corollary 3.4). It
should be stressed that generating an ultralter has the specic meaning stated
in Denition 1.3(3). In particular, having a small generating system does not
imply having small ultralter base.
The fourth section shows that some technical inconveniences of the proofs from
the third sections reect the delicate nature of our concepts, not necessarily our lack
of knowledge. We give an example of a nicely double abounding forcing notion
which kills ultralters generated by systems from the ground model. Then we show
that for an inaccessible cardinal , it is consistent that 2
=
++
and there is no
ultralter generated by a system of size
+
(see Corollary 3.4).
Studies of ultralters generated according to the schema introduced in [12] are
also carried out in Roslanowski and Shelah [10].
Notation: Our notation is rather standard and compatible with that of classical
textbooks (like Jech [5]). In forcing we keep the older convention that a stronger
condition is the larger one.
(1) Ordinal numbers will be denoted be the lower case initial letters of the
Greek alphabet (, , , . . .) and also by i, j (with possible sub- and su-
perscripts). Cardinal numbers will be called , , (with possible sub- and
superscripts). is always assumed to be regular, sometimes even
strongly inaccessible.
By we will denote a suciently large regular cardinal; H() is the
family of all sets hereditarily of size less than . Moreover, we x a well
ordering <
of H().
(2) A sequence is a function with the domain being a set of ordinals. For two
sequences , we write whenever is a proper initial segment of ,
and when either or = . The length of a sequence is the
order type of its domain and it is denoted by lh().
(3) We will consider several games of two players. One player will be called
Generic or Complete or just COM, and we will refer to this player as she.
Her opponent will be called Antigeneric or Incomplete or just INC and will
be referred to as he.
(4) For a forcing notion P, all Pnames for objects in the extension via P will
be denoted with a tilde below (e.g.,
, X
P
. The weakest element of P will be denoted by
P
(and we will always assume that there is one, and that there is no other
condition equivalent to it). We will also assume that all forcing notions
under consideration are atomless.
By support iterations we mean iterations in which domains of con-
ditions are of size . However, we will pretend that conditions in a
support iteration
Q = P
, Q
: <
and for
p lim(
Q) and
.
(5) For a lter D on , the family of all Dpositive subsets of is called D
+
.
(So A D
+
if and only if A and A B ,= for all B D.)
The club lter of is denoted by T
.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 3
1. More reasonable ultrafilters on
Here we recall some basic denitions and results from [12], and then we introduce
even stronger properties of ultralters and/or generating systems. We also show
that assumptions like
S
of such that
_
_
[
,
+1
) : C
_
/ D.
We want to investigate ultralters on which are generated by systems dening
largeness in by giving a condition based on largeness in intervals below .
The family Q
0
)
is a lter, and (<
+
)directness has the avour of Ppointness.
Denition 1.3 ([12, Def. 2.5]). (1) Let Q
0
: C
p
), d
p
: C
p
))
such that
(i) C
p
is a club of consisting of limit ordinals only, and for C
p
:
(ii) Z
p
=
_
, min
_
C
p
( + 1)
__
and
(iii) d
p
T(Z
p
.
(2) For q Q
0
we let
l(q)
def
=
_
A : ( < )( C
q
)(A Z
q
d
q
)
_
,
and for a set G
Q
0
we let l(G
)
def
=
l(p) : p G
. We also dene
a binary relation
0
on Q
0
by
p
0
q if and only if l(p) l(q).
(3) We say that an ultralter D on is reasonable if it is weakly reasonable
(see 1.1) and there is a directed (with respect to
0
) set G
Q
0
such that
D = l(G
). The family G
Q
0
such that
D = l(G
).
Denition 1.4. Suppose that
(a) X is a non-empty set and e is an ultralter on X,
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
4 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
(b) d
x
is an ultralter on a set Z
x
(for x X).
We let
e
xX
d
x
=
_
A
_
xX
Z
x
: x X : Z
x
A d
x
e
_
.
(Clearly,
e
xX
d
x
is an ultralter on
xX
Z
x
.)
Proposition 1.5 ([12, Prop. 2.9]). Let p, q Q
0
__
C
p
__
A Z
p
d
p
_
,
(c) there is < such that
if C
q
,
0
= sup
_
C
p
( + 1)
_
,
1
= min
_
C
p
min(C
q
( + 1))
_
,
then there is an ultralter e on [
0
,
1
) C
p
such that
d
q
=
_
A Z
q
: A
e
d
p
: [
0
,
1
) C
p
_
.
Observation 1.6 (Compare [12, Prop. 2.3(4)]). If p Q
0
, A , then there is
q Q
0
such that p
0
q and either A l(q) or A l(q).
Denition 1.7 ([12, Def. 2.10]). Let p Q
0
. Suppose that X [C
p
]
and C C
p
is a club of such that
if < are successive elements of C,
then [[, ) X[ = 1.
(In this situation we say that p is restrictable to X, C).) We dene the restriction
of p to X, C) as an element q = pX, C) Q
0
such that C
q
= C, and if <
are successive elements of C, x [, ) X, then Z
q
= [, ) and d
q
= A Z
q
:
A Z
p
x
d
p
x
.
Proposition 1.8 ([12, Prop. 2.11]). (1) If G
Q
0
is
0
directed and [G
[
, then G
has a
0
upper bound. (Hence, in particular, l(G
) is not an
ultralter.)
(2) Assume that G
Q
0
is
0
directed and
0
downward closed, p G
,
X [C
p
]
and C C
p
is a club of such that p is restrictable to X, C).
If
xX
Z
p
x
l(G
), then pX, C) G
.
The following denition is used here to simplify our notation in 1.11 only. How-
ever, these concepts play a more central role in [10].
Denition 1.9. (1) Let Q
we dene
l
(r) =
_
A :
_
<
__
(, Z, d) r
__
A Z d
__
,
and we dene a binary relation
on Q
by
r
1
r
2
if and only if (r
1
, r
2
Q
and) l
(r
1
) l
(r
2
).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 5
(2) For a set G
we let l
(G
) =
_
l
(r) : r G
.
(3) We say that an r Q
_
<
__
[(, Z, d) r : = [ < 2
_
, and
_
(
1
, Z
1
, d
1
), (
2
, Z
2
, d
2
) r
__
1
<
2
Z
1
2
_
.
(4) For p Q
0
we let #(p) = (, Z
p
, d
p
) : C
p
.
Observation 1.10. (1) If p Q
0
then #(p) Q
(#(p)). Also, if r Q
(r) = l(p)
for some p Q
0
.
(2) Let r, s Q
. Then r
, Z
, d
) r
__
A Z
__
.
The various denitions of super reasonable ultralters introduced in Denition
1.11 below are motivated by the proof of the Sacks forcing preserves Ppoints.
In that proof, a fusion sequence is constructed so that at a stage n < of the
construction one deals with nitely many nodes in a condition (the nodes that are
declared to be kept). We would like to carry out this kind of argument, e.g., for
forcing notions used in [9, B.8.3, B.8.5], but now we have to deal with < nodes in
a tree, and the ultralter we try to preserve is not that complete. So what do we do?
We deal with nitely many nodes at a time eventually taking care of everybody.
One can think that in the denition below the set I
) is an ultralter
by many forcing notions. We would also love to preserve that property itself, but
we failed to achieve it. The super reasonability is what we need to preserve the
ultralter (see 3.2), strong reasonability is what we can prove about G
in the
extension (see 3.3).
Denition 1.11. Let G
Q
0
and let =
for < .
(1) We dene a game
(G
(G
) lasts steps and at a stage < of the play the players choose
I
, i
, u
and r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
of cardinality <
and an
enumeration u
= u
,i
: i < i
) of [I
]
<
(so i
<
0
).
Next the two players play a subgame of length i
. In the i
th
move of
the subgame,
(a) COM chooses r
,i
G
, and then
(b) INC chooses r
,i
G
such that r
,i
0
r
,i
, and nally
(c) COM picks (
,i
, Z
,i
, d
,i
) #
_
r
,i
_
such that
,i
> .
In the end of the play COM wins if and only if
() there is r G
: < )
<
I
we have
(
,i
, Z
,i
, d
,i
) : < , j
u
,i
and i < i
#(r).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
6 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
A game
(G
) is dened similarly to
(G
the set
Z
,i
: < , i < i
and j
u
,i
belongs to l(G
).
(2) We say that the family G
reasonable,
respectively) if
(i) G
is (<
+
)directed (with respect to
0
), and
(ii) if s G
, r Q
0
= d
s
for C
r
, then r G
, and
(iii) INC has no winning strategy in the game
(G
) (
(G
), respec-
tively).
(3) We say that a uniform ultralter D on is super reasonable ( super
reason-
able, respectively) set G
Q
0
).
(4) If
= for all < , then we omit and say just super reasonable
or super
Q
0
instead of
.
Denition 1.12. Let G
Q
0
:
< ) be a sequence of cardinals, 2
for < .
(1) A game
(G
(G
, i
, u
and r
,i
,
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) applying
the following procedure.
First, INC chooses a non-empty set I
of cardinality <
, and then
COM chooses i
= u
,i
: i < i
) of non-empty
nite subsets of I
such that I
i<i
u
,i
.
Next the two players play a subgame of length i
. In the i
th
move of
the subgame,
(a) COM chooses r
,i
G
and then
(b) INC chooses
,i
< , and nally
(c) COM picks (
,i
, Z
,i
, d
,i
) #
_
r
,i
_
such that
,i
is above
,i
and .
In the end of the play COM wins if and only if
() there is r G
: < )
<
I
we have
(
,i
, Z
,i
, d
,i
) : < , j
u
,i
and i < i
#(r).
A game
(G
) is dened similarly to
(G
the set
Z
,i
: < , i < i
and j
u
,i
belongs to l(G
).
(2) If G
Q
0
is (<
+
)directed (with respect to
0
) and INC has no winning
strategy in the game
(G
is strongly reasonable.
Also, G
is said to be strongly
reasonable if it is (<
+
)directed and
INC has no winning strategy in the game
(G
).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 7
(3) We say that a uniform ultralter D on is strongly reasonable (
strongly
Q
0
). If
= for all < , then we omit and say just strongly reasonable or
strongly
reasonable.
Observation 1.13. Assume that 2
: <
), =
Q
0
reasonable super
reasonable strongly
reasonable
Proposition 1.14. Assume that 2
: < ). If
a uniform ultralter D on is strongly
reasonably family G
Q
0
).
Then G
is (<
+
)directed and the proof will be completed once we show that D
is weakly reasonable.
Let f
. We will argue that for some club C =
: < we have
[, +f()) : C / D, where
(G
: <
) so that at a stage < of the play, when
, i
, u
, r
,i
,
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
is the result of the play so far, then
if is limit, then
= sup(
: < ),
if is not limit, then
= sup
_
Z
,i
: i < i
, <
_
+ 1.
Now (at the stage ) st(f) instructs INC to choose I
, u
,i
=
+f(
) + sup
_
_
Z
,j
: j < i
_
+ 890.
(After this COM chooses (
,i
, Z
,i
, d
,i
) #
_
r
,i
_
with
,i
>
,i
.)
The strategy st(f) cannot be the winning one for INC, so there is a play
, i
, u
, r
,i
,
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
of
(G
def
=
_
_
Z
,i
: < , i < i
_
l(G
) = D
(note that necessarily u
,i
= I
,
,i
that
for each <
[
+f(
))
_
_
Z
,i
: < , i < i
_
= ,
and hence also
_
[
+ f(
)) : <
_
A
= . Consequently
_
[
+
f(
)) : <
_
/ D and one can easily nish the proof.
Proposition 1.15. Assume =
<
and
S
: <
+
) Q
0
such that
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
8 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
(i) ( < <
+
)(r
0
r
), and
(ii) the family
G
def
=
_
r Q
0
: ( <
+
)(r
0
r
)
_
is super reasonable and l(G
) is an ultralter on .
Proof. The sequence r
: <
+
) will be constructed inductively. At successor
stages we will use 1.6 to make sure that l(G
(G
(Q
0
) of the form
()
1
, i
, u
, r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
where each I
<
X
1
. Next, for
< , 0 < I < and an enumeration u = u
j
: j < i) of [I]
<
let X
2
,I, u
be the
set of all legal plays of
(Q
0
) of the form
()
2
,I, u
(I, i, u))
r
j
, r
j
, (
j
, Z
j
, d
j
) : j < j
r),
where X
1
, j
< i (and r
j
, r
j
, (
j
, Z
j
, d
j
) : j < j
j
, r Q
0
). Also let
X
2
=
_
X
2
,I, u
: < and 0 < I < and
u = u
j
: j < i) is an enumeration of [I]
<
_
.
Any strategy for INC in
(Q
0
(I, i, u))
r
j
, r
j
, (
j
, Z
j
, d
j
) : j < j
is such
that r
0
st( ).
Below, whenever we say a strategy for INC we mean a function st satisfying con-
ditions ()
3
()
5
.
Since [Q
0
[ = 2
2
<
=
+
, we may pick a bijection
0
: Q
0
11
+
and for <
+
let A
consist of all X
1
X
2
such that
0
(r) < for all elements r Q
0
consist of all
pairs ( , a) such that
A
) then
0
(a) < .
Note that [A
[ and [}
<
+
}
. Plainly
[}[ =
+
so we may x a bijection
1
:
+
onto
}. Let
C = <
+
:
1
[] = }
;
it is a club of
+
.
Let A
: <
+
) list all subsets of and let B
: S
) be a diamond
sequence on S
= <
+
: cf() = . By induction on <
+
we choose
a
0
increasing sequence r
: <
+
) Q
0
: < ).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 9
Case 0: = 0.
We let r
0
be the <
rst member of Q
0
.
Case 1: = + 1.
Pick r
Q
0
such that r
0
r
and either A
l(r
) or A
l(r
) (remember
Observation 1.6).
Case 2: is a limit ordinal, cf() < .
Pick r
Q
0
0
r
C and ( < )(
0
(r
(Q
0
) such that
1
[B
] = st }
= stA
.
If the answer to ()
6
Q
0
as in Case 2.
Suppose now that the answer to ()
6
] = st }
= stA
. Let
=
: < ) be
an increasing continuous sequence conal in . Consider a play
=
, i
, u
, r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
of
(Q
0
(Q
0
), at step i < i
(Q
0
))
COM chooses r
,i
= r
,i
by st, she picks the
<
rst (
,i
, Z
,i
, d
,i
) #(r
,i
) satisfying:
()
7
,,i
( )(A d
,i
)( C
r
)(A Z
r
d
r
)(Z
,j
,i
) and (j < i)(Z
,j
,i
).
The above rules fully determine the play and it should be clear that A
and
only (and not on st, provided it
is as required by ()
6
).
By the demands ()
8
,,i
, we may choose an increasing continuous sequence
:
< ) such that
0
= 0 and ( < )(i < i
)(Z
,i
[
,
+1
)). Now, for
< choose an ultralter e
on i
such that
()
9
,
_
j I
__
i < i
: j u
,i
e
_
and let d
be an ultralter on [
,
+1
) such that
()
10
,
e
_
d
,i
: i < i
_
d
.
Now let r
Q
0
be such that
C
r
: < , and
if =
, then Z
r
= [
,
+1
) and d
r
= d
.
One easily veries that r
0
r
0
r
for every
j = j
: < )
<
I
we have
_
(
,i
, Z
,i
, d
,i
) : < & j
u
,i
& i < i
,i
_
#(r
).
After the construction of r
: <
+
) is carried out we let
G
= r Q
0
: ( <
+
)(r
0
r
).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
10 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Plainly, G
) is an ultralter on
(remember Case 1 of the construction). We should argue that INC has no winning
strategy in
(G
is a strategy of INC in
(G
).
Pick S
) < ) and
1
[B
] = st
= st
.
Then when choosing r
and we constructed a
play of
(Q
0
,
so it is a play of
(G
means that r
witnesses that
COM wins the play and consequently st
= (Q
0
,
0
).
(1) Q
0
is a (<
+
)complete forcing notion of size 2
2
<
.
(2)
Q
0
Q
0
Q
0
) is an ultralter .
Proof. (1) Should be clear; see also Proposition 1.8(1).
(2) By the completeness of Q
0
Q
0
l(G
Q
0
) is a uniform ultralter on .
It should also be clear that G
Q
0
).
Let us argue that
Q
0
(G
Q
0
)
and to this end suppose p Q
0
and st
is a Q
0
st
is a strategy of INC in
(G
Q
0
) .
We are going to construct a condition q Q
0
(Q
0
) such that
q
Q
0
is a play of
(G
Q
0
, ()
2
,I, u
there). We may
assume that
p
Q
0
st
is a function satisfying ()
3
()
5
of the proof of 1.15 .
By induction on < we choose conditions p
Q
0
X
1
so that
()
1
p
0
p
0
p
and
Q
0
is a partial play of
(G
Q
0
,
()
3
if
, i
, u
, r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
, then for
every < < and j < i < i
we have
r
,i
0
p
and Z
,j
,i
and Z
,j
,0
.
Suppose that =
. Pick p
0
p
and I
, i
, u
such that p
st
) = (I
, i
, u
, r
,i
, r
,i
and (
,i
, Z
,i
, d
,i
) for i < i
we have
()
4
(i) p
0
= p
, p
i
0
p
j
, p
i
= r
,i
0
r
,i
0
p
i+1
, and
(ii) p
i+1
,i
is the answer by st
,i
),
(iv) (A d
,i
)(
)( C
p
)(A Z
p
d
p
).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 11
Then p
+1
is any
0
upper bound to p
i
: i < i
.
The limit stages of the construction should be clear.
After the construction is carried out and we have
: < , we dene
r Q
0
like r
there). Then r is
0
stronger
then all p
is a play of
(G
Q
0
.)
2. More on reasonably complete forcing
Denition 2.1. Let P be a forcing notion.
(1) For a condition r P let
0
(P, r) be the following game of two players,
Complete and Incomplete:
the game lasts at most moves and during a play the
players attempt construct a sequence (p
i
, q
i
) : i < ) of
pairs of conditions from P in such a way that (j < i <
)(r p
j
q
j
p
i
) and at the stage i < of the game,
rst Incomplete chooses p
i
and then Complete chooses q
i
.
Complete wins if and only if for every i < there are legal moves for both
players.
(2) We say that the forcing notion P is strategically (<)complete if Complete
has a winning strategy in the game
0
(P, p) for each condition p P.
(3) Let N (H(), , <
N for an ordinal
we have p
N .
(4) P is proper in the standard sense (or just: proper) if there is x H()
such that for every model N (H(), , <
) satisfying
<
N N, [N[ = and P, x N,
and every condition p N P there is an (N, P)generic condition q P
stronger than p.
Theorem 2.2 (See Shelah [11, Ch. III, Thm 4.1], Abraham [1, 2] and Eisworth
[2, 3]). Assume 2
=
+
,
<
= . Let
Q = P
i
, Q
i
: i <
++
) be support
iteration such that for all i <
++
we have
P
i
is proper,
Pi
[Q
i
[
+
.
Then
(1) for every <
++
,
P
=
+
, and
(2) the limit P
++ satises the
++
cc.
Proposition 2.3 ([9, Prop. A.1.6]). Suppose
Q = P
i
, Q
i
: i < ) is a support
iteration and, for each i < ,
Pi
Q
i
is strategically (<)complete .
Then, for each and r P
0
(P
0
(P
0
, r
0
) in which Complete follows the
strategy st(
0
, r
0
), then (p
i
r[
0
,
1
), q
i
r[
0
,
1
)) : i < ) is a play of
0
(P
1
, r) in which Complete uses st(
1
, r);
(ii) if (p
i
, q
i
) : i < ) is a play of
0
(P
1
, r) in which Complete plays ac-
cording to the strategy st(
1
, r), then (p
i
0
, q
i
0
) : i < ) is a play of
0
(P
0
, r
0
) in which Complete uses st(
0
, r
0
);
(iii) if (p
i
, q
i
) : i < i
) is a partial play of
0
(P
1
, r) in which Complete uses
st(
1
, r) and p
P
0
is stronger than all p
i
0
(for i < i
), then there is
p
P
1
such that p
= p
0
and p
p
i
for i < i
.
Denition 2.4 (Compare [8, Def. 2.2]). (1) Let be an ordinal, w . A
standard (w, 1)
: w ),
(T, ) is a tree with root ) and such that every chain in T has a
upper bound in T,
if t T, then there is t
T such that t t
and rk(t
) = .
We will keep the convention that T
x
y
is (T
x
y
, rk
x
y
).
(2) Let
Q = P
i
, Q
i
: i < ) be a support iteration. A standard tree of
conditions in
Q is a system p = p
t
: t T) such that
(T, rk) is a standard (w, 1)
i
: i < ) is a
support iteration such that for all i < we have
Pi
Q
i
is strategically (<)complete .
Suppose that p = p
t
: t T) is a standard tree of conditions in
Q, [T[ < , and
1 P
:
< ) be a sequence of regular cardinals such that
0
, p
t
, q
t
: t I
) : <
_
is constructed. Suppose that the players have arrived to a stage < of
the game. Now,
()
of cardinality <
and a
system p
t
: t I
) of conditions from Q,
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 13
()
t
: t I
) of condi-
tions from Q such that (t I
)(p
t
q
t
).
At the end, Generic wins the play
_
I
, p
t
, q
t
: t I
) : <
_
of
rcA
(p, Q) if and only if
()
rc
A
there is a condition p
Q
_
<
__
t I
__
q
t
G
Q
_
.
(2) We say that a forcing notion Q is reasonably Abounding over if
(a) Q is strategically (<)complete, and
(b) for any p Q, Generic has a winning strategy in the game
rcA
(p, Q).
Denition 2.7 (See [8, Def. 3.2]). Let
Q = P
, Q
: < ) be a support
iteration and let =
= lim(
treeA
(p,
Q) between two players, Generic and Antigeneric, as follows. A
play of
treeA
(p,
Q) lasts steps and in the course of a play a sequence
T
, p
, q
tree T
such that [T
[ <
and
a tree of conditions p
= p
t
: t T
) P
, picks also w = w
),
()
= q
t
: t T
) P
such that p
.
At the end, Generic wins the play T
, p
, q
: < ) of
treeA
(p,
Q) if and
only if
()
tree
A
there is a condition p
P
_
<
__
t T
__
rk
(t) = & q
t
G
P
_
(2) We say that P
= lim(
Q) is reasonably
A(
.
Theorem 2.8 (See [8, Thm 3.2]). Assume that
(a) is a strongly inaccessible cardinal,
(b) =
: < ), each
and
_
f
__
<
f()
<
_
,
(c)
Q = P
, Q
= lim(
Q) is reasonably
A(
is also
proper).
In [8, 3], in addition to Areasonable completeness game we considered its
variant called areasonable completeness game. In that variant, at stage <
of the game the players played a subgame to construct a sequence p
, q
: <
i
) (corresponding to p
t
, q
t
: t I
: < ) be a sequence
of cardinals such that
0
rc2a
(p, Q) between Generic and Antigeneric as follows. A play of
rc2a
(p, Q)
lasts at most steps and in the course of the play the players try to con-
struct a sequence
()
, p
, q
: <
) : <
_
.
(Here
and
, q
: <
). At a stage =
i +j (where i <
,
j <
Q stronger
than all conditions q
+ j (where i
< i),
and then Antigeneric answers with a condition q
stronger than p
.
At the end, Generic wins the play () of
rc2a
(p, Q) if and only if both
players had always legal moves and
()
rc
2a
there is a condition p
Q
_
<
__
j <
__
q
i+j
: i <
Q
_
.
(2) Games
rc2b
,U
(p, Q) (for p Q) are dened similarly, we only replace condi-
tion ()
rc
2a
by
()
rc
2b
there is a condition p
Q
_
< :
_
j <
__
q
i+j
: i <
Q
__
|
Q
,
where |
Q
is the (Qname for the) normal lter generated by | in V
Q
.
(3) A strategy st for Generic in
rc2a
(p, Q) (or
rc2b
,U
(p, Q)) is said to be nice
if for every play
, p
, q
: <
) : <
_
in which she uses st,
for every < , the conditions in p
: <
, p
mod
.)
(4) Let x a, b. A forcing notion Q is nicely double xbounding over (and
| if x = b) if
(a) Q is strategically (<)complete, and
(b) Generic has a nice winning strategy in the game
rc2a
(p, Q) (
rc2b
,U
(p, Q)
if x = b) for every p Q.
Remark 2.10. (1) Reasonable double xboundedness (for x a, b) is an
iterable relative of reasonable xboundedness introduced in [8, Denition
3.1, pp 206-207]. Technical dierences in the denitions of suitable games
are to achieve the preservation of the corresponding property in support
iterations (see Theorems 2.13, 2.14 below).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 15
(2) The game
rc2b
,U
(p, Q) is easier to win for Generic than
rc2a
(p, Q) (because
the winning criterion is weaker). Therefore, if we are interested in
properness for support iterations only, then 2.14 will cover a larger class
of forcing notions than 2.13.
Denition 2.11 (See [8, Def. 6.1]). Suppose that is inaccessible and =
:
< ) is a sequence of cardinals, 1 <
: C
p
.
The order
P
= of P
is given by:
p
P
q if and only if C
q
C
p
and f
p
f
q
.
Proposition 2.12. (1) Assume that , are as in 2.11 above and let a se-
quence =
<
.
(2) If
, then P
(p, P
), then the set B
def
=
_
< :
<
_
is in the
lter T
. In the game, the stages B are ignored and only those for B
are active. Also, at each stage we may create
<
=
|+1|
),
(c)
Q = P
, Q
: <
= lim(
is also proper).
Proof. Our arguments rene those presented in the proof of [8, Theorem 3.2, p.
217], but the dierences in the games involved eliminate the use of trees of condi-
tions. However, trees of conditions are implicitely present here too. The tree at
level of the argument is indexed by
T
=
_
_
: w
_
and it is formed in part by conditions played in the game for various t
w
=
t T
: rk(t) =
).
The strategy st instructs Generic to play the game
rc2a
on each relevant coordinate
<
. At stage < Generic will be concerned
with coordinates w
of size < . If
,
= p
,
: <
) and q
,
= q
,
: <
),
respectively. Generics innings in the subgame of
rc2a
(p, P
) will be associated
with sequences t
= t
j
: j <
) =
t
, q
(for <
as follows. If t = t
j
, w
and = (t)
<
rc2a
(p, P
i +j we will have p
() = p
,
and q
() = q
,
, where =
+.
To keep track of what happens at coordinates / w
.
Let us note that the construction of st presented in detail below would be some-
what simpler if we knew that all forcings Q
, r
and p
,
and/or p
)
have upper bounds. However, many natural forcing notions tend to have strategic
completeness only (see [9, Part B]).
Let us formalize the ideas presented above. For each <
pick a P
name st
such that
st
0
_
Q
_
such that
if Incomplete plays
as well .
In the course of a play of
rc2a
(p, P
, st
(for w
+1
w
), p
,
, q
,
, p
,
(for <
), and r
, r
.
These objects will be chosen so that if
, p
, q
: <
) : <
_
is a play of
rc2a
(p, P
, r
, r
0
(0) = r
0
(0) = p(0), w
, [w
[ = [ +1[,
<
Dom(r
) =
<
w
, w
0
= 0, w
w
+1
and if is limit then w
<
w
.
()
2
For each < < we have ( w
+1
)(r
() = r
() = r
()) and
p r
, and p
,
(for <
).
()
3
If
, then
r
the sequence r
(), r
0
_
Q
_
in which Complete follows st
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 17
and if w
+1
w
, then st
is a P
(), Q
). (And st
0
is a nice winning strategy of Generic
in
rc2a
(p(0), Q
0
).)
()
4
t
= t
j
: j <
) is an enumeration of
=
w
.
()
5
,
= p
,
: <
) and
q
,
= q
,
: <
) are P
of
length
(for
<
w
).
()
6
If w
+1
w
, p
,
, q
,
: <
) : ) is a partial play of
rc2a
(r
(), Q
.
()
7
If =
i +j, i <
, j <
, then
Dom(p
,
) = Dom(p
) = w
Dom(p)
_
<
Dom(r
)
_
<
Dom(q
),
and for each w
the condition p
,
is an upper bound to
p r
: <
q
+j
< & i
<
& j
<
& t
j
= t
j
.
()
8
If j <
, i <
, w
, (t
j
)
= and =
i +j, =
+ , then
p
,
() = p
() = p
,
and q
() = q
,
.
()
9
If =
i + j, i <
, j <
and t
: w
,
t t
j
, then
p
the sequence
p
,
(), p
() :
+j
& i
<
& j
<
& t t
j
)
is a legal partial play of
0
(Q
.
()
10
Dom(r
) = Dom(r
) =
Dom(q
) : <
and if
,
t
: w
, and q P
, q r
and q q
whenever
=
i +j, i <
, j <
and t t
j
, then
q
P
if the set
p() r
() : <
q
() : =
i +j & i <
& j <
& t t
, then r
() : < .
Assume that the two players arrived to stage of
rc2a
(p, P
) and
, p
, q
: <
) : <
_
is the play constructed so far, and that Generic followed st and determined objects
listed in ()
so that the
demands of ()
1
are satised (and that at the end we will have
<
Dom(r
) =
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
18 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
<
w
). If < and w
,
, q
,
for < (see
()
6
), but we have not yet dened those objects when =
0
+1 and w
w
0
.
So if =
0
+ 1 and w
w
0
then let p
,
= p
,
: <
) and
q
,
= q
,
: <
, p
,
, q
,
: <
rc2a
(r
0
(), Q
and
p
,
= q
,
for all < , <
.
Condition ()
4
and our rule of taking the <
= t
j
: j <
) of
) will correspond to
steps in the
subgames of
rc2a
(r
(), Q
) (when w
+1
w
, j <
,
and assume also that Generic (playing according to st) has already dened p
,
, q
,
for w
, <
and p
,
for
+j
>
+j
, w
and t
j
= t
j
,
then p
p
,
.
For each w
and < (t
j
)
let p
+
,
= q
+
,
be P
+
,
for = (t
j
)
, w
in
rc2a
(p, P
)
is chosen so that Dom(p
) = Dom(p
,
) and clauses ()
8
+ ()
9
hold. After this
Antigeneric answers with a condition q
+
,
for w
and = (t
j
)
by the demand in ()
8
. She
also picks p
+
,
= q
+
,
for w
and (t
j
)
< <
so that ()
6
holds.
This completes the description of what happens during the
steps of the
subgame. After the subgame is over and the sequence p
, q
: <
) is
constructed, Generic chooses conditions r
, r
by ()
1
()
3
and ()
10
.
(Note: since st
are names for nice strategies, if
, i
0
, i
1
<
, j
0
, j
1
<
0
=
i
0
+j
0
,
1
=
i
1
+j
1
, t
0
, t
1
: w
, t
0
t
j0
, t
1
t
j1
and
t
0
,= t
1
, then the conditions q
0
, q
1
are incompatible.)
This nishes the description of the strategy st.
Let us argue that st is a winning strategy for Generic. Suppose that
, p
, q
: <
) : <
_
is a play of
rc2a
(p, P
<
Dom(r
). For Dom(r)
let r() be a P
such that
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 19
()
11
if w
+1
w
r() r
() and r()
Q
_
<
__
j<
__
<
__
q
+j
,
G
_
.
Clearly r is well dened (remember ()
6
) and ( < )(r
r) and p r.
Suppose now that < and r
and a condition
r
)(q
i+j
r
) be the
increasing enumeration of w
. For
and q P
, let st(, q) be a
winning strategy of Complete in
0
(P
, r
and (t)
<
such that
()
12
r
,
()
13
if i <
, j <
and (t
j
)
= (t)
i+j
,
()
14
r
), r
0
(P
, r
)
in which Complete uses her winning strategy st(
, r
).
Suppose that
<
and
r
, r
= s
[,
) P
.
Then plainly r
and q
i+j
r
whenever
()
i,j,
15
i <
, j <
and (t
j
)
= (t)
we show that q
i+j
r
whenever ()
i,j,
15
holds. For we are already done, so assume [,
i+j
r
whenever ()
i,j,
15
holds. It follows from ()
7
+()
9
that the
condition r
forces in P
that
the set
p() r
() : <
_
q
() : =
i +j & i <
& j <
&
_
<
__
(t
j
)
= (t)
.
and therefore we may use ()
10
to conclude that
r
if ()
i,j,
15
holds, then q
i+j
() r
() r
() = r
() .
The limit stages are trivial and we may claim that q
i+j
whenever ()
i,j,
15
holds. Next, r
is determined by ()
14
.
Now suppose that = + 1
, r
and (t)
r(
)
Q
_
<
__
<
__
q
+
,
_
,
so we may pick = (t)
and a condition s P
+1
such that r
and
s
_
<
__
q
+
,
s(
)
_
.
It follows from ()
13
+()
8
that then also q
i+j
(
+ 1) s whenever i <
,
j <
and (t
j
)
= (t)
for . We let r
= s
and q
i+j
whenever i <
,
j <
and (t
j
)
= (t)
for . Again, r
is determined by ()
14
.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
20 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
After the induction is completed look at r
= r
and j <
such that t
j
=
(t)
: <
).
Theorem 2.14. Assume (a), (b) of 2.13. Suppose that | is a normal lter on
and
(c)
Q = P
, Q
: <
.
Then P
= lim(
).
We assume that
, p
, q
: <
) : <
_
is a play in which Generic
follows st and the objects listed in ()
<
Dom(r
) =
<
w
,
< , we have
r() r
() and
r()
Q
< : (j <
)( <
)(q
+j
,
G
) |
P
+1
.
Then, for each Dom(r), we choose P
+1
names A
i
for elements of | such that
r()
Q
(
<
A
i
)(j <
)( <
)(q
+j
,
G
) .
Finally, we show that for each limit ordinal < ,
r
P
( w
)(
<
A
i
) (j <
)(i <
)(q
i+j
G
) .
For this we start with arbitrary condition r
r such that
r
P
( w
)(
<
A
i
)
and we repeat the arguments from the end of the proof of 2.13 to nd j <
and
r
)(q
i+j
r
).
3. Reasonable ultrafilters with small generating systems
Our aim here is to show that, consistently, there may exist a very reasonable
ultralter on an inaccessible cardinal with generating system of size less than 2
.
Lemma 3.1. Assume that G
Q
0
) is
an ultralter on , r G
_
l(G
)
_
+
.
Then
Y
def
=
_
_
Z
r
: C
r
and p
P
A
Z
r
/ d
r
_
l(G
).
Proof. Assume towards contradiction that Y / l(G
). Then we may nd s G
such that r
0
s and Y l(s). Take < such that
if C
s
,
then Z
s
Y d
s
and (A d
s
)( C
r
)(A Z
r
d
r
).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 21
(Remember 1.5.) Now take a generic lter G P over V such that p G and
work in V[G]. Since A
G
l(s)
+
, we may pick C
s
such that < and
A
G
Z
s
d
s
. Then also Z
s
G
Y d
s
G
Z
r
Y d
r
. In particular, Z
r
Y ,= , so p A
Z
r
/ d
r
, and thus
A
G
Z
r
/ d
r
. Consequently Z
s
G
Z
r
Y / d
r
giving a contradiction.
Theorem 3.2. Assume that
(i) is strongly inaccessible, =
: < ), each
is a regular cardinal,
and
_
f
__
<
f()
<
_
for < ;
(ii)
Q = P
, Q
Q
0
is a
0
downward closed super reasonable family such that
l(G
) is an ultralter on .
Then
P
l(G
) is an ultralter on .
Proof. The proof is by induction on the length of the iteration
Q. So we assume
that (i)(iii) hold and for each <
()
l(G
) is an ultralter on .
Note that (by the strategic (<)completeness of P
) forcing with P
_
V
_
Q
0
_
V
P
.
Claim 3.2.1. Assume that
(a) A
is a P
_
l(G
)
_
+
,
(b) w []
<
and T is a nite standard (w, 1)
tree, and
(c) p = p
t
: t T) is a (nite) tree of conditions in
Q, and
(d) r G
Z
r
d
r
).
Then
Z
r
: X l(G
).
Proof of the Claim. Induction on [w[.
If w = and so T = ), then the assertion follows directly from Lemma 3.1
(with p, P there standing for p
, P
here).
Assume that [w[ = n + 1,
= max(w), w
= w
, p. Let P
be a P
= p[
, ) : p P
such that
if G P
,
then V[G] [= f
P
[G]
g if and only if (p G)(p f
P
p g).
Note that P
[G]
is dened in V[G] only. Also
P
.
We are going to dene a P
name Y
is generic over V and work in V[G]. For t T such that rk(t) = let X
t
consist
of all C
r
for which there is f P
such that
p
t
[
, )
P
[G]
f and f
P
[G]
A
Z
r
d
r
.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
22 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Let Y
t
=
_
Z
r
: X
t
_
(for t T such that rk(t) = ). It follows from Lemma
3.1 that each Y
t
belongs to l(G
) (remember that
P
l(G
) is an ultralter
by ()
). Hence
Y
def
=
_
Y
t
: t T & rk(t) =
_
l(G
).
Note that for each C
r
, either Z
r
Y
= or Z
r
Y
.
Going back to V, let Y
, Y
t
, X
t
be P
, Y
t
, X
t
above. Thus
P
l(G
, T
= t
: t T and p
= p
t
: t
) P
. Thus, if X
is the
set of all C
r
for which there is a tree of conditions q
= q
t
: t
) P
such that q
and
_
t
T
__
rk(t
) =
t
P
Z
r
d
r
_
,
then
_
Z
r
: X
l(G
).
Now suppose that X
is witnessed by q
and let t
T be such that
rk(t
) =
. Then q
t
P
Z
r
and hence q
t
P
t
for all t T with
rk(t) = , so we have P
names f
t
for elements of P
such that
q
t
P
p
t
[
, )
P
t
& f
t
P
Z
r
d
r
.
Now use 2.5 (or just nite induction) to get a tree of conditions
q
= q
t
: t
) P
and objects g
t
t
(for t
, t T, rk(t
) =
and
q
t
P
t
= g
t
t
. Now, for t T put
q
t
= q
t
if rk(t)
, and
q
t
= q
g
t
t
if rk(t) = .
It should be clear that q = q
t
: t T) is a tree of conditions in
Q, p q and for
every t T with rk(t) = we have q
t
P
A
Z
r
d
r
is
included in the set X dened in the assumption (d), and hence
_
Z
r
: X
_
l(G
).
Let A
be a P
_
l(G
)
_
+
and let
p P
. We will nd a condition p
p such that p
P
A
l(G
). It will be
provided by the winning criterion ()
tree
A
of the game
treeA
(p,
Q) (see Denition
2.7; remember P
is reasonably
A(
0
(P
(G
). In the
course of a play of
(G
(G
, i
, u
, r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
.
Also, let us assume that INC (playing according to st
, p
, q
: <
_
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 23
of
treeA
(p,
Q) (in which Generic plays according to st). Let a standard tree T
= p
t
: t T
2
(so [T
[ <
).
On the board of
(G
), the strategy st
def
= t T
: rk
(t) =
and the <
rst enumeration u
= u
,i
: i < i
) of [I
]
<
(so i
<
). Now
the two players start playing a subgame of length i
to determine a sequence
r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) of trees of conditions in P
so that
()
3
q
i
= q
t,i
: t T
, the sequence
q
0
t,i
, q
1
t,i
: i < i
) is a legal play of
0
(P
rk(t)
, p
t
) in which Complete uses
her winning strategy st(rk
(t), p
t
).
Suppose that COM and INC arrive at level i < i
(G
)) and
()
i
4
r
,j
, r
,j
, (
,j
, Z
,j
, d
,j
) : j < i) and q
0
j
, q
1
j
: j < i)
have been determined and COM has chosen r
,i
G
in
Q such that
()
a
5
q
= q
t
: t T
) and q
t
P
rk(t)
is an upper bound to the set p
t
q
1
t,j
:
j < i (for each t T
)
(remember ()
3
). Then INC lets X C
r,i
to be the set of all C
r,i
greater
than sup
_
<
j<i
Z
,j
j<i
Z
,j
_
+ 890 and such that
()
b
5
there is a tree of conditions q
in
Q such that q
and
if t u
,i
, then q
t
P
A
Z
r,i
d
r,i
.
Since u
,i
is nite, it follows from 3.2.1 that
_
Z
r,i
: X
_
l(G
). Then
INC picks also the club C of such that C C
r,i
and r
,i
is restrictable to
X, C) (see Denition 1.7) and min(C) = min(X), and his inning at the stage i of
the subgame of
(G
) is r
,i
= r
,i
X, C) (again, see Denition 1.7; note that
r
,i
G
by 1.8).
After this COM answers with (
,i
, Z
,i
, d
,i
) #
_
r
,i
_
, and then INC chooses
(for the construction on the side) the <
q
0
i
and
()
6
if t u
,i
, then q
0
t,i
P
A
Z
,i
d
,i
.
Then q
1
i
= q
1
t,i
: t T
(t), p
t
) (for t T
of Antigeneric in the
play of
treeA
(p,
Q) which he is constructing on the side:
()
7
q
is the <
t
: t T
) such that q
0
i
q
1
i
q
.
(There is such a tree of conditions by ()
3
; remember i
<
.)
This completes the description of the strategy st
. Since G
is super reason-
able, st
, i
, u
, r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
of
(G
, but
()
9
for some r G
, for every j
: < )
<
I
we have
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
u
,i
#(r).
Let T
, p
, q
t
: t T
& rk
. Note that if we
show that
()
10
it is forced in P
: < )
<
I
we have
_
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
u
,i
_
#(r),
then we will be able to conclude that p
l(r) (remember ()
6
+ ()
7
and
1.10), nishing the proof of the Theorem. So let us argue that ()
10
holds true.
It follows from the description of st
,0
sup
_
_
i<i
Z
,i
_
<
+1
_
.
Now, we will say that C
r
is a sick case whenever there are
0
<
1
< and
B d
r
such that Z
r
[
0
,
1
) and
_
[
0
,
1
)
__
t I
__
i < i
__
t / u
,i
or B Z
,i
/ d
,i
_
.
Using 1.10(2) one can easily verify that the following two conditions are equivalent:
()
one
11
there is j
: < )
<
I
such that
_
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
u
,i
_
#(r),
()
two
11
there are many sick cases of C
r
.
Since the forcing with P
and ( <
)(2
<
),
() G
Q
0
is super reasonable.
Then
P
G
is strongly reasonable .
Proof. First of all note that the forcing notion P
is reasonably
A(
Q)bounding
over and proper (see 2.8). Therefore
P
_
[G
_
V
is conal in [G
, and
consequently
P
G
is (<
+
)directed (with respect to
0
).
Suppose that st
is a P
name, p P
and
P
st
is a strategy of INC in
(G
) such that
all values given by it are from V .
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 25
We are going to nd a condition p
p and a P
name g
such that
p
P
g
is a play of
(G
but
COM wins the play .
The condition p
treeA
(p,
Q) (see Denition 2.7).
In the rest of the proof whenever we say INC chooses/picks x such that we
mean INC chooses/picks the <
0
(P
, q) (for , q P
) such
that the coherence conditions of 2.3 are satised.
We are going to describe a strategy st
(G
). In the
course of a play of
(G
(G
. Thus players
INC/COM will appear in the play of
(G
(G
) in
V
P
. To avoid confusion we will refer to them as COM
V
, INC
V
for
(G
) (in V)
and COM
V
P
, INC
V
P
for
(G
) (in V
P
).
So suppose that INC
V
and COM
V
arrived at a stage < of the play of
(G
)
(in V), and INC
V
(playing according to st
1
a partial play T
, p
, q
: < ) of
treeA
(p,
Q) in which Generic plays
according to st, and
()
2
a P
name g
= I
, i
, u
, x
(G
) (in
V
P
) in which INC
V
P
uses the strategy st
,
()
3
ordinals i
<
such that q
t
i
= i
for every t T
with rk
(t) =
(for < ).
Note that I
is a P
from V, u
is a P
name for
an i
and x
is a P
of level .
Let I
be a P
to the play g
of
(G
) (in V
P
).
Let T
and p
= p
t
: t T
1
. Let q
= q
t
: t T
) be a tree of conditions in
Q such that
()
a
4
p
and q
t0
, q
t1
are incompatible for distinct t
0
, t
1
T
with rk
(t
0
) =
rk
(t
1
),
()
b
4
for every t T
with rk
t
decides the value of I
,
say q
t
P
I
= I
t
.
(Note that
P
I
V by the choice of st
; remember 2.5.)
In the play of
(G
), the strategy st
instructs INC
V
to choose the set
I
I
t
: t T
& rk
(t) =
and an enumeration u
= u
,i
: i < i
) of [I
]
<
. Note that [I
t
[ <
for all
relevant t T
and [T
[ <
, so by our assumptions on
and
we know that
[I
[ <
(so also i
<
).
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
26 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Then, in the play of
(G
), INC
V
P
pretends that COM
V
P
played an ordinal
i
[i
, ) and u
= u
,i
: i < i
) such that
P
u
[I
]
<
and
_
u
,i
: i < i
= I
and for each t T
with rk
(t) = we have
q
t
P
i
= i
and u
,i
= c(t) : c u
,i
for i < i
.
Now, both in
(G
) of V
P
and in
(G
= i
) of trees of conditions in
Q such that (in addition to
demands stated later):
()
a
5
q
j
= q
t,j
: t T
), q
q
0
j
q
1
j
q
0
i
for < 2, j < i < i
, and
()
b
5
for each t T
, the sequence q
0
t,i
, q
1
t,i
: i < i
0
(P
rk(t)
, q
t
) in which Complete uses her winning strategy st(rk
(t), q
t
).
He (as INC
V
P
) will also construct a name for a play of a subgame of
(G
) of
V
P
for this stage.
Suppose that INC
V
and COM
V
have arrived to a stage i < i
of the subgame
and INC
V
has determined on the side q
j
for j < i, < 2 and a P
name z
j
: j < i)
for a partial play of the subgame of
(G
) of V
P
. Now COM
V
chooses r
,i
G
which INC
V
passes to INC
V
P
as an inning of COM
V
P
at the ith step of the
subgame of level of
(G
) in V
P
. There the strategy st
gives INC
V
P
an
answer
,i
< .
Next, INC
V
picks a tree of conditions q
0
i
= q
0
t,i
: t T
) in
Q such that
()
a
6
(j < i)( q
1
j
q
0
i
) and q
q
0
i
, and
()
b
6
for every t T
with rk
,i
,
say q
0
t,i
P
,i
=
t
,i
.
Then INC
V
lets
,i
= sup
_
_
t
,i
: t T
& rk
(t) =
_
_
<
_
j<i
Z
,j
_
j<i
Z
,j
_
+ 890
and in the subgame of
(G
,i
such that
C
r
,i
= C
r,i
,i
and d
r
,i
= d
r,i
for C
r
,i
.
(Note that r
,i
G
by 1.11(2)(ii).)
After this COM
V
chooses (
,i
, Z
,i
, d
,i
) #(r
,i
), so
,i
C
r,i
,
,i
,i
and d
,i
= d
r,i
,i
. Next INC
V
lets
q
1
i
be the tree of conditions in
Q fully determined by demand ()
b
5
and
z
i
be a P
(G
) in V
P
such that for each t T
with rk
(t) = we have
q
0
t,i
P
z
i
=
_
r
,i
,
,i
, (
,i
, Z
,i
, d
,i
)
_
.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 27
Then the subgame continues.
After all i
= q
t
: t T
) in
Q such that (i < i
)( q
1
i
q
be
a P
(G
) in V
P
such that
x
= z
i
: i < i
(G
).
Since G
, i
, u
, r
,i
, r
,i
, (
,i
, Z
,i
, d
,i
) : i < i
) : <
_
of
(G
, but
()
8
for some r G
, for every j
: < )
<
I
we have
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
u
,i
#(r).
Exactly as in the proof of Theorem 3.2 we may argue that then also
()
9
it is forced in P
that
_
<
I
___
(
,i
, Z
,i
, d
,i
): < & i < i
& j
u
,i
_
#(r)
_
.
(See ()
10
in the proof of 3.2.)
Let T
, p
, q
t
: t T
& rk
.
Also, let g
be the P
name of a play of
(G
) (in V
P
) constructed on the side
in the same run of
(G
) (see ()
2
). We are going to argue that
()
10
the condition p
forces (in P
) that
_
<
I
__
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
,i
#(r)
_
,
that is
p
P
COM
V
P
wins the play g
as witnessed by r .
Suppose that G P
is generic over V, p
such that rk
(t) = and q
t
G, and thus
_
I
_
G
= I
t
,
_
i
_
G
= i
and
_
u
_
G
= (u
,i
)
G
: i < i
), where
_
u
,i
_
G
= c(t) :
c u
,i
I
t
. Suppose that
j = j
: < )
<
I
t()
I
t
: t T
& rk
(t()) = j
. Note that if
j
u
,i
, i < i
, then j
_
u
,i
_
G
and therefore
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
_
u
,i
_
G
(
,i
, Z
,i
, d
,i
) : < & i < i
& j
u
,i
#(r)
(remember ()
9
). Now ()
10
follows and the proof of the theorem is complete.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
28 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Corollary 3.4. Assume that is a strongly inaccessible cardinal. Then there is a
forcing notion P such that
P
is strongly inaccessible and 2
=
++
and
there is a strongly reasonable family G
Q
0
such that
l(G
) is an ultralter on and [G
[ =
+
, in particular
there is a very reasonable ultralter on
with a generating system of size < 2
: <
+
) Q
0
such that G
def
= r Q
0
: ( <
+
)(r
0
r
) is super
reasonable and l(G
) is an ultralter on .
Let
Q = P
, Q
: <
++
) be a support iteration of the forcing notion
Q
tree
D
(K
1
,
1
) dened in the proof of [9, Prop. B.8.5]. This forcing is reasonably A
bounding (by [8, Prop. 4.1, p. 221] and [9, Thm B.6.5]), so we may use Theorems
3.2 and 3.3 to conclude that
++
G
is strongly reasonable, [G
[ =
+
< 2
and
l(G
) is ultralter on .
If one analyzes the proof of Theorem 3.3, one may notice that even
++
r
: <
+
is strongly reasonable .
) is not an
ultralter anymore.
In this section we assume that is a strongly inaccessible cardinal.
Denition 4.1. (1) Let P
is
dened by letting p q if and only if
() C
q
C
p
,
q
min(C
p
) =
p
min(C
p
), and
() for every successive members < of C
p
we have
_
[, )
__
q
() =
p
()
q
()
p
()
_
.
(2) For p P
and C
p
let
pos(p, )
def
=
_
q
: q P
& p q
_
.
(3) For p P
p = (
p
[, ), C
p
).
(Plainly,
p P
.)
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 29
Remark 4.2. P
: < ),
= 2
||+0
(for < ). Then P
is a
nicely double abounding over forcing notion. Also [P
[ = 2
.
Proof. One easily veries that the relation
P
is transitive and reexive, also
plainly [P
[ = 2
.
Claim 4.3.1. P
is (<)complete.
Proof of the Claim. Suppose that < and p
: < ) is a
P
increasing
sequence of conditions in P
. Let C =
<
C
p
(),
if < are successive members of the club C, < and = min
_
<
: < min
_
C
p
( + 1)
__
, then () =
p
()
p
().
Plainly, is well dened and q
def
= (, C) P
q).
To this end suppose < . Clearly C C
p
), then
() =
p
). Since p
, we have
() =
p
() and thus
p
() = ().
Next, suppose that < are successive members of C
p
and < . If
< min(C) and = min
_
< : < min(C
p
)
_
, then > , () =
p
() and
()
1
() =
p
() =
p
()
()
p
() =
p
()
()
p
().
So assume C ,= and let
<
<
. Let = min
_
< : < min
_
C
p
(
+ 1)
__
. If =
,
then and
()
2
() =
p
()
p
() =
p
()
p
()
()
p
() =
()
p
() =
p
()
()
p
()
(as () = (
) = 1). If
)
p
)
p
() =
()
()
p
() =
()
()
p
()
()
p
() =
p
()
()
p
().
Clearly ()
1
()
3
are what we need to justify 4.1(1) and conclude p
q.
Claim 4.3.2. Let p P
rc2a
(p, P
). When-
ever we say Generic chooses x such that we mean Generic chooses the <
rst x
such that (and likewise for other variants).
During a play of
rc2a
(p, P
: <
) and
=
and
: + = C
p
(
+
+ 1),
(b) if < , then p
and
p
+
=
p
+
,
(c)
: = C
p
: otp( C
p
) and p
0
= p,
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
30 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
(d)
++1
and p
+1
are determined right after stage of
rc2a
(p, P
).
So suppose that the two players have arrived to a stage < of a play of
rc2a
(p, P
: + and p
j
: j <
)
of pos(p
,
+
) such that
0
=
p
+
. Antigeneric picks a non-zero ordinal
. In the course
of the subgame, in addition to her innings p
<
C
q
and
[
+
,
) =
[
+
,
) for
< <
,
(f) if =
i + 2j, i <
and j <
, then
(i)
+1
,
=
q
, and
+1
() =
q
+1
() for
[
+
,
+1
),
(ii) p
0
+
p
, min(C
p
0
) >
+
, and (
for
< , and
(iii) q
+1
+1
+1
q
+1
.
So suppose that the two players have arrived to a stage =
i + 2j (i <
,
j <
, q
< . Let
=
<
[
+
,
< ) is
P
increasing, so Generic may choose an upper bound p
to it.
(Note that necessarily
p
, sup(
< ) min(C
p
in
the subgame and Antigeneric answers with q
C
q
be such that [C
q
=
q
1, 1 be dened by
+
=
j
and () =
() for [
+
,
). Then
Generic plays p
+1
=
+1
p
+1
. Finally, Generic picks
+1
C
q
+1
such
that [C
q
+1
+1
[ = 1 and she takes
+1
:
+1
1, 1 such that
+1
and
+1
() =
q
+1
() for [
,
+1
). Plainly, if i
< i,
+ 2j then
q
and q
+1
p
+1
so both p
and p
+1
are legal innings in
rc2a
(p, P
).
Also easily the demands in (e)+(f) are satised. Moreover, if j
< j <
then the
conditions p
+j
and p
+j
are incompatible.
After the subgame is over, Generic lets
=
[
+
,
) : <
,
and she picks a
P
upper bound p
+1
to the increasing sequence
(
: <
).
Note that
min
_
C
p
+1
_
and
+1
for all <
, so also
0
=
+
p
+1
. Also
(g) (
p
+1
+1
for all <
.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 31
Let p
+1
P
be such that C
p+1
=
: + C
p
+1
and
p+1
=
p
+1
(plainly p
p
+1
) and let
++1
= min
_
C
p
+1
_
.
This nishes the description of the strategy st. Let us argue that st is a winning
strategy for Generic. To this end suppose that
()
, p
, q
: <
) : <
_
is a result of a play of
rc2a
(p, P
, p
: <
),
j
: j <
)
(and the demands in (a)(g) are satised). Let C =
<
+
(clearly : 1, 1; remember (b)), and let
p
= (, C). It is a condition in P
(for < ) so
also p
, the condition p
i+j
for all
i <
)
++1
. Let j
<
be such that
+
=
j
. We consider two cases now.
Case 1:
p
(
+
) =
p
(
+
) =
p+1
(
+
).
Then
p
[
+
,
++1
) =
p+1
[
+
,
++1
). Let j = 2 j
<
, and we will
argue that q
i+j
p
. So let i <
, =
i + j. By the choice
of j
we know that
p
+
=
j
=
q
+
and also
[
+
,
) =
p+1
[
+
,
) =
p
[
+
,
).
Hence (by (f)(i))
p
=
q
and now
q
(
p
(
p
+
)
+
p
+1
=
(
p
++1
)
++1
p
+1
= (
p
++1
)
++1
p
+1
(
p
++1
)
++1
p
(
p
++1
)
++1
p
= p
(
+
) =
p
(
+
) =
p+1
(
+
).
Then
p
() =
p
() =
p+1
() for all [
+
,
++1
). Let j = 2 j
+ 1
and let us argue that q
i+j
p
. So let i <
, =
i +j. Like
in the previous case we show that
p
=
q
(
p
(
p
++1
)
++1
p
+1
(
p
++1
)
++1
p
= p
be a P
=
_
p
min(C
p
) : p G
P
.
Then
P
V,
P
< :
() = 1 l(s)
+
and < :
() = 1 l(s)
+
.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
32 ANDRZEJ ROS LANOWSKI AND SAHARON SHELAH
Proof. It should be clear that
P
, s Q
0
: < ) C
p
such that for every < there is = () C
s
such that
Z
s
,
+1
). Then let C =
,
+1
)
_
p
[
,
+1
),
p
[
,
+1
)
_
,
if < is even, then
_
Z
s
()
: () = 1
_
d
s
()
,
if < is odd, then
_
Z
s
()
: () = 1
_
d
s
()
.
Now note that (, C) P
that
_
< :
() = 1
_
l(s)
+
and
_
< :
() = 1
_
l(s)
+
.
P
is strongly inaccessible and 2
=
++
and there is no very
reasonable ultralter on with a generating system of size < 2
=
+
.
Let
Q = P
, Q
: <
++
) be a support iteration of the forcing notion P
(see Denition 4.1). This forcing is nicely double abounding over (where
=
2
||+0
; remember Proposition 4.3) and hence P
++ does not
collapse any cardinals and forces that 2
=
++
. Proposition 4.4 implies that
++
for no family G
Q
0
of size < 2
, l(G
) is an ultralter on .
Problem 4.6. (1) Is it consistent that for some uncountable regular cardinal
we have that there is no super-reasonable ultralter on ? Or even no
very reasonable one?
(2) In particular, are there super-reasonable ultralters on in the model con-
structed for Corollary 4.5?
(3) Do we need the inaccessibility of for the assertions of Corollaries 3.4, 4.5
(concerning ultralters on )?
References
[1] Uri Abraham. Lectures on proper forcing. In M. Foreman A. Kanamori and M. Magidor,
editors, Handbook of Set Theory.
[2] Todd Eisworth. On iterated forcing for successors of regular cardinals. Fundamenta Mathe-
maticae, 179:249266, 2003, math.LO/0210162.
[3] Moti Gitik. On nonminimal p-points over a measurable cardinal. Annals of Mathematical
Logic, 20:269288, 1981.
[4] Martin Goldstern and Saharon Shelah. Ramsey ultralters and the reaping numberCon(r <
u). Annals of Pure and Applied Logic, 49:121142, 1990.
[5] Thomas Jech. Set theory. Springer Monographs in Mathematics. Springer-Verlag, Berlin,
2003. The third millennium edition, revised and expanded.
[6] Andrzej Roslanowski and Saharon Shelah. Lords of the iteration. In Proceedings of the Confer-
ence on Boise Extravaganza in Set Theory (BEST 2009), volume accepted of Contemporary
Mathematics (CONM). math.LO/0611131.
8
9
0
r
e
v
i
s
i
o
n
:
2
0
1
0
-
0
9
-
2
0
m
o
d
i
f
i
e
d
:
2
0
1
0
-
0
9
-
2
0
REASONABLE ULTRAFILTERS, AGAIN 33
[7] Andrzej Roslanowski and Saharon Shelah. Norms on possibilities I: forcing with trees
and creatures. Memoirs of the American Mathematical Society, 141(671):xii + 167, 1999.
math.LO/9807172.
[8] Andrzej Roslanowski and Saharon Shelah. Reasonably complete forcing notions. Quaderni di
Matematica, 17, 2005. math.LO/0508272.
[9] Andrzej Roslanowski and Saharon Shelah. Sheva-Sheva-Sheva: Large Creatures. Israel Jour-
nal of Mathematics, 159:109174, 2007. math.LO/0210205.
[10] Andrzej Roslanowski and Saharon Shelah. Generating ultralters in a reasonable way. Math-
ematical Logic Quarterly, 54:202220, 2008. math.LO/0607218.
[11] Saharon Shelah. Proper and improper forcing. Perspectives in Mathematical Logic. Springer,
1998.
[12] Saharon Shelah. The combinatorics of reasonable ultralters. Fundamenta Mathematicae,
192:123, 2006. math.LO/0407498.
Department of Mathematics, University of Nebraska at Omaha, Omaha, NE 68182-
0243, USA
E-mail address: roslanow@member.ams.org
URL: http://www.unomaha.edu/logic
Einstein Institute of Mathematics, Edmond J. Safra Campus, Givat Ram, The Hebrew
University of Jerusalem, Jerusalem, 91904, Israel, and Department of Mathematics,
Rutgers University, New Brunswick, NJ 08854, USA
E-mail address: shelah@math.huji.ac.il
URL: http://shelah.logic.at